Kit-Wing Yu - A Complete Solution Guide To Real and Complex Analysis II-978-988-74156-5-7 (2021)

A Complete Solution Guide to Real
and Complex Analysis II
by Kit-Wing Yu, PhD
[email protected]
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.
ISBN: 978-988-74156-4-0 (eBook)

ISBN: 978-988-74156-5-7 (Paperback)
ii
About 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 was 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 seven books:
• A Complete Solution Guide to Complex Analysis
• A Complete Solution Guide to Real and Complex Analysis I
• A Complete Solution Guide to Principles of Mathematical Analysis
• Problems and Solutions for Undergraduate Real Analysis
• Problems and Solutions for Undergraduate Real Analysis I
• Problems and Solutions for Undergraduate Real Analysis II
• Mock Tests for the ACT Mathematics
iii
iv
Preface
This is the continuum of my book A Complete Solution Guide to Real and Complex Analysis I.
It covers the “Complex Analysis” part of Rudin’s graduate book. In fact, we study all exercises
of Chapters 10 to 20.
Same as A Complete Solution Guide to Real and Complex Analysis I, the primary aim of this
book is to help every mathematics student and instructor to understand the ideas and applica-
tions of the theorems in Rudin’s book. To accomplish this goal, I have adopted the way I wrote
the solution guides of Baby Rudin and the first part of Papa Rudin. 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 read 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 refer to some other analysis
or topology books such as [2], [9], [15], [18], [23], [42] and [65]. Other useful references are [3],
[27], [28], [37], [69], [79], [80], [82] and [83]. Of course, we will use the exercises in Baby Rudin
and the first part of Papa Rudin freely and if you want to read proofs of them, you are strongly
advised to read my books [77] and [78].
As you will expect, this book always keeps the main features of my previous books [77] and
[78]. In fact, its features are as follows:
• It covers all the 221 exercises from Chapters 10 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 29 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.
v
vi
• Many useful or relevant references are provided to some questions for your future research.
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
[email protected].
Then I will post the updated errata on my website
https://sites.google.com/view/yukitwing/
irregularly.
Kit Wing Yu
April 2021
List of Figures
10.1 The closed contour ΓA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

10.2 The contour ΓA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
10.3 The contours ΓA , Γ1 and Γ2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
10.4 The annulus A(r1 , r2 ) and the circles γ1 , γ2 . . . . . . . . . . . . . . . . . . . . . . 23
10.5 A non null-homotopic closed path Γ = γ1 − γ3 − γ2 + γ4 in Ω. . . . . . . . . . . . 32
11.1 The I divides ∂∆ into several triangles. . . . . . . . . . . . . . . . . . . . . . . . 48
12.1 The boundary ∂∆. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

12.2 The sectors ∆1 , ∆2 and the ray Lα . . . . . . . . . . . . . . . . . . . . . . . . . . 76
13.1 The simply connected set Ω. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

13.2 The compact sets Dn and En . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
13.3 The compact sets An , Bn and Cn . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
13.4 The disc ∆n , the arc Ln and its neighborhood Ωn . . . . . . . . . . . . . . . . . . 87
14.1 The region Ω bounded by C1 and C2 . . . . . . . . . . . . . . . . . . . . . . . . . 119

14.2 The constructions of Ωn−1 , D(0; rn ) and αn . . . . . . . . . . . . . . . . . . . . . . 122
14.3 The construction of the symmetric point z ∗ of z. . . . . . . . . . . . . . . . . . . 134

14.4 The conformal mapping ψ(z) = φ ϕ(z) . . . . . . . . . . . . . . . . . . . . . . . 143
π π
14.5 The conformal mapping f : U → A(e− 2 , e 2 ). . . . . . . . . . . . . . . . . . . . . 144
14.6 The locus of f (z) for t ∈ (0, π). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
14.7 The locus of f (z) for t ∈ (π, 2π). . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
14.8 The image of f (E) when r = 0.25. . . . . . . . . . . . . . . . . . . . . . . . . . . 146
15.1 The distribution of the zeros zk,n of exp(exp(z)). . . . . . . . . . . . . . . . . . . 151

15.2 The paths γ(z + h) and −γ(z). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
16.1 The paths βζ (I) and γζ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

16.2 The paths γz (I), γω (I) and γω,ζ (I). . . . . . . . . . . . . . . . . . . . . . . . . . . 182
16.3 The fundamental domain R of G. . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
16.4 The regions Ωα and Ωβ if α < β. . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
16.5 The regions of convergence of the two series. . . . . . . . . . . . . . . . . . . . . . 199
19.1 The closed contour Γr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
vii
viii List of Figures
20.1 The compact set X. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

Contents
Preface v
List of Figures viii
10 Elementary Properties of Holomorphic Functions 1

10.1 Basic Properties of Holomorphic Functions . . . . . . . . . . . . . . . . . . . . . 1
10.2 Evaluation of Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
10.3 Composition of Holomorphic Functions and Morera’s Theorem . . . . . . . . . . 13
10.4 Problems related to Zeros of Holomorphic Functions . . . . . . . . . . . . . . . . 17
10.5 Laurent Series and its Applications . . . . . . . . . . . . . . . . . . . . . . . . . . 22
10.6 Miscellaneous Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
11 Harmonic Functions 33
11.1 Basic Properties of Harmonic Functions . . . . . . . . . . . . . . . . . . . . . . . 33
11.2 Harnack’s Inequalities and Positive Harmonic Functions . . . . . . . . . . . . . . 49
11.3 The Weak∗ Convergence and Radial Limits of Holomorphic Functions . . . . . . 57
12 The Maximum Modulus Principle 69

12.1 Applications of the Maximum Modulus Principle . . . . . . . . . . . . . . . . . . 69
12.2 Asymptotic Values of Entire Functions . . . . . . . . . . . . . . . . . . . . . . . . 78
12.3 Further Applications of the Maximum Modulus Principle . . . . . . . . . . . . . 79
13 Approximations by Rational Functions 83

13.1 Meromorphic Functions on S2 and Applications of Runge’s Theorem . . . . . . . 83
13.2 Holomorphic Functions in the Unit Disc without Radial Limits . . . . . . . . . . 87
13.3 Simply Connectedness and Miscellaneous Problems . . . . . . . . . . . . . . . . . 93
14 Conformal Mapping 99
14.1 Basic Properties of Conformal Mappings . . . . . . . . . . . . . . . . . . . . . . . 99
14.2 Problems on Normal Families and the Class S . . . . . . . . . . . . . . . . . . . 112
14.3 Proofs of Conformal Equivalence between Annuli . . . . . . . . . . . . . . . . . . 118
14.4 Constructive Proof of the Riemann Mapping Theorem . . . . . . . . . . . . . . . 122
15 Zeros of Holomorphic Functions 149
ix
x Contents
15.1 Infinite Products and the Order of Growth of an Entire Function . . . . . . . . . 149
15.2 Some Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
15.3 Problems on Blaschke Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
15.4 Miscellaneous Problems and the Müntz-Szasz Theorem . . . . . . . . . . . . . . . 172
16 Analytic Continuation 179

16.1 Singular Points and Continuation along Curves . . . . . . . . . . . . . . . . . . . 179
16.2 Problems on the Modular Group and Removable Sets . . . . . . . . . . . . . . . 183
17 H p -Spaces 205
17.1 Problems on Subharmonicity and Harmonic Majoriants . . . . . . . . . . . . . . 205
17.2 Basic Properties of H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
17.3 Factorization of f ∈ H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
17.4 A Projection of Lp onto H p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
18 Elementary Theory of Banach Algebras 235

18.1 Examples of Banach Spaces and Spectrums . . . . . . . . . . . . . . . . . . . . . 235
18.2 Properties of Ideals and Homomorphisms . . . . . . . . . . . . . . . . . . . . . . 239
18.3 The Commutative Banach algebra H ∞ . . . . . . . . . . . . . . . . . . . . . . . . 249
19 Holomorphic Fourier Transforms 251

19.1 Problems on Entire Functions of Exponential Type . . . . . . . . . . . . . . . . . 251
19.2 Quasi-analytic Classes and Borel’s Theorem . . . . . . . . . . . . . . . . . . . . . 260
20 Uniform Approximation by Polynomials 271
Index 275
Bibliography 277
CHAPTER 10
Elementary Properties of Holomorphic Func-
tions
10.1 Basic Properties of Holomorphic Functions
Problem 10.1
Rudin Chapter 10 Exercise 1.
Proof. In fact, this is [61, Exercise 21, p. 101]. For a solution of it, please refer to [77, p. 75].
This completes the proof of the problem.
Problem 10.2
Proof. Let a ∈ C and

∞
X
f (z) = cn (z − a)n . (10.1)
n=0
By [62, 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
1
2 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
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
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
γ(t) = a + 2δeit (10.4)
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
for all t ∈ [0, 2π].

Next, for every n ∈ N, we define Fn : [0, 2π] → C and F : [0, 2π] → C by
Fn (t) = fn (γ(t)) and F (t) = f (γ(t))
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 3
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
/ γ ∗ , so Ind γ (z) = 1. Then we obtain from Theorem 10.15 (The Cauchy’s

Recall that z ∈
Formula in a Convex Set) and the limit (10.7) thata
f (z) = lim fn (z)

n→∞
Z
1 fn (ζ)
= lim dζ
n→∞ 2πi γ ζ − z
Z 2π
1 γ ′ (t)
= lim fn (γ(t)) · dt
n→∞ 2πi 0 γ(t) − z
Z 2π
1 γ ′ (t)
= lim Fn (t) · dt
n→∞ 2πi 0 γ(t) − z
Z 2π
1 γ ′ (t)
= F (t) · dt. (10.8)
2πi 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
Z 2π Z 2π
1 γ ′ (t) 1 γ ′ (t)
|fn (z) − f (z)| = Fn (t) · dt − F (t) · dt
2πi 0 γ(t) − z 2πi 0 γ(t) − z
Z 2π
1 γ ′ (t)
≤ |Fn (t) − F (t)| · dt
2π 0 γ(t) − z
<ǫ
for all n ≥ N . In other words, {fn } converges uniformly to f on D(a; δ).

Finally, every compact K ⊆ Ω can be covered by D(a1 , δ), D(a2 , δ), . . . , D(am , δ) for some
m ∈ N. Thus the previous analysis shows immediately that for all z ∈ D(ak , δ), the inequality
|fn (z) − f (z)| < ǫ (10.9)

a
We remark that we cannot apply Theorem 10.15 (The Cauchy’s Formula in a Convex Set) directly to f to
obtain the representation (10.8) because it is not clear whether f ∈ H(Ω) or not.
holds if n ≥ Nk , where k = 1, 2, . . . , m. Set N = max(N1 , N2 , . . . , Nm ). Therefore, if n ≥ N ,

then the inequality (10.9) remains valid on D(a1 , δ) ∪ D(a2 , δ) ∪ · · · ∪ D(am , δ), so particularly
on K. Hence we have shown that the convergence is in fact uniform on every compact subset
of Ω. This completes the proof of the problem.
Problem 10.6
Proof. We divide the proof into several parts.
• 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
zeF (z) F ′ (z) − eF (z)

g ′ (z) = =0
z2
for all z ∈ D(1; 1). Using [9, Exercise 5, p. 42], we know that g is a constant in D(1; 1).
Obviously, we have g(1) = eF (1) = 1, so it is true that g(z) = 1 in D(1; 1). Hence it follows
from the form (10.10) that
eF (z) = z (10.11)
for all z ∈ D(1; 1).
• The injection of exp. Set Ω = F (D(1; 1)). We claim that Ω is a region such that
exp(Ω) = D(1; 1). (10.12)
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.
• The number of such region Ω. For each n ∈ Z, we define Ωn = {z + 2nπi | z ∈ Ω}

which is also a region and satisfies
exp(Ωn ) = exp(Ω + 2nπi) = exp(Ω) = D(1; 1).
Since F is continuous on D(1; 1), Ω must be bounded. Therefore, there exists an N ∈ N

such that ΩN 6= Ω. Hence {ΩkN | k ∈ Z} is a set of distinct regions satisfying the set
equality (10.12).
• The derivative of log z. Define log z, for |z − 1| < 1, such that
elog z = z. (10.13)
By differentiating the equation (10.13), we get log′ (z)elog z = 1 and so

1 1
log′ (z) = = .
elog z z
10.1. Basic Properties of Holomorphic Functions 5
• The coefficients an and cn . It is easy to see that

X ∞ X ∞
1 1
= = (1 − z)n = (−1)n (z − 1)n ,
z 1 − (1 − z) n=0 n=0
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 = .
z
n=1 n=0
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=1
n
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.
We have completed the analysis of the problem.
Problem 10.7
b
Read [9, Theorem 2.9, pp. 28, 32] for details.
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) k! f (ζ)
f (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) = k+1
− . (10.16)
(ζ − z) (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) dζ. (10.17)
Γ (ζ − z) Γ (ζ − z)k+2
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.
10.2 Evaluation of Integrals
Problem 10.8
Proof. We note that

Z ∞ Z A
P (x)
R(x) dx = lim dx. (10.18)
−∞ A→∞ −A Q(x)
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} positively oriented, see Figure 10.1
10.2. Evaluation of Integrals 7
Figure 10.1: The closed contour ΓA .
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 [62, 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, [9, pp. 129, 130] or [65, pp. 75, 76].
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
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 [9, pp. 144 – 146], we see that
Z ∞ Z X
ix
R(x)e 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 + y2
eiy dy.
−∞ 1 + x −∞ t
t
Set R(z) = t2 +z 2 . Then it follows from the representation (10.23) that
Z ∞ t
eitx iz π
2
dx = 2πiRes 2 2
e ; ti = t. (10.24)
−∞ 1+x t +z e
Next, if t = −u for some u > 0, then we have

Z ∞ Z ∞ −iux
eitx e
2
dx = 2
dx
−∞ 1 + x −∞ 1 + x
Z ∞
u
= 2 2
eiy dy
−∞ u + y
u
iz
= 2πiRes 2 e ; ui
u + z2
π
= u. (10.25)
e
Combining the two expressions (10.24) and (10.25), we conclude that
Z ∞
eitx π
2
dx = |t| . (10.26)
−∞ 1+x e
pπ −|t| ,
Using the theory of Fourier transforms, we notice that if f (t) = 2e then we know from
Definition 9.1 that
Z ∞
1
fb(x) = √ f (t)e−ixt dt
2π −∞
Z ∞
1
= e−|t| e−ixt dt
2 −∞
Z Z
1h 0 ∞ i
= e(1−ix)t 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 b ixt 1 eixt
g(t) = √ f (x)e 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
Proof. Let f (z) = (ez − e−z )z −4 . Then f is holomorphic in C \ {0}. By the power series
expansion of ez (see [62, 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)!
Problem 10.11
6 1, we have either |α| < 1 or |α−1 | < 1. If z = eiθ , then cos θ = 21 (z + z1 ).

Proof. Since |α| =
Using [9, Eqn. (5), p. 150], we see that
Z 2π Z
dθ 1 dz
2
= 1 1 2
·
0 1 − 2α cos θ + α |z|=1 1 − 2α · 2 (z + z ) + α iz
Z
dz
=i 2 2
|z|=1 αz − (1 + α )z + α
 1

 −2πRes ; α , if |α| < 1;

 αz 2 − (1 + α2 )z + α
=

 1

 −2πRes ; α−1
, if |α−1 | < 1
αz 2 − (1 + α2 )z + α

 2π

 − α2 − 1 , if |α| < 1;

=

 2π

 , if |α−1 | < 1.
2
α −1
Hence we complete the analysis of the problem.
Problem 10.12
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.
Figure 10.2: The contour ΓA .
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 = e dx + e dx + e dx. (10.28)
−A x −A x2 −1 x
2
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
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
Next, we write 2i sin z = eiz − e−iz so that

Z Z
sin2 z itz e2iz − 2 + e−2iz itz
e dz = e dz. (10.31)
ΓA z2 ΓA −4z 2
Now we define Z
eisz
ϕA (s) = dz, (10.32)
ΓA z2
so the expression (10.31) becomes
Z
sin2 z itz 1 1
2
e dz = − [ϕA (t + 2) + ϕA (t − 2)] + ϕA (t). (10.33)
ΓA z 4 2
If we combine (10.30) and (10.33), then we have

Z A
sin2 x itx 1 1
2
e dx = ϕA (t) − [ϕA (t + 2) + ϕA (t − 2)]. (10.34)
−A x 2 4
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.
Figure 10.3: The contours ΓA , Γ1 and Γ2 .

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 ∞ 2  π, if t = 0;
sin 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
Proof. This has been solved on [76, p. 143] which completes the proof of the problem.
10.3. Composition of Holomorphic Functions and Morera’s Theorem 13
10.3 Composition of Holomorphic Functions and Morera’s Theorem
Problem 10.14
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 [61,
Exercise 26, p. 102].
Problem 10.15
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
g(z) = f (ϕ(z)) = [ϕ(z) − ϕ(z0 )]m h(ϕ(z)). (10.40)
Suppose that n ≥ 1 is the order of the zero of ϕ(z) − ϕ(z0 ). Now Theorem 10.18 implies that
ϕ(z) − ϕ(z0 ) = (z − z0 )n φ(z),
where φ ∈ H(Ω) and φ(z0 ) 6= 0. Assume that n ≥ 2. We follow from
ϕ′ (z) = (z − z0 )n−1 [(z − z0 )φ′ (z) + φ(z)] (10.41)
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 .
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
g(z) = (z − z0 )m(k+1) · φm (z)h(ϕ(z)).
In conclusion, g has a zero of order m(k + 1) at z0 . This completes the proof of the problem.
Problem 10.16
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
Z 2π
ϕ(z, p) − ϕ(ω, p) 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 2π
Mǫ
≤ 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
lim gn (x) = ϕ′ (z0 , x) · h(x).

n→∞
10.3. Composition of Holomorphic Functions and Morera’s Theorem 15
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→∞ zn − z0
ZX
= 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
Problem 10.17
Proof. We apply Problem 10.16 to the functions one by one.
• The function f (z). We write

Z
f (z) = ϕ(z, t) dm,
X
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).
• The function g(z). We have

Z
g(z) = ϕ(z, t) dm,
X
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.
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
i.e., g is continuous at z0 . Since z0 is arbitrary, g is continuous in Ω.

Next, we suppose that ∆ ⊆ Ω is a closed triangle and ∂∆ is parameterized by a piecewise
continuously differentiable curve γ : [a, b] → Ω. We have
Z Z b Z bZ ∞
′
g(z) dz = g(γ(x))γ (x) dx = ϕ(γ(x), t)γ ′ (x) dt dx. (10.46)
∂∆ a a 0
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,
∂∆
so the integral (10.47) reduces to

Z
g(z) dz = 0.
∂∆
Finally, we apply Theorem 10.17 (Morera’s Theorem) to obtain the desired conclusion that
g ∈ H(Ω).
• The function h(z). We have

Z
h(z) = ϕ(z, t) dm,
X
10.4. Problems related to Zeros of Holomorphic Functions 17
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.
We complete the proof of the problem.
10.4 Problems related to Zeros of Holomorphic Functions
Problem 10.18
Proof. By [76, Problem 10.11, p. 131], we know that

Z ′ X
1 f (z) p
z dz = mk zkp ,
2πi γ f (z)
k
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 [9, Eqn. (1), p. 129] that
Res (F ; zk ) = lim (z − zk )F (z) = mk ϕ(zk ). (10.49)

z→zk
Substituting the residues (10.49) into the formula (10.48), we get

Z ′ X
1 f (z)
ϕ(z) dz = mk ϕ(zk ),
2πi γ f (z)
k
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
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 [9, 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
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 [9, Theorem
10.13, p. 139] or [18, p. 152].
Problem 10.21
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
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
1 = |f (0)| ≥ min |f (reiθ )| > 2,

θ
a contradiction. Hence f has at least one zero in the unit disc, completing the proof of the
problem.
Problem 10.23
Proof. Here we list some observations about the zeros of Pn and Qn .
• 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.
• Since Qn (0) = 0, Qn always has a zero at 0 for every n.
• By [9, 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 [33] that every zero ζ of Pn lie in the annulus
n
< |ζ| < n
e2
for large enough n.
• Applying [13, Corollary 1.2.3, p. 13] to Qn , all the zeros z 6= 0 of
z2 zn z z n−1
Qn (z) = z + + ··· + = z 1 + + ··· +
2! n! 2! n!
lie inside the annulus
2= min (k + 2) ≤ |z| < max (k + 2) = n.

0≤k≤n−2 0≤k≤n−2
• We imitate the proof of [76, 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
|ez − 1 − Qn | = |ez − Pn | < ǫk . (10.51)
Thus the inequalities (10.50) and (10.51) give
|ez − 1 − Qn | < |ez − 1|
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)
outside C(0; Rk ) for all n ≥ Mk .

Similarly, since 2(k + 1)π < Rk + 2π < 2(k + 2)π, Theorem 10.43(b) (Rouché’s Theorem)
tells us that
NQn = Nez −1 = 2k + 3
inside C(0; Rk + 2π) for all n ≥ Mk+1 , or equivalently,
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,
[13, Chap. 1], [39] and [40, Chap. 3].
Problem 10.24
Proof. We have Ω = K ◦ which is an open set in C. Put
ϕ(z) = |f (z) − g(z)| − |f (z)| − |g(z)|,
E = {z ∈ K | ϕ(z) = 0} and {zn } ⊆ E with zn → z0 as n → ∞. We divide the proof into several

steps.
• Step 1: E ⊆ Ω. Our hypothesis implies that
|f (z) − g(z)| < |f (z)| + |g(z)| (10.54)
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
|f (z) − g(z)| < |f (z)| + |g(z)| (10.55)
in K \ E.
• Step 2: The numbers of zeros of f and g are finite. Suppose that
Z(f ) = {a ∈ Ω | f (a) = 0}.
Assume that Z(f ) was infinite. Then [79, 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 [63, IX.8 & 9, pp. 115 – 118] or [67, 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
for every f ∈ H(G) and z ∈ K.
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
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
Next, we apply Lemma 10.1 with f ≡ 1 and z = ak to get

Z
dζ
2πi = . (10.59)
Γ ζ − ak
By combining the integrals (10.58) and (10.59), we conclude that

Z X N
1 f ′ (ζ)
dζ = pk = |Z(f )|. (10.60)
2πi Γ f (ζ)
k=1
Similarly, we obtain
Z
1 g′ (ζ)
dζ = |Z(g)|. (10.61)
2πi Γ g(ζ)
By the strict inequality (10.55), it is impossible that f = −N ′ g for some N ′ ∈ N ∪ {0} on

Ω \ E. Therefore, the function h : Ω \ E → C \ (−∞, 0] given by
f (z)
h(z) = log
g(z)
is well-defined and holomorphic in the open set Ω \ E. Taking differentiation, we have
f ′ g′
h′ = −
f g
on Ω \ E. Since Z(f ), Z(g) * Ω \ E, h′ is continuous in Ω \ E. Since Γ is a cycle in Ω \ E,

Theorem 10.12 guarantees that Z
h′ (ζ) dζ = 0
Γ
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)|.
Now we complete the proof of the problem.
10.5 Laurent Series and its Applications
Problem 10.25
10.5. Laurent Series and its Applications 23
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 ,
where r > 0. Particularly, we have γ1 = −ρr1 +ǫ and γ2 = ρr2 −ǫ .
(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 [62, 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
Ind Γ (z) = Ind γ1 (z) + Ind γ2 (z) = 1.
See Figure 10.4 for an illustration below.
Figure 10.4: The annulus A(r1 , r2 ) and the circles γ1 , γ2 .
Hence, Theorem 10.35 (Cauchy’s Theorem) and [62, Eqn. (5), p. 217] assert that
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
Then the formula (10.64) simplifies to f = f1 + f2 .
– 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
Ind Γ1 (α) = Ind ρR1 (α) − Ind ρR′ (α) = 0

1
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
In other words, f1 (z) is uniquely determined by z, not by R1 . Since ǫ is arbitrary,

this means that f1 is actually well-defined in C \ D(0; r1 ).
Similarly, we take |z| < R2 < R2′ < r2 so that z ∈ D(0; R2 ) ⊆ D(0; R2′ ). Let
Γ2 = ρR2 − ρR′2 . By Theorem 10.11 again, we have
Ind Γ2 (α) = Ind ρR2 (α) − Ind ρR′ (α) = 0

2
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
is representable by power series in D(0; R11 ). Furthermore, for ρ 1 (t) = 1 it

R1 e , we
R1
have
Z Z 2π
1 dµ(ζ) 1 i it dt
= R1 e−it f (R1 e−it ) e · 1 it
2πi ρ 1
ζ −z 2πi 0 R1 R1 e−z
R1
g
In fact, we have applied the formula [62, Eqn. (2), p. 219] here.
Z 2π
1 R1 e−it dt
= if (R1 e−it ) ·
2πi 0 ( R11 eit − 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.
(c) – Existence of a Laurent series. On γ2 , we have |ζ| > |z| so that
1 1 1 z z2
= z = + 2 + 2 + ··· . (10.67)
ζ −z ζ(1 − ζ ) ζ ζ ζ
On −γ1 , since |ζ| < |z|, we have
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 ,
−∞
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)
dz = an z n−k−1 dz = 2πiak ,
γ z k+1 −∞ 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
n n
|cn z | ≤ |cn z | + |cn z n |
−∞ n=−∞ n=0
X∞ ∞
X
|cn |
≤ + |cn |(r2 − δ)n
|z|n
n=1 n=0
X∞ X ∞
|cn |
≤ + |cn |(r2 − δ)n
(r1 + δ)n
n=1 n=0
< ∞.
Hence it follows from the Weierstrass M -Test [9, Theorem 1.9, p. 15] or [61, 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
f1 = f − f2 on D(0; s1 ) \ D(0; r1 ). Clearly, we know that D(0; s1 ) \ D(0; r1 ) ⊆ D(0; s2 ), so

f2 is bounded there and thus f1 is also bounded there. Using the second assertion in the
previous paragraph, we see that f1 is bounded in C \ D(0; r1 ).
(e) All the foregoing parts remain valid if r1 = 0, or r2 = ∞ or both. In the case r2 = ∞,
we note that f2 represents an entire function and thus it reduces to 0 by Theorem 10.23
(Liouville’s Theorem).
(f) Suppose that 0 < r1 < r2 < r3 < · · · < rn−1 < rn < ∞. Each A(rk , rk+1 ) is an annulus,
n−1
[
where k = 1, 2, . . . , n − 1. Let A = A(rk , rk+1 ) and f ∈ H(A). Then the foregoing
k=1
results can be applied to f in each A(rk , rk+1 ) and obtain the corresponding Laurent series
in each A(rk , rk+1 ).
Hence we have completed the analysis of the problem.
Problem 10.26
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}.
Note that if |z| < 1, then we have

X ∞
1
= z 2n . (10.72)
1 − z2
n=0
If |z| > 1, then | 1z | < 1 so that

X 1 ∞
1 1 1
2
=− 2 · −2
=− . (10.73)
1−z z 1−z z 2n
n=1
Similarly, if |z| < 3, then we have | z3 | < 1 so that

∞
1 1 1 1 X zn
= · z = (10.74)
3−z 3 1− 3 3 n=0 3n
and if |z| > 3, then | 3z | < 1 which gives

∞
1 1 1 1 X 3n
=− · 3 =− . (10.75)
3−z z 1− z
z zn
n=0
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 .
3 3n 2 3n+1
n=0 n=0 n=0
Thus we have 
 1 + (−1)n 1
 + n+1 , if n = 0, 1, 2, . . .;
cn = 2 3


0, otherwise.
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
Problem 10.27
Proof. Suppose that

1 i
ζ = e2πiz or z= arg ζ − log |ζ|, (10.76)
2π 2π
where 0 < arg ζ < 2π. Set F (ζ) = f (z). We claim that F (ζ) does not depend on the choice of
arg ζ. In fact, it is easy to see from the second equation (10.76) that
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 .
Proof of Lemma 10.2. We consider
S = {z ∈ Ω2 | every ω ∈ Ω1 with f (ω) = z implies f ′ (ω) = 0}. (10.77)
Let β ∈ Ω2 \ S. Since f (Ω1 ) = Ω2 , there exists an α ∈ Ω1 such that f (α) = β. By

the definition (10.77), f ′ (α) 6= 0. Using Theorem 10.30, there exist neighborhoods
V ⊆ Ω1 and W ⊆ Ω2 of α and β respectively such that f : V → W is a bijection and a
holomorphic inverse f −1 : W → V exists. Thus we have g = g ◦ (f ◦ f −1 ) = h ◦ f −1 on
W which guarantees that g is holomorphic in W and thus g is holomorphic in Ω2 \ S.
Next, we want to show that every point b of S is isolated. This means that one can
find a neighborhood W ⊆ Ω2 of b such that W ∩ S = {b}. By the hypothesis, one can
find an a ∈ Ω1 such that f (a) = b. By the definition (10.77), we have f ′ (a) = 0. Since
f is nonconstant, f ′ is nonzero. Since f ′ ∈ H(Ω1 ), Theorem 10.18 implies that Z(f ′ )
has no limit point in Ω1 . In other words, Ω1 contains a neighborhood U of a such that
f ′ (z) 6= 0 (10.78)
for all z ∈ U \ {a}. By the Open Mapping Theorem, f (U ) ⊆ Ω2 is our wanted

neighborhood because if f (U ) contains a c ∈ S \ {b}, then we have f (ζ) = c for some
ζ ∈ U \ {a} but the definition (10.77) shows that f ′ (ζ) = 0 which contradicts the result
(10.78).
Given ǫ > 0. Since h is continuous at a, there exists a δ > 0 such that ω ∈ D(a; δ)
implies
h(ω) ∈ D(h(a); ǫ). (10.79)
Since f is holomorphic in Ω1 , f (D(a; δ)) is open in C by the Open Mapping Theorem.
Since b ∈ f (D(a; δ)), there is a δ′ > 0 such that D(b; δ′ ) ⊆ f (D(a; δ)). Obviously, if
z ∈ D(b; δ′ ), then we have f (ω) = z for some ω ∈ D(a; δ). Combining this fact with
the property (10.79), we conclude that
g(z) = g(f (ω)) = h(ω) ∈ D(h(a); ǫ) = D(g(f (a)); ǫ) = D(g(b); ǫ).
Consequently, g is continuous on S. According to the Riemann’s Principle of Remov-

able Singularities [9, Theorem 9.3, p. 118], every point of S is a removable singularity
of g. Hence g can be extended to a function holomorphic in Ω2 , as required.
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)
−∞
in A and this implies

∞
X
f (z) = F (ζ) = cn e2nπiz . (10.81)
−∞
Finally, we denote
Ω′ = {z ∈ C | a + ǫ ≤ Im z ≤ b − ǫ} and A′ = {ζ ∈ C | e−2π(b−ǫ) ≤ |ζ| ≤ e−2π(a+ǫ) }.
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.
10.6 Miscellaneous Problems
Problem 10.28
Proof. If we consider γ = Γ − α, then we observe immediately from [62, 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 [61,
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 [61, Exercise 26, p. 202]h , we conclude that
Ind Γn (0) = Ind Γm (0)
for all m, n ≥ N .
• The result is independent of the choice of {Γn }. Note that [77, Problem 8.26, pp.
204 – 206] also includes this part.
• Lemma 10.39 is true for closed curves. Again, [77, 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
H(s, 0) = Γ0 (s), H(s, 1) = Γ1 (s) and H(0, t) = H(1, t).
Put Γt (s) = H(s, t). According to [62, Eqn. (3) & (4), p. 223], we obtain
|Γt (s)| > 2ǫ (10.82)
for all (s, t) ∈ I 2 and there exists a positive integer n such that
|Γt (s) − Γt′ (s)| < ǫ (10.83)
for all t, t′ ∈ I with |t − t′ | ≤ n1 and all s ∈ I. Let t0 = 0, tn = 1 and tk = tk−1 + n1 , where

k = 1, 2, . . . , n. Then the set {t0 , t1 , . . . , tn } forms a partition of I and tk − tk−1 = n1 for
each k = 1, 2, . . . , n. Therefore, it follows from the inequalities (10.82) and (10.83) that
|Γtk (s) − Γtk−1 (s)| < ǫ < |Γtk−1 (s)|

h
The author has proven it in [77, Problem 8.26, pp. 204 – 206].
10.6. Miscellaneous Problems 31
for every s ∈ I, where k = 1, 2, . . . , n. By the improved version of Lemma 10.39, we see

that
Ind Γtk (0) = Ind Γtk−1 (0),
where k = 1, 2, . . . , n. Consequently, we have the desired result that
Ind Γ0 (0) = Ind Γt0 (0) = Ind Γt1 (0) = · · · = Ind Γtn (0) = Ind Γ1 (0).
Problem 10.29
Proof. Suppose that z 6= 0, so we can write

Z 1Z π
1 r
f (z) = dθ dr
π 0 −π reiθ+z
Z 1Z π
1 re−iθ
= dθ dr
π 0 −π r + ze−iθ
Z 1Z π r −iθ
1 ze
= r −iθ
dθ dr
π 0 −π z + e
Z 1 h Z π i
1 r −ieiθ
=− r
−iθ − (− )
dθ dr
πi 0 z −π e z
Z 1 h Z π i
r 1 −ieiθ
= −2 r
−iθ − (− )
dθ dr
0 z 2πi −π e z
Z 1 −r
r
= −2 · Ind γ dr (10.84)
0 z z
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.
If |z| < 1, then the expression (10.84) reduces to
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.
Next, if |z| ≥ 1, then we always have Ind γ (− zr ) = −1 so that

Z 1
2 1
f (z) = r dr = .
z 0 z
Finally, if z = 0, then it is easy to see that

Z Z
1 1 π −iθ
f (0) = e dθ dr = 0.
π 0 −π
Problem 10.30
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
Figure 10.5: A non null-homotopic closed path Γ = γ1 − γ3 − γ2 + γ4 in Ω.
Γ = γ1 − γ3 − γ2 + γ4 is a closed curve in Ω and we observe easily that
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
11.1 Basic Properties of Harmonic Functions
Problem 11.1
Proof. Let u, v : Ω → R. We prove the assertions one by one.
• uv is harmonic if and only if u + icv ∈ H(Ω) for some c ∈ R. Direct computation

gives
∆(uv) = (uv)xx + (uv)yy

= (uvxx + 2ux vx + uxx v) + (uvyy + 2uy vy + uyy v)
= 2(ux vx + uy vy )
= 2(ux , uy ) · (vx , vy ).
Thus uv is harmonic if and only if
(ux , uy ) · (vx , vy ) = 0. (11.1)
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,
ux = c(x, y)vy and uy = −c(x, y)vx (11.2)
in Ω. On the one hand, since ∆u = 0 in Ω, we obtain
cx vy − cy vx = 0 (11.3)
in Ω. On the other hand, uxy − uyx = 0 and ∆v = 0 imply that
cy vy + cvyy − (−cx vx − cvxx ) = 0
33
34 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
(c2x + c2y )vx = 0 and (c2x + c2y )vy = 0 (11.5)
in Ω. Since v is harmonic in Ω, the function f = vx − ivy is holomorphic in Ω by Theorem

11.2. If Z(f ) = Ω, then vx = vy = 0 in Ω so that v is constant in Ω, a contradiction.
By Theorem 10.18, Z(f ) has no limit point in Ω. Pick z ∈ Ω \ Z(f ). Then there exists
a δ > 0 such that D(z; δ) ∩ Z(f ) = ∅. By the equations (11.5), we have (c2x + c2y )f = 0
which implies that c2x + c2y = 0 in D(z; δ). Since z is arbitrary, we conclude that
c2x + c2y = 0 (11.6)
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.
• u2 cannot be harmonic in Ω unless u is constant. This part is shown in [76, Problem

16.3, pp. 199, 200].
• |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
∆(u2 + v 2 ) = (u2 + v 2 )xx + (u2 + v 2 )yy

= (2uux + 2vvx )x + (2uuy + 2vvy )y
= (2uuxx + 2u2x + 2vvxx + 2vx2 ) + (2uuyy + 2u2y + 2vvyy + 2vy2 )
= 2(u2x + vx2 + u2y + vy2 ).
Therefore, ∆(u2 + v 2 ) = 0 in Ω if and only if ux = uy = vx = vy = 0 in Ω if and only if

both u(x, y) and v(x, y) are constant in Ω. In other words, |f |2 is harmonic in Ω if and
only if f is constant in Ω.
Problem 11.2
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
u2x + u2y = vx2 + vy2 and ux vx + uy vy = 0
in Ω respectively. Consequently, they show that
(ux + ivx )2 + (uy + ivy )2 = 0

11.1. Basic Properties of Harmonic Functions 35
(ux + ivx )2 − [i(uy + ivy )]2 = 0

[ux + ivx + i(uy + ivy )][ux + ivx − i(uy + ivy )] = 0
(ux − vy + ivx + iuy )(ux + vy + ivx − iuy ) = 0
holds in Ω.
Next, the equation (ux − ivx )2 + (uy − ivy )2 = 0 gives
(ux − vy − ivx − iuy )(ux + vy − ivx + iuy ) = 0 (11.7)
in Ω. We denote
g = ux − vy − i(vx + uy ) and h = ux + vy + i(uy − vx ). (11.8)
It is obvious that
(ux − vy )x = uxx − vyx = −uyy − vxy = −(uy + vx )y
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 (i): g ≡ 0 in Ω. By the definition, g ≡ 0 implies that ux − vy = i(vx + uy ) in Ω.

