Applied Complex Analysis

Willie Rush Lim

2020
ii
 1

Introduction
Denote the set of complex numbers by

C = {x + iy | x, y ∈ R}
√
where i = −1 is defined such that i2 = −1.
Complex analysis is the study of functions of a complex variable. In the
first few chapters, we shall explore some introductory concepts, such as basic
properties of complex numbers and continuity of complex-valued functions.
The main emphasis is the concept of holomorphic functions, i.e. complex-
valued functions which are differentiable in a complex sense, and the many
applications of their somewhat magical properties. I used the word ’magical’
because holomorphicity is such a rigid condition that many of the results you
will see are somewhat unintuitive yet true.
We will start with some motivation. Basic algebra tells us that the num-
ber of roots of a polynomial with real coefficients is at most its degree. For
example, x2 + c has two real roots if c < 0, one root if c = 0, and no roots if
c > 0. Introducing the imaginary number i provides us with a more elegant
way of formulating this idea.

Theorem (Fundamental Theorem of Algebra). The field C is algebraically

closed, that is, any polynomial with coefficients in C of degree d > 1 has
exactly d roots in C, counting multiplicity.

Many initial attempts of proving the theorem by prominent mathemati-

cians D’Alembert, Euler, Gauss, Lagrange, and Laplace in 1700s were incom-
plete. In 1806, a Swiss accountant, Parisian bookstore manager and ’ama-
teur’ mathematician Jean-Robert Argand completed D’Alembert’s ideas and
hence became the first person to rigorously prove the fundamental theorem
of algebra. We will in fact use properties of holomorphic functions to give
3 different proofs of the theorem, including D’Alembert and Argand’s ap-
proach.
It is difficult to list the many applications of the fundamental theorem of
algebra. The main idea is that the field of complex numbers is the perfect
setting to solve equations!
A direct consequence in linear algebra is that every square matrix with
entries in C admits an eigenvalue. When a 2 × 2 matrix has imaginary eigen-
values, it acts as a rotation of the plane rather than expansion or contraction
in certain directions. In the study of continuous dynamics arising from me-
chanical systems, it is common to use complex numbers in order to capture
oscillations in the system.
2

One of the direct applications of the study of holomorphic functions is

contour integration. The integral of a complex function along a closed path is
not dependent on the path itself but rather on certain values called residues
of the function’s singularities. This means that it is often easier to integrate
a real function of a real variable by converting it into a problem involving a
contour integral in the complex plane.
Fourier series and Fourier transforms are useful in decomposing functions
into its frequency components. (Think of decomposing nice functions as a
sum or an integral of different sine and cosine waves.) Fourier analysis can
be easily formulated via complex analysis, and it comes up everywhere: in
differential equations, probability, quantum mechanics, signal processing, etc.
Mechanical and electrical engineers as well as computer musicians also
encounter complex variables in electrical circuits with alternating current.
Digital filters are designed by looking at the locations of zeros and poles of
rational functions called transfer functions, which essentially model a device’s
inputs and outputs.
Iterations of holomorphic functions have long been known to have many
applications. Complex polynomials, for example, can be used to model the
population of rabbits over time. Powerful basic results in complex analysis,
many of which do not apply to generic real differentiable functions, make up
one of the many reasons why the study of iterations of holomorphic func-
tions (holomorphic dynamics) is very well developed compared to the other
branches of the field of dynamical systems.
Conformal functions are holomorphic functions with strictly non-zero
derivative. Such functions have an amazing geometric property of angle
preservation at every point and are useful in transforming regions with com-
plicated boundary to those of a much nicer shape (square, disk, etc). You
may, for example, want to transform a mechanical problem on a complicated
domain into an equivalent problem on a circular disk. In cartography, confor-
mal maps are useful in creating a world map as well as local nautical charts
using Mercator and stereographic projections. More recently, conformal func-
tions are applied to the surface of the human brain for brain development
study and diagnosis of Alzheimer’s disease and schizophrenia.
Chapter 1

Complex Numbers

In this chapter, we will go through the basic algebraic and geometric prop-
erties of complex numbers.

1.1 The Algebra of C

The set C is equipped with the usual arithmetic operators, namely:

• addition +: (x + iy) + (a + ib) = (x + a) + i(y + b),

• multiplication ×: (x + iy) × (a + ib) = (xa − yb) + i(xb + ya).

Let’s denote by C∗ the set of non-zero complex numbers C\{0}. This set
is equipped with an additional operator:
x−iy
• inversion of a non-zero number: (x + iy)−1 = x2 +y 2
.

Similar to R, the set of complex numbers C is a field ; it satisfies the

following axioms:

1. (C, +) is an abelian group:

• + is associative and commutative,

• 0 is the identity element of +, i.e. z + 0 = z for all z ∈ C,
• Additive inverses exist, i.e. z + (−z) = 0 for all z ∈ C;

2. (C∗ , ×) is an abelian group:

• × is associative and commutative,

• 1 is the identity element of ×, i.e. z × 1 = z for all z ∈ C∗ ,

3
4 CHAPTER 1. COMPLEX NUMBERS

• Multiplicative inverses exist, i.e. z × z −1 = 1 for all z ∈ C∗ ;

3. × is distributive over +.

The set C of complex numbers can be identified with the real vector space
2
R by the vector space isomorphism:

C → R2 ,
z 7→ (Re z, Im z),
x + iy 7→ (x, y).

Unlike C, the real plane R2 is only equipped with addition operator +

but not a natural multiplication operator ×. Nonetheless, the mapping above
allows us to geometrically represent complex numbers as points on the plane.
This is typically known as Argand diagram.

1.2 The Geometry of C

Every complex number z = x + iy comes with a unique real part x and
imaginary part y. We shall denote them as follows:

Re z = x, Im z = y.

Geometrically, Re and Im can be thought as functions C → R acting as

projections onto the real and imaginary axes respectively.
The complex conjugate z̄ of a complex number z = x + iy is z̄ = x − iy.
Geometrically, the operation z 7→ z̄ is a reflection over the real axis. The
following identity can be thought of as a change of basis from (x, y) to (z, z̄).
z+z̄ z−z̄
Proposition 1.1. For any z ∈ C, Re z = 2
and Im z = 2i
.

Another straightforward algebraic exercise also gives us the following ba-

sic properties.

Proposition 1.2. For any z, w ∈ C, z + w = z̄ + w̄ and zw = z̄ w̄. If z 6= 0,

z −1 = z̄ −1 .

The absolute value / modulus of a complex number z = x + iy is

p
|z| = x2 + y 2 .

Phytagoras’ theorem indicates that geometrically the modulus |z| of z is

equal to the distance between 0 and z.
1.2. THE GEOMETRY OF C 5

Proposition 1.3. For any z, w ∈ C,

• |zw| = |z||w|,

• z z̄ = |z|2 ,

• |z + w| ≤ |z| + |w| (Triangle inequality).

The argument of z, arg(z), is defined to be the counterclockwise angle
(measured in radians) subtended by the positive real axis R+ and the line
segment joining 0 and z. See figure 1.1.
Notice that arg is a multivalued function. For example, both π and 3π
are arguments of i. We can refine this by defining the principal argument of
z, Arg(z), to be the unique argument of z lying in (−π, π].
Remark. The interval [0, 2π) is also often chosen to be the codomain of the
principal argument.
Proposition 1.4. For any z, w ∈ C∗ ,
• arg(zw) = arg(z) + arg(w),

• Arg(zw) = Arg(z) + Arg(w) mod 2π.

√
Example 1. Let z = 1 + i and w = −1 + 3i. The modulus and arguments
of z and w are:
√ π 2π
|z| = 2, |w| = 2, arg(z) = , arg w = .
4 3
√ √
Then, the modulus and argument of (1 + i)(−1 + 3i) are 2 2 and 11π
12
respectively.
For any non-zero complex number z = x + iy, if r = |z| and θ = Arg(z),
then basic trigonometry gives us the following change of variables:

x = r cos θ, y = r sin θ.

The expression z = r(cos θ + i sin θ) from above is the polar form of z.

Theorem 1.5 (Euler’s formula). For any θ, eiθ = cos θ + i sin θ.
Proof. We will give two different proofs of the result - one with differential
equations, and another with Maclaurin series. The expression eiθ is a non-
zero complex number, so there is a unique r > 0 and θ̂ such that

eiθ = r(cos θ̂ + i sin θ̂). (1.1)

6 CHAPTER 1. COMPLEX NUMBERS

Figure 1.1: A point z = x + iy = reiθ on the Argand diagram

Here, r and θ̂ are functions of θ. When θ = 0, r(0) = 1 and θ̂(0) = 0.

Differentiating (1.1) with respect to θ, we obtain

dr dθ̂
ieiθ = (cos θ̂ + i sin θ̂) + r (− sin θ̂ + i cos θ̂)
dθ dθ
iθ
dr e dθ̂
= + i eiθ
dθ r dθ!
dr dθ̂
= +i eiθ ,
dθ dθ

where the second equality above is obtained from (1.1). From above, we see
dr
that dθ = 0 and ddθθ̂ = 1. By our initial conditions, we obtain r(θ) ≡ 1 and
θ̂ ≡ θ.
(iθ)n
Alternatively, recall the following Maclaurin series: eiθ = ∞
P
n=0 n! . Us-
ing the fact that in = (−1)n/2 if n is even, and in = (−1)n−1/2 i if n is odd,

X (−1)n θn X (−1)n−1/2 θn
eiθ = +
even n
n! odd n
n!
θ2 θ4 θ3 θ5
   
= 1− + − ... + i θ − + − . . . = cos θ + i sin θ.
2! 4! 3! 5!

Example 2. When θ = π, we have Euler’s identity: eiπ = −1.

1.3. COMPLEX ROOTS 7

The polar form of a complex number z can alternatively be written in

the form of z = reiθ . This expression is particularly useful when performing
multiplication of complex numbers as we can use the laws of exponent. One
particular instance is the following.
Theorem 1.6 (De Moivre’s Theorem). For any θ and integer n ∈ Z,

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

 √ 10
Example 3. To compute and simplify − 21 + 23i , we can use De Moivre’s
2π
theorem. The term inside the bracket is essentially cos θ+i sin θ where θ = 3
.
Then,
√ !10     √
1 3i 20π 20π 1 3i
− + = cos + i sin =− + .
2 2 3 3 2 2

1.3 Complex Roots

Consider a complex number z0 and a positive integer n. A complex number
w satisfying wn = z0 is called an nth root of z0 .
Suppose z0 = 0. Regardless of n, there is only one root of 0, which is 0
itself. This is due to the fact that C is an integral domain, i.e. for any two
complex numbers z1 and z2 , if z1 z2 = 0 then either z1 = 0 or z2 = 0.
Suppose z 6= 0 now, then surely any root w is also non-zero. Using their
polar forms z = reiθ and w = seit , then the equation becomes:

sn eint = reiθ

Considering the modulus and the argument independently, we obtain two

real equations sn = r and nt = θ mod 2π. There are therefore n different
solutions to w:

wk = r1/n ei(θ+2πk)/n , k ∈ {0, 1, . . . n − 1}.

In the expression above, w0 is called the principal root of z0 . On the

complex plane, these roots are evenly spaced on the circle {z ∈ C | |z| = r1/n }
of radius r1/n centered at the origin.
When z0 = 1, the nth roots of 1 are called the nth roots of unity. They all
lie on the unit circle and are of the form e2πik/n , where k ∈ {0, 1, . . . n − 1}.
Example 4. The 3rd roots of √unity are 1, √e2πi/3 and e4πi/3 . The Cartesian
forms of these roots are 1, −1+i
2
3
, and −1−i
2
3
.
8 CHAPTER 1. COMPLEX NUMBERS

Figure 1.2: 3rd roots of unity

1.4 The Topology of C

An open disk of radius r > 0 centred at a complex number a ∈ C is a subset
of C of the form:
D(z, r) = {z ∈ C | |z − a| < r}.
The boundary of this disk is a circle of of radius r > 0 centred at a,
denoted with a partial sign in front:

C(z, r) = ∂D(z, r) = {z ∈ C | |z − a| = r}.

If we include the boundary, we obtain a closed disk typically denoted with

an overline:
D(z, r) = {z ∈ C | |z − a| ≤ r}.
Example 5. Let’s consider the sets

A = {reiθ | r = sin θ, θ ∈ R}, B = {reiθ | 0 < r < sin θ, θ ∈ R}.

If z = x + iy lies in A, then x = sin θ cos θ and y = sin2 θ for some θ. By

double angle formulas,

sin2 (2θ) + cos2 (2θ) = (2x)2 + (1 − 2y)2 = 1.

This equation represents a circle of radius 21 centered at 2i . Therefore, A is

the circle C( 2i , 12 ). For points z on the set B, we only need to consider the
case when sin θ > 0, or principally when 0 < θ < π. The set B is the open
disk D( 2i , 12 ).
1.4. THE TOPOLOGY OF C 9

The geometric and topological properties of the complex plane C are

essentially the same as those of the real plane R2 since we have the obvious
identification x + iy 7→ (x, y). We will give a brief introduction of necessary
topological terminology that we will use in the next few chapters.

Definition 1. A subset S ⊂ C is:

• open if for every s ∈ S, there is some r > 0 such that D(s, r) ⊂ S,

• closed if its complement C\S is open,

• bounded if there is some r > 0 where S ⊂ D(0, r),

• compact if S is closed and bounded.

Example 6. Below are some subsets of C which we will commonly encounter.

1. The empty set ∅ is trivially open and compact.

2. The complex plane C is both open and closed, but not bounded.

3. The punctured plane C∗ = C\{0} is open, but not closed nor bounded.

4. The unit disk D := D(0, 1) is open and bounded, but not closed.

5. The closed unit disk D and its boundary ∂D are compact.

6. The real axis R is closed and unbounded.

Definition 2. An open/closed set S ⊂ C is:

• connected if S cannot be expressed as a disjoint union of two open/closed

non-empty subsets of C,

• simply connected if it is connected and it has no ”holes”, i.e. the

complement C\S has no bounded connected component,

• multiply connected if it is connected but not simply connected.

We say that S is a domain if it is a non-empty open connected subset of C.

10 CHAPTER 1. COMPLEX NUMBERS

Figure 1.3: Four connected subsets of C. Solid boundary lines are included
in the colored set, whereas dashed boundary lines are not included. The
first (from the left) is a simply connected domain. The second is closed
and simply connected. The third is a punctured disk, which is a multiply
connected domain. The last is closed and multiply connected.

Example 7.
1. ∅, C, D, D and R are simply connected.
2. The punctured plane C∗ , the punctured unit disk D∗ := D\{0}, and
the unit circle ∂D are multiply connected.
3. The annulus {z ∈ C | r < |z| < R} of inner radius r and outer radius
R is multiply connected.
Example 8. Consider the set S = {z ∈ C | |Im z1 | < 1}. In polar form

z = reiθ , the inequality becomes

|Im(r−1 e−iθ )| = |r−1 sin(−θ)| = r−1 | sin θ| < 1
Therefore, | sin θ| < r. Similar to Example 5, this represents all the complex
numbers lying outside two closed disks D(± 2i , 12 ). The set S is illustrated in
Figure 1.4; it is open, unbounded, and multiply connected.

Figure 1.4: The set S.

1.4. THE TOPOLOGY OF C 11

Short Quiz 1
1+i
1. Simplify i−1
.

2. Find the modulus of (3 + 4i)(−4 + 3i).

3. Find the argument(s) of arg(−1 + i).

4. Express 2e−2πi/3 in the form of x + iy.

5. What are the 3th roots of 8i?

6. Find the value of (1 + i)6 .

z


7. If z 6= 0, express Im z+z̄ in terms of θ = Arg(z).

Consider the following subsets:

A = D(2, 2) ∪ D(−2, 2), B = D(i, 1) ∪ D(−i, 1),

C = D(2, 1) ∪ D(−2, 1), D = C(i, 1) ∪ D(−i, 1).

8. Identify subsets which are open.

9. Identify subsets which are connected.

10. Identify subsets which are simply connected.

√ √
Answers: 1. −i, 2. 25, 3. 3π4
+ 2πk, 4. −1 − i 3, 5. ± 3 + i & −2i, 6.
1
−8i, 7. 2 tan θ, 8. A and C, 9. B and D, 10. B.
12 CHAPTER 1. COMPLEX NUMBERS
Chapter 2

Complex Functions

2.1 Convergence and Continuity

Definition 3. A sequence of complex numbers {zn }n∈N converges to a limit
z if and only if:

for all  > 0, there exists N > 0 such that for all n ≥ N, |zn − z| < .

Convergence of a sequence zn to z can be denoted by zn → z, |zn −z| → 0,

or limn→∞ zn = z.
Proposition 2.1. The limit of a convergent sequence is unique.
Proof. Suppose for a contradiction that there are distinct limits z 6= w of a
sequence zn . Let  = 21 |z − w| > 0, then for all sufficiently high n, |zn − z| < 
and |zn − w| < . However, by triangle inequality,

|z − w| ≤ |z − zn | + |zn − w| < 2 = |z − w|.

