i i

“book” — 2014/6/4 — 17:28 — page i — #1

i i

Complex Analysis

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 17:28 — page iii — #3

i i

Complex Analysis

R. Roopkumar
Department of Mathematics
Alagappa University
Karaikudi
Tamilnadu

Chennai • Delhi

i i

i i
 i i

“book” — 2014/11/19 — 13:37 — page iv — #4

i i

Copyright © 2015 Pearson India Education Services Pvt. Ltd

Published by Pearson India Education Services Pvt. Ltd, CIN:

U72200TN2005PTC057128, formerly known as TutorVista Global Pvt. Ltd,
licensee of Pearson Education in South Asia.

No part of this eBook may be used or reproduced in any manner whatsoever

without the publisher’s prior written consent.

This eBook may or may not include all assets that were part of the print
version. The publisher reserves the right to remove any material in this
eBook at any time.

ISBN 978-93-325-3761-3
eISBN 978-93-325-4159-7

Head Office: A-8 (A), 7th Floor, Knowledge Boulevard, Sector 62, Noida
201 309, Uttar Pradesh, India.
Registered Office: Module G4, Ground Floor, Elnet Software City, TS-140,
Block 2 & 9, Rajiv Gandhi Salai, Taramani, Chennai 600 113, Tamil Nadu,
India.
Fax: 080-30461003, Phone: 080-30461060
www.pearson.co.in, Email: [email protected]

i i

i i
 i i

“book” — 2014/6/4 — 17:28 — page v — #5

i i

Contents

Preface vii

1. Preliminaries 1
1.1 Introduction and Brief Prerequisites 1
1.2 Complex Numbers and Geometrical Representations 9
1.3 Sequences and Series of Complex Numbers 24
1.4 Some Topological Properties of the Complex Plane 36
1.5 Extended Complex Numbers and Stereographic
Projection 52
1.6 Limit and Continuity 55

2. Analytic Functions 71
2.1 Differentiability 71
2.2 Cauchy–Riemann Equations 79
2.3 Power Series and Abel’s Theorems 94
2.4 Exponential and Trigonometric Functions 102
2.5 Hyperbolic Functions 119

3. Rational Functions and Multivalued Functions 123

3.1 Polynomials and Rational Functions 123
3.2 Linear Fractional Transforms 132
3.3 Branch of a Multivalued Function 148
3.4 Conformal Mapping 156
3.5 Elementary Riemann Surface 161

4. Complex Integration 165

4.1 Line Integral 165
4.2 Winding Number and Cauchy’s Theorems 182
4.3 Cauchy’s Integral Formula 193

i i

i i
 i i

“book” — 2014/6/4 — 17:28 — page vi — #6

i i

vi Contents

4.4 General Version of Cauchy’s Theorem 209

4.5 Local Correspondence Theorem and Its Consequences 212

5. Series Developments and Infinite Products 223

5.1 Taylor Series and Laurent Series 223
5.2 Zeroes, Poles, and Singularities 244
5.3 Partial Fraction of Entire Functions 257
5.4 Infinite Product 264
5.5 Gamma Function and Its Properties 278

6. Residue Calculus 287

6.1 Residue 287
6.2 Cauchy’s Residue Theorem 294
6.3 Argument Principle and Rouche’s Theorem 306
6.4 Evaluation of Real Integrals 314
6.5 Integrals of Multivalued Functions 326

7. Some Interesting Theorems 341

7.1 Mean Value Property of Harmonic Functions 341
7.2 Poisson’s Integral 354
7.3 Schwarz Reflection Principle 363
7.4 Riemann Mapping Theorem 368
7.5 Schwarz–Christoffel Formula 373

8. Elliptic Functions 391

8.1 Basic Concepts 391
8.2 Fundamental Parallelogram 398
8.3 Weierstrass ℘a,b Function 406
8.4 The Functions ζa,b and σa,b 423
8.5 Jacobi’s Elliptic Functions snk , cnk and dnk 432

Bibliography 455
Index 457

i i

i i
 i i

“book” — 2014/6/4 — 17:28 — page vii — #7

i i

Preface

There are many good text books on complex analysis, some of which are
listed in the bibliography. In my opinion, most of the books are written with
the assumption that the reader can understand the intricacies of the proofs
by filling the gaps in the arguments. However, based on my personal experi-
ence of offering several courses on complex analysis to the present generation
of students, I sensed the necessity to write a book on complex analysis by
explaining each and every argument in any proof in a lucid manner so that the
book would be an ideal self study material for the students. The present book
has been written to address this need. Since many concepts in complex analy-
sis are geometrical in nature, more geometrical arguments are given, without
any compromise in rigor. While the detailed proofs presented in the book
may appear to be self-evident to the experts in complex analysis, beginner
students who try to learn the subject with rigor and without any assumptions
will find such treatment helpful. At the same time, this book may also be used
as a hand book by a young teacher who needs explanations for some tedious
theorems in complex analysis.
This text book is intended for both under graduate and post graduate
courses in complex analysis. The first chapter consists of the basic concepts
of complex numbers, operations on the complex numbers, topological prop-
erties and limiting concepts with more details. All the results given in this
chapter are used at least once in the later part of this book. For an under-
graduate course, the chapters 2,3,4,5,6 may be prescribed by omitting big
theorems. For a postgraduate course, chapters 1 to 6 may be prescribed as
a first course on complex analysis. For those who have the desired level of
exposure to complex analysis in an under graduate course, directly chapters
4,5,6,7,8 may be prescribed in the post graduate course. In the last chapter, the
properties of Jacobian elliptic functions are having very long proofs. Indeed,
these proofs have been presented to motivate the young students and urge
them persevere instead of being bewildered by the subtle nuances and com-
plications involved. However, my personal opinion is that it is not advisable
to ask the proofs of those results in any examination.

i i

i i
 i i

“book” — 2014/6/4 — 17:28 — page viii — #8

i i

viii Preface

The content of the book is subdivided into 8 chapters with two objectives.
The first objective is to club similar topics under a chapter title and the second
one is to maintain the size of the chapters as uniform as possible. I have made
sure that each chapter has a sufficient number of results and problems of
varied lengths, so that it becomes suitable for setting questions of all types.
For doing this, I realized that I either had to compromise on at least one of
the above said objectives or I had to use two results from the later chapters in
earlier chapters. I have chosen the latter option. However, I have confirmed
that there is no begging of question by using a result before it is proved.
There are some concepts such as the Jordan curve Theorem, Hardamard’s
Theorem regarding the estimate of genus of an infinite product, introduction
to univalent and star-like functions that are not discussed here, since these
are to be taken up at an advanced level beyond the scope of this book. I have
taken stringent efforts to make this book error-free. Nonetheless, if any error
is found, please write to me so that they may be eliminated in future editions.
My e-mail address is roopkumar [email protected]
I am grateful to Dr R. Vembu for his constant encouragement, fruitful dis-
cussions and valuable suggestions while preparing the content for this book.
My thanks are also due to the editors at Pearson Education who worked on
this book, for their professionalism and diligence.

R. Roopkumar

i i

i i
 i i

“book” — 2014/6/4 — 17:28 — page ix — #9

i i

Complex Analysis

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 20:35 — page 1 — #1

i i

1
Preliminaries

1.1 INTRODUCTION AND BRIEF PREREQUISITES

Every mathematical structure has some good properties as well as some
deficiencies. To overcome a particular deficiency on a mathematical structure,
it is, therefore, customary to generalize the mathematical structure. How-
ever, it should be noted that any generalized mathematical structure having a
good property that is not in the earlier structure will not satisfy another good
property of an earlier structure. Consider the following examples:
1. Z is an ordered abelian group, whereas N is an ordered commutative
semi-group but not a group. However, every non-empty subset of N has
a minimum in N, which is not true in Z.
2. Q is an ordered field, whereas Z is an ordered integral domain but not
a field. However, for every element of Z, we can find previous element
and next element in Z, which is not true in Q.
3. R is an ordered field with least upper bound property, whereas Q is not
having least upper bound property. However, Q is countable, whereas
R is not countable.
The set R of all real numbers is a very good setup in which we have lot of
mathematical structures such as Archimedian field with least upper bound
property, complete metric space having Heine–Borel property and Banach
space (a complete normed linear space), and so on. In fact, it is identi-
fied with the set of all points on a straight line. However, in the algebraic
point of view, not all polynomials of degree n (for some n ∈ N) over R
have n roots in R. To be specific, x2 + 1 is a polynomial of degree 2 over
R, which has no solution in R. This is the motivation for introducing the
complex number system in which every polynomial of degree n over C has
exactly n roots in C. In fact, C is the splitting field of x2 + 1, considering

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 2 — #2

i i

2 Introduction and Brief Prerequisites

x2 + 1 as a polynomial over R. As in the earlier generalization of number

systems (just listed above), R is an ordered field, whereas there is no order
relation on C, which makes it as an ordered field. However, complex anal-
ysis has more good and different properties that cannot be expected in real
analysis.
As R and C are metric spaces, we can say all the topological proper-
ties that are common in a metric space such as the properties of open sets,
closed sets, limit point of a set, convergent sequences and series, limit of a
function, and continuous functions etc are similar in R and C. However, it is
interesting to note some of the good differences between real and complex
analysis.

1. If x → x0 in R, then x can approach only in two directions along the

real axis. However, if z → z0 in C, then z gets closer to z0 through
uncountable number of paths. See the following diagrams for some
examples.

x x0

z0
x0 x

z
z
z
z0
z0

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 3 — #3

i i

Preliminaries 3

2. We know that interior of A is connected for every connected subset A

of R. However, there is a connected subset A of C whose interior is not
connected.
3. If f = (u, v) : C → C is differentiable, then for the given u, we can
find v and for the given v, we can find u (uniquely up to a constant).
However, if f = ( f1 , f2 ) : R → R × R is differentiable, then neither f2
can be found from f1 nor f1 can be found from f2 .
4. If f :  → C is (once) differentiable (where  is an open connected
subset of C), then f is infinitely many times differentiable. However, if
f : I → R (where I is an interval) is differentiable, then f  need not be
even continuous.
5. If f : C → C is a bounded differentiable function, then f is a con-
stant function. However, it is not true for a bounded differentiable
real-valued function on R.
6. If f is a complex-valued differentiable function on  ⊆ C such that
f = 0 on a set having a limit point must be identically zero. However,
it is not true for a differentiable real-valued function on an interval of
R.
7. Every non-constant differentiable function f :  → C is an open map.
However, this is not true for a differentiable real-valued function on an
interval of R.
8. Although the above points are some positive properties of complex
analysis that are not in real analysis, graph of a complex-valued
function of a complex variable cannot be visualized, whereas the graph
of a real-valued function of a real variable can be plotted. However, the
geometric properties of a complex-valued function of a complex vari-
able can be studied by seeing the image of a curve or a region under
the given function.
To discuss any branch of Mathematics, it is necessary to know the required
basic definitions and results. Therefore, to make this this book self-contained,
we carefully selected the most useful definitions and results from set theory,
real analysis, and algebra, and presented them in this section.

1. We call a collection of ‘well defined’ objects a set.

2. A set containing no elements is called an empty set. To denote this
special set, we use the notation ∅.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 4 — #4

i i

4 Introduction and Brief Prerequisites

3. To say that x is an element of a set, we write x ∈ A.

4. We say that B is a subset of A if x ∈ A whenever x ∈ B. We write this

symbolically as B ⊆ A.

5. We say that A and B are equal if A ⊆ B and B ⊆ A.

6. Let X be a set, A, B and Aα ⊆ X , ∀α ∈ I for some index set I , then

(a) the union ∪ Aα of {Aα }α∈I is defined by {x : x ∈ Aα , for some

α∈I
α ∈ I}.
(b) the intersection ∩ Aα of {Aα }α∈I is defined by {x : x ∈ Aα ,
α∈I
∀α ∈ I}.
When I = {1, 2, 3, . . . , n}, we also use the following equivalent
notations:
n
∪ Aα = ∪ Aα = A1 ∪ A2 ∪ A3 ∪ · · · ∪ An
α∈I α=1

and
n
∩ Aα = ∩ Aα = A1 ∩ A2 ∩ A3 ∩ · · · ∩ An ,
α∈I α=1

(c) The complement Ac = X \ A of A in X is defined by the set {x ∈

X : x  A}.
(d) A and B are said to be disjoint if A ∩ B = ∅. We also note that
A ∩ B = ∅ iff A ⊆ Bc iff B ⊆ Ac .
(e) De-Morgan’s
 c laws  c
(i) ∪ Aα = ∩ Aα and (ii) ∩ Aα = ∪ Acα .
c
α∈I α∈I α∈I α∈I

7. Let {Xi }ni=1 be a finite collection of sets, then the Cartesian product
n
X1 × X2 × · · · × Xn (or)  Xi
i=1

consists of all elements of the form (x1 , x2 , x3 , . . . , xn ) called ordered

n-tuples, where xi ∈ Xi , ∀i = 1, 2, 3, . . . , n. We say that two elements
n
(x1 , x2 , x3 , . . . , xn ) and (y1 , y2 , y3 , . . . , yn ) are said to be equal in  Xi
i=1
n
if xi = yi and ∀i = 1, 2, 3, . . . , n. The members of  Xi are called
i=1
ordered pairs when n = 2 and ordered triplets when n = 3.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 5 — #5

i i

Preliminaries 5

One can say that mathematical analysis cannot be discussed without

inequalities. Therefore, now we recall the definitions of order relation,
lower bound, upper bound, infimum, and supremum and state some useful
inequalities.

1. A relation R on a non-empty set X is a subset of X × X .

2. A relation R on X is said to be an order relation if

(a) (x, y), (y, z) ∈ R ⇒ (x, z) ∈ R;

(b) for every x, y ∈ X , one of the following is true:
(i) x = y (ii) (x, y) ∈ R (iii) (y, x) ∈ R

If R is an ordered relation on X , then it is customary to write (x, y) ∈ R

by x < y and (X , <) is called an ordered set. In an ordered set (X , <),
the notations x ≤ y and a ≥ b mean that (x < y or x = y) and (b < a
or a = b), respectively.

3. Let (X , <) be an ordered set and ∅  A ⊆ X , then

(a) an element α ∈ X is said to be an upper bound of A, if

x ≤ α, ∀x ∈ A.

(b) A is said to be bounded above in X if A has an upper bound α ∈ X .

(c) least upper bound of A is an upper bound β of A with the property
that
β ≤ γ , for every upper bound γ of A.
least upper bound of A is also called supremum of A and is denoted
by sup A in X or simply by sup A.
(d) an element α ∈ X is said to be a lower bound of A, if

α ≤ x, ∀x ∈ A.

(e) A is said to be bounded below in X if A has a lower bound α ∈ X .

(f) greatest lower bound of A is a lower bound β of A with the property
that
γ ≤ β, for every lower bound γ of A.
greatest lower bound of A is also called infimum of A and is denoted
by inf A in X or simply by inf A.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 6 — #6

i i

6 Introduction and Brief Prerequisites

(g) • sup A is unique if it exists.

• inf A is unique if it exists.
• If α = sup A and α ∈ A, then α is called the maximum of A.
• If α = inf A and α ∈ A, then α is called the minimum of A.
(h) X is said to have least upper bound property, if for every bounded
above subset A of X , sup A exists in X .
(i) X is said to have greatest lower bound property, if for every
bounded below subset A of X , inf A exists in X .
(j) X has greatest lower bound property iff it has least upper bound
property.
(k) R is an ordered set with least upper bound property and Archime-
dian property1 .
(l) For given x, y ∈ R such that x < y, there exists p ∈ Q such that
x < p < y.
(m) For a given x > 0 there exists unique y > 0 such that yn = x. This
√ 1
y is called the nth root of x and is denoted by n x or x n .

Although the content of this book includes some topological properties of

subsets of the complex plane, mainly we concentrate on functions of a com-
plex variable. Therefore, now we recall some basic ideas from function
theory.

1. Let A and B be non-empty sets. We say that f is a function from A into

B (denoted by f : A → B) if f is a subset of A × B with the following
properties:

(a) for every a ∈ A, there exists b ∈ B such that (a, b) ∈ f .

(b) if (a, b1 ), (a, b2 ) ∈ f , then b1 = b2 .

2. Let f : A → B be a function.

(a) Then A and B are called the domain and codomain of f , respec-
tively.
(b) (a, b) ∈ f is equivalently denoted by f (a) = b.
(c) If f : A → A is defined by f (x) = x, ∀x ∈ A, then f is called the
identity function on A.
1 Archimedian property: Given x > 0 and y ∈ R, there exists n ∈ N such that nx > y.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 7 — #7

i i

Preliminaries 7

(d) If A1 ⊆ A, then the image f (A1 ) of A1 under f is defined by f (A1 ) =

{f (a) : a ∈ A1 }.
(e) If B1 ⊆ B, then the pre-image or inverse image f −1 (B1 ) of B1 is
defined by f −1 (B1 ) = {a ∈ A : f (a) ∈ B1 }.
(f) If g : B → C , then we define the composition g ◦ f : A → C of f
and g by (g ◦ f )(x) = g( f (x)), ∀x ∈ A.
(g) f is invertible if there exists a function g : B → A such that g ◦ f
is the identity function on A and f ◦ g is the identity function on B.
Furthermore, g is called the inverse of f and is denoted by f −1 .
3. A function f : A → B is said to be
(a) one-to-one (or injective) if f (x)  f (y) whenever x, y ∈ A with
x  y.
(b) onto (or surjective) if for every b ∈ B, there exists a ∈ A such that
f (a) = b.
(c) bijective if it is injective and surjective.
4. Let f : A → B. Then, f is invertible iff f is a bijection.
Furthermore, for a bijective function f : A → B, the inverse f −1 :
B → A of f is defined by f −1 (y) = x, where x ∈ A is unique such that
f (x) = y.
5. Let A, B ⊆ R. A function f : A → B is said to be
(a) an increasing function if x, y ∈ A such that x < y, then f (x) ≤ f (y).
(b) a strictly increasing function if x, y ∈ A such that x < y, then
f (x) < f (y).
(c) a decreasing function if x, y ∈ A such that x < y, then f (x) ≥ f (y).
(d) a strictly decreasing function if x, y ∈ A such that x < y, then
f (x) > f (y).
As algebra is inevitable to discuss analysis, we briefly recall some concepts
like group, ring, field, and so on.
Let A be a non-empty set.
1. Binary operator. A function f : A × A → A is called a binary operator
on A. We also denote a binary operator f by  and the value f (a, b) by
a  b.
2. Unary operator. A function f : A → A is called a unary operator on A.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 8 — #8

i i

8 Introduction and Brief Prerequisites

3. Let A be a non-empty set and  be a binary operator on A, then

(a)  is said to be associative if a  (b  c) = (a  b)  c, ∀a, b, c ∈ A.
(b)  is said to have an identity element in A if there exists e ∈ A such
that a  e = e  a = a, ∀a ∈ A.
(c) an element a of A is said to have an inverse b ∈ A with respect to 
if a  b = b  a = e . We know that inverse of a is unique.
(d)  is said to be commutative if a  b = b  a, ∀a, b ∈ A.
4. (A, ) is said to be an abelian group if  satisfies the above four
properties (a), (b), (c) for every element of A, and (d).
5. (A, +, ·) is said to be a field if (A, +) and (A \ {0}, ·) are abelian groups,
where 0 is the additive identity and (a + b) · c = (a · c) + (b · c), ∀
a, b, c ∈ A.
6. A four tuple (A, +, ·, <) is said to be an ordered field if (A, <) is an
ordered set and (A, +, ·) is a field with the following properties:
(a) If a, b ∈ A such that a < b, then a + c < b + c, ∀c ∈ A.
(b) If a, b ∈ A such that a > 0 and b > 0, then a · b > 0.
7. If (A, +, ·, <) is an ordered field, then
(a) a · a > 0, ∀a ∈ A \ {0}.
(b) a > 0 iff −a < 0, where −a is the additive inverse of a.
(c) a > 0 iff a−1 > 0, where a−1 is the multiplicative inverse of a.
(d) if a < b and c > 0, then ac < bc.
(e) if a < b and c < 0, then ac > bc.
(f) if 0 < a < b, then b−1 < a−1 .
8. The set R of all real numbers is an ordered field with respect to usual
+, ·, and <.

x if x ≥ 0
(a) The modulus function on R is defined by |x| =
−x if x < 0,
∀x ∈ R.
(b) The modulus function on R satisfies the following properties:
(i) |x| ≥ 0, ∀x ∈ R, (ii) |x| = 0 iff x = 0, (iii) |xy| = |x| |y|,
(iv) |x + y| ≤ |x| + |y|, and (v) x2 ≤ y2 iff |x| < |y|.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 9 — #9

i i

Preliminaries 9

1.2 COMPLEX NUMBERS AND GEOMETRICAL

REPRESENTATIONS
The terminology Complex numbers was due to C. F. Gauss in 1831. The rigor-
ous definition of complex number as an ordered pair of real numbers (a1 , a2 )
was introduced by W. R. Hamilton in 1837. The set of all complex numbers
is denoted by C. Then immediately it follows that C = R × R. In the set of
all complex numbers, there is a special element (0, 1), which is denoted by i.
Euler named this number i by the imaginary number. Note that this number i
was used as a number satisfying i2 = −1 before the rigorous construction of
complex numbers, and hence, it was called an imaginary number.
Since there is no element in R whose square is a negative number and
i2 = −1, this number is not in R and hence, it might be called an imaginary
number. As the numbers on a straight line are not imaginary, the name set of
real numbers is given to R.
We define addition and multiplication on C as follows:
(a1 , a2 ) + (b1 , b2 ) = (a1 + b1 , a2 + b2 ),
(a1 , a2 ) · (b1 , b2 ) = (a1 b1 − a2 b2 , a1 b2 + a2 b1 ), and
(a1 , a2 ), (b1 , b2 ) ∈ C.

THEOREM 1.2.1 The set C of all complex numbers is a field with respect to
addition and multiplication as defined above.

Proof: Let a = (a1 , a2 ), b = (b1 , b2 ), and c = (c1 , c2 ) ∈ C. Using the fact

that R is a field, we get the following:
1. Clearly, a + b = (a1 , a2 ) + (b1 , b2 )
= (a1 + b1 , a2 + b2 )
= (b1 + a1 , b2 + a2 )
= (b1 , b2 ) + (a1 , a2 )
= b+a
2. a + (b + c) = (a1 , a2 ) + [(b1 , b2 ) + (c1 , c2 )]
= (a1 , a2 ) + (b1 + c1 , b2 + c2 )
= (a1 + (b1 + c1 ), a2 + (b2 + c2 ))
= ((a1 + b1 ) + c1 , (a2 + b2 ) + c2 )
= (a1 + b1 , a2 + b2 ) + (c1 , c2 )
= [(a1 , a2 ) + (b1 , b2 )] + (c1 , c2 )
= (a + b) + c.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 10 — #10

i i

10 Complex Numbers and Geometrical Representations

3. (0, 0) ∈ C is the identity element with respect to + because (a1 , a2 ) +

(0, 0) = (a1 + 0, a2 + 0) = (a1 , a2 ) and (0, 0) + (a1 , a2 ) = (0 + a1 , 0 +
a2 ) = (a1 , a2 ).

4. If (a1 , a2 ) ∈ C, then (−a1 , −a2 ) ∈ C and (a1 , a2 ) + (−a1 , −a2 ) =

(a1 + (−a1 ), a2 + (−a2 )) = (0, 0).

5. (a1 , a2 ) · (b1 , b2 ) = (a1 b1 − a2 b2 , a1 b2 + a2 b1 )

= (b1 a1 − b2 a2 , b2 a1 + b1 a2 )
= (b1 a1 − b2 a2 , b1 a2 + b2 a1 )
= (b1 , b2 ) · (a1 , a2 ).

6. a · (b · c)
= (a1 , a2 ) · [(b1 , b2 ) · (c1 , c2 )]
= (a1 , a2 ) · [(b1 c1 − b2 c2 , b1 c2 + b2 c1 )]
= (a1 (b1 c1 − b2 c2 ) − a2 (b1 c2 + b2 c1 ), a1 (b1 c2 + b2 c1 )
+ a2 (b1 c1 − b2 c2 ))
= (a1 b1 c1 − a1 b2 c2 − a2 b1 c2 − a2 b2 c1 , a1 b1 c2 + a1 b2 c1
+ a2 b1 c1 − a2 b2 c2 )
= ((a1 b1 − a2 b2 )c1 − (a1 b2 + a2 b1 )c2 , (a1 b1 − a2 b2 )c2
+ (a1 b2 + a2 b1 )c1 )
= ((a1 b1 − a2 b2 ), (a1 b2 + a2 b1 )) · (c1 , c2 )
= ((a1 , b1 ) · (a2 , b2 )) · (c1 , c2 )
= (a · b) · c.

7. Clearly, (1, 0) ∈ C and (a1 , a2 ) · (1, 0) = (a1 · 1 − a2 · 0, a1 · 0 + a2 · 1) =

(a1 , a2 ).

8. If (a1 , a2 ) ∈ C  (0, 0), then at least

 one of a1 and  a2 is non-zero.
a1 −a2
Therefore, a21 + a22  0, and hence, , 2 ∈ C. Now,
a1 + a2 a1 + a22
2 2
   
a1 −a2 a21 + a22 −a1 a2 + a2 a1
(a1 , a2 ) · , = ,
a21 + a22 a21 + a22 a21 + a22 a21 + a22
= (1, 0).

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 11 — #11

i i

Preliminaries 11

9. a · (b + c)
= (a1 , a2 ) · ((b1 , b2 ) + (c1 , c2 ))
= (a1 , a2 ) · (b1 + c1 , b2 + c2 )
= ((a1 (b1 + c1 ) − a2 (b2 + c2 )), a1 (b2 + c2 ) + a2 (b1 + c1 ))
= (a1 b1 + a1 c1 − a2 b2 − a2 c2 , a1 b2 + a1 c2 + a2 b1 + a2 c1 )
= ((a1 b1 − a2 b2 ) + (a1 c1 − a2 c2 ), (a1 b2 + a2 b1 )+(a1 c2 +a2 c1 ))
= (a1 b1 − a2 b2 , a1 b2 + a2 b1 ) + (a1 c1 − a2 c2 , a1 c2 + a2 c1 )
= ((a1 , a2 ) · (b1 , b2 )) + ((a1 , a2 ) · (c1 , c2 ))
= (a · b) + (a · c).

Hence, (C, +, ·) is a field. 

We identify every real number x with the element (x, 0) of C. By this
identification, the set of all real numbers is regarded as a subset of C. It is
also customary to write a complex number z = (x, y) as x + iy, which is the
simplified form of the expression (x, 0)+(0, 1)·(y, 0). We call x, y the real and
imaginary parts of z and denote them by x = Re z and y = Im z, respectively.

LEMMA 1.2.2 i · i = (−1, 0).

Proof: By direct computation, we get i · i = (0, 1) · (0, 1) = (0 − 1, 0 + 0) =

(−1, 0). 

Remark 1.2.3: C is not an ordered field with respect to any order relation.

Reason: By the observation 7(a) in page 8, if C were an ordered field, then

we would get
−1 = i · i > 0 and 1 = 1 · 1 > 0,

which contradicts the observation 7(b) in page 8.

Example 1.2.4 Write the following numbers in a + ib form.

1. (3 + i5)(4 − i2).

(3 + i)2
2. .
1 − i2

3. sin(1 + i).

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 12 — #12

i i

12 Complex Numbers and Geometrical Representations

Solution:
1. (3 + i5)(4 − i2) = (12 + 10) + i(−6 + 20) = 22 + i14.
(3 + i)2 (9 − 1 + i6)(1 + i2) (8 − 12) + i(16 + 6)
2. = = =
1 − i2 5 5
−4 + i22
.
5
3. sin(1 + i) = sin(1) cos(i) + cos(1) sin(i) = sin(1) cosh(1) +
i cos(1) sinh(1).

Exercise 1.2.5 Write the following numbers in a + ib form.

(2 + i)(3 − i2)
1. .
1 + i4
2. (4 + i3)4 .
3. cos(1 − i2).
4. exp(i(1 + i)).
5. cos(i) exp(i3 + 2).

Answers:
4 − i33
1. .
17
2. −527 − i336.
3. cos(1) cosh(2) + i sin(1) sinh(2).
4. exp(−1) cos(1) + i exp(−1) sin(1).
5. cosh(1) exp(2) cos(3) + i cosh(1) exp(2) sin(3).

Definition 1.2.6 We define the complex conjugate z of z ∈ C by z = (x, −y)

whenever z = (x, y).

RESULT 1.2.7 (Properties of complex conjugate)

Let z, z1 , z2 ∈ C be arbitrary.
1. z + z = 2Re z and z − z = i2Im z.
2. z = z.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 13 — #13

i i

Preliminaries 13

3. z1 + z2 = z1 + z2 .
4. z1 · z2 = z1 · z2 .
 
z1 z1
5. = , when z2  (0, 0).
z2 z2

6. If z = x + iy then z · z = x2 + y2 .
7. z = z if and only if z is a real number.

Proof: Let z = x + iy, z1 = x1 + iy1 , and z2 = x2 + iy2 .

1. z +z = (x+iy)+(x−iy) = 2x = 2Re z and z −z = (x+iy) −(x−iy) =
i2y = i2 Im z.

2. z = x + iy = x − iy = x + iy = z.
3. (z1 + z2 ) = (x1 + iy1 ) + (x2 + iy2 )
= (x1 + x2 ) + i(y1 + y2 )
= (x1 + x2 ) − i(y1 + y2 )
= (x1 − iy1 ) + (x2 − iy2 )
= z1 + z2 .

4. z1 · z2 = (x1 + iy1 ) · (x2 + iy2 )

= (x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 )
= (x1 x2 − y1 y2 ) − i(x1 y2 + x2 y1 )
= (x1 x2 − (−y1 )(−y2 )) + i(x1 (−y2 ) + x2 (−y1 ))
= (x1 − iy1 ) · (x2 − iy2 )
= z1 · z2 .
 
1 1
5. First, we prove that = , for z  (0, 0).
z z
   
1 1
=
z x + iy
 
x − iy
=
(x − iy)(x + iy)
 
x y
= − i
x2 + y2 x2 + y2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 14 — #14

i i

14 Complex Numbers and Geometrical Representations

x y
= +i 2
x2 + y2 x + y2
x + iy
=
x2 + y2
x + iy
=
(x + iy)(x − iy)
1
=
x − iy
1
= .
z
Then, applying (4), we get
     
z1 1 1 1 z1
= z1 · = z1 · = z1 · = .
z2 z2 z2 z2 z2

6. By a direct computation, we get z · z = (x + iy) · (x − iy) = x2 + y2 .

7. z = z iff x + iy = x − iy iff y = −y iff y = 0 iff z is a real number. 
We also have one more unary operator called the modulus or absolute value
of a complex number, which is defined by
√ 
|z| = z · z or equivalently |(x, y)| = x2 + y2

here, we take the positive square root. This unary operator satisfies the
following properties:

RESULT 1.2.8 Let z, z1 , z2 ∈ C be arbitrary.

1. |Re z| ≤ |z| and |Im z| ≤ |z|.
2. |z| = |z|.
3. |z1 · z2 | = |z1 | · |z2 |.
 
 z1  |z1 |
4.   = if z2  0.
z2 |z2 |

Proof: Let z = x + iy, z1 = x1 + iy1 , and z2 = x2 + iy2 .

1. x2 ≤ x2 + y2 implies that |x| ≤ x2 + y2 = |z|. As x2 + y2 is
symmetric in x and y, by interchanging x and y, we get |y| ≤ |z|.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 15 — #15

i i

Preliminaries 15

2. |z| = x2 + y2 = x2 + (−y)2 = |z|.

3. |z1 · z2 | = |(x1 x2 − y1 y2 ) + i(x1 y2 + x2 y1 )| .
= (x1 x2 − y1 y2 )2 + (x1 y2 + x2 y1 )2
= x2 x2 + y21 y22 + x21 y22 + x22 y21
 1 2
= (x21 + y21 )(x22 + y22 )
= |z1 | · |z2 |.
4. As    
 1   x − iy 
= x +y =
2 2 1 1
 = = ,
 z   x2 + y2  x2 + y2 |z|
x2 + y2
we get      
 z1     
  = z1 · 1  = |z1 |  1  = |z1 | . 
z   z   z  |z |
2 2 2 2

RESULT 1.2.9 (Triangle inequality)

If z1 , z2 ∈ C, then |z1 + z2 | ≤ |z1 | + |z2 |. Furthermore, |z1 + z2 | = |z1 | + |z2 |
iff z1 · z2 is real and is non-negative.
Proof: Using Results 1.2.7 and 1.2.8, we get
|z1 + z2 |2 = (z1 + z2 ) · (z1 + z2 )
= (z1 + z2 ) · (z1 + z2 )
= (z1 · z1 + z1 · z2 + z2 · z1 + z2 · z2
= |z1 |2 + |z2 |2 + z1 · z2 + z1 · z2
= |z1 |2 + |z2 |2 + 2 Re (z1 · z2 )
≤ |z1 |2 + |z2 |2 + 2|z1 · z2 |
= |z1 |2 + |z2 |2 + 2|z1 | · |z2 |
= (|z1 | + |z2 |)2 .
Hence, we get |z1 + z2 | ≤ |z1 | + |z2 |.
Next, we find the necessary and sufficient condition to get equality in the
above inequality.
From the proof of the triangle inequality, we get
|z1 + z2 | = |z1 | + |z2 | ⇔ |z1 |2 + |z2 |2 + 2Re (z1 · z2 )
= |z1 |2 + |z2 |2 + 2|z1 · z2 |
⇔ Re (z1 · z2 ) = |z1 · z2 |
⇔ z1 · z2 is real and non-negative.
Hence, the result follows. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 16 — #16

i i

16 Complex Numbers and Geometrical Representations

As a corollary of the above result, we have another version of triangle

inequality, which follows.

COROLLARY 1.2.10 If z1 , z2 ∈ C, then | |z1 | − |z2 | | ≤ |z1 − z2 |.

Proof: Now, |z1 | = |z1 − z2 + z2 | ≤ |z1 − z2 | + |z2 |, which implies that

|z1 | − |z2 | ≤ |z1 − z2 |.

Interchanging z1 , z2 and using | − z| = | − 1||z| = |z|, we get

|z2 | − |z1 | ≤ |z2 − z1 | = |z1 − z2 |.

Hence, by the definition of the modulus function on R, the corollary

follows. 
 
 n 
  n
RESULT 1.2.11 If zj ∈ C, 1 ≤ j ≤ n, then  zj  ≤ |zj |.
j=1  j=1

Proof: We prove this by induction on n. For n = 1, this result holds obviously.

Assume that this result is true for some n. Now, using triangle inequality
(Result 1.2.9) and by induction hypothesis, we get
     
 n+1   n   n 
      n n+1
 z  =  z + z  ≤  z  + |z | ≤ |z | + |z | = |zj |.
 j  j n+1   j n+1 j n+1
 j=1   j=1   j=1  j=1 j=1

This completes the proof of this result. 

 
 n 
  n
Exercise 1.2.12 Prove that  zj  = |zj | iff zj zk is real and non-negative,
j=1  j=1
∀j, k ∈ {1, 2, . . . , n}.

RESULT 1.2.13 (Cauchy–Schwarz inequality)

If xi , yi ∈ C, ∀i = 1, 2, 3, . . . , n, then
 2
 n  n n
 
 xk yk  ≤ |xk |2 · |yk |2 . (1.1)
 
k=1 k=1 k=1

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 17 — #17

i i

Preliminaries 17

n
 n
1  n
1
2 2
Proof: Let c = xk yk , x = |xk |2 and y = |yk |2 .
k=1 k=1 k=1
Consider
n  2
 
0 ≤ (y2 xk − cyk )
k=1
n
= (y2 xk − cyk )(y2 xk − c yk )
k=1
n
= (y4 |xk |2 − y2 cxk yk − y2 cyk xk + ccyk yk )
k=1
= y4 · x2 − y2 · |c|2 − y2 · |c|2 + y2 · |c|2
= y2 (x2 y2 − |c|2 )

If y2 = 0, then equality holds in (1.1); otherwise, we get |c|2 ≤ x2 y2 .
This implies (1.1). 

RESULT 1.2.14 (Schwarz’s inequaltiy)

If x = (x1 , x2 , . . . , xn ), y = (y1 , y2 , . . . , yn ) ∈ Cn , then x + y ≤ x + y,
 n 1
2
where z = |zk |2 , ∀z = (z1 , z2 , . . . , zn ) ∈ Cn .
k=1

Proof:
n
x + y2 = |xk + yk |2
k=1
n
= (xk + yk )(xk + yk )
k=1
n
= (xk xk + xk yk + yk xk + yk yk )
k=1
n n
= x2 + y2 + xk yk + yk xk
k=1 k=1
n n
(note that xk yk + yk xk is real)
k=1 k=1

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 18 — #18

i i

18 Complex Numbers and Geometrical Representations

 n 
 n 
 
≤ x + y + 
2 2
xk yk + yk xk 
 
k=1 k=1
 n   n 
   
   
≤ x2 + y2 +  xk yk  +  yk xk 
   
k=1 k=1
≤ x + y + xy + xy
2 2

(by Cauchy–Schwarz inequality)

= (x + y)2 .
Thus, the result follows. 
Geometrically, the set of all complex numbers can be viewed as the
set of all points in a plane. We fix a point arbitrarily and name it the
origin and denote it by (0, 0). We draw two lines on the plane that pass
through (0, 0) horizontally and vertically. We call the horizontal line and
vertical line, respectively, the real axis and the imaginary axis. Now, every
element (x, y) ∈ C is represented by the point of intersection of the
horizontal line passing through (0, y) and the vertical line passing through
(x, 0). This representation is also called the Cartesian form of a complex
number.
Every non-zero complex number z = (x, y) can also be represented in
another form r exp(iθ) (or reiθ ) called polar form2 , where r = |z| is the length
of the line segment joining (0, 0) and z (called the modulus of z), and θ is
the angle between the positive side of the x-axis and the line segment joining
(0, 0) and z, measured in the anti-clock wise sense (called an argument θ of z).
Using the Pythagoras theorem, one can get the length of the line segment as
x2 + y2 = |(x, y)| = |z|.
Note that the modulus of a complex number is unique but the argument
is not. The reason for the second statement is that if θ is one of the values
of argument of (a, b), then for every k ∈ Z, θ + 2kπ is also an argument
of the same (a, b). Hence, we define the principal argument of a non-zero
complex number (a, b) as the argument θ of (a, b) satisfying the condition
−π < θ ≤ π .
The subsets {(a, b) : a > 0, b > 0}, {(a, b) : a < 0, b > 0}, {(a, b) :
a < 0, b < 0} and {(a, b) : a > 0, b < 0} are called first quadrant, second
quadrant, third quadrant, and fourth quadrant, respectively.

2 At present, treat r exp(iθ ) as just a symbol. The reason for using this notation is explained
in Corollary 2.4.25.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 19 — #19

i i

Preliminaries 19

 At this juncture, the reader is advised not to use arg (a +

Remark 1.2.15:
b
ib) = arctan , for all non-zero a + ib ∈ C.
a

To justify this remark, if we use the above formula, then we will get arg (1 + i)
π
= arg (−1 − i) = , but the first point is in the first quadrant and the second
4
point is in the third quadrant, and hence obviously, their arguments should be
different. See the following diagram.
1+i

arg(−1 − i) arg(1 + i)

−1 − i

Definition 1.2.16 (Principal argument)3 Let a + ib  0.

π −π
• Case (i): If a = 0, b > 0, then θ = ; If a = 0, b < 0, then θ = ;
2 2
If a < 0, b = 0, then θ = π .

 
b
• Case (ii): If a > 0, b ∈ R, then θ = arctan .
a
 
b
• Case (iii): If a < 0, b > 0, then θ = π − arctan .
|a|
 
|b|
• Case (iv): If a < 0, b < 0, then θ = −π + arctan .
|a|

3 For arctan and its properties, we refer the reader to Section 2.4.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 20 — #20

i i

20 Complex Numbers and Geometrical Representations

(a, b) (a, b)

p q
q
q

(|a|, |b|) (a, −b)

q q
−q
−p + q

(a, b) (a, b)

Example 1.2.17 Write the given numbers in the polar form with principal
1 √
arguments: (1) 2 + i3 and (2) (1 − 3).
2
 
√ √ 4
1. |3 + i4| = 9 + 16 = 25 = 5 and arg 3 + i4 = arctan
3
√ iarctan 43
⇒ 3 + i4 = 25 e .
 
1 √  1√ 1 √
2.  (1 − 3) =
 1 + 3 = 1 and arg (1 − 3) = − arctan
2 2 2
√ π 1 √ −π
3 = − ⇒ (1 − 3) = e 3 .
3 2

Definition 1.2.18 Given a non-zero complex number z = reiθ , we define

√ 1 i θ+2πk
n
z = rne n
, where k = 0, 1, 2, . . . , n − 1.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 21 — #21

i i

Preliminaries 21

Example 1.2.19 Find the cube roots of −8.

First, we write −8 = 8eiπ , and hence, (−8)1/3 = 2ei(π+2kπ)/3 , k = 0, 1, 2.
Hence, the three values of (−8)1/3 are as follows:
√
1. 2eiπ/3 = 2(cos(π/3) + i sin(π/3)) = 1 + i 3.

2. 2eiπ = 2(cos(π ) + i sin(π)) = −2.

√
3. 2ei4π/3 = 2(cos(4π/3) + i sin(4π/3)) = 1 − i 3.

The sum of two complex numbers (a, b) and (c, d) can be viewed geomet-
rically as follows. Construct a parallelogram using the line segment joining
(0, 0), (a, b) and the line segment joining (0, 0), (c, d), as the adjacent sides.
Then the fourth vertex of the parallelogram is (a, b)+(c, d). See the following
figure.

(a + c, b + d)

(a, b)

(c, d )

(0, 0)

The product of two complex numbers z1 and z2 can be understood easily

if we represent them in polar coordinates rather than in Cartesian coordinates.
Let r1 exp(θ1 ), r2 exp(θ2 ) be the given two non-zero complex numbers. (If at
least one of them is zero, then the product is obvious.) Then the product of
these two complex numbers is the complex number r1 · r2 exp(θ1 + θ2 ). Geo-
metrically, r1 ·r2 exp(θ1 +θ2 ) is obtained by rotating the line joining the origin
and r1 exp(θ1 ) through an angle θ2 and then multiplying it by r2 . Multiplying
r1 exp(θ1 + θ2 ) by r2 expands (or shrinks) the length of line segment joining
(0, 0) and r1 exp(θ1 + θ2 ) by the scale r2 , when r2 > 1 (or 0 < r2 < 1).
At this stage, take one of the end points of the resulting line segment, other
than origin, as the product of the given complex numbers. See the following
figure.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 22 — #22

i i

22 Complex Numbers and Geometrical Representations

r2 exp(iq 2)
r1r2 exp(i(q 1 + q 2))

r2 exp(i(q 1 + q 2)) r1 exp(iq 1)

q1 q2
q1

The complex conjugate of a complex number (a, b) can be viewed as the

point of reflection with respect to the real axis.

Geometrically, the additive inverse of a complex number can be viewed

as the rotation of the given non-zero complex number through an angle π or
equivalently, it is the reflection of z with respect to the real axis followed by
another reflection with respect to the imaginary axis.

−z

Exercise 1.2.20 Geometrically explain the multiplicative inverse of a com-

plex number.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 23 — #23

i i

Preliminaries 23

Now, we find the equation of the straight line passing through the given
two points ζ and ξ . Geometrically, the slope of the line segment joining ξ
and ζ is same as the slope of the line segment joining 0 and ζ − ξ .

z+x

Line segment joining x and z.

z
0

Line segment joining 0 and z − x.

z−x

−x

Therefore, the straight line passing through ξ and ζ is same as the straight
line passing through ξ with the slope as that of the line segment joining 0 and
ζ − ξ . Note that if t moves from 0 towards ∞, then t(ζ − ξ ) moves from 0
to ∞, and it passes through ζ − ξ when t = 1. If t moves from 0 to −∞,
then t(ζ − ξ ) moves from 0 to ∞ through −(ζ − ξ ) when t = −1. Hence, the
equation of the straight line passing through 0 and ζ − ξ is t(ζ − ξ ), t ∈ R.
If the straight line passing through 0 and ζ − ξ is translated by ξ , then the
resulting straight line passes through ξ and ζ . Hence, its equation becomes
ξ + t(ζ − ξ ), t ∈ R. Therefore, the equation of a straight line passing through
a and having slope as the slope of the line segment joining 0 and b, for some
a, b ∈ C with b  0, is given by

z = a + tb, t ∈ R. (1.2)

Now, we find an implicit equation of the same straight line.

z−a
z lies on the line (1.2) iff ∈R
b  
z−a z−a
iff =
b b
iff zb − zb + ab − ab = 0
iff z(ib) + zib + 2Im ab = 0.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 24 — #24

i i

24 Sequences and Series of Complex Numbers

The circle with centre a ∈ C and radius r > 0 can be defined as the set of all
points that are at distance r units from a. More explicitly {z : |z − a| = r}.
We shall prove latter (in Corollary 2.4.25) that z − a = r(exp(iθ )) for some
θ ∈ [0, 2π]. Thus, an equation of a circle can be given by z = a + r exp(iθ ),
θ ∈ [0, 2π].

The implicit equation of the circle |z − a| = r is given as follows.

z belongs to the circle iff |z − a|2 = r2
iff |z|2 − za − za + |a|2 − r2 = 0
iff |z|2 + z(−a) − z(−a) + |a|2 − r2 = 0.
The interesting fact is that there is a combined version of an equation for
straight lines and circles. This is simply
α|z|2 + βz + βz + γ = 0, where α, γ ∈ R and β ∈ C with not α = 0 = β.

straight line if α = 0.
The above equation represents a
circle if α  0.

1.3 SEQUENCES AND SERIES OF COMPLEX

NUMBERS
In this section, first we recall some definitions and results on sequences and
series of complex numbers.

Definition 1.3.1 A sequence (an ) of complex numbers is said to be

1. convergent if there exists a ∈ C such that given > 0, then there exists
m ∈ N such that |an − a| < , ∀n ≥ m. In this case, we call ‘a’ the
limit of (an ) and we write an → a as n → ∞ or a = lim an or (an )
n→∞
converges to a.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 25 — #25

i i

Preliminaries 25

2. a Cauchy sequence if given > 0, then there exists m ∈ N such that

|aj − ak | < , ∀j, k ≥ m.

3. bounded if there exists M > 0 such that |an | ≤ M, ∀n ∈ N.

RESULT 1.3.2 If (an ) is a convergent sequence, then (an ) is a Cauchy

sequence.

Proof: Let (an ) be a convergent sequence. Then, there exists a ∈ C such that
given > 0, there exists m ∈ N such that

|an − a| < , ∀n ≥ m.
2
For the same m ∈ N, if j, k ≥ m, then

|aj − ak | ≤ |aj − a| + |ak − a| < + = .

2 2
Hence, (an ) is a Cauchy sequence. 

RESULT 1.3.3 If (an ) is a Cauchy sequence, then (an ) is a bounded sequence.

Proof: For = 1, there exists m ∈ N such that

|aj − ak | < 1, ∀j, k ≥ m.

Then, put

K = max{1 + |am |, |an | : n = 1, 2, 3, . . . , m − 1}.

Then, K > 0 and for every n ≤ m − 1, obviously we have |an | ≤ K and for
n ≥ m,
|an | ≤ |an − am | + |am | < 1 + |am | ≤ K,
which implies that (an ) is a bounded sequence. 
From the above two results, we obtain the following corollary.

COROLLARY 1.3.4 If (an ) is a convergent sequence, then (an ) is a bounded

sequence.

Example 1.3.5 Every constant sequence is obviously a convergent sequence.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 26 — #26

i i

26 Sequences and Series of Complex Numbers

 
n
Example 1.3.6 converges to 1.
n+1

1
Given > 0, applying Archimedian property, we choose N > . If n ≥ N ,
then    
 n   1  1 1
   
 n + 1 − 1 =  n + 1  = n + 1 < n < .
n
Thus, → 1 as n → ∞.
n+1

Definition 1.3.7 Given a sequence (an ), we mean (ank ) a subsequence of (an )

if (nk ) is a strictly increasing sequence of natural numbers. That is, nk < nj if
k < j.

LEMMA 1.3.8 If (xn ) is a Cauchy sequence and it has a subsequence (xnk )

converging to x, then (xn ) also converges to the same x.

Proof: Let > 0 be given. Then, there exist N , j ∈ N such that

|xn − xm | < , ∀m, n ≥ N and |xnk − x| < , ∀k ≥ j.

2 2
If p = max{N, nj }, then for every n ≥ p, as np ≥ p ≥ N and p ≥ nj ≥ j, we
have,
|xn − x| ≤ |xn − xnp | + |xnp − x| < + = .
2 2
Hence, the lemma follows. 

THEOREM 1.3.9 Let (an ) and (bn ) be complex sequences and a, b ∈ C.

(1) an → a as n → ∞ iff Re an → Re a and Im an → Im a as n → ∞.
Let an → a and bn → b as n → ∞. Then,
(2) can → ca as n → ∞.
(3) Every subsequence of (an ) converges to a.
(4) an + bn → a + b as n → ∞.
(5) an · bn → a · b as n → ∞.
an a
(6) → as n → ∞ if b  0, bn  0, ∀n ∈ N.
bn b

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 27 — #27

i i

Preliminaries 27

Proof:

1. Assume that an → a as n → ∞. Then, for a given > 0, there exists

m ∈ N such that |an − a| < , ∀n ≥ m. For the same m ∈ N, if n ≥ m,
we have

|Re an − Re a| ≤ |an − a| < and |Im an − Im a| ≤ |an − a| <

Conversely, assume that Re an → Re a and Im an → Im a as n → ∞.

Given > 0, there exist m1 , m2 ∈ N such that

|Re an − Re a| < √ , ∀n ≥ m1
2
and
|Im an − Im a| < √ , ∀n ≥ m2 .
2
For n ≥ max{m1 , m2 }, we have

 2 2
|an − a| = |Re an − Re a|2 + |Im an − Im a|2 < + = .
2 2

Hence, an → a as n → ∞.

2. Using an → a as n → ∞ for a given > 0, we choose m ∈ N such

that
|an − a| < , ∀n ≥ m.
1 + |c|
For n ≥ m,

|can − ca| = |c||an − a| ≤ |c| ≤ .

1 + |c|
Therefore, can → ca as n → ∞.

3. Let (ank ) be a subsequence of the given sequence (an ). Then, (nk ) is

a strictly increasing sequence of natural numbers. Given > 0, there
exists m ∈ N such that |an − a| < , ∀n ≥ m. Choose j ∈ N such that
nj > m. If k ≥ j, then nk ≥ nj > m, and hence, |ank − a| < .

4. Given > 0, choose m1 , m2 ∈ N such that

|an − a| < , ∀n ≥ m1 and |bn − b| < , ∀n ≥ m2 .

2 2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 28 — #28

i i

28 Sequences and Series of Complex Numbers

If n ≥ max{m1 , m2 }, then

|(an + bn ) − (a + b)| ≤ |an − a| + |bn − b| < + = .

2 2
Thus, an + bn → a + b as n → ∞.
5. Using Corollary 1.3.4, we find an M > 0 such that |an | ≤ M , ∀n ∈ N.
Given > 0, choose m1 , m2 ∈ N, such that

|an − a| < , ∀n ≥ m1 and |bn − b| < , ∀n ≥ m2 .

2(1 + |b|) 2M
If n ≥ max{m1 , m2 }, then
|an · bn − a · b| = |an · bn − an · b + an · b − a · b|
≤ |an | · |bn − b| + |b| · |an − a|
< M + |b|
2M 2(1 + |b|)
≤ + = .
2 2
6. It is left as an exercise to the reader. For a hint, see Theorem 1.6.3. 

LEMMA 1.3.10 Let (an ) and (bn ) be real sequences such that an ≤ bn ∀n ∈ N.
If an → a and bn → b as n → ∞, then a ≤ b.

Proof: If a > b, then a−b

2 > 0. Then, for = a−b
2 , there exist N1 , N2 ∈ N
such that
a−b a−b
|an − a| < , ∀n ≥ N1 and |bn − b| < , ∀n ≥ N2 .
2 2
If m ≥ max{N1 , N2 }, then
a−b a−b
|am − a| < and |bm − b| < .
2 2
Therefore,
a−b a−b
a − b = a − am + am − b ≤ (a − am ) + (bm − b) < + = a − b.
2 2
which is a contradiction. Thus, a ≤ b. 

THEOREM 1.3.11 (Cauchy criterion)

Let (an ) be a sequence of complex numbers. Then, (an ) is a Cauchy sequence
iff (an ) is a convergent sequence.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 29 — #29

i i

Preliminaries 29

The half of the proof of the above theorem is already proved and the remain-
ing half follows from the fact that C is a complete metric space. (See Theorem
1.4.46.)

Definition 1.3.12 Let (an ) be a sequence of complex numbers. We write

an → ∞ as n → ∞ if for every K > 0, there exists m ∈ N such that
|an | > K, ∀n ≥ m.

Definition 1.3.13 Let (xn ) be a sequence of real numbers.

1. We say that xn → +∞ as n → ∞ if for every K > 0, there exists

m ∈ N such that xn > K, ∀n ≥ m.

2. We say that xn → −∞ as n → ∞ if for every K > 0, there exists

m ∈ N such that xn < −K, ∀n ≥ m.

Definition 1.3.14 Let (xn ) be a sequence of real numbers, then an extended

real number α is said to be the limit superior or upper limit, which is denoted
by lim sup xn , if it satisfies the following conditions:
n→∞

1. There exists a subsequence (xnk ) of (xn ) such that xnk → α as k → ∞.

2. Given > 0, there exists m ∈ N such that an < α + , ∀n ≥ m.

Definition 1.3.15 Let (xn ) be a sequence of real numbers, then an extended

real number α is said to be the limit inferior or lower limit, which is denoted
by lim inf xn , if it satisfies the following conditions:
n→∞

1. There exists a subsequence (xnk ) of (xn ) such that xnk → α as k → ∞.

2. Given > 0, there exists m ∈ N such that an > α − , ∀n ≥ m.

LEMMA 1.3.16 For a real sequence (xn ), lim inf xn ≤ lim sup xn .
n→∞ n→∞

Proof: Let > 0 be given. If α = lim inf xn and β = lim sup xn , then there
n→∞ n→∞
exist N1 , N2 ∈ N such that

xn > α − , ∀n ≥ N1 and xn < β + , ∀n ≥ N2 .

If m ≥ max{N1 , N2 }, then α− < xm and xm < β+ imply that α− < β+ .

As > 0 is arbitrary, we get α ≤ β . 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 30 — #30

i i

30 Sequences and Series of Complex Numbers

RESULT 1.3.17 Let (xn ) be a sequence of real numbers. (xn ) converges to x

iff lim sup xn = lim inf xn = x.
n→∞ n→∞

The proof of the above result is straightforward from the definitions.

LEMMA 1.3.18 If (xn ) and (yn ) be sequences of real numbers such that xn ≤
yn , ∀n ∈ N, then lim sup xn ≤ lim sup yn and lim inf xn ≤ lim inf yn .
n→∞ n→∞ n→∞ n→∞

Proof: Let x and y be the limit superior of (xn ) and (yn ), respectively, then by
definition, given > 0, there exists N ∈ N such that yn < y + , ∀n ≥ N .
From the hypothesis, we get

xn ≤ yn < y + , ∀n ≥ N.

Let (xnk ) be a subsequence of (xn ) such that

n1 ≥ N and xnk → x as k → ∞.

As the constant sequence converges to the same constant, in view of Lemma

1.3.10, from xnk < y + , ∀k ∈ N, we infer that lim sup xn = lim xnk x ≤
n→∞ k→∞
y + . Since > 0 is arbitrary, we get

lim sup xn = x ≤ y = lim sup yn .

n→∞ n→∞

Similarly, we can prove the other inequality. 

THEOREM 1.3.19 Let (xn ) and (yn ) be sequences of positive real numbers.
If (xn ) converges to a positive real number, then lim sup(xn · yn ) = lim xn ·
n→∞ n→∞
lim sup yn .
n→∞

Proof: Let x = lim xn and y = lim sup yn . First, we assume that y ∈ R.

n→∞ n→∞
Given ∈ (0, 1), we find m1 , m2 ∈ N such that

|xn − x| < ⇔x− < xn < x + ∀n ≥ m1

1+x+y 1+x+y 1+x+y

and
yn < y + , ∀n ≥ m2 .
1+x+y

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 31 — #31

i i

Preliminaries 31

For n ≥ max{m1 , m2 }, we get

 2
(x + y)
xn · yn < x·y+ +
1+x+y 1+x+y
(x + y)
< x·y+ + (as <1)
1+x+y 1+x+y 1+x+y
(1 + x + y)
< x·y+
1+x+y
= x·y+ .
If (ynk ) is a subsequence of (yn ) converges to y, then using Theorem 1.3.9, we
get (xnk · ynk ) converges to x · y. Thus, x · y = lim sup(xn · yn ).
n→∞
If y = +∞, then there exists a subsequence (ynk ) of y such that ynk → ∞
as k → ∞. Let K > 0 be given, then by assumption, there exists j ∈ N such
that ynk > 2K
x , ∀k ≥ j. For = 2 > 0, we choose m such that |xn − x| < 2 ,
x x

∀n ≥ m. Hence,
x x
x − xn ≤ |xn − x| < ⇒ xn > , ∀n ≥ m.
2 2
Next, choose p ∈ N such that np ≥ m. Now, for k ≥ max{j, p}, we get
x 2K
xnk · ynk > · = K.
2 x
Hence, xnk · ynk → ∞ as k → ∞. As we have
xn · yn < ∞ + , ∀ > 0, ∀n ∈ N.
we get lim sup(xn · yn ) = ∞ = x · y. 
n→∞

THEOREM 1.3.20 Let (xn ) be a sequence of positive real numbers, then

xn+1 √ √ xn+1
lim inf ≤ lim inf n xn ≤ lim sup n xn ≤ lim sup
n→∞ xn n→∞ n→∞ n→∞ xn

√ xn+1
Proof: Let α = lim sup n xn and β = lim sup . If β = +∞, then
n→∞ n→∞ xn
obviously we have α ≤ β . Therefore, we assume that β < +∞. Then by
definition of limit superior, for a given > 0, there exists m ∈ N such that
xn+1
< β + , ∀n ≥ m. For n ≥ m, we have
xn
xn xn−1 xm+1
xn = × × ··· × × xm < xm × (β + )n−m .
xn−1 xn−2 xm

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 32 — #32

i i

32 Sequences and Series of Complex Numbers

 1/n
√ xm xm
which implies that xn <
n (β + ). If c = , then we
(β + ) m (β + )m
get c1/n → 1 as n → ∞. (Because if c > 1, then c1/n = 1 + δn for some
δn > 0 ⇒ c = (1 + δn )n > nδn ⇒ δn < nc → 0 as n → ∞ ⇒ c1/n →
1 as n → ∞. If c < 1, then c−1 > 1 ⇒ c−1/n → 1 ⇒ c1/n → 1 as n → ∞).
Therefore, using Lemma 1.3.18, we get
√
α = lim sup n
xn ≤ β + .
n→∞

As > 0 is arbitrary, we get

√ xn+1
lim sup n
xn = α ≤ β = lim sup .
n→∞ n→∞ xn

Similarly, we can prove that

xn+1 √
lim inf ≤ lim inf n xn .
n→∞ xn n→∞

Using Lemma 1.3.16, we get

√ √
lim inf n
xn ≤ lim sup n xn .
n→∞ n→∞

Thus, the theorem follows. 

Definition 1.3.21 Let fn : A → C, ∀n ∈ N, where A ⊆ C. We say that the

sequence (fn ) of functions converges

1. point-wise to f on A, if for a given z ∈ A and a given > 0, there exists

mz ∈ N such that |fn (z) − f (z)| < , ∀n ≥ mz .

2. uniformly to f on A, if given > 0, there exists m ∈ N such that

|fn (z) − f (z)| < , ∀n ≥ m and ∀z ∈ A.

Definition 1.3.22 Let x0 ∈ [a, b] and f : (a, b) × (c, d) → C. We say that

f (x, y) →  as x → x0 uniformly in y on (c, d). If given > 0, then
there exists δ > 0 such that 0 < |x − x0 | < δ ⇒ |f (x, y) − f (x0 , y)| < ,
∀y ∈ (c, d).

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 33 — #33

i i

Preliminaries 33

∞
It is customary to define series of complex numbers by a formal sum an .
n=1
Before talking about its convergence, the sum has no meaning other than in
∞
the expression an , first term is a1 , second term is a2 , third term is a3 , and
n=1
so on. In the sequence (an ) of complex numbers also, we say that first term
is a1 , second term is a2 , third term is a3 , and so on. So what is the difference
∞
between (an ) and an ? To find the difference, we can redefine the series of
n=1
complex numbers as follows.

Definition 1.3.23 (Rigorous Definition)

Let (an ) be a sequence of complex numbers. We say that the sequence (an )
n
is summable if (Sn ) converges, where Sn = ak , ∀n ∈ N. In this case, we
k=1
∞ ∞
call lim Sn the sum of (an ) and is denoted by an . The sum an is also
n→∞ n=1 n=1
called a series (or) an infinite series.
However, the old definition of a series (formal sum) is in usage for a long
∞
time and most of the readers are accustomed with using the notation an
n=1
before discussing its convergence, a few similar less rigorous statements on
series are also used in this book, only to facilitate the reader

∞
Definition 1.3.24 A series an of complex numbers is said to be conver-
n=0
m
gent if sm = ak , ∀m ∈ N and sm → s as m → ∞ for some complex
k=0
∞
number s. In this case, we call ‘s’ the sum of the series an and is also
n=0
denoted by the series itself.

∞
Definition 1.3.25 A series an of complex numbers is said to be absolutely
n=0
∞
convergent if the series |an | converges.
n=0

∞ ∞
RESULT 1.3.26 A series an converges iff lim an = 0.
n=0 m→∞ n=m

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 34 — #34

i i

34 Sequences and Series of Complex Numbers

∞
Proof: an converges to s iff sm → s as m → ∞, where sm = a0 + a1 +
n=0
∞
a2 + · · · + am , ∀m ∈ N and s = an iff sm−1 − s → 0 as m → ∞ iff
n=0
∞
lim an = 0. 
m→∞ n=m
The following result gives a necessary condition for a convergent series
of complex numbers.
∞
RESULT 1.3.27 If an converges, then lim an = 0.
n=0 n→∞

∞
Proof: Let an converge. If
n=0

sm = a0 + a1 + a2 + · · · + am , ∀m ∈ N,
then (sm ) is a convergent sequence and by Result 1.3.2, (sm ) is a Cauchy
sequence. Let > 0 be given. Then, there exists m ∈ N such that
|sj − sk | < , ∀j, k ≥ m.
If n ≥ m + 1, then n, n − 1 ≥ m, and hence, |an | = |sn − sn−1 | < . Hence,
(an ) converges to 0. 

Definition 1.3.28 (Rearrangement of a series)

∞ ∞
Let an be a given series. If f : N → N is a bijection, then af (n) is
n=1 n=1
∞
called a rearrangement of an .
n=1

∞
RESULT 1.3.29 If an converges to S absolutely, then every rearrangement
n=1
∞
of an converges to the same S .
n=1

∞
Proof: Let af (n) be a rearrangement of an , where f : N → N is a
n=1 n=1
m m
bijection. Let Sm = |an | and Tm = af (n) , ∀m ∈ N. Since (Sm ) con-
n=1 n=1
verges, by Cauchy criterion (Theorem 1.3.11), we have (Sm ) is a Cauchy

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 35 — #35

i i

Preliminaries 35

sequence and hence for a given > 0, there exists N0 ∈ N such that
q
|Sp − Sq | = |an | < ∀q ≥ p ≥ N0 .
n=p
2
∞
Since an convergs to s, there exists N ∈ N such that N ≥ N0 and
n=1
 
 p 
 
 an − S  ≤ ∀p ≥ N. (1.3)
  2
n=1
Since f (N) = N we can choose M ∈ N such that {1, 2, . . . , N} ⊆ {f (n) : 1 ≤
n ≤ M}. Now for all p ∈ N with p ≥ M , we have
 p   q 
   q   
Tp − S  =  af (n) −
 
an  + 

an − S  (where q ≥ p.)
   
n=1 n=1 n=1

≤ |an | + (using q ≥ p ≥ M ≥ N and (1.3))

2
n A
where A ⊆ N such that K ≤ n ≤ J , ∀n A for some K, J N
with N ≤ K ≤ J .
J
≤ |an | +
2
K

< + = ,
2 2
∞
because by the choice of M , we have N ≤ K ≤ J . Thus af (n) converges
n=1
to S . 

THEOREM 1.3.30 (Comparison test)

If (an ) is a sequence of complex numbers and (bn ) is a sequence of non-
negative real numbers such that |an | ≤ bn , ∀n ≥ m, for some m ∈ N and
∞ ∞
bn converges, then an converges absolutely.
n=0 n=0

Proof: For each n ∈ N, let

sn = |a0 | + |a1 | + · · · + |an | and tn = b0 + b1 + · · · + bn .
By assumption, (tn ) is a convergent sequence, and hence, (tn ) is a Cauchy
sequence. Then, for a given > 0, there exists p ∈ N such that
|tj − tk | < , ∀j, k ≥ p.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 36 — #36

i i

36 Some Topological Properties of the Complex Plane

Now for the same p, if j > k ≥ p, then we have

j j
|sj − sk | = |an | ≤ bn = |tj − tk | < .
n=k+1 n=k+1

Hence, it follows that |sj − sk | < whenever j, k ≥ p. Therefore, (sn )

∞
converges, by Cauchy’s criterion, and hence, an converges absolutely. 
n=0

∞
Example 1.3.31 If z ∈ C with |z| < 1, then the geometric series zn
n=0
converges.

If z = 0, then obviously the series converges to 0. For z  0, then we have

1 1
0 < |z| < 1. Hence, we write |z| = , where r = − 1 > 0. For every
1+r |z|
n > 2, since
(1 + r)n > nC1 r = nr
we get
1 1
0 < |z|n = < → 0 as n → ∞.
(1 + r) n nr
If sn = 1 + z + z2 + · · · + zn , then

sn (1 − z) = (1 + z + z2 + · · · + zn )(1 − z) = 1 − zn+1 , ∀n ∈ N.

Therefore, it follows that

1 − zn+1 1
sn = → as n → ∞
1−z 1−z
∞ 1
and hence, zn = .
n=0 1−z

1.4 SOME TOPOLOGICAL PROPERTIES OF THE

COMPLEX PLANE
Although we assume that the reader is familiar with mathematical analy-
sis, we briefly recall some important definitions and results on topology of
complex plane, which will be required in the following sequel.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 37 — #37

i i

Preliminaries 37

Definition 1.4.1 The standard metric d : C × C → R is defined by d(z, w) =

|z − w|, ∀z, w ∈ C.

From the properties of | · |, we obtain the following:

1. |z − w| ≥ 0, ∀z, w ∈ C,

2. |z − w| = |w − z|, ∀z, w ∈ C,

3. |z − w| = 0 iff z = w,

4. |z − w| ≤ |z − ζ | + |ζ − w|, ∀z, ζ , w ∈ C.

Definition 1.4.2 For a given z0 ∈ C and r > 0, by a neighbourhood of

z0 or by the open ball with center z0 and radius r > 0, we mean the set
B(z0 , r) = {z ∈ C : |z − z0 | < r}.

Definition 1.4.3 Let U ⊆ C and a ∈ C. We say that a is

1. a limit point of U if for every r > 0, we have (B(a, r) \ {a}) ∩ U  ∅.

2. an interior point of U if there exists > 0 such that B(a, ) ⊆ U .

Definition 1.4.4 A subset U of C is said to be

1. an open set if every point of U is an interior point of U .

2. a closed set if every limit point of U is a point of U .

RESULT 1.4.5 Let U ⊆ C. U is an open set iff U c is a closed set.

Proof: U is open iff every point of U is an interior point of U

iff for every x ∈ U , there exists rx > 0 such that
B(x, rx ) ⊆ U
iff for every x  U c , there exists rx > 0 such that
B(x, rx ) ∩ U c = ∅
iff for every x  U c , there exists rx > 0 such that
(B(x, rx ) \ {x}) ∩ U c = ∅
iff every x  U c is not a limit point of U c
iff every limit point of U c is a point of U c
iff U c is closed.
Hence, the result. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 38 — #38

i i

38 Some Topological Properties of the Complex Plane

THEOREM 1.4.6 Let Aα , Aj ⊆ C for all α ∈  and for all j ∈ {1, 2, . . . , n},
where  is an arbitrary index set.

(i) If Aα is open ∀α ∈ , then Aα is open.
α∈


n
(ii) If Aj is open ∀j ∈ {1, 2, . . . , n}, then Aj is open.
j=1

 
Proof: To prove Aα is open, let x ∈ Aα be arbitrary. Then x ∈ Aβ
α∈ α∈
 some β ∈ 
for . As Aβ is open, there exists r > 0 such that B(x, r) ⊆ Aβ ⊆
Aα . Thus, Aα is open.
α∈ α∈

n
Next, if x ∈ Aj is arbitrary, then x ∈ Aj , ∀j ∈ {1, 2, . . . , n}. Since each
j=1
Aj is open, there exists rj > 0 such that B(x, rj ) ⊆ Aj , ∀j ∈ {1, 2, . . . , n}. If
s = min{rj : 1 ≤ j ≤ n}, then s > 0, and hence, B(x, s) ⊆ B(x, rj ) ⊆ Aj ,

n 
n
∀j ∈ {1, 2, . . . , n} ⇒ B(x, s) ⊆ Aj . Thus, Aj is open. 
j=1 j=1

THEOREM 1.4.7 Let Bα , Bj ⊆ C for all α ∈  and for all j ∈ {1, 2, . . . , n},
where  is an arbitrary index set.

(i) If Bα is closed ∀α ∈ , then Bα is closed.
α∈


n
(ii) If Bj is closed ∀j ∈ {1, 2, . . . , n}, then Bj is closed.
j=1

Proof: Proof of this theorem follows from Theorem 1.4.5, De-Morgan’s laws,
and Theorem 1.4.6. 

Definition 1.4.8 Let A ⊂ C. Define closure Cl A of A by A ∪ A , where A is

the set of all limit points of A.

LEMMA 1.4.9 x ∈ Cl A iff B(x, r) ∩ A  ∅ for every r > 0.

Proof of this lemma follows directly from the definition of Cl A.

RESULT 1.4.10 The closure operator satisfies the following properties:

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 39 — #39

i i

Preliminaries 39

1. A ⊆ B ⇒ Cl A ⊆ Cl B.

2. A is closed iff A = Cl A.

3. Cl A is the smallest closed set containing A.

4. Cl Cl A = Cl A.

5. Cl (A ∪ B) = (Cl A) ∪ (Cl B).

6. Cl (A ∩ B) ⊆ (Cl A) ∩ (Cl B).

Proof:

1. If A ⊂ B, then using the previous lemma, we get

x ∈ Cl A ⇒ B(x, r) ∩ A  ∅, ∀r > 0
⇒ B(x, r) ∩ B  ∅, ∀r > 0
⇒ x ∈ Cl B.

2. A = Cl A ⇔ A = A ∪ A ⇔ A ⊆ A ⇔ A is closed.

3. Let x  Cl A. Then by previous lemma, there exists r > 0 such that

B(x, r) ∩ A = ∅. For every y ∈ B(x, r), there exists sy = r − |x − y| > 0
such that B(y, sy ) ⊆ B(x, r), and hence, B(y, sy ) ∩ A = ∅. Hence, it
follows that no point of B(x, r) is a point of Cl A by the same lemma.
Therefore, B(x, r) ⊆ (Cl A)c . Thus, we have proved that (Cl A)c is an
open set, and hence, Cl A is a closed set.
Next if B is a closed subset of C such that A ⊂ B, then using the
properties (1) and (2), we have Cl A ⊆ Cl B. Therefore, Cl A ⊆ Cl
B = B. Thus, Cl A is the smallest closed set containing A.

4. Using the property (3), Cl A is a closed set, and by using the property
(2), we get Cl Cl A = Cl A.

5. Clearly, A ⊆ Cl A and B ⊆ Cl B imply that A ∪ B ⊆ (Cl A) ∪ (Cl B).

As (Cl A) ∪ (Cl B) is a closed set (being a union of two closed sets)
containing A ∪ B and Cl (A ∪ B) is the smallest closed set containing
A ∪ B, we get Cl (A ∪ B) ⊆ (Cl A) ∪ (Cl B). On the other hand, using the
property (1), A ⊆ A ∪ B and B ⊆ A ∪ B imply that Cl A ⊆ Cl (A ∪ B)
and Cl B ⊆ Cl (A ∪ B). Therefore, (Cl A) ∪ (Cl B) = Cl (A ∪ B). Thus,
the property (5) holds.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 40 — #40

i i

40 Some Topological Properties of the Complex Plane

6. Since A∩B ⊆ A and A∩B ⊆ B, using the property (1), we have Cl (A∩
B) ⊆ Cl A and Cl (A ∩ B) ⊆ Cl B. Thus, Cl (A ∩ B) ⊆ (Cl A) ∩ (Cl B).

Example 1.4.11 For every a ∈ C and r > 0, Cl B(a, r) = {x ∈ C :

|x − a| ≤ r}.
Suppose y  {x ∈ C : |x − a| ≤ r}. Then, |y − a| > r. If s ∈ R is such that 0 <
s < |y−a|−r, then we claim that B(y, s)∩B(a, r) = ∅. If z ∈ B(y, s)∩B(a, r),
then |z − y| < s and |z − a| < r which implies that

|y − a| ≤ |y − z| + |z − a| < s + r < |y − a|.

This is not possible. Hence, B(y, s) ∩ B(a, r) = ∅. Therefore, y  Cl B(a, r) by

Lemma 1.4.9. Conversely, if y  Cl B(a, r), then there exists t > 0 such that
y−a y−a
B(y, t) ∩ B(a, r) = ∅. As y − 2t |y−a| ∈ B(y, t), we have y − 2t |y−a|  B(a, r),
and hence by applying |y − a| > t ⇒ 0 < 2|y−a| < 1, we get
t

 
 
r ≤ y − t y − a − a
 2 |y − a| 
 
 t 
= |y − a| 1 − 
2|y − a| 
 
t
= |y − a| 1 −
2|y − a|
< |y − a|.

Therefore, y  {x ∈ C : |x − a| ≤ r}, and hence, Cl B(a, r) = {x ∈ C :

|x − a| ≤ r}.

Example 1.4.12 There exists A ⊂ C and B ⊂ C such that Cl (A ∩ B) 

(Cl A) ∩ (Cl A).
We consider A = B(1, 1) and B = B(2, 1). Clearly, Cl (A ∩ B) = Cl ∅ = ∅
but (Cl A) ∩ (Cl B) = {x ∈ C : |x − 1| ≤ 1} ∩ {x ∈ C : |x − 2| ≤ 1} = {1}.

Exercise 1.4.13 Prove that x ∈ Cl A iff there exists a sequence (xn ) in A such
that xn → x as n → ∞.

Exercise 1.4.14 If A is a bounded subset of R, then prove that sup A, inf A ∈

Cl A.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 41 — #41

i i

Preliminaries 41

Definition 1.4.15 A subset A of R is dense in R if Cl A = R. Similarly, if

A ⊆ C with Cl A = C, then A is called a dense subset of C.

Example 1.4.16 The set of all rational numbers is dense in R.

To justify this example, we shall show that every point of R is a limit

point of Q. Let r > 0 be given, then using the observation 3(l) in page 6, we
choose q ∈ Q such that x < q < x + r. Therefore, q ∈ [B(x, r) \ {x}] ∩ Q.
Thus, R ⊆ Q ⊆ Cl Q.

Definition 1.4.17 Let A ⊂ C. The interior Int A of A is defined by the set of

all interior points of A.

Exercise 1.4.18 Let A, B ⊆ C. Then prove that

(i) A ⊆ B ⇒ Int A ⊆ Int B.
(ii) A is open ⇔ A = Int A.
(iii) Int A is the largest open set contained in A.
(iv) Int Int A = A.
(v) Int (A ∩ B) = Int A ∩ Int B.
(vi) Int (A ∪ B) ⊇ Int A ∪ Int B.
Definition 1.4.19 Let A, B ⊂ C. A and B are said to be separated if (Cl A) ∩
B = ∅ = A ∩ (Cl B).

Definition 1.4.20 Let S be a subset of C. S is said to be connected if S cannot

be a subset of union of two separated sets A and B such that A ∩ S  ∅ and
B ∩ S  ∅.

Definition 1.4.21 (Interval) A subset I of R is said to be an interval if z ∈ I

whenever x, y ∈ I and x < z < y.

There are four types of intervals, which are as follows:

1. [a, b] = {x ∈ R : a ≤ x ≤ b}.
2. [a, b) = {x ∈ R : a ≤ x < b}.
3. (a, b] = {x ∈ R : a < x ≤ b}.
4. (a, b) = {x ∈ R : a < x < b}.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 42 — #42

i i

42 Some Topological Properties of the Complex Plane

THEOREM 1.4.22 A subset S of R is connected iff S is an interval.

Proof: Let S be a connected subset of R. Suppose S is not an interval, then

there exist x, y ∈ S and z ∈ R such that x < z < y but z  S .
Put A = (−∞, z) and B = (z, ∞). Clearly, it follows that

S ⊆ A ∪ B, A ∩ S  ∅ and B ∩ S  ∅.

As

A ∩ (Cl B) = (−∞, z) ∩ [z, ∞) = ∅ = (−∞, z] ∩ (z, ∞) = (Cl A) ∩ B,

we get that S is disconnected, which is a contradiction. Therefore, S is an

interval.
Conversely, suppose S is an interval and if there exist subsets A, B of R such
that

S ⊆ A ∪ B, S ∩ A  ∅  S ∩ B and (Cl A) ∩ B = ∅ = (Cl B) ∩ A.

As S ∩ A  ∅  S ∩ B, we choose α ∈ S ∩ A and γ ∈ S ∩ B. Without loss of

generality, we assume that α < γ , since α  γ . If β = sup([α, γ ] ∩ A), then
α ≤ β ≤ γ and α ∈ Cl A. As α, γ ∈ S and S is an interval, we get β ∈ S and
hence,
β ∈ A or β ∈ B (1.4)
Using β ∈ Cl A and (Cl A) ∩ B = ∅, we obtain β  B. Then, β ∈ A, and
hence, β  Cl B. Hence, (β, γ )  B; otherwise, β ∈ [β, γ ] = Cl (β, γ ) ⊆
Cl B, which is not possible. Therefore, there exists δ ∈ (β, γ ) such that δ  B.
From the definition of β and from the inequality β < δ , it follows that δ  A.
Thus, δ  S , as S ⊆ A∪B. This implies that S is not an interval, as α < δ < γ
and α, γ ∈ S . This is a contradiction, and hence S is connected. 

COROLLARY 1.4.23 The set R of all real numbers is a connected set.

Proof: As R itself is obviously an interval, it is connected by the previous

theorem. 

Definition 1.4.24 (Line segment) Let a, b be two points in C. We call the set
La,b = {(1 − t)a + tb : t ∈ [0, 1]} the line segment joining a and b. The point
a is called the initial point of the line segment and b is called the end point of
the line segment.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 43 — #43

i i

Preliminaries 43

Definition 1.4.25 (Polygon) Let {ai : i = 0, 1, 2, . . . , n} be a finite set of

n
complex numbers. The union ∪ Lai−1 ,ai of the given finite number of line
i=1
segments of the form Lai−1 ,ai , i = 1, 2, . . . , n is called a polygon.

THEOREM 1.4.26 If E is a subset of C such that every pair of points a, b can

be joined by a polygon Pa,b ⊆ E, then E is a connected subset of C.
Proof: Suppose E is disconnected, then there exists a pair of sets A, B in C
such that E ⊆ A ∪ B and A ∩ E  ∅  E ∩ B and (Cl A) ∩ B = A ∩ (Cl B) = ∅.
Choose a ∈ E ∩ A and b ∈ E ∩ B. By hypothesis, there exists Pa,b ⊆ E.
On examining the line segments from a to b, we can find a line segment
l = Lc,d ⊆ Pa,b such that c ∈ A and d ∈ B. Put

A1 = {t ∈ [0, 1] : (1 − t)c + td ∈ A ∩ E}

A2 = {t ∈ [0, 1] : (1 − t)c + td ∈ B ∩ E}
As the line segment l ⊆ E ⊆ A ∪ B, we get [0, 1] = A1 ∪ A2 by an easy
verification. We note that

c ∈ A ∩ E ⇒ 0 ∈ A1 ⇒ A1  ∅

and
d ∈ B ∩ E ⇒ 1 ∈ A2 ⇒ A2  ∅
Claim. (Cl A1 ) ∩ A2 = A1 ∩ (Cl A2 ) = ∅.
Suppose t ∈ (Cl A1 ) ∩ A2 . Then, there exists a sequence {tn } from A1
such that tn → t as n → ∞. Therefore, (1 − tn )c + tn d ∈ A, ∀n ∈ N,
and so,
(1 − t)c + td = lim (1 − tn )c + tn d ∈ Cl A
n→∞

Clearly, (1 − t)c + td ∈ B because t ∈ A2 . Similarly, if s ∈ A1 ∩ (Cl A2 ),

then we can show that (1 − s)c + sd ∈ A ∩ (Cl B). Therefore, (Cl A1 ) ∩ A2 =
A1 ∩ (Cl A2 ) = ∅. This gives a separation for the interval [0, 1], which is a
contradiction. 
Immediately, we get the following corollaries.
COROLLARY 1.4.27 Every line segment is connected in C.
COROLLARY 1.4.28 Every polygon is connected.

COROLLARY 1.4.29 In C, every open disc B(a, r) = {x ∈ C : |x − a| < r} is

connected.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 44 — #44

i i

44 Some Topological Properties of the Complex Plane

Proof: Let x, y ∈ B(a, r). We shall show that Lx,y = {(1 − t)x + ty : t ∈
[0, 1]} ⊆ B(a, r).
Now, if t ∈ [0, 1], then

|(1 − t)x + ty − a| = |(1 − t)x + ty − (1 − t)a + ta|

≤ (1 − t)|x − a| + t|y − a|
< (1 − t)r + tr = r

and hence, we get (1 − t)x + ty ∈ B(a, r). Hence, the corollary follows from
the above theorem. 

Definition 1.4.30 An open connected subset of C is called a region.

THEOREM 1.4.31 If E is a region, then given a pair of points x, y of E, there

exists a polygon Qx,y joining x and y such that Qx,y ⊆ E and whose line
segments are parallel to the coordinate axes.

Proof: Throughout this proof, by Qs,t we mean a polygon joining s and t

such that Qs,t ⊆ E and whose line segments are parallel to the coordinate
axes. Assume that E is connected. Fix a ∈ E arbitrarily. Define

A = {x ∈ E : there exists a polygon Qx,a ⊆ E} and B = E \ A.

Clearly, E = A ∪ B and A  ∅. To prove this theorem, we shall show that

E = A. (Because if it is proved and if x, y ∈ E is arbitrary, then there exist
Qx,a , Qy,a ⊆ E, and hence, their union is a polygon in E joining x and y.)

Claim. (Cl A) ∩ B = ∅.
If x ∈ (Cl A) ∩ B, then there exists r > 0 such that B(x, r) ⊆ E (as x ∈ E
and E is open). Then, there exists y ∈ A ∩ B(x, r), using x ∈ Cl A. If
x = (x1 , x2 ) and y = (y1 , y2 ), then let z = (x1 , y2 ). From

|x − z| = |(x1 , x2 ) − (x1 , y2 )| = |x2 − y2 | ≤ |x − y| < r

we get z ∈ B(x, r).

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 45 — #45

i i

Preliminaries 45

(x1, y2)

( y1, y2)

(x1, x2)

By applying the proof of Corollary 1.4.29 twice, we get Lx,z , Lz,y ⊆

B(x, r) ⊆ E, and hence, Px,y = Lx,z ∪ Lz,y is completely contained in E.
Since y ∈ A, there exists Py,a ⊂ E and hence Px,a = Px,y ∪ Py,a ⊆ E.
Thus, x ∈ A. This contradicts the fact that A ∩ B = ∅. Therefore,
(Cl A) ∩ B = ∅.

Claim. A ∩ (Cl B) = ∅.
If a ∈ A ∩ (Cl B), then choose s > 0 such that B(a, s) ⊆ E. As a ∈
Cl B there exists b ∈ B ∩ B(a, s). By a similar argument used in the
justification of previous claim, we can find a polygon Pa,b ⊆ E. This
implies that b ∈ A, which is a contradiction.

Hence, B = ∅, and hence, A = E. This completes the proof of the

theorem. 

Example 1.4.32 There exists a connected subset E of C such that it contains

a pair of points that cannot be joint by a polygon inside E.

If E = {z ∈ C : |z| = 1}, then E is connected because E = f ([0, 2π ]) where

f (t) = eit , t ∈ [0, 2π]; at this level, assume that the equality holds and the
fact that f is continuous. However, for any two distinct points z and w of E,
no point of Lz,w other than z and w belongs to E.
To justify the last statement, let z, w ∈ E be such that z  w. Then, there
exist θ1 , θ2 ∈ [0, 2π) such that z = eiθ1 and w = eiθ2 so that 0 < |θ1 − θ2 | <
2π . Hence, for any t ∈ (0, 1), (1 − t)ztw = (1 − t)tei(θ1 −θ2 ) , which is not
real and non-negative, and hence, |(1 − t)z + tw| < |(1 − t)z| + |tw| =
(1 − t)|z| + t|w| ≤ (1 − t) + t = 1. Thus, (1 − t)z + tw  E. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 46 — #46

i i

46 Some Topological Properties of the Complex Plane

Definition 1.4.33 Let S ⊆ C. A subset A of S is said to be a component of S

if A is a largest connected subset of C contained in S .

That is, there is no connected subset B of C such that A  B ⊆ S .

RESULT 1.4.34 The collection of all components of a set S is a partition of S .

That is, the union of all components of S is S , and components are pairwise
disjoint.

Proof: For each, x ∈ S , let CS [x] be the union of all connected subsets of S
containing x. Clearly, this union is non-empty as {x} is a connected subset of
S containing x. We claim that CS [x] is connected. Suppose, CS [x] ⊆ A ∪ B,
with CS [x] ∩ A  ∅, CS [x] ∩ B  ∅ and A ∩ Cl B = ∅ and B ∩ Cl A = ∅.
Let a ∈ A and b ∈ B. Then, there exist connected subsets Ca and Cb of S
such that a, x ∈ Ca ⊆ S and b, x ∈ Cb ⊆ S . Therefore, Ca ⊆ A ∪ B and
Cb ⊆ A ∪ B. Since Ca is a connected set either Ca ⊆ A or Ca ⊆ B, otherwise
Ca ∩ A  ∅ and Ca ∩ B  ∅, and hence, Ca becomes a disconnected set,
which is a contradiction. As a ∈ A ∩ Ca , we conclude that Ca ⊆ A. Similarly,
we conclude that Cb ⊆ B. Therefore, x ∈ Ca ∩ Cb ⊆ A ∩ B ⊆ A ∩ Cl B = ∅,
which is a contradiction. Hence, CS [x] is a connected set. Next we observe
that CS [x] cannot be properly contained in a connected subset of S because
if K is a connected subset of S containing CS [x], then x ∈ K , and hence by
definition of CS [x], we have K ⊆ CS [x] which implies CS [x] = K . Thus,
CS [x] is a component of S .
As for every x ∈ S , there exists a component CS [x] of S containing x;
to conclude this theorem, we show that any two component of S are either
identical or disjoint. If CS [x] and CS [y] are two components of S such that
CS [x] ∩ CS [y]  ∅, then let z ∈ CS [x] ∩ CS [y]. If CS [z] is the component of
S containing z, then CS [x] ⊆ CS [z], since CS [x] is a connected subset of S
containing z. Now, x ∈ CS [x] ⊆ CS [z] imply that CS [z] is a connected subset
of S containing x, and hence, CS [z] ⊆ CS [x] ⇒ CS [x] = CS [z]. Similarly, we
show that CS [y] = CS [z]. Thus, CS [x] = CS [z] = CS [y]. 

RESULT 1.4.35 Every non-empty connected subset A of a set S is contained

in one and only one component of S .

Proof: Let A be a non-empty connected subset S . Then, there exists x ∈ A ⊆

S . If CS [x] is the component of S containing x, then obviously CS [x] is a
component containing A, as CS [x] is the union of all connected subsets of S
containing x. Since the components of S are pairwise disjoint (by previous
result), CS [x] is the unique component of S containing A. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 47 — #47

i i

Preliminaries 47

Exercise 1.4.36 Let E ⊆ C. Then prove the following statements:

1. If E is written as a disjoint union of two non-empty open subsets of C,

then E is not connected.

2. If E is written as a disjoint union of two non-empty closed subsets of

C, then E is not connected.

Definition 1.4.37 A subset K of C is called a compact set if for a given

collection {Eα : α ∈ I} of open subsets of C such that K ⊂ ∪ Eα , then
α∈I
n
there exists α1 , α2 , . . . , αn ∈ I such that K ⊂ ∪ Eαj .
j=1

THEOREM 1.4.38 Let K be a compact subset of C. Then K is closed and

bounded.

|x − y|
Proof: Let x  K . Then, for every y ∈ K , we have y  x. If ry = ,
2
then ry > 0 and B(x, ry ) ∩ B(y, ry ) = ∅, ∀y ∈ K . As {B(y, ry ) : y ∈ K}
is a collection of open sets satisfying K ⊂ ∪ B(y, ry ). Hence, there exist
y∈K
m
y1 , y2 , . . . , ym ∈ K such that K ⊂ ∪ B(yk , ryk ). If
k=1

r = min{ryk : 1 ≤ k ≤ m}

then
m m
B(x, r) ∩ K ⊆ B(x, r) ∩ ∪ B(yk , ryk ) ⊆ ∪ B(x, ryk ) ∩ B(yk , ryk ) = ∅
k=1 k=1

and hence we get B(x, r) ⊆ K c . Thus, K c is an open set, and hence, K is

closed.
Consider the collection of open sets {B(x, 1) : x ∈ K} with K ⊂
∪ B(x, 1). By definition, there exist x1 , x2 , . . . , xn ∈ K such that K ⊂
x∈K
n
∪ B(xj , 1). If M = max{|xj | : j ∈ {1, 2, . . . , n}}, then we claim that
j=1
|x| ≤ M + 1, ∀x ∈ K . If x ∈ K , then x ∈ B(xj , 1) for some 1 ≤ j ≤ n.
Therefore,
|x| ≤ |x − xj | + |xj | ≤ 1 + M.
Hence, K is bounded. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 48 — #48

i i

48 Some Topological Properties of the Complex Plane

THEOREM 1.4.39 Every infinite subset of a compact set K has a limit point
in K .

Proof: Let S be an infinite subset of K . If S has no limit point in K , then

for each z ∈ K , there exists rz > 0 such that B(z, rz ) ∩ S ⊆ {z}. Obviously,
{B(z, rz ) : z ∈ K} is a collection of open sets such that ∪ B(z, rz ) = K . Since
z∈K
n
K is compact, there exists z1 , z2 , . . . , zn ∈ K such that ∪ B(zn , rzn ) = K , and
j=1
n n
hence, S = S ∩ K = S ∩ ∪ B(zn , rzn ) = ∪ S ∩ B(zn , rzn ) ⊆ {z1 , z2 , . . . , zn },
j=1 j=1
which is a contradiction to the assumption on S . 

THEOREM 1.4.40 Closed subset of a compact set is compact.

Proof: Let K be a compact subset of C and F ⊂ K be closed. If {Eα : α ∈ I}

is a collection of open sets such that F ⊆ ∪ Eα , then
α∈I

K ⊆ C = F ∪ F c ⊆ ∪ Eα ∪ F c .
α∈I

As F is closed, {Eα , F c : α ∈ I} is a collection of open sets; therefore, using

the compactness of K , we get α1 , α2 , . . . , αn ∈ I such that K ⊆ Eα1 ∪ Eα2 ∪
· · · ∪ Eαn ∪ F c . Thus, F ⊆ Eα1 ∪ Eα2 ∪ · · · ∪ Eαn , and hence, F is compact. 

THEOREM 1.4.41 Let A be a closed subset of C and B be a compact subset

of C. If A ∩ B = ∅, then inf{|z − w| : z ∈ A, w ∈ B} > 0.

Proof: Since A ∩ B = ∅, w is not a limit point of A, ∀w ∈ B. There-

fore,  exists rw > 0 such that B(w, rw ) ∩ A = ∅, ∀w ∈ B. As
 there 
B w, r2w : w ∈ B is a collection of open sets such that B ⊆ ∪ B w, r2w ,
w∈B
n rw
there exist wk ∈ B, k = 1, 2, . . . , n such that B ⊆ ∪ B wk , 2k . Let
r  k=1
μ = min 2 : k = 1, 2, . . . , n , then μ > 0. We claim that inf{|z − w| :
wk

z ∈ A, w ∈ B} ≥ μ. Otherwise, there exist z ∈ A and w ∈ B such that

rw
|z − w| < μ. Using w ∈ B, we find j ∈ {1, 2, . . . , n} such that w ∈ B wj , 2j .
rwj
Then we have |w − wj | < 2 . Now,

rwj
|z − wj | ≤ |z − w| + |w − wj | < μ + ≤ rwj
2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 49 — #49

i i

Preliminaries 49
   
which implies that z ∈ B wj , rwj . This is a contradiction to B wj , rwj ∩ A =
∅. Hence, the theorem follows. 

THEOREM 1.4.42 (Cantor’s intersection theorem)

If (Kn ) is a sequence of non-empty compact subsets of C such that Kn ⊇
∞
Kn+1 , ∀n ∈ N, and D(Kn ) → 0 as n → ∞, then ∩ Kn is a singleton set,
n=1
where D(Kn ) = sup{|z − w| : z, w ∈ Kn }, ∀n ∈ N.

∞ ∞
Proof: First, we show that ∩ Kn has at most one point. If x, y ∈ ∩ Kn , then
n=1 n=1
x, y ∈ Kn , ∀n ∈ N. Then, 0 ≤ |x − y| ≤ D(Kn ) → 0 as n → ∞, and hence,
x = y.  c
∞ ∞ ∞
Suppose that ∩ Kn = ∅. Then K1 ∩ ∩ Kn = ∅ ⇒ K1 ⊆ ∩ Kn =
n=1 n=2 n=2
∞
∪ Knc .
As each Kn is a closed set (Theorem 1.4.38), is open, ∀n ≥ 2.
Knc
n=2    c
m m
Hence, by the compactness of K1 , we have K1 ⊆ ∪ Knj = ∩ Knj for
c
j=1
  j=1
m
some 2 ≤ n1 < n2 < · · · < nm . Thus, Knm = K1 ∩ ∩ Knj = ∅, which is
j=1
a contradiction. Hence, the theorem follows. 

THEOREM 1.4.43 (Heine-Borel theorem)

Every closed and bounded interval in R is compact.

Proof: Let I = [a, b] be the given interval, where a, b ∈ R with a < b.

Suppose there exists a collection C of open sets in R such that [a, b] ⊆ ∪ E
E∈C
and I is not contained in the union of any finite sub-collection of C . Let
c1 = a+b 2 be the mid-point of a and b. Then at least one of the two intervals
[a, c1 ] and [c1 , b] is not contained in the union of any finite sub-collection
of C . Choose such an interval and denote it by I1 = [a1 , b1 ]. If D(I) is the
diameter of I , then D(I) = b − a and D(I1 ) = b1 − a1 = b−a 2 . Proceeding like
this, we can find a sequence {In } of intervals with the following properties:

1. In+1 ⊆ In , ∀n ∈ N.

2. In is not contained in the union of any finite sub-collection of C .

3. D(In+1 ) = 12 D(In ) = 1
2n D(I), ∀n ∈ N.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 50 — #50

i i

50 Some Topological Properties of the Complex Plane

If In = [an , bn ], ∀n ∈ N, then we have

an ≤ am+n ≤ bm+n ≤ bn , ∀m, n ∈ N.

Hence, it follows that for each m ∈ N, bm is an upper bound of {an : n ∈ N}.

Using the fact that R has least upper bound property, there exists α ∈ R such
that α = sup{an : n ∈ N}. Then immediately, we have an ≤ α ≤ bm , ∀m, n ∈
∞
N. Thus, α ∈ ∩ In . As α ∈ I , there exists E ∈ C such that α ∈ E. Since E
n=1  
is open, there exists r > 0 such that B(α, r) ⊆ E. As 21n D(I) converges to
0, there exists m ∈ N such that 21m D(I) < r. We now claim that Im ⊆ E. If
x ∈ Im , then |x − α| ≤ 21m D(I) < r, because x, α ∈ Im . Thus, x ∈ B(α, r) ⊆ E.
Hence, our claim follows. However, this is contradiction to a property (2) of
Im . Hence, [a, b] is compact. 

COROLLARY 1.4.44 Every closed and bounded subset of R is compact.

Proof: Let K be a closed and bounded subset of R. Since K is bounded,

there exists M > 0 such that |x| ≤ M, ∀x ∈ K . Therefore, K ⊆ [−M, M].
By previous theorem, [−M, M] is compact. As K is a closed subset of the
compact set [−M, M], K is compact. 

Exercise 1.4.45 Every closed and bounded subset of C is compact.

Hint: First prove that [a, b] × [c, d] is compact as in the proof of Heine–
Borel theorem. Next find positive real numbers M1 and M2 such that K ⊂
[−M1 , M1 ] × [−M2 , M2 ] and complete the remaining proof as in the proof of
previous corollary.

THEOREM 1.4.46 Every Cauchy sequence in C is a convergent sequence.

Proof: Let (zn ) be a Cauchy sequence in C. If zn = xn + iyn , ∀n ∈ N, then

|xn − xm | ≤ |zn − zm | and |yn − ym | ≤ |zn − zm |, ∀n ∈ N.

Therefore, (xn ) and (yn ) are Cauchy sequences of real numbers. Then by
Result 1.3.3, (xn ) and (yn ) are bounded sequences. Therefore, we can find
M > 0 such that xn ∈ [−M, M], ∀n ∈ N. If (xn ) has a subsequence as a
constant sequence, then obviously the subsequence is convergent (Example
1.3.5), and by Lemma 1.3.8, (xn ) itself is convergent. If (xn ) has no such
subsequence, then {xn : n ∈ N} is an infinite set of [−M, M], which is a

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 51 — #51

i i

Preliminaries 51

compact set by Heine–Borel theorem. Now applying Theorem 1.4.39, we get

{xn : n ∈ N} has a limit point x in [−M, M]. Hence, for each k ∈ N, there
exists xnk ∈ B(x, 1/k), with nk > nj whenever k > j. Therefore, (xn,k ) is
a convergent subsequence of (xn ). Again by using Theorem 1.4.39, we get
(xn ) is a convergent sequence. By the same argument, we can prove that (yn )
is a convergent sequence. Thus, by Theorem 1.3.9 (2), we get that (zn ) is a
convergent sequence. 
The above theorem can be rephrased as that C is a complete metric space.

THEOREM 1.4.47 (Bolzano–Weierstrass property) Every bounded infinite

subset of C has a limit point.
Proof of this theorem is a consequence of Exercise 1.4.45 and Theorem
1.4.39.

THEOREM 1.4.48 The set C of all complex numbers is second countable.

That is, there exists a countable collection B of open subsets of C such that
for every open subset G of C and z ∈ G, there exists B ∈ B such that
z ∈ B ⊆ G.

Proof: We know that the set of all rational numbers is countable and dense
in R. Therefore, the collection
 
B = B(wj , q) : wj ∈ Q × Q, q ∈ Q ∩ (0, ∞)

is also a countable collection of open subsets of C. For z = x + iy ∈ , there

exists a positive real number R such that B(z, R) ⊂ . Using the fact that Q
is dense in R, we can choose a rational number r such that 0 < r < R so that
B(z, r) ⊆ B(z, R) ⊆ G. Again using the denseness of Q in R, we can choose
u, v ∈ Q such that
r r
x<u<x+ and y < v < y + .
4 4
If w = u + iv, then
r r r
|w − z| = |(u + iv) − (x + iy)| ≤ |u − x| + |v − y| < + = .
4 4 2
We also have
r r
ζ ∈ B w, ⇒ |w − ζ | <
2 2
r r
⇒ |z − ζ | ≤ |z − w| + |w − ζ | < +
2 2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 52 — #52

i i

52 Extended Complex Numbers and Stereographic Projection

⇒ ζ ∈ B (z, r) .
   
Therefore, z ∈ B w, 2r ⊆ B (z, r) ⊆ G. We note that B w, 2r ∈ B . Hence,
the theorem follows. 

1.5 EXTENDED COMPLEX NUMBERS AND

STEREOGRAPHIC PROJECTION
The set C of all complex numbers along with a symbol ∞, denoted by C∞ ,
is called the set of all extended complex numbers (or the extended complex
plane) if the symbol is assigned with the following properties:

1. a ± ∞ = ∞ ± a = ∞, for all a ∈ C,

2. c · ∞ = ∞ · c = ∞, for all c ∈ C \ {0},

3. ∞ = ∞,
a
4. = ∞, for all a ∈ C∞ \ {0},
0
a
5. = 0, ∀a ∈ C.
∞

∞
∞ − ∞, and 0 · ∞ are not defined.
∞

Definition 1.5.1 (Topology on C∞ )

A set A ⊆ C∞ is said to be an open subset of C∞ if

1. A = C∞ \ K for some compact subset K of C or

2. A is an open subset of C.

THEOREM 1.5.2 (Stereographic projection)

There exists a bijection φ from the unit ball U in R3 onto C∞ .

Proof: Given a point (x1 , x2 , x3 ) ∈ U other than the north pole (0, 0, 1), we
define the image of φ((x1 , x2 , x3 ))) by the intersection of the line passing
through (x1 , x2 , x3 ) and (0, 0, 1) with the complex plane C.

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 53 — #53

i i

Preliminaries 53

To find φ((x1 , x2 , x3 ))) explicitly, first we consider the equation of the line
passing through (x1 , x2 , x3 ) and (0, 0, 1), which is given by
(1 − t)(0, 0, 1) + t(x1 , x2 , x3 ) = (tx1 , tx2 , (1 − t) + tx3 ), t ∈ R.
This line meets the x1 x2 -plane if 1 − t + tx3 = 0 ⇒ t = 1−x1
3
. Therefore,
⎧ 
⎨ x1 x2
, ,0 if x3  1
φ((x1 , x2 , x3 )) = 1 − x3 1 − x3 .
⎩
∞ if x3 = 1

(0, 0, 1)

(x1, x2, x3)

(0, 0, 0)
z

To prove φ : U → C∞ is onto let z ∈ C. If (x1 , x2 , x3 ) ∈ U such that

z = φ(x1 , x2 , x3 ), then using x21 + x22 + x23 = 1, we get
x21 + x22 1 − x23 1 + x3
|z|2 = = =
(1 − x3 )2 (1 − x3 )2 1 − x3
⇒ (1 − x3 )|z|2 = (1 + x3 )
⇒ x3 (1 + |z|2 ) = |z|2 − 1
|z|2 − 1
⇒ x3 = 2 .
|z| + 1
 
x1 z+z z+z |z|2 − 1 z+z
Next, = Re z = ⇒ x1 = 1− 2 = 2 .
1 − x3 2 2 |z| + 1 |z| + 1
Similarly, we obtain that
 
x2 z−z z−z |z|2 − 1 z−z
= Im z = ⇒ x2 = 1− 2 = .
1 − x3 i2 i2 |z| + 1 i(|z|2 + 1)
Next we show, that (x1 , x2 , x3 ) obtained as above belongs to U and it satisfies
φ(x1 , x2 , x3 ) = z. Now,
(z + z)2 + (z − z)2 + (|z|2 − 1)2
x21 + x22 + x23 =
(|z|2 + 1)2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 54 — #54

i i

54 Extended Complex Numbers and Stereographic Projection

2(z2 + z2 ) + |z|4 + 1 − 2|z|2

=
(|z|2 + 1)2
|z| + 1 + 2|z|2
4
=
(|z|2 + 1)2
= 1.

and hence, (x1 , x2 , x3 ) ∈ U . Clearly, we have φ((x1 , x2 , x3 )) = z. Moreover,

we have φ(∞) = (0, 0, 1). Thus, φ : U → C∞ is onto. To prove φ : U →
C∞ is one-to-one, let (x1 , x2 , x3 ), (y1 , y2 , y3 ) ∈ U such that φ((x1 , x2 , x3 )) =
φ((y1 , y2 , y3 )) = z. If z = ∞, then by definition, (x1 , x2 , x3 ) = (0, 0, 1) =
(y1 , y2 , y3 ). Hence, assume that z ∈ C. Then, from the previous argument,
 
z+z z−z |z|2 − 1
(x1 , x2 , x3 ) = , , = (y1 , y2 , y3 ).
|z|2 + 1 i(|z|2 + 1) |z|2 + 1
Thus, φ : U → C∞ is one-to-one. 