Since ux , uy , vx and vy are real functions, we get ux = vy and uy = −vx in Ω. By Theorem
11.2, we conclude that f ∈ H(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
Proof. Let u be a real function. Suppose that
V = {z ∈ Ω | grad u = 0} = {z ∈ Ω | ux (z) = uy (z) = 0} and f = ux − iuy .
Since ux and uy are continuous in Ω, V is closed in Ω. By Theorem 11.2, f ∈ H(Ω). It is trivial

that V = Z(f ). By Theorem 10.18, either V = Ω or V has no limit point in Ω.a This ends the
analysis of the problem.
a
Note that the case V = Ω implies that u is constant in Ω.
Problem 11.4
Proof. We prove the assertions one by one.
• Every partial derivative of a harmonic is harmonic. See [76, Problem 16.2, p. 199].
• 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
[62, Eqn. (2), §11.5, p. 233] can be written as P A = 1 − r 2 . Direct differentiation gives
Pr 2[r − cos(θ − t)] 2

=− P− ,
r rA A
4[r − cos(θ − t)] 2 2
Prr = − Pr − P − , (11.10)
A A A
Pθθ 8 sin2 (θ − t) 2 cos(θ − t)
= P− P.
r2 A2 rA
Put these equations (11.10) into the Laplacian equation (11.9) to get
4[r − cos(θ − t)] 2 2 2[r − cos(θ − t)] 2

∆P = − Pr − P − − P−
A A A rA A
8 sin2 (θ − t) 2 cos(θ − t)
+ P− P
A2 rA
8[r − cos(θ − t)]2 8r[r − cos(θ − t)] 4 4 8 sin2 (θ − t)
= P + − P − + P
A2 A2 A A A2
8[r − cos(θ − t)]2 − 4A + 8 sin2 (θ − t) 8r[r − cos(θ − t)] − 4A
= 2
P+
A A2
2
8[r − 2r cos(θ − t) + 1] − 4A 2 2
4[r + r − 2r cos(θ − t)] − 4A
= 2
P+
A A2
4P 2
4(r + A − 1) − 4A
= +
A A2
2
4(1 − r ) 4(r 2 − 1)
= +
A2 A2
= 0.
Hence Pr (θ − t) is a harmonic function of reiθ for a fixed t.
• P [ dµ] is harmonic in U . Suppose that µ is a finite Borel measure on T and u = P [ dµ].

It suffices to show that
Z Z Z
∂ ∂
ur = Pr (θ − t) dµ(eit ) = Pr (θ − t) dµ(eit ) = [Pr (θ − t)]r dµ(eit ) (11.11)
∂r T T ∂r T
and
Z Z Z
∂ it ∂
uθ = Pr (θ − t) dµ(e ) = Pr (θ − t) dµ(eit ) = [Pr (θ − t)]θ dµ(eit ) (11.12)
∂θ T T ∂θ T
because they certainly yield that

Z Z
urr = [Pr (θ − t)]rr dµ(e ) and uθθ = [Pr (θ − t)]θθ dµ(eit )
it
T T
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
= · −
h 1 − 2(r + h) cos(θ − t) + (r + h)2 1 − 2r cos(θ − t) + r 2
[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
Pr+h (θ − t) − Pr (θ − t) 2 + 2(2 + h) + 2(1 + h) 4(8 + 4h) 40

≤ ≤ ≤
h (1 − r)2 [1 − (r + h)]2 (1 − r)4 (1 − r)4
for every eit ∈ T . Since µ is a finite Borel measure on T , we have

Z
40 40
4
dµ(eit ) = · µ(T ) < ∞
T (1 − r) (1 − r)4
40 1
which means that (1−r) 4 ∈ L (T ). Hence Theorem 1.34 (The Lebesgue’s Dominated
Convergence Theorem) ensures that the order of integration and the limit in (11.13) can
be changed and this action shows that the formula (11.11) holds. Similarly, it is easily to
check that
Pr (θ + h − t) − Pr (θ − t) −4r(1 − r 2 ) sin(θ − t + h2 ) sin h2
=
h h[1 − 2r cos(θ + h − t) + r 2 ][1 − 2r cos(θ − t) + r 2 ]
2r(1 − r 2 ) sin h2
≤ · h
(1 − r)4 2
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
Proof. Let f = u + iv. Since f ∈ H(Ω), both u and v are harmonic in Ω. Since |f | = 6 0 in Ω,
1
2 2
log |f | is a well-defined real function and in fact log |f | = log(u + v ) . Using ∆u = ∆v = 0
2
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
(u + v )(ux + uy + vx + vy ) − 2[u2 (u2x + u2y ) + 2uv(ux vx + uy vy ) + v 2 (vx2 + vy2 )]
2 2 2 2 2 2
=
(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
in Ω. Hence log |f | is harmonic in Ω.

1 ′ f′
If f is holomorphic and non-vanishing in Ω, then f,f ∈ H(Ω). Thus we obtain f ∈ H(Ω).
f′
Since (log f )′
= f ,
we conclude log f ∈ H(Ω) so that its real part, which is log |f |, is harmonic
in Ω, completing the proof of the problem.
Problem 11.6
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 [59, Eqn. (6), p. 11].
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
Since |f ′ (z)|2 = f ′ (z) · f ′ (z), we see that

X
∞ X
∞
|f ′ (r cos θ, r sin θ)|2 = ncn r n−1 ei(n−1)θ × mcm r m−1 e−i(m−1)θ
n=1 m=1
∞
X
= nmcn cm r n+m−2 ei(n−m)θ .
m,n=1
Recall from Definition 4.23 that {einθ } forms an orthonormal set, the integral (11.15) becomes
Z 2π Z 1X
∞
2 2 2n−1
A(Ω) = n |cn | r 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→∞
which implies that p

lim sup n
n2 |cn |2 = 1.
n→∞
∞
X
Therefore, the radius of convergence of the series n2 |cn |2 r 2n−1 is 1 and then the order of
n=1
integration in the integral (11.16) can be switched so that
Z ∞ Z 1
2π X ∞
X ∞
X
1
A(Ω) = n2 |cn |2 r 2n−1 dr dθ = 2π n2 |cn |2 × =π n|cn |2 ,
0 0 2n
n=1 n=1 n=1
as required. This completes the proof of the problem.
Problem 11.7
Proof. Suppose that f = u + iv.
(a) Let ψ be a twice differentiable function on (0, ∞). Then f f = u2 + v 2 = |f |2 . Using the
formulas [62, 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
+ iψ ′′ (|f |2 ) · (uux + vvx )(uuy + vvy )

+ iψ ′ (|f |2 ) · (ux uy + uuxy + vx vy + vvxy )

+ −iψ ′′ (|f |2 ) · (uuy + vvy )(uux + vvx )
−iψ ′ (|f |2 )(uy ux + uuxy + vx vy + vvxy )
o
+ 2ψ ′′ (|f |2 ) · (uuy + vvy )2 + ψ ′ (|f |2 ) · (u2y + uuyy + vy2 + vvyy ) . (11.17)
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 [62, Eqn. (3), p. 232] to get
∆(|f |α ) = 4∂∂(|f |α ) = α2 |f |α−2 |f ′ |2 .
(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
= Φ(u, v) + Φ(u, v)
∂x2 ∂y 2
∂ ∂
= (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
∆[Φ ◦ f ] = Φuu · (u2x + vx2 ) + Φvv · (vx2 + u2x )

= (Φuu + Φvv ) · |f ′ |2
= [(∆Φ) ◦ f ] · |f ′ |2 (11.20)
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 .
This ends the analysis of the problem.
Problem 11.8
Proof. The proof of this problem will be divided into two steps as follows:
• Step 1: {fn (z)} converges at every point of Ω. Suppose that
S1 = {z ∈ Ω | {fn (z)} converges}
Let a ∈ Ω and R > 0 be such that
D(a; 2R) ⊆ Ω. (11.21)
We consider the functions ucn (z) = un (z + a) and c

fn (z) = fn (z + a) which are defined in
D(0; 2R) and in D(0; 2R) respectively. By [9, Theorem 16.9, p. 233]c , we have
Z π
c c 1 2Reit + z
fn (z) = iIm fn (0) + cn (2Reit ) dt
u
2π −π 2Reit − z
and thus Z
1 2Reit + z
π
fn (z + a) = iIm fn (a) + it − z
un (a + 2Reit ) dt,
2π −π 2Re
where z ∈ D(0; 2R). If we take z = Reiθ , then it is easy to see that
2Reit + z 2Reit + Reiθ

= ≤3
2Reit − z 2Reit − Reiθ
and this implies that
|fn (z + a) − fm (z + a)| ≤ Im [fn (a) − fm (a)]

Z π
1 2Reit + z
+ [un (a + 2Reit ) − um (a + 2Reit )] dt
2π −π 2Reit − z
Z π
3
≤ |fn (a) − fm (a)| + un (a + 2Reit ) − um (a + 2Reit ) dt
2π −π
c
In fact, this is a generalization of Theorem 11.9 and it is called the Schwarz Integral Formula. See, for
example, [37, p. 408].
≤ |fn (a) − fm (a)| + 3 · max un (a + 2Reit ) − um (a + 2Reit ) .

−π≤t≤π
Consequently, we conclude that
sup |fn (z + a) − fm (z + a)| ≤ 3 · max un (a + 2Reit ) − um (a + 2Reit )

z∈D(0;R) −π≤t≤π
+ |fn (a) − fm (a)|
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
|fn (z) − fm (z)| < ǫ (11.24)
for all z ∈ D(a; R) and all n, m ≥ N = max(N1 , N2 ). By [61, 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 [15, Theorem 1.28, p. 28]:
Lemma 11.1 (The Basic Connectedness Lemma)

Suppose that Ω is open in C and A is a nonempty subset of Ω. If there exists
a θ > 0 such that for every a ∈ A and D(a; 2R) ⊆ Ω for some R > 0, we have
D(a; 2θR) ⊆ A, then A is relatively open and closed in Ω. Furthermore, if Ω is a
region, then we have
A = Ω.
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.
• Step 2: {fn } converges uniformly on compact subsets of Ω. Let K be a compact

subset of Ω. Clearly, [
K⊆ D(a; Ra ),
a∈K
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
where Rj ∈ {Ra | a ∈ K} for 1 ≤ j ≤ p. In particular, if we take N = max N (aj , Rj ),

1≤j≤p
then it is true that n, m ≥ N imply
|fn (z) − fm (z)| < ǫ
for all z ∈ K. Using [61, Theorem 7.8, p. 147] again, the sequence {fn } converges
uniformly on K.
Problem 11.9
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
which gives exactly
Z r Z π ZZ
1 iθ 1
u(a) = 2 u(a + ρe )ρ dθ dρ = 2 u(x, y) dx dy. (11.27)
πr 0 −π πr
D(a;r)
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)
On the other hand, we have Z r

2
r u(a) = 2 ρu(a) dρ. (11.29)
0
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 [79, p. 161] that
Z π
1 r
ru(a) = · U (a, r, θ)r = u(a + reiθ ) dθ. (11.32)
2π 2π −π
After cancelling the r in the expression (11.32), we find that

Z π
1
u(a) = u(a + reiθ ) dθ
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.
= ǫ.
By the definition, u is continuous at z. Since z is arbitrary, u is actually continuous in Ω. Finally,

Theorem 11.13 asserts that u is harmonic in Ω, completing the proof of the problem.
Problem 11.10
Proof.
• By the definition, we have
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
a
Z ∞
= ϕ(x − t)Pǫ (t) dt, (11.34)
−∞
where
1 ǫ
Pǫ (t) = · 2
π t + ǫ2
relates to the formula [62, Eqn. (3), §9.7, p. 183] (in fact, it is the Poisson kernel for the
upper half-plane, see [66, p. 149]) and ϕ(t) = 0 if t ∈ R \ I. Using the convolution notation
introduced in Theorem 8.14, the expression (11.34) becomes
f (x + iǫ) − f (x − iǫ) = (ϕ ∗ Pǫ )(x). (11.35)
Since |ϕ(x)| ≤ max |ϕ(t)| for every x ∈ R, we get ϕ ∈ L∞ (R).

t∈I
To proceed, we need to modify Theorem 9.9 and givese
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 [62, 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.
Proof of Lemma 11.2. Following the proof of Theorem 9.9, we have

Z 0 h
g(x+) + g(x−) g(x+) + g(x−) i
(g ∗ hλ )(x) − = g(x − λs) − h1 (s) dm(s)
2 −∞ 2
Z ∞h
g(x+) + g(x−) i
+ g(x − λs) − h1 (s) dm(s).
0 2
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.
By Lemma 11.2, we conclude at once that
ϕ(x+) + ϕ(x−)
lim [f (x + iǫ) − f (x − iǫ)] = (11.36)
ǫ→0 2
ǫ>0
for every x ∈ R. Hence the formula (11.36) asserts that



 ϕ(x), if x ∈ (a, b);







 0, if x ∈ R \ I;


lim [f (x + iǫ) − f (x − iǫ)] = ϕ(a+)
ǫ→0 
 , if x = a;
ǫ>0 
 2






 ϕ(b−) ,

if x = b.
2
• The case when ϕ ∈ L1 . In this case, we apply Theorem 9.10 to the expression (11.35)
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
lim kfǫ+ − fǫ− k1 = kϕk1 .

ǫ→0
ǫ>0
• The case when ϕ(x+) and ϕ(x−) exist at x. This case has been settled already in the
formula (11.36).
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 [44].
Problem 11.11
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.
• By Theorem 10.17 (Morera’s Theorem), it suffices to prove that

Z
f (z) dz = 0 (11.38)
∂∆
for every closed triangle ∆ ⊆ Ω. There are three cases.
– Case (i): ∆ ∩ I = ∅. Then ∆ ⊆ Ω \ I. Since Ind ∂∆ (x) = 0 for every x ∈ I, Theorem

10.35 (Cauchy’s Theorem) implies that the result (11.38) holds trivially.
– Case (ii): ∂∆ has a side lying on I. Since ∆ is compact, C \ Ω is closed in C
and ∆ ∩ (C \ Ω) = ∅, Problem 10.1 ensures that there exists a δ1 > 0 such that
d(∆, C \ Ω) = δ1 > 0. Therefore, it is true that
i
∆n = ∆ + ⊆Ω (11.39)
n
for all n > δ11 . Thus we have ∆n ∩ I = ∅ so that the result (11.38) holds for every
∂∆n with n > δ11 . Suppose that ∂∆ = [a, b]+[b, c]+[c, a], where a, b and c are vertices
of the triangle ∆. Then we have
i i i
∂∆n = [a, b] + + [b, c] + + [c, a] + .
n n n
Using [62, Eqn. (4), p. 202], we know that
Z Z 1
i
f (z) dz = (b − a) f a + + (b − a)t dt
[a,b]+ ni 0 n
so that
Z Z
f (z) dz − f (z) dz
[a,b] [a,b]+ ni
Z 1h
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
Since f is continuous on Ω, it is uniformly continuous on any compact subset of Ω.

Given ǫ > 0, there exists a δ2 > 0 such that n > δ12 implies
i ǫ
f a + + (b − a)t − f a + (b − a)t <
n 3|b − a|
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.
(a) ∂∆ = ∂∆1 + ∂∆2 .

(b) ∂∆ = ∂∆1 + ∂∆2 + ∂∆3 .
Figure 11.1: The I divides ∂∆ into several triangles.

11.2. Harnack’s Inequalities and Positive Harmonic Functions 49
• Removable sets for holomorphic functions of class C (Ω). Let N ∈ N. It is clear

that the above argument can be applied to the compact set in the form
K = I1 ∪ I2 ∪ · · · ∪ IN
and f ∈ H(Ω \ K), where each Ij = [aj , bj ] is a subset of Ω for 1 ≤ j ≤ N .
We have completed the analysis of the problem.f
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 [12, 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 [19] 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 [74] 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.
11.2 Harnack’s Inequalities and Positive Harmonic Functions
Problem 11.12
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.
To see this, choose ρ > 0 such that r < ρ < R. Using the second set of inequalities on [62,
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
αun (z0 ) ≤ un (z) ≤ βun (z0 ) (11.44)
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.
• The positivity of {un } is essential. For each n ∈ N, we consider u(x, y) = nx − ny in C.

Then it is easily checked that ∆un = 0 so that un is harmonic in C. As un (0, 1) = − n1 < 0
and un (1, 0) = n > 0, each un is neither positive nor negative. Furthermore, un (0, 1) → 0
but un (1, 0) → ∞ as n → ∞. Hence this counterexample shows that the positivity of {un }
cannot be omitted for these results.
Problem 11.13
Proof. By the Poisson formula, we have

Z π
1 1 − r2
u(reiθ ) = u(eit ) dt, (11.46)
2π −π 1 − 2r cos(θ − t) + r 2
1
where 0 ≤ r < 1 and −π ≤ θ ≤ π. Pick r = 2 and θ = 0 in the formula (11.46) to get
Z Z
1 π 1 − 41 1 π
3
u( 12 ) = it
1 u(e ) dt = 2π u(eit ) dt. (11.47)
2π −π 1 − cos t + 4 −π 5 − 4 cos t
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
· it
u(e ) dt ≤ u( 12 ) ≤3· u(eit ) dt. (11.48)
3 2π −π 2π −π
Put r = 0 in the Poisson formula (11.46), we obtain

Z π
1
u(0) = u(eit ) dt,
2π −π
therefore, we conclude from the hypothesis u(0) = 1 and the inequalities (11.48) that
1
≤ u( 21 ) ≤ 3.
3
Problem 11.14
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 .
• Case (ii): L1 and L2 are perpendicular. Suppose that L2 : x = 0. Then it is easy to

check that the function u(x, y) = xy satisfies all the requirements.
• 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
xn = (x + i · 0)2 = un (x, 0) + ivn (x, 0)
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 [7, 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
|nαπ − mπ − θπ| < δ.

h
See, for example, [5, §7.4, pp. 148, 149].
Note that tan(nαπ − mπ) = (−1)m tan(nαπ). The continuity of tan x implies that
|(−1)m tan(nαπ) − tan(θπ)| < ǫ. (11.50)
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
|(−1)m+1 tan(nαπ) − tan(θπ)| < ǫ.
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
| tan(nαπ) − tan(θπ)| < ǫ (11.51)
for some n ∈ Z. If θ = 0, then we follow from the Dirichlet’s Approximation Theoremi

π
that for a positive integer N with N < δ, there exist m, n ∈ N with 0 < n ≤ N such that
π
|nαπ − mπ| < <δ
N
and again the continuity of tan x implies that
| 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
Proof. We first prove the following lemma:
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 [79, Problem 2.4, p. 12].
Proof of Lemma 11.4. Recall from the series expansion of the Poisson kernel that
Pr (t) = Pr (−t). We know from [62, 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 − 2(1 − ǫ) cos ǫ + (1 − ǫ)2 = 2(1 − ǫ) − 2(1 − ǫ) cos ǫ + ǫ2

= 2(1 − ǫ)(1 − cos ǫ) + ǫ2
ǫ2 ǫ4
= 2(1 − ǫ) − + · · · + ǫ2
2! 4!
ǫ 2
≤
M
for some constant M > 0. Thus they imply that
1 − (1 − ǫ)2 Mǫ
ǫP1−ǫ (ǫ) = ǫ · 2
≥ǫ· 2 =M
1 − 2(1 − ǫ) cos ǫ + (1 − ǫ) ǫ
which gives the desired result.
Let’s return to the proof of the problem. When u is considered in D(0; r) for 0 < r < 1,
[9, 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 iθ 1
sup kur k1 = sup u(re ) 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
iθ
1 M i(θ−t)
M M
u (1 − ǫ)e ≥ dµ e ≥ dµ ei(θ−t) ≥ µ I(eiθ ; 2ǫ) . (11.54)
2π [−ǫ,ǫ] ǫ ǫ [−ǫ,ǫ] ǫ
Take ǫ → 0 in the inequality (11.54), the condition u(reiθ ) → 0 as r → 1 implies that

µ I(eiθ ; 2ǫ)
lim =0
ǫ→0 ǫ
for every eiθ 6= 1. Consequently, we have µ(T \ {1}) = 0 so that µ is a positive point mass at 1
which means that Z
iθ 1
u(re ) = Pr (θ − t) dµ(eit ) = µ(1)Pr (θ),
2π T
completing the proof of the problem.
Problem 11.16
Proof. Assume that u was the Poisson integral of a complex measure µ on T , i.e., u = P [ dµ].
We know from [62, Eqn. (2), p. 244] that
kur k1 ≤ kµk = |µ|(T ) < ∞
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
0 ≤ cα (Nα u)(−1) ≤ (Mrad u)(−1) = sup{|u(−r)| | 0 ≤ r < 1} ≤ lim |u(−r)|. (11.56)

r→1
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
(Nα u)(−1) = 0. (11.57)
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 iθ 1 iθ 1
kur k1 = |u(re )| dθ ≤ v(re ) 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
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
is harmonic and also positive in U . In addition, it is easy to see that

Z Z
it it
u(0) = P (0, e ) dµ(e ) = dµ(eit ) = |µ|(T ) (11.60)
T 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 iθ 1
kur k1 = |u(re )| 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
it it
αu(z) + βv(z) = α P (z, e ) dµ(e ) + β P (z, eit ) dν(eit )
Z T T
= P (z, eit ) d[αµ + βν](eit ),

T
i.e., f (αu + βv) = αµ + βν = αf (u) + βf (v).
• 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 [62, 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 57
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 }.

11.3 The Weak∗ Convergence and Radial Limits of Holomorphic Functions
Problem 11.18
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
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, [7, p. 115, 116].
k
Recall from Remark 5.21 that X ∗ is the collection of all bounded linear functionals on X.
Problem 11.19
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) = 2
> = 1.
1 − 2r cos(1 − r) + r 1 + r(1 − r)2 − 2r + r 2
(b) Clearly, we have

Z Z
it it
δu(1 − δ) = δP (1 − δ, e ) dµ(e ) = δP1−δ (t) dµ(eit ). (11.62)
T T
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
Since µ ≥ 0, we have u(1 − δ) ≥ 0. By Definition 11.19, we have

µ(Iδ ) µ(Iδ )
= ≤ u(1 − δ) ≤ (Mrad u)(1) (11.63)
πσ(Iδ ) δ
for every δ > 0. Recall the definition [62, Eqn. (3), §11.19, p. 241], we conclude that
(M µ)(1) ≤ π(Mrad u)(1)
as desired.
(c) In fact, the results of part (b) are valid for every point eiθ on T if we apply the special case
to the rotated measure µθ (E) = µ(eiθ E). In particular, we deduce from the inequalities
(11.63) that
µ(Iδ )
≤ u (1 − δ)eiθ ≤ (Mrad u)(eiθ ), (11.64)
πσ(Iδ )
where eiθ ∈ T and Iδ is the open arc with center eiθ and length 2δ.
Next, we know from [62, Eqn. (2), §11.17 p. 240] that
kur k1 ≤ kµk = |µ|(T ) < ∞
for every 0 < r < 1. Therefore, we follow from Theorem 11.30(A) that µ is a Borel measure
on T . By the hypothesis, µ is positive. Since µ ⊥ m, it follows from Theorem 7.15 that
(Dµ)(eiθ ) = ∞ a.e. [µ]. (11.65)
Finally, we observe from the inequality (11.64), the result (11.65) and [62, Eqn. (4), p.
241] that
lim u (1 − δ)eiθ = ∞ a.e. [µ]
δ→0
which is exactly the required result.
11.3. The Weak∗ Convergence and Radial Limits of Holomorphic Functions 59
Problem 11.20
Proof. Since m(E) = 0, for each n ∈ N, we know from [61, 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
iθ 1
χVn (e ) dm = m(Vn ) ≤ = 1,
2n
n=1 T n=1 n=1
so Theorem 1.38 implies that

Z ∞ Z
X
iθ
|ϕ(e )| dm = χVn (eiθ ) dm = 1.
T n=1 T
In other words, ϕ ∈ L1 (T ) and kϕk1 = 1.

Define u : U → R by u = P [ϕ] which is harmonic in U by Theorem 11.7. Now, for each
eiθ ∈ E, we have eiθ ∈ Vn for every n ∈ N, so it follows from the definition that
Z π
iθ 1
lim u(re ) = lim P (reiθ , eit )ϕ(eit ) dt = ∞.
r→1 r→1 2π −π
Since U is simply connected, u is the real part of a holomorphic function g in U , see [9, Theorem
16.3, p. 226]. Let f = e−g . Then we have f ∈ H(U ) and |f | = e−Re g = e−u so that
lim f (reiθ ) = lim e−u = 0

r→1 r→1
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 [53, p. 295] and [84, pp. 105, 276].
Problem 11.21
t+is−1
Proof. We first show that g ∈ / H ∞ . In fact, fix s ∈ R, consider zt = t+is+1 for 0 < t < ∞. Then
it is easy to check that |zt | < 1 and 1+z t
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
1 + reiθ 1 − r 2 + 2i sin θ 1 − r 2 + 2i sin θ

= and f (reiθ ) = exp .
1 − reiθ 1 − 2r cos θ + r 2 1 − 2r cos θ + r 2
By the definition of g, we have

h 1 − r 2 + 2i sin θ i
g(reiθ ) = (1 − reiθ ) · exp − exp
1 − 2r cos θ + r 2
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
the representation (11.67) gives

h i sin θ i
lim g∗ (eiθ ) = lim (1 − eiθ ) exp − exp
θ→0 θ→0 1 − cos θ
h i sin θ i
= lim (1 − eiθ ) lim exp − exp
θ→0 θ→0 1 − cos θ
−1
=0·e
= 0.
As a consequence, we establish the fact that g∗ ∈ C(T ) and this completes the proof of the
problem.
Problem 11.22
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 − u− , where u+ and u− are the positive and negative parts of u , see
r r r r
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
+
ur dm = ur dm. (11.69)
−π Er+
Let N be the least positive integer such that 2π l

N ≤ δ. Define Ik to be the arc from e
(k−1)δi to
ekδi including e(k−1)δi but not ekδi , where k = 1, 2, . . . , N − 1. Similarly, we define IN to be the
arc from e(N −1)δi to e2πi = 1 including e(N −1)δi but not e2πi = 1. Clearly, the central angles of
I1 , I2 , . . . , IN −1 are exactly δ and the central angle of IN is less than or equal to δ. Next, we
denote
+
Er,k = Er+ ∩ Ik
+ + +
so that {Er,1 , Er,2 , . . . , Er,N } forms a disjoint measurable subsets of Er+ and
+
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 N Z
X N
X Z
u+
r dm = ur dm ≤ ur dm ≤ N
+ +
−π k=1 Er,k k=1 Er,k
for every 0 ≤ r < 1. Similarly, it can be shown that

Z π
u−
r dm ≤ N
−π
for every 0 ≤ r < 1. Thus they imply that

Z π Z Z π
1 1 π + iθ N
kur k1 = iθ
|ur (e )| dθ ≤ ur (e ) dθ + u−
r (eiθ
) dθ ≤
2π −π 2π −π −π π
for every 0 ≤ r < 1.

Hence we apply Theorem 11.30(a) to obtain a unique complex Borel measure µ such that
u = P [ dµ]. Next, we know form Theorem 11.24 that there exists a f ∈ L1 (T ) such that
lim u(zj ) = f (eiθ )

j→∞
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.
as j → ∞ or equivalently, we have
Z Z
iθ iθ
lim urj (e ) dσ(e ) = f (eiθ ) dσ(eiθ ). (11.70)
j→∞ T T
Finally, we take g = 1 in [62, Eqn. (3), p. 247] to get

Z Z
iθ iθ
lim urj (e ) dσ(e ) = dµ(eiθ ). (11.71)
j→∞ T T
Combining the results (11.70) and (11.71), we establish
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 [62, p. 341] to prove that
µ ≪ σ.
Problem 11.23
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
≤ 2
= < ∞.
|e − z| (1 − |z|) 1 − |z|
Thus the series

∞
X ∞
X
−2
v(z) = n P (z, e iθn
) and w(z) = n−2 P (z, e−iθn )
n=1 n=1
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
−2
vn (z) = k iθk
P (z, e ) 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 .
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
v = P [ dµ] and w = P [ dν].
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 [62, 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]
= +
| cos θ − 1 + sin θ|2 |(cos θ − 1 + sin θ) − 2i sin θ|2
h 1 1 i
= 2 sin θ(1 − sin θ) · −
(cos θ + sin θ − 1)2 (cos θ + sin θ − 1)2 + 4 sin2 θ
h 1
= 2 sin θ(1 − sin θ) · 2 θ
4 sin 2 (cos 2θ − sin 2θ )2
1 i
− . (11.72)
4 sin2 θ2 (cos 2θ − sin 2θ )2 + 4 sin2 θ
Since (cos θ2 − sin 2θ )2 = 1 − sin θ, the expression (11.72) reduces to
4 sin2 θ2 (1 − sin θ) + 4 sin2 θ − 4 sin2 θ2 (1 − sin θ)

P (z, eiθ ) − P (z, e−iθ ) = 2 sin θ ·
4 sin2 2θ [4 sin2 θ2 (1 − sin θ) + 4 sin2 θ]
sin3 θ
=
2 sin2 2θ [sin2 θ2 (1 − sin θ) + sin2 θ]
sin3 θ
≥
2 sin2 2θ (sin2 θ2 + 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 [62, Eqn. (6), p. 233] that
P (x + iy, eiθ ) − P (x + iy, e−iθ ) > 0 (11.75)
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
u(1 − ǫ + iǫ) > n−2 [P (1 − ǫ + iǫ, eiθn ) − P (1 − ǫ + iǫ, e−iθn )] (11.76)
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 ) > = ≥ (11.77)
n 2 ǫn n2 sin 21n 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
Proof. Most assertions of this problem come from [61, Exercise 15, p. 199] and their solutions
are shown by the author in [77, 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 [61, 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
If −π ≤ t ≤ 0, then we let t = −s so that 0 ≤ s ≤ π. In this case, we have

sin N t 2 sin N s 2 1 π2 π
2 2
= s ≤ ≤ = 2. (11.81)
sin t
2
sin 2 sin2 s
2
s2 t
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 [62, Eqn. (1), p. 101] that
Z Z
it

sn (f ; θ) = f (e )Dn (θ − t) dσ = f ei(θ−t) Dn (t) dσ,
T T
so it is easy to check that

Z Z
i(θ−t)
h 1 X
N i
σN (f ; θ) = f e Dn (t) dσ = f ei(θ−t) KN (t) dσ. (11.84)
T N +1 T
n=0
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 →∞
It is easy to check that the KN satisfies KN (t) ≥ 0, KN (t) is even and

Z
KN (t) dσ = 1
T
for every N ∈ N. Thus it follows from the expression (11.84) that

Z π
1
|σN (f ; θ)| = f ei(θ−t) KN (t) dt
2π −π
m
See, for instances, [61, Exercises 15, 16, p. 199] or [84, Theorem 3.4, p. 89].
Z π Z 0
1 i(θ−t)

= f e KN (t) dt + f ei(θ−t) KN (t) dt
2π 0 −π
Z π
1
≤ f ei(θ−t) + 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 [62, 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
h 1 Z δ
2
π G(δ) 2 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(θ+t)
i
≤ |f e | dt + |f e | dt
N δ2 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
for sufficiently large N . Since ǫ is arbitrary, we have obtained the desired result that
σN (f ; θ) → 0 as N → ∞.
Problem 11.25
Proof. Let z = x + iλ, where λ > 0. Using [62, 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
property for every t ∈ R:
Z π
1
ϕ(z, t) = ϕ(z + reiθ , t) dθ,
2π −π
where r is any positive number such that D(z, r) ⊆ Π+ .
Mz
• Step 2: ϕ(z, t) ≤ 1+t2
for some Mz > 0. We claim that there exists a constant Mz > 0
such that
Mz
0 < ϕ(z, t) = ϕ(x + iλ, t) ≤ (11.91)
1 + t2
for all t ∈ R. To see this, simple calculation shows that
λ Mz
2 2
≤
(x − t) + λ 1 + t2
holds for all t ∈ R if and only if
(λ − Mz )t2 + 2Mz xt + (λ − Mz x2 − Mz λ2 ) ≤ 0
for all t ∈ R if and only if
(2Mz x)2 − 4(λ − Mz )(λ − Mz x2 − Mz λ2 ) ≤ 0
λMz2 − (x2 + λ2 + 1)Mz + λ ≥ 0 (11.92)
and the inequality (11.92) is true for large Mz > 0. This proves the claim.
n
In fact, it can be shown directly from the definition (11.90) that, for each t ∈ R, we have
6λ(x − t)2 − 2λ3 [(x − t)2 + λ2 ](−2λ) − 4λ[(x − t)2 − λ2 ]

ϕxx = and ϕλλ =
[(x − t)2 + λ2 ]3 [(x − t)2 + λ2 ]3
which give ∆ϕ = 0.
• Step 3: ϕ(z, t) ∈ Lq (R) for every 1 ≤ q ≤ ∞. By Step 2, if 1 ≤ q < ∞, then we have

1 ≤ 1 + t2 ≤ (1 + t2 )q so that
Z ∞ Z ∞ Z ∞
q Mzq q 1
ϕ (z, t) dt ≤ 2 )q
dt ≤ Mz · dt < ∞. (11.93)
−∞ −∞ (1 + t −∞ 1 + t2
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 ∞
iθ
kϕ(z + re , ·)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 iθ 1
u(z + re ) 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).
• Step 5: u is continuous on Π+ . Let {zn } be a sequence of Π+ converging to z0 ∈ Π+ .

Then we must have ϕq (zn , t) → ϕq (z0 , t) as n → ∞. Now the inequality (11.93) guarantees
that we may use Theorem 1.34 (The Lebesgue’s Dominated Convergence Theorem) to
conclude that
Z
1 ∞
lim u(zn ) = lim ϕ(zn , t)f (t) dt
n→∞ n→∞ π −∞
Z ∞h i
1
= lim ϕ(zn , t) f (t) dt
π −∞ n→∞
Z
1 ∞
= ϕ(z0 , t)f (t) dt
π −∞
= u(z0 ).
Thus u is continuous on Π+ .
• Step 6: u is harmonic in Π+ . In fact, this follows immediately from Steps 4, 5 and

Theorem 11.13.
Hence we have completed the analysis of the proof.

CHAPTER 12
The Maximum Modulus Principle
12.1 Applications of the Maximum Modulus Principle
Problem 12.1
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
max |f (z)| = max |f (z)|.

z∈∆ z∈∂∆
Let L = |a − b| = |b − c| = |c − a| and t = t(z) = |z − a| ∈ [0, L], see Figure 12.1 below.
Figure 12.1: The boundary ∂∆.
If z ∈ [a, b], then we have
|f (z)|2 = |z − a|2 · |z − b|2 · |z − c|2

h √
2 2 L 2 3L 2 i
= t (L − t) · t − +
2 2
69
70 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
Problem 12.2
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 [9, 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
−1
constant λ with |λ| = 1. Since ϕ0 (z) = z, we conclude that f h (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 71
Problem 12.3
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
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)
for all z ∈ D. On the other hand, since f1 ∈ H(D) and 1

f ∈ C(D), we follow from Theorem
10.24 (The Maximum Modulus Theorem) that
1 1
≤ (12.6)
|f (z)| M
for all z ∈ D. Combining the inequalities (12.5) and (12.6), we conclude that |f (z)| = M
in D, or equivalently, f (D) ∈ {−M, M } which contradicts the Open Mapping Theorem.
Hence f has at least one zero in D.
(b) This part is proven in [76, Problem 7.5, pp. 85 – 87].

