Complex Analysis

Lecture Notes for MATH 4023 (Spring 2020)

F REDERICK T SZ -H O F ONG

Hong Kong University of Science and Technology

(Version: February 29, 2020)
Contents

Preface ix
Chapter 1. Preliminaries 1
1.1. Complex Numbers 1
1.2. Sequences and Series 12
1.3. Point-Set Topology of C 25

Chapter 2. Holomorphic Functions 33

2.1. Complex-Valued Functions 33
2.2. Complex Differentiability 39
2.3. Logarithmic and Trigonometric Functions 50
Chapter 3. Contour Integrals 57
3.1. Complex Integrations 57
3.2. Cauchy-Goursat’s Theorem 65
3.3. Cauchy’s Integral Formula I 72
3.4. Cauchy’s Integral Formula II 81
3.5. Morera’s Theorem 89
Chapter 4. Taylor and Laurent Series 95
4.1. Taylor Series 95
4.2. Laurent Series 103
4.3. Residue Calculus 110
Chapter 5. What is the Riemann Hypothesis? 131
5.1. Analytic Continuation 131
5.2. Riemann ζ Functions 135
Appendix A. Results from MATH 2043/3033 143

vii
Preface

「虛則實之，實則虛之。」

《孫子兵法》

This is the lecture note written for the course MATH 4023 - Complex Analysis, taught
by the author at the Hong Kong University of Science and Technology (HKUST) in Spring
2017 and Spring 2020.
The purpose of this lecture note and the course is to introduce both theory and
applications of complex-valued functions of one variable. It begins with basic notions of
complex differentiability (i.e. holomorphic) functions. The central part of the course is
the Cauchy’s integral formula, which is a fundamental theorem leading many important
and exciting results in the later half of the course. The last chapter of the course explains
the statement of the Riemann Hypothesis, a famous problem that worths US$1,000,000
in Pure Mathematics.
The prerequisites of the course include Multivariable Calculus in which students
should be familiar with line integrals and basic analysis of multivariable functions, and
Analysis I (as basic ε − δ languages are needed). It is also recommended that students
have taken Analysis II (MATH 3033) or Honors Analysis I (MATH 2043) before taking
this course, as some standard toolkits such as uniform convergence, Weierstass’s M-
test, Lebesgue Dominated Convergence Theorem, Fubini’s Theorem, etc. will appear
very frequently in the later half of the course. Students without MATH 2043/3033 are
recommended to first go through some examples in these topics. A list of theorems in
MATH 2043/3033 which are essential for this course can be found in the appendix of
this lecture note.
The author welcomes any students and/or readers to point out typographical errors
and mistakes of this lecture note.
Frederick Tsz-Ho Fong
15 January, 2020
HKUST, Clear Water Bay, Hong Kong

ix
 Chapter 1

Preliminaries

1.1. Complex Numbers

1.1.1. Basic Arithmetics. From middle/high school, we learned that the quadratic
equation x2 + 1 = 0 does not have any real root because x2 + 1 > 0 for any x ∈ R.
Complex numbers are introduced to make it possible for the equation x2 + 1 = 0 to have
roots. We denote:

√
i= −1 so that i2 = −1.

While complex numbers make their appearance for purely algebraic purposes, their uses
branch out to many scientific fields beyond Mathematics, including Quantum Mechanics,
String Theory, Electrical Engineering, Fluid Mechanics, etc.

Definition 1.1 (Complex Numbers). A complex number z is a number of the form:

z = a + bi
√
where a and b are real numbers, and i = −1. We call:
• a is the real part of z and is denoted by a =: Re(z); and
• b is the imaginary part of z and is denoted by b =: Im(z).
The set of all complex numbers is denoted by C. Precisely, we have:
C := {a + bi : a, b ∈ R}.

Remark 1.2. Note that a real number is also considered as a complex number, since
a = a + 0i. In other words, we have R ⊂ C.

A complex number z = x + yi can be geometrically represented by the point (x, y) in

R2 (see Figure 1.1). The x-axis is now called the real axis as it represents numbers of the
form a + 0i. Likewise, the y-axis is called the imaginary axis, which represents numbers
of the form 0 + bi.

1
2 1. Preliminaries

Im
z = x + yi
y

x Re

−y
z = x − yi

Figure 1.1. geometry of complex numbers

Given two complex numbers z1 = a + bi and z2 = c + di, the arithmetics between

them are defined by:

z1 + z2 = (a + c) + (b + d)i
z1 − z2 = (a − c) + (b − d)i
z1 z2 = (a + bi)(c + di)
= (ac − bd) + (ad + bc)i
z1 a + bi c − di
= · (where z2 6= 0)
z2 c + di c − di
(ac + bd) (bc − ad)i
= 2 + 2
c + d2 c + d2

1.1.2. Conjugate and Modulus. Two important operations on complex numbers

are taking conjugates and modulus:

Definition 1.3 (Conjugate and Modulus). Given z = a + bi ∈ C, we denote and define:

• z := a − bi as the conjugate of z; and
√
• |z| := a2 + b2 as the modulus of z.

Remark 1.4. It is important to note that complex numbers are un-ordered. It does not
make sense to say z1 < z2 or z1 > z2 . However, since |z| is a real number, it makes sense
to make comparison of |z1 | and |z2 |.

Remark 1.5. Geometrically, z is obtained by reflecting z across the Re-axis (see Figure
1.1), and |z| is the magnitude of the position vector representing z.

Listed below are some very useful properties of complex numbers. Given any z, z1 , z2 ∈ C,
we have:

2
zz = |z| z=z |z| = |z|
z+z z−z
Re(z) = Im(z) =
2 2i
 
z1 z1
z1 ± z2 = z1 ± z2 z1 z2 = z1 z2 =
z2 z2
1.1. Complex Numbers 3

The proofs are all straight-forward and hence omitted. Simply let z = x + yi and
verify LHS and RHS are equal in each property. Let’s look at some examples on how to
make good use of these properties:
Example 1.1. Show that for any z1 , z2 ∈ C, we have:
2 2 2
|z1 + z2 | = |z1 | + |z2 | + 2Re(z1 z2 ).

Solution
2
The key step is to use the property that |z| = zz for any z ∈ C.
2
|z1 + z2 | = (z1 + z2 )(z1 + z2 ) = (z1 + z2 )(z1 + z2 )
= z1 z1 + z1 z2 + z1 z2 + z2 z2
2 2
= |z1 | + z1 z2 + z1 z2 + |z2 |
2 2
= |z1 | + |z2 | + 2Re(z1 z2 )

β
Example 1.2. Let α, β ∈ C\{0}. Show that αβ ∈ R if and only if α ∈ R.

Solution

(=⇒) Suppose αβ ∈ R, then we have αβ = αβ, and so αβ = αβ. Since α, β 6= 0,

by rearrangement we get:
 
β β β
= =
α α α
β β
Therefore, α is equal to its conjugate. It concludes that α ∈ R.
β 2
(⇐=) Conversely, let α = λ ∈ R. Then: αβ = αλα = λαα = λ |α| ∈ R.

It is important to note that in general |z1 + z2 | =

6 |z1 | + |z2 |. However, we do have:

Proposition 1.6 (Triangle Inequality). Let z1 , z2 ∈ C, we have:

|z1 + z2 | ≤ |z1 | + |z2 | .

Proof. From Example 1.1, we have:

2 2 2
|z1 + z2 | = |z1 | + |z2 | + 2Re(z1 z2 ).
Let z1 z2 = u + vi, where u, v ∈ R. Then, we have:
p
2Re(z1 z2 ) = 2u ≤ 2 u2 + v 2 = 2 |z1 z2 | = 2 |z1 | |z2 | = 2 |z1 | |z2 | .
Finally, we get:
2 2 2 2
|z1 + z2 | ≤ |z1 | + |z2 | + 2 |z1 | |z2 | = (|z1 | + |z2 |)
and it completes the proof by taking square root on both sides. 

Exercise 1.1. Let z1 , z2 ∈ C, show that:

2 2 2 2
|z1 + z2 | + |z1 − z2 | = 2(|z1 | + |z2 | ).

Exercise 1.2. Let α, β ∈ C. Suppose αz + βz ∈ R for any z ∈ C. Show that α = β.

Exercise 1.3. Let z1 , z2 ∈ C. Show that ||z1 | − |z2 || ≤ |z1 − z2 |.

4 1. Preliminaries

Exercise 1.4. Let p be the polynomial p(z) = c0 + c1 z + · · · + cd z d where d ≥ 1 and

{c0 , c1 , c2 , . . . , cd } is a (monotone) decreasing sequence of positive real numbers.
Prove that the polynomial equation p(z) = 0 does not have any roots with modulus
(strictly) less than 1.

1.1.3. Polar Form. There are two common types of coordinates in R2 , namely
rectangular and polar. Apart from the standard (rectangular) form x + yi for representing
a complex number, we can also represent a complex number by a polar form. The
conversion rule between rectangular and polar coordaintes is given by:
x = r cos θ
y = r sin θ
Therefore, a complex number z = x + yi can be written as:
z = (r cos θ) + i(r sin θ) = r(cos θ + i sin θ).
The form z = r(cos θ + i sin θ) is commonly called the polar form of z.
p
Note that |cos θ + i sin θ| = cos2 θ + sin2 θ = 1. When z = r(cos θ + i sin θ), it is
easy to see that r = |z|. However, the value of θ is not unique as both sin and cos are
periodic functions of period 2π. We define the principal argument of a complex number
to be the angle θ with a specified range described below:

Definition 1.7 (Principal Argument). Given a complex number z, the principal argument
of z, denoted by Arg(z), is defined to be the angle θ0 ∈ (−π, π] such that:
z = |z| (cos θ0 + i sin θ0 ).

√
For example, −1 − 3i has modulus 2 and so the r-coordinate is 2:
√ !
√ 1 3
−1 − 3i = 2 − − i .
2 2
√
To find the θ-coordinate, we solve cos θ = − 21 and sin θ = − 23 . From standard
trigonometry, we get θ = 4π3 + 2kπ for any integer k. The only θ that falls into the range
(−π, π] is − 2π
3 = 4π
3 − 2π. Therefore, we have:
    
√ 2π 2π
−1 − 3i = 2 cos − + i sin −
3 3
√
and Arg(−1 − 3i) = − 2π 3 .

Im
z = r(cos θ + i sin θ)

r
θ
Re
2 − 2π
3

√
z = −1 − 3i
1.1. Complex Numbers 5

In general, Arg(x + yi) can be found using tan−1 xy since if x = r cos θ and y = r sin θ,
then tan θ = xy . However, it is important to note that Arg(x + yi) is NOT simply equal to
tan−1 xy because by definition of the inverse tangent, tan−1 xy takes the value in (− π2 , π2 )
only. Precisely, we have (when x 6= 0):

 −1 y
tan x if (x, y) is in 1st and 4th quadrants;
−1 y
Arg(x + yi) = tan x + π if (x, y) is in 2nd quadrant;

 −1 y
tan x − π if (x, y) is in 3rd quadrant;
π
Furthermore, Arg(0 + yi) = 2 when y > 0; and Arg(0 + yi) = − π2 when y < 0. Note that
Arg(0 + 0i) is undefined.
Im

y y
tan−1 x +π tan−1 x

Re

y y
tan−1 x −π tan−1 x

Exercise 1.5. Express the following complex numbers in polar form, and find their
principal arguments Arg:
(a) 1 + 2i
(b) 1 − 2i
(c) cos(−π) + i sin(−π)
(d) −i

Exercise 1.6. Given |z| = 1, show that:

 
1+z
(a) Re =0
1−z
z−ω
(b) = 1 for any ω ∈ C such that ωz 6= 1.
1 − ωz

Exercise 1.7. Given z, ω ∈ C such that |z + ω| = |z − ω|, show that:

(a) izω ∈ R
π 3π
(b) Arg(z) − Arg(ω) = ± or ±
2 2

Exercise 1.8. Show that the real-valued function f : R2 \{(x, 0) : x ≤ 0} defined by

f (x, y) := Arg(x + yi) is continuous.
6 1. Preliminaries

1.1.4. De Moivre’s Theorem. By expressing complex numbers using polar form,

one can see that multiplications and divisions between two complex numbers are rotations
in the complex plane C. It thanks to the fact that:

(1.1) (cos θ + i sin θ)(cos φ + i sin φ)

= (cos θ cos φ − sin θ sin φ) + i(cos θ sin φ + sin θ cos φ)
= cos(θ + φ) + i sin(θ + φ)

Using (1.1), we can see that given z1 = r1 (cos θ1 + i sin θ1 ) and z2 = r2 (cos θ2 +
i sin θ2 ), then we have:

z1 z2 = r1 r2 (cos θ1 + i sin θ1 )(cos θ2 + i sin θ2 )

= r1 r2 (cos(θ1 + θ2 ) + i sin(θ1 + θ2 ))

Therefore, z1 z2 is obtained by rotating z1 by Arg(z2 ), and lengthen (or shorten) z1

by a factor of |z2 |. See the figure below:

Im z1 z2

z1
θ2
θ1
Re

An important consequence of (1.1) is the following celebrated theorem:

Theorem 1.8 (De Moivre’s Theorem). For any θ ∈ R and n ∈ Z, we have:

n
(1.2) (cos θ + i sin θ) = cos(nθ) + i sin(nθ).

Proof. We prove by induction for positive n’s. Clearly (1.2) is true when n = 1. Assume
that (1.2) is true when n = k for some positive integer k. Then, for n = k + 1, we have:

(cos θ + i sin θ)k+1 = (cos θ + i sin θ)k (cos θ + i sin θ)

= (cos(kθ) + i sin(kθ))(cos θ + i sin θ) (induction assumption)
= cos(kθ + θ) + i sin(kθ + θ) (from (1.1))
= cos((k + 1)θ) + i sin((k + 1)θ)

Hence (1.2) is true when n = k + 1. By induction, (1.2) is true for all positive integer n.
When n = 0, (1.2) also holds because (cos θ + i sin θ)0 = 1.
1.1. Complex Numbers 7

Finally we consider negative integers n. When n < 0, let m = −n then m is a positive

integer. From above, (1.2) holds for this m:
(cos θ + i sin θ)m = cos(mθ) + i sin(mθ)
(cos θ + i sin θ)−n = cos(−nθ) + i sin(−nθ)
1
= cos(nθ) − i sin(nθ)
(cos θ + i sin θ)n
1
(cos θ + i sin θ)n =
cos(nθ) − i sin(nθ)
1 cos(nθ) + i sin(nθ)
= ·
cos(nθ) − i sin(nθ) cos(nθ) + i sin(nθ)
cos(nθ) + i sin(nθ)
= = cos(nθ) + i sin(nθ).
cos2 (nθ) + sin2 (nθ)
This proves (1.2) holds for negative integers n, and hence completing the proof of the
theorem. 

De Moivre’s Theorem can be used to derive some trigonometric identities. For

example, consider (cos θ + i sin θ)3 . On one hand, De Moivre’s Theorem shows that:
(cos θ + i sin θ)3 = cos 3θ + i sin 3θ
and on the other hand, by expanding (cos θ + i sin θ)3 we get:
(cos θ + i sin θ)3 = cos3 θ + 3(cos2 θ)(i sin θ) + 3 cos θ(i sin θ)2 + (i sin θ)3
cos 3θ + i sin 3θ = (cos3 θ − 3 cos θ sin2 θ) + i(3 cos2 θ sin θ − sin3 θ)
By equating the real and imaginary parts, we get:
cos 3θ = cos3 θ − 3 cos θ sin2 θ = cos3 θ − 3 cos θ (1 − cos2 θ) = 4 cos3 θ − 3 cos θ
sin 3θ = 3 cos2 θ sin θ − sin3 θ = 3(1 − sin2 θ) sin θ − sin3 θ = 3 sin θ − 4 sin3 θ

Exercise 1.9. Use De Moivre’s Theorem to show that:

n
[2] k
X X
n
cos nθ = C2k Crk (−1)k+r cosn−2k+2r θ
k=0 r=0
for any n ∈ N. Here [ n2 ] denotes the integer part of n
2.

1.1.5. Roots of Complex Numbers. In the real number system, the root equation
xn = a where a 6= 0 and n ∈ N, has at most two solutions. When √ n is odd (no matter
whether a is positive or negative), the only
√ real solution
√ is x = n
a. When n is even and
a > 0, there are two real solutions x = n a or − n a. The equation has no solution when
n is even and a < 0.
However, in the complex number system, the root equation z n = a, where a ∈ C\{0}
and n ∈ N, always has n solutions! Let’s first look at the simplest equation z n = 1:
Certainly, 1 is a solution to the equation. Furthermore, using De Moivre’s Theorem,
we get:
 n    
2π 2π 2π 2π
cos + i sin = cos · n + i sin · n = cos(2π) + i sin(2π) = 1.
n n n n
Clearly, this shows the complex number cos 2π 2π n
n + i sin n satisfies the equation z = 1. In
2kπ 2kπ
fact, any number which can be expressed in form of cos n + i sin n , where k is an
8 1. Preliminaries

integer, is a solution to the root equation z n = 1:

 n
2kπ 2kπ
cos + i sin = cos(2kπ) + i sin(2kπ) = 1.
n n
 
2kπ 2kπ
Note that the set of roots cos + i sin : k ∈ Z consists of n distinct ele-
n n
ments only (instead of infinitely many), since
2kπ 2kπ 2mπ 2mπ
cos + i sin = cos + i sin
n n n n
if and only if k − m is a multiple of n. In other words, when k = n, the root cos 2kπ n +
i sin 2kπ
n is the same as the one with k = 0. Likewise, the root when k = n + 1 gives the
same root as the one with k = 1, etc. Overall, the set of n-th roots of 1 is essentially given
by the finite set:
 
2kπ 2kπ
cos + i sin : k ∈ {0, 1, 2, . . . , n − 1}
n n
and these n numbers are called the n-th root of 1. In terms of notations, we write:
 
1 2kπ 2kπ
1 n = cos + i sin : k ∈ {0, 1, 2, . . . , n − 1} .
n n
It is important to note that unlike the real number system, the n-th root of 1 is no longer
1
a single number. In contrast, 1 n represents a set of roots for the equation z n = 1.
Due to this distinctive difference from the real number system, from now on we will
√ 1
use n a to denote the n-th root of a in the real number system; while we will use a n to
denote the n-th root of a in the complex number system, which will be discussed in the
next paragraph.
Now consider the general root equation z n = a where a 6= 0. Suppose a can be
expressed in polar form as:
a = |a| (cos θ + i sin θ)
Then, one can show that:
    
p
n θ + 2kπ θ + 2kπ
|a| cos + i sin , k∈Z
| {z } n n
real n-th root

are solutions to the root equation z n = a, since:

     n
p
n θ + 2kπ θ + 2kπ
|a| cos + i sin
n n
 p n     
n θ + 2kπ θ + 2kπ
= |a| cos · n + i sin ·n
n n
= |a| (cos(θ + 2kπ) + i sin(θ + 2kπ))
= |a| (cos θ + i sin θ)
=a





Again, two numbers cos θ+2kπ n + i sin θ+2kπ
n and cos θ+2mπ
n + i sin θ+2mπ
n are
equal if and only if k − m is a multiple of n. Therefore, we conclude that the following n
complex numbers:
    
p
n θ + 2kπ θ + 2kπ
|a| cos + i sin
n n
are all the solutions to the root equation z n = a. Similar to the case of roots of 1, we
write the n-th root of a as:
1.1. Complex Numbers 9

Definition 1.9 (Roots of a Complex Number). Given any a ∈ C\{0} and n ∈ N, the
n-th roots of a is a set given by:
      
1 p Arg(a) + 2kπ Arg(a) + 2kπ
a n = n |a| cos + i sin : k ∈ {0, 1, . . . , n − 1}
n n

1 √ 1
Example 1.3. Find i 3 and (1 − 3i) 2 .

Solution
First express i into polar form i = cos π2 + i sin π2 . Hence by Definition 1.9, we have:
 π π 
1 + 2kπ + 2kπ
i 3 = cos 2 + i sin 2 : k = 0, 1, 2
3 3
 

 
π π 5π 5π 3π 3π 
= cos + i sin , cos + i sin , cos + i sin

| 6 {z 6} | 6 {z 6} | 2 {z 2}

k=0 k=1 k=2
(√ √ )
3+i 3−i
= , , −i
2 2
n √ 1o
Similarly, to find (1 − 3i) 2 , we first express:
√   π  π 
1 − 3i = 2 cos − + i sin −
3 3
Hence, by Definition 1.9, we have:
   π   π  
√ 1 √ − 3 + 2kπ − 3 + 2kπ
(1 − 3i) = 2 2 cos + i sin : k = 0, 1
2 2
( √ ! √ !)
√ 3−i √ − 3+i
= 2 , 2
2 2
(√ √ )
3−i − 3+i
= √ , √
2 2

Exercise 1.10. First, show that the roots of z 4 + 1 = 0 are:

 
1+i 1−i −1 + i −1 − i
√ , √ , √ , √ .
2 2 2 2
Then, use this result to factorize z 4 + 1 into the product of two quadratic polynomials
with real coefficients.

Exercise 1.11. By considering the roots of the equation z n − 1 = 0 (where n > 2 is

an integer), show that z n − 1 can be factorized into a product of linear and quadratic
polynomials with real coefficients:
( Qk−1

n (z − 1)(z + 1) r=1 z 2 − 2z cos 2πr n +1 if n = 2k
z −1= Qk−1 2 2πr


(z − 1) r=1 z − 2z cos n + 1 if n = 2k − 1

Next we discuss a useful observation about the n-th root of 1. Let

2π 2π
ω = cos + i sin
n n
10 1. Preliminaries

where n is an integer with n ≥ 2, then one can show the following identity holds:
1 + ω + ω 2 + . . . + ω n−1 = 0.

(1 − ω)(1 + ω + ω 2 + . . . + ω n−1 )
= (1 + ω + ω 2 + . . . + ω n−1 ) − ω(1 + ω + ω 2 + . . . + ω n−1 )
= (1 + ω + ω 2 + . . . + ω n−1 ) − (ω + ω 2 + . . . + ω n−1 + ω n )
= 1 − ωn
 n
2π 2π
= 1 − cos + i sin
n n
= 1 − (cos(2π) + i sin(2π)) = 1 − 1 = 0.
Since ω 6= 1 as n ≥ 1, we conclude that:
1 + ω + ω 2 + . . . + ω n−1 = 0.
Using this result, one can derive some trigonometric identities. Express ω in terms of its
real and imaginary parts:

   2  n−1
2π 2π 2π 2π 2π 2π
1 + cos + i sin + cos + i sin + . . . + cos + i sin =0
n n n n n n
| {z }
ω
   
2π 2π 4π 4π
1 + cos + i sin + cos + i sin + ...
n n n n
 
2(n − 1)π 2(n − 1)π
+ cos + i sin =0
n n
By equating the real and imaginary parts, we obtain two trigonometric identities:
2π 4π 2(n − 1)π
cos + cos + . . . + cos = −1
n n n
2π 4π 2(n − 1)π
sin + sin + . . . + sin =0
n n n

Exercise 1.12. Show that for any z 6= 1, we have

1 − z n+1
1 + z + z2 + . . . + zn = ,
1−z
and use it to show:
 
(2n+1)θ
1 sin 2
1 + cos θ + cos 2θ + . . . + cos nθ = +
2 2 sin θ2
for any θ ∈ (0, 2π).

Exercise 1.13. Let n ≥ 2 be an integer.

(a) Solve the equation (z + 1)n − 1 = 0.
π 2π (n − 1)π n
(b) Hence, show that sin · sin · · · sin = n−1 .
n n n 2
(c) Consider a circle of radius 1, and let P1 , P2 , . . . , Pn be the vertices of a regular
n-sided polygon inscribed in the circle. Denote the distance between any pair of
1.1. Complex Numbers 11

points P and Q by P Q. Using (b), show that:

Yn
P1 Pk = n.
k=2

Exercise 1.14. Let Pk (xk , yk ), where k = 1, 2, 3, be three distinct points in C and let
zk := xk + yk i be the complex number representing Pk . Denote ω = cos 2π 2π
3 + i sin 3 .
Show that 4P1 P2 P3 is equilateral if and only if
z1 + ωz2 + ω 2 z3 = 0.
Using this, show that it is impossible for 4P1 P2 P3 being equilateral if xk , yk ∈ Q for
all k = 1, 2, 3.
12 1. Preliminaries

1.2. Sequences and Series

1.2.1. Sequences of Complex Numbers. In this section, we will extend the no-
tion of sequences and series to complex numbers. As we shall see, many results and
convergence tests which hold for real numbers will carry over to complex numbers. Let’s
begin with the definition of convergence of complex sequences:

Definition 1.10 (Limit of Sequences). Let {zn }∞

n=1 be a sequence of complex numbers.
We say zn converges to w as n → ∞ if for any ε > 0, there exists an integer N > 0 such
that whenever n ≥ N , we have |zn − w| < ε.

Remark 1.11. We may abbreviate “zn converges to w as n → ∞” by simply saying:

lim zn = w.
n→∞

Remark 1.12. It is easy to see that lim zn = w is equivalent to lim |zn − w| = 0.

n→∞ n→∞

Remark 1.13. The definition of convergence of complex sequences is almost the same
as the that of real sequences. The only difference is now |·| represents the modulus
while for real sequence it represents the absolute value. Therefore, many computational
rules about limits carry over to complex sequences. For instance, if lim zn = L and
n→∞
lim wn = M , then we have
n→∞
lim (zn ± wn ) = L ± M
n→∞
lim (zn wn ) = LM
n→∞
zn L
lim = (whenever M 6= 0)
n→∞ wn M

Example 1.4. Consider the sequence zn = z n where z ∈ C is a fixed complex

number. Show from the definition of limits that:
• if |z| < 1, then zn converges to 0 as n → ∞;
• if z = 1, then zn converges to 1 as n → ∞;

Solution
First consider the case |z| < 1: if z = 0, then zn = 0 for any n and the desired result
clearly holds. From now on we assume z 6= 0. For any ε > 0, we pick a positive
log ε
integer N > log|z| . Whenever, n ≥ N , we have:
n N
|zn − 0| = |z n | = |z| ≤ |z| .
Here we have used the fact that |z| < 1 and n ≥ N . By our choice of N , we have:
log ε
N
|z| < |z| log|z|
log|z| ε
= |z| = ε.
This shows lim zn = 0 in case of |z| < 1. The case of z = 1 is trivial.
n→0
1.2. Sequences and Series 13

When |z| ≥ 1 and z 6= 1, the sequence zn = z n can be shown to diverge using the
squeezing principle (see Exercise 1.16). It can also be proved using the following useful
fact:

Proposition 1.14. A sequence {zn } ∈ C converges to w as a complex sequence if and only

if {Re(zn )} converges to Re(w) and {Im(zn )} converges to Im(w) as real sequences.

Proof. (=⇒)-part follows from the inequalities:

|Re(zn ) − Re(w)| ≤ |zn − w| and |Im(zn ) − Im(w)| ≤ |zn − w|
and the squeezing principle.
(⇐=)-part follows from the fact that:
q
2 2
|zn − w| = |Re(zn ) − Re(w)| + |Im(zn ) − Im(w)|


Now given a complex number z expressed in polar form as z = r(cos θ + i sin θ), and
suppose |z| ≥ 1 (i.e. r ≥ 1) and z 6= 1. Consider again the sequence zn = z n . By De
Moivre’s Theorem, we have:
zn = rn (cos nθ + i sin nθ).
It is well known in real analysis that when θ 6= 2kπ (where k ∈ Z), at least one of the real
sequences {cos nθ} and {sin nθ} diverges as n → ∞. Hence, when r ≥ 1 and θ 6= 2kπ
(k ∈ Z), at least one of the real sequences {rn cos nθ} and {rn sin nθ} diverges. This
shows zn diverges.
Since z n → 0 when |z| < 1, we also have the geometric series formula for complex
numbers:
a
a + az + az 2 + az 3 + · · · = .
1−z
when |z| < 1 just like the real case.

Exercise 1.15. Show that if lim zn = L, then lim zn = L and lim |zn | = |L|.
n→∞ n→∞ n→∞

Exercise 1.16. Show (without using Proposition 1.14) that if |z| ≥ 1 and z 6= 1,
then the sequence {z n } must diverge. [Hint: First prove the following inequality:
|z − 1| ≤ z n+1 − w + |z n − w|
for any z ∈ C such that |z| ≥ 1, and any w ∈ C.]

In Real Analysis, there is a notion of Cauchy sequences which describe sequences that
are closer and closer to each other. It is a priori different from convergent sequences, which
are sequences that are closer and closer to a certain limit. However, it is well-known that
for sequences in R, the Cauchy condition will guarantee convergence. This important
fact is known as completeness of real numbers.
In Complex Analysis, we have a similar notion of Cauchy sequences and completeness,
to be discussed below.

Definition 1.15 (Cauchy Sequence). A sequence {zn }∞ n=1 of complex numbers is said
to be a Cauchy sequence if and only if for any ε > 0, there exists an integer N ∈ N such
that whenever m, n ≥ N , we have |zn − zm | < ε.
14 1. Preliminaries

Theorem 1.16 (Completeness of C). Every Cauchy sequence of complex numbers con-
verges to a certain complex number. In other words, C is complete.

Proof. Let {zn } be a Cauchy sequence of complex numbers. We need to show it converges.
Write zn = xn + iyn , where xn , yn ∈ R. Since we have:
|xn − xm | ≤ |zn − zm |
|yn − ym | ≤ |zn − zm |
and given that {zn } is a Cauchy sequence, the real sequences {xn } and {yn } are also
Cauchy sequences. By Completeness of R, both {xn } and {yn } converge to some real
numbers x∞ and y∞ respectively. By Proposition 1.14, the complex sequence {zn }
converges to x∞ + iy∞ . 

Exercise 1.17. Suppose {zn }∞ n=0 is a complex sequence. Suppose there exists a real
constant α ∈ [0, 1) such that:
|zn+1 − zn | ≤ α |zn − zn−1 | for any n ∈ N.
Show that the complex sequence {zn }∞
n=0 converges.

∞
X
1.2.2. Series of Complex Numbers. An (infinite) series zn of complex num-
n=1
N
X
bers zn ∈ C is the limit (if exists) of the N -th partial sums zn as N → ∞. In Real
n=1
Analysis, we learned that many series convergence tests rely on the fact that R is com-
plete. Now that we know C is also complete (Theorem 1.16), we can generalize many
series convergence tests for C.

Definition 1.17 (Absolute and Conditional Convergences). A series of complex numbers

∞
X ∞
X ∞
X
zn is said to converge absolutely if the series |zn | converges. A series zn is
n=1 n=1 n=1
said to converge conditionally if it converges but does not converge absolutely.

∞
X
Proposition 1.18 (Absolute Convergence Test). If the series |zn | converges, then the
n=1
∞
X
complex series zn also converges.
n=1

Proof. The proof is almost identical to the real case, mutatis mutandis. Given that
X∞ N
X
|zn | converges, its N -th partial sum |zn | is a Cauchy sequence. Now consider
n=1 n=1
N
X
the sequence of N -th partial sums zn . We want to show the later is also a Cauchy
n=1
sequence.
For any ε > 0, there exists an integer K > 0 such that whenever M > N ≥ K, we
have
XM N
X
|zn | − |zn | < ε.
n=1 n=1
1.2. Sequences and Series 15

It implies:

M
X N
X M
X M
X M
X N
X
zn − zn = zn ≤ |zn | = |zn | − |zn | < ε.
n=1 n=1 n=N +1 n=N +1 n=1 n=1

N
X
Therefore, zn is also a Cauchy sequence. By completeness of C (Theorem 1.16), the
n=1
N
X ∞
X
N -th partial sum zn (and hence the infinite series zn ) converges. 
n=1 n=1

X∞ n
i
Example 1.5. Does the series converge absolutely, conditionally, or does not
n=1
n
X∞
in
converge? How about the series ?
n=1
n2

Solution
X ∞ X∞
in 1
The series = diverges by p-test. The N -th partial sum can be decom-
n=1
n n=1
n
posed into:
 k
  k−1

XN
in  − 1 + 1 − . . . + (−1) + 1 − 1 + 1 + . . . + (−1) i if N = 2k
2 4 2k 3 5 2k−1
=  k
  k+1

n  − 1 + 1 − . . . + (−1) + 1 − 1 + 1 + . . . + (−1) i if N = 2k + 1
n=1 2 4 2k 3 5 2k+1

In either case, the real and imaginary parts converge by alternating series test. By
X∞ n
i
Proposition 1.14, the series converges, and so it converges conditionally.
n=1
n
X∞ X∞ X∞
in in 1
Now consider 2
. The series 2
= converges by p-test. There-
n=1
n n=1
n n=1
n2
X∞
in
fore, the series converges absolutely.
n=1
n2

One good property of an absolute convergent series is that we can rearrange the
terms as we wish without changing the value of the series. Precisely, given an absolute
∞
X
convergent series zn =: L and a bijection σ : N → N, then the rearranged series
n=1
∞
X
zσ(n) also converges absolutely to the limit L. The proof in the complex case is
n=1
∞
X
based on the analogous result in the real case: if the complex series zn converges
n=1
∞
X ∞
X
absolutely, then the real series Re(zn ) and Im(zn ) both converge absolutely by
n=1 n=1
the comparison test (as |Re(zn )| ≤ |zn | and |Im(zn )| ≤ |zn |). It then follows from the
16 1. Preliminaries

∞
X ∞
X
rearrangement of real absolute convergent series that Re(zσ(n) ) = Re(zn ) and
n=1 n=1
∞
X ∞
X X∞ ∞
X
Im(zσ(n) ) = Im(zn ). By Proposition 1.14, it proves zσ(n) = zn .
n=1 n=1 n=1 n=1
Recall from Real Analysis that the ratio test and root test follow from the absolute
convergence test and geometric series test. Now we learned that both hold on C, hence
the ratio test and root test can be extended to complex series:
∞
X
Proposition 1.19 (Ratio Test). Consider the complex series zn :
n=1

X∞
zn+1
• If lim < 1, then zn converges absolutely.
n→∞ zn n=1
X∞
zn+1
• If lim > 1, then zn diverges.
n→∞ zn n=1
zn+1
• If lim = 1, then no conclusion can be drawn.
n→∞ zn

∞
X
Proposition 1.20 (Root Test). Consider the complex series zn :
n=1

p ∞
X
n
• If lim |zn | < 1, then zn converges absolutely.
n→∞
n=1
p ∞
X
• If lim n |zn | > 1, then zn diverges.
n→∞
n=1
p
n
• If lim |zn | = 1, then no conclusion can be drawn.
n→∞

Remark 1.21. The proofs of the ratio and root tests are the same as in the real case. We
omit their proofs but we encourage readers to write down their proofs as an exercise.

X∞
zn
Example 1.6. Show that for any z ∈ C, the complex series converges
n=0
n!
absolutely.

Solution
We use the ratio test. Consider:
z n+1 /(n + 1)! z n+1 n!
lim n
= lim n
n→∞ z /n! n→∞ z (n + 1)!
z |z|
= lim = lim
n→∞ n + 1 n→∞ n + 1

= 0 < 1 for any z ∈ C.

X∞
zn
Hence, the series converges absolutely by ratio test (Proposition 1.19).
n=0
n!
1.2. Sequences and Series 17

Alternatively, we can also use the root test (Proposition 1.20) by showing that:
s
zn |z|
lim n = lim √ =0<1
n→∞ n! n→∞ n n!
√n
for any z ∈ C. Here we have used the fact that lim n! = ∞.
n→∞

∞
X
Example 1.7. Determine all complex numbers z such that the series nz n con-
n=0
verges.

Solution
(n + 1)z n+1 (n + 1)
Consider the limit lim n
= lim |z| = |z|. Therefore, by ratio
n→∞ nz n→∞ n
test (Proposition 1.19), the series converges absolutely when |z| < 1; and diverges
when |z| > 1.
When |z| = 1, the ratio test fails to conclude anything. In this case, we let
∞
X
z = cos θ + i sin θ where θ ∈ R. Then, the series is given by (n cos nθ + in sin nθ),
n=0
and the real and imaginary parts are:
∞
! ∞ ∞
! ∞
X X X X
n n
Re nz = n cos(nθ) and Im nz = n sin(nθ).
n=0 n=0 n=0 n=0
By Proposition 1.14, if the complex series converges, then both their real and
imaginary parts converge, and in particular we have:
lim n cos(nθ) = 0 and lim n sin(nθ) = 0.
n→∞ n→∞
By Squeeze Theorem, it will imply:
lim cos(nθ) = lim sin(nθ) = 0.
n→∞ n→∞

However, it would contradict the fact that cos2 (nθ) + sin2 (nθ) = 1; and so the series
∞
X
nz n does not converge when |z| = 1.
n=0
∞
X
Conclusion: the series nz n converges if and only if |z| < 1.
n=0

Below is a very nice example that demonstrates complex numbers could come into
play to solve a problem (about integers) which seems to have nothing to do with complex
numbers.

