Professional Documents
Culture Documents
kitwing@hotmail.com
Copyright
c 2021 by Kit-Wing Yu. All rights reserved. No part of this publication may be
reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic,
mechanical, photocopying, recording, or otherwise, without the prior written permission of the
author.
Dr. Kit-Wing Yu received his B.Sc. (1st Hons), M.Phil. and Ph.D. degrees in Math. at the
HKUST, PGDE (Mathematics) at the CUHK. After his graduation, he joined United Christian
College (UCC) to serve as a mathematics teacher between 2000 and 2020. From 2002 to 2020,
he also took the responsibility of the mathematics panel at UCC. Starting from Sept. 2020,
Dr. Yu has been promoted to be the Vice Principal (Academic) at Kiangsu-Chekiang College
(Kwai Chung). Furthermore, he was appointed as a part-time tutor (2002 – 2005) and then a
part-time course coordinator (2006 – 2010) of the Department of Mathematics at the OUHK.
Besides teaching, Dr. Yu has been appointed to be a marker of the HKAL Pure Mathematics
and HKDSE Mathematics (Core Part) for over thirteen years. Between 2012 and 2014, Dr. Yu
was invited to be a Judge Member by the World Olympic Mathematics Competition (China).
In the research aspect, he has published research papers in international mathematical journals,
including some well-known journals such as J. Reine Angew. Math., Proc. Roy. Soc. Edinburgh
Sect. A and Kodai Math. J.. His research interests are inequalities, special functions and
Nevanlinna’s value distribution theory. In the area of academic publication, he is the author of
the following eight books:
iii
iv
Preface
This present book A Complete Solution Guide to Real and Complex Analysis is the combined
volume of author’s two books [125] and [126]. It provides solutions to all exercises of Rudin’s
classical graduate book [100].
The primary aim of this book is to help every mathematics student and instructor to under-
stand the ideas and applications of the theorems in Rudin’s book. To accomplish this goal, I
have adopted the usual way I wrote all my published solution guides. In other words, I intend
writing the solutions as comprehensive as I can so that you can understand every detailed part
of a proof easily. Apart from this, I also keep reminding you what theorems or results I have
applied by quoting them repeatedly in the proofs. By doing this, I believe that you will become
fully aware of the meaning and applications of each theorem.
Before you go into details of this book, I have two gentle reminders for you. Firstly, as a
mathematics instructor at a college, I understand that the growth of a mathematics student
depends largely on how hard he/she does exercises. When your instructor asks you to do some
exercises from Rudin, you are not suggested to read my solutions unless you have tried your
best to prove them seriously yourselves. Secondly, when I prepared this book, I found that some
exercises require knowledge that Rudin did not cover in his book. To fill this gap, I suggest you
to check the references provided by the solutions or read directly the reference list of this book.
Of course, I will use my books [124], [127], [128], [125], [123] and [126] freely.
• It covers all the 397 exercises from Chapters 1 to 20 with detailed and complete solutions.
As a matter of fact, my solutions show every detail, every step and every theorem that I
applied.
• There are 40 illustrations for explaining the mathematical concepts or ideas used behind
the questions or theorems.
• Sections in each chapter are added so as to increase the readability of the exercises.
• Different colors are used frequently in order to highlight or explain problems, lemmas,
remarks, main points/formulas involved, or show the steps of manipulation in some com-
plicated proofs. (ebook only)
• Necessary lemmas with proofs are provided because some questions require additional
mathematical concepts which are not covered by Rudin.
• Many useful or relevant references are provided to some questions for your future research.
v
vi
Since the solutions are written solely by me, you may find typos or mistakes. If you really
find such a mistake, please send your valuable comments or opinions to
kitwing@hotmail.com.
https://sites.google.com/view/yukitwing/
irregularly.
Kit Wing Yu
April 2021
List of Figures
vii
viii List of Figures
π π
14.5 The conformal mapping f : U → A(e− 2 , e 2 ). . . . . . . . . . . . . . . . . . . . . 458
14.6 The locus of f (z) for t ∈ (0, π). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
14.7 The locus of f (z) for t ∈ (π, 2π). . . . . . . . . . . . . . . . . . . . . . . . . . . . 459
14.8 The image of f (E) when r = 0.25. . . . . . . . . . . . . . . . . . . . . . . . . . . 460
Preface v
1 Abstract Integration 1
1.1 Problems on σ-algebras and Measurable Functions . . . . . . . . . . . . . . . . . 1
1.2 Problems related to the Lebesgue’s MCT/DCT . . . . . . . . . . . . . . . . . . . 7
3 Lp -Spaces 69
3.1 Properties of Convex Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Relations among Lp -Spaces and some Consequences . . . . . . . . . . . . . . . . 71
3.3 Applications of Theorems 3.3, 3.5, 3.8, 3.9 and 3.12 . . . . . . . . . . . . . . . . . 87
3.4 Hardy’s Inequality and Egoroff’s Theorem . . . . . . . . . . . . . . . . . . . . . . 91
3.5 Convergence in Measure and the Essential Range of f ∈ L∞ (µ) . . . . . . . . . . 106
3.6 A Converse of Jensen’s Inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.7 The Completeness/Completion of a Metric Space . . . . . . . . . . . . . . . . . . 112
3.8 Miscellaneous Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
ix
x Contents
7 Differentiation 207
7.1 Lebesgue Points and Metric Densities . . . . . . . . . . . . . . . . . . . . . . . . 207
7.2 Periods of Functions and Lebesgue Measurable Groups . . . . . . . . . . . . . . . 210
7.3 The Cantor Function and the Non-measurability of f ◦ T . . . . . . . . . . . . . 216
7.4 Problems related to the AC of a Function . . . . . . . . . . . . . . . . . . . . . . 218
7.5 Miscellaneous Problems on Differentiation . . . . . . . . . . . . . . . . . . . . . . 232
17 H p -Spaces 519
17.1 Problems on Subharmonicity and Harmonic Majoriants . . . . . . . . . . . . . . 519
17.2 Basic Properties of H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
17.3 Factorization of f ∈ H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 531
17.4 A Projection of Lp onto H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
17.5 Miscellaneous Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 544
Index 589
xii Contents
Bibliography 591
CHAPTER 1
Abstract Integration
Problem 1.1
Rudin Chapter 1 Exercise 1.
Proof. Let X be a set. Assume that M was an infinite σ-algebra in X which has only countably
many members. Then we have M = {∅, X, A1 , A2 , . . .}. Let x ∈ X be fixed, we define the set
\
Sx = A, (1.1)
x∈A
where x ∈ A ∈ M. (By this definition (1.1), it is clear that x ∈ Sx .) Since M is countable, the
intersection in (1.1) is actually at most countable. Thus it follows from Comment 1.6(c) that
Sx ∈ M. Next we define the set
S = {Sx | x ∈ X}
so that S ⊆ M. We need a lemma about the set S:
Lemma 1.1
If Sx , Sy ∈ S and Sx ∩ Sy 6= ∅, then Sx = Sy .
Sx ⊆ Sz . (1.2)
On the other hand, if w ∈ Sz , then w ∈ A for all A containing z. Thus the hypothesis
z ∈ Sx ∩ Sy implies that w ∈ Sx ∩ Sy , i.e., Sz ⊆ Sx ∩ Sy . By this result and the
definition (1.1), we obtain
Sz ⊆ Sx ∩ Sy ⊆ Sx . (1.3)
Combining the observations (1.2) and (1.3), we have the desired result that Sz = Sx .
Similarly, we can show that Sz = Sy . Hence we conclude that Sx = Sy .
1
2 Chapter 1. Abstract Integration
Now we return to the proof of the problem. By Lemma 1.1, we may assume that all elements
in S are distinct. If B ∈ M, then the definition (1.1) certainly gives y ∈ Sy for every y ∈ B so
that [
B⊆ Sy . (1.4)
y∈B
Therefore, if the cardinality of S is N , then the cardinality of its power set 2S is 2N and so
there are at most 2N elements in M by the inclusion (1.4), a contradiction. Thus the set S must
be infinite.
Recall that S ⊆ M and M is countable, S is also countable by [99, Theorem 2.8, p. 26].
Since M is a σ-algebra, it must contain the power set of S. However, it is well-known that the
power set of an infinite countable set is uncountable [93, Problem 22, p. 16]. We conclude from
these facts that M is uncountable, a contradiction. This completes the proof of the problem.
Problem 1.2
Rudin Chapter 1 Exercise 2.
Proof. Put f (x) = (f1 (x), . . . , fn (x)). Since f1 , . . . , fn : X → R, f maps the measurable space
X into Rn . Let Y be a topological space and Φ : Rn → Y be continuous. Define
h(x) = Φ (f1 (x), . . . , fn (x)) .
Since h = Φ ◦ f , Theorem 1.7(b) shows that it suffices to prove the measurability of f .
To this end, we first consider f −1 (R) for an open rectangle R in Rn . By the definition,
R = I1 × · · · × In , where Ii is an open interval in R for 1 ≤ i ≤ n. Now we know from [74,
Exercise 3, p. 21] that
f −1 (R) = {x ∈ X | f (x) ∈ R}
= {x ∈ X | fi (x) ∈ Ii for 1 ≤ i ≤ n}
= {x ∈ X | f1 (x) ∈ I1 } ∩ · · · ∩ {x ∈ X | fn (x) ∈ In }
= f1−1 (I1 ) ∩ · · · ∩ fn−1 (In ).
Since each fi is measurable and each Ii is open in R, the set fi−1 (Ii ) is measurable in X by
Definition 1.3(c). Thus this implies that f −1 (R) is measurable in X by Comment 1.6(c).
Our proof will be complete if we can show that every open set V in Rn can be written as a
countable union of open rectangles Rj in Rn . We need some topology. By [74, Exercise 8(a), p.
83], the countable collection
B = {I = (a, b) | a < b and a, b ∈ Q}
is a basis that generates the standard topology on R. Then we follow from [74, Theorem 15.1,
p. 86] that the collection
C = {I1 × · · · × In | Ii = (ai , bi ), ai , bi ∈ Q, 1 ≤ i ≤ n} (1.5)
is a basis for the (product) topology of Rn .a
Since elements in the collection (1.5) are open
rectangles and it is countable by [99, Theorem 2.13, p. 29], every open set V in Rn is, in fact, a
countable union of open rectangles Rj , i.e,
[
∞ [∞
f −1 (V ) = f −1 Rj = f −1 (Rj ).
j=1 j=1
a
For details, please read [74, §13 and §15].
1.1. Problems on σ-algebras and Measurable Functions 3
Hence f −1 (V ) is also a measurable set in X by Definition 1.3(a)(iii). This completes the proof
of the problem.
Problem 1.3
Rudin Chapter 1 Exercise 3.
Proof. Let f : X → R ⊂ [−∞, ∞] and M be the σ-algebra of X. Let α be real. By [99, Theorem
1.20(b), p. 9], we can find a sequence {rn } of rational numbers such that α < rn for all n ∈ N,
rn → α as n → ∞. In other words, we have
∞
[
(α, ∞] = [rn , ∞]
n=1
Recall that
f −1 ([rn , ∞]) = {x | f (x) ≥ rn }
is assumed to be measurable for each n, so we have f −1 ([rn , ∞]) ∈ M and then we follow from
Definition 1.3(a)(iii) that f −1 ((α, ∞]) ∈ M. Since α is arbitrary, we conclude from Theorem
1.12(c) that f is measurable. This finishes the proof of the problem.
Problem 1.4
Rudin Chapter 1 Exercise 4.
Proof.
(a) Let αk = sup{−ak , −ak+1 , . . .} for k = 1, 2, . . .. By the definition, we have αk ≥ −an for
all n ≥ k and if α ≥ −an for all n ≥ k, then α ≥ αk . Note that this is equivalent to the
fact that −αk ≤ an for all n ≥ k and if −α ≤ an for all n ≥ k, then −α ≤ −αk . In other
words, we have
−αk = inf{ak , ak+1 , . . . , }
for k = 1, 2, . . .. Thus this implies that
Similarly, we have
inf (−cn ) = − sup(cn ) (1.7)
n≥k n≥k
for a sequence {cn } in [−∞, ∞]. By applying the equality (1.6) and then the equality
(1.7), we achieve that
n o n o n o
lim sup(−an ) = inf sup(−an ) = inf − inf (an ) = − sup inf (an ) = − lim inf (an )
n→∞ k≥1 n≥k k≥1 n≥k k≥1 n≥k n→∞
| {z }
cn
(b) This part is proven in [124, Problem 3.5, pp. 32, 33].
for all k = 1, 2, . . .. Thus {αn } and {βn } are two sequences in [−∞, ∞] such that αn ≤ βn
for all n = 1, 2, . . .. By a similar argument, we also have
Hence we obtain from the results (1.10) and (1.11) that the strict inequality can hold in part
(b). This completes the proof of the problem.
Problem 1.5
Rudin Chapter 1 Exercise 5.
Proof.
Since f is measurable and (n, ∞] is openb in [−∞, ∞], f −1 ((n, ∞]) = {x | f (x) > n} ∈ MX
by Theorem 1.12(b) for each n ∈ N. Thus Sf (∞) ∈ MX by Comment 1.6(c). Similarly,
all the other sets Sf (−∞), Sg (∞) and Sg (−∞) belong to MX too.
Next, we let X ′ = X\(Sf (∞)∪Sf (−∞)∪Sg (∞)∪Sg (−∞)). Since Sf (∞), Sf (−∞), Sg (∞)
and Sg (−∞) are measurable, X ′ ∈ M by Definition 1.3(a)(ii) and Comment 1.6(b). By
b
See the proof of Theorem 1.12(c) in [100, p. 13]
1.1. Problems on σ-algebras and Measurable Functions 5
the comment following Proposition 1.24, we know that X ′ is itself a measure space and if
we let MX ′ be a σ-algebra of X ′ , then
MX ′ ⊆ MX . (1.12)
We need to prove one more thing: the mapping −g : X → [−∞, ∞] is measurable. Since
g is measurable, we know from [99, Definition 11.13] that {x ∈ X | g(x) > −a} ∈ MX for
every real a. It is obvious that {x ∈ X | − g(x) < a} = {x ∈ X | g(x) > −a} for every real
a, so we deduce from [99, Theorem 11.15, p. 311] that −g is measurable.
Now we are ready to prove the desired results. Notice that
where h = f − g. Since f and −g are measurable, the new (real) function h = f − g is also
measurable by Theorem 1.9(c). Since
∞
\ 1 1
h−1 (0) = h−1 − ,
n n
n=1
and h−1 ((− n1 , n1 )) ∈ MX ′ for every n ∈ N, we yield from this and the relation (1.12) that
h−1 (0) ∈ MX ′ ⊆ MX . This shows the second assertion. For the first assertion, we note
that
Hence the facts (1.13) and (1.14) show that {x | f (x) < g(x)} ∈ MX .
Problem 1.6
Rudin Chapter 1 Exercise 6.
c
This can be seen, again, from the proof of Theorem 1.12(c) or from the comment following Proposition 1.24.
6 Chapter 1. Abstract Integration
which means that Ac is at most countable too. Hence Ac ∈ M and then M is a σ-algebra
in X.
• µ is a measure on M. We check Definition 1.18(a). Since ∅ is at most countable,
we have µ(∅) = 0 so that µ is not identically ∞. In fact, it is clear that we have
µ : M → {0, 1} ⊂ [0, ∞]. Let {An } be a disjoint countable collection of members of M.
Case (i): All An are at most countable. Recall that the set A given by (1.15)
is at most countable. By the definition of µ, we have µ(A) = µ(An ) = 0 for all
n = 1, 2, . . .. Thus we have
X∞
µ(A) = µ(An ) (1.16)
n=1
in this case.
Case (ii): There is at least one Ak is uncountable. Since Ak ∈ M, Ack must
be at most countable. Since An ∩ Ak = ∅ for all n 6= k, we have An ⊆ Ack for all
n 6= k. In other words, the measurable sets An are at most countable for all n 6= k.
Therefore, we have µ(Ak ) = 1 and µ(An ) = 0 for all n 6= k. Since
∞
\
Ac = Acn ⊆ Ack ,
n=1
Ac is at most countable and then µ(A) = 1. Hence the equality (1.16) also holds in
this case.
This completes the proof that µ is a measure on M.
• The determination of measurable functions and their integrals. Let f : X → R
be a measurable function. Since X is uncountable, we have µ(X) = 1 which is the only
thing that we know and start with. For every n ∈ Z, we know that
[n, n + 1) = R \ (−∞, n) ∪ [n + 1, ∞) .
By [99, Theorem 11.15, p. 311], each set on the right-hand side in (1.17) is measurable.
This implies that
f −1 [n, n + 1) ∈ M.
Let En = f −1 [n, n + 1) , where n ∈ Z. By the definition of M and then the definition of
µ, we have either µ(En ) = 0 or µ(En ) = 1. For every x ∈ X, we must have f (x) ∈ [n, n+1)
for some n ∈ Z, i.e., x ∈ En for some n ∈ Z. Therefore, we have
∞
[
En = X.
n=−∞
∞
[
It is clear that R = [n, n + 1), so {En } is a disjoint countable collection of members
n=−∞
of M. By Definition 1.18(a), we see that
[
∞ ∞
X
µ(X) = µ En = µ(En ). (1.18)
n=−∞ n=−∞
The fact µ(X) = 1 and the equality (1.18) force that there exists an integer n0 such that
µ(En0 ) = 1. Without loss of generality, we may assume that n0 = 0, i.e.,
µ(E0 ) = µ f −1 [0, 1) = 1. (1.19)
1 1
2 ) ∪ [ 2 , 1), then the above
Next if we write [0, 1) = [0, argument and the value (1.19)
imply that either µ f −1 1
[0, 2 ) = 1 or µ f −1 [ 21 , 1) = 1. This process can be done
continuously so that a sequence of intervals {[an , bn )} is constructed such that
1
µ f −1 [an , bn ) = 1, 0 ≤ bn − a n ≤ and lim (bn − an ) = 0,
2n n→∞
for some a ∈ R and all other real numbers b 6= a, but this is equivalent to saying that f −1 (a)
is uncountable and f −1 (b) is at most countable for all b 6= a. Now we have completely
characterized every measurable function on X.
Finally, it is clear that X \ f −1 (a) is a set of measure 0. Therefore, we have
Z Z
f dµ = f dµ = a.
X f −1 (a)
Problem 1.7
Rudin Chapter 1 Exercise 7.
8 Chapter 1. Abstract Integration
Proof. Let Ek = {x ∈ X | f1 (x) > k} = f1−1 (k, ∞] and E = {x ∈ X | f1 (x) = ∞}, where
k = 1, 2, . . .. It is clear that each (k, ∞] is a Borel set in [0, ∞], so each Ek is measurable by
Theorem 1.12(b). Since
\∞
E= Ek ,
k=1
E is also measurable by Comment 1.6(c). If µ(E) > 0, then we know from the definition that
Z
f1 dµ = ∞. (1.20)
E
By using the result (1.20), Proposition 1.24(b) and Theorem 1.33, we conclude that
Z Z Z
|f1 | dµ ≥ f1 dµ ≥ f1 dµ = ∞
X X E
which contradicts the hypothesis that f1 ∈ L1 (µ). In other words, we must have µ(E) = 0 and
thus f1 ∈ L1 (µ) on X \ E.
Here we may assume that the form of Theorem 1.34 (Lebesgue’s Dominated Convergence
Theorem) is also valid for measurable functions defined a.e. on X.e Now we see that the
measurable function f1 in the problem plays the role of g in the theorem and hence our desired
result follows from Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem) immediately.
For a counterexample, we consider X = R, In = (n, ∞), µ = m the Lebesgue measure (see
[99, Definition 11.5, pp. 302, 303]) and define for each n = 1, 2, . . .,
1, if x ∈ In ;
fn (x) = χIn (x) =
0, if x ∈
/ In .
/ L1 (m). Furthermore,
for every n = 1, 2, . . .. In particular, the integrals (1.21) show that f1 ∈
we have
f (x) = lim fn (x) = lim χIn (x) = 0 (1.22)
n→∞ n→∞
for every x ∈ R. Thus we deduce from the expression (1.22) and Proposition 1.24(d) that
Z
f dm = 0. (1.23)
R
Hence the inconsistence of the integrals (1.21) and (1.23) show that the condition “f1 ∈ L1 (µ)”
cannot be omitted. This completes the proof of the problem.
Problem 1.8
Rudin Chapter 1 Exercise 8.
e
Actually, Rudin [100, p. 29] assumed this fact in the proof of Theorem 1.38 or the reader may refer to the
comment between Theorem 11.32 and its proof in [99, p. 321].
1.2. Problems related to the Lebesgue’s MCT/DCT 9
Thus we have
lim inf fn (x) = 0
n→∞
for all x ∈ X. However, we have
Z
χ dµ, if n is odd;
Z
X E
fn dµ = Z
X
(1 − χE ) dµ, if n is even.
X
µ(E), if n is odd;
= (1.24)
µ(X \ E), if n is even.
if we assume that µ(E) > 0 and µ(X \E) > 0. Hence the example here shows that the inequality
in Fatou’s lemma can be strict, completing the proof of the problem.f
Problem 1.9
Rudin Chapter 1 Exercise 9.
Z
Proof. If µ(X) = 0, then Proposition 1.24(e) implies that c = f dµ = 0, a contradiction. Let
X
E = {x ∈ X | f (x) = ∞}. We claim that µ(E) = 0. Otherwise, Proposition 1.24(b) implies that
Z Z
c= f dµ ≥ f dµ = ∞
X E
is also measurable on X \ E.
Now we are going to show that when α ≥ 1, there exists a function g ∈ L1 (µ) such that
holds for all n = 1, 2, . . . and all x ∈ X \ E. We first show the following lemma:
Lemma 1.2
For each n ∈ N and α ≥ 1, we have
h x α i
n log 1 + ≤ αx (1.25)
n
on [0, ∞).
F ′ (x) < 0
for all x ∈ [0, ∞). By [99, Theorem 5.11, p. 108] and the continuity of F on [0, ∞), we
establish that F (x) is decreasing on [0, ∞) which implies the validity of the inequality
(1.25).
Let’s return to the proof of the problem. By Lemma 1.2, it is evident that for all n ∈ N and
α ≥ 1, we have
h f α i
fn = n log 1 + ≤ αf (1.27)
n
on X \ E. Furthermore, we apply Proposition 1.24(c) to get
Z Z
|αf | dµ = α f dµ = αc < ∞.
X X
Case (ii): α > 1. In this case, we apply L’Hospital’s rule [99, Theorem 5.13, p. 109] to
conclude that
h f α i log(1 + f α y α ) αf α y α−1 0
lim n log 1 + = lim = lim = =0 (1.28)
n→∞ n y→0+ y y→0+ 1 + f α y α 1+0
on X \ E.
Thus it follows from Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem) that
Z
f dµ, if α = 1;
Z h f α i
X
lim n log 1 + dµ = Z
n→∞ X n
0 dµ, if α > 1,
X
c, if α = 1;
=
0, if α > 1.
It remains the case that 0 < α < 1. In this case, Theorem 1.28 (Fatou’s Lemma) can be
applied directly to get
Z ( n h f α io
) Z h f α i
lim inf n log 1 + dµ ≤ lim inf n log 1 + dµ. (1.29)
X n→∞ n n→∞ X n
1
Since 0 < α < 1, we have y α−1 = y 1−α
so that the limit (1.28) becomes
h f α i αf α
lim n log 1 + = lim 1−α = ∞. (1.30)
n→∞ n y→0+ y (1 + f α y α )
Since lim xn = ∞ if and only if lim sup xn = lim inf xn = ∞, the inequality (1.29) and the limit
n→∞ n→∞ n→∞
(1.30) combine to imply that
n h f α io
lim inf n log 1 + =∞
n→∞ n
and then it certainly gives
Z h f α i
lim n log 1 + dµ = ∞.
n→∞ X n
This completes the proof of the problem.
Problem 1.10
Rudin Chapter 1 Exercise 10.
x n
h
We use the definition ex = lim 1+ here.
n→∞ n
12 Chapter 1. Abstract Integration
Proof. Since fn → f uniformly on X, there exists a positive integer N such that n ≥ N implies
and
|f (x)| ≤ |f (x) − fN (x)| + |fN (x)| ≤ |fN (x)| + 1 (1.32)
for all n ≥ N and x ∈ X. Combining the inequalities (1.31) and (1.32), we see that
Then it is easy to see from the inequality (1.33) and the definition (1.34) that
for x ∈ X and n = 1, 2, . . ..
Next we want to show that g ∈ L1 (µ). By Theorem 1.9(b), |f1 (x)|, . . . , |fN −1 (x)|, |fN (x)|
are measurable. Thus it follows from the corollaries following Theorem 1.14 that g is also
measurable. Furthermore, we know from the hypothesis that |fn (x)| ≤ Mn on X for some
constants Mn , where n = 1, 2, . . . , N . Therefore, this and the definition (1.34) certainly imply
that
g(x) = |g(x)| ≤ M
on X for some constant M . Since it is obvious that g ≥ 0 on X, we get from Proposition 1.24(a)
that Z Z
0≤ |g| dµ ≤ M dµ = M µ(X) < ∞
X X
so that g ∈ L1 (µ).In conclusion, our sequence of functions {fn } satisfies the hypotheses of The-
orem 1.34 (Lebesgue’s Dominated Convergence Theorem) and hence the desired result follows
immediately from this.
For a counterexample, consider X = R and µ = m so that m(R) = ∞. For each n ∈ N,
define fn : R → R by
1
fn (x) = .
n
Then it is easy to prove that fn → f ≡ 0 uniformly on R. Since
Z Z
m(R)
f dm = 0 and fn dm = =∞
R R n
Problem 1.11
Rudin Chapter 1 Exercise 11.
Recall that A is the set of all x ∈ X which lie in infinitely many Ek . On the one hand, if x ∈ A,
\∞
then x ∈ Bn for all n ∈ N so that x ∈ Bn . In other words, we have
n=1
∞
\
A⊆ Bn . (1.35)
n=1
∞
\
On the other hand, if x ∈ Bn , then x ∈ Bn for each positive integer n and this is equivalent
n=1
to the condition that x belongs to infinitely many Ek , i.e.,
∞
\
Bn ⊆ A. (1.36)
n=1
Hence the set relations (1.35) and (1.36) imply the desired result that
∞
\ ∞ [
\ ∞
A= Bn = Ek .
n=1 n=1 k=n
B1 ⊇ B2 ⊇ B3 ⊇ · · · .
Furthermore, our hypothesis and the subadditive property of a measure (see , for example, [105,
Corollary 4.6, p. 26]) show that
[
∞ X∞
µ(B1 ) = µ Ek ≤ µ(Ek ) < ∞.
k=1 k=1
Thus it follows from Theorem 1.19(e) and the subadditive property of µ again that
\
∞ [
∞ ∞
X
0 ≤ µ(A) = µ Bn = lim µ(Bn ) = lim µ Ek ≤ lim µ(Ek ) = 0.
n→∞ n→∞ n→∞
n=1 k=n k=n
Problem 1.12
Rudin Chapter 1 Exercise 12.
14 Chapter 1. Abstract Integration
0 ≤ fn (x) ≤ n (1.38)
if n < |f (x0 )| ≤ n + 1, then we know again from the definition (1.37) that
Thus we can conclude from the computations (1.39), (1.40) and (1.41) that the inequality
holds for all n ∈ N and x ∈ X. By Theorem 1.26 (Lebesgue’s Monotone Convergence Theorem),
we have Z Z
lim fn dµ = F dµ,
n→∞ X X
where F = lim fn . By the definition (1.37), we have F = |f | and so
n→∞
Z Z
lim fn dµ = |f | dµ
n→∞ X X
or equivalently, Z
lim |fn − f | dµ = 0.
n→∞ X
Thus for every ǫ > 0, there exists a positive integer N such that
Z
ǫ
|fn − f | dµ < (1.43)
X 2
for all n ≥ N .
We fix this N . Since 0 ≤ |f | ≤ |f −fN |+|fN | by the triangle inequalityi , we apply Proposition
1.24(a) and then Theorem 1.27 and the property (1.38) to the inequality (1.43) to derive
Z Z Z Z
ǫ ǫ
|f | dµ ≤ |f − fN | dµ + |fN | dµ < + N dµ = + N µ(E) (1.44)
E E E 2 E 2
ǫ
for every E ∈ M. Therefore, if we take δ = 2N and µ(E) < δ, then our inequality (1.44) becomes
Z
|f | dµ < ǫ
E
Problem 1.13
Rudin Chapter 1 Exercise 13.
fn (x) = nf (x).
It is clear that
n α io α i
fn−1 (α, ∞] = {x ∈ X | fn (x) ∈ (α, ∞]} = x ∈ X f (x) ∈ , ∞ = f −1 ,∞
n n
for all real α. Since f is measurable, we have f −1 ( αn , ∞] ∈ M for every real α, where M is a σ-
algebra in X. Thus we obtain from Theorem 1.12(c) that each fn is measurable for n = 1, 2, . . ..
Furthermore, we have
0 ≤ f1 (x) ≤ f2 (x) ≤ · · · ≤ ∞
on X and fn (x) → ∞ · f (x) as n → ∞. By Proposition 1.24(c), we have
Z Z Z
fn dµ = nf dµ = n f dµ (1.45)
X X X
for n = 1, 2, . . .. Hence we conclude from the equality (1.45) and Theorem 1.26 (Lebesgue’s
Monotone Convergence Theorem) that
Z Z Z Z
∞ · f dµ = lim fn dµ = lim n f dµ = ∞ · f dµ.
X n→∞ X n→∞ X X
Thus Proposition 1.24(c) also holds when c = ∞ and so we complete the proof of the problem.
16 Chapter 1. Abstract Integration
CHAPTER 2
Positive Borel Measures
Problem 2.1
Rudin Chapter 2 Exercise 1.
Proof. We have fn : R → [0, ∞) for all n ∈ N. In Proposition 2.5, we will prove a property
which is equivalent to Definition 2.8 and our proof of Statement (a) below becomes simpler.
However, we choose to apply Definition 2.8 to prove the statements here.
then there exist real β1 and β2 with β1 + β2 ≤ α and β1 , β2 > 0 so that x ∈ F1 (β1 )∩ F2 (β2 ).
By the definition (2.1), this x satisfies f1 (x) < β1 and f2 (x) < β2 and their sum implies
17
18 Chapter 2. Positive Borel Measures
On the other hand, if x ∈ E(α), then we let η = f1 (x) + f2 (x). Define the two numbers
β1 and β2 by
η+α η+α
β1 = − f2 (x) and β2 = − f1 (x).
2 2
By direct computation, we know that
f1 (x) + f2 (x) + f1 (x) + f2 (x)
βi > − fi (x) = f1 (x) + f2 (x) − fi (x) ≥ 0,
2
where i = 1, 2. Since η = f1 (x) + f2 (x), it is easy to check that f1 (x) + f2 (x) < α if and
only if [f1 (x) + f2 (x)] + [f1 (x) + f2 (x)] < α + η if and only if f1 (x) + f2 (x) < η+α
2 if and
only if
f1 (x) < β1 . (2.3)
Similarly, the above argument can be used to show that f1 (x) + f2 (x) < α if and only if
Now we deduce from the inequalities (2.3) and (2.4) that x ∈ F1 (β1 )∩F2 (β2 ). Furthermore,
since β1 + β2 = η + α − f1 (x) − f2 (x) = α, we have
[
x∈ [F1 (β1 ) ∩ F2 (β2 )],
β1 +β2 ≤α
β1 ,β2 >0
i.e., [
E(α) ⊆ [F1 (β1 ) ∩ F2 (β2 )].
β1 +β2 ≤α
β1 ,β2 >0
• Statement (b): By a similar argument as in part (a), we can show that the sum of two
lower semicontinuous functions is lower semicontinuous.
• Statement (c): This is not true in general. We use Proposition A.7 to give a counterex-
1
ample. For each n ∈ N, let Fn = [ n+1 , n1 ]. Then each χFn is upper semicontinuous because
Fn is closed in R. Consider the set
n X ∞
1o
E = x ∈ R χFn (x) < .
n=1
2
∞
[
In other words, we have Fn * E for n = 1, 2, . . .. Since Fn = (0, 1],a we see that
n=1
(0, 1] * E. (2.6)
a 1
If x ∈ (0, 1], then there exists a positive integer k such that k+1
< x which implies that x ∈ F1 ∪ F2 ∪ · · · ∪ Fk .
2.1. Properties of Semicontinuity 19
E = (−∞, 0] ∪ (1, ∞)
∞
X
which is not open in R. By Definition 2.8, χFn is not upper semicontinuous.
n=1
for every real α. Similar to the proof of part (a), we suppose that α > 0, E(α) 6= R and
FN (α) 6= R.
Suppose that x ∈ E(α), i.e., f (x) > α. Then there exists a ǫ > 0 such that
f (x) > α + ǫ.
nX
N o
Since each fn is nonnegative, fn is an increasing sequence. By this and the definition
n=1
N ′
X
of f in (2.8), there exists a positive integer N′ such that fn (x) > α + ǫ which implies
n=1
that x ∈ FN ′ (α), i.e.,
∞
[
E(α) ⊆ FN (α). (2.10)
N =1
∞
[
To prove the other side, if x ∈ FN (α), then x ∈ FN ′ (α) for some positive integer
N =1
′
N
X nX
N o
N ′, i.e., fn (x) > α. Again, the fact that fn is an increasing sequence implies
n=1 n=1
that ′
N
X
f (x) ≥ fn (x) > α,
n=1
20 Chapter 2. Positive Borel Measures
Hence the set relations (2.10) and (2.11) definitely imply the claim (2.9) is true and since
each FN (α) is open in R by Definition 2.8, E(α) is also open in R. Since α is arbitrary, f
is lower semicontinuous by Definition 2.8.
In the proof of Statements (a) and (b) above, we don’t use the property that f1 and f2 are
nonnegative. Therefore, they remain valid even if the word “nonnegative” is omitted. However,
Statement (c) cannot hold anymore if the word “nonnegative” is omitted. In fact, we consider
the sequence of real functions {fn } defined by
Problem 2.2
Rudin Chapter 2 Exercise 2.
2.1. Properties of Semicontinuity 21
Proof. We formulate and prove the general statement: Let X be a topological space, U an open
set in X containing the point x and f : X → C. Define
ϕ(p) < α.
This fact shows that there exists an open set U containing p andb
ϕ(p, U ) < α.
Pick q ∈ U \ {p}. Since U is open, there exists an open set V containing q such that
q ∈ V ⊆ U . Since p, q ∈ U , we know from the definition (2.15) that
Now these facts (2.16) and (2.17) imply that ϕ(q, V ) ≤ ϕ(q, U ) = ϕ(p, U ) < α and then
ϕ(x, Uǫ ) ≤ ǫ. (2.19)
By the definition (2.15), we have ϕ(x) ≤ ϕ(x, U ) for every open set U containing x.
Therefore, we deduce from this and the inequality (2.19) that ϕ(x) ≤ ǫ. Since ǫ is arbitrary,
we have the desired result that ϕ(x) = 0.
b
Otherwise, we have ϕ(p, U ) ≥ α for every open set U containing p and this implies that α ≤ ϕ(p), a
contradiction.
c
Note that |f (s) − f (t)| ≤ |f (s) − f (x)| + |f (x) − f (t)| < ǫ for every s, t ∈ Uǫ .
22 Chapter 2. Positive Borel Measures
Conversely, suppose that ϕ(x) = 0. Thus for every ǫ > 0, there exists a neighborhood
Uǫ of x such that ϕ(x, Uǫ ) < ǫ. By the definition (2.15), this implies that
|f (s) − f (t)| < ǫ (2.20)
for every s, t ∈ Uǫ . In particular, if we take s = x to be fixed and t = y vary, then the
inequality (2.20) can be rewritten as |f (x) − f (y)| < ǫ for every y ∈ Uǫ . By the definition
([74, Theorem 18.1, p. 104] or [100, p. 9]), f is continuous at x.
• The set of points of continuity of an arbitrary complex function is a Gδ . By the
previous analysis, we establish that
∞ n
\ 1o
G = {x ∈ X | f is continuous at x} = {x ∈ X | ϕ(x) = 0} = x ∈ X ϕ(x) < .
n
n=1
1
Since ϕ is upper semicontinuous, each set {x ∈ X | ϕ(x) < n} is open in X. By Definition
1.11, we see that G is actually a Gδ .
Problem 2.3
Rudin Chapter 2 Exercise 3.
Proof. The first part is proven in [124, Problem 4.20, pp. 73, 74]. The function in the question
can be used to prove Urysohn’s Lemma for any metric space X directly.
Proof of Lemma 2.1. This lemma is proven in [124, Problem 4.22, pp. 75, 76]. Readers
are recommended to read [74, Theorem 32.2, p. 202; Theorem 33.1, pp. 207 – 210].
Problem 2.4
Rudin Chapter 2 Exercise 4.
Proof.
(a) Recall from [100, Eqn. (2), p. 42] that
µ(E) = inf{µ(V ) | E ⊆ V and V is open} (2.21)
which is defined for every subset E of X.d By the result [100, Eqn. (4), p. 42], we have
µ(E1 ∪ E2 ) ≤ µ(E1 ) + µ(E2 ). (2.22)
d
Here we don’t require E to be a measurable set.
2.1. Properties of Semicontinuity 23
We need a lemma:
Lemma 2.2
For disjoint open sets V1 and V2 , we have
Proof of Lemma 2.2. The proof of the inequality µ(V1 ∪ V2 ) ≤ µ(V1 ) + µ(V2 ) can be
found in [100, STEP I, p. 42], so we only prove the other side. Suppose that g ∈ Cc (X)
and g ≺ V1 . Similarly, suppose that h ∈ Cc (X) and h ≺ V2 . By Definition 2.9, since
Cc (X) is a vector space, f = g + h ∈ Cc (X). Furthermore, we know from Definition
2.9(a) that
By assumption, we have V1 ∩V2 = ∅ which means that g(x) = 0 on X \V1 and h(x) = 0
on X \ V2 . Thus we deduce from these facts that
g(x), if x ∈ V1 ;
f (x) = h(x), if x ∈ V2 ;
0, if x ∈ X \ (V1 ∪ V2 ).
Next, the inequality (2.27) holds for every h ≺ V2 , we establish from the definition of
supremum that
µ(V1 ) + µ(V2 ) ≤ µ(V1 ∪ V2 ).
Hence we have µ(V1 ) + µ(V2 ) = µ(V1 ∪ V2 ).
Let’s go back to the original proof. We want to compute µ(E1 ∪ E2 ). By the definition
(2.21), it needs to consider all open sets containing E1 ∪ E2 . Since V1 ⊆ V2 certainly
implies µ(V1 ) ≤ µ(V2 ), we can restrict our attention to any open set W such that
E1 ∪ E2 ⊆ W ⊆ V1 ∪ V2 .
Furthermore, since W ∩ V1 and W ∩ V2 are open sets in X, it yields from Lemma 2.2 that
Since W is arbitrary, µ(E1 ) + µ(E2 ) is a lower bound of the set in the definition (2.21).
Hence we follow from this fact and the inequality (2.28) that
By the inequalities (2.22) and (2.29), we have the desired result that
(b) Define E1 = E. By [100, STEP V, p. 44], there exists a compact set K1 and an open set
V1 such that
1
K1 ⊆ E1 ⊆ V1 and µ(V1 \ K1 ) < 2 .
2
By [100, STEP II, p. 43], we have K1 ∈ MF . Then we follow from [100, STEP VI, p.
44] that E1 \ K1 ∈ MF . Define E2 = E1 \ K1 so that
E = E1 = E2 ∪ K1 ,
where µ(E2 ) = µ(E1 \ K1 ) ≤ µ(V1 \ K1 ) < 212 . By induction, we can show that there exists
a sequence of compact sets {Kn } and a sequence of open sets {Vn } such that
Kn ⊆ En ⊆ Vn , (2.30)
1
where En+1 = En \ Kn , En+1 ∈ MF and µ(En+1 ) < 2n+1
. Therefore, the set E satisfies
the following relations:
1
E = En+1 ∪ K1 ∪ K2 ∪ · · · ∪ Kn and µ(En+1 ) < (2.31)
2n+1
for every positive integer n. By the set relations (2.30), we deduce that
for all n ∈ N. In other words, Kn+1 is disjoint with all K1 , K2 , . . . , Kn and thus {K1 , . . . , Kn }
is a disjoint collection of compact sets in X. Now, by letting n → ∞ in the relations (2.31),
we achieve that E is given by
E = N ∪ K1 ∪ K2 ∪ · · · ,
By the equalities in (2.32), N ⊆ En+1 holds for all n ∈ N and the inequality in (2.31)
shows that
1
µ(N ) = lim µ(En+1 ) = lim n+1 = 0.
n→∞ n→∞ 2
Problem 2.6
Rudin Chapter 2 Exercise 6.
Proof. Recall from [99, Theorem 2.47, p. 42] that a subset F of R is connected if and only if
(x, y) ⊆ F for every x, y ∈ F with x < y. Therefore, if F is a connected subset of a totally
disconnected set K ⊂ R and F consists of more than one point, then K must contain a segment,
a contradiction. By this observation, we know that a totally disconnected set K ⊂ R contains
no segment and this gives a hint to our construction: It is well-known that the Cantor set E is
compact and contains no segment. However, Problem 2.5 shows that m(E) = 0, so we have to
“modify” the construction of E so as to fit our problem.
1
Let K0 = [0, 1]. The idea of the construction is to remove the “middle 22n
th” segments. In
fact, we first remove the segment ( 38 , 58 ) of length 212 from K0 and let
h 3i h5 8i
K1 = 0, ∪ , .
8 8 8
5 7
Next, we remove the middle segments ( 32 , 32 ) and ( 25 27
32 , 32 ) each of length
1
24
from each connected
component of K1 to get
h 5 i h 7 12 i h 20 25 i h 27 32 i
K2 = 0, ∪ , ∪ , ∪ , .
32 32 32 32 32 32 32
Continuing in this way, we obtain a sequence of compact sets Kn such that
K1 ⊃ K2 ⊃ K3 ⊃ · · · .
Therefore, we have
n
X n
X
1 1
m(Kn ) = 1 − 2k
· 2k−1 = 1 − . (2.33)
2 2k+1
k=1 k=1
We know that K 6= ∅ because of the corollary [99, Theorem 2.36, p. 38]. Applying an argument
similar to that used in [99, §2.44, pp. 41, 42], we can show that K is a totally disconnected
compact set in R. Furthermore, since m(K1 ) is finite, Theorem 1.19(e) and the Lebesgue measure
(2.33) ensure that
n
X 1
∞
X 1 1
m(K) = lim m(Kn ) = lim 1− =1− = > 0.
n→∞ n→∞ 2k+1 2k+1 2
k=1 k=1
26 Chapter 2. Positive Borel Measures
For the second assertion, let v be lower semicontinuous and v ≤ χK . Assume that there was
a p ∈ R such that v(p) > 0. Since v is lower semicontinuous, Definition 2.8 implies that the set
A = {x ∈ R | v(x) > 0} is open in R. By the assumption, we have p ∈ A, so there exists a δ > 0
such that
(p − δ, p + δ) ⊆ A. (2.34)
Now we follow from the hypothesis v ≤ χK and the relation (2.34) that
χK (x) > 0
on (p − δ, p + δ), therefore we have (p − δ, p + δ) ∈ K which contradicts to the fact that K
contains no segment. Hence we have v ≤ 0 and this completes the proof of the problem.
Problem 2.7
Rudin Chapter 2 Exercise 7.
Proof. By the idea used in Problem 2.6, it is not hard to see that the construction of the compact
ǫ
sets Kn and K still work for removing the “middle 22n−1 th” segments. In this case, instead of
the Lebesgue measure (2.33), we have
n
X n
X
ǫ 1
m(Kn ) = 1 − · 2k−1 = 1 − ǫ (2.35)
22k−1 2k
k=1 k=1
and so
m(K) = lim m(Kn ) = 1 − ǫ.
n→∞
Since K is closed in [0, 1], the complement
E = [0, 1] \ K (2.36)
is definitely open in [0, 1].
Now it remains to show that E is dense in [0, 1]. We prove an equivalent definition of a dense
set: Let A ⊆ B ⊆ R. We say A is dense in B means that for every x, y ∈ B with x < y, we
can find z ∈ A such that x < z < y. In particular, suppose that x, y ∈ [0, 1] with x < y. If
(x, y) ∩ E = ∅, then the definition (2.36) says that (x, y) ⊆ K which is impossible. Therefore,
we have
(x, y) ∩ E 6= ∅
and this means that there exists z ∈ E such that x < z < y. Hence E is dense in [0, 1],
completing the proof of the problem.
Problem 2.8
Rudin Chapter 2 Exercise 8.
Proof. Let {rn } be the enumeration of all rationals in R. Let Fn be the segment given by
1 1
Fn = rn − 2n , rn + 2n , (2.37)
2 2
where n = 1, 2, . . .. Define the sets
∞
[
Gn = Fn \ (Fn+1 ∪ Fn+2 ∪ · · · ) = Fn \ Fn+k . (2.38)
k=1
m(Fn )
• Step 1: m(Gn ) > 0. By the definition (2.37), we know that m(Fn+1 ) = 22
for every
n = 1, 2, . . .. By the definition (2.38), for every n ∈ N, we have
∞
[
Fn = Gn ∪ Fn+k
k=1
so that the subadditive property of a measure (see the proof of Problem 1.11) implies that
∞
X ∞
X 1 m(Fn )
m(Fn ) ≤ m(Gn ) + m(Fn+k ) = m(Gn ) + m(Fn ) = m(Gn ) + .
22k 3
k=1 k=1
Therefore, we have
2m(Fn )
m(Gn ) ≥ >0 (2.39)
3
for every positive integer n.
n)
• Step 2: Existence of a Borel subset An ⊂ Fn with m(An ) = m(F 2 . It is easy to
1
check from the definition (2.37) that m(Fn ) = 22n−1 . By Definition 2.19, we know thate
1 h 1 i
Fn′ = Fn − rn − 2n = (0, 1).
m(Fn ) 2
Thus, by the proof of Problem 2.7f , there exists a (totally disconnected) compact set
A′n ⊂ Fn′ such that m(A′n ) = 21 . Then Theorem 2.20(c) implies that the set An given by
1
An = m(Fn )A′n + rn − 2n
2
is a (totally disconnected) compact subset of Fn with measure
1 m(Fn )
m(An ) = m m(Fn )A′n + rn − 2n = m m(Fn )A′n = m(Fn )m(A′n ) = . (2.40)
2 2
Since An is compact, it is closed by the Heine–Borel theorem [99, Theorem 2.41, p. 40].
By Definition 1.11, it is also a Borel set, completing the proof of Step 2.
Now we can construct a Borel set with the desired properties. The construction is as
follows: For every positive integer n, we define
∞
[
En = Gn ∩ An and E = En . (2.41)
n=1
By the definition (2.38), since each Fn is Borel, each Gn is also Borel. Thus each En and
their countable union E are also Borel sets. It remains to show that the E satisfies the
requirements of the problem.
• Step 3: 0 < m(E ∩ Fn ). We first show that m(En ∩ Fn ) > 0 for every positive integer n.
To prove this claim, we note from the definitions (2.38) and (2.41) that En ⊆ Gn ⊆ Fn , so
we have
m(En ∩ Fn ) = m(En ) = m(Gn ∩ An ).
Furthermore, since Gn ⊆ Fn and An ⊂ Fn (by Step 2), we have
Combining the inequality (2.39) and the expression (2.40), we can reduce the inequality
(2.42) to
m(Gn ∪ An ) < m(Gn ) + m(An ). (2.43)
Now we apply a property of additive function [99, Eqn. (7), p. 302] to the inequality
(2.43) to obtain
m(Gn ∪ An ) < m(Gn ∪ An ) + m(Gn ∩ An )
which means that m(Gn ∩ An ) > 0 for every positive integer n. Therefore, we follow from
this and the fact that En ∩ Fn ⊆ E ∩ Fn
• Step 4: m(E ∩ Fn ) < m(Fn ). Fix an integer n. For m < n, we apply the identity
A \ B = B c ∩ A to the definition (2.38) to get
which implies that Fn ⊆ Gcm for m = 1, 2, . . . , n − 1. Recall from the definition (2.41) that
Enc = Gcn ∪ Acn , so we must have
c
Fn ⊆ Em
for m = 1, 2, . . . , n − 1. In other words, it says that Fn ∩ Em = ∅ for m = 1, 2, . . . , n − 1.
Therefore, this fact implies
[
∞ [
∞
[
∞
m(E ∩ Fn ) = m Em ∩ Fn = m Em ∩ Fn ≤ m Am ∩ Fn . (2.44)
m=1 m=n m=n
By the subadditive property of a measure again, we further deduce the inequality (2.44)
to
X∞ X∞
m(E ∩ Fn ) ≤ m(Am ∩ Fn ) ≤ m(Am ). (2.45)
m=n m=n
m(Fn )
Recall the result (2.40) and the fact m(Fn+1 ) = 22
, so we derive from the inequality
(2.45) that
X∞ ∞
m(Fn+m ) m(Fn ) X 1 2m(Fn )
m(E ∩ Fn ) ≤ = 2m
= < m(Fn )
m=0
2 2 m=0 2 3
• Step 5: 0 < m(E ∩ I) < m(I) for every nonempty segment I. By Step 3 and Step
4, the inequalities
0 < m(E ∩ Fn ) < m(Fn ) (2.46)
hold for every positive integer n, where Fn and E are given by (2.37) and (2.41) respectively.
Lemma 2.3
Let I = (α, β) with α < β. Then there is a positive integer n0 such that Fn0 ⊆ I.
2.2. Problems on the Lebesgue Measure on R 29
We return to the proof of the problem. By Lemma 2.3, we have E ∩ Fn0 ⊆ E ∩ I and the
left-hand side of the inequalities (2.46) shows that
Hence, by applying the right-hand side of the inequalities (2.46) and Theorem 1.19(b)
twice, we obtain
m(E ∩ I) = m(E ∩ Fn0 ) + m E ∩ (I \ Fn0 ) < m(Fn0 ) + m(I \ Fn0 ) = m(I). (2.52)
Now our desired result follows immediately if we combine the inequalities (2.51) and (2.52).
• Step 6: m(E) < ∞. By the subadditive property of the measure m, the expression (2.40)
and the definition (2.41), we get
∞
X ∞
X ∞ ∞
1X m(F1 ) X 1 2m(F1 )
m(E) ≤ m(En ) ≤ m(An ) = m(Fn ) = 2n
= < ∞.
2 2 2 3
n=1 n=1 n=1 n=0
Problem 2.9
Rudin Chapter 2 Exercise 9.
It is clear that each gn is continuous on [−1, 1] and 0 ≤ gn ≤ 1 on [−1, 1]. The graph of gn is
shown in Figure 2.1 below.
for every positive integer n and k = 0, 1, . . . , 2n. In fact, we know from the definitions (2.53)
and (2.54) that
1 1
2 + nx, if x ∈ [0, 2n );
1
1 − n|x|, if x ∈ [0, n );
3 1 3
gn,0 (x) = and gn,1 (x) = 2 − nx, if x ∈ [ 2n , 2n ]; (2.56)
0, if x ∈ [ n1 , 1]
3
0, if x ∈ ( 2n , 1].
Similarly, we have
0,
if x ∈ [0, 2n−3
2n );
gn,2n−1 (x) = (2.57)
1 − nx − 2n−1
2n , if x ∈ [ 2n−3
2n , 1]
and
0, if x ∈ [0, n−1
n );
gn,2n (x) = (2.58)
1 + n(x − 1), if x ∈ [ n−1
n , 1].
Finally, for k = 2, 3, . . . , 2n − 2, we have
0, if x ∈ [0, k−2
2n );
k
gn,k (x) = 1 − nx − 2n , if x ∈ [ k−2 k+2
2n , 2n ]; (2.59)
0, if x ∈ ( k+2
2n , 1].
In other words, it is clear from Figure 2.2 that {gn,0 , gn,1 . . . , gn,2n } is a family of tent functions
0 1
centered at { 2n , 2n , . . . , 2n
2n }. Now we have the following cases.
h
h
We use the fact that each gn,k is a part or the whole of an isosceles triangle with height 1 and base less than
or equal to n2 .
32 Chapter 2. Positive Borel Measures
• When k = 0, we have
Z 1 Z 1 Z 1
n 1
gn,0 (x) dx = gn (x) dx = (1 − nx) dx = . (2.60)
0 0 0 2n
• When k = 1, we have
Z 1 Z 1 1 Z 3 3
2n 2n 7
gn,1 (x) dx = + nx dx + − nx dx = . (2.61)
0 0 2 1 2 8n
2n
• When 2 ≤ k ≤ 2n − 2, we have
Z 1
1 2 1
gn,k (x) dx = × ×1= . (2.62)
0 2 n n
• When k = 2n − 1, we have
Z 1 Z
1
2n − 1 7
gn,2n−1 (x) dx = 1 − nx − dx = . (2.63)
0 2n−3 2n 8n
2n
Combining the expressions (2.60) to (2.64), we may conclude that the limit (2.55) holds.
Now it remains to show that {fn (x)} converges for no x ∈ [0, 1]. If x = 0, then we obtain
from the definitions (2.56) and (2.57) that
In other words, we can find subsequences {fnk (0)} and {fnl (0)} such that fnk (0) → 1 and
fnl (0) → 0 as k → ∞ and l → ∞ respectively. By Definition 1.13, they imply that
for all n ≥ N . By the density of rationals, there exists a sequence { pqnn } of rationals in (0, 1)
such that
pn
→x
qn
2.3. Integration of Sequences of Continuous Functions 33
for all n ∈ N. Combining the two results (2.67) and (2.68), we have shown that the limits (2.65)
remain valid for irrationals x, completing the proof of the problem.
Problem 2.10
Rudin Chapter 2 Exercise 10.
En ⊇ En+1 ⊇ · · · .
If En0 = ∅ for some positive integer n0 , then we know from the definition (2.69) that
fk (x) ≤ ǫ
for every x ∈ [0, 1] and every positive integer k ≥ n0 . Therefore it implies that
Z 1
0≤ fk (x) dx ≤ ǫ
0
for every positive integer k ≥ n0 . Since ǫ is arbitrary, we obtain from this that
Z 1
lim fn (x) dx = 0.
n→∞ 0
Without loss of generality, we assume that En 6= ∅ for all n ∈ N. Let’s prove two properties of
En first:
• Property 1: En is open. Let p ∈ En . Then we have fk (p) > ǫ for some k ≥ n. Define
ǫ′ = 12 (fk (p) − ǫ) > 0. Since fk is continuous on [0, 1], it is continuous at p. Thus for this
particular ǫ′ , there is a δ > 0 such that
for all points x ∈ [0, 1] with |x − p| < δ. Now the inequality (2.70) implies that
fk (p) + ǫ
fk (x) > fk (p) − ǫ′ = > ǫ.
2
In other words, we have (p − δ, p + δ) ⊆ En as desired.
34 Chapter 2. Positive Borel Measures
∞
\
• Property 2: En = ∅. Otherwise, there was a p ∈ [0, 1] such that p ∈ En for all
n=1
n ∈ N. Therefore, for each positive integer n, we have
a contradiction.
To finish our proof, we have to study the lengths of certain subsets of En . Let F be a finite
union of bounded (open or closed) intervals of [0, 1]. Then we may write
m
[ s
[
F = [ak , bk ] ∪ (ar , br ) (2.72)
k=1 r=1
m
X s m s
X X X
ℓ(F ) = ℓ [ak , bk ] + ℓ (ar , br ) = (bk − ak ) + (br − ar ).
k=1 r=1 k=1 r=1
Given a nonempty finite union of bounded interval F expressed in the form (2.72) and assume
ǫ
that F has at least one open interval. In addition, we suppose that ℓ(F ) > ǫ. Let δ = 4s . Then
the set
m
[ s
[
G= [ak , bk ] ∪ [ar + δ, br − δ]
k=1 r=1
is clearly a nonempty finite union of bounded and closed intervals of [0, 1]. Now it is easy to see
that
Xm Xs
ǫ
ℓ(G) = (bk − ak ) + (br − ar − 2δ) = ℓ(F ) − 2sδ = ℓ(F ) −
r=1
2
k=1
implying that
ℓ(G) > ℓ(F ) − ǫ.
Lemma 2.4
Let Sn = {F | F is a finite union of bounded intervals of En } for every positive in-
teger n and
Ln = sup{ℓ(F ) | F ∈ Sn }. (2.73)
Then we have
lim Ln = 0.
n→∞
2.3. Integration of Sequences of Continuous Functions 35
ℓ(F ) ≥ δ (2.75)
for each n ∈ N. Otherwise, there is a N ∈ N such that ℓ(F ) < δ for all F ∈ SN
which means that δ is an upper bound of LN , but this is a contradiction. Now
we are going to show that if Kn = ∅, then it is impossible to have the inequality
(2.75). This contradiction bases on the following two facts:
(F \ F1 ) ∪ · · · ∪ (F \ Fn ) = F \ (F1 ∩ · · · ∩ Fn ) = F \ Kn = F. (2.77)
Hence the limit must be 0 which completes the proof of the lemma.
36 Chapter 2. Positive Borel Measures
We return to the proof of the problem. By Lemma 2.4, for ǫ > 0, there is a positive integer
N such that for n ≥ N , we have
ℓ(F ) ≤ Ln < ǫ, (2.80)
where F is any finite union of bounded intervals of En . Now one may think that we can derive
Z 1
fn (x) dx < 2ǫ
0
for n ≥ N directly from the uniform boundedness of fn , the definition (2.69) and the inequality
(2.80). However, it fails because we have no way to estimate the integral
Z
fn (x) dx.
En \F
To overcome this problem, we play the trick that since fn is Riemann integrable on [0, 1], the
integral must be equal to its lower Riemann integral, see [4, §1.17, p. 74]. Thus we have to find
nZ 1 o
sup sn (x) dx 0 ≤ sn (x) ≤ fn (x) and sn is a step function ,
0
where the sup is taken over all step functions sn below fn on [0, 1].
For every n ≥ N , we define
Since sn is a step function, it is clear that En and Fn are finite unions of bounded intervals.
Furthermore, since 0 ≤ sn (x) ≤ fn (x), we have En ⊆ En and En ∈ Sn . Then we deduce from the
inequality (2.80) that
ℓ(En ) < ǫ
for all n ≥ N . Hence we obtain from this and repeated uses of [99, Theorem 6.12(c), p. 128]
that for all n ≥ N , we have
Z 1 Z Z Z Z
0≤ sn (x) dx = sn (x) dx + sn (x) dx ≤ dx + ǫ dx ≤ ℓ(En ) + ǫ < 2ǫ.
0 En Fn En Fn
Remark 2.1
There are many mathematicians who have provided different proofs of Problem 2.10. See,
for examples, [64], [66] and [106].
Problem 2.11
Rudin Chapter 2 Exercise 11.
2.4. Problems on Borel Measures and Lebesgue Measures 37
Proof. We follow the given hint. Suppose that Ω is the family of all compact subsets Kα of X
with µ(Kα ) = 1. Since X is compact and µ(X) = 1, Ω is not empty. Now we define
\
K= Kα . (2.81)
Kα ∈Ω
Since X is Hausdorff, each Kα is closed in X by the corollaries following Theorem 2.5. Thus K
is closed in X and then K is Borel (see Definition 1.11), i.e., K ∈ M. Since K ⊆ X, Theorem
2.4 says that K is also compact.
Suppose that K ⊆ V and V is open in X. Then V c is closed in X and so it is compact in X
by Theorem 2.4. As we have mentioned in the previous paragraph that each Kα is closed in X,
so each Kαc is open in X. By the definition (2.81) and the fact that K ⊆ V , we have
[
V c ⊆ Kc = Kαc .
Kα ∈Ω
In other words, {Kαc } forms an open cover of the compact set V c . Thus we have
for some positive integer n. Since µ(Kαm ) = µ(X) = 1 for m = 1, 2, . . . , n, we have µ(Kαc m ) = 0
for m = 1, 2, . . . , n. Therefore, it follows from these and the set relation (2.82) that
n
X
µ(V c ) ≤ µ(Kαc m ) = 0,
m=1
i.e., µ(V c) = 0 or equivalent µ(V ) = 1. Since µ is regular and K ∈ M, Definition 2.15 implies
that
µ(K) = inf{µ(V ) | K ⊆ V and V is open} = 1
which proves the first assertion.
For the second assertion, let H be a proper compact subset of K, i.e., H ⊂ K. Assume
that µ(H) = 1. Then it means that H ∈ Ω and the definition (2.81) shows that K ⊆ H, a
contradiction. Hence we must have µ(H) < 1. This completes the proof of the problem.
Problem 2.12
Rudin Chapter 2 Exercise 12.
Proof. Let B be the σ-algebra of all Borel sets in R and K be a nonempty compact subset of R.
We have to show that there exists a measure µ on B such that K is the smallest closed subset
with the property µ(K c ) = 0.
Since every compact set K has a countable base (see [99, Exercise 25, Chap. 2, p. 45]), there
exists a countable sequence F = {x1 , x2 , . . .} ⊆ K such that
K = F.
where χ{xn } is the characteristic function of the set {xn }. Suppose that {Ei } is a disjoint
countable collections of members of B and
∞
[
E= Ei .
i=1
where the summation runs through all n such that xn ∈ Ei for some i. Since the series
∞
X 1
2n
n=1
converges absolutely, its rearrangement converges to the same limit (see [99, Theorem 3.55, p.
78]). Therefore, we can rewrite the expression (2.84) as
∞ X
X 1
µ(E) = χ (Ei ), (2.85)
2ik {xik }
i=1 ik
where the inner summation runs through all ik such that xik ∈ Ei . It may happen that the
set F ∩ Ei is finite for some i = 1, 2, . . .. Since the inner summation in the expression (2.85) is
exactly µ(Ei ), we deduce from the expression (2.85) that
[
∞ ∞
X
µ Ei = µ(E) = µ(Ei ).
i=1 i=1
so that µ(K c ) = 0. Assume that there was a proper closed subset H ⊂ K in X such that
µ(H c ) = 0. Since H ⊂ K, we have K c ⊂ H c . If H c ∩ K = ∅, then H c ⊆ K c , a contradiction.
Thus H c ∩ K 6= ∅. Since H c is open in X, we have H c ∈ B. Therefore, we can conclude from
the condition (2.86) that
µ(H c ) > 0
which contradicts to our hypothesis. In other words, K is the smallest closed subset in R such
that µ(K c ) = 0, or equivalently,
K = supp (µ).
This completes the analysis of the problem.
Problem 2.13
Rudin Chapter 2 Exercise 13.
i
However, Ei may contain more than one element of F .
2.4. Problems on Borel Measures and Lebesgue Measures 39
Proof. It is clear that the point set {0} is a compact subset of R. Assume f : R → R was
a continuous function such that {0} = supp (f ). Let V = (−∞, 0) ∪ (0, ∞). We know from
Definition 2.9 that supp (f ) = f −1 (V ), so
{0} = f −1 (V ) (2.87)
(p − δ, p + δ) ⊆ f −1 (V ). (2.88)
Thus we deduce from the set relations (2.87) and (2.88) that
[p − δ, p + δ] ⊆ f −1 (V ) = {0},
ρF (x) = inf{|x − y| | y ∈ F }.
By Problem 2.3, we know that ρF is uniformly continuous on R and ρF (x) = 0 if and only if
x ∈ F . Therefore, we must have
K = {x ∈ R | ρF (x) 6= 0},
Problem 2.14
Rudin Chapter 2 Exercise 14.
40 Chapter 2. Positive Borel Measures
Proof. Let f : Rk → R. We follow the proof of Theorem 2.24 (Lusin’s Theorem). Suppose that
0 ≤ f < 1. Attach a sequence {sn } of simple measurable functions to f , as in the proof of
Theorem 1.17 (The Simple Function Approximation Theorem). Put t1 = s1 and tn = sn − sn−1
for n = 2, 3, . . .. Then 2n tn is the characteristic function of a measurable set Tn ⊆ Rk and
∞
X ∞
X
f (x) = tn (x) = 2−n χTn (x) (2.91)
n=1 n=1
We claim that g and h are Borel measurable functions on Rk . It is easy to check that
∅, if α ≤ 0;
{x ∈ Rk | χAn (x) < α} = Acn , if 0 < α ≤ 1;
R, if α > 1.
Thus each χAn is a Borel measurable function by Definition 1.11. Similarly, each χBn is also a
Borel measurable function. Since the partial sums gn and hn of g and h are just simple functions
and An , Bn ∈ B, they are Borel measurable functions by the comment following Definition 1.16.
Furthermore, for each n = 1, 2, . . ., we have
X∞
on Rk . Since 2−n converges, it follows from Weierstrass M -test (see [99, Theorem 7.10, p.
Xn=1 X
147]) that 2−n χAn and 2−n χBn converge uniformly on Rk to g and h respectively. By
the Corollary (a) following Theorem 1.14, g and h are Borel measurable. This proves our claim.
By the measure (2.92), we know that g = h a.e. [m] on Rk . Obviously, if F ⊆ E, then we
must have χF (x) ≤ χE (x). Therefore, we observe from the set relations (2.92) that
on Rk . Hence we apply the inequalities (2.94) to functions (2.91) and (2.93), we obtain
on Rk .
Next, if 0 ≤ f < M for some positive constant M on Rk , then the above argument can
f
be applied to M so that the inequalities (2.95) hold with g and h replaced by M g and M h
respectively.
j
In fact, each An is an Fσ and each Bn is a Gδ , see Definition 1.11.
2.4. Problems on Borel Measures and Lebesgue Measures 41
g(x) = lim sup gN (x) and h(x) = lim sup hN (x) (2.96)
N →∞ N →∞
are Borel measurable functions on Rk . Since gN (x) = hN (x) a.e. [m] on Rk , we have g(x) = h(x)
a.e. [m] on Rk . Furthermore, we note that
g(x) = lim sup gN (x) ≤ lim sup χEN f (x) ≤ lim sup hN (x) = h(x) (2.97)
N →∞ N →∞ N →∞
on Rk . Since
lim χEN f (x) = lim sup χEN f (x) = f (x)
N →∞ N →∞
on Rk ,
the inequalities (2.95) follow immediately from the inequalities (2.96) and the fact (2.97).
Hence our desired results also hold in this general case, completing the proof of the problem.
Problem 2.15
Rudin Chapter 2 Exercise 15.
It is easy to see that each fn is continuous on [0, ∞) and so Definition 1.11 shows that it is Borel
measurable. Fix a x ∈ R. Then the Archimedean Property ensures the existence of a positive
integer N such that N ≥ x. By this fact and the fact that (1 + nx )n → ex as n → ∞, we have
x n x x x
lim fn (x) = lim 1 − e 2 = e−x · e 2 = e− 2 .
n→∞ n→∞ n
x
Since e− 2 ∈ L1 (R) and
x
|fn (x)| ≤ e− 2 (2.98)
for all n = 1, 2, . . . and x ∈ [0, ∞), it follows from Theorem 1.34 (Lebesgue’s Dominated Con-
vergence Theorem) that Z ∞ Z ∞
x
lim fn dx = e− 2 dx = 2. (2.99)
n→∞ 0 0
By the inequality (2.98), we get
Z ∞ Z n Z ∞ Z ∞ Z ∞
x n
fn dx − fn dx = fn dx ≤ |fn | dx ≤ e− 2 dx = 2e− 2
0 0 n n n
Now we apply the same trick as obtaining the result (2.100), we are able to show that
Z n
lim gn dx = 1.
n→∞ 0
Problem 2.16
Rudin Chapter 2 Exercise 16.
T (α1 b1 + · · · + αk bk ) = (α1 , . . . , αk ).
T (Y ) = {(α1 , . . . , αr , 0, . . . , 0) | α1 , . . . , αr ∈ R}
so that
∞
[
T (Y ) = Wn , (2.101)
n=1
Wn ⊆ Wn+1
k
See, for example, [42, Theorem 30.19, p. 279]. Sometimes, this result is called the Basis Extension
Theorem.
2.5. Problems on Regularity of Borel Measures 43
for each n = 1, 2, . . .. In addition, since Rk is a Borel set, Y is also a Borel set. Hence we follow
from this fact, the expressions (2.101), (2.102) and Theorem 1.19(d) that
m T (Y ) = lim m(Wn ) = lim vol (Wn ) = 0. (2.103)
n→∞ n→∞
When we read the proof of Theorem 2.20(e) and §2.23 carefully, we see that the validity of
the formula
m T (E) = | det T |m(E) (2.104)
when T is an one-to-one map of Rk onto Rk is independent of whether T (Rk ) is a subspace of
lower dimension or not. As a consequence, we may apply the formula (2.104) here. Hence we
can conclude from m(Y ) = | det T |−1 m T (Y ) and the result (2.103) that m(Y ) = 0, completing
the proof of the problem.
Problem 2.17
Rudin Chapter 2 Exercise 17.
Proof. Let d be the mentioned distance in the problem, p = (x1 , y1 ) and q = (x2 , y2 ). We are
going to prove the assertions one by one.
• X is a metric space with metric d. We check the definition of a metric space, see [99,
Definition 2.15, p. 30]:
Similarly, we have
|y2 − y3 |, if x3 = x2 and y3 6= y2 ;
d(q, r) = (2.106)
1 + |y2 − y3 |, if x3 6= x2 .
Geometrically, B(p, 1) is the vertical line segment with half length less than 1 and (x, y)
as its midpoint. In addition, this line segment is open in R with the usual metric. Fur-
thermore, it is clear from the definition (2.108) that
We want to show that B(p, 1) is compact with respect to the metric d. To this end, let
{Vα } be an open cover of B(p, 1), i.e.,
[
B(p, 1) ⊆ Vα .
α
Let a ∈ B(p, 1), where a = (s, t) with s = x and t ∈ [y − 1, y + 1]. Then (x, t) ∈ Vα for
some α. Since Vα is open in X, there exists a δt ∈ (0, 1) such that
so that [ [
B(p, 1) ⊆ B(a, δt ) ⊆ Vα . (2.111)
a∈B(p,1) α
a=(s,t)
Applying the expressions (2.109) and (2.110) to the set relation (2.111), we see that
[ [
B(p, 1) = {x} × [y − 1, y + 1] ⊆ {x} × (t − δt , t + δt ) ⊆ Vα (2.112)
t∈[y−1,y+1] α
Suppose that (x, ti ) ∈ Vαi for i = 1, 2, . . . , k. Hence we follow immediately from the set
relation (2.112) that
B(p, 1) ⊆ Vt1 ∪ Vt2 ∪ · · · ∪ Vtk .
In other words, the closure B(p, 1) is compact with respect to the metric d and since p is
an arbitrary point in X, Definition 2.3(f) shows that X is locally compact.l
l
We remark that we cannot use the Heine-Borel Theorem directly to the closure (2.109) because we are
working in the space X with metric d, not R2 with the usual metric.
2.5. Problems on Regularity of Borel Measures 45
where B(p, δ) = {x} × (y − δ, y + δ). Now the compactness of K ensures that there exists
some positive integer n such that
n
[ n
[
K⊆ B(pi , δ) = {q ∈ X | u = xi and |yi − v| < δ}
i=1 i=1
is a positive linear functional. To this end, let f, g ∈ Cc (X). Since Cc (X) is a vector space,
we have αf + βg ∈ Cc (X) for any α, β ∈ C. Furthermore, let f (xi , y) 6= 0 and g(xj , y) 6= 0
for at least one y, where 1 ≤ i ≤ n and n + 1 ≤ j ≤ m, and f (zk , y)g(zk , y) 6= 0 for at least
one y, where m + 1 ≤ k ≤ r. By the definition, we have
f (xi , y) = g(xj , y) = 0,
• The measures of the x-axis and its compact subsets. In order to apply Theorem
2.14 (The Riesz Representation Theorem), we have to show that X is Hausdorff. Given
p, q ∈ X and p 6= q. Let p = (x, y) and q = (u, v). There are two cases:
E = {(x, 0) | x ∈ R}.
Then Uδ is clearly an open set in X and E ⊆ Uδ . Take (x1 , 0), . . . , (xn , 0) ∈ E and
construct
K = B (x1 , 0), δ ∪ B (x2 , 0), δ ∪ · · · ∪ B (xn , 0), δ
= {x1 } × [−δ, δ] ∪ {x2 } × [−δ, δ] ∪ · · · ∪ {xn } × [−δ, δ] .
Since each B (xi , 0), δ is compact, K is also compact and K ⊆ Uδ . By Theorem 2.12
(Urysohn’s Lemma), there exists an f ∈ Cc (X) such that
K ≺ f ≺ Uδ .
where x′1 , . . . , x′m are those values of x for which f (x, y) 6= 0 for at least one y and for
some positive integer m. Since K ≺ f , we have f (xi , y) = 1 for all i = 1, 2, . . . , n and
y ∈ (−δ, δ). Thus we have {x1 , x2 , . . . , xn } ⊆ {x′1 , x′2 , . . . , x′m } and then the inequality
(2.113) reduces to
Xn Z δ
µ(Uδ ) ≥ dy = nδ
i=1 −δ
for some positive integer m. In other words, we know from the definition (2.114) that
Besides, since W is open in X, Theorem 2.12 (Urysohn’s Lemma) again ensures there is a
g ∈ Cc (X) such that
K ≺ g ≺ W.
Recall from [100, Eqn. (7), p. 43] that
m Z
X ∞
µ(K) = inf{Λ(f ) | K ≺ f } ≤ Λ(g) = g(ai , y) dy = 2mδ. (2.115)
i=1 −∞
Since m is fixed and δ is arbitrary, it follows from the inequality (2.115) that µ(K) = 0.
Problem 2.18
Rudin Chapter 2 Exercise 18.
Proof. Recall that X has an order relation ‘<’ and every nonempty subset A of X has a
first/smallest element a, i.e., a ≤ x for all x ∈ A.m If ω0 = min X (ω0 is a smallest element in
X), then we have
X = [ω0 , ω1 ].
Next, we characteristic the topology on X as follows:
[ω0 , α), (β, ω1 ], [ω0 , α) ∩ (β, ω1 ] = (β, α) or a union of such sets. (2.116)
(a) X is a compact Hausdorff space. Suppose that V is an open cover of [ω0 , ω1 ]. Let
α0 = ω1 . Then there is an open set V0 ∈ V such that α0 ∈ V0 . If V0 = X, then X is
compact. Therefore, in the following discussion, we may assume that V0 6= X.
Since V0 is open in X, it follows from the uncountability of X and the order topology
(2.116) that there exists an α1 ∈ X such that (α1 , α0 ] ⊆ V0 . We note that α1 < α0 . Now
this α1 can be chosen such that α1 ∈ / V0 . Otherwise, α1 ∈ V0 and we may choose another
α2 ∈ X such that (α2 , α1 ] ⊆ V0 . Here we have α2 < α1 . Similarly, this α2 can be chosen
such that α2 ∈/ V0 . If not, then we continue this process in the same manner, but this must
stop in a finite number of steps. Otherwise, we can find an infinite decreasing sequence
α1 ∈
/ V0 and (α1 , α0 ] ⊆ V0 . (2.118)
m
See [74, p. 63].
48 Chapter 2. Positive Borel Measures
α2 ∈
/ V1 and (α2 , α1 ] ⊆ V1 . (2.119)
αn+1 ∈
/ V0 ∪ V1 ∪ · · · ∪ Vn and (αn+1 , α0 ] ⊆ V0 ∪ V1 ∪ · · · ∪ Vn .
If this process continues infinitely, then we can find an infinite decreasing sequence (2.117)
again, a contradiction. Therefore, the sequence must stop eventually at the N th step and
it must stop at αn+1 = ω0 for all n ≥ N , i.e.,
ω0 ∈
/ V0 ∪ V1 ∪ · · · ∪ VN and (ω0 , α0 ] ⊆ V0 ∪ V1 ∪ · · · ∪ VN .
[ω0 , ω1 ] = [ω0 , α0 ] ⊆ V0 ∪ V1 ∪ · · · ∪ VN ∪ V.
Lemma 2.5
Define the mapping γ : [ω0 , ω1 ] → [ω0 , ω1 ] by
min(Sα ), if α ∈ [ω0 , ω1 );
γ(α) =
ω1 , if α = ω1 .
Proof of Lemma 2.5. For every α ∈ [ω0 , ω1 ), Sα 6= ∅ so that the mapping γ is well-
defined. By the definition, we have
Sα = [γ(α), ω1 ]
with α < γ(α). Furthermore, there is no element between α and γ(α). Otherwise,
γ(α) is not a smallest element of Sα anymore. Hence we deduce from this that
Let α, β ∈ X and we may assume that α < β. Then we see that α ∈ Pγ(α) = [ω0 , α] and
β ∈ Sα . Thus Pγ(α) and Sα are neighborhoods of α and β respectively. By Lemma 2.5,
they are disjoint and this shows that X is Hausdorff, as required.
2.5. Problems on Regularity of Borel Measures 49
(b) X \ {ω1 } is open but not σ-compact. Let X \ {ω1 } = [ω0 , ω1 ) and α ∈ [ω0 , ω1 ). As we
have proven in part (a) that Pγ(α) = [ω0 , γ(α)) is a neighborhood of α. Hence X \ {ω1 } is
open in X.
Assume that X \ {ω1 } = [ω0 , ω1 ) was σ-compact. Let K be a compact subset of [ω0 , ω1 ).
We know that the family
{Px = [ω0 , x) | x < ω1 }
is an open cover of K so that there are finitely many indices x1 , x2 , . . . , xk such that
Without loss of generality, we may assume that x1 < x2 < · · · < xk . Thus we have
K ⊆ [ω0 , xk ) ⊂ [ω0 , xk ].
Since xk < ω1 , our hypothesis shows that K is at most countable. Since X \ {ω1 } is
σ-compact, it is a countable union of compact sets so that X \ {ω1 } is countable. However,
we have
X = X \ {ω1 } ∪ {ω1 }
which means that X is countable. Evidently, this result contradicts our hypothesis. Hence
X \ ω1 is not σ-compact.
(c) Every f ∈ C(X) is constant on Sα for some α 6= ω1 . Suppose that x = f (ω1 ) and
B(x, n1 ) = {z ∈ C | |z − x| < n1 }, where n is a positive integer. Since f is continuous on X
and B(x, n1 ) is open in C, f −1 B(x, n1 ) is an open subset of X and ω1 ∈ f −1 B(x, n1 ) .
By the topology (2.116), there exists an α1 ∈ X \ {ω1 } such that
1
(α1 , ω1 ] ⊆ f −1 B x, . (2.120)
n
By Lemma 2.5, we note that (α1 , ω1 ] is uncountable, so a sequence {αn } exists such that
Lemma 2.6
Every well-ordered set A has the least upper bound property.
By the relation (2.121), we see that the nonempty subset {αn } of X is bounded above
by ω1 . Therefore, it follows from Lemma 2.6 that sup{αn } exists in X. Call it α. It is
n∈N
clear that α ≤ ω1 and 1
Sα ⊆ Sαn ⊆ f −1 B x, (2.122)
n
n
See [74, Exercise 1, p. 66]
50 Chapter 2. Positive Borel Measures
Sω1 = {x ∈ Y | ω1 < x}
Lemma 2.7
Suppose that K is a nonempty compact subset of X. Then K is uncountable if
and only if ω1 is a limit point of K.
Proof of Lemma 2.7. Let K be uncountable. By the proof of part (b), we know that
K cannot lie inside [ω0 , α] for every predecessor α < ω1 . Otherwise, K will be at most
countable, a contradiction. Hence this implies that
K ∩ (α, ω1 ] 6= ∅
for every α < ω1 . By the definition (see [74, p. 97]), it follows that ω1 is a limit point
of K.
Conversely, suppose that ω1 is a limit point of K. By [74, Theorem 17.9, p. 99], K
must be infinite. Assume that K was countable. Then K has the following represen-
tation
K = {αn },
where α1 ≤ α2 ≤ · · · . Since K is compact and X is Hausdorff, K is closed in X by
Corollary (a) following Theorem 2.5. Thus it must be ω1 ∈ K, i.e., every neighborhood
Sα of ω1 satisfies
K ∩ Sα \ {ω1 } 6= ∅.
Let αN ∈ K ∩ Sα \ {ω1 } for some positive integer N , i.e., α < αN < ω1 . Recall that
{αn } is increasing, so we must have
αn ∈ Sα
for all n ≥ N . By the definition (see [74, p. 98]), {αn } converges to ω1 . However,
by an argument similar to the proof of part (c), this fact will imply the contradiction
that Sαn is uncountable. Hence K is uncountable and we complete the proof of the
lemma.
Let’s go back to the proof of part (d). Denote the intersection of {Kn } by K. Since
X is Hausdorff, each Kn is closed in X. Since K ⊆ Kn , Theorem 2.4 implies that K is
2.5. Problems on Regularity of Borel Measures 51
so it is true that
∞
\ ∞
\
K= Kn ⊆ Enc ∪ {ω1 } = E c ∪ {ω1 }.
n=1 n=1
Now the result of part (d) illustrates that K is uncountable compact, thus E c ∈ M
which implies that E ∈ M.
– M contains all Borel sets in X. If α < ω1 , then (α, ω1 ) 6= ∅. Let β ∈ (α, ω1 ).
Since X = Pβ ∪ [β, ω1 ] and Pβ is countable, [β, ω1 ] must be uncountable. Since
[β, ω1 ] = X \Pβ , [β, ω1 ] is closed in X and Theorem 2.4 verifies that [β, ω1 ] is compact.
Furthermore, we note that
[β, ω1 ] ⊆ Sα , (2.124)
so these facts show that Sα ∈ M.
Similarly, for every α > ω0 , we have X \ Pα = [α, ω1 ] and then X \ Pα ∈ M. Since
M is a σ-algebra, M also contains Pα . Hence M contains the order topology τ of X
which proves that B(X) ⊆ M as desired.
52 Chapter 2. Positive Borel Measures
On the other hand, we know from the facts λ(E1 ) = 1 and λ(Ei ) = 0 for i = 2, 3, . . .
that
X∞
λ(Ei ) = 1.
i=1
Therefore, we have
[
∞ X∞
λ Ei = λ(Ei ) (2.127)
i=1 i=1
in this case. Similarly, if there is no Ei ∪ {ω1 } contains an uncountable compact set,
then each Eic ∪ {ω1 } contains an uncountable compact set Ki so that
λ(Ei ) = 0
for every i = 1, 2, . . .. Furthermore, we have
∞
\ ∞
\ \∞ [
∞ c
K= Ki ⊆ Eic ∪ {ω1 } = Eic ∪ {ω1 } = Ei ∪ {ω1 }
i=1 i=1 i=1 i=1
and then the formula (2.127) still holds in this case. Finally, since λ(X) = 1 < ∞.
By Definition 1.18(a), λ is a positive measure.
2.5. Problems on Regularity of Borel Measures 53
– λ is not regular. For every α < ω1 , we know from the set relation (2.124) that
Sα ∪ {ω1 } must contain an uncountable compact set, so the definition (2.125) gives
that λ(Sα ) = 1. If E = {ω1 }, then E c ∪ {ω1 } = {ω1 }c ∪ {ω1 } = X so that {ω1 } ∈ M.
Therefore, we derive from the definition (2.125) that
λ({ω1 }) = 0.
(g) The validity of the integral. Let f ∈ Cc (X). By part (c), there exists an α0 < ω1 such
that f (x) = f (ω1 ) on Sα0 . Recall from part (e) that Sα0 ∈ M. By the set relation (2.124),
every Sα ∪ {ω1 } contains an uncountable compact set and thus, in particular, λ(Sαc 0 ) = 0.
By this, we gain
Z Z Z Z
f dλ = f dλ + f dλ = f dλ = f (ω1 )λ(Sα0 ) = f (ω1 )
X c
Sα Sα0 Sα0
0
(h) The regular µ associates with the linear functional in part (f ). It is clear that
Λ : Cc (X) → C defined by
Λ(f ) = f (ω1 ) (2.128)
By [100, Eqn. (1), p. 41], we have µ(V ) = sup{Λ(f ) | f ≺ V } = f (ω1 ) for every open set
V in X. Thus this implies that
for every E ∈ M.
Remark 2.2
The measure considered in Problem 2.18 is called the Dieudonné’s measure, see [19,
Exampl 7.1.3, pp. 68, 69] and [31].
Problem 2.19
Rudin Chapter 2 Exercise 19.
54 Chapter 2. Positive Borel Measures
Proof. Suppose that X is a compact metric space with metric ρ and Λ is a positive linear
functional on Cc (X), the space of all continuous complex functions on X with compact support.
To begin with the construction of the class µ, we use Λ to define a set function µ∗ on every open
set V in X by
µ∗ (V ) = sup{Λ(f ) | f ≺ V }
= sup{Λ(f ) | f ∈ Cc (X), 0 ≤ f ≤ 1 on X and supp (f ) ⊆ V } (2.129)
Now we are going to present the proof by quoting several facts (some are with proofs and some
are not). In fact, the idea of the following proof is stimulated by Feldman’s online article [38].
• Fact 1: The set function µ∗ is an outer measure. We check the definition [93, p.
346].
– Since supp (f ) ⊆ ∅ if and only if supp (f ) = ∅, we get from Definition 2.9 that f ≡ 0
on X which implies that
– Suppose that E, F ⊆ X and E ⊆ F . Since any open set V containing F must also
contain E, we obtain from the definition (2.130) that
• Fact 2: µ∗ (K) < ∞ for every compact set K ⊆ X. Since X is open in X, we follow
from the definition (2.129) that µ∗ (X) ≤ Λ(1). By Fact 1, we have µ∗ (K) ≤ µ∗ (X) ≤ Λ(1)
for every compact set K ⊆ X.
• Fact 3: A topological result in metric spaces. Now we need the following topological
result about metric spaces which will be used in Fact 4:
Lemma 2.8
Let X be a metric space with metric ρ and E a nonempty proper subset of X.
Then the set V = {x ∈ X | ρE (x) < ǫ} is open in X, where ǫ > 0 and ρE (x) is
defined in Problem 2.3.
2.5. Problems on Regularity of Borel Measures 55
Proof of Lemma 2.8. If V = ∅, then there is nothing to prove. Thus we assume that
V 6= ∅ and pick x0 ∈ V so that ρE (x0 ) < ǫ. Then there exists a δ > 0 such that
ǫ − δ > 0 and ρE (x0 ) < ǫ − δ. We consider the ball
and we claim that B(x0 , δ) ⊆ V . To see this, let x ∈ B(x0 , δ). For every y ∈ E, we
have
ρ(x, y) ≤ ρ(x, x0 ) + ρ(x0 , y) < δ + ρ(x0 , y)
and therefore we obtain
ρE (x) = inf{ρ(x, y) | y ∈ E}
≤ inf{ρ(x, x0 ) + ρ(x0 , y) | y ∈ E}
< δ + inf{ρ(x0 , y) | y ∈ E}
= δ + ρE (x0 )
< ǫ.
µ∗ (E) = µ∗ (E ∩ V ) + µ∗ (E ∩ V c ). (2.132)
µ∗ (E) ≥ µ∗ (E ∩ V ) + µ∗ (E ∩ V c ) − ǫ (2.133)
holds for every ǫ > 0 because the other side follows directly from the subadditivity (2.131).
By the definition (2.130), there is an open set E ′ such that E ⊆ E ′ and
ǫ
µ∗ (E) ≥ µ∗ (E ′ ) − . (2.134)
2
By the property in Fact 1, we have
µ∗ (E ∩ V ) ≤ µ∗ (E ′ ∩ V ) and µ∗ (E ∩ V c ) ≤ µ∗ (E ′ ∩ V c ). (2.135)
Thus it is easy to see from the inequalities (2.134) and (2.135) that one can obtain the
inequality (2.133) if we can show the following result holds
ǫ
µ∗ (E ′ ) ≥ µ∗ (E ′ ∩ V ) + µ∗ (E ′ ∩ V c ) − . (2.136)
2
Recall the fact that the positivity of Λ implies the monotonicity of Λ. In particular,
Λ(f ) ≤ Λ(1) for every f ≺ E ′ ∩ V . Thus we apply the results (2.138) and the monotonicity
of Λ to the inequality (2.137) to get
1 δ ǫ δ ǫ ǫ
µ∗ (E ′ ∩ V ) ≤ Λ(f1 )+ Λ(f1 )+ ≤ Λ(F1 )+ Λ(1)+ = Λ(F1 )+ . (2.139)
1+δ 1+δ 8 1+δ 8 4
We replace E by the closed set (E ′ ∩V )c in Lemma 2.8 to get the open set U . Given δ > 0.
Since f1 is uniformly continuous on X, there exists a η > 0 such that |f1 (x) − f1 (y)| ≤ δ
for all x, y ∈ X with ρ(x, y) < η. Recall that f1 vanishes on the closed set (E ′ ∩ V )c , so if
we take y ∈ (E ′ ∩ V )c , then we have
f1 (x) ≤ δ
for all x ∈ X with ρ(x, y) < η. Since ρ(E ′ ∩V )c (x) < η, Lemma 2.8 ensures that the set
is open in X. In other words, we have established from Fact 3 that there exists an open
set U such that
U ⊃ (E ′ ∩ V )c ⊇ V c and f1 (x) ≤ δ on U . (2.140)
Since E ′ ∩ U is open in X, the definition (2.129) again shows that we can find a continuous
function f2 : X → [0, 1] such that supp (f2 ) ⊆ E ′ ∩ U and furthermore, by using similar
argument as in the proof of the inequality (2.139), we obtain that
ǫ ǫ
µ∗ (E ′ ∩ V c ) ≤ µ∗ (E ′ ∩ U ) ≤ Λ(f2 ) + ≤ Λ(F2 ) + , (2.141)
4 4
f2
where F2 = 1+δ .
Now we have found the bounds (2.139) and (2.141) of µ∗ (E ′ ∩ V ) and µ∗ (E ′ ∩ V c )
respectively. Next, we want to show that F3 ∈ Cc (X), where F3 = F1 + F2 . To this
1
end, we recall the facts that F1 , F2 : X → [0, 1+δ ] because 0 ≤ f1 ≤ 1 and 0 ≤ f2 ≤ 1.
Furthermore, we note from the set relation (2.140) that
V ∩ U 6= ∅.
To have a better understanding of the above facts, we can draw some pictures. For
examples, Figure 2.3(a) shows the sets V, E ′ and E ′ ∩ V , the shaded blue part in Figure
2.3(b) indicates the closed set (E ′ ∩ V )c and the part inside the circle in Figure 2.4 is the
set (E ′ ∩ V )c \ U .
Thus we have shown that 0 ≤ F3 (x) ≤ 1 on X. Next, since F1 and F2 are continuous
on X, F3 is also continuous on X. By the definition, we know that x0 ∈ supp (F3 ) if and
only if x0 ∈ supp (F1 ) ⊆ E ′ ∩ V or x0 ∈ supp (F2 ) ⊆ E ′ ∩ U . In addition, we note that
(E ′ ∩ V ) ∪ (E ′ ∩ U ) ⊆ E ′ . On the other hand, if y ∈ E ′ but y ∈
/ E ′ ∩ V , then we have
′ c
y ∈ (E ∩ V ) and we follow from the set relation (2.140) that y ∈ U . Thus this implies
that y ∈ E ′ ∩ U and so
E ′ = (E ′ ∩ V ) ∪ (E ′ ∩ U ).
In other words, it gives x0 ∈ supp (F3 ) if and only if x0 ∈ supp (F1 ) ∪ supp (F2 ) ⊆ E ′ , i.e.,
supp (F3 ) ⊆ E ′ .
Finally, it yields from the inequalities (2.139) and (2.141) as well as the definition (2.129)
58 Chapter 2. Positive Borel Measures
that
ǫ ǫ ǫ
µ∗ (E ′ ∩ V ) + µ∗ (E ′ ∩ V c ) ≤ Λ(F1 ) + Λ(F2 ) + = Λ(F3 ) + ≤ µ∗ (E ′ ) +
2 2 2
which is exactly the inequality (2.136). Hence we obtain the desired result that (2.132).
B(X) ⊆ M∗
which is exactly Step VII, where B(X) denotes the set of all Borel sets in X. By
Carathéodory’s Theorem again, it shows that µ is a (complete) measure on B(X) and
this is the same as Step IX. Consequently, Steps III, IV, V, VI and VIII can all be
skipped.o
• Fact 6: Deduction of parts (b) and (d). Let K be a compact set of X. Since X is a
metric space, K is also Hausdorff. By Corollary (a) following Theorem 2.5, we know that
K is closed in X and so K ∈ B(X). Therefore, we deduce from Facts 2 and 5 that
which is exactly part (b) of Theorem 2.14 (The Riesz Representation Theorem).
If E is open in X, then E ∈ B(X). Thus we observe from the equation (2.132) and
then the definition (2.130) that
µ(E) = µ∗ (E)
= µ∗ (X) − µ∗ (E c )
= µ∗ (X) − inf{µ∗ (V ) | V is open in X and E c ⊆ V }
= µ∗ (X) − inf{µ∗ (X) − µ∗ (V c ) | V c is closed in X and V c ⊆ E}
= sup{µ∗ (V c ) | V c is closed in X and V c ⊆ E}
= sup{µ(V c ) | V c is closed in X and V c ⊆ E}. (2.142)
Since X is compact, V c is compact by Theorem 2.4 so that the expression (2.142) can be
expressed as
µ(E) = sup{µ(K) | K compact and K ⊆ E }
which is the same as part (d) of Theorem 2.14 (The Riesz Representation Theorem).
• Fact 7: Deduction of part (a). As Rudin pointed out in Step X that it suffices to
prove Z
Λ(f ) ≤ f dµ (2.143)
X
holds for f ∈ Cc (X) because the opposite inequality follows by changing the sign of f . In
fact, the proof of our inequality (2.143) for compact metric X is essentially the same as
that of Step X, except that Rudin applied Step II to show that
X
µ(K) ≤ Λ(hi ),
o
We cannot say at this stage that Step II can be omitted because the proof of Step X still needs it, but we
will see very soon in Fact 7 that it is eventually allowable to do so.
2.6. Miscellaneous Problems on L1 and Other Properties 59
P
where hi ≺ Vi and hi = 1 on K.p However, this kind of argument can be replaced
completely by using Fact 6 that µ(K) ≤ Λ(1) holds for every compact set K ⊆ X.
Therefore, we finally arrive at the expected inequality (2.143) and furthermore, Step II
can also be skipped.
Hence we conclude that Steps II to IX can be replaced by a simpler argument (outer measure
in the sense of Carathéodory). In other words, only Steps I and X should be kept. Of course,
the proof of the uniqueness of µ cannot be omitted. This completes the proof of the problem.
Remark 2.3
(a) We remark that a Radon measure is a measure µ defined on the σ-algebra M containing
all Borel sets B(X) of the Hausdforff space X and it satisfies µ(K) < ∞ for all compact
subsets K, outer regular on M and inner regular on open sets in X, see [40, p. 212]
or [93, p. 455]. Hence the unique positive measure µ in Theorem 2.14 (The Riesz
Representation Theorem) is in fact a Radon measure when X is Hausdforff.
(b) You are also advised to read the paper [102] for a similar proof of Problem 2.19.
Problem 2.20
Rudin Chapter 2 Exercise 20.
/ L1 on [0, 1] is equivalent to
Proof. Let f = sup fn : [0, 1] → [0, ∞]. By Definition 1.30, f ∈
n
Z 1
|f (x)| dx = ∞.
0
In fact, the inspiration of the construction of {fn } comes from the idea of Problem 2.9. For each
positive integer n, we consider the continuous functions gn : [−1, 1] → [0, ∞) given by
0, / [− n12 , n12 ];
if x ∈
gn (x) =
n − n3 |x|, if x ∈ [− n12 , n12 ].
Recall that fn (x) is the tent function centered at n1 and the graph of it is an isosceles triangle
with height n and base n22 so that each fn is continuous on [0, 1] and its area is n1 which implies
that Z 1
1
fn (x) dx = → 0
0 n
p
See Step X for the meanings of the notations used here.
60 Chapter 2. Positive Borel Measures
/ [ n1 −
as n → ∞. Furthermore, if x = 0, then x ∈ 1 1
,
n2 n
+ 1
n2
] for all n ∈ N. and thus
fn (0) = 0
for all n ∈ N by the definition (2.144). If x ∈ (0, 1], then there is a positive integer N such that
1 1
n + n2 < x for all n ≥ N . In other words, we have x ∈ / [ n1 − n12 , n1 + n12 ] for all n ≥ N . Therefore,
the definition (2.144) certainly shows that
fn (x) → 0
as n → ∞ for all x ∈ (0, 1].
Now it remains to show that f ∈ / L1 on [0, 1]. To this end, we study the behaviour of f as
x → 0. For every n ∈ N, we consider every x in [ n1 − n12 , n1 + n12 ]. If n is sufficiently large enough,
we have 1
1
x= +o , (2.145)
n n
where o(x) is the little-o notation.q Therefore, it follows from the definition (2.144) and the
relation (2.145) that
1
fn (x) = n + o(1) = + o(1) (2.146)
x
for large enough n and small positive x. By the definition and the expression (5.102), we have
1
f (x) = sup fn (x) ≥ fn (x) = + o(1) (2.147)
n x
for small positive x. Obviously, x1 ∈
/ L1 on [0, 1], so we conclude from the estimate that (2.147)
that f ∈ 1
/ L on [0, 1]. This completes the proof of the problem.
Problem 2.21
Rudin Chapter 2 Exercise 21.
Therefore, {Eα } has the finite intersection property and then since X is compact, we conclude
thatr \
Eα 6= ∅.
α
\
Thus suppose that x0 ∈ Eα which means that f (x0 ) ≥ α for all α ∈ f (X), i.e., f (x0 ) ≥ f (x)
α
for all x ∈ X. Hence this shows that f attains its maximum at some point of X, completing the
proof of the problem.
q
Recall that g(n) = o(f (n)) means fg(n)
(n)
→ 0 as n → ∞.
r
See, for instances, [74, Theorem 26.9, p. 169] or [127, Problem 4.23, p. 43].
2.6. Miscellaneous Problems on L1 and Other Properties 61
Remark 2.4
Problem 2.21 can be treated as the Extreme Value Theorem for upper semicontinuous
functions. Similarly, we can show that if f : X → R is lower semicontinuous and X is
compact, then f attains its minimum at some point of X.
Problem 2.22
Rudin Chapter 2 Exercise 22.
Proof. By the definition and the triangle inequality, for all p ∈ X, we have
In other words, gn (x) − nd(x, y) is a lower bound of {f (p) + nd(y, p) | p ∈ X}. Thus, by the
definition of infimum, we must have gn (x) − nd(x, y) ≤ gn (y) or equivalently,
Likewise, we have
gn (y) − gn (x) ≤ nd(x, y). (2.149)
Therefore, we conclude from the inequalities (2.148) and (2.149) that
Since f (x) ≥ 0 on X and d(x, y) ≥ 0 for every x, y ∈ X, we have gn (x) ≥ 0 for all positive
integers n. Fix x ∈ X, we observe that
for every x ∈ X and n = 2, 3, . . .. Besides, it is trivial that gn (x) ≤ f (x) holds for all n ∈ N
if f (x) = ∞. Without loss of generality, we may assume that f (x) < ∞. In this case, since
f (p) ≤ f (p) + nd(x, p) for all p ∈ X, if we take p = x in the definition of gn , then we have
gn (x) ≤ f (x)
Lemma 2.9
A function f : X → R is lower semicontinuous on X if and only if given x ∈ X, for
every {xn } ⊆ X \ {x} converging to x and for every ǫ > 0, there exists a positive
integer N such that n ≥ N implies
Proof of Lemma
2.9. For every α ∈ R, denote Eα = {x ∈ X | f (x) ≤ α}. Recall that
f −1 (α, ∞) is open in X if and only if X \ f −1 (α, ∞) = Eα is closed in X. In other
words, f is lower semicontinuous on X if and only if Eα is closed in X.
Let Eα be closed in X. Assume that the inequality (2.150) was invalid. Then it
means that there exists a x0 ∈ X and a sequence {xn } in X converging to x0 such that
for some ǫ > 0, the inequality
holds for infinitely many k. Take α ∈ (f (x0 ) − ǫ, f (x0 )). On the one hand, since
f (x0 ) > α, x0 ∈ / Eα . On the other hand, we gain from the inequality (2.151) that
f (xnk ) < α for infinitely many k and this implies that
{xnk } ⊆ Eα .
for all n ≥ N . Since f (xn ) ≤ α, we deduce immediately from the inequality (2.152)
that
f (x) < α + ǫ.
Since ǫ is arbitrary, we must have f (x) ≤ α or equivalently x ∈ Eα . This ends the
proof of the lemma.
Remark 2.5
(a) Similarly, a function f : X → R is upper semicontinuous on X if and only if given
x ∈ X, for every {xn } ⊆ X \ {x} converging to x and for every ǫ > 0, there exists a
positive integer N such that n ≥ N implies f (x) > f (xn ) − ǫ.
(b) Except the equivalent definition stated in Lemma 2.9, a lower semicontinuous function
f on X can also be reformulated in terms of the lower limit of f : lim inf f (y) ≥ f (x)
y→x
for every y → x in X \ {x}, see [122, pp. 69, 70] and [129, Definition 3.62, p. 64].
Now it is time to go back to the proof of the problem. On the one hand, since f is lower
semicontinuous on X, Lemma 2.9 implies that for every ǫ > 0, there exists a δ > 0 such that
for every positive integer n. Since ǫ > 0 is arbitrary, the inequality (2.154) becomes
On the other hand, we consider y ∈ / B(x, δ) so that d(x, y) ≥ δ. It is clear that nδ − ǫ > 0 holds
for large enough n,s thus we get from the inequality (2.153) that
for large enough n. Therefore, we conclude that the inequality (2.155) always holds for all y ∈ X
and for all large enough n. By the definition of limit inferior, we have
holds for every x ∈ X. Hence it deduces from the inequalities (2.156) and (2.157) that
on X.
By property (i), each gn is continuous on X. Next, property (ii) says that {gn } is increasing
and bounded by f . Finally, property (iii) ensures that {gn } converges to f on X pointwisely.
This ends the proof of the problem.
Remark 2.6
With the aid of Lemma 2.9, we can prove Problem 2.1(b) easily. In fact, given x ∈ X, for
every {xn } ⊆ X \ {x} converging to x and for every ǫ > 0, there exists a positive integer N1
such that n ≥ N1 implies that
ǫ
f1 (x) < f1 (xn ) + . (2.158)
2
Similarly, there exists a positive integer N2 such that n ≥ N2 implies
ǫ
f2 (x) < f2 (xn ) + . (2.159)
2
Take N = max(N1 , N2 ). If n ≥ N , then we derive from the inequalities (2.158) and (2.159)
that
f1 (x) + f2 (x) < f1 (xn ) + f2 (xn ) + ǫ.
Hence it follows from Lemma 2.9 that f is lower semicontinuous on X.
Problem 2.23
Rudin Chapter 2 Exercise 23.
B(a, ǫ) ⊆ V.
In addition, it follows from the inequality (2.162) and the application of Proposition 1.24(f)
again that
Z Z Z
lim inf χV +xn dµ ≥ χV +x (y) dµ = dµ = µ(V + x). (2.165)
Rk n→∞ Rk V +x
Thus, by substituting the results (2.164) and (2.165) into the inequality (2.163), we achieve that
lim inf µ(V + xn ) ≥ µ(V + x)
n→∞
or equivalently,
lim inf µ(V + y) ≥ µ(V + x)
y→x
Problem 2.24
Rudin Chapter 2 Exercise 24.
Proof. Suppose that S1 and S2 are the sets of simple functions and step functions on R respec-
tively. We divide the proof into several steps.
where {tn } is a sequence of nonnegative increasing simple functions such that for every
x ∈ R, tn (x) → f − (x) as n → ∞. Hence {gn = sn − tn } is a sequence of simple functions
such that
Z ∞ Z ∞
|f − gn | dx = |f + − f − − sn + tn | dx
−∞ −∞
t
Since |f (x)| ≤ |f (x)| on R and f ∈ L (R), we have f + ∈ L1 (R). Thus this makes the second equality holds
+ 1
Z ∞ Z ∞
≤ |f + − sn | dx + |f − − tn | dx
−∞ −∞
→0
as n → ∞.
• Step 2: S2 is dense in S1 . To this end, it suffices to prove that for every measurable
set E ⊂ R with m(E) < ∞ and every ǫ > 0, there exists a step function g on R such that
Z ∞
|χE − g| dx < ǫ. (2.168)
−∞
By Theorem 2.20(b), there exists an open set V in R such that E ⊆ V and m(V \ E) < 2ǫ .
By [99, Exercise 29, Chap. 2, p. 45], we have
∞
[
V = (an , bn ), (2.169)
n=1
where (ai , bi ) ∩ (aj , bj ) = ∅ if i 6= j. Assume that there was an unbounded segment in the
expression (2.169). Then we have m(V ) = ∞. However, since m(V ) = m(V \ E) + m(E),
the facts m(V ) = ∞ and m(V \ E) < 2ǫ will force that m(E) = ∞, a contradiction.
Furthermore, the expression (2.169) implies that
∞
X
m(V ) = m (an , bn ) ,
n=1
Define g : R → R by
N
X
g(x) = χ(an ,bn ) (x)
n=1
and VN = (a1 , b1 ) ∪ · · · ∪ (aN , bN ). Let x ∈ (E \ VN ) ∪ (VN \ E). If x ∈ E \ VN , then
we have χE (x) = 1 and g(x) = 0. Similarly, if x ∈ VN \ E, then we have χE (x) = 0 and
g(x) = 1. Therefore, we have
|g − χE | = 1 (2.171)
on (E \ VN ) ∪ (VN \ E). On the other hand, we let
x ∈ (E \ VN )c ∩ (VN \ E)c = (VN ∪ E c ) ∩ (E ∪ VNc ) = (VN ∩ E) ∪ (VNc ∩ E c ).
If x ∈ VN ∩ E, then we have χE (x) = g(x) = 1. Similarly, x ∈ VNc ∩ E c implies that
χE (x) = g(x) = 0. Both cases give
|g − χE | = 0 (2.172)
on (E \ VN )c ∩ (VN \ E)c . Now the facts E, VN ⊆ V and (2.170) definitely imply that
ǫ ǫ
m(VN \ E) ≤ m(V \ E) < and m(E \ VN ) ≤ m(V \ VN ) < . (2.173)
2 2
Hence we deduce from the results (2.171), (2.172) and the estimates (2.173) that
Z ∞ Z Z
|g − χE | dx = dx + dx = m(VN \ E) + m(E \ VN ) < ǫ.
−∞ E\VN VN \E
Lemma 2.10
Suppose that A, B and C are subsets of a metric space X. If A ⊆ B ⊆ C, A is
dense in B and B is dense in C, then A is dense in C.
A ⊆ B ⊆ A and B ⊆ C ⊆ B
B(p, δ) ∩ (B \ {p}) 6= ∅
A⊆C⊆A
Let’s return to the original proof. By applying Lemma 2.10 to the results in Step 1
and Step 2, we conclude easily that S2 is dense in L1 (R) which is our desired result.
Remark 2.7
By Definition 1.16, we remark that a simple function s is a finite linear combination of
characteristic functions of an arbitrary set Ai and when all Ai are intervals, then s becomes
a step function. In fact, the class of step functions is one of the building blocks for the theory
of Riemann integration, see [99, Chap. 6].
Problem 2.25
Rudin Chapter 2 Exercise 25.
Proof.
(a) We note that et > 0 and log(1 + et ) > 0 for every t > 0. Furthermore, we know that the
functions ex and log x are increasing in their corresponding domains respectively. Thus
68 Chapter 2. Positive Borel Measures
the inequality log(1 + et ) < c + t is equivalent to 1 + et < ec+t and finally, it is equivalent
to
0 < 1 + e−t < ec . (2.175)
Apply log to both sides of the inequality (2.175), we have
for every t > 0. Since the left-hand side of the inequality (2.176) is a decreasing function
of t, the smallest value of c for the validity of the inequality (2.176) on (0, ∞) is found by
(b) Let f ∈ L1 on [0, 1]. Define E = {x ∈ [0, 1] | f (x) > 0} = f −1 (0, ∞) . By Theorem
1.12(b), E is a measurable set in [0, 1]. Thus we follow from part (a) that, on E,
nf (x) = log enf (x) < log 1 + enf (x) < log 2 + nf (x)
on [0, 1]\E. Since [0, 1]\E is also a measurable set in [0, 1], we deduce from the inequalities
(2.178) that Z
1 log 2
0≤ log(1 + enf ) dµ ≤ .
n [0,1]\E n
Now the sum of the inequalities (2.177) and (2.178) give
Z Z 1 Z
1 2 log 2
f dµ < log(1 + enf ) dµ < + f dµ. (2.179)
E n 0 n E
Problem 3.1
Rudin Chapter 3 Exercise 1.
ϕ = sup{ϕα } (3.1)
and assume that it is finite. Suppose that x, y ∈ (a, b) and 0 ≤ λ ≤ 1. By the definition (3.1)
and the convexity of ϕα , we have
ϕα (1 − λ)x + λy ≤ (1 − λ)ϕα (x) + λϕα (y) ≤ (1 − λ)ϕ(x) + λϕ(y)
Now for x, y ∈ (a, b) and 0 ≤ λ ≤ 1, we follow from the limit (3.2) that
and
lim ϕn (1 − λ)x + λy = ϕ (1 − λ)x + λy . (3.4)
n→∞
Since ϕn (1 − λ)x + λy ≤ (1 − λ)ϕn (x) + λϕn (y) for every n = 1, 2, . . ., we take limits to both
sides and apply the results (3.3) and (3.4) to conclude that
ϕ (1 − λ)x + λy ≤ (1 − λ)ϕ(x) + λϕ(y).
69
70 Chapter 3. Lp -Spaces
For each positive integer n, suppose that fn : (a, b) → R is convex. Define f : (a, b) → R by
where gk = sup{fk , fk+1 , . . .} for k = 1, 2, . . .. Thus if each fn is convex on (a, b), then the
first assertion shows that each gk is convex on (a, b). By the definition (3.5), since f is the
pointwise limit of the sequence of convex functions {g1 , g2 , . . .} defined on (a, b), we deduce from
the second assertion that f is also convex on (a, b). However, the lower limit of a sequence of
convex functions may not be convex. For example, consider the functions fn : R → R defined
by
fn (x) = (−1)n x.
It is clear that each fn is convex on R. However, for each x ∈ R, we have
This shows that f is not convex on R and we complete the proof of the problem.
Problem 3.2
Rudin Chapter 3 Exercise 2.
Proof. For every x, y ∈ (a, b) and 0 ≤ λ ≤ 1, since ϕ is convex on (a, b), we have
ϕ (1 − λ)x + λy ≤ (1 − λ)ϕ(x) + λϕ(y). (3.6)
Since ψ is convex and nondecreasing on ϕ (a, b) , we obtain from the inequality (3.6) that
ψ ϕ (1 − λ)x + λy ≤ ψ (1 − λ)ϕ(x) + λϕ(y) ≤ (1 − λ)ψ ϕ(x) + λψ ϕ(y) .
ϕ(x) = x
is convex on (0, ∞), but log x is not convex on (0, ∞) because (log x)′′ = −x−2 < 0 on (0, ∞),
i.e., (log x)′ is not a monotonically increasing function on (0, ∞). We have completed the proof
of the problem.
Problem 3.3
Rudin Chapter 3 Exercise 3.
Problem 3.4
Rudin Chapter 3 Exercise 4.
Proof.
