Peize Liu

St. Peter’s College

University of Oxford

Notes on
Complex Analysis

January, 2020
This page is intentionally left blank.
 Notes on Complex Analysis*
Peize Liu
St Peter’s College, University of Oxford
October 4, 2020

0 Preliminaries 1
0.1 Complex Numbers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.2 Complex Differentiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
0.3 Branch Cuts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.4 Paths and Integration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.4.1 Paths. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
0.4.2 Integration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
0.A Appendix: Background in Metric and Topological Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

1 Cauchy’s Integral Theorem 18

1.1 Proof of Cauchy’s Theorem, Basic Track . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.2 Homotopy and Cauchy’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
1.3 Cauchy’s Integral Formulae . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.3.1 Cauchy’s Integral Formulae. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.3.2 Consequences of Cauchy’s Integral Formulae. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.4 Winding Numbers and Cauchy’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.1 Winding Numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.2 Dixon’s Proof of Cauchy’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
1.A Appendix: Proof of Jordan Curve Theorem* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2 Series Representation of Functions 38

2.1 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.1 Identity Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.1.2 Argument Principle & Rouché’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.1.3 Maximum Modulus Principle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.2 Laurent Series and Isolated Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.1 Laurent Series. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.2 Isolated Singularities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3 Weierstrass Factorization Theorem* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.1 Weierstrass Factorization Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.3.2 Mittag-Leffler Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.3.3 Interpolation Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

3 Calculus of Residues 53
3.1 Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 Semicircular Contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Jordan’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Keyhole Contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.5 Infinite Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.6 Some More Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4 Conformal Mappings 68
4.1 Extended Complex Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.1.1 Riemann Sphere. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.1.2 Projective Line. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.2 Conformal Equivalence and Möbius Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
* These notes mainly follow Oxford second year complex analysis. Some off-syllabus topics are starred. These notes do not aim to be self-contained. The readers
should be familiar with first-year introductory analysis and the language of metric spaces. Some linear algebra, group theory, and the basic arithmetic of complex
numbers are also assumed. In addition, we may sometimes use intuitions in geometry and topology without giving rigorous proofs.
 4.2.1 Conformal Equivalence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.2.2 Möbius Transformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.3 Examples of Conformal Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
4.3.1 Using Möbius Transformations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
4.3.2 Using other Elementary Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
4.4 Automorphism Groups* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.1 Automorphisms of the Unit Disk. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
4.4.2 Automorphisms of the Upper Half Plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.4.3 Automorphisms of the Complex Plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4.4.4 Automorphisms of the Extended Complex Plane. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.4.5 Automorphisms of an Annulus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.5 Schwarz Reflection Principle* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.6 Riemann Mapping Theorem* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.7 Boundary Correspondence* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.8 Schwarz-Christoffel Mappings* . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.9 Harmonic Functions and Dirichlet Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.9.1 Harmonic Functions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.9.2 Poisson Kernel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.9.3 Dirichlet Boundary Value Problems. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Reading List

Lecture Notes
• K. McGerty, A2: Metric Spaces and Complex Analysis (2018-2019). [McGerty]

The lecture notes for 2018-2019 Oxford second year math course A2: Metric Spaces and Complex Analysis, on which my notes
is mainly based.

Basic Textbooks
• H. A. Priestley, Introduction to Complex Analysis (2nd edition). [Priestley]

The primary textbook for Oxford second year Complex Analysis course. Very thorough overall and especially for residue
calculus. Yet it does not contain advanced topics such as Riemann Mapping Theorem.

• J. Shi, Functions of Complex Variables. [SJH]

An undergraduate complex analysis textbook adapted from Gong’s book. Written in Chinese and is already out of print.
Contains more detail and examples compared to Gong’s book. Very typical Chinese–style textbook and is one of my favorite.

Advanced Textbooks
• D. Belyaev, C4.8 Complex Analysis: Conformal Maps and Geometry (2018-2019). [Belyaev]

The lecture notes for 2018-2019 Oxford fourth year math course C4.8 Complex Analysis: Conformal Maps and Geometry. This
is the second course on undergraduate complex analysis and it focuses mainly on conformal mappings, as suggested in the
course title.

• E. M. Stein, Princeton Lectures in Analysis II: Complex Analysis. [Stein]

One of the most popular book for undergraduate complex analysis. Contains a lot of advanced materials beyond this notes.

• S. Gong, Concise Complex Analysis (Revised edition). [GS]

A relatively advanced book. Very concise and insightful. The way of introducing differential geometry into undergraduate
complex analysis is very novel.

• W. Rudin, Real and Complex Analysis (3rd edition). [Rudin]

• S. Lang, Complex Analysis (GTM103) (4th edition). [Lang]

Papers
• Scott, Anne Elizabeth. Cauchy Integral Theorem: a Historical Development of Its Proof. Diss. Oklahoma State University, 1978.

A paper that summarizes the historical development of the proofs of Cauchy’s Theorem.

i
Chapter 0

Preliminaries

0.1 Complex Numbers 1

0.2 Complex Differentiation 2
0.3 Branch Cuts 4
0.4 Paths and Integration 7
0.4.1 Paths. 7
0.4.2 Integration. 8
0.A Appendix: Background in Metric and Topological Spaces 12

0.1 Complex Numbers

Definition 0.1. Field of Complex Numbers C

p
The field of complex numbers C := R [x] /〈x 2 + 1〉 = R( −1) is a quadratic extension of the field of real numbers. For conve-
p
nience we write −1 = i. Any complex number can be represented as z = a + bi where a, b ∈ R, or as z = r eiθ where r ∈ R and
θ ∈ R/2πZ. The addition and multiplication are as follows:

(a + bi) + (c + d i) = (a + c) + (b + d )i

(a + bi)(c + d i) = (ac − bd ) + (ad + bc)i

Remark. It is natural to identify C as R2 with extra multiplication structure. We can define the modulus, distance, and all associated
concepts for metric spaces on the complex plane.

Definition 0.2. Norm

p
For z = a + bi ∈ C, we define its norm (modulus) to be |z| := a2 + b2.

Remark. It is easy to check that (C, | · |) is a normed vector space.

Definition 0.3. Complex Conjugation

For z = a + bi ∈ C, we define its complex conjugation to be z̄ = a − bi.

Remark. |z|2 = z z̄.

Remark. The complex conjugation is the only non-trivial R-module automorphism in C, as Gal(C|R) = {idC , z̄}. See Galois Theory
for detailed discussions.

I am not going to repeat the definition of exponential / trigonometric functions or the closed / compact / connected sets on the
complex plane, because they have been thoroughly studied in the preceding courses.

1
2 CHAPTER 0. PRELIMINARIES

Definition 0.4. Domain.

We call U ⊆ C a domain, if U is open and connected.

Remark. A path-connected subset U ⊆ Rn is said to be simply-connected, if its fundamental group is trivial. That is, any closed
path in U is homotopic to a constant path (a singleton) in U . See Section 1.2 for discussion of homotopy.

Theorem 0.5. C is algebraically closed.

Any non-constant polynomial has a root in C.

Remark. This is the famous Fundamental Theorem of Algebra. We can use Extreme Value Theorem (continuous mapping between
metric spaces preserves compactness) and some elementary estimation to proof this. An more elegant proof of this theorem makes
use of Liouville’s Theorem in Chapter 1. See Corollary 1.18.

Definition 0.6. Extended Complex Plane.

We introduce ∞ into the complex plane: C∞ := C ∪ {∞}. The arithmetic of ∞ is as follows:

∀z ∈ C : z +∞ = ∞+z = ∞
∀ z ∈ C\{0} : z · ∞ = ∞ · z = ∞; z/0 = ∞; z/∞ = 0.

This is an example of one-point compactification of the complex plane. See Section 4.1 for detailed discussion of the extended
plane.

0.2 Complex Differentiation

Definition 0.7. Complex Differentiability, Holomorphicity

f (z) − f (z 0 )
Suppose U ⊆ C is a domain. f : U → C is (complex) differentiable at z 0 ∈ U , if lim exists. We denote the derivative
z→z 0 z − z0
f (z) − f (z 0 )
f 0 (z 0 ) = lim .
z→z 0 z − z0
If f is differentiable at every z ∈ U , then we say that f is holomorphic in U .

Remark. We will present a necessary condition for complex differentiability, namely the Cauchy-Riemann Equations.

Theorem 0.8. Cauchy-Riemann Equations.

Suppose U ∈ C is a domain. Let f : U → C be a complex-valued function. We can treat it as a mapping from R2 to R2 , f :

(x, y) 7→ (u, v). If f is complex differentiable, then u and v are real differentiable, and the partial derivatives satisfy:

∂u ∂v ∂u ∂v
= , =−
∂x ∂y ∂y ∂x

Proof. The standard basis of C as a vector space over R is {1, i}. The Jacobian matrix of f with respect to this basis is:

∂u/∂x ∂u/∂y
µ ¶

∂v/∂x ∂v/∂y

But with respect to the standard basis, multiplying a complex number w = r + si is equivalent to left multiplying a matrix:
µ ¶
r −s
s r
0.2. COMPLEX DIFFERENTIATION 3

∂u ∂v ∂u ∂v
Hence we must have = , =− by the complex differentiability.
∂x ∂y ∂y ∂x

Corollary 0.9

∂f 1 ∂f
The Cauchy-Riemann Equations is also equivalent to = .
∂x i ∂y

Corollary 0.10

If f : U → C is complex differentiable, then the derivative is given by:

∂u ∂v ∂u ∂v
f 0 (z) = (z) + i (z) = −i (z) + (z)
∂x ∂x ∂y ∂y

Remark. We know from introductory analysis that a mapping f : U → R2 is (real) differentiable given that the partial derivatives are
continuously differentiable. Next proposition provides a sufficient condition for complex differentiability similar to this.

Proposition 0.11

Suppose U ∈ C is a domain. Let f : U → C be a complex-valued function. Write f : (x, y) 7→ (u, v). If u, v are continuously
differentiable on U and satisfy the Cauchy-Riemann Equations, then f is holomorphic in U .

Proof. Trivial.

Remark. Actually we only require the continuity of either of the partial derivatives.

Definition 0.12. Laplacian, Harmonic Functions.

∂2 ∂2
The differential operator ∇2 := + 2 acting on twice differentiable functions in R2 is called the Laplacian. f : U → R is
∂x 2 ∂y
called a harmonic function if f ∈ ker ∇2 .

Definition 0.13. Wirtinger Derivatives.

The Wirtinger derivatives are defined to be:

∂ 1 ∂ ∂ ∂ 1 ∂ ∂
µ ¶ µ ¶
:= −i , := +i
∂z 2 ∂x ∂y ∂z̄ 2 ∂x ∂y

Remark. If we factorize the Laplacian, we will obtain:

∂2 ∂2 ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂
µ ¶µ ¶
∇2 = + = − i + i =4 =4
∂x 2 ∂y 2 ∂x ∂y ∂x ∂y ∂z ∂z̄ ∂z̄ ∂z

Proposition 0.14

Suppose U ∈ C is a domain. Let f = u + vi : U → C be holomorphic on U . Then:

∂f ∂f ∂u
= 0, f 0 = =2
∂z̄ ∂z ∂z

Sketch of Proof. It directly follows from the Cauchy-Riemann Equations and the definition of Wirtinger derivatives.
4 CHAPTER 0. PRELIMINARIES

Proposition 0.15

Suppose U ∈ C is a domain. Let f = u + vi : U → C be holomorphic on U . If u, v are twice continuously differentiable on U ,

then they are harmonic on U .

Proof.
∂ ∂
µ ¶
∇2 u = ∇2 (Re f ) = Re(∇2 f ) = Re 4 f =0
∂z ∂z̄
∂ ∂
µ ¶
∇2 v = ∇2 (Im f ) = Im(∇2 f ) = Im 4 f =0
∂z ∂z̄
Hence u, v are harmonic.

Proposition 0.16

Suppose U ∈ C is a domain. Suppose that g : U → R is a twice continuously differerntiable function and ∂g /∂z is holomorphic.
Then g is harmonic on U .

∂g ∂ ∂g
Sketch of Proof. is holomorphic =⇒ = 0 =⇒ ∇2 g = 0 =⇒ g is harmonic.
∂z ∂z̄ ∂z

Definition 0.17. Harmonic Conjugate.

Suppose U ∈ C is a domain. Let u : U → R be harmonic. Then v : U → R is said to be a harmonic conjugate of u, if f = u + vi is

holomorphic on U .

0.3 Branch Cuts

Definition 0.18. Multifunctions, Branches.

A multi-valued function or a multifunction on U ⊆ C is a mapping f : U → P (C). That is, every point of U is assigned to
multiple points in C.

A branch of f on V ⊆ U is a function g : V → C such that g (z) ∈ f (z) for z ∈ V . We will be interested in the holomorphic
branches of a multifunction.

Remark. In order to distinguish multifunctions and their branches, we use [ f (x)] to emphasize that the image of x is a set instead
of a number. This notation is not generally accepted.

Remark. We will see that the complex analogues of some real functions, such as power and logarithm, are in nature multifunctions.
The many-valuedness often arises from the argument function.

Example 0.19. Argument.

The argument is a multifunction arg : C\{0} → P (R) such that for z ∈ C,

[arg(z)] := {θ ∈ R : z = |z| eiθ }

We know that [arg(z)] is in the form of {θ + 2πk : k ∈ Z}.

We can restrict the image to a 2π segment on R to eliminate the many-valuedness. For example, the choice (−π, π] gives a
single-valued function, namely the principal argument:

Arg(z) ∈ [arg(z)] ∩ (−π, π]

However, Arg : C\{0} → (−π, π] is not continuous on C\{0}. It has a jump discontinuity across the negative real axis. In fact, we
can never impose a restriction on the image which make the argument into a continuous function.
0.3. BRANCH CUTS 5

Proof. Suppose that θ : C\{0} → R is such a continuous function. Define k : R → R by:

1 ¡ it
θ(e ) + θ(e−it )
¢
k(t ) =
2π
Then k is continuous on R. Moreover, there exists m, n : R → Z such that
1
k(t ) = ((t + 2m(t )π) + (−t + 2n(t )π)) = m(t ) + n(t )
2π
We can see that k is continuous and takes integer values. Hence k is constant on R. But for t = π, θ(eiπ ) = θ(e−iπ ). Hence
2(π + 2m(π)π)
k(π) = = 2m(π) + 1
2π
is odd. On the other hand, for t = 0,
2(0 + 2m(0)π)
k(0) = = 2m(0)
2π
is even. Contradiction.

Example 0.20. Logarithm.

We know that the exponential function exp : R → (0, +∞) is differentiable and bijective. So we can define its inverse function,
namely the logarithm, on (0, +∞), which is also bijective and differentiable. However, the exponential function on the complex
plane is not injective, so its "inverse" would be a multifunction.

We define the complex logarithm for z 6= 0:

[log z] := exp−1 ({z}) = {w ∈ C : z = ew }

If we write z = |z| eiθ and w = a + bi, then

ew = z =⇒ ea eib = |z| eiθ =⇒ a = ln |z|, b ∈ θ + 2πZ

That is,
[log z] = {ln |z| + iθ : θ ∈ arg(z)} = ln |z| + i · arg(z)

The imaginary part of the complex logarithm is exactly the argument. So by Example 0.19, there is no continuous branch of
the complex logarithm.

When we pick the principal argument, the corresponding single-valued logarithm is called principal logarithm:

Log z := ln |z| + iArg(z)

The principal logarithm is holomorphic on C\R− , with (Log z)0 = 1/z. The question that on which domain does the complex
logarithm have a holomorphic branch is fully addressed in Corollary 1.39.

Example 0.21. Power Functions.

Similar to real functions, the complex power function is defined in terms of logarithm:

[z α ] := [eα log z ] for z ∈ C\{0}

Therefore power functions are also multifunctions. Especially, the rule [z α z β ] = [z α+β ] still holds, whereas [(z 1 z 2 )α ] = [z 1α z 2α ]
does not hold generally.

For fractional powers, [z 1/n ] (n ∈ Z+ ), the image generally contains n points and can be described by the n-th roots of unity.
Let ωn := e2πi/n . Then [11/n ] = {1, ωn , ω2n , ..., ωn−1
n }.

=⇒ [z 1/n ] = |z| eiArg(z)/n {1, ωn , ω2n , ..., ωn−1

p
n
n }

Remark. Given a multifunction f , suppose the many-valuedness arises because the definition of f depends explicitly or implicitly
on the argument of one or more points a. Such points are usually excluded from the domain of definition. But from the reasoning
in Example 0.19, we can never choose a continuous branch of f in the neighborhood of a. Such points are called branch points.
The formal definition is presented below.
6 CHAPTER 0. PRELIMINARIES

Definition 0.22. Branch Points, Branch Cuts.

Suppose f : U → P (C) is a multifunction defined on open set U ⊆ C. z 0 ∈ U is said to be a branch point of f , if for r > 0, there
is no continuous branch of f defined on B (z 0 , r )\{z 0 }.

It is also useful to work in the extended complex plane. If U is unbounded, we define f˜(1/z) = f (z). We say ∞ is a branch point
of f , if 0 is a branch point of f˜. Equivalently, ∞ is a branch point of f if only if for r > 0, there is no continuous branch of f
defined on U ∩ (C\B (0, r )).

A branch cut of f is a curve in C on whose complement we can choose a continuous branch of f . A branch cut must contain
all branch points.

Remark. The argument function and complex logarithm both have branch points at 0 and ∞. When we pick the principal argu-
ment, we are in fact performing a branch cut at the negative real axis R− = (−∞, 0]. Generally we may only consider cuts that are
between branch points (or infinity).

Example 0.23

Consider the multifunction [(z 2 − 1)1/2 ]. We observe that 1 and −1 are two branch points. If we shift to the polar coordinates
and write:
z = 1 + r eiθ = −1 + s eiϕ
p
Then [(z 2 − 1)1/2 ] = [ r s ei(θ+ϕ)/2 ]. If f (z) = F (r, s, θ, ϕ) is a holomorphic branch of [(z 2 − 1)1/2 ], then F (r, s, θ, ϕ) should be
uniquely determined by z. More explicitly, we require that ei(θ+ϕ)/2 stays unchanged after f (z) goes along a closed path γ in
the cut plane. The branch cut is chosen to restrict the movement on the plane such that all inadmissible closed paths are
outlawed.

The idea of winding number introduced in Section 1.4 can help determine the admissibility of closed paths. In this case,
suppose γ(t ) = 1 + r (t ) eiθ(t ) continuously parametrizes the closed path. Then the value of θ remains unchanged if 1 is in the
exterior of γ, and the value increases by an integer multiple of 2π if 1 is in the interior of γ. This is similar for −1. If we want
ei(θ+ϕ)/2 be unchanged, then the closed paths that go around either (but not both) of the branch points should be outlawed.
We conclude that a holomorphic branch of [(z 2 − 1)1/2 ] exists if we perform the branch cut at [−1, 1].

In this cut plane, there are two holomorphic branches:

f ± (z) = ±|z 2 − 1|1/2 ei(θ+ϕ)/2 , θ ∈ [arg(z − 1)] ∩ (−π, π], ϕ ∈ [arg(z + 1)] ∩ [0, 2π)

Im
z

θ
ϕ cut
−1 1 Re

Figure 1: Branch cut along [−1, 1].

Example 0.24

Consider the multifunction [log(z 2 − 1)]. The branch points are ±1 and ∞. The polar form:

[log(z 2 − 1)] = [ln(r s) + i(θ + ϕ)] for z = 1 + r eiθ = −1 + s eiϕ

In this case, any closed path that encloses either or both of ±1 is outlawed. We can achieve this by performing a branch cut at
(−∞, −1] ∪ [1, +∞). Holomorphic branches are given, for k ∈ Z, by:

f k (z) = ln(r s) + i(θ + ϕ + 2kπ), θ ∈ [arg(z − 1)] ∩ [0, 2π), ϕ ∈ [arg(z + 1)] ∩ (−π, π]
0.4. PATHS AND INTEGRATION 7

0.4 Paths and Integration

0.4.1 Paths.
Recall that a path in Rn is described by a continuous parametrization γ : [a, b] → Rn . We adopt this concept in C. We say that a path
γ : [a, b] → C is:

1. closed, if γ(a) = γ(b);

2. simple, if ∀x, y ∈ (a, b) : γ(x) 6= γ(y);

3. C1 or smooth, if γ is continuously differentiable on (a, b);

4. piecewise smooth, if there exists a partition of [a, b]: a = x 0 < x 1 < ... < x n = b such that γ¯[x ,x ] is C 1 . A piecewise-smooth
¯
i −1 i
simple closed curve is also called a contour.

Sometimes we denote the image of the path γ∗ := γ([a, b]).

Remark. A smooth path does not necessarily have a well-defined tangent at every point, especially where γ0 (t ) = 0. Some texts
insist that a smooth path must have non-vanishing derivative everywhere. We shall not adopt this convention.

Definition 0.25. Length, Rectifiability.

Suppose γ : [a, b] → C is a path. Given a partition of [a, b], P : a = t 0 < t 1 < ... < t n = b, we define:
n ¯
Λ(γ; P ) := ¯γ(t i ) − γ(t i −1 )¯
X ¯
i =1

We say that γ is rectifiable, or is of bounded variation, if supP Λ(γ; P ) exists, where the supremum is taken over all partitions
P of [a, b]. We denote L(γ) := supP Λ(γ; P ) the length of the path γ.

Proposition 0.26

Suppose γ : [a, b] → C is a smooth path, then it is rectifiable and its length is given by:
Z b
L(γ) = |γ0 |
a

Proof. Please refer to the real analysis for the proof.

Theorem 0.27. Jordan Curve Theorem.

Suppose that γ : [0, 1] → C is a simple closed path. Then C\γ∗ has two connected components, one bounded and one un-
bounded. Each of the components has γ∗ as its boundary.

A simple closed curve on the plane is also called a Jordan curve.

Proof. See Section 1.A for a complete proof.

Remark. Without Jordan Curve Theorem we can still define the interior and exterior of a piecewise-smooth closed path. But we do
not know whether the interior is non-empty, nor do we know the connectivity of it. See Section 1.4 for detail.

Definition 0.28. Oriented Curves.

Let γ1 : [a, b] → C and γ2 : [c, d ] → C be two paths. We define γ1 ∼ γ2 , if there exists a diffeomorphism (continuously differen-
tiable bijection with continuously differentiable inverse) ϕ : [a, b] → [c, d ] such that ∀ t ∈ [a, b] : ϕ0 (t ) > 0 and γ1 = γ2 ◦ ϕ.

It is easy to check that this defines an equivalence relation. We denote the equivalence class of γ to be [γ] and call it an oriented
curve.
8 CHAPTER 0. PRELIMINARIES

Remark. The condition that ϕ0 (t ) > 0 ensures that the curve is traversed in the same direction for every parametrization.

Definition 0.29. Circles.

For a circle centered at a with radius r , the parametrization γ(t ) = a + r e2πit is said to be positively oriented and is denoted
γ(a, r ). We denote by γ+ (a, r ) the upper semicircle of γ(a, r ) and γ− (a, r ) the lower semicircle of γ(a, r ). On the other hand, the
parametrization γ(t ) = a + r e−2πit is said to be negatively oriented.

0.4.2 Integration.
Next we define integration on the complex place and investigate the line integral along paths.

Definition 0.30. Integrability of Complex-Valued Functions.

Complex-valued function f = u +vi : [a, b] → C is Riemann (resp. Lebesgue) integrable if and only if u and v are Riemann (resp.
Lebesgue) integrable. Moreover we define:
Z b Z b Z b
f := u +i v
a a a

Proposition 0.31. Triangular Inequality.

¯Z b ¯ Z b
Suppose f : [a, b] → C, then ¯¯
¯ ¯ ¯ ¯
f ¯¯ É ¯ f ¯.
a a

Z b
Proof. Suppose f = z = |z| eiθ where θ = Arg(z). Decompose f with respect to the orthonormal basis {eiθ , i eiθ }: f (t ) = u(t ) eiθ +iv(t ) eiθ
a
where u, v : [a, b] → R.
Z b µZ b Z b ¶
Integrate: f = eiθ u +i v = |z| eiθ .
a a a
b b b b b bp b
Z Z ¯Z ¯ ¯Z ¯ Z Z Z
¯ ¯ ¯ ¯
Then we have v = 0, u = |z| = ¯¯ f ¯¯ =⇒ ¯
¯ f ¯¯ = uÉ u2 + v 2 = | f | as required.
a a a a a a a

Remark. We shall give a definition of line integral, which is a special case of the Riemann-Stieltjes integral.

Definition 0.32. Line Integral (Riemann-Stieltjes).

Suppose U ⊆ C. f : U → C is a complex function and γ : [a, b] → U is a path. Given a partition of [a, b], P : a = t 0 < t 1 < ... < t n =
b, we consider the following Riemann-Stieltjes sum:
n
Σ( f , γ; ξ, P ) := f ◦ γ(ξi ) γ(t i ) − γ(t i −1 )
X ¡ ¢
i =1

where ξ := (ξ1 , ..., ξn ) and ξi ∈ [t i −1 , t i ] for each i = 1, ..., n.

If for all ε > 0 there exists δ > 0 such that for all partitions P with max |t i − t i −1 | < δ and all intermediate points ξ, we have
1Éi Én
|Σ( f , γ; ξ, P ) − A| < ε, then we say that the line integral of f along γ exists. The limit A ∈ C is the value of the line integral of f
along γ. We denote: Z Z
f (z) dz = f := A = lim Σ( f , γ; ξ, P )
γ γ mesh P →0

Proposition 0.33
Z
If f : U → C is continuous and γ : [a, b] → U is rectifiable, then the line integral f exists.
γ
0.4. PATHS AND INTEGRATION 9

Proof. Please refer to the real analysis for the proof.

Remark. The definition of line integral is purely conceptual and we shall never use it in these notes. We are only interested in
smooth or piecewise smooth paths. In such case the line integral can be computed as follows:

Proposition 0.34

Suppose γ : [a, b] → C is a smooth path and f : C → C is integrable. Then we have:

Z Z b
f := ( f ◦ γ) · γ0
γ a

Remark. If γ is only piecewise smooth, we can find a partition a = x 0 < x 1 < ... < x n = b such that each γ¯(x ,x ) is smooth. Then
¯
i −1 i
the line integral is given by
Z n
X xi
Z
f := ( f ◦ γ) · γ0
γ i =1 x i −1

Proposition 0.35

Z γ : [a,
If Z b] → C and γ̃ : [c, d ] → C are two piecewise smooth equivalent paths, then for continuous function f : C → C, we have
f = f . That is, the line integral only depends on the oriented curve [γ].
γ γ̃

Proof. Suppose ϕ is a diffeomorphism such that γ = γ̃ ◦ ϕ. Then:

Z Z b
f = ( f ◦ γ) · γ0
γ a
Z b
= ( f ◦ γ̃ ◦ ϕ) · (γ̃ ◦ ϕ)0
a
Z b
= ( f ◦ γ̃ ◦ ϕ) · (γ̃0 ◦ ϕ) · ϕ0
a
Z d
= ( f ◦ γ̃) · γ̃0 (change of variable)
c
Z
= f
γ̃

Proposition 0.36. Properties of Line Integral.

Suppose f , g : C → C are continuous functions. γ, η : [a, b] → C are piecewise smooth paths. Then:
Z Z Z
(i) Linearity. ∀ α, β ∈ C : (α f + βg ) = α f + β g
γ γ γ

(ii) Opposite Path. The opposite path of γ is defined by γ− : [a, b] → C, γ− (t ) = γ(a + b − t ). Then we have:
Z Z
f =− f
γ− γ

(iii) Additivity. Suppose γ(b) = η(a). We define the concatenation of γ and η to be: γ ? η : [a, b] → C,
(
γ(2t − a), t ∈ [a, (a + b)/2]
γ ? η(t ) =
η(2t − b), t ∈ [(a + b)/2, b]

We have: Z Z Z
f = f+ f
γ?η γ η
10 CHAPTER 0. PRELIMINARIES

(iv) Estimation Lemma. ¯Z ¯

¯ ¯
¯ f ¯ É sup | f (z)| · L(γ)
γ ∗
¯ ¯
z∈γ

where L(γ) is the length of γ.

Proof. (i) follows from the linearity of Riemann integral.