We end the proof of the problem.
Problem 12.5
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 [61, Theorem 7.8,
p. 147], the inequality (12.8) ensures that {fn } converges uniformly on Ω which ends the proof
of the problem.
Problem 12.6
Proof. By the definitions, Γ∗ is closed in C and then C \ Γ∗ is open in C. By Definition 10.1,

C \ Γ∗ is a union of disjoint open connected sets, and hence components. Suppose that V is a
component of C \ Γ∗ such that Ind Γ (z) 6= 0 for every z ∈ V . Let Γ = γ1 +̇ · · · +̇γn , where each γj
is a closed path in Γ. Assume that V was unbounded. If z ∈ V , then it follows from Theorem
10.10 that Ind γj (z) = 0 for every 1 ≤ j ≤ n. Since
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 [42, 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
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.
Problem 12.8
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
M (x) = sup{|F (x + iy)| | α ≤ x ≤ β and y ∈ R}

= sup{|f (ex · eiy )| | α ≤ x ≤ β and y ∈ R}
= sup{|f (reiy )| | a ≤ r ≤ b and y ∈ R}
= M (r).
Therefore, it deduces from the inequality (12.12) that
(β − α) log M (r) ≤ (β − x) log M (a) + (x − α) log M (b)

log rb log r
a
log M (r) ≤ log M (a) + log M (b) (12.13)
log ab log b
a
which is the required result.

We claim that the equality (12.13) holds if and only if f (ζ) = Aζ λ for some A ∈ C and λ ∈ Z.
Obviously, the equality holds if f is of this form. Conversely, the we can rewrite the equality
(12.13) as
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) + b
log . (12.14)
log a 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
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
|ζ0λ f (ζ0 )| = aλ M (a). (12.16)
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
|eλz0 f (ez0 )| = aλ M (a) = eλα M (α),
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
eλz f (ez ) = c (12.17)
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
• |f (z)| ≤ 1 in the open right half plane Π. If α = 0, then |f (z)| ≤ Ae in Π. Since

Π is unbounded and ∂Π is exactly the imaginary axis, we may apply Problem 12.11 to
conclude that |f (z)| ≤ 1 in Π. Next, if α < 0, then there exists a R > 0 such that |z| ≥ R
implies
|f (z)| ≤ 1. (12.18)
Since D(0; R) ∩ Π is compact, it is bounded. Therefore, Theorem 10.24 (The Maximum
Modulus Theorem), the Extreme Value Theorem and the hypothesis |f (iy)| ≤ 1 for all
y ∈ R ensure that the inequality (12.18) is also valid for all z ∈ Π. Thus we may assume
that 0 < α < 1 in the following discussion.
Denote Ω = {x + iy | x ∈ R, |y| < π2 }. Consider the mapping ϕ : Ω → Π defined by
ϕ(z) = ez (12.19)
which is clearly an isomorphism and ϕ ∈ H(Ω). Next, we define g : Ω → C by g = f ◦ ϕ.

By Problem 10.14, we know that g ∈ H(Ω). By the definition of (12.19), we see that
ϕ(∂Ω) = ∂Π. Since ϕ and f are continuous on Ω and Π respectively, g is continuous on
Ω. Furthermore, we have
πi πi
g x± = f exp x ± = |f (±iex )| ≤ 1
2 2
for all x ∈ R. Finally, if z ∈ Ω, then we have

|g(z)| = |f ϕ(z) | < A exp(|ϕ(z)|α ) < A exp(|ez |α ) = A exp(eαx ) ≤ A exp(eα|x| ). (12.20)
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)
Combining the inequalities (12.20) and (12.21), the inequality
|g(z)| < exp(Beα|x| )
holds for all z = x + iy ∈ Ω. Hence Theorem 12.9 (The Phragmen-Lindelöf Theorem)

asserts that |g(z)| ≤ 1 in Ω which means that the inequality (12.18) holds in Π.
z
• The conclusion is false for α = 1. It is easy to check that the function f (z) = ee gives
a counterexample to the result.
• 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
|f (z)| < A exp(|z|α ) (12.22)
for all z ∈ ∆. Furthermore, if |f (z)| ≤ 1 on ∂∆, then we have |f (z)| ≤ 1 in ∆.

To see this, let θ be the angle between the real axis and the ray nearest to it. Then we see
that the mapping φ : ∆ → Π defined by
φ(z) = −i(e−iθ z)β = −ie−iβθ exp(β log z)
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
−1
have φ ∈ 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
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:
Figure 12.2: The sectors ∆1 , ∆2 and the ray Lα .
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)
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
|f (z)| < e−nr cos θ
for all n ∈ N which means f (z) = 0 because cos θ > 0 for θ ∈ (− π2 , π2 ), i.e., f = 0 as required.

Problem 12.11
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)
If fe(ω) 6= 0, then we deduce from the inequality (12.27) that
1
CN
|f (ω)| ≤ 1 . (12.28)
|fe(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.
12.2 Asymptotic Values of Entire Functions
Problem 12.12
Proof. Let E1 = {z ∈ C | |f (z)| > 1}. Since f is nonconstant, E1 6= ∅. Let F1 be a component

of E1 . By Definition 10.1, F1 is an open set. By the continuity of f , it is true that |f (z)| ≥ 1
on ∂F1 . Assume that |f (z0 )| > 1 for some z0 ∈ ∂F1 . Obviously, z0 ∈ E1 . Furthermore, we
also have |f (ω)| > 1 for all ω ∈ D(z0 ; δ) for some δ > 0 by the Sign-preserving Property [79,
Problem 7.15, p. 112]. Therefore, D(z0 ; δ) ⊆ E1 . By [42, Theorem 25.1, p. 159], we know that
D(z0 ; δ) intersects only F1 which is impossible by the definition of a boundary point of a set
[4, Definition 3.40, p. 64]. Hence we have |f (z)| = 1 on ∂F1 . Assume that F1 was bounded.
Since F1 is compact, it follows from Theorem 10.24 (The Maximum Modulus Theorem) and
the Extreme Value Theorem that f attains its maximum on ∂F1 . Thus |f (z)| ≤ 1 on F1 ,
contradicting to the fact that F1 ⊆ E1 . Consequently, every component of E1 is unbounded.
Now for every n = 2, 3, . . ., we define
En = {z ∈ C | |f (z)| > n}.
By similar argument, it can be shown easily that every component of En is unbounded.

Since f is unbounded on F1 , En 6= ∅ for every n = 2, 3, . . .. Clearly, we have En+1 ⊆ En for
every n ∈ N. Let Fn+1 be a component of En+1 . By similar argument of the previous paragraph,
Fn+1 is open in C so that it is a region. By the definition, we have Fn+1 ⊆ En and then it must
lie entirely in a component of En , namely Fn . Hence, we obtain a sequence of regions
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 [9, 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
Proof. If z = x < ∞, then |ez | = e−x → 0 as |z| → ∞. By the definition, 0 is an asymptotic

value of ez . Since ez is nonconstant entire, ∞ is also one of its asymptotic value by Problem
12.12. Hence it remains to show that if α is an asymptotic value of ez , then α is either 0 or
∞. Let γ : [0, 1) → C be a continuous curve such that γ(t) → ∞ and exp(γ(t)) → α as t → 1.
Let γ(t) = a(t) + ib(t), where a and b are continuous real-valued functions. Then we note that
a2 (t) + b2 (t) → ∞ as t → 1.
12.3. Further Applications of the Maximum Modulus Principle 79
If a(t) → +∞ as t → 1, then it is clear that | exp(γ(t))| = exp(a(t)) → ∞ as t → 1 so that

α = ∞. Next, if a(t) → −∞ as t → 1, then we | exp(γ(t)) = exp(a(t)) → 0 as t → 1 and so
α = 0 in this case. Finally, if a(t) → A as t → 1 for some finite A, then b(t) → ∞ as t → 1.
Since b is continuous, there exist sequences {tn } and {t′n } in [0, 1) such that b(tn ) = 2nπ and
b(t′n ) = (2n + 1)π respectively. Thus we have
′ ′ ′
eγ(tn ) = ea(tn ) × eib(tn ) → A and eγ(tn ) = ea(tn ) × eib(tn ) → −A
as n → ∞. These imply that α = A = −A and so α = A = 0. In conclusion, we have shown

that exp has exactly two asymptotic values: 0 and ∞.
For the entire functions sin z and cos z, we notice that
eiz − e−iz eiz + e−iz

sin z = and cos z = .
2i 2
Therefore, ∞ is the only asymptotic value of sin z and cos z, completing the proof of the problem.

Problem 12.14
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.
12.3 Further Applications of the Maximum Modulus Principle
Problem 12.15
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− n ) |f (z)|
1
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.
• Case (i): Z(f ) is infinite. Since Z(f ) ⊆ U which is bounded, the Bolzano-Weierstrass
Theorem [79, 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
(z − ζk )mk
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
mk
|f (zn )| ≤ M |zn − ζk | ≤M (1 + |ζk |) < ∞
k=1 k=1
for all n ∈ N.
Consequently, we have completed the proof of the problem.
Problem 12.16
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
Proof. By the definitions, we have
Φ = {f ∈ H(U ) | 0 < |f (z)| < 1 for all z ∈ U } and Φc = {f ∈ Φ | f (0) = c},
where 0 < c < 1.

12.3. Further Applications of the Maximum Modulus Principle 81
• 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
|f ′ (0)| ≤ 2|f (0) log f (0)|. (12.34)
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.
Hence we complete the analysis of the problem.
Remark 12.1
The first assertion of Problem 12.17 is called Rogosinski’s Theorem. See, for instances,
[57] and [15, Exercise 6.36, pp. 213, 214] for a different proof.
CHAPTER 13
Approximations by Rational Functions
13.1 Meromorphic Functions on S 2 and Applications of Runge’s Theorem
Problem 13.1
Proof. Suppose that f is meromorphic on S 2 and A ⊆ S 2 is the set of poles of f . If A is infinite,

then the compactness of S 2 implies that A has a limit point in S 2 . However, this contradicts
the note following Definition 10.41 and so A must be finite. Let A = {a1 , a2 , . . . , aN } for some
N ∈ N and mk be the order of ak for 1 ≤ k ≤ N . If we define
N
Y
P (z) = f (z) · (z − ak )mk ,
k=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
(z − ak )mk
k=1
f must be rational which completes the proof of the problem.

Problem 13.2
Proof.
(a) It is clear that Ω is simply connected, so S 2 \ Ω is connected by Theorem 13.11. Hence

Theorem 13.9 (Runge’s Theorem) implies that there exists a sequence {Pn } of polynomials
such that Pn → f uniformly on compact subsets of Ω. See Figure 13.1 for details.
83
84 Chapter 13. Approximations by Rational Functions
Figure 13.1: The simply connected set Ω.
(b) The answer is negative. Consider the function

1
f (z) = 1 (13.1)
z− 2
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
a contradiction. Hence no such sequence exists.

(c) The answer is still negative. The function (13.1) considered in part (b) is in fact holomor-
phic in C \ { 12 } which is open in C and it contains Ω.
Problem 13.3
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 85
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
Figure 13.2: The compact sets Dn and En .

Problem 13.4
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.
Figure 13.3: The compact sets An , Bn and Cn .

13.2. Holomorphic Functions in the Unit Disc without Radial Limits 87
13.2 Holomorphic Functions in the Unit Disc without Radial Limits
Problem 13.5
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 .
Figure 13.4: The disc ∆n , the arc Ln and its neighborhood 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
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 [61,
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
we obtain f ∈ H(U ) and Qn → f uniformly on compact subsets of U . Finally, we define

Pn = Qn − Qn−1 for every n = 1, 2, . . .. Each Pn is also a polynomial and
n
X
Pk = Qn .
k=1
∞
X
Therefore, we have constructed a sequence of polynomials {Pn } such that the series f = Pn
n=1
is holomorphic in U .
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
|P1 (z)| = |Q1 (z) − Q0 (z)| = |Q1 (z)| < δ
on ∆1 . By the inequality (13.7) again, we obtain
|Pn (z)| = |Qn (z) − Qn−1 (z)| < δn
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
|f (z)| ≤ |f (z) − Q2n (z)| + |Q2n (z)|

≤ |Q2n+1 (z) − Q2n (z)| + |f (z) − Q2n+1 (z)| + |Q2n (z)|
≤ |Q2n+1 (z) − Q2n (z)| + |Q2n+2 (z) − Q2n+1 (z)| + |f (z) − Q2n+2 (z)| + |Q2n (z)|
∞
X
≤ |Qk+1 (z) − Qk (z)| + |Q2n (z)|
k=2n
∞
X 1 1
< +
2k+1 22n
k=2n
1
≤ . (13.8)
22n−1
1
On the other hand, if z ∈ Ω2n+1 ⊆ Ω′2n+1 , then we have |Q2n+1 (z) − 1| < 22n+1
so that
∞
X
|f (z) − 1| ≤ |Qk+1 (z) − Qk (z)| + |Q2n+1 (z) − 1|
k=2n+1
X∞
1 1
< +
2k+1 22n+1
k=2n+1
1
≤ . (13.9)
22n
Assume that lim f (reiθ0 ) existed for some θ0 ∈ [0, 2π). Since every Ln intersects the radius
r→1
ℓ0 = {reiθ0 | 0 ≤ r < 1}, every Ωn also intersects ℓ0 . In other words, we can choose a sequence
{zn = rn eiθ0 } tending to eiθ0 ∈ T with the property that z2n+1 ∈ Ω2n+1 and z2n ∈ Ω2n . Then
the above estimates (13.8) and (13.9) show that
lim f (z2n ) = 0 and lim f (z2n+1 ) = 1

n→∞ n→∞
which are contrary. Hence this contradiction means that our f has no radial limit on T , com-
pleting the analysis of the problem.
Problem 13.6
Proof. Let’s prove the assertions one by one.
• The series converges if |z| < 1. It is easily seen that

 k−1
 22 if 1 ≤ k ≤ 3;
nk ≥

(k − 1)! · 2k if k ≥ 4.
For large k, since

k k
≤ ,
nk (k − 1)! · 2k
k
we have nk → 0 as k → ∞ and then
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
iθ 4 4iθ 9 9iθ
h e = 5(0.75) e + 25(0.75) e + 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
+ 1 − e
5m nm
k=m+1
X 5k
h m−1 1 nk iθnk i
= 1 − e − S m−1 + (Sm−1 + am )
5m nm
k=1
5k 1 nk iθnk
∞
X
+ 1 − e ,
5m 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:
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
|Sm−1 + am | ≥ |am | − |Sm | > 0.11 (13.11)
for all m = 3, 4, . . ..
Next, we know that
5k 1 nk iθnk X 5k h 1 nk i
m−1
X m−1
1− e − Sm−1 ≤ 1− 1− . (13.12)
5m nm 5m nm
k=1 k=1
Using differentiation, we always have 1 − (1 − x)n ≤ nx for every x ∈ [0, 1] and n ∈ N. By

the definition of nm , we see that
nm > 2j+1 (m − 1)(m − 2) · · · (m − j − 1)nm−j−1
for j = 0, 1, . . . , m − 2. Applying these to the inequality (13.12) to obtain
5k 1 nk iθnk
m−1
X m−1
1 X k nk
1− e − Sm−1 ≤ m 5 ·
5m nm 5 nm
k=1 k=1
m−1
X
1 5k
≤
5m 2m−k (m − 1)(m − 2) · · · k
k=1
m−1
X
1 1
≤ · 5k−m
m−1 2m−k
k=1
1 1
≤ · 1 − m−1
8(m − 1) 5
< 0.06 (13.13)
for every m = 3, 4, . . ..
Clearly, we deduce from the upper bound (13.10) that
5k 1 nk iθnk
∞
X ∞
X ∞
5k nk
nm 5 1 X 5k
1 − e = ·a ≤ nm+1 + m nk . (13.14)
5m nm 5m k 5
e nm k=m+2 e m
n
k=m+1 k=m+1
nm+1
As m ≥ 3, we get nm ≥ 6 which implies that
5 5
nm+1 ≤ < 0.0124. (13.15)
e nm e6
Since nk > 2(k − 1)nm for k = m + 2, m + 3, . . ., we have

nk
e nm > e2(k−1)
and then the inequality (13.14) becomes
5k 1 nk iθnk e2 X 5 k
∞
X ∞
25
1 − e ≤ = 2 < 0.026 (13.16)
5m nm 5m e2 (e − 5)e2m
k=m+2 k=m+2
for every m ≥ 3. Finally, by combining the bounds (13.11), (13.13), (13.15) and (13.16),
we conclude immediately that

h (1 − n1m )eiθ X 5k
m−1
1 nk iθnk
≥ |S m−1 + a m | − 1 − e − Sm−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.
Consequently, there exists a constant C > 0 such that
|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
m m z − αk
ϕ(z) = z ϕαk (z) = z (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 93
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
for all z ∈ D(0; 1 − n1m ) and m ≥ 3. Since nm → ∞ as m → ∞, we conclude from this

1
that f (z) = 0 in U which is impossible.
• The function h assumes every complex number α infinitely many times in U .

Define b
h(z) = f (z) − α. Then b
h ∈ H(U ). For large enough m, 5m > C2 |α| so that
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, [84,
Chap. V].
(b) In [29, 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 [16], 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
[26], [43] and [75]
13.3 Simply Connectedness and Miscellaneous Problems
Problem 13.7
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
component of S 2 \ Kn . By the proof of Theorem 13.9 (Runge’s Theorem), it suffices to prove

that V ∩ A 6= ∅.
Since every component of S 2 \ Kn contains a component of S 2 \ Ω, we have
U ⊆V (13.20)
for at least one component U of S 2 \ Ω. Since A intersects every component of S 2 \ Ω, we have
A ∩ U 6= ∅. (13.21)
Now the set relations (13.20) and (13.21) together imply that V ∩ A 6= ∅. If V ∩ A 6= ∅, then
we are done. Otherwise, p ∈ V ∩ A′ , where A′ is the set of limit points of A. By the openness
of V and the definition of limit points, there exists a δ > 0 such that q ∈ A ∩ D ′ (p; δ) and
D(p; δ) ⊆ V . This means that V ∩ A 6= ∅. Hence we have obtained the requirement and this
completes the analysis of the problem.
Problem 13.8
Proof. Let Ω = C. For every n = 1, 2, . . ., we denote Kn = D(0; n) which is compact. Put

A1 = A ∩ K1 and An = A ∩ (Kn \ Kn−1 ) for n = 2, 3, . . .. Since An ⊆ Kn and A has no limit
point in C (hence none in Kn ), every An is a finite set. Put
X
Qn (z) = Pα (z),
α∈An
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.
Problem 13.9
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 .
Take g = exp( nh ) which is obviously holomorphic in Ω. Hence we have
g n = eh = f
which ends the proof of the problem.
Problem 13.10
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
f (z) = (z − a)m g(z) (13.24)
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
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
Vk = {z ∈ U | |fn (z)| ≤ k for all n ∈ N} ⊆ U . (13.26)

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 [48]. In fact, Problem
10.5 is the well-known Vitali Convergence Theorem, see [72, p. 168].
Problem 13.12
Proof. We regard C as R2 and consider the two-dimensional Lebesgue measure m2 . Suppose

that f : R2 → R2 is measurable. Let f = u + iv and RN = [−N, N ] × [−N, N ]. Consider the
measurable function
u|RN : RN → R,
where N ∈ N. By [67, Theorem 4.3, p. 32], there exists a sequence {ψN,k } of step functions
converging to u|RN for almost every z ∈ RN , i,e., there corresponds an pN ∈ N such that k ≥ pN
implies
1
u|RN (z) − ψN,k (z) < N +1 (13.27)
2
for almost every z ∈ RN .
Note that each ψN,k has the form
mk
X
ψN,k (z) = αN,j χRN,j (z),
j=1
a
Without loss of generality, we may assume further that D(z0 ; δ) ⊆ Ω.
where {RN,j } forms a set of disjoint open rectangles and

!
m
[k mk
[ m
[k 1
m2 RN ∆ RN,j = m2 RN \ RN,j ∪ RN,j \ RN < . (13.28)
2N +1
j=1 j=1 j=1
Suppose that
m
[k
ΩN = RN,j .
j=1
Then we can represent the inequality (13.28) as

1
m2 (RN ∆ΩN ) = m2 (RN \ ΩN ) ∪ (ΩN \ RN ) < .
2N +1
By [67, Theorem 3.4, p. 21], we know that one can find a compact subset KN of ΩN such that
1
m2 (ΩN \ KN ) < .
2N +1
Since we have RN \ KN ⊆ (RN \ ΩN ) ∪ (ΩN \ KN ) and KN \ RN ⊆ ΩN \ RN , we get

m2 (RN ∆KN ) = m2 (RN \ KN ) ∪ (KN \ RN )

≤ m2 [(RN \ ΩN ) ∪ (ΩN \ KN )] ∪ (ΩN \ RN )
≤ m2 (RN ∆ΩN ) + m2 (ΩN \ KN )
1
< N. (13.29)
2
Obviously, each ΩN is open in C because each RN,j is an open set. Besides, as the complement
C \ ΩN is path-connected, it is also connected.b Since ψN,k is constant on each RN,j , it is
holomorphic in RN,j and hence in ΩN , i.e., ψN,k ∈ H(ΩN ). By Theorem 13.9 (Runge’s Theorem),
for each pair of fixed N and k, there is a polynomial sequence {QN,k,n } such that QN,k,n → ψN,k
as n → ∞ uniformly on compact subsets of ΩN . Fix the KN as above, there is an qN ∈ N such
that n ≥ qN implies that
1
|QN,k,n (z) − ψN,k (z)| < N +1 (13.30)
2
for all z ∈ KN . Combining the estimates (13.27) and (13.30), for k ≥ pN and n ≥ qN , we can
establish that
1
u|RN (z) − QN,k,n (z) ≤ u|RN (z) − ψN,k (z) + ψN,k (z) − QN,k,n (z) < (13.31)
2N
for almost every RN ∩ KN . Recall that
∞
[
C= RN .
N =1
By this fact and the inequality (13.29), we obtain
lim KN = C.
N →∞
If we pick UN (z) = QN,pN ,qN (z), then it can be established from the inequality (13.31) that
lim UN (z) = u(z) (13.32)

N →∞
b
Refer to the paragraph following the definition in [42, p. 155].
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
lim VN (z) = v(z) (13.33)

N →∞
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
CHAPTER 14
Conformal Mapping
14.1 Basic Properties of Conformal Mappings
Problem 14.1
Proof. The linear fractional transformation
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 [76, Problem 13.16, p. 181], completing the proof of the problem.

Problem 14.2
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 [9, 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 [18, pp. 50, 51]
99
100 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}.
If f ∈ H(Ω) and |f (zn )| → 1 for every {zn } in Ω which converges to a point of L,

then there exists a function F , holomorphic in Ω ∪ L ∪ Ω− , such that


 f (z), if z ∈ Ω ∪ L;

F (z) = 1 (14.2)

 , if z ∈ Ω −.

f (z)
Proof of Lemma 14.1. Fix a point t ∈ L. By the hypothesis |f (z)| → 1 as z → t ∈ L,

it is legitimate to select a disc Dt so small that f (z) 6= 0 in Π+ ∩ Dt . Then the function
g(z) = i log f (z) is well-defined and holomorphic in Π+ ∩ Dt . Furthermore, we know
that
Im g(zn ) = log |f (zn )| → 0
for every {zn } in Ω converging to a point of L. By the application of Theorem 11.14
(The Schwarz Reflection Principle), we see that one can find a function G, holomorphic
in Ω ∪ L ∪ Ω− , such that G(z) = g(z) in Ω and satisfies
G(z) = G(z) (14.3)
for all z ∈ Ω ∪ L ∪ Ω− . Define F (z) = e−iG(z) . Since f (z) = e−ig(z) , if z ∈ Ω, then we

have
F (z) = e−iG(z) = e−ig(z) = f (z).
Next, if z ∈ Ω− , then z ∈ Ω and we deduce from the equation (14.3) and the definition
of F that
1 1
F (z) = e−iG(z) = e−iG(z) = eiG(z) = =
F (z) f (z)
which is exactly the equation (14.2).
(b) Recall from [62, Eqn. (1), p. 281] that
z−i
ψ(z) =
z+i
is a conformal one-to-one mapping of Π+ onto U and ψ(R) ⊆ T . So the inverse
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
i(1 + eiθ ) sin θ

ψ −1 (eiθ ) = iθ
=− ∈ R.
1−e 1 − cos θ
14.1. Basic Properties of Conformal Mappings 101
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

 , if z ∈ Ωb −,

b
f (z)

 f ψ(z) , if z ∈ Ω b ∪ L;
b


= 1 (14.5)

 , if z ∈ Ωb −.

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}.
(c) We have the following analogous reflection theorem for U :
Lemma 14.2 (The Schwarz Reflection Principle for U )
Every eit ∈ L ⊆ T is the center of an open disc Dt such that Dt ∩ U lies in Ω.

Denote Ω∗ to be the reflection of Ω, i.e.,
n 1 o
Ω∗ = z ∈ C z ∗ = ∈ Ω .
z
Suppose that f ∈ H(Ω) and Im f (zn ) → 0 for every sequence {zn } in Ω which
converges to a point of L. Then there exists a function F , holomorphic in the set
Ω ∪ L ∪ Ω∗ , such that


 f (z), if z ∈ Ω ∪ L;

F (z) = 1 (14.8)


 f ∗
, if z ∈ Ω .
z
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
b and z → L. According to Theorem 11.14 (The Schwarz Reflection Principle),

as z ∈ Ω
there is a function Fb , holomorphic in Ω
b ∪Lb∪Ω b − , such that


 f ψ(z) , if z ∈ Ω b ∪ L;
b
b
F (z) = (14.9)
 f ψ(z ), if z ∈ Ω
 b −.
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
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.
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
m z − αn
B(z) = z ·
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
as desired. This ends the proof of the problem.
Problem 14.4
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)
where α, β ∈ U . Obviously, if |z| = 1, then z · z = |z|2 = 1 and 1 − αz 6= 0 which imply that
z (z − α)(1 − αz) (1 − αz)(1 − αz) |1 − αz|2

f (α, β, z) = · = = > 0.
z (z − β)(1 − βz) (1 − βz)(1 − βz) |1 − βz|2
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
has the same numbers of zeros and poles as those of R inside U .

Consequently, the quotient
R(z)
F (z) =
Q(z)
is a rational function without zeros or poles in U , i.e., F ∈ H(U ). Since f (αn , βn , z) > 0 on
|z| = 1 for every n = 1, 2, . . . , N , we also have Q(z) > 0 on |z| = 1 and hence F (z) > 0 on
|z| = 1. In other words, Im F ≡ 0 on |z| = 1. Recall that Im F is a continuous real-valued
function on U and is harmonic in U , so we follow from [7, Corollary 1.9, p. 7] that Im F ≡ 0 in

U . Finally, using [9, Proposition 3.6, p. 39], there is a positive constant c such that
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)
in Ω, completing the analysis of the problem.
Problem 14.5
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)
in Ω = C \ {β1 , β2 , . . . , βn , β1 , β1 , . . . , β1 }. Here {α1 , α2 , . . . , αn } and {β1 , β2 , . . . , βn } are sets of

1 2 n
zeros and poles of g inside U respectively. Since ζ n g(ζ) is a polynomial, the expression (14.13)
implies that β1 = β2 = · · · = βn = 0. Hence we obtain
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
n
√ Y 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
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.
• Case (i): d = 0. Since ϕ(z) → ac as z → ∞ along R, we have ac ∈ R or c = 0. In the

latter case, the transformation is ϕ(z) = ∞ for all z ∈ R ∪ {∞} and we can rewrite it as
1
ϕ(z) = 0·z+0 . For the case ac ∈ R, since ϕ(1) = ac + bc ∈ R with c 6= 0, we have bc ∈ R.
a
az + b z + bc
ϕ(z) = = c .
cz 1·z+0
b
• Case (ii): d ∈ R. Note that d 6= 0. Let B = db . Then we have
az + Bd
ϕ(z) = . (14.14)
cz + d
Using ϕ(z) → ac as z → ∞ along R again, we know that A = ac ∈ R or c = 0. In the latter
case, we have ϕ(z) = ad z + db . Since ϕ(1) = ad + db ∈ R and db ∈ R, we have ad ∈ R so that
a
dz + db
ϕ(z) = .
0z + 1
a
For the case A = c ∈ R, the representation (14.14) becomes
Acz + Bd
ϕ(z) = . (14.15)
cz + d
If c + d = 0, then c = −d 6= 0 so that the representation (14.15) reduces to
−Adz + Bd −Az + B
ϕ(z) = = .
−dz + d −z + 1
Otherwise, ϕ(1) = Ac+Bd d d
c+d = A + (B − A) · c+d ∈ R if and only if c+d ∈ R. Since d 6= 0,
d c
c+d ∈ R if and only if d ∈ R. Let c = Cd for some C ∈ R. Now the representation (14.15)
reduces to
ACdz + Bd ACz + B
ϕ(z) = = .
Cdz + d Cz + 1
This proves our claim.
Return to our original problem. Suppose that ϕα maps a straight line L into itself. It is clear
that eiθ L = R for some θ ∈ [0, 2π], so we assume that eiθ ϕα maps R into R. Write
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.
Problem 14.7
Proof. Suppose that z, ω ∈ U . By the definition, fα (z) = fα (0) = 0 so that z = 0 and α is

arbitrary in this case. Thus, without loss of generality, we may assume that z, ω 6= 0. Now
fα (z) = fα (ω) if and only if (z − ω)(1 − αzω) = 0 if and only if
1
z=ω or α= . (14.16)
zω
1
Since |z| < 1 and |ω| < 1, we have |zω| > 1. Suppose that |α| ≤ 1. Then the result (14.16) leads
to us that z = ω, i.e., fα is one-to-one in U .
Let |α| ≤ 1. It is easy to check that fα (0) = 0 and fα′ (0) = 1. Besides, we have f ∈ H(U ).
Combining these facts and the previous result, we know that fα ∈ L and
f (z) = z − αz 3 + α2 z 5 + · · · .
Therefore, it deduces from Theorem 14.14 that D(0; 14 ) ⊆ fα (U ). This ends the analysis of the
problem.
Problem 14.8
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
(r + 1r )2 (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).
If θ = π2 , 3π
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.
Problem 14.9
Proof. Define Ω1 = {z ∈ C | 0 < Im z < π} and Π+ = {z ∈ C | Im z > 0}.
(a) Notice that the map f1 : Ω → Ω1 defined by f1 (z) = πi ′ πi

2 (z + 1) satisfies f (z) = 2 6= 0
in Ω. By Theorem 14.2, it is a one-to-one conformal mapping of Ω onto Ω1 . Next, we
know from [9, p. 176] that f2 (z) = ez is a one-to-one conformal mapping of Ω1 onto the
upper half plane Π+ . Recall from [62, Eqn. (1), p. 281] that f3 (z) = z−i
z+i is a one-to-one
conformal mapping of Π+ onto U . Hence the mapping f = f3 ◦ f2 ◦ f1 : Ω → U is the
required mapping. Explicitly, we have
exp( iπ
2 z) − 1
f (z) = iπ
.
exp( 2 z) + 1
(b) The inverse f −1 : U → Ω is given by

2i 1+z 2 1 + z 2i 1+z
f −1 (z) = − log = arg − log . (14.18)
π 1−z π 1−z π 1−z
If f −1 = u + iv, then we have
2 1 + z 2 1+z
u(z) = arg and v(z) = − log .
π 1−z π 1−z
Since f −1 ∈ H(U ), u must be bounded and harmonic in U . Its harmonic conjugate v
is unbounded in U because |v(z)| → ∞ as z → ±1. It remains to prove that u can
be extended continuously to U . The definition of f shows that it can be extended to a
continuous function of Ω onto U . Hence its inverse, which is u, can also be extended to a
continuous function on U .
(c) Since our f −1 and g satisfy the hypotheses of Problem 14.10, we establish immediately
that
g D(0; r) ⊆ f −1 D(0; r) (14.19)
for all 0 < r < 1. Of course, we observe from the definition (14.18) that
π 1+z
− Im f −1 (z) = log (14.20)
2 1−z
∞
X
for all z ∈ U . Let f −1 (z) = cn z n , where z ∈ U . Since n!cn = (f −1 )(n) (0) for every
n=0
n = 0, 1, 2, . . ., it is easy to see that f −1 has the form
∞
4i X z 2n−1
f −1 (z) = −
π 2n − 1
n=1
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 2n − 1 π 1 − |z| π 1−r
n=1
Finally, the set relation (14.19) asserts that

2 1+r
|g(z)| ≤ log
π 1−r
for all z ∈ D(0; r).
(d) Now suppose that Ω = {x + iy | − π2 < y < π2 } and h : Ω → Ω is a conformal bijective

mapping such that h(a + iβ) = 0. It is known from Problem 14.32 that the mapping
1+z
ψ(z) = log
1−z
sends U conformally, one-to-one and onto the horizontal strip Ω with ψ(0) = 0. Thus its
inverse
ez − 1
ψ −1 (z) = z
e +1
is a conformal one-to-one mapping of Ω onto U and ψ −1 (0) = 0. Denote ψ −1 (α + iβ) = A
so that
eα+iβ − 1
= A. (14.21)
eα+iβ + 1
It is well-known [9, Theorem 13.15, p. 183] that the conformal mapping ϕ of U onto itself
and ϕ(A) = 0 is represented by
z−A
ϕ(z) = eiθ ·
1 − Az
for some θ ∈ R. Therefore, the composition h = ψ ◦ ϕ ◦ ψ −1 is a conformal one-to-one
mapping from Ω onto itself and

h α + iβ = ψ ϕ ψ −1 (α + iβ) = ψ ϕ(A) = ψ(0) = 0.
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× × . (14.22)
1 − |A|2 (eα+iβ + 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.
Consequently, we have completed the proof of the problem.
Remark 14.1
Problem 14.9 contributes to the theory of the Principle of Subordination, read [15, Chap.
VI, §5, pp. 207 - 215 ], [21, §1.5, pp. 10 – 13] or [47, Chap V, §9, pp. 226 – 236].
Problem 14.10
Proof. Since f is one-to-one, f −1 : Ω → U exists. Let h = f −1 ◦ g. Then it is clear that

h ∈ H(U ), h(U ) = f −1 g(U ) ⊆ U and h(0) = 0. Thus Theorem 12.2 (Schwarz’s Lemma)
ensures that
|h(z)| ≤ |z| (14.23)
for all z ∈ U . For every 0 < r < 1, if z ∈ D(0; r), then the inequality (14.23) implies that
h(z) ∈ D(0; r). Equivalently, this means that g(z) ∈ f D(0; r) and henceb
g(D(0; r)) ⊆ f (D(0; r)),
Problem 14.11
Proof. Now we have Ω = {z ∈ U | Im z > 0}. Using [9, 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
Problem 14.12
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 [61, Definition 1.3, p. 3].
• f is either one-to-one or constant in Ω when u(z) ≥ 0 for all z ∈ Ω. Since f ∈ H(Ω),

u has continuous derivative of all orders. Thus the set
S = {z ∈ Ω | u(z) = 0} = u−1 (0) (14.26)
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
iθ iθ
u(p + Re ) dθ = u(p + Re ) 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 [9, 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.
• The condition “convex” cannot be replaced by “simply connected”. The fol-

lowing example can be found in [30].c Take β = π2 + δ for very small δ > 0. Define
1+ π
Ω = {z ∈ C | − β < arg z < β} and f (z) = z 2β . Then it is clear that f ∈ H(Ω) and
π
π
f ′ (z) = (1 + 2β )z 2β so that
π 2βπ
π arg z
Re f ′ (z) = 1 + |z| cos . (14.28)
2β 2β
Since −β < arg z < β, we have − π2 < π arg z π
2β < 2 and it follows from the expression (14.28)
that Re f ′ (z) > 0 for every z ∈ Ω. Suppose that z1 = reiθ1 and z2 = reiθ2 are points of Ω,
where θ1 6= θ2 . Clearly,
π π
1+ 1+
z1 2β = z2 2β . (14.29)
π

if and only if exp i(θ1 − θ2 )(1 + 2β ) = 1 if and only if
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 [71, 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).
Problem 14.13
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 ) .
By Theorem 10.18, we conclude immediately that f is constant in Ω.