(a) Since 0 < r < p < s, we can find λ ∈ (0, 1) such that p = λr + (1 − λ)s. By Theorem 3.5
(Hölder’s Inequality), we have
Z
ϕ(p) = |f |p dµ
ZX
= |f |λr × |f |(1−λ)s dµ
X
nZ 1 oλ n Z 1
o1−λ
≤ |f |λr λ dµ × [|f |(1−λ)s ] 1−λ dµ
X X
= [ϕ(r)]λ × [ϕ(s)]1−λ . (3.7)
Since r, s ∈ E, ϕ(r) and ϕ(s) are finite. Hence the inequality (3.7) ensures that ϕ(p) is
also finite, i.e., p ∈ E.
(b) We prove the assertion one by one.
– Case (i): ln ϕ is convex in E ◦ . If E ◦ = ∅, then there is nothing to prove. Assume
that E ◦ 6= ∅. By [99, Theorem 2.47, p. 42] and part (a), the set E is connected.
Since E ⊆ (0, ∞), E is either an interval in one of the forms [a, b], [a, b), (a, b] or (a, b)
for some positive a and b with a < b. In each of the cases, we always have E ◦ = (a, b).
Let x, y ∈ (a, b) and λ ∈ [0, 1]. Since (a, b) is a convex set, λx + (1 − λ)y ∈ (a, b).
Thus it follows from the inequality (3.7) that
ϕ λx + (1 − λ)y ≤ [ϕ(x)]λ × [ϕ(y)]1−λ . (3.8)
If ϕ(p) = 0 for some p ∈ (0, ∞), then we have
Z
|f |p dµ = 0
X
on X. In conclusion, we have shown that the sequence {|f |pn } satisfies the hypotheses
of Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem) and then
Z Z
lim ϕ(pn ) = lim |f |pn dµ = |f |a dµ = ϕ(a).
n→∞ n→∞ X X
By definition, ϕ is continuous at a. Similarly, ϕ is continuous at b if b ∈ E. Conse-
quently, ϕ is continuous on E.
(c) We claim that E can be any open or closed connected subset of (0, ∞). We consider
X = (0, ∞), µ = m and 0 < a ≤ b. We need the following result (see [4, Examples 1 & 5,
pp. 417, 419]):
Lemma 3.1
Let x > 0 and b > 1. Then we have
Z 1
1 − x1−α
, if α 6= 1;
t−α dt = 1−α
x
− ln x if α = 1
and 1−β
Z
b −1
b , if β =
6 1;
−β
t dt = 1 − β
1
ln b, if β = 1.
Furthermore, the improper integrals
Z 1 Z ∞
x−α dx and x−β dx
0 1
Apply Lemma 3.1 to the two improper integrals in the expression (3.9), we know that
ϕ(p) < ∞ if and only if pb < 1 and ap > 1 if and only if a < p < b. In other words, we
have E = (a, b) and such split is allowable.
– Case (ii): E = [a, b]. In this case, we need another lemma:a
Lemma 3.2
The improper integral
Z e−1
dx
0 xα | ln x|β
converges if and only if (i) α < 1 or (ii) α = 1 and β > 1. The improper integral
Z ∞
dx
x α (ln x)β
e
We consider
1
1 , if x ∈ (0, e−1 );
1
x | ln x|1+ b
b
1 1
f (x) = 1 e− a − e b (3.10)
e b + (x − e−1 ), if x ∈ (e−1 , e);
e − e−1
1
1 1 , if x ∈ [e, ∞).
x a (ln x)1+ a
By Lemma 3.2, the first integral in the expression (3.11) converges if and only if (i)
p p 1
b < 1 or (ii) b = 1 and p(1 + b ) > 1. The first condition gives p < b. If p = b, then
1
p(1 + b ) = 1 + b > 1 so that the second condition is actually p = b. In this case, we
have p ≤ b.
Similarly, we apply Lemma 3.2 to the third integral in the expression (3.11), so it
converges if and only if (i) ap > 1 or (ii) ap = 1 and p(1 + a1 ) > 1. The first condition
a
The integrals in Lemma 3.2 are called Bertrand’s integrals, see the following webpage
https://fr.wikipedia.org/wiki/Int%C3%A9grale_impropre.
74 Chapter 3. Lp -Spaces
shows p > a. If p = a, then p(1 + a1 ) = 1 + a > 1 so that the second condition implies
that p = a. In this case, we have p ≥ a.
Since the second integral in the expression (3.11) is finite, we combine these obser-
vations to conclude that ϕ(p) < ∞ if and only if p ∈ [a, b].
– Case (iii): E = {a}. If we take b = a in the definition (3.10), then we obtain p ≥ a
and p ≤ a. Hence we establish E = {a}.
– Case (iv): E = ∅. Consider f ≡ 1 on (0, ∞). Since the integral
Z ∞
|1|p dx
0
is obviously divergent for every p ∈ (0, ∞), we conclude that E = ∅ in this case.
(d) Let r < p < s. By the proof of part (a), we see that p = λr + (1 − λ)s for some λ ∈ (0, 1)
so that the inequality (3.7) holds. In other words, we have
λ 1−λ
kf kpp ≤ kf krr × kf kss . (3.12)
It is clear that
λ 1−λ
kf krs × kf kss , if kf kr ≤ kf ks ;
λ 1−λ
kf krr × kf kss ≤
kf kr λ × kf ks 1−λ , if kf k ≤ kf k
r r s r
p
kf ks , if kf kr ≤ kf ks ;
= (3.13)
kf kpr , if kf ks ≤ kf kr .
kf kp ≤ max(kf kr , kf ks ). (3.14)
Hence we follow from these and the inequality (3.14) that kf kp < ∞, i.e., f ∈ Lp (µ) and
then Lr (µ) ∩ Ls (µ) ⊆ Lp (µ).
(e) Recall that kf k∞ > 0.
– Case (i): kf k∞ < ∞. Let Eα = {x ∈ X | |f (x)| ≥ α}, where α ∈ (0, kf k∞ ). It is
clear that Eα is measurable because |f | is measurable. By Definition 3.7, kf k∞ is the
smallest number such that
Thus we have µ(Eα ) > 0 for every α ∈ (0, kf k∞ ). Assume that µ(Eα ) = ∞. Then
we have
Z Z Z
kf krr = |f |r dµ = |f |r dµ + |f |r dµ ≥ αr µ(Eα ) = ∞
X Eα X\Eα
which contradicts to our hypothesis. Thus we have 0 < µ(Eα ) < ∞ and then Propo-
sition 1.24(b) implies that
Z Z
kf kpp = |f |p dµ ≥ |f |p dµ ≥ αp µ(Eα )
X Eα
3.2. Relations among Lp -Spaces and some Consequences 75
lim inf kf kp ≥ α.
p→∞
By Definition 3.7, since |f (x)| ≤ kf k∞ for almost all x ∈ X, we get from the expres-
sion (3.17) and Proposition 1.24(b) that
Z Z
kf kpp ≤ kf kp−r r p−r
∞ × |f | dµ = kf k∞ × |f |r dµ = kf kp−r r
∞ × kf kr , (3.18)
X\E X\E
Hence our desired result follows immediately from the inequalities (3.16) and (3.19).
– Case (ii): kf k∞ = ∞. Then we deduce immediately from the inequality (3.16) that
Problem 3.5
Rudin Chapter 3 Exercise 5.
Proof.
(a) Suppose that s = ∞. Since |f (x)| ≤ kf k∞ holds for almost all x ∈ X, we have
so that Z Z
kf krr = |f |r dµ ≤ kf kr∞ dµ = kf kr∞ · µ(X \ E) = kf kr∞ . (3.20)
X\E X\E
(b) Suppose that 0 < r < s < ∞. Since kf kr = kf ks , it means that the equality holds in the
inequality (3.21). Recall that the inequality (3.21) is derived from Theorem 3.5 (Hölder’s
Inequality), so there are constants α and β, not both 0, such that
s
α(|f |r ) r = β a.e. on X.
β 1
s
|f | = a.e.
α
Next, suppose that s = ∞ and kf kr = kf k∞ < ∞, where 0 < r < ∞. Then the equality
holds in the inequality (3.20) so that |f |r = kf kr∞ < ∞ a.e. on X, i.e.,
|f | = kf k∞ a.e. on X.
Consequently, we conclude that 0 < r < s ≤ ∞ and the conditions kf kr = kf ks < ∞ hold
if and only if |f | is a constant a.e. on X
(c) If f ∈ Ls (µ), then kf ks < ∞ and we know from part (a) that kf kr < ∞, i.e., f ∈ Lr (µ).
Thus it is true that Ls (µ) ⊆ Lr (µ). Now the other inclusion Lr (µ) ⊆ Ls (µ) is a consequence
of the following lemmab :
Lemma 3.3
Suppose that (X, M, µ) is a measure space and M0 = {E ∈ M | µ(E) > 0}. Then
the following conditions are equivalent:
b
This lemma is part of the statements in [120, Theorem 1] and our proof follows part of the argument there.
3.2. Relations among Lp -Spaces and some Consequences 77
Proof of Lemma 3.3. Suppose that Condition (1) is true. Suppose that f ∈ Lr (µ)
and En = {x ∈ X | |f (x)| ≥ n} for every n ∈ N. We claim that there exists an N ∈ N
such that µ(EN ) = 0. Otherwise, there exists ǫ > 0 such that µ(En ) ≥ ǫ for each
∞
\
n ∈ N. Let E = En . By the construction of En , we have E1 ⊇ E2 ⊇ · · · . Since
n=1
µ(E1 ) ≤ µ(X) < ∞, it yields from Theorem 1.19(e) that
Since r < s, we have 1 − rs < 1. Note that f (x) = 0 if x ∈ X \ E. Thus it reduces from
the bounds (3.23) and Theorem 1.29 that
Z Z ∞ Z
X ∞
X r
−r 1− rs 21− s
kf krr = |f |r dµ + |f |r dµ = αn s dµ = αn < r < ∞
X\E E n=1 En n=1
1 − 21− s
Lr (µ) * Ls (µ).
−1 − 1
If s = ∞, then we replace the coefficients αn s by αn 2r in the expression (3.24). By
1
− 2r n
similar argument, since αn > 2 2r for all n ∈ N, we conclude kf k∞ = ∞ and then
Lr (µ) * L∞ (µ).
78 Chapter 3. Lp -Spaces
(d) If 0 < p < q, then part (a) says that kf kp ≤ kf kq < ∞. Let E0 = {x ∈ X | |f (x)| = 0}.
By Problem 1.5, E0 is measurable.
Since µ(X) = 1 and |f (x)| is nonzero finite on X \ (E∞ ∪ E0 ), the first integral on
the right-hand side of the equation (3.25) is finite. However, the second integral on
the right-hand side of the equation (3.25) is in the form −∞ so that the right-hand
side of our desired result is in the form exp{−∞} which is defined to be 0. On the
other hand, we notice that
nZ o1 n Z o1
p p
kf kp = |f |p dµ = |f |p χX\E0 dµ . (3.26)
X X
Apply Theorem 3.5 (Hölder’s Inequality) with u > 1 and v > 1 as fixed conjugate
exponents to the right-hand side of the expression (3.26), we see that
nZ o1 nZ o 1 nZ o1
p pu pv
|f |p χX\E0 dµ ≤ (|f |p )u dµ (χX\E0 )v dµ
X X X
1
= µ(X \ E0 ) pv kf kpu . (3.27)
Since µ(X) = 1 and µ(E0 ) > 0, we have µ(X \ E0 ) < 1 and then
1
lim µ(X \ E0 ) pv = 0. (3.28)
p→0
In addition, when p is chosen to be small enough such that pu < r, part (a) implies
that kf kpu ≤ kf kr < ∞. Hence we observe from the limit (3.28) and this fact that
1 1
lim kf kp ≤ lim µ(X \ E0 ) pv kf kpu ≤ kf kr lim µ(X \ E0 ) pv = 0.
p→0 p→0 p→0
Lemma 3.4
Let |f | > 0 and kf kr < ∞ for some r > 0. Define ψ : R → R by ψ(p) = |f |p . Then
we have
|f |p − 1
∈ L1 (µ)
p
for all p ∈ (0, r).
Let’s return to the proof of the problem. By Theorem 1.34 (Lebesgue’s Dominated
Convergence Theorem), we obtain that
Z Z |f |p − 1
|f |p − 1
lim dµ = lim dµ. (3.34)
p→0+ F p F p→0+ p
|f |p − 1
lim = lim |f |p log |f | = log |f | (3.35)
p→0+ p p→0+
on X. Therefore, when we combine the results (3.32), (3.33), (3.34) and (3.35), we
establish that
Z |f |p − 1 Z Z
lim log kf kp ≤ lim dµ = log |f | dµ = log |f | dµ. (3.36)
p→0+ F p→0+ p F X
Since log x is continuous for x > 0, we get from the inequality (3.36) that
nZ o
lim kf kp ≤ exp log |f | dµ . (3.37)
p→0+ X
Hence the required result follows immediately from the inequalities (3.30) and (3.37).
This ends the proof of the problem.
Problem 3.6
Rudin Chapter 3 Exercise 6.
Proof. Let x > 0 and 0 ≤ c ≤ 1. We claim that the function Φ : [0, ∞) → R satisfies the relation
Obviously, the relation (3.38) holds for c = 0 and c = 1. Without loss of generality, we may
assume that 0 < c < 1 in the following discussion.
Since f is bounded measurable and positive on [0, 1], there exists a positive constant M such
that Z 1
kf k1 = |f | dx ≤ M < ∞.
0
We define f : [0, 1] → R by
x, if x ∈ [0, c];
f (t) = (3.41)
1, if x ∈ (c, 1].
3.2. Relations among Lp -Spaces and some Consequences 81
Since x > 0, the f is bounded measurable and positive on [0, 1]. Therefore, we substitute the
function (3.41) into the relation (3.40) to yield
Φ(xc ) = Φ exp(c log x)
Z c
= Φ exp log x dt
0
Z c Z 1
= Φ(x) dt + Φ(1) dt
0 c
= cΦ(x) + (1 − c)Φ(1).
Lemma 3.5
The relation (3.38) holds for all x > 0 and c ≥ 0.
Proof of Lemma 3.5. The case 0 ≤ c ≤ 1 is clear by the above analysis. Let c > 1 so
that 0 < 1c < 1. Then the relation (3.38) becomes
1 1 1
Φ x c = Φ(x) + 1 − Φ(1). (3.42)
c c
1
Take y = x c . Since 1c > 0, y will take all positive real numbers. Therefore, it follows
from the expression (3.42) that
1 1
Φ(y) = Φ(y c ) + 1 − Φ(1)
c c
and after simplification, we have
We return to the proof of the problem. Define Ψ : [0, ∞) → R by Ψ(x) = Φ(x) − Φ(1), then
the relation (3.38) can be rewritten as
where x > 0 and c ≥ 0. Take x = e and c = ln y, where y ≥ 1. Then the expression (3.43)
becomes
Ψ(y) = Ψ(e) ln y, (3.44)
where y ≥ 1. For 0 < y < 1, we put x = e−1 and c = ln(y −1 ) > 0 into the expression (3.43) so
that
−1
Ψ(y) = Ψ e− ln(y ) = −Ψ(e−1 ) ln y. (3.45)
Combining the results (3.44) and (3.45), we establish that
Ψ(e) ln x, if x ≥ 1;
Ψ(x) =
−Ψ(e−1 ) ln x, if 0 < x < 1
82 Chapter 3. Lp -Spaces
Problem 3.7
Rudin Chapter 3 Exercise 7.
Proof. In this problem, we assume that µ is a positive measure. We give examples one by one:
Example 3.1. Let X = N with the counting measure µ. By Definition 3.6, we have
n X
∞ 1 o
p
ℓp (N) = x = {ξn } kxkp = |ξn |p <∞ .
n=1
Let r < s. Thus if {ξn } ∈ ℓr (N), then {|ξn |r } is a convergent real sequence so that there exists
a positive integer N such that |ξn | ≤ 1 for all n ≥ N , see [99, Theorem 3.23, p. 60]. Therefore,
we must have |ξn |s ≤ |ξn |r for all n ≥ N and this implies that
∞
X ∞
X
|ξn |s ≤ |ξn |r < ∞.
n=N n=N
In other words, we prove that {ξn } ∈ ℓs (N) and then ℓr (N) ⊆ ℓs (N).
Example 3.2. We consider the Lebesgue measure m on X = [0, 1]. Since m(X) = 1, Problem
3.5(c) shows that Ls ([0, 1]) ⊆ Lr ([0, 1]) if 0 < r < s.
Example 3.3. Consider X = (0, ∞) and the Lebesgue measure m. Let 1 ≤ r < s ≤ ∞. Take
1 1
s < α < r and define f : (0, ∞) → R by
Then we have
nZ ∞ o1 n Z 1 o1 nZ ∞ o1 nZ 1 o1
x−αr dx x−αs dx
r r s s
|f |r dx = and |f |s dx = . (3.46)
0 0 0 0
By Lemma 3.1, the integrals on the right-hand sides of the two expressions (3.46) converge if
and only if αr < 1 and αs < 1 respectively. Since 1s < α < 1r , we conclude that f ∈ Lr ((0, ∞))
/ Ls ((0, ∞)), i.e., Lr ((0, ∞)) * Ls ((0, ∞)).
but f ∈
Similarly, we define g : (0, ∞) → R by
Then we have
nZ ∞ o1 n Z ∞ o1 nZ ∞ o1 nZ ∞ o1
r r s s
|f |r dx = x−αr dx and |f |s dx = x−αs dx .
0 1 0 1
Next, we are going to find conditions on µ under which these situations will occur:
• Lr (µ) ⊆ Ls (µ) for 0 < r < s ≤ ∞. Recall from Lemma 3.3 that it is equivalent to the
condition that inf{µ(E) | E ∈ M0 } > 0, where M0 = {E ∈ M | µ(E) > 0}.
• Ls (µ) ⊆ Lr (µ) for 0 < r < s < ∞. Similar to the proof of Lemma 3.3, the following result
is part of the statements in [120, Theorem 2]:
Lemma 3.6
Suppose that (X, M, µ) is a measure space and M∞ = {E ∈ M | µ(E) is finite}.
Then the following conditions are equivalent:
Proof of Lemma 3.6. Suppose that Condition (1) holds. To begin with, let f ∈ Ls (µ)
1
and En = {x ∈ X | n+1 ≤ |f (x)| < n1 } for each n = 1, 2, . . .. Then it is clear that En ∈ M.
Now for each fixed n, we see that
Z Z Z
1 µ(En )
|f |s dµ ≥ |f |s dµ ≥ dµ =
X En (n + 1)s En (n + 1)s
[
∞ [
∞ X∞
lim µ(Fn ) = µ Fn = µ En = µ(En ). (3.47)
n→∞
n=1 n=1 n=1
Furthermore, by combining Condition (1), the expression (3.47) and the inequality
(3.48), it yields that
∞
X
µ(En ) = lim µ(Fn ) ≤ sup{µ(E) | E ∈ M∞ } < ∞.
n→∞
n=1
84 Chapter 3. Lp -Spaces
∞
[ ∞
[
On X \ En , we claim that |f (x)| ≥ 1. Otherwise, there is a x0 ∈ X \ En such that
n=1 n=1
|f (x0 )| < 1. By the definition, x0 ∈ EN for some N ∈ N, a contradiction. This observation
implies that
|f (x)|r ≤ |f (x)|s
∞
[
on X \ En . Finally, we have
n=1
Z Z ∞ Z
X Z ∞
X
r r r µ(En )
|f | dµ = S |f | dµ + |f | dµ ≤ |f |s dµ + < ∞.
X X\ ∞n=1 En
nr
n=1 En X n=1
Problem 3.8
Rudin Chapter 3 Exercise 8.
Proof. For every n ∈ N, since g(x) → ∞ as x → 0, there exists a sequence {ǫn } of positive
numbers such that
g(x) ≥ n (3.49)
for every x ∈ (0, ǫn ). Without loss of generality, we may assume that {ǫn } is decreasing and
ǫn
ǫn+1 < (3.50)
2
for every n = 1, 2, . . .. Now we define hn : (0, 1) → R and h : (0, 1) → R by
x
hn (x) = n 1 − χ (x) and h(x) = sup hn (x)
ǫn (0,ǫn ) n
respectively.
Let x ∈ (0, 1) be fixed but arbitrary. Since ǫn → 0 as n → ∞, there exists a positive integer
Nx such that x ≤ ǫn for all 1 ≤ n ≤ Nx and x ≥ ǫn for all n > Nx . Thus we have x ∈ (0, ǫn )
for all 1 ≤ n ≤ Nx and x ∈ / (0, ǫn ) for all n > Nx . See Figure 3.1 for an illustration of the
distribution of x and ǫn .
3.2. Relations among Lp -Spaces and some Consequences 85
so that x
h(x) = sup hn (x) = sup{h1 (x), h2 (x), . . .} = max n 1 − < ∞,
n 1≤n≤Nx ǫn
i.e., h is finite. Since each hn is convex on (0, 1), Problem 3.1 shows that the h is also convex
on (0, 1).
Next, we want to compare the magnitudes of hn (x) and g(x) at the fixed point x ∈ (0, 1).
On the one hand, for 1 ≤ n ≤ Nx , we have x ∈ (0, ǫn ), so n(1 − ǫxn ) ≤ n and we deduce from
the inequality (3.49) and the definition (3.51) that
for 1 ≤ n ≤ Nx . On the other hand, if n > Nx , then x ∈ / (0, ǫn ) and we have hn (x) = 0, but the
inequality (3.49) still holds for n = Nx (i.e., x ∈ (0, ǫNx )) so that the inequality (3.52) remains
valid in this case. Hence we have established that
for the fixed point x ∈ (0, 1). Since x is arbitrary, the inequality (3.53) holds on (0, 1).
Take n = Nx − 1 in the definition (3.51) to get
x
hNx −1 (x) = (Nx − 1) 1 − . (3.54)
ǫNx −1
By the hypothesis (3.50), we have 2ǫNx < ǫNx −1 . Then since x < ǫNx , we have
x x 1
< < . (3.55)
ǫNx −1 2ǫNx 2
Nx − 1
hNx −1 (x) > .
2
Since x → 0 if and only if Nx → ∞, we conclude that hNx (x) → ∞ as x → 0 and the definition
implies that h(x) → ∞ as x → 0 as required.
1
The second assertion is false. To see this, consider g(x) = x 3 which satisfies g(x) → ∞ as
x → ∞. Assume that there was a convex function h : (0, ∞) → R such that
1
h(x) ≤ x 3 (3.56)
86 Chapter 3. Lp -Spaces
on (0, ∞) and h(x) → ∞ as x → ∞. We know from [131, Exercise 6(c), p. 262]c that if
h : (0, ∞) → R is convex on (0, ∞), then the ratio
h(x)
x
tends to a finite limit or to infinity as x → ∞. If this limit is finite, since h(x) → ∞ as x → ∞,
this limit must be positive. Thus it must be true that
h(x) ≥ kx (3.57)
as x → ∞ for some positive constant k. However, the inequality (3.57) definitely contradicts
the inequality (3.56). This completes the proof of the problem.
Problem 3.9
Rudin Chapter 3 Exercise 9.
Proof. Let Φ : (0, ∞) → (0, ∞) be such that Φ(p) → ∞ as p → ∞. Suppose that E1 = (0, 21 )
and h1 1 1 1 1 1 h 1 1
En = + 2 + · · · + n−1 , + 2 + · · · + n = 1 − n−1 , 1 − n ,
2 2 2 2 2 2 2 2
where n ≥ 2. Obviously, we have En ⊂ (0, 1) for every n ∈ N and Ei ∩ Ej = ∅ for i 6= j.
Furthermore, each En is measurable and
1
m(En ) = > 0.
2n
Suppose that x ∈ (0, 1). If x < 21 , then x ∈ E1 . Otherwise, 12 ≤ x < 1 and there exists a δ > 0
such that x < δ < 1. Take N to be least positive integer such that
log(1 − δ)
N >− > 0.
log 2
1
Then we have x < 1 − 2N
< 1 so that x ∈ E2 ∪ E3 ∪ · · · ∪ EN . In other words, we get
∞
[
(0, 1) = En .
n=1
respectively. Since each En is measurable, each χEn is measurable by Proposition 1.9(d). Thus
each function fn is measurable and then Theorem 1.14 implies that f is also measurable. Now
it remains to show that the function f satisfies the required properties of the problem.
|f (x)| ≥ M
/ L∞ ((0, 1))
on EN , but this contradicts our assumption (3.59). Therefore, we must have f ∈
and in fact kf k∞ = ∞. Furthermore, it follows from the Ratio Test (see [99, Theorem
3.34]) that
Z 1 X∞ X∞
n
kf k1 = f (x) dx = n × m(En ) = n
< ∞,
0 n=1 n=1
2
kf kp → kf k∞ = ∞
as p → ∞.
Consequently, we follow from the inequality (3.61) and the expression (3.62) that
∞ h
n ip
∞
X X X∞
kf kpp 1 p 1
p
= p
n × m(En ) = × m(En ) ≤ =1
[Φ(p)] [Φ(p)] n=1 n=1
Φ(p) n=1
2n
kf kp ≤ Φ(p)
Problem 3.10
Rudin Chapter 3 Exercise 10.
88 Chapter 3. Lp -Spaces
Proof. First of all, we have to show that f ∈ Lp (µ). Recall that Lp (µ) is a metric space with
metric k · k. By the triangle inequality
kfn − fm kp ≤ kfn − f kp + kf − fm kp
and the hypothesis kfn − f kp → 0 as n → ∞ together imply that {fn } is a Cauchy sequence in
Lp (µ). Now the completeness of Lp (µ) ensures that {fn } converges to an element of Lp (µ). By
the uniqueness of limits, this element must be f so that f ∈ Lp (µ). By Theorem 3.12, {fn } has
a subsequence {fnk } such that fnk → f a.e. on X. Since fn → g a.e. on X as n → ∞, every
subsequence of {fn } converges to g a.e. on X. In particular, we take this subsequence {fnk }
and the uniqueness of limits again imply that
f =g
Problem 3.11
Rudin Chapter 3 Exercise 11.
√
Proof. Since f, g : Ω → [0, ∞], the hypothesis f g ≥ 1 implies that f g ≥ 1. By Theorem 3.5
(Hölder’s Inequality), we see that
Z p 2 Z Z
f g dµ ≤ f dµ × g dµ. (3.63)
Ω Ω Ω
Now our desired result follows immediately by combining the inequalities (3.63) and (3.64). This
completes the analysis of the problem.
Problem 3.12
Rudin Chapter 3 Exercise 12.
√
Proof. Suppose that A = ∞. Since h ≤ 1 + h2 on Ω, we have
Z p
1 + h2 dµ = ∞,
Ω
so the
√ equalities
√ definitely hold in this case. Next, suppose that A = 0. By the fact that
1 ≤ 1 + h2 ≤ 1 + 2h + h2 = 1 + h, we achieve that
Z Z p Z Z
dµ ≤ 1 + h2 dµ ≤ (1 + h) dµ = dµ. (3.65)
Ω Ω Ω Ω
Suppose that 0 < A < ∞ which means h(x) ∈ (0, ∞) a.e. on Ω. We consider the function
ϕ : R → (0, ∞) defined by
p
ϕ(x) = 1 + x2
which is exactly
√ the left-hand side inequality. The right-hand side inequality is easy because we
always have 1 + h2 ≤ 1 + h. Hence we have verified the validity of the inequalities.
For the second assertion, we suppose that µ = m on Ω = [0, 1], h = f ′ and h is continuous
on Ω. Then we have Z 1
A= f ′ (x) dx = f (1) − f (0).
0
which is the arc length of the curve from (0, f (0)) to (1, f (1)).d Thus the inequalities give
bounds of such arc length. Geometrically, the upper bound consists of the length from (0, f (0))
to (1, f (0)) (which is 1) and the length from (1, f (0)) to (1, f (1)) (which is A). The lower bound
is just the hypotenuse of the right triangle with vertices (0, f (0)), (0, f (1)) and (1, f (1)), see
Figure 3.2 below:
d
See, for examples, [4, p. 535] and [99, Theorem 6.27, p. 137].
90 Chapter 3. Lp -Spaces
Our first inequality (3.67) comes from the application of Theorem 3.3 (Jensen’s Inequality),
so the equality of the first inequality (3.67) holds if and only if the equality in [100, Eqn. (2),
p. 62] holds if and only if
ϕ(A) = ϕ h(x) . (3.68)
We see from the first derivative in (3.66) that ϕ′ (x) > 0 in (0, ∞) so that it is injective in (0, ∞).
Recall that A ∈ (0, ∞) and h(x) ∈ (0, ∞) a.e. on Ω, so we deduce that the equality (3.68) holds
if and only if
h = A a.e. on Ω.
√
Next we recall the second inequality is derived from the fact that 1 + h2 ≤ 1 + h and some
simple computations show that the equality holds if and only if
h≡0 a.e. on Ω.
Problem 3.13
Rudin Chapter 3 Exercise 13.
• Case (i): 1 < p < ∞. Then the inequality in Theorem 3.8 is just Hölder’s inequality. If
kf kp = 0 or kgkp = 0, then we have f = 0 a.e. on X or g = 0 a.e. on X by Theorem
1.39(a). In either case, the equality holds trivially. Therefore, we may assume that both
kf kp > 0 and kgkp > 0. Furthermore, by the remark following the proof of Theorem 3.5
(Hölder’s Inequality), we may further assume that
In this case, we see that the equality in Theorem 3.8 holds if and only if there are positive
constants α and β such that
αf p = βgq
a.e. on X.
In Theorem 3.9, the inequality is in fact Minkowski’s inequality. By simple observation,
it is clear that the equality holds if f = 0 a.e. on X or g = 0 a.e. on X. Therefore,
we may assume that f 6= 0 and g 6= 0 on measurable sets E and F with µ(E) > 0 and
µ(F ) > 0 respectively. By examining the proof of [100, Eqn. (2), Theorem 3.5, pp. 64,
65], we know that it applies Hölder’s inequality to the functions f and (f + g)p−1 as well
as the functions g and (f + g)p−1 . Thus we deduce from the previous paragraph that the
equality in
Z nZ o1 n Z o1
p q
f · (f + g)p−1 dµ ≤ f p dµ × (f + g)(p−1)q dµ
X X X
holds if and only if there are positive constants α and β such that
holds if and only if there are positive constants λ and ν such that
β βλ p
fp = (f + g)q = g a.e.
α αν
which means that f = Kg a.e. for some positive constant K.
• Case (ii): p = ∞. The inequality in Theorem 3.8 comes from the inequalitye
|f (x)g(x)| ≤ kf k∞ · |g(x)|
for almost all x. Therefore, it is easy to see that the equality in Theorem 3.8 holds if and
only if |f (x)g(x)| = kf k∞ · |g(x)| for almost all x if and only if g(x) = 0 for almost x or
f (x) = kf k∞ for almost x.f
Now the triangle inequality
|f + g| ≤ |f | + |g|
implies the inequality in Theorem 3.9. Thus the equality in Theorem 3.9 holds if and only
if f and g are either nonnegative functions or nonpositive functions for almost all x.
Problem 3.14
Rudin Chapter 3 Exercise 14.
for every x ∈ (0, ∞). By Integration by Partsg and the expression (3.71), we have
Z ∞ Z ∞
F p (x) dx = [xF p (x)]∞
0 − pxF p−1 (x)F ′ (x) dx. (3.72)
0 0
e
This is [100, Eqn. (2), Theorem 3.8, p. 66].
f
The latter case means that f is a nonnegative constant for almost all x.
g
Here we assume that the formula for integration by parts is also valid for improper integrals, see [5, p. 278].
92 Chapter 3. Lp -Spaces
(xF )p
on (0, ∞). Since xF p = xp−1
and p > 1, it reduces from these and the bound (3.74)
that
lim xF p (x) = 0. (3.75)
x→∞
Therefore, we deduce from the formula (3.72) and the two limits (3.73) and (3.75)
that
Z ∞ Z ∞
F p (x) dx = −p F p−1 (x)[f (x) − F (x)] dx
0 0
Z ∞ Z ∞
= −p F p−1 (x)f (x) dx + p F p (x) dx
0 0
so that Z ∞ Z ∞
(p − 1) F p (x) dx = p F p−1 (x)f (x) dx. (3.76)
0 0
Note that 1 < p < ∞, let q be its conjugate exponent. Obviously, our hypotheses
make sure that f and F p−1 have range [0, ∞], so we may apply Theorem 3.5 (Hölder’s
Inequality) to them. In fact, it yields from the fact (p − 1)q = p and the expression
(3.76) that
Z ∞
(p − 1)kF kpp = (p − 1) F p (x) dx
0
nZ ∞ o1n Z ∞ o1
F (p−1)q (x) dx
p q
≤p f p (x) dx
0 0
p
q
= pkf kp · kF kp
which is exactly
p− pq p
kF kp = kF kp ≤ kf kp . (3.77)
p−1
– General Case: f ∈ Lp (0, ∞) . By Definition 3.6, we always have
kf kp =
|f |
p ,
so we may assume that f ≥ 0 on (0, ∞). Since 1 < p < ∞, Theorem 3.14 says that
Cc (0, ∞) is dense in Lp (0, ∞) . Thus there exists a sequence {fn } ⊆ Cc (0, ∞)
such that
kfn − f kp → 0 (3.78)
3.4. Hardy’s Inequality and Egoroff ’s Theorem 93
as n → ∞. Notice that fn may not be nonnegative, but since |fn | − |f | ≤ |fn − f |
and f ≥ 0, we have
|fn (x)| − f (x) ≤ |fn (x) − f (x)| (3.79)
for every x ∈ (0, ∞). Certainly, the inequality (3.79) implies that
|fn | − f
≤ kfn − f kp (3.80)
p
for every n = 1, 2, . . .. Now the result (3.78) and the inequality (3.80) ensure that
|fn | − f
→ 0
p
where Z x
1
Fn (x) = fn (t) dt
x 0
and n = 1, 2, . . .. Suppose that
Z x
1
F (x) = f (t) dt.
x 0
By the result (3.78), we know that for every ǫ > 0, there exists a positive integer N
such that n ≥ N implies
kfn − f kp < ǫ.
Apply Theorem 3.5 (Hölder’s Inequality) with 1p + 1q = 1 to the inequality (3.82), we
get immediately that
Z o1 n Z x
1n x p
o1
q
|Fn (x) − F (x)| ≤ |fn (t) − f (t)|p dt × 1q dt
x 0 0
1 1
≤ kfn − f kp × x q
x
1
= x− p kfn − f kp
1
< ǫx− p (3.83)
for all n ≥ N . Since x is fixed and independent of ǫ, the estimate (3.83) verifies the
truth of the claim.
Next, recall that each fn is continuous on [a, x] for every a > 0, so the First
Fundamental Theorem of Calculus shows that each Fn is continuous on [a, x]. As a
continuous function, every Fn is measurable on [a, x].h Hence, it follows from the
h
See §1.11 or [99, Example 11.14].
94 Chapter 3. Lp -Spaces
pointwise convergence of {Fn }, Theorem 1.28 (Fatou’s Lemma), the inequality (3.81)
and then the result (3.78) that
Z ∞
kF kpp = F p (x) dx
0
Z ∞
= lim inf Fnp (x) dx
0 n→∞
Z ∞
≤ lim inf Fnp (x) dx
n→∞ 0
Z ∞
≤ lim Fnp (x) dx
n→∞ 0
p
p
= kf kp × kF kpq
p−1
= kF kpp .
Hence the equality in the inequality (3.87) must hold. Consequently, there are constants
α and β, not both 0, such that αf p = β(F p−1 )q = βF p a.e. on (0, ∞) and so
1 1
αpf = β pF a.e. on (0, ∞). (3.88)
– Case (i): β = 0. Then we have α 6= 0 and f = 0 a.e. on (0, ∞). Thus we are done.
– Case (ii): α = 0. Then we have β 6= 0 and F = 0 a.e. on (0, ∞). By the hypothesis
(3.84), we see that kf kp = 0 which gives f = 0 a.e. on (0, ∞) and we are done again.
– Case (iii): αβ 6= 0. By Theorem 3.8, we see that f ∈ L1 ([a, b]), where [a, b] is any
bounded interval of (0, ∞). By the comment following the proof of Theorem 11.3 on
[99, p. 324], we know that F is differentiable a.e. on [a, b] and
′
xF (x) = f (x) (3.89)
a.e. on [a, b]. Since [a, b] is arbitrary, the formula (3.89) holds a.e. on (0, ∞). Com-
bining the equations (3.88) and (3.89), we see immediately that xF ′ + xF = f which
implies
xf ′ = (c − 1)f a.e. on (0, ∞) (3.90)
for some nonzero constant c. By solving the differential equation (3.90), we obtain
that
f (x) = γxc−1 a.e. on (0, ∞)
for some constant γ. Since γxc−1 ∈ Lp ((0, ∞)), it forces that γ = 0. Hence we have
shown that the equality holds only if f = 0 a.e. on (0, ∞).
1
(c) Take f (x) = x− p χ[1,A] (x) for large A. Then we have
nZ ∞ o1 nZ A o1
− p1 1
x−1 dx
p p
kf kp = |x χ[1,A] (x)|p dx = = (log A) p . (3.91)
0 1
1
Note that A is large and 1 − p > 0, so we have
1
A1− p − 1 > 0. (3.93)
Therefore, we deduce from the expressions (3.92) and the fact (3.93) that
Z ∞
kF kpp = |F (x)|p dx
0
Z A Z ∞
p
= |F (x)| dx + |F (x)|p dx
1 A
p p Z A p p Z ∞ A1− p1 − 1 p
− p1 −1 p
= |x − x | dx + dx
p−1 1 p − 1 A x
Z
p p A 1 pp 1 p
= (x− p − x−1 )p dx + p+1
A1− p − 1 A1−p
p−1 1 (p − 1)
p p Z A 1
> (x− p − x−1 )p dx. (3.94)
p−1 1
p δp
Fix p first. Assume that the constant p−1 could be replaced by a smaller number p−1
for some δ < 1, i.e.,
δp
kF kp ≤ kf kp .
p−1
1
Take ǫ > 0 such that δ < ǫ < 1. Then it is clear that for large A, we have x p −1 < 1 − ǫ
for x ≥ A2 or equivalently,
− p1 − p1
x − x−1 > ǫx >0 (3.95)
A
for x ≥ 2. By this estimate (3.95), we obtain
Z Z
A
− p1 −1 p
A
A
(x −x ) dx > ǫp x−1 dx = ǫp log A − log > ǫp log A. (3.96)
1 A 2
2
By substituting the inequality (3.96) into the inequality (3.94) and then using the fact
(3.91), we can show that
ǫp p δp p
kF kpp > log A > kf kpp ≥ 0,
p−1 p−1
but this implies
δp
kF kp > kf kp ,
p−1
a contradiction.
(d) Since f (x) > 0 on (0, ∞), there exists a δ > 0 such that we can find an α ∈ (0, ∞) with
Z α
f (t) dt > δ > 0.
0
/ L1 . In other words, this shows that Hardy’s inequality does not hold when
Hence F ∈
p = 1.
3.4. Hardy’s Inequality and Egoroff ’s Theorem 97
Problem 3.15
Rudin Chapter 3 Exercise 15.
Proof. We follow the hint. Suppose that {an } is decreasing. Let f : (0, ∞) → (0, ∞) be given
by
∞
X
f (x) = an χ(n−1,n] (x). (3.97)
n=1
Firstly, we note that
Z ∞ ∞ Z
X n ∞
X
f p (x) dx = apn dx = apn (3.98)
0 n=1 n−1 n=1
∞
X
which implies that f ∈ Lp (0, ∞) if and only if apn < ∞. This shows that the desired
n=1
∞
X
inequality holds if f ∈
/ Lp (0, ∞) . Without loss of generality, we assume that apn < ∞ in
n=1
the following discussion.
Secondly, let N be a positive integer and x ∈ (N − 1, N ]. Then we deduce from the definition
(3.97) that
Z
1 x
F (x) = f (t) dt
x 0
N −1 Z Z x
1h X n i
= f (t) dt + f (t) dt
x N −1
n=1 n−1
Z Z
1h X n i
N −1 x
= an dt + aN dt
x n=1 n−1 N −1
1
= [a1 + a2 + · · · + aN −1 + aN (x − N + 1)]. (3.99)
x
It is clear from the expression (3.99) and the fact {an } is decreasing that
1h X i 1X
N −1 N
an + aN (x − N + 1) = an − N aN + xaN
x x
n=1 n=1
1X
N
= an − N aN + aN
x
n=1
1 X
N
≥ an − N aN + aN
N n=1
N
1 X
= an . (3.100)
N
n=1
Thus we obtain from the expression (3.99) and the inequality (3.100) that
nZ ∞ o1 ∞ Z
nX N o1 ∞
nX 1 X p o 1p
N
p p
kF kp = F p (x) dx = F p (x) dx ≥ an . (3.101)
0 N n=1
N =1 N −1 N =1
98 Chapter 3. Lp -Spaces
Hence our desired inequality follows immediately from the inequalities (3.101) and (3.102).
X∞
For the general case, recall the assumption that an > 0 for all n ∈ N and apn < ∞, so the
n=1
sequence {apn } converges absolutely and every rearrangement converges to the same sum (see
[99, Theorem 3.55, p. 78]). On the other hand, if we fix a positive integer N , then
N X
X k p a p a + a p a + a + a p
1 1 1 2 1 2 3
an = + +
k n=1
1 2 3
k=1
a + a + · · · + a p
1 2 N
+ ··· + ,
N
so it follows from this form that the sum attains its maximum if and only if a1 ≥ a2 ≥ · · · ≥ aN .
Therefore, we have
∞
1 X p
∞
1 X p p p X p p p X p
X N X N ∞ ∞
an ≤ αn ≤ αn = an ,
N N p−1 p−1
N =1 n=1 N =1 n=1 n=1 n=1
| {z }
By the special case.
where {αn } is the decreasing rearrangement of {an }. This completes the proof of the problem.
Problem 3.16
Rudin Chapter 3 Exercise 16.
is evident. To prove the reverse direction, we note that for each k ∈ N and x ∈ X, we
yield from our hypothesis fn (x) → f (x) that there exists a positive integer N such that
i, j > N implies that
1
|fi (x) − fj (x)| < .
k
In other words, we must have x ∈ S(n, k), i.e.,
∞
[
X⊆ S(n, k).
n=1
Thus these prove the validity of our claim (3.104) and Theorem 1.19(d) gives
lim µ S(n, k) = µ(X)
n→∞
for every k = 1, 2, . . ..
Given that ǫ > 0. For each k ∈ N, this fact allows us to choose nk such that
ǫ
|µ S(nk , k − µ(X)| < k .
2
Define
∞
\
E= S(nk , k). (3.105)
k=1
Then it is clear that
∞
[
X \E = X \ S(nk , k)
k=1
which implies that
∞
X ∞ ∞
X X ǫ
µ(X \ E) ≤ µ X \ S(nk , k) ≤ µ(X) − µ S(nk , k) < = ǫ.
2k
k=1 k=1 k=1
1
Pick a k with k < ǫ. If x ∈ S(nk , k), then we have
for all i, j > nk . By the definition (3.105), we know that E ⊆ S(nk , k) for every k ∈ N,
so the inequality (3.106) holds on E. By the definition, fn → f uniformly on E, proving
Egoroff’s Theorem.
• A counterexample on a σ-finite space. Recall from Definition 2.16 that X is said to
be a σ-finite space if X is a countable union of sets Ei with µ(Ei ) < ∞. It is clear that
∞
[ ∞
[
R= [n − 1, n) ∪ (−n, 1 − n],
n=1 n=1
F (t) = ft (x)
is continuous.
Since fr is measurable, f is also measurable by Corollary (a) following Theorem 1.14. Thus
|fr − f | is measurable by Proposition 1.9(b) and then the set
is measurable for every a ∈ R by Definition 1.3(c). By the expression (3.108), since there
are countable such measurable sets, Comment 1.6(c) implies that Ea (n) is measurable for
every real a. By Definition 1.3(c) again, each gn is measurable and thus Egoroff’s Theorem
can be applied to conclude that for every ǫ > 0, there exists a measurable set E ⊆ X such
that µ(X \ E) < ǫ and {gn } converges uniformly to 0 on E. By the definition (3.107), {ft }
converges to f uniformly on E.
Problem 3.17
Rudin Chapter 3 Exercise 17.
Proof.
(a) Since the inequality is clearly holds if either α = 0 or β = 0, we assume, without loss of
generality, that α and β cannot be both zero in the following discussion.
Let 0 < p ≤ 1. We claim that the continuous function ϕ : [0, 1] → R defined by
ϕ(x) = xp + (1 − x)p
3.4. Hardy’s Inequality and Egoroff ’s Theorem 101
ψ(x) = xp
has second derivative p(p − 1)xp−2 > 0 on (0, 1). By Definition 3.1, it is convex on (0, 1),
i.e.,
[(1 − λ)x + λy]p ≤ (1 − λ)xp + λy p , (3.112)
|α| |β|
where x, y ∈ (0, 1) and λ ∈ [0, 1]. Put λ = 12 , x = |α|+|β| and y = |α|+|β| into the inequality
(3.112), we obtain
1 |α| + |β| p 1 |α|p 1 |β|p
· ≤ · + ·
2 |α| + |β| 2 (|α| + |β|)p 2 (|α| + |β|)p
which reduces to
(|α| + |β|)p ≤ 2p−1 (|α|p + |β|p ).
Again, the triangle inequality and this show that
Hence the desired inequality follows from combining the inequalities (3.111) and (3.113).
(b) (i) If µ(X) = 0, then Proposition 1.24(e) implies that kf kp = 0 for every f ∈ Lp (µ)
and there is nothing to prove. Thus, without loss of generality, we may assume that
µ(X) > 0. We divide the proof into several steps:
102 Chapter 3. Lp -Spaces
Proof of Lemma 3.7. Note that |f | must be bounded a.e. on X. Otherwise, there
exists an E ∈ M such that µ(E) > 0 and |f | = ∞ on E. Then Proposition 1.24(b)
shows that Z Z
|f |p dµ ≥ |f |p dµ = ∞,
X E
a contradiction. Thus there exists a positive constant M such that |f | ≤ M a.e.
on X. Given ǫ > 0. Now for every E ∈ M with µ(E) < Mǫ p , we have
Z
|f |p dµ ≤ M p µ(E) ≤ ǫ a.e. on X
E
ǫ 1
p
|fn (x) − f (x)| ≤
µ(B)
for n ≥ N .
∗ Step 2: Lemma 3.8 and its application.
Lemma 3.8
If f ∈ L1 (µ), then for every ǫ > 0, there exists an E ∈ M such that F = X \ E,
µ(F ) < ∞ and Z
|f | dµ ≤ ǫ.
E
3.4. Hardy’s Inequality and Egoroff ’s Theorem 103
Proof of Lemma 3.8. Since |f | is measurable, Theorem 1.17 (The Simple Function
Approximation Theorem) ensures that there exists a sequence of simple measur-
able functions {sk } on X such that
0 ≤ s1 ≤ s2 ≤ · · · ≤ |f | (3.117)
Hence, for every ǫ > 0, there exists a positive integer N such that
Z Z
(|f | − sk ) dµ = (|f | − sk ) dµ ≤ ǫ. (3.118)
X X
Then the previous analysis shows that µ(F ) < ∞. Furthermore, it is clear that
sk (x) > 0 if and only if x ∈ Fi for some i, so this fact implies that
X \ E = F.
Thus we deduce from the estimate (3.118) and the definition of E that
Z Z Z Z Z
|f | dµ = (|f | − sk ) dµ + sk dµ ≤ (|f | − sk ) dµ + sk dµ ≤ ǫ.
E E E X E
and since E ∩ (F \ B) = ∅, we obtain from Theorem 1.29 and the two estimates
(3.119) and (3.115) that
Z Z Z
|f |p dµ = |f |p dµ + |f |p dµ ≤ ǫ. (3.120)
A E F \B
= |f |p dµ
A
≤ ǫ. (3.122)
Putting the limit (3.124) into the inequality (3.123), we obtain from Problem 1.4 that
Z Z
2γp |f |p dµ = lim hn dµ
X n→∞
ZX
= lim inf hn dµ
X n→∞
Z
= lim inf hn dµ
n→∞ X
h Z Z i
≤ lim inf γp (|f |p + |fn |p ) dµ − |fn − f |p dµ
n→∞ X X
hZ i Z
p p
≤ γp lim inf (|f | + |fn | ) dµ − lim sup |fn − f |p dµ
n→∞ X n→∞ X
Z
= γp lim inf kfn kpp + kf kpp − lim sup |fn − f |p dµ. (3.125)
n→∞ n→∞ X
as desired.
(c) Consider X = (0, 1) and µ = m the Lebesgue measure. For each n = 1, 2, . . ., we let
En = (0, n1p ) and fn = nχEn : (0, 1) → R. If x ∈ (0, 1), then there exists a positive integer
N such that x ∈ / En for all n ≥ N . In this case, we have fn (x) = 0 for all n ≥ N . In other
words, we have
fn (x) → f (x) ≡ 0
for every x ∈ (0, 1). It is clear that f ∈ Lp (0, 1) and kf kp = 0. Besides, we have
nZ 1 o1 nZ o1
p p
kfn kp = |fn (x)|p dx = np dx =1
0 En
kfn kp 9 kf kp
Remark 3.1
We remark that there is a short and elementary proof of Problem 3.17(b) in [82].
106 Chapter 3. Lp -Spaces
Problem 3.18
Rudin Chapter 3 Exercise 18.
Proof.
Since fn (x) → f (x) a.e. on X, we deduce from the expression (3.126) that
µ(X \ E) = 0.
This result ensures that there exists a N ∈ N such that µ(X \ EN ) < ǫ.
i.e., fn → f in measure.
Let Fn = {x ∈ X | |fn (x) − f (x)| > ǫ} for each n = 1, 2, . . .. Now it is easy to check thati
Z Z
|fn (x) − f (x)|p dµ ≥ ǫp dµ = ǫp µ(Fn ). (3.128)
Fn Fn
i
The inequality (3.128) is a consequence of the so-called Chebyshev’s Inequality: If f is a nonnegative,
extended real-valued measurable function on X with measure µ, 0 < p < ∞ and ǫ > 0, then
Z
1
µ({x ∈ X | f (x) ≥ ǫ}) ≤ p f p dµ.
ǫ X
See [40, Theorem 6.17, p. 193].
3.5. Convergence in Measure and the Essential Range of f ∈ L∞ (µ) 107
(c) Suppose that fn → f in measure, i.e., for every ǫ > 0, there exists a positive integer N
such that n ≥ N implies that
In particular, for each positive integer k, there exists a positive integer Nk such that
for all n ≥ Nk . Now we may choose nk > Nk freely in the estimate (3.129) and consider
the subsequence {fnk }. We define
∞ [
\ ∞
Ek = {x ∈ X | |fnk (x) − f (x)| > 2−k } and E = Ek . (3.130)
m=1 k≥m
• Failure of the converse of part (a). By Problem 2.9, there exists a sequence of
continuous functions fn : [0, 1] → R such that 0 ≤ fn ≤ 1 and
kfn − 0k1 → 0
as n → ∞, but there is no x ∈ [0, 1] such that {fn (x)} converges. In other words, fn (x)
does not converge to 0 a.e. on [0, 1]. However, since each fn is continuous [0, 1], it is R
on [0, 1] so that fn ∈ L1 [0, 1] by [99, Theorem 11.33, p. 323]. Thus the sequence of
functions {fn } satisfies the hypotheses of part (b) which shows that fn → 0 in measure.
Hence this example implies the the converse of part (a) is false.
108 Chapter 3. Lp -Spaces
• Failure of the converse of part (b). Suppose that X = [0, 1], µ is the Lebesgue
measure and
fn = en χ[0, 1 ]
n
for each n = 1, 2, . . .. Then fn → 0 a.e. on [0, 1] and part (a) implies that fn → 0 in
measure. However, we note that
nZ o1 n Z n1 o1 en
p p p
kfn kp = |fn (x)| dµ = enp dx = 1 →∞
X 0 np
p
as n → ∞, so the sequence {fn } fails to converge in L [0, 1] .
Finally, by examining the proofs of parts (b) and (c), we see that they remain to be true in
the case µ(X) = ∞. However, part (a) does not hold in this case. For instance, let X = [0, ∞)
and En = [0, n] for each n ∈ N. Consider the functions fn = χEn . Then for every x ∈ [0, ∞),
there exists a positive integer N such that x ∈ En for all n ≥ N , so it is true that fn (x) = 1 for
all n ≥ N , i.e., fn (x) → f (x) ≡ 1 on [0, ∞). However, for every ǫ ∈ (0, 1), since fn (x) = 0 if and
only if x ∈ (n, ∞), we see that
Problem 3.19
Rudin Chapter 3 Exercise 19.
Since ǫ is arbitrary, we know from the estimate (3.132) that w ∈ Rf . In other words, Rf
is closed in C.
Recall that kf k∞ < ∞, so we pick a complex number w0 such that |w0 | > kf k∞ and
then consider the positive number ǫ = |w0 | − kf k∞ . If x ∈ Eǫ (w0 ), then the triangle
inequality shows that
which implies that |f (x)| > kf k∞ . Thus we have Eǫ (w0 ) ⊆ {x ∈ X | |f (x)| > kf k∞ }. By
Definition 3.7, µ({x ∈ X | |f (x)| > kf k∞ }) = 0, so µ Eǫ (w0 ) = 0 too. By the definition
(3.131), we have w0 ∈/ Rf and hence the set Rf is bounded by kf k∞ . By the Heine–Borel
Theorem, we conclude that Rf is compact.
• A relation between Rf and kf k∞ . By the previous analysis, it is obvious that the
relation
Rf ⊆ {w ∈ C | |w| ≤ kf k∞ } (3.133)
holds, i.e., Rf lies in B(0, kf k∞ ). We claim that
kf k∞ = max{|z| | z ∈ Rf }. (3.134)
– Fact 1: It is easy to show that V is the largest open subset of C such that
µ f −1 (V ) = µ({x ∈ X | f (x) ∈ V }) = 0,
Hence we follow immediately from these two observations that the equality actually holds
in the set relation (3.133). Since they are equal, our claim (3.134) is established.
• Relations between Af and Rf . By the definition, we have
n 1 Z o
Af = f dµ E ∈ M and µ(E) > 0 . (3.136)
µ(E) E
In other words, w is a limit point of Af and this means that w ∈ Af . Hence this proves
the claim that Rf ⊆ Af .
• Af is not always closed. For example, we consider X = [0, 1], f (x) = x with µ = m the
Lebesque measure. By Definition 3.7, f ∈ L∞ [0, 1] . In addition, for a measurable set E
in [0, 1] with m(E) > 0, we have
Z
1
w(E) = f (x) dx. (3.137)
m(E) E
We claim that w(E) can take any value in (0, 1). Indeed, if a, b ∈ (0, 1) and E = (a, b),
then we have m(E) = b − a > 0 and it follows from the definition (3.137) that
Z Z b
1 1 1 b2 − a 2 a+b
w(E) = f dx = x dx = × =
m(E) E b−a a b−a 2 2
which implies the claim.
Next, we claim that w(E) 6= 0. Assume that w(E0 ) = 0 for a measurable set E0 ⊆ [0, 1]
with m(E0 ) > 0. Then we have Z
f (x) dx = 0,
E0
but Theorem 1.39(a) implies that f (x) = 0 a.e. on E0 which contradicts the fact that
f (x) = 0 only at x = 0. Hence 0 is not a limit point of Af and Af is not closed in R.
• A measure µ such that Af is convex for every f ∈ L∞ (µ). Consider X = {0}
and µ the counting measure (see Example 1.20(a)). Then we have M = {∅, X} and
µ(X) = µ({0}) = 1 > 0. By the definition, kf k∞ = |f (0)| < ∞ for every f ∈ L∞ (µ) so
that f (0) is either kf k∞ or −kf k∞ . Now for every f ∈ L∞ (µ), we have
Z
1
w(X) = f dµ = f (0).
µ(X) X
Therefore, the definition (3.136) gives either Af = {kf k∞ } or Af = {−kf k∞ }, but both
cases are also convex sets.
• A measure µ such that Af is not convex for some f ∈ L∞ (µ). We consider the set
X = {0, 1}, µ the counting measure and f (x) = x. Then we have M = {∅, {0}, {1}, X},
µ({0}) = µ({1}) = 1 > 0 and µ(X) = 2 > 0. Now we have
Z Z
1 1
w({0}) = f dµ = f (0) = 0, w({1}) = f dµ = f (1) = 1,
µ({0}) {0} µ({1}) {1}
Z
1 1 1 1
w(X) = f dµ = [f (0)µ({0}) + f (1)µ({1})] = [f (0) + f (1)] = .
µ(X) X 2 2 2
Thus we have Af = {0, 1, 21 } is not convex.
• The situations when L∞ (µ) is replaced by L1 (µ). We consider f : (0, ∞) → C by
1
√ , if x ∈ (0, 1];
x
f (x) =
0, if x > 1
3.6. A Converse of Jensen’s Inequality 111
and take µ = m the Lebesgue measure. By Lemma 3.1, we have f ∈ L1 (0, ∞) . For any
large positive integer N , if x ∈ (0, N12 ), then we have f (x) > N and
1 1
m({x ∈ (0, ∞) | f (x) > N }) = m 0, 2 = 2 > 0.
N N
By Definition 3.7, we have kf k∞ / L∞ (0, ∞) . Given that ǫ > 0 and
= ∞, i.e., f ∈
sufficiently large N ∈ N.
1
– Rf is unbounded and not compact. If x ∈ , 1
(N +ǫ)2 (N −ǫ)2
⊆ (0, 1), then we
have |f (x) − N | < ǫ so that
1 1
m({x ∈ (0, ∞) | |f (x) − N | < ǫ}) = m , > 0.
(N + ǫ)2 (N − ǫ)2
Thus N ∈ Rf and then N ⊆ Rf , i.e., Rf is unbounded and not compact.
– Af is unbounded. Next, if E = (0, N42 ), then m(E) = 4
N2 > 0 and
Z Z
√ 42
4
1 N2 N2 1 N2
w(E) = f (x) dx = x− 2 dx = × 2 x N = N.
m(E) E 4 0 4 0
Problem 3.20
Rudin Chapter 3 Exercise 20.
Proof. We check Definition 3.1. Let p, q ∈ R and λ ∈ [0, 1]. By the trick of Proposition 1.24(f),
we see that
Z Z
λp + (1 − λ)q = p χ[0,λ] (x) dx + q χ[λ,1] (x) dx
[0,1] [0,1]
Z 1
= [pχ[0,λ] (x) + qχ[λ,1] (x)] dx. (3.138)
0
Suppose that
f (x) = pχ[0,λ] (x) + qχ[λ,1] (x) (3.139)
which is clearly a real and bounded function. Besides, since [0, λ] and [λ, 1] are Borel sets in
[0, 1], χ[0,λ] and χ[λ,1] are measurable by Proposition 1.9(d). By the last paragraph in [100, §1.22,
112 Chapter 3. Lp -Spaces
p. 19], the function f , as the sum of two measurable functions, is also measurable. Hence, by
substituting the expressions (3.138) and (3.139) into the inequality in question, we obtain
Z 1
ϕ λp + (1 − λ)q ≤ ϕ pχ[0,λ] (x) + qχ[λ,1] (x) dx
0
Z λ Z 1
= ϕ(p) dx + ϕ(q) dx
0 λ
= λϕ(p) + (1 − λ)ϕ(q).
Remark 3.2
We note that Problem 3.20 is a converse of Theorem 3.3 (Jensen’s Inequality).
Problem 3.21
Rudin Chapter 3 Exercise 21.
Proof. Let (X, d) and (X ∗ , ρ) be two metric spaces. An isometry of X into X ∗ is a mapping
ϕ : X → Y such that
ρ ϕ(p), ϕ(q) = d(p, q) (3.140)
for all p, q ∈ X. We notice immediately that an isometry ϕ is necessarily injective and continuous.
If ϕ is surjective, then we call ϕ an isomorphism. Two metric spaces X and Y are called
isomorphic if there is an isomorphism between them. Before stating the statement, we need:
Lemma 3.9
Let (X, d) be a metric space with metric d and p ∈ X. Then the function f : X → R
defined by f (x) = d(x, p) is continuous.
Proof of Lemma 3.9. Given ǫ > 0. Let x ∈ X. If y ∈ X satisfies d(x, y) < ǫ, then the
triangle inequality implies that
Lemma 3.10
Suppose that (X, d) and (Y, ρ) are metric spaces and Y is complete. If E is dense
in X and f : E → Y is an isometry, then there exists an isometry g : X → Y such
that g|E = f .
3.7. The Completeness/Completion of a Metric Space 113
y′
Since ρ is a metric, we must have y = and so we obtain a mapping g : X → Y given
by
f (x), if x ∈ E;
g(x) = (3.141)
lim f (xn ), if x ∈ X \ E, {xn } ⊆ X and xn → x.
n→∞
• Case (i): x, y ∈ E. In this case, we know from the definition (3.141) that
g(x) = f (x) and g(y) = f (y) and then the use of the expression (3.140) implies
that
ρ g(x), g(y) = ρ f (x), f (y) = d(x, y).
Choose a sequence {xn } in E such that xn → x. By Lemma 3.9 and the expres-
sion (3.140), we establish that
ρ g(x), f (y) = lim ρ f (xn ), f (y) = lim d(xn , y) = d(x, y). (3.142)
n→∞ n→∞
Lemma 3.11
The mapping g in Lemma 3.9 is unique up to isometry.
f = ϕ2 ◦ ϕ−1
1 . (3.144)
g|ϕ1 (X) = f.
zn = ϕ1 (xn ) (3.145)
for n ∈ N. Since {yn } is Cauchy in ϕ2 (X) (and hence in Y2 ) and ϕ2 is an isometry, the expression
(3.140) forces that {xn } is Cauchy in X. Then, since ϕ1 is an isometry, the expression (3.140)
again forces that {zn } is Cauchy in ϕ1 (X) (and hence in Y1 ). Since Y1 is complete, we can find
z ∈ Y1 such that zn → z as n → ∞. We claim that
g(z) = y.
3.8. Miscellaneous Problems 115
Thus we deduce from Lemma 3.10, the definition (3.145) and the fact g|ϕ1 (X) = f that
ρ2 g(z), y = lim ρ2 g(zn ), y = lim ρ2 g ϕ1 (xn ) , y = lim ρ2 f ϕ1 (xn ) , y . (3.146)
n→∞ n→∞ n→∞
Therefore, g(z) = y and g is surjective. Hence g is an isomorphism and our claim follows,
completing the proof of the problem.
Problem 3.22
Rudin Chapter 3 Exercise 22.
Problem 3.23
Rudin Chapter 3 Exercise 23.
Proof. If αn0 = 0 for some n0 , then we must have f = 0 a.e. on X by Theorem 1.39(a). However,
this implies that the measure of
f −1 (α, ∞] = {x ∈ X | f (x) > α}
is zero for every α > 0. By Definition 3.7, it means that kf k∞ = 0 which contradicts the
hypothesis. Thus we have αn 6= 0 for all n ∈ N so that the limit in question is well-defined.
By Definition 3.7, we see that |f (x)| ≤ kf k∞ holds for almost all x ∈ X. Therefore, we have
for all n ∈ N. Since kf k∞ > 0, we can find a ǫ > 0 such that kf k∞ > ǫ > 0. By the definition,
the measurable set
F = {x ∈ X | |f (x)| ≥ kf k∞ − ǫ > 0}
satisfies µ(F ) > 0. Apply Theorem 3.5 (Hölder’s Inequality) to the measurable functions f n
and 1 with p = n+1n , we obtain
Z nZ o n nZ o 1
n+1 n+1 n+1 n 1
αn = |f |n dµ ≤ (|f |n ) n 1n+1 dµ = (αn+1 ) n+1 × µ(X) n+1
X X X
so that
αn+1 1 −1
≥ (αn+1 ) n+1 × µ(X) n+1
αn
nZ o 1 −1
n+1
≥ |f |n+1 dµ × µ(X) n+1
F
1 −1
≥ (kf k∞ − ǫ)n+1 µ(F ) n+1 × µ(X) n+1
116 Chapter 3. Lp -Spaces
µ(F ) 1
n+1
= × (kf k∞ − ǫ). (3.148)
µ(X)
Problem 3.24
Rudin Chapter 3 Exercise 24.
Proof. Let x ≥ 0 and y ≥ 0. Put α − β = x and α = y into the equality (3.111), we gain
if 0 < p < 1.
If x = y, then the inequality
certainly holds. Without loss of generality, we may assume that x < y. For p > 1, we consider
the function ϕ(t) = tp on [x, y]. Since ϕ is differentiable in (x, y), the Mean Value Theorem
implies that
|y p − xp | ≤ |x − y||ϕ′ (ξ)| ≤ p|x − y|ξ p−1 (3.153)
for some ξ ∈ (x, y). Clearly, we have ξ p−1 ≤ xp−1 + y p−1 , so it reduces from the inequality
(3.153) that
|xp − y p | ≤ p|x − y|(xp−1 + y p−1 ). (3.154)
Hence we establish from the inequalities (3.152) and (3.154) that
|x − y|p , if 0 < p < 1;
p p
|x − y | ≤
p|x − y|(xp−1 + y p−1 ), if 1 ≤ p < ∞.
3.8. Miscellaneous Problems 117
– The truth of the inequality. Put x = |f | ≥ 0 and y = |g| ≥ 0 into the inequality
(3.152) and then take integration, we get
Z Z
p
|f | − |g|p dµ ≤ |f | − |g|p dµ. (3.155)
Since |a| − |b| ≤ |a − b| holds for every a, b ∈ R, we follow immediately from the
inequality (3.155) that
Z Z Z
p
|f | − |g|p dµ ≤ |f | − |g|p dµ ≤ |f − g|p dµ, (3.156)
for all f, g ∈ Lp (µ). Next, Theorem 1.39(a) ensures that ∆(f, g) = 0 if and only if
|f − g|p = 0 a.e. on X, i.e., f = g a.e. on X. It is clear that
Z Z
∆(f, g) = |f − g|p dµ = |g − f |p dµ = ∆(g, f ).
Finally, since |f |p − |g|p ≤ |f |p − |g|p , it is clear from the inequality (3.156) that
Z Z Z
p
p p
(|f | − |g| ) dµ ≤ p
|f | − |g| dµ ≤ |f − g|p dµ. (3.157)
= ∆(f, h).
After rearrangement, we have ∆(f, g) ≤ ∆(f, h) + ∆(h, g). Hence, by the definition,
∆ is in fact a metric.
– (Lp (µ), ∆) is a complete metric space. Suppose that {fn } is a Cauchy sequence
in Lp (µ) with respect to the metric ∆, i.e., for every ǫ > 0, there exists a positive
integer N such that n, m ≥ N implies that
Although ∆(f, g) = kf −gkpp here, we cannot apply Theorem 3.11 directly to conclude
that Lp (µ) is complete with respect to ∆ because p ≥ 1 in Theorem 3.11.
118 Chapter 3. Lp -Spaces
We imitate the proof of Theorem 3.11. To start with, there is a subsequence {fni }
such that
∆(fni+1 , fni ) < 2−i (3.159)
k
X ∞
X
gk = |fni+1 − fni | and g = |fni+1 − fni |
i=1 i=1
X
k
∆(gk , 0) = ∆ |fni+1 − fni |, 0
i=1
Z X
k p
= |fni+1 − fni | dµ
i=1
k nZ
hX o 1 ip
p
≤ |fni+1 − fni |p dµ
i=1
hX
k
1
ip
= ∆(fni+1 , fni ) p (3.160)
i=1
for each k = 1, 2, . . ..
To proceed further, we need the following result:
Lemma 3.12
Let p ∈ (0, 1). For nonnegative real numbers a1 , a2 , . . . , ak , we have
(α + β)p ≤ αp + β p (3.161)
Apply Lemma 3.9 to the inequality (3.160) and then using the inequality (3.159)
to get
k
X k
X
∆(gk , 0) ≤ ∆(fni+1 , fni ) < 2−i ≤ 1 (3.162)
i=1 i=1
for each k = 1, 2, . . ..
3.8. Miscellaneous Problems 119
converges absolutely for almost every x ∈ X. Denote the sum (3.163) by f (x) for
those x at which the sum (3.163) converges and put f (x) = 0 on the remaining set
of measure zero.
Since we have
k−1
X
fn1 (x) + [fni+1 (x) − fni (x)] = fnk (x),
i=1
lim ∆(f, fm ) = 0.
m→∞
By Theorem 3.5 (Hölder’s Inequality) and the facts that kf kp ≤ R and kgkp ≤ R, we
observe that
Z Z Z
|f − g|(|f |p−1 + |g|p−1 ) dµ ≤ |f − g| · |f |p−1 dµ + |f − g| · |g|p−1 dµ
nZ o1 n Z p
o p−1
p p
≤ |f − g|p dµ (|f |p−1 ) p−1 dµ
120 Chapter 3. Lp -Spaces
nZ o1 n Z p
o p−1
p p
+ |f − g|p dµ (|g|p−1 ) p−1 dµ
Problem 3.25
Rudin Chapter 3 Exercise 25.
Proof. Suppose that E ⊆ X and 0 < µ(E) < ∞. Let ME be a σ-algebra in E. Then (E, ME , µ)
is a measure space.j Define
Z
1
ϕ(F ) = dµ (F ∈ ME ).
F µ(E)
for every measurable f on E with range in [0, ∞]. It is trivial that ϕ(E) = 1 and
Z Z
1 1
0≤ f dϕ ≤ f dµ = <∞
E µ(E) X µ(E)
or equivalently, Z Z
log f dϕ ≤ log f dϕ . (3.168)
E E
Now, by using the fact (3.167) to the inequality (3.168), we gain
Z 1 Z 1 Z
1 1
log f dµ ≤ log f dµ ≤ log f dµ = log
µ(E) E µ(E) E µ(E) X µ(E)
which implies the first inequality.
For the second inequality, we define ψ : (0, ∞) → R by
ψ(x) = −xp ,
j
See the paragraphs following Proposition 1.24.
3.8. Miscellaneous Problems 121
where 0 < p < 1. Since ψ ′′ (x) = p(1 − p)xp−2 > 0 on (0, ∞), we apply Theorem 3.3 (Jensen’s
Inequality) with ψ and measure ϕ to conclude that
nZ op Z
− f dϕ ≤ − f p dϕ. (3.169)
E E
so that Z
f p dµ ≤ µ(E)1−p ,
E
completing the proof of the problem.
Problem 3.26
Rudin Chapter 3 Exercise 26.
Proof. Let M be a σ-algebra in [0, 1]. Suppose that there exists an E ∈ M such that f (x) = ∞
and m(E) > 0, where m is the Lebesgue measure. Then
Z 1 Z Z
f (x) log f (x) dx = f (x) log f (x) dx + f (x) log f (x) dx. (3.170)
0 E [0,1]\E
On the one hand, since f > 0 on [0, 1] \ E, we know from Proposition 1.24(a) that the second
integral on the right-hand side of the expression (3.170) is nonnegative. Since f (x) log f (x) = ∞
on E and m(E) > 0, we get from the expression (3.170) that
Z 1
f (x) log f (x) dx = ∞. (3.171)
0
in this case.
Next, we suppose that 0 < f (x) < ∞ a.e. on [0, 1]. Define ϕ : (0, ∞) → R and ψ : (0, ∞) → R
by
ϕ(x) = x log x and ψ(x) = − log x
respectively. Since ϕ′′ (x) = x1 > 0 and ψ ′′ (x) = x−2 > 0 on (0, ∞), they are convex on (0, ∞).
Since 0 < f (x) < ∞ a.e. on [0, 1] and m([0, 1]) = 1, we apply Theorem 3.3 (Jensen’s Inequality)
with ϕ to f to obtain
nZ 1 on Z 1 o Z 1
f (x) dx log f (x) dx ≤ f (x) log f (x) dx (3.173)
0 0 0
122 Chapter 3. Lp -Spaces
In the following problems, we use hx, yi to denote the inner product of the complex vectors x
and y and k · k the norm with respect to the Hilbert space H. Besides, we use span (S) to denote
the span of a set S.
Problem 4.1
Rudin Chapter 4 Exercise 1.
M = (M ⊥ )⊥ .
To see this, we recall from §4.8 that M is a closed subspace of H, so the first assertion implies
that
⊥
M = (M )⊥ . (4.1)
⊥
Let x ∈ M . By §4.9, we see that x ⊥ M so that x ⊥ M particularly. In other words, we have
⊥
M ⊆ M ⊥. (4.2)
⊥
For the other direction, suppose that x ∈ M ⊥ , we want to show that x ∈ M , i.e., x ⊥ M . To
this end, given y ∈ M , there exist a sequence {yn } ⊆ M such that yn → y as n → ∞. Since
123
124 Chapter 4. Elementary Hilbert Space Theory
Problem 4.2
Rudin Chapter 4 Exercise 2.
Proof. We note that this is actually the Gram-Schmidt Process. Suppose that n = 1. Then
the set {u1 } is clearly orthonormal and span (u1 ) = span (x1 ). Thus the statement is true for
n = 1.
Assume that the statement is true for n = k for some k ∈ N, i.e., {u1 , u2 , . . . , uk } is an
orthonormal set and
span (u1 , u2 , . . . , uk ) = span (x1 , x2 , . . . , xk ). (4.4)
Let n = k + 1. Note that xk+1 ∈ / span (x1 , x2 , . . . , xk ) because {x1 , x2 , . . .} is linearly inde-
pendent. By the assumption (4.4), it implies that xk+1 ∈ / span (u1 , u2 , . . . , uk ) so that vk+1 6= 0.
By this fact and Definition 4.1, for j = 1, 2, . . . , k, we derive that
vk+1
huk+1 , uj i = , uj
kvk+1 k
* k
+
1 X
= xk+1 − hxk+1 , ui i ui , uj
kvk+1 k
i=1
k
X
1 1
= hxk+1 , uj i − hxk+1 , ui i hui , uj i
kvk+1 k kvk+1 k
i=1
1 1
= hxk+1 , uj i − hxk+1 , uj i huj , uj i
kvk+1 k kvk+1 k
= 0.
Therefore, {u1 , u2 , . . . , uk+1 } is orthonormal. Next, by the assumption (4.4), we know that
x ∈ span (x1 , x2 , . . . , xk , xk+1 ) if and only if
Since
k
X vk+1
vk+1 = xk+1 − hxk+1 , ui i ui and uk+1 = ,
kvk+1 k
i=1
the xk+1 in the result (4.5) can be replaced by vk+1 and ultimately by uk+1 . Thus we have
Hence the statement is true for n = k + 1 if it is true for n = k. By induction, the construction
yields an orthonormal set {u1 , u2 , . . .} with
span (u1 , u2 , . . . , un ) = span (x1 , x2 , . . . , xn )
for all n ∈ N. This completes the proof of the problem.
Problem 4.3
Rudin Chapter 4 Exercise 3.
Proof. Recall the definition that a space is separable if it contains a countable dense subset.
Suppose that 1 ≤ p < ∞. Since T is compact, the second paragraph following Definition
3.16 says that Cc (T ) = C0 (T ) = C(T ). Thus we deduce from this and Theorem 3.14 that C(T )
is dense in Lp (T ) in the norm k · kp . Let P be the set of all trigonometric polynomials on T .
By Theorem 4.25 (The Weierstrass Approximation Theorem) and Definitions 4.23 (or Remark
3.15), P is dense in C(T ) in the norm k · k∞ . By Definition 3.7, we know that |f (x)| ≤ kf k∞
a.e. on T and then
n 1 Z π o1 n 1 Z π o1
p p
kf kp = |f (t)|p dt ≤ kf kp∞ dt = kf k∞ . (4.6)
2π −π 2π −π
In other words, the inequality (4.6) implies that P is also dense in C(T ) in the norm k · kp . By
Lemma 2.10, we conclude that P is dense in Lp (T ) in the norm k · kp .
Next we suppose that PQ is the set of all trigonometric polynomials with rational coefficients.
Note thata
n Xn o
PQ = P (t) = (ak + ibk )eikt a−n , . . . , an , b−n , . . . , bn ∈ Q and n ∈ N
k=−n
∞ n
[ n
X o
= P (t) = (ak + ibk )eikt a−n , . . . , an , b−n , . . . , bn ∈ Q
n=1 k=−n
[∞
∼ Q2(2n+1) . (4.7)
n=1
For each n ∈ N, since each Q2(2n+1) is countable, it follows from the set relation (4.7) that PQ
is also countable. Given f ∈ P. Since the set {p + iq | p, q ∈ Q} is dense in C and f must be in
the form
n
X
f (t) = ck eikt
k=−n
for some positive integer n, there exists a sequence {fnk } ⊆ PQ such that kfnk − f k∞ → 0 as
nk → ∞. By the inequality (4.6), it is true that
kfnk − f kp → 0
as nk → ∞. Consequently, PQ is dense in P in the norm k · kp and an application of Lemma 2.10
shows that PQ is dense in Lp (T ) in the norm k · kp . Since PQ is countable, Lp (T ) is separable.
By Definition 4.23, every function defined on T can be identified with a 2π-periodic function
on R, so we may identity L∞ (T ) with L∞ [0, 2π] . Consider the set
E = {fθ = χ[0,θ] | θ ∈ [0, 2π]}.
a
We write A ∼ B if there is a bijection between the two sets A and B, see [99, Definition 2.3, p. 25].
126 Chapter 4. Elementary Hilbert Space Theory
It is clear that χ[0,θ] ∈ L∞ [0, 2π] so that E ⊆ L∞ [0, 2π] . Since χ[0,θ1] 6= χ[0,θ2 ] if θ1 6= θ2 ,
E ∼ [0, 2π] and then it is uncountable. In fact, we have
if θ1 < θ2 .
Assume that L∞ [0, 2π] was separable. Let F = {fn } be a countable dense subset of
∞
L [0, 2π] . Then we must have
[ 1
E ⊆ L∞ [0, 2π] ⊆ B fn , , (4.9)
2
fn ∈F
a contradiction. Thus each B(fn , 12 ) contains at most one element of E. Since F is countable
but E is uncountable, we have
[ 1
E* B fn ,
2
fn ∈F
which definitely contradicts the set relation (4.9). Hence L∞ [0, π] is not separable and we
have completed the proof of the problem.
Problem 4.4
Rudin Chapter 4 Exercise 4.
Proof. Suppose that H is separable. By the definition, it has a countable dense subset {un }.
By Problem 4.2, we may assume that {un } is also an orthonormal set. Let {unk } be a maximal
linearly independent subset of {un }. Note that {unk } is at most countable. Since we have
span ({unk }) = {un }, span ({unk }) is also dense in H. By Theorem 4.18, the set {unk } is in fact
an at most countable maximal orthonormal set in H.
Conversely, we suppose that E = {un } is an at most countable maximal orthonormal set in
H. By Theorem 4.18 again, span (E) is dense in H. Let span Q (E) be the span of the set E
with rational coefficients. It is easy to see that
nX
n o
span Q (E) = (ak + ibk )uk a1 , . . . , an , b1 , . . . , bn ∈ Q and n ∈ N
k=1
∞ nX
[ n o
= (ak + ibk )uk a1 , . . . , an , b1 , . . . , bn ∈ Q
n=1 k=1
[∞
∼ Q2n .
n=1
As a result, span Q (E) must be countable. Next, since Q is dense in R, we conclude that
span Q (E) is dense in span (E). By Lemma 2.10, span Q (E) is also dense in H and hence it is
separable. This completes the proof of the problem.
4.1. Basic Properties of Hilbert Spaces 127
Problem 4.5
Rudin Chapter 4 Exercise 5.
L(x) = hx, zi
for all x ∈ H. This z must be nonzero, otherwise we have L = 0 which contradicts our
assumption. It is clear that
Since hx, αzi = α hx, zi, the two expressions (4.10) and (4.11) give
⊥
M = span (z)
and then ⊥ ⊥
M⊥ = span (z) . (4.12)
We claim that span (z) is a closed subspace of H. In fact, it is a subspace of H follows easily
from the definition of span (z). Suppose that {xn } is a sequence in span (z). Then we know that
each xn has the form xn = αn z, where αn ∈ C. Let α = lim αn and x = αz. We want to show
n→∞
that xn → x in H with respect to the norm k · k. Indeed, we have
lim xn = x
n→∞
as desired. In conclusion, span (z) contains all its limit points and therefore it is closed in H.
⊥ ⊥
By Problem 4.1, we have span (z) = span (z). Combining this fact and the expression
(4.12), we have
M ⊥ = span (z)
and since {z} is a basis of span (z), we have dim span (z) = 1 and our required result follows.
We end the proof of the problem.
Problem 4.6
Rudin Chapter 4 Exercise 6.
128 Chapter 4. Elementary Hilbert Space Theory
• E = {un } is closed and bounded, but not compact. Since E is orthonormal, kun k ≤ 1
for every n ∈ N. Thus E is bounded by 1. If n 6= m, then it follows from Definition 4.1
that
kun − um k2 = hun − um , un − um i = hun , un i + hum , um i = 2. (4.13)
Assume that u was a limit point√ of E. Then there exists a positive integer N such that
n ≥ N implies that kun − uk < 22 . If n, m ≥ N , then we have
√
kun − um k ≤ kun − uk + ku − xm k < 2
which contradicts the result (4.13). In other words, E has no limit point and then it is
closed in H.
Suppose that B(un , 1) = {x ∈ H | kun −xk < 1} which is clearly open in H. Furthermore,
it is obvious that {B(un , 1)} is an open cover of E. By the fact (4.13), we conclude that
each B(un , 1) contains exactly one element of E, so it is impossible to have
k
[
E⊆ B(uni , 1)
i=1
where k = 1, 2, . . .. Note that we have |ckn | ≤ δn for all k ∈ N. Since {ck1 } ⊆ B(0, δ1 ) ⊆ C
for all k ∈ N and B(0, δ1 ) is compact, B(0, δ1 ) is sequentially compact and then there
exists a subsequence k1 (m) such that
k (m)
c11 → c1 ∈ B(0, δ1 )
k (m)
as m → ∞. Next, we consider the sequence {c21 } ⊆ B(0, δ2 ). Then we can find a
subsequence {k2 (m)} ⊆ {k1 (m)} such that
k (m)
c22 → c2 ∈ B(0, δ2 )
We claim that the sequence {xkm (m) }, which is clearly a subsequence of {xk } converges
to the point x. To this end, given ǫ > 0, there is a N ′ ∈ N such that
X ǫ
δn2 < . (4.16)
′
8
n>N
Then we obtain from the definition (4.14), the fact |ckn | ≤ δn for all k and the estimate
(4.16) that
X∞
2
kxkm (m) − xk2 =
[cknm (m) − cn ]un
n=1
∞
X
= |cknm (m) − cn |2
n=1
N ′ ∞
X X
= |cknm (m) − cn |2 + |cnkm (m) − cn |2
n=1 n=N ′ +1
N ′ ∞
X X
≤ |cknm (m) − cn |2 + (|cknm (m) |2 + 2|cknm (m) | · |cn | + |cn |2 )
n=1 n=N ′ +1
N′
X ∞
X
≤ |cknm (m) − cn |2 + 4 × δn2
n=1 n=N ′ +1
N ′
X ǫ
< |cknm (m) − cn |2 + . (4.17)
2
n=1
To finish our proof, we have to find an upper bound of the summation in the inequality
(4.17). Recall from the limit (4.15) that for each n ≤ N ′
k ′ (m)
cnN → cn ∈ B(0, δn )
N
X
where N ∈ N. Then we have kxN k2 = δn2 . Since S is a compact metric space, it is
n=1
bounded. Thus there exists a positive constant M such that
N
X
δn2 ≤ M
n=1
∞
X
for all N ∈ N and hence δn2 < ∞.
n=1
• H is not locally compact. Recall that H is itself a vector space, so it contains the zero
element 0. Assume that H was locally compact. Then the closure
B(0, r) = {x ∈ H | kxk ≤ r}
must be compact for every r > 0. Note that {run } ⊆ B(0, r). Since Er = {run } is closed
in H, Theorem 2.4 implies that Er is compact. Similar to part (a), {B(run , r)} is an open
cover of Er . A direct computation shows that
√
krun − rum k = 2r. (4.19)
which contradicts the fact (4.19). Therefore, each B(run , r) contains exactly one element
of Er so that {B(run , r)} does not have a finite subcover for Er . In other words, Er is not
compact, a contradiction. Hence we establish the fact that H is not locally compact.
Problem 4.7
Rudin Chapter 4 Exercise 7.
P
Proof. We follow the suggestion given by Rudin. Assume that a2n = ∞. Therefore, for every
k ∈ N, there are disjoint sets E1 , E2 , . . . such that
X
a2n > k > 0. (4.20)
n∈Ek
P
which is divergent, a contradiction. Hence we must have a2n < ∞ and it completes the proof
of the problem.
Problem 4.8
Rudin Chapter 4 Exercise 8.
Proof. By Problem 4.2, H contains an orthonormal set A and then Theorem 4.22 (The Hausdorff
Maximality Theorem) implies that A is contained in a maximal orthonormal set in H. Suppose
that {uα | α ∈ A} and {vβ | β ∈ B} are maximal orthonormal sets in H1 and H2 respectively.
Without loss of generality, we may assume that |A| ≤ |B|. Then we take a subset B ′ of B which
is of the same cardinality as A. Now we consider
H = span ({vβ | β ∈ B ′ }) ⊆ H2 .
By Definition 4.13, span ({vβ | β ∈ B ′ }) is a subspace of H2 . Thus we follow from the last
paragraph in §4.7 that H is a closed subspace of H2 . By the given hint, H is also Hilbert.
Finally, we know from §4.19 that H1 and H are Hilbert space isomorphic to ℓ2 (A) and ℓ2 (B ′ )
respectively. Since A and B have the same cardinality, ℓ2 (A) is Hilbert space isomorphic to
ℓ2 (B ′ ). In other words, H1 is Hilbert space isomorphic to H ⊆ H2 , completing the proof of the
problem.
Problem 4.9
Rudin Chapter 4 Exercise 9.
X Z
2π
hun , χA i 2 = kχA k2 = |χA (x)|2 dx = m(A) < ∞.
n∈Z 0
and Z Z π
lim sin nt dt = lim (sin nt) · χA (t) dt = 2π lim hsin nx, χA i = 0.
n→∞ A n→∞ −π n→∞
Problem 4.10
Rudin Chapter 4 Exercise 10.
Problem 4.11
Rudin Chapter 4 Exercise 11.
n+1
Proof. For each n = 1, 2, . . ., we define xn = n un and E = {xn } which is obviously a subset
of L2 (T ) because, by Definition 4.23,
n+1
kxn k = kxn k2 = <∞
n
for every n ∈ N. It is clear that E 6= ∅. Now we are going to apply a similar argument as in
the proof of the first assertion in Problem 4.6. In fact, we note that hxn , xm i = 0 for n 6= m, so
we follow from Definition 4.1 that
kxm − xn k2 = hxm − xn , xm − xn i
= hxm , xm − xn i − hxn , xm − xn i
= hxm , xm i − hxm , xn i − hxn , xm i + hxn , xn i
m + 1 2 n + 1 2
= hum , um i + hun , un i
m n
m + 1 2 n + 1 2
= +
m n
≥2
which contradicts the result (4.23). Thus E has no limit point and hence it is closed in L2 (T ).
Furthermore, assume that xN was an element of E such that kxN k ≤ kxn k for all xn ∈ E.
However, since we have
N +1 N +2
kxN k = > = kxN +1 k,
N N +1
this contradiction shows that E has no element of smallest norm. Hence we end the proof of
the problem.
Remark 4.1
Problem 4.11 indicates that the condition “convexity” in Theorem 4.10 cannot be omitted.
Problem 4.12
Rudin Chapter 4 Exercise 12.
If we substitute y = z 2 and use the integral form of the beta function (see [99, Theorem 8.20,
p. 193]), then the expression (4.25) can be written as
Z
ck 1 1 1 ck Γ(k + 21 )Γ( 12 ) ck Γ(k + 21 )
1= (1 − y)k− 2 y − 2 dy = · =√ · . (4.26)
π 0 π Γ(k + 1) π kΓ(k)
Γ(k + c)
lim =1 (4.27)
k→∞ kc Γ(k)
for every real constant c. Take c = 12 in the limit (4.27) and then substitute the corresponding
result into the expression (4.26), we get
1
1 √ k 2 Γ(k) √
lim k− 2 ck = lim π· = π.
k→∞ k→∞ Γ(k + 12 )
Problem 4.13
Rudin Chapter 4 Exercise 13.
134 Chapter 4. Elementary Hilbert Space Theory
Proof. We follow the hint. Consider the function f (t) = e2πikt , where k = 0, ±1, ±2, . . .. When
k = 0, we have f ≡ 1 and then the formula holds trivially. Suppose that k 6= 0. Since α is
/ Z and then e2πikα 6= 1. By this, we have
irrational, kα ∈
N N
1 X 2πiknα 1 X 2πikα n e2πikα e2πikN α − 1
e = (e ) = · 2πikα
N n=1 N n=1 N e −1
so that
1 X
N 2
lim e2πiknα ≤ lim = 0. (4.28)
N →∞ N N →∞ N |e2πikα − 1|
n=1
On the other hand, we have Z 1
e2πikt 1
e2πikt dt = = 0. (4.29)
0 2πik 0
Hence we deduce from the inequality (4.28) and the integral (4.29) that
N Z 1
1 X 2πiknα
lim e = e2πikt dt (4.30)
N →∞ N 0
n=1
for every k = 0, ±1, ±2, . . .. Therefore the formula (4.30) holds for every trigonometric polyno-
mials in the form
N
X
P (t) = ck e2πikt , (4.31)
k=−N
where c−N , . . . , cN ∈ C.
Suppose that f : R → R is continuous and f (x + 1) = f (x) for every x ∈ R. By Theorem
4.25 (The Weierstrass Approximation Theorem), for every ǫ > 0, there exists a trigonometric
polynomial P in the form (4.31) such that
ǫ
|f (t) − P (t)| < (4.32)
3
for every t ∈ R. Next, the analysis in the previous paragraph shows that there is a positive
integer N such that n ≥ N implies
1 X N Z 1 ǫ
P (nα) − P (t) dt < . (4.33)
N n=1 0 3
Hence we follow from the estimates (4.32) and (4.33) that n ≥ N implies
1 X N Z 1 1 X N N N
1 X 1 X
f (nα) − f (t) dt = f (nα) − P (nα) + P (nα)
N 0 N N N
n=1 n=1 n=1 n=1
Z 1 Z 1 Z 1
− P (t) dt + P (t) dt − f (t) dt
0 0 0
Z 1
1 X 1 X
N N
≤ f (nα) − P (nα) + P (nα) − P (t) dt
N N 0
n=1 n=1
Z 1
+ |f (t) − P (t)| dt
0
< ǫ.
Hence the formula (4.30) holds for every continuous function on R with period 1, completing
the proof of the problem.
4.2. Application of Theorem 4.14 135
Remark 4.2
A question similar to Problem 4.13 is also proven in [124, Problem 8.19, pp. 194 – 196].
In fact, Problem 4.13 is a very important tool in the study of the Weyl’s Equidistribution
Theorem, see [110, §4.2, pp. 105 – 113].
Problem 4.14
Rudin Chapter 4 Exercise 14.
Proof. Let H be the Hilbert space L2 [−1, 1] . Define the inner product of f, g ∈ H by
Z 1
hf, gi = f (x)g(x) dx. (4.34)
−1
It is clear that xn ∈ H for every n = 0, 1, 2, . . . and the set {1, x, x2 , . . .} is a linearly independent
set of vectors in H. By Problem 4.2, we can obtain an orthonormal set {u1 (x), u2 (x), . . .} such
that {1, x, . . . , xN −1 } and {u1 (x), u2 (x), . . . , uN (x)} have the same span for all positive integers
N . Next, we let F = {0, 1, 2} and we consider the closed subspace
of H. Then the orthonormal set {u1 (x), u2 (x), u3 (x)} obtained by {1, x, x2 } is given by
r r
1 3 3 5 2 1
u1 (x) = √ , u2 (x) = x and u3 (x) = x − .
2 2 2 2 3
Let x3 ∈ H. Define
sF (x3 ) = x3 , u1 (x) · u1 (x) + x3 , u2 (x) · u2 (x) + x3 , u3 (x) · u3 (x).
q
Since x3 , u1 (x) = 0, x3 , u2 (x) = 52 32 and x3 , u3 (x) = 0, we have
3
sF (x3 ) = x
5
and then r
hZ 1 3 2 i 12
3 3
3 8
x − x
= x − x dx = . (4.37)
5 −1 5 175
Combining the results (4.36) and (4.37), it yields that
Z 1
8
min |x3 − a − bx − cx2 |2 dx = . (4.38)
a,b,c −1 175
136 Chapter 4. Elementary Hilbert Space Theory
For the second assertion, we first notice that the definite integral must be a real number
because we are considering its maximum. Secondly, the conditions on g actually mean that
h1, gi = hx, gi = x2 , g = 0 and kgk = 1.
Therefore, we combine the expressions (4.39) and (4.40) and then using the value (4.38) to get
Z 1 r
8
max x3 g(x) dx ≤ . (4.41)
g∈MF⊥ −1 175
kgk=1
q
175 3 − 3 x , then we have
If g(x) = 8 x 5
Z r
1
8
x3 g(x) dx =
−1 175
so that the equality in the inequality (4.41) is attainable. Hence we have completed the proof
of the problem.
Problem 4.15
Rudin Chapter 4 Exercise 15.
Proof. Now we work on the space H = L2 [0, ∞] and it is easy to check that the formula given
by Z ∞
hf, gi = f ge−x dx (4.42)
0
satisfies Definition 4.1, so it is indeed an inner product for the space H.
Now we follow the idea of proof in Problem 4.14. Evidently, we have xn ∈ H for every
n = 0, 1, 2, . . . and the set {1, x, x2 , . . .} is a linearly independent set of vectors in H. By
Problem 4.2, we can obtain an orthonormal set {u1 (x), u2 (x), . . .} such that {1, x, . . . , xN −1 }
and {u1 (x), u2 (x), . . . , uN (x)} have the same span for all positive integers N . Next, we let
F = {0, 1, 2} and we consider the closed subspace
MF = span ({1, x, x2 })
of H. Then the orthonormal set {u1 (x), u2 (x), u3 (x)} obtained by {1, x, x2 } is given by
1
u1 (x) = 1, u2 (x) = x − 1 and u3 (x) = (x2 − 4x + 2).
2
d
We assume its truth here.
4.3. Miscellaneous Problems 137
Let x3 ∈ H. Define
sF (x3 ) = x3 , u1 (x) · u1 (x) + x3 , u2 (x) · u2 (x) + x3 , u3 (x) · u3 (x).
so that
hZ 1 i1
2
kx3 − 9x2 + 18x − 6k = (x3 − 9x2 + 18x − 6)2 e−x dx = 6. (4.44)
−1
For the second assertion, we observe again that the definite integral is a real number and the
conditions on g imply that
g ∈ MF⊥ and kgk = 1.
Therefore, it yields that
Z 1 Z 1
max x3 g(x)e−x dx ≤ max x3 g(x)e−x dx
g∈MF ⊥ ⊥ g∈MF
−1 −1
kgk=1 kgk=1
By combining the expressions (4.46) and (4.47) and then using the value (4.45), we gain
Z 1
max x3 g(x)e−x dx ≤ 6. (4.48)
g∈MF⊥ −1
kgk=1
so that the equality in the inequality (4.48) is attainable and we end the proof of the problem.
Problem 4.16
Rudin Chapter 4 Exercise 16.
138 Chapter 4. Elementary Hilbert Space Theory
Proof. Since M is a closed linear subspace of H, we follow from Theorem 4.11(a) that x0 has
the unique decomposition
x0 = P x0 + Qx0 ,
where P x0 ∈ M and Qx0 ∈ M ⊥ . In addition, Theorem 4.11(b) ensures that
kP x0 − x0 k = min{kx − x0 k | x ∈ M }. (4.49)
If Qx0 = 0, then it means that x0 = P x0 ∈ M . In this case, we deduce from the expression
(4.49) and the inequality (4.51) that
Hence our desired result follows from the comparison between the expressions (4.49) and (4.52)
and the fact that kP x0 − x0 k = kQx0 k. We have ended the proof of the problem.
Problem 4.17
Rudin Chapter 4 Exercise 17.
Proof. Take H = L2 [0, 1] which is a Hilbert space with the inner product given by
Z 1
hf, gi = f (t)g(t) dt, (4.53)
0
where f, g ∈ H, see Example 4.5(b). For t ∈ [0, 1], we note that χ[0,t] ∈ H because
nZ 1 o1 nZ t o1 √
2 2 2
kχ[0,t] k = |χ[0,t] (x)| dx = dx = t ≤ 1.
0 0
γ(t) = χ[0,t]
4.3. Miscellaneous Problems 139
which is well-defined. It is trivial that γ is injective. Furthermore, for every p ∈ [0, 1], if t > p,
then we have
nZ 1 o1 n Z t o1 1
2 2 2
kγ(t) − γ(p)k = |χ(p,t] (x)| dx = dx = (t − p) 2 . (4.54)
0 p
Similarly, if c = d, then hγ(b) − γ(a), γ(d) − γ(c)i = 0. Without loss of generality, we may
assume that a < b and c < d. Then (a, b] ∩ (c, d] = ∅ and so we have
Z 1
hγ(b) − γ(a), γ(d) − γ(c)i = χ(a,b] , χ(c,d] = χ(a,b] (x)χ(c,d] (x) dx = 0.
0
Hence the desired result holds in the special case that H = L2 [0, 1] .
To treat the general case, we first consider the function un (t) = e2πint for every n ∈ Z. It is
clear that un ∈ H and
Z 1 1, if n = m;
2πi(n−m)t
hun , um i = e dt =
0
0, if n 6= m.
Thus {un | n ∈ Z} is an orthonormal set in H. By Theorem 3.14, C [0, 1] is dense in H with
respect to the norm induced by the inner product (4.53). By Theorem 4.25 (The Weierstrass
Approximation Theorem) and Definition 4.23, we know that the set P of all trigonometric
polynomials in the form
XN N
X
P (t) = cn un (t) = cn e2πint
n=−N n=−N
is dense in C [0, 1] with respect to the norm k · k∞ . Note that P = span ({un | n ∈ Z}). By
Definition 3.7, it can be shown easily that P is also dense in C [0, 1] with respect to the norm
induced by the inner product (4.53). Therefore, we derive from Lemma 2.10 that P is also dense
in H in the norm induced by the inner product (4.53). Consequently, we conclude from Theorem
4.18(i) that {un | n ∈ Z} is in fact maximal and thene
H∼
= ℓ2 (N). (4.56)
Suppose that H ′ is a Hilbert space with an infinite maximal orthonormal set {vα | α ∈ A},
where A is countable or uncountable infinite. Then we have
H′ ∼
= ℓ2 (A). (4.57)
e
We say A ∼
= B if A is Hilbert space isomorphic to B.
140 Chapter 4. Elementary Hilbert Space Theory
Applying Problem 4.8 to the Hilbert space isomorphisms (4.56) and (4.57), we conclude that H
must be Hilbert space isomorphic to a subspace of H ′ and then the special case is applicable to
this general case. Indeed, let Λ : H → K be a Hilbert space isomorphism, where K is a Hilbert
subspace of H ′ . If we define λ : [0, 1] → H ′ by
λ = Λ ◦ γ,
then λ is injective and continuous. Furthermore, since hΛ(f ), Λ(g)i = hf, gi for all f, g ∈ H, we
deduce from the special case above that
hλ(b) − λ(a), λ(d) − λ(c)i = Λ γ(b) − Λ γ(a) , Λ γ(d) − Λ γ(c)
= Λ γ(b) , Λ γ(d) − Λ γ(b) , Λ γ(c)
− Λ γ(a) , Λ γ(d) + Λ γ(a) , Λ γ(c)
= hγ(b), γ(d)i − hγ(b), γ(c)i − hγ(a), γ(d)i + hγ(a), γ(c)i
= hγ(b) − γ(a), γ(c) − γ(d)i
= 0.
Problem 4.18
Rudin Chapter 4 Exercise 18.
Therefore, we obtain
Z A Xn X
m Z AX
n X
m
1 1
lim cp eisp t dq e−irq t dt = lim cp dq ei(sp −rq )t dt
A→∞ 2A −A A→∞ 2A −A
p=1 q=1 p=1 q=1
which is undoubtedly finite by the values (4.58) and (4.59). As a consequence, we have shown
that hf, gi is well-defined for all f, g ∈ X.
Next we are going to show that h · i satisfies Definition 4.1. Firstly, it is clear that
Z A Z A
1 1
hf, gi = lim f (t)g(t) dt = lim g(t)f (t) dt = hg, f i.
A→∞ 2A −A A→∞ 2A −A
Z A Z A
1 1
= lim f (t)h(t) dt + lim g(t)h(t) dt
A→∞ 2A −A A→∞ 2A −A
= hf, hi + hg, hi .
Fourthly, if f takes the representation (4.60), then we establish from the results (4.58) and (4.59)
that
Z A Z A n
X
1 1
hf, f i = lim f (t)f (t) dt = lim |f (t)|2 dt = |cp |2 ≥ 0. (4.61)
A→∞ 2A −A A→∞ 2A −A
p=1
H∼
= ℓ2 (N). (4.63)
Now the two isomorphic relations (4.62) and (4.63) imply that ℓ2 (R) ∼
= ℓ2 (N), but then R
and N have the same cardinal number, a contradiction. Hence we complete the proof of the
problem.
Problem 4.19
Rudin Chapter 4 Exercise 19.
Proof. If N = 1, then ω = e2πi = 1 and k can only take the value 0. Thus the orthogonality
relation holds in this special case. Suppose that N > 1. If k = 0, then ω 0 = 1 so that
N
1 X 0
ω = 1.
N
n=1
ω k(N −1) + · · · + ω k + 1 = 0
142 Chapter 4. Elementary Hilbert Space Theory
or equivalently
ω kN + ω k(N −1) + · · · + ω k = 0.
In conclusion, we have
1, if k = 0;
1 X nk
N
ω =
N
n=1 0, if 1 ≤ k ≤ N − 1.
Let N ≥ 3. It is clear from the orthogonality relations and the properties in Definition 4.1
that
N N
1 X 1 X
kx + ω n yk2 ω n = hx + ω n y, x + ω n yi ω n
N N
n=1 n=1
1 hX
N N
X N
X
= hx, xi ω n + hx, ω n yi ω n + hω n y, xi ω n
N
n=1 n=1 n=1
N
X i
+ hω n y, ω n yi ω n
n=1
1h X NX N X N
= hx, xi ωn + hx, yi ω −n × ω n + hy, xi ω n × ω n
N
n=1 n=1 n=1
N
X i
+ hy, yi ωn
n=1
1h i
XN N
X N
X
= hx, xi ω n + N hx, yi + hy, xi ω 2n + hy, yi ωn
N
n=1 n=1 n=1
= hx, yi (4.66)
as desired.
Finally, Definition 4.1 gives
D E
kx + eiθ yk2 = x + eiθ y, x + eiθ y = hx, xi + e−iθ hx, yi + eiθ hy, xi + hy, yi
so that
Z π Z π
1 2 iθ 1
iθ
kx + e yk e dθ = kxk2 + kyk2 + e−iθ hx, yi + eiθ hy, xi eiθ dθ
2π −π 2π
Z−π
π Z π
1 1
= hx, yi dθ + e2iθ hy, xi dθ
2π −π 2π −π
= hx, yi . (4.67)
Remark 4.3
We note that the polarization identity (see [100, p. 86]) can be written in the form
3
1 X −n
hx, yi = i kx + in yk2 .
4
n=0
Thus the two identities (4.66) and (4.67) in Problem 4.19 are actually generalizations of this.
CHAPTER 5
Examples of Banach Space Techniques
Problem 5.1
Rudin Chapter 5 Exercise 1.
B = {f ∈ Lp (µ) | kf kp ≤ 1}.
Therefore, it follows from the inequality (5.1) that the unit ball B is a circle if and only if
p = 2. Besides, the unit ball B becomes a square if and only if p = 1.
which is a square.
143
144 Chapter 5. Examples of Banach Space Techniques
which cannot be a circle or a square for any p ∈ (0, ∞). However, we note that the expression
(5.2) is still valid even in the case µ({a}) 6= µ({b}), so we have the inequality (5.3) and then B
is a square when p = ∞. This ends the analysis of the problem.
Problem 5.2
Rudin Chapter 5 Exercise 2.
Proof. Let X be a normed linear space with norm k · k and B(0, 1) = {x ∈ X | kxk < 1} be the
open unit ball. For every x, y ∈ B(0, 1) and t ∈ [0, 1], we follow from Definition 5.2 that
so that (1 − t)x + ty ∈ B(0, 1). By §4.8, B(0, 1) is convex. Similarly, the convexity of the closed
unit ball B(0, 1) = {x ∈ X | kxk ≤ 1} can be proven similarly. Hence we complete the proof of
the problem.
Problem 5.3
Rudin Chapter 5 Exercise 3.
Z Z
1
kgk1 = |g(x)| dx = χF (x) dx = 1,
µ(F ) F
f + g
ZX Z Z
f (x) + g(x) 1 1
= dx = f (x) dx + g(x) dx = 1.
2 1 X 2 2 E 2 F
These show that the invalidity of strictly convexity in L1 (µ). For L∞ (µ), we replace the functions
f and g in the definition (5.4) by
f (x) = χE (x) and g(x) = χE∪F (x)
respectively. Then it is also true that f 6= g. By Definition 3.7, we have
f + g
kf k∞ = kgk∞ =
= 1.
2 ∞
∞
Therefore, the strictly convexity does not hold in L (µ).
Finally, we consider the space C(X), where X is Hausdorff. By Definition 3.16, we see that
X is assumed to be compact and so the norm k · k∞ of C(X) is given by
kf k∞ = sup |f (x)|. (5.5)
x∈X
Take f ∈ C(X) and suppose that f is nonconstant. Thus one can find a x0 ∈ X such that
|f (x0 )| < kf k∞ . (5.6)
Otherwise, we have |f (x)| ≥ kf k∞ on X, but this and the definition (5.5) will imply that
|f (x)| = kf k∞ on X, i.e., f is a constant which is a contradiction. Define
3 1 1 3
a = kf k∞ + |f (x0 )| and b = kf k∞ + |f (x0 )|.
4 4 4 4
Consider the sets
n o n o
K = x ∈ X |f (x)| ≥ a = |f |−1 [a, ∞) and E = x ∈ X |f (x)| ≤ b = |f |−1 (−∞, b] .
and
f + g
1 1 1
= kf + hf k∞ = sup [1 + h(x)]f (x) = |1 + h(p)| · |f (p)| = |f (p)| = kf k∞ .
2 ∞ 2 2 x∈X 2
Hence strictly convexity does not hold for C(X), completing the proof of the problem.
146 Chapter 5. Examples of Banach Space Techniques
Problem 5.4
Rudin Chapter 5 Exercise 4.
Then it is easy to see that f ∈ M , i.e., M is nonempty. Suppose that f, g ∈ M , t ∈ [0, 1] and
h = (1 − t)f + tg. Then we have
Z 1 Z 1 Z 1 Z 1
2 2
h(x) dx − h(x) dx = [(1 − t)f (x) + tg(x)] dx − [(1 − t)f (x) + tg(x)] dx
1 1
0 2
0 2
hZ 1 Z 1 i
= (1 − t) f (x) dx − f (x) dx
1
0 2
hZ 1 Z 1 i
+t g(x) dx − g(x) dx
1
0 2
=1
as n → ∞. Then it follows froma [99, Theorem 7.9, p. 148] that fn → f uniformly on [0, 1].
Since each fn is continuous on [0, 1], f is also continuous on [0, 1]. Furthermore, we have
hZ 1
2
Z 1 i
lim fn (t) dt − fn (t) dt = 1
n→∞ 0 1
2
Z 1 Z 1
2
lim fn (t) dt − lim fn (t) dt = 1
0 n→∞ 1 n→∞
2
Z 1 Z 1
2
f (t) dt − f (t) dt = 1.
1
0 2
a
Or we can see the rephrased Theorem 7.9 on p. 151 in [99].
5.2. Failure of Theorem 4.10 and Norm-preserving Extensions 147
for every f ∈ M . Assume that there was a f ∈ M such that kf k∞ = 1. Then we deduce from
the inequality (5.8) that
Z 1 Z 1
|f (t)| dt = 1 or (1 − |f (t)|) dt = 0. (5.9)
0 0
Since kf k∞ = 1, we have |f (x)| ∈ [0, 1] for all x ∈ [0, 1] so that the function g(x) = 1 − |f (x)| is
continuous and nonnegative on [0, 1]. Applying Theorem 1.39(a) to the second integral in (5.9),
we obtain
g(x) = 0 a.e. on [0, 1].
Now the continuity of g forces that g(x) = 0 on [0, 1], i.e., |f (x)| = 1 on [0, 1]. In particular, we
have
−1 ≤ Re f (x) ≤ 1 (5.10)
on [0, 1].
On the other hand, since f ∈ M , the definition gives
Z 1
2
Z 1 Z 1
2
Z 1
1 = Re f (t) dt − f (t) dt = Re f (t) dt − Re f (t) dt
1 1
0 2
0 2
or equivalently,
Z 1 Z 1
2
Re f (t) − 1 dt + −1 − Re f (t) dt = 0. (5.11)
1
0 2
Observe from the inequalities (5.10) that Re f (x) − 1 ≤ 0 and −1 − Re f (x) ≤ 0 on
[0, 1].
Therefore, the equation (5.11) and the continuities of Re f (x) − 1 and −1 − Re f (x) imply
that
1, if [0, 12 ];
Re f (x) =
−1, if [ 21 , 1].
However, this says that Re f (x) is discontinuous at x = 21 , a contradiction. Hence no such f
exists and this completes the proof of the problem.
Problem 5.5
Rudin Chapter 5 Exercise 5.
Proof. It is clear that 1 ∈ M , so M 6= ∅. Suppose that f, g ∈ M ⊆ L1 [0, 1] , t ∈ [0, 1] and
h = (1 − t)f + tg. Then we have
Z 1 Z 1 Z 1
|h(x)| dx ≤ (1 − t) |f (x)| dx + t |g(x)| dx < ∞
0 0 0
which means that h ∈ L1 [0, 1] . Besides, we also have
Z 1 Z 1 Z 1 Z 1
h(x) dx = [(1 − t)f (x) + tg(x)] dx = (1 − t) f (x) dx + t g(x) dx = 1
0 0 0 0
so that h ∈ M , i.e., M is convex. Next, suppose that {fn } ⊆ M and there exists a function f on
[0, 1] such that kfn − f k1 → 0 as n → ∞. Thus fn − f ∈ L1 [0, 1] and it yields from Theorem
1.33 that Z 1 Z 1 Z 1
fn (t) dt − f (t) dt ≤ |fn (t) − f (t)| dt = kfn − f k1
0 0 0
148 Chapter 5. Examples of Banach Space Techniques
Problem 5.6
Rudin Chapter 5 Exercise 6.
for all x ∈ M . Recall from Definition 5.3 that kf k is the smallest number such that
|f (x)| ≤ kf k · kxk
for every x ∈ M . In particular, we must have kf (x0 )k ≤ kf k · kx0 k. On the other hand, the
representation (5.12) gives
so we must have
kx0 k = kf k. (5.13)
F (x) = hx, x0 i
b
In fact, such fe is uniformly continuous on M . See also [74, Exercise 2, p. 270].
c
Let U and W be two vector subspaces of the vector space V . Then V is said to be the direct sum of U and
W , denoted by V = U ⊕ W , if V = U + W and U ∩ W . See [7, pp. 95, 96].