RESULT 1.5.3 The image of a circle on the sphere U under the map φ is
either a circle or a straight line in C∞ .

Proof: Any circle on the sphere can be interpreted as the intersection of U

by a plane (say) α1 x1 + α2 x2 + α3 x3 = α0 . Using the values of x1 , x2 , and x3
in terms of z, we get
α2
α1 (z + z) + (z − z) + α3 (|z|2 − 1) = α0 (|z|2 + 1).
i
Putting z = x+iy in the last equation, we have 2α1 x+2α2 y+α3 (x2 +y2 −1) =
α0 (x2 + y2 + 1) ⇒ (α0 − α3 )(x2 + y2 ) − 2α1 x − 2α2 y + α0 + α3 = 0, which is
the equation of a circle if (α0 − α3 )  0 and is the equation of a straight line
if (α0 − α3 ) = 0. 

Definition 1.5.4 Given any two points z, w ∈ C∞ , define the distance D(z, w)
by the actual distance between φ −1 (z) and φ −1 (w) in R3 .

PROBLEM 1.5.5 Find D(z, w) explicitly.

Proof:
Case 1: z, w ∈ C.
Let φ −1 (z) = (x1 , x2 , x3 ) and φ −1 (w) = (y1 , y2 , y3 ). Then,

(x1 − y1 )2 + (x2 − y2 )2 + (x3 − y3 )2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 55 — #55

i i

Preliminaries 55

= x21 + y21 − 2x1 y1 + x22 + y22 − 2x2 y2 + x23 + y23 − 2x3 y3

= (x21 + x22 + x23 ) + (y21 + y22 + y23 ) − 2(x1 y1 + x2 y2 + x3 y3 )
= 2 − 2(x1 y1 + x2 y2 + x3 y3 )
(as x21 + x22 + x23 = 1 = y21 + y22 + y23 )
(z + z)(w + w) − (z − z)(w − w) + (|z|2 − 1)(|w|2 − 1)
=2−2
(|z|2 + 1)(|w|2 + 1)
 
2(zw + zw) + (|z|2 |w|2 − |z|2 − |w|2 + 1)
=2−2
(|z|2 + 1)(|w|2 + 1)
 
2(−|z| − |w|2 + zw + zw) + (|z|2 |w|2 + |z|2 + |w|2 + 1)
2
=2−2
(|z|2 + 1)(|w|2 + 1)
 
−2|z − w|2 + (|z|2 + 1)(|w|2 + 1)
=2−2
(|z|2 + 1)(|w|2 + 1)
4|z − w|2
= .
(|z|2 + 1)(|w|2 + 1)
Case 2: z ∈ C and w = ∞.
In this case, φ −1 (w) = (0, 0, 1) and let φ −1 (z) = (x1 , x2 , x3 ). Therefore,
(x1 − 0)2 + (x2 − 0)2 + (x3 − 1)2 = 2 − 2x3
|z|2 − 1
= 2−2 2
|z| + 1
4
=
|z|2 + 1
⎧
⎪ 2|z − w|
⎪
⎨ if z, w ∈ C
Hence, D(z, w) = |(z|2 + 1)(|w|2 + 1) 
⎪
⎪
2
if z ∈ C and w = ∞.
⎩
|z|2 + 1

1.6 LIMIT AND CONTINUITY

Definition 1.6.1 Let  be a region, z0 ∈ , l ∈ C and f :  → C. We say
that f (z) → l as z → z0 if given > 0, then there exists δ > 0 such that
|f (z) − l| < whenever 0 < |z − z0 | < δ .
In this case, we write lim f (z) = l.
z→z0

THEOREM 1.6.2 Let f :  → C, z0 ∈ , l, c ∈ C and lim f (z) = l. Then,

z→z0

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 56 — #56

i i

56 Limit and Continuity

1. lim Re (f (z)) = Re l,
z→z0

2. lim Im (f (z)) = Im l,
z→z0

3. lim f (z) = l,
z→z0

4. lim |f (z)| = |l|,

z→z0

5. lim (f + c)(z) = l + c,
z→z0

6. lim (cf )(z) = cl.

z→z0

Proof: Let > 0. Then, there exists δ > 0 such that |f (z) − l| < whenever
0 < |z − z0 | < δ . We choose the same δ for proving (1), (2), (3), (4), (5). If
0 < |z − z0 | < δ , then

1. |Re (f (z)) − Re l| = |Re (f (z) − l)| ≤ |f (z) − l| <

2. |Im (f (z)) − Im l| = |Im (f (z) − l)| ≤ |f (z) − l| <

3. |f (z) − l| = |f (z) − l| ≤ |f (z) − l| <

4. | |f (z)| − |l| | ≤ |f (z) − l| < , by using Corollary 1.2.10

5. |(f (z) + c) − (l + c)| = |f (z) − l| <

6. For proving the last assertion, we first choose δ > 0 such that

|f (z) − l| < whenever 0 < |z − z0 | < δ

|c| + 1
|c|
If 0 < |z − z0 | < δ , then |(cf (z)) − (cl)| = |c||f (z) − l| ≤ < .
|c| + 1


THEOREM 1.6.3 Let f and g be functions on , z0 ∈  and lim f (z) = l1 ,

z→z0
lim g(z) = l2 , where l1 , l2 ∈ C. Then,
z→z0

1. lim (f + g)(z) = l1 + l2 ,
z→z0

2. lim (fg)(z) = l1 l2 ,
z→z0

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 57 — #57

i i

Preliminaries 57
 
f l1
3. lim (z) = , provided l2  0.
z→z0 g l2

Proof: Let > 0 be given.

1. For a given > 0, there exist δ1 > 0 and δ2 > 0 such that

|f (z) − l1 | < whenever 0 < |z − z0 | < δ1

2
and
|g(z) − l2 | <whenever 0 < |z − z0 | < δ2 .
2
Let 0 < δ < min{δ1 , δ2 }. If 0 < |z − z0 | < δ , then 0 < |z − z0 | < δ1
and 0 < |z − z0 | < δ2 . Thus, we get

|(f + g)(z) − (l1 + l2 )| ≤ |f (z) − l1 | + |g(z) − l2 | < + =

2 2

2. Consider

|(fg)(z) − (l1 l2 )| ≤ |f (z)g(z) − f (z)l2 )| + |f (z)l2 − l1 l2 |

= |f (z)||g(z) − l2 | + |l2 ||f (z) − l1 |. (1.5)

Given > 0 choose δ1 > 0 such that

|f (z) − l1 | < 1 if 0 < |z − z0 | < δ1 .

Hence, |f (z)| < |l1 | + 1, whenever 0 < |z − z0 | < δ1 . For the same ,
find δ2 > 0 such that

|g(z) − l2 | < whenever 0 < |z − z0 | < δ2 .

2(|l1 | + 1)
Next choose δ3 > 0 such that

|f (z) − l1 | < whenever 0 < |z − z0 | < δ3 .

2(|l2 | + 1)
If 0 < δ < min{δ1 , δ2 , δ3 }, then from (1.5), we get

|(fg)(z) − (l1 l2 )| ≤ |f (z)||g(z) − l2 | + |l2 ||f (z) − l1 |

≤ (|l1 | + 1) + |l2 |
2(|l1 | + 1) 2(|l2 | + 1)
< + = .
2 2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 58 — #58

i i

58 Limit and Continuity

1 1
3. First we prove that lim = if l2  0. Given > 0, choose δ > 0
z→z0 g(z) l2
such that
 
|l2 | |l2 |2
0 < |z − z0 | < δ ⇒ |g(z) − l2 | < min , .
2 2
Thus, if 0 < |z − z0 | < δ , then
|l2 | |l2 | 1 2
| |l2 | − |g(z)| | ≤ |g(z) − l2 | < ⇒ |g(z)| > ⇒ < .
2 2 |g(z)| |l2 |
We also have
 
 1 1  |g(z) − l2 | |l2 |2 2

 g(z) − l  = |g(z)l | < 2 · |l |2 = .
2 2 2

Hence, our claim follows. Next by using (2) of this theorem, we get
f (z) 1 1 l1
lim = lim f (z) = lim f (z) lim = . 
z→z0 g(z) z→z0 g(z) z→z0 z→z0 g(z) l2

Definition 1.6.4 Let f : C → C. We say that f (z) → ∞ as z → ∞ if given

M > 0, then there exists K > 0 such that |f (z)| > M whenever |z| > K . In
this case, we write lim f (z) = ∞.
z→∞

Definition 1.6.5 Let f : C → C and a ∈ C. We say that f (z) → ∞ as z → a

if given M > 0, then there exists δ > 0 such that 0 < |z − a| < δ implies that
|f (z)| > M . We write this by lim f (z) = ∞.
z→a

Similarly, we define lim f (z) = a.

z→∞

THEOREM 1.6.6 Let f :  → C, g :  → C, A ∈ C and a ∈ . If

lim f (z) = ∞ and lim g(z) = A, then
z→a z→a

1. lim (cf )(z) = ∞ for every non-zero complex number c,

z→a

2. lim (f + g)(z) = ∞,
z→a

3. lim (f · g)(z) = ∞ if A  0,
z→a

f (z)
4. lim = ∞,
z→a g(z)

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 59 — #59

i i

Preliminaries 59

g(z)
5. lim = 0.
z→a f (z)

Proof: Let M > 0 be given.

1. We choose δ > 0 such that

M
0 < |z − a| < δ ⇒ |f (z)| > .
|c|

M
If 0 < |z − a| < δ , then |cf (z)| = |c||f (z)| > |c| = M.
|c|
2. We choose δ1 > 0 and δ2 > 0 such that

0 < |z − a| < δ2 ⇒ |f (z)| > M + 1 + |A|

and

0 < |z−a| < δ2 ⇒ |g(z)−A| < 1 ⇒ |g(z)| ≤ |g(z)−A|+|A| < 1+|A|.

If 0 < |z − a| < min{δ1 , δ2 }, then

|f (z) + g(z)| ≥ |f (z)| − |g(z)| > M + 1 + |A| − 1 − |A| = M.

Hence, lim (f + g)(z) = ∞.

z→a

3. We choose δ1 > 0 and δ2 > 0 such that

2M
0 < |z − a| < δ2 ⇒ |f (z)| >
|A|

and
|A| |A|
0 < |z − a| < δ2 ⇒ |g(z) − A| < ⇒ |g(z)| ≥ .
2 2
If 0 < |z − a| < min{δ1 , δ2 }, then

2M |A|
|f (z)g(z)| > = M.
|A| 2

Hence, lim f (z)g(z) = ∞.

z→a

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 60 — #60

i i

60 Limit and Continuity

4. We choose δ1 > 0 and δ2 > 0 such that

0 < |z − a| < δ2 ⇒ |f (z)| > M(1 + |A|)
and
0 < |z−a| < δ2 ⇒ |g(z)−A| < 1 ⇒ |g(z)| ≤ |g(z)−A|+|A| < 1+|A|.
If 0 < |z − a| < min{δ1 , δ2 }, then
 
 f (z)  M(1 + |A|)
 
 g(z)  > 1 + |A| = M.

f (z)
Hence, lim = ∞.
z→a g(z)
5. Given > 0 we choose δ1 > 0 and δ2 > 0 such that
1 + |A|
0 < |z − a| < δ1 ⇒ |f (z)| >

and
0 < |z − a| < δ2 ⇒ |g(z) − A| < 1 ⇒ |g(z)| ≤ 1 + |A|.
If 0 < |z − a| < min{δ1 , δ2 }, then
 
 g(z) 
 
 f (z)  < 1 + |A| (1 + |A|) = .

g(z)
Hence, lim = 0. 
z→a f (z)

Example 1.6.7 Let f : C → C be defined by

⎧
⎨ xy(x + iy) z = x + iy  0
f (z) = x3 + y3 , ∀z ∈ C.
⎩
0 z=0
Prove that lim f (z) does not exist.
z→0
If we allow z → 0 along the line y = cx, then we get
x(cx)(x + i(cx)) c(1 + ic) c(1 + ic)
lim f (z) = lim = lim = .
z→0 x→0 x3 + (cx)3 x→0 1 + c3 1 + c3
which is depending on c. That is, if we let z → 0 along different lines in
the family y = cx of straight lines, then f (z) approaches different values, and
hence, lim f (z) does not exist.
z→0

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 61 — #61

i i

Preliminaries 61

LEMMA 1.6.8 Let f :  → C and let (a, b) ∈ , where  ⊆ C is open.

If lim f (x, y) = A for some A ∈ C, then lim lim f (x, y) = A =
(x,y)→(a,b) x→a y→b
lim lim f (x, y).
y→b x→a

Proof: First we choose r > 0 such that B((a, b), r) ⊆ . For the given > 0,
there exists 0 < δ < r such that

0 < |(x, y) − (a, b)| < δ ⇒ |f (x, y) − A| < ,

2
which implies that

|f (x, y) − f (u, v)| ≤ |f (x, y) − A| + |A − f (u, v)| < + = . (1.6)

2 2
whenever (x, y), (u, v) ∈ B((a, b), δ) \ {(a, b)}. Now for a fixed x ∈ R such that
|x − a| < δ ,

if 0 < |y − b| < δ, then |(x, y) − (x, b)| = |y − b| < δ

and hence, |f (x, y) − f (x, b)| < . In other words, we have proved that
lim f (x, y) = f (x, b) for every x ∈ R with |x − a| < δ . Using the definition of
y→b
lim f (x, b) = A , we get
(x,y)→(a,b)

0 < |x − a| < δ ⇒ |(x, b) − (a, b)| = |x − a| < δ ⇒ |f (x, b) − A| < .

Thus, lim lim f (x, y) = lim f (x, b) = A. Similarly, we can prove that
x→a y→b x→a
lim lim f (x, y) = A. 
y→b x→a

Exercise 1.6.9 Let f :  → C, a ∈ , and A ∈ C. Prove that lim f (z) = A

z→a
iff f (zn ) → A as n → ∞ whenever zn → a as n → ∞ in  with zn  a,
∀n ∈ .
Now we discuss left limit and right limit of a real-valued function on an
interval I of R.

Definition 1.6.10 Let f : I → R and x ∈ I , where I is an interval of R. We

write that
1. lim f (y) = l exists for some l ∈ R if given > 0, then there exists
y→x−
δ > 0 such that (x − δ, x) ⊂ I and x − δ < y < x ⇒ |f (y) − l| < .

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 62 — #62

i i

62 Limit and Continuity

2. lim f (y) = l exists for some l ∈ R if given > 0, then there exists
y→x+
δ > 0 such that (x, x + δ) ⊂ I and x < y < x + δ ⇒ |f (y) − l| < .
3. lim f (y) = +∞ exists if given M > 0, then there exists δ > 0 such
y→x−
that (x − δ, x) ⊂ I and x − δ < y < x ⇒ f (y) > M .
4. lim f (y) = +∞ exists if given K > 0, then there exists δ > 0 such
y→x+
that (x, x + δ) ⊂ I and x < y < x + δ ⇒ f (y) > K .
Similarly, we can define lim f (y) = −∞ and lim f (y) = −∞.
y→x− y→x+

Exercise 1.6.11 Let f :  → C, g :  → C, and a ∈ . If lim f (z) = ∞

z→a
and g is bounded in a neighbourhood of a, then prove that lim (fg)(z) = ∞.
z→a

Remark 1.6.12: Let f :  → C, g :  → C, and a ∈ . If lim f (z) = ∞

z→a
and lim g(z) = ∞, then lim (f + g)(z) need not be ∞.
z→a z→a ⎧
⎨1
z0
For example, for a fixed c ∈ C, if f (z) = z and g(z) =
⎩1 z = 0
⎧
⎨ 1
− +c z0
z , then clearly, lim f (z) = ∞ and lim g(z) = ∞ but
⎩0 z=0 z→0 z→0

lim (f + g)(z) = c  ∞.
z→0

Definition 1.6.13 Let f :  → C.

1. f is said to be continuous at z0 ∈  if lim f (z) = f (z0 ).
z→z0

2. f is said to be continuous on  if f is continuous at every point z0 ∈ .

f
THEOREM 1.6.14 If f and g are continuous at z0 ∈ , then f + g, fg, and
g
are continuous at z0 .

Proof: Proof of this theorem follows from Theorem 1.6.3. 

Example 1.6.15 Every constant function on C is continuous.

Let z0 ∈ C be arbitrary. Given > 0, we choose δ > 0 arbitrarily. If
|z − z0 | < δ , then |f (z) − f (z0 )| = 0 < . Hence, f is continuous at z0 .

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 63 — #63

i i

Preliminaries 63

Example 1.6.16 If f (z) = z, ∀z ∈ C, then f is continuous on C.

For a given > 0, we choose δ = . Then obviously, we get |z − z0 | <
when |z − z0 | < δ .

Example 1.6.17 Every polynomial is continuous on C.

Using Theorem 1.6.14 and the above two examples, we get every polynomial
is a continuous function on C.

Example 1.6.18 Re z, Im z, z, and |z| are continuous on C.

For justification, see Example 1.6.16 and Theorem 1.6.2.

LEMMA 1.6.19 f is continuous iff Re f and Im f are continuous iff f is

continuous.

Proof: Let f be continuous at z0 . For a given > 0, there exists δ > 0 such
that |f (z) − f (z0 )| < , whenever |z − z0 | < δ . For the same δ ,

if |z − z0 | < δ, then |Re f (z) − Re f (z0 )| ≤ |f (z) − f (z0 )| < .

Similarly, for the same δ ,

if |z − z0 | < δ, then |Im f (z) − Im f (z0 )| ≤ |f (z) − f (z0 )| < .

Conversely, assume that Re f and Im f are continuous at z0 . Then, for a given

> 0, there exist δ1 > 0 and δ2 > 0 such that |Re f (z) − Re f (z0 )| < √ ,
2
whenever |z − z0 | < δ1 and |Im f (z) − Im f (z0 )| < √ , whenever |z − z0 | <
2
δ2 . If δ = min{δ1 , δ2 }, then δ > 0, and if |z − z0 | < δ , then

|f (z) − f (z0 )| = |Re f (z) − Re f (z0 )|2 + |Im f (z) − Im f (z0 )|

2 2
< + = .
2 2
Thus, f is continuous at z0 .
Using |z| = |z|, one can prove that f is continuous iff f is continuous. 

THEOREM 1.6.20 Composition of two continuous functions is continuous.

Proof: Let f :  → C and g : f () → C be continuous. If h = g ◦ f , then

we show that h is continuous on . Let z0 ∈  be arbitrary. Then f (z0 ) ∈

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 64 — #64

i i

64 Limit and Continuity

f (). Using the continuity of g at f (z0 ) for a given > 0, choose r > 0
such that
|g(w) − g(f (z0 ))| < if |w − f (z0 )| < r.
Next, using the continuity of f at z0 , for this r > 0, there exists δ > 0 such
that
|f (z) − f (z0 )| < r if |z − z0 | < δ.
Therefore, combining these two statements, we get
if |z − z0 | < δ, then |h(z) − h(z0 )| = |g(f (z)) − g(f (z0 ))| < .
Hence, h is continuous at z0 . 

RESULT 1.6.21 Let f : A → C and x ∈ A, then f is continuous at x iff

f (xn ) → f (x) as n → ∞, whenever xn → x as n → ∞.

Proof: Assume that f is continuous at x. For a given > 0, we can find δ > 0
such that
|y − x| < δ ⇒ |f (y) − f (x)| < .
If xn → x0 as n → ∞, for the δ > 0, there exists N ∈ N such that
|xn − x| < δ, ∀n ≥ N.
Therefore, if n ≥ N , then
|xn − x| < δ ⇒ |f (xn ) − f (x)| < .
Conversely, assume that f is not continuous at x. Then, there exists >
0 such that for every n ∈ N, there exists xn ∈ A with |xn − x| < 1n and
|f (x) − f (xn )| ≥ . Hence, there exists a sequence (xn ) such that xn → x as
n → ∞ but f (xn ) → f (x) as n → ∞. Hence, the result follows. 

Definition 1.6.22 Let m, n ∈ N, E ⊆ Rn , (x1 , x2 , . . . , xn ) ∈ E, and f =

(f1 , f2 , . . . , fm ) : E → Rm be continuous at (x1 , x2 , . . . , xn ) if given > 0,
then there exists δ > 0 such that

 m

 |fk (t1 , t2 , . . . , tn ) − fk (x1 , x2 , . . . , xn )|2 <
k=1

whenever 
 n

(t1 , t2 , . . . , tn ) ∈ Rn and  |tj − xj |2 < δ.
j=1

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 65 — #65

i i

Preliminaries 65

Limit of a function f = (f1 , f2 , . . . , fm ) : E → Rm at a point

(x1 , x2 , . . . , xn ) can also be defined in an analogous way. As in the earlier dis-
cussions, one can prove that f = (f1 , f2 , . . . , fm ) : E → Rm is continuous at
(x1 , x2 , . . . , xn ) iff fk (x1 (ν), x2 (ν), . . . , xn (ν)) → fk (x1 , x2 , . . . , xn ) as ν → ∞,
∀k = 1, 2, . . . , m, whenever xj (ν) → xj as ν → ∞, ∀j = 1, 2, . . . , n.

THEOREM 1.6.23 Let φ : U → C∞ be the stereographic projection. Then

• φ is a continuous map on U \ {(0, 0, 1)} and

φ(x1 , x2 , x3 ) → ∞ as (x1 , x2 , x3 ) → (0, 0, 1),

• φ −1 is a continuous map on C and

φ −1 (z) → (0, 0, 1) as z → ∞.

x1 x2
Proof: If φ1 (x1 , x2 , x3 ) = and φ2 (x1 , x2 , x3 ) = ∀(x1 , x2 , x3 ) ∈
1 − x3 1 − x3
U \ {(0, 0, 1)}, then φ = (φ1 , φ2 ). (Theorem 1.5.2.)

1. (a) As x3  1, ∀(x1 , x2 , x3 ) ∈ U \ {(0, 0, 1)}, whenever xj (ν) → xj as

ν → ∞ for all j = 1, 2, 3, we have

x1 (ν) x1
→ as ν → ∞.
1 − x3 (ν) 1 − x3

and
x2 (ν) x2
→ as ν → ∞.
1 − x3 (ν) 1 − x3
Therefore, φ is continuous on U \ {(0, 0, 1)}.
(b) If xj (ν) → 0 for j = 1, 2 and x3 (ν) → 1 as ν → ∞, then

φk (x1 (ν), x2 (ν), x3 (ν)) → ±∞ as ν → ∞ for k = 1, 2

in the extended real number system, and hence,

φ(x1 (ν), x2 (ν), x3 (ν)) → ∞ in C∞ as ν → ∞.

2. As the remaining part of the proof of this theorem is similar, we leave

it as an exercise to the reader.

Therefore, φ and φ −1 are continuous functions. 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 66 — #66

i i

66 Limit and Continuity

THEOREM 1.6.24 Let f : A → C, where A is open in C. Then, f is

continuous on A iff f −1 (E) is open in A whenever E is open in C.

Proof: Assume that f is continuous on A. Let x ∈ f −1 (E). Then f (x) ∈ E. As

E is open, there exists > 0 such that B(f (x), ) ⊆ E. Since f is continuous
at x, for this > 0, there exists δ > 0 such that

|z − x| < δ ⇒ |f (z) − f (x)| < .

Therefore, if z ∈ B(x, δ) ⇒ f (z) ∈ B(f (x), ). Hence,

f (B(x, δ)) ⊆ B(f (x), )) ⊆ E ⇒ B(x, δ) ⊆ f −1 (E).

Thus, f −1 (E) is open in A.

Conversely, assume that f −1 (E) is open in A, whenever E is open in C.
Let x ∈ A and > 0 be given. Then B(f (x), ) is an open subset of C. Then by
hypothesis, f −1 (B(f (x), )) is open in A and x ∈ f −1 (B(f (x), )). Then, there
exists δ > 0 such that B(x, δ) ⊆ f −1 (B(f (x), )) ⇒ f (B(x, δ)) ⊆ B(f (x), ).
Thus, we have proved that

|z − x| < δ ⇒ |f (z) − f (x)| <

and hence, f is continuous on A. 

Definition 1.6.25 Let f : E → C, where E ⊂ C. f is said to be uniformly

continuous on E if given > 0, there exits δ > 0 such that x, y ∈ E, |x − y| <
δ ⇒ |f (x) − f (y)| < .

THEOREM 1.6.26 If f : K → C is continuous and K is a compact subset of

C, then f is uniformly continuous.

Proof: Let > 0 be given. For every w ∈ K , there exists δw > 0 such that

|z − w| < δ ⇒ |f (z) − f (w)| < . (1.7)

3
 
Since K is compact and B w, δ2w : w ∈ K is a collection of open sets such
that K ⊂ ∪ B w, δ2w , there exist
w∈K
 
n δwj
w1 , w2 , w3 , . . . , wn ∈ K such that K ⊂ ∪ B wj , .
j=1 2

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 67 — #67

i i

Preliminaries 67
 δw 
If δ = min 2 : j
j = 1, 2, . . . , n , then δ > 0. If x, y ∈ K with |x − y| < δ ,
δwj  
then x ∈ B wj , 2 for some 1 ≤ j ≤ n. We note that y ∈ B wj , δwj for the
same j, because

δwj δwj δwj

|y − wj | ≤ |y − x| + |x − wj | ≤ δ + < + = δwj .
2 2 2
Hence, using (1.7), we get

|f (x) − f (y)| ≤ |f (x) − f (wj )| + |f (wj ) − f (y)| < + < .

3 3
Thus, f is uniformly continuous on K . 

THEOREM 1.6.27 Continuous image of a compact set is compact.

Proof: Let {Eα : α ∈ I} be a collection of open sets such that f (K) ⊆ ∪ Eα .

α
By Theorem 1.6.24, f −1 (Eα ) is open for every α ∈ I . Now, if x ∈ K , then
f (x) ∈ f (K). Then f (x) ∈ Eα for some α . This implies x ∈ f −1 (Eα ) for this α .
Therefore,
K ⊆ ∪ f −1 (Eα ).
α
n
As K is compact, we have K ⊆ ∪ f −1 (Eαi ) for some suitable αi ∈ I , i =
i=1
1, 2, . . . , n. Then
n n
f (K) ⊆ ∪ f (f −1 (Eαi )) ⊆ ∪ Eαi .
i=1 i=1

Therefore, f (K) is compact. 

THEOREM 1.6.28 If f is a real-valued continuous function on a compact

subset K of C, then f is bounded and there exist x, y ∈ K such that
f (x) = sup f (K) and f (y) = inf f (K).

Proof: By previous theorem, f (K) is a compact subset of R, and hence, it

is closed and bounded. Thus, f is a bounded function on K . As sup f (K) ∈
Cl f (K) = f (K) (cf. Exercise 1.4.14), then there exists x ∈ K such that f (x) =
sup f (K). Similarly, we can prove that there exists y ∈ K such that f (y) =
inf f (K). 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 68 — #68

i i

68 Limit and Continuity

THEOREM 1.6.29 Continuous image of a connected set is connected.

Proof: Let C be a connected subset of C. Suppose if f (C) is not connected in

C, then there exist non-empty separated sets A and B in C such that

f (C) ⊆ A ∪ B and f (C) ∩ A  ∅  f (C) ∩ B.

Therefore, clearly we have

C ⊆ f −1 (A) ∪ f −1 (B) and C ∩ f −1 (A)  ∅  C ∩ f −1 (B).

Now we claim that f −1 (A) and f −1 (B) are separated sets in C. Suppose
f −1 (A) ∩ (Clf −1 (B))  ∅, then there exists a point x ∈ f −1 (A) and a sequence
(xn ) in f −1 (B) such that xn → x as n → ∞, indeed choose
 
1
xn ∈ B x, ∩ f −1 (B), ∀n ∈ N.
n
As f is continuous, using Result 1.6.21, we have

f (xn ) → f (x) as n → ∞.

Note that f (x) ∈ A and f (xn ) ∈ B ∀n ∈ N. Therefore, f (x) ∈ Cl B as for each

r > 0,
f (xn ) ∈ B(f (x), r) ∩ B for all but finitely many n,
and hence, A ∩ Cl B  ∅. This contradicts the fact that A and B are separated
sets. By a similar argument, we can show that

(Cl f −1 (A)) ∩ f −1 (B)  ∅.

This implies that C is not connected, which is a contradiction. Therefore, f (C)

is connected. 

THEOREM 1.6.30 (Intermediate value theorem)

If f : [a, b] → R is a continuous function and λ ∈ R lies between f (a) and
f (b), then there exists c ∈ (a, b) such that f (c) = λ.

Proof: As [a, b] is a connected subset of R (as well as connected subset of

C), by previous theorem, f ([a, b]) is a connected subset of R. Hence, f ([a, b])
is an interval by Theorem 1.4.22. Thus, by definition of an interval, if λ lies
between f (a) and f (b), then λ ∈ f ([a, b]). That is, there exists c ∈ [a, b]
such that f (c) = λ. As f (a) < f (c) < f (b), we have a  c  b, and hence,
c ∈ (a, b). 

i i

i i
 i i

“book” — 2014/6/4 — 20:35 — page 69 — #69

i i

Preliminaries 69

THEOREM 1.6.31 Let (fn ) be a sequence of complex-valued continuous func-

tions on E ⊆ C. If fn → f as n → ∞ uniformly on E for some f : E → C,
then f is also continuous on [a, b].

Proof: Let z0 ∈ E and > 0 be arbitrary. By hypothesis, given > 0, there

exists N ∈ N such that

|fn (z) − f (z)| < , ∀z ∈ E, ∀n ≥ N. (1.8)

3
As fN is continuous at z0 , given > 0, there exists δ > 0 such that

|z − z0 | < δ ⇒ |fN (z) − fN (z0 )| < . (1.9)

3
For the same δ > 0, if |z − z0 | < δ , then using (1.8) and (1.9), we get

|f (z)−f (z0 )| ≤ |f (z)−fN (z)|+|fN (z)−fN (z0 )|+|fN (z0 )−f (z0 )| < + + = .
3 3 3
Therefore, f is continuous on E. 

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 20:46 — page 71 — #1

i i

2
Analytic Functions

2.1 DIFFERENTIABILITY
Definition 2.1.1 Let  be a region, z0 ∈  and f :  → C. f is said to be dif-
ferentiable at z0 if lim f (z)−f
z−z0
(z0 )
exists in C or equivalently, lim f (z0 +h)−f
h
(z0 )
z→z0 h→0
exists in C, and this limit is called the derivative of f at z0 , which is denoted
by f  (z0 ).

Definition 2.1.2 A function f is said to be differentiable on a region  if f is

differentiable at every point of .

Definition 2.1.3 Let  be a region, z0 ∈  and f :  → C.

1. f is said to be analytic at z0 if f is differentiable on some neighbourhood
of z0 .
2. f is said to be analytic on  if f is analytic at every point of .

Remark 2.1.4: Suppose f is a function on a region . Then, f is differen-

tiable on  iff f is analytic on .

RESULT 2.1.5 If f is differentiable at z0 , then f is continuous at z0 .

Proof: Using Theorem 1.6.3, we get

( f (z) − f (z0 ))
lim ( f (z) − f (z0 )) = lim · (z − z0 )
z→z0 z→z0 z − z0
( f (z) − f (z0 ))
= lim · lim (z − z0 )
z→z0 z − z0 z→z0

= f (z0 ) · 0 = 0.
Thus, f is continuous at z0 . 

71

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 72 — #2

i i

72 Differentiability

THEOREM 2.1.6 If f and g are differentiable at z0 , then f + g and fg are

differentiable at z0 . If g (z0 )  0, then gf is differentiable at z0 .

Proof: By assumption, we have

f (z) − f (z0 ) g(z) − g(z0 )
lim = f  (z0 ) and lim = g (z0 ).
z→z0 z − z0 z→z0 z − z0
Using Theorems 1.6.2 and 1.6.3 and Result 2.1.5, we get
( f + g)(z) − ( f + g)(z0 )
1. lim
z→z0 z − z0
f (z) − f (z0 ) g(z) − g(z0 )
= lim + lim
z→z0 z − z0 z→z0 z − z0
= f  (z0 ) + g (z0 ).
( fg)(z) − ( fg)(z0 )
2. lim
z→z0 z − z0
f (z)g(z) − f (z0 )g(z) + f (z0 )g(z) − f (z0 )g(z0 )
= lim
z→z0 z − z0
g(z)( f (z) − f (z0 )) f (z0 )(g(z) − g(z0 ))
= lim + lim
z→z0 z − z0 z→z0 z − z0
f (z) − f (z0 ) g(z) − g(z0 )
= lim g(z) lim + f (z0 ) lim
z→z0 z→z0 z − z0 z→z0 z − z0
= f  (z0 )g(z0 ) + f (z0 )g (z0 ).
   
f f
(z) − (z0 )
g g
3. lim
z→z0 z − z0

f (z)g(z0 ) − f (z0 )g(z)

= lim
g(z)g(z0 )(z − z0 )
z→z0
f (z)g(z0 ) − f (z0 )g(z0 ) + f (z0 )g(z0 ) − f (z0 )g(z)
= lim
z→z0 g(z)g(z0 )(z − z0 )
 
g(z0 )( f (z) − f (z0 )) f (z0 )(g(z) − g(z0 ))
lim −
z→z0 z − z0 z − z0
=
lim g(z)g(z0 )
z→z0
g(z0 )f  (z 
0 ) − f (z0 )g (z0 )
= 2
.
(g(z0 )) 

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 73 — #3

i i

Analytic Functions 73

Example 2.1.7 Every constant function is differentiable on C and its deriva-

tive is 0.
Let z0 ∈ C be arbitrary. If f is a constant function, then obviously
lim f (z)−f
z−z0
(z0 )
= lim z−z
0
0
= 0.
z→z0 z→z0

Example 2.1.8 If f (z) = zn , ∀z ∈ C, where n ∈ N, then f is differentiable on

C and f  (z) = nzn−1 , ∀z ∈ C.
For an arbitrary z0 ∈ C, consider

f (z) − f (z0 ) zn − z0n

lim = lim
z→z0 z − z0 z→z0 z − z0
(z − z0 )(zn−1 + zn−2 z0 + · · · + zz0n−2 + z0n−1 )
= lim
z→z0 z − z0
 
= lim z n−1
+ z z0 + · · · + zz0n−2 + z0n−1
n−2
z→z0

= nz0n−1 .

Thus, f  (z) = nzn−1 , ∀z ∈ C.

Example 2.1.9 Every polynomial is differentiable on C.

Applying Theorem 2.1.6 repeatedly and using the above two examples, we
get every polynomial to be differentiable on C.

THEOREM 2.1.10 (Chain rule)

If f is differentiable on , g is differentiable on f (), and h = g ◦ f , then h is
differentiable on  and h (z) = g ( f (z))f  (z), ∀z ∈ .

Proof: Let z0 ∈  be arbitrary, then we have

f (z) − f (z0 ) g(w) − g( f (z0 ))

lim = f  (z0 ) and lim = g ( f (z0 )).
z→z0 z − z0 w→f (z0 ) w − f (z0 )

Thus, if we set

f (z) − f (z0 )
R(z) = − f  (z0 ) in B(z0 , r) \ {z0 }
z − z0

and
g(w) − g( f (z0 ))
S(w) = − g ( f (z0 )) in B( f (z0 ), r) \ {f (z0 )}
w − f (z0 )

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 74 — #4

i i

74 Differentiability

for a small r > 0, then

f (z) − f (z0 ) = (z − z0 )( f  (z0 ) + R(z)),

g(w) − g( f (z0 )) = (w − f (z0 ))[g ( f (z0 )) + S(w)],

and R(z) → 0 as z → z0 , S(w) → 0 as w → f (z0 ). As f is continuous at z0

(Result 2.1.5), we have f (z) → f (z0 ) as z → z0 . Therefore,
h(z) − h(z0 ) g( f (z)) − g( f (z0 ))
lim = lim
z→z0 z − z0 z→z0 z − z0
[g ( f (z0 )) + S( f (z))]( f (z) − f (z0 ))
= lim
z→z0 z − z0
[g ( f (z0 )) + S( f (z))][f  (z0 ) + R(z)](z − z0 )
= lim
z→z0 z − z0
= lim [g ( f (z0 )) + S( f (z))]( f  (z0 ) + R(z))

z→z0
= g ( f (z0 )) f  (z0 ).


Hence, h is differentiable on  and h (z) = g ( f (z))f  (z), ∀z ∈ . 

Definition 2.1.11 (Higher order derivatives)

Let f :  → C, where  be a region.
1. The second derivative of f is defined by the derivative of f  and is
denoted by f (2) or f  .
2. The third derivative of f is defined by the derivative of f (2) and is
denoted by f (3) or f  .
3. Proceeding further inductively, we define the k th derivative of f by the
derivative of f (k−1) and is denoted by f (k) , ∀k > 1.
We shall also use the notation f (0) to denote the function f . We also mean that
f is k times differentiable, by writing that f (k) exists, for k ∈ {1, 2, 3, . . .}.

Remark 2.1.12: In the above definition, the k th derivative of f could be

defined if f (k−1) is differentiable. However, we shall prove that if f is analytic,
then f (k) exists, for all k ∈ N. (See Corollary 4.3.3.)

RESULT 2.1.13 (Leibniz rule)


k
Let f and g be analytic functions, then ( f · g)(k) = kCj f ( j) g(k−j) , for all
j=0

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 75 — #5

i i

Analytic Functions 75

k × (k − 1) × · · · × (k − j + 1)
k ∈ N, where kCj =
1 × 2 × ··· × j

Proof: We prove this theorem by induction on k . For k = 1, the result follows

from Theorem 2.1.6 immediately. Assume that this result holds for some k ∈
N. First, we note that
kCj + kCj−1

k × (k − 1) × · · · × (k − j + 1)
=
1 × 2 × ··· × j

k × (k − 1) × · · · × (k − j + 2)
×
1 × 2 × ··· × j − 1
 
k × (k − 1) × · · · × (k − j + 2) (k − j + 1)
= × +1
1 × 2 × ··· × j − 1 j

k × (k − 1) × · · · × (k − j + 2) (k + 1)
= ×
1 × 2 × ··· × j − 1 j

= (k + 1)Cj .
Again applying Theorem 2.1.6 and by using nC0 = nCn = 1, ∀n ∈ N, we get
( f · g)(k+1) = (( f · g)(k) )
⎛ ⎞
k
= ⎝ kCj f ( j) · g(k−j) ⎠
j=0

k  
= kCj f ( j+1) · g(k−j) + f ( j) · g(k−j+1)
j=0

k+1 k
= kCj−1 f ( j) · g(k−j+1) + kCj f ( j) · g(k−j+1)
j=1 j=0

= kCk f (k+1) · g + kC0 f · g(k+1)

k
+ kCj + kCj−1 f ( j) · g(k−j+1)
j=1

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 76 — #6

i i

76 Differentiability

k
= (k + 1)C0 f · g(k+1) + (k + 1)Cj f ( j) · g(k+1−j)
j=1

+ (k + 1)Ck+1 f (k+1)
·g
(k+1)
= (k + 1)Cj f ( j) · g(k+1−j) .
j=0

Thus, the result follows. 

A function f of a complex variable z can also be viewed as a function of
two real variables x and y. Therefore, we can discuss about the partial deriva-
tives of f with respect to x and y, and the relation between the differentiability
of f and partial differentiabilities of f .

Definition 2.1.14 Let f :  → C and (x0 , y0 ) ∈ , where  is a region.

1. f is said to be partially differentiable with respect to x at (x0 , y0 )
f (x0 + s, y0 ) − f (x0 , y0 ) ∂f
if lim exists and is denoted by (x0 , y0 ) or
s→0 s ∂x
fx (x0 , y0 ).
2. f is said to be partially differentiable with respect to y at (x0 , y0 )
f (x0 , y0 + t) − f (x0 , y0 ) ∂f
if lim exists and is denoted by (x0 , y0 ) or
t→0 t ∂y
fy (x0 , y0 ).
In the above two limits, s and t approach 0 through reals.

THEOREM 2.1.15 If f and g are partially differentiable with respect to x (with

respect to y), then
1. f + g is partially differentiable with respect to x (with respect to y) and
( f + g)x = fx + gx (( f + g)y = fy + gy ).
2. cf is partially differentiable with respect to x (with respect to y) and
(cf )x = cfx ((cf )y = cfy ), where c ∈ C.

Proof: Proof of this theorem is analogous to that of Theorem 2.1.6. 

As an immediate consequence, we have the following corollary.

COROLLARY 2.1.16 Let f = u + iv, then f is partially differentiable with

respect to x (with respect to y) iff u and v are partially differentiable with
respect to x (with respect to y). In this case, fx = ux + ivx (fy = uy + ivy ).

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 77 — #7

i i

Analytic Functions 77

Example 2.1.17 In general, though fxy and fyx exist at a point, they need not
be equal to each other.
⎧ 3
⎨ x y
(x, y)  (0, 0)
Consider f (x, y) = x2 + y2 .
⎩
0 (x, y) = 0
Now for s, t ∈ R \ {0}, we have
f (0 + s, 0) − f (0, 0) 0−0
fx (0, 0) = lim = lim =0
s→0 s s→0 s
f (0 + s, t) − f (0, t) s2 t
fx (0, t) = lim = lim 2 =0
s→0 s s→0 s + t2
f (0, 0 + t) − f (0, 0) 0−0
fy (0, 0) = lim = lim =0
t→0 t t→0 t
f (s, 0 + t) − f (s, 0) s3
fy (s, 0) = lim = lim 2 = s.
t→0 t t→0 s + t2

Therefore, we have
fy (s, 0) − fy (0, 0) s
fxy (0, 0) = lim = lim = 1
s→0 s s→0 s

and
fx (0, t) − fx (0, 0) 0−0
fyx (0, 0) = lim = lim = 0.
t→0 t t→0 t
Thus, fxy (0, 0)  fyx (0, 0).

Next we recall mean-value theorem from real analysis, which will be applied
in the following sequel.

THEOREM 2.1.18 (Mean-value theorem)

Let f : [a, b] → R be a continuous function. If f is differentiable on (a, b),
then there exists x ∈ (a, b) such that f (b) − f (a) = f  (x)(b − a).

THEOREM 2.1.19 (Complex version of mean-value theorem)

Let f : [a, b] → C be a continuous function. If f is differentiable on (a, b),
then there exists x ∈ (a, b) such that | f (b) − f (a)| ≤ | f  (x)|(b − a).

Proof: If f = (u, v), then u and v are real-valued continuous functions on

[a, b], and they are differentiable on (a, b). Furthermore,
f  (t) = (u (t), v (t)), ∀ t ∈ [a, b]

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 78 — #8

i i

78 Differentiability

Define φ : [a, b] → R by

φ(t) = xu(t) + yv(t), ∀ t ∈ [a, b].

where (x, y) = f (b) − f (a) = (u(b) − u(a), v(b) − v(a))

Therefore, φ is a real-valued continuous function on [a, b] and φ  (t) =
xu (t) + yv (t), ∀t ∈ (a, b).


Now applying mean-value theorem (Theorem 2.1.18), we choose t ∈

(a, b) such that
φ(b) − φ(a) = φ  (t)(b − a).

Therefore,

|(x, y)|2 = x2 + y2
= x(u(b) − u(a)) + y(v(b) − v(a))
= xu(b) + yv(b) − (xu(a) + yv(a))
= φ(b) − φ(a)
= φ  (t)(b − a)
= (xu (t) + yv (t))(b − a)
 
≤ x2 + y2 (u (t))2 + (v (t))2 (b − a)
(by Cauchy–Schwarz inequality. )
= |(x, y)|| f  (t)|(b − a).

If |(x, y)|  0, then we have |(x, y)| ≤ | f  (t)|(b − a). If |(x, y)| = 0, then
|(x, y)| = 0 ≤ | f  (t)|(b − a). 

THEOREM 2.1.20 (Young’s theorem)

Let f :  → R and (a, b) ∈ . If fx exists in a neighbourhood of (a, b) and
fxy is continuous at (a, b), then fyx exists at (a, b) and fyx (a, b) = fxy (a, b).

Proof: Let fx , fy , and fxy exist on B ((a, b), r). Fix (s, t) ∈ C such that (a +
s, b + t) ∈ B ((a, b), r). If F : [b, b + t] → R is defined by

F(y) = f (a + s, y) − f (a, y), ∀y ∈ [b, b + t],

then F is differentiable on [b, b + t], and hence, by applying mean-value

theorem (Theorem 2.1.18) twice, we get

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 79 — #9

i i

Analytic Functions 79

f (a + s, b + t) − f (a, b + t) − [f (a + s, b) − f (a, b)]

= F(b + t) − F(b)
= tF  (b + λt), for some λ ∈ (0, 1)
 
= t fy (a + s, b + λt) − fy (a, b + λt)
= stfxy (a + μs, b + λt), for some μ ∈ (0, 1).

Using the continuity of fxy at (a, b), we obtain

fxy (a, b) = lim fxy (a + μs, b + λt)

(s,t)→(0,0)
 
1 f (a + s, b + t) − f (a, b + t) f (a + s, b) − f (a, b)
= lim lim −
t→0 s→0 t s s
(using Lemma 1.6.8)
fx (a, b + t) − fx (a, b)
= lim
t→0 t
= fyx (a, b).

Thus, the theorem follows. 

2.2 CAUCHY–RIEMANN EQUATIONS

Definition 2.2.1 (Cauchy–Riemann equations)
Let u and v be real-valued functions of two real variables x and y, then we say
that u and v satisfy Cauchy–Riemann equations (or simply C–R equations) if
ux = vy and vx = −uy .

THEOREM 2.2.2 Let  be a region, z0 ∈  and f :  → C. If f = u + iv is

differentiable at z0 , then f is partially differentiable at z0 = x0 +iy0 = (x0 , y0 ),
and u and v satisfy C–R equations.

f (z0 + h) − f (z0 )
Proof: As f is differentiable at z0 , we have f  (z0 ) = lim
h→0 h
exists or equivalently,

f ((x0 + s) + i(y0 + t)) − f (x0 + iy0 )

f  (x0 + iy0 ) = lim exists. (2.1)
s+it→0 s + it

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 80 — #10

i i

80 Cauchy–Riemann Equations

By letting t = 0 and s → 0 in equation (2.1), we get

f ((x0 + s) + iy0 ) − f (x0 + iy0 )

f  (x0 + iy0 ) = lim
s→0 s
f (x0 + s, y0 ) − f (x0 , y0 )
= lim
s→0 s
= fx (x0 , y0 ).

If s = 0 and t → 0 in equation (2.1), then we get

f (x0 + i(y0 + t)) − f (x0 + iy0 )

f  (x0 + iy0 ) = lim
t→0 it
1 f (x0 , y0 + t) − f (x0 , y0 )
= lim
i t→0 t
= −ify (x0 , y0 ).

Hence, fx and fy exist at (x0 , y0 ), and we also have fx = f  = −ify . As f =

u + iv, using Corollary 2.1.16, we have

ux + ivx = fx = −ify = −i(uy + ivy ) = −iuy + vy .

Equating the real and imaginary parts on both sides, we get ux = vy and
vx = −uy . Thus, u and v satisfy C–R equations. 

Remark 2.2.3: From the above theorem, we conclude that a necessary con-
dition on f has to be differentiable at (x, y) such that f should satisfy the C–R
equations at (x, y) (i.e., fx = −ify ).

Example 2.2.4 If f (z) = x2 − iy3 and g(z) = exp(2x + i3y), ∀z ∈ C, where

x = Re z and y = Im z, then justify that f and g are not differentiable on C.
We now check the C–R equations for f .
If u(x, y) = x2 , v(x, y) = −y3 , ∀(x, y) ∈ C, then f = u + iv and we have

∂u ∂u
= 2x, = 0
∂x ∂y
∂v ∂v
= 0, = −3y2 .
∂x ∂y

∂u ∂v
Therefore,  . Thus, f does not satisfy C–R equations at any (x, y) ∈
∂x ∂y
C\{(0, 0)}, and hence, it is not differentiable on C.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 81 — #11

i i

Analytic Functions 81

If u(x, y) = exp(2x) cos(3y), v(x, y) = exp(2x) sin(3y), ∀(x, y) ∈ C, then g =

u + iv and we have
∂u ∂u
= 2 exp(2x) cos(3y), = −3 exp(2x) sin(3y)
∂x ∂y
∂v ∂v
= 2 exp(2x) sin(3y), = 3 exp(2x) cos(3y).
∂x ∂y

Therefore, f does not satisfy C–R equations at any (x, y) ∈ C, and hence, it is
not differentiable on C.

Definition 2.2.5 Let  be a region, then a function u on  is called a

harmonic function if u satisfies the Laplace equation uxx + uyy = 0 on .

RESULT 2.2.6 If f = u + iv and f is differentiable on , then u and v are

harmonic functions on .

Proof: (At present, assume the fact that partial derivatives of u and v of all
orders exist and they are continuous. This will be obtained as a consequence
of Theorem 4.3.3.) By Theorem 2.2.2, if f is differentiable on  and u and
v satisfy C–R equations, then ux = vy and uy = −vx . Now applying Young’s
theorem (Theorem 2.1.20), we get
∂ux ∂uy ∂vy −∂vx
1. uxx + uyy = + = + = vyx − vxy = 0. Hence, u is
∂x ∂y ∂x ∂y
harmonic.
∂vx ∂vy −∂uy ∂ux
2. vxx + vyy = + = + = uxy − uyx = 0. Hence, v is
∂x ∂y ∂x ∂y
harmonic. 

Definition 2.2.7 Given a harmonic function u on a region , another har-

monic function v on  is called a harmonic conjugate of u if u + iv is a
differentiable function on .

RESULT 2.2.8 v is a harmonic conjugate of u iff −u is a harmonic conjugate

of v.

Proof: v is a harmonic conjugate of u iff u + iv is an analytic function

iff −i(u + iv) = v − iu is an analytic function iff −u is a harmonic
conjugate of v. 

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 82 — #12

i i

82 Cauchy–Riemann Equations

Algorithm 2.2.9 (To find harmonic conjugate.)

Step 1. Let the given function be u and find ux .
Step 2. Put vy = ux and integrate vy partially with respect to y and write
v = vy dy + φ , where φ is a function of x.
Step 3. Using this v, find vx .
Step 4. From the given u, find uy .
Step 5. Using uy = vx , find the value of φ  and then φ .

Now, we present two more algorithms (by Milne-Thompson) to find an ana-

lytic function f whose real part u (or imaginary part v) is the given harmonic
function. Replacing z by x + iy and finding the real and imaginary parts of
f (x + iy), we can also find the harmonic conjugate of u (or v).

Algorithm 2.2.10 (Finding an analytic function f = u + iv from the given u)

Step 1. Check whether the given function u is a harmonic function. If yes,
then go to next step.
Step 2. Find ux and uy .
Step 3. Put f  (z) = ux (z, 0) − iuy (z, 0).

Step 4. Find f = f  (z) dz + c, where c is a constant.
Step 5. If you want to find the harmonic conjugate of u, then put z = x + iy
and find real and imaginary parts of f .
Step 6. Write the harmonic conjugate of u by Im f .

Algorithm 2.2.11 (Finding an analytic function f = u + iv from the given v )

Step 1. Check whether given function v is a harmonic function.
Step 2. Find vx and vy .
Step 3. Put f  (z) = vy (z, 0) + ivx (z, 0).

Step 4. Find f = f  (z) dz + c, where c is a constant.
Step 5. If you want to find the harmonic conjugate of v, then put z = x + iy
and find real and imaginary parts of f .
Step 6. Write the harmonic conjugate of v by −Re f .

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 83 — #13

i i

Analytic Functions 83

Example 2.2.12 If u(x, y) = x3 − 3xy2 + 2x2 − 2y2 + x, then prove that u is

a harmonic function and find its harmonic conjugate.
As
ux = 3x2 − 3y2 + 4x + 1, uxx = 6x + 4
uy = −6xy − 4y and uyy = −6x − 4
we have
uxx + uyy = 6x + 4 − 6x − 4 = 0.
Hence, u is a harmonic function. If v is a harmonic conjugate of u, then u and
v satisfy C–R equations (i.e., ux = vy , uy = −vx ). The first equation

vy = ux = 3x2 − 3y2 + 4x + 1

implies
 
v = vy dy = (3x2 − 3y2 + 4x + 1) dy = 3x2 y − y3 + 4xy + y + φ(x)

for some function φ of x. Using uy = −vx , we get

−6xy − 4y = −6xy − 4y − φ  (x).

Therefore, φ  (x) = 0, and hence, φ is a constant. Thus,

v = 3x2 y − y3 + 4xy + y + c

for some real constant c.

sin(2x)
Example 2.2.13 If u(x, y) = , then find the analytic
cosh(2y) − cos(2x)
function whose real part is u.
∂u (cosh(2y) − cos(2x))2 cos(2x) − sin(2x)2 sin(2x)
=
∂x (cosh(2y) − cos(2x))2
2(cos(2x) cosh(2y) − 1)
=
(cosh(2y) − cos(2x))2
∂u 2 sin(2x) sinh(2y)
= − .
∂y (cosh(2y) − cos(2x))2
If f = u + iv, then
∂u ∂u
f  (z) = (z, 0) − i (z, 0)
∂x ∂y

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 84 — #14

i i

84 Cauchy–Riemann Equations

2(cos(2z) − 1)
=
(1 − cos(2z))2
2
=
cos(2z) − 1
2
=
cos2 (z) − sin2 (z) − cos2 (z) − sin2 (z)
2
=
−2 sin2 (z)
= − csc2 (z)

and hence, we get f (z) = cot(z) + c for some complex constant c.

Example 2.2.14 If u(x, y) = cosh(x) sin(y), ∀(x, y) ∈ C, then prove that u

is a harmonic function and find the analytic function f whose real part is u.
Find also its harmonic conjugate.

ux = sinh(x) sin(y), uxx = cosh(x) sin(y)

uy = cosh(x) cos(y) uyy = − cosh(x) sin(y).
Therefore, uxx + uyy = cosh(x) sin(y) − cosh(x) sin(y) = 0, and hence, u is a
harmonic function.
As ux = sinh(x) sin(y) and uy = cosh(x) cos(y), we put

f  (z) = ux (z, 0) − iuy (z, 0) = −i cosh(z).

 
Therefore, f (z) = f  (z) dz = −i cosh(z) dz + c = −i sinh(z) + c.
Writing z = x + iy, we get

f (x + iy) = −i sinh(x + iy) + c

= − sin(i(x + iy)) + c
= − sin(ix − y) + c
= − sin(ix) cos(y) + sin(y) cos(ix) + c
= −i sinh(x) cos(y) + sin(y) cosh(x) + c.

Hence, the harmonic conjugate of cosh(x) sin(y) is − sinh(x) cos(y) + r for

some real constant r.

Example 2.2.15 Prove that v(x, y) = exp(x)(x sin(y) + y cos(y)), ∀(x, y) ∈ C

is a harmonic function. Find an analytic function whose imaginary part is v
and also find its harmonic conjugate.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 85 — #15

i i

Analytic Functions 85

vx = exp(x)(x sin(y) + y cos(y)) + exp(x) sin(y)

= exp(x)((x + 1) sin(y) + y cos(y))
vxx = exp(x)((x + 1) sin(y) + y cos(y)) + exp(x) sin(y)
= exp(x)((x + 2) sin(y) + y cos(y))
vy = exp(x)(x cos(y) − y sin(y) + cos(y))
= exp(x)((x + 1) cos(y) − y sin(y))
vyy = exp(x)(−(x + 1) sin(y) − y cos(y) − sin(y))
= exp(x)(−(x + 2) sin(y) − y cos(y)).

As

vxx + vyy = exp(x)((x + 2) sin(y) + y cos(y) − (x + 2) sin(y) − y cos(y))

= 0

v is a harmonic function. We know that f  = fx = ux + ivx = vy + ivx ;

therefore,

f = exp(x)((x + 1) cos(y) − y sin(y))

+ i(exp(x)((x + 1) sin(y) + y cos(y)))
= exp(z)(z + 1) (by replacing x and y by z and 0 respectively).

Now,

f (z) = exp(z)(z + 1) dz

= (z + 1) exp(z) − exp(z) dz
= (z + 1) exp(z) − exp(z) = z exp(z).

Therefore, f (z) = z exp(z)+c, ∀z ∈ C for some constant c. Writing z = x+iy

and finding the real part of f (z), we get

u(x, y) = Re [(x + iy) exp(x + iy)]

= Re [(x + iy) exp(x)(cos(y) + i sin(y))]
= exp(x)(x cos(y) − y sin(y)).

Therefore, the harmonic conjugate of v(x, y) is

−u(x, y) = exp(x)(y sin(y) − x cos(y))

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 86 — #16

i i

86 Cauchy–Riemann Equations

Example 2.2.16 Find an analytic function f whose real and imaginary parts
are u and v, respectively, if u + v = exp(−y) (cos(x) − sin(x)).
We find
ux + vx = exp(−y) (− sin(x) − cos(x)) , (2.2)
uy + vy = − exp(−y) (cos(x) − sin(x)) ,
−vx + ux = exp(−y) (sin(x) − cos(x)) . (2.3)
Now, equations (2.2) + (2.3) and (2.2) − (2.3) imply that
ux = − exp(−y) cos(x),
vx = − exp(−y) sin(x).
Therefore,
f  (z) = ux (z, 0) + ivx (z, 0)
= −(cos(z) + i sin(z)) = − exp(iz)
and hence,
f (z) = i exp(iz) + c = exp(iz) + c
for some c ∈ C.

Exercise 2.2.17

1. Prove that the following functions are harmonic and find the harmonic
conjugates.
(a) u(x, y) = x3 − 3xy2 .
(b) u(x, y) = x4 + y4 − 6x2 y2 + 2xy − x.
(c) u(x, y) = exp(−x)(x sin(y) − y cos(y)).
(d) u(x, y) = cosh(x) sin(y).
(e) u(x, y) = exp(−x)(x cos(y) + y sin(y)).
y
2. Verify that u(x, y) = 2 , ∀(x, y) ∈ C \ {(0, 0)} is a harmonic
x + y2
function and find an analytic function whose real part is u.
3. Verify that v(x, y) = −3 exp(2x) sin(2y), ∀(x, y) ∈ C is a harmonic
function and find an anlytic function whose imaginary part is v.
4. Verify that u(x, y) = − exp(x2 − y2 ) sin(2xy), ∀(x, y) ∈ C and find an
analytic function whose real part is u. Find also the harmonic conjugate
of u.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 87 — #17

i i

Analytic Functions 87

5. Verify that v is a harmonic function and find the analytic function

whose imaginary part is v, where v(x, y) = x2 − y2 + 2y, ∀(x, y) ∈ C.
Find also its harmonic conjugate.
6. Determine the analytic function f = u + iv if
(a) (u − v)(x, y) = x3 − y3 + 3x(x − y).
sin(2x) + sinh(2y)
(b) (u + v)(x, y) = .
cosh(2y) + cos(2x)
Answers: (1) (a) v(x, y) = 3x2 y − y3 + c.
(b) 4x3 y − 4xy3 − (x2 − y2 ) − y.
(c) v(x, y) = exp(−x)( y sin(y) + x cos( y)) + c.
(d) − sinh(x) cos( y) + c.
(e) exp(−x)( y cos( y) − x sin( y)).
i
(2) .
z
(3) −3 exp(2z).
(4) − exp(z2 ), exp(x2 − y2 ) cos(2xy).
(5) iz2 + 2z, 2xy − 2x.
(6) (a) f (z) = −iz3 + c.
(b) tan z + c.

LEMMA 2.2.18 Let φ : (a, b) → R be a differentiable function.

1. If φ  (t) = 0, ∀t ∈ (a, b), then φ is constant on (a, b).
2. If φ  (t) ≥ 0, ∀t ∈ (a, b), then φ is increasing on (a, b).
3. If φ  (t) ≤ 0, ∀t ∈ (a, b), then φ is decreasing on (a, b).
4. If φ  (t) > 0, ∀t ∈ (a, b), then φ is strictly increasing on (a, b).
5. If φ  (t) < 0, ∀t ∈ (a, b), then φ is strictly decreasing on (a, b).

Proof: Let x, y ∈ (a, b) be arbitrary such that x  y. We assume that x < y

without loss of generality. Then, φ is a differentiable function on [x, y], and
then by using mean-value theorem (Theorem 2.1.18), there exists t ∈ (x, y)
such that
φ(y) − φ(x) = φ  (t)(x − y) = 0.
1. If φ  = 0, then φ(y) = φ(x). Thus, φ is a constant.
2. If φ  ≥ 0, then φ(y) ≥ φ(x). Thus, φ is increasing.
The proof of the remaining assertions are similar. 

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 88 — #18

i i

88 Cauchy–Riemann Equations

THEOREM 2.2.19 If f is analytic on a region  such that f  = 0, then f is

constant.

Proof: Let z0 ∈  be arbitrary. Choose r > 0 such that B(z0 , r) ⊆ . If

f = u + iv, then f  = ux + ivx = vy − iuy implies ux = vx = uy = vy = 0.
From ux = 0, we get that u is a constant cy on Hz ∩ B(z0 , r), where Hz is the
horizontal line passing through z = (x, y) for every z ∈ B(z0 , r). Similarly,
from uy = 0, we get u, which is a constant cx on Vz ∩ B(z0 , r), where Vz is the
vertical line passing through z = (x, y) for every z ∈ B(z0 , r).
If z1 = (x1 , y1 ) and z2 = (x2 , y2 ) ∈ B(z0 , r), then the polygon passing
through (x1 , y1 ), (x1 , y2 ), and (x2 , y2 ) is completely lying inside B(z0 , r). See
Corollary 1.4.29 and the following diagram.

(x1, y2)
(x2, y2)

(x1, y1)

As (x1 , y1 ) and (x1 , y2 ) lie on a vertical line segment in B(z0 , r), we have
u(x1 , y1 ) = u(x1 , y2 ). As (x1 , y2 ) and (x2 , y2 ) lie on a horizontal line segment
in B(z0 , r), we have u(x1 , y2 ) = u(x2 , y2 ). Thus, we get u(z1 ) = u(z2 ), and
hence, u is constant on B(z0 , r).
By the same argument, from vx = vy = 0, we can prove that v is constant
on B(z0 , r), and so f is a constant function on B(z0 , r).
If a, b are any two points in , then by Theorem 1.4.31, we can find a
polygon such that
1. it passes through the vertices a = a0 , a1 , a2 , . . . an = b,

2. for each i = 1, 2, . . . , n, the line segment joining ai−1 and ai is either

vertical or horizontal,

3. for each i = 1, 2, . . . , n, the line segment joining ai−1 and ai is

contained in B(z, r) ⊆  for some z ∈  and r > 0.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 89 — #19

i i

Analytic Functions 89

Therefore, f (a) = f (a0 ) = f (a1 ) = f (a2 ) = · · · = f (an ) = f (b).

Hence, f is constant on . 

RESULT 2.2.20 Let f = u + iv be an analytic function on a region , then f

is a constant if any one of the following statements holds:
1. u is constant.
2. v is constant.
3. | f | is constant.
4. arg f is constant.

Proof: Assume that u is constant. Then, ux = uy = 0. As vx = −uy , we have

f  (z) = fx = ux + ivx = 0 on . Therefore, by using Theorem 2.2.19, we get
that f is constant.
By a similar argument, using the C–R equations, one can prove that if v is
constant, then u is constant, and hence, f is constant.
If | f | = 0, then obviously f = 0, and hence, f is constant. If | f | is con-
stant and non-zero, then we have u2 + v2 as constant and non-zero. Partially
differentiating with respect to x and y, we get
2uux + 2vvx = 0, 2uuy + 2vvy = 0.
Using C–R equations, we get
uux − vuy = 0 and vux + uuy = 0.
 
 u −v 
As determinant of the coefficient matrix   = u2 + v2  0, this
v u 
system of homogeneous linear equations in the two variables ux and uy has
a unique solution and is ux = 0 = uy . Hence, u is constant. Therefore, f is
constant. v
If arg f is constant, then is constant, say c. It follows that v − cu is constant,
u
but v − cu = Im (1 − ci)f , and hence, (1 − ci)f is constant. Thus, f is constant.


RESULT 2.2.21 If f = u + iv is differentiable, then we have ux vx + uy vy = 0.

Geometrically, this result is rephrased by the family u(x, y) = c of curves is
orthogonal with the family v(x, y) = d of curves.

Proof: As u and v satisfy C–R equations, we have ux = vy and vx = −uy .

Therefore, ux vx + uy vy = −ux uy + ux uy = 0. 

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 90 — #20

i i

90 Cauchy–Riemann Equations

∂ 2 u2 ∂ 2 u2
RESULT 2.2.22 If f = u + iv is analytic, then + = 2| f  |2 .
∂x2 ∂y2

∂u ∂v
Proof: First we note that u and v satisfy the C–R equations, = and
∂x ∂y
∂u ∂v
=− .
∂y ∂x
Now,
   2
∂ 2 u2 ∂ ∂u ∂u ∂ 2u
= 2u = 2 + 2u .
∂x2 ∂x ∂x ∂x ∂x2
 2
∂ 2 u2 ∂u ∂ 2u
Similarly, we get = 2 + 2u . Therefore, using the fact that
∂y2 ∂y ∂y2
u is harmonic, we get
 2  2   
∂ 2 u2 ∂ 2 u2 ∂u ∂u ∂ 2u ∂ 2u
+ = 2 + + 2u + 2
∂ 2 x2 ∂y2 ∂x ∂y ∂x2 ∂y
 2  2 
∂u ∂u
= 2 +
∂x ∂y
 2  2 
∂u ∂v
= 2 + (using C–R equations)
∂x ∂x
 
∂f ∂u ∂v
= 2| f  |2 as f  = = +i .
∂x ∂x ∂x

Exercise 2.2.23 If f is an analytic function, then

 
∂2 ∂2
1. + | f (z)|2 = 4| f  (z)|2
∂x2 ∂y2
 2  2 
∂ ∂
2. + | f (z)|2 = | f  (z)|2 .
∂x ∂y

RESULT 2.2.24 (Cauchy–Riemann equations in polar form)

∂u ∂v ∂v ∂u
If u and v satisfy C–R equations, then r = and r =− .
∂r ∂θ ∂r ∂θ

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 91 — #21

i i

Analytic Functions 91

∂u ∂v
Proof: We know that C–R equations in Cartesian form are given by =
∂x ∂y
∂u ∂v
and = − . From the relation x = r cos(θ ) and y = r sin(θ ) between the
∂y ∂x
Cartesian coordinates and polar coordinates, we get,
 
∂u ∂u ∂x ∂u ∂y
r = r +
∂r ∂x ∂r ∂y ∂r
∂v ∂v
= − r sin(θ ) + r cos(θ )
∂x ∂y
∂v ∂x ∂v ∂y
= +
∂x ∂θ ∂y ∂θ
∂v
= ,
∂θ 
∂v ∂v ∂x ∂v ∂y
r = r +
∂r ∂x ∂r ∂y ∂r
∂u ∂u
= r sin(θ) − r cos(θ )
∂x ∂y
∂u ∂x ∂u ∂y
= − −
∂x ∂θ ∂y ∂θ
∂u
= − .
∂θ

Hence, C–R equations in polar form are obtained. 

RESULT 2.2.25 If f is a complex-valued function on a region  satisfying

∂f ∂f ∂f
the C–R equation = − , then = 0.
∂x ∂y ∂z

1 1
Proof: If z = x + iy, then we have x =(z + z) and y = (z − z). Therefore,
2 i2
 
∂f ∂f ∂x ∂f ∂y 1 ∂f ∂f
= + = +i = 0.
∂z ∂x ∂z ∂y ∂z 2 ∂x ∂y

Hence, the result follows. 

Example 2.2.26 There exists a function f on C such that fx and fy exist and
satisfy C–R equation at a point, but it is not differentiable at the same point.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 92 — #22

i i

92 Cauchy–Riemann Equations

Define
⎧
⎨ xy(x + iy) if (x, y)  (0, 0)
f (x, y) = x2 + y2 , ∀(x, y) ∈ C
⎩
0 if (x, y) = (0, 0)
As f (x, 0) = 0, ∀x ∈ R, we obtain fx (0, 0) = 0, and as f (0, y) = 0, ∀y ∈ R,
we get fy (0, 0) = 0. Hence, the C–R equation fx = −ify is satisfied at (0, 0).
However,
f (2h, h) − f (0, 0) 2h2 (2h + ih)
lim = lim
h→0 2h + ih h→0 5h2 (2h + ih)
2
=
5
1

2
h2 (h + ih)
= lim 2
h→0 2h (h + ih)
f (h, h) − f (0, 0)
= lim
h→0 h + ih
Hence, f is not differentiable at (0, 0).

Example 2.2.27 There exists a function f on C such that fx and fy exist and
are continuous at a point, but it is not differentiable at the same point.
Consider the function
f (x, y) = x − iy, ∀(x, y) ∈ C
Now, fx = 1 and fy = −i, ∀(x, y) ∈ C. Clearly, fx and fy are continuous
functions. However, it is not differentiable at (0, 0) as f does not satisfy C–R
equations at (0, 0).

THEOREM 2.2.28 If f = u+iv has continuous partial derivatives with respect

to x, y on a region  and u, v satisfy C–R equations, then f is differentiable
on .
Proof: Let z0 = (x0 , y0 ) ∈  be arbitrary, then by applying mean-value
theorem, we get
u(x0 + s, y0 + t) − u(x0 , y0 )
= u(x0 + s, y0 + t) − u(x0 , y0 + t) + u(x0 , y0 + t) − u(x0 , y0 )
= sux (x0 + ps, y0 + t) + tuy (x0 , y0 + qt)
( for some p, q ∈ (0, 1))

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 93 — #23

i i

Analytic Functions 93

= s(ux (x0 , y0 ) + r1 ) + t(uy (x0 , y0 ) + r2 ), where r1 → 0, r2 → 0

as (s, t) → (0, 0), (by using the continuity of ux and uy )
= s(ux + r1 ) + t(uy + r2 )
(after writing ux (x0 , y0 ) and uy (x0 , y0 ) simply by ux and uy ).

Similarly, we get v(x0 + s, y0 + t) − v(x0 , y0 ) = s(vx + r3 ) + t(vy + r4 ), where

r3 → 0, r4 → 0 as (s, t) → (0, 0).
Therefore,

f (x0 + s, y0 + t) − f (x0 , y0 )
= s(ux + r1 ) + t(uy + r2 ) + is(vx + r3 ) + it(vy + r4 )
= s(ux + r1 ) + t(−vx + r2 ) + is(vx + r3 ) + it(ux + r4 )
= (s + it)ux + i(s + it)vx + s(r1 + ir3 ) + t(r2 + ir4 )
= (s + it)(ux + ivx ) + s(r1 + ir3 ) + t(r2 + ir4 )
= (s + it)fx + s(r1 + ir3 ) + t(r2 + ir4 ).

Hence,
 
 f (x0 + s, y0 + t) − f (x0 , y0 ) 
0 
≤  − fx 
s + it
 
 s t 

≤  (r1 + ir3 ) + (r2 + ir4 )
s + it s + it
≤ |r1 | + |r3 | + |r2 | + |r4 | → 0 as (s, t) → (0, 0).
f ((x0 + iy0 ) + (s + it)) − f (x0 + iy0 )
Thus, f  (z0 ) = lim exists and is
s+it→0 s + it
equal to fx (z0 ). 

RESULT 2.2.29 Let f :  → C, where  is a region such that z ∈  ⇒ z ∈

. If g(z) = f (z), ∀z ∈ , then f is differentiable iff g is differentiable.

Proof: Let f = u + iv and g = ϕ + iψ . As

ϕ(x, y) = u(x, −y) and ψ(x, y) = −v(x, −y), ∀(x, y) ∈ ,

we get

ϕx (x, y) = ux (x, −y), ϕy (x, y) = −uy (x, −y), ∀(x, y) ∈ 

and
ψx (x, y) = −vx (x, −y), ψy (x, y) = vy (x, −y), ∀(x, y) ∈ .

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 94 — #24

i i

94 Power Series and Abel’s Theorems

Hence, ux , uy , vx and vy exist and are continuous iff ϕx , ϕy , ψx and ψy exist

and are continuous. Furthermore, ux = vy and uy = −vx imply

ϕx (x, y) = ux (x, −y) = vy (x, −y) = ψy (x, y)

and
ϕy (x, y) = −uy (x, −y) = vx (x, −y) = −ψx (x, y).
Similarly, we can prove that ϕx = ψy and ϕy = −ψx imply ux = vy and uy =
−vx . Hence, using Theorems 2.2.2 and 2.2.28, we get that f is differentiable
iff ux , uy , vx and vy are continuous and satisfy C–R equations iff ϕx , ϕy , ψx
and ψy are continuous and satisfy C–R equations iff g is differentiable. 

2.3 POWER SERIES AND ABEL’S THEOREMS


∞
Definition 2.3.1 The series of functions an (z−a)n is called a power series,
n=0
where a ∈ C and (an ) is a sequence of complex numbers.

The convergence of a power series depends on the sequence (an ) and the
point z.

THEOREM 2.3.2 (Abel’s theorem on convergence of a power series)


∞
For a given power series an (z − a)n , we assign an extended real number
n=0
R, called the radius of convergence, satisfying the following properties:

∞
1. an (z − a)n converges absolutely on B(a, R).
n=0


∞
2. If 0 < S < R, then an (z − a)n converges absolutely and uniformly
n=0
on {z : |z − a| ≤ S}.

∞
3. If |z − a| > R, then an (z − a)n does not converge.
n=0

 −1
1
Proof: Define R = lim sup |an | n ∈ [−∞, ∞] with the convention that
n→∞
1 1
= ∞ and = 0. We claim that this R has the desired properties.
0 ∞

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 95 — #25

i i

Analytic Functions 95

1. Let |z − a| < R. Then, choose r ∈ R such that |z − a| < r < R. As

1 1 1 1 1 1
< , we can write = + , where = − > 0. Then, by
R r r R r R
1
definition of lim sup |an | n , there exists m ∈ N such that
n→∞
1 1 1
|an | n < + = , ∀n ≥ m.
R r
1
Hence, it follows that |an | < n , ∀n ≥ m. Therefore, we have
r
 
|z − a| n
|an (z − a) | ≤
n
, ∀n ≥ m
r
 
|z − a| 
∞ |z − a| n
As < 1, we get that converges. (See Theorem
r n=0 r
1.3.31.) Thus, by comparison test (Theorem 1.3.30), we get that the
∞
series an (z − a)n converges absolutely.
n=0

2. If |z − a| < S < R, then choose r ∈ R such that S < r < R. Arguing

1
as before, we get that there exists m ∈ N such that |an | < n , ∀n ≥ m.
r
Hence, it follows that
 n
S
|an (z − a) | ≤
n
, ∀z with |z − a| ≤ S, ∀n ≥ m.
r
 n
S 
∞ S
Since < 1, we get that the series converges. Again by
r n=0 r
∞
comparison test, we get that an (z − a)n converges absolutely and
n=0  n
∞ S
uniformly on |z − a| ≤ S as the convergence of the series
n=0 r
used in the comparison test is independent of z.
3. If |z − a| > R, then we choose t ∈ R such that |z − a| > t > R. Then,
1 1 1
< . From the definition of lim sup |an | n , there exists a subsequence
t R n→∞
1 1 1 1 1
(|ank | ) of (|an | n ), which converges to . Then for = − > 0,
nk
R R t
there exists m ∈ N such that
 
 
|an | nk − 1  < 1 − 1 , ∀k ≥ m.
1

 k R R t

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 96 — #26

i i

96 Power Series and Abel’s Theorems

1 1
Therefore, we have |ank | nk > , ∀k ≥ m, which implies
t
1
|ank | > , ∀k ≥ m.
t nk
Hence, for |z − a| > t and k ≥ m, we have
  nk
|z − a|
|ank (z − a) | >
nk
> 1, ∀k ≥ m.
t

Thus, the subsequence (ank (z−a)nk ) of (an (z−a)n ) does not converge to

∞
0. Hence, (an (z −a)n ) itself does not converge to 0. Thus, an (z −a)n
n=0
diverges by using Result 1.3.27. 


∞
THEOREM 2.3.3 Let the radius of convergence of an (z − a)n be R. Then
n=0


∞
1. the radius of convergence of nan (z − a)n−1 is also R,
n=1


∞
2. if f (z) = an (z − a)n on |z − a| < R, then f is analytic on |z − a| < R
n=0

∞
and f  (z) = nan (z − a)n−1 .
n=1

1 1
Proof: Recall that = lim sup |an | n .
R n→∞

1 1
1. First, we prove that n n → ∞ as n → ∞. If rn = n n − 1, ∀n ∈ N, then
we have for each n ≥ 2,

n(n − 1) 2
n = (rn + 1)n = 1 + nC1 rn + nC2 rn2 + · · · + rnn > rn ,
2

2 1
which implies that rn < → 0 as n → ∞. Therefore, n n −1 →
n−1
0 as n → ∞. Hence, our claim follows. Using Theorem 1.3.19, we get
1 1 1 1
lim sup n n |an | n = lim n n lim sup |an | n = R.
n→∞ n→∞ n→∞

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 97 — #27

i i

Analytic Functions 97


∞ 
m
2. Put g(z) = nan (z − a)n−1 , Sm (z) = an (z − a)n , and Rm (z) =
n=1 n=0

∞
an (z − a)n , ∀z ∈ B(a, R) and ∀m ∈ N. Obviously, we have the
n=m+1
following:
(a) f (z) = Sm (z) + Rm (z), ∀m ∈ N on B(a, R).
 (z) → g(z) as m → ∞ on B(a, R).
(b) Sm

∞
(c) n|an |rn−1 → 0 as m → ∞ for 0 ≤ r < R, using Result
n=m+1
1.3.26.
For a fixed z0 ∈ B(a, R) and any z ∈ B(a, R), choose r ∈ R such that

max{|z − a|, |z0 − a|} < r < R.

Choose m1 , m2 ∈ N such that

  
S (z0 ) − g(z0 ) < , ∀m ≥ m1 (2.4)
m
3
and
∞
n|an |rn−1 < , ∀m ≥ m2 . (2.5)
3
n=m+1
For m ≥ m2 , we have
  ∞
 Rm (z) − Rm (z0 )  |(z − a)n − (z0 − a)n |
  ≤ |an |
 z − z0  |z − z0 |
n=m+1
∞ n−1
≤ |an | |(z − a)k (z0 − a)n−k−1 |
n=m+1 k=0
n−1
(using (A − B)n = (A − B) Ak Bn−k )
k=0
∞
< n|an |rn−1 < . (2.6)
3
n=m+1

From the definition of Sm  (z ), given > 0, there exists δ > 0 such that
0
 
 Sm (z) − Sm (z0 ) 
 − Sm (z0 ) < whenever 0 < |z − z0 | < δ. (2.7)

 z − z0 3

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 98 — #28

i i

98 Power Series and Abel’s Theorems

 m > max{m1 , m2 } and for z with 0 < |z − z0 | < δ , we get

For
 f (z) − f (z0 ) 
 − g(z0 )
 z−z
0
 
 Sm (z) − Sm (z0 ) Rm (z) − Rm (z0 ) 
=  
− Sm (z0 ) + + Sm 
(z0 ) − g(z0 )
z − z0 z − z0
   
 Sm (z) − Sm (z0 )   Rm (z) − Rm (z0 )    
≤  
− Sm (z0 ) +   + Sm (z0 ) − g(z0 )
z − z0 z − z0
< + + =
3 3 3
Hence, f is differentiable and f  = g. 


∞
COROLLARY 2.3.4 Let the power series an (z−a)n converge in |z−a| < R
n=0

∞
f (n) (a)
for some R > 0. If f (z) = an (z − a)n , ∀z ∈ B(a, R), then an = n! ,
n=0
∀n = 0, 1, 2, 3, . . .

Proof: Applying the previous theorem repeatedly, first we note that f (m)
exists for every m = 0, 1, 2, 3, . . . and
m−1 ∞

d m
f (m) (z) = an (z − a)n + an (z − a)n
dzm n=m
n=0
∞
= n(n − 1) · · · (n − m + 1)an (z − a)n−m
n=m
∞
= am m! + n(n − 1) · · · (n − m + 1)an (z − a)n−m
n=m+1

and hence, f (m) (a) = am m!. Thus, the corollary follows. 


∞ 
∞
COROLLARY 2.3.5 an (z − a)n = bn (z − a)n , ∀z ∈ B(a, R) iff an =
n=0 n=0
bn , ∀n = 0, 1, 2, 3, . . .

Proof: If an = bn , ∀n = 0, 1, 2, 3, . . ., then obviously we get

∞ ∞
an (z − a)n = bn (z − a)n .
n=0 n=0

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 99 — #29

i i

Analytic Functions 99


∞
To prove the converse, let f (z) = an (z − a)n , ∀z ∈ B(a, R). If an =
n=0
f (n) (a)
bn , ∀n = 0, 1, 2, 3, . . . then by using previous corollary, we get an = =
n!
bn , ∀n = 0, 1, 2, 3, . . . 

Remark 2.3.6: If f is an analytic function on B(a, R), then

∞
f (n) (a)
(z − a)n
n!
n=0

is called the Taylor’s series of f . The corollary 2.3.5 states that if

∞
f (z) = an (z − a)n , ∀z ∈ B(a, R),
n=0

∞
then the power series an (z − a)n is same as the Taylor’s series of f .
n=0

Example 2.3.7 Find the radius of convergence of the following power series:

∞
1. nn zn ;
n=0

∞
2. n2 zn ;
n=0
  n2

∞ 1
3. 1− zn ;
n=0 n

∞
4. (4 + i3)n zn .
n=0
Solution:
1. Let an = nn , ∀n = 0, 1, 2, . . . . As |an |1/n = n → ∞ as n → ∞, the
radius of convergence of this power series is 0.
2. Let an = n2 , ∀n = 0, 1, 2, . . . . As |an |1/n = n2/n → ∞ as n → ∞, the
radius of convergence of this power series is 0.

 2  1n  
1 n 1 n 1
3. As lim 1− = lim 1 − = , the radius of
n→∞ n n→∞ n e
convergence of this power series is e.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 100 — #30

i i

100 Power Series and Abel’s Theorems

1
4. As lim (|4 + i3|n ) n = |4 + i3| = 5, the radius of convergence of
n→∞
∞ 1
(4 + i3)n zn is .
n=0 5

Exercise 2.3.8

∞ 2n
1. zn ;
n=0 n!


∞ √
n zn ;
2. 2
n=0


∞ n+1 an+1 √
3. zn (Hint. Find lim which is same as lim n an .);
n=0 (n + 2)(n + 3) n→∞ an n→∞

∞ 1
4. 2
zn ;
n=0 n

∞ (n!)2
5. zn .
n=0 (2n)!

Answer: 1. ∞; 2. 1; 3. 1; 4. 1; 5. 4.

THEOREM 2.3.9 (Abel’s limit theorem)


∞ 
∞
Let an converge. If f (z) = an zn , then f (z) → f (1) as z → 1 such that
n=0 n=0
|1 − z|
is bounded.
1 − |z|


∞
Proof: First we note that, as an converges at 1, the radius R of conver-
n=0

∞
gence of an is at least 1 by Abel’s theorem. If R > 1, then f is continuous
n=0
at 1, and hence, the theorem follows. Therefore, assume that R = 1.

∞
Case 1: f (1) = an = 0.
n=0

n 
n
If αn = ak and Sn (z) = ak zk , ∀z ∈ B(0, 1) and ∀n = 0, 1, 2, . . .,
k=0 k=0

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 101 — #31

i i

Analytic Functions 101

|1 − z|
then for z ∈ B(0, 1) with ≤ K , for some K > 0, we get
1 − |z|

n
Sn (z) = ak zk
k=0
n
= α0 + (αk − αk−1 )zk
k=1
n n
= α0 + αk zk − αk−1 zk
k=1 k=1
n−1 n−1
= α0 + αk zk + αn zn − α0 z − αk zk+1
k=1 k=1
n−1
= α0 (1 − z) + αk (zk − zk+1 ) + αn zn
k=1
n−1
= α0 (1 − z) + αk zk (1 − z) + αn zn
k=1
n−1
= (1 − z) αk zk + αn zn .
k=0

Allowing n → ∞ on both sides, we get

∞ ∞
f (z) = (1 − z) αk zk + lim αn zn = (1 − z) αk zk ,
n→∞
k=0 k=0

as αn → f (1) = 0 as n → ∞ and |z| < 1. (cf. Proof of Example

1.3.31.)
Again using αn → 0 as n → ∞, given > 0, there exists m ∈ N such
that |αn | < , ∀n ≥ m and choose δ such that
2K

0<δ<  .

m−1
2 1+ |αn |
k=1

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 102 — #32

i i

102 Exponential and Trigonometric Functions

|1 − z|
If 0 < |z − 1| < δ and ≤ K , then
1 − |z|
m−1 ∞

| f (z)| ≤ |1 − z| |αk zk | + |αk zk |
k=0 k=m
m−1 ∞

< |1 − z| |αk | + |z |
k
2K
k=0 k=m
m−1 ∞
< δ |αk | + |zm ||1 − z| |zk |
2K
k=0 k=0
|1 − z|
< +
2 2K 1 − |z|
≤ + = .
2 2
Hence, the theorem follows in this case.
Case 2: f (1)  0

∞
If g(z) = f (z) − f (1) = a0 − f (1) + an zn , then g(1) = 0. Then, by
n=1
Case 1, we obtain
|1 − z|
g(z) → g(1) as z → 1 with is bounded,
1 − |z|
and hence,
|1 − z|
f (z) − f (1) → 0 as z → 1 with is bounded.
1 − |z|

Thus, the proof is complete. 

2.4 EXPONENTIAL AND TRIGONOMETRIC

FUNCTIONS
Definition 2.4.1 The unique solution of the initial value problem f  = f with
f (0) = 1 is called the exponential function and is denoted by exp.

∞ zn
LEMMA 2.4.2 exp(z) = for z ∈ C at which the series converges.
n=0 n!

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 103 — #33

i i

Analytic Functions 103


∞
Proof: Let exp(z) = an zn . As it is the solution of the above initial value
n=0
problem, we have
∞ ∞ ∞
an zn = an−1 zn−1 = nan zn−1 , and a0 = f (0) = 1.
n=0 n=1 n=1
an−1
Hence, an = , n ∈ {0, 1, 2, . . .}, and a0 = f (0) = 1. We claim that
n
1
an = , ∀n = 0, 1, 2, 3, . . .
n!
We prove our claim by induction on n. For n = 0, we have a0 = 1. Assume
that the claim holds for some n ∈ N. Now,
an 1 1
an+1 = = = .
n+1 (n + 1)n! (n + 1)!
∞ zn
Thus, our claim follows. Therefore, exp(z) = , ∀z ∈ B(0, R), where R
n=0 n!
∞ zn
is the radius of convergence of . 
n=0 n!

∞ zn
LEMMA 2.4.3 The radius of convergence of exp(z) = is ∞.
n=0 n!

Proof: First, we note that

1
(n + 1)! 1
lim = lim = 0.
n→∞ 1 n→∞ n + 1
n!
Hence, applying Theorems 1.3.17 and 1.3.20, we get
1
(n + 1)!
0 = lim inf
n→∞ 1
n!
 1
1 n
≤ lim inf
n→∞ n!
 1
1 n
≤ lim sup
n→∞ n!

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 104 — #34

i i

104 Exponential and Trigonometric Functions

1
(n + 1)!
≤ lim sup
n→∞ 1
n!
= 0
 1
1 n
and hence, lim sup = 0.
n→∞ n!
∞ zn
Therefore, the radius of convergence of is ∞. 
n=0 n!

LEMMA 2.4.4 For every z1 , z2 ∈ C, exp(z1 + z2 ) = exp(z1 ) · exp(z2 ).

Proof: As
d
(exp(z) · exp(w − z)) = exp(z) · exp(w − z) − exp(z) · exp(w − z) = 0,
dz
we get exp(z) · exp(w − z) as a constant, say K . Putting z = 0 and using
exp(0) = 1, we get K = exp(w). Thus, we obtain

exp(z) · exp(w − z) = exp(w).

Replacing z by z1 and w by z1 + z2 , we get exp(z1 + z2 ) = exp(z1 ) ·

exp(z2 ). 
The above property of exponential function is called the addition theorem
for exp and is the reason for denoting exp(z) by ez .

COROLLARY 2.4.5 For every z ∈ C, exp(z)  0, and its multiplicative inverse

is exp(−z).

Proof: As for every z ∈ C, 1 = exp(0) = exp(z − z) = exp(z) · exp(−z), we

get exp(z)  0, and its multiplicative inverse is exp(−z). 

LEMMA 2.4.6 exp : R → (0, ∞) is a bijection.

∞ xn
Proof: First, note that exp(0) = 1, for x > 0, we have exp(x) = >
n=0 n!
1 and by previous corollary, for x < 0, exp(x) = exp(−x)
1
∈ (0, 1), since
−x > 0 ⇒ exp(−x) > 1. Therefore, we have exp(x) ≥ 1 if x ≥ 0 and
0 < exp(x) < 1 if x < 0. So, exp is a mapping from R into (0, ∞). To prove
exp : R → (0, ∞) is onto, let y > 0 be arbitrary.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 105 — #35

i i

Analytic Functions 105

Case (1): y = 1. Then, by definition, we have 0 ∈ R such that

exp(0) = 1.

∞ xn
Case (2): y > 1. Then, by definition, we have exp (x) = > x > 0, and
n=0 n!
hence, exp(x) → ∞ as x → ∞. Using this fact, we can choose s >
0 such that exp(s) > y > 1 = exp(0). As exp is differentiable, it
is continuous. Therefore, by intermediate value theorem (Theorem
1.6.30), there exists x ∈ (0, s) such that exp(x) = y.
1
Case (3): y < 1. Then, > 1. Therefore, by Case (2), there exists t > 0
y
1
such that exp(t) = . Therefore, exp(−t) = (exp(t))−1 = y.
y
For every x ∈ R, we have exp (x) = exp(x) > 0, and hence, exp is a strictly
increasing function (by Lemma 2.2.18), that is, x < y ⇒ exp(x) < exp(y).
Hence, exp is an one-to-one function. 

LEMMA 2.4.7 | exp(z)| = 1 iff z = iy for some y ∈ R.

Proof: Assume that z = iy for some y ∈ R. From the power series

representation of exp(z), it is obvious that exp(z) = exp(z), ∀z ∈ C.
Therefore,

| exp(iy)|2 = exp(iy) · exp(iy) = exp(iy − iy) = exp(0) = 1

using Lemma 2.4.4.

Conversely, assume that if z = x + iy and | exp(z)| = 1. Then, 1 =
| exp(x) · exp(iy)| = exp(x) implies that x = 0 since exp is a bijection from R
onto (0, ∞). Thus, z = iy for some real y ∈ R. 

Definition 2.4.8 For every z ∈ C, define

exp(iz) + exp(−iz) exp(iz) − exp(−iz)
cos(z) = and sin(z) = .
2 i2

By definition, both cos and sin are differentiable functions on C, and the
∞ (−1)n z2n
power series representations of these two functions are cos(z) =
n=0 (2n)!

∞ (−1)n z(2n+1)
and sin(z) = .
n=0 (2n + 1)!

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 106 — #36

i i

106 Exponential and Trigonometric Functions

LEMMA 2.4.9 For every z ∈ C,

1. sin(−z) = − sin(z) and cos(−z) = cos(z).

2. exp(iz) = cos(z) + i sin(z). (Euler’s formula)1 .

3. (cos(z) + i sin(z))n = cos nz + i sin nz, ∀n ∈ Z (de Moivre’s formula).

Proof: By direct computation, we get (1) and the Euler’s formula as given
below.
exp(−iz) − exp(iz) exp(iz) − exp(−iz)
sin(−z) = =− = − sin(z)
i2 i2
exp(−iz) + exp(iz)
cos(−z) = = cos(z)
2

exp(iz) + exp(−iz) exp(iz) − exp(−iz)

cos(z) + i sin(z) = +i = exp(iz).
2 i2
Next, by using Lemma 2.4.4 and Euler’s formula, for n ∈ N, we get

(cos(z) + i sin(z))n = (exp(iz))n = exp(niz) = cos(nz) + i sin(nz).

Hence, for n = −1, we get

(cos(z) + i sin(z))−1 = (exp(iz))−1

= exp(−iz)
= cos(−z) + i sin(−z)
= cos(z) − i sin(z).

If n is a negative integer, then −n ∈ N and n = −(−n). Therefore,

(cos(z) + i sin(z))n = (cos(z) + i sin(z))−(−n)

= (cos(−nz) + i sin(−nz))−1
= cos(nz) + i sin(nz).

Thus, the de Moivre’s formula is obtained. 

LEMMA 2.4.10 For every z ∈ C, sin2 (z) + cos2 (z) = 1.

1 Warning! cos(z)  Re eiz and sin(z)  Im eiz for z ∈ C.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 107 — #37

i i

Analytic Functions 107

Proof: By direct computation, we get

sin2 (z) + cos2 (z)
(exp(iz) − exp(−iz))2 (exp(iz) + exp(−iz))2
= +
(i2)2 4
−(exp(iz))2 − (exp(−iz))2 + 2 + (exp(iz))2 + (exp(−iz))2 + 2
=
4
4
= = 1. 
4

Example 2.4.11 Expand cos(7θ ) as a polynomial in cos(θ ).

Solution:
By de Moivre’s formula, we have
cos(7θ ) = Re (cos(7θ ) + i sin(7θ))
= Re (cos(θ) + i sin(θ))7
7
= Re 7Ck cosk (θ )(i sin(θ ))7−k
k=0
= cos7 (θ ) − 21 cos5 (θ ) sin2 (θ ) + 35 cos3 (θ ) sin4 (θ )
− 7 cos(θ) sin6 (θ )
= cos7 (θ ) − 21 cos5 (θ )(1 − cos2 (θ )) + 35 cos3 (θ )(1 − cos2 (θ ))2
− 7 cos(θ )(1 − cos2 (θ ))3
= cos7 (θ ) − 21 cos5 (θ )(1 − cos2 (θ ))
+ 35 cos3 (θ )(1 − 2 cos2 (θ ) + cos4 (θ ))
−7 cos(θ)(1 − 3 cos2 (θ ) + 3 cos4 (θ ) − cos6 (θ ))
= 64 cos7 (θ ) − 112 cos5 (θ ) + 56 cos3 (θ ) − 7 cos(θ ).

LEMMA 2.4.12 For every z, w ∈ C, (1) sin(z + w) = sin(z) cos(w) +

cos(z) sin(w) and (2) cos(z + w) = cos(z) cos(w) − sin(z) sin(w).

Proof: As
i2 sin(z + w) = exp(i(z + w)) − exp(−i(z + w))
= exp(iz) exp(iw) − exp(−iz) exp(−iw)
= (cos(z) + i sin(z))(cos(w) + i sin(w))
− (cos(z) − i sin(z))(cos(w) − i sin(w))
= (cos(z) cos(w) − sin(z) sin(w)) + i(cos(z) sin(w)

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 108 — #38

i i

108 Exponential and Trigonometric Functions

+ sin(z) cos(w)) − (cos(z) cos(w) − sin(z) sin(w))

+ i(cos(z) sin(w) + sin(z) cos(w))
= i2(cos(z) sin(w) + sin(z) cos(w)),
we get sin(z + w) = sin(z) cos(w) + cos(z) sin(w).
Similarly,
2 cos(z + w) = exp(i(z + w)) + exp(−i(z + w))
= exp(iz) exp(iw) + exp(−iz) exp(−iw)
= (cos(z) + i sin(z))(cos(w) + i sin(w))
+ (cos(z) − i sin(z))(cos(w) − i sin(w))
= (cos(z) cos(w) − sin(z) sin(w)) + i(cos(z) sin(w)
+ sin(z) cos(w)) + (cos(z) cos(w) − sin(z) sin(w))
− i(cos(z) sin(w) + sin(z) cos(w))
= 2(cos(z) cos(w) − sin(z) sin(w)).


COROLLARY 2.4.13 Prove that for z, w ∈ C, sin(z − w) = sin(z) cos(w) −

cos(z) sin(w) and cos(z − w) = cos(z) cos(w) + sin(z) sin(w).

Proof: In the previous lemma, replacing w by −w and using

sin(−w) = − sin(w), cos(−w) = cos(w),
we obviously get this corollary. 

d d
LEMMA 2.4.14 (1) sin(z) = cos(z) and (2) cos(z) = − sin(z).
dz dz

Proof: By direct calculation, we obtain

 
d d exp(iz) − exp(−iz) i exp(iz) + i exp(−iz)
1. sin(z) = =
dz dz i2 i2
exp(iz) + exp(−iz)
= = cos z.
2
 
d d exp(iz) + exp(−iz) i exp(iz) − i exp(−iz)
2. cos z = =
dz dz 2 2
 
exp(iz) − exp(−iz)
= − = − sin z.
i2


i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 109 — #39

i i

Analytic Functions 109

RESULT 2.4.15 There exists a least r0 > 0 such that cos(r0 ) = 0.

Proof: For every r > 0, by mean-value theorem, we have sin(r) − sin(0) =

cos(t), for some t ∈ (0, r), which implies that sin(r) ≤ r, ∀r > 0 as | cos(t)| ≤
exp(0) − exp(0) r
| exp(it)| = 1 and sin(0) = = 0. Applying on both sides
i2 0
of sin(r) < r, we get
 2 r
r r2 r2
[− cos(r)]r0 ≤ ⇒ − cos(r) + 1 ≤ ⇒ cos(r) ≥ 1 − .
2 0 2 2
r
Applying on both sides of the resulting inequalities repeatedly, we get
0

r3 r2 r4 r2 r4
sin(r) ≥ r − ⇒ − cos(r) + 1 ≥ − ⇒ cos(r) ≤ 1 − + .
6 2 24 2 24
√ 3 9
Therefore, cos( 3) ≤ 1 − + < 0, and from the definition, we obtain
2 24
cos(0) = 1. Therefore,
√ by intermediate value theorem (Theorem 1.6.30),
there exists r0 ∈ (0, 3) such that cos(r
√0 ) = 0.
Next, we claim that if 0 < r < r0 < 3, then cos(r)  0. From one of the
above inequalities, we get
   √ 
r3 r2 ( 3)2 r
sin(r) ≥ r − =r 1− >r 1− = > 0.
6 6 6 2

d
Therefore, cos(r) = − sin(r) < 0, and hence, cos is strictly decreasing on
dr
(0, r0 ), applying Lemma 2.2.18. Since sin(r) > 0, ∀r ∈ (0, r0 ) and sin2 (r) +
cos2 (r) = 1, we get sin, and it is strictly increasing on (0, r0 ). Therefore,

0 < sin(r) < sin(r0 ) = 1 − cos2 (r0 ) = 1.

Therefore, cos2 (r) = 1 − sin2 (r) > 0, and hence, cos(r)  0. 

Definition 2.4.16 Define π = 2r0 , where r0 is the least positive real number
such that cos(r0 ) = 0.
π  π 
Example 2.4.17 Prove that cos = 0, sin = 1, cos(π ) = −1,
2 2
sin(π ) = 0, cos(2π ) = 1, and sin(2π) = 0.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 110 — #40

i i

110 Exponential and Trigonometric Functions

By definition, π = 2r0 , where r0 is the least positive real number such that
cos(r0 ) = 0. Therefore,
π  π   π 
cos = cos(r0 ) = 0, sin = ± 1 − cos2 = ±1,
2 2 2
π  π  π
but sin = sin(r0 ) > 0, we have sin = 1. Therefore, exp i = i,
2 2 2
which implies that
exp(iπ) = i2 = −1 and exp(i2π ) = (−1)2 = 1.
Equating real and imaginary parts in the above two equations, we can get the
required identities.

Exercise 2.4.18 Prove that for every z ∈ C in the following:

π
1. sin z + 2 = cos(z).
π
2. cos z + 2 = − sin(z).
π
3. sin z − 2 = − cos(z).
π
4. cos z − 2 = sin(z).
5. sin (z + π) = − sin(z).
6. cos (z + π ) = − cos(z).
7. sin (z − π) = − sin(z).
8. cos (z − π ) = − cos(z).
Hint. Apply Result 2.4.12 and Corollary 2.4.13.

LEMMA 2.4.19 If 0 < t < 2π, then exp(it)  1.

Proof: If 0 < t < 2π and exp(it/4) = u + iv, then using the following:
t π
1. 0 < < ,
4 2
2. exp(it/4) = cos(t/4) + i sin(t/4),
π π 
3. is the least positive real number such that cos = 0,
2 2
 π
4. sin > 0 on 0, ,
2

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 111 — #41

i i

Analytic Functions 111

we get 0 < u < 1 and 0 < v < 1. As u + iv is not equal to any of 1, −1, i,
and −i, which are the solutions of z4 = 1, we obtain that exp(it/4) = u + iv
is not a solution of z4 = 1. Therefore, exp(it)  1. 

RESULT 2.4.20 exp(z) = 1 iff z = 2kiπ for some k ∈ Z.

Proof: Let z = 2kiπ for some k ∈ Z. If k = 0, then by definition,

exp(2kiπ ) = exp(0) = 1. If k is a positive integer, then

exp(2kiπ) = [exp(i2π )]k = 1.

If k is a negative integer, then k = −(−k) and −k is a positive integer.

Therefore, exp(2kiπ) = [exp(2(−k)iπ )]−1 = 1.
To prove the converse, let z be such that exp(z) = 1. Then, 1 = | exp(z)|.
Therefore, using Lemma 2.4.7, we have z = iy for some y ∈ R. We claim that
y = 2kπ for some k ∈ Z. Suppose if it is not so, then there exists a unique
integer n such that 2nπ < y < 2(n + 1)π . Therefore,

1 = exp(iy)
= exp(i(y − 2nπ + 2nπ))
= exp(i(y − 2nπ)) exp(i2nπ)
= exp(i(y − 2nπ)),

which is a contradiction to Lemma 2.4.19 as 0 < y − 2nπ < 2π . Thus,

z = 2kiπ for some k ∈ Z. 
Now, we define periodic function, which will be studied in detail in
Chapter 8.

Definition 2.4.21 Let f : → C and let a ∈ . We say that ‘a’ is a period

of f if f (z + a) = f (z), ∀z ∈ , where = C or R. A function is said to be
a periodic function if it has a non-zero period.