Z Z b
(ii): f = f ◦ γ− (t )(γ− )0 (t ) dt
γ− a
Z b
= − f ◦ γ(a + b − t ) · γ0 (a + b − t ) dt
a
Z a Z b
= f ◦ γ(t ) · γ0 (t ) dt = − f ◦ γ(t ) · γ0 (t ) dt
b a
Z
=− f
γ

Z Z b
(iii): f = f ◦ (γ ? η)(t ) · (γ ? η)0 (t ) dt
γ?η a
a+b
Z
2
Z b
= f ◦ γ(2t − a) · 2γ0 (2t − a) dt + f ◦ η(2t − b) · 2η0 (2t − b) dt
a+b
a 2
Z b Z b
0
= f ◦ γ(t ) · γ (t ) dt + f ◦ η(t ) · η0 (t ) dt
Za Z a

= f+ f
γ η

(iv) Since [a, b] is compact and γ is continuous, γ∗ is compact. f is continuous so that | f |(γ∗ ) is also compact. Hence
sup | f (z)| exists. We have:
z∈γ∗

¯Z ¯ ¯Z b ¯ Z b
¯ f ¯ = ¯ ( f ◦ γ) · γ0 ¯ É ¯( f ◦ γ) · γ0 ¯
¯ ¯ ¯ ¯ ¯ ¯
(by Proposition 0.31)
γ
¯ ¯ ¯ ¯
a a
Z b Z b¯
¯ ¯ 0¯
¯( f ◦ γ)¯ · ¯γ ¯ É sup ¯ f ◦ γ(t )¯ · ¯γ0 ¯ = sup | f (z)| · L(γ)
¯ ¯ ¯ ¯
=
a t ∈[a,b] a z∈γ∗

Corollary 0.37

If a sequence of continuous functions { f n } converge uniformly to f on γ∗ , then we have:

Z Z
lim fn = f
n→∞ γ γ

Proof. By uniform convergence, we have sup ( f n (z) − f (z)) → 0. By the Estimation Lemma, we have:
z∈γ∗
¯Z Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
¯ f n − f ¯ = ¯ ( f n − f )¯ É sup ( f n (z) − f (z)) · L(γ) → 0
γ γ ∗ γ
¯ ¯ ¯ ¯
z∈γ

as n → ∞. Then the result follows.

Definition 0.38. Primitive.

Suppose U ⊆ C is a domain. f : U → C is a complex function. If there exists a holomorphic function F : U → C such that F 0 = f ,
then we say that F is a primitive of f .

Remark. The existence of primitive on C is analogous to the existence of scalar potential on R2 .

0.4. PATHS AND INTEGRATION 11

Theorem 0.39. Fundamental Theorem of Line Integral.

Suppose U ⊆ C is a domain. f : U → C is continuous and F : U → C is a primitive of f . γ : [a, b] → U is a piecewise smooth

path. Then we have: Z
f = F ◦ γ(b) − F ◦ γ(a)
γ

Proof. Suppose there exists a partition a = x 0 < x 1 < ... < x n = b such that each γ¯(x
¯
is smooth. Then:
i −1 ,x i )

Z n Z xi n Z xi n Z xi
( f ◦ γ) · γ0 = (F 0 ◦ γ) · γ0 = (F ◦ γ)0 by chain rule.
X X X
f =
γ i =1 x i −1 i =1 x i −1 i =1 x i −1

By the second Fundamental Theorem of Calculus, we have:

Z n ¡
F ◦ γ(x i ) − F ◦ γ(x i −1 ) = F ◦ γ(b) − F ◦ γ(a)
X ¢
f =
γ i =1

Corollary 0.40

SupposeIU ⊆ C is a domain. f : U → C is integrable and has a primitive. γ : [a, b] → U is a piecewise smooth closed path. Then
we have f =0
γ

Example 0.41

f (z) = 1/z does not have a primitive on C\{0}.

Proof. Let the unit circle be parametrized as γ : [0, 1] → C, t 7→ e2πit . Then

Z Z 1 d 2πit
Z 1
f = e−2πit (e ) dt = 2πi dt = 2πi 6= 0
γ 0 dt 0

Hence 1/z cannot have a primitive.

Corollary 0.42

Suppose U ⊆ C is path-connected. f : U → C is holomorphic and satisfies f 0 = 0 on U . Then f is constant on U .

Theorem 0.43

Suppose U ⊆ C is a domain. f : U → C is a continuous function. The following statements are equivalent:

(i) f has a primitive on U ;

I
(ii) f = 0 along any piecewise smooth closed path γ : [a, b] → C.
γ
Z
(iii) The value of f = 0 only depends on the endpoints of γ.
γ

Proof. (i)=⇒(ii): This is just Corollary 0.40.

(ii)=⇒(iii): Suppose γ, η : [a, b] → are piecewise smooth paths such that γ(a) = η(a) and γ(b) = η(b). Then γ ? η− is a closed
path. Moreover by (ii) we have: Z Z I
f− f = f =0
γ η γ?η−
R R
Hence γ f = η f . The value of the integral only depends on the endpoints of the path.

(iii)=⇒(i): Fix z 0 ∈ U . Let γ : [a, b] → C be a piecewise smooth path such that γ(a) = z 0 and γ(b) = z. Define F (z) :=
R
γ f . We
shall show that F 0 = f .
12 CHAPTER 0. PRELIMINARIES

Fix z ∈ U . ∃ ε > 0 (B (z, ε) ⊆ U ). For w ∈ B (z, ε), consider a line segment γ starting from z to w given by s : [0, 1] → U such that
s(t ) = z + t (w − z). For a path γ1 from z 0 to z, γ2 := γ1 ? s is a path from z 0 to w. We have:
Z Z Z
F (w) − F (z) = f− f = f
γ2 γ1 s
Z 1
= f (z + t (w − z)) · (w − z) dt
0

¯ ¯ ¯Z 1 ¯
¯ F (w) − F (z) ¯ ¯ ¯
=⇒ ¯
¯ − f (z)¯ = ¯
¯ ¯ f (z + t (w − z)) dt − f (z)¯¯
w −z 0
Z 1
É | f (z + t (w − z)) − f (z)| dt
0
É sup | f (z + t (w − z)) − f (z)| → 0
t ∈[0,1]

as w → z, by the continuity of f . Hence by definition F 0 = f . F is a primitive of f .

0.A Appendix: Background in Metric and Topological Spaces

Definition 0.44. Metric Spaces.

Suppose that X is a set and d : X × X → R is a map satisfying

1. Symmetry: ∀ x, y ∈ X : d (x, y) = d (y, x);

2. Positivity: ∀ x, y ∈ X : d (x, y) Ê 0; d (x, y) = 0 ⇐⇒ x = y;

3. Triangular Inequality: ∀ x, y, z ∈ X : d (x, y) É d (x, z) + d (y, z).

(X , d ) is called a metric space.

Definition 0.45. Open Balls, Closed Balls.

Suppose that (X , d ) is a metric space. We define

B (x 0 , r ) := {x ∈ X : d (x, x 0 ) < r }, B (x 0 , r ) := {x ∈ X : d (x, x 0 ) É r }

which are the open ball and the closed ball centered at x 0 with radius r , respectively.

Definition 0.46. Interior Points, Open Sets, Neighbourhoods.

Suppose that (X , d ) is a metric space and U ⊆ X . We say that x ∈ U is an interior point of U , if:

∃r > 0 : B (x, r ) ⊆ U

The set of interior points of U is denoted by Ů . U is called an open subset of X , if U = Ů .

For x ∈ X , an open set U ⊆ X such that x ∈ U is called a(n open) neighbourhood of x.

Proposition 0.47. Properties of Open Sets

Suppose that (X , d ) is a metric space.

1. ∅ and X are open sets in X ;

2. If U1 ,U2 ⊆ X are open, then so is U1 ∪U2 ;

\
3. If {Ui }i ∈I ⊆ 𝒫 (X ) is a collection of open subsets of X , then Ui is open in X .
i ∈I
0.A. APPENDIX: BACKGROUND IN METRIC AND TOPOLOGICAL SPACES 13

Definition 0.48. Limit Points, Isolated Points, Closed Sets.

Suppose that (X , d ) is a metric space and U ⊆ X . We say that x ∈ X is a limit point of U , if

∀r > 0 : B (x, r ) ∩U \{x} 6= ∅

The set of limit points of U is denoted by U 0 . U is called a closed subset of X , if U 0 ⊆ U .

If x ∈ U is not a limit point of U , then it is called an isolated point of U .

Proposition 0.49. Closed Sets are the Complement of Open Sets.

Suppose that (X , d ) is a metric space and U ⊆ X . U is open in X if and only if X \U is closed in X .

Proposition 0.50. Open/Closed Balls are Open/Closed Sets

Suppose that (X , d ) is a metric space. For x ∈ X and r > 0, B (x, r ) is open in X and B (x, r ) is closed in X .

Definition 0.51. Topology.

Suppose that X is a set and 𝒯 ⊆ 𝒫 (X ) is a collection of subsets of X satisfying

1. ∅, X ∈ 𝒯 ;

2. U1 ,U2 ∈ 𝒯 =⇒ U1 ∪U2 ∈ 𝒯 ;
\
3. {Ui }i ∈I ⊆ 𝒯 =⇒ Ui ∈ 𝒯 .
i ∈I

Then (X , 𝒯 ) is called a topological space. 𝒯 is called a topology on X . The elements of 𝒯 are called open sets in X .

Proposition 0.52. Metric Spaces induce Topology.

Suppose that (X , d ) is a metric space. Then the open sets in X form a topology on X .

Definition 0.53. Closure.

Suppose that (X , 𝒯 ) is a topological space and U ⊆ X . The closure of X is the intersection of all closed sets in X that contain
U: \
X := V
V closed,
U ⊆V ⊆X

Proposition 0.54. Closure are Closed Sets.

Suppose that (X , 𝒯 ) is a topological space and U ⊆ X . Then U is closed. In particular, U = U if and only if U is closed.

Definition 0.55. Denseness, Separability.

Suppose that (X , 𝒯 ) is a topological space and U ⊆ X . U is called a dense subset of X if U = X . X is said to be separable, if it
has a countable dense subset.

Definition 0.56. Boundary.

Suppose that (X , d ) is a metric space and U ⊆ X . The boundary of U is defined by ∂U := U \Ů .

14 CHAPTER 0. PRELIMINARIES

Definition 0.57. Subspace Topology.

Suppose that (X , 𝒯 ) is a topological space. For Y ⊆ X , we define 𝒮 := {Y ∩U : U ∈ 𝒯 }. Then 𝒮 is a topology on Y , called the
subspace topology.

Definition 0.58. Convergence in Topological Spaces

Suppose that (X , 𝒯 ) is a topological space and {a i }i ∈N is a sequence in X . We say that {a i } converges to a, if for any open
neighbourhood U of a, there exists N ∈ N such that {a i }∞ i =N
is in U .

Definition 0.59. Continuity.

Suppose that (X , d X ) and (Y , d Y ) are metric spaces. f : X → Y is a map. The following statements are equivalent:

1. ∀ x 0 ∈ X ∀ ε > 0 ∃ δ > 0 ∀ x ∈ X : d X (x, x 0 ) < δ =⇒ d Y ( f (x), f (x 0 )) < ε;

−1
2. ∀U ⊆ Y : U is open in Y =⇒ f (U ) is open in X .

Any map satisfying the properties are called continuous functions from X to Y .

Remark. If (X , 𝒯 ) and (Y , 𝒮 ) are only topological spaces, then we take the second statement as the definition of a continuous
function f : X → Y .

Definition 0.60. Uniform Continuity.

Suppose that (X , d X ) and (Y , d Y ) are metric spaces. f : X → Y is said to be uniform continuous, if

∀ ε > 0 ∃ δ > 0 ∀ x, y ∈ X : d X (x, y) < δ =⇒ d Y ( f (x), f (y)) < ε

Definition 0.61. Homeomorphisms.

Suppose that (X , 𝒯 ) and (Y , 𝒮 ) are topological spaces. f : X → Y is called a homeomorphism, if f is continuous and bijective,
with a continuous inverse f −1 : Y → X . X and Y are said to be homeomorphic if there exists a homeomorphism f : X → Y .

Theorem 0.62. Invariance of Domain.

Suppose that U ⊆ Rn is open. If f : U → Rn is continuous and injective, then f (U ) is open, and f : U → f (U ) is a homeomor-
phism.

Remark. A similar theorem holds for holomorphic functions on C. See Open Mapping Theorem 2.8 and Inverse Function Theorem
2.9.

Definition 0.63. Completeness.

Suppose that (X , d ) is a metric space. It is said to be complete if every Cauchy sequence in X converges.

Remark. C and Rn are complete metric spaces.

Theorem 0.64. Cantor’s Intersection Theorem.

\
Suppose that (X , d ) is a complete metric space. {Ui }i ∈N is a descending chain of non-empty closed sets in X . Then Ui is
i ∈N
non-empty.
0.A. APPENDIX: BACKGROUND IN METRIC AND TOPOLOGICAL SPACES 15

Theorem 0.65. Contraction Mapping Theorem.

Suppose that (X , d ) is a non-empty complete metric space. Let f : X → X be a map such that

∃ K ∈ [0, 1) ∀ x, y ∈ X : d ( f (x), f (y)) É K d (x, y)

f is called a contraction mapping. There exists a unique x 0 ∈ X such that f (x 0 ) = x 0 .

Definition 0.66. Compactness.

Suppose that (X , 𝒯 ) is a topological space. X is said to be compact if for any open cover {Ui }i ∈I (a collection of open sets that
covers X ), there exists a finite subcover {Ui 1 , ...,Ui n }.

Definition 0.67. Sequential Compactness.

Suppose that (X , 𝒯 ) is a topological space. X is said to be sequentially compact if any sequence in X has a convergent subse-
quence.

Definition 0.68. Boundedness, Total Boundedness.

Suppose that (X , d ) is a metric space. X is said to be bounded if {d (x, y) : x, y ∈ X } is bounded. X is said to be totally bounded if
n
∀ ε > 0 ∃ x 1 , ..., x n ∈ X : B (x i , ε)
[
X=
i =1

Theorem 0.69

Suppose that (X , d ) is a metric space. The following statements are equivalent:

1. X is compact;

2. X is sequentially compact;

3. X is complete and totally bounded.

Theorem 0.70. Heine-Borel Theorem.

Suppose that X ⊆ Rn is equipped with the Euclidean metric. Then X is compact if and only if X is closed and bounded.

Theorem 0.71. Continuity Functions preserve Compactness.

Suppose that (X , 𝒯 ) and (Y , 𝒮 ) are topological spaces. Let f : X → Y be a continuous function. If X is compact, then f (X ) is
compact.

Theorem 0.72. Heine-Cantor Theorem.

Suppose that (X , d X ) and (Y , d Y ) are metric spaces. Let f : X → Y be a continuous function. If X is compact, then f is
uniformly continuous.

Definition 0.73. Equicontinuity.

Suppose that (X , d ) is a metric space and F is a collection of functions from X to R. F is said to be equicontinuous on X , if

∀ ε > 0 ∃ δ > 0 ∀ f ∈ F ∀ x1 , x2 ∈ X : d (x 1 , x 2 ) < δ =⇒ | f (x 1 ) − f (x 2 )| < ε

16 CHAPTER 0. PRELIMINARIES

Definition 0.74. Uniform Boundedness.

Suppose that (X , d ) is a metric space and F is a collection of functions from X to R. F is said to be uniformly bounded on X ,
if there exists M ∈ R such that | f (x)| É M for all x ∈ X and f ∈ F .

Theorem 0.75. Arzelà-Ascoli Theorem.

Suppose that (X , d ) is a compact metric space and F is a collection of continuous functions from X to R which is equicon-
tinuous and uniformly bounded. Then any sequence { f i }i ∈N in F contains a subsequence { f i k }k∈N that converges uniformly
on X .

Remark. The theorem can be generalized to complex-valued functions without difficulty.

Definition 0.76. Connectivity.

Suppose that (X , 𝒯 ) is a topological space. X is said to be disconnected if there exists non-empty open subsets A, B ⊆ X such
that A ∪ B = X and A ∩ B = ∅.

X is said to be connected if it is not disconnected.

Proposition 0.77

Suppose that (X , 𝒯 ) is a topological space. The following statements are equivalent:

1. X is connected;

2. Any continuous function f : X → {0, 1} (with the discrete topology) is constant;

3. The only subsets that are both open and closed are ∅ and X .

Proposition 0.78. Connected Subsets of R.

I ⊆ R is connected if and only if I is an interval.

Theorem 0.79. Continuity Functions preserve Connectedness.

Suppose that (X , 𝒯 ) and (Y , 𝒮 ) are topological spaces. Let f : X → Y be a continuous function. If X is connected, then f (X ) is
connected.

Definition 0.80. Path-Connectivity.

Suppose that (X , 𝒯 ) is a topological space. X is said to be path-connected, if for any x, y ∈ X , there exists a continuous path
γ : [0, 1] → X such that γ(0) = x and γ(1) = y.

Theorem 0.81

Suppose that (X , 𝒯 ) is a topological space.

1. If X is path-connected, then it is connected;

2. If X is a normed vector space, then X is path-connected if and only if it is connected.

Theorem 0.82. Continuity Functions preserve Path-Connectedness.

Suppose that (X , 𝒯 ) and (Y , 𝒮 ) are topological spaces. Let f : X → Y be a continuous function. If X is path-connected, then
f (X ) is path-connected.
0.A. APPENDIX: BACKGROUND IN METRIC AND TOPOLOGICAL SPACES 17

Definition 0.83. Connected Components, Path Components.

Suppose that (X , 𝒯 ) is a topological space.

The equivalence relation given by

x ∼ y ⇐⇒ ∃ connected subset U ⊆ X : x, y ∈ U

partitions X . The equivalence classes are called connected components of X .

The equivalence relation given by

x ∼ y ⇐⇒ ∃ continuous path γ : [0, 1] → X γ(0) = x, γ(1) = y

partitions X . The equivalence classes are called path components of X .

Proposition 0.84

Suppose that (X , 𝒯 ) is a topological space.

1. The connected components of X are connected;

2. The path components of X are path-connected.

Definition 0.85. Local Connectivity, Local Path-Connectivity.

Suppose that (X , 𝒯 ) is a topological space. X is said to be locally connected (resp. locally path-connected), if any point x ∈ X
is contained in a connected (resp. locally path-connected) neighbourhood of x.

Proposition 0.86

Suppose that (X , 𝒯 ) is a topological space.

1. X is locally connected if and only if for any open subset U ⊆ X , the connected components of U are open in X ;

2. X is locally path-connected if and only if for any open subset U ⊆ X , the path components of U are open in X ;
Chapter 1

Cauchy’s Integral Theorem

Informally
I stated, Cauchy’s Theorem says that, if f : U → C is holomorphic on U and γ is a simple closed curve contained in U ,
then f = 0 under some conditions. We shall build up this theorem by a sequence of lemmata and propositions. There are three
γ
tracks of proving the theorem. We will develop them respectively in Section 1.1, 1.2 and 1.4. Here is a dendrogram showing the
interdependence of each form Cauchy’s Theorem.

Cauchy’s Theorem for a Triangle

(Lemma 1.2)
Fu
nd
am
fo
r L en
in ta
eI lT
nt he
eg o
ra rem
ls

Cauchy’s Theorem for a Polygon Cauchy’s Theorem for a Star Domain

(Corollary 1.3) (Lemma 1.11)
y
op
ot
om
H

Cauchy’s Integral Formula

Cauchy’s Theorem Deformation Theorem
for a Circle
(Theorem 1.5) (Theorem 1.9)
(Theorem 1.15)

Cauchy’s Theorem for

Simply-Connected Domains, Liouville’s Theorem
Homotopy Form (Theorem 1.17)
(Theorem 1.13)
Numbers
Winding

Cauchy’s Theorem &

Cauchy’s Integral Formula,
Homology Form
(Theorem 1.34 & 1.35)

1.1 Proof of Cauchy’s Theorem, Basic Track 19

1.2 Homotopy and Cauchy’s Theorem 22
1.3 Cauchy’s Integral Formulae 24
1.3.1 Cauchy’s Integral Formulae. 24
1.3.2 Consequences of Cauchy’s Integral Formulae. 28
1.4 Winding Numbers and Cauchy’s Theorem 30
1.4.1 Winding Numbers. 30
1.4.2 Dixon’s Proof of Cauchy’s Theorem. 32
1.A Appendix: Proof of Jordan Curve Theorem* 34

18
1.1. PROOF OF CAUCHY’S THEOREM, BASIC TRACK 19

1.1 Proof of Cauchy’s Theorem, Basic Track

Theorem 1.1. Cauchy’s Theorem, 1846

Suppose U ⊆ C is simply-connected. f : U → C is holomorphic on U and its derivative f 0 is continuous on U . γ is a piecewise-

smooth simple closed curve contained in U . Then we have:
I
f =0
γ

Remark. This is the original form of Cauchy’s Theorem when it was first proposed. The extra condition that f 0 is continuous on U
make it a direct corollary of Green’s Theorem for the plane.

Proof. Consider f as a mapping in R2 . f : (x, y) 7→ (u, v). The line integral on the complex plane has a corresponding form:
I I I I
f (z) dz = (u + iv)(dx + idy) = (udx − vdy) + i (vdx + udy)
γ γ γ γ

Apply Green’s Theorem to f along γ:

∂u ∂v
Ï µ ¶ I
− dxdy = (vdx + udy)
S ∂x ∂y γ

∂v ∂u
Ï µ ¶ I
− − dxdy = (udx − vdy)
S ∂x ∂y γ

where S is the region enclosed by γ (this is well-defined by Jordan Curve Theorem).

∂u ∂v ∂v ∂u
I
But from Cauchy-Riemann Equations we know that = and = − . Hence f (z) dz = 0 as required.
∂x ∂y ∂x ∂y γ

Remark. Next we shall loosen the condition on f , of which the holomorphicity on U is sufficient. The result is given by Goursat in
1900. The following proof is adapted from Pringsheim’s work published a year after Goursat’s.

Lemma 1.2. Cauchy’s Theorem for a triangle.

Suppose U ⊆ C is a domain. f : U → C is holomorphic on U . Let T be a triangular path whose interior is contained in U . Then
we have: I
f =0
T

Figure 1.1: Bisecting a triangle.

¯I ¯
¯ ¯
Proof. "Divide and Conquer". Suppose that ¯¯ f ¯¯ = M . We are going to prove that M = 0.
T

As shown in Figure 1.1, the image of T is a triangle, which can be divided into four congruent sub-triangles, denoted by
T1 , T2 , T3 , T4 . Notice that the integral along the boundary of the inner sub-triangle is cancelled. Hence we have:
I I I I I
f = f+ f+ f+ f
T T1 T2 T3 T4

or: ¯I ¯
¯ ¯ X4 ¯¯I ¯
¯
¯ f ¯É ¯ f ¯¯
¯ ¯ ¯
T i =1 Ti
20 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM
¯I ¯
¯ M
f ¯¯ Ê . We denote this sub-triangle T (1) and the original triangle T (0) . Start from T (1) and repeat
¯
Then ∃ i ∈ {1, 2, 3, 4} : ¯¯
Ti 4
the process, we will inductively obtain a sequence of triangles T (n) such that:

1. ∆(0) ⊆ ∆(1) ⊆ ... ⊆ ∆(n) where ∆(n) denotes the region enclosed by T (n) ;

2. diam(∆(n) ) = 2−n diam(∆(0) );

3. L(T (n) ) = 2−n L(T (0) );

¯I ¯
f ¯¯ Ê 4−n M .
¯ ¯
4. ¯¯
T (n)
∞
∆(n) . Since
\
By Cantor Intersection Theorem, the first and the second properties imply that there exists a unique point z 0 ∈
n=0
U is simply-connected, z 0 ∈ U . Hence f is differentiable at z 0 . For ε > 0 there exists δ > 0 such that for z ∈ B (z 0 , δ) ∩U :
¯ ¯
¯ f (z) − f (z 0 ) 0
0 ¯<ε
¯
¯ − f (z ) ¯
¯ z − z0

Choose n ∈ N such that ∆(n) ∈ B (z 0 , δ) ∩U . Then:

| f (z) − f (z 0 ) − f 0 (z 0 )(z − z 0 )| < ε(z − z 0 ) < ε · diam(∆(n) )

Now perform the line integral along T (n) :

I
( f (z) − f (z 0 ) − f 0 (z 0 )(z − z 0 )) dz É ε · diam(∆(n) ) · L(T (n) )
T (n)
= 4−n ε · diam(∆(0) ) · L(T (0) )

But notice that: I I

( f (z) − f (z 0 ) − f 0 (z 0 )(z − z 0 )) dz = f (z) dz Ê 4−n M
T (n) T (n)
Hence I
n
M É4 f É ε · diam(∆(0) ) · L(T (0) )
T (n)
Since ε is arbitrary, we conclude that M = 0 as required.

Corollary 1.3. Cauchy’s Theorem for a polygon.

Suppose U ⊆ C is a domain. f : U → C is holomorphic on U . Let P be a piecewise linear closed path whose interior is contained
in U . Then we have: I
f =0
P

Proof. First we can prove that every simple polygon admits a triangulation by induction on the number of vertices, as shown in
Figure 1.2. If P ∗ is not simple (i.e. it intersects itself ), it can be expressed as a finite union of simple polygons, the union of
interior of which is the interior of P ∗ .

After triangulation,
I the integral of f along the paths in the interior of P ∗ will be cancelled. What is left is the integral along P .
XI
Hence f = f = 0.
P T

Remark. We use the interior of a (not necessarily simple) polygon without defining this concept properly. One way is to invoke
the Jordan’s Curve Theorem for polygons, which is an elementary fact. The other way is using the winding number introduced in
Section 1.4.

Lemma 1.4

Suppose U ⊆ C is a domain. f : U → C is continuous on U and γ is a piecewise smooth path in U . Then for ε > 0 there exists a
polygonal (piecewise linear) path P in U such that:

(i) The vertices of P ∗ are on γ∗ ;

¯Z Z ¯
(ii) ¯ f − f ¯¯ < ε
¯ ¯
¯
γ P
1.2. HOMOTOPY AND CAUCHY’S THEOREM 21

Figure 1.2: Triangulation of a polygon.

Proof. We shall make use of the uniform continuity of f . Since U is open and γ∗ is compact, we can find a compact set G such that
γ∗ ⊆ G ⊆ U . Then f is uniformly continuous on G by the Heine-Cantor Theorem. Hence:
ε
∀ ε > 0 ∃ η > 0 ∀ z, w ∈ G : |z − w| < η =⇒ | f (z) − f (w)| <
2L(γ)
n
Let ρ := dist(γ∗ , ∂U ) > 0 and δ := min{η, ρ}. Since γ∗ is compact, there is a finite open covering: γ∗ ⊆ B (z i , δ) where
[
i =1
z 1 , ..., z n ∈ γ∗ . Let the endpoints of the path be z 0 and z n+1 . Now {z 0 , z 1 , ..., z n+1 } partitions the curve γ into n + 1 parts. We
connect these points by line segments and denote the corresponding path by P . Since |z i −1 − z i | < δ < ρ, the line segment P i
between z i −1 and z i is contained in U . For the curve γi and line segment P i , we estimate the difference of the integral:
¯Z Z ¯ ¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯ ¯ ¯
¯ f − f ¯É¯ f (z) dz − f (z i )(z i − z i −1 ) ¯+¯ f (z) dz − f (z i )(z i − z i −1 ) ¯
γi γi
¯ ¯ ¯ ¯ ¯ ¯
Pi Pi
¯Z ¯ ¯Z ¯
¯ ¯ ¯ ¯
= ¯¯ ( f (z) − f (z i )) dz ¯¯ + ¯¯ ( f (z) − f (z i )) dz ¯¯
γi Pi
ε ε
< · L(γi ) + · |z i − z i −1 |
2L(γ) 2L(γ)
ε
É · L(γi )
L(γ)

Hence we have:
¯ X ε
¯Z Z ¯ n+1 ¯Z Z ¯ n+1
· L(γi ) = ε
¯ ¯ X¯
¯ f − f ¯É ¯ f− f ¯¯ <
γ γi i =1 L(γ)
¯ ¯ ¯
P i =1 Pi

Remark. Combining Corollary 1.3 and Lemma 1.4 we finally reach the landmark theorem:

Theorem 1.5. Cauchy-Goursat Theorem.