5.2. Failure of Theorem 4.10 and Norm-preserving Extensions 149
Furthermore, by the same analysis in obtaining the representation (5.13), we get kF k = kx0 k
and then using the representation (5.13) again, we conclude that
kF k = kf k = kx0 k, (5.14)
F ′ (x) = hx, x1 i
By putting the numbers (5.14) and (5.15) into the formula (5.16), we see that
kf k2 = kf k2 + kx0 − x1 k2
and this reduces to kx0 − x1 k2 = 0. By Definition 4.1, we must have x0 = x1 , i.e., F ′ = F . This
proves the uniqueness part and so we have completed the proof of the problem.
Problem 5.7
Rudin Chapter 5 Exercise 7.
Proof. Consider
X = [−1, 1] with µ = m, the Lebesgue measure. Then the vector space
L1 [−1, 1] consists of all complex measurable function on [−1, 1] such that
Z 1
kf k1 = |f (x)| dx < ∞.
−1
We define
M = {f ∈ L1 [−1, 1] | f is real-valued and f (x) = 0 on [−1, 0]}.
which is linear. By Definition 5.3, the definition (5.17) and the fact that χ[0,1] ∈ M , we have
Thus Λ is bounded.
Let δ ∈ (0, 1). If we define Λδ : L1 [−1, 1] → C by
Z 1 Z 1
Λδ (f ) = Re [f (x)]χ(−δ,1) (x) dx = Re [f (x)] dx,
−1 −δ
then it satisfies Definition 2.1 so that Λδ is linear on L1 [−1, 1] . It is also clear that Λδ |M = Λ.
A direct computation also shows that
Z 1 Z 1 Z 1
|Λδ (f )| = Re [f (x)] dx ≤ |Re [f (x)]| dx ≤ |f (x)| dx
−δ −δ −1
Problem 5.8
Rudin Chapter 5 Exercise 8.
Proof.
(a) We first show that X ∗ is a normed linear space by checking Definition 5.2. For all f, g ∈ X ∗ ,
since |f (x) + g(x)| ≤ |f (x)| + |g(x)| for all x ∈ X, it must be true that
kf + gk ≤ kf k + kgk.
Next, for a scalar α and f ∈ X ∗ , we have |αf (x)| = |α| · |f (x)| for all x ∈ X which implies
that
kαf k = |α| · kf k.
Finally, suppose that f ∈ X ∗ is such that kf k = 0. By [100, Eqn. (3), p. 96], we have
|f (x)| ≤ kf k · kxk = 0
kΛn − Λm k < ǫ
5.3. The Dual Space of X 151
and for arbitrary but fixed x ∈ X, this and [100, Eqn. (3), p. 96] together imply that
|Λn (x) − Λm (x)| = |(Λn − Λm )(x)| ≤ kΛn − Λm k · kxk < ǫkxk. (5.18)
Thus {Λn (x)} ⊆ C is Cauchy. Since C is complete, {Λn (x)} converges to a point in C and
it is reasonable to define the pointwise limit Λ : X → C by
Now our target is to show that Λ ∈ X ∗ and {Λn } converges to Λ in X ∗ , i.e., kΛn −Λk → 0
as n → ∞. Let α, β ∈ C and x, y ∈ X. Since X is a normed linear space, αx + βy ∈ X
and thus
kΛn k ≤ M (5.19)
for every n = 1, 2, . . .. Therefore, it follows from the triangle inequality, then [100, Eqn.
(3), p. 96] and finally the inequalities (5.19) that
for all n > N and for all x ∈ X. Recall that if we take m → ∞ in the inequality (5.18),
then we have
|Λn (x) − Λ(x)| ≤ ǫkxk (5.21)
for all n > N . Thus, by putting the inequality (5.19) into the inequality (5.20), we get
|Λ(x)| ≤ (M + ǫ)kxk
for every x ∈ X. Hence we obtain kΛk < ∞, i.e., Λ ∈ X ∗ . Since the inequality (5.21) is
true for all x ∈ X, it must be true for those x with kxk ≤ 1 and so n > N implies that
Λ∗x (f ) = f (x).
If x = 0, then it is clear that Λ∗0 (f ) = f (0) = 0 on X ∗ . Therefore, Λ∗0 must be linear and of
norm 0. Without loss of generality, we may assume that x 6= 0 in the following discussion.
For any f, g ∈ X ∗ and α, β ∈ C, we have
Λ∗x (αf + βg) = (αf + βg)(x) = αf (x) + βg(x) = αΛ∗x (f ) + βΛ∗x (g).
152 Chapter 5. Examples of Banach Space Techniques
By Definition 2.1, Λ∗x is a linear functional on X ∗ . On the one hand, by [100, Eqn. (3), p.
96], we know that
|Λ∗x (f )| = |f (x)| ≤ kf k · kxk ≤ kxk (5.22)
for all f ∈ X ∗ with kf k = 1. Thus it means that kΛ∗x k ≤ kxk. On the other hand, since
X is a normed linear space, and x 6= 0, Theorem 5.20 ensures the existence of a g ∈ X ∗
such that kgk = 1 and g(x) = kxk, so we have
where kgk = 1. Combining the inequalities (5.22) and (5.23), we get the desired result
that
kΛ∗x k = kxk.
(c) Suppose that {xn } ⊆ X and {f (xn )} is bounded for every f ∈ X ∗ . We consider the
mapping Λ∗xn : X ∗ → C given by
Λ∗xn (f ) = f (xn ).
By part (a), X ∗ is a Banach space. By part (b), kΛ∗xn k = kxn k for every n ∈ N. By
the hypothesis, for each f ∈ X ∗ , the set {Λ∗xn (f )} = {f (xn )} is bounded by a positive
constant Mn , so you can’t find a f ∈ X ∗ such that
sup |Λ∗xn (f )| = ∞.
n∈N
kΛ∗xn k ≤ M
for all n ∈ N, where M is a positive constant. Hence the sequence {kxn k} is bounded.
This completes the proof of the problem.
Problem 5.9
Rudin Chapter 5 Exercise 9.
Proof.
so that the series (5.24) converges (absolutely) and thus the map Λ is well-defined.
5.3. The Dual Space of X 153
By the analysis in the previous paragraph, the series (5.25) is allowed to split into
two series so that
∞
X ∞
X
Λ(x + z) = α ξi ηi + β ωi ηi = αΛ(x) + βΛ(z),
i=1 i=1
ξn,i → 0
for every n ∈ N, i.e., kΛk ≥ kyk1 . In conclusion, we have kΛk = kyk1 < ∞ and thus
Λ ∈ (c0 )∗ .
ei = {δi,k }∞
k=1 , where δi,k is the Kronecker delta function. Then it is routine to
check that ei ∈ c0 and kei k∞ = 1.
Next, we define the sequence y = {ηi }∞ i=1 by
ηi = Λ(ei ) (5.27)
Therefore, we apply [100, Eqn. (3), p. 96] to the expression (5.28) to obtain
n
X
|ηi | = |Λ(xn )| ≤ kΛk · kxn k∞ = kΛk
i=1
i.e, y ∈ ℓ1 .
Finally, for every x = {ξi } ∈ c0 , we have
∞
X
x= ξi ei
i=1
and the application of the expressions (5.27) and the linearity of Λ indicate that
X
∞ X∞ ∞
X
Λ(x) = Λ ξi ei = ξi Λ(ei ) = ξi ηi
i=1 i=1 i=1
∗ Step 4: Φ is an isometry. This follows from the first assertion directly because
kΦ(y)k = kΛy k = kyk1 .
for every x = {ξi } ∈ ℓ1 . Similar to the proof of the first assertion in part (a), we can show
that the series (5.29) converges (absolutely) and Θ is linear. It is clear from the definition
that |Θ(x)| ≤ kyk∞ · kxk1 , so we have
for every i = 1, 2, . . .. Put z = {ηi }. Then the inequality (5.31) immediately implies that
z ∈ ℓ∞ . In fact, it shows further that kzk∞ ≤ kΘk. Now for every x = {ξi } ∈ ℓ1 , we have
∞
X
x= ξi ei
i=1
and then
∞
X
Θ(x) = ξi ηi
i=1
which is exactly in the form (5.29), so the inequality (5.30) holds for z. Therefore, we have
kΘk = kzk∞ , i.e., Θ ∈ ℓ∞ .
If we define the mapping Ψ : ℓ∞ → (ℓ1 )∗ by
Ψ(y) = Θy ,
where Θy is given by the form (5.29), then the previous paragraph actually shows that Ψ
is surjective. By using similar argument as in the proofs of Step 1, Step 3 and Step 4
in part (a), it is easy to conclude that Ψ is an isometric vector space isomorphism.
f (x0 ) 6= 0. (5.32)
If f ∈ ℓ1 , then we have f ∈ (c0 )∗ by part (a). Since f (x) = 0 on c0 , f is in fact the trivial
bounded functional on c0 . By the assumption and part (a), (c0 )∗ is isometric isomorphic
to (ℓ∞ )∗ so that the trivial bounded functional on c0 corresponds to the trivial bounded
functional on ℓ∞ . In other words, f (x) = 0 on ℓ∞ which contradicts the result (5.32).
Hence we conclude that ℓ1 is not isometric isomorphic to (ℓ∞ )∗ .
∞ N ′ ∞
X X X ǫ ǫ
kx − yk1 = |ξi − ηi | ≤ |ξi − ηi | + |0 − ηi | < + = ǫ.
2 2
i=1 i=1 i=N ′
– ℓ∞ is not separable. Let A be the set of all sequences x = {ξi } such that ξi ∈ {0, 1}.
Clearly, A is uncountable (see [99, Theorem 2.14]) and A ⊆ ℓ∞ . Let x, y ∈ A be such
that x = {ξi } and y = {ηi }. If x 6= y, then there is at least one i such that ξi 6= ηi
which implies that
kx − yk∞ = 1.
By this and the fact that B(x, 21 ) ∩ B(y, 21 ) = ∅ if x 6= y, we conclude that the
collection of open balls {B(x, 12 ) | x ∈ A} is uncountable. If S was a dense subset of
ℓ∞ , then we have 1
S ∩ B x, 6= ∅
2
for every x ∈ A so that S must be uncountable. Hence ℓ∞ is not separable.
Remark 5.1
We say two normed linear spaces X and Y isometric isomorphism if there is a bijective
linear map T : X → Y such that
kT (x)k = kxk
for every x ∈ X.
Problem 5.10
Rudin Chapter 5 Exercise 10.
where x = {ξi } ∈ c0 . By the hypotheses of Problem 5.9, we know that c0 is a Banach space. On
the one hand, we note from the definition (5.37) that Λn is linear and the analysis in Problem
5.9(a) first assertion shows that
n
X
|Λn (x)| ≤ kxk∞ · |αi |
i=1
On the other hand, if we define the sequence xn = {ξn,i } by (5.26) with ηi replaced by αi , then,
using the argument as in the proof of Problem 5.9(a), we conclude that xn ∈ c0 . Thus we obtain
from Definition 5.2 that
n
X
kΛn k = sup{|Λn (x)| | x ∈ c0 and kxk∞ ≤ 1} ≥ |Λn (xn )| = |αi | (5.39)
i=1
158 Chapter 5. Examples of Banach Space Techniques
for every n ∈ N. Combining the inequalities (5.38) and (5.39), we conclude that
n
X
kΛn k = |αi | (5.40)
i=1
which is bounded for each n ∈ N. By the hypothesis and the definition (5.37), there is no x ∈ c0
such that sup |Λn (x)| = ∞. Thus Theorem 5.8 (The Banach-Steinhaus Theorem) implies that
n∈N
there exists a positive constant M such that
kΛn k ≤ M (5.41)
for every n ∈ N. Consequently, we deduce from the two results (5.40) and (5.41) that
∞
X
|αi | < ∞,
i=1
Problem 5.11
Rudin Chapter 5 Exercise 11.
• Step 2: Lip α is a normed linear space with the mentioned norms. If we define
kf k1 = |f (a)| + Mf , then for f, g ∈ Lip α, it is easy to see from the inequality (5.42) that
Finally, if kf k1 = 0, then |f (a)| = Mf = 0 so that |f (s) − f (a)| = 0 for all s ∈ (a, b]. Thus
we establish that f (x) = 0 on [a, b] and we can say that Lip α is a normed linear space
with respect to the first norm.
For the second norm, if f, g ∈ Lip α, then we see that
kµf k2 = Mµf + sup |µf (x)| = |µ| · Mf + |µ| · sup |f (x)| = |µ| · kf k2 .
x∈[a,b] x∈[a,b]
Finally, if kf k2 = 0, then we have Mf = sup |f (x)| = 0 implying that f (x) = 0 on [a, b].
x∈[a,b]
Hence Lip α is also a normed linear space with respect to the second norm.
• Step 3: Lip α is complete. Suppose that {fn } ⊆ Lip α is Cauchy with respect to the
norm k · k2 . Given ǫ > 0. There exists a positive integer N such that n, m ≥ N imply that
ǫ
sup |fn (x) − fm (x)| + Mfn −fm = kfn − fm k2 < . (5.43)
x∈[a,b] 2
Particularly, we have
ǫ
|fn (x) − fm (x)| < (5.44)
2
for all x ∈ [a, b]. Thus {fn (x)} is a Cacuhy sequence in C. Since C is complete, there
exists a function f : [a, b] → C such that
On the other hand, we know from the inequality (5.43) that Mfn −fm < 2ǫ , or explicitly
for all s, t ∈ [a, b] with s 6= t. Taking m → ∞ in the inequality (5.48), we observe that
where N is the positive integer which makes the inequality (5.43) holds. Therefore, we
deduce from the definition and the inequality (5.49) that
ǫ
Mf ≤ MfN −f + MfN ≤ + MfN < ∞,
2
i.e., f ∈ Lip α. In other words, Lip α is complete with respect to the norm k · k2 .
For the first norm k · k1 , we note that
If {fn } ⊆ Lip α is Cauchy with respect to the norm k · k1 , then we deduce from the
inequality (5.50) that
kfn − f k1 ≤ kfn − f k2 < ǫ
for large enough n, where the f is defined by the limit (5.45). Since f ∈ Lip α, we have
shown that Lip α is complete with respect to the norm k · k1 .
Hence Lip α is also a Banach space with respect to the two norms and this completes the proof
of the problem.
Problem 5.12
Rudin Chapter 5 Exercise 12.
K = Conv (P, Q, R)
= {λP + τ Q + νR | λ, τ, ν ≥ 0 and λ + τ + ν = 1}
= {(λx1 + τ x2 + νx3 , λy1 + τ y2 + νy3 ) | λ, τ, ν ≥ 0 and λ + τ + ν = 1}.
and thus
To prove the uniqueness of µ, we suppose that the integral representation (5.53) holds
for the measure µ′ . Then the integral (5.53) implies that
f (x0 , y0 ) = µ′ (P )f (x1 , y1 ) + µ′ (Q)f (x2 , y2 ) + µ′ (R)f (x3 , y3 )
= f µ′ (P )x1 + µ′ (Q)x2 + µ′ (R)x3 , µ′ (P )y1 + µ′ (Q)y2 + µ′ (R)y3 . (5.54)
By the definition of f , the expression (5.54) means that
(x0 , y0 ) = µ′ (P ) · P + µ′ (Q) · Q + µ′ (R) · R. (5.55)
Comparing the expressions (5.51) and (5.55), we derive that
[µ′ (P ) − λ]P + [µ′ (Q) − τ ]Q + [µ′ (R) − ν]R = 0. (5.56)
Since the set {P, Q, R} is linearly independent, it follows from the equation (5.56) that
µ′ (P ) = λ, µ′ (Q) = τ and µ′ (R) = ν.
In other words, we establish the fact that µ′ = µ.
• Case (ii): K is a square. Without loss of generality, we may assume that the vertices
of K are P (0, 0), Q(1, 0), R(1, 1) and S(0, 1). Let A(a, b) be a point of K, see Figure 5.2
below.
– Subcase (i): A belongs to Region I. Here we have 0 < b ≤ a < 1 and a+b−1 ≤ 0.
If θ = 0, then we obtain from the equations (5.57) that ν = b and τ = a − b ≥ 0. If
λ = 1 − a ≥ 0, then it is easy to see that
Hence, every point in Region I has two different convex combinations of P, Q, R and
S.
– Subcase (ii): A belongs to Region II. In this case, we have 0 < b ≤ a < 1 and
a + b − 1 ≥ 0. Since we have
so every point in Region III also has two different convex combinations of P, Q, R and
S.
– Subcase (iv): A belongs to Region IV. Now we have 0 < a ≤ b < 1 and
a + b − 1 ≤ 0. Then we certainly have
which means that it has two different convex combinations of P, Q, R and S in Region
IV.
– Subcase (v): A belongs to the boundary of K. Suppose that A = (a, b). We
note that
(a, 0) = (1 − a) · P + aQ + 0 · R + 0 · S = 1 · P + aQ + 0 · R + 0 · S,
thus every point on P Q can be expressed as two distinct linear (but not convex)
combinations of P, Q, R and S. Similarly, all points lie on QR, RS and SP also have
two different linear combinations of P, Q, R and S.
In conclusion, every point (x0 , y0 ) in the square K has at least two distinct linear
combinations of P, Q, R and S. Now, instead of the expression (5.52), we have
where {λ, τ, ν, θ} and {λ′ , τ ′ , ν ′ , θ ′ } are two sets of distinct numbers corresponding to two
distinct linear combinations of P, Q, R and S. Thus {λ, τ, ν, θ} and {λ′ , τ ′ , ν ′ , θ ′ } give
distinct measures µ and µ′ such that the representation (5.53) holds.
5.4. Applications of Baire’s and other Theorems 163
Problem 5.13
Rudin Chapter 5 Exercise 13.
Proof.
(a) For each N = 1, 2, . . ., we define
EN = {x ∈ X | |fn (x)| ≤ N for all n ∈ N}.
Since each fn is continuous on X, each EN is closed in X. On the one hand, we have
EN ⊆ X. (5.58)
On the other hand, for each fixed x ∈ X, since f (x) is well-defined as a complex number,
it must be true that
|f (x)| ≤ m
for some m > 0. By the convergence of {fn (x)}, there is a positive integer N0 such that
n ≥ N0 implies that
|f (x) − fn (x)| < 1
Thus, for n ≥ N0 , the triangle inequality shows that
|fn (x)| ≤ |f (x)| + |f (x) − fn (x)| < m + 1. (5.59)
Take m′ = max f1 (x), f2 (x), . . . , fN0 −1 (x) . Combining this and the inequality (5.59), we
obtain that
|fn (x)| < max(m′ , m + 1)
for all n ∈ N. Hence we have x ∈ EN1 for some N1 ≥ max(m′ , m + 1) and then we follow
from the set relation (5.58) that
[∞
X= EN .
N =1
By Theorem 5.6 (Baire’s Theorem), X is not of the first category, i.e., X is not a countable
union of nowhere dense sets. Therefore, some EN ′ contains a nonempty open subset V of
X, i.e., for all x ∈ V and n = 1, 2, . . ., we have
|fn (x)| ≤ N ′ ,
where N ′ < ∞.
164 Chapter 5. Examples of Banach Space Techniques
Since fm and fn are continuous on X, fm − fn is also continuous on X. Thus for each pair
n, m ∈ N, the set
S(m, n) = (fm − fn )−1 [−ǫ, ǫ] = {x ∈ X | |fm (x) − fn (x)| ≤ ǫ}
is closed in X. Since \
AN = S(m, n),
n,m≥N
AN is definitely closed in X.
It is clear that
AN ⊆ X (5.60)
for each N = 1, 2, . . .. Now for every x ∈ X, by the convergence of {fn (x)}, it is Cauchy
in C, i.e., there exists a N0 ∈ N such that n, m ≥ N0 imply that
By Theorem 5.6 (Baire’s Theorem), X is not of the first category, i.e., X is not a countable
union of nowhere dense sets. Therefore, some AN ′ contains a nonempty open subset V of
X. Particularly, if x ∈ V and n, m ≥ N ′ , then we have
|f (x) − fn (x)| ≤ ǫ
Problem 5.14
Rudin Chapter 5 Exercise 14.
Proof. Let C = {f : [0, 1] → R | f is continuous on [0, 1]} and kf k = sup |f (x)|. Furthermore,
x∈[0,1]
for a positive integer n, let
Xn = {f ∈ C | there exists a t ∈ I such that |f (s) − f (t)| ≤ n|s − t| for all s ∈ I}.
5.4. Applications of Baire’s and other Theorems 165
To prove the first assertion, we fix n ∈ N. We follow the hint given by Rudin and the construction
of g will be shown in several steps below.
Let V be an open set in C and f ∈ V . Then there exists a ǫ > 0 such that B(f, 2ǫ) ⊆ V .
Since f is continuous on the compact [0, 1], it is uniformly continuous on [0, 1], i.e., there exists
a k ∈ N such that
|f (x) − f (y)| < ǫ (5.63)
for all x, y ∈ [0, 1] with |x − y| ≤ k1 . In particular, we consider the partition P = {0 =
x0 , x1 , . . . , xk = 1} of [0, 1], where xi = ki for i = 0, 1, . . . , k. Then we know from the inequality
(5.63) that
|f (x) − f (xi )| < ǫ (5.64)
for all x ∈ [xi−1 , xi+1 ]. To prove the first assertion, we are going to divide the proof into several
steps:
Lemma 5.1
For each n ∈ N, if we define
e
Such zigzag function ϕ is something looks like the graph in Figure 2.1.
166 Chapter 5. Examples of Banach Space Techniques
In other words, we may assume that 4n x ∈ [0, 1). Let h > 0 be so small such that
4n (x + h) ∈ [0, 1]. Then we obtain
ϕn (x + h) − ϕn (x)
h
1 ϕ 4n (x + h) − ϕ(4n x)
= n·
2 h
n (x + h) − 2 · 4n x
1 2 · 4
n·
, if 4n x, 4n (x + h) ∈ [0, 12 ];
2 h
1 1 − 2 · 4n h − 1
= · , if 4n x = 12 and 4n (x + h) = 1
+ 4n h;
2 n h 2
n n
1 · −2 · 4 (x + h) + 2 · 4 x , 4n x, 4n (x + h) ∈ [ 1 , 1],
n 2
2 n h
n x, 4n (x + h) ∈ [0, 1 ];
2 · 2 , if 4 2
= −2 · 2n , if 4n x = 12 and 4n (x + h) = 21 + 4n h;
−2 · 2n , 4n x, 4n (x + h) ∈ [ 12 , 1]
which implies that |ϕ′n (x+)| ≥ 2n . Similarly, |ϕ′n (x−)| ≥ 2n , completing the proof of
Lemma 5.1.
• Step 3: Construction of a e
g ∈ C with large slopes and ke
g − f k < 2ǫ. We define
ge = g + ϕk ,
where the k will be determined later. We claim that gb satisfies the required properties.
To this end, let
Mk = max k|f (xi+1 ) − f (xi )|.
0≤i≤k−1
Now for every x ∈ [0, 1], if we take h > 0 small enough such that x + h ∈ [xi , xi+1 ], then
we see from the definition (5.65) and the expression (5.66) that
g(x + h) − g(x) |λ
x+h − λx | · |f (xi+1 ) − f (xi )|
= = k|f (xi+1 ) − f (xi )| ≤ Mk (5.68)
h h
which implies that |g′ (x+)| ≤ Mk on [0, 1]. Similarly, we have |g′ (x−)| ≤ Mk for every
x ∈ [0, 1].f
f
If x = 1, then g ′ (1+) does not exist. Similarly, if x = 0, then g ′ (0−) does not exist.
5.4. Applications of Baire’s and other Theorems 167
Suppose that N ′ ∈ N. Since f is bounded on [0, 1], we must have Mk ≤ kM for some
constant M > 0 and then k can be chosen so large that
1 ǫ
2k ≥ kM ≥ Mk + N ′ and < . (5.69)
2k 2
Since g and ϕk are continuous on [0, 1], ge is also continuous on [0, 1]. Next, for x ∈ [0, 1)
and h > 0 so small such that x + h ∈ [0, 1], we derive from Lemma 5.1 and the inequality
(5.68) that
ge(x + h) − ge(x) g(x + h) − g(x) ϕ(x + h) − ϕ(x)
= +
h h h
ϕ(x + h) − ϕ(x) g(x + h) − g(x)
≥ −
h h
≥ 2k − Mk
> N′ (5.70)
which yields |eg ′ (x+)| > N ′ for all x ∈ [0, 1). Similarly, we have |e
g′ (x−)| > N ′ for all
x(0, 1]. Since N ′ can be chosen arbitrary large, our e
g is continuous function on [0, 1] with
large slopes.
Finally, it remains to show that ke
g − f k < ǫ. On [0, 1], we know from Steps 1 and 2
that
ǫ
g(x) − f (x)| ≤ |g(x) − f (x)| + |ϕk (x)| < ǫ + 2−k |ϕ(4k x)| < ǫ +
|e < 2ǫ.
2
Thus we conclude that ke g − f k < 2ǫ. By the definition of Xn and the lower bound (5.70),
/ Xn if we pick N ′ ≥ n.
we note that ge ∈
for every s ∈ [0, 1] and k ∈ N. Since [0, 1] is compact, we may assume that tk → t ∈ [0, 1].
By Theorem 3.17, C is a metric space and then we recall from the rephrased Theorem 7.9
[99, p. 151] that fk → f uniformly on [0, 1]. Hence it follows from [99, Exercise 9, p. 166]
that
lim fk (tk ) = f (t)
k→∞
for every s ∈ [0, 1]. In other words, we conclude that f ∈ Xn which means Xn is closed in
C.
1
g − f k < ǫ′ + ke
kh − f k ≤ kh − gek + ke g − fk ≤ (2ǫ − ke
g − f k) + ke
g − f k < 2ǫ
2
g, ǫ′ ) ⊆ B(f, 2ǫ) ⊆ V .
which means B(e
168 Chapter 5. Examples of Banach Space Techniques
is a dense Gδ in C. Suppose that f ∈ C and f is differentiable at p ∈ [0, 1). Then there exists
a δ > 0 such that for all 0 < |h| < δ, we haveg
f (p + h) − f (p)
− f ′ (p) ≤ 1
h
which implies that
f (p + h) − f (p)
≤ 1 + |f ′ (p)|. (5.72)
h
Obviously, if |h| ≥ δ, then
f (p + h) − f (p) |f (p + h)| + |f (p) 2kf k
≤ ≤ . (5.73)
h h δ
Combining the inequalities (5.72) and (5.73), we obtain that
|f (x) − f (p)| ≤ N |x − p|
or equivalently, f ∈
/ XN which is a contradiction. Hence we have completed the proof of the
problem.
Problem 5.15
Rudin Chapter 5 Exercise 15.
Proof. Recall from Problem 5.9 that c0 is the subspace of ℓ∞ consisting of all x = {ξi } ∈ ℓ∞ for
which ξi → 0 as i → ∞. By the definition, we have
{σi } = A(x)
whenever the series converges (but ξi not necessarily converging to 0). We are going to prove
the results one by one:
• Necessity part: Suppose that A transforms every convergent sequence {sj } to a sequence
{σi } which converges to the same limit.
To prove Condition (a), it suffices to prove the case when both {sj } and {σi } converge
to 0 because we may replace sj and σi by sj − L and σi − L respectively, where L is the
g
Of course, only one-sided derivative f ′ (0+) will be considered if p = 0.
5.4. Applications of Baire’s and other Theorems 169
common limit of {sj } and {σi }. In other words, we work on the space c0 . In fact, for each
fixed i ∈ N, we define Λi : c0 → C by
∞
X
Λi (x) = σi = aij sj
j=0
for all x ∈ c0 so that A(x) = {Λi (x)}. Since {σi } is convergent, it is bounded so that
X∞ X ∞ nX
k+1
|aij |
|Λi (x)| = |σi | = aij sj = =∞
k
j=0 k=1 j=nk +1
By the inequality (5.74) again, we have Λi ∈ c∗0 and since c0 is Banach, we may apply
Theorem 5.8 (The Banach-Steinhaus Theorem) to conclude that
∞
X
for all i ∈ N. Recall that |aij | < ∞ for every i = 1, 2, . . ., so yi = {aij } ∈ ℓ1 for every
j=0
i = 1, 2, . . . and Problem 5.9(a) implies that
∞
X
kΛi k = kyi k1 = |aij | (5.77)
j=0
for every i = 1, 2, . . .. Hence Condition (b) follows from the inequality (5.76) and the
equality (5.77).
• Sufficiency part: Let x = {sj } and sj → L as j → ∞. By Condition (b), the series
X∞
σi = aij sj converges absolutely for every i ∈ N. We write
j=0
∞
X ∞
X
σi = aij (sj − L) + L · aij . (5.78)
j=0 j=0
Let the supremum in Condition (b) be M . Given ǫ > 0, we choose a N ∈ N such that
|sj − L| < Mǫ for all n ≥ N . Thus we see that
X∞ X N ∞
X N
X
aij (sj − L) ≤ |aij | · |sj − L| + |aij | · |sj − L| ≤ |aij | · |sj − L| + ǫ.
j=0 j=0 n=N +1 j=0
so that
X∞
lim aij (sj − L) ≤ ǫ.
i→∞
j=0
Since ǫ is arbitrary, we conclude from the representation (5.78) and the limit (5.79) that
lim σi = L.
i→∞
• Two examples. It is easy to check that the examples satisfy the conditions of the
problem.
To show the last assertion, we consider the sequence {sj } defined by
√
k, if j = 2k;
sj =
√
− k, if j = 2k − 1.
Then it is clear that {sj } is unbounded and direct computation shows that, for each
i = 1, 2, . . .,
∞
X
σi = aij sj
j=0
5.5. Miscellaneous Problems 171
i
X 1
= sj
i+1
j=0
2k
1 X
sj , if i = 2k;
2k + 1 j=0
=
2k−1
1 X
sj , if i = 2k − 1;
2k
j=0
0, if i = 2k;
=
1
− √ , if i = 2k − 1.
2 k
as i → ∞.
Remark 5.2
Classically, Problem 5.15 is called the Silverman-Toepliotz Theorem. For more information
or other proofs about this theorem, please refer to [49, §3.2], [80, Chap. 4] and [133, §1.2].
Problem 5.16
Rudin Chapter 5 Exercise 16.
Proof. We follow Rudin’s hint. Let X ⊕Y = {(x, y) | x ∈ X and y ∈ Y } with addition and scalar
multiplication defined componentwise. We define the norm k · k on X ⊕ Y byh
kα(x, y)k = k(αx, αy)k = kαxkX + kαykY = |α| · kxkX + |α| · kykY = |α| · k(x, y)k.
Finally, if k(x, y)k = 0, then kxkX + kykY = 0. Since kxkX and kykY are nonnegative for every
x ∈ X and y ∈ Y , kxkX + kykY = 0 implies that kxkX = kykY = 0 and thus (x, y) = (0, 0).
Hence X ⊕ Y is a normed linear space.
Suppose that {(xn , yn )} ⊆ X ⊕ Y is Cauchy. Given ǫ > 0, Then there exists a positive integer
N such that n, m ≥ N imply that
By the definition (5.80), we must have kxn − xm kX < ǫ and kyn − ym kY < ǫ for all n, m ≥ N .
In other words, {xn } ⊆ X and {yn } ⊆ Y are also Cauchy. Since X and Y are Banach, there
exist x ∈ X and y ∈ Y such that
Thus G is a linear subspace of X ⊕ Y and it is also a metric space. We claim that G is closed
in X ⊕ Y . To see this, let { xn , Λ(xn ) } ⊆ G and k xn , Λ(xn ) − (x, y)k → 0 as n → ∞. By the
definition (5.80), we have kxn − xkX → 0 and kΛ(xn ) − ykY → 0 as n → ∞, i.e.,
Then it is clear that Φ is linear and bijective. Furthermore, for every x ∈ X, we see that
kΦ x, Λ(x) kX = kxkX ≤ kxkX + kΛ(x)kY = k x, Λ(x) k
i.e., Φ is bounded. By Theorem 5.10, Φ−1 is also a bounded linear transformation of X onto G
and it follows from Theorem 5.10’s proof that there exists a positive constant δ such that
1
kΦ−1 k ≤ .
δ
5.5. Miscellaneous Problems 173
Besides, Theorem 5.4 shows that Φ−1 is continuous. Combining these two facts and [100, Eqn.
(3), p. 96], we obtain
1
kxkX + kΛ(x)kY = k x, Λ(x) k = kΦ−1 (x)k ≤ kΦ−1 k · kxkX ≤ kxkX < ∞ (5.81)
δ
for every x ∈ X. Since kxkX and kΛ(x)kY are nonnegative, we get from the inequality (5.81)
that δ < 1 and 1
0 ≤ kΛ(x)kY ≤ − 1 · kxkX (5.82)
δ
on X. Hence we conclude from the inequality (5.82) that
1
kΛk = sup{kΛ(x)kY | x ∈ X and kxkX = 1} ≤ < ∞ − 1.
δ
Consequently, Λ is bounded and it is continuous on X by Theorem 5.4, completing the proof of
the problem.
Problem 5.17
Rudin Chapter 5 Exercise 17.
• Measures µ with kMf k = kf k∞ for all f ∈ L∞ (µ). We call the measure µ semifinite
if for each E ∈ M with µ(E) = ∞ one can find a F ∈ M with F ⊂ E such that
0 < µ(F ) < ∞, see [40, p. 25].
Now we claim that kMf k = kf k∞ for all f ∈ L∞ (µ) if and only if the measure µ is
semifinite.
f g = 0 a.e. on E
kMf k = 0.
• Functions f ∈ L∞ (µ) such that Mf is onto. Suppose that µ satisfies the condition
that every measurable set E of positive measure contains a measurable subset F with
0 < µ(F ) < ∞.i We claim that the map Mf is onto if and only if f1 ∈ L∞ (µ).
We claim that µ(E) = 0. Assume that µ(E) > 0. By the hypothesis, there exists
a F ⊆ E with 0 < µ(F ) < ∞. Thus χF ∈ L2 (µ). Since f g = 0 on F for every
g ∈ L2 (µ), it implies that
χF ∈/ Mf L2 (µ)
i
Some books take this as the definition of a semifinite measure µ. See, for example, [14, Exercise 25.9]
5.5. Miscellaneous Problems 175
g = 0 a.e. on X
Problem 5.18
Rudin Chapter 5 Exercise 18.
Proof. Let x ∈ X and ǫ > 0. Since E is dense in the normed linear space X, we can find y ∈ E
such that
ǫ
kx − yk < . (5.89)
2M
Since {Λn (y)} converges in the Banach space Y , there exists a positive integer N such that
n, m ≥ N imply that
ǫ
kΛn (y) − Λm (y)k < . (5.90)
2
Therefore, if n, m ≥ N , then we deduce immediately from the inequalities (5.89) and (5.90) that
kΛn (x) − Λm (x)k ≤ kΛn (x) − Λn (y)k + kΛn (y) − Λm (y)k + kΛm (y) − Λm (x)k
ǫ
< kΛn (x − y)k + + kΛm (y − x)k
2
ǫ
≤ kΛn k · kx − yk + kΛm k · kx − yk +
2
ǫ ǫ
< +
2 2
= ǫ.
This means that {Λn (x)} is Cauchy in Y . Since Y is Banach, {Λn (x)} converges in Y and we
complete the analysis of the problem.
176 Chapter 5. Examples of Banach Space Techniques
Problem 5.19
Rudin Chapter 5 Exercise 19.
By §5.11, C(T ) is Banach relative to the supremum norm kf k∞ .j Next, since f ∈ C(T ), we
follow from the equations (5.91) that sn (f ; x) ∈ C(T ) for each n ∈ N.
Let X = Y = C(T ) and for each n = 2, 3, . . ., we define Λn : C(T ) → C(T ) by
sn (f ; x)
Λn (f ) = .
log n
Since f ∈ C(T ), there exists a M > 0 such that |f (x)| ≤ M for all x ∈ T so that
Z π Z
1 M π
|sn (f ; x)| ≤ |f (t)| · |Dn (x − t)| dt ≤ |Dn (x − t)| dt. (5.92)
2π −π 2π −π
Since Dn (x − t) = eikx Dn (t) and Dn (t) is an even function, the estimate (5.92) can be replaced
by
Z Z Z
M π M π M π sin(n + 21 )t
|sn (f ; x)| ≤ |Dn (t)| dt = |Dn (t)| dt = dt. (5.93)
2π −π π 0 π 0 sin 2t
Next, we consider f (x) = sin x2 − x4 on [0, π]. Using differentiation, we can show that sin x2 ≥ x4
on [0, π]. Thus the estimate (5.93) can be further written as
Z
4M π | sin(n + 12 )t|
|sn (f ; x)| ≤ dt
π 0 t
Z 1
4M (n+ 2 )π | sin t|
= dt
π 0 t
Z n Z (k+1)π
4M π sin t X | sin t|
≤ dt + dt . (5.94)
π 0 t kπ t
k=1
1 1
On the interval [kπ, (k + 1)π], it is easy to see that t ≤ kπ , where k = 1, 2, . . . , n, so we have
n Z
X (k+1)π X 1
n Z (k+1)π 2X n
| sin t| 1
dt ≤ | sin t| dt = . (5.95)
kπ t kπ kπ π k
k=1 k=1 k=1
2 sin t
Besides, we get from [99, Problem 8.6, p. 197] that π < t < 1 for 0 < t < π2 , so the integral
Z π
2 sin t
dt
0 t
π
is bounded by 2 and thus
Z π Z π Z π Z π
sin t 2 sin t sin t π dt π π π
dt = dt + dt ≤ + = + ln π − ln = + ln 2. (5.96)
0 t 0 t π t 2 π t 2 2 2
2 2
j
To see this, since T is compact, we follow from Definition 3.16 that C(T ) = C0 (T ) and Theorem 3.17 implies
that C(T ) is complete. By Definition 5.2, we see that C(T ) is Banach.
5.5. Miscellaneous Problems 177
Now, by putting the estimates (5.95) and (5.96) into the estimate (5.94), we derive that
s (f ; x) 4M π 8M
n
X 1
n
|Λn (f )| = ≤ × ( + ln 2) + 2 · . (5.97)
log n π log n 2 π log n k
k=1
When kf k∞ = 1, we may take M = 1 in the estimate (5.97). It is well-known (see [124, Eqn.
Xn
1
(6.27), p. 124]) that log n and are of the same growth as n → ∞, so we follow from the
k
k=1
estimate (5.97) that there exists a M ′ > 0 such that
kΛn k ≤ M ′
for all n = 2, 3, . . .. That is, {Λn } satisfies the first hypothesis of Problem 5.18.
By Theorem 4.25 (The Weierstrass Approximation Theorem), the set of all trigonometric
polynomials, namely P, is dense in C(T ). Let Pm (t) = eimt for some m ∈ Z. If n ≥ m, then we
follow from the result [100, Eqn. (8), p. 89] that
Z π n
X n
X Z π
1 1
sn (Pm ; x) = e−imt eikt dt = ei(k−m)t dt = 1. (5.98)
2π −π 2π −π
k=−n k=−n
N
X
Therefore, if P (t) = cm Pm (t), then we get immediately from the result (5.98) that
m=−N
N
X N
X
sn (P ; x) = cm sn (Pm ; x) = cm
m=−N m=−N
as n → ∞. Thus {Λn } satisfies the second hypothesis of Problem 5.18. Hence we establish from
Problem 5.18 that {Λn (f )} converges to 0 for each f ∈ C(T ) and as a consequence, we have
ksn k∞
lim = lim kΛn k = 0.
n→∞ log n n→∞
4 log n γ
n
kDn k1 4 X1 4
≥ 2 ≥ 2 (log n + γ) = 2 + →∞ (5.99)
|λn | π |λn | k π |λn | π |λn | |λn |
k=1
k
We have applied the estimate of kDn k1 used in [100, p. 102] in the first inequality in (5.99).
178 Chapter 5. Examples of Banach Space Techniques
as n → ∞, where γ is the famous Euler constant. Hence Theorem 5.8 (the Banach-Steinhaus
Theorem) ensures that the sequence
n s (f ; 0) o
n
λn
is unbounded for every f in some dense Gδ set in C(T ), completing the proof of the problem.
Problem 5.20
Rudin Chapter 5 Exercise 20.
Proof.
(a) Assume that such a sequence of continuous positive functions {fn } existed. Let x ∈ R and
n, k ∈ N. Furthermore, let
We claim that
∞
\
U= Uk . (5.100)
k=1
On the one hand, if x ∈ Uk for all k ∈ N, then for every k ∈ N, there exists a n ∈ N such
that fn (x) > k. In other words, we have x ∈ U . On the other hand, if x ∈ U , then since
{fn (x)} is unbounded, for every k ∈ N, there exists a n ∈ N such that fn (x) > k which
is equivalent to saying that x ∈ Uk for all k ∈ N. Thus this proves the claim. Next, it is
easy to see that
∞
[
Uk = {x ∈ R | fn (x) > k}.
n=1
Since each fn is continuous, the set {x ∈ R | fn (x) > k} is open in R and therefore, each Uk
is open in R. By the definition (5.100), U is a Gδ set. Thus we follow from the assumption
that
Q=U
is also a Gδ set. We deduce from the definition (5.100) that
∞
[ ∞
[ ∞
[
R = Q ∪ Qc = Q ∪ Ukc = {qk } ∪ Ukc (5.101)
k=1 k=1 k=1
Clearly, each qk is a nowhere dense subset of R. Furthermore, it is trivial that each Ukc is
closed in R. Assume that Ukc0 was not a nowhere dense subset of R for some k0 . Let Vk0
be a nonempty open subset of Ukc0 . Then there exists a p ∈ Vk0 and a δ > 0 such that
(p − δ, p + δ) ⊆ Vk0 ⊆ Ukc0 ⊆ Qc ,
but this means that Q ∩ Qc 6= ∅, a contradiction. Thus every Ukc is also a nowhere dense
subset of R and the representation (5.101) shows that R is a set of the first category which
contradicts Theorem 5.6 (Baire’s Theorem) that no complete metric space is of the first
category.
It is clear that fn is continuous on R and fn (x) ≥ 1 for all x ∈ R, i.e., {fn } is a sequence
of continuous positive functions on R.
Suppose that θ is irrational and N ∈ N. Consider the number
α = min(|θ − q1 |, |θ − q2 |, . . . , |θ − qN |) > 0.
Then the Archimedean Property ([99, Theorem 1.20(a)]) implies that there is a positive
integer n such that nα > N and thus
(c) The sequence of the functions given by (5.102) shows that the assertion is true for irra-
tionals. For the rational numbers, we first prove the case on [0, 1].l To begin with, suppose
that Qn = {q1 , q2 , . . . , qn } ⊆ [0, 1] ∩ Q, where q1 = 0. Furthermore, we let
1
δn = min{|qi − qj | | 1 ≤ i < j ≤ n} > 0.
2n+1
Clearly, we have
Thus fn is a continuous function on [0, 1], zig-zag on (qi − δn , qi + δn ) and fn (qi ) = n for
each i = 1, 2, . . . , n.m Therefore, if x ∈ [0, 1] ∩ Q, then x = qk for some k ∈ N and thus we
obtain
lim fn (x) = ∞.
n→∞
Next, suppose that x is irrational. Assume that there was an N ∈ N such that x ∈ En
for all n ≥ N . By the definition, it means that
x ∈ (qk − δN , qk + δN ) (5.104)
l
The following argument is stimulated by the papers of Fabrykowski [37] and Myerson [77].
m
The graph of the fn looks like the graph shown in Figure 2.1.
180 Chapter 5. Examples of Banach Space Techniques
for some k ∈ {1, 2, . . . , N }. Since x ∈ En for all n ≥ N , the set relation (5.104) shows that
x ∈ (qk − δn , qk + δn )
Problem 5.21
Rudin Chapter 5 Exercise 21.
e = h(E) ∩ E 6= ∅
E
for every measurable E ⊆ R with m(E) = 0. Given Q = {qk } and ǫ > 0. We first construct
a particular measurable set E with m(E) = 0. To this end, for all qk ∈ Q, we consider the
neighborhoods (qk − 2−k ǫ, qk + 2−k ǫ) and their union
∞
[
E(ǫ) = (qk − 2−k ǫ, qk + 2−k ǫ).
k=1
It is easy to see that every E(ǫ) is nonempty open in R. In addition, since Q ⊆ E(ǫ), each E(ǫ)
is dense in R with m E(ǫ) ≤ 2ǫ. Hence it follows from Theorem 5.6 (Baire’s Theorem) that
the set
∞
\ 1
E= E (5.105)
n
n=1
Problem 5.22
Rudin Chapter 5 Exercise 22.
Proof. Suppose that f ∈ C(T ), f ∈ Lip α and f (0) = 0. We need to show that
To achieve the goal, we first quote the following stronger form of the Riemann-Lebesgue Lemma
whose proof can be found in [5, Theorem 11.16, p. 313].
f (t)
In other words, we have t ∈ L1 (T ). Since we have
1 1 h i(n+ 1 )t 1
i
sin n + t= e 2 − e−i(n+ 2 )t ,
2 2i
it yields
1 Z π sin(n + 12 )t 1
Z π
f (t) i(n+ 1 )t 1
Z π
f (t) −i(n+ 1 )t
f (t) · dt = e 2 dt − e 2 dt
π −π t 2πi −π t 2πi −π t
1 Z Z
π
f (t) i(n+ 1 )t 1 π
f (t) −i(n+ 1 )t
≤ e 2 dt + e 2 dt. (5.107)
2π −π t 2π −π t
By Lemma 5.2, each of the integrals on the right-hand side of the inequality (5.107) tends to 0
as |n| → ∞. Thus it is true that
Z
1 π sin(n + 12 )t
f (t) · dt → 0 = f (0)
π −π t
as |n| → ∞.o
Next, we claim that
sn (f ; 0) → f (0) (5.108)
Z b
o sin αt
Integrals of the form g(t) dt are called Dirichlet integrals, where α > 0 and g is defined on [0, b].
0 t
182 Chapter 5. Examples of Banach Space Techniques
Clearly, F is continuous on [−π, π] and so F f ∈ L1 (T ). Using Lemma 5.2 to the right-hand side
of the equation (5.109), we see that
Z π
1 1
lim F (t)f (t) sin n + t dt = 0
n→∞ 2π −π 2
for every x ∈ R. Clearly, the real function g satisfies the conditions g ∈ C(T ), g ∈ Lip α and
g(0) = 0. By the above argument, we obtain
By applying the limit (5.111) to the equation (5.112), we obtain our desired result (5.106),
completing the proof of the problem.
CHAPTER 6
Complex Measures
Problem 6.1
Rudin Chapter 6 Exercise 1.
183
184 Chapter 6. Complex Measures
Since the inequality (6.2) holds for every partition of E, we have |µ|(E) ≤ λ(E) + ǫ. Since ǫ is
arbitrary, we obtain |µ|(E) ≤ λ(E). Hence we conclude that λ = |µ| which completes the proof
of the problem.
Problem 6.2
Rudin Chapter 6 Exercise 2.
Proof. Let µ be Lebesgue measure on (0, 1) and λ be the counting measure on the σ-algebra M
of all Lebesgue measurable sets in (0, 1). Assume that λ was σ-finite. Then we have
∞
[
(0, 1) = En , (6.3)
n=1
where λ(En ) < ∞ for each n ∈ N. Clearly, we have λ(E) < ∞ if and only if E is finite.
Therefore, the representation (6.3) implies that (0, 1) is countable, a contradiction. Hence λ is
not σ-finite.
Next, we check µ ≪ λ. Suppose that λ(E) = 0, where E ∈ M. By the definition of λ, we
know that E = ∅ which implies definitely that µ(E) = µ(∅) = 0. In addition, the definition of
µ ensures that µ (0, 1) = 1, i.e., µ is a bounded measure.
Assume that there was a h ∈ L1 (λ) such that dµ = h dλ, i.e.,
Z
µ(E) = h dλ (6.4)
E
for every E ∈ M. Note that µ(E) ≥ 0, so h(x) ≥ 0 a.e. [λ] on E.a Particulary, if E = (0, 1),
then h(x) ≥ 0 a.e. [λ] on (0, 1). Suppose that En = {x ∈ (0, 1) | h(x) ≥ n1 } for every n ∈ N.
Since h ∈ L1 (λ), Proposition 1.24(a) implies that
Z
1
∞> h dλ ≥ λ(En ) ≥ 0.
En n
Thus we must have λ(En ) ∈ [0, ∞) or equivalently, En is finite or En = ∅. Let
∞
[
E= En = {x ∈ (0, 1) | h(x) > 0}. (6.5)
n=1
Therefore, the set E is countable. There are two cases for consideration:
• Case (i): E 6= ∅. Let E = {x1 , x2 , . . .}. On the one hand, we know from [99, Remark
11.11(f), p. 309] that µ(F ) = 0 for every countable subset F ⊂ (0, 1), so we deduce from
the representation (6.4) and the definition (6.5) that
Z ∞
X
0 = µ(E) = h dλ = h(xn ) 6= 0,
E n=1
a contradiction.
• Case (ii): E = ∅. In this case, we have h(x) = 0 a.e. [λ] on (0, 1) and the representation
(6.4) again shows that Z
1 = µ (0, 1) = h dλ = 0,
(0,1)
a contradiction.
a
See Definition 1.35 for the meaning of the notation a.e. [λ].
6.1. Properties of Complex Measures 185
Hence no such h exists and this ends the proof of the problem.
Problem 6.3
Rudin Chapter 6 Exercise 3.
Proof. We first show that if µ and λ are complex regular Borel measures, then both µ + λ and
αµ are complex regular Borel measures too. By §6.18, it is equivalent to show that both |µ + ν|
and |αµ| are regular Borel measures. To this end, we follow from §6.18 again that both |µ| and
|λ| are regular Borel measures. Let E ∈ B and ǫ > 0. By Definition 2.15, there exist open sets
V1 , V2 ⊇ E such that
ǫ ǫ
|µ|(V1 ) < |µ|(E) + and |λ|(V2 ) < |λ|(E) + . (6.6)
2 2
Similarly, there are compact sets K1 , K2 ⊆ E such that
ǫ ǫ
|µ|(E) < |µ|(K1 ) + and |λ|(E) < |λ|(K2 ) + . (6.7)
2 2
The triangle inequality certainly implies |µ(E) + λ(E)| ≤ |µ(E)| + |λ(E)| and then Definition
6.1 gives
X∞ ∞
X
|µ + λ|(E) = sup |µ(E) + λ(E)| ≤ sup |µ(E)| + |λ(E)| , (6.8)
i=1 i=1
∞
X
where the supremum being taken over all partitions {Ei } of E. Since the series |µ(E)| and
i=1
∞
X
|λ(E)| converge, we deduce from the expression (6.8) that
i=1
∞
X ∞
X
|µ + λ|(E) ≤ sup |µ(E)| + sup |λ(E)| = |µ|(E) + |λ|(E).
i=1 i=1
Thus we have
|µ + λ| ≤ |µ| + |λ|. (6.9)
Since ǫ and E are arbitrary, the measure |µ + λ|(E) is in fact outer regular. Similarly, we have
In other words, |µ + λ| is also inner regular. By Definition 2.15, µ + λ is a regular complex Borel
measure, i.e., µ + λ ∈ M (X). Now the regularity of |αµ| is easy to prove, so we omit the details
here.
186 Chapter 6. Complex Measures
Now it is time to prove the assertion in the question. Since X is a locally compact Haus-
dorff space, Theorem 6.19 (The Riesz Representation Theorem) ensures every Φ ∈ C0 (X)∗ is
represented by a unique µΦ ∈ M (X) in the sense that
Z
Φ(f ) = f dµΦ (6.10)
X
and Z Z
f dµαΦ = αΦ(f ) = f d(αµΦ ).
X X
Therefore, they imply that
By the previous analysis, we see that µΦ+Ψ , µαµ ∈ M (X), so we may define the mapping
F : C0 (X)∗ → M (X) by
F (Φ) = µΦ
and it is easy to see from the results (6.11) that
Furthermore, Theorem 6.19 (The Riesz Representation Theorem) also implies that F is a bi-
jection and kΦk = |µΦ |(X) = kµΦ k = kF (Φ)k. Consequently, F is actually an isometric vector
space isomorphism, i.e.,
C0 (X)∗ ∼
= M (X).
Since C0 (X) is a Banach space with the supremum norm, Problem 5.8 guarantees that C0 (X)∗
is Banach. Hence M (X) is Banach and we have completed the proof of the problem.
Problem 6.4
Rudin Chapter 6 Exercise 4.
where {Xn } is an increasing sequence of measurable sets and µ(Xn ) < ∞ for all n ∈ N, see [105,
Definition 4.22, p. 22]. Let A = {x ∈ X | |g(x)| = ∞} and An = {x ∈ Xn | |g(x)| = ∞}, where
n ∈ N. Since g is measurable, every An and A are measurable by Problem 1.5. Besides, it is
evident that A1 ⊆ A2 ⊆ · · · and if x ∈ A, then x ∈ XN0 for some N0 ∈ N so that x ∈ AN0 . As
a result, we get
∞
[
A= An .
n=1
• Step 1: µ(A) = 0. Otherwise, it follows from the construction and Theorem 1.19(c) that
This means that one can find a N1 ∈ N such that 0 < µ(AN1 ) ≤ µ(XN1 ) < ∞. We know
that χAN1 ∈ Lp (µ), so our hypothesis implies that χAN1 g ∈ L1 (µ), but
Z Z
kχAN1 gk1 = |χAN1 g| dµ = |g| dµ = µ(AN1 ) · ∞ = ∞
X AN1
• Step 2: g ∈ Lq (µ) when 1 < p < ∞. In this case, we have q ∈ (1, ∞). Here we define
En = {x ∈ Xn | |g(x)| ≤ n}.
for each n = 1, 2, . . .. Thus we have gn ∈ Lq (µ) for all n ∈ N. Next, we define Λn : Lp (µ) →
C by Z
Λn (f ) = f gn dµ
X
which is linear for n ∈ N. Since f gn = (f χEn )g and f χEn ∈ Lp (µ), we have f gn ∈ L1 (µ).
By Theorem 1.33 and Theorem 3.8, we obtain
Z Z
f gn dµ ≤ |f gn | dµ = kf gn k1 ≤ kf kp × kgn kq , (6.14)
X X
Z Z Z
kf0 kp = |f0 |p dµ = kgn k−q
q |gn |(q−1)p
dµ = kgn k−q
q |gn |q dµ = 1
X X X
and
Z
−q
Λn (f0 ) = kgn kq p |gn |q−2 gn · gn dµ
X
188 Chapter 6. Complex Measures
Z
− pq
= kgn kq |gn |q dµ
X
− pq
= kgn kq × kgn kqq
= kgn kq (6.16)
for all n ∈ N. Combining the results (6.13), (6.15) and (6.16), we have established the fact
that
kΛn k = kgn kq < ∞, (6.17)
where n ∈ N. Hence {Λn } is a family of bounded linear transformations of Lp (µ) into C.
Since Lp (µ) is Banach (see Definition 5.2) and
Z Z Z Z
|Λn (f )| = f gn dµ ≤ |f gn | dµ = |f g| dµ ≤ |f g| dµ = kf gk1 < ∞
X X En X
for every f ∈ Lp (µ) and n ∈ N, Theorem 5.8 (The Banach-Steinhaus Theorem) implies
that there is a M > 0 such that
kΛn k ≤ M (6.18)
for all n ∈ N. Using the results (6.17) and (6.18), we conclude that
kgn kq ≤ M (6.19)
for all n ∈ N.
By Step 1 and the definition of gn , we have |gn (x)| ≤ |g(x)| < ∞ a.e. on X for all
n ∈ N and
|g1 (x)| ≤ |g2 (x)| ≤ · · ·
for every x ∈ X. Now the Monotone Convergence Theorem [99, Theorem 3.14, p. 55]
implies that |gn (x)| → |g(x)| < ∞ a.e. on X as n → ∞. Hence we gain from Theorem
1.26 (Lebesgue’s Monotone Convergence Theorem) that
Z Z
q q
lim kgn kq = lim |gn | dµ = |g|q dµ = kgkqq . (6.20)
n→∞ n→∞ X X
Hence it follows from the inequality (6.19) and the limit (6.20) that
kgkq ≤ M < ∞,
i.e. g ∈ Lq (µ).
Since |f1 (x)| ≤ 1 for all x ∈ X, we get from Definition 3.7 that kf1 k∞ ≤ 1, i.e., f1 ∈ L∞ (µ).
By the hypothesis, we have f1 g ∈ L1 (µ). Since f1 g = |g|, we conclude that g ∈ L1 (µ).
6.2. Dual Spaces of Lp (µ) 189
Problem 6.5
Rudin Chapter 6 Exercise 5.
Problem 6.6
Rudin Chapter 6 Exercise 6.
Proof. We want to show that Lp (µ)∗ ∼ = Lq (µ) for 1 < p < ∞. Equivalently, we have to show
∗
that for each Λ ∈ L (µ) , there exists a unique g ∈ Lq (µ) such that
p
Z
Λf = f g dµ (6.22)
X
for all f ∈ Lp (µ). The following proof follows mainly Follan’s argument ([40, p. 190]):
Combining the results (6.24) and (6.25) and using the fact p ∈ (1, ∞), we derive that
lim kχFn − χE kp = 0.
n→∞
Since χE ∈ Lp (µ) and each χEi g is measurable, we deduce from the representation (6.22)
and then Theorem 1.27 that
Z Z X ∞ ∞ Z
X ∞
X
λ(E) = Λ(χE ) = χE g dµ = (χEi g) dµ = χEi g dµ = λ(χEi ).
X X i=1 i=1 X i=1
By [40, Theorem 6.14, p. 189], we conclude that g ∈ Lq (µ). Given that f ∈ Lp (µ). By
Theorem 3.8, we have f g ∈ L1 (µ). By Theorem 3.13, there exists a sequence of simple
functions {sn } such that |sn | ≤ |f | for every n ∈ N and sn → f in Lp (µ) as n → ∞.
Since |sn g| ≤ |f g| and f g ∈ L1 (µ), we obtain from Theorem 1.34 (Lebesgue’s Dominated
Convergence Theorem) and Theorem 5.4 that
Z Z
Λ(f ) = lim Λ(sn ) = lim sn g dµ = f g dµ.
n→∞ n→∞ X X
M ≤ kΛk
holds. By the definition, we may select a sequence {En } of σ-finite subsets of X such that
[∞
kgEn kq → M as n → ∞. Define F = En which is obviously σ-finite subset of X so
n=1
that kgF kq ≤ M . Furthermore, we deduce from the inequality (6.27) that
kgEn kq ≤ kgF kq
for all n ∈ N. Therefore, the definition of M implies that M ≤ kgF kq and then
M = kgF kq .
gA = gF a.e. on X. (6.28)
Hence we may pick g = gF and the above argument makes sure this g is unique.
Problem 6.7
Rudin Chapter 6 Exercise 7.
Since |µ| is real, the previous paragraph and the expression (6.29) imply that
Z Z Z
b(−n) = eint h d|µ| = e−int h(t) d|µ|(t) = e−int [h(t)]2 dµ(t)
µ
192 Chapter 6. Complex Measures
for every n ∈ Z. By Theorem 3.14, C(T ) is dense in L1 (|µ|). Using Theorem 4.25 (The
Weierstrass Approximation Theorem) and then Lemma 2.10, we deduce that the set of all
trigonometric polynomials P is dense in L1 (|µ|).
We want to apply Problem 5.18. For any f ∈ L1 (|µ|), we define Λn : L1 (|µ|) → C by
Z
Λn (f ) = f (t)e−int dµ(t)
Λn (f ) → 0 (6.31)
as n → ∞ for each f ∈ P. Therefore, Problem 5.18 says that the limit (6.31) also holds for
2
every f ∈ L1 (|µ|). Obviously, we have h ∈ L1 (|µ|) and so
Z
2
b(−n) = e−int [h(t)]2 dµ(t) = Λn (h ).
µ
Problem 6.8
Rudin Chapter 6 Exercise 8.
b is periodic
Proof. Suppose that X = [0, 2π), B is the collection of all Borel sets in X and µ
with period k ∈ Z. Denote dλ(t) = (e−ikt − 1) dµ(t). Clearly, we have
Z Z
b(n + k) − µ
µ b
b(n) = e−int (e−ikt − 1) dµ(t) = e−int dλ(t) = λ(n) (6.32)
for all n ∈ Z. Particularly, take n = 0 in the result (6.32) and use the periodicity of µ
b and
Theorem 6.12 to get Z Z
h d|λ| = dλ = 0, (6.33)
λ(E) = 0 (6.34)
6.3. Fourier Coefficients of Complex Borel Measures 193
for every E ∈ B.
Using Theorem 6.12 to write dµ = H d|µ|, where H is a measurable function with |H(x)| = 1
on X. Next, we recall the meaning of the notation dλ(t) = (e−ikt − 1) dµ(t) and we observe
from the result (6.34) that
Z Z
0 = λ(E) = (e−ikt − 1) dµ = (e−ikt − 1)H d|µ| (6.35)
E E
A = {t ∈ X | e−ikt − 1 = 0} = A1 ∩ A2 .
Since f1 (t) = cos kt − 1, f2 (t) = sin kt and g(t) = 0 are continuous on X, they are Borel
measurable.b By Problem 1.5(a), we have A1 , A2 ∈ B and also A ∈ B.
We claim that µ is concentrated on A. Assume that it was not the case, i.e., there is E0 ∈ B
such that a E0 ∩ A = ∅ but µ(E0 ) 6= 0. We recall from the result (6.35) that (e−ikt − 1)H = 0
a.e. on any E with |µ|(E) > 0. Since |H| = 1 on X, we have e−ikt = 1 a.e. on any E with
|µ|(E) > 0. Particularly, since µ(E0 ) 6= 0, we have |µ|(E0 ) > 0 and then
e−ikt = 1 (6.36)
for every n ∈ Z. Combining the expression (6.32) and the result (6.38), we may conclude that
b(n + k) = µ
µ b(n)
for all n ∈ Z.
b
Hence we have shown that µ is a complex Borel measure with periodic Fourier coefficient µ
with period k if and only if µ is concentrated on
n 2nπ o
A = {t ∈ X | e−ikt − 1 = 0} = t = n ∈ N ∪ {0} ,
k
completing the proof of the problem.
Problem 6.9
Rudin Chapter 6 Exercise 9.
b
See §1.11.
194 Chapter 6. Complex Measures
as n → ∞. This shows that condition (i) holds for this gn . Since each gn,k is in fact a trapezium
(see Figure 6.1 again) with area
Z 1
(n + 1) n + 2 2n + 3 1
gn,k (x) dx = δn + δn = δn =
0 2 n+1 2 n
which implies that
Z 1 n Z
X 1
gn (x) dx = gn,k (x) dx = 1.
0 k=1 0
Thus this gn satisfies condition (ii). Finally, note that for large n, we have
1 n(2n + 3)
= ≈ 1.
(n + 1)2 δn 2(n + 1)2
Since f ∈ C(I), there is a M > 0 such that |f (x)| ≤ M on I. On the linear part, we have
gn,k = ak x + bk for some ak , bk ∈ R. Then it asserts that, for example,
Z k δn Z k δn
n+1 − 2 n+1 − 2 δn
(ak x + bk )f (x) dx ≤ M |ak | · x dx + M |bk | · (6.40)
k − δn − δn k − δn − δn 2n + 2
n+1 2 2n+2 n+1 2 2n+2
for every k = 1, 2, . . . , n. By routine computation, it can be shown easily that the integral on
′
the right-hand side of the inequality (6.40) is of the growth M n4
, so the integral on the left-hand
′
side of the inequality (6.40) is bounded by M n 3 for some positive constant M ′ . Using this fact,
the inequality (6.39) can be further reduced to
Z
1 1 X k
n+1
f gn dx − f
0 n+1 n+1
k=1
n " Z k δn #
X k M
f kf k∞
+ 2
n+1 1
≤ (n + 1)f (x) dx − f + +
k δ
− 2n n + 1 n + 1 n2 n+1
k=1 n+1
n " Z k δn Z #
X k δn
k M
f kf k∞
+ 2 + 2
n+1 1 n+1
= (n + 1)f (x) dx − f dx + 2 +
k δ
− 2n (n + 1)δn k δ
− 2n n+1 n n+1
k=1 n+1 n+1
" Z k δn #
X n + 2 1 k Mf kf k∞
n+1
≤ (n + 1)f (x) − 2
f dx + 2 +
k δn
− 2 (n + 1) δn n + 1 n n+1
k=1 n+1
" Z
δn
#
X n k
+ 2 k f kf k∞
M
n+1
≤ (n + 1) f (x) − f dx + 2 + , (6.41)
k δn n+1 n n+1
k=1 n+1
−2
196 Chapter 6. Complex Measures
f is a positive constant.
where M
Applying the Mean-Value Theorem for Integrals (see [5, Theorem 7.30, pp. 160, 161]) to the
integral in the inequality (6.39), we see that
Z
k
+ δ2n k k
n+1
f (x) − f dx = f (ξ) − f δn , (6.42)
k
− δ2n n+1 n+1
n+1
k δn k δn
where ξk ∈ [ n+1 − 2 , n+1 + 2 ]. Substituting the expression (6.42) into the inequality (6.41),
we get
Z
1 1 X k X
n+1 n k kf k∞
f gn dx − f ≤ (n + 1)f (ξk ) − f δn +
0 n+1 n+1 n+1 n+1
k=1 k=1
f kf k∞
M
≤ n(n + 1)δn Mn + + , (6.43)
n2 n+1
where
1 2
n
Mn = max f (ξ1 ) − f , f (ξ2 ) − f , . . . , f (ξn ) − f .
n+1 n+1 n+1
k
When n → ∞, ξk → n+1 for each k = 1, 2, . . . , n so that Mn → 0 as n → ∞. By this observation
and the fact lim n(n + 1)δn = 1, we derive from the inequality (6.43) that
n→∞
Z 1 Z 1
lim f gn dx = f dx
n→∞ 0 0
Problem 6.10
Rudin Chapter 6 Exercise 10.
Proof.
(a) Let Φ = {f1 , f2 , . . . , fN } ⊆ L1 (µ) for some N ∈ N and ǫ > 0. By Problem 1.12, there
exists a δk > 0 such that Z
|fk | dµ < ǫ (6.44)
Ek
Lemma 6.1
Suppose that Φ ⊆ L1 (µ) is uniformly integrable. Given ǫ > 0, there exists a δ > 0
such that Z
|f | dµ < ǫ
E
whenever µ(E) < δ and f ∈ Φ.
for all f ∈ Φ and µ(E) < δ. In other words, Lemma 6.1 holds for subsets of real
measurable functions in L1 (µ).
Next, suppose that Φ is a collection of complex measurable functions in L1 (µ). If
f ∈ Φ, then f = u + iv, where u and v are real measurable functions in L1 (µ). Let
Since 0 < |u| ≤ |f | and 0 < |v| ≤ |f |, we see that u, v ∈ L1 (µ) and then Ψ, Ω ⊆ L1 (µ).
Besides, by the estimate (6.46), we have
Z Z
ǫ ǫ
|u| dµ < and |v| dµ < (6.47)
E 2 E 2
when µ(E) < δ and for all u ∈ Ψ and v ∈ Ω. By the definition, the sets Ψ and Ω are
uniformly integrable. Since |f | ≤ |u| + |v|, we deduce from the estimates (6.47) that
Z Z Z
|f | dµ ≤ |u| dµ + |v| dµ < ǫ
E E E
whenever µ(E) < δ and for all f ∈ Φ. This completes the proof of Lemma 6.1.
whenever µ(E) < δ and for all n ∈ N. Employing Theorem 1.28 (Fatou’s Lemma) to
the estimate (6.48), we know that
Z Z
|f | dµ ≤ lim inf |fn | dµ ≤ ǫ (6.49)
E n→∞ E
Now using the estimates (6.51), (6.52) and the fact fN ∈ L1 (µ) that
Z Z Z Z Z Z
|f | dµ = |f | dµ + |f | dµ ≤ |fN − f | dµ + |fN | dµ + |f | dµ < ∞.
X E X\E E E X\E
(e) Suppose that the hypotheses of Theorem 1.34 (Lebesgue’s Dominated Convergence The-
orem) hold. Then hypotheses (i) and (iii) are true definitely. On the one hand, since
g ∈ L1 (µ), Problem 1.12 says that each ǫ > 0 there exists a δ > 0 such that
Z
|g| dµ < ǫ (6.56)
E
whenever µ(E) < δ. On the other hand, |fn (x)| ≤ g(x) for all n ∈ N will imply that
Z Z
|fn | dµ ≤ |g| dµ < ∞. (6.57)
X X
Therefore, {fn } ⊆ L1 (µ) and then we gain from this, Theorem 1.33 and the estimate (6.56)
that Z Z
fn dµ ≤ |fn | dµ < ǫ
E E
whenever µ(E) < δ and all n ∈ N. Hence {fn } is uniformly integrable, i.e., hypothesis
(ii) is true. Since |fn | are measurable by Proposition 1.9(b), it follows from Theorem 1.28
(Fatou’s Lemma) and the estimate (6.57) that
Z Z
|f | dµ ≤ lim inf |fn | dµ < ∞.
X n→∞ X
Thus |f (x)| < ∞ a.e. on X, i.e., hypothesis (iv) is valid. Hence part (b) verifies that
Vitali’s Convergence Theorem implies Theorem 1.34 (Lebesgue’s Dominated Convergence
Theorem) if µ(X) < ∞.
Now we are going to construct an example in which Vitali’s Theorem applies, but
not Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem). For every n ∈ N, we
consider fn : (0, 1) → R defined by
1
fn (x) = χ 1 1 ≥ 0.
x ( n+1 , n )
Evidently, we have m (0, 1) = 1 and |f (x)| ≤ on (0, 1) which are hypotheses (i) and (iv)
respectively. For each x ∈ (0, 1), we have x ∈
/ χ( 1 , 1 ) for all sufficiently large n so that
n+1 n
fn (x) → 0 = f (x) as n → ∞. Thus we have hypothesis (iii). It is trivial to check that
Z Z 1
1
1 n 1
|fn | dx = dx = ln 1 + <∞ (6.58)
0 1 x n
n+1
6.4. Problems on Uniformly Integrable Sets 201
so that {fn } ∈ L1 (0, 1) . Given ǫ > 0. One can find a positive integer N such that
ln(1 + n1 ) < ǫ for all n ≥ N . By this result, we observe from the representation (6.58) that
Z Z 1 Z 1
fn dx ≤ fn dx ≤ |fn | dx < ǫ
E 0 0
for every n ≥ N and every measurable subset E of (0, 1). By part (a), the finite set
{f1 , f2 , . . . , fN −1 } is uniformly integrable. Therefore, these two facts confirm that {fn } is
uniformly integrable, i.e., hypothesis (ii). Hence the set {fn } satisfies all the hypotheses
of Vitali’s Convergence Theorem.
However, for every x ∈ (0, 1), we note from Definition 1.13 that
1
sup(fn (x)) = ,
n∈N x
Notice that each fn is measurable and satisfies fn (0) = fn (1) = 0. If x ∈ (0, 1), then there
/ (0, n1 ) and x ∈
exists a positive integer N such that x ∈ / (1 − n1 , 1) for all n ≥ N . Thus we
obtain fn (x) = 0 for all n ≥ N . Next,
Z 1 Z 1 Z 1
n 1 1
fn (x) dx = n dx − n dx = n × −n× = 0
0 0 1
1− n n n
for each n ∈ N.
However, {fn } is not uniformly integrable. To see this, assume that there corresponds
a δ > 0 such that Z
fn dx < 1 (6.59)
E
1
whenever m(E) < δ and for all n ∈ N. In particular, fix n = N > δ and take EN = (0, N1 )
so that m(En ) = N1 < δ and
Z Z 1 Z 1
N N
fN dx = fN dx = N dx = 1
EN 0 0
(g) If µ(X) = 0, then there is nothing to prove. Without loss of generality, we may assume
that µ(X) > 0. We follow the hint and prove it into several steps.
ρ(En , Em ) < ǫ.
for all n, m ≥ N , i.e., {χEn } is Cauchy in L1 (µ). By Theorem 3.11, there exists
f ∈ L1 (µ) such that {χEn } converges to f in L1 (µ). By Theorem 3.12, {χEn } has a
subsequence {χEnk } such that
ρ(En , E) = kχEn − χE k1 → 0
so that µ(E \ F ) < δ and µ(F \ E) < δ. Since (χE − χF )fn ∈ L1 (µ), Theorem 1.33
and the estimate (6.61) imply that
Z Z
|ϕn (E) − ϕn (F )| = fn dµ − fn dµ
E F
6.4. Problems on Uniformly Integrable Sets 203
Z Z
= χE fn dµ − χF fn dµ
X X
Z
= (χE − χF )fn dµ
Z X
≤ |χE − χF | · |fn | dµ
ZX Z
= |fn | dµ + |fn | dµ
E\F F \E
< 2ǫ.
for all n > N . By part (a), the set {f1 , f2 , . . . , fN } is uniformly integrable, i.e., there
corresponds a δ′ > 0 such that
Z
fn dµ < ǫ (6.65)
A
n = N + 1, N + 2, . . ., since µ(A) < δ, it yields from the estimates (6.64) and (6.65)
that Z Z Z
fn dµ ≤ (fn − fN ) dµ + fN dµ < 3ǫ. (6.66)
A A A
if µ(A) < δ′′ and for all n ∈ N, i.e., {fn } is uniformly integrable.
We have completed the proof of the problem.
Problem 6.11
Rudin Chapter 6 Exercise 11.
Proof. We note that if µ(X) = 0, then Proposition 1.24(e) shows that the result holds trivially.
Therefore, without loss of generality, we may assume that µ(X) 6= 0. Now we want to apply
Problem 6.10(b). By our hypotheses, it suffices to show that Φ = {fn } is uniformly integrable
and |f (x)| < ∞ a.e. on X.
Since fn ∈ L1 (µ) for every n ∈ N, we apply Theorem 1.33 and then Theorem 3.5 (Hölder’s
Inequality) to obtain
Z Z nZ o1 n Z o1
p q 1 1
fn dµ ≤ |fn | dµ ≤ |fn |p dµ dµ < C p × µ(E) q (6.67)
E E E E
−q
for every E ∈ M. Since p > 1, q < ∞. Take δ = ǫq C p . If µ(E) < δ, then it follows from the
inequality (6.67) that
Z
fn dµ < ǫ.
E
Since µ(X) 6= 0, the inequality (6.68) ensures that |f (x)| < ∞ a.e. on X. Hence we deduce
from Problem 6.10(b) that Z
lim |f − fn | dµ = 0
n→∞ X
Problem 6.12
Rudin Chapter 6 Exercise 12.
6.5. Dual Spaces of Lp (µ) Revisit 205
Proof. We note that Problem 1.6 establishes that M is a σ-algebra. Let E = ( 14 , 12 ) which is open
in [0, 1]. It is clear that g −1 (E) = E. Since E and E c are uncountable, we have g−1 (E) ∈ / M
and thus g is not M-measurable by Definition 1.3(c). We have completed the proof of the
problem.
Problem 6.13
Rudin Chapter 6 Exercise 13.
Proof. Let C(I) be the space of all continuous functions on I. It is clear that χ[0,δ] ∈ L∞ but
χ[0,δ] ∈
/ C(I) for some δ > 0. We need the following result:
Lemma 6.2
The space C(I) is closed in the metric space L∞ .
Proof of Lemma 6.2. Let {fn } ⊆ C(I) and {fn } converges to f in L∞ . Given ǫ > 0.
It means that there exists a positive integer N such that n ≥ N implies
We return to the proof of the problem. Since L∞ is a Banach space, we get from Lemma 6.2
and the equivalent form of Theorem 5.19 that there exists a bounded linear functional Λ on L∞
such that
Λ(f ) = 0 (6.69)
for every f ∈ C(I), but Λ(χ[0,δ] ) 6= 0.
Assume that there was a g ∈ L1 (µ) such that
Z
Λ(f ) = f g dm (6.70)
I
for every f ∈ L∞ . In particular, the result (6.69) and the representation (6.70) imply that
Z
f g dm = 0 (6.71)
I
on C(I). Since f ∈ C(I) and g ∈ L1 (µ), we have f g ∈ L1 (µ) and then we apply Theorem
1.39(b) to the integral (6.71) to conclude that f g = 0 a.e. on I. If we take f (x) = ex on I which
belongs to C(I), then we obtain the fact that
g=0
a.e. on I. Hence we follow from the representation (6.70) that the result (6.69) actually holds
on the whole L∞ , a contradiction to the fact that Λ(χ[0,δ] ) 6= 0. This completes the proof of the
problem.
206 Chapter 6. Complex Measures
CHAPTER 7
Differentiation
Problem 7.1
Rudin Chapter 7 Exercise 1.
Proof. Let x be a Lebesgue point of f . Since f ∈ L1 (Rk ), we deduce from the triangle inequality
and Theorem 1.33 that
Z 1 Z
1
|f (x)| ≤ f (x) − f (y) dy + f (y) dy
m(Br ) Br m(Br ) Br
1 Z 1
Z
≤ [f (x) − f (y)] dy + |f (y)| dy
m(Br ) Br m(Br ) Br
Z
1
≤ |f (y) − f (x)| dy + (M f )(x). (7.1)
m(Br ) Br
|f (x)| ≤ (M f )(x)
Problem 7.2
Rudin Chapter 7 Exercise 2.
Proof. We have I = (−δ, δ). Now we are going to prove the existence of a measurable set E ⊆ R
such that
m E ∩ I(δ) m E ∩ I(δ)
lim inf = α and lim sup = β. (7.2)
δ→0 2δ δ→0 2δ
In fact, it suffices to prove that
m E ∩ [0, δ) m E ∩ [0, δ)
lim inf =α and lim sup = β, (7.3)
δ→0 δ δ→0 δ
where E ⊆ [0, ∞) because we can get the desired result (7.2) by reflecting about the origin. To
start our proof, we consider five cases:
207
208 Chapter 7. Differentiation
• Case (i): 0 < α < β < 1. In this case, we consider the numbers
α(1 − β) α β−α n
θ= < <1 and dn = ·θ >0
β(1 − α) β 1−α
En = [θ n − dn , θ n ].
satisfies the requirement (7.3). To see this, we first note from the inequality (7.4) that
dn → 0 as n → ∞ and En ∩ En+1 = ∅ for every n ∈ N, see Figure 7.1 below:
Next, choose δ′ > δ > 0 as shown in Figure 7.1, then it is easy to know that
∞
[ ∞
[ ∞
[
[0, θ n − dn ) ∩ E = Ek , [0, θ n ) ∩ E = Ek , [0, δ) ∩ E = Ek
k=n+1 k=n k=n+1
and
∞
[
[0, δ′ ) ∩ E = [θ n − dn , δ′ ) ∪ Ek
k=n+1
so that
∞
m [0, θ n − dn ) ∩ E 1−α X β−α θ
= dk = · = α,
θ n − dn (1 − β)θ n 1−β 1−θ
k=n+1
∞
m [0, θ n ) ∩ E 1 X β−α 1
= dk = · = β,
θn θn 1−α 1−θ
k=n
∞
m [0, δ) ∩ E m [0, θ n+1 ) ∩ E) 1 X βθ n+1
= = dk =
δ δ δ δ
k=n+1
7.1. Lebesgue Points and Metric Densities 209
and
m [0, δ′ ) ∩ E m [0, θ n ∩ E − m [δ′ , θ n ] βθ n + δ′ − θ n θn
= = = 1 − (1 − β) .
δ′ δ′ δ′ δ′
Therefore, the above analysis shows trivially that the function f : (0, ∞) → R defined by
m([0, δ) ∩ E)
f (δ) = (7.5)
δ
attains the values α and β at the end-points of each En , decreases from β to α in the
interval [θ n+1 , θ n − dn ] and increases from α to β in En . Hence, we have
and
2n − 1 h 1 1 m E ∩ [0, 1 )
(2n−1)!
= (2n − 1)! × m , ≤ 1 ≤ 1.
2n (2n)! (2n − 1)! (2n−1)!
N
X 1 1 1
Since = − , we have
k(k + 1) n N
k=n
∞
X 1 1
= . (7.7)
k(k + 1) n
k=n
Now we put the infinite series (7.7) into the expression (7.6) to get
m E ∩ [0, n1 )
1 =α
n
• Case (iv): 0 = α < β < 1. Similar to the construction in Case (i), we suppose that
h1 − β β 1 i
En = + , (7.8)
(2n)! (2n + 2)! (2n)!
∞
[
where n ∈ N. Similarly, we have En ∩ En+1 = ∅ for all n ∈ N. Let E = En . On the
n=1
one hand, since
h 1−β β ∞ h
X 1 1 i β
m E ∩ 0, + =β − = ,
(2n)! (2n + 2)! (2k)! (2k + 2)! (2n + 2)!
k=n+1
we have
1−β β β
m E ∩ [0, (2n)! + (2n+2)! ) (2n+2)! β
1−β β
= 1−β β
= →0
+ (1 − β)(2n + 2)(2n + 1) + β
(2n)! (2n+2)! (2n)! + (2n+2)!
we have
1 β
m E ∩ [0, (2n)! ) (2n)!
1 = 1 = β.
(2n)! (2n)!
1 1−β β
Besides, the function (7.5) decreases from β to 0 in the interval [ (2n+2)! , (2n)! + (2n+2)! ] and
increases from 0 to β in En . Hence these imply the results (7.3).
• Case (v): 0 < α < β ≤ 1. Instead of the intervals (7.8), we consider the intervals
h 1 α 1−α i
En = , + ,
(2n)! (2n)! (2n + 2)!
where n ∈ N. Thus it can be shown similar to Case (iv) that the limits (7.3) hold for the
∞
[
measurable set E = En and we omit the details here.
n=1
Remark 7.1
The limits defined in (7.2) are called the upper and lower densities of E at 0 respectively.
Problem 7.3
Rudin Chapter 7 Exercise 3.
7.2. Periods of Functions and Lebesgue Measurable Groups 211
for x > α. If y > x > α + pi , then we deduce from Theorem 2.20(c) and the fact
[α − pi , α + pi ] = [α − pi , x] \ (α + pi , x]
that
F (x + pi ) − F (x − pi ) 1 h F (x + pi ) − F (x) F (x − pi ) − F (x) i
lim = lim + = F ′ (x). (7.10)
i→∞ 2pi i→∞ 2 pi −pi
Combining the results (7.9) and (7.10), we see that F ′ (x) = F ′ (y) for all x, y > α and thus
F ′ (x) = cα (7.11)
for all x > α, where cα is a constant. Next, we recall from the definition of F that
Z x
F (x) = m(E ∩ [α, x]) = χE (t) dt.
α
Since χE ∈ L1 (R), it establishes from Theorem 7.11 that F is differentiable and F ′ (x) = χE (x)
a.e. for x > α. Using this and the fact (7.11), we get
χE (x) = cα (7.12)
a.e. for x > α. Since χE takes only 0 or 1, we see that cα ∈ {0, 1}. If x > β > α, then
so cα does not depend on α and we can simply replace cα by c in the results (7.11) and (7.12).
Finally, if c = 0, then we conclude from the result (7.12) that
Z
m(E) = χE dm = 0.
R
Otherwise, c = 1 and the result (7.12) implies that χE c (x) = 0 and hence m(E c ) = 0. We have
completed the proof of the problem.
Problem 7.4
Rudin Chapter 7 Exercise 4.
212 Chapter 7. Differentiation
Proof. Our proof does not follow Rudin’s hint. In fact, we mainly follow the argument of Cignoli
and Hounie [27] and now we first quote one of their results that will be used very soon:
Lemma 7.1
Let D be a dense subset of R and µ a Borel measure on R such that
µ d + [a, b) = µ [a, b)
for every d ∈ D and a, b ∈ R. Then we have µ = km, where k = µ (0, 1] .
Next, it is clear that the set G(f ) = {ns + mt | n, m ∈ Z} is an additive group of R. Without
loss of generality, we assume that t > 0. Since α = s/t is irrational, it follows from the
Kronecker’s Approximation Theorema that for every θ ∈ R and every ǫt−1 > 0, there exists
n, m ∈ Z such that
θ ǫ
nα + m − <
t t
which is equivalent to saying that
|ns + mt − θ| < ǫ.
Hence G(f ) is dense in R.
We first consider the special case that f is nonnegative and bounded a.e. on R. Note that
if F (x) = f (xt), then F (x + 1) = f (xt + t) = f (xt) = F (x), so F is a function of period 1 and
we may assume that f has period 1. Consequently, it suffices to prove the problem when f is a
nonnegative, bounded a.e. function on [0, 1] having period 1. Suppose that µ : B → [0, ∞) is
defined by Z
µ(E) = f (x) dx (7.13)
E
for all E ∈ B. By Theorem 1.29, µ is a Borel measure. Let p be a period of f . Then we deduce
from the representation (7.13) that
Z Z Z
µ(E + p) = f (x) dx = f (x + p) dx = f (x) dx = µ(E).
E+p E E
In other words, µ has the same periods as f . Furthermore, if p ∈ G(f ), then we obtain from
Theorem 7.26 (The Change-of-variables Theorem)b that
Z b+p Z b Z b
µ p + [a, b) = µ [a + p, b + p) = f (x) dx = f (x + p) dx = f (x) dx = µ [a, b) .
a+p a a
then it is trivial that f has periods s and t as well as f = 1 a.e. on R.c This completes the
proof of the problem.
Problem 7.5
Rudin Chapter 7 Exercise 5.
• A + B contains a segment. Since m(A) > 0 and m(B) > 0, A and B are Lebesgue
measurable. By the comment in §7.12, the metric densities of A and B are 1 at almost
every point of A and B. Thus we pick a0 ∈ A and b0 ∈ B such that
m A ∩ B(a0 , r) m B ∩ B(b0 , r)
lim = lim = 1. (7.15)
r→0+ m B(a0 , r) r→0+ m B(b0 , r)
Since χA , χB ∈ L1 (R), we may assume with the aid of Theorem 7.7 that a0 and b0 are
Lebesgue points of χA and χB respectively.
Set c0 = a0 + b0 . For each ǫ > 0, define
We claim that A ∩ Bǫ 6= ∅ for all small enough ǫ > 0. Otherwise, no matter how small
ǫ > 0 is, we can find a η ∈ (0, ǫ) such that A ∩ Bη = ∅. In fact, we may pick η small
enough so that the metric density of A at a0 in (7.15) implies
7
m A ∩ B(a0 , 4η) > m B(a0 , 4η) = 7η. (7.21)
8
Since A ∩ Bη = ∅, we have A ∩ [B(a0 , 4η) \ Bη ] = [A ∩ B(a0 , 4η)] \ [A ∩ Bη ] = A ∩ B(a0 , 4η)
so that
m A ∩ B(a0 , 4η) = m A ∩ [B(a0 , 4η) \ Bη ] ≤ m B(a0 , 4η) \ Bη . (7.22)
Recall from the set relation (7.17), we have the fact B(a0 , 4η) = [B(a0 , 4η) \ Bη ] ∪ Bη . By
this fact and the inequality (7.20), we may further reduce the inequality (7.22) to
m A ∩ B(a0 , 4η) ≤ m B(a0 , 4η) − m(Bη ) < 8η − 2η = 6η. (7.23)
However, the inequalities (7.21) and (7.23) are contradictory and this proves our claim
that A ∩ Bǫ 6= ∅ for all small ǫ > 0.
Pick x ∈ (c0 − ǫ′ , c0 + ǫ′ ) for sufficiently small ǫ′ and without loss of generality, suppose
that ǫ = x − c0 > 0. Then we have 0 < ǫ < ǫ′ and the above claim implies that
A ∩ Bǫ 6= ∅.
Choose a ∈ A ∩ Bǫ . By the definition (7.16) and then using the fact ǫ = x − c0 , we have
a = c0 + ǫ − b = x − b
for some b ∈ B with |b − b0 | < 2ǫ. In other words, we have the representation
x = a + b ∈ A + B,
i.e., (c0 − ǫ′ , c0 + ǫ′ ) ⊆ A + B.
C C
• 2 + 2 = [0, 1]. Recall from [99, Exercise 19, p. 81] that if c ∈ C, then
∞
X αn
c= ,
3n
n=1
Then it is always true that an = βn + γn for all n = 1, 2, . . .. Thus the x given in (7.24)
can be represented as
x = y + z.
where βn , γn ∈ {0, 1} for all n = 1, 2, . . ..d In other words, the analysis implies that
[0, 1] ⊆ 21 C + 21 C.
On the other hand, let y, z ∈ 12 C be expressed by the series (7.25), where βn , γn ∈ {0, 1}.
Then we have
X∞
βn + γn
y+z = ,
3n
n=1
Otherwise, y + z ∈ (0, 1). Therefore, we have 12 C + 12 C ⊆ [0, 1]. Hence we have obtained
the required result that 21 C + 12 C = [0, 1].
Remark 7.2
The result in Problem 7.5 is called the Steinhaus Theorem. It has a generalization in Rk
which says that if A ∈ B k and mk (A) > 0, then 0 is an interior point of A − A, see [13,
Theorem 26.6, p. 163] for details.
Problem 7.6
Rudin Chapter 7 Exercise 6.
Proof. Assume that m(G) > 0. By Problem 7.5, G + G contains a segment (a, b). Since G is
a group with addition, we have G + G ⊆ G so that (a, b) ⊆ G. Let m = a+b 2 ∈ (a, b) so that
m ∈ G. Since G is a group, −m ∈ G so that (a − m, b − m) ⊆ G. Since m is the mid-point of a
and b, the segment (a − m, b − m) is in the form (−c, c) for some fixed c ∈ R \ {0}. Recall that
G + G ⊆ G, so we have (−2c, 2c) ⊆ G. Repeat this kind of argument, we may conclude that
(−nc, nc) ⊆ G
for all n ∈ N, but this implies that G = R, a contradiction. Hence m(G) = 0 and this completes
the proof of the problem.
d
Since the series (7.25) converge, we can add them together.
216 Chapter 7. Differentiation
Problem 7.7
Rudin Chapter 7 Exercise 7.
Proof. The construction has been done in [46, Example 15, pp. 96 – 98]. We won’t repeat it
here. See also [46, Example 30, pp. 105, 195].
Problem 7.8
Rudin Chapter 7 Exercise 8.
(a) Obviously, V is open in R and Lebesgue measurable. Since 0 ≤ ϕn (x) < 2−n in V and
∞
X
2−n < ∞, the Weierstrass M -test [99, Theorem 7.10, p. 148] indicates that
n=1
∞
X
ϕ= ϕn (7.26)
n=1
for every x ∈ [a, b]. Therefore, it follows from the First Fundamental Theorem of Calculus
that T is differentiable at every point of the continuity of ϕ in [a, x]. Consequently, T is
differentiable (and hence continuous) at every point of V .
Next, if T (x) = T (y) for x < y, then we have
Z y Z y
T (y) = ϕ(t) dt = T (x) + ϕ(t) dt
a x
a contradiction to the result (7.27). Thus T is one-to-one and condition (ii) is satisfied.
Since T is differentiable at every point of V , it is continuous on V and condition (i) is also
satisfied. Condition (iii) is obvious because X \ X = ∅.
7.3. The Cantor Function and the Non-measurability of f ◦ T 217
(b) By the First Fundamental Theorem of Calculus, we know that T ′ = ϕ which is continuous
/ Wn for all n ∈ N which implies that ϕn (p) = 0 for all n ∈ N.
on V . If p ∈ K, then p ∈
Therefore, we have
∞
X
T ′ (p) = ϕ(p) = ϕn (p) = 0. (7.28)
n=1
For the third assertion, we first show that T (K) is a Borel set. To this end, since T is
injective, we note from [74, Exercise 2(h), p. 21] that
∞
[
T (K) = T (V \ W ) = T (V ) \ T (W ) = T (V ) \ T (Wn ).
n=1
where |JT (x)| denotes the Jacobian of T at x. Recall that T is injective, so T (x) ∈ T (K)
if and only if x ∈ K and then we reduce from the expression (7.29) that
Z Z
m T (K) = χT (K) T (x) |JT | dm = χK (x)|JT | dm. (7.30)
V K
Notice from the result (7.28) that |JT (x)| = | det T ′ (x)| = 0 on K, so we obtain from the
integral (7.30) that m T (K) = 0.
(c) Since m(K) > 0, it follows from Corollary to Theorem 2.22 that K contains a nonmeasur-
able subset E. Let A = T (E). Since E ⊂ K, we have A ⊂ T (K). Since m T (K) = 0, we
recall from Theorem 1.36 that all subsets of sets of measure 0 are measurable, thus A is a
measurable set and χA is a measurable function. By the definition, we know that
1, if x ∈ E;
χA T (x) = χT (E) T (x) =
0, otherwise.
In other words, we have χA ◦ T = χE . Since E is nonmeasurable, χE is not measurable.
(d) By part (b), we know that T ′ (x) = ϕ(x) on V . Thus if we can construct a ϕ which
is infinitely differentiable, then we are done. To see this, we are going to construct an
infinitely differentiable function ϕn on Wn for each positive integer n. Let’s recall the
bump function ψ : R → R defined by
1
exp − , if x ∈ (−1, 1);
1 − x2
ψ(x) =
0, otherwise.
The ψ has derivatives of all orders in R, i.e., ψ is an element of C ∞ (R) the space of all
infinitely differentiable functions on R. Since each Wn is a segment, we let Wn = (an , bn )
for some an , bn ∈ (a, b) with an < bn . Therefore the bump function ϕn : V → R given by
2 x − an +bn
ϕn (x) = 2−n ψ (x − bn ) + 1 = 2−n ψ 2
bn −an
bn − a n 2
218 Chapter 7. Differentiation
Problem 7.9
Rudin Chapter 7 Exercise 9.
Proof. The function f discussed in §7.16 is called a singular function (see [17, p. 161] or [59,
Definition 7.1.44, p. 170]). Let 0 < α < 1. Pick t so that tα = 2. Let δn = ( 2t )n = 2n t−n for all
n ∈ N.
We remark that comprehensive materials about the relations between singular functions and
Cantor sets can be found in [58, Chap. 4] and [59, Chap. 8]. In fact, our proof here relies
heavily on the known results there.
We fix some notations first. Let n ≥ 0. At the nth step, we have
2n
[
En = In (k), (7.31)
k=1
where In (1), In (2), . . . , In (2n ) are the 2n disjoint closed intervals, each of length 2−n δn = t−n .
Next the deleted open segment between the intervals In (k) and In (k +1) is denoted by J(k ·2−n ),
where k = 1, 2, . . . , 2n − 1. Suppose that
D = {m · 2−n | 0 ≤ m ≤ 2n and n ≥ 0}
for all x ∈ J(k · 2−n ) and k · 2−n ∈ D. To this end, the hypothesis x ∈ J(k · 2−n ) implies that
there are k closed disjoint intervals In (1), In (2), . . . , In (k) on its left-hand side, so we derive from
this and the fact [100, Eqn. (5), p. 145] that
Z x k Z
X
fn (x) = gn (t) dt = gn (t) dt = k · 2−n . (7.33)
0 m=1 In (m)
Since x ∈
/ En by the definition (7.31), we see from [100, Eqn. (6), p. 145] and the values (7.33)
that
f (x) = lim fn (x) = k · 2−n .
n→∞
7.4. Problems related to the AC of a Function 219
∞
[
This proves our claim. By [100, Eqn. (3), p. 145], we have E c = Enc . Since x ∈
/ En , we have
n=1
x ∈ Enc ⊆ E c . Therefore, the f is well-defined on E c by the values (7.32).
By [58, p. 84] or [59, p. 202], E is nowhere dense in [0, 1] so that E c is a dense subset of
[0, 1]. Next, since f is uniformly continuous on [0, 1] and particularly on E c , it follows from [99,
Exercise 13, pp. 99, 100] that f has a unique continuous extension on [0, 1] which agrees with
f on E c . Hence the uniqueness shows that our f considered in §7.16 is the so-called Lebesgue
function associated with the Cantor-like set E. Let h : [0, 1] → R be defined by
h(x) = xα .
Since α ∈ (0, 1), it is immediate from basic calculus that h is strictly concave on [0, 1]. Clearly,
we know that
1
h(2−n δn ) = (2−n δn )α = nα = 2−n , (7.34)
t
where n = 0, 1, 2, . . .. Thus it gives
h 2−(n−1) δn−1 = 2−(n−1) = 2 · 2−n = 2h(2−n δn ) > h(2−n+1 δn )
for all n = 1, 2, . . .. Since h is strictly increasing on [0, 1], we obtain 2−(n−1) δn−1 > 2−n+1 δn
which is equivalent to δn−1 > δn for each n = 1, 2, . . .. By the formula (7.34), we know that
h(δ0 ) = 1 which implies δ0 = 1. In other words, the Cantor-like set E in §7.16 is constructed
by means of our fixed concave function h. By [58, p. 97] or [59, Theorem 8.5.12, p. 205], it
concludes that
|f (x) − f (y)| ≤ h |x − y| = |x − y|α
for all x, y ∈ [0, 1]. Hence f belongs to Lip α on [0, 1], completing the proof of the problem.
Problem 7.10
Rudin Chapter 7 Exercise 10.
Problem 7.11
Rudin Chapter 7 Exercise 11.
Since f ′ ∈ Lp , we apply Theorem 3.5 (Hölder’s Inequality) to the right-hand side of the inequality
(7.36) to obtain
nZ x o1 n Z x o1
p q
|f (x) − f (y)| ≤ |f ′ (t)|p dt dt ≤ kf ′ kp · |x − y|α
y y
and thus
|f (x) − f (y)|
sup ≤ kf ′ kp < ∞.
x,y∈[a,b] |x − y|α
x6=y
By the definition (see Problem 5.11), we conclude that f ∈ Lip α, completing the proof of the
problem.
Problem 7.12
Rudin Chapter 7 Exercise 12.
Proof.
(a) Since ϕ is nondecreasing, we follow from [99, Theorems 4.29 & 4.30, pp. 95, 96] that ϕ(x+)
and ϕ(x−) exist for every point x of [a, b] and the set of discontinuities of ϕ is at most
countable. Let this set be A = {x1 , x2 , . . .} and x1 < x2 < · · · . We define f : [a, b] → R by
f (x) = sup ϕ(t). In fact, we have
t<x
ϕ(x), if x ∈ [a, b] \ A;
f (x) = (7.37)
lim ϕ(xk − t) = ϕ(xk −), if x = xk for some k ∈ N.
t→0
t>0
Let x < y and x, y ∈ [a, b]. Then the nondecreasing property of ϕ ensures that
Clearly, {x ∈ [a, b] | f (x) 6= ϕ(x)} ⊆ A is at most countable. Given x ∈ (a, b] and ǫ > 0.
By the definition (7.37), if x = xk , then there is a δ > 0 such that
for all t ∈ (0, δ). By the construction of A, the number δ can be chosen so small such that
xk − t ∈/ A for all t ∈ (0, δ), i.e., xk − t ∈ [a, b] \ A for all t ∈ (0, δ). Then ϕ(x) is continuous
at every xk − t and ϕ(xk − t) = f (xk − t). Thus the inequality (7.38) reduces to
Now our claim (7.40) follows directly by taking limit N → ∞ in the inequality (7.42).
Secondly, we define h : [a, b] → R by
X ∞
X
h(x) = g(t)χ(t,b] (x) = g(xk )χ(xk ,b] (x). (7.43)
t∈A k=1
222 Chapter 7. Differentiation
By the claim (7.40) (or the second inequality in (7.42)) and the Weierstrass M -test, we
know that the sum (7.43) actually converges uniformly on [a, b]. If a ≤ x < y ≤ b, then
we have χ(t,b] (x) ≤ χ(t,b] (y) and this shows that
X X
h(x) = g(t)χ(t,b] (x) ≤ g(t)χ(t,b] (y) = h(y),
t∈A t∈A
Lemma 7.2
The function F is continuous on [a, b].
and
∞
X m−1
X
h(xm −) = lim h(xm + δ) = lim g(xk )χ(xk ,b] (xm + δ) = [ϕ(xk +) − ϕ(xk −)].
δ→0 δ→0
δ<0 δ<0 k=1 k=1
Lemma 7.3
The function F is nondecreasing on [a, b].
Proof of Lemma 7.3. Let x, y ∈ [a, b] be x < y. Suppose that xm < x ≤ xm+1 for some
m ∈ N. If xm < x < y ≤ xm+1 , then the definition (7.43) gives
∞
X ∞
X
h(y) − h(x) = g(xk )χ(xk ,b] (y) − g(xk )χ(xk ,b] (x)
k=1 k=1
Xm Xm
= g(xk )χ(xk ,b] (y) − g(xk )χ(xk ,b] (x)
k=1 k=1
=0
≤ f (y) − f (x).
Suppose that
xm < x ≤ xm+1 < · · · < xm+p ≤ y (7.44)
for some p ∈ N, then we get from the definition (7.39) that
∞
X ∞
X
h(y) − h(x) = g(xk )χ(xk ,b] (y) − g(xk )χ(xk ,b] (x)
k=1 k=1
Xm m+p
X m
X
= g(xk )χ(xk ,b] (y) + g(xk )χ(xk ,b] (y) − g(xk )χ(xk ,b] (x)
k=1 k=m+1 k=1
m+p
X
= g(xk )χ(xk ,b] (y)
k=m+1
m+p
X
≤ [ϕ(xk +) − ϕ(xk −)]. (7.45)
k=m+1
If we apply the same trick (the sequence (7.41) is replaced by the sequence (7.44) with
a and b are replaced by x and y respectively) as in proving the inequality (7.42), we
can easily deduce from the inequality (7.45) that h(y) − h(x) ≤ f (y) − f (x).
In conclusion, we have shown that
Fourthly, we express F in terms of a finite and positive Borel measure. By Lemmas 7.2
and 7.3, we know that the function
g(x) = x + F (x)
is continuous and strictly increasing on [a, b]. Then g is one-to-one on [a, b] so that its in-
verse function g −1 : [g(a), g(b)] → [a, b] is a continuous and bijective function on [g(a), g(b)]
224 Chapter 7. Differentiation
by [99, Theorem 4.17, p. 90]. For every E ⊆ [a, b] and E ∈ B, since g−1 is continuous on
[g(a), g(b)], Definition 1.2(c) tells us that g(E) = (g −1 )−1 (E) ∈ B so that we are able to
define
ν(E) = m g(E) . (7.46)
We have to show that ν is a positive Borel measure, but it follows exactly the same
reasons as in [100, p. 147]. Furthermore, since m is a finite measure, the construction
(7.46) implies that ν is also a finite measure. By Theorem 6.10 (The Lebesgue-Radon-
Nikodym Theorem), there exists a unique pair of positive and finite Borel measures νac
and νsc such that
νac ≪ m, νsc ⊥ m and ν = νac + νsc .
Now the function F and the measures ν and m are linked up in the following lemma:
Lemma 7.4
Let µ′ = ν − m. Then µ′ is finite and positive Borel measure. Furthermore, we
have
µ′ [a, x) = F (x) − F (a). (7.47)
Proof of Lemma 7.4. Clearly, the measure µ′ is a complex Borel measure. The finite-
ness of µ′ is also clear. If x ∈ [a, b], then we deduce from the definition (7.46) that
µ′ [a, x) = m g[a, x) − m([a, x)) = g(x) − g(a) − x + a = F (x) − F (a).
It remains to verify that µ′ is positive. To this end, it is obvious from the definition
and Lemma 7.3 that for x, y ∈ [a, b] with x < y, we have
ν (x, y) = m g (x, y) = g(y) − g(x) = y + F (y) − x − F (x) ≥ y − x = m (x, y) .
Suppose that V is an open set in [a, b]. By [99, Exercise 29, p. 45], V can be expressed
as a union of an at most countable collection of disjoint segments {Vi }. Since µ and
m are measures, we have
[
∞ [∞ ∞
[ [
∞
ν(V ) = ν Vi = ν(Vi ) ≥ m(Vi ) = m Vi = m(V ). (7.48)
i=1 i=1 i=1 i=1
Since ǫ is arbitrary, we establish the fact that ν(E) ≥ m(E) and then µ′ (E) ≥ 0 for
all E ∈ B and E ⊆ [a, b]. This proves the positivity of the measure µ′ and ends the
proof of Lemma 7.4.
7.4. Problems related to the AC of a Function 225
We come to the final step of the proof of part (b). Recall that a complex Borel measure
µ′′ is called a P P is a countable set {xj } ⊂ R and complex numbers
discrete measure if there
cj such that |cj | < ∞ and µ′′ = cj δxj , where δx is the point mass at x, see [40, p.
106] or [100, Example 1.20(b), p. 17]. Now if we let
∞
X
µ′′ (E) = g(xk )δxk (E) ≥ 0
k=1
for any E ⊆ [a, b] and E ∈ B, then it follows from the fact (7.40) that µ′′ is finite. Besides,
we deduce from this and the definition (7.43) easily that
∞ ∞
X X
µ′′ [a, x) = g(xk )δxk [a, x) = g(xk )χ(xk ,b] (x) = h(x) − h(a). (7.50)
k=1 k=1
Therefore, if we further define µ = µ′ + µ′′ , then µ must be a finite positive Borel measure
on [a, b] and it yields from the representations (7.47) and (7.50) that
µ([a, x)) = µ′ ([a, x)) + µ′′ ([a, x)) = F (x) − F (a) + h(x) − h(a) = f (x) − f (a). (7.51)
µ = µac + µsc ,
where µac ≪ m and µsc ⊥ m. Suppose that ω is the Radon-Nikodym derivative of µac
with respect to m. Since ω ∈ L1 (m) on [a, b], almost every x ∈ [a, b] is a Lebesgue point
of ω by Theorem 7.7.
Let x ∈ (a, b) be a Lebesgue point of ω. Define r = min(x − a, b − x) and we can choose
{ri } ⊆ (0, r) such that ri → 0 as i → ∞. Then the two sequences of sets {Ei } and {Ei′ }
given by
Ei = [x, x + ri ) and Ei′ = [x − ri , x)
satisfy the inequalities
1 1 1
m(Ei ) = ri ≥ · 2ri = m (x − ri , x + ri ) and m(Ei′ ) ≥ m (x − ri , x + ri )
2 2 2
for all i = 1, 2, . . .. In other words, they mean that {Ei } and {Ei′ } shrink to x nicely.
Since µ is a complex Borel measure on [a, b] and dµ = ω dm + dµs , Theorem 7.14 and the
formula (7.51) imply that
f (x + ri ) − f (x) µ [x, x + ri )
f+′ (x) = lim = lim = ω(x) (7.52)
i→∞ ri i→∞ m [x, x + ri )
a.e. [m] on [a, b]. Now the results (7.52) and (7.53) definitely give f ′ (x) exists a.e. [m] on
[a, b] and f ′ = ω ∈ L1 (m) on [a, b].
Finally, since dµac = ω dm, we have
Z x Z x Z x
µac [a, x) = ω dm = f ′ dm = f ′ (t) dt
a a a
226 Chapter 7. Differentiation
so that
Z x
f (x) − f (a) = µ [a, x) = µac [a, x) + µsc [a, x) = f ′ (t) dt + s(x),
a
where s(x) = µsc [a, x) for all x ∈ [a, b]. Now µsc [a, x) is positive finite because µ is
positive finite, so s is nondecreasing on [a, b]. Since f ′ ∈ L1 (m) on [a, b], Theorem 7.11
shows that Z x ′
s′ (x) = [f (x) − f (a)]′ − f ′ (t) dt = f ′ (x) − f ′ (x) = 0
a
a.e. [m] on (a, b).
(d) By Theorem 6.8(e), µ ⊥ m if and only if µac = 0 if and only if
Z
f ′ dm = 0 (7.54)
E
for every E ∈ B and E ⊆ [a, b]. Recall that f ′ ∈ L1 (m). By Theorem 1.39(b), the result
(7.54) is equivalent to f ′ (x) = 0 a.e. [m] on [a, b].
Similarly, µ ≪ m if and only if µsc = 0 if and only if s = 0 on [a, b] if and only if
Z x
f (x) − f (a) = f ′ dm (7.55)
a
for all x ∈ [a, b]. As Rudin pointed out in the discussion preceding Definition 7.17 (on p.
145), one may apply Theorem 6.11 to show that f is AC on [a, b] if the formula (7.55)
holds. Conversely, if f is AC on [a, b], then f is automatically continuous on [a, b] and
since f is nondecreasing, Theorem 7.18 implies that the formula (7.55) holds.
(e) Suppose that f ′ (p) exists at p ∈ (a, b) \ A. Then f is continuous at p and the definition
(7.37) implies the existence of a δ > 0 such that
f (x) = ϕ(x)
for all x ∈ (p − δ, p + δ). If h ∈ (0, δ), then p + h ∈ (p − δ, p + δ) and so
ϕ(p + h) − ϕ(p) f (p + h) − f (p)
− f ′ (p) = − f ′ (p). (7.56)
h h
By taking h → 0+ in the expression (7.56), we get
|ϕ′+ (p) − f ′ (p)| = |f+′ (p) − f ′ (p)| = 0,
i.e., ϕ′+ (p) = f ′ (p). The other side ϕ′− (p) = f ′ (p) can be done similarly. Thus we conclude
that ϕ′ (p) = f ′ (p). Since f ′ (x) exists a.e. [m] on [a, b], we have ϕ′ (x) = f ′ (x) a.e. [m] on
[a, b].
Hence we have completed the proof of the problem.
Remark 7.3
By the result of part (b) in Problem 7.12, we have
singular
continuous
′ ′′ ′′ z}|{
µ = µ + µ = v − m + µ = (νa − m) + νs + µ′′ .
| {z } |{z}
absolutely discrete
continuous
Problem 7.13
Rudin Chapter 7 Exercise 13.
where the supremum is taken over all N and over all choices of {ti } such that
Thus f is bounded on [a, b]. By the definition (7.59), f1 and f2 are also bounded on
[a, b]. If f is nonincreasing, then the function −f is nondecreasing but the functions
F −(−f ) F +(−f )
f1 = F +f
2 = 2 and f2 = F −f
2 = 2 remain nondecreasing.f
F (x) − F (x − h) = F (x − h; x)
e
The conclusion of this assertion does not depend on the assumption that f is AC on [a, b].
f
If we write f = (−f2 ) − (−f1 ), then −f1 and −f2 are nonincreasing.
228 Chapter 7. Differentiation
for every small h > 0 such that x − h ∈ [a, b]. By the definition, if {ti } and {sj } are
partitions of [a, x − h] and [x − h, x] respectively, then {ti } ∪ {sj } is certainly a partition
of [a, b] and so
F (x) ≥ F (x − h) + F (x − h; x). (7.60)
On the other hand, let {t0 , t1 , . . . , tN } be a partition of [a, x]. If tM < x − h < tM +1 , then
we may insert a number t′ such that
Since the partition {t0 , t1 , . . . , tN } is arbitrary, the inequality (7.61) asserts that
Hence our desired claim follows immediately from the inequalities (7.60) and (7.62).
Since f is left-continuous at p ∈ (a, b], given ǫ > 0, there exists a δ > 0 such that
h ∈ (0, δ) implies
ǫ
|f (p) − f (p − h)| < .
2
For this same ǫ, we can find a partition {ti } of [a, p], say
such that
N
ǫ X
F (a; p) − < |f (ti ) − f (ti−1 )| (7.63)
2
i=1
Since adding more points to the partition {ti } results in an increase of the summation
(7.63), we may assume that tN − tN −1 = h ∈ (0, δ) and this implies that
ǫ
|f (tN ) − f (tN −1 )| = |f (tN ) − f (tN − h)| = |f (p) − f (p − h)| <
2
and the inequality (7.63) becomes
N −1
ǫ ǫ X ǫ
F (a; p) − < + |f (ti ) − f (ti−1 )| ≤ + F (a; p − h)
2 2 2
i=1
or equivalently,
F (a; p) − F (a; p − h) < ǫ.
However, we note that
which is exactly the expected result (7.59). Hence F and then f1 as well as f2 are left-
continuous on [a, b].