If fn (z) = n1 ez for every n ∈ N, then each fn is entire and one-to-one in C. Since fn (z) → 0
pointwise in C, we have f ≡ 0. On each compact set K of C, since ez is bounded on K, fn → f
1
uniformly on K. In this case, our limit function f is a constant. Next, we consider fn (z) = ez+ n
in C, then each fn is entire and one-to-one in C. Besides, it is easily checked that fn → f = ez
uniformly on every compact subset of C. In this case, we have f (z) = ez which is one-to-one in
C. This completes the proof of the problem.
Problem 14.14
Proof. Let’s answer the questions step by step.
• 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
lim f (xn + iy ′ ) = L. (14.31)

n→∞
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
F (x) = lim fnk (x) = lim f (x + xnk ) = 0.

k→∞ k→∞
On the other hand, it follows from the limit (14.31) that
F (iy ′ ) = lim fnk (iy ′ ) = lim f (xnk + iy ′ ) = L.

k→∞ k→∞
Thus it is a contradiction and we obtain our desired result.

• 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.
14.2 Problems on Normal Families and the Class S
Problem 14.15
Proof. Let Π be the right half plane. Recall that the ϕ given by [62, 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
g(0) = ϕ−1 (f (0)) = ϕ−1 (1) = 0.

d
Recall from Theorem 11.32 (Fatou’s Theorem) that our g has radial limits almost everywhere on T .
14.2. Problems on Normal Families and the Class S 113
According to Theorem 12.2 (Schwarz’s Lemma), we always have
|g(z)| ≤ |z| (14.34)
for all z ∈ U . Thus the auxiliary family G = {g = ϕ−1 ◦ f | f ∈ F } is uniformly bounded on

each compact subset of U . Particularly, Theorem 14.6 (Montel’s Theorem) implies that G is a
normal family.
Let K be a compact subset of U . Then there exists a constant 0 < R < 1 such that
K ⊆ D(0; R), so the inequality (14.34) gives |g(z)| ≤ R forevery g ∈ G and all z ∈ K. Since
ϕ is conformal, it is continuous on U . Therefore, ϕ D(0; R) is bounded by a positive constant
M and then
f (z) = ϕ g(z) ∈ ϕ D(0; R) ⊆ D(0; M ) (14.35)
for all z ∈ K. As g runs through G , f runs through F . Hence, it yields from the result (14.35)
that F is uniformly bounded on K. Again, Theorem 14.6 (Montel’s Theorem) implies that F
is normal.
The condition “f (0) = 1” can be omitted or replaced by “|f (0)| ≤ 1”. In fact, we suppose
that F ′ = {f ∈ H(U ) | Re f > 0} and the auxiliary family G ′ = {g = e−f | f ∈ F ′ }. It is
evident that
1
|g(z)| = |e−f (z) | = Re f (z) ≤ 1
e
for all z ∈ U , the family G ′ is uniformly bounded on U . Using similar argument as the previous
paragraph, it can be shown that the family F ′ is also normal. This completes the proof of the
problem.
Problem 14.16
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
Applying Theorem 3.5 (Hölder’s Inequality) to the expression (14.36), we obtain

Z R Z 2π
1
|f (z)| = rf (z + reit ) dt dr
πR2 0 0
Z Z o 1 n Z R Z 2π √
1 n R 2π √ 2 2 it
2 o1
2
≤ ( r) dr dt · r|f (z + re )| dr dt
πR2 0 0 0 0
n ZZ o1
1 √ 2
= · πR · |f (z)|2 dx dy
πR2
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
K ⊆ D(p1 ; R1 ) ∪ D(p2 ; R2 ) ∪ · · · ∪ D(pN ; RN ).
If R = min(R1 , R2 , . . . , RN ), then we conclude from the inequality (14.37) that F is uniformly

1
bounded by √πR on K. Hence Theorem 14.6 (Montel’s Theorem) shows that F is a normal
family and we complete the proof of the problem.
Problem 14.17
Proof. The conclusion is affirmative. To this end, we need the following version of Hurwitz’s
Theorem [18, p. 152]:
Lemma 14.3 (Hurwitz’s Theorem)

Let Ω be a region and fn ∈ H(Ω) for n = 1, 2, . . .. If {fn } converges to f uniformly
on compact subsets of Ω, f 6≡ 0, D(a; R) ⊆ Ω and f (z) 6= 0 on C(a; R), then there
is an N ∈ N such that f and fn have the same number of zeros in D(a; R) for all
n ≥ N.
Fix an z0 ∈ Ω and let a ∈ Ω. Then the functions gn = fn − fn (z0 ) converge uniformly to

g = f − f (z0 ) on compact subsets of Ω. Since f is one-to-one in Ω, g(z) 6= 0 on C(a; r) for
every r > 0 such that D(a; r) ⊆ Ω. By Lemma 14.3 (Hurwitz’s Theorem), there corresponds an
N (a, r) ∈ N such that gn and g have the same number of zeros in D(a; r) for all n ≥ N (a, r). In
other words, we obtain
fn (z) 6= fn (z0 ) (14.38)
in D(a; r) \ {z0 } and for all n ≥ N (a, r). Suppose that K ⊆ Ω is compact, p ∈ K and {D(a; r)}
is an open covering of K, where each D(a; r) is a subset of Ω. Then we have
K ⊆ D(a1 ; r1 ) ∪ D(a2 ; r2 ) ∪ · · · ∪ D(am ; rm )
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
Proof. Suppose that f, g : Ω → U and f (z0 ) = g(z0 ) = 0. Then F = g ◦ f −1 : U → U is a

bijective conformal mapping such that
g′ (z0 )
F (0) = g f −1 (0) = g(z0 ) = 0 and F ′ (0) = ′ . (14.40)
f (z0 )
By [9, Theorem 13.15, p. 183], we know that
z−α
F (z) = eiθ · (14.41)
1 − αz
g ′ (z0 )
for some |α| < 1 and θ ∈ [0, 2π]. By the conditions (14.40), we have α = 0 and eiθ = f ′ (z0 )
which imply that
g′ (z0 ) g′ (z0 )
g(z) = ′ f (z) = ′ ϕ0 f (z)
f (z0 ) f (z0 )
for all z ∈ Ω.
For the case that f (z0 ) = g(z0 ) = a, the conditions (14.40) are replaced by
g ′ (z0 )
F (a) = a and F ′ (a) = .
f ′ (z0 )
g ′ (z0 ) (1−αa)2
By the form (14.41) again, we can show that eiθ = f ′ (z0 ) · 1−|α|2 which gives
g′ (z0 ) (1 − αa)2
g(z) = ′
· 2
ϕα f (z)
f (z0 ) 1 − |α|
for all z ∈ Ω. This ends the proof of the problem.
Problem 14.19
Proof. We claim that the mapping given by

i
f (z) = z exp (14.42)
1 − |z|
is a homeomorphism of U onto U . If we consider z = reiθ and represent the function (14.42) as

1
f (r, θ) = r, θ + , (14.43)
1−r
then the inverse function f −1 is given by
1
f −1 (r, θ) = r, θ − .
1−r
Now it is easy to see that both f and f −1 are continuous and bijective. In other words, f
is a homeomorphism. However, this homeomorphism cannot be extended to U continuously.
Otherwise, we assume that F : U → U was a continuous extension to f , i.e., F = f on U .
′
Therefore, we have F C(0; 1) = C(0; 1). Take F (eiθ ) = (1, 0) for some θ ′ ∈ [0, 2π] and V any
′
neighborhood of (1, 0). Suppose that i ∈/ V . Then there exists a neighborhood W of eiθ such
′
that F (W ) ⊆ V .e Indeed, we can find a sequence {rn eiθ } in W such that rn < 1 for all n ∈ N,
′
rn → 1 and F (rn eiθ ) = f (rn , θ ′ ) → (1, 0)
e
See, for example, [42, Theorem 18.1, p. 104].
as n → ∞. Using the formula (14.43), we have

1 h 1 i
f (rn , θ ′ ) = rn , θ ′ + = rn exp i θ ′ + . (14.44)
1 − rn 1 − rn
1

Since the function s(r) = θ ′ + 1−r is continuous on [0, 1) and s [0, 1) = [θ ′ + 1, ∞), if we take
1
the points θ ′ + 1−r n
= (2n + 1)π for all large n, then we deduce from the formula (14.44) that
h 1 i
f (rn , θ ′ ) = i 1 −
(2n + 1)π − θ ′
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
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
g(z) = zh(z n ) (14.45)
in U . Then we know that
g n (z) = z n hn (z n ) = z n ϕ(z n ) = f (z n ) (14.46)
for every z ∈ U . It is clear that g(0) = 0 and g ′ (0) = h(0) = 1.

To prove g ∈ L , it suffices to show that g is one-to-one in U . Suppose that z, ω ∈ U and
g(z) = g(ω). Since f is one-to-one in U , the formula (14.46) ensures that z n = ω n or equivalently,
2kπi
z=e n ω,
where k = 0, 1, . . . , n − 1. Put this into the equation (14.45) to get

2kπi 2kπi 2kπi
g(z) = e n ωh(e2kπi ω n ) = e n · ωh(ω n ) = e n g(ω). (14.47)
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
Proof. By Definition 14.10, we have

∞
X
f (z) = z + an z n .
n=2
(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 [62, 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
Consequently, f must be in the form

z
f (z) = .
(1 + eiθ z)2
Problem 14.22
Proof. Let f : U → S be a one-to-one conformal mapping, where S is a square with center at 0.

Now it is clear that both if : U → S and f −1 : S → U are conformal. We consider
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
if (z) = f (λz) (14.48)

f
Or it can be seen directly from [9, Theorem 13.8, p. 174].
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
if (z) = f (iz) (14.49)

P
for every z ∈ U . Suppose that f (z) = cn z n . By the expression (14.49), we have
∞
X
icn (1 − in−1 )z n = 0
n=1
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
considering g = f −1 ◦ [exp( 2πiN )f ] : U → U which is clearly a one-to-one conformal mapping of

U onto itself. Obviously, we have
2πi
g(0) = f −1 exp f (0) = 0.
N
We observe from Remark 10.3 that
exp( 2πi ′
N )f (z)
g′ (z) = ,
f ′ exp( 2πi
N )f (z)
so |g ′ (0)| = 1. Hence Theorem 12.2 (Schwarz’s Lemma) implies that

2πi
e N f (z) = f (λz) (14.50)
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.
14.3 Proofs of Conformal Equivalence between Annuli
Problem 14.23
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 119
Figure 14.1: The region Ω bounded by C1 and C2 .
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(1 + ab)
α2 − α + 1 = 0.
a+b
Solving this equation to get

r
1 + ab 1 + ab 2
α± = ± − 1. (14.52)
a+b 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.
Remark 14.2
An example of this kind of conformal mappings can be found in [76, Problem 13.20, pp. 182,
183].
Problem 14.24
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
f1 = f and fn+1 = f ◦ fn : A(1; R1 ) → A(1; R2 )
for all n ∈ N. Thus each fn is bijective.

By the definition, we have

f A(1, R1 ) = A(1, R2 ) ⊆ A(1, R1 ) and fn A(1, R1 ) ⊆ A(1, R1 ) (14.53)
for every n = 1, 2, . . ., so the family F is uniformly bounded on compact subsets of A(1, R1 ).

Therefore, it follows from Theorem 14.6 (Montel’s Theorem) that F is normal and then there
exists a subsequence {fnk } converging uniformly on compact subsets of A(1, R1 ) to a holomorphic
function
g : A(1, R1 ) → A(1, R2 ).
Denote Ω = A(R2 , R1 ). We claim that
fn (Ω) ∩ fm (Ω) = ∅, (14.54)
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 121
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
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
Next, by considering the function

f1 (R1 z)
g1 (z) = ,
R2
we see that g1 is holomorphic in the region Ω = {z ∈ C | R1−2 < |z| < 1} such that |g1 (z)| → 1
as |z| → 1 and |g(z)| → R2−2 as |z| → R1−2 . Thus Problem 14.2(b) may be applied to extend g1
to a conformal mapping f2 : A(R1−2 , R12 ) → A(R2−2 , R22 ) and f2 satisfies
lim |f2 (z)| = R2−2 , lim |f2 (z)| = 1 and lim |f2 (z)| = R22 .
|z|→R−2
1
|z|→1 |z|→R21
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.
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 [6].
(b) Besides the analytical proofs provided in the text and the problems, one can find a
pure algebraic proof in [58].
14.4 Constructive Proof of the Riemann Mapping Theorem
Problem 14.26
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
rn = inf{|z| | z ∈ C \ Ωn−1 }. (14.56)
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.
Figure 14.2: The constructions of Ωn−1 , D(0; rn ) and αn .

h
Geometrically, αn is a point on ∂Ωn−1 nearest the origin.
14.4. Constructive Proof of the Riemann Mapping Theorem 123
Choose βn2 = −αn and put
Fn = ϕ−αn ◦ s ◦ ϕ−βn : U → U, (14.57)
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 .
(b) By the definition and Theorem 12.4, we may write
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.
(c) We prove the assertions one by one.
– Each ψn : Ω → Ωn is bijective. By the definition, we have ψn = fn ◦ fn−1 ◦ · · · ◦ f1 .

Since Ω = Ω0 and fn (Ωn−1 ) = Ωn ⊆ U , we have
ψn (Ω) = fn (fn−1 (· · · f1 (Ω0 ))) = fn (fn−1 (· · · f2 (Ω1 ))) = fn (Ωn−1 ) = Ωn ⊆ U.
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)
so we know from Theorem 12.2 (Schwarz’s Lemma) that

|Fn (ω)| ≤ |ω| and |Fn′ (0)| ≤ 1 (14.61)
for all ω ∈ U . Put ω = Gn (z) into the inequality (14.61), we get that
|Gn (z)| ≥ |z| and |G′n (0)| ≥ 1 (14.62)
hold for all z ∈ Ωn−1 .
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→∞
as desired. This ends the proof of the lemma.

z
Now Lemma 14.4 ensures that D(0; r1 ) lies in every Ωn . Since φ(z) = maps D(0; r1 )
r1
one-to-one and onto U , the map Ψn = ψn ◦ φ−1 : U → U satisfies the hypotheses of
Theorem 12.2 (Schwarz’s Lemma) so that |Ψ′n (0)| ≤ 1 which means that
1
|ψn′ (0)| ≤ (14.63)
r1
for every n = 1, 2, . . .. In other words, {ψn′ (0)} is bounded.
– A formula of ψn′ (0). By the aid of the Chain Rule and part (b), we establish easily
that
n
Y
′ ′ ′ ′ 1 + rk
ψn (0) = fn (0) × fn−1 (0) × · · · × f1 (0) = √ . (14.64)
2 rk
k=1
– The sequence {rn } converges to 1. For m > n ≥ 1, we define

ψm,n = fm ◦ fm−1 ◦ · · · ◦ fn+1
which is holomorphic in Ωn and hence on D(0; rn+1 ). In view of the value (14.60),
we know that Gn (0) = 0 and then fn (0) = 0 for every n ∈ N. Recalling the fact
ψn = fn ◦ fn−1 ◦ · · · ◦ f1 , we therefore have ψn (0) = 0 for every n ∈ N which implies
ψm,n (0) = 0. Using similar attack as in proving the inequality (14.63),i we can obtain
′ 1
|ψm,n (0)| ≤ .
rn+1
i
That is, D(0; rn+1 ) lies in every Ωm for all m ≥ n + 1, the map Ψm,n = ψm,n ◦ φ−1 : U → U satisfies the
−1
hypotheses of Theorem 12.2 (Schwarz’s Lemma), where φ(z) = rn+1 z maps D(0; rn+1 ) one-to-one and onto U .
Now the Chain Rule and part (b) assert that

m
Y m
Y
′ 1 + rk
|ψm,n (0)| = |fk′ (0)| = √
2 rk
k=n+1 k=n+1
so that
m
Y 1 + rk 1
1< √ ≤ (14.65)
2 rk rn+1
k=n+1
which implies the convergence of

∞
Y 1 + rn
√
2 rn
n=1
by taking m → ∞. By a basic fact about the convergence theory of infinite products

[9, Note 1, p. 242], we see that
1 + rn
lim √ = 1.
n→∞ 2 rn
Combining this fact and the identity

√
1 + rn (1 − rn )2
√ −1= √ ,
2 rn 2 rn
we conclude immediately that rn → 1 as n → ∞.
(d) We verify the parts one by one.
| for each n ∈ N. By the definition of ψn , we have ψn+1 (z) =

– |hn | ≤ |hn+1
fn+1 ψn (z) so that
zhn+1 (z) = fn+1 zhn (z) . (14.66)
Applying the first inequality (14.62) to the right-hand side of the expression (14.66),
we assert that
|zhn+1 (z)| ≥ |zhn (z)|
which implies our expected result immediately.
– ψn → ψ converges uniformly on compact subsets of Ω. Recall from part
(c) that each ψn is injective on Ω, so every hn is zero-free in Ω and we can define
gn = log hn as a holomorphic function in Ω. Define un = Re gn = log |hn | which is
harmonic in Ω. Since each ψn (Ω) is a subset of U , |hn (z0 )| < ∞ for some z0 ∈ Ω \ {0}
and all n ∈ N. Now the inequality |hn | ≤ |hn+1 | guarantees that
u1 ≤ u2 ≤ · · · .
According to Theorem 11.11 (Harnack’s Theorem), we see immediately that {un }

converges uniformly on compact subsets of Ω. Since ψn′ (z) = hn (z) + zh′n (z), the
second fact in part (c) assures us that {hn (0)} is bounded and therefore {hn (0)}
converges. Consequently, {gn (0)} converges and we conclude from Problem 11.8 that
{gn } converges uniformly on compact subsets of Ω to g. By the definition, we know
that
ψn (z) = zegn (z) ,
so {ψn } also converges uniformly on compact subsets of Ω to ψ. In view of Theorem
10.28, it is true that ψ ∈ H(Ω).
– The map ψ : Ω → U is surjective. Let ω ∈ U . By Lemma 14.4 and the fact

rn → 1 as n → ∞, we may select an N ∈ N such that
|ω| < rn
for all n ≥ N . Therefore, for p ≥ 1, we have
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)
which means that there is a zp ∈ Ω such that
ω = fN +p−1 ◦ fN +p−2 ◦ · · · ◦ fN +1 ◦ ψN (zp ) (14.67)

= ψN +p−1 (zp ). (14.68)
Applying the inequality (14.62) repeatedly to the expression (14.67), we see that
|ω| ≥ |ψN (zp )| (14.69)
−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 [79, 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
lim ψN +pj −1 (zpj ) = ψ(z). (14.70)

j→∞
Combining the expression (14.68) and the limit (14.70), we conclude at once that
ω = ψ(z)
which means that ψ is surjective.

– The map ψ : Ω → U is injective. Since ψ is surjective, it is not constant. Recall
that each ψn is injective, so Problem 14.13 indicates easily that ψ is also injective.
Problem 14.27
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
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
By elementary calculus again, we know that log(1+x)

x is strictly decreasing for x > −1. Combining
this fact and Lemma 14.4, it is true that for all n ≥ 1,
log rn+1 log[1 + (rn+1 − 1)] log[1 + (r1 − 1)] log r1

= ≤ =
rn+1 − 1 rn+1 − 1 r1 − 1 r1 − 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
√
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
where B = 3A. Using Lemma 14.4, the inequality (14.75) gives

1 − r 2 1 − rn
2n
n < .
2B B
After some algebra, we can show that it is actually equivalent to
h 1 − r i2 1 − r
2n n
2n <n .
16B 16B
Particularly, for p = 1, 2, . . ., we obtain
h 1 − r p i2 1 − r p−1
2 2
2p < 2p−1
16B 16B
h 1 − r p−1 i2 1 − r p−2
2 2
2p−1 < 2p−2
16B 16B
..
.
h 1 − r 1 i2 1 − r
2 1
21 <
16B 16B
and they certainly imply that
h 1 − r p i2p 1 − r1
2
2p < .
16B 16B
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
[79, 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
[49].
(b) For more properties of the so-called Koebe mapping and details of the constructive
proof of the Riemann Mapping Theorem, please refer to [52, §3, pp. 15, 16, 183 – 186]
and [10, §1.7, pp. 180 – 214] respectively.
Problem 14.28
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)| = 2
= = p > 1.
2(1 − |βn | )|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
where m > n ≥ 1. By the right-most inequality (14.76), we may write

1 + rn
= |αn | + δn ,
2
1−rn
where 0 ≤ δn < 2 . Since we have
1 + |αk | 1 + rk + δk
p = √ ,
2 |αk | 2 rk + δk
we can show that the infinite product

Y∞
1 + rn + δn
√
n=1
2 rn + δn
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
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
fk′ (0) = [f ′ (0)]k (14.79)
for some k ∈ N. By the definition, we have

′

fk+1 (0) = f ′ fk (0) · fk′ (0). (14.80)
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
fk+1 (0) = [f ′ (0)]k+1 .
Hence the claim follows from induction.

Since Ω is bounded, there exists a positive constant M such that |z| ≤ M for all z ∈ Ω.
Since Ω is a region, one can find a r > 0 such that D(0; r) ⊆ Ω. By the hypothesis
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
′
[fn′ (0)] n
n
|f (0)| = ≤
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
fn (z) = z + cN z N gn (z) and gn (0) = n. (14.83)
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.
Otherwise, we need the following form of the Kronecker’s Approximation Theorem [5,
Theorem 7.8, p. 149]:
Lemma 14.5 (The Kronecker’s Approximation Theorem)

Given any real α, any irrational β and any ǫ > 0, there exist integers m and n with
n > 0 such that
|nβ − m − α| < ǫ.
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
nk θ 1
− 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
z = g(z) = lim fnk (z) = lim fnk (ω) = g(ω) = ω.

k→∞ k→∞
Thus f is one-to-one. If p ∈ Ω \ f (Ω), then p ∈ Ω \ fn (Ω) for every n ∈ N because we

always have fn (Ω) ⊆ f (Ω). Particularly, the function fnk (z) − p 6= 0 in Ω. However,
fnk (p) − p → g(p) − p = 0 as k → ∞. By Problem 10.20 again, we establish that g(z) = p
for all z ∈ Ω which certainly contradicts the result (14.84). Hence we have f (Ω) = Ω, i.e.,
f is onto.
Problem 14.30
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.
(a) If α, β and γ are distinct, then the definition gives
(z − α)(β − γ) (β − γ)z − α(β − γ)

ϕ(z) = [z, α, β, γ] = = . (14.85)
(z − γ)(β − α) (β − α)z − γ(β − α)
By direct computation, we see that
−γ(β − γ)(β − α) + α(β − γ)(β − α) = (β − γ)(β − α)(α − γ) 6= 0
which means ϕ ∈ Λ in this case. Next, if α = ∞ and β 6= γ, then we have
0 · z + (β − γ)
ϕ(z) = [z, ∞, β, γ] = (14.86)
z−γ
so that 0 · (−γ) − 1 · (β − γ) 6= 0. Thus we also have ϕ ∈ Λ in this case.j

Furthermore, the representations (14.85) and (14.86) imply easily that ϕ maps {α, β, γ}
to {0, 1, ∞}.
(b) Suppose that {α, β, γ} and {a, b, c} are two groups of distinct complex numbers. By the
definition (14.85), we have
(b − c)ω − a(b − c) (β − γ)z − α(β − γ)

=
(b − a)ω − c(b − a) (β − α)z − γ(β − α)
which implies
a(b − c)(β − α)(z − γ) − c(b − a)(β − γ)(z − α)

ω=
(b − c)(β − α)(z − γ) − (b − a)(β − γ)(z − α)

a(b − c)(β − α) − c(b − a)(β − γ) z + cα(b − a)(β − γ) − aγ(b − c)(β − α)
=
(b − c)(β − α) − (b − a)(β − γ) z + α(b − a)(β − γ) − γ(b − c)(β − α)

ab(β − α) + ac(α − γ) + bc(γ − β) z + bcα(β − γ) + acβ(γ − α) + abγ(α − β)
=
b(γ − α) + a(β − γ) + c(α − β) z + bβ(α − γ) + aα(γ − β) + cγ(β − α)
= ϕ(z). (14.87)
Direct checking gives the determinant of ϕ as follows

ab(β − α) + ac(α − γ) + bc(γ − β) · bβ(α − γ) + aα(γ − β) + cγ(β − α)

− bcα(β − γ) + acβ(γ − α) + abγ(α − β) · b(γ − α) + a(β − γ) + c(α − β)
= ab2 β(α − γ)(β − α) + abcβ(α − γ)2 + b2 cβ(γ − β)(α − γ) + a2 bα(γ − β)(β − α)
+ ac2 α(γ − β)(α − γ) + abcα(γ − β)2 + abcγ(β − α)2 + ac2 γ(β − α)(α − γ)
+ bc2 γ(γ − β)(β − α) − b2 cα(γ − α)(β − γ) − abcβ(γ − α)2 ab2 γ(γ − α)(α − β)
− abcα(β − γ)2 − a2 cβ(β − γ)(γ − α) − a2 bγ(α − β)(β − γ) − bc2 α(β − γ)(α − β)
− ac2 β(α − β)(γ − α) − abcγ(α − β)2
= (α − β)(β − γ)(γ − α)(ab2 − b2 c − a2 b + a2 c − ac2 + bc2 )
= (α − β)(β − γ)(γ − α)(a − b)(b − c)(c − a)
6= 0
so that ϕ ∈ Λ in this case.

j
The cases for β = ∞ or γ = ∞ can be done similarly, so we omit the details here.
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
(b − c)ω − a(b − c) β−γ

=
(b − a)ω − c(b − a) z−γ
which gives
a(b − c)z − aγ(b − c) − c(b − a)(β − γ)
ω= = ϕ(z). (14.88)
(b − c)z − γ(b − c) − (β − γ)(b − a)
Now simple algebra shows that
a(b − c)[−γ(b − c) − (β − γ)(b − a)] − (b − c)[−aγ(b − c) − c(b − a)(β − γ)]

= −aγ(b − c)2 − a(b − a)(b − c)(β − γ) + aγ(b − c)2 + c(b − a)(b − c)(β − γ)
= (b − a)(b − c)(c − a)(β − γ)
6= 0
which means ϕ ∈ Λ in this case.

Furthermore, if {∞, β, γ} and {∞, b, c} are two groups of distinct complex numbersl , then
we have
b−c β−γ
=
ω−c z−γ
and so
(b − c)z + cβ − bγ
ω= = ϕ(z). (14.89)
β−γ
Since (b − c)(β − γ) 6= 0, we also have ϕ ∈ Λ.
Finally, by the formulas (14.87), (14.88) and (14.89), it is easy to see that ϕ maps {α, β, γ}
to {a, b, c}.
(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 [76, Problem 13.18, p. 182].
(e) Now we have

[z ∗ , α, β, γ] = [z, α, β, γ]. (14.90)
If C is a straight line, then we choose γ = ∞ in the equation (14.90) to get
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
which means that z and z ∗ lie in different half planes determined by C.

k
The other cases can be done similarly, so we only consider this case and omit the others.
l
Again, we have omitted other similar cases.
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 ∗ .
Figure 14.3: The construction of the symmetric point z ∗ of z.
(f) Let α, β, γ ∈ C. Then it follows from part (c) and the definition that
[ϕ(z ∗ ), ϕ(α), ϕ(β), ϕ(γ)] = [z ∗ , α, β, γ] = [z, α, β, γ] = [ϕ(z), ϕ(α), ϕ(β), ϕ(γ)].
Hence ϕ(z ∗ ) and ϕ(z) are symmetric with respect to ϕ(C).
This finishes the analysis of the problem.
Problem 14.31
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
and its determinant is

(αa + βc)(γb + δd) − (γa + δc)(αb + βd) = (ad − bc)(αδ − βγ) 6= 0
so that ψ ◦ ϕ ∈ Λ.
– Associativity. Simple algebra verifies
[ϕ(z) + φ(z)] + ψ(z) = ϕ(z) + [φ(z) + ψ(z)].
– 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 [25, 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
In conclusion, ϕ has either one or two fixed points on S 2 .
(c) We consider two cases.
– 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
which carries z1 to ∞. Therefore, the linear fractional transformation
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)
By combining the expressions (14.95) and (14.96), we conclude that
ϕ1 = (P −1 ◦ S) ◦ ϕ ◦ (S −1 ◦ P ) = (S −1 ◦ P )−1 ◦ ϕ ◦ (S −1 ◦ P ).
Hence we have ψ = S −1 ◦ P . Since P, S ∈ Λ, we have ψ ∈ Λ.

– Case (ii): ϕ has two distinct fixed points. Consider the linear fractional trans-
formation
φα (z) = αz,
where α is a non-zero complex number which will be determined soon. By part (b),
we have either “c = 0 and a 6= d” or “c 6= 0 and a + d 6= ±2”.
∗ Subcase (i): c = 0 and a 6= d. We notice that
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
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
(az2 + b)z − (az1 + b)

ϕ S −1 (z) = .
(cz2 + d)z − (cz1 + d)
and
(az − dz1 + 2b)z + cz12 + (d − a)z1 − b
T (z) = 22 . (14.98)
− cz2 + (d − a)z2 − b z − (az1 − dz2 + 2b)
Since z1 , z2 are roots of the equation (14.92), the formula (14.98) simplifies to
(az2 − dz1 + 2b)
T (z) = − z. (14.99)
(az1 − dz2 + 2b)
Using the formula (14.94), the expression (14.99) can further reduce to
p
(a + d)2 − 4 + (a + d)
T (z) = − p z.
(a + d)2 − 4 − (a + d)
Hence we obtain the formula
p
(a + d)2 − 4 + (a + d)
α = −p . (14.100)
(a + d)2 − 4 − (a + d)
Finally, α is determined by either (14.97) or (14.100).
(d) We prove the assertions one by one.
– 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
b − dα = cα2 − aα = α(cα − a). (14.102)

Obviously, cα − a 6= 0. Otherwise, put cα = a into the equation (14.101) will give

a + d = 0 which is impossible. Now we note that
1 1 cz + d cz + d
= az+b
= = (14.103)
ϕ(z) − α cz+d −α az + b − cαz − dα (a − cα)z + (b − dα)
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)
– Gα is a subgroup of Λ. Let Gα = {ϕ ∈ Λ | ϕ(α) = α} ∪ {id} ⊆ Λ. Let ϕ and φ be

elements of Gα with the corresponding constant βϕ and βφ respectively. Therefore,
we see that
1 1 1
= + βϕ = + βϕ + βφ (14.106)
ϕ φ(z) − α φ(z) − α z − α
so that ϕ ◦ φ ∈ Gα . By the definition, we have id ∈ Gα . Furthermore, since ϕ has α
as its only finite fixed point, so is ϕ−1 . Thus we have
1 1
= − β.
ϕ−1 (z) −α z−α
In other words, it means that ϕ−1 ∈ Gα . By [25, Theorem 5.14, p. 52], Gα is a

subgroup of Λ.
– Gα is isomorphic to (C, +). Define f : Gα → C by f (ϕ) = βϕ , where βϕ is the
complex number satisfying the equation
1 1
= + βϕ . (14.107)
ϕ(z) − α z−α
As the expression (14.106) shows definitely that
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, [25, p. 132].
(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−1 (z) − α 1 z−α

−1
= · .
ϕ (z) − β γϕ z − β
Consequently, these imply that ϕ−1 ∈ Gα,β . Hence Gα,β is a subgroup of Λ.

– Gα,β is isomorphic to (C \ {0}, ×). Define g : Gα,β → C \ {0} by g(ϕ) = γϕ , where
γϕ is the complex number satisfying the equation
ϕ(z) − α z−α
= γϕ · . (14.111)
ϕ(z) − β z−β
The equation (14.110) implies that g(ϕ ◦ φ) = γϕ × γφ so that g is a homomorphism.

If g(ϕ) = 1, then we have a − αc = a − βc so that c = 0 and ϕ takes the form
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 [24, pp. 15, 16].
Next, we apply the fact

√ 1
γ+ √ =a+d
γ
to represent a, c and d in terms of α, β and γ as follows:
α − βγ √ 1
a= γ+√ ,
(1 + γ)(α − β) γ
αγ − β √ 1
d= γ+√ , (14.112)
(1 + γ)(α − β) γ
1−γ √ 1
c= γ+√ .
(1 + γ)(α − β) γ
Finally, we employ the formula αβ = − cb to obtain
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 [20]
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
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)
and ∆ = (α − βγ)(αγ − β) + αβ(1 − γ)2 = γ(α − β)2 6= 0. Now we have

√
(α − βγ)[(1 − γ)z + (αγ − β)] − ∆ ∆
(1 − γ)ϕ(z) = = (α − βγ) − ,
(1 − γ)z + (αγ − β) ζ
where
(1 − γ)z + (αγ − β)
ζ= √ . (14.120)
∆
1
Define ϕ∗ = ∆− 2 [(1 − γ)ϕ + (αγ − β)]. Then we have
h √ i
∗ − 12 ∆ 1+γ 1
ϕ (ζ) = ∆ (α − β)(1 + γ) − = √ − . (14.121)
ζ γ ζ
Similar to Case (i), this change of variables establishes a one-to-one correspondence

between the invariant circles of the linear fractional transformations (14.119) and
(14.121). Instead of the expressions (14.118), we have
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
R2 = ρ2 + 1 and χ = iρ sin θ (14.122)
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.
∗ Subcase (i): χ2 < 0. Here χ = it, where t = | 21+γ

√ |. By the equations (14.122),
γ
∗
the invariant circle of ϕ has the form
p
ζ − ± R2 − t2 − 1 + it = R,
where R2 ≥ 1 − χ2 = 1 + t2 . Transforming back to the original system (using

(14.120)), we get
(1 − γ)z + (αγ − β) p
√ − ± R2 − t2 − 1 + it = R
∆
√ √ p
(β − αγ) + ∆ ± R2 − t2 − 1 + it R|α − β| |γ|
z− =
1−γ |1 − γ|
√ q
p
(β − αγ) + ∆ ± R2 − | 21+γ√ |2 − 1 + i| 1+γ
γ 2 γ|
√
R|α − β| |γ|
z− = .
1−γ |1 − γ|
∗ Subcase (ii): χ2 = 0. In this subcase, we know that γ = −1. Furthermore, it

can be seen from the expressions involving χ in (14.122) and (14.123) that ρ = 0
and then R = 1. Consequently, we have |ζ| = 1 which gives
α+β |α − β|
z− = .
2 2
∗ Subcase (iii): χ2 > 0. Then χ = 21+γ
√ is real and we get from the expression
γ
(14.123) that p
ζ − χ ± i R2 + 1 − χ2 = R,
where R2 ≥ χ2 − 1. Hence, after transforming back to the original system, we
assert that
√ q
p
√ ± i R2 + 1 − ( 1+γ
(β − αγ) + ∆ 21+γ
γ
√ )2
2 γ R|α − β| |γ|
z− = ,
1−γ |1 − γ|
(1+γ)2 (1−γ)2
where R2 ≥ 4γ −1 = 4γ .
Now we have completed the analysis of the problem.
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 [24, pp. 15 – 23] for further details.
Problem 14.32
Proof. We notice from §14.3 that the linear fractional transformation

1+z
ω = ϕ(z) =
1−z
is a conformal one-to-one mapping of U onto the open right half plane
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
S = {ζ = u + iv ∈ C | u ∈ R and − π2 < v < π2 }.

π
Furthermore, the positive Y -axis is mapped onto the line v = 2 and the negative Y -axis is
mapped onto the line v = − π2 . Consequently, the mapping
1+z
ψ(z) = φ ϕ(z) = log (14.124)
1−z
sends U conformally onto the horizontal strip S, the images of the upper semi-circle and the
lower semi-circle under ψ are the lines v = π2 and v = − π2 respectively. See Figure 14.4 for the
illustration.