Remark 2.4.22: From addition theorem of exp function and Result 2.4.20,
we conclude that i2kπ are periods of exp.
Now, we prove a simple but useful lemma on the integrals of a periodic
function.

LEMMA 2.4.23 If f : R → R is a periodic function of period p > 0 and


a+p p
a ∈ R, then f (x) dx = f (x) dx.
a 0

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 112 — #42

i i

112 Exponential and Trigonometric Functions

Proof: By Archimedian property, we can find a unique m ∈ Z such that

m ≤ ap < (m + 1). Then, we have (m + 1)p < a + p and 0 ≤ a − mp < p.
Hence,
a+p 
(m+1)p a+p
f (x) dx = f (x) dx + f (x) dx
a a (m+1)p
p 
a−mp

= f (y − mp) dy + f (z − (m + 1)p) dz
a−mp 0
(by putting y = x − mp and z = x − (m + 1)p)
p 
a−mp

= f (y) dy + f (z) dz (since p is a period of f )

a−mp 0
p
= f (y) dy.
0

Hence, the lemma follows. 

RESULT 2.4.24 |z| = 1 iff z = exp(iy) for some y ∈ R.

Proof: If z = exp(iy) for some y ∈ R, then using Lemma 2.4.7, we get

|z| = 1. Conversely, assume that |z| = 1. As cos(0) = 1 and cos( π2 ) = 0 (see
Example 2.4.17), by intermediate value theorem (Theorem 1.6.30), for every
t ∈ [0, 1], there exists θ ∈ [0, π2 ] such that cos θ = t. Since |Re z| ≤ |z|, we
have |Re z| ≤ 1; similarly, we have |Im z| ≤ 1.
Case 1: If z = 1, then we choose θ = 0; obviously, we have exp(0) = 1.
If z = i, then we choose θ = π2 so that exp i π2 = cos π2 +i sin π2 =
i as cos π2 = 0 and sin π2 = 1.
If z = −1, then we choose θ = π so that exp(iπ ) = cos(π )+i sin(π ) =
−1 as cos(π ) = −1 and sin(π) = 0.
If z = −i, then we choose θ = − π2 so that exp −i π2 = cos π2 −
i sin π2 = −i as cos π2 = 0 and sin π2 = 1.
Case 2: 0 < Re z < 1, 0 < Im z < 1. Then, there exists θ ∈ 0, π2 such
that Re z = cos(θ ). Using sin(θ) > 0 in 0, π2 , we get
 
Im z = 1 − (Re z)2 = 1 − cos2 (θ ) = sin(θ ),

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 113 — #43

i i

Analytic Functions 113

and hence,

exp(iθ) = cos(θ ) + i sin(θ) = Re z + iIm z = z

Case 3: −1 < Re z < 0, 0 < Im z < 1. Then, −Re z = cos(θ ) for some
θ ∈ 0, π2 . As sin(θ) > 0 in 0, π2 , we get
 
Im z = 1 − (Re z)2 = 1 − cos2 (θ ) = sin(θ ),

and hence,

exp(i(π − θ )) = exp(iπ ) exp(−iθ )

= (cos(π) + i sin(π ))(cos(θ ) − i sin(θ ))
= − cos(θ ) + i sin(θ)
= Re z + iIm z
= z.

Case 4: −1 < Re z < 0, −1 < Im z < 0. Then, −Re z = cos(θ ) for some
θ ∈ 0, π2 . As sin(θ) > 0 in 0, π2 , we get
 
−Im z = 1 − (Re z)2 = 1 − cos2 (θ ) = sin(θ ),

and hence,

exp(i(−π + θ )) = exp(−iπ ) exp(iθ )

= (−1)(cos(θ ) + i sin(θ ))
= − cos(θ ) − i sin(θ )
= Re z + iIm z
= z.

Case 5: 0 < Re z < 1, −1 < Im z < 0. Then, Re z = cos(θ ) for some

θ ∈ (0, π2 ). As sin(θ) > 0 in (0, π2 ), we get
 
−Im z = 1 − (Re z)2 = 1 − cos2 (θ ) = sin(θ ),

and hence,

exp(−iθ ) = cos(θ ) − i sin(θ) = Re z + iIm z = z.

Hence, the result follows. 

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 114 — #44

i i

114 Exponential and Trigonometric Functions

COROLLARY 2.4.25 Every non-zero complex number, z can be written as

z = |z| exp(iθ ) for some θ ∈ (−π , π].

 
z  z 
Proof: As z  0, we write z = |z| , where   = 1. Therefore, from the
|z| |z|
z
proof of previous result, we get = exp(iy) for some y ∈ (−π , π ]. Hence,
|z|
the corollary follows. 

RESULT 2.4.26 If θ ∈ (0, π2 ), then, in a right-angled triangle with an angle θ ,

1. sin θ is equal to opposite side divided by hypotenuse.

2. cos θ is equal to adjacent side divided by hypotenuse.

Proof: Let the given triangle be ABC with ∠A = θ , ∠B = π2 , and

 = r. Fix the origin as A and the real axis as the line passing
hypotenuse |AC|
through A and B.

r
r sin (q )

q
A = (0,0) r cos (q ) B

Then by Euler’s formula, we have C = r exp(iθ ) = (cos(θ ) + i sin(θ )).

(Note that the polar form r exp(iθ) was used just as a symbol in Chapter 1
and by the previous corollary it is justified that it actually means in terms
of the exponential function. Therefore, we can use the Euler’s formula.)

Hence, AB = r cos(θ ) and |BC| = r sin(θ ). Therefore, sin(θ ) = |BC| =

|AC|
opposite side 
|AB| adjacent side
and cos(θ ) = = . 
hypotenuse 
|AC| hypotenuse

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 115 — #45

i i

Analytic Functions 115

Definition 2.4.27 Define the following:

sin(z)
1. tan(z) = , ∀z ∈ C \ {z : cos(z) = 0}.
cos(z)
1
2. csc(z) = , ∀z ∈ C \ {z : sin(z) = 0}.
sin(z)
1
3. sec(z) = , ∀z ∈ C \ {z : cos(z) = 0}.
cos(z)
1 cos(z)
4. cot(z) = = , ∀z ∈ C \ {z : sin(z) = 0}.
tan(z) sin(z)

Example 2.4.28 Find the domains of tan, csc, sec, and cot.
π
We have already proved that is the least positive real number at which cos
2
vanishes. See Result 2.4.15. Therefore, for any k ∈ Z,
 π   π 
π  exp i + kπ + exp −i + kπ
cos + kπ = 2 2
2 2
i exp (ikπ ) − i exp (−ikπ )
=
2
i(−1)k − i(−1)−k
=
2
= 0 (since ± 1 = (±1)−1 ).

Conversely, if z ∈ C such that cos(z) = 0, then sin(z) = ±1 from sin2 (z) +

cos2 (z) = 1. Hence, using Result 2.4.20, it follows that
 π  
3π
cos(z) + i sin(z) = ±i ⇒ exp(iz) = exp i or exp i
2 2
  π    π 
⇒ exp i z − = 1 or exp i z − 3 =1
2 2
π 3π
⇒ z − = 2nπ or z − = 2nπ, n ∈ Z
2 2
π π
⇒ z = (4n + 1) or z = (4n + 3) , n ∈ Z
2 2
π
⇒ z = (2(2n) + 1) or
2
π
z = (2(2n + 1) + 1) , n ∈ Z
2

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 116 — #46

i i

116 Exponential and Trigonometric Functions

π
⇒ z = (2k + 1) , k ∈ Z
2
π
⇒ z = kπ + , k ∈ Z.
2
Therefore,  
π
{z ∈ C : cos(z) = 0} = kπ + : k ∈ Z .
2
Similarly, we get
sin(z) = 0 ⇔ cos(z) = ±1
⇔ exp(iz) = ±1 = exp(0) or exp(iπ )
⇔ exp(iz) = 1 or exp(i(z − π )) = 1
⇔ z = 2nπ or z = 2nπ + π , n ∈ Z
⇔ z = kπ , k ∈ Z.
Therefore,
{z ∈ C : sin(z) = 0} = {kπ : k ∈ Z} .
 π 
Therefore, the domain of tan and sec is same as C \ kπ + : k ∈ Z and
2
that of cot and csc is C \ {kπ : k ∈ Z}.

THEOREM 2.4.29 lim tan(x) = +∞, lim tan(x) = −∞, and tan :
x→ π2 − x→− π2 +
 π π
− , → R is a bijection.
2 2

Proof: We know that

π   π
cos(0) = 1, cos = 0, and cos(x)  0, ∀x ∈ 0, .
2 2
 π π
Hence, cos(x) > 0, ∀x ∈ 0, because if cos(y) < 0 for some 0 < y < ,
2 2
then by intermediate value theorem, we get cos, which vanishes ata point
π
between 0 and y, which is a contradiction to the fact that cos  0 on 0, .
 π 2
As sin = cos > 0 on 0, , by Lemma 1.6.8, we have sin, which is strictly
 π 2  π
increasing on 0, . As sin(0) = 0, we have sin > 0 on 0, . Now
2 2
π
using the continuity of these functions at , we get limπ sin(x) = 1 and
2 x→ 2
limπ cos(x) = 0. Therefore, for a given M > 0, we choose δ1 > 0 and δ2 > 0
x→ 2
such that
 π 
 1 1 1
0 < x −  < δ1 ⇒ | sin(x) − 1| < ⇒ | sin(x)| > 1 − =
2 2 2 2

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 117 — #47

i i

Analytic Functions 117

and 
 π  1
0 < x −  < δ2 ⇒ | cos(x)| < .
2 2M
If δ = min{δ1 , δ2 , π2 }, then δ > 0, and
π π | sin(x)| sin(x) 1
−δ <x< ⇒ = > > M.
2 2 | cos(x)| cos(x) 2 cos(x)
Therefore, lim tan(x) = +∞.
z→ π2 −
π
Similarly, using the continuity of cos and sin at , we obtain that for a given
2
K < 0, there exists δ3 > 0 such that
 π 
 1 1
0 < x −  < δ3 ⇒ | cos(x)| < ⇒ 2 cos x <
2 −2K −k
1 1
⇒ > −k ⇒ < k.
2 cos x −2 cos x
π π
If r = min{δ1 , δ3 , π2 }, then r > 0, and − < x < − + r implies
2 2
| sin(x)| − sin(x) 1 sin(x) 1
= > ⇒ < <K
| cos(x)| cos(x) 2 cos(x) cos(x) −2 cos(x)
 π 
since sin < 0 and cos > 0 on − , 0 . Therefore, lim tan(x) = −∞.
2 z→− π2 +
sin(0) 0  π π
We know that tan(0) = = = 0 and is continuous on − , .
cos(0) 1  π 2 2
Given any y > 0, using lim tan(x) = +∞, we find t ∈ 0, such that
x→ π2 − 2
tan(t) > y. Then by using intermediate value theorem, there exists x ∈ (0, t)
such that tan(x) = y. Similarly, for a given y < 0, using lim tan(x), we can
x→− π +
 π   π π2 
find x ∈ − , 0 such that tan(x) = y. Thus, tan : − , → R is onto.
2 2 2
cos(x) cos(x) + sin(x) sin(x) 1
Since, tan (x) = 2 (x)
= 2
and cos(x)  0, ∀x ∈
 π π cos cos
 (x)
π π
− , , we get that tan is strictly increasing on − , . Therefore, tan :
 π 2 2
π
2 2
− , → R is one-to-one. Thus, the theorem follows. 
2 2

 π π  2.4.30 The inverse arctan of tan is continuous from R onto

THEOREM
− , .
2 2

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 118 — #48

i i

118 Exponential and Trigonometric Functions

 π π
Proof: Let x ∈ R be arbitrarily fixed. Since tan : − , → R is a
 π π 2 2
bijection, there exists unique t ∈ − , such that tan(t) = x. For a given
2 2  π π
> 0, we choose 1 > 0 such that 0 < 1 < min , − t, t + . As
 π π 2 2
tan : − , → R is continuous and strictly increasing, we obtain the
2 2
image of (t − 1 , t + 1 ) under tan is exactly equal to (tan(t − 1 ), tan(t + 1 ))
and it contains x = tan(t). Hence, we find δ > 0 such that

(x − δ, x + δ) ⊆ (tan(t − 1 ), tan(t + 1 )) = tan((t − 1, t + 1 ))

which implies that

arctan ((x − δ, x + δ)) ⊆ (t − 1, t + 1) = (arctan(x) − 1 , arctan(x) + 1 )

In other words, we have

|y − x| < δ ⇒ | arctan(y) − arctan(x)| < 1 <

As x ∈ R is arbitrary, the theorem follows. 

Exercise 2.4.31 Prove that 1 + tan2 (z) = sec2 (z), 1 + cot2 (z) = csc2 (z), and
tan(z) + tan(w)
tan(z + w) = , ∀z, w ∈ C.
1 − tan(z) tan(w)
π π
Exercise 2.4.32 Prove that arctan x → as x → +∞ and arctan x → −
2 2
as x → −∞.

Exercise 2.4.33

1. Prove that tan(−z) = − tan(z), cot(z) = − cot(z), sec(−z) = sec(z),

and csc(−z) = − csc(z).

2. Prove the following identities:

π
(a) tan z − 2 = − cot(z)
π
(b) cot z − 2 = − tan(z)
π
(c) tan z + 2 = − cot(z)
π
(d) cot z + 2 = − tan(z)

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 119 — #49

i i

Analytic Functions 119

π
(e) sec z − 2 = csc(z)
π
(f) csc z − 2 = − sec(z)
π
(g) sec z + 2 = − csc(z)
π
(h) csc z + 2 = sec(z)

(i) tan (z − π ) = tan(z)

(j) cot (z − π ) = cot(z)
(k) tan (z + π ) = tan(z)
(l) cot (z + π ) = cot(z)
(m) sec (z − π ) = − sec(z)
(n) csc (z − π ) = − csc(z)
(o) sec (z + π ) = − sec(z)
(p) csc (z + π ) = − csc(z).

Hint. Use Exercise 2.4.18.

2.5 HYPERBOLIC FUNCTIONS

Definition 2.5.1 Define
exp(z) + exp(−z) exp(z) − exp(−z)
cosh(z) = and sinh(z) = , ∀z ∈ C.
2 2

LEMMA 2.5.2 For every z ∈ C, we have cosh(z) + sinh(z) = exp(z).

Proof: For each z ∈ C, we get

exp(z) + exp(−z) exp(z) − exp(−z)
cosh(z) + sinh(z) = +
2 2
2 exp(z)
= = exp(z).
2
Hence, the lemma follows. 

LEMMA 2.5.3 cosh2 (z) − sinh2 (z) = 1, ∀z ∈ C.

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 120 — #50

i i

120 Hyperbolic Functions

Proof: For z ∈ C, we have

cosh2 (z) − sinh2 (z)
(exp(z) + exp(−z))2 (exp(z) − exp(−z))2
= −
4 4
(exp(z)) + (exp(−z)) + 2 − (exp(z))2 − (exp(−z))2 + 2
2 2
=
4
4
= = 1. 
4

LEMMA 2.5.4 sinh and cosh are differentiable and sinh = cosh and cosh =
sinh.

Proof: As exp is a differentiable function on C, cosh and sinh are differen-

tiable functions on C and
 
d exp(z) + exp(−z) exp(z) − exp(−z)
cosh (z) = = = sinh(z)
dz 2 2
 
 d exp(z) − exp(−z) exp(z) + exp(−z)
sinh (z) = = = cosh(z).
dz 2 2
Hence, the lemma follows. 

Exercise 2.5.5 Prove that the power series representations of cosh(z) and
∞ z2n ∞ z2n+1
sinh(z) are and .
n=0 (2n)! n=0 (2n + 1)!

Exercise 2.5.6 For every z, w ∈ C, prove the following

1. sinh(−z) = − sinh(z).

2. cosh(−z) = cosh(z).

3. sinh(z + w) = sinh(z) cosh(w) + sinh(z) cosh(w).

4. cosh(z + w) = cosh(z) cosh(w) + sinh(z) sinh(w).

5. sinh(z − w) = sinh(z) cosh(w) − sinh(z) cosh(w).

6. cosh(z − w) = cosh(z) cosh(w) − sinh(z) sinh(w).

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 121 — #51

i i

Analytic Functions 121

Definition 2.5.7 Define the following:

sinh(z)
1. tanh(z) = , ∀z ∈ {z ∈ C : cosh(z)  0}.
cosh(z)
1
2. csch (z) = , ∀z ∈ {z ∈ C : sinh(z)  0}.
sinh(z)
1
3. sech(z) = , ∀z ∈ {z ∈ C : cosh(z)  0}.
cosh(z)
1
4. coth(z) = , ∀z ∈ {z ∈ C : sinh(z)  0}.
tanh(z)

LEMMA 2.5.8 For z, w ∈ {z ∈ C : cosh(z)  0},

tanh(z) + tanh(w) tanh(z) − tanh(w)

tanh(z + w) = and tanh(z − w) = .
1 + tanh(z) tanh(w) 1 − tanh(z) tanh(w)

Proof: As
sinh(z + w) sinh(z) cosh(w) + sinh(z) cosh(w)
tanh(z + w) = = ,
cosh(z + w) cosh(z) cosh(w) + sinh(z) sinh(w)

dividing the numerator and denominator by cosh(z) cosh(w), we get

sinh(z) sinh(w)
+
cosh(z) cosh(w) tanh(z) + tanh(w)
tanh(z + w) = = . (2.8)
sinh(z) sinh(w) 1 + tanh(z) tanh(w)
1+
cosh(z) cosh(w)

Since
sinh(−w) sinh(w)
tanh(−w) = =− = − tanh(w),
cosh(−w) cosh(w)
by replacing w by −w in equation (2.8), we get

tanh(z) − tanh(w)
tanh(z − w) =
1 − tanh(z) tanh(w)

Exercise 2.5.9 Prove that for every z ∈ C, 1 − tanh2 (z) = sech2 (z) and
coth2 (z) − 1 = csch2 (z).

i i

i i
 i i

“book” — 2014/6/4 — 20:46 — page 122 — #52

i i

122 Hyperbolic Functions

Finally, we present the relation between the trigonometric functions and

hyperbolic functions.

RESULT 2.5.10 For every z ∈ C, sin(iz) = i sinh(z) and cos(iz) = cosh(z).

The proof of the above result is easy, and hence, it is left to the reader as
a simple exercise.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 123 — #1

i i

3
Rational Functions
and Multivalued
Functions

3.1 POLYNOMIALS AND RATIONAL FUNCTIONS

Definition 3.1.1 A polynomial is a function P : C → C, which is of the form
P(z) = a0 + a1 z + a2 z2 + · · · + an zn , ∀z ∈ C, where ai ∈ C, ∀i = 1, 2, . . . , n
and n ∈ N.

1. If n is the largest positive integer such that an  0, then we call n

the degree of the polynomial P. We define the degree of a constant
polynomial by 0. We denote the degree of P by deg P.
2. If z0 ∈ C such that P(z0 ) = 0, then we call z0 a zero of P.

Now we state (without proof) a particular version of the Euclidean algorithm,

which is an useful result from algebra.

RESULT 3.1.2 (Division algorithm for polynomials)

Let P and Q be polynomials with complex coefficients. Then there exist
unique polynomials A and B such that P = AQ + B with either B = 0 or
deg B < deg Q.

LEMMA 3.1.3 If z0 is a zero of P, then we can write P(z) = (z − z0 )Q(z),

∀z ∈ C for some polynomial Q.

123

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 124 — #2

i i

124 Polynomials and Rational Functions

Proof: By division algorithm, we can find polynomials Q(z) and R(z) such
that
P(z) = (z − z0 )Q(z) + R(z), ∀z ∈ C
with 0 < deg R(z) < deg (z − z0 ) or R(z) = 0, ∀z ∈ C. As deg (z − z0 ) is 1,
R(z) must be a constant and is equal to

R(z0 ) = P(z0 ) − (z0 − z0 )Q(z0 ) = 0.

Therefore, the lemma follows. 

Definition 3.1.4 If m ∈ N is such that

P(z) = (z − z0 )m Q(z), ∀z ∈ C

for some polynomial Q such that Q(z0 )  0, then m is called the order of the
zero z0 of P.

RESULT 3.1.5 If m is the order of a zero of a non-constant polynomial P,

then P(k) (z0 ) = 0, ∀k ∈ {1, 2, 3, . . . , m − 1} and P(m) (z0 )  0.

Proof: Let z0 be a zero of order m, then by definition, we have

P(z) = (z − z0 )m Q(z), ∀z ∈ C

for some polynomial Q. Using the Leibniz rule(Result 2.1.13), we have

dk   k
dj d k−j
k
(z − z0 ) m
Q(z) = kCj j (z − z0 )m · k−j Q(z), ∀k ∈ N.
dz dz dz
j=0

For 1 ≤ k ≤ m − 1 and 0 ≤ j ≤ k , we have

dj
(z − z0 )m = m(m − 1) · · · (m − j + 1)(z − z0 )m−j and m − j > 0
dzj
and hence, we get P(k) (z0 ) = 0, 1 ≤ k ≤ m − 1. Furthermore, using the same
observation, we obtain
 m 
d
P (z0 ) =
(m)
(z − z0 ) Q(z) = m!Q(z0 )  0.
m
dzm
Thus, the result follows. 
Converse of the above result is also true and is given by the following result.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 125 — #3

i i

Rational Functions and Multivalued Functions 125

RESULT 3.1.6 Let P be a non-constant polynomial and z0 ∈ C. If m ∈ N is

such that P(k) (z0 ) = 0, ∀k ∈ {1, 2, 3 . . . , m − 1} and P(m) (z0 )  0, then the
order of zero of P at z0 is m.

Proof: If m = 1, then using Lemma 3.1.3, we have

P(z) = (z − z0 )P1 (z), ∀z ∈ C

for some polynomial P1 . Using

P (z) = (z − z0 )P1 (z) + P1 (z), ∀z ∈ C,

we get 0  P (z0 ) = P1 (z0 ).

If m = 2, using P(z0 ) = 0, then we write

P(z) = (z − z0 )P1 (z), ∀z ∈ C

and we get 0 = P (z0 ) = P1 (z0 ). Therefore,

P1 (z) = (z − z0 )P2 (z), ∀z ∈ C

for some polynomial P2 , and hence,

P(z) = (z − z0 )2 P2 (z), ∀z ∈ C.

Using

P (z) = (z − z0 )2 P2 (z) + 4(z − z0 )P2 (z) + 2P2 (z), ∀z ∈ C,

we get 0  P (z0 ) = P2 (z0 ).

Proceeding further at the (m − 1)st stage using P(m−1) (z0 ) = 0, we get

P(z) = (z − z0 )m Pm (z), ∀z ∈ C

for some polynomial Pm . Therefore, for each z ∈ C, we have


m j (m−j)
P(m) (z) = d
mCj dz j [(z − z0 ) ] Pm
m (z)
j=0 
m j−1
(m−j)
= mCj (m − ν) (z − z0 )m−j Pm (z).
j=0 ν=0

Thus, 0  P(m) (z0 ) = m!Pm (z0 ) ⇒ Pm (z0 )  0. Therefore, m is the order of

zero at z0 for P. 

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 126 — #4

i i

126 Polynomials and Rational Functions

Remark 3.1.7: Another simple proof of the previous result is given in

Corollary 6.1.4.

Next we discuss a theorem that states a relation between the zeroes of a poly-
nomial P with the zeroes of its derivative. For that purpose, now we discuss
how to represent a half plane described by a straight line. We know that the
equation of a straight line, which has the slope as the argument of a non-zero
complex number b and it passes through the complex number a, is given by
z = a + tb, t ∈ R. Hence, we have

z−a
z lies on the straight line iff is a real number
b
z−a
iff Im = 0.
b
z−a
Therefore, z is not on the straight line iff Im  0. Hence,
b
z−a z−a
z ∈ C : Im >0 and z ∈ C : Im <0
b b
are the two half planes determined by the straight line z = a + tb, t ∈ R.

THEOREM 3.1.8 (Lucas’ theorem)

If every zero of a polynomial P is enclosed by a closed polygon K in C, then
every zero of P is also enclosed by K .

Proof: Let the closed polygon K is formed by the straight lines z = aj +

tbj , t ∈ R, j = 1, 2, . . . , m. That is, the region enclosed by the given polygon
is the intersection of m half planes each of which is determined by one of
these m straight lines z = aj + tbj , t ∈ R.

Closed polygon

Therefore, so to conclude this theorem, we shall show that every zero of P

lies in a half plane described by z = aj + tbj , t ∈ R and every zero of P also
lies in the same half plane for all j = 1, 2, . . . , m. If α1 , α2 , α3 , . . . , αn ∈ C

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 127 — #5

i i

Rational Functions and Multivalued Functions 127

are the zeroes of P including multiplicities. Then, using Lemma 3.1.3 and
fundamental theorem of algebra (Theorem 4.3.12), we get

P(z) = c(z − α1 )(z − α2 ) · · · (z − αn ), ∀z ∈ C and for some c ∈ C.

d
Therefore, either by direct calculation or by finding log P(z) (with less
dz
rigor), we get

P (z) 1 1 1
= + + ··· + , ∀z ∈ C. (3.1)
P(z) z − α1 z − α2 z − αn

z − aj
Let 1 ≤ j ≤ m and let Hj = z ∈ C : Im > 0 be a half plane
bj
described by the straight line z = aj + tbj , t ∈ R. Now assume that αk ∈ Hj ,
∀k ∈ {1, 2, . . . , n}. Therefore, we have

αk − aj
Im > 0, ∀1 ≤ k ≤ n.
bj

z0 − aj
If z0  Hj , then Im ≤ 0, and hence,
bj

z0 − αk z0 − aj αk − aj
Im = Im − Im < 0.
bj bj bj

As (Im w) × (Im w−1 ) < 0, for any complex number w with Im w  0,

bj
we have Im > 0. Therefore,
z0 − αk

bj P (z) 
n
bj
Im = Im > 0 ⇒ bj P (z0 )  0 ⇒ P (z0 )  0.
P(z) z0 − bk
k=1

Thus, if P (z0 ) = 0, then z0 ∈ Hj . Therefore, every zero of P lies in Hj . As j

is arbitrary, every zero of P lies in the closed polygon K . 

Definition 3.1.9 A function R : C∞ → C∞ is called a rational function if it

is a quotient of two polynomials.
Usually, we write a rational function in the reduced form in the sense that
P
if R = , then P and Q are polynomials without common zeroes.
Q

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 128 — #6

i i

128 Polynomials and Rational Functions

LEMMA 3.1.10 If R is any rational function, then lim R(z) exists in C∞ .

z→∞

P
Proof: Let R = , where
Q

m 
n
P(z) = ak z and Q(z) =
k
bj z j , ∀z ∈ C
k=0 j=0

with am  0 and bn  0. We also assume that P and Q have no common

zeroes. Then, we have
m
ak zk
k=0
lim R(z) = lim n
z→∞ z→∞ 
bj zj
j=0

m
ak zk−m
k=0
= lim zm−n n
z→∞ 
bj zj−n
j=0
m−n am
= lim z
z→∞ bn
⎧ an
⎨ bm if n = m
= 0 if n > m
⎩
∞ if n < m.

Definition 3.1.11 The value of a rational function R at ∞ is defined by

lim R(z).
z→∞

Definition 3.1.12 Let R be a rational function and z0 ∈ C∞ , then

1. z0 is said to be a zero of R if R(z0 ) = 0.
2. z0 is said to be a pole of R if R(z0 ) = ∞.

Note that if R = Q P
is written in the reduced form, then every zero of P is a
finite zero of R and every zero of Q is a finite pole of R. Furthermore, ∞ is a
zero of R if deg P < deg Q and is a pole if deg P > deg Q.

P
Definition 3.1.13 Let R = be a rational function, z0 ∈ C∞ be a zero of R,
Q
and k ∈ N.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 129 — #7

i i

Rational Functions and Multivalued Functions 129

1. If z0 ∈ C, then we say that R has a zero of order k at z0 if z0 is a zero

of order k for the polynomial P; and we say that R has a pole of order
k at z0 if z0 is a zero of order k for the polynomial Q.
2. We say that R has a zero of order k at ∞ if 0 is a zero of order k for the
rational function R1 ; and we say that R has a pole of order k at ∞  if0
is a zero of order k for the rational function R1 , where R1 (z) = R 1z .

P
Note that if R = , then R has a zero of order k at ∞ if deg P = k + deg Q
Q
and has a pole of order k at ∞ if deg Q = k + deg P.

Remark 3.1.14: Let zj , j = 1, 2, . . . , n be the distinct zeroes (poles) of a

rational function R. If kj is the order of the zero (pole) at zj , ∀j = 1, 2, . . . , n,
then we mean the number of zeroes (poles) of R including multiplicities by
n
kj .
j=1

RESULT 3.1.15 Let R = Q P

be in the reduced form, where P and Q are poly-
nomials of degree m and n, respectively. If μ is the number of zeroes of R
including multiplicities in C∞ and ν is the number of poles of R including
multiplicities in C∞ , then μ = ν = max{m, n}.

Proof: Obviously, there are m finite zeroes of R including multiplicities and

n finite poles of R including multiplicities.
Case 1: m = n. In this case, ∞ is neither a pole nor a zero for R. Therefore,
μ = m = n = ν.
Case 2: m > n. In this case, it is easy to verify that ∞ is a pole for R of order
m − n. Therefore, μ = m and ν = n + m − n = m.
Case 3: m < n. In this case, it is easy to verify that ∞ is a zero for R of order
n − m. Therefore, μ = m + n − m = n and ν = n.
From the above discussions, we get μ = ν = max{m, n}. 

LEMMA 3.1.16 Let R be a rational function. R is a polynomial iff the only

possible pole of R is ∞.

Proof: Let R be a polynomial. If R is a constant polynomial, then it has no

poles. If R is a non-constant polynomial, then obviously R(∞) = ∞, and

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 130 — #8

i i

130 Polynomials and Rational Functions

hence, ∞ is a pole of R. For every c ∈ C, as lim R(z) = R(c)  ∞, it has

z→c
P
no other poles. Conversely, assume that R has only one pole at ∞. If R = ,
Q
where P and Q are polynomials without common zeroes. If Q has a zero z0
in C, then z0 becomes a pole for R, which is a contradiction. Therefore, Q
must not have a zero in C, and hence, by the fundamental theorem of algebra
(Theorem 4.3.12), we get that Q is a constant. Thus, R is a polynomial. 

Definition 3.1.17 Let z1 , z2 , . . . , zn be the distinct finite poles of a rational

function R, then the expression


n  
1
R(z) = G(z) + Fj , ∀z ∈ C
z − zj
j=1

is called the partial fraction expansion of R if G is a polynomial and Fj are

polynomials with Fj (0) = 0.

THEOREM 3.1.18 Every rational function has partial fraction expansion.

P
Proof: Let R = be in reduced form. Hence, by division algorithm for
Q
polynomials (Result 3.1.2), we can write P = AQ + B, where A and B are
B
polynomials such that either B = 0 or deg B < deg Q. Therefore, R = A + .
Q
B
Let C = A − A(0) and D = A(0) + . Hence, we can write R = C + D,
Q
where C is a polynomial with C(0) = 0 and D is a rational function such that
A(0)Q + B
D(∞)  ∞, as D = and degree of (A(0)Q + B) ≤ degree of Q.
Q
We note that if R(∞) is finite, then C = 0, and if R(∞) = ∞, then degree of
P > degree of Q ⇒ C is a non-constant polynomial. . . . . . . . . . . . . (I)
 
For every j = 1, 2, . . . , n, let Rj (w) = R zj + w1 , ∀w ∈ C∞ , where
z1 , z2 , . . . , zn are distinct finite poles of R. Note that Rj (∞) = ∞; then by
previous argument, there exist non-constant polynomial Cj with Cj (0) = 0
and rational function Dj with Dj (∞)  ∞ such that

Rj (w) = Cj (w) + Dj (w), ∀w ∈ C∞ , ∀j = 1, 2, . . . , n.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 131 — #9

i i

Rational Functions and Multivalued Functions 131

1
Applying the change of variable w = , we get
z − zj
 
R(z) = R zj + w1
= Rj (w)
 + Dj(w)
= Cj (w)  
1 1
= Cj + Dj
z − zj z − zj
= Ej (z) + Fj (z), (say) ∀z ∈ C∞
and Fj (zj )  ∞, j = 1, 2, . . . , n. Now the rational function

n
R−C− Ej (3.2)
j=1

has possible poles from {z1 , z2 , . . . , zn , ∞}.

Note that
1. R(zj ) = ∞,
2. from Rj (∞) = ∞, by the observation (I ), we get Cj is a non-constant
polynomial, and hence, Cj (∞) = ∞ ⇒ Ej (zj ) = Cj (∞) = ∞.
Although R(zj ) = ∞ and Ej (zj ) = ∞, as R(zj ) − Ej (zj ) = Fj (zj )  ∞, it
follows that zj is not a pole for the rational function in equation(3.2). Next
from the following statements:
1. If R(∞) = ∞, then using (I ), we get C(∞) = ∞, but as
(R − C)(∞) = D(∞)  ∞,
we get that ∞ is not a pole for R.
2. If R(∞)  ∞, then using (I ), we get C = 0, and hence,
(R − C)(∞)  ∞.
we conclude that ∞ is not a pole for equation(3.2).
n
Hence, R − C − Ej has no poles in C∞ . Therefore, by Result 3.1.15,
j=1

n
we get R − C − Ej is a constant say α . Hence,
j=1


n 
n  
1
R = (C + α) + Ej ⇒ R(z) = G(z) + Fj , ∀z ∈ C
z − zj
j=1 j=1

where G = C + α . 

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 132 — #10

i i

132 Linear Fractional Transforms

3.2 LINEAR FRACTIONAL TRANSFORMS

Definition 3.2.1 A linear fractional transform (or Möbius transfrom) T is of
az + b
the form T(z) = , where a, b, c, and d ∈ C such that ad − bc  0.
cz + d
The condition ad − bc  0 states that az + b and cz + d do not have a common
−b −d −b
zero. For if ad − bc = 0, then = . However, as is the zero of
a c a
−d
az + b and is the zero of cz + d , it would follow that az + b and cz + d
c
have a common zero.

Definition 3.2.2 (Elementary linear fractional transforms)

1. Translation: T(z) = z + a, ∀z ∈ C∞ for some a ∈ C.

2. Rotation: T(z) = eiθ z, ∀z ∈ C∞ for some θ ∈ [0, 2π).

3. Scaling: T(z) = rz, ∀z ∈ C∞ for some r > 0.

(a) If r > 1, then this transform is called magnification.

(b) If r < 1, then this transform is called contraction.
1
4. Inversion: T(z) = , ∀z ∈ C∞ .
z

LEMMA 3.2.3 Every linear fractional transform can be written as a suitable

composition of these elementary linear fractional transforms.
az + b
Proof: Let T(z) = be given.
cz + d
a b
Case 1: If c = 0, then T(z) = z + . Here, we note that a  0 and d  0
d d
a
from ad − bc  0. Let = reiθ . Therefore, T = T1 ◦ T2 ◦ T3 , where
d
b
T3 (z) = rz, T2 (z) = eiθ z, and T1 (z) = z + .
d
Case 2: If c  0, then
a b
z+
T(z) = c c
c
z+
d

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 133 — #11

i i

Rational Functions and Multivalued Functions 133

 
a d b ad
z+ + − 2
c c c c
=
d
z+
 c
a bc − ad 1
= + .
c c2 d
z+
c
bc − ad bc − ad
As 2
 0, we can write = ρeiφ . Therefore, T =
c c2
d 1
T1 ◦ T2 ◦ T3 ◦ T4 ◦ T5 , where T5 (z) = z + , T4 (z) = , T3 (z) = eiφ z,
c z
a
T2 (z) = ρz, and T1 (z) = z + . 
c

THEOREM 3.2.4 Every linear fractional transform maps circles and lines to
circles and lines.

Proof: To prove this theorem, we shall first show that the elementary linear
fractional transforms map circles and lines to circles and lines.
1. The translation operator w = z + c maps the line z = a + tb into the
line w = (a + c) + tb, and it maps the circle z = a + reit into the circle
w = (a + c) + reit . Geometrically,

(a) the line passing through a with slope m is mapped into the line
passing through a + c with slope m.
(b) the circle with centre a and radius r is mapped to the circle with
centre a + c and radius r.

2. The rotation operator w = eiθ z maps the line z = a + tb into the line
w = (eiθ a) + t(eiθ b), and it maps the circle z = a + reit into the circle
z = (eiθ a) + reit . Geometrically,

(a) the line passing through a with slope m is mapped to the line
passing through eiθ a with slope m + θ .
(b) the circle with centre a and radius r is mapped to the circle with
centre eiθ a and radius r.

3. The scaling operator ρz maps the line z = a + tb into the line

z = (ρa) + tb, and it maps the circle z = a + reit into the circle
z = (ρa) + rρeit . Geometrically,

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 134 — #12

i i

134 Linear Fractional Transforms

(a) the line passing through a with slope m is mapped to the line
passing through ρa with slope m.

(b) the circle with centre a and radius r is mapped to the circle with
centre ρa and radius rρ .

4. The combined equation of a line and circle (with non-zero centre) is

azz + bz + bz + c = 0, whose image under the inversion operator
1
w = is cww + bw + bw + a = 0, which is again a combined equation
z
of a line and circle (with non-zero centre). As azz + bz + bz + c = 0
represents a circle (with non-zero centre) if a  0 and represents a line
if a = 0 and it passes through c, geometrically we have the following
statements:

(a) a line passing through 0 is mapped onto a line passing through 0.

(b) a line not passing through 0 is mapped onto a circle (with non-zero
centre) passing through 0.

(c) a circle (with non-zero centre) passing through 0 is mapped onto a

line not passing through 0.

(d) a circle (with non-zero centre) not passing through 0 is mapped

onto a circle (with non-zero centre) not passing through 0.

If we consider a circle with centre zero radius r > 0, then it is mapped

1
to a circle with centre zero and radius , as
r

1
z = r exp(iθ ), θ ∈ [−π, π] ⇒ w = exp(−iθ ), θ ∈ [−π , π ].
r

As every linear fractional transform is a composition of these elementary

linear fractional transforms, the theorem follows. 

RESULT 3.2.5 Let T and S be linear fractional transforms, then

1. T ◦ S is a linear fractional transform.

2. T −1 exists, and T −1 is also a linear fractional transform.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 135 — #13

i i

Rational Functions and Multivalued Functions 135

a1 z + b1 a2 z + b2
Proof: Let T(z) = and S(z) = , where ai , bi , ci , and
c1 z + d1 c2 z + d2
di ∈ C with ai di − bi ci  0, for i = 1, 2. As
 
a2 z + b2
a1 + b1
c2 z + d2 (a1 a2 + b1 c2 )z + (a1 b2 + b1 d2 )
(T ◦ S)(z) =   = ,
a2 z + b2 (c1 a2 + c2 d1 )z + (c1 b2 + d1 d2 )
c1 + d1
c2 z + d2
Az + B
which is of the form , and
Cz + D
AD − BC = (a1 a2 + b1 c2 )(c1 b2 + d1 d2 ) − (a1 b2 + b1 d2 )(c1 a2 + c2 d1 )
= a1 a2 c1 b2 + a1 a2 d1 d2 + b1 c2 c1 b2 + b1 c2 d1 d2
− a1 b2 c1 a2 − a1 b2 c2 d1 − b1 d2 c1 a2 − b1 d2 c2 d1
= a1 a2 d1 d2 + b1 c2 c1 b2 − a1 b2 c2 d1 − b1 d2 c1 a2
= a1 d1 (a2 d2 − b2 c2 ) + b1 c1 (b2 c2 − a2 d2 )
= (a1 d1 − b1 c1 )(a2 d2 − b2 c2 )  0.

Therefore, T ◦ S is a linear fractional transform.

az + b
The equation w = implies that
cz + d
dw − b
w(cz + d) = az + b ⇒ z(cw − a) = −dw + b ⇒ z =
−cw + a
which is denoted by T −1 (w). Clearly, T −1 is a linear fractional transform as
da − (−b)(−c)  0. 
From the previous result, we get the collection of linear fractional trans-
forms, which becomes a group with respect to composition ‘◦’ of functions,
as the other axioms of group are obvious. It is not an abelian group because
1
if T(z) = 2z and S(z) = , ∀z ∈ C∞ , then T ◦ S  S ◦ T .
z

THEOREM 3.2.6 If T is a linear fractional transform such that T is not

identity, then T has at most two fixed points in C.

az + b
Proof: Let T(z) = , ∀z ∈ C∞ with ad − bc  0. The fixed points of
cz + d
az + b
T are the solutions of z = ⇔ cz2 + (d − a)z − b = 0.
cz + d
If c  0, then the equation has at most two solutions.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 136 — #14

i i

136 Linear Fractional Transforms

If c = 0 and d − a  0, then from ad − bc  0, we get a  0 and d  0.

Hence,
a  b a
T(z) − z = − 1 z + with 0   1.
d d d
 −1 b
Therefore, z = 1 − da d is the only fixed point of T in C.
If c = 0, d − a = 0, and b  0, then d  0, and hence, T(z) = z + db with
d  0, which has no fixed point.
b

The case c = 0, d − a = 0, and b = 0 is not possible as T is not the identity

transformation. Thus, the theorem follows. 

Example 3.2.7 If T : C∞ → C∞ is a bijection such that T maps lines and

circles to lines and circles, then T need not be a linear fractional transform.
If T(z) = z, ∀z ∈ C∞ , then T is a bijection from C∞ onto itself and T maps
the line (circle) a|z|2 +bz+bz+c = 0 onto the line (circle) a|z|2 +bz+bz+c =
0. By previous theorem, T is not a linear fractional transform, as it has every
real number as a fixed point.

THEOREM 3.2.8 Every linear fractional transform w = T(z), which has only
w−α z−α
two distinct fixed points α, β ∈ C, satisfies the equation =γ
w−β z−β
for some γ ∈ C.

az + b
Proof: Let w = T(z) = , ∀z ∈ C∞ . As α and β are the fixed points of
cz + d
aα + b aβ + b
T , we have α = and β = . Therefore,
cα + d cβ + d
az + b aα + b (ad − bc)(z − α)
w−α = − =
cz + d cα + d (cz + d)(cα + d)
az + b aβ + b (ad − bc)(z − β)
w−β = − =
cz + d cβ + d (cz + d)(cβ + d)

Therefore,
w−α z−α
=γ
w−β z−β

cβ + d
where γ = . 
cα + d

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 137 — #15

i i

Rational Functions and Multivalued Functions 137

THEOREM 3.2.9 Every linear fractional transform w = T(z), which has only
1 1
one fixed point α ∈ C, satisfies the equation = + γ for some
w−α z−α
γ ∈ C.

az + b
Proof: As in the proof of previous theorem, if w = T(z) = , ∀z ∈ C∞ ,
cz + d
then we have
aα + b
α= ⇒ α(cα − a) = b − dα.
cα + d
Since,
az + b
w−α = −α
cz + d
az + b − αcz − αd
=
cz + d
(a − cα)z + α(cα − a)
=
cz + d
(a − cα)(z − α)
=
cz + d
we have
 
1 1 1 cz + d 1 cz + d − a + cα
− = −1 = .
w−α z−α z−α a − cα z−α a − cα

As, α is the double root of the quadratic equation, cz2 + (d − a)z − b = 0, we

a−d
have 2α = ⇒ a − d = 2αc. Therefore,
c
1 1 1 cz − 2αc + cα
− =
w−α z−α z−α a − cα
1 c(z − α)
=
z − α a − cα
c
=
a − cα
= γ (say)
1 1
Hence, = + γ. 
w−α z−α

PROBLEM 3.2.10 Find the image of |z − 2| < 2 under the linear fractional
transform T(z) = 2z−8
z
.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 138 — #16

i i

138 Linear Fractional Transforms

Solution: As the point 4 lies on the circle |z − 2| = 2 and T(4) = ∞, the

image of the circle is a straight line. We see that 0 and 2 + 2i lie on the circle
−i
and f (0) = 0, f (2+2i) = 2+2i
4i−4 = 2 . Hence, the image of the circle |z−2| = 2
is the imaginary axis. As the centre 2 of this circle is mapped to T(2) = −1 2 ,
which lies in the left half plane, the image of |z − 2| < 2 under the linear
fractional transform T(z) = 2z−8 z
is the left half plane {z ∈ C : Re z < 0}.

PROBLEM 3.2.11 Determine the image of {z : Re z < 0 and Im z > 0} under

the linear fractional transform S(z) = z+i
z−i .

Solution: We first find the images of the real and imaginary axes under z+iz−i .
 
 x+i 
As |T(x)| =  x−i  = 1, for all x ∈ R, the image of the real axis is the
unit circle. As T(i) = ∞, the image of the imaginary axis under the linear
fractional transform is a straight line. Seeing T(2i) = 3 and T(3i) = 2, we
conclude that the imaginary axis is mapped onto the real axis. As −1 + i lies
on the second quadrant {z : Re z < 0 and Im z > 0} and T(−1 + i) = 1 − 2i
(which lies outside the circle and below the real axis), we conclude that the
image of the second quadrant under z+iz−i is {z ∈ C : |z| > 1 and Im z < 0}.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 139 — #17

i i

Rational Functions and Multivalued Functions 139

Definition 3.2.12 For given four distinct extended complex numbers z1 , z2 ,

z3 , and z4 , we define the cross ratio (z1 , z2 , z3 , z4 ) by T(z1 ), where T is a linear
fractional transform that maps z2 , z3 , and z4 to 1, 0, and ∞, respectively. More
explicitly,

⎧
⎪ z 1 − z3 z2 − z4
⎪
⎪ × if z1 , z2 , z3 , z4 ∈ C
⎪
⎪ z − z z 2 − z3
⎪
⎪
1 4
⎪
⎪ z2 − z 4
⎪
⎪
⎪
⎪ if z1 = ∞, z2 , z3 , z4 ∈ C
⎪
⎪ z 2 − z3
⎪
⎪
⎪
⎨ z1 − z3
(z1 , z2 , z3 , z4 ) = if z2 = ∞, z1 , z3 , z4 ∈ C
⎪
⎪ z 1 − z4
⎪
⎪
⎪
⎪ z2 − z4
⎪
⎪
⎪
⎪ if z3 = ∞, z1 , z2 , z4 ∈ C
⎪
⎪ z 1 − z4
⎪
⎪
⎪
⎪
⎪
⎪ z1 − z3
⎩ if z4 = ∞, z1 , z2 , z3 ∈ C.
z 2 − z3

Example 3.2.13 Evaluate the following cross ratios:

1. (1 + i, 2 − i, 3, −i).

2. (2 + i, i, 5 − i2, ∞).

3. (1 + i2, 2 + i3, ∞, i).

4. (2, ∞, 1 − i, 3 + i).

5. (∞, 1 − i, 1 + i, i).

Solution:
1+i−3 2−i+i
1. (1 + i, 2 − i, 3, −i) = ×
1+i+i 2−i−3

−4 + i2
= = −1 − i.
1 − i3

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 140 — #18

i i

140 Linear Fractional Transforms

2 + i − 5 + i2
2. (2 + i, i, 5 − i2, ∞) =
i − 5 + i2

−3 − i3 3
= = (4 − i).
−5 + i3 17

2 + i3 − 1
3. (1 + i2, 2 + i3, ∞, 1) =
1 + i2 − 1

1 + i3 1
= = (3 − i).
i2 2
2−1+i
4. (2, ∞, 1 − i, 3 + i) =
2−3−i
1+i
= = −1.
−1 − i
1−i−i
5. (∞, 1 − i, 1 + i, i) =
1−i−1−i
1 − i2 1
= = (2 + i).
−i2 2

Exercise 3.2.14 Find the following cross ratios:

1. (2 + i3, 1 + i, 3 − i, 1 − i4).

2. (1 + i3, 5 + i, 2 − i3, ∞).

3. (3 + i2, 1 − i2, ∞, i5).

4. (1 − i2, ∞, 1 + i4, 2 + i).

5. (∞, 2 − i, 3 + i4, i3).

Answers: 1. 19
20 − i 15 , 2. 25 (21 + i22),
1
3. 4
3 − i, 4. 15 (9 + i3), 5. 13 (9 + i7).
1

RESULT 3.2.15 Let z1 , z2 , z3 , and z4 be extended complex numbers, then

1. (z1 , z2 , z3 , z4 ) = (z1 , z2 , z3 , z4 ).

2. (S(z1 ), S(z2 ), S(z3 ), S(z4 )) = (z1 , z2 , z3 , z4 ) for every linear fractional

transform S .

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 141 — #19

i i

Rational Functions and Multivalued Functions 141

Proof: Let T be the linear fractional transform that maps z2 , z3 , and z4 to 1, 0,

and ∞, respectively.
az + b az + b
1. If T(z) = and if T1 (z) = , then T1 (z) = T(z) for every
cz + d cz + d
extended complex number z. Hence,

T1 (z2 ) = T(z2 ) = 1 = 1,
T1 (z3 ) = T(z3 ) = 0 = 0,
T1 (z4 ) = T(z4 ) = ∞ = ∞.

Therefore, T1 is the linear fractional transform that maps z2 , z3 , and

z4 to 1, 0, and ∞, respectively. Therefore, (z1 , z2 , z3 , z4 ) = T1 (z1 ) =
T(z1 ) = (z1 , z2 , z3 , z4 ).
2. Using Result 3.2.5, we get T ◦ S −1 , which is a linear fractional trans-
form. Clearly, we have T ◦ S −1 maps Sz2 , Sz3 , and Sz4 to 1, 0, and ∞,
respectively. Therefore, (S(z1 ), S(z2 ), S(z3 ), S(z4 )) = (T ◦S −1 )(S(z1 )) =
T(z1 ) = (z1 , z2 , z3 , z4 ). 

THEOREM 3.2.16 Let z1 , z2 , z3 , and z4 be distinct extended complex num-

bers, then z1 , z2 , z3 , and z4 lie on a circle or on a straight line iff (z1 , z2 , z3 , z4 )
is real.

Proof: Let T be the linear fractional transform that maps z2 , z3 , and z4 to

1, 0, and ∞, respectively, then T(z1 ) = (z1 , z2 , z3 , and z4 ). Assume that
(z1 , z2 , z3 , z4 ) is real. Then, T(z1 ), T(z2 ) = 1, T(z3 ) = 0, T(z4 ) = ∞ lie on
the real line. As, T −1 is also a linear fractional transform, by Theorem 3.2.4,
it maps the real line to a circle or to a straight line. In particular,

T −1 (T(z1 )) = z1 , T −1 (T(z2 )) = z2 , T −1 (T(z3 )) = z3 , and T −1 (T(z4 )) = z4

lie on a circle or on a straight line.

Conversely, assume that z1 , z2 , z3 , and z4 lie on a circle or on a straight
line. Since T is a linear fractional transform, T(z1 ), T(z2 ) = 1, T(z3 ) =
0, and T(z4 ) = ∞ lie on a circle or on a straight line, say C . As ∞ lies
on C , it follows that C is a straight line. Using the fact that it passes through
1 and 0, we conclude that C must be the real line. Hence, T(z1 ) lies on the
real line itself. Thus, z1 , z2 , z3 , and z4 are real. 

LEMMA 3.2.17 Let C be a circle and z and w be any two points not on C .
If (w, z1 , z2 , z3 ) = (z, z1 , z2 , z3 ) for some three distinct points z1 , z2 , and z3

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 142 — #20

i i

142 Linear Fractional Transforms

on C , then (w, a1 , a2 , a3 ) = (z, a1 , a2 , a3 ) for any other three distinct points

a1 , a2 , and a3 on C .

Proof: Let T be the linear fractional transform that maps z1 , z2 , and z3 to

1, 0, and ∞, respectively. Therefore, (w, z1 , z2 , z3 ) = (z, z1 , z2 , z3 ) can be
rewritten as T(w) = T(z). As, a1 , z1 , z2 , and z3 lie on C , by Theorem 3.2.16,
T(a1 ) is real. Similarly, we can show that T(aj ) is real for j = 2 and 3. Hence,
using Result 3.2.15, we get

(z, a1 , a2 , a3 ) = (T(z), T(a1 ), T(a2 ), T(a3 ))

= (T(z), T(a1 ), T(a2 ), T(a3 ))
= (T(z), T(a1 ), T(a2 ), T(a3 ))
= (T(w), T(a1 ), T(a2 ), T(a3 ))
= (w, a1 , a2 , a3 ).

Hence, the lemma follows. 

Definition 3.2.18 Let C be a circle or a straight line. If z, w  C , then z and

w are said to be symmetric with respect to C if (w, z1 , z2 , z3 ) = (z, z1 , z2 , z3 )
for any three distinct points z1 , z2 , and z3 on C .

By previous lemma, the symmetry of the points does not depend on the choice
of the three points on C .

Example 3.2.19 Explain symmetry with respect to a straight line.

If C is a straight line and z and w are symmetric with respect to C , then by
definition, we have
(w, z1 , z2 , ∞) = (z, z1 , z2 , ∞)
for any pair of distinct points z1 and z2 in C . Hence, we have
 
z − z2 w − z2
= .
z 1 − z2 z 1 − z2

This implies that |z − z2 | = |w − z2 | for every z2 on C and

   
z − z2 w − z2
Im = −Im
z 1 − z2 z 1 − z2

which imply geometrically that z and w lie on different half planes determined
by the line C , and they are at equal distance from an arbitrary point on C .

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 143 — #21

i i

Rational Functions and Multivalued Functions 143

z
w
C

Example 3.2.20 Explain symmetry with respect to a circle.

If C is the circle |z − a| = r and z and w are symmetric with respect to C ,
then we have (w, z1 , z2 , z3 ) = (z, z1 , z2 , z3 ) for any three distinct points on C .
Using Result 3.2.15, we get
 2 
r r2 r2 r2
(w, z1 , z2 , z3 ) = , , ,
w − a z1 − a z2 − a z3 − a
 2 
r
= , z1 − a, z2 − a, z3 − a
w−a
 2 
r
= + a, z1 , z2 , z3
w−a
  
r2
= + a , z 1 , z2 , z3 .
w−a
 
r2
Therefore, z = + a , which implies that (z − a)(w − a) = r2 and
w−a
z−a
w−a = r2 . Hence, |z−a|·|w−a| = r2 and arg (w−a) = arg (z−a),
|z − a|2
provided z, w ∈ C. If z = a, then w = ∞.

Geometrically, one of the two points z and w lies inside the circle and the
other point lies outside the circle satisfying |z − a||w − a| = r2 . Furthermore,
a, w, and z are collinear, such that both z and w lie on a same ray starting
from a.
w
z
C a

RESULT 3.2.21 (Principle of symmetry)

If z and w are symmetric with respect to a circle or to a straight line C , then
T(z) and T(w) are symmetric with respect to T(C).
Proof: Let z1 , z2 , and z3 be three distinct points on C . As z and w are sym-
metric with respect to C , we have (w, z1 , z2 , z3 ) = (z, z1 , z2 , z3 ). By Result
3.2.15, we get

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 144 — #22

i i

144 Linear Fractional Transforms

(T(w), T(z1 ), T(z2 ), T(z3 )) = (T(z), T(z1 ), T(z2 ), T(z3 )).

Using Result 3.2.5, we get that T is one-to-one, and hence,
T(z1 ), T(z2 ), and T(z3 ) are three distinct points on T(C). Hence, T(z)
and T(w) are symmetric with respect to T(C). 

Algorithm 3.2.22 (To find the LFT that maps a, b , and c to p, q , and r ,
respectively.)

Step 1: Let T be the linear fractional transform that maps a, b, and c into
(z − b) (a − c)
1, 0, and ∞, respectively. That is, T(z) = .
(z − c) (a − b)
Step 2: Let S be the linear fractional transform that maps p, q, and r into
(w − q) (p − r)
1, 0, and ∞, respectively. That is, S(w) = .
(w − r) (p − q)

Step 3: Then S −1 ◦T is the required linear fractional transform, which can be

(w − q) (p − r) (z − b) (a − c)
obtained by solving for w from =
(w − r) (p − q) (z − c) (a − b)
in terms of z.

Example 3.2.23 Find the linear fractional transform that maps 1, 2, and 3 to
0, i, and − i, respectively.

Solution:
The linear fractional transform that maps 1, 2, and 3 into 1, 0, and ∞,
(z − 2) (1 − 3)
respectively, is given by , and the linear fractional trans-
(z − 3) (1 − 2)
form that maps 0, i, and − i into 1, 0, and ∞, respectively, is given by
(w − i) (0 + i)
.
(w + i) (0 − i)

(w − i) (0 + i) (z − 2) (1 − 3)
=
(w + i) (0 − i) (z − 3) (1 − 2)
(w − i) (z − 2)
⇒ − =2
(w + i) (z − 3)
⇒ −(w − i)(z − 3) = 2(z − 2)(w + i)
⇒ −w(z − 3) − 2w(z − 2) = i2(z − 2) − i(z − 3)
⇒ w(−3z + 7) = i(z − 1)
i(z − 1)
⇒ w= .
−3z + 7

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 145 — #23

i i

Rational Functions and Multivalued Functions 145

i(z − 1)
Hence, R(z) = , ∀z ∈ C∞ , is the required linear fractional
−3z + 7
transform.

Exercise 3.2.24 Find the linear fractional transform that maps a, b, and c to
p, q, and r, respectively, in the following:

1. 1, 0, and − 1 to 1, 1 − i2, and − 1

2. 1, −1, and i to 1, i, and − 1

3. 0, 1, and ∞ to −1, −i, and 1

4. ∞, −1, and − 2 to 1, ∞, and 3

5. 1 − i, i, and 0 to −i, 1 + i, and 1

1
6. 1, ∞, and − 2 to −1, 2, and
2
7. −i, 0 and ∞ to ∞, −i2, and 1

8. −1, 0, and 2 to −1, 4, and 2

9. −1, ∞, and i to 3 − i3, 2 + i3, and − 4 + i2

3 + i4 −1 + i 2 + i3
10. 1 + i, 1, and i to , , and .
5 2 2
Answers:
(1 + i2)z + 1 (i − 1)z + (3 + i)
1. , 2. ,
z + (1 + i2) (1 + i3)z + (1 − i)
−iz − 1 z−1
3. , 4. ,
−iz + 1 z+1
1 2z + 1
5. , 6. ,
−z + 1 z−4
z+2 3z + 4
7. , 8. ,
z+i 2z + 1
(2 + i3)z − 1 2z − 3
9. , 10. .
z + (1 − i) z+i

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 146 — #24

i i

146 Linear Fractional Transforms

Example 3.2.25 Check the symmetry of given z, and w with respect to the
given C in the following:

1. z = −4 + i7, w = 8 + i and C is the straight line y = 2x.

2. z = 2 + i5, w = 3 and C is the straight line 3x − 4x + 1 = 0.
3. z = 8 + i7, w = 17 + i19 and C is the circle |z − 5 − i3| = 10.

Solution:
1. We choose the three points ∞, 0, and 1 + 2i on y = 2x. As
11 − i3 11 + i3
(−4 + i7, ∞, 0, 1 + 2i) = and (8 + i, ∞, 0, 1 + 2i) = ,
10 10
−4 + i7 and 8 + i are symmetric with respect to y = 2x.
2. We choose the three points ∞, i, and 5 + 4i on 3x − 4y + 1 = 0. As
1 + 7i 1 + 7i
(2 + i5, ∞, i, 5 + i4) = and (3, ∞, i, 5 + 4i) = ,
5 10
2 + i5 and 3 are not symmetric with respect to 3x − 4y + 1 = 0.
3. We choose the three points 15 + 3i, −5 + 3i, and 5 + 13i on |z −
7 + 5i
5 − i3| = 10. As (8 + 7i, 15 + i3, −5 + i3, 5 + i13) = and
6
7 − 5i
(17 + i19, 15 + i3, −5 + i3, 5 + i13) = , 8 + 7i and 17 + 19i are
6
symmetric with respect to |z − 5 − i3| = 10.

r(z − a)
Example 3.2.26 Prove that if w = T(z) = , then T maps |z − a| = r
r2 − az
onto |w| = 1 and T(B(0, r)) = B(0, 1).

Proof: Let |z| = r. If w = T(z), then we can write z = reiθ . Therefore,

     
 r(z − a)   r(reiθ − a)   r − ae−iθ 
|w| =  2 = =  = 1.
r − az   r2 − areiθ   r − aeiθ 
Therefore, T maps |z| = r into |w| = 1. As every linear fractional transform
maps circles to circles, we conclude that T maps |z − a| = r onto |w| = 1.
Since T(B(0, r)) is a connected subset of {w ∈ C : |w|  1} and the latter
is having exactly two components namely, B(0, 1) and {w ∈ C : |w| > 1}.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 147 — #25

i i

Rational Functions and Multivalued Functions 147

Using the observation T(a) = 0, it follows that T(B(0, r)) ⊆ B(0, 1). Since,
T −1 is also a linear fractional transform, such that T −1 maps |w| = 1 onto
|z − a| = r and T −1 (0) = a, by a similar argument, we obtain T −1 (B(0, 1)) ⊆
B(0, r). Thus, we get T(B(0, r)) = B(0, 1). 

PROBLEM 3.2.27 Find all linear fractional transform that maps |z| = r onto
|w| = ρ .

Solution: Let w = T(z) = az+b cz+d be a required transformation, where

a, b, c, d ∈ C with ad − bc  0. Clearly, 0 and ∞ are symmetric points
with respect to the circle |w| = ρ ; therefore, by symmetric principle, if
T −1 (0) = α and T −1 (∞) = β , then α and β are symmetric points with
respect to |z| = r.
Case 1: 0  α  ∞.
By one of the properties of the symmetric points of a circle, we
2
have β = rα . Putting w = 0 and w = ∞ in z = −cw+adw−b
, we get
2
α = − ba and rα = − dc . From these observations, we note that c  0,
otherwise α = 0, which contradicts the fact that α  0. Therefore,
b
a z+ a
w= c z+ d = a z−α
c z− r2 = aα z−α
c αz−r2 .
As the image T(r) of r must lie on
c α    
 r−α   aα 
the circle |w| = ρ , we have ρ = |T(r)| =  aαc αr−r2  = cr , which
 
 aα 
implies  crρ  = 1. Hence, by Result 2.4.24, we can find θ ∈ R such
that aα
crρ = exp(iθ ). Thus, T(z) = aα z−α
c αz−r2 = rρ exp(iθ ) αz−r
z−α
2.

Case 2: α = 0
In this case, immediately, we have β = ∞ using the symmetry of α
and β with respect to |z| = r. Now, T(0) = 0 and T(∞) = ∞ imply
b = 0 and c = 0, and hence, a  0  d. As T maps |z| = r onto
|w| = ρ , we have ρ = |w| = |T(z)| =  da  r. Therefore, there exists
ϕ ∈ R such that da = ρr exp(iϕ). Thus, T(z) = ρr exp(iϕ)z.
Case 3: α = ∞.
Then, we have β = 0. Now, T(0) = ∞ and T(∞) = 0 imply that
d = 0 and a = 0. As T maps |z| = r onto |w| = ρ , we have
 
ρ = |w| = |T(z)| =  bc  1r . Therefore, there exists ψ ∈ R such that
b
c = rρ exp(iψ). Thus, T(z) = rρ exp(iψ) 1z .

Remark 3.2.28: The linear fractional transforms obtained in Case 1 and

Case 2 of the previous problem map |z| < r onto |w| < ρ . The

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 148 — #26

i i

148 Branch of a Multivalued Function

transforms obtained in Case 3 of the previous problem map |z| < r onto
|w| > ρ .

PROBLEM 3.2.29 Find all linear fractional transform that maps the upper
half-plane onto the unit circle.

Solution: Let w = T(z) = az+bcz+d be a linear fractional transform, which maps

upper half-plane onto the unit circle. Clearly, 0 and ∞ are symmetric points
with respect to the circle |w| = 1. If T −1 (0) = α and T −1 (∞) = β , then α
and β are symmetric points with respect to the real axis. Therefore, obviously
we have α, β  {0, ∞} and β = α . Using w = 0 and w = ∞ in z = −cw+a dw−b
,
z+ b
we get α = − ba and α = − dc . Therefore, w = ac da = ac z−α
z−α . As T(0) lies on
a z+ c
 
|w| = 1, we have 1 = |T(0)| = c , and hence, c = exp(iθ ) for some θ ∈ R.
a

Thus, T(z) = exp(iθ ) z−α

z−α .

3.3 BRANCH OF A MULTIVALUED FUNCTION

We shall start this section with the definition logarithm and then we introduce
the concept of branch of a multivalued function.

Definition 3.3.1 The inverse function of exp : R → (0, ∞) is called the

natural logarithm and is denoted by ln.

Note that the above definition is well defined as exp : R → (0, ∞) is a

bijection. See Lemma 2.4.6.
a
LEMMA 3.3.2 If a, b ∈ (0, ∞), then ln(ab) = ln a + ln b and ln =
b
ln a − ln b.

Proof: Let x = ln a and y = ln b, then exp(x) = a and exp(y) = b. By

Theorem 2.4.4, we have

exp(x + y) = and
exp(x) exp(y)
exp(x)
exp(x − y) = exp(x) exp(−y) = .
exp(y)
Therefore,

ln a + ln b = x + y = ln(exp(x + y)) = ln(exp(x) exp(y)) = ln(ab)

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 149 — #27

i i

Rational Functions and Multivalued Functions 149

and
  a
exp(x)
ln a − ln b = x − y = ln(exp(x − y)) = ln = ln .
exp(y) b

Hence, the lemma follows. 

1
RESULT 3.3.3 ln : (0, ∞) → R is differentiable and ln a = .
a

Proof: First, we prove that ln : (0, ∞) → R is continuous on (0, ∞).

For arbitrary a, b ∈ [1, ∞), we choose x, y ∈ [0, ∞) such that exp(x) = a and
exp(y) = b and then we get x = ln a, y = ln b. Now,

|a − b| = | exp(x) − exp(y)|
∞ 
 xn − y 
 n
=  
 n! 
n=1
 
 
 ∞  xk yn−1−k 
n−1
 
 k=0 
= |x − y|  
 n! 
 n=1 
 

n−1
∞ xk yn−1−k
 k=0
= |x − y|
n!
n=1
≥ |x − y|
= | ln a − ln b|.

Therefore, ln b → ln a as b → a, whenever a, b ∈ [1, ∞).

1 1
Suppose, a, b ∈ (0, 1], then , ∈ [1, ∞), and hence,
a b
1 1 1 1
b→a⇒ → ⇒ ln → ln ⇒ − ln b → − ln a ⇒ ln b → ln a.
b a b a
Hence, ln is continuous on (0, ∞).
To prove ln is differentiable on (0, ∞), fix a, b ∈ (0, ∞) arbitrarily. Then let
x = ln a and ln b = y. Now, using b → a ⇒ y → x, we get

ln b − ln a y−x
lim = lim
b→a b−a y→x exp(y) − exp(x)

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 150 — #28

i i

150 Branch of a Multivalued Function

1
= lim
y→x exp(y) − exp(x)
y−x
1
=
exp(y) − exp(x)
lim
y→x y−x
1 1 1
= 
= = .
exp (x) exp(x) a
1
Thus, ln is differentiable on (0, ∞) and ln a =
. 
a
We would like to define complex logarithm as the inverse function of
exp on C. However, unfortunately, exp : C → C is neither one-to-one nor
onto. However, we can define the complex logarithm of a non-zero complex
number as a multivalued function as follows.

Definition 3.3.4 For every w ∈ C \ {0}, we define log w = ln |w| + iarg w.

Note that in the above definition, the imaginary part of log is not single
valued.

Definition 3.3.5 Let f be a multivalued function on a subset of . We mean

a branch (an analytic branch) of f by a single-valued continuous (or analytic)
function g on a sub-region of , if value of g at a point is one of the multi-
values of f at the same point.
To define an analytic branch of log, we have to choose a unique value of
its imaginary part so that log becomes a single-valued analytic function.

LEMMA 3.3.6 The principal argument given in Definition 1.2.16 is continu-

ous on C \ {(a, 0) : a ≤ 0}.

Proof: Using the continuity of arctan, we get that arg is continuous on each
of the following regions.

1. {(a, b) : a > 0, b ∈ R}
2. {(a, b) : a < 0, b > 0}
3. {(a, b) : a < 0, b < 0}

Hence to conclude the proof of this lemma, we shall show that arg is
continuous on the positive imaginary axis and on the negative imaginary axis.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 151 — #29

i i

Rational Functions and Multivalued Functions 151

Let (0, b) be fixed, where b > 0 is arbitrary. If (an , bn ) → (0, b) as n → ∞,

with (an , bn )  (0, b), ∀n ∈ N, then, an → 0 and bn → b as n → ∞, and
hence, there exists M ∈ N such that bn > 0 for all n ≥ M .

Case 1: an > 0, ∀n ∈ N
bn bn π
As → +∞ as n → ∞, we have arctan → as n → ∞.
an an 2
Therefore,
 
π bn
arg ((0, b)) = = lim arctan = lim arg ((an , bn )).
2 n→∞ an n→∞

Case 2: an < 0, ∀n ∈ N  
bn π
Therefore, using the fact that arctan → as n → ∞, we
|an | 2
obtain
 
π π bn
arg ((0, b)) = = π − = π − lim arctan
2 2 n→∞ |an |
= lim arg ((an , bn )).
n→∞

Case 3: an ∈ R, ∀n ∈ N.
Given > 0 using Case 1 and Case 2, we find N1 , N2 ∈ N such that
N1 > M , N2 > M , and
 π 

|arg ((an , bn )) − arg ((0, b))| = arg ((an , bn )) −  < ,
2
∀n ≥ N1 if an > 0
 π 

|arg ((an , bn )) − arg ((0, b))| = arg ((an , bn )) −  <
2
∀n ≥ N2 if an < 0.

As
π π 

|arg ((an , bn )) − arg ((0, b))| =  − =0< ,
2 2
∀n ∈ N if an = 0,
 π 

it follows that arg ((an , bn )) −  < , ∀n ≥ max{N1 , N2 } if
2
an ∈ R. 

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 152 — #30

i i

152 Branch of a Multivalued Function

Remark 3.3.7: The principal branch of arg is not continuous at any point of
negative real axis.
Indeed, if (x, 0) ∈ C with x < 0, we have
   
1 1
x + , 0 → (x, 0) and x − , 0 → (x, 0) as n → ∞
n n

but
   
1 π 1 π
arg x + , 0 → and arg x − , 0 → − as n → ∞.
n 2 n 2

LEMMA 3.3.8 If 0  w ∈ C, then there exists a region containing w in which

some suitable branch of arg is continuous.

Proof: If 0  w ∈ C, then we can define argw on w = C\{r exp(iθ ) : r > 0},

by argw (z) = arg(z)+π +θw , ∀z ∈ w , where arg(z) is the principal argument
of z as in Definition 1.2.16, and θw ∈ R \ {arg(w) + 2kπ : k ∈ Z}. Being argw
is a translate of principal argument, it is also continouous on w . 

RESULT 3.3.9 (Principal branch for log)

If = C \ {(x, 0) : x ≤ 0} and if log(w) = ln |w| + iarg (w), with Im log(w)
∈ (−π, π ), then log : → C is analytic.

Proof: First, we note that for every w ∈ , principal argument of w belongs

to (−π , π), and hence, log is well defined on . As ln is continuous on (0, ∞)
and | · | is continuous on C (Example 1.6.18), we have Re log, which is con-
tinuous on . By Lemma 3.3.6, we have Im log, which is continuous on .
Therefore, log is continuous on by Lemma 1.6.19. Next, we prove that log
is differentiable on . For w0 , w ∈ , using the continuity of log on , we get

log(w) − log(w0 ) z − z0
lim = lim ,
w→w0 w − w0 z→z0 exp(z) − exp(z0 )

where z = log(w) and z0 = log(w0 )

1
= lim
z→z0 exp(z) − exp(z0 )
z − z0
1
=
exp(z) − exp(z0 )
lim
z→z0 z − z0

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 153 — #31

i i

Rational Functions and Multivalued Functions 153

1 1 1
= 
= = .
exp (z0 ) exp(z0 ) z0
1
Thus, log is differentiable on and log (a) = , ∀a ∈ . 
a
√
Definition 3.3.10 For a non-zero complex number z, we define w =  z by

√ √ θ
the solution of the equation w2 = z. More explicitly, z = ± r exp i ,
2
where z = r exp(iθ ).
√
RESULT 3.3.11 (Analytic branch for√ ·) √
If = C \ {(x, 0) : x ≤ 0} and √ if z is defined by the unique value of z
whose real part is positive, then · : → C is analytic.

Proof: First, we note that every z ∈ is never

√ √ real and negative,
√ and hence,
z is never purely
√ imaginary. Therefore, Re z  0 . If √z is defined by the
unique value of z whose
√ real part is positive, and then z is single
√ valued.
First, we claim that · is continuous on . Let z0 , z ∈ and √ w = z, w0 =
√
z0 . Hence, w2 = z and w20 = z0 . Using the definition of z, we get

|z − z0 | = |w2 − w20 | = |w − w0 | · |w + w0 | ≥ |w − w0 | · Re (w + w0 )
> |w − w0 | · Re w0

which implies that

√ √ 1
| z − z0 | = |w − w0 | < |z − z0 |.
Re w0
√ √ √
Hence, z → z0 as z → z0 . Thus, · is continuous on .
As
√ √
z − z0 w − w0 1 1 1
lim = lim = lim = = √ ,
z→z0 z − z0 w→w0 w − w
2 2
0
w→w0 w + w0 2w0 2 z0

√ 1
we get · is differentiable and its derivative is √. 
2 ·
 √
Example 3.3.12 Define an analytic branch for 1 + z and justify that it is
analytic. √
We know that z is defined on√ 1 = C \ {(x, 0) : x ≤ 0} as an analytic
function by the unique value of z, whose real part is positive. As the range

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 154 — #32

i i

154 Branch of a Multivalued Function

√ √
G of · is {z ∈ C : Re z > 
0}, we get 1 + z ∈ G, ∀z ∈ 1 . Furthermore,
√ √
G ⊆ 1 , and we can define 1 + z using the above definition of · twice
as an analytic function.
 √ 
RESULT 3.3.13 arccos(w) = ±i log w + w2 − 1

exp(iz) + exp(−iz)
Proof: Let z = arccos w. Then w = cos z = . Hence, it
2
follows that

2w = exp(iz) + exp(−iz) ⇒ exp(i2z) − 2w exp(iz) + 1 = 0

which is a quadratic equation in exp(iz). Therefore,

√
2w ± 4w2 − 4 
exp(iz) = = w ± w2 − 1.
2
√ √
Therefore, iz = log(w ± w2 − 1) ⇒ z = −i log(w ± w2 − 1). As
 
(w + w2 − 1) · (w − w2 − 1) = w2 − (w2 − 1) = 1
√
arccos(w) = z = ±i log(w + w2 − 1) 
√
RESULT 3.3.14 arcsin(w) = −i log(iw ± 1 − w2 ).

exp(iz) − exp(−iz)
Proof: Let z = arcsin(w). Then w = sin(z) = . Hence, it
i2
follows that

i2w = exp(iz) − exp(−iz) ⇒ exp(i2z) − i2w exp(iz) − 1 = 0

which is a quadratic equation in exp(iz). Therefore,

√
i2w ± −4w2 + 4 
exp(iz) = = iw ± 1 − w2 .
2
√ √
Therefore, iz = log(iw ±√ 1 − w ) ⇒ z = −i log(iw ± 1 − w ). Thus,
2 2
arcsin w = −i log(iw ± 1 − w ) 2 
 √ 
RESULT 3.3.15 sinh−1 w = log w ± w2 + 1

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 155 — #33

i i

Rational Functions and Multivalued Functions 155

Proof: If z = sinh−1 w, then

exp(z) − exp(−z)
w = sinh z =
2
which implies that exp(2z) − 2w exp(z) − 1 = 0. Therefore,
√
2w ± 4w2 + 4 
exp(z) = = w ± w2 + 1
2
 √ 
and hence, z = log w ± w2 + 1 . 
 
1 1+w
RESULT 3.3.16 tanh−1 w = log .
2 1−w

exp(z) − exp(−z)
If z = tanh−1 w, then w = tanh z = ,
exp(z) − exp(−z)

⇒ exp(z) − exp(−z) = w(exp(z) + exp(−z))

⇒ exp(2z) − 1 = w(exp(2z) + 1)
⇒ exp(2z)(1 − w) = 1 + w
1+w
⇒ exp(2z) = .
1−w
 
1 1+w
Therefore, z = log . 
2 1−w
 √ 
Exercise 3.3.17 Prove that cosh−1 w = log w ± w2 − 1 and tan−1 w =
 
1 1 + iw
log .
i2 1 − iw

Example 3.3.18 Find a branch for arccos. √

Let 1 = C \ {(x, 0) : |x| ≥ 1}. Define arccos(w)
√ = i√ log(w + w2 − 1)
with its real part lies between (−π, 0) and w2 − 1 = i 1 − w2 , ∀w ∈ 1 .
√ that this branch of arccos is analytic, we first show that 1 − w and
To prove 2

w + w2 − 1 are never real and negative whenever w ∈ 1 , so that from the

principal branch of log, arccos becomes analytic.

1 − w2 is real and less than or equal to 0 iff w2 is real and w2 ≥ 1

iff w is real and |w| ≥ 1
iff w  1.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 156 — #34

i i

156 Conformal Mapping

√ √
As (w + w2 − 1)(w − w2 − 1) = w2 − (w2 − 1) = 1, we have
√ √
(w + w2 − 1) is real iff (w − √w2 − 1) is real
iff w and w2 − 1 are real
iff w is real and |w| ≥ 1
iff w  1 .
√
If g(w) = w+ √ w − 1, ∀w ∈ 1 , then g is a continuous function on√ 1 , and
2
hence, {w+ w2 − 1 : w ∈√ 1 } is a connected subset of C. As w+ w2 − 1
is never real on 1 , {w + w2 − 1 : w ∈ 1 } lies either√in the upper half-
√ As i ∈ 1 and g(i) = i+i 2, which belongs
plane or in the lower half-plane.
to the upper half-plane, {w + w2 − √ 1 : w ∈ 1 } should be in the upper half-
plane, and hence, we define log(w + w2 − 1) as an analytic function √ on 1
whose imaginary part lies in (0, π). Hence, arccos(w) = i log(w + w2 − 1),
∀w ∈ 1 is analytic if its real part lies in (−π , 0).

Further, the derivative of arccos is obtained by

 
d i 2w
(arccos(w)) = √ × 1+ √
dw w + w2 − 1 2 w2 − 1
i i 1
= √ = √ =√ .
w −1
2 i 1−w 2 1 − w2

Exercise 3.3.19 Define an analytic branch for arcsin.

√
Exercise 3.3.20 Define
 an analytic branch for n
·. (Hint. Use the branch of
√
log and z = exp n log(z) .)
n 1

3.4 CONFORMAL MAPPING

Definition 3.4.1 (Curve)
If ϕ : [a, b] → C is a continuous function, then the image γ = ϕ([a, b]) of ϕ
is called a curve with initial point ϕ(a) and end point ϕ(b) in C.
We call ϕ(t), t ∈ [a, b] a parametric equation of the curve γ.

Definition 3.4.2 Let be a region and let z0 ∈ . A continuous function

f : → C is said to be a
1. conformal mapping of first type at z0 if for every curve γ with a para-
metric equation ϕ(t), t ∈ [a, b] such that ϕ(t0 ) = z0 and ϕ  (t0 )  0,

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 157 — #35

i i

Rational Functions and Multivalued Functions 157

we get ( f ◦ ϕ) (t0 )  0 and

( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 )
lim arg
t→t0 ϕ(t) − ϕ(t0 )

exists for a suitable branch of arg and is independent of the curve γ .

| f (z) − f (z0 )|
2. conformal mapping of second type at z0 if lim exists
z→z0 |z − z0 |
and is non-zero.

3. conformal mapping if f is conformal of first type at z0 as well as

conformal of second type at z0 .

THEOREM 3.4.3 If f is differentiable at z0 and f  (z0 )  0, then f is conformal

at z0 .
Proof: By assumption, we have

f (z) − f (z0 )
lim  0.
z→z0 z − z0

Let γ be any curve with a parametric equation ϕ(t), t ∈ [a, b] such that
ϕ(t0 ) = z0 and ϕ  (t0 )  0. By chain rule, we have

( f ◦ ϕ) (t0 ) = f  (z0 )ϕ  (t0 )  0

and as f  (z0 )  0, then arg is continuous on a region that contains f  (z0 ) with
respect to a suitable branch (Lemma 3.3.8), and hence, we get

( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 )
lim arg
t→t0 ϕ(t) − ϕ(t0 )
( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 )
= arg lim
t→t0 ϕ(t) − ϕ(t0 )
( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 ) t − t0
= arg lim ·
t→t0 t − t0 ϕ(t) − ϕ(t0 )
( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 ) t − t0
= arg lim ·
t→t0 t − t0 ϕ(t) − ϕ(t0 )
( f ◦ ϕ) (t0 )
= arg
ϕ  (t0 )

= arg f (z0 )

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 158 — #36

i i

158 Conformal Mapping

which is independent of the curve γ . Again using the hypothesis and the
continuity of the modulus function (Example 1.6.18), we get
| f (z) − f (z0 )|
lim  0.
z→z0 |z − z0 |
Thus, f is conformal at z0 . 

THEOREM 3.4.4 If f is conformal at z0 , then f is differentiable at z0 and

f  (z0 )  0.

Proof: By assumption, we have

1. for every curve γ with a parametric equation ϕ(t), t ∈ [a, b] such that
ϕ(t0 ) = z0 and ϕ  (t0 )  0, we get ( f ◦ ϕ) (t0 )  0 and
( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 )
lim arg =θ
t→t0 ϕ(t) − ϕ(t0 )
for some θ ∈ R with respect to a suitable branch of arg and is
independent of the curve γ.
| f (z) − f (z0 )|
2. conformal of second type at z0 . That is, lim = A for
z→z0 |z − z0 |
some A  0.
To prove this theorem, we show that if

zn → z0 as n → ∞ with zn  z0 , ∀n ∈ N

then
f (zn ) − f (z0 )
→ A exp(iθ ) as n → ∞.
z n − z0
First, we choose a curve γ with a parametric equation ϕ(t), t ∈ [a, b] with
the following properties:
1. ϕ(t0 ) = z0 , for some t0 ∈ [a, b] and ϕ(tn ) = zn , ∀n ∈ N for some
sequence (tn ) of distinct points of [a, b] such that tn → t0 as n → ∞.
2. ϕ  (t0 )  0.
Now, using the assumptions and Lemma 3.3.8,
f (zn ) − f (z0 )
lim
n→∞ z n − z0

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 159 — #37

i i

Rational Functions and Multivalued Functions 159

    
 f (zn ) − f (z0 ) 
= lim   exp iarg f (zn ) − f (z0 )
n→∞ z n − z0  z n − z0
    
 f (zn ) − f (z0 ) 
= lim   exp i lim arg f (ϕ(tn )) − f (ϕ((t0 ))
n→∞ z n − z0  n→∞ ϕ(tn ) − ϕ(t0 )
= A exp (iθ ) .
Thus, f is differentiable at z0 and f  (z0 ) = A exp (iθ )  0. 

THEOREM 3.4.5 Let be a region, z0 ∈ and let f : → C be continuous

at z0 .
1. If f is conformal of first type at z0 and fx , fy exist at z0 and they are
continuous, then f is differentiable at z0 .
2. If f is conformal of second type at z0 and fx , fy exist at z0 and they are
continuous, then f or f is differentiable at z0 .

Proof: Let γ be a curve with a parametric equation ϕ(t), t ∈ [a, b] such that
ϕ(t0 ) = z0 and ϕ  (t0 )  0. If ψ(t) = f (ϕ(t)), ∀t ∈ [a, b] and if x (t0 ) and y (t0 )
are the real and imaginary parts of ϕ  (t0 ), respectively, then ψ is differentiable
at t0 and
ψ  (t0 ) = fx (ϕ(t0 ))x (t0 ) + fy (ϕ(t0 ))y (t0 )
   
ϕ  (t0 ) + ϕ  (t0 ) ϕ  (t0 ) − ϕ  (t0 )
= fx (z0 ) + fy (z0 )
2 i2
   
fx (z0 ) − ify (z0 ) fx (z0 ) + ify (z0 )
= ϕ  (t0 ) + ϕ  (t0 ) .
2 2
Thus, we have
   
ψ  (t0 ) fx (z0 ) − ify (z0 ) fx (z0 ) + ify (z0 ) ϕ  (t0 )
= + . (3.3)
ϕ  (t0 ) 2 2 ϕ  (t0 )
fx (z0 ) − ify (z0 )
which represents an equation of a circle with centre and radius
2
fx (z0 ) + ify (z0 )
, as ϕ varies through the continuous function on [a, b] with
2
ϕ(t0 ) = z0 for some t0 and ϕ  (t0 )  0.
1. Assume that f is conformal of first type at z0 .
Then, we have
ψ  (t0 ) ( f ◦ ϕ) (t0 ) ( f ◦ ϕ)(t) − ( f ◦ ϕ)(t0 )
arg = arg = lim arg
ϕ  (t0 ) ϕ(t0 ) t→t0 ϕ(t) − ϕ(t0 )

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 160 — #38

i i

160 Conformal Mapping

which exists and independent of ϕ . Therefore, every point on the circle

represented by the equation (3.3) has same argument. This is possible
iff the radius of the circle is 0. That is, we have

fx (z0 ) + ify (z0 )

= 0 ⇒ fx (z0 ) = −ify (z0 ).
2
Therefore, f satisfies C–R equations at z0 whence by using Theorem
3.1.48, we get f is differentiable at z0 .

2. Assume that f is conformal of second type at z0 .

By a similar argument, we get the absolute value of every point on the
circle given in the equation (3.3) is constant. This is possible iff the
radius of the circle is 0 or the centre of the circle is 0. That is, we have

fx (z0 ) + ify (z0 ) fx (z0 ) − ify (z0 )

=0 or = 0.
2 2
fx (z0 ) + ify (z0 )
Just now we have seen that = 0 implies that f is
2
fx (z0 ) − ify (z0 )
differentiable at z0 . If = 0, then we have
2
   
fx (z0 ) = ify (z0 ) ⇒ f (z0 ) = −i f (z0 )
x y

and hence, f satisfies the C–R equation at z0 . Furthermore, as f

has continuous partial derivatives at z0 , f also has continuous partial
derivatives at z0 . Thus, f is differentiable at z0 , by Theorem 2.2.28. 

Example 3.4.6 The function exp maps the rectangle {z ∈ C : 1 < Re z <
π
2, < Im z < π3 } onto the part of angular sector {w ∈ C : a < |w| <
6
b, π6 < arg w < π3 }.
Let z = x + iy. Under the exponential function,

1. the horizontal line segment {(x, π6 ) : 1 < x < 2} is mapped onto

{exp(x) exp(i π6 ) : 1 < x < 2}, which is the line segment with slope π6
with the positive real axis from |w| = exp(1) to |w| = exp(2).

2. similarly, the horizontal line segment {(x, π3 ) : 1 < x < 2} is mapped

onto {exp(x) exp(i π3 ) : 1 < x < 2}, which is the line segment with
slope π3 with the positive real axis from |w| = exp(1) to |w| = exp(2).

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 161 — #39

i i

Rational Functions and Multivalued Functions 161

3. the vertical line segment {(1, y) : π6 < y < π3 } is mapped onto

{exp(1) exp(iy) : π6 < y < π3 }, which is the part of the circular arc
with radius exp(1) with argument varies from π6 to π3 .

4. similarly, the vertical line segment {(2, y) : π6 < y < π3 } is mapped

onto {exp(2) exp(iy) : π6 < y < π3 }, which is the part of the circular arc
with radius exp(2) with argument varies from π6 to π3 .

Thus, the image of the rectangle under exp is shown as in the following
diagram.

Exercise 3.4.7

1. Prove that the map f (z) = z2 , ∀z ∈ C maps the first quadrant {(x, y) ∈
C : x > 0, y > 0} onto the upper half-plane {(x, y) ∈ C : y > 0}.
z−1
2. Prove that the map f (z) = , ∀z ∈ C \ {−1} maps the upper
z+1
half-plane {(x, y) ∈ C : y > 0} onto the unit circle {z ∈ C : |z| < 1}.

3.5 ELEMENTARY RIEMANN SURFACE

The functions z → z1/n , log and arccos, and so on are not single-valued
functions because z → zn , exp and cos are not injective. Therefore, in the
previous section, we have defined branches of each of these multivalued func-
tions by omitting all but one of its images at every point. However, Riemann
introduced an alternate concept, called elementary surfaces corresponding to
every multivalued function, using which the multivalued function becomes a
bijective single-valued function between a region of the complex plane onto
a suitable surface.
A Riemann surface corresponding to a non-injective function f , intu-
itively speaking, is a surface obtained by attaching domains (commonly

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 162 — #40

i i

162 Elementary Riemann Surface

called sheets) of all the different branches of the multivalued function f −1

so that f becomes a bijection from C onto the surface (or equivalently, f −1
can be defined as a single-valued function from the surface into C) without
losing any of the multivalued information.
The rigorous definition of elementary Riemann surface uses terminolo-
gies from topology, which are not within the scope of this book. For the
sake of completeness, here we provide the rigorous definition of a Riemann
surface.

Definition 3.5.1 A Riemann surface is a two-dimensional real analytic

manifold.1

Example 3.5.2 Construct the Riemann surface corresponding to z → zn for

some n ∈ N with n > 1.
We first note that z → zn is not injective; for each θ ∈ R, all of the
 indeed,
θ+2π
following pairwise distinct n points r exp i k , k = 0, 1, 2, . . . , n − 1,
are mapped to the same value r exp (iθ). We also observe that the function
z → zn maps positive real axis onto itself. Although any sector
2π
z ∈ Z \ {0} : a < arg z < a +
n
is mapped onto the same set C \ {(x, 0) : x ≥ 0}, for an arbitrary a ∈ R, we
distinguish the images of the sectors
2kπ 2(k + 1)π
z ∈ Z \ {0} : < arg z <
n n
by denoting as
{w ∈ C \ {0} : 2kπ < arg w < 2(k + 1)π }
in n different complex planes, and we call them by sheet-k for every k ∈
0, 1, 2, . . . , n − 1. As the boundary arg z = 2(k+1)π
n in sheet-k and the bound-
ary arg z = 2(k+1)π n in sheet-(k + 1) are common, by using these n sheets, we
construct the Riemann surface as follows:
1. The edge arg z = 2(k+1)π n at the bottom of the slit in sheet-k and arg
2(k+1)π
z= n at the top of the slit in sheet-(k + 1) should be pasted, for
all k = 0, 1, 2, . . . , n − 2.
1 A studious reader can refer to any book on topology to know the definitions of the
terminologies used in the above definition.

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 163 — #41

i i

Rational Functions and Multivalued Functions 163

2. The boundary arg z = 2π at the bottom of the slit in sheet-(n − 1)

should be pasted with the boundary arg z = 0 at the top of the slit in
sheet-0.
3. The origin at every sheet should be glued as a single point in the
surface.
Now, if arg z traverses in the z-plane from 0 to 2π through 2π k , k =
1, 2, . . . , n−1, correspondingly, arg zn navigates from 0 to 2π in sheet-0, then
it moves to the successive sheets through the pasted edges arg z = 2(k+1)π n ,
and finally, it comes back to the top edge of the slit in sheet-0. In other
words, we have shown geometrically that the function z → zn is a continuous
bijection from C onto the surface.

Example 3.5.3 Construct the Riemann surface associated with cos.

We observe that the non-negative imaginary axis {(0, y) : y ≥ 0} is

mapped into {(u, 0) : u ≥ 1} because

cos(0 + iy) = cos(0) cosh(y) − i sin(0) sinh(y) = cosh(y)

exp(y) + exp(−y)
= ≥ 1.
2
In fact, we see that cos : {(0, y) : y ≥ 0} → {(u, 0) : u ≥ 1} is onto. Indeed,
for a given u > 1, using cosh(y) → +∞ as y → ∞, we find y > 0 such
that cosh(y) > u. Therefore, applying the intermediate value theorem, we can
find y0 ∈ (0, y) such that cosh(y0 ) = u. Being cosh an even function, we also
observe that {(0, y) : y ≤ 0} is mapped onto {(u, 0) : u ≥ 1}. By a similar
argument, one can verify that cos maps
• both of {(π , y) : y ≥ 0} and {(π , y) : y ≤ 0} onto {(u, 0) : u ≤ 1},
• {(x, y) : 0 < x < π , y > 0} onto the lower half-plane {(u, v) : v < 0},
• {(x, y) : 0 < x < π , y < 0} onto the upper half-plane {(u, v) : v > 0},
• {(x, y) : 0 < x < π , y = 0} onto the line segment {(u, 0) : |u| < 1}.
Now we claim that cos : {(x, y) : 0 < x < π } → C \ {(u, v) ∈ C : |u| ≥
0, v = 0} is a bijection. As, the injectivity of cos on {(x, y) : 0 < x < π }
is obvious, we only verify that surjectivity of cos. Let w ∈ C
 \ {(u, v) ∈ C :
√
|u| ≥ 0, v = 0} be arbitrary and let θ = w ± w − 1 . If θ ∈ (0, π ),
2
 √ 
 
then we take z = θ − i log w ± w2 − 1 so that z ∈ {(x, y) : 0 < x < π}

i i

i i
 i i

“book” — 2014/6/4 — 20:56 — page 164 — #42

i i

164 Elementary Riemann Surface

and cos z = w. Otherwise, θ ∈ (π , 2π). In this case,

 we choose the required
 √ 
pre-image by z = 2π − θ − i log w ± w − 1.2

By a similar argument, we obtain cos : {(x, y) : (k − 1)π < x < kπ } →

C \ {(u, v) ∈ C : |u| ≥ 0, v = 0} is a bijection, ∀k ∈ N, with the additional
information that
• the upper and lower half-strips are mapped onto the lower and upper
half-planes, respectively, when k is odd,
• the upper and lower half-strips are mapped onto themselves when k is
even.
Now we are ready to construct the Riemann surface associated with cos.
• We take countably infinite number of complex planes with two slits
from 1 to ∞ and −1 to ∞ and we denote them by

{w ∈ C\{0} : (2k+1)π  arg w ∈ (2kπ , (2k+2)π )}∪{(u, 0) ∈ C : |u| ≤ 1},

called sheet-k , ∀k ∈ Z.
• Paste upper edge on the slit on the left side of sheet-k with lower edge
on the slit on the left side of sheet-(k + 1) and lower edge on the slit
on the left side of sheet-k with upper edge on the slit on the left side of
sheet-(k + 1).
• Paste upper edge on the slit on the right side of sheet-k with lower edge
on the slit on the right side of sheet-(k + 1) and lower edge on the slit
on the right side of sheet-k with upper edge on the slit on the right side
of sheet-(k + 1).
• By gluing the line segment {(u, 0) : |u| ≤ 1} in every sheet altogether,
identify it as a single line segment on the surface.
Then, cos becomes a continuous bijection from C onto the surface.

Exercise 3.5.4 Construct the Riemann surface corresponding to exp(z).

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 165 — #1

i i

4
Complex Integration

4.1 LINE INTEGRAL

In this section, first we recall some definitions and results (without proofs)
from the theory of Riemann integration of real-valued function on [a, b].

Definition 4.1.1 Let f : [a, b] → R be a bounded function, where −∞ <

a < b < +∞.
1. By a partition of [a, b], we mean a finite subset {t0 , t1 , t2 , . . . , tn } with
a = t0 < t1 < t2 < · · · < tn = b.
 
n
2. Define U(P, f ) = sup f (t) (tk − tk−1 ) and L(P, f ) =
k=1 t∈[tk−1 , tk ]
 
n
inf f (t) (tk − tk−1 ).
k=1 t∈[tk−1 , tk ]

3. f is Riemann integrable if inf U(P, f ) = sup L(P, f ) and this value is

P P
b
denoted by f (t) dt.
a

THEOREM 4.1.2 Let f : [a, b] → R be a bounded real-valued function. Then

f is Riemann integrable on [a, b] iff given  > 0, there exists a partition P of
[a, b] such that U(P, f ) − L(P, f ) <  .

THEOREM 4.1.3 If f , f1 , and f2 are Riemann integrable functions on [a, b],

then
b b b
1. ( f1 + f2 )(t) dt = f1 (t) dt + f2 (t) dt,
a a a

165

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 166 — #2

i i

166 Line Integral

b b
2. (αf )(t) dt = α f (t) dt, ∀α ∈ R,
a a

b b
3. f1 (t) dt ≤ f2 (t) dt, whenever f1 ≤ f2 on [a, b],
a a

b b c
4. f (t) dt = f (t) dt + f (t) dt, provided a < c < b.
a c a

RESULT 4.1.4 If f : [a, b] → R is bounded and continuous except at a finite

number of points of [a, b], then f is Riemann integrable.

RESULT 4.1.5 If f : [a, b] → R is Riemann  integrable, then | f | is also

b  b
 
Riemann integrable over [a, b] and  f (t) dt ≤ |f (t)| dt.
a  a

Definition 4.1.6 Let f = u + iv be a complex-valued continuous function on

[a, b], where u and v are real-valued continuous functions on [a, b]. We define
b b b

f (t) dt = u(t) dt + i v(t) dt,

a a a

b b
where u(t) dt and v(t) dt are Riemann integrals of u and v, respectively, on
a a
[a, b].

RESULT 4.1.7 If f : [a, b] → C, g : [a, b] → C are continuous and c ∈ C,

b b b b b
then ( f + g)(t) dt = f (t) dt + g(t) dt and (cf )(t) dt = c f (t) dt.
a a a a a

Proof: Let f = u1 + iv1 , g = u2 + iv2 and c = α + iβ . Then

b

( f + g)(t) dt
a
b b

= (u1 + u2 )(t) dt + i (v1 + v2 )(t) dt

a a

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 167 — #3

i i

Complex Integration 167

⎛ b b
⎞ ⎛ b b
⎞

=⎝ u1 (t) dt + u2 (t) dt⎠ + i ⎝ v1 (t) dt + v2 (t) dt⎠

a a a a
⎛ b b
⎞ ⎛ b b
⎞

=⎝ u1 (t) dt + i v1 (t) dt⎠ + ⎝ u2 (t) dt + i v2 (t) dt⎠

a a a a
b b

= f (t) dt + g(t) dt,

a a
b

(cf )(t) dt,

a
b

= ((α + iβ)(u + iv))(t)

a
b b

= (αu(t) − βv(t)) dt + i (αv(t) + βu(t)) dt

a a
b b
⎛ b b
⎞

=α u(t) dt − β v(t) dt + i ⎝α v(t) dt + β u(t) dt⎠

a a a a
⎛ b b
⎞

= (α + iβ) ⎝ u(t) dt + i v(t) dt⎠

a a
b

=c f (t) dt.
a

Hence, the result follows. 

 
b 
 
LEMMA 4.1.8 If f : [a, b] → C is a continuous function, then  f (t) dt ≤
a 
b
| f (t)| dt.
a

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 168 — #4

i i

168 Line Integral

b b
Proof: If f (t) dt = 0, then obviously the inequality follows. If f (t) dt  0,
a a
b
then we can write it as f (t) dt = reiθ . Then, using the previous theorem, we
a
get
 
 b 
 
 f (t) dt = r

 
a
b
−iθ
= e f (t) dt
a
b

= e−iθ f (t) dt
a
⎛ b
⎞

= Re ⎝ e−iθ f (t) dt⎠

a
b

= Re (e−iθ f (t)) dt
a
b
 −iθ 
≤ e f (t) dt
a
b

= | f (t)| dt.

a
RESULT 4.1.9 If f : [a, b] → C is Riemann integrable and | f (t)| ≤ M ,
b
∀t ∈ [a, b], then | f (t)| dt ≤ M(b − a).
a

Proof of this result follows immediately from the previous Lemma.

RESULT 4.1.10 Let fn and f be complex-valued Riemann integrable func-

b
tions on [a, b], ∀n ∈ N. If fn → f uniformly on [a, b], then fn (t) dt →
a
b
f (t) dt as n → ∞.
a

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 169 — #5

i i

Complex Integration 169

Proof: Let ε > 0 be given. Since fn → f uniformly on [a, b], there exists
N ∈ N such that | fn (x) − f (x)| < /(b − a) ∀ n ≥ N and ∀ x ∈ [a, b].

Now for n ≥ N ,
 b 
 b  b b
  
 fn (x)dx − f (x)dx ≤ | fn (x) − f (x)|dx < dx = .
  b−a
 
a a a a

b b
Therefore, fn (x) dx → fdx as n → ∞. 
a a

t
THEOREM 4.1.11 If f : [a, b] → C is continuous and g(t) = f (τ ) dτ , ∀t ∈
a
[a, b], then g is differentiable on [a, b] and g = f.

Proof: Let t ∈ [a, b) be arbitrary. Since f is continuous at t, given  > 0, there

exists δ > 0 such that [t, t + δ] ⊂ [a, b) and t ≤ s < t + δ ⇒ | f (s) − f (t)| <  .
For 0 < h < δ , we have
 t 
 
t+h 
t+h 
   f (τ ) dτ − f (τ ) dτ − f (t) dτ 
 g(t + h) − g(t)  
 a 
− f (t) = 
a t
 
h  h 
 
 
 t+h 
 
1  
=  ( f (τ ) − f (t)) dτ 
h 
t
t+h
1
≤ | f (τ ) − f (t)| dτ
h
t
1
= h = , since |τ − t| < h < δ.
h
Similarly,
 we can prove that  for every t ∈ (a, b], there exits δ > 0 such
 g(t − h) − g(t) 
that  − f (t) <  , whenever 0 < h < δ . Therefore, we have
−h
proved that
g(t + h) − g(t)
lim = f (t), ∀t ∈ [a, b].
h→0 h
Thus, g = f on [a, b]. 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 170 — #6

i i

170 Line Integral

THEOREM 4.1.12 (Fundamental theorem of calculus)

If f : [a, b] → C is differentiable on [a, b] and f  is Riemann itegrable on
b
[a, b], then f  (t) dt = f (b) − f (a).
a

Proof: First, we prove this theorem for the case that f is a real-valued
Riemann integrable function on [a, b]. By Theorem 4.1.2, given  > 0,
there exists a partition P = {t0 , t1 , t2 , . . . , tn } of [a, b] such that U(P, f  ) −
L(P, f  ) < . By mean value theorem, there exists sk ∈ (tk−1 , tk ) such
that
f (tk ) − f (tk−1 ) = f  (sk )(tk − tk−1 ), ∀k = 1, 2, . . . , n. (4.1)

If Mk = sup f  (t) and mk = inf f  (t), ∀k = 1, 2, . . . , n, then we have

tk−1 <t<tk tk−1 <t<tk
mk ≤ f  (sk ) ≤ Mk , ∀k = 1, 2, . . . , n, which implies that


n
L(P, f  ) = mk (tk − tk−1 )
k=1
n
≤ f  (sk )(tk − tk−1 )
k=1

n
≤ Mk (tk − tk−1 )
k=1
= U(P, f  ).