7.4. Problems related to the AC of a Function 229
(d) By parts (b) and (c), f = f1 − f2 , where f1 and f2 are left-continuous and nondecreasing
on [a, b]. Now Problem 7.12(b) shows that there exist positive and finite Borel measures
λ and ν on [a, b] such that
f1 (x) − f1 (a) = λ [a, x) and f2 (x) − f2 (a) = ν [a, x) ,
where x ∈ [a, b]. Then the difference µ = λ − ν is easily seen to be a Borel measure on
[a, b] satisfying the equation in the question. In fact, the finiteness of λ and ν imply the
finiteness of µ. (However, µ may not be positive.)
As Theorem 6.10 (The Lebesgue-Radon-Nikodym Theorem) tells us that there exists a
unique pair of positive and finite measures µa ≪ m and µs ⊥ m such that µ = µa + µs .
Furthermore, there is a unique h ∈ L1 (m) on [a, b] such that
Z
µa (E) = h dm (7.64)
E
for every E ∈ B.
If µ ≪ m, then Proposition 6.8(e) implies that µs = 0 and thus the µa in the represen-
tation (7.64) can be replaced by µ. Now we put E = [a, x) into the representation and use
the result f (x) − f (a) = µ([a, x)) to obtain
Z Z x
f (x) − f (a) = µ([a, x)) = h dm = h dm.
[a,x) a
Since h ∈ L1 (m)
on [a, b], we follow from [18, Theorem 5.3.6, p. 339] or [40, Theorem 3.35,
p. 106] that f is AC on [a, b].g Conversely, suppose that f is AC on [a, b]. By Theorem
7.19, F, f1 and f2 are AC and nondecreasing on [a, b]. By Problem 7.12(d), we conclude
that
λ ≪ m and ν ≪ m.
Since µ = λ − ν, Proposition 6.8(c) ensures that µ ≪ m.
(e) By part (b), we have f = f1 − f2 , where f1 and f2 are nondecreasing. By Problem 7.12(e),
f1′ and f2′ exist a.e. [m] on [a, b]. By Problem 7.12(c), we further know that f1′ , f2′ ∈ L1 (m)
on [a, b]. Hence we conclude that f ′ exists a.e. [m] on [a, b] and f ′ ∈ L1 (m).
Problem 7.14
Rudin Chapter 7 Exercise 14.
Proof. Suppose that f and g are AC on [a, b]. By Definition 7.17, they are continuous on [a, b],
so the Extreme Value Theorem ([99, Theorem 4.16, p. 89]) ensures that we may assume that
there exists a M > 0 such that
for every n and any disjoint collection of segments (α1 , β1 ), . . . , (αn , βn ) in [a, b] with
n
X
(βk − αk ) < δ.
k=1
It is clear that
n
X n
X
|f (βk )g(βk ) − f (αk )g(αk )| = |f (βk )g(βk ) − f (βk )g(αk ) + f (βk )g(αk ) − f (αk )g(αk )|
k=1 k=1
Xn n
X
≤ |f (βk )g(βk ) − f (βk )g(αk )| + |f (βk )g(αk ) − f (αk )g(αk )|
k=1 k=1
n
X n
X
≤M |g(βk ) − g(αk )| + M |f (βk ) − f (αk )|. (7.67)
k=1 k=1
By the bounds (7.65), we see that f ′ g, f g′ ∈ L1 on [a, b]. Furthermore, since (f g)′ = f ′ g + f g′
a.e. on [a, b], we deduce from the formula (7.68) that
Z b Z b
f ′ (x)g(x) dx = f (b)g(b) − f (a)g(a) − f (x)g′ (x) dx,
a a
Problem 7.15
Rudin Chapter 7 Exercise 15.
Proof. It is well-known [99, Example 5.6(b), p. 106] that the function f : R → R defined by
1
x2 sin , if x 6= 0;
f (x) = x
0, otherwise
Thus f is differentiable at all points but f ′ is not a continuous function. However, such function
oscillates around 0 so that it is not monotonic and we have to “modify” it. To do this, we notice
that since
1 sin( x1 )
lim x sin = lim = 1,
x→±∞ x x→±∞ x1
f ′ (x) is bounded on R. Let M > 0 be a bound of f ′ on R, i.e.,
−M ≤ f ′ (x) ≤ M (7.69)
F (x) = f (x) + M x.
Now F ′ (x) = f ′ (x) + M which exists finitely on R. In addition, the inequalities (7.69) implies
that
F ′ (x) ≥ −M + M ≥ 0
for all x ∈ R. Hence [99, Theorem 5.11, p. 108] ensures that F is monotonically increasing on R.
Since f ′ is not continuous at 0, F ′ is also discontinuous at 0 and thus F is our desired function.
This completes the proof of the problem.
Problem 7.16
Rudin Chapter 7 Exercise 16.
Proof. We follow the hint given by Rudin. Since m is outer regular (see Theorem 2.14(c) and
Theorem 2.20(b)) and m(E) = 0, we can find a sequence {Vn } of open subsets of R such that
E ⊆ · · · ⊆ V2 ⊆ V1 and m(Vn ) < 2−n . This construction gives
∞
\
E⊆ Vn . (7.70)
n=1
which is clearly nondecreasing. By Theorem 7.11, f is differentiable a.e. on [a, b] and then we
follow from Theorem 7.18 that f is AC on [a, b].
Now it remains to show that f ′ (x) = ∞ on E. To see this, if x ∈ E, then the set relation
(7.70) tells us that x ∈ Vn for all n ∈ N. Since each Vn is open, there exists a δn > 0 such that
(x − δn , x + δn ) ⊆ Vn
232 Chapter 7. Differentiation
for all n ∈ N. In fact, we may assume that {δn } is a decreasing sequence of positive numbers.
Pick y ∈ (x − δn , x + δn ) so that
Z y Z n
yX
|f (y) − f (x)| = F (t) dt ≥ [χV1 (t) + · · · + χVn (t)] dt (7.72)
x x k=1
By the definition (7.71), we conclude that f ′ (x) = ∞ on E. We have completed the proof of the
problem.
Problem 7.17
Rudin Chapter 7 Exercise 17.
where
µa , µn,a ≪ m and µs , µn,s ⊥ m (7.73)
for all n ∈ N. Let
∞
X ∞
X
λa = µn,a and λs = µn,s .
n=1 n=1
h
Or t ∈ [y, x].
7.5. Miscellaneous Problems on Differentiation 233
Since µ and µn are positive, so are µa , µn,a , µs and µn,s and it deduces from the first
assertion that both λa and λs are positive Borel measures of Rk . Furthermore, we have
∞
X ∞
X ∞
X ∞
X
µ= µn = (µn,a + µn,s ) = µn,a + µn,s = λa + λs .
n=1 n=1 n=1 n=1
By Theorem 6.8 and the relations (7.73), we assert that λa ≪ m and λs ⊥ m. Recall that
the Lebesgue decomposition of µ relative to m is unique, it must be true that
∞
X ∞
X
µ a = λa = µn,a and µs = λs = µn,s .
n=1 n=1
for all E ∈ B. Similarly, for each n ∈ N, there exists a unique hn ∈ L1 (m) such that
Z
µn,a (E) = hn dm (7.75)
E
for all E ∈ B. Since µa , µn,a ≪ m for all n ∈ N, we observe from Theorem 7.8 that
a.e. [m] for all n ∈ N. Since µs ⊥ m and µn,s ⊥ m for all n ∈ N, Theorem 7.14 ensures
that
Dµs = 0 and Dµn,s = 0 (7.77)
a.e. [m] for all n ∈ N. Hence the two results (7.76) and (7.77) say that
Dµ = h and Dµn = hn
for all E ∈ B. Since hn ∈ L1 (m), each hn is measurable and Theorem 1.27 implies that
the expression (7.78) can be further reduced to
Z Z X
∞
h dm = hn dm
E E n=1
for all E ∈ B.
234 Chapter 7. Differentiation
N
X
Since µa (E) ≥ µn,a(E) for every N ∈ N, the integrals (7.74) and (7.75) show that
n=1
Z N
X
h− hn dm ≥ 0 (7.80)
E n=1
N
X
for every E ∈ B. Assume that h < hn on E ′ with m(E ′ ) > 0. For this Borel set E ′ ,
n=1
we derive from Proposition 1.24(a) that
Z N
X
h− hn dm < 0
E′ n=1
N
X
which contradicts the result (7.80). Consequently, we always have h ≥ hn a.e. [m] on
n=1
Rk and since N is arbitrary, this implies that
∞
X
h− hn ≥ 0
n=1
a.e. [m] on Rk . Now we apply Theorem 1.39(a) to the integral (7.79) to conclude that
∞
X
h(x) = hn (x)
n=1
a.e. [m] on Rk .
a.e. [m] on R. By the definition, it is trivial that f is nondecreasing and 0 < f (x) < ∞
on R. Let [a, b] be a compact interval of R. Suppose that
By Problem 7.12(e), we see that f and each fn are differentiable a.e. [m] on [a, b] so that
[a, b] \ E is of measure zero.
Let x ∈ E and h > 0 be sufficiently small so that x + h ∈ [a, b]. Then for any N ∈ N,
we have
N
f (x + h) − f (x) X fn (x + h) − fn (x)
≥
h h
n=1
7.5. Miscellaneous Problems on Differentiation 235
a.e. [m] on [a, b]. Recall that x ∈ E so that the inequality (7.83) ensures that the series of
its right-hand side is convergent a.e. [m] on [a, b]. Consequently, it implies that
1
0 ≤ f (b) − sNk (b) < . (7.85)
2k
Obviously, we have
∞
X
gk (x) = f (x) − sNk (x) = fn (x),
n=Nk +1
so each gk is a positive nondecreasing function on [a, b]. To apply the first part of the
X∞
proof, we have to show that the series gk (x) converges on [a, b]. Combining this fact
k=1
and the inequality (7.85), we obtain
1
0 ≤ gk (x) <
2k
on [a, b]. As a result, the series
∞
X ∞
X
gk = (f − sNk )
k=1 k=1
converges (uniformly) to g(x) on [a, b] by the Weierstrass M -test. Now we are able to apply
the result (7.84) to the sequence {gk } of positive nondecreasing functions and conclude
that
lim gk (x) = lim [f ′ (x) − s′Nk (x)] = 0
k→∞ k→∞
Nk
X
lim fn′ (x) = f ′ (x) (7.86)
k→∞
n=1
236 Chapter 7. Differentiation
a.e. [m] on [a, b]. Since each fn is nondecreasing on [a, b], fn′ (x) ≥ 0 for all x ∈ E. Hence
the limit (7.86) implies that
∞
X
f ′ (x) = fn′ (x)
n=1
a.e. [m] on [a, b]. Since any real number is contained in some compact interval [a, b], our
desired result (7.81) follows from this.
Remark 7.4
The last assertion in Problem 7.17 is called Fubini’s Theorem on Differentiation. See
[58, Sec. C, pp. 527 – 529].
Problem 7.18
Rudin Chapter 7 Exercise 18.
Proof. First of all, we show that {ϕn } is orthonormal on [0, 1]. Suppose that m, n ∈ N and
m ≥ n. If m = n, then ϕn (t)ϕn (t) = ϕ2n (t) = 1 and so
Z 1
ϕn (t)ϕm (t) dt = 1.
0
If m > n, then we deduce from Theorem 7.26 (The Change-of-variables Theorem) that
Z 1 Z 1
ϕn (t)ϕm (t) dt = ϕ0 (2n t)ϕ0 (2m t) dt
0 0
Z 2n
1
= n ϕ0 (y)ϕ0 (2m−n y) dy
2 0
2n−1 Z
1 X 2i
= n ϕ0 (y)ϕ0 (2m−n y) dy. (7.87)
2 2(i−1)
i=1
Since ϕ0 (y) is periodic with period 2, we have ϕ0 (y) = 1 on [2i − 2, 2i − 1) and ϕ0 (y) = −1 on
[2i − 1, 2i). Therefore, we can reduce the expression (7.87) to
Z 2n−1 Z Z 2i
1
1 X h 2i−1 m−n
i
ϕn (t)ϕm (t) dt = n
ϕ0 (2 y) dy − ϕ0 (2m−n y) dy . (7.88)
0 2 2(i−1) 2i−1
i=1
Note that the interval [2m−n+1 i − 2m−n , 2m−n+1 i) is of length 2m−n , so the integrals on the
right-hand side of the equation (7.88) are 0 and this gives
Z 1
ϕn (t)ϕm (t) dt = 0
0
for some f ∈ L2 [0, 1] , i.e., the series in (7.89) is the Fourier series of f .
Consider a = j · 2−N , b = (j + 1) · 2−N , a < t < b and sN = c1 ϕ1 + · · · + cN ϕN , where
j = 0, 1, 2, . . .. On the one hand, if 1 ≤ k ≤ N , then a ≤ t ≤ b implies that
j · 2k−N ≤ 2k t ≤ (j + 1)2k−N and m [2k a, 2k b] = 2k−N ≤ 1
so that each ϕk (t) = ϕ0 (2k t) (of course, it is 1 or −1) is constant in [a, b]. This implies that
Z b Z b
1 1
sN (t) dt = sN (t) · dt = sN (t).
b−a a b−a a
On the other hand, if k > N , then 2k−N > 1 and we apply Theorem 7.26 (The Change-of-
variables Theorem) to get
Z b Z b Z 2k b Z (j+1)·2k−N
1 1
ϕk (t) dt = ϕ0 (2k t) dt = ϕ0 (y) dy = ϕ0 (y) dy.
a a 2k 2k a 2k j·2k−N
Using the same argument as in evaluating the integrals in the equation (7.88), we can show that
Z (j+1)·2k−N
1
ϕ0 (y) dy = 0.
2k j·2k−N
In conclusion, we arrive at
Z b Z b
1 1
sN (t) = sN dm = sn dm (7.90)
b−a a b−a a
for every n = 1, 2, . . .. By the relation (7.89), sn → f in L2 on [0, 1]. By Theorem 3.5 (Hölder’s
Inequality), we obtain
Z b nZ b o1 n Z b o1
2 2 1
|sn − f | dm ≤ |sn − f |2 dm dm = (b − a) 2 ksn − f k2 → 0
a a a
as n → ∞. Thus this implies that sn − f ∈ L1 on [0, 1] for large enough n and so we derive from
Theorem 1.33 that Z b Z b
lim sn dm = f dm. (7.91)
n→∞ a a
Since f ∈ L2 on [0, 1], we have f ∈ L1 on [0, 1] by Theorem 3.8. By Theorem 7.7, almost every
x ∈ [0, 1] is a Lebesgue point of f . Let p ∈ [0, 1] be such a point and for each N ∈ N, we define
EN (p) = [2−N , 2−N +1 ]. Clearly, we have
∞
X
m EN (p) = 2−N and 2−N = 1.
N =1
Therefore, one can find a N ∈ N such that p ∈ [2−N , 2−N +1 ]. Furthermore, it is easy to check
that the sequence {EN (p)} shrinks to p nicely because
1
EN (p) ⊆ B(p, 2−N +1 ) and m EN (p) = m B(p, 2−N +1 )
4
238 Chapter 7. Differentiation
for every N ∈ N. Using Theorem 7.10 (Lebesgue Differentiation Theorem) and then the repre-
sentation (7.92), we find that
Z ∞
X
1
f (p) = lim f dm = lim sN (p) = cN ϕN (p).
N →∞ m [2−N , 2−N +1 ] [2−N ,2−N+1 ] N →∞
N =1
∞
X
In other words, the series cn ϕn converges to f a.e. in [0, 1].
n=1
Finally, we know that
ϕn (t + 1) = ϕ0 2n (t + 1) = ϕ0 (2n t + 2n ) = ϕ0 (2n t) = ϕn (t)
for every n ∈ N, the series and f are functions with period 1. Now for every t ∈ R, we can write
t = t′ + m with t′ ∈ [0, 1] and m ∈ Z. Furthermore, if t is a Lebesgue point of f , then it can be
shown from the definition and the periodicity of f that t′ is also a Lebesgue point of f . By this
fact (for the third equality below), we obtain
∞
X ∞
X ∞
X
cn ϕn (t) = cn ϕn (t′ + m) = cn ϕn (t′ ) = f (t′ ) = f (t′ + m) = f (t),
n=1 n=1 n=1
Problem 7.19
Rudin Chapter 7 Exercise 19.
hc (x) = 0
and then Z Z 1
0≤ hc (x) dx = hc (x) dx ≤ M.
R 0
M
h1 (x) ≤ (7.96)
x
for x ∈ (0, 1) and h1 (x) = 0 otherwise. Take λ > 0. If the real number x satisfies
|h1 (x)| = h1 (x) > λ, then we must have x ∈ (0, 1) and furthermore 0 < x < M
λ by the
inequality (7.96). In other words, we have
n Mo
{x ∈ R | h1 (x) > λ} ⊆ x ∈ R 0 < x <
λ
and so n
Mo M
λ · m{|h1 | > λ} ≤ λ · m x ∈ R 0 < x < =λ· = M.
λ λ
By the definition in §7.5, we see that h1 is in weak L1 .
Recall the fact (7.93), so the Extreme Value Theorem [99, Theorem 4.16, p. 89] ensures
that M = f (p) for some p ∈ (0, 1). Given ǫ > 0. Consider the function g : R → R defined
by
g(x) = f (x) − M + ǫ
which is continuous on R. Since g(p) = f (p) − M + ǫ = ǫ > 0, the sign-preserving
property for continuous functions [127, Problem 7.15, p. 112] says that there exists a
δ > 0 small enough such that g(x) > 0 on (p − δ, p + δ) ⊂ (0, 1), i.e., f (x) > M − ǫ on
(p − δ, p + δ) ⊂ (0, 1). Particularly, we may take ǫ = M
2 so that
M
f (x) > (7.97)
2
which is equivalent to saying that Ny y ∈ (p−δ, p+δ). Hence it follows from the inequalities
(7.97) and (7.98) that
Ny M (p − δ)M
h1 (y) ≥ Ny f (Ny y) > > .
2 2y
Z 2δ
dt
However, we know that / L1 (R).
= ∞, so we conclude that h1 ∈
0 t
(c) Let c > 1 and δ > 0 be the number such that the inequality (7.97) holds on the segment
(p − δ, p + δ) ⊂ (0, 1). Now for each k ∈ N, if x ∈ ( p−δ p+δ
k , k ), then kx ∈ (p − δ, p + δ) and
thus
M kc
hc (x) ≥ kc f (kx) >
2
240 Chapter 7. Differentiation
c c
which means ( p−δ p+δ
k , k ) ⊂ {x ∈ R | hc (x) >
Mk
2 }. Consequently, if we take λk =
Mk
2 ,
then we obtain
p − δ p + δ M k c 2δ
λk · m{hc > λk } > λk · m , = · = M δkc−1 . (7.99)
k k 2 k
As M and δ are fixed as well as c > 1, the inequality (7.99) implies that
λk · m{hc > λk } → ∞
Problem 7.20
Rudin Chapter 7 Exercise 20.
(a) If m(∂E) = 0, then we have m(E \E ◦ ) = 0. Since E ◦ is the union of all open sets contained
in E and E is the intersection of all closed sets containing E, we have E ◦ ⊆ E ⊆ E and
E ◦ , E ∈ B. In fact, we have
m(E) = m(E ◦ )
by Theorem 1.36.i Furthermore, since m is a complete measure, E is also Lebesgue
measurable.j
(b) To prove this part, we need stronger versions of Theorems 7.7 and 7.10 (The Lebesgue
Differentiation Theorem). Roughly speaking, the hypothesis f ∈ L1 (Rk ) can be replaced
by f ∈ L1loc (Rk ), where L1loc (Rk ) is the space of all locally integrable functions:
Z
|f | dm < ∞
K
for every bounded measurable set K ⊆ Rk , see [40, p. 95]. For convenience, we state the
stronger versions of these theorems in a single lemma:
Lemma 7.5
(a) If f ∈ L1loc (Rk ), then almost every x ∈ Rk is a Lebesgue point of f .
(b) Associate to each x ∈ Rk a sequence {Ei (x)} that shrinks to x nicely, and
let f ∈ L1loc (Rk ). Then
Z
1
f (x) = lim f dm
i→∞ m Ei (x) Ei (x)
i
See also [99, Theorem 2.27, p. 35; Exercise 9, p. 43].
j
Notice the difference between a Borel measurable set and a Lebesgue measurable set: Not every subset of a
Borel set with measure 0 is also Borel measurable, but Lebesgue measure is obtained by enlarging B to include
all subsets of sets of Borel measure 0.
7.5. Miscellaneous Problems on Differentiation 241
DE ◦ (p) = 0. (7.101)
Lemma 7.6
There is a sequence {Ei (p)} that shrinks to p nicely.
Proof of Lemma 7.6. For every i ∈ N, we have E ∩ B(p, 1i ) 6= ∅. (In fact, this is an
equivalent definition of a boundary point of E, see [5, Definition 3.40, p. 64].) Pick
{qi } to be an arbitrary sequence temporarily such that qi ∈ E ∩ B(p, 1i ). By the
1
definition (7.100), we know that qi ∈ B(xαi , rαi ) for some αi ∈ A. If |p − qi | ≤ 2i ,
then we have
1 1
B qi , ⊂ B p, (7.102)
2i i
1
and we define Ei (p) = B(qi , 2i ). Otherwise, we consider the point
1
s2i ∈ B(xβ2i , rβ2i ) ∩ B p,
2i
1
which satisfies |p − s2i | < 2i so that if we let qi = s2i and αi = β2i , then we gain
the set inclusion (7.102) in this case. (See Figure 7.2 below.) Now this process can be
done continually and then we obtain a sequence {qi } satisfying
1 1 1
qi ∈ E ∩ B p, and Ei (p) = B qi , ⊂ B p, .
i 2i i
Clearly, we have
1 21 1
m(Ei (p)) = π × = m B p, (7.103)
2i 4 i
for every i ∈ N, therefore {Ei (p)} shrinks to p nicely by the definition in §7.9, com-
pleting the proof of the lemma.
We go back to the proof of part (b). Recall that qi ∈ B(xαi , rαi ). There are two cases:
242 Chapter 7. Differentiation
– Case (i): qi ∈ ∂B(xαi , rαi ). Geometrically, the area of B(qi , r) ∩ B(xαi , rαi ) is
always greater than that of 14 B(qi , r) whenever 0 < r < rαi so that
1
m B(qi , r) ∩ B(xαi , rαi ) ≥ m B(qi , r) . (7.104)
4
Since B(α, rα ) ⊆ E for all α ∈ A, we have B(α, rα ) ⊆ E ◦ for all α ∈ A. Furthermore,
1
since rα ≥ 1 for all α ∈ A, we certainly have 2i < rαi for all i. Hence these
observations allow us to apply the equality (7.103) and the inequality (7.104) to get
m E ◦ ∩ B(p, 1i )
DE ◦ (p) = lim
i→∞ m B(p, 1i )
m E ◦ ∩ B(p, 1i )
= lim
i→∞ 4m Ei (p)
1
1 m B(xαi , rαi ) ∩ B(qi , 2i )
≥ · lim 1
4 i→∞ m B(qi , 2i )
1
≥
16
which contradicts the result (7.101).
1
– Case (ii): qi ∈ B(xαi , rαi ). Since |p − qi | < 2i and qi ∈ B(xαi , rαi ) ∩ B(p, 1i ),
1
the geometry tells us that rαi > 2i so that the inequality (7.104) holds in this case.
1
Finally, we have DE ◦ (p) ≥ 16 , a contradiction again.
In conclusion, we have proved that m(∂E) = 0.
1
(c) The crux of the analysis in part (b) is the inequalities rαi > 2i . For unrestricted radii
rα , we can pick a sequence {rαi } satisfying this condition for sufficiently large i and thus
1
DE ◦ (p) ≥ 16
(d) By the comments in §2.21, there exists a non-Borel subset of R. Let this set be A. Suppose
that [
E= B (x, 0), 1 .
x∈A
7.5. Miscellaneous Problems on Differentiation 243
Assume that E was Borel in R2 . If we can show that R × {1} is Borel in R2 , then since
E ∩ (R × {1}) = A × {1},
Lemma 7.7
Let A and B be Boreal subsets of R. Then A × B is also Borel in R2 .
∞
[ n ∞
[ o
An × R = (a, y) a ∈ An , y ∈ R
n=1 n=1
= {(a1 , y) | a1 ∈ A1 , y ∈ R} ∪ {(a2 , y) | a2 ∈ A1 , y ∈ R} ∪ · · ·
= (A1 × R) ∪ (A2 × R) ∪ · · ·
[∞
= (An × R),
n=1
we establish that
∞
[
An ∈ M1 .
n=1
(e) As Balcerzak and Kharazishvili [12, p. 205] pointed out that the union of an arbitrary
family of convex bodies (compact convex sets with non-empty interiors) in R2 is Lebesgue
measurable. Therefore, the closed discs can be replaced by arbitrary closed polygons
provided that they are convex.k
Problem 7.21
Rudin Chapter 7 Exercise 21.
[
k
More generally, it is known that if E = Xα and for each Xα , there exists an open ball Bα such that
Bα ⊆ Xα ⊆ Bα , then E is Lebesgue measurable. See [58, Problem 22, p. 482].
244 Chapter 7. Differentiation
Proof. Notice that the triangle inequality implies that f, g ∈ BV on [a, b] if and only if f +g ∈ BV
on [a, b]. Clearly, the function g(t) = t belongs to BV on [0, 1], so γ ∈ BV on [0, 1] if and only
if f ∈ BV on [0, 1].
By the definition, the length of the graph of f is the total variation of γ on [0, 1], say Vγ (1).
We want to show that Z 1p
Vγ (1) = 1 + [f ′ (t)]2 dt. (7.105)
0
Suppose that f is AC on [0, 1]. Then γ is also AC on [0, 1]. Without loss of generality, we may
assume that f (0) = 0; otherwise, we consider the function fe : [0, 1] → R defined by fe = f − f (0).
This assumption implies that γ(0) = 0. By Theorem 7.18,l we know that γ ′ ∈ L1 a.e. [m] on
[0, 1] and Z x
γ(x) = γ ′ (t) dt.
0
Therefore, if x, y ∈ [0, 1] and x < y, then the above fact and Theorem 1.33 establish
Z y Z y Z yp
|γ(y) − γ(x)| = γ ′ (t) dt ≤ |γ ′ (t)| dt = 1 + f ′ (t) dt. (7.106)
x x x
Finally, it follows from the definition and the inequality (7.106) that if 0 = t0 < t1 < · · · < tN =
1, then we have
N
X N Z
X ti p Z 1p
Vγ (1) = sup |γ(ti ) − γ(ti−1 | ≤ sup 1 + f ′ (t) dt = 1 + f ′ (t) dt.
i=1 i=1 ti−1 0
Employing the notation in the proof of Problem 7.13(c), if 0 ≤ x < y ≤ 1, then we have
N
X
Vγ (y) − Vγ (x) = Vγ (x; y) = sup |γ(ti ) − γ(ti−1 )|, (7.108)
i=1
where the supremum is taken over all positive integer N and over all partitions {ti } such that
x = t0 < t1 < · · · < tN = y. By repeated use of the triangle inequality, we certainly have
p N
X
(y − x)2 + [f (y) − f (x)]2 = |γ(y) − γ(x)| ≤ |γ(ti ) − γ(ti−1 )| ≤ Vγ (x; y). (7.109)
i=1
and then s
h f (y) − f (x) i2 Vγ (y) − Vγ (x)
1+ ≤ . (7.110)
y−x y−x
l
Monotonicity of γ is not necessary for the implication (a) → (c).
7.5. Miscellaneous Problems on Differentiation 245
Recall that f and Vγ are AC on [0, 1], f ′ and Vγ′ exist a.e. [m] on [0, 1] by Theorem 7.18, so the
inequality (7.110) ensures that p
1 + [f ′ (x)]2 ≤ Vγ′ (x)
a.e. [m] on [0, 1]. Substituting this into the expression (7.107), we get
Z 1p
1 + [f ′ (t)]2 dt ≤ Vγ (1)
0
which proves the other direction. Hence we have completed the proof of the problem.
Problem 7.22
Rudin Chapter 7 Exercise 22.
Proof.
(a) Assume that f 6= 0 on a measurable set E ⊆ Rk of positive measure. We claim that there
corresponds a constant c = c(f ) > 0 such that
(M f )(x) ≥ c|x|−k
for sufficiently large |x|. If it was not the case, then for each n ∈ N, there exists a xn ∈ Rk
with |xn | > n such that
|xn |−k
(M f )(xn ) < .
n
Recall the basic fact that the Lebesgue measure of the ball B(x, r) in Rk is given by
k
π2
k
rk .
Γ( 2 + 1)
Thus it follows from the definition [100, Eqn. (4), p. 138] and this fact that
Z k
π2 |xn |−k
|f | dm ≤ m(Br ) × (M f )(xn ) <
k
rk × (7.111)
B(xn ,r) Γ( 2 + 1) n
√
for all r > 0. Particularly, we pick the sequence {rn }, where rn = 2k n · |xn |, and the
inequality (7.111) becomes
Z Z k
π2 1
χB(xn ,rn ) |f | dm = |f | dm < ×√ . (7.112)
Rk B(xn ,rn ) Γ( k2 + 1) n
By the Archimedean Property [99, Theorem 1.20(a), p. 9], for every x ∈ Rk , there is a
positive integer n such that n ≥ |x|, so we have |x| < n < |xn | and then
√
|x − xn | ≤ |x| + |xn | < 2|xn | < 2k n · |xn |
for large enough n (remember that k is fixed). In other words, this means that x ∈
√
B(xn , 2k n · |xn |) and thus
lim χB(xn , 2k√n|xn |) (x) = 1. (7.113)
n→∞
Let |fn | = χB(xn ,rn ) |f | for all n ∈ N. Then each |fn | is measurable, |fn | ≤ |f | and the
limit (7.113) implies
|f |(x) = lim |fn |(x)
n→∞
246 Chapter 7. Differentiation
for every x ∈ Rk . Since f ∈ L1 (Rk ), the sequence {|fn |} satisfies the hypotheses of
Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem). Thus we have
Z Z Z
lim |f | dm = lim |fn | dm = |f | dm. (7.114)
n→∞ B(x ,r ) n→∞ Rk Rk
n n
Now we combine the inequality (7.112) and the result (7.114) to get
Z k
π2 1
|f | dm ≤ lim × √ = 0.
Rk n→∞ Γ( k + 1) n
2
(M f )(x) ≥ c|x|−k
(b) For every x ∈ (0, 41 ), if r ∈ (0, x], then we have x − r ≥ 0 and x + r < 12 . Particularly, we
take r = x so that
Z Z x+r Z 2x
1 1 1 1
|f | dm = |f (t)| dt = dt. (7.115)
m(Br ) B(x,r) 2r x−r 2x 0 t(log t)2
Since t(log t)2 > 0 on (0, 21 ) and t−1 (log t)−2 ∈ R on [ǫ, 2x], we apply [99, Theorem 11.33,
p. 323] to obtain
Z Z Z
2x
1 2x
1 2x
1 1 2x 1 1
dt ≥ dt = R dt = − =− +
0 t(log t)2 ǫ t(log t)2 ǫ t(log t)2 log t ǫ log(2x) log ǫ
1
for every ǫ > 0. As ǫ → 0+, we know that log ǫ → 0− and so
Z 2x
1 1
dt ≥ − . (7.116)
0 t(log t)2 log(2x)
as required.
Furthermore, direct computation gives
Z 1
4 dx
= ∞.
0 |2x log(2x)|
as desired.
Hence we have completed the proof of the problem.
7.5. Miscellaneous Problems on Differentiation 247
Problem 7.23
Rudin Chapter 7 Exercise 23.
Proof. We first show that (SF )(x) is well-defined. To this end, let α1 and α2 be two complex
numbers such that
Z Z
1 1
lim |f1 − α1 | dm = lim |f2 − α2 | dm = 0 (7.117)
r→0+ m(Br ) B(x,r) r→0+ m(Br ) B(x,r)
Since f and g belong to the same equivalence class F , f = g a.e. on Rk and the last integral in
the inequality (7.118) is actually 0. By taking r → 0+ to the integrals in the inequality (7.118)
and using the hypotheses (7.117), we obtain α1 = α2 , i.e., (SF )(x) is unique if x is a Lebesgue
point of F .
Next, if x is a Lebesgue point of f and f ∈ F , then Definition 7.6 gives
Z
1
lim |f (y) − f (x)| dm(y) = 0
r→0+ m(Br ) B(x,r)
so that x is also a Lebesgue point of F and the uniqueness of Lebesgue points of F in the
previous paragraph shows immediately that (SF )(x) = f (x). This completes the proof of the
problem.
248 Chapter 7. Differentiation
CHAPTER 8
Integration on Product Spaces
Problem 8.1
Rudin Chapter 8 Exercise 1.
• Case (i): Ai = (−∞, ai ). Then the condition Ai ⊂ Ai+1 forces that Ai+1 is in one of the
following forms:
(−∞, ai ], (−∞, ai+1 ) and (−∞, ai+1 ] (8.1)
for some ai+1 > ai . Thus A is in one of the following forms:
• Case (ii): Ai = (−∞, ai ]. Now Ai+1 is still in one of the forms (8.1), so similar analysis
shows that A is in one of the forms (8.2). Therefore, A ∈ M.
• Case (iii): Ai = (bi , ∞). In this case, Ai+1 is in one of the forms:
for some bi+1 < bi . Then it can be shown easily that A is in one of the following forms:
• Case (iv): Ai = [b, ∞). The Ai+1 is expressed in one of the forms (8.3) so that A is in
one of the forms (8.4). In conclusion, we have A ∈ M.
249
250 Chapter 8. Integration on Product Spaces
∞
[
Next, let Bi ∈ M, Bi ⊃ Bi+1 and B = Bi . Similar to the sequence of sets {Ai }, there are
i=1
also four cases for {Bi }. Since the analysis of these cases is very similar to those above, so we
omit the details here. By Definition 8.1, we see that M is a monotone class.
Clearly, R ∈ M. Since R \ (−∞, a) = [a, ∞), R \ (−∞, a] = (a, ∞), R \ (b, ∞) = (−∞, b] and
R \ [b, ∞) = (−∞, b), we immediately have R \ A ∈ M for every A ∈ M. However, M is not
a σ-algebra because (−∞, 0), (1, ∞) ∈ M but (−∞, 0) ∪ (1, ∞) ∈ / M which violates Definition
1.3(a). This completes the proof of the problem.
Problem 8.2
Rudin Chapter 8 Exercise 2.
(a) Since f is Lebesgue measurable, Theorem 1.17 (The Simple Function Approximation The-
orem) ensures the existence of simple measurable functions sn on R such that 0 ≤ s1 ≤
s2 ≤ · · · ≤ f and sn (x) → f (x) as n → ∞ for every x ∈ R.
Suppose that s is a nonnegative simple function on R. Then, by Definition 1.6, we have
n
X
s= αi χAi ,
i=1
Lemma 8.1
If f and g are functions on R such that f (x) ≤ g(x) for every x ∈ R, then
A(f ) ⊆ A(g).
holds trivially because of the fact that sn ≤ f for every n = 1, 2, . . . and Lemma 8.1. For
the other direction, let (x, y) ∈ A(f ). This implies that x ∈ R and 0 < y < f (x). Since
sn (x) → f (x) as n → ∞, there exists a N ∈ N such that 0 < y < sN (x) ≤ f (x) which
means (x, y) ∈ A(sN ) so that
[∞
A(f ) ⊆ A(sn )
n=1
also holds. Hence we have established the claim (8.6). Finally, since each A(sn ) is a
Lebesgue measurable subset of R2 , the relation (8.6) implies that A(f ) is also a Lebesgue
measurable subset of R2 .
Since R is obviously σ-finite, it follows from Theorem 8.6 (see also Definition 8.7) that
Z Z
m2 A(f ) = (m × m) A(f ) = m [A(f )]x dx = f (x) dx.
R R
G = {(x, y) ∈ R2 | y = f (x)}.
For each i ∈ N, the set A(f + 1i ) is Lebesgue measurable in R2 by part (a). Then the set
∞
\ 1
A= A f+ = {(x, y) ∈ R2 | 0 < y ≤ f (x)}
i
i=1
as required.
Problem 8.3
Rudin Chapter 8 Exercise 3.
in (0, 1) × (0, 1). It is clear that f is a positive continuous function in (0, 1) × (0, 1). If x = 21 ,
then f ( 12 , y) = y −1 so that
1 Z 1
ϕ = y −1 dy = ln y|10 = ∞.
2 0
If x 6= 21 , then we have
Z 1 1 2 1
ϕ(x) = y (x− 2 ) −1
dy =
0 (x − 21 )2
and this implies that
Z Z
1 1
1 1 −1 1
ϕ(x) dx = dx = − x − = −4.
0 0 (x − 21 )2 2 0
Problem 8.4
Rudin Chapter 8 Exercise 4.
Proof.
(a) As Rudin pointed out, we may assume that f and g are Borel functions on R. If p = 1,
then it is in fact Theorem 8.14. If p = ∞, then kgk∞ < ∞ and so
Z ∞
|(f ∗ g)(x)| = f (x − t)g(t) dt
−∞
Z ∞
≤ |f (x − t)| · |g(t)| dt
−∞
Z ∞
≤ kgk∞ |f (x − t)| dt
Z−∞∞
= kgk∞ |f (t)| dt
−∞
= kgk∞ · kf k1 .
1
Suppose that 1 < p < ∞. Since f and g are measurable on R, the functions |f (x − t)| q
1
and |f (x − t)| p · |g(y)| are also measurable on R by Theorem 1.7, where q is the conjugate
exponent of p. By Theorem 3.5 (Hölder’s Inequality), we derive that
Z ∞ Z ∞
1 1
|f (x − y)g(y)| dy = |f (x − y)| q · |f (x − y)| p |g(y)| dy
−∞ −∞
nZ ∞ o1 n Z ∞ o1
q p
≤ |f (x − y)| dy × |f (x − y)| · |g(y)|p dy
−∞ −∞
1 nZ ∞ o1
p
q
= kf k1 · |f (x − y)| · |g(y)|p dy
−∞
which gives
nZ ∞ op p
Z ∞
|f (x − y)g(y)| dy ≤ kf k1q |f (x − y)| · |g(y)|p dy. (8.7)
−∞ −∞
Define F : R2 → C by
F (x, y) = f (x − y)g(y).
Using similar argument as in the proof of Theorem 8.14, we are able to show that the
function F ∈ L1 (R2 ). Since R is a σ-finite measure space, we get from Theorem 8.8 (The
Fubini Theorem) that
Z ∞ Z ∞
ϕ(x) = Fx (y) dy = f (x − y)g(y) dy = (f ∗ g)(x)
−∞ −∞
a.e. is in L1 (R) so that the integral defining (f ∗ g)(x) exists for almost all x. This proves
the first assertion.
To verify the second and the third assertions, we recall that |f |, |g|p ∈ L1 (R), so Theorem
8.14 ensures that the function h : R → [0, ∞] given by
Z ∞
h(x) = |f (x − y)| · |g(y)|p dy
−∞
belongs to L1 (R)
which means that h is a measurable function on R. By Theorem 1.33,
we observe from the inequality (8.7) that
Z ∞ p
|(f ∗ g)(x)|p = f (x − y)g(y) dy
−∞
nZ ∞ op
≤ |f (x − y)g(y)| dy
−∞
p
Z ∞
≤ kf k1q |f (x − y)| · |g(y)|p dy. (8.8)
−∞
Now we apply Theorem 8.8 (The Fubini Theorem) to the right-hand side of the inequality
(8.8) to get
Z ∞ p
Z ∞Z ∞
|(f ∗ g)(x)|p dx ≤ kf k1q |f (x − y)| · |g(y)|p dy dx
−∞ −∞ −∞
p
Z ∞Z ∞
= kf k1q |f (x − y)| · |g(y)|p dx dy
−∞ −∞
p
Z ∞
= kf k1q · kf k1 |g(y)|p dy
−∞
= kf kp1 · kgkpp
which implies that kf ∗ gkp ≤ kf k1 · kgkp < ∞.
254 Chapter 8. Integration on Product Spaces
kf ∗ gk1 = khk1
Z ∞
= h(x) dx
Z−∞
∞ Z ∞
= f (x − y)g(y) dy dx
Z−∞
∞
−∞
Z ∞
= g(y) dy f (x − y) dx
−∞ −∞
= kgk1 · kf k1
as required.
Let p = ∞. If f ≥ 0 and g = c ≥ 0, then the inequality (8.9) holds in this case. Recall
that f ∈ L1 (R) and g ∈ L∞ (R), so we have
Z ∞
|h(x)| = f (x − y)c dy = ckf k1 < ∞
−∞
(c) Suppose that 1 < p < ∞ and kf ∗ gkp = kf k1 · kgkp . By the definition of f ∗ g and the
application of part (a), we observe that
kf ∗ gkp ≤
|f | ∗ |g|
p ≤
|f |
1 ·
|g|
p = kf k1 · kgkp = kf ∗ gkp
which forces
|f | ∗ |g|
p =
|f |
1 ·
|g|
p . Thus we may assume that f, g ≥ 0.
Recall that the inequality kf ∗ gkp ≤ kf k1 · kgkp derives from (see the deduction of the
inequality (8.8))
nZ ∞ op
p
(f ∗ g)(x) = f (x − y)g(y) dy
−∞
nZ ∞ 1 1
op
= f (x − y) q · [f (x − y) p g(y)] dy
−∞
nZ ∞ op n Z ∞ o
q
≤ f (x − y) dy × f (x − y)g(y)p dy (8.10)
−∞ −∞
p nZ ∞ o
= kf k1q × f (x − y)g(y)p dy
−∞
and so the equality kf ∗ gkp = kf k1 · kgkp forces that the equality (8.10) holds, but this
means that the equality of Theorem 3.5 (Hölder’s Inequality) holds there i.e., there are
constants (with respect to y) α(x) and β(x), not both zero, such that
q 1
α(x)f (x − y) = α(x)f (x − y) q = β(x)[f (x − y) p g(y)]p = β(x)f (x − y)g(y)p (8.11)
8.2. Applications of the Fubini Theorem 255
for almost all y ∈ R. Let E = {z ∈ R | f (−z) > 0}. If kf k1 = 0, then since m(R) = ∞,
f = 0 a.e. on R and we are done. Therefore, without loss of generality, we may assume
that kf k1 = 1 which gives m(E) > 0.
On the one hand, if β(x) = 0 for some x ∈ R, then α(x) 6= 0 and so the equation (8.11)
implies that f (x − y) = 0 for almost all y ∈ R which is equivalent to f = 0 a.e. on R. On
the other hand, if α(x) = 0 for some x ∈ R, then β(x) 6= 0 and so
f (x − y)g(y)p = 0 (8.12)
for almost all y ∈ R. Now if y ∈ Ex = x+E = {x+z | z ∈ E}, then y = x+z or −z = x−y
for some z ∈ E so that
f (x − y) > 0
for all y ∈ Ex . Therefore, it follows from the equation (8.12) that g(y) = 0 for almost all
y ∈ Ex . Assume that m(Ex ) < ∞. Then m(Exc ) = ∞. Furthermore, we have g(y) > 0 for
all y ∈ Exc and this implies that
Z Z
kgkpp = g(y)p dy = g(y)p dy = ∞
R Exc
(d) Suppose that p = ∞. We simply take f (x) = χ[0,1] and g(x) = 1 on R. Then we have
kf k1 = kgk∞ = 1. Since f ≥ 0 and g ≥ 0 on R, we deduce from Theorem 7.26 (The
Change-of-variables Theorem) that
Z ∞ Z 1
(f ∗ g)(x) = f (y)g(x − y) dy = dy = 1.
−∞ 0
Next, we suppose that p < ∞. Let 0 < α < 1 − (1 − ǫ)p < 1 and define
1 1
f (x) = χ (x) and g(x) = χ (x).
2α [−α,α] 2p [−1,1]
It is easy to see that
Z ∞ Z α
1
kf k1 = |f (x)| dx = χ (x) dx = 1
−∞ −α 2α [−α,α]
and Z Z
∞ 1
1
kgkp = |g(x)|p dx = χ (x) dx = 1.
−∞ −1 2 [−1,1]
Since f ≥ 0 and g ≥ 0 on R, we deduce from Theorem 7.26 (The Change-of-variables
Theorem) that
Z ∞ Z α
1
(f ∗ g)(x) = f (y)g(x − y) dy = 1 χ[−1,1] (x − y) dy. (8.13)
−∞ 2α · 2 p −α
Since −1 ≤ x − y ≤ 1 if and only if x − 1 ≤ y ≤ x + 1, we follow from the expression (8.13)
that
Z α
1
(f ∗ g)(x) = 1 χ[x−1,x+1](y) dy
2α · 2 p −α
256 Chapter 8. Integration on Product Spaces
1
= 1 m [−α, α] ∩ [x − 1, x + 1]
2α · 2
p
0, if x ∈ (−∞, −1 − α];
x+1+α
if x ∈ (−1 − α, −1 + α];
1
2α · 2 p
2α
= 1 if x ∈ (−1 + α, 1 − α];
2α · 2 p
1+α−x
if x ∈ (1 − α, 1 + α];
1
2α · 2
p
0, if x ∈ (1 + α, ∞).
Since (x+1+α)p ≥ 0 on [−1−α, −1+α], the integral in the expression (8.14) is nonnegative
so that
kf ∗ gkpp ≥ 1 − α > (1 − ǫ)p
Remark 8.1
The result in Problem 8.4(a) is sometimes called Minkowski’s Inequality which is a special
case of the Young’s Convolution Inequality.
Problem 8.5
Rudin Chapter 8 Exercise 5.
Proof.
Lemma 8.2
It is true that α−1 (E) ∈ B2 if E ∈ B1 . (See Problem 8.11 for the meaning of the
notation Bk for any k ∈ N.)
– R ∈ M.
Let’s return to the proof of the problem. We want to check Definition 6.1:
∞
X
(µ ∗ λ)(E) = (µ ∗ λ)(E i ) (8.15)
i=1
Then we have
∞
[
E2 = E2i .
i=1
Z nZ o
= χAx (y) dλ(y) h1 (x) d|µ|(x)
ZX n ZY o
= χA (x, y)h2 (y) d|λ|(y) h1 (x) d|µ|(x). (8.17)
X Y
Thus, by Theorem 8.8 (The Fubini Theorem), the integral (8.17) can be reduced to
Z Z
(µ × λ)(A) = χA (x, y)h2 (y)h1 (x) d|λ|(y) d|µ|(x)
ZX Y
= χA (x, y)h2 (y)h1 (x) d(|µ| × |ν|)
ZX×Y
= f (x, y) d(|µ| × |ν|), (8.18)
A
where f (x, y) = h2 (y)h1 (x). Obviously, Theorem 8.8 (The Fubini Theorem) also says that
f (x, y) ∈ L1 (|µ| × |λ|) so that Theorem 6.13 may be applied to the result (8.18) to get
Z Z
|µ × λ|(A) = |f (x, y)| d(|µ| × |λ|) = d(|µ| × |λ|) = (|µ| × |λ|)(A), (8.19)
A A
Hence we follow from the definition of the norm in M and Definition 8.7 (R is σ-finite)
that
(b) Firstly, for E ∈ B1 , we know from the definition of µ ∗ λ, Definition 8.7 and then [100,
Eqn. (3), p. 163] that
Z
χE (x) d(µ ∗ λ) = (µ ∗ λ)(E)
R
= (µ × λ)(E2 )
Z
= λ(E2y ) dµ(x)
R
Z Z
= χE (x + y) dµ(x) dλ(y).
R R
8.2. Applications of the Fubini Theorem 259
n
X
Thus the formula holds for χE and then it also holds for any simple function s = αi χAi ,
i=1
where each Ai is a Borel set in R, i.e.,
Z Z Z
s(x) d(µ ∗ λ) = s(x + y) dµ(x) dλ(y). (8.22)
R R R
By Theorem 6.12, there is a Borel measurable function h such that |h(x)| = 1 on R and
d(µ ∗ λ) = h d|µ ∗ λ|. If we consider the sequence {sn h} of Borel measurable functions,
then the inequalities (8.23) are still valid. Furthermore, we have sn (x)h(x) → f (x)h(x) as
n → ∞ for every x ∈ R. By the inequality (8.21), we see that
Z
kf hk1 = |f (x)h(x)| d|µ ∗ λ| ≤ M |(µ ∗ λ)|(R) ≤ M |µ|(R)| · λ|(R) < ∞.
R
Now we may apply similar analysis to show that the limit of the right-hand side of the
formula (8.24) is exactly Z Z
f (x + y) dµ(x) dλ(y).
R R
Hence we conclude that
Z Z Z
f d(µ ∗ λ) = f (x + y) dµ(x) dλ(y). (8.25)
R R R
is a bounded linear functional on C0 (R). Consequently, Theorem 6.19 (The Riesz Repre-
sentation Theorem) ensures that µ ∗ ν is the unique complex Borel measure satisfying the
formula (8.25).
Thus the associativity of the convolution in M follows immediately from the formulas
(8.26) and (8.27), i.e., (µ ∗ λ) ∗ ν = µ ∗ (λ ∗ ν).
By the formula (8.25) again, we see that
Z Z Z
f d µ ∗ (λ + ν) = f (x + y) dµ(x) d[λ(y) + ν(y)]
R ZR ZR Z Z
= f (x + y) dµ(x) dλ(y) + f (x + y) dµ(x) dν(y)
ZR R Z R R
= f d(µ ∗ λ) + f d(µ ∗ ν)
ZR R
= f d(µ ∗ ν + µ ∗ ν)
R
which shows immediately that the convolution in M is distributive with respect to addition,
i.e., µ ∗ (λ + ν) = µ ∗ λ + µ ∗ ν.
E2y = {x ∈ R | (x, y) ∈ E2 } = {x ∈ R | x + y ∈ E} = {x − y ∈ R | x ∈ E} = E − y.
as required.
(e) Suppose that µ and λ are concentrated on the countable sets A and B respectively, i.e.,
µ(E) = λ(F ) = 0
Lemma 8.3
E2 ∩ (A × B) = ∅ if and only if E ∩ (A + B) = ∅.
where α and β are real. Thus (X1 , +, ·) is a vector space. Next, by part (e), we know
that the map X1 × X1 → X1 given by
(µ, λ) 7→ µ ∗ λ
is well-defined. Thus it can be seen that Conditions (1) to (3) come directly from
the results of part (c). By the definition, (X1 , +, ∗, ·) is an algebra.
– Case (2): X2 is an ideal in M . Recall that Y is called an ideal of an algebra X
if Y is a vector subspace of X and xY, Y x ⊆ Y for every x ∈ X. Here
xY = {x ∗ y | y ∈ Y } and Y x = {y ∗ x | y ∈ Y }.
Let µ, λ ∈ X2 so that µ({x}) = λ({x}) = 0 for every x ∈ R, but these imply trivially
that
(αµ + βλ)({x}) = αµ({x}) + βλ({x}) = 0
for every real α and β. In other words, (X2 , +, ·) is a vector space. Since M is
commutative by part (c), it suffices to prove X2 λ ⊆ X2 for every λ ∈ M , but this
fact follows immediately from part (e). Hence we may conclude that X2 is an ideal
in M .
– Case(3): X3 is an ideal in M . Let µ, λ ∈ X3 . Then µ(E) = λ(E) = 0 whenever
m(E) = 0. For every α, β ∈ R, if m(E) = 0, then
Therefore, it means that (αµ + βλ) ≪ m and so (X3 , +, ·) is a vector space. Next,
for any µ ∈ X3 and λ ∈ M , µ ∗ λ ≪ m by part (e), so we have µλ ∈ X3 and X3 is an
ideal in M by the definition.
Recall that m is a positive σ-finite measure, thus for every µ ∈ X3 , we haveµ ≪ m
so that Theorem 6.10 (The Lebesgue-Radon-Nikodym Theorem) ensures that there
8.2. Applications of the Fubini Theorem 263
Φ(µ ∗ ν) = f ∗ g,
where E ∈ B1 . Since u±
and v±
are measurable and nonnegative, Theorem 1.29
guarantees that every integral in the expression (8.32) is a measure on B1 . Thus, if
m(E) = 0, then Proposition 1.24(e) tells us that µ(E) = 0. By Definition 6.7, µ ≪ m
or equivalently µ ∈ X3 and Φ is surjective.
In conclusion, Φ is an isomorphism, as required.
(i) Suppose that M is defined to be the Banach space of all complex Borel measures on Rk
with norm kµk = |µ|(Rk ) and associate to each Borel set E ⊆ Rk the set
E2 = {(x, y) | x + y ∈ E} ⊆ Rk × Rk .
If µ, λ ∈ M , we define
(µ ∗ λ)(E) = (µ × λ)(E2 )
for every Borel set E ⊆ Rk .
Then all the assertions from parts (a) to (h) also hold when we
replace R and B1 by Rk and Bk respectively because Rk is always a commutative group
and mk is a translation invariant Borel measure on Rk . The proofs go exactly the same,
so we omit the details here.
For the k-dimensional torus T k , we need to seek a measure similar to the Lebesgue
measure mk on Rk . Note that T k is a locally compact group with the σ-algebra B(T k ),
so the Haar measure denoted by σk is what we want. (For the existence, uniqueness
and basic properties of such a measure, the reader is suggested to read, for examples, [28,
264 Chapter 8. Integration on Product Spaces
Chap. 9], [40, §11.1, pp. 339 – 348] and [96, pp. 128 – 132, 211].) Once we have such a
measure, we can follow the same line as above to obtain the corresponding results. Again,
we omit the details here.
This completes the analysis of the problem.
Problem 8.6
Rudin Chapter 8 Exercise 6.
Proof. Let x ∈ Rk \ {0} and r = |x|. Then we have r > 0 so that it is meaningful to define
u = xr which certainly gives the representation
Suppose that x = r ′ u′ , where r ′ > 0 and |u′ | = 1. Then the equality ru = r ′ u′ implies that
r = |ru| = |r ′ u′ | = r ′
and so u = u′ . In other words, the representation (8.33) is unique. Therefore, we may “identify”
Rk \ {0} as (0, ∞) × Sk−1 , i.e., the map
Lemma 8.4
Let MX and MY be σ-algebras in X and Y respectively. Suppose that ϕ : X → Y
is measurable and µ : MX → [0, ∞] is a measure. Define ν = ϕ∗ µ by
ν(E) = µ ϕ−1 (E)
for E ∈ MY . Then the set function ν is a measure on MY and for every measurable
function g : Y → [0, ∞], we have
Z Z
g dν = (g ◦ ϕ) dµ.
Y X
Let ρ be the measure on (0, ∞) given by dρ = r k−1 dr. Now it suffices to prove that
ν(E) = mk ϕ−1 (E) = (ρ × σk−1 )(E) (8.35)
for every E ∈ B (0, ∞) × Sk−1 , where σk−1 is the measure defined on Sk−1 in the question.
This is because if the formula (8.35) holds, then for every nonnegative Borel function f on Rk ,
c
The measure ν in Lemma 8.4 is called the push-forward measure of µ. See [18, §3.6].
8.2. Applications of the Fubini Theorem 265
we deduce immediately from Lemma 8.4 (with X = Rk \ {0}, Y = (0, ∞) × Sk−1 , ϕ is given by
the formula (8.34), µ = mk and g = f ◦ ϕ−1 ) and Theorem 8.8 (The Fubini Theorem) that
Z Z
f dmk = (f ◦ ϕ−1 ) d(ρ × σk−1 )
Rk \{0} (0,∞)×Sk−1
Z
= f ϕ−1 (r, u) dσk−1 dρ
(0,∞)×S
Z ∞ Z k−1
= r k−1 f (ru) dσk−1 dr. (8.36)
0 Sk−1
Since {0} is a Borel set in Rk and mk (0) = 0, the integral on the left-hand side of the equation
(8.36) can be replaced by Z
f dmk
Rk
and we have what we want. To prove the formula (8.35), we have several steps:
• Step 1: The formula (8.35) holds for (r1 , r2 ) × A, where A ∈ B(Sk−1 ). On the one
hand, it is clear from Definition 8.7 that
(ρ × σk−1 ) (r1 , r2 ) × A = ρ (r1 , r2 ) σk−1 (A)
Z r2
= σk−1 (A) r k−1 dr
r1
r k − r1k
= 2 σk−1 (A). (8.37)
k
On the other hand, we see from the definition (8.34) that
e \ [(r1 A)
ϕ (r1 , r2 ) × A = {ru | 0 < r1 < r < r2 and u ∈ A} = (r2 A) e ∪ {r1 u | u ∈ A}]
so that
ν (r1 , r2 ) × A = mk ϕ (r1 , r2 ) × A
e \ [(r1 A)
= mk (r2 A) e ∪ {r1 u | u ∈ A}]
e − mk (r1 A)
= mk (r2 A) e −0
e
= (r k − r k )mk (A)
2 1
r k − r1k
= 2 σk−1 (A). (8.38)
k
Hence our claim follows by combining the two expressions (8.37) and (8.38).
K1 ⊆ K2 ⊆ · · · ⊂ U ⊂ · · · ⊂ V2 ⊂ V1 (8.39)
266 Chapter 8. Integration on Product Spaces
as n → ∞. Therefore, we apply the set relations (8.39), then the fact that ϕ−1 sends Borel
sets in (0, ∞) × Sk−1 to Borel sets in Rk \ {0} and finally Definition 8.7 to get
ν (Vn \ U ) × A ≤ ν (Vn \ Kn ) × A
= (ρ × σk−1 ) (Vn \ Kn ) × A
= ρ(Vn \ Kn )σk−1 (A). (8.41)
By using the inequality (8.40), the inequality (8.41) shows that ν(Vn \ U ) × A) → 0 as
n → ∞. In other words, it means that
as n → ∞. Furthermore, if we suppose
∞
[ ∞
\
K= Kn and V = Vn
n=1 n=1
which are an Fσ set and a Gδ set respectively, then Theorem 1.19(d) and (e) imply that
Thus we establish from Definition 8.7 and the limits (8.42) and (8.43) that
Finally, if U is any bounded or unbounded Borel set in (0, ∞), then we consider the sets
Un = U ∩ (n − 1, n]
for n ∈ N. It is trivial that each Un is a bounded Borel set, so the preceding paragraph
gives
(ρ × σk−1 )(Un × A) = ν(Un × A) (8.44)
for each n = 1, 2, . . .. Since Ui ∩ Uj = ∅ if i 6= j, (Ui × A) ∩ (Uj × A) = ∅ for all i 6= j and
then the result (8.44) shows that
[
∞
(ρ × σk−1 )(U × A) = (ρ × σk−1 ) Un × A
n=1
8.2. Applications of the Fubini Theorem 267
[
∞
= (ρ × σk−1 ) (Un × A)
n=1
∞
X
= (ρ × σk−1 )(Un × A)
n=1
X∞
= ν(Un × A)
n=1
[
∞
=ν Un × A
n=1
= ν(U × A),
i.e., the formula (8.35) holds for every U × A, where U ∈ B (0, ∞) and A ∈ B(Sk−1 ).
• Step 3: The formula (8.35) holds for E ∈ B (0, ∞) × Sk−1 . Suppose that
M = {E ∈ B (0, ∞) × Sk−1 | the formula (8.35) holds for E} ⊆ B (0, ∞) × Sk−1 .
(8.45)
By using similar argument as in the proof of Problem 8.11, we can prove that
B (0, ∞) × B(Sk−1 ) = B (0, ∞) × Sk−1 .
By Step 2, we see that B (0, ∞) × B(Sk−1 ) ⊆ M. Thus it concludes from the set
inclusion (8.45) that
M = B (0, ∞) × Sk−1 .
for every r ∈ (0, ∞). To this end, let u ∈ S1 . Then u is a point on the unit circle so that
u = ( √ 2x 2 , √ 2y 2 ). We define F : B(0, 1) \ {0} → [0, ∞] by
x +y x +y
rx ry
F (u) = f (ru) = f p ,p = f (r cos θ, r sin θ),
x2 + y 2 x2 + y 2
where θ ∈ [0, 2π). Recall the definition of σ1 so that
Z Z
f (ru) dσ1 = 2 F (u) dm2
S1 B(0,1)\{0}
Z 1 Z 2π
=2 a f (r cos θ, r sin θ) dθ da
0 0
Z 2π
= f (r cos θ, r sin θ) dθ
0
where θ ∈ [0, π) and ϕ ∈ [0, 2π). This completes the proof of the problem.
268 Chapter 8. Integration on Product Spaces
Remark 8.2
We note that it can be further shown that the Borel measure σk−1 on Sk−1 is unique, see
[40, Theorem 2.49, pp. 78, 79] for a proof of this. For similar or other proofs of Problem
8.6, you are suggested to read [60, pp. 172, 173, 322], [105, Theorem 15.13, pp. 154, 155] or
[111, §3.2, pp. 279 – 281].
Problem 8.7
Rudin Chapter 8 Exercise 7.
Proof. Since X and Y are σ-finite, X is the union of countably many disjoint sets Xn with
µ(Xn ) < ∞, and that Y is the union of countably many disjoint sets Ym with λ(Ym ) < ∞. For
each Xn × Ym , we have
ψ(Xm × Ym ) = µ(Xn )λ(Ym ) < ∞. (8.47)
Suppose that Ω is the class of all E ∈ S × T for which the conclusion of the problem holds,
i.e.,
Ω = {E ∈ S × T | ψ(E) = (µ × λ)(E)}.
If E = A × B for some A ∈ S and B ∈ T , then Ex = B if x ∈ A and Ex = ∅ if x ∈ B. Thus
we deduce from Definition 8.7 that
Z Z
(µ × λ)(E) = λ(Ex ) dµ(x) = λ(B) dµ(x) = µ(A)λ(B) = ψ(A × B) = ψ(E). (8.48)
X A
∞
[
In other words, A × B ∈ Ω. Besides, if Ai ∈ Ω, Ai ∩ Aj = ∅ for i 6= j and A = Ai , then since
i=1
ψ and µ × µ are measures,
[
∞ X∞ ∞
X [
∞
ψ(A) = ψ Ai = ψ(Ai ) = (µ × λ)(Ai ) = (µ × λ) Ai = (µ × λ)(A). (8.49)
i=1 i=1 i=1 i=1
Suppose that
Our goal is to show that M is a monotone class containing E so that Theorem 8.3 (The Monotone
Class Theorem) can be applied to obtain the reverse direction of the set inclusion (8.50).
[∞
We claim that M is a monotone class. To see this, let Ai ∈ M, Ai ⊆ Ai+1 and A = Ai .
i=1
By the definition (8.50), we know that
ψ Ai ∩ (Xn × Ym ) = (µ × λ) Ai ∩ (Xn × Ym ) (8.51)
Furthermore, since Ai ∩ (Xn × Ym ) ⊆ Ai+1 ∩ (Xn × Ym ) for all n, m ∈ N, Theorem 1.19(d) and
the equality (8.51) imply that
[
∞
ψ A ∩ (Xn × Ym ) = ψ [Ai ∩ (Xn × Ym )]
i=1
= lim ψ Ai ∩ (Xn × Ym )]
i→∞
= lim (µ × λ) Ai ∩ (Xn × Ym )
i→∞
[
∞
= (µ × λ) [Ai ∩ (Xn × Ym )]
i=1
= (µ × λ) A ∩ (Xn × Ym ) .
∞
\
In other words, A ∩ (Xn × Ym ) ∈ Ω so that A ∈ M. Next, if Bi ∈ M, Bi ⊇ Bi+1 and B = Bi .
i=1
By the definition (8.50) again, we have
ψ Bi ∩ (Xn × Ym ) = (µ × λ) Bi ∩ (Xn × Ym ) (8.52)
and
(µ × λ) B1 ∩ (Xn × Ym ) ≤ (µ × λ)(Xn × Ym ) < ∞.
Since Bi ∩ (Xn × Ym ) ⊇ Bi+1 ∩ (Xn × Ym ) for all n, m ∈ N, Theorem 1.19(e) and the equality
(8.52) imply that
\
∞
ψ B ∩ (Xn × Ym ) = ψ [Bi ∩ (Xn × Ym )]
i=1
= lim ψ Bi ∩ (Xn × Ym )]
i→∞
= lim (µ × λ) Bi ∩ (Xn × Ym )
i→∞
\
∞
= (µ × λ) [Bi ∩ (Xn × Ym )]
i=1
= (µ × λ) B ∩ (Xn × Ym ) .
Remark 8.3
Problem 8.7 is sometimes called the Product Measure Theorem.
Problem 8.8
Rudin Chapter 8 Exercise 8.
1
0 ≤ x − αn (x) <
n
for every n ∈ N and αn (x) → x as n → ∞. Therefore, it is illegal to say that there are
n ∈ N and i(n) ∈ Z such that
i(n) − 1 i(n)
= ai(n)−1 ≤ x ≤ ai(n) =
n n
and ai(n) → x as n → ∞. Now we simply write ai = ai(n) and define fn : R2 → R by
ai − x x − ai−1
fn (x, y) = f (ai−1 , y) + f (ai , y) (8.54)
ai − ai−1 ai − ai−1
ai − x x − ai−1
= fai−1 (y) + fa (y).
ai − ai−1 ai − ai−1 i
By the hypotheses, fai−1 (y) and fai (y) are Borel measurable. Obviously, both
ai − x x − ai−1
and
ai − ai−1 ai − ai−1
are continuous in x so that they are Borel measurable. By Proposition 1.19(c), each
fn (x, y) is Borel measurable. Furthermore, it satisfies
a −x x − ai−1 ai − x
i
|fn (x, y) − f (x, y)| = f (ai−1 , y) + f (ai , y) − f (x, y)
ai − ai−1 ai − ai−1 ai − ai−1
8.3. The Product Measure Theorem and Sections of a Function 271
x − ai−1
− f (x, y)
ai − ai−1
≤ |f (ai−1 , y) − f (x, y)| + |f (ai , y) − f (x, y)|
= |f y (ai−1 ) − f y (x)| + |f y (ai ) − f y (x)|, (8.55)
(b) We prove it by induction on k. The case for k = 1 is obviously true. Assume that the
statement is also true for k = m, i.e., if g(x1 , x2 , . . . , xm ) is continuous in each of the m
variables separately, then g is a Borel function in Rm . For k = m+1, g(x1 , x2 , . . . , xm+1 ) is
continuous in each of the m + 1 variables separately. Now for each xm+1 ∈ R, the function
gxm+1 (x1 , x2 , . . . , xm )
is continuous in x1 .
i−1 i
Now if we define n = ai−1 ≤ x1 ≤ ai = n and gn : Rm+1 → R by
ai − x 1 x1 − ai−1
gn (x, y) = g(ai−1 , y) + g(ai , y),
ai − ai−1 ai − ai−1
then we use a similar argument as part (a) to conclude that each gn is Borel measurable
in Rm+1 and the limit
g(x, y) = lim gn (x, y)
n→∞
implies that g is a Borel function in Rm+1 . Hence our desired result follows from induction.
Problem 8.9
Rudin Chapter 8 Exercise 9.
Proof. For each x ∈ R, let {pn } ⊆ E such that pn → x. By the hypotheses, let N be the
Lebesgue measurable set such that m(N ) = 0 and f y is continuous on R for all y ∈ R \ N . Now
for every y ∈ R \ N , these facts imply that
lim fpn (y) = lim f (pn , y) = lim f y (pn ) = f y (x) = f (x, y) = fx (y).
n→∞ n→∞ n→∞
Since each fpn is Lebesgue measurable, Theorem 1.14 ensures that fx is also Lebesgue measur-
able. In other words, what we have shown is that for each x ∈ R, the function fx is Lebesgue
measurable on R \ N .
272 Chapter 8. Integration on Product Spaces
Problem 8.10
Rudin Chapter 8 Exercise 10.
By using similar argument as in the proof of Problem 8.8, we can show that
which implies that h is Lebesgue measurable on R by Theorem 1.14, completing the proof of
the problem.
d
Remember that i actually depends on n.
8.4. Miscellaneous Problems 273
Problem 8.11
Rudin Chapter 8 Exercise 11.
Proof. Recall from the definition on [74, p. 86] that the product topology on Tm+n = Rm+n is
Rm × Rn which is the topology having as basis the collection
Bm+n = {U × V | U ∈ Tm and V ∈ Tn }.
Furthermore, the result [74, Theorem 15.1, p. 86] shows that the basis Bm+n can be replaced
by the basis
′
Bm+n = {U × V | U ∈ Bm and V ∈ Bn }.
For every k ≥ 1, since Rk is second-countable, it has a countable basis.e Thus we may assume
that Bm and Bn are countable and consequently Bm+n ′ is also countable. Obviously, we have
U ∈ Bm ⊆ Tm ⊆ Bm and V ∈ Bn ⊆ Tn ⊆ Bn so that
U × V ∈ B m × B n ⊆ Tm × Tn ⊆ B m × B n . (8.56)
′
If W ∈ T m+n , then it can be expressed as a union of basis elements of Bm+n ′
. Since Bm+n
is countable, this union must be countable too. Therefore, it follows from this fact, the fact
Bm × Bn is an σ-algebra and the set relation (8.56) that Tm+n ⊆ Bm × Bn and this means
that
Bm+n ⊆ Bm × Bn . (8.57)
Problem 8.12
Rudin Chapter 8 Exercise 12.
By [99, Exercise 7, p. 197], we have π2 < sinx x < 1 on (0, π2 ). Therefore, the integral (8.60)
becomes
Z A Z ∞ Z A | sin x|
dx |f (x, t)| dt = dx
0 0 0 x
Z π2 sin x Z A | sin x|
≤ dx + dx
0 x π x
2
Z A
π dx
< +
2 π x
2
π π
= + ln A − ln
2 2
<∞
Then the given relation implies that e−t ∈ L1 (m) on (0, ∞). Obviously, for t ≥ 0, we note from
the second inequality in (8.62) that
e−At
lim (cos A + t sin A) = 0,
A→∞ 1 + t2
so we deduce from this and Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem) that
Z ∞
e−At
lim (cos A + t sin A) dt = 0. (8.63)
A→∞ 0 1 + t2
Hence our desired result follows from the representation (8.61) and the limit (8.63), completing
the proof of the problem.
Remark 8.4
The improper integral in Problem 8.12 is known as Dirichlet integral.
f
See the Notes on [100, p. 165]
8.4. Miscellaneous Problems 275
Problem 8.13
Rudin Chapter 8 Exercise 13.
dµ = eiθ d|µ|.
Given E ∈ M and we define Aα to be the subset of E where cos(θ − α) > 0. Then we have
Z Z Z
µ(Aα ) = dµ = eiθ d|µ| = χAα eiθ d|µ|
Aα Aα E
Since χAα cos(θ − α) = cos+ (θ − α) by the Corollary following Theorem 1.14, the inequality
(8.64) becomes Z Z
|µ(Aα )| ≥ χAα cos(θ − α) d|µ| = cos+ (θ − α) d|µ|. (8.65)
E E
As in the proof of Lemma 6.3 from Rudin, we may choose α0 so as to maximize the integral
in the inequality (8.65). Since this maximum is at least as large as the average of the sum over
[−π, π], we obtain Z π Z
1
|µ(Aα0 )| ≥ cos+ (θ − α) d|µ| dθ.
2π −π E
Since |µ|(E) < ∞ by Theorem 6.4, E is σ-finite. It is clear that [−π, π] is also σ-finite. Thus
we deduce from Theorem 8.8 (The Fubini Theorem) and the integral in Lemma 6.3 that
Z π Z Z h Z π i
1 1 1
|µ(Aα0 )| ≥ cos+ (θ − α) d|µ| dθ = cos+ (θ − α) dθ d|µ| = |µ|(E).
2π −π E E 2π −π π
By [28, Example 9.2.1(c), p. 285], if T is the circle group and f : [0, 2π] → T is given by
f (t) = (cos t, sin t), then the Haar measure µ on T is given by
1
µ(S) = m f −1 (S) . (8.66)
2π
In particular, we have µ(T ) = 1. Since f is continuous and 2 < 2π, if we consider f ([0, 2]) = A,
then we see from the definition (8.66) that
1 1 1 1
µ(A) = m f −1 (A) = m([0, 2]) = = µ(S).
2π 2π π π
Since µ is a positive measure, we have |µ| = µ so that the Haar measure considered here gives an
example showing that π1 is the beast constant. We have completed the proof of the problem.
Problem 8.14
Rudin Chapter 8 Exercise 14.
276 Chapter 8. Integration on Product Spaces
Proof. Since f (t)tα and t−α satisfy the hypotheses of Theorem 3.5 (Hölder’s Inequality), we
have
Z x nZ x o1 n Z x o1 n Z x o 1 x−αq+1 1
f (t)tα t−α dt ≤
p q p q
xF (x) = f ptαp dt tαq dt = f p tαp dt ×
0 0 0 0 −αq + 1
which gives Z x
F p (x) ≤ (1 − αq)1−p x−1−αp f p tαp dt
0
and then
Z ∞ Z ∞ Z x
F p (x) dx ≤ (1 − αq)1−p x−1−αp f p tαp dt dx (8.67)
0 0 0
Z ∞ Z ∞
= (1 − αq)1−p x−1−αp χ(0,x) (t)f p tαp dt dx. (8.68)
0 0
As the space (0, ∞) is σ-finite, χ(0,x) x−1−αp f p tαp ≥ 0 and measurable on (0, ∞) × (0, ∞),
Theorem 8.8 (The Fubini Theorem) allows the inequality (8.67) can be rewritten as
Z ∞ Z ∞ Z ∞
F p (x) dx ≤ (1 − αq)1−p f p tαp χ(0,x) (t)x−1−αp dx dt
0 0 0
Z ∞ Z ∞
= (1 − αq)1−p f p tαp x−1−αp dx dt
0 t
Z ∞
1−p t−αp
= (1 − αq) f p tαp × dt
0 αp
Z ∞
= (1 − αq)1−p (αp)−1 f p dt.
0
If we define g(α) = (1 − αq)1−p (αp)−1 on (0, 1q ), then we can apply basic calculus to verify
1
that g will attain the absolute minimum at α = pq and its minimum value is
p p
.
p−1
Hence we have again proven the result of Problem 3.14 and this completes the proof of the
problem.
Problem 8.15
Rudin Chapter 8 Exercise 15.
|ϕ(x) − ϕ(y)| ≤ |x − y|
for all x, y ∈ [0, 2π]. Next, for x ∈ [0, 2π], we follow from the definition (8.69) that
Z x Z x
h(x) = ϕ(t) dt = (1 − cos t) dt = x − sin x.
−∞ 0
It is obvious that ϕ(y) = 1 − cos y > 0 on (0, x). Similarly, if y ∈ (0, x), then x − y ∈ (0, x) and
so ϕ(x − y) = 1 − cos(x − y) > 0 for all y ∈ (0, x). Consequently, we deduce that
on (0, 2π]. Suppose that x ∈ (2π, 4π). Then the definition (8.69) implies that
Z x Z x−2π
(ϕ ∗ ϕ)(x) = ϕ(y)ϕ(x − y) dy = ϕ(y)ϕ(x − y) dy.
0 0
Obviously, ϕ(y) = 1 − cos y > 0 on (0, x − 2π). If 0 < y < x − 2π, then 2π < x − y < x and so
ϕ(x − y) = 1 − cos(x − y) > 0 for all y ∈ (0, x − 2π). Therefore, the inequality (8.71) is till valid
whenever x ∈ (2π, 4π). In conclusion, we establish that
On the one hand, Statement (i) says that f ∗ g = 0 which implies that (f ∗ g) ∗ h = 0. On
the other hand, if x ∈ R, since f (x) = 1 for all x ∈ R, we have
Z ∞ Z ∞
[f ∗ (g ∗ h)](x) = f (y)(g ∗ h)(x − y) dy = (g ∗ h)(x − y) dy. (8.72)
−∞ −∞
By Theorem 7.26 (The Change-of-variables Theorem) (or its special case on p. 156), we see that
the last integral in the expression (8.72) becomes
Z ∞
(g ∗ h)(y) dy. (8.73)
−∞
Recall from the definition (8.69) that ϕ(y) = 0 outside [0, 2π] and ϕ(x−y) = 0 outside [x−2π, x].
If x ≤ 0 or x ≥ 4π, then [0, 2π] ∩ [x − 2π, x] is of measure 0 so that (ϕ ∗ ϕ)(x) = 0 outside (0, 4π).
Hence it follows from this fact and the expression (8.73) that
Z ∞ Z 4π Z 4π
[f ∗ (g ∗ h)](x) = (g ∗ h)(y) dy = (g ∗ h)(y) dy = (ϕ ∗ ϕ)(y) dy > 0
−∞ 0 0
by Statement (ii).
/ L1 (R). This completes
Now the inconsistence in Statement (iii) is due to the fact that f ∈
the proof of the problem.
Problem 8.16
Rudin Chapter 8 Exercise 16.
Proof. Suppose that (X, L , µ) and (Y, T , ν) are σ-finite measure spaces. Let f be an (L × T )-
measurable function on X × Y . If 0 ≤ f ≤ ∞ and 1 ≤ p < ∞, then we have the analogy of
Minkowski’s inequality in the question.
If p = 1, then it is exactly Theorem 8.8 (The Fubini Theorem). Suppose that 1 < p < ∞ and
q is the conjugate exponent of p. Since the inequality holds trivially when
Z hZ i1
p
f p (x, y) dµ(x) dλ(y) = ∞,
so without loss of generality, we assume further that this integral is finite. Let g ∈ Lq (µ). By
Theorem 8.8 (The Fubini Theorem) and Theorem 3.5 (Hölder’s Inequality), we have
Z hZ i Z Z
f (x, y) dλ(y) · |g(x)| dµ(x) = f (x, y) · |g(x)| dµ(x) dλ(y)
Z nZ o1n Z o1
p q
≤ f p (x, y) dµ(x) |g|q dµ(x) dλ(y)
Z hZ i1
p
= kgkq f p (x, y) dµ(x) dλ(y). (8.74)
then the inequality (8.74) and the assumption imply that Φ is a bounded linear functional on
Lq (µ). Applying Theorem 6.16 to Φ, there exists a unique h ∈ Lp (µ) such that
Z
Φ(g) = hg dµ(x). (8.76)
8.4. Miscellaneous Problems 279
Since Φ(|g|) = Φ(g), the uniqueness of h and the comparison of the representations (8.75) and
(8.76) show that Z
h = f (x, y) dλ(y).
Furthermore, Theorem 6.16 gives that kΦk = khkp and we deduce from this and the inequality
(8.74) that
nZ hZ ip o1
p
f (x, y) dλ(y) dµ(x) = khkp
= kΦk
= sup{|Φ(g)| | g ∈ Lq (µ) and kgkq = 1}
Z hZ i1
p
≤ f p (x, y) dµ(x) dλ(y)
Problem 9.1
Rudin Chapter 9 Exercise 1.
Proof. For every y ∈ R, |f (x)e−ixy | = |f (x)| = f (x) so that f (x)e−ixy ∈ L1 . By Definition 9.1
and Theorem 1.33, we have
Z Z ∞
1 ∞
1
|fb(y)| = √ f (x)e−ixy dx ≤ √ f (x) dx = fb(0)
2π −∞ 2π −∞
a.e. on R. Since f (x) > 0 on R, the expression (9.1) implies that αe−ixy = 1 a.e. on R. Thus it
gives y = 0, a contradiction. This completes the proof of the problem.
Problem 9.2
Rudin Chapter 9 Exercise 2.
281
282 Chapter 9. Fourier Transforms
(e−ibt − e−iat )
− √ , if t 6= 0;
2πit
= (9.2)
b−a
√ , otherwise.
2π
If x ≤ −n − 1, then x + n ≤ −1 so that
m [x − n, x + n] ∩ [−1, 1] = 0.
If x ≥ n + 1, then x − n ≥ 1 so that
m [x − n, x + n] ∩ [−1, 1] = 0.
2c
• Proof of fn = gn ∗ h. We first claim that fn ∈ L1 for each n = 1, 2, . . .. Recall the
π
sin x
basic fact that → 1 as x → 0. Thus if we define fn (0) = n, then fn is continuous on
x
R. Particularly, fn is bounded by a positive constant Mn on [−1, 1]. Since |fn (x)| ≤ x12 if
|x| ≥ 1, we have
Z Z Z
|fn (x)| dx = |fn (x)| dx + |fn (x)| dx
R |x|≥1 |x|≤1
Z ∞ Z −1
dx dx
≤ + + Mn
1 x2 −∞ x2
= 2 + Mn .
f[
− g = F(f − g) = F(f ) − F(g) = 0.
By Theorem 9.12 (The Uniqueness Theorem), we obtain f (x) = g(x) for almost all x ∈ R
which means that F is injective.
Assume that F was surjective. Since both L1 and C0 are Banach, we know from Theorem
5.10 that there is a δ > 0 such that
kfbk∞ ≥ δkf k1 (9.8)
for all f ∈ L1 . Recall the facts from the above assertions that
r r
2c π π 2 2
fn ∈ L , 1
gn ∗ h ∈ L ,1
fn = gn ∗ h and kc
fn k∞ = kgn ∗ hk∞ = · = ,
π 2 2 π π
so we get from the inequality (9.8) that
r
1 2
kfn k1 ≤ <∞
δ π
but this contradicts the fact kfn k1 → ∞. Hence F is not surjective.
• The set B = {fb| f ∈ L1 } is dense in C0 . To solve this part, we need the locally compact
Hausdorff version of the Stone-Weierstrass Theorema . In fact, it is
f[
∗ g = fb · b
g ∈ B.
By the definition, B is an algebra. Suppose that p, q ∈ R and we have two cases for
consideration.
– Case (i): p < q and p 6= −q. Consider the function
f (x) = e−|x| . (9.9)
By [100, Eqn. (3), p. 183], since f is even in R, we have
Z ∞ Z ∞ r
b −ixt ixt 2 1
f (t) = f (x)e dm(x) = f (x)e dm(x) = h1 (t) = · . (9.10)
−∞ −∞ π 1 + t2
Since fb(x) = fb(y) if and only if x = ±y, we obtain fb(p) 6= fb(q) in our case.
a
See [99, pp. 159 – 165] for the compact cases and [40, §4.7] for the locally compact Hausdorff cases
9.1. Properties of The Fourier Transforms 285
– Case (ii): p < q and p = −q. We consider g(x) = f (x)eix , where f is the function
(9.9). By Theorem 9.2(a), we have
r
b 2 1
gb(t) = f (t − 1) = · .
π 1 + (t − 1)2
In other words, B separates points on R. Furthermore, we see from the expression (9.10)
that fb(t) 6= 0 for each t ∈ R. Thus B vanishes at no point of R. Hence our desired result
follows immediately from Lemma 9.1.
Remark 9.1
The result in Problem 9.2 that F(L1 ) ⊂ C0 is called the Riemann-Lebesgue Lemma
([111, Theorem 1.4, p. 80]). See also §5.14 from Rudin.
Problem 9.3
Rudin Chapter 9 Exercise 3.
Note that b / L1 , so we can’t apply any inversion theorems (Theorems 9.11 or 9.14) directly.
gλ ∈
To solve this problem, we apply Theorem 9.13.
It is clear that gλ , gbλ ∈ L2 . Using the same notation as in Theorem 9.13(d), we have
Z A r Z A Z
ixt 1 2 sin λt ixt 1 A sin λt ixt
ψA (x) = gbλ (t)e dm(t) = √ · e dt = e dt
−A 2π π −A t π −A t
theorem for piecewise smooth functionsc . Indeed, we have (see [39, Theorem 7.6, p. 220] or
[107, Theorem 6.6.2, p. 330])
Lemma 9.2
If f is a piecewise smooth function on R such that f ∈ L1 , then we have
Z A
π
lim fb(t)eixt dm(t) = [f (x−) + f (x+)]
A→∞ −A 2
for every x ∈ R.
Problem 9.4
Rudin Chapter 9 Exercise 4.
Lemma 9.3
b b
If f ∈ L2 , then we have fb = f − , where fb denotes the Fourier transform of fb and
−
f (x) = f (−x).
Problem 9.5
Rudin Chapter 9 Exercise 5.
|fb(t)| ≤ |tfb(t)|
a.e. on R.
It remains to show that g is differentiable with the mentioned derivative in the question. To
this end, we define F (t) = fb(t) and G(t) = −itfb(t) so that
for all x ∈ R. Since |G(t)| = |tfb(t)|, the hypotheses give G ∈ L1 . Thus we deduce from Theorem
9.2(f) that Fb is differentiable and
Fb′ (t) = G(t)
b (9.15)
for all t ∈ R. By the integral (9.14) and Definition 9.1, we see that
Z ∞
g(x) = F (t)eixt dm(t) = Fb(−x).
−∞
Problem 9.6
Rudin Chapter 9 Exercise 6.
a.e. on R and
Z Z r
∞ 1
ti ti 2 sin t 2
tifb(t) = ti −ixt
f (x)e dm(x) = √ −ixt
e dx = √ · =i sin t.
−∞ 2π −1 2π t π
d
Hence we conclude that (f ′ )(t) 6= tifb(t) which completes the proof of the problem.
Problem 9.7
Rudin Chapter 9 Exercise 7.
Proof. Let f ∈ S. Denote F[f (x)](t) = fb(t), or simply F[f (x)] = fb, to be the Fourier transform
of f . We want
to show that, for every m and n = 0, 1, 2, . . ., there are positive numbers
Amn F[f (x)] < ∞ such that
|xn D m F[f (x)]| ≤ Amn (F[f (x)]).
Since |x2 f (x)| ≤ A02 (f ) for every x ∈ R, it means that
A02
|f (x)| ≤
x2
for every x 6= 0. By this, we have f ∈ L1 and thus Theorem 9.6 gives
F[f (x)](t) ≤
F[f (x)]
≤ kf k1 < ∞.
∞
It is easy to prove
that if f ∈ S, then f ′ (x)
∈ S and xf (x) ∈ S. These facts show that
D n (−ix)m f (x) belongs to S for every m, n = 0, 1, 2, . . . and so
n
F D (−ix)m f (x) < ∞ (9.16)
for every m, n = 0, 1, 2, . . ..
By Theorem 9.2(f), we observe that
d d2
{F[f (x)](t)} = F[−ixf (x)](t) and {F[f (x)](t)} = F[(−ix)2 f (x)](t).
dt dt2
Therefore, we see that
dm
D m F[f (x)](t) = {F[f (x)](t)} = F[(−ix)m f (x)](t) (9.17)
dtm
for every m = 0, 1, 2, . . .. To continue the proof, we need the following lemma:
Lemma 9.4
Let f ∈ S. Then we have F[f ′ (x)](t) = itF[f (x)](t).
9.1. Properties of The Fourier Transforms 289
for every n = 0, 1, 2, . . .. Particularly, by replacing f (x) by (−ix)m f (x) in the formula (9.19),
we gain
F[D n (−ix)m f (x) ](t) = (it)n F[(−ix)m f (x)](t)
or equivalently,
F[(−ix)m f (x)](t) = (it)−n F[D n (−ix)m f (x) ](t). (9.20)
Now we combine formulas (9.17) and (9.20) to conclude
(it)n D m F[f (x)](t) = F[D n (−ix)m f (x) ](t)
Remark 9.2
The class S in Problem 9.7 is called the Schwartz space or space of rapidly decreasing
functions on R.
Problem 9.8
Rudin Chapter 9 Exercise 8.
It is clear that either p or q must be finite. Suppose that 1 ≤ p < ∞. We need the
following result:
Lemma 9.5
Let p, q ∈ [1, ∞] be conjugate exponents. If f ∈ Lp and g ∈ Lq , then f ∗ g ∈ L∞
and
kf ∗ gk∞ ≤ kf kp · kgkq . (9.21)
Proof of Lemma 9.5. Suppose that σ(f )(x) = f (−x). By Theorem 3.8 and the fact
that σ(f−x )(y) = f−x (−y) = f (−y + x) for every y ∈ R, if x ∈ R, then
Z ∞
|(f ∗ g)(x)| = f (x − y)g(y) dm(y) = kσ(f−x ) × gk1 ≤ kσ(f−x )kp · kgkq . (9.22)
−∞
In addition, we have
nZ ∞ o1 nZ ∞ o1
p p
kσ(fx )kp = |f (−y − x)|p dy = |f (y)|p dy = kf kp . (9.23)
−∞ −∞
Hence we derive (9.23) by simply putting the result (9.23) back into the inequality
(9.22). This proves Lemma 9.5.
Now it is easily seen that the convolution of functions f and g is distributive, so for every
x ∈ R, it follows from Lemma 9.5 that
• h ∈ C0 if 1 < p < ∞. Recall from Theorem 3.14 that Cc (R) is dense in Lp and Lq . Choose
{fn } ⊆ Cc (R) ⊂ Lp and {gn } ⊆ Cc (R) ⊂ Lq such that kfn − f kp → 0 and kgn − gkq → 0
as n → ∞.
Let M be a positive constant such that supp (fn ), supp (gn ) ⊆ [−M, M ]. Thus for every
|x| > 2M and every y ∈ R, we have either y ∈ R \ [−M, M ] or x − y ∈ R \ [−M, M ].
Otherwise, it will give the contradiction that |x| = |x − y + y| ≤ 2M . Thus this fact gives
fn (x − y)gn (y) = 0 and then
(fn ∗ gn )(x) = 0
for every x such that |x| > 2M , i.e., supp (fn ∗ gn ) ⊆ [−2M, 2M ] which means that each
fn ∗ gn is an element of Cc (R).
Next, by Lemma 9.5 and then Theorem 3.9, we obtain
≤ (kfn − f kp + kf kp ) × kgn − gm kq
+ kfn − fm kp × (kgm − gkq + kgkq ). (9.24)
for every x ∈ R. In other words, {fn ∗gn } is Cauchy in Cc (R) relative to the metric induced
by the supremum norm.
Finally, using similar argument as in proving the inequality (9.24), we can show that
one can find a positive integer N ′ such that n ≥ N ′ implies
Problem 9.9
Rudin Chapter 9 Exercise 9.
Proof. We have Z Z
g(x) = χ[x,x+1] (t)f (t) dt = χ[0,1] (t − x)f (t) dt. (9.26)
R R
By the change of variable formula [100, p. 156], the integral (9.26) reduces to
Z Z
g(x) = χ[0,1] (t)f (t + x) dt = χ[0,1] (t)f−x (t) dt.
R R
For every x ∈ R, by Theorem 9.5, the mapping x 7→ f−x is continuous at x. In other words,
given ǫ > 0, there exists a δ > 0 such that kf−x − f−y kp < ǫ for all y ∈ R with |x − y| < δ.
Since f−x − f−y ∈ Lp and χ[0,1] ∈ Lq , where q be the conjugate exponent of p, we deduce from
Theorem 3.8 and then Theorem 1.33 that
Z
|g(x) − g(y)| = χ[0,1] (t)[f−x (t) − f−y (t)] dt
R
=
χ[0,1] · (f−x − f−y )
1
292 Chapter 9. Fourier Transforms
Thus there is a positive integer N such that for all n ∈ Z, |n| ≥ N implies that
Z n+1
|f (t)|p dt < ǫp . (9.29)
n
Finally, we use Theorem 1.33 and then apply the limit (9.30) to the inequality (9.27) with a = x
and b = x + 1 to establish
Z x+1 n Z x+1 o1
p
lim |g(x)| ≤ lim |f (t)| dt ≤ lim |f (t)|p dt = 0.
|x|→∞ |x|→∞ x |x|→∞ x
Z y Z y+1
≤ |f (t)| dt + |f (t)| dt
x x+1
≤ 2(y − x)kf k∞ .
for almost every x ∈ R so that kgk∞ ≤ kf k∞ . This completes the proof of the problem.
Problem 9.10
Rudin Chapter 9 Exercise 10.
• Cc∞ does not consist of 0 alone. In Problem 7.8(d), we define the bump function
ψ : R → [0, ∞) by
1
exp − , if x ∈ (−1, 1);
1 − x2
ψ(x) = (9.31)
0, otherwise.
This is an element of C ∞ . Since supp (ψ) = [−1, 1], it must be true that ψ ∈ Cc∞ .
• If f ∈ L1loc and g ∈ Cc∞ , then f ∗ g ∈ C ∞ . To begin with, we have to show that f ∗ g is
well-defined first. Since supp (g) is compact in R, let M be a positive number such that
supp (g) ⊆ [−M, M ]. Thus we have
Z Z M
(f ∗ g)(x) = f (x − y)g(y) dy = f (x − y)g(y) dy. (9.32)
R −M
Similarly, we may extend the lower limit and the upper limit of the second integral in
(9.33) to x − (M + 1) and x + (M + 1) respectively. Therefore, the expression (9.33) can
be reduced to
Z x+(M +1)
(f ∗ g)(x + δ) − (f ∗ g)(x) g(x − y + δ) − g(x − y)
= · f (y) dy. (9.34)
δ x−(M +1) δ
Now we fix x ∈ R. Let g = g1 + ig2 . Since g ∈ Cc∞ , g1 and g2 are real differentiable
functions. It follows from the Mean Value Theorem [99, Theorem 5.10, p. 108] that there
exist ξ, η ∈ (x − y, x − y + δ) such that
Then they imply that g(x − y + δ) − g(x − y) = δ[g1′ (ξ) + ig2′ (η)] which gives
g(x − y + δ) − g(x − y) q p
≤ [g1′ (ξ)]2 + [g2′ (η)]2 ≤ |g′ (ξ)|2 + |g ′ (η)|2 . (9.35)
δ
Since g ∈ Cc∞ , g′ is continuous and thus bounded on [x − y, x − y + δ]. Recall that
y ∈ [x − (M + 1), x + (M + 1)] and 0 < δ < 1. Thus, there is an A > 0 such that |g ′ (t)| ≤ A
for every t ∈ [−M − 2, M + 2]d and then we observe from the inequality (9.35) that
g(x − y + δ) − g(x − y) √
· f (y) ≤ 2A|f (y)|.
δ
Since f is integrable on [x − (M + 1), x + (M + 1)], Theorem 1.34 (Lebesgue’s Dominated
Convergence Theorem) ensures that the expression (9.34) gives
(f ∗ g)(x + δ) − (f ∗ g)(x)
(f ∗ g)′ (x) = lim
δ→0 δ
Z x+(M +1)
g(x − y + δ) − g(x − y)
= lim · f (y) dy
δ→0 x−(M +1) δ
Z x+(M +1)
g(x − y + δ) − g(x − y)
= lim · f (y) dy
x−(M +1) δ→0 δ
Z x+(M +1)
= g′ (x − y) · f (y) dy
x−(M +1)
= (g′ ∗ f )(x).
(f ∗ g)(n) = g(n) ∗ f.
where ψ is the bump function (9.31). It is clear that c is a finite positive number because
ψ > 0. If we define ϕ : R → [0, ∞) by
then ϕ satisfies the conditions that ϕ ∈ Cc∞ , supp (ϕ) = [−1, 1], ϕ ∈ L∞ and kϕk1 = 1.
Next, for ǫ > 0, we definee
ϕǫ (x) = ǫ−1 ϕ(ǫ−1 x).
We observe that
Z Z
f ∗ ϕǫ (x) − f (x) = f (x − y)ϕǫ (y) dy − f (x)ϕǫ (y) dy
ZR R
= [f (x − y) − f (x)]ϕǫ (y) dy
ZR
= [f (x − y) − f (x)]ϕ(ǫ−1 y)ǫ−1 dy. (9.36)
R
Since the integrand in the expression (9.38) is nonnegative, we may apply Theorem 8.8 (The
Fubini Theorem) to change the order of integration in the expression (9.38) to conclude
that Z
kf ∗ ϕǫ − f k1 ≤ |ϕ(z)| · kfǫz − f k1 dz. (9.39)
R
kf ∗ ϕǫ − f k1 → 0
1
as ǫ → 0. Now, if we let ǫ = n and gn = ϕ 1 , then we can obtain our desired sequence.
n
– 0 ≤ ψi ≤ 1 for 1 ≤ i ≤ s;
– supp (ψi ) ⊂ Vn ;
– ψ1 (x) + ψ2 (x) + · · · + ψs (x) = 1 for every x ∈ Kn .
e
The collection {ϕǫ } is called an approximate identity or a sequence of mollifiers on R, see [96,
Definition 6.31, p. 173] or [22, p. 108].
f
For a proof of this, please read [124, Problem 10.6, pp. 266, 267]. Of course, our s depends on n.
296 Chapter 9. Fourier Transforms
If we define gn : R → R by
n
gn =
(ψ1 + ψ2 + · · · + ψs ),
2
then the above conditions imply that gn ∈ Cc∞ , gn (x) = n2 on Kn , gn (x) = 0 outside Vn
and 0 ≤ gn (x) ≤ n2 for all x ∈ R.
Let x ∈ R and f (x) ≥ 0 on R. On the one hand, we have
Z 1 Z
n n 1
(f ∗ gn )(x) ≥ f (x − y) dy = f (y) dy. (9.40)
2 −1
n
m B 0, n1 ) B(x, 1 ) n
Since f ∈ L1loc , the integral (9.40) is finite. On the other hand, we have
Z 1 −n
n n +2
(f ∗ gn )(x) ≤ f (x − y) dy
2 − 1 −2−n
n
Z
n 1
= 1+ n × f (y) dy. (9.41)
2 m B(0, n1 + 2−n ) B(x, n1 +2−n )
Similarly, the integral (9.41) is finite because of the fact that f ∈ L1loc . Now if x ∈ R is a
Lebesgue point of f , then we combine the inequalities (9.40) and (9.41) to get
Z
1
lim (f ∗ gn )(x) = lim f (y) dy = f (x). (9.42)
n→∞ n→∞ m B(0, 1 ) 1
B(x, n )
n
function such that it is decreasing on (0, ∞) and h1 ∈ L1 . With the aid of [108, Theorem
1.25, p. 13], we see that
Z
lim (f ∗ hλ )(x) = f (x) h1 (t) dm(t) = f (x)
λ→0 R
for every Lebesgue point x of f . By Theorem 7.7, we obtain our desired result that
(f ∗ hλ )(x) → f (x) a.e. on R if f ∈ L1 as λ → 0.
It remains to prove that f ∗ hλ ∈ C ∞ . To see this, the definition of hλ shows clearly that
(n)
hλ ∈ C ∞ . In addition, direct computation gives easily that hλ is bounded on R for every
positive integer n. As a consequence of [40, Proposition 8.10, p. 242], we may conclude
immediately that f ∗ hλ ∈ C ∞ .
Problem 9.11
Rudin Chapter 9 Exercise 11.
Proof. Suppose that f is a Schwartz function. We are going to show that the Poisson summation
formula holds in this situation and consider the limiting case as α → 0 under the extra hypothesis
that fb ∈ L1 .
Since f ∈ S, there exists a positive constant M such that |f (x)| ≤ M and |x2 f (x)| ≤ M
for all x ∈ R. In particular, if k ≥ 1, then
M M
|f (x + 2kπ)| ≤ ≤
(x + 2kπ)2 (2k − 1)2 π 2
for all x ∈ [−π, π]. By the Weierstrass M -test [99, Theorem 7.10, p. 148], we see that the
series
∞
X
f (x + 2kπ)e−inx
k=1
converges uniformly and absolutely on [−π, π] so that the integral and summation in the
expression (9.43) can be interchanged (by [99, Corollary, p. 152]) and we get
∞ Z π
1 X
Fb(n) = f (x + 2kπ)e−inx dx
2π −π
k=−∞
∞ Z (2k+1)π
1 X
= f (x)e−inx dx
2π (2k−1)π
k=−∞
298 Chapter 9. Fourier Transforms
Z ∞
1
= f (x)e−inx dx
2π −∞
= ϕ(n)
Let γ > 0 and g(x) = f (γx). Since g ∈ S, the preceding formula (9.46) also gives
∞
X ∞
X
g(2kπ) = ψ(n), (9.47)
k=−∞ n=−∞
where Z ∞
1
ψ(t) = g(x)e−itx dx.
2π −∞
and after substituting this back into the formula (9.47) and then using the definition of g
, we obtain
1 n
∞
X X∞
f (2γkπ) = ϕ . (9.48)
n=−∞
γ γ
k=−∞
1 2π
Suppose that α = γ > 0 and β = α > 0. Then we have αβ = 2π and the formula (9.48)
becomes
∞
X ∞
X
f (kβ) = α ϕ(nα). (9.49)
k=−∞ n=−∞
• The case when α → 0. Now we suppose further that the inequality (9.51) in Remark
9.3 below holds. On the one hand, we know that
Z ∞ Z ∞
1 X n
∞
X ∞
lim α ϕ(nα) = lim ϕ = ϕ(x) dx = fb(x) dm(x)
α→0 N →∞ N N −∞ −∞
n=−∞ n=−∞
∞
X ∞
X 2πk ∞
X
lim f (kβ) = lim f = lim f (2πkN ). (9.50)
α→0 α→0 α N →∞
k=−∞ k=−∞ k=−∞
holds for all x 6= 0, where M is a positive constant. Thus this shows that
∞
X X∞ ∞
X M
f (2πkN ) ≤ |f (2πkN )| ≤ →0
|2πkN |2
k=−∞ k=−∞ k=−∞
k6=0 k6=0 k6=0
Remark 9.3
If there exists a constant M and p > 1 such that
M
|f (x)| ≤ (9.51)
|x|p
for all sufficiently large x, then the series (9.44) still converges uniformly and absolutely on
[−π, π] so that the Poisson summation formula (9.49) also holds in this case.
Problem 9.12
Rudin Chapter 9 Exercise 12.
Proof. We note that f is not a Schwartz function, but the result [99, Theorem 8.6(f)] ensures
that f satisfies the inequality (9.51), so we may apply the first assertion of Problem 9.11. First
of all, since e−|x| cos tx and e−|x| sin tx are even and odd functions in x respectively, we have
Z ∞ Z
1 1 ∞ −x
ϕ(t) = e−|x| e−itx dx = e cos tx dx.
2π −∞ π 0
∞
1 + e−β 1 X α
−β
=
1−e π n=−∞ 1 + n2 α2
∞
eβ + 1 1 X α
= , (9.53)
eβ − 1 π n=−∞ 1 + n2 α2
where αβ = 2π. If we take β = 2πγ, then α = γ1 . Hence we deduce from the formula (9.53) that
∞
e2πγ + 1 1 X γ
2πγ
= (9.54)
e −1 π n=−∞ γ 2 + n2
and the desired result follows if we replace γ by α in the formula (9.54). This completes the
proof of the problem.
Problem 9.13
Rudin Chapter 9 Exercise 13.
Proof.
(a) By Definition 9.1 and similar argument as in the proof of Problem 9.12, we have
Z ∞ Z ∞ r Z
b −ixt −cx2 −ixt 2 ∞ −cx2
fc (t) = fc (x)e dm(x) = e e dm(x) = e cos tx dx. (9.55)
−∞ −∞ π 0
1 t2
fbc (t) = √ exp − .
2c 4c
1 x2
√ exp − = exp(−cx2 )
2c 4c
4c2 − 1 √
exp x2 = 2c (9.56)
4c
for every x ∈ R. Since the right-hand side of the equation (9.56) is constant, 4c2 − 1 = 0
and then c = 21 . Direct checking immediately shows that c = 21 satisfies the equation
(9.56).
Z ∞
= exp{−[(a + b)y 2 − 2axy + ax2 ]} dy. (9.57)
−∞
Using [47, §2.33, Eqn. 1, p. 108] and the facts thatg erf(∞) = 1 and erf(−∞) = −1, we
are able to represent the integral (9.57) as
r
1 π ab 2 √ ax y=∞
(fa ∗ fb )(x) = exp − x × erf a + by − √
2 a+b a+b a + b y=−∞
r
π ab 2
= exp − x .
a+b a+b
(d) By Problem 9.7, we know that fc ∈ S. Thus the definition and the formula [47, §2.33,
Eqn. 1, p. 108] together give
Z ∞ Z ∞ t2
1 2 1 1
ϕ(t) = e−cx e−itx dx = exp[−(cx2 + itx)] dx = √ exp − .
2π −∞ 2π −∞ 2 πc 4c
Problem 9.14
Rudin Chapter 9 Exercise 14.
We first show the analogue of Theorem 9.6 that fb ∈ C0 (Rk ) and kfbk∞ ≤ kf k1 . To this end, we
imitate the proof of Theorem 9.6. The required inequality is obvious from the definition (9.58)
and the integrability of f .
To prove fb ∈ C0 (Rk ), we need the analogue of Theorem 9.5 for Rk :h
g
Here erf(x) is the error function.
h
Its proof is very similar to that of Theorem 9.5, so we omit the details here.
302 Chapter 9. Fourier Transforms
Lemma 9.6
For any function f on Rk and every y ∈ Rk , let fy be the translate of f defined by
fy (x) = f (x − y)
y 7→ fy
Let t ∈ Rk . If tn → t, then
Z
|fb(tn ) − fb(t)| ≤ |f (x)| · | exp(−itn · x) − exp(−it · x)| dmk (x). (9.59)
Rk
Since the integrand of the integral (9.59) is bounded by 2|f (x)| ∈ L1 (Rk ), Theorem 1.34
(Lebesgue’s Dominated Convergence Theorem) implies that
fb(tn ) → fb(t)
= − exp(−ix · y)
which tends to 0 as |y1 | → 0 or equivalently, as η1 → ±∞. Or we can say this way: Given ǫ > 0,
there exists a δ > 0 such that
kf0 − fy1 k1 < 2ǫ
9.3. Fourier Transforms on Rk and its Applications 303
for all y1 ∈ Rk for which |y1 | < δ or |η1 | > πδ . Notice also that η1 → ±∞ means that |y| → ∞.
By similar argument as the previous part, we note that the result (9.61) also holds if y1 is
replaced by any yj = (0, . . . , 0, ηπj , 0, . . . , 0) with ηj 6= 0, where j = 2, 3, . . . , k. Write it explicitly,
we have
kf0 − fyj k1 < 2ǫ (9.62)
for all yj ∈ Rk for which |ηj | > πδ , where j = 1, 2, . . . , k. Now for arbitrary y ∈ Rk \ {0}, there
is at least one j such that |ηj | ≥ |y| |y|
k . Otherwise, we have |ηj | < k for every j = 1, 2, . . . , k, but
2
this means that k < 1, a contradiction. Thus if we take
n kπ o
K = y ∈ Rk |y| ≤ ,
δ
then K is compact and for y ∈
/ K, we have
|y| π
|ηj | ≥ >
k δ
for some 1 ≤ j ≤ k. Hence the inequality (9.62) always holds for this j and then we conclude
from the inequality (9.61) that
|fb(y)| < ǫ
where t = (t1 , t2 , . . . , tk ). Let λ > 0. Then 0 < H(t) ≤ 1 and H(λt) → 1 as λ → 0. Define
Z
hλ (x) = H(λt) exp(it · x) dmk (t).
Rk
A simple computation with repeated applications of Theorem 8.8 (The Fubini Theorem), it gives
k 1
Y 2 2 λ
hλ (x) = .
j=1
π λ2 + ξj2
By Theorem 8.8 (The Fubini Theorem) and [100, Eqn. (4), p. 183], we get
Z "Z #
k
Y 2 1 λ
2
hλ (x) dmk (x) = dm(ξ1 ) = 1.
Rk R π
j=1
λ2 + ξj2
Similarly, we have the following analogue of Theorem 9.10: If 1 ≤ p < ∞ and f ∈ Lp (Rk ), then
lim kf ∗ hλ − f kp = 0. (9.64)
λ→0
• The Inversion Theorem for Rk . Suppose that f, fb ∈ L1 (Rk ). The integrand on the
right-hand side of the formula (9.63) are bounded by |fb(t)|. By Theorem 1.34 (Lebesgue’s
Dominated Convergence Theorem) and the fact that H(λt) → 1 as λ → 0, the right-hand
side of the formula (9.63) converges to
Z
g(x) = fb(t) exp(ix · t) dmk (t).
Rk
for all x ∈ Rk . By the limit (9.64) and Theorem 3.12, there is a sequence {λn } such that
λn → 0 and
lim (f ∗ hλn )(x) = f (x)
n→∞
a.e. on Consequently, we have f (x) = g(x) a.e. on Rk . Since fb ∈ L1 (Rk ) and g(−x)
Rk .
is the Fourier transform of fb, the analogue of Theorem 9.6 above implies that g ∈ C0 (Rk )
as desired.
• The Plancherel Theorem for Rk . Fix f ∈ L1 (Rk ) ∩ L2 (Rk ). Put fe(x) = f (−x) and
g = f ∗ fe. Then we have
Z Z
g(x) = f (x − y)f (−y) dmk (y) = f (x + y)f (y) dmk (y) = hf−x , f i .
Rk Rk
By the use of Lemma 9.6 and Theorem 4.6, it can be shown that g is a continuous function
on Rk and thus Theorem 4.2 (The Schwarz Inequality) gives
Since g is continuous and bounded on Rk , the analogue of Theorem 9.9j shows that
Hence it follows from the expression (9.65) and the two limits (9.66) and (9.67) that
Z
kfbk22 = lim H(λt)bg (t) dmk (t) = lim (g ∗ hλ )(0) = kf k22 < ∞. (9.68)
λ→0 Rk λ→0
Suppose that
Y = {fb| f ∈ L1 (Rk ) ∩ L1 (Rk )}.
i
The analogue of Theorem 8.14 is that f, g ∈ L1 (Rk ) implies f ∗ g ∈ L1 (Rk ), but we won’t give a proof here.
j
If g ∈ L∞ (Rk ) and g is continuous at a point x ∈ Rk , then (g ∗ hλ )(x) → g(x) as λ → 0.
9.3. Fourier Transforms on Rk and its Applications 305
The equality (9.68) implies that Y is a subspace of L2 (Rk ). We claim that Y is dense in
L2 (Rk ). To this end, we recall from Problem 4.1 that (Y ⊥ )⊥ = Y . Now we observe that
Y ⊥ = {0} if and only if
Thus it suffices to show that Y = {0}⊥ . Let a ∈ Rk and λ > 0. Denote the mapping
x 7→ exp(ia · x)H(λx) by F . It is clear that F (x) ∈ L1 (Rk ) ∩ L2 (Rk ) for all x ∈ Rk and
Z Z
Fb(t) = F (x) exp(−ix · t) dmk (t) = exp i(a − t) · x H(λx) dmk (x) = hλ (a − t)
Rk Rk
for all a ∈ Rk . By the limit (9.64), we see that w = 0, i.e., Y ⊥ = {0} and thus Y = L2 (Rk ).
Now we define Φ : L1 (Rk ) ∩ L2 (Rk ) → Y by
Φ(f ) = fb.
Then the equality (9.68) certainly shows that Φ is an L2 (Rk )-isometry. Since both L1 (Rk )∩
L2 (Rk ) and Y are dense in L2 (Rk ), as the paragraph before [100, Eqn. (10), p. 187]
indicates that Φ can be extended to an isometry
e : L2 (Rk ) → L2 (Rk ).
Φ
If we denote fb = Φ(f
e ) (of course, it is the Plancherel transform), then the equality (9.68)
holds for every f ∈ L2 (Rk ). Using a similar argument as in the proof of Theorem 4.18, we
can prove the Parserval formula
Z Z D E
hf, gi = f (x)g(x) dmk (x) = fb(t)b
g (t) dmk (t) = fb, gb
Rk Rk
for every f, g ∈ L2 (Rk ). Finally, for the symmetric relation in Theorem 9.13(d), the ϕA
and ψA should be replaced by
Z Z
fA (t) =
ϕ f (x) exp(−ix · t) dmk (x) and ψ fA (x) = fb(t) exp(ix · t) dmk (t)
[−A,A]k [−A,A]k
as A → ∞.
ϕ(f ) = fb(t).