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}
and the lines v = ± π2 onto the lines X = ∓ π2 respectively.
(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.
π π
Figure 14.5: The conformal mapping f : U → A(e− 2 , e 2 ).
(b) Refer to part (a).
(c) Refer to part (a).
(d) The circular arc in U from −1 to 1 can be parametrized as
x = cos t and y = b sin t,
where b ∈ (−1, 0) ∪ (0, 1) and t ∈ (0, π) ∪ (π, 2π). Straightforward computation gives
1+z 1 + cos t + ib sin t 1 − cos t + ib sin t

= ×
1−z 1 − cos t − ib sin t 1 − cos t + ib sin t
sin2 t − b2 sin2 t + 2ib sin t
=
(1 − cos t)2 + b2 sin2 t
(1 − b2 ) cos2 2t b cos 2t
= +i· .
sin2 2t + b2 cos2 2t sin 2t (sin2 2t + b2 cos2 2t )
which implies
1+z 1 1+z 2 1+z
log = log + i arg
1−z 2 1 − z 2 2 2 t 1 −2z t t
1 (1 − b ) sin 2 cos 2 + b2 cos2 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].
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:
Figure 14.6: The locus of f (z) for t ∈ (0, π).
Similarly, Figure 14.7 shows the locus of f (z) for t ∈ (π, 2π):
Figure 14.7: The locus of f (z) for t ∈ (π, 2π).
(e) On the radius [0, 1), z is real so that log 1+z

1−z is also real. In this case, |f (z)| = 1 for every

z ∈ [0, 1), so f [0, 1) starts at z = 1 and runs through the unit circle T anticlockwise
infinitely many times.
(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
cot θ
through [− π2 , π2 ]. 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.
Figure 14.8: The image of f (E) when r = 0.25.
(g) Suppose that γ : [0, 1] → U is a curve in U such that γ(t) → 1 as t → 1. Generally

speaking, f γ(t) is also a curve in C and we have to study the behaviour of f γ(t) as
t → 1. The fact γ(t) → 1 as t → 1 implies that γ(t) ∈ D(r; 1 − r) for any 0 < r < 1 and
for all t sufficiently close to 1. Furthermore, γ(t) → 1 as t → 1 also implies that θ(t) → 0
so that
cot θ(t) π
exp − arg 2
→ e− 2
r
as t → 1. Therefore, it follows from part (f) that

π
lim f γ(t) = e− 2 .
t→1
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 [65, p.
206].
Problem 14.33
Proof.
(a) We may write ϕα = u + iv = (u, v), so we may think of ϕα : U → U is given by

ϕα (x, y) = u(x, y), v(x, y)
which is an invertible C 2 mapping. Now the Jacobian of ϕα is

 ∂u ∂u 
 ∂x ∂y 
 
Jϕα (x, y) =  .
 ∂v ∂v 
∂x ∂y
Therefore, it follows from this and the Cauchy-Riemann equations that
∂u ∂v ∂v ∂u ∂u 2 ∂v 2 2
det Jϕα (x, y) = · − · = + = ϕ′α (x, y) 6= 0
∂x ∂y ∂x ∂y ∂x ∂x
for all (x, y) ∈ U . Finally, we follow from the Change of Variables Theorem [61, Theorem
10.9, pp. 252, 253] that
Z Z Z Z
′ 2 2
π= 1 dx dy = det Jϕα (x, y) dx dy = ϕα (x, y) dx dy = ϕ′α dm
U U U U
which is exactly our required result.
(b) An easy computation gives
1 − |α|2
ϕ′α (z) =
|1 − αz|2
1 − |α|2
=
(1 − αz)(1 − αz)
nX
∞ o nX
∞ o
= (1 − |α|2 ) · (αz)n · (αz)k (14.126)
n=0 k=0
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 1 − |α|2
ϕ′α dm = (αz)n (αz)k dm
π U π U n,k=0
∞ Z
1− |α|2 X
= (αz)n (αz)k dm
π
n,k=0 U
∞ Z 2π Z
1− |α|2 X 1
= (αreiθ )n (αre−iθ )k r dr dθ
π
n,k=0 0 0
∞ Z Z
1− |α|2 X 2π 1
= · (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 ∞ Z Z
1 1 − |α|2 X 2n 2π 1 2n+1
ϕ′α dm = · |α| r dr dθ
π U π n=0 0 0
∞
1 − |α|2 X 2n π
= · |α|
π n+1
n=0
∞
2
X |α|2n
= 1 − |α| . (14.128)
n+1
n=0
Consider the power series expansion of log(1 + z) about z = 0, we know that

∞
X z n+1
log(1 + z) = (−1)n
n=0
n+1
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 1 − |α|2 1
ϕ′α dm = 2
log .
π U |α| 1 − |α|2
We have ended the proof of the problem.
Remark 14.7
Problem 14.33(a) is a special case of the classical result Lusin Area Integral, see [34, p.
150].
CHAPTER 15
Zeros of Holomorphic Functions
15.1 Infinite Products and the Order of Growth of an Entire Function
Problem 15.1
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
for every z ∈ S. By Theorem 15.4, the infinite product
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.
149
150 Chapter 15. Zeros of Holomorphic Functions
Problem 15.2
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 | ≤ n (15.2)
r
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 1
at r = n λ 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
z
Proof. The part of finding solutions of ee = 1 has been solved in [76, 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)

 3π
 ln(−2kπ) + i 2nπ + , 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 151
Figure 15.1: The distribution of the zeros zk,n of exp(exp(z)).
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

• 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
which gives the desired claim. Since
z z 2z
+ = 2 ,
n(z − n) (−n)(z + n) z − n2
the function f can be expressed as
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 [9, 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
so the function f also has the form
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
/ Z, then it is obvious to check that

• Both functions are periodic. If z ∈
eπi(z+1) + e−πi(z+1) eπiz + e−πiz
π cot π(z + 1) = πi · = πi · = π cot πz.
eπi(z+1) − e−πi(z+1) eπiz − e−πiz
By the representation (15.4), we have
1 Xh 1 1i
f (z + 1) = + +
z+1 z − (n − 1) n
n∈Z
n6=0
1 X 1 1
= + +
z+1 z−n n+1
n∈Z
n6=−1
1 1 X 1 1 1 1
= +1+ + + − +
z+1 z z−n n n n+1
n∈Z
n6=−1,0
1 1 X 1 1 X 1 1
= +1+ + + − −
z+1 z z−n n n n+1
n∈Z n∈Z
n6=−1,0 n6=−1,0
1 1 X 1 1
= −1+ + +
z+1 z z−n n
n∈Z
n6=−1,0
1 X 1 1
= + +
z z−n n
n∈Z
n6=0
= f (z).
Hence both functions are periodic.

• Their difference is a constant. Now the function ∆(z) = π cot πz − f (z) must be
entire and periodic (i.e., ∆(z + 1) = ∆(z)). To show that it is bounded, it is enough to
show that ∆(z) is bounded in the strip |Re z| ≤ 21 . By Theorem 10.24 (The Maximum
Modulus Theorem), ∆ is bounded in the rectangle {z ∈ C | |Re z| ≤ 21 and |Im z| ≤ 1}. Let
z = x + iy, where |x| ≤ 12 and |y| > 1. On the one hand, we have
e−2πy + e−2πix
cot πz = i ·
e−2πy − e−2πix
so that
|e−2πy | + 1
| cot πz| ≤ < ∞.
|e−2πy | − 1
On the other hand, we write
X ∞
1 2(x + iy)
f (z) = + .
x + iy x − y 2 − n2 + 2ixy
2
n=1
√
If y > 1 and |x| ≤ 21 , 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 +
|x2 − y 2 − n2 + 2ixy|
n=1
∞
√ X y
≤1+4 2
y2 + 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 2
=
0 y +x 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 −y 2 − n2
n=1
∞
X y
= −2i lim
y→∞ y2 + n2
n=1
Z ∞
dt
= −2i
0 1 + t2
= −πi
and
lim π cot iπy = −πi.
y→∞
Thus ∆(z) ≡ 0 and then

∞
1 X 2z
π cot πz = + . (15.6)
z z 2 − n2
n=1
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
Next, we consider the infinite product

∞
Y z2
P (z) = πz 1− 2 . (15.8)
n=1
n
2
Now each Pn (z) = 1 − nz 2 ∈ H(C \ Z) and Pn 6≡ 0 in C \ Z. In addition, if K is a compact
subset of C \ Z, then the Weierstrass M -test [61, Theorem 7.10, p. 148] guarantees that
∞
X ∞
X |z|2
|1 − Pn (z)| =
n2
n=1 n=1
converges uniformly on K. By Theorem 15.6, the product (15.8) is holomorphic in C \ Z.

Using [18, Exercise 10, p. 174], we have
∞ ∞ ∞
P ′ (z) 1 X Pn′ (z) 1 X − n2z2 1 X 2z
= + = + = + (15.9)
P (z) z n=1 Pn (z) z n=1 1 − z 22 z n=1 z 2 − n2
n
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)
and then it gives

h P (z) i′
= 0.
g(z)
P (z)
Therefore, we have P (z) = cg(z) for some constant c in C \ Z. Since z → 1 as z → 0,
we have c = 1 and eventually,
∞
Y z2
sin πz = πz 1− 2 (15.10)
n=1
n
∞
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
where ζ ∈ C and the zk denotes a pole of g(z) = cot πz

z−ζ inside CN . Take ζ ∈ / Z and ζ to be
a point inside CN . Then it is clear that the poles of the function g(z) occur at z = ζ and
at z = n ∈ {−N, −N + 1, . . . , 0, 1, . . . , N }, and they are all simple. By the basic method of
evaluating residue [9, p. 129], we know that
cot πz cot πζ cot πz 1
Res ;ζ = = cot πζ and Res ;n = .
z−ζ 1 z−ζ π(n − ζ)
By putting these values into the equation (15.11), we get

Z N
X
1 cot πz 1
dz = π cot πζ − .
2i CN z−ζ ζ −n
n=−N
It can be shown that Z

cot πz
lim dz = 0
N →∞ CN z−ζ
which implies the desired result. For details, please refer to [54, Lemma 7.22, pp. 181, 182].
Problem 15.5
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
= Re − z k+1 − z k+2 − · · ·
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
log |Ek (z)| < M1 |z|k+1 (15.13)
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
k+1
< and ≤ N.
|zn | 2M |zn |k+1
n=N +1 n=1
Using the inequality (15.15), we obtain

∞
X z ∞
X z k+1 |z|k+1
log Ek ≤M < . (15.16)
zn zn 2
n=N +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 , and it is equivalent to saying that
|f (z)| < exp(|z|k+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
Proof. Given ǫ > 0. Notice that
X ∞
X X X∞

|zn |−p−ǫ = |zn |−p−ǫ ≤ 2−k(p+ǫ) n 2k+1 . (15.18)
|zn |≥1 k=0 2k ≤|z n |<2k+1 k=0
Since |f (z)| < exp(|z|p ), it follows from [62, 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
for every ǫ > 0, completing the proof of the problem.
Problem 15.7
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 [62, 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.
15.2 Some Examples
Problem 15.8
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)
Since the summation is a positive integer, it is true that

nZ o n N
X Z o nZ o
exp g(ζ) dζ = exp 2πi mn + g(ζ) dζ = exp g(ζ) dζ
γ(z) n=1 η(z) η(z)
and this means that f is independent of the choice of γ(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 159
Figure 15.2: The paths γ(z + h) and −γ(z).
Now we consider the difference

Z Z
G(z + h) − G(z) = g(ζ) dζ − g(ζ) dζ
γ(z+h) γ(z)
XZ
m Z
= g(ζ) dζ + g(ζ) dζ (15.20)
k=1 ∂∆k [z,z+h]
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]
the last integral actually tends to 0 as h → 0. Consequently, we have proved that

G(z + h) − G(z)
lim = g(z),
h→∞ h
i.e., G is holomorphic at z. Since z is arbitrary, G ∈ H(C \ A) which implies the desired
result that f ∈ H(C \ A).
• f has a removable singularity at each zn . It suffices to verify that
lim (z − zn )f (z) = 0. (15.22)
z→zn
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

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)
By the definition, we parameterize γ2 (z) : [0, 1] → C as
γ2 (z; t) = z + (1 − t)(zn − δn eiθn − 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)
Z
1
iθn iθn
= mn ln{t[z − (zn − δn e )] − δn e } + h(ζ) dζ
0 γ2 (z)
Z

= 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)
Combining the estimate (15.24) and the expression (15.26), we establish

nZ o nZ o
|f (z)| = exp g(ζ) dζ × exp g(ζ) dζ
γ1 (z) γ2 (z)
M1 +M2
≤e · exp(−mn ln δn ) × |z − zn |mn
which implies the limit (15.22).
• The extension of f has a zero of order mn at zn . Since f has a removable singularity

at each zn , it follows from Definition 10.19 that f can be extended to be holomorphic at
zn . The analysis in the previous part further shows that
nZ o nZ o
f (z) = exp g(ζ) dζ × exp g(ζ) dζ
γ1 (z) γ2 (z)
nZ o nZ o
= exp g(ζ) dζ × exp h(ζ) dζ × δn−mn e−imn (θn +π) (z − zn )mn .
γ1 (z) γ2 (z)
In other words, f has a zero of order mn at zn .

a
See, for example, [9, Corollary 9.11, p. 124].
b
If zn is real, then θn = 0 and z is also real. In this case, the integrated result (15.25) becomes ln |z −zn |−ln δn .
Thus we have completed the analysis of the problem.
Problem 15.9
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
Then g is clearly holomorphic in a neighbourhood V containing U and |g(z)| = 1 on T by

Theorem 12.4. Now the function
f (z)
h(z) =
g(z)
must be holomorphic in U . Since f (U ) ⊆ U , we deduce from Theorem 10.24 (The Maximum
Modulus Theorem) that
|h(z)| ≤ 1
for all z ∈ U . If z = 0, then we get
α
≤1
|z1 | × |z2 | × · · · × |zn |
which implies 0 < α ≤ β n . Consequently, we obtain
log α
n≤ . (15.27)
log β
1
(a) Put α = β = 2 into the inequality (15.27), we conclude that n ≤ 1.
(b) Similarly, by the inequality (15.27) again, we know that n ≤ 2.
(c) In this case, we have n = 0.
(d) In this case, we have n ≤ 3.
Problem 15.10
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
fN1 gN1 + fN2 hN2 gN1 + · · · + fNk hNk gN1 = f gN1 ,

where f is entire. Thus we have
I = {f gN | f is entire and N ≥ 1}. (15.29)
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
z−1
Proof. Recall from [62, 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
∞ ∞ p
X i log n X 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
∞
i log n X
X ∞ ∞
2 2X1
1− ≥ 2
≥ .
n=3
2 + i log n n=3
3(log n) 3 n=3
n
Hence the series (15.31) diverges and thus no such bounded holomorphic function in Π.
(b) In this case, we know from the A.M. ≥ G.M. that

∞ √ ∞ √ √
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
∞
X 1
=2 5 1
n=1 (n + 4) 4 · n 4
X∞
1
<2 3
n=1 n2
< ∞,
so the answer is affirmative.
(c) The answer to this part is affirmative because

∞ ∞ √
X in X 4 + n2 − n
1− = √
2 + in 4 + n2
n=1 n=1
X∞
4
= √ √
4 + n 2 · 4 + n 2+n
n=1
X∞
2
< 2
n=1
n
< ∞.
(d) The answer to this part is affirmative because

∞
in2 X |2 + in2 | − n2 X 2 + n2 − n2 X 2
X ∞ ∞ ∞
1− = ≤ = <∞
2 + in2 |2 + in2 | n2 n2
n=1 n=1 n=1 n=1
We complete the analysis of the problem.
Problem 15.12
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) = ·
n=1
1 − αn z αn
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
αn + |αn |z 1 + |z|
· 1 − |αn | ≤ 1 − |αn | · . (15.34)
(1 − αn z)αn |1 − αn z|
As K is compact, it is bounded by a positive constant M so that |z| ≤ M for all z ∈ K. Now

there exists an N ∈ N such that n ≥ N implies
δ
|αn − e−iθ | ≤ . (15.35)
2M
Combining the estimates (15.32) and (15.35), if n ≥ N , then we gain
δ
δ ≤ |1 − e−iθ z| ≤ |1 − αn z| + |αn − e−iθ | · |z| ≤ |1 − αn z| +
2
so that
δ
>0
|1 − αn z| ≥
2
for all z ∈ K. Thus the inequality (15.34) reduces to
αn + |αn |z 2(1 + M )
· 1 − |αn | ≤ 1 − |αn | · .
(1 − αn z)αn δ
for all z ∈ K and all n ≥ N . Using the Blaschke condition and the Weierstrass M -test, we
conclude immediately that the series (15.33) converges uniformly on K. Eventually, Theorem
10.28 ensures that B ∈ H(Ω), and we end the analysis of the problem.
15.3 Problems on Blaschke Products
Problem 15.13
Proof. Since αn are real, we have

∞
Y αn − z
B(z) = z k
1 − αn z
n=1
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
αn − r αn − r
= <1
1 − αn r 1 − αn r
for every n = N, N + 1, . . .. Consequently, we have
Y∞ N
Y −1 ∞
Y N
Y −1
k αn − r αn − r αn − r r − αn
|B(r)| = r < × < . (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−
N2
n=1
15.3. Problems on Blaschke Products 165
h N
Y −1
n2 i
≤ exp log 1−
N2
n=1
h NX
−1 n2 i
= exp log 1 − 2
n=1
N
N X−1 2
n
≤ exp −
n=1
N2
h (N − 1)(2N − 1) i
≤ exp −
6N
h (N − 1)2 i
< exp −
3N
2 N
−
< e3 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
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 [8, Eqn. (15), p. 12] that we can write
p
Y 4p−1
Y αn − x2p ∞
Y
x2p − αn αn − x2p
|B(x2p )| = × ×
1 − αn x2p 1 − αn x2p 1 − αn x2p
n=1 n=p+1 n=4p
= T1 (p) · T2 (p) · T3 (p). (15.39)
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
lim T1 (p) = 1. (15.40)

p→∞
Furthermore, for sufficiently large p, the inequalities

∞
X n

exp − 8 e− 2 < T3 (p) < 1
n=4p
give
lim T3 (p) = 1. (15.41)
n→∞
However, the function T2 (p) only satisfies
lim T2 (p) > 0.

p→∞
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
{αpk +1 , αpk +2 , . . . , α4pk −1 } ∩ {αpk+1 +1 , αpk+1 +2 , . . . , α4pk+1 −1 } = ∅
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
Thus it follows from Theorem 15.21 that

∞
Y α′k − z
B(z) = (15.43)
1 − α′k z
k=1
is an element of H∞ and has no zeros except at α′k . It is no doubt that |B(α′n )| → 0 as n → ∞,

so
lim inf |B(r)| = 0. (15.44)
r→1
Besides, we apply the representation (15.39) to our Blaschke product (15.43) to obtain
pk
Y 4pY
k −1 ∞
Y
x2pk − α′n α4pk − x2pk α′n − x2pk
|B(x2pk )| = × ×
1 − α′n x2pk 1 − α4pk x2pk 1 − α′n x2pk
n=1 n=pk +1 n=4pk
= T1 (pk ) · S2 (pk ) · T3 (pk ).
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
Proof. As a linear fractional transformation, ϕ is one-to-one and ϕ ∈ H(U ). Since ϕ(U ) = U ,

it follows from Theorem 12.6 that
z−α
ϕ(z) = λ · (15.46)
1 − αz
for some α ∈ U and |λ| = 1.
(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
for some nonzero constant A. Now ψn (z) = z + nA for every z ∈ U , so ψn (z) → ∞

as n → ∞. Since
ϕn = µ−1 ◦ ψn ◦ µ,
it establishes that
(1 + nAb)z − nAb2
ϕn (z) =
nAz + (1 − nAb)
for every z ∈ U . Put z = 0 to get
−nAb2 b
ϕn (0) = =b− ,
1 − nAb 1 − nAb
so for sufficiently large n, we have
1
|ϕn (0)| ≈ 1 − ,
n|A| − 1
but it implies that
1
1 − |ϕn (0)| ≈ .
n|A| − 1
Therefore, such ϕ cannot satisfy the Blaschke condition by [79, Theorem 6.10, p. 77].
– Case (ii): Suppose that ϕ has two distinct fixed points on T . Call them b1 and b2
respectively. Using Problem 14.31(c), we have ψ = ν ◦ ϕ ◦ ν −1 , where
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.
b2 z−b1
Note that ν −1 (z) = z−1 , so
z − b1
ϕn (z) = ν −1 kn ·
z − b2
c
Or we may apply Problem 14.31(c) directly here.
z−b1
b2 k n · z−b2 − b1
= z−b1
kn · z−b2 − 1
(b2 kn − b1 )z + (1 − kn )b1 b2
= . (15.48)
(kn − 1)z + (b2 − b1 kn )
∗ Subcase (i): |k| < 1. Then for sufficiently large enough n, the expression
(15.48) gives
(kn − b1 )z + (b1 − kn )b2

|ϕn (z)| ≈ ≈ 1 − |k|n
−z + b2
so that 1 − |ϕn (z)| ≈ |k|n . In other words, ϕ satisfies the Blaschke condition by
[79, Theorem 6.9, p. 77].
∗ Subcase (ii): |k| > 1. In this case, for sufficiently large enough n, the expression
(15.48) implies that
b1 1
(b2 − k n )z + ( k n − 1)b1 b2
|ϕn (z)| =
(1 − k1n )z + ( kbn2 − b1 )
(b2 − k1n )z + ( k1n − b2 )b1
≈
z − b1
1
≈1−
|k|n
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 (15.48), we have
1 1
ϕN =
2 2
so that
∞ h
X 1 i ∞ h
X 1 i
1 − ϕn ≥ 1 − ϕpN = ∞.
2 2
n=1 p=1
Thus ϕ does not satisfy the Blaschke condition in this case.

∗ Subcase (iv): |k| = 1 and k is not an nth root of unity for all n. We
claim that the set S = {kn | n ≥ 0} is dense on T . To see this, we write k = eiθ ,
where θ is an irrational multiple of 2π.d Therefore, we obtain
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
e(p−q)iθ < <ǫ
N
for some p < q. This means that the points en(p−q)iθ with n ≥ 0 are distributed
on T at successive angles less then ǫ. Consequently, we have en(p−q)iθ ∈ ℓ for
some n ≥ 0.
2πp
d
Otherwise, we have θ = q
for some p ∈ Z and q ∈ N. This implies that kq = 1, a contradiction.
Put z = 0 into the expression (15.48) to get

|1 − kn |
|ϕn (0)| = .
|b2 − b1 kn |
Now the claim ensures that there exists a sequence {nm } such that |b1 − b2 knm | <
1
m and then
m
|ϕnm (0)| > m · |1 − knm | >
2
for large enough m. This shows that ϕ does not satisfy the Blaschke condition.
(b) We claim that ϕ satisfies

ϕn (z) = z (15.49)
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

(15.49) holds immediately for N . Otherwise, we fix z = z0 ∈ U \ {0} so that
f (λn z0 ) = f (z0 )
for every n ≥ 0. By Subcase (iv), the set S ′ = {λn | n ≥ 0} is dense on T which

implies that
f (z) = f (z0 ) (15.50)
on C(0; |z0 |). By Theorem 10.18, the result (15.50) 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→∞
for every z ∈ U , but it also contradicts the Open Mapping Theorem.

– Case (iii): ϕ has two distinct fixed points b1 and b2 on T . Then it yields from
the equation (15.48) that for every z ∈ U , we have

 b1 , if |k| < 1;
lim ϕn (z) =
n→∞ 
b2 , if |k| > 1.
By similar argument to Case (ii), we can show that it is impossible for these two
cases so that |k| = 1.
Suppose that k is an N th root of unity for some N ∈ N. Then we see from the
equation (15.48) again that ϕN (z) = z. Otherwise, Subcase (iv) ensures that the
set S = {kn | n ≥ 0} is dense on T . Using similar argument as Case (i), we conclude
that this is impossible.
Consequently, the linear fractional transformation ϕ must satisfy the claim (15.49).
Problem 15.16
Proof. Note that

Z 1
1 − |αj | = dr,
|αj |
so we have
∞
X ∞ Z
X 1
(1 − |αj |) = dr. (15.51)
j=1 j=1 |αj |
Define the characteristic function


 1, if r ≥ |αj |;
χj (r) =

0, otherwise.
Then formula (15.51) becomes

∞
X ∞ Z
X 1
(1 − |αj |) = χj (r) dr. (15.52)
j=1 j=1 0
Observe that
∞
X
χj (r) = n(r). (15.53)
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 (15.52) and then use the formula
(15.53) 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
∞
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 [62, p. 66]
Furthermore, we also have An (eiθ ) = e−inθ B(eiθ ) ∈ L2 (T ) for every n = 1, 2, . . .. Clearly, we

have L∞ (T ) ⊆ L2 (T ).
Recall that L2 (T ) is an inner product space. On the one hand, we may apply [62, Eqn. (7)
& (8), p. 89] to get
Z π
1
hB, An i = B(eiθ ) · An (eiθ ) dθ
2π −π
Z π X∞ X ∞
1 ikθ
= ck e × ck ei(n−k)θ dθ
2π −π
k=0 k=0
∞
X
= ck cn+k . (15.54)
k=0
On the other hand, the fact |B(eiθ )| = 1 a.e. on T gives

h 1 Z π i2 1 Z 2
inθ iθ 2
hB, An i = e |B(e )| dθ = einθ dθ , (15.55)
2π −π 2π T \N
6 1 on N with m(N ) = 0. Recall from Remark 3.10 that L2 (T ) is a space whose

where |B(eiθ )| =
elements are equivalence classes of functions, so we can express (15.55) by
1 Z 2
hB, An i = einθ dθ = 0. (15.56)
2π T
Combining the two results (15.54) and (15.56), we conclude that ck = 0 for all k ∈ N, which is
impossible. Hence no such Blaschke product exists and we complete the proof of the problem.
Problem 15.18
Proof. Obviously, we have
f ′ (z) = 2(z − 1)B(z) + (z − 1)2 B ′ (z).
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 ) < ∞. (15.57)
n=1
Without loss of generality, we may assume that
0 < α1 ≤ α2 ≤ · · · < 1. (15.58)
Furthermore, suppose that

∞
Y ∞
Y
αn − z αk − z
B(z) = and Bn (z) = .
1 − αn z 1 − αk z
n=1 k=1
k6=n
We yield from [18, Exercise 10, p. 174] that
B ′ (z) X αn − z ′ αn − z −1
∞
=
B(z) 1 − αn z 1 − αn z
n=1
∞
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 (15.58)
that
δ = min(α1 , α2 , . . .) > 0.
Geometrically, if z ∈ U and p > 1, then
|p − z| > |1 − z|. (15.59)
Put p = α1n into the estimate (15.59) to get |1 − αn z|2 ≥ δ2 |1 − z|2 for every z ∈ U . Hence, the
condition (15.57) implies
∞
X X∞
1 − α2n 1
|B ′ (z)| ≤ · |B n (z)| ≤ (1 − α2n ) < ∞
n=1
(1 − αn z)2 δ2 |1 − z|2 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 [17].
15.4 Miscellaneous Problems and the Müntz-Szasz Theorem
Problem 15.19
Proof. Let 0 < r < 1. On the one hand, we notice that

reiθ + 1 r2 − 1
log |f (reiθ )| = log exp Re iθ = 2 .
re − 1 r − 2r cos θ + 1
Thus [62, Eqn, (3), §11.5, p. 233] asserts that
Z π Z π
1 r2 − 1 1
µr (f ) = dθ = − Pr (θ) dθ = −1. (15.60)
2π −π r 2 − 2r cos θ + 1 2π −π
On the other hand, we have

reiθ + 1 eiθ + 1
f ∗ (eiθ ) = lim f (reiθ ) = lim exp = exp
r→1 r→1 reiθ − 1 eiθ − 1
so that
eiθ + 1
log |f ∗ (eiθ )| = Re = 0.
eiθ − 1
µ∗ (f ) = 0. (15.61)
15.4. Miscellaneous Problems and the Müntz-Szasz Theorem 173
Hence it follows from the results (15.60) and (15.61) that
lim µr (f ) < µ∗ (f ),
r→1
Problem 15.20
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 in C(I) of the set of all finite linear combinations of the
functions 1 1 1
1, t λ1 , t λ2 , t λ3 , . . . .
By Theorem 15.26 (The Müntz-Szasz Theorem), we have the following results:
∞
X
• If λn = ∞, then X = C(I).
n=1
∞
X 1
• If λn < ∞, λ ∈
/ {λn } and λ 6= ∞, then X does not contain the function t λ .
n=1
This ends the analysis of the problem.
Problem 15.21
Proof. Let {λn } be a sequence of distinct real numbers and λn > − 12 . Then the set of all finite
linear combinations of the functions
tλ0 , tλ1 , tλ2 , . . .
if dense in L2 (I) if and only if

∞
X 2λn + 1
2+1
= ∞.
n=0
(2λ n + 1)
To this end, let m ∈ N and m ∈

/ {λn }. By [13, p. 173], we know that
n−1
X n−1
Y
m λk 1 m − λk
min t − ak t =√ .
ak ∈C 2 1 + 2m k=0 m + λk + 1
k=0
Thus we see that

tm ∈ span {tλ0 , tλ1 , . . .} (15.62)
if and only if
n−1
Y m − λk
lim sup =0
n→∞ m + λk + 1
k=0
if and only if
n−1
Y n−1
Y
2m + 1 2λk + 1
lim sup 1− × 1− = 0.
n→∞ m + λk + 1 m + λk + 1
k=0 k=0
λk >m − 12 <λk ≤m
Hence the set relation (15.62) is true if and only if

n−1
Y n−1
Y
2m + 1 2λk + 1
lim sup 1− = 0 or lim sup 1− = 0. (15.63)
n→∞ m + λk + 1 n→∞ m + λk + 1
k=0 k=0
λk >m − 21 <λk ≤m
2λk +1
Since m + λk + 1 > 0 and λk > − 12 , we have 2m+1
m+λk +1 , m+λk +1 ∈ (0, 1). By Theorem 15.5, the
results (15.63) hold if and only if
∞
X ∞
X
2m + 1 2λk + 1
=∞ or = ∞. (15.64)
m + λk + 1 m + λk + 1
k=0 k=0
λk >m − 12 <λk ≤m
1
Since λk + 2 < m + λk + 1 < 2λk + 1, we get
1 1 2
< <
2λk + 1 m + λk + 1 2λk + 1
which means that the first summation (15.64) is equivalent to
∞
X 1
= ∞. (15.65)
2λk + 1
k=0
λk >m
Since 1 < m + λk + 1 ≤ 2m + 1, the second summation (15.64) is equivalent to

∞
X
(2λk + 1) = ∞. (15.66)
k=0
− 21 <λk ≤m
If λk > m, then it is clear that

2m + 1 2m + 1 2λk + 1
< 2
< < 2λk + 1.
2λk + 1 (2λk + 1) + 1 (2λk + 1)2 + 1
Similarly, if − 12 < λk ≤ m, then it is easy to see that
2λk + 1 2λk + 1 2m + 1 2m + 1
2
≤ 2
≤ 2
< .
(2m + 1) + 1 (2λk + 1) + 1 (2λk + 1) + 1 2λk + 1
Therefore, we establish that one of the summations (15.65) and (15.66) holds if and only if
∞
X 2λk + 1
=∞
(2λk + 1)2 + 1
k=0
holds. Finally, our desired result is proven if we combine Theorem 3.14 and the Weierstrass
Approximation Theorem [61, Theorem 7.26, p. 159]. This completes the proof of the problem.

Remark 15.3
The proof of Problem 15.21 follows basically the proof of Theorem 2.2 in [14]. For the version
of a complex sequence {λn }, please refer to [50, §12, pp. 32 – 36].
Problem 15.22
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
for each n = 1, 2, . . .. Define Z ∞

F (z) = e−tz g(t) dt
0
on Π = {z ∈ C | Re z > 0}. For every fixed z ∈ Π, since
Z ∞ Z ∞
1
|e−tz |2 dt = e−2tRe z dt = √ < ∞,
0 0 2 Re z
we have e−tz ∈ L2 (0, ∞). Using Theorem 3.8, we see that
|F (z)| = ke−tz gk1 ≤ ke−tz k2 · kgk2 < ∞.
In other words, F is well-defined in Π.

Next, we have to compute F ′ (z) in Π. To this end, we consider
Z ∞ −tω
F (ω) − F (z) e − e−tz
= · g(t) dt. (15.67)
ω−z 0 ω−z
Given ǫ > 0. Let t > 0 and

ω = z − ζ, (15.68)
where |ζ| < ǫ. Then we have
e−t(z−ζ) − e−tz etζ − 1

= −e−tz · .
z−ζ−z ζ
etζ −1
Suppose that h(ζ) = ζ . The power series expansion of h is given by
∞ 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! ǫ
n=1
By the Mean Value Theorem, 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 etζ − 1

= e−tz · < 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
Since z is arbitrary, we have proven that F ∈ H(Π).

In fact, this process can be repeated to get
Z ∞
(n)
F (z) = (−1) n
tn e−tz g(t) dt, (15.71)
0
where n ∈ N and z ∈ Π. By the hypothesis and the formula (15.71), we have

Z ∞
(n)
F (1) = (−1) n
tn e−t g(t) dt = (−1)n hfn , gi = 0
0
so that F ≡ 0. Particularly, if we denote G(t) = χ(0,∞) (t)e−t g(t), then we see that
√ Z ∞ √ Z ∞ √
b
G(y) = 2π −ity
G(t)e dt = 2π e−t g(t)e−ity dt = 2πF (1 + iy) = 0,
−∞ 0
where y ∈ R. As G ∈ L1 (0, ∞), we observe from Theorem 9.12 (The Uniqueness Theorem) that
G(t) = 0 a.e. on R which implies that
g(t) = 0 (15.72)
a.e. on (0, ∞). Recall that L2 (0, ∞) is Hilbert and M is a subspace of L2 (0, ∞). If we have
M 6= L2 (0, ∞), according to the Corollary of Theorem 4.11, there corresponds a g ∈ L2 (0, ∞)
such that g 6= 0 and hfn , gi = 0 for every n = 1, 2, . . .. However, this definitely contradicts the
above conclusion (15.72). Hence M = L2 (0, ∞), as desired. This completes the analysis of the
problem.
Problem 15.23
Proof. Since f (0) = 0, λn 6= 0 for all n = 1, 2, . . . , N . Suppose that
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
λn − eiθ
=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
CHAPTER 16
Analytic Continuation
16.1 Singular Points and Continuation along Curves
Problem 16.1
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
an 1 k
∞ X
X ∞
= 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 [61, 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
179
180 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
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
0 = s0 < s1 < · · · < sn = 1 and 0 = t0 < t1 < · · · < tm = 1

′ ,
such that γ(0) is the center of D0 = D0′ , γ(1) is the center of Dn and Dm

γ [sj , sj+1 ] ⊆ Dj and γ [tk , tk+1 ] ⊆ Dk′
for j = 0, 1, . . . , n − 1 and k = 0, 1, . . . , m − 1. We notice that there are function elements

′
(fj , Dj ) ∼ (fj+1 , Dj+1 ) and (gk , Dk′ ) ∼ (gk+1 , Dk+1 ) for j = 0, 1, . . . , n−1 and k = 0, 1, . . . , m−1,
where f0 = f and g0 = g.