Example 1.8. An (infinite) arithmetic sequence is of the form: {a, a + d, a + 2d, . . .}

where d is called the common difference of the arithmetic sequence.
Show that it is impossible to partition N into a finite disjoint union of arithmetic
sequences with mutually distinct common differences (greater than 1).
18 1. Preliminaries

Solution
What is needed to show is that for any integer k ≥ 2, if ai , di ∈ N where 1 ≤
i ≤ k are positive integers such that whenever i 6= j, we have di 6= dj . Denote
S1 , S2 , . . . , Sk are the sets of infinite arithmetic sequences:
S1 = {a1 , a1 + d1 , a1 + 2d1 , . . .}
..
.
Sk = {ak , ak + dk , ak + 2dk , . . .}
We need to show Si ∩ Sj = ∅ for any i 6= j, then:
k
[
N 6= Si .
i=1
X
For each 1 ≤ i ≤ k, we first express the sum z n where |z| < 1 in terms of ai , di
n∈Si
and z with no summation sign.

X
z n = z ai + z ai +di + z ai +2di + · · ·
n∈Si


= z ai 1 + z di + (z di )2 + · · ·
z ai
=
1 − z di
di
using the geometric series formula (note that the common ratio z di = |z| < 1).
Now suppose on the contrary N = {1, 2, 3, · · · } can be decomposed into disjoint
union of Si ’s, then X X X
zn = zn + · · · + zn
n∈N n∈S1 n∈Sk
for any |z| < 1. Here we have used the absolute convergence of geometric series,
and the rearrangement theorem of absolute convergent series. However, it will
imply:
z z a1 z ak
= + · · · + .
1−z 1 − z d1 1 − z dk
Assume without loss of generality that d1 < d2 < · · · < dk so that dk is the largest
among all d’s. Letting z → cos 2π 2π
dk + i sin dk which is an dk -th root of 1, then for any
j = 1, 2, · · · , k − 1:
z cos 2π 2π
dk + i sin dk
lim =
z→cos 2π
dk +i sin 2π
dk
1−z 1 − (cos 2π 2π
dk + i sin dk )
2a π 2a π
z aj cos djk + i sin djk
lim = 2a π 2a π .
z→cos 2π
dk +i sin 2π
dk
1 − z dj 1 − (cos djk + i sin djk )
Note that the both limits exist because the denominators do not go to zero.
However, for the last term, we will get:
z ak cos 2adkkπ + i sin 2adkkπ
lim = which does not exist.
z→cos 2π
dk +i sin 2π
dk
1 − z dk 1−1
This gives a contradiction. Hence N cannot be decomposed into a finite disjoint
union of arithmetic sequences of distinct common differences.
1.2. Sequences and Series 19

Exercise 1.18. Determine whether each of the following complex series converges
absolutely, conditionally, or diverge:
X∞
(1 − 3i)n
(a)
n=0
(4 + i)2n
∞
X n2
(b)
n=1
n + n3 i
∞
X
(c) (cos n − i sin n)
n=1

Exercise 1.19. In each of the following complex series: (i) determine all complex
numbers z such that the series converges, (ii) sketch the range of these z’s on the
complex plane C.
∞
X
(a) zn
n=1
X∞  n
z
(b)
n=1
z+1
X∞
(−1)n z 5n
(c)
n=1
n!
X∞
z n!
(d)
n=1
n2

Exercise 1.20. Suppose z ∈ C.

(a) Assume |z| =
6 1 and z 6= 0, show that for any n ∈ N, we have:
zn 1
≤
1 + z 2n n
|z| − |z|
−n

 ∞
zn
(b) Using (a), or otherwise, find all z ∈ C such that the sequence
1 + z 2n n=1
converges.
∞
X zn
(c) Find all z ∈ C such that the series converges.
n=1
1 + z 2n

X∞
zn
1.2.3. Euler’s Identity. The series considered in Example 1.6 is an important
n=0
n!
one – it defines the complex exponential function. When z = x is a real number, the value
of the series is given by ex . Given that the series converges for any z ∈ C, we define ez to
be the limit of this series:

Definition 1.22 (Complex Exponential). Let z ∈ C, the exponential ez , or equivalently

exp(z), of z is defined by:
X∞
z zn
e := .
n=0
n!
20 1. Preliminaries

Remark 1.23. Some books (especially those about physics and engineering) define
ex+yi = ex (cos y + i sin y), but here we will first define ez as an infinite series first, and
justify that ex+yi = ex (cos y + i sin y) from the series definition of ez .

Here is the famous Euler’s identity that relates complex exponentials with the polar
form of a complex number:

Theorem 1.24 (Euler’s Identity). For any θ ∈ R, we have:

(1.3) eiθ = cos θ + i sin θ.

Proof. The key idea is to split the defining series into real and imaginary parts.
X∞ 2N n n
X
(iθ)n i θ
eiθ = = lim
n=0
n! N →∞
n=0
n!
N N −1 2k+1 2k+1
!
X i2k θ2k X i θ
= lim + [by rearrangement]
N →∞ (2k)! (2k + 1)!
k=0 k=0
N
! N −1
X (−1)k θ2k (−1)k θ2k+1 X
= lim +i lim [using i2k = (i2 )k = (−1)k ]
N →∞ (2k)! N →∞ (2k + 1)!
k=0 k=0
| {z } | {z }
=cos θ =sin θ
= cos θ + i sin θ

Remark 1.25. From (1.3), it is evident that we have:
eiπ + 1 = 0
which is a single identity involving 5 most important constants in mathematics, namely 1,
0, e, π and i.
Remark 1.26. From the Euler’s identity, we can now write down the polar form of a
complex number in a simpler way: if z = r(cos θ + i sin θ), then we can write:
z = reiθ .
In particular, any z ∈ C such that |z| = 1 can be expressed as z = eiθ for some θ ∈ R.

We are going to show that the complex exponential has the property that ez ew = ez+w
just like the real case. Informally, we express both ez and ew into two infinite series. After
multiplying the two series, we express the double sum diagonally:
∞
! ∞ ! ∞ X ∞
X zn X wm X z n wm
z w
e e = =
n=0
n! m=0
m! n=0 m=0
n!m!
∞
X X ∞
X X z n wk−n k
z n wm
= = [since m = k − n]
n!m! n=0
n!(k − n)!
k=0 m+n=k k=0
∞ X
X k
Cnk z n wk−n k!
= [since Cnk = ]
n=0
k! (n − k)!n!
k=0
X∞
(z + w)k
= = ez+w [Binomial Theorem]
k!
k=0

Although this “proof” above seems convincing and neat, there isPa little
P step we need
to justify, namely why we can rearrange the infinite double sum n m in a diagonal
1.2. Sequences and Series 21

P P P P
way: k m+n=k ? We have seen in Real Analysis that even switching n and m may
sometimes result in a different sum. Below we give a rigorous (and more refined) proof
of this fact:

Proposition 1.27. For any z, w ∈ C, we have ez ew = ez+w .

XN XN
zn wm
Proof. Consider the N -th partial sums and , then:
n=0
n! m=0
m!
N
! N
! N X N
X zn X wm X z n wm
=
n=0
n! m=0
m! n=0 m=0
n!m!
| {z }
Region I in Fig. 1.2
2N
X X N
X −1 2N
X −m N −1 2N −n
z w n m
z n wm X X z n wm
= − −
n!m! m=0
n!m! n=0
n!m!
k=0 m+n=k n=N +1 m=N +1
| {z } | {z } | {z }
Region I+II+III in Fig. 1.2 Region II in Fig. 1.2 Region III in Fig. 1.2
PN PN
Here we break down the finite double sum n m into three triangular sums. See
Figure 1.2 for illustration. For the sum corresponding to the large triangle (Region
I+II+III in Figure 1.2), we can rewrite it as:
2N
X X 2N k
X X z n wk−n X X C k z n wk−n 2N k
X (z + w)k 2N
z n wm n
= = = → ez+w
n!m! n=0
n!(k − n)! n=0
k! k!
k=0 m+n=k k=0 k=0 k=0

as N → ∞.
For Region II in Figure 1.2, we can show that it converges to 0 as N → ∞:
N
X −1 2N
X −m N
X −1 2N
X −m
z n wm z n wm
≤
m=0 n=N +1
n!m! m=0
n!m!
n=N +1
N
X −1
X∞ n m
|z| |w|
≤
m=0 n=N +1
n! m!
N −1
! ∞
!
X |w|
m X |z|
n
=
m=0
m! n!
n=N +1
∞
!
X |z|
n
≤ e|w| →0
n!
n=N +1
∞
X n
|z|
as N → ∞ since the infinite sum converges (to e|z| ). The sum corresponding to
n=0
n!
Region III in Figure 1.2 can be shown to converge to 0 by switching m and n, and z and
w in the above argument.
Overall, we have shown:
N
! N !
X zn X wm
lim
N →∞
n=0
n! m=0
m!
2N
X X N
X −1 2N
X −m N
X −1 2N
X −n
z n wm z n wm z n wm
= lim − lim − lim
N →∞ n!m! N →∞
m=0
n!m! N →∞
n=0
n!m!
k=0 m+n=k n=N +1 m=N +1
z+w
=e − 0 − 0,
22 1. Preliminaries

which implies ez ew = ez+w as desired. 

2N

II

I III
m
O N 2N

Figure 1.2

∞
X ∞
X
Exercise 1.21. Given two series zn and wn which converge absolutely to A
n=0 n=0
and B respectively, show that the series below converges absolutely to AB:
∞ k
!
X X
zn wk−n
k=0 n=0

Exercise 1.22. Suppose {an }∞

n=1 is a (monotonically) decreasing sequence of real
numbers such that lim an = 0, and {zn }∞n=1 be a sequence of complex numbers
n→∞
N
X
with the property that there is a constant C > 0 such that zn ≤ C for any N .
n=1
∞
X
Show that the series an zn converges. [Hint: First prove the following summation-
n=1
by-parts formula
N
X +1 N
X +1 N X
X n
an zn = aN +1 zn + (an − an+1 )zk ,
n=1 n=1 n=1 k=1
and make good use of the given conditions.]
Furthermore, use the above result to prove the alternating series test in Real
Analysis.

X∞
zn
Exercise 1.23. Using the result from Exercise 1.22, show that the series
n=1
n
converges for any z such that |z| = 1 and z 6= 1.

Using the multiplicative property ez ew = ez+w , one can show the following properties
about the complex exponential function. We leave the proofs for readers.
Remark 1.28. For any z = x + yi ∈ C where x, y ∈ R, we have:
n
• (ez ) = enz for any integer n.
1.2. Sequences and Series 23

• ez = ex eiy = ex (cos y + i sin y), and hence |ez | = ex .

• ez 6= 0.

The complex exponential az with other real base a > 0 is defined via the natural
exponential ez . Recall that a = eln a , and we define:

az := e(ln a)z .

Using this definition, some properties of ez extend to complex exponentials az with an

arbitrary real base a > 0. Proofs are again left for readers.

Remark 1.29. For any real a, b > 0 and z, w ∈ C we have:

n
• (az ) = anz for any integer n.
• az aw = az+w
• |az | = aRe(z)
• az 6= 0
• (ab)z = az bz

Remark 1.30. For any positive integer n, the rational number n1 is no doubt also a
1
complex number. Therefore, now e n could mean two different things, namely the value
∞
k
X 1
n
of the series , or the n-th roots of e. It is a confusing ambiguity but fortunately
k!
k=0
we seldom deal with both of them in the same context. One way to avoid such a confusion
1
is to represent the n-th roots of e by e n , and use exp( n1 ) to represent the value of the
aforesaid series.

1.2.4. Riemann ζ Function: the first encounter. The Riemann zeta function, de-
noted by ζ(z), is of central importance in Complex Analysis and Number Theory. It is an
infinite series defined as:
X∞
1
ζ(z) := z
n=1
n

for Re(z) > 1. This complex series motivates the discussions of the famous Riemann
Hypothesis, which is a conjecture purposed by Riemann in 1859 and remains unsolved
as of today (February 29, 2020). The statement of the Riemann Hypothesis will be
explained after we learn about analytic continuation of holomorphic functions. The
Riemann zeta function has deep connections with Number Theory, in particular on the
study of distribution of prime numbers. It is used to show the renowned Prime Number
Theorem, which asserts that:
π(x)
lim =1
x→∞ x/ ln x

where π(x) is the number of positive prime numbers less than or equal to x. The
deep connection between ζ(z) and prime numbers is beyond the scope of this course.
Interested readers may consult Stein-Shakarchi’s book.
Meanwhile, we show that this series converges absolutely when Re(z) > 1 by the
(real) p-test. The main reason is as follows. Write z = x + yi where x, y ∈ R, then we
have:
1 1 1 1 1
= z log n = x log n iy log n = x = Re(z) .
nz e |e e | n n
24 1. Preliminaries

∞
X1
By (real) p-test, the series Re(z)
converges if and only if Re(z) > 1. Therefore, by
n=1
n
X∞
1
the (complex) absolute convergence test, the series converges absolutely when
n=1
nz
Re(z) > 1.
1.3. Point-Set Topology of C 25

1.3. Point-Set Topology of C

In this section, we will introduce several terminologies and topological concepts about
subsets of C. These topological concepts will come up from time to time in the course.
To begin, let’s define some standard notations we will use in the rest of the course.
Let z0 ∈ C and r > 0. From now on, we will denote:
Br (z0 ) = {z ∈ C : |z − z0 | < r}
Br (z0 ) = {z ∈ C : |z − z0 | ≤ r}
∂Br (z0 ) = {z ∈ C : |z − z0 | = r}
which are respectively the open ball, closed ball and circle with radius r centered at z0 .
In the literature of Complex Analysis, it is often that the term disc is used instead of ball.

1.3.1. Open and Closed Subsets. Intuitively, an open subset Ω in C is one that
does not have a boundary. However, this “definition” is not rigorous enough since the
term “boundary” has not been defined so far. We are going to give a rigorous definition
of open and closed subsets here. We first define:

Definition 1.31 (Interior, Boundary and Exterior Points). Consider a set U ⊂ C. We

say that z ∈ C is an interior point of U if there exists ε > 0 such that Bε (z) ⊂ U . We say
that w ∈ C is a boundary point of U if for any ε > 0, both Bε (w) ∩ U and Bε (w) ∩ (C\U )
are non-empty. We say η ∈ C is an exterior point of U if there exists δ > 0 such that
Bδ (η) ⊂ C\U .

In the figure below, the yellow set is the subset U ⊂ C. The point z ∈ U is an interior
ball because by drawing a ball with a small enough radius (i.e. the blue ball), the ball is
completely inside U . In layman terms, an interior point of U is a point z whose “nearby”
points are contained in U .
On the other hand, the point w in the figure below is a boundary point. No matter
how small the ball you draw around w, that ball must contain some points in U , and
some points not in U . In layman terms, a boundary point of U is a point w at which if
you look around it, you can see “nearby” some points in U and some point not in U .
The point η in the figure is an exterior point of U . In layman terms, it is a point
whose “nearby” are outside U .

η ε
δ z

w
26 1. Preliminaries

Remark 1.32. Since z ∈ Bε (z) for any z ∈ C and ε > 0, if z is an interior point of U , it
is automatically that z ∈ U . In other words, an interior point of a set must belong to that
set. However, a boundary point of a set can be contained or not contained in the set.
Furthermore, according to the definitions, interior points, boundary points and exterior
points are mutually exclusive.

Example 1.9. Find all interior, boundary and exterior points of the set:
U = {z ∈ C : Re(z) > 1}.

Solution
We claim that the set of interior points is U itself. For any z ∈ U , we have Re(z) > 1.
Write z = x + yi, then we have x > 1. We need to find a small ε > 0 such that
Bε (x + yi) ⊂ U . According to the figure below, an appropriate choice of ε should
be ε = x−1
2 . We next verify that it is indeed Bε (z) ⊂ U for this choice of ε.
x−1
For any α ∈ Bε (z), we have |α − z| < ε = . Then, by Re(z − α) ≤ |z − α|,
2
we know that:
x−1 x−1
Re(z − α) < =⇒ x − Re(α) < .
2 2
x−1 x+1 1+1
By rearrangement, we get Re(α) > x − = > = 1, which is equiv-
2 2 2
alently to saying that α ∈ U . This shows Bε (z) ⊂ U , and hence z is an interior
point.

η z

1 x

Next we show that every point w with Re(w) = 1 is a boundary point of U .

Given any ε > 0, we consider the ball Bε (w). The point w − 2ε lies in the ball Bε (w)
and has real part 1 − 2ε and hence is not in U ; while the point w + 2ε is also in the
ball Bε (w) but has real part 1 + 2ε and so is inside U . Therefore, both Bε (w) ∩ U
and Bε (w) ∩ (C\U ) are non-empty, it concludes that w is a boundary point of U .
Finally, we claim that any point η ∈ C with Re(η) < 1 is an exterior point of U .
To prove this claim, we choose a δ = 1−Re(η)
2 and show that Bδ (η) ⊂ C\U : Given
any β ∈ Bδ (η), we have:
1 − Re(η) 1 + Re(η)
Re(β − η) ≤ |β − η| < δ = =⇒ Re(β) < < 1.
2 2
Therefore, β 6∈ U , and it shows Bδ ⊂ C\U . It completes the claim that η is an
exterior point of U .
1.3. Point-Set Topology of C 27

Exercise 1.24. Find all the interior, boundary and exterior points of each set below:
(a) U1 = {z ∈ C : Re(z) ≥ 0 and Im(z) < 0}.
(b) U2 = Br (z0 ) where z0 ∈ C is a fixed number and r > 0
(c) U3 = Br (z0 ) where z0 ∈ C is a fixed number and r > 0.
(d) U4 = ∂Br (z0 ) where z0 ∈ C is a fixed number and r > 0.
(e) U5 = C.
From now on, given any set U ⊂ C, we denote and define:
U c := C\U = the complement of U in C
U ◦ := set of all interior points of U
∂U := set of all boundary points of U
U := U ∪ ∂U = the closure of U
There is no standard notation for the set of all exterior points though. According to the
definition of interior points, we must have U ◦ ⊂ U .
We are now ready to define what are open sets and closed sets. The way we define
open sets is very common in many other textbooks, while the way we define closed sets
may sound different from some textbooks but it is more intuitive and is nonetheless
equivalent to the definition found in other textbooks.

Definition 1.33 (Open and Closed Sets). A set Ω ⊂ C is said to be open if every point
z ∈ Ω is an interior point of Ω (i.e. Ω = Ω◦ ). A set E ⊂ C is said to be closed if all
boundary points of E belong to E (i.e. ∂E ⊂ E).

Remark 1.34. Note that it is always true that Ω◦ ⊂ Ω.

Let’s look at some examples. Consider the set Ω = {z ∈ C : Re(z) > 1}:
Ω◦ = {z ∈ C : Re(z) > 1} = Ω
∂Ω = {z ∈ C : Re(z) = 1} 6⊂ Ω
Therefore, Ω is an open set, but is not closed.
Let’s look at another example: E = {z ∈ C : Re(z) ≥ 1}. By inspection (we left the
detail for readers), we can see:
E ◦ = {z ∈ C : Re(z) > 1} =
6 E
∂E = {z ∈ C : Re(z) = 1} ⊂ E
Therefore, E is not open, but is closed.
There are sets which are not open and not closed! For instance, consider the unit
circle W = {z ∈ C : Re(z) ≥ 0 and Im(z) > 0}. We can see from Figure 1.3 that:
W ◦ = {z ∈ C : Re(z) > 0 and Im(z) > 0} =
6 W
∂W = {x + 0i ∈ C : x ≥ 0} ∪ {0 + yi ∈ C : y ≥ 0} 6⊂ W.
W is neither open nor closed.
28 1. Preliminaries

(a) W (b) W ◦ (c) ∂W

Figure 1.3. The set W = {z ∈ C : Re(z) ≥ 0 and Im(z) > 0} and its interior and boundary sets

Surprisingly, there are sets which are both open and closed (so “open” and “closed”
are not exactly opposite, which is a linguistic nightmare)! For subsets of C, there are not
many though. They are the empty set ∅ and the whole C. It is easy to see that C◦ = C
and ∂C = ∅ ⊂ C (the empty-set is a subset of every set). This shows C is both open and
closed.
The argument that shows ∅ is both open and closed has a bit of philosophical favor.
We claim that ∅◦ = ∅. Suppose otherwise, then we must have ∅◦ 6⊂ ∅ (since ∅ is a subset
of every set). This means there exists z ∈ ∅◦ such that z 6∈ ∅. Then, z being an interior
point of ∅ implies there exists ε > 0 such that Bε (z) ⊂ ∅, which is clearly impossible!
This shows ∅◦ = ∅ and so the empty set is open. To show ∅ is closed as well, we claim
∂∅ = ∅. Suppose ∂∅ is non-empty, then we can pick w ∈ ∂∅, then for any δ > 0, both
Bδ (w) ∩ ∅ and Bδ (w) ∩ (C\∅) are non-empty. However, the former cannot happen! This
concludes ∂∅ = ∅, and so ∅ is closed as well!
Remark 1.35. There is an interesting YouTube video titled “Hitler learns Topology”.

Exercise 1.25. Determine whether each set U1 to U5 in Exercise 1.24 is open, closed,
neither or both.
Readers who have learned a bit point-set topology may have seen another definition
of closed sets, namely a set E is closed if its complement E c is open. We are going to
show that this is equivalent to our definition:

Proposition 1.36. For any set E ⊂ C, we have

∂E ⊂ E ⇐⇒ E c is open.

Proof. (=⇒)-part: Suppose ∂E ⊂ E. Consider z ∈ E c , by the given condition ∂E ⊂ E,

we know z 6∈ ∂E. This shows there exists ε > 0 such that at least one of the sets Bε (z)∩E
or Bε (z) ∩ E c is empty. Since z ∈ E c , we must have Bε (z) ∩ E = ∅, which is equivalent
to saying Bε (z) ⊂ E c . This shows E c is open.
(⇐⇒)-part: Suppose E c is open. Consider w ∈ ∂E, and we need to show w ∈ E.
Suppose not, then w ∈ E c . By the openness of E c , there exists δ > 0 such that
Bδ (w) ⊂ E c . However, it would imply Bδ (w) ∩ E = ∅, contradicting to the fact that
w ∈ ∂E. This shows w ∈ E, completing the proof that ∂E ⊂ E. 

Therefore, from now on we can say a set is closed if and only if its complement is
open, which is more convenient sometimes. For instance, this fact can be used to show
an important and nice property about a closed set E: if there is a convergent sequence in
E, then the limit must be inside E.

Proposition 1.37. Let E ⊂ C be a closed set. Suppose {zn }∞

n=1 is a complex sequence
such that zn ∈ E for any n. If lim zn = w, then w ∈ E.
n→∞
1.3. Point-Set Topology of C 29

Proof. We prove by contradiction. The key idea is that if w 6∈ E, then one can draw a
small ball around w such that the ball is completely outside E. However, then zn which
approaches w must go within the ball, and hence outside E, when n is large (see Figure
1.4).
Here we present the detail: suppose w 6∈ E, then w ∈ E c . By Proposition 1.36, E c
is open and so there exists ε > 0 such that Bε (w) ⊂ E c . By the fact that zn → w, there
exists N ∈ N such that whenever n ≥ N , we have |zn − w| < ε. However, it implies:
zn ∈ Bε (w) ⊂ E c =⇒ zn 6∈ E
which is clearly a contradiction. It proves w ∈ E. 

w
zn

z4
z3 E
z2 z1

Figure 1.4. If E is closed, w 6∈ E and zn → w, then zn must go outside E for large n.

Below is a list of useful facts about open and closed sets. We will prove some of them
and leave the others as exercises for readers.
Proposition 1.38. Open and closed sets in C have the following properties:
[
• The union Uα of any family (finite or infinite) of open sets {Uα } in C is open.
α
N
\
• The intersection Uk of a finite family of open sets U1 , . . . , UN in C is open.
k=1
N
[
• The union Ek of a finite family of closed sets E1 , . . . , EN in C is closed.
k=1
\
• The intersection Eα of any family (finite or infinite) of closed sets {Eα } in C is
α
closed.

Proof. Let’s prove the second statement only, that if U1 , . . . , UN are open, then their
N
\
intersection is also open. Let z ∈ Uk , then z ∈ Uk for any k = 1, . . . , N . For each
k=1
k, since Uk is open, z is an interior point of Uk and so there exists εk > 0 such that
Bεk (z) ⊂ Uk . Let ε = min{ε1 , . . . , εN }, which is positive, then ε ≤ εk for any k, and so
we have:
Bε (z) ⊂ Bεk (z) ⊂ Uk for any k = 1, . . . , N .
N
\ N
\ N
\
Therefore, Bε (z) ⊂ Uk . This shows z is an interior point of Uk . As a result, Uk
k=1 k=1 k=1
is an open set.
30 1. Preliminaries

We leave the proof of the first statement as an exercise for readers. Once the
first two statements are established, the third and fourth statements about
!c closed sets
[ \
easily follow from Proposition 1.36 and De Morgan’s Rule: Ek = Ekc and
!c k k
\ [
Eα = Eαc . 
α α

Exercise 1.26. Prove all the other three statements in Proposition 1.38. Give an
example of a family of open sets whose intersection is not open. Also give an
example of a family of closed sets whose union is not closed.

Exercise 1.27. Given any two sets U and V in C, show that:

(a) ∂(U ∪ V ) ⊂ ∂U ∪ ∂V
(b) ∂(∂U ) ⊂ ∂U
(c) U := U ∪ ∂U is always closed.

Here are two more terminologies which we will use sometimes:

• A set Ω in C is said to be bounded if there exists M > 0 such that |z| < M for any
z ∈ Ω, i.e. Ω ⊂ BM (0).
• A set Ω in C is said to be compact if it is closed and bounded.

Exercise 1.28. Use the Bolzano-Weierstrass’s Theorem for R to show the Bolzano-
Weierstrass’s Theorem for C, which asserts that if {zn }∞
n=1 is a complex sequence in
a bounded set Ω, then there exists a convergent subsequence {znk }∞ ∞
k=1 of {zn }n=1 .

Exercise 1.29. Supoose Ω1 ⊃ Ω2 ⊃ Ω3 ⊃ · · · is an infinite sequence of non-empty

compact sets in C. Show that:
\∞
Ωk 6= ∅.
k=1
[Hint: Pick zk ∈ Ωk for each k. What can you say about {zk }∞
k=1 ?]

1.3.2. Connected Sets. Intuitively, a connected set is one that is in one “piece”.
However, such a definition is not rigorous as the word “piece” is quite vague. To define
connectedness, we first need to understand what it means by a disconnected set:

Definition 1.39 (Disconnected Sets). A set Ω ⊂ C is said to be disconnected if there

exists two disjoint open sets U and V (disjoint means U ∩ V = ∅) such that:
Ω ⊂ U ∪ V, Ω ∩ U 6= ∅ and Ω ∩ V 6= ∅.

Remark 1.40. The condition Ω ⊂ U ∪ V means that U and V together cover the whole
set Ω. The condition Ω ∩ U and Ω ∩ V being non-empty means that Ω has something
inside U and something inside V . Since the definition requires U and V are disjoint (i.e.
separated in some sense), these sets U and V create a separation for the set Ω, and hence
we say Ω is disconnected (see Figure 1.5).

A set Ω is said to be connected if it is not disconnected, meaning that whenever there

are disjoint open sets U and V covering the set Ω, then at least one of Ω ∩ U or Ω ∩ V
must be empty. In practice, it is not straight-forward to verify that a set is connected using
the definition, even for simple examples such as an open ball Br (z), an open rectangle or
1.3. Point-Set Topology of C 31

Figure 1.5. Ω is the yellow set. It is disconnected with disjoint open sets U and V that
separate Ω.

an annulus 1 < |z| < 2. However, thanks for a proposition that we will state, one can
verify that they are all connected easily. Before we state the proposition, we need to
define:

Definition 1.41 (Polygonally Path-Connected Sets). A non-empty set Ω ⊂ C is said to

be polygonally path-connected if any pair of points in Ω can be joined by a continuous
path consisting of finitely many line segments contained inside Ω.

For instance, any convex set is polygonally path-connected since every pair of points
can be joined by a single line segment contained inside the set. The annulus 1 < |z| < 2
is also polygonally path-connected (see the figure below):

z1

z2

The following proposition asserts that for any open set Ω, connectedness and polygonal-
path-connectedness are equivalent:

Proposition 1.42. An open set Ω in C is connected if and only if it is polygonally path-

connected.

We omit the proof in this lecture note. Interested readers may consult Stein-
Shakarchi’s book (Exercise 5 in P.25) for an outline of the proof and try to complete the
detail as an exercise. Using this proposition, it is easy to see that any convex open sets
(and many other non-convex open sets) are connected.
The last notion about sets in C to be introduced is simply-connectedness. Readers
should have encountered this concept in Multivariable Calculus (typically in the chapter
about conservative vector field).

Definition 1.43 (Simply-Connected Sets). A set Ω is said to be simply-connected if Ω is

connected and that every closed loop in Ω can continuously contract to a point without
leaving Ω.
32 1. Preliminaries

The concept of simply-connectedness will come up frequently when we talk contour

integrals and Cauchy’s Integral Formula.
A ball and a rectangle (either open or closed) are simply-connected, while an annulus
1 < |z| < 2 is not, because the red circle in the figure below cannot shrink to a point
unless it steps into the “hole” which is not a part of the annulus.

On C, simply-connected sets have one nice property concerning simple closed curves
(“simple closed” means closed without self-intersections). If γ is a simple closed curve
contained inside a simply-connected set Ω, then the region enclosed by γ will be a subset
of Ω. Some textbooks put this as the definition of simply-connected sets in C.
Exercise 1.30. For each set described below, sketch the region on C, and determine
whether it is (i) open, (ii) closed, (iii) bounded, (iv) compact, (v) connected and
(vi) simply-connected or not.
(a) Ω1 = {z ∈ C : |z + 1| ≥ 4 |z − 1|}
(b) Ω2 = {z ∈ C : |z + 1| < 4 |z − 1|}
(c) Ω3 = {z ∈ C : |z| ≤ Re(z) + 1}
(d) Ω4 = {ez ∈ C : 1 ≤ Re(z) ≤ 2}
(e) Ω5 = {z ∈ C : z 2 − 1 ≤ 1}
 Chapter 2

Holomorphic Functions

2.1. Complex-Valued Functions

A real-valued function f : (a, b) → R with domain (a, b) maps a real number x ∈ (a, b) to
a unique real number f (x) ∈ R. To visualize such a function, we can consider its graph
y = f (x) in the xy-plane.
In this chapter and in this course, we are mostly concerned about complex-valued
functions. They are functions f : Ω → C with inputs z inside an open domain Ω ⊂ C, and
also with complex numbers f (z) as the outputs. By writing the inputs as x + yi and the
output as u + vi, then a complex-valued function f : Ω → C can be generally expressed
as:
f (x + yi) = u(x, y) + iv(x, y)
where u(x, y) and v(x, y) are real-valued functions. Essentially, both inputs and outputs
are two-dimensional, and so the graph of f is four dimensional! It is not possible for us
to visualize such a graph. In this section, we will learn how to visualize a complex-valued
function by various other ways.

2.1.1. Examples of Complex-Valued Functions. From now on, unless otherwise

is stated, we will write z = x + yi, and f (z) = u + iv.

Example 2.1. An easy example of a complex-valued function is f (z) = z 2 . The

domain of this function is C. By writing z = x + yi, we can see that:
f (z) = (x + yi)2 = x2 + 2xyi + (yi)2 = (x2 − y 2 ) + 2xyi
Therefore, we denote its real and imaginary parts by:
u(x, y) = x2 − y 2
v(x, y) = 2xy.

Example 2.2. Consider another function f (z) = ez . By writing z = x + yi, we get:

f (z) = ex+yi = ex eyi = ex (cos y + i sin y).
Therefore, u(x, y) = ex cos y and v(x, y) = ex sin y.

33
34 2. Holomorphic Functions

Exercise 2.1. For each function below, state its domain and find its real and imagi-
nary parts:
1
(a) f (z) =
z
2
(b) f (z) = |z|
(c) f (z) = e2z
iz + 1
(d) f (z) =
z−i

2.1.2. Visualizing Complex-Valued Functions using Graphs. Although one can-

not visualize the graph w = f (z) of a complex-valued function, we can separately visualize
the graphs of the real and imaginary parts of f (z).
Take f (z) = z 2 = (x2 − y 2 ) + 2xyi as an example. One can plot two separate graphs
u(x, y) = x2 − y 2 and v(x, y) = 2xy to represent this function:

The orange graph represents the real part u(x, y) = x2 − y 2 , whereas the blue graph
represents the imaginary part v(x, y) = 2xy.
Similarly, the exponential function ez = ex cos y + iex sin y can be visualized as two
separate graphs

Again the orange graph is the real part, and the blue graph is the imaginary part.
1 x yi
Some functions such as f (z) = = 2 − 2 are not defined everywhere
z x + y2 x + y2
on C. Let’s see how its graphs look:
2.1. Complex-Valued Functions 35

From the graph, one can see easily that both real and imaginary parts tend to ±∞ as
(x, y) approaches (0, 0).

2.1.3. Visualizing Complex-Valued Functions using Level Curves. Recall that

a level set of a real-valued function u(x, y) : R2 → R is a set in R2 such that u(x, y) =
constant. Different constants will give different curves or points on the plane, and a
collection of these level sets is called a level set diagram of the function.
Level set diagrams are another good way to visualize a complex-valued function.
Below are level-sets of some complex-valued functions. The orange curves are level
curves of the real part u(x, y), and the blue curves are level curves of the imaginary part.
1.0 1.0
6

5
0.5 0.5

0.0 0.0
3

-0.5 -0.5

-1.0 0 -1.0
-1.0 -0.5 0.0 0.5 1.0 0 1 2 3 4 5 6 -1.0 -0.5 0.0 0.5 1.0

(a) f (z) = z 2 (b) f (z) = ez (c) f (z) = 1/z

One can see the orange and blue level curves intersect each other orthogonally at
almost all points. It is in fact not coincident! This orthogonality phenomenon is related
to complex differentiability as we will see in the next section.

2.1.4. Visualizing Complex-Valued Functions via Mappings. One elegant way

to visualize a complex-valued function is by its mapping properties. A function f : Ω ⊂
C → C can be regarded as a map that assigns a point z in the domain Ω to a unique
point in f (z) in C.

Example 2.3. Take f (z) = z 2 = (x2 − y 2 ) + 2xyi as an example. We are going

to see how it maps a unit square in the xy-plane to the uv-plane. Consider the
straight-line L1 parametrized by (x, y) = (t, 0) where t ∈ [0, 1]. The image of L1
under the mapping f is given by:
f (t + 0i) = t2 + 0i
which is a straight-line joining 0 + 0i and 1 + 0i in the uv-plane. Similarly, consider
the straight-line L2 parametrized by (x, y) = (1, t), where t ∈ [0, 1]. The image of
36 2. Holomorphic Functions

L2 under the mapping f is:

f (1 + ti) = (1 − t2 ) + 2ti.
Consider the (u, v)-coordinates, we have u = 1 − t2 and v = 2t, which simplifies to:
v2
u=1−
4
which is a parabola in the uv-plane joining (1, 0) and (0, 2)
Likewise, one can figure out that f maps the straight-line L3 joining (0, 0) and
(1, 1) in the xy-plane to the parabola
v2
−1 u=
4
joining (0, 2) and (−1, 0). It maps the straight-line L4 joining (0, 1) and (0, 0) in the
xy-plane to the straight-line joining (−1, 0) and (0, 0) in the uv-plane.

1.0 2.0

0.8
1.5

0.6

1.0

0.4

0.2 0.5

0.0

0.0 0.2 0.4 0.6 0.8 1.0 - 1.0 - 0.5 0.5 1.0

1
Example 2.4. Consider another example f (z) = z + . We want to find the image
z
of the circle |z| = r0 under this map. Write z = r0 eiθ where 0 ≤ θ ≤ 2π, then we
have:
1
f (z) = r0 eiθ +
r0 eiθ
1
= r0 eiθ + e−iθ
r0
1
= r0 (cos θ + i sin θ) + (cos θ − i sin θ)
r
  0 
1 1
= r0 + cos θ +i r0 − sin θ,
r0 r0
| {z } | {z }
u v
which gives the following ellipse in the uv-plane (when r0 6= 1):
u2 v2
 2 +  2 = 1.
1 1
r0 + r0 r0 − r0

As r0 → 1, the image of the circle |z| = r0 degenerates to a straight-line on the

real-axis:
2.1. Complex-Valued Functions 37

3 3

2 2

1 1

0 0

-1 -1

-2 -2

-3 -3
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3

Exercise 2.2. Describe and sketch the image of the semi-infinite strip:
Σ = {z ∈ C : 0 ≤ Re(z) ≤ a and Im(z) ≥ 0}
under the map f (z) = z 2 . Here a is a positive real number.