Here it is seen that the content up to Eqn. (6) in §9.22 can be extended to Rk easily. If
x = ξe1 + ξ2 e2 + · · · + ξk ek , where {e1 , e2 , . . . , ek } is the usual basis of Rk , then we have
Thus each βj satisfies the hypotheses qon [100, p. 192] which implies βj (ξj ) = exp(−itj ξj )
for a unique tj ∈ R. By the formula (9.69), we obtain
Problem 9.15
Rudin Chapter 9 Exercise 15.
Proof. Let A : Rk → Rk be a linear operator with det A > 0. Thus A−1 exists. Let y = A−1 (x).
Then we follow from Theorem 7.26 (The Change-of-variables Theorem) (with X = Y = V = Rk ,
T = A−1 which is one-to-one and differentiable on Rk , T (Rk ) = A−1 (Rk ) = Rk and y = A−1 (x))
that
g (z) = f[
b ◦ A(z)
Z
= f A(y) exp(−iy · z) dmk (y)
|R
k
{z
Z
}
This is f dm.
Y
Z
= f (x) exp[−iA−1 (x) · z] × | det A−1 | dmk (x) (9.70)
Rk | {z }
−1
This is (f ◦ T )(x) = (f ◦ A )(x).
Lemma 9.7
If A is a k × k matrix, then for all x, y ∈ Ck , we have hA(x), yi = x, AT (y) ,
where AT is the transpose of A.
Proof of Lemma 9.7. Using [71, p. 109; Example 5.3.1, p. 286], it can be seen easily
that
hA(x), yi = [A(x)]T y = xT AT (y) = x, AT (y) ,
completing the proof of Lemma 9.7.
f\
◦ A = | det A| · (fb ◦ A−T ). (9.71)
Particularly, let A be a rotation matrix of Rk . Then it preserves the distance of x from the
origin, i.e., hA(x), A(x)i = hx, xi. By [7, Proposition 5.1.13, p. 134], A is orthogonal which
means
AT = A−1 .
Furthermore, this implies that | det A| = 1, so the formula (9.71) can be simplified to
f\
◦ A = fb ◦ A. (9.72)
In other words, if f is invariant under rotations, then f ◦A = f and the expression (9.72) implies
fb ◦ A = fb,
i.e., fb is also invariant under rotations. This completes the proof of the problem.
Problem 9.16
Rudin Chapter 9 Exercise 16.
Proof. We make the following assumption: Suppose that f ∈ S(Rk ), where S(Rk ) is the class
of all functions f : Rk → C such that f ∈ C ∞ and xβ ∂ α f is bounded for all multi-indices α
∂f
and β.k Let 1 ≤ j ≤ k, x = (xj , xj ) and F (x) = , where x = (x1 , . . . , xj , . . . , xk ) and
∂xj
xj = (x1 , . . . , xj−1 , xj+1 , . . . , xk ). We note that
x · y = x1 y1 + x2 y2 + · · · + xk yk = xj · yj + xj yj .
For every y ∈ Rk , we follow from this, the definition in Problem 9.14 and Lemma 9.4 that
Z
∂f
Fb(y) = (x) exp(−ix · y) dmk (x)
Rk ∂xj
Z h Z ∂f i
= exp(−ixj · yj ) dmk−1 (xk ) (xj , xj ) exp(−ixj · yj ) dm(xj )
Rk−1 ∂xj
| R {z }
Apply Lemma 9.4 to this.
Z h Z i
= exp(−ixj · yj ) dmk−1 (xk ) iyj f (x) exp(−ixj · yj ) dm(xj )
Rk−1
Z R
k
In fact, S is the Schwartz space on Rk , see Remark 9.2.
308 Chapter 9. Fourier Transforms
for all x ∈ Rk .
Suppose that f has continuous second derivatives so that the Laplacian ∆f is well-defined.
Suppose further that supp (f ) is compact. In this case, f ∈ L1 (Rk ) so that fb is well-defined on
Rk . Furthermore, we suppose that A is a rotation of Rk about the origin 0 and gA = ∆(f ◦ A).
Then we deduce from the formulas (9.72) and (9.73) that
2 \
gc 2 b 2 b
b(Ax) = g[
A (x) = −|x| · f ◦ A(x) = −|Ax| · (f ◦ A)(x) = −|Ax| · f (Ax) = g ◦ A(x).
\
By the definition, we have ∆(f \
◦ A) = (∆f ) ◦ A. To go further, we need the analogue of
Theorem 9.12 (The Uniqueness Theorem):
Lemma 9.8
f (x) = 0 (9.74)
Proof of Lemma 9.8. Since fb = 0 on Rk , we have fb ∈ L1 (Rk ) and the result (9.74)
follows from the Inversion Theorem for Rk (Problem 9.14). If f is continuous on Rk ,
the definition of continuity ensures that the relation (9.74) holds everywhere on Rk .
Thus we have ∆(f ◦ A) ∈ L1 (Rk ) and (∆f ) ◦ A ∈ L1 (Rk ) so that F ∈ L1 (Rk ). Since Fb(t) = 0
for all t ∈ Rk and F is continuous on Rk , Lemma 9.8 implies that
holds everywhere on Rk . It is well-known that any rotation about a point p is the composition of
a rotation about the origin 0 and a translation. By this fact, the analysis preceding Lemma 9.8
and the hypothesis that ∆ commutes with translations, we conclude that the expression (9.75)
is still valid for every rotation about every point p.
It is time to consider the general situation. Fix a p ∈ Rk . There exists a r > 0 such
that p ∈ Kr = B(0, r) ⊂ Vr = B(0, 2r). By [99, Exercise 6, p. 289], there exists functions
ψ1 , . . . , ψs ∈ C ∞ (Rk ) such thatl
• 0 ≤ ψi ≤ 1 for 1 ≤ i ≤ s;
• supp (ψi ) ⊂ Vr ;
If we define Ψ : Rk → R by
Ψ = ψ1 + ψ2 + · · · + ψs ,
l
For a proof of this, please read [124, Problem 10.6, pp. 266, 267 ].
9.3. Fourier Transforms on Rk and its Applications 309
then the above conditions imply that Ψ ∈ Cc∞ (Rk ) and Ψ(x) = 1 on Kr . Thus we have f · Ψ = f
on Kr . Particularly, we have
∂Ψ ∂2Ψ
(f · Ψ)(p) = f (p) and (p) = (p) = 0
∂xj ∂x2j
for 1 ≤ j ≤ k. By the definition of the Laplacian and the result (9.76), we yield
Xk Xk
∂ 2 (f · Ψ) ∂2f
∆(f · Ψ) (p) = 2 (p) = (p) = (∆f )(p). (9.77)
j=1
∂xj j=1
∂x2j
Note that f · Ψ is compactly supported in this case. If Ap = q, then q ∈ Kr . On the one hand,
the expressions (9.75) (with f replaced by f · Ψ) and (9.77) give
∆ f · Ψ) ◦ A (p) = [(∆(f · Ψ)) ◦ A](p) = ∆(f · Ψ) (q) = (∆f )(q) = [(∆f ) ◦ A](p). (9.78)
on Kr , we have (f · Ψ) ◦ A = f ◦ A on Kr so that
∆ (f · Ψ) ◦ A (p) = ∆(f ◦ A) (p) (9.79)
Finally, by combining the expressions (9.78) and (9.79) and using the fact that p is arbitrary,
we may conclude that our desired formula (9.75) also holds everywhere on Rk in this general
case. This completes the proof of the problem.
Problem 9.17
Rudin Chapter 9 Exercise 17.
Proof. We prove the case for Rk directly. Suppose that ϕ : Rk → C is a Lebesgue measurable
character. We define Φ : L1 (Rk ) → C by
Z
Φ(f ) = f (x)ϕ(x) dmk (x). (9.80)
Rk
Since ϕ is a character, we have ϕ(x) = ϕ(x − y)ϕ(y). Substituting this into the expression
(9.81), we see that
Z Z
Φ(f ∗ g) = f (x − y)g(y)ϕ(x − y)ϕ(y) dmk (y) dmk (x)
ZR h RZ
k k
i
= f (x − y)ϕ(x − y) dmk (x) g(y)ϕ(y) dmk (y)
Rk Rk
310 Chapter 9. Fourier Transforms
Z
= Φ(f )g(y)ϕ(y) dmk (y)
Rk
= Φ(f ) · Φ(g).
holds for all f ∈ L1 (Rk ). Since f (x) = e−|x| is in L1 (Rk ) and f (x) 6= 0 on Rk , the result (9.83)
implies that
ϕ(x) = exp(−iy · x) (9.84)
a.e. on Rk . To proceed further, we need the following result which is a generalization of Problem
7.6 and its proof uses the Steinhaus Theorem in Rk (see Remark 7.2), but we don’t present
it here.
Lemma 9.9
Suppose that G is a subgroup of Rk (relative to addition), G 6= Rk , and G is
Lebesgue measurable. Then mk (G) = 0.
To apply this lemma, we consider h(x) = ϕ(x) − exp(−iy · x) and G = {x ∈ Rk | h(x) = 0}.
For a, b ∈ G, if a + b ∈
/ G, then we have h(a + b) 6= 0 and
Problem 9.18
Rudin Chapter 9 Exercise 18.
As a consequence, Zorn’s Lemma can be used to prove that every vector space has a
basis. In fact, if A is a linearly independent subset of a vector space V over a field K, then
there is a basis B of V that contains A.m By this result, there exists a basis of R over Q
with the usual addition and scalar multiplication.n
Let x ∈ R and H be a basis of R over Q. Thus x has a unique representation
x = r 1 b1 + r 2 b2 + · · · + r n bn ,
f (x) = r1 + r2 + · · · + rn .
It is obvious that f satisfies the equation (9.85). Assume that f was continuous on R.
Then f (R) must be connected (see [99, Theorem 4.22, p. 93]), but f (R) = Q which is a
contradiction.
• If f is Lebesgue measurable and satisfies the equation (9.85), then f is contin-
uous. We use the following special form of Theorem 2.24 (Lusin’s Theorem):o
Lemma 9.11
Suppose that f : [a, b] → R is Lebesgue measurable. For every ǫ > 0, there is a
compact set K ⊆ [a, b] such that m([a, b] \ K) < ǫ and f |K is continuous.
By Theorem 2.20(c), m(K − η) = m(K), so we deduce from the inequalities (9.86) and
(9.87) that 1 + η ≥ 2m(K) > 43 and thus η > 13 , a contradiction. Hence we have
K ∩ (K − η) 6= ∅.
m
For a hint of this proof, please refer to [74, Exercise 8, p. 72]
n
This kind of basis is called a Hamel basis.
o
Read [40, Exercise 44, p. 64] or [93, p. 66]
312 Chapter 9. Fourier Transforms
In other words, f is continuous at 0. Now for any x ∈ [0, 1], the inequality (9.88) implies
clearly that
• If graph (f ) is not dense in R2 and satisfies the equation (9.85), then f is con-
tinuous. Suppose that f is discontinuous. Then f cannot be of the form f (x) = cx for
any constant c. We have
graph (f ) = { x, f (x) | x ∈ R}.
Choose a nonzero real number x1 . Then there exists another nonzero real number x2 such
that
f (x1 ) f (x2 )
6=
x1 x2
or equivalently,
x1 f (x1 )
x2 f (x2 ) 6= 0.
In other words, the vectors u = x1 , f (x1 ) and v = x2 , f (x2 ) are linearly independent
and thus span the space R2 . Let p ∈ R2 . Given ǫ > 0. Since Q2 is dense in R2 , one can
find q1 , q2 ∈ Q such that
|p − q1 u − q2 v| < ǫ.
Clearly, we have
q1 u + q2 v = q1 x1 + q2 x2 , f (q1 x1 + q2 x2 ) .
Therefore, the set
G = { x, f (x) ∈ R2 | x = q1 x1 + q2 x2 and q1 , q2 ∈ Q}
• The form of all continuous functions satisfying the equation (9.85). In fact, by a
similar argument as in the proof used in [124, Problem 8.6, p. 178], it can be shown that
the function f satisfies
f (r) = rf (1)
for every r ∈ Q. Now for every x ∈ R, since Q is dense in R, we can find a sequence
{rn } ⊆ Q such that rn → x as n → ∞. Since f is continuous, we have
Hence every continuous solution of the equation (9.85) must be in the form
f (x) = f (1) · x.
9.4. Miscellaneous Problems 313
Problem 9.19
Rudin Chapter 9 Exercise 19.
so the integral on the right-hand side in the expression (9.89) can be simplified to
Z Z
(f ∗ g)(x) dm(x) = ψ(y) dm(y)
R R
Z " Z #
= g(y) f (x) dm(x) dm(y)
R R
nZ o nZ o
= f (x) dm(x) × g(y) dm(y)
R R
1
= m(A)m(B) > 0.
2π
Hence it is impossible that χA ∗ χB = 0.
Recall the definition that
Z ∞
(χA ∗ χB )(x) = χA (x − y)χB (y) dm(y). (9.90)
−∞
By the previous paragraph, there exists a x0 ∈ R such that (χA ∗ χB )(x0 ) > 0, so the expression
(9.92) gives
m B ∩ (x0 − A) > 0.
In other words, we have B ∩ (x0 − A) 6= ∅ and we can find a b ∈ B such that x0 = a + b for
some a ∈ A. Consequently, we have
x0 ∈ A + B. (9.93)
Finally, since χA ∗ χB is continuous at x0 , there exists a neighborhood I around x0 such that
(χA ∗ χB )(x) > 0 on I. Hence we conclude from the relation (9.93) that
I ⊆ A + B,
Problem 10.1
Rudin Chapter 10 Exercise 1.
Proof. In fact, this is [99, Exercise 21, p. 101]. For a solution of it, please refer to [124, p. 75].
This completes the proof of the problem.
Problem 10.2
Rudin Chapter 10 Exercise 2.
By [100, Eqn. (8), p. 199], we have n!cn = f (n) (a) for all n = 0, 1, 2, . . . and so the hypothesis
implies that f (n) (a) = 0 for some n ∈ N ∪ {0}.
For every n ∈ N ∪ {0}, let Zn = {z ∈ C | f (n) (z) = 0} ⊆ C. Now the previous paragraph
implies that
∞
[
C= Zn . (10.2)
n=0
Since f is entire, every f (n) is also entire. If f (n) 6≡ 0 for every n ∈ N ∪ {0}, then Zn 6= C for
every n ∈ N ∪ {0} by Theorem 10.18. Furthermore, each Zn is at most countable and it deduces
from the set relation (10.2) that C is countable, a contradiction. Thus there exists an N ∈ N
such that f (n) ≡ 0 for all n > N and so the representation (10.1) implies that f is a polynomial
of degree at most N , completing the proof of the problem.
Problem 10.3
Rudin Chapter 10 Exercise 3.
315
316 Chapter 10. Elementary Properties of Holomorphic Functions
Proof. We claim that f (z) = cg(z) for some |c| ≤ 1. If g 6≡ 0 in C, then we follow from Theorem
10.18 that Z(g) has no limit point in C. Consider h : C \ Z(g) → C given by
f (z)
h(z) = .
g(z)
For each a ∈ Z(g), we have h ∈ H(D ′ (a; r)) and |h(z)| ≤ 1 in D ′ (a, r) for some r > 0. Now we
follow from Theorem 10.20 that h has a removable singularity at a and then h can be defined at
a so that it is holomorphic in D(a; r). Since it is true for every a ∈ Z(g), h is in fact entire and
|h(z)| ≤ 1 in C. Therefore, Theorem 10.23 (Liouville’s Theorem) asserts that h(z) = c for some
constant c such that |c| ≤ 1. Consequently, we obtain f (z) = cg(z) as required. This proves the
claim and we end the proof of the problem.
Problem 10.4
Rudin Chapter 10 Exercise 4.
Proof. For every n = 0, 1, 2, . . ., we apply the hint given in Problem 10.2, Theorem 10.26 (The
Cauchy’s Estimates) and the hypothesis to get
f (n) (0) A + Rk
|cn | = ≤ , (10.3)
n! Rn
where R > 0. If n > k, then we take R → ∞ to both sides of the inequality (10.3) to conclude
cn = 0 for all n > k. In other words, f is a polynomial of degree at most k. This completes the
proof of the problem.
Problem 10.5
Rudin Chapter 10 Exercise 5.
Proof. Since {fn } is uniformly bounded in an open subset Ω ⊆ C, there exists a M > 0 such
that |fn (z)| ≤ M for all z ∈ Ω and all n ∈ N. Let a ∈ Ω and f (z) = lim fn (z). Since Ω is open
n→∞
in C, one can find a δ > 0 such that D(a; 4δ) ⊆ Ω. Define γ : [0, 2π] → Ω by
which is clearly a circle centered at a with radius 2δ. If z ∈ D(a; δ), then we have |γ(t) − z| > δ
for all t ∈ [0, 2π]. Therefore, we get from this fact and the representation (10.4) that
γ ′ (t) 2δ
< =2 (10.5)
γ(t) − z δ
respectively. Clearly, the pointwise convergence of {fn } on Ω implies the pointwise convergence
of {Fn } on [0, 2π]. Besides, since all fn and γ are continuous functions, every Fn is complex
10.1. Basic Properties of Holomorphic Functions 317
measurable on the measurable space [0, 2π]. Since 2M ∈ L1 (m) and |Fn (t)| ≤ M for all n ∈ N,
we deduce from Theorem 1.34 (The Lebesgue’s Dominated Convergence Theorem) that
Z 2π Z 2π
lim Fn (t) dt = F (t) dt. (10.6)
n→∞ 0 0
γ ′ (t)
Now the bound (10.5) ensures that γ(t)−z is continuous on [0, 2π] so that for all z ∈ D(a; δ), we
certainly have
γ ′ (t) γ ′ (t)
Fn (t) · → F (t) ·
γ(t) − z γ(t) − z
γ (t)′
pointwisely on [0, 2π] and each Fn (t)· γ(t)−z is complex measurable on [0, 2π]. Since 2M ∈ L1 (m)
′
γ (t)
and |Fn (t) · γ(t)−z | ≤ 2M for all n ∈ N, further application of Theorem 1.34 (The Lebesgue’s
Dominated Convergence Theorem) also gives
Z 2π Z 2π
γ ′ (t) γ ′ (t)
lim Fn (t) · dt = F (t) · dt. (10.7)
n→∞ 0 γ(t) − z 0 γ(t) − z
Given that ǫ > 0. The result (10.6) guarantees that there is an N ∈ N such that n ≥ N implies
Z 2π
|Fn (t) − F (t)| dt < ǫπ.
0
In this case, for every z ∈ D(a, δ), we obtain from the expression (10.8) that
1 Z 2π γ ′ (t) 1
Z 2π
γ ′ (t)
|fn (z) − f (z)| = Fn (t) · dt − F (t) · dt
2πi 0 γ(t) − z 2πi 0 γ(t) − z
Z 2π γ ′ (t)
1
≤ |Fn (t) − F (t)| · dt
2π 0 γ(t) − z
<ǫ
Problem 10.6
Rudin Chapter 10 Exercise 6.
• The existence of Ω. Since f (z) = 1z is continuous on D(1; 1), we follow from Theorem
10.14 (The Cauchy’s Theorem in a Convex Set) that one can find an F0 ∈ H(D(1; 1)) such
that F0′ (z) = z1 in D(1; 1). If F (z) = F0 (z) − F0 (1), then we also have F ′ (z) = F0′ (z) = z1
in D(1; 1) and F (1) = 0. Next, we define g : D(1; 1) → C by
eF (z)
g(z) = . (10.10)
z
Direct computation gives
• The injection of exp. Set Ω = F (D(1; 1)). We claim that Ω is a region such that
To this end, since F ′ (z) = z1 , Ω is not a point set. Then Ω is a region follows from the Open
Mapping Theorem directly. Finally, the expression (10.11) implies that the set equality
(10.12). Furthermore, the expression (10.11) implies immediately that exp is one-to-one
in Ω.
elog z = z. (10.13)
so we derive that an = (−1)n for all n = 0, 1, 2, . . .. To find cn , we first consider the power
series
X∞
(−1)n−1
h(z) = (z − 1)n .
n
n=1
1
Since lim sup √ = 1, the radius of convergence of the power series h is 1, so termwise
n→∞
n
n
differentiation can be performed to obtainb
∞
X ∞
X 1
h′ (z) = (−1)n−1 (z − 1)n−1 = (−1)n (z − 1)n = .
n=1 n=0
z
Since log′ (z) = z1 , we have h(z) = log z + C for some constant C. Since h(1) = log 1 = 0,
we get C = 0 and
X∞
(−1)n−1
log z = h(z) = (z − 1)n .
n
n=1
Hence we establish
(−1)n−1
, if n ≥ 1;
cn = n
0, if n = 0.
• What other discs can this be done? This can be done in every disc D(a; |a|), where
a ∈ C \ {0}. In fact, we pick b ∈ C such that eb = a. By similar argument as the proof of
the first assertion, there exists an F1 ∈ H(D(a; |a|)) such that F1′ (z) = z1 in D(a; |a|). If
we set F2 (z) = F1 (z) − F1 (a) + b, then we also have
1
F2′ (z) = F1′ (z) = and F2 (a) = b
z
in D(a; |a|). Furthermore, the function G : D(a; |a|) → C given by
eF2 (z)
G(z) = (10.14)
z
satisfies G′ (z) = 0 in D(a; |a|). Thus G is a constant in D(a; |a|) and since G(a) = 1, we
conclude that G(z) = 1 in D(a; |a|) and we get from the definition (10.14) that
eF2 (z) = z
in D(a; |a|). According to this construction, all the above assertions can be proven similarly
and we won’t repeat the argument here.
Problem 10.7
Rudin Chapter 10 Exercise 7.
b
Read [11, Theorem 2.9, pp. 28, 32] for details.
320 Chapter 10. Elementary Properties of Holomorphic Functions
Proof. The conditions should be that Γ is a cycle in the open set Ω and
0, if α ∈ / Ω;
Ind Γ (α) =
1, if α ∈ Ω \ Γ∗ .
In fact, with the above hypotheses, we get from Theorem 10.35 (Cauchy’s Theorem) that
Z
1 f (ζ)
f (z) = dζ
2πi Γ ζ − z
for all z ∈ Ω \ Γ∗ . This proves the formula for the case n = 0. Assume that
Z
k! f (ζ)
f (k)(z) = dζ (10.15)
2πi Γ (ζ − z)k+1
on Ω \ Γ∗ for some k ∈ N ∪ {0}. Next, we define g : Ω \ {z} → C by
f (ζ)
g(ζ) = ,
(ζ − z)k+1
where ζ ∈ Ω \ Γ∗ . Then it is clear that
f ′ (ζ) (k + 1)f (ζ)
g′ (ζ) = − . (10.16)
(ζ − z)k+1 (ζ − z)k+2
Since f ∈ H(Ω), we have f ′ ∈ H(Ω) and thus the formula (10.16) ensures that g ∈ H(Ω \ {z}).
Notice that Ω \ {z} is also an open set in C, so a combined application of Theorem 10.35
(Cauchy’s Theorem) and the formula (10.16) implies that
Z
g(ζ) dζ = 0
Γ
Z Z
f ′ (ζ) f (ζ)
k+1
dζ = (k + 1) k+2
dζ. (10.17)
Γ (ζ − z) Γ (ζ − z)
Applying the induction step (10.15) to the left-hand side of the formula (10.17) (with f replaced
by f ′ ), we see immediately that
Z
2πi (k+1) f (ζ)
f (z) = (k + 1) k+2
dζ
k! Γ (ζ − z)
which gives the desired result for the case k + 1. By induction, we have completed the proof of
the problem.
Problem 10.8
Rudin Chapter 10 Exercise 8.
Clearly, if A is large enough, then ΓA will contain all zeros of Q(z) lying in the upper half
plane. Hence it follows from Theorem 10.42 (The Residue Theorem) that
Z X
R(z) dz = 2πi Res (R; zk ), (10.19)
ΓA k
where {zk } is the set of all zeros of Q in the upper half plane. In fact, we can write the expression
(10.19) in the form
Z Z A X
R(z) dz + R(x) dx = 2πi Res (R; zk ).
CA −A k
Since CA is a semi-circle of radius A, its length is πA. Using this fact and deg Q − deg P ≥ 2,
we obtain from the estimate [100, Eqn. (5), Definition 10.8, p. 202] that
Z M πM
R(z) dz ≤ 2 · πA = (10.20)
CA A A
for some positive constant M . Taking A → ∞ in the inequality (10.20), we get
Z
lim R(z) dz = 0. (10.21)
A→∞ CA
Finally, we combine the results (10.18), (10.19) and (10.21) to conclude that
Z ∞ X
R(x) dx = 2πi Res (R; zk ). (10.22)
−∞ k
For the analogous statement for the lower half plane, the formula (10.22) will be replaced by
Z ∞ X
R(x) dx = −2πi Res (R; zk ),
−∞ k
where the set {zk } now consists of all zeros of Q in the lower half plane.c To compute the
integral, we note from the formula (10.22) and some basic facts of calculating residuesd that
Z ∞ h z2 πi z2 3πi i
x2
4
dx = 2πi Res ; exp + Res ; exp
−∞ 1 + x 1 + z4 4 1 + z4 4
c
Here we have the negative sign in the formula because the corresponding semi-circle in the lower half plane
is negatively oriented.
d
See, for examples, [11, pp. 129, 130] or [109, pp. 75, 76].
322 Chapter 10. Elementary Properties of Holomorphic Functions
πi h πi 3πi i
= exp − + exp −
2 4 4
πi −2i
= × √
2 2
π
=√ .
2
This completes the analysis of the problem.
Problem 10.9
Rudin Chapter 10 Exercise 9.
Proof. Let ΓA be the closed contour consisting of the real segment [−A, A] and the upper semi-
circle CA = {z ∈ C | |z| = A and Im z > 0}, see Figure 10.1. Furthermore, we let P and Q be
polynomials such that deg Q − deg P ≥ 1, Q(x) 6= 0 (except perhaps at zeros of cos x or sin x)
P (x)
and R(x) = Q(x) . By the discussion of Type II integrals in [11, pp. 144 – 146], we see that
Z ∞ Z X
R(x)eix dx = lim R(z)eiz dz = 2πi Res (R(z)eiz ; zk ), (10.23)
−∞ A→∞ ΓA
k
where the points zk are the poles of R(z) in the upper half plane.
Suppose that t ≥ 0. By the substitution y = tx, we have
Z ∞ Z ∞
eitx t
2
dx = 2 2
eiy dy.
−∞ 1 + x −∞ t + y
t
Set R(z) = t2 +z 2
. Then it follows from the representation (10.23) that
Z ∞ t
eitx π
2
dx = 2πiRes 2 2
eiz ; ti = t . (10.24)
−∞ 1+x t +z e
Z ∞
1
= e−|t| e−ixt dt
2 −∞
1 hZ 0 Z ∞ i
(1−ix)t
= e dt + e−(1+ix)t dt
2 −∞ 0
1 n exp[(1 − ix)t] 0 exp[−(1 + ix)t] ∞ o
= · +
2 1 − ix −∞ −1 + ix 0
1 1 1
= +
2 1 − ix 1 + ix
1
= .
1 + x2
Now it is clear that f, fb ∈ L1 (R), so we follow from Theorem 9.11 (The Inversion Theorem) that
g(t) = f (t), where
Z ∞ Z ∞
1 1 eixt
g(t) = √ fb(x)eixt dx = √ dx.
2π −∞ 2π −∞ 1 + x2
Consequently, we obtain
Z r
∞
eixt √ π −|t| π
2
dx = 2π · e = |t|
−∞ 1+x 2 e
which is consistent with the result (10.26). Hence we have completed the proof of the problem.
Problem 10.10
Rudin Chapter 10 Exercise 10.
Proof. Let f (z) = (ez − e−z )z −4 . Then f is holomorphic in C \ {0}. By the power series
expansion of ez (see [100, Eqn. (1), p. 1]), we have
ez − e−z 1 1 z
= 2 + + + · · ·
z4 z 3 3!z 5!
so that f has a pole of order 3 at 0. By Theorem 10.21(b), the difference
1 1
f (z) − 2 3 +
z 3!z
has a removable singularity at 0. Thus there exists an entire function g such thate
1 1
f (z) − 2 3 + = g(z)
z 3!z
which gives Z Z Z
1 1 1 1 1
f (z) dz = + dz + g(z) dz. (10.27)
2πi γ πi γ z 3 3!z 2πi γ
Applying Theorems 10.10 and 10.12 to the right-hand side of the expression (10.27), we establish
that Z z Z Z
1 e − e−z 1 1 dz 1
dz = f (z) dz = = .
2πi γ z4 2πi γ 6πi γ z 3
This ends the proof of the problem.
e
Obviously, we have
∞
X z 2n−3
g(z) = 2 .
n=2
(2n + 1)!
324 Chapter 10. Elementary Properties of Holomorphic Functions
Problem 10.11
Rudin Chapter 10 Exercise 11.
Problem 10.12
Rudin Chapter 10 Exercise 12.
Proof. Let ΓA be the path obtained by going from −A to −1 along the real axis, from −1 to
1 along the lower half of the unit circle C and from 1 to A along the real axis, see Figure 10.2
below.
We note that
Z A Z −1 Z 1 Z A
sin2 x itx sin2 x itx sin2 x itx sin2 x itx
2
e dx = 2
e dx + 2
e dx + e dx. (10.28)
−A x −A x −1 x 1 x2
Since z −2 · sin2 z · eitz is entire for every t ∈ R, it follows from Theorem 10.14 (The Cauchy’s
Theorem in a Convex Set) that
Z Z −1
sin2 z itz sin2 x itx
2
e dz + e dx = 0
C z 1 x2
10.2. Evaluation of Integrals 325
or equivalently,
Z 1 Z
sin2 x itx sin2 z itz
e dx = e dz. (10.29)
−1 x2 C z2
Combing the integral relations (10.28) and (10.29), we see immediately that
Z Z
A
sin2 x itx sin2 z itz
e dx = e dz. (10.30)
−A x2 ΓA z2
Now we define Z
eisz
ϕA (s) = dz, (10.32)
ΓA z2
so the expression (10.31) becomes
Z
sin2 z itz 1 1
e dz = − [ϕA (t + 2) + ϕA (t − 2)] + ϕA (t). (10.33)
ΓA z2 4 2
Complete ΓA to a closed path in two different ways: Firstly, we consider the semi-circle Γ1
from A to −Ai and then to −A; secondly, we consider the semi-circle Γ2 from A to Ai and then
to −A, see Figure 10.3.
It is easily checked that the function eisz · z −2 has a pole of order 2 at 0, and so the residue
is is. Thus in the first way, we have
Z Z
eisz eisz
2
dz + 2
dz = 0
ΓA z Γ1 z
so that if z = Aeiθ , where θ ∈ [−π, 0], then we deduce from the definition (10.32) that
Z Z 0
eisz exp(isAeiθ )
ϕA (s) = 2
dz = i dθ (10.35)
ΓA z −π Aeiθ
and in the second way, Theorem 10.42 (The Residue Theorem) yields
Z Z eisz
eisz eisz
2
dz + 2
dz = 2πiRes ; 0 = −2πs
ΓA z Γ2 z z2
which implies that
Z Z π
eisz exp(isAeiθ )
ϕA (s) = dz = −2πs − i dθ. (10.36)
ΓA z2 0 Aeiθ
Since exp(isAeiθ ) exp(−As sin θ)
≤ →0
Aeiθ A
as A → ∞ if s and sin θ have the same sign. Thus it follows from Theorem 1.34 (The Lebesgue’s
Dominated Convergence Theorem) that the integral (10.35) tends to 0 if s < 0 and the one in
(10.36) tends to 0 if s > 0. In other words, we obtain
−2πs, if s > 0;
lim ϕA (s) = (10.37)
A→∞
0, if s < 0.
Finally, we apply the result (10.37) to the expression (10.34) to get
Z ∞ 2
sin x itx 0,
if |t| > 2;
2
e dx = (10.38)
−∞ x π − πt , if −2 < t < 0 or 0 < t < 2.
2
When t = 0, we know from the integral (10.35) that ϕA (0) = − A2 , so we establish from the
expression (10.34) that
Z ∞ π, if t = 0;
sin2 x itx
2
e dx = (10.39)
−∞ x
0, if t = ±2.
By combining the results (10.38) and (10.39), we achieve
0, if |t| ≥ 2;
Z ∞
sin2 x itx
πt
2
e dx = π− , if −2 < t < 0 or 0 < t < 2;
−∞ x
2
π, if t = 0.
We have completed the proof of the problem.
Problem 10.13
Rudin Chapter 10 Exercise 13.
Proof. This has been solved on [123, p. 143] which completes the proof of the problem.
10.3. Composition of Holomorphic Functions and Morera’s Theorem 327
Problem 10.14
Rudin Chapter 10 Exercise 14.
Proof. Both cases can be negative. Let Ω1 = C \ {0} and Ω2 = C. Define f (z) = z in Ω1 and
z, if z 6= 0;
g(z) =
1, otherwise.
Then f (Ω1 ) ⊆ Ω2 and h(z) = g(f (z)) = g(z) = z in Ω1 . Hence f and h are holomorphic in Ω1 ,
but g is discontinuous in Ω2 .
Next, we consider Ω1 = Ω2 = C. Define
−1, if z 6= 0;
f (z) =
1, otherwise
and g(z) = z 2 . Obviously, we have f (Ω1 ) = {±1} ⊆ Ω2 and h(z) = g(f (z)) = 1 for all z ∈ Ω1 .
It is clear that both g and h are holomorphic in Ω2 and Ω1 respectively, but f is not continuous
in Ω1 . This ends the analysis of the proof.
Remark 10.1
A problem similar to Problem 10.14 but for (uniform) continuity has been discussed in [99,
Exercise 26, p. 102].
Problem 10.15
Rudin Chapter 10 Exercise 15.
Proof. According to Theorem 10.18, we have f (z) = (z − ω0 )m h(z), where h ∈ H(ϕ(Ω)) and
h(ω0 ) 6= 0. Then we have
Suppose that n ≥ 1 is the order of the zero of ϕ(z) − ϕ(z0 ). Now Theorem 10.18 implies that
that ϕ′ (z0 ) = 0, a contradiction. Consequently, n = 1 and we can write the expression (10.40)
as
g(z) = (z − z0 )m · φm (z)h(ϕ(z)). (10.42)
Finally, since φm (z0 )h(ϕ(z0 )) = φm (z0 )h(ω0 ) 6= 0, the representation (10.42) ensures that g has
a zero of order m at z0 .
328 Chapter 10. Elementary Properties of Holomorphic Functions
If ϕ′ has a zero of order k at z0 , then the expression (10.41) will imply that n = k + 1 so the
representation (10.40) becomes
In conclusion, g has a zero of order m(k + 1) at z0 . This completes the proof of the problem.
Problem 10.16
Rudin Chapter 10 Exercise 16.
Proof. Since ϕ is bounded on Ω × X, there exists a M > 0 such that |ϕ(z, t)| ≤ M for all
(z, t) ∈ Ω × X. Let z0 ∈ Ω. Since Ω is open in C, there exists a ǫ > 0 such that D(z0 ; 3ǫ) ⊆ Ω.
Then we have D(z0 ; 2ǫ) ⊆ Ω.
We claim that for every pair z, ω ∈ D(z0 ; ǫ), z 6= ω and p ∈ X, we have
ϕ(z, p) − ϕ(ω, p) 2M
≤ . (10.43)
z−ω ǫ
To this end, we consider the closed curve γ(t) = z0 + 2ǫeit , where t ∈ [0, 2π]. Obviously, since
ϕ(z, t) ∈ H(D(z0 ; 3ǫ)) for each t ∈ X, we establish from Theorem 10.15 (The Cauchy’s Formula
in a Convex Set) that if z ∈ D(z0 ; 2ǫ) ⊆ D(z0 ; 3ǫ) and p ∈ X, then z ∈/ γ ∗ and
Z Z 2π
1 ϕ(ζ, p) 1 ϕ(γ(t), p) ′
ϕ(z; p) = dζ = · γ (t) dt. (10.44)
2πi γ ζ − z 2πi 0 γ(t) − z
Note that z ∈ D(z0 ; ǫ) implies |γ(t) − z| > ǫ. Thus we follow from the formula (10.44) that
ϕ(z, p) − ϕ(ω, p) Z 2π
1 1 1
≤ · − · |ϕ(γ(t), p)| · |γ ′ (t)| dt
z−ω 2π|z − ω| 0 γ(t) − z γ(t) − ω
Z 2π
1 z−ω
≤ · · 2M ǫ dt
2π|z − ω| 0 γ(t) − z γ(t) − ω
Z 2π
Mǫ dt
≤ ·
π 0 |γ(t) − z| · |γ(t) − ω|
Z
M ǫ 2π
≤ 2 dt
ǫ π 0
2M
=
ǫ
which is exactly the inequality (10.43).
Recall that µ is a complex measure, so Theorem 6.12 tells us that there is a measurable
function h such that |h(x)| = 1 in X and dµ = h d|µ|. This fact ensures that
Z
2M 2M |µ|(X)
d|µ| = < ∞,
ǫ X ǫ
2M
i.e., ǫ ∈ L1 (|µ|). Suppose that {zn } ⊆ D(z0 ; ǫ) \ {z0 } satisfies zn → z0 . Define
ϕ(zn , x) − ϕ(z0 , x)
gn (x) = · h(x).
zn − z0
By the hypotheses, we know that each gn is measurable of x and
In other words, the sequence {gn } satisfies the conditions of Theorem 1.34 (The Lebesgue’s
Dominated Convergence Theorem), so we conclude that ϕ′ (z0 , x) · h(x) ∈ L1 (|µ|) and further-
more,
Z
f (zn ) − f (z0 ) ϕ(zn , x) − ϕ(z0 , x)
lim = lim · h(x) d|µ|
n→∞ zn − z0 n→∞ X zn − z0
Z
= lim gn (x) d|µ|
n→∞ X
Z
= ϕ′ (z0 , x) · h(x) d|µ|.
X
Consequently, f ′ (z0 ) exists. Since z0 is arbitrary, we establish that f ∈ H(Ω), completing the
proof of the problem.
Problem 10.17
Rudin Chapter 10 Exercise 17.
were X = [0, 1] and ϕ(z, t) = (1 + tz)−1 . Notice that if z0 ∈ (−∞, −1], then ϕ(z0 , − z10 ) is
unbounded. Thus it is reasonable to take Ω = C\(−∞, −1]. Now the function ϕ(z, t) satis-
fies all hypotheses of Problem 10.16 except the boundedness condition because ϕ(z, 1) → ∞
as z → −1 in Ω. Instead, it is really bounded locally. In fact, for every z ∈ Ω, there exists
a δ > 0 such that D(z; δ) ⊂ D(z; 2δ) ⊆ Ω so that ϕ is bounded on D(z; δ) × X. Hence
Problem 10.16 implies that f ∈ H(D(z; δ)). Since z is arbitrary, it yields that f ∈ H(Ω).
where X = [0, ∞) and ϕ(z, t) = etz (1 + t2 )−1 . Since |etz | = etRe z , if Re (z0 ) > 0, then
etRe (z0 )
lim |ϕ(z0 , t)| = lim = ∞.
t→∞ t→∞ 1 + t2
In other words, ϕ(z, t) is unbounded in any set containing a point of the right half plane.
Thus we may take Ω to be the left half plane which is an open set in C. We want to apply
Morera’s Theorem and Fubini’s Theorem.f
Let z0 , zn ∈ Ω for all n ∈ N, where zn → z0 as n → ∞. Fix t ∈ [0, ∞), then it is easy to
see that ϕ(z, t) is continuous at z0 . In addition, we know that
1
|ϕ(z, t)| ≤ (10.45)
1 + t2
f
Problem 10.16 cannot be applied directly in this case because m(X) = ∞, i.e., m is not a complex measure
on X.
330 Chapter 10. Elementary Properties of Holomorphic Functions
1 1
for all z ∈ Ω and 1+t 2 ∈ L (m). Then we deduce from Theorem 1.34 (The Lebesgue’s
Dominated Convergence Theorem) that
Z Z
lim g(zn ) = lim ϕ(zn , t) dm = ϕ(z0 , t) dm = g(z0 ),
n→∞ n→∞ X X
Define φ(x, t) = ϕ(γ(x), t) · γ ′ (x) on [a, b] × [0, ∞). Clearly, both [a, b] and [0, ∞) are σ-
finite measure spaces. Since γ is piecewise continuously differentiable on [a, b], there exists
a M > 0 such that
|γ ′ (x)| ≤ M
on [a, b]. Furthermore, φ(x, t) is piecewise continuous on [a, b] × [0, ∞) so that φ is a
measurable function on [a, b] × [0, ∞). Finally, for each x ∈ [a, b], we know from the
inequality (10.45) that
M
|φ|x = |ϕ(γ(x), t) · γ ′ (x)| ≤
1 + t2
which gives Z Z
∞ ∞
dt Mπ
φ∗ (x) = |φ|x dt ≤ M 2
=
0 0 1+t 2
and Z b
φ∗ dx < ∞.
a
Consequently, we may apply Theorem 8.8 (The Fubini Theorem) to change the order of
integration in the integral (10.46) and get
Z Z ∞hZ b i Z ∞Z
g(z) dz = ϕ(γ(x), t)γ ′ (x) dx dt = ϕ(z, t) dz dt. (10.47)
∂∆ 0 a 0 ∂∆
Since ϕ(z, t) is holomorphic in Ω for every t ∈ [0, ∞), we conclude from Theorem 10.13
(The Cauchy’s Theorem for a Triangle) that
Z
ϕ(z, t) dz = 0,
∂∆
Finally, we apply Theorem 10.17 (Morera’s Theorem) to obtain the desired conclusion that
g ∈ H(Ω).
where X = [−1, 1] and ϕ(z, t) = etz (1 + t2 )−1 . For every z ∈ C, we have z ∈ D(z; 1) ⊂ C.
We know that
|ϕ(z, t)| ≤ e|Re z| < ∞
in D(z; 1) × X. Thus the function ϕ satisfies all the requirements of Problem 10.16 in
D(z; 1) × X, so h ∈ H(D(z; 1)). Since z is arbitrary, we conclude that h ∈ H(C), i.e., h is
entire.
Problem 10.18
Rudin Chapter 10 Exercise 18.
where the sum is taken over all the zeros of f inside γ and mk is the multiplicity of the zero zk
of f . This is the answer of the first assertion.
′(z)
For the second assertion, let F (z) = ff (z) ϕ(z). Then F is a meromorphic function in Ω. Let
A ⊆ Ω be the set of poles of F . Since f 6= 0 on γ ∗ , γ is a cycle in Ω \ A and Ind γ (α) = 0 for
all α ∈ Ω. In other words, our F satisfies Theorem 10.42 (The Residue Theorem) which implies
that Z ′
1 f (z) X
ϕ(z) dz = Res (F ; zk ), (10.48)
2πi γ f (z)
k
where the zk denote the isolated singularities of F inside γ which are exactly the zeros of f
inside γ. By Theorem 10.18, there exists a g ∈ H(Ω) and a unique positive integer mk such that
f (z) = (z − zk )mk g(z) and g(zk ) 6= 0. Clearly, we have
f ′ (z) mk ϕ(z) g′ (z)
F (z) = ϕ(z) = + ϕ(z).
f (z) z − zk g(z)
If ϕ(zk ) = 0, then F is holomorphic at zk . Otherwise, F has a simple pole at zk and it yields
from [11, Eqn. (1), p. 129] that
where the sum is taken over all the zeros of f which are not zeros of ϕ inside γ. This completes
the proof of the problem.
Problem 10.19
Rudin Chapter 10 Exercise 19.
332 Chapter 10. Elementary Properties of Holomorphic Functions
Proof. We claim that f = cg for some nonzero constant c. To this end, we consider h = fg in U .
Since f (z) 6= 0 on U , we know that h(z) 6= 0 on U . In addition, since f, g ∈ H(U ) and g(z) 6= 0
on U , we conclude that h ∈ H(U ). Direct differentiation gives
f ′ (z)g(z) − f (z)g ′ (z) f (z) h f ′ (z) g′ (z) i
h′ (z) = = · −
g 2 (z) g(z) f (z) g(z)
so that h′ ( n1 ) = 0 on {1, 2, . . .}. By Theorem 10.18, we conclude that h′ (z) = 0 in U . Next, the
Fundamental Theorem of Calculus [11, Proposition 4.12, p. 51] shows that
Z
h(z) = h(0) + h′ (ζ) dζ = h(0) 6= 0
[0,z]
for every z ∈ U , where [0, z] is a path connecting 0 and z in U . By the definition, we establish
that
f (z) = h(0)g(z)
on U . This ends the proof of the problem.
Problem 10.20
Rudin Chapter 10 Exercise 20.
Proof. Suppose that f (z0 ) = 0 for some z0 ∈ Ω. Assume that f 6≡ 0. Then there exists a circle
C(z0 ; R) for some R > 0 such that f (z) 6= 0 on C(z0 ; R). By Theorem 10.28, we know that
fn′ → f ′ uniformly on compact subsets of Ω. Thus the convergence
fn′ f′
→
fn f
is also uniform on C(z0 ; R). Next, Theorem 10.43(a) gives
Z Z
1 f ′ (z) 1 fn′ (z)
Nf = dz and Nfn = dz.
2πi C(z0 ;R) f (z) 2πi C(z0 ;R) fn (z)
Our hypotheses give Nf = 1 but Nfn = 0 which is a contradiction, so we establish the result
f ≡ 0 on Ω.
For the second assertion, suppose that
∞
[
fn (Ω) ⊆ Ω′ .
n=1
Choose a point a ∈ C \ Ω′ .
Define Fn : Ω → C by Fn = fn − a for every n = 1, 2, . . .. Then each
Fn is holomorphic in Ω, Fn (z) 6= 0 for all z ∈ Ω and Fn → f − a uniformly on compact subsets
of Ω. Thus the first assertion and the fact that f is nonconstant imply that f (z) 6= a in Ω. In
other words, we must have
f (Ω) ⊆ Ω′
which completes the proof of the problem.
Remark 10.2
We note that Problem 10.20 is classically called Hurwitz’s Theorem, see [11, Theorem
10.13, p. 139] or [30, p. 152].
10.4. Problems related to Zeros of Holomorphic Functions 333
Problem 10.21
Rudin Chapter 10 Exercise 21.
Proof. Let g(z) = f (z) − z and h(z) = −z on Ω. Since D(0; 1) ⊆ Ω and |f (z)| < 1 on |z| = 1,
we have
|g(z) − h(z)| = |f (z)| < 1 = |h(z)|
on |z| = 1. By Theorem 10.43(b) (Rouché’s Theorem), we conclude that Ng = Nh = 1 in the
disc D(0; 1). This completes the analysis of the proof.
Problem 10.22
Rudin Chapter 10 Exercise 22.
Proof. Assume that f (z) 6= 0 for all z ∈ Ω. By the Corollary to Theorem 10.24 (The Maximum
Modulus Theorem), we see that
a contradiction. Hence f has at least one zero in the unit disc, completing the proof of the
problem.
Problem 10.23
Rudin Chapter 10 Exercise 23.
• By Theorem 10.25 (The Fundamental Theorem of Algebra), both Pn and Qn have precisely
n zeros in C.
• The definitions of Pn and Qn guarantee that Pn and Qn cannot have any common zeros.
• By [11, Exercise 12, p. 142], we know that for every R > 0, if n is large enough, Pn (z) has
no zeros in |z| ≤ R, i.e., all zeros ζ of Pn satisfy |ζ| > R for large n. In fact, it has been
shown further in [56] that every zero ζ of Pn lie in the annulus
n
< |ζ| < n
e2
for large enough n.
z2 zn z z n−1
Qn (z) = z + + ··· + = z 1 + + ··· +
2! n! 2! n!
lie inside the annulus
• We imitate the proof of [123, Problem 10.12, pp. 131, 132]. First, we notice that ez − 1 = 0
if and only if z = ±2kπi for all k ∈ N, so if we take 2kπ < Rk < 2(k + 1)π, then there is a
ǫk > 0 such that
|ez − 1| > ǫk . (10.50)
Since Pn (z) → ez uniformly in D(0; Rk ), there exists a Mk ∈ N such that n ≥ Mk implies
for all n ≥ Mk and for all z ∈ C(0; Rk ). By Theorem 10.43(b) (Rouché’s Theorem), we
conclude that
NQn = Nez −1 = 2k + 1
inside C(0; Rk ) for all n ≥ Mk . This also means that
NQn = n − 2k − 1 (10.52)
NQn = n − 2k − 3 (10.53)
outside C(0; Rk + 2π) for all n ≥ Mk+1 . Now if we combine the results (10.52) and (10.53),
there exists a Mk′ ∈ N such that
NQ n = 2
in the annulus A = {z ∈ C | Rk < |z| < Rk + 2π} for all large enough n ≥ Mk′ , where
k = 1, 2, 3, . . ..
Remark 10.3
There are some books concerning the location of the zeros of polynomials. For instances,
[20, Chap. 1], [69] and [72, Chap. 3].
Problem 10.24
Rudin Chapter 10 Exercise 24.
on ∂Ω = K \ Ω and this ensures that f and g cannot have any zero on ∂Ω. In addition,
the continuity of f and g imply the continuity of ϕ and then z0 ∈ E. Thus E is a closed,
hence compact, subset of K by Theorem 2.4. By the inequality (10.54), we know that
E ∩ ∂Ω = ∅ which means that E ⊆ Ω and
in K \ E.
Assume that Z(f ) was infinite. Then [127, Problem 5.25, p. 68] leads to us that Z(f ) has
a convergent subsequence and Theorem 10.18 says that Z(f ) = Ω, but the continuity of
f implies immediately that f ≡ 0 on K which is impossible. Consequently, Z(f ) is finite.
Furthermore, since f (a) = 0 implies ϕ(a) = 0, all zeros of f lie in E. Similarly, Z(g) is
also finite and all zeros of g belong to E.
• Step 3: A lemma and its application. Here we need the following result whose proof
can be found in [103, IX.8 & 9, pp. 115 – 118] or [111, Lemma 5.8, pp. 61, 62]:
Lemma 10.1
Let G be an open subset of C and K a compact subset of G. Then there exists a
cycle Γ in G \ K such that K ⊆ int Γ ⊆ G and
Z
1 f (ζ)
f (z) = dζ (10.56)
2πi Γ ζ − z
By Step 1, we see that E and Ω satisfy the roles of K and G of Lemma 10.1 respectively,
so it ensures that there exists a cycle Γ in Ω \ E such that E ⊆ int Γ ⊆ Ω and the formula
(10.56) holds for every f ∈ H(Ω) and every z ∈ E.
As an application, let Fe ∈ H(Ω), z ∈ E and F (ζ) = (ζ − z)Fe(ζ). Obviously, we have
F ∈ H(Ω) and F (z) = 0, so the formula (10.56) gives
Z
Fe(ζ) dζ = 0. (10.57)
Γ
• Step 4: The calculation of |Z(f )| and |Z(g)|. Recall from the hypotheses and Step
2 that f ∈ H(Ω) and f has only finitely many zeros a1 , a2 , . . . , aN with multiplicities
p1 , p2 , . . . , pN respectively in E. Then the function
N
f ′ (z) X pk
fe(z) = −
f (z) z − ak
k=1
336 Chapter 10. Elementary Properties of Holomorphic Functions
can be shown to have a removable singularity at each ak by Theorem 10.18 and Definition
10.19. Thus fe is holomorphic in Ω. Hence it follows from the result (10.57) that
Z X N Z
f ′ (ζ) dζ
dζ = pk . (10.58)
Γ f (ζ) Γ ζ − ak
k=1
Similarly, we obtain
Z
1 g′ (ζ)
dζ = |Z(g)|. (10.61)
2πi Γ g(ζ)
f (z)
h(z) = log
g(z)
f ′ g′
h′ = −
f g
or equivalently,
Z Z
f ′ (ζ) g′ (ζ)
dζ = dζ. (10.62)
Γ f (ζ) Γ g(ζ)
Finally, by substituting the results (10.60) and (10.61) into the expression (10.62), we have
established that
|Z(f )| = |Z(g)|.
Problem 10.25
Rudin Chapter 10 Exercise 25.
10.5. Laurent Series and its Applications 337
Proof. Define A(r1 , r2 ) = {z ∈ C | r1 < |z| < r2 }. Let ǫ > 0 be such that r1 + ǫ < r2 − ǫ.
Furthermore, we define ρr : [0, 2π] → C by
ρr (t) = reit ,
(a) Note that γ1 and γ2 are negatively oriented and positively oriented circles respectively. By
Theorem 10.11, we have
−1, if z ∈ D(0; r1 + ǫ);
Ind γ1 (z) =
0, if z ∈
/ D(0; r1 + ǫ)
and
1, if z ∈ D(0; r2 − ǫ);
Ind γ2 (z) =
0, if z ∈
/ D(0; r2 − ǫ).
Let Γ = γ1 + γ2 which is the sum of two circles in A(r1 , r2 ). Let α ∈
/ A(r1 , r2 ). If |α| ≤ r1 ,
then we have α ∈ D(0; r1 + ǫ) ⊆ D(0; r2 − ǫ). Thus we follow from [100, Eqn. (8), p. 218]
that
Ind Γ (α) = Ind γ1 (α) + Ind γ2 (α) = 0. (10.63)
Similarly, if |α| ≥ r2 , then we still have the result (10.63). Next if z ∈ A(r1 + ǫ, r2 − ǫ),
then z ∈
/ D(0; r1 + ǫ) and z ∈ D(0; r2 − ǫ). Thus it is easy to check that
Hence, Theorem 10.35 (Cauchy’s Theorem) and [100, Eqn. (5), p. 217] assert that
338 Chapter 10. Elementary Properties of Holomorphic Functions
Z Z Z
1 f (ζ) 1 f (ζ)
f (z) = dζ = + dζ (10.64)
2πi Γ ζ −z 2πi γ1 γ2 ζ −z
for every z ∈ A(r1 + ǫ, r2 − ǫ).
(b) Let R1 = r1 + ǫ and R2 = r2 − ǫ so that r1 < R1 < |z| < R2 < r2 . Define
Z Z
1 f (ζ) 1 f (ζ)
f1 (z) = dζ and f2 (z) = dζ. (10.65)
2πi −ρR ζ − z 2πi ρR ζ − z
1 2
– Step 1: f1 and f2 are well-defined. To see this, we first recall from the hypothesis
that f ∈ H(A(r1 , r2 )). Next, we fix z and take r1 < R1 < R1′ < |z| which means that
z∈/ D(0; R1′ ). Let Γ1 = ρR1 − ρR′1 . By Theorem 10.11, we know that
for every α ∈ / A(r1 , r2 ). Thus it yields from Theorem 10.35 (Cauchy’s Theorem)g
that Ind Γ1 (z) = 0 − 0 = 0 and then
Z Z
1 f (ζ) 1 f (ζ)
dζ = dζ.
2πi ρR ζ − z 2πi ρR′ ζ − z
1 1
for every α ∈
/ A(r1 , r2 ). Hence we have Ind Γ2 (z) = 1 − 1 = 0 and similar argument
shows that f2 (z) is well-defined in D(0; r2 ).
– Step 2: f2 ∈ H(D(0; r2 )). As suggested by the proof of Theorem 10.16, we may
apply Theorem 10.7 to the integral representation (10.65) of f2 with X = [0, 2π],
ϕ = ρR2 and dµ(t) = f (ρR2 (t))ρ′R2 (t) dt to establish the fact that f2 is representable
by a power series in D(0; R2 ). Since ǫ is arbitrary, f2 is representable by a power
series in D(0; r2 ) and hence Theorem 10.6 concludes that f2 ∈ H(D(0; r2 )).
– Step 3: f1 ∈ H(C \ D(0; r1 )). For f1 , we consider the function g(ζ) = ζ1 f ( 1ζ ) on
|ζ| = R11 . Tale X = [0, 2π], ϕ = ρ 1 and dµ(t) = g ρ 1 (t) ρ′ 1 (t) dt in Theorem 10.7.
R1 R1 R1
Obviously, we have ϕ([0, 2π]) = ρ 1 ([0, 2π]) = C(0; R11 ), so ϕ([0, 2π]) ∩ D(0; R11 ) = ∅
R1
and thus the function Z
1 dµ(ζ)
g(z) =
2πi ρ 1
ζ−z
R1
g
In fact, we have applied the formula [100, Eqn. (2), p. 219] here.
10.5. Laurent Series and its Applications 339
Z 2π
1 R1 e−it dt
= if (R1 e−it ) ·
1 it
2πi 0 ( R1 e − z)R1 e−it
Z 2π
1 1 if (R1 e−it )R1 e−it dt
= · 1 −it
z 2πi 0 z − R1 e
Z
1 f (ζ)
= dζ
z −ρ 1 ζ − z1
R1
1 1
= f1 .
z z
Therefore, the function 1z f1 ( z1 ) is representable by power series in D(0; R11 ). Again,
since ǫ is arbitrary, z1 f1 ( 1z ) can be represented by power series in D(0; r11 ) and then
Theorem 10.6 ensures that z1 f1 ( 1z ) ∈ H(D(0; r11 )). Hence, this certainly shows that
f1 ∈ H(C \ D(0; r1 )) as required.
– Step 4: Uniqueness of the decomposition. Suppose that f = g1 + g2 , where
g1 ∈ H(C \ D(0; r1 )) and g2 ∈ H(D(0; r2 )). Then we have
g1 − f1 = f2 − g2 (10.66)
in A(r1 , r2 ). Define h : C → C by
f2 (z) − g2 (z), if z ∈ D(0; r2 );
h(z) =
g1 (z) − f1 (z), if z ∈ C \ D(0; r1 ).
Now the equation (10.66) shows that the two definitions of h actually agree in
A(r1 , r2 ), so h is well-defined in C and in fact it is entire. Thus it suffices to prove
that h ≡ 0. Since f1 (z), g1 (z) → 0 as |z| → ∞, we have h(z) → 0 as |z| → ∞ and thus
h is bounded. By Theorem 10.23 (Liouville’s Theorem), we conclude that h(z) = 0
for all z ∈ C and this means that f1 (z) = g1 (z) and f2 (z) = g2 (z) in C \ D(0; r1 ) and
D(0; r2 ) respectively. This proves the uniqueness of the decomposition.
1 1 1 z z2
= z = + 2 + 2 + ··· . (10.67)
ζ −z ζ(1 − ζ ) ζ ζ ζ
1 1 1 ζ ζ2
=− ζ
= − − 2 − 3 − ··· . (10.68)
ζ−z z(1 − z ) z z z
Substituting the series (10.67) and (10.68) into the formula (10.64), we get
Z ∞
X Z −∞
X
1 f (ζ)z n 1 f (ζ)z n
f (z) = dζ + dζ. (10.69)
2πi γ2 n=0 ζ n+1 2πi −γ1 n=−1 ζ n+1
Since the convergence of the series (10.67) and (10.68) are uniform, we can switch
the order of integration and summation in the expression (10.69) to obtain
∞
X
f (z) = cn z n ,
−∞
340 Chapter 10. Elementary Properties of Holomorphic Functions
where
Z γ2 , if n = 0, 1, 2, . . .;
1 f (ζ)
cn = dζ and γ = (10.70)
2πi γ ζ n+1
−γ1 , if n = −1, −2, −3, . . ..
∞
X
– Uniqueness of the Laurent series. We claim that if an z n converges to f in
−∞
∞
X
A, then an = cn for all n ∈ Z. To this end, since an z n converges uniformly to f
−∞
on γ, we have
Z X ∞ Z
f (z)
k+1
dz = an z n−k−1 dz = 2πiak ,
γ z −∞ γ
where k ∈ Z. Hence it asserts from the definitions (10.70) that cn = an hold for all
n ∈ Z. This proves the claim and then the uniqueness follows.
– Uniform convergence on compact subsets of A. Let K be a compact subset of
A. It is easy to check that C \ A is closed in C and (C \ A) ∩ K = ∅. Thus we deduce
from Problem 10.1 that there is a δ > 0 such that d(K, C \ A) = 2δ > 0. In addition,
we have
K ∩ C(0; r1 + δ) = ∅ and K ∩ C(0; r2 − δ) = ∅
which imply that K ⊆ A(r1 + δ, r2 − δ) ⊆ A(r1 , r2 ). For r1 + δ ≤ |z| ≤ r2 − δ, we
know that
∞
X −1
X ∞
X
|cn z n | ≤ |cn z n | + |cn z n |
−∞ n=−∞ n=0
X∞ ∞
|cn | X
≤ + |cn |(r2 − δ)n
|z|n
n=1 n=0
X∞ X ∞
|cn |
≤ + |cn |(r2 − δ)n
n=1
(r1 + δ)n n=0
< ∞.
Hence it follows from the Weierstrass M -Test [11, Theorem 1.9, p. 15] or [99, Theorem
7.10, p. 148] that the series converges to f uniformly on K.
(d) Let r1 < s1 < s2 < r2 . Firstly, since f2 ∈ H(D(0; r2 )) and D(0; s2 ) is a compact subset of
D(0; r2 ), f2 is bounded in D(0; s2 ). Secondly, since f1 (z) → 0 as |z| → ∞, there exists a
M > s1 such that
|f1 (z)| < 1 (10.71)
on C \ D(0; M ). The set D(0; M ) \ D(0; s1 ) = {z ∈ C | s1 ≤ |z| ≤ M } is a closed and
bounded subset in C, so it is compact. Since f1 ∈ H(C \ D(0; r1 )), f1 is bounded in
D(0; M ) \ D(0; s1 ). This fact and the bound (10.71) combine to say that f1 is bounded in
C \ D(0; s1 ).
Now we write
f2 = f − f1
on D(0; r2 ) \ D(0; s2 ). Since D(0; r2 ) \ D(0; s2 ) ⊆ C \ D(0; s1 ), f1 is bounded there.
Since f is bounded in A, f2 is bounded in D(0; r2 ) \ D(0; s2 ). Using the first result in
the previous paragraph, we conclude that f2 is bounded in D(0; r2 ). Similarly, we write
10.5. Laurent Series and its Applications 341
Problem 10.26
Rudin Chapter 10 Exercise 26.
Proof. Let f be the function in the question. The function f is holomorphic in C \ {−1, 1, 3}.
By Problem 10.25, we have to consider the following three regions
D(0; 1) = {z ∈ C | |z| < 1}, A = {z ∈ C | 1 < |z| < 3} and B = {z ∈ C | |z| > 3}.
Therefore, for every z ∈ D(0; 1), we follow from the series (10.72) and (10.74) that
1 X z n X h 1 + (−1)n 1 i n
∞
X ∞ ∞
f (z) = z 2n + = + z .
n=0
3 n=0 3n n=0 2 3n+1
Thus we have
1 + (−1)n 1
+ n+1 , if n = 0, 1, 2, . . .;
cn = 2 3
0, otherwise.
342 Chapter 10. Elementary Properties of Holomorphic Functions
Next, if z ∈ A, then we deduce from the series (10.73) and (10.74) that
∞ ∞ ∞
X 1 1 X zn X
f (z) = − 2n
+ n
= cn z n ,
n=1
z 3 n=0
3 n=−∞
where
1
3n+1 ,
if n ≥ 0;
cn =
n+1
−1 + (−1)
, otherwise.
2
Finally, if z ∈ B, then the series (10.73) and (10.75) imply that
∞
X ∞ ∞
1 1 X 3n X
f (z) = − − = cn z n ,
z 2n z z n n=−∞
n=1 n=0
where
0, if n ≥ 0;
cn = h n i
− 1 + (−1) + 3|n|−1 , otherwise.
2
Hence we have completed the analysis of the problem.
Problem 10.27
Rudin Chapter 10 Exercise 27.
arg ζ + 2π i arg ζ i
− log |ζ| = +1− log |ζ| = z + 1.
2π 2π 2π 2π
Since f is a function of period 1, we have F (ζ) = f (z) = f (z + 1) = F (ζe2πi ) which proves our
claim. Denote the annulus A = {ζ ∈ C | e−2πb < |ζ| < e−2πa }. Then the mapping G(z) = e2πiz
clearly maps the horizontal strip Ω = {z ∈ C | a < Im z < b} onto A. Furthermore, G is
holomorphic in Ω and f = F ◦ G. Here we need a positive result to Problem 10.14:
Lemma 10.2
Suppose that Ω1 and Ω2 are two regions, f and g are nonconstant complex functions
defined in Ω1 and Ω2 respectively. Put h = g ◦ f . If f and h are holomorphic in
Ω1 and f (Ω1 ) = Ω2 , then g is also holomorphic in Ω2 .
10.5. Laurent Series and its Applications 343
f ′ (z) 6= 0 (10.78)
Since f and G are holomorphic in Ω and G(Ω) = A, Lemma 10.2 ensures that F is analytic
in A. By Problem 10.25(c), F admits the Laurent series
∞
X
F (ζ) = cn ζ n (10.80)
−∞
Finally, we denote
Then we have G(Ω′ ) = A′ and A′ is a compact subset of A. By Problem 10.25(c), the series
(10.80) converges uniformly in A′ and hence the corresponding series (10.81) converges uniformly
in Ω′ . This completes the proof of the problem.
344 Chapter 10. Elementary Properties of Holomorphic Functions
Problem 10.28
Rudin Chapter 10 Exercise 28.
Proof. If we consider γ = Γ − α, then we observe immediately from [100, Eqn. (2), p. 203] that
Z 2π Z 2π
1 Γ′ (s) 1 γ ′ (s)
Ind Γ (α) = ds = ds = Ind γ (0).
2πi 0 Γ(s) − α 2πi 0 γ(s)
Thus, without loss of generality, we may assume that α = 0 in the following discussion. Now we
prove the assertions one by one.
• Ind Γn (0) = Ind Γm (0) if m and n are sufficiently large. Since Γ : [0, 2π] → C is a closed
curve, it is a continuous function with period 2π. Thus the Stone-Weierstrass Theorem [99,
Theorem 8.15, p. 190] asserts that there exists a sequence of trigonometric polynomials
{Γn } converges to Γ uniformly in [0, 2π]. In other words, choose |Γ(θ)| > δ > 0, there
exists an N ∈ N such that n ≥ N implies
δ
|Γ(θ) − Γn (θ)| <
4
for all θ ∈ [0, 2π]. According to [99, Exercise 26, p. 202]h , we conclude that
for all m, n ≥ N .
• The result is independent of the choice of {Γn }. Note that [124, Problem 8.26, pp.
204 – 206] also includes this part.
• Lemma 10.39 is true for closed curves. Again, [124, Problem 8.26, pp. 204 – 206]
contains this part.
• Another proof of Theorem 10.40. By the definition, there exists a continuous map
H : I 2 → Ω such that
Put Γt (s) = H(s, t). According to [100, Eqn. (3) & (4), p. 223], we obtain
for all (s, t) ∈ I 2 and there exists a positive integer n such that
Ind Γ0 (0) = Ind Γt0 (0) = Ind Γt1 (0) = · · · = Ind Γtn (0) = Ind Γ1 (0).
Problem 10.29
Rudin Chapter 10 Exercise 29.
where γ(θ) = e−iθ with −π ≤ θ ≤ π which is the negatively oriented circle with center at 0 and
radius 1. By Theorem 10.11, we know that
r
−r −1, if |z| < 1;
Ind γ =
z
0, r
if |z| > 1.
hZ |z|
r −r Z 1
r −r i
f (z) = −2 · Ind γ dr + · Ind γ dr
0 z z |z| z z
Z |z|
r
=2 dr
0 z
Z
2 |z|
= r dr
z 0
|z|2
=
z
= z.
346 Chapter 10. Elementary Properties of Holomorphic Functions
Problem 10.30
Rudin Chapter 10 Exercise 30.
Proof. Without loss of generality, we may assume that Ω = C \ {−1, 1} which is clearly open in
C. Consider the boundaries γ1 , γ2 , γ3 and γ4 of the discs D(−1; 1), D(1; 1), D(−2; 2) and D(2; 2)
respectively. Each starts and ends at 0 with the orientation as shown in Figure 10.5. Then
Ind Γ (±1) = 0.
Assume that Γ was null-homotopic to the constant map 0 in Ω. By §10.38, there exists a
continuous map Γt : [0, 1] → Ω connecting Γ and 0 such that Γ0 = 0 and Γ1 = Γ. However,
the continuity of Γt forces that it must pass through one of the omitted points 1 and −1, a
contradiction. This completes the proof of the problem.
CHAPTER 11
Harmonic Functions
Problem 11.1
Rudin Chapter 11 Exercise 1.
If u + icv ∈ H(Ω), then it yields from the Cauchy-Riemann equations that ux = cvy and
uy = −cvx so that ux vx + uy vy = cvy vx − cvx vy = 0 in Ω. This means that ∆(uv) = 0.
Conversely, suppose that uv is harmonic in Ω. If u is constant, then Theorem 11.2 ensures
that ux = uy = 0 in Ω which gives the equation (11.1). The case for v being constant
is similar. Therefore, without loss of generality, we may assume that both u and v are
nonconstant. Clearly, the equation (11.1) implies that there exists a function c : R2 → R
such that (ux , uy ) = c(x, y)(vy , −vx ) in Ω, or equivalently,
cx vy − cy vx = 0 (11.3)
347
348 Chapter 11. Harmonic Functions
and thus
cy vy − cx vx = 0. (11.4)
Eliminating vx and vy from the equations (11.3) and (11.4), we get
in Ω \ Z(f ). Since f = vx − ivy ∈ H(Ω), f has continuous derivatives of all orders. Thus
both vx and vy have continuous partial derivatives of all orders in Ω. Similarly, both ux
and uy have continuous partial derivatives of all orders in Ω. Therefore, it follows from any
one of the equations (11.2) that both cx and cy are continuous on Ω. Hence the equation
(11.6) holds in Ω so that cx = cy = 0 in Ω. In conclusion, c ∈ R and hence u + icv satisfies
the Cauchy-Riemann equations.
• |f |2 is harmonic. Since f ∈ H(Ω), we know from Theorem 11.4 that both u and v are
harmonic in Ω. Note that |f |2 is harmonic in Ω if and only if u2 + v 2 is harmonic in Ω.
Clearly, we have
Problem 11.2
Rudin Chapter 11 Exercise 2.
Proof. Let f = u + iv. The result is obvious if f is constant, so without loss of generality, we
may assume that f is nonconstant. Since f is harmonic in Ω, both u and v are harmonic in Ω.
Similarly, since f 2 = u2 − v 2 + 2iuv is harmonic in Ω, both u2 − v 2 and uv are harmonic in Ω.
Thus ∆(u2 − v 2 ) = 0 and ∆(uv) = 0 imply that
holds in Ω.
Next, the equation (ux − ivx )2 + (uy − ivy )2 = 0 gives
in Ω. We denote
It is obvious that
and
(ux − vy )y = uxy − vyy = uyx + vxx = −[−(uy + vx )]x
hold in Ω. In other words, g ∈ H(Ω). Similarly, we have h ∈ H(Ω).
Assume that g 6≡ 0 and h 6≡ 0 in Ω. Then Theorem 10.18 ensures that Z(g) and Z(h) are
at most countable. Therefore, Z(gh) = Z(g)Z(h) is also at most countable. Note that zeros
of h and h are identical. Therefore, the equation (11.7) says that Z(gh) = Ω, a contradiction.
Hence we have g ≡ 0 or h ≡ 0 in Ω. If g ≡ h ≡ 0 in Ω, then we deduce from the definition
(11.8) that ux = uy = 0 in Ω, i.e., u is constant in Ω. By this and the fact f ∈ H(Ω), we obtain
immediately from Theorem 11.2 that v is also constant in Ω. Hence f is also constant in H(Ω),
a contradiction. In other words, we have either g ≡ 0 or h ≡ 0 in Ω.
• Case (ii): h ≡ 0 in Ω. Similarly, this shows that ux = −vy and uy = vx in Ω and these
mean that f = u − iv ∈ H(Ω).
Problem 11.3
Rudin Chapter 11 Exercise 3.
Problem 11.4
Rudin Chapter 11 Exercise 4.
• Pr (θ − t) is a harmonic function of reiθ for a fixed t. It can be shown easily that the
Laplacian equation in polar form is given by
1 1
∆u = urr + ur + 2 uθθ . (11.9)
r r
Fix t. Write P = Pr (θ − t) and A = A(r, θ) = 1 − 2r cos(θ − t) + r 2 for convenience. Then
[100, Eqn. (2), §11.5, p. 233] can be written as P A = 1 − r 2 . Direct differentiation gives
and
Z Z Z
∂ ∂
uθ = Pr (θ − t) dµ(eit ) = Pr (θ − t) dµ(eit ) = [Pr (θ − t)]θ dµ(eit ) (11.12)
∂θ T T ∂θ T
11.1. Basic Properties of Harmonic Functions 351
so that Z
∆u = ∆Pr (θ − t) dµ(eit ) = 0.
T
In other words, u = P [ dµ] is harmonic in U .
Fix r ∈ [0, 1). For very small h > 0 such that 1 − r − 2h > 0, we note that
Z
u(r + h, θ) − u(r, θ) Pr+h (θ − t) − Pr (θ − t)
= dµ(eit )
h T h
Z
Pr+h (θ − t) − Pr (θ − t)
ur = lim dµ(eit ). (11.13)
h→0 T h
We observe that
Pr+h (θ − t) − Pr (θ − t) ∂
lim = Pr (θ − t)
h→0 h ∂r
and
Pr+h (θ − t) − Pr (θ − t)
h
1 h 1 − (r + h)2 1 − r2 i
= · 2
− 2
h 1 − 2(r + h) cos(θ − t) + (r + h) 1 − 2r cos(θ − t) + r
[1 − (r + h)2 ][1 − 2r cos(θ − t) + r 2 ] − (1 − r 2 )[1 − 2(r + h) cos(θ − t) + (r + h)2 ]
=
h[1 − 2(r + h) cos(θ − t) + (r + h)2 ] · [1 − 2r cos(θ − t) + r 2 ]
2h cos(θ − t) + 2r 2 − 2(r + h)2 + 2hr(r + h) cos(θ − t)
=
h[1 − 2(r + h) cos(θ − t) + (r + h)2 ][1 − 2r cos(θ − t) + r 2 ]
2h cos(θ − t) − 2h(2r + h) + 2hr(r + h) cos(θ − t)
=
h[1 − 2(r + h) cos(θ − t) + (r + h)2 ] · [1 − 2r cos(θ − t) + r 2 ]
2 cos(θ − t) − 2(2r + h) + 2r(r + h) cos(θ − t)
= . (11.14)
[1 − 2(r + h) cos(θ − t) + (r + h)2 ] · [1 − 2r cos(θ − t) + r 2 ]
1−r
Since 1 − r − 2h > 0 implies 1 − (r + h) > 2 , this and the expression (11.14) give
P (θ − t) − P (θ − t) 2 + 2(2 + h) + 2(1 + h) 4(8 + 4h) 40
r+h r
≤ ≤ ≤
h (1 − r)2 [1 − (r + h)]2 (1 − r)4 (1 − r)4
4 3
≤ ·
(1 − r)3 2
6
= .
(1 − r)3
6 1
Again, the finiteness of µ implies that (1−r)3 ∈ L (T ) and we can apply Theorem 1.34
(The Lebesgue’s Dominated Convergence Theorem) to conclude that the formula (11.12)
holds.
Problem 11.5
Rudin Chapter 11 Exercise 5.
Proof. Let f = u + iv. Since f ∈ H(Ω), both u and v are harmonic in Ω. Since |f | = 6 0 in Ω,
1
log |f | is a well-defined real function and in fact log |f | = log(u2 + v 2 ) 2 . Using ∆u = ∆v = 0
and the Cauchy-Riemann equations, we get
1 1
∆(log |f |) = [log(u2 + v 2 ) 2 ]xx + [log(u2 + v 2 ) 2 ]yy
1
= log(u2 + v 2 ) xx + log(u2 + v 2 ) yy
2
uu + vv uu + vv
x x y y
= +
u2 + v 2 x u2 + v 2 y
(u2 + v 2 )(u∆u + u2x + u2y + v∆v + vx2 + vy2 ) − 2(uux + vvx )2 − 2(uuy + vvy )2
=
(u2 + v 2 )2
(u2 + v 2 )(u2x + u2y + vx2 + vy2 ) − 2[u2 (u2x + u2y ) + 2uv(ux vx + uy vy ) + v 2 (vx2 + vy2 )]
=
(u2 + v 2 )2
2(u + v )(vx + vy ) − 2(u + v )(vx + vy2 )
2 2 2 2 2 2 2
=
(u2 + v 2 )2
=0
Problem 11.6
Rudin Chapter 11 Exercise 6.
Proof. Let A(Ω) be the area of the region Ω = f (U ). Referring to [2, §2.4, pp. 75, 76],b we
know that ZZ
A(Ω) = |f ′ (z)|2 dx dy,
U
b
See also the discussion on [97, Eqn. (6), p. 11].
11.1. Basic Properties of Harmonic Functions 353
where f (z) = u(x, y) + iv(x, y). Put z = reiθ , where 0 ≤ r < 1 and 0 ≤ θ ≤ 2π. Then the
formula of A(Ω) becomes
Z 2π Z 1
A(Ω) = |f ′ (r cos θ, r sin θ)|2 r dr dθ. (11.15)
0 0
Recall from Definition 4.23 that {einθ } forms an orthonormal set, the integral (11.15) becomes
Z 2π Z 1X
∞
A(Ω) = n2 |cn |2 r 2n−1 dr dθ. (11.16)
0 0 n=1
∞
X
Since f ′ (z) = ncn z n−1 ∈ H(U ), we have
n=1
p
n
lim sup n|cn | = 1
n→∞
Problem 11.7
Rudin Chapter 11 Exercise 7.
(a) Let ψ be a twice differentiable function on (0, ∞). Then f f = u2 + v 2 = |f |2 . Using the
formulas [100, Eqn. (3), p. 231], we see that
1 ∂ ∂ ∂ ∂
∂∂[ψ ◦ (f f )] = −i +i ψ(|f |2 )
4 ∂x ∂y ∂x ∂y
1 ∂ ∂ ′
= −i ψ (|f |2 ) · (uux + vvx ) + iψ ′ (|f |2 ) · (uuy + vvy )
2 ∂x ∂y
1 n ′′
= 2ψ (|f |2 ) · (uux + vvx )2 + ψ ′ (|f |2 ) · (u2x + uuxx + vx2 + vvxx )
2
354 Chapter 11. Harmonic Functions
Since f ∈ H(Ω), both u and v are harmonic in Ω. Furthermore, we note from Theorem
11.2 that |f ′ |2 = u2x + vx2 = u2y + vy2 , so the expression (11.17) becomes
1 n ′′
∂∂[ψ ◦ (f f )] = 2ψ (|f |2 ) · u2 (u2x + u2y ) + v 2 (vx2 + vy2 )
2
o
+ ψ ′ (|f |2 ) · u2x + u2y + vx2 + vy2 + u(uxx + uyy ) + v(vxx + vyy )
1 ′′
= 2ψ (|f |2 ) · |f |2 · |f ′ |2 + 2ψ ′ (|f |2 ) · |f ′ |2
2
= [|f |2 ψ ′′ (|f |2 ) + ψ ′ (|f |2 )] · |f ′ |2
= (ϕ ◦ |f |2 ) · |f ′ |2 (11.18)
as required.
α
Now the function ψ(t) = t 2 is clearly twice differentiable on (0, ∞) with
αα α α α α2 α −1
ϕ(t) = − 1 t 2 −1 + t 2 −1 = t2 ,
2 2 2 4
so we combine the formula (11.18) and [100, Eqn. (3), p. 232] to get
(b) Suppose that Φ : f (Ω) → C is defined by Φ(f ) = Φ(u + iv) = Φ(u(x, y), v(x, y)). Since
uxx + uyy = vxx + vyy = 0, we have
∆[Φ ◦ f ] = ∆Φ(f )
∂2 ∂2
= 2
Φ(u, v) + 2 Φ(u, v)
∂x ∂y
∂ ∂
= (Φu · ux + Φv · vx ) + (Φu · uy + Φv · vy )
∂x ∂y
∂ ∂
= ux Φu + Φu · uxx + vx Φv + Φv · vxx
∂x ∂x
∂ ∂
+ uy Φu + Φu · uyy + vy Φv + Φv · vyy
∂y ∂y
= ux (Φuu · ux + Φuv · vx ) + Φu · uxx + vx (Φuv · ux + Φvv · vx ) + Φv · vxx
+ uy (Φuu · uy + Φuv · vy ) + Φu · uyy + vy (Φuv · uy + Φvv · vy ) + Φv · vyy
= Φuu · (u2x + u2y ) + Φvv · (vx2 + vy2 ) + 2Φuv · ux vx + 2Φuv · uy vy . (11.19)
According to the Cauchy-Riemann equations, we can further reduce the expression (11.19)
to
as desired.
Finally, we take Φ(w) = |w| so that Φ(w) = Φ(|w|). Suppose that f = u + iv and
α
w = |f |α = (u2 + v 2 ) 2 . Direct differentiation implies that
α α α
Φuu + Φvv = 2α(u2 + v 2 ) 2 −1 + α(α − 2)u2 (u2 + v 2 ) 2 −2 + α(α − 2)v 2 (u2 + v 2 ) 2 −2
α
= α2 (u2 + v 2 ) 2 −1
and thus
(∆Φ) ◦ f = α2 |f |α−2 .
Substituting this into the formula (11.20), we have established that
∆(|f |α ) = α2 |f |α−2 · |f ′ |2 .
Problem 11.8
Rudin Chapter 11 Exercise 8.
Proof. The proof of this problem will be divided into two steps as follows:
≤ |fn (a) − fm (a)| + 3 · max un (a + 2Reit ) − um (a + 2Reit ).
−π≤t≤π
or equivalently,
sup |fn (z) − fm (z)| ≤ 3 · max |un (ζ) − um (ζ)| + |fn (a) − fm (a)|. (11.22)
z∈D(a;R) ζ∈C(a;2R)
Given ǫ > 0. Now our hypothesis asserts that there is an N1 ∈ N such that n, m ≥ N1
imply
ǫ
|un (ζ) − um (ζ)| <
6
for all ζ ∈ D(a; 2R). Thus it follows from the inequality (11.22) that
ǫ
|fn (z) − fm (z)| ≤ |fn (a) − fm (a)| + (11.23)
2
for all z ∈ D(a; R).
Particularly, if a ∈ S1 , then there exists an N2 ∈ N such that n, m ≥ N2 imply
ǫ
|fn (a) − fm (a)| <
2
so that the inequality (11.23) gives
for all z ∈ D(a; R) and all n, m ≥ N = max(N1 , N2 ). By [99, Theorem 7.8, p. 147], the
sequence {fn } converges uniformly in D(a; R) and this definitely implies
D(a; R) ⊆ S1 . (11.25)
To finish the proof that S1 = Ω, we need a result from [23, Theorem 1.28, p. 28]:
Put A = S1 , the above set relations (11.21) and (11.25) mean that we can take θ = 21
in Lemma 11.1 (The Basic Connectedness Lemma). Hence we conclude immediately that
S1 = Ω, as desired.
where Ra > 0 is a number satisfying the set relation (11.21). Therefore, it follows from
Step 1 that the inequality (11.24) holds in D(a; Ra ) and all n, m ≥ N (a, Ra ) (of course,
the positive integer N (a, Ra ) depends on a and Ra ) for every a ∈ K. Since K is compact,
there exists a finite set {a1 , a2 , . . . , ap } ⊆ K such that
p
[
K⊆ D(aj ; Rj ) ⊆ Ω,
j=1
for all z ∈ K. Using [99, Theorem 7.8, p. 147] again, the sequence {fn } converges
uniformly on K.
Problem 11.9
Rudin Chapter 11 Exercise 9.
Proof. Let D(a; r) ⊆ Ω. Since u is locally in L1 , the double integral considered in the question
is well-defined. If u is harmonic in Ω, then Definition 11.12 says that it satisfies
Z π
1
u(a) = u(a + reiθ ) dθ (11.26)
2π −π
for every r > 0 with D(a; r) ⊆ Ω. Multiplying both sides of the expression (11.26) by ρ and
integrating from 0 to r, we obtain
Z r
r2
u(a) = u(a)ρ dρ
2
Z0 r Z π
1
= u(a + ρeiθ ) dθ ρ dρ
0 2π
Z r Z π−π
1
= u(a + ρeiθ )ρ dθ dρ
2π 0 −π
Conversely, suppose that the formula (11.27) holds for any D(a; r) ⊆ Ω. On the one hand,
applying the polar coordinates, we can write
ZZ Z rZ π
u(x, y) dx dy = u(a + ρeiθ )ρ dθ dρ. (11.28)
0 −π
D(a;r)
Substituting the expressions (11.28) and (11.29) into the formula (11.27), we get
Z r Z r
u(a) ρ dρ = ρu(a) dρ (11.30)
0 0
Z rZ π
1
= u(a + ρeiθ )ρ dθ dρ
2π 0 −π
Z r
1
= U (a, ρ, θ)ρ dρ, (11.31)
2π 0
where Z π
U (a, ρ, θ) = u(a + ρeiθ ) dθ.
−π
As the integral on the left-hand side in the formula (11.30) is differentiable with respect to r,
so is the integral (11.31). Consequently, we derive from the First Fundamental Theorem of
Calculus [127, p. 161] that
Z π
1 r
ru(a) = · U (a, r, θ)r = u(a + reiθ ) dθ. (11.32)
2π 2π −π
whenever D(a; r) ⊆ Ω.
To finish the proof, we have to show that u is continuous in Ω. To this end, fix z ∈ Ω. Given
that ǫ > 0. Since u ∈ L1loc (Ω),d we must have u ∈ L1 (D(z; 2r ′ )) for some r ′ > 0 such that
D(z; 2r ′ ) ⊆ Ω. We fix this r ′ . By Problem 1.12, there exists a δz > 0 such that
Z
|u| dm < π(r ′ )2 ǫ (11.33)
E
whenever m(E) < δz and E ⊆ D(z; 2r ′ ). Clearly, we may assume that δz < r ′ . For every
ω ∈ D(z; r ′ ), it is always true that D(ω; r ′ ) ⊆ D(z; 2r ′ ) and
m D(ω; r ′ ) \ D(z; r ′ ) = m D(z; r ′ ) \ D(ω; r ′ ) < δz
if ω is very close to z. Therefore, we follow from the formula (11.27) and the inequality (11.33)
that
ZZ ZZ
1
|u(z) − u(ω)| = u(x, y) dx dy − u(x, y) dx dy
π(r ′ )2
D(z;r ′ ) D(ω;r ′ )
Z Z
1
= u dm − u dm
π(r ′ )2 D(z;r′ ) D(ω;r ′ )
Z Z
1
= u dm − u dm
2π(r ′ )2 D(z;r′ )\D(ω;r′ ) ′ ′
D(ω;r )\D(z;r )
"Z Z #
1
≤ |u| dm + |u| dm
2π(r ′ )2 D(z;r′ )\D(ω;r′ ) D(ω;r ′ )\D(z;r ′ )
1 ′ 2
< ′ 2
π(r ) ǫ + π(r ′ )2 ǫ
2π(r )
d
The notation L1loc (Ω) is the set of all locally integrable functions on Ω.
11.1. Basic Properties of Harmonic Functions 359
= ǫ.
Problem 11.10
Rudin Chapter 11 Exercise 10.
Proof.
Z b
1 1 1
f (x + iǫ) − f (x − iǫ) = ϕ(t) − dt
2πi a t − x − iǫ t − x + iǫ
Z b
ǫ ϕ(t)
= · dt
a π (x − t)2 + ǫ2
Z b
= Pǫ (x − t)ϕ(t) dt
Za ∞
= ϕ(x − t)Pǫ (t) dt, (11.34)
−∞
where
1 ǫ
Pǫ (t) = · 2
π t + ǫ2
relates to the formula [100, Eqn. (3), §9.7, p. 183] (in fact, it is the Poisson kernel for
the upper half-plane, see [110, p. 149]) and ϕ(t) = 0 if t ∈ R \ I. Using the convolution
notation introduced in Theorem 8.14, the expression (11.34) becomes
Lemma 11.2
If g ∈ L∞ , then we have
g(x+) + g(x−)
lim (g ◦ hλ )(x) = ,
λ→0 2
where hλ is the formula [100, Eqn. (3), §9.7, p. 183].
e
If g is continuous at x, then g(x+) = g(x−) and it is exactly Theorem 9.9.
360 Chapter 11. Harmonic Functions
Since the integrands are dominated by 2kgk∞ h1 (s) and the integrals converge pointwise
for every s as λ → 0, we deduce from Theorem 1.34 (The Lebesgue’s Dominated
Convergence Theorem) that
h g(x+) + g(x−) i
lim (g ∗ hλ )(x) −
λ→0 2
Z 0 h
g(x+) + g(x−) i
= lim g(x − λs) − h1 (s) dm(s)
λ→0 −∞ 2
Z ∞h
g(x+) + g(x−) i
+ lim g(x − λs) − h1 (s) dm(s)
λ→0 0 2
Z 0 h g(x+) + g(x−) i
= lim g(x − λs) − h1 (s) dm(s)
−∞ λ→0 2
Z ∞ h g(x+) + g(x−) i
+ lim g(x − λs) − h1 (s) dm(s)
0 λ→0 2
Z 0 h
g(x+) + g(x−) i
= g(x+) − h1 (s) dm(s)
−∞ 2
Z ∞h
g(x+) + g(x−) i
+ g(x−) − h1 (s) dm(s)
0 2
Z 0 Z ∞
g(x+) − g(x−) g(x−) − g(x+)
= h1 (s) dm(s) + h1 (s) dm(s)
−∞ 2 0 2
Z ∞ Z ∞
g(x+) − g(x−) g(x−) − g(x+)
= h1 (s) dm(s) + h1 (s) dm(s)
2 0 2 0
=0
which implies the desired result. We complete the proof of Lemma 11.2.
ϕ(x+) + ϕ(x−)
lim [f (x + iǫ) − f (x − iǫ)] = (11.36)
ǫ→0 2
ǫ>0
• The case when ϕ ∈ L1 . In this case, we apply Theorem 9.10 to the expression (11.35)
11.1. Basic Properties of Harmonic Functions 361
to get
Z ∞
lim |f (x + iǫ) − f (x − iǫ) − ϕ(x)| dm(x) = lim kϕ ∗ Pǫ − ϕk1 = 0. (11.37)
ǫ→0 −∞ ǫ→0
ǫ>0 ǫ>0
Denote f (x + iǫ) = fǫ+ (x) and f (x − iǫ) = fǫ− (x). Then the result (11.37) means that
• The case when ϕ(x+) and ϕ(x−) exist at x. This case has been settled already in the
formula (11.36).
We have completed the proof of the problem.
Remark 11.1
The integral considered in Problem 11.10 is an example of the so-called Cauchy type
integrals. For more details of this subject, please refer to Muskhelishvili’s book [76].
Problem 11.11
Rudin Chapter 11 Exercise 11.
Proof. We note that the following proof uses only the techniques from Chapter 10, not from
Chapter 11. Actually, the author admits that he is not able to apply the theory of harmonic
functions to prove this problem.
so that
Z Z
f (z) dz − f (z) dz
[a,b] [a,b]+ ni
Z 1 h i i
= |b − a| · f a + + (b − a)t − f a + (b − a)t dt
0 n
Z 1
i
≤ |b − a| · f a + + (b − a)t − f a + (b − a)t dt.
0 n
and then Z Z ǫ
f (z) dz − f (z) dz < . (11.40)
[a,b] [a,b]+ ni 3
Thus, if n > max( δ11 , δ12 ), then both the conditions (11.39) and (11.40) hold simulta-
neously. Since the inequality (11.40) also holds for [b, c] and [c, a] for large enough n,
we obtain immediately that
Z
f (z) dz < ǫ
∂∆n
which means Z Z
f (z) dz = lim f (z) dz = 0.
∂∆ n→∞ ∂∆
n
– Case (iii): ∆ intersects with I at only two points. Then I divides ∆ into a
triangle and a quadrangle or another triangle. If it is a quadrangle, then it can be
further divided into two triangles, see Figure 11.1 for an illustration. Since ∂∆ is a
sum of two or three boundaries of triangles, it follows from Case (i) and Case (ii)
that our result (11.38) remains true in this case.
K = I1 ∪ I2 ∪ · · · ∪ IN
Remark 11.2
(a) Classically, Problem 11.11 is a topic of the so-called removable sets for holomorphic
functions. To say it more precisely, let F be a class of functions from Ω to C. Then
a compact set K ⊆ Ω is said to be removable for holomorphic functions of class F if
every f ∈ F such that f ∈ H(Ω \ E) can be extended to a holomorphic function in Ω.
Examples of F are L∞ (Ω), C (Ω) and Lip α (Ω), where 0 < α ≤ 1.
(b) It is clear that Theorem 10.20 is a positive result for bounded and holomorphic func-
tions. Because of this, Painlevé was motivated and studied some more general prob-
lems: “Which subsets of C are removable? What geometric characterization(s) must
these subsets satisfy?” In fact, Painlevé proved a sufficient condition that if a compact
set K ⊆ Ω is of one-dimensional Hausdorff measure [18, pp. 215, 216], then it is
removable for bounded and holomorphic functions in Ω \ K.
(c) For the class Lip α (Ω) with 0 < α < 1, Dolženko [32] has shown in 1963 that a compact
set K is removable for holomorphic functions of this class if and only if the (1 + α)-
dimensional Hausdorff measure is zero. For the remaining case α = 1, Uy [119] verified
in 1979 that K is removable if and only if m(K) = 0.
(d) Besides the approach of Hausdorff measure, Ahlfors [1] introduced the analytic ca-
pacity (a purely complex-analytic concept) of a compact set K to study removable
compact sets for holomorphic functions of class L∞ . In fact, he proved that K is
removable for bounded analytic functions if and only if its analytic capacity vanishes.
Problem 11.12
Rudin Chapter 11 Exercise 12.
Proof.
• Proof of Harnack’s Inequalities. We first prove the special case that if u is harmonic
in D(a; R) and u > 0 in D(a; R), then for every 0 ≤ r < R and z = a + reiθ ∈ D(a; R), we
have
R−r R+r
u(a) ≤ u(z) ≤ u(a). (11.41)
R+r R−r
f
See also Problem 16.10.
364 Chapter 11. Harmonic Functions
To see this, choose ρ > 0 such that r < ρ < R. Using the second set of inequalities on
[100, p. 236], we know that
ρ−r ρ+r
u(a) ≤ u(z) = u(a + reiθ ) ≤ u(a) (11.42)
ρ+r ρ−r
for every θ ∈ R. Letting ρ → R in the inequalities (11.42), we obtain the desired results
(11.41).
Since Ω is a region and K ⊆ Ω, C \ Ω is closed in C and (C \ Ω) ∩ K = ∅. By Problem 10.1,
there exists a δ > 0 (depending on K and Ω) such that d(C \ Ω, K) = 2δ > 0. Clearly, we
have [ [
K⊆ D(z; δz ) ⊆ D(z; δ) ⊆ Ω,
z∈K z∈K
where 0 < δz < δ. Since K is compact, one can find a finite set {z1 , z2 , . . . , zm } with
positive numbers δ1 , δ2 , . . . , δm such that
m
[ m
[
K⊆ D(zk ; δk ) ⊆ D(zk ; δ) ⊆ Ω.
k=1 k=1
Now we apply the special case (11.41) to each disc D(zk ; δ) and consider only points
z ∈ D(zk ; δk ) to get
δ − δk δ−r δ+r δ + δk
u(zk ) ≤ u(zk ) ≤ u(z) ≤ u(zk ) ≤ u(zk ).
δ + δk δ+r δ−r δ − δk
Take positive numbers α and β such thatg
1 δ − δk 1 δ + δk
α= · min u(zk ) and β = · max u(zk ).
u(z0 ) 1≤k≤m δ + δk u(z0 ) 1≤k≤m δ − δk
Then we establish
αu(z0 ) ≤ u(z) ≤ βu(z0 ) (11.43)
m
[
for every z ∈ D(zk ; δk ). In particular, the inequalities (11.43) are true for all z ∈ K.
k=1
• The behavior of {un } in Ω \ {z0 } if un (z0 ) → 0. Let un (z) → u(z) for every z ∈ Ω and
a ∈ Ω \{z0 }. Since Ω \{z0 } is open in C, there exists a R > 0 such that D(a; R) ⊆ Ω \{z0 }.
Obviously, D(a; R) is a compact subset of Ω. By the inequalities (11.43), we have
for every z ∈ D(a; R) and n = 1, 2, . . ., where α and β depend on z0 , K and Ω only. Take
n → ∞ in the inequalities (11.44) and then use the hypothesis, we get
u(z) = 0 (11.45)
for all z ∈ D(a; R). Particularly, u(a) = 0. Since a is arbitrary, we conclude that u ≡ 0 in
Ω \ {z0 }. Finally, the continuity of u ensures that u(z0 ) = 0 and then u ≡ 0 in Ω.
• The behavior of {un } in Ω \ {z0 } if un (z0 ) → ∞. Instead of the result (11.45), we
obtain
u(z) = ∞
for all z ∈ D(a; R). Hence, using similar argument as the previous assertion, we conclude
that u(z) = ∞ in Ω.
g
Since δ depends on K and Ω, α and β trivially depend on z0 , K and Ω.
11.2. Harnack’s Inequalities and Positive Harmonic Functions 365
Problem 11.13
Rudin Chapter 11 Exercise 13.
3 3 3
Since 5+4 ≤ 5−4 cos t ≤ 5−4 on [−π, π], the integral (11.47) gives
Z π Z π
1 1 1
· u(eit ) dt ≤ u( 12 ) ≤ 3 · u(eit ) dt. (11.48)
3 2π −π 2π −π
therefore, we conclude from the hypothesis u(0) = 1 and the inequalities (11.48) that
1
≤ u( 21 ) ≤ 3.
3
This completes the proof of the problem.
Problem 11.14
Rudin Chapter 11 Exercise 14.
Proof. By translation and/or rotation, we may assume that L1 is the real axis (i.e., y = 0) and
if L1 and L2 intersect, the intersection point is the origin.
• Case (i): L1 and L2 are parallel. Suppose that the equation of L2 is y = A for some
A ∈ R \ {0}. Then we consider the function
πx πy
u(x, y) = e A sin .
A
It is obvious that u(x, 0) = u(x, A) = 0 for all x ∈ R. Furthermore, direct computation
gives uxx + uyy = 0 in R2 so that u is a harmonic function in R2 .
366 Chapter 11. Harmonic Functions
• Case (iii): L1 and L2 are non-parallel and non-perpendicular. Suppose that the
angle between L1 and L2 is a rational multiple of π, say mπ n , where m, n ∈ N and n is
m
1 n
not a multiple of 2 . Since z = un (x, y) + ivn (x, y) is entire, its imaginary part vn (x, y) is
harmonic in R2 by Theorem 11.4. On L1 , we have
for all x ∈ R. This implies that vn (x, 0) = 0 on R. Similarly, since points on L2 are in the
form z = cos mπ mπ
n + i sin n , so we have
mπ mπ n mπ mπ mπ mπ
(−1)m = cos + i sin = un cos , sin + ivn cos , sin
n n n n n n
which implies that vn (x, y) = 0 on L2 .
Now suppose that the angle between them is an irrational multiple of π, say απ for some
α ∈ R \ Q. Assume that u(x, y) was a harmonic function in R2 vanishing on L1 ∪ L2 . We
need to develop a harmonic version of Theorem 11.14 (The Schwarz reflection principle).
To this end, we recall the following concept: Let L be a straight line passing through
the origin. We say that a pair of points are symmetric with respect to L if L is the
perpendicular bisector of the line segment joining these points. For each z = (x, y) ∈ C,
it is easy to see that there exists a unique zL = (xL , yL ) ∈ C such that z and zL are
symmetric with respect to L. Next, the following result is taken from [9, Theorem 4.12,
p. 68]:
Lemma 11.3
Let z0 = (a, b) ∈ C and consider the line L = {(x, y) ∈ C | (x, y) · (a, b) = c} for
some a, b, c ∈ R. Define L+ = {(x, y) ∈ C | (x, y) · (a, b) > c}. Suppose that Ω ⊆ C
is a region symmetric with respect to L. If u is continuous on Ω∩L+, u is harmonic
on Ω ∩ L+ and u = 0 on Ω ∩ L, then the function
u(x, y), if (x, y) ∈ Ω ∩ L+ ;
U (x, y) =
−u(xL , yL ), if (x, y) ∈ Ω ∩ L−
is harmonic in Ω.
Applying Lemma 11.3 to our u with L = L2 , we see immediately that u vanishes on the
line L3 : y = (tan απ)x. In fact, repeated applications of Lemma 11.3 show that u vanishes
on lines in the form
y = (tan nαπ)x (11.49)
for every n ∈ Z.
To finish the proof, we have to show that the collection of the straight lines (11.49), denoted
by L, is dense in R2 . To see this, let α > 0. Given that 0 < θ < 12 , δ > 0 and ǫ > 0. By
the Kronecker’s Approximation Theoremh , we find that there exist m, n ∈ N such that
Note that tan(nαπ − mπ) = (−1)m tan(nαπ). The continuity of tan x implies that
If m is odd, then we can replace (−1)m tan(nαπ) by tan(−nαπ) in the estimation (11.50).
If − 21 < θ < 0, then a similar argument gives the following estimation
In this case, if m is even, then (−1)m+1 tan(nαπ) will be replaced by tan(−nαπ). In other
words, for every θ ∈ (− 12 , 12 ) \ {0}, we always have
π
|nαπ − mπ| < <δ
N
| tan(nαπ)| < ǫ.
Consequently, the inequality (11.51) actually holds for all − 21 < θ < 21 . Since the range of
tan x on (− π2 , π2 ) is R, our inequality (11.51) means that the collection L is dense in R2
for the case α > 0. For the case α < 0, we just consider −α > 0 and then the inequality
(11.51) remains valid with the integer −n.
Finally, since u is continuous on R2 and vanishes on L, we conclude that u ≡ 0 on R2 .
Problem 11.15
Rudin Chapter 11 Exercise 15.
Lemma 11.4
Given ǫ ∈ (0, 1). There exists a constant M > 0 such that
M
P1−ǫ (t) ≥ (11.52)
ǫ
for all t ∈ [−ǫ, ǫ].
i
Read [127, Problem 2.4, p. 12].
368 Chapter 11. Harmonic Functions
Proof of Lemma 11.4. Recall from the series expansion of the Poisson kernel that
Pr (t) = Pr (−t). We know from [100, Eqn. (4), p. 233] that Pr (t) is decreasing
on [0, π] and [−π, 0]. Therefore, it suffices to prove that the inequality (11.52) holds
for t = ǫ. In fact, we have 1 − (1 − ǫ)2 = 2ǫ − ǫ2 ≥ ǫ and
1 − (1 − ǫ)2 Mǫ
ǫP1−ǫ (ǫ) = ǫ · ≥ǫ· 2 =M
1 − 2(1 − ǫ) cos ǫ + (1 − ǫ)2 ǫ
Let’s return to the proof of the problem. When u is considered in D(0; r) for 0 < r < 1,
[11, Theorem 16.5, p. 227] implies that u attains its maximum on C(0; r). This fact and the
positivity of u imply that
u(reiθ ) ≤ u(seiθ )
for every θ ∈ [−π, π] if 0 ≤ r ≤ s < 1. Using this result and the hypothesis, if θ 6= 0, then we
obtain
Z Z
1 1
sup kur k1 = sup u(reiθ ) dθ ≤ lim u(reiθ ) dθ = 0.
0<r<1 2π [−π,π]\{0} 0<r<1 2π [−π,π]\{0} r→1
Hence it follows from Theorem 11.30 that one can find a unique positive Borel measure on T
such that u = P [ dµ]. Next, it is easy to see that
Z Z
1 1
u (1 − ǫ)eiθ = P1−ǫ (θ − t) dµ(eit ) = P1−ǫ (t) dµ ei(θ−t) , (11.53)
2π T 2π I(eiθ ;2π)
where θ 6= 0 and I(eiθ ; 2π) ⊆ T denotes the open arc centred at eiθ 6= 1 with arc length 2π. We
notice that [−ǫ, ǫ] ⊆ I(eiθ ; 2π), so we apply Lemma 11.4 to the representation (11.53) and get
Z Z
1 M M M
u (1 − ǫ)eiθ ≥ dµ ei(θ−t) ≥ dµ ei(θ−t) ≥ µ I(eiθ ; 2ǫ) . (11.54)
2π [−ǫ,ǫ] ǫ ǫ [−ǫ,ǫ] ǫ
Problem 11.16
Rudin Chapter 11 Exercise 16.
11.2. Harnack’s Inequalities and Positive Harmonic Functions 369
Proof. Assume that u was the Poisson integral of a complex measure µ on T , i.e., u = P [ dµ].
We know from [100, Eqn. (2), p. 244] that
for every 0 < r < 1. Thus it deduces from Theorem 11.30(a) that µ is a (unique) complex Borel
measure. Next, we express u as
1 − |z|2
u(z) = −4 · Im (z) · . (11.55)
|1 + z|4
On the one hand, as the first inequality of Theorem 11.20 only requires that µ is a Borel measure
on T , so we may apply it with a fixed 0 < α < 1 and one can find a constant cα > 0 such that
If z ∈ (−1, 0], then we have Im (z) = 0 and we see from the representation (11.55) that u(z) = 0.
Consequently, the inequalities (11.56) force that
On the other hand, given 0 < ǫ < sin−1 α, if z = (−1 + ǫ) + iǫ2 , then it is easily checked that
z ∈ −Ωα and the expression (11.55) becomes
4 2 − ǫ − ǫ3
u(z) = u (−1 + ǫ) + iǫ2 = − · .
ǫ (1 + ǫ2 )2
Therefore, we obtain
4 2 − ǫ − ǫ3
lim u(z) = lim − · = −∞
z→−1 ǫ→0 ǫ (1 + ǫ2 )2
z∈−Ωα
which implies (Nα u)(−1) = ∞, a contradiction to the result (11.57). This proves the first
assertion that u cannot be a Poisson integral of any measure on T .
For the second assertion, if u = v − w, where both v and w are positive harmonic functions
in U , then we have |u(z)| ≤ |v(z)| + |w(z)| = v(z) + w(z) for all z ∈ U so that
Z π Z π Z π
1 1 1
kur k1 = |u(reiθ )| dθ ≤ v(reiθ ) dθ + w(reiθ ) dθ (11.58)
2π −π 2π −π 2π −π
for every 0 < r < 1. Since v and w are harmonic in U , they satisfy the mean value property, so
the inequality (11.58) gives
Z π
1
kur k1 = |u(reiθ )| dθ ≤ v(0) + w(0)
2π −π
for every 0 < r < 1. In other words, sup kur k1 is bounded and Theorem 11.30(a) shows that
0<r<1
u = P [ dµ] for a unique complex Borel measure on T . However, it is impossible by the first
assertion and hence u is not the difference of two positive harmonic functions in U . We have
completed the analysis of the problem.
Problem 11.17
Rudin Chapter 11 Exercise 17.
370 Chapter 11. Harmonic Functions
Proof. Set Φ = {u : U → R | u is positive, harmonic and u(0) = 1} and let C be the set whose
members are the positive Borel measures µ on T of |µ|(T ) = 1. We divide the proof into several
steps:
• Step 1: There is an isomorphism between Φ and C. On the one hand, for each
µ ∈ C, we know from §11.17 that
Z
u(z) = P (z, eit ) dµ(eit ) (11.59)
T
which implies u(0) = 1, i.e., u ∈ Φ. On the other hand, if u ∈ Φ, then the mean value
property shows that
Z π Z π
1 1
kur k1 = |u(reiθ )| dθ = u(reiθ ) dθ = u(0) < ∞
2π −π 2π −π
so that u = P [ dµ] for a unique positive Borel measure on T by Theorem 11.30(a). Besides,
the expression (11.60) gives |µ|(T ) = 1, so µ ∈ C. Consequently, the mapping f : Φ → C
defined by
f (u) = µ
is bijective. In fact, f is a homemorphism because for every u, v ∈ Φ, if their corresponding
positive Borel measures are µ and ν respectively, then
Z Z
αu(z) + βv(z) = α P (z, eit ) dµ(eit ) + β P (z, eit ) dν(eit )
Z T T
• Step 2: Both Φ and C are convex. Suppose that u1 , u2 ∈ Φ and u = λu1 + (1 − λ)u2 ,
where 0 ≤ λ ≤ 1. Since u1 and u2 are positive and harmonic in U , u is also positive and
harmonic in U . Furthermore, we have u(0) = λu1 (0) + (1 − λ)u2 (0) = 1. In other words,
u ∈ Φ and thus Φ is a convex set. By Step 1, C is also convex.
• Step 3: Extreme points of C. Denote ext C to be the set of extreme points of C.j
Let µ ∈ ext C. Since u is positive in U , it follows from Step 1 that µ(E) > 0 for every
Borel subset E of T . Now we claim that supp µ = {eit ∈ T | µ(eit ) 6= 0} is a unit mass
concentrated at eit (see [100, Example 1.20(b), p. 17] for the definition). Otherwise,
there was a measurable set E ⊆ T such that 0 < µ(E) < 1. Then we have 0 < µ(T \E) < 1.
We define
µ(F ∩ E) µ(F \ E)
µE (F ) = and µT \E (F ) = . (11.61)
µ(E) µ(T \ E)
Since µ is a positive Borel measure on T , both µE and µT \E are also positive Borel measures
on T . Furthermore, it is clear that
|µ|(T ∩ E) |µ|(T \ E)
|µE |(T ) = = 1 and |µT \E |(T ) = = 1.
|µ|(E) |µ|(T \ E)
j
A point x of a convex set X is an extreme point of X if x cannot be written as a proper convex combination
x = λx1 + (1 − λ)x2 , where 0 < λ < 1, x1 , x2 ∈ X and x1 6= x2 .
11.3. The Weak∗ Convergence and Radial Limits of Holomorphic Functions 371
Thus µE and µT \E belong to C. If we set λ = µ(E), then for every measurable subset
F ⊆ T , we get from the definitions (11.61) that
µ(F ∩ E) µ(F \ E)
λµE (F ) + (1 − λ)µT \E (F ) = µ(E) · + [1 − µ(E)] ·
µ(E) µ(T \ E)
= µ(F ∩ E) + µ(F \ E)
= µ(F )
which means µ ∈
/ ext C, a contradiction. Hence, we obtain
ext C = {δeit | eit ∈ T },
where δx is the Dirac delta function at x.
• Step 4: Extreme points of Φ. By Step 2, the restriction fext Φ is certainly an iso-
morphism between ext Φ and ext C. Using Step 3 and the definition (11.59), we see
that
ext Φ = {P (z, eit ) | eit ∈ T }.
Problem 11.18
Rudin Chapter 11 Exercise 18.
Proof. Let X be a Banach space and X ∗ be its dual space.k Recall Definition 5.3 that
kΛn k = sup{|Λn (x)| | x ∈ X and kxk = 1}.
Suppose that {Λn } ⊆ X ∗ converges weakly to Λ ∈ X ∗ , so Λn (x) → Λ(x) for every x ∈ X and
then
sup |Λn (x)| = Mx < ∞
n∈N
for each x ∈ X. By the definition, {Λn } is a collection of bounded linear functionals. Hence
Theorem 5.8 (The Banach-Steinhaus Theorem) asserts the existence of a positive constant M
such that
sup kΛn k ≤ M,
n∈N
completing the proof of the problem.
Remark 11.3
We note that the convergence in Problem 11.18 is called the weak∗ convergence. In
fact, there are two important convergences in functional analysis. They are called strong
convergence and weak convergence. To be more precisely, let X be a Banach space and
fn , f ∈ X for every n ∈ N. If kfn − f k → 0, then {fn } is said to converge strongly to f .
Similarly, we say {fn } converges weakly to f if we have g(fn ) → g(f ) for every g ∈ X ∗∗ .
Read, for example, [9, p. 115, 116].
k
Recall from Remark 5.21 that X ∗ is the collection of all bounded linear functionals on X.