On the other hand, by the definition of Riemann integral, we also have

b
L(P, f  ) ≤ f  (t) dt ≤ U(P, f  ), and hence,
a

 
 b
 
 n

 
f (t) dt − f (sk )(tk − tk−1 ) ≤ U(P, f ) − L(P, f ).


 k=1 
a

Therefore, using equation (4.1), we get

   
 b   b
 
   n

 f  (t) dt − f (a) + f (b) =  f  (t) dt − [ f (tk ) − f (tk−1 )]
 
   k=1 
a a

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 171 — #7

i i

Complex Integration 171

 
 b 
 
n

=  f  (t) dt − f  (sk )(tk − tk−1 )

 k=1 
a
≤ U(P, f ) − L(P, f ) < .

b
Since  > 0 is arbitrary, we get f  (t) dt = f (b) − f (a).
a
If f is a complex-valued function, let f = u + iv, where u and v are real-
valued Riemann integrable functions. Hence, using this theorem for u and v,
we obtain
b b b
 
f (t) dt = u (t) dt + i v (t)) dt
a a a
= (u(b) − u(a)) + i(v(b) − v(a))
= f (b) − f (a).

Hence, the theorem follows. 

2
Example 4.1.13 Find [(1 + 9t2 ) + i(4t + 2)] dt.
0

2 2 2

[(1 + 9t ) + i(4t + 2)] dt

2
= (1 + 9t ) dt + i
2
(4t + 2) dt
0 0 0
 2  2
= t + 3t 3
+ i 2t + 2t
2
0 0
= 26 + 12i.

Definition 4.1.14 (Piecewise smooth curve)

A curve γ with a parametric equation ϕ(t), t ∈ [a, b] is called a piecewise
smooth curve if ϕ is differentiable and ϕ  is continuous for all but finite num-
ber of points {tj : 1 ≤ j ≤ n} of [a, b]. Furthermore, ϕ has both left limit and
right limit at each tj , 1 ≤ j ≤ n. Hereafter by a curve, we mean a piecewise
smooth curve.

Definition 4.1.15 A curve with a parametric equation ϕ(t), t ∈ [a, b] is said

to be

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 172 — #8

i i

172 Line Integral

1. a closed curve if ϕ(a) = ϕ(b).

2. a simple curve if ϕ is one-to-one on [a, b].
3. a simple closed curve if it is closed and φ(t) = φ(s) ⇒ either t = s or
{t, s} = {a, b}.

Definition 4.1.16 Let be a region and let γ be a curve in . If f is a

continuous function on , then define the line integral
b

f (z) dz = f (ϕ(t))ϕ  (t) dt,

γ a

where ϕ(τ ), τ ∈ [a, b] is a parametric equation of γ .

Definition 4.1.17 Let γ be a curve in a region . If f is a continuousfunction

on and g has a continuous partial derivatives on , then define f dg =
γ
 
b ∂g ∂g
f dx − dy . More explicitly, if ϕ(t), t ∈ [α, β] is the parametric
a ∂x ∂y
equation of γ , then
β  
∂g ∂g
f dg = f (φ(t)) (φ(t)) (Re ϕ) (t) − (φ(t)) (Im ϕ) (t) dt.
∂x ∂y
γ α

Example 4.1.18 Find f (z) dz, where f (z) = z + 1, ∀z ∈ C and ϕ(t) =
γ
t + it, t ∈ [0, 1].
1

f (z) dz = f ((ϕ(t))ϕ  (t) dt

γ 0
1

= (ϕ(t) + 1)(1 + i) dt
0
1

= ((1 + t) + it)(1 + i) dt
0
1

= [((1 + t) − t) + i(1 + t + t)] dt

0

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 173 — #9

i i

Complex Integration 173

= (1 + i(2t + 1)) dt
0
1 1

= dt + i (2t + 1) dt
0 0
= 1 + i2.

Example 4.1.19 If f (z) = z2 and γ is the parabola given by 2t + i(t2 + 1)

t ∈ R, then find f (z) dz, where σ is the part of γ from i to 2 + i2.
σ
First, we note that the points i and 2 + i2 correspond to t = 0 and t = 1,
respectively, and hence, the parametric equation of σ is ϕ(t) = 2t + i(t2 + 1),
t ∈ [0, 1]. Now, for every t ∈ [0, 1], we have

ϕ  (t) = 2 + i2t
f (ϕ(t)) = (2t + i(t2 + 1))2
= (4t2 − (t2 + 1)2 ) + i2t(t2 + 1)
= (−t4 + 2t2 − 1) + i(2t3 + 2t)
f (ϕ(t))ϕ  (t) = 2(−t4 + 2t2 − 1) − 2t(2t3 + 2t)
+ i(4t3 + 4t − 2t5 + 4t3 − 2t)
= (−6t4 − 2) + i(−2t5 + 8t3 + 2t).
1

f (z) dz = f ((ϕ(t))ϕ  (t) dt

σ 0
1 1

= (−6t − 2) dt + i
4
(−2t5 + 8t3 + 2t) dt
0 0
   
6 1
= − −2 +i − +2+1
5 3
16 8
= − +i .
5 3


Example 4.1.20 Find (z + 3) dz, where γ is the circular arc given by ϕ(t) =
γ
i + 2eit , t ∈ [0, π ].

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 174 — #10

i i

174 Line Integral

Solution:
π

(z + 3) dz = [(i + 2eit ) + 3]i2eit dt

γ 0
π

= i2 [(3 + i)eit + 2ei2t )] dt

0
 π
= i2 −(3 + i)ieit + −iei2t
0
= −12 − i4.

Exercise 4.1.21

1. Evaluate f (z) dz, where f and the parametric equation ‘ϕ(t),
γ
t ∈ [a, b]’ of γ are given as follows:

(1) f (z) = z3 , ϕ(t) = 1 + it, t ∈ [0, 1],

(2) f (z) = z, ϕ(t) = exp(iπt), t ∈ [0, 1],
1
(3) f (z) = , ϕ(t) = 1 + 2t + it2 , t ∈ [0, 1],
z
√
(4) f (z) = z exp(z2 ), ϕ(t) = t, t ∈ [1, 4],
(5) f (z) = cos(z), ϕ(t) = t + it2 , t ∈ [0, 1].
−5 exp(1)
Answers: (1) ; (2) iπ ; (3) log(3 + i); (4) (exp(3) − 1);
4 2
(5) sin(1 + i).

2. Evaluate f (z) dz, where f (z) = z2 + z, where γ is the part of the
γ
parabola x = y2 from (0, 0) to (4, 2).
2
Answer: (17 + i56).
3
 1
3. Evaluate f (z) dz, where f (z) =
, where γ is the part of the circle
γ z−a
with center a ∈ C and radius r > 0 from a + r to a + ir.
π
Answer: i .
2

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 175 — #11

i i

Complex Integration 175

Definition 4.1.22 Let be a region and let γ be a curve in . If f is a

 b
continuous function on , then define f (z) |dz| = f (ϕ(t))|ϕ  (t)| dt, where
γ a
ϕ(t), t ∈ [a, b] is a parametric equation of γ .

THEOREM 4.1.23 If f is continuous on and γ is a curve in , then

  
 
 f (z) dz ≤ | f (z)| |dz|.
γ  γ

Proof: Using Lemma 4.1.8, if ϕ(t), t ∈ [a, b] is a parametric equation of γ ,

then
   b 
    b
   
 f (z) dz =  f (ϕ(t))ϕ (t) dt ≤ | f (ϕ(t))ϕ  (t)| dt = | f (z)| |dz|.

   
   
γ a a γ

Hence, the theorem follows. 

Definition 4.1.24 If a curve γ is given by ϕ(t), t ∈ [a, b], then the opposite
curve −γ is defined by ψ(t) = ϕ(a + b − t), t ∈ [a, b].

Geometrically, −γ is obtained from γ just by changing its direction in the

opposite sense.

g −g

LEMMA
 4.1.25 If f is continuous on and γ is a curve in , then
f (z) dz = − f (z) dz.
−γ γ

Proof: If ϕ(t), t ∈ [a, b] is a parametric equation of γ and ψ(t) = ϕ(a +

b − t), t ∈ [a, b] is the parametric equation of −γ , then using the change of
variable s = a + b − t in the following, we get
b
f (z) dz = f (ψ(t))ψ  (t) dt
−γ a
b
= f (ϕ(a + b − t))ϕ  (a + b − t) (−dt)
a

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 176 — #12

i i

176 Line Integral

a
= f (ϕ(s))ϕ  (s) ds
b
b
= − f (ϕ(s))ϕ  (s) ds
a

= − f (z) dz.
γ

Hence, the lemma follows. 

Definition 4.1.26 (Arc length)

If γ is curve with parametric equation z(t), t ∈ [a, b], then the length of
the curve is defined by L(γ ) = sup |z(tk ) − z(tk−1 )|, where P varies over all
P
partitions of [a, b].
A curve with finite length is called a rectifiable curve.

THEOREM 4.1.27 If γ is a smooth curve with the parametric equation

b
z(τ ), τ ∈ [a, b], then it is rectifiable and L(γ ) = a |z (t)| dt.

Proof: For each fixed t ∈ [a, b], let γa,t be the part of the curve γ with
parametric equation z(τ ) = u(τ ) + iv(τ ), τ ∈ [a, t], where u and v are
real-valued differentiable function on [a, b]. Define S : [a, b] → R by
S(t) = L(γa,t ), ∀t ∈ [a, b]. We shall show that S is a differentiable function
and its S  (t) = |z (t)|, ∀t ∈ [a, b]. Therefore, using the uniform continuity of
z , given  > 0, there exists δ > 0 such that

|τ − t| < δ ⇒ |z (τ ) − z (t)| < √ (4.2)
2
 
⇒ |u (τ ) − u (t)| < √ and |v (τ ) − v (t)| < √ .
 
2 2
Therefore, for |τ − t| < δ and |σ − t| < δ , we have
    
 |(u (τ ), v (σ ))| − |z (t)|  =  |(u (τ ), v (σ ))| − |(u (t), v (t))| 
≤ |(u (τ ), v (σ )) − (u (t), v (t))|

= (u (τ ) − u (t))2 + (v (σ ) − v (t))2

2 2
< +
2 2
= .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 177 — #13

i i

Complex Integration 177

Hence, it follows that

|z (t)| −  < |(u (τ ), v (σ ))| < |z (t)| + , ∀σ , τ ∈ (t − δ, t + δ). (4.3)

Let 0 < s < δ and let P = {t0 , t1 , . . . , tn } be an arbitrary partition of [t, t + s].
Then, applying mean-value theorem (Theorem 2.1.18) for u and v, we get

n n 

|z(tk ) − z(tk−1 )| = (u(tk ) − u(tk−1 ))2 + (v(tk ) − v(tk−1 ))2
k=1 k=1

n
= (tk − tk−1 )|(u (xk ), v (yk ))|, (4.4)
k=1

for some xk , yk ∈ (tk−1 , tk ).

Therefore,

s(|z (t)| − ) = (tn − t0 )(|z (t)| − )

 n
= (tk − tk−1 )(|z (t)| − )
k=1

n
< (tk − tk−1 )|(u (xk ), v (yk ))| (by using (4.3))
k=1

n
= |z(tk ) − z(tk−1 )| (by using (4.4))
k=1
n
≤ (tk − tk−1 )|z (sk )|, for some sk ∈ (tk−1 , tk )
k=1
(by using Theorem 2.1.19 for the function z.)
< (tn − t0 )(|z (t)| + )
(since |sk − t| < δ, by using (4.3))
= s(|z (t)| + ).

Hence, we proved that


n
s(|z (t)| − ) ≤ sup |z(tk ) − z(tk−1 )| ≤ s(|z (t)| + ),
P k=1

where P varies over all partitions of [t, t + s]. Therefore, if γt, t+s is the part of
gamma, whose parametric equation is z(τ ), τ ∈ [t, t + s], then it follows that

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 178 — #14

i i

178 Line Integral

s(|z (t)| − ) ≤ L(γt, t+s ) = (S(t + s) − S(t)) ≤ s(|z (t)| + ),

 
 S(t + s) − S(t) 
which implies that   − |z (t)| <  . Similarly, we can show

 s 
 S(t − s) − S(t) 
that if 0 < s < δ ,   − |z (t)| <  . That is, we have proved that

(−s)
 S(t + s) − S(t)
S (t) = lim = |z (t)|. Therefore, by Theorem 4.1.11, we get
s→0 s
t t

S(t) = S (τ ) dτ = S  (τ ) dτ , ∀t ∈ [a, b].
a a

b
In particular, L(γ ) = S(b) = S  (τ ) dτ . 
a

LEMMA 4.1.28 Let f (z) be a continuous function on a region . If there

exists an analytic function F on such that F  = f , then

f (z) dz = F(ϕ(b)) − F(ϕ(a)),

γ

for every curve γ in , where ϕ(t), t ∈ [a, b] is a parametric equation of γ .

Proof: Define h(t) = F(ϕ(t)), ∀t ∈ [a, b], then h is piecewise differen-

tiable and h (t) = F  (ϕ(t))ϕ  (t) = f (ϕ(t))ϕ  (t) for every t at which ϕ is
differentiable. By fundamental theorem of calculus (Theorem 4.1.12),
b

f (z) dz = f (ϕ(t))ϕ  (t) dt

γ a
b

= h (t) dt
a
= h(b) − h(a)
= F(ϕ(b)) − F(ϕ(a)).
Hence, the lemma follows. 

COROLLARY 4.1.29 Let f (z) be a continuous function on a region . If there

exists an analytic function F on such that F  = f , then f (z) dz = 0, for
γ
every closed curve γ in .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 179 — #15

i i

Complex Integration 179

Proof: Since γ is a closed curve, if ϕ(t), t ∈ [a, b] is a parametric equation

of γ , then ϕ(a) = ϕ(b). Hence, using the previous lemma, we get

f (z) dz = F(ϕ(b)) − F(ϕ(a)) = 0.

γ

Converse of the above corollary is also true, which follows. 

RESULT
 4.1.30 If f is a continuous function on a region such that
f (z) dz = 0, for every closed curve γ in , then f is the derivative of an
γ
analytic function F on .

Proof: Let w = (u, v) ∈ be arbitrary. Then, for a given z0 = (x0 , y0 ) ∈

and an  > 0, choose δ > 0 such that B(z0 , δ) ⊆ and

| f (z) − f (z0 )| <  whenever |z − z0 | < δ. (4.5)

Let Pz0 be the polygon joining w and z0 such that

1. the line segments of Pz0 are parallel to the coordinate axes.

2. the line segment incident with z0 is horizontal and completely con-

tained in B(z0 , δ).

3. Pz0 ⊂ .

Such a polygon exists by Theorem 1.4.31.

(x, y0)
(x0, y0)

(u, u)

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 180 — #16

i i

180 Line Integral

 ∂F
Define F(z0 ) = f (z) dz. We claim that ∂x (x0 , y0 ) = f (x0 , y0 ). Let
Pz0
(x, y0 ) be the point adjacent with (x0 , y0 ) in the polygon Pz0 . As the parametric
equation of L(x, y0 ),(x0 , y0 ) and L(x, y0 ),(x0 +h, y0 ) are
ϕ(t) = t + iy0 = (t, iy0 ), t ∈ [x, x0 ]
and
ψ(t) = t + iy0 = (t, iy0 ), t ∈ [x, x0 + h],
respectively, for each h ∈ R such that |h| < δ,
 
f (z) dz − f (z) dz
F(x0 + h, y0 ) − F(x0 , y0 ) Pz0 +h Pz0
=
h  h 
f (z) dz − f (z) dz
L(x, y0 ),(x0 +h, y0 ) L(x, y0 ),(x0 , y0 )
=
h
x0+h x0
f (t, y0 ) dt − f (t, y0 ) dt
x x
=
h
x0 +h
1
= f (t, y0 ) dt
h
x0
h
1
= f (x0 + s, y0 ) ds. (4.6)
h
0
Now, using the equation (4.6), we get
 
 F(x0 + h, y0 ) − F(x0 , y0 ) 
 − f (x0 , y0 )
 h
 h 
 
1 
=  f (x0 + s, y0 ) ds − f (x0 , y0 )
h 
0
 
 h 
1 
=  ( f (x0 + s, y0 ) − f (x0 , y0 )) ds
h 
0
h
1
≤ | f (x0 + s, y0 ) − f (x0 , y0 )| |ds|
|h|
0

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 181 — #17

i i

Complex Integration 181

h
1
<  |ds|
|h|
0
|h|
= = ,
|h|
as |(x0 + s, y0 ) − (x0 , y0 )| = |t| < |h| < δ and by using equation (4.5).
 Thus,
our claim is proved. Next for every z0 ∈ , we define G(z0 ) = f (z) dz,
Qz0
where Qz0 is a polygon joining w and z0 such that
1. the line segments of Qz0 are parallel to the coordinate axes.
2. the line incident with z0 is vertical and completely contained in B(z0 , δ).
3. Qz0 ⊆ .
Next we claim that ∂G
∂y (x0 , y0 ) = if (x0 , y0 ). We fix  > 0, δ > 0 and h ∈ R as
before. As the parametric equation of L(x0 , v),(x0 , y0 ) is
ϕ(t) = x0 + it = (x0 , t), t ∈ [v, y0 ],
by a similar argument, we get
 
 G(x0 , y0 + h) − G(x0 , y0 ) 
 − if (x0 , y0 )
 h
 h 
 
1 
=  f (x0 , y0 + t) idt − if (x0 , y0 )
h 
0
h
1
≤ | f (x0 , y0 + t) − f (x0 , y0 )| |dt| < .
|h|
0
Since

F(z0 ) − G(z0 ) = f (z) dz − f (z) dz = f (z) dz

Pz0 Qz0 Pz0 ∪(−Qz0 )

and Pz0 ∪ (−Qz0 ) is a closed curve, by hypothesis, we get f (z) dz =
Pz0 ∪(−Qz0 )
0, and hence, F(z0 ) − G(z0 ) = 0. In other words, F = G on . There-
fore, F has continuous partial derivatives, and they satisfy the C–R equation
∂F ∂F
∂x = −i ∂y (= f ). Thus, F is analytic on (by Theorem 2.2.28) and
∂F
F = ∂x = f. 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 182 — #18

i i

182 Winding Number and Cauchy’s Theorems

4.2 WINDING NUMBER AND CAUCHY’S THEOREMS

1  dz
LEMMA 4.2.1 For a curve γ and a point z0 not on γ , is an
i2π γ z − z0
integer.

Proof: Let ϕ(t), t ∈ [a, b] be a parametric equation of the closed curve γ .

Then we have ϕ(a) = ϕ(b) and

b
1 dz 1 ϕ  (t) dt
= .
i2π z − z0 i2π ϕ(t) − z0
γ a

s ϕ  (t) dt
If F(s) = ϕ(t)−z0 , ∀s ∈ [a, b], then F is a differentiable function on
a
ϕ (s) 
[a, b] and F  (s)
= ϕ(s)−z 0
, ∀s ∈ [a, b], by Theorem 4.1.11. To conclude
the theorem, we shall show that F(b) is an integral multiple of i2π . If
ψ(s) = exp(−F(s))(ϕ(s) − z0 ), ∀s ∈ [a, b], then

ψ  (s) = − exp(−F(s))F  (s)(ϕ(s) − z0 ) + exp(−F(s))ϕ  (s)

= − exp(−F(s))ϕ  (s) + exp(−F(s))ϕ  (s) = 0,

we get that ψ is a constant function on [a, b]. Therefore,

exp(−F(a))(ϕ(a) − z0 ) = exp(−F(b))(ϕ(b) − z0 ),

and hence, exp(−F(a)) = exp(−F(b)) ⇒ exp( F(b)) = 1, as F(a) = 0. Using

Lemma 2.4.20, it follows that F(b) = i2kπ for some k ∈ Z. 

Definition 4.2.2 Let γ be a closed curve and z0 be a point not on γ , then

we define the winding number of γ with respect to z0 or the index of z0 with
1  dz
respect to γ , by WN(γ , z0 ) = .
i2π γ z − z0

Definition 4.2.3 A region determined by a closed curve γ is defined by a

component of the complement of γ in the extended complex plane.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 183 — #19

i i

Complex Integration 183

The above definition can be understood easily by the following diagram.

S3
R2 S4

S1
S2
R1

g Γ

Remark 4.2.4: For a given closed curve, among the regions determined by
γ , there must be only one unbounded region determined by γ . That region is
treated as the region containing ∞ in C∞ .
In the above diagrams, R2 and S4 are the unbounded regions determined by
γ and  , respectively.

THEOREM 4.2.5 Let γ be a closed curve in C and z0 ∈ C not on γ . Then,

1. WN (−γ , z0 ) = −WN(γ , z0 ),

2. WN (γ , z0 ) = WN (γ , w0 ) if z0 and w0 belong to a same region

determined by γ ,

3. WN (γ , z0 ) = 0 if z0 belongs to the unbounded region determined by γ .

Proof: Let z0  γ .

1. Using Lemma 4.1.25, we get

1 dz 1 dz
WN (−γ , z0 ) = =− = −WN(γ , z0 ).
i2π z − z0 i2π z − z0
−γ γ

2. Case 1. First, we show that if the line segment Lz0 ,w0 joining z0 and
w0 is completely contained
 in a same region determined by γ . We
 z − z0 1 1
know that log = − , provided the expression
z − w0 z − z0 z − w0
z − z0
belongs to the domain of log, where it is analytic. From Result
z − w0

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 184 — #20

i i

184 Winding Number and Cauchy’s Theorems

3.3.9, we recall that log is analytic on = C \ {(x, 0) : x ≤ 0}. As

z − z0 z − z0
is real and ≤ 0 ⇒ = −k for some k > 0
z − w0 z − w0
z0 + kw0
⇒ z=
1+k
⇒ z lies on Lz0 ,w0 .
 
z − z0
Hence, log is differentiable on the complement of Lz0 ,w0 .
z − w0
Since Lz0 ,w0 does not intersect γ , we obtain that γ belongs to the
complement of Lz0 ,w0 . Thus, using Corollary 4.1.29, we get
   
1 1  z − z0
− dz = log dz = 0.
z − z0 z − w0 z − w0
γ γ

Case 2. Next, let z0 and w0 be two points in the same region determined
n
by γ . Then, we can join z0 and w0 by a polygon ∪ Laj−1 ,aj contained in
j=1
the same region determined by γ , where a0 = z0 , an = w0 , and Laj−1 ,aj
is the line segment joining ai−1 and ai . Applying case 1, repeatedly, we
get

WN (γ , z0 ) = WN (γ , a0 )= WN(γ , a1 ) = · · · = WN(γ , an )
= WN (γ , w0 ).

3. As γ is the continuous image of [a, b], it is a compact subset of C,

and hence, it is bounded. Therefore, we choose M > 0 such that γ ⊆
B(0, M). Let z0 be the given point belonging to the unbounded region
determined by γ . Then for any w0  B(0, M), z0 and w0 belong to the
1
same region determined by γ . Since is an analytic function in
z − w0
B(0, M), using (2) and Cauchy’s theorem for simply connected region
(Theorem 4.2.14),1 we have

1 dz
WN (γ , z0 ) = WN(γ , w0 ) = = 0.
i2π z − w0
γ

1 Though Theorem 4.2.14 is proved later, and its proof does not depend on the present
theorem.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 185 — #21

i i

Complex Integration 185

From the above properties. WN(γ , z0 ) is called the winding number of γ with
respect to z0 . 

1 if |z − a| < r
Example 4.2.6 If C is |ζ − a| = r, then WN (C, z) = .
0 if |z − a| > r
Since the parametric equation of the circle |ζ − a| = r is given by ζ =
a + r exp(iθ ), θ ∈ [0, 2π], we get

2π
1 dζ 1 ir exp(iθ )dθ
= = 1.
i2π ζ −a i2π r exp(iθ )
|ζ −a|=r 0

If |z − a| < r, then a and z are lying in the same region determined by C , and
using Theorem 4.2.5(2), we get WN(C, z) = WN(C, a) = 1. If |z − a| > r,
then by using Theorem 4.2.5(3), we get WN(C, z) = 0.
There is an interesting theorem, namely, Jordan curve theorem, which
states that the complement of a simple closed curve has exactly to regions.
Although it can be realized geometrically, its proof is too lengthy. Hence, we
prefer to omit this theorem.

Definition 4.2.7 A region in C is called a simply connected region if the

complement of in the extended complex plane is also connected.
Geometrically, every closed curve in a simply connected region
encloses only the points of . We prove this statement rigorously through
the following theorem.

A simply connected region

THEOREM 4.2.8 Let be a region in C. is simply connected iff

WN (γ , z0 ) = 0, for every z0  and for every curve γ in .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 186 — #22

i i

186 Winding Number and Cauchy’s Theorems

Proof: Assume that is simply connected, that is, complement of in

C∞ is connected. Since γ ⊆ , we have C∞ \ ⊆ C∞ \γ . How-
ever, C∞ \γ is a disjoint union of the regions determined by γ and
∞ ∈ C∞ \ . Hence, C∞ \ is contained in the unbounded region de-
termined by γ . Hence, by Theorem 4.2.5(3), we get WN(γ , z0 ) = 0,
∀z0  .
Conversely, assume that WN(γ , z0 ) = 0, ∀z0  , for every closed curve
γ in . Suppose, C∞ \ is not connected. Then, C∞ \ = A ∪ B, where
A  ∅, B  ∅, A ∩ B = ∅ and A and B are closed subsets of C∞ \ .
Using ∞ ∈ C∞ \ , without loss of generality, we assume that ∞ ∈ A and
let b ∈ B. Being C∞ \B an open set containing ∞, it will be of the form
C∞ \K for some compact subset K of C. (Cf. Definition 1.5.1.) Thus, B is a
compact subset of C, and hence, it is bounded. As A \ {∞} and B are disjoint
closed subsets of C with one of them is compact, if we let δ = inf{|x − y| :
x ∈ A \ {∞} and y ∈ B}, then δ > 0 (since A \ {∞} is closed in C and by
Theorem 1.4.41). Now, we cover the entire complex plane by a net consisting
of squares of diameter less than δ such that the point b lies at the centre of
a square. Let {Sα : α ∈ } be the collection of all squares, which
 intersect
B. Being B is bounded,  is a finite set. Then, let σδ = ∂Sα , where
α∈
∂Sα is the boundary of the square Sα . After removing the common edges
with opposite directions present in σδ , it is a closed curve, and it does not
intersect B.

sd

Ω B

We claim that σδ ∩ A = ∅. Otherwise, there exists z0 ∈ σδ ∩ A. Then,

there exists a β ∈  such that Sβ ∩ B  ∅ and z0 ∈ Sβ . Then, we can
choose w0 ∈ Sβ ∩ B, and hence, |z0 − w0 | ≥ δ . However, z0 , w0 ∈ Sβ ⇒
|z0 − w0 | < δ , which is a contradiction. Hence, σδ does not intersect A.
Therefore, σδ ⊆ C\(A ∪ B) = . From b ∈ B ⊆ (C∞ \ ), we have b  .
Therefore, by assumption, WN(σδ , b) = 0. On the other hand, WN(σδ , b) =

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 187 — #23

i i

Complex Integration 187


 1 if α = α0
WN (∂Sα , b) = 1, since WN(∂Sα , b) = where b ∈ Int Sα0 .
α∈ 0 if α  α0 ,
This is a contradiction. Therefore, C∞ \ is connected. Thus, is simply
connected. 

Definition 4.2.9 A region is called a multiply connected region, if it is not

simply connected.
That is, the complement of in C∞ has more than two components or
equivalently; there exists a closed curve γ in , which encloses a point of
complement of .

A multiply connected region

Definition 4.2.10 Let γ be a closed curve in a region . We say that γ is

homologous to 0 in if n(γ , a) = 0, ∀a  . In this case, we write γ ∼ 0
in .
The following result is an immediate consequence of Theorem 4.2.8.

RESULT 4.2.11 Let be a region. Then, is simply connected iff γ ∼ 0 in

, ∀ closed curve γ in .
In the following theorem, we mean a rectangle R by [a, b] × [c, d], for
some a, b, c, d ∈ R with a < b and c < d . We also use the notation ∂R to
denote the boundary of the rectangle R, which is a closed curve.

THEOREM 4.2.12 (Cauchy’s theorem for rectangle)

If
 f is analytic on a region and R is a rectangle contained in , then
f (z) dz = 0.
∂R

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 188 — #24

i i

188 Winding Number and Cauchy’s Theorems


Proof: Let I(R), D(R), and L(R) denote f (z) dz, the diameter of R, and
∂R
the perimeter of R, respectively. If R is subdivided into four equal rectangles
S1 , S2 , S3 , and S4 of same size, then we have


4 
4
I(R) = I(Sk ) ⇒ |I(R)| ≤ |I(Sk )|.
k=1 k=1

This implies that there exists at least one k ∈ {1, 2, 3, 4} such that |I(Sk )| ≥
1
|I(R)|. Denote that rectangle Sk by R1 . Next, we subdivide R1 into four
4
equal subrectangles of same size as before and choose the one say R2
1
such that |I(R2 )| ≥ |I(R1 )|. Proceeding further, we get a sequence of
4
rectangles R ⊇ R1 ⊇ R2 ⊇ · · · such that
1 1
|I(Rn+1 )| ≥ |I(Rn )| ⇒ |I(Rn )| ≥ n |I(R)|. (4.7)
4 4
1 1
L(Rn+1 ) = L(Rn ) ⇒ L(Rn ) = n L(R). (4.8)
2 2
1 1
D(Rn+1 ) = D(Rn ) ⇒ D(Rn ) = n D(R). (4.9)
2 2
S3 = R1

S4
R3
R2

S1 S2

Since {Rn } is a decreasing sequence of compact sets with D(Rn ) → 0 as

n → ∞, by Cantor’s intersection theorem (Theorem 2.4.33),
∞
∩ Rn = {z0 }, for some z0 ∈ R. (4.10)
n=1

As f is analytic at z0 , given  > 0, there exists δ > 0 such that

 
 f (z) − f (z0 )  
0 < |z − z0 | < δ ⇒   − f (z0 ) <

.
z − z0 D(R)L(R)

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 189 — #25

i i

Complex Integration 189

Therefore, we have
  
0 < |z − z0 | < δ ⇒  f (z) − f (z0 ) − (z − z0 )f  (z0 ) < |z − z0 |.
D(R)L(R)
Using equations (4.9) and (4.10), we choose m ∈ N such that D(Rm ) < δ , and
hence, Rm ⊆ B(z0 , δ). If f1 (z) = 1 and f2 (z) = z, ∀z ∈ C, then f1 and f2 are
the derivatives of the analytic functions F1 and F2 defined by F1 (z) = z and
z2
F2 (z) = , ∀z ∈ C, respectively. Therefore, by Corollary 4.1.29, we have
 2 
dz = 0 and z dz = 0. Now,
∂Rm ∂Rm

|I(R)| ≤ 4m |I(Rm )| (by equation (4.7))

 
 
 
m 
= 4  f (z) dz
 
∂Rm 
 
 
 
m  
= 4  [f (z) − f (z0 ) − (z − z0 )f (z0 )] dz
 
∂Rm 

≤ 4m | f (z) − f (z0 ) − (z − z0 )f  (z0 )| |dz|

∂Rm

< 4m |z − z0 | |dz|
D(R)L(R)
∂Rm

< 4m D(Rm ) |dz|
D(R)L(R)
∂Rm

= 4m D(Rm )L(Rm ) (by Theorem 4.1.27)
D(R)L(R)

= 4m 4−m D(R)L(R) (by equations (4.8) and (4.9))
D(R)L(R)
= .
As  > 0 is arbitrary, we get |I(R)| = 0. 

THEOREM 4.2.13 Let be a region, F ⊆ be a finite set and let f be an

analytic function on \F . If f is bounded in B(a, r)\{a}, for some
 ra > 0, for
every a ∈ F , then for every rectangle R such that ∂R ⊂ \F , f (z) dz = 0.
∂R

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 190 — #26

i i

190 Winding Number and Cauchy’s Theorems

Proof: Let F∩R = {ak : k = 1, 2 . . . , m}. Then, ak ∈ Int R, ∀k = 1, 2, . . . , m.

By assumption, there exists rk > 0 such that B(ak , rk ) ⊆ Int R, and
0 < |z − ak | < rk ⇒ | f (z)| ≤ Mk , for some Mk > 0, ∀1 ≤ k ≤ m.

 each k = 1, 2, . . . , m, we choose δk ∈ R such that 0 < δk <

For

min , rk .
mMk
Find a subrectangle Sk of R such that ak ∈ Int Sk , and its circumference
L(Sk ) of Sk is less than δk , ∀1 ≤ k ≤ m. If D(Sk ) is the diameter of Sk , then
we get
z ∈ ∂Sk ⇒ z ∈ Sk ⇒ |z − ak | ≤ D(Sk ) ≤ L(Sk ) < δk . (4.11)
Now, extend all the line segments that are used to construct the rectangles Sk
up to the boundary of R so that the original rectangle R is subdivided into
finite number of rectangles.

Hence, by a similar argument used in the proof of the previous theorem, we

get

m 
p
f (z) dz = f (z) dz + f (z) dz,
∂R k=1∂S j=1 ∂T
k j

where Tj ’s are subrectangles

 of R not containing any ak . Applying the previ-
ous theorem, we get f (z) dz = 0, ∀1 ≤ j ≤ p. Now, for every 1 ≤ k ≤ m,
∂Tj
invoking equation(4.11), we get
 
   
 
 f (z) dz ≤ | f (z)| |dz| ≤ | f (z)| |dz|
∂Sk  ∂S ∂Sk
k
≤ Mk |dz| = Mk L(Sk )
∂Sk

< Mk δk < Mk mM k
= m .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 191 — #27

i i

Complex Integration 191

Therefore,  
   
   m   
   
m

 f (z) dz ≤  f (z) dz < = .

  k=1  
 k=1
m
∂R ∂Sk

Thus, f (z) dz = 0. 
∂R

THEOREM 4.2.14 (Cauchy’s theorem for a simply connected region)

Let F be a finite2 subset of a simply connected region and f be an analytic
function on \F . If f is bounded in B(a, r)\{a}, for some
 ra > 0, for every
a ∈ F , then for every closed curve γ ⊂ \F , we get f (z) dz = 0.
γ

Proof: First we show that there exists an analytic function  on such that
 = f . This part of this theorem is almost similar to that of the proof of
Result 4.1.30. Let (a, b) ∈ \F be arbitrarily fixed. For every (x0 , y0 ) ∈ \F ,
let γ(x0 ,y0 ) ⊆ \F be a polygon joining (a, b) and (x0 , y0 ) consisting of the
line segments parallel to the coordinate axes and the line segment incident
with (x0 , y0 ), which is horizontal (such a polygon exists by Theorem 1.4.31).
Furthermore, as F is a finite set, we can choose thepolygon such that it does
not pass through any a ∈ F . Define (x0 , y0 ) = f (z) dz. First, we show
γ(x0 ,y0 )
that the definition of (x0 , y0 ) is independent of the choice of the polygon
γ(x0 ,y0 ) .

(a,b) (x0,y0)

If σ(x0 ,y0 ) is another polygon joining (a, b) and (x0 , y0 ), then γ(x0 ,y0 ) ∪
(−σ(x0 ,y0 ) ) is a finite union of boundary of rectangles. As is simply con-
nected, all rectangles enclosed by γ(x0 ,y0 ) ∪ (−σ(x0 ,y0 ) ) are contained in .
2 Note that empty set is a finite set, and hence, this theorem is also true for the case F = ∅.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 192 — #28

i i

192 Winding Number and Cauchy’s Theorems

Hence, by Theorem 4.2.13, we get

f (z) dz = 0 ⇒ f (z) dz = f (z) dz.

γ(x0 ,y0 ) ∪(−σ(x0 ,y0 ) ) γ(x0 ,y0 ) σ(x0 ,y0 )

Thus,  is well defined. Let  > 0 be given. Using the continuity of f at

(x0 , y0 ), choose δ > 0 such that

(x, y) ∈ B((x0 , y0 ), δ) ⊂ \F ⇒ | f (x, y) − f (x0 , y0 )| < .

Let h ∈ R with 0 < |h| < δ . Then, the line segment L(x0 ,y0 ),(x0 +h,y0 ) ⊆ and
is horizontal. Therefore, for |h| < δ , we have

(x0 + h, y0 ) = f (z) dz,

γ(x0 ,y0 ) ∪L(x0 ,y0 ),(x0 +h,y0 )

and hence,
 
 (x0 + h, y0 ) − (x0 , y0 ) 
 − f (x , y ) 
 h
0 0 
 
 
1 
 
= f (z) dz − f (x0 , y0 )
h 
 L(x ,y ),(x +h,y ) 
0 0 0 0
 h 
 
1 
=  f (x0 + t, y0 ) dt − f (x0 , y0 )
h 
0
h
1
≤ | f (x0 + t, y0 ) − f (x0 , y0 )| |dt|
|h|
0
1
< |h| = .
|h|
∂
Thus, (x0 , y0 ) = f (x0 , y0 ).
∂x
∂
Similarly, we can prove that (x0 , y0 ) = if (x0 , y0 ). Furthermore, we get
∂y
∂ ∂ ∂ ∂
(x0 , y0 ) = −i (x0 , y0 ) = f (x0 , y0 ) (the C–R equation), and ,
∂x ∂y ∂x ∂y
are continuous at (x0 , y0 ). Therefore, by Theorem 2.2.28, we get that  is

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 193 — #29

i i

Complex Integration 193

differentiable at (x0 , y0 ). As (x0 , y0 ) ∈ \F is arbitrary, we get that  is

∂
analytic on and  = = f . Therefore, using Corollary 4.1.29, we get
 ∂x
f (z) dz = 0, for all closed curve γ in . 
γ

THEOREM 4.2.15 (Cauchy–Goursat theorem)

 γ be a simple closed curve and f be analytic inside and on γ , then
Let
f (z) dz = 0.
γ

Proof: As f is analytic inside and on γ , for every point z on or inside γ , there

exists rz > 0 such that f is analytic on B(z, rz ). If we put as the union of
all these open balls B(z, rz ) along with the region enclosed by γ , then is a
simply connected region, γ ⊂ , and f is analytic on . Hence, by Cauchy’s
theorem for simply connected region, we get f (z) dz = 0. 
γ

4.3 CAUCHY’S INTEGRAL FORMULA

THEOREM 4.3.1 (Cauchy’s integral formula)
Let f be an analytic function on a simply connected region . Then,
 f (z)
WN (γ , z0 )f (z0 ) = i2π dz, for every closed curve γ in \{z0 }.
1
γ z − z0

f (z) − f (z0 )
Proof: Let F(z) = , ∀z ∈ \{z0 }. Then, F is analytic on \{z0 }.
z − z0
Since
f (z) − f (z0 )
lim F(z) = lim = f  (z0 ),
z→z0 z→z0 z − z0

there exists δ > 0 such that

 
 f (z) − f (z0 ) 

0 < |z − z0 | < δ ⇒  − f (z0 ) < 1.

z − z0

Hence, for every z ∈ B(z0 , r)\{z0 }, we have

 
 f (z) − f (z0 ) 

| f (z)| =   ≤ 1 + | f  (z0 )| < +∞.
z − z0 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 194 — #30

i i

194 Cauchy’s Integral Formula

Therefore,
 by Cauchy’s theorem for simply connected region, we get
F(z) dz = 0. Hence, from the definition of WN(γ , z0 ), it follows that
γ

1 f (z) 1 f (z) − f (z0 )

dz − WN(γ , z0 )f (z0 ) = dz
i2π z − z0 i2π z − z0
γ γ
1
= F(z) dz = 0.
i2π
γ

Hence, the theorem follows. 

THEOREM 4.3.2 Let γ be a curve in C and ϕ be a continuous function on γ .

If
φ(ζ )
Fn (z) = dζ ,
(ζ − z)n
γ

for every z  γ and n ∈ N, then Fn is differentiable on the complement of γ

and Fn = nFn+1 , ∀n ∈ N.

Proof: Let z0  γ be arbitrary. First, we show that if

φ(ζ )
Gm,n (z) = , ∀z  γ and ∀m, n ∈ N,
(ζ − z)m (ζ − z0 )n
γ

then lim Gm,n (z) = Gm,n (z0 ). We choose δ > 0 such that B(z0 , δ) ∩ γ = ∅.
z→z0
 
δ δ
Then, we have |ζ − z| ≥ and |ζ − z0 | ≥ δ, ∀z ∈ B z0 , and ∀ζ ∈ γ . If
 2 2
M = |φ(ζ )| |dζ |, then M < ∞ and
γ

|Gm,n (z) − Gm,n (z0 )|

 
 1 1 
≤   −  |φ(ζ )| |dζ |
(ζ − z)m (ζ − z0 )n (ζ − z0 )m+n 
γ
 
1  1 1 
=  − 
|ζ − z0 |n (ζ − z) (ζ − z0 ) 

γ
m−1 
 
 1 1 
×  |φ(ζ )| |dζ |
 (ζ − z)m−1−k (ζ − z0 )k 
k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 195 — #31

i i

Complex Integration 195


m−1
(by using am − bm = (a − b) am−1−k bk )
k=0

|z − z0 | 
m−1
|φ(ζ )| |dζ |
≤
δn |ζ − z|m−k |ζ − z0 |k+1
k=0 γ

|z − z0 | 
m−1
2m−k
≤ |φ(ζ )| |dζ |
δn δ m−k δ k+1
k=0 γ


m−1
M 2m−k
k=0
≤ |z − z0 | → 0 as z → z0 .
δ m+n+1
 
For z ∈ B z0 , 2δ , we have

Fn (z) − Fn (z0 )
z − z0
 
1 1 1
= − φ(ζ ) dζ
z − z0 (ζ − z)n (ζ − z0 )n
γ

 
n−1
1 1 1 1 1
= − φ(ζ ) dζ
z − z0 (ζ − z) (ζ − z0 ) (ζ − z)n−1−k (ζ − z0 )k
γ k=0

 
n−1
1 z − z0 1 1
= φ(ζ ) dζ
z − z0 (ζ − z)(ζ − z0 ) (ζ − z)n−1−k (ζ − z0 )k
γ k=0


n−1
1 1
= φ(ζ ) dζ
(ζ − z)n−k (ζ − z0 )k+1
k=0 γ


n−1 
n−1 
n−1
φ(ζ )
= Gn−k,k+1 (z) → Gn−k,k+1 (z0 ) = dζ
(ζ − z0 )n+1
k=0 k=0 k=0 γ

= nFn+1 (z0 ), as z → z0 .

This completes the proof of the theorem. 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 196 — #32

i i

196 Cauchy’s Integral Formula

COROLLARY 4.3.3 (Cauchy’s integral formula for derivatives)

Let f be an analytic function on a simply connected region and z0 ∈ .
Then for every n ∈ N, f (n) exists and

n! f (z)
WN (γ , z0 )f
(n)
(z0 ) = dz,
i2π (z − z0 )n+1
γ

for every closed curve not passing through z0 in , and for every n ∈ N.

Proof: Cauchy’s integral formula states that

1 f (ζ )
WN (γ , z0 )f (z0 ) = dζ ,
i2π ζ − z0
γ

where γ is the circle |z − z0 | = r, where r > 0 is such that Cl B(z0 , r) ⊆ .

1  f (ζ )
As f is continuous on γ , by previous theorem, we get that dζ is
i2π γ ζ − z
analytic in B(z0 , r), and for every z ∈ B(z0 , r),
⎛ ⎞
d ⎝ 1 f (ζ ) 1 f (ζ )
dζ ⎠ = dζ
dz i2π ζ −z i2π (ζ − z)2
γ γ
⎛ ⎞
d2 ⎝ 1 f (ζ ) 1×2 f (ζ )
dζ ⎠ = dζ
dz2 i2π (ζ − z) i2π (ζ − z)3
γ γ
⎛ ⎞
d3 ⎝ 1 f (ζ ) 1×2×3 f (ζ )
dζ ⎠ = dζ
dz3 i2π (ζ − z) i2π (ζ − z)4
γ γ

..
.
⎛ ⎞
dn ⎝ 1 f (ζ ) n! f (ζ )
dζ ⎠ = dζ .
dzn i2π (ζ − z) i2π (ζ − z)n+1
γ γ

 f (z)
Therefore, f (n) (z0 ) = n!
dz, ∀n ∈ N. 
i2π
γ (z − z0 )n+1

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 197 — #33

i i

Complex Integration 197

THEOREM 4.3.4 (Weierstrass theorem on convergence of sequence of

analytic functions)
If ( fn ) is a sequence of analytic functions on a region such that ( fn ) con-
verges to f uniformly on every compact subset of , then f is analytic and
( fn ) converges to f  uniformly on every compact subset of .

Proof: Let z ∈ be arbitrary. Then, choose r > 0 such that Cl B(z, r) ⊆ .

Then, by Cauchy’s integral formula, we have
1 fn (ζ )
fn (w) = dζ , ∀w ∈ B(z, r) ∀n ∈ N.
i2π ζ −w
|ζ |=r

By assumption, we have fn → f uniformly on |ζ | = r as n → ∞. For all

w ∈ B(z, r), we have
1 f (ζ )
f (w) = dζ .
i2π ζ −w
|ζ |=r

Since fn → f as n → ∞ uniformly on |ζ | = r and each fn is continuous,

we have f is continuous on |ζ | = r, by Theorem 1.6.31. Hence, by Theorem
4.3.2, we obtain that f is analytic on B(z, r). As z ∈ is arbitrary, f is analytic
on . Again by using ( fn ) converges uniformly to f on |ζ | = r, given  > 0,
there exists N ∈ N such that | fn (ζ ) − f (ζ )| <  , ∀ζ with |ζ | = r, whenever
n ≥ N . By Cauchy’s integral formula for derivatives, we have
1 fn (ζ )
fn (w) = dζ , ∀w ∈ B(z, r)
i2π (ζ − w)2
|ζ |=r

and
1 f (ζ )
f  (w) = dζ , ∀w ∈ B(z, r),
i2π (ζ − w)2
|ζ |=r
 r
For all w ∈ Cl B z, , and for all n ≥ N ,
2
   | fn (ζ ) − f (ζ )| 
 f (w) − f  (w) ≤ 1 |dζ | ≤
1
|dζ | =
4
.
n
2π |ζ − w| 2 2π r2 r
|ζ |=r |ζ |=r 4

Therefore, for every z ∈ , there exists ρz > 0 such that fn → f  uniformly
on Cl B(z, ρz ). Let K be a compact subset of . Then, {B(z, ρz ) : z ∈ K} is

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 198 — #34

i i

198 Cauchy’s Integral Formula

a collection of open sets with K ⊆ ∪ B(z, ρz ). Since K is compact, there

z∈K
m m
exist z1 , z2 , . . . , zm ∈ K such that K ⊆ ∪ B(zj , ρzj ) ⊆ ∪ Cl B(zj , ρzj ). Since
j=1 j=1
fn → f  uniformly on Cl B(zj , ρzj ) as n → ∞, for a given  > 0, there exists
Nj ∈ N such that

| fn (w) − f  (w)| < , ∀w ∈ Cl B(zj , ρzj ), ∀n ≥ Nj , ∀1 ≤ j ≤ m.

If N0 = max{Nj : 1 ≤ j ≤ m}, then | fn (w) − f  (w)| <  , ∀w ∈ K , ∀n ≥ N0 .

Thus, ( fn ) converges to f  uniformly on every compact subset of . 
The following algorithm is useful to evaluate integrals over a given simple
closed curve γ .

Algorithm 4.3.5 (To find φ (z) dz by Cauchy’s integral formula)
γ
Step 1: Let the given function be φ and find its poles. (If φ is a rational
function, the zeroes of the denominator of φ are the poles of φ .)

Step 2: Check which of the poles are lying outside γ .


Step 3: (a) If there is no pole lying inside γ , then φ(z) dz = 0.
γ
(b) If there is only one pole z0 , which is enclosed by γ , then put

f (z) = (z − z0 )m φ(z),

where m is the order of the pole z0 of φ . Then, we find

  f (z)
φ(z) dz = dz using Cauchy’s integral formula as
γ γ (z − z0 )m
follows.
f (z) i2π
φ(z) dz = dz = f (m−1) (z0 ).
(z − z0 )m (m − 1)!
γ γ

(c) If there are more than one pole, say z1 , z2 , . . . , zn , enclosed by γ ,

then put

f (z) = (z − z1 )m1 (z − z2 )m2 · · · (z − zn )mn φ(z),

where m1 , m2 , . . . , mn are the orders of the poles z1 , z2 , . . . , zn ,

respectively.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 199 — #35

i i

Complex Integration 199

1
(d) Split (z−z1 )m1 (z−z2 )m2 ···(z−zn )mn into partial fractions as follows.


m1
A1,k 2 m
A2,k n
An,k
m
+ + · · · + .
(z − z1 ) k (z − z2 ) k (z − zn )k
k=1 k=1 k=1

(Note that in the field of complex numbers, there is no non-

1
splittable factor. For example, 2 should be written as
z + a2
A B
+ . Hence, in a partial fraction expansion, a general
z + ia z − ia
C
term should be of the form .)
(z − a)p
 
n 
mj  f (z)
(e) Then write φ(z) dz as Aj,k (z−zj )k
dz.
γ j=1 k=1 γ

(f) Doing Step 3(b) repeatedly, we can find the given integral.

Example 4.3.6 Evaluate the following integrals using Cauchy’s integral

formula:
 z3 + 1  3z + 1
(1) dz. (6) dz.
|z|=3 z + 3z − 10 |z|=5 (z − 1)(z − 2)
2 2

 z2 + 1  exp(z)(3z2 + 4)
(2) dz. (7) dz.
|z−3|=1 z−1 |z−(1−i)|=3 (z − 1)3
 z4 − 4z − 6  z2 + 5z + 6
(3) dz, where γ is (8) dz.
γ z2 − 6z + 5 |z+1|=2 (z2 + 1)(z − 1 + i)
 exp(z)
the square with vertices (3, 3), (9) π
  dz.
(−3, 3), (−3, −3), and (3, −3). |z|=4 z z − 2 (z − π )
 2z + 3  z+i
(4) dz. (10) dz, where γ
|z|=5 z 2 − 2z − 3
γ (z + 4)(z−1−i)
 z2 − 3
(5) dz. is the boundary of [0, 2]×[0, 3].
|z−1|=4 (z2 + 3z + 2)(z + 6)

Solution:
z3 + 1
(1) Let φ(z) = . As z2 + 3z − 10 = 0 ⇒ z = −5, z = 2, the
+ 3z − 10
z2
poles of φ are −5 and 2. We know that 2 is enclosed by |z| = 3, and
−5 is not enclosed by |z| = 3 because |2| = 2 < 3 and | − 5| = 5 > 3.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 200 — #36

i i

200 Cauchy’s Integral Formula

z3 + 1
We put f (z) = . Therefore,
z+5
z3 + 1 f (z) 23 + 1 18iπ
dz = dz = i2π f (2) = i2π = .
z2 + 3z − 10 z−2 2+5 7
|z|=3 |z|=3

z2 + 1
(2) Let φ(z) = . Then, clearly 1 is the only pole of φ and is not
z−1
enclosed by |z − 3| = 1 (because |1 − 3| = 2 > 1.) Therefore,
 z2 + 1
dz = 0.
|z−3|=1 z − 1

z4 − 4z − 6
(3) Let φ(z) = . Since z2 − 6z + 5 = 0 ⇒ z = 5, z = 1,
z2 − 6z + 5
the poles of φ are 5 and 1. Obviously, 1 is enclosed by γ and 5 is
z4 − 4z − 6
not enclosed by γ (check geometrically). We put f (z) = .
z−5
Therefore,

z4 − 4z − 6 f (z) 1−4−6 9iπ

dz = dz = i2π f (1) = i2π = .
z2 − 6z + 5 z−5 1−5 2
γ γ

2z + 3
(4) Let φ(z) = . Now, z2 − 2z − 3 = 0 implies that z = −1,
z2 − 2z − 3
and z = 3 are the poles of φ , and both are enclosed by |z| = 5, because
| − 1| = 1 < 5 and |3| = 3 < 5. Therefore, we put f (z) = 2z + 3. Now,
1
we split into partial fractions as follows.
(z + 1)(z − 3)
1 A B
Let = + ⇒ 1 = A(z − 3) + B(z + 1).
(z + 1)(z − 3) z+1 z−3
1
Putting z = 3, we get 4B = 1 ⇒ B = .
4
−1
Putting z = −1, we get −4A = 1 ⇒ A = .
 4
1 1 1 1
Hence, = − . Therefore,
(z + 1)(z − 3) 4 z−3 z+1
2z + 3 f (z)
dz = dz
z2 − 2z − 3 (z + 1)(z − 3)
|z|=5 |z|=5

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 201 — #37

i i

Complex Integration 201

⎛ ⎞
1⎜ f (z) f (z) ⎟
= ⎝ dz − dz⎠
4 z−3 z+1
|z|=5 |z|=5

i2π iπ
= ( f (3) − f (−1)) = (9 − 1) = 4iπ .
4 2

z2 − 3
(5) Let φ(z) = . As (z2 + 3z + 2)(z + 6) = 0, it follows
(z2 + 3z + 2)(z + 6)
that z = −1, z = −2, and z = −6 are the poles of φ . Furthermore,
−1 and −2 are enclosed by |z − 1| = 4, and −6 is not enclosed by
|z − 1| = 4 because | − 1 − 1| = 2 < 4, | − 2 − 1| = 3 < 4, and
z2 − 3
| − 6 − 1| = 7 > 4. Hence, we take f (z) = . Now, we split
z+6
1
into partial fractions as follows.
(z + 1)(z + 2)
1 A B
= + ⇒ 1 = A(z + 2) + B(z + 1).
(z + 1)(z + 2) z+1 z+2
Putting z = −2, we get −B = 1 ⇒ B = −1.
Putting z = −1, we get A = 1.
1 1 1
Hence, = − . Therefore,
(z + 1)(z + 2) z+1 z+2

z2 − 3 f (z)
dz = dz
(z2 + 3z + 2)(z + 6) (z + 1)(z + 3)
|z−1|=4 |z−1|=4

f (z) f (z)
= − dz + dz
z+2 z+1
|z−1|=4 |z−1|=4

= −i2π ( f (−2) − f (−1))

 
(−2)2 − 3 (−1)2 − 3
= −i2π −
−2 + 6 −1 + 6
 
iπ 1 2 3iπ
= − − =− .
2 4 5 40

3z + 1
(6) If φ(z) = , then z = 1 and 2 are the poles of φ , and
(z − 1)(z − 2)2
both are enclosed by |z| = 5. Let f (z) = 3z + 1. Now, we split

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 202 — #38

i i

202 Cauchy’s Integral Formula

1
into partial fractions.
(z − 1)(z − 2)2
1 A B C
= + +
(z − 1)(z − 2)2 z − 1 z − 2 (z − 2)2

⇒ 1 = A(z − 2)2 + B(z − 1)(z − 2) + C(z − 1). Putting z = 1, z = 2, and

z = 0, we get, respectively, A = 1, C = 1, 1 = 4A+2B−C ⇒ B = −1.
1 1 1 1
Therefore, = − + . Hence,
(z − 1)(z − 2)2 z − 1 z − 2 (z − 2)2
3z + 1 f (z)
dz = dz
(z − 1)(z − 2)2 (z − 1)(z − 2)2
|z|=5 |z|=5
f (z) f (z) f (z)
= dz − + dz
z−1 z−2 (z − 2)2
|z|=5 |z|=5 |z|=5
 

= i2π f (1) − f (2) + f (2)
= i2π (4 − 7 + 3) = 0.

exp(z)(3z2 + 4)
(7) Let φ(z) = . Then, 1 is the only pole of order 3 for φ ,
(z − 1)3
which is enclosed by |z − (1 + i)| = 3, as |1 − (1 − i)| = 1 < 3. If
f (z) = exp(z)(3z2 + 4), then

f  (z) = exp(z)(3z2 + 6z + 4),

f  (z) = exp(z)(3z2 + 12z + 10) ⇒ f  (1) = 25e.
 exp(z)(3z2 + 4) i2π 
Therefore, dz = f (1) = 25π exp(1)i.
|z−(1+i)=3| (z − 1) 3 2!

z2 + 5z + 6
(8) Let φ(z) = . Then, i, −i, and 1 − i are the poles
+ 1)(z − 1 + i)
(z2
for φ , among
√ which i and −i√ are enclosed by |z − 1| √ = 2, as
|i + 1| = 2 < 2, | − i + 1| = 2 < 2, and |1 − i + 1| = 5 > 2. Let
z2 + 5z + 6 1
f (z) = . Now, we split into partial fractions
z−1+i (z − i)(z + i)
as follows.
1 A B
= + ⇒ 1 = A(z + i) + B(z − i).
(z − i)(z + i) z−i z+i

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 203 — #39

i i

Complex Integration 203

1 1
Putting z = i and z = −i we get, respectively, A = and B = − .
i2 i2
Therefore,

z2 + 5z + 6
dz
(z2 + 1)(z − 1 + i)
|z−1|=2
⎛ ⎞
1 ⎜ f (z) f (z) ⎟
= ⎝ dz − dz⎠
i2 z−i z+i
|z−1|=2 |z−1|=2

= π( f (i) − f (−i))
2 
i + i5 + 6 i2 − i5 + 6
=π −
i−1+i −i − 1 + i
 
i5 + 5 5 − i5
=π −
i2 − 1 −1
= π(1 − i3 + 5 − i5) = π (6 − i8).

exp(z)
(9) Let φ(z) = . Then, 0, π , and π/2 are the poles for
z(z − π/2)(z − π)
φ , and all are enclosed by |z − 1| = 2, as |0| < 5, |π/2| < 5, and
1
|π | < 5. Let f (z) = exp(z). Now, we split into
z(z − π/2)(z − π )
partial fractions as follows.
1 A B C
= + + ⇒ 1 =
z(z − π/2)(z − π ) z z − π/2 z−π
A(z − π/2)(z − π) + Bz(z − π ) + Cz(z − π/2).

Substituting z = 0, z = π/2, and z = π , we get, respec-

2 4 2
tively, A = 2 , B = − 2 , and C = . Therefore, we have
π  π π2 
1 2 1 2 1
= 2 − + . Hence,
z(z − π/2)(z − π ) π z z − π/2 z−π

exp(z)
dz
z(z − π/2)(z − π )
|z|=5

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 204 — #40

i i

204 Cauchy’s Integral Formula

⎛ ⎞
2 ⎜ f (z) f (z) f (z) ⎟
= ⎝ dz − 2 dz + dz⎠
π2 z z − π/2 z−π
|z|=5 |z|=5 |z|=5

2
= i2π (f (0) − 2f (π/2) + f (π ))
π2
i4
= (1 − 2 exp(π/2) + exp(π )).
π

z+i
(10) Let φ(z) = . Then, −4 and 1 + i are the poles for φ .
(z + 4)(z − 1 − i)
Since the given curve γ is the boundary of the rectangle with vertices
(0, 3), (2, 0), (2, 3), and (0, 3), 1 + i is enclosed by γ , and −4 is not
z+i
enclosed by γ . Therefore, let f (z) = . Hence,
z+4

z+i f (z)
dz = dz
(z + 4)(z − 1 − i) z−1−i
γ γ
1 + i2 π
= i2πf (1+i) = i2π = (−9 + i7).
5+i 13

Exercise 4.3.7 Evaluate the following integrals using Cauchy’s integral

formula:
 exp(z)
(1) dz.
|z|=2 (z + 1)(z − 3)2

 dz
(2) dz.
|z+i2|=2 z2 +1

 sin(z)
(3) dz.
|z|=2 z+i

 exp(z)
(4) dz.
|z|=1 (z − 2)3

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 205 — #41

i i

Complex Integration 205

 z2 exp(z)
(5) dz.
|z+ 12 |=1
(z + 1)(z − 2)

 z+1
(6) dz.
|z−1|=2 z2 − 9z + 20

 z exp(z)
(7) dz.
|z|=3 z−1

 z
(8) dz.
|z−(2+i)|=2 (z − (1 + i))

 exp(z)
(9) dz.
|z|=4 (z + 1)(z − 3)2

 z2 + 1
(10) dz.
|z|=3 (z2 − 1)(z − 2)

 z
(11) dz.
|z|=2 (z − 1)2

 2z + 3
(12) dz.
|z−i2|=4 (z − (1 − i))2

 ez
(13) dz.
|z|=4 (z2 + 1)3

 z+i
(14) dz.
|z−1|=4 (z − 1)2 (z − 2)(z − 6)

 exp(z + 1)
(15) dz.
|z−2|=3 (z + 6)4

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 206 — #42

i i

206 Cauchy’s Integral Formula

iπ exp(−1) i2π exp(−1)

Answers: (1) ; (2) −π ; (3) 2π sinh(1); (4) 0; (5) ;
8 −3
iπ
(6) 0; (7) i2π exp(1); (8) 2π(−1 + i); (9) (exp(−1) + 3 exp(3)); (10) i2π ;
8
π π
(11) i2π ; (12) 4iπ ; (13) (2 sin(1) − 3 cos(1); (14) (1 − 6i); (15) 0
4 50

RESULT 4.3.8 (Cauchy’s estimate)

If f is analytic on a closed disk Cl B(a, r), then for each n ∈ N, and 0 < r <  ,
Mr n!
we have | f (n) (a)| ≤ n , where Mr = sup | f (z)|.
r |z|=r

Proof: Let γ be the circle |z − a| = r. By Cauchy’s integral formula for

derivatives, we get
 
   
 (n)   n! f (z) 
 f (a) =  dz
 i2π (z − a) n+1
 
γ
 
n!  f (z) 
≤  
2π  (z − a)n+1  dz
γ
 
n!  Mr 
≤  
2π  rn+1  dz
γ
Mr n! Mr n!
≤ 2π r = n .
2πrn+1 r

Hence, the result follows. 

THEOREM 4.3.9 (Morera’s theorem) 

Let f be a continuous function on such that f (z) dz = 0 for every closed
γ
curve γ in , then f is analytic on .

Proof: As f (z) dz = 0 for every closed curve γ in , using Result 4.1.30,
γ
there exists an analytic function F on such that F  = f . By Corollary 4.3.3,
F  is analytic on , and hence, f is analytic on . 

Definition 4.3.10 An analytic function on C is called an entire function.

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 207 — #43

i i

Complex Integration 207

Clearly, every polynomial, exp, sin, and cos are entire functions. Obviously,
every constant function is a bounded entire function but the interesting fact
is that the converse of this statement is also true, which follows.

THEOREM 4.3.11 (Liouville’s theorem)

Every bounded entire function is constant.

Proof: Let f be a bounded entire function such that sup | f (z)| ≤ M for some
z∈C
M > 0. For an arbitrary z ∈ C and for an arbitrary r > 0, from Cauchy’s
estimate, we get
Mr
| f  (a)| ≤ , where Mr = sup | f (z)|.
r |z|=r

As Mr ≤ M, ∀r > 0, we have
M
| f  (a)| ≤ → 0 as r → ∞.
r
Thus, f  (z) = 0, ∀z ∈ C. Hence, by Theorem 2.2.19, we get that f is a constant
function. 

THEOREM 4.3.12 (Fundamental theorem of algebra)

Every polynomial of degree n with complex coefficients has exactly n zeroes
in C including multiplicities.

Proof: First, we prove that every non-constant polynomial has a zero in C.

1
Let P be a non-constant polynomial. If P has no zero in C, then is an entire
P
function, as P is an entire function (See Example 2.1.9). Now, we claim that
1 n
is bounded on C. If P(z) = ak zk , with an  0, then for |z| ≥ 1, we have
P k=0
 n−1 
 n   

|P(z)| ≥ an z  −  k
ak z 
 
k=0

n−1  
 n   k
≥ an z  − ak z 
k=0

 n 
n−1
≥ an z  − |z|n−1 |ak |
k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 208 — #44

i i

208 Cauchy’s Integral Formula

(since |z|k ≤ |z|n−1 if k ≤ n − 1)

 

n−1
≥ |z| n−1
|an z| − |ak |
k=0


n−1
≥ |an z| − |ak |
k=0
 
1 
n−1
≥ 1 if |z| ≥ 1+ |ak | .
|an |
k=0
 
1
As   is a continuous real-valued function on the compact set
P
   
1 
n−1
z ∈ C : |z| ≤ max 1, 1+ |ak | ,
|an |
k=0
 
 1 
there exists M > 0 such that   ≤ M on the compact set. Therefore,
P(z) 
 
1
  ≤ M + 1 < ∞ on C,
P

1
and hence, is a bounded entire function. Therefore, by Liouville’s theorem
P
1
(Theorem 4.3.11), we get that is a constant function. This implies that P is
P
a constant function. This is a contradiction. Therefore, P has a zero in C. If
α1 is a zero of P, then using Lemma 3.1.3, we can write
P(z) = (z − α1 )Q1 (z), ∀z ∈ C.
Certainly, degree of Q1 is n − 1. If n − 1 > 0, then Q1 has a zero say α2 , and
hence, we can write
P(z) = (z − α1 )(z − α2 )Q2 (z), ∀z ∈ C.
Proceeding further, at the nth stage, we get
P(z) = (z − α1 )(z − α2 ) · · · (z − αn )Qn (z), ∀z ∈ C,
where α1 , α2 , . . . , αn are the zeroes of P, and Qn is a polynomial of degree 0,
which means that Qn is a constant. Therefore, P has exactly n zeroes in C. 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 209 — #45

i i

Complex Integration 209

4.4 GENERAL VERSION OF CAUCHY’S THEOREM

THEOREM 4.4.1 (General version of Cauchy’s theorem)

 f is analytic on and γ is a closed curve in such that γ ∼ 0 in
If , then
f (z) dz = 0.
γ

Proof: As γ is closed and bounded subset of C, it is compact. Since c is

closed and c ∩ γ = ∅, if

r = d(γ , c
) = inf{|z − w| : z ∈ γ , w ∈ c
},

then r > 0, by Theorem 1.4.41. If z = φ(t), t ∈ [a, b] is the parametric

equation of γ , then φ is a continuous function on the compact set [a, b], and
hence, it is uniformly continuous on [a, b], by Theorem 1.6.26. Then, there
exists δ > 0 such that

t, s ∈ [a, b] with |t − s| < δ ⇒ |z(t) − z(s)| < r.

If {t0 , t1 , t2 , . . . , tn } ⊆ [a, b] such that a = t0 < t1 < t2 < . . . < tn = b and

ti − ti−1 < δ , then let γk = φ([tk−1 , tk ]). Then, clearly,

n
γ = ∪ γk and γk ⊂ B(φ(tk ), r), 1 ≤ k ≤ n.
k=1

Moreover, by the choice of r, we have B(φ(tk ), r), which does not intersect c ,
as φ(tk ) ∈ γ ⊂ . Since φ(tk ), φ(tk−1 ) ∈ γk ⊆ B(φ(tk ), r), and B(φ(tk ), r) is a
convex set,3 we can connect φ(tk ) and φ(tk−1 ) by a polygon λk ⊆ B(φ(tk ), r),
which consists of line segments parallel to coordinate axes. Hence, γk ∪(−λk )
is a closed curve contained in the simply connected region B(φ(tk ), r), ∀1 ≤
k ≤ n.

3 A set C is said to be a convex set if for every pair of points of C , the line segment joining
the two points is contained in C .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 210 — #46

i i

210 General Version of Cauchy’s Theorem

f(tk) l
r
f(tk−1)

Hence, by Cauchy’s theorem for simply connected region, we have

f (z) dz = 0 ⇒ f (z) dz = f (z) dz, 1 ≤ k ≤ n.

γk ∪(−λk ) γk λk

n
Therefore, if λ = ∪ λk , then
k=1

f (z)dz = f (z) dz = f (z) dz = f (z) dz. (4.12)

γ n n λ
∪ γk ∪ λk
k=1 k=1

Now, we subdivide the entire complex plane into finite number of bounded
rectangles Bp , 1 ≤ p ≤ μ, and finite number of unbounded rectangles Uq ,
1 ≤ q ≤ ν , by extending each line segment involved in the construction of λ
into straight lines. Now, for each p ∈ {1, 2, . . . , μ} and q ∈ {1, 2, . . . , ν}, we
μ
fix one point bp ∈ Int Bq and uq ∈ Int Uq . If we denote ∪ WN (λ, bp )∂Bp by
p=1
χ , then we claim that after removing the line segments, which are common
sides of two rectangles in χ , we get χ = λ. First, we note that for 1 ≤ j ≤ μ
and 1 ≤ l ≤ ν , we have
 
μ
WN (χ , bj ) = WN ∪ WN(λ, bp )∂Bp , bj
p=1
μ

= WN (λ, bp ) WN(∂Bp , bj )
p=1

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 211 — #47

i i

Complex Integration 211


1 if p = j
= WN(λ, bj ), since WN(∂Bp , bj ) = , (4.13)
0 if p  j

and
 
μ
WN (χ , ul ) = WN ∪ WN(λ, bp )∂Bp , ul
p=1
μ

= WN (λ, bp ) WN (∂Bp , ul )
p=1
= 0, since WN(∂Bp , ul ) = 0. (4.14)
Next, we claim that the multiplicity m of any line segment Lp, j in λ ∪ (−χ ),
which is common to two bounded rectangles Bp and Bj , is zero for some
1 ≤ p, j ≤ μ. Since λ ∪ (−χ ) ∪ (−m∂Bp ) does not have the line Lp, j , the
points bp and bj belong to the same region determined by λ∪(−χ )∪(−m∂Bp ).
Therefore,
WN (λ ∪ (−χ ) ∪ (−m∂Bp ), bp ) = WN(λ ∪ (−χ ) ∪ (−m∂Bp ), bj ).
Hence, by using equation(4.13), we get
WN (λ, bp ) − WN (λ, bj ) − m = WN(λ, bp ) − WN(λ, bj ) − 0 ⇒ m = 0.
Similarly, if Lp,q is the common side of a bounded rectangle Bp and an
unbounded rectangle Uq , then arguing as before by applying equation(4.14),
we get the multiplicity of the line segment Lp,q in λ ∪ (−χ ) as zero. Hence,
our claim holds. 
By Cauchy’s theorem for rectangle, we have f (z) dz = 0, provided
∂Bp
Bp ⊆ . If Bp  , then there exists a ∈ Int Bp and a  . Since γ is
homologous to zero in , we get WN(γ , a) = 0. As γ ∪(−λ) does not enclose
a, we get
WN (γ ∪ (−λ), a) = 0 ⇒ WN(λ, a) = WN(γ , a) = 0.
Since a and bp belong to the same region determined by λ, we get
WN (λ, bp ) = WN(λ, a) = 0.
This implies that for each p = 1, 2, . . . , μ,

either f (z) dz = 0 or WN(λ, bp ) = 0.

∂Bp

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 212 — #48

i i

212 Local Correspondence Theorem and Its Consequences

  μ
 
Hence, it follows that f (z) dz = f (z) dz = WN (λ, bp ) f (z) dz = 0.
λ χ p=1 ∂Bp

m
Definition 4.4.2 A cycle  is a finite union of closed curves ∪ γj .
j=1
 m 
 
m
We define f (z) dz = f (z) dz, so that we get WN(, a) = WN (γj , a).
 j=1 γj j=1

Then, the general version of Cauchy’s theorem can be further extended to the
following form easily.

THEOREM 4.4.3 If f is analytic

 on a region and  is a cycle, which is
homologous to zero in , then f (z) dz = 0.


4.5 LOCAL CORRESPONDENCE THEOREM

AND ITS CONSEQUENCES
In this section, we shall use the following result, which will be proved in
Chapter 5. See Result 5.2.3.

RESULT 4.5.1 If f is an analytic function on a region with a zero of order k

at z0 ∈ , then there exists an analytic function g on such that g is analytic
on , f (z) = (z − z0 )k g(z) on and g(z0 )  0.

THEOREM 4.5.2 Let f be a non-zero analytic function on a region with

zeroes z1 , z2 , . . . , zn , including multiplicities. If γ is a closed curve in such
that γ ∼ 0 in and it is not passing through any zj , then

1 f  (z) n
dz = WN (γ , zj ).
i2π f (z)
γ j=1

Proof: Let 1 ⊆ be a region such that γ and each region determined by

γ are contained in 1 . We also assume that the only zeroes of f in 1 are
{zj : j = 1, 2, . . . , n}. Since zj ’s are the zeroes of f in , by Result 4.5.1, we
can write

f (z) = (z − z1 )(z − z2 ) · · · (z − zn )g(z), ∀z ∈ 1,

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 213 — #49

i i

Complex Integration 213

where g is analytic on 1 and is nowhere zero on 1. Hence,

f  (z) 1 1 1 g (z)
= + + ··· + + , ∀z on γ .
f (z) z − z1 z − z2 z − zn g(z)

We note that gg is analytic on 1 because g and g are analytic and g is
nowhere zero on 1 . Therefore,

1 f  (z)  1
n
dz 1 g (z)  n
dz = + dz = WN (γ , zj ),
i2π f (z) i2π z − zj i2π g(z)
γ j=1 γ γ j=1

 g (z)
since we have dz = 0, by general version of Cauchy’s theorem.
γ g(z)
Note that if we take γ as a simple closed curve in the above theorem, then
1  f  (z)
dz counts the number of zeroes of f inside γ . 
i2π γ f (z)

THEOREM 4.5.3 (Local correspondence theorem)

Let f be a non-constant analytic function on a region , z0 ∈ and f (z0 ) =
w0 . For each w ∈ f ( ), if gw (z) = f (z)−w0 , ∀z ∈ and the order of gw0 at z0
is n, then there exists r > 0 and ρ > 0 such that for each w ∈ B(w0 , ρ)\{w0 },
there exists exactly n distinct roots for gw in B(z0 , r)\{z0 }.

Proof: Since f is non-constant, gw0 is not identically zero, and hence, by the
principle of analytic continuation, it follows that z0 is isolated. Therefore,
there exists r > 0 such that
1. Cl B(z0 , r) ⊆ .
2. gw0  0 on B(z0 , r)\{z0 }, since z0 is an isolated zero of gw0 .
3. f   0 on B(z0 , r)\{z0 }. (This is possible because of the following rea-
son. Since f is non-constant, f  is not identically 0. If f  (z0 ) = 0, then
z0 must be an isolated zero of f  . Then, there exists r > 0. If f  (z0 )  0,
using the continuity of f  , we can find such an r > 0.)
Let γ be the circle |z − z0 | = r. Since z0  γ , we have w0 = f (z0 )  f (γ ).
Hence, we can choose ρ > 0 such that B(w0 , ρ) ∩ f (γ ) = ∅. Then, for each
w ∈ B(w0 , ρ), we get that w and w0 lie in a same region determined by f (γ ).
Therefore, from a property of winding number (Thereom 4.2.5), it follows
that
WN ( f (γ ), w) = WN ( f (γ ), w0 ). (4.15)

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 214 — #50

i i

214 Local Correspondence Theorem and Its Consequences

For each w ∈ B(w0 , ρ), if nw denotes the number of zeroes of gw inside γ

including the multiplicities and if Jw = {1, 2, . . . , nw }, then using Theorem
4.5.2 and using the change of variable ζ = f (z), we get
 1 gw (z)
WN (γ , zj ) = dz
i2π gw (z)
j∈Jw γ
1 f  (z)
= dz
i2π f (z) − w
γ
1 dζ
=
i2π ζ −w
f (γ )
= WN ( f (γ ), w).

 the only one zero of gw0 inside γ is z0 and its multiplicity is n, we have
Since
WN (γ , zj ) = n. Therefore, from equation (4.15), we get
j∈Jw0

 
n= WN (γ , zj ) = WN( f (γ ), w0 ) = WN( f (γ ), w) = WN (γ , zj ).
j∈Jw0 j∈Jw

In other words, the number of zeroes of gw inside B(z0 , r) is n. Furthermore,

gw = f   0 on B(z0 , r)\{z0 } implies that every zero of gw in B(z0 , r) \ {z0 }
is simple, and hence, for each w ∈ B(w0 , ρ) \ {w0 }, gw has exactly n distinct
roots inside B(z0 , r)\{z0 }. 

THEOREM 4.5.4 (Open mapping theorem)

Every non-constant analytic function on a region is an open mapping.

Proof: Let f be a non-constant analytic function on a region . To prove that

f is an open mapping, we have to show that if U is open in , then f (U)
is an open subset of C. Let U be an open subset of and let w0 ∈ f (U).
Then, there exists z0 ∈ U such that w0 = f (z0 ). Choose  > 0 such that
B(z0 , ) ⊆ U . By local correspondence theorem, there exists 0 < r <  and
ρ > 0 such that for every w ∈ B(w0 , ρ), there exists z ∈ B(z0 , r) such that
f (z) = w. Therefore, we have proved that

B(w0 , ρ) ⊆ f (B(z0 , r)) ⊆ f (B(z0 , )) ⊆ f (U).

Thus, f (U) is an open set. 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 215 — #51

i i

Complex Integration 215

COROLLARY 4.5.5 If f : 1 → is an analytic bijection, then f −1 : 2 →

2
−1 ) (w ) = 1
1 is also an analytic function and (f 0 , where z0 = f −1 (w0 ),
f  (z0 )
∀w0 ∈ 2 .

Proof: First, we claim that f  is nowhere zero on 1 . Suppose f  (z0 ) = 0

for some z0 ∈ 1 , then f (z) − f (z0 ) has at least two zeroes at z0 . Therefore,
by local correspondence theorem, there exist r > 0 and ρ > 0 such that for
each w ∈ B(w0 , ρ) \ {w0 }, there exist at least two distinct zeroes for f (z) − w
from B(z0 , r), which is a contradiction to the injectivity of f . Therefore, our
claim holds. If g is the inverse of f , then by open mapping theorem, g is a
continuous map. In fact, g is analytic; indeed, if w = f (z) and f (z0 ) = w0 ,
then w → w0 whenever z → z0 , and hence,

g(w) − g(w0 ) z − z0 1 1
lim = lim = lim =  .
w→w0 w − w0 z→z0 f (z) − f (z0 ) z→z0 f (z) − f (z0 ) f (z0 )
z − z0

Thus, the theorem follows. 

THEOREM 4.5.6 (Maximum modulus principle)

If f is a non-constant analytic function on a region , then | f (z)| has no
maximum in .

Proof: To prove this theorem, we show that for each z0 ∈ , there exists
z1 ∈ such that | f (z1 )| > | f (z0 )|. Since f is a non-constant analytic function
on , then by open mapping theorem, f is an open map. Therefore, f ( ) itself
is an open subset of C. If w0 = f (z0 ), then w0 ∈ f ( ), and hence, there exists
 > 0 such that B(w0 , ) ⊆ f ( ). If |w0 | < r <  and if

r exp(iarg w0 ) if w0  0
w1 = ,
r if w0 = 0

then w1 ∈ B(w0 , ) ⊆ f ( ) and |w1 | > |w0 |. Thus, there exists z1 ∈ such
that f (z1 ) = w1 , and hence, | f (z1 )| = |w1 | > |w0 | = | f (z0 )|. 

COROLLARY 4.5.7 [Minimum modulus principle]

Let f be a non-constant analytic function on a region . If f is nowhere zero
on , then |f | has no minimum in .

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 216 — #52

i i

216 Local Correspondence Theorem and Its Consequences

1
Proof: Since f is nowhere zero on , is also an analytic function on .
f 
1
Hence, by maximum modulus principle,   has no maximum in . That is,
f    
 1   1 
for every z ∈ , there exists z1 ∈ such that    <  , equivalently,
f (z)   f (z1 ) 
|f (z1 )| < |f (z)|. Thus, |f | has no minimum on . 

RESULT 4.5.8 (Schwarz lemma)

If f : B(0, 1) → Cl B(0, 1) is an analytic function such that f (0) = 0, then

1. | f (z)| ≤ |z| on B(0, 1) and | f  (0)| ≤ 1.

2. If | f (z0 )| = |z0 | for some z0  0 or if | f  (0)| = 1, then f (z) = exp(iθ )z,

on B(0, 1) for some θ ∈ R.

f (z)
Proof: If g(z) = for z  0, then g is analytic on B(0, 1)\{0}, and 0 is a
z
removable singularity for g since

f (z) − f (0)
lim g(z) = lim = f  (0)
z→0 z→0 z−0
exists in C. Hence, if we define g(0) = f  (0), then g becomes analytic on
B(0, 1), by Lemma 5.1.2. This implies that g is analytic on Cl B(0, r) for every
0 < r < 1. Since |g| is a continuous function on the compact set Cl B(0, r),
there exists zr with |zr | ≤ 1 such that |g(zr )| = sup |g(z)|. If g is a
z∈Cl B(0,r)
constant function on B(0, 1), then we can choose this zr with |zr | = 1, and if
g is non-constant, then by maximum modulus principle, |g| does not attain its
maximum on |z| < r, and hence, |zr | = r. Since 0 < |zr | < 1, we have
 
 f (zr )  | f (zr )| 1
sup |g(z)| = |g(zr )| =  = ≤ ,
z  r r
z∈Cl B(0,r) r

and hence, by allowing r → 1, we get sup |g(z)| ≤ 1. Hence, by definition

  |z|≤1
 
of g, we get  f (z)z  ≤ 1 on B(0, 1)\{0} and | f  (0)| ≤ 1. Thus, | f (z)| ≤ |z| on

B(0, 1) and | f (0)| ≤ 1.
If | f (z0 )| = |z0 | for some z0 ∈ B(0, 1)\{0} or | f  (0)| = 1, then we get
|g(ζ0 )| = 1 for some ζ0 (= z0 or 0) in B(0, 1). Hence, by maximum modulus
principle, we conclude that g is a constant say c. Since |g(ζ0 )| = 1, we obtain

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 217 — #53

i i

Complex Integration 217

|c| = 1. Therefore, using Corollary 2.4.25, there exists θ ∈ R such that

c = exp(iθ ), and hence, f (z) = g(z) · z = cz, ∀z in B(0, 1). 

RESULT 4.5.9 (Generalized Schwarz lemma)

Let f be analytic on |z| < R such that | f (z)|
 ≤ S , ∀|z| < R. If |a| < R and
 S( f (z) − b)   R(z − a) 
f (a) = b, then  ≤ .
S 2 − bf (z)   R2 − az 

Proof: Consider the linear fractional transform

R(z − a)
w = T(z) = , ∀w ∈ C∞ .
R2 − az
By Example 3.2.26, we get T(B(0, R)) = B(0, 1).
S(ζ − b)
Similarly, if ξ = (ζ ) = , ∀ζ ∈ C∞ , then we have (B(0, S)) =
S 2 − bζ
B(0, 1). If F = ( ◦ f ◦ T −1 ), then F : B(0, 1) → B(0, 1) is analytic and

F(0) = ( f (T −1 (0))) = ( f (a)) = (b) = 0.

Hence, by Schwarz lemma, we get | F(w)| ≤ |w|, ∀w ∈ B(0, 1). If z ∈ B(0, R),
then |T(z)| < 1, and hence,
   
 S( f (z) − b)   R(z − a) 
| F(T(z))| ≤ |T(z)| ⇒ |( f (z))| ≤ |T(z)| ⇒  ≤ .
S 2 − bf (z)   R2 − az 

Thus, the result follows. 

Now, we can see some applications of Schwarz lemma, which are used to
prove the uniqueness of Riemann mapping between a region and open unit
disc, in Result 7.4.5.

COROLLARY 4.5.10 If f : B(0, 1) → B(0, 1) is an analytic bijection, satis-

fying f (0) = 0, then f (z) = cz, ∀z ∈ B(0, 1) for some c ∈ C with |c| = 1.

Proof: Let f : B(0, 1) → B(0, 1) is the given bijection. Then, by Corol-

lary 4.5.5, we get that f −1 : B(0, 1) → B(0, 1) is an analytic function and
1
(f −1 ) (0) =  . Applying Schwarz lemma to f and f −1 , we have
f (0)
 
 1 
|f (0)| ≤ 1 and    = |(f −1 ) (0)| ≤ 1.
 
f (0)

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 218 — #54

i i

218 Local Correspondence Theorem and Its Consequences

The second inequality implies that |f  (0)| ≥ 1. Therefore, |f  (0)| = 1. Then

using the condition for equality in Schwarz lemma, we get f (z) = cz, ∀z ∈ C
for some c ∈ C. 

COROLLARY 4.5.11 If f : B(0, 1) → B(0, 1) is an analytic bijection, then

z−α
f (z) = c , ∀z ∈ B(0, 1) for some c ∈ C with |c| = 1, where α =
1 − αz
−1
f (0) ∈ B(0, 1).

z−α
Proof: If Tα (z) = , ∀z ∈ C, then we claim that Tα : B(0, 1) → B(0, 1)
1 − αz
is an analytic bijection such that Tα (α) = 0 and Tα−1 is also analytic.

If |z| = 1, then let z = exp(iθ) for some θ ∈ R so that

     
 exp(iθ ) − α   exp(iθ) − α   exp(iθ ) − α 

|Tα (z)| =   =   =   = 1.
1 − α exp(iθ )   exp(iθ )(exp(−iθ ) − α)   (exp(−iθ ) − α) 

As every linear fractional transform maps circles to circles, if C is the

unit circle, then Tα (C) ⊆ C , which implies that Tα (C) = C . Since Tα
has a unique singularity α1 , which is not in Cl B(0, 1), Tα is an analytic
function on B(0, 1), and hence, Tα (B(0, 1)) is a connected subset of {z ∈
C : |z|  1}. Therefore, either Tα (B(0, 1)) = B(0, 1) or Tα (B(0, 1)) =
{z ∈ C : |z| > 1}.
Since Tα (α) = 0 (by definition) and α ∈ B(0, 1), we get Tα (B(0, 1)) =
B(0, 1). As every linear fractional transform is a one-to-one map, Tα :
B(0, 1) → B(0, 1) is a bijection, satisfying Tα (α) = 0, and by Corollary
4.5.5, Tα−1 is also an analytic map on B(0, 1) satisfying Tα−1 (0) = α .
If F = f ◦ Tα−1 , then F : B(0, 1) → B(0, 1) is an analytic bijection and
F(0) = (f ◦ Tα−1 )(0) = f (α) = 0. Therefore, by the previous corollary, we
have F(w) = cw, ∀w ∈ C, for some c ∈ C with |c| = 1. Thus, for an arbitrary
z ∈ B(0, 1), Tα (z) ∈ B(0, 1) so that f (z) = f (Tα−1 (Tα (z))) = F(Tα (z)) =
z−α
cTα (z) = c . 
1 − αz

THEOREM 4.5.12 Let f be an analytic function on a simply connected region

. If f is nowhere zero, then log( f ) can be suitably defined on so that it is
analytic on .

Proof: Using Theorem 4.2.8, we get that every closed curve γ in is

homologous to zero in . Since f is analytic and is nowhere zero, using

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 219 — #55

i i

Complex Integration 219

Theorem 4.5.5, we have

1 f  (ζ )
dζ = 0,
i2π f (ζ )
γ

for every closed curve γ is . Then by Result 4.1.30, there exists an analytic
f
function g on , such that g = . If log( f ) is defined, then its derivative
f
f
must be . Hence, we expect that log( f ) would be defined by g + c for some
f
suitable constant c. We find c from
log( f ) = g + c ⇔ exp c = f exp(−g).
To conclude this proof, first we show that f exp(−g) is a constant. This can
be obtained since
d
( f exp(−g)) = f  exp(−g) − f exp(−g)g
dz
f
= f  exp(−g) − f exp(−g)
f
= 0
and by using Theorem 2.2.19. Hence, let f exp(−g) = exp(c) for some con-
stant c, and this is possible since f exp(−g) is nowhere zero. Then, for a fixed
z0 ∈ ,
f (z0 ) exp(−g(z0 )) = exp(c) ⇒ c = log( f (z0 )) − g(z0 ),
where log( f (z0 )) is any one of the infinite number of its values. Thus, if log( f )
is defined by g + log( f (z0 )) − g(z0 ), then log( f ) is differentiable. 

COROLLARY 4.5.13 If f and are as in the previous theorem, then for every
n ∈ N, there exists ananalytic function g on such that gn = f . In other
words, we can define n f as an analytic function on .

Proof: Using the previous theorem,  log( f ) as an analytic func-

 we can define
tion on . If we define g(z) = exp 1n log( f (z) , ∀z ∈ , then clearly g is an
analytic function on and
  n
1
g (z) = exp
n
log( f (z) ) = exp (log( f (z)) = f (z), ∀z ∈ ,
n

and hence, g is the required analytic function such that g = n f . 

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 220 — #56

i i

220 Local Correspondence Theorem and Its Consequences

THEOREM 4.5.14 (Hurwitz theorem)

Let ( fn ) be a sequence of analytic functions such that fn is nowhere zero for
all n ∈ N. If fn → f uniformly on every compact subset of as n → ∞,
then either f is identically zero or f is nowhere zero.

Proof: Assume that f is not identically zero. Let z ∈ be arbitrary. Then,

we claim that there exists δ > 0 such that f  0 on Cl B(z, δ)\{z}.
| f (z)|
If f (z)  0, then using the continuity of f at z, for  = , there exists
2
r > 0 such that
| f (z)|
| f (w) − f (z)| < whenever |w − z| < r.
2
If 0 < δ < r and w ∈ Cl B(z, δ), then
| f (z)| | f (z)|
| f (z)| − | f (w)| ≤ | f (w) − f (z)| < ⇒ | f (w)| > > 0.
2 2
Thus, our claim holds, in this case.
If f (z) = 0, then by Theorem 4.5.1, there exist m ∈ N and an analytic function
g on such that

f (w) = (w − z)m g(w), ∀w ∈ and g(z)  0.

Then by the previous argument, there exists δ > 0 such that g  0 on

Cl B(z, δ). This implies that f  0 on 0 < |w − z| ≤ δ . Thus, our claim
holds.
Since fn → f uniformly on every compact subset of as n → ∞, then
by Theorem 4.3.4, we get fn → f  uniformly on every compact subset of
as n → ∞. By a similar technique employed in Theorem 1.3.9, we obtain
fn f
→ uniformly on |ζ − z| = δ , as n → ∞.
fn f
Indeed, since |f | is real, continuous, and nowhere zero on |ζ − z| = δ ,
there exists μ > 0 such that μ = inf{|f (ζ )| : |ζ − z| = δ}. Let M1 and M2 be
positive real numbers such that

M1 > sup{|f (ζ )| : |ζ − z| = δ} and M2 > sup{|f  (ζ )| : |ζ − z| = δ}.

Then, for a given  > 0, there exist N1 and N2 ∈ N such that

 
|μ| |μ|2
|fn (ζ ) − f (ζ )| < min , , ∀n ≥ N1
2 4M2

i i

i i
 i i

“book” — 2014/6/4 — 21:03 — page 221 — #57

i i

Complex Integration 221

and
|μ|2
|fn (ζ ) − f  (ζ )| < , ∀n ≥ N2 .
4M1
|μ|
Therefore, for n ≥ N1 , we have |f | − |fn | ≤ |fn − f | < 2 , and hence,

|μ| |μ| |μ|

= |μ| − ≤ |f | − < |fn |.
2 2 2
For every ζ ∈ C with |ζ − z| = δ and for every n ≥ max{N1 , N2 },
  
 fn (ζ ) f  (ζ ) 
 fn (ζ ) − f (ζ ) 
  
 f (ζ )f (ζ )−f (ζ )f  (ζ ) 
=  n fn (ζ )f n(ζ ) 
    !
≤ |fn (ζ )f (ζ )| fn (ζ )f (ζ ) − f (ζ )f  (ζ ) + f (ζ )f  (ζ ) − fn (ζ )f  (ζ )
1 
  !
≤ |μ|2 2 M1 fn (ζ ) − f  (ζ ) + M2 |f (ζ ) − fn (ζ )|
 
≤ |μ|2 2 M1 |μ| |μ|2
2
4M1 + M 2 4M2
= 2 + 2 = .

Therefore, by Theorem 4.5.2, we get that the number of zeroes of f in B(z, δ)

is equal to

1 f  (ζ ) 1 fn (ζ )
dζ = lim dζ
i2π f (ζ ) n→∞ i2π fn (ζ )
|ζ −z|=δ |ζ −z|=δ
 
1 fn (ζ )
= lim dζ
i2π n→∞ fn (ζ )
|ζ −z|=δ
= 0,

since each fn has no zeroes. Therefore, f  0 on B(z, δ). In particular, f (z)  0.

Using that z ∈ is arbitrary, we get that f is nowhere zero on . 

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 21:09 — page 223 — #1

i i

5
Series Developments
and Infinite Products

5.1 TAYLOR SERIES AND LAURENT SERIES

Definition 5.1.1 Let f be an analytic function on \{a}. Then, the point a is
said to be a removable singularity of f if lim f (z) exists in C.
z→a

LEMMA 5.1.2 Let f be an analytic function on \{a} for some a ∈ . Then,

the point a is a removable singularity of f if and only if there exists an analytic
function g on  such that f (z) = g(z), ∀z ∈ \{a}.

Proof: Assume that there exists an analytic function g on  such that

f (z) = g(z), ∀z ∈ \{a}. Then, g is continuous at a. Therefore, lim f (z) =
z→a
lim g(z) = g(a), which exists in C.
z→a
Conversely, assume that a is a removable singularity. We choose r > 0
such that B(a, r) is contained in . Then, for each z ∈ B(a, r)\{a}, we define
f (ζ ) − f (z)
gz (ζ ) = , ∀ζ ∈ \{a, z}.
ζ −z
Then, gz is analytic on \{a, z},
f (ζ ) − f (z)
lim gz (ζ ) = lim = f  (z)
ζ →z ζ →z ζ −z
and
lim f (ζ ) − f (z)
f (ζ ) − f (z) ζ →a
lim gz (ζ ) = lim = .
ζ →a ζ →a ζ −z a−z

223

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 224 — #2

i i

224 Taylor Series and Laurent Series

Therefore, gz is bounded on some neighbourhoods of a and z. Using Cauchy’s

theorem for open disk with exceptional points, if 0 < s < r and 0 < |z − a| <
s, then we get
  
1 f (ζ ) f (z) dζ
gz (ζ ) dζ = 0 ⇒ dζ = = f (z).
i2π ζ −z i2π ζ −z
|ζ −a|=s |ζ −a|=s |ζ −a|=s

1  f (ζ )
Since f is continuous on |z − a| = s, if f1 (z) = dζ , then f1
i2π |ζ −a|=s ζ − z
is analytic on B(a, r) (including at a) and f1 (z) = f (z), ∀z ∈ B(a, r). Hence, if
g :  → C is defined by

f (z) if z  a
g(z) = ,
f1 (a) if z = a

then g is analytic on  and g(z) = f (z), ∀z ∈ \{a}. 

THEOREM 5.1.3 (Taylor’s theorem)

Let f be an analytic function on  and z0 ∈ . Then, for every n ∈ N, there
exists an analytic function φn on  such that


n−1 (k)
f (z0 )
f (z) = (z − z0 )k + φn (z)(z − z0 )n , ∀z ∈ .
k!
k=1

Furthermore,

1 f (ζ )
φn (z) = , dζ , ∀z inside C,
i2π (ζ − z)(ζ − z0 )n
C

where C is a circle |z − z0 | = r contained in .

Proof: If we define
f (z) − f (z0 )
F(z) = , ∀z ∈ \{z0 },
z − z0

then F is analytic on \{z0 } and z0 is a removable singularity for F , as

f (z) − f (z0 )
lim F(z) = lim = f  (z0 ).
z→z0 z→z0 z − z0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 225 — #3

i i

Series Developments and Infinite Products 225

Then, by Lemma 5.1.2, there exists an analytic function φ1 on  such that

φ1 (z) = F(z), ∀z ∈ \{z0 }. Hence, for every z ∈ \{z0 }, clearly we have
f (z)−f (z0 ) = φ1 (z)(z−z0 ) and f (z0 )−f (z0 ) = 0 = φ1 (z0 )(z0 −z0 ). Therefore,

f (z) = f (z0 ) + φ1 (z)(z − z0 ), ∀z ∈ .

Applying the same technique, we can write φ1 (z) = φ1 (z0 ) + φ2 (z)(z − z0 )

for some analytic function φ2 on . Proceeding like this, at the nth stage, we
get φn−1 (z) = φn−1 (z0 ) + φn (z)(z − z0 ) for some analytic function φn on .
Therefore, for every z ∈ ,

f (z) = f (z0 ) + φ1 (z)(z − z0 )

= f (z0 ) + (φ1 (z0 ) + φ2 (z)(z − z0 )) (z − z0 )
= f (z0 ) + φ1 (z0 )(z − z0 ) + φ2 (z)(z − z0 )2
= f (z0 ) + φ1 (z0 )(z − z0 ) + φ2 (z0 )(z − z0 )2 + φ3 (z)(z − z0 )3
..
.

n−1
= f (z0 ) + φk (z0 )(z − z0 )k + φn (z)(z − z0 )n .
k=1

Differentiating n times, we get


n
dj (n−j)
f (n)
(z) = nCj ((z − z0 )n )φn (z), ∀z ∈ .
dzj
j=0

dj
For each j = 0, 1, 2, . . . , n − 1, − z0 )n ) has the factor (z − z0 ), and
dz j
((z
dn f (n) (z0 )
dzn ((z − z0 ) ) = n!, we get f (z0 ) = n!φn (z0 ). Hence, φn (z0 ) =
n (n) ,
n!
∀n ∈ N. Therefore, we have


n−1 (k)
f (z0 )
f (z) = f (z0 ) + (z − z0 )k + φn (z)(z − z0 )n , ∀z ∈ .
k!
k=1

Next, we choose r > 0 such that ClB(z0 , r) ⊆ . If C is the circle with centre
z0 and radius r > 0, then by Cauchy’s integral formula, we have

1 φn (ζ )
φn (z) = dζ , ∀z ∈ B(z0 , r).
i2π ζ −z
C

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 226 — #4

i i

226 Taylor Series and Laurent Series

Since  
1 
n−1 (k)
f (z0 )
φn (z) = f (z) − (z − z0 ) k
,
(z − z0 )n k!
k=0
we have
   n−1
1 f (ζ ) f (k) (z0 )
φn (z) = − dζ . (5.1)
i2π (ζ − z0 )n (ζ − z) k !(ζ − z0 )n−k (ζ − z)
C k=0

 dζ
Now, we claim that = 0, ∀j ∈ N and ∀z ∈ B(z0 , r). For
C (ζ − z0 ) (ζ − z)
j
each j ∈ R and z ∈ B(z0 , r), define

dζ
Fj,z (w) = , ∀w ∈ B(z0 , r).
(ζ − w)j (ζ − z)
C

Applying partial fraction technique, we get

1 1 1 1
= − ,
(ζ − w)(ζ − z) z−w ζ −z ζ −w

and hence, using Example 4.2.6, for every w ∈ B(z0 , r), we have
 
dζ 1 1 1
F1,z (w) = = − dζ
(ζ − w)(ζ − z) z−w ζ −z ζ −w
C C
i2π
= (WN(C, z) − WN(C, w)) = 1 − 1 = 0.
z−w
Using Theorem 4.3.2, we get
(1) (2)
Fj−1,z (w) Fj−2,z (w)
Fj,z (w) = =
j j(j − 1)
(3) (j−1)
Fj−3,z (w) F1,z (w)
= = ··· = = 0,
j(j − 1)(j − 2) j!
for all j > 1. Thus, Fj,z (w) = 0, ∀j > 0 and ∀w ∈ B(z0 , r). In particular,

dζ
= Fj,z (z0 ) = 0, ∀j = 1, 2, 3, . . . , n − 1.
(ζ − z0 )j (ζ − z)
C

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 227 — #5

i i

Series Developments and Infinite Products 227

Using this observation in equation (5.1), we get


1 f (ζ )
φn (z) = dζ , ∀z ∈ B(z0 , r).
i2π (ζ − z0 )n (ζ − z)
C

Hence, the theorem follows. 

COROLLARY 5.1.4 Let P be a non-constant polynomial and z0 ∈ C. If m ∈ N

such that P(k) (z0 ) = 0, ∀k = 0, 1, 2 . . . , m − 1 and P(m) (z0 )  0, then m is the
order of the zero z0 for P.

Proof: By Taylor’s theorem, we have P(z) = (z − z0 )m φ(z) for some analytic

P(z)
function φ on C such that φ(z0 )  0. Then, φ(z) = (z−z 0)
m , ∀z ∈ C, which is
a rational function and has no pole in C. Hence, using Lemma 3.1.16, we get
φ , which is a polynomial. Therefore, m is the order of the zero of P at z0 . 

THEOREM 5.1.5 (Infinite Taylor series)

Let f be analytic in a region  and z0 ∈ . Then, we have
∞ (k)
 f (z0 )
f (z) = (z − z0 )k , ∀z ∈ B(z0 , r),
k!
k=0

where r is a positive real number such that ClB(z0 , r) ⊆ .

Proof: For every n ∈ N, from Taylor theorem, we get


n−1 (k)
f (z0 )
f (z) = (z − z0 )k + φn (z)(z − z0 )n , ∀z ∈ ,
k!
k=0

where 
1 f (ζ )
φn (z) = dζ , ∀z ∈ B(z0 , r),
i2π (ζ − z)(ζ − z0 )n
C
where C is the circle |ζ − z0 | = r. If M = sup | f (ζ )|, then for z ∈ B(z0 , r),
ζ ∈C


n−1 (k)
f (z0 )
f (z) − (z − z0 )k = |z − z0 |n |φn (z)|
k!
k=0

|z − z0 |n | f (ζ )|
≤ |dζ |
2π |ζ − z||ζ − z0 |n
C

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 228 — #6

i i

228 Taylor Series and Laurent Series


|z − z0 |n M
≤ |dζ |
2π rn (r − |z − z0 |)
C
n
Mr |z − z0 |
≤ → 0 as n → ∞
r −|z − z0 | r
(as |dζ | = 2πr).
C

n−1 f (k) (z ∞ f (k) (z0 )

0)
Thus, f (z) = lim (z − z0 )k = (z − z0 )k , ∀z ∈ B(z0 , r).
n→∞ k=0 k! k=0 k!


Remark 5.1.6: At this juncture, it should be noted that finite Taylor series
n−1 f (k) (z )
0
expansion f (z) = (z − z0 )k + φn (z)(z − z0 )n of an analytic function
k=0 k !
f on  is valid on the entire region , but the infinite Taylor series expan-
∞ f (k) (z )
0
sion (z − z0 )k is valid only in a largest possible neighbourhood
k=0 k!
B(z0 , r) ⊆ .

∞ ∞
Definition 5.1.7 Let the power series an (z − a)n and bn (z − a)n con-
n=0 n=0
verge on B(a, R). Then, the Cauchy product of the two power series is defined
by
∞
 n 
 
ak bn−k (z − a)n , ∀z ∈ B(a, R).
n=0 k=0

∞ ∞
THEOREM 5.1.8 If f (z) = an (z − a)n and g(z) = bn (z − a)n ,∀z ∈
n=0 n=0
∞ n
B(a, R), then ( f · g)(z) = ak bn−k (z − a)n , ∀z ∈ B(a, R).
n=0 k=0

Proof: We know that f and g are analytic on B(a, R), and hence, f · g is also
analytic on B(a, R). By Taylor’s theorem, we have
∞
 ( f · g)(n) (a)
( f · g)(z) = (z − a)n , ∀z ∈ B(a, R).
n!
n=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 229 — #7

i i

Series Developments and Infinite Products 229

Now using Leibniz rule (Result 2.1.13) and Corollary 2.3.4, we have

1 
n
( f · g)(n) (a)
= nCk f (k) (a) · g(n−k) (a)
n! n!
k=0

1 
n
= nCk f (k) (a) · g(n−k) (a)
n!
k=0

1 
n
n!
= (ak k !) · (bn−k (n − k)!)
n! k !(n − k)!
k=0

n
= ak bn−k .
k=0

Hence, for every z ∈ B(a, R), using Corollary 2.3.5, we get ( f · g)(z) =
∞ ( f · g)(n) (a)
(z − a)n . 
n=0 n!

∞ ∞
COROLLARY 5.1.9 If an (z − a)n and bn (z − a)n converges on B(a, R),
n=0 n=0
then the Cauchy product of the two power series converge on B(a, R).