Exercise 2.3. Describe and sketch the image of the square:

Σ = {z ∈ C : 0 ≤ Re(z) ≤ a and 0 ≤ Im(z) ≤ b}
under the map f (z) = ez . Here a and b are positive real numbers.

2.1.5. Multi-Valued “Functions”. In first year courses, we learned that a function

needs to be well-defined, meaning that one input z gives exactly one output f (z). In
Complex Analysis, we often come across the term multi-valued functions, which are
“functions” with more than one outputs for a given input. For example, the cubic root
1
f (z) = z 3 is multi-valued as discussed in Section 1.1. Given any z = reiθ with r 6= 0,
there are three possible cubic roots:
1
  13 n 1 Arg(z) 1 Arg(z)+2π 1 Arg(z)+4π
o
z 3 = |z| eiArg(z) = |z| 3 e 3 i , |z| 3 e 3 i , |z| 3 e 3 i .
1
Therefore, the map z 7→ z 3 is not rigorously a function from C to C, but is a function
with a set as the output. To visualize such a function, one can separate the graph into
different branches. Below are the graphs of the real parts:

     
1 Arg(z) 1 Arg(z)+2π 1 Arg(z)+4π
(a) z 7→ Re |z| 3 e 3 i (b) z 7→ Re |z| 3 e 3
i
(c) z 7→ Re |z| 3 e 3
i

Each branch is then a well-defined function (each input gives a unique output). If
we plot all three branches in a single graph, we obtain a beautiful surface below:
Another example of a multi-valued function is the argument map with domain C\{0}.
It is denoted and defined as:
arg(z) = {Arg(z) + 2kπ : k ∈ Z}.
38 2. Holomorphic Functions

 1
Figure 2.3. Three branches of z 7→ Re z 3 .

Recall that Arg(z) is defined to be the unique angle θ0 ∈ (−π, π] such that z = |z| eiθ0 .
An element in the set arg(z) is any angle θ such that z = |z| eiθ . For example,
n π π π π o
arg(i) = . . . , − 2π, , + 2π, + 4π, . . .
n 2 2 2 2 o
π π π π
arg(1 + i) = . . . , − 2π, , + 2π, + 4π, . . .
4 4 4 4
For each fixed k ∈ Z, we regard z 7→ Arg(z) + 2kπi as a branch of the multi-valued
function arg(z). Below is the graph of five of its branches:

The “infinite spiral” is call a helicoid, which is a minimal surface in Differential

Geometry.
2.2. Complex Differentiability 39

2.2. Complex Differentiability

In Real Analysis, we learned about continuity and differentiability of functions F (x, y) :
R2 → R2 . The function F is
psaid to be continuous at (x0 , y0 ) if for any ε > 0, there exists
δ > 0 such that whenever (x − x0 )2 + (y − y0 )2 < δ, we have:
|F (x, y) − F (x0 , y0 )| < ε.

Analogously, a complex-valued function f : C → C is said to be continuous at z0 if

for any ε > 0, there exists δ > 0 such that whenever |z − z0 | < δ, we have:
|f (z) − f (z0 )| < ε.
If we match z = x + yi with (x, y), z0 = x0 + y0 i with (x0 , y0 ), and F (x, y) with f (z), we
can see the notions of real continuity and complex continuity are essentially the same.
However, we will see that complex differentiability is very distinguished from real
differentiability (and this is why we have a separate course on Complex Analysis). In
single-variable calculus, the derivative of a function f : R → R is defined to be:
f (x + h) − f (x)
(2.1) f 0 (x) = lim .
h→0 h
If such a limit exists, then we say f 0 (x) is the derivative of f at x, and that f is said to be
(real) differentiable at x.
Now for a complex-valued function f : C → C, we define the derivative f 0 (z) in an
analogous way, except that we replace h → 0 (where h ∈ R) by w → 0 where w ∈ C:
f (z + w) − f (z)
(2.2) f 0 (z) = lim .
w→0 w
If such a limit exists, then we say f is complex differentiable at z.
Example 2.5. Let f (z) = z n where n is a positive integer. Find f 0 (z) from the
definition of complex derivative, i.e. (2.2).

Solution

(z + w)n − z n
f 0 (z) = lim
w→0 w
(z+w)n
z }| {
z n + C1n z n−1 w + C2n z n−2 w2 + . . . + Cn−1
n
zwn−1 + wn −z n
= lim
w→0 w
C1n z n−1 w + C2n z n−2 w2 + . . . + Cn−1
n
zwn−1 + wn
= lim
w→0 w

n n−1 n n−2 n
= lim C1 z + C2 z w + . . . + Cn−1 zwn−2 + wn−1
w→0
= C1n z n−1 + 0 + . . . + 0
= nz n−1 .

A natural question: How is it the different from single-variable calculus? The key
distinction is that (2.2) is a multivariable limit since w is a complex number! By writing
w = h + ki, z = x + yi and f (x + yi) = u(x, y) + iv(x, y) where u, v, x, y, h, k are real,
the limit in (2.2) can be rewritten as:
f (z+w)=f ((x+yi)+(h+ki)) f (z)
z }| { z }| {
(u(x + h, y + k) + iv(x + h, y + k)) − (u(x, y) + iv(x, y))
f 0 (x + yi) = lim .
(h,k)→(0,0) h + ki
40 2. Holomorphic Functions

Past experience in MATH 2023/3033/3043 tell us that a multivariable limit exists less
likely than a single variable limit, since we require the limit not only exist, but also equal
when (h, k) approaches (0, 0) in all possible directions in the hk-plane. Therefore, there
are many complex-valued functions, which look simple and elementary, are not complex
2
differentiable. For instance, let f (z) = |z| = x2 + y 2 , then:

f (z + w) − f (z) (x + h)2 + (y + k)2 − (x2 + y 2 )

lim = lim
w→0 w (h,k)→(0,0) h + ki
2hx + h2 + 2ky + k 2
= lim
(h,k)→(0,0) h + ki

Along the path {k = 0}, the limit goes to:

2hx + h2 + 2ky + k 2 2hx + h2

lim = lim = 2x.
h→0 h + ki h→0 h
k=0

However, along the path {h = 0}, the limit goes to:

2hx + h2 + 2ky + k 2 2ky + k 2

lim = lim = −2yi.
k→0 h + ki k→0 ki
h=0

In general, 2x 6= −2yi (unless x = y = 0) and so f 0 (z) does not exist for almost all
x + yi ∈ C. Nonetheless, f (z) = x2 + y 2 is a polynomial of x and y and so it is real
differentiable everywhere on the xy-plane R2 . From this example, we can see that complex
differentiability is much more restrictive than real differentiability!

Exercise 2.4. Show that f (z) = Re(z) is nowhere complex differentiable on C.

Exercise 2.5. Show that f (z) = 1/z is complex differentiable at any z ∈ C\{0}.

Exercise 2.6. Find all possible constants α, β ∈ C such that f (z) = αz + βz is

complex differentiable at every z ∈ C.

2.2.1. Cauchy-Riemann Equations. In this section, we will examine a necessary

condition for a function being complex differentiable. This necessary condition is called
the Cauchy-Riemann equations (or Cauchy-Riemann relations in some textbooks).

Proposition 2.1. If f (z) = u(x, y) + iv(x, y) is complex differentiable at z0 = x0 + y0 i,

then all first derivatives of u and v exist, and the following equations (which are called
Cauchy-Riemann equations hold at (x0 , y0 ):
∂u ∂v ∂v ∂u
(2.3) = =−
∂x ∂y ∂x ∂y
Furthermore, u(x, y) and v(x, y) are (real) differentiable at (x0 , y0 ).

Proof. Given that f is complex differentiable at z0 = x0 + y0 i, the following limit exists:

f ((x0 + y0 i) + (h + ki)) − f (x0 + y0 i)
(2.4) lim
(h,k)→(0,0) h + ki

and the values of the limit are equal as (h, k) approaches (0, 0) from any direction.
2.2. Complex Differentiability 41

In particular, along the path {k = 0}, the above limit equals to:
f (x0 + y0 i + h) − f (x0 + y0 i)
lim
h→0 h
u(x0 + h, y0 ) + iv(x0 + h, y0 ) − u(x0 , y0 ) − iv(x0 , y0 )
= lim
h→0 h
   
u(x0 + h, y0 ) − u(x0 , y0 ) v(x0 + h, y0 ) − v(x0 , y0 )
= lim +i
h→0 h h
∂u ∂v
= (x0 , y0 ) + i (x0 , y0 ).
∂x ∂x
Along the path {h = 0}, the limit (2.4) equals to:
f (x0 + y0 i + ki) − f (x0 + y0 i)
lim
k→0 ki
u(x0 , y0 + k) + iv(x0 , y0 + k) − u(x0 , y0 ) − iv(x0 , y0 )
= lim
k→0 ki
   
u(x0 , y0 + k) − u(x0 , y0 ) v(x0 , y0 + k) − v(x0 , y0 )
= lim +i
k→0 ki ki
 
1 ∂u ∂v
= (x0 , y0 ) + i (x0 , y0 )
i ∂y ∂y
∂v ∂u
= (x0 , y0 ) − i (x0 , y0 )
∂y ∂y

As the limits along these two paths must be equal, we get:

∂u ∂v ∂v ∂u
(x0 , y0 ) + i (x0 , y0 ) = (x0 , y0 ) − i (x0 , y0 )
∂x ∂x ∂y ∂y
which proves the desired Cauchy-Riemann equation (2.3).
Next, we prove that u and v are (real) differentiable at (x0 , y0 ). By the complex
differentiability of f at z0 , we have
f (z) = f (z0 ) + f 0 (z0 ) (z − z0 ) + (E1 (x, y) + iE2 (x, y))
where E1 and E2 are real-valued functions such that |Ei (x, y)| ≤ o(|z − z0 |) for i = 1, 2.
In the proof above (the computation along the path {k = 0}), we also have
∂u ∂v
f 0 (z0 ) = (x0 , y0 ) + i (x0 , y0 ).
∂x ∂x
Then by direct computation and the Cauchy-Riemann equation (that we have just proved),
one can check that
f (z) − f (z0 ) − f 0 (z0 ) (z − z0 )
= u(x, y) + iv(x, y) − (u(x0 , y0 ) + iv(x0 , y0 ))
| {z } | {z }
f (z) f (z0 )

− [ux (x0 , y0 ) + ivx (x0 , y0 )] [(x − x0 ) + i(y − y0 )]

| {z }| {z }
f 0 (z0 ) z−z0

= u(x, y) − u(x0 , y0 ) − ux (x0 , y0 ) (x − x0 ) + vx (x0 , y0 ) (y − y0 )

+ i (v(x, y) − v(x0 , y0 ) − vx (x0 , y0 ) (x − x0 ) − ux (x0 , y0 ) (y − y0 ))
= u(x, y) − u(x0 , y0 ) − ux (x0 , y0 ) (x − x0 ) − uy (x0 , y0 ) (y − y0 )
+ i (v(x, y) − v(x0 , y0 ) − vx (x0 , y0 ) (x − x0 ) − vy (x0 , y0 ) (y − y0 )) .
42 2. Holomorphic Functions

Therefore, we get
E1 (x, y) = u(x, y) − u(x0 , y0 ) − ux (x0 , y0 ) (x − x0 ) − uy (x0 , y0 ) (y − y0 )
E2 (x, y) = v(x, y) − v(x0 , y0 ) − vx (x0 , y0 ) (x − x0 ) − vy (x0 , y0 ) (y − y0 ).
The fact that |Ei (x, y)| ≤ o(|z − z0 |) is equivalent to u and v being (real) differentiable
at (x0 , y0 ) (simply by the definition of real differentiability).


It is notable that the converse of Proposition 2.1 is not true. If f (z) = u(x, y)+iv(x, y)
satisfies the Cauchy-Riemann equations (2.3) at (x, y), it may not be true that f is complex
differentiable at z = x + yi. Here is one counter-example:
p p
f (z) = |Re(z)Im(z)| = |xy|.
p
Then u(x, y) = |xy| and v(x, y) = 0. We claim that ∂u ∂u
∂x (0, 0) = ∂y (0, 0) = 0:

∂u u(0 + h, 0) − u(0, 0)
(0, 0) = lim
∂x h→0 h
0−0
= lim = 0.
h→0 h
Similarly, one can also show ∂u
∂y (0, 0) = 0. It is obvious that
∂v
∂x = ∂v
∂y = 0. Therefore, the
Cauchy-Riemann equation (2.3) holds at (x, y) = (0, 0).
However, the function f (z) is not complex differentiable at z = 0 + 0i. It is because
when computing the limit (2.4) at (x, y) = (0, 0), we get:
p
f ((0 + 0i) + (h + ki)) − f (0 + 0i) |hk|
lim = lim .
(h,k)→(0,0) h + ki (h,k)→(0,0) h + ki

Along the path {h = 0}, the limit equals to 0. However, along the path {h = k}, the limit
becomes:
√
h2 |h|
lim = lim
h→0 h + hi h→0 h(1 + i)
1 1
which does not exist as it approaches 1+i as h → 0+ , while it approaches − 1+i as
h → 0− .
This f (z) is one example that the Cauchy-Riemann equation holds at a point, but
the function is not complex differentiable at that point. Fortunately, if we assume further
that u(x, y) and v(x, y) are (real) differentiable functions, then the Cauchy-Riemann
equations imply complex differentiability.

Proposition 2.2. Let f : Ω → C be a complex-valued function defined on an open set

Ω ⊂ C such that u(x, y) := Ref (x + yi) and v(x, y) := Imf (x + yi) are both (real)
differentiable functions on Ω. If the Cauchy-Riemann equations (2.3) hold at (x0 , y0 ) ∈ Ω,
then f is complex differentiable at z0 = x0 + y0 i.

Proof. We define:
∂u ∂u
E1 (x, y) = u(x, y) − u(x0 , y0 ) − (x − x0 ) − (y − y0 )
∂x (x0 ,y0 ) ∂y (x0 ,y0 )
∂v ∂v
E2 (x, y) = v(x, y) − v(x0 , y0 ) − (x − x0 ) − (y − y0 ).
∂x (x0 ,y0 ) ∂y (x0 ,y0 )
2.2. Complex Differentiability 43

Since both u and v are differentiable at (x0 , y0 ), we have (by definition of differentiabil-
ity):
E1 (x, y) E2 (x, y)
lim = lim = 0.
(x,y)→(x0 ,y0 ) |z − z0 | (x,y)→(x0 ,y0 ) |z − z0 |

We first verify that

f (z) − f (z0 ) ∂u ∂v E1 (x, y) + iE2 (x, y)
= +i + .
z − z0 ∂x (x0 ,y0 ) ∂x (x0 ,y0 ) z − z0
It basically follows from the Cauchy-Riemann equations:
f (z) − f (z0 )
z − z0
u(x, y) + iv(x, y) − u(x0 , y0 ) − iv(x0 , y0 )
=
z − z0
E1 (x, y) + ux (x0 , y0 ) · (x − x0 ) + uy (x0 , y0 ) · (y − y0 )
= (by definition of E1 )
z − z0
E2 (x, y) + vx (x0 , y0 ) · (x − x0 ) + vy (x0 , y0 ) · (y − y0 )
+i (by definition of E2 )
z − z0
E1 (x, y) + ux (x0 , y0 ) · (x − x0 ) − vx (x0 , y0 ) · (y − y0 )
= (Cauchy-Riemann)
z − z0
E2 (x, y) + vx (x0 , y0 ) · (x − x0 ) + ux (x0 , y0 ) · (y − y0 )
+i (Cauchy-Riemann)
z − z0
E1 (x, y) + iE2 (x, y) ux (x0 , y0 ) · [(x − x0 ) + i(y − y0 )]
= +
z − z0 z − z0
vx (x0 , y0 ) · [(x − x0 ) + i(y − y0 )]
+i (Re-grouping)
z − z0
Note that z − z0 = (x − x0 ) + i(y − y0 ). By cancellations, we get:
f (z) − f (z0 ) E1 (x, y) + iE2 (x, y)
= + ux (x0 , y0 ) + ivx (x0 , y0 )
z − z0 z − z0
as desired.
Since
E1 (x, y) + iE2 (x, y)
lim = 0,
z→z0 z − z0
we finally get:
f (z) − f (z0 )
lim = ux (x0 , y0 ) + ivx (x0 , y0 ) = fx (x0 , y0 ).
z→z0 z − z0
Hence, f is complex differentiable at z0 . 

Remark 2.3. The condition of (real) differentiability is sometimes difficult to verify.

Fortunately, if u and v are C 1 on Ω, i.e. ∂u ∂u ∂v ∂v
∂x , ∂y , ∂x and ∂y all exist and are continuous
on Ω, then from MATH 3033/3043 we learned that u and v are then (real) differentiable
on Ω. Many textbooks state a weaker version of Proposition 2.2, namely C 1 plus Cauchy-
Riemann equations imply complex differentiability.

In short, combining Propositions 2.1 and 2.2, for any functions f = u + iv such
that u and v are (real) differentiable on an open domain Ω, complex differentiability
is equivalent to the Cauchy-Riemann equations. Many functions we will encounter are
(real) differentiable. Therefore, to show such a function is complex differentiable it
suffices to verify the Cauchy-Riemann equations. Let’s look at some examples.
44 2. Holomorphic Functions

Example 2.6. Determine all z at which the following functions are complex differ-
entiable.
(a) f (z) = z
2
(b) f (z) = |z|
(c) f (z) = ez

Solution
(a) f (z) = z = x − yi. Hence u = x and v = −y, which are clearly C 1 .
∂u ∂u
=1 =0
∂x ∂y
∂v ∂v
=0 = −1
∂x ∂y
Cauchy-Riemann equations do not hold at every point, hence f is not complex
differentiable at any point z ∈ C.
2
(b) f (z) = |z| = x2 + y 2 . Hence u = x2 + y 2 and v = 0, which are clearly C 1 .
∂u ∂u
= 2x = 2y
∂x ∂y
∂v ∂v
=0 =0
∂x ∂y
Cauchy-Riemann equations hold if and only if (x, y) = (0, 0). Therefore, f is
complex differentiable at z = 0 only.
(c) f (z) = ez = ex cos y + iex sin y. Hence u = ex cos y and v = ex sin y, which are
C 1 functions
∂u ∂u
= ex cos y = −ex sin y
∂x ∂y
∂v ∂v
= ex sin y = ex cos y
∂x ∂y
which are all continuous. Clearly the Cauchy-Riemann equations hold at every
(x, y), hence by Proposition 2.2, the function f (z) = ez is complex differentiable
at every z ∈ C.

From now on, we will use a more professional term to describe complex differentiable
functions:

Definition 2.4 (Holomorphic Functions). Let Ω be an open set of C. A complex-valued

function f is said to be holomorphic on Ω if f is complex differentiable at every z ∈ Ω.
A function which is holomorphic on C is said to be an entire function.

Remark 2.5. It is a custom to say a function is holomorphic on an open domain while

saying a function is complex differentiable at a point. Therefore, whenever we say
holomorphic on a set Ω, the set Ω is assumed to be open (and non-empty). We seldom
say a function is holomorphic on the real-axis in C, since the real-axis is not an open
set. Instead, we should say the function is complex differentiable at every point on the
real-axis.
2.2. Complex Differentiability 45

Example 2.7. Prove the following facts about holomorphic functions:

(a) Consider a complex-valued function f (z) = u(x, y) + iv(x, y) : C → C, and its
conjugate function f (z) := u(x, y) − iv(x, y). Suppose both f (z) and f (z) are
holomorphic on C, show that f must be a constant function.
(b) Consider a complex-valued function g(z) = u(x, y) + iv(x, y) : C → C. Define
g : C → C by h(z) := g(z) for any z ∈ C. Show that if g is holomorphic on C,
then h is also holomorphic on C.

Solution
(a) Given that f is holomorphic on C, the Cauchy-Riemann equations show:
∂f ∂f
=i .
∂y ∂x
Similarly, f is also holomorphic on C, we have:
∂f ∂f ∂f ∂f ∂f ∂f
=i =⇒ =i =⇒ = −i .
∂y ∂x ∂y ∂x ∂y ∂x
Combining both results, we have:
∂f ∂f ∂f
i = −i =⇒ = 0.
∂x ∂x ∂x
Since f 0 (z) = ∂f 0
∂x for holomorphic functions, we have f (z) = 0 for any z ∈ C.
Therefore, f (z) is a constant function.
(b) Denote h(z) = U (x, y) + iV (x, y). Since h(z) = g(z) = u(x, −y) − iv(x, −y), we
get:
U (x, y) = u(x, −y) and V (x, y) = −v(x, −y).
By Proposition 2.1, g is holomorphic on C implies u and v are real differentiable on
R2 , and hence U and V are real differentiable on R2 as well. Hence, according to
Proposition 2.2, it suffices to show h satisfies the Cauchy-Riemann equations:

∂h ∂ ∂ ∂ ∂
(z) = U (x, y) + i V (x, y) = u(x, −y) + i (−v(x, −y))
∂x ∂x ∂x ∂x ∂x
∂u ∂v ∂u ∂u
= (x, −y) − i (x, −y) = (x, −y) + i (x, −y)
∂x ∂x ∂x ∂y
∂h ∂ ∂ ∂ ∂
(z) = U (x, y) + i V (x, y) = u(x, −y) − i v(x, −y)
∂y ∂y ∂y ∂y ∂y
∂u ∂v ∂u ∂u
= (x, −y) + i (x, −y) = − (x, −y) + i (x, −y)
∂y ∂y ∂y ∂x
 
∂u ∂u ∂h
=i (x, −y) + i (x, −y) = i (z).
∂x ∂y ∂x
Therefore, h satisfies the Cauchy-Riemann equations and that U and V are real
differentiable on R2 , by Proposition 2.2, h is holomorphic on C.

Exercise 2.7. Determine all z’s in the complex plane at which the following functions
are complex differentiable:
(a) f (z) = 1/z
2
(b) f (z) = z |z|
(c) f (z) = z 2
46 2. Holomorphic Functions

Exercise 2.8. Consider the function:

( 4
e−1/z 6 0
if z =
f (z) =
0 if z = 0
Show that the Cauchy-Riemann equations hold everywhere on C, but f is not
complex differentiable at z = 0.

Exercise 2.9. Suppose f (z) is complex differentiable at every z ∈ C. Prove that any
one of the following conditions imply that f is constant:
(i) Re(f ) is constant.
(ii) Im(f ) is constant.
(iii) |f | is constant.

Exercise 2.10. Find a function f : C → C which is complex differentiable every-

where in C such that
Re(f (x + yi)) = y 3 − 2x2 y.

Exercise 2.11. Show that the Cauchy-Riemann equations (2.3) for the function
f = u + iv is equivalent to:
∂f ∂f
=i .
∂y ∂x
Give a geometric interpretation of this result.

Exercise 2.12. Show that any holomorphic function f = u + iv : Ω → C satisfies

s
∂f ∂f ∂(u, v)
|∇u| = |∇v| = = = |f 0 (z)| = det
∂x ∂y ∂(x, y)
on the domain Ω.

Exercise 2.13. Fix a complex number w such that |w| < 1. Consider the map
f : B1 (0) → C defined by:
w−z
f (z) = .
1 − wz
Show that:
(a) f is well-defined on B1 (0);
(b) f holomorphic on B1 (0);
(c) The image of f is B1 (0);
(d) f is bijective as a map f : B1 (0) → B1 (0).

Exercise 2.14. Let f : C → C be the function defined by:

(
z 2 if Re(z) ∈ Q and Im(z) ∈ Q
f (z) =
0 otherwise
Show that f is complex differentiable at z = 0, but it is not holomorphic on any
open set containing 0.
2.2. Complex Differentiability 47

Example 2.8. Find the largest open subset Ω ⊂ C such that the function:
f (x + yi) = x2 − y 2 + 2xyi
is holomorphic on Ω.

Solution
First we divide the complex plane C into open regions:
U = {x + yi : x2 − y 2 > 0} and V = {x + yi : x2 − y 2 < 0}.
Im

Re

On U , we have f (z) = (x2 − y 2 ) + 2xyi which is clearly C 1 on U . It clearly

satisfies the Cauchy-Riemann equations on U (straight-forward computations, left
as an exercise). Hence f is holomorphic on U .
On the other hand, we have f (z) = (y 2 − x2 ) + 2xyi on V , which does not
satisfy the Cauchy-Riemann equations (straight-forward computations). Therefore,
f is not complex differentiable at any z0 ∈ V .
The largest open set Ω on which f is holomorphic is U , since any open set which
is larger than Ω must contain some point in V .

Exercise 2.15. Find the largest open subset Ω ⊂ C such that the function:
f (x + yi) = |x| + |y| i
is holomorphic on Ω.

2.2.2. Geometric Interpretation of Cauchy-Riemann Equations. Recall from

MATH 2023 that ∇u(a, b), where u(x, y) is a real-valued functions of (x, y), is a normal
vector to the level curve of u at (a, b) whenever ∇u(a, b) 6= 0.
Now consider a complex-valued function f = u + iv. If f is holomorphic on its
domain Ω, then the Cauchy-Riemann equations, i.e. ux = vy and uy = −vx hold on Ω.
As result, we get easily see that:
   
∂u ∂u ∂v ∂v ∂u ∂v ∂u ∂v ∂u ∂u ∂u ∂u
∇u · ∇v = i+ j · i+ j = + =− + = 0.
∂x ∂y ∂x ∂y ∂x ∂x ∂y ∂y ∂x ∂y ∂y ∂x

Therefore, ∇u and ∇v are perpendicular to each other provided that they are both non-
zero. Geometrically speaking, it means that the level sets of u and v are perpendicular!
There are more we can say about ∇u and ∇v. Define the matrix:
 
0 −1
J= .
1 0
48 2. Holomorphic Functions

Given any vector x ∈ R2 , the product J(x) is the vector obtained by rotating x counter-
clockwisely by π2 . The Cauchy-Riemann equations can be rewritten as:
      
vx −uy 0 −1 ux
= = .
vy ux 1 0 uy

As a result, we have ∇v = J(∇u). It means that for a holomorphic function f , the vector
∇v can be obtained by rotating ∇u counter-clockwisely by π2 .

2.2.3. Conformal Mappings. Another important geometric significance of holo-

morphic functions is that it preserves angles. Let γ1 (t) = x1 (t) + iy1 (t) and γ2 (t) =
x2 (t) + iy2 (t) be two C 1 curves in the complex plane, and assume that they intersect
at t = 0, i.e. γ1 (0) = γ2 (0) =: z0 , then the angle between the curves at the point z0 is
measured by the angle between the tangent vectors γ10 (0) and γ20 (0).
Now consider a map f : C → C. The images of the curves γ1 and γ2 under the map
f are the curves f ◦ γ1 (t) and f ◦ γ2 (t), and hence that angle between them at the point
f (z0 ) is measured by the angle between vectors (f ◦ γ1 )0 (0) and (f ◦ γ2 )0 (0). We will
show that if f is complex differentiable at z0 and that f 0 (z0 ) 6= 0, then the angle between
γ10 (0) and γ20 (0), is the same as that between (f ◦ γ1 )0 (0) and (f ◦ γ2 )0 (0).
First, we leave the following elementary fact as an exercise:

Exercise 2.16. Let z1 , z2 , w1 , w2 ∈ C such that z2 6= 0 and w2 6= 0. Suppose we have

z1 w1
= . Show that the angle between z1 and z2 , is the same as that between w1
z2 w2
and w2 .

Proposition 2.6. Let γk (t) = xk (t) + iyk (t) where k = 1, 2 be two C 1 curves in C which
intersect at t = 0 at the point z0 ∈ C. Suppose f : Bε (z0 ) → C is complex differentiable at
z0 and f 0 (z0 ) 6= 0 then we have:
(f ◦ γ1 )0 (0) γ10 (0)
= .
(f ◦ γ2 )0 (0) γ20 (0)
As a result, the angle between γ1 and γ2 at z0 is preserved under f . See Figure 2.4.

Proof. Let f = u + iv (as usual). Recall from Exercise 2.11 that the Cauchy-Riemann
equation is equivalent to
∂f ∂f
=i .
∂y ∂x
Using this and the chain rule, we have:
d ∂f dx1 ∂f dy1
(f ◦ γ1 ) = +
dt ∂x dt ∂y dt
∂f ∂f
(f ◦ γ1 )0 (0) = x0 (0) + i y10 (0)
∂x z0 1 ∂x z0
∂f
= γ10 (0).
∂x z0

Likewise, we have:
∂f
(f ◦ γ2 )0 (0) = γ20 (0)
∂x z0
2.2. Complex Differentiability 49

∂f ∂f
By Exercise 2.12, |f 0 (z0 )| = (z0 ) . Given f 0 (z0 ) 6= 0, we also have (z0 ) 6= 0 and
∂x ∂x
hence:
(f ◦ γ1 )0 (0) fx (z0 ) γ10 (0) γ 0 (0)
0
= 0 = 10
(f ◦ γ2 ) (0) fx (z0 ) γ2 (0) γ2 (0)
as desired. 

Figure 2.4. A holomorphic map preserves angles

50 2. Holomorphic Functions

2.3. Logarithmic and Trigonometric Functions

In the real world, the exponential function x 7→ ex is injective, and hence it makes perfect
sense to define its inverse function, which is known as the logarithm x 7→ log x.
However, in the complex world, the exponential function is no longer injective. In
particular, ez+2kπi = ez for any integer k. Therefore, z 7→ ez is an infinity-to-one map,
so just like the n-th root function z 7→ z 1/n and the argument function arg, the complex
logarithm is multi-valued.
Another type of functions that is conceptually different from the real world is trigono-
metric functions such as sin, cos and tan. In the real world, sin θ and cos θ are defined
in a geometric way using a right-angled triangle of angle θ. However, if z is a complex
number, it doesn’t make any sense to say a triangle with a “complex” angle z, and so the
complex trigonometric functions such as sin z, cos z and tan z are defined in a different
way.
In this section, we will study the logarithmic and trigonometric functions in detail.

2.3.1. Logarithmic Functions. As discussed before, the complex exponential func-

tion f (z) = ez is not injective, since for any integer k, we have:

ez+2kπi = ez e2kπi = ez (cos 2kπ + i sin 2kπ) = ez (1 + 0i) = ez .

As such, the complex logarithm is multi-valued and so for each z 6= 0, log z is a set
rather than a single number.

Definition 2.7 (Complex Logarithm). Given any z ∈ C and z 6= 0, we define the

complex logarithm to be the set:
log z := {w ∈ C : ew = z}.

Example 2.9. If w is a complex number such that ew = 1, by writing w = u + vi,

then we have:
1 = eu+iv = eu eiv = eu (cos v + i sin v) = eu cos v + ieu sin v.
The only possible solutions for (u, v) are u = 0 and v = 2kπ where k ∈ Z. Therefore,
we get:
log 1 = {w ∈ C : ew = 1} = {2kπi : k ∈ Z} = {· · · , −4πi, −2πi, 0, 2πi, 4πi, · · · }.

Let’s derive a general formula for log z. Given any z 6= 0, we first express it in polar
form z = |z| eiArg(z) . Then, w = u + iv satisfies ew = z if and only if eu+iv = |z| eiArg(z) :

eu eiv = |z| eiArg(z) ⇐⇒ u = ln |z| and v = Arg(z) + 2kπ where k ∈ Z.

As a result, we have w = ln |z| + i(Arg(z) + 2kπ) for any k ∈ Z, and using set notations:

log z = {ln |z| + i(Arg(z) + 2kπ) : k ∈ Z}.

Remark 2.8. To avoid confusion, from now on we will denote ln as the real, single-valued
logarithm, and use log for the complex multi-valued logarithm.

Recall that arg(z) = {Arg(z) + 2kπ : k ∈ Z}, we can also write:

log z = ln |z| + i arg(z)

2.3. Logarithmic and Trigonometric Functions 51

for any z 6= 0. For example:

log(1) = ln |1| + i arg(1) = {0 + i(Arg(1) +2kπ) : k ∈ Z} = {2kπi : k ∈ Z}
| {z }
0
log(2i) = ln |2i| + i arg(2) = {ln 2 + i(Arg(2i) +2kπ) : k ∈ Z}
| {z }
π
2
n π  o
= ln 2 + + 2kπ i : k ∈ Z
2
log(−1) = ln |−1| + i arg(−1) = {0 + i(Arg(−1) +2kπ) : k ∈ Z}
| {z }
π
= {(2k + 1)πi : k ∈ Z}
Remark 2.9. Recall that for real logarithms, ln(a) is undefined if a < 0. However, for
complex logarithms, log(a) is defined even for a < 0 since ez can be a negative real
number. On the other hand, both log 0 and ln 0 are undefined, since ez 6= 0 for any z ∈ C.

Recall that for the argument function arg(z), we define the principal argument Arg(z)
as the unique angle θ ∈ (−π, π] such that z = |z| eiθ . Similarly, we define the principal
logarithm to be:

Definition 2.10 (Principal Logarithm). For any z 6= 0, we define its principal logarithm
to be:
Log(z) := ln |z| + iArg(z).

For example, we have:

Log(1) = 0
π
Log(2i) = ln 2 + i
2
Log(−1) = πi

Exercise 2.17. Find log(z) and Log(z) for each of the following z’s:
(a) z = −2
√
(b) z = 1 + 3i
(c) z = −i

Exercise 2.18. Give an example of z1 and z2 such that

Log(z1 z2 ) 6= Log(z1 ) + Log(z2 ).

Exercise 2.19. For any two subsets S and T of C, we define the sum of the two sets
to be:
S + T := {s + t : s ∈ S and t ∈ T }.
Show that for any z1 , z2 6= 0, we have:
arg(z1 z2 ) = arg(z1 ) + arg(z2 )
log(z1 z2 ) = log(z1 ) + log(z2 ).
52 2. Holomorphic Functions

Exercise 2.20. Let z 6= 0, and n ∈ N.

(a) Show that:
 
1/n 1 Arg(z) + 2(pn + k)π
log(z ) = ln |z| + i : p ∈ Z, k ∈ {0, 1, . . . , n − 1} .
n n
1
(b) Hence, show that log(z 1/n ) = log z.
n

2.3.2. Complex Derivatives of Logarithms. Recall that if f (z) is holomorphic,

∂f
then from the proof of Proposition 2.1, we have f 0 (z) = . We will use this fact to
z
∂x
compute the complex derivatives of both e and log(z).
For the exponential function f (z) = ez , we have already shown in Example 2.6 that
it is holomorphic on C. Its complex derivative is given by:
d z ∂f
(2.5) e =
dz ∂x
∂ x
= (e cos y + iex sin y)
∂x
= ex cos y + iex sin y
= ez .

d
We next compute log(z). Note that log(z) is multi-valued. Differentiating log(z)
dz
basically means differentiating each branch of log(z) individually. We first verify that
every branch of log(z) is holomorphic on C\{x + 0i : x ≤ 0}:
log(z) = {ln |z| + i(Arg(z) + 2kπ) : k ∈ Z}
1
Let u(x, y) = ln |z| = ln(x2 + y 2 ) and v(x, y) = Arg(z) + 2kπ. We leave it as an exercise
2
for readers to verify that on C\(−∞, 0]:
∂u x ∂u y
= 2 = 2
∂x x + y2 ∂y x + y2
∂v y ∂v x
=− 2 = 2 .
∂x x + y2 ∂y x + y2
Clearly, they are all continuous and satisfy the Cauchy-Riemann equation. By Proposition
2.2, each branch ln |z| + i(Arg(z) + 2kπ) is holomorphic on C\{x + 0i : x ≤ 0}, and so:
d ∂u ∂v x − yi z 1
(ln |z| + i(Arg(z) + 2kπ)) = +i = 2 = = .
dz ∂x ∂x x + y2 zz z
Therefore, every branch of log(z) has the same complex derivative, so we may simply
write:
d 1
(2.6) log(z) = .
dz z

2.3.3. Complex Powers. For any z, w ∈ C where z 6= 0, we define z to the power

of w as follows:
(2.7) z w := ew log z .
For example, we have
π π
ii = ei log(i) = ei(ln|i|+i arg(i)) = {ei(i( 2 +2kπ)) : k ∈ Z} = {e− 2 −2kπ : k ∈ Z}.
Maybe to your surprise, ii is real-valued!
2.3. Logarithmic and Trigonometric Functions 53

Using (2.7), one can recover the n-th root formula stated in Definition 1.9. For any
complex number a 6= 0, according to (2.7), we have:
1 1
a n = e n log a
n 1 o
= e n (ln|a|+(Arg(a)+2kπ)i) : k ∈ Z
n 1 Arg(a)+2kπ
o
= e n ln|a| e n i
:k∈Z
   
p
n Arg(a) + 2kπ Arg(a) + 2kπ
= |a| cos + i sin :k∈Z
n n
   
p
n Arg(a) + 2kπ Arg(a) + 2kπ
= |a| cos + i sin : k = 0, 1, 2, . . . , n − 1
n n

Using the chain rule, one can derive the differentiation formula for complex powers.
Regard w as a fixed complex numbers, we can derive:
d w d w log z d
(2.8) z = e = ew log z w log z
dz dz dz
w
= z w · = wz w−1
z
d z d z log w d
(2.9) w = e = ez log w z log w
dz dz dz
= wz log w

Exercise 2.21. Find the following complex powers:

(a) (1 + i)1−i
(b) 21−i
(c) 33i

Exercise 2.22. Compute the complex derivative of each function below:

(a) f (z) = (1 − z)1/2
(b) f (z) = (1 − z 2 )1/3

Exercise 2.23. Fix a non-zero complex number w. On what domains are the
following functions holomorphic?
(a) f (z) = wz
(b) g(z) = z w

For complex logarithms, we use log z and Log z to distinguish the multi-valued
logarithm, or the principal branch of logarithm. Similarly, for a complex power z w , which
is defined via logarithms, we can also define its principal branch by ewLog(z) . For instance,
π
the principal branch of ii is e− 2 . Unlike logarithms and arguments, we do not introduce
a new symbol to denote the principal branch of z w . With abuse of notations, we may
π π
sometimes simply write, for instance ii = e− 2 , to mean e− 2 is the principal value of ii .
It may cause ambiguity, but such an ambiguity will cause less nuisance if we state clearly
in the context whether z w means a multi-valued power or the principal branch.