372 Chapter 11. Harmonic Functions
Problem 11.19
Rudin Chapter 11 Exercise 19.
Proof.
(a) Since 0 < r < 1, 1 − r > 0. By the power series of the cosine function, we have
(1 − r)2
cos(1 − r) > 1 −
2
which implies that
(1 − r)(1 − r 2 ) 1 − r − r2 + r3
(1 − r)Pr (1 − r) = > = 1.
1 − 2r cos(1 − r) + r 2 1 + r(1 − r)2 − 2r + r 2
We apply part (a) and the fact that Pr (t) is an even function of t to the integral (11.62)
to get Z
δu(1 − δ) ≥ dµ(eiθ ) = µ(Iδ ).
Iδ
Problem 11.20
Rudin Chapter 11 Exercise 20.
Proof. Since m(E) = 0, for each n ∈ N, we know from [99, Remark 11.11(b), p. 309] that there
exists an open set Vn ⊆ T containing E such that m(Vn ) ≤ 21n . Define
1, if eiθ ∈ Vn ;
χVn (eiθ ) =
0, otherwise
and ϕ : T → R by
∞
X
ϕ(eiθ ) = χVn (eiθ ).
n=1
Obviously, we have
∞ Z
X ∞
X ∞
X 1
χVn (eiθ ) dm = m(Vn ) ≤ = 1,
2n
n=1 T n=1 n=1
Since U is simply connected, u is the real part of a holomorphic function g in U , see [11, Theorem
16.3, p. 226]. Let f = e−g . Then we have f ∈ H(U ) and |f | = e−Re g = e−u so that
for every eiθ ∈ E. By the definition of f , f (z) 6= 0 for every z ∈ U . Thus f is nonconstant.
Since ϕ ≥ 0, we have u ≥ 0 and consequently, f ∈ H ∞ . If f (0) 6= 1, then we can replace f by
fe(z) = ff (z) e
(0) so that f (0) = 1. This completes the proof of the problem.
Remark 11.4
For further information, the reader can refer to [89, p. 295] and [133, pp. 105, 276].
Problem 11.21
Rudin Chapter 11 Exercise 21.
374 Chapter 11. Harmonic Functions
t+is−1
Proof. We first show that g ∈ / H ∞ . In fact, fix s ∈ R, consider zt = t+is+1 for 0 < t < ∞. Then
1+zt
it is easy to check that |zt | < 1 and 1−zt = t + is. Therefore, we have
2
g(zt ) = exp(−et+is )
(t + 1) + is
which gives
2
|g(zt )| = p exp(−et cos s). (11.66)
(t + 1)2 + s2
Now we put s = π in the expression (11.66) to get |g(zt )| → ∞ as t → ∞. Consequently,
/ H ∞.
g∈
Next, if eiθ ∈ T and 0 ≤ r < 1, then we have
and so
h i sin θ i
(1 − eiθ ) exp − exp , if θ =
6 0;
1 − cos θ
g∗ (eiθ ) = lim g(reiθ ) = (11.67)
r→1
0, if θ = 0.
Hence g∗ (eiθ ) exists for every eiθ ∈ T . By the representation (11.67), we know that g ∗ (eiθ ) is
continuous at every θ ∈ (0, 2π). Thus it suffices to show that g∗ (eiθ ) is continuous at θ = 0. To
see this, since
sin θ cos θ − sin θ
lim = lim = lim = 0,
θ→0 1 − cos θ θ→0 sin θ θ→0 cos θ
As a consequence, we establish the fact that g∗ ∈ C(T ) and this completes the proof of the
problem.
Problem 11.22
Rudin Chapter 11 Exercise 22.
11.4. Miscellaneous Problems 375
Proof. For each 0 ≤ r < 1, the function ur : T → C is continuous, so u−1 r (R) is measurable.
Suppose that ur = u+ − + −
r − ur , where ur and ur are the positive and negative parts of ur , see
Definition 1.15. Since {ur } is uniformly integrable, there exists a δ > 0 such that
Z
ur dm < 1 (11.68)
E
whenever 0 ≤ r < 1 and m(E) < 2δ. Let Er+ = {θ ∈ [0, 2π] | u+ iθ +
r (e ) > 0}. Then Er is
measurable and Z π Z
u+
r dm = ur dm. (11.69)
−π Er+
+
m(Er,k ) ≤ m(Ik ) ≤ δ < 2δ
for k = 1, 2, . . . , N . Hence we follow from the inequality (11.68) and the result (11.69) that
Z π XN Z X N Z
u+
r dm =
ur dm ≤ ur dm ≤ N
+ +
−π k=1 Er,k k=1 Er,k
for almost all points eiθ ∈ T , where zj → eiθ and zj ∈ eiθ Ωα for α < 1. Particularly, this implies
that
urj (eiθ ) = u(rj eiθ ) → f (eiθ )
as j → ∞ a.e. on T , where {rj } ⊆ [0, 1) and rj → 1 as j → ∞. Furthermore, since σ(T ) < ∞
and {ur } ⊆ L1 (T ) is uniformly integrable, Problem 6.10(d) ensures that
Z
kurj − f k1 = |urj (eiθ ) − f (eiθ )| dσ(eiθ ) → 0
T
l
It is obvious that N is independent of r.
376 Chapter 11. Harmonic Functions
as j → ∞ or equivalently, we have
Z Z
lim urj (eiθ ) dσ(eiθ ) = f (eiθ ) dσ(eiθ ). (11.70)
j→∞ T T
dµ = f dσ
and hence u = P [f ] for some f ∈ L1 (T ) which completes the proof of the problem.
Remark 11.5
We can also apply Theorem 17.13 (F. and M. Riesz Theorem) on [100, p. 341] to prove that
µ ≪ σ.
Problem 11.23
Rudin Chapter 11 Exercise 23.
2
Proof. Given z ∈ U , since |eiθ − z|2 ≥ |eiθ | − |z| = (1 − |z|)2 , we have
1 − |z|2 1 − |z|2 1 + |z|
|P (z, eiθ )| = iθ 2 ≤ = < ∞.
|e − z| (1 − |z|)2 1 − |z|
converge absolutely so that we may split the representation of u into the difference of v and w.
For each n ∈ N, we define
n
X n
X
vn (z) = k−2 P (z, eiθk ) and wn (z) = k−2 P (z, e−iθk ).
k=1 k=1
Using Problem 11.4, we know that P (z, eiθk ) and P (z, e−iθk ) are harmonic in U for k = 1, 2, . . . , n
so that vn and wn are also harmonic in U . Fix 0 ≤ r < 1. If z ∈ D(0, r), then we have
1 − |z| > 1 − r and thus
X∞ ∞
P (z, eiθn ) 1 + r X 1
≤ · < ∞.
n=1
n2 1 − r n=1 n2
By the Weierstrass M -test, we see that {vn } converges uniformly to v in D(0; r). Let K be a
compact subset of U . Then there exists a 0 < r < 1 such that K ⊆ D(0, r), so it follows from
Theorem 11.11 (Harnack’s Theorem) that v is harmonic in U . By a similar argument, we are
able to show that w is also harmonic in U .
11.4. Miscellaneous Problems 377
Since P (z, eit ) > 0 for every z ∈ U and eit ∈ T , we have v > 0 and w > 0. Now, for every
0 ≤ r < 1, we observe that
Z π
1
kvr k1 = v(reiθ ) dθ
2π −π
Z π hX ∞ i
1
= n−2 P (reiθ , eiθn ) dθ
2π −π
n=1
X∞ h 1 Z π i
= n−2 P (reiθ , eiθn ) dθ
n=1
2π −π
∞
X
= n−2
n=1
< ∞.
Similarly, we have sup kwr k1 < ∞ and hence v and w satisfy the requirements of Theorem
0<r<1
11.30. Then there exist unique positive Borel measures µ and ν on T such that
Since u = v − w, we establish
u = P [ d(µ − ν)],
where µ − ν is clearly a measure on T .
Next, if x ∈ (−1, 1), then we get from [100, Eqn. (2), p. 233] that
1 − x2
P (x, eiθn ) = P (x, e−iθn ) = .
1 − 2x cos θn + x2
Consequently, we conclude that u(x) = 0 if −1 < x < 1. Finally, suppose that z = 1 − ǫ + iǫ,
where ǫ = sin θ > 0. Then it is easy to check that
P (z, eiθ ) − P (z, e−iθ ) = P (1 − sin θ + i sin θ, eiθ ) − P (1 − sin θ + i sin θ, e−iθ )
1 − [(1 − sin θ)2 + sin2 θ] 1 − [(1 − sin θ)2 + sin2 θ]
= 2
+
| cos θ − 1 + sin θ| |(cos θ − 1 + sin θ) − 2i sin θ|2
h 1 1 i
= 2 sin θ(1 − sin θ) · 2
− 2 2
(cos θ + sin θ − 1) (cos θ + sin θ − 1) + 4 sin θ
h 1
= 2 sin θ(1 − sin θ) ·
4 sin2 θ2 (cos 2θ − sin 2θ )2
1 i
− . (11.72)
4 sin2 θ2 (cos 2θ − sin 2θ )2 + 4 sin2 θ
sin θ
≥ . (11.73)
2 sin2 2θ
By the definition, ǫ is small if and only if θ > 0 is small. In this case, cos2 θ
2 > 12 , so the inequality
(11.73) becomes
sin2 θ 1 2 cos2 θ2 1 1
P (z, eiθ ) − P (z, e−iθ ) ≥ ·
2 θ sin θ
= > = . (11.74)
2 sin 2 sin θ sin θ ǫ
If z = x + iy ∈ U with x > 0 and y > 0, then it is easy to see from [100, Eqn. (6), p. 233] that
for every θ ∈ (0, π2 ). By the definition of u, if we put z = 1 − ǫ + iǫ, then we may apply the
estimate (11.75) to get
for every n ∈ N. Now we take ǫn = sin θn = sin 21n in the inequality (11.76) and then apply the
estimate (11.74) as well as the fact that sin θn ≤ θn for every n ∈ N to obtain
1 1 2n
u(1 − ǫn + iǫn ) > = 1 ≥ (11.77)
n 2 ǫn n2 sin 2n n2
2n
for every n ∈ N. Since n2
→ ∞ as n → ∞, we conclude immediately from the inequality (11.77)
that
lim u(1 − ǫn + iǫn ) = ∞,
n→∞
Problem 11.24
Rudin Chapter 11 Exercise 24.
Proof. Most assertions of this problem come from [99, Exercise 15, p. 199] and their solutions
are shown by the author in [124, Problem 8.15, pp. 187 – 190], so we just prove the necessary
assertions here.
1−cos θ
• Proof of KN −1 (t) ≤ LN (t). Since sin 2θ = 2 , we know that
1 sin N2t 2
KN −1 (t) = · . (11.78)
N sin 2t
By [99, Exercise 8, p. 197], we always have | sin nθ| ≤ n| sin θ| for n = 0, 1, 2, . . . so that
sin N t 2 | sin N2t |2
2
t = ≤ N2 (11.79)
sin 2 | sin 2t |2
for N2 |t| ≤ π
2. If 0 ≤ t ≤ π, then the power series of sin x implies that sin 2t ≥ t
π ≥ 0 so
that
sin N t 2 1 π2
2
≤ ≤ . (11.80)
sin t
2 sin2 t
2
t2
11.4. Miscellaneous Problems 379
Hence, by putting the inequalities (11.79), (11.80) and (11.81) into the expression (11.78),
π
we conclude that if |t| ≤ N , then
1
KN −1 (t) ≤ · N2 = N; (11.82)
N
π
if N ≤ |t| ≤ π, then
π2
. KN −1 (t) ≤ (11.83)
N t2
By the definition of LN , we see that KN −1 (t) ≤ LN (t) for every t ∈ [−π, π].
R
• Proof of T LN dσ ≤ 2. Using the estimates (11.82) and (11.83), we have
Z Z π
1
LN dσ = Ln (t) dt
T 2π −π
Z Z
1 π dt
= N dt +
2π |t|≤ π 2N π ≤|t|≤π t2
N N
1
=2−
N
≤2
for every N ∈ N.
• Proof of Fejér’s Theorem. The result about the convergence of the arithmetic means
is called Fejér’s Theorem.m Recall from [100, Eqn. (1), p. 101] that
Z Z
sn (f ; θ) = f (eit )Dn (θ − t) dσ = f ei(θ−t) Dn (t) dσ,
T T
Suppose that eiθ is a Lebesgue point of f ∈ L1 (T ). Imitating the proof of Theorem 11.23
(Fatou’s Theorem), we may assume without loss of generality that f (eiθ ) = 0. Thus it
suffices to prove that
lim σN (f ; θ) = 0.
N →∞
Z Z 0
1 π
= f ei(θ−t) KN (t) dt + f ei(θ−t) KN (t) dt
2π 0 −π
Z π
1 i(θ−t)
≤ f e + f ei(θ+t) · KN (t) dt.
2π 0
Put g(t) = f ei(θ−t) + f ei(θ+t) and
Z x
G(x) = g(t) dt,
0
where 0 ≤ x ≤ π. We have
Z x Z x
G(x) 1 1
0< ≤ |f ei(θ−t) | dt + |f ei(θ+t) | dt
x x 0 x 0
and since eiθ is a Lebesgue point of f , we know from the definition [100, Eqn. (5), p. 241]
that
G(x)
lim = 0. (11.85)
x→0 x
Given ǫ > 0. Firstly, the result (11.85) means that we may choose a δ > 0 such that
G(x) < ǫx for all 0 < x < δ. If N > 1δ , then we deduce from (11.82) that
Z 1 Z 1
N N N +1
g(t)KN (t) dt ≤ (N + 1) g(t) dt < ǫ < 2ǫ. (11.86)
0 0 N
Secondly, the property (11.83) actually holds for all 0 < |t| ≤ π so that
Z δ Z
π 2 δ g(t)
g(t)KN (t) dt < dt
1 N 1 t2
N N
Z
π 2 δ dG(t)
=
N 1 t2
N
Z δ
π 2 h G(δ) 2
1 G(t) i
= − N G + 2 dt
N δ2 N 1 t3
N
Z δ
π2 ǫ dt
< + 2ǫ
N δ 1 t2
N
π2 ǫ
< + 2ǫN
N δ
< 3π 2 ǫ. (11.87)
Thirdly, we note that
Z π Z π
π2
g(t)KN (t) dt ≤ g(t) dt
δ N δ2 δ
Z Z
π2 h i(θ−t)
i
≤ 2
|f e | dt + |f ei(θ+t) | dt
Nδ T T
π2
= · 2kf k1
N δ2
<ǫ (11.88)
for large enough N . Finally, by combining the inequalities (11.86), (11.87) and (11.88),
we conclude immediately that
Z π
1 1 3(1 + π 2 )
|σN (f ; θ)| ≤ g(t)KN (t) dt < (3ǫ + 3π 2 ǫ) = ǫ
2π 0 2π 2π
11.4. Miscellaneous Problems 381
for sufficiently large N . Since ǫ is arbitrary, we have obtained the desired result that
σN (f ; θ) → 0 as N → ∞.
We have completed the proof of the problem.
Problem 11.25
Rudin Chapter 11 Exercise 25.
Proof. Let z = x + iλ, where λ > 0. Using [100, Eqn. (3), §9.7, p. 183], we find that
Z ∞ Z
1 1 ∞ λf (x − y)
u(z) = u(x, λ) = (f ∗ hλ )(x) = √ f (x − y)hλ (y) dy = dy. (11.89)
2π −∞ π −∞ y 2 + λ2
We prove the problem by showing the following steps:
• Step 1: ϕ(z, t) is harmonic in Π+ for every t ∈ R. By the change of variable t = x− y,
the expression (11.89) can be written as
Z Z
1 ∞ λf (t) 1 ∞
u(z) = u(x, λ) = dt = ϕ(x + iλ, t)f (t) dt,
π −∞ (x − t)2 + λ2 π −∞
where ϕ : Π+ × R is given by
λ 1
ϕ(x + iλ, t) = = Im . (11.90)
(x − t)2 + λ2 t − (x + iλ)
1
Since t−(x+iλ) is holomorphic in Π+ for every t ∈ R, Theorem 11.4 ensures that ϕ is
harmonic in Π .n Hence ϕ(z, t) is continuous on Π+ and then it satisfies the mean value
+
In other words, we have ϕ(z, t) ∈ Lq (R1 ). By the inequality (11.91), since ϕ(z, t) is
obviously bounded by Mz , we know that ϕ(z, t) ∈ L∞ (R1 ).
• Step 4: u satisfies the mean value property. Suppose that z ∈ Π+ and r is a positive
number such that D(z; r) ⊆ Π+ . If q is the conjugate exponent of p, then we obtain from
Theorem 3.8 and Step 3 that
Z ∞
kϕ(z + reiθ , ·)f (·)k1 = |ϕ(z + reiθ , t)f (t)| dt ≤ kf kp × kϕ(z + reiθ , ·)kq < ∞.
−∞
Next, we observe from the inequality (11.92) that we can pick Mζ in such a way that the
set {Mζ | ζ = z + reiθ and −π ≤ θ ≤ π} is bounded. Therefore, we get
Z π
kϕ(z + reiθ , ·)f (·)k1 dθ < ∞.
−π
Consequently, we apply Theorem 8.8 (The Fubini Theorem) and Step 1 to conclude that
Z π Z π Z ∞
1 1
u(z + reiθ ) dθ = 2 ϕ(z + reiθ , t)f (t) dt dθ
2π −π 2π −π −∞
Z Z π
1 ∞ 1
= ϕ(z + reiθ , t) dθ f (t) dt
π −∞ 2π −π
Z
1 ∞
= ϕ(z, t)f (t) dt
π −∞
= u(z).
Thus u is continuous on Π+ .
Problem 12.1
Rudin Chapter 12 Exercise 1.
Proof. Let f (z) = (z − a)(z − b)(z − c). Then f is nonconstant and entire. By Theorem 10.24
(The Maximum Modulus Theorem), we know that
383
384 Chapter 12. The Maximum Modulus Principle
= t2 (L − t)2 (t2 − Lt + L2 ).
Define the function F : [0, L] → R by F (t) = t2 (L − t)2 (t2 − Lt + L2 ). Elementary differentiation
shows that
F ′ (t) = 6t5 − 15Lt4 + 16L2 t3 − 9L3 t2 + 2L4 t
= t(2t − L)(3t3 − 6Lt2 + 5L2 t − 2L3 )
= t(2t − L)(t − L)(3t2 − 3Lt + 2L2 ).
Thus F ′ (t) = 0 if and only if t = 0, L2 , L. By the First Derivative Test, it is easily seen that F
3 6
attains its maximum 64 L at t = L2 . Hence we conclude that
r √
p L 3 3
max |f (z)| = max |f (z)| = max F (t) = F = L .
z∈∆ z∈∂∆ t∈[0,L] 2 8
This completes the analysis of the problem.
Problem 12.2
Rudin Chapter 12 Exercise 2.
Proof. Suppose that f (i) = α. If |α| = 1, then Theorem 10.24 (The Maximum Modulus The-
orem) forces that f (z) = α in Π+ . In this case, we have |f ′ (i)| = 0. Suppose that α ∈ U . By
Definition 12.3, ϕα (α) = 0. Furthermore, we know from [11, Theorem 13.16, p. 183] that the
mapping h : Π+ → U given by
z − β
h(z) = eiθ , (12.1)
z−β
where Im β > 0 and θ ∈ R, is a bijection. Particularly, we take β = i in the definition (12.1).
Clearly, we have h(i) = 0. Next, we consider the mapping F = ϕα ◦ f ◦ h−1 : U → U . Then we
have F ∈ H ∞ , kF k∞ ≤ 1 and
F (α) = ϕα f h−1 (0) = ϕα f (i) = ϕα (α) = 0.
Hence it follows from Theorem 12.2 (Schwarz’s Lemma) that
|F ′ (0)| ≤ 1. (12.2)
z+eiθ
Since h−1 (z) = i · eiθ −z
, we have (h−1 )′ (0) = 2ie−iθ . Consequently, we see from Theorem 12.4
that
2ie−iθ f ′ (i)
F ′ (0) = ϕ′α (α) × f ′ (i) × (h−1 )′ (0) = , (12.3)
1 − |α|2
so the inequality (12.2) implies that
1 − |α|2
|f ′ (i)| ≤ .
2
Thus |f ′ (i)| attains the maximum 12 when α = 0. In this case, we observe from the expression
(12.3) that |F ′ (0)| = 1, so Theorem 12.2 (Schwarz’s Lemma) implies that
F (z) = λz for some
constant λ with |λ| = 1. Since ϕ0 (z) = z, we conclude that f h−1 (z) = λz. Now if we put
z = h(ζ), then it asserts that
ζ − i
f (ζ) = λh(ζ) = λeiθ . (12.4)
ζ +i
Since λ = eiφ for some φ ∈ R, we may simply replace λeiθ by eiθ in the representation (12.4)
which gives all extremal functions with |f ′ (i)| = 12 . This completes the proof of the problem.
12.1. Applications of the Maximum Modulus Principle 385
Problem 12.3
Rudin Chapter 12 Exercise 3.
Proof. If f is constant, then there is nothing to prove. Thus, without loss of generality, we may
assume that f is nonconstant. In this case, f has a local minimum in Ω if and only if f has a
zero in Ω. Assume that f was a non-vanishing function in Ω. Then f1 ∈ H(Ω) and |f | has a
local minimum at z0 ∈ Ω if and only if |f1 | has a local maximum at z0 . By Theorem 10.24 (The
Maximum Modulus Theorem), f1 is forced to be constant which is impossible by our hypothesis.
Hence f has a zero in Ω, completing the proof of the problem.
Problem 12.4
Rudin Chapter 12 Exercise 4.
Proof.
(a) Assume that f was non-vanishing in D. By Theorem 10.24 (The Maximum Modulus
Theorem), f 6= 0 on ∂D. Thus it is true that f 6= 0 in D. Denote M = |f (z)| on ∂D. On
the one hand, Theorem 10.24 (The Maximum Modulus Theorem) again implies that
|f (z)| ≤ M (12.5)
Problem 12.5
Rudin Chapter 12 Exercise 5.
Proof. Given that ǫ > 0. Since fn → f uniformly on ∂Ω, there exists an N ∈ N such that
m, n ≥ N imply that
|fn (z) − fm (z)| < ǫ (12.7)
for every z ∈ ∂Ω. Since Ω is bounded, we may apply Theorem 10.24 (The Maximum Modulus
Theorem) to assert that the inequality (12.7) holds for every z ∈ Ω, i.e.,
max |fn (z) − fm (z)| = max |fn (z) − fm (z)| < ǫ (12.8)
z∈Ω z∈∂Ω
for m, n ≥ N . According to the Cauchy Criterion for Uniform Convergence [99, Theorem 7.8,
p. 147], the inequality (12.8) ensures that {fn } converges uniformly on Ω which ends the proof
of the problem.
386 Chapter 12. The Maximum Modulus Principle
Problem 12.6
Rudin Chapter 12 Exercise 6.
n
X
Ind Γ (z) = Ind γj (z),
j=1
we have Ind Γ (z) = 0, a contradiction. Therefore, V must be bounded. Since Ind Γ (α) = 0 for
all α ∈
/ Ω, we get
V ⊆ Ω. (12.9)
Let ζ ∈ ∂V . Then ζ ∈ / V . If ζ lies in another component U , then there exists a δ > 0 such that
D(ζ; δ) ∩ V 6= ∅ and D(ζ; δ) ∩ U 6= ∅, but this is a contradiction by [74, Theorem 25.1, p. 159].
Hence ζ ∈/ C \ Γ∗ , i.e., ζ ∈ Γ∗ and then ∂V ⊆ Γ∗ . Since Γ is a cycle in Ω, we have
Γ∗ ⊆ Ω. (12.10)
Combining the set relations (12.9) and (12.10), we conclude that V ⊆ Ω. Recall that V is
bounded, V is compact. By the hypothesis, we have |f (ζ)| ≤ 1 for every ζ ∈ ∂V ⊆ Γ∗ . By
Theorem 10.24 (The Maximum Modulus Theorem), we observe that
|f (z)| ≤ 1
for every z ∈ V with Ind Γ (z) 6= 0. This completes the proof of the problem.
Problem 12.7
Rudin Chapter 12 Exercise 7.
Proof. Suppose that Ω = {x + iy | a < x < b, y ∈ R} and M (a) = 0. We have to show that
M (x) = 0 for all x ∈ (a, b). Consider the function g(x + iy) = f (y + a + ix) which is defined
in the horizontal strip Ω+
g = {x + iy | −∞ < x < ∞ and 0 < y < b − a}. By the definition, we
know that
|g(x)| = |f (a + ix)| = 0
for all x ∈ R. In particular, g is real on the segment L = (0, 1). By Theorem 11.14 (The
Schwarz Reflection Principle), there exists a function G holomorphic in Π = Ω+ −
g ∪ L ∪ Ωg such
that G(x) = g(x) = 0 for every x ∈ L. By Theorem 10.18, we have G(z) = 0 in Π. Since
G(z) = g(z) = 0 in Ω+
g , we obtain
f (z) = 0
in Ω which implies the required result that M (x) = 0 for all x ∈ (a, b) and hence Theorem 12.8
(The Hadamard’s Three-Line Theorem) is also true if M (a) = 0. This completes the proof of
the problem.
12.1. Applications of the Maximum Modulus Principle 387
Problem 12.8
Rudin Chapter 12 Exercise 8.
Proof. Suppose that Ω = {z = x + iy | c < x < d and y ∈ R} for some −∞ < c < d < ∞. By
the hypothesis, the exponential function ζ = ez maps Ω onto A(R1 , R2 ). We are given that
R1 < a < r < b < R2 . Then there exist c < α < β < d such that ζ = ez maps the closed strip
Ω′ = {z = x + iy | α ≤ x ≤ β and y ∈ R} onto the closed annulus A(a, b). Thus we have
eα = a and eβ = b. (12.11)
We define F : Ω′ → C by
F (z) = f (ez ) = f (ζ)
which is continuous on Ω′ and F ∈ H(Ω′ ). By Theorem 10.24 (The Maximum Modulus Theorem)
and the Extreme Value Theorem, there is a positive constant B > 0 such that |F (z)| < B for
all z ∈ Ω′ . In other words, we may apply Theorem 12.8 (The Hadamard’s Three-Line Theorem)
to our function F to get
M (x)β−α ≤ M (α)β−x × M (β)x−α , (12.12)
where the two expressions (12.11) give
b b r
log log M (r) = log log M (a) + log log M (b)
a r a
b r r
= log − log · [log M (a)] + log log M (b)
a a a
b r M (b)
= log log M (a) + log log
a a M (a)
log ar M (b)
log M (r) = log M (a) + log . (12.14)
log ab M (a)
M (a)
If we denote λ = (log ab )−1 log M (b) ∈ R, then the expression (12.14) can be further simplified to
r
log M (r) = log M (a) − λ log
a
388 Chapter 12. The Maximum Modulus Principle
a λ
M (r) = M (a). (12.15)
r
If we combine the Extreme Value Theorem and the expression (12.15), then one can show that
there exists an |ζ0 | = r ∈ (a, b) such that |f (ζ0 )| = M (r) = ( ar )λ M (a) which can be rewritten as
Notice that ζ λ f (ζ) may not be holomorphic because λ may not be an integer. However, remem-
ber that ζ = ez , so we can express the expression (12.16) as
where ζ0 = ez0 ∈ Ω′ . Now eλz f (ez ) ∈ H(Ω′ ) and continuous on Ω′ , so we may apply Theorem
12.4 (The Maximum Modulus Theorem) to this function to conclude that
in Ω′ , where c is some constant. Since Ω′ is a region, Theorem 10.18 asserts that the result
(12.17) also holds in Ω. Using the substitution ζ = ez again, the result (12.17) becomes
f (ζ) = cζ −λ
in A(R1 , R2 ). Recall that f ∈ H A(R1 , R2 ) , it forces that λ ∈ Z. This ends the analysis of the
problem.
Problem 12.9
Rudin Chapter 12 Exercise 9.
ϕ(z) = ez (12.19)
Choose B such that B − 1 ≥ log A. Note that eα|x| ≥ 1 for all x ∈ R. Then it is easy to
see that
log A ≤ (B − 1)eα|x|
log A + eα|x| ≤ Beα|x|
A exp(eα|x| ) ≤ exp(Beα|x| ). (12.21)
• The modified result. Suppose that ∆ is an open sector between two rays from the origin
with sectoral angle βπ < π for some β > 1. Suppose that f is continuous on ∆, f ∈ H(∆)
and there are constants A < ∞ and α ∈ (0, β) such that
is clearly an isomorphism such that φ maps the boundary of ∆ onto the boundary of Π.
Since 0 ∈/ ∆, we can define a branch for log z so that φ ∈ H(∆). By Theorem 10.33, we
have φ−1 ∈ H(Π). Next, the map F : Π → C defined by
F = f ◦ φ−1
1 1 1
is continuous on Π and F ∈ H(Π). Since φ−1 (ζ) = eiθ i β ζ β , we have |φ−1 (ζ)| = |ζ| β and
thus the hypothesis (12.22) implies
α
|F (ζ)| = |f φ−1 (ζ) | < A exp(|φ−1 (ζ)|α ) = A exp(|ζ| β ),
α
where β < 1. If ζ = iy for some y ∈ R, then since φ−1 (iy) lies on ∂∆, we get
|F (iy)| = |f φ−1 (ζ) | ≤ 1.
Hence we establish from our first assertion that |F (ζ)| ≤ 1 in Π or equivalently, |f (z)| ≤ 1
in ∆.
Problem 12.10
Rudin Chapter 12 Exercise 10.
Proof. For each n = 1, 2, 3, . . ., we define gn (z) = f (z)enz . Suppose that Lα = {z = reiα | r ≥ 0},
n π o n πo
∆1 = z ∈ Π − < arg z < α and ∆2 = z ∈ Π α < arg z < .
2 2
π π
Certainly, ∆1 and ∆2 have the sectoral angles α + 2 < π and 2 − α < π respectively. Refer to
Figure 12.2:
In order to use Problem 12.9 as stated in the hint, it is necessary to assume that f is
continuous on the closure of Π. Since f ∈ H(Π) and ∆1 , ∆2 are proper subsets of Π, each gn is
2π
continuous on ∆j and gn ∈ H(∆j ) for j = 1, 2. Choose an γ ∈ (0, 2α+π ) and let Fn (x) = nx − xγ
for x ≥ 0. By elementary calculus, the Fn attains its maximum value
n 1 γ − 1
γ−1
Mn = n
γ γ
1
at x = (nγ −1 ) γ−1 > 0. Therefore, if we pick An = 1 + eMn , then we have
exp(n|z| − |z|γ ) = eFn (|z|) ≤ eMn < An (12.23)
for all z ∈ ∆1 . Since |f (z)| < 1 for all z ∈ Π, for each fixed n ∈ N, we note from the inequality
(12.23) that
|gn (z)| < exp(n|z|) < An exp(|z|γ )
for all z ∈ ∆1 . Consequently, each gn satisfies the inequality (12.22). Furthermore, we observe
from the definition of gn that
log |gn (reiα )| log |f (reiα )|
= + n cos α → −∞
r r
as r → ∞, so |gn (z)| is bounded on Lα . Without loss of generality, we may assume that the
bound is 1. Obviously, we know from the additional assumption that
π π
|gn re±i 2 | = |f re±i 2 | < 1 (12.24)
12.1. Applications of the Maximum Modulus Principle 391
for all r ≥ 0. In other words, gn is bounded by 1 on ∂∆1 . By the modified result of Problem
12.9, we conclude that each gn is bounded by 1 in ∆1 . Similarly, each gn is also bounded by
1 on ∂∆2 and then in ∆2 . Since Π = ∆1 ∪ Lα ∪ ∆2 , what we have shown is that each gn is
bounded by 1 in Π, so for every z ∈ Π, this implies that
for all n ∈ N which means f (z) = 0 because cos θ > 0 for θ ∈ (− π2 , π2 ), i.e., f = 0 as required.
This completes the proof of the problem.
Problem 12.11
Rudin Chapter 12 Exercise 11.
Proof. Since the result is trivial if f is constant, we assume that f is nonconstant in the following
discussion. Besides, without loss of generality, we may assume that |f (z)| ≤ 1 on Γ = ∂Ω. It
suffices to prove that
|f (ω)| ≤ 1 (12.25)
for every ω ∈ Ω. We choose a ∈ Ω and consider
f (z) − f (a)
fe(z) = .
z−a
Using Theorem 10.16 and then Theorem 10.6, we see that fe ∈ H(Ω). Furthermore, the continuity
of f on Ω ∪ Γ certainly implies that fb is also continuous on Ω ∪ Γ. Now the boundedness of f
guarantees that fe(z) → 0 as z → ∞. In other words, there is a positive constant C such that
|fe(z)| ≤ C (12.26)
in Ω ∪ Γ. Next, let ΩR = D(0; R) ∩ Ω ⊆ Ω for R > 0 and Fe(z) = f N (z)fe(z) for some N ∈ N.
Clearly, we have
Fe ∈ H(Ω).
By the boundedness of f and the fact fe(z) → 0 as z → ∞, we can find R large enough such
that ω ∈ ΩR and |Fe(z)| ≤ C for all z ∈ ∂ΩR = C(0; R) ∩ Γ ⊆ Γ from the bound (12.26). By
Theorem 10.24 (The Maximum Modulus Theorem) and the Extreme Value Theorem, we assert
that
|Fe (ω)| ≤ C. (12.27)
e
If f (ω) 6= 0, then we deduce from the inequality (12.27) that
1
CN
|f (ω)| ≤ 1 . (12.28)
e
|f (ω)| N
Taking N → ∞ in the inequality (12.28) which yields the required result (12.25) immediately.
Consequently, the inequality
|f (ω)| ≤ 1
holds for all ω ∈ Ω \ Zfe.
Assume that Zfe had a limit point in Ω. Now Theorem 10.18 ensures that fe ≡ 0 in Ω and then
f (z) = f (a) for all z ∈ Ω, a contradiction to our hypothesis that f is nonconstant. Therefore,
Zfe is discrete and hence the continuity of f forces definitely that the inequality (12.25) remains
valid in Ω which completes the proof of the problem.
392 Chapter 12. The Maximum Modulus Principle
Problem 12.12
Rudin Chapter 12 Exercise 12.
F1 ⊇ F2 ⊇ · · · . (12.29)
For each k = 1, 2, . . ., pick zk ∈ Fk . Now the sequence (12.29) ensures that zn ∈ Fk for all
n ≥ k. Using [11, Proposition 1.7, p. 14], one can find a continuous mapping γk : [k − 1, k] → Fk
connecting zk and zk+1 . We note that the definition
of Ek asserts that |f (z)| > k for all z ∈ Fk , so
particularly, it is also true on γk , i.e., |f γ(t) | > k for every t ∈ [k −1, k]. Define γ : [0, ∞) → F1
by
γ = γ1 +̇ γ2 +̇ · · · .
t
e(t) = γ( 1−t
Then it is easy to see that f (γ(t)) → ∞ as t → ∞. Finally, if we set γ e is a
), then γ
well-defined continuous function on [0, 1) and satisfies
lim f (e
γ (t)) = ∞,
t→1
Problem 12.13
Rudin Chapter 12 Exercise 13.
Problem 12.14
Rudin Chapter 12 Exercise 14.
Proof. Suppose that f is nonconstant. Since f (z) 6= α for all z ∈ C, the function
1
F (z) =
f (z) − α
is nonconstant entire. Now Problem 12.12 implies that F has ∞ as an asymptotic value which
means that α is an asymptotic value of f . This completes the proof of the problem.
Problem 12.15
Rudin Chapter 12 Exercise 15.
Proof. The case is trivial if f is constant. So we may assume that f is nonconstant. Suppose
first that Z(f ) = ∅. Then we have f1 ∈ H(U ). For every n ∈ N, we have 0 < 1 − n1 < 1. By
combining Theorem 10.24 (The Maximum Modulus Theorem) and the Extreme Value Theorem,
we see that
1 1
≤ max
|f (0)| z∈C(0;1− n1 ) |f (z)|
or equivalently,
|f (0)| ≥ min |f (z)|.
1
z∈C(0;1− n )
We simply take zn = 1 − n1 which gives |f (zn )| ≤ |f (0)| for all n ∈ N. Obviously, |zn | → 1 as
n → ∞, so our assertion is true in this case.
Next, we suppose that Z(f ) 6= ∅. Then there are two cases.
394 Chapter 12. The Maximum Modulus Principle
• Case (i): Z(f ) is infinite. Since Z(f ) ⊆ U which is bounded, the Bolzano-Weierstrass
Theorem [127, Problem 5.25, pp. 68, 69] ensures that Z(f ) has a convergent subsequence
{ζk }. Now Theorem 10.18 forces that ζk → ζ ∈ C(0, 1). Since f (ζk ) = 0 for all k ∈ N, the
assertion remains true in this case.
• Case (ii): Z(f ) is finite. Suppose that Z(f ) = {ζ1 , ζ2 , . . . , ζN } for some N ∈ N. Suppose
further that mk is the order of zero of f at ζk , where 1 ≤ k ≤ N . Consider
f (z)
g(z) = N
. (12.30)
Y
mk
(z − ζk )
k=1
Therefore, we know that g ∈ H(U ) and Z(g) = ∅. Thus the special case implies that there
is a sequence {zn } ⊆ U and a positive constant M such that |zn | → 1 and |g(zn )| ≤ M for
all n ∈ N. Hence it follows from the representation (12.30) that
N
Y N
Y
|f (zn )| ≤ M |zn − ζk |mk ≤ M (1 + |ζk |) < ∞
k=1 k=1
for all n ∈ N.
Problem 12.16
Rudin Chapter 12 Exercise 16.
Proof. The result is always true if f is a constant function. Without loss of generality, we may
assume that f is nonconstant. Let α = sup{|f (z)| | z ∈ Ω}. If |f (ζ)| = α for some ζ ∈ Ω, then
f is constant by Theorem 10.24 (The Maximum Modulus Theorem), a contradiction. Thus we
always have
|f (z)| < α (12.31)
for all z ∈ Ω. By the definition, there is a sequence {zn } ⊆ Ω such that |f (zn )| → α as n → ∞.
Since Ω is bounded, the Bolzano-Weierstrass Theorem ensures that {zn } contains a convergent
subsequence {znk }. Let znk → z0 . If z0 ∈ Ω, then α = |f (z0 )| which is impossible. Therefore,
we must have z0 ∈ ∂Ω and our hypothesis gives α ≤ M . By the inequality (12.31), we conclude
that
|f (z)| < α ≤ M
for all z ∈ Ω. This ends the proof of the analysis.
Problem 12.17
Rudin Chapter 12 Exercise 17.
• The value of M . Without loss of generality, we may assume that f (0) > 0. Otherwise,
we can consider the function fb = eiθ f , where θ = − arg f (0). Then fb(0) = |f (0)| > 0 and
fb ∈ Φ.
Let Ω− = {z ∈ C | Re z < 0}. Since 0 < |f (z)| < 1 for z ∈ U , the mapping f1 = log f
maps U into Ω− . Next, the mapping f2 (z) = −iz clearly maps Ω− onto the upper half
plane Π+ . Finally, for Im α > 0, we know that the mapping
z−α
f3 (z) =
z−α
maps Π+ into U . Hence the mapping F = f3 ◦ f2 ◦ f1 maps U into U . Since f1 ∈ H(U ),
f2 ∈ H(Ω− ) and f3 ∈ H(Π+ ), it is true that F ∈ H(U ). Clearly, the definition of F
implies that F ∈ H ∞ , kF k∞ ≤ 1 and
−i log f (0) − α
F (0) = . (12.32)
−i log f (0) − α
If we take α = −i log f (0), then Im α = − log f (0) > 0 because 0 < f (0) < 1. Thus the
expression (12.32) gives F (0) = 0 in this case. By Theorem 12.2 (Schwarz’s Lemma), we
have |F (z)| ≤ |z| for all z ∈ U and |F ′ (0)| ≤ 1.
Now the explicit formula of F is given by
(z)
−i log f (z) + i log f (0) log ff (0)
F (z) = =
−i log f (z) − i log f (0) log[f (0)f (z)]
so that
f ′ (z) log f (0)2
F ′ (z) = · . (12.33)
f (z) {log[f (0)f (z)]}2
Since |F ′ (0)| ≤ 1, we see from the formula (12.33) that
Elementary calculus shows that the function g : (0, 1) → R defined by g(x) = x log x
attains its absolute minimum −e−1 at x = e−1 . Therefore, we have M ≤ 2e−1 .
Next, we claim that M = 2e−1 . To see this, we consider the function f (z) = e−2z−1 which
is holomorphic in U . Since e−2 < |e2z | = e2r cos θ < e2 for all z = reiθ ∈ U , it is easy to see
that
0 < e−3 < |f (z)| < e−1 < 1
in U so that f ∈ Φ. As f ′ (z) = −2e−2z−1 , we have |f ′ (0)| = 2e−1 = M as required.
• The value of M (c). Since f (0) = c, it follows from the inequality (12.34) that
2c| log c|, if c < e−1 ;
M (c) =
−1
2e , if e−1 ≤ c < 1.
Remark 12.1
The first assertion of Problem 12.17 is called Rogosinski’s Theorem. See, for instances,
[94] and [23, Exercise 6.36, pp. 213, 214] for a different proof.
396 Chapter 12. The Maximum Modulus Principle
CHAPTER 13
Approximations by Rational Functions
Problem 13.1
Rudin Chapter 13 Exercise 1.
then this expression implies that P (z) is entire and has at most a pole at ∞. If ∞ is not a pole
of f , then P is a constant by Theorem 10.23 (Liouville’s Theorem). Otherwise, the function
P ( z1 ) has a pole at 0. If P (z) = c0 + c1 z + c2 z 2 + · · · , then we have P ( 1z ) = c0 + cz1 + zc22 + · · ·
and the nature of its singularity at 0 implies that cp = cp+1 = · · · = 0 for some p ∈ N. In other
words, P must be a polynomial. Since
P (z)
f (z) = N
,
Y
mk
(z − ak )
k=1
Problem 13.2
Rudin Chapter 13 Exercise 2.
Proof.
397
398 Chapter 13. Approximations by Rational Functions
which belongs to H(Ω). Assume that there was a sequence {Pn } of polynomials such that
Pn → f uniformly in Ω. Take γ = C(− 34 ; 14 ). Then we have γ ⊆ Ω, see Figure 13.1 again.
On the one hand, we have
Z 1
f (z) dz = 2πi · Ind γ = 2πi.
γ 2
Furthermore, Theorem 10.12 gives
Z
Pn (z) dz = 0
γ
for every n ∈ N. On the other hand, the uniform convergence shows that
Z Z
0 = lim Pn (z) dz = f (z) dz = 2πi,
n→∞ γ γ
Problem 13.3
Rudin Chapter 13 Exercise 3.
Proof. For every n ∈ N, let Dn = {z ∈ D(0; n) | |Im z| ≥ n1 } and En = [ n1 , n] × {0}, see Figure
13.2. It is clear that both Dn and En are compact, so the set Kn = Dn ∪ En ∪ {0} is compact
1
too. Furthermore, we note that S 2 \ Kn is connected. Take 0 < δn < 2n . Then the sets
n 1 o 1
Dn′ = z ∈ D(0; n + δn ) |Im z| ≥ − δn and En′ = − δn , n + δn × (−δn , δn )
n n
13.1. Meromorphic Functions on S 2 and Applications of Runge’s Theorem 399
are open sets containing Dn and En respectively. Obviously, the set Dn′ ∪ En′ is open in C and
is disjoint from Un = (−δn , δn ) × (−δn , δn ). Define the function fn : Ωn = Dn′ ∪ En′ ∪ Un → C by
1, if z ∈ Un ;
fn (z) =
0, if z ∈ Dn′ ∪ En′ .
Then we have fn ∈ H(Ωn ) and Ωn is an open set containing Kn . According to Theorem 13.7,
one can find a polynomial Qn such that |Qn (z) − fn (z)| < n1 for all z ∈ Kn . In fact, we get
|Qn (z) − 1|, if z ∈ Un ;
|Qn (z) − fn (z)| = (13.2)
|Qn (z)|, if z ∈ Dn′ ∪ En′ .
If we define Pn (z) = Qn (z) − Qn (0) + 1, then the definition (13.2) implies immediately that
Pn (0) = Qn (0) − Qn (0) + 1 = 1 for n = 1, 2, . . .. Since we have
∞
[
C \ {0} = (Dn′ ∪ En′ ),
n=1
if z 6= 0, then there exists an N ∈ N such that z ∈ Dn′ ∪ En′ for all n ≥ N and thus the definition
(13.2) implies that
2
|Pn (z)| = |Qn (z) − Qn (0) + 1| ≤ |Qn (z)| + |Qn (0) − 1| <
n
for all n ≥ N . Consequently, we conclude from this that Pn (z) → 0 as n → ∞, completing the
proof of the problem.
400 Chapter 13. Approximations by Rational Functions
Problem 13.4
Rudin Chapter 13 Exercise 4.
Proof. For every n ∈ N, we consider the three sets An = [−n, n] × [ n1 , n], Bn = [−n, n] × {0}
and Cn = [−n, n] × [− n1 , −n], see Figure 13.3 for an illustration. Let
Kn = An ∪ Bn ∪ Cn .
1
Obviously, each Kn is compact and the set S 2 \ Kn is connected. Choose 0 < δn < 2n . Then
′ 1 ′
the three sets An = (−n − δn , n + δn ) × ( n − δn , n + δn ), Bn = (−n − δn , n + δn ) × (−δn , δn ) and
Cn′ = (−n − δn , n + δn ) × (− n1 − δn , n + δn ) are open sets containing An , Bn and Cn respectively.
Besides, they are disjoint and Kn ⊆ Ωn = A′n ∪ Bn′ ∪ Cn′ . Define
1, if z ∈ A′n ;
fn (z) = 0, if z ∈ Bn′ ;
−1, if z ∈ Cn′ .
Since fn ∈ H(Ωn ), it follows from Theorem 13.7 that there exists a polynomial Pn such that
|Pn (z) − fn (z)| < n1 for all z ∈ Kn . In fact, we have
|Pn (z) − 1|, if
z ∈ An ;
|Pn (z) − fn (z)| = |Pn (z)|, if z ∈ Bn ;
|Pn (z) + 1|, if z ∈ Cn .
Therefore, such sequence {Pn } of polynomials satisfy the requirements of the problem. This
completes the proof of the problem.
13.2. Holomorphic Functions in the Unit Disc without Radial Limits 401
Problem 13.5
Rudin Chapter 13 Exercise 5.
1 1
Proof. For each n ∈ N, suppose that ∆n = D(0; 1 − 2n ), Cn = {(1 − 2n+1 )eiθ | π2 ≤ θ ≤ 2π},
1 1 1 iθ π
Dn = [1 − 2n+1 , 1 − 2n+2 ] and En = {(1 − 2n+2 )e | 0 ≤ θ ≤ 2 }. Then the union
Ln = Cn ∪ Dn ∪ En
is an arc in U \ ∆n with the property that each Ln intersects every radius of U . Finally, we
1
suppose that Ωn is a tubular region of Ln with width 2(2n+2) 4 . The construction ensures that
Ωn ∩ ∆n = ∅, see Figure 13.4 for an illustration. Similar to Problem 13.4, we can select very
1
small δn > 0 such that the modified tubes Ω′n of Ln with widths 2(2n+2) 4 + δn and Πn =
1
D(0; 1 − 2n + δn ) also satisfy Ω′n ∩ Πn = ∅. Notice that Ω′n ⊆ Ωn .
We apply induction to construct the sequence of polynomials {Qn } and a holomorphic func-
tion f such that Qn → f uniformly on U : Consider Q0 ≡ 0 and
Q0 (z), if z ∈ Π◦1 ;
f1 (z) =
1, if z ∈ Ω′1 .
Obviously, we have f1 ∈ H(Π◦1 ∪ Ω′1 ). Since C \ (Π◦1 ∪ Ω′1 ) is connected, it follows from Theorem
13.9 (Runge’s Theorem) that one can find a polynomial Q1 such that
1
|Q1 (z) − f1 (z)| < (13.3)
2
402 Chapter 13. Approximations by Rational Functions
on compact subsets of Π◦1 ∪Ω′1 . Particularly, the estimate (13.3) holds for all z ∈ ∆1 ∪Ω1 because
∆1 ⊆ Π◦1 and Ω1 ⊆ Ω′1 .
Assume that the polynomial Qn is chosen with the property that
1
|Qn (z) − fn (z)| < (13.4)
2n
on ∆n ∪ Ωn . We consider the function defined by
Q (z), if z ∈ Π◦n+1 ;
n
fn+1 (z) = n (13.5)
1 + (−1) , if z ∈ Ω′ .
n+1
2
Then we have fn+1 ∈ H(Π◦n+1 ∩ Ω′n+1 ). Since C \ (Π◦n+1 ∩ Ω′n+1 ) is connected, Theorem 13.9
(Runge’s Theorem) guarantees that one can find a polynomial Qn+1 such that
1
|Qn+1 (z) − fn+1 (z)| < (13.6)
2n+1
on compact subsets of Π◦n+1 ∩ Ω′n+1 and thus on ∆n+1 ∪ Ωn+1 because ∆n+1 ⊆ Π◦n+1 and
Ωn+1 ⊆ Ω′n+1 . Consequently, we have constructed the polynomial Qn+1 and then a sequence of
polynomials {Qn } satisfying the property (13.4).
Next, let N be a positive integer. Then we have ∆N ⊆ ∆n for all n ≥ N . Therefore, it
deduces from the definition (13.5) and the inequality (13.6) that
1
|Qn+1 (z) − Qn (z)| < (13.7)
2n+1
holds for all z ∈ ∆N ⊆ ∆n ⊆ ∆n+1 and all n ≥ N . Given ǫ > 0. We pick the N large enough so
that 21N < ǫ. For all positive integers p and n ≥ N , we observe from the inequality (13.7) that
|Qn+p (z) − Qn (z)| ≤ |Qn+p (z) − Qn+p−1 (z)| + · · · + |Qn+1 (z) − Qn (z)|
1 1 1
< n+p + n+p−1 + · · · + n+1
2 2 2
1
< n
2
<ǫ
for all z ∈ ∆N . Hence it follows from the Cauchy Criterion for Uniform Convergence (see [99,
Theorem 7.8, p. 147]) that the sequence of polynomials {Qn } converges uniformly to a function
f on ∆N . By Theorem 10.28, we have f ∈ H(∆◦N ). Since it is true for every large N and
∞
[
U= ∆N ,
N =1
∞
X
Therefore, we have constructed a sequence of polynomials {Pn } such that the series f = Pn
n=1
is holomorphic in U .
13.2. Holomorphic Functions in the Unit Disc without Radial Limits 403
1
The above construction of {Qn } starts with the inequality (13.3), where 2 can be actually
replaced by any small δ > 0 so that
on ∆n for every n = 2, 3, . . .. This means that the polynomials Pn are very small on ∆n .
Furthermore, we can also replace the values 0 and 1 by another Qn + g ∈ H(Ω′n+1 ) in the
definition (13.5), and hence Qn + g ∈ H(Ωn+1 ). As a result, we obtain from the inequality
(13.6) that
1
|Pn+1 (z) − g(z)| < n+1
2
in Ωn+1 . In other words, it means that the sequence {Pn } can be chosen more or less arbitrary
on Ln , i.e., any holomorphic g can be approximated by the sequence of polynomials {Pn } in a
neighborhood of Ln .
Now it remains to show that f has no radial limit at any point of T . To this end, by the
definitions of ∆n and Ωn , we see easily that Ω2n ⊆ ∆p for all p ≥ 2n + 2. On the one hand,
1
suppose that z ∈ Ω2n ⊆ Ω′2n . Then we recall from the definition (13.5) that |Q2n (z)| < 22n .
Combining this fact, Qn → f in U and the inequality (13.7), we know that
which are contrary. Hence this contradiction means that our f has no radial limit on T , com-
pleting the analysis of the problem.
404 Chapter 13. Approximations by Rational Functions
Problem 13.6
Rudin Chapter 13 Exercise 6.
k
lim sup 5 nk = 1.
k→∞
Hence it asserts from §10.5 that the radius of convergence of the series of h is 1, as required.
• There is a constant c > 0 such that |h(zm )| > c · 5m for all z with |zm | = 1 − n1m
and m ≥ 3. We remark that the hint is not true for the cases m = 1, 2. For example, take
n1 = 4 and n2 = 9, so we have
3 ∞
X
h eiθ = 5(0.75)4 e4iθ + 25(0.75)9 e9iθ + 5k (0.75)nk eiθnk .
4
k=3
Since 0.754 < 0.32 and 5(0.75)9 > 0.375, the first term is not the dominant term in the
series defining h(z) which means the hint is not correct anymore. Examples for the case
m = 2 can be found similarly. However, we can show affirmative result if m ≥ 3. Suppose
that zm = (1 − n1m )eiθ .
5k 1 nk iθnk 1 nm iθnm
1 m−1
X
h (1 − nm eiθ )
= 1 − e + 1 − e
5m 5m nm nm
k=1
5k 1 nk iθnk
∞
X
+ m
1− e
5 nm
k=m+1
h m−1
X 5k 1 nk iθnk i
= 1 − e − S m−1 + (Sm−1 + am )
5m nm
k=1
5k 1 nk iθnk
∞
X
+ m
1− e ,
5 nm
k=m+1
where
m−1
X 5k iθnk 1 nm iθnm
Sm−1 = e and am = 1 − e
5m nm
k=1
for every m = 3, 4, . . .. By elementary calculus, we can show easily the following result:
13.2. Holomorphic Functions in the Unit Disc without Radial Limits 405
Lemma 13.1
If α ≥ 1, then the function 1 αx
fα (x) = 1 −
x
is strictly increasing and 0 < fα (x) ≤ e−α on [1, ∞). In addition, fα (x) → e−α as
x → ∞.
Particularly, Lemma 13.1 says that {|am |} is a strictly increasing sequence of positive
numbers such that
1 1
≤ |am | ≤ (13.10)
4 e
for all m = 1, 2, . . . and |am | → e−1 as m → ∞. In fact, if m ≥ 3, then |am | > 0.35. Simple
algebra gives
1 1
|Sm−1 | ≤ × 1 − m−1
4 5
for all m = 3, 4, . . .. Thus it is true that
for all m = 3, 4, . . ..
Next, we know that
m−1 1 nk iθnk m−1 h 1 nk i
X 5k X 5k
1 − e − S m−1 ≤ 1 − 1 − . (13.12)
5m nm 5m nm
k=1 k=1
for every m = 3, 4, . . ..
Clearly, we deduce from the upper bound (13.10) that
X
∞
5k 1 nk iθnk X 5k
∞ nk
5
∞
1 X 5k
nm
1 − e = · a ≤ n + nk . (13.14)
5m nm 5m k m+1
5m
e nm k=m+2 e m
n
k=m+1 k=m+1
406 Chapter 13. Approximations by Rational Functions
nm+1
As m ≥ 3, we get nm ≥ 6 which implies that
5 5
nm+1 ≤ < 0.0124. (13.15)
e nm e6
for every m ≥ 3. Finally, by combining the bounds (13.11), (13.13), (13.15) and (13.16),
we conclude immediately that
h (1 − 1 )eiθ m−1 1 nk iθnk
nm X 5k
≥ |S m−1 + a m | − 1 − e − S m−1
5m 5m nm
k=1
X
∞
5k 1 nk iθnk
− 1 − e
5m nm
k=m+1
≥ 0.11 − 0.06 − 0.0124 − 0.026
> 0.
|h(z)| > C · 5m
1
holds for every z with |z| = 1 − nm and m ≥ 3.
1 iθ
• Proof that h has no finite radial limits. If zm = (1 − nm )e , then the previous
assertion shows that
1 iθ
h 1 − e > C · 5m
nm
for every m ≥ 3 and θ ∈ [0, 2π]. Therefore, it guarantees that
1 iθ
lim h 1 − e = ∞.
m→∞ nm
In other words, h has no finite radial limits.
• The h has infinitely many zeros in U . Assume that h had only finitely many zeros
α1 , α2 , . . . , αp in U . Then the function
p
Y p
Y z − αk
ϕ(z) = z m ϕαk (z) = z m (13.17)
1 − αk z
k=1 k=1
has exactly the same zeros as h counted with multiplicity, where m is the order of zero of
h at the origin. If h has no zero in U , then we let ϕ = 1. Now the function f = ϕh satisfies
f ∈ H(U ). By the proof of Theorem 12.4, we know that |ϕαk (z)| < 1 if |z| < 1, so the
definition (13.17) implies that
|f (z)| > C · 5m (13.18)
13.3. Simply Connectedness and Miscellaneous Problems 407
for every z with |z| = 1− n1m and m ≥ 3. Furthermore, f has no zero in U so that f1 ∈ H(U ).
Combining this fact, the inequality (13.18) and Theorem 10.24 (The Maximum Modulus
Theorem), we see immediately that
1 1
<
f (z) C · 5m
C
|b
h(z)| = |h(z) − α| ≥ |h(z)| − |α| > C · 5m − |α| > · 5m
2
for |z| = 1 − n1m . Therefore, we can apply similar argument as above to obtain the desired
result.
Remark 13.1
(a) A sequence {nk } of positive integers is said to be lacunary if there is a constant c > 1
such that nk+1 > cnk for all k ∈ N. A power series
∞
X
ak z n k (13.19)
k=1
is called a lacunary power series or a power series with Hadamard gaps. Thus
our h is an example of this type of power series with c = 2. See, for instance, [133,
Chap. V].
(b) In [52, Problem 5.36 & Update 5.36, p. 113], it is pointed out that the best known
result concerning the number of zeros of a lacunary power series inside U is due to
Chang [26], who proved that if
∞
X
|ak |2+ǫ = ∞
k=0
for some ǫ > 0, then the series (13.19) has infinitely many zeros in any sector. See also
[44], [75] and [121]
Problem 13.7
Rudin Chapter 13 Exercise 7.
Proof. Suppose that A intersects each component of S 2 \ Ω. Choose a sequence of compact sets
Kn in Ω with the properties specified in Theorem 13.3. Fix a positive integer n. Let V be a
408 Chapter 13. Approximations by Rational Functions
Problem 13.8
Rudin Chapter 13 Exercise 8.
where n = 1, 2, . . .. Since An is finite, each Qn is a rational function and the poles of Qn lie
in An for n ≥ 2. In particular, Qn is holomorphic in an open set V containing Kn−1 . By the
known fact given in Definition 10.5, the power series of Qn at 0 converges uniformly to Qn in
Kn−1 . This means that for each n = 2, 3, . . ., there exists a polynomial Rn such that
1
|Rn (z) − Qn (z)| < (13.22)
2n
for all z ∈ Kn−1 .
By imitating the remaining part of the proof of Theorem 13.10 (The Mittag-Leffler Theorem),
it can be seen that the function
X∞
f (z) = Q1 (z) + [Qn (z) − Rn (z)],
n=2
where z ∈ C, has the desired properties. In fact, we fix a positive integer N first. On KN , we
have
XN X ∞
f = Q1 + (Qn − Rn ) + (Qn − Rn ). (13.23)
n=2 n=N +1
Using the inequality (13.22), each term in the last sum in the expression (13.23) is less than 2−n
on KN , hence this last series converges uniformly on KN to a function gN +1 which is holomorphic
◦ . Since each R is a polynomial, the function
in KN n
N
X
f − (Q1 + Q2 + · · · + QN ) = gN +1 − Rn
n=2
is holomorphic in KN ◦ and therefore, f has precisely the prescribed principal parts in K ◦ . Since
N
N is arbitrary, it is actually true in C. This completes the analysis of the problem.
13.3. Simply Connectedness and Miscellaneous Problems 409
Problem 13.9
Rudin Chapter 13 Exercise 9.
Proof. As f (z) 6= 0 for all z ∈ Ω, we have f1 ∈ H(Ω). Since Ω is simply connected, Theorem
13.11 says that there exists an h ∈ H(Ω) such that
f = eh .
g n = eh = f
Problem 13.10
Rudin Chapter 13 Exercise 10.
Proof. We want to show that there exists a g ∈ H(Ω) such that f = exp g if and only if there
exists a ϕn ∈ H(Ω) such that f = ϕnn for every positive integer n.
Suppose that there exists a g ∈ H(Ω) such that f = exp g. Then the function ϕn = exp( ng )
satisfies ϕn ∈ H(Ω) and ϕnn = exp g = f for every positive integer n. Conversely, suppose that
there exists a ϕn ∈ H(Ω) such that f = ϕnn for every n = 1, 2, . . .. We claim that f (z) 6= 0 for
all z ∈ Ω. Assume that there was an a ∈ Ω such that f (a) = 0. Since f 6≡ 0, Theorem 10.18
asserts that there is a unique positive integer m such that
for some g ∈ H(Ω) and g(a) 6= 0. Particularly, we take n = m + 1. Since ϕm+1 (a) = 0, Theorem
10.18 ensures that
ϕm+1 (z) = (z − a)φm+1 (z) (13.25)
for some φ ∈ H(Ω). If we combine the representations (13.24) and (13.25), then we conclude
that
g(z) = (z − a)φm+1
m+1 (z)
for all z ∈ Ω, but this implies that g(a) = 0, a contradiction which proves our claim. In other
words, we have f1 ∈ H(Ω). Next, we may take n = 2 so that f = ϕ22 for some ϕ2 ∈ H(Ω). Hence
it follows from Theorem 13.11 that f has a holomorphic logarithm g in Ω and thus we complete
the analysis of the problem.
Problem 13.11
Rudin Chapter 13 Exercise 11.
Proof. Put ϕ(z) = sup |fn (z)| in Ω. Now ϕ is well-defined because fn → f pointwise in Ω.
n∈N
Suppose that U is a nonempty open subset of Ω with U ⊆ Ω. Such a set exists because Ω is an
open set. For k = 1, 2, . . ., we define
By the hypothesis, we know that fn (z) → f (z) for every z ∈ U , the set {fn (z) | n ∈ N} must be
bounded for each fixed z. Thus each z ∈ U lies in some Vk , i.e.,
∞
[
U= Vk .
k=1
Since U is a compact subset of the metric space C, it is a complete metric space. By Theorem
5.6 (Baire’s Theorem) (see also §5.7), it is impossible that all Vk are nowhere dense sets. Thus
there exists an N ∈ N such that the closure VN contains a nonempty disc DU of U , so it yields
from the definition (13.26) that
|fn (z)| ≤ N
for all n ∈ N and z ∈ DU . Equivalently, it means that ϕ(z) is bounded on DU and we may apply
Problem 10.5 to conclude that fn → f uniformly on every compact subset of DU . According to
Theorem 10.28, we have f ∈ H(DU ).
Now we let [
V = DU ,
U
where U is any arbitrary nonempty open subset of Ω with U ⊆ Ω. Obviously, we observe that
f ∈ H(V ). If D(z0 ; δ) ⊂ Ω and D(z0 ; δ) ∩ V = ∅ for some z0 ∈ Ω and δ > 0,a then the above
argument shows that there exists a nonempty open subset D of D(z0 ; δ) such that f ∈ H(D).
Therefore, D ⊆ V which implies the contradiction D(z0 ; δ) ∩ V 6= ∅. As a result, it means that
V is a dense open subset of Ω which ends the proof of the problem.
Remark 13.2
The result of Problem 13.11 is sometimes called Osgood’s Theorem [83]. In fact, Problem
10.5 is the well-known Vitali Convergence Theorem, see [117, p. 168].
Problem 13.12
Rudin Chapter 13 Exercise 12.
a
Without loss of generality, we may assume further that D(z0 ; δ) ⊆ Ω.
13.3. Simply Connectedness and Miscellaneous Problems 411
Suppose that
m
[k
ΩN = RN,j .
j=1
lim KN = C.
N →∞
If we pick UN (z) = QN,pN ,qN (z), then it can be established from the inequality (13.31) that
for almost every z ∈ C. Now the same result also holds for the imaginary part, i.e., there exists
a sequence of polynomials {VN } such that
for almost every z ∈ C. If we let PN = UN + iVN , then each PN is a polynomial and therefore,
our desired result follows immediately from the limits (13.32) and (13.33). Hence we have ended
the proof of the problem.
CHAPTER 14
Conformal Mapping
Problem 14.1
Rudin Chapter 14 Exercise 1.
az + b
f (z) = (14.1)
cz + d
maps Π+ onto itself if and only if a, b, c and d are real numbers such that ad − bc > 0.
Suppose that a, b, c and d are real numbers such that ad − bc > 0. By §14.3, the f in the
form (14.1) is already a one-to-one mapping of S 2 onto S 2 . Since a, b, c, d ∈ R, f must map the
real axis onto itself. Furthermore, we have
ad − bc
Im f (i) = > 0,
c2 + d2
so i is mapped into Π+ which proves that the transformation f maps Π+ onto itself. The converse
part can be found in [123, Problem 13.16, p. 181], completing the proof of the problem.
Problem 14.2
Rudin Chapter 14 Exercise 2.
Proof. Denote Π+ to be the upper half plane. Let z ∈ U . We have to make clear the meaning
of the reflection, namely z ∗ , of z with respect to the arc L. In fact, by the discussion in [11, pp.
102, 103]a , we know that z ∗ lies on the same ray as z and |z ∗ | = |z|−1 . In other words, we have
1
z∗ = .
z
a
See also [30, pp. 50, 51]
413
414 Chapter 14. Conformal Mapping
(a) We have the following analogous reflection theorem for this part:
Lemma 14.1
Suppose that Ω ⊆ Π+ , L = R and every point t ∈ L is the center of an open disc
Dt such that Π+ ∩ Dt ⊆ Ω. Let Ω− be the reflection of Ω, i.e.,
Ω− = {z ∈ C | z ∈ Ω}.
z−i
ψ(z) =
z+i
i(1 + ζ)
ψ −1 (ζ) =
1−ζ
is a conformal one-to-one mapping of U onto Π+ . For every θ ∈ [0, 2π], we know that
b = ψ −1 (L) and Ω
Thus L b = ψ −1 (Ω) are a segment of R and a region in Π+ respectively.
Define the map
fb = f ◦ ψΩb : Ω
b ⊆ Π+ → C. (14.4)
b converging to a z0 ∈ L,
Then for every {zn } ∈ Ω b the points ζn = ψ(zn ) ∈ Ω ⊆ U converging
to ζ0 = ψ(z0 ) ∈ L ⊆ T , so the hypothesis guarantees that
|fb(zn )| = f ψ(zn ) = |f (ζn )| → 1
as n → ∞. Hence it follows from Lemma 14.1 that there exists a function Fb, holomorphic
b ∪L
in Ω b∪Ω b − , such that
fb(z), if z ∈ Ω b ∪ L;
b
Fb(z) = 1
b −,
, if z ∈ Ω
b
f (z)
f ψ(z) , if z ∈ Ω b ∪ L;
b
= 1 (14.5)
b −.
, if z ∈ Ω
f ψ(z)
b − = {z ∈ C | z ∈ Ω}.
Here Ω b If z ∈ Ω
b − , then z ∈ Ω
b and so we note from the definition of ψ
that
1
ψ(z) = . (14.6)
ψ(z)
Hence the formula (14.5) becomes
f (ζ), if ζ ∈ Ω ∪ L;
F (ζ) = 1 (14.7)
, if ζ ∈ Ω∗ ,
−1
f ζ
−1
where Ω∗ = {ζ ∈ C | ζ ∈ Ω}.
Proof of Lemma 14.2. With the same function ψ : Π+ → U as in part (b), we see that
“z ∈ Ωb and z → L” b is equivalent to saying that “ψ(z) ∈ Ω and ψ(z) → L”. This
property implies that the function (14.4) satisfies
Im fb(z) = Im f ψ(z) → 0
Using the formula (14.6), the function (14.9) can be expressed in the following form:
f (ζ), if ζ ∈ Ω ∪ L;
F (ζ) = 1
f , if ζ ∈ Ω∗ .
ζ
This completes the proof of Lemma 14.2.
Now, since ( α1 )−1 = α, it is easy to conclude from the expression (14.7) that if f (α) = 0 for
some α ∈ Ω, then α1 ∈ Ω∗ so that
1 1
F = = ∞,
α f (α)
i.e., F has a pole at α1 . By the expression (14.8), the analogue of part (c) is that F ( α1 ) = f (α) = 0.
For part (a), if f (α) = 0 for some α ∈ Ω, then the expression (14.2) implies that
1
F (α) = = ∞,
f (α)
i.e., F has a pole at α. This finishes the proof of the problem.
Problem 14.3
Rudin Chapter 14 Exercise 3.
b
Proof. If |z| = 1, then it is easy to see that z = z1 and the rational function R(z) = R(z) · R( z1 )
satisfies
1
b
R(z) = R(z) · R = R(z) · R(z) = |R(z)|2 = 1
z
if |z| = 1. Since R b = P , where P and Q are polynomials, we have P (z) = Q(z) on the unit
Q
circle. Now the Corollary following Theorem 10.18 guarantees that P (z) ≡ Q(z) in C and this
implies that R(z) = 1 for every z ∈ C, i.e.,
1 1
R =
z R(z)
for every z ∈ C. Therefore, ω is a zero of order m of R if and only if ω1 is a pole of order m of
R. This fact shows that the zeros and poles of R inside U completely determines all zeros and
poles of R in C.
14.1. Basic Properties of Conformal Mappings 417
Next, suppose that R has a zero at z = 0 of order m. Suppose, further, that {0, α1 , α2 , . . . , αk }
are the distinct zeros and poles of R inside U . We consider the product
Yk
z − αn
B(z) = z m ·
n=1
1 − αn z
which is a rational function having the same zeros and poles of the same order as R. Recall
z−αn
from the proof of Theorem 12.4, we know that | 1−α nz
| = 1 for |z| = 1 and thus |B(z)| = 1 for
|z| = 1. Consequently, the quotient
R(z)
f (z) =
B(z)
|R(z)|
is a rational function without zeros or poles in D(0; r) for some r > 1. Since |f (z)| = |B(z)| =1
on |z| = 1, we get from the Corollary following Theorem 10.18 that f (z) = c for some constant
c with |c| = 1 in D(0; r) and hence in C \ { α11 , α12 , . . . , α1n }, i.e.,
Yk
z − αn
R(z) = cB(z) = cz m ·
n=1
1 − αn z
Problem 14.4
Rudin Chapter 14 Exercise 4.
Proof. We have R(z) > 0 on |z| = 1. By the hint, R must have the same number of zeros as
poles in U . Let α1 , α2 , . . . , αN and β1 , β2 , . . . , βN be the zeros and poles of R inside U , where
N is a positive integer. Next, we consider the rational function
(z − α)(1 − αz)
f (α, β, z) = , (14.10)
(z − β)(1 − βz)
Now the representation (14.10) indicates that α and β are the only zeros and poles of f (α, β, z)
inside U respectively. Therefore, the rational function
N
Y
Q(z) = f (αn , βn , z)
n=1
F (z) = c (14.11)
in U . By the Corollary following Theorem 10.18, we conclude that the result (14.11) holds in
Ω = C \ {β1 , β2 , . . . , βN , β1 , β1 , . . . , β1 } which means that
1 2 N
YN
(z − αn )(1 − αn z)
R(z) = cQ(z) = c (14.12)
n=1
(z − βn )(1 − βn z)
Problem 14.5
Rudin Chapter 14 Exercise 5.
n
X
Proof. Suppose that g(ζ) = ak ζ n which is a rational function. Then g(eiθ ) = f (θ), so g is
k=−n
positive on T . By the representation (14.12), one can find a positive constant c such that
Yn
(ζ − αj )(1 − αj ζ)
g(ζ) = c (14.13)
j=1
(ζ − βj )(1 − βj ζ)
f (θ) = g(eiθ )
Yn
(eiθ − αj )(1 − αj eiθ )
=c
eiθ
j=1
Yn
=c (eiθ − αj )(e−iθ − αj )
j=1
Yn
=c (eiθ − αj )(eiθ − αj )
j=1
Yn
=c |eiθ − αj |2
j=1
√ Yn 2
= c (eiθ − αj )
j=1
= |P (eiθ )|2 ,
√
where P (z) = c(z − α1 )(z − α2 ) · · · (z − αn ). We have completed the proof of the problem.
Problem 14.6
Rudin Chapter 14 Exercise 6.
14.1. Basic Properties of Conformal Mappings 419
Proof. If α = 0, then ϕ0 (z) = z, so the fixed points of ϕ0 are U . Next, we suppose that α 6= 0.
By Definition 12.3, ϕα (z) = z if and only if z 2 = αα if and only if
α
z=± .
|α|
This gives our first assertion.
For the second assertion, we note that since ϕα is a special case of the linear fractional
transformation ϕ(z) = az+bcz+d , we consider the general case for ϕ. By a suitable rotation, we may
assume that the straight line is the real axis. We claim that ϕ maps R ∪ {∞} into R ∪ {∞}
if and only if a, b, c and d are real. It is easy to see that if a, b, c and d are real, then we have
ϕ(R∪{∞}) ⊆ R∪{∞}. Conversely, suppose that ϕ(R∪{∞}) ⊆ R∪{∞}. Since ϕ(0) ∈ R∪{∞},
we have either d = 0 or db ∈ R.
eiθ z − eiθ α
eiθ ϕα (z) = .
(eiθ α)z + eiθ
Then the above claim says that eiθ , eiθ α and eiθ α are real, or equivalently, both eiθ and α are
real. Hence we end the analysis of the problem.
420 Chapter 14. Conformal Mapping
Problem 14.7
Rudin Chapter 14 Exercise 7.
Problem 14.8
Rudin Chapter 14 Exercise 8.
Proof. Let z = reiθ , where r > 0 and 0 ≤ θ < 2π. Then we have
e−iθ 1 1
f (reiθ ) = reiθ + = r+ cos θ + i r − sin θ.
r r r
Suppose that 1 1
x= r+ cos θ and y = r − sin θ. (14.17)
r r
If r = 1, then x = 2 cos θ and y = 0 so that f C(0; 1) = [−2, 2]. Suppose, otherwise, that r 6= 1,
so we obtain
x2 y2
1 2 + =1
(r + r ) (r − 1r )2
which is an ellipse.
Next, suppose that θ is a fixed number. Denote Lθ = {reiθ | 0 ≤ r < ∞}. If θ = 0, then
cos θ = 1 and sin θ = 0 and thus it easily follows from the representations (14.17) that
f (L0 ) = (0, ∞).
Similarly, if θ = π, then cos θ = −1 and sin θ = 0 so that
f (Lπ ) = (−∞, 0).
π 3π
If θ = 2, 2 , then cos θ = 0 and sin θ = ±1. Simple computation verifies that
f L π2 = f L 3π = iR.
2
π 3π
Finally, if θ ∈ [0, 2π) \ {0, 2 , π, 2 }, then we have cos θ · sin θ 6= 0 and we deduce from the
representations (14.17) that
x2 y2
− =4
cos2 θ sin2 θ
which is trivially a hyperbola. This completes the analysis of the problem.
14.1. Basic Properties of Conformal Mappings 421
Problem 14.9
Rudin Chapter 14 Exercise 9.
for all z ∈ U . Thus for every z ∈ D(0; r), we deduce from the expression (14.20) that
∞
4 X |z|2n−1 2 1 + |z| 2 1+r
|f −1 (z)| ≤ = −Im f −1 (|z|) = log = log .
π n=1 2n − 1 π 1 − |z| π 1−r
Notice that
2 2ez 1 − |A|2
ψ ′ (z) = , (ψ −1 )′ (z) = and ϕ′ (z) = eiθ · .
1 − z2 (ez + 1)2 (1 − Az)2
By the Chain Rule, we have
h′ (α + iβ) = ψ ′ ϕ ψ −1 (α + iβ) · ϕ′ ψ −1 (α + iβ) · (ψ −1 )′ (α + iβ)
2eα+iβ
= ψ ′ (0) × ϕ′ (A) ×
(eα+iβ + 1)2
eiθ 2eα+iβ
=2× 2
× α+iβ . (14.22)
1 − |A| (e + 1)2
According to the value (14.21) and the expression (14.22), we see that
|eα+iβ − 1|2 −1 2eα
|h′ (α + iβ)| = 2 1 − α+iβ ×
|e + 1|2 |eα+iβ + 1|2
4eα
= α+iβ
|e + 1|2 − |eα+iβ − 1|2
1
=
cos β
which is the desired result.
Remark 14.1
Problem 14.9 contributes to the theory of the Principle of Subordination, read [23, Chap.
VI, §5, pp. 207 - 215 ], [34, §1.5, pp. 10 – 13] or [81, Chap V, §9, pp. 226 – 236].
14.1. Basic Properties of Conformal Mappings 423
Problem 14.10
Rudin Chapter 14 Exercise 10.
Problem 14.11
Rudin Chapter 14 Exercise 11.
Proof. Now we have Ω = {z ∈ U | Im z > 0}. Using [11, Example 1, p. 180; Theorem 13.16, p.
183], if f : Ω → U is a conformal bijective mapping, then f has the representation
(z − 1)2 + 4α(z + 1)2
f (z) = eiθ · , (14.24)
(z − 1)2 + 4α(z + 1)2
where θ ∈ [0, 2π] and Im α > 0. Putting f (−1) = −1, f (0) = −i and f (1) = 1 into the formula
(14.24), we obtain that θ = π and α = 4i , so
(z − 1)2 + i(z + 1)2
f (z) = − (14.25)
(z − 1)2 − i(z + 1)2
is the desired conformal mapping.√By the representation (14.25), it is easily seen that if z ∈ Ω
satisfies f (z) = 0, then z = (−1 + 2)i. Furthermore, simple algebra gives f ( 2i ) = 7i which ends
the proof of the problem.
Problem 14.12
Rudin Chapter 14 Exercise 12.
Proof. For convenience, we let u(z) = Re f ′ (z) : C → R. We prove the assertions as follows:
• f is one-to-one in Ω when u(z) > 0 for all z ∈ Ω. Choose a, b ∈ Ω and a 6= b. Since
Ω is convex, the path γ(t) = a + (b − a)t for all t ∈ [0, 1] is in Ω. Then we know from the
Fundamental Theorem of Calculus that
Z Z 1
f ′ (z) dz = (b − a) f ′ a + (b − a)t dt = f (b) − f (a)
γ 0
so that Z
h f (b) − f (a) i 1
Re = u a + (b − a)t dt > 0.
b−a 0
Consequently, Re f (a) 6= Re f (b) which implies that f (a) 6= f (b). Since the pair of points
{a, b} is arbitrary, we assert that f is one-to-one in Ω.
b
Rudin used the notation “⊂” to mean “⊆”, see [99, Definition 1.3, p. 3].
424 Chapter 14. Conformal Mapping
is closed in C. Let p ∈ S. Since u is harmonic in Ω, it follows from the mean value property
that Z 2π
1
0 = u(p) = u(p + Reiθ ) dθ (14.27)
2π 0
for some R > 0 such that D(p; R) ⊆ Ω. If there exists a measurable set E ⊆ [0, 2π] such
that m(E) > 0 and u(p + Reiθ ) > 0 for every θ ∈ E, then we have
Z 2π Z Z
u(p + Reiθ ) dθ = u(p + Reiθ ) dm + u(p + Reiθ ) dm > 0
0 E [0,2π]\E
which contradicts the result (14.27). Therefore, no such E exists and then u(p + Reiθ ) = 0
a.e. on [0, 2π]. Now the continuity of u on Ω forces that u(z) = 0 for every z ∈ D(p; R).
In other words, D(p; R) ⊆ S which means that S is open in C.
Since S is both open and closed in C, we have either S = ∅ or S = C. Suppose that S = ∅,
then Re f ′ (z) > 0 in Ω so that f is one-to-one in Ω by the previous assertion. Next, we
suppose that S = C, then the definition (14.26) implies f ′ (Ω) is purely imaginary. Since
f ′ ∈ H(Ω), the Open Mapping Theorem ensures that f ′ (Ω) = A for some constant A.
Finally, if we consider g(z) = f (z) − Az, then g ∈ H(Ω) and g ′ ≡ 0 there. Thus it follows
from [11, Exercise 5, p. 42] that g is a constant B. Consequently, we have f (z) = Az + B.
If A 6= 0, then f is linear and so it is one-to-one. Otherwise, f ≡ B in Ω.
4πβ π
θ1 − θ2 = = · (π + 2δ). (14.30)
2β + π π+δ
δ2
If − π2 − δ < θ2 < − π2 + π+δ < π
2 + δ, then the inequality (14.30) yields
π π δ2 π π
θ1 = θ2 + · (π + 2δ) < − + + · (π + 2δ) = + δ.
π+δ 2 π+δ π+δ 2
Now we have found distinct θ1 , θ2 ∈ (−β, β), and hence distinct z1 and z2 of Ω such that
the equation (14.29) holds. Consequently, our f is not one-to-one in Ω.
c
In fact, it was shown by Tims [116, Theorem 2] that for any simply connected non-convex region Ω whose
boundary contains more than one point, there exists distinct points α, β ∈ Ω and an f ∈ H(Ω) such that Re f ′ (z)
is nonzero in Ω and f (α) = f (β).
14.1. Basic Properties of Conformal Mappings 425
Problem 14.13
Rudin Chapter 14 Exercise 13.
Proof. Assume that z1 6= z2 but f (z1 ) = f (z2 ). Pick a δ > 0 such that z2 ∈ / D(z1 ; δ) and
D(z1 ; δ) ⊆ Ω. Since every fn is one-to-one in D(z1 ; δ), none of the functions fn (z) − fn (z2 )
has a zero in D(z1 ; δ). By the hypotheses, fn (z) − fn (z2 ) → f (z) − f (z2 ) uniformly on every
compact subset in D(z1 ; δ), so we deduces from Problem 10.20 (Hurwitz’s Theorem) that either
f (z) − f (z2 ) 6= 0 for all z ∈ D(z1 ; δ) or f (z) − f (z2 ) = 0 in D(z1 ; δ). Since f (z1 ) = f (z2 ), we
have f (z) = f (z2 ) in D(z1 ; δ) and this means that
D(z1 ; δ) ⊆ Z f − f (z2 ) .
Problem 14.14
Rudin Chapter 14 Exercise 14.
• f (x + iy) → 0 as x → ∞ for all y ∈ (−1, 1). Assume that one could find an y ′ ∈
(−1, 1) \ {0} such that the limit lim f (x + iy ′ ) is nonzero. Since |f | < 1, the Bolzano-
x→∞
Weierstrass Theorem ensures the existence of a sequence {xn } and a nonzero complex
number L such that xn → ∞ as n → ∞ and
Consider the family F = {fn } ⊆ H(Ω), where fn (z) = f (z + xn ). Now the boundedness
of f implies that F is uniformly bounded on each compact subset of Ω. By Theorem 14.6
(Montel’s Theorem), it is a normal family and then there exists a subsequence {nk } and
an F ∈ H(Ω) such that fnk → F uniformly on every compact subset of Ω. It is clear that
{x, iy ′ } is compact. On the one hand, the hypothesis gives
• The passage to the limit is uniform if y is confined to [−α, α], where α < 1.
Assume that the limit was not uniform in Kα = {x + iy | x ∈ R and y ∈ [−α, α]} for some
α < 1. Then there exists some ǫ > 0 so that for all N ∈ N, one can find xN ≥ N and
yN ∈ [−α, α] such that
|fN (iyN )| = |f (xN + iyN )| > ǫ. (14.32)
As we have shown above that {fN } has a subsequence {fNk } which converges uniformly
on compact subsets of Ω to a holomorphic function g, and g 6≡ 0 in view of the inequality
(14.32). By the previous assertion, we have
fNk (x + iy) = f xNk + (x + iy) → 0
as k → ∞ for every (x, y) ∈ [−α, α]2 ⊆ Kα ⊆ Ω, so this means that g(z) = 0 for all
z ∈ [−α, α]2 and then the Corollary following Theorem 10.18 implies that g(z) = 0 in Ω,
a contradiction. Hence the limit must be uniform in Kα .
• Boundary behavior of a function g ∈ H ∞ with a radial limit.d Let g ∈ H ∞ .
Without loss of generality, we may assume that |g(z)| < C1 for all z ∈ U and g(reiθ ) → C2
as r → 1 for some θ, where C1 and C2 are some constants. Using the mapping (14.124),
we know that π
e2z − 1
κ(z) = π z
e2 + 1
is a conformal one-to-one mapping of Ω onto U . Since κ(x) → 1 as x → ∞, the composite
g κ(z)eiθ − C2
h(z) =
C1 + C2
is a mapping from Ω into U satisfying h ∈ H(Ω), |h(z)| < 1 for all z ∈ Ω and
g κ(x)eiθ − C2
h(x) = →0
C1 + C2
as x → ∞. Hence the first assertion implies that
lim h(x + iy) = 0 or lim g κ(x + iy)eiθ = C2 (14.33)
x→∞ x→∞
for every y ∈ (−1, 1). We observe from the definition that |κ(x+iy)| < 1 and κ(x+iy) → 1
as x → ∞. By §11.21, the limit (14.33) means that g has non-tangential limit C2 at eiθ .
The analogue of the second assertion can be stated similarly and we omit the details here.
Problem 14.15
Rudin Chapter 14 Exercise 15.
Proof. Let Π be the right half plane. Recall that the ϕ given by [100, Eqn. (6), p. 281] is a
conformal one-to-one mapping of U onto Π. Thus ϕ−1 : Π → U is conformal and bijective. Since
ϕ−1 (z) = z−1
z+1 and f : U → Π is holomorphic, we have g = ϕ
−1 ◦ f : U → U is holomorphic and
Problem 14.16
Rudin Chapter 14 Exercise 16.
Proof. Let p ∈ U . Then there exists a R > 0 such that D(p; 2R) ⊆ U . For every z ∈ D(p; R),
we have D(p; r) ⊆ U for all 0 < r ≤ R. Since f ∈ H(U ), it is harmonic in U by Theorem 11.4.
By the mean value property, we have
Z 2π
1
f (z) = f (z + reit ) dt
2π 0
which implies
Z 2π
1
rf (z) = rf (z + reit ) dt
2π 0
Z R Z R Z 2π
1
rf (z) dr = rf (z + reit ) dt dr
0 2π 0 0
Z R Z 2π
R2 1
f (z) = rf (z + reit ) dt dr. (14.36)
2 2π 0 0
n ZZ o1
1 √ 2
= 2
· πR · |f (z)|2 dx dy
πR
U
1
≤√ (14.37)
πR
for all z ∈ D(p; R). Let K ⊆ Ω be compact and {D(p; 2R)} be an open cover of K, where
D(p; 2R) ⊆ U . Then there exist finitely many points p1 , p2 , . . . , pN and positive numbers
R1 , R2 , . . . , RN such that
Problem 14.17
Rudin Chapter 14 Exercise 17.
Proof. The conclusion is affirmative. To this end, we need the following version of Hurwitz’s
Theorem [30, p. 152]:
for some positive integer m. Let N (K) = max{N (a1 , r1 ), N (a2 , r2 ), . . . , N (am , rm )}. It follows
from the result (14.38) that
fn (z) 6= fn (p) (14.39)
in K \ {p} for all n ≥ N (K). Since p is an arbitrary point of K, the result (14.39) means that
fn is one-to-one in K for all n ≥ N (K), completing the proof of the problem.
Problem 14.18
Rudin Chapter 14 Exercise 18.
14.2. Problems on Normal Families and the Class S 429
g′ (z0 ) (1 − αa)2
g(z) = · ϕα f (z)
f ′ (z0 ) 1 − |α|2
for all z ∈ Ω. This ends the proof of the problem.
Problem 14.19
Rudin Chapter 14 Exercise 19.
which means that f (rn , θ ′ ) cannot be contained in the neighborhood V of (1, 0) for all large n.
Hence no such continuous extension exists and we complete the proof of the problem.
Problem 14.20
Rudin Chapter 14 Exercise 20.
Proof. Write f (z) = zϕ(z). Then ϕ ∈ H(U ), ϕ(0) = 1 and ϕ has no zero in U . By Problem
13.9, there exists an h ∈ H(U ) such that hn (z) = ϕ(z) and h(0) = 1. Put
Recall that g(z) = g(ω), so it follows from the equation (14.47) that we have g(z) = g(ω) = 0
2kπi
or e n = 1. In the latter case, we have z = w Otherwise, since g(z) = 0 if and only if z = 0 in
U , we have z = w = 0, completing the proof of the problem.
Problem 14.21
Rudin Chapter 14 Exercise 21.
(a) Since f ∈ S , it is one-to-one. By Theorems 10.33 and 14.2, f is conformal and its inverse
f −1 : f (U ) → U exists and is conformal.f As U ⊆ f (U ), we consider g = f −1 |U : U → U .
Clearly, we have
1 1
g(0) = f −1 (0) = 0 and g ′ (0) = = = 1.
f′ f −1 (0) f ′ (0)
By Theorem 12.2 (Schwarz’s Lemma), we conclude that g(z) = z and consequently, f (z) =
z in U .
(b) We claim that there is no element f ∈ S with U ⊆ f (U ). Assume that f was such a
function. By part (a), we know that f (z) = z so that f (U ) = U which contradicts our
assumption that U ⊆ f (U ).
(c) Consider the function F given in [100, Eqn. (1), p. 286]. If |α1 | = 1, then Theorem 14.13
(The Area Theorem) implies that αn = 0 for all n = 2, 3, . . .. In this case, we have
1
F (z) = + α0 + eiθ z.
z
Now we know from the proof of Theorem 14.14 that |a2 | = 2 is equivalent to |α1 | = 1, so
we have
1 1
= G(z) = + α0 + eiθ z.
g(z) z
By Theorem 14.12, we have f (z 2 ) = g 2 (z) which implies definitely that α0 = 0 and then
z2
f (z 2 ) = .
(1 + eiθ z 2 )2
Problem 14.22
Rudin Chapter 14 Exercise 22.
g = f −1 ◦ (if ) : U → U
which is also a one-to-one conformal mapping of U onto itself. Clearly, g(0) = f −1 if (0) = 0.
By Remark 10.3, we have
if ′ (z)
g′ (z) = ′
f if (z)
which gives |g′ (0)| = |i| = 1. Hence Theorem 12.2 (Schwarz’s Lemma) implies that the formula
holds in U for some constant λ with |λ| = 1. By the note following Theorem 14.2, we have
f ′ (0) 6= 0. Now we observe from this fact and the expression (14.48) that if ′ (z) = λf ′ (λz) so
that λ = i if we put z = 0 into this equation. Thus we get what we want
for all z ∈ U . If n − 1 is not a multiple of 4, then in−1 6= 1, so it yields from Theorem 10.18 that
cn = 0 for such n.
A generalization is as follows: Let S be a simply connected region with rotational symmetry
of order N and center at 0, i.e., exp( 2πi N )S = S. Let f : U → S be a one-to-one conformal
mapping with f (0) = 0. Thus both exp( 2πi N )f : U → S and f
−1 : S → U are conformal. By
holds for some constant λ with |λ| = 1. Since f ′ (0) 6= 0, we take differentiation to both sides of
the formula (14.50) to conclude that λ = exp( 2πi
N ) and hence
2πi 2πi
e N f (z) = f e N z (14.51)
P
for every z ∈ U . Next, if n − 1 is not a multiple of N and f (z) = cn z n , then the formula
(14.51) gives
X∞
2πi 2π(n−1)i
e N cn 1 − e N zn = 0
n=1
2π(n−1)i
for all z ∈ U . Since e N 6= 1 if n − 1 is not a multiple of N , Theorem 10.18 implies that
cn = 0 for such n. This completes the analysis of the problem.
Problem 14.23
Rudin Chapter 14 Exercise 23.
Proof. Suppose that ∂Ω = C1 ∪C2 , where C1 lies in the inside of C2 . By appropriate translation,
rotation and homothety, we assume that C2 = T (the unit circle) and the center of C1 lies on
the real axis with x-intercepts a and b, where |a| < b < 1. See Figure 14.1 below:
14.3. Proofs of Conformal Equivalence between Annuli 433
By Theorem 12.4, ϕα carries T onto itself and U onto U , where |α| < 1. Since ϕα is a linear
fractional transformation, §14.3 ensures that it will send C1 ⊂ U onto a circle or a line. Now
the condition ϕα (U ) = U implies that ϕα (C1 ) must be a circle.
Suppose that α ∈ R. Since C1 intersects the real axis at a and b perpendicularly and ϕα is
a conformal map, ϕα (a) and ϕα (b) are the end-points of a diameter of ϕα (C1 ). Furthermore,
since ϕα (R) = R, both ϕα (a) and ϕα (b) are real. Thus if ϕα (C1 ) is a circle centered at 0, then
we must have
ϕα (a) = −ϕα (b)
which gives
a−α b−α
=−
1 − αa 1 − αb
2 2(1 + ab)
α − α + 1 = 0.
a+b
Since |a| < b < 1, we always have a(1 − b) < 1 − b or equivalently, 1 + ab > a + b > 0. Combining
this and the formulas (14.52), it follows that
r
1 + ab 1 + ab 2
α+ = + − 1 > 1.
a+b a+b
Since α+ · α− = 1, we conclude that 0 < α− < 1. Takethis α− . Then the conformal map
ϕα− carries U onto U , T onto T and C1 onto
C 0; ϕα− (a) . In other words, it is a one-to-one
conformal mapping of Ω onto A ϕα− (a), 1 . This completes the proof of the problem.
434 Chapter 14. Conformal Mapping
Remark 14.2
An example of this kind of conformal mappings can be found in [123, Problem 13.20, pp.
182, 183].
Problem 14.24
Rudin Chapter 14 Exercise 24.
Proof. Since 1 < R2 < R1 , we have A(1, R2 ) ⊆ A(1, R1 ). Assume that f : A(1, R1 ) → A(1, R2 )
was a bijective conformal mapping. By the first half of the proof of Theorem 14.22, we may
assume without loss of generality that |f (z)| → 1 as |z| → 1 and |f (z)| → R2 as |z| → R1 .
Consider the family of holomorphic functions F = {fn }, where
where n, m ∈ N and n 6= m. Fix m. Assume that ω ∈ fm (Ω) ∩ fn (Ω). Then we can find p, q ∈ Ω
such that fn (p) = ω and fm (q) = ω. If n < m, then we have
fn (p) = fm (q) = fn fm−n (q)
which gives fm−n (q) = p ∈ Ω, but it contradicts the fact that fm−n A(1, R1 ) ∩ Ω = ∅. If
n > m, then we have
fm (q) = fn (p) = fm fn−m(p)
which gives fn−m (p) = q ∈ Ω, a contradiction again. Consequently, we prove the claim (14.54).
Assume that the range of g contained a nonempty open set. This means that g is not
constant and so we pick
an p ∈ Ω ⊆ A(1, R1 ) such that g(p) = ω. By the Open Mapping
Theorem, g A(1, R1 ) is an open set so that there exists a δ > 0 such that 0 ∈ / g D(p; δ) .
Suppose that
hk (z) = fnk (z) − ω and h(z) = g(z) − ω.
We observe that hk , h ∈ H A(1, R1 ) , {hk } converges to h uniformly on compact subsets of
A(1, R1 ) and h 6≡ 0. Furthermore, we can select δ if necessary so that h(z) 6= 0 on C(p; δ).g
Hence it follows from Lemma 14.3 (Hurwitz’s Theorem) that there corresponds an N ∈ N such
g
Otherwise, h(z) = 0 for all z ∈ A(1, R1 ) by Theorem 10.18 which means that g is constant.
14.3. Proofs of Conformal Equivalence between Annuli 435
that if k ≥ N , then hk and h have the same number of zeros in D(p; δ). Since h(p) = g(p)−ω = 0,
there exists an zk ∈ D(p; δ) ⊆ Ω such that k ≥ N implies
fnk (zk ) = ω
but this contradicts the fact (14.54). Therefore, the range of g cannot contain any nonempty
open set and the Open Mapping Theorem ensures that g is constant.
√
On the other hand, g cannot be constant on the circle C(0; R1 ). Otherwise, Theorem√10.24
(The Maximum Modulus Theorem)√and its Corollary establish that g is constant in A(1, R1 ).
Let K be a compact subset of A(1, R1 ). Since fnk → g uniformly on K, fnk is also constant in
K for large enough k. However, this contradicts
√ the fact that fn is injective for every n = 1, 2, . . ..
Therefore, g is not constant on C(0; R1 ).
Now the above two results are contrary, so they force that no such f exists and we have
completed the proof of the problem.
Problem 14.25
Rudin Chapter 14 Exercise 25.
Proof. We have f : A(1, R1 ) → A(1, R2 ), where 1 < R2 < R1 . By the first half of the proof of
Theorem 14.22, we may assume without loss of generality that
lim |f (z)| = 1 and lim |f (z)| = R2 .
|z|→1 |z|→R1
Applying Problem 14.2(b), the reflection across the inner circle (i.e., the unit circle) extends f
to a conformal mapping
f1 : A(R1−1 , R1 ) → A(R2−1 , R2 )
and f1 satisfies
lim |f1 (z)| = R2−1 , lim |f1 (z)| = 1 and lim |f1 (z)| = R2 .
|z|→R−1
1
|z|→1 |z|→R1
This process can be repeated infinitely many times and finally we obtain a conformal mapping
F of the punctured plane C \ {0}. If F has a pole of order m at the origin, then z m F (z) is
entire and |F (z)| = 1 whenever |z| = 1. Now Problem 12.4(b) asserts that F (z) = αz n for some
|α| = 1 and some n ∈ Z \ {0}. Assume that n ≥ 2. Let a ∈ A(1, R1 ). Take ζ 6= 1 to be an n-root
of unity. Then we have
F (a) = αan = α(ζa)n = F (ζa), (14.55)
but a 6= ζa. This contradicts the fact that F is one-to-one in A(1, R1 ). Assume that n ≤ −1.
The equation (14.55) also holds in this case which in turn contradicts the injectivity of F again.
In other words, F (z) = αz which implies that R1 = R2 , a contradiction. Next, if F has a
removable singularity at the origin, then F is actually entire and we use the same argument as
above to show that no such F exists. Hence no such mapping f exists which completes the proof
of the problem.
436 Chapter 14. Conformal Mapping
Remark 14.3
(a) Theorem 14.22 is sometimes called Schottky’s Theorem. Besides the proofs given
in the text (Problems 14.24 and 25), you can also find a simple and elegant proof ofs
this theorem in [8].
(b) Besides the analytical proofs provided in the text and the problems, one can find a
pure algebraic proof in [95].
Problem 14.26
Rudin Chapter 14 Exercise 26.
Proof.
(a) Suppose that the regions Ω0 , Ω1 , . . . , Ωn−1 and functions f1 , f2 , . . . , fn are constructed such
that Ωk = fj (Ωk−1 ), where k = 1, 2, . . . , n. Define
Then rn is the largest number such that D(0; rn ) ⊆ Ωn−1 and the definition shows that
there is an αn ∈ ∂Ωn−1 with |αn | = rn .h See Figure 14.2 below.
where
z−α
ϕα (z) = and s(ω) = ω 2 .
1 − αz
By the Chain Rule, we have
Fn′ (z) = ϕ′−αn s ϕ−βn (z) × s′ ϕ−βn (z) × ϕ′−βn (z)
= 2ϕ′−αn s ϕ−βn (z) × ϕ′−βn (z) × ϕ−βn (z).
By Theorems 10.33 and 12.4, ϕ′−αn (z) 6= 0 and ϕ′−βn (z) 6= 0 for every z ∈ Ωn−1 . Thus
Fn′ (z) 6= 0 for all z ∈ Ωn−1 and it follows from Theorem 10.30(c) that Fn has a holomorphic
inverse Gn in Ωn−1 .
Gn = ϕ−1
−βn ◦ s
−1
◦ ϕ−1
−αn = ϕβn ◦ s
−1
◦ ϕαn : Ωn−1 → U, (14.58)
√
where s−1 (z) = z. Combining the Chain Rule and Theorem 12.4, we get
G′n (0) = ϕ′βn s−1 ϕαn (0) × (s−1 )′ ϕαn (0) × ϕ′αn (0)
1
= ϕ′βn s−1 (−αn ) × √ × (1 − |αn |2 )
2 −αn
1 − rn2
= × ϕ′βn (βn )
2βn
1 − rn2
= . (14.59)
2(1 − |βn |2 )βn
|G′n (0)|
Put fn = λn Gn : Ωn−1 → U , where λn = G′n (0) . Now the formula (14.59) implies that
1 − rn2 1 − rn2 1 + rn
fn′ (0) = λn G′n (0) = |G′n (0)| = 2
= √ = √ .
2(1 − |βn | ) · |βn | 2(1 − rn ) rn 2 rn
Using the A.M. ≥ G.M., it is easy to see that fn′ (0) > 1.
By the representation (14.58) and Theorem 12.4, each Gn and hence each fn is injec-
tive on Ωn−1 . Consequently, each
ψn : Ω → Ωn ⊆ U
is injective on Ω.
– {ψn′ (0)} is bounded. Since ϕ−αn , ϕ−βn , s ∈ H(U ), the definition (14.57) gives Fn ∈
H(U ). Furthermore, it is clear that
Fn (0) = ϕ−αn ϕ2−βn (0) = ϕ−αn (βn2 ) = ϕ−αn (−αn ) = 0, (14.60)
438 Chapter 14. Conformal Mapping
Lemma 14.4
For every n = 1, 2, . . ., we have 0 < r1 ≤ r2 ≤ · · · ≤ 1.
Proof of Lemma 14.4. Notice that we have fn (Ωn−1 ) = Ωn , D(0; rn ) ⊆ Ωn−1 and
D(0; rn+1 ) ⊆ Ωn . By the definition (14.56), there exists a boundary point α ∈ ∂Ωn
such that |α| = rn+1 . Select {zn−1,k } ⊆ Ωn−1 such that fn (zn−1,k ) → α as k → ∞.
Assume that {zn−1,k } had a limit point β in Ωn−1 . Since fn is obviously continuous on
Ωn−1 , we have α = fn (β) which means that Ωn ∩ ∂Ωn 6= ∅, a contradiction. Therefore,
the sequence {zn−1,k } cannot have a limit point in Ωn−1 . Since D(0; rn ) ⊆ Ωn−1 , we
must have
lim sup |zn−1,k | ≥ rn .
k→∞
Since |fn (z)| = |Gn (z)|, we derive from the first inequality (14.62) that
rn+1 = |α| = lim |fn (zn−1,k )| = lim |Gn (zn−1,k )| ≥ lim sup |zn−1,k | ≥ rn
k→∞ k→∞ k→∞
so that
m
Y 1 + rk 1
1< √ ≤ (14.65)
2 rk rn+1
k=n+1
u1 ≤ u2 ≤ · · · .
|ω| < rn
ω ∈ D(0, rN +p )
⊆ ΩN +p−1
= fN +p−1 (ΩN +p−2 )
= ···
= fN +p−1 ◦ fN +p−2 ◦ · · · ◦ fN +1 ◦ fN (ΩN −1 )
= fN +p−1 ◦ fN +p−2 ◦ · · · ◦ fN +1 ◦ ψN (Ω)
Applying the inequality (14.62) repeatedly to the expression (14.67), we see that
−1
for all p ≥ 1. Clearly, D(0; |ω|) is a compact subset of ΩN , so ψN D(0; |ω|) is also
a compact subset of Ω. Denote this set by K. By the result (14.69), we establish
that {zp } ⊆ K. Then the Bolzano-Weierstrass Theorem [127, Problem 5.25, p. 68]
suggests that there corresponds a subsequence {zpj } such that zpj → z ∈ K. Since
ψn → ψ uniformly on K, it observes that
Combining the expression (14.68) and the limit (14.70), we conclude at once that
ω = ψ(z)
Problem 14.27
Rudin Chapter 14 Exercise 27.
Proof. Taking logarithms in the inequality (14.64) with m = 2n and then using the hint to get
h √
(1 − rk )2 i
2n
X 2n
X 1 + rk 1
log 1 + √ = log √ ≤ log = − log rn+1 . (14.71)
2 rk 2 rk rn+1
k=n+1 k=n+1
14.4. Constructive Proof of the Riemann Mapping Theorem 441
x
Using the Mean Value Theorem for Derivatives, one can show that log(1 + x) > 1+x for x > 0.
√
(1− rk )2 1
Since 0 < 2 < 2 for all k ∈ N, the inequality (14.71) reduces to
h √ √
(1 − rk )2 i
2n
X 2n
X (1 − rk )2
− log rn+1 > log 1 + √ > > 0. (14.72)
2 rk 3
k=n+1 k=n+1
which implies
log r1 log r1−1
− log rn+1 ≤ (1 − rn+1 ) · ≤ (1 − rn ) · . (14.73)
r1 − 1 1 − r1
√
As (1 + rk )2 ≤ 4, we have
√ 2 (1 − rk )2 (1 − rk )2
(1 − rk ) = √ 2 ≥ . (14.74)
(1 + rk ) 4
log r −1
Let A = 1−r11 . Now we observe by substituting the inequalities (14.73) and (14.74) into the
inequality (14.72) that
√
1 X 1 − rk 2
2n 2n
X (1 − rk )2
0< ≤ < − log rn+1 < A(1 − rn )
3 2 3
k=n+1 k=n+1
or equivalently,
2n
X 1 − rk 2 1 − rn
0< < , (14.75)
2B B
k=n+1
1
1−r1
Since 16B is constant and a n → 1 as n → ∞ if a > 0, there exists a positive constant M such
that
M
0 < 1 − r2p <
2p
for every p = 1, 2, . . .. Thus it gives
M2
0 < 2p (1 − r2p )2 <
2p
∞
X M2
for every p = 1, 2, . . .. Obviously, the series converges, so we deduce from the results
p=1
2p
[127, Theorems 6.5, 6.6, p. 76] that
∞
X
(1 − rn )2 < ∞.
n=1
Remark 14.4
(a) The proofs of Problems 14.26 and 14.27 are more or less the same as Ostrowski’s paper
[84].
(b) For more properties of the so-called Koebe mapping and details of the constructive
proof of the Riemann Mapping Theorem, please refer to [88, §3, pp. 15, 16, 183 – 186]
and [15, §1.7, pp. 180 – 214] respectively.
Problem 14.28
Rudin Chapter 14 Exercise 28.
Proof. Since αn ∈ U \ Ωn−1 , we have |αn | ≥ rn , see Figure 14.2 again. By Problem 14.26, we
may assume further that
1 + rn
rn < |αn | ≤ . (14.76)
2
In this case, we also have β 2 = −αn . Furthermore, recall from the definition (14.57) that
Fn′ (z) 6= 0 for all z ∈ Ωn−1 so that Fn has a holomorphic inverse Gn in Ωn−1 . Next, we observe
from the formula (14.59) and the A.M. ≥ G.M. that
1 − |αn |2 1 + |βn |2 1 + |αn |
|G′n (0)| = = = p > 1.
2(1 − |βn |2 )|βn | 2|βn | 2 |αn |
In conclusion, we always have |fn′ (0)| > 1 for n = 1, 2, . . .. Now it is easy to see that Lemma
14.4 remains valid, each ψn : Ω → Ωn is bijective and {ψn′ (0)} is bounded by r11 . It remains to
prove that rn → 1 as n → ∞.
Instead of the formula (14.64) and the inequalities (14.65), we obtain
Yn
1 + |αk |
ψn′ (0) = fn′ (0) × fn−1
′
(0) × · · · × f1′ (0) = p
k=1
2 |αk |
and
Ym
1 + |αk | 1
1< p ≤ ,
2 |αk | rn+1
k=n+1
14.4. Constructive Proof of the Riemann Mapping Theorem 443
1 + |αk | 1 + rk + δk
p = √ ,
2 |αk | 2 rk + δk
converges and so
1 + rn + δn
1 = lim √ . (14.77)
n→∞ 2 rn + δn
By Lemma 14.4, {rn } converges to a positive number r. Suppose that δn → δ as n → ∞, where
0 ≤ δ ≤ 1−r
2 . Then it follows from the limit (14.77) that
1+r+δ
1= √
2 r+δ
which implies r + δ = 1. If r 6= 1, then δ 6= 0. Now the expression δ = 1 − r contradicts the fact
0 < δ ≤ 1−r
2 . Hence we have r = 1 and we complete the proof of the problem.
Problem 14.29
Rudin Chapter 14 Exercise 29.
Proof.
(a) By translating Ω by −a, we may assume without loss of generality that a = 0. We claim
that
fn′ (0) = [f ′ (0)]n (14.78)
for all n = 1, 2, . . .. We use induction and the case n = 1 is trivial. Assume that
The hypothesis f (0) = 0 implies that fn (0) = 0 for all n ∈ N, so the expression (14.80)
′
reduces to fk+1 (0) = f ′ (0) · fk′ (0). By the inductive step (14.79), we conclude that
f (Ω) ⊆ Ω, it is true that fn (Ω) ⊆ Ω for every n = 1, 2, . . .. Hence we have |fn (z)| ≤ M
for all z ∈ D(0; r) and n ∈ N, and then we yield from Theorem 10.26 (Cauchy’s Estimate)
that
k!M
|fn(k) (0)| ≤ k (14.81)
r
for k = 1, 2, . . .. Combining the expression (14.78) and the estimate (14.81), we get
1
M 1
|f ′ (0)| = [fn′ (0)] n ≤
n
r
for every n = 1, 2, . . .. By taking n → ∞, we establish that |f ′ (0)| ≤ 1 as required.
(b) If f (k) (0) = 0 for all k ≥ 2, then since f ∈ H(Ω), Theorem 10.16 implies that f (z) = a + bz
for some a, b ∈ C. Since f (0) = 0 and f ′ (0) = 1, we have a = 0 and b = 1. Thus f (z) = z
as desired. Suppose that N ≥ 2 is the smallest positive integer such that f (N ) (0) 6= 0 and
f (k) (0)
let ck = k! for k ≥ N . In particular, we have cN 6= 0. Since f ′ (0) = 1, we have
∞
X
f (z) = z + ck z k = z + cN z N g(z), (14.82)
k=N
where g(0) = 1.
We claim that for each n ∈ N, there is a rn > 0 and an gn ∈ H D(0; rn ) such that for all
z ∈ D(0; rn ), we have
The expression (14.82) is just the case n = 1, so we assume that the result (14.83)
holds for some positive
integer n. Since fn (0) = 0, we can pick rn+1 ∈ (0, rn ) such
that fn D(0; rn+1 ) ⊆ D(0; rn ). Then it follows from the expressions (14.82) and (14.83)
that if z ∈ D(0; rn+1 ), then
fn+1 (z) = f fn (z)
= fn (z) + cN [fn (z)]N g fn (z)
N
= z + cN z N gn (z) + cN z + cN z N gn (z) g fn (z)
N
= z + cN z N gn (z) + 1 + cN z N −1 gn (z) g fn (z) .
N
Suppose that gn+1 (z) = gn (z) + 1 + cN z N −1 gn (z) g fn (z) . Recall that N − 1 ≥ 1, so
we obtain
gn+1 (0) = gn (0) + g fn (0) = n + 1.
By induction, our claim follows. Next, we differentiate the expression (14.83) N times and
put z = 0, we have
fn(N ) (0) = N !cN gn (0) = nN !cN
(N )
for every n = 1, 2, . . .. Since cN 6= 0, we have fn (0) → ∞ as n → ∞ which contradicts
the fact (14.81). Consequently, this means that f (k) (0) = 0 for all k ≥ 2 which implies
f (z) = z in D(0; r). By the Corollary following Theorem 10.18, it is actually true in Ω.
(c) By the hypothesis, we know that f ′ (0) = eiθ for some θ ∈ [0, 2π]. If eiθ is an N -root of
unity, then the integer nk = kN satisfies
[f ′ (0)]nk = (eiN θ )k = 1.