We then have a contradiction.

Theorem 2.2. If zn → z and wn → w, then
• zn + wn → z + w,

• zn wn → zw.
Proof. Let’s pick  > 0. There are some high N1 , N2 ∈ N such that |zn − z| <
/2 if n ≥ N1 , and |wn − w| < /2 if n ≥ N2 . By triangle inequality, when
n ≥ max{N1 , N2 },
 
|zn + wn − z − w| ≤ |zn − z| + |wn − w| < + = .
2 2
13
14 CHAPTER 2. COMPLEX FUNCTIONS

This shows that zn + wn → z + w.

Set M = max{|w1 |, . . . |wN2 |, |w| + }. The sequence {wn }n∈N is bounded
because we have the following inclusions:
{wn }n∈N ⊂ {w1 , . . . wN2 } ∪ D(w, /2) ⊂ D(0, M ).
There are some N3 , N4 ∈ N such that |zn − z| < /2M if n ≥ N3 , and
|wn − w| < /2 max{1, |z|} if n ≥ N4 . Then, when n ≥ max{N3 , N4 },
|zn wn − zw| ≤ |zn wn − zwn | + |zwn − zw| = |wn ||zn − z| + |z||wn − w|
   
<M· + |z| · ≤ + = .
2M 2 max{1, |z|} 2 2
This shows that zn wn → zw.
In particular, a sequence of complex numbers converges exactly when the
real parts and the imaginary parts converge respectively.
Corollary 2.3. xn + iyn → x + iy if and only if xn → x and yn → y.
Proof. The ⇐ direction is immediate from the previous proposition. The ⇒
direction comes from the following inequality:
p
max{|xn − x|, |yn − y|} ≤ |xn − x|2 + |yn − y|2 = |xn + iyn − x − iy|.
As |xn + iyn − x − iy| → 0, sandwich rule forces both |xn − x| and |yn − y|
to converge to 0 too.
Definition 4. Let U and V be non-empty subsets of C. A function f : U →
V is continuous at a ∈ U if and only if:
for all  > 0, there exists δ > 0 such that
if z ∈ U ∩ D(a, δ), then f (z) ∈ D(f (a), ).
We say that f is continuous if it is continuous at every point in U .
Proposition 2.4. A function f : U → V is continuous at a ∈ U if and only
if for any sequence zn in U , if zn → a, then f (zn ) → f (a).
Proof. Let f be continuous at a and pick the pair (, δ) in the definition of
continuity. Suppose zn → a, then there is some high N ∈ N such that if
n ≥ N , |zn − a| < δ. By continuity, if n ≥ N , |f (zn ) − f (a)| < . Therefore,
f (zn ) → f (a).
Suppose for any sequence zn converging to a, f (zn ) → f (a). Suppose for
a contradiction that f is not continuous at a, then there is some  > 0 and
sequence of points zn ∈ U ∩ D(a, n1 ) for n ∈ N such that |f (zn ) − f (a)| ≥ .
Since |zn − a| < n1 → 0 as n → ∞, then zn → a. The assumption implies
that f (zn ) → f (a), but this cannot happen because f (zn ) is always at least
 away from f (a). This gives the contradiction.
2.2. HOLOMORPHIC FUNCTIONS 15

The statement above gives an equivalent way of defining continuity at a

point. A shorter way of saying this is

lim f (z) = f (a).

z→a

The following is a direct consequence of 2.2.

Proposition 2.5. Let f, g : U → V and h : V → W be continuous functions,
then the sum f +g, the product f ·g and the composition h◦f are continuous.
Example 9. Constant functions f (z) = a are trivially continuous. The
identity function Id(z) = z is also continuous. By taking products and sums,
we can inductively obtain that every complex polynomial ad z d + ad−1 z d−1 +
. . . a1 z + a0 is continuous.
Example 10. The modulus function m(z) = |z| is a continuous function on
C. Indeed, for any a ∈ C, if z → a, then by triangle inequality,

m(z) − m(a) ≤ |z| − |a| ≤ |z − a| → 0.

By sandwich rule, m(z) → m(a) too.

By the previous proposition, if f (z) is a continuous function on a subset
of C, then so is the composition |f (z)|.
Example 11. The functions Re and Im are continuous. (Refer to Corollary
2.3.) Since z̄ = Re(z) − i Im(z), complex conjugation is also continuous.
Continuous functions behave nicely on compact subsets of C.
Theorem 2.6. Let f : K → V be a continuous function and K be a compact
subset of C. Then, f attains a maximum and a minimum on K, i.e. there
are points a, b ∈ K where |f (a)| ≤ |f (z)| ≤ |f (b)| for all z ∈ K.
The theorem above is a consequence of a result from topology. In partic-
ular, the image |f (K)| of a compact set K under a continuous function |f | is
compact. Any compact subset of R contains maximum and minimum points
because it must be a finite union of closed finite intervals.

2.2 Holomorphic Functions

We can define differentiability of complex-valued functions the same way as
we define that of functions of one real variable. However, we will emphasise
in the next few sections that complex differentiability is actually a much more
rigid notion than the usual multivariable real differentiability.
16 CHAPTER 2. COMPLEX FUNCTIONS

Definition 5. Let U, V ⊂ C be open and non-empty. A complex function

f is (complex) differentiable at a point a if and only if the following limit
exists:
f (z) − f (a)
f 0 (a) = lim
z→a z−a
0
If so, then f (a) is called the (complex) derivative of f at a. The function f
is said to be holomorphic on U if it is holomorphic at every point in U , and
entire if additionally U = C.
Remark. The term ”analytic” and ”complex differentiable” are often used
interchangeably with ”holomorphic”.
Example 12.
1. Constant functions f (z) = a are entire with derivative 0 everywhere.
2. The identity function Id(z) = z is an entire function and its derivative
is 1 everyhere.
3. The inversion function τ (z) = 1/z is is holomorphic on C∗ and its
derivative is −z −2 . Indeed, if we choose any angle θ, then if z = reiθ +a,
−re iθ
1
τ (z) − τ (a) reiθ +a
− a1 a(reiθ +a)
lim = lim = lim
z→a z−a r→0 (reiθ + a) − a r→0 reiθ
1 1
= lim − = − .
r→0 a(reiθ + a) a2
4. Complex conjugation f (z) = z̄ has no derivative at any point. If we
choose any angle θ, then using z = reiθ + a,
f (z) − f (a) (re−iθ + ā) − ā
lim = lim = e−2iθ ,
z→a z−a r→0 (reiθ + a) − a

but the value of this limit is not the same if we choose different values
of θ. For example, the limit is 1 when θ = 0, but it is −1 if θ = π2 .
Proposition 2.7. Every holomorphic function is continuous.
Proof. Let f : U → V be holomorphic. If a ∈ U ,
f (z) − f (a)
lim f (z) − f (a) = lim (z − a)
z→a z→a z−a
f (z) − f (a)
= lim (z − a) · lim
z→a z→a z−a
0
= 0 · f (a) = 0,
where the second equality follows from Theorem 2.2. Therefore, f is contin-
uous at a. As a is arbitrary, f is continuous on U .
2.2. HOLOMORPHIC FUNCTIONS 17

The rules for differentiation of complex-valued functions is more or less

the same as those of functions of one real variable.
Proposition 2.8. Let f, g : U → V and h : V → W be holomorphic. Then,
(a) the sum f + g is holomorphic and (f + g)0 (z) = f 0 (z) + g 0 (z),

1. the product f · g is holomorphic and (f · g)0 (z) = f 0 (z)g(z) + f (z)g 0 (z),

2. the composition h ◦ f is holomorphic and (h ◦ f )0 (z) = h0 (f (z))f 0 (z).

Proof. (a) follows immediately from Proposition 2.5. For (b),

f (z)g(z) − f (a)g(a)
(f · g)0 (a) = lim
z→a z−a


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

For (c), we use the fact that f is continuous:

h(f (z)) − h(f (a))

(h ◦ f )0 (a) = lim
z→a z−a
h(f (z)) − h(f (a)) f (z) − f (a)
= lim ·
z→a f (z) − f (a) z−a
h(w) − h(f (a)) f (z) − f (a)
= lim · lim
w→f (a) w − f (a) z→a z−a
0 0
= h (f (a))f (a).

Example 13.
1. Every polynomial p(z) = dn=0 an z n with complex coefficients an ∈ C is
P
an entire function. We can show this inductively by product rule above
that every monomial an z n is holomorphic with derivative an nz n−1 and
by the addition rule, p is holomorphic.

2. Every rational function, i.e. a function of the form f (z) = p(z)/q(z)

where p and q are polynomials, is holomorphic on C\{z ∈ C|q(z) = 0}.
Every complex function f (z) admits a unique real-imaginary splitting
f (x + iy) = u(x, y) + iv(x, y), where u and v are real-valued functions defined
on an open subset of R2 given by:

u(x, y) = Ref (x + iy), v(x, y) = Imf (x + iy).

18 CHAPTER 2. COMPLEX FUNCTIONS

We say that f is (real) differentiable if the partial derivatives of u and v

with respect to x and y exist. When we are given u and v, we will see that
holomorphic functions are precisely solutions of a system of partial differential
equations.

Theorem 2.9 (Cauchy-Riemann Equations). Let f = u + iv be a complex

function on an open set U ⊂ C. Then, f is holomorphic if and only if u and
v are continuously differentiable and ux = vy and vx = −uy .

Proof. Let f be holomorphic at a point a = a1 + ia2 ∈ U , then

u(a1 + h1 , a2 + h2 ) − u(a1 , a2 ) v(a1 + h1 , a2 + h2 ) − v(a1 , a2 )

f 0 (a) = lim +i .
h→0 h h

The dummy variable h = h1 + ih2 can converge to zero in various directions,

but we will only consider two cases. Suppose h2 = 0, then

u(a1 + h1 , a2 ) − u(a) v(a1 + h1 , a2 ) − u(a)

f 0 (a) = lim +i
h1 →0 h1 h1
= ux + ivx . (2.1)

Similarly, if we consider the limit in the direction satisfying h1 = 0,

u(a1 , a2 + h2 ) − u(a) v(a1 , a2 + ih2 ) − v(a)

f 0 (a) = lim +i
ih2 →0 ih2 ih2
= −iuy + vy . (2.2)

Comparing (2.1) and (2.2), it is clear by taking the real and imaginary parts
of f 0 (a) that the partial derivatives of u and v at a exist and satisfy ux = vy
and vx = −uy . To show that these partial derivatives are continuous, we
need continuity of f 0 . We will obtain this for granted in Corollary 4.4.
Conversely, suppose u and v are continuously differentiable satisfying
ux = vy and vx = −uy . The Taylor series of u and v at a can be expressed
as:

u(a1 + h1 , a2 + h2 ) = u(a1 , a2 ) + ux h1 + uy h2 + |h|ψ(h),

v(a1 + h1 , a2 + h2 ) = v(a1 , a2 ) + vx h1 + vy h2 + |h|φ(h),

for some functions ψ and φ where ψ(h), φ(h) → 0 as h → 0. All partial

derivatives of u and v are evaluated at (a1 , a2 ). Then,
2.2. HOLOMORPHIC FUNCTIONS 19

u(a1 + h1 , a2 + h2 ) − u(a1 , a2 ) v(a1 + h1 , a2 + h2 ) − v(a1 , a2 )

lim +i
h→0 h h
[ux h1 + uy h2 + |h|ψ(h)] + i[vx h1 + vy h2 + |h|φ(h)]
= lim
h→0 h

(ux + ivx )(h1 + ih2 ) + |h| ψ(h) + iφ(h)
= lim
h→0 h
= ux + ivx .
As the limit converges to f 0 = ux + ivx , f is indeed holomorphic.
It is natural to consider a change of variables from (x, y) to (z, z̄) =
(x+iy, x−iy). By using multivariable chain rule, we can obtain an expression
∂
for partial derivatives ∂z and ∂∂z̄ in terms of ∂x
∂ ∂
and ∂y .
Definition 6. The Wirtinger derivatives are defined as follows:
   
∂ 1 ∂ ∂ ∂ 1 ∂ ∂
= −i , = +i
∂z 2 ∂x ∂y ∂ z̄ 2 ∂x ∂y
It is straightforward to check that the following identities hold:
∂z ∂ z̄ ∂ z̄ ∂z
= = 1, = = 0.
∂z ∂ z̄ ∂z ∂ z̄
Given a complex function f = u+iv with differentiable real and imaginary
parts u and v,
∂f 1 1

= (fx + ify ) = (ux − vy ) + i(vx + uy )
∂ z̄ 2 2
Clearly, the Cauchy-Riemann equations hold if and only if the expression
above is 0. The equation
∂f
=0
∂ z̄
is the complex form of the Cauchy-Riemann equations, as it allows us to
deduce holomorphicity without having to know the real and imaginary parts
u and v, but rather by the absence of the variable z̄. Roughly speaking, if f
is not a function of z̄, then it is holomorphic!
If f is holomorphic, the notation for the complex derivative of f is also
consistent since
∂f 1 1
= (fx − ify ) = (ux + vy ) + i(vx − uy ) = ux + ivx = f 0 ,


∂z 2 2
where the last inequality follows from (2.1).
Example 14. Re and Im are differentiable but not holomorphic. (Refer to
Proposition 1.1.)
20 CHAPTER 2. COMPLEX FUNCTIONS

2.3 Exponential and Logarithmic Functions

One of the many elementary functions we will commonly encounter is the
exponential function
exp(z) := ez = ex eiy .
The real and imaginary parts are u = ex cos(y) and v = ex sin(y), and we can
easily check that the Cauchy-Riemann equations are satisfied. Thus, this is
an entire function and its derivative is itself. Below are some of its properties:

1. | exp(z)| = ex ,

2. arg(exp(z)) = y + 2πik where k ∈ Z,

3. exp(z + 2πik) = exp(z) for any k ∈ Z,

4. The image of exp on C is C∗ .

We would now like to find the inverse of exp. If we denote the inverse by
log, then
log(z) := ln |z| + i arg(z)
for any z 6= 0. However, as arg is multivalued, log is multi-valued and
therefore it is not a well-defined function. This is consistent with the fact
that exp is not injective due to property 3.
This problem can be fixed by using the principle argument Arg. Doing so
will introduce a ray of discontinuity (−∞, 0] corresponding to the points in C
with argument π. We can replace arg with Arg and define the principal value
of log(z), denoted by pv log(z). If we choose the codomain to be Arg(z) ∈
(−π, π], we then have the principal branch of the logarithmic function

Log : C\(−∞, 0] → C, Log(z) := ln |z| + iArg(z).

The ray (−∞, 0] is often called a branch cut. Using this choice of branch cut,
Log is holomorphic on C\(−∞, 0], with image {x + iy | x ∈ R, −π < y < π}
and derivative z1 . (To verify this, compute the Wirtinger derivatives in polar
coordinates.)

Example 15. log(i) is purely imaginary and multivalued since

πi
log(i) = ln(1) + i arg(i) = + 2πik, where k ∈ Z.
2
πi
Its principal value is Log(i) = 2
.
2.4. TRIGONOMETRIC FUNCTIONS 21

Figure 2.1: Various branch cuts of the logarithm

Remark. In general, the branch cut can be taken to be any unbounded curve
from 0 which does not intersect itself. Straight rays of different angles are
often used if necessary.
For a non-integer c ∈ C, we define the power function z c to be the mul-
tivalued function on C∗ given by
z c := exp(c log(z)).
Similar to the logarithm, we can take the principal value of z c to be
pv z c := exp(c Log(z)).
Again, this becomes a holomorphic function outside a chosen branch cut
(−∞, 0].
Example 16. ii is (perhaps surprisingly) real and multivalued since
ii = e−π/2−2πk , where k ∈ Z.
Its principal value is pv ii = e−π/2 .

2.4 Trigonometric Functions

Euler’s formula allows us to express sin θ and cos θ in terms of e±iθ as follows:
eiθ + e−iθ eiθ − e−iθ
cos θ = , sin θ = .
2 2i
It is then natural to define trigonometric functions of a complex variable
z using the exponential function:
eiz + e−iz eiz − e−iz sin(z)
cos(z) := , sin(z) := , tan(z) = .
2 2i cos(z)
22 CHAPTER 2. COMPLEX FUNCTIONS

As exp is entire, so are cos and sin. However, the function tan is holomorphic
everywhere except at points z such that cos(z) = 0. You may check that the
usual trigonometric identities still hold.
The generalisation of hyperbolic functions are also clear:

ez + e−z ez − e−z sinh(z)

cosh(z) := , sinh(z) := , tanh(z) = .
2 2 cosh(z)

The functions cosh and sinh can be viewed as the even part and the odd
part of the exponential function respectively. Both are entire functions, but
tanh is only holomorphic everywhere except at points z such that cosh(z) = 0.
The following property can be easily deduced by definition.