2.3.4. Trigonometric Functions. From Euler’s formula (1.3), we saw that:

eix = cos x + i sin x and e−ix = cos x − i sin x
54 2. Holomorphic Functions

for any x ∈ R. Hence by writing sin x and cos x in terms of the exponentials, we get:

eix − e−ix eix + e−ix

sin x = cos x = .
2i 2

When z is a complex number, it does not make sense to define sin z and cos z by
regarding z as an “angle”. Thanks for the above identities, we define the complex sin and
cos functions as:

Definition 2.11 (Complex Sine and Cosine). For any z ∈ C, we define:

eiz − e−iz eiz + e−iz
sin z = cos z = .
2i 2

Using the chain rule, we get:

d iz d −iz
e = ieiz e = −ie−iz ,
dz dz
and so it is easy to verify that both sin z and cos z are entire functions and their derivatives
are:
d d
(2.10) sin z = cos z cos z = − sin z.
dz dz

The other trigonometric functions are defined similarly as in the real case:

sin z 1
tan z = sec z =
cos z cos z
cos z 1
cot z = csc z =
sin z sin z
Using the product and quotient rules, one can easily derive that:

d d
(2.11) tan z = sec2 z sec z = sec z tan z
dz dz
d d
(2.12) cot z = − csc2 z csc z = − csc z cot z
dz dz

Exercise 2.24. Prove (2.10), (2.11) and (2.12).

Exercise 2.25. Prove that for any z, z1 , z2 ∈ C:

(a) sin(z1 + z2 ) = sin z1 cos z2 + cos z1 sin z2
(b) cos(z1 + z2 ) = cos z1 cos z2 − sin z1 sin z2
(c) sin2 z + cos2 z = 1
(d) 1 + tan2 z = sec2 z
(e) 1 + cot2 z = csc2 z
(f) sin z = sin z
(g) cos z = cos z
(h) sin(−z) = − sin z
(i) cos(−z) = cos z
2.3. Logarithmic and Trigonometric Functions 55

2.3.5. Inverse Trigonometric Functions. It is well-known that trigonometric func-

tions are not injective even in the real case. The real inverse sine sin−1 x is defined
to be the number θ ∈ (−π/2, π/2] such that x = sin θ, making sin−1 a well-defined,
single-valued function of x.
In the complex world, we accept multi-valued functions. Therefore, given any z ∈ C,
we regard sin−1 z to be:

sin−1 z := {w ∈ C : sin w = z}.

It is possible to rewrite sin−1 z using complex logarithms. Suppose sin w = z, let’s

try to solve for w in terms of z. By definition of sin z, we have:
eiw − e−iw
= z.
2i
By rearrangement, we get:

eiw − e−iw − 2iz = 0

eiw eiw − e−iw − 2iz = 0

2

eiw − 2iz eiw − 1 = 0

which can be regarded as a quadratic equation of eiw . Solving it, we get:

1/2
iw 2iz + (−2iz)2 − 4(1)(−1)
1/2
e = = iz + 1 − z 2 .
2

1/2
1/2
Note that 1 − z 2 is multi-valued and it is not necessary to write iz ± 1 − z 2 .
From this, we get:

1/2  
1/2 
iw = log iz + 1 − z 2 =⇒ w = −i log iz + 1 − z 2 .

As a result, we have:

1/2 
sin−1 z = −i log iz + 1 − z 2 .

For example,

sin−1 (1) = −i log(i)

 √ 
sin−1 (i) = −i log(−1 + 21/2 ) = −i log −1 ± 2

Exercise 2.26. Show that:


1/2  i i+z
cos−1 z = −i log z + i 1 − z 2 tan−1 z = log
2 i−z

Exercise 2.27. Derive the following differentiation rules:

d 1
sin−1 z =
dz (1 − z 2 )1/2
d 1
cos−1 z = −
dz (1 − z 2 )1/2
d 1
tan−1 z =
dz 1 + z2
56 2. Holomorphic Functions

2.3.6. Mapping Properties. Recall that given any z = x+yi, we have ez = ex eiy =
e cos y + iex sin y. Therefore, along the straight-path z = x0 + yi (where x0 is fixed and
x

y varies), the image under the map z 7→ ez is given by (u, v) = ex0 (cos y + i sin y), which
is a circle with radius ex0 centered at the origin. In other words, the complex exponential
maps vertical lines in the xy-plane to concentric circles in the uv-plane. On the other hand,
along the horizontal path z = x + y0 i, the image is given by (u, v) = ex (cos y0 + i sin y0 ),
which is a half-line from the origin (but not passing through it).

1.0

2.0

0.8

1.5
0.6

0.4 1.0

0.2
0.5

0.0

0.0 0.2 0.4 0.6 0.8 1.0 1.0 1.5 2.0 2.5

Exercise 2.28. Show that the complex sine function f (z) = sin z maps horizontal
lines in the xy-plane to ellipses in the uv-plane, and maps vertical lines in the
xy-plane to hyperbolas in the uv-plane.
 Chapter 3

Contour Integrals

We start discussing complex integrations in this chapter. Given a function f : Ω ⊂ C → C

and a C 1 curve γ in the domain of f , the contour integral of f over γ is denoted by:
Z
f (z) dz.
γ

We will learn how they are defined and how they can be computed soon. In the first
glance, it appears quite similar to line integrals in Multivariable Calculus. However,
when combining with properties of holomorphic functions, there are many beautiful
and amazing results concerning complex contour integrals which did not appear in line
integrals. One notable result is Cauchy’s integral formula, an elegant theorem which
leads to many important results in Complex Analysis and beyond.

3.1. Complex Integrations

3.1.1. Contour Integrals. Consider a C 1 curve γ in C parametrized by:
z(t) = x(t) + iy(t), t ∈ [a, b].
The differential dz is regarded as:
dz
dz = dt = (x0 (t) + iy 0 (t)) dt.
dt
For example, if γ is the unit circle centered at the origin, then it is parametrized by:
z(t) = cos t + i sin t = eit , t ∈ [0, 2π].
it
d(e )
Hence, we have dz = dt = ieit dt.
dt

Definition 3.1 (Contour Integrals). Let f : Ω → C be a continuous function on the

open domain Ω ⊂ C, and γ be a C 1 curve in Ω. Suppose γ is parametrized by
z(t) = x(t) + iy(t), t ∈ [a, b],
then the contour integral of f over γ is denoted and defined by:
Z Z b
f (z) dz := f (z(t)) z 0 (t) dt .
γ a | {z }
dz

57
58 3. Contour Integrals

Remark 3.2. If γ is a piecewise C 1 curve, meaning that it can be decomposed into

γ = γ1 + . . . + γk where each of γ1 , . . . , γk is C 1 , and that the whole curve γ is continuous,
then we define:
Z Z Z
f (z) dz = f (z) dz + . . . + f (z) dz.
γ γ1 γk

Furthermore, if γ is closed, we usually denote the contour integral by:

I
f (z) dz.
γ

I
Example 3.1. Compute the line integral f (z) dz for each of the functions below.
γ
Here γ is the circle with radius 2 centered at the origin.
(a) f (z) = z 2
1
(b) f (z) =
z
(c) f (z) = z

Solution
γ can be parametrized by:
z(t) = 2eit , t ∈ [0, 2π].
Therefore dz = 2ieit dt.
(a)
I Z 2π Z 2π
2


it 2 it
z dz = 2e · 2ie dt = 8ie3it dt
γ 0 | {z } | {z } 0
dz
z2
 t=2π
1 3it
= 8i e
3i t=0
8 6πi
8
= e − e0 = (1 − 1) = 0.
3 3
(b)
I Z 2π Z t=2π
1 1
dz = · 2ieit dt = i dt = 2πi.
γ z 0 2eit t=0
(c)
I Z 2π Z 2π
−it it
z dz = 2e · 2e dt = 4 dt = 8πi.
γ 0 0

d 1 3it


Remark 3.3. In part (a) of the above example, we have used the fact that dt 3i e =
e3it , and also Fundamental Theorem of Calculus. In general, just like in the real case,
if F (t) is a differentiable function of t on [a, b] such that F 0 (t) = ϕ(t) on [a, b], then we
have
Z t=b
ϕ(t) dt = F (b) − F (a).
t=a
3.1. Complex Integrations 59

However, we sometimes need to be more careful when applying this. Try to find out
what’s wrong with the calculation below:
I Z 2π  2π
1 1 it 1 it
dz = ie dt = − Log(1 − 2e )
|z|=1 1 − 2z 0 1 − 2eit 2 0
1
= − (Log(−1) − Log(−1)) = 0???
2

Example 3.2. Consider the line segment L from a point z1 to a point z2 in C.

Compute the following contour integral (in terms of z1 and z2 ):
Z
ez dz.
L

Solution
First we parametrize L:
z(t) = (1 − t)z1 + tz2 , t ∈ [0, 1].
Then, we have dz = (z2 − z1 ) dt, and so:
Z Z 1
z
e dz = ez1 +t(z2 −z1 ) · (z2 − z1 ) dt
L 0
 1
1
= ez1 +t(z2 −z1 ) · (z2 − z1 )
z2 − z1 0
= ez2 − ez1 .

3
Γ2
L1

1 Γ1

x
1 L2 3

Figure 3.1. the path in Example 3.1

Exercise 3.1. Compute the contour integrals

I I I
1
2
dz, z dz and |z| dz
γ z γ γ
where γ = Γ1 + L2 + Γ2 + L1 is the curve in Figure 3.1.

3.1.2. Primitive Functions. In Calculus I, we learned that if F 0 (x) = f (x) on x ∈

[a, b], then:
Z b
f (x) dx = F (b) − F (a).
a
This is the celebrated Fundamental Theorem of Calculus. In Complex Analysis, we have
an analogous result:
60 3. Contour Integrals

Theorem 3.4. Let f : Ω → C be a continuous function defined on an open domain

Ω ⊂ C, and γ be a piecewise C 1 curve in Ω with starting point z1 and ending point z2 .
If F : Ω → C is a (single-valued) holomorphic function on Ω such that F 0 (z) = f (z) for
every z ∈ Ω, then we have:
Z
f (z) dz = F (z2 ) − F (z1 ).
γ

Proof. First assume that γ is C 1 . Suppose the path γ can be parametrized by:
z(t) = x(t) + iy(t), t ∈ [a, b].
0
Then, we have dz = z (t) dt, and hence:
Z Z b
f (z) dz = f (z(t)) · z 0 (t) dt
γ a
Z b
= F 0 (z(t)) ·z 0 (t) dt
a | {z }
f (z(t))
Z b
d
= F (z(t)) dt (chain rule)
a dt
= F (z(b)) − F (z(a))
= F (z2 ) − F (z1 ).
If γ is only piecewise C 1 , we can decompose γ = γ1 + · · · + γk so that each γi is C 1 .
Then, one can argue as above for each γi , and finally obtain the desired result by adding
a telescope sum. 

Remark 3.5. If such an F (z) in Theorem 3.4 exists, then we call F (z) a primitive function
of f (z).

The above theorem is particularly useful when the anti-derivative of f is easy to find.
For example, if γ is any continuous piecewise C 1 path from z1 to z2 , we can find easily
that:
Z  3 z2
z z 3 − z13
2
z dz = = 2
γ 3 z1 3
Z
z
ez dz = [ez ]z21 = ez2 − ez1 .
γ

In particular, if C is a closed path, then we have:

I I
2
z dz = 0 and ez dz = 0.
C C

Exercise 3.2. Let γ1 be the path which starts from (0, 0), first to (1, 1), then to (0, 2).
Let γ2 be the path which starts from (0, 0), then straight to (0, 2). Verify the following
by direct computations:
Z Z
πz πz
cos dz = cos dz.
γ1 2 γ2 2
Then, verify that Theorem 3.4 gives the same result.

However, it is important to note that Theorem 3.4 requires the curve γ to be inside
Ω (on which F 0 (z) = f (z) holds). Let’s consider the function f (z) = z1 . Although we
3.1. Complex Integrations 61

d
usually simply write dz Log(z) = z1 , it is only true for z ∈ C\{x + 0i : x ≤ 0} since Log(z)
is not continuous on the negative x-axis.
Therefore, we can only apply Theorem 3.4 when the curve γ lies inside Ω :=
C\{x + 0i : x ≤ 0}. For instance, we still have
I
1
=0
γ1 z

where γ1 is the unit circle centered at 2 + 0i with radius 1. This closed curve γ1 is
contained inside Ω.
H
However, it is incorrect to claim γ2 z1 = 0 where γ2 is the unit circle centered at the
origin. The reason is that this closed curve passes through the negative x-axis (hence not
contained inside Ω). In fact we can directly verify that:
I
1
= 2πi.
γ2 z
y

γ2

γ1

Fortunately, we can still apply Theorem 3.4 on f (z) = z12 when the integration curve
γ does not pass through the origin. The reason is that F (z) = − z1 is a primitive function
for f such that F 0 (z) = f (z) holds on C\{0}. Therefore, we have:
I
1
2
=0
γ z
for any closed curve γ not passing through the origin. Also, for a path L in C\{0}
connecting z1 to z2 , we have:
Z  z
1 1 2 1 1
2
dz = − = − .
L z z z1 z1 z2

Exercise 3.3. Consider the path γ parametrized by:

z(t) = cos3033 t + i sin2033 t, where t ∈ [0, π].
Z Z
1
Find the contour integrals 1014
dz and (1 + iz)1013 dz.
γ z γ

Z
Exercise 3.4. Evaluate the integral |z| dz where γ is each of the following:
γ

(a) a line segment joining −i to i.

(b) a counter-clockwise semi-circular path joining −i to i
Does it exist an entire function F : C → C such that F 0 (z) = |z| for any z ∈ C? Why
or why not?
62 3. Contour Integrals

Exercise 3.5. First verify that on an appropriate domain, we have:

d 2
i (Log(i + z) − Log(i − z)) = .
dz 1 + z2
Using this, show that:
I
1
dz = 0 when r < 1.
|z|=r 1 + z2
In your solution, explain clearly where the condition r < 1 is needed.

3.1.3. Integral Estimates. Estimation of a contour integral is an important tech-

nique in Complex Analysis. It will appear in many parts of the course. If we know an
upper bound for |f (z)| on the curve γ, and
Z the upper bound for the length of γ, then we
are able to bound the contour integral f (z) dz without calculating it.
γ

Lemma 3.6. Let f : Ω → C be defined on an open domain Ω. Suppse γ is a curve in Ω such

that:
• |f (z)| ≤ M for any z ∈ γ, and
• the arc-length of γ is bounded above by L.
Then, we have:
Z
f (z) dz ≤ M L.
γ

Proof. There is a nice trick in the proof that readers are recommended to learn. Let
Z
I= f (z) dz.
γ

Express I in polar form: I = |I| eiθ , then we have e−iθ I = |I| which is real! Suppose γ is
parametrized by z(t) = x(t) + iy(t) where a ≤ t ≤ b, then:
Z Z
−iθ −iθ
e I=e f (z) dz = e−iθ f (z) dz
γ γ
Z b 


= Re e−iθ f (z) + iIm e−iθ f (z) (x0 (t) + iy 0 (t)) dt
a
Z b 


= Re e−iθ f (z) x0 (t) − Im e−iθ f (z) y 0 (t) dt.
a

The last equality above follows from the fact that e−iθ I is real.
Then, we use Cauchy-Schwarz’s inequality to bound the integrand:



Re e−iθ f (z) x0 (t) − Im e−iθ f (z) y 0 (t)
q q
2 2 2 2
≤ (Re (e−iθ f (z))) + (Im (e−iθ f (z))) (x0 (t)) + (y 0 (t))
= e−iθ f (z) |z 0 (t)| = |f (z)| |z 0 (t)| ≤ M |z 0 (t)| .

Finally, we get:
Z b
e−iθ I ≤ M |z 0 (t)| = M L,
a
and hence |I| ≤ M L, completing the proof. 
3.1. Complex Integrations 63

Z
Remark 3.7. If we estimate the integral f (z) dz in a more direct way by writing
γ
f = u + iv and then consider the following:
Z Z Z b
f (z) dz = (u + iv)(dx + idy) = (ux0 − vy 0 ) + i(vx0 + uy 0 ) dt
γ γ a
v !2 !2
u Z Z
u b b
= t 0 0
(ux − vy ) dt + (vx0 + uy 0 ) dt .
a a

then after applying Cauchy-Schwarz’s inequality to each integral, the best we can achieve
is Z √
f (z) dz ≤ 2M L,
γ
which is weaker than the result in Lemma 3.6.

Example 3.3. Find an upper bound for the contour integral:

I
1
e z dz .
|z|=1

Solution
For any z ∈ C such that |z| = 1, we have:
x y
1 −i x2 +y
e z = e x2 +y2 2
= ex−iy = ex e−iy ,
1
e z = ex ≤ e1 = e.
Here we have used the fact that −1 ≤ x ≤ 1 along the curve |z| = 1.
Therefore, by Lemma 3.6, we have:
I
1
e z dz ≤ |{z}
2π |{z}
e .
|z|=1
L M

Example 3.4. Show that:

I
1
lim dz = 0.
R→+∞ |z|=R (z − 1)2

Solution
We are interested in the limit when R → +∞, so we can assume R > 1 so that
the contour circle |z| = R does not pass through 1 (at which the integrand is
undefined).
On the contour |z| = R, we have |z − 1| ≥ R − 1 (draw a diagram to convince
yourself on that), so we have:
1 1 1
(z − 1)2
= 2 ≤ (R − 1)2 on |z| = R.
|z − 1|
| {z }
M
64 3. Contour Integrals

The length of the contour |z| = R is 2πR. Hence, by Lemma 3.6, we get
I
1 1
2
dz ≤ 2πR · .
|z|=R (z − 1) (R − 1)2
2πR
From elementary calculus, we have lim = 0, and the desired result
R→+∞ (R − 1)2
follows from the squeeze theorem.

Example 3.5. Show that:

I
Log(z)
lim dz = 0.
R→+∞ |z|=R z2

Solution
For |z| = R  1, we have
Log(z) ln(R) + iArg(z) 1 1
= ≤ 2 (|ln(R)| + |Arg(z)|) ≤ 2 (ln(R) + π).
z2 R2 R R
Using ML inequality
I
Log(z) Log(z) 2π
2
dz ≤ 2
· 2πR ≤ (ln(R) + π) → 0 as R → +∞.
|z|=R z z R
Here we have used the l’Hospital’s rule. We conclude:
I
Log(z)
lim dz = 0.
R→+∞ |z|=R z2

Exercise 3.6. Let f : C → C be a continuous function, and consider a fixed point

α ∈ C. Show that:
I
f (z) 2πR
dz ≤ max |f (z)| when R > |α| .
|z|=R z − α R − |α| |z|=R

Exercise 3.7. Suppose f : C → C is a continuous function such that:

|f (z)|
lim sup = 0.
R→+∞ |z|≥R R
Show that: I
f (z)
lim dz = 0.
R→+∞ |z|=R z2

Exercise 3.8. Let f : C → R be a continuous real-valued function such that |f (z)| ≤ 1

for any z ∈ C. Show that:
I
f (z) dz ≤ 4.
|z|=1
I
[Hint: Define I = f (z) dz, then write I = |I| eiθ . ]
|z|=1
3.2. Cauchy-Goursat’s Theorem 65

3.2. Cauchy-Goursat’s Theorem

In this section, we will prove a very fundamental theorem in Complex Analysis, the
Cauchy-Goursat’s Theorem, which asserts that if f : Ω → IC is a holomorphic function on
a simply-connected domain Ω, then the contour integral f (z) dz must be zero for any
γ
closed curve γ in Ω. The statement of the theorem sounds simple, but the proof is quite
delicate. We will discuss the proof of this theorem in detail.
Cauchy-Goursat’s Theorem is fundamental because it is used to prove the Cauchy’s
integral formula,
I which provides a very elegant way for computing contour integral of
f (z)
the form dz and leading many exciting results. We will see later in the course
γ z−α
that the Cauchy integral formula is the heart of complex analysis.

Theorem 3.8 (Cauchy-Goursat’s Theorem). Let Ω ⊂ C be a simply-connected open

domain, γ be any closed piecewise C 1 curve in Ω, and f : Ω → C be any holomorphic
function defined on Ω, then we have:
I
f (z) dz = 0.
γ

Using Cauchy-Goursat’s Theorem, we can immediately conclude that all the integrals
below over any closed curve γ ∈ C are zero, without performing any calculation:
I I I
ez dz, sin z dz, z 2 dz, etc.
γ γ γ

Both conditions of Ω being simply-connected and f being holomorphic on Ω are

essential. If Ω is not simply-connected, say Ω = C\{0}, Cauchy-Goursat’s Theorem does
not hold. Here is a quick counter-example:
I
1
dz = 2πi 6= 0.
|z|=1 z
Moreover, the holomorphic condition on f is also necessary, and here is a counter-
example: I
z dz = 2πi 6= 0.
|z|=1
We will prove this theorem soon. The proof consists of several steps:
Step 1: First prove a special case when the contour γ is a triangle (while Ω is any
simply-connected open domain);
Step 2: Then prove a special case when Ω is convex (while γ is any closed piecewise
C 1 contour).
Step 3: Use results from previous steps to deduce the general case: Ω is any simply-
connected open domain, and γ is any closed piecewise C 1 contour.

3.2.1. Step 1: Cauchy-Goursat’s Theorem for Triangle Contours. Let’s begin

by assuming that T is a triangle contour in Ω. We bisect each side of the triangle T to
(1) (1) (1) (1)
create four smaller triangles T1 , T2 , T3 and T4 as shown in the Figure 3.2.
By cancellations of common sides, we have:
I X4 I
f (z) dz = f (z) dz.
(1)
T j=1 Tj
66 3. Contour Integrals

1. Goursat’s theorem 35

(1)
T3

T (1)
T2
(1) (1)
T1 T4

(0)
Figure Figure
3.2. Divide1.
theBisection of 4Ttriangles
contour T into

Triangle inequality then shows:

for otherwise (2) would
I be contradicted.
4 I
We choose a triangle that
satisfies this inequality,f (z) X
anddzrename (1)
. Observe
it T f (z) that if d(1) and
(1)
≤ (1)
dz . (1)
p denote the diameter T and perimeter j=1 Tof
(1)
j
T , respectively, then d =
(0) (1) (0)
(1/2)d and p = (1/2)p . We now repeat this process for the trian-
Letgle T (1)
T (1) be, the
bisecting
triangleit among
into four (1)
all smaller
Tj ’s (where triangles.j = 1,Continuing
2, 3, 4) withthis
the process,
largest value
I we obtain a sequence of triangles
of f (z) dz , then one has:
Tj
(1)
T (0) , T (1) , . . . , T (n) , . . .
I I
with the properties that f (z) dz ≤ 4 f (z) dz
!" T ! T!(1)
" !
! ! ! !
! f (z) dz !
(1) ! ≤ 4 !
n!
(z) dz !!
f(1)
Repeat the above procedure ! on T : sub-divide T into four congruent triangles
T (0) T (n) I
(2)
Tj (whereand j = 1, 2, 3, 4), and pick the one with the largest value of f (z) dz and
(2)
Tj
(2)
label it as T . Then, one has: d(n) = 2−n d(0) , p(n) = 2−n p(0)
I I I I
where d(n) and p(n) denote the diameter and perimeter2 of T (n) , respec-
f (z) dz ≤ 4 f (z) dz
(n) =⇒ f (z) dz ≤ 4 f (z) dz .
tively.
T (1) We also denoteT (2) by T the solid Tclosed triangle with
T (2) boundary
T (n) , and observe that our construction yields a sequence of nested com-
Continuing
pact setsthis process, we obtain a sequence of triangles:
(0) (1) (2)
T (3)
T (0)T ⊃ ,TT(1) ,⊃T · · ·, ⊃ T ,(n)
. . .⊃ · · ·
(wherewhose
we denote T (0) :=
diameter T )to
goes such
0. that
By Proposition 1.4 in Chapter 1, there exists
I I
a unique point z0 that belongs to all the solid triangles T (n) . Since f is
(3.1) f (z) dz ≤ 4n f (z) dz for any n ≥ 0.
holomorphic Tat(0)z0 we can write T (n)
Denote ∆(j) to be thef (z)
closed (z0 ) + f ′ (z
= ftriangular region enclosed
0 )(z − by T (j)
z0 ) + ψ(z)(z −. zThen,
0 ) , we have:
(0)
where ψ(z) → 0 as z →∆ ⊃ ∆(1)the
z0 . Since ∆(2) ⊃ . . .f (z0 ) and the linear func-
⊃ constant
′
tion f1.29,
By Exercise (z0 )(z − zis0 )athave
there leastprimitives,
one point zwe can integrate the above (n) equality
0 contained inside all of ∆ .
using Corollary 3.3 in the previous chapter, I and obtain
Our goal is to bound"the RHS term 4"n f (z) dz of (3.1), so as to show that
I T (n)
(3) f (z) dz = ψ(z)(z − z0 ) dz.
f (z) dz is arbitrarilyTsmall,
(n) concluding
T (n)that it must be zero. To achieve our goal,
T (0)
we recall that f is holomorphic on Ω, and in particular, it is complex differentiable
at z0 (which is a point in all of ∆(n) ’s). By considering the derivative f 0 (z0 ), and by
rearrangement:
f (z) − f (z0 ) f (z) − f (z0 ) − f 0 (z0 )(z − z0 )
f 0 (z0 ) = lim =⇒ lim = 0.
z→z0 z − z0 z→z0 z − z0
3.2. Cauchy-Goursat’s Theorem 67

For simplicity, denote the numerator by E(z) := f (z) − f (z0 ) − f 0 (z0 )(z − z0 ), then we
have:
E(z)
(3.2) lim = 0.
z→z0 z − z0
0
Since the function f (z0 ) + f 0 (z0 )(z − z0 ) has a primitive function zf (z0 ) + f (z0 )
2 (z − z0 )2
(note that z0 is a fixed point), we have

I I I
E(z) dz = [f (z) − f (z0 ) − f 0 (z0 )(z − z0 )] dz = f (z) dz.
T (n) T (n) T (n)

Therefore, to bound the RHS of (3.1), we can consider the integral of E(z) instead,
which is very small according to (3.2).
Now, given any ε > 0, by (3.2), there exists δ > 0 such that whenever z ∈ Bδ (z0 ),
E(z)
we have < ε. Recall that {∆(n) }∞ n=0 is a strictly decreasing sequence of triangles
z − z0
“converging” to the point z0 . Hence, for sufficiently large n, ∆(n) must lie inside the ball
Bδ (z0 ), and so |E(z)| < ε |z − z0 | for any z ∈ ∆(n) ⊂ Bδ (z0 ).
Recall that |z − z0 | is the distance between z and z0 , both of which are in ∆(n) . By
elementary geometry, the distance between any two points in a triangle must be bounded
by the perimeter of the triangle. Hence, we have for any z ∈ ∆(n) ,
εL0
(3.3) |E(z)| < ε |z − z0 | ≤ εLn =
2n
where Ln denotes the perimeter of the triangle T (n) .
Using (3.3), we can apply Lemma 3.6 to show:
I
εL0 εL2
E(z) dz ≤ n · Ln = n0 .
T (n) 2 4
Finally, by considering (3.1), we have proved:
I I I
n n εL20
f (z) dz ≤ 4 f (z) dz = 4 E(z) dz ≤ 4n · = εL20 .
T T (n) T (n) 4n
Since ε > 0 is arbitrary, by letting ε → 0+ , we get:
I
f (z) dz = 0,
T

completing Step 1.

Exercise 3.9. Using the result proved so far, show that Cauchy-Goursat’s Theorem
holds for any closed polygon γ.

Exercise 3.10. Show that if 4ABC is contained inside a simply-connected open set
Ω on which f is holomorphic, then we have:
Z Z Z
f (z) dz = f (z) dz + f (z) dz.
L(A,C) L(A,B) L(B,C)

Here L(A, B), for instance, is the straight path from A to B.

Exercise 3.11. Which part in the proof of Step 1 will break down if f is not
holomorphic? Also, why will the proof break down if Ω is not simply-connected?
68 3. Contour Integrals

3.2.2. Step 2: Cauchy-Goursat’s Theorem for Convex Domains. Now we are

given any closed piecewise C 1 curve γ (not necessarily a triangle) in an open convex
domain Ω. We want to show that if f : Ω → C is holomorphic, then
I
f (z) dz = 0.
γ

We show that by finding a primitive function F : Ω → C such that F 0 (z) = f (z) on

Ω, then this step is proved using Theorem 3.4. To define such a function F , we first fix
a point z0 ∈ Ω, and denote L(z0 , z) to be the straight path from z0 to z. Note that by
convexity of Ω, such a path must be contained in Ω. Next, we define:
Z
F (z) := f (ξ) dξ.
L(z0 ,z)

F (z + w) − F (z)
We claim that F 0 (z) = f (z) by showing that the quotient tends to f (z)
w
as w → 0.
z+w
z

z0

From Step 1 (note that z0 , z and z + w form a triangle), we know that:

Z Z !
F (z + w) − F (z) 1
= f (ξ) dξ − f (ξ) dξ
w w L(z0 ,z+w) L(z0 ,z)
Z
1
= f (ξ) dξ.
w L(z,z+w)
Z
ξ=z+w
By observing that f (z) dξ = [f (z) ξ]ξ=z = w f (z), we have:
L(z,z+w)
Z Z
F (z + w) − F (z) 1 1
(3.4) = f (ξ) dξ = (f (ξ) − f (z)) + f (z) dξ
w w L(z,z+w) w L(z,z+w)
Z
1
= (f (ξ) − f (z)) dξ + f (z).
w L(z,z+w)
Z
1
The next task will be to show that (f (ξ) − f (z)) dξ tends to 0 as w → 0.
w L(z,z+w)
For any ε > 0, by the continuity of f , there exists δ > 0 such that whenever ξ ∈ Bδ (z),
we have |f (ξ) − f (z)| < ε. In particular, if |w| < δ, then the path L(z, z + w) ⊂ Bδ (z),
and so for any ξ ∈ L(z, z + w), we have:
|f (ξ) − f (z)| < ε.
3.2. Cauchy-Goursat’s Theorem 69

Z
Applying Lemma 3.6 on the integral (f (ξ) − f (z)) dξ, we have:
L(z,z+w)
Z
(f (ξ) − f (z)) dξ ≤ ε · |w| ,
L(z,z+w) |{z}
length of contour
Z
1
which implies (f (ξ) − f (z)) dξ ≤ ε (whenever 0 < |w| < δ), or equivalently,
w L(z,z+w)
Z
1
lim (f (ξ) − f (z)) dξ = 0.
w→0 w L(z,z+w)

Finally, from (3.4), we conclude:

F (z + w) − F (z)
lim = f (z) =⇒ F 0 (z) = f (z).
w→0 w
This shows f (z) has a primitive function on Ω, and hence
I
f (z) dz = 0
γ

for any closed curve γ in Ω, completing Step 2.

Remark 3.9. It is worthwhile to note that the whole argument in Step 2 remains valid
as long as f is continuous on Ω, and that
I
f (z) dz = 0
T
for any triangle T in the domain Ω. These two conditions are enough to prove, using the
same argument, that F 0 (z) = f (z) on Ω, even if we don’t assume f is holomorphic. This
observation will be important in the proof of Morera’s Theorem in later section.

Exercise 3.12. Discuss: In the above proof, we require Ω to be convex so that

L(z0 , z) is contained in Ω. Now suppose Ω is not convex, but is polygonally path-
connected, and we define F as:
Z
F (z) = f (ξ) dξ
γ(z0 ,z)

where γ(z0 , z) is any polygonal path from z0 to z. Can we still claim that F 0 (z) = f (z)
with the same proof? If not, where does the proof break down?

3.2.3. Step 3: Completion of the Proof. We have by far proved that Cauchy-
Goursat’s Theorem holds when at least one of the conditions holds:
(i) γ is a closed polygon; or
(ii) Ω is convex.
Now we deduce the general case based on these special cases.
Given any simply-connected domain Ω and any closed piecewise
I C 1 curve γ ⊂ Ω,
and a holomorphic function f : Ω → C, the key idea to show f (z) dz = 0 is to break
γ
the region enclosed by γ into small rectangles {Rj }N M
j=1 and “partial rectangles” {γk }k=1
(see Figure 3.3). By breaking the region into small enough of these rectangles and partial
rectangles, we may assume that these partial rectangles are contained inside an convex
subset of Ω. This is intuitively true, but the proof involves some deep knowledge on
analysis and topology beyond the scope of this course. Interested readers may read
70 3. Contour Integrals

Stein’s book Appendix B. For most of our applications, we will deal with closed curves
whose enclosed regions are as simple as rectangles and balls, and the union of them.

Figure 3.3

For each rectangle Rj and partial rectangle γk , results from Steps 1 and 2 show
I I
f (z) dz = f (z) dz = 0.
Rj γk

Note that by cancellation of common sides, we can see:

I XI XI
f (z) dz = f (z) dz + f (z) dz = 0.
γ j Rj k γk

It completes the proof of Cauchy-Goursat’s Theorem.