Suppose U ⊆ C is a domain. f : U → C is holomorphic on U . γ is a piecewise smooth simple closed curve whose interior is
contained in U . Then we have: I
f =0
γ

¯Z Z ¯
Proof. Fix ε > 0. By Lemma 1.4 we can find a closed piecewise linear path P such that ¯¯ f − f ¯¯ < ε. By Corollary 1.3 we have
¯ ¯

¯I ¯ γ P
I I
f = 0, Hence ¯¯ f ¯¯ < ε. But ε is arbitrary, we can conclude that f = 0 as required.
¯ ¯
P γ γ

Remark. The Cauchy-Goursat Theorem can be also stated using the concept of primitives: a holomorphic function on a simply-
connected domain has a primitive on the domain.

Remark. Again we use the concept of the interior of a simple closed path. In general, Jordan Curve Theorem addresses this problem
(see Section 1.A for a complete proof). However, as we are only interested in piecewise smooth paths, we may use winding numbers
to define the interior of such closed paths. The discussions are in Section 1.4, where another proof of Cauchy’s Theorem is presented.
22 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

1.2 Homotopy and Cauchy’s Theorem

In this section, we shall develop an alternative way of formulating Cauchy’s Theorem. First we introduce the homotopy of curves.
Informally, we say that two paths are homotopic in a domain, if one can continuously deform to another.

Definition 1.6. Homotopy.

Suppose U ⊆ C is a domain and γ, η : [0, 1] → U are two paths in U with the same endpoints, i.e. γ(0) = η(0) = z 0 , γ(1) = η(1) = z 1 .
We say that γ and η are homotopic, if there exists a continuous function h : [0, 1]2 → U , (t , s) 7→ z, such that:

∀t ∈ [0, 1] : h(0, s) = z 0 , h(1, s) = z 1

∀s ∈ [0, 1] : h(t , 0) = γ(t ), h(t , 1) = η(t )

One should think of h as a family of paths in U indexed by the second variable s which continuously deform γ into η.

Remark. It follows immediately from the definition that homotopy is an equivalence relation. We call the equivalence classes the
homotopic classes.

Definition 1.7. Constant Path, Null Homotopy.

If a path γ : [0, 1] → U is a constant function, then the image γ∗ is just a point and we call this a constant path, denoted c a as its
image is a ∈ U .

A closed path γ starting and ending at a ∈ U is said to be null homotopic, if it is homotopic to the constant path c a .

Definition 1.8. Simple-Connectivity

A domain U ⊆ C is simply-connected, if ∀z, w ∈ U , any two paths starting at z and ending at w are homotopic.

Remark. Equivalently, a domain U is simply-connected if all closed paths starting and ending at a given point z 0 ∈ U are null-
homotopic.

Remark. In the next theorem we shall prove that the line integral only depends on the homotopic class given that the function is
holomorphic.

Theorem 1.9. Deformation Theorem.

Suppose U ⊆ C is a domain. γ, η : [0, 1] → U are piecewise-smooth paths in U which are homotopic. f : U → C is holomorphic
on U . Then we have: Z Z
f = f
γ η

Remark. We need some form of Cauchy’s Theorem before we prove Theorem 1.9. For now we forget Corollary 1.3 and Theorem
1.5, which depend on some forms of Jordan Curve Theorem. We shall begin with Lemma 1.2 and generalize it to a broader class of
domains.

Definition 1.10. Star Domain.

A domain U ⊆ C is a star domain, if there exists z ∈ U such that for all w ∈ U , the line segment [z, w] is entirely contained in U .

Remark. Notice the following inclusion relation for concepts:

Convex Domain ⊆ Star Domain ⊆ Simply-Connected Domain ⊆ Domain.

1.2. HOMOTOPY AND CAUCHY’S THEOREM 23

Figure 1.3: Dissecting the homotopy.

Lemma 1.11. Cauchy’s Theorem for a star domain.

Suppose U ⊆ C is a star domain. f : U → C is holomorphic on U . γ : [a, b] → U is a piecewise-smooth closed path. Then we

have: I
f =0
γ

Proof. It suffices to prove the existence of the primitive of f on U . Suppose z 0 ∈ U is th point that satisfies the definition
Z of star
domain. For every z ∈ U , consider the line segment parametrized by γz (t ) = z 0 + t (z − z 0 ). We claim that F (z) = f is a
γz
primitive of f (z). To show this, we fix z ∈ U . ∃ ε > 0 (B (z, ε) ⊆ U ). For w ∈ B (z, ε), consider the line segment parametrized by
η(t ) =Iz + t (w − z). Notice that the interior of the triangle T with vertices z 0 , z, w is entirely contained in U . By Lemma 1.2 we
have f = 0. But T is the concatenation γz ? η ? γ−
w . Hence:
T
Z Z Z
f = f− f = F (w) − F (z)
η γw γz

And we have:
¯ ¯ ¯ ¯
¯ F (w) − F (z) ¯ ¯ 1
Z
¯
¯ − f (z)¯ = ¯
¯ ¯ f (ζ) dζ − f (z)¯¯
¯ w −z w −z η
¯ Z 1 ¯
¯ 1 ¯
=¯¯ f (z + t (w − z)) · (w − z) dt − f (z)¯¯
w −z 0
¯ ¯
É sup ¯( f (z + t (w − z)) − f (z))¯ → 0
t ∈[0,1]
I
0
as w → z by the continuity of f . Hence F = f on U . By Corollary 0.40, we have f = 0 for any piecewise-smooth closed
γ
path γ in U .

Proof of Theorem 1.9. Let h : [0, 1] × [0, 1] → U be a homotopy of γ and η. Since [0, 1]2 is compact and h is continuous, h is also
uniformly continuous so that we have cover the image h([0, 1]2 ) with finitely many disks.

∀ ε > 0 ∃ δ > 0 ∀(t 1 , s 1 ), (t 2 , s 2 ) ∈ [0, 1]2 : k(t 1 , s 1 ) − (t 2 , s 2 )k < δ =⇒ |h(t 1 , s 1 ) − h(t 2 , s 2 )| < ε

Choose N ∈ N such that N > 1/δ. Let us dissect [0, 1]2 into N 2 squares. In order to deform γ(t ) = h(t , 0) to η(t ) = h(t , 1), we
connect the image of these vertices byµ piecewise
¶ µ linear ¶ paths. µ More ¶ specifically, for k ∈ {0, 1, ..., N }, let µk be the piecewise
k 1 k k
linear path that connects the points h 0, ,h , , ..., h 1, . We claim that:
N N N N
Z Z Z Z Z
f = f = f = ... = f = f
γ µ0 µ1 µN η

It suffices to prove the second equality and the proof for others are similar. Consider two adjacent squares whose vertices are

j −1 j j +1 j −1 1 j 1 j +1 1
µ ¶ µ ¶ µ ¶ µ ¶ µ ¶ µ ¶
,0 , ,0 , ,0 , , , , , ,
N N N N N N N N N

For the left square, by the compactness condition, its image can be covered by a disc B (h(p j ), ε) where p j is the center of the
square. Therefore by Lemma 1.11, f has a primitive F j on the disc (disks are star domains). Similarly we cover the image of
24 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

the right square with another disc B (h(p j +1 ), ε) and find a primitive F j +1 of f . Since the two disks intersect, F j and F j +1 only
differ by a constant. In particular, since µ0 ( j /N ), µ1 ( j /N ) ∈ B (h(p j ), ε) ∩ B (h(p j +1 ), ε), we have:

F j ◦ µ0 ( j /N ) − F j ◦ µ1 ( j /N ) = F j +1 ◦ µ0 ( j /N ) − F j +1 ◦ µ1 ( j /N ) (1.1)

By the Fundamental Theorem for Line Integral, we have:

j j −1
Z µ¶ µ ¶
f = F j ◦ µ0 − F j ◦ µ0
µ0 |[ j −1, j ] N N
j j −1
Z µ ¶ µ ¶
f = F j ◦ µ1 − F j ◦ µ1
µ1 |[ j −1, j ] N N

After we have covered two paths with N disks, we have:

Z N Z
X
f = f
µ0 j =1 µ0 |[ j −1, j ]
N µ
j
µ ¶ µ
j −1
¶¶
F j ◦ µ0 − F j ◦ µ0
X
=
j =1 N N
NX −1 µ j
µ ¶ µ
j +1
¶¶
= F N ◦ µ0 (1) − F 1 ◦ µ0 (0) + F j ◦ µ0 − F j +1 ◦ µ0
j =1 N N
NX −1 µ µ
j
¶ µ
j +1
¶¶
= F N ◦ µ1 (1) − F 1 ◦ µ1 (0) + F j ◦ µ1 − F j +1 ◦ µ1
j =1 N N
Z
= f
µ1

where the fourth equality follows from Equation 1.1 and that µ0 and µ1 have the same endpoints.

Remark. We use this cumbersome piecewise linear approximation because we only know the continuity of h, and the integrability
of f along γk (t ) = h(t , k/N ) is not assumed. 1

Corollary 1.12

Suppose U ⊆ C is a domain. γ is a piecewise-smooth closed path which is null homotopic. f : U → C is holomorphic on U .

Then we have: I
f =0
γ

Theorem 1.13. Cauchy-Goursat Theorem.

Suppose U ⊆ C is simply-connected. f : U → C is holomorphic on U . γ is a piecewise-smooth closed curve contained in U.

Then we have: I
f =0
γ

Proof. It immediately follows from the previous corollary and the definition of a simply-connected domain.

1.3 Cauchy’s Integral Formulae

1.3.1 Cauchy’s Integral Formulae.

We are now ready to present some important consequences of Cauchy’s Theorem. All results in this section are based on Cauchy’s
Theorem for star domain (Lemma 1.11). The most general form of the theorem via the winding number will be postponed after we
present Liouville’s theorem and Riemann’s Removable Singularity Theorem.
1 See https://math.stackexchange.com/questions/44306 for detail.
1.3. CAUCHY’S INTEGRAL FORMULAE 25

Lemma 1.14

Let γ(a, r ) be a positively oriented circle. Then for w ∈ B (a, r ) we have:

1
I
dz = 2πi
γ(a,r ) z − w

Proof.
1 1 1 ³ w − a ´−1 1 X ∞ ³ w − a ´n
= = 1− =
z − w (z − a) − (w − a) z − a z −a z − a n=0 z − a
Since (w − a) < (z − a) for z ∈ γ∗ , the series converges uniformly on the image of the circle. Hence by Corollary 0.37 we can
integrate term by term:
I
1 X∞ I 1 ³ w − a ´n
dz = dz
γ(a,r ) z −w n=0 γ(a,r ) z − a z − a
∞ I
1
(w − a)n
X
= n+1
dz
n=0 γ(a,r ) (z − a)
∞ Z 1
(w − a)n (r e2πit )−(n+1) · 2πi · r e2πit dt
X
=
n=0 0
∞ Z 1
(w − a)n r −n e−2nπit dt
X
= 2πi
n=0 0
∞ µZ 1 Z 1 ¶
(w − a)n r −n
X
= 2πi + 2πi cos(−2nπt ) dt + i sin(−2nπt ) dt
n=1 0 0

= 2πi
R1 R1
The last equality follows from that the integrals 0 cos(−2nπt )dt and 0 sin(−2nπt )dt are obviously 0 for n 6= 0.

Theorem 1.15. Cauchy’s Integral Formula for a Circle.

Suppose f : U → C is holomorphic in a domain U that contains B (a, r ). Then for any w ∈ B (a, r ), we have:

1 f (z)
I
f (w) = dz
2πi γ(a,r ) z −w

Proof. Fix w ∈ B (a, r ). Define g : U → C by

 f (z) − f (w)

z ∈ U \{w}
g (z) := z −w
 0
f (w) z=w
Then g is continuous on U and holomorphic on U \{w}. U \{w} is not a star domain. Alternatively, we consider the closed
paths
I Γ1 and Γ2 as shown
I in Figure. Note that Γ1 is in a star domain which is contained in U \{w}. By Lemma 1.11 we have
g = 0. Similarly, g =0
Γ1 Γ2

The integrals over the linear segments cancel. We have:

I I I I
0= g+ g= g+ g
Γ1 Γ2 γ(a,r ) γ(w,ε)−
I I
. But g → 0 as ε → 0, by the continuity of g at w. Hence g = 0.
γ(w,ε)− γ(a,r )

f (z) − f (w)
I
=⇒ dz = 0
γ(a,r ) z −w
f (z) dz
I I
=⇒ dz = f (w)
γ(a,r ) z − w γ(a,r ) z − w
f (z)
I
=⇒ dz = 2πi · f (w) by Lemma 1.14.
γ(a,r ) z − w

And the result follows.

26 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

Γ1

w
w
γ(w, ε)−
a
γ(a, r )
a

Γ2

Figure 1.4: Cauchy’s Integral Formula.

Remark. In fact this result holds for any simple closed positively path. The generalization of Cauchy’s Integral Formula will be
presented after we introduce winding number and make sense about orientation and interior in the next section.

Definition 1.16. Entire Functions.

If f : C → C is holomorphic on the whole C, then we call it an entire function.

Theorem 1.17. Liouville’s Theorem.

A bounded entire function is constant.

Proof. Suppose f : C → C is entire and bounded by M . For w ∈ C, choose r > |w| and consider the integral:

1 f (z) 1 f (z)
I I
f (w) = dz, f (0) = dz
2πi γ(0,r ) z −w 2πi γ(0,r ) z

Hence
¯ µ ¶ ¯
¯ 1 1 1
I
¯
| f (w) − f (0)| = ¯¯ f (z) − dz ¯¯
2πi γ(a,r ) z −w z
¯I ¯
1 ¯¯ w f (z) ¯
= dz ¯
2π γ(0,r ) z(z − w) ¯
¯
¯ ¯
¯ w f (z) ¯
É r · sup ¯¯ ¯
z(z − w) ¯
z∈∂B (0,r )
|w|M 1
Ér · =M· →0
r (r − |w|) r /|w| − 1

as r → ∞. Hence f (w) = f (0) for all w ∈ C. It follows that f is constant.

Remark. There is a result much stronger than Liouville’s Theorem, namely Picard’s Little Theorem, which states that the image of
a non-constant entire function is either C or C\{z 0 } for some point z 0 ∈ C.

Corollary 1.18. Fundamental Theorem of Algebra.

Any non-constant polynomial has a root in C.

1.3. CAUCHY’S INTEGRAL FORMULAE 27

Proof. Suppose p(z) = nj=0 a j z j ∈ C[z] has no roots in C. Then f (z) = 1/p(z) is defined on the whole complex plane and is entire.
P

Without loss of generality we may assume that a n = 1. Note that

¯ ¯ ¯ ¯
¯n−1 ¯ ¯ n−1
X |a j | ¯¯
n ¯X j¯ n ¯
|p(z)| = |z | + ¯ a z ¯ Ê |z | ¯1 − ¯→∞
¯ j =0 j ¯ ¯ j =0 |z|
n− j ¯

as |z| → ∞. Hence f → 0 as |z| → ∞. Hence f is bounded. By Liouville’s Theorem, f is constant. Then p(z) is also constant.

Remark. The following theorems shows the powerful aspects of complex analysis. Any holomorphic function is in fact infinitely
differentiable and even analytic. This is much more well-behaved than real functions.

Theorem 1.19. Taylor Expansion.

∞
If f : U → C is holomorphic on domain U ⊆ C which contains B (a, r ), then the power series c n (z − a)n converges uniformly
X
n=0
to f on B (a, r ), where
f (n) (a) 1 f (z)
I
cn = = n+1
dz
n! 2πi γ(a,r ) (z − a)

Proof. We will use the same technique as in Lemma 1.14. For w ∈ B (a, r ), by Cauchy’s Integral Formula,

1
I
f (z) 1
I
f (z) X∞ ³ w − a ´n
f (w) = dz = dz
2πi γ(a,r ) z − w 2πi γ(a,r ) z − a n=0 z − a
∞ µ 1 I f (z)
¶ ∞ f (n) (a)
n
(w − a)n
X X
= n+1
dz (w − a) =
n=0 2πi γ(a,r ) (z − a) n=0 n!

Hence the Taylor series of f converges to the function on B (a, r ). The absolute and uniform convergence follows immedi-
ately.

Remark. The previous theorem demonstrates that all holomorphic functions are analytic. From now on we shall use these two
words interchangably. However, some physicists like to call complex differentiable functions "analytic" from the very beginning.
From my perspective, this is not appropriate until we prove the Taylor expansion of holomorphic functions.

Remark. We know that any power series is in fact infinitely differentiable. The Taylor expansion of a holomorphic function not only
gives a proof that it is analytic, but also gives the explicit formulae for the derivatives of the function.

Corollary 1.20. Infinite Differentiability of Holomorphic Functions.

If f : U → C is holomorphic on domain U ⊆ C, then f is infinitely differentiable on U . Moreover, if U contains B (a, r ), then for
any w ∈ B (a, r ), we have:
n! f (z)
I
f (n) (w) = dz
2πi γ(a,r ) (z − w)n+1

Remark. These integral Representations for derivatives are also called Cauchy’s Integral Formulae.

Remark. The next theorem is an immediate corollary of Corollary 1.20 and is a converse to Cauchy’s Theorem.

Corollary 1.21. Morera’s Theorem.

I
Suppose f : U → C is continuous on a domain U ⊆ C. f = 0 for any closed path in U . Then f is holomorphic on U .
γ

Proof. By Theorem 0.43, f has a primitive F on U . But by Corollary 1.20, the second derivative F 00 = f 0 exists on U . Hence f is
holomorphic on U .
I
Remark. In fact the condition for Morera’s Theorem can be weakened as follows. Instead of any closed path, f = 0 for any triangle
T
T is sufficient to deduce that f is holomorphic.
28 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

Corollary 1.22. Cauchy’s Inequalities.

If f : U → C is holomorphic on domain U ⊆ C which contains B (a, r ), then the modules of the derivatives of f at a are con-
trolled by:
¯ (n) ¯ n!
¯ f (a)¯ É sup | f (z)|
r n z∈∂B (a,r )

Proof.
¯ ¯
¯ (n) ¯ ¯ n! f (z)
I
¯
¯ f (a)¯ = ¯ dz ¯
n+1
γ(a,r ) (z − a)
¯ 2πi ¯
¯ ¯
n! ¯ f (z) ¯
É · 2πr sup ¯¯ ¯
n+1 ¯
2π z∈∂B (a,r ) (z − a)
n!
É sup | f (z)|
rn z∈∂B (a,r )

1.3.2 Consequences of Cauchy’s Integral Formulae.

Theorem 1.23. Riemann’s Removable Singularity Theorem.

Suppose U ⊆ C is open and z 0 ∈ U . f : U \{z 0 } → C is holomorphic in U \{z 0 } and is bounded near z 0 . Then f can be holomor-
phically extended on the whole U .

Remark. For this reason, z 0 is called a removable singularity of f . We will discuss the classification of isolated singularities in detail
in Section 2.2.

Proof. We define g : U → C by
(
(z − z 0 )2 f (z), z ∈ U \{z 0 }
g (z) :=
0, z = z0
Clearly g is holomorphic in U \{z 0 }. Since f is bounded near z 0 , we have

g (z) − g (z 0 )
= (z − z 0 ) f 0 (z) → 0
z − z0

as z → z 0 . It follows that g is in fact holomorphic on U . Choose r > 0 such that B (z 0 , r ) ⊆ U . By Theorem 1.19, g has the
power series expansion in B (z 0 , r ):
∞
c n (z − z 0 )n .
X
g (z) =
n=0
∞
Since g (z 0 ) = g 0 (z 0 ) = 0, we have c 0 = c 1 = 0. Therefore g (z) = (z − z 0 )2 c n−2 (z − z 0 )n . Now we define f (z 0 ) = c 2 . Then we
X
n=0
have
∞
c n−2 (z − z 0 )n .
X
f (z) =
n=0

which implies that f is holomorphic on B (z 0 , r ). We conclude that f is holomorphic on U = (U \{z 0 }) ∪ B (z 0 , r ).

Remark. The next theorem suggests that the uniform limit of holomorphic functions is holomorphic. This is a very strong result
which has no analogy in real analysis.

Theorem 1.24. Weierstrass’ Theorem.

Suppose U ⊆ C is a domain. f n : U → C is a sequence of holomorphic functions on U . If f n → f uniformly on any compact

subset of U , then f is holomorphic on U . Moreover, f n(k) → f (k) uniformly on every compact subset of U for any k ∈ N.

Proof. Fix z 0 ∈ U . It suffice to prove that f is holomorphic on a neighborhood of z 0 . Let r > 0 such that B (z 0 , r ) ⊆ U . For any
piecewise-smooth closed path γ : [a, b] → B (z 0 , r ), since f n is holomorphic in B (z 0 , r ), we have:
I
fn = 0
γ
1.3. CAUCHY’S INTEGRAL FORMULAE 29

Since γ∗ is compact, by the assumption, f n → f uniformly on γ∗ . Hence:

I I
f = lim fn = 0
γ n→∞ γ

By Morera’s Theorem, f is holomorphic in B (z 0 , r ). Hence f is holomorphic on U .

To prove the second part, note that by Corollary 1.20 f is infinitely differentiable. Fix compact subset K ⊆ U . Since U is open
[ ρ
and K is compact, we can find a open subset G such that K ⊆ G ⊆ G ⊆ U . More explicitly, we define G := B (z, ) where
z∈K 2
ρ := inf |z − w|. By Cauchy’s Inequalities, for z ∈ K we have:
z∈K ,w∈∂U

k! k!
f n(k) (z) É sup | f n (w)|, f (k) (z) É sup | f (w)|
(ρ/2)k w∈∂B (z,ρ/2) (ρ/2)k w∈∂B (z,ρ/2)

Since G is compact, f n → f uniformly on G:

∀ ε > 0 ∃ N ∈ N ∀ n > N : sup | f n (z) − f (z)| < ε

z∈G

Hence:
¯ ¯ k! k!
sup ¯ f n(k) (z) − f (k) (z)¯ É ε
¯ ¯
sup ¯ f n (z) − f (z)¯ É
¯ ¯
z∈K (ρ/2)k w∈G (ρ/2)k

Therefore we conclude that f n(k) → f (k) uniformly on every compact K ⊆ U .

Remark. The property "uniformly convergent on any compact subset" is an important property that we shall exploit in Section 4.6.
In the language of Definition 4.52, we say that { f n } is a normal family and it converges normally.

Proposition 1.25. Holomorphic Function defined in terms of Integrals.

Suppose U ⊆ C is a domain. F : U × [a, b] → C is a continuous function. Suppose z 7→ F (z, s) is holomorphic on U for every
s ∈ [a, b]. Then the function defined by:
Z b
f (z) := F (z, s) ds
a
is holomorphic on U .

Proof using Fubini’s Theorem. For any triangle T contained in U ,

I I µZ b ¶
f (z) dz = F (z, s) ds dz
T T a
Z b µI ¶
= F (z, s) dz ds (by Fubini’s Theorem)
a T
Z b
= 0 ds = 0 (since F (z, s) is holomorphic in z)
a

Then by Morera’s Theorem, f is holomorphic on U .

A more rigorous proof. We shall find a sequence of holomorphic functions f n that converges uniformly to f , so that by the pre-
vious theorem we can assert that f is holomorphic. To construct f n we use partitions of [a, b], which is analgous to the way we
approximate the Riemann integral of continuous functions.
1 Xn j
Without loss of generality we shift [a, b] to [0, 1]. Let f n (z) := F (z, ). We claim that f n (z) → f (z) uniformly on every compact
n j =1 n
subsets K ⊆ U . For compact K , note that F is continuous on compact set K × [0, 1], and hence is uniformly continuous. There-
fore:
∀ ε > 0 ∃ δ > 0 ∀ z ∈ K ∀ s, t ∈ [0, 1] : |s − t | < δ =⇒ |F (z, s) − F (z, t )| < ε

For n ∈ N such that n > 1/δ, we have:

¯ ¯
¯1 X n j
Z 1 ¯
| f n (z) − f (z)| = ¯ F (z, ) − F (z, s) ds ¯
¯ ¯
¯ n j =1 n 0 ¯
30 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM
¯ Z ¶ ¯¯
¯X n j /n µ
j
=¯ F (z, ) − F (z, s) ds ¯
¯ ¯
¯ j =1 ( j −1)/n n ¯
n j /n
¯ ¯
¯F (z, j ) − F (z, s)¯ ds
X Z
¯ ¯
É
j =1 ( j −1)/n
¯ n ¯
n
ε/n = ε
X
É
j =1

Hence f n → f on every compact subset K ⊆ U . By Weierstrass Theorem f is holomorphic on U .

1.4 Winding Numbers and Cauchy’s Theorem

In this section we wish to define the interior of a simple closed curve by introducing the winding number of a closed curve with
respect to a certain point. This will allow us to present the ultimate form of Cauchy’s theorem, as well as many consequences of it.
Informally stated, the winding number is the number of anti-clockwise rotations that a path goes around a point.

1.4.1 Winding Numbers.

Lemma 1.26. Continuous Choices of Argument.

Suppose γ : [0, 1] → C is a path and z 0 ∈ C\γ∗ . Then there exists a continuous function θ : [0, 1] → R such that:

γ(t ) = z 0 + |γ(t ) − z 0 | eiθ(t )

Moreover, if θ and ϕ are two such functions, then

∃ k ∈ Z ∀ t ∈ [0, 2π] : θ(t ) − ϕ(t ) = 2πk

Proof. Let η : [0, 1] → S 1 ⊆ C be the path such that η(t ) = (γ(t )−z 0 )/|γ(t )−z 0 |. Again we appeal to the uniform continuity of the path.
Let δ > 0 such that p
∀ s, t ∈ [0, 1] : |s − t | < δ =⇒ |η(s) − η(t )| < 3
p
Let n ∈ N such that n > 1/δ. Note that for |z| = |w| = 1 and |z − w| < 3, we must have |Arg(z) − Arg(w)| < 2π/3. Therefore we
can define a holomorphic branch of the argument function on this subinterval. That is, for 1 É j É n, there is a continuous
j −1 j
function θ j : [ n , n ] → R such that η|[ j −1 , j ] (t ) = eiθ j (t ) .
n n

iθ j ( j /n)
But for t = j /n, we must have η( j /n) = e = eiθ j +1 ( j /n) , which means that |θ j ( j /n) − θ j +1 ( j /n)| = 2πk j for some k j ∈ Z.
We may choose each k j such that θ j ( j /n) = θ j +1 ( j /n) for j ∈ {1, ..., n − 1} and obtain a continuous function θ : [0, 1] → R such
that η(t ) = eiθ(t ) as required.

Moreover, if θ and ϕ are two such functions, then we have

ei(θ(t )−ϕ(t )) = 1 =⇒ θ(t ) − ϕ(t ) ∈ 2πZ

But θ(t ) − ϕ(t ) is continuous, it must take constant value on each connected domain. [0, 1] is connected, so θ(t ) − ϕ(t ) is a
constant integer.

Remark. The Lemma shows that, if γ is a closed path, then ei(θ(1)−θ(0)) = 1. It follows that θ(1) − θ(0) ∈ 2πZ. We shall demonstrate
that this integer is an important parameter of the path, which is called the winding number.

Definition 1.27. Winding Number.

Suppose γ : [0, 1] → C is a closed path and z 0 ∈ C\γ∗ . θ : [0, 1] → R is a continuous function such that γ(t ) = z 0 + |γ(t ) − z 0 | eiθ(t ) .
θ(1) − θ(0)
Then we define I (γ, z 0 ) := to be the winding number or index of γ about z 0 .
2π
By the previous lemma, we know that I (γ, z 0 ) ∈ Z and is independent of the choice of θ(t ).