Proposition 2.10. For any z ∈ C, cos(iz) = cosh(z) and sin(iz) = i sinh(z).

A point w is a zero of a function f if f (w) = 0. It turns out that the zeros

of trigonometric functions in C are the same as the zeros of trigonometric
functions in R.

Proposition 2.11. The zeros of trigonometric and hyperbolic functions are

as follows:

{z ∈ C | sin(z) = 0} = {nπ}n∈Z ,
nπ o
{z ∈ C | cos(z) = 0} = + nπ ,
2 n∈Z
{z ∈ C | sinh(z) = 0} = {inπ}n∈Z ,
n π o
{z ∈ C | cosh(z) = 0} = i + nπ .
2 n∈Z

Proof. By addition rule and Proposition 2.10, sin(x + iy) = sin x cosh y +
i sinh y cos x. Thus, if z = x + iy,

| sin(z)|2 = sin2 x cosh2 y + sinh2 y cos2 x

= sin2 x(1 + sinh2 y) + sinh2 y(1 − sin2 x)
= sin2 x + sinh2 y. (2.3)

As such, sin(z) = 0 if and only if sin x = 0 and sinh y = 0, and the latter
occurs exactly when x ∈ {nπ}n∈Z , and y = 0.
The zeros of cos can be obtained from those of sin using the identity
cos(z) = sin(π/2 − z), and the zeros of hyperbolic functions can be obtained
from those of trigonometric functions by Proposition 2.10.
2.4. TRIGONOMETRIC FUNCTIONS 23

Tgonometric and hyperbolic functions are definitely not surjective, but

we are still able to find their local inverses on some restricted domains. Since
they are made up of the exponential function, their inverses can be described
in terms of logarithms.
Proposition 2.12. The inverses of trigonometric and hyperbolic functions
are multivalued functions:
sin−1 (z) = −i log iz + [1 − z 2 ]1/2 ,

cos−1 (z) = −i log z + i[1 − z 2 ]1/2 ,

i i+z
tan−1 (z) = log ,
2 i−z
sinh−1 (z) = log z + [z 2 + 1]1/2 ,

cosh−1 (z) = log z + [z 2 − 1]1/2 ,

1 1+z
tanh−1 (z) = log .
2 1−z
iw −iw
Proof. Let sin−1 (z) = w, then z = e −e 2i
. This can be written in quadratic
form:
(eiw )2 − 2izeiw − 1 = 0.
The quadratic formula gives us eiw = iz + [1 − z 2 ]1/2 , and using logarithm,
w = −i log iz + [1 − z 2 ]1/2 .

The resulting function is multivalued since the square root and the logarithm
are multivalued. Similar algebraic methods can be applied to obtain the
inverses of other functions and will be left to the reader as an exercise.
For each of the functions above, we can pick a branch cut of the square
root and the logarithm in order to obtain a holomorphic branch of the func-
tion. However, finding a nice branch cut for the inverse of the function
requires a more involved argument and we shall not attempt to find it.
Example 17. Let’s find the solution for the equation sin(2πz) = 2. By the
proposition above,
i i √

z = − log 2i + (−3)1/2 = − log i(2 ± 3)


2π  2π
i √ πi
=− ln(2 ± 3) + + 2πik ,
2π 2
√
i ln(2 ± 3) 1
=− + + k, k ∈ Z,
2π 4
24 CHAPTER 2. COMPLEX FUNCTIONS

Short Quiz 2

1. Does the sequence in converge as n → ∞? If so, what is the limit?

n
2. Does the sequence ni converge as n → ∞? If so, what is the limit?

n
3. Find the limit of the sequence n1 + 1 + ni .

4. Which of the following functions are continuous on C?

z3, 1/z, |z − 2| + |z + 2|, arg(z)

5. At which of the following points is Arg(z) continuous?

1, i, −1, −i, 0

6. What are the solutions of the equation ez = −π?

7. What is the value of (ii )i ?

8. What are the solutions of the equation tanh(iz) = 0?

9. Which of these functions are holomorphic on a domain containing 2πi?

sin z, sinh z, csc z cschz cot z coth z

10. Which of these complex numbers correspond to log(1)?

0, 1, 2πi, 1 + 2πi, −2πi, 1 − 2πi.

Answers: 1. No, 2. Yes, to 0, 3. ei , 4. z 3 and |z − 2| + |z + 2|, 5. 1, i, −i.

6. ln π + i(π + 2πik). 7. −i. 8. kπ. 9. sin z, sinh z, csc z, cot z. 10. 0 and
±2πi.
Chapter 3

Contour Integration

3.1 Curves in C
Definition 7. An arc / curve / path in C is subset of C parametrized by a
continuous function γ : [a, b] → C defined on a closed interval [a, b] ⊂ R.

A curve γ : [a, b] → C, can be expressed as γ(t) = u(t) + iv(t) where u

and v are real-valued functions. We say that γ is differentiable when both u
and v are differentiable on [a, b] as real functions. The derivative of γ at t is

γ 0 (t) = u0 (t) + iv 0 (t).

For each t, when γ 0 (t) 6= 0, γ 0 (t) represents

p a tangent vector of the curve at
0
the point γ(t) of magnitude |γ (t)| = u (t)2 + v 0 (t)2 .
0

Definition 8. A curve γ : [a, b] → C is:

• closed if γ(a) = γ(b),

• simple if γ is injective on the open interval (a, b),

• smooth if γ is differentiable and γ 0 (t) 6= 0 for all t ∈ (a, b),

• a contour if γ is piecewise smooth, i.e. γ can be partitioned into finitely

many smooth curves.

Example 18. A circle C(z, r) of radius r > 0 centered at w ∈ C is a simple

closed smooth curve. It can be parametrised by γ(t) = z0 + reit , 0 ≤ t ≤ 2π.

Example 19. The function σ(t) = eit sin(2t), 0 ≤ t ≤ 2π, parametrises the
locus of the equation r = sin(2θ). This curve is a closed and smooth but
clearly not simple. See the leftmost curve in Figure 3.1.

25
26 CHAPTER 3. CONTOUR INTEGRATION

Figure 3.1: A closed non-simple smooth curve, a non-closed non-piecewise-

smooth curve (Devil’s Staircase), and a closed piecewise smooth curve.

We will state without proof two results from topology.

Proposition 3.1. Any domain U ⊂ C is path-connected. That is, for any

two points z and w in a domain U ⊂ C, there is a smooth curve γ : [0, 1] → U
such that γ(0) = z and γ(1) = w.

Theorem 3.2 (Jordan Curve Theorem). The complement of any simple

closed curve in C has exactly two connected components, exactly one of which
is bounded.

Figure 3.2: A simple closed curve splits its complement into a bounded do-
main U and an unbounded domain V .

Remark. Due to this theorem, simple closed curves are often called Jordan
curves. You may think that the theorem seems very intuitive, but the proof
is rather involved. The very first proof was out in 1910s (not by Camille
Jordan), relying on heavy machineries such as the theory of homology groups
3.2. INTEGRATION ALONG A CONTOUR 27

in algebraic topology. The shortest proof I know is by Maehara (1984),

relying only on basic knowledge of topology, including the Brouwer fixed
point theorem.

Definition 9. A point w or a subset S of C is said to be enclosed by a simple

closed curve γ if {w} or S is contained in the bounded connected component
of the complement of γ in C.

Every curve γ has two possible orientations. Given a curve γ : [a, b] → C,

we can reverse its orientation to obtain γ − : [a, b] → C defined by γ − (t) =
γ(a + b − t).

Figure 3.3: Reversing the orientation of γ.

When γ is a simple closed curve, the orientation of γ is positive if for any

point w enclosed by γ, γ(t) runs anticlockwise with respect to the basepoint w
as t increases. The orientation is negative if γ(t) runs clockwise. Unless stated
otherwise, we always assume that every simple closed curve γ is positively
oriented, and γ − is negatively oriented.

3.2 Integration Along a Contour

Any closed interval [a, b] can be parametrized by a trivial curve γ(t) = t,
t ∈ [a, b]. The integral of a continuous function f along γ is taken to be
Rb
a
f (z)dz. We shall introduce a way to generalise line integrals along an
arbitrary contour.

Definition 10. Let f be a complex-valued continuous function defined on a

smooth curve parametrized by γ : [a, b] → C. The integral of f along γ is
Z Z b
f γ(t) γ 0 (t)dt.


f (z)dz :=
γ a
28 CHAPTER 3. CONTOUR INTEGRATION

If γ is a contour which is smooth on [aj−1 , aj ] for j = 1, . . . N where a0 = a

and aN = b, then the integral of f along γ is a finite sum of the integral over
the smooth parts:
Z N Z
X aj
f γ(t) γ 0 (t)dt.


f (z)dz :=
γ j=1 aj−1

Remark. When γ is a simple closed curve, it is common to denote the integral

of f along γ by the notation
I
f (z)dz.
γ

Two parametrizations γ : [a, b] → C and σ : [c, d] → C of a smooth curve

are equivalent if there is a continuously differentiable bijection h : [a, b] →
[c, d] such that h(a) = c, h(b) = d, and γ = σ ◦ h.
Our definition of contour integral is robust because it is independent
of our choice of parametrization of the curve. If two curves γ and σ are
equivalent, convince yourself by change of variables t 7→ h(t) defined above
that the integral of f along γ and σ must be equal:
Z Z
f (z)dz = f (z)dz.
γ σ

If we reverse the orientation of γ, we can again apply change of variables

t 7→ a + b − t and check that the integral of a continuous function f on γ − is
Z Z
f (z)dz = − f (z)dz. (3.1)
γ− γ
H
Example 20. Let’s compute the integral I = γ f (z)dz of the function
f (z) = 1/(z − z0 ) along the curve γ(t) = z0 + reit where 0 ≤ t ≤ 2π.
Z 2π Z 2π Z 2π
γ 0 (t) ireit
I= dt = dt = idt = 2πi.
0 γ(z) − z0 0 (z0 + reit ) − z0 0

Notice that the value I is independent of the radius r and the center z0 .
3.2. INTEGRATION ALONG A CONTOUR 29

The length L(γ) of a smooth curve γ : [a, b] → C can be computed by the

following integral: Z b
L(γ) = |γ 0 (t)|dt.
a
We can think of L(γ) as the integral of the continuous function f along
γ defined by f (z) = |γ 0 (t)|/γ 0 (t) where z = γ(t). As such, the length is
invariant under parametrization and change of orientation.
Example 21. The length of any circle of radius r is 2πr. Indeed, using the
parametrisation γ(t) = z0 + reit where 0 ≤ t ≤ 2π,
Z 2π Z 2π
it
L(γ) = |ire |dt = rdt = 2πr.
0 0

An equivalent parametrization that is commonly used is σ(t) = z0 +re2πit , 0 ≤

t ≤ 1. Can you show that the length computed using this parametrization is
the same?
Proposition 3.3. Let f and g be continuous functions on a contour γ and
let α, β ∈ C. Then,
R R R
• γ αf (z) + βg(z)dz = α γ f (z)dz + β γ g(z)dz, (linearity)
R
• γ f (z)dz ≤ L(γ) · maxz∈γ |f (z)|. (ML inequality)
Proof. Linearity is trivial. The curve parametrized by γ is a compact subset
of C and therefore, by Theorem 2.6, |f | always attains its maximum along
γ. Let M := maxz∈γ |f (z)| and pick a parametrization γ : [a, b] → C. Then,
Z Z b Z b
0
f (z)dz ≤ f (γ(t))γ (t)dt ≤ |f (γ(t))| · |γ 0 (t)|dt
γ a a
Z b Z b
≤ M |γ 0 (t)|dt = M |γ 0 (t)|dt = M L(γ).
a a

The second inequality follows from the fact thatRfor any continuous function
b Rb
h : [a, b] → C, we always have the inequality | a h(t)dt| ≤ a |h(t)|dt. We
Rb
shall prove this below. Let reiθ be the polar form of a h(t)dt. Then,
Z b Z b Z b
−iθ
h(t)dt = e h(t)dt = Re e−iθ h(t)dt
a a a
Z b Z b Z b
−iθ −iθ
= Re[e h(t)]dt ≤ |e h(t)|dt = |h(t)|dt,
a a a

and we are done.

30 CHAPTER 3. CONTOUR INTEGRATION

3.3 Primitives
Definition 11. An antiderivative / primitive of a continuous function f on
a domain U ⊂ C is a holomorphic function F : U → C such that F 0 = f .
The existence of primitives makes computation of contour integrals much
easier. Regardless of the shape of the contour, it turns out that the integral
only depends on the endpoints of the contour.
Lemma 3.4. Suppose f is continuous on a domain U and has a primitive
F . Then, for any contour γ : [a, b] → U ,
Z
f (z)dz = F (γ(b)) − F (γ(a)),
γ

Proof. Let γ : [a, b] → U be a smooth curve, then

Z Z b Z b
0 0 d
f (z)dz = F (γ(t))γ (t)dt = F (γ(t))dt = F (γ(b)) − F (γ(a)),
γ a a dt

If γ is piecewise smooth, we can sum up the integral over all smooth parts
to obtain a similar result.
Corollary 3.5. If F is holomorphic on a domain U and F 0 ≡ 0, then F is
a constant function.
Proof. For any two points u and v in U , we can apply the lemma above by
setting f = 0 to obtain f (u) = f (v) because the integral over f over any
curve is 0.
Theorem 3.6. Suppose f : U → C is continuous on a domain U . The
following are equivalent:
(a) f has a primitive F : U → C,
H
(b) For any closed contour γ in U , σ
f (z)dz = 0.
Proof. Suppose (a) is true. Let γ : [a, c] → U be a closed contour, then
F (γ(b)) = F (γ(b)) because γ(b) = γ(a). Then, Lemma 3.4 immediately
gives us (a) ⇒ (b).
Suppose (b) is true. Pick a basepoint z0 ∈ U and for each z ∈ U , let
γz : [0, 1] → U be a smooth curve from γz (0) = z0 to γz (1) = z. Let’s define
F by Z
F (z) = f (w)dw.
γz
3.3. PRIMITIVES 31

To prove that F is a well-defined function, we must show that the value

F (z) is independent of the choice of the smooth curve γz . Let γz and σz
be two such curves, then reverse the orientation σz to obtain σz− , a smooth
curve from z to z0 . The two curves γz and σz− glue together to form a closed
contour Γz . By (3.1) and (b),
Z Z Z Z I
f (w)dw − f (w)dw = f (w)dw + f (w)dw = f (z)dw = 0.
γz σz γz σz− Γz

Therefore, the integrals of f along γz and σz coincide, proving that F is a

function. The proof is complete once we show that F is holomorphic and
F 0 = f . Pick any point z ∈ U , then U contains a small ball D(z, ). Pick
h ∈ C such that |h| < , and define a line segment αh (t) = z + th, 0 ≤ t ≤ 1.
By a similar argument as above, (b) implies
Z
F (z + h) − F (z) = f (w)dw.
αh
R
Then, using the fact that αh
dz = h and L(αh ) = |h|, by ML inequality,

F (z + h) − F (z)
f (z) −
h
Z Z
1 1

= f (z) − f (w)dw = f (z) − f (w) dw
h αh h αh
1
≤ L(αh ) max |f (z) − f (w)| = max |f (z) − f (w)|.
|h| w∈αh w∈αh

By continuity of f , as h → 0, maxw∈αh |f (z) − f (w)| → 0. The limit of the

left hand side of the inequality above is also 0, thus proving holomorphicity
of F near z.
Example 22. The function f (z) = (z − z0 )−n on C\{z0 } admits a primitive
(z−z0 )−n+1
−n+1
when n 6= 1. As such, if n 6= 1 and γ is any closed contour not
passing through z0 , I
dz
n
= 0.
γ (z − z0 )

Example 23. The derivative of log : C\(−∞, 0] → C is the function

τ (z) = 1/z, so then τ has primitive log in the domain C\(−∞, 0]. This
domain cannot be extended any further since log becomes discontinuous on
the branch cut. Alternatively, we can say that τ has no primitive in C∗ be-
cause, from Example 20, the integral of τ , along any circle centered at 0 is
non-zero.
32 CHAPTER 3. CONTOUR INTEGRATION

3.4 Cauchy-Goursat Theorem

Theorem 3.6 does not give a nice criterion for the existence of primitives
because it can be rather troublesome to compute integrals over all possible
closed contours.