The proof of this corollary follows immediately from the previous

theorem.
∞ ∞
Definition 5.1.10 Let an (z − a)n and 0  bn (z − a)n converge ∀z ∈
n=0 n=0
B(a, R). Then, the quotient
∞
∞ an (z − a)n
 n=0
dn (z − a)n = ∞
n=0 bn (z − a)n
n=0

f
of the power series is defined as the Taylor series of , where
g
∞
 ∞

f (z) = an (z − a)n and 0  g(z) = bn (z − a)n , ∀z ∈ B(a, R).
n=0 n=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 230 — #8

i i

230 Taylor Series and Laurent Series

Finding the coefficients of the Taylor series of gf in terms of an and bn are not
easy as that of f · g. However, the first few coefficients of the Taylor series of
f
g can be calculated successively.
In the following sequel, we use the notation ANsr (a) to denote the annulus
{z ∈ C : r < |z − a| < s}.

∞ ∞ ∞
Definition 5.1.11 We say that an is convergent if an and a−n
n=−∞ n=0 n=1
∞ ∞ ∞
are convergent; in this case, we write an = a−n + an .
n=−∞ n=1 n=0

∞
THEOREM 5.1.12 For a given an (z − z0 )n , if r = lim sup |a−n |1/n , s =
n=−∞ n→∞
lim sup |an |1/n in [−∞, +∞], and r < s, then
n→∞

∞
1. an zn converges on r < |z − z0 | < s,
n=−∞

∞
2. an zn converges uniformly on r1 ≤ |z − z0 | ≤ s1 , if r < r1 <
n=−∞
s1 < s,
∞
3. an zn diverges on |z − z0 | < r or |z − z0 | > s.
n=−∞

Proof of this theorem follows immediately from Abel’s theorem on conver-

gence of power series (Theorem 2.3.2).

THEOREM 5.1.13 (Laurent’s Theorem)

Let f be analytic on an annulus ANRR21 (z0 ) = {z ∈ C : R1 < |z − z0 | < R2 }.
Then, f can be written as
∞

f (z) = ak (z − z0 )k , ∀z ∈ ANRR21 (z0 ).
k=−∞

Proof: Choose positive real numbers r1 and r2 such that R1 < r1 < r2 < R2 .
If Cj is the circle with centre z0 and radius rj , for j = 1, 2, then we claim that
C2 ∪ (−C1 ) ∼ 0 in ANRR21 (z0 ). Let w  ANRR21 (z0 ).

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 231 — #9

i i

Series Developments and Infinite Products 231

Case 1: |w − z0 | > R2 .
Then, |w − z0 | > r1 and |w − z0 | > r2 , which implies WN (C2 ∪
(−C1 ), w) = WN(C2 , w) − WN(C1 , w) = 0.
Case 2: |w − z0 | < R1 . In this case, |w − z0 | < r1 , |w − z0 | < r2 , and hence,
WN (C2 ∪ (−C1 ), w) = WN (C2 , w) − WN (C1 , w) = 1 − 1 = 0.

Hence, our claim holds. Then, by the general version of Cauchy’s integral
formula, we have

1 f (ζ )
WN (C2 ∪ (−C1 ), z)f (z) = dζ ∀z  C2 ∪ (−C1 ).
i2π ζ −z
C2 ∪(−C1 )

In particular, as WN (C2 ∪ (−C1 ), z) = 1 whenever r1 < |z − z0 | < r2 , we

have
  
1 f (ζ ) 1 f (ζ ) 1 f (ζ )
f (z) = dζ = dζ − dζ .
i2π ζ −z i2π ζ −z i2π ζ −z
C2 ∪(−C1 ) C2 C1

|ζ − z0 | r1
Since = < 1, ∀ζ ∈ C1 , we get
|z − z0 | |z − z0 |
1 1
=
ζ −z (ζ − z0 ) −(z − z0 ) 
∞
 k
−1 1 −1 ζ − z0
= = .
z − z0 1 − ζ −z0
z−z0
z − z0 z − z0
k=0

We note that the power series on the right-hand side converges uniformly on
|z − z0 |
C1 . Similarly, using < 1, ∀ζ ∈ C2 , we get
|ζ − z0 |
1 1
=
ζ −z (ζ − z0 ) −(z − z0 ) 
∞
 k
1 1 1 z − z0
= = ,
ζ − z0 1 − ζz−z
−z
0 ζ − z0 ζ − z0
0 k=0

and the series uniformly converges on C2 , ∀z ∈ ANrr21 (z0 ).

Therefore, we get
∞
 
(z − z0 )k f (ζ )
f (z) = dζ
i2π (ζ − z0 )k+1
k=0 C2

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 232 — #10

i i

232 Taylor Series and Laurent Series

∞
 
(z − z0 )−k−1 f (ζ )
+ dζ
i2π (ζ − z0 )−k
k=0 C1
∞
 
(z − z0 )k f (ζ )
= dζ
i2π (ζ − z0 )k+1
k=0 C2
∞
 
(z − z0 )−j f (ζ )
+ dζ
i2π (ζ − z0 )−j+1
j=1 C1
∞

= ak (z − z0 )k ,
k=−∞

⎧ 1  f (ζ )
⎪
⎪ dζ if k ≥ 0
⎨ i2π (ζ − z0 )k+1
C2
where ak = 1  , ∀k ∈ Z. 
⎪
⎪
f (ζ )
if k < 0
⎩ dζ
i2π C1 (ζ − z0 )k+1

RESULT 5.1.14 (Uniqueness)

∞
If f (z) = ck (z − z0 )k , ∀z ∈ ANsr (z0 ) and if the Laurent series of f is
k=−∞
∞
f (z) = ak (z − z0 )k , ∀z ∈ ANsr (z0 ), then ck = ak , ∀k ∈ Z.
k=−∞

Proof: Let n be a non-negative integer. Then, by definition, we have


1 f (ζ )
an = dζ
i2π (ζ − z0 )n+1
C2
∞
 ck (ζ − z0 )k
1 k=−∞
= dζ
i2π (ζ − z0 )n+1
C2
∞
 
1
= ck (ζ − z0 )k−n−1 dζ
i2π
k=−∞ C2
(as the series converges uniformly on C2 )

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 233 — #11

i i

Series Developments and Infinite Products 233

∞
 2π
1
= ck (r2 exp(it))k−n−1 ir2 exp(it) dt
i2π
k=−∞ 0
(as C2 is given by ζ (t) = z0 + r2 exp(it), t ∈ [0, 2π])
∞ 2π
k−n 1
= ck ir2 exp(i(k − n)t) dt
i2π
k=−∞ 0
= cn ,


1 2π 1 k=n
since exp(i(k − n)t) dt = . By a similar argument, we get
2π 0 0 kn
an = cn for every negative integer also. Hence, the result follows. 

Algorithm 5.1.15 (To find Laurent series of rational functions)

Algorithm to find Laurent series
P(z)
Step 1. If the given rational function is , where P(z), Q(z) are polyno-
Q(z)
1
mials, then find the partial fraction expansion of if Q(z) has
Q(z)
more than one root. In this case, a typical summand in the partial
A
fraction expansion is of the form for some A, a ∈ C.
(z ± a)m
1
Step 2(a). To expand , in |z − z0 | < R for some R > 0 and for some
(z ± a)m
a ∈ C with R < |b|, where b = −(z0 ± a).
Then write
1 1
=
(z ± a)m ((z − z0 ) − b)m
1 1
=  
|b|m 1 − (z−z0 ) m
b
∞
 k−m
1 1 (z − z0 )
= k(k − 1) · · · (k − m + 1) .
|b|m m! b
k=m

1
Step 2(b). To expand , in |z − z0 | > R for some R > 0 and for some
(z ± a)
a ∈ C with |b| < R, where b = −(z0 ± a).

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 234 — #12

i i

234 Taylor Series and Laurent Series

Then write
1 1
=
(z ± a)m ((z − z0 ) − b)m
1 1
=  
(z − z0 )m 1 − b m
(z−z0 )
∞
 k−m
1 1 b
= k(k − 1) · · · (k−m+1) .
(z − z0 )m m! (z − z0 )
k=m

Step 3. Finally, replace all the summands by the corresponding series and
simplify.

From the following examples, one can understand the algorithm still better.

Example 5.1.16 Find the series development of the following functions in

the specified regions:
1
1. in (i) 0 < |z| < 1; (ii) 1 < |z| < 4; (iii) |z| > 4.
z2 − 5z + 4
z+1
2. in (i) 0 < |z| < 2; (ii) 2 < |z| < 3; (iii) |z| > 3.
−z−6
z2
z
3. 2 in (i) 0 < |z−2| < 1; (ii) 1 < |z−2| < 3; (iii) |z−2| > 3.
z − 8z + 15
z+5
4. in 1 < |z + 1| < 3.
z2 − 6z + 8

Solution:
1. Applying partial fraction technique, we write
1 1/3 1/3
= − .
z2 − 5z + 4 z−4 z−1
(a) If 0 < |z| < 1, then we have1

1   z k
∞
1/3 1
=−  z  =− (5.2)
z−4 12 1 − 12 4
k=0
4
1 Here note that |z| < 1 < 4 so we have taken −4 and −1 as the common factors in equations
(5.2) and (5.3), respectively.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 235 — #13

i i

Series Developments and Infinite Products 235

and
∞
−1/3 1 1 k
= = z . (5.3)
z−1 3 (1 − z) 3
k=0

Thus, we get

1   z k 1  k
∞ ∞
1
= − + z
z − 5z + 4
2 12 4 3
k=0 k=0
∞
1 1
= 1− zk , 0 < |z| < 1.
3 4k+1
k=0

(b) If 1 < |z| < 4, then we have2

1   z k
∞
1/3 1
=−  z  =− (5.4)
z−4 12 1 − 12 4
k=0
4
and
∞ ∞
−1/3 1 1  1 k
1 1 k+1
=− =− =− .
z−1 1 3z z 3 z
3z 1 − k=0 k=0
z
(5.5)
Thus, we get

1   z k 1  1 k+1
∞ ∞
1
= − −
z − 5z + 4
2 12 4 3 z
k=0 k=0
∞ ∞
1  1 k  −k
= − z + z , 1 < |z| < 4.
3 4k+1
k=0 k=1

(c) If |z| > 4, then we have

∞ ∞
1/3 1 1  4 k
1  4k
= = =
z−4 4 3z z 3 zk+1
3z 1 − k=0 k=0
z
2 Here note that 1 < |z| < 4 so we have taken −4 and z as the common factors in equations
(5.4) and (5.5), respectively.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 236 — #14

i i

236 Taylor Series and Laurent Series

and
∞ ∞
−1/3 1 1  1 k
1 1 k+1
=− =− =− .
z−1 1 3z z 3 z
3z 1 − k=0 k=0
z
Thus, we get
∞ ∞
1 1  4k 1 1 k+1
= −
z − 5z + 4
2 3 z k+1 3 z
k=0 k=0
∞
1 k
= − (4 − 1)z−(k+1) , 0 < |z| < 1.
3
k=1

2. Applying partial fraction technique, we write

1 1/5 1/5
= − .
z2 − z − 6 z−3 z+2
(a) If 0 < |z| < 2, then we have
z+1 z+1
= −  z
5 (z − 3) 15 1 −
3
z + 1   z k
∞
= −
15 3
∞ k=0
1   z k+1   z k
∞
= − 3 +
15 3 3
 k=0 k=0
1   z k+1
∞ ∞  
z k+1
= − 3 +1+
15 3 3
k=0 k=0
4   z k+1
∞
1
= − +
15 15 3
k=0

and
−(z + 1) −(z + 1)
=  z
5(z + 2) 10 1 +
2
∞
z+1 z k
= −
10 −2
k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 237 — #15

i i

Series Developments and Infinite Products 237

∞ ∞
1  z k+1  z k
= − (−2) +
10 −2 −2
k=0 k=0
∞ ∞
1  z k+1  z k+1
= − (−2) +1+
10 −2 −2
k=0 k=0
∞
−1 1  z k+1
= + .
10 10 −2
k=0

Thus, we get

4   z k+1 −1
∞ ∞
z+1 1 1  z k+1
= − + + +
z2 − z − 6 15 15 3 10 10 −2
k=0 k=0
∞  
−1  4 −k 1
= + 3 + (−2)k zk , 0 < |z| < 2.
6 15 10
k=1

(b) If 2 < |z| < 3, then using (i), we have

4   z k+1
∞
z+1 z+1 1
=−  z  = − +
5 (z − 3) 15 1 − 15 15 3
k=0
3
and
z+1 z+1
− = −
5(z + 2) 2
5z 1 +
z
∞
z+1 2 k
= − −
5z z
k=0
∞ ∞
1  2 k 1 2 k+1
= − − − −
5 z 2 z
k=0 k=0
 ∞ ∞
1  2 k+1 1  2 k+1
= − 1+ − − −
5 z 2 z
k=0 k=0
∞
 k+1
1 1 2
= − + − .
5 10 z
k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 238 — #16

i i

238 Taylor Series and Laurent Series

Thus, we get
4   z k+1 1
∞ ∞
z+1 1 1  2 k+1
= − + − + −
z −z−6
2 15 15 3 5 10 z
k=0 k=0
∞

4 4
= − + 3−k zk
15 15
k=1
∞

1
+ (−2)k z−k , 2 < |z| < 3.
10
k=1

(c) If |z| > 3, then we have

z+1 z+1
=
5(z − 3) 3
5z 1 −
z
∞
z+1 3 k
=
5z z
k=0
∞ ∞
1  3 k  1 3 k+1
= +
5 z 3 z
k=0 k=0
 ∞ ∞
1  3 k 1 3 k
= + 1+ +
5 z 3 z
k=1 k=1
∞
1 4  3 k
= + ,
5 15 z
k=1

and from case (ii), we have

∞
z+1 z+1 1 1  2 k+1
− =− =− + − .
5(z + 2) 2 5 10 z
5z 1 + k=0
z
Thus, we get
∞ ∞
z+1 4  k −k 1 
= 3 z + (−2)k z−k
z −z−6
2 15 10
k=1 k=1
∞
 4 k−1 1
= 3 + (−2)k z−k, |z| > 3.
5 10
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 239 — #17

i i

Series Developments and Infinite Products 239

3. Applying partial fraction technique, we write

1 1/2 1/2
= − .
z2 − 8z + 15 z−5 z−3
(a) If 0 < |z − 2| < 1, then we have
z z
=
2(z − 5) 2(z − 2 − 3)
z−2+2 1
=  
−6 1 − z−2 3
z−2 1 1
= − +  
6 3 1− z−2
3
∞
z−2 1  z−2 k
= − +
6 3 3
k=0
∞ ∞
1 z−2 k+1
1 z−2 k
= − −
2 3 3 3
k=0 k=0
∞ ∞
1 z−2 k
1 1  z−2 k
= − − −
2 3 3 3 3
k=1 k=1
∞
1 5  −k
= − − 3 (z − 2)k
3 6
k=1

and
z z
− = −
2(z − 3) 2(z − 2 − 1)
z−2+2 1
=
2 (1 − (z − 2))
z−2 1
= +1
2 (1 − (z − 2))
 ∞
z−2
= +1 (z − 2)k
2
k=0
∞ ∞
1 
= (z − 2)k+1 + (z − 2)k
2
k=0 k=0
∞ ∞
1 
= (z − 2)k + 1 + (z − 2)k
2
k=1 k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 240 — #18

i i

240 Taylor Series and Laurent Series

∞
3
= 1+ (z − 2)k .
2
k=1

Therefore,
∞
1 1 5  −k
= − − 3 (z − 2)k + 1
z2 − 8z + 15 3 6
k=1
∞
3
+ (z − 2)k
2
k=1
∞
2 3 5 −k
= + − 3 (z − 2)k .
3 2 6
k=1

(b) If 1 < |z − 2| < 3, then we have

z z
=
2(z − 5) 2(z − 2 − 3)
z−2+2 1
=  
−6 1 − z−2 3
z−2 1 1
= − +  
6 3 1− z−2
3
∞
z−2 1  z−2 k
= − +
6 3 3
k=0
∞ ∞
1 z−2 k+1
1 z−2 k
= − −
2 3 3 3
k=0 k=0
∞ k ∞ k
1 z−2 1 1 z−2
= − − −
2 3 3 3 3
k=1 k=1
∞
1 5
= − − 3−k (z − 2)k
3 6
k=1

and
z z
− = −
2(z − 3) 2(z − 2 − 1)
z−2+2 1
= −  
2(z − 2) 1 − 1
z−2

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 241 — #19

i i

Series Developments and Infinite Products 241

1 1 1
= − −  
2 z−2 1− 1
z−2
∞ ∞
−1  
= (z − 2)−k + (z − 2)−k
2
k=0 k=1
∞
1 1
= − + (z − 2)−k .
2 2
k=1

Therefore,

∞ ∞
1 1 5  −k 1 1
= − − 3 (z − 2)k − + (z − 2)−k
z − 8z + 15
2 3 6 2 2
k=1 k=1
∞ ∞
5 5  −k 1
= − − 3 (z − 2)k + (z − 2)−k .
6 6 2
k=1 k=1

(c) If |z − 2| > 3, then we have

z z
=
2(z − 5) 2(z − 2 − 3)
z−2+2 1
=  
2(z − 2) 1 − 3
z−2
1 1 1
= +  
2 z−2 1− 3
z−2
∞
 k
1 1 3
= +
2 z−2 z−2
k=0
∞ ∞
1 3 k
1  3 k
= +
2 z−2 3(z − 2) z−2
k=0 k=0
∞
 k ∞
 k
1 1 3 1 3
= + +
2 2 z−2 3 z−2
k=1 k=1
∞
1 5
= + 3k (z − 2)−k
2 6
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 242 — #20

i i

242 Taylor Series and Laurent Series

and
z z
− = −
2(z − 3) 2(z − 2 − 1)
z−2+2 1
= −  
2(z − 2) 1 − 1
z−2
1 1 1
= − −  
2 z−2 1− 1
z−2
∞
1 1
= − + (z − 2)−k .
2 2
k=1

Therefore,
∞ ∞
1 1 5 k 1 1
= + 3 (z − 2)−k − + (z − 2)−k
z − 8z + 15
2 2 6 2 2
k=1 k=1
∞
 ∞

5 1
= 3−k (z − 2)k + (z − 2)−k .
6 2
k=1 k=1

4. Applying partial fraction technique, we write

1 1/2 1/2
= − .
z2 − 6z + 8 z−4 z−2
(a) If 1 < |z + 1| < 3, then we have
z+5 z+5
=
2(z − 4) 2(z + 1 − 5)
z+5 1
= −
10 z+1
1−
5
∞
z+1+4 z+1 k
= −
10 5
k=0
 ∞ ∞
1  z + 1 k+1  z+1 k
= − 5 +4
10 5 5
k=0 k=0
∞
 k+1 ∞
 k
1 z+1 2 z+1
= − −
2 5 5 5
k=0 k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 243 — #21

i i

Series Developments and Infinite Products 243

∞ ∞
1 z+1 k
2 2 z+1 k
= − − −
2 5 5 5 5
k=1 k=1
∞
2 7  −k
= − − 5 (z + 1)k
5 10
k=1

and
z+5 z+5
− = −
2(z + 2) 2(z + 1 + 1)
z+1+4
= −
1
2(z + 1) 1 +
z+1
∞
 k
1 2 −1
= − +
2 z+1 z+1
k=0
∞ ∞
1 −1 k  −1 k+1
= − +2
2 z+1 z+1
k=0 k=0
∞
 k ∞
 k
1 1 −1 −1
= − − +2
2 2 z+1 z+1
k=1 k=1
∞

1 3
= − + (−1)k (z + 1)−k .
2 2
k=1

Thus, we get
∞
1 1 7  −k
= − − 5 (z + 1)k + 2
z2 − 5z + 4 2 10
k=1
∞
1
− (−1)k (z + 1)−k
2
k=1
∞
7  −k
= −1 − 5 (z + 1)k
10
k=1
∞
3
+ (−1)k (z + 1)−k , ∀z with 1<|z + 1|<3.
2
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 244 — #22

i i

244 Zeroes, Poles, and Singularities

Exercise 5.1.17 Find the Laurent series expansions of the given functions in
the specified regions.
1
1. f (z) = in 0 < |z| < 1.
z2 (1 − z)
1 √ √
2. f (z) = in (i) 0 < |z − i| < 2, (ii) |z − i| > 2.
1+z
1
3. f (z) = 2 in 0 < |z| < 2.
z − 5z + 6
(1 − z)3
4. f (z) = in |z − 1| > 1.
z−2
1
5. f (z) = in (i) 0 < |z − 1| < 2, (ii) 0 < |z − 3| < 2.
(z − 1)2 (z − 3)
Answers:
∞
 1
1. .
zn
n=−2
∞
 n ∞
 n
z−i 1+i
2. (i) (−1)n , (ii) (−1)n+1 .
1+i z−i
n=0 n=1
∞
 1 1
3. − n zn−1 .
2n 3
n=1
∞
 1
4. −(z − 1)3 .
(z − 1)n
n=1
∞
 n ∞
   n−1
1 z−1 1 z−3
5. (i) − , (ii) n − .
2(z − 1) 2 2 4(z − 3) 2
n=0 n=1

5.2 ZEROES, POLES, AND SINGULARITIES

Definition 5.2.1 If f is an analytic function on a region  and f (z0 ) = 0 for
some z0 ∈ , then z0 is called a zero of f . If there exists k ∈ N, such that

f (z0 ) = f  (z0 ) = f  (z0 ) = · · · = f (k−1) (z0 ) = 0 and f (k) (z0 )  0,

then k is called the order of the zero of f at z0 .

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 245 — #23

i i

Series Developments and Infinite Products 245

THEOREM 5.2.2 Let f be an analytic function, which is not identically zero

on a region . If f (z0 ) = 0 for some z0 ∈ , then order of zero of f at z0
exists.
 
Proof: Let A = z ∈  : f (z) = 0 and f (k) (z) = 0, ∀k ∈ N and B = \A.
 
That is, B = z ∈  : f (z)  0 or f (k) (z)  0 for some k ∈ N . Then clearly,
 = A ∪ B and A ∩ B = ∅. Since f (k) is analytic (see Corollary 4.3.3), f (k)
is continuous for every k = 0, 1, 2, 3, . . . . Therefore,
   −1
z ∈  : f (k) (z) = 0 = f (k) ({0})
 
is a closed set, and hence, z ∈  : f (k) (z)  0 is an open set for every k =
∞  
0, 1, 2, 3, . . . . Therefore, B = ∪ z ∈  : f (k) (z)  0 is an open set.
k=0
Next, we show that A is also an open set. If a ∈ A is arbitrary, then for every
n ∈ N, by Taylor’s theorem, we have


n−1 (k)
f (a)
f (z) = (z − a)k + φn (z)(z − a)n = φn (z)(z − a)n , ∀z ∈ ,
k!
k=0

where 
1 f (ζ )
φn (z) = dζ , ∀z inside C,
i2π (ζ − a)n (ζ − z)
|ζ −a|=r

where r > 0 is such that ClB(a, r) ⊆ . If M = sup | f (ζ )|, then M < ∞ and
ζ ∈C
for every z ∈ B(a, r),

|z − a|n | f (ζ )|
| f (z)| ≤ |dζ |
2π |ζ − a|n |ζ − z|
|ζ −a|=r

M|z − a|n |dζ |
≤
2π rn (r − |z − a|)
|ζ −a|=r
(as |ζ − a| ≤ |z − ζ | + |z − a| ⇒ r − |z − a| ≤ |z − ζ |)
M|z − a|n 2πr
≤
2π r (r − |z − a|)
n

|z − a| n Mr
≤ → 0 as n → ∞.
r r − |z − a|

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 246 — #24

i i

246 Zeroes, Poles, and Singularities

Therefore, f = 0 on B(a, r), and hence, f (k) (z) = 0, for all k = 0, 1, 2, 3, . . . .

Thus, B(a, r) ⊆ A. Given that z0 ∈  such that f (z0 ) = 0. Suppose f (k) (z0 ) =
0, ∀k ∈ N, then A  ∅. By using the connectedness of , we get

B = ∅ ⇒  = A ⇒ f = 0 on ,

which is a contradiction. Hence, there exists a least positive integer k such

that f (k) (z0 )  0. In other words, order of the zero at z0 exists for f . 

RESULT 5.2.3 If f is an analytic function on a region  with a zero of order k

at z0 ∈ , then there exists an analytic function g on  such that g is analytic
on , f (z) = (z − z0 )k g(z) on  and g(z0 )  0.

Proof: By Taylor’s theorem, there exists an analytic function g on  such

that


k−1 (j)
f (z0 )
f (z) = (z − z0 ) j + g(z)(z − z0 )k = g(z)(z − z0 )k , ∀z ∈ .
j!
j=0

Suppose g(z0 ) = 0, then using

 dν  
k
f (k)
(z) = kCν g(k−ν) (z) (z − a)k
dzν
ν=0

and 
dν   0 ν<k
ν
(z − a)k = ,
dz z=a k! ν=k

we obtain f (k) (a) = g(a)k ! = 0, which contradicts the fact that the order of
the zero of f at z0 is k . Thus, g(a)  0. 

THEOREM 5.2.4 (Principle of analytic continuation)

Let f and g be analytic functions on a region . If {z ∈  : f (z) = g(z)} has
a limit point in , then f (z) = g(z), ∀z ∈ .

Proof: Since f and g are analytic, clearly f − g is an analytic function. If z0 is

a limit point of {z ∈  : f (z) = g(z)}, then there exists a sequence (zn ) from
\{z0 } such that

f (zn ) − g(zn ) = 0, ∀n ∈ N and zn → z0 as n → ∞.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 247 — #25

i i

Series Developments and Infinite Products 247

Therefore, ( f − g)(z0 ) = lim ( f − g)(zn ) = 0. If f (w)  g(w) for some

n→∞
w ∈ , then f − g is not identically zero. Then, by the previous result, there
exists k ∈ N such that
( f − g)(z) = (z − z0 )k φ(z), ∀z ∈ B(z0 , r),
for some analytic function φ on  such that φ(z0 )  0. Since φ is continuous
|φ(z0 )|
at z0 , for > 0, we find 0 < δ < r such that
2
|φ(z0 )|
|z − z0 | < δ ⇒ |φ(z) − φ(z0 )| <
2
|φ(z0 )|
⇒ |φ(z0 )| − |φ(z)| ≤ |φ(z) − φ(z0 )| <
2
|φ(z0 )|
⇒ |φ(z)| ≥ .
2
Therefore, φ is nowhere zero on B(z0 , δ). Since zn → z0 as n → ∞, for this
δ > 0, we find an m ∈ N such that
|zm − z0 | < δ ⇒ zm ∈ B(z0 , δ)
⇒ (zm − z0 )k φ(zm ) = ( f − g)(zm ) = 0
⇒ φ(zm ) = 0.
which is a contradiction. Hence, f (z) = g(z), ∀z ∈ . 

COROLLARY 5.2.5 If f is a non-constant analytic function on a region ,

then f cannot have uncountable number of distinct zeroes.

Proof: Let Z( f ) = {z ∈  : f (z) = 0}. By principle of analytic continuation,

Z( f ) has no limit point in . Hence, for every z ∈ Z( f ), there exists rz > 0
such that B(z, rz )∩Z( f ) = {z}. From Theorem 2.4.34, there exists a countable
collection B = {Bn : n ∈ N} of open balls such that for every z ∈ Z( f ), there
exists Bnz ∈ B such that z ∈ Bnz ⊆ B(z, rz ). Therefore, it follows that
{z} ⊆ Bnz ∩ Z( f ) ⊆ B(z, rz ) ∩ Z( f ) = {z},
and hence, Bnz ∩ Z( f ) = {z}, ∀z ∈ Z( f ). If we define
φ : Z( f ) → N by φ(z) = nz , ∀z ∈ Z( f ),
where nz ∈ N is the least positive integer such that z ∈ Bnz ⊆ B(z, rz ), then
claim that φ is one-to-one. If z, w ∈ Z( f ) be such that φ(z) = φ(w), then
Bnz = Bnw ⇒ {z} = Bnz ∩ Z( f ) = Bnw ∩ Z( f ) = {w} ⇒ z = w.
Thus, φ is an injection. Hence, Z( f ) is at most countable. 

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 248 — #26

i i

248 Zeroes, Poles, and Singularities

LEMMA 5.2.6 Let f be a non-zero analytic function on a region  and γ be

a closed curve not passing through any zero of f . Then, the number of zeroes
of f enclosed by γ is finite.

Proof: We note that the regions enclosed by a closed curve are bounded sets.
Suppose, there are infinitely many zeroes of f , which are enclosed by γ , then
the set {z ∈  : f (z) = 0} has a limit point, by Bolzano–Weierstrass property
(Theorem 1.4.47). Hence, by the principle of analytic continuation (Theorem
5.2.4), it would follow that f is identically zero. Hence, there must be only
finitely many zeroes of f enclosed by γ . 

PROBLEM 5.2.7 If f is an entire function such that |f | = 1 on {z ∈ C : |z| =

1}, then f = czn , ∀z ∈ C, for some constant c ∈ C with |c| = 1 and for some
n ∈ {0, 1, 2, . . .}.

Solution: If f is a constant function, then the proof is obvious. Hence,

assume that f is not a constant. Let z1 , z2 , . . . , zn be the zeroes of f includ-
ing multiplicities on Cl B(0, 1). (Suppose f has infinite number of zeroes in
Cl B(0, 1), being a compact set, then by Theorem 1.4.47, the zeroes of f can
have limit point in Cl B(0, 1). Thus, by the principle of analytic continuation
(Theorem 5.2.4), f becomes identically zero on C, which is not possible.)
Moreover, note that |zk | < 1, ∀k ∈ {1, 2, . . . , n}, as |zk | ≤ 1 and |f (z)| = 1 on
|z| = 1.

n z−z
k
Consider the rational function R(z) = , whose finite zeroes
k=1 1 − z k z
1
are z1 , z2 , . . . , zn and its finite poles are , k = 1, 2, 3, . . . , n, with zk  0.
zk
f (z)
If F(z) = , then after removing the common zeroes z1 , z2 , . . . , zn (which
R(z)
are removable singularities for F ), F becomes an entire function with zeroes
1
at the poles of R. Since > 1, F is nowhere zero on Cl B(0, 1). As |F(z)|
zk
is continuous real-valued function on the compact set Cl B(0, 1), there exist
w1 , w2 ∈ Cl B(0, 1) such that

|F(w1 )| ≤ |F(z)| ≤ |F(w2 )|, ∀z ∈ Cl B(0, 1).

Hence, applying maximum and minimum principle for analytic functions, we

see |w1 | = |w2 | = 1, and hence, by hypothesis,

1 = |F(w1 )| ≤ |F(z)| ≤ |F(w2 )| = 1 on Cl B(0, 1).

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 249 — #27

i i

Series Developments and Infinite Products 249

In particular, maximum of |F| is attained at all points of B(0, 1); hence, again
by using maximum modulus principle, we get F = c on Cl B(0, 1) for some
c ∈ C with
 |c| = 1. Then, by principle of  analytic continuation, we get F = c
1
on C \ : k = 1, 2, . . . n with zk  0 . Therefore,
zk
n  
z − zk 1
f (z) = c , ∀z ∈ C \ : k = 1, 2, . . . n with zk  0 .
1 − zk z zk
k=1

If zk  0 for some k , then f should have a pole at zk , which is a contradiction

to the assumption on f . Hence, zk = 0, ∀k = 1, 2, 3, . . . , n. In other words,
f (z) = czn for all z ∈ C.

Definition 5.2.8 Let f be a complex-valued function on a region  and a ∈

. We say that a is a singularity of f if f is not differentiable at a. A singularity
a of f is called
1. a removable singularity if lim f (z) exists in C.
z→a
2. a pole if lim f (z) = ∞.
z→a
3. an essential singularity if lim f (z) does not exist in C∞ . That is, a is
z→a
neither a removable singularity nor a pole.
4. an isolated singularity if there exists r > 0 such that f is analytic in
B(a, r) \ {a}.

RESULT 5.2.9 Let f be an analytic function on \{a}. Then, the following

statements are equivalent:
1. a is a removable singularity for f ,
2. f is bounded on B(a, r)\{a} for some r > 0,
3. lim (z − a)f (z) = 0,
z→a
4. There exists an analytic function φ on  such that φ(z) = f (z), ∀z ∈
\{a}.

Proof:
(1) ⇒ (2) Since a is a removable singularity of f , we have lim f (z) = for
z→a
some ∈ C. Therefore, for = 1, there exists r > 0 such that
0 < |z − a| < r ⇒ | f (z) − | < 1.
Therefore, | f (z)| ≤ 1 + | |, ∀z ∈ B(a, r)\{a}. Hence, (2) follows.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 250 — #28

i i

250 Zeroes, Poles, and Singularities

(2) ⇒ (3) By assumption, there exists r > 0 and M > 0 such that

| f (z)| ≤ M, ∀z ∈ B(a, r)\{a}.

 
For a given > 0, choose δ = min , r so that δ > 0 and
M

0 < |z − a| < δ ⇒ |z − a| | f (z)| ≤ δM ≤ M= .

M
Therefore, (3) holds.
(3) ⇒ (4) Assume that lim (z − a)f (z) = 0. Choose r > 0 such that
z→a
ClB(a, r) ⊆ . Define
⎧
⎪  f (ζ )
⎨ dζ , ∀z ∈ B(a, r)
φ(z) = |ζ −a|=r ζ − a .
⎪
⎩f (z) ∀z ∈ \B(a, r)

Then, by Theorem 4.3.2 and by Cauchy’s integral formula (Theorem

4.3.1), we get φ , which is analytic on  and φ(z) = f (z), ∀z ∈ \{a}.
(4) ⇒ (1) In particular, φ is continuous at a, and hence,

φ(a) = lim φ(z) = lim f (z).

z→a z→a

Thus, proof (1) follows.

Hence, the result follows.

Definition 5.2.10 Let f be a complex-valued function on a region . f is said

to be a meromorphic function if for each z ∈ , either f is differentiable at z
or f has a pole at z.

RESULT 5.2.11 Let f be an analytic function on a region  except at a pole

a. Then, there exist δ > 0 and p ∈ N such that

f (z) = (z − a)−p ψ(z), ∀z ∈ B(a, δ)\{a},

where ψ is analytic on B(a, δ) and ψ(a)  0.

Proof: Since lim f (z) = ∞, there exists > 0 such that B(a, ) ⊆  and
z→a

0 < |z − a| < ⇒ | f (z)| > 1.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 251 — #29

i i

Series Developments and Infinite Products 251

Hence, f (z)  0, ∀z ∈ B(a, )\{a}. If

1
g(z) = , ∀z ∈ B(a, )\{a},
f (z)
then g is analytic and nowhere zero on B(a, )\{a}. Since
1
lim g(z) = lim = 0,
z→a z→a f (z)
a is a removable singularity for g. Therefore, if we define
1
g(a) = lim g(z) = = 0,
z→a f (z)
then g becomes analytic on B(a, ), and hence, there exist p ∈ N and an
analytic function ϕ on B(a, ) such that
g(z) = (z − a)p ϕ(z), ∀z ∈ B(a, ) and ϕ(a)  0.
Since ϕ is non-zero at a and g is nowhere zero on B(a, δ) \ {a}, it follows that
ϕ is nowhere zero on B(a, r). If we define
1
ψ(z) = , ∀z ∈ B(a, δ),
ϕ(z)
then ψ is a nowhere zero analytic function on B(a, δ), and
f (z) = (z − a)−p ψ(z), ∀z ∈ B(a, δ).
Hence, the result follows. 

COROLLARY 5.2.12 If f is a meromorphic function on , then f has at most

countable number of poles.

Proof: From Result 5.2.11, every pole of f is isolated. Hence, by a simi-

lar argument employed in the proof of Corollary 5.2.5, we can prove this
corollary. 

Definition 5.2.13 If a is a pole of f , then the order of the pole at a is defined

by the natural number p such that
f (z) = (z − a)−p ψ(z), ∀z ∈ B(a, δ)\{a},
for some δ > 0 and for some analytic function ψ on B(a, δ) with ψ(a)  0.
If f is analytic on  except at an isolated singularity a then the nature of
singularity at a can be described easily from the Laurent series expansion of
f in B(a, r)\{a} as follows, where r > 0 is such that ClB(a, r) ⊆ .

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 252 — #30

i i

252 Zeroes, Poles, and Singularities

RESULT 5.2.14 If f has an isolated singularity at a and

∞

f (z) = cn (z − a)n , in B(a, r)\{a},
n=−∞

for some r > 0, then a is

1. a removable singularity iff cn = 0, ∀n < 0.
2. a pole iff there exists m ∈ N such that c−m  0 and cn = 0, ∀n < −m.
Furthermore, this m is the order of the pole of f at a.
3. an essential singularity iff cn  0 for infinitely many n < 0.
∞
Proof: Let a be an isolated singularity of f (z) = cn (z − a)n in 0 <
n=−∞
|z − a| < r.
1. Assume that a is a removable singularity. Suppose there exists at least
one k < 0 such that ck  0, then
∞

lim f (z) = lim cn (z − a)n
z→a z→a
n=−∞

does not exist in C, which is a contradiction. Hence, cn = 0, ∀n < 0.

∞
Conversely, assume that f (z) = cn (z − a)n . Then, lim f (z) = f (a)
n=0 z→a
exists in C. Hence, f has a removable singularity at a.
2. Assume that a is a pole of order m. Then, using Theorem 5.2.11, we
get
f (z) = (z − a)−m ψ(z), ∀z ∈ B(a, r)\{a},
where ψ is analytic on B(a, δ) and ψ(a)  0. As ψ is analytic on B(a, δ),
we have the Taylor series expansion
∞

ψ(z) = bn (z − a)n , ∀z ∈ B(a, r).
n=0

Hence, we have
∞
 ∞

−m
f (z) = (z − a) bn (z − a) =
n
bn (z − a)n−m , ∀z ∈ B(a, r)\{a}.
n=0 n=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 253 — #31

i i

Series Developments and Infinite Products 253

Since the coefficient of (z − a)−m is b0 = ψ(a)  0, the proof of this

part follows. Conversely, assume that c−m  0 and cn = 0, ∀n < −m
for some m ∈ N. Then, let

(z − a)m f (z), if z  a
ψ(z) = ∀z ∈ B(a, r).
c−m if z = a,

Since
∞

ψ(z) = (z − a)m cn (z − a)n
n=−m
∞

= cn (z − a)n+m
n=−m
∞
= cn−m (z − a)n , |z − a| < r
n=0
∞
and the power series cn−m (z − a)n is analytic inside B(a, r), we get
n=0
that ψ is an analytic function in B(a, r) with
f (z) = (z − a)−m ψ(z), ∀z ∈ B(a, r)\{a} and ψ(a) = c−m  0.
Hence, f has a pole at a of order m.
3. a is essential singularity iff a is neither a removable singularity nor a
pole iff cn  0 for some n < 0 and for every m ∈ N, there exists n ∈ Z
such that n < −m and cn  0 iff cn  0 for infinitely many n < 0.
Hence, the result follows. 

RESULT 5.2.15 Let f be an analytic function on a region  except at an

isolated singularity a.
1. If lim |z − a|r | f (z)| = 0 for some r ∈ R
z→a
or
2. if lim |z − a|s | f (z)| = ∞ for some s ∈ R,
z→a

then there exists an integer m such that

lim |z − a| j | f (z)| = 0, ∀j > m and lim |z − a|k | f (z)| = ∞, ∀k < m.
z→a z→a

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 254 — #32

i i

254 Zeroes, Poles, and Singularities

Proof: Let (1) hold. Then, choose an integer n ≥ r. Now,

lim |z − a|n+1 | f (z)| = lim |z − a|n−r+1 lim |z − a|r | f (z)| = 0.

z→a z→a z→a

Hence, if φ(z) = (z − ∀z ∈ \{a}, then φ is an analytic function

a)n f (z),
except at a, and it has removable singularity at a and

|φ(a)| = lim |φ(z)|

z→a
= lim |z − a|n | f (z)|
z→a
= lim |z − a|n−r lim |z − a|r | f (z)|
z→a z→a
= 0.

Hence, φ(a) = 0. Since φ is not identically zero3 using Result 5.2.3, we have

φ(z) = (z − a)ν φ1 (z), ∀z ∈ B(a, r) ⊆ ,

for some ν ∈ N, for some r > 0 and for an analytic function φ1 on B(a, r)
with φ1 (a)  0. Now, let m = n − ν . Then,

lim |z − a|m | f (z)| = lim |φ1 (z)| = |φ1 (a)|,

z→a z→a

which is neither 0 nor ∞. Hence, for j > m,

lim |z − a| j | f (z)| = lim |z − a| j−m lim |z − a|m | f (z)| = 0 · |φ1 (a)| = 0,

z→a z→a z→a

and for k < m,

lim |z − a|k | f (z)| = lim |z − a|k−m lim |z − a|m | f (z)| = ∞ · |φ1 (a)| = ∞.
z→a z→a z→a

Next, assume that (2) holds. Then, choose an integer l such that l < s. Then,
lim |z − a|l | f (z)| = lim |z − a|l−s lim |z − a|s | f (z)| = ∞,
z→a z→a z→a

and hence, a is a pole for ψ , where

ψ(z) = (z − a)l f (z), ∀z ∈ \{a}.

Hence, by using Result 5.2.11, there exist a negative integer λ, δ > 0 and an
analytic function ψ1 on B(a, δ) ⊆  such that
ψ(z) = (z − a)λ ψ1 (z), ∀z ∈ B(a, δ)\{a}.
3 As f has a singularity, f should not be identically zero, and hence, φ is not identically zero.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 255 — #33

i i

Series Developments and Infinite Products 255

Therefore,
f (z) = (z − a)−l ψ(z) = (z − a)λ−l ψ1 (z), ∀z ∈ B(a, δ) \ {a}.

If m = l − λ, then
lim |z − a|m | f (z)| = lim |ψ1 (z)| = |ψ1 (a)|,
z→a z→a

which is neither 0 nor ∞. Thus, by a similar argument, we get the result.

Note that if there exist two integers m1 and m2 such that
lim |z − a| j | f (z)| = 0, ∀j > mi
z→a

and
lim |z − a|k | f (z)| = ∞, ∀k < mi , for i = 1, 2,
z→a

then obviously, m1 = m2 . 

Definition 5.2.16 Let f be an analytic function on a region  except at an

isolated singularity a. If there exists an integer m such that lim |z−a|j | f (z)| =
z→a
0, ∀j > m and lim |z − a|k | f (z)| = ∞, ∀ k < m, then m is called the
z→a
algebraic order of f at a. It is easy to see that algebraic order of f at a is m iff
lim |z − a|m | f (z)| exists and is neither 0 nor ∞.
z→a
The following lemma characterizes the singularity in terms of algebraic
order.

LEMMA 5.2.17 Let f be an analytic function on a region  except at an

isolated singularity a, and the algebraic order of f at a is m. Then,
1. m < 0 iff a is a removable singularity of f and f (a) = 0,
2. m = 0 iff a is a removable singularity of f and f (a)  0,
3. m > 0 iff a is a pole of f .

Proof: Since m is the algebraic order of f at a, we have

lim |z − a| j | f (z)| = 0, ∀j > m and lim |z − a|k | f (z)| = ∞, ∀k < m.

z→a z→a

1. m < 0 ⇒ lim |z − a|0 | f (z)| = 0. Hence, a is a removable singularity

z→a
of f and f (a) = 0. Conversely, assume that a is a removable singularity

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 256 — #34

i i

256 Zeroes, Poles, and Singularities

of f and f (a) = 0. Then, f (z) = (z − a)n g(z) in a neighbourood of a for

some n ∈ N and for some analytic function g on the neighbourhood of
a such that g(a)  0. Therefore,

lim |z − a|−n | f (z)| = lim | f (z)| = |g(a)|,

z→a z→a

which is neither 0 nor ∞; hence, −n is the algebraic order of f at a.

2. m = 0 ⇒ lim |z − a|0 | f (z)|  0. Hence, a is a removable singularity

z→a
of f and f (a)  0. Conversely, assume that a is a removable singularity
of f and f (a)  0. Then, lim |z − a|0 | f (z)| = | f (a)|, which is neither 0
z→a
nor ∞, and hence, 0 is the algebraic order of f at a.

3. If m > 0, then lim |z − a|0 | f (z)| = ∞ ⇒ f has a pole at a. Conversely,

z→a
assume that a is a pole for f . Then, we have f (z) = (z − a)−n g(z) in a
neighbourhood of a for some n ∈ N and for an analytic function g on
the neighbourhood of a such that g(a)  0. Therefore,

lim |z − a|n | f (z)| = lim |g(z)| = |g(a)|,

z→a z→a

which is neither 0 nor ∞. Hence, the algebraic order of f at a

is n > 0. 

COROLLARY 5.2.18 If f is an analytic function on a region  except at an

isolated singularity a, then the algebraic order of f at a does not exist iff a is
an essential singularity of f .

Proof: The algebraic order m of f at a exists iff m < 0 or m = 0 or m > 0 iff

a is a removable singularity or a pole iff a is not an essential singularity of f .
Thus, the corollary follows. 

THEOREM 5.2.19 (Weierstrass theorem for essential singularity)

Let f be an analytic function on a region \{a}. If f has an essential singu-
larity at a, then given c ∈ C, given > 0, and given δ > 0, there exists
z ∈ B(a, δ) such that | f (z) − c| < .

Proof: Suppose that this theorem is not true. That is, there exist c ∈ C, > 0
and δ > 0 such that

|z − a| < δ ⇒ | f (z) − c| ≥ .

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 257 — #35

i i

Series Developments and Infinite Products 257

Clearly,
lim |z − a|−1 | f (z) − c| ≥ lim |z − a|−1 = ∞.
z→a z→a
Hence, by Result 5.2.15, there exists r > 0, such that

lim |z − a|r | f (z) − c| = 0.

z→a

Therefore,

lim |z − a|r | f (z)| ≤ lim |z − a|r | f (z) − c| + lim |z − a|r |c| = 0.

z→a z→a z→a

It follows that f has algebraic order at a. Then, a is a pole of f or a removable

singularity of f . This is a contradiction to the assumption that a is an essential
singularity of f . Hence, the theorem follows. 

Exercise 5.2.20 Prove that the algebraic order of f at a is the least integer
∞
m ∈ Z such that cm  0 in the Laurent series expansion of f (z) = ck (z−
k=−∞
z0 )k , 0 < |z − z0 | < δ for some δ > 0.

5.3 PARTIAL FRACTION OF ENTIRE FUNCTIONS

Definition 5.3.1 Let f be a meromorphic function on C. If {an : n ∈ N} is
the set of all poles of f , then the following representation of f is called the
partial fraction expansion of f .
∞
 1
f (z) = Pn − Qn (z) + φ(z), ∀z ∈ C\{an : n ∈ N},
n=−∞
z − an

where Pn is a polynomial without constant term, Qn is a polynomial ∀n ∈ N,

and φ is an entire function.

1
In the partial fraction of f , Pn is called the singular part of f at an .
z − an

THEOREM 5.3.2 (Mittag-Leffler theorem)

Let {an } be a sequence of complex numbers such that an → ∞ as n → ∞ and
let {Pn } be an arbitrary sequence of polynomials without constant term. Then,
there exists a meromorphic function such that whose poles are precisely {an :
1
n ∈ N} and the singular part of the function at an is Pn .
z − an

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 258 — #36

i i

258 Partial Fraction of Entire Functions

Proof: We shall construct a meromorphic function f such that

∞
 1
f (z) = Pn − Qn (z) , ∀z ∈ C\{an : n = 0, 1, 2, . . .},
z − an
n=0

for some suitable polynomial Qn such that the above series converges
uniformly on every compact subset of C\{an : n = 0, 1, 2, . . .}.
Let M > 0 be arbitrary. Since an → ∞ as n → ∞, we choose N ∈ N
1
such that 4|an | > M , ∀n ≥ N . For n ≥ N , if gn (z) = Pn , ∀z ∈
z − an
B(0, |an |), then gn is analytic on B(0, |an |). Then, the finite Taylor series
expansion of gn is given by

1 
mn
gn (z) = Pn = ck zk + φmn +1 (z)zmn +1 , ∀z ∈ B(0, |an |),
z − an
k=0

where mn ∈ N is such that

1
2mn −n > Mn = sup Pn < +∞
|z|≤ |a2n |
z − an

and
1
 Pn
1 z − an |an |
φmn +1 (z) = +1
dζ , ∀z ∈ B 0, .
i2π ζ mn (ζ − z) 2
|ζ |= |a2n |

mn  
Let Qn (z) = ck zk , ∀z ∈ C. Now for z ∈ B 0, |a4n | ,
k=0

1
Pn − Qn (z) = φmn +1 (z)zmn +1
z − an

Mn |dζ |
≤ |z|mn +1  mn +1
2π |an |
|ζ − z|
|ζ |= |a2n | 2

Mn |dζ |
≤ |z|mn +1  mn +1  
2π |an | |an | |an |
−
|ζ |= |a2n | 2 2 4

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 259 — #37

i i

Series Developments and Infinite Products 259

|an | mn +1 Mn |an | 2 mn +1
4
≤ 2π
4 2π 2 |an | |an |
= 2−mn Mn < 2−n .
∞ 1
Hence, by comparison test, Pn − Qn (z) converges uni-
n=N z − an
formly on |z| ≤ M . As every compact subset of C\{an : n =
0, 1, 2, . . .} is contained in B (0, M) for some M > 0, we conclude that
∞ 1
Pn − Qn (z) converges uniformly on every compact subset
n=N z − an
of C\{an : n = 0, 1, 2, . . .}. Thus, there exists a meromorphic function
∞
 1
f (z) = Pn − Qn (z) , +φ(z), ∀z ∈ C\{an : n = 0, 1, 2, . . .},
z − an
n=0

1
whose singular part at an is Pn , ∀n = 0, 1, 2, . . ., where φ is an
z − an
arbitrary entire function. 

Example 5.3.3 Prove the following:

π2 ∞ 1
1. = , ∀z ∈ C\Z.
k=−∞ (z − k)
2 2
sin (πz)
n 1
2. π cot(π z) = lim , ∀z ∈ C\Z.
n→∞ k=−n z−k

π n (−1)k
3. = lim , ∀z ∈ C\Z.
sin(π z) n→∞ k=−n z − k

π2
Let f (z) = , ∀z ∈ C\N. Clearly, f has double pole at every integer.
sin2 (π z)
Therefore, the Laurent series expansion of f around 0 is of the form
∞
A2 A1 
f (z) = 2
+ + An zn , 0 < |z| < δ for some δ > 0,
z z
n=0

A2 A1
and hence, the singular part of f at 0 is 2
+ , where
z z
π2
A−2 = lim z2 = 1 and A−1 = 0,
z→0 sin2 (πz)

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 260 — #38

i i

260 Partial Fraction of Entire Functions

as f is an even function, and hence, all odd coefficients in the Laurent series
1
expansion of f around 0 are zeroes. Thus, the singular part of f at 0 is 2 .
z
1
Since sin2 (π (z − n)) = sin2 (πz), the singular part of sin2 (π z) at n is .
(z − n)2
Therefore, by Mittag-Leffler theorem, we have
∞
 1
f (z) = + φ(z), ∀z ∈ C\N,
n=−∞
(z − n)2
∞1
for some entire function φ . (As itself is convergent, we have
(z − n)2
n=−∞
taken Qn (z) = 0, ∀z ∈ C and n ∈ N.) We claim that φ is identically zero.
π2 ∞ 1
First, we note that as and are periodic functions of
2
sin (πz) n=−∞ (z − n)2
period 1, φ is also a periodic function of period 1. Now, writing z = x + iy,
we get
| sin(π z)|2 = | sin(π(x + iy))|2
= | sin(πx) cosh(π y) + i sinh(πy) cos(π x)|2
= sin2 (πx) cosh2 (πy) + sinh2 (π y) cos2 (π x)
= (1 − cos2 (π x)) cosh2 (π y) + sinh2 (π y) cos2 (π x)
= cosh2 (πy) − cos2 (π x)(cosh2 (π y) − sinh2 (π y))
= cosh2 (πy) − cos2 (π x) → ∞ as |y| → +∞ uniformly in x.
Therefore,
g(x + iy) → 0 as |y| → +∞ uniformly on C\Z. (5.6)
In particular, g is bounded in the strip {(x, y) : 0 ≤ x ≤ 1, y ∈ R}. Since
g is a periodic function of period 1, g is bounded on C. Thus, by Liouville’s
theorem (Theorem 4.3.11), g is a constant function. Moreover, using equation
(5.6), we conclude that g is identically 0. Thus,
∞
 
n
π2 1 1
= = lim .
sin2 (π z) (z − k)2 n→∞ (z − k)2
k=−∞ k=−n

Integrating the above equation, we get

− cot(π z) 
n
−1
π2 = lim ,
π n→∞ z−k
k=−n

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 261 — #39

i i

Series Developments and Infinite Products 261

which implies the required series representation of π cot(π z). Next, consider

n
(−1)k 
(2j+1)
(−1)k
lim = lim
n→∞ z−k j→∞ z−k
k=−n k=−(2j+1)
⎛ ⎞

j
1  j
1
= lim ⎝ − ⎠
j→∞ z − 2k z − (2k + 1)
k=−j k=−j−1
⎛ ⎞
1  j
1  j
1 ⎠
= lim ⎝ −
2 j→∞ z
− k z−1
−k
k=−j 2 k=−j−1 2
π  πz  π (z − 1)
= cot − cot
2 2 2
⎛  πz  ⎞
π (z − 1)
cos cos
π⎜ ⎜ 2 ⎟
⎟
= ⎝  π2z  −
2 sin π (z − 1) ⎠
2 sin
⎛ 2
 πz  πz ⎞
π ⎜ cos 2 sin
⎟
= ⎝ πz +  π2z  ⎠
2 sin cos
2 2
π 1
=    
2 sin π z cos π z
2 2
π
= .
sin(πz)

PROBLEM 5.3.4 Find the partial fraction expansion for sec and deduce the
∞ (−1)n+1
Gregory–Leibniz–Madhava series, which is given by .
n=0 2n + 1

Solution:
From Example 5.3.3(3), we have
∞
π π (−1)n
$π %= = .
cos 2 − πz sin(πz) n=−∞ z − n
π
By substituting w = 2 − π z, we get
∞
π (−1)n
= ,
cos(w) n=−∞ − πw + 12 − n

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 262 — #40

i i

262 Partial Fraction of Entire Functions

which implies that

∞
 (−1)n
sec(w) = −  
n=−∞ w + n − 2 π
1

∞
  ∞
(−1)n (−1)n
= −   −  
n=1 w + n − 12 π n=0 w + −n − 12
∞
  ∞
(−1)n (−1)m
= −   +  
n=1 w + n − 12 π m=1 w − m − 12
(using the change of variable m = n + 1 in the second sum)
⎡ ⎤
∞
1 1
= (−1)n ⎣−   +   ⎦
n=1 w + n − 1
2 π w − n − 2 π
1
 
∞ 2 n − 12 π
= (−1)n  2
n=1 w2 − n − 12 π 2
∞
 (2n − 1)π
= 4 (−1)n .
4w2 − (2n − 1)2 π 2
n=1
Substituting w = 0 in the above series representation of sec(w), we get
∞
 ∞

(−1)n+1 (−1)n+1 π
1=4 ⇒ = .
(2n − 1)π (2n − 1) 4
n=1 n=1

Exercise 5.3.5
$ % ∞ 1
1. Prove that π tan πz2 = 4z . Hint: Use the identity
n=1 (2n − 1) − 4z
2 2
$ πz % $ πz %
tan 2 = cot 2 − 2 cot(πz) and the partial fraction expansion of
π cot(π z).
2. Obtain the value of Gregory–Leibniz–Madhava series from the partial
π 1
fraction expansions of πz cot(π z) and . Hint: Substitute z =
sin(π z) 4
in both expansions.

∞ 1 π2
PROBLEM 5.3.6 Prove that 2
= .
k=1 k 6

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 263 — #41

i i

Series Developments and Infinite Products 263

Solution:
From Example 5.3.3, we have
∞

π2 1
=
sin2 (π z) (z − k)2
k=−∞
 1 1 1
= $ % + 2
k2 1 − z 2 z
0k∈Z k
 1  ∞  z m−1 1
= 2
m + 2. (5.7)
k k z
0k∈Z m=1

1
Since the Laurent series of 2
is given by
sin (πz)
⎡ ⎤2
⎢ 1 ⎥
1 ⎢ 1 ⎥
= ⎢ ⎥
sin2 (πz) ⎣ πz (πz)2 (π z)4 ⎦
1− + − ···
3! 5!
 ∞ n 2
1  (πz)2 (π z)4
= + − ···
πz 3 5!
n=0
 ∞ n 2
1  (πz)2 (π z)4
= + − ···
πz 3! 5!
n=0
 ∞
1 2
= + + Am zm ,
(πz)2 3!
m=1

for some Am ∈ C, m ∈ N. Hence,

∞
π2 1 2π 2 
= + + Am zm . (5.8)
sin2 (πz) z2 3!
m=1

π2
Equating the constant terms of the expansions of in equations
sin2 (π z)
(5.7) and (5.8), we get
 1 ∞
 ∞

2π 2 1 1 π2
= = 2 ⇒ = .
6 k2 k2 k2 6
0k∈Z k=1 k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 264 — #42

i i

264 Infinite Product

5.4 INFINITE PRODUCT

Definition 5.4.1 (Rigorous definition) A sequence (ζk ) of non-zero complex
n
numbers is said to be multiplyable if ζk converges in C\{0}. Then, the
k=1

n
product of (ζk ) is denoted by ζk .
k=1

Although the above definition is rigorous, it is customary to define the

infinite product of complex numbers as follows.


n
Definition 5.4.2 (Customary definition) ζk is said to be convergent if
k=1

n
lim ζk exists in C\{0}.
n→∞ k=1


∞ 1 
n 1
Note that according to this definition, does not converge, as =
k=1 2 k=1 2
1
→ 0 as n → ∞.
2n

n
LEMMA 5.4.3 If ζk converges, then ζk → 1 as k → ∞.
k=1


n
Proof: Let n = ζk , ∀n ∈ N. Then, by assumption, we have lim n =
k=1 n→∞
k
A for some A ∈ C\{0}. As ζk = , ∀k > 1, allowing k → ∞, we get
k−1
A
lim ζk = = 1. 
k→∞ A


∞ ∞
LEMMA 5.4.4 ζk converges iff log(ζk ) converges, for a suitable
k=1 k=1
branch of log.

Proof: Using Corollary 2.4.25, for every ζk , there exists unique θk ∈ (0, 2π ]
such that arg ζk = θk . Since {θk : k ∈ N} is countable, there exists θ ∈ (0, 2π )
such that arg ζk  θ , ∀k ∈ N. Now, we define a branch of log as follows. Let
θ = C\ ({0} ∪ {r exp(iθ ) : r > 0}) and define

log z = ln |z| + i arg z with arg z ∈ (θ − 2π , θ ),

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 265 — #43

i i

Series Developments and Infinite Products 265

so that θ contains positive real axis and arg 1 = 0. Then, as in the proof
of Result 3.3.9, we can prove that log is analytic on θ . Now, assume that
∞
log(ζk ) converges. Then, using the continuity of exp, we get
k=1
 ∞
  
 
n
exp log(ζk ) = exp lim log(ζk )
n→∞
k=1 k=1
 n 

= lim exp log(ζk )
n→∞
k=1

n
= lim exp (log(ζk ))
n→∞
k=1
n
= lim ζk .
n→∞
k=1

∞
Therefore, ζk converges.
k=1

∞ 
n
Conversely, assume that ζk converges to A. If n = ζk , ∀n ∈ N, then
k=1 k=1
we have
n 
n
log = log(ζk ) − log A + i2π νn , where νn ∈ Z, ∀n ∈ N.
A
k=1
 
Using the continuity of log at 1, we get log An → log(1) = 0 as n → ∞.
Therefore,
 
n+1 n
0 = lim log − log
n→∞ A A
 n+1
 n
= lim log(ζk ) − log(ζk ) + i2π (νn+1 − νn )
n→∞
k=1 k=1
= lim [log(ζn+1 ) + i2π(νn+1 − νn )]
n→∞
= lim i2π(νn+1 − νn ) (by Lemma 5.4.3).
n→∞
n
Using → 1 and ζn → 1 as n → ∞, we obtain
A
n+1 n
arg − arg → 0 and arg ζn+1 → 0 as n → ∞.
A A

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 266 — #44

i i

266 Infinite Product

Since
n+1 n
(νn+1 − νn )2π = arg − arg + arg (1 + ζn+1 ),
A A
we get νn+1 − νn → 0 as n → ∞. Therefore, by definition of a convergent
1 1
sequence, for = > 0, there exists N ∈ N such that |νn+1 − νn | < .
2 2
Since νn+1 − νn is an integer, for each n ∈ N, we have

νn = νn+1 = ν (say), ∀n ≥ N.

Therefore, for n ≥ N , we get


n
n
log(ζk ) = log + log A − i2π ν → log A − i2π ν as n → ∞.
A
k=1

∞
Thus, log(ζk ) converges. 
k=1
It is interesting to note that we can use any branch of log which is analytic
on the positive real axis, in particular the principal logarithm.
∞ ∞

Indeed, ζk → 1 iff log(ζk ) → 0 implies that both ζk and log(ζk )
k=1 k=1
do not converge whenever ζk → 1 (or equivalently log(ζk ) → 0). In the other
case, as ζk → 1 as k → ∞, we can very well define principal logarithm
of ζk by log(ζk ) = In(|ζk |) + i arg(ζk ) where ζk ∈ (−π , π ]. However, we
can find N ∈ N such that ζk ∈ C\{(x, 0) : x ≤ 0} ∀k ≥ N , using ζk → 1 as
k → ∞. This is sufficient to get the proof of the previous lemma for principal
logarithm.


∞
Definition 5.4.5 An infinite product ζk is said to be absolutely convergent
k=1
∞
if log(ζk ) converges absolutely.
k=1
As in the case of infinite series, we can define the concept of rearrange-
ment of an infinite product as follows.


∞
Definition 5.4.6 If f : N → N is a bijection, then ζf (k) is called an
k=1

∞
rearrangement of ζk .
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 267 — #45

i i

Series Developments and Infinite Products 267


∞
RESULT 5.4.7 If ζk converges absolutely, then every rearrangement of
k=1

∞
ζk converges.
k=1


∞ ∞
Proof: By definition, ζk converges absolutely means that log(ζk ) con-
k=1 k=1
verges absolutely. Then, using Result 1.3.29, we get every rearrangement of
∞ 
∞
log(ζk ) converges, and hence, every rearrangement of ζk converges.
k=1 k=1


∞ ∞
THEOREM 5.4.8 ζk converges absolutely iff (ζk − 1) converges
k=1 k=1
absolutely.


∞ ∞
Proof: First, we note that if ζk converges absolutely or (ζk − 1) con-
k=1 k=1
verges absolutely, then by Lemma 5.4.3 or by Result 1.3.27, we have ζk → 1
as k → ∞. ∞ ∞
We claim that (ζk − 1) converges absolutely iff log(ζk ) converges
k=1 k=1
absolutely. We note that Taylor series expansion of log(ζk ) around 1 is given
by
∞
 ∞

log(m) (1) (−1)m+1
log(ζk ) = (ζk − 1)m = (ζk − 1)m .
m! m
m=0 m=1
Therefore,
∞

log(ζk ) (−1)m+1
lim = lim (ζk − 1)m−1 = 1.
k→∞ ζk − 1 k→∞ m
m=1

Therefore, for a given > 0, there exists N ∈ N such that

log(ζk )
k≥N ⇒ −1 <
ζk − 1
log(ζk )
⇒ 1− < <1+
ζk − 1
⇒ (1 − )(ζk − 1) < | log(ζk )| < (1 + )(ζk − 1).

Therefore, by comparison test (Theorem 1.3.30), our claim holds.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 268 — #46

i i

268 Infinite Product

Thus, by definition of absolute convergence of infinite product and by our

claim, we have

∞ ∞
ζk converges absolutely iff log(ζk ) converges absolutely
k=1 k=1
∞
iff (ζk − 1) converges absolutely. 
k=1

∞ 1
PROBLEM 5.4.9 Prove that 1− converges and find its value.
n=2 n2

Solution:

m 1
Let Pm = 1− , ∀m ∈ N with m ≥ 2. Therefore,
n=2 n2


m
1 1
Pm = 1− 1+
n n
n=2
m
n−1 n+1
=
n n
n=2
1 3 2 4 m−1 m+1
= ···
2 2 3 3 m m
1 m+1 1
= →  0 as m → ∞.
2 m 2

∞ 1 1
Hence, 1− converges to .
n=2 n2 2


∞ n 1
PROBLEM 5.4.10 Prove that (1 + z2 ) = , when |z| < 1.
n=1 1−z

Solution:

m−1 n
2m −1
If Pm (z) = (1 + z2 ), ∀m ∈ N, then we claim that Pm (z) = zk ,
n=1 k=0
∀m ∈ N. For m = 1, our claim is true. Assume that our claim is true for some
m ∈ N, then for

m
n
Pm+1 (z) = (1 + z2 )
n=1
m
= Pm (z) (1 + z2 )

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 269 — #47

i i

Series Developments and Infinite Products 269

2m −1 
 m
= zk (1 + z2 ) (by assumption)
k=0
m −1
2 m −1
2
m
= z +
k
zk+2
k=0 k=0
m −1
2 −1
2m+1
= zk + z j (using the change of variable j = k + 2m )
k=0 j=2m
m+1−1
2
= z j.
j=0

Hence, by induction, our claim holds. Since |z| < 1,

m −1
2
1
lim Pm (z) = lim zk =
m→∞ m→∞ 1−z
k=0

using Example 1.3.31.

∞ 
 z  −z
PROBLEM 5.4.11 Prove that 1+ e n converges absolutely and
n=1 n
uniformly on every compact subset of C.

Solution: ,
∞ z - −z 
To prove this statement, we shall obtain that 1+
e n converges
log
n=1 n
absolutely and uniformly on {z ∈ C : |z| ≤ R}, ∀R > 0 with respect to
the principal branch of log. As principal logarithm is analytic at 1, we can
∞ 1
expand the Taylor series expansion of log(1+w) = − wk , ∀w ∈ B(0, 1).
k=1 k
Therefore, for each |z| ≤ R and each n > R, we have
, z - −z   z z
log 1 + en = log 1 − −
n n n

(−1)k  z k z
∞

= −
k n n
k=1

(−1)k  z k
∞

=
k n
k=2

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 270 — #48

i i

270 Infinite Product

∞
 k
1 R
≤
k n
k=2

∞
1 R2  R k
≤
2 n2 n
k=0
⎛ ⎞
1R2 ⎜ 1⎟
≤ ⎝
2n2 R⎠
1−
n
R2
= ,
2n(n − R)

which is the nth term of a convergent series of real numbers (i.e., this
∞ , z - −z 
term is independent of z). Hence, by comparison test, log 1 + en
n=1 n
converges absolutely and uniformly on {z ∈ C : |z| ≤ R}.

Exercise 5.4.12 Prove the following:


∞ 2n + 1 1
1. 1+ = .
n=2 n2 − 1 3


∞ 2n + 1
2. 1− = 3.
n=2 n(n + 2)


∞ (−1)n
3. 1+ = 1.
n=2 n

Definition 5.4.13 Let f be an entire function with zeroes {an : n ∈ N}. Then,
the representation
∞
 z
f (z) = zm exp(g(z)) 1− exp(Pn (z)), ∀z ∈ C
ak
n=1

is called the canonical product representation of f , where g is an entire

function, Pn are some polynomials, and m ∈ {0, 1, 2, 3, . . .}.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 271 — #49

i i

Series Developments and Infinite Products 271

THEOREM 5.4.14 (Weierstrass theorem on existence of entire functions

with prescribed zeroes)
Let {an } be a sequence of complex numbers such that an → ∞ as n → ∞.
Then, there exists an entire function f whose zeroes are {an : n ∈ N}.

Proof:

Case (i) Let f be an entire function without any zeroes in C. In view of

Theorem 4.5.12, we observe that if f is an entire function and
has no zeroes, then there exists an entire function g such that
f  (z)
g(z) = log( f (z)), ∀z ∈ C and g (z) = , ∀z ∈ C. Now, we
f (z)
claim that f exp(−g) is a constant function on C.

d
(f (z) exp(−g(z))) = f  (z) exp(−g(z)) − exp(−g(z))g (z)f (z)
dz
f  (z)
= f  (z) exp(−g(z)) − exp(−g(z)) f (z)
f (z)
= f  (z) exp(−g(z)) − exp(−g(z))f  (z)
= 0.

Hence, f exp(−g) = k for some 0  k ∈ C. Then, choose c ∈ C

such that exp(c) = k . Therefore, f (z) exp(−g(z)) = exp(c), ∀z ∈ C.
Thus, g + c is an entire function such that f = exp(g + c).

Case (ii) If f has finite number of zeroes, say for example, and
a1 , a2 , a3 , . . . , an ∈ C\{0} including multiplicities and 0 is a zero
of order m (where m ≥ 0), then applying Result 5.2.3 repeatedly,
we get an entire function h, which has no zeroes in C, such that

f (z) = zm (z − a1 )(z − a2 ) · · · (z − an )h(z)

n
z  n
= zm 1− × (−1)n ak × h(z), ∀z ∈ C.
ak
k=1 k=1

Therefore, by Case (i), there exists an entire function g such that


n
(−1)n ak h(z) = exp(g(z)), ∀z ∈ C.
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 272 — #50

i i

272 Infinite Product

Thus, we can write


n
z
f (z) = zm exp(g(z)) 1− , ∀z ∈ C.
ak
k=1

Case (iii) Let f has countably infinite number of zeroes {an : n ∈ N} such
that an → ∞ as n → ∞. In this case, with the motivation from
Case (ii), we shall find an entire function f such that
∞
 z
f (z) = z exp(g(z))
m
1− , ∀z ∈ C.
ak
n=1


∞ z
However, the convergence of 1−
is not guaranteed
n=1 a k
for an arbitrary sequence of complex numbers with an →
∞ as n → ∞. Thus, we have to find a suitable sequence

∞ z
of polynomials (Pn ) such that 1− exp(Pn (z)) con-
n=1 a k

∞ z
verges. We know that 1− exp(Pn (z)) converges iff
ak
∞
 
n=1
z
log 1 − + Pn (z) converges with respect to a suitable
n=1 a k
branch of log. Let R > 0 be given. Choose n0 ∈ N  such that

|an | > 2R, ∀n ≥ n0 . If |z| ≤ R and n ≥ n0 , then log 1 − azn
is analytic and its Taylor series is given by
∞
 k
z 1 z
log 1 − = −
an k an
k=1
n k ∞
 k
1 z 1 z
= − − .
k an k an
k=1 k=n+1

n k
1 z
Therefore, for n ≥ n0 , if Pn (z) = , ∀z ∈ B(0, R) then
k=1 k an
∞
 k
z 1 z
log 1 − + Pn (z) ≤
an k an
k=n+1
∞
 k
1 R
≤
n+1 |an |
k=n+1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 273 — #51

i i

Series Developments and Infinite Products 273

∞
n+1 k
R R 1
≤ since ≤1
|an | |an | n+1
⎛
k=0 ⎞
n+1 ⎜ ⎟
R ⎜ 1 ⎟
= ⎝
|an | R ⎠
1−
|an |
n+1
1
≤ M for some M > 0.
2
⎛ ⎞
⎜ 1 ⎟
Since ⎜
⎝
⎟ converges to 1, such an M > 0 exists, by
R ⎠
1−
|an |
∞  
Corollary 1.3.4. Hence, by comparison test, log 1 − azn +
n=1
Pn (z) converges absolutely on B(0, R). Therefore, a most general
entire function that has the zeroes {an : n ∈ N}, is of the form
∞
 z
f (z) = zm exp(g(z)) 1− exp(Pn (z)), ∀z ∈ C,
ak
n=1
for some entire function g. 

COROLLARY 5.4.15 Every meromorphic function on C is a quotient of two

entire functions at all points of C except at the zeroes of f .

Proof: Let f be an arbitrary meromorphic function. From Corollary 5.2.12,

we observe that f has at most countable number of poles. If there are
countably infinite number of poles, then they must tend to ∞, otherwise,
it should have a finite limit point, which is a contradiction to the fact that the
poles are isolated. Hence, we can apply the previous theorem, and we can find
an entire function G such that its zeroes are the poles of f . Hence, f · G has
removable singularity at every pole of f, and it is analytic at every other points
of C. If F is an entire function such that ( f · G)(z) = F(z), for all point of C
F
except at the poles of f , then clearly, f = at all points of C except at the
G
poles of f . 

THEOREM 5.4.16 (Euler’s product for sin)


∞ z2
sin(π z) = π z 1− 2 .
n=1 n

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 274 — #52

i i

274 Infinite Product

Proof: As sin(πn) = 0, for every n ∈ Z, from Theorem 5.4.14, we write

∞ 
 z z
sin(πz) = z exp(g(z)) 1− exp , ∀z ∈ C,
n=−∞
n n
n0

for some entire function g. Applying logarithmic derivative on both sides, we

get
∞
π cos(π z) 1 1 1
= + g (z) + + .
sin(πz) z n=−∞
z−n n
n0
From Example 5.3.3, we have
∞

1 1 1
π cot(π z) = + + .
z n=−∞ z − n n
n0

Hence, it follows that g (z)

= 0, ∀z ∈ C ⇒ g(z) = c, ∀z ∈ C for some
constant c ∈ C. Therefore,
∞ 
 z z
sin(πz) = z exp(c) 1− exp , ∀z ∈ C.
n=−∞
n n
n0

sin(πz)
Using the easy fact that → 1 as z → 0, we observe that
πz
sin(πz)
1 = lim
z→0 πz
∞  z $ %
z exp(c) 1− exp nz
n=−∞ n
n0
= lim =1
z→0 πz
exp(c)
⇒ =1
π
⇒ exp(c) = π
∞ 
 z z
⇒ sin(πz) = π z 1− exp
n=−∞
n n
n0
∞ 
 z z z  z
⇒ sin(π z) = πz 1− 1+ exp exp −
n n n n
n=1
∞
z2
⇒ sin(πz) = π z 1− .
n2
n=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 275 — #53

i i

Series Developments and Infinite Products 275


∞ 4z2
COROLLARY 5.4.17 cos(π z) = 1−
n=1 (2n − 1)2

Proof: Using sin(2πz) = 2 sin(πz) cos(π z), we have

sin(2π z)
cos(π z) =
2 sin(π z)

∞ 4z2
2π z 1− 2
n=1 n
=

∞ z2
2πz 1−
n=1 n2

∞ 4z2
1−
n=1 n2
= . (5.9)

∞ z2
1− 2
n=1 n

Applying the following observation

∞
 ∞
 ∞

4z2 4z2 4z2
1− = 1− 1−
n2 (2k − 1)2 (2k)2
n=1 k=1 k=1
∞ ∞
4z2 z2
= 1− 1−
(2k − 1)2 k2
k=1 k=1

in equation (5.9), we get

∞
 4z2
cos(π z) = 1− .
(2n − 1)2
n=1

COROLLARY 5.4.18 (Wallis’ formulae)

∞

π 4n2
= .
2 4n2 − 1
n=1

1  2j
n
√
π = lim √ .
n→∞ n 2j − 1
j=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 276 — #54

i i

276 Infinite Product

1
Proof: Substituting z =in the Euler’s product for sin, we get
2
π  π ∞
1
∞
π  4n2 − 1
sin = 1− 2 ⇒1= ,
2 2 4n 2 4n2
n=1 n=1
which implies the first formula.
We rewrite the first formula as
∞
π 4n2
=
2 4n2 − 1
n=1

n
(2j)2 2j − 1
= lim
n→∞ (2j − 1)2 2j + 1
j=1
⎡ ⎤
n
(2j)2
1 3 2n − 1 ⎦
= lim ⎣ ×
· ···
n→∞ (2j − 1)2
3 5 2n + 1
j=1
⎡ ⎤
n
(2j)2 1 ⎦
= lim ⎣ ×
n→∞ (2j − 1) 2 2n + 1
j=1
 
1  (2j)2
n
1
= lim as lim =1 .
n→∞ 2n (2j − 1)2 n→∞ 1 + 1
2n
j=1

√ 1  n 2j
This implies that π = lim √ . 
n→∞ n j=1 2j − 1

Exercise 5.4.19
1. Prove the following:

∞ z2
(a) sinh(π z) = πz 1+ .
n=1 n2

∞ z2
(b) cosh(πz) = 1+ .
n=1 (2n − 1)2
√ 
∞ (−1)n+1 π
2. Prove that 2= 1+ . Hint: Substitute z = in the
n=1 2n − 1 4
expansion of cos(π z).

π 
∞ (2n)2 1
3. Prove that = . Hint: Substitute z = in the
2
n=1 (2n − 1)(2n + 1) 2
sin(πz)
expansion of .
πz

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 277 — #55

i i

Series Developments and Infinite Products 277

∞ 1 π2 ∞ 1 π4
RESULT 5.4.20 2
= and 4
= .
n=1 n 6 n=1 n 90

Proof: 4 Recall that the Euler’s product for sin is given by

∞
sin(πx)  x2
= 1− 2 , ∀x ∈ [0, 1).
πx n
n=1

∞ x2
If g(x) = log 1 − , then from the proof of Lemma 5.4.4, we see
n=1 n2
sin(π x)
that = exp(g(x)). Therefore,
πx
 2 k
∞ ∞ 
∞ x
sin(πx)  x2  n2
log = g(x) = log 1 − = , ∀x ∈ [0, 1).
πx n2 k
n=1 n=1 k=1

Since
 2 k
∞ 
∞ x ∞
 n2  |x|2
= log 1 − = g(|x|) < +∞,
k n2
n=1 k=1 n=1

we can apply Fubini’s theorem5 to get

∞
∞  ∞
∞ 
sin(π x)   1 x2k   1 x2k
log = = , ∀x ∈ [0, 1).
πx n2k k n 2k k
k=1 n=1 k=1 n=1
(5.10)
On the other hand, we have
 
sin(πx) (πx)2 (πx)4
log = log 1 − − + ···
πz 3! 5!
   2
(πx)2 (πx)4 1 (π x)2 (π x)4
= − +··· + − + ··· + ···
3! 5! 2 3! 5!
1 1 1
= (πx)2 + − + (π x)4
3! 5! 2 × (3!)2
4 There are plenty of proofs for these identities. We present here the Euler’s proof.
5 Fubini’s theorem is a well-known big theorem on integral over a product measure space. As
it is beyond the scope of this book, we have not even stated this theorem. Interested readers can
refer to any book on measure and integration to see this theorem.

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 278 — #56

i i

278 Gamma Function and Its Properties

1 1 1
+ − + (π x)6 + · · · . (5.11)
7! 3! × 5! 3 × (3!)3

Comparing the coefficients of x2 and x4 from equations (5.10) and (5.11), we

π2 ∞ 1 1 ∞ 1
get = and π 4 − 1 + 1 = .
6 2 120 72 2 n=1 n4
n=1 n
π2 ∞ 1 π4 ∞ 1
Thus, we get = 2
and = 4
. 
6 n=1 n 90 n=1 n

π6
Exercise 5.4.21 Prove that . Hint: Compare the coefficients of x6 from
945
equations (5.10) and (5.11).

5.5 GAMMA FUNCTION AND ITS PROPERTIES

Definition 5.5.1 The gamma function  is defined by

exp(−γ z)   z −1 z

∞
(z) = 1+ exp , ∀z ∈ C\{−n : n ∈ N},
z n n
n=1

1 1 1
where γ = lim 1+ + + · · · + − ln(n) , which is called the
n→∞ 2 3 n
Euler’s constant.
∞ 
 z  z
First, we note that the infinite product 1+ exp − converges
n=1 n n
uniformly on the compact subsets of C. If |z| ≤ r and n > 2r, then
 z z z z2 z3 z
log 1 + − = − 2 + 3 − ··· −
n n n 2n 3n n
∞
 rk
≤
knk
k=2
∞ k
 r
<
nk
k=2
∞
r2  rk
=
n2 nk
k=0

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 279 — #57

i i

Series Developments and Infinite Products 279

r2  r −1 r 1
= 1 − (as 0 < < < 1)
n2 n n 2
r2  r −1
= 2 2 (as 1 − < 2).
n n
∞  
Hence, log 1 + 1n − z
n converges uniformly on compact subsets
n=1
∞ 
 z  z
of C. Equivalently, the infinite product 1+ exp − converges
n=1 n n
∞ 
 z  z
uniformly on the compact subsets of C. If G(z) = 1+ exp − ,
n=1 n n
then (z) = z exp(γ1z)G(z) , and hence,  is well defined on C \ {−n : n ∈ N}.

THEOREM 5.5.2 (Properties of  )

The  function satisfies the following properties:

1. (z + 1) = z(z), ∀z ∈ C,

2. (n) = (n − 1)!, ∀n ∈ N,
  √
3.  12 = π ,

4. Legendre’s duplication formula

√ 1
π (2z) = 22z−1 (z) z + , ∀z ∈ C.
2

Proof:
∞ 
 z  z
1. Let G(z) = 1+ exp − , ∀z ∈ C. Hence, the zeroes of
n=1 n n
G are −1, −2, −3, . . . . This implies that the zeroes of G(z − 1) are
0, −1, −2, −3, . . . . Therefore, from Theorem 5.4.14, we have

G(z − 1) = z exp(φ(z))G(z), ∀z ∈ C,

for some entire function φ . That is, for each z ∈ C, we have

∞
z  z
∞
 
z−1 (z − 1)
1+ exp − = z exp(φ(z)) 1+ exp − .
n n n n
n=1 n=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 280 — #58

i i

280 Gamma Function and Its Properties

Finding the logarithmic derivative of the above equation, we get

∞
  ∞
1 1 1 1 1
− = + φ  (z) + − . (5.12)
z−1+n n z z+n n
n=1 n=1

Now,
∞
  ∞
1 1 1 1 1
− = −1+ −
z−1+n n z z−1+n n
n=1 n=2
 ∞
1 1 1
= −1+ −
z z+m m+1
m=1
∞
1 1 1
= −1+ −
z z+n n
n=1
∞
 1 1
+ −
n n+1
n=1
 ∞
1 1 1
= −1+ −
z z+n n
n=1
1
+ lim 1−
n→∞ n+1
∞
1  1 1
= + − .
z z+n n
n=1

Applying the last observation in equation (5.12), we get φ  (z) =

0, ∀z ∈ C, and hence, it is a constant, say, C . To find the value of
C , substitute z = 1 in G(z − 1) = z exp(φ(z))G(z) to get
∞
 1 1
G(0) = exp(C)G(1) ⇒ 1 = exp(C) 1+ exp − ,
n n
n=1

which implies that

∞
 1 1
−C = ln 1 + −
n n
n=1

N
1 1
= lim ln 1 + −
N→∞ n n
n=1

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 281 — #59

i i

Series Developments and Infinite Products 281

1 1 1
= lim + + ··· +
ln (N + 1) − 1 +
N→∞ 2 3 N
N +1 1 1 1
= lim ln (N) + ln − 1 + + + ··· +
N→∞ N 2 3 N
1 1 1
= lim ln (N) − 1 + + + · · · + ,
N→∞ 2 3 N
 
as lim ln N+1
N = ln(1) = 0. Thus, C = γ , the Euler’s constant.
N→∞
Thus, we have
G(z − 1) = z exp(γ )G(z), ∀z ∈ C.
From the definition of  and G, we can easily write that
1
(z) =
z exp(γ z)G(z)
1
(z + 1) =
(z + 1) exp(γ (z + 1))G(z + 1)
1
=
(z + 1) exp(γ (z + 1))G(z)(z + 1)−1 exp(−γ )
1
=
exp(γ z)G(z)
= z(z).

2. We prove this identity by induction on n.

1
(1) = = 1 = 0!, as 1 = G(0) = exp(γ )G(1).
exp(γ )G(1)
Assume that this identity holds for some n ∈ N. Now, by using (1) and
induction hypothesis, we get (n + 1) = n(n) = n(n − 1)! = n! thus,
(n) = n!, ∀n ∈ N.
3. From the definition of G and Example 5.3.3, we have zG(z)G(−z) =

∞ z2 sin(π z)
z 1− 2 = . Therefore, for every z ∈ C,
n=1 n π
(1 − z)(z) = (−z)(−z)(z)
1 1
= (−z)
(−z) exp(−γ z)G(−z) z exp(γ z)G(z)
1 π
= = .
zG(−z)G(z) sin(π z)

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 282 — #60

i i

282 Gamma Function and Its Properties

1
Substituting z = , we get
2
1 1 π 1 √
 1−  = ⇒ = π.
2 2 1 2

4. Finding the logarithmic derivative on both sides of

exp(−γ z)   z −1 z

∞
(z) = 1+ exp ,
z n n
n=1

we get
 −1 ∞
  (z) −1 1
= −γ + + .
(z) z z+n n
n=1

Finding the derivative on both sides of the last equation, we get

∞
  ∞
d   (z) 1 1 1
= + = .
dz (z) z 2 (z + n) 2 (z + n)2
n=1 n=0

Therefore,
⎛ ⎞
1
 z +
d   (z) d ⎜
⎜ 2 ⎟ ⎟
+
dz (z) dz ⎝ 1 ⎠
 z+
2
∞ ∞
1 1
= +
(z + n)2 2
n=0 n=0 z + 1 + n
2
∞ ∞

 1  1
=4 +
(2z + 2n)2 (2z + 2n + 1)2
n=0 n=0
∞
 1
=4
(2z + m)2
m=0
1 d   (2z)
=4×
2 dz (2z)
d   (2z)
=2 .
dz (2z)

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 283 — #61

i i

Series Developments and Infinite Products 283

Integrating on both sides, we get

1
 z +
  (z) 2   (2z)
+ =2 + a,
(z) 1 (2z)
 z+
2

for some constant a ∈ C Again integrating on both sides, we get

1
log ((z)) + log  z + = log ((2z)) + az + b,
2

which implies the following equation.

1
(z) ×  z + = exp(az + b)(2z). (5.13)
2

1
Substituting z = and z = 1 in equation (5.13), we get
2
1 a  a √
 (1) = exp + b (1) ⇒ + b = log( π ), (5.14)
2 2 2
√
3 π
(1) = exp(a + b)(2) ⇒ a + b = log . (5.15)
2 2

√
a π √
Equation (5.15)−equation (5.14) ⇒ = log − log( π )
2 2
= − log(2) ⇒ a = −2 log 2.
1
Using this value in equation (5.15), we get b = log(2) + log(π).
2
Therefore, from equation (5.13), we get

1 1
(z) z + = exp (−2 log(2))z + log(2) + log(π) (2z)
2 2
√
= 2−2z+1 π (2z).

Thus, we get

√ 1
π (2z) = 22z−1 (z) z + .
2 

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 284 — #62

i i

284 Gamma Function and Its Properties

N zN !
Example 5.5.3 Prove that (z) = lim , ∀z ∈
N→∞ z(z + 1) · · · (z + N)
C\{−1, −2, −3, . . .}.
Solution:
For z ∈ C\{0, −1, −2, −3, . . .}, we have

exp(−γ z)   z −1 z

∞
(z) = 1+ exp
z n n
n=1
N 
exp − n − ln(N) z
1
N 
n=1
 z −1 z
= lim · lim 1+ exp
N→∞ z N→∞ n n
n=1
⎛ N  ⎞
⎜ exp − 1
− ln(N) z  z ⎟
⎜ n=1
n 
N
z + n −1 ⎟
= lim ⎜ · exp ⎟
N→∞ ⎝ z n n ⎠
n=1


N  z  exp (z ln(N)) N
n z
= lim exp − · exp
N→∞ n z z+n n
n=1 n=1

Nz 
N
n
= lim ·
N→∞ z z+n
n=1
N zN !
= lim .
N→∞ z(z + 1) · · · (z + N)

In the following example, we justify that  has some other integral rep-
resentation. However, this justification depends on dominated convergence
theorem. Hence, the following example is for those who are familiar with the
dominated convergence theorem.

∞
Example 5.5.4 Prove that (z) = tz−1 exp(−t) dt, ∀z = (x, y) ∈ C with
0
x > 0.
Solution:
First, we claim that the above integral exists for all z = (x, y) ∈ C with x > 0.
Since exp(−t) ≤ 1, ∀t ∈ [0, ∞], we have

tz−1 exp(−t) < |exp((z − 1) log t)|

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 285 — #63

i i

Series Developments and Infinite Products 285

= |exp((x − 1) log t) exp(iy log t)|

= |exp((x − 1) log t)| .
Since for x > 0,
∞ 1 ∞
t x−1
exp(−t) dt = t x−1
exp(−t) dt + tx−1 exp(−t) dt
0 0 1
1 ∞
= lim t x−1
exp(−t) dt + tx−1 exp(−t) dt
→0
1
∞
1− x
= lim + tx−1 exp(−t) dt
→0 x
1
∞
1
< + exp(−t/2) dt < +∞,
x
1
∞
our claim holds. Next, we claim that if G(z) = tz−1 exp(−t) dt, ∀z ∈ {z ∈
0
C : Re z > 0}, then G is an analytic function. Fix z0 ∈ C such that Re z > 0.
Now, for z0 ∈ C such that Re z > 0, by applying dominated convergence
theorem, we get
∞
G(z) − G(z0 )
lim − tz−1 log(t) exp(−t) dt
z→z0 z − z0
0
∞
tz−1 − tz0 −1
= lim − tz0 −1 log(t) exp(−t) dt
z→z0 z − z0
0
∞
tz−1 − tz0 −1
= lim − tz0 −1 log(t) exp(−t) dt
z→z0 z − z0
0
= 0.
Hence, G is analytic on {z∈ C : Re z > 0}.
n
Using exp(−t) = lim , ∀t > 0, we get G(z) = lim Gn (z),
1− 1
n
n→∞ n→∞
n z−1 $ %
t n
∀z ∈ {z ∈ C : Re z > 0}, where Gn (z) = t 1 − n dt, ∀z ∈ C,
0
with Re z > 0, ∀n ∈ N. We note that

i i

i i
 i i

“book” — 2014/6/4 — 21:09 — page 286 — #64

i i

286 Gamma Function and Its Properties

n n
t
Gn (z) = tz−1 1 − dt
n
0
1
= n z
sz−1 (1 − s)n ds
0
(by using the change of variable t = ns)
⎧  
⎨ 1 n z+1
Gn+1 (z + 1), n ≥ 2
= z n−1 .
⎩ 1
n=1
z(z+1) ,
(by applying integration by parts)

By iterating n − 1 times, we get

nz n!
Gn (z) = , ∀z ∈ C with Re z > 0, ∀n ≥ 1,
z(z + 1)(z + 2) · · · (z + n)

and hence,
nz n!
G(z) = lim , ∀z ∈ C with Re z > 0.
n→∞ z(z + 1)(z + 2) · · · (z + n)

Thus, by using the previous example, we get

G(z) = (z), z ∈ C with Re z > 0.

∞
Remark 5.5.5: The integral tz−1 exp(−t) dt does not exist for z ∈ C with
0
Re z < 0.
Indeed, if z = x + iy with x < 0 and > 0, then

∞ 1
1− x
|t z−1
exp(−t)| dt ≥ tx−1 exp(−t) dt = exp(−1) → ∞ as → 0.
x
0

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 287 — #1

i i

6
Residue Calculus

6.1 RESIDUE
Definition 6.1.1 Let f be an analytic function except at a singularity ‘a’.
The residue of f at ‘a’ is defined to be a complex number R, which makes
R
f (z) − as the derivative of an analytic function on B(a, δ)\{a} for some
z−a
δ > 0.

Residue of f at a is unique, and hence, it is denoted by (Res f )(a).

LEMMA 6.1.2 If f is an analytic function on  except at an isolated

singularity a, then residue of f at a exists and is unique.

Proof: Since a is an isolated singularity, then there exists δ > 0 such that f is
analytic on (Cl B(a, δ))\{a}. For 0 < r < δ  circle |z − a| = r
, let C denote the
1   R
and R = f (z) dz. We claim that f (z) − dz = 0 for every
i2π C γ z−a
closed curve γ in B(a, δ)\{a}.

Case 1: γ is homologous to 0 in B(a, δ)\{a}.

If
R
F(z) = f (z) − , ∀z ∈ Cl B(a, δ)\{a}.
z−a
then F is analytic on Cl B(a, δ)\{a}, and by general version of
Cauchy’s theorem, we get F(z) dz = 0.
γ

Case 2: γ is not homologous to 0 in B(a, δ)\{a}.

Then, WN(γ , a) = m for some m ∈ Z\{0}. We claim that γ −mC ∼ 0

287

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 288 — #2

i i

288 Residue

in B(a, δ)\{a}. If z0  B(a, δ)\{a}, then either z0 = a or z0  B(a, δ).

If z0 = a, then
WN (γ − mC, z0 ) = WN(γ , a) − mWN(C, a) = m − m × 1 = 0.
If z0  B(a, δ), then
WN (γ , z0 ) = 0 and WN(C, z0 ) = 0 and WN(γ − mC, z0 ) = 0.
Hence, by general version of Cauchy’s theorem, we get
  
R
0 = f (z) − dz
z−a
γ −mC
  
R
= f (z) − dz
z−a
γ −mC
     
R R
= f (z) − dz − m f (z) − dz,
z−a z−a
γ C

which implies
⎛ ⎞
    
R dz ⎠
f (z) − dz = m⎝ f (z) dz − R
z−a z−a
γ
⎛C C ⎞

dz ⎠
= m ⎝i2π R − R
z−a
C
= m (i2π R − Ri2π ) = 0.
Hence, by Result 4.1.30, F is the derivative of an analytic function.
1 
Thus, residue of f at a exists, and (Res f )(a) = R = f (z) dz.
i2π C
R S
If there are R, S ∈ C such that f (z) − and f (z) −
z−a z−a
are derivatives of the analytic functions  and  on B(a, δ)\{a}
and B(a, )\{a}, respectively, then obviously, R − S ∈ C and
S−R
( − ) (z) = in B(a, min{δ, }) \ {a}. Hence, by Corollary
z−a
4.1.29, we get

S−R
dz = 0 ⇒ (R − S)i2π = 0 ⇒ R = S.
z−a
C
Thus, the lemma follows. 

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 289 — #3

i i

Residue Calculus 289

LEMMA 6.1.3 If ‘a’ is an isolated singularity of f and if the Laurent series

∞
expansion of f in B(a, r)\{a} ( for some r > 0) is f (z) = cn (z − a)n ,
n=−∞
then (Res f )(a) = c−1 .

(z − a)n+1
Proof: We know that (z − a)n is the derivative of inside
n+1
B(a, r)\{a}. Using the fact that
1 1
cn n 1 c−n n 1
lim sup = lim sup |cn | n and lim sup = lim sup |c−n | n ,
n→∞ n+1 n→∞ n→∞ −n + 1 n→∞

∞ cn
we conclude that (z − a)n converges in the neighbourhood of a
n=−∞ n +1
n−1
∞
(except at a) on which cn (z − a)n+1 converges (cf. the proof of Theorem
n=−∞
n−1
2.3.2). Hence,
⎛ ⎞
∞ ∞
c1 d ⎜ cn ⎟
f (z) − = cn (z − a)n = ⎝ (z − a)n+1 ⎠ ,
z − a n=−∞ dz n=−∞ n + 1
n−1 n−1

for every 0 < |z − a| < r. Thus, (Res f )(a) = c−1 . 

RESULT 6.1.4 If f is a meromorphic function with a pole ‘a’ of order m, then

1 d m−1
(Res f ) (a) = lim m−1 ((z − a)m f (z)).
(m − 1)! z→a dz

Proof: If f has a pole of order m at a, then by Result 5.2.14, we have

∞
f (z) = cn (z − a)n , ∀z ∈ B(a, r)\{a},
n=−m

for some r > 0. From the previous lemma, we get

1 d m−1  
lim m−1 (z − a)m f (z)
(m − 1)! z→a dz
 ∞

1 d m−1
= lim (z − a) m
cn (z − a) n
(m − 1)! z→a dzm−1 n=−m

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 290 — #4

i i

290 Residue
 ∞

1 d m−1
= lim m−1 cn−m (z − a) n
which is equal to c−1 if m = 1.
(m−1)! z→a dz
n=0
∞
m−2 
1 
= lim cn−m (n − ν) (z − a)n−m+1 if m ≥ 2
(m − 1)! z→a
n=m−1 ν=0
1
= (m − 1)!c−1
(m − 1)!
= c−1
= (Res f )(a).

Thus, the result follows. 

Example 6.1.5 Find the residue of f at all poles.

z2 + 3 z
(1) f (z) = ; (2) f (z) = ;
z−1 (z + 1)(z − 2)
2z + 3 z2 + 2
(3) f (z) = ; (4) f (z) = .
(z + 2)2 (z − 3) (z + 2)3
Solution:
z2 + 3
1. If f (z) = , then 1 is the only simple pole of f . Then,
z−1
(Res f )(1) = lim (z − 1)f (z) = lim (z2 + 3) = 4.
z→1 z→1

z
2. If f (z) = , then −1 and 2 are the simple poles of f . Then,
(z + 1)(z − 2)
z 1
(Res f )(−1) = lim (z + 1)f (z) = lim =
z→−1 z→−1 z − 2 3
z 2
and (Res f )(2) = lim (z − 2)f (z) = lim = .
z→2 z→2 z + 1 3

2z + 3
3. If f (z) = , then −2 is a double pole of f , and 3 is a
(z + 2)2 (z − 3)
simple pole of f . Then,  
d d 2z + 3
(Res f )(−2) = lim [(z + 2) f (z)] = lim
2
z→−2 dz z→−2 dz z−3
−9 −9
= lim =
z→−2 (z − 3)2 25
2z + 3 9
and (Res f )(3) = lim (z − 3)f (z) = lim = .
z→3 z→3 (z + 2)2 25

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 291 — #5

i i

Residue Calculus 291

z2 + 2
4. If f (z) = , then −2 is the only pole of f of order 3. Then,
(z + 2)3
1 d2
(Res f )(−2) = lim [(z + 2)3 f (z)]
2 z→−2 dz2
1 d2  2 
= lim 2
z +2
2 z→−2 dz
1
= × 2 = 1.
2

Example 6.1.6 Evaluate the residues of the following functions at their

poles:
z2 + 1 z+1 exp(z)
(1) 2 ; (2) 2 ; (3) .
z + 5z + 6 (z + 1) 2 (z + 3)(z − 2)2
Solution:
z2 + 1
1. Let f (z) = , ∀z ∈ C. The poles of f are the zeroes of the
z2 + 5z + 6
polynomial z2 + 5z + 6. Therefore, z = −2 and z = −3 are the simple
poles of f . Hence,
(Res f ) (−2) = lim (z + 2)f (z)
z→−2
z2 + 1
= lim (z + 2)
z→−2 z2 + 5z + 6
z2 + 1
= lim
z→−2 z + 3
= 5.
(Res f ) (−3) = lim (z + 3)f (z)
z→−3
z2 + 1
= lim (z + 3)
z→−3 z2 + 5z + 6
z2 + 1
= lim
z→−3 z + 2
= −10.
z+1
2. Let f (z) = , ∀z ∈ C. z = i and z = −i are the poles of the
(z2 + 1)2
order 2 for f .
 
1 d z+1
(Res f ) (i) = lim (z − i)2 2
(2 − 1)! z→i dz (z + 1)2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 292 — #6

i i

292 Residue
 
d z+1
= lim
z→i dz (z + i)2
 
(z + i)2 − (z + 1)2(z + i)
= lim
z→i (z + i)4
(i2)2 − (i + 1)2(i2)
=
(i2)4
−4 − i4 + 4
=
16
i
= − .
4
 
1 d z+1
(Res f ) (−i) = lim (z + i)2 2
(2 − 1)! z→−i dz (z + 1)2
 
d z+1
= lim
z→−i dz (z − i)2
 
(z − i)2 − (z + 1)2(z − i)
= lim
z→−i (z − i)4
i
= .
4

exp(z)
3. Let f (z) = , ∀z ∈ C. z = −3 is a simple pole, and
(z + 3)(z − 2)2
z = 2 is a pole of the order 3 for f .
 
exp(z)
(Res f ) (−3) = lim (z + 3)
z→−3 (z + 3)(z − 2)2
exp(z)
= lim
z→−3 (z + 3)(z − 2)2
exp(−3)
= .
25
 
1 d2 exp(z)
(Res f ) (2) = lim (z − 2)3
(3 − 1)! z→2 dz2 (z + 3)(z − 2)2
2  
1 d exp(z)
= lim 2
2 z→2 dz z+3
 
1 d (z + 2) exp(z)
= lim
2 z→2 dz (z + 3)2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 293 — #7

i i

Residue Calculus 293

1 exp(z)[(z + 3)2 − 2(z + 2)]

= lim
2 z→2 (z + 3)3
17
= exp(2).
250

Example 6.1.7 Find the residues of f at the given singularities.

 
sin(3z) cos(z) 2 1 π
(1) at 0; (2) at 0; (3) exp at 0; (4) at .
z 2z 3z cos(z) 2
Solution:
sin(3z) ∞ (−1)k (3z)2k−1
1. Let f (z) = , ∀z  0. Since sin(3z) = , we
z (2k − 1)!
  k=0
sin(z) ∞ (−1)k (3z)2k
have lim = lim 3 + = 3 in C. Hence, 0 is
z→0 z z→0 k=1 3(2k + 1)!
a removable singularity for f , and hence, (Res f )(0) = 0.
cos(z)
2. Let f (z) = , ∀z  0. By a similar argument we get
2z
cos 3z 1 ∞ (−1)k (z)2k−1
= + , which is the Larurent series
z 2z k=1 2(2k)!
1
expansion of f in 0 < |z| < ∞. Hence, (Res f )(0) = (the coefficient
2
1
of ).
z
 
2 k
  ∞
2 3z
3. If f (z) = exp , then f (z) = . Hence, 0 is an essential
3z k=0 k!
2
singularity of f , and hence, (Res f )(0) = .
3

1 π π
4. Let f (z) = , ∀z  (2k − 1) , k ∈ Z. Clearly, z =
cos(z) 2 2
 π 1
is a simple pole. Hence, (Res f )(0) = lim z − =
π 2 cos(z)
z→
 2
π −1
lim z− = −1.
π 2 sin(z − π )
z→
2 2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 294 — #8

i i

294 Cauchy’s Residue Theorem

Exercise 6.1.8 Evaluate the residues of f at its singularities.

sin(z) 1 exp(z) − 1
(1) 2 at 0; (2) at π ; (3) at z = 0.
z sin2 (z) z4
1
Answers: (1) 1; (2) 0; (3) .
6

RESULT 6.1.9 If f and g are analytic functions  on a region  such that
f f (z0 )
f (z0 )  0 and g has a simple zero at z0 , then Res (z0 ) =  .
g g (z0 )

f
Proof: By assumption, it is clear that has a simple pole at z0 . Hence, by
g
Lemma 6.1.4,
 
f f (z)
Res (z0 ) = lim (z − z0 )
g z→z0 g(z)
f (z)
= lim
z→z0 g(z)
z−z0
f (z)
= lim
z→z0 g(z)−g(z0 )
z−z0
lim f (z)
z→z0
=
lim g(z)−g(z
z−z0
0)
z→z0
f (z0 )
= .
g (z0 )
Hence, the result follows. 

6.2 CAUCHY’S RESIDUE THEOREM

THEOREM 6.2.1 (Cauchy’s residue theorem)
Let f be an analytic function on a region  except at a finite number of
singularities. Then,
 n
f (z) dz = i2π WN (γ , aj )(Res f )(aj ),
γ j=1

for every closed curve γ in  \ {a1 , a2 , . . . , an } such that γ ∼ 0 in .