Remark. The winding number can also be interpreted by logarithm. Since log(η(t )) = ln |η(t )| + iθ(t ), we can define θ(t ) locally by
choosing holomorphic branches of [Log(η(t ))]. Hence we have the line integral form of the winding number:
1.4. WINDING NUMBERS AND CAUCHY’S THEOREM 31

Proposition 1.28

Suppose γ is a piecewise-smooth closed path and z 0 ∈ C\γ∗ . The winding number of γ about z 0 is given by:

1 dz
I
I (γ, z 0 ) :=
2πi γ z − z0

Proof. By Lemma 1.26, we may write γ(t ) = z 0 + r (t ) eiθ(t ) where r (t ) = |γ(t ) − z 0 |. Then γ0 (t ) = (r 0 (t ) + ir (t )θ 0 (t )) eiθ(t ) . Compute the
integral:

dz 1 (r 0 (t ) + ir (t )θ 0 (t )) eiθ(t )
I Z
= dt
γ z − z0 0 r (t ) eiθ(t )
Z 1 0 Z 1
r (t )
= dt + i θ 0 (t )dt
0 r (t ) 0
= (ln r (t ) + iθ(t ))10 = 0 + i(θ(1) − θ(0))
= 2πi · I (γ, z 0 ) (by Definition 1.27.)

and the result follows.

Remark. The next corollary reveals the connection between homotopy and winding number.

Corollary 1.29

If γ and η are homotopic piecewise-smooth paths via the homotopy h : [0, 1]2 → C and z 0 ∉ h([0, 1]2 ). Then I (γ, z 0 ) = I (η, z 0 ).

Proof.
1 dz 1 dz
I I
I (γ, z 0 ) = = = I (η, z 0 )
2πi γ z − z 0 2πi η z − z0

Proposition 1.30

Suppose U ⊆ C is a domain. γ : [0, 1] → U is a piecewise-smooth closed path and f is a function continuous on γ∗ . Then the
function defined by
f (z)
I
I f (γ, w) := dz
γ −w
z
is holomorphic on U .

Proof. The proof is similar to the one in Taylor Expansion. Since C\γ∗ is open and holomorphicity is a local property, it suffices to
show that I f (γ, w) is holomorphic in B (z 0 , r ) for each z 0 ∈ C\γ∗ and some r > 0.
1 ¯w −z ¯ 1
0¯
Now fix z 0 ∈ C\γ∗ . Let r = inf |γ(t ) − z 0 |. Then for w ∈ B (z 0 , r ) and z ∈ γ∗ we have ¯ ¯ < . Moreover, since γ∗ is
¯
2 t ∈[0,1] z − z0 2
compact, M = sup | f (z)| exists. Hence:
z∈γ∗
¯ n ¯¯ µ ¶n
¯ f (z) (w − z 0 ) ¯ É M 1
¯
¯ (z − z 0 )n+1 ¯ 2r 2
∞ (w − z 0 )n f (z)
on γ∗ . Hence we have:
X
By Weierstrass M-test, f (z) converges uniformly to
n=0 (z − z 0 )n+1 z −w
I
f (z) ∞ µI
f (z)
¶
(w − z 0 )n
X
dz = n+1
dz
γ z −w n=0 γ (z − z 0 )

Since I f (γ, w) is given by a power series, it is analytic and of course holomorphic.

Corollary 1.31

Fix the piecewise-smooth closed path γ. The winding number as a function z 7→ I (γ, z) is continuous on C\γ∗ . Therefore it
takes constant value on connected components of C\γ∗ .
32 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

Definition 1.32. Interior of Closed Path.

Suppose γ is a piecewise-smooth closed path. We define the interior of γ to be {z ∈ C\γ∗ : I (γ, z) 6= 0}.

By the previous lemma, it is the union of bounded connected components of C\γ∗ (if it is not empty).

We define the exterior of γ to be {z ∈ C\γ∗ : I (γ, z) = 0}. Since lim I (γ, z) = 0, the exterior is made of exactly one unbounded
z→∞
connected component of C\γ∗ .

Definition 1.33. Orientation.

A closed path γ is said to be positively oriented, if I (γ, z 0 ) > 0 for z 0 in the interior of γ; γ is said to be negatively oriented, if
I (γ, z 0 ) < 0 for z 0 in the interior of γ.

Remark. We can see that the definition of orientation above is consistent with our definition of the orientation of circles. In fact,
for a simple closed positively oriented curve, I (γ, z) = 1 for z in the interior and I (γ, z) = 0 for z in the exterior.

1.4.2 Dixon’s Proof of Cauchy’s Theorem.

Remark. We shall end this chapter with our final goal: Cauchy’s Theorem and Cauchy’s Integral Formula in the form of winding
numbers. The proof uses Liouville’s Theorem, Riemann’s Removable Singularity Theorem, and Proposition 1.25. These are con-
sequences of Cauchy’s Integral Formula for a circle, which is based on Cauchy’s Theorem for star domains. We can see that it is
independent of the homotopy form of Cauchy’s Theorem.

Theorem 1.34. Cauchy-Goursat Theorem, Homology Form.

Suppose U ⊆ C is open. f : U → C is holomorphic on U . γ : [a, b] → U is a piecewise-smooth closed path whose interior is

entirely contained in U , i.e. I (γ, z) = 0 for all z ∉ U . Then we have:
I
f =0
γ

Theorem 1.35. Cauchy’s Integral Formula, Homology Form.

Suppose U ⊆ C is open. f : U → C is holomorphic on U . γ is a piecewise-smooth closed curve whose interior is entirely

contained in U . Then for all w ∈ U \γ∗ :
1 f (z)
I
dz = I (γ, w) · f (w)
2πi γ z − w

Proof of Theorem 1.34 and 1.35. We only prove the Cauchy’s Integral Formula.
1 f (w)
I
Since I (γ, w) = dz, the formula to be proved can be written as:
2πi γ z − w

f (z) − f (w)
I
dz = 0
γ z −w

Define g : U ×U → C by:
 f (z) − f (w)

z 6= w;
g (z, w) := z −w
 0
f (z) z = w.
Then g is continuous on U × U . Fix z ∈ U , then w 7→ g (z, w) is holomorphic on U \{z}. But by continuity, w 7→ g (z, w) is
bounded near z. Hence by Riemann’s Removable Singularity Theorem, w 7→ g (z, w) is actually holomorphic on the whole U .
Then the line integral:
I Z b
F (w) := g (z, w) dz = g (γ(t ), w)γ0 (t ) dt
γ a

is holomorphic on U by Proposition 1.25.

1.4. WINDING NUMBERS AND CAUCHY’S THEOREM 33

f (z)
I
Let V denote the exterior of γ, i.e. I (γ, w) = 0 for w ∈ V . Define G : V → C by G(w) := dz. Then F (w) and G(w) agrees
γ z −w
on U ∩ V :
f (z) − f (w)
I
F (w) = dz = G(w) − f (w)I (γ, w) = G(w)
γ z −w
Hence the function H : C → C defined by:
(
F (w), w ∈U;
H (w) :=
G(w), w ∈ V.
is entire. But
¯I ¯ L(γ)sup | f (z)|
¯ f (z) ¯ z∈γ∗
|H (w)| = ¯
¯ dz ¯ É
¯ →0
γ z −w |z| − sup |γ(t )|
t ∈[a,b]

as |z| → ∞. Then by Liouville’s Theorem, H is constant on C. That is H (w) = 0 for w ∈ C. And the formula follows immediately.

Definition 1.36. Cycles.

n
Suppose γ1 , · · · , γn are closed paths in C and a 1 , · · · , a n ∈ C. We define a cycle to be the formal sum Γ := a i γi . The line
X
i =1
integral along a cycle is defined by:
I n
X I
f = ai f
Γ i =1 γi

Since winding numbers can be expressed as integrals, we can naturally define the winding number of Γ to be I (Γ, z) :=
n n
a i I (γi , z), which is well-defined for z ∉ Γ∗ := γ∗i . We define the interior of Γ to be the set of z ∈ C such that I (Γ, z) 6= 0.
X [
i =1 i =1

Corollary 1.37

Theorem 1.34 and 1.35 also holds for cycles.

Remark. Recall that we first define simple-connectivity in terms of the interior of a closed path. After introducing homotopy,
we redefine simply-connected domains to be domains such that paths with the same endpoints are homotopic. We summarize
the properties related to simply-connectivity in the next proposition. To complete the whole proof, we have to use the Riemann’s
Mapping Theorem in Chapter 3, which states that simply-connected domains are not only homeomorphic, but also conformally
equivalent to the unit disk.

Proposition 1.38. Equivalent Formulations of Simple-Connectivity.

Suppose U ⊆ C is a domain. Then the following statements are equivalent:

(i) U is homeomorphic to the unit disk D;

(ii) U is simply-connected (all paths with the same endpoints are homotopic);

(iii) Any piecewise-smooth closed path in U is null-homotopic;

(iv) The interior of any piecewise-smooth closed path is contained in U ;

I
(v) For any piecewise-smooth closed path γ in U and holomorphic function f on U , we have: f = 0;
γ

(vi) Any holomorphic function f on U has a primitive.

(vii) If f and 1/ f are both holomorphic on U , then there exists a holomorphic function g on U , such that f = exp ◦ g .

Proof. (i)=⇒(iii): By homeomorphism, there exists a continuous bijection ϕ : U → D such that ϕ−1 is also continuous. For any
piecewise-smooth closed path γ : [0, 1] → U , suppose γ(0) = γ(1) = z 0 and define the homotopy by:

H (t , s) := ϕ−1 (s · ϕ ◦ γ(t ) + (1 − s)ϕ(z 0 ))

34 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

Then H is obviously continuous. Moreover, H (t , 0) = z 0 ; H (t , 1) = γ(t ); and H (0, s) = H (1, s) = z 0 . We claim that the image
H ([0, 1]2 ) ⊆ U . To show this, it suffices to show that the line segment s · ϕ ◦ γ(t ) + (1 − s)ϕ(z 0 ), s ∈ [0, 1] is contained in D for
all t ∈ [0, 1]. This is true because ϕ(γ∗ ) ⊆ D and D is convex. Hence we conclude that H is a homotopy between γ and z 0 . γ is
null-homotopic.

(ii)⇔(iii): Trivial.

(iii)=⇒(iv): Suppose γ is a closed path in U . We have to assume that γ is piecewise-smooth.

1
For w ∈ C\U , note that is holomorphic in U . Then by Cauchy’s Theorem,
z −w
1
I
I (γ, w) = dz = 0
γ z −w

Hence the interior of γ {z ∈ C\γ∗ : I (γ, z) 6= 0} ⊆ U .

(iv)=⇒(v): This is Theorem 1.34.

(v)=⇒(vi): This is Theorem 0.43.

(vi)=⇒(vii): Since f has no roots in U , f 0 / f is holomorphic on U . By (vi), there exists a function g : U → C such that g 0 = f 0 / f .
By adding a constant to g , we can have f (z 0 ) = exp(g (z 0 )) for some z 0 ∈ U . But

d ¡ −g ¢
fe = f 0 e−g − f g 0 e−g = 0
dz

Hence f e−g = const on U . The condition that f (z 0 ) = eg (z0 ) implies that the constant is 1. Hence f = e g on U as required.

(vii)=⇒(i): This is a direct corollary of Riemann’s Mapping Theorem 4.51. See the remark after the proof of the theorem.

Corollary 1.39. Global Existence of Holomorphic Logarithm.

Suppose U ⊆ C is simply connected and 0 ∉ U , then there exists a holomorphic branch of complex logarithm on U .

Proof. Let f (z) = z in part (vii) of the previous theorem. Then there exists a holomorphic function g on U such that eg (z) = z. Hence
g is a holomorphic branch of the complex logarithm on U .

1.A Appendix: Proof of Jordan Curve Theorem*

We shall formulate a quick proof of the full Jordan Curve Theorem using the tools from fundamental groups and covering spaces in
algebraic topology.1 We shall quote a few topological theorems without proof:

Theorem 1.40. Seifert-van Kampen Theorem.

Suppose that X is a topological space. Let X = X 1 ∪ X 2 , where X 1 and X 2 are path-connected open subsets of X such that
X 1 ∩ X 2 is also path-connected. For b ∈ X 1 ∩ X 2 , the push-out of the based set
ι1 ι2
(X 1 , b) (X 1 ∩ X 2 , b) (X 2 , b)

induces the push-out of the fundamental group

ι1∗ ι2∗
π1 (X 1 , b) π1 (X 1 ∩ X 2 , b) π1 (X 2 , b)

which is isomorphic to π1 (X , b).

Theorem 1.41. Homotopy Extension Lemma.

Suppose that X is a topological space such that X × [0, 1] satisfies the T4 axiom. Let A be a closed subset of X , and Y be an
open subset of Rn . If f : A → Y is a null-homotopic continuous map, then it can be extended to a continuous map f˜ : X → Y
which is also null-homotopic.

We state the Jordan Curve Theorem for S2 :

1 The proof is adapted from Munkres’ Topology.
1.A. APPENDIX: PROOF OF JORDAN CURVE THEOREM* 35

Theorem 1.42. Jordan Curve Theorem for S2 .

Suppose that γ : [0, 1] → S2 is a simple closed path. Then S2 \γ∗ has two connected components, each of which has γ∗ as its
boundary.

We break down the proof of Jordan Curve Theorem into a sequence of lemmata.

Lemma 1.43

Suppose that γ : [0, 1] → S2 is an injective path. Then S2 \γ∗ is connected.

Proof. Suppose that a, b ∈ S2 \γ∗ . We shall show that a and b lie in the same connected component.

We first apply a stereographic projection S2 \{b} → C (see Section 4.1). Let z 0 ∈ C be the image of a, and C be the image of γ∗ .
Since C is compact, there is a unique unbounded component of C\C . Let U be the connected component of C\C containing
z 0 . It suffices to prove that U is unbounded.

Suppose for contradiction that U is bounded. Let V be the union of other connected components of C\C . Then V is un-
bounded. Consider the identity map idC : C → C . Since C is the image of an injective path, it is contractible. So idC is
null-homotopic. Since C × [0, 1] is a metric space, it satisfies the T4 axiom. C is a closed subset of U ∪ C . By Homotopy
Extension Lemma, idC : C → C extends to a continuous map α : U ∪ C → C , which is also null-homotopic. We extend α to
β : C → V ∪C by setting β|V = idV .

Since U ∪ C is bounded, there exists r > 0 such that U ∪ C ⊆ B (z 0 , r ). By restricting β on B (z 0 , r ), we obtain a continuous
map δ : B (z 0 , r ) → V ∪ C where δ|∂B (z0 ,r ) = id∂B (z0 ,r ) . Note that z 0 ∉ V ∪ C . Consider the retraction f : C\{z 0 } → ∂B (z 0 , r ),
z − z0
f (z) = z 0 + r . It is easy to see that f ◦ δ is a deformation retraction from B (z 0 , r ) to ∂B (z 0 , r ). But it is impossible, as
|z − z 0 |
π1 (B (z 0 , r )) = {e} whereas π1 (∂B (z 0 , r )) = Z.

Lemma 1.44

Suppose that γ : [0, 1] → S2 is a simple closed path. Then S2 \γ∗ has at least two connected components.

Proof. S2 \γ∗ is an open subset of a normed vector space. The path components of S2 \γ∗ are exactly the connected components.

Let x 1 and x 2 be two distinct points on γ∗ . Let C 1 and C 2 be the arcs between x 1 and x 2 . That is, γ∗ = C 1 ∪ C 2 and C 1 ∩
C 2 = {x 1 , x 2 }. Let X 1 = S2 \C 1 and X 2 = S 2 \C 2 . By the previous lemma, X 1 and X 2 are both connected open subsets of S2 .
X 1 ∪ X 2 = S2 \{x 1 , x 2 }.

Suppose for contradiction that S2 \γ∗ is connected. Then X 1 ∩ X 2 = S2 \γ∗ is path-connected. Fix x 0 ∈ S2 \γ∗ , By Seifert-van
Kampen Theorem, π1 (X ) is isomorphic to the push-out
ι1∗ ι2∗
π1 (X 1 , x 0 ) π1 (S2 \γ∗ , x 0 ) π1 (X 2 , x 0 )

We consider the group homomorphism j 1∗ : π1 (X 1 , x 0 ) → π1 (X 1 ∪ X 2 , x 0 ) induced by the inclusion map j 1 : X 1 ,→ X 1 ∪ X 2 . We

claim that it is a trivial homomorphism.

Consider a closed path η : [0, 1] → X 1 with η(0) = η(1) = x 0 . We shall show that j 1 ◦η is null-homotopic in S2 \{x 1 , x 2 }. As in the
previous lemma we consider the stereographic projection p : S2 \{x 1 } → C. Let z 0 := p(x 2 ). Then p ◦ j 1 ◦ η is a closed path in
C\{z 0 }, whose image is denoted by C . Note that x 1 and x 2 are path-connected in S2 via C 1 , away from η∗ . Since p maps x 1 to
infinity, z 0 lies in the unbounded connected component of C\C . C is bounded on C. There exists r > 0 such that C ⊆ B (z 0 , r ).
Pick z 1 ∈ C\B (z 0 , r ). Then z 0 and z 1 are path-connected in C\C . Let ξ : [0, 1] → C\C be a path connecting z 0 and z 1 . We can
define a homotopy G : [0, 1]2 → C\{z 0 }

G(t , s) = p ◦ j 1 ◦ η(t ) − ξ(s) + z 0 .

It is clear that G(t , s) 6= z 0 since p ◦ j 1∗ ◦ η and ξ are disjoint paths.

Finally we define a homotopy H : [0, 1]2 → C\{z 0 }

H (t , s) = (1 − s)(p ◦ j 1∗ ◦ η(t )) − z 1 + z 0 .

It is clear that H (t , s) 6= z 0 since |p ◦ j 1 ◦ η(t ) − z 0 | < r < |z 1 − z 0 |.

36 CHAPTER 1. CAUCHY’S INTEGRAL THEOREM

Since p ◦ j 1 ◦ η(t ) = G(t , 0), G(t , 1) = H (t , 0), and H (t , 1) = z 0 − z 1 , we deduce that p ◦ j 1∗ ◦ η is homotopic to a constant path in
C\{z 0 }. Hence j 1 ◦η is null-homotopic in S2 \{x 1 , x 2 }. It follows that j 1∗ is a trivial homomorphism. Similarly, j 2∗ : π1 (X 2 , x 0 ) →
π2 (X 1 ∪ X 2 , x 0 ) is also a trivial homomorphism.

Next we claim that π1 (X 1 ∪ X 2 ) is a trivial group. Consider the universal property of the push-out of fundamental groups:

π1 (X 1 )
ι1∗ j 1∗
θ1
∃! σ
π1 (X 1 ∩ X 2 ) π1 (X 1 ) ∗π1 (X 1 ∩X 2 ) π1 (X 2 ) π1 (X 1 ∪ X 2 )

ι2∗ θ2
j 2∗
π1 (X 2 )

By Seifert-van Kampen Theorem, π1 (X 1 ) ∗π1 (X 1 ∩X 2 ) π1 (X 2 ) ∼

= π1 (X 1 ∪ X 2 ). Since j 1∗ and j 2∗ are trivial, the maps θ1 and θ2
are also trivial. The unique induced map σ has to be both a trivial homomorphism and an isomorphism. We conclude that
π1 (X 1 ∪ X 2 ) = {e}.

But this leads to a contradiction, because π1 (X 1 ∪ X 2 ) = π1 (S2 \{x 1 , x 2 }) ∼

= Z. We conclude that S2 \γ∗ is not path-connected.

Lemma 1.45

Suppose that γ : [0, 1] → S2 is a simple closed path. Then S 2 \γ∗ has at most two connected components.

Proof. As in the previous lemma, let x 1 and x 2 be two distinct points on γ∗ . Let C 1 and C 2 be the arcs between x 1 and x 2 . Let
U = S2 \C 1 and V = S 2 \C 2 . U ∪ V = S2 \{x 1 , x 2 } and U ∩ V = S2 γ∗ . Suppose for contradiction that U ∩ V = S2 \γ∗ has more
than two connected components. Let X 1 , X 2 be two of them and W be the union of the rest.

First we shall construct a covering space for U ∪ V . Let Y := (U × 2Z) t (V × (2Z + 1)). Define an equivalence relation:

∀ n ∈ Z ∀ x ∈ X 1 ∪ X 2 : (x, 2n) ∼ (x, 2n − 1)

∀ n ∈ Z ∀ x ∈ W : (x, 2n) ∼ (x, 2n + 1)

Define the quotient space E := Y / ∼. Let π : Y  E be the quotient map. Let ρ : Y  U ∪ V , ρ(x, n) = x induces the map
p : E → U ∪ V by ρ = p ◦ π. It is not hard to verify that p is a covering map. Therefore E is a covering space of U ∪ V .

We fix a 1 ∈ X 1 , a 2 ∈ X 2 and b ∈ B . Construct the following paths:

α : [0, 1] → U , α(0) = a 1 , α(1) = b;

β : [0, 1] → V, β(0) = b, β(1) = a 1 ;
δ : [0, 1] → U , δ(0) = a 1 , δ(1) = a 2 ;
λ : [0, 1] → V, λ(0) = a 2 , λ(1) = a 1 .

Let f := α ? β and g := δ ? λ. They are loops in U ∪ V based at a 1 . Since E is connected, f and g has unique based liftings in
E , which we are going to construct.

Now we consider the lifting of f . Let α̃n (t ) = π(α(t ), 2n) and β̃n (t ) = π(β(t ), 2n + 1). They are liftings of α and β repsectively.
Let f˜n := α̃n ? β̃n . Then f˜n is a path in E such that f˜n (0) = π(a 1 , 2n) and f˜n (1) = π(a 1 , 2(n + 1)). If we fix the based point
π(a 1 , 0) ∈ E , the path f m has the unique lifting in E

f ˜m := f˜0 ? · · · ? f˜m−1

where f ˜m (0) = π(a 1 , 0) and f ˜m (1) = π(a 1 , 2m).

Now we consider the lifting of g . Let δ̃n (t ) = π(δ(t ), 2n) and λ̃n (t ) = π(λ(t ), 2n − 1). Since π(a 1 , 2n) = π(a 1 , 2n − 1) and
π(a 2 , 2n) = π(a 2 , 2n − 1), g̃ n is a loop in E such that g̃ n (0) = g̃ n (1) = (a 1 , 2n) The path g k has the unique lifting in E

g˜k := g̃ 0 ? · · · ? g̃ 0

where g˜k (0) = g˜k (1) = π(a 1 , 0).

Since g˜k (1) = f ˜m (1) if and only if m = 0, it follows that [ f ]m 6= [g ]k for all m, k ∈ Z\{0}. [ f ] ∈ π1 (U ∪V ) is non-trivial as we have
shown. But [g ] ∈ π1 (U ∪ V ) is also non-trivial because X 1 ∪ X 2 is disconnected: construct another covering space of U ∪ V ,
where the roles of X 1 ∪ X 2 and W are replaced by X 1 ∪ W and X 2 .
1.A. APPENDIX: PROOF OF JORDAN CURVE THEOREM* 37

In summary, [ f ], [g ] ∈ π1 (U ∪V ) are non-trivial elements such that [ f ]m 6= [g ]k for all m, k ∈ Z\{0}. This is impossible because
π1 (U ∪ V ) = π1 (S2 \{x 1 , x 2 }) = Z.

Lemma 1.46

Suppose that γ : [0, 1] → S2 is a simple closed path. Let X 1 and X 2 be the two connected components of S2 \γ∗ . Then ∂X 1 =
∂X 2 = γ∗ .

Proof. Since S2 is locally connected, X 1 and X 2 are open. In particular ∂X 1 ∩ X 1 = ∅ and ∂X 2 ∩ X 2 = ∅. It is clear that ∂X 1 ∩ X 2 = ∅
and ∂X 2 ∩ X 1 = ∅. Therefore ∂X 1 , ∂X 2 ⊆ γ∗ .

For the reverse inclusion, suppose that x ∈ γ∗ . We shall show that x ∈ ∂X 1 . Let U be an open neighbourhood of x. Choose
x 1 , x 2 ∈ U such that one of the arcs connecting x 1 and x 2 is entirely in U . Denote it by C 1 and denote the other arc by C 2
Choose a ∈ X 1 and b ∈ X 2 . Since S2 \C 2 is path-connected, there exists a path α : [0, 1] → S 2 \C 2 such that α(0) = a and
α(1) = b. Since X 1 and X 2 are open and disjoint, and α∗ is connected, then α∗ ∩ ∂X 1 6= ∅. But ∂X 1 ⊆ γ∗ and α∗ ∩ C 2 = ∅.
Hence α∗ ∩ C 1 6= ∅. In other words, there exists y ∈ C 1 ⊆ U such that y ∈ ∂X 1 . It follows that x ∈ pa X 1 = ∂X 1 . Similarly,
x ∈ ∂X 2 .

In conclusion, we have ∂X 1 = ∂X 2 = γ∗ .

Theorem 0.27. Jordan Curve Theorem for C.

Suppose that γ : [0, 1] → C is a simple closed path. Then C\γ∗ has two connected components, one bounded and one un-
bounded. Each of the components has γ∗ as its boundary.

Proof. Choose x 0 ∈ S2 \γ∗ and consider the stereographic projection S2 \{x 0 } → C.

Chapter 2

Series Representation of Functions

2.1 Taylor Series 38

2.1.1 Identity Theorem. 38
2.1.2 Argument Principle & Rouché’s Theorem. 39
2.1.3 Maximum Modulus Principle. 40
2.2 Laurent Series and Isolated Singularities 42
2.2.1 Laurent Series. 42
2.2.2 Isolated Singularities. 44
2.3 Weierstrass Factorization Theorem* 47
2.3.1 Weierstrass Factorization Theorem. 47
2.3.2 Mittag-Leffler Theorem. 50
2.3.3 Interpolation Theorem. 51

2.1 Taylor Series

We have shown in Section 1.3 that holomorphic functions on a disk can be expanded into Taylor Series:

Theorem 1.19. Taylor Expansion.

∞
If f : U → C is holomorphic on domain U ⊆ C which contains B (a, r ), then the power series c n (z − a)n converges to f
X
n=0
absolutely on B (a, r ) and uniformly on any compact subset of B (a, r ), where

f (n) (a) 1 f (z)

I
cn = = dz
n! 2πi γ(a,r ) (z − a)n+1

2.1.1 Identity Theorem.

Lemma 2.1

Suppose f is holomorphic in a domain U ⊆ C. Let S := f −1 ({0}) be the set of zeros of f . z 0 ∈ S is an isolated point in S, then
there exists a unique k ∈ N and a holomorphic function g : U → C such that f (z) = (z − z 0 )k g (z) for all z ∈ U and g (z 0 ) 6= 0.

n
Proof. f is analytic at z 0 . So f (z) = ∞
P
i =0 c n (z − z 0 ) for all z ∈ B (z 0 .r ) ⊆ U . Since z 0 is an isolated zero of f , not all c n are zero. Let k
be the smallest integer such that c n 6= 0. Clearly c 0 = 0 so k Ê 1.
n
Define g (z) := (z − z 0 )−k f (z) = ∞
P
i =0 c n+k (z − z 0 ) , which is holomorphic on U \{z 0 } and continuous on U . g (z 0 ) = c k 6= 0, and
by continuity there exists ε > 0 such that g (z) 6= 0 on B (z 0 , ε). Hence f (z) = (z − z 0 )k g (z) is non-zero on B (z 0 , ε)\{z 0 }.

To prove the uniqueness of k, let f (z) = (z −z 0 )k g (z) = (z −z 0 )l h(z). If k < l , then g (z) = f (z)(z −z 0 )−k = (z −z 0 )l −k h(z). Since
h(z 0 ) 6= 0, letting z → z 0 we have g (z) → 0, which contradicts that g (z 0 ) 6= 0. Similarly we cannot have k > l . Hence k = l .

38
2.1. TAYLOR SERIES 39

Definition 2.2. Multiplicity of zeros.

Suppose f is holomorphic in a domain U ⊆ C and z 0 ∈ f −1 ({z 0 }) is an isolated. Then we define the multiplicity of z 0 to be the
unique integer k in the previous lemma.

Theorem 2.3. Identity Theorem.

Suppose f is holomorphic in a domain U ⊆ C. The following statements are equivalent:

(i) f (z) = 0 for all z ∈ U ;

(ii) ∃ a ∈ U ∀ k ∈ N : f (k) (a) = 0;

(iii) The set f −1 ({0}) has a limit point in U .

Proof. (i)=⇒(ii): Trivial.

(ii)=⇒(iii): There exists r > 0 such that B (a, r ) ⊆ U . By Theorem 1.19, we have:

∞ f (n) (a)
(z − a)n = 0
X
f (z) =
n=0 n!

for all z ∈ B (a, r ). Then f −1 ({0}) contains B (a, r ) and hence has a limit point.