Theorem 3.7 (Cauchy-Goursat). Let f be holomorphic on a simply con-

nected domain U ⊂ C and let γ be a closed contour in U , then
I
f (z)dz = 0.
γ

Cauchy proved the theorem using Green’s theorem from multivariable

calculus, whereas Goursat’s proof, albeit rather lengthly, only uses continuity
of partial derivatives. We will only present the proof using Green’s theorem.
Proof. Let’s assume first that γ is a simple closed contour and let V be the
bounded domain enclosed by γ. Let f (x + iy) = u(x, y) + iv(x, y). Assume
that γ is positively oriented. By the usual change of variables z = x + iy and
by Green’s theorem,
I I
f (z)dz = f (x + iy)dx + if (x + iy)dy
γ
ZγZ ZZ
= ifx − fy dxdy = (iux − vx ) − (uy + ivy )dxdy = 0,
V V

where the last equality follows from Cauchy-Riemann equations.

When γ is negatively oriented (clockwise), we can replace it by γ − and
the minus sign will not change the value zero. When γ is not a simple curve,
there is a way to partition γ into a finite number of components consisting of
simple closed curves and a degenerate closed curves, i.e. those which enclose
a region of zero area. Taking the sum of the integrals over each of these
components, we will still obtain 0.
Remark. Simply connectedness is an essential criterion. If the domain U
is multiply connected, there is always a contour γ such that the domain V
enclosed by U is not contained in U and thus the integral of f over the region
V \U will not make sense. Example 20 shows that this theorem fails when
the domain is multiply connected.
Combining this theorem with Theorem 3.6, we obtain the following.

Corollary 3.8. Any holomorphic function on a simply connected domain

has a primitive.
3.4. CAUCHY-GOURSAT THEOREM 33

Theorem 3.9 (Deformation Theorem). Let f be holomorphic on a domain

U ⊂ C, and let γ and σ be two simple closed contours such that σ lies in the
domain V enclosed by γ and that both contours have the same orientation.
If the annulus A enclosed by γ and σ is contained in U , then
I I
f (z)dz = f (z)dz.
γ σ

Proof. Pick a simple contour α in A joining a point on the outer boundary

to another on the inner boundary. (See Figure 3.4.) Removing α from A
gives us a region A0 and its boundary can be parametrized by the closed
contour Γ obtained by gluing in order the curves γ, α, σ − , and α− . Then,
by Cauchy-Goursat,
I I I I I I
f (z)dz − f (z)dz = f (z)dz + f (z)dz + f (z)dz + f (z)dz
γ σ γ α σ− α−
I
= f (z)dz = 0.
Γ

Figure 3.4: Curve α joining γ and σ. Figure 3.5: The square γ can be re-
placed by the smaller circle.

Example 24. Let’s compute the integral I of (z 2 − 4)−1 along γ, a simple

closed contour parametrising the square of side length 4 centered at 1 + i.
The partial fraction decomposition of I is
I I I
1 1 1 1 1
I= 2
dz = dz − dz
γ z −4 4 γ z−2 4 γ z+2

Since the function 1/(z + 2) is holomorphic on and inside γ, the second inte-
1
gral is zero by Cauchy-Goursat. The function z−2 is holomorphic everywhere
34 CHAPTER 3. CONTOUR INTEGRATION

except at 2, so then by deformation theorem, we can replace γ on the first in-

tegral with any circle centered at 2, e.g. C(2, 0.5). (See Figure 3.5.) Example
20 immediately tells us that the first integral is equal to 41 · 2πi. Therefore,
I = 21 πi.
In fact, the deformation theorem can be applied to any pair of arbitrary
closed contours in U which are homotopic. This means that one curve can
be continuously deformed in U to the other curve. If the two contours are
not closed, then they must be homotopic relative to their endpoints, i.e. we
need the additional assumption that both have the same endpoints. (Lemma
3.4 hints at why fixing endpoints are important.) Below is a simple pictorial
explanation of homotopic curves.

Figure 3.6: On the gray annular domain above, curves 1, 2 and 3 have the
same endpoints but only 1 and 2 are homotopic relative to their endpoints.
Among closed curves 4 - 7, the only pair of homotopic curves is 5 and 6.

Short Quiz 3
1. What is the integral of 1/z along the circle C(0, 2)?
2. Compute the length of the curve γ(t) = cos(t)e( it) where 0 ≤ t ≤ π.
3. Compute the contour integral of the function 8z 3 along an L-shaped
contour which starts from 1 to 2i and passes through 0.
Answers: 1. 2πi, 2. π, 3. 30.
Chapter 4

Integration Formulas

4.1 Cauchy’s Formulas

Theorem 4.1 (Cauchy’s Integral Formula). Let f : U → V be a holomorphic
function, γ be a simple closed contour in U , and W be the domain enclosed
by γ such that U ⊂ W . For any point z0 in W ,
I
1 f (z)
f (z0 ) = dz.
2πi γ z − z0
Proof. We can assume that z0 = 0 without loss of generality because when
z0 6= 0, we can replace f with the function f (z + z0 ) on the domain {z ∈ C :
z + z0 ∈ U } and the contour γ(t) with γ(t) − z0 .
By the deformation theorem, we can replace γ with γr , a contour parametriz-
ing the circle C(0, r) for arbitrarily small radius r > 0. Recall that
I
1 1
dz = 1.
2πi γr z
Then, by taking the limit as r → 0,
I I
1 f (z) 1 f (z)
dz − f (0) = dz − f (0)
2πi γ z 2πi γr z
f (z) − f (0)
I
1
= dz
2πi γr z
1 f (z) − f (0)
≤ · L(γr ) · max
2π z∈γr z
f (z) − f (0)
= r · max −→ 0 · f 0 (0) = 0.
z∈γr z
As the term of the left hand side is independent of r, it is identically 0.

35
36 CHAPTER 4. INTEGRATION FORMULAS

The case where the closed contour is chosen to be a circle yields an inter-
esting property of holomorphic functions.
Corollary 4.2 (Mean Value Property). Let f be holomorphic on a domain
U . For any closed disk D(z0 , r) in U ,
Z 2π
1
f (z0 ) = f (z0 + reit )dt.
2π 0
Proof. Let γ(t) = z0 + reit , 0 ≤ t ≤ 2π parametrize the circle C(z0 , r). Since
f is holomorphic on a simply connected open neighbourhood of D(z0 , r), by
Cauchy’s integral formula,
I
1 f (z)
f (z0 ) = dz
2πi γ z − z0
Z 2π
1 f (z0 + reit ) it
= ire dt
2πi 0 reit
Z 2π
1
= f (z0 + reit )dt.
2π 0
The reason why the corollary above deserves its name is clear if you view
the integral as the average value of f along the circle γ centered at z0 . Recall
that the length element ds in polar coordinates is ds2 = dr2 + r2 dt2 . When
r is constant, ds =  dt and the equation can be rewritten as:
Z
1
f (z0 ) = f (z)ds.
L(γ) γ
Example 25. Let’s evaluate the integral I along γ parametrizing C(0, 2),
given by
ez
I
I= 2
dz.
γ z −1
By partial fractions decomposition, we can split I into I1 + I2 where
ez ez
I I
1 1
I1 = dz, I2 = dz.
2 γ z−1 2 γ z+1
By Cauchy’s formula, we immediately obtain I1 = πie and I2 = πie−1 . Thus,
I = 2πi cosh(1).
Theorem 4.3 (Cauchy’s Differentiation Formula). Let f : U → V be a
holomorphic function, γ be a simple closed contour in U , and W be the
domain enclosed by γ such that U ⊂ W . Then, the nth derivative f (n) (z0 ) of
f at a point z0 in W satisfies the following formula:
I
(n) n! f (z)
f (z0 ) = dz.
2πi γ (z − z0 )n+1
4.2. APPLICATIONS OF CAUCHY’S FORMULAS 37

Proof. We can again assume without loss of generality that z0 = 0. The base
case where n = 0 is exactly the previous theorem. Suppose the formula holds
for some natural number n. Then, for some small non-zero a,

f (n) (a) − f (n) (0)

I I
n! f (z) n! f (z)
= dz − dz
a 2πi γ a(z − a)n+1 2πi γ az n+1
f (z) · [z n+1 − (z − a)n+1 ]
I
n!
= dz
2πi γ az n+1 (z − a)n+1

Using the algebraic identity An+1 − B n+1 = (A − B) nk=0 Ak B n−k , the

P
term inside the square brackets simplifies to
n
X
a· z k (z − a)n−k .
k=0

Taking the limit as a → 0, this simplifies to the desired formula:

f (n) (a) − f (n) (0)

f (n+1) (0) = lim
I a Pn
a→0
n! f (z) k=0 z k (z − a)n−k
= lim dz
a→0 2πi γ z n+1 (z − a)n+1
f (z)(n + 1)z n
I
n!
= dz
2πi γ z 2n+2
I
(n + 1)! f (z)
= n+2
dz.
2πi γ z

By induction over n, the formula works for all n.

Example 26. We know that the function f (z) = sin z3
z
is holomorphic on C∗ .
Let γ be the unit circle C(0, 1). By Cauchy’s differentiation formula,
d2 sin
I I
2! sin z
f (z)dz = πi · dz = πi (0) = −πi sin 0 = 0.
γ 2πi γ z 3 dz 2

4.2 Applications of Cauchy’s Formulas

Cauchy’s formulas have many implications and we shall state a number of
them below. For each result presented, we shall see how holomorphic func-
tions are much more rigid compared to real differentiable functions in general.
Corollary 4.4. Every holomorphic function is infinitely complex differen-
tiable and all of its derivatives f 0 , f 00 , . . . are holomorphic.
38 CHAPTER 4. INTEGRATION FORMULAS

Proof. Let f : U → V be a holomorphic function. For every point z ∈ U , we

can pick a small radius  > 0 such that the closed disk D(z, ) lies in U . Take
γ to be the circle C(z, ) and obtain f (n) (z) for every n ∈ N by Cauchy’s
differentiation formula. This proves that derivatives of all orders exist. The
existence of f (n+1) automatically implies that f (n) is holomorphic on U .
Remark. Recall that holomorphic functions are often called analytic. The
term ”analytic” refers to functions (real or complex) which are infinitely (real
or complex) differentiable. There are plenty of examples of real differentiable
functions which are not analytic (e.g. the indefinite integral of any continuous
non-differentiable function).
Below is yet another important criterion of holomorphicity.
Theorem 4.5 (Morera). Let f be a continuous function on a domain U . If
I
f (z)dz = 0
γ

for every closed contour γ in U , then f is holomorphic.

Proof. By Theorem 3.6, the vanishing integral assumption guarantees the
existence of a primitive F of f on U . By the previous corollary, the derivative
of F , which is f , is holomorphic on U .
Corollary 4.6 (Cauchy’s Inequality). Let f be a holomorphic function on a
domain U and let D(z0 , r) be a closed disk contained in U . Then
n!
|f (n) (z0 )| ≤ max |f (z)|.
rn z∈C(z0 ,r)
Proof. Apply the ML inequality to Cauchy’s differentiation formula, taking
γ to be the circle C(z0 , r).
Cauchy’s inequality itself has many important implications.
Definition 12. A function f : U → V is bounded if there is some M > 0
such that |f (z)| ≤ M for all z ∈ U .
Theorem 4.7 (Liouville). Every bounded entire function is constant.
Proof. Suppose f is entire and bounded. There is some M > 0 where |f (z)| ≤
M for all z ∈ C. The complex plane C contains closed disks D(z, r) centered
at any point z of arbitrarily large radius r > 0. By Cauchy’s inequality,
M
|f 0 (z)| ≤ → 0, as r → ∞.
r
Since |f 0 (z)| is independent of r, f 0 is identically zero in C. By Corollary 3.5,
f must be constant.
4.2. APPLICATIONS OF CAUCHY’S FORMULAS 39

Example 27. All polynomials of degree ≥ 1, exp, sin, cos, sinh, and cosh
are all unbounded.
Remark. There is no such analogue of Liouville’s theorem for real functions.
For example, the functions tanh and x21+1 are bounded and infinitely differ-
entiable (i.e. real analytic) in the whole R.
One consequence of Liouville’s theorem is the following.
Corollary 4.8. The image of a non-constant entire function is dense in C,
i.e. it intersects every open disk in the plane.
Proof. Suppose for a contradiction that the image avoids some open disk
1
D(z0 , r), i.e. |f (z) − z0 | ≥ r for all z ∈ C. Then, f (z)−z 0
is an entire function
whose modulus is bounded above by r. By Liouville’s theorem, this fraction
must be a constant function, but then this raises a contradiction because f
is not constant.
Liouville’s theorem also gives us a standard proof of the fundamental
theorem of algebra.
Theorem 4.9 (Fundamental Theorem of Algebra). Every complex polyno-
mial p(z) of degree d ≥ 1 has exactly d roots (counting multiplicity).
Proof. Let p(z) = dn=0 an z n where ad 6= 0 and suppose for a contradiction
P
that p has no roots. Then, 1/p(z) is an entire function. As |z| → ∞,
1 1 1 1
lim = lim d ad−1 a0 =0· = 0.
|z|→∞ p(z) |z|→∞ |z | |ad + + ... + | |ad |
z zd
Let’s pick any small  > 0. By the definition of continuity, there is some
1
R > 0 such that p(z) ≤  whenever |z| > R. Since the closed disk D(0, R)
is a compact disk, the maximum value
1
M := max
|z|≤R p(z)
1
exists and is finite. Clearly, 1/p(z) is bounded because p(z) ≤ max{, M }.
By Liouville’s theorem, 1/p is constant, but this contradicts the fact that
p(z) is a non-constant polynomial.
Therefore, p has some root z1 ∈ C. This allows us to express p as a
product p(z) = (z − z1 )q1 (z) for some polynomial q1 (z) of degree d − 1. By
the same reasoning, q1 has a root z2 ∈ C and q1 (z) = (z − z2 )q2 (z) for some
polynomial q2 (z) of degree d − 2. Inductively, we can find d roots of p(z),
namely z1 , z2 , . . . zd (may be repeated).
Above is one of the many ways for us to prove the theorem. We will
present a much shorter one in section 5.5.
40 CHAPTER 4. INTEGRATION FORMULAS

4.3 Maximum Modulus Principle

Lemma 4.10. Let f be a holomorphic function on a domain U . If the
modulus |f (z)| is a constant function on U , then f is constant too.
Proof. Suppose |f (z)| = c for all z ∈ U . If c = 0, then f is identically zero.
Let’s assume c > 0. Writing f as f (x + iy) = u(x, y) + iv(x, y), observe
that u2 + v 2 = c2 . Taking the partial derivatives with respect to x and y
respectively, we obtain:
uux + vvx = 0, uuy + vvy = 0.
By Cauchy Riemann equations,
c2 (u2x + u2y ) = (u2 + v 2 )(u2x + u2y ) = (uux − vuy )2 + (uuy + vux )2
= (uux + vvx )2 + (uuy + vvy )2 = 0
As c2 6= 0, then ux = vy = 0 and uy = −vx = 0. Therefore, u and v are
constant.
Lemma 4.11 (Maximum Modulus Principle - Local Version). Let f be a
holomorphic function on an open disk D(z0 , R). If |f | attains a maximum at
z0 , i.e. |f (z)| ≤ |f (z0 )| for all z, then f is constant.
Proof. Suppose for a contradiction that |f (z)| is not constant. There must
be some point z in D(z0 , R) such that |f (z)| < |f (z0 )|. Let reiθ be the polar
form of z − z0 . By continuity of |f (z)|, there is a small δ > 0 such that
|f (z0 + reit )| < |f (z0 )| whenever θ − δ < t < θ + δ. Thus,
Z θ+δ Z θ+δ
it
|f (z0 + re )|dt < |f (z0 )|dt = 2δ|f (z0 )|. (4.1)
θ−δ θ−δ

The subset I = [0, 2π]\(θ − δ, θ + δ) has length 2π − 2δ. By ML inequality,

Z
|f (z0 + reit )|dt ≤ (2π − 2δ)|f (z0 )|.
I

Let γ be the circle C(z0 , r). By the mean value property,

Z 2π Z 2π
1 it 1
|f (z0 )| = f (z0 + re )dt ≤ |f (z0 + reit )|dt
2π 0 2π 0
Z Z θ+δ 
1 it it
= |f (z0 + re )|dt + |f (z0 + re )|dt
2π I θ−δ
1 
< (2π − 2δ)|f (z0 )| + 2δ|f (z0 )| = |f (z0 )|,
2π
where the strict inequality comes from (4.1). The above statement raises a
contradiction. Thus, |f | is constant. By the previous lemma, so is f .
4.3. MAXIMUM MODULUS PRINCIPLE 41

Theorem 4.12 (Maximum Modulus Principle - Global Version). If f be a

non-constant holomorphic function on a domain U . Then,

• |f | cannot attain a maximum on U .

• if additionally U is bounded and f is continuous along the boundary

∂U of U , then f attains a maximum at some point in ∂U .

Proof. Suppose for a contradiction that f attains a maximum at some point

z0 in U . Pick any point w ∈ U . Since U is connected, we can always find
some integer N > 0 and some finite sequence1 of open disks Dn := D(zn , rn )
in U for n = 0, . . . N such that zN = w and zn+1 ∈ Dn for all n < N .