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 295 — #9

i i

Residue Calculus 295

Proof: Let a1 , a2 , a3 , . . . , an be the isolated singularities of f such that aj 

ak if j  k . Choose δj > 0 such that
1. Cl B(aj , δj ) ⊆  \ γ and
2. Cl B(aj , δj ) ∩ Cl B(ak , δk ) = ∅ for every j, k ∈ {1, 2, . . . , n} with j  k .
In view of proof of Lemma 6.1.2, if Cj is the circle |z − aj | = δj , then we
1  n
have (Res f )(aj ) = f (z) dz. We claim that γ − WN (γ , aj )Cj ∼ 0
i2π Cj j=1
in ∗ = \{a1 , a2 , a3 , . . . , an }. If z0  ∗ , then either z0   or z0 = aj
for some j ∈ {1, 2, 3, . . . , n}. If z0  , then WN(γ , z0 ) = WN(Cj , z0 ) = 0
n
for all j ∈ {1, 2, 3, . . . , n}, and hence, WN γ − WN (γ , aj )Cj = 0. If
j=1

1 if j = k
z0 = ak , then WN(Cj , z0 ) = for all j ∈ {1, 2, 3, . . . , n}, and
0 if j  k
 
n
hence, WN γ − WN (γ , aj )Cj = WN(γ , ak ) − WN(γ , ak ) = 0.
j=1
Hence, our claim holds. Therefore, by general version of Cauchy’s theorem,
we get
  n 
0= f (z) dz = f (z) dz − WN (γ , aj ) f (z) dz,
n γ j=1 Cj
γ− WN (γ ,aj )Cj
j=1

 n
which implies that f (z) dz = i2π WN (γ , aj )(Res f )(aj ). 
γ j=1

Example 6.2.2 Evaluate the following integrals using Cauchy’s residue the-
orem:
 z+1
1. dz.
|z|=3 z + 7z + 10
2

 3z2 − 2z + 1
2. dz.
|z|=4 z2 − 2z

 z+3
3. dz.
|z|=3 (z2 + 6z + 8)(z2 + 1)

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 296 — #10

i i

296 Cauchy’s Residue Theorem

 z2 + 2
4. dz.
|z−1|=3 (z − 2)2 (z + 4)
 z
5. dz.
|z+1−i|=6 (z2 − 2z − 15)(z2 + 1)
 sin(z)
6. 2 + 1)2
dz, where γ is the rectangle with vertices (−1, 0), (1, 0),
γ (z
(1, 2), and (−1, 2).
 4z + 1
7. dz.
|z|=1 (z + 1)(z + 2)
2 2

 z3 + 1
8. dz.
|z|=4 z(z + 1)(z + 2)
 exp(z) − 1
9. dz.
|z−1|=5 sin(z)
 exp(z)
10. dz.
|z|=1 sin(z)

Solution:
z+1
1. Let f (z) = , ∀z ∈ C\{−2, −5}. Clearly, −2 and −5 are
z2 + 7z + 10
the simple poles of f . We note that −2 is enclosed by |z| = 3 and −5
is not enclosed by |z| = 3. Therefore, by Cauchy’s residue theorem, we
get

z+1
dz = i2π (Res f )(−2)
z2 + 7z + 10
|z|=3
= i2π lim (z + 2)f (z)
z→−2
z+1
= i2π lim
z→−2 z+5
−i2π
= .
3

3z2 − 2z + 1
2. Let f (z) = dz, ∀z ∈ C\{0, 2}. Clearly, 0 and 2 are the
z2 − 2z
simple poles of f and both are enclosed by |z| = 4. Therefore, by
Cauchy’s residue theorem, we get

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 297 — #11

i i

Residue Calculus 297


3z2 − 2z + 1
dz = i2π [(Res f )(0) + (Res f )(2)] .
z2 − 2z
|z|=4

Since

(Res f )(0) = lim zf (z)

z→0
3z2 − 2z + 1
= lim
z→0 z−2
1
= −
2
(Res f )(2) = lim (z − 2)f (z)
z→2
3z2 − 2z + 1
= lim
z→2 z
9
= ,
2
we get  

3z2 − 2z + 1 −1 9
dz = i2π + = i8π .
z2 − 2z 2 2
|z|=4

z+3
3. Let f (z) = , ∀z ∈ C\{−2, −4, i, −i}. Clearly,
+ 6z + 8)(z2 + 1)
(z2
−2, −4, i, and −i are simple poles of f , and −2, i, and −i are enclosed
by |z| = 3, but −4 is not enclosed by |z| = 3. Therefore, by Cauchy’s
residue theorem,

z+3
dz
(z2 + 6z + 8)(z2 + 1)
|z|=3
= i2π [(Res f )(−2) + (Res f )(−i) + (Res f )(i)] .

Since
z+3 −1
(Res f )(−2) = lim (z + 2)f (z) = lim = ,
z→−2 (z + 4)(z2 + 1)
z→−2 10
z+3
(Res f )(−i) = lim (z + i)f (z) = lim 2
z→−i z→−i (z + 6z + 8)(z − i)
−i + 3 −11 + i27
= = ,
(7 − i6)(−i2) 170

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 298 — #12

i i

298 Cauchy’s Residue Theorem

and
z+3
(Res f )(i) = lim(z − i)f (z) = lim
z→i (z2 + 6z + 8)(z + i)
z→i
i+3 −11 − i27
= = ,
(7 + i6)(i2) 170
we get

z+3
dz
(z2 + 6z + 8)(z2 + 1)
|z|=3
 
−1 −11 + i27 −11 − i27
= i2π + +
10 170 170
 
−39
= i2π
170
−39πi
= .
85

z2 + 2
4. Let f (z) = , ∀z ∈ C\{2, −4}. Obviously, 2 is a double
(z − 2)2 (z + 4)
pole and −4 is a simple pole for f . 2 is enclosed by |z − 1| = 3, and
−4 is not enclosed by |z − 1| = 3. Since
 
d z2 + 2 (z + 4)2z − (z2 + 2) 1
(Res f )(2) = lim = = ,
z→2 dz z+4 (z + 4)2 2
by Cauchy’s residue theorem, we get

z2 + 2 1
dz = i2π [(Res f )(2)] = i2π = iπ .
(z − 2) (z + 4)
2 2
|z−1|=3

z
5. Let f (z) = , ∀z ∈ C\{5, −3, i, −i}. Clearly,
− 2z − 15)(z2 + 1)
(z2
5, −3, i, and −i are poles of f , and |z + 1 − i| = 6 encloses −3, i,
and −i but not 5. By Cauchy’s residue theorem, we have

z
dz
(z − 2z − 15)(z2 + 1)
2
|z+1−i|=6
= i2π [(Res f )(−3) + (Res f )(i) + (Res f )(−i)] .

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 299 — #13

i i

Residue Calculus 299

Since
z 3
(Res f )(−3) = lim = ,
z→−3 (z − 5)(z + 1)
2 80
z
(Res f )(i) = lim 2
z→i (z − 2z − 15)(z + i)
i −8 + i
= = ,
(−16 − i2)(i2) 260
and
z
(Res f )(−i) = lim
z→−i(z2 − 2z − 15)(z − i)
1 −8 − i
= = ,
2(−16 + i2) 260
we have

z
dz
(z2− 2z − 15)(z2 + 1)
|z+1−i|=6
 
3 −8 + i −8 − i
= i2π + +
 80 260 260 
3 −8 + i −8 − i
= i2π + +
80 260 260
iπ
= .
52

sin(z)
6. Let f (z) = , ∀z ∈ C\{i, −i}. Clearly, i and −i are double
(z2 + 1)2
poles of f and both are enclosed by γ . Therefore, by Cauchy’s residue
theorem,

sin(z)
dz = i2π [(Res f )(i) + (Res f )(−i)] .
(z2 + 1)2
γ

Since
 
d sin(z)
(Res f )(i) = lim
z→i dz (z + i)2
(z + i) cos(z) − 2 sin(z)
= lim
z→i (z + i)3
exp(i)
=
−4

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 300 — #14

i i

300 Cauchy’s Residue Theorem

and
 
d sin(z)
(Res f )(−i) = lim
z→−i dz (z − i)2
(z − i) cos(z) − 2 sin(z)
= lim
z→−i (z − i)3
exp(i)
= ,
−4
we have

sin(z) 2 exp(i)
dz = i2π = −iπ exp(i).
(z + 1)
2 2 −4
γ

4z + 1 √ √ √
7. Let , ∀z ∈ C\{i, −i, −i 2, i 2} . Clearly, i, −i, i 2,
+ 1)(z2 + 2)
(z2 √
and −i 2 are simple poles of f , and i and −i are enclosed by |z| = 1.
By Cauchy’s residue theorem, we have

4z + 1
dz = i2π [(Res f )(i) + (Res f )(−i)] .
(z2 + 1)(z2 + 2)
|z|=1

Since
4z + 1
(Res f )(i) = lim
z→i (z + i)(z2 + 2)
1 + i4
=
i2
4−i
=
2
and
4z + 1
(Res f )(−i) = lim
z→−i (z − i)(z2 + 2)
1 − i4
=
−i2
4+i
= ,
2
we get
  
4z + 1 4−i 4+i
dz = i2π + = i8π .
(z2 + 1)(z2 + 2) 2 2
|z|=1

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 301 — #15

i i

Residue Calculus 301

z3 + 1
8. Let f (z) = , ∀z ∈ C\{0, −1, −2}. Clearly, 0, −1, and
z(z + 1)(z + 2)
−2 are simple poles of f and all are enclosed by |z| = 4. Therefore, by
Cauchy’s Residue theorem,

z3 + 1
dz = i2π [(Res f )(0) + (Res f )(−1) + (Res f )(−2)] .
z(z + 1)(z + 2)
|z|=4

Since

z3 + 1 1
(Res f )(0) = lim = ,
z→0 (z + 1)(z + 2) 2
z3 + 1
(Res f )(−1) = lim = 0,
z→−1 z(z + 2)

and
z3 + 1 −7
(Res f )(−2) = lim = ,
z→−2 z(z + 1) 2
we get
  
z3 + 1 1 −7
= i2π +0+ = −3iπ .
z(z + 1)(z + 2) 2 2
|z|=4

exp(z) − 1
9. Let f (z) = , ∀z ∈ C\{nπ : n ∈ Z}. Clearly, 0 is a removable
sin(z)
singularity for f as

∞ zk
exp(z) − 1 k=1 k !
lim f (z) = lim = lim = 0,
z→0 z→0 sin(z) z→0 ∞ (−1)k−1 z2k−1
k=1 (2k − 1)!

and nπ is a pole of order 1, for every 0  n ∈ Z, as

d
lim f (z) = ∞ and sin(z) = cos(nπ) = (−1)n  0.
z→nπ dz nπ

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 302 — #16

i i

302 Cauchy’s Residue Theorem

Among the singularities of f , 0 and π are enclosed by |z − 1| = 4.

Therefore, by Cauchy’s residue theorem, we have

exp(z) − 1
dz = i2π [(Res f )(0) + (Res f )(π )] .
sin(z)
|z−1|=4

(Res f )(0) = 0.
(z − π )(exp(z) − 1)
(Res f )(π ) = lim
z→π sin(z)
= (−1)(exp(π) − 1).

Therefore,

exp(z) − 1
dz = i2π (1 − exp(π )).
sin(z)
|z−1|=4

exp(z)
10. Let f (z) = , ∀z ∈ C\{0}. Clearly, 0 is the simple pole of f and is
sin(z)
enclosed by |z| = 1. Therefore, by Cauchy’s residue theorem, we have

exp(z)
dz = i2π (Res f )(0)
sin(z)
|z|=1
z
= i2π lim
z→0 sin(z)
= i2π .

Note that so far we have evaluated integrals over simple curves. Now, we
evaluate an integral over a closed curve, which is not simple.

 2z + 3
Example 6.2.3 Evaluate dz, where γ is given in the
γ (z4 − 1)(z2 − 5z + 6)
following diagram.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 303 — #17

i i

Residue Calculus 303

−2 −1 −0 1

−i

2z + 3
Let f (z) = , ∀z ∈ C\{−1, 0, i, −i, 1, 2, 3}.
− 1)(z3 − 5z2 + 6z)
(z4
Clearly, 0, 1, −1, i, −i, 2, and 3 are poles of f . From the diagram, we find1
that WN(γ , −1) = −3, WN(γ , 0) = 2, WN(γ , i) = −1, WN(γ , −i) = 0,
WN (γ , 1) = −1, WN (γ , 2) = 0, and WN (γ , 3) = 0.
Therefore, by Cauchy’s residue theorem, we get

2z + 3
dz
(z4 − 1)(z3 − 5z2 + 6z)
γ
= i2π [WN(γ , −1) (Res f )(−1)
+ WN(γ , 0) (Res f )(0) + WN(γ , i) (Res f )(i)
+ WN(γ , −i) (Res f )(−i) + WN(γ , 1) (Res f )(1)
+ WN(γ , 2) (Res f )(2) + WN(γ , 3) (Res f )(3)]
= i2π [−3 (Res f )(−1) + 2 (Res f )(0) − (Res f )(i) − (Res f )(1)] .

Since
2z + 3 1
(Res f )(−1) = lim = ,
z→−1 (z − 1)(z2 + 1)(z3 − 5z2 + 6z) 48
1 WN (γ , a) is found by counting the number of times the curve γ winds around a, and the sign
is given as + (−) if it winds in anti-clockwise sense (clockwise sense).

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 304 — #18

i i

304 Cauchy’s Residue Theorem

2z + 3 −1
(Res f )(0) = lim = ,
z→0 (z4− 1)(z − 5z + 6)
2 2
2z + 3 1 + i5
(Res f )(i) = lim 2 = ,
z→i (z − 1)(z + i)(z − 5z + 6z)
3 2 40

and
2z + 3 5
(Res f )(1) = lim = ,
z→1 (z + 1)(z2 + 1)(z3 − 5z2 + 6z) 8
we get
  
2z + 3 1 −1 1 + i5 5
dz = i2π −3 × +2× − −
(z − 1)(z2 − 5z + 6)
4 48 2 40 8
γ
 
1 −1 1 + i5
= i π −8− −5
4 2  5
π i137
= 1− .
4 10

Example 6.2.4 Evaluate tan(z) dz using Cauchy’s residue theorem.
|z−1|=2

Solution:
We note that the only pole of tan inside |z−1| = 2 is π2 , which is a simple pole.
 π 
Hence, by Cauchy’s residue theorem, tan(z) dz = i2π (Res tan) . As
|z|=2 2
sin(z) π  π 
tan(z) = , sin = 1  0, and cos = 0, then by applying
cos(z) 2 π  2
π  sin 1
Result 6.1.9, we get (Res tan) = 2π  = = −1. Therefore,
2 cos −1
2

tan(z) dz = −i2π .
|z|=2

Exercise 6.2.5 Evaluate the following integrals using Cauchy’s residue

theorem:
 exp(z)
1. dz.
|z|=2 z2 − 2z − 3

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 305 — #19

i i

Residue Calculus 305

−iπ
Answer:
2 exp(1)
 z2 − 1
2. dz.
|z−1|=2 z2 − 6z + 8

Answer: −i3π
 z2 + 3z
3. dz.
|z−1|=3 (z − 1)2 (z − 2)

Answer: i2π
 z−1
4. dz.
|z|=2 (z + 1) (z − 3)
2

−iπ
Answer:
4
 z−1
5. dz.
|z+1|=5 (z + 2)2 (z + 3)2

Answer: 0
 exp(z)
6. dz.
|z|=2 (z + 1) (z − 3)
2

−i5π
Answer:
8 exp(1)
 z+4
7. dz.
|z−1−i|=3 (z − 7)(z − 2)

−i12π
Answer:
5
 sin(z)
8. dz.
|z|=1 z2

Answer: i2π
 z3
9. dz.
|z|=4 z2 − 2z + 2

Answer: i4π

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 306 — #20

i i

306 Argument Principle and Rouche’s Theorem

 exp(z)
10. dz.
|z|=3 z(z + 1 + i)
2

 
Answer: πi e−(1+i) (−1 + 2i) − i

6.3 ARGUMENT PRINCIPLE AND ROUCHE’S

THEOREM
THEOREM 6.3.1 (Argument principle)
Let f be a meromorphic function on a region  with zeroes zj , 1 ≤ j ≤ n and
poles wk , 1 ≤ k ≤ m including multiplicities. Then, for every closed curve
γ ∼ 0 in  and not passing through any zj or wk ,
 n m
1 f  (ζ )
dζ = WN (γ , zj ) − WN (γ , wk ).
i2π f (ζ )
γ j=1 k=1

f
Proof: We know that f  is also analytic wherever f is analytic. Hence, is
f
analytic on  except at zj ’s and wk ’s. By Cauchy’s residue theorem, we have
   
1 f (ζ ) f
dζ = WN (γ , z) Res (z)
i2π f (ζ ) z
f
γ
 
f
+ WN (γ , w) Res (w), (6.1)
w
f

where z varies through the distinct zeroes of f and w varies through the dis-
tinct poles of f .
 
f
Claim. If f has a zero of order pz at z, then Res (z) = pz .
f
Since z is a zero for f of order pz , by Result 5.2.3, there exists an analytic
function g on a neighbourhood B(z, r) of z such that

f (ζ ) = (ζ − z)pz g(ζ ) and g(ζ )  0, ∀ζ ∈ B(z, r).

Hence, by applying Theorem 4.5.12, we get that log(g) is analytic in B(z, r),
and hence,

f  (ζ ) pz (ζ − z)pz −1 g(ζ ) + (ζ − z)pz g (ζ )

=
f (ζ ) (ζ − z)pz g(ζ )

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 307 — #21

i i

Residue Calculus 307

pz g (ζ )
= +
g(ζ ) g(ζ )
pz d
= + [log(g(ζ ))] .
g(ζ ) dζ
f pz
This implies that − is the derivative of an analytic function on B(z, r),
 f ζ − z
f
and hence, Res (z) = pz .
f
 
f
Claim. If f has a pole of order qw at w, then Res (w) = qw .
f
Since w is a pole for f of order qw , by Result 5.2.11, there exists an analytic
function g on a neighbourhood B(w, r) of w such that
f (ζ ) = (ζ − w)−qw g(ζ ) and h(ζ )  0, ∀ζ ∈ B(w, r) \ {w}.
Hence, log(h) is analytic in B(z, r) (by Theorem 4.5.12) and hence,
f  (ζ ) (−qw )(ζ − w)−qw −1 g(ζ ) + (ζ − w)−qw h (ζ )
=
f (ζ ) (ζ − w)−qw h(ζ )
−qw 
h (ζ )
= +
h(ζ ) h(ζ )
−qw d
= + [(log(h(ζ )))] .
h(ζ ) dζ
 
f
This implies that Res (w) = −qw . Using these observations in equation
f
(6.1), we get
     
1 f (ζ ) f f
dζ = WN (γ , z) Res (z) + WN (γ , w) Res (w)
i2π f (ζ ) z
f w
f
γ
(where z varies through distinct zeroes of f and w
varies through distinct poles of f .)
= WN (γ , z)pz + WN (γ , w)qw
z w
n m
= WN (γ , zj ) − WN (γ , wk ).
j=1 k=1

Thus, the theorem follows. 

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 308 — #22

i i

308 Argument Principle and Rouche’s Theorem

THEOREM 6.3.2 Let f be a meromorphic function on a simply connected

region . If every zero of f is of even order and if every pole of f is of even
order, then there exists a meromorphic function g on  such that f = g2 .

Proof: Let z1 , z2 , . . . , zn be the distinct zeroes of f such that the order of the
zero of f at zj is 2pj for some pj ∈ N, ∀j = 1, 2, . . . , , n, and let w1 , w2 , . . . , wm
be the distinct poles of f such that the order of the pole of f at wk is 2qk for
some qk ∈ N, ∀k = 1, 2, . . . , m. By argument principle, for every closed
curve γ in ∗ :=  \ {zj , wk : 1 ≤ j ≤ n; 1 ≤ k ≤ m}, we have
  n m
1 f (ζ )
dζ = 2pj WN(γ , zj ) − 2qj WN(γ , wk ).
2π i f (ζ )
γ j=1 k=1

1
 f  (ζ )
Therefore, 2 f (ζ ) dζ is an integral multiple of 2π i. For an arbitrarily fixed
γ
z0 ∈ ∗ , we define
⎛ ⎞
 
1 f  (ζ ) ⎠
g(z) = exp ⎝ f (z0 ) + dζ , ∀z ∈ ∗ , (6.2)
2 f (ζ )
σz

where σz is a curve in ∗ joining z0 and z, and f (z0 ) is any one value of
square root of f (z0 )  0. The definition of g is well defined. Indeed, if z0 and
z are connected by two curves σz and τz , respectively, in ∗ , then σz ∪ (−τz )
 f  (ζ )
is a closed curve in ∗ . Therefore, 12 f (ζ ) dζ is an integral multiple
σz ∪(−τz )
 
 f  (ζ )
f (ζ ) dζ = 1, which implies that
1
of 2π i, and hence, exp 2
σz ∪(−τz )
⎛ ⎞ ⎛ ⎞
  
1 f  (ζ ) ⎠ 1 f (ζ )
exp ⎝ dζ = exp ⎝ dζ ⎠ .
2 f (ζ ) 2 f (ζ )
σz τz

Thus, the definition of g is independent of the curve σz and hence well de-

fined. Next, we claim that g is analytic on ∗ and g = gf2f on ∗ . Let z ∈ ∗
be arbitrary. If r > 0 is such that B(z, r) ⊂ ∗ and if w ∈ B(z, 3r ), then
z ∈ B(w, 2r ∗
3 ) ⊂ B(z, r) ⊂  . Hence, we can write
⎛ ⎞
 
⎜ 1 
f (ζ ) 1 
f (ζ ) ⎟
g(z) = exp ⎝ f (z0 ) + dζ + dζ ⎠
2 f (ζ ) 2 f (ζ )
σw Lw,z

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 309 — #23

i i

Residue Calculus 309

(where Lw,z
⎛ is the line segment
⎞ joining w and z)
 
⎜1 f (ζ ) ⎟
= g(w) exp ⎝ dζ ⎠ .
2 f (ζ )
Lw,z

f  (z) 2r
Since f (z) is analytic on B(w, 3 ), from the proof of Cauchy’s theorem for
 (z)  f  (ζ )
disc, ff (z) is the derivative of the analytic function f (ζ ) dζ in B(w, 2r
3 ).
Lw,z
Therefore, g is analytic on ∗ and
⎛ ⎞

⎜1 f  (ζ ) 
⎟ 1 f (z) g(z)f  (z)
g (z) = g(w) exp ⎝ dζ ⎠ = , ∀z ∈ ∗ .
2 f (ζ ) 2 f (z) 2f (z)
Lw,z

(g(z0 ))2
Since f (z0 ) = 1 and

 
d (g(z))2 f (z)2g(z)g (z) − (g(z))2 f  (z)
=
dz f (z) (f (z))2
g(z)f  (z)
f (z)2g(z) 2f (z) − (g(z))2 f  (z)
=
(f (z))2
= 0,

g2
f is a constant on ∗ and the constant value is 1. Thus, f = g2 on ∗ . It
is also easy to see that lim g2 (z) = lim f (z) = f (zj ) = 0 and lim g2 (z) =
z→zj z→zj z→wk
lim f (z) = ∞, where g is the required meromorphic function on  such that
z→wk
g2 = f. 

Definition 6.3.3 If f is a meromorphic function on a simply connected region

, then we define the ‘square root’ of f , which is definedby the meromorphic
function g on  such that g2 = f . We denote this g by f .

RESULT 6.3.4 Let f be a meromorphic function on a region  such that

− = . If f 2 is an even function and lim f (z) = c for some c ∈ C \ {0},
z→0
then f must be an even function.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 310 — #24

i i

310 Argument Principle and Rouche’s Theorem

Proof: First, we note that f 2 is analytic at 0 (since lim f 2 (z) = c2 ), and hence,
z→0
we can write
∞
f 2 (z) = an zn , ∀z ∈ B(0, r) for some r > 0. (6.3)
n=0

Since f is even, we have

∞ ∞ ∞
0 = f 2 (z) − f 2 (−z) = an zn − an (−z)n = 2 a2n+1 z2n+1 ,
n=0 n=0 n=0

which implies that a2n+1 = 0, ∀n ∈ {0, 1, 2, 3, . . .}, by uniqueness theorem

∞
for power series. Therefore, we have f 2 (z) = a2n z2n , ∀z ∈ B(0, r). Then,
n=0
 ∞= f (z)
 =
we claim that f must be even. As f is analytic at 2
∞
 0∞, then  f (z)
bn zn , ∀z ∈ B(0, r) with b0 = c  0. Since bn zn bn zn =
n=0 n=0 n=0
∞
f 2 (z) = a2n z2n , equating the coefficient of z on both sides, we get 2b0 b1 =
n=0
0 ⇒ b1 = 0. Next by equating the coefficients of z3 on both sides, we get
2(b0 b3 +b0 b1 ) = 0 ⇒ 2b0 b3 = 0 ⇒ b3 = 0. Furthermore, we can prove that
∞
b2n+1 = 0, ∀n ∈ {0, 1, 2, . . .}. Thus, we have f (z) = b2n z2n , ∀z ∈ B(0, r).
n=0
Therefore, f is even in B(0, r). Then, by analytic continuation, f is an even
function on . 

THEOREM 6.3.5 (Rouche’s theorem)

Let f and g be analytic functions on a region  and γ be a simple closed
curve in  such that γ ∼ 0 in . If

| f (z) − g(z)| < | f (z)|, ∀z ∈ γ , (6.4)

then f and g have same number of zeroes inside γ including multiplicities.

Proof: Let aj , 1 ≤ j ≤ n be the zeores of f and bk , 1 ≤ k ≤ m be the zeores

of g including multiplicities.
Case 1: Assume that f and g have no common zeroes.
g
Let F = , then F is a meromorphic function on  such that F has
f
zeores at the zeores of g, and it has poles at the zeroes of f . We claim

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 311 — #25

i i

Residue Calculus 311

that f and g do not vanish on γ . If f (z) = 0 for some z ∈ γ , then

from equation (6.4), we get |g(z)| < 0. If g(z) = 0 for some z ∈ γ ,
then again from equation (6.4), we get | f (z)| < | f (z)|. As both are
not possible, our claim holds. Now, we can apply argument principle
to F and we get
  n m
1 F (z)
dz = WN (γ , aj ) − WN (γ , bk ). (6.5)
i2π F(z)
γ j=1 k=1

As γ is a simple closed curve, WN(γ , z) = 1 or 0, whenever γ

encloses z or γ does not enclose z. Hence, the right-hand side of
equation (6.5) is simply the number of zeroes of F enclosed by γ
including multiplicities − the number of poles of F enclosed by γ
including multiplicities which is equal to the number of zeroes of f
enclosed by γ including multiplicities − the number of zeroes of g
enclosed by γ including multiplicities.
On the other hand, if F(γ ) = , then using equation (6.4), we have
| f (z) − 1| < 1, ∀z ∈ γ . Thus, ⊂ B(1, 1). By Cauchy’s theorem for
1
simply connected region, since (w) = , ∀w ∈ B(1, 1) is analytic
w
and is a closed curve in B(1, 1), we get
  
1 dw F (z)
0= = dz,
i2π w F(z)
γ

by applying the change of variable w = F(z). Thus, we get that

left-hand side of equation (6.5) is zero. Hence, the theorem follows.

Case 2: Let f and g have the common zeroes α1 , α2 , . . ., αs . Then, by

Result 5.2.3, we can write

f (z) = (z − α1 )(z − α2 ) . . . (z − αs )φ(z),

and
g(z) = (z − α1 )(z − α2 ) . . . (z − αs )ψ(z),
where φ and ψ are analytic functions on  with the following
properties:

1. the number of zeroes of φ including multiplicities = n − s.

2. the number of zeroes of ψ including multiplicities = m − s.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 312 — #26

i i

312 Argument Principle and Rouche’s Theorem

3. φ and ψ have no common zeroes.

g(z) ψ(z)
4. = , ∀z ∈  \ {aj : j = 1, 2, . . . , n}.
f (z) φ(z)
Then, we have

n−m = (n − s) − (m − s)
 
 ψ (z)
φ
=   dz
ψ
γ φ (z)
= 0 (by applying Case(1)).

Thus, the theorem follows. 

Example 6.3.6 Find the number of zeroes of z7 − 4z3 + z − 1 enclosed by

|z| = 1.
Let f (z) = −4z3 and g(z) = z7 − 4z3 + z − 1, ∀z ∈ C. On |z| = 1, then we
have

| f (z) − g(z)| = |z7 + z − 1| ≤ |z|7 + |z| + 1 ≤ 2 < 4 = | − 4z3 | = | f (z)|,

whence we can apply Rouche’s theorem and we get that f and g have same
number of zeroes enclosed by |z| = 1. As three zeroes of f are enclosed by
|z| = 1, including multiplicites, g also has three zeroes enclosed by |z| = 1.
 
3
Example 6.3.7 Prove that one zero of z5 + 15z + 1 belongs to B 0, and
2
four zeroes of z5 + 15z + 1 belong to AN23/2 (0).
Solution: Let f (z) = z5 and g(z) = z5 + 15z + 1, ∀z ∈ C. Then we note that

| f (z) − g(z)| = |15z + 1| ≤ 15|z| + 1 = 31 < 32 = 25 = |z5 | = | f (z)|,

on |z| = 2. Hence, we can apply the Rouche’s theorem (Theorem 6.5) and
we get that f (z) and g(z) have same number of zeroes in B(0, 2) including
multiplicities. We know that f has five zeroes (which are obviously 0, 0, 0, 0,
and 0) in B(0, 2) including multiplicities, whence g has five zeroes in B(0, 2).
As g has exactly five zeroes, every zero of g belongs to B(0, 2). On the other
3
hand, if we put f1 (z) = 15z, then on the circle |z| = , we have
2
243 45
| f1 (z) − g(z)| = |z5 + 1| ≤ |z|5 + 1 = +1< = |15z| = | f1 (z)|,
32 2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 313 — #27

i i

Residue Calculus 313

and hence, again by using Rouche’s  theorem,

 we conclude that f1 and g
3
have same number of zeroes in B 0, . Since f1 has exactly one zero
  2  
3 3
in B 0, , g also has exactly one zero in B 0, . From the proof of
2 2
3
Rouche’s theorem, we have g  0 on |z| = . Hence, the remaining four
2
zeroes of g must lie in AN23/2 (0).

Example 6.3.8 Prove that z6 + 5z4 + 3 has exactly 2 zeroes in AN31 (0).
Solution: Let f (z) = z6 and g(z) = z6 + 5z4 + 3, ∀z ∈ C. As for |z| = 3,

| f (z) − g(z)| = |5z4 + 3| ≤ 5|z|4 + 3 = 408 < 729 = 36 = |z6 | = | f (z)|,

we can apply Rouche’s theorem, and we conclude that f and g have same
number of zeroes enclosed by |z| = 3, including multiplicities. As f has
six zeroes inside B(0, 3), g also has six zeroes inside B(0, 3). If f1 (z) = 5z4 ,
∀z ∈ C, then

| f1 (z) − g(z)| = |z6 + 3| ≤ |z|6 + 3 = 4 < 5 = 5|z4 | = | f1 (z)| on |z| = 1.

Therefore, again using Rouche’s theorem, we get that the number of zeroes
of f1 enclosed by |z| = 1 (including multiplicities) is same as that of g. Since
f1 has four zeroes enclosed by |z| = 1, g has two zeroes enclosed by |z| = 1.
Thus, g has two zeroes inside the annulus AN31 (0).

Exercise 6.3.9

1. Prove that z3 − 9z + 27 has all zeroes in B(0, 4).

2. Prove that one zero of z4 − 7z − 1 lies in B(0, 1) and three zeroes are in
AN21 (0).

3. Find the number of zeroes of z8 − 3z5 + 7z3 − 2 in the annulus AN21 (0).
Answer: 5.

4. Prove that all zeroes of z7 − 5z3 + 12 lie in the annulus AN21 (0).

5. Prove that exactly 3 zeroes of z5 − 10z2 − 3z + 1 lie in AN32 (0).

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 314 — #28

i i

314 Evaluation of Real Integrals

6.4 EVALUATION OF REAL INTEGRALS

Now, we present the formulae to evaluate certain definite real integrals using
Cauchy’s residue theorem.

THEOREM 6.4.1 (Type 1): If S is a quotient of polynomials in two variables,


2π
then S(cos(θ ), sin(θ )) dθ = i2π × sum of the residues of G at the poles
0     
i 1 1 1 1
enclosed by |z| = 1, where G(z) = − S z+ , z− .
z 2 z i2 z

Proof: If z = exp(iθ ), then

 
1 1 1
cos(θ ) = Re z =
(z + z) = z+ ,
2 2 z
 
1 1 1
sin(θ) = Im z = (z − z) = z− ,
i2 i2 z
i
dθ = − dz (as dz = i exp(iθ ) dθ = iz dθ ),
z

and when θ varies along the interval [0, 2π], z varies along the circle |z| = 1.
Therefore, by Cauchy’s residue theorem,

2π        
1 1 1 1 i
S(cos(θ), sin(θ )) dθ = S z+ , z− − dz
2 z i2 z z
0 |z|=1

= G(z) dz
|z|=1
= i2π × sum of the residues of G
at the poles enclosed by |z| = 1.

Hence, the theorem follows. 

THEOREM 6.4.2 (Type 2): If R is a rational function with zero at ∞ of order

∞
at least 2 and R has no poles on real axis, then R(x) dx = i2π × sum of
−∞
the residues of R at the poles in the upper half-plane.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 315 — #29

i i

Residue Calculus 315

Proof: First, we construct a closed semicircle C(r) = γ (r) ∪ L(−r,0),(r,0) with

centre 0 and radius r such that C(r) encloses all poles of R in the upper half-
plane.
Then, by Cauchy’s residue theorem, we get

R(z) dz = i2π × sum of the residues of R
C(r) at the poles enclosed by C(r).
= i2π × sum of the residues of R
at the poles in the upper half-plane.
y

g (r)

−r O r x

From the diagram, we have

  r
R(z) dz = R(z) dz + R(x) dx. (6.6)
C(r) γ (r) −r

We claim that R(z) dz → 0 as r → ∞.
γ (r)
Since R has a zero at ∞ of order at least 2, lim z2 R(z) is finite and hence
z→∞
there exist M > 0 and s > 0 such that |z2 R(z)| ≤ M , ∀|z| > s. Therefore,
using Theorem 4.1.27, for r > s,
  
M M Mπ
R(z) dz ≤ |R(z)| |dz| ≤ |dz| = πr = → 0 as r → ∞.
r2 r2 r
γ (r) γ (r) γ (r)

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 316 — #30

i i

316 Evaluation of Real Integrals


Hence, allowing r → ∞ in equation (6.6), we get lim R(z) dz =
r→∞
C(r)
∞ ∞ 
R(x) dx. Thus, R(x) dx = lim R(z) dz = i2π × sum of the residues
−∞ −∞ r→∞
C(r)
of R at the poles in the upper half-plane. 

THEOREM 6.4.3 (Type 3(a)): If R is a rational function with zero at ∞ of

∞
order at least 2 and R has no poles on real axis, then R(x) exp(ix) dx =
−∞
i2π× sum of the residues of H at the poles in the upper half-plane, where
H(z) = R(z) exp(iz), ∀z ∈ C.

Proof: First, we construct a closed semicircle C(r) = γ (r) ∪ L(−r,0),(r,0) with

centre 0 and radius r such that C(r) encloses all poles of R in the upper half-
plane.
y

g (r)

−r O r x

Then by Cauchy’s residue theorem, we get


R(z) exp(iz) dz = i2π × sum of the residues of H
C(r) at the poles enclosed by C(r).
= i2π × sum of the residues of H
at the poles in the upper half-plane.
Now, we have
  r
R(z) exp(iz) dz = R(z) exp(iz) dz + R(x) exp(ix) dx. (6.7)
C(r) γ (r) −r

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 317 — #31

i i

Residue Calculus 317


We claim that R(z) exp(iz) dz → 0 as r → ∞.
γ (r)
Since R has a zero at ∞ of order at least 2, there exist M > 0 and s > 0 such
that |z2 R(z)| ≤ M , ∀|z| > s. Using the fact that

Im z > 0 on γ (r) ⇒ exp(−Im z) < 1 on γ (r),

for r > s, we get

  M 
R(z) exp(iz) dz ≤ |R(z)| exp(−Im z) |dz| ≤ |dz|
γ (r) γ (r) r2 γ (r)
(using Theorem 4.1.27)
M Mπ
= 2
πr = → 0 as r → ∞.
r r
Hence, allowing r → ∞ in equation (6.6), we get
 ∞
lim R(z) dz = R(x) exp(ix) dx.
r→∞
C(r) −∞

∞ 
Thus, R(x) exp(ix) dx = lim R(z) exp(iz) dz = i2π × sum of the
−∞ r→∞
C(r)
residues of H at the poles in the upper half-plane. 

THEOREM 6.4.4 (Type 3(b)): If R is a rational function with zero at ∞ of

∞
order 1 and R has no poles on real axis, then R(x) exp(ix) dx = i2π ×
−∞
sum of the residues of H at the poles in the upper half-plane, where H(z) =
R(z) exp(iz), ∀z ∈ C.

Proof: First, we construct a rectangle γ1 ∪ γ2 ∪ γ3 ∪ γ4 with vertices X2 ,

X2 + iY , X2 − iY , and −X2 , where X1 , X2 , and Y are positive real numbers
such that the rectangle encloses all poles of R in the upper half-plane.
−X1 + iY X2 + iY
g3
g4 g2

g1

−X1 X2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 318 — #32

i i

318 Evaluation of Real Integrals

Then by Cauchy’s residue theorem, we get


R(z) exp(iz) dz = i2π × sum of the residues of H
4 4
∪ γj at the poles enclosed by ∪ γj .
j=1 j=1
= i2π × sum of the residues of H at the
poles in the upper half equation plane. (6.8)

We claim the following:


1. R(z) exp(iz) dz → 0 as X2 → ∞, ∀Y > 0, uniformly,
γ2

2. R(z) exp(iz) dz → 0 as X1 → ∞, ∀Y > 0, uniformly,
γ4

3. R(z) exp(iz) dz → 0 as Y → ∞, for each fixed X1 > 0 and fixed
γ3
X2 > 0.

Since R has a simple zero at ∞, there exist M > 0 and K > 0 such that
|zR(z)| ≤ M , ∀|z| > K . Therefore, for X2 > K we get

 Y
R(z) exp(iz) dz = R(z) exp(iz) dy where z = X2 + iy
γ2 0
Y
≤ |R(z)| exp(−y) dy
0
Y
1
≤ M exp(−y) dy
|z|
0
Y
M
≤ exp(−y) dy (since |z| ≥ X2 )
X2
0
M
= (− exp(−Y ) + 1)
X2
M
< → 0 as X2 → ∞.
X2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 319 — #33

i i

Residue Calculus 319

The second claim can also be proved by a similar argument.

For Y > K,

 −X1

R(z) exp(iz) dz = R(z) exp(iz) dx where z = x + iY
γ3 X2

X2
≤ |R(z)| exp(−Y ) dx
−X1
X2
1
≤ M exp(−Y ) dx
|z|
−X1
X2
M
≤ exp(−Y ) dx (as |z| ≥ Y )
Y
−X1
M
= (X2 + X1 ) → 0 as Y → ∞
Y
for each fixed X1 > 0 and X2 > 0.

Hence, applying lim , lim , and lim in equation (6.8) and using claims
X1 →∞ X2 →∞ Y →∞
(1), (2), and (3), the theorem follows. 

THEOREM 6.4.5 (Type 3(c)): If R is a rational function with sim-

ple zero at ∞, a simple pole at 0, and no other real poles, then
∞
P.V. R(x) exp(ix) dx = i2π×sum of the residues of H at the poles in
−∞
1
the upper half-plane + (Res H)(0), where H(z) = R(z) exp(iz), ∀z ∈ C,
2
∞
where P.V. R(x) exp(ix) dx is called the principal value of the integral and
−∞  

−δ ∞
is defined by lim + R(x) exp(ix) dx.
δ→0 −∞ δ

Proof: First, we construct a closed curve γ1 ∪ γ2 ∪ γ3 ∪ γ4 ∪ γ5 ∪ γ6 as in the

following diagram such that the closed curve encloses all poles of R in the
upper half-plane, where γ6 is the upper semicircle with centre 0 and radius δ ,
for some δ ≥ 0.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 320 — #34

i i

320 Evaluation of Real Integrals

−X1 + iY X2 + iY

g3

g4 g2
g6
g5 g1

−d 0 d
−X1 X2

Then by Cauchy’s residue theorem, we get


R(z) exp(iz) dz = i2π × sum of the residues of H
6
∪ γj
j=1
6
at the poles enclosed by ∪ γj .
j=1
= i2π × sum of the residues of H
at the poles in the upper half-plane. (6.9)
We claim the following:

1. R(z) exp(iz) dz → 0 as X2 → ∞, ∀Y > 0, uniformly,
γ2

2. R(z) exp(iz) dz → 0 as X1 → ∞, ∀Y > 0, uniformly,
γ4

3. R(z) exp(iz) dz → 0 as Y → ∞, for each fixed X1 > 0 and fixed
γ3
X2 > 0,

4. R(z) exp(iz) dz → −iπ(Res H)(0) as δ → 0.
γ6
As the first three claims are same as in the previous theorem, we prove the
last claim only. Next, let β = (Res H)(0). Then, there exists δ0 > 0 such that
β
R(z) exp(iz) = + S(z), 0 < |z| < δ0 ,
z
where S is analytic on |z| < δ0 . Since R(z) exp(iz) has a simple pole at ‘0’, the
β β
singular part of R(z) exp(iz) around 0 is and hence R(z) exp(iz) − = S(z)
z z
has 0 as a removable singularity. Now,
   
β
R(z) exp(iz) dz = + S(z) dz
z
γ6 γ6

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 321 — #35

i i

Residue Calculus 321

 
β
= dz + S(z) dz
z
γ6 γ6
0 
β
= δi exp(it) dt + S(z) dz
δ exp(it)
π γ6

= −iβπ + S(z) dz.
γ6

If 0 < δ < δ0 , then |S(z)| ≤ M for all z with |z| ≤ δ , for some M > 0.
Therefore, using Theorem 4.1.27, we get
  
S(z) dz ≤ |S(z)| |dz| ≤ M |dz| = Mπ δ → 0 as δ → 0.
γ6 γ6 γ6

Hence, applying lim , lim , lim , and lim in equation (6.9) and using
δ→0 X1 →∞ X2 →∞  Y →∞ 

−δ ∞
claims (1), (2), (3), and (4), we get lim + R(x) exp(ix) dx − iβπ =
δ→0 −∞ δ
i2π× sum of the residues of H at poles in the upper half-plane. Since
⎛ −δ ∞⎞
  ∞
lim ⎝ + ⎠ R(x) exp(ix) dx = P.V. R(x) exp(ix) dx,
δ→0
−∞ δ −∞

the theorem follows. 

π 1
Example 6.4.6 Find dθ .
0 5 + 3 cos(θ )
π 1
Solution: Let us first calculate dθ .
−π 5 + 3 cos(θ )
 
1 1
Using the change of variable z = exp(iθ), we get cos(θ ) = z+ and
2 z
dz
dθ = . Hence,
iz
π 
1 1 dz
dθ =  
5 + 3 cos(θ ) 3 1 iz
−π |z|=1 5+ z+
2 z

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 322 — #36

i i

322 Evaluation of Real Integrals


1
= −i2 dz.
3z2 + 10z + 3
|z|=1

1 1 1
The simple poles of G(z) = are −3 and − . Since − is
+ 10z + 33z2 3 3
enclosed by |z| = 1, by Cauchy’s residue theorem, we get
 
1 i2 1
−i2 dz = −   dz
3z + 10z + 3
2 3 1
|z|=1 |z|=1 (z + 3) z +
3
 
1
= −i2i2π (Res G) −
3
 
1
= 4π lim z + G(z)
1 3
z→−
3
4π 1
= lim
3 1 z+3
z→−
3
4π 3 π
= × = .
3 8 2
1
Since is an even function,
5 + 3 cos(θ )

π π
1 1 1 1 π π
dθ = dθ = × = .
5 + 3 cos(θ ) 2 5 + 3 cos(θ ) 2 2 4
0 −π


2π dθ
Example 6.4.7 Evaluate .
0 2 + cos(θ )
1
Solution: Let S(cos(θ ), sin(θ )) = . This integral is of Type 1.
2 + cos(θ )
Hence,

2π
dθ
= i2π × sum of the residues of G
2 + cos(θ )
0
at the poles enclosed by |z| = 1,

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 323 — #37

i i

Residue Calculus 323

where
    
i 1 1 1 1
G(z) = − S z+ , z−
z 2 z i2 z
i 1
= − ×  
z 1 1
2+ z+
2 z
−i2
= .
z2 + 4z + 1
√
The poles of√G are −2 ± 3, which are the roots of z2 + 4z + 1 = 0. We note
that −2 + 3 lies inside |z| = 1 and the other root lies out side the circle.
Hence,
2π √
dθ
= i2π × (Res G)(−2 + 3)
2 + cos(θ )
0
√
= i2π lim √ (z + 2 − 3)G(z)
z→−2+ 3
−i2
= i2π lim √ √
z→−2+ 3 z+2+ 3
2π
= √ .
3

∞ cos(3x)
Example 6.4.8 Evaluate dx.
−∞ (x + 1)
2 2
∞ exp(i3x)
Solution: First, we find dx. As this integral is of Type 3(a), we
−∞ (x + 1)
2 2
have
∞
cos(3x)
dx = i2π (Res H)(i),
(x2 + 1)2
−∞
exp(i3z)
where H(z) = , ∀z ∈ C. Now,
(z2 + 1)2
 
d 2 exp(i3z)
(Res H)(i) = lim (z − i) 2
z→i dz (z + 1)2
 
d exp(i3z)
= lim
z→i dz (z + i)2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 324 — #38

i i

324 Evaluation of Real Integrals

 
d exp(i3z)
= lim
z→i dz (z + i)2
exp(−3)(−8)
= = −i exp(−3),
(−i8)
 
d exp(i3z) i3 exp(i3z)(z + i)2 − 2(z + i) exp(i3z)
as = =
dz (z + i)2 (z + i)4
exp(i3z)(i3z − 5)
.
(z + i)4
∞ exp(i3x)
Therefore, dx = i2π × (−i) exp(−3) = 2π exp(−3). Equating
−∞ (x + 1)
2 2
∞ cos(3x)
the real parts on both sides, we get dx = 2π exp(−3).
−∞ (x + 1)
2 2

∞ sin(x)
Example 6.4.9 Evaluate dx.
−∞ x
∞ exp(ix)
Solution: First, we evaluate dx. This integral comes under Type
−∞ x
3(c). Hence,
∞
exp(ix)
dx = i2π × sum of the residues of H at the poles
x
−∞
1
in the upper half-plane + (Res H)(0),
2
exp(iz)
where H(z) = , ∀z ∈ C. We note that H has only one simple pole
z
exp(iz)
at 0. Hence, (Res H)(0) = lim z = lim exp(iz) = 1. Therefore,
z→0 z z→0
∞ exp(ix) ∞ sin(x) ∞ exp(ix)
dx = iπ ⇒ dx = Im dx = π .
−∞ x −∞ x −∞ x

Exercise 6.4.10 Evaluate the following integrals:


2π dθ
1. .
0 3 − 2 cos(θ ) + sin(θ)

Answer: π

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 325 — #39

i i

Residue Calculus 325


2π cos(2θ ) dθ
2. .
0 5 + 4 sin(θ)

Answer: − π6
π dθ
3. .
0 4 + cos2 (θ )
π
Answer: √
5

2π dθ
4. .
0 (2 + cos(θ ))2
4π
Answer: √
3 3

2π cos(3θ )dθ
5. .
0 5 − cos(2θ)

Answer: 0
∞ dx
6. .
−∞ x2 + 2x + 6
π
Answer: √
2 5
∞ x2
7. dx.
−∞ (x2 + 4)3
π
Answer:
32
∞ dx
8. .
−∞ x4 + 1
π
Answer: √
2
∞ x sin(x)
9. dx.
0 x2 + 9
π
Answer: exp(−3)
2

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 326 — #40

i i

326 Integrals of Multivalued Functions

∞ x exp(i2x)
10. dx.
−∞ 1 + x
2

Answer: iπ exp(−2)

∞ cos(x)
11. dx.
0 (1 + x )
2 3

7π
Answer:
16 exp(1)

∞ sin2 (x)
12. dx.
−∞ (x + 4)
2

π
Answer: (1 − exp(−4))
4

6.5 INTEGRALS OF MULTIVALUED FUNCTIONS

THEOREM 6.5.1 (Type (4)): If 0 < α < 1, R is a rational function with a

∞ α
zero at infinity of order at least 2 and has no real poles, then x R(x) dx =
−∞
i4π
× sum of the residues of  at poles in the upper half-plane,
1 − exp(i2π α)
where (z) = z2α+1 R(z2 ), with a suitable branch in C.

∞
Proof: First, we evaluate x2α+1 R(x2 ) dx. Using the branch of log defined
−∞
on 2 = C\{(0,
 y) : y ≤ 0} by the unique value of log whose imaginary
π 3π
part lies in − , , if z2α is defined on 2 by exp(2α log(z)), then z2α is
2 2
analytic on 2 . Consider the closed curve

γ = L(−ρ,0),(−δ,0) ∪ γ1 ∪ L(δ,0),(ρ,0) ∪ γ2

as shown in the following diagram.

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 327 — #41

i i

Residue Calculus 327

g2

g1

−r −d O d r x

We note that if Q(z) = R(z2 ), ∀z ∈ C, then Q is a rational function with zero

of order at least 4 at ∞ and Q has no real poles. Then, Q is bounded on B(0, s)
for some s > 0. By Cauchy’s residue theorem, if 0 < δ < ρ , then we have
⎛ −δ ⎞
  ρ 
⎝ + + + z2α+1 R(z) dz⎠ = i2π × sum of the residues of  at
−ρ γ1 δ γ2 poles in the upper half-plane. (6.10)

where (z) = z2α+1 R(z2 ), ∀z ∈ C\{(0, y) : y ≤ 0}. If we choose ρ > 0 and

δ > 0 such that γ encloses all poles of  in the upper half-plane. We claim
the following:

1. z2α+1 R(z2 ) dz → 0 as δ → 0,
γ1

2. z2α+1 R(z2 ) dz → 0 as ρ → ∞.
γ2

If h(z) = z2α+1 Q(z), ∀z ∈ C\{(0, y) : y ≤ 0}, then h is bounded on A =

{r exp(iθ ) : 0 < r < s, 0 < θ < π } because Q is bounded on B(0, s) and
|z2α+1 | ≤ s2α+1 , for all z ∈ A. Therefore, there exists M1 > 0 such that

|z2α+1 R(z2 )| ≤ M1 , ∀z ∈ A.

If 0 < δ < s, then

 
z2α+1 R(z) dz ≤ |z2α+1 R(z2 )| |dz|
γ1 γ1

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 328 — #42

i i

328 Integrals of Multivalued Functions


≤ δM1 |dz|
γ1

≤ M1 π δ 2 → 0 as δ → 0.

As Q has zero at ∞ of order at least 4 at ∞, there exist M2 > 0 and K > 0

such that |z4 R(z2 )| ≤ M2 whenever |z| ≥ K . Therefore, for ρ > K , we have

 
z2α+1 R(z) dz ≤ |z2α+1 ||R(z2 )| |dz|
γ2 γ1

|dz|
≤ ρ 2α+1
|z4 R(z2 )|
|z4 |
γ2

ρ 2α+1
≤ M2 |dz|
ρ4
γ2

ρ 2α+1
= M2 πρ
ρ4
= M2 πρ 2α−2 → 0 as ρ → ∞,

as 2α − 2 < 0.
Allowing ρ → ∞ and δ → 0 on both sides of equation (6.10), we get
∞ 2α+1 2
z R(z ) dz = i2π × sum of residues of  at poles in the upper half-
−∞
plane. Now,

∞ 0 ∞
z 2α+1 2
R(z ) dz = z 2α+1
R(z ) dz +
2
z2α+1 R(z2 ) dz
−∞ −∞ 0
0 ∞
= (−t) 2α+1
R(t ) (−dt) +
2
z2α+1 R(z2 ) dz
∞ 0
  ∞
= 1 + (−1)2α+1 t2α+1 R(t2 ) dt.
0

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 329 — #43

i i

Residue Calculus 329

Since (−1)2α = − exp(i2απ ), according to this branch of z2α , we have

∞
xα R(x) dx
0
∞
=2 t2α+1 R(t2 ) dt (using the change of variable x = t2 )
0

∞
1
=2× z2α+1 R(z2 ) dz
1 − exp(i2απ )
−∞

i4π
= sum of residues of 
1 − exp(i2απ )
at poles in the upper half-plane.

Thus, the theorem follows. 

π
THEOREM 6.5.2 (Type (5)): log(sin(x)) dx = −π log(2).
0

π
Proof: First, we show that log(1−exp(i2z)) dz = 0. We know that principal
0
branch of log which is defined on 1 = C\{(x, 0) : x ≤ 0} by the unique value
of log whose imaginary part lies in (−π , π), so that log is analytic on 1 . If
z = x + iy, then
1 − exp(i2z) is real and ≤ 0

⇔ Re (1 − exp(i2z)) ≤ 0 and Im (1 − exp(i2z)) = 0.

⇔ 1 − exp(−2y) cos(2x) ≤ 0 and exp(−2y) sin(2x) = 0.
⇔ exp(−2y) cos(2x) ≥ 1 and sin(2x) = 0.
⇔ y ≤ 0, 2x = 2kπ , k ∈ Z and 2x = nπ , n ∈ Z.
⇔ y ≤ 0, x = kπ , k ∈ Z.

Therefore, log(sin(z)) is defined for all z ∈ C\{(kπ , y) : k ∈ Z, y ≤ 0}. If

γ = L(δ,0),(π−δ,0) ∪ γ1 ∪ γ2 ∪ γ3 ∪ γ4 ∪ γ5

is as shown in the following diagram,

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 330 — #44

i i

330 Integrals of Multivalued Functions

g3
iY p + iY

g4 g2

i g5 g1 p + id

0 p
d p−d

then by Cauchy’s theorem, we have

⎛ π −δ ⎞
     
⎝ + + + + + ⎠ log(1 − exp(i2z)) dz = 0. (6.11)
δ γ1 γ2 γ3 γ4 γ5

We claim the following:


1. log(1 − exp(i2z)) dz → 0 as δ → 0,
γ5

2. log(1 − exp(i2z)) dz → 0 as δ → 0,
γ1

3. log(1 − exp(i2z)) dz → 0 as Y → ∞,
γ3
 
4. log(1 − exp(i2z)) dz = − log(1 − exp(i2z)) dz.
γ2 γ4

(i2z)n ∞

1 − exp(i2z) ∞ (i2z)zn−1
k=1 n!
1. We know that =− =− → −i2
z z k=1 n!
as z → 0. Since modulus function is continuous on C and log is con-
1 − exp(i2z)
tinuous at 2, we get log → log 2 as z → 0. Therefore,
z
there exists M > 0 such that
1 − exp(i2z)
log ≤ M, ∀z ∈ B(0, r) for some r > 0.
z

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 331 — #45

i i

Residue Calculus 331

For 0 < δ < r, we have


log (1 − exp(i2z)) dz
γ5

≤ |log (1 − exp(i2z))| |dz|
γ5
  
1 − exp(i2z)
≤ log + log(z) |dz|
z
γ5
    
1 − exp(i2z)
≤ log + |log(z)| + |arg (1 − exp(i2z))| |dz|
z
γ5

≤ (M + log(δ) + π) |dz|
γ5
δ
≤ (M + log(δ) + π )π → 0 as δ → 0.
2
1
log(δ) δ
Since lim δ log(δ) = lim = lim = 0, by L’Hospital’s rule.
δ→0 δ→0 1 δ→0 −1
δ δ2

2. Arguing as before, we can prove that log(1 − exp(i2z)) dz → 0 as
γ1
δ → 0.
3. We know that | exp(i2(x + iY ))| = exp(−Y ) → 0 as Y → ∞ uni-
formly on x ∈ R. Therefore, log(1 − exp(x + iY )) dx → 0 as Y → ∞.
Therefore, using Result 4.1.5, we get
 0
log(1 − exp(i2z)) dz = log(1 − exp(i2z)) dz → 0 as Y → ∞.
γ3 π

4. By a direct computation, we get

 δ
log(1 − exp(i2z)) dz = log(1 − exp(−2y)) dz
γ4 Y
Y
= − log(1 − exp(−2y)) dz
δ

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 332 — #46

i i

332 Integrals of Multivalued Functions

Y
= − log(1 − exp(i2(π + iy))) dz
δ
= − log(1 − exp(i2z)) dz.
γ2

Allowing Y → ∞ and δ → 0 on both sides of equation (6.11), we get

π
log(1 − exp(i2z)) dz = 0.
0

Since

1 − exp(i2z) = exp(iz) (exp(−iz) − exp(iz)) = −i2 exp(iz) sin(z),

we have
π π
log(sin(x)) dx = log(1 − exp(i2x)) dx
0 0
π
− [log(2) + log(−i) + log(exp(ix))] dx
0
 
π π2
= 0 − π log(2) − i π + i
2 2
= −π log(2).

Hence, the theorem follows. 

∞ dx
Example 6.5.3 Evaluate √ .
0 x(x + 1)

1
Solution: Let f (z) = √ . Then, z = −1 is a pole for f . Note that if
√ z(z + 1)
g(z) = z, then g is not a single-valued function. Hence, first we define a
branch of g as follows so that it is analytic at −1.  
√ θ
We take  = C \ {(x, 0) : x ≥ 0} and define z = r exp i , 1/2
2
where θ ∈ (0, 2π). As there is a unique argument of z in this range (0, 2π),

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 333 — #47

i i

Residue Calculus 333

it is a well-defined function. Using the continuity of this√branch of argument

function (See Lemma 3.3.8), we get that this branch
 of z is analytic.
To evaluate the integral, first we consider f (z) dz, where γ is the closed
γ
curve as in the following diagram.

CR

−1 C D

B A
Cr

Although it seems that there is a gap between the line segments AB and CD,
actually they are lying on the positive real axis. However, they are given dif-
ferent notation because their parametric equations are given in different ways
as AB = {t exp(i2π) : R ≤ t ≤ r} and CD = {t exp(i0) : r ≤ t ≤ R}. √
(The difference between AB and CD is that the image of AB under z is on
the negative real axis, whereas the same for CD is on the positive real axis.)
Since −1 is the only pole of f inside γ , by Cauchy’s residue theorem, we
have 
dz
√ = i2π (Res f )(−1) = i2π (−i) = 2π ,
z(z + 1)
γ
as  π
dz
(Res f )(−1) = lim √ = (−1)1/2 = exp −i = −i.
z→−1 z 2
Since
    
1 dz
2π = √ = + + + √ , (6.12)
z(z + 1) z(z + 1)
γ CR AB Cr CD

we find
 r r
dz exp(i2π) dt dt
√ = =−
z(z + 1) t1/2 exp(iπ)(t exp(i2π ) + 1) t1/2 (t + 1)
AB R R

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 334 — #48

i i

334 Integrals of Multivalued Functions

and
 R R
dz exp(i0) dt dt
√ = = .
z(z + 1) t1/2 exp(i0)(t exp(i0) + 1) t1/2 (t + 1)
CD r r

We also note that

 2π
dz Ri exp(iθ ) dθ
√ =  θ
z(z + 1) R1/2 exp i 2 (R exp(iθ ) + 1)
CR 0

2π
R dθ
≤
R1/2 |R exp(iθ ) + 1|
0
2π
−1/2 dθ
≤ R
| exp(iθ ) + R1 |
0
1
≤ R−1/2 2π → 0 as R → ∞
1− 1
R

(In the above inequality, we have used | exp(iθ ) + R1 | ≥ || exp(iθ )| − | R1 || =

1 − R1 because 1 > R1 for sufficiently large R.)
and

 0
dz ri exp(iθ ) dθ
√ =  θ
z(z + 1) r1/2 exp i 2 (r exp(iθ ) + 1)
Cr 2π

2π
dθ
≤ r 1/2
1−r
0
1
= r 1/2
2π → 0 as r → ∞.
1−r
Using these observations in equation (6.12), we get

R R 
dt dt dz
2 = 2 + lim √
t (t + 1)
1/2 t (t + 1) r→0
1/2 z(z + 1)
r r Cr

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 335 — #49

i i

Residue Calculus 335


dz
+ lim √ = 2π .
R→∞ z(z + 1)
CR

R dt
Hence, = π.
r t1/2 (t+ 1)

∞ exp(−t2 )
LEMMA 6.5.4 For any c ∈ C with Re c > 0, dt < ∞.
−∞ 1 + exp(−2ct)

Proof: Let φ(t) = 1 + exp(−2ct), t ∈ R. As φ is continuous on R and

φ(0) = 2 for any  ∈ (0, 2), there exists δ > 0 such that

|t| < δ ⇒ |1 + exp(−2ct) − 2| <  ⇒ |1 + exp(−2ct)| > 2 − .

If t > δ , then |1 + exp(−2ct)| ≥ 1 − | exp(−2ct)| = 1 − exp(−2t Re c) >

1−exp(−2δ Re c), and if t < −δ , then |1+exp(−2ct)| ≥ exp(−2t Re c)−1 >
exp(2δ Re c) − 1.
So, if μ = min{2 − , 1 − exp(−2δ Re c), exp(2δ Re c)} − 1, then |1 +
exp(−2ct)| ≥ μ, ∀t ∈ R. Therefore,
∞ exp(−t2 )
dt
−∞ 1 + exp(−2ct)

∞
exp(−t2 )
≤ dt
|1 + exp(−2ct)|
−∞
∞
1
≤ exp(−t2 ) dt
μ
−∞
∞
2
= exp(−t2 ) dt
μ
0
⎛ ⎞
1 ∞
2⎝
= exp(−t2 ) dt + exp(−t2 ) dt⎠
μ
⎛0 ⎞
1
∞
2⎝
< 1 + exp(−t) dt⎠
μ
1
(as t < 1 ⇒ exp(−t2 ) < 1 and t > 1 ⇒ exp(−t2 ) < exp(−t))

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 336 — #50

i i

336 Integrals of Multivalued Functions

 
2 1
= 1+ < ∞.
μ e 
∞ √
PROBLEM 6.5.5 Prove that exp(−t2 ) dt = π.
−∞


√ π exp(−z2 )
Solution: Let c = iπ = (1 + i). If f (z) = , then f has
2 1 + exp(−2cz)
poles at every w ∈ {w ∈ C : exp(−2cw)+1 = 0}. Then, 2cw = i(2k −1)π ⇒
i(2k − 1) i(2k − 1)
w= for some k ∈ Z. Moreover, each is a simple pole
2c 2c
d
of f because it is a simple zero of 1 + exp(−2cz) as (1 + exp(−2cz))  0.
dz 
Now, we consider the following rectangle R with vertices −ρ, r, r + i π2 ,
 
and −ρ + i π2 , which passes through the point c = π
2 (1 + i), where

r > 1 + π2 and ρ > 0 (later, both of these positive real numbers r and ρ
will be tending to +∞).

−r + i p r+i p
2 c 2

c
2
−r r
0
  
π
Since Im z ∈ 0, , ∀z ∈ Int R, the only pole of f in Int R is 2c . In other
2
words, 2c is the only pole of f enclosed by the boundary ∂R of R. Therefore,
by Cauchy’s residue theorem, we have
 c
exp(−z2 ) √
dz = i2π(Res f ) = π, (6.13)
1 + exp(−2cz) 2
∂R

and by applying Result 6.1.9, we have

 2
c  π
c exp − exp −i
4 4
(Res f ) =  = √
2 (−2c) exp −c2 (− 2π (1 + i)) exp (−iπ )

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 337 — #51

i i

Residue Calculus 337

1
√ (1 − i)
2 −i
= √ = √ .
2π(1 + i) 2 π
By direct calculation, we also get
√π
 r  2
f (z) dz = f (t) dt + i f (r + it) dt
∂R −ρ 0
−ρ    0
π
+ f t+i dt + i f (−ρ + it) dt. (6.14)
2 √
r π
2

We note that
√π √π
 2  2
| exp(−(r2 − t2 + i2rt))|
f (r + it) dt ≤ dt
|1 + exp(−2a(r + it))|
0 0
√π
 2
exp(−(r2 − t2 ))
≤ √ dt
1 − exp(− 2π (r − t))
0
[ as |1 − exp(−2a(r + it))| ≥ 1 − | exp(−2a(r + it))|
√
= 1 − Re exp(−2a(r + it)) = 1 − exp(− 2π (r − t)) ]
√π
 2
exp(t2 )
≤ exp(−r2 ) √ dt
1 − exp(− 2π )
0
 
π π
[ as r > 1 + and t ≤ ⇒r−t >1
2 2
1 1
⇒ √ ≤ √ ]
1 − exp(− 2π(r − t)) 1 − exp(− 2π )
 √
π   π2
exp −r2 +
≤ √2 dt
1 − exp(− 2π)
0
 
 exp −r2 + π
π
≤ √ 2 → 0 as r → +∞.
2 1 − exp(− 2π )

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 338 — #52

i i

338 Integrals of Multivalued Functions

Similarly, we can show that

0
f (−ρ + it) dt → 0 as ρ → +∞.
√π
2

Therefore, applying lim and lim on both sides of equation (6.14) and
r→+∞ ρ→+∞
using equation (6.13), we get
 ∞    
√ π
π = lim lim f (z) dz = f (t) − f t+i dt. (6.15)
r→+∞ ρ→+∞ 2
∂R −∞

Applying the following observations in equation (6.15)

∞    ∞       
π π π π
f t+i dt = f s+ +i ds using s = t +
2 2 2 2
−∞ −∞

∞
= f (s + c) ds
−∞

and
exp(−(t + c)2 )
f (t + c) =
1 + exp(−2c(t + c))
exp(−(t2 + c2 + 2ct)
=
1 + exp(−2ct − 2c2 )
− exp(−t2 ) exp(−2ct)
=
1 + exp(−2ct)
(as c2 = π i, exp(−iπ ) = −1, exp(−i2π ) = 1.)
− exp(−t2 )(1 + exp(−2ct) exp(−t2 )
= +
1 + exp(−2ct) 1 + exp(−2ct)
= − exp(−t2 ) + f (t),
we obtain
∞ ∞
√
π = f (t) dt − (− exp(−t2 ) + f (t)) dt
−∞ −∞

i i

i i
 i i

“book” — 2014/6/4 — 21:16 — page 339 — #53

i i

Residue Calculus 339

∞ ∞ ∞
= f (t) dt + exp(−t ) dt −
2
f (t) dt (using Lemma 6.5.4)
−∞ −∞ −∞
∞
= exp(−t2 ) dt.
−∞

The integral in the previous problem is called the error integral.

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 21:22 — page 341 — #1

i i

7
Some Interesting
Theorems

7.1 MEAN VALUE PROPERTY OF HARMONIC

FUNCTIONS
In Chapter 2, we have already introduced the definition of harmonic func-
tions, and in this section, we shall see some properties of harmonic functions.

RESULT 7.1.1 (Laplace equation in polar form)  

∂ ∂u ∂ 2u
If u is a harmonic function given in polar form, then r r + 2 = 0.
∂r ∂r ∂θ

Proof: We know that the Cartesian coordinates and polar coordinates are
related by x = r cos(θ ) and y = r sin(θ ). As
∂u ∂u ∂x ∂u ∂y ∂u ∂u
= + = cos(θ ) + sin(θ ),
∂r ∂x ∂r ∂y ∂r ∂x ∂y
we get
∂u ∂u ∂u ∂u ∂u
r = r cos(θ ) + r sin(θ ) = x +y ,
∂r ∂x ∂y ∂x ∂y
which implies
   
∂ ∂ ∂ ∂u ∂u
r r = r x +y
∂r ∂r ∂r ∂x ∂y
   
∂ ∂u ∂u ∂ ∂u ∂u
= x x +y +y x +y
∂x ∂x ∂y ∂y ∂x ∂y

341

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 342 — #2

i i

342 Mean Value Property of Harmonic Functions

   
∂u ∂ 2u ∂ 2u ∂ 2u ∂u ∂ 2u
= x +x 2 +y +y x + +y 2
∂x ∂x ∂x∂y ∂y∂x ∂y ∂y
∂ 2u ∂ 2u 2∂ u
2 ∂u ∂u
= x2 + 2xy + y +x +y .
∂x 2 ∂x∂y ∂y 2 ∂x ∂y
Similarly, using
∂u ∂u ∂x ∂u ∂y ∂u ∂u ∂u ∂u
= + = − r sin(θ ) + r cos(θ ) = x − y ,
∂θ ∂x ∂θ ∂y ∂θ ∂x ∂y ∂y ∂x
we get
 
∂ 2u ∂ ∂u ∂u
= x −y
∂θ 2 ∂θ ∂y ∂x
   
∂ ∂u ∂u ∂ ∂u ∂u
= x x −y −y x −y
∂y ∂y ∂x ∂x ∂y ∂x
 2   2 
∂ u ∂ u
2 ∂u ∂ u ∂u ∂ 2u
= x x 2 −y − −y x + −y 2
∂y ∂y∂x ∂x ∂x∂y ∂y ∂x
∂ 2u 2∂ u
2 ∂ 2u ∂u ∂u
= x2 + y − 2xy −x −y .
∂y 2 ∂x 2 ∂x∂y ∂x ∂y
∂ 2u ∂ 2u
Using the Laplace equation in Cartesian form + 2 = 0, we get
∂x2 ∂y
     
∂u ∂u ∂ 2u 2 ∂ u
2 ∂ 2u 2 ∂ u
2 ∂ 2u
r r + 2 =x + 2 +y + 2 = 0.
∂r ∂r ∂θ ∂x2 ∂y ∂x2 ∂y
Hence, the result follows. 

LEMMA 7.1.2 Let v be a function on a region  such that it has continuous

partial derivatives. Then, dv = 0 for every closed curve γ in .
γ

Proof: By direct computation, if z(t), t ∈ [a, b] is a parametric equation of γ ,

 
  ∂v ∂v b
then dv = dx + dy = z (t) dt = z(x(b), y(b))−z(x(a), y(a)) =
γ γ ∂x ∂y a
0, as γ is a closed curve. 

RESULT 7.1.3 Let u1 and u2 be harmonic functions on a region .

If v1 and v2 are harmonic conjugates of u1 and u2 , respectively, then
(u1 dv2 − u2 dv1 ) = 0 for every cycle γ ∼ 0 in .
γ

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 343 — #3

i i

Some Interesting Theorems 343

Proof: If f1 = u1 + iv1 and f2 = u2 + iv2 , then f1 and f2 are analytic func-

tions on . Therefore, f2 is analytic, and hence, f1 f2 is also analytic on .
Therefore, by general version of Cauchy’s theorem (Theorem 4.4.3), we get
 
f1 f2 dz = 0 ⇒ Im f1 f2 dz = 0.
γ γ

Using
Im ( f1 f2 dz) = Im ( f1 d( f2 )) = Im ((u1 + iv1 )(du2 + idv2 ))
= u1 dv2 + v1 du2
= u1 dv2 + d(u2 v1 ) − u2 dv1

and using Lemma 7.1.2, we obtain d(u2 v1 ) = 0, and hence,
γ

(u1 dv2 − u2 dv1 ) = 0.
γ

Thus, the result follows. 

THEOREM 7.1.4 (Mean value property for harmonic function)


1 2π
If u is a harmonic function on Cl B(z, r), then u(z) = u(z+r exp(iθ )) dθ .
2π 0

Proof: Let v be a harmonic conjugate of u. Then, f = u + iv is an analytic

function on Cl B(z, r). Then, by Cauchy’s integral formula, we have

1 f (ζ )
f (z) = dζ .
i2π ζ −z
|ζ −z|=r

Since the parametric equation of |ζ − z| = r is ζ (θ ) = z + r exp(iθ ), ∀θ ∈

[0, 2π], we have
2π 2π
1 f (z + r exp(iθ )) 1
f (z) = ir exp(iθ )dθ = f (z + r exp(iθ )) dθ .
i2π r exp(iθ ) 2π
0 0


1 2π
Hence, we get u(z) = u(z + r exp(iθ )) dθ by equating the real parts on
2π 0
both sides. 

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 344 — #4

i i

344 Mean Value Property of Harmonic Functions

PROBLEM 7.1.5 If u is a harmonic function on B(0, s)\{0}, then

1 
u dθ = a log(r) + b, ∀0 < r < s for some A, B ∈ R.
2π |z|=r

Solution:
Let u1 = log(r) and u2 = u. Applying Theorem
  7.1.1, u1 is a harmonic
∂ ∂u1 ∂ 2 u1 ∂ 1
function, as r r + = r r + 0 = 0. Therefore, by
∂r ∂r  ∂θ 2 ∂r r
Theorem 7.1.3, we have (u1 dv2 − u2 dv1 ) = 0 for every cycle γ ∼ 0 in
γ
, where v1 and v2 are harmonic conjugates of u1 and u2 , respectively. For a
harmonic function u and its harmonic conjugate v, we first convert u dv into
the polar form (on |z| = r) as follows. Using C–R equations, we have
∂v ∂v ∂u ∂u
dv = dx + dy = − dx + dy.
dx dy dy dx
As in the proof of Theorem 7.1.1, from x = r cos(θ ) and y = r sin(θ ), we
have
 
∂u ∂u ∂u ∂u ∂u
r dθ = r cos(θ ) + r sin(θ ) dθ = x dθ + y dθ,
∂r ∂x ∂y ∂x ∂y
and using the fact that r is constant on |z| = r, we also have

dx = −r sin(θ ) dθ = −y dθ and dy = r cos(θ ) dθ = x dθ.

From the above two equations, we get

∂u ∂u ∂u
r dθ = dy − dx = dv on |z| = r.
∂r dx dy
If C1 and C2 are concentric circles with centre 0 and radii r1 and r2 ,
respectively, with 0 < r1 < r2 < s, then we claim that C1 ∪ (−C2 ) ∼ 0 in
B(0, s)\{0}. If w  B(0, s)\{0}, then either
 w = 0 or |w| > s. As WN(C1 ∪
1 − 1 = 0, if w = 0
(−C2 ), z) = WN(C1 , z) − WN(C2 , z) = , our claim
0 − 0 = 0, if |w| > s
holds.
Therefore, we have

0 = (u1 dv2 − u2 dv1 )
C1 ∪(−C2 )

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 345 — #5

i i

Some Interesting Theorems 345

  
∂u2 ∂u1
= u1 r dθ − u2 r dθ
∂r ∂r
C1 ∪(−C2 )
   
∂u
= log(r) r dθ − u dθ
∂r
C1 ∪(−C2 ) C1 ∪(−C2 )
       
∂u ∂u
= log(r1 ) r1 dθ − log(r2 ) r2 dθ − u dθ + u dθ ,
∂r ∂r
C1 C2 C1 C2

which implies that

   
∂u ∂u
u dθ − log(r1 ) r1 dθ = u dθ − log(r2 ) r2 dθ .
∂r ∂r
C1 C1 C2 C2

 ∂u 
Therefore, u dθ − log(r) dθ is a constant, say b. (Reason: After
r
|z|=r |z|=r ∂r
integrating these functions with respect to θ , we get only a function of r and
by previous equation, it is also independent of r.)
By a same argument using dv = 0, whenever γ ∼ 0 in B(0, r)\{0} (see
γ
∂u 
Lemma 7.1.2.), we get r dθ , which is a constant, say a.
|z|=r ∂r
1  a b
Thus, we get u dθ = A log(r) + B, where A = and B = .
2π |z|=r 2π 2π

Exercise 7.1.6 Deduce mean value property for a harmonic function on

B(0, r) from the previous problem.

THEOREM 7.1.7 (Maximum principle for harmonic function)

If u is a non-constant harmonic function on a region , then u cannot have
maximum in .

Proof: Suppose u attains maximum at z ∈ , then we have u(w) ≤ u(z), ∀w ∈

. Now, choose r > 0 such that Cl B(z, r) ⊆ . By mean value property for
harmonic function, we have

2π
1
u(z) = u(z + r exp(iθ )) dθ.
2π
0

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 346 — #6

i i

346 Mean Value Property of Harmonic Functions

We claim that u(w) = u(z), ∀w with |w − z| = r. If there exists w0 with

|w0 − z| = r and u(w0 ) < u(z), then let w0 = z + r exp(iθ0 ) for some
θ0 ∈ [0, 2π ]. Using the continuity of u and the exponential function, we
choose an interval [a, b] such that θ0 ∈ [a, b] ⊆ [0, 2π], where a < b, and for
every θ ∈ [a, b],
u(z) − u(z + r exp(iθ0 ))
|u(z + r exp(iθ )) − u(z + r exp(iθ0 ))| < ,
2
which implies
u(z) − u(z + r exp(iθ0 ))
u(z + r exp(iθ )) < u(z + r exp(iθ0 )) +
2
u(z) + u(z + r exp(iθ0 ))
=
2
u(z) + u(w0 )
= .
2
Hence, we get

2π
1
u(z) = u(z + r exp(iθ )) dθ
2π
0
⎛ a
 b
1 ⎝
= u(z + r exp(iθ)) dθ + u(z + r exp(iθ )) dθ
2π
0 a
⎞
2π
+ u(z + r exp(iθ)) dθ ⎠
b
⎛ a ⎞
 b   2π
1 ⎝ u(z) + u(w0 )
≤ u(z) dθ + dθ + u(z) dθ ⎠
2π 2
0 a b
u(z)
< (a + b − a + 2π − b) (since u(z) + u(w0 ) < 2u(z))
2π
= u(z),

which is a contradiction. Hence, u is constant on {w : |w − z| = r}. Therefore,

u(w) = u(z) on {w : |w − z| = s}, ∀s ∈ (0, r), and hence, u(w) = u(z)
on B(z, r). If A = {w   : u(w) = u(z)}, then A is an open set, by the
previous argument. As u is continuous and {z} is a closed set, we see that
A = u−1 ({u(z)}) is a closed set, and hence, Ac is an open set. Using that 

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 347 — #7

i i

Some Interesting Theorems 347

is connected and  = A ∪ Ac with both A and Ac are open, either A = ∅ or

Ac = ∅. Since z ∈ A, we get Ac = ∅, which implies that  = A. Therefore,
u is a constant on , which is a contradiction. Hence, there is no maximum
for u in . 

THEOREM 7.1.8 (Minimum principle for harmonic function)

If u is a non-constant harmonic function on a region , then u cannot have
minimum in .
Proof of this theorem is an immediate consequence of the previous theorem.
If we examine the proof of the above two theorems, we can notice that
the mean value property and the continuity of the function u are sufficient to
obtain these theorems. Hence, we can have the following theorem.

THEOREM 7.1.9 If u is a real-valued continuous function on  such that u is

continuous and satisfies
2π
1
u(z) = u(z + r exp(iθ )) dθ
2π
0

whenever Cl B(z, r) ⊆ , then u has neither maximum nor minumum in .

THEOREM 7.1.10 (Poisson’s formula)

If u is harmonic on B(0, r) and continuous on Cl B(0, r), then
2π
1 r2 − |z|2
u(z) = u(r exp(iθ )) dθ , ∀z ∈ B(0, r).
2π |r exp(iθ ) − z|2
0

Proof: Poisson’s formula at z = 0 is same as the mean value property of the

harmonic function u. So, we prove this result for z ∈ B(0, r)\{0}. First, we
note that
 
w+z 1 w+z w+z r2 − |z|2
Re = + = , ∀w with |w| = r.
w−z 2 w−z w−z |w − z|2
Case 1. u is harmonic on Cl B(0, r).
r(w − z)
We know that η = (w) = maps Cl B(0, r) onto Cl B(0, 1)
r2 − zw
(see Example 3.2.26)). As (by Result 3.2.5)

−1 r(rη + z)
w= (η) = ,
r + zη

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 348 — #8

i i

348 Mean Value Property of Harmonic Functions

We also know that −1 analytically maps Cl B(0, 1) onto Cl B(0, r).

If f is an analytic function on Cl B(0, r) such that Re f = u, then by
Chain rule (Theorem 2.1.10), f ◦ −1 is analytic on Cl B(0, 1), and
hence, Re ( f ◦ −1 ) is a harmonic function on Cl B(0, 1).
Using w = r exp(iθ ) ⇒ dw = ir exp(iθ )dθ = iwdθ and η =
exp(iφ), we get
d(exp(iφ))
dφ =
i exp(iφ)
dη
=
iη  
r(w − z)
= −id log 2
r − zw
 
r z
= −i + 2 dw
r(w − z) r − zw
 
1 z
= −i + iw dθ
w − z ww − zw
 
w z
= + dθ
w−z w−z
 2 
r − |z|2
= dθ
|w − z|2
 
r2 − |z|2
= dθ .
|r exp(iθ) − z|2
Then, by mean value property for harmonic function, we get
2π
−1 1 −1
u( (0)) = u (exp(iφ)) dφ
2π
0
2π  
1 r2 − |z|2
= u (r exp(iθ )) dθ.
2π |r exp(iθ ) − z|2
0

Case 2. u is harmonic on B(0, r) and continuous on Cl B(0, r).

Then, for every s ∈ (0, 1), if us (w) = u(sw), ∀w ∈ Cl B(0, r), then
us is harmonic on Cl B(0, r), and hence, by Case 1, we get
2π
1 s2 r2 − |z|2
u(sz) = u (sr exp(iθ )) dθ .
2π |sr exp(iθ ) − z|2
0

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 349 — #9

i i

Some Interesting Theorems 349

By using the uniform continuity of u on Cl B(0, r), we get

2π
1 r2 − |z|2
u(z) = lim u(sz) = u (r exp(iθ )) dθ .
s→1 2π |r exp(iθ ) − z|2
0

Thus, the theorem follows. 

Remark 7.1.11: The Poisson’s formula can also be written in the following
ways.


a+2π
1 r2 − |z|2
u(z) = u (r exp(iθ )) dθ, (a ∈ R). (7.1)
2π |r exp(iθ) − z|2
a
2π  
1 r exp(iθ ) + z
= Re u(r exp(iθ )) dθ . (7.2)
2π r exp(iθ) − z
0
⎛ ⎞

1 ⎜ w + z u(w) ⎟
= Re ⎝ dw⎠ , ∀z ∈ B(0, r). (7.3)
2π w − z iw
|w|=r

THEOREM 7.1.12 (Jensen’s formulae)

Let f be an analytic function on Cl B(0, r) such that f is not identically zero.
Case 1. If f  0 on B(0, r), then

2π
log(| f (0)|) = log(| f (r exp(iθ ))|) dθ.
0

Case 2. If the zeroes of f in B(0, r) are aj ∈ B(0, r)\{0}, j = 1, 2, . . . , n,

including multiplicities, then


n   2π
r
log(| f (0)|) = − log + log(| f (r exp(iθ ))|) dθ .
|aj |
j=1 0

Case 3. If f has a zero of order m at 0 and the zeroes of f in B(0, r) are

aj ∈ B(0, r)\{0}, j = 1, 2, . . . , n, including multiplicities, then

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 350 — #10

i i

350 Mean Value Property of Harmonic Functions


n   2π
r
log(|cm |) + m log(r)=− log + log(| f (r exp(iθ ))|) dθ,
|aj |
j=1 0

where cm is the coefficient of zm in the Taylor’s series expansion of

f in B(0, r).

Proof:

Case 1. Subcase (i) If f is nowhere zero on Cl B(0, r), then log( f ) is an

analytic function on Cl B(0, r) (by Theorem 4.5.12). Therefore,
by Result 2.2.6, Re log( f ) = log(| f |) is harmonic on Cl B(0, r).
Therefore, by mean value property of harmonic function, we have

2π
1
log(| f (0)|) = log( f (r exp(iθ ))) dθ .
2π
0

Subcase (ii) Suppose f is nowhere zero on B(0, r) and the zeroes of

f on {z : |z| = r} are r exp(iθ1 ), r exp(iθ2 ), . . . , r exp(iθν ), including
the multiplicities, for some θk ∈ [0, 2π], ∀k = 1, 2, . . . , ν. Then,
f (z)
let f1 (z) = ν . Then, f1 has no zeroes on Cl B(0, r) after
(z−r exp(iθk ))
k=1
removing the removable singularities at r exp(iθk ), k = 1, 2, . . . , ν .
Therefore, by replacing f by f1 in the Subcase (i), we get

⎛ − ν log r
log(| f (0)|) ⎞
⎜ | f (0)| ⎟
= log ⎜
⎝
⎟
⎠

ν
|0 − r exp(iθk )|
k=1
= log(| f1 (0)|)
2π
1
= log( f1 (r exp(iθ))) dθ
2π
0 ⎛ ⎞
2π
1 ⎜ | f (r exp(iθ ))| ⎟
= log ⎜
⎝
⎟ dθ
⎠
2π 
ν
0 |r exp(iθ ) − r exp(iθk )|
k=1

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 351 — #11

i i

Some Interesting Theorems 351

2π 
1
= log(| f (r exp(iθ ))|)
2π
ν
0 

− [log(r) + log(| exp(iθ ) − exp(iθk )|)] dθ.
k=1
Thus, we have
2π 
1
log(| f (0)|) = log(| f (r exp(iθ ))|)
2π
ν
0 

− log(| exp(iθ ) − exp(iθk )|) dθ . (7.4)
k=1


2π
We claim that log(| exp(iθ) − exp(iθk )|) = 0, ∀k ∈ {1, 2, . . . , ν}.
0
Using the fact that exp(it) is a periodic function on R of period 2π
(Result 2.4.20) and using Lemma 2.4.23, we obtian
2π
log(| exp(iθ) − exp(iθk )|)
0
2π
= log(|1 − exp(i(θ − θk )|))
0
θk+2π

= log(|1 − exp(it)|)
θk
(using the change of variable t = θ − θk )
2π
= log(|1 − exp(it)|) dt. (7.5)
0
Since
1
log(|1 − exp(it)|) = log((1 − cos(t))2 + sin2 (t))
2
1
= log(1 + cos2 (t) − 2 cos(t) + sin2 (t))
2
1
= log (2(1 − cos(t)))
2

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 352 — #12

i i

352 Mean Value Property of Harmonic Functions

1
= [log(2) + log(1 − cos(t))]
2
1
= [log(2) + log(2 sin2 (t/2))]
2
= log(2) + log(sin(t/2)),
the right-hand side of equation (7.5) is equal to
2π π
[log(2) + log(sin(t/2))] dt = 2π log(2) + 2 log(sin(u)) du
0 0
(by the change of variable u = t/2)
= 2π log(2) − 2π log(2)
(by Theorem 6.5)
= 0.
Using these observations in equation (7.4), we get
2π
1
log(| f (0)|) = log(| f (r exp(iθ ))|) dθ.
2π
0


n r2 − a z
j
Case 2. Let f2 (z) = f (z) , ∀z ∈ Cl B(0, r). Clearly, f2 becomes
j=1 r(z − aj )
an analytic function on Cl B(0, r), and it has no zeroes in B(0, r). If
|z| = r, then let z = r exp(iθ ) for some θ ∈ [0, 2π], and hence,
 
 r2 − a z   r exp(iθ )(r exp(−iθ ) − a )   r exp(−iθ ) − a 
 j 
=
j   j

 r(z − aj )   r(r exp(iθ ) − aj )  =  r exp(iθ ) − a  = 1.
j

Hence it follws that | f2 | = | f | on |z| = r Therefore, replacing f (z)

by f2 (z) in Case 1, we obtain
⎛ ⎞
n   
n
r r
log(| f (0)|) + log = log ⎝| f (0)| ⎠
|aj | |aj |
j=1 j=1

2π
1
= log(| f2 (r exp(iθ ))|) dθ
2π
0
2π
1
= log(| f (r exp(iθ ))|) dθ .
2π
0

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 353 — #13

i i

Some Interesting Theorems 353

m
Case 3. Now, let f3 (z) = f (z) rzm , ∀z ∈ Cl B(0, r) so that f3 is analytic and
f3 (0)  0. Hence, applying Case 2, we get


n   2π
r 1
log(| f3 (0)|) = − log + log |f3 (r exp(iθ ))| dθ .
|aj | 2π
j=1 0


n   2π
r 1
= − log + log | f (r exp(iθ ))| dθ ,
|aj | 2π
j=1 0
since | f3 | = |f | on | z| = r. If the coefficient of zm in the Taylor’s se-
ries expansion of f is cm , then we have f3 (0) = cm rm , and hence, the
right-hand side of the above equation becomes log(|cm |) + m log(r).


Remark 7.1.13: In the Jensen’s formulae, if f has zeroes on {z ∈ C : |z| =

r}, then the integral on the right-hand side has no meaning as in Definition
4.1.6, and hence, the value of the integral should be understood by the above
formulae.
THEOREM 7.1.14 (Poisson–Jensen’s formula)
Let f be an analytic function on Cl B(0, r) such that f is not identically zero.
If the zeroes of f in Cl B(0, r) are aj ∈ B(0, r)\{0}, j = 1, 2, . . . , n, including
multiplicities, then
 
n  r2 − a z  1 2π r exp(iθ ) + z
 j 
log(| f (z)|) = − log  + Re log(| f (r exp(iθ ))|) dθ
 r(z − aj )  2π r exp(iθ ) − z
j=1 0
provided f (z)  0.

n r2 − a z
j
Proof: Let F(z) = f (z) ∀z ∈ Cl B(0, r). Then, clearly F is
j=1 r(z − a j)
nowhere zero analytic on Cl B(0, r). (For details,  see the proof of the pre-
 r2 − a z 
 j 
vious theorem.) Using the fact that   = 1 on |z| = r, we obtain
 r(z − aj ) 
| F(z)| = | f (z)| on |z| = r. Hence, log(F) is analytic and by applying the
Poisson’s formula (Theorem 7.1.10) for log(| f |), we get
 
  r2 − a z 
n
 j 
log(| f (z)|) + log  
j=1  r(z − aj ) 
= log(|F(z)|)

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 354 — #14

i i

354 Poisson’s Integral

2π
1 r exp(iθ ) + z
= Re log |F(r exp(iθ ))| dθ
2π r exp(iθ ) − z
0
2π
1 r exp(iθ ) + z
= Re log | f (r exp(iθ ))| dθ .
2π r exp(iθ ) − z
0

Thus, the theorem follows. 

7.2 POISSON’S INTEGRAL

Definition 7.2.1 Let u be a real-valued bounded piecewise continuous
function on the unit circle C . Then, the Poisson integral of u is defined by
⎛ ⎞

1 ζ + z u(ζ )
Pu (z) = Re ⎝ dζ ⎠ , ∀z ∈ B(0, 1).
2π ζ − z iζ
C

LEMMA 7.2.2 Let u be a real-valued bounded piecewise continuous function

on the unit circle C , and let γ be a circular arc given by γ (t) = exp(it), t ∈
[a, b], where a, b ∈ R with 0 ≤ b − a ≤ 2π .
 
1  ζ + z u(ζ )
1. If Pu,γ (z) = Re dζ , ∀z ∈ γ c , then Pu,γ is
2π γ ζ − z iζ
harmonic on γ c .
2. If u is continuous at ζ0 ∈ C , then lim Pu (z) = u(ζ0 ).
z→ζ0

Proof:
1. If
 
1 u(ζ ) 1 u(ζ )
f1 (z) = dζ , f2 (z) = dζ , ∀z ∈ γ c ,
2πi ζ −z 2π i ζ (ζ − z)
γ γ

then as in the proof of Theorem 4.3.2, we get that f1 and f2 are analytic
on γ c .
(Note that the only difference between the current situation and Theo-
rem 4.3.2 is that here, we have a piecewise continuous function u on γ ,
whereas φ is a continuous function on γ in Theorem 4.3.2. However,
to follow the proof of Theorem 4.3.2, we need continuity of φ to get

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 355 — #15

i i

Some Interesting Theorems 355


|φ(ζ )||dζ | < +∞. As u is bounded piecewise continuous on γ , we
γ 
also have |u(ζ )||dζ | < +∞. Thus, the same proof will work in the
γ
current situation also.)
Therefore, if f (z) = f1 (z) + zf2 (z), ∀z ∈ γ c , then f is analytic on γ c and
⎛ ⎞
  
1 u(ζ ) zu(ζ ) dζ
Re f (z) = Re ⎝ + ⎠
2π ζ − z ζ (ζ − z) i
γ
⎛ ⎞
  
1 (ζ + z)u(ζ )
= Re ⎝ dζ ⎠
2π iζ (ζ − z)
γ
= Pu,γ .
Thus, Pu,γ is a harmonic function on γ c .
2. Let ζ0 = exp(it0 ) for some t0 ∈ R. First, we assume that u(ζ0 ) = 0.
If v : R → R is defined by v(t) = u(exp(it)), ∀t ∈ R, then v is
continuous at t0 . Hence, for a given  > 0, there exists r > 0 such that
 
t ∈ (t0 − r, t0 + r) ⇒ |v(t) − v(t0 )| < ⇒ |u(exp(it))| < .
2 2
Let a = t0 − r and b = t0 + r, so that C = C1 ∪ C2 , where
C1 = {exp(it) : b − 2π ≤ t ≤ a} and C2 = {exp(it) : a < t < b}.
Since u is a piecewise continuous function on C1 and on Cl C2 Pu,C1 is
harmonic on C1c and Pu,C2 is harmonic on (Cl C2 )c . Now for each ξ ∈
 
exp(it) + ξ 1 − |ξ |2
C2 , using Re = (Cf. Proof of Theorem
exp(it) − ξ | exp(it) − ξ |2
7.1.10), we have
⎛ ⎞

1 ⎜ ζ + ξ u(ζ ) ⎟
Pu,C1 (ξ ) = Re ⎝ dζ ⎠
2π ζ − ξ iζ
C1
⎛ ⎞
a
1 exp(it) + ξ
= Re ⎝ u(exp(it)) dt⎠
2π exp(it) − ξ
b−2π
a
1 1 − |ξ |2
= u(exp(it)) dt = 0,
2π | exp(it) − ξ |2
b−2π

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 356 — #16

i i

356 Poisson’s Integral

as u is real valued and |ξ | = 1. In particular, Pu,C1 (ζ0 ) = 0. Being

Pu,C1 is harmonic on C1c , it is continuous on C1c . Therefore, Pu,C1 is
continuous at ζ0 . Hence, for a given  > 0, there exists δ > 0 such

that |z − ζ0 | < δ ⇒ |Pu,C1 (z)| < . By the choice of C2 , we have
2

|u(ζ )| < , ∀ζ ∈ C2 . Therefore, for each z ∈ B(0, 1), we have
2
 ⎛ ⎞
  
1  ζ + ξ u(ζ ) ⎟
⎜
|Pu,C2 (z)| = Re ⎝ dζ ⎠
2π  ζ − ξ iζ 
 Cl C2 
 b 
   
1  exp(it) + z 
= Re u(exp(it)) dt 
2π  exp(it) − z 

a
(as u is real valued)
b   
 
≤
1 Re exp(it) + z  |u(exp(it))| dt
2π  exp(it) − z 
a
b
1 1 − |z|2
= |u(exp(it))| dt
2π | exp(it) − z|2
a
b
 1 1 − |z|2
≤ dt
2 2π | exp(it) − z|2
a
b
 1 1 − |z|2
≤ dt
2 2π | exp(it) − z|2
b−2π

= .
2

The last equality is obtained by using the Poisson’s formula in equation

(7.1) for the constant function 1. Thus, for z ∈ B(0, 1) with |z−ζ0 | < δ ,
we have
 
|Pu (z)| = |Pu,C1 (z) + Pu,C2 (z)| ≤ |Pu,C1 (z)| + |Pu,C2 (z)| < + = .
2 2

Therefore, we have proved that lim u(z) = 0.

z→ζ0
Suppose u(ζ0 )  0, if v = u−u(ζ0 ), then v is also a real-valued bounded
piecewise continuous function on C with v(ζ0 ) = 0. So, applying the

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 357 — #17

i i

Some Interesting Theorems 357

previous case for v, we get that Pv is a harmonic function on B(0, 1)

and lim Pv (z) = 0. However,
z→ζ0
⎛ ⎞

1 ζ + z v(ζ ) ⎠
Pv (z) = Re ⎝ dζ
2π ζ − z iζ
C
⎛ ⎞ ⎛ ⎞
 
1 ζ + z u(ζ ) ⎠ 1 ζ + z u(ζ0 ) ⎠
= Re ⎝ dζ − Re ⎝ dζ
2π ζ − z iζ 2π ζ − z iζ
C C
⎛ ⎞

1 ζ + z dζ ⎠
= Pu (z) − u(ζ0 ) Re ⎝ , (as u(ζ0 ) ∈ R)
2π ζ − z iζ
C
b
1 1 − |z|2
= Pu (z) − u(ζ0 ) dt
2π | exp(it) − z|2
b−2π
= Pu (z) − u(ζ0 ) (applying Poisson’s formula for u).
Thus, we have lim Pu (z) = u(z0 ). 
z→ζ0
From Poisson’s formula and from the proof of previous theorem, we can
see that if u is a harmonic function on B(0, r) and continuous on Cl B(0, r),
then the analytic function f on B(0, r), whose real part is the given u, can be
obtained by

1 ζ + z u(ζ )
f (z) = dζ , ∀z ∈ B(0, r),
i2π ζ −z ζ
|ζ |=r

which is called the Schwarz’s formula. The following theorem is also due to
Schwarz.

THEOREM 7.2.3 (Schwarz’s theorem)

For a fixed z ∈ B(0, 1), let uz : [0, 2π ) → [0, 2π ) be the function defined by
uz (θ ) = θz∗ , where θz∗ is such that exp(iθ ), exp(iθz∗ ) and z are collinear. If
Uz : C → [0, 2π) is defined by Uz (ζ ) = uz (θ ) = θz∗ ,
where θ ∈ [0, 2π) is unique (however depending on ζ ) such that ζ = exp(iθ),
∀ζ ∈ C , where C = {ζ ∈ C : |ζ | = 1}. Then,
2π 2π
PUz (z) = Uz (exp(iθ )) dθz∗ = Uz (exp(iθz∗ )) dθ , ∀z ∈ B(0, 1).
0 0

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 358 — #18

i i

358 Poisson’s Integral

Proof: Note that uz : [0, 2π) → [0, 2π) is a bijection, with uz = u−1
z .

B = exp(iq )

A*z
z
0

B*z = exp(iq *z )

By the intersecting chord theorem1 , we obtain that the area of the rectangle
with sides |z − A| and |z − A∗z | is equal to the area of the rectangle with sides
|z−B| and |z−B∗z |. From the diagram, |z−A| is the length of the line segment
joining z and A, which is equal to |z|+|A| = |z|+1 and |z−A∗ | = |A∗ |−|z| =
1 − |z|, and hence, we have

1 − |z|2 = (1 − |z|)(1 + |z|) = |z − exp(iθ )| |z − exp(iθz∗ )|. (7.6)

As z lies between the points exp(iθ) and exp(iθz∗ ), we get that z − exp(iθ ) and
z − exp(iθz∗ ) lie on the same line but in opposite directions from z. Therefore,
z − exp(iθ )
it follows that < 0. If arg (z − exp(iθ )) = ϕ and arg (z −
z − exp(iθz∗ )
exp(iθz∗ )) = ψ , then we have ϕ − ψ = π (if the arguments are written in
[0, 2π)). Moreover, we have

(z − exp(iθ ))(z − exp(−iθz∗ )) = |z − exp(iθ )||z exp(−iθz∗ )| exp(i(ϕ − ψ))

= −|z − exp(iθ )| |z − exp(iθz∗ )|
= |z|2 − 1 (using equation 7.6).

Hence, by differentiating on both sides of the above equation, we get

i exp(iθ) dθ (z − exp(−iθz∗ )) − i exp(−iθz∗ )dθz∗ (z − exp(iθ )) = 0

dθ z − exp(iθ )
⇒ = exp(−i(θ + θz∗ )) .
dθz∗ z − exp(−iθz∗ )
1 It is a theorem in 2D analytic geometry, proved by Euclid, which says that in a circle, two
straight lines cut one another, and the rectangle contained by the segments of the one is equal to
the rectangle contained by the segments of the other.

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 359 — #19

i i

Some Interesting Theorems 359

Using the fact that dθ and dθz∗ are the usual Lebesgue measures on [0, 2π ]
(positive measures) and using equation (7.6), we have
   
dθ  dθ   z − exp(iθ )  1 − |z|2
=   =  =
dθz∗  dθ ∗   z − exp(−iθ ∗ )  |z − exp(−iθ ∗ )|2 .
z z z

Therefore, by definition of PUz , we have

  
1 ζ +z dζ
PUz (z) = Re Uz (ζ )
2π ζ −z iζ
C
2π  
1 exp(iθ ) + z
= Re Uz (exp(iθ)) dθ
2π exp(iθ ) − z
0
2π
1 1 − |z|2
= Uz (exp(iθ )) dθ
2π | exp(iθ ) − z|2
0
2π
1
= Uz (exp(iθ )) dθz∗ .
2π
0

Using the fact that u−1

z = uz and interchanging θ and θz∗ in the above
expression, we also get
2π
1
PUz (z) = Uz (exp(iθz∗ )) dθ.
2π
0

Hence, the theorem follows. 

It is interesting to note that the converse of the mean value property for
harmonic functions is also true.

THEOREM 7.2.4 If u is a real-valued continuous function on , satisfying


2π
the mean value property, u(a) = 2π
1
u(a + r exp(it)) dt, ∀a ∈ , for some
0
r > 0 such that Cl B(a, r) ⊆ , then u is harmonic on .

Proof: Let a ∈  be arbitrary and r > 0 be such that Cl B(a, r) ⊆ . If

v : Cl B(0, 1) → R is defined by
v(z) = u(a + rz), ∀z ∈ Cl B(0, 1),

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 360 — #20

i i

360 Poisson’s Integral

then v is a continuous function on Cl B(0, 1). Therefore, by Lemma 7.2.2, if

⎧  
⎪
⎨ 1 Re  ζ + z v(ζ )
dζ if |z| < 1
Pv (z) = 2π |ζ |=1 ζ − z iζ
⎪
⎩
v(z) if |z| = 1,

then Pv is a harmonic function on B(0, 1) and continuous on Cl B(0, 1). We

define φ : Cl B(a, r) → R by
 
w−a
φ(w) = Pv , ∀z ∈ Cl B(a, r).
r
If f is an analytic function on B(0, 1) such that Pv = Re f , then we have
φ = Re ( f ◦ T), where T : Cl B(a, r) → Cl B(0, 1) is defined by T(w) = w−ar ,
∀w ∈ Cl B(a, r). Since T is analytic, we get that φ is harmonic on B(a, r) and
is continuous on Cl B(a, r). In particular, if |w − a| = r, then
      
w−a w−a w−a
φ(w) = Pv =v =u a+r = u(w).
r r r
By hypothesis and by Theorem ??, both u and φ satisfy mean value property,
and hence, u − φ is a real function on Cl B(a, r), satisfying the following
properties:
1. u − φ is a continuous function on Cl B(a, r),
2. u − φ = 0 on |w − a| = r,
3. u − φ satisfies the mean value property.
As u − φ is a continuous function on the compact set Cl B(a, r), it attains its
maximum and minimum on Cl B(a, r) (Theorem 1.6.28). However, applying
Theorem 7.1.9, it follows that u − φ has neither maximum nor minimum on
B(a, r). Hence, u − φ attains its maximum and minimum on |w − a| = r
(on which this function vanishes). Thus, u − φ = 0 on Cl B(a, r). Therefore,
u = φ is a harmonic function on B(a, r). As a ∈  is arbitrary, u is harmonic
on . 