It could be to many people’s surprise that complex analysis in fact has real applications
such as evaluating a complicated real integral which cannot be easily done by finding
anti-derivatives.
Example 3.6. Show that Z ∞
1 − cos x
dx = π
−∞ x2
1 − eiz
by considering the contour below and the function f (z) = .
z2
y

CR

−Cε
x
−R −ε ε R

The contour CR is the counter-clockwise semi-circle centered at the origin with

radius R, and −Cε is the clockwise semi-circle centered at the origin with radius ε.
Denote γR,ε to be the closed contour:
γR,ε = CR + [−R, −ε] − Cε + [ε, R].
3.2. Cauchy-Goursat’s Theorem 71

Solution
iz
The function 1−ez2 is holomorphic on (for instance) {z ∈ C : |z| > 0, Arg(z) 6= 3π 2 }
which is a simply connected open domain containing γR,ε . Using Cauchy-Goursat’s
Theorem: I
1 − eiz
dz = 0 ∀0 < ε < R
γR,ε z2
Therefore, we have:
Z Z Z Z
1 − eiz 1 − eiz 1 − eiz 1 − eiz
2
dz − 2
dz + 2
dz + dz = 0.
CR z Cε z [−R,−ε] z [ε,R] z2
We will then take R → +∞ and ε → 0+ .
We first estimate the integral over CR . For all z ∈ CR , we have Im(z) ≥ 0 =⇒
Re(iz) = −Im(z) ≤ 0 =⇒ eiz = eRe(iz) ≤ 1. Therefore, the numerator satisfies
1 − eiz ≤ 1 + eiz ≤ 2. Using ML inequality,
Z
1 − eiz 2 4π
2
dz ≤ 2 · 2πR = → 0 as R → +∞.
CR z R R
This shows Z
1 − eiz
lim dz = 0.
R→+∞ CR z2
For the integral over Cε , we study the behavior near z = 0. By Taylor’s
expansion of eiz , we have
∞
1 − eiz 1
1 − (1 + iz + 2! (iz)2 + · · · ) i X (iz)n
= = − − .
z2 z2 z n=2 z 2 n!
Since
X∞ X∞ n−2 X∞
(iz)n |z| 1
2 n!
≤ ≤ ≤ e,
n=2
z n=2
n! n=2
n!
we have
1 − eiz i
+ < e.
z2 z
Using ML inequality,
Z Z
1 − eiz i + 1 − eiz i
2
+ dz ≤ e · π → 0 as ε → 0 =⇒ lim 2
+ dz = 0
Cε z z ε→0 +
Cε z z
Using the parametrization zε (t) = εeit where t ∈ [0, π], we have:
Z Z π Z π
i i
lim+ dz = lim+ zε0 (t) dt = lim+ ie−it ieit dt = −π
ε→0 Cε z ε→0 0 zε (t) ε→0 0
Combine the above results
Z Z Z 
1 − eiz 1 − eiz i i
lim dz = lim+ + dz − dz = 0 − (−π) = π.
ε→0+ Cε z2 ε→0 Cε z2 z Cε z

Combining the previous result and by letting R → +∞ and ε → 0+ :

Z 0 Z +∞
1 − eix 1 − eix
0−π+ 2
dx + dx = 0
−∞ x 0 x2
By comparing the real part of both side:
Z ∞
1 − cos x
dx = π.
−∞ x2
72 3. Contour Integrals

3.3. Cauchy’s Integral Formula I

Cauchy-Goursat’s Theorem requires that the function f involved is defined and holo-
morphic in the region enclosed by the closed curve γ. When the integrand has some
“singularities” such as f (z) = z1 , Cauchy-Goursat’s Theorem may not hold.
Consider the closed curves γ1 and γ2 shown below:
y

γ2

γ1

For γ1 , there is no issue to apply Cauchy-Goursat’s Theorem by taking Ω to be the

green region, and it shows
I
1
dz = 0
γ1 z

since z1 is holomorphic on the green region. However, we cannot do the same for γ2 . Any
simply-connected region containing γ2 must contain 0 at which z1 is undefined. In this
section,
I we will introduce Cauchy’s integral formula to deal with contour integrals of the
f (z)
form dz.
γ z−α

Theorem 3.10 (Cauchy’s Integral Formula). Let f : Ω → C be a holomorphic function

defined on a simply-connected domain Ω, and γ be a simple closed curve in Ω. Then, we
have: I (
1 f (z) f (α) if γ encloses α
dz =
2πi γ z − α 0 if γ does not enclose α

For instance, given an entire function f : C → C, a point α, and two closed curves γ1
and γ2 below. Cauchy’s Integral Formula asserts that:
I I
f (z) f (z)
dz = 0 whereas dz = 2πif (α).
γ1 z − α γ2 z −α
γ2

γ1

It is a very powerful theorem as it tells us that the evaluation of some contour

integrals can be done by just substituting a point into the numerator function. Let’s first
see some examples, and then we will prove the theorem.
3.3. Cauchy’s Integral Formula I 73

3.3.1. Elementary Examples. We first illustrate the use of Cauchy’s integral for-
mula by a toy example:
I I (
1 1 2πi · 1 = 2πi if γ encloses 0
dz = dz =
γ z γ z−0 0 if γ does not enclose 0
Here we take f (z) = 1 which is an entire function on C.
Example 3.7. Evaluate the following contour integrals:
I
z
(a) dz
|z|=2 (z + 3i)(z − i)
I
z
(b) dz
|z|=4 (z + 3i)(z − i)
I
ez
(c) 2
dz
|z|=2 z + 1

Solution
(a) The integrand has two singularities: z = −3i and z = i. First observe that
the curve |z| = 2 enclose i only, and hence near the simply-connected region
z
|z| ≤ 2, the function f (z) := z+3i is holomorphic. Apply Cauchy’s integral
formula with this f , we get:
I I z
z z+3i z
dz = dz = 2πi ·
|z|=2 (z + 3i)(z − i) |z|=2 z − i z + 3i z=i
i πi
= 2πi · = .
i + 3i 2
(b) Note that the curve |z| = 4 enclose both singularities −3i and i of the integrand.
We cannot apply Cauchy’s integral formula by writing the integrand as either:
z z
z+3i z−i
or .
z−i z + 3i
The way out is to do partial fractions for the denominator. Let A and B be
complex numbers such that:
1 A B
= + .
(z + 3i)(z − i) z + 3i z − i
We need to solve for A and B:
1 A(z − i) + B(z + 3i)
=
(z + 3i)(z − i) (z + 3i)(z − i)
1 = (A + B)z + (−Ai + 3Bi)
Equating coefficients, we need A + B = 0 and (−Ai + 3Bi) = 1. Solving these
equations, we get A = 41 i and B = − 41 i, and hence:
1 1
1 i i
= 4 − 4 .
(z + 3i)(z − i) z + 3i z − i
74 3. Contour Integrals

Now applying Cauchy’s integral formula:

I I 1 1
z 4 zi zi
dz = − 4 dz
|z|=4 (z + 3i)(z − i) |z|=4 z + 3i z − i
    !
1 1
= 2πi zi − zi
4 z=−3i 4 z=i
 
1 1
= 2πi · (−3i)i − i2 = 2πi.
4 4
(c) The integrand has z 2 + 1 as the denominator. Be careful that it can be zero in
ez
the complex world and so 2 is NOT holomorphic everywhere. By partial
z +1
fractions, we get:
 
1 1 1 1 1
= = − .
z2 + 1 (z − i)(z + i) 2i z − i z + i
Hence, Cauchy’s integral formula shows:
I I  z 
ez 1 e ez
2
dz = − dz
|z|=2 z + 1 2i |z|=2 z − i z + i
1

= · 2πi · ei − e−i
2i
= π ((cos 1 + i sin 1) − (cos 1 − i sin 1))
= 2πi sin 1.

Exercise 3.13. Use Cauchy’s integral formula to evaluate the following contour
integrals:
I
1
(a) 2+i
dz
|z|=2 z
I
1
(b) 2
dz
|z−eπi/4 |=1 z + i
I
1
(c) 3−1
dz
|z|=2 z
Try to do the problems in a rather tedious way using partial fractions. We will provide
another approach soon.

Exercise 3.14. Let f : Ω → C be a holomorphic function defined on a domain Ω

containing Br (α). Prove the following Mean-Value Identity:
Z 2π
1
f (α) = f (α + reiθ ) dθ.
2π 0

The result in Exercise 3.14 implies an important property about holomorphic func-
tions:

Proposition 3.11 (Maximum Principle, a.k.a. Maximum Module Theorem). If f is

holomorphic on a connected open domain Ω such that |f (z)| achieves an interior maximum
at some z0 ∈ Ω, i.e. |f (z)| ≤ |f (z0 )| for any z ∈ Ω, then f is constant on Ω.

We first prove the following lemma:

3.3. Cauchy’s Integral Formula I 75

Lemma 3.12. If f is holomorphic on an open ball BR (α) such that |f (z)| ≤ |f (α)| for
any z ∈ BR (α), then f is constant on BR (α).

Proof. By Exercise 3.14, we have for any r < R:

Z 2π Z 2π
1 iθ 1
|f (α)| ≤ f (α + re ) dθ ≤ |f (α)| dθ = |f (α)| .
2π 0 2π 0
Therefore, the inequality above is in fact an equality. This implies:
Z 2π Z 2π
1 1

0 = |f (α)| − f (α + reiθ ) dθ = |f (α)| − f (α + reiθ ) dθ.
2π 0 2π 0
Since it is given that f (α + reiθ ) ≤ |f (α)| for any α + reiθ ∈ BR (α), the integrand is
non-negative and we must have
|f (α)| = f (α + reiθ )
for any r < R and θ ∈ R. It implies |f (z)| is constant which equals |f (α)| on BR (α). By
Exercise 2.9(c), f (z) itself must be a constant function on BR (α). 

Now we use the above lemma to prove the Maximum Principle:

Proof of Proposition 3.11. Consider the set S := {z ∈ Ω : |f (z)| = |f (z0 )|}. By conti-
nuity of |f (z)|, the set S is obviously closed. We will argue that S is also open. Then, by
connectedness of Ω, we must have S = Ω (note that z0 ∈ S so S is non-emtpy).
Take any α ∈ S, so that |f (α)| = |f (z0 )|. Since Ω is open, there exists δ > 0 such
that Bδ (α) ⊂ Ω. Now it is given that |f (z0 )| is the maximum value of |f (z)| among all
z ∈ Ω. In particular, we have |f (α)| = |f (z0 )| ≥ |f (z)| for any z ∈ Bδ (α). Using Lemma
3.12, we conclude that f is constant on Bδ (α), which implies |f (z)| = |f (α)| = |f (z0 )|
for any z ∈ Bδ (α). In other words, Bδ (α) ⊂ S. This shows S is open. 

3.3.2. Proof of Cauchy’s Integral Formula. The proof of Cauchy’s integral for-
mula is a reminiscence of the proof of generalized (i.e. with holes) Green’s Theorem in
Multivariable Calculus. Fix α ∈ C and consider a simple closed curve γ enclosing α. We
want to find out the value of the integral:
I
f (z)
dz.
γ z −α
We drill a circular hole near α in the region enclosed by γ, so that the following “key-hole”
contour Γε is produced.

−∂Bε (α)
ε
α

L −L
76 3. Contour Integrals

The contour Γε = γ + L − ∂Bε (α) − L encloses a simply-connected region on which

f (z)
is holomorphic (since z 6= α in this key-hole region). Therefore, we have:
z−α
I I I I I
f (z) f (z) f (z) f (z) f (z)
0= dz = dz + − dz −
Γε z − α γ z − α L z − α |z−α|=ε z − α L z −α
| {z }
orientation!
I I
f (z) f (z)
= dz − dz.
γ z−α |z−α|=ε z − α
I I
f (z) f (z)
Therefore, we have dz = dz for any sufficiently small ε > 0.
γ z − α |z−α|=ε z −α
To prove the desired result, we try to figure out the contour integral over the circle
|z − α| = ε. The key trick is to write f (z) = f (z) − f (α) + f (α), so that:
I I  
f (z) f (z) − f (α) f (α)
(3.5) dz = + dz
|z−α|=ε z − α |z−α|=ε z−α z−α
I I
f (z) − f (α) 1
= dz + f (α) dz
|z−α|=ε z − α |z−α|=ε z − α
The second integral can be computed directly by parametrizing the circle: z = α + εeit ,
where t ∈ [0, 2π]:
I Z 2π
1 1
dz = it
· εieit dt
|z−α|=ε z − α 0 εe
Z 2π
= i dt = 2πi.
0
For the first term, we claim that it tends to 0 as ε → 0+ : since f is complex differentiable
at z = α, and so
f (z) − f (α)
lim = f 0 (α).
z→α z−α
By definition of limit, there exists δ > 0 such that whenever z ∈ Bδ (α) we have:
f (z) − f (α)
− f 0 (α) < 1,
z−α
and hence
f (z) − f (α)
< 1 + |f 0 (α)| =: M.
z−α
As a result, when ε < δ, the contour |z − α| = ε lies completely inside the ball Bδ (α),
then by Lemma 3.6, we have:
I
f (z) − f (α)
≤ M · 2πε → 0 as ε → 0+ .
|z−α|=ε z−α
Finally, from (3.5), we have:
I
f (z)
lim dz = 2πif (α).
ε→0+ |z−α|=ε z−α
I I
f (z) f (z)
Recall that dz = dz for any sufficiently small ε > 0, so we
γ z−α |z−α|=ε z − α
have: I I
f (z) f (z)
dz = lim dz = 2πif (α),
γ z − α ε→0 +
|z−α|=ε z −α
completing the proof of Cauchy’s integral formula.
3.3. Cauchy’s Integral Formula I 77

3.3.3. Cauchy’s Integral Formula with Multiple Holes. We have seen how to
1
apply Cauchy’s integral formula on fractions such as 2 which is not defined on z = i
z +1
and z = −i. If a simple closed contour γ enclosesboth singularities,
 then we performed
1 1 1
partial fractions so that the fraction becomes − .
2i z − i z + i
Sometimes, partial fractions can be time-consuming especially when there are many
singularities. However, using the hole-drilling technique demonstrated in the proof of
Cauchy’s integral formula, we can break down the contour integral into a sum of several
contour integrals, each of which is over a contour that encloses only one singularity. Let’s
look at some examples.
Example 3.8. Evaluate the contour integral:
I
z
dz
|z|=4 (z + 3i)(z − i)
without using partial fractions.

Solution
The two singularities are z = −3i and z = i, both are contained inside the contour
|z| = 4. Draw two little circles with small radii ε around each singularity and
consider the key-hole contour:
Γ = γ1 + L1 − ∂Bε (−3i) − L1 + γ2 + L2 − ∂Bε (i) − L2

−L2 L2

ε
i
γ1 γ2

ε
−3i
L1 −L1

Then, the key hole contour Γ encloses a simply-connected region not containing
any singularity of the integrand. Therefore, Cauchy-Goursat’s Theorem asserts that
I
z
dz = 0.
Γ (z + 3i)(z − i)
On the other hand, by cancellation of the common sides, we have:
I Z Z I I I I I
= + − − = − − .
Γ γ1 γ2 |z+3i|=ε |z−i|=ε |z|=4 |z+3i|=ε |z−i|=ε
78 3. Contour Integrals

Therefore,
I
z
0= dz
(z + 3i)(z − i)
IΓ I
z z
= dz − dz
|z|=4 (z + 3i)(z − i) |z+3i|=ε (z + 3i)(z − i)
I
z
− dz.
|z−i|=ε (z + 3i)(z − i)

Therefore, we can break the required integral into the sum of two integrals:
I I I
z z z
dz = dz + dz
|z|=4 (z + 3i)(z − i) |z+3i|=ε (z + 3i)(z − i) |z−i|=ε (z + 3i)(z − i)
z
Since ε is very small, the function is holomorphic on |z + 2i| < ε, and so
z−i
Cauchy’s integral formula asserts that:
I I z
z z−i −3i 3πi
dz = dz = 2πi · = .
|z+3i|=ε (z + 3i)(z − i) |z+3i|=ε z − (−3i) −3i − i 2
For the second integral, we have:
I I z
z z+3i i πi
dz = dz = 2πi · =
|z−i|=ε (z + 3i)(z − i) |z−i|=ε z − i i + 3i 2
Adding up the results, we get:
I
z 3πi πi
dz = + = 2πi.
|z|=4 (z + 3i)(z − i) 2 2

Example 3.9. Evaluate the contour integral:

I
1
3−1
dz
|z|=2 z
without using partial fractions.

Solution
First factorize the integrand:
1 1
3
=
z −1 (z − 1)(z − ω)(z − ω 2 )
2πi
where ω := e 3 is the cubic root of unity. There are three singularities, namely 1, ω
and ω 2 , all are enclosed by the given contour |z| = 2. By mimicking the hole-drilling
3.3. Cauchy’s Integral Formula I 79

argument, one can arrive at:

I
1
2
dz
|z|=2 (z − 1)(z − ω)(z − ω )
I I
1 1
= 2)
dz + dz
|z−1|=ε (z − 1)(z − ω)(z − ω |z−ω|=ε (z − 1)(z − ω)(z − ω 2 )
I
1
+ dz
2
|z−ω |=ε (z − 1)(z − ω)(z − ω 2 )
I 1 I 1 I 1
(z−ω)(z−ω 2 ) (z−1)(z−ω 2 ) (z−1)(z−ω)
= dz + dz + dz
|z−1|=ε z−1 |z−ω|=ε z−ω |z−ω 2 |=ε z − ω2
 
1 1 1
= 2πi + +
(1 − ω)(1 − ω 2 ) (ω − 1)(ω − ω 2 ) (ω 2 − 1)(ω 2 − ω)
ω − (1 + ω) + 1
= 2πi ·
ω(1 − ω)2 (1 + ω)
= 0.

Exercise 3.15. Evaluate the following contour integrals:

I
1
(a) 3
dz
|z|=24601 z + 1
I
1
(b) 2 + 1)(z 2 + 9)
dz
|z|=2 (z
I
ez
(c) 4
dz
|z−1|=1 z + 1
I
z
(d) dz
|z|=4 1 − ez

Exercise 3.16. Let n be a positive integer, and ω := e2πi/n denote the n-th root of
unity. Express the contour integral:
I
1
n−1
dz
|z|=2 z
in terms of ω.

Exercise 3.17. Given any real constant a ∈ R, by considering the contour integral
I
eaz
dz, prove the following integration formula:
|z|=1 z
Z π
ea cos θ cos(a sin θ) dθ = π.
0
80 3. Contour Integrals

Exercise 3.18. Suppose γ is a simple-closed curve, and f is a holomorphic function

on C\{α1 , · · · , αk }, where α1 , · · · , αk are enclosed by the curve γ. Let R0 > 0 be a
large real number so that the circle |z| = R0 encloses γ.
Z I
(a) Show that f (z) dz = f (z) dz for any R ≥ R0 .
|z|=R γ
I
(b) Suppose further that lim f (z) dz = 0, show that
R→+∞ |z|=R
I
f (z) dz = 0.
γ

(c) Consider the quotient p(z)

q(z) where p(z) and q(z) are polynomials, and deg q ≥
deg p+2. Using (a) and (b), show that given any simple-closed curve γ enclosing
all roots of q, we have I
p(z)
dz = 0.
γ q(z)
3.4. Cauchy’s Integral Formula II 81

3.4. Cauchy’s Integral Formula II

Recall that Cauchy’s integral formula asserts that if f : Ω → C is holomorphic on a
simply-connected domain Ω and γ is a closed curve in Ω, then we have:
I
1 f (z)
f (α) = dz
2πi γ z − α
if γ encloses α.
f (z)
If the integrand is of the form whenever α 6= β, we can still use
(z − α)(z − β)
Cauchy’s integral formula in a modified way: either by partial fractions, or by a hole-
drilling argument illustrated in the previous section.
f (z)
However, if the integrand is of the form , then both partial fractions and
(z − α)2
the hole-drilling
I argument do not work well (think about why). Indeed, the contour
f (z)
integral 2
dz is related to f 0 (α), and this fact has many deep and surprising
γ (z − α)
consequences as we will see later. These include the celebrated Liouville’s Theorem
(which implies Fundamental Theorem of Algebra).
Our goal is to prove and discuss the following higher-order Cauchy’s integral formula:

Theorem 3.13 (Higher-Order Cauchy’s Integral Formula). Let f : Ω → C be a holomor-

phic function defined on a simply-connected domain Ω, and α be any point in Ω. Then, for
any simple closed curve γ enclosing α, the n-th derivative of f at α is equal to:
I
n! f (z)
f (n) (α) = dz, n = 0, 1, 2, . . .
2πi γ (z − α)n+1

Corollary 3.14. If f is holomorphic on an open domain Ω, then f is complex differentiable

for infinitely many times on Ω, i.e. f (n) exists on Ω for any n ≥ 0.

The corollary is a very remarkable and surprising result. In Real Analysis, there are
many functions which are differentiable for one time but not the second time or further.
However, this theorem and the corollary assert that once f is complex differentiable on a
simply-connected domain (say an open ball), then it is infinitely differentiable on that
domain!

3.4.1. Elementary Examples. Again, we will first see some examples of using the
higher-order Cauchy’s integral formula, then we will give a proof for it. As a quick
example: I
1
2
dz.
|z|=1 z
One way of evaluating it is to argue that its primitive function is − z1 , which is well defined
and holomorphic near the contour |z| = 1. Then by Proposition 3.4, the contour integral
is 0.
Let’s see how to obtain the same result using Theorem 3.13 (with n = 1, and
f (z) ≡ 1): I
1 1 d
1+1
dz = 1 = 0.
2πi |z|=1 z dz z=0
82 3. Contour Integrals

Example 3.10. Evaluate the contour integral using higher-order Cauchy’s integral
formula: I
e2z
3
dz.
|z|=1 z

Solution
In practice, it may be helpful to write the higher-order Cauchy’s integral formula as:
I
f (z) 2πi (n)
n+1
dz = f (α).
γ (z − α) n!
Let f (z) = e2z which is entire, then f 0 (z) = 2e2z and f 00 (z) = 4e2z . By Theorem
3.13 (with n = 2), we get:
I 2z I
e e2z
3
dz = 2+1
dz
γ z γ (z − 0)
2πi 00 2πi
= f (0) = · 4 = 4πi.
2! 2

Example 3.11. Evaluate the contour integral:

I
1
2 3
dz.
|z|=3 (z + i) (z − 2i)

Solution
The contour |z| = 3 encloses two singularities of the integrand, namely −i and 2i.
By the hole-drilling technique, we can pick a small ε > 0 such that:
I I I !
1 1
2 (z − 2i)3
dz = + 2 (z − 2i)3
dz.
|z|=3 (z + i) |z+i|=ε |z−2i|=ε (z + i)
Then we calculate each integral on the RHS individually:
I I 1
1 (z−2i)3
2 3
dz = 1+1
dz
|z+i|=ε (z + i) (z − 2i) |z+i|=ε (z + i)
2πi d 1 2πi
= =− 3
1! dz z=−i (z − 2i)3 3
I I 1
1 (z+i)2
dz = dz
|z−2i|=ε (z + i)2 (z − 2i)3 |z−2i|=ε (z − 2i)2+1
2
2πi d 1
=
2! dz z=2i (z + i)2
2
 
6
= πi ·
(z + i)4 z=2i
2πi
= 3
3
Therefore, I
1 2πi 2πi
dz = − 3 + 3 = 0.
|z|=3 (z + i)2 (z − 2i)3 3 3
3.4. Cauchy’s Integral Formula II 83

The higher-order Cauchy’s Integral Formula has some interesting applications on the
proof of some combinatorial identities. It is because for any non-negative integers n and
k (where n ≥ k), we have
I
1 zn 1 dk n n · (n − 1) · · · (n − k + 1)
k+1
dz = z = = Ckn
2πi |z−1|=r (z − 1) k! dz k z=1 k!
for any r > 0. Using this, one can derive some combinatorial identities in the example
below:
Example 3.12. Prove that
(a) C0n − C1n + C2n − · · · + (−1)n Cnn = 0 (where n ≥ 1).
m+n+1
(b) Cnn + Cnn+1 + Cnn+2 + · · · + Cnn+m = Cn+1

Solution

C0n − C1n + C2n − · · · + (−1)n Cnn

I Xn
1 (−1)k z n
= dz
2πi |z−1|=r (z − 1)k+1
k=0
I X n  1+k
1 −1
= −z n dz
2πi |z−1|=r z−1
k=0
I   −1 n+1
1 −1 1 − ( z−1 )
= −z n −1 dz
2πi |z−1|=r z−1 1 − ( z−1 )
I  n+1 !
1 −1
= z n−1 1 − dz
2πi |z−1|=r z−1
I
1 (−1)n+2 z n−1
= z n−1 + dz
2πi |z−1|=r (z − 1)n+1
1 dn n−1
= 0 + (−1)n+2 z
n! dz n z=1
=0

Cnn + Cnn+1 + Cnn+2 + · · · + Cnn+m

I Xm
1 z n+k
= dz
2πi |z−1|=r (z − 1)n+1
k=0
I Xm
1 zn
= z k dz
2πi |z−1|=r (z − 1)n+1
k=0
I n
1 z 1 − z m+1
= n+1
dz
2πi |z−1|=r (z − 1) 1−z
I
1 z n+m+1 + z n
= dz
2πi |z−1|=r (z − 1)n+2
m+n+1 m+n+1
= Cn+1 + 0 = Cn+1 .
84 3. Contour Integrals

Exercise 3.19. Prove that C0n + C1n + · · · + Cn−1

n
+ Cnn = 2n using complex analysis.

Exercise 3.20. Evaluate the following contour integrals:

I
sin z
(a) 2
dz
|z|=2 (z − π)
I
zetz
(b) 3
dz where t > 0 is real.
|z|=3 (z + 1)
I  
1 f (z)
(c) 2+z+ dz, where f is entire and f (0) = 1.
|z|=1 z z

Exercise 3.21. Evaluate the contour integral (where n is a positive integer):

I  2n
1 1
z+ dz.
|z|=1 z z
Hence, show that: Z 2π
(2n − 1)!!
cos2n θ dθ = 2π .
0 (2n)!!

Exercise 3.22. Let f : Ω → C be a holomorphic function defined on a simply-

connected domain Ω. Suppose BR (z0 ) ⊂ Ω, show that:
n!
f (n) (z0 ) ≤ n sup |f (z)|
R |z−z0 |=R
for any integer n ≥ 0.

3.4.2. Proof of Higher Order Cauchy’s Integral Formula. Now we discuss the
proof of Theorem 3.13. From the (zeroth order) Cauchy’s integral formula, we know:
I
1 f (z)
f (α) = dz,
2πi γ z − α
where α is a point on the domain Ω, and γ is a simple closed curve in Ω enclosing α.
Note that if w ∈ C is very small, α + w will still be enclosed by γ, and so we have:
I
1 f (z)
f (α + w) = dz.
2πi γ (z − α − w)
Our first goal is to show Theorem 3.13 holds for f 0 (α), i.e. n = 1. Recall that:
f (α + w) − f (α)
f 0 (α) = lim .
w→0 w
We will use the zeroth order Cauchy’s integral formula to evaluate such a limit:
I I 
1 1 f (z) f (z)
(3.6) f 0 (α) = lim dz − dz
2πi w→0 w γ z−α−w γ z−α
I  
1 1 1 1
= lim f (z) · − dz.
2πi w→0 γ w z−α−w z−α
By straight-forward computation, we get:
 
1 1 1 1
− = .
w z−α−w z−α (z − α − w)(z − α)
3.4. Cauchy’s Integral Formula II 85

f (z)
The integrand of (3.6) becomes , which is bounded as z ∈ γ is away
(z − α − w)(z − α)
from α and α + w when w is small, and that the holomorphic function f is bounded on γ
by Extreme-Value Theorem. The length of γ is also bounded. Using Lebesgue Dominated
Covergence Theorem (commonly called LDCT in short), we can switch the limit and the
integral sign of (3.6), and get:
I  
1 1 1 1
f 0 (α) = lim f (z) · − dz
2πi γ w→0 w z−α−w z−α
I
1 f (z)
= lim dz
2πi γ w→0 (z − α − w)(z − α)
I
1 f (z)
= dz,
2πi γ (z − α)2
proving Theorem 3.13 when n = 1.
The second and higher order cases of Theorem 3.13 can be proved by induction.
Assume the theorem holds for some integer n:
I
n! f (z)
f (n) (α) = dz
2πi γ (z − α)n+1
for any α enclosed by γ. When w is very small, α + w is also enclosed by γ, hence it is
also true that: I
n! f (z)
f (n) (α + w) = dz.
2πi γ (z − α − w)n+1
Our next goal is to determine f (n+1) (α) from the definition:
f (n) (α + w) − f (n) (α)
f (n+1) (α) = lim .
w→0 w
We leave it as an exercise:
Exercise 3.23. Follow the outline listed below, and complete the inductive proof of
Theorem 3.13:
(a) Show that:
 
1 1 1
−
w (z − α − w)n+1 (z − α)n+1
Xn
1 1
=
(z − α − w)(z − α) j=0
(z − α − w)j (z − α)n−j
(b) Using the induction assumption and LDCT, show that
f (n+1) (α)
I Xn
n! f (z) 1
= lim dz.
2πi γ w→0 (z − α − w)(z − α) j=0 (z − α − w)j (z − α)n−j
(c) Finally, complete the proof.

3.4.3. Liouville’s Theorem. We now discuss an important consequence (Liouville’s

Theorem) of the higher order Cauchy’s integral formula. Using this theorem, one can
give a very short and elegant proof that every non-constant complex polynomial must
have at least one root!

Theorem 3.15 (Liouville’s Theorem). Any bounded entire function must be constant.
86 3. Contour Integrals

Proof #1. The proof is a consequence of 1st-order Cauchy’s integral formula. Suppose
f : C → C is a entire function and that there exists M > 0 such that |f (z)| ≤ M for any
z ∈ C.
Take an arbitrary α ∈ C, and consider the contour |z − α| = R. By Theorem 3.13
with n = 1, we know: I
1 f (z)
f 0 (α) = dz.
2πi |z−α|=R (z − α)2
Then on the contour, we have:
f (z) M
≤ 2,
(z − α)2 R
and by Lemma 3.6, we can estimate that:
I
f (z) M 2πM
2
dz ≤ 2πR · 2 = .
|z−α|=R (z − α) R R
Therefore, we have for any α ∈ C and R > 0:
I
0 1 f (z) 1 2πM
|f (α)| = 2
dz ≤ · → 0 as R → +∞.
2πi |z−α|=R (z − α) 2π R
This shows f 0 ≡ 0, and hence f is a constant function. 

Proof #2. Another proof make use of the function

f (z)
g(z) =
z(z − α)
where α 6= 0 is an arbitrary fixed complex number. Suppose again |f (z)| ≤ M on C.
Then, given any R > |α|, we have on |z| = R:
M M
|g(z)| ≤ ≤ .
|z| |z − α| R(R − |α|)
By Lemma 3.6 again, we can estimate that:
I
M 2πM
g(z) dz ≤ 2πR · = → 0 as R → +∞.
|z|=R R(R − |α|) R − |α|
Suppose g(z) is holomorphic on R0 ≤ |z| ≤ R for any R > R0 > |α|, a standard
hole-drilling argument shows I
g(z) dz
|z|=R
is independent of R for any R > |α|. Therefore, by the above limit we have in fact
I I
g(z) dz = lim g(z) dz = 0
|z|=R0 R→+∞ |z|=R

for any R0 > |α|.

On the other hand, the function g can be expressed as
 
1 f (z) f (z)
g(z) = − ,
α z−α z
so by Cauchy’s Integral Formula we also have:
I
2πi
g(z) dz = (f (α) − f (0)).
|z|=R0 α
Hence f (α) = f (0). Since α is arbitrary, we conclude that f (z) = f (0) for any z ∈ C. In
other words, f is a constant function. 
3.4. Cauchy’s Integral Formula II 87

Exercise 3.24. Prove the following general version of Liouville’s Theorem: Suppose
f : C → C is an entire function, and there exists M > 0 and a nonnegative integer k
such that:
k
|f (z)| ≤ M |z| for any z ∈ C.
Show that f is a polynomial of degree at most k.

Exercise 3.25. Suppose f : C → C is an entire function satisfying:

|f (z)|
lim sup = 0.
R→+∞ |z|≥R R
Show that f is a constant function.

Liouville’s Theorem is a “luxury" for holomorphic functions. There are many non-
constant bounded functions f : R → R that are (real) differentiable everywhere, while
Liouville’s Theorem says there is no non-constant bounded functions f : C → C which
are complex differentiable everywhere.

3.4.4. Fundamental Theorem of Algebra. In this subsection we showcase THREE

proofs of the Fundamental Theorem of Algebra using complex analysis.

Theorem 3.16 (Fundamental Theorem of Algebra). Every non-constant complex polyno-

mial p(z) = an z n + an−1 z n−1 + . . . + a1 z + a0 must have at least one complex root.

1
Proof #1. We prove by contradiction. If p(z) has no root, then p(z) is an entire function.
1 1
Note that |p(z)| → ∞ as |z| → ∞, we have: p(z) → 0 as |z| → ∞. In particular, p(z) is
1
bounded. By Liouville’s Theorem, p(z) is constant, which is a contradiction. 

1 1
Proof #2. Suppose p(z) has no root, so that p(z) is an entire function. Since p(z) →0
1 1
as |z| → ∞, there exists R > 0 such that p(z) < p(0) whenever |z| ≥ R. Therefore,
1
supz∈C p(z) is achieved at some point in BR (0). By Maximum Principle (Proposition
1
3.11), the function p(z) must be a constant, which is a contradiction. 
0
Proof #3. Suppose p(z) has no root, then pp(z)
(z)
is an entire function. For any R > 0, we
have I  0 
p (z) n
− dz = 0 − 2πni = −2πni.
|z|=R p(z) z
p0 (z) n
On the other hand, one can easily compute that p(z) − z is a rational function of the
deg≤n−1
form deg=n+1 . Hence, by Exercise 3.18, we have
I  0 
p (z) n
− dz = 0 for large R.
|z|=R p(z) z
It leads to a contradiction. 
Remark 3.17. There are many proofs of Fundamental Theorem of Algebra, at least one
in almost all important fields in mathematics. We will give one more using Rouche’s
Theorem later. There is one in Topology using the concept of homotopy. There is even
one geometric proof using Gauss-Bonnet’s Theorem in Differential Geometry! Ironically,
despite the name of the theorem, a purely algebraic proof has not yet been found. The
most purest algebraic proof uses Galois Theory, but that proof is based on the fact that
88 3. Contour Integrals

every real number has a real cubic root (which has to be justified using Intermediate-Value
Theorem in Real Analysis).

Exercise 3.26. Using Liouville’s Theorem, show that if the image of an entire
function f : C → C is disjoint from an open ball Bδ (z0 ), then f is a constant
function.

Exercise 3.27. In the proof of Fundamental Theorem of Algebra (Theorem 3.16), we

used the fact that |p(z)| → ∞ as |z| → ∞. Although this fact is intuitively clear since
the dominant term an z n of p becomes very large when |z| → ∞, try to prove this fact
in a more rigorous way. Hint: try to show that if p(z) = an z n + an−1 z n−1 + . . . + a0 ,
then
n−1
|p(z)| ≥ |z| (|an z| − |an−1 | − . . . − |a0 |)
whenever |z| > 1.

The Liouville’s Theorem implies that if f : C → C is entire and the image omits a
set with non-empty interior, then it must be a constant. In fact, there is a much stronger
result called the Picard’s Little Theorem, which says that a non-constant entire function
f : C → C must have its image omitting at most 1 point in C.
3.5. Morera’s Theorem 89

3.5. Morera’s Theorem

Before we stated Morera’s Theorem, let’s recall the proof Iof Cauchy-Goursat’s Theorem.
Using the holomorphic condition on f , Step 1 shows that f (z) dz = 0 for any triangle
T Z
contour T in the domain. Using this fact, Step 2 shows F (z) := f (ξ) dξ, where
L(z0 ,z)
L(z0 , z) is the straight path from a fixed point z0 to z, is a primitive function for f , i.e.
F 0 (z) = f (z) on the convex domain Ω.
It is a nice observation that the proof in Step 2 requires only two facts about f ,
namely:
(1) f is continuous on Ω; and
I
(2) f (z) dz = 0 for any triangle T in Ω.
T
Under these two conditions, the entire argument in Step 2 is still valid even if we don’t
assume that f is holomorphic on Ω. Step 2 shows F 0 (z) = f (z) on Ω, hence proving
I
f (z) dz = 0 for any closed curve γ in Ω.
γ
The result that F 0 (z) = f (z) on Ω has another implication: since the primitive
function F is holomorphic on Ω (and its derivative is f ), the higher order Cauchy’s integral
formula (Theorem 3.13) and Corollary 3.14 tell us that F is complex differentiable on Ω
for infinitely many times. Certainly, it shows f = F 0 is also complex differentiable on Ω
for infinitely many times too. In particular, f is holomorphic on Ω.
To summarize, the preceding discussion proves the following remarkable result:

Theorem 3.18 (Morera’s Theorem). If f : Ω → C is a continuous function on an open

domain Ω, and I
f (z) dz = 0
T
for any triangle contour T in Ω, then f is holomorphic on Ω.

Remark 3.19. Although convexity of the domain is needed in Step 2 of the proof of
Cauchy-Goursat’s Theorem, we do not need to assume Ω is convex when using Morera’s
Theorem. It is because complex differentiability is a local property. One can first restrict f
on an open ball Bε (z0 ) which is convex, then prove f is holomorphic on Bε (z0 ). Simply
repeat the same argument on all other open balls in the domain. It will show f is
holomorphic on the whole Ω.
I
In practice, it seems more difficult to verify f (z) dz = 0 for any triangle T than to
T
show f is holomorphic directly. Nonetheless, Morera’s Theorem can come in handy if we
want to show holomorphicity of a function which is not quite explicit. In the last chapter,
we may encounter functions defined in an integral form, such as the Gamma’s function:
Z ∞
Γ(z) = tz−1 e−t dt.
0
It is almost impossible to find an explicit, integral-free expression. Nonetheless, it is
possible toIshow it is a holomorphic function using Morera’s Theorem. The key idea is to
show that Γ(z) dz = 0 for any triangle T in the domain under consideration.
T
90 3. Contour Integrals

Example 3.13. Define f : Ω := {z : Re(z) < 0} → C by:

Z ∞ zt
e
f (z) = dt.
0 t+1
Show that f (z) is holomorphic on Ω.