Figure 4.1: A chain of 8 disks.

By the local version of the maximum principle, f ≡ f (z0 ) in D0 . Since

the center z1 of D1 lies in D0 , then the same lemma guarantees that f ≡
f (z0 ) in D1 . Inductively, if f ≡ f (z0 ) in Dn for every n, and in particular,
f (w) = f (z0 ). As the point w picked is arbitrary, f must be a constant
function equal to f (z0 ) on U . This proves the first part of the theorem.
Suppose f is non-constant and U is a bounded domain. The union U ∪∂U
of U and its boundary ∂U is a compact subset. By Theorem 2.6, |f (z)| must
attain a maximum at some point z0 ∈ U ∪ ∂U . We conclude that z0 ∈ ∂U
because the first part of the theorem states that z0 6∈ U .

There are plenty of examples of real differentiable functions which violate

the maximum modulus principle. One of such is the function x21+1 which has
a global maximum at 0.

Corollary 4.13 (Minimum Modulus Principle). Let f be a non-constant

holomorphic function on a domain U . Then,
1
This chain of disks is commonly known as kreisketten in German.
42 CHAPTER 4. INTEGRATION FORMULAS

• f cannot attain a minimum on U except at its zeros,

• if additionally U is bounded, f extends continuously to the boundary

∂U and f (z) 6= 0 for all z ∈ U , then f attains a minimum at some
point in ∂U .

Proof. If f (z) 6= 0 for all z ∈ U , then 1/f is a non-constant holomorphic

function on U . Apply the maximum modulus principle to 1/f .

Example 28. Let’s find the maximum and minimum of sin z on the closed
square S with vertices πi, π + πi, π + 2πi and 2πi. Recall from 2.3 that
| sin z|2 = sin2 x + sinh2 y. Observe the following.

• If 0 ≤ x ≤ π, sin2 x achieves a maximum value of 1 at x = π

2
and a
minimum value of 0 at x = 0, π.

• sinh2 y is monotonically increasing on π ≤ y ≤ 2π.

p
Therefore, on S, | sin z| achieves a maximum value of 1 + sinh2 2π at z =
π
2
+ 2πi and a minimum value of sinh π at z = πi, π + πi. These extremal
values are achieved along the boundary of S.

Example 29. Let’s find the maximum and minimum of |f |, where f (z) =
z 2 − 7z + 12, on the closed unit disk D̄. Since f only attains 0 outside of D̄
at 3 and 4, it is sufficient to check maxima and minima along the boundary,
which is the unit circle. When |z| = 1,

|f (z)| = |z − 3||z − 4| ≤ (|z| + 3)(|z| + 4) = 20,

|f (z)| = |z − 3||z − 4| ≥ (3 − |z|)(4 − |z|) = 6.

The triangle inequalities used above achieve equality at z = −1 and z = 1.

Thus, the extremal values of |f (z)| on D̄ are 20 and 6.
4.3. MAXIMUM MODULUS PRINCIPLE 43

Short Quiz 4
sin z
1. What is the integral of z−π
along the circle C(0, 5)?
2z 5
2. What is the integral of (2z−1)3
along the circle C(0, 5)?

3. Which of these functions are bounded on C?

z 1 cos(π/2 − z)
|z|, , , , sin(z 2 ).
|z| + 1 z−1 sin z

4. How many roots does (z 2 + 1)2 + 1 have?

5. Which of these functions are entire with bounded derivative?

iz, iz 2 , eiz , sin iz

6. Locate the extrema points of | cos z| on the square {x+iy|0 ≤ x, y ≤ π}.

Answers: 1. 0, 2. 5πi/8, 3. The second and the fourth. 4. 4, 5. iz. 6.

π
2
, πi, π
+ πi.
44 CHAPTER 4. INTEGRATION FORMULAS
Chapter 5

Series, Zeros, and Poles

5.1 Taylor Series

1
Example 30. The function z−1 is holomorphic on the unit disk D. We can
always this function in terms of a complex power series. Specifically, for
z ∈ D,
∞
1 X
= zn.
1−z n=0
Indeed, basic arithmetic results in sequences and series tell us that the partial
sums can be expressed as
N
X 1 − z N +1
zn = .
n=0
1−z

Since |z| < 1, |z N +1 | → 0 and clearly z N +1 → 0 too, as N → ∞. Therefore,

1
the partial sums converge to 1−z as N → ∞.
Example 31. The principal branch of the logarithm Log(1 − z) is holomor-
1
phic on D as well. As this is a primitive of z−1 , we can obtain a power series
1
for this function by integrating the power series for z−1 obtained. It is given
by:
∞
X zn
Log(1 − z) = −
n=0
n

Suppose aPholomorphic function f (z) on a domain U coincides with some

∞ n
power series n=0 an (z − z0 ) about some point z0 in U . Differentiating
both functions at z0 n times, we find that the coefficients an are unique as
they must satisfy f (n) (z0 ) = n!an . Therefore, this power series must also be
unique.

45
46 CHAPTER 5. SERIES, ZEROS, AND POLES

Theorem 5.1 (Taylor’s Theorem). If f is a holomorphic function on an

open disk D(z0 , r), then for all z ∈ D(z0 , r),
∞
X f (n) (z0 )
f (z) = an (z − z0 )n , where an = .
n=0
n!

Proof. We can always assume that z0 = 0 without loss of generality because

when z0 6= 0, we can replace the function f with f (z + z0 ) where z ∈ D(0, r).
Pick any point z ∈ D(0, r). Let γ be a circle C(0, s) centered at 0 of
some radius s such that |z| < s < r. This curve γ separates the point z from
the boundary of the disk D(0, r). For any point w along this curve, since
| wz | = |z|
s
< 1, we have the following identity:
∞
1 1 1 1 X  z n
= · z = .
w−z w 1− w
w n=0 w

By Cauchy’s formula and the above identity,

I
1 f (w)
f (z) = dw
2πi γ w − z
I ∞
1 f (w) X  z n
= dw
2πi γ w n=0 w
∞  I 
X 1 f (w)
= n+1
dw z n
n=0
2πi γ w
∞ ∞
X f (n) (0) n
X
= z = an z n .
n=0
n! n=0

In particular, Taylor’s theorem tells us that if we know the derivatives of

f of every order at a point z0 , then we know f (z) for every z z0 .
Corollary 5.2. Let f be an entire function and z0 be any point in C. Then,
for any z ∈ C,
∞
X f (n) (z0 )
f (z) = an (z − z0 )n , where an = .
n=0
n!

Definition 13. Given a holomorphic function f on a domain U , The Taylor

series for f about a point z0 ∈ U is the infinite sum
∞
X f (n) (z0 )
(z − z0 )n .
n=0
n!
5.2. ZEROS 47

When z0 = 0, the expansion is often called the MacLaurin series for f . The
radius of convergence of the Taylor series is the largest possible radius R > 0
such that the series converges for |z − z0 | < r. If there is no such maximum
value (e.g. when f is entire), we say that R = ∞.
1
Example 32. Examples 30 and 31 show the MacLaurin series of 1−z and
Log(1 − z). Both series cannot be extended beyond the unit disk since imme-
diately at z = 1, both functions are not well-defined. Therefore, both have
radius of convergence 1.

Example 33. The exponential function ez−z0 is an entire function. Since its
nth derivative at z0 is 1 for all n, it has a Taylor series about z0 with infinite
radius of convergence given below:
∞
X (z − z0 )n
ez−z0 = .
n=0
n!

Using this series, we can also derive the corresponding series for the functions
sin, cos, sinh and cosh.
1
Example 34. Let us compute the Taylor series for f (z) = 2i−z
about 2.

1 1 1 1
= = · z−2
2i − z (2i − 2) − (z − 2) 2i − 2 1 − 2i−2
∞  n X ∞
1 X z−2 (z − 2)n
= =
2i − 2 n=0 2i − 2 n=0
(2i − 2)n+1
∞  n+1
X 1+i
= − (z − 2)n .
n=0
4
√ z−2
The radius of convergence is 2 2 because the series converges when | 2i−2 |<
√
1, or equivalently, |z − 2| < 2 2.

5.2 Zeros
Definition 14. Suppose f is a holomorphic function on a domain U . We say
that f has a zero at a point z0 ∈ U order k if f (k) (z0 ) 6= 0 and f (n) (z0 ) = 0
for all n < k. If k = 1, we say that the zero is simple.

Proposition 5.3. Let f be a holomorphic function on a domain U and let

z0 ∈ U . The following are equivalent.
48 CHAPTER 5. SERIES, ZEROS, AND POLES

(a) f has a zero at z0 of order k > 0,

P∞
(b) the Taylor series of f about z0 is of the form n=k an (z − z0 )n , ak 6= 0,

(c) there is some holomorphic function g on U such that g(z0 ) 6= 0 and for
each z ∈ U , f (z) = (z − z0 )k g(z).

Proof. (a) ⇐⇒ (b) follows immediately from Taylor’s theorem, and (c) ⇒ (b)
follows from direct computation of derivatives using Leibniz’s rule.
f (z)
Assume (b) holds. The function g(z) = (z−z k is holomorphic on U \{z0 }.
0)
About z0 , the Taylor series of f gives us a well-defined power series repre-
sentation of g:
P∞ ∞
an (z − z0 )n X
g(z) = n=k
= an+k (z − z0 )n .
(z − z0 )k n=0

Hence, g is holomorphic at z0 and g(0) = ak 6= 0. We then have (b) ⇒ (c).

Lemma 5.4. Let f be holomorphic on some disk D(z0 , r) and let {zn }n∈N be
an infinite sequence of distinct zeros of f such that zn → z0 . Then, f ≡ 0.

Proof. By continuity of f , f (z0 ) = 0. Let ∞ n

P
n=1 an (z − z0 ) be the Taylor
series of f about z0 . Assume for a contradiction that f 6≡ 0, then there
must be some non-zero coefficient within the Taylor series. Let k > 1 be the
smallest number such that ak 6= 0, then f has a zero of order k at z0 .
Let g be a holomorphic function in Proposition 5.3. By continuity of g,
since g( z0 ) 6= 0, there is a small δ > 0 such that g(z) 6= 0 for z ∈ D(z0 , δ).
However, since zn ∈ D(z0 , δ) for sufficiently high n,

0 = f (zn ) = g(zn )(z − zn )k 6= 0.

This is a contradiction.

Theorem 5.5 (Identity Theorem / Coincidence Principle). Let f and g

be holomorphic functions on a domain U and let {zn }n∈N be a sequence of
distinct points in U such that zn → z0 ∈ U and f (zn ) = g(zn ) for all N .
Then, f ≡ g.

Proof. Define the holomorphic function h = f − g. To prove the theorem,

it is sufficient to show that h ≡ 0 in U . Pick any point w ∈ U . Since U is
connected, there is some finite sequence of open disks Dn := D(wn , rn ) in U
for n = 0, . . . N such that w0 = z0 , wN = w and wn+1 ∈ Dn . (See Figure
4.1.)
5.3. LAURENT SERIES 49

The function h has a zero at every zn By the previous lemma, h ≡ 0 on

D0 . Since D0 ∩ D1 contains w1 as well as a sequence of points converging
to w1 , the same lemma tells us that h ≡ 0 on D1 . Inductively, we conclude
that h ≡ 0 for all z ∈ DN and in particular, h(w) = 0. As w is arbitrary,
h ≡ 0.

The theorem essentially says that a holomorphic function on some domain

is completely determined by its values on a countable subset of the domain.
Such property again hints at how rigid holomorphic functions are compared
to real differentiable functions.

Example 35. The only holomorphic function f (z) which has zeros on the
set of rational numbers Q is the zero function.

Example 36. The only holomorphic function f (z) satisfying f ( n1 ) = 1

n
for
all n ∈ N is the identity function f (z) = z.

5.3 Laurent Series

It is often useful to include terms with negative powers in a power series. This
allows the possibility of expressing a holomorphic function with singularity
at a single point as a power series.
2−z
Example 37. The function g(z) = z 2 −z 3
has a singularity at 0. The power
series about 0 for 0 < |z| < 1 is
∞
2 1 2 1X n
g(z) = + = + z
z 2 z(1 − z) z 2 z n=0
= 2z −2 + z −1 + 1 + z + z 2 + . . . .

We say that the series above is the Laurent series of g at 0.

Theorem 5.6 (Laurent’s Theorem). Let f be a holomorphic function on an

annular domain A = {z ∈ C : r < |z − z0 | < R} of inner and outer radii
r ∈ [0, ∞) and R ∈ (0, ∞] centered at z0 ∈ C. For all z ∈ A,
∞ I
X
n 1 f (z)
f (z) = an (z − z0 ) , where an = dz,
n=−∞
2πi γ (z − z0 )n+1

for any simple closed contour γ in A enclosing the disk D(z0 , r).
50 CHAPTER 5. SERIES, ZEROS, AND POLES

Figure 5.1: The curve α joining γ1 and γ2 .

Proof. Let’s assume again that z0 = 0 because when z0 6= 0, we can replace

the function f with f (z + z0 ) where r < |z| < R. Pick any point z ∈ A
and positive numbers s and S such that r < s < |z| < S < R. Let γ1
and γ2 be simple closed curves parametrizing the circles C(0, S) and C(0, s)
respectively.
If w ∈ γ1 , then wz < 1 and thus,

∞ ∞
1 1 1 1 X  z n X z n
= · z = = n+1
. (5.1)
w−z w 1− w
w n=0 w n=0
w

w
However, if w ∈ γ2 , then z
< 1 and thus,

∞ −1
1 1 1 1 X  w n X zm
= · w = = m+1
. (5.2)
z−w z 1− z
z n=0 z m=−∞
w

Pick an angle θ ∈ (−π, π]\{Arg(z)} and define a radial line segment

α : [s, S] → A where α(t) = teiθ . (See Figure 5.1.) Let σ be the closed curve
obtained by concatenating γ1 , α− , γ2− and α in order, and let γ0 be any small
circle centered at z lying entirely between γ1 and γ2 . Then, since w 7→ fw−z
(w)
is
holomorphic on the domain enclosed between σ and γ0 , by Cauchy’s formula
5.3. LAURENT SERIES 51

and deformation theorem,

I I
1 f (w) 1 f (w)
f (z) = dw = dw
2πi γ0 w − z 2πi σ w − z
I I I I 
1 f (w)
= − − + dw
2πi γ1 α γ2 α w−z
I I
1 f (w) 1 f (w)
= dw + dw.
2πi γ1 w − z 2πi γ2 z − w

By (5.1) and (5.2), we can convert the two integrals into the desired series.
∞
! −1
!
z n f (w) z m f (w)
I I
1 X 1 X
f (z) = dw + dw
2πi γ1 n=0 wn+1 2πi γ2 m=−∞ wm+1
∞  I  −1  I 
X 1 f (w) n
X 1 f (w)
= n+1
dw z + m+1
dw z m
n=0
2πi γ1 w m=−∞
2πi γ2 w
∞
X −1
X ∞
X
= an z n + am z m = an z n .
n=0 m=−∞ n=−∞

Definition 15. The bi-infinite series in the theorem above is called the Lau-
rent series of f about the point z0 .

The Laurent series of a holomorphic function P∞ about a pointn is unique.

This is because if it coincides with some series n=−∞ an (z − z0 ) , then each
an necessarily satisfies the equation given in the theorem. Note that the an ’s
are independent of the choice of γ by deformation theorem.
1
Example 38. The Laurent series of z−1
about 0 is

∞  n −1
1 1 1 1X 1 X
= · 1 = = zn.
z−1 z 1− z
z n=0 z n=−∞

This series converges when | z1 | < 1, or equivalently, when 1 < |z| < ∞.
3z
Example 39. Let’s find a Laurent series for f (z) = z2 +z−2 converging on
1 2
the annulus {1 < |z + 1| < 2}. By partial fractions, f (z) = z−1 + z+2 has
1
singularities at −1 and 2. The Taylor series of z−1 about −1 is
∞  n ∞
1 1 1 1X z+1 X
=− · z+1 = − =− 2−n−1 (z + 1)n .
z−1 2 1− 2 2 n=0 2 n=0
52 CHAPTER 5. SERIES, ZEROS, AND POLES

convergent when | z+1

2
| < 1, i.e. in the domain U1 = {z : |z + 1| < 2}. The
2
Laurent series of z+2 about −1 is
∞  n −1
2 2 1 2 X 1 X
= · 1 = − = 2(−1)n+1 (z + 1)n .
z+2 z + 1 1 + z+1 z + 1 n=0 z+1 n=−∞

1
convergent when | z+1 | < 1, i.e. in the domain U1 = {z | 1 < |z + 1|}.
Combining the two series, we deduce that the Laurent series of f about −1
is
∞
(
X −2−n−1 , if n ≥ 0,
f (z) = an (z + 1)n , where an = n+1
n=−∞
2(−1) , if n < 0,
convergent in the annulus U1 ∩ U2 = {z | 1 < |z + 1| < 2}.