THEOREM 7.2.5 (Harnack’s inequality)

Let u be a non-negative real-valued continuous function on Cl B(a, ρ) and
harmonic on B(a, ρ). If |z − a| ≤ r < ρ , then
ρ−r ρ+r
u(a) ≤ u(z) ≤ u(a).
ρ+r ρ−r

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 361 — #21

i i

Some Interesting Theorems 361

Proof: Let z0 ∈ C be arbitrary such that |z0 − a| = r. As

ρ−r = |ρ exp(it)|−|z0 −a| ≤ |a+ρ exp(it)−z0 | ≤ |ρ exp(it)|+|z0 −a| ≤ ρ+r,

we have (ρ − r)2 ≤ |a + ρ exp(it) − z0 |2 ≤ (ρ + r)2 . Therefore, we have

ρ−r ρ 2 − r2 ρ 2 − r2 ρ 2 − r2 ρ+r
= ≤ ≤ = .
ρ+r (ρ + r) 2 |a + ρ exp(it) − z0 | 2 (ρ − r) 2 ρ−r

As u(ζ ) ≥ 0 for all ζ ∈ C with |ζ − a| = ρ , we have

ρ−r ρ 2 − r2
u (a + ρ exp(it)) ≤ u (a + ρ exp(it))
ρ+r |a + ρ exp(it) − z0 |2
ρ+r
≤ u (a + ρ exp(it)) ,
ρ−r

which implies that

2π
ρ−r 1
u (a + ρ exp(it)) dt
ρ + r 2π
0
2π
1 ρ 2 − r2
≤ u (a + ρ exp(it)) dt
2π |a + ρ exp(it) − z0 |2
0
2π
ρ+r 1
≤ u (a + ρ exp(it)) dt, ∀t ∈ [0, 2π).
ρ − r 2π
0

Now, using mean value property for harmonic functions, we have

2π
ρ−r 1 ρ 2 − r2 ρ+r
u(a) ≤ u (a + ρ exp(it)) dt ≤ u(a).
ρ+r 2π |a + ρ exp(it) − z0 |2 ρ−r
0

If v(w) = u(a + w), ∀w ∈ B(0, ρ), then using the Poisson’s formula for v as
in equation (7.1), we obtain that the integral in the above inequality is equal
to v(z0 ) = u(a + z0 ), and hence, we have
ρ−r ρ+r
u(a) ≤ u(a + z0 ) ≤ u(a), ∀z0 ∈ C with |z0 − a| = r.
ρ+r ρ−r

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 362 — #22

i i

362 Poisson’s Integral

Applying the maximum and minimum principles for harmonic functions, we

get
ρ−r ρ+r
u(a) ≤ u(z) ≤ u(a), ∀z ∈ Cl B(a, r).
ρ+r ρ−r
Thus, the inequality is proved. 

THEOREM 7.2.6 (Harnack’s principle)

Let un be harmonic on n , ∀n ∈ N. Let  be a region such that for each a ∈
, there exists r > 0 and N ∈ N such that Cl B(a, ρ) ⊆ n and un ≤ un+1
on Cl B(a, ρ), ∀n ≥ N . Then, either un → ∞ uniformly on every compact
subset of  as n → ∞ or un → u uniformly on every compact subset of 
as n → ∞ for some harmonic function u on .

Proof: For an arbitrary a ∈ , there exist ρ > 0 and N ∈ N such that

Cl B(a, ρ) ⊆ n and un ≤ un+1 on Cl B(a, ρ), ∀n ≥ N . If n ≥ m ≥ N ,
then un − um is a non-negative harmonic function on Cl B(a, ρ), and hence,
by Harnack’s inequality for un − um ,

R1 (un − um )(a) ≤ (un − um )(z) ≤ R2 (un − um )(a), ∀z ∈ Cl B(a, r) (7.7)

holds, where R1 = ρ−r ρ+r

ρ+r , R2 = ρ−r , and 0 < r < ρ be arbitrary but fixed.
2 Since (u
n+N (a)) is an increasing sequence of real numbers, either
un (a) → +∞ as n → ∞ or un (a) → b as n → ∞, for some b ∈ R.
We first assume that un (a) → +∞ as n → ∞.
If μ = inf{um (z) : z ∈ Cl B(a, r)} (which exists because um is a continu-
ous function and Cl B(a, r) is a compact set), then using the first inequality in
equation (7.7), for every z ∈ Cl B(a, r), we get

un (z) ≥ R1 (un − um )(a) + um (z)

≥ R1 un (a) + [μ − R1 um (a)] → +∞ as n → ∞,

as R1 > 0. Therefore, un → +∞ uniformly on Cl B(a, r) as n → ∞. As

every compact subset of  is contained in a finite union of such closed discs,
we conclude that un → +∞ uniformly on every compact subset of , as
n → ∞.

Suppose un (a) → b as n → ∞ for some b ∈ C. Then, arguing as before,

using the second inequality in equation (7.7), for every z ∈ Cl B(a, r), we get

0 ≤ un (z) − um (z) ≤ R2 (un − um )(a) → 0 as m, n → ∞

2 It is a simple exercise to the reader.

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 363 — #23

i i

Some Interesting Theorems 363

Therefore, (un ) is a uniformly Cauchy sequence on Cl B(a, r), and hence, it

is uniformly convergent on Cl B(a, r) and also on every compact subset of
. If u = lim un , then u is a uniform limit of a sequence of continuous
n→∞
functions on every compact subset of  and so it is a continuous function
on . Furthermore, for every z ∈ , if r > 0 and n0 ∈ N are such that
Cl B(z, r) ⊆ n , for all n ≥ n0 , then we have

2π
1
un (z) = un (z + r exp(it)) dt, ∀n ≥ n0 .
2π
0

Allowing n → ∞ on both sides, the uniform convergence of un on compact

subsets of  implies that

2π
1
u(z) = u(z + r exp(it)) dt.
2π
0

Hence, applying Theorem 7.2.4, we get that u is a harmonic function on .



7.3 SCHWARZ REFLECTION PRINCIPLE

In the following sequel, we shall use + =  ∩ {z ∈ C : Im z > 0},
− =  ∩ {z ∈ C : Im z < 0} and σ =  ∩ R, where  is a region.

THEOREM 7.3.1 (Schwarz reflection principle)

Let  be a region such that  = {z : z ∈ }.
1. If v : + ∪ σ → R is continuous on + ∪ σ and harmonic on + and
if v(z) = 0, ∀z ∈ σ , then there exists a harmonic function V :  → R
such that V (z) = v(z), ∀z ∈ + ∪ σ and V (z) = −V (z), ∀z ∈ .
2. Furthermore, if f = u + iv is analytic on + , where v is as in (1), then
there exists an analytic function F on  such that F(z) = f (z), ∀z ∈ +
and F(z) = F(z), ∀z ∈ .


v(z) if z ∈ + ∪ σ
Proof: Define V :  → R by V (z) = ∀z ∈ .
−v(z) if z ∈ −
Immediately, it follows that V (z) = −V (z), ∀z ∈ ; indeed,

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 364 — #24

i i

364 Schwarz Reflection Principle

⎧
⎪
⎨v(z) if z ∈ +
V (z) = 0 if z ∈ σ
⎪
⎩
−v(z) if z ∈ −
⎧
⎪
⎨v(z) if z ∈ −
= 0 if z ∈ σ
⎪
⎩
−v(z) if z ∈ +
= −V (z).

To prove that V is harmonic on , we shall show that V is harmonic on

∂V ∂ 2V
σ ∪ − . For (x, y) ∈ − , we have (x, y) = −vx (x, −y), (x, y) =
∂x ∂x2
∂V ∂ V
2
−vxx (x, −y), (x, y) = vy (x, −y), and 2 (x, y) = −vyy (x, −y). Therefore,
∂x ∂x
∂ 2V ∂ 2V
(x, y) + (x, y) = −vxx (x, −y) − vyy (x, −y) = 0,
∂x2 ∂x2
as v is harmonic on + and (x, −y) ∈ + whenever (x, y) ∈ − . Therefore,
V is harmonic on − . Let x0 ∈ σ ⊆  be arbitrary. Since  is open, choose
r > 0 such that B(x0 , r) ⊆ , and hence, B(x0 , r) ∩ R = (x0 − r, x0 + r).
z − x0
If T(z) = , ∀z ∈ C, then T maps B(x0 , r) onto B(0, 1). Define ϕ :
r
B(0, r) → R by

ϕ(w) = (V ◦ T −1 )(w) = V (rw + x0 ), ∀w ∈ B(0, 1).

Clearly, ϕ is a continuous function on {exp(iθ ) : 0 < θ < 2π}; in other

words, it is a piecewise continuous function on the unit circle. Therefore, by
Lemma 7.2.2, Pϕ is a harmonic function on B(0, 1), where
⎛ ⎞

1 ⎜ w + ξ ϕ(ξ ) ⎟
Pϕ (w) = Re ⎝ dξ ⎠
2π w − ξ iξ
|ξ |=1

1 1 − |w|2 ϕ(ξ )
= dξ , ∀w ∈ B(0, 1).
2π |ξ − w|2 ξ
|ξ |=1

Clearly, Pϕ ◦ T is a harmonic function on B(x0 , r), and for z ∈ B(x0 , r),

 
z − x0
(Pϕ ◦ T)(z) = Pϕ
r

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 365 — #25

i i

Some Interesting Theorems 365


1 r2 (1 − | z−x
r | ) ϕ(ξ )
0 2
= dξ
2π |rξ − (z − x0 )|2 ξ
|ξ |=1

1 r2 − |z − x0 |2 V (ζ )
= dζ
2π |ζ − z|2 (ζ − x0 )
|ζ −x0 |=r
ζ − x0
(by the change of variable ξ = T(ζ ) =
r
and using the fact that ϕ ◦ T = V ).

We also see that


1 r2 − |z − x0 |2 V (ζ )
(Pϕ ◦ T)(z) = dζ
2π |ζ − z|2 (ζ − x0 )
|ζ −x0 |=r

π
1 r2 − |z − x0 |2
= V (x0 + r exp(iθ )) idθ
2π |x0 + r exp(iθ ) − z|2
−π

π
1 r2 − |z − x0 |2
= V (x0 + r exp(−iϕ)) idϕ
2π |x0 + r exp(−iϕ) − z|2
−π

(using the change of variable ϕ = −θ)

π
1 r2 − |z − x0 |2
= V (x0 + r exp(iϕ)) idϕ
2π |x0 + r exp(iϕ) − z|2
−π

(using x0 ∈ R)
π
1 r2 − |z − x0 |2
= [−V (x0 + r exp(iϕ))] idϕ
2π |x0 + r exp(iϕ) − z|2
−π

1 r2 − |z − x0 |2 V (ω)
= − dω
2π |ω − z|2 (ω − x0 )
|ω−x0 |=r

= −(Pϕ ◦ T)(z).

Now, V − Pϕ ◦ T is a harmonic function on B(x0 , r) ∩ + with the following

properties:

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 366 — #26

i i

366 Schwarz Reflection Principle

1. (V − Pϕ ◦ T)(z) = 0, ∀z ∈ {x0 + r exp(iθ ) : 0 < θ < π}.

Justification. Using Lemma 7.2.2, for 0 < s < r, (Pϕ ◦ T)(x0 +
s exp(iθ )) = Pϕ ( rs exp(iθ)) → ϕ(exp(iθ )) = V (x0 + r exp(iθ )) as
s → r. Hence, (Pϕ ◦ T)(x0 + r exp(iθ )) = lim (Pϕ ◦ T)(x0 + s exp(iθ )) =
s→r
V (x0 + r exp(iθ )).
2. (V − Pϕ ◦ T)(z) = 0, ∀x ∈ B(x0 , r) ∩ σ .
Justification. By definition, V (x) = 0, ∀x ∈ σ . From the fact that Pϕ ◦T
is continuous on B(x0 , r) ∩ σ and (Pϕ ◦ T)(z) = (Pϕ ◦ T)(z), we have
(Pϕ ◦ T)(x) = lim (Pϕ ◦ T)(x + iy) = − lim (Pϕ ◦ T)(x − iy) = −(Pϕ ◦
y→0 y→0
T)(x) ⇒ (Pϕ ◦ T)(x) = 0, ∀x ∈ B(x0 , r) ∩ σ .
3. (V − Pϕ ◦ T)(z) = 0, ∀z ∈ B(x0 , r) ∩ + .
Justification. From the above two properties, we have V − Pϕ ◦ T ,
which is a harmonic function on B(x0 , r) ∩ + that vanishes on
∂(B(x0 , r) ∩ + ). Hence, by maximum and minimum principles for
harmonic functions, V − Pϕ ◦ T = 0 on B(x0 , r) ∩ + .

Hence, V = Pϕ ◦ T on B(x0 , r) ∩ + . Applying the same technique, we can

prove that V = Pϕ ◦ T on B(x0 , r) ∩ − . Thus, V = Pϕ ◦ T is harmonic on
σ , and hence, it is harmonic on .
Next, for the given analytic function f = u + iv on + , we define
⎧
⎪ +
⎨f (z) if z ∈ 
F(z) = U(z) if z ∈ σ
⎪
⎩
f (z) if z ∈ − , ∀z ∈ ,

where −U is a harmonic conjugate of V such that f = U + iV on + .

Using Result 2.2.29, we immediately get that F is analytic on − . As −U
is a harmonic conjugate of V , U + iV is an analytic function, wherever they
are defined. In particular, U is analytic on σ , as V = 0 on σ . Therefore, F
is analytic on , and by definition, F(z) = f (z), ∀z ∈ + and F(z) = F(z),
∀z ∈  \ σ . Let z0 ∈ σ be arbitrary. Find r > 0 such that B(z0 , r) ⊆ . If ψ :
∂ψ ∂ψ
B(z0 , r) → R is defined by ψ(z) = U(z) − U(z) and (z) = (z) − i (z),
∂x ∂y
∀z ∈ B(z0 , r), then we claim that ψ is an harmonic function and  is an
analytic function on B(z0 , r). For (x, y) ∈ B(x0 , r),
 2 
∂ ∂2
ψxx (x, y) + ψyy (x, y) = + [ψ(x, y)]
∂x2 ∂y2

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 367 — #27

i i

Some Interesting Theorems 367

 
∂2 ∂2
= + 2 [U(x, y) − U(x, −y)]
∂x2 ∂y
= Uxx (x, y) + Uyy (x, y) − [Uxx (x, −y) + Uyy (x, −y)]
= 0 (as U is harmonic).
Therefore, ψ is a harmonic function on B(x0 , r). Clearly,  has continuous
partial derivatives with respect to x and y. Therefore, to prove  is analytic,
we verify C–R equations only.  
∂ ∂ ∂ψ ∂ 2ψ
(Re ) = =
∂x ∂x ∂x ∂x2
∂ 2ψ
= − 2 (as ψ is harmonic)
∂y
 
∂ ∂ψ ∂
= − = (Im )
∂y ∂y ∂y
and  
∂ ∂ ∂ψ ∂ 2ψ
(Re ) = =
∂y ∂y ∂x ∂y∂x  
∂ 2ψ ∂ ∂ψ
= = − −
∂x∂y ∂x ∂y
∂
= − (Im ) .
∂x
Therefore, our claim holds.
∂ψ
Since ψ(x) = 0, ∀x ∈ B(x0 , r) ∩ σ ⊆ R, we have = 0, ∀x ∈ B(x0 , r).
∂x
∂ψ ∂
Furthermore, = [U(x, y) − U(x, −y)] = Uy (x, y) + Uy (x, −y) which
∂y ∂y
∂ψ
implies that (x, 0) = 2Uy (x, 0) = −2Vx (x, 0) = 0, ∀(x, 0) ∈ B(x0 , r) ∩
∂y
σ , as V is a harmonic conjugate of U and V = 0 on σ . Thus,  vanishes
on B(x0 , r) ∩ σ , and hence, it vanishes on B(x0 , r), by analytic continuation.
∂ψ ∂ψ
Hence, = = 0 on B(x0 , r). Therefore, ψ is constant on B(x0 , r). As
∂x ∂y
x = x, ∀x ∈ σ , we have ψ(x) = U(x) − U(x) = 0, for all x ∈ B(x0 , r) ∩ σ , and
hence, we obtain ψ(z) = 0, ∀z ∈ B(x0 , r). Thus, U(z) = U(z), ∀z ∈ B(x0 , r).
Note that for each x0 ∈ σ , we have found a neighbourhood B(x0 , r) in
which U is a harmonic conjugate with the following properties:
(1) U(z) + iV (z) is analytic on B(x0 , r),
(2) f (z) = U(z) + iV (z), on B(x0 , r) ∩ + ,
(3) U(z) = U(z), ∀z ∈ B(x0 , r).

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 368 — #28

i i

368 Riemann Mapping Theorem

From the properties (2) and (3) of U , we can also have

f (z) = U(z) + iV (z) = U(z) + iV (z) = f (z), ∀z ∈ B(x0 , r) ∩ − .
Therefore, if xj ∈ σ and Uj harmonic conjugate of Vj in B(xj , rj ) for some rj >
0, j = 1, 2 with B(x1 , r1 )∩B(x2 , r2 )  ∅, then U1 (z)+iV (z) = f (z) = U2 (z)+
iV (z), ∀z ∈ B(x1 , r1 ) ∩ B(x2 , r2 ) ∩ + . Therefore, by analytic continuation,
U1 (z) + iV (z) = f (z) = U2 (z) + iV (z), ∀z ∈ B(x1 , r1 ) ∩ B(x2 , r2 ).
Therefore, F is a well-defined analytic function on , satisfying the desired
properties. 

7.4 RIEMANN MAPPING THEOREM

Definition 7.4.1 Let ( fn ) be a sequence of analytic functions on a region 
and let A ⊆ . Then, we say that ( fn ) is uniformly bounded on A if there
exists M > 0 such that | f (z)| ≤ M , ∀z ∈  and ∀n ∈ N.

THEOREM 7.4.2 (Montel’s theorem)

Let ( fn ) be a sequence of analytic functions on a region . If ( fn ) is uniformly
bounded on every compact subset of , then there exists a subsequence of
( fn ), which converges uniformly on every compact subset of .

Proof: Fix w ∈ . Choose δ(w) > 0 such that Cl B (w, δ(w)) ⊆ . First,
we show that
 we can find a subsequence that converges uniformly on
δ(w)
B w, . Now applying Theorem 5.1.5, we can expand each fm as
2
∞ (n)
  
fm (w) δ(w)
fm (z) = (z − w) , ∀z ∈ Cl B w,
n
.
n! 2
n=0
   
δ(w)
If M = sup | fm (z)| : z ∈ Cl B w, , m ∈ N , then M < +∞ and by
2
Cauchy’s estimate, we have
 
 f (n) (w)  M2n
m 
 ≤ , ∀m ∈ N.
 n!  (δ(w))n

As ( fm (w)) is a bounded sequence in C, there exists a subsequence ( fm,0 ) of

( fm ) such that
( fm,0 (w)) := f1,0 (w), f2,0 (w), f3,0 (w), . . .

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 369 — #29

i i

Some Interesting Theorems 369

  (w)

fm,0
converges. Using the fact that is bounded, then there exists a
1!
subsequence ( fm,1 ) of ( fm,0 ) such that
    (w) f  (w) f  (w)
fm,1 (w) f1,1 2,1 3,1
:= , , ,...
1! 1! 1! 1!

converges. Proceeding further, we get that for every n ∈ N, there exists a

subsequence ( fm,n ) of ( fm,n−1 ) such that
 (n)
 (n) (n) (n)
fm,n (w) f1,n (w) f2,n (w) f3,n (w)
:= , , ,...
n! n! n! n!

converges. Then, the diagonal subsequence ( fm, m ) of ( fm ),

 (k)
 (k) (k) (k)
fm, m (w) f1,1 (w) f2,2 (w) f3,3 (w)
:= , , ,...
k! k! k! k!

converges, for every k = 0, 1, 2, 3, .. . . Now,we claim that if (gk ) = ( fm, m ),

δ(w)
then (gk ) converges uniformly on B w, . For a given  > 0, we choose
2
N1 , N2 ∈ N such that

   n!2n−1
1  (n) (n) 
< and (g − g )(w)  < , ∀k, j ≥ N2 , ∀1 ≤ n ≤ N1 .
2N1 4M j k (N1 + 1)(δ(w))n
 
δ(w)
For j, k ≥ N2 and z ∈ B w, , we have
2
∞ 
 (g − g )(n) 
 j k 
|gj (z) − gk (z)| =  (z − w)n 
 n! 
n=0
N1 
   
  ∞  (g − g )(n) 
 (gj − gk )
(n)
n  j k n
≤  (z − w)  +  (z − w) 
 n!   n! 
n=0 n=N1 +1

N1 ∞
2n−1 2M
≤ |z − w| +n
|z − w|n
(N1 + 1)(δ(w))n (δ(w))n
n=0 n=N1 +1
∞
 2M  1
< + N +1
2 2 1 2n
n=0

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 370 — #30

i i

370 Riemann Mapping Theorem

 4M
= + N +1
2 2 1
 
< + = .
2 2
Since C is second countable (Theorem 1.4.48), we can choose a sequence
∞ δ(wj )
of points wj ∈  such that  = ∪ B wj , 2 . Given a sequence of
j=1
functions ( fm ), which is uniformly bounded on compact sets, we find a
subsequence (φn, 1 ) of ( fm ), which converges uniformly on B w1 , δ(w2 1 ) .
Next, we find a subsequence (φn, 2 ) of (φn, 1 ), which converges uniformly on
B w2 , δ(w2 2 ) , and so on. The diagonal subsequence (φn, n ) converges uni-
δ(w )
formly on B wj , 2 j , ∀j ∈ N. Thus, (φn, n ) converges uniformly on each
compact subset of  because every compact subset of  is contained in a
δ(w )
finite union of the B wj , 2 j . 

LEMMA 7.4.3 (Koebe’s lemma)

Let  be a simply connected region such that 0 ∈  ⊆ B(0, 1). Then, there
exists a one-to-one analytic function g from  into B(0, 1) such that g(0) = 0
and |g(z)| > |z|, ∀z ∈ \{0}.

Proof: For a given ζ ∈ B(0, 1)\, we choose ξ ∈ C such that ξ 2 = −ζ .

 z−w
Define g1 (z) = T−ζ [T−ξ (z)]2 , ∀z ∈ C, where Tw (z) = , ∀z ∈
1 − wz
C. Now, g1 (0) = T−ζ (ξ ) = T−ζ (−ζ ) = 0. We note that g1 is not one-to-
2

one because using the bijectiveness of T−ξ on C∞ , we can find a, b1 , b2 ∈ C

such that a  0, b1  b2 , T−ξ (b1 ) = a, and T−ξ (b2 ) = −a, so that
g1 (b1 ) = g1 (b2 ). Therefore, there is no constant c ∈ C such that g1 (z) =
cz, ∀z ∈ C. Applying Schwarz’s lemma, we have |g1 (z)| < |z|, ∀z ∈ . As
Tζ is nowhere zero on  and it is analytic, using Corollary 4.5.13, there exists
an analytic function g2 on  such that g22 = Tζ and g2 (0) = ξ . (By choosing
a suitable value of log(−ζ ). This is possible because [g2 (0)]2 = Tζ (0) = −ζ
and ξ 2 = −ζ ). Define g(z) = Tξ (g2 (z)) , ∀z ∈ . Then, g is analytic on 
and
g(0) = Tξ (g2 (0)) = Tξ (ξ ) = 0.
As Tw−1 = T−w , ∀w ∈ C, for z ∈ \{0}, we have g1 (z)  0 and
|g(z)| >|g1 (g(z))|
  
 2
= T−ζ [T−ξ ◦ Tξ ◦ g2 ](z) 

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 371 — #31

i i

Some Interesting Theorems 371

 
 
= T−ζ (g2 (z))2 
  
= T−ζ Tζ (z) 
= |z|.

Thus, the lemma follows. 

THEOREM 7.4.4 (Riemann mapping theorem)

If   C is a simply connected region, then there exists an analytic bijection
from  onto B(0, 1).

Proof: Let ζ ∈ C\ be fixed. If

√
f1 (z) = z, z ∈ C\{(x, 0) : x ≤ 0}
√
is the principal branch of · and

f2 (z) = z − a, ∀z ∈ ,

then f1 ◦ f2 is analytic and one-to-one on . Furthermore,

 if ξ ∈ ( f1 ◦ f2 )(),
then the principal argument of ξ should be in − π2 , π2 , and hence, −ξ 
( f1 ◦ f2 )(). By open mapping theorem (Theorem 4.5.4), ( f1 ◦ f2 )() turns to
be an open set, and hence, for a given ξ0 ∈ ( f1 ◦ f2 )(), there exists  > 0
such that B(ξ0 , ) ⊆ ( f1 ◦ f2 )(). Since B(−ξ0 , ) = −B(ξ0 , ), no point
of B(−ξ0 , ) lies in ( f1 ◦ f2 )(). Therefore, for every z ∈ , ( f1 ◦ f2 )(z) 
B(−ξ0 , ), and hence, |( f1 ◦ f2 )(z) + ξ0 | ≥ . Hence, it follows that if

1
f3 (z) = , ∀z ∈ ,
( f1 ◦ f2 )(z) + ξ0

then f3 :  → Cl B 0,  −1 is analytic and injective on . For a fixed z0 ∈ ,
if
( f3 (z) − f3 (z0 ))
f (z) = , ∀z ∈ ,
3
then f is analytic on , and f () is a region containing 0, by open mapping
theorem (Theorem 4.5.4) and by Theorem 1.6.29.  Since f3 is injective on ,
f is also injective on . As f3 :  → Cl B 0,  −1 , we have f :  →
Cl B 0, 23 r . Thus, f :  → G is an analytic bijective map, where G =
f ()  B(0, 1) is a region containing 0.
Let H = {h : G → B(0, 1) : h be analytic injective with h(0) = 0}.
Let a ∈ G\{0} be arbitrarily fixed. Using f ∈ H , we have H  ∅. If

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 372 — #32

i i

372 Riemann Mapping Theorem

M = sup |h(a)|, then 0 < M (since 0  f ∈ H ) and M ≤ 1 (as |h(a)| <

h∈H
1, ∀z ∈ G, ∀h ∈ H ). From the definition of M , we can find a sequence {hn }
from H such that
|hn (a)| → M as n → ∞.
As {hn } is uniformly bounded on every compact subset of G, by Montel’s
theorem (Theorem 7.4.2), there exists an analytic function h on G such that
hnj → h as j → ∞ for some subsequence {hnj } of {hn }. Using the fact
that each hnj is analytic, injective and hnj (0) = 0, we observe that h is also
analytic (by Theorem 4.3.4), injective and h(0) = 0. For every z ∈ G, h(z) =
lim hnj (z) ∈ Cl B(0, 1). Furthermore,
j→∞

|h(a)| = lim |hnj (a)| = M > 0

j→∞

implies that h is non-constant. Therefore, by open mapping theorem, h(G)

should be an open subset, and hence, no point of G should be in {w ∈ C :
|w| = 1}. Therefore, h : G → B(0, 1), and hence, h ∈ H . Now, we claim that
h : G → B(0, 1) is onto. Suppose h(G)  B(0, 1), then by Koebe’s lemma,
there exists an analytic injective function g : h(G) → B(0, 1) such that

g(0) = 0 and |g(z)| > |z|, ∀ ∈ f (G).

Then, (g ◦ h) : G → B(0, 1) belongs to H and

|(g ◦ h)(z)| > |h(z)| = M,

which contradicts the choice of M . Hence, h is onto. Therefore, h ◦ f :  →

B(0, 1) is the required analytic bijection. Thus, the proof is completed. 
The map constructed in the previous theorem is usually called a Riemann
mapping from  onto B(0, 1).

RESULT 7.4.5 (Uniqueness of Riemann mapping)

If Fj :  → B(0, 1), j = 1, 2 are two Riemann mappings such that Fj (z0 ) = 0,
j = 1, 2, and arg (F1 (z0 )) = arg (F2 (z0 )), for some z0 ∈ , then F1 = F2
on .

Proof: By Corollary 4.5.5, F2−1 : B(0, 1) →  is an analytic bijection. Then,

clearly F1 ◦ F2−1 : B(0, 1) → B(0, 1) is an analytic bijection, satisfying

(F1 ◦ F2−1 )(0) = F1 (F2−1 )(0) = F1 (z0 ) = 0.

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 373 — #33

i i

Some Interesting Theorems 373

Hence, by Corollary 4.5.10, we get (F1 ◦ F2−1 )(z) = cz, ∀z ∈ B(0, 1) for some
c ∈ C with |c| = 1. For a given z ∈ , let w = F2 (z) so that F1 (z) = (F1 ◦
F2−1 )(w) = cw = cF2 (z). Using arg (F1 (z0 )) = arg (F2 (z0 )) and F1 (z0 ) =
cF2 (z0 ), we get arg (c) = 0. Thus, c = 1. Hence the result follows. 

7.5 SCHWARZ–CHRISTOFFEL FORMULA

Note that the Riemann mapping theorem states the existence of an analytic
bijection between a simply connected region and the open unit ball; for a
practical purpose, this theorem does not give any procedure for construct-
ing such a function. Such a formula is obtained by Schwarz and Christoffel
separately. In this section, we shall see how this formula is obtained.

Definition 7.5.1 Let  be a region. We say that a sequence {zn } in  tends

to the boundary ∂ of ; if for a given z ∈ , then there exist z > 0 and
Nz ∈ N such that
|zn − z| ≥ z , ∀n ≥ Nz .
We write this by zn → ∂ as n → ∞.

Definition 7.5.2 Let  be a region and γ be a curve in  given by γ (t), t ∈

[a, b]. We say that a curve γ in  tends to the boundary ∂ of ; if for a
given z ∈ , there exist z > 0 and tz ∈ [a, b) such that
|γ (t) − z| ≥ z , ∀t ≥ tz .
In this case, we write γ (t) → ∂ as t → b.

LEMMA 7.5.3 Let {zn } be a sequence in a region . Then, zn → ∂ as

n → ∞ iff for each compact set K ⊆ , there exists NK ∈ N such that
zn  K , ∀n ≥ NK .

Proof: Assume that zn → ∂ as n → ∞ and K ⊆  is the given compact

set. For each z ∈ K , there
 exists z > 0 and Nz ∈ N such that |zn − z| ≥
z , ∀n ≥ Nz . Since {B x, 2x : x ∈ K} is a collection of open sets satisfying

K ⊆ ∪ B x, 2x and K is compact, there exist x1 , x2 , . . . , xn ∈ K such that
x∈K
n xj
K ⊆ ∪ B xj , 2 . If NK = max Nxj , then for each x ∈ K and for n ≥ NK ,
j=1 1≤j≤n
there exists j ∈ {1, 2, . . . , n} such that x ∈ B(xj , xj ) so that
x j
|x − zn | ≥ |zn − xj | − |xj − x| > xj − > 0.
2

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 374 — #34

i i

374 Schwarz–Christoffel Formula

Thus, zn  K , ∀n ≥ NK .
Conversely, if for each compact set K ⊆ , there exists NK ∈ N such
that zn  K , ∀n ≥ NK , then for a given z ∈ , we choose r > 0 such
that Cl B(z, r) ⊆ . Clearly, Cl B(z, r) is a compact subset of , then by
assumption, there exists  > 0 and N ∈ N such that zn  Cl B(z, r), ∀n ≥ N ,
and hence, |z − zn | >  . Thus, zn → ∂ as n → ∞. 

THEOREM 7.5.4 Let f be a homeomorphism3 of a region  onto another

region G. Then, zn → ∂ as n → ∞ iff f (zn ) → ∂G as n → ∞.

Proof: Assume that zn → ∂ as n → ∞. For a given compact set K of G,

using the continuity of f −1 , we conclude that f −1 (K) is compact (by Theorem
1.6.27). Then, by hypothesis, there exists N ∈ N such that zn  f −1 (K) for all
n ≥ N . Since f is a bijection, we have f (zn )  K , ∀n ≥ N . Thus, f (zn ) → ∂G
as n → ∞.
Conversely, assume that f (zn ) → ∂G as n → ∞. For a given compact set
T of , we have f (T), which is compact in G. Therefore, there exists N ∈ N
such that f (zn )  f (T) for all n ≥ N . Thus, zn  T , ∀n ≥ N .
Similarly, we can have the following theorem. 

THEOREM 7.5.5 If f :  → G is a homeomorphism and γ (t), t ∈ [a, b] is a

curve in , then γ (t) → ∂ as t → b iff f (γ (t)) → ∂G as t → b.

Definition 7.5.6 Let  be a region such that ∂ contains a line segment γ

with a parametric equation γ (t) = (1 − t)ζ + tξ , t ∈ (0, 1). (Note that we use
open interval (0, 1) and not [0, 1] in the parametric equation.) We say that γ
is a free boundary arc if for every ζ ∈ γ and for a sufficiently small r > 0,
then B(ζ , r) ∩ ∂ is an actual diameter4 of B(ζ , r).
If γ is a free boundary arc, for ζ ∈ γ , then there exists r0 > 0 such that
B(ζ , r) ∩ γ is a diameter of B(ζ , r) for every r ∈ (0, r0 ). Hence, the two open
half-discs (say H1 and H2 ) determined by this diameter does not intersect ∂.
Then, we have two possible cases.

Case 1. H1 ⊆  and H2 ⊆ c .

Case 2. H1 ⊆  and H2 ⊆ .

See the following diagrams, to understand the above possibilities.

3 Homeomorphism is a continuous bijection with continuous inverse.
4 Here, we mean the diameter by any chord passing through the centre of the disc.

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 375 — #35

i i

Some Interesting Theorems 375

H1 g

H2
H1 H2

g
Ω Ω

Definition 7.5.7 Let ζ ∈ γ and r > 0 be such that B(ζ , r) ∩ ∂ is a diameter

of B(ζ , r).
1. If one of the two half-open discs determined by the diameter of B(ζ , r)
lies in  and the other half-disc lies in (Cl )c , then ζ is called a one-
sided free boundary point of .
2. If both of the half-open discs determined by the diameter of B(ζ , r) lie
in , then ζ is called a two-sided free boundary point of .

Definition 7.5.8 Let γ be a free boundary arc in . Then,

1. ζ is a one-sided free boundary point of , ∀ζ ∈ γ iff γ is called a
one-sided free boundary arc.
2. ζ is a two-sided free boundary point of , ∀ζ ∈ γ iff γ is called a
two-sided free boundary arc.

LEMMA 7.5.9 Let γ be a free boundary arc in . Then,

1. γ is a one-sided free boundary arc of  iff γ has a one-sided free
boundary point.
2. γ is a two-sided free boundary arc  iff γ has a two-sided free
boundary point.

Proof: As the first statement of this theorem is equivalent to the second state-
ment, we prove only the first statement. Obviously, it follows that if γ is a
one-sided free boundary arc, then γ has a one-sided free boundary point.
Conversely, assume that γ has a one-sided free boundary point ζ . By defi-
nition of free boundary arc, there exists r > 0 such that B(ζ , r) ∩ ∂ is a
diameter of B(ζ , r), and by the definition of one-sided free boundary point
of , if H1 and H2 are the two half-open discs determined by the diameter,

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 376 — #36

i i

376 Schwarz–Christoffel Formula

then let H1 ⊆ , H2 ⊆ (Cl )c . Then, for every ξ ∈ B(ζ , r) ∩ ∂ and for
every s ∈ (0, r − d(ζ , ξ )), we have B(ξ , s) ∩ ∂, which is a diameter of B(ξ , s).
Furthermore, one of the two half-open discs determined by this diameter lies
in H1 ⊆  and the other half-open disc lies in H2 ⊆ (Cl )c .

H1

z
x
H2

Thus, we have proved that if A = {ζ ∈ γ : ζ is a one-sided free boundary

point}, then A is an open subset of γ . By the same argument, we can show
that if B = {ζ ∈ γ : ζ is a two-sided free boundary point}, then B is an open
subset of γ . However, γ is connected (begin a continuous image of (0, 1)),
γ = A ∪ B and A ∩ B = ∅. Given that A  ∅. Therefore, B = ∅ (otherwise,
 is disconnected), and hence, A = γ . 

THEOREM 7.5.10 Let  be a simply connected region such that ∂ contains

a one-sided free boundary arc γ . If f is a Riemann mapping of  onto B(0, 1),
then there exist a simply connected region 0 containing  ∪ γ and an
analytic function F0 on 0 such that F0 (z) = f (z), ∀z ∈ , and F0 is one-to-
one on γ and F0 (γ ) ⊆ ∂B(0, 1) = {w ∈ C : |w| = 1}.

Proof:

Case 1. γ ⊆ R and  is a subset of the upper half-plane.

Using that f :  → B(0, 1) is a bijection and 0 ∈ B(0, 1), we find a
unique point z0 ∈  such that f (z0 ) = 0. As γ is a one-sided free
boundary arc, for each x0 ∈ γ , find r > 0 such that z0  B(x0 , r)
and B(x0 , r) ∩ σ is a diameter of B(x0 , r), and let H be the semi-open
disc B(x0 , r) ∩ . Since H is a simply connected region and f  0 on
H , log( f ) can be suitably defined as an analytic function on H (see
Theorem 4.5.12). Let (zn ) be a sequence in H such that zn → γ as

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 377 — #37

i i

Some Interesting Theorems 377

n → ∞ be arbitrary. Using Theorem 7.5.4, we get | f (zn )| → 1 as

n → ∞. Therefore,

Re log( f (zn )) → 0 as n → ∞.

If g = i log( f ), then Im g(zn ) = log(| f |)(zn ) → 0 as n → ∞,

whenever zn → γ as n → ∞. Therefore, using Schwarz reflec-
tion principle (Theorem 7.3.1), there exists an analytic function G
on B(x0 , r) such that G(z) = g(z), ∀z ∈ H and G(z) = G(z),
∀z ∈ B(x0 , r). If F = exp(−iG), then F is analytic on B(x0 , r)
such that F(z) = f (z), ∀z ∈ H . As x0 ∈ γ is arbitrary, for each
x ∈ γ , there exists rx > 0, and analytic function Fx is analytic
on B(x,
 rx ) and F x (z) = f (z), ∀z ∈ B(x, rx ) ∩ . We take 0 =

∪ ∪ B(x, rx ) . Clearly, 0 is a simply connected region con-

x∈γ
taining  ∪ γ . If x1 , x2 ∈ γ such that B(x1 , rx1 ) ∩ B(x2 , rx2 )  ∅, then
Fx1 (z) = Fx2 (z) = f (z), ∀z ∈ B(x1 , rx1 ) ∩ B(x2 , rx2 ) ∩ . Therefore,
by analytic continuation, we have Fx1 = Fx2 on B(x1 , rx1 )∩B(x2 , rx2 ).
Therefore, if we define F0 : 0 → C by

f (z) if z ∈ 
F0 (z) =
Fx (z) if z ∈ B(x, rx ),

then F0 is a well-defined analytic function on 0 with F0 (z) = f (z),

∀z ∈ . Since F0 (z) = f (z) on  and

| f (zn )| → 1 as n → ∞, whenever zn → γ as n → ∞, (7.8)

we have F0 (γ ) ⊆ ∂B(0, 1).

To conclude the proof of this theorem, it remains to show that F0 is
one-to-one on γ . Suppose there exist z1 , z2 ∈ γ such that F(z1 ) =
F(z2 ). Then, choose a simple curve μ (with a parametric equation
μ(t), t ∈ [a, b]) in 0 such that μ(t) ∈ , ∀t ∈ (a, b), μ(a) = z1 and
μ(b) = z2 .

Ω
F(m)
F(z) F(z1) = F(z2)
R1
Ω2
m (0, 0)
R2
Ω1
z2 g z1

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 378 — #38

i i

378 Schwarz–Christoffel Formula

Clearly, μ∪Lz2 ,z1 is a simple closed curve in ∪γ and F(μ)∪{F(z1 )}

is also a simple closed curve in {w ∈ C : |w| ≤ 1}; because F(μ \
{z1 , z2 }) = f (μ \ {z1 , z2 }) ⊆ B(0, 1), f is one-to-one and F(z1 ) =
F(z2 ) ∈ ∂B(0, 1). The simple curve μ in  determines two regions
1 and 2 such that  \ μ = 1 ∪ 2 (see the diagram). Similarly,
the closed curve F(μ) ∪ {F(z1 )} determines two regions R1 and R2
in B(0, 1) such that B(0, 1) \ f (μ) = R1 ∪ R2 . As f :  → B(0, 1) is a
continuous bijection, we have f (1 ) ∪ f (2 ) = R1 ∪ R2 , and f (1 )
and f (2 ) are connected subsets of B(0, 1) \ f (μ). Therefore, either
f (1 ) = R1 or f (1 ) = R2 . In both cases, (Cl f (1 )) ∩ ∂B(0, 1) =
{F(z1 )} (see the above diagram).
If z ∈ γ is such that z lies between z1 and z2 , then choose a sequence
(zn ) in 1 such that zn → z as n → ∞. This is possible because
z ∈ Cl 1 (see the diagram). By Theorem 7.5.4, we have | f (zn )| →
1 as n → ∞, and by the continuity of f , we also have | f (zn )| →
| f (z)| as n → ∞. Therefore, | f (z)| = 1. As f (z) ∈ (Cl f (1 )) ∩
∂B(0, 1) = {F(z1 )}, we must have f (z) = F(z) = F(z1 ). Hence,
F is a constant function on the line segment Lz2 ,z1 joining z2 and z1 .
Thus, by analytic continuation, F becomes constant on 0 ⇒ F = f ,
which is constant on . This is a contradiction to the bijectiveness
of f . Therefore, F is one-to-one on γ .

Case 2. For an arbitrary γ and an arbitrary  as in the statement of the

theorem, first we find a linear fractional transform S such that

1. S maps the straight line passing along γ onto the real axis,
2. the entire region  is mapped into the upper half-plane,
3. its unique finite pole lies in the complement of Cl .

If S = S(), then S is a region such that S(γ ) is a part of its boundary on

the real axis. If g = f ◦ S −1 on S , then g is an analytic function on S , and g
and S are suitable for applying Case 1 of proof of this theorem. Therefore,
we get an analytic function G on a region S,0 containing S ∪ S(γ ) such
that G(z) = g(z), ∀z ∈ S . Thus, we take 0 = S −1 (S,0 ) and the required
analytic function as F = G ◦ S on 0 . 

Remark 7.5.11: We point out that according to the notations in the proof of
previous theorem, |F0 (z)| > 1, ∀z ∈ 0 \ Cl .
We justify the above statement for the case γ ⊆ R, and the general case
follows by applying a similar technique used in Case 2 of proof of previous

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 379 — #39

i i

Some Interesting Theorems 379

theorem. If z ∈ B(x, rx ) ∩ (Cl )c for some x ∈ γ , then F0 (z) = Fx (z), and
by the construction of Fx ( from the previous proof), Fx (z) = exp(−iG(z)) for
some analytic function G on B(x, rx ), thereby satisfying

G(z) = exp(i log( f )), ∀z ∈ B(x, rx ) ∩  and G(z) = G(z), ∀z ∈ B(x, rx ).

Therefore, for z ∈ B(x, rx ) ∩ (Cl )c , find z1 ∈ B(x, rx ) ∩  such that z1 = z.

We have

|F(z)| = exp(−iG(z))
= | exp(−iG(z1 ))|
= | exp(−iG(z1 ))|
= | exp(−i[i log( f (z1 ))])|
= | exp(− log(f (z1 )))|
= exp(Re [− log(f (z1 ))])
= exp(log(|f (z1 )|)−1 )
= | f (z1 )|−1 > 1, as f :  → B(0, 1) and z1 ∈ .

COROLLARY 7.5.12 If f is a Riemann mapping of B(a, r) onto B(0, 1) and

μ is an circular arc on the boundary of B(a, r) with a parametric equation
a + r exp(iθ ), θ1 < θ < θ2 for some θ1 , θ2 ∈ R, then there exist a simply
connected region U0 containing B(a, r) ∪ μ and an analytic function F0 on
U0 such that F0 (z) = f (z), ∀z ∈ B(a, r), where F0 is one-to-one on μ and
F0 (μ) ⊆ ∂B(0, 1).
Proof of this corollary is similar to the proof of Theorem 7.5.10 with a
small change. Here, first we choose a linear fractional transform, which maps
the unit circle to the real line (as in Case 2 of proof of Theorem 7.5.10), and
then, we follow the proof of Case 1 as in the proof of the same theorem.

THEOREM 7.5.13 Let  be a simply connected region enclosed by the

n−1
given closed polygon ∪ Lzj ,zj+1 ∪ Lzn ,z1 . If the interior angle at zj is αj π ,
j=1
∀j = 1, 2, . . . , n, then, the unique Riemann mapping f :  → B(0, 1)
can be extended as a homeomorphism F : Cl  → Cl B(0, 1) such that
F(zj ) = wj for some wj with |wj | = 1, and F(Lzj ,zρ(j) ) (including the end
points) is the circular arc joining wj and wρ(j) (including the end points),
where ρ : {1, 2, . . . , n} → {1, 2, . . . , n} is the bijection defined by ρ(j) =

j + 1 if j < n
, ∀j = 1, 2, . . . , n so that F(∂) = ∂B(0, 1).
1 if j = n

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 380 — #40

i i

380 Schwarz–Christoffel Formula

Proof: If γj = {(1 − t)zj + tzρ(j) : (0, 1)} and ρ(j) is as in the statement of
the theorem, ∀j = 1, 2, . . . , n, then γj is a one-sided free boundary arc. (Note
that γj does not include the end points zj and zρ(j) , whereas Lzj ,zρ(j) includes
the end points zj and zρ(j) .) Then, applying the previous theorem repeatedly,
f can be extended as an analytic function G on a region containing  and
γj , ∀j = 1, 2, . . . , n and G(γj ) ⊆ ∂B(0, 1). Note that G is not defined at zj ,
∀j = 1, 2, . . . , n. Fix j ∈ {1, 2, . . . , n} arbitrarily. Now, find rj > 0 such that
zk  B(zj , rj ) if j  k ∈ {1, 2, . . . , n}. Denote B(zj , rj ) ∩  by Sj . As z − zj  0,
∀z ∈ Sj , log(z − zj ) is defined as an analytic function on Sj , and hence, if

ζ = φj (z) = (z − zj )1/αj = exp((1/αj ) log(z − zj )), ∀z ∈ Sj ,

then φj : Sj → Hj is an analytic bijection, where
 
1/α θj θj
Hj = rj j exp(iθ) ∈ C : <θ < + π , where θj = arg (zρ(j) − zj ),
αj αj
1/αj
an open semi-disc with centre 0 and radius rj . (Because
z ∈ Sj ⇔ θj < arg (z − zj ) < θj + αj π and |z − zj | < rj .
θj θj 1/α
⇔ π < arg (z − zj )1/αj < + π and |z − zj |1/αj < rj j .
αj αj
⇔ φj (z) ∈ Hj ).
Then, the inverse of φj is also an analytic function, which is given by

z = φj−1 (ζ ) = zj + ζ αj = zj + exp(αj log(ζ )), ∀ζ ∈ Hj .

w3
w4

F w5 0 w2
1
w6 w1
z5
a5 z4 y3
z6
a6 a4
f Y3
a1 Ω
z1
a3
a2 S3 1/a3
z3 0
r3 r3
z2

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 381 — #41

i i

Some Interesting Theorems 381

Define ψj : Hj → C by ψj = f ◦ φj−1 . Then, ψj is analytic on Hj . If ζ tends

to the diameter of Hj , then z → [Lzρ −1 (j) ,zj ∪ Lzj ,zρ(j) ] ∩ B(zj , rj ) (that is, z tends
to the union of two line segments in the boundary of Sj ).
⇒ f (z) → ∂U (by Theorem 7.5.4) ⇒ |ψj (ζ )| = | f (φj−1 (ζ ))| = | f (z)| → 1.
Therefore, as in the proof of previous theorem, by reflection principle,
1/α
there exists an analytic function j : B(0, rj j ) → C such that j (ζ ) =
ψj (ζ ), ∀ζ ∈ Hj . Now, we define F : Cl  → C by

G(z) if z ∈ (Cl ) \ {zj : j = 1, 2, . . . , n}
F(z) =
j (0) if z = zj , j = 1, 2, . . . , n.
To prove F is continuous on Cl , it is sufficient to check its continuity at
zj , ∀j = 1, 2, . . . , n. Let j ∈ {1, 2, . . . , n} be arbitrary and (xm ) be a sequence
from Cl  such that xm → zj as m → ∞. We first choose M ∈ N such that
xM+m ∈ B(zj , rj ), ∀m ∈ N.
Case 1. xM+m ∈ , ∀m ∈ N.
Using lim φj (z) = 0, we get φj (xM+m ) → 0 as m → ∞. As j
z→zj
is continuous at 0, we have j (φj (xM+m )) → j (0) as m → ∞.
However,
j (φj (xM+m )) = ψj (φj (xM+m )) = f (xM+m ) = G(xM+m ) = F(xM+m ),
and hence, F(xM+m ) → j (0) = F(zj ) as m → ∞.
Case 2. xM+m ∈ γj ∩ B(zj , rj ), ∀m ∈ N.
Since G is continuous on γj for each m ∈ N, there exists 0 < δm < m1
such that
1
|z − xM+m | < δm ⇒ |G(z) − G(xM+m )| < .
m
As γj ⊆ ∂Sj , we choose uM+m ∈ B(xM+m , δm ) ∩ Sj ⊆ , ∀m ∈ N.
Using
|uM+m − zj | ≤ |uM+m − xM+m | + |xM+m − zj |
1
≤ + |xM+m − zj | → 0 as m → ∞,
m
it follows that uM+m → zj as m → ∞. Therefore, by Case 1, we
have F(uM+m ) → F(zj ) as m → ∞. Therefore,
|F(xM+m ) − F(zj )| ≤ |F(xM+m ) − F(uM+m )| + |F(uM+m ) − F(zj )|
1
≤ + |F(uM+m ) − F(zj )| → 0 as m → ∞.
m

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 382 — #42

i i

382 Schwarz–Christoffel Formula

Case 3. xM+m ∈ γρ −1 (j) ∩ B(zj , rj ), ∀m ∈ N.

This case is similar to the previous case.

Thus, F is continuous on Cl . If we denote j (0) = wj , then clearly we have

F(zj ) = wj and |wj | = 1, as F(zj ) = f (zj ) = lim zj,m for some zj,m ∈  such
m→∞
that zj,m → zj as m → ∞, and by Theorem 7.5.4, | f (zj,m )| → 1 as m → ∞,
∀j = 1, 2, . . . , n. Therefore, F(∂) ⊆ ∂B(0, 1). As ∂ is a closed curve in
Cl  whose image under the continuous function F should be a closed curve
in ∂B(0, 1), there are only two possibilities:

1. F(∂) = ∂B(0, 1).

2. F(∂) is a circular arc contained in ∂B(0, 1), which is traversed in both

directions as in the following diagram.

wr ( j)

0
1

wj

(In the above diagram, actually there is a single circular arc from wj and wρ(j) ,
which is traversed in both directions. To make it clear, we have drawn the arc
using double lines.)
We claim that the second case is not possible because in the second case,
we have

0 = WN (F(∂), 0)

1 dw
=
i2π w
F(∂)

1 f  (z)
= dz
i2π f (z)
∂
(by the change of variable w = f (z))

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 383 — #43

i i

Some Interesting Theorems 383

= WN (∂, z0 ), where z0 ∈  such that F(z0 ) = 0

(by Theorem 4.5.2 )
= 1,
which is a contradiction. Therefore, our claim holds. Therefore, the images
of γj and γk under F are pairwise disjoint and wj  wk if j  k . Hence,
F : Cl  → Cl B(0, 1) is a bijection. As Cl  is a closed and bounded
subset of C, it is a compact set. Therefore, for every open set E of Cl ,
(Cl ) \ E is closed, and hence, it is compact, whose image under the contin-
uous function F is also a compact subset of Cl B(0, 1), by Theorem 1.6.27.
Therefore, F((Cl ) \ E) = (Cl B(0, 1)) \ F(E) is closed in Cl B(0, 1). There-
fore, (F −1 )−1 (E) = F(E) is open in Cl B(0, 1) whenever E is open in Cl .
Thus, we get that F −1 : Cl B(0, 1) → Cl  is a continuous map, by Theorem
1.6.24, and hence, F : Cl  → Cl B(0, 1) is a homeomorphism. 

THEOREM 7.5.14 (Schwarz–Christoffel formula for open unit disc)

Let  be a bounded simply connected region enclosed by a polygon with
vertices zj and interior angles αj π at zj , j = 1, 2, . . . , n. If f is a Riemann
mapping from  onto B(0, 1), then there exist constants c1 and c2 such that
w 
n
−1
f (w) = c1 (ζ − wj )−βj dζ + c2 ,
0 j=1

where the integral is defined over any curve joining 0 and w in B(0, 1), βj =
1 − αj , wj = lim f (z), ∀j = 1, 2, . . . , n.
z→zj

Proof: From geometry, we first note that sum of the exterior angles of a

n
closed polygon is 2π . That is, βj = 2. We shall use the notations as in
j=1
the proof of previous theorem. By applying Corollary 7.5.12, repeatedly, for
the the circular arc μj joining wj and wρ(j) (which excludes the end points),
j = 1, 2, . . . , n, there exists a simply connected region U0 containing Cl
B(0, 1) \ {wj : 1 ≤ j ≤ n} and an analytic function F on U0 such that
F = F −1 on B(0, 1).
Next, we prove that the function j (0)  0. Suppose that j (0) = 0.
1/α
Then, by local correspondence theorem, there exist 0 < δ < rj j and 0 <
 < 1 such that for each w ∈ B(w, ), there exist at least two distinct points
a, b ∈ B(0, δ) such that j (a) = w = j (b). By Theorem 7.5.10 and by
Remark 7.5.11, we see that

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 384 — #44

i i

384 Schwarz–Christoffel Formula

1/αj
|j | = 1 on the diameter of the disc B(0, rj )

and
|j | > 1 on {z ∈ C : Im z < 0}.

Therefore, a, b ∈ Hj . Observing, ψj : Hj → C is defined by ψj = f ◦ φj−1 and


f is one-to-one on φ −1 Hj , ψj is one-to-one on Hj . Since j = ψj , ∀z ∈
Hj , j is one-to-one on Hj . Therefore, j (a) = j (b) implies that a = b,
which is a contradiction. Hence, our claim holds. Then, again using local
1/α
corresponding theorem, there exist 0 < δj < rj j and 0 < j < 1 such that
j : B(0, δj ) → B(wj , j ) is an analytic bijection. Then, by Corollary 4.5.5,
1
we have j−1 , which is also differentiable at wj and (j−1 ) (wj ) =   0.
j (0)
Observing j−1 (wj ) = 0, we conclude that j−1 has simple zero at wj , and
hence, there exists an analytic function Pj on B(wj , tj ) for some 0 < tj < 
such that Pj (w)  0 and ζ = j−1 (w) = (w − wj )Pj (w), ∀w ∈ B(wj , tj ). Then,
we can write,
α
ζ αj = (j−1 (w))αj = (w − wj )αj Pj j (w), ∀w ∈ B(wj , tj ) \ Lwj ,wj −tj ,

where (j−1 (w))αj , and (w−wj )αj are analytic functions on B(wj , tj )\Lwj ,wj −tj ,
α
and Pj j is an analytic function on B(wj , tj ), each of which is defined by using
exp and principal branch of log.
By the previous theorem, F : Cl  → Cl B(0, 1) is a homeomorphism
such that F(z) = f (z), ∀z ∈ . Therefore, for each w ∈ B(wj , tj ), first we
choose ζ ∈ B(0, δj ) such that j (ζ ) = w, and hence,

F −1 (w) = F −1 (j (ζ )) = (F −1 ◦ F ◦ φ −1 )(ζ ) = φ −1 (ζ ) = zj + ζ αj .

Therefore, we have

F −1 (w) − zj = ζ αj = (w − wj )αj Qj (w), ∀w ∈ B(wj , tj ) \ Lwj ,wj −tj , (7.9)

α
where Qj = Pj j . By differentiating on both sides of equation (7.9), we get

(F −1 ) (w) = αj (w − wj )αj −1 Qj (w) + (w − wj )αj Qj (w)

⇒ (w − wj )1−αj (F −1 ) (w) = αj Qj (w) + (w − wj )Qj (w).
⇒ (w − wj )βj (F −1 ) (w) = αj Qj (w) + (w − wj )Qj (w). (7.10)

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 385 — #45

i i

Some Interesting Theorems 385

We note that the function on the right-hand side of equation (7.10) is an an-
alytic function on B(wj , tj ), whereas the function on the left-hand side is an
analytic function on B(wj , tj ) \ Lwj ,wj −tj . Define Ij : B(wj , tj ) → C by

Ij (w) = αj Qj (w) + (w − wj )Qj (w), ∀w ∈ B(wj , tj ). (7.11)

Using equations (7.10) and (7.11) and by analytic continuation, we have

F  (w)(w − wj ) = Ij (w)(w − wj ), ∀w ∈ B(wj , tj ). (7.12)

n
If V0 = U0 ∪ ∪ B(wj , tj ), then H : V0 → C is defined by
j=1


n
H(w) = F  (w) (w − wj )βj , ∀w ∈ V0 . (7.13)
j=1

We claim that H is analytic on Cl B(0, 1). As |wj | = 1, ∀j = 1, 2, . . . , n,

clearly, each (w − wj )βj is analytic on B(0, 1), and hence, H is analytic on
B(0, 1). If j  k , then (w − wj )βj is analytic on B(wk , tk ) and by equation
(7.12),
F  (w)(w − wk )βk = Iρ (w),
which is analytic on B(wk , tk ). Therefore,
 H is analyticon B(wk , tk ), ∀k =
n
1, 2, . . . , n. If w0 ∈ (Cl B(0, 1)) \ ∪ B(wj , tj ) ∪ B(0, 1) , then |w0 | = 1 but
j=1
w0  B(wj , tj ) for any j. Therefore, w − wj  0 in a neighbourhood of w0
in which (w − wj )βj is an analytic function, ∀j = 1, 2, . . . , n and F  is also
analytic in a neighbourhood of w0 . Thus, our claim follows.
We claim that H is nowhere zero on Cl B(0, 1). Using F = F −1 on
B(0, 1) and F −1 is a one-to-one function, we get that F  is nowhere zero on
Cl B(0, 1). (The proof of this statement is similar to the proof of j (0)  0 at
the beginning of this proof.) Furthermore, w − wj  0, ∀w ∈ B(0, 1) ⇒ (w −
wj )βj  0, ∀w ∈ B(0, 1) ⇒ H(w)  0, ∀w ∈ B(0, 1). Hence, the only possible
zeroes of H are wj , j = 1, 2, . . . , n. However, the value of F  (w)(w − wk )βk at
wk is Iρ (wk ) = αk Qρ (wk ), which is non-zero and (wk − wj )βj  0, if j  k .
Thus, H is nowhere zero on Cl B(0, 1).
Next, we find s > 0 such that H is analytic and nowhere zero on
B(0, 1 + s). For each u ∈ ∂B(0, 1), find su > 0 such that F is analytic
and nowhere zero on B(u, su ). As ∂B(0, 1) is a closed and bounded sub-
set of C, it is a compact set, and hence, there exist u1 , u2 , . . . , uν such that
ν  s !s "
∂B(0, 1) ⊆ ∪ B up , 2p . If s = min 2p : 1 ≤ p ≤ ν , then s > 0, and for
p=1

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 386 — #46

i i

386 Schwarz–Christoffel Formula

each w ∈ B(0, 1 + s) \ Cl B(0, 1), let u = exp(iθ ), where θ = arg w. Then,

s
u ∈ ∂B(0, 1) ⇒ u ∈ B up , 2p for some p ∈ {1, 2, . . . , ν}. As

sp sp sp s p
|w − up | ≤ |w − u| + |u − up | < |w| − 1 + < 1 + r − 1 + ≤ + = sp
2 2 2 2

w ∈ B up , sp , and hence, F is analytic and non-zero at w. Therefore, F is
analytic and nowhere zero on B(0, 1 + s), which is a simply connected region.
Therefore, by Theorem 4.5.12, we can find a suitable branch for log(H), so
that it is analytic on B(0, 1 + s). In particular, with respect to this branch,
arg (H(w)) =Im H(w) is continuous on B(0, 1 + s). If j ∈ {1, 2, . . . , m} and
w ∈ {exp(iθ ) : θj < θ < θρ(j) }, where arg (wj ) = θj and arg (wρ(j) ) = θρ(j) ,
then from equation (7.13), we have


n
arg H(exp(iθ)) = arg (F  (exp(iθ ))) + arg (exp(iθ ) − exp(iθj )). (7.14)
j=1

Since

1. arg F  (exp(iθ)) is the angle between the tangent of the line segment γj
joining zj and zρ(j) at F(exp(iθ)) and the tangent of the circular arc μj
joining wj = exp(iθj ), wρ(j) = exp(iθρ(j) ) at w = exp(iθ ),

2. the slope of the the tangent of line segment γj is constant (say ϕj ) at

each point on γj ,

3. the slope of the tangent of the circular arc μj at exp(iθ ) is θ + π2 ,

we have
π
arg (F  (exp(iθ ))) = ϕj − θ + .
2
As
exp(iθ ) − exp(iθj )
         
θ + θj θ − θj θ − θj
= exp i exp i − exp −i
2 2 2
    
θ + θj θ − θj
= exp i 2i sin
2 2
    
θ − θj θ + θj + π
= 2 sin exp i
2 2

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 387 — #47

i i

Some Interesting Theorems 387

θ−θ θ+θj +π
and 2 sin 2 j > 0, we have arg (exp(iθ ) − exp(iθj )) = 2 .
Consolidating the above observations, we get

π  θ + θj + π
n
arg H(exp(iθ)) = ϕj − θ + + βj
2 2
j=1

θ +π  
n n
π θj
= ϕj − θ + + βj + βj
2 2 2
j=1 j=1
⎛ ⎞
π  n
θj 
n
= ϕj − θ + +θ +π + βj ⎝ using βj = 2⎠
2 2
j=1 j=1

π 
n
θj
= ϕj + + βj ,
2 2
j=1

which is independent of θ . Hence, arg H is a constant function on μj ,

∀j ∈ {1, 2, . . . , n}. As arg H is continuous, its value on each circular arc
μj must be the same, and hence, it is constant on ∂B(0, 1). Since arg H is
the imaginary part of the analytic function log(H), it is a harmonic func-
tion and by maximum and minimum principles of harmonic functions, arg
H is constant on Cl B(0, 1). Hence, by Result 2.2.20, log(H) is a con-
stant function on Cl B(0, 1). Thus, H is a constant function on B(0, 1),
and hence,


n
F  (w) = ( f −1 ) (w) = c1 (w − wj )−βj , ∀w ∈ B(0, 1),
j=1

for some c1 ∈ C. Integrating on both sides of the above equation over a curve
joining 0 and w in B(0, 1), we get

w 
n
−1
f (w) = c1 (w − wj )−βj dw + c2 , ∀w ∈ B(0, 1),
0 j=1

for some c1 , c2 ∈ C. 

THEOREM 7.5.15 (Schwarz–Christoffel formula for upper half-plane)

Let  be a bounded simply connected region enclosed by a polygon with
vertices zj and interior angles αj π at zj , j = 1, 2, . . . , n. If f is a Riemann

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 388 — #48

i i

388 Schwarz–Christoffel Formula

mapping from  onto the upper half-plane H + , then there exist constants C1
and C2 such that
ζ n−1

−1
f (ζ ) = C1 (ζ − ζj )−βj dζ + C2 ,
i j=1

where the integral is defined over any curve joining i and ζ in H + and βj =
1 − αj , ζj = lim f (z), ∀j = 1, 2, . . . , n, with ζn = ∞.
z→zj

Proof: To find a linear fractional transform, which maps the unit disc onto
the the upper half-plane, we find a linear fractional transform T , which maps
1, −i, i to 1, 0, ∞, respectively, by using the definition of cross ratio as given
below.
w+i1−i w+i
T(w) = (w, 1, −i, i) = = −i .
w−i1+i w−i
i
As T maps the unit circle onto the real axis and T(0) = −i = i, it maps
−i
+
B(0, 1) onto H . We find wj ∈ ∂B(0, 1) such that T(wj ) = ζj , j = 1, . . . , n−1,
and we know that if wn = i, then T(wn ) = ∞, and T −1 ◦ f :  → B(0, 1)
is a bijective analytic map such that lim (T −1 ◦ f )(z) = wj , j = 1, 2, . . . , n.
z→zj
Hence, by applying the proof of previous theorem, we obtain


n
c1 (w − wj )−βj = ( f −1 ◦ T) (w) = ( f −1 ) (T(w))T  (w),
j=1

which implies that

c1 
n
( f −1 ) (T(w)) = 
(w − wj )−βj .
T (w)
j=1

Now, using the change of variable ζ = T(w) in the last equation, we also get


n−1
( f −1 ) (ζ ) = c1 (T −1 ) (ζ )(T −1 (ζ ) − T −1 (ζn ))−βn (T −1 (ζ ) − T −1 (ζj ))−βj .
j=1
(7.15)

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 389 — #49

i i

Some Interesting Theorems 389

Using the following

−iζ − 1 iζ + 1 ζ −i
w = T −1 (ζ ) = = =i ,
−ζ − i ζ +i ζ +i
(ζ − i) − (ζ + i) −2
(T −1 ) (ζ ) = i = ,
(ζ + i)2 (ζ + i)2
and  
−1 −1 ζ − i ζj − i
T (ζ ) − T (ζj ) = i −
 ζ + i ζj + i 
ζ ζ + iζ − iζj + 1 − (ζ ζj − iζ + iζj + 1)
= i
(ζ + i)(ζj + i)
−2(ζ − ζj )
=
(ζ + i)(ζj + i)
in equation (7.15), we get
 −βn n−1
 −βj
−2c1 2 −2(ζ − ζj )
( f −1 ) (ζ ) =
(ζ + i)2 ζ +i (ζ + i)(ζj + i)
j=1
 −βn  
− n−1
−2c1 2 −2 
βj n−1
ζ − ζj
−βj
= j=1
(ζ + i)2 ζ +i ζ +i ζj + i
j=1

n−1
 −βn  −2+βn (ζ − ζj )−βj
−2c1 2 −2 j=1
=
(ζ + i)2 ζ +i ζ +i 
n−1
(ζj + i)−βj
j=1

n−1
c1 
n−1
= C1 (ζ − ζj )−βj , where C1 = − (ζj + i)βj .
2
j=1 j=1

Therefore, applying integral over a curve joining i and ζ in H + , we get

ζ 
n
−1
f (ζ ) = C1 (ζ − ζj )−βj dζ + C2 , ∀ζ ∈ H + ,
i j=1

for some C1 , C2 ∈ C. 

Remark 7.5.16: The Schwarz–Christoffel formula is valid for semi-infinite

regions determined by a polygon with vertices zj , j = 1, 2, · · · , n − 1. In this
case, zn should be considered as a point at infinity.

i i

i i
 i i

“book” — 2014/6/4 — 21:22 — page 390 — #50

i i

390 Schwarz–Christoffel Formula

Example 7.5.17 Find a Reimann mapping that maps the upper half-plane to
the infinite strip {(x, y) : x ≥ 0, −1 ≤ y ≤ 1}.

Solution: The polygon considered in this example is as follows.

i
p
a2 =
2

p
a1 =
2
−i

1
We take z1 = −i, z2 = i; α1 = α2 = and ζ1 = −1, ζ2 = 1. (The choice
2
of ζ1 and ζ2 are arbitrary but distinct.) Then, by Schwarz–Christoffel formula
(Theorem 7.5.15), the required conformal mapping g is given by

g(ζ ) = C1 (ζ +1)−1/2 (ζ −1)−1/2 dζ +C2 = C1 (ζ 2 −1)−1/2 +C2 , ∀ζ ∈ H + ,

for some C1 , C2 ∈ C, where H + denotes +

# the upper half-plane. Since ζ ∈ H ,
we find the value of (ζ − 1)
2 1/2
√ i 1 − ζ , where the latter function is
by 2
defined according to the branch of ·, as defined in Result 3.3.11. Therefore,

1
g (ζ ) = −iC1 # dζ +C2 ⇒ g(z) = −iC1 arcsin(ζ )+C2 , ∀ζ ∈ H + .
1 − ζ2

Now, we find the values of C1 and C2 by using g(−1) = −i and g(1) = i as

follows:
π
−iC1 arcsin(−1) + C2 = iC1 + C2 = −i,
2
π
−iC1 arcsin(1) + C2 = −iC1 + C2 = i.
2
Adding these two equations, we get 2C2 = 0 ⇒ C2 = 0. Substitut-
2
ing this value in the first equation, we get C1 = − . Hence, g(ζ ) =
π
2i
arcsin(ζ ), ∀ζ ∈ H.
π

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 391 — #1

i i

8
Elliptic Functions

8.1 BASIC CONCEPTS

We would first like to make clear to the readers the common features among
ellipse, elliptic curve, elliptic integrals, and elliptic functions. We know that
x2 y2
an ellipse is a closed curve satisfying the equation 2 + 2 = 1, for some
a b
a, b ∈ R  {0}, whose parametric equations are given by
x(t) = a cos(t), y(t) = b sin(t), ∀t ∈ [0, 2π ].
(As there is no loss of generality, we assume that a ≤ b).
Hence, the arc length of this ellipse at a given point t0 ∈ [0, 2π ], from the
point on the ellipse corresponding to t = 0, is given by the integral
t0
S (t0 ) = |(x (t), y (t))| dt
0
t0 
= a2 sin2 (t) + b2 cos2 (t) dt
0
t0 
= a2 sin2 (u) + b2 (1 − sin2 (u)) du
0
t0 
= b2 − (b2 − a2 ) sin2 (u) du
0
t0 
(b2 − a2 )
= b 1 − k 2 sin2 (u) du, where k = .
b
0

391

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 392 — #2

i i

392 Basic Concepts

φ 
Hence, the integrals of the form 1 − k 2 sin2 (u) du are called elliptic inte-
0
grals. More specifically, they are called the elliptic integrals of second kind.
There are two more kinds of elliptic integrals, namely, elliptic integrals of
first kind and elliptic integrals of third kind.
φ du
Elliptic integrals of first kind :  .
0 1 − k 2 sin2 (u)
φ du
Elliptic integrals of third kind:   .
0 1 + n sin2 (u) 1 − k 2 sin2 (u)
By making the change of variable x = sin u, we get
x dx
Elliptic integrals of first kind : √ √ ,
0 1√− x2 1 − k 2 x2
x 1 − k 2 x2 dx
Elliptic integrals of second kind : √ , and
0 1 − x2
x dx
Elliptic integrals of third kind: √ √ √ .
0 1 + nx 1 −√x2 1 − k 2 x2
2
A general elliptic integral is of the form R(x, P(x)) dx, where R is a ra-
tional function in two variables, and P is a cubic or quadratic polynomial.
It is interesting to see from the literature on elliptic integrals that every el-
liptic integral can be expressed by using those standard forms, together with
elementary integrals.
An elliptic curve is a curve in C, which is given by the following equation;

y2 = x3 + ax + b where a, b ∈ Q with the discriminant 4a3 + 27b2  0.

The condition 4a3 + 27b2  0 is equivalent to say that x3 + ax + b = 0 has

no multiple roots in C. We are going to see that the Weierstrass function ℘a,b
is an elliptic function satisfying the differential equation

(℘a,b (z))2 = 4(℘(z))3 − 60G4 ℘(z) − 140G6

and hence, ℘a,b can be viewed as the inverse of an elliptic integral. We shall
show that every elliptic function is a rational function of ℘a,b and its deriva-
tive, for some suitable choices of a’s and b’s. From the literature of elliptic
curves, we see that every elliptic curve can be parameterized by using ℘a,b .
This is the history behind the similarity in the nomenclature. It should be
noted that ellipse is not an elliptic curve. Now, we are going to discuss the
elliptic functions in this chapter.

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 393 — #3

i i

Elliptic Functions 393

Definition 8.1.1 Let f : C → C and let a ∈ C. We say that ‘a’ is a period of

f if f (z + a) = f (z), ∀z ∈ C.
The set of all periods of f is denoted by Pf .

Definition 8.1.2 A non-zero period ‘a’ of f is said to be a fundamental period

a
of f if there is no b ∈ Pf \ {0} such that ∈ Z with |b| < |a|.
b
By the above definition, it is clear that ‘a’ is a fundamental period of f if
and only if ‘ − a’ is a fundamental period of f .

LEMMA 8.1.3 For a given f : C → C,

1. ma + nb ∈ Pf , whenever m, n ∈ Z and a, b ∈ Pf .
And if f is a non-constant meromorphic function on C, then
2. Pf has no limit point in C.
3. Pf is exactly equal to one of the following three sets:
(a) {0},
(b) {ma : m ∈ Z} for some a ∈ C,
a
(c) {ma + nb : m, n ∈ Z} for some a, b ∈ C with  R.
b

Proof:
1. First, we prove that if a ∈ Pf and m ∈ Z, then ma ∈ Pf . For m = 0,
clearly ma = 0 ∈ Pf . For m > 0, using the condition a ∈ Pf repeatedly,
we get

f (z + ma) = f (z + (m − 1)a + a)
= f (z + (m − 1)a)
= f (z + (m − 2)a)
..
.
= f (z + a)
= f (z), ∀z ∈ C

and hence, ma ∈ Pf . For m < 0,

f (z + ma) = f (z + ma + a)
= f (z + (m + 1)a)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 394 — #4

i i

394 Basic Concepts

= f (z + (m + 2)a)
..
.
= f (z − a)
= f (z − a + a)
= f (z), ∀z ∈ C.

Hence, ma ∈ Pf , ∀m ∈ Z. Clearly, p, q ∈ Pf ⇒ f (z+p+q) = f (z+q) =

f (z), ∀z ∈ C ⇒ p + q ∈ Pf . Thus, (1) follows.

2. If Sf is the set of all poles of f and if

f = {z ∈ C : f is analytic at z} = C \ Sf ,

then either Pf ⊆ f or Pf ⊆ Sf . If Pf ⊆ f and if Pf has a limit point,

then f has a limit point, and hence, f becomes a constant function on
f , by analytic continuation. If w is a pole of f , then it is isolated, and
hence, there exists r > 0 such that B(w, r) \ {w} ⊂ f . As f is constant
on f , f is constant on B(w, r)\{w}. Then, lim f (z) must be the constant
z→w
(and is not infinity), which is a contradiction. This implies that f cannot
have any pole. Thus, f is constant on C. This contradiction leads to
conclude that Pf has no limit point in C. If Pf ⊆ Sf , then obviously Pf
has no limit point since every pole of f is isolated.

3. If Pf  {0}, then we choose a ∈ Pf \ {0} such that |a| = inf{|p| :

p ∈ Pf \ {0}}. Such an a exists by using (2). If there are more than one
such a, then we choose any one of them. Using (1), we have {ma : m ∈
Z} ⊆ Pf . If {ma : m ∈ Z} = Pf , then the theorem is over. Otherwise, we
choose b ∈ Pf \ {ma : m ∈ Z} such thatI |b| = inf{|p| : p ∈ Pf \ {ma :
m ∈ Z}}. Then by using (1), {ma + nb : m, n ∈ Z} ⊆ Pf . Now, we
a a a
first claim that  R. Suppose that ∈ R, then either ∈ R \ Z or
b b b
b a a
∈ R \ Z. (Note that  0.) If ∈ R \ Z, then we can find m ∈ Z
a b b
such that
a
m< < m + 1 ⇒ |mb − a| < |b| and mb − a ∈ Pf ,
b
which is a contradiction to the choice of b. Similarly, we get a
b
contradiction by assuming that ∈ R \ Z.
a

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 395 — #5

i i

Elliptic Functions 395

Next, we claim that for each c ∈ Pf , there

 exist r, s ∈ R such that c = ra + sb.
a a  a b 
Using  , it follows that   = ab − ab  0, and hence, the system
b b a b 
of linear equations ra + sb = c; ra + sb = c in the variables r and s has a
unique solution (r, s) ∈ C2 . We note that (r, s) is a solution of the system of
linear equations if and only if (r, s) is also a solution of the same system of
linear equations. Hence, by the uniqueness of the solution, we get
(r, s) = (r, s) ⇒ r = r and s = s ⇒ r, s ∈ R.
Thus, our claim holds.
Now, we choose m, n ∈ Z such that
1 1
|r + m| ≤ and |s + n| ≤ .
2 2
If d = c + ma + nb, then d ∈ Pf , and using the necessary condition for
equality in triangle inequality, as ba  R, we get
1 1
|d| = |(r+m)a+(s+n)b| < |(r+m)a|+|(s+n)b| ≤ (|a|+|b|) ≤ (2|b|) = |b|.
 2 2
If d is not an integral multiple of a, then this inequality contradicts the mini-
mality of |b|. Hence, it follows that d = la for some l ∈ Z. Therefore, from
the definition of d , we have
c = d − ma − nb = (l − m)a − nb.
Thus, the proof is complete. 
From the above proof, we see the following observations:
1. If Pf = {ma : m ∈ Z}, then a is a fundamental period of f .
2. If Pf = {ma + nb : m, n ∈ Z} with a
b  R, then a and b are fundamental
periods of f .

Definition 8.1.4 A meromorphic function f on C is called an elliptic function

if Pf = {ma + nb : m, n ∈ Z} for some a, b ∈ C with ab  R, where Pf is the
set of all periods of f .
If f is a meromorphic function on C, then there exists Sf ⊂ C such that
f : C \ Sf → C and lim f (z) = ∞, ∀w ∈ Sf . In this case, by “a is a period
z→w
of f ”, we mean that f (z + a) = f (z), ∀z ∈ C \ Sf and w ∈ Sf if and only if
a + w ∈ Sf .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 396 — #6

i i

396 Basic Concepts

In the language of linear algebra, the condition ab  R can be stated as

‘a and b are linearly independent in C over R’ or briefly ‘a and b are inde-
pendent’. Therefore, in the above definition, as there exist two independent
fundamental periods of f , an elliptic function is also called a doubly periodic
function.

Definition 8.1.5 For a given pair of complex numbers a, b ∈ C with ab  R,

a period module generated by a and b is defined by Qa,b = {ma + nb :
m, n ∈ Z}.

Remark 8.1.6: Hereafter, we write ‘f is an elliptic function with periodic

module Qa,b ’ to mean that f is an elliptic function with Pf = Qa,b .

THEOREM 8.1.7 If f and g are elliptic functions with Pf = Pg and c ∈ C,

then
1. f + g is an elliptic function.
2. cf is an elliptic function.
3. fg is an elliptic function.
1 f
4. g is an elliptic function, and hence, g is an elliptic function.

Proof:
1. As f and g are meromorphic functions, for each w ∈ C, the Laurent
series expansions of f and g around w have at most finite number of
non-zero terms in their singular parts. Hence, the Laurent series ex-
pansion of f + g also has at most finite number of terms in its singular
part. If w is a pole of exactly one of f and g, then w is a pole of f + g. If
w is a pole of both f and g, then either w is a removable singularity of
f + g (and hence, it may be assumed as a regular point after removing
its singularity) or w is a pole of f + g. Hence, Sf +g ⊆ Sf ∪ Sg . Thus,
f + g is a meromorphic function on C. As f and g have same periods,
for each a ∈ Pf = Pg ,
( f + g)(z + a) = f (z + a) + g(z + a) = f (z) + g(z), ∀z ∈ C \ Sf +g .

We know that for each w ∈ C, there exists r > 0 such that f is analytic
on B(w, r) \ {w}. Therefore
lim ( f + g)(z) = lim ( f + g)(z + a) = lim ( f + g)(ζ ), ∀w ∈ C.
z→w z→w ζ →w+a

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 397 — #7

i i

Elliptic Functions 397

Thus, w is a pole of f + g if and only if w + a is a pole of f + g. Hence,

f + g is an elliptic function.
2. Clearly, w is a pole of f if and only if w is a pole of cf . Hence, cf is a
meromorphic function and Sf = Scf . Furthermore, if a ∈ Pf , then

(cf )(z + a) = cf (z + a) = cf (z), ∀z ∈ C \ Scf = Sf .

3. First, we show that if f and g are meromorphic functions on C, then fg

is also a meromorphic function. If w is a pole of order m for f , then
(a) w is a pole of order m for fg if w ∈ g and g(w)  0.
(b) w is a pole of order m − k for fg if w is a zero of order k for g with
k < m.
(c) w is a removable singularity for fg with ( fg)(w)  0 if w is a zero
of order m for g.
(d) w is a removable singularity for fg and after removing the singular-
ity, w is a zero of order k − m for fg if w is a zero of order k for g
with k > m.
Thus, every pole of f is either a pole of fg or a removable singularity of
fg. Similarly, every pole of g is either a pole or removable singularity
of fg. Moreover, if z  Sf ∪ Sg , then f and g are analytic at z, and
hence, fg is also analytic at fg. Thus, fg is a meromorphic function on
C. Using the periodicity of f and g as in (1), one can prove that fg is
doubly periodic with same fundamental periods of f and g.
4. If g is a meromorphic function, then obviously, every zero of g is a
pole of g1 . Every pole w of g is a removable singularity of g1 ; and it
becomes a zero of g1 , after removing the singularity at w. Hence, 1
g is a
meromorphic function and for every a ∈ Pf ,

1 1 1 1
(z + a) = = = (z).
g g(z + a) g(z) g
1
Thus, g is an elliptic function. Finally, using this with (3), we conclude
f
that g is also an elliptic function. 

Exercise 8.1.8 If f and g are elliptic functions such that Pf  Pg , then prove
that Pf +g = Pfg = P f = Pg .
g

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 398 — #8

i i

398 Fundamental Parallelogram

LEMMA 8.1.9 If f is an elliptic function, then f  is also an elliptic function.

Proof: Let z0 ∈ C and a ∈ Pf be fixed. If f is differentiable at z0 , then we get

f (z) − f (z0 )
f  (z0 ) = lim
z→z0 z − z0
f (z + a) − f (z0 + a)
= lim (using a ∈ Pf )
z→z0 (z + a) − (z0 + a)
f (w) − f (z0 + a)
= lim (by substituting w = z + a)
w→z0 +a w − (z0 + a)
= f  (z0 + a).

Therefore, f  (z0 ) = f  (z0 + a). If f has a pole at w of order m, then we can

write f (z) = (z − w)−m g(z), ∀z ∈ B(w, r) \ {w} for some analytic function g
and for some r > 0. Therefore,

f  (z) = (z − w)−m g (z) − m(z − w)−m−1 g(z)

= (z − w)−m−1 [(z − w)g (z) − mg(z)], ∀z ∈ B(w, r) \ {w}.

As (z − w)g (z) − mg(z) is analytic, we obtain f  has a pole of order m + 1 at

w. Thus, f  is also an elliptic function. 

8.2 FUNDAMENTAL PARALLELOGRAM

Definition 8.2.1 Let a and b be complex numbers such that a
b  R. For each
z0 ∈ C, the semi-open parallelogram

Da,b (z0 ) = {z0 + λa + μb : 0 ≤ λ < 1, 0 ≤ μ < 1}

with vertices z0 , z0 + a, z0 + b, and z0 + a + b is called a fundamental

parallelogram generated by a and b.

If f is an elliptic function with period module Qa,b , then we also call

Da,b (z0 ) a fundamental parallelogram of f . If f is a constant (elliptic) func-
tion, then there is no fundamental period for f . In this case, for every pair
a, b ∈ C with ab  R and for every z0 ∈ C, we call Da,b (z0 ) a fundamental
parallelogram of f .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 399 — #9

i i

Elliptic Functions 399

z0 + a + b
r
z0 + a



r



























 
r
 z0 + b


r
z0
A fundamental parallelogram

LEMMA 8.2.2 Let a and b be such that a

b  R and let z0 ∈ C.
1. For each z ∈ C, there exists unique w ∈ Da,b (0) such that z − w ∈ Qa,b .
2. There exists a bijection φ : Da,b (z0 ) → Da,b (0) such that φ(z) − z ∈
Qa,b , ∀z ∈ Da,b (z0 ).
3. f (Da,b (z0 )) = f (Da,b (0)) = f (C), ∀f with Pf = Qa,b .

Proof:
1. As in the proof of Lemma 8.1.3(3), for the given z ∈ C, we get z = ra+
sb for some r, s ∈ R. Next, we choose m, n ∈ Z such that 0 ≤ r+m < 1
and 0 ≤ s + n < 1 so that

z + na + mb = (r + m)a + (s + n)b ∈ Da,b (0).

If we take w = z + na + mb, then obviously, z − w ∈ Qa,b . Suppose

there exists ζ ∈ Da,b (0) with z − ζ ∈ Qa,b , then let ζ = z + la + kb
for some l, k ∈ Z. Then, 0 ≤ r + l < 1 and 0 ≤ s + k < 1, and hence,
1 > |(r + m) − (r + l)| = |m − l|. Being m − l ∈ Z, the only option is
m − l = 0. Similarly, we can prove that n = k . Thus, w = ζ , and hence,
the uniqueness follows.
2. Using (1), for the given z0 ∈ C, there exist unique w0 ∈ Da,b (0) and
p ∈ Qa,b such that w0 = z0 + p. If ζ0 = a + b − p, then we claim that

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 400 — #10

i i

400 Fundamental Parallelogram

ζ0 ∈ Cl Da,b (z0 ). As w0 ∈ Da,b (0), there exist r0 , s0 ∈ [0, 1) such that

w0 = r0 a + s0 b. Therefore, if u0 = 1 − r0 and v0 = 1 − s0 , then
ζ0 = a+b−p
= a + b − w0 + z0
= z0 + (1 − r0 )a + (1 − s0 )b
= z0 + u0 a + v0 b ∈ Cl Da,b (z0 )
as u0 , v0 ∈ (0, 1].
Case 1: w0 = 0.
In this case, ζ0 = a + b + z0 , p = z0 , and hence, the required
map φ : Da,b (z0 ) → Da,b (0) is simply the translation map
given by φ(z) = z + p, ∀z ∈ Da,b (z0 ).

Case 2: w0  0.
Then, we assume that ζ0  ∂Da,b (z0 ), and hence, we can
4
partition Da,b (z0 ) as ∪ Rj , where
j=1

R1 = {z0 + λa + μb : λ ∈ [0, u0 ), μ ∈ [0, v0 )}

R2 = {z0 + λa + μb : λ ∈ [u0 , 1), μ ∈ [0, v0 )}
R3 = {z0 + λa + μb : λ ∈ [u0 , 1), μ ∈ [v0 , 1)}
R4 = {z0 + λa + μb : λ ∈ [0, u0 ), μ ∈ [v0 , 1)}
The following diagram explains this proof geometrically.
w0 + a + b

w0 + b R3
R4
a + b R2
w0 + a
b R1
z0 + a + b S2 w
Da,b(0)
Da,b(z0) p 0

+ b− S4
z0 + b R3 a S3 a
ζ0
R4 ζ0
R2 z0 + a w0 0
R1 +p
z 0

z0

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 401 — #11

i i

Elliptic Functions 401

⎫ ⎧ 
R1 = R1 + p ⎪
⎪ ⎪
⎪ R = R1 + p
⎬ ⎨ 1
S2 + a = R2 = R2 + p S2 = R2 + p − a
=⇒
S3 + a + b = R3 = R3 + p ⎪
⎪ ⎪
⎪ S3 = R3 + p − a − b
⎭ ⎩
S4 + b = R4 = R4 + p S4 = R4 + p − b
Now, if we define φ : Da,b (z0 ) → Da,b (0) by
⎧
⎪
⎪ z+p if z ∈ R1
⎪
⎨z + p − a if z ∈ R2
φ(z) =
⎪
⎪z + p − a − b if z ∈ R3
⎪
⎩
z+p−b if z ∈ R4
then φ is the required bijection between Da,b (z0 ) and Da,b (0).
If ζ0  a + b + z0 and ζ0 ∈ ∂Da,b (0), then we can prove this
part of the lemma by splitting the rectangle Da,b (z0 ) into two suitable
subrectangles. This part is a simple exercise to the reader.
3. Obviously, we have f (Da,b (0)) ⊆ f (C). If ζ ∈ f (C), then there exists
z ∈ C such that f (z) = ζ . Using part (1) of this lemma, there exists
w ∈ D(a, b)(0) such that z − w ∈ Qa,b = Pf . Thus, ζ = f (z) = f (w) ∈
f (Da,b (0)). Hence, we get f (Da,b (0)) = f (C). Therefore, it immediately
follows that f (Da,b (z0 )) ⊆ f (Da,b (0)). If w ∈ Da,b (0), then f (w) =
f (φ −1 (w)) ∈ f (Da,b (z0 )), as φ −1 (w) − w ∈ Qa,b , from the proof of part
(2) of the lemma. Therefore, f (Da,b (z0 )) = f (Da,b (0)). 
Using the previous lemma, if f is an elliptic function, then the number of
poles (including multiplicities) of f in a fundamental parallelogram of f is
independent of z0 . Therefore, we can define the following.

Definition 8.2.3 Let f be an elliptic function. Then, the number of poles (in-
cluding multiplicities) of f in a fundamental parallelogram of f is called the
order of the elliptic function f .

LEMMA 8.2.4 Let f be an elliptic function. The number of poles of f in a

fundamental parallelogram of f is finite, and the sum of the residues of f at
all poles in Da,b (z0 ) is 0.

Proof: We first note that if f is a meromorphic function on a region , then

for each z ∈ , there exists rz > 0 such that f is analytic on B(z, rz ) \ {z}.
(If z is a pole, then this statement follows from Lemma 5.2.11, and if it is not
a pole, then this statement is straightforward.) Let Da,b (z0 ) be a fundamental
parallelogram of f . If there exist infinite number of poles of f in Cl Da,b (z0 )
(which is compact), then they should have a limit point w in Cl Da,b (z0 ) by

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 402 — #12

i i

402 Fundamental Parallelogram

Theorem 1.4.39. Therefore, f becomes analytic on B(w, r) \ {w}, for some

r > 0. By the choice of w, for this r > 0, we can find a pole p of f such
that p ∈ Cl Da,b (z0 )B(w, r) \ {w}, which is a contradiction. Thus, there exist
at most finite number of poles of f in Da,b (z0 ).
Since every pole p of f in Da,b (z0 ) corresponds to a pole q of f in a
translate Da,b (z + z0 ) of Da,b (z0 ) such that p − q ∈ Qa,b , by Lemma 8.2.2,
instead of finding the sum of residues of f in Da,b (z0 ), we can find the same
in Da,b (z + z0 ) for any z ∈ C. As there are only finite number of zeroes and
poles in Da,b (z0 ), by translating Da,b (z0 ) with a suitable z ∈ C, we can obtain
that there is no zero or pole on the boundary of Da,b (z + z0 ). Therefore, we
now assume that there is no pole on ∂Da,b (z0 ).
If γ is the boundary of Da,b (z0 ), then by Cauchy’s residue theorem,

1
f (z) dz = Sum of the residues of f at poles enclosed by γ .
i2π
γ = Sum of the residues of f at poles in Da,b (z0 ).
On the other hand, as Da,b (z0 ) = {z0 + λa + μb : λ, μ ∈ [0, 1)}, where
a, b ∈ Pf with ab  R and z0 ∈ C is fixed, then
 4 
1 1
f (z) dz = f (z) dz.
i2π i2π
γ j=1 γj

where
γ1 (t) = z0 + ta, t ∈ [0, 1]
γ2 (t) = z0 + a + tb, t ∈ [0, 1]
γ3 (t) = z0 + (1 − t)a + b, t ∈ [0, 1]
γ4 (t) = z0 + (1 − t)b, t ∈ [0, 1]
z0 + a + b

γ3
γ2
z0 + b

z0 + a
γ4
γ1
z0

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 403 — #13

i i

Elliptic Functions 403

Therefore, using a, b ∈ Pf ,

 1
f (z) dz = f (z0 + ta) adt
γ1 0
0
= f (z0 + (1 − s)a) a(−ds)
1
1
= − f (z0 + (1 − s)a) a(−ds)
0

= − f (z) dz.
γ3

  
Similarly, we can prove that f (z) dz = − f (z) dz, and hence, f (z) dz = 0.
γ2 γ4 γ
Thus, we get the sum of the residues of f at poles in Da,b (z0 ) = 0. 

LEMMA 8.2.5 If Da,b (z0 ) is a fundamental parallelogram of a non-zero ellip-

tic function, then the number of zeroes of f in Da,b (z0 ) including multiplicities
is same as the number of zeroes of f in Da,b (z0 ) including multiplicities.

f
Proof: Using Theorem 8.1.7 and Lemma 8.1.9, we get that is also an
f
elliptic function. Therefore, as in the proof of the previous lemma, we can
 f  (z)
prove that dz = 0. However, using argument principle, we get
∂Da,b (z0 ) f (z)
 f  (z)
1
i2π dz = number of zeroes of f (including multiplicities) en-
∂Da,b (z0 ) f (z)
closed by ∂Da,b (z0 )− number of poles of f (including multiplicities) enclosed
by ∂Da,b (z0 ). This completes the proof of the lemma. 

LEMMA 8.2.6 If {z1 , z2 , . . . , zn } and {w1 , w2 , . . . , wn } are the zeroes and poles
of an elliptic function f including multiplicities in a fundamental parallelo-
gram Da,b of f , respectively, then z1 + z2 + · · · + zn − w1 − w2 − · · · −
wn ∈ Pf .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 404 — #14

i i

404 Fundamental Parallelogram

Proof: Without loss of generality, we assume that no pole or zero of f is on

∂Da,b because, if necessary, as the number of poles and zeroes are finite, we
can translate this fundamental parallelogram to another fundamental paral-
lelogram with this property. Furthermore, by Lemma 8.2.2, we conclude that
proving this result for one fundamental parallelogram implies this result for
all fundamental parallelogram. As {z1 , z2 , . . . , zn } and {w1 , w2 , . . . , wn } are
the zeroes and poles of f including multiplicities in Da,b , applying Results
n
(z − wj )
j=1
5.2.3 and 5.2.11 repeatedly, we observe that if φ(z) = f (z), then
n
(z − zj )
j=1
φ has removable singularity at each ζ ∈ {z1 , z2 , . . . , zn , w1 , w2 , . . . , wn }, and
it has neither a pole nor a zero at any point. Hence,

n
(z − zj )
j=1
f (z) = φ(z), ∀z ∈ Da,b \ {w1 , w2 , . . . , wn }.
n
(z − wj )
j=1

Therefore, finding the logarithmic derivative of f , we get

n  
f  (z) 1 1 φ  (z)
= − + .
f (z) (z − zj ) (z − wj ) φ(z)
j=1

φ
As φ is analytic, by applying Cauchy’s theorem and Cauchy’s integral
formula, we get
⎛ ⎞
  n  
1 zf (z) ⎜ 1 z 1 z ⎟
dz = ⎝ dz − dz⎠
i2π f (z) i2π (z − zj ) i2π (z − wj )
∂Da,b j=1 ∂Da,b ∂Da,b

1 φ  (z)
+ dz
i2π φ(z)
∂Da,b
= z1 + z2 + · · · + zn − w1 − w2 − · · · − wn .
Next, using the parametric equation of ∂Da,b (from Lemma 8.2.4), we get
 4 
1 zf  (z) 1 zf  (z)
dz = dz,
i2π f (z) i2π f (z)
∂Da,b j=1 γj

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 405 — #15

i i

Elliptic Functions 405

where γ1 (t) = ζ + ta, γ2 (t) = ζ + a + tb, γ3 (t) = ζ + (1 − t)a + b, γ4 (t) =

ζ + (1 − t)b, t ∈ [0, 1]. Therefore, using a, b ∈ Pf = Pf  , we get

 1
zf  (z) (ζ + ta)f  (ζ + ta)
dz = adt
f (z) f (ζ + ta)
γ1 0

1
(ζ + b + ta)f  (ζ + b + ta)
= adt
f (ζ + b + ta)
0

1
bf  (ζ + b + ta)
− adt
f (ζ + b + ta)
0

0
(ζ + b + (1 − s)a)f  (ζ + b + (1 − s)a)
= a(−ds)
f (ζ + b + (1 − s)a)
1

1
bf  (ζ + ta)
− adt
f (ζ + ta)
0

1
(ζ + b + (1 − s)a)f  (ζ + b + (1 − s)a)
= − a(−ds)
f (ζ + b + (1 − s)a)
0

f  (z)
−b dz
f (z)
γ1
 
zf  (z) f  (z)
= − dz − b dz.
f (z) f (z)
γ3 γ1

Now, using a is a period of f and f has no zero or pole on ∂Da,b , we get that
f (γ1 ) is a closed curve not passing through zero. Then, by using the change
of variable w = f (z), we get
  
f (z) dw
dz = dw = WN( f (γ1 ), 0) ∈ Z.
f (z) w−0
γ1 f (γ1 )

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 406 — #16

i i

406 Weierstrass ℘a,b Function

 zf  (z)
Hence, dz = nb for some n ∈ Z. Similarly, we can prove that
γ1 +γ3 f (z)
 zf  (z)
dz = ma, for some m ∈ Z. Thus,
γ2 +γ4 f (z)

1 zf  (z)
z1 + z2 + · · · + zn − w1 − w2 − · · · − wn = dz = ma + nb
i2π f (z)
∂Da,b

which belongs to Pf . 

LEMMA 8.2.7 Every entire elliptic function is a constant function.

Proof: Let f be an entire elliptic function. First, we show that f is a bounded

function on C. If a and b are independent periods of f , then consider the
closed parallelogram
K = {λa + μb : 0 ≤ λ ≤ 1, 0 ≤ μ ≤ 1}
with vertices 0, a, b, and a + b. As f is continuous on the compact set K , we
get that f is bounded on K . Next, we show that for every z ∈ C, there exist
m, n ∈ Z such that z + na + mb ∈ K .
As in the proof of Lemma 8.2.2, we get z = ra + sb for some r, s ∈ R. Next,
we choose m, n ∈ C such that
0 ≤ r + m ≤ 1 and 0 ≤ s + n ≤ 1
so that
z + na + mb = (r + m)a + (s + n)b ∈ K.
Clearly, f (z + na + mb) = f (z), using a, b ∈ Pf . Thus, f (C) = f (K), which is
bounded. Therefore, f is a bounded entire function, and hence, it is constant,
by Liouville’s theorem. 

Exercise 8.2.8 Prove that the order of a non-constant elliptic function is at

least 2.
Hint: Apply Lemma 8.2.4.

8.3 WEIERSTRASS ℘a,b FUNCTION

There are no obvious examples of non-constant elliptic functions. However,
the Weierstrass function is an example of elliptic function with desirable
fundamental periods, which is defined as follows.

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 407 — #17

i i

Elliptic Functions 407

Definition 8.3.1 For given a, b ∈ C with a

b  R, we define
 
1 1 1
℘a,b (z) = 2 + −
z m,n∈Z,
(z − (ma + nb))2 (ma + nb)2
m0 or n0

∀z ∈ Z \ {ma + nb : m, n ∈ Z}.

The above infinite series means lim SJ , whenever the limit exists,
J →∞
J   
where SJ = 1
(z−(ma+nb))2
− 1
(ma+nb)2
, ∀J ∈ N. For notational
N=1 m,n∈Z
|m|+|n|=N
convenience, we write ℘a,b as follows.
 
1 1 1
℘a,b (z) = 2 + − 2 , ∀z ∈ Z \ Qa,b ,
z (z − p)2 p
p∈Qa,b \{0}

where Qa,b = {ma + nb : m, n ∈ Z}.

LEMMA 8.3.2 If a, b ∈ C with a

b  R, then there exists μ > 0 such that

|ra + sb| ≥ μ(|r| + |s|), ∀r, s ∈ R.

Proof: If φ : R × R → R is defined by

φ(r, s) = |ra + sb|, ∀(r, s) ∈ R × R

then φ is continuous on R × R. Indeed for a given > 0, we choose δ =

, so that whenever |(r, s) − (u, v)| < δ , we have |r − u| ≤ |(r, s) −
|a| + |b|
(u, v)| < δ , |s − v| ≤ |(r, s) − (u, v)| < δ , and hence,

|φ(r, s) − φ(u, v)| = | |ra + sb| − |ua + vb| |

≤ | (ra + sb) − (ua + vb) |
= |(r − u)a + (s − v)b|
≤ |r − u||a| + |s − v||b|
< δ(|a| + |b|)
= .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 408 — #18

i i

408 Weierstrass ℘a,b Function

Next, we define ψ : R × R → R by ψ(r, s) = |r| + |s|, ∀(r, s) ∈ R × R. We

also see that ψ is continuous because for a given > 0, we choose δ = 2 so
that |(r, s) − (u, v)| < δ implies that

|ψ(r, s) − ψ(u, v)| = | |r| + |s| − |u| − |v| |

≤ | |r| − |u| | + | |s| − |v| |
≤ |r − u| + |s − v|
≤ 2|(r, s) − (u, v)|
< 2δ
= .

If K = {(r, s) ∈ R : |r| + |s| = 1}, then K is a closed set because ψ is

−1
continuous,  {1} is a closed√set, and K = ψ ({1}). Clearly, K is bounded, as
|(r, s)| = |r|2 + |s|2 ≤ 2, ∀(r, s) ∈ K . Therefore, K is a compact subset
of C. Since φ is a continuous real-valued function on K , if μ = inf φ(K), then
μ = φ(r0 , s0 ) for some (r0 , s0 ) ∈ K .
We claim that μ > 0. Suppose that μ = 0. As (r0 , s0 ) ∈ K , we have
|r0 | + |s0 | = 1, and hence, at least one of them is non-zero.

Case 1: r0  0.
−s0
Then, |r0 a + s0 b| = 0 ⇒ r0 a + s0 b = 0 ⇒ a
b = r0 ∈ R, which is
a contradiction.

Case 2: s0  0.
−r0
Here, |r0 a + s0 b| = 0 ⇒ r0 a + s0 b = 0 ⇒ a = s0
b
∈ R, which is a
contradiction to b
a  R. (Note that b
a R ⇔ ab  R.)

Thus, μ > 0. Now, for (r, s) = (0, 0), the inequality

  |ra
 + sb| ≥ μ(|r| + |s|)
 r   s 
holds and for (r, s) ∈ R × R \ (0, 0),  |r|+|s|  +  |r|+|s|  = 1, and hence,
 
 r s 
 
 |r| + |s| a + |r| + |s| b ≥ μ ⇒ |ra + sb| ≥ μ(|r| + |s|).