Solution
First we show that f is defined on Ω: for any z ∈ Ω and t ∈ [0, ∞), we have:
ezt
≤ ezt ≤ ext
t+1
(as usual, we denote z = x + yi). Note that:
Z ∞  ∞
1 xt 1
ext dt = e = − < ∞.
0 x 0 x
Z ∞ zt
e
Hence, dt is integrable.
0 t+1
It is quite difficult to find an explicit formula for f (z), let alone its derivative.
To show it is holomorphic, we are going to use Morera’s Theorem.
One requirement of applying Morera’s Theorem is that f needs to be a con-
tinuous function. To verify this, we fix z0 ∈ Ω and consider a sequence zj ∈ Ω
converging to z0 . We need to show f (zj ) → f (z0 ) as j → ∞. Since f (z) is defined
using an improper integral, we cannot use uniform convergence to justify the swap-
ping of limit and integral signs. Instead, we use LDCT. As −µ := Re(z0 ) < 0, we
can assume that for sufficiently large j, we have Re(zj ) < − µ2 < 0. Then, for any
t ∈ [0, ∞), and sufficiently large j, we have
µ
ezj t etRe(zj ) e− 2 t
≤= ≤ .
1+t 1+t 1+t
µ
− t µ
As e1+t2
≤ o(e− 4 t ) as t → +∞, and it is a continuous function on [0, ∞), there
exists a constant C > 0 such that
µ
e− 2 t µ
≤ Ce− 4 t for t ∈ [0, ∞).
1+t
µ
e− 2 t
Hence, is integrable over [0, ∞). By LDCT, we have
1+t
Z ∞ zj t Z ∞ Z ∞ z0 t
e ezj t e
lim f (zj ) = lim dt = lim dt = dt = f (z0 ).
j→+∞ j→+∞ 0 1+t 0 j→+∞ 1 + t 0 1 +t
Since this holds for any sequence zj converging to z0 , f is continuous at z0 . Also,
since z0 is arbitrary in Ω, we complete the verification that f is continuous on Ω.
Next we check the major
Z condition of Morera’s Theorem: given any triangle T
in Ω, we want to show f (z) dz = 0.
T
3.5. Morera’s Theorem 91

Z Z Z ∞
ezt
f (z) dz = dt dz
T T 0 t+1
Z ∞Z
ezt
= dz dt (Fubini’s Theorem)
0 T t+1
Z ∞
ezt
= 0 dt (since is holomorphic)
0 t+1
=0
To justify the legitimacy of using Fubini’s Theorem, the integral
Z Z ∞
ezt
dt |dz|
T 0 t+1
Z ∞
ezt 1
needs to be finite. To verify this, we consider dt ≤ − , so that
0 t + 1 x
Z Z ∞ Z
ezt 1
dt |dz| ≤ − |dz|,
T 0 t+1 T x
which is finite since x is away from 0 when z is on any triangle T ⊂ Ω.
Hence, by Morera’s Theorem, f is holomorphic on Ω.

Example 3.14. Define the Gamma function Γ : Ω → C on Ω := {z : Re(z) > 0} by:

Z ∞
Γ(z) := tz−1 e−t dt.
0
The purpose of this example is to show that Γ is holomorphic on Ω using a combina-
tion of Morera’s Theorem, Fubini-Tonelli’s Theorem, uniform convergence property,
and LDCT. [All in one!]

Solution
First we show that Γ is well-defined (i.e. the integral is finite) on Ω by showing that
both Z 1 Z ∞
tz−1 e−t dt and tz−1 e−t dt
0 1
are integrable if Re(z) > 0. For the first integral, we observe that tz−1 e−t =
tx−1 e−t ≤ tx−1 for any t ∈ [0, 1]. Whenever x > 0, we have
Z 1
1
tx−1 dt = < ∞,
0 x
and hence the first integral is integrable if x := Re(z) > 0. For the second integral,
we observe that:
tz−1 e−t
lim = lim tx−1 e−t/2 = 0.
t→+∞ e−t/2 t→+∞
−t/2
Since e is integrable on [1, ∞), we conclude that the second integral is also
integrable on [1, ∞). This shows Γ is well-defined on Re(z) > 0. Also, since
Z ∞
fn (z) := χ[ n1 ,∞) tz−1 e−t dt
0
where χ[ n1 ,∞) is the indicator function of the set [ n1 , ∞). By the fact that

χ[ n1 ,∞) tz−1 e−t dt ≤ tz−1 e−t ,

92 3. Contour Integrals

each fn (z) is also well-defined on Re(z) > 0.

We consider fn ’s here because when t = 0, the integrand tz−1 e−t is not well-
defined at z = 1 (hence Hnot holomorphic
R∞ functionH on Ω). Even if Fubini’s Theorem
allows the switching of T and 0 , the integral T tz−1 e−t may not be zero when
t = 0. We tackle this subtle issue by considering fn instead (which avoids t = 0).
Note that fn → Γ. We will show each fn is holomorphic, and using uniform
convergence to show that its limit function Γ is also holomorphic.
Next, show that each fn is continuous on Ω using LDCT. For any α ∈ Ω and any
sequence zj ∈ Ω converging to α as j → ∞, we need to justify that
lim fn (zj ) = fn (α).
j→∞

Let ε > 0 such that Re(α) > ε. For sufficiently large j, we have |zj − α| < 2ε , and
hence Re(zj ) > 2ε and Re(zj ) < Re(α) + 2ε . In order to apply LDCT on fn (zj ), we
need to bound the integrand. For any t ∈ [ n1 , ∞), we have for any large j, and
t ∈ [0, ∞):
ε ε
tzj −1 e−t = χ[ n1 ,1) txj −1 e−t + χ[1,∞) txj −1 e−t ≤ χ(0,1) t 2 −1 + χ[1,∞) tRe(α)+ 2 −1 e−t .
| {z }
=:g(t)
ε
2 −1 Re(α)+ 2ε −1
As χ(0,1) t is integrable on (0, 1) and χ[1,∞) t e−t is integrable on [1, ∞),
the function g(t) is integrable on (0, +∞). By LDCT, we conclude:
Z ∞ Z ∞
lim fn (zj ) = lim tzj −1 e−t dt = lim tzj −1 e−t dt = fn (α).
j→∞ j→∞ 1 1 j→∞
n n

Since α is arbitrary on Ω, and zj is an arbitrary sequence converging to α, we

conclude that fn is continuous on Ω.
Now we are ready to apply Morera’s Theorem to show that the function fn (z)
is holomorphic on Ω. We have already shown fn is continuous in the previous part,
and we are left to show that:
I
fn (z) dz = 0
T
for any triangle T in Ω. Note that
I I Z ∞ Z ∞ I Z ∞
fn (z) dz = tz−1 e−t dt = tz−1 e−t dz dt = 0 dt = 0.
T T 1/n 1/n T 1/n

Below we justify the switching of order of integration by Fubini-Tonelli’s Theorem.

We need to verify the following integral is finite:
I Z ∞ I Z ∞
tz−1 e−t dt |dz| = tx−1 e−t dt |dz| .
T 1/n T 1/n

By compactness of T , there exists δ, M > 0 such that x := Re(z) ∈ (δ, M ) for any
z ∈ T . Hence,
Z 1 Z 1
1 − n1δ
tx−1 e−t dt ≤ tδ−1 dt = < ∞,
1/n 1/n δ
and Z ∞ Z ∞
x−1 −t
t e dt ≤ tM −1 e−t dt < ∞.
1 1
These combine to show:
I Z ∞
tx−1 e−t dt |dz| < ∞,
T 1/n
3.5. Morera’s Theorem 93

H R∞
hence the switching of T
and 1/n
is justified. By Morera’s Theorem, fn is holo-
morphic on Ω.
Finally, by showing that fn converges uniformly to Γ on Ωε := {z : Re(z) > ε}
for any ε > 0, it will complete the proof that Γ is holomorphic on Ω. We show
uniform convergence from the definition. First we estimate |fn (z) − Γ(z)|. For any
z ∈ Ωε , we have:
Z 1/n
|fn (z) − Γ(z)| = tz−1 e−t dt
0
Z 1/n
≤ tz−1 e−t dt
0
Z 1/n Z 1/n
x−1 −t
= t e dt ≤ tx−1 dt
0 0
1 1
=< ε.
xnx εn
The last step follows from x := Re(z) > ε for any z ∈ Ωε . Taking supremum over
all z ∈ Ωε , we get
1
sup{|fn (z) − Γ(z)| : z ∈ Ωε } ≤ ε
εn
for any n ≥ 1. By squeeze theorem, we conclude:
1
lim sup{|fn (z) − Γ(z)| : z ∈ Ωε } ≤ lim = 0,
n→∞ n→∞ εnε
concluding that fn converges uniformly to Γ on Ωε for any fixed ε > 0.
By Morera’s Theorem (and its corollary), fn being holomorphic on every Ωε
S Γ is holomorphic on every Ωε (where ε > 0). This shows Γ is holomorphic
implies
on ε>0 Ωε (which is Ω).

Exercise 3.28. Define f : C\[0, 1] → C by:

Z 1 √
t
f (z) = dt.
0 t−z
Show that f is holomorphic on its domain.

Exercise 3.29. Suppose {fn }∞

n=1 is a sequence of holomorphic functions on an open
domain Ω, and that fn converges uniformly to f on Ω. Show that the limit function
f is also holomorphic on Ω.

Exercise 3.30. Recall that the Riemann’s zeta function ζ : Ω → C is defined on

Ω := {z : Re(z) > 1} and by:
X∞ X∞
1 1
ζ(z) := z
= z ln n
.
n=1
n n=1
e
X∞
1
(a) Show that the series z
converges uniformly on Ωε := {z : Re(z) > 1 + ε}
n=1
n
for any ε > 0.
(b) Show that ζ is holomorphic on Ω.
94 3. Contour Integrals

Exercise 3.31. Let f : C → C be a continuous function. Consider the integral

Z 1
F (z) := tz−1 f (t) dt.
0
Show that F is holomorphic on Ω := {z : Re(z) > 0}.
 Chapter 4

Taylor and Laurent Series

4.1. Taylor Series

4.1.1. Taylor Series for Holomorphic Functions. In Real Analysis, the Taylor series
of a given function f : R → R is given by:
f 00 (x0 ) f 000 (x0 )
f (x0 ) + f 0 (x0 ) (x − x0 ) + (x − x0 )2 + (x − x0 )3 + . . .
2! 3!
We have examined some convergence issues and applications of Taylor series in MATH
2033/2043. We also learned that even if the function f is infinitely differentiable
everywhere on R, its Taylor series may not converge to that function. In contrast, there is
no such an issue in Complex Analysis: as long as the function f : C → C is holomorphic
on an open ball Bδ (z0 ), we can show the Taylor series of f :
f 00 (z0 ) f 000 (z0 )
f (z0 ) + f 0 (z0 ) (z − z0 ) + (z − z0 )2 + (z − z0 )3 + . . .
2! 3!
converges pointwise to f (z) on Bδ (z0 ), and uniformly on any smaller ball. As we shall
see, it thanks to Cauchy’s integral formula. Moreover, the proof of Taylor Theorem in
Complex Analysis is also much easier than that in Real Analysis, again thanks to Cauchy’s
integral formula.
In this chapter, it is more convenient to re-label the variables in the Cauchy’s integral
formula:
I I
n! f (z) n! f (ξ)
f (n) (α) = dz −→ f (n)
(z) = dξ.
2πi γ (z − α)n+1 2πi γ (ξ − z)n+1
For the re-labelled Cauchy’s integral formula, we require the point z to be enclosed by the
simple closed curve γ.

Theorem 4.1 (Taylor Theorem for Holomorphic Functions). Given a complex-valued

function f which is holomorphic on an open ball BR (z0 ), the series:
X∞
f (n) (z0 )
(z − z0 )n
n=0
n!
converges (pointwise) to f (z) for any z ∈ BR (z0 ).

Proof. Given any z ∈ BR (z0 ), we let ε > 0 be small enough so that |z − z0 | < R − ε. For
simplicity, denote R0 = R − ε.

95
96 4. Taylor and Laurent Series

By Cauchy’s integral formula, for any z ∈ BR0 (z0 ), we have:

I
1 f (ξ)
f (z) = dξ.
2πi |ξ−z0 |=R0 ξ − z

Then, the contour |z − z0 | = R0 lies inside the open ball BR (z0 ). The key trick to prove
1
the Taylor Theorem is rewriting as a geometric series. Recall that:
ξ−z
1
= 1 + w + w2 + . . . whenever |w| < 1.
1−w
1
We first rewrite into this form:
ξ−z
1 1 1 1
= = ·
ξ−z (ξ − z0 ) − (z − z0 ) ξ − z0 1 − z−z0
ξ−z0
∞  n
1 X z − z0
=
ξ − z0 n=0 ξ − z0

z−z0
Here we have used the fact that ξ−z0 < 1. See the diagram below. The yellow ball is
BR (z0 ), and the red circle is |ξ − z0 | = R0 .

|ξ − z0 | = R0

z0
|z − z0 | < R0
z

Then, whenever z ∈ BR0 (z0 ), the function f (z) can be expressed as:
I
1 1
(4.1) f (z) = f (ξ) · dξ
2πi |ξ−z0 |=R0 ξ−z
I ∞  n
1 f (ξ) X z − z0
= dξ
2πi |ξ−z0 |=R0 ξ − z0 n=0 ξ − z0
I ∞
X
1 f (ξ) (z − z0 )n
= dξ
2πi |ξ−z0 |=R0 n=0 (ξ − z0 )n+1
I
Next we want to see whether we can switch the integral sign and the infinite
|ξ−z0 |=R0
∞
X
summation . For this we need to show uniform convergence of the series below.
n=0

X∞
f (ξ) (z − z0 )n
.
n=0
(ξ − z0 )n+1
4.1. Taylor Series 97

We use Weiestrass’s M-test: for any ξ on the circle {|ξ − z0 | = R0 }, we have:

f (ξ) (z − z0 )n (z − z0 )n
≤ sup |f (ξ)|
(ξ − z0 )n+1 (ξ − z0 )n+1 |ξ−z0 |=R0
| {z }
=:CR0
 n
CR0 |z − z0 |
=
R0 R0
Since |z − z0 | < R0 , the series
X∞  n
CR0 |z − z0 |
n=0
R0 R0
converges. Note that the above series does not depend on ξ (the integration variable).
X∞
f (ξ) (z − z0 )n
Hence by Weiestrass’s M-test, the series converges uniformly on the
n=0
(ξ − z0 )n+1
circle {|ξ − z0 | = R0 }, thus allowing the switch between the integral sign and the
summation sign in (4.1):
∞ I
1 X f (ξ) (z − z0 )n
f (z) = dξ
2πi n=0 |ξ−z0 |=R0 (ξ − z0 )n+1
∞ I !
X 1 f (ξ)
= n+1
dξ (z − z0 )n
n=0
2πi |ξ−z0 |=R 0 (ξ − z0 )
X∞
f (n) (z0 )
= (z − z0 )n .
n=0
n!
In the last step we have used the higher order Cauchy’s integral formula. 

Example 4.1. The function f (z) = sin z is an entire function. By straight-forward

computations, its derivatives are given by:
f 0 (z) = cos z f 00 (z) = − sin z
f (3) (z) = − cos z f (4) (z) = sin z
.. ..
. .
Inductively, it is easy to deduce that f (2k+1) (0) = (−1)k , and f (2k) (0) = 0 for any
integer k ≥ 0. Hence, the Taylor series of f about 0 is given by:
∞
X ∞
f (2k+1) (0) 2k+1 X (−1)k 2k+1
f (z) = z = z
(2k + 1)! (2k + 1)!
k=0 k=0
z3 z5 z7
=z− + − + ...
3! 5! 7!
This series converges to sin z for any z ∈ C, because sin z is entire (i.e. holomorphic
on every ball BR (0)).

Example 4.2. Consider the function f (z) = Log(z) which is holomorphic on Ω :=

C\{x + 0i : x ≤ 0}. Note that we can only apply Theorem 4.1 if the ball BR (z0 ) is
contained inside Ω.
98 4. Taylor and Laurent Series

Let’s take z0 = 1 as an example.

1
f 0 (z) = f 0 (1) = 1
z
00 1
f (z) = − 2 f 00 (1) = −1
z
(3) 2
f (z) = 3 f (3) (1) = 2
z
(4) 2×3
f (z) = − 4 f (4) (1) = −2 × 3
z
.. ..
. .
Inductively, we deduce that f (n) (1) = (−1)n−1 · (n − 1)! for n ≥ 1.
Therefore, the Taylor series for f about 1 is given by:
X∞ ∞
X
(−1)n−1 · (n − 1)! (−1)n−1
Log(z) = Log(1) + (z − 1)n = (z − 1)n
n=1
n! n=1
n
1 1 1
= (z − 1) − (z − 1)2 + (z − 1)3 − (z − 1)4 + . . .
2 3 4
Since f is holomorphic on B1 (1) (but not on any larger ball centered at 1), the
above Taylor series converges to Log(z) on B1 (1).

2
Example 4.3. The Taylor series for some composite functions, such as ez , can
be derived by substitution instead of deducing the general n-th derivative of the
function. For example:
z2 z3 z4
ez = 1 + z + + + + ...
2! 3! 4!
2 (z 2 )2 (z 2 )3 (z 2 )4
ez = 1 + z2 + + + + ...
2! 3! 4!
z4 z6 z8
= 1 + z2 + + + + ...
2! 3! 4!
2
Since the series for ez converges for any z ∈ C, the series for ez converges for any
z ∈ C as well.
Similarly, by replacing z by 1 − z in the Taylor series for Log(z), we get:
1 1 1
Log(1 − z) = −z − z 2 − z 3 − z 4 − . . .
2 3 4
The series for Log(z) about 1 converges when |z − 1| < 1, and so the above series
for Log(1 − z) converges when |(1 − z) − 1| < 1, i.e. |z| < 1.

Apart from using Theorem 4.1 to find the Taylor series of a given holomorphic
function, we can also make use of the geometric series formula directly:
1
= 1 + w + w2 + . . . where |w| < 1.
1−w
This method is particularly useful for functions whose n-th derivatives are tedious to
compute.
4.1. Taylor Series 99

Example 4.4. Consider the function:

z−2
f (z) = .
(z + 2)(z + 3)
We are going to derive its Taylor series about 0. First, we do partial fractions on the
function:
5 4
f (z) = − .
z+3 z+2
a
Then, we try to rewrite each term above in the form of 1−w . Note that:
5 5 1 5 1
= · z = ·

z+3 3 3 + 1 3 1 − − z3
5 X
∞
z n X
∞
5
= − = (−1)n n+1 z n (where |z| < 3)
3 n=0 3 n=0
3
4 4 1 2
= · z =

z+2 2 2 + 1 1 − − z2
∞ 
X z n
=2 − (where |z| < 2)
n=0
2
∞
X (−1)n n
= z .
n=0
2n−1
Hence, for |z| < 2, we have:
X∞ X∞ ∞  
5 (−1)n n X 5 1
f (z) = (−1)n n+1 z n − n−1
z = (−1) n
n+1
− n−1
zn.
n=0
3 n=0
2 n=0
3 2
5
To derive the Taylor series of f about other center (say 1), we can express z+3
4
and z+2 into:
5 5 5 1
= = ·

z+3 (z − 1) + 4 4 1 − − z−1
4
X∞  n
5 z−1
= − (where |z − 1| < 4)
n=0
4 4
X∞
(−1)n 5
= n+1
(z − 1)n
n=0
4
4 4 4 1
= = ·

z+2 (z − 1) + 3 3 1 − − z−1
3
X∞  n
4 z−1
= − (where |z − 1| < 3)
n=0
3 3
X∞
(−1)n 4
= n+1
(z − 1)n .
n=0
3
Therefore, on |z − 1| < 3, we have:
X∞  
n 5 4
f (z) = (−1) n+1
− n+1 (z − 1)n .
n=0
4 3
100 4. Taylor and Laurent Series

Exercise 4.1. Derive the Taylor series of each function below about the given center
z0 :
2π
(a) f (z) = sin 2z; z0 = 3
(b) f (z) = cos 3z; z0 = π
−z 3
(c) f (z) = e ; z0 = 0
(d) f (z) = Log(3 − 2z); z0 = 1

Exercise 4.2. Find the Taylor series about 0 of the functions below up to the z 4
term:
(a) f (z) = e−z cos z
(b) f (z) = Log(1 − ez )

Exercise 4.3. Find the Taylor series about z0 of the function below without using
Theorem 4.1. State its radius of convergence.
1
(a) f (z) = , z0 = 0
(z − 1)(z − 2)
1
(b) f (z) = , z0 = i
(z − 1)(z − 2)

Exercise 4.4. Let α, β and z0 be three distinct complex numbers. Consider the
function
1
f (z) = .
(z − α)(z − β)
Find the Taylor series about z0 of the above function, and state its radius of conver-
gence.

Exercise 4.5. Let α be a fixed non-zero complex number. Consider the principal
branch of (1 + z)α :
(1 + z)α := eαLog(1+z) .
Show that its Taylor series about 0 is given by:
X∞
α(α − 1) · (α − n + 1) n
(1 + z)α = 1 + z .
n=1
n!
State its radius of convergence.

4.1.2. Taylor Series with Remainder Term. In Real Analysis, the Taylor Theorem
with a remainder term asserts that for any smooth (C ∞ ) function f : R → R, we have:
N
X −1 Z x
f (n) (a) 1
f (x) = (x − a)n + (x − t)N −1 f (N ) (t) dt
n! (N − 1)! a
n=0 | {z }
=:RN (x)

The last integral term, commonly denoted as RN (x), measures how fast the Taylor
series converges to f (x) as N → ∞. If lim RN (x) → 0 for any x in an interval I, then
N →∞
X∞
f (n) (a)
the Taylor series (x − a)n converges (pointwise) to f (x) for any x ∈ I. If
n=0
n!
4.1. Taylor Series 101

furthermore, we have:
lim sup |RN (x)| → 0,
N →∞ x∈I
then the Taylor series converges uniformly to f on I. However, it is often not easy to
show RN → 0 as the N -th derivative f (N ) may not be easy to find.
Back to Complex Analysis, we will soon derive the remainder term for the Taylor
series for holomorphic functions. One good thing about the complex version is that the
remainder involves only f , but not its derivatives, making it much easier to handle the
convergence issue of complex Taylor series. It again thanks to Cauchy’s integral formula.

Proposition 4.2. Let f be a holomorphic function defined on BR (z0 ), then for any
z ∈ BR (z0 ), and any simple closed curve γ in BR (z0 ) enclosing both z and z0 , we have:
N
X −1 (n) I  N
f (z0 ) n 1 f (ξ) z − z0
f (z) = (z − z0 ) + dξ
n=0
n! 2πi γ ξ − z ξ − z0
| {z }
=:RN (z)

Proof. We only outline the proof since it is modified from the proof of Theorem 4.1.
Using Cauchy’s integral formula, we first have:
I
1 f (ξ)
f (z) = dξ.
2πi γ ξ − z
The key step in the proof of Theorem 4.1 is to write:
1 1 1 1
= = · ,
ξ−z (ξ − z0 ) − (z − z0 ) ξ − z0 1 − z−z
ξ−z0
0

z − z0
so that when < 1, we have:
ξ − z0
∞ 
X n
1 z − z0
z−z0 = .
1− ξ−z0 n=0
ξ − z0
Now, to prove this proposition, we modify the above key step a bit, by considering:
 N
1 − z−z 0 N
X −1  n
ξ−z0 z − z0
= .
1 − z−z
ξ−z0
0
n=0
ξ − z0
We leave the rest of the proof for readers (which is a good exercise to test your under-
standing of the proof of Theorem 4.1). 

Exercise 4.6. Complete the proof of Proposition 4.2.

Exercise 4.7. Consider the remainder term in Proposition 4.2:

I  N
1 f (ξ) z − z0
RN (z) = dξ.
2πi γ ξ − z ξ − z0
Let γ be the circle |ξ − z0 | = R0 such that |z − z0 | < R0 < R. Show that:
 N
R0 |z − z0 |
|RN (z)| ≤ 0 sup |f (ξ)| .
R − |z − z0 | R0 |ξ−z0 |=R0
102 4. Taylor and Laurent Series

Exercise 4.8. Let f be a holomorphic function on BR (z0 ). Using this estimate ob-
X∞
f (n) (z0 )
tained in Exercise 4.7, deduce that the Taylor series (z − z0 )n converges
n=0
n!
uniformly to f (z) on any smaller ball Br (z0 ) where 0 < r < R.

X∞
f (n) (z0 )
Remark 4.3. The uniform convergence of (z − z0 )n has many remarkable
n=0
n!
consequences as discussed in MATH 3033/2043. For instance, one can integrate a Taylor
series term-by-term.

Exercise 4.9. Consider the Taylor series for −Log(1 − ξ):

ξ2 ξ3 ξn
−Log(1 − ξ) = ξ + + + ··· + + ··· where |ξ| < 1
2 3 n
Show that:
z2 z4 z n+1
+ + ··· + + · · · = (1 − z)Log(1 − z) + z
2 3×4 n(n + 1)
for any z ∈ B1 (0).

Exercise 4.10. Show that for any z ∈ C, we have:

Z z X∞
2 (−1)n z 2n+1
e−ξ dξ = .
0 n=0
n!(2n + 1)
4.2. Laurent Series 103

4.2. Laurent Series

A Laurent series is a “power series” with negative powers of z − z0 as well. The general
form of a Laurent series about z0 is:
X∞ X∞
a−n
n
+ an (z − z0 )n
n=1
(z − z0 ) n=0
a−2 a−1
= ··· + 2
+ + a0 + a1 (z − z0 ) + a2 (z − z0 )2 + · · ·
(z − z0 ) z − z0
which can be abbreviated as:
∞
X
an (z − z0 )n .
n=−∞
∞
X X∞
a−n
A Laurent series is said to be convergent if both n
and an (z − z0 )n
n=1
(z − z0 ) n=0
converge.
If a−n = 0 for any negative −n, then the Laurent series is a Taylor series. On the other
hand, if a−n 6= 0 for some negative −n, then the Laurent series is undefined when z = z0 .
As such, a Laurent series is usually defined on an annular region {r < |z − z0 | < R}
instead of a ball centered at z0 . From now on, we denote such an annular region by:
AR,r (z0 ) := {z ∈ C : r < |z − z0 | < R}
where R, r ∈ [0, ∞]. Note that:
AR,0 (z0 ) = BR (z0 )\{z0 }
A∞,r (z0 ) = C\Br (z0 ) for r > 0
A∞,0 (z0 ) = C\{z0 }.

4.2.1. Examples of Laurent Series. While a Taylor series gives an analytic expres-
sion for a holomorphic function on a ball, a Laurent series gives an analytic expression
for a function that has a singularity at the center of a ball. Before we discuss a general
theorem about Laurent series, let’s first look at some examples of writing a function as a
Laurent series:
Example 4.5. Consider the function f : C\{1, 2} → C defined by:
1
f (z) = .
(z − 1)(z − 2)
It is holomorphic on its domain C\{1, 2}. Let’s express the above function as a
Laurent series about 1:
1 1 1
= =−
z−2 (z − 1) − 1 1 − (z − 1)
X∞
=− (z − 1)n where |z − 1| < 1.
n=0
Hence, on ∈ A1,0 (1), i.e. the green annulus in the figure below, we have:
∞ ∞
1 1 1 X X
f (z) = · =− (z − 1)n = − (z − 1)n−1
z−1 z−2 z − 1 n=0 n=0
1
=− − 1 − (z − 1) − (z − 1)2 + · · · .
z−1
104 4. Taylor and Laurent Series

x
1 2

However, the green annulus A1,0 (1) is not the only annulus centered at 1 on
which f is holomorphic. There is another one A∞,1 (1) = {1 < |z − 1|} centered at
1, i.e. the yellow annulus in the above figure, on which f is also holomorphic. It is
also possible to express f as a Laurent series on this yellow annulus:

1 1 1 1
= = · 1
z−2 (z − 1) − 1 z − 1 1 − z−1
∞   n
1 X 1
= (where |z − 1| > 1)
z − 1 n=0 z − 1
∞
X 1
=
n=0
(z − 1)n+1
Hence, on the yellow annulus A∞,1 (1), the function f can be expressed as the
following Laurent series:
∞
X
1 1 1 1
f (z) = = · = .
(z − 1)(z − 2) z − 1 z − 2 n=0 (z − 1)n+2

Example 4.6. Find the Laurent series about 0 of the function:

1
f (z) = z 2 e z
defined on C\{0}.

Solution
First recall that the Taylor series for ew is:
X∞
wn
ew = for any w ∈ C.
n=0
n!
Substitute w = z1 , where z 6= 0, we get:
∞
X
1 1
ez = ,
n=0
n!z n
4.2. Laurent Series 105

and hence:
1
f (z) = z 2 e z
∞
X 1
= z2
n=0
n!z n
∞
X 1
= n−2
n=0
n!z
1 1 1
= z2 + z + + + + ···
2 3!z 4!z 2

Exercise 4.11. Express the function:

1
f (z) =
z(z − 1)(z − 2)
as a Laurent series about 0 in each of the following annuli:
A1,0 (0), A2,1 (0), A∞,2 (0).
Also, express the function as a Laurent series about 1 in each of the following annuli:
A1,0 (1), A∞,1 (1).
[Hint: First expand f into partial fractions.]

Exercise 4.12. Find all possible Laurent (or Taylor) series about 1 for the function:
1
f (z) = 2 .
z − 2z
For each series, state the annulus or ball on which it converges.

Exercise 4.13. Find all possible Laurent (or Taylor) series about each z0 below for
1
the function f (z) = .
z
(a) z0 = 0
(b) z0 = 1
(c) z0 = i
For each series, state the annulus or ball on which it converges.

Exercise 4.14. Show that for any w such that |w| < 1, we have:
X∞
1 n(n − 1) n−2
3
= (−1)n w .
(1 + w) n=2
2
[Hint: use Exercise 4.5]
Hence, find all possible Laurent or Taylor series about i for the function:
1
f (z) = 3 .
z
For each series, state the annulus or ball on which it converges.
106 4. Taylor and Laurent Series

Exercise 4.15. Find the Laurent series about 1 on the annulus A∞,0 (1) for the
functions:
1 1
f (z) = sin and g(z) = cos .
z−1 z−1
Hence, find the Laurent series about 1 on A∞,0 (1) for:
z
h(z) = sin .
z−1

Exercise 4.16. What’s wrong with the following argument?

X∞
z
=z zn = z + z2 + z3 + · · ·
1−z n=0
X∞
z 1 1 1 1
=− 1 =− n
= −1 − − 2 − · · ·
1−z 1− z n=0
z z z
By subtraction, we get:
X∞
1 1 2
0 = ··· + + + 1 + z + z + · · · = zn.
z2 z n=−∞

4.2.2. Existence Theorem of Laurent Series. We have learned how to express a

function into a Laurent series through examples. Next, we proved a general existence
theorem of Laurent series for any holomorphic function on any annular region.

Theorem 4.4 (Laurent Theorem). Let f be a holomorphic function defined on an annulus

AR,r (z0 ) := {r < |z − z0 | < R} where R, r ∈ [0, ∞], then f can be expressed as a Laurent
series about z0 on the annulus AR,r (z0 ):
∞
X
f (z) = cn (z − z0 )n
n=−∞

for some complex numbers cn ’s.

Proof. The proof is similar to that of Taylor’s series, but is a bit trickier since an annulus
is not simply-connected and so Cauchy’s integral formula cannot be applied directly.
Fix z ∈ AR,r (z0 ), we first consider a simple closed curve Γ in AR,r (z0 ) which encloses
both z and z0 (just like in the proof of Taylor’s Theorem). However, we cannot apply
Cauchy’s integral formula on the integral:
I
1 f (ξ)
dξ
2πi Γ ξ − z0

since f is not holomorphic on BR (z0 ). However, we can construct a “key-hole” contour:

C =Γ+L−γ−L

where −γ is the clockwise circle, and L is a straight-path as shown in the figure below.
We can pick Γ to be the circle with radius slightly smaller than R, and γ with radius
slightly bigger than r so that C encloses z.
4.2. Laurent Series 107

Γ
R z

r
z0

−γ

L −L

Under such a construction, the contour C = Γ + L − γ − L is a simple closed curve

and the region enclosed by C becomes simply connected. We can then apply Cauchy’s
integral formula:
I
1 f (ξ)
(4.2) f (z) = dξ
2πi C ξ − z
I I I I 
1 f (ξ)
= + − − dξ
2πi Γ L γ L ξ−z
I I
1 f (ξ) 1 f (ξ)
= dξ − dξ.
2πi Γ ξ − z 2πi γ ξ − z

The key idea of the proof is to express the integral over Γ as a series of non-negative
powers, and the integral over γ as a series of negative powers.
When ξ ∈ Γ, we have |z − z0 | < |ξ − z0 |, so:
∞  n
1 1 1 1 1 X z − z0
= = · =
ξ−z (ξ − z0 ) − (z − z0 ) ξ − z0 1 − z−z
ξ−z0
0
ξ − z0 n=0 ξ − z0

Hence, the first integral becomes:

I I X ∞  n
1 f (ξ) 1 f (ξ) z − z0
dξ = dξ.
2πi Γ ξ − z 2πi Γ n=0 ξ − z0 ξ − z0

In order to switch the infinite summation and the integral sign, we justify that the series
converges uniformly on ξ ∈ Γ. Suppose Γ has radius R0 , then for any ξ ∈ Γ:
 n  n
f (ξ) z − z0 1 |z − z0 |
≤ 0 sup |f | .
ξ − z0 ξ − z0 R R0 Γ

Note that supΓ |f | is finite by compactness of Γ. Since |z − z0 | < R0 , the geometric series
X∞  n
1 |z − z0 |
sup |f |
n=0
R0 R0 Γ

converges. By Weierstrass’s M-test, the series

X∞  n
f (ξ) z − z0
n=0
ξ − z0 ξ − z0
108 4. Taylor and Laurent Series

converges uniformly on ξ ∈ Γ, so one can switch the summation and integral signs and
get:
I X∞  I 
1 f (ξ) 1 f (ξ)
(4.3) dξ = n+1
dξ (z − z0 )n .
2πi Γ ξ−z n=0
2πi Γ (ξ − z0 )

The second integral can be handled similarly. The difference is that when ξ ∈ γ, we
have |ξ − z0 | < |z − z0 | instead. We instead write:
∞  n
1 1 1 1 1 X ξ − z0
= =− · ξ−z0
= −
ξ−z (ξ − z0 ) − (z − z0 ) z − z0 1 − z−z z − z0 n=0 z − z0
0

Hence,
I I X ∞  n
1 f (ξ) 1 f (ξ) ξ − z0
dξ = − dξ.
2πi γ ξ−z 2πi γ n=0 z − z0 z − z0

We leave it as an exercise for readers to argue that the series converges uniformly on
ξ ∈ γ so that we can switch the integral and summations signs:
I X∞  I 
1 f (ξ) 1 n 1
(4.4) dξ = − f (ξ)(ξ − z0 ) dξ n+1
.
2πi γ ξ−z n=0
2πi γ (z − z0)

Combining (4.2), (4.3) and (4.4), we obtain:

X∞  I 
1 1
f (z) = f (ξ)(ξ − z0 )n−1 dξ
n=1
2πi γ (z − z0 )n
| {z }
(4.4)
X∞  I 
1 f (ξ)
+ n+1
dξ (z − z0 )n
n=0
2πi Γ (ξ − z0 )
| {z }
(4.3)

It completes the proof by defining

I
1
c−n = f (ξ)(ξ − z0 )n−1 dξ for −n = −1, −2, −3, · · ·
2πi γ
I
1 f (ξ)
cn = dξ for n = 0, 1, 2, 3, · · ·
2πi Γ (ξ − z0 )n+1

Exercise 4.17. Justify the claim in the above proof that the series:
∞
X  n
f (ξ) ξ − z0
n=0
z − z0 z − z0
converges uniformly on ξ ∈ γ (and z, z0 are considered to be fixed).

Remark 4.5. Although from the proof of Theorem 4.4 one can express the coefficient cn ’s
of a Laurent series in terms of contour integrals, we do not usually find the coefficients
this way since these contour integrals may not be easy to compute.
4.2. Laurent Series 109

4.2.3. Laurent Series with Remainders. Similar to Taylor series, one can refine
Theorem 4.4 a bit by deriving the remainder terms. Using the remainder terms, one can
argue that for a holomorphic function f defined on an annulus AR,r (z0 ), the Laurent
series converges uniformly to f on every smaller annulus AR0 ,r0 (z0 ) (where r < r0 < R0 <
R). This result is remarkable as it allows us to integrate a Laurent’s series term-by-term.