5.4 Singularities
We will use an asterisk on a disk to introduce a puncture at its centre:

D(z0 , r)∗ := D(z0 , r)\{z0 } = {z ∈ C : 0 < |z − z0 | < r}.

Definition 16. A point z0 is a singularity of a function f if

• f is not holomorphic at z0 , and
• for every r > 0, there exists a point in the punctured neighbourhood
D(z0 , r)∗ at which f is holomorphic.
Additionally, if there is some R > 0 such that f is holomorphic on some
punctured neighbourhood D(z0 , r)∗ , then z0 is an isolated singularity of f .
Example 40. The function (z − z0 )−d for any integer d > 0 has an isolated
singularity at z0 . In fact, the singularities of every rational function p(z)
q(z)
are
isolated since there are only finitely many of them.
1
Example 41. The function csc(2π/z) has singularities at 0 and k
for each
k ∈ Z. In particular, 0 is not isolated.
For the rest of the chapter, we will only focus on isolated singularities.
When isolated, Laurent’s theorem allows us to express the function in terms
of a Laurent series on a punctured neighbourhood of the singularity.
Definition 17. Suppose the Laurent P∞series of a holomorphic function f
n
about an isolated singularity z0 is n=−∞ an (z − z0 ) valid in a for 0 <
|z − z0 | < R for some radius R > 0. We say that z0 is:
5.4. SINGULARITIES 53

• an essential singularity if an 6= 0 for infinitely many integers n < 0,

• a pole of order k if a−k 6= 0 and an = 0 for all n < −k,

• a removable singularity if an = 0 for all n < 0.

We say that a pole is simple if it has order k = 1.

Example 42. e1/z has an essential singularity at 0 because for z ∈ C∗ ,

∞
1/z
X 1 −1 z −2 z −3
e = =1+z + + ...
n=0
n!z n 2 6

Proposition 5.7. Let f be a holomorphic function on a punctured domain

U \{z0 }. The following are equivalent:

(a) f has a pole at z0 of order k,

(b) there is a holomorphic function g on U such that g(z0 ) 6= 0 and for all
z 6= z0 ,
g(z) = (z − z0 )k f (z).

Proof. Assume (a) holds. Let the Laurent series of f about z0 be ∞

P
n=−k an (z−
n
z0 ) where a−k 6= 0. Define a holomorphic function g on PU∞\{z0 } by g(z) =
(z − z0 ) f (z). Since the Laurent series of g about z0 is n=0 an+k (z − z0 )n ,
k

g has a removable singularity at z0 . Setting g(z0 ) = a−k 6= 0, then g is

holomorphic at z0 as well. P Thus, (a) ⇒ (b).
Assume (b) holds. Let ∞ n
n=0 bn (z − z0 ) be the Taylor series of g about
z0 , then the Laurent series of f about z0 is
∞
g(z) X
f (z) = = bn+k (z − z0 )n .
(z − z0 )k n=−k

As b0 6= 0, f has a pole of order k. This gives us (b) ⇒ (a).

Definition 18. A function f is meromorphic on a domain U if it is holo-

morphic except at some countably many number of poles.

When z0 is a removable singularity of f , the Laurent series has no terms

of negative powers and hence becomes a Taylor series. We can then remove
the singularity and make f holomorphic at z0 by defining f (z0 ) = a0 , where
a0 = limz→z0 f (z) is the zeroth coefficient of the Taylor series.
54 CHAPTER 5. SERIES, ZEROS, AND POLES

Example 43. has a removable singularity at 0 because for z ∈ C∗ ,

sin z
z

z3 z5 z2 z4
 
sin z 1
= · z− + − ... = 1 − + − ...
z z 3! 5! 3! 5!
sin z sin 0
Since limz→0 z
= 1, we may set 0
= 1 so that it becomes an entire
function.
Theorem 5.8 (Riemann’s Removable Singularity). Suppose U is a domain
and f is a holomorphic function on U \{z0 } with a singularity at z0 . The
following are equivalent.
(a) z0 is a removable singularity.

(b) f is continuously extendable to z0 ,

(c) f is bounded on a small punctured disk D(z0 , r)∗ centered at z0 .

Proof. The implication (a) ⇒ (b) is clear and (b) ⇒ (c) follows from con-
tinuity at z0 . Suppose (c) holds and assume without loss of generality that
z0 = 0. There is some upper bound M > 0 for |f | on the punctured disk.
When |z| → 0, |zf (z)| ≤ |z|M → 0. As such, the function
(
z 2 f (z), z ∈ D(0, r)∗ ,
g(z) =
0, z = 0.

is continuous at 0. In fact, it is also holomorphic at 0 since

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

The Taylor series of h will be of the form b2 z 2 + b3 z 3 + . . .. Since f (z) =

z −2 h(z) for any z ∈ D(0, r)∗ , the Laurent series for f about 0 is b2 + b3 z + . . .
and it has no negative power terms. Therefore, 0 is a removable singularity.

5.5 Counting Zeros and Poles

Consider a closed curve γ : [a, b] → C avoiding the origin 0. In polar coor-
dinates, γ(t) = r(t)e2πiθ(t) for some continuous functions r(t) and θ(t) such
that r(a) = r(b) and θ(a) = θ(b) mod 1.
Definition 19. The winding number W (γ) of γ is the number of counter-
clockwise turns γ makes around 0, i.e. W (γ) = θ(b) − θ(a).
5.5. COUNTING ZEROS AND POLES 55

Figure 5.2: Winding number of various closed curves

Lemma 5.9. Let γ : [a, b] → C be a closed contour avoiding the origin 0.

Then, I
1 1
W (γ) = dz.
2πi γ z

Proof. Express γ in terms of polar coordinates: γ(t) = r(t)e2πiθ(t) . Since

γ 0 (t) = r0 (t)e2πiθ(t) + 2πiθ(t)γ(t),
b
γ 0 (t)
I Z
1 1 1
dz = dt
2πi γ z 2πi a γ(t)
b 0
Z
1 r (t)
= + 2πiiθ0 (t)dt
2πi a r(t)
1
= [log r(b) − log r(a)] + θ(b) − θ(a)
2πi
= θ(b) − θ(a) = W (θ).

Example 44. Let γ(t) = e2πit , 0 ≤ t ≤ 1 parametrise the unit circle. The
image of γ under the power map f (z) = z n , where n ≥ 1, is f ◦ γ(t) = e2πint .
The winding number of f ◦ γ is n, which coincides with the order of the zero
of f at 0.

This observation is generalised by the argument principle.

Theorem 5.10 (Argument Principle). Let f be a meromorphic function on

a simply connected domain U and γ be a simple closed contour in U along
which f has no zeros or poles. Let V be the domain enclosed by γ. Then,
I 0
1 f (z)
W (f ◦ γ) = dz = Z − P,
2πi γ f (z)

where Z is the number of zeros of f in V counting multiplicities (i.e. each

is counted as many times as its order), and P is the number of poles of f in
V counting multiplicities.
56 CHAPTER 5. SERIES, ZEROS, AND POLES

Proof. Let the parametrization of γ be γ : [a, b] → U . From the previous

lemma,
Z b
(f ◦ γ)0 (t)
I
1 1 1
W (f ◦ γ) = dz =
dt
2πi f ◦γ z 2πi a f γ(t)
Z b 0
I 0
1 f γ(t) 0 1 f (z)
=
γ (t)dt = dz
2πi a f γ(t) 2πi γ f (z)

Therefore, we have the first part of the equation. Suppose {z1 , . . . zm } and
{w1 , . . . wn } are the sets of zeros and poles in V respectively.
Pick any zero zj and let kj be its order. There is some meromorphic
function gj such that f (z) = (z −zj )kj gj (z) and gj is holomorphic at zj where
gj (zj ) 6= 0. Pick a small radius j > 0 so that inside the closed disk D(zj , j )
gj have no poles nor zeros aside from zj . Let γj be the circle C(z1 , j ), then

f 0 (z) kj (z − zj )kj −1 gj (z) + (z − zj )k gj0 (z)

I I
1 1
dz = dz
2πi γj f (z) 2πiγj (z − zj )k gj (z)
gj0 (z)
I I
kj 1 1
= dz + dz = kj ,
2πi γj z − zj 2πi γj gj (z)

where the last equality follows from the fact that gj0 (z)/gj (z) is holomorphic
on D(zj , j ) and Cauchy-Goursat.
Pick any pole wj and let lj be its order. There is some meromorphic
function hj such that f (z) = hj (z)(z − wj )−lj and hj is holomorphic at wj
where hj (zj ) 6= 0. Pick a small radius δj > 0 so that inside the closed
disk D(wj , δj ) g have no poles nor zeros aside from wj . Let σj be the circle
C(wj , δj ), then

f 0 (z) −lj (z − zj )lj −1 hj (z) + (z − wj )lj h0j (z)

I I
1 1
dz = dz
2πi σj f (z) 2πiσj (z − wj )lj hj (z)
h0j (z)
I I
lj 1 1
=− dz + dz = −lj ,
2πi σj z − wj 2πi σj hj (z)

where the last inequality follows from the fact that h0j (z)/hj (z) is holomorphic
on D(wj , δj ).
The curve γ can be split into some m + n simple closed contours each of
which encloses exactly one zero or pole. By deformation theorem, the integral
along γ is equal to the sum of the integrals along each of the contours γj for
j = 1 . . . m and σj for j = 1 . . . n.
5.5. COUNTING ZEROS AND POLES 57

Figure 5.3: The contour γ can be split into a collection of m+n simple closed
contours each of which encloses exactly one zero or pole and deformed into
γj ’s and σj ’s.

The argument principle gives us two ways of computing the difference

between the number of zeros and the number of poles inside some domain.
One way is more geometric: by computing the winding number of the image
of the boundary of the domain. The other is analytic: by computing a
contour integral along the boundary. Both ways are equally useful.

Example 45. Let’s compute the integral of sec z along a square γ of side
length 7 centered at 0 by using the fact that sec0 z = sec z tan z. There are
four simple poles of sec enclosed by γ, namely ± π2 , ± 3π
2
. sec has no zeros.
By the argument principle,
I I
sec z tan z
tan z dz = dz = 2πi · (−4) = −8πi.
γ γ sec z

Theorem 5.11 (Rouché’s Theorem). Let f and g be holomorphic functions

on a domain U , γ be a simple closed contour on U , and V be the domain
enclosed by γ such that V ⊂ U . If |g(z)| < |f (z)| for all z along γ, then f
and f + g have the same number of zeros, counting multiplicities, inside V .

g(z)
Proof. Define a meromorphic function h(z) = f (z)
+ 1 on U . Along γ, h is
holomorphic and |h(z) − 1| = fg(z)
(z)
< 1. Therefore, the contour h ◦ γ lies in
the disk D(1, 1) disjoint from 0 and consequently has zero winding number.
Denote by Zf +g and Zf the respective numbers of zeros of f + g and f inside
58 CHAPTER 5. SERIES, ZEROS, AND POLES

the domain enclosed by γ. For any z along γ,

f 0 (z) + g 0 (z) f 0 (z) g 0 (z)f (z) − f 0 (z)g(z)

− =

f (z) + g(z) f (z) f (z) f (z) + g(z)
g 0 (z)f (z) − f 0 (z)g(z) 1 h0 (z)
= · g(z)
= .
f (z)2 +1 h(z)
f (z)

Therefore, by the argument principle,

I 0
1 f (z) + g 0 (z) f 0 (z)
Zf +g − Zf = − dz
2πi γ f (z) + g(z) f (z)
I 0
1 h (z)
= dz = W (h ◦ γ) = 0.
2πi γ h(z)

Example 46. The polynomial p(z) = z 5 + 3z 2 + 6z + 1 has exactly one

zero inside the unit disk D. Indeed, let’s split p into f (z) = 6z and g(z) =
z 5 + 3z 2 + 1. The function f has a simple zero at 0. When |z| = 1,

|g(z)| = |z 5 + 3z 2 + 1| ≤ |z|5 + 3|z|2 + 1 = 5 < 6 = |6z| = |f (z)|.

As such, inside D, p has the same number of zeros as f , which is 1.

Rouché’s theorem also provides a much shorter proof of the fundamental

theorem of algebra. In fact, it also gives us a rough estimate of where the
zeros of a polynomial can be found.
Pd n
proof of the Fundamental Theorem of Algebra. Let p(z) = n=0 an z be a
d−1
degree d polynomial. Split p into f (z) = ad z d and g(z) = n=0 an z n . Clearly,
P
f has a zero of order d at 0. Pick any positive number R such that
1
an d d−n
R> max ,
n=0,1,...d−1 ad
|ad |Rd
then |an |Rn < d
for each n = 0, 1, . . . d − 1. For |z| = R, by triangle
inequality,
d−1
X d−1
X
n
|g(z)| ≤ |an z| = |an |Rn < |ad |Rd = |f (z)|.
n=0 n=0

By Rouché’s theorem, p = f + g has d zeros, counting multiplicities, all of

which lie inside the disk D(0, R).
5.5. COUNTING ZEROS AND POLES 59

Short Quiz 5

1. Suppose the Taylor series of cos(iz+π) at π is a+b(z−π)+c(z−π)2 +. . ..

What is the value of c?

2. Suppose f is an entire function. Which of the following criteria imply

that f is the zero function?

(a) f (z) = 0 for all natural numbers z

(b) f (z) = 0 for all integers z
(c) f (z) = 0 for all rational numbers z
(d) f (z) = 0 for all real numbers z

sin z cos z −2
, , e(z−π) , tan((z − π)−1 )
z−π z−π
3. Which of the four functions above have a removable singularity at π?

4. Which of the four functions about have an essential singularity at π?

5. How many zeros of z 4 + 2z 2 + 5z + 2 lie inside the disk D(0, 3) centered

at 0 of radius 3?
sin z −2
Answers: 1. cos(iπ + π)/2, 2. (c), (d), 3. z−π
, 4. e(z−pi) , 5. 2.
60 CHAPTER 5. SERIES, ZEROS, AND POLES
Chapter 6

Evaluation of Integrals

6.1 Residue Theory

Let f be a meromorphic function on a domain P U and supposen its Laurent
series representation at some point z0 ∈ U is ∞
n=−∞ an (z − z0 ) .

Definition 20. The residue Resf (z0 ) of f at z0 ∈ U is defined to be the

coefficient a−1 .
If f is holomorphic at z0 , then clearly Resf (z0 ) = 0. Otherwise, let’s
suppose z0 is a pole of order k. If g(z) = (z − z0 )k f (z), then the Taylor series
of g about z0 is

g(z) = a−k + a−k+1 (z − z0 ) + . . . + a−1 (z − z0 )k−1 + . . .

To extract out a−1 from the series, we can take the (k − 1)th derivative of g
evaluated at z0 :
g (k−1) (z0 ) = (k − 1)!a−1 .
Therefore, when we are given the order k of the pole, the residue can be
computed as follows:

1 dk−1 h i
Resf (z0 ) = lim k−1 (z − z0 )k f (z) .
(k − 1)! z→z0 dz

Theorem 6.1 (Residue Theorem). Let f be a meromorphic function on a

simply connected domain U and γ be a simple closed contour in U along
which f has no poles. If {w1 , . . . wm } is the set of poles of f enclosed by γ,
then I m
X
f (z)dz = 2πi Resf (wj ).
γ j=1

61
62 CHAPTER 6. EVALUATION OF INTEGRALS

Proof. Pick a pole wj and let kj be its order. Let j > 0 be small enough such
that the only pole of f inside the closed disk D(wj , j ) is wj , and parametrize
the boundary of this disk by
γj (t) = wj + j e2πit , 0 ≤ t ≤ 1.
If the Laurent series of f about wj is ∞ n
P
n=−kj an (z − wj ) , then
I ∞
X I
f (z)dz = an (z − wj )n dz
γj n=−kj γj
∞
X Z 1
= an nj e2πitn · γj0 (t)dt
n=−kj 0

X Z 1 Z 1
= an 2πin+1
j e2πit(n+1) dt + a−1 2πidt
n≥−kj ,n6=−1 0 0

X e2πit(n+1) 1
= an n+1
j + 2πia−1
n≥−kj ,n6=−1
n+1 0

= 2πi Resf (wj ).