LEMMA 8.3.3 If a and b are independent complex numbers, then the series
 
1 1
− 2
(z − p)2 p
p∈Qa,b \{0}

converges uniformly on every compact subset of C \ Qa,b .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 409 — #19

i i

Elliptic Functions 409

Proof: Let K be a compact subset of C \ Qa,b . Since K is bounded, there

exists C > 0 such that |z| ≤ C, ∀z ∈ K . We choose N1 ∈ N, such that

|p| > 2C, ∀p = ma + nb such that m, n ∈ Z with |m| + |n| ≥ N1 . (8.1)

For a given z ∈ K and p as in equation (8.1), we have |p| > 2C ≥ 2|z|, and
|p| |p|
hence, |z − p| ≥ |p| − |z| > |p| − = . Therefore,
2 2
   
 1 1   2pz − z2 
 − =  
 (z − p)2 p2   p2 (z − p)2 
   
 pz   z(p − z) 
≤  2 + 
p (z − p)2   p2 (z − p)2 
|z| |z|
< +
|p| 2 |p|
|p| |p|2
4 2
6C
≤ . (8.2)
|p|3

Now, using Lemma 8.3.2, we choose μ > 0 such that

|ma + nb| > μ(|m| + |n|), ∀m, n ∈ Z. (8.3)

Next, we note that for every N ∈ N, there exist exactly 4N ordered pairs
(m, n) such that |m| + |n| = N . Indeed, the required ordered pairs are given
as follows.
If N is even, then

(4 pairs) (±N, 0), (0, ±N)

(8 pairs) (±(N − 1), ±1), (±1, ±(N − 1))
(8 pairs) (±(N − 2), ±2), (±2, ±(N − 2))
..
.         
N N N N
(8 pairs) ± + 1 ,± −1 , ± − 1 ,± +1
2 2 2 2
 
N N
(4 pairs) ± , ±
2 2
   
N
(Total 4N pairs) = 4 + −1 ×8+4
2

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 410 — #20

i i

410 Weierstrass ℘a,b Function

and if N is odd, then

(4 pairs) (±N, 0), (0, ±N)
(8 pairs) (±(N − 1), ±1), (±1, ±(N − 1))
(8 pairs) (±(N − 2), ±2), (±2, ±(N − 2))
..
.   
N +1 N −1 N −1 N +1
(8 pairs) ± ,± , ± ,±
2 2 2 2
 
N −1
(Total 4N pairs) =4+ ×8 .
2
Now, for a given > 0, we can find N2 ∈ N such that N2 > N1 and

∞
1
N2
< . For this N1 , using the above observation and the equations
N=N2
(8.2)
 and (8.3), we get 
 
   1   N2    
 − 1
− 1
− 1 
 (z−p) 2 p 2 (z−(ma+nb)) 2 (ma+nb) 
2
p∈Qa,b \{0} N=1 m,n∈Z 
|m|+|n|=N
 
 ∞  

 1 1 
=  − 2 
 (z − (ma + nb)) 2 (ma + nb)
N=N2 m,n∈Z 
|m|+|n|=N
∞
6C
≤
|ma + nb|3
N=N2 m,n∈Z
|m|+|n|=N
∞
1
≤ 6C
μ3 (|m| + |n|)3
N=N2 m,n∈Z
|m|+|n|=N
∞
4N
≤ 6C
μ3 N 3
N=N2
∞
24C 1 24C
≤ < 3 .
μ3 N 2 μ
N=N2

Thus, the series converges uniformly on every compact subset of C \ Qa,b . 

RESULT 8.3.4 The Weierstrass ℘a,b function is an

1. even function;
2. elliptic function with P℘a,b = Qa,b .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 411 — #21

i i

Elliptic Functions 411

Proof: For z ∈ C \ Qa,b , replacing −(m, n) by (u, v) and using

{(m, n) : m, n ∈ Z, (m, n)  (0, 0)} = {−(m, n) : m, n ∈ Z, (m, n)  (0, 0)},

we get
∞  
1 1 1
℘a,b (−z) = + −
(−z)2 (−z − (ma + nb))2 (ma + nb)2
N=1 m,n∈Z
|m|+|n|=N
∞  
1 1 1
= + −
z2 (z + (ma + nb))2 (ma + nb)2
N=1 m,n∈Z
|m|+|n|=N
∞  
1 1 1
= + −
z2 (z − (ua + vb))2 (ua + vb)2
N=1 u,v∈Z
|u|+|v|=N

= ℘a,b (z).

From the previous lemma, ℘a,b is an analytic function on C \ Qa,b and

 2 2 −2 −2
℘a,b (z) = − − = = .
z3 (z − p)3 (z − p)3 (z − (ma + nb))3
p∈Qa,b \{0} p∈Qa,b m,n∈Z

Hence, for any ja + kb ∈ Qa,b , using {(m + k, n + j) : m, n ∈ Z} = {(m, n) :

m, n ∈ Z}, we get

 −2
℘a,b (z + ja + kb) =
(z + ja + kb − (ma + nb))3
m,n∈Z
−2
=
(z + (j − m)a + (k − n)b)3
m,n∈Z
−2
=
(z + ra + sb)3
r,s∈Z

= ℘a,b (z).
 is a doubly periodic with P  = Q . If
Therefore, ℘a,b ℘ a,b a,b

F(z) = ℘a,b (z + a) − ℘a,b (z), ∀z ∈ C \ Qa,b ,

then
F  (z) = ℘a,b
 
(z + a) − ℘a,b (z) = 0, ∀z ∈ C \ Qa,b .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 412 — #22

i i

412 Weierstrass ℘a,b Function

Thus, F is constant on C\Qa,b , and hence, on C. Therefore, by taking z = − a2 ,

we get F(− a2 ) = ℘a,b ( a2 ) − ℘a,b (− a2 ) = 0, as ℘a,b is an even function. Thus,
F is identically zero on C, and hence, ℘a,b (z + a) = ℘a,b (z), ∀z ∈ C \ Qa,b .
Similarly, we can show that ℘a,b (z + b) = ℘a,b (z), ∀z ∈ C \ Qa,b . Therefore,
℘a,b is an elliptic function with P℘a,b = Qa,b . 

LEMMA 8.3.5 If f and g are elliptic functions such that both have same ze-
roes and poles of the same orders, then prove that f = cg for some constant
c ∈ C.

Proof: As f and g are elliptic functions with same zeroes and poles, it is ob-
vious that they have same fundamental periods. Then, gf is an entire function
after removing the singularities at the zeroes and poles of f . Thus, by Lemma
8.2.7, gf is a constant, and hence, the lemma follows. 

THEOREM 8.3.6 Every even elliptic function with fundamental periods a and
b is a rational function of ℘a,b .

Proof: Let f be an even elliptic function with fundamental periods a and b.

Case 1: f has neither a pole nor a zero at 0.
Then, using the doubly periodicity of f, we immediately conclude
that f has neither a pole nor a zero at w, ∀w ∈ P℘a,b . If w ∈
{ a2 , b2 , a+b 
2 }, then w−(−w) = 2w ∈ {a, b, a+b}, and hence, ℘a,b (w) =
℘a,b (−w). On the other hand, as ℘  is an odd function (equivalently,
a,b
℘a,b is an even function), we get ℘a,b  (w) = −℘ (−w). Therefore,
a,b

℘a,b vanishes at each w ∈ { 2 , 2 , 2 }. As 0 is the only pole for ℘a,b
a b a+b  in

the fundamental parallelogram Da,b (0) and its order is 3, by Lemma

8.2.5, the number of zeroes of ℘a,b  in D (0) must be 3. Therefore,
a,b
a b a+b
, , must be simple zeroes of ℘  , and there is no other zero
2 2 2 a,b
for ℘a,b  in D (0). Hence, obviously, if w ∈ { a , b , a+b }, then w
a,b 2 2 2
must be a zero of order 2 for ℘a,b − ℘a,b (w). Using the doubly pe-
riodicity of ℘a,b , we obtain

that ℘a,b − ℘a,b (w) has a zero of order
2 at each w ∈ 12 P℘a,b = { p2 : p ∈ P℘a,b } . Next, if w  12 P℘a,b , then
℘a,b − ℘a,b (w) has two distinct simple zeroes at w and −w.
Since f is an even function, for a given zero z0 of f in Da,b (0),
−z0 is also a zero of f, and hence, there exists a unique point w0
in Da,b (0) such that w0 + z0 ∈ Qa,b and hence, f (−z0 ) = f (w0 ). If
w0  z0 , then we get two distinct zeroes (z0 and w0 ) of f in Da,b (0)
with z0 +w0 ∈ Qa,b . If w0 = z0 (equivalently z0 ∈ 12 P℘a,b ), then f has

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 413 — #23

i i

Elliptic Functions 413

a zero at z0 of even order. (Because, being f is even, f (2k−1) is an odd

function for every k ∈ N, and hence, f (2k−1) (z0 ) = −f (2k−1) (−z0 ).
From z0 ∈ 12 P℘a,b and by the fact that f (2k−1) is a doubly periodic
function of periods a and b, we also have f (2k−1) (z0 ) = f (2k−1) (−z0 ).
Therefore, f (2k−1) (z0 ) = 0. Therefore, if the positive integer m at
which f (m) (z0 )  0, then m must be even.) Thus, if the distinct zeroes
of f in Da,b (0) are z1 , z2 , . . . , zn , then we can rename the zeroes such
that
1 1
z1 , z2 , . . . , z2k  P℘a,b , z2k+1 , z2k+2 , . . . , zn ∈ P℘a,b
2 2
and
zk+j + zj ∈ Qa,b , ∀j = 1, 2, . . . , k
for some k ∈ N with 2k ≤ n.
If the order of the zero of f at zj is mj , then clearly mj ∈ 2N for
2k + 1 ≤ j ≤ n and

k 
n mj
(℘a,b − ℘a,b (zj ))mj · (℘a,b − ℘a,b (zj )) 2 (8.4)
j=1 j=2k+1

has exactly the same zeroes z1 , z2 , . . . , zn with the same orders mj .

Similarly, if ζ1 , ζ2 , . . . , ζν are the distinct poles of order μj , respec-
tively, for f in Da,b (0), then by the previous argument, we can say
that
λ
 ν
 μj
(℘a,b − ℘a,b (ζj ))μj · (℘a,b − ℘a,b (ζj )) 2 (8.5)
j=1 j=2λ+1

has exactly the same zeroes ζ1 , ζ2 , . . . , ζν with the orders μj , for a

suitable λ ∈ N.
n ν
Then by Lemma 8.2.5, we have mj = μj , and hence, both of
j=1 j=1
the functions in equations (8.4) and (8.5) are having same poles with
same orders. Thus, after removing the removable singularities (the
poles of the functions in equations (8.4) and (8.5)),

k 
n mj
(℘a,b − ℘a,b (zj ))mj · (℘a,b − ℘a,b (zj )) 2
j=1 j=2k+1

λ 
ν μj
(℘a,b − ℘a,b (ζj ))μj · (℘a,b − ℘a,b (ζj )) 2
j=1 j=2λ+1

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 414 — #24

i i

414 Weierstrass ℘a,b Function

has the zeroes zj with orders mj , 1 ≤ j ≤ n, and poles ζj , 1 ≤ j ≤ μ,

with orders μj in Da,b (0). Thus, by applying Lemma 8.3.5, we
conclude that

k n mj
(℘a,b − ℘a,b (zj ))mj · (℘a,b − ℘a,b (zj )) 2
j=1 j=2k+1
f =c .

λ 
ν μj
(℘a,b − ℘a,b (ζj ))μj · (℘a,b − ℘a,b (ζj )) 2
j=1 j=2λ+1

for some c ∈ C.
Case 2: f can have a zero or a pole at 0.
If f has a zero of order n at 0, then n ∈ 2N, as f is even. We note that
n
℘a,b has a double pole at 0, and hence, ℘a,b
2
has a pole at 0 of order
n
n. Therefore, f ℘a,b
2
has neither a pole nor a zero at 0 after removing
the singularity at 0.
If f has a pole of order m, then m ∈ 2N, as f is even. Therefore,
−m
f ℘a,b2 has neither a pole or a zero at 0 after removing the singularity
at 0.
Hence, the theorem follows. 

 ),
THEOREM 8.3.7 Every elliptic function can be written as R(℘a,b , ℘a,b
where R is a rational function in two variables.

Proof: Let f be a given elliptic function. If we define f1 = f (z)−f2 (−z) and

f2 (z) = f (z)+f2 (−z) , ∀z ∈ f , then f1 is an odd function, f2 is an even function,
and f = f1 + f2 on f . (Recall that f is the set of all points at which f is
f1
analytic.) As  and f2 are elliptic even functions, applying the previous
℘a,b
f1
theorem, we get  = R1 (℘a,b ) and f2 = R2 (℘a,b ), where R1 and R2 are
℘a,b
rational functions in single variable. Thus, f = ℘a,b  R (℘ ) + R (℘ ),
1 a,b 2 a,b
which is a rational function in ℘a,b and ℘a,b .  

In the following sequel, we use the notation Gn = pn , ∀n ≥ 3.
1
p∈Qa,b \{0}

RESULT 8.3.8 The Weierstrass ℘a,b function satisfies the following differen-
tial equation.

(℘a,b (z))2 = 4(℘a,b (z))3 − 60G4 ℘a,b (z) − 140G6 .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 415 — #25

i i

Elliptic Functions 415

Proof: We first find the Laurent series expansion of ℘a,b in B(0, r) \ {0}, for
some r > 0.
 
1 1 1
℘a,b (z) = + −
z2 (z − p)2 p2
p∈Qa,b \{0}
   
1 d 1 1
= + −
z2 dz (p − z) p2
p∈Qa,b \{0}
!∞   " #
1 1 d z k 1
= + − 2
z2 p dz p p
p∈Qa,b \{0} k=0
∞  k−1 #
1 1 z 1
= + k − 2
z2 p2 p p
p∈Qa,b \{0} k=1
! ∞  k−1 " #
1 1 z 1
= + 1+ k − 2
z2 p2 p p
p∈Qa,b \{0} k=2
∞  k−1
1 1 z
= + k
z2 p2 p
p∈Qa,b \{0} k=2
∞
1 1 k
= + (k + 1) z
z2 pk+2
k=1 p∈Qa,b \{0}
∞
1
= + (k + 1)Gk+2 zk
z2
k=1
∞
1
= + (2n + 1)G2n+2 z2n
z2
n=1

as G2n+1 = 0, ∀n ∈ N. Differentiating on both sides, we get

∞
 −2
℘a,b (z) = + 2n(2n + 1)G2n+2 z2n−1 .
z3
n=1

 (z))2 − 4(℘ (z))3 + 60G ℘ (z) + 140G is an elliptic

Now, clearly (℘a,b a,b 4 a,b 6
function, which can have a possible pole only at 0 in Da,b (0). The Laurent
series expansion of this function around 0 is obtained as follows.

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 416 — #26

i i

416 Weierstrass ℘a,b Function


∞
 (z))2 , then we first
If An zn is the Laurent series expansion of (℘a,b
n=−∞
0
find An zn (the singular part along with A0 ) as
n=−∞

 2
 2
(℘a,b (z))2 = − 3 + 6G4 z + 20G6 z + · · ·
3
z
4 1
= − 24G4 2 − 80G6 + A2 z2 + A3 z3 + · · ·
z6 z

(In the first line of the above expression, we have written the series expansion
 (z) up to the term corresponding to z3 explicitly. The reason is that
of ℘a,b
only these three terms which contribute to get the terms corresponding to zn,
n ≤ 0 in the square of this expansion.
Similarly, we write only the terms corresponding to zn , n ≤ 0, in the
expansion of ℘a,b (z))3 as follows.

 3
1
(℘a,b (z))3 = + 3G 4 z 2
+ 5G 6 z 6
+ · · ·
z2
 
1 9G4
= + + 15G6 + · · · .
z6 z2

Therefore, the Laurent series expansion of

 (z))2 − 4(℘ (z))3 + 60G ℘ (z) + 140G is
(℘a,b a,b 4 a,b 6

   
4 24G4 1 9G4
− 2 − 80G6 + · · · − 4 6 + 2 + 15G6 + · · ·
z6 z z z
1
+60G4 2 + 140G6 + B1 z + B2 z2 + B3 z3 + · · · ,
z

which is equal to B1 z + B2 z2 + B3 z3 + · · · for some B1 , B2 , B3 , · · · ∈ C.

 (z))2 − 4(℘ (z))3 + 60G ℘ (z) + 140G = 0. Thus,
Therefore, lim (℘a,b a,b 4 a,b 6
z→0
 (z))2 −4(℘ (z))3 +60G ℘ (z)+140G has no poles in D (0). Hence,
(℘a,b a,b 4 a,b 6 a,b
it is an entire elliptic function, and hence, it is a constant function, by Lemma
8.2.7. As the value of the constant is same as the limit of the function at 0, we
conclude that (℘a,b  (z))2 − 4(℘ (z))3 + 60G ℘ (z) + 140G = 0. Hence,
a,b 4 a,b 6
the result follows. 

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 417 — #27

i i

Elliptic Functions 417

 = 6℘
Exercise 8.3.9 Prove that ℘a,b a,b − 30G4 .

RESULT 8.3.10 The Weierstrass function ℘a,b satisfies the following differ-
ential equation
 2    

℘a,b = 4 ℘a,b − e1 ℘a,b − e2 ℘a,b − e3 ,

a    
where e1 = ℘a,b 2 , e2 = ℘a,b b
2 and e3 = ℘a,b a+b
2 .

Proof: From Case 1 of proof Theorem 8.3.6, we have the following observa-
tions:
 in the fundamental parallelogram D (0) are a ,
1. The only zeroes of ℘a,b a,b 2
b a+b  2
2 and 2 , and each is a simple zero. Thus, the only zeroes of (℘a,b )
are double zeroes at a2 , b2 and a+b
2 .
   
2. The only zeroes of ℘a,b − e1 ℘a,b − e2 ℘a,b − e3 in Da,b (0) are
a b a+b
2 , 2 and 2 , and each zero has multiplicity 2.

Since 0 is the only pole of ℘a,b in Da,b (0), which is of order 2,

   
℘a,b − e1 ℘a,b − e2 ℘a,b − e3

has a pole at 0 in Da,b (0) of order 6. Similarly, ℘a,b  has the only pole at
 2
0 in Da,b (0) of order 3, and hence (℘a,b ) has a pole at 0 in Da,b (0) which
 )2 and ℘
   
is of order 6. Therefore, (℘a,b a,b − e1 ℘a,b − e2 ℘a,b − e3
have same number of zeroes and poles of same orders in C. Therefore,
 )2
(℘a,b
 2
is a constant, say c . Thus, we have ℘  =
(℘a,b −e1 )(℘a,b−e
 2 )(℘a,b −e3)  a,b
c ℘a,b − e1 ℘a,b − e2 ℘a,b − e3 . Next, we find the value of c as follows.
As
⎛ ⎞2
2
lim z6 (℘a,b 
(z))2 = lim z6 ⎝ ⎠ = 22 = 4
z→0 z→0 (z − (ma + nb))3
m,n∈Z

and
⎛ ⎞
 
  1 1 1
lim z 2
℘a,b (z) − e1 = lim z2 ⎝ + − 2 ⎠ = 1,
z→0 z→0 z2 (z − p)2 p
p∈Qa,b \{0}

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 418 — #28

i i

418 Weierstrass ℘a,b Function

we get
 )2
(℘a,b
c =    
℘a,b − e1 ℘a,b − e2 ℘a,b − e3
 (z))2
(℘a,b
= lim    
z→0 ℘a,b (z) − e1 ℘a,b (z) − e2 ℘a,b (z) − e3
 (z))2
lim z6 (℘a,b
z→0
=    
lim z6 ℘a,b (z) − e1 ℘a,b (z) − e2 ℘a,b (z) − e3
z→0
= 4.

THEOREM 8.3.11 Prove that the unique elliptic

 function  of order 2 with the
fundamental periods a and b satisfying lim f (z) − z2 = 0 is ℘a,b .
1
z→0

Proof: From the definition of ℘a,b , clearly we have that ℘a,b is an elliptic
function of order 2 (as the only pole of ℘a,b in Da,b (0) is at 0 and it is a
double pole) and
   
1 1 1
lim ℘a,b (z) − 2 = lim − = 0.
z→0 z z→0 (z − p)2 p2
p∈Qa,b \{0}

Conversely, let f be an elliptic

 function of order 2 with the fundamental pe-
riods a and b satisfying lim f (z) − z12 = 0. First, we note that f should
z→0
have a pole of order 2 at 0. As a and b are the fundamental periods of f , f
should have double poles at every point of Qa,b . Since f is an elliptic function
of order 2, it should not have non-zero poles in Da,b (0). Thus, the set of all
poles of f are same as Qa,b . Now, we have
   
  1 1
lim f (z) − ℘a,b (z) = lim f (z) − − ℘a,b (z) − =0
z→0 z→0 z2 z2
and hence, f − ℘a,b has a removable singularity at 0. As f − ℘a,b is elliptic
with fundamental periods a and b, it has removable singularity at every point
of Qa,b . By defining f (z) − ℘a,b (z) = 0, ∀z ∈ Qa,b , the singularity at every
point of Qa,b is removed and f − ℘a,b becomes an entire elliptic function.
Thus, f − ℘a,b becomes a constant function. By seeing its value at a point of
Qa,b , we conclude that f = ℘a,b . 

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 419 — #29

i i

Elliptic Functions 419

THEOREM 8.3.12 (First addition theorem for ℘a,b )

For any u, v ∈ C \ Qa,b , we have
  (u) 
 ℘a,b ℘a,b (u) 1 
 
  (v)
℘a,b ℘a,b (v) 1  = 0.

 −℘  (u + v) ℘a,b (u + v) 1 
a,b

Proof: Since ℘a,b is doubly periodic and even if u ± v ∈ Qa,b , then ℘a,b (u) =
 (u) = ℘  (v), and hence, the determinant is 0. Therefore, we
℘a,b (v) and ℘a,b a,b
assume that u ± v  Qa,b . Now, we consider the following system of linear
equations in two variables α and β .
 
℘a,b (u) = α℘a,b (u) + β and ℘a,b (v) = α℘a,b (v) + β.

As the determinant of the coefficient matrix

 
 ℘a,b (u) 1 
 
 ℘a,b (v) 1  = ℘a,b (u) − ℘a,b (v)  0,

the above system of linear equations has a unique solution, say (α0 , β0 ). If

f (z) = ℘a,b (z) − α0 ℘a,b (z) − β0

then clearly f is an elliptic function of order 3 and it has a triple pole at 0, as

 is of order 3 and it has a pole of order 3 at 0. Hence, by Lemmas 8.2.5
℘a,b
and 8.2.6, there are only three zeroes including multiplicities (say z1 , z2 and
z3 ) in Da,b (0), and z1 + z2 + z3 ∈ Qa,b . By the choice of α0 and β0 , we know
that u and v are the two known zeroes of f , and hence, the other zero must be
−(u + v) + p for some p ∈ Qa,b . In other words, we have

℘a,b (−(u + v)) − α0 ℘a,b (−(u + v)) − β0 = 0.

We combine this equation with the known two equations

 
℘a,b (u) − α0 ℘a,b (u) − β = 0 and ℘a,b (v) − α0 ℘a,b (v) − β0 = 0

into a matrix equation as follows.

⎛  (u) ⎞⎛ ⎞ ⎛ ⎞
℘a,b ℘a,b (u) 1 1 0
⎝  (v)
℘a,b ℘a,b (v) 1 ⎠ ⎝ −α0 ⎠ = ⎝ 0 ⎠ .
 (−(u + v)) ℘ (−(u + v)) 1
℘a,b −β0 0
a,b

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 420 — #30

i i

420 Weierstrass ℘a,b Function

Thus, the system of homogeneous linear equation

⎛  (u) ⎞⎛ ⎞ ⎛ ⎞
℘a,b ℘a,b (u) 1 A 0
⎝  (v)
℘a,b ℘a,b (v) 1 ⎠⎝ B ⎠ = ⎝ 0 ⎠
 (−(u + v)) ℘ (−(u + v)) 1
℘a,b C 0
a,b

in three variables A, B and C has a non-trivial solution A = 1, B = −α0 and

C = α0 . Hence, from the theory of system linear equations, the coefficient
matrix should be singular or equivalently
  (u) 
 ℘a,b ℘a,b (u) 1 
 
  (v)
℘a,b ℘a,b (v) 1  = 0.

 −℘  (u + v) ℘a,b (u + v) 1 
a,b

 is an odd
Here, we have used the fact that ℘a,b is an even function and ℘a,b
function. 

THEOREM 8.3.13 (Second addition theorem for ℘a,b )

1. For every u, v ∈ C \ Qa,b with u ± v  Qa,b , we have

 (u) − ℘  (v)
#2
1 ℘a,b a,b
℘a,b (u + v) = − ℘a,b (u) − ℘a,b (v).
4 ℘a,b (u) − ℘a,b (v)

2. For every u ∈ C \ Qa,b with 2u  Qa,b , then

 (u)
#2
1 ℘a,b
℘a,b (2u) =  (u) − 2℘a,b (u).
4 ℘a,b

Proof: Let u, v ∈ C \ Qa,b with u ± v  Qa,b . Retaining the notations used

in the previous theorem, we have ℘a,b (z) = α0 ℘a,b (z) + β0 , ∀z ∈ S , where
S consists of all points of C \ Qa,b with z − u ∈ Qa,b or z − v ∈ Qa,b or
z + u + v ∈ Qa,b . Hence, we have

(℘a,b (z))2 = (α0 ℘a,b (z) + β0 )2 , ∀z ∈ S.

Using Result 8.3.8, we have

4(℘a,b (z))3 − 60G4 ℘a,b (z) − 140G6 − (α02 (℘a,b (z))2 + 2α0 β0 ℘a,b (z) + β02 ) = 0

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 421 — #31

i i

Elliptic Functions 421

and hence,
4(℘a,b (z))3 −α02 (℘a,b (z))2 −(2α0 β0 +60G4 )℘a,b (z)−(140G6 +β02 ) = 0, ∀z ∈ S.
By the choice of u and v, the three complex numbers ℘a,b (u), ℘a,b (v), and
℘a,b (u + v) are pairwise distinct, and hence, they are the only solutions of the
cubic polynomial equation
4X 3 − α02 X 2 − (2α0 β0 + 60G4 )X − (140G6 + β02 ) = 0.
Therefore, the sum of the roots of the equation ℘a,b (u) + ℘a,b (v) + ℘a,b (u + v)
 (u)−℘  (v)
℘a,b
equals 14 α02 . Now, we find α0 = a,b
℘a,b (u)−℘a,b (v) by solving from the system of
linear equations
 
℘a,b (u) = α0 ℘a,b (u) + β0 and ℘a,b (v) = α0 ℘a,b (v) + β0 .
Thus, we get
 (u) − ℘  (v)
#2
1 ℘a,b a,b
℘a,b (u + v) = − ℘a,b (u) − ℘a,b (v). (8.6)
4 ℘a,b (u) − ℘a,b (v)

If u ∈ C \ Qa,b such that 2u  Qa,b , then allowing v → u on both sides of

equation (8.6), we obtain
℘a,b (2u) = lim ℘a,b (u + v)
v→u
 (u) − ℘  (v)
#2
1 ℘a,b a,b
= lim − ℘a,b (u) − lim ℘a,b (v)
v→u 4 ℘a,b (u) − ℘a,b (v) v→u
⎛ ℘  (v)−℘  (u) ⎞2
a,b a,b
1⎝ v−u ⎠ − ℘a,b (u) − ℘a,b (u)
= lim ℘a,b (v)−℘a,b (u)
v→u 4
v−u
 (u)
#2
1 ℘a,b 
=  (u) − 2℘a,b (u)(as 2u  Qa,b ⇒ ℘a,b (u)  0).
4 ℘a,b

Thus, the theorem follows. 

The second assertion of the above theorem is called the duplication
formula for ℘a,b .

COROLLARY 8.3.14 For u, v ∈ C \ Qa,b , ℘a,b satisfies

(2℘a,b (u)℘a,b (v)−30G4 )(℘a,b (u)+℘a,b (v))−140G6
℘a,b (u+v)+℘a,b (u−v) = .
(℘a,b (u)−℘a,b (v))2

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 422 — #32

i i

422 Weierstrass ℘a,b Function


Proof: Applying Result 8.3.8 in equation (8.6) and using the oddity of ℘a,b
and the evenness of ℘a,b , we get
℘a,b (u + v) + ℘a,b (u − v)
 
#2
1 ℘a,b (u) − ℘a,b (v)
= − ℘a,b (u) − ℘a,b (v)
4 ℘a,b (u) − ℘a,b (v)
 
#2
1 ℘a,b (u) + ℘a,b (v)
+ − ℘a,b (u) − ℘a,b (v)
4 ℘a,b (u) − ℘a,b (v)
 
1 (℘a,b (u)) + (℘a,b (v))
2 2
=   − 2(℘a,b (u) + ℘a,b (v))
2 ℘a,b (u) − ℘a,b (v) 2

 (u))2 + (℘  (v))2 − 4(℘ (u) + ℘ (v)) ℘ (u) − ℘ (v) 2

(℘a,b a,b a,b a,b a,b a,b
=  2
2 ℘a,b (u) − ℘a,b (v)
1 $
=  2 4(℘a,b (u)) − 60G4 ℘a,b (u) − 140G6
3
2 ℘a,b (u) − ℘a,b (v)
%
+ 4(℘a,b (v))3 − 60G4 ℘a,b (v) − 140G6
$ %
− 4 (℘a,b (u))3 + (℘a,b (v))3 − ℘a,b (u)(℘a,b (v))2 − (℘a,b (u))2 ℘a,b (v)
1 
=  2 −60G4 (℘a,b (u) + ℘a,b (v)) − 280G6
2 ℘a,b (u) − ℘a,b (v)

+ 4℘a,b (u)℘a,b (v)(℘a,b (u) + ℘a,b (v))
(2℘a,b (u)℘a,b (v) − 30G4 )(℘a,b (u) + ℘a,b (v)) − 140G6
=  2 .
℘a,b (u) − ℘a,b (v)
Hence, the corollary follows. 

COROLLARY 8.3.15 For u, v ∈ C \ Qa,b , ℘a,b satisfies

℘a,b (u)℘a,b (v)
℘a,b (u + v) − ℘a,b (u − v) = − .
(℘a,b (u) − ℘a,b (v))2
The proof of this corollary is similar to that of the previous one.

COROLLARY 8.3.16 For every u ∈ C \ Qa,b ,

 a  (e1 − e2 )(e1 − e3 )
℘a,b u + = + e1 ,
2 ℘a,b (u) − e1
     
where e1 = ℘a,b a2 , e2 = ℘a,b b2 and e3 = ℘a,b a+b 2 .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 423 — #33

i i

Elliptic Functions 423

 2
Proof: Using the value of ℘a,b (u) from Result 8.3.10 in equation (8.6), we
get  
℘a,b u + a2
 (u)
#2
1 ℘a,b
= − ℘a,b (u) − e1
4 ℘a,b (u) − e1
(℘a,b (u) − e1 )(℘a,b (u) − e2 )(℘a,b (u) − e3 )
= − ℘a,b (u) − e1
(℘a,b (u) − e1 )2
(℘a,b (u) − e2 )(℘a,b (u) − e3 )
= − (℘a,b (u) + e1 )
(℘a,b (u) − e1 )
(℘a,b (u))2 − (e2 + e3 )℘a,b (u) + e2 e3 − (℘a,b (u))2 + e21
=
(℘a,b (u) − e1 )
e1 ℘a,b (u) + e2 e3 + e21
=
(℘a,b (u) − e1 )
(as e1 + e2 + e3 = 0)
e1 ℘a,b (u) + (e1 − e2 )(e1 − e3 ) − 2e21 + e21
=
(℘a,b (u) − e1 )
(as (e1 − e2 )(e1 − e3 ) = e21 + e2 e3 − (e2 + e3 )e1 = 2e21 + e2 e3 )
(e1 − e2 )(e1 − e3 )
= + e1 .
(℘a,b (u) − e1 )
Thus, the corollary follows. 
a √
COROLLARY 8.3.17 ℘a,b 4 = e1 ± (e1 − e2 )(e1 − e3 ).

Proof: In the previous corollary, substituting u = − a4 , we get

a  a a  (e − e )(e − e )
1 2 1 3
℘a,b = ℘a,b − + =   + e1
4 4 2 (℘a,b a4 − e1 )
&   '2
and hence, ℘a,b a4 − e1 = (e1 − e2 )(e1 − e3 ). Thus, the corollary
follows. 

8.4 THE FUNCTIONS ζa,b AND σa,b

Definition 8.4.1 Define
 
1 1 1 z
ζa,b (z) = + + + 2 , ∀z ∈ C \ Qa,b .
z z−p p p
0p∈Qa,b

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 424 — #34

i i

424 The Functions ζa,b and σa,b

LEMMA 8.4.2 The series

 
1 1 z
+ +
z − p p p2
0p∈Qa,b

converges uniformly in C \ Qa,b , and the anti-derivative of ℘a,b is −ζa,b .

Proof: We shall show that the series converges uniformly on Cl B(0, r) for
every r > 0. Using Lemma 8.3.2, first we find μ > 0 such that |ma + nb| ≥
2r
μ(|m| + |n|), ∀(m, n) ∈ Z. Now, choose N0 ∈ N such that N0 > so that
μ

|ma + nb| ≥ μ(|m| + |n|) ≥ μN0 > 2r, ∀(m, n) ∈ Z2 with |m| + |n| ≥ N0 .
a
(Note that as  R, we have a  0  b.)
b  
 z 
Now, for |z| ≤ r and for (m, n) ∈ Z2 with |m| + |n| ≥ N0 , we have  ma+nb <
1
2 , and hence,
 
 1 1 z 
 
 z − (ma + nb) + ma + nb + (ma + nb)2 
 ! " 
 −1 
 1 1 z 
=  + + 
 ma + nb 1 − ma+nb
z
ma + nb (ma + nb)2 
 ! " 
 −1 ∞
zk 
 z 1 z 
=  1+ + + + 
 ma + nb ma + nb |ma + nb|k ma + nb (ma + nb)2 
k=2
 
 −1 ∞
zk 
 
=  
 ma + nb (ma + nb) 
k
k=2
∞
|z|2 |z|k
≤
|ma + nb|3 |ma + nb|k
k=0
|z|2 1
≤
|ma + nb| 1 − |z|
3
|ma+nb|
|z|2 1
≤
|ma + nb|3 1 − 1
2
2|z|2 1
≤ ≤ 4r2 ,
|ma + nb| 3 |ma + nb|3

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 425 — #35

i i

Elliptic Functions 425

 1
which is a typical term of the convergent series |ma+nb|3
. (See the
m,n∈Z
|m|+|n|0
proof of Lemma 8.3.3.) Thus, by comparison test, the series
 
1 1 z
+ +
m,n∈Z
z − (ma + nb) ma + nb (ma + nb)2
|m|+|n|0

converges uniformly on Cl B(0, r) \ Qa,b , and hence, on every compact subset

 −1   −1 
of C \ Qa,b . Therefore we have, ζa,b (z) = z2 + (z−a)2
+ 1
p2
=
0p∈Qa,b
−℘a,b (z). 

RESULT 8.4.3 There exist constants α, β ∈ C such that ζa,b (z+a)−ζa,b (z) =
α and ζa,b (z + b) − ζa,b (z) = β , ∀z ∈ C \ Qa,b . Furthermore, they satisfy
αb − βa = 2πi.
 = −℘ . As a is a period of
Proof: From the previous lemma, we have ζa,b a,b
℘a,b , we get
 
ζa,b (z + a) − ζa,b (z) = 0 ⇒ ℘a,b (z + a) − ℘a,b (z) = 0, ∀z ∈ C \ Qa,b .
Thus, there exists a constant α ∈ C such that
ζa,b (z + a) − ζa,b (z) = α, ∀z ∈ C \ Qa,b .
Similarly, we can prove that there exists β ∈ C such that
ζa,b (z + b) − ζa,b (z) = β, ∀z ∈ C \ Qa,b
Let Da,b (z0 ) be a fundamental parallelogram such that ∂Da,b (z0 ) does not have
any pole of ζa,b . Since the set of all poles of ζa,b is same as Qa,b , Da,b (z0 )
should contain exactly one pole of ζa,b . We also know that Res (ζa,b , p) = 1,
∀p ∈ Qa,b , as the coefficient of z−p
1
in the Laurent series expansion of ζa,b in
B(a, r) \ {a} is 1, for some r > 0, ∀p ∈ Qa,b . Note that a parametric equation
of ∂Da,b (z0 ) is given as follows:
z0 + a + b

γ3 γ 1 (t) = z0 + ta, t ∈ [0, 1]

γ2 γ 2 (t) = z0 + a + tb, t ∈ [0, 1]
z0 + b
γ 3 (t) = z0 − ta + b, t ∈ [− 1, 0]
z0 + a γ 4 (t) = z0 − ta, t ∈ [− 1, 0].
γ4
γ1
z0

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 426 — #36

i i

426 The Functions ζa,b and σa,b

Therefore, by Cauchy’s residue theorem, we get


i2π = ζa,b (z) dz
∂Da,b (z0 )

1 1
= ζa,b (z0 + ta) adt + ζa,b (z0 + a + tb) bdt
0 0
0 0
− ζa,b (z0 − ta + b) adt − ζa,b (z0 − tb) bdt
−1 −1
1 1
& '
= a ζa,b (z0 + ta) dt + b ζa,b (z0 + tb) + α dt
0 0
1 1
& '
−a ζa,b (z0 + sa) + β ds − ζa,b (z0 + sb) ds
0 0
1 1
= −a β dt + b α dt
0 0
= αb − βa.

Remark 8.4.4: From the previous proposition, we note that ζa,b is not a
periodic function.

PROBLEM 8.4.5 Prove that ζa,b is an odd function.

Solution: Let z ∈ C \ Qa,b be arbitrary. Now,

 
1 1 1 −z
ζa,b (−z) = − + + + 2
z −z − p p p
p∈Qa,b \{0}
⎡ ⎤
 
1 1 1 z ⎦
= −⎣ + − +
z z + p p p2
p∈Qa,b \{0}

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 427 — #37

i i

Elliptic Functions 427

⎡ ⎤
 
1 1 1 z ⎦
= −⎣ + + +
z z − p p p2
p∈Qa,b \{0}
(as Qa,b \ {0} = −(Qa,b \ {0}))
= −ζa,b (z).

PROBLEM 8.4.6 Prove that

& '2
℘a,b (z + w) + ℘a,b (z) + ℘a,b (w) = ζa,b (z + w) − ζa,b (z) − ζa,b (w) ,
∀z, w ∈ C \ Qa,b .

Solution: Let w ∈ C be fixed arbitrarily in C \ Qa,b . If

& '2
f (z) = ℘a,b (z + w) + ℘a,b (z) + ℘a,b (w) − ζa,b (z + w) − ζa,b (z) − ζa,b (w)

for every z ∈ C \ Qa,b , then f is analytic on C \ Qa,b . Now, we show that f is

an entire function after removing the removable singularities. If

f1 (z) = ℘a,b (z + w) + ℘a,b (w)

f2 (z) = ζa,b (z + w) − ζa,b (w)
 
1 1
Ra,b (z) = − 2
(z − p)2 p
p∈Qa,b \{0}
 
1 1 z
ρa,b (z) = + + 2 , ∀z ∈ C
(z − p) p p
p∈Qa,b \{0}

then we can write

 2
1 1
f (z) = + R a,b (z) + f1 (z) − f2 (z) − − ρa,b (z)
z2 z
1  2 1 2 
= 2
+ Ra,b (z) + f1 (z) − f2 (z) − ρa,b (z) − 2 + f2 (z) − ρa,b (z)
z z z
 2 2  
= Ra,b (z) + f1 (z) − f2 (z) − ρa,b (z) + f2 (z) − ρa,b (z) .
z
Since
 
1 1
f2 (0) − ρa,b (0) = ζa,b (w) − ζa,b (w) − + = 0,
(−p) p
p∈Qa,b \{0}

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 428 — #38

i i

428 The Functions ζa,b and σa,b

z ( f2 (z) − ρa,b (z)) has a removable singularity at 0. Hence, after removing the
2

singularity at 0, 2z ( f2 (z) − ρa,b (z)) becomes analytic in a neighbourhood of 0.

Hence, f becomes analytic in a neighbourhood of 0. By a similar argument,
we can prove that f is analytic in a neighbourhood of p, ∀p ∈ Qa,b . Thus, f
becomes analytic on C.
Next, we claim that f is a periodic function with Pf = Qa,b . To prove this
claim, first we prove that ζa,b (z + w) − ζa,b (z) − ζa,b (w) is a periodic function
of periods a and b (as a function of z only). As in the proof of Result 8.3.4,
let
F(z) = ζa,b (z + w + a) − ζa,b (z + a) − ζa,b (z + w) + ζa,b (z), ∀z ∈ C \ Qa,b .
Then, F is analytic on C \ Qa,b and
F  (z) = 
ζa,b 
(z + w + a) − ζa,b 
(z + a) − ζa,b 
(z + w) + ζa,b (z)
= −℘a,b (z + w + a) + ℘a,b (z + a) + ℘a,b (z + w) − ℘a,b (z)
( by Lemma 8.4.2 )
= 0. (as a is a period of ℘a,b ).
Therefore, F is constant on C \ {0}, and hence, it is constant on C. To find the
value of the constant, using the fact that ζa,b is an odd function, we find
       
w+a w+a −w + a w−a
F − = ζa,b − ζa,b − ζa,b
2 2 2 2
 
−w − a
+ζa,b
2
     
w+a w−a w−a
= ζa,b + ζa,b − ζa,b
2 2 2
 
w+a
−ζa,b
2
= 0.
Therefore, F is identically zero on C, and hence, ζa,b (z+w)−ζa,b (z)−ζa,b (w)
is a periodic function of period a. Similarly, we can prove that b is also a
period of ζa,b (z + w) − ζa,b (z) − ζa,b (w). Recalling that P℘a,b = Qa,b , we
conclude that Pf = Qa,b . 

   z
+ 12 z2
LEMMA 8.4.7 The infinite product 1− z
p ep p2 converges uni-
p∈Qa,b \{0}
formly on every compact subset of C.

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 429 — #39

i i

Elliptic Functions 429

Proof: For an arbitrarily fixed w ∈ C with |w| < 1, if we define ϕw (t) :

[0, 1] → C by

(tw)2
ϕ(t) = (1 − tw)etw+ 2 − 1, t ∈ [0, 1]

then ϕ is a differentiable function on [0, 1]. Applying complex version of

mean value theorem (Theorem 2.1.19), we first observe that if |w| < 1, then
 
 2 
(1 − w)ew+ w2 − 1 = |ϕ(1) − ϕ(0)|
 
 
≤ ϕ  (s) ( for some s ∈ (0, 1) )
|sw|2
= s2 e|sw|+ 2 |w|3
< |w|3 e3/2 .

Let K be a compact subset of C. Then, we choose M > such that |z| ≤

M, ∀z ∈ K . Now, for z ∈ K , we have
  z 1 z2   3
   z  3/2
 z +    e ≤ M 3 e3/2 1
 1− e p 2 p2 −1 ≤ p .
 p  |p|3
p∈Qa,b \{0} p∈Qa,b \{0} p∈Qa,b \{0}

The last series is convergent. (Indeed, see the proof of Lemma

8.3.3). Hence, by Theorem 5.4.5, we get that the product
   z + 1 z2
1 − pz e p 2 p2 converges absolutely and uniformly on K
p∈Qa,b \{0}
 
   z + 1 z2 

(because  1− p ez p 2 p2
− 1 converges uniformly on K ). 
p∈Qa,b \{0}  

Definition 8.4.8 Define

   z 1 z2
z +
σa,b (z) = z 1− e p 2 p2 , ∀z ∈ C
p
p∈Qa,b \{0}

Clearly, σa,b is an entire function with simple zeroes at every p ∈ Qa,b .

 (z)
σa,b
RESULT 8.4.9 The function σa,b is an odd function and it satisfies σa,b (z) =
ζa,b (z), ∀z ∈ C \ Qa,b .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 430 — #40

i i

430 The Functions ζa,b and σa,b

Proof: For z ∈ C, using the fact that Qa,b \ {0} = −(Qa,b \ {0}), we get

   (−z) 1 (−z)2
(−z) +
σa,b (−z) = (−z) 1−
e p 2 p2
p
p∈Qa,b \{0}
  z
 z 1 z2
+
= −z 1− e (−p) 2 p2
(−p)
p∈Qa,b \{0}
  z
 z 1 z2
+
= −z 1− e p 2 p2
p
p∈Qa,b \{0}
= −σa,b (z).

Thus, σa,b is an odd function.

Finding the logarithmic derivative of σa,b , we immediately get
 (z)
!   "
σa,b 1 1 1 z −1
= + + + 2
σa,b (z) z z
1− p p p
p∈Qa,b \{0} p
 
1 1 1 z
= + + +
z z − p p p2
p∈Qa,b \{0}
= ζa,b (z).

Hence, the result follows. 

RESULT 8.4.10 Let f be an elliptic function such that

1. Pf = Qa,b ,

2. the only zeroes of f in Da,b (0) are z1 , z2 , . . . , zn including multiplicities,

3. the only poles of f in Da,b (0) are w1 , w2 , . . . , wn including

multiplicities.

Then, there exists c ∈ C such that


n
σa,b (z − zj )
f (z) = c , ∀z ∈ C \ Sf ,
σa,b (z − wj )
j=1

where Sf is the set of all singularities of f .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 431 — #41

i i

Elliptic Functions 431

Proof: As σa,b is an entire function with simple zero at 0, we observe that

σa,b (z − s) is also an entire function with a simple zero at s, for each fixed
n
σa,b (z−zj )
s ∈ C. If g(z) = σa,b (z−wj ) , ∀z ∈ C \ Sf , then g has zeroes at zj and poles
j=1
at wj . We claim that a and b are periods of g. From Result 8.4.3, we have

ζa,b (z + a) − ζa,b (z) = α and ζa,b (z + b) − ζa,b (z) = β

for some α, β ∈ C. Using the previous result, we obtain

 (z + a)
σa,b  (z)
σa,b  (z + b)
σa,b  (z)
σa,b
− = α and − = β,
σa,b (z + a) σa,b (z) σa,b (z + b) σa,b (z)

which imply that

 
σa,b (z + a)σa,b (z) − σa,b (z)σa,b (z + a) − ασa,b (z + a)σa,b (z) = 0. (8.7)
 
σa,b (z + b)σa,b (z) − σa,b (z)σa,b (z + b) − βσa,b (z + b)σa,b (z) = 0. (8.8)

For$ z ∈ C \ Sf , consider
d σa,b (z+a)
  %
dz σa,b (z) exp −α z + a
2

!  "
 
a  σa,b (z + a) σa,b (z + a)σa,b (z)

= exp −α z + −
2 σa,b (z) σa,b
2 (z)
  a  σa,b (z + a)
−α exp −α z +
2 σa,b (z)
   $
exp −α z + 2 a
 
= σa,b (z + a)σa,b (z) − σa,b (z + a)σa,b (z)
σa,b
2 (z)
'
−ασa,b (z + a)σa,b (z)
= 0(by applying equation (8.7)).

σ (z+a)   
σa,b (z) exp −α z + 2 is a constant function on C \ Qa,b , and
a
Therefore, a,b
hence, it is a constant function (say C ) on C. By substituting z = − a2 , we get
σ (a)  
C = σ a,b −2a = −1 because σa,b is an odd function and σa,b a2  0. Thus,
a,b ( 2 )
we have   a 
σa,b (z + a) = −σa,b (z) exp α z + (8.9)
2

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 432 — #42

i i

432 Jacobi’s Elliptic Functions snk , cnk and dnk

For z ∈ C \ Sf , using equation (8.9), it follows that

σa,b (z − zj + a) σa,b (z − zj )
=
σa,b (z − wj + a) σa,b (z − wj )

and hence, g(z + a) = g(z), ∀z ∈ C \ Qa,b . By a similar argument, we prove

that b is also a period of g. Thus, g is an elliptic function with fundamental1
periods a and b. Therefore, the zeroes of g and the poles of g are precisely
{zj + p : p ∈ Qa,b , 1 ≤ j ≤ n} and {wj + p : p ∈ Qa,b , 1 ≤ j ≤ n},
respectively, as the the zeroes and poles of g inside the fundamental region
Da,b (0) are {zj : 1 ≤ j ≤ n} and {wj : 1 ≤ j ≤ n}, respectively. Thus, by
Lemma 8.3.5, we get f = cg for some c ∈ C. 

8.5 JACOBI’S ELLIPTIC FUNCTIONS snk , cnk

AND dnk

We start this section with the motivation for introducing elliptic integrals
from the problem of simple pendulum.
A simple pendulum consists of a weightless thread of length l, whose one
end is attached to a sphere of mass m and the other end is attached to a fixed
point on the roof. By drawing the sphere aside from the equilibrium position,
we assume that the sphere oscillates in a same vertical plane. The problem
is to determine the angle θ between the current position of thread with its
equilibrium position, as a function of time t.

l cos (q )
q
l

l (1−cos (q ))
)
(q
sin
mg

1 This statement can be obtained from the fact that a and b are fundamental periods of ℘
a,b
and by the relations among ℘a,b , ζa,b and σa,b .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 433 — #43

i i

Elliptic Functions 433

Applying the Newton’s second law, the force acting upon the sphere due
to this movement is same as the tangential component of force acting upon
the sphere due to gravity. Thus, we get
d2 d2θ g
m 2
(lθ) = −mg sin(θ), ⇒ 2
+ sin(θ ) = 0,
dt dt l
where lθ is the arc length of the path of the sphere from equilibrium position
to the present time. Multiplying on both sides of the above equation by ml2 dθ
dt ,
we get
!   "
2 dθ d θ
2 dθ d 1 dθ 2
ml +mgl sin θ =0⇒ m l + mgl(1 − cos(θ )) = 0.
dt dt2 dt dt 2 dt

Thus, we get the total energy (kinetic energy + potential energy) acting upon
the sphere at time t, which is constant. As the energy required to move the
sphere from the equilibrium position (corresponding to θ = 0) to the highest
position (corresponding to θ = π ) is k 2 (2mgl) for some 0 < k < 1 (because
of the oscillatory motion), we have
 
1 dθ 2
l + gl(1 − cos(θ )) = k 2 (2gl). (8.10)
2 dt
If the initial velocity of the
 sphere is v0 , then by the law of conservation of
 2 
energy, then we have 12 m v20 − l dθdt = mgl(1 − cos(θ )) and so we have

 2
dθ
l = v20 − 2gl(1 − cos(θ )). (8.11)
dt
v20
Comparing equations (8.10) and (8.11), we find that k 2 = 4gl . Thus, using

the change of variable x = gl t in equation (8.11), we get
 2   
dθ θ
= 4 k 2 − sin2 . (8.12)
dx 2
To obtain the Euler’s normal form and Jacobi’s normal form  of
 equation
(8.12), we consider the change of variable k sin (φ) = sin θ2 , which is
 θ  dθ
introduced by Euler and Jacobi. As k cos(φ) dφ
dx = 2 cos 2 dx , we have
1

   2   
4k 2 cos2 (φ) dφ 2 dθ 2 θ
θ  = = 4 k 2
− sin = 4k 2 (1 − sin2 (φ))
2
cos 2 dx dx 2

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 434 — #44

i i

434 Jacobi’s Elliptic Functions snk , cnk and dnk

 2    
and so dφ
dx = cos2 θ2 = 1 − sin2 θ2 = 1 − k 2 sin2 (φ). Thus, the Euler’s
normal form is obtained as
 2
dφ
= 1 − k 2 sin2 (φ). (8.13)
dx
In this equation, if we use another change of variable (due to Jacobi) y =
  k −1
 θ  dθ
sin(φ) = k −1 sin θ2 , then we obtain dy
dx = 2 cos 2 dx , and hence, using
equation (8.12), we also get
 2     
dy k −2 θ θ
= cos2 4 k 2 − sin2
dx 4 2 2
     
θ θ
= 1 − sin2 1 − k −2 sin2
2 2
= (1 − k 2 y2 )(1 − y2 ), (8.14)
which is called Jacobi’s normal form. Now, we find the solution of equation
(8.13) as follows.
From equation (8.13), we have

dφ
= 1 − k 2 sin2 (φ). (8.15)
dx
dθ
Using the initial conditions θ = 0 and = 0 at t = 0, we have θ = 0
dt
dθ
and = 2k , and hence, we get φ = 0 and dφ dφ
dx = 1 at x = 0. Thus, dx is
dx
invertible in a small neighbourhood of x = 0, and hence, the derivative of x
with respect to φ is obtained simply by
dx 1
=  , (8.16)
dφ
1 − k 2 sin2 (φ)

which has the solution

φ
du
x = Fk (φ) =  . (8.17)
0 1 − k 2 sin2 (u)

The integral equation (8.17) is called the elliptic integral of first kind (See
first section of this chapter). The constant k is called the modulus, and x is
called the argument. If Fk is the function denoting φ → x, then Fk is an odd
and increasing function. Indeed,

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 435 — #45

i i

Elliptic Functions 435

1. using the change of variable v = −u in what follows, we get Fk (−φ) =

−φ
 φ
√ du 2 = − √ dv 2 = −Fk (φ).
2
1−k sin (u) 2 1−k sin (v)
0 0

2. the derivative dFk

dφ = √ 1
> 0 (by Theorem 4.1.11), and
1−k 2 sin2 (u)
hence, Fk is an increasing function (by Lemma 2.2.18).
Hence, the function Fk is invertible, and hence, its inverse map is also odd
and increasing function. It is customary to denote the inverse of Fk by amk .
We also note that

amk (x) = 1 − k 2 sin2 (amk (x)). (8.18)

If we denote Fk ( π2 ) by Kk , then Kk is the time taken (in the variable x) when

φ varies from 0 to π2 . Now, we see that the map Fk is a bijection from R onto
 
itself when 0 ≤ k < 1 and is a bijection from R onto − π2 , π2 when k = 1.

Definition 8.5.1 The basic Jacobian elliptic functions are defined as follows:
1. snk (x) = sin(amk (x)) = sin(φ),

2. cnk (x) = cos(amk (x)) = cos(φ),

 
3. dnk (x) = 1 − k 2 sin2 (φ) = 1 − k 2 sn2k (x),

where x and φ are related by the one-to-one correspondence x = Fk (φ).

φ
For the case k = 0, as x = F0 (φ) = du = φ , we have, by definition,
0

π
sn0 (x) = sin(x), cn0 (x) = cos(x), dn0 (x) = 1, and K0 = .
2
When k = 1, as

φ
x = F1 (φ) = sec(u)du = ln(sec(φ) + tan(φ))
0

we have
1 + sin(φ)
exp(x) = sec(φ) + tan(φ) =
cos(φ)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 436 — #46

i i

436 Jacobi’s Elliptic Functions snk , cnk and dnk

and hence,
(1 + sin(φ))2 (1 + sin(φ))2 1 + sin(φ)
exp(2x) = = = .
2
cos (φ) (1 − sin (φ))
2 1 − sin(φ)
Solving for sin(φ) from the above two equations, we get
exp(2x) − 1
(1 − sin(φ)) exp(2x) = 1 + sin(φ) ⇒ sin(φ) = = tanh(x).
exp(2x) + 1
Hence,
sn1 (x) = sin(φ) = tanh(x),
 
cn1 (x) = cos(φ) = 1 − sin2 (φ) = 1 − tanh2 (x) = sech(x),
 
dn1 (x) = 1 − sin2 (φ) = 1 − tanh2 (x) = sech(x),
π π
2 2
du
K1 =  = sec2 (u) du = ∞.
0 1 − sin (u)
2
0

RESULT 8.5.2 For any 0 ≤ k < 1,

1. snk (0) = 0 and cnk (0) = dnk (0) = 1.
√
2. snk (Kk ) = 1, cnk (Kk ) = 0 and dnk (Kk ) = j, where j = 1 − k 2 and
is called the complementary modulus.
3. snk (−x) = −snk (x), cnk (−x) = cnk (x) and dnk (−x) = dnk (x).
4. sn2k + cn2k = 1 and k 2 sn2k + dn2k = 1.

Proof: Directly from the definition of snk , cnk , and dnk , we get the follow-
ing:
1. From the definition of Fk , we have Fk (0) = 0, and hence, amk (0) =
0. Therefore, snk (0) = sin(amk (0)) = sin(0) = 0, cnk (0) =
cos(amk (0)) = cos(0) = 1, and dnk (0) = 1 − k 2 sn2k (0) = 1.
 
2. We note that Fk π2 = Kk , and hence, amk (Kk ) = π2 . Therefore,
π 
snk (Kk ) = sin(amk (Kk )) = sin = 1,
2
π 
cnk (Kk ) = cos(amk (Kk )) = cos = 0,
 2

dnk (Kk ) = 1 − k 2 sn2k (Kk ) = 1 − k 2 .

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 437 — #47

i i

Elliptic Functions 437

3. We know that amk and sin are odd functions, and hence,

snk (−x) = sin(amk (−x)) = sin(−amk (x)) = − sin(amk (x)) = −snk (x),

cnk (−x) = cos(−amk (x)) = cos(amk (x)) = cnk (x),

 
dnk (−x) = 1 − k 2 snk (−x) = 1 − k 2 sn2k (x) = dnk (x).

4. Finally, we also have

sn2k (x) + cn2k (x) = sin2 (amk (x)) + cos2 (amk (x)) = 1

and
dn2k (x) = 1 − k 2 sn2k (x) ⇒ dn2k (x) + k 2 sn2k (x) = 1.

Thus, the result follows. 

THEOREM 8.5.3 For 0 ≤ k ≤ 1,

snk (x + 2Kk ) = −snk (x), cnk (x + 2Kk ) = −cnk (x), dnk (x + 2Kk ) = dnk (x).

1
Proof: First, we note that  is a periodic function of period π
1 − k 2 sin2 (u)
(as sin2 is a periodic function of period π ). Let φ ∈ R be arbitrary. Applying
Lemma 2.4.23, twice (for a = − π2 and for a = φ ), we get
π
2 φ+π

du du
 =  . (8.19)
− π2 1 − k 2 sin (u)
2
φ 1 − k 2 sin2 (u)

For the given φ ∈ R, if x = Fk (φ), then

π
φ 2
du du
x + 2Kk =  +2 
0 1 − k 2 sin2 (u) 0 1 − k 2 sin2 (u)
π
φ 2
du du
=  + 
0 1 − k 2 sin2 (u) − π2 1 − k 2 sin2 (u)
(as the integrand of last integral is an even function.)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 438 — #48

i i

438 Jacobi’s Elliptic Functions snk , cnk and dnk

φ φ+π

du du
=  + 
0 1 − k 2 sin (u)
2
φ 1 − k 2 sin2 (u)
(Using equation (8.19))
φ+π

du
=  .
0 1 − k 2 sin2 (u)

Therefore, we have proved that

Fk (φ + π ) = x + 2Kk . (8.20)

Hence, snk (x + 2Kk ) = sin(φ + π) = − sin(φ) = −snk (x). Similarly, we can

prove that cnk (x+2Kk ) = −cnk (x) is a periodic function of period 4Kk . Now,
 
dnk (x + 2Kk ) = 1 − k 2 sn2k (x + 2Kk ) = 1 − k 2 sn2k (x) = dnk (x).

Hence, the theorem follows. 

As an immediate consequence of the previous theorem, we have

snk (x + 4Kk ) = −snk (x + 2Kk ) = snk (x).

Similarly, we also get cnk (x + 4Kk ) = cnk (x), and hence, the the following
corollary holds.

COROLLARY 8.5.4 snk and cnk are periodic functions of period 4Kk , and dnk
is a periodic function of period 2Kk .

RESULT 8.5.5 snk = cnk dnk , cnk = −snk dnk , and dnk = −k 2 snk cnk .

Proof: Let x and φ be related by x = Fk (φ). By a direct calculation, we get

snk (x) = d
$dx snk (x) % = d
dx (sin(φ))
dφ
= d
dφ sin(φ) = cos(φ) dam k
 dx dx

= cos(φ) 1 − k 2 sn2k (x) = cos(φ)dnk (x)

= $cnk (x)dnk (x),
%
cnk (x) = d
dφ cos(φ) dφ
dx = − sin(φ)dnk (x).

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 439 — #49

i i

Elliptic Functions 439


d
dnk (x) = 1 − k 2 sn2k (x)
dx
1 1
=  (−k 2 )2snk (x)snk (x)
2 1 − k 2 sn2 (x)
k
1
= −k 2  snk (x)cnk (x)dnk (x)
1 − k 2 sn2k (x)
= −k 2 snk (x)cnk (x).

Hence, the result follows. 

The quotients of the functions from {snk , cnk , dnk , 1} are denoted by
Glaisher as follows:

1 1 1
nsk = , nck = , ndk = ,
snk cnk dnk
snk snk cnk
sck = , sdk = , csk = ,
cnk dnk snk
cnk dnk dnk
cdk = , dsk = , dck = .
dnk snk cnk

So far the Jacobi’s elliptic functions are defined from R into R. To extend
them as functions from C into C, we use the Jacobi’s imaginary transforma-
tion which is given as follows:

sin(ψ) = i tan(φ). (8.21)

RESULT 8.5.6 If φ √ and ψ are related by equation (8.21), then Fk (ψ) =

iFj (φ), where j = 1 − k 2 . Moreover, ψ = log(sec(φ) + tan(φ)).

Proof: Using the change of variable sin(v) = i tan(u), we get the following
identities;
 
1. cos(v) = 1 − sin2 (v) = 1 + tan2 (u) = sec(u), and hence,
cos(u) = sec(v).
 
2. sin(u) = 1 − cos2 (u) = 1 − sec2 (v) = i tan(v).

3. Differentiating sin(v) = i tan(u) and using the first identity, we obtain

cos(v)dv = i sec2 (u)du ⇒ dv = i sec(u)du.

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 440 — #50

i i

440 Jacobi’s Elliptic Functions snk , cnk and dnk

 
4. 1 − k 2 sin2 (v) = 1 + k 2 tan2 (u)

= sec2 (u) − tan2 (u) + k 2 tan2 (u)

= sec2 (u) − (1 − k 2 ) tan2 (u)

1 − j2 sin2 (u)
= .
cos(u)
Therefore,
ψ
dv
Fk (ψ) = 
0 1 − k 2 sin2 (v)
φ
i sec(u)du
= 
1 + k 2 tan2 (u)
0
φ
i sec(u) cos(u)du
= 
0 1 − j2 sin2 (u)
= iFj (φ).
Next, from the third identity, replacing u and v, respectively, by φ and ψ ,
we get
dψ = i sec(φ)dφ ⇒ ψ = i log(sec(φ) + tan(φ)).
Thus, the result follows. 
Using the previous result, we can define the function amk , snk , cnk , and
dnk at ix, for x ∈ R, consistently with their definitions at x.
√
Definition 8.5.7 Let 0 ≤ k ≤ 1 and j = 1 − k 2 . For a given x ∈ R, if
φ = amj (x), then we define the following:
1. amk (ix) = ψ = i log(sec(φ) + tan(φ)).
snj (x)
2. snk (ix) = sin(ψ) = i tan(φ) = i = iscj (x).
cnj (x)
1 1
3. cnk (ix) = cos(ψ) = = = ncj (x).
cos(φ) cnj (x)

 1 − j2 sin2 (φ) dnj (x)
4. dnk (ix) = 1 − k 2 sin2 (ψ) = = .
cos(φ) cnj (x)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 441 — #51

i i

Elliptic Functions 441

THEOREM 8.5.8 For every x ∈ R, we have

snk (ix + i4Kj ) = snk (ix); cnk (ix + i4Kj ) = cnk (ix); dnk (ix + i4Kj ) = dnk (ix),
π
2 du √
where Kj =  and j = 1 − k2.
0 1 − j2 sin2 (u)

Proof: Let x ∈ R be arbitrary. Using Definition 8.5.7, we have

snj (x + 4Kj )
snk (i(x + 4Kj )) = i
cnj (x + 4Kj )
snj (x)
= i (as 4Kj is a period for snj and cnj ),
cnj (x)
= snk (ix),
1
cnk (i(x + 4Kj )) =
cnj (x + 4Kj )
1
= (as 4Kj is a period for cnj )
cnj (x)
= cnk (ix),
dnj (x + 4Kj )
dnk (i(x + 4Kj )) =
cnj (x + 4Kj )
dnj (x)
= (as 4Kj is a period for cnj and dnj )
cnj (x)
= dnk (ix).

Hence, the theorem follows. 

Next we shall prove the addition theorems for the Jacobi’s elliptic func-
tions using which these functions can be extended to the complex plane.
Although there are different proofs of the following theorem in the litera-
ture, the following proof is relatively easier and is not using the properties of
any other elliptic functions such as theta function.

THEOREM 8.5.9
cnk (x)cnk (y) − snk (x)snk (y)dnk (x)dnk (y)
cnk (x + y) =
1 − k 2 sn2k (x)sn2k (y)
snk (x)cnk (y)dnk (y) + snk (y)cnk (x)dnk (x)
snk (x + y) =
1 − k 2 sn2k (x)sn2k (y)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 442 — #52

i i

442 Jacobi’s Elliptic Functions snk , cnk and dnk

dnk (x)dnk (y) − k 2 snk (x)snk (y)cnk (x)cnk (y)

dnk (x + y) = .
1 − k 2 sn2k (x)sn2k (y)

Proof: Let Fk (φ) = x, Fk (ψ) = y, and Fk (μ) = x + y. Then, using the fact
that Fk is an odd function, we have

Fk (φ) + Fk (ψ) + Fk (−μ) = 0.

We claim that the expression


cos(φ) cos(ψ) − sin(φ) sin(ψ) 1 − k 2 sin2 (μ) = cos(μ) (8.22)

(where μ is a constant) satisfies the following differential equation

dφ dψ
 +  = 0. (8.23)
1 − k 2 sin2 (φ) 1 − k 2 sin2 (ψ)

Dividing on both sides of equation (8.22) by sin(φ) sin(ψ) and then differen-
tiating, we get
   
cos(φ) cos(ψ) cos(μ)
d =d .
sin(φ) sin(ψ) sin(φ) sin(ψ)
As
 
cos(φ) cos(ψ) 1
d = [sin(φ) sin(ψ)(− sin(φ)dφ cos(ψ)
sin(φ) sin(ψ) sin (φ) sin2 (ψ)
2

− cos(φ) sin(ψ)dψ) − cos(φ) cos(ψ)(cos(φ)dφ

sin(ψ) + sin(φ) cos(ψ)dψ)]
−1
= [sin(ψ) cos(ψ)dφ+ sin(φ) cos(φ)dψ]
sin (φ) sin2 (ψ)
2
 
cos(ψ)dφ cos φdψ
= − +
sin(ψ) sin2 (φ) sin(φ) sin2 (ψ)
and
 
cos(μ) − cos(μ)
d = [cos(φ)dφ sin(ψ)+ sin(φ) cos(ψ)dψ]
sin(φ) sin(ψ) sin (φ) sin2 (ψ)
2
 
cos(μ) cos(φ)dφ cos(μ) cos(ψ)dψ
= − +
sin2 (φ) sin(ψ) sin(φ) sin2 (ψ)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 443 — #53

i i

Elliptic Functions 443

we have
dφ dψ
2
[cos(ψ) − cos(μ) cos(φ)] +
sin (φ) sin(ψ) sin(φ) sin2 (ψ)
[cos(φ) − cos(μ) cos(ψ)] = 0
which implies that
   
cos(ψ) − cos(μ) cos(φ) cos(φ) − cos(μ) cos(ψ)
dφ + dψ = 0.
sin(φ) sin(ψ)
(8.24)
The expression in equation (8.22) can be rewritten as
(cos(φ) cos(ψ) − cos(μ))2 = sin2 (φ) sin2 (ψ)(1 − k 2 sin2 (μ)) (8.25)
⇒ cos2 (φ) cos2 (ψ) + cos2 (μ) − 2 cos(φ) cos(ψ) cos(μ)
= (1 − cos2 (φ))(1 − cos2 (ψ)) − k 2 sin2 (φ) sin2 (ψ) sin2 (μ)
= 1 − cos2 (φ) − cos2 (ψ) + cos2 (φ) cos2 (ψ)
− k 2 sin2 (φ) sin2 (ψ) sin2 (μ)
⇒ cos (φ) + cos2 (ψ) + cos2 (μ) − 2 cos(φ) cos(ψ) cos(μ)
2

+ k 2 sin2 (φ) sin2 (ψ) sin2 (μ) = 1.

The last equation is unaltered by permuting φ , ψ , and μ; equivalently, equa-
tion (8.22) is also unaltered by replacing μ by −μ and then by permuting φ ,
ψ , and −μ. Therefore, by interchanging ψ and −μ, and by interchanging φ
and −μ in equation (8.22), we get the following two more equations:

1. cos(φ) cos(μ) + sin(φ) sin(μ) 1 − k 2 sin2 (ψ) = cos(ψ)
cos(ψ) − cos(μ) cos(φ) sin(μ)
⇒ =  .
sin(φ)
1 − k 2 sin2 (ψ)
(8.26)

2. cos(μ) cos(ψ) + sin(μ) sin(ψ) 1 − k 2 sin2 (φ) = cos(φ)
cos(φ) − cos(μ) cos(ψ) sin(μ)
⇒ =  .
sin(ψ)
1 − k 2 sin2 (φ)
(8.27)
Using equations (8.26) and (8.27) in equation (8.24), we get the differential
equation (8.23). Thus, our claim holds.
Now substituting ζ = cos(μ) in equation (8.22), we get
(cos(φ) cos(ψ) − ζ )2 = sin2 (φ) sin2 (ψ)(1 − k 2 (1 − ζ 2 ))

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 444 — #54

i i

444 Jacobi’s Elliptic Functions snk , cnk and dnk

⇒ cos2 (φ) cos2 (ψ) + ζ 2 − 2 cos(φ) cos(ψ)ζ = (1 − k 2 ) sin2 (φ) sin2 (ψ)
+k 2 sin2 (φ) sin2 (ψ)ζ 2
⇒ (1 − k 2 sin2 (φ) sin2 (ψ))ζ 2 − 2 cos(φ) cos(ψ)ζ + cos2 (φ) cos2 (ψ)
+(k 2 − 1) sin2 (φ) sin2 (ψ) = 0.
Solving for ζ from √ the above quadratic equation, we get ζ =
2 cos(φ) cos(ψ) ± D
, where
2(1 − k 2 sin2 (φ) sin2 (ψ))
D = 4 cos2 (φ) cos2 (ψ) − 4(1 − k 2 sin2 (φ) sin2 (ψ))[cos2 (φ) cos2 (ψ)
+(k 2 − 1) sin2 (φ) sin2 (ψ)]
= 4[(1 − k 2 ) sin2 (φ) sin2 (ψ) + k 2 sin2 (φ) sin2 (ψ) cos2 (φ) cos2 (ψ)
+k 2 (k 2 − 1) sin4 (φ) sin4 (ψ)]
= 4[sin2 (φ) sin2 (ψ) − k 2 sin2 (φ) sin2 (ψ)
+k 2 sin2 (φ) sin2 (ψ)(1 − sin2 (φ) − sin2 (ψ) + sin2 (φ) sin2 (ψ))
+k 4 sin4 (φ) sin4 (ψ) − k 2 sin4 (φ) sin4 (ψ)]
= 4[sin2 (φ) sin2 (ψ) − k 2 sin2 (φ) sin4 (ψ) − k 2 sin4 (φ) sin2 (ψ)
+k 4 sin4 (φ) sin4 (ψ)]
= 4 sin2 (φ) sin2 (ψ)(1 − k 2 sin2 (ψ) − k 2 sin2 (φ) + k 4 sin2 (φ) sin2 (ψ))
= 4 sin2 (φ) sin2 (ψ)(1 − k 2 sin2 (φ))(1 − k 2 sin2 (ψ)).
Therefore,
cos(φ + ψ) = ζ

cos(φ) cos(ψ) ± sin(φ) sin(ψ) (1 − k 2 sin2 (ψ))

(1 − k 2 sin2 (φ))
=
1 − k 2 sin2 (φ) sin2 (ψ)
cnk (x)cnk (y) ± snk (x)snk (y)dnk (x)dnk (y)
⇒ cnk (x + y) = .
1 − k 2 sn2k (x)sn2k (y)
By observing
cn0 (x + y) = cos(x + y) = cos(x) cos(y) − sin(x) sin(y)
= cn0 (x)cn0 (y) − sn0 (x)sn0 (y)
we get
cnk (x)cnk (y) − snk (x)snk (y)dnk (x)dnk (y)
cnk (x + y) = .
1 − k 2 sn2k (x)sn2k (y)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 445 — #55

i i

Elliptic Functions 445

Using the identity sn2k + cn2k = 1 in the following straightforward but lengthy
manipulation, we get
sn2k (x + y)
= 1 − cn2k (x + y)2
! "2
cnk (x)cnk (y) − snk (x)snk (y)dnk (x)dnk (y)
= 1−
1 − k 2 sn2k (x)sn2k (y)
[1 − k snk (x)snk (y)]2 − [cnk (x)cnk (y) − snk (x)snk (y)dnk (x)dnk (y)]2
2 2 2
=
& Dr'2
( where Dr = 1 − k 2 sn2k (x)sn2k (y) )
= Dr 1
[1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y) − cn2k (x)cn2k (y)
−sn2k (x)sn2k (y)dn2k (x)dn2k (y) + 2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
[1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y)
−(1 − sn2k (x) − sn2k (y) + sn2k (x)sn2k (y))
−sn2k (x)sn2k (y)(1 − k 2 sn2k (x) − k 2 sn2k (y) + k 4 sn2k (x)sn2k (y))
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
[sn2k (x) + sn2k (y) + k 2 sn2k (x)sn4k (y) + k 2 sn4k (x)sn2k (y)
−2sn2k (x)sn2k (y) − 2k 2 sn2k (x)sn2k (y)
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
[(sn2k (x) − sn2k (x)sn2k (y))(1 − k 2 sn2k (y))
+(sn2k (y) − sn2k (x)sn2k (y))(1 − k 2 sn2k (x))
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
[sn2k (x)(1 − sn2k (y))(1 − k 2 sn2k (y))
+sn2k (y)(1 − sn2k (x))(1 − k 2 sn2k (x))
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
[sn2k (x)cn2k (y)dn2k (y) + sn2k (y)cn2k (x)dn2k (x)
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= Dr 1
(snk (x)cnk (y)dnk (y) + snk (y)cnk (x)dnk (x))2 .
Hence,

snk (x)cnk (y)dnk (y) + snk (y)cnk (x)dnk (x)

snk (x + y) = ± . (8.28)
1 − k 2 sn2k (x)sn2k (y)

Comparing this equation with (particular case k = 0),

sn0 (x + y) = sn0 (x)cn0 (y) + sn0 (y)cn0 (x),

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 446 — #56

i i

446 Jacobi’s Elliptic Functions snk , cnk and dnk

we choose the sign in equation (8.28) and we get

snk (x)cnk (y)dnk (y) + snk (y)cnk (x)dnk (x)
snk (x + y) = .
1 − k 2 sn2k (x)sn2k (y)

Next,
dn2k (x + y)
= 1 − k 2 sn2k (x + y)
(snk (x)cnk (y)dnk (y) + snk (y)cnk (x)dnk (x))2
= 1 − k2
(1 − k 2 sn2k (x)sn2k (y))2
Nr
= ,
(1 − k sn2k (x)sn2k (y))2
2
where
Nr = 1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y)
−k 2 [sn2k (x)cn2k (y)dn2k (y) + sn2k (y)cn2k (x)dn2k (x)
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= 1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y)
−k 2 [sn2k (x)(1 − sn2k (y))(1 − k 2 sn2k (y)) + sn2k (y)
(1 − sn2k (x))(1 − k 2 sn2k (x))
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)]
= 1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y)
−k 2 [(sn2k (x) − sn2k (x)sn2k (y))(1 − k 2 sn2k (y))
+(sn2k (y) − sn2k (x)sn2k (y))(1 − k 2 sn2k (x))
+2snk (x)snk (y)cnk (x)cnk (y)]dnk (x)dnk (y)]
= 1 + k 4 sn4k (x)sn4k (y) − 2k 2 sn2k (x)sn2k (y)
−k 2 [sn2k (x) + sn2k (y) − 2sn2k (x)sn2k (y) − 2k 2 sn2k (x)sn2k (y)
+2snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)
= (1 + k 4 sn2k (x)sn2k (y) − k 2 sn2k (x) − k 2 sn2k (y))
+k 4 sn2k (x)sn2k (y) − k 4 sn4k (x)sn2k (y) − k 4 sn2k (x)sn4k (y)
+k 4 sn4k (x)sn4k (y) − 2k 2 snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)
= (1 − k 2 sn2k (x))(1 − k 2 sn2k (y)) + k 4 sn2k (x)sn2k (y)
+k 4 sn2k (x)sn2k (y)[1 − sn2k (x) − sn2k (y) + sn2k (x)sn2k (y)]
−2k 2 snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)
= dn2k (x)dn2k (y) + k 4 sn2k (x)sn2k (y)(1 − sn2k (x))(1 − sn2k (y))
−2k 2 snk (x)snk (y)cnk (x)cnk (y)dnk (x)dnk (y)
= (dnk (x)dnk (y) − k 2 snk (x)snk (y)cnk (x)cnk (y))2 .
Therefore,

dnk (x)dnk (y) − k 2 snk (x)snk (y)cnk (x)cnk (y)

dnk (x + y) = .
1 − k 2 sn2k (x)sn2k (y)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 447 — #57

i i

Elliptic Functions 447

In the last expression, we have chosen the sign by considering the particular
case k = 0 as before. 
sck (x)dnk (x) + sck (y)dnk (y)
Exercise 8.5.10 Prove that sck (x + y) = .
1 − sck (x)sck (y)dnk (x)dnk (y)
snk (x + y)
Hint: Using the previous theorem, find and simplify.
cnk (x + y)

The following identities are the immediate consequences of Theorem

8.5.9 and Result 8.5.2:
cnk (x)cnk (y) ∓ snk (x)snk (y)dnk (x)dnk (y)
1. cnk (x ± y) =
1 − k 2 sn2k (x)sn2k (y)
snk (x)cnk (y)dnk (y) ± snk (y)cnk (y)dnk (x)
2. snk (x ± y) =
1 − k 2 sn2k (x)sn2k (y)
dnk (x)dnk (y) ∓ k 2 snk (x)snk (y)cnk (x)cnk (y)
3. dnk (x ± y) =
1 − k 2 sn2k (x)sn2k (y)

cn2k (x) − sn2k (x)dn2k (x)

4. cnk (2x) =
1 − k 2 sn4k (x)
1 − sn2k (x) − sn2k (x)(1 − k 2 sn2k (x))
=
1 − k 2 sn4k (x)
1 − sn2k (x) − sn2k (x) + k 2 sn4k (x)
=
1 − k 2 sn4k (x)
1 − 2sn2k (x) + k 2 sn4k (x)
=
1 − k 2 sn4k (x)
2snk (x)cnk (x)dnk (x)
5. snk (2x) =
1 − k 2 sn4k (x)
dn2k (x) − k 2 sn2k (x)cn2k (x)
6. dnk (2x) =
1 − k 2 sn4k (x)
1 − k 2 sn2k (x) − k 2 sn2k (x)(1 − sn2k (x))
=
1 − k 2 sn4k (x)
1 − 2k 2 sn2k (x) − k 2 sn4k (x)
= .
1 − k 2 sn4k (x)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 448 — #58

i i

448 Jacobi’s Elliptic Functions snk , cnk and dnk

x 
cnk (x)+dnk (x)
PROBLEM 8.5.11 Prove that cnk 2 = 1+dnk (x) .

Solution: As
x x        
cn2k −sn2k 2 dnk 2
2 x dn2k 2x −k 2 sn2k 2x cn2k 2x
2
  +  
cnk (x) + dnk (x) 1 − k 2 sn4k 2x 1 − k 2 sn4k 2x
=      
1 + dnk (x) dn2k 2x − k 2 sn2k 2x cn2k 2x
1+  
1 − k 2 sn4k 2x
           
cn2k 2x −sn2k 2x dn2k 2x +dn2k 2x − k 2 sn2k 2x cn2k 2x
=        
1 − k 2 sn4k 2x + dn2k 2x − k 2 sn2k 2x cn2k 2x
       
cn2k 2x 1 − k 2 sn2k 2x + dn2k 2x 1 − sn2k 2x
=        
1 − k 2 sn2k 2x sn2k 2x + cn2k 2x + dn2k 2x
   
2cn2k 2x dn2k 2x
=    
1 − k 2 sn2k 2x + dn2k 2x
   
2cn2k 2x dn2k 2x
=  
2dn2k 2x
x
= cn2k .
2
x 
(x)+dnk (x)
Thus, we get cnk 2 = cnk1+dn k (x)
.
By adopting the same technique, the reader can prove the following
identities.

Exercise 8.5.12 Prove the following:

,
x 1 − cnk (x)
1. snk =
2 1 + dnk (x)
,
x k 2 + dnk (x) + k 2 cnk (x)
2. dnk = .
2 1 + dnk (x)

By seeing the Theorem 8.5.9, we extend the Jacobi’s functions snk , sck ,
and dnk as follows.

Definition 8.5.13 Define the following:

cnk (x)cnk (iy) − snk (x)snk (iy)dnk (x)dnk (iy)
1. cnk (x + iy) =
1 − k 2 sn2k (x)sn2k (iy)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 449 — #59

i i

Elliptic Functions 449

1 snj (y) dnj (y)

cnk (x) − snk (x)i dnk (x)
cnj (y) cnj (y) cnj (y)
=
sn2j (y)
1 + k 2 sn2k (x)
cn2j (y)
cnk (x)cnj (y) − isnk (x)snj (y)dnk (x)dnj (y)
= .
cn2j (y) + k 2 sn2k (x)sn2j (y)
snk (x)cnk (iy)dnk (iy) + snk (iy)cnk (x)dnk (x)
2. snk (x + iy) =
1 − k 2 sn2k (x)sn2k (iy)
1 dnj (y) snj (y)
snk (x) +i cnk (x)dnk (x)
cnj (y) cnj (y) cnj (y)
=
sn2j (y)
1 + k 2 sn2k (x)
cn2j (y)
snk (x)dnj (y) + isnj (y)cnj (y)cnk (x)dnk (x)
= .
cn2j (y) + k 2 sn2k (x)sn2j (y)
dnk (x)dnk (iy) − k 2 snk (x)snk (iy)cnk (x)cnk (iy)
3. dnk (x + iy) =
1 − k 2 sn2k (x)sn2k (iy)
dnj (y) snj (y) 1
dnk (x) − ik 2 snk (x) cnk (x)
cnj (y) cnj (y) cnj (y)
=
sn2j (y)
1 + k 2 sn2k (x)
cn2j (y)
cnj (y)dnk (x)dnj (y) − ik 2 snk (x)snj (y)cnk (x)
= .
cn2j (y) + k 2 sn2k (x)sn2j (y)

PROBLEM 8.5.14 Prove that if 0 < k < 1, then cnk , snk , and dnk are elliptic
functions.

Solution: First, we note that the values of the three functions, namely, cnk ,
snk , and dnk at x+iy have the same denominator cn2j (y)+k 2 sn2k (x)sn2j (y), and
it is zero iff cnj (y) = 0 and ksnk (x)snj (y) = 0 iff cnj (y) = 0 and snk (x) = 0,
as sn2j (y) = 1.
From Theorems 8.5.2 and 8.5.3, we have

snk (2mKk ) = 0, ∀m ∈ Z and cnj ((2n + 1)Kj ) = 0, ∀n ∈ Z

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 450 — #60

i i

450 Jacobi’s Elliptic Functions snk , cnk and dnk

Furthermore, Fk (0) = 0 and Fk (φ + π ) = Fk (φ) + 2Kk (See equation (8.20))

imply that Fk (π) = 2Kk . Therefore, Fk maps (0, π ) onto (0, 2Kk ). As sin  0
on (0, π ), snk  0 on (0, 2Kk ), and hence,

snk (x) = 0 for some x ∈ R iff x ∈ {2mKk : m ∈ Z}.

Similarly, using the facts (i) cos  0 on (0, π2 ) and (ii) Fk maps (0, π2 ) onto
(0, Kk ), we obtain cnk  0 on (0, Kk ). Therefore,

cnk (x) = 0 for some x ∈ R iff x ∈ {(2n + 1)Kj : n ∈ Z}.

Thus, if  = C \ {(2mKk , (2n + 1)Kj ) : m, n ∈ Z}, then cnj (y)  0 and

snk (x)  0, and hence, these three functions are finite valued on .
Using Theorem 8.5.3, for x ∈ R, we have  
snj (x + 2Kj ) −snj (x)
snk (i(x + 2Kj )) = i = i = snk (ix)
cnj (x + 2Kj ) −cnj (x)
1 1
cnk (i(x + 2Kj )) = = = −cnk (ix)
cnj (x + 2Kj ) −cnj (x) 

dnj (x + 2Kj ) dnj (x)
dnk (i(x + 2Kj )) = = = −dnk (ix).
cnj (x + 2Kj ) −cnj (x)
Thus, replacing x ∈ R by z ∈ , we get

snk (z + i2Kj ) = snk (z), (8.29)

cnk (z + i2Kj ) = −cnk (z), (8.30)
dnk (z + i2Kj ) = −dnk (z). (8.31)

Combining the above equation with Corollary 8.5.4, we get

cnk (z + 2Kk + i2Kj ) = −cnk (z + 2Kk ) = cnk (z),

dnk (z + i4Kj ) = −dnk (z + i2Kj ) = dnk (z).

Thus, i2Kj and 4Kk are periods of snk ; 2Kk + i2Kj and 4Kk are periods of
cnk ; and i4Kj and 2Kk are periods of dnk .
Next, we show that snk is analytic on . We denote the real and imaginary
parts of snk (x + iy) (see Definition 8.5.13.), respectively, by

snk (x)dnj (y) snj (y)cnj (y)cnk (x)dnk (x)

U(x, y) = ; V (x, y) = .
cnj (y) + k 2 sn2k (x)sn2j (y)
2 cn2j (y) + k 2 sn2k (x)sn2j (y)

If Dr = (cn2j (y) + k 2 sn2k (x)sn2j (y))2 , then

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 451 — #61

i i

Elliptic Functions 451

∂U
(x, y)
∂x
1 $ 2
= (cnj (y) + k 2 sn2k (x)sn2j (y))(cnk (x)dnk (x)dnj (y))
Dr %
−snk (x)dnj (y)(2k 2 snk (x)cnk (x)dnk (x)sn2j (y))
1 $
= cnk (x)dnk (x)cn2j (y)dnj (y) + k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)dnj (y)
Dr %
−2k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)dnj (y)
1 $ %
= cnk (x)dnk (x)cn2j (y)dnj (y) − k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)dnj (y)
Dr
and
∂V
(x, y)
∂y
1 $ 2
= (cnj (y) + k 2 sn2k (x)sn2j (y))(cnk (x)dnk (x)(cn2j (y)dnj (y)
Dr
−sn2j (y)dnj (y))) − snj (y)cnj (y)cnk (x)dnk (x)(−2cnj (y)snj (y)dnj (y)
'
+2k 2 sn2k (x)snj (y)cnj (y)dnj (y)
1 $
= cnk (x)dnk (x)cn4j (y)dnj (y) − cnk (x)dnk (x)sn2j (y)cn2j (y)dnj (y)
Dr
+k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)cn2j (y)dnj (y) − k 2 sn2k (x)cnk (x)dnk (x)
sn4j (y)dnj (y) + 2cnk (x)dnk (x)sn2j (y)cn2j (y)dnj (y)
%
−2k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)cn2j (y)dnj (y)
1 $
= cnk (x)dnk (x)cn2j (y)dnj (y)(cn2j (y) + sn2j (y))
Dr %
−k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)dnj (y)(sn2j (y) + cn2j (y))
1 $ %
= cnk (x)dnk (x)cn2j (y)dnj (y) − k 2 sn2k (x)cnk (x)dnk (x)sn2j (y)dnj (y)
Dr
∂U
= (x, y).
∂x
Next, we verify the other C–R equation. Although it not as simple as the
previous one, we verify it as follows:
∂U
Dr (x, y)
∂y
= (cn2j (y) + k 2 sn2k (x)sn2j (y))(−j2 snk (x)snj (y)cnj (y))
−(snk (x)dnj (y))(−2cnj (y)snj (y)dnj (y) + 2k 2 sn2k (x)snj (y)cnj (y)dnj (y))
= −j2 snk (x)snj (y)cn3j (y) − j2 k 2 sn3k (x)sn3j (y)cnj (y)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 452 — #62

i i

452 Jacobi’s Elliptic Functions snk , cnk and dnk

+2snk (x)snj (y)cnj (y)dn2j (y) − 2k 2 sn3k (x)snj (y)cnj (y)dn2j (y)
= (k 2 − 1)snk (x)snj (y)cn3j (y) − (1 − k 2 )k 2 sn3k (x)sn3j (y)cnj (y)
+(2snk (x)snj (y)cnj (y) − 2k 2 sn3k (x)snj (y)cnj (y))(1 − (1 − k 2 )sn2j (y))
= −snk (x)snj (y)cn3j (y) + k 2 snk (x)snj (y)cn3j (y) − k 2 sn3k (x)sn3j (y)cnj (y)
+k 4 sn3k (x)sn3j (y)cnj (y) + 2snk (x)snj (y)cnj (y) − 2snk (x)sn3j (y)cnj (y)
+2k 2 snk (x)sn3j (y)cnj (y) − 2k 2 sn3k (x)snj (y)cnj (y)
+2k 2 sn3k (x)sn3j (y)cnj (y) − 2k 4 sn3k (x)sn3j (y)cnj (y)
= −snk (x)snj (y)cnj (y)[cn2j (y) + sn2j (y)] − snk (x)sn3j (y)cnj (y)
+k 2 snk (x)snj (y)cn3j (y) + k 2 sn3k (x)sn3j (y)cnj (y)−k 4 sn3k (x)sn3j (y)cnj (y)
+2snk (x)snj (y)cnj (y) + 2k 2 snk (x)sn3j (y)cnj (y)−2k 2 sn3k (x)snj (y)cnj (y)
= −snk (x)snj (y)cnj (y) − snk (x)sn3j (y)cnj (y) + k 2 snk (x)snj (y)cnj (y)
[cn2j (y) + sn2j (y)] + k 2 snk (x)sn3j (y)cnj (y) + k 2 sn3k (x)sn3j (y)cnj (y)
+2snk (x)snj (y)cnj (y) − 2k 2 sn3k (x)snj (y)cnj (y) − k 4 sn3k (x)sn3j (y)cnj (y)
= snk (x)snj (y)cnj (y) − snk (x)sn3j (y)cnj (y) + k 2 snk (x)snj (y)cnj (y)
+k 2 sn3k (x)sn3j (y)cnj (y) − 2k 2 sn3k (x)snj (y)cnj (y)
+k 2 snk (x)sn3j (y)cnj (y) − k 4 sn3k (x)sn3j (y)cnj (y)
= snk (x)snj (y)cnj (y)(1 − sn2j (y)) − k 2 sn3k (x)snj (y)cnj (y)(1 − sn2j (y))
+k 2 snk (x)snj (y)cnj (y) + k 2 snk (x)sn3j (y)cnj (y) − k 4 sn3k (x)sn3j (y)cnj (y)
−k 2 sn3k (x)snj (y)cnj (y)
= snk (x)snj (y)cn3j (y) − k 2 sn3k (x)snj (y)cn3j (y) + k 2 snk (x)snj (y)cnj (y)
+k 2 snk (x)sn3j (y)cnj (y) − k 4 sn3k (x)sn3j (y)cnj (y) − k 2 sn3k (x)snj (y)cnj (y)
and
∂V
Dr (x, y)
∂x
= (cn2j (y) + k 2 sn2k (x)sn2j (y))snj (y)cnj (y)[−k 2 snk (x)cn2k (x) − snk (x)dn2k (x)]
−(snj (y)cnj (y)cnk (x)dnk (x))(2k 2 sn2j (y)snk (x)cnk (x)dnk (x))
= −k 2 snk (x)cn2k (x)snj (y)cn3j (y) − snk (x)dn2k (x)snj (y)cn3j (y)
−k 4 sn3k (x)cn2k (x)sn3j (y)cnj (y) − k 2 sn3k (x)dn2k (x)sn3j (y)cnj (y)
−2k 2 snk (x)cn2k (x)dn2k (x)sn3j (y)cnj (y)
$
= −k 2 snk (x)snj (y)cn3j (y) − k 4 sn3k (x)sn3j (y)cnj (y)
%
−2k 2 snk (x)(1 − k 2 sn2k (x))sn3j (y)cnj (y) (1 − sn2k (x))
$ %
+ −snk (x)snj (y)cn3j (y) − k 2 sn3k (x)sn3j (y)cnj (y) (1 − k 2 sn2k (x))
= −k 2 snk (x)snj (y)cn3j (y) − k 4 sn3k (x)sn3j (y)cnj (y) − 2k 2 snk (x)sn3j (y)cnj (y)

i i

i i
 i i

“book” — 2014/6/4 — 21:28 — page 453 — #63

i i

Elliptic Functions 453

+2k 4 sn3k (x)sn3j (y)cnj (y) + k 2 sn3k (x)snj (y)cn3j (y) + k 4 sn5k (x)sn3j (y)cnj (y)
+2k 2 sn3k (x)sn3j (y)cnj (y) − 2k 4 sn5k (x)sn3j (y)cnj (y) − snk (x)snj (y)cn3j (y)
−k 2 sn3k (x)sn3j (y)cnj (y) + k 2 sn3k (x)snj (y)cn3j (y) + k 4 sn5k (x)sn3j (y)cnj (y)
= −k 2 snk (x)snj (y)cn3j (y) − 2k 2 snk (x)sn3j (y)cnj (y) + k 4 sn3k (x)sn3j (y)cnj (y)
+k 2 sn3k (x)sn3j (y)cnj (y) + 2k 2 sn3k (x)snj (y)cn3j (y) − snk (x)snj (y)cn3j (y)
= −k 2 snk (x)snj (y)cnj (y)[cn2j (y) + sn2j (y)] − k 2 snk (x)sn3j (y)cnj (y)
+k 4 sn3k (x)sn3j (y)cnj (y) + k 2 sn3k (x)snj (y)cnj (y)[sn2j (y) + cn2j (y)]
+k 2 sn3k (x)snj (y)cn3j (y) − snk (x)snj (y)cn3j (y)
= −k 2 snk (x)snj (y)cnj (y) − k 2 snk (x)sn3j (y)cnj (y) + k 4 sn3k (x)sn3j (y)cnj (y)
+k 2 sn3k (x)snj (y)cnj (y) + k 2 sn3k (x)snj (y)cn3j (y) − snk (x)snj (y)cn3j (y).
Therefore, the real and imaginary parts of snk satisfy the C–R equations
and obviously the partial derivatives are continuous on , and hence, snk is
analytic on  and meromorphic on C, as snk has poles on C \ . Thus, snk
is an elliptic function. Using Theorem 6.3.2, we have cnk and dnk , which are
meromorphic functions on , and hence, they are also elliptic functions.

Exercise 8.5.15 Prove the following:

1. The fundamental periods of snk are i2Kj and 4Kk .
2. The fundamental periods of cnk are 2Kk + i2Kj and 4Kk .
3. The fundamental periods of dnk are i4Kj and 2Kk .

Remark 8.5.16: All the identities involving the Jacobian elliptic functions
holds for all z ∈  also, using the principle of analytic continuation.

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 17:29 — page 455 — #1

i i

Bibliography

1. L.V. Ahlfors, Complex Analysis, Third Edition, McGraw-Hill, Inc.,

New York (1979).
2. J.V. Armitage and W.F. Eberlein, Elliptic Functions, Cambridge Uni-
versity Press, Cambridge (2006).
3. J.B. Bak and D.J. Newman, Complex Analysis, Third Edition, Springer,
New York (2010).
4. A. Cayley, An Elementary Treatise on Elliptic Functions, Second
Edition, Cambridge University Press, Cambridge (1895).
5. R.V. Churchill, J.W. Brown and R.F. Verhey, Complex Variables and
Applications, Third Edition, McGraw-Hill, Inc., New York (1974).
6. J. Conway, Functions of One Complex Variable, Second Edition,
Springer-Verlag (1978).
7. H. Hancock, Elliptic Integrals, John Wiley & Sons, New York (1917).
8. V. Karunakaran, Complex Analysis, Second Edition, Alpha Science
International Ltd, New Delhi (2005).
9. S. Lang, Complex Analysis, Fourth Edition, Springer, New York
(1999).
10. R. Remmert, Theory of Complex Functions, Springer-Verlag, New
York (1991).
11. W. Rudin, Real and Complex Analysis, McGraw-Hill, New York
(1969).
12. E.T. Whittaker and G.N. Watson, A Course of Modern Analysis, Fourth
Edition, Cambridge University Press, London (1963).

455

i i

i i
This page is intentionally left blank
 i i

“book” — 2014/6/4 — 17:30 — page 457 — #3

i i

Index

A C
Abel’s limit theorem, 100 canonical product, 270
Abel’s theorem on convergence of a Cantor’s intersection theorem, 49
power series, 94 Cartesian form, 18
Absolutely, 33 Cartesian product, 4
abelian group, 8 Cauchy criterion, 28
addition, 9 Cauchy product, 228
addition theorem for exp, 104 Cauchy sequence, 25
addition theorems for the Jacobi’s elliptic Cauchy’s estimate, 206
functions, 441 Cauchy’s integral formula, 193
adjacent side, 114 Cauchy’s integral formula for
Algorithm, 82, 144 derivatives, 196
analytic, 74 Cauchy’s residue theorem, 294
Analytic branch, 150 Cauchy’s theorem for a simply connected
√ region, 191
analytic branch for ·, 153
arc length, 176 Cauchy’s theorem for rectangle, 187
Archimedian property, 6 Cauchy-Goursat theorem, 193
Arccos, 154 Cauchy-Riemann equations, 79
arcsin(w), 154 Cauchy-Riemann equations in polar
arctan, 117 form, 90
Cauchy-Schwarz inequality, 16
B Chain rule, 73
Circle, 24
basic Jacobian elliptic functions, 435 closed curve, 172
bijective, 7 Closed set, 39
Binary operator, 7 closure, 38
Bolzano-Weierstrass property, 51 codomain, 6
Bounded, 25 Complete metric space, 51
bounded above, 5, 6 compact set, 48
bounded below, 5, 6 comparison test, 35
bounded sequence, 25, 50, 368 complex conjugate, 12
branch, 150 complex logarithm, 150

457

i i

i i
 i i

“book” — 2014/6/4 — 17:30 — page 458 — #4

i i

458 Index

Complex Numbers, 9 E
complex version of mean-value
elementary linear fractional
theorem, 77
transforms, 132
component, 46
elliptic curve, 392
conformal, 157 elliptic function, 395
Conformal mapping, 156 elliptic integral, 392
conformal of first type, 156 Elliptic integrals of first kind, 392
conformal of second type, 157 elliptic integrals of second kind, 392
Continuity, 55 Elliptic integrals of third kind, 392
Continuous, 62 Equation of the straight line, 23
continuous at z0 , 62 Equation of a circle, 24
continuous on , 62 entire functions, 207
convergent, 24 essential singularity, 249
convergent sequence, 25 Euler’s formula, 106
Connected, 41 Euler’s normal form, 434
cos(z), 105 evaluation of real integrals, 314
cot(z), 115 exponential function, 102
coth(z), 121 Extended complex numbers, 52
cosh(z), 119
F
cross ratio, 139
csc(z), 115 field, 8
csch(z), 121 First addition theorem for ℘a,b , 419
curve, 156 fixed points, 135
cycle, 212 function, 6
Functions za,b and sa,b , 426
D fundamental parallelogram, 398
fundamental period, 393
de Moivre’s formula, 106, 107 fundamental periods of cnk , 453
De-Morgan’s laws, 4 fundamental periods of dnk , 453
decreasing function, 7 fundamental periods of snk , 453
Define π , 109 Fundamental theorem of algebra, 207
degree, 123 Fundamental theorem of calculus, 170
Dense subset, 41
derivative, 73 G
Degree of the polynomial, 123 general version of Cauchy’s
differentiable, 73 theorem, 209
disconnected, 42 generalized Schwarz lemma, 217
Division algorithm for polynomials, 123 geometric series, 36
domain, 6 greatest lower bound property, 6

i i

i i
 i i

“book” — 2014/6/4 — 17:30 — page 459 — #5

i i

Index 459

H Laurent series, 232, 289

Laurent’s Theorem, 230
Half plane, 126
least upper bound, 5
harmonic conjugate, 81
least upper bound property, 6
harmonic function, 81
Left limit, 61
harmonic functions, 341
Legendre’s duplication formula, 279
Heine-Borel theorem, 49
Leibniz rule, 74
higher order derivatives, 74
Limit, 55
homologous, 187
Limit point, 37
Hurwitz theorem, 220
limit inferior, 29
hyperbolic functions, 119
limit superior, 29
hypotenuse, 114
line segment, 42
linear fractional transform, 132
I
linearly independent, 396
image, 7 Liouville’s theorem, 207
imaginary number, 9 Local correspondence theorem, 213
Imaginary part is v, 84 lower bound, 5
increasing function, 7 lower limit, 29
infimum, 5 Lucas’ theorem, 126
Infinite Taylor’s series, 227
injective, 7 M
interior, 41 Möbius transform, 132
Interior point, 37 Maximum, 6
Intermediate value theorem, 68 maximum modulus principle, 215
interval, 41 maximum principle for harmonic
Invertible, 7 function, 345
inverse of Fk , 435 mean-value property of a harmonic
isolated singularity, 249 function, 343
mean-value theorem, 77
J meromorphic function, 250
Jacobi’s functions snk , sck , and dnk , 448 Milne-Thompson, 82
Jacobi’s normal form, 434 Minimum, 6
Minimum modulus principle, 215
K Minimum principle for harmonic
function, 347
Koebe’s lemma, 370
Mittag-Lefler theorem, 257
Modulus function, 8
L
modulus, 14
Laplace equation, 81 Montel’s theorem, 368
Laplace equation in polar form, 341 Morera’s theorem, 206

i i

i i
 i i

“book” — 2014/6/4 — 17:30 — page 460 — #6

i i

460 Index

multiplication, 9 principal branch for log, 152

multiplicative inverse, 22 principle of symmetry, 143
multiply connected region, 187 problem of simple pendulum, 432
multiplyable, 264
R
N
radius of convergence, 94
natural logarithm, 148 rational function, 127
Neighbourhood, 37 Rearrangement of a series, 34
rectifiable curve, 176
O region, 44
one-to-one, 7 relation, 5
onto, 7 removable singularity, 223, 249
open ball, 37 residue, 287
open mapping theorem, 214 Riemann surface, 161
open set, 37 Riemann integrable, 166, 168
opposite curve, 175 Riemann mapping theorem, 371
opposite side, 114 Right limit, 61
order of the elliptic function, 401 Rouche’s theorem, 312
order of a zero, 124
order relation, 5 S
ordered field, 8 Schwarz lemma, 216
ordered set, 5 Schwarz’s inequality, 17
Second addition theorem for ℘a,b , 420
P separated, 41
partial fraction, 130, 257 sequence, 24
partial fraction expansion, 130 series, 24, 33
partially differentiable, 76 set, 3
period, 393 simple closed curve, 172
period module, 396 simple curve, 172
piece-wise smooth curve, 171 simply connected region, 185
Poisson’s formula, 347 sin(z), 86
polar form, 18 sinh(z), 119
pole, 129 singularity, 249
Pole of order k , 129 square root, 309
polygon, 43 standard metric, 37
polynomial, 123 stereographic projection, 52
power series, 94 strictly decreasing function, 7
pre-image, 7 strictly increasing function, 7
Principal argument, 19 Subsequence, 26

i i

i i
 i i

“book” — 2014/6/4 — 17:30 — page 461 — #7

i i

Index 461

summable, 33 uniformly bounded, 368

supremum, 5 upper bound, 5
surjective, 7 upper limit, 29
symmetry, 142
W
T
Weierstrass function, 406
tan(z), 118
Weierstrass theorem for essential
tanh(z), 121
singularity, 256
Taylors’s theorem, 224
Weierstrass theorem on convergence of
The function σa,b , 429
analytic functions, 197
The function Fk , 435
winding number, 182, 185, 213
To find the LFT, 144
To find harmonic conjugate, 82
Topology on C∞ , 52 Y
Triangle inequality, 15
Young’s theorem, 78
trigonometric functions, 102

U Z
Unary operator, 7 zero, 212, 246, 328
unbounded region determined by γ , 183 Zero of order k , 129, 397

i i

i i
This page is intentionally left blank