Proposition 4.6. Let f be a holomorphic function on the annulus AR,r (z0 ), where 0 ≤
r < R ≤ ∞. Then, for each positive integer N and z ∈ AR,r (z0 ), we have:
XN  I  I  N
1 1 1 f (ξ) ξ − z0
f (z) = f (ξ)(ξ − z0 )n−1 dξ + dξ
n=1
2πi γ (z − z0 )n 2πi γ z − ξ z − z0
| {z }
=:rN (z)
N
X −1  I  I  N
1 f (ξ) 1 f (ξ) z − z0
+ dξ (z − z0 )n + dξ
2πi Γ (ξ − z0 )n+1 2πi Γ ξ−z ξ − z0
n=0 | {z }
=:RN (z)

where Γ and γ are any pair of circles in AR,r (z0 ) centered at z0 such that z is bounded
between Γ and γ.

Proof. We leave the proof of Proposition 4.6 as an exercise. It is very similar to the
proof of Proposition 4.2 for Taylor series. Readers should first digest the whole proof of
Proposition 4.2, then write up a coherent proof for this proposition. 

Exercise 4.18. Prove Proposition 4.6. Using this, show that the Laurent series
about z0 for f converges uniformly on every smaller annulus AR0 ,r0 (z0 ) where
r < r0 < R0 < R. Give the proof in two different ways:
(a) Using Weierstrass’ M-test
(b) By estimating the reminder terms RN and rN directly.

One practical use of uniform convergence is term-by-term integration. For example,

1
consider the function f (z) = z 2 e z , which can be expressed as a Laurent series:
1 1 1 1
z2e z = z2 + z + + + + ···
2 3!z 4!z 2
Then, to integrate f (z) over the circle |z| = 1, we can integrate the Laurent series
term-by-term:
I
1
z 2 e z dz
|z|=1
I I I I I
1 1 1
= z 2 dz + z dz + dz + dz + 2
dz + · · ·
|z|=1 |z|=1 |z|=1 2 |z|=1 3!z |z|=1 4!z
2πi πi
=0+0+0+ + 0 + 0 + ··· = .
3! 6
Recall that for any simple closed γ enclosing the origin, the contour integral
I
z n dz
γ
is non-zero only when n = −1.
From the above example, we see the significance of expressing a function as a Laurent
series. To compute a contour integral, it often amounts to finding the coefficient c−1 of
the Laurent series. It leads to the develop of residue theory to be discussed in the next
section.
110 4. Taylor and Laurent Series

4.3. Residue Calculus

In this section we discuss both theory and applications of an important topic in Complex
Analysis: residue calculus. It has many powerful applications on evaluations of some
complicated real integrals that physicists and engineers often encounter.

4.3.1. Classification of Singularities. A singular point, or singularity, refers to a

point z0 at which a function f fails to be complex differentiable. For instance, 1 and 2
are singularities of the function:
1
f (z) = .
(z − 1)(z − 2)

It is possible for a function to have infinitely many singularities, such as:

1
g(z) =
sin z
whose singularities are 0, ±π, ±2π, etc.
Some functions even have singularities that form a “cluster”. For instance, consider:
1
h(z) =
sin z1
1 1
which is singular when z ∈ { πn : n ∈ Z} ∪ {0}. The singular set { πn : n ∈ Z} ∪ {0} form
a cluster around 0, meaning there is no way to find an annulus AR,0 (0) centered at 0
such that h is holomorphic on AR,0 (0). Hence, it is not possible to analyze the function h
by a Laurent series about 0 on AR,0 (0).
In order to utilize Laurent series, we focus on those singularities that can be isolated
from others. We have the following terminology:

Definition 4.7 (Isolated Singularity). A point z0 is said to be an isolated singularity for a

function f (z) if there exists ε > 0 such that f is holomorphic on Aε,0 (z0 ) = Bε (z0 )\{z0 }.

1
For the function g(z) = , all singularities are isolated as depicted in the diagram
sin z
below:
y

Around every isolated singularity z0 of a function f (z), it is possible (thanks to

Theorem 4.4) to express the function f as a Laurent series on a small annulus Aε,0 (z0 ):
∞
X
f (z) = cn (z − z0 )n .
n=−∞

Depending on the smallest n such that cn 6= 0, we have the following terminology:

• If c−1 = c−2 = c−3 = · · · = 0, then z0 is said to be a removable singularity of f . For
instance, 0 is such a singularity for the function:
sin z z2 z4 z6
=1− + − + ···
z 3! 5! 7!
4.3. Residue Calculus 111

• If k is a positive integer such that c−k 6= 0 while c−(k+1) = c−(k+2) = · · · = 0, then

z0 is said to be a pole of order k of f . For instance, 0 is a pole of order 3 for the
function:
sin z 1 1 z z3
4
= 3− + − + ···
z z 3!z 5! 7!
Moreover, a pole of order 1 is usually called a simple pole.
• If c−n 6= 0 for infinitely many negative integers −n, then z0 is said to be an essential
singularity. For instance, 0 is such a singularity for the function:
1 1 1 1
ez = 1 + + + + ···
z 2!z 2 3!z 3
If z0 is a removable singularity for f : BR (z0 )\{z0 } → C, then one can define
f (z0 ) := c0 so that f extends to become a holomorphic function on BR (z0 ). That’s why
we say z0 is removable. Similarly, if z0 is a pole of order n for f : AR,0 (z0 ) → C, then
(z − z0 )n f (z) extends to become a holomorphic function on BR (z0 ). However, a function
with an essential singularity cannot be extended to become a holomorphic function in a
similar way (that’s why we call it essential).
To determine the order of a pole, we may simply find its Laurent series expansion.
1
However, sometimes it is not easy to do so, such as 0 for the function . An alternative
sin z
way to find the order of a pole is to consider the limit:
lim (z − z0 )k f (z).
z→z0

If k is an integer such that:

lim (z − z0 )k f (z) exists and is non-zero,
z→z0

then the order of the pole z0 is k. For example, since:

z
lim = 1 6= 0,
z→0 sin z

1
0 is a pole of order 1 for the function . Hence, one can express this function as a
sin z
Laurent series on a small annulus Aε,0 (0):
1 c−1
= + c0 + c1 z + c2 z 2 + · · ·
sin z z
Multiplying z on both sides, we get:
z
= c−1 + c0 z + c1 z 2 + c2 z 3 + · · ·
sin z
and by letting z → 0, we can also conclude that c−1 = 1. Therefore, if γ is a simple close
curve enclosing 0 in this small annulus Aε,0 (0), then we have:
I I   I
1 1 1
dz = + c0 + c1 z + c2 z 2 + · · · dz = dz + 0 + 0 + · · · = 2πi.
γ sin z γ z γ z

Exercise 4.19. Find all isolated singularities of each function below, and classify
the nature of these singularities. For poles, state also their orders.
ez − 1
(a) f (z) =
z
Log(z)
(b) g(z) =
(z − 3)5
(c) h(z) = z 4023 cos z1
112 4. Taylor and Laurent Series

4.3.2. Residues. As illustrated in many examples, the coefficient c−1 of a Laurent

series plays a special role in evaluating a contour integral. It is special in a sense that for
an integer n, (
I
n 2πi if n = −1
(z − z0 ) dz =
|z−z0 |=ε 0 otherwise
c−1
Hence, to integrate a Laurent series, one only needs to integrate the term , which
z − z0
can be done by Cauchy’s integral formula. In view of the special role of c−1 , we define:

Definition 4.8 (Residues). Let z0 be an isolated singularity of f (z) such that the Laurent
series about z0 for f on some annulus Aε,0 (z0 ) is given by:
∞
X
f (z) = cn (z − z0 )n ,
n=−∞

then we denote and define the residue of f at z0 by:

Res(f, z0 ) := c−1 .

Example 4.7. Find the residue of the function

z 2 − 2z
f (z) =
(z + 1)2 (z 2 + 4)
at each of its isolated singularity.

Solution
The denominator has roots −1, 2i and −2i, hence they are isolated singularities of
f . In this solution, we will decompose f (z) into partial fractions. It may not be a
pleasant way finding residues, but we will later provide an easier way.
Note that f is a rational function, we can break it into partial fractions:
A B C D
f (z) = + + + .
(z + 1)2 z + 1 z − 2i z + 2i
We leave it as an exercise for readers to determine the value of A, B, C and D. One
should be able to get:
3
5 − 14
15
7+i 7−i
f (z) = + + 25 + 25 .
(z + 1)2 z + 1 z − 2i z + 2i
| {z }
holomorphic near −1
7+i 7−i
25 25
On a small annulus Aε,0 (−1) about −1, the last two terms are +
z − 2i z + 2i
holomorphic. Therefore, if one express them as a Laurent series about −1, only
1
non-negative powers of z + 1 will appear, and the coefficient of z+1 will not be
affected. Hence, we have:
14
Res(f, −1) = − .
15
By a similar reason, we have:
7+i 7−i
Res(f, 2i) = and Res(f, −2i) = .
25 25
4.3. Residue Calculus 113

Exercise 4.20. Determine all isolated singularities of the function

z2 + 1
f (z) = .
(z + 1)(z − 1)2
and find the residue at each isolated singularity.

It is no doubt that partial fraction decompositions are time-consuming and not fun
(it may remind you the computational nightmare you might have encountered in MATH
1014). Fortunately, there is a better way for finding residues for poles (does not work for
essential singularity).
If we know already that z0 is a pole of order 1 (i.e. simple pole) of a function f (z),
then
c−1
f (z) = + c0 + c1 (z − z0 ) + c2 (z − z0 )2 + · · ·
z − z0
It is then easy to see that
c−1 = lim (z − z0 )f (z).
z→z0
Therefore, in order to find Res(f, z0 ) for a simple pole z0 , we simply need to compute the
above limit.
Now consider the case if z0 is a pole of order k for f , then its Laurent series about z0
is given by:
c−k c−(k−1) c−1
f (z) = k
+ k−1
+ ··· + + c0 + c1 (z − z0 ) + c2 (z − z0 )2 + · · ·
(z − z0 ) (z − z0 ) z − z0
Our goal is to find c−1 . By multiplying both sides by (z − z0 )k , we can get:
(z − z0 )k f (z) = c−k + c−(k−1) (z − z0 ) + · · · + c−1 (z − z0 )k−1 + c0 (z − z0 )k + · · ·
By differentiating both sides for k − 1 times, all terms involving (z − z0 )n with n < k − 1
will disappear:
dk−1
(z − z0 )k f (z) = c−1 (k − 1)! + e c1 (z − z0 )2 + · · ·
c0 (z − z0 ) + e
dz k−1
dk−1
We have used the fact that k−1 (z − z0 )k−1 = (k − 1)!, and e c0 , e
c1 , . . . are some complex
dz
numbers (which we do not need to know their values).
By letting z → z0 , we get:
dk−1
lim (z − z0 )k f (z) = c−1 (k − 1)!
z→z0 dz k−1

which provides a good way to find c−1 without expanding a Laurent series:

Proposition 4.9. Suppose z0 is a pole of order k < ∞ for a function f , then we have:
1 dk−1
Res(f, z0 ) = lim (z − z0 )k f (z).
(k − 1)! z→z0 dz k−1
In particular, for a simple pole z0 , we have:
Res(f, z0 ) = lim (z − z0 )f (z).
z→z0

Proof. See the preceding paragraph. 

114 4. Taylor and Laurent Series

Example 4.8. Find the residue of the function

z 2 − 2z
f (z) =
(z + 1)2 (z 2 + 4)
of each isolated singularity using Proposition 4.9.

Solution
As discussed before, the isolated singularities are −1, 2i and −2i. Observe that:
z 2 − 2z 3
lim (z + 1)2 f (z) = lim 2
= 6= 0.
z→−1 z→−1 z + 4 5
Hence −1 is a pole of order 2. From Proposition 4.9, we have:
1 d2−1
Res(f, −1) = lim (z + 1)2 f (z)
(2 − 1)! z→−1 dz 2−1
d z 2 − 2z
= lim
z→−1 dz z 2 + 4
2z 2 + 8z − 8 14
= lim =− .
z→−1 (z 2 + 4)2 25
Both 2i and −2i are simple poles, so we have:
z 2 − 2z 7+i
Res(f, 2i) = lim (z − 2i)f (z) = lim =
z→2i z→2i (z + 1)2 (z + 2i) 25
2
z − 2z 7−i
Res(f, −2i) = lim (z + 2i)f (z) = lim =
z→−2i z→−2i (z + 1)2 (z − 2i) 25

Example 4.9. Find the residue at 0 of each function below:

ez ez − 1 ez
f (z) = g(z) = h(z) =
sin z sin z sin2 z

Solution
For each function, we first determine whether 0 is a pole, and find out its order.
For f (z), we consider:
z
lim zf (z) = lim · ez = 1 · e0 = 1 6= 0.
z→0 sin z
z→0
Hence 0 is a simple pole for f , and Res(f, 0) = 1.
For g(z), note that:
z2 z3 z z2
z+ 2! + 3! + ··· 1+ 2! + 3! + ···
lim g(z) = lim z3 z5
= lim z2 z4
= 1 < ∞.
z→0 z→0 z− 3! + 5! − ··· z→0 1− 3! + 5! − ···
Hence 0 is a removable singularity of g(z), and there is no z1 -term in the Laurent
series, and so Res(g, 0) = 0.
For h(z):
 z 2
lim z 2 h(z) = lim ez = 1 6= 0.
z→0 z→0 sin z
4.3. Residue Calculus 115

Hence 0 is a pole of order 2 for h. By Proposition 4.9, we can find:

1 d 2 d z 2 ez
Res(h, 0) = lim z h(z) = lim
1! z→0 dz z→0 dz sin2 z

sin2 z (2zez + z 2 ez ) − z 2 ez · 2 sin z cos z

= lim 4
 sin z
z→0
 2 z 
z e z sin z − z cos z
= lim + 2ze
z→0 sin2 z sin3 z
 3 5
  2 4

z − z3! + z5! − · · · − z 1 − z2! + z4! − · · ·
= 1 + lim 2zez  
z→0 sin3 z
 1 1 1 1 
( 2! − 3! )z 3 − ( 4! − 5! )z 5 + · · ·
= 1 + lim 2zez
z→0 sin3 z
  3   5 
z 1 1 z 1 1 z
= 1 + lim 2ze − − − + ···
z→0 2! 3! sin3 z 4! 5! sin3 z
 
1 1
= 1 + 0 · e0 · − + 0 + 0 + ···
2! 3!
= 1.

Exercise 4.21. For each function below, find its residue at each isolated singularity
using any method:
z2 − 1 1 z 1997 − 1
z 3 (z 2 + 1) 2
6z + 8z + 9 z 2047 − 1
1 e2zi e2zi − 1
z
e −1 sin z sin z
1 z2 sin z
z sin z e1/z z 2 (z − π)3

Exercise 4.22. Compute the following residues:

 
1
(a) Res , 0
2 cos z − 2 + z 2
 
z 2n
(b) Res , 1
(z − 1)n

4.3.3. Residue Theorem. The residue I Res(f, z0 ) of an isolated singularity z0 deter-

mines the value of a contour integral f (z) dz where γ is a tiny simple closed curve so
γ I
that z0 is the only singularity it encloses. Namely, we have f (z) dz = 2πi Res(f, z0 ).
γ
If a simple closed curve γ encloses more than one isolated singularities {z1 , · · · , zN },
then we may first express the contour integral over γ as the sum of contour integrals:
I I I
f (z) dz = f (z) dz + · · · + f (z) dz
γ γ1 γN

where each γj is a small simple-closed curve so that zj is the only singularity it encloses.
Then, each γj -integral is given by 2πi Res(f, zj ), and hence we have the following
theorem:
116 4. Taylor and Laurent Series

Theorem 4.10 (Residue Theorem). Let f : Ω → C be a complex-valued functions whose

singularities are all isolated. Let γ be a simple closed curve, and z1 , · · · , zN ∈ Ω be all the
singularities enclosed by γ. Then, we have:
I XN
f (z) dz = 2πi Res(f, zj ).
γ j=1

Proof. Let ε > 0 be sufficiently such that each circle {|z − zj | = ε}, denoted by γj ,
encloses zj as the only singularity of f (see figure below).

z1
γ1

γ2 γ3
z2 z3

Then, by the standard hole-drilling argument, we have:

I I I
f (z) dz = f (z) dz + · · · + f (z) dz
γ γ1 γN

Each γj encloses zj as the only singularity of f . Express f as a Laurent series on Aε,0 (zj ):
∞
X
f (z) = cn (z − zj )n .
n=−∞
I
Recall that (z − zj )n dz 6= 0 only when n = −1, and by uniform convergence of
γj
Laurent series, we get:
I
f (z) = 2πic−1 = 2πi Res(f, zj ).
γj

Therefore, we have:
I N
X
f (z) dz = 2πi Res(f, zj ),
γ j=1
completing the proof. 
4.3. Residue Calculus 117

Example 4.10. Use Residue Theorem to evaluate the contour integral:

I
z 2 − 2z
2 2
dz
|z|=R (z + 1) (z + 4)

where R is in the range of:

(a) 0 < R < 1
(b) 1 < R < 2
(c) 2 < R.

Solution
z 2 − 2z
Denote f (z) = . The singularities of f are −1, 2i and −2i. We have
(z + 1)2 (z 2 + 4)
calculated in Example 4.8 that
14 7+i 7−i
Res(f, −1) = − Res(f, 2i) = Res(f, −2i) =
15 25 25
y

2i

(a) (b) (c) x

−1

−2i

(a) When 0 < R < 1, the circle |z| = R does not enclose any singularities, hence
I
f (z) dz = 0.
|z|=R

(b) When 1 < R < 1, the circle |z| = R encloses the singularity −1 only, hence
I
28πi
f (z) dz = 2πi Res(f, −1) = −
|z|=R 15
(c) When R > 2, the circle |z| = R encloses all three singularities, hence
I
f (z) dz = 2πi (Res(f, −1) + Res(f, 2i) + Res(f, −2i))
|z|=R
 
14 7 + i 7 − i 56πi
= 2πi − + + =− .
15 25 25 75
118 4. Taylor and Laurent Series

Example 4.11. Let N be a positive integer and γN be the square contour with
vertices ±(N + 21 ) ± (N + 12 )i. Use Residue Theorem to show:
XN I
1 π2 1 π
2 2
= + cot πz dz.
n=1
n 3 2πi γN z 2
Hence, deduce that:
1 1 π2
1+ + + · · · = .
22 32 6
1 1
The problem about the exact value of 1 + 22 + 32 + · · · was famously known as
“Basel’s Problem”. It was first solved by Euler in 1734.

Solution
π π cos πz
Denote f (z) := cot πz = 2 . Its singularities are the set of all integers n.
z2 z sin πz
First we observe that
πz
lim z 3 f (z) = lim · cos πz = 1,
z→0 z→0 sin πz
so 0 is a pole of order 3 for f . By Proposition 4.9, its residue is given by:
1 d2 1 d2
Res(f, 0) = lim 2 z 3 f (z) = lim 2 πz cot πz
2! z→0 dz 2 z→0 dz
π 2 (πz cos πz − sin πz)
= lim
z→0
 sin3 πz 
π2 z2 π3 z3
π 2 πz(1 − 2! + · · · ) − (πz − 3! + ···)
= lim
z→0 sin3 πz
3 3
− π 3z + higher-order terms π2
= lim π 2 · 3 =− .
z→0 sin πz 3
For any non-zero integer n, observe that:
π(z − n) cos πz (−1)n 1
lim (z − n)f (z) = lim 2 n = 2 = 2.
z→n z→n z · (−1) sin(π(z − n)) n · (−1)n n
| {z }
=sin πz
1
Hence, n is a simple pole of f (for any n 6= 0), and Res(f, n) = .
n2
Now consider the contour γN . The singularities it encloses are:
0, ±1, ±2 · · · , ±N.
4.3. Residue Calculus 119

y
(N + 21 )i

x
−(N + 12 ) N+ 1
2

−(N + 21 )i

By Residue Theorem, we have:

I N
X
f (z) dz = 2πi Res(f, n)
γN n=−N
  
1 1
= 2πi Res(f, 0) + 2 1 + 2 + · · · + 2
2 N
N
!
π2 X 1
= 2πi − +2 .
3 n=1
n2
By rearrangement, we have the desired result:
XN I
1 π2 1
= + f (z) dz.
n=1
n2 6 2πi γN

The remaining task is to show:

I
π
lim cot πz dz = 0.
N →∞ γN z2
We do so by estimating the contour integral:
When z ∈ γN , we have |z| ≥ N + 12 > N . It is also possible to show that
|cot πz| < 2 for any z ∈ γN (this is left as an exercise). Therefore, on γN , we have
the bound:
π 2π
cot πz ≤ 2 .
z2 N
The length of γN is 8N + 4. By Lemma 3.6, we get:
I
π 2π
2
cot πz dz ≤ (8N + 4) · 2 → 0 as N → ∞,
γN z N
completing the proof.

Exercise 4.23. Complete the detail of the above example that:

|cot πz| < 2
for any z ∈ γN . [Hint: Write z = x + yi, and find an expression for cot πz in terms
of x and y. Then, maximize |cot πz| on each side of the contour γN .]
120 4. Taylor and Laurent Series

Exercise 4.24. Show, using residues, that for any positive integer k, the Riemann’s
zeta function takes the value:
(−1)k+1 (2π)2k
ζ(2k) = · B2k
2 · (2k)!
where B2k is the 2k-th Bernoulli’s number defined as follows:
The Bernoulli’s numbers (where n ∈ N) are defined to be coefficient of the series:
X∞
z Bn n
= z .
ez − 1 n=0 n!
z 2 3
Note that has a removable singularity at 0 (since ez − 1 = z + z2! + z3! + · · · ).
ez − 1
Therefore, its Laurent series about 0 is indeed a Taylor’s series. The general formula
of these Bernoulli’s numbers can hardly be found, but by direct computations, one
can find:
1 1 1 1
B0 = 1 B1 = − B2 = B3 = 0 B4 = − B5 = 0 B6 = ···
2 6 30 42
π
Hint: consider the function 2k cot πz, and mimic the previous example (which
z
is the case of ζ(2)).

Exercise 4.25. Use Residue Theorem to evaluate the following contour integrals:
I
1
(a) 2+1
dz
|z|=3 z
I
z 3 + 3z + 1
(b) 4 2
dz
|z|=2 z − 5z
I
ez + z
(c) 4
dz
|z−i|=2 (z − 1)
I
sin z
(d) 4023
dz
|z−i|=2 (z − i)
I
(e) tan πz dz where γ is the rectangle contour with vertices:
γ

(−2, 0), (2, 0), (2, 1), (−2, 1).

Exercise 4.26. Let γN be the square contour with vertices ±(N + 21 )π ± (N + 12 )πi
where N is a positive integer. Show that:
I N
1 1 1 2 X (−1)n
csc z dz = + .
2πi γN z 2 6 π 2 n=1 n2
Hence, show that:
1 1 1 π2
1− + − + · · · = .
22 32 42 12
4.3. Residue Calculus 121

Exercise 4.27. Determine the residues of all isolated singularities of the function:
1
f (z) = .
(2z − 1) sin πz
By considering a suitable contour integral of f , show that:
1 1 1 π
1 − + − + ··· = .
3 5 7 4

4.3.4. Evaluation of Real Integrals. Residues are often used to evaluate some
difficult real integrals that physicists and engineers may encounter.

Example 4.12. Evaluate the real definite integral:

Z 2π
1
dθ
0 a − b cos θ
where a and b are real numbers such that 0 < b < a.

Solution
The key trick is to express the real integral as a complex integral of the circle contour
|z| = 1, which is parametrized by z = eiθ where 0 ≤ θ ≤ 2π.
When z = eiθ is on the contour {|z| = 1}, we have
 
eiθ + e−iθ 1 1
cos θ = = z+
2 2 z
1 1
dz = ieiθ dθ =⇒ dθ = iθ dz = dz
ie iz
Therefore, the real integral can be written as a complex integral as:
Z 2π I I
1 1 1 1
dθ = b 1

· dz = 2i 2 − 2az + b
dz.
0 a − b cos θ |z|=1 a − 2 z + z
iz |z|=1 bz
We can then use residue theory to evaluate the complex integral. The singularities
of the integrand are roots of the quadratic equation bz 2 − 2az + b = 0, which are:
√ √
a − a 2 − b2 a + a2 − b2
ω1 = and ω2 = .
b b
Note that a > b, so both roots are real. We further observe that:
a+0
ω2 > > 1 and ω1 ω2 = 1,
b
and so |ω1 | < 1. Therefore, ω1 is the only singularity enclosed by the contour |z| = 1.
As ω1 and ω2 are distinct, they are simple poles, and so the contour integral is given
by:
I I
1 1
2
dz = dz
|z|=1 bz − 2az + b |z|=1 b(z − ω1 )(z − ω2 )
I 1  
b(z−ω2 ) 1
= dz = 2πi
|z|=1 z − ω1 b(z − ω2 ) z=ω1
2πi πi
= = −√ .
b(ω1 − ω2 ) a2 − b2
Hence, the real integral is given by:
Z 2π
1 πi 2π
dθ = −2i · √ =√ .
0 a − b cos θ a2 − b2 a2 − b2
122 4. Taylor and Laurent Series

Before we proceed to the next example, let’s first prove a useful observation which
will come in handy later on.

Exercise 4.28. Show that the function eiz is bounded on the upper-half plane, i.e.
there exists M > 0 such that eiz ≤ M whenever Im(z) ≥ 0. On the other hand,
show that the function = cos z is unbounded on the upper-half plane.

Example 4.13. Evaluate the following real integral:

Z ∞
cos x
2
dx.
−∞ 1 + x

Solution
Let’s consider the following semi-circle contour:
y

CR
i

x
−R LR R

Denote CR to be the (open) semi-circle with radius R, LR to be the straight-path

from −R to R, and γR to be the closed semi-circular path CR + LR . We consider
this contour because
Z ∞ Z R Z
cos x cos x cos z
2
dx = lim 2
dx = lim dz.
−∞ 1 + x R→+∞ −R 1 + x R→+∞ LR 1 + z2
Note that: I Z Z
cos z cos z cos z
2
dz = 2
dz + dz.
γR 1 + z LR 1 + z CR 1 + z2
The γR -integral can be computed using residues. If we are able to show the
CR -integral
Z tends to 0 as R → +∞, then one can determine our desired limit
cos z
lim dz.
R→+∞ L 1 + z 2
R
cos z
Unfortunately, it is not possible to bound as cos z is unbounded according
1 + z2
to Exercise 4.28. One trick to get around with this issue is to consider the following
function instead:
eiz
f (z) = .
1 + z2
When z ∈ LR , we have z = x + 0i and so:
Z Z R
eix cos x + i sin x cos x + i sin x
f (z) = = =⇒ f (z) dz = dx.
1 + x2 1 + x2 LR −R 1 + x2
Z
If we are able to find out lim f (z) dz, then one can recover the value of
R→+∞ L
Z ∞ R Z
cos x
2
by simply taking the real-part of lim f (z) dz.
−∞ 1 + x R→+∞ L
R
4.3. Residue Calculus 123

By considering the contour γR = CR + LR , we have:

I Z Z
f (z) dz = f (z) dz + f (z) dz.
γR LR CR
The only singularity enclosed by γR is i (when R is sufficiently large), so:
I
1 π
f (z) dz = 2πi Res(f, i) = 2πi · = .
γR 2ie e
Next we show the CR -integral converges to 0 as R → ∞. From Exercise 4.28,
the term eiz is bounded on the upper-half plane, and so whenever z ∈ CR , we have:
eiz M M M
≤ ≤ = 2 ,
1 + z2 |1 + z 2 | 2
|z| − 1 R −1

where M is an upper bound of eiz on the upper-half plane. Therefore, by Lemma

3.6, we get the estimate:
Z
eiz M
2
dz ≤ πR · 2 → 0 as R → +∞.
CR 1 + z R −1
Therefore, we get:
Z I Z
lim f (z) dz = lim f (z) dz − lim f (z) dz
R→∞ LR R→∞ γR R→∞ CR
Z ∞
cos x + i sin x π π
dx = − 0 = .
−∞ 1 + x2 e e
This shows: Z ∞
cos x π
= .
−∞ 1 + x2 e

Before we give another example, we recall some fundamental facts that:

• For any z 6= 0, the principal argument Arg(z) is in (−π, π].
• Log(z) = ln |z| + iArg(z) for any z 6= 0
• Log(z) is holomorphic on C\(−∞, 0].
Therefore, if we apply Cauchy’s integral formula or residue theory for an integral involving
Log(z), then we need to make sure the closed curve γ lies in C\(−∞, 0]. As such, we
cannot apply residue methods with a semi-circle contour as in the previous example.
Nonetheless, this kind of semi-circle contour is very useful when dealing with real
integrals over (−∞, ∞).
To get around with this issue, we can define a different branch of logarithm by the
following. For any z 6= 0, we let
(4.5) Log−π/2 (z) := ln |z| + iθ(z)

where θ(z) is the unique angle in [− π2 , 3π 2 ) such that z = |z| e

iθ(z)
. By doing so, we
Log−π/2 (z)
still have e = z. The notable difference from Log(z) is that now Log−π/2 (z) is
holomorphic on C\{0 + yi : y ≤ 0}, the yellow region below.

Exercise 4.29. Determine the value of Log−π/2 (z) when:

(a) z = i
(b) z = x + 0i where x > 0
(c) z = x + 0i where x < 0
124 4. Taylor and Laurent Series

CR
i
Cε
x
−R −ε ε R

Figure 4.1. Domain of Log−π/2 (z)

Example 4.14. Let α be a real constant in (0, 1). Evaluate the real integral:
Z ∞
1
I := dx.
0 x (1 + x2 )
α

Solution
First observe that for any x > 0, we have:
xα = eα ln x = eαLog−π/2 (x)
where Log−π/2 is the special branch of logarithm defined in (4.5). It prompts us to
consider a contour integral of the function:
1
f (z) = αLog (z)
.
e −π/2 (1 + z 2 )
We pick a contour as shown in Figure 4.1, where CR and Cε are semi-circles with
radii R and ε respectively. Since the closed contour γR,ε := [−R, −ε] + Cε + [ε, R] +
CR lies completely inside the domain of Log−π/2 (z), by Residue Theorem, we have:
I
1 1
αLog−π/2 (z)
dz = 2πi Res(f, i) = 2πi · αLog
2 −π/2 (z) (z + i)
γR,ε e (1 + z ) e z=i
2πi π
= π = απ i .
2ieα(ln|i|+ 2 i) e2
On the other hand, the γR,ε -integral can break down into:
(4.6)
I Z −ε Z Z R Z !
1 1
αLog−π/2 (z)
dz = + + + αLog−π/2 (z)
dz
2 (1 + z 2 )
γR,ε e (1 + z ) −R Cε ε CR e
When z = x + 0i ∈ [ε, R], the integrand is simply:
1 1
= α .
e αLog−π/2 (x) 2
(1 + x ) x (1 + x2 )
Hence,
Z R Z ∞
1 1
lim lim dz = dx =: I.
ε→0 R→∞ ε eαLog−π/2 (z) (1 + z 2 ) 0 xα (1 + x2 )
4.3. Residue Calculus 125

When z = x + 0i ∈ [−R, −ε], the integrand becomes:

1 1 1 1
= α(ln|x|+πi) = απi · α .
eαLog−π/2 (x) (1 + x2 ) e (1 + x2 ) e |x| (1 + x2 )
Hence,
Z −ε Z 0
1 1 1 I
lim lim dz = α dx = απi .
ε→0 R→∞ −R e αLog−π/2 (z)
(1 + z2) eαπi −∞ |x| (1 + x 2) e
| {z }
even function

We are left to analyze the two semi-circular integrals. We will show that they
tend to 0 as ε → 0 and R → ∞.
When z ∈ Cε , we have:
1 1 e−α ln ε ε−α
≤ e−α(ln|z|+iθ(z)) · = = 2
= .
e αLog−π/2 (z)
(1 + z 2 ) 1 − |z|
2 1−ε 1 − ε2

By Lemma 3.6, we get:

Z
1 ε−α πε1−α
dz ≤ · πε = →0 as ε → 0.
Cε eαLog−π/2 (z) (1 + z 2 ) 1 − ε2 1 − ε2

Similarly when z ∈ CR , we have:

1 1 R−α
≤ e−α(ln|z|+iθ(z)) · = .
eαLog−π/2 (z) (1 + z 2 ) 1 − |z|
2 R2 − 1

By Lemma 3.6, we have the estimate:

Z
1 R−α πR1−α
dz ≤ · πR = →0 as R → ∞.
Cε e
αLog−π/2 (z)
(1 + z 2 ) R2 − 1 R2 − 1

Finally, by letting ε → 0 and R → ∞ on both sides of (4.6), we get:

π I
απ = I + απi .
e2i e
Solving for I, we get:
π π π π απ
I = απ i −απi
= απ i − απ
i
= απ = sec .
e 2 (1 + e ) e 2 +e 2 2 cos 2 2 2

Exercise 4.30. Show that for any t ∈ R, we have:

Z ∞
eitx
2
dx = πe−|t| .
−∞ x + 1

Exercise 4.31. Show that:

Z ∞
1 (2n − 1)!!
dx = π.
−∞ (1 + x2 )n+1 (2n)!!

Exercise 4.32. Evaluate the following real integrals using residue methods:
Z 2π
1
(a) dθ where a > b > 0.
0 (a + b cos θ)2
126 4. Taylor and Laurent Series

Z 2π
1
(b) dθ where a ∈ R and a 6= ±1.
0 1 − 2a cos θ + a2
Z ∞
x2
(c) dx where a > 0.
(x + a2 )2
2
Z0 ∞
1
(d) dx where n ∈ N.
(x2 + 1)n
Z0 ∞
cos ax
(e) dx where a and b are positive real numbers
x 2 + b2
Z0 ∞
sin ax
(f) dx where a is a positive real number.
x(x2 + 1)
Z0 ∞
ln x
(g) dx where a > 0.
x2 + a2
Z0 ∞
1
(h) dx where α ∈ (0, 1).
0 xα (1 + x4 )

4.3.5. Argument Principle and Rouche’s Theorem. The Argument Principle deals
with functions of the following type:

Definition 4.11 (Meromorphic Functions). Let Ω be an open domain in C. A function

f : Ω → C is said to be meromorphic on Ω if f is holomorphic on Ω\S where S is a
discrete set of isolated singularities in Ω which are all poles of f .

We have seen a lot of examples of meromorphic functions already. All rational

functions (fractions of polynomials) are meromorphic.
The Argument Principle concerns about the following integral of type:
I 0
f (z)
dz,
γ f (z)

where f is a meromorphic function on a simply-connected open domain Ω containing the

simple closed curve γ. We assume that the zeros and poles of f do not lie on the curve
γ. Since γ is simple closed, one can parametrize it by a circle, i.e. z = γ(eit ), where
0 ≤ t ≤ 2π. Then, one can rewrite the above integral as:
Z 2π 0
f (γ(eit )) 0 it it
γ (e ) ie dt.
0 f (γ(eit ))
It can be verified that it is the same as the integral over the image curve f ◦ γ:
I
1
dw,
f ◦γ w

as one can parametrize f ◦ γ by w = f (γ(eit )), t ∈ [0, 2π]. This integral (up to a factor of
2πi) measures the winding number of the closed curve f ◦ γ around 0.

Exercise 4.33. Verify the above claim that

I 0 I
f (z) 1
dz = dw.
γ f (z) f ◦γ w

The Argument Principle is a nice formula relating this integral and the orders of
poles and zeros of f enclosed by γ. The order of a zero z0 of f is defined to be the order
of the pole z0 of f1 .
4.3. Residue Calculus 127

Theorem 4.12 (Argument Principle). Let f be a meromorphic function on a simply-

connected open domain Ω. Suppose γ is a simple closed curve in Ω not passing through
any poles and zeros of f . Denote by {α1 , · · · , αM } the zeros of f enclosed by γ, and
{β1 , · · · , βN } the poles of f enclosed by γ. Then, we have:
I 0 XM N
X
1 f (z)
(4.7) dz = ord(αj ) − ord(βk )
2πi γ f (z) j=1 k=1

where ord means the order of the poles and zeros.