Similar to the proof of the argument principle, we finish this proof by splitting
γ into k simple closed curves, each of which contains only one pole, and
subsequently using deformation theorem to reduce the integral computation
to a sum of the above integrals for each j = 1 . . . m.
In the proof, we do not actually use the properties of wj ’s being poles.
As such, the residue theorem also applies to removable singularities (poles of
zero order) and essential singularities (poles of infinite order).
1
H
Example 47. Let’s evaluate I = γ f (z)dz where f (z) = z5 −z 3 and γ is the

circle C(0, 1/2). The only pole of f enclosed by γ is 0 and its order is 3. By
Residue Theorem,
d2
 
2πi 1
I = 2πi Resf (0) = lim
2! z→0 dz 2 z 2 − 1
−6z 2 + 2
= πi lim 2 = −2πi.
z→0 (z − 1)3

6.2 Jordan’s Lemma

Lemma 6.2 (Jordan’s Lemma). Suppose g is a holomorphic function on the
domain U = {z ∈ C : |z| > R0 , Imz > 0} for some radius R0 > 0. Let
6.3. DEFINITE INTEGRALS 63

γR = {Reit | 0 ≤ t ≤ π} be a semicircular curve of radius R > R0 . Then, for

any α > 0, Z
π
g(z)eiαz dz ≤ max |g(z)|.
γR α z∈γR
Proof. Let MR = maxz∈γR |g(z)|. Then,
Z Z π
iαz
g(z)e dz = g(Reit )eαR(i cos t−sin t) · iReit dt
γR
Z π0
≤ g(Reit )eαR sin t · iRei(t+αR cos t) dt
0
Z π
=R |g(Reit )|e−αR sin t dt
0
Z π
≤ RMR e−αR sin t dt.
0

To prove the theorem, it is sufficient to prove the following inequality:

Z π
π
e−αR sin t dt ≤ . (6.1)
0 αR
Since sin t is convex on [0, π2 ], sin t ≥ 2t π
on this interval. Since αR > 0,
e−αR sin t ≤ e−2αRt/π for 0 ≤ t ≤ π2 . Then, by symmetry of sin t about t = π2 ,
Z π Z π/2 Z π/2
−αR sin t −αR sin t π(1 − e−αR ) π
e dt = 2 e dt ≤ 2 e−2αRt/π dt = < .
0 0 0 αR αR
Thus, (6.1) holds and subsequently, the proof is done.
The lemma is particularly useful when g(z) → 0 as |z| → ∞. This
additional assumption is equivalent to saying that MR → 0 as R → ∞, and
by Jordan’s lemma, Z
g(z)eiαz dz → 0.
γR
When α = 0, the upper bound in Jordan’s lemma is undefined, but we
already have ML inequality in hand.

6.3 Definite Integrals

Residue theory, ML inequality and Jordan’s lemma are common tools used
to evaluate real integrals. We will start with a simple integral I of the form
Z 2π
I= F (cos(x), sin(x))dx,
0
64 CHAPTER 6. EVALUATION OF INTEGRALS

for some generic rational function F . With the substitution z = eix where
0 ≤ x ≤ 2π, we can transform I into a contour integral

z + z −1 z − z −1 dz
I  
I= F , ,
C(0,1) 2 2i iz

which can be solved by residue theorem.

Example 48. Let’s evaluate

Z 2π
1
I= dx.
0 2 − cos x

By the above substitution,

I I
1 dz 1
I= z+z −1 iz
= 2i 2
dz.
C(0,1) 2 − 2 C(0,1) z − 4z + 1
√
√ integrand by f . Observe that f has simple poles at 2 ±
Denote the new 3,
but only 2 − 3 is inside the unit disk. By residue theorem,
√ 1 2π
I = 2i · 2πiResf (2 − 3) = −4π lim√ √ =√ .
z→2− 3 z − 2 − 3 3
R
Our aim is to generalise to an improper integral I = J f (x)dx where J
is either (0, ∞) or (−∞, ∞). The general method is as follows:

1. Pick any arbitrarily large R > 0.

2. Cook up a simple closed contour γ in C consisting of γ1 = (0, R) or

(−R, R) and a few other smooth pieces γ2 , . . . γm , such that as R → ∞,
the number of poles enclosed by γ remains constant.
H
3. Evaluate I0 = γ f (z)dz using residue theorem.
R
4. Reduce the limit as R → ∞ of each Ik = γk f (z)dz, k 6= 1, to either
a multiple of I or a constant value (typically 0) by Jordan’s lemma or
ML inequality.
R
5. Evaluate I as the limit as R → ∞ of I1 = γ1 f (z)dz.

We will explore in detail different types of integrals you can solve accord-
ing to the type of integrand f and the shape of the contour γ used.
6.3. DEFINITE INTEGRALS 65

Semicircular Contour
Consider integrals of the type
Z ∞
I= f (x)dx,
−∞

where the function f has no singularities in R and has a finite number of

poles throughout C. To evaluate I, we need to use a semicircular contour γ
consisting of

γ1 = [−R, R] and γ2 = {Reit | 0 ≤ t ≤ π},

The most crucial part is to show that I2 → 0 as R → ∞. Not all functions

will satisfy this, but when it does, the value of I follows immediately:

I = lim I1 = lim I0 − I2 = I0 .
R→∞ R→∞

Example 49. Suppose the integrand is f (x) = (x6 + 1)−1 . This function
can be extended to a meromorphic function on C and it has 6 simple poles
forming the set {z | z 6 = −1} = {±i, e±πi/6 , e±5πi/6 }. The ones on the upper
half plane are i, eπi/6 , e5πi/6 .
By L’Hopital’s rule, the residue of f at any pole c is
z−c 1 c c
Resf (c) = lim 6
= lim 5 = 5 = − .
z→c z + 1 z→c 6z 6c 6
Let’s pick a large R > 0 and define γ as outlined before. By residue
theorem,
I
I0 = f (z)dz = 2πi Resf (i) + Resf (eπi/6 ) + Resf (e5πi/6 )


γ
eπi/6 e5π/6
 
i πi
2π
= 2πi − − − =− i + eπi/6 + e5πi/6 = .
6 6 6 3 3
66 CHAPTER 6. EVALUATION OF INTEGRALS

Next, we will show that I2 → 0. By ML and triangle inequalities, as R → ∞,

Z
πR πR
f (z)dz ≤ L(γ2 ) max |f (z)| = 6
≤ 6 → 0.
γ2 z∈γ2 min|z|=R,Imz≥0 |z + 1| R −1
2π
Therefore, I = I0 = 3
.

Example 50. Calculating the integrals of functions such as

cos x sin x
and
x2 + 1 x2 + 1
over R is common in Fourier analysis. To do so, it is easier to firstly integrate

eiz
f (z) =
z2 + 1
and then take its real and imaginary parts (since cos z = Re eiz and sin z =
Im eiz ). The function f has simple poles at ±i. With the same setup as
before,
e−1
I
I0 = f (z)dz = 2πi Resf (i) = 2πi = πe−1 .
γ i+i
Since
1 1
lim ≤ lim 2 = 0,
R→∞ z 2+1 R→∞ R − 1

we can immediately conclude by Jordan’s lemma that I2 → 0 as R → ∞.

Therefore, I = πe−1 . Taking the real and imaginary parts,
Z ∞ Z ∞
cos x −1 sin x
2
dx = πe , and 2
dx = 0.
−∞ x + 1 −∞ x + 1

Indented Semicircular Contour

Consider integrals of the type
Z ∞
I= f (x)dx,
−∞

where the function f has a finite number of poles throughout C, with 0 being
the only real pole. Due to the presence of singularity at 0, we need to add
a little dent to our semicircular contour. We pick an arbitrarily small ρ > 0
and construct γ by gluing together the following curves:

γ1 = [ρ, R], γ2 = {Reit |0 ≤ t ≤ π}, γ3 = [−R, −ρ], γ4 = {ρeit |0 ≤ t ≤ π}.

6.3. DEFINITE INTEGRALS 67

Once we show that I2 as R → ∞, the value of I can be computed by

taking the following limit:

I= lim I1 + I3 = I0 − lim I4 .
R→∞,ρ→0 ρ→0

Remark. When f has a singularity at a non-zero point in R, we may introduce

a dent around that point similar to our construction above.
Example 51. Let’s compute the integral I of f (x) = eix /x over R. f has a
single pole at 0. Introduce the indented semicircular contour γ as outlined
above. Since f is holomorphic along γ and on the region enclosed by γ,
I0 = 0 by Cauchy-Goursat.
We can show by Jordan’s lemma that I2 → 0 as R → ∞. When ρ → 0,
Z iz Z π iρeit Z π
e e it it

lim I4 = lim dz = − lim it
iρe dt = −i lim eiρe dt = −πi.
ρ→0 ρ→0 γ z ρ→0 0 ρe 0 ρ→0
4

Therefore, I = πi. Taking the real and imaginary parts, we obtain the
following two integrals as well:
Z ∞ Z ∞
cos x sin x
dx = 0, and dx = π.
−∞ x −∞ x

Sector Contour
Consider integrals of the type
Z ∞
I= f (x)dx,
0

where f has a finite number of poles throughout C, none is located along

[0, ∞), and f exhibits some kind of rotational symmetry. More precisely,
there is some angle θ and constant c such that f (eiθ z) = cf (z).
68 CHAPTER 6. EVALUATION OF INTEGRALS

Let γ be a sector contour consisting of

γ1 = [0, R], γ2 = {Reit | 0 ≤ t ≤ θ}, and γ3 = {reiθ | 0 ≤ r ≤ R}.

Rotational symmetry will imply that there is some constant C such that
Z Z
f (z)dz = C f (z)dz.
γ3 γ1
After showing by ML inequality that I2 → 0 as R → ∞,
I = lim I1 = lim I0 − I2 − I3 = I0 − CI
R→∞ R→∞
−1
Therefore, I = (1 + C) I0 .
Example 52. Let’s evaluate I when f (x) = (1 + x3 )−1 . This integrand has
rotational symmetry of angle θ = 2π 3
since
1 1
f (e2πi/3 z) = = = f (z).
1 + e2πi z 3 1 + z3
The poles of f are −1 and e±iπ/3 , each of which is simple. Pick a large R
and define a sector contour γ of angle 2π 3
as outlined above. Since eiπ/3 is the
only pole enclosed by γ,
z − eiπ/3 1 2πeπi/3
I0 = 2πi Resf (eiπ/3 ) = 2πi lim = 2πi lim = .
z→eiπ/3 z3 + 1 z→eiπ/3 3z
2 3
Let’s parametrize γ3 by γ3 (r) = re2πi/3 as r varies from R to 0. Then,
Z 0 2πi/3 Z R
e 2πi/3 1
I3 = 3
dr = −e 3
dr = −e2πi/3 I1 .
R 1 + r 0 1 + r
By ML inequality, we have that I2 → 0 because
Z
2πR/3 2πR
f (z)dz ≤ L(γ2 ) max |f (z)| = 3
≤ → 0.
γ2 z∈γ2 min|z|=R,Imz≥0 |z + 1| 3(R3 − 1)
Thus,
2π
I = (1 − e2πi/3 )−1 I0 = √ .
3 3
6.3. DEFINITE INTEGRALS 69

Keyhole Contour
Consider integrals of the type
Z ∞
I= f (x)dx.
0

where f contains some term xa where 0 < a < 1 is a rational number. We

have to pick a branch cut and make sure that our contour avoids it. The
best choice of branch cut is usually the positive real axis [0, ∞). There are a
few variants of the contour γ, but it is typically split into four parts:

γ1 = {rei | ρ ≤ r ≤ R}, γ2 = {Reit |  ≤ t ≤ 2π − },

γ3 = {rei(2π−) | ρ ≤ r ≤ R}, γ4 = {ρeit |  ≤ t ≤ 2π − }.

Additional parameters ρ and  are needed and they are taken to be arbitrarily
small.

Example 53. Let’s evaluate the integral I of

√
x
f (x) =
(x + 1)2

over (0, ∞). Taking the branch cut of the square root to be arg z = 0, we
complexify f to a meromorphic function on C\[0, ∞) with a pole at −1 of
order 2. By residue theorem,

d√
I0 = 2πi Resf (−1) = 2πi lim z = π.
z→−1 dz
70 CHAPTER 6. EVALUATION OF INTEGRALS

The limits of I1 and I3 are:

Z √ Z R √ i/2
z re
lim I1 = lim dz = lim ei dr
R→∞,ρ,→0 R→∞,ρ,→0 γ (z + 1)2 R→∞,ρ,→0 ρ (rei + 1)2
Z ∞ √ 1
r
= dr = I.
0 (r + 1)2
Z √ Z ρ √ i(π−/2)
z re
lim I3 = lim dz = lim ei(2π−) dr
R→∞,ρ,→0 R→∞,ρ,→0 γ (z + 1)2 R→∞,ρ,→0 R (re−i + 1)2
3
Z ∞ √
r
= −eiπ dr = I.
0 (r + 1)2

By ML inequality and triangle inequality, as R → ∞ and ρ,  → 0,

√
z R3/2
|I2 | ≤ (2π − 2)R max ≤ (2π − 2) → 0,
z∈γ2 (z + 1)2 (R − 1)2
√
z ρ3/2
|I4 | ≤ (2π − 2)ρ max ≤ (2π − 2) → 0.
z∈γ4 (z + 1)2 (1 − ρ)2

Collecting all the integrals together,

I1 + I2 + I3 + I4 = I0
I +0+I +0=π
π
∴I= .
2
Chapter 7

Harmonic Functions

7.1 Harmonicity
Definition 21. Let U ⊂ R2 be a non-empty open subset. A real-valued
function u : U → R is harmonic if it is continuously twice differentiable and
it satisfies the Laplace’s equation:

∆u := uxx + uyy = 0.
2
∂ 2
∂
The operator ∆ = ∂x 2 + ∂y 2 is known as the Laplacian / Laplace operator.

There are many instances in which polar coordinates are used. With the usual
change of variables (x, y) = (r cos θ, r sin θ), this operator becomes

∂2 1 ∂ 1 ∂2
∆= + +
∂r2 r ∂r r2 ∂θ2
in polar coordinates.

Example 54. Affine functions u(x, y) = ax + by + c for some real constants

a, b, c are harmonic on R2 .

Example 55. The function u(x, y) = sin x cosh y is harmonic on R2 .

Example 56. Denote the real and imaginary parts of the principal logarithm
Log : C\(−∞, 0] as u and v accordingly. In Cartesian coordinates,
p y
u(x, y) = ln x2 + y 2 , v(x, y) = tan−1 .
x
It may not seem clear at first that both u and v are harmonic, but in polar
coordinates, u(r, θ) = ln r and v(r, θ) = θ and it is much easier to show that
these harmonic whenever r > 0 and −π < θ < π.

71
72 CHAPTER 7. HARMONIC FUNCTIONS

Harmonic functions naturally appear in many fields of applied mathemat-

ics and physics. In probability theory, the probability of that a Brownian
motion inside a domain hits part of the boundary is governed by harmonic
functions. In fluid dynamics, the velocity potential in an incompressible
irrotational fluid satisfies the Laplace’s equation. In physics, electrostatic
and gravitational potential satisfies the Poisson equation, a generalisation of
Laplace’s equation. One of the main reasons complex analysis is often an es-
sential tool for many physicists is that holomorphic functions and harmonic
functions are very much interrelated.
Proposition 7.1. Let f = u + iv be a holomorphic function on a domain
U ⊂ C, then both u and v are harmonic.
Proof. By Cauchy-Riemann equations,
∆u = uxx + uyy = vyx − vxy = 0, ∆v = vxx + vyy = −uyx + uxy = 0.
Hence, u and v are harmonic.
Proposition 7.2. Let u be a harmonic function on a simply connected do-
main U , then there is a harmonic function v on U such that f = u + iv is
holomorphic on U , and it is unique up to an additive constant.
Proof. It is sufficient to find a solution v to the first order partial differential
equations vx = −uy and vy = ux . By (2.1) and (2.2), this is equivalent to
finding a primitive f = u + iv of the holomorphic function f 0 = ux − iuy . By
Corollary 3.8, such a primitive exists since U is simply connected, and it is
unique up to an additive constant a + ib. Since we fix the real part of f to
be u, then a = 0 and all harmonic conjugates of u are v(x, y) + b for some
real constant b.
In the theorem above, we say that v is a harmonic conjugate to u.
Example 57. The function u(x, y) = x2 − y 2 is a harmonic function on R2 .
Suppose v is a harmonic conjugate of u. The partial derivatives of u are
ux = 2x and uy = −2y. By Cauchy-Riemann,
vx = −uy = 2y, vy = 2x.
By integrating vx with respect to x, v(x, y) = 2xy + c(y) for some real
differentiable function c(y). Differentiating v with respect to y, we obtain
that vy = 2x + c0 (y) = 2x, which implies that c0 (y) = 0. Therefore, any
harmonic conjugate of u is of the form
v(x, y) = 2xy + c,
for any real constant c. Indeed, the corresponding holomorphic function is a
quadratic f (z) = z 2 + ic.
7.2. KEY PROPERTIES 73

Example 58. The function u(r, θ) = ln r, written in polar coordinates, is

harmonic in R2 \{0}. The proposition above fails because the punctured
plane is multiply connected. Indeed, we see that v(r, θ) = θ is the harmonic
conjugate of u but only after we introduce a branch cut; v cannot be extended
to a harmonic conjugate on the whole R2 \{0}.