Since P f0 (z), g0 (ζ) = 0 for all z, ζ ∈ D0 , we have P f1 (z), g0 (ζ) = 0 for all z ∈ D0 ∩ D1
and ζ ∈ D0 . For each fixed ζ ∈ D0 , P is a polynomial in z so that P ∈ H(D1 ). By the Corollary
to Theorem 10.18, we have
P f1 (z), g0 (ζ) = 0 (16.3)
for all z ∈ D1 . Since ζ is arbitrary, the equation (16.3) is actually true for all ζ ∈ D0 . Repeat
this process, we conclude that P (fn (z), g0 (ζ)) = 0 for all z ∈ Dn and ζ ∈ D0 . Next, we fix a
z ∈ Dn and now P is a polynomial in ζ so that P ∈ H(D1′ ). Similar argument shows that

P fn (z), g1 (ζ) = 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
Proof. By Theorem 11.2, the function f = ux − iuy is holomorphic in Ω. Since Ω is simply

connected, Theorem 13.11 ensures that there corresponds an F ∈ H(Ω) such that F ′ = f . If
F = A + iB, then we have
F ′ (z) = Ax + iBx = Ax − iAy = ux − iuy

16.1. Singular Points and Continuation along Curves 181
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
|eF (z) | = eRe F (z) = elog |f (z)| = |f (z)|
which means that

|f e−F | = 1
in Ω. According to [9, Proposition 3.7, p. 39], f (z)e−F (z) = eiθ for some constant θ in Ω. Now
we can write
f (z) = eg(z)
for some g ∈ H(Ω). By Theorem 13.11, Ω is simply connected, a contradiction. Hence this shows
that the statement of the problem fails in every region that is not simply connected, completing
Problem 16.4
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) .
Figure 16.1: The paths βζ (I) and γζ .
If γ is a path in X from x0 to x1 , and if λ is a path in X from x1 to x2 , then we define the

product γ ∗ λ of γ and λ to be the path η given by the equations

 γ(2t), if t ∈ [0, 21 ];
η(t) =

λ(2t − 1), otherwise.
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 [42, 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)
Figure 16.2: The paths γz (I), γω (I) and γω,ζ (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 183
ℓ(z, ω) be the length of γω,z . Hence, if |z − ω| < δ, then we obtain

Z 1 ′ Z 1
γz (t) dt γω (t) dt
|g(z) − g(ω)| = −
0 γ z (t) 0 γω (t)
Z 1 ′ Z 1 ′
(γω ∗ γω,z ) (t) dt γω (t) dt
= −
(γω ∗ γω,z )(t) γω (t)
Z0 Z 0
dt dt
= −
γ ∗γ t γω t
Z ω ω,z Z Z
dt dt dt
= + −
γ t γω,z t γω t
Z ω
dt
= . (16.8)
γω,z t
Applying the definition of m to the integral (16.8), we get

ℓ(z, ω) κ
|g(z) − g(ω)| ≤ < < ǫ.
m m
In other words, g is continuous at ω so that it is actually continuous on X, as required. This
completes the proof of the problem.
16.2 Problems on the Modular Group and Removable Sets
Problem 16.5
Proof. Let τ (z) = z + 1, σ(z) = − z1 and

az + b
ϕ(z) = ∈ G.
cz + d
Then we have a, b, c, d ∈ Z and ad − bc = 1.a Furthermore, we notice that τ −1 (z) = z − 1 and
σ −1 (z) = − z1 .
• Case (i): a = 0. We have bc = −1 so that b = −c. Obviously, b = ±1 if and only if

c = ∓1. If b = 1, then c = −1 and we have
1 1 1
ϕ(z) = =− = − −d = σ τ −d (z) ,
−z + d z−d τ (z)
i.e., ϕ = σ ◦ τ −d . Similarly, we have ϕ = σ ◦ τ d if b = −1 and c = 1.
• Case (ii): a = ±1. Since −az−b az+b
−cz−d = cz+d , we may only consider the case that a = 1. Thus
z+b
we have d − bc = 1 and ϕ(z) = cz+d . We note that
−cz − d −cz − d −cz − d −(c − 1)z − (d − b)
σ ϕ(z) = and τ = +1=
z+b z+b z+b z+b
which imply that
−(d − bc) 1
τ c σ ϕ(z) = =− = σ τ b (z) .
z+b z+b
Hence we have ϕ = σ −1 ◦ τ −c ◦ σ ◦ τ b .
a
With the aid of Problem 16.7, we remark that G = SL2 (Z).

• 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
where either am = 0 or am = ±1. Therefore, the ϕm , and hence ϕ, is generated by τ and

σ.
Consequently, this proves the first assertion that τ and σ generate the modular group G.
For the second assertion, suppose that
R1 = {z = x + iy | |x| < 12 , y > 0 and |z| > 1},

R2 = {z = x + iy | − 12 ≤ x ≤ 0 and |z| = 1}, (16.9)
n 1 o
R3 = z = − + iy y > 0 and |z| ≥ 1 .
2
Then we have R = R1 ∪ R2 ∪ R3 , see Figure 16.3.
Figure 16.3: The fundamental domain R of G.
We check Theorem 16.19(a) and (b). Based on Apostol’s description [5, 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.
Lemma 16.1
Suppose that z1 , z2 ∈ R, z1 6= z2 and z2 = ϕ(z1 ) for some ϕ ∈ G. Then we have
Re z1 = ± 21 and z2 = z1 ∓ 1 or |z1 | = 1 and z2 = − z11 .
Proof of Lemma 16.1. Without loss of generality, we may assume that Im z2 ≥ Im z1

by symmetry. Let ϕ(z) = az+b
cz+d . Combing the assumption and the relation
Im z
Im ϕ(z) = (16.10)
|cz + d|2
to get the condition

|cz1 + d|2 ≤ 1. (16.11)
√ √
Since z1 ∈ R, it is easy to see that Im z1 ≥ 23 . Thus |c| · 2
3
≤ |c|Im z1 ≤ |cz1 + d| ≤ 1.
As c ∈ Z, this forces that either c = 0 or |c| = 1.
• Case (i): c = 0. Then ad = 1 and since a, d ∈ Z, we have a = d = ±1. In this

case, the relation (16.10) shows that Im z2 = Im z1 . Furthermore, ϕ(z) = z ± b
so that Re z2 = Re z1 ± b. Since b is an integer, the definition (16.9) shows b = 1
which implies that Re z1 = ± 21 and hence z2 = z1 ∓ 1.
• 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.
– Subcase (i): |d| = 1. Using the inequality (16.13), we have |Re z1 ± 1| = 21

and then Re z1 = ± 21 . Next, it follows from the inequality (16.12) that
√ √
3 3
0 ≤ Im z1 ≤√ 2 , so actually we have Im z1 = 2 . Consequently, we have
i 3
z1 = ± 12 + 2 and then both our results hold in this case.
– Subcase (ii): d = 0. Now the condition (16.11) implies that |z1 | ≤ 1.
Since z1 ∈ R by the definition (16.9), we actually have |z1 | = 1. Simple
calculation gives
1
z2 = ±a − ,
z1
where a ∈ Z. Let z1 = x + iy. Thus z2 = ±a − x + iy. If a 6= 0, then since
z1 , z2 ∈ R and z1 6= z2 , we get z1 = − 21 + iy and z2 = 21 + iy. Otherwise,
a = 0 so that z1 = x + iy and z2 = −x + iy for every 0 < x ≤ 12 . Obviously,
this case satisfies |z1 | = 1 and z2 = − z11 .
This completes the proof of the lemma.

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 [5, 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

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
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.
• An algebraic proof of Theorem 16.19(c). Now the group Γ is generated by the

matrices
1 0 1 2
A= and B = .
2 1 0 1
If M ∈ Γ, then we have
M = An1 Bm1 An2 Bm2 · · · Anp Bmp , (16.15)
where the nk , mk are integers. Direct computation gives

−1 1 0 −1 1 −2
A = and B =
−2 1 0 1
so that
nk 1 0 mk 1 (−2)mk
A = and B =
(−2)nk 1 0 1
and then

nk mk 1 (−2)mk 1 2Nk
A B = = ,
(−2)nk (−2)mk +nk + 1 2Mk 2Lk + 1
where Nk , Mk and Lk are integers. Thus we obtain

1 + 2Nk,j 2Mk,j
Ank Bmk Anj Bmj = ,
2Pk,j 1 + 2Lk,j
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
respectively. We claim that if M ∈ G, then it has the form
M = Tn1 STn2 S · · · Tnp S, (16.16)
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
with c ≥ 0. If c = 0, then ad = 1 or equivalently, a = d = ±1 so that

±1 b 1 ±b
M= = = T±b .
0 ±1 0 1
b
Remember that we have identified I = −I.
Next, if c = 1, then ad − b = 1 so that b = ad − 1 and

a ad − 1 1 a 0 −1 1 d
M= = = Td STd .
1 d 0 1 1 0 0 1
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

a b 1 −q a −aq + b
MT−q = = ,
c d 0 1 c r
we have
−q −aq + b −a
MT 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
Proof. Since E ⊆ R is compact, we can define
A = max x and B = min x.

x∈E x∈E
Notice that Ω = C \ E. Denote R = max(|A|, |B|).
(a) Let x, y ∈ R such that x < y < A. Then x 6= y and

Z Z
1 1 dt
f (x) − f (y) = − dt = (x − y) . (16.18)
E t − x t − y E (t − x)(t − y)
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
i.e., f (x) 6= f (y). Consequently, f is nonconstant.
(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|
on E. Since |z| > R, we have |z − t| ≥ |z| − |t| ≥ |z| − R > 0 so that
|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
for all t ∈ E. This proves our claim which asserts that

Z
|zf (z) + m(E)| = g(z, t) dt + m(E)
E
Z
= [g(z, t) + 1] dt
Z E
≤ |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
lim zf (z) = −m(E).

z→∞
(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 [42, Theorem
23.4, p. 150] that Π+ ∪ {a} is also connected. Similarly, the set Π− ∪ {a} is also connected.
According to [42, 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 [42, 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.
(f) Suppose that z = a + ib ∈ Ω. Then it is easy to see that

Z Z Z
dt (t − a) dt b dt
f (z) = = 2 2
+i 2 2
. (16.21)
E (t − a) − ib E (t − a) + b E (t − a) + b
For every z ∈ Ω, we deduce from the expression (16.21) that

Z
b dt
|Im f (z)| =
(t − a)2 + b2
ZE∞
b dt
≤
−∞ (t − a)2 + b2
t − a ∞
= tan−1
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 Z Z
dt 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
(h) By parts (e) and (f), we see that
f (Ω) ⊆ {z = x + iy | x ∈ R and −π ≤ y ≤ π}.
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
which implies that |t − z| ≥ δ for every z ∈ D(z0 ; δ) and t ∈ E. Therefore, we get

Z Z
1 dt m(E)
sup dt ≤ = < ∞.
z∈D(z0 ;δ) E |t − z| E δ δ
Z
In other words, |ψ(z, t)| dt is locally bounded. Hence we conclude that f ∈ H(Ω).c
E
Problem 16.9
c
See the online paper http://www.nieuwarchief.nl/serie5/pdf/naw5-2001-02-1-032.pdf or [22].
Proof.
(a) It is easy to see that

Z 1
dt 1 1 5
f (−2) = = log(t + 2) = log 3 and f (−4) = log(t + 4) = log .
−1 t+2 −1 −1 3
Thus f is not constant in Ω.
(b) Similar to Problem 16.8(b), f cannot be extended to an entire function.
(c) According to Problem 16.8(c), the value of the limit is −2 because m(E) = 2.
(d) Similar to Problem 16.8(d), f has no holomorphic square root in Ω.
(e) Similar to Problem 16.8(e), Re f is unbounded in Ω.
(f) Since E = [−1, 1], we see that

Z 1 1 − a −1 − a
b −1 t − a
1
Im f = dt = tan = tan−1 − tan−1 .
−1 (t − a)2 + b2 b −1 b b
(g) We have the exact result

Z Z 1
1
dt dz = 2πm(E)i = 4πi.
γ −1 t−z
(h) By Problem 16.18(h), the nonconstant bounded holomorphic function ϕ in Ω is given by

Z 1
dt z − 1
ϕ(z) = exp i = exp i log .
−1 t−z z+1
Problem 16.10
Proof.
(a) We first need the following lemma:
Lemma 16.3
Suppose that E is compact and has no interior, and K satisfies the following two
conditions:
– Condition (1): K ⊆ E is compact (K can possibly be empty) and
– Condition (2): Each f ∈ H(C \ E) can be extended to an fK ∈ H(C \ K).
Let E ′ be the intersection of all such compact subsets K of E. Then E ′ also

satisfies the conditions.
Proof of Lemma 16.3. Obviously, we have E ′ ⊆ K and E ′ is compact. This means

that E ′ satisfies Condition (1). To show that E ′ also satisfies Condition (2), let
f ∈ H(C \ E) and z ∈ C \ E ′ . Then z ∈ / K (or equivalently z ∈ C \ K) for some
compact K ⊆ E satisfying Condition (2). Thus our f ∈ H(C \ E) can be extended
to an fK ∈ H(C \ K). Since E has no interior, there exists a sequence {zn } ⊆ C \ E
such that zn → z and
fK (z) = lim f (zn )
n→∞
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
fb(z) = fK (z) (16.22)
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
Since E ′ is compact, E ′ is closed in C so that it is a complete metric space by [61,

Theorem 3.11, p. 53]. If each E ′ ∩ {zn } has no interior, then it is nowhere dense and
the Baire Category Theorem shows that E ′ is of the first category, a contradiction. Thus
{zN } = E ′ ∩ {zN } has a nonempty interior for some N ∈ N which is impossible. Hence we
have E ′ = ∅ and we deduce from Lemma 16.3 that every f ∈ H(C \ E) can be extended to
an entire function and we denote it by the same notation f . Particularly, if f is bounded,
then the corresponding entire function f is also bounded and Theorem 10.23 (Liouville’s
Theorem) forces that it is a constant. By the definition, E is removable.
(b) Let E ⊆ R be compact and m(E) = 0. Let f ∈ H(C \ E) be bounded by a positive

constant M . By the proof of Theorem 13.5, there exists a cycle Γ in C \ E such that
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
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
It is easy to see that Z

1 f (ζ) 1 M πǫ
dζ ≤ × .
2πi Γǫ ζ −α 2π d(α, E)
Since ǫ is arbitrary small, the second integral in the equation (16.23) is actually zero, we
get
f (α) = F (α)
for every α ∈ V \ E. Since F ∈ H(V ), we conclude immediately that f ∈ H(C) which
implies that it is a constant. By the definition, E is removable.
(c) Suppose thatf : Ω \ E → C is bounded by M . We fix z0 ∈ Ω \ E. Let Γ1 be a cycle in

Ω \ E ∪ {z0 } with winding number 1 around E ∪ {z0 } and zero around C \ Ω. Similarly,
suppose that Γ2 is a cycle in Ω \ E ∪ {z0 } with winding number 1 around E and zero
around (C \ Ω) ∪ {z0 }. Since Ind Γ1 (α) = Ind Γ2 (α) = 0 for every α ∈/ Ω, Theorem 10.35
(Cauchy’s Theorem) asserts that our construction guarantees
Z Z
1 f (ζ) 1 f (ζ)
f (z0 ) = dζ − dζ.
2πi Γ1 ζ − z0 2πi Γ2 ζ − z0
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).
(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
f (∞) = lim f (z) = 0.

z→∞
Consequently, we establish
f ′ (∞) = lim z[f (z) − f (∞)]

z→∞
z[g(z) − g(z0 )]
= lim
z→∞ z − z0
z z
= lim g(z) − g(z0 ) lim
z→∞ z − z0 z→∞ z − z0
= g(∞) − g(z0 ) 6= 0,
completing the proof of Lemma 16.4.
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 [73, Eqn. (1.2), p. 16], we have
Z n
X
′ 1
|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.
(e) Suppose first that E ⊂ C is compact and removable. If F ⊆ E is a connected component

of E containing more than one point, then it follows from Theorem 14.8 (The Riemann
Mapping Theorem) that there exists a conformal mapping f : C \ F → U which is non-
constant. Thus f |C\E ∈ H(C \ F ) and f |C\E must be bounded. By the definition, E is
d
Here m2 denotes the Lebesgue measure in two dimensional space C.
non-removable, a contradiction. Hence connected components of E are one-point sets, i.e.,

E is totally disconnected, see [42, Exercise 5, p. 152].
Now if E ⊂ C is a connected subset with more than one point, then the above paragraph
ensures that E must be non-removable.e
We have completed the analysis of the proof of the problem.
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
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.
Figure 16.4: The regions Ωα and Ωβ if α < β.
e
Recall from point-set topology [45, p. 3] that a nonempty compact connected metric space is a continuum.
• fβ is an analytic continuation of fα if α < β. Let ζ ∈ Ωα . Then there exists a δ > 0

such that D(ζ; δ) ⊆ Ωα and D(ζ; δ) ∩ Γ∗α = ∅. Define ψα : D(ζ; δ) × Γ∗α → C by
exp(eω )
ψα (z, ω) = .
ω−z
If ω = t + πi for t ∈ [α, ∞), then we have eω = −et so that
1
| exp(eω )| = exp(−et ) ≤ <∞ (16.26)
eeα
for all t ∈ [α, ∞). In fact, this bound (16.26) remains true on −t − πi for all t ∈ (−∞, −α].
If ω = α + πitα for t ∈ [−α, α], then we see that
πt
| exp(eω )| = exp eα cos ≤ exp(eα ) < ∞.
α
In other words, the function exp(eω ) is bounded on Γ∗α . Next, it is obvious that ψα (z, ω)
is a measurable function of ω, for each fixed z ∈ D(ζ; δ), and ψα (z, ω) is holomorphic in
D(ζ; δ), for each fixed ω ∈ Γ∗α . By Problem 10.18, we conclude immediately that fα is
holomorphic at ζ. Since ζ is arbitrary, we obtain fα ∈ H(Ωα ).
Recall that Ωα ⊂ Ωβ if α < β, so fβ is an analytic continuation of fα if α < β.
• Existence of an entire function f such that f = fα on Ωα . We fix an α > 0 and an
ζ ∈ Ωα . Since Ωα is open in C, one can find a δζ > 0 such that
D(ζ; δζ ) ⊆ Ωα .
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)
for all z ∈ D(ζ; δζ ). By the Corollary to Theorem 10.18, we have f = fα on Ωα .
• f (reiθ ) → 0 as r → ∞ for every eiθ 6= 1. Suppose that r > 0 and θ is real. By the second
assertion, we know that f (z) = f1 (z) on Ω1 . By the assumption, we have reiθ ∈ Ω1 for
large enough r > 0. Write Γα = γα− + Lα + γα+ , where γα− = −t − πi for t ≤ −α, γα+ = t + πi
for t ≥ α and Lα = α + πit α for t ∈ [−α, α]. Therefore, we see that
Z
1 exp(eω )
|f (reiθ )| = dω
2π Γ1 ω − z
Z Z Z
1 exp(eω ) 1 exp(eω ) 1 exp(eω )
≤ dω + dω + dω
2π γ1− ω − z 2π L1 ω − z 2π γ1+ ω − z
Z −1 Z 1
1 exp(−e−t ) 1 exp(e cos πt) · exp(ie sin πt)
= iθ
dt + πi dt
2π −∞ −t − πi − re 2π −1 1 + πit − reiθ
Z ∞
1 exp(−et )
+ dt . (16.27)
2π 1 t + πi − reiθ
Since |ω − reiθ | ≥ r sin θ − 1 for large enough r > 0 and for every ω ∈ Γ∗1 , the inequality
(16.27) reduces to
1 h Z −1 Z 1 Z ∞ i
|f (reiθ )| ≤ exp(−e−t ) dt + exp(e cos πt) dt + exp(−et ) dt
2π(r sin θ − 1) −∞ −1 1
h Z ∞ i
1
≤ 2 exp(−et ) dt + 2ee . (16.28)
2π(r sin θ − 1) 1
Since et ≤ exp(et ) for every t ≥ 0, the inequality (16.28) further reduces to
1
|f (reiθ )| ≤ (ee + e−1 ).
π(r sin θ − 1)
Since eiθ 6= 1, sin θ 6= 0 which implies that
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
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
= 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
Using the results (16.29) and (16.30), we immediately see that

Z ∞
− exp(−et )
dt = 0 (16.31)
R (t − r)2 + π 2
1 1
for every R > r. Since exp(−et ) ≤ e−t and (t−r)2 +π 2
≤ (R−r)2 +π 2
, we get
Z ∞ Z ∞
− exp(−et ) 1 1
dt ≥ − e−t dt = >0
R (t − r)2 + π 2 (R − r)2 + π 2 R eR [(R − r)2 + π 2 ]
which contradicts the result (16.31). Hence f (r) 6= 0.

• g(reiθ ) → 0 as r → ∞ for every eiθ . If eiθ 6= 1, then the third assertion implies that
lim g(reiθ ) = lim f (reiθ ) exp[−f (reiθ )] = 0 · 1 = 0.

r→∞ r→∞
Next, suppose that eiθ = 1. Since f (r) → ∞ as r → ∞, we see immediately that
f (r)
lim g(r) = lim = 0.
r→∞ r→∞ exp[f (r)]
This gives the fifth assertion.
• 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
lim G(reiθ ) = 0 (16.32)

r→∞
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.
We have ended the analysis of the problem.
Problem 16.12
Proof. Suppose that f is represented by the series

∞
X z − z 2 3k
. (16.33)
2
k=1
2
Evidently, if |z − z 2 | < 2, then | z−z
2 | < 1 so that the series (16.33) converges by [61, 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) = k Pk (z).
k=1
23
f
This is (x − x2 + y)2 + (y − 2xy)2 < 4.
Figure 16.5: The regions of convergence of the two series.
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 [61, 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
= k k
2π · 3 2+1 · 3 2−1
2 3k
= ·
π 32k − 1
2 1
∼ · k (16.36)
π 3
and
k
(3k )3 k
3k +1 3k −1
∼ 23 . (16.37)
3k +1 3k −1
( 2 ) 2 ( 2 ) 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
Proof. For each positive integer n, we have
Xn = {f ∈ H(Ω) | f = g(n) for some g ∈ H(Ω)}.
(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
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
the integral (16.38) implies that c−1 = 0. Thus we obtain

∞
X
f1 (z) = c−n z −n .
n=2
If we let ω = z1 , then the function
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
holds in C \ D(0; 21 ). Similarly, we can show that
d X cn n+1
∞
f2 (z) = z (16.40)
dz n=0 n + 1
holds in D(0; 2). Finally, if we define
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=−∞
(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
Proof. Our proof here basically follows that in [64, §2.7, pp. 54 – 56]. Since normality is a local
property, we may assume that Ω is the unit disc U . Suppose that
F = {f ∈ H(U ) | |f (p)| ≤ R and 0, 1 ∈

/ f (U )}.
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 [79, Problem
5.25, pp. 68, 69] ensures that there is a convergent subsequence {fnk (p)}. Let this limit be ℓ.
• 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 (−ifc c c
nk ) = Im fnk > 0 so that the family {−ifnk | 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→∞
Recall that the domain of λ is Π+ , so the Open Mapping Theorem implies that g(U ) ⊆ Π+
and λ ◦ g : U → C is well-defined such that

lim fnk (z) = lim λ fcnk (z) = λ g(z)
k→∞ k→∞
for all z ∈ U . Hence {fnk } is the required subsequence.
• 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
lim hk (p) = −1. (16.43)

k→∞
√
Since fnk = h2k , we have 0, 1 ∈
/ hk (U ) and |hk (p)| ≤ R. Consider the family
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
lim fnkj (z)

j→∞
exists for all z ∈ U .
• Case (iii): ℓ = 0. In this case, we may apply Case (ii) to the sequence {1 − fnk | k ∈ N}.
Hence we have completed the proof of the problem.
Remark 16.2
Problem 16.14 is classically called the Fundamental Normality Test.
Problem 16.15
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
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 [9, §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
17.1 Problems on Subharmonicity and Harmonic Majoriants
Problem 17.1
Proof. Let u : Ω → R be an upper semicontinuous subharmonic function. By Definition 2.8, for

every real α, the set {z ∈ C | u(z) < α} is open in C.
• Let K ⊂ Ω be compact and h : K → R be continuous such that h is harmonic in V = K ◦

and u(z) ≤ h(z) for all boundary points of K. Put u1 = u − h. Assume that u1 (ζ) > 0
for some ζ ∈ V . Since h is continuous on K, −h is upper semicontinuous on K. Thus u1
is also upper semicontinuous on K by [78, Problem 2.1, pp. 17, 18]. Since K is compact,
[78, Problem 2.21, p. 60] ensures that u1 attains its maximum m on K. Since u1 ≤ 0 on
the boundary of K, the set E = {z ∈ K | u1 (z) = m} is a nonempty compact subset of V .
Let z0 be a boundary point of E. Since E is compact, V is open in C and E ⊂ V , there
exists an r > 0 such that D(z0 ; r) ⊂ V . Now some subarc of the boundary of D(z0 ; r) lies
in V \ E. Hence we have
Z π
1
u1 (z0 ) = m > u1 (z0 + reiθ ) dθ
2π −π
which means that u1 is not subharmonic in V . However, since u and h are subharmonic
and harmonic in V respectively, it follows from the mean value property that u1 is also
subharmonic in V , a contradiction. This proves that no such ζ exists and then u(z) ≤ h(z)
for all z ∈ K.
• 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
Lemma 17.1 (Baire’s Theorem on Semicontinuous Functions)

Let K ⊂ C be compact and −∞ ≤ u(z) < ∞ on K. If u is upper semicontinuous,
then it is the limit of a monotone decreasing sequence of continuous functions {un }
on K.
a
Read https://encyclopediaofmath.org/wiki/Baire_theorem#Baire.27s_theorem_on_semi-continuous_functions .
205
206 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
u(z) ≤ (Hun )(z) (17.1)
for all z ∈ K. Furthermore, the mean value property gives

Z π Z π
1 1
(Hun )(0) = (Hun )(r2 eit ) dt = un (r2 eit ) dt. (17.2)
2π −π 2π −π
Combining the inequality (17.1) and the formula (17.2) we obtain

Z π Z π Z π
1 1 1
u(r1 eit ) dt ≤ (Hun )(r1 eit ) dt = (Hun )(0) = un (r2 eit ) dt. (17.3)
2π −π 2π −π 2π −π
Finally, we apply Problem 1.7b to the inequality (17.3) to get

Z π Z π Z π
1 1 1
u(r1 eit ) dt ≤ lim un (r2 eit ) dt = u(r2 eit ) dt.
2π −π n→∞ 2π −π 2π −π
Problem 17.2

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
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.
• f ∈ H p if and only if |f (z)|p ≤ u(z) for some harmonic function u in U . Suppose

that there exists a harmonic function u in U such that |f (z)|p ≤ u(z) for all z ∈ U .
Combining this and the mean value property, we get
nZ o1 n Z o1 1
iθ p p iθ p
kfr kp = |f (re )| dσ ≤ u(re ) dσ = [u(0)] p < ∞
T T
for every 0 ≤ r < 1. By Definition 17.7, we have f ∈ H p .

b
In fact, it is
17.1. Problems on Subharmonicity and Harmonic Majoriants 207
Conversely, suppose that f ∈ H p . Now it is easy to see that

Z Z
p p
kfr k1 = |fr | dσ = |fr |p dσ = kfr kpp
T T
which means kf p k1 = kf kpp < ∞. As a consequence, we have f p ∈ H 1 . Applying [58, Eqn.

(5), p. 344] directly to f p , we have
|f (z)|p ≤ |Qf p (z)|

n 1 Z eit + z o
p ∗ it
= exp log |(f ) (e )| dt
2π T eit − z
n h 1 Z eit + z io
p ∗ it
= exp Re log |(f ) (e )| dt
2π T eit − z
n 1 Z o
= exp Pr (θ − t) log |(f p )∗ (eit )| dt (17.5)
2π T
for all z ∈ U . Using the same argument as in proving the inequality in the proof of
Theorem 17.16(c), we obtain from the inequality (17.5) that
Z
1
|f (z)|p ≤ Pr (θ − t)|(f p )∗ (eit )| dt = P [(f p )∗ ](z)
2π T
in U .c By Theorem 17.11(b), (f p )∗ ∈ L1 (T ), so Theorem 11.7 ensures that P [(f p )∗ ] is

harmonic in U .
• The existence of a least harmonic majorant. Let uf = P [(f p )∗ ]. In fact, this is

a least harmonic majorant. To see this, let u be a harmonic majorant in U , i.e., u is
harmonic in U and |f (z)|p ≤ u(z) for every z ∈ U . For any ρ < 1, we have
Z Z
1 it p 1
Pr (θ − t)|f (ρe )| dt ≤ Pr (θ − t)u(ρeit ) dt = u(ρz),
2π T 2π T
where 0 ≤ r < 1. As ρ → 1, we getd

Z
1
uf (z) = Pr (θ − t)|(f p )∗ (eit )| dt
2π T
Z
1
= Pr (θ − t)lim |f p (ρeit )| dt
2π T ρ→1
h 1 Z i
= lim Pr (θ − t)|f (ρeit )|p dt
ρ→1 2π T
≤ 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).
Problem 17.4
Proof. On the one hand, if log+ |f | has a harmonic majorant in U , then there exists a harmonic
function u in U such that 0 ≤ log+ |f (z)| = log+ |f (z)| ≤ u(z). Combining this and the mean
value property, we obtain
Z Z
+ it
0 ≤ kfr k0 = exp log |fr (e )| dσ ≤ u(reit ) dσ = u(0) < ∞
T T
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
log+ |f (z)| ≤ −u(z)
in U . By Theorem 11.4 and g ∈ H(U ), −u is harmonic in U . Hence, log+ has a harmonic

majorant in U and this completes the proof of the problem.
17.2 Basic Properties of H p
Problem 17.5
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
p
f ϕ(z) ≤ u ϕ(z)
for all z ∈ U . Applying Problem 11.7(b) with Φ = u and f = ϕ there, we see that
∆[u ◦ ϕ] = [(∆u) ◦ ϕ] × |ϕ′ |2 = 0.
By the definition, u ◦ ϕ is harmonic in U and then Problem 17.3 asserts that f ◦ ϕ ∈ H p .

17.2. Basic Properties of H p 209
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
Proof. Let α > 0 and consider the functione

1
fα (z) = .
(1 − z)α
Put z = reit , we have

Z Z π
it 1
Iα (r) = |fα (re )| dt = dt.
T −π |1 − reit |α
We want to estimate Iα (r) as r → ∞. Of course, it depends on the value of α.

Since r → 1, we may assume that r > 21 . By considering the triangle formed by 1, r and reit .
Clearly, this is an obtuse triangle. If t ∈ [−π, π], then we have
|1 − reit | > max{1 − r, r|1 − eit |}

1
≥ 1 − r + r|1 − eit |
2
1 1
> 1 − r + |1 − eit |
2 2
1 t
= 1 − r + sin . (17.8)
2 2
Recall the fact [61, Exercise 7, p. 197] that | sin x| ≥ π2 |x| for every x ∈ [− π2 , π2 ], so the inequality
(17.8) can further reduce to
1 |t| 1
|1 − reit | ≥ 1−r+ ≥ (1 − r + |t|). (17.9)
2 π 2π
On the other hand, the triangle inequality gives
|1 − reit | ≤ |1 − eit | + |eit − reit | = |1 − eit | + 1 − r ≤ |t| + 1 − r. (17.10)
Combining the inequalities (17.9) and (17.10), we obtain
1 1 (2π)α
α ≤ ≤ α
|t| + 1 − r |1 − reit |α |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 < arg(1 − z) < π2 .

Z π Z π
dt dt
α ≤ Iα (r) ≤ (2π)α α . (17.11)
−π |t| + 1 − r −π |t| + 1 − r
Since |t| is an even function in t, we get

Z π Z π
dt dt
α = 2 α
−π |t| + 1 − r 0 t+1−r

 2 log(1 − r + π) − log(1 − r) , if α = 1;
=
 2
1−α − (1 − r)1−α , otherwise.
1−α (1 − r + π)
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 r ∞
α = 2r , then we have f 1 ∈ H but f 1 ∈ / H . This completes the analysis of the problem.
2r 2r
Problem 17.7
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
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
which shows Z
iθ
fr (e ) = f (re ) = iθ
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
iθ p
|fr (e )| = Pr (θ − t)f ∗ (eit ) dσ
ZT p
≤ Pr (θ − t)|f ∗ (eit )| dσ
ZT 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
∗ it p p−1
≤ Pr (θ − t)|f (e )| 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
ZT T
= |f ∗ (eit )|p dσ
T
= kf ∗ kp ,
where 0 ≤ r < 1. By the hypothesis, kf ∗ kp < ∞ so that f ∈ H p as required. We complete the

analysis of the problem.
Problem 17.9
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
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
The Cauchy-Schwarz inequality and Theorem 17.12 combine to yield

∞
X X
∞ 1 X
∞ 1 X
∞ 1
2 2 2n 2 2
|f (α)| = n
an α ≤ |an | × |α| = |α|2n · kf k2 < ∞.
n=0 n=0 n=0 n=0
By Definition 5.3, Φα is a bounded linear functional. By Theorem 5.4, Φα is continuous and

since H 2 is a Hilbert space, Theorem 4.12 (The Riesz Representation Theorem) ensures the
existence of an g ∈ H 2 such thatf
Z Z
f (α) = Φα (f ) = hf, gi = lim f (reiθ )g(reiθ ) dσ = f ∗ (reiθ )g∗ (reiθ ) dσ, (17.12)
r→1 T T
where f ∗ , g ∗ ∈ L2 (T ) by Theorem 17.11. If we let

∞
X
g(z) = bn z n ,
n=0
∞
X
then Theorem 17.12 implies |bn |2 < ∞. By the Parseval Theorem [62, Eqn. (6), p. 91], the
n=0
formula (17.12) becomes
∞
X ∞
X
n
an α = f (α) = an bn .
n=0 n=0
As a result, we have bn = αn for all n = 0, 1, 2, . . .. Hence g has the representation

∞
X
g(z) = αn z n .
n=0
Problem 17.11
Proof. Suppose that

∞
X
f (z) = an z n
n=0
in U . By the Parseval Theorem [62, Eqn. (6), p. 91], we have

Z ∞
X
it 2
|f (re )| dσ = |an |2 r 2n ,
T n=0
f
Or you may apply [62, Eqn. (6), p. 347] directly.
where 0 ≤ r < 1, so that

X
∞ 1 nZ o1
2 2 2
|an | = sup |f (reit )|2 dσ = kf k2 ≤ 1.
n=0 r<1 T
By the Cauchy-Schwarz inequality, we can establish

∞
X 2
′ 2
|f (α)| = nan αn−1
n=1
X
∞ X
∞
2
≤ |an | × n2 |α|2(n−1)
n=1 n=1
∞
X
≤ n2 |α|2(n−1) . (17.13)
n=1
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
kFα k22 = · n2 |α|2(n−1)
n=0
Cα
∞
(1 − |α|2 )3 X 2 2(n−1)
= · n |α|
1 + |α|2
n=0
(1 − 1 + |α|2
|α|2 )3
= ×
1 + |α|2 (1 − |α|2 )3
= 1.
Finally, we know that

s s
∞
1 X 2 2(n−1) (1 − |α|)3 1 + |α|2 1 + |α|2
|Fα′ (α)| = n |α| = × = .
|Cα | 1 + |α|2 (1 − |α|)3 (1 − |α|)3
n=1
If α = 0, then instead of the function (17.16), we consider
F0 (z) = eiθ z
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
(n) 2
|f (α)| = m(m − 1) · · · (m − n + 1)am αm−n
m=n
X
∞ 2 nX
∞ o2
2
≤ |am | × [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
(n)
|f (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
Then we claim that

∞
eiθ X
Fα (z) = m(m − 1) · · · (m − n + 1)|α|m−n z m
Cα,n m=n
are the extremal functions for α 6= 0. In fact, we know that

∞
X
1 1
kFα k22 = |m(m − 1) · · · (m − n + 1)αm−n |2 = × |Cα,n |2 = 1.
|Cα,n |2 m=n
|Cα,n |2
Furthermore, direct computation gives

∞
X
1
|Fα(n) (α)| = × |m(m − 1) · · · (m − n + 1)αm−n |2 = Cα,n
Cα,n m=n
and we prove the claim. If α = 0, then it is easily seen that
F0 (z) = eiθ n!z n
(n)
are the extremal functions in this case because |F0 (α)| = (n!)2 . Therefore, we have completed
Problem 17.12
Proof. Since p ≥ 1, we know that f ∈ H 1 and Theorem 17.11 tells us that

Z π
1
f (z) = P (z, eit )f ∗ (eit ) dt
2π −π
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
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
µ(π) = f ∗ (eπi ) = lim f (reπi ) = lim f (re−πi ) = f ∗ (e−πi ) = µ(−π). (17.17)

r→1 r→1
On the one hand, if

∞
X
f (z) = an z n ,
n=0
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
n c∗
|r an − f (n)| = e−int it ∗ it
[f (re ) − f (e )] dσ ≤ |f (reit ) − f (eit )| dσ = kfr − f ∗ k1 → 0
T T
as r → 1 by Theorem 17.11. In other words, we get


 an , if n ≥ 0;
c∗
f (n) =

0, if n < ∞.
This implies that Z Z

π π
eint
µ(t) dt = eint f ∗ (eit ) dt = 0 (17.20)
−π −π
for every n = 1, 2, . . ..
Using the two facts (17.17) and (17.20), we find

Z π Z π Z π
int int π int
e dµ(t) = [e µ(t)]−π − µ(t) d(e ) = −in eint µ(t) dt = 0
−π −π −π
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) = it
P (z, e )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 [62, 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
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
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
kf k∞ = sup{|f (z)| | z ∈ K}.
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 217
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 P ∈ PK . In particular, we have

Z
e−int dµ = 0
T
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
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)
By the definition of an inner function, we know that

2 2
n 2 Z eit + z o 2
n Z eit + z o
p
Mh (z) = c exp −
p dµ h (t) = c p exp − dµ f (t) , (17.24)
p T eit − z it
T e −z
where µh is a finite positive Borel measure on T and µf = p2 µh . Clearly, µf is also a finite

positive Borel measure on T , so it follows from Theorem 17.15 that the right-most function is
in fact an inner function. Let it be Mf .
Next, according to [62, Eqn. (1), p. 344], we see that
2 n 1 Z eit + z 2 o
p
Qh (z) = exp · · log |h∗ (eit )| dt
2π T eit − z p
n 1 Z eit + z 2
o
= exp log (h∗ ) p (eit ) dt
2π T eit − z
n 1 Z eit + z o
∗ it
= exp log |f (e )| dt . (17.25)
2π T eit − 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 .
Finally, we note that the inequality

Z
1
log |h(0)| ≤ log |h∗ (eit )| dt (17.26)
2π T
is equivalent to the inequality

Z
1
log |f (0)| ≤ log |f ∗ (eit )| dt. (17.27)
2π T
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
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 [62, Eqn. (6), p. 91] and also the proof of Theorem 17.12, we
get
X∞ Z Z Z
2 2 1 1
|an | = lim |fr | dσ = lim p
dσ = ∗ p
dσ.
r→1 T r→1 T |ϕ| 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
In particular, we have |f (0)| = |a0 | ≤ 1 or equivalently, |ϕ(0)| ≥ 1. By Theorem 17.15,

every inner function M satisfies |M (z)| ≤ 1 in U . Combining this fact and Theorem
10.24 (The Maximum Modulus Theorem), we establish that ϕ is constant, a contradiction.
Consequently, ϕ1 ∈
/ H p for all p > 0.
(b) By the hypotheses, we have

n Z o
eit + z
ϕ(z) = c exp − dµ(t) ,
T eit − z
where |c| = 1, µ is a finite positive Borel measure on T , and µ ⊥ m. By Theorem 17.15

and Theorem 10.24 (The Maximum Modulus Theorem), we know that log |ϕ| is always
negative, i.e., 0 < |ϕ(z)| < 1 for every z ∈ U . Since ϕ ∈ H(U ) and ϕ(z) 6= 0 for all z ∈ U ,
it follows from Problem 11.5 that log |ϕ| is harmonic in U . Recall from [62, Eqn. (2),
§11.5, p. 233] that
Z eit + z Z
u(z) = − log |ϕ(z)| = Re dµ(t) = P (z, eit ) dµ(t)
T eit − z T
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
Proof. Suppose that

ϕ(z) − α
ϕα (z) =
1 − αϕ(z)
for every z ∈ U . It is clear that ϕα has no zero in U because α ∈
/ ϕ(U ). If we can show that ϕα
is nonconstant and inner, then it follows directly from Problem 17.17(b) that there is at least
one eiθ ∈ T such that ϕα (reiθ ) → 0 as r → 1. Equivalently, it means that
lim ϕ(reiθ ) = α. (17.28)

r→1
To this end, we directly apply the following result from [68, 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
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 [58, 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
Q1 (z) = cQ2 (z)
in U for some constant c with |c| = 1. Since Q is unique up to a constant of modulus 1, M is

also unique up to a constant of modulus 1 in the factorization (17.29), as required.
By the definition of g and Theorem 17.17 (The Canonical Factorization Theorem), we know
that Qg = Q1f . Consequently, this fact and the above claim show immediately that Qf Qg =
Mf Mg = 1. By Theorem 17.15 again, Mf Mg = 1 implies that Mf = 1 and hence f = Qf ,
Problem 17.20
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
∗ it
fǫ (z) = c exp log (f ǫ ) (e ) 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
∗ it
f (z) = c exp log |f (e )| dt = Qf (z)
2π −π eit − z
for all z ∈ U . This completes the proof of the problem.