Proof. First define a function F such that:

QM ord(αj )
j=1 (z − αj )
f (z) = QN · F (z).
ord(βk )
k=1 (z − βk )

One can check that F is holomorphic on the region enclosed by γ, by showing that all
αj ’s and βk ’s are removable singularities of F :

QN
(z − βk )ord(βk )
F (z) =f (z) · Qk=1
M
j=1 (z − αj )ord(αj )
QN
(αm − βk )ord(βk ) f (z)
lim F (z) = QM k=1 · lim ord(αm )
z→αm
j=1,j6=m (αm − αj )ord(αj ) z→αm (z − αm )
QN ord(βk )
k=1,k6=n (βn − βk ) ord(βn )
lim F (z) = QM · lim f (z) · (z − βn ) .
ord(αj )
j=1 (βn − αj )
z→βn z→βn

ord(βn ) (z−αm )ord(αm )

By the definition of pole, we have lim f (z)·(z − βn ) and lim f (z) exists
z→βn z→αm
and not equal to zero for all 0 ≤ n ≤ N , 0 ≤ m ≤ M . That implies both limits lim F (z)
z→αm
and lim F (z) exist, so F (z) has removable singularity at all αm and βn , which means
z→βn
F (z) can be extended to a holomorphic function near every pole and zero of f enclosed
by γ. Similar argument also shows F is non-zero at every point enclosed by γ.
0 QM
Next, we related ff with the poles and zeros of f . Let g(z) = j=1 (z − αj )ord(αj )
QN
and h(z) = k=1 (z − βk )ord(βk ) . Then, by direct computations:
g(z)
f (z) = · F (z)
h(z)
 0 
f 0 (z) g (z)F (z) g(z)F (z)h0 (z) g(z)F 0 (z) h(z)
= − + ·
f (z) h(z) h(z)2 h(z) g(z)F (z)
g 0 (z) h0 (z) F 0 (z)
= − + .
g(z) h(z) F (z)
By product rule, we get:
PM  ord(αm )−1
QM ord(αj )

0
g (z) m=1 ord(α m )(z − αm ) j=1,j6=m (z − αj )
= QM
g(z) j=1 (z − αj )
ord(αj)

M
X ord(αm )
=
j=1
z − αm
128 4. Taylor and Laurent Series

Similarly, we have:
N
h0 (z) X ord(βk )
= .
h(z) z − βk
k=1
Combining both results, we conclude that:
M N
f 0 (z) X ord(αj ) X ord(βk ) F 0 (z)
= − +
f (z) j=1
z − αj z − βk F (z)
k=1

Finally, the desired result (4.7) follows from the Cauchy’s Integral Formula and the fact
0
that FF is holomorphic on the region enclosed by γ.

Remark 4.13. The Argument Principle formula (4.7) can be generalized to the famous
Riemann-Roch’s Theorem for Riemann surfaces.

The Rouche’s Theorem below is closely related to the Argument Principle:

Theorem 4.14 (Rouche’s Theorem). Let f and h be meromorphic functions on a simply-

connected open domain Ω, and let γ be a simple closed curve in Ω which does not pass
through any pole and zero of f and h. Suppose |h(z)| < |f (z)| for any z ∈ γ, then we
have: I 0 I 0
f (z) + h0 (z) f (z)
dz = dz.
γ f (z) + h(z) γ f (z)

f (z) + h(z) h(z)

Proof. Define g(z) := =1+ . By our assumptions, g has no poles and
f (z) f (z)
h(z)
zeros on γ. Furthermore, for any z ∈ γ, we have < 1 and so g maps the curve γ
f (z)
to a closed curve inside the ball {w : |w − 1| < 1}. As the curve g ◦ γ does not enclose 0,
we have I
1
dw = 0,
g◦γ w
or equivalently, I 0
g (z)
dz = 0.
γ g(z)
By direct computations, we can check that
g 0 (z) f 0 (z) + h0 (z) f 0 (z)
= − .
g(z) f (z) + h(z) f (z)
It completes the proof. 
Remark 4.15. Combining the Rouche’s Theorem and the Argument Principle, one can
claim that if one add a small term h to a function f along a simple closed curve, then the
RHS of (4.7) is unchanged. It gives a powerful tool for us to count the number of zeros
and poles for a function, by just considering its dominant term.
4.3. Residue Calculus 129

Example 4.15. Show that all zeros of z 5 + 3z + 1 are inside the circle |z| = 2.

Solution
The dominant term of z 5 + 3z + 1 is z 5 , so we let f (z) = z 5 and h(z) = 3z + 1. On
the circle |z| = 2, we have
5
|f (z)| = |z| = 25
whereas
|h(z)| = 3 |z| + 1 = 7 < 25 = |f (z)| .
By Rouche’s Theorem, we have
I I
f 0 (z) f 0 (z) + h0 (z)
dz = dz
|z|=2 f (z) |z|=2 f (z) + h(z)

and by Argument Principle (noting that f and f + h have no poles), we conclude

that the sum of orders of zeros enclosed by |z| = 2 is the same for both f and
f + h = z 5 + 3z + 1.
The only zero for f (z) = z 5 is 0 which is of order 5, so f + h also has sum of
orders of zeros enclosed by |z| = 2 is 5, which is the maximum order for a 5-th
degree polynomial. Therefore, all zeros of z 5 + 3z + 1 are inside the circle |z| = 2.

Exercise 4.34. Show that z + ez + 3 has exactly one root with negative real part.

Exercise 4.35. Show that all roots of the equation z 6 + (1 + i)z + 1 = 0 lie in the
annulus 12 ≤ |z| < 54 .

The technique above gives another (yes, another one) proof of the Fundamental
Theorem of Algebra.

Proof #4 of Fundamental Theorem of Algebra. Consider the polynomial p(z) = z n +

an−1 z n−1 + · · · + a0 . The dominant term is z n , so we let f (z) = z n , and h(z) =
1/2 1/n
an−1 z n−1 + · · · + a0 . Pick R > max{1, n |an−1 | , n1/2 |an−2 | , · · · , n1/n |a0 | }, then
on |z| = R we have
|h(z)| ≤ |an−1 | Rn−1 + |an−2 | Rn−2 + · · · + |a1 | R + |a0 |
R R2 Rn
< · Rn−1 + · Rn−2 + · · · +
n n n
= Rn = |f (z)| .
By Rouche’s Theorem and Argument Principle, p(z) = h(z) + f (z) and f (z) have the
same sum of order of zeros inside |z| = R. Clearly, f (z) has a root 0 inside |z| = R, so
p(z) also has roots inside |z| = R.

 Chapter 5

What is the Riemann

Hypothesis?

5.1. Analytic Continuation

We will end this course by an introduction to the Riemann Hypothesis, a long-standing
unresolved problem in Pure Mathematics, and is a topic of central importance in Complex
Analysis, Number Theory, and related fields.
The Riemann Hypothesis concerns about the Riemann zeta function which is a priori
defined by the following infinite sum:

X∞ X∞
1 1
ζ(z) := z
= z ln n
.
n=1
n n=1
e

Here nz is regarded as a single-valued function of z. This sum converges absolutely on

the domain Ω = {z ∈ C : Re(z) > 1}, and converges uniformly on every smaller domain
Ωε = {z ∈ C : Re(z) > 1 + ε}. Therefore, Morera’s Theorem shows that ζ is holomorphic
on Ω.
Although ζ is a priori defined on Ω, we will soon learn that it can be extended to a
holomorphic function on C\{1}. In other words, there exists a function ζ̂ : C\{1} → C
such that ζ̂(z) = ζ(z) for any z ∈ Ω, and that ζ̂ is holomorphic on C\{1}. This new
function ζ̂ is called the analytic continuation of ζ.
Such an analytic continuation can be shown to be unique, and it is common to abuse
the notations a bit by simply writing ζ (instead of ζ̂) for the analytic continuation of ζ.
In this section, we will collect some useful facts about analytic continuations. We will
then describe how to extend ζ in the next section.

Definition 5.1 (Analytic Continuations). Given a holomorphic function f : Ω → C, a

function fˆ : Ω̂ → C defined on a connected domain Ω̂ ⊃ Ω is said to be an analytic
continuation of f on Ω̂ if:
• fˆ(z) = f (z) for any z ∈ Ω; and
• fˆ is holomorphic on Ω̂.

131
132 5. What is the Riemann Hypothesis?

While a (real) differentiable function defined on a smaller domain can be easily

extended to a (real) differentiable function defined on a larger domain, it is very difficult
to do so for a holomorphic function. One reason is that holomorphic functions are very
rigid, in a sense that if any two holomorphic functions coincide on an open set, then the
two function must be equal elsewhere! As a corollary, if an analytic continuation exists,
then it must be unique! Let’s state and prove this fact:

Theorem 5.2 (Identity Theorem). Let f : Ω → C be a holomorphic function on a

connected domain Ω. If there exists a non-empty open set U ⊂ Ω such that f (z) = 0 for
any z ∈ U , then f ≡ 0 on Ω.

Proof. Consider the set

S := {z ∈ Ω : f (n) (z) = 0 for any n ≥ 0}.
Since f (z) = 0 on U which is an open set, we have f (z) = f 0 (z) = f 00 (z) = · · · = 0 for
any z ∈ U . This shows U ⊂ S, and so S is non-empty. The proof goes by showing S is
both closed and open. Together with the fact that S is non-empty and Ω is connected, it
will prove S = Ω which implies our claim.
To show S is closed, we recall the fact that a holomorphic function f must be
infinitely differentiable, and hence f (n) are all continuous functions. The set S can be
written as:
\∞  −1
S := f (n) (0).
n=0

−1
The single set {0} is closed, and hence the pre-image f (n) (0) is closed for each n ≥ 0.
Since the intersection of any family of closed sets is closed, we conclude that S is closed.
To show S is open, we consider Taylor series expansions. For any z0 ∈ Ω, we consider
the Taylor series about z0 of f :
X∞
f (n) (z0 )
f (z) = (z − z0 )n
n=0
n!
which is defined on an open ball Bε (z0 ) for some ε > 0 (according to Taylor’s Theorem).
If z0 ∈ S, then we will have f (n) (z0 ) = 0 for any n ≥ 0, and as such, the above Taylor
series shows f (z) = 0 for any z ∈ Bε (z0 ). In other words, Bε (z0 ) ⊂ S. This shows S is
open.
Finally, S is non-empty, open and closed, and Ω is connected, so S = Ω. 

Corollary 5.3. Suppose g : Ω → C and h : Ω → C are two holomorphic functions defined

on a connected domain Ω, and that g and h coincide on a smaller open set U ⊂ Ω, then it
is necessary that g ≡ h on Ω.

Proof. Apply f := g − h to Identity Theorem. 

Exercise 5.1. Why is it necessary for f to be holomorphic in the proof of Identity

Theorem? Point out which part of the proof is no longer valid if f is just assumed to
smooth (differentiable for infinitely many times).
5.1. Analytic Continuation 133

Example 5.1. Consider the series:

∞
X
f (z) = zn
n=0
which converges pointwise on B1 (0), and uniformly on every smaller ball B1−ε (0)
where ε > 0. Therefore, f : B1 (0) → C is a holomorphic function on B1 (0).
On the other hand, the infinite sum is:
∞
X 1
zn = ,
n=0
1−z
1
and the function fˆ(z) = is defined on every z ∈ C\{1}, not only those in
1−z
B1 (0). Therefore, fˆ : C\{1} → C is the analytic continuation of f on C\{1}.

Exercise 5.2. What’s wrong with the following claim?

From fˆ(−1) = f (−1) (where f and fˆ are defined as in Example 5.1), we have:
∞
X 1 1
(−1)n = = .
n=0
1 − (−1) 2
Hence:
1
1 − 1 + 1 − 1 + 1 − 1 + ··· = .
2

Exercise 5.3. Consider the following function defined by the sum:

∞
X
1 1 1
f (z) = 1 + + 2 + · · · = .
z z n=0
zn
What is the largest possible domain on which f is holomorphic? Find the analytic
continuation of f on the larger domain C\{1}. Is it possible to further extend the
function to become an entire function on C?

Another common way of extending a holomorphic function is through a functional

equation. Let’s consider the following example. Suppose f : Ω → C is a holomorphic
function on Ω := {z ∈ C : Re(z) > 1}. If it can be shown that f satisfies an equation
such as:
f (z + 1) = 2f (z) for any z ∈ Ω,
then one can define an analytic continuation of it by the following way:
1
fˆ(z) := f (z + 1).
2
Since f (z + 1) is well-defined as long as z + 1 ∈ Ω, or equivalently, Re(z) > 0, the extend
function fˆ(z) is now defined on a larger domain Ω̂ := {z ∈ C : Re(z) > 0}. Note that
fˆ(z) = 12 f (z + 1) is holomorphic on {Re(z) > 0} since f is so on {Re(z) > 1}. Also,
when Re(z) > 1, we have
1
fˆ(z) = f (z + 1) = f (z)
2
by the given functional equation. Therefore, fˆ is the analytic continuation of f on
{Re(z) > 0}.
134 5. What is the Riemann Hypothesis?

Furthermore, the same functional equation holds for fˆ. Let’s verify this. For any z
such that Re(z) > 0, we have:
1 1
fˆ(z + 1) − 2fˆ(z) = f (z + 2) − 2 · f (z + 1)
2 2
1
= (f (z + 2) − 2f (z + 1)) = 0.
2
Now that fˆ is holomorphic on {z : Re(z) > 0} and satisfies the functional equation
fˆ(z + 1) = 2fˆ(z).
One can then repeat the same procedure as before to extend fˆ to a holomorphic function
ˆ
fˆ defined on {z : Re(z) > −1}, which is given by:
ˆ 1
fˆ(z) = fˆ(z + 1), for any z ∈ {Re(z) > −1}.
2
Inductively, we can repeat the same procedure over and over again, and extend f to
a function F : C → C that is holomorphic on the whole complex plane C.
y y

z z+1
1 f (z
fˆ(z) 2
+ 1)

x x
1

fˆ(z) = f (z)
f is originally defined here fˆ(z) = 21 f (z + 1)

x
−1 0 1

f can be inductively extended to an entire function F

Exercise 5.4. Given that f : Ω → C is a holomorphic on Ω := {z : Re(z) > 1}, and

that it satisfies the relation f (z + 1) = zf (z) for any z ∈ Ω. Show that there is an
analytic continuation fˆ on C\{0, −1, −2, −3, · · · }. Classify the type of singularities
(pole, removable or essential singularity) of each non-positive integer −n for fˆ.
5.2. Riemann ζ Functions 135

5.2. Riemann ζ Functions

5.2.1. Analytic Continuation of Γ. In this section we discuss the Γ (Gamma) and
ζ (zeta) functions, as well as their analytic continuations. These two functions are closely
related. The Gamma function Γ : Ω → C is a priori defined on Ω := {z : Re(z) > 0} by:
Z ∞
Γ(z) := tz−1 e−t dt for Re(z) > 0.
0

We have already proved that it is holomorphic function in Example 3.14.

Using integration-by-parts, one can derive a functional equation for Γ which can be
used to extend Γ beyond the domain {Re(z) > 0}. For any Re(z) > 0, we consider:
Z ∞ Z ∞
Γ(z + 1) = tz e−t dt = tz d(−e−t )
0 0
Z ∞
 t=∞
= −tz e−t t=0 + e−t d(tz )
0
Z ∞
=0+ ztz−1 e−t dt
0
= zΓ(z).
t=∞
We leave the part [−tz e−t ]t=0 = 0 as an exercise for readers:

Exercise 5.5. Show that whenever Re(z) > 0, we have:

lim tz e−t = 0 and lim tz e−t = 0.
t→0+ t→∞

Exercise 5.6. Show that for any positive integer n, we have:

Γ(n) = (n − 1)!

From the functional equation Γ(z + 1) = zΓ(z), one can define:

1
Γ1 (z) := Γ(z + 1)
z
for any z such that z 6= 0 and Re(z + 1) > 0. Then, Γ1 is an holomorphic function on
{z : Re(z) > −1}\{0}, and when Re(z) > 0, we have Γ1 (z) = Γ(z). In other words, Γ1 is
an analytic continuation of Γ.
y y

z z+1
Γ1 (z) 1 Γ(z + 1)
z

x x
−1

Γ1 (z) = Γ(z)
Γ is originally defined here Γ1 (z) = z1 Γ(z + 1)
136 5. What is the Riemann Hypothesis?

The functional equation for Γ then induces a new functional equation for Γ1 . When-
ever Re(z) > −1, we have:
1
Γ1 (z + 1) = Γ(z + 2) (Definition of Γ1 )
z+1
1
= · (z + 1)Γ(z + 1) (Functional equation for Γ)
z+1
= Γ(z + 1) = zΓ1 (z) (Definition of Γ1 ).
Therefore, one can define:
1
Γ2 (z) := Γ1 (z + 1)
z
for any z ∈ C such that z + 1 is in the domain of Γ1 , i.e. Re(z) > −2 and z 6= −1. As
such, Γ2 is an analytic continuation of Γ1 (and hence of Γ) on {Re(z) > −2}\{0, −1}.
Repeat the above process indefinitely, one can define analytic continuations Γm on
{Re(z) > −m}\{0, −1, −2, · · · , −(m − 1)}, and eventually an analytic continuation Γ̂ of
Γ on the domain C\{0, −1, −2, −3, · · · }.
y

x
−2 −1

Γ can be inductively extended to Γ̂

Exercise 5.7. Show that for any integer m ≥ 1 and z in the domain of Γm , we have:
Γ(z + m)
Γm (z) =
z(z + 1) · · · (z + m − 1)

Exercise 5.8. Show that each non-positive integer −n is a simple pole of Γ̂, and
that:
(−1)n
Res(Γ̂, −n) = .
n!
Here is a summary of facts about the Gamma function:
• Γ is a priori defined on {z : Re(z) > 0}.
• By the relation Γ(z + 1) = zΓ(z), one can define an analytic continuation Γ̂ of Γ on
C\{0, −1, −2, −3, · · · }.
(−1)n
• Each non-positive integer −n is a simple pole of Γ̂, with residue equal to n! .
Recall that since the analytic continuation must be unique, some textbooks denote the
analytic continuation by simply Γ.
5.2. Riemann ζ Functions 137

Exercise 5.9. Show that when Re(z) > 0, the Gamma function can be decomposed
into: Z ∞
X∞
(−1)n
Γ(z) = + e−t tz−1 dt.
n=0
n!(z + n) 1
X∞
(−1)n
Show also that the infinite sum converges for any z 6= 0, −1, −2, −3, · · ·
n=0
n!(z + n)
Z ∞
and the integral e−t tz−1 dt is an entire function of z.
1

5.2.2. Relation between Γ and ζ. Recall that the Riemann zeta function ζ : {z :
Re(z) > 1} → C is defined by the infinite series:
X∞
1
ζ(z) =
n=1
nz

which converges when Re(z) > 1. The following lemma shows a relation between Γ and
ζ.

Lemma 5.4. For any z ∈ C such that Re(z) > 1, we have:

Z ∞ z−1
t
(5.1) ζ(z)Γ(z) = dt.
0 et − 1

Proof. The key step of the proof is the change of variables t = nτ in the integral that
defines Γ:
Z ∞ Z ∞
Γ(z) = tz−1 e−t dt = (nτ )z−1 e−nτ d(nτ )
0 0
Z ∞
= nz τ z−1 e−nτ dτ
0
Z ∞
1
Γ(z) = tz−1 e−nt dt.
nz 0

Here we have used the fact that τ is a dummy variable. Summing up over n, we get:
X∞ X∞ Z ∞
1
(5.2) z
Γ(z) = tz−1 e−nt dt.
n=1
n n=1 0

Next we want to switch the integral and summation signs. It has to be justified using
LDCT. Consider:
tz−1 e−nt ≤ tx−1 e−nt
for any t ∈ [0, ∞). Note that:
∞ Z
X ∞ X∞
1
tx−1 e−nt dt = Γ(x)
n=1 0 n=1
nx

which converges since x > 1. Hence, LDCT shows we can switch the summation and
integral signs of (5.2), and it yields:
X∞ Z ∞ X∞
1
z
Γ(z) = tz−1
e−nt dt.
n=1
n 0 n=1
138 5. What is the Riemann Hypothesis?

∞
X ∞
X
−nt

n
Observing that e = e−t is a geometric series, we get:
n=1 n=1
∞
X e−t 1
e−nt = = t .
n=1
1 − e−t e −1

From (5.2), we get our desired result (5.1). 

5.2.3. Analytic Continuation of ζ. The relation (5.1) will be used to extend ζ

beyond the domain {Re(z) > 1}. We have already shown that Γ can be extended to
almost all of C. If we are able to extend the integral:
Z ∞ z−1
t
t−1
dt
0 e
beyond {Re(z) > 1}, then ζ can also be extended accordingly.
First break down the integral into two part:
Z ∞ z−1 Z 1 z−1 Z ∞ z−1
t t t
t−1
dt = t−1
dt + t−1
dt.
0 e 0 e 1 e
The second integral is well-defined for any z ∈ C. To see this, we first note that
tx−1 et/2
tz−1 = tx−1  et/2 as t → ∞, and hence t  t ∼ e−t/2 . The function e−t/2
e −1 e − 1
Z ∞ z−1
t
is integrable over [1, ∞). By comparison, the integral dt is finite for any
1 et − 1
z ∈ C (not only those with Re(z) > 1). By Morera’s Theorem, the integral is an entire
function of z.
Z 1 z−1
t
Next we handle the first integral t−1
dt. The key trick is to consider the
0 e
1
denominator t , and expand it as a series. Consider the function:
e −1
1 1
f (w) = w − .
e −1 w
Although it is not defined when w = 0, we can see that 0 is a removable singularity:
!
1 1
lim f (w) = lim 2 3 −
w→0 w→0 w + w2! + w3! + · · · w
w2 w3
w−w− 2! − 3! − ···
= lim 2 3
w→0 w(w + w + w + ···)
2! 3!
− 12 − w6 − · · ·
= lim 2
w→0 1 + w + w + · · ·
2 6
1
=− .
2
Therefore, by declaring that f (0) = − 12 , it becomes a holomorphic function defined on
B2π (0) (why 2π?). Consider its Taylor series about 0:
1 f 00 (0) 2 f (3) (0) 3
f (w) = − + f 0 (0)w + w + w + ···
2 2! 3!
∞
X
1 1 f (n) (0) n
− = w .
ew − 1 w n=0 n!
5.2. Riemann ζ Functions 139

Substitute w = t ∈ [0, 1], then we get:

∞
1 1 X f (n) (0) n
= + t .
et − 1 t n=0 n!

X ∞
f (n) (0) n
Recall from Exercise 4.8 that the Taylor’s series w converges uniformly
n=0
n!
on every ball B2π−ε (0) slightly smaller than B2π (0), say B2 (0). In particular, since
X∞
f (n) (0) n
[0, 1] ⊂ B2 (0), the convergence of the series t is also uniform on [0, 1]. When
n=0
n!
z is a fixed complex number such that Re(z) > 1, we have tz−1 ≤ tx−1 ≤ 1. Therefore,
X∞
f (n) (0) n+z−1
the series t also converges uniformly on t ∈ [0, 1] regarding z as fixed.
n=0
n!
Using the fact, one can write the first integral as:
Z 1 z−1 Z 1 ∞
!
t 1 X f (n)
(0)
t
dt = tz−1 + tn dt
0 e −1 0 t n=0 n!
Z 1 ∞
!
X f (n) (0) z+n−1
z−2
= t + t dt
0 n=0
n!
 z−1 t=1 X ∞  t=1
t f (n) (0) tz+n
= +
z − 1 t=0 n=0 n! z + n t=0
∞
X f (n) (0)
1 1
= + · .
z − 1 n=0 n! z+n

Here we have integrated term-by-term thanks to uniform convergence of the series.

Z 1 z−1
t
Although the integral t−1
dt on the LHS is defined only when Re(z) > 1, the
0 e
RHS series is defined whenever z 6= 1, 0, −1, −2, −3, · · · . Furthermore, the RHS series
is holomorphic on Ω := C\{1, 0, −1, −2, −3, · · · }. To show this, it suffices to prove
X∞
f (n) (0) 1
· converges uniformly on any small ball Br (z0 ) ⊂ Ω. Note that the
n=0
n! z + n
singularities {1, 0, −1, −2, −3, · · · } are isolated, points in Br (z0 ) must be well away from
the singularities. There exists δ > 0 such that |z + n| ≥ δ for any z ∈ Br (z0 ) and
n = 0, 1, 2, 3, · · · . As a result, we have:
f (n) (0) 1 f (n) (0) 1
· ≤ · .
n! z+n n! δ
X∞
f (n) (0) 1
By Weierstrass’s M-test, the series · converges uniformly on any small
n=0
n! z+n
ball Br (z0 ) ⊂ Ω. By Morera’s Theorem, it defines a holomorphic function on any small
ball Br (z0 ) ⊂ Ω, and so is holomorphic on Ω.
Combining the result (5.1), we have so far established that on {Re(z) > 1}:
 
Z ∞ z−1
1  
∞
X
 1 f (n) (0) 1 t 
ζ(z) =  + · + t−1
dt
Γ(z)  z − 1 n=0 n! z+n 1 e 
| {z } | {z }
entire
extendable to Ω
140 5. What is the Riemann Hypothesis?

Since Γ has an analytic continuation Γ̂ on C\{0, −1, −2, −3, · · · }. From the above
relation, we can then define an analytic continuation of ζ on C\{1, 0, −1, −2, −3, · · · } as:
∞ Z ∞ z−1 !
1 1 X f (n) (0) 1 t
(5.3) ζ̂(z) = + · + dt
Γ̂(z) z − 1 n=0 n! z+n 1 et − 1
X∞ Z ∞ z−1
1 f (n) (0) 1 1 t
= + + dt.
(z − 1)Γ̂(z) n=0 n! (z + n)Γ̂(z) Γ̂(z) 1 et − 1

It appears (5.3) has singularities at every 1, 0, −1, −2, −3, · · · , yet we can show
0, −1, −2, −3, · · · are all removable. It is because Γ̂ has a simple pole at every of
1/Γ̂
{0, −1, −2, −3, · · · }, so they are zeros of 1/Γ̂. Therefore, z+n has a removable singularity
at −n. Precisely, for any integers m, n ∈ {0, 1, 2, 3, · · · } we have:
(
1
1 if m = n
lim = Res(Γ̂,−n)
z→−m (z + n)Γ̂(z) 0 if m 6= n

This shows {0, 1, 2, 3, · · · } are all removable singularities of ζ̂ since the following limit is
finite for any m = 0, 1, 2, 3, · · ·
f (m) (0) 1
lim ζ̂(z) =
z→−m m! Res(Γ̂, −m).
Therefore, ζ(z) can be holomorphically defined on C\{1} by declaring that
f (m) (0) 1
ζ̂(−m) :=
m! Res(Γ̂, −m)

for any m = 0, 1, 2, · · · . Note that 1 is a simple pole of ζ̂.

5.2.4. Special Values of ζ̂. We will determine the value of ζ̂ at some special z ∈ C.
When z = −m where −m is a non-positive integer, then we have already discussed that
f (m) (0) 1
ζ̂(−m) = .
m! Res(Γ̂, −m)
Here f is the function:
1 1
f (w) = − .
ew − 1 w
(−1)m
We have already figured out that Res(Γ̂, −m) = , so ζ̂(−m) = (−1)m f (m) (0).
m!
However, it is not straight-forward to find a general expression for f (m) (0), but by direct
computations one can verify that the first few terms of f (m) (0) are given as follows:
1 1 1
f (0) = − f 0 (0) = f 00 (0) = 0 f (3) (0) = − .
2 12 120
Therefore, the extended Riemann zeta function ζ̂ takes the following values:
1 1
ζ̂(0) = − ζ̂(−1) = −
2 12
1
ζ̂(−2) = 0 ζ̂(−3) =
120
1
To many people’s surprise, the fact that ζ̂(−1) = − 12 is used in String Theory!
However, many “muggles” misunderstand the meaning of it. They misinterpret it as
5.2. Riemann ζ Functions 141

P∞ 1 1
P∞
n=1 n−1 = − 12 , which is mathematically wrong as ζ̂(z) = n=1 n1z only when Re(z) >
1. It would lead to the following awkward and non-sense expression:

1
1 + 2 + 3 + 4 + ··· = − .
12
P∞
Similarly, those “muggles” also mix up ζ̂(0) = − 21 with n=1 n10 = − 12 , and ζ̂(−2) = 0
P∞ 1
with n=1 n−2 = 0, both would lead to awkward expressions:

1
1 + 1 + 1 + 1 + ··· = − .
2
12 + 22 + 32 + 42 + · · · = 0.

Exercise 5.10. Prove that for all positive integers n, one has
Bn+1
ζ̂(−n) = (−1)n
n+1
where Bn ’s are Bernoulli’s numbers defined as in Exercise 4.24.

Exercise 5.11. Suppose f : C → C is continuous on C, and holomorphic on BR (0)

for some R > 1. Consider the function
Z 1
F (z) := tz−1 f (t) dt.
0
We have already shown in Exercise 3.31 that F is holomorphic on {z : Re(z) > 0}.
Show that F has an analytic continuous to the domain O := C\{0, −1, −2, −3, · · · }.

5.2.5. Riemann Hypothesis. Finally, we are ready to understand the statement of

the Riemann Hypothesis. It is a conjecture about the zeros of the (extended) Riemann
zeta function ζ̂. To begin, let’s first recall that for any negative integer −m, we have:

ζ̂(−m) = (−1)m f (m) (0),

1 1
where f (w) = − . It is not difficult to show that f (m) (0) = 0 for any even
ew − 1 w
integer m:

Exercise 5.12. Show that

1
g(w) := f (w) +
2
is an odd function, and hence deduce that f (m) (0) = 0 for any even integer m ≥ 2.

Therefore, we have ζ̂(−2) = ζ̂(−4) = ζ̂(−6) = · · · = 0. These negative even integers

{−2, −4, −6, · · · } are called trivial zeros of ζ̂.
Any complex number z0 which is not a negative even integer is called a non-trivial
zero of ζ̂ whenever ζ̂(z0 ) = 0. The Riemann Hypothesis is concerned with the locations
of these non-trivial zeros. It is conjectured by Bernhard Riemann in 1859 that:

“All non-trivial zeros z0 of ζ̂ must have real part equal to 21 .”

142 5. What is the Riemann Hypothesis?

x
1
2

The zeros of ζ̂ have deep connections with the distribution of prime numbers. The
renowned Prime Number Theorem asserts that:
π(x)
lim =1
x→∞ x/ ln x

where π(x) is the number of positive prime numbers less than or equal to x. A corollary
of the theorem is that the n-th prime number pn is approximately equal to n ln n. The
proof of Prime Number Theorem relies surprisingly on the fact that there is no zero of
ζ̂ with real part equal to 1. If the Riemann Hypothesis is proven to be true, then the
Prime Number Theorem can be substantially improved, and many mysteries about the
distribution of primes will be revealed.
As of today (February 29, 2020), the mathematics community widely believes that the
conjecture is unsolved. It is one of the most important open problem in Pure Mathematics
nowadays. In 2000, the Clay Mathematics Institute compiled a list of 7 problems, called
Millennium Prize Problems. For each problem in the list, the institute promises to award
US$1,000,000 to the first person who solves or disproves it. Riemann Hypothesis is
one of the problems in the list. The other 6 problems are: P versus NP Problem, Hodge
Conjecture, Poincaré Conjecture, Yang-Mills Existence and Mass Gap, Navier-Stokes
Existence and Smoothness, and Birch Swinnerton-Dyer Conjecture. The only Millennium
Prize Problem that was solved is the Poincaré Conjecture, by Grigori Perelman in 2002-03
using the idea of Ricci flow developed by Richard Hamilton in 1982.
* End of MATH 4023 *
** I hope you have learned a lot and/or enjoyed the course. **
 Appendix A

Results from MATH 2043/3033

In this appendix we list some important concepts and theorems from MATH 2043/3033
that we will use frequently in this course. Proofs are all omitted since they are essentially
the same as in the real case. This appendix is intended to be brief (no worked example
here). For detail, please consult Chapter 10 of MATH 3033, or Chapter 4 in MATH 2043.

Definition A.1 (Uniform Convergence). A sequence of functions fn (z) is said to con-

verge to f (z) uniformly on Ω if
sup{|fn (z) − f (z)| : z ∈ Ω} → 0 as n → ∞.
X∞
A (pointwise) convergent series fn (z) is said to converge uniformly on Ω if the N -th
n=1
N
X
partial sum fn (z) converges uniformly on Ω as N → ∞. In other words:
n=1
( N ∞
)
X X
sup fn (z) − fn (z) : z ∈ Ω →0 as N → ∞.
n=1 n=1

It is sometimes difficult to show a series converges uniformly from the definition.

Fortunately, we have the following useful test:
∞
X
Theorem A.2 (Weierstrass’ M-test). Consider a series fn (z) defined on Ω. If there
n=1
exists a sequence of real numbers Mn ∈ R, independent of z, such that:
• |fn (z)| ≤ Mn for any z ∈ Ω and any n, and
∞
X
• the series Mn converges,
n=1
∞
X
then fn (z) converges uniformly on Ω.
n=1

143
144 A. Results from MATH 2043/3033

There are many nice consequences if a series or sequence converges uniformly,

namely we can switch the integral, limit and summation signs quite freely:

Proposition A.3. Suppose fn (z) converges uniformly on Ω to the limit function f (z),
then:
• If fn are continuous on Ω for all n, then f is also continuous on Ω.
• For any α ∈ Ω, we have
lim lim fn (z) = lim lim fn (z).
z→α n→∞ n→∞ z→α
• Let [a, b] be a bounded interval in R, and fn (t)’s be integrable functions on [a, b] If
fn (t) converges uniformly to f (t) on [a, b], then
Z b Z b
lim fn (t) dt = lim fn (t) dt.
n→∞ a a n→∞
• Let γ be a curve in C of finite length, and fn (z)’s be integrable functions on γ. If fn (z)
converges uniformly to f (z) on Ω, then
Z Z
lim fn (z) dz = lim fn (z) dz.
n→∞ γ γ n→∞
∞
X
Analogous results hold for uniform convergence series. For instance, if fn (z) converges
n=1
uniformly on Ω, then for any curve γ in Ω of finite length, we have:
Z X∞ X∞ Z
fn (z) dz = fn (z) dz.
γ n=1 n=1 γ

In the above proposition, the conditions that [a, b] is a finite interval, and γ is
a curve of finite length are necessary. For unbounded intervals or curves, uniform
convergence is not sufficient to guarantee the switching of the integral and summation
signs! Fortunately, there is another tool to deal with improper integrals, namely Lebesgue
Dominated Convergence Theorem (LDCT), which stems from measure theory:

Theorem A.4 (Lebesgue Dominated Convergence Theorem). Let fn (t) : (a, b) → C be

a sequence of measurable functions (including continuous functions) defined on a possibly
infinite interval (a, b) ⊂ R. Suppose:
• fn (t) → f (t) pointwise on every t ∈ (a, b), and
• there exists an integrable function h : (a, b) → R independent of n such that
|fn (t)| ≤ h(t) for any t ∈ (a, b) and any n,
then we have Z Z
b b
lim fn (t) dt = lim fn (t) dt.
n→∞ a a n→∞
∞
X
Consider a series gn (t) where gn : (a, b) → C are measurable. Suppose
n=1
∞ Z
X b
|gn (t)| dt < ∞,
n=1 a

then we have: Z ∞
bX ∞ Z
X b
gn (t) dt = gn (t) dt.
a n=1 n=1 a
A. Results from MATH 2043/3033 145

Recall from MATH 3033/2043 that even if fn (x) converges uniformly on (a, b) to
f (x), the derivatives fn0 (x) may not converge to f 0 . Sometimes,
P∞ the limit of fn0 may not
even be differentiable. Likewise, even when the sum n=1 fn (x) converges uniformly
on (a, b), term-by-term differentiation
∞ ∞
d X X
fn (x) = fn0 (x)
dx n=1 n=1
may not hold. Instead, what is needed is pointwise convergence of fn (z), and uniform
convergence of fn0 (z):

Proposition A.5. Let fn (z) : Ω → C be a sequence of functions defined on an open

domain Ω ⊂ C, and suppose
• fn (z) converges pointwise to a function f (z) on Ω; and
• fn0 converges uniformly on Ω to a function g(z).
Then, f 0 (z) = g(z) on Ω. In other words:
d
lim fn (z) = lim fn0 (z).
dz n→∞ n→∞
An analogous result holds for series of functions: suppose
∞
X
• fn (z) converges pointwise on Ω; and
n=1
X∞
• fn0 (z) converges uniformly on Ω.
n=1

Then, we have
∞ ∞
d X X d
fn (z) = fn (z).
dz n=1 n=1
dz

When using Morera’s Theorem, we often consider a double integral of the form:
I Z b
f (z, t) dtdz.
T a
It we can switch the two integral signs, and it happens that f (z, t) is a holomorphic
function for each fixed t ∈ [a, b], then we have:
I Z b Z bI Z b
f (z, t) dtdz = f (z, t) dzdt = 0 dt = 0.
T a a T a
The question is whether we can switch the two integral signs. It thanks for the following
(special case) of Fubini’s Theorem

Theorem A.6 (Fubini-Tonelli’s Theorem). Suppose f (z, t) : Ω × I → C is a continuous

function, where Ω ⊂ C and I is an interval (possibly infinite) in R. Let γ be a curve in Ω.
If one of the following is finite:
Z Z Z Z
|f (z, t)| dt |dz| or |f (z, t)| |dz| dt.
γ I I γ
Then, we have: Z Z Z Z
f (z, t) dtdz = f (z, t) dzdt.
γ I I γ
p
Here |dz| means (dx)2 + (dy)2 .