7.2 Key Properties

We shall transfer a number of properties on holomorphic functions we have
already known to harmonic functions. From now on, harmonic functions
u(x, y) will sometimes be expressed as a function u(x + iy) of one complex
variable without any ambiguity.
Proposition 7.3 (Mean Value Property). Let u be a harmonic function on
a domain U . For any closed disk D(z0 , ) in U ,
Z 2π
1
u(z0 ) = u(z0 + eit )dt.
2π 0
Proof. Let f be a holomorphic function on U such that u(z) = Ref (z). By
Corollary 4.2,
 Z 2π 
1 it
u(z0 ) = Ref (z0 ) = Re f (z0 + e )dt
2π 0
Z 2π
1
= u(z0 + eit )dt.
2π 0
We may also take the mean value of u over the whole disk enclosed by
the circle and obtain the same result.
Corollary 7.4 (Volume Mean Value Property). Let u be a harmonic function
on a simply connected domain U ⊂ C. The average value of u inside any
closed disk D(z0 , r) lying inside U is equal to u(z0 ), i.e.
Z
1
u(z0 ) = 2 u(z)dA.
π D(z0 ,)
Proof. By expressing the area element dA in polar coordinates,
Z Z  Z 2π
1 1
2
u(z)dA = 2 u(z0 + reit )rdtdr
π D(z0 ,) π 0 0
2  1
Z  Z 2π 
it
= 2 u(z0 + re )dt rdr
 0 2π 0
2 
Z
= 2 u(z0 ) · rdr = u(z0 ).
 0
74 CHAPTER 7. HARMONIC FUNCTIONS

It is worth noting that the only ingredients of the proof of the maximum
modulus principle of holomorphic functions in Lemma 4.11 and Theorem 4.12
are continuity, mean value property and connectivity of the domain. Since
these three properties are satisfied by every harmonic function on a domain,
harmonic functions automatically satisfy the maximum modulus principle.
Nonetheless, we shall give a slicker proof using harmonic conjugates.
Theorem 7.5. Harmonic functions satisfy the maximum modulus principle.
Proof. Let u be a harmonic function on a domain U ⊂ C. Pick a har-
monic conjugate v of u on U and define f (x + iy) = u(x, y) + iv(x, y) on U .
The function is a non-constant holomorphic function ef (z) with modulus is
|ef (z) | = eu (x, y). Applying the maximum modulus principle to ef (z) , then
eu and therefore u(x, y) do not attain maximum on U . We can apply the
same argument on −u to conclude that u does not attain minimum on U .
Therefore, |u| does not attain maximum on U , and if U is bounded, maxima
are achieved only on the boundary.

7.3 Poisson Integral Formula

Mean value property is an analogue of the Cauchy integral formula, but it
only tells us the value of the function at the center of the circle. In this
section, we wish to obtain a better analogue of the Cauchy integral formula
to obtain the value of the function at any point enclosed by the circle.
Definition 22. The Poisson kernel P (r, θ) for the unit disk D is given by
1 − r2
P (r, θ) = ,
1 − 2r cos θ + r2
where reiθ ∈ D.
By the cosine rule, the denominator is equal to the square of the length
of a side of a triangle whose other sides have length 1 and r and make an
interior angle θ (mod 2π). Let z = reiθ ∈ D, then this triangle can be chosen
to have vertices 0, 1 and z, and the denominator is equal to |1 − z|2 . In
particular, P (r, θ) > 0.
The Poisson kernel can also be rewritten as the real part of a rational
function of z:
1 − |z|2
 
1+z
P (r, θ) = = Re .
|1 − z|2 1−z
In particular, the Poisson kernel is harmonic on D because this rational func-
tion is holomorphic on D.
7.3. POISSON INTEGRAL FORMULA 75

Theorem 7.6 (Poisson Integral Formula). Let u(r, θ) be a harmonic function

on a neighbourhood of D, written as a function in polar coordinates. For any
point z = reiθ ∈ D,
Z 2π
1
u(r, θ) = P (r, t − θ) u(1, t)dt.
2π 0
Proof. Let z1 = 1/z̄ = r−1 eiθ be the reflection of z in the unit circle. Since
z1 is outside the closed unit disk, if we denote by f a holomorphic function
whose real part is u, then by Cauchy-Goursat,
I
f (w)
dw = 0.
∂D w − z1

By Cauchy integral formula,

I
1 f (w)
f (z) = dw
2πi ∂D w − z
I I
1 f (w) 1 f (w)
= dw − dw
2πi ∂D w − z 2πi ∂D w − z1
I  
1 1 1
= f (w) − dw. (7.1)
2πi ∂D w − z w − z1
For any w = eit on the unit circle ∂D,
   
1 1 1 1
w − =w −
w − z w − z1 w − z w − ww̄/z̄
w z̄
= +
w − z w̄ − z̄
ww̄ − z z̄
=
ww̄ − wz̄ − w̄z + z z̄
1 − r2
= = P (r, t − θ). (7.2)
1 − 2r cos(t − θ) + r2
Using parametrization w = eit , 0 ≤ t ≤ 2π, of the unit circle ∂D, dw = iw dt.
Then, combining (7.1) and (7.2),
Z 2π  
1 it w w
f (z) = f (e ) − dt
2π 0 w − z w − z1
Z 2π
1
= f (eit )P (r, t − θ)dt.
2π 0
Taking the real part of both sides of the equation, we immediately obtain
the Poisson integral formula.
76 CHAPTER 7. HARMONIC FUNCTIONS

7.4 Dirichlet Problem for D

Solving Laplace’s equation with given boundary conditions goes under the
name of Dirichlet’s problem, and is one of the fundamental problems of partial
differential equations (PDE). Specifically, we wish to find a solution u of the
equation ∆u = 0 on some domain such that along the boundary of the
domain, u takes certain prescribed values.
Theorem 7.7 (Dirichlet Problem on RD). For any real-valued function F :
2π
[0, 2π] → R which is integrable, i.e. 0 |F (t)|dt < ∞, there is a unique
solution u : D → R to the following PDE problem:
∆u(r, θ) = 0, if 0 ≤ r < 1, 0 ≤ θ < 2π,
lim u(r, θ) = F (θ), for all 0 ≤ θ < 2π at which F is continuous.
r→1

Proof. Assume u and v are two such solutions, then u − v is a harmonic

function on D which is identically zero along ∂D. By the maximum modulus
principle, u − v ≡ 0. Therefore, u ≡ v. We have then proven uniqueness.
What remains is to prove existence of the solution u. If it does exist,
then it will satisfy the Poisson kernel formula we previously obtained. Let’s
define u using the Poisson kernel:
Z 2π
1
u(r, θ) = P (r, t − θ) F (t)dt.
2π 0
On D, u is continuously twice differentiable because P is continuously twice
differentiable. By chain rule, for any fixed value of t,
∂ ∂P ∂ ∂P
P (r, t − θ) = (r, t − θ), P (r, t − θ) = − (r, t − θ).
∂r ∂r ∂θ ∂θ
Therefore, since P is harmonic,

∂ 2P 1 ∂P 1 2 ∂
2
∆ P (r, t − θ) = (r, t − θ) + (r, t − θ) + (−1) (r, t − θ)
∂r2 r ∂r r2 ∂θ2
= (∆P )(r, t − θ) = 0.
This immediately implies that u is harmonic on D:
Z 2π
1
∆u(r, θ) = ∆P (r, t − θ) F (t)dt = 0.
2π 0
What remains is to check that u satisfies the boundary condition. By the
mean value property,
Z 2π
1
P (r, t − θ)dt = P (0, t − θ) = 1. (7.3)
2π 0
7.4. DIRICHLET PROBLEM FOR D 77

Consequently,
Z 2π
1
u(r, θ) − F (θ) = P (r, t − θ) (F (t) − F (θ))dt.
2π 0

Since P is positive on D,
Z 2π
1
|u(r, θ) − F (θ)| ≤ P (r, t − θ) |F (t) − F (θ)|dt. (7.4)
2π 0

To evaluate the last integral, we will split the interval into I = (θ − δ, θ + δ)

(mod 2π) and J = [0, 2π]\I. On I,
Z Z 
1 1
P (r, t − θ) |F (t) − F (θ)|dt ≤ P (r, t − θ)dt max |F (t) − F (θ)|
2π I 2π I θ−δ≤tθ+δ
1
≤ max |F (t) − F (θ)| (7.5)
2π θ−δ≤tθ+δ

The last inequality follows from (7.3). By continuity of F , maxθ−δ≤tθ+δ |F (t)−

F (θ)| → 0 as δ → 0.
When t ∈ J, P (r, t − θ) is well defined because it avoids the singularity
at t = θ. The distance between eiθ and the set {z ∈ D | arg z ∈ J} is sin δ,
so if t ∈ J, |1 − rei(t−θ) | = |eiθ − reit | ≥ sin δ. Therefore,

1 − r2
Z Z
1 1
P (r, t − θ) |F (t) − F (θ)|dt = |F (t) − F (θ)|dt
2π J 2π J |1 − rei(t−θ) |2
1 − r2
Z
≤ |F (t) − F (θ)|dt
2π sin2 δ J
D(1 − r2 )
≤ .
2π sin2 δ
R 2π
where D = 0 |F (t)|dt. This upper bound converges to 0 as r → 1. Combin-
ing this with (7.5), we have shown that the upper bound in (7.4) converges
to 0 as r → 1 and subsequently δ → 0. In other words, u(r, θ) → F (θ) as
r → 1. Hence, u extends continuously to the boundary condition F .

Remark. By rescaling and translation, it is straightforward that the unique-

ness and existence of solution of the Dirichlet problem for any arbitrary disk
D(z0 , r0 ) also hold. In general, the theorem also holds for arbitrary bounded
simply connected domain U if its boundary ∂U is a smooth curve. Prov-
ing such generalisation requires more involved argument and we shall not
attempt to do it here.
78 CHAPTER 7. HARMONIC FUNCTIONS

Example 59. For any real constant c ∈ R, every harmonic function u : D →

R that is identically c along the boundary ∂D must be constant, i.e. u ≡ c
on D.

Example 60. Let F (θ) = 1 when 0 ≤ θ < π and F (θ) = 0 when π ≤ θ < 2π.
The unique harmonic function on the unit disk with boundary value F (θ) is
Z π  π
1 1 −1 1+r t−θ
u(r, θ) = P (r, t − θ)dt = tan tan
2π 0 π 1−r 2 0
   
1 −1 1 + r π − θ 1 −1 1 + r θ
= tan tan + tan tan
π 1−r 2 π 1−r 2
 2 
1 r −1
= tan−1 .
π 2r sin θ

One implication of the principle is the fact that continuous functions

which satisfy the mean value property are automatically harmonic. This
holds even without the assumption that the function is differentiable.

Corollary 7.8. Let u : U → R be a continuous function on a domain U .

Then, u is harmonic if and only if it satisfies the mean value property (MVP).

Proof. It is clear from Proposition 7.3 that harmonicity implies mean value
property. Assume now that u satisfies the MVP. Let’s pick a point z0 ∈ U
in complex coordinates. By openness, there is some r > 0 such that U
contains the closed disk D(z0 , r). By the Dirichlet principle, there is a unique
continuous function v on D(z0 , r) that coincides with u on the circle C(z0 , r)
and is harmonic on the open disk D(z0 , r).
The function u − v on D(z0 , r) satisfies the MVP because both u and v
satisfy the MVP. Therefore, u − v satisfies the maximum modulus principle
too. Since u − v ≡ 0 on the boundary, u − v ≡ 0 on D(z0 , r). In particular,
u is harmonic on the open disk D(z0 , r). Since z0 is an arbitrary point in U ,
u must be harmonic everywhere on U .

7.5 Applications in Fluid Dynamics

We shall discuss one immediate application of the study of harmonic and
holomorphic functions in fluid mechanics. Consider the case of a fluid flowing
on a planar domain Ω ⊂ R2 . The (instantaneous) velocity of the flow of the
fluid at each point (x, y) ∈ Ω is assumed to be determined by a smooth vector
field V (x, y) = (p(x, y), q(x, y)).
7.5. APPLICATIONS IN FLUID DYNAMICS 79

Definition 23. The flow is incompressible if and only if the vector field V
has zero divergence:

div V = ∇ · V = px + qy ≡ 0.

The flow is irrotational if and only if V has zero curl:

curl V = ∇ × V = qx − py ≡ 0.

Incompressibility essentially means that the fluid has constant uniform

density throughout. The reason behind the vanishing divergence relies on
what is known in physics as the ”continuity equation”. Irrotationality means
that the flow has zero local circulation (often known as vorticity). This can
be described more mathematically in the following way.
Let γ be a simple closed contour on Ω and suppose the region R enclosed
by γ is contained in Ω. The circulation of the fluid along γ is defined to be
the line integral of V along γ, i.e.
I
p(x, y)dx + q(x, y)dy.
γ

By Green’s theorem, this integral can be rewritten as

Z Z
(qx − py )dxdy
R

Notice that the integrand coincides with the curl of V . Irrotationality essen-
tially ensures that the circulation along γ is zero because this integral always
vanishes. (Alternatively, you may say that the vector field is conservative.)
When both assumptions are met, we say that the fluid has an ideal fluid
flow. Observe that they very much resemble the Cauchy-Riemann equations.

Proposition 7.9. The vector field V = (p, q) on the domain Ω induces an

ideal fluid flow if and only if the complex function

f (x + iy) = p(x, y) − iq(x, y)

is holomorphic on Ω ⊂ C.

For an ideal fluid flow, the function f above is often called the complex
velocity of the fluid flow.
One implication of irrotationality is that f admits a primitive F (x+iy) =
φ(x, y) + iψ(x, y) on Ω. By Cauchy-Riemann equations and equation 2.1,
80 CHAPTER 7. HARMONIC FUNCTIONS

dF
p − iq = f = = φx + iψx = φx − iφy .
dz
In particular, φ satisfies ∇φ = (φx , φy ) = (p, q) = V . Moreover, both
φ and ψ are necessarily harmonic because they are the real and imaginary
parts of the holomorphic function F .

Definition 24. A complex potential of the fluid flow is a primitive F of f

on Ω. The real part φ of F is called the velocity potential and the imaginary
part ψ of F is called the stream function.

In summary, if we complexify V and use the notation V = p + iq instead,

we have the equation V = f¯ = F 0 . By decomposing F , we obtain two
harmonic functions φ and ψ which are harmonic conjugates of each other
and they satisfy the equations φx = ψy = p and φy = −ψx = q.

Definition 25. The level sets of the velocity potential {(x, y) ∈ Ω | φ(x, y) =
c} for some real constant c are called the equipotentials of the flow. The level
sets of the stream function {(x, y) ∈ Ω | ψ(x, y) = d} for some real constant
d are called the streamlines of the flow.

Proposition 7.10. Equipotentials and streamlines are always perpendicular

to each other at their points of intersection.

Proof. The gradient ∇φ is always normal to any equipotential {φ = c} and

The gradient ∇ψ is always normal to any streamline {ψ = d}. By Cauchy-
Riemann equations,

∇φ · ∇ψ = φx ψx + φy ψy = ψy ψx + (−ψx )ψy = 0.

As the dot product always vanishes, perpendicularity is shown.

The proof above also shows that streamlines have tangent vectors ∇φ.
Given any fluid particle (x0 , y0 ) in Ω, since the vector field V coincides
with ∇φ and determines its flow trajectory, the particle always flows along
a unique streamline {(x, y) | ψ(x, y) = ψ(x0 , y0 )}.

Example 61. A constant vector field throughout the plane V = (a, b) is

trivially an ideal flow. An easy choice of complex potential would be F (z) =
(a − ib)z on C. The corresponding complex potential and stream function
are φ(x, y) = ax + by and ψ(x, y) = bx − ay. As such, all equipotentials are
straight lines of slope −a/b and all streamlines are straight lines of slope b/a.
7.5. APPLICATIONS IN FLUID DYNAMICS 81

Example 62. The vector field V (x, y) = (x, −y) induces an ideal fluid flow
2
on the plane. The complex potential can be chosen to be F (z) = z2 =
x2 −y 2
2
+ ixy. V induces hyperbolic streamlines {xy = d}. The equipotentials
of V are hyperbolas of the form {x2 − y 2 = 2c}. At the origin, the flow has
a unique fixed point, which is a saddle point of the flow. Perpendicularity
of the equipotentials and the streamlines fails here because the vector field
vanishes.

Short Quiz 7

1. Which of the following functions are harmonic on R2 ?

x3 − y 3 , (x − y)2020 , cos x cosh y, cosh x cosh y

Answers: 1. cos x cosh y.

82 CHAPTER 7. HARMONIC FUNCTIONS