Problem 17.21
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
+ + it
log |fr | dt = log |f (re )| dt ≤ log |h(reit )| dt (17.31)
−π −π −π
for 0 ≤ r < 1. Since h(z) 6= 0 in U , Theorem 15.18 (Jensen’s Formula) gives

Z π
1
log |h(reit )| dt = log |h(0)|. (17.32)
2π −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
= it
log |fr (e )| 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 it
log |fr (e )| dt + log− |fr (eit )| dt
0≤r<1 2π T 2π T
<∞
By Theorem 11.30(a), there is a unique complex Borel measure µ on T such that

Z
u(z) = P [ dµ] = P (z, eit ) dµ(eit ). (17.34)
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 .
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
dµ(eit ).
T e −z
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+ )
Problem 17.22
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≤ iθ
log |f (r1 e )| 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
iθ iθ 1
0 = lim log |f (re )| dθ = lim log |g(re )| dθ = lim log dθ.
r→1 −π r→1 −π r→1 −π |g(reiθ )|
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
Problem 17.23
Proof. Let M1 6= 0 and M2 6= 0 be two closed S-invariant subspaces of H 2 . By Theorem 17.21

(Beurling’s Theorem), there exist inner functions ϕ1 and ϕ2 such that
M1 = ϕ1 H 2 and M2 = ϕ2 H 2 .
By Theorem 17.15, we have

n Z π o
eit + z
ϕ1 (z) = c1 B1 (z) exp − dµ 1 (t) (17.38)
−π eit − z
and Z
n π
eit + z o
ϕ2 (z) = c2 B2 (z) exp − dµ 2 (t) (17.39)
−π eit − 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, B2
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
where c, B and µ satisfy the conditions of Theorem 17.15. Then we have
B1 (z) = B2 (z)B(z) (17.40)
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
for all z ∈ U . Put z = 0 and recall that the measures are finite positive, we get C = 0. Recall
from the formula following [62, Eqn. (8), p. 111] that
X∞
eit + z
= 1 + 2 (ze−it )n , (17.42)
eit − z n=1
where z ∈ U . Hence we obtain

Z π X∞ Z π ∞
X
n −int

1+2 z e dµ1 (t) = 1+2 z n e−int d µ(t) + µ2 (t)
−π n=1 −π n=1
Z π ∞ hZ π
X i Z π
dµ1 (t) + 2 e−int dµ1 (t) z n = d µ(t) + µ2 (t)
−π n=1 −π −π
∞ hZ
X π i
+2 e−int d µ(t) + µ2 (t) z n .
n=1 −π
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
−int

e 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
for every n ∈ Z. By Theorem 17.13 (The F. and M. Riesz Theorem), we have µ + µ2 − µ1 ≪ m.

We know that µ + µ2 − µ1 ⊥ m, so Proposition 6.8 implies that µ + µ2 − µ1 = 0 or µ1 = µ + µ2 .
This means that
µ2 (E) ≤ µ1 (E)
for every Borel subset E of T , as desired. This completes the analysis of the problem.
17.4 A Projection of Lp onto H p
Problem 17.24
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 [21,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 ,
where v is the harmonic conjugate of u.

17.4. A Projection of Lp onto H p 225
Now we consider the function

1+z
f (z) = i log = u(z) + iv(z)
1−z
which maps U conformally onto the vertical strip {z ∈ C | − π2 < Re z < π2 }. In other words, u
is bounded, but v is not so that M. Riesz’s Theorem fails in the case p = ∞.
For p = 1, recall the facts [62, Eqn. (2) & (3), §11.5, p. 233] that
1 + z Z π
1
Pr (θ) = Re and sup kPr (θ)k1 = Pr (θ) dθ < ∞.
1−z 0≤r<1 2π −π
1+z
Denote f (z) = 1−z . If f ∈ H 1 , then Theorem 17.11 gives
Z Z Z
1 π
1 + eit 1 π
dt 1 π
dt 2h t iπ
kf k1 = kf ∗ k1 = dt = t = π = ln sin =∞
2π −π 1−e it 2π −π | tan 2 | 0 tan 2t π 2 0
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 [21, Theorem 4.2, p. 57].
Problem 17.25
Proof.
(a) Let z ∈ U . Note that

1 X z n
∞
1
= ,
eit − z eit n=0 eit
so we obtain
Z
eit + z X z n it
π ∞
1
(ψf )(z) = f (e ) dt
2π
−π eit n=0 eit
∞ h
X Z π i X∞ h Z π i
1 it −int 1
= f (e )e n
dt · z + f (eit )e−int dt · z n
2π −π 2π −π
n=0 n=1
X∞ ∞
X
= fb(n)z n + fb(n)z n
n=0 n=1
X∞
= 2fb(n)z n − fb(0). (17.45)
n=0
Define
∞
(ψf )(z) + fb(0) X b
F (z) = = f (n)z n .
2
n=0
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
b 1
= f (m) · ei(m−n)t dt
2π −π
m=0

 fb(n), for all n ≥ 0;
=

0, for all n < 0.
Next, we notice that Theorem 3.5 (Hölder’s Inequality) gives

Z π
b 1
|f (n)| = f (eit )e−int dt ≤ kf k1 ≤ kf kp (17.46)
2π −π
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
is a bounded linear projection.
(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
Similarly, we have F ∈ H p , g = F ∗ ∈ Lp (T ) and then


X∞
1
Z π  fb(n), for all n ≥ k;
gb(n) = b
f (m) · ei(m−n)t
dt = (17.47)
2π −π 
m=k 0, for all 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
= Cp,k · kf kp ,
where Cp,k = Ap + k + 21 , a constant depending only on p and k.

Similarly, if k < 0, then we define
1
F (z) = (ψf )(z) + fb(0) + 2fb(−1)z −1 + 2fb(−2)z −2 + · · · + 2fb(k)z k
2
X∞
= fb(n)z n .
n=k
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
kg1 kp ≤ Cp kf kp and kg2 kp ≤ Cp kf kp
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 every t ∈ T . By Theorem 3.5 (Minkowski’s Inequality), we see that
ksn (f ) − f kp ≤ kf − gkp + kg − sn (g)kp + ksn (g) − sn (f )kp .
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→∞
(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
|n| 1
= r × 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 [31, p. 152].
(b) We may regard H p as a subspace of Lp (T ) by Theorem 17.11(d), where 1 ≤ p ≤ ∞.

In addition, H p is closed in Lp (T ). To see this, let f be a limit point of H p . Given
ǫ > 0. There exists an N ∈ N such that kfn − f kp < 2ǫ , where fn ∈ H p . For m, n ≥ N ,
the inequality
kfm − fn kp ≤ kfm − f kp + kf − fn kp < ǫ
implies that {fn } is Cauchy. Note that H p is Banach by Remark 17.8(c), so f ∈ H p
and then H p is closed in Lp (T ).
Problem 17.26
Proof. Take p = 2. Denote H = ψh which is holomorphic in U . Therefore, we follow from the

expression (17.42) that
Z π ∞
X
1
H(z) = 1+2 z n e−int h(eit ) dt
2π −π n=1
Z π ∞
X h 1 Z π i
1
= it
h(e ) dt + 2 zn
h(eit )e−int dt
2π −π 2π −π
n=1
∞
X
=b
h(0) + 2b
h(n)z n
n=1
and then
∞
X ∞
X
H(re ) = b
iθ
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 [62, 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
for n = 0, −1, −2, . . ..

Problem 17.27
Proof. Define g : U → C by
∞
X
g(z) = |an |z n .
n=0
By Theorem 10.6, we have f, g ∈ H(U ) so that

∞
X ∞
X
′
|f (z)| = nan z n−1
≤ n|an | · |z|n−1 = g ′ (|z|)
n=1 n=1
and thus for every x ∈ (0, 1), we have

Z x Z x ∞
X
′ iθ
|f (re )| dr ≤ g′ (r) dr = g(x) − g(0) ≤ |an | < ∞.
0 0 n=0
Since this is true for every 0 < x < 1, we actually have

Z 1
|f ′ (reiθ )| dr < ∞,
0
Problem 17.28
Proof. Suppose that {nk } is a sequence of positive integers.

nk 1 1
• |f ′ (z)| > 10k for all 1 − nk < |z| < 1 − 2nk . Let n1 be a sufficiently large positive integer
and √
nk > 4k(k − 1) e · exp(nk−1 ) (17.53)
1 1
for k ≥ 2. Then we have nk > k which implies nk > k > k > nk and
1
1 1 n
1 < 1 < nk k .
nk k
n
k nk
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
k−1
X ∞
X
nk |z|nk −1 np z np −1 np z np −1
|f ′ (z)| ≥ − − (17.54)
k p p
p=1 p=k+1
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
and the property (17.53) shows that

k−1
X k−1
X
np z np −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 n
≈ √p
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
np h 1 2nk i 2nk
∞
X ∞
X np −1
np z 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
for every p ≥ k + 1. Then the estimate (17.57) becomes

∞
X ∞
np z np −1 nk X 1 nk
< = . (17.58)
p 4ek 2p−k 4ek
p=k+1 p=k+1
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
∞ 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
for every θ. Hence we have Z 1

|f ′ (reiθ )| dr = ∞ (17.59)
0
for every θ.
• The convergence of the limit. Let 0 < R < 1. By the definition of f , we have
Z R Z RX∞
′ iθ nk r nk −1 i(nk −1)θ
f (re ) dr = e dr
0 0 k=1 k
Z
nk R nk −1 i(nk −1)θ
X∞
= r dr e
k 0 k=1
∞
X R nk
= ei(nk −1)θ
k
k=1
−iθ
=e f (Reiθ ). (17.60)
∞
X 1
As < ∞, Theorem 17.12 implies that f ∈ H 2 and then we follow from Theorem
k2
k=1
17.11 that f ∗ (eiθ ) exists a.e. on T . Thus we deduce from the expression (17.60) that
Z R
lim f ′ (reiθ ) dr = e−iθ f ∗ (eiθ )
R→1 0
exists for almost all θ.

• The geometrical meaning of the integral (17.59). If f ∈ H(U ), we denote
Z 1
V (f ; θ) = |f ′ (reiθ )| dr
0
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
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
and fc∗ (n) be the Fourier coefficients of its boundary function f ∗ (eit ), i.e.,
Z π
c 1
∗
f (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
1 1
n c∗
r an − f (n) = e−int it ∗ it
f (re ) − f (e ) dt ≤ f (reit ) − f ∗ (eit ) dt
2π −π 2π −π
for every n ∈ Z. Applying Theorem 17.11, we obtain
r n 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 −int 1
e it
g(e ) 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
iθ 1
f (re ) = 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
it follows from the formula (17.63) that

Z π
1
f (reiθ ) = Pr (θ − t)g(eit ) dt
2π −π
X∞ Z π X∞ Z π
1 1
= n in(θ−t)
r e it
g(e ) dt + e−in(θ−t) g(eit ) dt
n=0
2π −π n=1
2π −π
X∞ h 1 Z π i
= (reiθ )n e−int g(eit ) dt
2π −π
n=0
X∞
= gb(n)(reiθ )n
n=0
which means
∞
X
f (z) = g(n)z n
b
n=0
for every z ∈ U and then f ∈ H(U ). By the integral (17.64) and [62, 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
We observe from Theorem 17.11 that

Z π
lim |f (ρeit ) − f ∗ (eit )| dt → 0,
ρ→1 −π
so we obtain
f (z) = lim f (ρz) = Φ(z)
ρ→1
for every z ∈ U , i.e., Z π

1
f (reiθ ) = Pr (θ − t)f ∗ (eit ) dt. (17.65)
2π −π
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
18.1 Examples of Banach Spaces and Spectrums
Problem 18.1
Proof. Let X be a Banach space and
B(X) = {A : X → X | A is linear and bounded with mentioned conditions}. (18.1)
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
α(A1 A2 )(x) = αA1 (A2 x) = A1 (αA2 x) = (αA1 )(A2 x),
i.e., α(A1 A2 ) = A1 (αA2 ) = (αA1 )A2 . Thus B(X) is a complex algebra.

We check Definition 5.2. By the definition, kAk must be nonnegative. By the definition
(18.1), there exists a positive constant M such that kAxkX ≤ M kxkX for all x ∈ X. Therefore,
we have
kAxkX
kAk = sup ≤ M,
kxkX
i.e., kAk is a real number. Next, for all A1 , A2 ∈ B(X), we apply the fact that X is a Banach,
so we have
k(A1 + A2 )(x)kX
kA1 + A2 k = sup
kxkX
kA1 x + A2 xkX
= sup
kxkX
kA1 xkX + kA2 xkX
≤ sup
kxkX
= kA1 k + kA2 k.
Furthermore, if α ∈ C and A ∈ B(X), then we obtain
k(αA)(x)kX kAxkX
kαAk = sup = |α| sup = |α| · kAk.
kxkX kxkX
235
236 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.
Hence B(X) is also a normed complex algebra.

We claim that B(X) is a complete metric space. Fix x ∈ X. Given ǫ > 0. Let {An } be
Cauchy with respect to the norm k · k. Then it suffices to show that there exists an A ∈ B(X)
such that
kAn − Ak → 0
as n → ∞. Since there exists an N ∈ N such that kAm − An k < 1 for all n, m ≥ N . Using the
triangle inequality of the norm k · k, we see that
kAn k < 1 + kAN k
for all n ≥ N . Denote M = max{kA1 k, kA2 k, . . . , kAN −1 k, 1 + kAN k}. Thus we get
kAn k < 1 + kAN k
for all n ∈ N. This means that

kAn xkX ≤ M kxkX (18.2)
for all x ∈ X and n ∈ N. Now, for all x ∈ X, we can establish
kAm x − An xkX = k(Am − An )xkX ≤ kAm − An k · kxkX → 0
as n, m → ∞. In other words, {An x} is a Cauchy sequence in X. Since X is Banach, the sequence

converges to an element in X, namely Ax. Then we can define the operator A : X → X by
this and the linearity of taking limits implies the linearity of A. Besides, we follow from the
kAxkX = lim kAn xkX ≤ M kxkX
n→∞
for every x ∈ X. Consequently, it means that A ∈ B(X).
Now it remains to verify that kAm − Ak → 0 as m → ∞. Recall that {An } is Cauchy, so
given ǫ > 0, there exists an N ∈ N such that
kAm − An k ≤ ǫ
whenever n, m ≥ N . Therefore, for every x ∈ X, we have
kAm x − An xkX ≤ kAm − An k · kxkX < ǫ · kxkX
whenever n, m ≥ N . If n → ∞, then we see that
kAm x − AxkX ≤ ǫ · kxkX
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
18.1. Examples of Banach Spaces and Spectrums 237
Problem 18.2
Proof. Suppose that we have
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 [35, 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
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.
By Definition 5.3, Mϕ is bounded.

For the second assertion, recall from [62, Exercise 19, p. 74] that

Rϕ = λ ∈ C m {x | |ϕ(x) − λ| < ǫ} > 0 for every ǫ > 0 . (18.4)
Let I be the identity operator on L2 . We are required to prove that
σ(Mϕ ) = {λ ∈ C | Mϕ − λI is not invertible} = Rϕ . (18.5)

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
Thus we have M 1 is the inverse of Mϕ − λI which means λ ∈

/ σ(Mϕ ).
ϕ−λ
On the other hand, let λ ∈ Rϕ . For any n ∈ N, we denote
Sn = {x | |ϕ(x) − λ| < 2−n }.
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
k(Mϕ − λI)(φn )k2 = kϕφn − λφn k2 = |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
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) = {λ ∈ C | S − λI is not invertible}.
Since kSk = 1, Corollary 3 to Theorem 18.4 implies that
σ(S) ⊆ U . (18.6)
Take 0 < |λ| < 1. Assume that λ ∈

/ σ(S). Then S − λI is invertible so that the equation
(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 239
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.
18.2 Properties of Ideals and Homomorphisms
Problem 18.5
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 .
Problem 18.6
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
I = {f ∈ C(X) | f (p) = 0}. (18.9)
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
F (x) = f12 (x) + f22 (x) + · · · + fN

2
(x). (18.10)
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.
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
Problem 18.7
/ σ(x). Then x − λe is invertible so that (x − λe)−1 ∈ A.

Proof. Let e be the unit of A. Given λ ∈
Since A is generated by a single element x, this means that there are polynomials Pn such that
Pn (x) → (x − λe)−1 (18.11)
as n → ∞ in A. If z ∈ σ(x), then Theorem 18.17(b) ensures that h(x) = z for some h ∈ ∆.

Since h is a complex homomorphism of A, we have h(xm ) = z m for every m ∈ Z. By this and
Theorem 18.17(e), we establish that

|Pn (z) − (λ − z)−1 | = h Pn (x) − (λ − x)−1 ≤ kPn (x) − (λ − x)−1 k. (18.12)
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
−1
dz ℓLn
2πi = Pn (z) − (z − λ) dz ≤ ℓ · Ln = · 2πδ = 2πℓLn . (18.15)
C(λ;δ) C(λ;δ) |z − λ| δ
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
∞
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 int
= an e = 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
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 [60, 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
for all f ∈ I and x ∈ R. On the one hand, we note that

Z
(f ∗ g ∗ φ)(0) = [(f ∗ g) ∗ φ](0) = (f ∗ g)(0 − y)φ(y) dm(y). (18.17)
R
On the other hand, recall from [62, 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
∗
isomorphic to the dual space of L∞ (R), i.e., L1 (R) ∼ = L∞ (R). By Remark 5.21, every
∞ 1
φ ∈ L (R) is a bounded linear functional on L (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 [60, 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
• L1 (T ) is a commutative Banach algebra. Suppose that f, g, h ∈ L1 (T ). It is clear

that f ∗ (g + h) = f ∗ g + f ∗ h, (f + g) ∗ h = f ∗ h + g ∗ h and α(f ∗ g) = f ∗ (αg) = (αf ) ∗ g
for every α ∈ C. It also satisfies the associative law by an application of Theorem 8.8 (The
Fubini Theorem)d Thus L1 (T ) is a complex algebra.
b
By Theorem 8.14, we know that f ∗ g ∈ L1 (R).
c
Recall that fx0 (y) = f (y − x0 ), see Theorem 9.5.
d
See also [62, Example 9.19(d), p. 190].
Recall from [62, 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 π
1 1
= 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 .
In other words, L1 (T ) is a Banach algebra by Definition 18.1. The definition implies

immediately that f ∗ g = g ∗ f . Hence L1 (T ) is commutative.
• 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 [62, §5.14, p. 103].
• Complex homomorphisms of L1 (T ). Denote ∆T to be the set of all complex homo-

morphisms of L1 (T ). Let ϕ ∈ ∆T and ϕ 6= 0. By Theorem 18.17(e), ϕ is bounded by 1,
so it follows from Theorem 6.16 that there is a unique β ∈ L∞ (T ) such that
Z π
1
ϕ(f ) = f (t)β(t) dt. (18.23)
2π −π
On the one hand, we have

Z π Z π Z π
1 1
ϕ(f ∗ g) = (f ∗ g)(t)β(t) dt = f (t − s)g(s)β(t) ds dt.
2π −π (2π)2 −π −π
On the other hand, we obtain

h 1 Z π i h 1 Z π i
ϕ(f )ϕ(g) = f (t)β(t) dt × g(s)β(s) ds
2π −π 2π −π
Z πZ π
1
= f (t − s)g(s)β(t − s)β(s) ds dt.
(2π)2 −π −π
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
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 a unique integer n.
• IE is a closed ideal in L1 (T ). Let E ⊆ Z and

IE = f ∈ L1 (T ) fb(n) = 0 for all n ∈ E . (18.24)
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
for every n ∈ E. Thus we have f ∗ g ∈ IE and IE is an ideal in L1 (T ). For every α ∈ R,

we follow from the definition that
fb(n − α) = fb(n)e−iαn = 0
for every n ∈ E, so f (x − α) ∈ IE . This means that IE contains every translate of f and

Remark 18.1 ensures that IE is closed.
• 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
c int c
exists an fn ∈ I such that fn (n) 6= 0. Put gn (t) = e fn (n) 1
∈ L (T ). Then we have
Z π Z π
1 1 eint 1
(gn ∗ fn )(t) = · in(t−s)
fn (s)e ds = · fn (s)e−ins ds = eint .
c
fn (n) 2π −π c
fn (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.,
E = {n ∈ Z | fb(n) = 0 for all f ∈ I}. (18.25)
Thus it is easy to see that I ⊆ IE . If n ∈ / E, then there exists an g ∈ I such that b

g(n) 6= 0.
For simplicity, we may assume that this number is 1. Regarding eint as an element of
L1 (T ), we see that (g ∗ ein )(t) = −eint . Since I is an ideal, we have g ∗ ein ∈ I and then
eint ∈ I. Thus I contains every trigonometric polynomial of the form
X
an eint
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 [23, Theorem 8.19, p. 247].
i.e., E ⊂ Z(f ∗ g). Using [60, 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 [81,
§9, pp. 39 – 43].
Problem 18.11
Proof. Notice that λ, µ ∈ C \ σ(x). On the one hand, we have

(x − λe) R(λ, x) − R(µ, x) (x − µe) = (x − λe)(λe − x)−1 − (x − λe)(µe − x)−1 (x − µe)
= −e(x − µe) − (x − λe)(µe − x)−1 (x − µe)
= −x + µe + x − λe
= (µ − λ)e. (18.26)
On the other hand, we see that

(x − λe) (µ − λ)R(λ, x)R(µ, x) (x − µe) = (µ − λ) (x − λe)R(λ, x)R(µ, x)(x − µe)
= (µ − λ)e. (18.27)
It yields from the results (18.26) and (18.27) that
R(λ, x) − R(µ, x) = (µ − λ)R(λ, x)R(µ, x) (18.28)
holds for all λ, µ ∈ C \ σ(x).

We follow from the identity (18.28) that
(x − µe)−1 − (x − λe)−1 R(λ, x) − R(µ, x)

= = R(λ, x)R(µ, x) → R(λ, x)2 = (x − λe)−2
µ−λ µ−λ
as µ → λ. This is exactly [62, 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
Proof. Denote M be the set of maximal ideals of A. Let M be a maximal ideal of A. By

Theorem 18.17, we have M = ker h for some h ∈ ∆. Conversely, if h ∈ ∆, then it follows from
the First Isomorphism Theorem [25, Theorem 16.2, p. 145] that
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
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
2 t2
T (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
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.
• The mapping G is a homomorphism. Suppose that x, y ∈ A, α ∈ C and h ∈ ∆. Then

we see that
d
(αx)(h) = h(αx) = αh(x) = (αb
x)(h)
\
(x b(h) + yb(h) = (b
+ y)(h) = h(x + y) = h(x) + h(y) = x x + yb)(h)
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.
Thus if rad A = {0},f then we get A ∼

=Ab and G becomes an isomorphism.
• ρ(x) = kbxk∞ = sup{|b

x(h)| | h ∈ ∆}. By Theorem 18.17(e), we have ρ(x) ≥ |h(x)| = |b
x(h)|
for every h ∈ ∆ which means that
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
ρ(x) = sup{|λ| | λ ∈ σ(x)} ≤ sup{|b

x(h)| | h ∈ ∆} = kxk∞ .
• 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
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
(x, λ)(y, µ) = (xy + µx + λy, λµ). (18.31)

f
In this case, A is called semisimple.
• A1 is a commutative Banach algebra with unit. It is easily checked that this is

associative and distributive. Thus A1 is a complex algebra. Furthermore, the element
(0, 1) is a unit for this multiplication because
(x, λ)(0, 1) = (x · 0 + 1 · x + λ · 0, λ · 1) = (x, λ) = (0, 1)(x, λ).
For any (x, λ), (y, µ) ∈ A1 , we see that
k(x, λ) + (y, µ)k = k(x + y, λ + µ)k

= kx + yk + |λ + µ|
≤ kxk + kyk + |λ| + |µ|
= k(x, λ)k + k(y, µ)k.
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
k(x, λ)(y, µ)k = k(xy + µx + λy, λµ)k

= kxy + µx + λyk + |λµ|
≤ kxyk + kµxk + kλyk + |λ| · |µ|
≤ kxkkyk + |µ| · kxk + |λ| · kyk + |λ| · |µ|

= kxk + |λ| · kyk + |µ|
= k(x, λ)k · k(y, µ)k.
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).
• The mapping x 7→ (x, 0) is an isometric isomorphism of A onto a maximal ideal

of A1 . Let Φ be this mapping. It is trivial surjective. If Φ(x) = Φ(y), then (x, 0) = (y, 0)
which means that (x − y, 0) = 0. Since k(x − y, 0)k = 0, we get x = y and thus Φ is
injective. Is is easily checked that Φ satisfies
Φ(x + y) = Φ(x) + Φ(y),
so Φ is an isomorphism onto Φ(A). It is also isometric because
kΦ(x) − Φ(y)k = kΦ(x − y)k = k(x − y, 0)k = kx − yk.
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.

18.3. The Commutative Banach algebra H ∞ 249
18.3 The Commutative Banach algebra H ∞
Problem 18.16
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 [62, 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
Proof. By Problem 18.16, H ∞ is a commutative Banach algebra. Let I = {(z − 1)2 f | f ∈ H ∞ }.

For any f, g ∈ H ∞ , we know that f g ∈ H ∞ . Thus if (z − 1)2 f ∈ I and g ∈ H ∞ , then we have
g · (z − 1)2 f = (z − 1)2 f g ∈ I.
By Definition 18.12, I is an ideal of H ∞ .

Given ǫ > 0. The function fǫ (z) = (1 + ǫ − z)−1 belongs to H ∞ because
|1 + ǫ − z| ≥ |1 + ǫ| − |z| > ǫ
so that |fǫ (z)| < ǫ−1 for all z ∈ U . We observe that

ǫ(1 − z)
(1 − z)2 (1 + ǫ − z)−1 − (1 − z) = <ǫ
1−z+ǫ
for all z ∈ U . This means that (z − 1)2 fǫ (z) converges uniformly to 1 − z in U . However, we
know that 1− z ∈ / H ∞ which means that I is not closed. This ends the proof of the problem.
Problem 18.18
Proof. Denote I = {ϕf | f ∈ H ∞ }. Of course, we have ϕf ∈ H ∞ . Since f g ∈ H ∞ for any

f, g ∈ H ∞ , the space I is an ideal of H ∞ by Definition 18.12. Let {fn } be a sequence in H ∞
such that
kϕfn − gk∞ → 0
as n → ∞, where g ∈ H ∞ . Thus {ϕfn } is a Cauchy sequence by [61, Theorem 3.11(a), p. 53]
and hence so is {fn }. Observing from Remark 17.8(c) that H ∞ is Banach, so it is complete.
Then we have fn → f ∈ H ∞ as n → ∞ and this means that ϕf ∈ I. Hence we complete the

CHAPTER 19
Holomorphic Fourier Transforms
19.1 Problems on Entire Functions of Exponential Type
Problem 19.1
Proof. By the hypothesis, we know that there exist some constants A and C such that
|f (z)| ≤ CeA|z|
for all z ∈ C. Suppose for simplicity that ϕ(0) < ∞, i.e.,

Z ∞
|f (x)|2 dx < ∞.
−∞
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
for all z ∈ C. Define F1 and F2 by

 
 F (t), if 0 ≤ t < A;  F (t), if −A < t ≤ 0;
F1 (t) = and F2 (t) =
 
0, if t ≥ A 0, if t ≤ −A.
Then we may express f as

Z ∞ Z 0
itz
f (z) = f1 (z) + f2 (z) = F1 (t)e dt + F2 (t)eitz dt. (19.1)
0 −∞
By the definition, it is clear that F1 ∈ L2 (0, ∞), so we know from [62, 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
251
252 Chapter 19. Holomorphic Fourier Transforms
where F f2 (t) = F2 (−t). Since F2 ∈ L2 (−∞, 0), we have F f2 ∈ L2 (0, ∞). By similar argument as
in [62, 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 f 2 2ty f2 (t) 2 dt < ∞
|f (x + iy)| dx = F2 (t) e dt ≤ F (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 ∞
2 1n ∞ 2 2
o1
2
≤ |F1 (t)| dt + |f1 (x + iy)| dx · |f2 (x + iy)|2 dx
0 π −∞ −∞
Z ∞
+ f2 (t) 2 dt
F
0
f2 f2 2
= kF1 k22 + 2kF1 k2 · F 2
+ F 2
2
= F1 k2 + kF f2
2
<∞
for every real y. This proves the first assertion.

For the second assertion, we suppose that ϕ is a bounded function. By Theorem 19.3 (The
Paley and Wiener Theorem), there exists an F ∈ L2 (−A, A) such that
Z A Z A
f (x + iy) = f (z) = F (t)eitz dt = F (t)e−ty · e−itx dt (19.4)
−A −A
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.
−∞
Observing from Theorem 9.13 (The Plancherel Theorem), we have

Z ∞ Z ∞ Z A
2 −ty 2
ϕ(y) = |f (x + iy)| dx = 2π |F (t)e | dt = 2π e−2ty |F (t)|2 dt.
−∞ −∞ −A
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
19.1. Problems on Entire Functions of Exponential Type 253
Proof. Since f is of exponential type, there exist some constants A and C such that
|f (z)| ≤ C exp(A|z|) (19.5)
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
−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
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
Proof. Now f is an entire function of exponential type.
• 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 [11, 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).
This means that

1
lim sup |n!an | n = A
n→∞
and so the power series Φ converges if

1 1 1
< 1 = ,
|w| lim sup |n!an | n A
n→∞
i.e., |w| > A.b
• The function f can be expressed as an integral. By the power series of Φ, we see

that
Z Z X
n!an X z k wk
∞ ∞
1 1
Φ(w)ewz dw = × dw
2πi Γ 2πi Γ wn+1 k!
n=0 k=0
Z X∞ m
1 am z
= dw
2πi Γ w
m=0
∞ Z
1 X m dw
= am z
2πi Γ w
m=0
∞
X
= am z m
m=0
= f (z),
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)
for every real α.d If Re (weiα ) > 3A, then we have

| exp(−wseiα )| = exp − sRe (weiα ) < e−3As . (19.7)
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 [11, §5.3, p. 73] or
[36, §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}.
of the power series of f satisfies

∞
X s k Mf (r) s n+1
|Rn (s)| ≤ Mf (r) = · .
r 1 − rs r
k=n+1
By putting r = 2s, we get

e2s
|Rn (s)| ≤ (19.8)
2n
Therefore, we deduce from the inequalities (19.7) and (19.8) that the series
∞
X
an e−wz z n
n=0
converges uniformly on the ray Γα = {seiα | s ≥ 0}. Consequently, for every w ∈ Πα , we

obtain
Z ∞
X Z ∞
X
−wz n!an
Φα (ω) = e f (z) dz = an z n e−wz dz = = Φ(w)
Γα Γα wn+1
n=0 n=0
which is exactly the expression (19.6).
Problem 19.5
Proof. Suppose that f ∈ H(Π+ ) and

Z ∞
1
sup |f (x + iy)|2 dx = C < ∞. (19.9)
0<y<∞ 2π −∞
• 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:
Figure 19.1: The closed contour Γr .

Using Theorem 10.15 (The Cauchy’s Formula in a Convex Set), we have

Z
1 f (ω)
f (z) = dω
2πi Γr ω − z
Z r Z π
1 f (ξ + iǫ) 1 ireiθ f (reiθ + iǫ)
= dξ + dθ. (19.10)
2πi −r ξ + iǫ − z 2πi 0 reiθ + iǫ − z
By Theorem 19.2 (The Paley and Wiener Theorem), there exists an F ∈ L2 (0, ∞) such
that Z ∞
f (z) = F (t)eitz dt,
0
where z ∈ Π+ . Take z = reiθ + iǫ = r cos θ + i(r sin θ + ǫ), where 0 < θ < π. Since
itz = −t(r sin θ + ǫ) + itr cos θ, we have
|eitz | = e−t(r sin θ+ǫ) ≤ e−tr sin θ , (19.11)
where 0 < θ < π. According to Theorem 3.8 and the estimate (19.11), we obtain
Z ∞
2 2 e−2tr sin θ ∞ kF k22
iθ
|f (re + iǫ)| ≤ kF k2 × e−2tr sin θ dt = kF k22 × = . (19.12)
0 −2r sin θ 0 2r sin θ
Substituting the bound (19.12) into the second integral in the formula (19.10), we obtain
Z π Z π
ireiθ f (reiθ + iǫ) rkF k2 dθ
iθ + iǫ − z
dθ ≤ √ 1
0 re 2r(r − |iǫ − z|) 0 sin 2 θ
√ Z π
2rkF k2 2 dθ
= . (19.13)
r − |iǫ − z| 0 sin 21 θ
By the inequality π2 ≤ sinθ θ for 0 < θ ≤ π2 , we can show that the integral in the estimate
(19.13) is bounded so that
Z π
ireiθ f (reiθ + iǫ)
lim dθ = 0
r→∞ 0 reiθ + iǫ − z
and consequently, the expression (19.10) gives
Z ∞
1 f (ξ + iǫ)
f (z) = dξ
2πi −∞ ξ + iǫ − z
for every 0 < ξ < y.
• f ∗ (x) = lim f (x + iy) exists for almost all x. Notice that the linear fractional trans-
y→0
1+z
formation ϕ(z) = i 1−z maps U conformally onto Π+ and ϕ(T ) = R ∪ {∞}. If we consider
g = f ◦ ϕ ∈ H(U ), then the condition (19.9) implies definitely that
Z
|g(z)|2 dθ dr < ∞.
U
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
exists for almost all x.