14.4. Constructive Proof of the Riemann Mapping Theorem 445
Otherwise, we need the following form of the Kronecker’s Approximation Theorem [6,
Theorem 7.8, p. 149]:
θ
Take α = 0 and β = 2π . For every k ∈ N, we obtain from Lemma 14.5 that there exist
integers nk and mk with nk > 0 such that
n θ 1
k
− mk <
2π 2kπ
or equivalently
1
|nk θ − 2mk π| <
k
which implies that |nk θ − 2mk π| → 0 as k → ∞. Consequently, we have eink θ → 1 as
k → ∞.
Next, we consider the family F = {fnk }. Recall the fact that fn (Ω) ⊆ Ω, so F is bounded
on Ω and Theorem 14.6 (Montel’s Theorem) ensures that F is normal and thus it has
a convergent subsequence which we also call {fnk } for convenience. Let g be its limit
function. By Theorem 10.28, g ∈ H(Ω) and fn′ k (0) → g′ (0) as k → ∞. By the formula
(14.78) and the hypothesis f ′ (0)]nk → 1 as k → ∞, we get fn′ k (0) → 1 as k → ∞ and
thus g′ (0) = 1. In other words, g is not constant and it follows from Problem 10.20 that
g(Ω) ⊆ Ω. By part (b), we see that
g(z) = z (14.84)
in Ω.
Finally, if z, ω ∈ Ω and f (z) = f (ω), then we have
Problem 14.30
Rudin Chapter 14 Exercise 30.
Proof. We have
n az + b o
Λ = ϕ(z) = a, b, c, d ∈ C and ad − bc 6= 0 .
cz + d
The number ad − bc is called the determinant of ϕ.
446 Chapter 14. Conformal Mapping
0 · z + (β − γ)
ϕ(z) = [z, ∞, β, γ] = (14.86)
z−γ
(b) Suppose that {α, β, γ} and {a, b, c} are two groups of distinct complex numbers. By the
definition (14.85), we have
which implies
Next, if {∞, β, γ} and {a, b, c} are two groups of distinct complex numbersk , then we follow
from the definition (14.86) (or by taking α → ∞ in the definition (14.87) that
(c) Suppose that φ(z) = [z, β, γ, δ] which sends {β, γ, δ} to {0, 1, ∞} by part (a). Then the
mapping φ ◦ ϕ−1 carries {ϕ(β), ϕ(γ), ϕ(δ)} to {0, 1, ∞}. By §14.3, this map φ ◦ ϕ−1 is
unique, so (φ ◦ ϕ−1 )(z) = [z, ϕ(β), ϕ(γ), ϕ(δ)] and then we have
[ϕ(α), ϕ(β), ϕ(γ), ϕ(δ)] = (φ ◦ ϕ−1 ) ϕ(α) = φ(α) = [α, β, γ, δ].
(d) This part has been solved in [123, Problem 13.18, p. 182].
z∗ − β z−β
= . (14.91)
α−β α−β
This certainly gives |z ∗ − β| = |z − β|. Since β is an arbitrary point on C, z and z ∗ are in
fact equidistance from each point on C. Furthermore, the equation (14.91) implies that
z∗ − β z−β z−β
Im = Im = −Im
α−β α−β α−β
Next, we suppose that C = {z ∈ C | |z| = 1}. By the equation (14.90) and applications of
the invariance property of ϕ ∈ Λ as verified in part (c), we see that
h 1 1 1 i h1 i
[z ∗ , α, β, γ] = [z, α, β, γ] = [z, α, β, γ] = z, , , = , α, β, γ
α β γ z
1
which implies z ∗ = z or z ∗ · z = 1. Thus we have |z ∗ | = |z|−1 and furthermore, it deduces
from the ratio
z∗ 1
= 2 >0
z |z|
that z ∗ lies on the ray L = {tz | t ∈ R}. Geometrically, see Figure 14.3 for the construction
of the point z ∗ .
(f) Let α, β, γ ∈ C. Then it follows from part (c) and the definition that
Problem 14.31
Rudin Chapter 14 Exercise 31.
Proof.
(a) Given ϕ, ψ, φ ∈ Λ by
az + b αz + β Az + B
ϕ(z) = , ψ(z) = and φ(z) = .
cz + d γz + δ Cz + D
– Composition as group operation. It is easy to see that
(αa + βc)z + αb + βd
ψ ϕ(z) =
(γa + δc)z + γb + δd
14.4. Constructive Proof of the Riemann Mapping Theorem 449
– The identity element. Now the usual identity map id : C → C is the identity
element of Λ because its determinant is 1 and
id ◦ ϕ = ϕ ◦ id = ϕ.
– The inverse of ϕ. The equation ω = ϕ(z) has exactly one solution and indeed, it is
dω − b
z = ϕ−1 (ω) = .
cω − a
Since the determinant of ϕ−1 is −ad + bc 6= 0, we have ϕ−1 belongs to Λ. Clearly, we
know that ϕ−1 ◦ ϕ = ϕ ◦ ϕ−1 = id.
By the definition (see [42, Definition 4.1, pp. 37, 38]),
Λ is indeed a group.
If1 we take
ϕ(z) = z + 1 and ψ(z) = z1 , then we see that ϕ ψ(z) = z1 + 1 and ψ ϕ(z) = z+1 which
imply that Λ is not commutative.
(b) Let ϕ ∈ Λ be given by
az + b
ϕ(z) =
cz + d
and ϕ 6= id. Define ∆ = ad − bc 6= 0. Since we may write
√a z + √b
∆ ∆
ϕ(z) = ,
√c z + √d
∆ ∆
we may assume without loss of generality that ad − bc = 1. Now the equation z = ϕ(z) is
equivalent to saying that
cz 2 + (d − a)z − b = 0. (14.92)
– Case (i): c = 0. Thus we have ad 6= 0 and ϕ(∞) = ∞ so that ∞ is a fixed point of
ϕ. If a 6= d, then by solving the equation (14.92), we get one more (finite) fixed point
which is
b
z= .
d−a
Otherwise, a = d implies that
b
ϕ(z) = z + (14.93)
d
whose fixed point is also ∞. Since ϕ 6= id, b 6= 0 so that ϕ has only a unique (infinite)
fixed point in this case.
– Case (ii): c 6= 0. Then the equation (14.92) has two roots
p p
a − d ± (d − a)2 + 4bc a − d ± (a + d)2 − 4
z= = . (14.94)
2c 2c
Since ϕ(∞) = ac , ∞ is not transformed into itself. This means that ϕ has either one
or two finite fixed points on S 2 depending on whether a + d = ±2 or not. In the case
of the unique finite fixed point, it is given by
a−d
z= .
2c
450 Chapter 14. Conformal Mapping
– Case (i): ϕ has a unique fixed point. Given ϕ1 (z) = z + 1 which is obviously an
element of Λ.
If c = 0, then we follow from part (b) that ϕ has a unique (infinite) fixed point if and
only if it takes the form (14.93). Define ψ(z) = db z. Recall that bd 6= 0, so ψ ∈ Λ and
ψ −1 (z) = db z. Furthermore, it is clear that
d b b
ψ −1 ϕ ψ(z) = · z+ = z + 1 = ϕ1 (z).
b d d
Hence we have shown that ϕ is conjugate to ϕ1 in this case.
Next, if c 6= 0, then we observe from the roots (14.94) that ϕ has a unique fixed point
z1 = a−d
2c if and only if a + d = ±2. Define the linear fractional transformation
1
S(z) =
z − z1
T = S ◦ ϕ ◦ S −1 (14.95)
has ∞ as its only fixed point because if p is a fixed point of T , then S −1 (p) will be a
fixed point of ϕ so that
p = S(z1 ) = ∞.
Hence it follows from part (b) that T (z) = z + B for some B ∈ C \ {0}. If we take
P (z) = Bz, then P −1 (z) = Bz and so
ϕ1 = P −1 ◦ T ◦ P. (14.96)
ϕ1 = (P −1 ◦ S) ◦ ϕ ◦ (S −1 ◦ P ) = (S −1 ◦ P )−1 ◦ ϕ ◦ (S −1 ◦ P ).
az + b
ϕ(z) = .
d
b
Take ψ(z) = z + d−a which is obviously a linear fractional transformation and
−1 b
ψ (z) = z − d−a . Direct computation gives
a b b a b
ϕ ψ(z) = z+ + = z+
d d−a d d d−a
14.4. Constructive Proof of the Riemann Mapping Theorem 451
and then
a b b a
ψ −1 ϕ ψ(z) = z + − = z = φ ad (z).
d d−a d−a d
Consequently, ϕ is conjugate to φα with
a
α= . (14.97)
d
Particularly, 0 is the finite fixed point if and only if b = 0, so we have ϕ(z) = ad z
and ψ(z) = z.
∗ Subcase (ii): c 6= 0 and a + d 6= ±2. The two distinct finite fixed points
are given by (14.94). Let z1 and z2 be the roots corresponding to the negative
square root and the positive square root respectively. Now the linear fractional
transformation
z − z1
S(z) =
z − z2
maps the ordered pair {z1 , z2 } into {0, ∞}. Then the linear fractional transfor-
mation T = S ◦ ϕ ◦ S −1 fixes 0 and ∞. By the particular case of Subcase (i),
we know that
φα = ψ −1 ◦ T ◦ ψ = T
z2 z−z1
for some complex α. Since S −1 (z) = z−1 , we obtain from the definition that
– The existence of β. Since ϕ has only a unique finite fixed point, the analysis of
part (b) leads us to the result that c 6= 0 and a + d = ±2. In this case, we have
α = a−d
2c . Then we have
d = a − 2αc. (14.101)
Since α is a root of the equation (14.92), we have
Substituting the values (14.101) and (14.102) into the expression (14.103) to get
1 cz + a − 2αc c(z − α) + a − αc 1 c
= = = + (14.104)
ϕ(z) − α (a − cα)z + α(cα − a) (z − α)(a − αc) z − α a − αc
which means that
c
β= ∈ C.
a − αc
a−d
With the aid of α = 2c and a + d = ±2, we can further show that
β = c. (14.105)
f (ϕ ◦ φ) = βϕ + βφ ,
1 1
so f is a homomorphism. Next, suppose that β = 0. Then ϕ(z)−α = z−α if and only
if ϕ = id. In other words, the kernel of f is {id}. Finally, given β ∈ C \ {0}. we
consider a = 1 + αβ, b = −α2 β, c = β and d = 1 − αβ. Direct computation gives
ad − bc = 1. Besides, the linear fractional transformation
(1 + αβ)z − α2 β
ϕβ (z) = (14.108)
βz + (1 − αβ)
fixes α only and satisfies the equation (14.107).m Consequently, we have ϕβ ∈ Gα
and f (ϕβ ) = β, i.e., f is surjective. Hence f is in fact an isomorphismn .
m
In fact, we establish from the value (14.105) that the representation (14.108) becomes
(1 + cα)z − cα2
ϕ(z) = .
cz + (1 − cα)
n
See, for instance, [42, p. 132].
14.4. Constructive Proof of the Riemann Mapping Theorem 453
(e) Now we have Gα,β = {ϕ ∈ Λ | ϕ(α) = α and ϕ(β) = β}. This refers to the case c 6= 0 and
a + d 6= ±2.
– Every ϕ ∈ Gα,β satisfies the required equation. Since α and β are roots of
the equation (14.92), the formula (14.102) also holds forpβ. Obviously, a − βc 6= 0.
Otherwise, it implies the contradiction that a + d = ± (a + d)2 − 4. We observe
that
az+b
ϕ(z) − α cz+d −α
= az+b
ϕ(z) − β cz+d −β
(a − αc)z + (b − dα)
=
(a − βc)z + (b − dβ)
(a − αc)z + α(cα − a)
=
(a − βc)z + β(cβ − a)
z−α
=γ· ,
z−β
where
a − αc
γ= ∈ C.o (14.109)
a − βc
– Gα,β is a subgroup of Λ. Since id fixes α and β, we have id ∈ Gα,β . For every
ϕ, φ ∈ Gα,β , let γϕ and γφ be their corresponding complex numbers respectively.
Since
ϕ(φ(z)) − α φ(z) − α z−α
= γϕ · = γϕ · γφ · , (14.110)
ϕ(φ(z)) − β φ(z) − β z−β
we have ϕ ◦ φ ∈ Gα,β . Assume that ϕ ∈ Gα,β was a constant map. Then it implies
that α = β, a contradiction. In addition, γϕ 6= 0. Otherwise, ϕ(z) = α for all z ∈ S 2
which is impossible. Next, if ϕ ∈ Gα,β , then ϕ−1 also fixes α and β, and we have
ϕ(z) − α z−α
= γϕ · . (14.111)
ϕ(z) − β z−β
az + b
ϕ(z) = .
d
Put this into the equation (14.111) with γϕ = 1 and after simplification, we conclude
that ϕ(z) = z, i.e., the kernel of g is {id}. Let γ ∈ C \ {0}. By changing the subject
of the formula (14.109) to c and using the formula α + β = a−d c , we can show that
αγ − β
d= a.
α − βγ
o
This number is called the multipler of the transformation ϕ, read [41, pp. 15, 16].
454 Chapter 14. Conformal Mapping
αβ(1 − γ) √ 1
b=− γ+√ . (14.113)
(1 + γ)(α − β) γ
Now it is a routine task to check that the linear fractional transformation ϕ with
the coefficients given by the formulas (14.112) and (14.113) has α and β as its fixed
points, ad − bc = 1 and satisfies
g(ϕ) = γ.
In other words, the map g is surjective and hence, an isomorphism.
(f) By the hypothesis, the fixed points are finite. The following proof is due to Drazin [33]
az+b
who verified that the linear fractional transformation ϕ(z) = cz+d has invariant circles if
(a+d)2
and only if its determinant ∆ 6= 0 and ∆ is real.
– Case (i): ϕ has a unique finite fixed point. Recall from the explicit form
(14.108) that
(1 + βα)z − βα2
ϕ(z) = , (14.114)
βz + (1 − βα)
where β = c 6= 0. Simple algebra gives
1
βϕ(z) = 1 + βα − . (14.115)
βz + (1 − βα)
Define
ϕ∗ = βϕ + (1 − βα) and ζ = βz + (1 − βα). (14.116)
Then the equation (14.115) becomes
1
ϕ∗ (ζ) = 2 − . (14.117)
ζ
Consequently, this change of variables establishes a one-to-one correspondence be-
tween the invariant circles of the linear fractional transformations (14.114) and (14.117).
Let C ∗ = {ζ ∈ C | |ζ − ρeiθ | = R} be an invariant circle of ϕ∗ , where ρ, θ, R are
real and R > 0. Consider the two points (ρ − R)eiθ and (ρ + R)eiθ which are the
endpoints of a diameter of C ∗ . Since ϕ∗ is conformal, the points P = ϕ∗ (ρ − R)eiθ
and Q = ϕ∗ (ρ + R)eiθ are also endpoints of a diameter of C ∗ so that |P − Q| = 2R
and 21 (P + Q) = ρeiθ . Notice that
1 1
P =2− and Q = 2 − , (14.118)
(ρ − R)eiθ (ρ + R)eiθ
14.4. Constructive Proof of the Riemann Mapping Theorem 455
so we have
ρe−iθ
ρ2 − R2 = ±1 and 2 = ρeiθ + .
ρ2 − R2
If ρ2 − R2 = −1, then we have 1 = iρ sin θ which is impossible. Therefore, we must
have
R2 = ρ2 − 1 and 1 = ρ cos θ.
In this case, C ∗ are circles with centers 1 ± iR. By the transformation (14.116), the
invariant circles of ϕ satisfy the equations
R R
z − α ± i = ,
β |β|
where R > 0.
– Case (ii): ϕ has two finite fixed points. By the expressions (14.112) and (14.113),
the explicit form of ϕ (after the cancellation of the common coefficient) is given by
(α − βγ)z − αβ(1 − γ)
ϕ(z) = (14.119)
(1 − γ)z + (αγ − β)
where
(1 − γ)z + (αγ − β)
ζ= √ . (14.120)
∆
1
Define ϕ∗ = ∆− 2 [(1 − γ)ϕ + (αγ − β)]. Then we have
h √ i
1 ∆ 1+γ 1
ϕ∗ (ζ) = ∆− 2 (α − β)(1 + γ) − = √ − . (14.121)
ζ γ ζ
1+γ 1 1+γ 1
P = √ − and Q = √ −
γ (ρ − R)eiθ γ (ρ + R)eiθ
so that
1+γ ρe−iθ
ρ2 − R2 = ±1 and √ = ρeiθ + 2 .
γ ρ − R2
1+γ
Denote χ = 2 γ.
√ Thus we have either
or
R2 = ρ2 − 1 and χ = ρ cos θ. (14.123)
Since ρ and θ are real, the expressions involving χ in (14.122) and (14.123) show that
it is either purely real or purely imaginary.
456 Chapter 14. Conformal Mapping
(1+γ)2 (1−γ)2
where R2 ≥ 4γ −1 = 4γ .
Remark 14.5
(a) The expressions (14.107) and (14.111) are called the normal forms of the linear
fractional transformation ϕ.
(b) The number of finite fixed points can be used to classify the linear fractional transfor-
mations ϕ. In fact, we rewrite the multiplier (14.109) as ρeiθ . If ρ 6= 1 and θ = 2nπ,
then ϕ is called a hyperbolic transformation. If ρ = 1 and θ 6= 2nπ, then it is
called an elliptic transformation. If ρ > 0 but ρ 6= 1 and θ 6= 2nπ, then it is
called a loxodromic transformation. The case for one finite fixed point is called
a parabolic transformation and it can be thought as corresponding to ρ = 1 and
θ = 2nπ. See [1, §3.5, pp. 84 – 89] and [41, pp. 15 – 23] for further details.
Problem 14.32
Rudin Chapter 14 Exercise 32.
14.4. Constructive Proof of the Riemann Mapping Theorem 457
Π = {ω = X + iY ∈ C | X > 0}.
The images of the upper semi-circle and the lower semi-circle under ϕ are the positive Y -axis
and the negative Y -axis respectively. Next, the mapping
1+z
ζ = φ(ω) = log ω = log
1−z
maps Π conformally onto the horizontal strip
Figure 14.4: The conformal mapping ψ(z) = φ ϕ(z)
Finally, it is easy to see that the mapping ζ 7→ iζ carries the horizontal strip S conformally
onto the vertical strip
H = {ω = X + iY ∈ C | − π2 < X < π
2 and Y ∈ R}
(a) Since ez maps the horizontal strip {z ∈ C | a < Re z < b} conformally onto the annulus
π π
A(ea , eb ), our conformal mapping f carries U onto the annulus A(e− 2 , e 2 ).p Further-
more, the images of the upper semi-circle and the lower semi-circle under f are the circles
π π
C 0, e− 2 and C 0, e 2 respectively, see Figure 14.5 for details below.
p
By direct differentiation, we see that
2i n 1 +zo
f ′ (z) = 2
exp i log 6= 0
1−z 1−z
in U , so f is conformal in U by Theorem 14.2.
458 Chapter 14. Conformal Mapping
π π
Figure 14.5: The conformal mapping f : U → A(e− 2 , e 2 ).
where b ∈ (−1, 0) ∪ (0, 1) and t ∈ (0, π) ∪ (π, 2π). Straightforward computation gives
which implies
1+z 1 1 + z 2 1+z
log = log + i arg
1−z 2 1 − z 1 −z
1 (1 − b2 )2 sin2 2t cos2 2t + b2 cos2 t
2b
= log 2
+ i tan−1 .
2 sin2 2t (sin2 2t + b2 cos2 2t )2 (1 − b2 ) sin t
Consequently, we have
h 2b i
f (z) = exp − tan−1
(1 − b2 ) sin t
ni
(1 − b2 )2 sin2 2t cos2 2t + b2 cos2 2t o
× exp log (14.125)
2 sin2 2t (sin2 2t + b2 cos2 2t )2
so that h i
2b
|f (cos t, b sin t)| = exp − tan−1 .
(1 − b2 ) sin t
Furthermore, suppose that
(1 − b2 )2 sin2 2t cos2 2t + b2 cos2 t
2
F (b, t) =
sin2 2t (sin2 2t + b2 cos2 2t )2
which is continuous on [(−1, 0) ∪ (0, 1)] × [(0, π) ∪ (π, 2π)]. Now for any fixed b, we have
F (b, t) → ∞ as t → 0. This implies that arg f (z) takes any value in [0, 2π].
14.4. Constructive Proof of the Riemann Mapping Theorem 459
1
For example, put b = 2 into the equation (14.125), we see that
4 hi (9 sin2 2t cos2 t
+ 4) cos2 2t i
f (z) = exp − tan−1 × exp log 2
,
3 sin t 2 sin2 2t (4 sin2 t
2 + cos2 2t )2
where t ∈ (0, π) ∪ (π, 2π). For t ∈ (0, π), the locus of f (z) is pictured in Figure 14.6:
Similarly, Figure 14.7 shows the locus of f (z) for t ∈ (π, 2π):
(f) Suppose that z = r + (1 − r)eiθ , where 0 < r < 1 and −π ≤ θ ≤ π. Then it is easy to
check that
1+z r cot θ2
ω = ϕ(z) = = +i
1−z 1−r 1−r
r
so that ϕ maps the circle C(r; 1 − r) onto the vertical line Re ω = 1−r . Hence ϕ maps the
r
disc E = D(r; 1 − r) conformally onto the vertical strip {ω | 0 < Re ω < 1−r }. Next, we
have
1 r 2 + cot2 2θ cot 2θ
log ϕ(z) = log + i arg
2 (1 − r)2 r
so that
cot 2θ hi r 2 + cot2 2θ i
f (z) = exp − arg × exp log .
r 2 (1 − r)2
θ
θ cot
Fix an r. As θ runs through [−π, π], 2 will run through [− π2 , π2 ] so that arg r 2 runs
π π cot θ
through [− 2 , 2 ]. Since |f (z)| = exp(− arg r 2 ), this means that f C(r; 1 − r) is a curve
π π
connecting the two circles C(0; e− 2 ) and C(0; e 2 ), and thus f maps E onto the annulus
π π
A(e− 2 , e 2 ), see Figure 14.8 when r = 0.25.
Remark 14.6
In the literature, there are two ways to define conformal maps. The first definition says that
a holomorphic function f to be conformal at all points with f ′ (z) 6= 0. This is the only
Rudin uses in his book, see also [2, p. 73]. The second way to define a conformal map f is
that it is one-to-one and holomorphic on an open set in the plane, see for example [109, p.
206].
Problem 14.33
Rudin Chapter 14 Exercise 33.
Proof.
holds for all z ∈ U . Since |α| < 1, the radii of convergence of the power series in the
1
expression (14.126) are |α| > 1. Consequently, they converge absolutely and uniformly in
U so that an interchange of integration and summation is legitimate. In terms of polar
coordinates, this means that
Z Z ∞
X
1 ′ 2
ϕα dm = 1 − |α| (αz)n (αz)k dm
π U π U n,k=0
∞ Z
X
1−|α|2
= (αz)n (αz)k dm
π
n,k=0 U
∞ Z 2π Z
1− |α| X
2 1
= (αreiθ )n (αre−iθ )k r dr dθ
π
n,k=0 0 0
2 X∞ Z 2π Z 1
1−|α|
= · (αn αk ) r n+k+1 ei(n−k)θ dr dθ. (14.127)
π 0 0
n,k=0
If n 6= k, then Z Z
2π 1
r n+k+1 ei(n−k)θ dr dθ = 0.
0 0
Thus the expression becomes
Z 2 X∞ Z 2π Z 1
1 ′
ϕα dm = 1 − |α| · |α|2n
r 2n+1 dr dθ
π U π 0 0 n=0
X∞
1− |α|2 π
= · |α|2n
π n=0
n+1
∞
X |α|2n
= 1 − |α|2 . (14.128)
n+1
n=0
which has radius of convergence 1. Substituting this with z = −|α|2 into the right-hand
side of the formula (14.128), we obtain finally that
Z
1 ′ 2
ϕα dm = 1 − |α| log 1
.
π U |α|2 1 − |α|2
Remark 14.7
Problem 14.33(a) is a special case of the classical result Lusin Area Integral, see [61, p.
150].
CHAPTER 15
Zeros of Holomorphic Functions
Problem 15.1
Rudin Chapter 15 Exercise 1.
bn −an
Proof. Let S be the set in which the infinite product converges uniformly. Define un (z) = z−bn .
Note that
z − an bn − a n
=1+ = 1 + un (z).
z − bn z − bn
Suppose that
δ = d S, {bn } = inf{|z − ω| | z ∈ S and ω ∈ {bn }} > 0.
Then it is easy to see that
1
|un (z)| ≤ |bn − an | < ∞
δ
for every n ∈ N and z ∈ S. Furthermore, we know that
∞
X ∞
X ∞
|bn − an | 1X
|un (z)| = ≤ |bn − an | < ∞
|z − bn | δ
n=1 n=1 n=1
Y∞
z − an
f (z) = (15.1)
n=1
z − bn
converges uniformly on S.
z−an
Clearly, S ◦ is an open set in C, fn (z) = z−bn ∈ H(S ◦ ) for n = 1, 2, . . . and fn 6≡ 0 in any
component of S ◦ . By the above paragraph,
∞
X ∞
X
|1 − fn (z))| = |un (z)|
n=1 n=1
converges uniformly on every compact subset of S ◦ . By Theorem 15.6, the infinite product
(15.1) is holomorphic in S ◦ . This ends the proof of the problem.
463
464 Chapter 15. Zeros of Holomorphic Functions
Problem 15.2
Rudin Chapter 15 Exercise 2.
Proof. Denote λ to be the order of the entire function f . By the definition, we have
λ = inf{ρ | |f (z)| < exp(|z|ρ ) holds for all large enough |z|}.
f (n) (0)
Using the fact an = n! and Theorem 10.26 (Cauchy’s Estimates), we have
λ
er
|an | ≤ (15.2)
rn
for large enough r. Let g(r) = r −n exp(r λ ), where r > 0. Applying elementary differentiation,
we can show that g attains its minimum
λn n
λ
exp
n λ
1
− λ1
at r = n λ
λ . Note that r is large if and only if n is large. Thus it follows from the inequality
(15.2) that
λn n eλ n
λ λ
|an | ≤ exp =
n λ n
holds for all large enough n.
Consider the entire functions f (z) = exp(z k ), where k = 1, 2, . . .. It is clear that λ = k. By
k 1
the power series expansion of ez , we have ank = n! . By induction, we obtain
1 e n
|ank | = <
n! n
for every large enough n. Consequently, the above bound is not close to best possible. This
completes the analysis of the proof.
Problem 15.3
Rudin Chapter 15 Exercise 3.
z
Proof. The part of finding solutions of ee = 1 has been solved in [123, Problem 3.19, pp. 44 –
45]. In fact, they are given by
π
ln(2kπ) + i 2nπ + , if k > 0 and n ∈ Z;
2
z= (15.3)
ln(−2kπ) + i 2nπ + 3π , if k < 0 and n ∈ Z.
2
Denote the zeros (15.3) by zk,n , where k, n ∈ Z and k 6= 0, see Figure 15.1.
Assume that f was an entire function of finite order having a zero at each (15.3). Suppose
further that f 6≡ 0. Consider the disc D(0, RN ), where RN = N π and N is a sufficiently large
positive integer. Then we have zk,n ∈ D(0, RN ), where
exp(N π) exp(N π)
−1≤k ≤ .
2π 2π
15.1. Infinite Products and the Order of Growth of an Entire Function 465
Therefore, we gain
exp(N π)
n RN ≥
4π
so that
log n RN N π − log 4π
≥ →∞
log RN log N π
as N → ∞, but it means that f is of infinite order, a contradiction. Hence no such entire
function exists and we finish the proof of the problem.
Problem 15.4
Rudin Chapter 15 Exercise 4.
• Both functions have a simple pole with residue 1 at each integer. Since sin πz
has a simple zero at every integer N , we have
π cos πz π cos πN
Res (π cot πz; N ) = Res ;N = =1
sin πz π cos πN
which shows that π cot πz has a simple pole with residue 1 at each integer.
We claim that
X 1 1 X z
+ = (15.4)
z−n n n(z − n)
n∈Z n∈Z
n6=0 n6=0
converges absolutely and uniformly on compact subsets of C \ Z. To see this, let |z| ≤ R
with R > 0. Then we have
X |z| X R X R X 1
≤ ≤ n ≤ 2R <∞
|n| · |n − z| |n| · (|n| − R) |n| · | 2 | n2
|n|≥2R |n|≥2R |n|≥2R n∈Z
n6=0
z z 2z
+ = 2 ,
n(z − n) (−n)(z + n) z − n2
1 X 1 1
∞
1 X 2z 1 X z
f (z) = + = + = + + .
z z 2 − n2 z n(z − n) z z−n n
n=1 n∈Z n∈Z
n6=0 n6=0
By Theorem 10.28, we see that f ∈ H(C \ Z). Furthermore, for each N ∈ N, we see that
hz − N z−N X 1 1 i
lim (z − N )f (z) = lim +1+ + (z − N ) + =1
z→N z→N z N z−n n
n∈Z
n6=0,N
and
h (z − N )2 (z − N )2 X 1 1 i
lim (z−N )2 f (z) = lim +(z−N )+ +(z−N )2 + = 0,
z→N z→N z N z−n n
n∈Z
n6=0,N
so it follows from [11, Theorem 9.5, p. 118] that f also has a simple pole with residue 1
at each integer N . Simple computation gives
N
1
X N
X
1 1 1
+ =− + ,
z−n n z z−n
n=−N n=−N
n6=0
N
1
X N
X
1 1 1
f (z) = + lim + = lim .
z N →∞ z−n n N →∞ z−n
n=−N n=−N
n6=0
15.1. Infinite Products and the Order of Growth of an Entire Function 467
= f (z).
e−2πy + e−2πix
cot πz = i ·
e−2πy − e−2πix
so that
|e−2πy | + 1
| cot πz| ≤ −2πy < ∞.
|e | − 1
On the other hand, we write
X ∞
1 2(x + iy)
f (z) = + .
x + iy x2 − y 2 − n2 + 2ixy
n=1
1
√
If y > 1 and |x| ≤ 2, we have |x + iy| ≤ 2y and
p 1 y 2 + n2
|x2 − y 2 − n2 + 2ixy| = [x2 − (y 2 + n2 )]2 + 4x2 y 2 ≥ (y 2 + n2 ) − ≥ .
4 2
Thus they imply that
∞
X 2|x + iy|
|f (z)| ≤ 1 + 2 − y 2 − n2 + 2ixy|
n=1
|x
468 Chapter 15. Zeros of Holomorphic Functions
∞
√ X y
≤1+4 2
y 2 + n2
n=1
√ Z ∞
y dx
≤1+4 2 . (15.5)
0 y 2 + x2
By the change of variable x = yt, it is easily checked that the integral in the inequality
(15.5) becomes Z ∞ Z ∞
y dx dt
2 + x2
=
0 y 0 1 + t2
which implies that |f (z)| ≤ M for some M > 0 and for all |x| ≤ 12 and y > 1. Similarly,
f (z) is also bounded for all |x| ≤ 21 and y < −1. Consequently, we have shown that ∆(z)
is a bounded entire function and Theorem 10.23 (Liouville’s Theorem) says that it is in
fact a constant.
To find this constant, we note that
h1 ∞
X 2iy i
lim f (iy) = lim +
y→∞ y→∞ iy n=1
−y 2 − n2
∞
X y
= −2i lim
y→∞ y 2 + n2
n=1
Z ∞
dt
= −2i
0 1 + t2
= −πi
and
lim π cot iπy = −πi.
y→∞
sin πz
• The product representation of . If g(z) = sin πz, then it is clear that
πz
g′ (z)
= π cot πz.
g(z)
Consequently, we observe from the representation (15.6) that
∞
g′ (z) 1 X 2z
= + . (15.7)
g(z) z z 2 − n2
n=1
for every z ∈ C \ Z. Substituting the result (15.9) into the formula (15.7), we see that
g′ (z) P ′ (z)
=
g(z) P (z)
∞
Y z2
in C \ Z. Since sin πz and πz (1 − ) agree on Z, the formula (15.10) holds in the
n=1
n2
whole plane.
Remark 15.1
Another way to prove Problem 15.4 is to consider the square CN with vertices (N + 12 )(±1±i)
for N ∈ N. According to Theorem 10.42 (The Residue Theorem), we have
Z X cot πz
1 cot πz
dz = Res ; zk , (15.11)
2πi CN z − ζ z−ζ
k
Problem 15.5
Rudin Chapter 15 Exercise 5.
Proof. By Theorem 15.9, our f is an entire function having a zero at each point zn . We claim
that
M (r) < exp(|z|k+1 )
for sufficiently large enough |z|. If |z| < 12 , then
h z2 zk i
log |Ek (z)| = Re log(1 − z) + z + + ··· +
2 k
1 1
k+1 k+2
= Re − z − z − ···
k+1 k+1
1 |z| |z|2
≤ |z|k+1 + + + ···
k+1 k+2 k+3
1 1
≤ |z|k+1 1 + + 2 + · · ·
2 2
≤ 2|z|k+1 . (15.12)
|z|2 |z|k
Since |Ek (z)| ≤ 1 + |z| exp |z| + 2 + ··· + k , we have
|z|2 |z|k
log |Ek (z)| ≤ log 1 + |z| + |z| + + ··· +
2 k
which gives
log |Ek (z)|
lim =0
z→∞ |z|k+1
so that if M1 > 0, then there exists a R > 0 such that
for all |z| > R. On the set S = {z ∈ C | 12 ≤ |z| ≤ R}, the function g(z) = |z|−(k+1) log |Ek (z)| is
continuous except at z = 1, where g(z) → −∞ as z → 1. Thus there is a constant M2 > 0 such
that
log |Ek (z)| ≤ M2 |z|k+1 (15.14)
for all z ∈ S.
Let M = max{2, M1 , M2 }. Now we combine the inequalities (15.12), (15.13) and (15.14) to
get
log |Ek (z)| ≤ M |z|k+1 (15.15)
for every z ∈ C. By the hypothesis, one can find an N ∈ N such that
∞
X N
X
1 1 1
< and ≤ N.
|zn |k+1 2M
n=1
|zn |k+1
n=N +1
Since M1 can be chosen arbitrary, we deduce from the inequality (15.13) that there exists a
R1 > 0 such that
|z|k+1
log |Ek (z)| ≤
2N
for all |z| > R1 . Let R2 = max{|z1 |R1 , |z2 |R1 , . . . , |zN |R1 }. Then it is obvious that
N
X z |z|k+1
log Ek ≤ (15.17)
zn 2
n=1
for all |z| > R2 . Finally, the inequalities (15.16) and (15.17) give
∞
X z
log |f (z)| = log Ek ≤ |z|k+1
zn
n=1
for all |z| > R2 . This proves our claim and thus f is of finite order. This completes the analysis
of the problem.
Problem 15.6
Rudin Chapter 15 Exercise 6.
X ∞
X X X∞
|zn |−p−ǫ = |zn |−p−ǫ ≤ 2−k(p+ǫ) n 2k+1 . (15.18)
|zn |≥1 k=0 2k ≤|zn |<2k+1 k=0
Since |f (z)| < exp(|z|p ), it follows from [100, Eqn. (2) & (3), p. 309] that
n(r) ≤ Cr p (15.19)
for some constant C > 0 and all sufficiently large enough r. Combining the inequalities (15.18)
and (15.19), we see that
X ∞
X ∞
X 1 k
|zn |−p−ǫ < C 2−k(p+ǫ) · 2(k+1)p = C · 2p < ∞.
2ǫ
|zn |≥1 k=0 k=0
Hence we have
∞
X
|zn |−p−ǫ < ∞
n=1
Problem 15.7
Rudin Chapter 15 Exercise 7.
472 Chapter 15. Zeros of Holomorphic Functions
Proof. Without loss of generality, we may assume that f 6≡ 0. By the definition (see Problem
15.2), f is of finite order. In the disc D 0; N + 21 for large enough positive integer N , the
number of zeros of f inside D 0; N + 12 is at least N 2 , i.e., n(N + 21 ) ≥ N 2 . By the hypothesis,
we know that M (r) < exp(|z|α ), so it follows from [100, Eqn. (4), p. 309] that
log n(N + 21 )
2 ≤ lim sup ≤ α.
N →∞ log N
Therefore, if 0 < α < 2, then no entire function can satisfy the hypotheses of the problem. In
other words, f (z) = 0 for all z ∈ C if 0 < α < 2, completing the proof of the problem.
Problem 15.8
Rudin Chapter 15 Exercise 8.
Proof. Let A = {zn }. We are going to verify the results one by one.
• f is independent of the choice of γ(z). Suppose that γ, η : [0, 1] → C are paths from
0 to z and they pass through none of the points zn . Then Γ = γ − η is a simple closed
path. Let {z1 , z2 , . . . , zN } be the set of points which are surrounded by Γ for some N ∈ N.
Now we follow from Theorem 10.42 (The Residue Theorem) that
Z N
X N
X
1
g(ζ) dζ = Res (g; zn ) = mn
2πi Γ(z) n=1 n=1
which gives
Z N
X Z
g(ζ) dζ = 2πi mn + g(ζ) dζ.
γ(z) n=1 η(z)
• f ∈ H(C \ A). The definition of f shows that the holomorphicity of f depends on the
holomorphicity of the integral. Let z ∈ C \ A and denote
Z
G(z) = g(ζ) dζ.
γ(z)
Now there exists a disc D(0; R) containing z. Let h be so small that z + h ∈ D(0; R)
and [z, z + h] ∩ A = ∅. By Definition 10.41, D(0; R) contains only finitely many points
of A. Without loss of generality, we may assume that {z1 , z2 , . . . , zN } ⊆ D(0; R) for some
positive integer N . Then both γ(z) and γ(z + h) must lie in Ω = D(0; R) \ {z1 , z2 , . . . , zN }.
Suppose that γ(z) consists of only horizontal or vertical line segments in Ω and γ(z + h)
shares the same path with γ(z) until the point z. Since [z, z +h]∩A = ∅, we can connect z
and z + h by another set of horizontal or vertical line segments in a way that only triangles
are produced. Figure 15.2 illustrates this setting.
15.2. Some Examples 473
where ∆1 , ∆2 , . . . , ∆m are the triangles produced. Obviously, the compactness of the set
∆1 ∪ ∆2 ∪ · · · ∪ ∆m ensures that there corresponds an open set Ω′ in Ω such that ∆k ⊆ Ω′
for every k = 1, 2, . . . , m. Since g ∈ H(Ω), Theorem 1013 (The Cauchy’s Theorem for a
Triangle) implies that each integral in the summation (15.20) is 0. As g is continuous at
z, we can write g(ζ) = g(z) + ǫ(ζ), where ǫ(ζ) → 0 as ζ → z. Therefore, the expression
(15.20) becomes
Z Z Z
G(z + h) − G(z) = g(z) dζ + ǫ(ζ) dζ = hg(z) + ǫ(ζ) dζ. (15.21)
[z,z+h] [z,z+h] [z,z+h]
Since Z
ǫ(ζ) dζ ≤ sup |ǫ(ζ)| · |h|,
[z,z+h] ζ∈[z,z+h]
Since zn is a simple pole of g with residue mn , there exists a δn > 0 such that
mn
g(z) = + h(z) (15.23)
z − zn
474 Chapter 15. Zeros of Holomorphic Functions
for all |z − zn | < 2δn and h ∈ H D(zn ; 2δn ) .a In fact, we can choose δn small enough
such that D(zn ; 2δn ) contains only the pole zn . Suppose that z lies on the line segment
joining 0 and zn , z ∈ D(zn ; δn ) and θn = arg zn . We split the path γ(z) into two paths
γ1 (z) and γ2 (z) as follows: The path γ1 (z) is the line segment from 0 to zn − δn eiθn . If it
passes through a pole of g, then we can make a small circular arc around that pole. The
path γ2 (z) is the line segment from zn − δn eiθn to z.
On γ1 (z), since δn is fixed and g is continuous on γ1 (z), there exists a positive constant
M1 such that Z
g(ζ) dζ ≤ M1 . (15.24)
γ1 (z)
so that γ2 (z; 0) = zn − δn eiθn and γ2 (z; 1) = z. Substitute γ2 (z; t) into the Laurent series
(15.23) to getb
Z Z 1n o Z
z − zn − δn eiθn dt
g(ζ) dζ = mn + h(ζ) dζ
γ2 (z) 0 t z − zn − δn eiθn − δn eiθn γ2 (z)
1 Z
= mn ln{t[z − (zn − δn eiθn )] − δn eiθn } + h(ζ) dζ
0 γ (z)
Z 2
= mn ln(z − zn ) − i(θn + π) − ln δn + h(ζ) dζ. (15.25)
γ2 (z)
Since h ∈ H D(zn ; 2δn ) , there exists a positive constant M2 such that
Z
h(ζ) dζ ≤ M2
γ2 (z)
which gives
nZ o
exp g(ζ) dζ ≤ |z − zn |mn eM2 × exp(−mn ln δn ). (15.26)
γ2 (z)
Problem 15.9
Rudin Chapter 15 Exercise 9.
Proof. Suppose that z1 , z2 , . . . , zn are the zeros of f , listed according to their multiplicities, such
that z1 , z2 , . . . , zn ∈ D(0; β). Define
n
Y z − zk
g(z) = .
1 − zk z
k=1
log α
n≤ . (15.27)
log β
1
(a) Put α = β = 2 into the inequality (15.27), we conclude that n ≤ 1.
Problem 15.10
Rudin Chapter 15 Exercise 10.
Proof. Let I be the ideal generated by the set {gN | N ∈ N}. By Definition 15.14, every element
of I is of the form
fN1 gN1 + fN2 gN2 + · · · + fNk gNk (15.28)
for some increasing sequence {Nk } of positive integers, where fN1 , fN2 , . . . , fNk are entire. By
the definition of gN , if N < M , then there exists an entire function hM such that gN = hM gM .
Consequently, the element (15.28) can be expressed as
Assume that I was principal. One can find an entire function g ∈ I such that I = [g]. By the
representation (15.29), we know that g = f gN for some entire f and some N ∈ N. Thus we
may assume that there exists some positive integer N such that I = {f gN | f is entire}. Then
we have
gN +1 = f gN (15.30)
for some entire f . However, since gN (N ) = 0 but gN +1 (N ) 6= 0, the equation (15.30) is a
contradiction. Hence I is not principal and we have completed the proof of the problem.
Problem 15.11
Rudin Chapter 15 Exercise 11.
z−1
Proof. Recall from [100, Eqn. (6), p. 281] that ϕ−1 (z) = z+1 is a conformal one-to-one mapping
of Π = {z ∈ C | Re z > 0} onto U . Therefore, we can reduce the problem to the existence of a
bounded holomorphic function f in U which is not identically zero and its zeros are precisely at
iyn
αn = ,
2 + iyn
where n = 1, 2, . . .. In this case, §15.22 shows that {αn } satisfies
∞
X
(1 − |αn |) < ∞.
n=1
(a) We have
X∞ i log n X ∞ p
4 + (log n)2 − log n
1− = p
n=1
2 + i log n n=1 4 + (log n)2
∞
X 4
= p p . (15.31)
n=1 4 + (log n)2 · 4 + (log n)2 + log n
p
For n ≥ 3, we have 4 + (log n)2 ≤ 2 log n so that
X∞ i log n X ∞
2 2X1
∞
1− ≥ ≥ .
n=3
2 + i log n n=3
3(log n)2 3 n=3 n
Hence the series (15.31) diverges and thus no such bounded holomorphic function in Π.
X∞ i√n X ∞ √ √
4+n− n
1− √ = √
n=1
2+i n n=1
4+n
∞
X 4
= √ √ √
n=1
4+n· 4+n+ n
X∞
2
≤ √ p
4
n=1 4+n· (4 + n)n
15.2. Some Examples 477
∞
X 1
=2 5 1
n=1 (n + 4) 4 · n 4
X∞
1
<2 3
n=1 n 2
< ∞,
X∞ in X ∞ √
4 + n2 − n
1− = √
n=1
2 + in n=1
4 + n2
∞
X 4
= √ √
2
4+n · 4 + n2 + n
n=1
X∞
2
< 2
n=1
n
< ∞.
Problem 15.12
Rudin Chapter 15 Exercise 12.
Proof. Let E = { α1n } and Ω = C \ E. The Blaschke condition implies that αn → eiθ as n → ∞
for some real θ, so eiθ ∈ E. Let K be a compact subset of Ω. Since E is closed and K ∩ E = ∅,
Problem 10.1 shows that δ = d(K, E) > 0. In particular, we have |eiθ − z| ≥ δ for every z ∈ K
or equivalently
|1 − e−iθ z| ≥ δ (15.32)
for all z ∈ K. It is easy to see from the representation
∞
Y αn − z |αn |
B(z) = ·
1 − αn z αn
n=1
1
that B has a pole at each point αn . Next, the nth term of the series
∞
X αn − z |αn |
1 − · (15.33)
1 − αn z αn
n=1
is given by
α + |α |z 1 + |z|
n n
· 1 − |αn | ≤ 1 − |αn | · . (15.34)
(1 − αn z)αn |1 − αn z|
478 Chapter 15. Zeros of Holomorphic Functions
Problem 15.13
Rudin Chapter 15 Exercise 13.
for some k ∈ N. Suppose that αN −1 < r < αN . Since 0 < αn < 1 and αn < αn+1 for all positive
integers n, we obtain r k < 1 and
α −r αn − r
n
= <1
1 − αn r 1 − αn r
for every n = N, N + 1, . . .. Consequently, we have
Y ∞
αn − r Y αn − r Y αn − r
N −1 ∞ N
Y −1
r − αn
|B(r)| = r k < × < . (15.36)
n=1
1 − αn r n=1
1 − αn r 1 − αn r n=1
1 − αn r
n=N
Simple algebra shows that αN −αn > r−αn > 0 and 1−αn r > 1−αn for each n = 1, 2, . . . , N −1,
the inequality (15.36) reduces to
N
Y −1
αN − αn
|B(r)| < . (15.37)
1 − αn
n=1
αN −αn n2
Since 1−αn =1− N2
and log x ≤ x − 1 for x > 0, the inequality (15.37) becomes
−1
n2
N
Y
|B(r)| < 1−
n=1
N2
15.3. Problems on Blaschke Products 479
h N
Y −1
n2 i
≤ exp log 1− 2
N
n=1
h NX
−1 n2 i
= exp log 1 − 2
n=1
N
N X−1 2
n
≤ exp −
N2
n=1
h (N − 1)(2N − 1) i
≤ exp −
6N
h (N − 1)2 i
< exp −
3N
2 N
< e 3 e− 3
N
< 2e− 3 . (15.38)
Hence, by combining the fact r → 1 if and only if N → ∞ and the estimate (15.38), we have
established that B(r) → 0 as r → 1 and r ∈ (0, 1), as required. This completes the proof of the
problem.
Problem 15.14
Rudin Chapter 15 Exercise 14.
Proof. Consider αn = 1 − e−n and xn = 1 − 12 e−n for n = 1, 2, . . .. Now we are going to modify
the sequence {αn } and select a subsequence of {xn } so that a Blaschke product with zeros at
the modified sequence satisfies the requirement.
We note from [10, Eqn. (15), p. 12] that we can write
p
Y x2p − αn Y αn − x2p
4p−1 ∞
Y αn − x2p
|B(x2p )| = × ×
1 − αn x2p 1 − αn x2p 1 − αn x2p
n=1 n=p+1 n=4p
For n = 1, 2, . . . , p, since
x2p − αn x2p − αp
≥ ≥ 1 − e−p ,
1 − αn x2p 1 − αp x2p
we obtain (1 − e−p )p < T1 (p) < 1 which implies that
give
lim T3 (p) = 1. (15.41)
n→∞
Now here is the trick: For a positive integer p, we first replace αp+1 , αp+2 , . . . , α4p−1 by α4p .
Then we have
4p−1
Y α4p − x2p α − x 3p−1 4 3p−1
4p 2p
1 > S2 (p) = = ≥ 1 − 2p
1 − α4p x2p 1 − α4p x2p e +2
n=p+1
so that
lim S2 (p) = 1. (15.42)
p→∞
Next, we select a subsequence {pk } of positive integers such that 4pk − 1 < pk+1 + 1 for each
k = 1, 2, . . .. This makes sure that
for each k = 1, 2, . . .. Then the modified sequence {α′n } will be the one that replaces only the
terms αpk +1 , αpk +2 , . . . , α4pk −1 by α4pk from the original sequence {αn } for k = 1, 2, . . .. Since
∞
X ∞
X
(1 − |α′n |) ≤ (1 − |αn |) < ∞,
n=1 n=1
Combining the limits (15.40), (15.41) and (15.42), we conclude immediately that
lim |B(x2pk )| = 1
k→∞
which means
lim sup |B(r)| = 1. (15.45)
r→1
Hence the two results (15.44) and (15.45) imply that the function (15.43) has no radial limit at
z = 1, completing the analysis of the problem.
Problem 15.15
Rudin Chapter 15 Exercise 15.
(a) If α = 0, then λ 6= 1 because ϕ is not the identity function. In this case, 0 is the unique
fixed point and ϕ(z) = λz. However, it implies that
∞
X
1<∞
n=1
which is impossible. Thus α 6= 0 and we deduce from the expression (15.46) that ϕ(z) = z
is equivalent to
αz 2 − (1 − λ)z − λα = 0. (15.47)
It is clear from the equation (15.47) that any fixed point must be of modulus 1 because if
z is a root of the equation (15.47), then the fact λ = λ1 implies that z 6= 0 and 1z is also
one of its roots.
– Case (i): Suppose that ϕ has a unique fixed point on T . Let it be b. Consider
the linear fractional transformation µ−1 (z) = 1+bz −1
z . Clearly, µ (∞) = b, so the
−1
conjugate ψ = µ ◦ ϕ ◦ µ fixes ∞ and then it can be expressed asc
ψ(z) = z + A
−nAb2 b
ϕn (0) = =b− ,
1 − nAb 1 − nAb
so for sufficiently large n, we have
1
|ϕn (0)| ≈ 1 − ,
n|A| − 1
z − b1
ν(z) = .
z − b2
Since ψ fixes 0 and ∞, it can be expressed as ψ(z) = kz for some nonzero complex
constant k. Since ψn (z) = kn z, we have
ϕn = ν −1 ◦ ψn ◦ ν.
c
Or we may apply Problem 14.31(c) directly here.
482 Chapter 15. Zeros of Holomorphic Functions
b2 z−b1
Note that ν −1 (z) = z−1 , so
z − b1
ϕn (z) = ν −1 kn ·
z − b2
b2 kn · z−b
z−b2 − b1
1
= z−b1
kn · z−b2
−1
(b2 kn − b1 )z + (1 − kn )b1 b2
= . (15.49)
(kn − 1)z + (b2 − b1 kn )
∗ Subcase (i): |k| < 1. Since |(b2 kn − b1 )z − (kn − b1 )z| = |(b2 − 1)kn z| and
|(1 − kn )b1 b2 − (b1 − kn )bn | = |(1 − b1 )b2 kn |, for sufficiently large enough n, the
expression (15.49) gives
(kn − b )z + (b − kn )b
1 1 2
|ϕn (z)| ≈ ≈ 1 − |k|n
−z + b2
so that 1 − |ϕn (z)| ≈ |k|n . In other words, ϕ satisfies the Blaschke condition by
[127, Theorem 6.9, p. 77].
∗ Subcase (ii): |k| > 1. In this case, for sufficiently large enough n, the expression
(15.49) implies that
(b − b1 )z + ( 1 − 1)b b
2 kn kn 1 2
|ϕn (z)| =
(1 − k1n )z + ( kbn2 − b1 )
(b − 1 )z + ( 1 − b )b
2 kn kn 2 1
≈
z − b1
1
≈1− n
|k|
which means 1 − |ϕn (z)| ≈ |k|1n . Hence the ϕ also satisfies the Blaschke condition
in this case.
∗ Subcase (iii): |k| = 1 and k is an N th root of unity for some N . Put
z = 21 into the expression (15.49), we have
1 1
ϕN =
2 2
so that
∞ h
X 1 i X∞ h 1 i
1 − ϕn ≥ 1 − ϕpN = ∞.
2 2
n=1 p=1
S = {einθ | n ∈ N}.
Given ǫ > 0. Let ℓ be an arc on T with angle ǫ. Choose N ∈ N such that 2π N < ǫ.
By the hypothesis, the (N + 1) points 1, eiθ , e2iθ , . . . , eN iθ are all distinct. As a
result, two of them must have a counterclockwise angle less than 2π N , i.e.,
2π
(p − q)θ < <ǫ
N
2πp
d
Otherwise, we have θ = q
for some p ∈ Z and q ∈ N. This implies that kq = 1, a contradiction.
15.3. Problems on Blaschke Products 483
for some p < q. This means that the points en(p−q)iθ with n ≥ 0 are distributed
on T at successive angles less than ǫ. Consequently, we have en(p−q)iθ ∈ ℓ for
some n ≥ 0.
Put z = 0 into the expression (15.49) to get
|1 − kn |
|ϕn (0)| = .
|b2 − b1 kn |
1
The claim ensures that there exists a sequence {nm } satisfying |b1 − b2 knm | < m
and then
m
|ϕnm (0)| > m · |1 − knm | >
2
for large enough m. This shows that ϕ satisfies the Blaschke condition.
(b) We claim that ϕ satisfies
ϕn (z) = z (15.50)
for some n ∈ N. To this end, if f ◦ ϕ = f for some nonconstant f ∈ H ∞, then we must
have
f ϕn (z) = f ϕ ϕn−1 (z) = f ϕn−1 (z) = · · · = f (z)
for every z ∈ U and every n = 1, 2, . . ..
– Case (i): ϕ has a unique fixed point in U . Recall from the equation (15.46) that
ϕ has a unique fixed point in U if and only if α = 0. In this case, we have ϕ(z) = λz
for some λ such that |λ| = 1. Obviously, we know that
ϕn (z) = λn z
for every n ∈ N and z ∈ U . If λ is an N th root of unity, then the expected result
(15.50) holds immediately for some N ∈ N. Otherwise, we fix z = z0 ∈ U \ {0} so
that
f (λn z0 ) = f (z0 )
for every n ≥ 0. By part (a) Subcase (iv) above, the set S ′ = {λn | n ≥ 0} is dense
on T which implies that
f (z) = f (z0 ) (15.51)
on C(0; |z0 |). By Theorem 10.18, the result (15.51) shows that f (z) = f (z0 ) for every
z ∈ U and this contradicts the Open Mapping Theorem.
– Case (ii): ϕ has a unique fixed point b on T . In this case, it follows from the
equation (15.48) that ϕn (z) → b as n → ∞ for every z ∈ U . This means that
f (z) = lim f ϕn (z) = f (b)
n→∞
Consequently, the linear fractional transformation ϕ must satisfy the claim (15.50).
Problem 15.16
Rudin Chapter 15 Exercise 16.
so we have
∞
X ∞ Z
X 1
(1 − |αj |) = dr. (15.52)
j=1 j=1 |αj |
Observe that
∞
X
χj (r) = n(r). (15.54)
j=1
Since each χj : [0, 1] → [0, ∞] is measurable for every j = 1, 2, . . ., we apply Theorem 1.27 to
interchange the summation and the integration in the equation (15.53) and then use the formula
(15.54) to gain
X∞ Z 1 X
∞ Z 1
(1 − |αj |) = χj (r) dr = n(r) dr.
j=1 0 j=1 0
Problem 15.17
Rudin Chapter 15 Exercise 17.
∞
X
Proof. Assume that B(z) = ck z k was a Blaschke product with ck ≥ 0 for all k = 0, 1, 2, . . .
k=0
and B(α) = 0 for some α ∈ U \ {0}. Combining Theorem 15.24 and the facte that |f (x)| ≤ λ
holds for almost all x if and only if kf kL∞ ≤ λ, we have
kB ∗ kL∞ (T ) ≤ 1 or B(eiθ ) ∈ L∞ (T ).
e
Refer to [100, p. 66]
15.3. Problems on Blaschke Products 485
Problem 15.18
Rudin Chapter 15 Exercise 18.
Thus it suffices to show that (z − 1)B ′ (z) is bounded in U . Suppose that {αn } is the sequence
of zeros of B. Then we know that {αn } ⊆ (0, 1) and it satisfies the Blaschke condition
∞
X
(1 − α2n ) < ∞. (15.58)
n=1
B ′ (z) X αn − z ′ αn − z −1
∞
=
B(z) n=1
1 − αn z 1 − αn z
486 Chapter 15. Zeros of Holomorphic Functions
∞
X 1 − α2n
B ′ (z) = − Bn (z).
(1 − αn z)2
n=1
It is well-known that |Bn (z)| < 1 for every z ∈ U . Now we observe from the assumption (15.59)
that
δ = min{α1 , α2 , . . .} = α1 > 0.
Geometrically, if z ∈ U and p > 1, then
Put p = α1n into the estimate (15.60) to get |1 − αn z|2 ≥ δ2 |1 − z|2 for every z ∈ U . Hence, the
condition (15.58) implies
X∞ ∞
1 − α2n 1 X
|(z − 1)2 B ′ (z)| ≤ |1 − z|2 · 2 · |B n (z)| ≤ 2
(1 − α2n ) < ∞
n=1
(1 − αn z) δ n=1
holds for all z ∈ U . Consequently, the modulus |(z − 1)2 B ′ (z)| is bounded in U which shows the
boundedness of f ′ in U , completing the proof of the problem.
Remark 15.2
Blaschke products serve as an important subclass of H(U ). If you are interested in the
literature of Blaschke products, you are suggested to read the book by Colwell [29].
Problem 15.19
Rudin Chapter 15 Exercise 19.
µ∗ (f ) = 0. (15.62)
15.4. Miscellaneous Problems and the Müntz-Szasz Theorem 487
lim µr (f ) < µ∗ (f ),
r→1
Problem 15.20
Rudin Chapter 15 Exercise 20.
Proof. If λN < 0 for some N ∈ N, then there exists a δ > 0 such that λN < δ < 0. By the
hypothesis, we actually have λn < δ < 0 for all n ≥ N , but this contradicts another hypothesis
that λn → 0 as n → ∞. Therefore, we have λ1 > λ2 > · · · > 0 which implies that
1 1
0< < < ··· .
λ1 λ2
Suppose that X is the closure of the C(I), the set of all finite linear combinations of the functions
1 1 1
1, t λ1 , t λ2 , t λ3 , . . . .
Problem 15.21
Rudin Chapter 15 Exercise 21.
Proof. Let {λn } be a sequence of distinct real numbers and λn > − 12 . Then the set of all finite
linear combinations of the functions
n−1
X
1 Y m − λk
n−1
min
tm − ak tλk
= √ .
ak ∈C
k=0
2 1 + 2m k=0 m + λk + 1
if and only if
Y
n−1
m − λk
lim sup =0
n→∞ m + λk + 1
k=0
if and only if
n−1
Y 2m + 1
n−1
Y 2λk + 1
lim sup 1 − × 1 − = 0.
n→∞ m + λk + 1 m + λk + 1
k=0 k=0
λk >m − 12 <λk ≤m
2m+1
Clearly, we have m + λk + 1 > 0. If λk > m, then we have m+λk +1 ∈ (0, 1). Otherwise, since
2λk +1
λk > − 12 , we have m+λk +1 ∈ (0, 1). By Theorem 15.5, the results (15.64) hold if and only if
∞
X ∞
X
2m + 1 2λk + 1
=∞ or = ∞. (15.65)
m + λk + 1 m + λk + 1
k=0 k=0
λk >m − 12 <λk ≤m
1
Suppose that λk > m. Then we have λk + 2 < m + λk + 1 < 2λk + 1 and we get
1 1 2
< <
2λk + 1 m + λk + 1 2λk + 1
which means that the first summation (15.65) is equivalent to
∞
X 1
= ∞. (15.66)
2λk + 1
k=0
λk >m
If − 12 < λk < m, then we have 1 < m + λk + 1 ≤ 2m + 1 so that the second summation (15.65)
is equivalent to
X∞
(2λk + 1) = ∞. (15.67)
k=0
− 21 <λk ≤m
holds. Finally, our desired result is proven if we combine Theorem 3.14 and the Weierstrass
Approximation Theorem [99, Theorem 7.26, p. 159]. This completes the proof of the problem.
15.4. Miscellaneous Problems and the Müntz-Szasz Theorem 489
Remark 15.3
The proof of Problem 15.21 follows basically the proof of Theorem 2.2 in [21]. For the version
of a complex sequence {λn }, please refer to [85, §12, pp. 32 – 36].
Problem 15.22
Rudin Chapter 15 Exercise 22.
Proof. Let M be the set of all finite linear combinations of the functions fn . It is clear that M
is a subspace of L2 (0, ∞). Let g ∈ L2 (0, ∞) be orthogonal to each fn , i.e.,
Z ∞
hfn , gi = fn (t)g(t) dt = 0
0
∞ n
X t
h(ζ) = ζ n−1 , (15.69)
n!
n=1
so we have h ∈ H D(0; ǫ) by Theorem 10.6. Since h is continuous on D(0; ǫ), we obtain from
Theorem 10.24 (The Maximum Modulus Theorem) and the Extreme Value Theorem that the
maximum of |h(ζ)| occurs on the boundary |ζ| = ǫ. By the expansion (15.69), we see that
∞ n
X t etǫ − 1
max |h(ζ)| ≤ · |ζ|n−1 = . (15.70)
|ζ|=ǫ
n=1
n! ǫ
490 Chapter 15. Zeros of Holomorphic Functions
By the Mean Value Theorem for Derivatives, we know that etǫ − 1 = ǫtetξ for some ξ ∈ (0, ǫ), so
we induce from the estimate (15.70) that
etζ − 1
≤ tetξ < tetǫ ,
ζ
where |ζ| < ǫ and t > 0. If z ∈ Π, then Re z > 2ǫ for some ǫ > 0. With this ǫ > 0, we may pick
ω ∈ Π satisfying the condition (15.68). Therefore, we have
e−tω − e−tz
−tz etζ − 1
= e · < te−tRe z · etǫ = te−t(Re z−ǫ) ≤ te−tǫ
ω−z ζ
for t > 0. Since te−tǫ , g ∈ L2 (0, ∞), Theorem 3.8 implies that te−tǫ ·g ∈ L1 (0, ∞). Consequently,
Theorem 1.34 (Lebesgue’s Dominated Convergence Theorem) can be applied to show the limit
and the integral can be interchanged in the following deduction:
F (ω) − F (z)
F ′ (z) = lim
ω→z ω−z
Z ∞ −ωt
e − e−tz
= lim · g(t) dt
ω→z 0 ω−z
Z ∞
e−ωt − e−tz
= lim · g(t) dt
ω→z ω−z
Z0 ∞
= −tetz g(t) dt.
0
so that F ≡ 0. Particularly, if we denote G(t) = χ(0,∞) (t)e−t g(t), then we see that
√ Z ∞ √ Z ∞ −t √
b
G(y) = 2π G(t)e−ity dt = 2π e g(t)e−ity dt = 2πF (1 + iy) = 0,
−∞ 0
Problem 15.23
Rudin Chapter 15 Exercise 23.
15.4. Miscellaneous Problems and the Müntz-Szasz Theorem 491
YN
λn − z |λn |
B(z) = · .
n=1
1 − λn z λn
N
X
Since (1 − |λn |) < ∞, Theorem 15.21 implies that B ∈ H ∞ and B has no zeros except at
n=1
B
the points λn . Consider the function g = 1−f . Then we have g ∈ H(U ). By Theorem 12.4, g is
continuous on U and λ − eiθ
n
=1
1 − λn eiθ
for every n = 1, 2, . . . , N . Therefore, we see that
|B(eiθ )| 1 1
|g(eiθ )| = ≤ ≤ .
|1 − f (eiθ )| |f (eiθ )| − 1 2
Consequently, it yields from Theorem 10.24 (The Maximum Modulus Theorem) that
1
|λ1 λ2 · · · λN | = |g(0)| < .
2
We end the proof of the problem.
492 Chapter 15. Zeros of Holomorphic Functions
CHAPTER 16
Analytic Continuation
Problem 16.1
Rudin Chapter 16 Exercise 1.
Proof. By Theorem 16.2, f has a singularity at some point eiθ . If we consider the power series
for f about the point 21 , then the representation
∞
X 1 k X f (k) ( 21 )
∞
1 k
f (z) = bk z − = z− (16.1)
2 k! 2
k=0 k=0
holds in D( 21 ; 21 ). Thus the radius of convergence of the power series (16.1) must be 12 . Otherwise,
the power series would define a holomorphic extension of f beyond eiθ , a contradiction.
Assume that f was regular at z = 1. Then the series
f (k) ( 21 ) 1 k
∞
X
x−
k! 2
k=0
P
converges at some x > 1. For every k ≥ 1, we derive from the representation f (z) = an z n
that
1 X∞
n(n − 1) · · · (n − k + 1)an
f (k) = .
2 2n−k
n=k
Now we deduce from the binomial theorem that
f (k) ( 1 ) 1 k X h X n(n − 1) · · · (n − k + 1) an i 1 k
∞
X ∞ ∞
2
x− = × n−k × x −
k! 2 k! 2 2
k=0 k=0 n=k
∞ X
X ∞
an 1 k
= Ckn · · x− . (16.2)
2n−k 2
k=0 n=k
an
Since an ≥ 0, Ckn · 2n−k ·(x− 12 )k ≥ 0. Consequently, the order of the summation in the expression
(16.2) can be switched (see [99, Exercise 3, p. 196]) so that
f (k) ( 12 ) 1 k X h X n 1 n−k 1 k i X
∞
X ∞ n ∞
x− = an Ck −0 · x− = an xn
k! 2 2 2
k=0 n=0 k=0 n=0
493
494 Chapter 16. Analytic Continuation
P
which means that the radius of convergence of f (z) = an z n is greater than 1. This is a
contradiction to our hypothesis and we have completed the analysis of the problem.
Problem 16.2
Rudin Chapter 16 Exercise 2.
Proof. Suppose that (f, D) and (g, D) can be analytically continued along γ to (fn , Dn ) and
(gm , Dm ) respectively. According to Definition 16.9, there exist chains Cf = {D0 , D1 , . . . , Dn }
′ }, where D = D ′ = D. Furthermore, there are numbers
and Cg = {D0′ , D1′ , . . . , Dm 0 0
holds for all ζ ∈ D1′ and hence also for every z ∈ Dn . Repeat the process also implies that the
equation
P fn (z), gm (ζ) = 0 (16.4)
holds in Dn and Dm . Finally, we can establish the required equation of the problem if we replace
fn and gm by f1 and g1 in the equation (16.4).
Obviously, this can be extended to n function elements (f1 , D), (f2 , D), . . . , (fn , D) provided
that P (z1 , z2 , . . . , zn ) is a polynomial in n variables, f1 , f2 , . . . , fn can be analytically continued
along a curve γ to g1 , g2 , . . . , gn and P (f1 , f2 , . . . , fn ) = 0 in D. In fact, our above proof only
uses the holomorphicity of the polynomial P , so similar results can be established if we only
require that P is a function of n variables such that P (. . . , zj , . . .) is holomorphic in each variable
zj for j = 1, 2, . . . , n. This completes the analysis of the problem.
Problem 16.3
Rudin Chapter 16 Exercise 3.
so that A(x, y) = u(x, y) + C for some constant C. Hence u is the real part of F − C ∈ H(Ω).
Next, we suppose that Ω is a region, but not simply connected. Let f ∈ H(Ω) and f (z) 6= 0
for every z ∈ Ω. Then log |f | is harmonic in Ω by Problem 11.5. If it has harmonic conjugate,
then there corresponds an F ∈ H(Ω) such that
Problem 16.4
Rudin Chapter 16 Exercise 4.
Proof. Denote I = [0, 1]. Let α : I → C \ {0} be an arbitrary path from 1 to f (0). By Definition
10.8, without loss of generality, we may assume further that α′ is continuous on I. Define
Z 1 ′
α (t) dt
g(0) = . (16.5)
0 α(t)
We note that Z 1
g(0) = d ln α(t) = log f (0)
0
and it means that f (0) = eg(0) . Let ζ ∈ X \ {0} and βζ : I → X be the line segment joining 0
and ζ. Next, we define γζ : I → C \ {0} by
α(2t), if t ∈ [0, 21 ];
γζ (t) = (16.6)
f βζ (2t − 1) , otherwise.
Finally, we define
Z 1 γζ′ (t) dt
g(ζ) = . (16.7)
0 γζ (t)
Clearly, we know that
Z 1
g(ζ) = d log γζ (t) = log γζ (1) − log γζ (0) = log f βζ (1) − log α(0) = log f (ζ)
0
so that f (ζ) = eg(ζ) . See Figure 16.1 for the paths βζ (I) and γζ .
Now it suffices to prove that the function g : X → C defined by the equations (16.5) and
(16.7) is continuous on X. To this end, let z, ω ∈ X and suppose that βω,z : I → X is the line
segment from ω to z. We also define γω,z : I → C \ {0} by
γω,z (t) = f βω,z (t) .
496 Chapter 16. Analytic Continuation
Thus both βz and βω ∗ βω,z are paths in X from 0 to z. Since X is simply connected, we follow
from Theorem 16.14 or [74, p. 323] that βz and βω ∗ βω,z are (path) homotopic in X, i.e.,
βz ≃p βω ∗ βω,z in X. Since f is continuous on X, we must have f (βz ) ≃p f βω ∗ βω,z in f (X).
Recall that α is arbitrary in the definition (16.6), it establishes that
γz ≃p γω ∗ γω,z
in f (X) ∪ α(I), see Figure 16.2 for the paths γz (I), γω (I) and γω,z (I)
Since X is compact and f (z) 6= 0 for all z ∈ X, there exists a constant m > 0 such that
κ
m = min |f (z)|. Given ǫ > 0. Choose κ > 0 such that m < ǫ. By the continuity of f , there
z∈X
corresponds a δ > 0 such that for z, ω ∈ X and |z − ω| < δ, the length of γω,z is less than κ. Let
16.2. Problems on the Modular Group and Removable Sets 497
Problem 16.5
Rudin Chapter 16 Exercise 5.
• Case (iii): |a| > 1. We may take |a| > |c|. Otherwise, we consider σ ϕ(z) = − cz+d az+b
az+b
instead of ϕ(z) = cz+d . Now it is easy to see that one can find an N ∈ Z satisfying
0 ≤ |a − N c| < |c| < |a|. Since
az + b (a − c)z + (b − d)
τ −1 ϕ(z) = −1= ,
cz + d cz + d
we establish that
−cz − d a1 z + b1
ϕ1 (z) = σ τ −N ϕ(z) = = .
(a − N c)z + (b − N d) c1 z + d1
Simple algebra shows that ϕ1 ∈ G, |a1 | < |a| and 0 ≤ |c1 | < |c|. If |c1 | = 0, then a1 d1 = 1
so that a1 = ±1 which goes back to Case (ii). Otherwise, we can repeat the above process
finitely many times, say m times, to get
am z + bm
ϕm (z) = ∈ G,
cm z + dm
Consequently, this proves the first assertion that τ and σ generate the modular group G.
For the second assertion, suppose that
We check Theorem 16.19(a) and (b). Based on Apostol’s description [6, p. 30], two points
ω, ω ′ ∈ Π+ = {z ∈ C | Re z > 0} are said to be equivalent under G if ω ′ = ϕ(ω) for some
ϕ ∈ G. With this terminology, property (a) means that no two distinct points of R are equivalent
under G and property (b) implies that for every ω ∈ Π+ , there exists a z ∈ R such that z is
equivalent to ω.
16.2. Problems on the Modular Group and Removable Sets 499
Lemma 16.1
Suppose that z1 , z2 ∈ R, z1 6= z2 and z2 = ϕ(z1 ) for some ϕ ∈ G. Then we have
Im z
Im ϕ(z) = (16.10)
|cz + d|2
• Case (ii): |c| = 1. Then the condition (16.11) becomes |z1 ± d|2 ≤ 1 or equiva-
lently,
(Re z1 ± d)2 + (Im z1 )2 ≤ 1. (16.12)
Further reduction implies that
3 1
(Re z1 ± d)2 ≤ 1 − (Im z1 )2 ≤ 1 − = ,
4 4
so that
1
|Re z1 ± d| ≤ . (16.13)
2
Since − 21 ≤ Re z1 ≤ 12 , we have |d| ≤ 1 which means either d = 0 or |d| = 1.
Combining the definition (16.9) of R and Lemma 16.1, we see immediately that no two
distinct points of R are equivalent under G which is property (a). For proving property (b), we
need the following result whose proof can be found in [6, Lemma 1, pp. 31, 32]:
Lemma 16.2
ω′
Given ω1′ , ω2′ ∈ C with ω2′ not real. Let Ω = {mω1′ + nω2′ | m, n ∈ Z}. Then there
1
exist ω1 , ω2 ∈ C such that ω2 = aω2′ + bω1′ and ω1 = cω2′ + dω1′ , where ad − bc = 1,
|ω2 | ≥ |ω1 | and |ω1 ± ω2 | ≥ |ω2 |.
Now we go back to the proof of our problem. If ω1′ = 1 and ω2′ = ω ∈ Π+ , then it is easy to
ω′
/ R. By Lemma 16.2, there exist ω1 and ω2 with |ω2 | ≥ |ω1 | and |ω1 ± ω2 | ≥ |ω2 |
see that ω2′ ∈
1
such that
ω2 = aω + b and ω1 = cω + d.
ω2
Let z = ω1 . These relations give
aω + b
z= = ϕ(ω) (16.14)
cω + d
with ad − bc = 1, |z| ≥ 1 and |z ± 1| ≥ |z|. The relation (16.14) means that there exists a point
z ∈ R equivalent to ω ∈ Π+ under G which is exactly property (b). Hence we obtain the result
that R is a fundamental domain of G and we end the proof of the problem.
Problem 16.6
Rudin Chapter 16 Exercise 6.
Proof. Since ψ ϕ(z) = z + 1, it follows from Problem 16.5 that G is also generated by ϕ and
ψ. It is easy to see that
ϕ2 (z) = ϕ ϕ(z) = z
and
1
ψ 2 (z) = − and ψ 3 (z) = z.
z−1
Hence ϕ has period 2 and ψ has period 3. This completes the proof of the problem.
Problem 16.7
Rudin Chapter 16 Exercise 7.
az+b
Proof. For each linear fractional transformation ϕ(z) = cz+d , we associate the 2 × 2 matrix
a b
Mϕ = .
c d
Here we identify each matrix with its negative because Mϕ and −Mϕ represent the same trans-
formation. If Mϕ and Mψ are the matrices associated with the linear fractional transformations
ϕ and ψ respectively, then it is easy to see that the matrix product Mϕ Mψ is associated with
the function composition ϕ ◦ ψ.
16.2. Problems on the Modular Group and Removable Sets 501
so that
nk 1 0 mk 1 (−2)mk
A = and B =
(−2)nk 1 0 1
and then
1 (−2)mk 1 2Nk
Ank Bmk = = ,
(−2)nk (−2)mk +nk + 1 2Mk 2Lk + 1
where Nk,j , Mk,j , Lk,j and Pk,j are integers. Hence we apply this to the expression (16.15)
to conclude immediately that if
a b
M= ,
c d
then a and d are odd, b and c are even.
• Proof of the first part of Problem 16.5. Note that the transformations z 7→ z + 1
and z 7→ − z1 correspond to the matrices
1 1 0 −1
T= and S =
0 1 1 0
where the nk are integers. To this end, we first notice that S2 = I,b so this explains why
only the S appears in the form (16.16). Next, it suffices to prove those matrices
a b
M=
c d
Assume that the form (16.16) is true for all matrices with lower left-hand element less
than c for some c ≥ 1. Since ad − bc = 1, c and d must be coprime so that d = cq + r for
some q ∈ Z and 0 < r < c. Since
−q a b 1 −q a −aq + b
MT = = ,
c d 0 1 c r
we have
−aq + b −a
MT−q S = . (16.17)
r −c
By the hypothesis, the matrix (16.17) has the form (16.16) which implies that M can be
expressed in the form (16.17).
Problem 16.8
Rudin Chapter 16 Exercise 8.
Since x, y ∈ R and E ⊂ R, the integrals in the equation (16.18) are real integrals. Clearly,
1 1 1 1 1 1
t−x ≥ A−x > 0 and t−y ≥ A−y > 0 for all t ∈ E. Let δx = A−x and δy = A−y . Then it
follows from the expression (16.18) that
Z
f (x) − f (y) ≥ (x − y) δx δy dt = (x − y)δx δy m(E) > 0,
E
(b) The answer is negative. Assume that f could be extended to an entire function. Since
1
|t| ≤ R on E, we have | t−z | → 0 as |z| → ∞ for every t ∈ E. In other words, we see
that f (z) → 0 as |z| → ∞ which means f is bounded in C. By Theorem 10.23 (Liouville’s
Theorem), f is constant which contradicts part (a). Hence we conclude that f cannot be
extended to an entire function.
z
(c) Given ǫ > 0 and z ∈ C \ D(0; R + Rǫ ). Define g(z, t) = − z−t on E. We claim that
g(z, t) → −1 uniformly on E. In fact, we see that
|t|
|g(z, t) + 1| =
|z − t|
16.2. Problems on the Modular Group and Removable Sets 503
|t| R
|g(z, t) + 1| = ≤ (16.19)
|z − t| |z| − R
R
on E. Since |z| > R + ǫ, the inequality (16.19) implies that
|g(z, t) + 1| < ǫ
≤ |g(z, t) + 1| dt
E
< ǫm(E) (16.20)
R
for every |z| ≥ R + ǫ. Since ǫ is arbitrary, we conclude from the estimate (16.20) that
(d) The compactness of E implies that Ω is open in C. We have to show that Ω is connected.
Since E ⊂ R, we have C \ R ⊆ Ω. Thus the upper half plane Π+ lies in a component of
Ω. Similarly, the lower half plane Π− must lie in a component of Ω. Since E 6= R, one
can have a real number a lying in Ω. Since Π+ is connected, it follows from [74, Theorem
23.4, p. 150] that Π+ ∪ {a} is also connected. Similarly, the set Π− ∪ {a} is also connected.
According to [74, Theorem 23.3, p. 150], the union Π+ ∪ {a} ∪ Π− = (C \ R) ∪ {a} is
connected. Finally, since
(C \ R) ∪ {a} ⊆ Ω ⊆ C,
the connectedness of Ω can be deduced again from [74, Theorem 23.4, p. 150].
Assume that f had a holomorphic square root in Ω. By the definition, Ω is a region.
Furthermore, we observe from Theorem 13.11 that Ω is simply connected. However, the
closed curve C(0; 2R) is not null-homotopic in Ω because E lies inside C(0; 2R). By the
definition, Ω is not simply connected and hence f has no holomorphic square root in Ω.
(e) Assume that Re f was bounded in Ω. We use part (f) in advance that f will be bounded
in Ω. This implies that f can be extended to a bounded entire function because E is
compact. Hence it contradicts part (a) and then Re f is unbounded in Ω.
∞
−1 t − a
= tan
b −∞
π π
= − −
2 2
= π.
(g) Suppose that γ is a positively oriented circle which has E in its interior. Since γ ∗ (the
range of γ) is closed in C and γ ∗ ∩ E = ∅, we have δ = inf∗ |t − z| > 0 so that
z∈γ
t∈E
Z Z dt
Z Z
1 ℓ(γ)m(E)
dz ≤ dt dz = < ∞,
γ E t−z γ E δ δ
where ℓ(γ) is the circumference of the circle γ. Hence Theorem 8.8 (The Fubini Theorem)
and Theorem 10.11 together assert that
Z Z Z Z Z Z
1 1
dt dz = dz dt = 2πiInd γ (t) dt = 2πi dt = 2πm(E)i.
γ E t−z E γ t−z E E
Define g(z) = eiz and ϕ = g ◦ f : Ω → C. Now part (a) ensures that ϕ is not constant.
Furthermore, it is clear that
ϕ(Ω) = g f (Ω) ⊆ {reiθ | e−π ≤ r ≤ eπ and θ ∈ [0, 2π]}
so that ϕ is bounded on Ω. Thus it remains to show that ϕ ∈ H(Ω) and this follows from
the result f ∈ H(Ω). To see this, we write
Z
f (z) = ψ(z, t) dt,
E
1
where ψ(z, t) = t−z . For each fixed z ∈ Ω, ψ(z, t) is measurable. For each fixed t ∈ E,
we have ψ(z, t) ∈ H(Ω). Furthermore, for each z0 ∈ Ω, we have inf |t − z0 | > 0. Let this
t∈E
number be 2δ. Then we have
δ= inf |t − z|
z∈D(z0 ;δ)
t∈E
Problem 16.9
Rudin Chapter 16 Exercise 9.
c
See the online paper http://www.nieuwarchief.nl/serie5/pdf/naw5-2001-02-1-032.pdf or [36].
16.2. Problems on the Modular Group and Removable Sets 505
Proof.
(c) According to Problem 16.8(c), the value of the limit is −2 because m(E) = 2.
Problem 16.10
Rudin Chapter 16 Exercise 10.
Proof.
Lemma 16.3
Suppose that E is compact and has no interior, and K satisfies the following two
conditions:
which implies that the value fK (z) is uniquely determined by the limit, i.e., all the
values fK (z) must be equal for all compact K ⊆ E satisfying z 6∈ K and the conditions.
Hence we may define fb : C \ E ′ → C by
for any such compact K. Recall that fK ∈ H(C \ K) and E ′ ⊆ K, so the expression
(16.22) ensures that fb ∈ H(C \ E ′ ), completing the proof of Lemma 16.3
Suppose that E is countable compact, i.e., E = {z1 , z2 , . . .}. Since {zn } has no interior, it
is nowhere dense and we observe from the Baire Category Theorem (see §5.7) that E has
no interior. Let E ′ be the set in Lemma 16.3. Assume that E ′ 6= ∅. Notice that
∞
[
E′ = E ′ ∩ {zn } .
n=1
Ind Γ (z) = 1
for every z ∈ E. Suppose that V is the union of the collection of those components of
C \ Γ intersecting E. Thus we have E ⊆ V . Define g : V → C by
Z
1 f (ζ)
F (z) = dζ.
2πi Γ ζ − z
for every z ∈ V . By an argument similar to part (c) below, we see that F ∈ H(V ).
Fix α ∈ V \ E. Since the set (C \ V ) ∪ {α} is closed in C and [(C \ V ) ∪ {α}] ∩ E = ∅,
d E, (C\V )∪{α} > 0. Let 0 < ǫ < d E, (C\V )∪{α} . Since E is compact and m(E) = 0,
E can be covered by a finite number of open intervals I1 = (a1 , b1 ), I2 = (a2 , b2 ), . . .,
In = (an , bn ) whose total length is less than ǫ. Without loss of generality, we may assume
that I1 , I2 , . . . , In are pairwise disjoint and intersect E. Let γk denote the counterclockwise
circle having Ik as its diameter and
n
[
Γǫ = γk .
k=1
16.2. Problems on the Modular Group and Removable Sets 507
We notice that the length of Γǫ is less than πǫ. By applying Theorem 10.35 (Cauchy’s
Theorem) to the cycle Γ − Γǫ in C \ E, we obtain
Z Z
1 f (ζ) 1 f (ζ)
f (α) = dζ = F (α) − dζ. (16.23)
2πi Γ−Γǫ ζ − α 2πi Γǫ ζ − α
Define f1 , f2 : Ω \ E → C by
Z Z
f (ζ) f (ζ)
f1 (z) = dζ and f2 (z) = dζ.
Γ1 ζ−z Γ2 ζ −z
We claim that f1 ∈ H(Ω) and f2 ∈ H(C \ E). To this end, we first note from Theorem
10.35 (Cauchy’s Theorem) that f1 is independent of Γ1 . Next, we take z ∈ Ω and fix the
cycle Γ1 as constructed above. Denotethe length of Γ1 to be ℓ(Γ1 ). Since E ∪ {z} lies
entirely inside Γ1 , we have Γ∗1 ∩ E ∪ {z} = ∅. Recall that E is compact, so is E ∪ {z} and
then d Γ∗1 , E ∪{z} > 0. Let this number be 2δ. If h is very small such that z +h ∈ D(z; δ),
then we have Z
f1 (z + h) − f1 (z) f (ζ)
= dζ. (16.24)
h Γ1 (ζ − z)(ζ − z − h)
Clearly, for every ζ ∈ Γ1 , we have
f (ζ) M
≤ 2.
(ζ − z)(ζ − z − h) 2δ
Using this and the fact that ℓ(Γ1 ) < ∞, we may apply Theorem 1.34 (Lebesgue’s Domi-
nated Convergence Theorem) to the expression (16.24) to conclude that
Z
f (ζ)
f1′ (z) = 2
dζ.
Γ1 (ζ − z)
Since z ∈ Ω is arbitrary, we get the desired result that f1 ∈ H(Ω). Using a similar
argument, we can show that f2 ∈ H(C \ E) which proves the desired claim.
Therefore, we have f = f1 − f2 on Ω \ E. Now the boundedness of f certainly implies
the boundedness of f2 . Since E is removable, f2 is a constant. Consequently, we obtain
f ∈ H(Ω).
508 Chapter 16. Analytic Continuation
(d) Suppose that E ⊂ C is compact and m2 (E) = 0.d Then E is removable. Here we need
the following lemma to prove this result.
Lemma 16.4
E ⊆ C is removable if and only if every bounded holomorphic function f on C \ E
satisfies f ′ (∞) = 0.
Proof of Lemma 16.4. By the definition, we know that E is removable if and only if
every bounded holomorphic function f on C\E is constant. Obviously, if f ∈ H(C\E)
is constant, then f ′ (∞) = 0. Conversely, let g : C \ E → C be nonconstant and
bounded. Then there exists an z0 ∈ C \ E such that g(z0 ) 6= g(∞). Define
g(z) − g(z0 )
f (z) =
z − z0
on C \ E. Obviously, f is also a bounded and nonconstant function on C \ E and
Consequently, we establish
= g(∞) − g(z0 ) 6= 0,
We return to the proof of the problem. Let f be a bounded holomorphic function on C \E,
i.e., |f (z)| ≤ M on C \ E for some positive constant M . Given ǫ > 0. Then E can be
covered by open discs D1 , D2 , . . . , Dn of radii r1 , r2 , . . . , rn respectively such that
n
X
rk < ǫ.
k=1
Let Γ = ∂D1 ∪ ∂D2 ∪ · · · ∪ ∂Dn . Using [118, Eqn. (1.2), p. 16], we have
1 Z n
X
|f ′ (∞)| = f (z) dz ≤ M rk < M ǫ. (16.25)
2πi Γ
k=1
Since ǫ is arbitrary, the inequality (16.25) guarantees that f ′ (∞) = 0. Now we conclude
from Lemma 16.4 that E is in fact removable.
Remark 16.1
Recall that we have studied the special case of removable sets in Problem 11.11. See also
Remark 11.2.
Problem 16.11
Rudin Chapter 16 Exercise 11.
Proof. By the definition, we have Ωα ⊂ Ωβ if α < β. In Figure 16.4, Ωβ is the union of Ωα and
the region shaded by straight lines.
e
Recall from point-set topology [78, p. 3] that a nonempty compact connected metric space is a continuum.
510 Chapter 16. Analytic Continuation
Let γ be a curve in C with parameter interval [0, 1] that starts at the center of D(ζ; δζ ).
Note that this may happen that γ([0, 1]) * Ωα . However, the compactness of γ([0, 1])
ensures that there corresponds
an β > α such that γ([0, 1]) ⊆ Ωβ . Hence the first assertion
guarantees that fα , D(ζ; δ) can be analytically continued along the curve γ in C. By
Theorem 16.15 (The Monodromy Theorem), there exists an entire function f such that
f (z) = fα (z)
h Z ∞ i
1
≤ 2 exp(−et ) dt + 2ee . (16.28)
2π(r sin θ − 1) 1
1
|f (reiθ )| ≤ (ee + e−1 ).
π(r sin θ − 1)
lim f (reiθ ) = 0.
r→∞
• f is not constant. Fix r > 0. Let 0 < r < α < R and Γ = Γα ∪ LR , where LR = R + πit R
for t ∈ [−R, R]. Assume that f was constant. Since f is entire, the third assertion forces
that f (z) = 0 in C. In particular, we have
Z
1 exp(eω )
0 = f (r) = dω (16.29)
2πi Γα ω−r
for every α > r. It is clear that Γ is closed and Ind Γ (r) = 0. Using Theorem 10.35
(Cauchy’s Theorem), we know that
Z Z Z
1 exp(eω ) 1 exp(eω ) 1 exp(eω )
dω + dω = dω = 0
2πi Γα ω−r 2πi LR ω−r 2πi Γ ω−r
which implies
Z
exp(eω )
dω = 0 (16.30)
LR ω−r
for every R > r. Write
Z
1 exp(eω )
f (r) = dω
2πi ΓR ω − r
Z Z Z
1 exp(eω ) 1 exp(eω ) 1 exp(eω )
= dω + dω + dω
2πi γR+ ω − r 2πi γR− ω − r 2πi LR ω − r
Z ∞ Z ∞ Z
1 exp(−et ) 1 exp(−et ) 1 exp(eω )
= dt − dt + dω
2πi R t + πi − r 2πi R t − πi − r 2πi LR ω − r
Z ∞ Z
exp(−et ) 1 exp(eω )
=− 2 2
dt + dω.
R (t − r) + π 2πi LR ω − r
• g(reiθ ) → 0 as r → ∞ for every eiθ . If eiθ 6= 1, then the third assertion implies that
f (r)
lim g(r) = lim = 0.
r→∞ r→∞ exp[f (r)]
• Existence of an entire function h with the required properties. By the fourth and
the fifth assertions, we know that g is a nonconstant entire function such that g(reiθ ) → 0
as r → ∞ for every eiθ . If g has a zero of order N at z = 0, then we write g(z) = z N G(z).
Thus G is nonconstant entire, G(0) 6= 0 and
G(z)
for every eiθ . Define h(z) = G(0) . Therefore, h is nonconstant entire and h(0) = 1.
Furthermore, if z 6= 0, then we write z = reiθ for some r > 0. Since
G(nreiθ )
h(nz) = h(nreiθ ) = ,
G(0)
we follow from the limit (16.32) that h(nz) → 0 as n → ∞. If g(0) 6= 0, then we consider
the nonconstant entire function h(z) = g(z)
g(0) which satisfies h(0) = 1 and h(nz) → 0 as
n → ∞. In conclusion, there exists an entire function h such that
1, if z = 0;
lim h(nz) =
n→∞
0, if z 6= 0.
Problem 16.12
Rudin Chapter 16 Exercise 12.
2
Evidently, if |z − z 2 | < 2, then | z−z
2 | < 1 so that the series (16.33) converges by [99, Theorem
3.26, p. 61]. Furthermore, if |z − z 2 | > 2, then the series (16.33) diverges. The red shaded part
in Figure 16.5 indicates the regionf of convergence of the power series (16.33).
k
Next, suppose that Pk (z) = [z(1 − z)]3 , so
X∞
1
f (z) = P (z).
3k k
k=1
2
f
This is (x − x2 + y)2 + (y − 2xy)2 < 4.
16.3. Miscellaneous Problems 513
Note that the highest power and the lowest power of z in Pk (z) and in Pk+1 (z) are 2 · 3k and
3k+1 respectively. Since 3k+1 − 2 · 3k = 3k > 0 for every k ≥ 1, the polynomial Pk (z) contains
no power of z that appear in any other Pj (z) for all j 6= k. If we replace every Pk (z) by its
expansion in powers of z, then we get the power series
∞
X
f (z) = an z n (16.34)
n=1
with the property that a1 = a2 = 0 and for each positive integer k, we have
3k
(−1)n+1 Cn−3k
k
, if n = 3k , 3k + 1, . . . , 2 · 3k ;
23
an = (16.35)
0, if 2 · 3k < n < 3k+1 .
If both n and r tend to infinity, then it follows from Stirling’s formula [99, Eq. (103), p. 194]
that r
n nn
Crn ∼ · r .
2πr(n − r) r (n − r)n−r
3k 3k+1 +1
Recall that Cn−3k takes its maximum value when nk = 2 , so it is true that
3k 3k
=
2π(nk − 3k )[3k − (nk − 3k )] 2π(nk − 3k )(2 · 3k − nk )
3k
= 3k +1 3k −1
2π · 2 · 2
2 3k
= ·
π 32k − 1
2 1
∼ · k (16.36)
π 3
514 Chapter 16. Analytic Continuation
and
k
(3k )3 k
3 k +1 3 k −1 ∼ 23 . (16.37)
k k
( 3 2+1 ) 2 ( 3 2−1 ) 2
1 1
Since g(x) = x x is decreasing for x > e and x x → 1 as x → ∞ and nk ∼ 1.5 × 3k for large k, it
yields from the estimates (16.36) and (16.37) that
C 33k +1 ! 1.5×3
k 1
k
1
2
lim |ank | nk
= lim
k→∞ k→∞ 23k
1 k 1
= lim 2 × C 33k +1 1.5×3k
k→∞ 2 3 2
1 2 1 1 k+1
1
2
3k+1 3
= lim 2 · · k
· 23
k→∞ 2 3 π 3
= 1.
In other words, the radius of convergence of the power series is 1. Check the blue shaded part
in Figure 16.5.
Let λ = 3. Let pk = 2 · 3k and qk = 3k+1 for k = 1, 2, . . .. Then they satisfy λqk > (λ + 1)pk
and an = 0 for pk < n < qk for all positive integers k. Now the power series (16.33) and (16.34)
assert that there exists a δ > 0 such that
∞
z − z 2 3k
X ∞
X
= an z n
2
k=1 n=1
for all z ∈ D(0; 1) ∩ D(1; δ). By Definition 16.1, it means that 1 is a regular point of f . Finally,
it concludes from Theorem 16.5 that the sequence {spk (z)} converges in a neighborhood of 1,
where sp (z) is the pth partial sum of the power series (16.34).
By Figure 16.5 again, we know that all boundary points of T , except z = −1, are regular
points of f . Observe from the representation (16.34) and the definition (16.35) that
∞
X
−f (−z) = bn z n ,
n=1
where bn ≥ 0 for every n ≥ 1. Thus Problem 16.1 ensures that −f (−z) has a singularity at
−z = 1. Hence z = −1 is the singular point of f which is nearest to the origin, so we have
completed the proof of the problem.
Problem 16.13
Rudin Chapter 16 Exercise 13.
(a) If f ∈ X1 , then f = g′ for some g ∈ H(Ω). Since γ is a closed path lying in Ω, Theorem
10.12 implies that Z Z
f (z) dz = g′ (z) dz = 0. (16.38)
γ γ
16.3. Miscellaneous Problems 515
Conversely, suppose that the integral (16.38) holds. Since f ∈ H(Ω) and Ω is an annulus,
Problem 10.25 shows that f admits the Laurent series
−1
X
f (z) = f1 (z) + f2 (z) = cn z n + f2 (z),
−∞
where f1 ∈ H C \ D(0; 21 ) and f2 ∈ H D(0; 2) . Since
Z
1 f (z)
cn = dz,
2πi γ z n+1
1 ∞
X
F1 (ω) = f1 = c−n ω n
ω
n=2
√ 1
is holomorphic in {ω ∈ C | |ω| < 21 }. Therefore, it is true that lim sup n c
−n ≥ which
n→∞ 2
implies that the radius of convergence of the series
X∞
c−n n−1
G(ω) = ω
n=2
1−n
is at least 21 too. Next, it is clear from Theorem 10.6 that G′ (ω) = −ω −2 F1 (ω) in the
disc{ω ∈ C | |ω| < 12 }. By transforming back to the variable z, we see that
d X c−n 1−n
∞
f1 (z) = z (16.39)
dz n=2 1 − n
d X cn n+1
∞
f2 (z) = z (16.40)
dz n=0 n + 1
X∞ ∞
c−n 1−n X cn n+1
g(z) = z + z
n=2
1−n n=0
n+1
for all z ∈ Ω, then the facts (16.39) and (16.40) combine to imply immediately that
f (z) = g ′ (z) in Ω, i.e., f ∈ X1 .
(b) Since f ∈ H(Ω) and Ω is an annulus, Problem 10.25 shows that f admits a representation
∞
X
f (z) = an z n . (16.41)
n=−∞
516 Chapter 16. Analytic Continuation
(m)
If f ∈ Xm , then f = gm for some gm ∈ H(Ω). Again, gm has the Laurent series in Ω,
i.e.,
∞
X
gm (z) = bn,m z n
n=−∞
which gives
∞
X
(m)
an z n = gm (z)
n=−∞
∞
X
= n(n − 1) · · · (n − m + 1)bn,m z n−m
n=−∞
X∞
= (n + m)(n + m − 1) · · · (n + 1)bn+m,m z n .
n=−∞
Therefore, it means that a−1 = a−2 = · · · = a−m = 0. As m runs through all positive
integers, the Laurent series (16.41) reduces to
∞
X
f (z) = an z n
n=0
which implies that f ∈ H D(0; 2) .
Conversely, suppose that there exists an g ∈ H D(0; 2) such that f (z) = g(z) for all
z ∈ Ω. By Theorem 13.11, the simply connectedness of D(0; 2) ensures that one can find
an g1 ∈ H D(0; 2) such that g1′ = g. In fact, this argument can be repeated to achieve
(n)
the existence of an gn ∈ H D(0; 2) with gn = g for each positive integer n. Hence we
obtain
f (z) = gn(n) (z)
for all z ∈ Ω and this means that f ∈ Xn for every positive integer n.
Problem 16.14
Rudin Chapter 16 Exercise 14.
Proof. Our proof here basically follows that in [104, §2.7, pp. 54 – 56]. Since normality is a
local property, we may assume that Ω is the unit disc U . Suppose that
Recall from Theorem 16.20 that the modular function λ is invariant under Γ (i.e., λ ◦ ϕ = λ for
all ϕ ∈ Γ) and maps Π+ onto C \ {0, 1}. Since f (U ) ⊆ C \ {0, 1}, the function λ−1 ◦ f has a local
branch defined in a sufficiently small neighbourhood of f (0). Then this function element may
be analytically continued in U , so we assert from Theorem 16.15 (The Monodromy Theorem)
that there exists a holomorphic function fb : U → Π+ such that
λ ◦ fb = f. (16.42)
Let {fn } ⊆ F . Since |fn (p)| ≤ R for all n ∈ N, the Bolzano-Weierstrass Theorem [127, Problem
5.25, pp. 68, 69] ensures that there is a convergent subsequence {fnk (p)}. Let this limit be ℓ.
16.3. Miscellaneous Problems 517
• Case (i): ℓ 6= 0 and ℓ 6= 1. Then we can fix a branch of λ−1 in a neighborhood of ℓ and
use this to define the functions fc c
nk by the equation (16.42). Since Im fnk > 0, we have
Re (−ifcnk ) = Im c
f nk > 0 so that the family {−i fcnk | k ∈ N} is normal by Problem 14.15.
Consequently, the family
Fc = {fcnk | k ∈ N}
is also normal. For simplicity, we may assume that fcnk converges normally to g ∈ H(U ).
Clearly, we have g(U ) ⊆ Π+ . By the equation (16.42) again, we conclude that
g(p) = lim fbnk (p) = lim λ−1 fnk (p) = λ−1 (ℓ).
k→∞ k→∞
• Case (ii): ℓ = 1. Since fnk ∈ H(U ) and 0 ∈ / fnk (U ), we have fn1 ∈ H(U ). Since U is
k
simply connected, we deduce from Theorem 13.11 that each fnk has a holomorphic square
root hk in U . We choose the branch such that
H = {hk | k ∈ N}.
Now the limit (16.43) guarantees that we can apply Case (i) to H to obtain a convergent
subsequence hkj in U . In conclusion, the limit
• Case (iii): ℓ = 0. In this case, we may apply Case (ii) to the sequence {1 − fnk | k ∈ N}.
Remark 16.2
Problem 16.14 is classically called the Fundamental Normality Test.
Problem 16.15
Rudin Chapter 16 Exercise 15.
Proof. Without loss of generality, we may assume that D is the unit disc and D ⊆ Ω = D(0; R)
for some R > 1. Let (f, D) be analytically continued along every curve in Ω that starts at the
origin 0. Since f ∈ H(D), f has a power series expansion at 0, i.e.,
∞
X
f (z) = an z n .
n=0
518 Chapter 16. Analytic Continuation
Suppose that r is the radius of convergence of this power series. Clearly, we have 1 ≤ r. If r = 1,
then we know from Theorem 16.2 that f has at least one singular point on the unit circle T . Let
ω be a singular point of f on T and γ be a curve in Ω starting at 0 and passing through ω. Now
the assumption ensures that (f, D) can be analytically continued along γ in Ω, so ω is a regular
point of f , a contradiction. As a result, 1 < r ≤ R. Next, if r < R, then similar argument can
be applied to conclude that no point on C(0; r) is a singular point, but this contradicts Theorem
16.2. Hence we must have r ≥ R and it means that Theorem 16.5 holds in this special case.
This proves the first assertion.
Let Ω be any simply connected region (other than the complex plane itself), D ⊆ Ω and
(f, D) be analytically continued along every curve in Ω that starts at the center of D. Let
z0 ∈ D. By Theorem 14.8 (The Riemann Mapping Theorem) or [11, §14.2, pp. 200 – 204], one
can find a (unique) conformal mapping F : Ω → U such that F (z0 ) = 0 and F ′ (z0 ) > 0. Let
S = F (D). Obviously, we have G = F −1 |S : S ⊂ U → D so that G(0) = z0 . We consider the
mapping
g = f ◦ G : S ⊂ U → C. (16.44)
Since S is an open set containing the origin, we may assume that it is an open disc centered at
0 so that S and U are concentric. Since f ∈ H(D), we conclude that g ∈ H(S). Furthermore,
since (f, D) can be analytically continued along every curve in Ω, the function element (g, S) can
also be analytically continued along every curve in U . Therefore, the first assertion guarantees
that there corresponds a h ∈ H(U ) such that h(z) = g(z) for all z ∈ S. Using the definition
(16.44), we have h ◦ F ∈ H(Ω) and for all z ∈ D,
h F (z) = h G−1 (z) = h g −1 f (z) = f (z).
This proves the second assertion and we end the analysis of the problem.
CHAPTER 17
H p-Spaces
Problem 17.1
Rudin Chapter 17 Exercise 1.
• By Definition 17.1, it suffices to prove that Theorem 17.5 is true for subharmonic function
in U . Let 0 ≤ r < 1. Then K = D(0; r) ⊂ U and u : K → R is subharmonic. In particular,
u is an upper semicontinuous function. We need the following result:a
a
Read https://encyclopediaofmath.org/wiki/Baire_theorem#Baire.27s_theorem_on_semi-continuous_functions .
519
520 Chapter 17. H p -Spaces
Let 0 ≤ r1 < r2 < 1. Using this lemma, we know that u(z) ≤ un (z) on C(0; r2 ). By
Theorem 11.8, Hun ∈ C(K), Hun is harmonic in D(0; r2 ) and (Hun )|C(0;r2 ) = un . Clearly,
we have u(z) ≤ un (z) = (Hun )(z) on C(0; r2 ), so the first assertion implies that
Problem 17.2
Rudin Chapter 17 Exercise 2.
Proof. Let u(z) = log 1 + |f (z)| = log 1 + elog |f (z)| . Define ϕ : R → R by ϕ(x) = log(1 + ex ).
Then we can write
u(z) = ϕ log |f (z)| . (17.4)
Since f ∈ H(Ω), it follows from Theorem 17.3 that log |f (z)| is subharmonic in Ω. Evidently,
ex
ϕ′ (x) = 1+e ′ ′
x > 0 for all x ∈ R and ϕ (s) < ϕ (t) if s < t, so ϕ is a monotonically increasing
convex function on R. By applying Theorem 17.2 to the function (17.4), we conclude that u is
subharmonic in Ω, as required. This completes the analysis of the problem.
Problem 17.3
Rudin Chapter 17 Exercise 3.
Proof. It seems that the hypothesis 0 < p ≤ ∞ should be replaced by 0 < p < ∞ because if
p = ∞, then the function f (z) ≡ 2 belongs to H ∞ . In this case, we have |f (z)|∞ = ∞ for any
z ∈ U.
≤ u(z)
for every z ∈ U .
1
• kf kp = uf (0) p . Let 0 < p < ∞. By the previous assertion, we know that
Z
uf (z) = P (z, eit )|f (eit )|p dσ.
T
1
When z = 0, we have P (0, eit ) = 1 so that uf (0) = kf kpp , i.e., kf kp = uf (0) p .
c
Recall that P [f ] is the Poisson integral of f .
d
We can interchange the limit and the integral because Theorem 10.24 (The Maximum Modulus Theorem)
asserts that Fρ (t) = Pr (θ − t)|f (ρeit )|p is increasing with respect to ρ, so we may apply Theorem 1.26 (The
Lebesgue’s Monotone Convergence Theorem).
522 Chapter 17. H p -Spaces
Problem 17.4
Rudin Chapter 17 Exercise 4.
for all z = reit ∈ U and all 0 ≤ r < 1. By Definition 17.7, we get kf k0 < ∞ so that f ∈ N .
On the other hand, suppose that f ∈ N . If f ≡ 0 so that log+ |f (z)| = 0 in U , then
there is nothing to prove. With the aid of the result in §17.19, there correspond two functions
b1 , b2 ∈ H ∞ such that b2 has no zero in U and
b1
f= .
b2
Without loss of generality, we may assume that kb1 k∞ ≤ 1 and kb2 k∞ ≤ 1. Since U is simply
connected and b12 ∈ H(U ), we deduce from Theorem 13.11 that there exists an g ∈ H(U ) such
that b2 = eg . Since kb2 k∞ = eRe g ≤ 1, the function u(z) = Re g(z) is less than or equal to zero
in U and this implies
log |f (z)| = log |b1 (z)| − log |b2 (z)| ≤ − log eu(z) = −u(z) (17.6)
for all z ∈ U . Since −u(z) ≥ 0, the inequality (17.6) and the definition in §15.22 yield
Problem 17.5
Rudin Chapter 17 Exercise 5.
Proof. Since f ∈ H(U ), it is true that f ◦ ϕ ∈ H(U ). Since f ∈ H p , Problem 17.3 ensures that
there is a harmonic function u in U such that |f (z)|p ≤ u(z) for all u ∈ U . Thus this shows that
f ϕ(z) p ≤ u ϕ(z)
for all z ∈ U . Applying Problem 11.7(b) with Φ = u and f = ϕ there, we see that
The assertion is also true when we replace H p by N . To see this, Problem 17.4 ensures that
there exists a harmonic function u in U such that log+ |f (z)| ≤ u(z) holds for all z ∈ U . Then
we have
log+ f ϕ(z) ≤ u ϕ(z) (17.7)
in U . Applying Problem 11.7 to u ◦ ϕ and then using the fact that ∆u = 0 in U , we see
immediately that
∆[u ◦ ϕ] = [(∆u) ◦ ϕ] · |ϕ′ |2 = 0.
In other words, u ◦ ϕ is harmonic in U . Combining the inequality (17.7) and Problem 17.4, we
conclude that f ◦ ϕ ∈ N , completing the proof of the problem.
Problem 17.6
Rudin Chapter 17 Exercise 6.
1 |t| 1
|1 − reit | ≥ 1−r+ ≥ (1 − r + |t|). (17.9)
2 π 2π
On the other hand, the triangle inequality gives
1 1 (2π)α
α ≤ it α
≤ α
|t| + 1 − r |1 − re | |t| + 1 − r
Here we note that 1 − z 6= 0 in U , so we may take the branch such that (1 − z)α = exp α log(1 − z) , where
e
− π2 π
< arg(1 − z) < 2 .
524 Chapter 17. H p -Spaces
Z π Z π
dt dt
α ≤ Iα (r) ≤ (2π)α α . (17.11)
−π |t| + 1 − r −π |t| + 1 − r
1−α
If α < 1, then the integral in the inequalities (17.11) tends to 2π1−α as r → 1 so that Iα (r)
is bounded as r → 1. If α = 1, then we observe from the inequalities (17.11) that
Iα (r)
0 < m1 ≤ ≤ M1
log(1 − r)
as r → 1 for some positive constants m1 and M1 . Similarly, if α > 1, then the inequalities
(17.11) tells us that
Iα (r)
0 < m2 ≤ ≤ M2
(1 − r)1−α
as r → 1 for some positive constants m2 and M2 .
We notice that n 1 Z o 1 I (r) 1
p αp p
k(fα )r kp = |fα (reit )|p dt = ,
2π T 2π
so the previous paragraph indicates that fα ∈ H p if and only if αp < 1. Thus, for 0 < r < s < ∞,
if we take α ∈ ( 1s , 1r ), then it is easy to see that fα ∈ H r but fα ∈ / H s , i.e., H s ⊂ H r . The
case for s = ∞ is obvious because we have fα ∈ / H ∞ for every α > 0. In particular, if we take
1
α = 2r , then we have f 1 ∈ H r but f 1 ∈ / H ∞ . This completes the analysis of the problem.
2r 2r
Problem 17.7
Rudin Chapter 17 Exercise 7.
Proof. Now Problem 17.6 guarantees that H ∞ ⊂ H p for every 0 < p < ∞. By Definition 17.7,
H p ⊆ N holds for every 0 < p < ∞, so we have H ∞ ⊂ N . Hence we conclude easily that
\ \
H∞ ⊂ N ∩ Hp ⊆ H p.
0<p<∞ 0<p<∞
Problem 17.8
Rudin Chapter 17 Exercise 8.
Proof. If p = 1, then there is nothing to prove. If 0 < p < 1, then Problem 17.6 implies that
H 1 ⊂ H p so that the result is trivial. Thus we may assume that 1 < p < ∞. Since f ∈ H 1 ,
Theorem 17.11 guarantees that f is the Poisson integral of f ∗ , i.e.,
Z
f (z) = P (z, eit )f ∗ (eit ) dσ
T
17.2. Basic Properties of H p 525
which shows Z
fr (eiθ ) = f (reiθ ) = Pr (θ − t)f ∗ (eit ) dσ,
T
where 0 ≤ r < 1. Since Pr (t) > 0 for all t ∈ T , we may apply Theorem 3.5 (Hölder’s Inequality)
to obtain
Z p
|fr (eiθ )|p = Pr (θ − t)f ∗ (eit ) dσ
T
Z p
≤ Pr (θ − t)|f ∗ (eit )| dσ
T
Z h p−1 i h 1 i p
= Pr p (θ − t) × Prp (θ − t)|f ∗ (eit )| dσ
T
nZ o n Z h p−1 i p op−1
p−1
≤ Pr (θ − t)|f ∗ (eit )|p dσ × Pr p (θ − t) dσ
T T
nZ o nZ op−1
∗ it p
= Pr (θ − t)|f (e )| dσ × Pr (θ − t) dσ
Z T T
∗ it p
= Pr (θ − t)|f (e )| dσ
T
which implies
Z
1
kfr kpp = |fr |p dt
2π T
Z Z
1
≤ Pr (θ − t)|f ∗ (eit )|p dσ dθ
2π T T
Z h Z i
1
≤ Pr (θ − t) dθ |f ∗ (eit )|p dσ
2π T
ZT
= |f ∗ (eit )|p dσ
T
= kf ∗ kp ,
Problem 17.9
Rudin Chapter 17 Exercise 9.
Proof. Since f (U ) is not dense in the complex plane, there exists a point α ∈ C and r > 0 such
1
that D(α; r) ⊆ C \ f (U ), i.e., |f (z) − α| > r for all z ∈ U . Let F (z) = f (z)−α . Then F is
∞
a bounded holomorphic function in U so that F ∈ H . Thus it follows from Theorem 11.32
(Fatou’s Theorem) that
F ∗ (eiθ ) = lim F (reiθ )
r→1
exists at almost all points of T . This implies that the same conclusion also holds for the function
f , completing the proof of the problem.