(iii)=⇒(i): Let S be the set of limit points of f −1 ({0}) in U . Since f is continuous on U , f −1 ({0}) is closed. Therefore we have
n
S ⊆ f −1 ({0}). Suppose z 0 is a limit point of f −1 ({0}). f is analytic at z 0 . So f (z) = ∞
P
i =0 c n (z − z 0 ) for all z ∈ B (z 0 .r ) ⊆ U .
If there exists a non-zero coefficient c n , then by the previous lemma f is non-zero on some deleted neighbourhood of z 0 ,
contradicting that z 0 is a limit point. Hence f (z) = 0 on B (a, r ). Hence z 0 is an interior point of S and S is open in U . But by
definition S is closed in U . Since U is connected and S 6= ∅, we must have S = U ⊆ f −1 ({0}). That is, f (z) = 0 for all z ∈ U .

Corollary 2.4

Suppose f and g are holomorphic in a domain U ⊆ C. The following statements are equivalent:

(i) f (z) = g (z) for all z ∈ U ;

(ii) ∃ a ∈ U ∀ k ∈ N : f (k) (a) = g (k) (a);

(iii) The set S := {z ∈ U : f (z) = g (z)} has a limit point in U .

Proof. Simply apply the Identity Theorem to f − g .

2.1.2 Argument Principle & Rouché’s Theorem.

Theorem 2.5. Argument Principle for Holomorphic Functions.

Suppose U ⊆ C is a domain and f : U → C is holomorphic. Let γ be a piecewise-smooth simple closed path in U and f is
non-zero on γ. Then we have:
I 0
1 f (z)
N= dz
2πi γ f (z)
where N is the number of zeros of f inside γ (counting multiplicity).

Proof. Suppose f has zeros a 1 , · · · , a n with multiplicity m 1 , · · · , m n in the interior of γ. Without loss of generality, suppose that γ is
positively oriented. Since a 1 , · · · , a n are isolated zeros, we can choose r 1 , · · · , r n such that B (a i , r i ) are mutually disjoint and
n n
B (a i , r i ) is in the interior of γ. Now consider the cycle Γ := γ − γ(a i , r i ), we observe that all the zeros are in the
[ X
that
i =1 i =1
exterior of Γ. By Cauchy’s Theorem:
I
f 0 (z)
I
f 0 (z) X n I f 0 (z)
0= dz = dz − dz
Γ f (z) γ f (z) i =1 γ(a i ,r i ) f (z)
40 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

For each i ∈ {1, · · · , n}, g i (z) = (z − a i )−mi f (z) is holomorphic and non-zero on U . Moreover, we have:
m −1 m 0 0
f 0 (z) m i (z − a i ) i g i (z) + (z − a i ) i g i (z) g i (z) mi
= = +
f (z) (z − a i )mi g i (z) g i (z) z − a i

But g i0 /g i is holomorphic in B (a i , r i ), we have:

g i0 (z) f 0 (z) 1
I I I
0= dz = dz − m i dz
γ(a i ,r i ) g i (z) γ(a i ,r i ) f (z) γ(a i ,r i ) z − ai
0
f (z)
I
=⇒ dz = 2πim i
γ(a i ,r i ) f (z)

Hence I
f 0 (z) X n I f 0 (z) X n
dz = dz = 2πi m i
γ f (z) i =1 γ(a i ,r i ) f (z) i =1

which completes the proof.

Remark. To see why the theorem is called "argument principle", observe that the winding number of f ◦ γ about the origin is just:

dw b f 0 ◦ γ(t ) · γ0 (t ) f 0 (z)
I Z I
I ( f ◦ γ, 0) = = dt = dz
f ◦γ w a f ◦ γ(t ) γ f (z)

So the number of zeros of f inside γ is the same as the winding number of f ◦ γ about the origin.

Remark. Theorem 2.5 can be generalized to meromorphic functions. See Theorem 2.23.

Theorem 2.6. Rouché’s Theorem.

Suppose U ⊆ C is a domain and f , g are holomorphic functions on U . Suppose γ is a piecewise-smooth closed path in U . If
| f (z)| > |g (z)| for all z ∈ γ∗ , then f and f + g have same number of zeros (counting multiplicity) in the interior of γ.

Proof. Since | f (z)| > |g (z)| for all z ∈ γ∗ , we can see that f + t g have no zeros on γ∗ for all t ∈ [0, 1]. Define:

f 0 (z) + t g 0 (z)
F (z, t ) =
f (z) + t g (z)

Then F (z, t ) is continuous on γ∗ × [0, 1] and z 7→ F (z, t ) is holomorphic for all t ∈ [0, 1]. Hence the function n(t ) defined by
the integral: I
n(t ) := F (z, t ) dz
γ

is continuous on [0, 1].

But n(t ) is the number of zeros of f + t g in the interior of γ, by argument principle, and hence is integer-valued. Since [0, 1]
is connected, n(t ) is constant. n(0) = n(1). Therefore f and f + g have the same number of zeros.

Remark. As an application of the argument principle, Rouché’s Theorem implies that a holomorphic function can be slightly per-
turbed without changing the number of its zeros.

2.1.3 Maximum Modulus Principle.

Definition 2.7. Open Mapping.

A mapping is said to be open if it maps open sets to open sets.

Theorem 2.8. Open Mapping Theorem.

A holomorphic and non-constant function on a domain U ⊆ C is an open mapping.

Proof. Suppose f : U ⊆ C is holomorphic and non-constant. To show that f (U ) is open, it suffices to show that
2.1. TAYLOR SERIES 41

∀ w 0 ∈ f (U ) ∃ ε > 0 : B (w 0 , ε) ⊆ f (U ).

Let z 0 ∈ U such that w 0 = f (z 0 ). Then the function g (z) := f (z) − w 0 has an isolated zero at z 0 . Then there exists r > 0 such
that g (z) 6= 0 on B (z 0 , r )\{z 0 }. Since ∂B (z 0 , r ) is compact, there exists ε > 0 such that |g (z)| > ε on ∂B (z 0 , r ). Then for all
w ∈ B (w 0 , ε), we have |g (z)| > ε > |w 0 − w| on ∂B (z 0 , r ). By Rouché’s Theorem, h(z) := g (z) + (w 0 − w) = f (z) − w also has a
zero in B (z 0 , r ). Hence w ∈ f (U ). Hence B (w 0 , ε) ⊆ f (U ) as required.

Theorem 2.9. Inverse Function Theorem.

Suppose f : U → C is holomorphic.

1. If f is injective, then f 0 (z) 6= 0 for all z ∈ U ;

2. If f 0 (z) 6= 0 for all z ∈ U , then f is locally injective.

For the first case, the inverse function g : f (U ) → U is holomorphic on f (U ). Moreover, we have g 0 = 1/( f 0 ◦ g ). If γ is a
piecewise smooth closed path in U , then
z f 0 (z)
I
g (w) = dz
γ f (z) − w

for w in the interior of γ.

Proof. The proof of (1) and (2) are exactly the same as in real analysis. But there are simple proofs for holomorphic functions that
make use of Rouché’s Theorem.

(1). Suppose for contradiction that there exists a ∈ U with f 0 (a) = 0. We claim that there exists R > 0 such that f (z) 6= f (a)
and f 0 (z) 6= 0 in the deleted neighborhood B (a, R)\{a}. Indeed, if the first condition does not hold, then for each sufficiently
large n ∈ N there exists z n ∈ B (a, 1/n)\{a} such that f (a) = f (z n ). Then a is a limit point in the set {z ∈ C : f (z) = f (a)}. By
identity theorem f (z) = f (a) for all z ∈ U , contradicting that f is injective. If the second condition does not hold, for similar
reason we must have f 0 (z) = 0 for all z ∈ U , again contradicting that f is injective.

Let ε = inf | f (z) − f (a)| > 0 and w ∈ B ( f (a), ε)\{ f (a)}. For z ∈ ∂B (a, R):
z∈∂B (a,R)

| f (z) − f (a)| Ê ε > |w − f (a)| = |( f (z) − w) − ( f (z) − f (a))|

By Rouché’s Theorem, f (z)− f (a) and f (z)− w have the same number of zeros in B (a, R). However, f (z)− f (a) has exactly at
least two zeros in B (a, R) because f 0 (a) = 0, whereas f (z) − w has exactly one zero because f is injective and f 0 (z) 6= 0. This
is a contradiction. Hence f 0 (a) 6= 0.

(2). For the beginning the proof is the same as the previous part. The continuity of f at a implies that there exists δ > 0 such
that f (B (a, δ)) ⊆ B ( f (a), ε), where ε is defined as above. Let r := min{R, δ}. We claim that f is injective in B (a, r ). Suppose
for contradiction that there exists z 1 , z 2 ∈ B (a, r ) such that w = f (z 1 ) = f (z 2 ). By Rouché’s Theorem, f (z) − f (a) and f (z) − w
have the same number of zeros in B (a, R). However, f (z) − f (a) has exactly one simple zero in B (a, R) because f 0 (a) 6= 0,
whereas f (z) − w has at least two zeros, contradiction. Hence f is injective in B (a, r ).

The continuity of the inverse function follows from the open mapping theorem. By open mapping theorem, for any open set
V ⊆ U , the pre-image of V under g is f (V ), which is open. Hence g is continuous.

g 0 = 1/( f 0 ◦ g ) follows directly from the definition of derivatives.

Let Γ := f ◦ γ. Γ is a closed path. By Cauchy’s Integral Formula:

g (ζ)
I
g (w) = dζ
Γ ζ−w

As ζ = f (z), dζ = f 0 (z)dz, we have:

g ◦ f (z) 0 z f 0 (z)
I
g (w) = f (z) dz = dz
γ f (z) − w f (z) − w

Remark. Sometimes we call an injective holomorphic function a biholomorphism or a univalent function. In Corollary 4.13 we
shall show that biholomorphisms are angle-preserving mappings.

Theorem 2.10. Maximum Modulus Principle.

Suppose U ⊆ C is a domain and f : U → C is holomorphic. Then | f | cannot attain a maximum value in U .

42 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

Proof. Suppose there exists z 0 ∈ U such that | f (z 0 )| attains a maximum value. But f (U ) is open by the open mapping theorem, so
f (z 0 ) is an interior point. There exists z ∈ U such that | f (z)| > | f (z 0 )|, which is a contradiction.

Corollary 2.11

Suppose U ⊆ C is a domain and U is compact. f is holomorphic on U and continuous on U . Then | f | attains a maximum
value on ∂U .

Proof. Since U is compact and | f | is continuous, f (U ) is compact. | f | attains maximum on U . But by maximum modulus principle,
| f | cannot attain maximum in U , so it attains maximum on U \U = ∂U .

2.2 Laurent Series and Isolated Singularities

2.2.1 Laurent Series.

+∞
c n (z − z 0 )n a Laurent series. The power
X
A Laurent series is a generalization of the power series. We called a series of the form
n=−∞
∞ −∞
n
c n (z − z 0 )n is called the principal part. We shall
X X
series part c n (z − z 0 ) is called the holomorphic part and the negative part
n=0 n=−1
see that many properties of the Laurent series depend on the principal part. Just as holomorphic functions can be expanded into
Taylor series in an open disk, we shall prove that they can be expanded into Laurent series in an open annulus.

Definition 2.12. Open Annulus.

A open annulus A(z 0 , r, R) is a open set on the complex plane defined by

A(z 0 , r, R) = B (z 0 , R)\B (z 0 , r ) = {z ∈ C : r < |z − z 0 | < R}

Theorem 2.13. Laurent Expansion.

+∞
If f is holomorphic on a domain U ⊆ C which contains the annulus A(z 0 , r, R), then the Laurent series c n (z − a)n con-
X
n=−∞
verges to f uniformly on any compact subset of A(z 0 , r, R), where

1 f (z)
I
cn = dz
2πi γ(z 0 ,ρ) (z − z 0 )n+1

for any ρ ∈ (r, R). Moreover, the expansion is unique.

Proof. For w ∈ A(z 0 , r, R), we apply the homology form of the Cauchy’s Integral Formula to the cycle Γ := γ(z 0 , R) − γ(z 0 , r ):

f (z) f (z)
I I
2πi · f (w) = dz − dz
γ(z 0 ,R) z −w γ(z 0 ,r ) z −w

1 1 X ∞ w − z ¶n µ
0
On the circle ∂B (z 0 , R), since |w| < |z|, the series = converges uniformly. We can integrate term by
z − w z − z 0 n=0 z − z 0
term: I
f (z) ∞ µI f (z)
¶
dz (w − z 0 )n
X
dz = n+1
γ(z 0 ,R) z − w n=0 γ(z 0 ,R) (z − z 0 )

1 1 ∞ µ z − z ¶m
0
On the circle ∂B (z 0 , r ), since |w| > |z|, the series
X
= converges uniformly. We can integrate term
w − z w − z 0 m=0 w − z 0
by term:

f (z) ∞ (z − z 0 )m
I I X
dz = − f (z) m+1
dz
γ(z 0 ,r ) z −w γ(z 0 ,r ) m=0 (w − z 0 )
I X (w − z 0 )n
−∞
=− f (z) n+1
dz (let n = −m − 1)
γ(z 0 ,r ) n=−1 (z − z 0 )
2.2. LAURENT SERIES AND ISOLATED SINGULARITIES 43

−∞ µI
f (z)
¶
(w − z 0 )n
X
= n+1
dz
n=−1 γ(z 0 ,r ) (z − z 0 )

If we define
1 f (z)
 I

 dz, n Ê 0
2πi Iγ(z0 ,R) (z − z 0 )n+1

c n := 1 f (z)
dz, n < 0


2πi γ(z0 ,r ) (z − z 0 )n+1


+∞
c n (w − z 0 )n . To see that the coefficients can be expressed by integral
X
Then the above equations implies that f (w) =
n=−∞
along γ(z 0 , ρ), notice that f (z)/(z − z 0 )n+1 is holomorphic on tha annulus A(z 0 , r, R). So when we apply Cauchy’s Theorem to
the cycle Γ := γ(z 0 , ρ) − γ(z 0 , r ), we have:

f (z) f (z)
I I
0= n+1
dz − n+1
dz
γ(z 0 ,ρ) (z − z 0 ) γ(z 0 ,r ) (z − z 0 )

1 f (z)
I
And similar for the cycle γ(z 0 , r ) − γ(z 0 , R). Hence we have c n = dz as required.
2πi γ(z 0 ,ρ) (z − z 0 )n+1
+∞
d n (z − z 0 )n on A(z 0 , r, R). Then:
X
To prove uniqueness, suppose f has another Laurent series
n=−∞

f (z)
I
2πic n = n+1
dz
γ(z 0 ,ρ) (z − z 0 )
+∞ d k (z − z 0 )k
I X
= n+1
dz
γ(z 0 ,ρ) k=−∞ (z − z 0 )
I +∞ I +∞
d k (z − z 0 )k−n−1 dz + d −k (z − z 0 )−k−n−1 dz
X X
=
γ(z 0 ,ρ) k=0 γ(z 0 ,ρ) k=1
+∞ I
(z − z 0 )k−n−1 dz
X
= dk (since both power series converges uniformly)
k=−∞ γ(z 0 ,ρ)
+∞
X
= d k · 2πiδn,k = 2πid n
k=−∞
I
The fifth equality follows from that the integral (z − z 0 )n dz is zero except for n = −1, where the value is given in Lemma
γ(z 0 ,ρ)
1.14.

Remark. The following examples illustrates some techniques in calculating Laurent series. For a specific function, using the inte-
grals given in the previous theorem is not economical or even feasible. One shall exploit the properties of some known expansions.

Example 2.14
1
Compute the Laurent series of f (z) = in the annulus A(0, 1, 2) := {z ∈ C : 1 < |z| < 2}.
(z − 1)2 (z + 2)

Solution. We know that 1/(1 − z) can be expanded into Taylor series for |z| < 1:

1 ∞
zn
X
=
1 − z n=0

1 1 1 w
For |z| > 1, let w := 1/z. Then =− −1
=− can be expanded into Taylor series:
z −1 z 1−z 1−w

1 w ∞ ∞
wn = − z −n
X X
=− = −w
1−z 1−w n=0 n=1

To expand f , we first split it into partial fractions:

1 1 1 1
f (z) = = − +
(z − 1)2 (z + 2) 9(z + 2) 9(z − 1) 3(z − 1)2
44 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

f has singularities at z = 1 and z = −2. Since 1 < |z| < 2 we expand 1/z and z/2 respectively:

1 1 1
f (z) = − −
18(1 + z/2) 9z(1 − z −1 ) 3z 2 (1 − z −1 )2
1 X∞ 1 1 X∞ 1 X∞
= (− )n z n − z −n + 2 (n + 1)z −n
18 n=0 2 9z n=0 3z n=0
∞
cn z n
X
=
n=−∞

where  µ ¶n
 1 −1 nÊ0
c n = 18 2
−(3n + 4)/9 n < 0


This gives the desired Laurent series.

2.2.2 Isolated Singularities.

Definition 2.15. Singularities.

Suppose U ⊆ C is a domain and D ⊆ U is the set of points at which f : U → C are holomorphic. We say that z 0 is a singularity of
f , if z 0 ∉ D and z 0 is a limit point of D. Especially, z 0 is said to be an isolated singularity, if there exists a deleted neighborhood
B (z 0 , r )\{z 0 } ⊆ D.

Definition 2.16. Classification of Isolated Singularities.

Suppose f : U → C has an isolated singularity at z 0 . We say that z 0 is a:

1. removable singularity, if lim f (z) exists and finite;

z→z 0

2. pole, if lim f (z) = ∞;

z→z 0

3. essential singularity, if lim f (z) does not exists in C∞ .

z→z 0

Remark. Often we concern the behavior of a function at infinity. We say that ∞ is a removable singularity (resp. pole/essential
singularity) of f , if 0 is a removable singularity (resp. pole/essential singularity) of g , where g is defined by g (z) = f (1/z).

Definition 2.17. Meromorphic Functions.

Suppose U is open in C and S ⊆ U is at most countable with no limit points in U . Then f : U \S → C is said to be a meromorphic
function, if f is holomorphic on U \S and has poles at the points in S.

Remark. It is easy to prove that f has the same Laurent expansion at A(z 0 , 0, r ) regardless of the choice of r . So we can safely say
the Laurent expansion "near the singularity". We can now show the connection between the isolated singularities and the Laurent
expansion near them. The general result is as follows:

1. removable singularity: no principal part;

2. pole: finitely many terms in the principal part;

3. essential singularity: infinitely many terms in the principal part.

Remark. First, for removable singularities, we already have the Riemann’s Removable Singularity Theorem (1.23), which will be
restated below.
2.2. LAURENT SERIES AND ISOLATED SINGULARITIES 45

Theorem 2.18. Riemann’s Removable Singularity Theorem.

Suppose f : U \{z 0 } → C is holomorphic. The following statements are equivalent:

(i) z 0 is a removable singularity of f ;

(ii) f is bounded near z 0 ;

(iii) f can be extended to a holomorphic function on U ;

+∞
c n (z − z 0 )n is a Laurent expansion of f in a deleted neighborhood of z 0 , then c n = 0 for all n < 0. That is, the
X
(iv) If
n=−∞
Laurent expansion coincides with the Taylor expansion.

Remark. Next we turn to poles. We will see that the characterization of poles plays an important role in computing integrals in the
next section.

Proposition 2.19

z 0 is a pole of f if and only if z 0 is a zero of 1/ f .

Proof. Trivial by algebra of limits.

Definition 2.20. Multiplicity of Poles.

Suppose z 0 is a pole of f . The multiplicity or order of z 0 of f is defined to be the multiplicity of z 0 as a zero of 1/ f .

A pole of order 1 is called a simple pole.

Proposition 2.21

f has a pole of order k at z 0 if and only if the Laurent expansion of f in a deleted neighborhood of z 0 is
∞
c n (z − z 0 )n (c −k 6= 0).
X
n=−k

Proof. "=⇒": By definition, z 0 is a zero of 1/ f with multiplicity k. That is, 1/ f (z) = (z − z 0 )k g (z) with g holomorphic in a neighbor-
1 ∞
a n (z − z 0 )n .
X
hood B (z 0 , r ) and g (z 0 ) 6= 0. Hence 1/g is also holomorphic in B (z 0 , r ). Suppose its Taylor expansion is =
g (z) n=0
Then in B (z 0 , r )\{z 0 } we have:
1 (z − z 0 )k ∞
a n+k (z − z 0 )n
X
= P∞ n
=⇒ f (z) =
f (z) n=0 a n (z − z 0 ) n=−k
∞
c n (z − z 0 )n in B (z 0 , r )\{z 0 } with c −k 6= 0. Then
X
"⇐=": Suppose f has a Laurent series
n=−k
∞
c n−k (z − z 0 )n = (z − z 0 )−k g (z)
X
f (z) = (z − z 0 )−k
n=0

g (z) is defined by a power series and hence is holomorphic in B (z 0 , r ). Moreover, g (z 0 ) = c −k 6= 0. Therefore 1/g (z) is also
holomorphic in B (z 0 , r ) and non-zero at z 0 . We have:

1 1
= (z − z 0 )k
f (z) g (z)

Hence z 0 is a zero of 1/ f of multiplicity k.

Remark. The next proposition is helpful in classifying the singularities of some known functions.
46 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

Proposition 2.22

Suppose f : U → C and g : U → C are holomorphic functions. If f has a zero of multiplicity m at z 0 ∈ U , and g has a zero of
multiplicity n at z 0 ∈ U , then the function h := f /g has a

1. removable singularity at z 0 , if m Ê n;

2. pole of order n − m at z 0 , if m < n.

Proof. It follows immediately from Proposition 2.21.

Theorem 2.23. Argument Principle for Meromorphic Functions.

Suppose U ⊆ C is a domain and S ⊆ U is an at most countable subset. f : U \S → C is meromorphic function with poles at the
points of S. Let γ be a piecewise-smooth simple closed path in U and f is non-zero on γ. Then we have:

1 f 0 (z)
I
N −P = dz
2πi γ f (z)

where N is the number of zeros of f inside γ and P is the number of poles of f inside γ (both counting multiplicity).

Proof. The proof is essential the same as 2.5. Suppose f has zeros a 1 , · · · , a n with multiplicity m 1 , · · · , m n and poles b 1 , · · · , b k with
multiplicity p 1 , · · · , p k in the interior of γ. We have:
I 0
f (z) X n I f 0 (z) X k I f 0 (z)
dz = dz + dz
γ f (z) i =1 γ(a i ,r i ) f (z) i =1 γ(b i ,s i ) f (z)

At each pole b i , g i (z) = (z − b i )p i f (z) is a power series and hence is holomorphic on U . Moreover, g i (b i ) = c −p i 6= 0. We have:
0
f 0 (z) g i (z) pi
= −
f (z) g i (z) z − b i
Therefore:
f 0 (z) g i0 (z) pi
I I I
dz = dz − dz = 0 − 2πip i
γ(b i ,s i ) f (z) γ(b i ,s i ) g i (z) γ(b i ,s i ) z − bi
We have already known the behavior of the integral near the zeros. Finally,

1
I 0
f (z) X n X k
dz = mi − pi = N − P
2πi γ f (z) i =1 i =1

Remark. We have proven in the preceding theorems that the Laurent series of f has no principal part near a removable singularity
and finite terms of principal part near a pole. The case of essential singularity is more complicated. The Laurent series has infinite
terms of principal parts. In fact we have a deeper result:

Theorem 2.24. Casorati-Weierstrass Theorem.

Suppose U ⊆ C is a domain and z 0 ∈ U . f : U \{z 0 } → C is a holomorphic function with an essential singularity at z 0 . Then for
all r > 0 with B (z 0 , r ) ⊆ U , the image set f (B (z 0 , r )\{z 0 }) is dense in C.

Proof. We argue by contradiction. Suppose:

∃ r > 0 ∃ w ∈ C ∃ ε > 0 ∀ z ∈ B (z 0 , r )\{z 0 } : | f (z) − w| > ε

Then g (z) = 1/( f (z) − w) is holomorphic in B (z 0 , r )\{z 0 } and bounded near z 0 . Therefore z 0 is a removable singularity of g
and lim g (z) exists. Hence
z→z 0
1
lim f (z) = w +
z→z 0 lim g (z)
z→z 0

exists in C or is equal to ∞, contradicting that z 0 is an essential singularity of f .

Remark. A significant generalization of Casorati-Weierstrass Theorem is the Picard’s Great Theorem, which states that, if z 0 is an
essential singularity of f , then for any deleted neighborhood of z 0 , f (z) assumes all possible complex values, with at most one
exception, infinitely many times.
2.3. WEIERSTRASS FACTORIZATION THEOREM* 47

2.3 Weierstrass Factorization Theorem*

2.3.1 Weierstrass Factorization Theorem.

Now we begin with discussion of the zeros of an entire function. Suppose f : C → C is an entire function with zeros at 0, a 1 , · · · , a n
with multiplicities m, m 1 , · · · , m n . (If f (0) 6= 0, then we put m = 0.) Consider the polynomial

z m1 z mn
µ ¶ µ ¶
m
p(z) = z 1 − ··· 1−
a1 an

Then the function g (z) = f (z)/p(z) has removable singularities at 0, a 1 , · · · , a n , so we can extend g to an entire function with no
zeros in C. By Proposition 1.38 (vii), since C is simply-connected, there exists an entire function h such that g (z) = eh(z) . Hence we
have expressed f as a product of a polynomial, whose zeros and multiplicities are the same as f , and an entire function without
zeros.
z m1 z mn
µ ¶ µ ¶
m h(z)
f (z) = z e 1− ··· 1−
a1 an
If f has infinitely many zeros in C, we can generalize the above formula to an infinite product:

Theorem 2.25. Weierstrass Factorization Theorem.

Suppose f : C → C is an entire function with zeros at a 1 , a 2 , · · · (counting multiplicity). {a n } ∈ C\{0} forms an infinite set which
has no limit points. Suppose furthermore that 0 is a zero of order m of f . (If f (0) 6= 0, then we put m = 0.) Then there exists an
entire function h : C → C such that
∞ µ µ ¶n−1 ¶
z z z
¶ µ
1
f (z) = z m eh(z)
Y
1− exp +···+
n=1 a n a n n − 1 a n

Before giving the formal proof, we first introduce the so-called elementary factors, named by Weierstrass.

Definition 2.26. Elementary Factors.

Let E 0 (z) = 1 − z. For p ∈ Z+ , we define

à !
p
zp zi
µ ¶
X
E p (z) = (1 − z) exp z + · · · + = (1 − z) exp
p i =1 i

Then Weierstrass Factorization Theorem can be written as

∞
f (z) = z m eh(z)
Y
E n−1 (z/a n )
n=1

Lemma 2.27

1
∃c > 0 ∀p ∈ N ∀z ∈ C : |z| É =⇒ |1 − E p (z)| É c|z|p+1
2

Proof. Observe that

à !
p p
zn
µ ∞ n
X z zn ∞ zn
µ ¶ ¶
X X X
E p (z) = exp log(1 − z) + = exp − + = exp −
n=1 n n=1 n n=1 n n=p+1 n

=: exp(w)

For |z| É 1/2, we have

zn
¯∞ ¯ ¯∞ ¯
p+1 ¯
¯X 1 ¯
¯ É |z|p+1 ¯ ¯ = 2|z|p+1
¯X ¯
|w| = |z| n¯
n=0 n + p + 1 n=0 2
¯ ¯ ¯

In particular, |w| É 2|z|p+1 É 1/2p É 1. Hence

¯ ∞ n−1 ¯ ∞ |w|n−1
¯X w
|1 − E p (z)| = |1 − ew | = |w| ¯¯
¯ X
¯ É |w|
n=1 n! n!
¯
n=1
48 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

∞ 1
= (e − 1)|w| É 2(e − 1)|z|p+1
X
É |w|
n=1 n!