• The relation between f ∗ and F . The second assertion implies

Z ∞
∗
f (x) = lim f (x + iy) = lim F (t)e−ty+itx dt. (19.15)
y→0 y→0 0
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
−t 2 kF k2
|f1 (t)| dt = |F (t)|e 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 almost all x ∈ R.
• The case when ǫ = 0. Let z = x + iy with y > 0. Notice that

f (ξ + iǫ) |f (ξ + iǫ)|
≤
ξ + iǫ − z |ξ − z|
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|
nZ ∞ o1 n Z ∞ dξ o 12
2
≤ |f (ξ + iǫ)|2 dξ × 2
−∞ −∞ (ξ − x)
Z
n ∞
√ dξ o 21
≤ 2Cπ · 2
−∞ (x − ξ)
< ∞,
so Theorem 1.34 (The Lebesgue’s Dominated Convergence Theorem) ensures that

Z ∞ ∗
1 f (ξ)
f (z) = dξ.
2πi −∞ ξ − z
Problem 19.6
Proof. Here we follow mainly [50, 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 )
In other words, we have Z ∞

| log ϕ(x)|
dx < ∞.
−∞ 1 + x2
We write z = x + iy with y > 0 and consider
Z
1 ∞ y
u(z) = log ϕ(t) dt. (19.17)
π −∞ (x − t)2 + y 2
Using the half-plane version of Fatou’s Theorem [55, Theorem 5.5, pp. 86, 87], we see that the
function (19.17) is harmonic in Π+ and
lim u(x + iy) = log ϕ(x) or lim |f (x + iy)| = ϕ(x) (19.18)

y→0 y→0
holds for almost all x ∈ R.e Let v(z) be its harmonic conjugate and write
f (z) = exp(u(z) + iv(z)).
Because of [62, Eqn. (7), p. 63], we see that

Z ∞
1 ϕ(t)y
|f (x + iy)| = eu(z) ≤ dt,
π −∞ (x − t)2 + y 2
it follows from Theorem 3.5 (Hölder’s Inequality) that
Z Z ∞
2 1 ∞ ϕ(t)y ϕ(s)y
|f (x + iy)| ≤ 2 2 2
dt × 2 2
ds
π −∞ (x − t) + y −∞ (x − s) + y
Z o 1 nZ ∞
1n ∞ |ϕ(t)|2 y 2 y o1
2
≤ 2 2 + y2
dt 2 + y2
dt
π −∞ (x − t) −∞ (x − t)
nZ ∞ |ϕ(s)|2 y o 1 nZ ∞ y o1
2 2
× 2 + y2
ds 2 + y2
ds
−∞ (x − s) −∞ (x − s)
Z
1 ∞ |ϕ(t)|2 y
= dt
π −∞ (x − t)2 + y 2
which implies
Z ∞ Z Z
1 ∞ ∞ |ϕ(t)|2 y
|f (x + iy)|2 dx ≤ dt dx
−∞ π −∞ −∞ (x − t)2 + y 2
Z Z ∞
1 ∞ 2 y
= |ϕ(t)| dt · 2 2
dx
π −∞ (x − t) + y
Z ∞−∞
= |ϕ(t)|2 dt.
−∞
In other words, we have established

Z ∞
sup |f (x + iy)|2 dx < ∞.
y>0 −∞
According to Theorem 19.2 (The Paley-Wiener Theorem), there exists an F ∈ L2 (−∞, ∞)

vanishing on (−∞, 0) such that Z ∞
−∞
e
Here the function
y
(x − t)2 + y 2
is the Poisson kernel in the upper half plane Π+ . See also [7, pp. 145 – 147; Theorem 7.28, pp. 160, 161].
for all z ∈ Π+ . In particular, we have

Z ∞ Z ∞
lim f (x + iy) = f (x) = F (t)e itx
dt = F (t)eitx dt = Fb(−x) (19.19)
y→0 0 −∞
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 [78, Lemma 9.3, p. 286] that G(x) b = F (−x) which
vanishes on [0, ∞).
Conversely, we suppose that there exists an f with fb = 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 Z Z
π ∞
|fb(x)|2 ∞
2
|K(eiθ )|2 dθ = 2 dx ≤ 2 |fb(x)|2 dx = 2 fb ≤ 2kf k22 = 2kϕk22 < ∞.
−π −∞ 1 + x2 −∞
2
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 ).
By Definition 11.6, k is the Poisson integral of K, i.e., k = P [K]. Since K ∈ L2 (T ) and

k = P [K], it follows from Theorem 11.16 that
Z π Z π Z π
log+ |k(reiθ )| dθ ≤ |k(reiθ )|2 dθ ≤ |K(reiθ )|2 dθ. (19.22)
−π −π −π
If k(0) 6= 0, then we apply Theorem 15.18 (Jensen’s Formula) to obtain

Z π
1
log |k(0)| ≤ log |k(reiθ )| dθ. (19.23)
2π −π
Now the formula
Z π Z π Z π
1 1 1
log |k(reiθ )| dθ = log+ |k(reiθ )| dθ + log− |k(reiθ )| dθ
2π −π 2π −π 2π −π
clearly implies
Z π Z π Z π
1 1 1
log |k(reiθ )| dθ = log+ |k(reiθ )| dθ − log− |k(reiθ )| dθ
2π −π 2π −π 2π −π
Z Z π
1 π + iθ 1
= log |k(re )| dθ − log |k(reiθ )| dθ. (19.24)
π −π 2π −π
Substituting the inequalities (19.22) and (19.23) into the formula (19.24), we obtain
Z π Z
1 1 π
log |k(reiθ )| dθ ≤ |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
1 1 π k(ζ)
iθ
log |k(re )| dθ ≤ |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
1 ∞
| log ϕ(x)| 1 ∞ log |fb(x)| 1 π
2
dx = 2
dx = log |K(eiθ )| dθ < ∞.
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.
19.2 Quasi-analytic Classes and Borel’s Theorem
Problem 19.7
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
It is clear that E ⊂ Ω and Ω is an open set. Now we define F : Ω → C as follows: Given z ∈ Ω.

Then we have z ∈ Vα for some Vα and we define
F (z) = Fα (z). (19.27)

19.2. Quasi-analytic Classes and Borel’s Theorem 261
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.
Furthermore, it is clear that F is holomorphic at every point of Ω. Finally, if ζ ∈ E, then ζ ∈ Vζ .
Since Fζ (z) = f (z) for all z ∈ Vζ ∩ E, we must have
F (ζ) = Fζ (ζ) = f (ζ).
Problem 19.8
Proof. Since n! ≤ nn for every n ≥ 1, we have kD n f k∞ ≤ βf Bfn n! ≤ βf Bfn nn . By Definition

19.6, we have C{n!} ⊆ C{nn }. Recall the approximation to Stirling’s formula [61, Exercise 20,
p. 200] that
nn 1
∼√
en n! 2πn
for large n. Thus there exists a constant M > 0 such that nn ≤ M en n! for all n ≥ 1. If
f ∈ C{nn }, then we have
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 },
Problem 19.9
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 we have
C{n!} ⊆ C{Mn }.
The construction of an example f belonging to C{Mn }, but f ∈ / C{n!} is basically motivated
by [70, Theorem 1, p. 4]. Put mn = MMn+1
n
for every n = 0, 1, 2, . . .. Since
Mn+1 Mn Mn+1 Mn−1 − Mn2

mn − mn−1 = − = ≥0
Mn Mn−1 Mn Mn−1
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
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
for x ∈ R. We first show that f ∈ C{Mn }. For every k ≥ 1, we have
ik Mn
fn(k) (x) = e2mn ix .
(2mn )n−k
Combining this and the estimate (19.28), we get

∞
X ∞
X X Mn Mk ∞ X 1 ∞
(k) Mn
|f (x)| ≤ |fn(k) (x)| = ≤ · = Mk · ≤ 2 · 2k Mk (19.29)
(2mn )n−k 2n−k Mn 2n−k
n=0 n=0 n=0 n=0
for every k ≥ 1 and x ∈ R. Furthermore, we also have

∞
X ∞
X ∞
X 1 M n+1
Mn
|f (x)| ≤ |fn (x)| = n
= · n .
n Mn
(19.30)
n=0 n=0
(2m n ) n=0
2 n+1
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
holds for every k ≥ 0 so that f ∈ C{Mn } as required.

Next, we want to show that |f (k) (0)| ≥ Mk for every k ≥ 0. It is obvious that
∞
X Mn M1
f (0) = = M0 + + · · · ≥ 1 = M0 .
(2mn )n 2m1
n=0
Mn
In addition, if k ≥ 1, then since every term (2mn )n−k
is positive for every n = 0, 1, 2, . . ., we have
∞
X
(k) k Mn
|f (0)| = |i | ≥ 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
|f (k) (0)| ≤ βf Bfk k!
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
∞
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)
By the definition (19.32) again, we see that

Z λ1 Z x+λ1
1 1
g1 (x) = g0 (x − t1 ) dt1 = g0 (t) dt,
2λ1 −λ1 2λ1 x−λ1
so the Fundamental Theorem of Calculus yields that g1 (x) is differentiable in R and
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
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
For every n ≥ 2 and any x, y ∈ R, the bound (19.35) asserts that
|gn (x) − gn (y)| = |x − y| · |gn′ (ξ)| ≤ |x − y| · max |g2′ (x)|

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 [61, Exercise 16, p. 168] that {gn } converges
uniformly to a continuous function g on [−3λ, 3λ], i.e.,
g(x) = lim gn (x)

n→∞
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
is true for all x ∈ [−3λ, 3λ] which means that

1
g′ (x) = g λ2 , λ3 , . . . ; g0 (x + λ1 ) − g λ2 , λ3 , . . . ; g0 (x − λ1 )
2λ1
g0 (x + λ1 ) − g0 (x − λ1 )
= g λ2 , λ3 , . . . ; (19.36)
2λ1
holds in [−3λ, 3λ]. Using similar reasoning as the previous paragraph, it can be shown that g is
also differentiable at the end points ±3λ and the formula (19.36) holds in R. In conclusion, we
have g ∈ C ∞ .
Since |λ1 + λ2 + · · · + λn | < λ, we obtain g0 (−t1 − t2 − · · · − tn ) = 1 for all −λk ≤ tk ≤ λk ,
where 1 ≤ k ≤ n. Thus we note that
Z λ1 Z λ2 Z λn
1
g(0) = lim n dt1 dt2 · · · dtn = 1,
n→∞ 2 λ1 λ2 · · · λn −λ −λ2 −λn
1
so g is not identically zero in R.

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 [38, pp. 79 – 84]. See also [32].
Problem 19.11
Proof. An example of a function ϕ ∈ C ∞ with the required properties can be found in [77,
Problem 10.6, pp. 266, 267]. In fact, we start with
 −1
 e x , if x > 0;
g(x) =

0, if x ≤ 0.
It is known that g ∈ C ∞ and g(m) (0) = 0 for all m = 1, 2, . . .. Define ϕ : R → R by

g(2 − |x|)
ϕ(x) = . (19.38)
g(2 − |x|) + g(|x| − 1)
Now it is easy to see that ϕ ∈ C ∞ . Furthermore, we have ϕ(x) = 1 for −1 ≤ x ≤ 1, ϕ(x) = 0
for |x| ≥ 2 and 0 ≤ ϕ(x) ≤ 1 if x ∈ [−2, −1] ∪ [1, 2] so that
supp ϕ ⊆ [−2, 2].
Problem 19.12
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
where λn is large enough. Fix the non-negative integer n, we notice that

k
X n! k k−m n−m (m)
(D k fn )(x) = βn Cm λn x ϕ (λn x), (19.39)
m=0
(n − m)!
where k = 0, 1, 2, . . . , n − 1. Recall from the definition of ϕ in Problem 19.11 that
supp ϕ(m) ⊆ supp ϕ ⊆ [−2, 2]
/ [−2, 2], then ϕ(m) (λn x) = 0 so that (D k f )(x) = 0. If

holds for every m = 0, 1, . . . , k. If λn x ∈
2 (m)
λn x ∈ [−2, 2], then |x| ≤ λn . Since ϕ is continuous on [−2, 2], there is a positive constant M
(m)
such that |ϕ (λn x)| ≤ M . Thus we obtain
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 [61, 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)

+ (D n fn+1 )(0) + (D n fn+2 )(0) + · · ·
= n!βn
= αn
for every n = 0, 1, 2, . . ., as required. This completes the proof of the problem.f

f
Please also read [41] and [56].
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, [46, Theorem 1.5.4, p. 30] and [51].
Problem 19.13
Proof. It suffices to prove that

|(D n f )(a)| 1
n
lim sup = ∞. (19.42)
n→∞ n!
We follow the suggestion. Let ck = λ1−k

k , where the sequence {λk } satisfies
k−1
X
ck λkk = λk > 2 cj λkj = 2(λk1 + λ2k−1 + · · · + λ2k−1 ) and λk > k2k > 1.
j=1
Fix the non-negative integer n, we have

∞
X n+1
X ∞
X n+1
X ∞
X 1
ck λnk = λkn+1−k + λkn+1−k < λkn+1−k + < ∞.
k2k
k=1 k=1 k=n+2 k=1 k=n+2
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
|(D n fk )(a)| = ck λnk = λkn+1−k
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
Combining this and Stirling’s formula, for large enough n, we get

|(D n f )(a)| 1 n2n 1 en n2n 1 en
n n n
> ≈ √ = 2 2
n! 2n! n
2n 2πn (8π) n · n n
which implies the result (19.42). Hence the power series

∞
X (D n f )(a)
(x − a)n
n!
n=0
has radius of convergence 0 for every a ∈ R, completing the proof of the problem.
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
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
f ′ (α) = lim f ′ (ξn ) = 0.

n→∞
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
Proof. Suppose that

X = f ∈ H(C) |f (z)| ≤ Ceπ|z| for some C > 0, f ∈ L2 [−π, π] .
Recall from Definition 3.6 that the sequence space ℓ2 is given by

n ∞
X o
ℓ2 = {f (n)} |f (n)|2 < ∞ .
n=−∞
Define the map Φ : X → ℓ2 by

Φ(f ) = {f (n)}.
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
π|z|
positive constants C1 and C2 . Therefore, we have |αf (z) + βg(z)| ≤ |α|C1 + |β|C2 e and
kαf + βgk2 ≤ |α| · kf k2 + β · kgk2 < ∞

g
See [79, Problem 5.25, pp. 68, 69]
by Theorem 3.9. This means that αf + βg ∈ X. Besides, the fact

∞
X ∞
X ∞
X
|αf (n) + βg(n)|2 ≤ 2|α|2 |f (n)|2 + 2|β|2 |g(n)|2 < ∞
n=−∞ n=−∞ n=−∞
implies that {αf (n) + βg(n)} ∈ ℓ2 . Thus the relation
Φ(αf + βg) = {αf (n) + βg(n)} = α{f (n)} + β{g(n)} = αΦ(f ) + βΦ(g)
holds, i.e, Φ is linear.

For every f ∈ L2 (T ), we define
Z π
1
F (z) = f (t)e−izt dt.
2π −π
Using the analysis in §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 = × =
2π −π 2π i(n − z) (z − n)π
holds for every n ∈ Z. Recall from [62, 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 , (19.43)
so it makes X Hilbert. Of course, it follows from the expression (19.43) that

 1, if n = m;
hUn , Um iX = hun , um iT =

0, otherwise.
In other words, {Un | n ∈ Z} forms an orthonormal set in X. Since {un | n ∈ Z} is maximal in

L2 (T ), {Un | n ∈ Z} is also maximal in X. By Theorem 4.18 (The Riesz-Fischer Theorem), it is
true that X
F (z) = ck Uk (z),
k∈Z
where c1 , c2 , . . . are some scalars. Since F is entire, we have
X
F (n) = lim ck Uk (z) = cn Un (n) = cn
z→n
k∈Z
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.

Problem 19.16
Proof. Since |f (x)| ≤ e−|x| on R, f ∈ L2 (−∞, ∞). By the analysis in §19.1, its Fourier transform
Z ∞
b
f (z) = f (t)eitz dt
−∞
is holomorphic in Π+ . In particular, we have

Z ∞
b
f (x) = f (t)eitx dt
−∞
for every x ∈ R. For every n ≥ 0, the hypothesis implies that differentiation under the integral
sign is legitimateh so that
Z ∞ Z ∞
b (n)
|(f ) (x)| = n itx
f (t)(it) e dt ≤ e−|t| · |t|n dt = 2n!.
−∞ −∞
Consequently, the power series

∞
X
cn (z − a)n
n=0
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
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
|f (z) − R(z)| < ǫ (20.1)
for all z ∈ K.
Indeed, everything up to [62, 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.
Figure 20.1: The compact set X.
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.
271
272 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 [62, pp. 393, 394], we can obtain
|F (z) − Φ(z)| < 6000ω(z) and |f (z) − Φ(z)| < ω(δ) (20.2)
for all z ∈ Ω, where Ω = S 2 \ (E1 ∪ E2 ∪ · · · ∪ En ) which is an open set containing K. By the

definition, we have
S 2 \ Ω = E1 ∪ E2 ∪ · · · ∪ En .
Since Ej ⊂ Sj for j = 1, 2, . . . , n, one gets the set A = {p1 , p2 , . . . , pn }. Since F ∈ H(Ω) and
K ⊆ Ω, Theorem 13.9 (Rung’s Theorem) implies that there exists a rational function R(z) with
poles only in A such that
|F (z) − R(z)| < ω(δ) (20.3)
for all z ∈ K. Combining this and the inequalities (20.2) and (20.3), we have
|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
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 [27, pp. 344, 345]. In fact, we can also assume
that
X∞
rn2 < 1. (20.4)
n=1
• L is a bounded linear functional on C(K). It is easy to see that L is a linear functional

on C(K). According to [9, Theorem 4.10, p. 49], we have
Z Z
f (z) dz ≤ 2π · kf k∞ and f (z) dz ≤ 2πrn · kf k∞
Γ γn
for every f ∈ C(K) and n ∈ N. Therefore, we get

∞
X
|L(f )| ≤ 2π 1 + rn · kf k∞ < ∞
n=1
which shows that L is bounded.

• L(R) = 0 for every rational function R whose poles are outside K. Let z0 be a
pole of R. Since K = U \ V , we have either z0 lies outside U or z0 ∈ Dn for exactly one n.
If z0 lies outside U , then since Γ([0, 2π]), γn ([0, 2π]) ⊆ U , the integrals in L(R) are both
zero so that L(R) = 0 in this case. If z0 ∈ Dm , then we have z0 ∈ / Dn for every n 6= m.
In this case, we know that Ind Γ (z0 ) = Ind γm (z0 ) = 1 and Ind γn (z0 ) = 0 for every n 6= m.
Consequently, the integrals cancel for the principal part of R at the pole z0 which gives
L(R) = 0 when we express R in its partial fraction decomposition.
a
This shows that Mergelyan’s Theorem does not hold anymore if the finiteness of the components of S 2 \ K
is dropped.
273
• 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
for every n ∈ N. Therefore, we obtain

∞
X
L(f ) = 2πi 1 − rn2 6= 0
n=1
by the hypothesis (20.4).
Problem 20.3
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) = − 2 + ncn ω n−1 .
ω n=1
Since zf (z) → 1 as z → ∞, f has the form

∞
1 X a−n
f (z) = +
z zn
n=0
so that
∞
′ −1 X −na−n
f (z) = 2 + .
z z n+1
n=0
274 Chapter 20. Uniform Approximation by Polynomials
By the definition (20.5), we get

ϕ′ (0) = lim af ′ F (ω) · F ′ (ω) = 1.
ω→0
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) g(z)
n=1 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 n
b= g(z) dz + cn a · gn+1 (z) dz = c0
2πi Γ 2πi Γ
n=1
as desired. This proves the second assertion.

To prove the third assertion, we notice that since F (0) = ∞, we observe that F maps
CR = {ω | |ω| = R} into the disk D(0; r) for some R < 1 and sufficiently close to 1. Therefore,
we obtain Z
1 F (ω) 1 2πRr
|b| = dω < × =r
2πi CR ω 2π R
as desired, completing the proof of the problem.
Index
A Hurwitz’s Theorem, 18, 111, 114

analytic capacity, 49 hyperbolic transformation, 142
annihilate, 241
K
B Koebe mapping, 128
Baire’s Theorem on Semicontinuous Kolmogorov’s Theorem, 225
Functions, 205 Kronecker’s Approximation Theorem, 52,
Bolzano-Weierstrass Theorem, 80, 111, 202, 131
268
Borel transform, 254 L
Borel’s Theorem, 267 lacunary power series, 93
Leibniz’s Rule, 270
C locally compact abelian, 242
Cauchy type integrals, 47 loxodromic transformation, 142
continuum, 195 Lusin Area Integral, 148
converge strongly, 57
M
converge weakly, 57
multipler of the transformation, 139
Convolution Theorem, 243
N
D
normal form, 142
determinant of ϕ, 131
Dirac delta function at x, 57 O
Dirichlet’s Approximation Theorem, 53 Osgood’s Theorem, 96
E P
elliptic transformation, 142 Pólya representation, 254
equivalent under G, 184 parabolic transformation, 142
extreme point, 56 Poisson kernel, 258
Poisson kernel for the upper half-plane, 45
F Principle of Subordination, 108
Fejér’s Theorem, 65
Fundamental Normality Test, 203 R
Fundamental Theorem of Algebra, 237 Removable sets for holomorphic functions,
49
H Riemann-Lebesgue Lemma, 243
Hadamard gaps, 93 right-shift operator, 238
Hadamard’s Three-Line Theorem, 72 Rogosinski’s Theorem, 81
harmonic conjugate, 181, 224
Harnack’s inequalities, 49 S
Hausdorff measure, 49 Schottky’s Theorem, 122
Hilbert’s identity, 245 Schwarz Integral Formula, 41
275
276 Index
Schwarz Reflection Principle for U , 101 total variation of f , 232

semisimple, 247 totally disconnected, 195
Stone-Weierstrass Theorem, 30
strong convergence, 57 V
Swiss cheese set, 272 Vitali Convergence Theorem, 96
W
T weak convergence, 57
The Basic Connectedness Lemma, 42 weak∗ convergence, 57
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.
[4] T. M. Apostol, Mathematical Analysis, 2nd ed., Addison-Wesley Publishing Company, 1974.
[5] T. M. Apostol, Modular Functions and Dirichlet Series in Number Theory, 2nd ed., N. Y.:
Springer-Verlag, 1990.
[6] 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.
[7] S. Axler, P. Bourdon and W. Ramey, Harmonic Function Theory, 2nd ed., N. Y.: Springer,
2001.
[8] F. Bagemihl and W. Seidel, Sequential and continuous limits of meromorphic functions,
Ann. Acad. Sei. Fenn. Ser. Al., Vol. 280, pp. 1 – 17, 1960.
[9] J. Bak and D. J. Newman, Complex Analysis, 3rd ed., N. Y.: Springer, 2010.
[10] E. Bishop and D. Bridges, Constructive Analysis, Berlin: Springer-Verlag, 1985.
[11] R. P. Boas, Entire Functions, N. Y.: Academic Press, 1954.
[12] V. I. Bogachev, Measure Theory Volume I, Berlin; N. Y.: Springer, 2007.
[13] P. Borwein and Tamás Erdélyi, Polynomials and Polynomial Inequalities, N. Y.: Springer-
Verlag, 1995.
[14] 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.
[15] R. B. Burckel, An Introduction to Classical Complex Analysis, Vol. 1, Basel: Birkhäuser

Basel, 1979.
[16] 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.
[17] P. Colwell, Blaschke products: Bounded Analytic Functions, Ann Arbor: University of
Michigan Press, 1985.
277
278 Bibliography
[18] J. B. Conway, Fucntions of One Complex Variable, 2nd ed., N. Y.: Springer-Verlag, 1978.
[19] 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.
[20] M. P. Drazin, The invariant circles of a bilinear transformation, Math. Gaz., Vol. 38, pp.
26 – 29, 1954.
[21] P. L. Duren, Theory of H p Spaces, Boston: Academic Press, 1970.
[22] W. N. Everitt, W. K. Hayman and G. Nasri-Roudsari, On the Representation of Holomor-

phic Functions by Integrals, Appl. Anal., Vol. 65, pp. 95 – 102, 1997.
[23] G. B. Folland, Real Analysis: Modern Techniques and Their Applications, 2nd ed., N. Y.:
Wiley, 1999.
[24] L. R. Ford, Automorphic Functions, Mc-Graw Hill Inc., 1929.
[25] J. B. Fraleigh, A First Course in Abstract Algebra, 7th ed., Boston: Addison-Wesley, 2003.
[26] W. H. J. Fuchs, On the zeros of power series with Hadamard gaps, Nagoya Math. J., Vol.
29, pp. 167 – 174, 1967.
[27] T. W. Gamelin, Complex Analysis, N. Y.: Springer-Verlag, 2001.
[28] G. H. Hardy, A Course of Pure Mathematics, 10th ed., Cambridge: Cambridge University
Press, 2002.
[29] W. K. Hayman and E. F. Lingham, Research Problems in Function Theory: Fiftieth An-
niversary Edition, Cham: Springer International Publishing, 2019.
[30] 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.
[31] K. Hoffman, Banach Spaces of Analytic Fucntions, N. Y.: Dover Publications, 1988.
[32] 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
[33] K. S. K. Iyengar, A note on the zeros of = 0, The Mathematics Student, Vol. 6, pp.
r!
r=0
77, 78, 1938.
[34] S. G. Krantz, Handbook of Complex Variables, Boston, Mass.: Birkhäuser, 1999.
[35] A. N. Kolmogorov and S. V. Fomin, Elements of the Theory of Functions and Functional
Analysis, Mineola, N. Y. Dover Publications, 1999.
[36] B. Ja. Levin, Distribution of Zeros of Entire Functions, Revised d., Providence, R. I.: Amer.
Math. Soc., 1980.
[37] I. H. Lin, Classical Complex Analysis: A Geometric Approach - Vol. 2, Singapore: World
Scientific Publishing Company Co. Pte. Ltd., 2010.
[38] S. Mandelbrojt, Analytic functions and classes of infinitely differentiable functions, Rice
Inst. Pamphlet 29, Houston, Texas, 1942.
Bibliography 279
[39] M. Marden, Geometry of Polynomials, 2nd ed., Mathematical Surveys, No. 3, Providence,
R. I.: Amer. Math. Soc., 1966.
[40] G. V. Milovanović, D. S. Mitrinović and Th. M. Rassias, Topics in polynomials: Extremal

Problems, Inequalities, Zeros, Singapore; Hong Kong: World Scientific Publishing Co.,
1994.
[41] H. Mirkil, Differentiable functions, formal power series, and moments, Proc. Amer. Math.
Soc., Vol. 7, pp. 650 – 652, 1956.
[42] J. R. Munkres, Topology, 2nd ed., Upper Saddle River, N. J.: Prentice-Hall, 2000.
[43] T. Murai, The value distribution of lacunary series and a conjecture of Paley, Ann. Inst.
Fourier (Grenoble), Vol. 31, pp. 136 – 156, 1981.
[44] N. I. Muskhelishvili, Singular Integral Equations: Boundary Problems of Function Theory

and their Applications to Mathematical Physics, 2nd ed., N. Y.: Dover Publications, 1992.
[45] S. B. Nadler, Continuum Theory: An Introduction, N. Y.: Marcel Dekker, 1992.
[46] R. Narasimhan, Analysis on Real and Complex Manifolds, Paris: Masson, Amsterdam:
North-Holland Pub. Co., 1968.
[47] Z. Nehari, Conformal Mapping, N. Y.: Dover Publications, 1972.
[48] W. F. Osgood, Note on the functions defined by infinite series whose terms are analytic
functions, Ann. Math., Vol. 3, pp. 25 – 34, 1901.
[49] A. Ostrowski, Mathematische Miszellen XV. Zur konformen Abbildung einfachzusamm en-
hiingender Gebiete, Jahresbericht Deut. Math., Vereiningung 38, pp. 168 – 182, 1929.
[50] R. C. Paley and N. Wiener, Fourier Transforms in the Complex Domain, N. Y.: Amer.
Math. Soc. Colloquium Publ. 19, 1934.
[51] H.-J. Petzsche, On E. Borel’s Theorem, Math. Ann., Vol. 282, pp. 299 – 313, 1988.
[52] G. Pólya and G. Szegö, Problems and Theorems in Analysis II: Theory of Functions. Zeros.
Polynomials. Determinants. Number Theory. Geometry, Berlin: Springer-Verlag, 1998.
[53] I. I. Privalov, Randeigenschaften Analytischer Funktionen, VEB Deutscher Verlag der Wis-
senschaften, Berlin, 1956.
[54] R. E. Rodrı́guez, I. Kra and J. P. Gilman, Complex Analysis: In the Spirit of Lipman Bers,
2nd ed., N. Y.: Springer-Verlag, 2013.
[55] M. Rosenblum and J. Rovnyak, Topics in Hardy Classes and Univalent Functions, Basel;
Boston: Birkhäuser, 1994.
[56] A. Rosenthal, On functions with infinitely many derivatives, Proc. Amer. Math. Soc., Vol.
4, pp. 600 – 602, 1953.
[57] W. W. Rogosinski, On the order of the derivatives of a function analytic in an angle, J.

Lond. Math. Soc., Vol. 20, pp. 100 – 109, 1945.
[58] 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.
[59] W. Rudin, Function Theory in the Unit Ball of Cn , N. Y.: Springer-Verlag, 1980.
280 Bibliography
[60] W. Rudin, Fourier Analysis on Groups, Mineola, N. Y.: Dover Publications, Inc., 2017.
[61] W. Rudin, Principles of Mathematical Analysis, 3rd ed., Mc-Graw Hill Inc., 1976.
[62] W. Rudin, Real and Complex Analysis, 3rd ed., Mc-Graw Hill Inc., 1987.
[63] D. Sarason, Complex Function Theory, 2nd ed., Providence, R. I.: Amer. Math. Soc., 2007.
[64] J. L. Schiff, Normal Families, N. Y.: Springer-Verlag, 1993.
[65] E. M. Stein and R. Shakarchi, Complex Analysis, Princeton, N. J.: Princeton University
Press, 2003.
[66] E. M. Stein and R. Shakarchi, Fourier Analysis: an Introduction, Princeton, N. J.: Prince-
ton University Press, 2003.
[67] E. M. Stein and R. Shakarchi, Real Analysis: Measure Theorey, Integration and Hilbert
Spaces, Princeton, N. J.: Princeton University Press, 2005.
[68] K. Stephenson, Isometries of the Nevanlinna Class, Indiana Univ. Math. J., Vol. 26, No. 2,
307 – 324, 1977.
[69] R. S. Strichartz, The Way of Analysis, revised ed., Boston, Mass.: Jones and Bartlett,
2000.
[70] V. Thilliez, On quasianalytic local rings, Expo. Math., Vol. 26, pp. 1 – 28, 2008.
[71] S. R. Tims, A theorem on functions schlicht in convex domains, Pro. Lond. Math. Soc., Vol
3, pp. 200 – 205, 1951.
[72] E. C. Titchmarsh, The Theory of Functions, Oxford University Press, N. Y., 1960.
[73] X. Tolsa, Analytic Capacity, the Cauchy Transform, and Non-homogeneous

Calderón–Zygmund Theory, Cham: Springer International Publishing, 2014.
[74] N. X. Uy, Removable sets of analytic functions satisfying a Lipschitz condition, Ark. Mat.,
Vol. 17, No. 1 – 2, pp. 19 – 27, 1979.
[75] M. Weiss and G. Weiss, On the Picard property of lacunary power series, Studia Math.,
Vol. 22, pp. 221 – 245, 1963.
[76] K. W. Yu, A Complete Solution Guide to Complex Analysis, Amazon.com, 2019.
[77] K. W. Yu, A Complete Solution Guide to Principles of Mathematical Analysis, Ama-

zon.com, 2018.
[78] K. W. Yu, A Complete Solution Guide to Real and Complex Analysis I, Amazon.com, 2019.
[79] K. W. Yu, Problems and Solutions for Undergraduate Real Analysis I, Amazon.com, 2018.
[80] K. W. Yu, Problems and Solutions for Undergraduate Real Analysis II, Amazon.com, 2019.
[81] W. Żelazki, Banach Algebras, Warszawa: Elsevier Pub. Comp. and Polish Sc. Pub., 1973.
[82] V. A. Zorich, Mathematical Analysis I, 2nd ed., Berlin: Springer, 2016.
[83] V. A. Zorich, Mathematical Analysis II, 2nd ed., Berlin: Springer, 2016.
[84] A. Zygmund, Trigonometric Series, 3rd ed., Cambridge: Cambridge University Press, 2002.

Kit-Wing Yu - A Complete Solution Guide To Real and Complex Analysis II-978-988-74156-5-7 (2021)

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

Kit-Wing Yu - A Complete Solution Guide To Real and Complex Analysis II-978-988-74156-5-7 (2021)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Kit-Wing Yu - A Complete Solution Guide To Real and Complex Analysis II-978-988-74156-5-7 (2021)

Uploaded by

Copyright:

Available Formats

A Complete Solution Guide to Real

and Complex Analysis II

by Kit-Wing Yu, PhD

ISBN: 978-988-74156-4-0 (eBook)

• A Complete Solution Guide to Complex Analysis

• A Complete Solution Guide to Real and Complex Analysis I

• A Complete Solution Guide to Principles of Mathematical Analysis

• Problems and Solutions for Undergraduate Real Analysis

• Problems and Solutions for Undergraduate Real Analysis I

• Problems and Solutions for Undergraduate Real Analysis II

• Mock Tests for the ACT Mathematics

Then I will post the updated errata on my website

10.1 The closed contour ΓA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

11.1 The I divides ∂∆ into several triangles. . . . . . . . . . . . . . . . . . . . . . . . 48

12.1 The boundary ∂∆. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

13.1 The simply connected set Ω. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

14.1 The region Ω bounded by C1 and C2 . . . . . . . . . . . . . . . . . . . . . . . . . 119

15.1 The distribution of the zeros zk,n of exp(exp(z)). . . . . . . . . . . . . . . . . . . 151

16.1 The paths βζ (I) and γζ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

19.1 The closed contour Γr . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255

20.1 The compact set X. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271

List of Figures viii

10 Elementary Properties of Holomorphic Functions 1

12 The Maximum Modulus Principle 69

13 Approximations by Rational Functions 83

15 Zeros of Holomorphic Functions 149

16 Analytic Continuation 179

18 Elementary Theory of Banach Algebras 235

19 Holomorphic Fourier Transforms 251

20 Uniform Approximation by Polynomials 271

10.1 Basic Properties of Holomorphic Functions

Proof. Let a ∈ C and

γ(t) = a + 2δeit (10.4)

for all t ∈ [0, 2π].

Fn (t) = fn (γ(t)) and F (t) = f (γ(t))

/ γ ∗ , so Ind γ (z) = 1. Then we obtain from Theorem 10.15 (The Cauchy’s

f (z) = lim fn (z)

for all n ≥ N . In other words, {fn } converges uniformly to f on D(a; δ).

|fn (z) − f (z)| < ǫ (10.9)

holds if n ≥ Nk , where k = 1, 2, . . . , m. Set N = max(N1 , N2 , . . . , Nm ). Therefore, if n ≥ N ,

Proof. We divide the proof into several parts.

zeF (z) F ′ (z) − eF (z)

exp(Ω) = D(1; 1). (10.12)

• The number of such region Ω. For each n ∈ Z, we define Ωn = {z + 2nπi | z ∈ Ω}

exp(Ωn ) = exp(Ω + 2nπi) = exp(Ω) = D(1; 1).

Since F is continuous on D(1; 1), Ω must be bounded. Therefore, there exists an N ∈ N

• The derivative of log z. Define log z, for |z − 1| < 1, such that

By differentiating the equation (10.13), we get log′ (z)elog z = 1 and so

• The coefficients an and cn . It is easy to see that

We have completed the analysis of the problem.

10.2 Evaluation of Integrals

Proof. We note that

Figure 10.1: The closed contour ΓA .

Next, if t = −u for some u > 0, then we have

6 1, we have either |α| < 1 or |α−1 | < 1. If z = eiθ , then cos θ = 21 (z + z1 ).

Figure 10.2: The contour ΓA .

Next, we write 2i sin z = eiz − e−iz so that