Problem 17.10
Rudin Chapter 17 Exercise 10.
526 Chapter 17. H p -Spaces
Proof. Define Φα : H 2 → C by
Φα (f ) = f (α).
It is obvious that Φα is a linear functional on H 2 . Let
∞
X
f (z) = an z n .
n=0
∞
X
then Theorem 17.12 implies |bn |2 < ∞. By the Parseval Theorem [100, Eqn. (6), p. 91], the
n=0
formula (17.12) becomes
∞
X ∞
X
an αn = f (α) = an bn .
n=0 n=0
Problem 17.11
Rudin Chapter 17 Exercise 11.
The identity
∞
X 1+x
n2 xn−1 = (17.14)
(1 − x)3
n=1
is valid for |x| < 1, so we may apply this to the inequality (17.13) to obtain
s
′ 1 + |α|2
|f (α)| ≤ . (17.15)
(1 − |α|2 )3
We claim that the estimation (17.15) is sharp. To see this, let α 6= 0 and Cα be the number
in the estimation (17.15). Consider the function
∞
eiθ X
Fα (z) = n|α|n−1 z n , (17.16)
Cα
n=1
where z ∈ U . Direct computation indicates that the radius of convergence of the series (17.16)
1
is |α| > 1 so that Fα ∈ H(U ). Next, using the identity (17.14), we see that
∞
X
1 2 2 2(n−1)
kFα k22 = · n |α|
Cα
n=0
∞
(1 − |α|2 )3 X 2 2(n−1)
= · n |α|
1 + |α|2 n=0
(1 − |α|2 )3 1 + |α|2
= 2
×
1 + |α| (1 − |α|2 )3
= 1.
F0 (z) = eiθ z
528 Chapter 17. H p -Spaces
which satisfies kF0 k22 = 1 and |F0′ (0)| = 1. Hence this proves the claim and the functions (17.16)
are extremal.
Next, we consider the general n. Firstly, we have
∞
X
f (n) (α) = m(m − 1) · · · (m − n + 1)am αm−n
m=n
so that
X∞ 2
|f (n) (α)|2 = m(m − 1) · · · (m − n + 1)am αm−n
m=n
X
∞ 2 nX
∞ o2
≤ |am |2 × [m(m − 1) · · · (m − n + 1)]2 |α|2(m−n)
m=n m=n
nX
∞ o2
≤ |m(m − 1) · · · (m − n + 1)αm−n |2
m=n
which implies
∞
X
|f (n) (α)| ≤ |m(m − 1) · · · (m − n + 1)αm−n |2 .
m=n
We denote v
u ∞
uX
Cα,n =t |m(m − 1) · · · (m − n + 1)αm−n |2 > 0.
m=n
(n)
are the extremal functions in this case because |F0 (α)| = (n!)2 . Therefore, we have completed
the proof of the problem.
Problem 17.12
Rudin Chapter 17 Exercise 12.
17.2. Basic Properties of H p 529
for all z ∈ U . Now the hypothesis guarantees that f is real a.e. in U . By the Open Mapping
Theorem, f must be constant.
1+z
Consider f (z) = i 1−z in U . It is easy to see that f ∈ H p for every 0 < p < 1 and
t
f ∗ (eit ) = lim f (reit ) = − cot
r→1 2
is real a.e. on T , but f is not constant. Hence we have completed the proof of the problem.
Problem 17.13
Rudin Chapter 17 Exercise 13.
Proof. Since |f (reit )| = |γr (t)| ≤ M for every 0 ≤ r < 1 and t ∈ [−π, π], f is bounded in U .
Thus f ∈ H ∞ so that f ∈ H 1 . Furthermore, we also have the fact that f ∗ is bounded on T .
Let f ∗ (eit ) = µ(t). Since f (reπi ) = f (re−πi ), it is easy to see that
then we have Z Z π
1 f (ζ) 1
an = dζ = r −n e−int f (reit ) dt (17.18)
2πi C(0;r) ζ n+1 2π −π
for every n ≥ 0 and 0 < r < 1. On the other hand, it follows from Theorem 17.11 that
f ∗ ∈ L1 (T ), so the Fourier coefficients of f ∗ are given by
Z π
1
fc∗ (n) = e−int f ∗ (eit ) dt (17.19)
2π −π
for every n ∈ Z. By observing the coefficients (17.18) and (17.19), we know that
Z Z
|r n an − fc∗ (n)| = e−int [f (reit ) − f ∗ (eit )] dσ ≤ |f (reit ) − f (eit )| dσ = kfr − f ∗ k1 → 0
T T
for every n = 1, 2, . . ..
530 Chapter 17. H p -Spaces
holds for every n = 1, 2, . . .. Next, Theorem 17.13 (The F. and M. Riesz Theorem) shows that
µ is absolutely continuous with m, i.e., µ(E) = 0 if m(E) = 0. Recall that f ∗ (eit ) = µ(t), so
f ∗ ∈ C(T ). In fact, f ∗ is AC on T because of Theorem 7.18 because f maps sets of measure 0
to sets of measure 0. Finally, according to Theorem 17.11, we have
Z π Z π
1 1
f (z) = P (z, eit )µ(t) dt = P (z, eit )f ∗ (eit ) dt = P [f ∗ ](z)
2π −π 2π −π
for all z ∈ U . Thus if we define the function Hf ∗ : U → C as [100, Eqn. (1), p. 234], then
Theorem 11.8 shows that Hf ∗ ∈ C(U ) and the restriction of Hf ∗ to T is exactly f ∗ . This is a
required extension of f and we end the analysis of the problem.
Problem 17.14
Rudin Chapter 17 Exercise 14.
Proof. Recall from Problem 2.11 that the support of a measure µ is the smallest closed set
K ⊆ T such that µ(T \ K) = 0. Without loss of generality, we may assume that µ 6≡ 0. Given
that K is a proper closed subset of T . Note that T \ K is nonempty open in T , so σ(T \ K) > 0.
Our goal is to show that
µ(T \ K) > 0. (17.21)
Now we observe from the proof of Theorem 17.13 (The F. and M. Riesz Theorem) that
dµ = f ∗ (eit ) dσ, i.e., Z
1
µ(E) = f ∗ (eit ) dt
2π E
holds for every measurable subset E of T , where f = P [f ∗ ] ∈ H 1 and f ∗ ∈ L1 (T ). Thusit
suffices to show that f ∗ doesn’t vanish on T \ K. Let g(t) = f ∗ (eit ). Assume that g(T \ K) = 0.
Then it means that log |g| = ∞ on T \ K, but it contradicts Theorem 17.17 (The Canonical
Factorization Theorem) that
log |g| = log |f ∗ | ∈ L1 (T ).
Hence f ∗ does not vanish on T \ K with σ(T \ K) > 0 which implies the result (17.21). This
completes the analysis of the problem.
Problem 17.15
Rudin Chapter 17 Exercise 15.
Proof. Denote CK to be the set of all continuous functions on K.g Then the problem is equivalent
to show that the set of polynomials PK on K is dense in CK with respect to the norm
This is well-defined because f is continuous on the compact set K. Assume that PK was not
/ PK . It is clear that PK
dense in CK . In other words, there exists an g ∈ CK such that g ∈
g
See Definition 3.16, p. 70.
17.3. Factorization of f ∈ H p 531
is a linear subspace of the normed linear space CK . Then Theorem 5.19 implies that there is a
bounded linear functional Φ on CK such that
Φ(P ) = 0
for all P ∈ PK and Φ(g) 6= 0. According to Theorem 6.19, there exists a unique regular complex
Borel measure ν on K such that Z
Φ(f ) = f dν (17.22)
K
for every f ∈ CK . This measure ν must be nonzero because of Φ(g) 6= 0.
Define µ(E) = ν(E ∩ K) for E ∈ MT . Then it is easily checked that µ is also a complex
Borel measure on T . Combining this and the representation (17.22), we get
Z Z Z
P dµ = P dν + P dµ = Φ(P ) = 0
T K T \K
for every n = 1, 2, . . .. Since µ 6≡ 0, Problem 17.14 asserts that the support of µ is exactly all of
T so that µ(T \ K) > 0 because K is a proper compact (hence closed) subset of T . However, the
definition of µ implies that µ(T \ K) = ν(∅) = 0, a contradiction. This completes the analysis
of the problem.
17.3 Factorization of f ∈ H p
Problem 17.16
Rudin Chapter 17 Exercise 16.
Proof. Suppose that 0 < p < 1. Recall from the first paragraph of the proof of Theorem 7.17
that we may assume that f has no zeros in U . Therefore, we deduce from Theorem 17.10 that
2
one can find a zero-free function h ∈ H 2 such that f = h p . Note that h = Mh Qh by Theorem
17.17, so we have
2 2
f = Mhp Qhp . (17.23)
n 1 Z eit + z o
= exp it
log |f ∗ (eit )| dt . (17.25)
2π T e − z
Since log |h∗ | ∈ L1 (T ), we immediately have log |f ∗ | ∈ L1 (T ). By Definition 17.14, the function
(17.25) is an outer function and we let it be Qf . Since Qh ∈ H 2 , Theorem 17.16 implies that
|h∗ | ∈ L2 (T ) and thus |f ∗ | ∈ Lp (T ). Using Theorem 17.16 again, we conclude that Qf ∈ H p .
By substituting the expressions (17.24) and (17.25) into the formula (17.23), we obtain
f = Mf Qf .
Hence equality holds in (17.27) if and only if equality holds in (17.26) if and only if Mh is
constant if and only if Mf is constant too. Consequently, we have completed the proof of the
problem.
Problem 17.17
Rudin Chapter 17 Exercise 17.
Proof.
(a) Assume that ϕ1 ∈ H p for some p > 0. Since ϕ1 has no zero in U and nonconstant, Theorem
17.10 (The Riesz Factorization Theorem) implies that there is a zero-free function f ∈ H 2
such that
1 2
= f p.
ϕ
1
Therefore, p = f is in H 2 . Let
ϕ2
∞
X
f (z) = an z n .
n=0
By the Parseval Theorem [100, Eqn. (6), p. 91] and also the proof of Theorem 17.12, we
get
X∞ Z Z Z
1 1
|an |2 = lim |fr |2 dσ = lim dσ = ∗ |p
dσ.
r→1 T r→1 T |ϕ|p T |ϕ
n=0
Since ϕ is an inner function in U , Definition 17.14 implies that |ϕ∗ | = 1 a.e. on T and
then
X∞
|an |2 = 1.
n=0
which means u = P [ dµ]. Since µ ⊥ m, it follows from Problem 11.19 that u(reiθ ) → ∞
a.e. [µ]. Consequently, there exists an eiθ ∈ T such that
lim ϕ(reiθ ) = 0.
r→1
Problem 17.18
Rudin Chapter 17 Exercise 18.
To this end, we directly apply the following result from [112, p. 323]:
Lemma 17.2
If ψ and ϕ are inner functions in U , then ψ ◦ ϕ is also inner.
z−α
Since ψα (z) = 1−αz is clearly inner and ϕα = ψα ◦ ϕ, we follow from Lemma 17.1 that ϕα is
also inner. Now ϕα is nonconstant because ϕ is also nonconstant. Therefore, we conclude that
the result (17.28) holds and we have completed the proof of the problem.
Problem 17.19
Rudin Chapter 17 Exercise 19.
534 Chapter 17. H p -Spaces
Proof. Let g = f1 . Since f, g ∈ H 1 and f, g are not identically 0, we deduce from Theorem 17.17
(The Canonical Factorization Theorem) that
f = Mf Qf and g = Mg Qg
which imply that 1 = (Mf Mg )(Qf Qg ). By Definition 17.14 and Theorem 17.15, finite products
of inner functions and outer functions remain inner and outer respectively. Thus we can write
1 = MQ (17.29)
for some inner and outer functions M and Q, where M has no zero in U .
We claim that this factorization (17.29) is unique up to a constant of modulus 1. Suppose
that we have 1 = M1 Q1 = M2 Q2 . By Theorem 17.15, M1 and M2 can be expressed in the
form [95, Eqn. (1), p. 342] which gives |M1 (z)| = |M2 (z)| = 1 on T . Therefore, we also have
|Q1 (z)| = |Q2 (z)| = 1 on T . Since
Q1 M2 Q2 M1
= and = ,
Q2 M1 Q1 M2
Q1 Q2 Q1 Q2
both Q2 and Q1 are inner functions without zero in U . In other words, we have Q2 , Q1 ∈ H(U ).
Q1 (z) Q2 (z)
Since | Q 2 (z)
| = |Q 1 (z)
| = 1 on T , it follows from Theorem 10.24 (The Maximum Modulus
Q1 (z) 2 (z)
Theorem) that | Q2 (z) | ≤ 1 and | Q
Q1 (z) | ≤ 1 in U which imply that
Problem 17.20
Rudin Chapter 17 Exercise 20.
1 1
Proof. Given ǫ > 0. Define fǫ (z) = f (z) + ǫ in U . Then fǫ is bounded in U so that fǫ ∈ H 1.
Clearly, Problem 17.18 implies that fǫ = Qfǫ , i.e.,
n 1 Z π eit + z o
fǫ (z) = c exp log (fǫ )∗ (eit ) dt (17.30)
2π −π eit − z
for some constant c with |c| = 1. By the definition, the functions log |(fǫ )∗ | decrease to log |f ∗ |
as ǫ → 0. Next, we know from Theorem 17.17 (The Canonical Factorization Theorem) that
log |f ∗ | ∈ L1 (T ). Finally, we apply Problem 1.7 to the expression (17.30) to conclude that
n 1 Z π eit + z o
f (z) = c exp it
log |f ∗ (eit )| dt = Qf (z)
2π −π e − z
Problem 17.21
Rudin Chapter 17 Exercise 21.
Proof. Suppose first that f = hg , where g, h ∈ H ∞ . There is no loss of generality to assume that
|g(z)| ≤ 1 and |h(z)| ≤ 1. By the definition of log+ , it is easy to see that
Z π Z π Z π
log+ |fr | dt = log+ |f (reit )| dt ≤ log |h(reit )| dt (17.31)
−π −π −π
Combining the inequality (17.31) and the result (17.32), we conclude that f ∈ N .
Conversely, let f ∈ N and f 6≡ 0. According to Theorem 17.9, we may assume that f has no
zero in U . By Problem 11.5, u = log |f | is harmonic in U . Thus the mean value property gives
Z
1
u(0) = u(r) dt
2π T
Z
1
= log |fr (eit )| dt
2π T
Z Z
1 1
= log+ |fr (eit )| dt − log− |fr (eit )| dt
2π T 2π T
Z
1
≤ log kf k0 − log− |fr (eit )| dt (17.33)
2π T
for all 0 ≤ r < 1. Since f ∈ N and the left-hand side of the equation (17.33) is independent of
r, we know immediately that
Z
1
sup log− |fr (eit )| dt < ∞
0≤r<1 2π T
so that
sup kur k1 = sup
log |f |
1
0≤r<1 r<1
Z
1
= sup log |fr (eit )| dt
0≤r<1 2π T
n 1 Z 1
Z o
= sup log+ |fr (eit )| dt + log− |fr (eit )| dt
0≤r<1 2π T 2π T
<∞
By the definition, u is real, so µ is in fact real. Using Theorem 6.14 (The Hahn Decomposition
Theorem), we can write µ = µ+ −µ− . Put u± = P [ dµ± ]. Then both u+ and u− are nonnegative
harmonic functions in U .
536 Chapter 17. H p -Spaces
Since U is simply connected and f has no zero in U , it follows from Theorem 13.11 that
f = eg for some g ∈ H(U ). We may assume that g(0) = log |f (0)| so that g is unique. Recall
that u = log |f |, so we have u = Re g and then the expression (17.34) implies that
Z it
e +z
g(z) = it − z
dµ(eit ).
T e
If we denote Z
eit + z ± it
g ± (z) = dµ (e ),
T eit − z
then we have g = g + − g− and Re g ± = u± ≥ 0. Suppose that
− +
g1 = e−g and g2 = e−g .
Clearly, g1 , g2 ∈ H(U ) and they have no zero in U . In addition, we have
−
|g1 | = e−Re g = e−u− ≤ 1.
Similarly, we have |g2 | = e−u+ ≤ 1. These mean that g1 , g2 ∈ H ∞ and furthermore,
g1 exp(−g− )
= = exp(g+ − g − ) = eg = f.
g2 exp(−g+ )
This completes the proof of the problem.
Problem 17.22
Rudin Chapter 17 Exercise 22.
Proof. Since log+ t ≤ | log t| for every 0 < t < ∞, the hypothesis gives
Z π
lim log+ |f (reiθ )| dθ = 0. (17.35)
r→1 −π
Since f ∈ H(U ), Theorem 17.3 says that log+ |f | is subharmonic in U . Since log+ |f | is contin-
uous and nonnegative in U , it follows from Theorem 17.5 that
Z π Z π
0≤ log+ |f (r1 eiθ )| dθ ≤ log+ |f (r2 eiθ )| dθ (17.36)
−π −π
if 0 ≤ r1 < r2 < 1. Combining the limit (17.35) and the inequality (17.36), we deduce that
Z π
log+ |f (reiθ )| dθ = 0
−π
for every 0 ≤ r < 1. In other words, we have log+ |f (z)| = 0 and so |f (z)| ≤ 1 for every z ∈ U ,
i.e., f ∈ H ∞ .
Recall from Definition 17.7 that H ∞ ⊆ N . The hypothesis yields that f ≡ 0, so Theorem
17.9 asserts that
f = Bg, (17.37)
where B is the Blaschke product formed with the zeros of f and g ∈ N . Clearly, we always have
|g(z)| ≤ 1 in U . Furthermore, we see from the representation (17.37) that
Z π Z π Z π
1
0 = lim iθ
log |f (re )| dθ = lim iθ
log |g(re )| dθ = lim log dθ.
r→1 −π r→1 −π r→1 −π iθ
|g(re )|
1
Therefore, the previous paragraph implies that | g(z) | ≤ 1 in U so that |g(z)| = 1 in U . This
means that g is a constant of modulus 1 and then f is a Blaschke product. We have completed
the analysis of the problem.
17.3. Factorization of f ∈ H p 537
Problem 17.23
Rudin Chapter 17 Exercise 23.
M1 = ϕ1 H 2 and M2 = ϕ2 H 2 .
and Z
n π
eit + z o
ϕ2 (z) = c2 B2 (z) exp − it
dµ2 (t) (17.39)
−π e −z
where c1 and c2 are constants such that |c1 | = |c2 | = 1, B1 and B2 are Blaschke products, and
µ1 and µ2 are finite positive Borel measures on T such that µ1 , µ2 ⊥ m.
ϕ1
We claim that M1 ⊆ M2 if and only if the quotient ϕ = ϕ 2
is an inner function. This is
equivalent to saying that every zero of B2 is also a zero of B1 with at least the same multiplicity
and µ2 (E) ≤ µ1 (E) for every Borel subset E of T .
To prove the claim, we first suppose that every zero of B2 is also a zero of B1 with at least
B1
the same multiplicity and µ2 (E) ≤ µ2 (E) for every Borel subset E of T . Clearly, B 2
is another
Blaschke product B. Next, it follows from the representations (17.38) and (17.39) that
ϕ1 (z) c1 n Z π eit + z o
ϕ(z) = = B(z) exp − it
d µ 1 (t) − µ 2 (t) .
ϕ2 (z) c2 −π e − z
Define the measure µ on T by µ(E) = µ1 (E) − µ2 (E) for every Borel subset E of T . Since
µ2 (E) ≤ µ1 (E), µ is a finite positive measure. Recall that µ1 , µ2 ⊥ m, so Proposition 6.8 gives
µ ⊥ m. By Theorem 17.15, the function ϕ is inner.
Conversely, let ϕ1 H 2 ⊆ ϕ2 H 2 . Since 1 ∈ H 2 , it follows that ϕ1 ∈ ϕ2 H 2 and then ϕ1 = ϕ2 h
for some h ∈ H 2 . Since |ϕ∗1 | = |ϕ∗2 | = 1 a.e. on T , we have |h∗ | = 1 a.e. on T . Consequently,
h∗ ∈ L∞ (T ). By Problem 17.6, we know that H 2 ⊂ H 1 . Then we deduce from Problem 17.8
that h ∈ H ∞ . By Definition 17.14, h is an inner function, so we may write
n Z π eit + z o
h(z) = cB(z) exp − it
dµ(t) ,
−π e − z
and n Z o n Z π eit + z
π
eit + z o
exp − dµ 1 (t) = exp − d µ(t) + µ 2 (t) . (17.41)
−π eit − z it
−π e − z
Now the relation (17.40) implies that every zero of B2 is also a zero of B1 with at least the same
multiplicity. Next, the expression (17.41) implies that there exists a real number C such that
Z π it Z π it
e +z e +z
it
dµ 1 (t) = it
d µ(t) + µ2 (t) + iC
−π e − z −π e − z
538 Chapter 17. H p -Spaces
for all z ∈ U . Put z = 0 and recall that the measures are finite positive, we get C = 0. Recall
from the formula following [100, Eqn. (8), p. 111] that
X∞
eit + z
= 1 + 2 (ze−it )n , (17.42)
eit − z n=1
It is easy to see that the power series converge in U , so they are holomorphic in U by Theorem
10.6. Since the power series representation of any f ∈ H(U ) is unique, we have
Z π Z π
dµ1 (t) = d µ(t) + µ2 (t) (17.43)
−π −π
and Z π Z π
e−int dµ1 (t) = e−int d µ(t) + µ2 (t) (17.44)
−π −π
for every n ∈ N. Since all the measures are positive, it yields from taking complex conjugates
to the expression (17.44) that it also holds for all nonzero integers n. Combining this with the
expression (17.43), we may conclude that
Z π
eint d[µ(t) + µ2 (t) − µ1 (t)] = 0
−π
Problem 17.24
Rudin Chapter 17 Exercise 24.
Proof. By the explanation in §17.24, Theorem 17.26 (M. Riesz’s Theorem) is equivalent to saying
thath
sup kvr kp < ∞.
0≤r<1
h
See also [34,Z Theorem 4.1, p. 54]. More precisely, it says that if u is harmonic in U and it satisfies
π
1
kukp = sup |u(reiθ )|p dθ < ∞, then there exists a constant Ap , depending only on p, such that
0≤r<1 2π −π
kvkp ≤ Ap kukp ,
which is impossible. Hence we have f ∈/ H 1 and Theorem 17.26 (M. Riesz’s Theorem) also fails
in the case. This completes the proof of the problem.
Remark 17.1
Although Theorem 17.26 (M. Riesz’s Theorem) fails to say that sup kvr k1 < ∞, it is true
0≤r<1
that sup kvr kp < ∞ for all p < 1. In fact, this is the content of the so-called Kolmogorov’s
0≤r<1
Theorem, see [34, Theorem 4.2, p. 57].
Problem 17.25
Rudin Chapter 17 Exercise 25.
Proof.
Define
∞
(ψf )(z) + fb(0) X b
F (z) = = f (n)z n .
2 n=0
540 Chapter 17. H p -Spaces
Now Theorem 17.26 (M. Riesz’s Theorem) implies that F ∈ H p . By Theorem 17.11,
g = F ∗ ∈ Lp (T ). Direct computation gives
Z π
1
gb(n) = g(eit )e−int dt
2π −π
Z π
1
= F ∗ (eit )e−int dt
2π −π
X∞ Z π
1
= fb(m) · ei(m−n)t dt
2π −π
m=0
fb(n), for all n ≥ 0;
=
0, for all n < 0.
for every n ∈ Z. Finally, we obtain from the inequality (17.46), Theorem 17.11 and
Theorem 17.26 (M. Riesz’s Theorem) that
|fb(0)| kψf kp kf kp
kgkp = kF ∗ kp = kF kp ≤ + ≤ + Ap kf kp = Cp kf kp ,
2 2 2
where Cp = 21 + Ap which is a constant depending only on p. In addition, this means that
the mapping Φ : Lp (T ) → Lp (T ) defined by
Φ(f ) = g
(b) Let k > 0. It follows from the representation (17.45) that we may define
1
F (z) = (ψf )(z) + fb(0) − 2fb(0) − 2fb(1)z − · · · − 2fb(k − 1)z k−1
2
X∞
= fb(n)z n .
n=k
Finally, if we combine the inequality (17.46), Theorem 17.11 and Theorem 17.26 (M.
Riesz’s Theorem), then we get
kgkp = kF kp
k−1
|fb(0)| X b
≤ Ap kf kp + + |f (n)|
2
n=0
k−1
X
kf kp
≤ Ap kf kp + + kf kp
2 n=0
17.4. A Projection of Lp onto H p 541
= Cp,k · kf kp ,
Now if we define g = F ∗ , then it is easy to check that g satisfies the condition (17.47) so
that
k
|fb(0)| X b
kgkp ≤ Ap kf kp + + |f (n)| = Cp,k · kf kp ,
2 n=0
3
where Cp,k = Ap + k + 2 in this case.
(c) We have to show that there exists a constant M > 0 such that ksn kp ≤ M for n = 1, 2, . . ..
Recall that
Xn
sn (t) = fb(k)eikt = g1 (t) − g2 (t),
k=−n
where
∞
X ∞
X
g1 (t) = fb(k)eint and g2 (t) = fb(k)eint .
k=−n k=n+1
Using part (b), there exists a constant Cp , depending only on p, such that
so that
ksn kp ≤ kg1 kp + kg2 kp ≤ 2Cp kf kp (17.48)
for all n ∈ N. Since f ∈ Lp (T ), we may take M = 2Cp kf kp and we are done.
Given ǫ > 0. Recall from Theorems 3.14 and 4.25 (Weierstrass Theorem) that there exists
a trigonometric polynomial g such that
ǫ
kf − gkp < . (17.49)
1 + 2Cp
Clearly, if n ≥ deg g, then we have
n
X
sn (g) = g(n)eint = g(t)
b (17.50)
−n
For sufficiently large enough n, we get from the inequalities (17.48), (17.49) and the ex-
pression (17.50) that
ǫ ǫ
ksn (f ) − f kp < + ksn (f − g)kp ≤ + 2Cp kf − gkp < ǫ
1 + 2Cp 1 + 2Cp
which implies the desired result
lim kf − sn (f )kp = 0.
n→∞
542 Chapter 17. H p -Spaces
(d) Suppose that f ∈ Lp (T ) and set fb(n) = 0 for all n < 0. Using the estimate (17.46), we
know that
1 1
lim sup |fb(n)| n ≤ lim sup kf kpn = 1
n→∞ n→∞
which implies that F ∈ H D(0; R) for some R ≥ 1, where F is the function defined in
part (a). In fact, F is the Poisson integral of f because
Z π
1
P [f ](z) = Pr (θ − t)f (eit ) dt
2π −π
X∞ Z π
1
= r |n| × f (eit )ein(θ−t) dt
n=−∞
2π −π
∞
X
= fb(n)r n einθ
n=0
X∞
= fb(n)(reiθ )n
n=0
= F (z).
Thus it follows from Theorem 11.16 that kFr kp ≤ kf kp for all 0 ≤ r < 1. By Definition
17.7, we conclude that F ∈ H p .
Conversely, we suppose that F ∈ H p . By Theorem 17.11, we have f = F ∗ ∈ Lp (T ). Since
1 < p < ∞, we have H p ⊂ H 1 so that F is the Cauchy integral of f , i.e.,
Z π
1 f (eiθ )
F (z) = −iθ z
dθ
2π−π 1 − e
Z π X ∞
1
= e−inθ z n f (eiθ ) dθ
2π −π n=0
∞ h
X Z π i
1
= f (eiθ )e−inθ dθ z n
2π −π
n=0
X∞
= fb(n)z n .
n=0
Remark 17.2
(a) Problem 17.25(a) can be treated as an equivalent form of Theorem 17.26 (M. Riesz
Theorem), see the first paragraph of the proof of the theorem in [54, p. 152].
Problem 17.26
Rudin Chapter 17 Exercise 26.
and then
∞
X ∞
X
H(reiθ ) = b
h(0) + 2b
h(n)(reiθ )n = b
H(n)(reiθ n
) , (17.51)
n=1 n=−∞
where
b
2h(n), if n = 1, 2, . . .;
b
H(n) = b
h(0) if n = 0;
0, if n = −1, −2, . . ..
Using the Parseval Theorem [100, Eqn. (6), p. 91] and the Fourier series (17.51), we see that
Z π
1
kHk22 = |H(reiθ )|2 dθ
2π −π
X∞
= b
|H(n)|2
n=−∞
∞
X
= |b
h(0)|2 + |2b
h(n)|2
n=1
∞
X
≤4 |b
h(n)|2 + |b
h(0)|2 (17.52)
n=−∞
Z π
1
=4· |h(eit )|2 dt + |b
h(0)|2
2π −π
= 4khk22 + |b
h(0)|2 .
When bh(n) = 0 for n = 0, −1, −2, . . .,i the equality (17.52) holds and we get kHk2 = 2khk2 .
Hence the best possible value of A2 is 2 and we complete the proof of the problem.
i
This is equivalent to Z π
h(eiθ )e−inθ dθ = 0
−π
Problem 17.27
Rudin Chapter 17 Exercise 27.
Proof. Define g : U → C by
∞
X
g(z) = |an |z n .
n=0
Problem 17.28
Rudin Chapter 17 Exercise 28.
Thus the radius of convergence of the power series will be 1 and Theorem 10.6 implies that
∞
X k−1
X ∞
X
nk z nk −1 np z np −1 nk z nk −1 np z np −1
f ′ (z) = = + +
k p=1
p k p
k=1 p=k+1
which implies
nk |z|nk −1 X np z np −1 X np z np −1
k−1 ∞
|f ′ (z)| ≥ − − (17.54)
k p p
p=1 p=k+1
17.5. Miscellaneous Problems 545
1 1
for all z satisfying 1 − nk < |z| < 1 − 2nk . Since nk is sufficiently large enough, we have
nk |z|nk −1 nk 1 nk −1 nk
> 1− ≈ (17.55)
k k nk ek
Xk−1
np z np −1 X
k−1
≤ np |z|np −1
p=1
p p=1
np 1 np −1
k−1
X
≤ 1−
p=1
p 2np
np h 1 2np −2 i 12
k−1
X
≤ 1−
p=1
p 2np
k−1
X np
≈ √
p=1
ep
(k − 1)nk−1
≤ √
e
nk
< . (17.56)
4ek
If p ≥ k + 1, then np − 1 = exp(exp · · · exp nk ) − 1. Therefore, we obtain
| {z }
(p − k) iterations
X
∞
np z np −1
∞
X np h 1 2nk i 2nk
np −1
≤ 1−
p p 2nk
p=k+1 p=k+1
X∞
np
≈ np −1 . (17.57)
p=k+1 p exp( 2nk )
Besides the property (17.53), we require that the sequence {nk } satisfies
np − 1 4eknp
exp > 2p−k ×
2nk pnk
Substituting the estimates (17.55), (17.56) and (17.58) into the inequality (17.54), we get
nk nk nk nk nk
|f ′ (z)| > − − = >
ek 4ek 4ek 2ek 10k
1 1
in 1 − nk < |z| < 1 − 2nk .
• The divergence of the integral. It is clear from the first assertion that
Z 1 ∞ Z
X 1− 2n1
k
|f ′ (reiθ )| dr ≥ |f ′ (reiθ )| dr
0 k=1 1− n1
k
546 Chapter 17. H p -Spaces
∞ Z
X 1− 2n1
k nk
≥ dr
1− n1 10k
k=1 k
nk 1 1
∞
X
= −
10k nk 2nk
k=1
∞
X 1
=
20k
k=1
which is the total variation of f on the radius of U terminating at the point eiθ . Hence
the result (17.59) tells us that the length of the curve which is the image of the radius
with angle θ under f is always infinite.
Problem 17.29
Rudin Chapter 17 Exercise 29.
Proof. Let g ∈ Lp (T ). Suppose that g(eit ) = f ∗ (eit ) a.e. for some f ∈ H p . Let
∞
X
f (z) = an z n
n=0
17.5. Miscellaneous Problems 547
and fc∗ (n) be the Fourier coefficients of its boundary function f ∗ (eit ), i.e.,
Z π
1
fc∗ (n) = e−int f ∗ (eit ) dt (17.61)
2π −π
for all n ∈ Z. The Taylor coefficients of f can be expressed in the form
Z π Z π
1 f (reit ) 1
an = dt = r −n e−int f (reit ) dt (17.62)
2π −π r n eint 2π −π
for every 0 < r < 1. Combining the coefficients (17.61) and (17.62), we get
Z π Z π
n
r an − fc∗ (n) = 1 e−int f (reit ) − f ∗ (eit ) dt ≤
1 f (reit ) − f ∗ (eit ) dt
2π −π 2π −π
for every n ∈ Z. Applying Theorem 17.11, we obtain
n
r an − fc∗ (n) ≤ kfr − f ∗ k1 → 0
as r → 1. Therefore, we have
an , if n ≥ 0;
fc∗ (n) =
0, if n < 0.
By the hypothesis, we conclude that
Z π Z π
1 1
e−int g(eit ) dt = e−int f ∗ (eit ) dt = 0 (17.63)
2π −π 2π −π
for all negative integers n.
Conversely, we suppose that the formula (17.63) holds for all negative integers n. In other
words, gb(n) = 0 for all negative integers n. Let
Z π
1
f (reiθ ) = Pr (θ − t)g(eit ) dt, (17.64)
2π −π
where 0 < r < 1. Since the Poisson kernel has the expansion
∞
X
Pr (t) = 1 + r n (eint + e−int ),
n=1
which means
∞
X
f (z) = g(n)z n
b
n=0
548 Chapter 17. H p -Spaces
for every z ∈ U and then f ∈ H(U ). By the integral (17.64) and [100, Eqn. (3), p. 233], it is
easy to see that
Z π
1
kf k1 = |f (reiθ )| dθ
2π −π
Z πh Z π i
1 1
≤ Pr (θ − t)|g(eit )| dt dθ
2π −π 2π −π
Z πh Z π i
1 1
= Pr (θ − t) dθ · |g(eit )| dt
2π −π 2π −π
Z π
1
= |g(eit )| dt
2π −π
so that f ∈ H 1 .
By Theorem 17.11, we see that f ∗ ∈ L1 (T ). Now we consider
Z π
1
Φ(z) = Pr (θ − t)f ∗ (eit ) dt = P [f ∗ ](z).
2π −π
Let z ∈ U . For any fixed 0 < ρ < 1, since f ∈ H(U ), it follows from Theorem 11.4 and the
mean value property that
Z π
1
f (ρz) = Pr (θ − t)f (ρeit ) dt.
2π −π
so we obtain
f (z) = lim f (ρz) = Φ(z)
ρ→1
Comparing the two integrals (17.64) and (17.65), we conclude immediately that g(z) = f ∗ (z)
a.e. on T . Since g ∈ Lp (T ), we know that f ∗ ∈ Lp (T ) and it yields from Problem 17.8 that
f ∈ H p , as required. Now we have completed the proof of the problem.
CHAPTER 18
Elementary Theory of Banach Algebras
Problem 18.1
Rudin Chapter 18 Exercise 1.
Denote k · kX to be the norm of the Banach space X. The hypotheses ensure that an associative
and distributive multiplication is well-defined in B(X). For every α ∈ C, A1 , A2 ∈ B(X) and
x ∈ X, we see that
k(αA)(x)kX kAxkX
kαAk = sup = |α| sup = |α| · kAk.
kxkX kxkX
549
550 Chapter 18. Elementary Theory of Banach Algebras
Let kAk = 0. This means that kAxkX = 0 for all x ∈ X so that Ax = 0 for all x ∈ X.
Consequently, it must be the case A = 0 and then B(X) is a normed linear space.
For A1 , A2 ∈ B(X), we see that
k(A1 A2 )(x)kX
kA1 A2 k = sup
kxkX
kA1 (A2 x)kX kA2 xkX
= sup ×
kA2 xkX kxkX
kA1 (A2 x)kX kA2 xkX
≤ sup × sup
kA2 xkX kxkX
≤ kA1 k · kA2 k.
for all n ≥ N . Denote M = max{kA1 k, kA2 k, . . . , kAN −1 k, 1 + kAN k}. Thus we get
kAm − An k ≤ ǫ
for all m ≥ N and all x ∈ X. By the definition, we conclude that kAm − Ak ≤ ǫ for all m ≥ N .
Hence this proves our claim that B(X) is complete and then it is a Banach algebra, completing
the proof of the problem.
18.1. Examples of Banach Spaces and Spectrums 551
Problem 18.2
Rudin Chapter 18 Exercise 2.
X = {v = (z1 , z2 , . . . , zn ) | z1 , z2 , . . . , zn ∈ C} = Cn
is equipped with the norm k · kCn and B(Cn ) is the algebra of all bounded linear operators on
Cn . By Problem 18.1, B(Cn ) is also a Banach algebra with the norm k · k given by
kAvkCn
kAk = sup .
kvkCn
Our target is to find
σ(A) = {λ ∈ C | A − λI is not invertible}.
According to the explanation in [62, pp. 96, 97], every bounded linear operator A can be
represented by a matrix with entries in C. We also denote this matrix by A. Therefore, A − λI
is not invertible if and only if
det(A − λI) = 0. (18.3)
Since det(A−λI) = 0 is an equation in λ of order n, the Fundamental Theorem of Algebra ensures
that it has at most n complex roots. Hence σ(A) consists of at most n complex numbers and
they are exactly the solutions of the equation (18.3), completing the proof of the problem.
Problem 18.3
Rudin Chapter 18 Exercise 3.
Proof. Suppose that C is a positive constant such that |ϕ(x)| ≤ C a.e. on R. Define the mapping
Mϕ : L2 → L2 by
Mϕ (f ) = ϕ × f.
Of course, it is true that ϕf ∈ L2 , so Mϕ is well-defined. Furthermore, the linearity of Mϕ is
clear. Recall that L2 is a normed linear space with the norm kf k2 . Since we have
Z
kϕf k2 = |ϕ(x)f (x)|2 dx ≤ Ckf k2 ,
R
so we get
kMϕ k = sup kMϕ (f )k | f ∈ L2 and kf k2 = 1
= sup kϕf k2 | f ∈ L2 and kf k2 = 1
≤ C.
On the one hand, let λ ∈ / Rϕ . Then |ϕ(x) − λ| ≥ ǫ for some ǫ > 0 a.e. on R, so we have
1 ∞ (R) and this implies that the operator M
ϕ−λ ∈ L 1 is bounded. Furthermore, it is easy to
ϕ−λ
see that
M 1 Mϕ (f ) − λf = M 1 (ϕf − λf ) = f.
ϕ−λ ϕ−λ
The definition (18.4) reveals that m(Sn ) > 0. Suppose that there exists an N ≥ n such that
0 < m(SN ) < ∞. Otherwise, m(Sn ) = ∞ for all n ≥ 1 and this means that ϕ(x) = λ for almost
all x ∈ R. In this case, we know that Rλ = {λ}. Clearly, Mλ f − λf = 0, so λ ∈ σ(Mλ ). If
µ 6= λ, then Mλ (f ) − µf = (λ − µ)f so that
M 1 Mλ (f ) − µf = f.
λ−µ
Consequently, we obtain σ(Mλ ) = {λ} and then the equality (18.5) holds. Let 0 < m(SN ) < ∞.
Take φn = χSN . Then we have
Z
2 2
k(Mϕ − λI)(φn )k = kϕφn − λφn k = |ϕ(x) − λ|2 · |φn (x)|2 dx ≤ 2−2n kφn k2
SN
so that the operator (Mϕ − λI)−1 is not bounded, i.e., Mϕ − λI is not invertible. Hence we
conclude that λ ∈ σ(Mϕ ) and we have established the equality (18.5), completing the proof of
the problem.
Problem 18.4
Rudin Chapter 18 Exercise 4.
n nX
∞ o1 o
2
Proof. Recall that ℓ2 = x = {ξ0 , ξ1 , ξ2 , . . .} kxk = |ξn |2 < ∞ and S : ℓ2 → ℓ2 is
n=0
given by
Sx = {0, ξ0 , ξ1 , . . .}
which is a bounded linear operator on ℓ2 and kSk = 1.a We want to determine
σ(S) ⊆ U . (18.6)
(S − λI)x = y (18.7)
has a unique solution x ∈ ℓ2 for every y ∈ ℓ2 . If y = (1, 0, 0, . . .), then the equation (18.7) takes
the system
−λξ0 = 1 and ξn − λξn+1 = 0
a
In fact, S is called the right-shift operator.
18.2. Properties of Ideals and Homomorphisms 553
for every n = 0, 1, 2, . . .. Solving it, we obtain ξn = −λ−(n+1) for every n ∈ N. Since |λ| < 1, we
have |ξn | = |λ|−(n+1) > and thus x ∈ / ℓ2 which is a contradiction. Hence we must have λ ∈ σ(S),
i.e.,
{λ ∈ C | 0 < |λ| < 1} ⊆ σ(S). (18.8)
Finally, we observe from Theorem 18.6 that σ(S) is a closed set, so we conclude from the set
relations (18.6) and (18.8) that
σ(S) = U .
We end the analysis of the problem.
Problem 18.5
Rudin Chapter 18 Exercise 5.
Proof. Let M be an ideal of the commutative complex algebra A. Then M is a vector space and
we note from §4.7 that M is also a vector space. Thus it remains to show that aM ⊆ M and
M a ⊆ M for all a ∈ A. Fix a ∈ A and consider b ∈ M . Then there exists a sequence {bn } ⊆ M
such that bn → b as n → ∞. By considering the sequences {abn } and {bn a}, since M is an ideal,
it is true that
abn ∈ M and bn a ∈ M
for all n = 1, 2, . . .. Since a is fixed, the mapping x →
7 ax and x 7→ xa are both continuous on
M . Therefore, abn → ab and bn a → ba as n → ∞. In other words, it is true that ab, ba ∈ M .
This completes the proof of the problem.
Problem 18.6
Rudin Chapter 18 Exercise 6.
Proof. Let C(X) be the algebra of all continuous complex functions on X with pointwise addition
multiplication and the supremum norm. The constant function 1 is the unit element. Let I be
an ideal in C(X). We claim that either I = C(X) or there exists a p ∈ X such that
Assume that, for every p ∈ X, there exists a continuous function f ∈ I such that f (p) 6= 0.
Since X is compact Hausdorff, the continuity of f implies that there exists an open set Vp
containing p such that f (x) 6= 0 for all x ∈ Vp . Therefore, the collection {Vp } forms an open
covering of X. Since X is compact, there must exist a finite subcover. Call this subcover
V1 , V2 , . . . , VN and the corresponding functions f1 , f2 , . . . , fN for some N ∈ N. Define
Since fk ∈ I and I is an ideal, it follows from Definition 18.12 that F ∈ I. For every p ∈ X,
we have fk (p) 6= 0 for some k ∈ {1, 2, . . . , N } so that F (p) 6= 0. Since F is continuous on the
compact set X, it must attain a minimum. By the form (18.10), it is trivial that F (x) > 0 for
1
all x ∈ X. This implies that F −1 (x) = F (x) is the inverse of F in C(X). However, we note
from Definition 18.12 that no proper ideal contains an invertible element, so we have I = C(X).
Consequently, we have obtained our claim.
554 Chapter 18. Elementary Theory of Banach Algebras
If I is maximal, then it has the form (18.9) for some p ∈ X. Assume that I ⊂ J for some
ideal J in C(X), where I 6= J. Then there corresponds an f ∈ J such that f (p) 6= 0. Since f is
continuous, one can find a neighborhood Vp of p such that f (x) 6= 0 for all x ∈ Vp . The point
set {p} is compact by Theorem 2.4. According to Urysohn’s Lemma, there exists an g ∈ C(X)
such that
{p} ≺ g ≺ Vp ,
i.e., g(p) = 1 and g(x) = 0 for all x ∈ X \ Vp . Take h = 1 − g which is also an element of C(X)
and it satisfies h(p) = 0 and h(x) = 1 for all x ∈ X \ Vp . As J has the unit, we have h ∈ J.
Next, we define
H(x) = f 2 (x) + h2 (x)
which is an element of J. Obviously, it is easy to check that H(x) > 0 on X. Since X is compact,
H attains its minimum in X and thus H is bounded from below by a positive number. Therefore,
its inverse H1 belongs to C(X) which asserts that J = C(X) by the previous paragraph. Hence
we have the expected conclusion that the ideal in the form (18.9) is maximal. This completes
the analysis of the problem.
Problem 18.7
Rudin Chapter 18 Exercise 7.
Applying the result (18.11) to the inequality (18.12), we obtain the result that Pn (z) → (λ−z)−1
uniformly on σ(x).
Assume that C \ σ(x) was disconnected. Let Ω be a (non-empty) bounded component of it.
/ σ(x). Choose {Pn } as above. For every n ∈ N and z ∈ σ(x), we have
Fix λ ∈ Ω, i.e., λ ∈
|(z − λ)Pn (z) − 1| = |z − λ| · Pn (z) − (z − λ)−1 ≤ ℓ · Ln , (18.13)
where
ℓ = sup |z − λ| and Ln = sup |Pn (z) − (z − λ)−1 |.
z∈σ(x) z∈σ(x)
The compactness of σ(x) by Theorem 18.6 asserts that both ℓ and Ln are finite. Furthermore,
∂Ω ⊆ σ(x). Next, we use Theorem 10.24 (The Maximum Modulus Theorem) to see that the
inequality (18.13) also holds on Ω. In particular, we get
ℓ · Ln
|Pn (z) − (z − λ)−1 | ≤ (18.14)
|z − λ|
for all z ∈ Ω \ {λ}. Since Ω is a component, one can find a δ > 0 small enough such that the
circle C(λ; δ) lies in Ω. Therefore, we conclude from the estimate (18.14) that
Z
Z
dz ℓLn
2πi = Pn (z) − (z − λ)−1 dz ≤ ℓ · Ln = · 2πδ = 2πℓLn . (18.15)
C(λ;δ) C(λ;δ) |z − λ| δ
18.2. Properties of Ideals and Homomorphisms 555
Notice that Ln → 0 as n → ∞, so the inequality (18.15) implies a contradiction. Hence the set
C \ σ(x) must be connected, as required. This completes the proof of the problem.
Problem 18.8
Rudin Chapter 18 Exercise 8.
∞
X
Proof. Since |cn | < ∞, there exists a positive constant M such that |cn | ≤ M . This implies
n=0
that
1
lim sup |cn | n ≤ 1,
n→∞
i.e., the radius of convergence R of the power series satisfies R ≥ 1. By Theorem 10.6, both f
and then f1 are holomorphic in a region containing U .
Define Cn = cn for all n ≥ 0 and Cn = 0 for all n < ∞. It is clear that
∞
X ∞
X ∞
X ∞
X
f (eit ) = cn eint = Cn eint and |Cn | = |cn | < ∞.
n=0 n=−∞ n=−∞ n=0
Now the hypothesis |f (z)| > 0 for every z ∈ U implies that f (eit ) 6= 0 for every real t, so
Theorem 18.21 (Wiener’s Theorem) guarantees that f satisfies
X∞ ∞
X
1
= γn eint and |γn | < ∞. (18.16)
f (eit ) n=−∞ n=∞
1
Since f is holomorphic in a region containing U , we have γn = 0 for all n < 0 and
X ∞ X ∞
1
= an eint = γn eint
f (eit )
n=0 n=0
which implies immediately that an = γn for all n ≥ 0 by the Corollary following Theorem 10.18.
Hence the second condition (18.16) gives the desired result that
∞
X
|an | < ∞.
n=0
Problem 18.9
Rudin Chapter 18 Exercise 9.
Proof. We note from Example 9.19(d) that we define the multiplication in L1 (R) by convolution.
Let f, g ∈ L1 (R) and φ ∈ L∞ (R). Here we employ the proof of [98, pp. 157, 158]: If I is a
translation invariant subspace of L1 (R), then we say that φ annihilates I if
Z
(f ∗ φ)(x) = f (x − y)φ(y) dm(y) = 0
R
On the other hand, recall from [100, Example 9.19(d)] that f ∗ g = g ∗ f , so we have
Z
(f ∗ g ∗ φ)(0) = [g ∗ (f ∗ φ)](0) = g(0 − y)(f ∗ φ)(y) dm(y). (18.18)
R
In other words, the two integrals (18.17) and (18.18) are equal, i.e.,
Z Z
(f ∗ g)(0 − y)φ(y) dm(y) = g(0 − y)(f ∗ φ)(y) dm(y). (18.19)
R R
Let I be a closed translation invariant subspace of the Banach space L1 (R) and φ ∈ L1 (R)
annihilate f ∈ I. Then f ∗ φ = 0 and we see from the right-hand side of the expression (18.19)
that
(f ∗ g) ∗ φ = 0 (18.20)
for every g ∈ L1 (R).b Recall the basic fact from Theorem
∗ 6.16 that L1 (R) is isometrically
∞ 1 ∼
isomorphic to the dual space of L (R), i.e., L (R) = L∞ (R). By Remark 5.21, every
φ ∈ L∞ (R) is a bounded linear functional on L1 (R). Assume that f ∗ g ∈ / I. Since I = I, there
corresponds an ϕ ∈ L∞ (R) such that f ∗ ϕ = 0 and (f ∗ g) ∗ ϕ 6= 0 by Theorem 5.19, but this
contradicts the result (18.20). Hence we conclude that f ∗ g ∈ I so that I is an ideal.
Conversely, let I be a closed ideal and f ∗ φ = 0 for all f ∈ I. Assume that F = fx0 ∈
/I=I
for some x0 ∈ R.c By Theorem 5.19, there exists an ϕ ∈ L∞ (R) such that f ∗ ϕ = 0 for all f ∈ I
but
F ∗ ϕ 6= 0. (18.21)
Now the hypotheses show that we have f ∗ g ∈ I for every g ∈ L1 (R), so the left-hand side of
the expression (18.19) gives f ∗ φ annihilates every g ∈ L1 (R). This implies that f ∗ φ = 0 or
Z
0 = (f ∗ φ)(x) = f (x − y)φ(y) dm(y)
R
for x ∈ R. In other words, φ annihilates every translate of f . Particularly, this implies that
F ∗ ϕ = 0 which contradicts the result (18.21). Consequently, fx ∈ I for every x ∈ R which
means it is translation invariant. Hence we have completed the analysis of the problem.
Remark 18.1
As [98, Theorem 7.1.2, p. 157] indicates, the result of Problem 18.9 is also valid if we replace
R by any locally compact abelian group. In particular, Problem 18.9 remains true for
the unit circle T .
Problem 18.10
Rudin Chapter 18 Exercise 10.
Recall from [100, p. 96] that L1 (T ) is a Banach space normed by kf k1 . Using Theorem
8.8 (The Fubini Theorem) again, we know that
Z π
1
kf ∗ gk1 = |(f ∗ g)(t)| dt
2π −π
Z π Z
1 1 π
= f (t − s)g(s) ds dt
2π −π 2π −π
Z π Z π
1 1
≤ |f (t − s)| · |g(s)| ds dt
2π −π 2π −π
Z π Z π
1 1
= |f (t − s)| dt · |g(s)| ds
2π −π 2π −π
Z π
1
= kf k1 · |g(s)| ds
2π −π
= kf k1 · kgk1 .
• L1 (T ) does not have a unit. Assume that e ∈ L1 (T ) was a unit. Then e∗f = f for every
f ∈ L1 (T ). Using similar argument as in the proof of Theorem 9.2(c) (The Convolution
Theorem), we can show that
b
h(n) = fb(n) · gb(n) (18.22)
if f, g ∈ L1 (T ). Therefore, we have
eb(n) = 1
for every n ∈ Z, but it contradicts the Riemann-Lebesgue Lemma [100, §5.14, p. 103].
The fact ϕ(f ∗ g) = ϕ(f )ϕ(g) asserts that β(t) = β(t − s)β(s) a.e. on T or equivalently,
β(x + y) = β(x)β(y)
a.e. on T . Employing similar analysis as in the proof of Theorem 9.23, since β is periodic
with period 1, it has the form
β(x) = e−iαx
558 Chapter 18. Elementary Theory of Banach Algebras
for a unique α ∈ R.e Since we must have β(x + 2π) = β(x), α must be an integer. Put
α = n. Substituting this back into the integral (18.23), we get
Z π
1
ϕ(f ) = f (t)e−int dt = fb(n)
2π −π
For any f ∈ IE and g ∈ L1 (T ), if we write h = f ∗ g, then the formula (18.22) implies that
b
h(n) = fb(n) · gb(n) = 0
fb(n − α) = fb(n)e−iαn = 0
• Every closed ideal I in L1 (T ) has the form (18.24). Let I be a closed ideal of L1 (T ).
We have to prove that I = IE for some set E ⊆ Z. Suppose −1that for each n ∈ Z, there
exists an fn ∈ I such that c
fn (n) 6= 0. Put gn (t) = eint fc
n (n) ∈ L 1 (T ). Then we have
Z π Z π
1 1 eint 1
(gn ∗ fn )(t) = · fn (s)ein(t−s) ds = · fn (s)e−ins ds = eint .
c
fn (n) 2π −π fc
n (n)
2π −π
Since I is an ideal, we have eint ∈ I for every n ∈ Z. Using Theorems 3.14, 4.25 and
the fact that I is closed, we know that the set {eint | n ∈ Z} is dense in L1 (T ). Thus we
conclude that I = L1 (T ).
Without loss of generality, we may assume that I 6= L1 (T ). Then there exists an n ∈ Z
such that fb(n) = 0 for all f ∈ I. Let the collection of such integers be E, i.e.,
provided that an = 0 for all n ∈ E. Suppose that f ∈ IE and we consider the set
Z(f ) = n ∈ Z fb(n) = 0 .
Now the definitions (18.24) and (18.25) imply that E ⊆ Z(f ), so it follows from Theorem
9.2(c) that if P is a trigonometric polynomial on T and n ∈ E, then we have
(f[
∗ P )(n) = fb(n) × Pb(n) = 0,
e
See also [40, Theorem 8.19, p. 247].
18.2. Properties of Ideals and Homomorphisms 559
i.e., E ⊂ Z(f ∗ g). Using [98, Theorem 2.6.6, p. 51], we see that kf − f ∗ P k1 can be made
as small as we want. Therefore, the closeness of I asserts that f ∈ I which implies IE ⊆ I.
Hence we conclude that
I = IE
as desired.
Remark 18.2
For other classes of complex homomorphisms of specific Banach algebras, please refer to
[130, §9, pp. 39 – 43].
Problem 18.11
Rudin Chapter 18 Exercise 11.
as µ → λ. This is exactly [100, Eqn. (3), p. 359], so the argument in the proof of Theorem 18.5
can be applied directly. This completes the proof of the problem.
Remark 18.3
The result in Problem 18.11 is called Hilbert’s identity.
Problem 18.12
Rudin Chapter 18 Exercise 12.
560 Chapter 18. Elementary Theory of Banach Algebras
A/ ker h ∼
= h(A) = C.
Since C is a field, ker h is a maximal ideal of A. In other words, there exists an one-to-one
correspondence between M and ∆.
Let rad A be the radical of A. Suppose that x ∈ rad A, i.e.,
\
x∈ M. (18.29)
M ∈M
Now the previous paragraph yields that the set relation (18.29) is equivalent to the condition
\
x∈ h−1 (0)
h∈∆
which means that h(x) = 0 for every h ∈ ∆. Consequently, statements (a) and (c) are equivalent.
Next, Theorem 18.17(b) means that the spectrum σ(x) is exactly the set {h(x) | h ∈ ∆}.
Since x ∈ rad A if and only if h(x) = 0 for every h ∈ ∆, this implies that x ∈ rad A if and
1
only if σ(x) = {0} if and only if kxn k n → 0 as n → ∞ by Theorem 18.9 (The Spectral Radius
Formula). Hence we have shown the three statements are equivalent, completing the proof of
the problem.
Problem 18.13
Rudin Chapter 18 Exercise 13.
Proof. Let X = C([0, 1]). Then X is a Hilbert space (and hence a Banach space). For each
f ∈ X, we define Z t
T (f )(t) = f (s) ds,
0
where t ∈ [0, 1]. Since f ∈ X, T (f ) ∈ X so that T ∈ B(X), the algebra of all bounded linear
operations on X. By Problem 18.1, B(X) is a Banach space.
Particularly, we take f (x) = 1. Then we observe
Z tZ s Z t
t2
T 2 (f )(t) = du ds = s ds = .
0 0 0 2
More generally, for n = 1, 2, . . ., we obtain
tn
T n (f )(t) =
n!
which implies that T n (f ) 6= 0 for all n > 0 and
kT n (f )k∞ 1
kT n k = sup = . (18.30)
kf k∞ n!
1
Since (n!) n → ∞ as n → ∞, we conclude immediately from the result (18.30) that
1
lim kT n k n = 0.
n→∞
Problem 18.14
Rudin Chapter 18 Exercise 14.
b : ∆ → C is given by x
Proof. By the definition, the function x b(h) = h(x) and we have the set
b = {b
A x | x ∈ A}. Denote the surjective mapping G : A → Ab by G(x) = x
b.
and
cy(h) = h(xy) = h(x)h(y) = x
x b(h)b xyb)(h).
y (h) = (b
Therefore, the map G is a homomorphism. Its kernel consists of those x ∈ A such that
b=
h(x) = 0 for every h ∈ ∆. By Problem 18.12, it is exactly the radical of A, i.e., ker x
rad A.
Combining the First Isomorphism Theorem and the previous result, we see that
b∼
A/rad A = A/ ker x b
= A.
ρ(x) ≥ kb
xk∞ .
b means that λ = x
For the other direction, λ belongs to the range of x b(h) = h(x) for some
h ∈ ∆ and it follows from Theorem 18.17(b) that this happens if and only if λ ∈ σ(x), so
Definition 18.8 establishes
• The range of x b is σ(x). The analysis in the previous part also implies that the range of
b is exactly the spectrum σ(x).
the function x
Problem 18.15
Rudin Chapter 18 Exercise 15.
Proof. Let A1 = {(x, λ) | x ∈ A and λ ∈ C} and k(x, λ)k = kxk + |λ|. For any (x, λ), (y, µ) ∈ A1 ,
we define the multiplication in A1 by
If α ∈ C, then we have
kα(x, λ)k = k(αx, αλ)k = kαxk + |αλ| = |α| · kxk + |α| · |λ| = |α| · k(x, λ)k.
As k(x, λ)k = 0 if and only if kxk + |λ| = 0 if and only if kxk = 0 if and only if x = 0 and
λ = 0, A1 is a normed linear space by Definition 5.2. Since kxyk ≤ kxk · kyk, we see that
Since the spaces A and C are complete, A1 is obviously complete and then it is a Banach
algebra with unit by Definition 18.1. It is commutative because A and C are commutative
so that (y, µ)(x, λ) equals to the right-hand side of the expression (18.31).
Now we may identify A with Φ(A) ⊆ A1 . Since (x, λ)(y, 0) = (xy + λy, 0) ∈ Φ(A) for
every (x, λ) ∈ A1 and (y, 0) ∈ Φ(A), Φ(A) is an ideal of A1 by Definition 18.12. Since
A1 = A ⊕ C, we have A1 /Φ(A) ∼ = A1 /A ∼
= C which implies that Φ(A) is a maximal ideal
of A1 as required.
Problem 18.16
Rudin Chapter 18 Exercise 16.
Proof. It is clear that H ∞ is a commutative complex algebra. Recall from §11.31 that its norm
is defined by
kf k∞ = sup{|f (z)| | z ∈ U }.
This norm makes H ∞ satisfy Definition 5.2. Thus H ∞ is a normed linear space, so it is a normed
complex algebra. The fact that H ∞ is complete has been shown in [100, Remark 17.8(c), p.
338], so H ∞ is a commutative Banach algebra by Definition 18.1. The element 1 ∈ H ∞ is easily
seen to be its unit which gives the first assertion.
Suppose that |α| < 1. Define Φα : H ∞ → C by
Φα (f ) = f (α). (18.32)
Then it satisfies Φα (f g) = f (α)g(α) = Φα (f )Φα (g). For every constant a, the function f (z) = a
gives Φα (f ) = a so it is surjective. In other words, Φα ∈ ∆. To see that there are complex
homomorphisms of H ∞ other than the point homomorphisms (18.32), we let I be the set of
functions f ∈ H ∞ such that f (α) → 0 as α → 1 and α > 0. Then it is easy to see that I is
a proper ideal of H ∞ . By Theorem 18.13, I is contained in a maximal ideal J of H ∞ which
means that there exists a complex homomorphism of H ∞ , say ϕ ∈ ∆ such that ϕ(f ) = 0 for all
f ∈ I by Theorem 18.17(a). However, ϕ 6= Φα for all α ∈ U because there is no α such that
Φα (f ) = f (α) = 0 for every f ∈ I. We have finished the proof of this problem.
Problem 18.17
Rudin Chapter 18 Exercise 17.
g · (z − 1)2 f = (z − 1)2 f g ∈ I.
|1 + ǫ − z| ≥ |1 + ǫ| − |z| > ǫ
Problem 18.18
Rudin Chapter 18 Exercise 18.
564 Chapter 18. Elementary Theory of Banach Algebras
Problem 19.1
Rudin Chapter 19 Exercise 1.
Proof. By the hypothesis, we know that there exist some constants A and C such that
|f (z)| ≤ CeA|z|
By Theorem 19.3 (The Paley and Wiener Theorem), there exists an F ∈ L2 (−A, A) such that
Z A
f (z) = F (t)eitz dt
−A
By the definition, it is clear that F1 ∈ L2 (0, ∞), so we know from [100, Eqn. (3), p. 372] that
Z ∞ Z ∞
1
|f1 (x + iy)|2 dx ≤ |F1 (t)|2 dt < ∞ (19.2)
2π −∞ 0
for every y > 0. For the second integral of the equation (19.1), we write
Z 0 Z ∞
f2 (z) = F2 (t)eitz dt = f2 (t)e−itz dt,
F
−∞ 0
565
566 Chapter 19. Holomorphic Fourier Transforms
where Ff2 (t) = F2 (−t). Since F2 ∈ L2 (−∞, 0), we have F f2 ∈ L2 (0, ∞). By similar argument as
in [100, pp. 371, 372], we can show that f2 is holomorphic in the lower half plane Π− and if we
write Z ∞
f2 (x + iy) = f2 (t)ety · e−itx dt,
F
0
regard y as fixed, then Theorem 9.13 (The Plancherel Theorem) implies that
Z ∞ Z ∞ Z ∞
1 2 2
|f (x + iy)|2 dx = fF2 (t) e2ty dt ≤ f
F2 (t) dt < ∞ (19.3)
2π −∞ 0 0
for every y < 0. Now we substitute the estimates (19.2) and (19.3) into the expression (19.1),
we see from Theorem 3.8 that
Z ∞
1 1
ϕ(y) = |f (x + iy)|2 dx
2π 2π −∞
Z ∞ Z
1 1 ∞
≤ |f1 (x + iy)|2 dx + |f1 (x + iy)| · |f2 (x + iy)| dx
2π −∞ π −∞
Z ∞
1
+ |f2 (x + iy)|2 dx
2π −∞
Z ∞ Z o1 n Z ∞
1n ∞ 2
o1
2
≤ |F1 (t)|2 dt + |f1 (x + iy)|2 dx · |f2 (x + iy)|2 dx
0 π −∞ −∞
Z ∞
2
+ fF2 (t) dt
0
2
= kF1 k22 + 2kF1 k2 ·
f F2
2 +
f F2
2
2
=
F1 k2 + kF f2
2
<∞
for all z = x + iy ∈ C. If we extend F (t) = 0 for all t ≤ −A and t ≥ A, then the integral (19.4)
becomes Z ∞
f (x + iy) = [F (t)e−ty ] · e−itx dt.
−∞
If we consider y → −∞, then the boundedness of ϕ implies that F (t) = 0 a.e. on [0, ∞).
Similarly, if y → ∞, then we have F (t) = 0 a.e. on (−∞, 0]. Thus we have F (t) = 0 a.e. on R,
so the integral representation (19.4) implies that f = 0 which ends the proof of the problem.
Problem 19.2
Rudin Chapter 19 Exercise 2.
19.1. Problems on Entire Functions of Exponential Type 567
Proof. Since f is of exponential type, there exist some constants A and C such that
for all z ∈ C. Using rotation, we may assume that one of the nonparallel lines is the real line,a
so we also have Z ∞
|f (x)|2 dx < ∞.
−∞
Now Theorem 19.3 (The Paley and Wiener Theorem) asserts that there exists an F ∈ L2 (−A, A)
such that Z A
f (z) = F (t)eitz dt
−A
for all z. Particularly, we take z = x ∈ R so that
Z A nZ A o1
1 2
|f (x)| ≤ |F (t)| dt ≤ (2A) 2 · |F (t)|2 dt <∞
−A −A
by Theorem 3.8. Thus f is bounded on the real line. Similarly, it can be shown that f is also
bounded on the other line. Then we may apply Problem 19.3 to conclude that f is actually a
constant. Since f ∈ L2 , this constant is in fact zero. Hence we obtain our desired result and
this ends the proof of the problem.
Problem 19.3
Rudin Chapter 19 Exercise 3.
Proof. Without loss of generality, we assume that the two nonparallel lines pass through the
origin and f is bounded by 1 on these lines. Suppose that ∆ is the open sector between the two
lines with sectoral angle πβ < π for some β > 1. Since f is entire, it is continuous on ∆ and
holomorphic in ∆. Since f is of exponential type, the inequality (19.5) holds for all complex z.
Take 1 < α < β. Then there exists a positive number R such that A|z| ≤ |z|α for all z with
|z| > R. For |z| ≤ R, we let M = CeAR so that
|f (z)| ≤ M exp(|z|α )
for all z ∈ ∆. Thus Problem 12.9 reveals that |f (z)| is bounded in ∆ which implies that f is a
bounded entire function. Hence it follows from Theorem 10.23 (Liouville’s Theorem) that f is
constant, as desired. This ends the proof of the problem.
Problem 19.4
Rudin Chapter 19 Exercise 4.
• The series converges if |w| > A. By the hypothesis, it is true that |f (z)| < exp(|z|λ )
for all large enough |z|, where λ > 1. According to Problem 15.2, f is of order 1. We note
from [16, Eqns. (2.1.6) & (2.2.12), pp. 8, 12] that
1
lim sup |f (n) (0)| n = A.
n→∞
a
Notice that f (eiθ z) is also an entire function satisfying the inequality (19.5).
568 Chapter 19. Holomorphic Fourier Transforms
where the term-by-term integration being justified by the uniform convergence of the series
on Γ.c
• Φ is the function which occurred in the proof of Theorem 19.3. Recall from
Remark 19.4 that the functions Φα are restrictions of a function holomorphic in the com-
plement of the interval [−iA, iA]. Thus it suffices to prove that Φ is such function.
Our first assertion and Theorem 10.6 ensure that the Borel transform Φ is holomorphic in
the complement of [−iA, iA]. It remains to show that
Φ|Πα = Φα (19.6)
Furthermore, if we define Mf (r) = max |f (z)|, then we follow from Theorem 10.26
z∈D(0;r)
(Cauchy’s Estimates) that
Mf (r)
|an | ≤
rn
and the remainder
∞
X
Rn (s) = ak s k
k=n+1
b
The function Φ(w) in question is called the Borel transform of the function f (z), see [16, §5.3, p. 73] or
[63, §20, p. 84].
c
The integral is sometimes called the Pólya representation of the function f .
d
Indeed, the half plane Πα is given by {w = x + iy | x cos α − y sin α > A}.
19.1. Problems on Entire Functions of Exponential Type 569
Problem 19.5
Rudin Chapter 19 Exercise 5.
• The Cauchy formula holds in Π+ . Let 0 < ǫ < y and z = x + iy. Let r > 0 be large
and γr be the semicircle in Π+ with radius r and centered at iǫ. Define Γr to be the union
/ Γ∗r .
of γr and the line segment [−r + iǫ, r + iǫ]. Since r is large, we may assume that z ∈
See Figure 19.1 below:
Therefore, g is bounded for almost all z ∈ U which means kgk∞ < ∞, i.e., g ∈ H ∞ . By
Theorem 11.32 (Fatou’s Theorem), the limit
lim g(reiθ ) (19.14)
r→1
exists for almost everywhere on [−π, π]. After transforming back to f , the result (19.14)
means that
f ∗ (x) = lim f (x + iy)
y→1
t
Let fn (t) = F (t)e− n +itx . Then we have |f1 | ≥ |f2 | ≥ · · · ≥ 0 and fn (t) → F (t)eitx as
n → ∞ for almost every t ∈ (0, ∞). It is clear from Theorem 3.8 that
Z ∞ Z ∞ nZ ∞ o1
2 kF k2
|f1 (t)| dt = |F (t)|e−t dt ≤ kF k2 × e−2t dt = √ < ∞,
0 0 0 2
i.e., |f1 | ∈ L1 (0, ∞). Hence we may apply Problem 1.7 to the limit (19.15) to get
Z ∞
f ∗ (x) = F (t)eitx dt (19.16)
0
for almost all real x. Since we may assume that F vanishes on (−∞, 0) (see §19.1), we
follow from the expression (19.16) that
√
f ∗ (x) = 2π · Fb(−x)
for every ξ ∈ (−∞, ∞) and 0 < ǫ < y. Thus it follows from Theorem 3.8 that
Z ∞ Z ∞
f (ξ + iǫ) |f (ξ + iǫ)|
dξ ≤ dξ
−∞ ξ + iǫ − z −∞ |ξ − x|
Z
n ∞ o1 n Z ∞
2 dξ o 12
≤ |f (ξ + iǫ)|2 dξ × 2
−∞ −∞ (ξ − x)
Z
n ∞
√ dξ o 2 1
≤ 2Cπ · 2
−∞ (x − ξ)
< ∞,
Problem 19.6
Rudin Chapter 19 Exercise 6.
Proof. Here we follow mainly [85, Theorem XII, pp. 16 – 20]. Since 0 < ϕ < eϕ , we have
log ϕ < ϕ. Combining the hypothesis and Theorem 3.5 (Hölder’s Inequality), we obtain
Z ∞ Z ∞ nZ ∞ o1 n Z ∞
log ϕ(x) ϕ(x) 2 2 dx o 21
−∞ < 2
dx < 2
dx ≤ |ϕ(x)| dx × 2 2
< ∞.
−∞ 1 + x −∞ 1 + x −∞ −∞ (1 + x )
572 Chapter 19. Holomorphic Fourier Transforms
holds for almost all x ∈ R.e Let v(z) be its harmonic conjugate and write
for x ∈ R. If we denote G(x) = Fb (−x), then we combine the results (19.18) and (19.19) to get
|G(x)| = ϕ(x).
Since G ∈ L2 (−∞, ∞), we derive from Lemma 9.3 that G(x) b = F (−x) which vanishes on [0, ∞).
b
Conversely, we suppose that there exists an f with f = ϕ such that f (x) = 0 for all x ≤ 0.
Let us write Z ∞
1
fb(x) = √ f (t)e−ixt dt (19.20)
2π −∞
and Z ∞
1
ψ(z) = √ f (t)e−izt dt, (19.21)
2π −∞
where z ∈ Π+ and the integral in (19.21) is taken along a horizontal line in the z-plane. Certainly,
we have ψ ∈ H(Π+ ) by §19.1. Suppose that we map Π+ (conformally) onto U by z = i ζ+1 ζ−1 .
Write ζ + 1
k(ζ) = ψ(z) = ψ i and K(eiθ ) = fb(x),
ζ −1
iθ
where ζ = reiθ and 0 ≤ r < 1 and x = i eeiθ −1
+1
. Then it is easily seen from Theorem 12.12 (The
Hausdorff-Young Theorem) that
Z π Z ∞ b 2 Z ∞
|f (x)|
2
|K(eiθ )|2 dθ = 2 2
dx ≤ 2 |fb(x)|2 dx = 2
fb
2 ≤ 2kf k22 = 2kϕk22 < ∞.
−π −∞ 1 + x −∞
Therefore, we have K ∈ L2 (T ). On the other hand, if z = x + iy, then the integral (19.20)
implies that
Z π Z
1 1 ∞ y
K(eiθ )Pr (θ − φ) dθ = fb(t) dt
2π −π π −∞ (x − t)2 + y 2
Z Z ∞
1 ∞ y 1
= × √ f (ξ)e−itξ dξ dt
π −∞ (x − t)2 + y 2 2π 0
Z ∞ 1 Z ∞
1 e−itξ y
=√ f (ξ) × dt dξ
2π 0 π −∞ (x − t)2 + y 2
Z ∞
1
=√ f (ξ)e−ixξ+yξ dξ
2π 0
Z ∞
1
=√ f (ξ)e−izξ dξ
2π 0
= ψ(z)
= k(reiθ ).
Substituting the inequalities (19.22) and (19.23) into the formula (19.24), we obtain
Z π Z π
1
log |k(reiθ )| dθ ≤ 1 |K(reiθ )|2 dθ − log |k(0)| (19.25)
2π −π π −π
for all 0 ≤ r < 1. If k has zero at 0 with multiplicity m, then the inequality (19.25) becomes
Z π Z π k(ζ)
1
log |k(reiθ )| dθ ≤ 1
|K(reiθ )|2 dθ − log m
− m log r. (19.26)
2π −π π −π ζ ζ=0
for all 0 ≤ r < 1. By the definition, log |k(reiθ )| → log |K(eiθ )| as r → 1 almost everywhere, so
we conclude from the inequality (19.26) that
Z Z Z π
1 ∞ | log ϕ(x)| 1 ∞ log |fb(x)| 1
log |K(eiθ )| dθ < ∞.
2
dx = 2
dx =
π −∞ 1 + x π −∞ 1 + x 2π −π
Consequently, this ensures that
Z ∞
dx
log ϕ(x) > −∞.
−∞ 1 + x2
Hence we have completed the proof of the problem.
Problem 19.7
Rudin Chapter 19 Exercise 7.
Proof. It is trivial that Condition (a) implies Condition (b). Conversely, suppose that Condition
(b) holds. For each α ∈ E, we fix a neighborhood Vα of α and put
[
Ω= Vα .
α∈E
We claim that F ∈ H(Ω) and F (z) = f (z) for z ∈ E. If z ∈ Vβ for β 6= α, then we have
Vα ∩ Vβ 6= ∅. Since Vα ∩ Vβ is an open set, there exists a δ > 0 such that z ∈ D(0; δ) ⊆ Vα ∩ Vβ ,
so Theorem 10.18 says that Fα ≡ Fβ in Vα ∩ Vβ and thus the formula (19.27) is well-defined.
19.2. Quasi-analytic Classes and Borel’s Theorem 575
Problem 19.8
Rudin Chapter 19 Exercise 8.
kD n f k∞ ≤ βf Bfn nn ≤ (M βf )(eBf )n n!
which means f ∈ C{n!}. Consequently, we obtain the desired result that C{n!} = C{nn },
completing the proof of the problem.
Problem 19.9
Rudin Chapter 19 Exercise 9.
Proof. Let M0 = 1, M1 = 1, M2 = 2 and Mn = n!(log n)n for every n ≥ 3. Thus we always have
Mn2 ≤ Mn−1 Mn+1 for all n = 1, 2, . . .. Using Theorem 19.11 (The Denjoy-Carleman Theorem),
we know that C{Mn } is quasi-analytic. If f ∈ C{n!}, then there exist positive constants βf and
Bf such that
kD n f k∞ ≤ βf Bfn n!
for all n ≥ 0. By the definition of {Mn }, we also have
kD n f k∞ ≤ βf Bfn Mn
for every n ≥ 0. Thus {mn } is a positive increasing sequence. It is clear that f ∈ C ∞ . For every
n, k ∈ N, if k ≤ n, then we see that
1 1 1 1
= × × ··· ×
mnn−k mn mn mn
576 Chapter 19. Holomorphic Fourier Transforms
1 1 1
≤ × × ··· ×
mn−1 mn−2 mk
Mn−1 Mn−2 Mk
= × × ··· ×
Mn Mn−1 Mk+1
Mk
= .
Mn
If k > n, then we have
1
= mnk−n
mnn−k
≤ mn × mn+1 × · · · × mk−1
Mn+1 Mn+2 Mk
= × × ··· ×
Mn Mn+1 Mk−1
Mk
= .
Mn
In other words, we obtain the estimate
1 Mk
≤ . (19.28)
mnn−k Mn
Define
∞
X
Mn 2mn ix
fn (x) = e and f (x) = fn (x)
(2mn )n n=0
ik Mn
fn(k) (x) = e2mn ix .
(2mn )n−k
Mnn+1
By induction, we can show that n
Mn+1 ≤ 1 for each n ≥ 0. Therefore, the estimate (19.30) gives
|f (x)| ≤ 2 = 2 · 20 M0
for every x ∈ R. Now we conclude from the estimates (19.29) and (19.30) that
kD k f k∞ ≤ 2 · 2k Mk
Mn
In addition, if k ≥ 1, then since every term (2mn )n−k
is positive for every n = 0, 1, 2, . . ., we have
∞
X Mn
|f (k) (0)| = |ik | ≥ Mk . (19.31)
(2mn )n−k
n=0
Hence we have obtained what we want. If f ∈ C{n!}, then it must be true that
for some positive constants βf and Bf . However, if k is sufficiently large so that (log k)k ≥ βf Bfk ,
then this will certainly contradict the estimate (19.31). Hence we conclude that f ∈ / C{n!} and
then we complete the proof of the problem.
Problem 19.10
Rudin Chapter 19 Exercise 10.
∞
X
Proof. Suppose that λ = λn is positive finite. Recall from Definition 2.9 that Cc (R) is the
n=1
collection of all continuous complex functions on R whose support is compact. Now we let g0
to be the function modified from the (19.38) in such the way that g0 (x) = 1 for −λ ≤ x ≤ λ,
g0 (x) = 0 for |x| ≥ 2λ and 0 ≤ g0 (x) ≤ 1 if x ∈ [−2λ, −λ] ∪ [λ, 2λ]. Then g0 ∈ Cc (R) and g0 is
integrable in R. Write
gn (x) = g λ1 , λ2 , . . . , λn ; g0 (x)
Z λ1 Z λ2 Z λn
1
= n dt1 dt2 · · · g0 (x − t1 − t2 − · · · − tn ) dtn . (19.32)
2 λ1 λ2 · · · λn −λ1 −λ2 −λn
Since |g0 (x)| ≤ 1 for every x ∈ R, the definition (19.32) ensures that |gn (x)| ≤ 1 for every
n = 0, 1, 2, . . . and x ∈ R. In other words, the family {gn } is (uniform) bounded in R. Besides,
if |x| ≥ 3λ, then |x − λ1 − λ2 − · · · − λn | ≥ 2λ so that gn (x) = 0 there.
Obviously, we have
g λ1 , λ2 , . . . , λn ; g0 (x) = g λ1 , λ2 , . . . , λk ; g λk+1 , λk+2 , . . . , λn ; g0 (x) . (19.33)
1
g1′ (x) = [g0 (x + λ1 ) − g0 (x − λ1 )].
2λ1
Next, we assume that the function gn−1 (x) is continuous for any n ≥2 in R and if n ≥ 2, then
the function gn (x) = g λn ; g(λ1 , λ2 , . . . , λn−1 ; g0 (x) = g λn ; gn−1 (x) is differentiable in R and
it follows from the formula (19.33) that
d
gn′ (x) = g λ1 , λ2 , . . . , λn ; g0 (x)
dx
578 Chapter 19. Holomorphic Fourier Transforms
d
= g λ1 ; g λ2 , λ3 , . . . , λn ; g0 (x)
dx
1
= g λ2 , λ3 , . . . , λn ; g0 (x + λ1 ) − g λ2 , λ3 , . . . , λn ; g0 (x − λ1 ) (19.34)
2λ1
1
= g λ2 , λ3 , . . . , λn ; g0 (x + λ1 ) − g0 (x − λ1 ) .
2λ1
This also implies that gn (x) has continuous derivatives of order 0, 1, . . . , n−1 in R. Furthermore,
if we combine the formula (19.34) and the Mean Value Theorem for Derivatives, we get, for every
x ∈ R, that
|gn′ (x)| ≤ max |g′ λ2 , λ3 , . . . , λn ; g0 (x) | = max |gn−1
′
(x)| ≤ · · · ≤ max |g2′ (x)| < ∞. (19.35)
x∈R x∈R x∈R
which shows that the family {gn } is equicontinuous on R. Recall that gn (x) = 0 outside [−3λ, 3λ],
so {gn } is actually equicontinuous on [−3λ, 3λ]. The fact |gn (x)| ≤ 1 in R guarantees that {gn }
converges pointwise on R. Hence it asserts from [99, Exercise 16, p. 168] that {gn } converges
uniformly to a continuous function g on [−3λ, 3λ], i.e.,
for every x ∈ [−3λ, 3λ]. Since the argument also applies to any compact interval of R, the
function g is also continuous at the end points ±3λ. It is trivial that if x ∈ / [−3λ, 3λ], then
g(x) = 0. Hence we must have g(±3λ) = 0.
Denote g(x) = g λ1 , λ2 , . . . ; g0 (x) . Then we know that
g(x) = lim g λ1 , λ2 , . . . , λn ; g0 (x)
n→∞
Z λ1
1
= lim g λ2 , λ3 , . . . , λn ; g0 (x − t) dt
n→∞ 2λ1 −λ
1
Z λ1
1
= g λ2 , λ3 , . . . ; g0 (x − t) dt
2λ1 −λ1
Put
G0 (x) = g0 (x)
G0 (x + λ1 ) − G0 (x − λ1 )
G1 (x) = ,
2λ1
..
.
Gn−1 (x + λn ) − Gn−1 (x − λn )
Gn (x) = .
2λn
Thus the formula (19.36) can be written as
g′ (x) = g λ2 , λ3 , . . . ; G1 (x) .
In fact, it is true that
g (n) (x) = g λn+1 , λn+2 , . . . ; Gn (x) (19.37)
for every n = 0, 1, 2, . . . and x ∈ R. Recall that |g0 (x)| ≤ 1 on R, so we have
1
|Gn (x)| ≤ = Mn .
λ1 λ2 · · · λn
on R. Thus it follows from the formula (19.37) that
|g(n) (x)| ≤ Mn
for all n = 0, 1, 2, . . . and x ∈ R. Consequently, g ∈ C{Mn } which completes the proof of the
problem.
Remark 19.1
The construction in Problem 19.10 follows basically the unpublished work of H. E. Bray
which was quited in Mandelbrojt’s article [68, pp. 79 – 84]. See also [55].
Problem 19.11
Rudin Chapter 19 Exercise 11.
Proof. An example of a function ϕ ∈ C ∞ with the required properties can be found in [124,
Problem 10.6, pp. 266, 267]. In fact, we start with
−1
e x , if x > 0;
g(x) =
0, if x ≤ 0.
Problem 19.12
Rudin Chapter 19 Exercise 12.
αn
Proof. Let ϕ be as in Problem 19.11. Set β = n! and gn (x) = βn xn ϕ(x). Take
gn (λn x)
fn (x) = = βn xn ϕ(λn x),
λnn
k
X n−m
n! k k−m 2
|(D k fn )(x)| ≤ |βn |M Cm λn · n−m
(n − m)! λn
m=0
Xk
2n (n!)2
≤ |βn |M
m=0
λnn−k
n2n (n!)2 |βn |M
≤ (19.40)
λn
for all x ∈ R and k = 0, 1, . . . , n − 1. Since λn can be chosen large enough, we observe from the
estimate (19.40) that
1
kD k fn k∞ < n (19.41)
2
for all k = 0, 1, . . . , n − 1. Take f = f0 + f1 + · · · .
It is clear that f0 (0) + f1 (0) + · · · = α0 . Besides, the result (19.41) ensures that the series
{f0′ + f1′ + · · · + fn′ } converges uniformly on R. Using [99, Theorem 7.17, p. 152], termwise
differentiation is legitimate so that f ′ = f0′ + f1′ + · · · . Now this argument can be applied
repeatedly to show that f ∈ C ∞ . Next, it follows from the expression (19.39) that (D k fn )(0) = 0
for k = 0, 1, . . . , n − 1. Since ϕ(x) = 1 on [−1, 1], ϕ(n) (0) = 0 for all n = 1, 2, . . . and this implies
that (D n fm )(0) = 0 for m = 0, 1, . . . , n − 1. Hence we have
(D n f )(0) = (D n f0 )(0) + (D n f1 )(0) + · · · + (D n fn−1 )(0) + (D n fn )(0)
n
+ (D fn+1 )(0) + (D n fn+2 )(0) + · · ·
= n!βn
= αn
Remark 19.2
Problem 19.12 is called Borel’s Theorem which says that every power series is the Taylor
series of some smooth function, see, for examples, [79, Theorem 1.5.4, p. 30] and [86].
Problem 19.13
Rudin Chapter 19 Exercise 13.
k−1
X
ck λkk = λk > 2 cj λkj = 2(λk1 + λ2k−1 + · · · + λ2k−1 ) and λk > k2k > 1.
j=1
We put
∞
X
f (z) = ck eiλk x ,
k=1
where z = x + iy. Let fk (x) = ck eiλk x , where k ∈ N. For each n = 0, 1, 2, . . ., we observe that
for every a ∈ R. Using similar argument as in the proof of Problem 19.12, it is easy to show
that f ∈ C ∞ . The choices of our {ck } and {λk } reveal that
X∞ X∞ n2n
|(D n f )(a)| ≥ |cn λnn | − ck λnk = λn − ck λnk > .
2
k=1 k=1
k6=n k6=n
has radius of convergence 0 for every a ∈ R, completing the proof of the problem.
582 Chapter 19. Holomorphic Fourier Transforms
Remark 19.3
Let S = {2n | n ∈ N}. Define X √
g(x) = e− k
cos(kx).
k∈S
Then it can be shown that g also satisfies the requirements of Problem 19.13.
Problem 19.14
Rudin Chapter 19 Exercise 14.
Proof. Suppose that f ∈ C{Mn } has infinitely many zeros {xn } in [0, 1]. Then {xn } has a
convergent subsequence by the Bolzano-Weierstrass Theoremg . Without loss of generality, we
may assume that {xn } is itself convergent, distinct, increasing and its limit is α. The continuity
of f gives
f ′ (α) = 0.
By the Mean Value Theorem for Derivatives, we see that f ′ (ξn ) = 0 for some ξn ∈ (xn , xn+1 )
for all n = 1, 2, . . .. The fact xn → α as n → ∞ ensures that ξn → α as n → ∞. Since f ∈ C ∞ ,
the continuity of f ′ implies that
This argument can be repeated to show that f (n) (α) = 0 for all n = 0, 1, 2, . . .. Since C{Mn }
is quasi-analytic, Definition 19.8 gives f (x) ≡ 0 for all x ∈ R, completing the proof of the
problem.
Problem 19.15
Rudin Chapter 19 Exercise 15.
For any α, β ∈ C and f, g ∈ X, it is clear that |f (z)| ≤ C1 eπ|z| and |g(z)| ≤ C2 eπ|z| for some
positive constants C1 and C2 . Therefore, we have |αf (z) + βg(z)| ≤ (|α|C1 + |β|C2 )eπ|z| and
Φ(αf + βg) = {αf (n) + βg(n)} = α{f (n)} + β{g(n)} = αΦ(f ) + βΦ(g)
Using the analysis of §19.2, one can show that F is entire, |F (z)| ≤ Ceπ|z| for all z ∈ C and
F ∈ L2 (−∞, ∞). In other words, this means that F ∈ X. Furthermore, we note that F = fb.
Next, for each n ∈ Z, recall from Definition 4.23 that {un (t) = eint | n ∈ Z} forms an orthonormal
set in L2 (T ). Furthermore, we have
Z π
1 1 ei(n−z)π − e−i(n−z)π sin[(z − n)π]
Un (z) = ubn (z) = ei(n−z)t dt = × = (19.44)
2π −π 2π i(n − z) (z − n)π
holds for every n ∈ Z. Recall from [100, Example 4.5(b), p. 78] that
Z π
1
hf, giT = f (t)g(t) dt
2π −π
defines an inner product in L2 (T ). Then it is easily checked that we can induce a norm k · k to
X by defining p p
kF kX = hF, F iX = hf, f iT = kf kT , (19.45)
so it makes X to be Hilbert. Of course, it follows from the expression (19.45) that
1, if n = m;
hUn , Um iX = hun , um iT =
0, otherwise.
for every n ∈ Z. Finally, as a Hilbert space X with a maximal orthonormal set {Un | n ∈ Z}, we
conclude from §4.19 that the mapping
F 7→ hF, Un i = cn = F (n)
is a Hilbert space isomorphism of X onto ℓ2 (Z). This means that our map Φ is a bijection,
completing the analysis of the problem.
584 Chapter 19. Holomorphic Fourier Transforms
Problem 19.16
Rudin Chapter 19 Exercise 16.
Proof. Since |f (x)| ≤ e−|x| on R, f ∈ L2 (−∞, ∞). By the analysis of §19.1, its Fourier transform
Z ∞
fb(z) = f (t)eitz dt
−∞
for every x ∈ R. For every n ≥ 0, the hypothesis implies that differentiation under the integral
sign is legitimateh so that
Z ∞ Z ∞
|fb(n) (x)| = f (t)(it)n eitx dt ≤ e−|t| · |t|n dt = 2n!.
−∞ −∞
has at least 1 as its radius of convergence for every a ∈ R which means that fb is also holomorphic
on R. If fb has compact support, then fb vanishes on a set with a limit point. Hence Theorem
10.18 forces that f ≡ 0 a.e. on R. This completes the proof of the problem.
h
Of course, it follows from the Leibniz’s Rule by Problem 10.16, where ϕ(z, t) = f (t)eitz . See also [3,
Theorem 24.5, pp. 193, 194].
CHAPTER 20
Uniform Approximation by Polynomials
Problem 20.1
Rudin Chapter 20 Exercise 1.
Proof. We want to prove that if ǫ > 0, S 2 \ K has finitely many components, f ∈ C(K) and
f ∈ H(K ◦ ), then there exists a rational function R such that
for all z ∈ K.
Indeed, everything up to [100, p. 392] in the proof of Theorem 20.5 (Mergelyan’s Theorem)
remains the same. Let S1 , S2 , . . . , Sm be the (connected) components of S 2 \ K, i.e.,
S 2 \ K = S1 ∪ S2 ∪ · · · ∪ Sm .
Pick δ > 0 very small. Recall also that X = {z ∈ supp Φ | dist(z, S 2 \ K) ≤ δ} is compact, so X
contains no point which is “far within” K, see Figure 20.1 which shows that X is exactly the
yellow part.
Now we can cover X by finitely many open discs D1 (p1 ; 2δ), D2 (p2 ; 2δ), . . . , Dn (pn ; 2δ), where
p1 , p2 , . . . , pn are points in S 2 \ K. We may assume that n = m and pj ∈ Sj for j = 1, 2, . . . , n.
585
586 Chapter 20. Uniform Approximation by Polynomials
Since each Sj is connected, there must be a curve from pj to a point of ∂Dj (pj ; 2δ) that is not of
K. In other words, one can find a set Ej ⊂ Dj (pj ; 2δ) such that Ej is a compact and connected
subset of Sj , diam Ej ≥ 2δ, S 2 \ Ej is connected and K ∩ Ej = ∅, where j = 1, 2, . . . , n. We
apply Lemma 20.2 with r = 2δ and follow the proof in [100, pp. 393, 394], we can obtain
|F (z) − Φ(z)| < 6000ω(z) and |f (z) − Φ(z)| < ω(δ) (20.2)
|f (z) − R(z)| ≤ |f (z) − Φ(z)| + |Φ(z) − F (z)| + |F (z) − R(z)| < 10000ω(δ)
for all z ∈ K. Since ω(δ) → 0 as δ → 0, it yields the inequality (20.1) by choosing sufficiently
small δ. Hence we have completed the proof of the problem.
Problem 20.2
Rudin Chapter 20 Exercise 2.
Proof. The set K is known as a Swiss cheese set.a The construction of such a sequence {Dn }
in U with the specific properties can be found in [45, pp. 344, 345]. In fact, we can also assume
that
X∞
rn2 < 1. (20.4)
n=1
• There exists an f ∈ C(K) for which L(f ) 6= 0. We take f (z) = z which belongs to
C(K). Obviously, we have
Z Z
z dz = 1 and z dz = 2πirn2
Γ γn
Problem 20.3
Rudin Chapter 20 Exercise 3.
Proof. Suppose that E ⊆ D(0; r) is compact and connected, where r > 0. Let diam E ≥ r and
Ω = S 2 \ E be connected. Denote X = {f ∈ H(Ω) | zf (z) → 1 as z → ∞}. Now we recall the
definitions of the conformal mappings F : U → Ω and g : U → D(0; |a|−1 ) that
∞
a X 1 −1
F (ω) = + cn ω n and g(z) = F (z), (20.5)
ω n=0 a
where ω ∈ U and z ∈ Ω. Without loss of generality, we may assume that a > 0. Assume that
there was an f ∈ X such that
kgk∞ > kf k∞ . (20.6)
Since F −1 is a conformal mapping of Ω onto U , we have kF −1 k∞ = 1 and thus the definition
(20.5) gives
kgk∞ = a−1 . (20.7)
These two facts (20.6) and (20.7) combine to give f (Ω) ⊆ D(0; a−1 ). Next, we define the
mapping ϕ : U → U by
ϕ(ω) = af F (ω) .
Then it is easily checked that ϕ ∈ H ∞ and ϕ(0) = af F (0) = af (∞) = 0. Besides, we observe
that
ϕ′ (ω) = af ′ F (ω) · F ′ (ω).
By the definition (20.5), we have
X∞
a
F ′ (ω) = − + ncn ω n−1 .
ω 2 n=1
so that
∞
−1 X −na−n
f ′ (z) = + .
z2 n=0
z n+1
588 Chapter 20. Uniform Approximation by Polynomials
Hence it follows from Theorem 12.2 (The Schwarz Lemma) that ϕ(ω) = λω for some constant
λ with |λ| = 1 so that af F (ω) = λω. Substituting ω = F −1 (z) into this equation, we obtain
F −1 (z)
f (z) = λ = λg(z)
a
which implies that kf k∞ = kgk∞ , a contradiction to the inequality (20.6).
Put ω = F −1 (z). Then the definitions (20.5) imply
∞
X X∞
a n 1
z = F F −1 (z) = + c0 + cn F −1 (z) = + c0 + cn an gn (z).
F −1 (z) n=1
g(z) n=1
Rewrite it as
∞
X
zg(z) = 1 + c0 g(z) + cn an g n+1 (z). (20.8)
n=1
Since Z
1
b= zg(z) dz,
2πi Γ
where Γ is the positively oriented circle with center 0 and radius r, we may substitute the formula
(20.8) into the integral to get
Z h ∞
X i
1
b= 1 + c0 g(z) + cn an gn+1 (z) dz.
2πi Γ n=1
Since g ∈ X, g has a simple zero at ∞ which shows that Res (g; ∞) = 1 and Res gn+1 (z); ∞ = 0
for all n ≥ 1. Hence we conclude from Theorem 10.42 (The Residue Theorem) that
Z ∞
X h 1 Z i
c0
b= g(z) dz + cn an · gn+1 (z) dz = c0
2πi Γ 2πi Γ
n=1
589
590 Index
M T
Mean Value Theorem, 116 tent function, 59
Minkowski’s Inequality, 256 The Basic Connectedness Lemma, 356
mollifiers, 295 The Steinhaus Theorem, 215, 310
multipler of the transformation, 453 total variation of f , 546
totally disconnected, 509
N triangle inequality, 14, 43
normal form, 456
U
O upper densities of E at 0, 210
order topology, 47 Urysohn’s Lemma, 22
Osgood’s Theorem, 410 usual metric, 44
outer measure, 54
V
P Vitali Convergence Theorem, 410
Pólya representation, 568
parabolic transformation, 456 W
Poisson kernel, 296, 572 weak convergence, 371
Poisson kernel for the upper half-plane, 359 weak∗ convergence, 371
power set, 2 Weierstrass Approximation Theorem, 125,
Principle of Subordination, 422 139
Product Measure Theorem, 270 Weyl’s Equidistribution Theorem, 135
push-forward measure, 264
Y
R Young’s Convolution Inequality, 256
Radon measure, 59
Removable sets for holomorphic functions, Z
363 Zorn’s Lemma, 311
Bibliography
[1] L. Ahlfors, Bounded analytic functions, Duke Math. J., Vol. 14, pp. 1 – 11, 1947.
[2] L. Ahlfors, Complex Analysis, 3rd ed., Mc-Graw Hill Inc., 1979.
[3] C. D. Aliprantis and O. Burkinshaw, Principles of Real Analysis, 3rd ed., San Diego: Aca-
demic Press, 1998.
[5] T. M. Apostol, Mathematical Analysis, 2nd ed., Addison-Wesley Publishing Company, 1974.
[6] T. M. Apostol, Modular Functions and Dirichlet Series in Number Theory, 2nd ed., N. Y.:
Springer-Verlag, 1990.
[7] M. Artin, Algebra, 2nd ed., Boston: Pearson Prentice Hall, 2011.
[8] K. Astala, T. Iwaniec, G. Martin and J. Onninen, Schottky’s theorem on conformal map-
pings between annuli: a play of derivatives and integrals, Complex analysis and dynamical
systems III, 35 – 39, Contemp. Math., 455, Israel Math. Conf. Proc., Providence, R. I.:
Amer. Math. Soc., 2008.
[9] S. Axler, P. Bourdon and W. Ramey, Harmonic Function Theory, 2nd ed., N. Y.: Springer,
2001.
[10] F. Bagemihl and W. Seidel, Sequential and continuous limits of meromorphic functions,
Ann. Acad. Sei. Fenn. Ser. Al., Vol. 280, pp. 1 – 17, 1960.
[11] J. Bak and D. J. Newman, Complex Analysis, 3rd ed., N. Y.: Springer, 2010.
[17] R. P. Boas, A Primer of Real Functions, 4th ed., Washington, D. C.: The Mathematical
Association of America Inc., 1996.
[19] V. I. Bogachev, Measure Theory Volume II, Berlin; N. Y.: Springer, 2007.
591
592 Bibliography
[20] P. Borwein and Tamás Erdélyi, Polynomials and Polynomial Inequalities, N. Y.: Springer-
Verlag, 1995.
[21] P. Borwein and Tamás Erdélyi, The full Müntz Theorem in C[0, 1] and L1 [0, 1], J. Lond.
Math. Soc., Vol. 54, No. 2, pp. 102 – 110, 1996.
[22] H. Brezis, Functional Analysis, Sobolev Spaces and Partial Differential Equations, N. Y.:
Springer, 2011.
[24] M. R. Burke, Weakly dense subsets of the measure algebra, Proc. Amer. Math. Soc., Vol
106, No. 4, pp. 867 – 874, 1989.
[25] F. S. Cater, A partition of the unit interval, Amer. Math. Monthly, Vol. 91, No. 9, pp. 564
– 566, 1984.
[26] I. L. Chang, On the zeros of power series with Hadamard gaps-distribution in sectors, Trans.
Amer. Math. Soc., Vol. 178, pp. 393 – 400, 1973.
[27] R. Cignoli and J. Hounie, Functions with arbitrarily small periods, Amer. Math. Monthly,
Vol. 85, No. 7, pp. 582 – 584, 1978.
[28] D. L. Cohn, Measure Theory, 2nd ed., New York, N. Y.: Birkhäuser, 2013.
[29] P. Colwell, Blaschke products: Bounded Analytic Functions, Ann Arbor: University of
Michigan Press, 1985.
[30] J. B. Conway, Fucntions of One Complex Variable, 2nd ed., N. Y.: Springer-Verlag, 1978.
[31] J. Dieudonné, Un exemple d’un espace normal non susceptible d’une structure uniforme
d’espace complet, C. R. Acad. Sci. Paris, 209, pp. 145 – 147, 1939.
[32] E. P. Dolženko, On the removal of singularities of analytic functions, Uspehi Mat Nauk,
Vol. 18, No. 4 (112), pp. 135 – 142, 1963; English transl., Amer. Math. Soc. Transl., Vol.
97, pp. 33 – 41, 1970.
[33] M. P. Drazin, The invariant circles of a bilinear transformation, Math. Gaz., Vol. 38, pp.
26 – 29, 1954.
[35] C. Eckhardt, Free products and the lack of state-preserving approximations of nuclear C ∗ -
algebras, Proc. Amer. Math. Soc., Vol. 141, No. 8, pp. 2719 – 2727, 2013.
[39] G. B. Folland, Fourier Analysis and its Applications, Providence, R. I.: American Mathe-
matical Society, 2009.
Bibliography 593
[40] G. B. Folland, Real Analysis: Modern Techniques and Their Applications, 2nd ed., N. Y.:
Wiley, 1999.
[42] J. B. Fraleigh, A First Course in Abstract Algebra, 7th ed., Boston: Addison-Wesley, 2003.
[43] X. Fu and C. K. Lai, Translational absolute continuity and Fourier frames on a sum of
singular measures, J. Funct. Anal., Vol. 274, No. 9, pp. 2477 – 2498, 2018.
[44] W. H. J. Fuchs, On the zeros of power series with Hadamard gaps, Nagoya Math. J., Vol.
29, pp. 167 – 174, 1967.
[47] I. S. Gradshteyn and I. M. Ryzhik, Table of Integrals, Series, and Products, 7th ed., Aca-
demic Press, 2007.
[48] G. H. Hardy, A Course of Pure Mathematics, 10th ed., Cambridge: Cambridge University
Press, 2002.
[52] W. K. Hayman and E. F. Lingham, Research Problems in Function Theory: Fiftieth An-
niversary Edition, Cham: Springer International Publishing, 2019.
[53] F. Herzog and G. Piranian, On the univalence of functions whose derivative has a positive
real part, Proc. Amer. Math. Soc., Vol. 2, pp. 625 – 633, 1951.
[54] K. Hoffman, Banach Spaces of Analytic Fucntions, N. Y.: Dover Publications, 1988.
[55] R. B. Hughes, Zero sets of functions from non-quasi-analytic classes, Proc. Amer. Math.
Soc., Vol. 27, No. 3, pp. 539 – 542, 1971
n
X xr
[56] K. S. K. Iyengar, A note on the zeros of = 0, The Mathematics Student, Vol. 6, pp.
r!
r=0
77, 78, 1938.
[57] R. B. Kirk, Sets which split families of measurable sets, Amer. Math. Monthly, Vol. 79, No.
8, pp. 884 – 886, 1972.
[58] F. Jones, Lebesgue Integration on Euclidean Space, Rev. ed., Boston: Jones and Bartlett,
2001.
[59] R. Kannan and C. K. Krueger, Advanced Analysis on the Real Line, New York: Springer,
1996.
[60] A. W. Knapp, Basic Real Analysis, Boston, M. A.: Birkhäuser Boston, 2005.
[62] A. N. Kolmogorov and S. V. Fomin, Elements of the Theory of Functions and Functional
Analysis, Mineola, N. Y. Dover Publications, 1999.
[63] B. Ja. Levin, Distribution of Zeros of Entire Functions, Revised d., Providence, R. I.: Amer.
Math. Soc., 1980.
[64] J. W. Lewin, A Truly Elementary Approach to the Bounded Convergence Theorem, Amer.
Math. Monthly, Vol. 93, No. 5, pp. 395 – 397, 1986.
[65] I. H. Lin, Classical Complex Analysis: A Geometric Approach - Vol. 2, Singapore: World
Scientific Publishing Company Co. Pte. Ltd., 2010.
[66] W. A. Luxemburg, Arzelà’s Dominated Convergence Theorem for the Tiemann Integral,
Amer. Math. Monthly, Vol. 78, No. 9, pp. 970 – 979, 1971.
[67] B. Magajna, C ∗ -convex sets and completely positive maps, Integr. Equ. Oper. Theory, Vol.
85, No. 1, pp. 37 – 62, 2016.
[68] S. Mandelbrojt, Analytic functions and classes of infinitely differentiable functions, Rice
Inst. Pamphlet 29, Houston, Texas, 1942.
[69] M. Marden, Geometry of Polynomials, 2nd ed., Mathematical Surveys, No. 3, Providence,
R. I.: Amer. Math. Soc., 1966.
[70] R. E. Megginson, An Introduction to Banach Space Theory, New York; Hong Kong:
Springer, 1998.
[71] C. D. Meyer, Matrix Analysis and Applied Linear Algebra, Philadelphia: SIAM, 2000.
[73] H. Mirkil, Differentiable functions, formal power series, and moments, Proc. Amer. Math.
Soc., Vol. 7, pp. 650 – 652, 1956.
[74] J. R. Munkres, Topology, 2nd ed., Upper Saddle River, N. J.: Prentice-Hall, 2000.
[75] T. Murai, The value distribution of lacunary series and a conjecture of Paley, Ann. Inst.
Fourier (Grenoble), Vol. 31, pp. 136 – 156, 1981.
[77] G. Myerson, First-class Functions, Amer. Math. Monthly, Vol. 98, No. 3, pp. 237 – 240,
1991.
[79] R. Narasimhan, Analysis on Real and Complex Manifolds, Paris: Masson, Amsterdam:
North-Holland Pub. Co., 1968.
[80] P. N. Natarajan, An Introduction to Ultrametric Summability Theory, 2nd ed., New Delhi:
Springer, 2015.
[82] W. P. Novinger, Mean convergence in Lp spaces, Proc. Amer. Math. Soc., Vol. 34, No. 2,
pp. 627 – 628, 1972.
[83] W. F. Osgood, Note on the functions defined by infinite series whose terms are analytic
functions, Ann. Math., Vol. 3, pp. 25 – 34, 1901.
[84] A. Ostrowski, Mathematische Miszellen XV. Zur konformen Abbildung einfachzusamm en-
hiingender Gebiete, Jahresbericht Deut. Math., Vereiningung 38, pp. 168 – 182, 1929.
[85] R. C. Paley and N. Wiener, Fourier Transforms in the Complex Domain, N. Y.: Amer.
Math. Soc. Colloquium Publ. 19, 1934.
[86] H.-J. Petzsche, On E. Borel’s Theorem, Math. Ann., Vol. 282, pp. 299 – 313, 1988.
[87] D. R. Pitts and V. Zarikian, Unique pseudo-expectations for C ∗ -inclusions, Illinois J. Math.,
Vol. 59, No. 2, pp. 449 – 483, 2015.
[88] G. Pólya and G. Szegö, Problems and Theorems in Analysis II: Theory of Functions. Zeros.
Polynomials. Determinants. Number Theory. Geometry, Berlin: Springer-Verlag, 1998.
[89] I. I. Privalov, Randeigenschaften Analytischer Funktionen, VEB Deutscher Verlag der Wis-
senschaften, Berlin, 1956.
[90] R. E. Rodrı́guez, I. Kra and J. P. Gilman, Complex Analysis: In the Spirit of Lipman Bers,
2nd ed., N. Y.: Springer-Verlag, 2013.
[91] M. Rosenblum and J. Rovnyak, Topics in Hardy Classes and Univalent Functions, Basel;
Boston: Birkhäuser, 1994.
[92] A. Rosenthal, On functions with infinitely many derivatives, Proc. Amer. Math. Soc., Vol.
4, pp. 600 – 602, 1953.
[93] H. L. Royden and P. M. Fitzpatrick, Real Analysis, 4th ed., Boston: Prentice Hall, 2010.
[95] L. A. Rubel, Conformal Inequivalence of Annuli and the First-Order Theory of Subgroups
of P SL(2, R), Proc. Amer. Math. Soc., Vol. 88, No. 4, pp. 679 – 683, 1983.
[96] W. Rudin, Functional Analysis, 2nd ed., Mc-Graw Hill Inc., 1991.
[97] W. Rudin, Function Theory in the Unit Ball of Cn , N. Y.: Springer-Verlag, 1980.
[98] W. Rudin, Fourier Analysis on Groups, Mineola, N. Y.: Dover Publications, Inc., 2017.
[99] W. Rudin, Principles of Mathematical Analysis, 3rd ed., Mc-Graw Hill Inc., 1976.
[100] W. Rudin, Real and Complex Analysis, 3rd ed., Mc-Graw Hill Inc., 1987.
[101] W. Rudin, Well-distributed measurable sets, Amer. Math. Monthly, Vol. 90, No. 1, pp. 41
– 42, 1983.
[102] S. Saks, Integration in Abstract Metric Spaces, Duke Math. J., Vol. 4, No. 2, pp. 408 –
411, 1938.
[103] D. Sarason, Complex Function Theory, 2nd ed., Providence, R. I.: Amer. Math. Soc.,
2007.
596 Bibliography
[106] Nadish de Silva, A Concise, Elementary Proof of Arzelà’s Bounded Convergence Theorem,
Amer. Math. Monthly, Vol. 117, No. 10, pp. 918 – 920, 2000.
[108] E. M. Stein and G. L. Weiss, Introduction to Fourier Analysis on Euclidean Spaces, Prince-
ton, N. J.: Princeton University Press, 1971.
[109] E. M. Stein and R. Shakarchi, Complex Analysis, Princeton, N. J.: Princeton University
Press, 2003.
[110] E. M. Stein and R. Shakarchi, Fourier Analysis: an Introduction, Princeton, N. J.: Prince-
ton University Press, 2003.
[111] E. M. Stein and R. Shakarchi, Real Analysis: Measure Theorey, Integration and Hilbert
Spaces, Princeton, N. J.: Princeton University Press, 2005.
[112] K. Stephenson, Isometries of the Nevanlinna Class, Indiana Univ. Math. J., Vol. 26, No.
2, 307 – 324, 1977.
[113] R. S. Strichartz, The Way of Analysis, revised ed., Boston, Mass.: Jones and Bartlett,
2000.
[114] V. Thilliez, On quasianalytic local rings, Expo. Math., Vol. 26, pp. 1 – 28, 2008.
[115] B. S. Thomson, Strong Derivatives and Integrals, Real Anal. Exchange, Vol. 39, No. 2, pp.
469 – 488, 2013/14.
[116] S. R. Tims, A theorem on functions schlicht in convex domains, Pro. Lond. Math. Soc.,
Vol 3, pp. 200 – 205, 1951.
[117] E. C. Titchmarsh, The Theory of Functions, Oxford University Press, N. Y., 1960.
[119] N. X. Uy, Removable sets of analytic functions satisfying a Lipschitz condition, Ark. Mat.,
Vol. 17, No. 1 – 2, pp. 19 – 27, 1979.
[120] A. Villani, Another Note on the Inclusion of Lp (µ) ⊂ Lq (µ), Amer. Math. Monthly, Vol.
92, No. 7, pp. 485 – 487, 1985.
[121] M. Weiss and G. Weiss, On the Picard property of lacunary power series, Studia Math.,
Vol. 22, pp. 221 – 245, 1963.
[122] R. L. Wheeden and A. Zygmund, Measure and Integral: An Introduction to Real Analysis,
2nd ed., Boca Raton, F. L.: CRC Press, 2015.
[125] K. W. Yu, A Complete Solution Guide to Real and Complex Analysis I, Amazon.com,
2019.
[126] K. W. Yu, A Complete Solution Guide to Real and Complex Analysis II, Amazon.com,
2021.
[127] K. W. Yu, Problems and Solutions for Undergraduate Real Analysis I, Amazon.com, 2018.
[128] K. W. Yu, Problems and Solutions for Undergraduate Real Analysis II, Amazon.com, 2019.
[129] W. P. Ziemer, Modern Real Analysis, 2nd ed., Cham: Springer, 2017.
[130] W. Żelazki, Banach Algebras, Warszawa: Elsevier Pub. Comp. and Polish Sc. Pub., 1973.
[132] V. A. Zorich, Mathematical Analysis II, 2nd ed., Berlin: Springer, 2016.
[133] A. Zygmund, Trigonometric Series, 3rd ed., Cambridge: Cambridge University Press,
2002.