Proof of Theorem 2.25. We shall show that the infinite product converges. Since {a n } is an infinite set in C without limit points, we
have lim |a n | = +∞. Fix any z ∈ C,
n→∞
{n ∈ Z+ : |z/a n | > 1/2}
is a finite set. By Lemma 2.27,
∞
|z/a n |n É c |1/2|n = c < +∞
X X X
|1 − E n−1 (z/a n )| É c
n: |z/a n |É1/2 n: |z/a n |É1/2 n=1

∞
X
Hence the series |1 − E n−1 (z/a n )| converges.
n=1

For x > 0, we have 1 + x < ex . Hence

∞
Y ∞
Y ∞
Y
|E n−1 (z/a n )| É (1 + |1 − E n−1 (z/a n )|) É exp(|1 − E n−1 (z/a n )|)
n=1 n=1 n=1
µ ∞
X
¶
= exp |1 − E n−1 (z/a n )| < +∞
n=1

∞
E n−1 (z/a n ) converges absolutely on C and uniformly on any compact subset of C. Now
Y
Hence the infinite product P (z) :=
n=1
f (z)
we observe that z m P (z) is an entire function with the same zeros and multiplicities as f (z). Therefore g (z) := has
z m P (z)
only removable singularities, and can be extended to an entire function without zeros. By Proposition 1.38 (vii), there exists
an entire function h(z) such that g (z) = eh(z) . Hence we have

∞ ∞ µ ¶n−1 ¶
z z z
µ ¶ µ
m h(z)
Y m h(z)
Y 1
f (z) = z e E n−1 (z/a n ) = z e 1− exp +···+
n=1 n=1 a n a n n − 1 a n

as required.

Remark. Weierstrass Factorization Theorem can also be stated as follows: we can always find an entire function with prescribed
zeros and multiplicities.

∞ ¯¯ z ¯n
X ¯
Remark. In the proof of Theorem 2.25, a key step is the convergence of ¯
¯a
¯ . This motivates us to make the following general-
¯
n=1 n
ization:

Corollary 2.28

Suppose f : C → C is an entire function with zeros at a 1 , a 2 , · · · (counting multiplicity). {a n } ∈ C\{0} forms an infinite set which
has no limit points. Suppose furthermore that 0 is a zero of order m of f . (If f (0) 6= 0, then we put m = 0.) If {k n } ⊆ Z is a integer
sequence such that
X∞ µ r ¶k n

n=1 |a n |

converges for all r > 0. Then there exists an entire function h : C → C such that
à µ ¶kn −1 !
∞ ∞ µ z
¶
z 1 z
m h(z) m h(z)
Y Y
f (z) = z e E kn −1 (z/a n ) = z e 1− exp +···+
n=1 n=1 an an n − 1 an

Proof. Trivial.

Corollary 2.29

Any meromorphic function on C is the quotient of two entire functions.

2.3. WEIERSTRASS FACTORIZATION THEOREM* 49

Proof. Suppose f is a meromorphic function on C with poles at a 1 , a 2 , · · · of order m 1 , m 2 · · · . By Weierstrass Factorzation Theorem,
we can construct an entire function g such that g has zeros at a 1 , a 2 , · · · with multiplicities m 1 , m 2 · · · .
g (z)
Then near each a n , lim (z − a n )mn f (z) exists and is finite, and lim 6= 0. Let h(z) = f (z)g (z). The limit
z→a n z→a n (z − a n )mn

g (z)
lim h(z) = lim (z − a n )mn f (z) · lim
z→a n z→a n z→a n (z − a n )mn

exists and is finite. Hence h has only removable singularities and can be extended to an entire function. Hence f = h/g is
the quotient of two entire functions.

Next we shall use the Weierstrass Factorization Theorem to prove some famous identities:

Example 2.30
1
1. π cot πz = (z ∉ Z);
X
n∈Z z + n
∞ µ z2
¶
Y
2. sin z = z 1− 2 2 ;
n=1 n π
X∞ 1 π2
3. 2
= .
n=1 n 6

∞
X 1 −∞
X 1 X 1 XN 1
Remark. Note that both and diverges. The sum should be interpreted as lim .
n=1 z + n n=−1 z + n n∈Z z + n N →∞ n=−N z + n
1
1. Let f (z) = π cot πz − defined on C\Z. First we observe that f is periodic:
X
Proof.
n∈Z z + n

1 X 1
f (z + 1) = π cot π(z + 1) − = π cot πz −
X
= f (z)
n∈Z z + n + 1 n∈Z z + n

Second, note that

1
lim πz cot πz = 1 and
X
lim z =1
z→0 z→0 n∈Z z + n
We have lim z f (z) = 0. Therefore z = 0 is a removable singularity of f . By periodicity, Z are removable singularities of
z→0
f . Hence f can be extended to an entire function on C.

We claim that f is bounded in {z ∈ C : | Re z| É 1/2}. Since f is holomorphic in that domain, it suffices to show that
f (x + yi) is bounded as |y| → ∞ for |x| É 1/2. We have
¯ −2πy
+ e−2πix ¯¯
¯
¯ e
|cot πz| = ¯¯i −2πy → 1 as |y| → ∞
e − e−2πix ¯

Also, for |x| É 1/2 and |y| > 1:

¯ ¯
¯ X 1 ¯ ¯¯ 1 X ∞ ¯ ¯
2z ¯¯ ¯¯ 1 ∞
X 2(x + yi)
¯
¯
¯=¯ + = +
¯ ¯ ¯ ¯
2 2 2 2 2
¯
¯n∈Z z + n ¯ z n=1 z − n ¯ ¯ x + yi n=1 x − y − n + 2x yi ¯
¯ ¯ ∞ ¯ ¯ ∞
¯ 1 ¯ X ¯ 2(x + yi) ¯
¯ É A +B
X y
ɯ¯ ¯ + ¯
x + yi n=1 x 2 − y 2 − n 2 + 2x yi
¯ ¯ ¯
n=1 y 2 + n2
Z ∞
y y
µ ¶
É A +B 2 2
dx is decreasing in x
0 x +y x2 + y 2
µ ¶x=∞
x 1
= A + B arctan = A ± πB
y x=0 2

where A, B > 0 are some constants. Hence f is bounded in {z ∈ C : | Re z| É 1/2}.

By periodicity, f is bounded on C. By Liouville’s Theorem, f is constant on C. But

à !
1 X 1
f (−z) = π cot(−πz) − = − π cot(πz) −
X
= − f (z)
n∈Z −z + n n∈Z z + n
50 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

1
for z ∈ C. We conclude that f (z) = 0 for all z ∈ C. Hence π cot(−πz) = for z ∈ C\Z.
X
n∈Z z + n

2. Note that sin z has zeros at z ∈ πZ. If we fix k n = 2, we see that

¶kn ∞ 1
r
µ
X 2r X
=
n∈Z\{0} |nπ| π n=1 n 2
2

converges for all r > 0. Hence by Corollary 2.28, we have

³ z ´ Y ³ z ´ z/nπ
sin z = z eh(z) = z eh(z)
Y
E1 1− e
n∈Z\{0} nπ n∈Z\{0} nπ

for some entire function h. Note that the infinite product is absolutely convergent, we have
∞ ³ z ´ z/nπ ³ z ´ −z/nπ ∞ µ z2
¶
h(z) h(z)
Y Y
sin z = z e 1− e 1+ e =ze 1− 2 2
n=1 nπ nπ n=1 n π

It remains to determine the function h. Taking logarithm on both sides:

∞ z2
µ ¶
X
log sin z = log z + h(z) + log 1 − 2 2
n=1 n π

Taking derivative:
1 ∞ 2z
+ h 0 (z) +
X
cot z = 2 − n 2 π2
z n=1 z
or
1 ∞ 2z
π cot πz = + h 0 (πz) +
X
z z 2 − n2
n=1

Combining with the first part, we deduce that h 0 (πz) = 0 for z ∈ C\Z. Hence h(z) = const. By evaluating at z = 0 we
∞ µ z2
¶
Y
deduce that h(z) = 0. Hence sin z = z 1− 2 2 .
n=1 n π
3. By expanding sin z into Taylor series at z = 0, we obtain:
∞ µ
∞ (−1)n 2n+1 z2
Y ¶
X
z =z 1− 2 2
n=0 (2n + 1)! n=1
n π

By comparing the coefficient of z 3 , we obtain

∞
X 1 1
− =−
n=1 n 2 π2 6
Thus the result follows.

X∞ 1 π2
Remark. 2
= can also be proven by calculus of residue in the next chapter. See Example 3.18 for detail.
n=1 n 6

2.3.2 Mittag-Leffler Theorem.

Theorem 2.31. Mittag-Leffler Theorem.

kn c n, j
Suppose {a n } is a sequence of complex numbers in C without limit points. ϕn :=
X
is a sequence of rational func-
(z − a n ) j j =1
tions with c n, j ∈ C. Then there exists a meromorphic function f : U → C such that f has poles at {a n } and principal parts ϕn at
each pole.

Remark. Mittag-Leffler Theorem states that we can always find a meromorphic function with prescribed poles and principal parts
near each pole.
2.3. WEIERSTRASS FACTORIZATION THEOREM* 51

∞ ϕ(k) (0)
n
Proof. For each n ∈ Z+ , ϕn is a polynomial of 1/(z − a n ). Hence ϕn is holomorphic in B (0, |a n |) and has Taylor series zk ,
X
k=0 k!
which converges uniformly to ϕn in B (0, |a n |/2). Hence there exists s n ∈ N such that
¯ ¯
¯ sn
ϕ(k)
n (0) k ¯
¯ 1
¯ϕ n −
X
z ¯< n
¯
¯ k=0 k! ¯ 2
Ps n
for z ∈ B (0, |a n |/2). We denote k=0
ϕ(k) k
n (0)z /k! by p n (z).

Fix R > 0. Since {a n } is an infinite set in C without limit points, we have lim |a n | = +∞. Let
n→∞

N := max {n ∈ Z+ : |a n | É 2R}

Then for z ∈ B (0, R) and n > N , |z| É R < |a n |, the function defined by ϕn (z) − p n (z) is holomorphic. Moreover,
∞ ¯ ∞
¯ϕn (z) − p n (z)¯ É
X ¯ X
<1
n=N +1 n=N +1

∞ ¡
ϕn (z) − p n (z) converges absolutely and uniformly in B (0, R). Hence it defines a holomorphic function in
X ¢
The series
n=N +1
B (0, R).

Therefore the series

∞ ¡ N ¡ ∞ ¡
ϕn (z) − p n (z) = ϕn (z) − p n (z) + ϕn (z) − p n (z)
X ¢ X ¢ X ¢
n=1 n=1 n=N +1

defines a meromorphic function in B (0, R), with prescribed poles at {a n : |a n | < R} and corresponding principal parts {ϕn :
∞ ¡
ϕn (z) − p n (z) is a meromorphic function on C with the desired properties.
X ¢
|a n | < R}. Since R is arbitrary,
n=1

2.3.3 Interpolation Theorem.

Mittag-Leffler Theorem may be combined with Weierstrass Factorization Theorem to give a solution to the interpolation problem:
Given a infinite set {a n } ∈ C without limit points, we want to find an entire function with prescribed values on the set. In fact the
result is much stronger. We can also prescribe finitely many derivatives at each a n .

Theorem 2.32. Interpolation Theorem.

Suppose {a n } is a sequence of complex numbers in C without limit points. For each a n we associate a non-negative integer
m n and a sequence of complex numbers c k,n , (0 É k É m n ). Then there exists an entire function f such that f (k) (a n ) = k!c k,n
for each a n and 0 É k É m n .

In other words, f has the prescribed m n terms of Taylor expansion at each a n .

Proof. By Weierstrass Factorization Theorem, we can find an entire function g such that g has zeros at each a n with multiplicities
m n + 1. We claim that we can associate each a n with a rational function
mX
n +1 d k,n
ϕn (z) =
k=1 (z − a n )k

such that g (z)ϕn (z) has a Taylor expansion near a n :

m
Xn
c k,n (z − a n )k + O (z − a n )mn +1
¡ ¢
g (z)ϕn (z) =
k=0

∞
b k (z − a n )k+mn +1 near a n . Then
X
Suppose that we are given that g (z) =
k=0
à ! à !
mX
n +1 d k,n ∞
k+m n +1
X
g (z)ϕn (z) = · b k (z − a n )
k=1 (z − a n )k k=0
à !
∞
mn k
¡ ¢ X
= d mn +1,n + d mn ,n (z − a n ) + · · · + d 1,n (z − a n ) · b k (z − a n )
k=0
52 CHAPTER 2. SERIES REPRESENTATION OF FUNCTIONS

Comparing the coefficients, we have:

b 0 · d mn +1,n = c 0,n
b 1 · d mn +1,n + b 0 · d mn ,n = c 1,n
······
k
X
b k− j · d mn +1− j ,n = c k,n
j =0

Thus we can determine d mn +1,n , · · · , d 1 successively, since b 0 6= 0. In this way we have obtained the desired functions ϕn .

By Mittag-Leffler Theorem, we can find a meromorphic function h on C such that h has poles at each a n with principal
parts ϕn . Now f (z) := g (z)h(z) has only removable singularities and can be extended to an entire function with the desired
properties.

Remark. The Interpolation Theorem can also help determine the structure of the ring of holomorphic functions. Suppose U ∈
C is a domain. Let H(U ) denotes the set of all holomorphic functions on U . Then H(U ) is a ring under function additions and
multiplications. We have the following proposition:

Proposition 2.33

Every finitely-generated ideal in H(U ) is principal.

Proof. Given f 1 , · · · , f n ∈ H(U ). The corollary states that there exists f ∈ H(U ) such that 〈 f 1 , · · · , f n 〉 = 〈 f 〉. But first we shall prove the
following lemma: If g 1 , · · · , g n ∈ H(U ), if none of them is identically zero in U , and if they have no common zeros in U , then
〈g 1 , · · · , g n 〉 = 〈1〉 = H(U ).

We use induction on n. Suppose that the lemma holds for any n − 1 functions with no common zeros. Let m(g i ; α) denotes
the multiplicities of zero of g i at α ∈ U . By Weierstrass Factorization Theorem (applied to a domain rather than the whole C,
which is still valid), we can find ϕ ∈ H(U ) such that m(ϕ, α) = min m(g i , α) for each α ∈ U . This is practical because the
1Éi Én−1
common zeros of g 1 , · · · , g n−1 is a discrete point set.

Let h i := g i /ϕ for i = 1, · · · , n−1. Then h 1 , · · · , h n−1 have no common zeros. By the induction hypothesis, we have 〈h 1 , · · · , h n−1 〉 =
H(U ). Hence 〈g 1 , · · · , g n−1 , g n 〉 = 〈ϕ, g n 〉.

Since g 1 , · · · , g n have no common zeros, ϕ and g n have no common zeros. By Interpolation Theorem (again applied to a
domain U ), we can find ψ ∈ H(U ) such that m(1−ψg n ; α) Ê m(ϕ; α) for all α ∈ U . This could be done by prescribing the value
of ψ(k) (α) for 0 É k É m(ϕ; α) at each α ∈ U such that m(ϕ; α) > 0.

Hence ξ := (1 − ψg n )/ϕ has only removable singularities and can be extended to a holomorphic function in U . That is,
∃ ψ, ξ ∈ H(U ) : 1 = ψg n + ϕξ.

Then 1 ∈ 〈ϕ, g n 〉. 〈g 1 , · · · , g n 〉 = 〈ϕ, g n 〉 = H(U ), which completes the proof.

Chapter 3

Calculus of Residues

3.1 Residue Theorem 53

3.2 Semicircular Contour 54
3.3 Jordan’s Lemma 56
3.4 Keyhole Contour 59
3.5 Infinite Series 61
3.6 Some More Examples 63

3.1 Residue Theorem

Definition 3.1. Residue.

∞
Suppose f has an isolated singularity at z 0 ∈ C. The Laurent series of f in a deleted neighborhood is c n (z − z 0 )n . We call
X
n=−∞
the coefficient c −1 the residue of f at z 0 and denote it by Res( f , z 0 ).

Proposition 3.2. Residue at a Pole.

If f has a pole of order n at z 0 , then the residue of f at z 0 is given by:

µ ¶n−1
1 d
(z − z 0 )n f (z)
¡ ¢
Res( f , z 0 ) = lim
(n − 1)! z→z 0 dz

∞ ∞
c m (z −z 0 )m . Then (z −z 0 )n f (z) = c m−n (z −z 0 )m is can be holomorphically
X X
Proof. Suppose the Laurent series of f near z 0 is
m=−n m=0
extended to z 0 . We have:
¶n−1
d n−1 X ∞
µ µ ¶
d
(z − z 0 )n f (z) = lim c m−n (z − z 0 )m
¡ ¢
lim
z→z 0 dz z→z 0 dz
m=0
∞ m!
c m−n (z − z 0 )m−n+1
X
= lim
m=n−1 (n + 1)!
z→z 0

= (n − 1)! c −1 = (n − 1)! Res( f , z 0 )

Corollary 3.3. Residue at a Simple Pole.

If f has a simple pole at z 0 , then the residue of f at z 0 is given by:

Res( f , z 0 ) = lim (z − z 0 ) f (z)

z→z 0

Proposition 3.4

Suppose g , h are holomorphic at z 0 ∈ C, with g (z 0 ) 6= 0, h(z 0 ) = 0, and h 0 (z 0 ) 6= 0. Then z 0 is a simple pole of f (z) := g (z)/h(z).
Moreover, we have Res( f , z 0 ) = g (z 0 )/h 0 (z 0 ).

53
54 CHAPTER 3. CALCULUS OF RESIDUES

h(z 0 ) h 0 (z 0 )g (z 0 ) − h(z 0 )g 0 (z 0 ) h 0 (z 0 )
Proof. 1/ f (z 0 ) = = 0 and (1/ f )0 (z 0 ) = = 6= 0. So z 0 is a simple zero of 1/ f and hence a simple
g (z 0 ) g (z 0 )2 g (z 0 )
pole of f . By Corollary 3.3, we have:

g (z) g (z) g (z 0 )
Res( f , z 0 ) = lim (z − z 0 ) = lim =
z→z 0 h(z) z→z0 h(z)−h(z 0 ) h 0 (z 0 )
z−z 0

Theorem 3.5. Residue Theorem.

Suppose U ⊆ C is a domain and S ⊆ U is an at most countable subset with no limit points in U . f : U \S → C is holomorphic
with isolated singularities at the points of S. γ is piecewise-smooth closed path contained in U with S ∩γ∗ = ∅. Then we have:
I X
f = 2πi I (γ, a) Res( f , a)
γ a∈S

Proof. Since γ∗ is bounded, without loss of generality we can assume

X that S is finite. For a ∈ S, let p a be the principal part of f near
a. Then f − p a is holomorphic at a ∈ S. Consequently f − p a is holomorphic on S. Apply Cauchy’s Theorem:
a∈S
I XI
0= f (z) dz − p a (z) dz
γ a∈S γ

But we also have:

I I −∞ +∞
XI c −n
c n (z − z 0 )n dz =
X
p a (z) dz = n
dz
γ γ n=−1 n=1 γ (z − z 0 )
c −1
I
= dz = 2πi · I (γ, a) Res( f , a)
γ z − z0

The second equality follows from the uniform convergence of the principal part. The third equality follows from that (z −z 0 )n
always has a primitive in U except for n = −1.
I X
Therefore we have f (z) dz = 2πi I (γ, a) Res( f , a) as required.
γ a∈S

Remark. Residue Theorem provides a powerful tool that transforms the calculation of integrals into the calculation of the residue
of singularities, which, by Proposition 3.2, is essentially doing differentiation.

Now we will give a number of examples to demonstrate the power of Residue Theorem in computing integrals, especially improper
integrals on the real line. For more techniques and examples of residue calculus, please refer to [Priestley].

3.2 Semicircular Contour

Z +∞
We first consider improper integrals of the type f (x) dx:
−∞

Theorem 3.6

Suppose f is holomorphic on the upper half plane {z ∈ C : Im z Ê 0} except at a finite set S. If lim z · f (z) = 0, then we have:
z→∞
Z +∞ X
f (x) dx = 2πi Res( f , a)
−∞ a∈S

provided the improper integral on the left side exists.

Proof. We will use a semicircular contour in the upper half plane that consists of a line segment [−R, R] and a semicircle γR :=
γ+ (0, R), as shown in Figure 3.1. By Residue Theorem, for sufficiently large R, we have:
Z R Z X
f (x) dx + f (z) dz = 2πi Res( f , a) (3.1)
−R γR a∈S
3.3. JORDAN’S LEMMA 55

Let M (R) := sup | f (z)|. We know that lim R · M (R) = 0. Then:

z∈γ∗ R→∞
R

π
¯Z ¯ ¯Z ¯
iθ iθ
dθ ¯¯ É πR M (R) → 0
¯ ¯ ¯ ¯
¯ f (z) dz ¯¯ = ¯¯ f (R e ) · Ri e
γR
¯
0

as R → ∞.

By letting R → ∞ in the Equation (3.1), we have

Z +∞ X
f (x) dx = 2πi Res( f , a)
−∞ a∈S

as required.
P
Remark. We must prove the existence of the improper integral before we apply the theorem. If the integral diverges, then 2πi a∈S Res( f , a)
only gives the Cauchy principal value of the integral. We should write:
Z +∞ X
P.V. f (x) dx = 2πi Res( f , a)
−∞ a∈S

Remark. There is nothing special about our choice of the semicircle in the upper half plane. One gets the same conclusion if one
uses the semicircle in the lower half plane, with the other pole and the appropriate residue.

Im
γR

Re

−R R

Figure 3.1: A semicircular contour.

Example 3.7
Z +∞
1
Evaluate dx.
−∞ 1 + x2 + x4

Solution. Let us consider the function f (z) = 1/(1 + z 2 + z 4 ) defined on the upper half plane. Notice that 1/ f (z) = 1 + z 2 + z 4 =
(z 6 −1)/(z 2 −1) has simple zeros at eπi/3 , e2πi/3 , e4πi/3 and e5πi/3 . Then, in the upper half plane, f has simple poles at ω = eπi/3
and ω2 = e2πi/3 .

We calculate the residues by using Corollary 3.3:

z −ω 1 1
Res( f , ω) = lim = =−
z→ω z4 + z2 + 1 (ω − ω2 )(ω + ω)(ω + ω2 )
2(1 − ω2 )
2
z −ω 1 1
Res( f , ω2 ) = lim 4 = 2 =
z→ω 2 z + z 2 + 1 (ω − ω)(ω2 + ω)(ω2 + ω2 ) 2ω(1 − ω2 )
Note that
1
lim z f (z) = lim =0
z→∞ z→∞ z −1 + z + z 3
Now we can apply Theorem 3.6:
Z +∞ 1
dx = 2πi Res( f , ω) + Res( f , ω2 )
¡ ¢
2
1+x +x 4
−∞
πi
µ ¶
1 1
= 2πi · − 1 = 2
2(1 − ω2 ) ω ω +ω
π
=p
3
56 CHAPTER 3. CALCULUS OF RESIDUES

3.3 Jordan’s Lemma

Z +∞
When evaluating integrals of the type eiαx f (x) dx, we need the following lemma:
−∞

Lemma 3.8. Jordan’s Lemma.

Suppose f is continuous on the upper half plane (with possible exceptions at a finite set). If lim f (z) = 0, then:
z→∞
Z
lim eiαz f (z) dz = 0
R→∞ γR

for any α > 0, where γR (t ) = γ+ (0, R)(t ) = R eit , t ∈ [0, π] is a semicircular path.

Proof. Since sin x is convex on [0, π/2], by Jensen’s Inequality, we have:

2x π π 2
µµ ¶ ¶ µ ¶
2x 2 2x
sin x = sin 1 − 0+ · Ê 1− sin 0 + sin = x
π π 2 π π 2 π

for x ∈ [0, π/2].

Let M (R) := sup | f (z)|. We know that lim M (R) = 0. Then:

z∈γ∗ R→∞
R

π
¯Z ¯ ¯Z ¯
eiαz f (z) dz ¯¯ = ¯¯ eiαR cos θ e−αR sin θ f (R eiθ )Ri eiθ dθ ¯¯
¯ ¯ ¯ ¯
¯
γR
¯
0
Z π Z π/2
2α
É R · M (R) e−αR sin θ dθ É 2R · M (R) e− π Rθ dθ
0 0
π ¡
1 − e−αR → 0
¢
= 2M (R) ·
2α
as R → ∞.

Corollary 3.9

Suppose f is holomorphic on the upper half plane {z ∈ C : Im z Ê 0} except at a finite set S. If lim f (z) = 0, then for α > 0 we
z→∞
have: Z +∞
eiαx f (x) dx = 2πi Res eiαz f (z), a
X ¡ ¢
−∞ a∈S

Proof. This is exactly like the proof of Theorem 3.6. We know from Jordan’s Lemma that the integral on the upper semicircular path
tends to zero, and the result follows.

Remark. We can also translate the complex exponential into real trigonometrics:
Z +∞ Ã !
¡ iαz
cos αx f (x) dx = Re 2πi
X ¢
Res e f (z), a
−∞ a∈S
à !
Z +∞ ¡ iαz
sin αx f (x) dx = Im 2πi
X ¢
Res e f (z), a
−∞ a∈S

Example 3.10
Z +∞
sin x
Evaluate dx.
−∞ x

Solution. Normally we will consider eiz /z on the upper half plane. But it has a simple pole at the origin, which is on the contour
we are about to integrate along. So we consider f (z) = (eiz −1)/z instead. By expanding the exponential function we can see
that
eiz −1
lim =i
z→0 z
3.3. JORDAN’S LEMMA 57

Therefore z = 0 is a removable singularity of f , which can be extended to an entire function on the plane. By Cauchy’s
Theorem, along the semicircular contour we have:

R eix −1 eiz 1
I Z Z Z
0= f (z) dz = dx + dz − dz (3.2)
−R x γR z γR z

The third term in the Equation (3.2) is just:

Z
1
Z π iR eiθ
dz = dθ = iπ
γR z 0 R eiθ
Letting R → ∞, the second term in the Equation (3.2) vanishes by Jordan’s Lemma. Hence we have:

∞ eix R eix R eix −1

Z Z Z
dx = lim dx = lim dx = iπ
−∞ x R→∞ −R x R→∞ −R x

Take the imaginary part: Z ∞ sin x

dx = π
−∞ x

Remark. If the integrand function has singularities on the real line. We often indent the contour by a small circular arc around the
singularity. The following lemma shows that it works for simple poles.

Lemma 3.11. Indentation Lemma.

Suppose f is holomorphic on the sector z = z 0 + r eiθ : r ∈ (0, R], θ ∈ [θ1 , θ2 ] and has a simple pole at z 0 . Then:
© ª

Z
lim f (z) dz = i(θ2 − θ1 ) Res( f , z 0 )
r →0 γr

where γr (t ) = z 0 + r eit , t ∈ [θ1 , θ2 ].

Proof. Suppose Res( f , z 0 ) = A. Let g (z) = (z − z 0 ) f (z) − A. Then lim g (z) = 0. We have
z→z 0

g (z) 1 g (z)
Z Z Z Z
f (z)dz = dz + A dz = iA(θ2 − θ1 ) + dz
γr γr z − z0 γr z − z0 γr z − z0

where
¯ ¯¯Z θ ¯
¯ ¯ 2 g (z 0 + r eiθ )
¯Z
¯ g (z) iθ
¯
dz ¯¯ = ¯ r i e dθ
¯ ¯
r eiθ
¯
¯
γr z − z0 ¯ θ1 ¯
¯Z θ ¯
¯ 2
ig (z 0 + r eiθ )dθ ¯¯ É sup |g (z)|(θ2 − θ1 ) → 0
¯
= ¯¯
θ1 z∈γ∗
r

f (z) dz → iA(θ2 − θ1 ) as r → 0 as required.

R
as r → 0. Hence γr

Remark. The indentation only works for simple poles and fails at other types of singularities. This is because the contour has length
O(ε) and the integrand grows as O(ε−1 ) near a simple pole.

Im
γR

γρ

Re

−R −ρ ρ R

Figure 3.2: A semi-annular contour.

58 CHAPTER 3. CALCULUS OF RESIDUES

Example 3.12
Z +∞ sin x
Use Lemma 3.11 to evaluate dx.
−∞ x

Solution. Let f (z) = eiz /z. We integrate it along the semi-annular contour as shown in Figure 3.2. Since f is holomorphic inside
and on the contour, by Cauchy’s Theorem we have:

−ρ eix eiz R eix eiz

Z Z Z Z
0= dx − dz + dx + dz
−R x γρ z ρ x γR z

where γρ (t ) = ρ eit , t ∈ [0, π] and γR (t ) = R eit , t ∈ [0, π].

By Jordan’s Lemma, the fourth term vanishes as R → ∞.By Lemma 3.11, since lim z · eiz /z → 1, the second term
z→0

eiz
Z
dz → iπ
γρ z

as ρ → 0. Therefore we have:
+∞ eix −ρ eix R eix
Z Z Z
dx = lim dx + lim dx = iπ
−∞ x R→∞ −R x R→∞ ρ x
ρ→0 ρ→0

We can see that this method produces the same answer as the previous one.

Example 3.13
Z ∞
ln x
Evaluate dx.
0 (1 + x 2 )2

Solution. Let f (z) = log z/(1+z 2 )2 . f has a simple pole at 0 and double poles at i and −i. We wish to indent at z = 0 and use the semi-
annulur contour in Figure 3.2. Since logarithm is multi-valued, we have to choose a holomorphic branch of the logarithm
such that the cut line does not cross the contour. We use the principal logarithm here. First let us calculate the residue. By
Proposition 3.2,

d log z d log z (z + i)2 /z − 2(z + i) log z π i

Res( f , i) = lim (z − i)2 2 2
= lim 2
= lim = +
z→i dz (1 + z ) z→i dz (z + i) z→i (z + i)4 8 4

For x ∈ R− , we have ln x = ln |x| + iπ. Apply the Residue Theorem:

Z −ρ ln |x| + iπ
Z
log z
Z R ln x
Z
log z
dx − dz + dx + dz = 2πi Res( f , i)
−R (1 + x 2 )2 γρ (1 + z 2 )2 ρ (1 + x 2 )2 γR (1 + z 2 )2

log z
For the second term, since lim z · = 0, 0 is a simple pole of f . By Lemma 3.11, we have:
z→0 (1 + z 2 )2

log z
Z
lim dz = 0
ρ→0 γρ (1 + z 2 )2

log z
For the fourth term, lim z · = 0, we have:
z→∞ (1 + z 2 )2

log z
Z
lim dz = 0
R→∞ γR (1 + z 2 )2

Combining the preceding equations we have:

Z −ρ ln |x| + iπ ln x
Z R
lim dx + lim dx = 2πi Res( f , i)
ρ→0 −R (1 + x 2 )2 ρ→0 2 2
ρ (1 + x )
R→∞ R→∞
π i
Z ∞ Z 0
ln x
µ ¶
1
=⇒ 2 2 2
dx + iπ 2 2
dx = 2πi +
0 (1 + x ) −∞ (1 + x ) 8 4
3.4. KEYHOLE CONTOUR 59

Im

γ2 γR

π/4
O γ1 R Re

Figure 3.3: A sectorial contour

Taking the real part:

∞ ln x π
Z
2 2
dx = −
0 (1 + x ) 4
By taking the imaginary part, we also obtain that:
∞ 1 π
Z
dx =
0 (1 + x 2 )2 4

Remark. We will see later that the previous example can also be solved by a keyhole contour.

Example 3.14. Fresnel Integrals.

Z ∞ Z ∞
Evaluate sin x 2 dx and cos x 2 dx.
0 0

2
Solution. We consider the function f (z) = eiz , which is entire on C. We wish to apply Cauchy’s Theorem. A semicircular contour
2
would not work, because eiz does not tends to zero as R → ∞. Alternatively we can consider a contour that goes around a
sector, as shown in Figure 3.3. Let the contour be Γ := γ1 ? γR ? γ2 , where γR (t ) = R eit , t ∈ [0, π/4] and γ2 (t ) = −t eπi/4 , t ∈
[−R, 0]. By Cauchy’s Theorem:
Z R Z Z
2 2 2
eix dx + eiz dz + eiz dz = 0
0 γR γ2
Now the second term tends to zero as R → ∞, following from the same argument in the proof of Jordan’s Theorem (applying
the Jensen’s Inequality). For the third term,
Z Z ∞ Z ∞
2 2 πi/2 2
lim eiz dz = − eit e eπi/4 dt = − eπi/4 e−t dt
R→∞ γ2 0 0
p Ãp p !
πi/4 π 2π 2π
= −e =− +i
2 4 4
Hence p
Z ∞ Z ∞ 2π
2 2
sin x dx = cos x dx =
0 0 4

Remark. In the previous example we have used the well-known result:

Z +∞
2 p
e−x dx = π
−∞

which is usually derived from a double integral in the polar coordinates. An alternative way which utilizes residue calculus is given
in Example 3.23.

3.4 Keyhole Contour

Z ∞
We consider the integrals of the type f (x) dx, where f depends explicitly or implicitly on logarithm. We need to select a holo-
0
morphic branch by drawing a cut line on the plane. For example we choose R+ as the cut line and consider the following contour Γ.
Let γ(0, R) and γ(0, ρ)− be two circular paths, where we make R → ∞ and ρ → 0. We join the two circles by two line segments with
a narrow neck in between, as shown in Figure 3.4. Our choice will make log z holomorphic inside the contour. Moreover, we have
log z = ln |z| just above R+ and log z = ln |z| + 2πi just below R+ .
60 CHAPTER 3. CALCULUS OF RESIDUES

Im

γR

γρ γ+

γ− Re

Figure 3.4: A keyhole contour.

Example 3.15
∞ x p−1
Z
Evaluate dx, where m ∈ Z+ and p ∈ (0, m)\Z.
0 (1 + x)m

z p−1 e(p−1) log z

Solution. Let f (z) = = . Since p ∉ Z, f is multi-valued. We choose the positive real axis as the cut line and con-
(1 + z)m (1 + z)m
sider the keyhole contour Γ = γR ? γ− ? γ− ρ ? γ+ in Figure 3.4.

On γ+ , z p−1 = e(p−1) ln x = x p−1 . On γ− , z p−1 = e(p−1) log z = e(p−1)(ln x+2πi) = e2pπi x p−1 . We can see that f has a pole of order m
at z = −1. By Residue Theorem, we have:
R x p−1 z p−1 R e2pπi x p−1 z p−1
Z Z Z Z
dx + dz − dx − dz = 2πi Res( f , −1)
ρ (1 + x)m γR (1 + z)m ρ (1 + x)m γρ (1 + z)m

Since γR (t ) = R eit , t ∈ [0, 2π], we have:

z p−1
¯Z ¯ ¯Z 2π p−1 (p−1)it ¯
¯ ¯ ¯ R e it
¯
¯ dz =
¯ ¯ Ri e dt ¯¯
¯
γR (1 + z)m ¯ ¯ 0 (1 + R eit )m
¯ p−1 (p−1)it
Rp
¯
¯R e it ¯
¯
É 2π · sup ¯
¯ Ri e É 2π →0
it m
z∈γ∗ (1 + R e )
¯ (R − 1)m
R

as R → ∞, because p < m.

Similarly on γρ , we have: ¯Z ¯
¯ z p−1 ¯ ρp
dz É 2π →0
¯ ¯
¯ γρ (1 + z)m ¯ (ρ − 1)m
¯ ¯

as ρ → 0.

Then by letting R → ∞ and ρ → 0 in the Residue Theorem, we obtain that

∞ x p−1
Z
1 − e2pπi
¡ ¢
dx = 2πi Res( f , −1)
0 (1 + x)m
Now we compute the residue. If m = 1, then

z p−1
Res( f , −1) = lim (z + 1) = (−1)p−1 = − epπi
z→−1 (1 + z)
If m > 1, then

dm−1 m z
p−1 ¶
µ
1
Res( f , −1) = lim (z + 1)
(m − 1)! z→−1 dz m−1 (1 + z)m
1
= lim (p − 1)(p − 2) · · · (p − m + 1)z p−m
(m − 1)! z→−1
3.5. INFINITE SERIES 61

Im
(N + 21 )(−1 + i) (N + 12 )(1 + i)

ΓN

−N − 1 N +1

−N ··· −1 O 1 ··· N Re

(N + 21 )(−1 − i) (N + 12 )(1 − i)

Figure 3.5: A square contour.

epπi (1 − p)(2 − p) · · · (m − 1 − p) m−1 p

= − epπi
Y
=− (1 − )
(m − 1)! j =1 j

Hence we have:
∞ x p−1 − epπi 1 π
Z
dx = 2πi = 2πi · =
0 1+x 1 − e2pπi 2i sin pπ sin pπ
and for m > 1:
∞ x p−1 π m−1 p
Z Y
m
dx = (1 − )
0 (1 + x) sin pπ j =1 j

3.5 Infinite Series

Residue Theorem are also useful in calculating the sum of an infinite series. We will use the function cot πz. Since sin z has simple
zeros at πZ, cot πz has simple poles at Z. We have the following lemma:

Lemma 3.16

Suppose that ϕ is holomorphic near n ∈ Z with ϕ(n) 6= 0. Then f (z) := ϕ(z) cot πz has a simple pole at z = n with residue
Res( f , n) = ϕ(n)/π.

Proof. It is easy to see that z = n is a simple pole of f . So the residue is calculated by

(z − n) cos πz
Res( f , n) = lim (z − n)ϕ(z) cot πz = ϕ(n) lim
z→n sin πz z→n
cos πz cos πn + sin πz sin πn
= ϕ(n) lim (z − n) (sin πn = 0)
z→n sin πz cos πn − cos πz sin πn
(z − n) cos π(z − n) ϕ(n) w(1 + O(w 2 ))
= ϕ(n) lim = lim (w := π(z − n))
z→n sin π(z − n) π w→0 w + O(w 3 )
= ϕ(n)/π

Now let us consider the positively oriented square contour ΓN with vertices at (N + 21 )(±1 ± i). The next lemma shows that cot πz is
uniformly bounded on all Γ∗N .

Lemma 3.17

There exists a C > 0 such that sup | cot πz| < C for all N ∈ N.
z∈Γ∗N

Proof. We consider the horizontal and vertical sides of the square separately. On the horizontal sides, we have z = x ± (N + 21 )i, x ∈
[−(N + 12 ), N + 12 ].
¯ ¯ 1 1
¯
¯ e + e−iπz ¯ ¯¯ eiπ(x±i(N + 2 )) + e−iπ(x±i(N + 2 )) ¯¯
¯ iπz
| cot πz| = ¯¯ iπz ¯=¯
e − e−iπz ¯ ¯ eiπi(x±(N + 21 )) − e−iπ(x±i(N + 12 )) ¯
¯
62 CHAPTER 3. CALCULUS OF RESIDUES

¯ eiπx∓π(N + 12 ) + e−iπx±π(N + 21 ) ¯
¯ ¯
=¯
¯ ¯
¯ eiπx∓π(N + 12 ) − e−iπx±π(N + 21 ) ¯
¯

¯ eπ(N + 12 ) + e−π(N + 12 ) ¯
¯ ¯
ɯ ¯ = | coth π(N + 1/2))|
¯ ¯
¯ eπ(N + 12 ) − e−π(N + 12 ) ¯
¯ π ¯¯
É ¯coth ¯ (since coth x is decreasing when x > 0)
¯
2

On the vertical sides, we have z = ±(N + 21 ) + yi, y ∈ −(N + 21 ), N + 12 . From elementary trigonometric equalities, we know
£ ¤

that cot(z + π/2) = − tan z and that tan(z + N π) = tan z for N ∈ Z. We have:

| cot πz| = | cot(±(π/2 + N π) + iπy)| = | tan iπy| = | tanh πy| < 1

³ ¯ π ¯¯´
Hence we have sup | cot πz| < min 1, ¯coth ¯ for all N ∈ N.
¯
z∈Γ∗ 2
N

Remark. Given the two preceding lemmata, we are now able to sum some infinite series.

Example 3.18
X∞ 1
Evaluate 2
.
n=1 n

Solution. By Cauchy’s integral test the series converges. Let f (z) = cot πz/z 2 . We consider the square contour ΓN defined above. By
1
Lemma 3.16, we know that f has simple poles at z = n ∈ Z\{0} with residues Res( f , n) = . In addition, z = 0 is a triple pole
πn 2
of f . The residue is given by

d2 π π2 z cos πz
µ ¶
1 1
Res( f , 0) = lim 2 z cot πz = lim −2 2 +2
2 z→0 dz 2 z→0 sin πz sin3 πz
πz cos πz − sin πz
= π lim
z→0 sin3 πz
πz(1 − π2 z 2 /2 + O(z 4 )) − (πz − π3 z 3 /6 + O(z 5 ))
= π lim
z→0 (πz + O(z 3 ))3
3 3
−π /2 + π /6 π
=π =−
π3 3
Apply the Residue Theorem to f along ΓN :

N π N 1
I X X
f (z) dz = Res( f , n) = +2
n=1 πn
3 2
ΓN n=−N

By Lemma 3.17, we have

cot πz ¯¯
¯I ¯ ¯I ¯
¯ ¯ ¯ C
¯ f (z) dz ¯¯ = ¯¯ 2
dz É 4 · (2N + 1) · →0
¯
ΓN ΓN z ¯ (N + 1/2)2
as N → ∞.

Therefore:

π X∞ 1 X∞ 1 π2
0=− +2 =⇒ =
n=1 πn
2 2
3 n=1 n 6

Remark. In the previous example, the residue of cot πz/z 2 at z = 0 can also be calculated by directly expand sin z and cos z at z = 0:
¶−1
cos z z2 z3
µ ¶µ
cot z = = 1− + O(z 4 ) z − + O(z 5 )
sin z 2 6
¶¶−1
z2
µ ¶ µ µ
1 1
= 1− + O(z 4 ) 1 − z2 + O(z 2 )
2 z 6
z2
µ ¶µ µ ¶ ¶
1 1
= 1− + O(z 4 ) 1 + z 2 + O(z 2 ) + O(z 4 )
z 2 6
3.6. SOME MORE EXAMPLES 63

1 z
= − + O(z 3 )
z 3
Therefore we obtain the Laurent expansion of f at z = 0:
cot πz 1 π
f (z) = = − + O(z)
z 2 πz 3 3z
Hence we have Res( f , 0) = −π/3, which is the same as in the example.

Remark. We will give another example, which sums an alternating series. Instead of cot πz we use csc πz, for which we have the
similar lemmata:

Lemma 3.19

Suppose ϕ is holomorphic near n ∈ Z with ϕ(n) 6= 0. Then f (z) := ϕ(z)/ sin πz has a simple pole at z = n with residue Res( f , n) =
(−1)n ϕ(n)/π.

Lemma 3.20

There exists a C > 0 such that sup | csc πz| < C for all N ∈ N.
z∈Γ∗N

The proofs are very much similar (and in fact easier).

Example 3.21
X∞ (−1)n
Evaluate 2
.
n=1 n + 1

1
Solution. By Leibniz’s alternating series test the series converges. Let f (z) = . By Lemma 3.19, f has simple poles at
(z 2 + 1) sin πz
n
(−1)
every integer n ∈ Z with residues Res( f , n) = . In addition, f also has simple poles at z = ±i, where the residues are
π(1 + n 2 )
given by:
(z − i) 1 1
Res( f , i) = lim = =−
z→i (z 2 + 1) sin πz
2i sin(iπ) 2 sinh π
(z + i) 1 1
Res( f , −i) = lim 2 =− =−
z→−i (z + 1) sin πz 2i sin(−iπ) 2 sinh π
The integration of f along ΓN :

(−1)n ¯¯
¯I ¯ ¯I ¯
¯ ¯ ¯ C
¯ f (z) dz ¯¯ = ¯¯ dz É 4 · (2N + 1) · →0
¯
ΓN ΓN z2 + 1 ¯ (N + 1/2)2 + 1
as N → ∞.

Therefore by Residue Theorem:

I N
X
f (z) dz = Res( f , i) + Res( f , −i) + Res( f , n)
ΓN n=−N
N →∞ 1 1 X∞ 1 X∞ (−1)n π 1
====⇒ 0 = 2 · − + +2 =⇒ = −
2 sinh π π n=1 π(n 2 + 1)
n=1 n 2 +1 2 sinh π 2

3.6 Some More Examples

Example 3.22
Z +∞
x
Evaluate dx
−∞ sinh x
64 CHAPTER 3. CALCULUS OF RESIDUES

Solution. Let f (z) = z/ sinh z. This is a function with infinitely many poles, so we will not be able to choose a contour that encloses
all the poles. However, we can use the periodicity of the function. At z = 0, observe that
z z
lim = lim P =1
z→0 sinh z z→0 ∞ n=0 z 2n+1 /(2n + 1)!

Then z = 0 is a removable singularity of f . Notice the periodicity of sinh z: sinh(x) = sinh(x + iπ) for x ∈ R. sinh z has simple
zeros at iπZ. f has simple poles at iπZ\{0} with residues:

z(z − nπi) z
Res( f , nπi) = lim = nπi lim = nπi
z→nπi sinh z z→0 sinh z

We consider the rectangular contour with indentation at z = iπ, as shown in Figure 3.6.

Im
−R + iπ iπ R + iπ

γρ
γ− γ+

−R R Re

Figure 3.6: A rectangular contour with indentation at z = iπ.

We observe that f is holomorphic inside and along the above contour. Apply Cauchy’s Theorem:
R Z ρ
x z x + iπ
Z Z
dx + dz + dx
−R sinh x γ+ sinh z R sinh(x + iπ)
Z −R
z x + iπ z
Z Z
+ dz + dx + dz = 0
γρ sinh z −ρ sinh(x + iπ) γ− sinh z

First we inspect the vertical sides of the contour:

¯Z ¯ ¯Z π ¯ ¯Z π ¯
¯ z ¯ ¯ R + it ¯ ¯ R + it ¯
¯ dz ¯ = ¯
¯ ¯ idt ¯ = ¯
¯ ¯ idt ¯¯
γ+ sinh z 0 sinh(R + it ) 0 sinh R cos t + i cosh R sin t
¯
R +π
É π· →0
sinh R
as R → 0. And it is similar for γ− .

For the indentation near z = iπ, by Lemma 3.11, we have:

z
Z
lim dz = iπ Res( f , iπ) = −π2
ρ→0 γρ sinh z

Let R → ∞ and ρ → 0 in Cauchy’s Theorem. Substitute the preceding equations into the equation:
Z +∞ x
Z +∞ x + iπ
dx − π2 − dx = 0
−∞ sinh x −∞ − sinh x

Taking the real part, we have:

+∞ x π2
Z
dx =
−∞ sinh x 2

Example 3.23. Gaussian Integral.

Z +∞
2
Evaluate e−x dx.
−∞

2
Solution. Normally we may consider the function e−z . But this is entire and is not well-behaved as z → ∞. Here we define f (z) =
2
eiπz / sin πz and consider its integral on a parallelogram contour, as shown in Figure 3.7.
3.6. SOME MORE EXAMPLES 65

Im iπ iπ
− 12 + R e 4 1 +R e 4
2

γ−

γ2

π/4
−1 O 1 Re
γ1

γ+
iπ iπ
− 12 − R e 4 1 −R e 4
2

Figure 3.7: Contour for the Gaussian Integral.

The nominator of f is entire. By Lemma 3.19, we know that f has a simple pole at z = 0 with residue Res( f , 0) = 1/π. By
Residue Theorem, we have
Z Z Z Z
f (z) dz + f (z) dz + f (z) dz + f (z) dz = 2πi Res( f , 0) = 2i
γ1 γ− γ2 γ+

On γ1 , we have γ1 (t ) = 12 + t eiπ/4 , t ∈ [−R, R]. The integral:

2
eiπz R exp(iπ( 12 + t eiπ/4 )2 )
Z Z
dz = · eiπ/4 dt
γ1 sin πz −R sin(π( 12 + t eiπ/4 ))
R exp(iπ( 12 + it 2 + t eiπ/4 ))
Z
= dt
−R cos(πt eiπ/4 )
R iπ/4
−πt 2 exp(iπt e )
Z
=i e dt
−R cos(πt eiπ/4 )

Similarly on γ2 , we have
2
eiπz R exp(−iπt eiπ/4 )
Z Z
2
dz = i e−πt dt
γ2 sin πz −R cos(πt eiπ/4 )
Adding up:
2 2
eiπz eiπz R exp(iπt eiπ/4 ) + exp(−iπt eiπ/4 )
Z Z Z
2
dz + dz = i e−πt dt
γ1 sin πz γ2 sin πz −R cos(πt eiπ/4 )
R 2 cos(πt eiπ/4 )
Z Z R
2 2
=i e−πt dt = 2i e−πt dt
−R cos(πt eiπ/4 ) −R

On γ− , the integral:
2
¯Z ¯ ¯Z
¯ eiπz ¯ ¯ −1/2 exp(iπ(x + R eiπ/4 )2 ) ¯¯
dz ¯ = ¯ dx ¯¯
¯ ¯
¯ γ− sin πz ¯ ¯ 1/2 sin(π(x + R eiπ/4 ))
¯
¯Z p p ¯
¯ 1/2 eiπx(x+ 2R) e−πR(R+ 2x) ¯
= 2¯ p dx ¯
¯ ¯
p p p
¯ −1/2 eiπ(x+R/ 2) e−πR/ 2 + e−iπ(x+R/ 2) eπR/ 2 ¯
¯ p ¯
¯ e−πR(R+ 2x) ¯
É 2 sup ¯ p ¯→0
¯ ¯
p
πR/ 2 −πR/ 2
x∈[− 21 , 21 ] e −e
¯ ¯

as R → ∞. Similarly, the integral along γ+ also vanishes as R → ∞.

By letting R → ∞ in the Residue Theorem, we have:

Z +∞ 2
Z +∞ 2 p
2i e−πx dx = 2i =⇒ e−x dx = π
−∞ −∞
66 CHAPTER 3. CALCULUS OF RESIDUES

Example 3.24
Z 2π
dθ
Evaluate .
0 3 + cos θ + 2 sin θ

Z 2π
Remark. This is an example of integral with finite interval. More generally, consider integrals of the form R(sin θ, cos θ) dθ,
0
iθ
where the integrand is a rational function of sin θ and cos θ. We use the substitution z = e . Then the integral can be regarded as a
contour integral along the unit disk. Moreover, we have:

1 iθ 1
sin θ = (e − e−iθ ) = (z − z −1 )
2i 2i
1 1
cos θ = (eiθ + e−iθ ) = (z + z −1 )
2 2
iθ −1 iθ
dθ = (i e ) d(e ) = −iz −1 dz

So we can use the Residue Theorem to compute the integral.

Solution. Let γ(0, 1) be the positively oriented contour along the unit disk. We have:
Z 2π dθ
I
1 1
= dz
0 3 + cos θ + 2 sin θ γ(0,1) 3+ 1 −1 1 −1
2 (z + z ) + i (z − z )
iz
10 dz
I
=
i+2 γ(0,1) (5z + 1 + 2i)(z + 1 + 2i)

Let f be the integrand above. Then f has two simple poles: z = − 15 (1 + 2i) and z = −(1 + 2i). By Residue Theorem, we have:

dz 1
I
= 2πi Res( f , − (1 + 2i))
γ(0,1) (5z + 1 + 2i)(z + 1 + 2i) 5
1 πi
= 2πi lim =
z→− 5 (1+2i) 5(z + 1 + 2i)
1 2(1 + 2i)

Hence
2π dθ 10 πi
Z
= · =π
0 3 + cos θ + 2 sin θ i + 2 2(1 + 2i)

Example 3.25
Z 1
dx
Evaluate 2/3
.
−1 (1 + x) (1 − x)1/3

Im

γR

γ+ γρ 1

γρ 2 γ− R Re
−1 1

Figure 3.8: A dumbbell contour.

3.6. SOME MORE EXAMPLES 67

Solution. Let f (z) = (1−z)−1/3 (1+z)−2/3 . This is a multifunction with branch points at z = 1 and z = −1. We shift to polar coordinates
(see Example 0.23 for detail) and write z = 1 + r eiθ = −1 + s eiϕ . Then f (z) = r −1/3 s −2/3 e−i(θ+2ϕ)/3 . The admissible contours
are those which winds around both or none of the two branch points, so we can obtain a holomorphic branch of f by doing
the branch cut at [−1, 1]. We consider a "dumbbell-shaped" contour γd := γρ 1 ? γ− ? γρ 2 ? γ+ that encloses both poles, as
shown in Figure 3.8. Here γρ 1 := γ(1, ρ) and γρ 2 := γ(−1, ρ). Let γR := γ(0, R) be the positively oriented circular contour. Then
f is holomorphic in the interior of the cycle Γ := γR + γd . By Cauchy’s Theorem we have:
I Z Z Z Z
f (z) dz + f (z) dz + f (z) dz + f (z) dz + f (z) dz = 0
γR γρ 1 γρ 2 γ+ γ−

Notice that
z
lim z f (z) = eiπ/3 lim
z→∞ z→∞ (z − 1)1/3 (1 + z)2/3
¯ z ¯1/3 ¯ z ¯2/3 µ
i 2i
¶
= eiπ/3 lim ¯ ¯ exp − arg(z − 1) − arg(1 + z) + i arg(z)
¯ ¯ ¯ ¯
¯ ¯
z→∞ z − 1 1+z 3 3
µ ¶
i 2i
= eiπ/3 lim exp − arg(z) − arg(z) + i arg(z) = eiπ/3
z→∞ 3 3

On the circular contour γR , the calculation is very similar to the proof of the Indentation Lemma. Let g (z) = z f (z) − e−iπ/3 .
Then lim g (z) = 0. Hence:
z→∞

g (z) + e−iπ/3
I I
lim f (z) dz = lim dz
R→∞ γR R→∞ γR z
g (z) 1
I I
= lim dz + e−iπ/3 lim dz
R→∞ γR z R→∞ γR z

= 2πi eiπ/3

On the indented circular arcs γρ 1 and γρ 2 , since

z − 1 2/3
¶ µ
−iπ/3
lim (z − 1) f (z) = e lim =0
z→1 z→1 z + 1
¶1/3
z +1
µ
lim (z + 1) f (z) = lim =0
z→−1 z→−1 1 − z

By the Indentation Lemma, we have:

Z Z
lim f (z) dz = 0, lim f (z) dz = 0
ρ→0 γρ ρ→0 γρ
1 2

On the line segment γ+ , arg(1 − z) = 0, arg(z + 1) = 0, therefore

f (z) = (1 − x)−1/3 (1 + x)−2/3

On the line segment γ− , arg(1 − z) = −2π, arg(z + 1) = 0, therefore

f (z) = (e−2πi (1 − x))−1/3 (1 + x))−2/3 = e2πi/3 (1 − x)−1/3 (1 + x)−2/3

Letting R → ∞ and ρ → 0 in Cauchy’s Theorem, we have:

1 dx −1 e2πi/3 dx
Z Z
+ + 2πi eiπ/3 = 0
−1 (1 − x)1/3 (1 + x)2/3 1 (1 − x)1/3 (1 + x)2/3
1 dx −2πi eiπ/3 π 2π
Z
=⇒ = = =p
−1 (1 − x)1/3 (1 + x)2/3 1−e 2πi/3 sin(π/3) 3
Chapter 4

Conformal Mappings

4.1 Extended Complex Plane 68

4.1.1 Riemann Sphere. 68
4.1.2 Projective Line. 70
4.2 Conformal Equivalence and Möbius Transformations 72
4.2.1 Conformal Equivalence. 72
4.2.2 Möbius Transformations. 73
4.3 Examples of Conformal Mappings 76
4.3.1 Using Möbius Transformations. 77
4.3.2 Using other Elementary Functions. 80
4.4 Automorphism Groups* 82
4.4.1 Automorphisms of the Unit Disk. 82
4.4.2 Automorphisms of the Upper Half Plane. 84
4.4.3 Automorphisms of the Complex Plane. 84
4.4.4 Automorphisms of the Extended Complex Plane. 85
4.4.5 Automorphisms of an Annulus. 85
4.5 Schwarz Reflection Principle* 85
4.6 Riemann Mapping Theorem* 88
4.7 Boundary Correspondence* 91
4.8 Schwarz-Christoffel Mappings* 94
4.9 Harmonic Functions and Dirichlet Problem 99
4.9.1 Harmonic Functions. 99
4.9.2 Poisson Kernel. 100
4.9.3 Dirichlet Boundary Value Problems. 102

4.1 Extended Complex Plane

In the classification of singularities, we are motivated to process functions in the extended plane C∞ := C ∪ {∞} rather than in C.
This is an example of one-point compactification of C. To fully make sense of continuity and holomorphicity in C∞ , we shall use
two approaches:

1. Riemann Sphere S2 . We will identify points on S2 := {x ∈ R3 : kxk = 1} with points on C through the stereograhic projection.
The north pole on S2 will be identified as ∞.

2. Projective Line CP1 . We denote the set of all one-dimensional subspaces in C2 by CP1 . We shall indentify points z ∈ C with
the subspace 〈(z, 1)〉 ⊆ C2 . Then the point of infinity is identified as the subspace 〈(1, 0)〉.

4.1.1 Riemann Sphere.

We begin our discussion with the stereograhic projection.

68
4.1. EXTENDED COMPLEX PLANE 69

S2
N

P
C

Figure 4.1: Stereographic Projection.

Definition 4.1. Stereographic Projection.

Consider the unit sphere S2 := {x ∈ R3 : kxk = 1} in R3 . Let N (0, 0, 1) be the north pole of S 2 . We view the complex plane C as a
copy of R2 in R3 : {(x, y, 0) ∈ R3 : x, y ∈ R}. Given a point M ∈ S 2 \{N }, the line connecting N and M has a unique intersection
with C at P . The mapping M 7→ P is a bijection between S2 \{N } and C, and is called the stereographic projection.

Remark. By some simple calculations the stereographic projection is given explicitly by:
X +Y i
(X , Y , Z ) 7→
1− Z
The inverse mapping is given by:
1 1
2x, 2y, x 2 + y 2 − 1 = (2 Re z, 2 Im z, |z|2 − 1)
¡ ¢
x + iy 7→
x2 + y 2 + 1 |z|2 + 1

Definition 4.2. Riemann Sphere.

If we identify N with ∞ on the plane, then the stereographic projection becomes a bijection between S2 and C∞ . The sphere
S2 is called the Riemann sphere.

Lemma 4.3

As a subset R3 , the Riemann sphere S2 is naturally a metric space. It induces a metric on C∞ :

2|z − w| 2
d (z, w) = p p , d (z, ∞) = p
1 + |z|2 1 + |w|2 1 + |z|2

for any z, w ∈ C.

Proof. For convenience, we denote the image of z ∈ C under the inverse mapping of stereographic projection by S(z). Then from
the previous remark we have
1
S(z) = 2 (2Re(z), 2Im(z), |z|2 − 1)
|z| + 1
For z, w ∈ C,
p
d (z, w) := kS(z) − S(w)k = 2 − 2S(z) · S(w)
s
2
= 2− (2Re(z), 2Im(z), |z|2 − 1) · (2Re(w), 2Im(w), |w|2 − 1)
(|z| + 1)(|w|2 + 1)
2
s
8Re(zw) + 8Im(zw) + 2(|z|2 − 1)(|w|2 − 1)
= 2−
(|z|2 + 1)(|w|2 + 1)
s
4|z|2 + 4|w|2 − 4(zw + zw) 2|z − w|
= =p
(|z|2 + 1)(|w|2 + 1)
p
1 + |z|2 1 + |w|2
70 CHAPTER 4. CONFORMAL MAPPINGS

p
d (z, ∞) := kS(z) − (0, 0, 1)k = 2 − 2S(z) · (0, 0, 1)
s
|z|2 − 1 2
= 2−2 =p
1 + |z|2 1 + |z|2

Remark. We have made C∞ into a metric space, so that we can define continuity of functions on C̃. We can also see that S(z) →
N (0, 0, 1) as |z| → ∞, so it is legitimate to identify N with ∞. In particular, a meromorphic function f : U → C naturally extends to
a continuous function from U to C∞ .

Remark. The geometry of Riemann sphere nicely unites lines and circles on the complex plane, as shown in the following proposi-
tion:

Proposition 4.4

The mapping S : C → S2 induces the following bijective correspondence:

(i) Lines in C ⇔ circles in S2 containing N ;

(ii) Circles in C ⇔ circles in S2 not containing N .

Proof. This is purely calculation in analytic geometry. A circle on S2 is obtained by intersecting the sphere with a plane {(X , Y , Z ) ∈
R3 : a X + bY + c Z = d } where a 2 + b 2 + c 2 > d 2 . The plane contains N (0, 0, 1) if and only if c = d .

For z = x + iy ∈ C, S(z) lies in the above plane, if and only if:

2ax + 2b y + c(x 2 + y 2 − 1) = d (x 2 + y 2 + 1)
⇔(c − d )(x 2 + y 2 ) + 2ax + 2b y − (c + d ) = 0

We can see that this is the equation of a line for c = d . If c 6= d , the equation simplifies to

a ´2 b 2 a2 + b2 + c 2 − d 2
³ µ ¶
x+ + y+ = (4.1)
c −d c −d (c − d )2

The RHS>0. It is indeed an equation of a circle. Hence circles on S2 corresponds to circles and lines in C.

Conversely, for a line in C, its equation can certainly be expressed as 2ax + 2b y = c + d , which corresponds to a circle on S2
containing N . For a circle in C of the form (x + A)2 + (y + B )2 = C 2 , we put c − d = 1 and a = A, b = B, c = (C 2 − A 2 − B 2 + 1)/2.
Then the equation of the circle becomes Equation (4.1). So it corresponds to a circle on S2 not containing N .

Proposition 4.5

Suppose U ∈ C is a domain. Then U is simply-connected if and only if C∞ \U is connected.

Proof. "⇐=": Suppose C∞ \U is connected. Let γ be a closed path in U and V be the exterior of γ. Then V is connected and has ∞ as
its limit point. Since C∞ \U ⊆ C∞ \γ∗ is connected and contains ∞, we have C∞ \U ⊆ V ∪ {∞}. Therefore, we have I (γ, z) = 0
for all z ∈ C\U . Hence the interior of γ is contained in U . Since γ is arbitrary, by Proposition 1.38, U is simply-connected.

"=⇒": See the Appendix B.1 of [Stein].

4.1.2 Projective Line.

Definition 4.6. Projective Line.

The complex projective line CP1 is defined by the set of all one-dimensional subspaces of C2 . The subspace generated by
(z, w) ∈ C2 \{0} would be denoted by [z : w]. The coordinates z, w are called homogeneous coordinates, which are determined
up to simultaneous rescaling.

The mapping that identifies z ∈ C with [z : 1] ∈ CP1 is a bijection between C and CP1 \{[1 : 0]}. If we identify ∞ with [1 : 0], then
the mapping becomes a bijection between C∞ and CP1 .
4.1. EXTENDED COMPLEX PLANE 71

Lemma 4.7

C̃ (equipped with the mertic induced by S2 ) induces a metric on CP1 :

s
|〈u, v〉|2
d (L 1 , L 2 ) = 2 1−
kuk2 kvk2

for any u ∈ L 1 \{0} and v ∈ L 2 \{0}.

Proof. Just some simple calculations.

Definition 4.8. Differentiability on CP1 .

Now we wish to introduce a differential structure on CP1 .

Let U0 := CP1 \{[1 : 0]} and U∞ := CP1 \{[0 : 1]} be two subsets of CP1 . Then the mapping ι0 (z) := [z : 1] and ι∞ (z) := [1 : z] are
two embeddings from C into CP1 , whose images are exactly U0 and U∞ . We also have CP1 = U0 ∪U∞ . We say that V ∈ CP1 is
1
an open set, if both ι−1
0 (V ) and ι∞ (V ) are open sets in C. This makes CP a one-dimensional smooth complex manifold or a
−1
1
Riemann surface. More explicitly, (U0 , ι0 ) and (U∞ , ι∞ ) are two coordinate charts, and {(U0 , ι−1
−1 −1
0 ), (U∞ , ι∞ )} is an altas of CP
−1

(check that it meets all the definitions).

Suppose V ⊆ CP1 is open. Suppose f : V → CP1 is continuous. For L ∈ V and f (L) ∈ CP1 , we have L ∈ Uα and f (L) ∈ Uβ for
some α, β ∈ {0, ∞}. Now f induces a mapping f˜ : ι−1
α (V ) → C via the following diagram:

ι−1
α
V ∩Uα ι−1
α (V )

f f˜

Uβ ιβ C

We say that f is differentiable at L ∈ CP1 , if f˜ := ι−1

β
◦ f ◦ ια : ι−1
α (V ) → C is differentiable at ια (L) ∈ C. (One should check that
−1

this is well-defined and is independent of the choice of Uα and Uβ .)

If we identify C∞ with CP1 , then the above discussion defines holomorphic functions on C∞ .

Remark. The above definition uses the language of differential geometry. It is essentially saying that, a function f : C∞ → C is
differentiable at z 0 ∈ C∞ , if:

1. z 0 , f (z 0 ) 6= ∞: f is differentiable under the usual definition;

2. z 0 = ∞, f (z 0 ) 6= ∞: g (z) := f (1/z) is differentiable at 0;

3. z 0 6= ∞, f (z 0 ) = ∞: g (z) := 1/ f (z) is differentiable at z 0 ;

4. z 0 = f (z 0 ) = ∞: g (z) := 1/ f (1/z) is differentiable at 0.

Proposition 4.9

(i) f : C∞ → C is holomorphic (in the sense of Definition 4.8) if and only if f ◦ ι0 is constant;

(ii) Suppose f : C → C∞ is non-constant. Then f is holomorphic (in the sense of Definition 4.8) if and only if ι−1
0 ◦ f is
meromorphic;

(iii) Suppose f : C∞ → C∞ is non-constant. Then f is holomorphic (in the sense of Definition 4.8) if and only if ι−1
0 ◦ f ◦ ι0 is
a rational function.

Proof. (i): The backward argument is trivial. For the forward argument, since C∞ is compact and f is continuous, then f (C∞ ) is
bounded. By Liouville’s Theorem, f is constant.

(ii): The backward argument is trivial. For the forward argument, it suffices to show that if f (z 0 ) = ∞ = [1 : 0] then z 0 is a pole
of ι−1
0 ◦ f . By continuity of f at z 0 , there exists r > 0 such that f (z) 6= 0 = [0 : 1] for z ∈ B (z 0 , r ). Hence f (B (z 0 , r )\{0}) ⊆ U∞ ∩U0 .
Let f ∞ := ι−1
∞ ◦ f and f 0 := ι0 ◦ f . Then f ∞ (z) = 1/ f 0 (z) for z ∈ B (z 0 , r )\{z 0 } and f ∞ (z) = 0. Hence z 0 is a pole of f 0 .
−1
72 CHAPTER 4. CONFORMAL MAPPINGS

(iii): The backward argument is trivial. For the forward argument, since f is holomorphic, then f¯ := ι−1
0 ◦ f ◦ ι0 is a meromor-
phic function with ∞ as a pole or removable singularity. Since C∞ is compact and the singularities of f¯ are isolated, f¯ can
only have finitely many poles. By Mittag-Leffler Theorem, there exists a meromorphic function g with the same poles and
orders. Hence f¯ − g is holomorphic and bounded on C. By Liouville’s Theorem, f¯(z) − g (z) = c is constant. Moreover, g is
given by a finite sum of some rational functions. Hence f¯(z) = g (z) + c is also a rational function.

Remark. A meromorphic function is said to be transcendental, if it is not a rational function. As a corollary of the previous propo-
sition, a transcendental meromorphic function either has ∞ as an essential singularity, or as a limit of poles.

4.2 Conformal Equivalence and Möbius Transformations

In this chapter we will shift our focus from analysis to geometry. The main problem is that, given two open sets U and V in C, if
there exists a holomorphic bijection between these sets. We begin with the definition of conformal mappings, which are mappings
that preserve angles. Informally stated, suppose there are two rays γ1 and γ2 starting at z 0 . We can define the angle between the two
rays by the difference of their arguments at z 0 . We say that f is angle-preserving at z 0 , if f ◦ γ1 and f ◦ γ2 make the same angle at z 0
as γ1 and γ2 do. The formal definition is as follows:

4.2.1 Conformal Equivalence.

Definition 4.10. Angle Preservation.

Suppose that U ⊆ C is open and f : U → C is a function. For z 0 ∈ U , suppose z 0 has a deleted neighborhood B (z 0 , r )\{z 0 } such
that f (z 0 ) ∉ f (B (z 0 , r )\{z 0 }) (i.e. f is locally injective at z 0 ). We say that f preserves angle at z 0 , if the limit

f (z 0 + r eiθ ) − f (z 0 )
lim e−iθ ¯ ¯
r →0 ¯ f (z 0 + r eiθ ) − f (z 0 )¯

exists and is independent of θ.

Proposition 4.11

Suppose f : U ⊆ C is a function. For z 0 ∈ U ,

(i) If f is complex differentiable at z 0 and f 0 (z 0 ) 6= 0, then f preserves angle at z 0 ;

(ii) If f is real differentiable (as a mapping U → R2 ) at z 0 with d f z0 6= 0 and preserves angle at z 0 , then f is complex differen-
tiable at z 0 and f 0 (z 0 ) 6= 0.

Proof. (i): Suppose f 0 (z 0 ) = a. Then the limit

f (z 0 + r eiθ ) − f (z 0 )
lim =a
r →0 r eiθ
holds for any θ ∈ R.

Then the limit is given by

f (z 0 + r eiθ ) − f (z 0 ) r a
lim e−iθ ¯ ¯ = a lim ¯ ¯=
r →0 ¯ f (z 0 + r eiθ ) − f (z 0 )¯ r →0 ¯ f (z 0 + r eiθ ) − f (z 0 )¯ |a|
Hence f preserves angle at z 0 .

(ii): Since f is real differentiable at z 0 , we can write

f (z) = f (z 0 ) + λ(z − z 0 ) + η z − z 0 + |z| · O(z) λ, η ∈ C

¡ ¢

One can check that this is equivalent to real differentiability at z 0 .

Angle preservation at z 0 implies that:

f (z 0 + r eiθ ) − f (z 0 ) λr eiθ +ηr e−iθ λ + η e−2iθ

α = lim e−iθ ¯ ¯ = lim e−iθ ¯ ¯=¯
¯λr eiθ +ηr e−iθ ¯ ¯λ + η e−2iθ ¯
¯
r →0 ¯ f (z 0 + r eiθ ) − f (z 0 )¯ r →0

The value of α is independent of θ. We can infer that η = 0. Since d f 0 6= 0, we must have λ 6= 0. Hence

f (z) = f (z 0 ) + λ(z − z 0 ) + |z| · O(z)

4.2. CONFORMAL EQUIVALENCE AND MÖBIUS TRANSFORMATIONS 73

We conclude that f is complex differentiable at z 0 with f 0 (z 0 ) = λ 6= 0.

Remark. Under our definition, there exists angle-preserving mappings which are not holomorphic. For example, f (z) = z|z| pre-
serves angle at z = 0 and the differential at that point is a zero map, but clearly f is not holomorphic at z = 0.

Definition 4.12. Conformal Mapping.

We say that f is a conformal mapping, if f is holomorphic with non-vanishing derivatives.

Corollary 4.13

A conformal mapping preserves angle at every point in its domain.

Remark. If f : U → C is holomorphic and injective, then we must have f 0 (z) 6= 0 for all z ∈ U . Hence biholomorphisms (and their
inverses) are conformal mappings.

Definition 4.14. Conformal Equivalence.

Suppose U ,V ∈ C are two open sets. We say that U and V are conformally equivalent, if there exists a bijective conformal
mapping f : U → V .

4.2.2 Möbius Transformations.

Definition 4.15. Projective General Linear Group PGL(2, C).

The General Linear Group GL(2, C) is the group of all invertible linear operators in C2 with compositions. That is,
½µ ¶ ¾
a b
GL(2, C) := : a, b, c, d ∈ C; ad − bc 6= 0
c d

GL(2, C) acts naturally on C2 and hence on CP1 . The operators that fix all elements of CP1 are exactly all the scalar multiplica-
tions, Z(2, C), which is exactly the center of GL(2, C). The quotient group, GL(2, C)/Z (2, C), which is the group of all invertiable
projective transformations in CP1 , is called the projective general linear group and is denoted by PGL(2, C).

Remark. Now let us calculate the action of PGL(2, C) on CP1 . For

µ ¶
a b
f = ∈ GL(2, C)
c d

it induces f¯ ∈ PGL(2, C). For [z, 1] ∈ CP1 :

az + b
·µ ¶ µ ¶¸ · ¸
a b z
f¯([z, 1]) = = [az + b : c z + d ] = : 1 (z 6= −d /c)
c d 1 cz + d

(If z = −d /c, then f¯([z : 1]) = [1 : 0].)

For [1, 0] ∈ CP1 ,

f¯([1, 0]) = [a : c] = [a/c : 1]

If we identify CP1 with C∞ , then PGL(2, C) is isomorphic to a group of automorphisms on C∞ , namely the Möbius transformations.
74 CHAPTER 4. CONFORMAL MAPPINGS

Definition 4.16. Möbius Transformations.

For a, b, c, d ∈ C such that ad −bc 6= 0, the following mapping on C∞ are called the fractral linear transformations or the Möbius
transformations:
az + b a
z 7→ , ∞ 7→
cz + d c
All Möbius transformations forms a group under mapping compositions, which is denoted by Mob. From the previous remark
we have Mob ∼ = PGL(2, C).

Definition 4.17. Dilations, Translations, and Inversion

There are three special types of Möbius transformations:

• For a ∈ C\{0}, z 7→ az is called a dilation.

• For b ∈ C, z 7→ z + b is called a translation.

• z 7→ z −1 is called inversion.

Lemma 4.18

Any Möbius transformation is a composition of dilations, translations, and an inversion.

az + b
Proof. Suppose f (z) = is a Möbius transformation. If c 6= 0, we observe that
cz + d
az + b a ad
µ ¶
1
f (z) = = + b− = t 1 ◦ d 1 ◦ i ◦ t 2 ◦ d 2 (z)
cz + d c c cz + d

where
d 2 (z) = c z, t 2 (z) = z + d , i (z) = z −1 , d 1 (z) = (b − ad /c)z, t 1 (z) = z + a/c

If c = 0, we trivially have f (z) = a/d + (b/d )z.

Remark. The subgroup of Mob generated by translations and dilations is the group of C-affine transformations Aff(C) := { f (z) =
az + b : a 6= 0} of the complex plane. It is the automorphism group of C (see 4.44) and is the stablizer of ∞ in Mob.

Proposition 4.19

az + b
Möbius transformation f (z) = is a biholomorphism on C∞ .
cz + d

Proof. Trivial.

Now we shall explore some geometric properties of the Möbius transformations.

Proposition 4.20

Möbius transformations preserve circles in S2 (which are circles or lines on C by Proposition 4.4).

Proof. It suffices to prove that circles in S2 are preserved under dilations, translations, and inversion. The first two cases are trivial.
For the inversion mapping f (z) = 1/z = z̄/|z|2 , we apply the inverse stereographic mapping:

1
z 7→ (2 Re z, 2 Im z, |z|2 − 1);
|z|2 + 1
z̄
µ ¶
1 1 1 1 1
→
7 1
2 Re z, −2 Im z, − 1 = 2 (2 Re z, −2 Im z, 1 − |z|2 )
|z|2 2 + 1 |z| 2 |z| 2 |z| 2 |z| + 1
|z|

Hence f induces a transformation (x, y, z) 7→ (x, −y, −z) on S2 , which is a rotation about the x-axis by π and certainly pre-
serves circles in S2 .
4.2. CONFORMAL EQUIVALENCE AND MÖBIUS TRANSFORMATIONS 75

Proposition 4.21

Given two triples of distinct points z 1 , z 2 , z 3 and w 1 , w 2 , w 3 in C∞ , there exists a unique Möbius transformation f such that
f (z i ) = w i for i = 1, 2, 3.

Proof. Consider the mapping f : C∞ → C∞ defined by

(z − z 1 )(z 2 − z 3 )
f (z) =
(z − z 3 )(z 2 − z 1 )
It is a Möbius transformation and sends z 1 , z 2 , z 3 to 0, 1, ∞. Similarly, consider g : C∞ → C∞ defined by
(z − w 1 )(w 2 − w 3 )
g (z) =
(z − w 3 )(w 2 − w 1 )
which sends w 1 , w 2 , w 3 to 0, 1, ∞.

Hence g −1 ◦ f is a Möbius transformation that sends z 1 , z 2 , z 3 to w 1 , w 2 , w 3 .

For the uniqueness part, suppose h is another such mapping. Then g ◦ h ◦ f −1 is a Möbius transformation that sends 0, 1, ∞
to 0, 1, ∞. Suppose
az + b
g ◦ h ◦ f −1 (z) =
cz + d
We observe that
g ◦ h ◦ f −1 (0) = 0 =⇒ b = 0
g ◦ h ◦ f −1 (1) = 1 =⇒ a = (c + d )
g ◦ h ◦ f −1 (∞) =∞ =⇒ c = 0
Therefore we have a = d and g ◦ h ◦ f −1 = idC∞ =⇒ h = g −1 ◦ f .

Remark. The Möbius transformation f given in the previous proposition is very crucial. It motivates us to give the following
concept:

Definition 4.22. Cross Ratio.

For a quadruple of points z, z 1 , z 2 , z 3 ∈ C∞ , we define the cross ratio of it to be

(z − z 1 )(z 2 − z 3 )
(z, z 1 , z 2 , z 3 ) =
(z − z 3 )(z 2 − z 1 )

Proposition 4.23. Möbius transformation preserves cross ratio.

Suppose f is a Möbius transformation that maps z 1 , z 2 , z 3 , z 4 ∈ C∞ to w 1 , w 2 , w 3 , w 4 ∈ C∞ , then the cross ratio:

(z 1 , z 2 , z 3 , z 4 ) = (w 1 , w 2 , w 3 , w 4 )

Proof. The only non-trivial case is that z 2 , z 3 , z 4 are distinct. By Proposition 4.21, the unique mapping g −1 ◦ f maps z 2 , z 3 , z 4 to
w 2 , w 3 , w 4 . Since f (z) = (z, z 2 , z 3 , z 4 ), g (z) = (z, w 2 , w 3 , w 4 ), and g (w 1 ) = f (z 1 ), then we have
(z 1 , z 2 , z 3 , z 4 ) = (w 1 , w 2 , w 3 , w 4 )
as required.

Definition 4.24. Symmetric Points.

Suppose that C ⊆ C is a circle centered at a with radius r . We say that z, w ∈ C∞ is a pair of symmetric points with respect to C ,
if they lies on the same ray starting from a and satisfies

|z − a| · |w − a| = r 2

Suppose that L ⊆ C is a line. r . We say that z, w ∈ C∞ is a pair of symmetric points with respect to L if L is the perpendicular
bisector of z and w.
76 CHAPTER 4. CONFORMAL MAPPINGS

Lemma 4.25

z and w are symmetric with respect to a circle C ∈ C∞ (which is a circle or a line in C) if and only if the cross ratio

(w, z 1 , z 2 , z 3 ) = (z, z 1 , z 2 , z 3 )

for any z 1 , z 2 , z 3 ∈ C .

Proof. We have to do some explicit calculations.

First, suppose that C is a line in C that passes through a and makes angle θ with the real axis. We have |z − a| = |w − a| and
arg(z − a) + arg(w − a) = 2θ (draw a diagram and do some middle school geometry). Hence

w − a = |w − a| ei arg(w−a) = |z − a| ei(2θ−arg(z−a)) = (z − a) e2iθ

=⇒ w = a + (z̄ − ā) e2iθ

Compute the cross ratio:
(w − z 1 )(z 2 − z 3 ) (a + (z̄ − ā) e2iθ −z 1 )(z 2 − z 3 )
(w, z 1 , z 2 , z 3 ) = =
(w − z 3 )(z 2 − z 1 ) (a + (z̄ − ā) e2iθ −z 3 )(z 2 − z 1 )
Since z 1 , z 2 , z 3 ∈ C , we have z j = a + (z̄ j − ā) e2iθ for j = 1, 2, 3. Then

(a + (z̄ − ā) e2iθ −(a + (z̄ 1 − ā) e2iθ ))(a + (z̄ 2 − ā) e2iθ −(a + (z̄ 3 − ā) e2iθ ))
(w, z 1 , z 2 , z 3 ) =
(a + (z̄ − ā) e2iθ −(a + (z̄ 3 − ā) e2iθ ))(a + (z̄ 2 − ā) e2iθ −(a + (z̄ 1 − ā) e2iθ ))
(z̄ − z̄ 1 )(z̄ 2 − z̄ 3 )
= = (z, z 1 , z 2 , z 3 )
(z̄ − z̄ 3 )(z̄ 2 − z̄ 1 )

Second, suppose that C is a circle in C centered at a with radius r . We have |z − a| · |w − a| = r 2 and arg(z − a) = arg(w − a).
Hence
r2 r2
w − a = |w − a| ei arg(w−a) = ei arg(z−a) =
|z − a| (z − a)
r2
=⇒ w = a +
z̄ − ā
Compute the cross ratio:
r 2
(w − z 1 )(z 2 − z 3 ) (a + z̄−ā − z 1 )(z 2 − z 3 )
(w, z 1 , z 2 , z 3 ) = =
(w − z 3 )(z 2 − z 1 ) (a + r 2 − z 3 )(z 2 − z 1 )
z̄−ā
³ ´³ ´
1 1 1 1
z̄−ā − z̄ 1 −ā z̄ 2 −ā − z̄ 3 −ā (z̄ − z̄ 1 )(z̄ 2 − z̄ 3 )
=³ ´³ ´=
1 1 1 1 (z̄ − z̄ 3 )(z̄ 2 − z̄ 1 )
z̄−ā − z̄ 3 −ā z̄ 2 −ā − z̄ 1 −ā

= (z, z 1 , z 2 , z 3 )

Proposition 4.26. Möbius transformation preserves symmetric points.

Suppose f is a Möbius transformation. Suppose z 1 and z 2 are symmetric with respect to a circle C ∈ C∞ . Then f (z 1 ) and f (z 2 )
are symmetric with respect to f (C ).

Proof. This is immediate by Lemma 4.25 and Proposition 4.23.

In next section, we shall see how we utilize Möbius transformations and other elementary functions to build up common conformal
mappings.

4.3 Examples of Conformal Mappings

Definition 4.27. Upper Half Plane; Unit Disk.

We will frequently encounter these special subsets of C in this chapter. For notational convenience, we write H := {z ∈ C : Im z >
0} and D := B (0, 1) = {z ∈ C : |z| < 1}.
4.3. EXAMPLES OF CONFORMAL MAPPINGS 77

4.3.1 Using Möbius Transformations.

Example 4.28

Find a conformal mapping from H onto D.

Solution. We express H as follows:

¯ ¯
¯z −i¯
½ ¾
H = {z ∈ C : Im(z) > 0} = {z ∈ C : |z − i| < |z + i|} = z ∈ C : ¯
¯ ¯ <1
z +i¯
We immediately observe that the Möbius transformation
z −i
f (z) =
z +i
is a conformal mapping from H onto D. The inverse mapping from D onto H is given by

z −1
f −1 (z) = i
z +1

Remark. Some conformal mappings from other half planes to D are given by:
z −i
• From lower half plane to D: f (z) = ;
z +i
z +1
• From left half plane to D: f (z) = ;
z −1
z −1
• From right half plane to D: f (z) = (notice that this mapping is self-inverse).
z +1

Example 4.29

Find a conformal mapping from H onto D that maps a ∈ H to 0.

Solution. We can make use of Proposition 4.26 to find a Möbius transformation satisfying the conditions. Suppose f is such a
mapping. For a ∈ H, the symmetric point of a is ā. As f should maps the real axis to the unit circle, f (a) and f (ā) are
symmetric with respect to the unit circle. f (a) = 0 implies f (ā) = ∞. Hence we can choose
z −a
f (z) =
z − ā
It is easy to observe that | f (z)| É 1 and that f 0 (z) 6= 0. Then f is a conformal mapping. Moreover, f is unique up to composi-
tion of dilations (this is obvious after we prove Riemann Mapping Theorem).

Example 4.30

For |a| < 1, find the conformal mapping from D onto D that swaps a with 0.

Solution. We again use Proposition 4.26 to construct a Möbius transformation. Suppose ϕa is such a mapping. For a ∈ D, the
symmetric point of a is 1/ā. ϕa should fix the unit circle and maps a to 0. Hence ϕa (1/ā) = ∞. We write
a−z
ϕa (z) = λ
1 − āz
where λ ∈ C. Moreover, we have ϕa (0) = a. Hence a = λa =⇒ λ = 1. The mapping with desired properties is given by
a−z
ϕa (z) =
1 − āz

Remark. The mapping ϕa constructed in the previous example is an automorphism of the unit disk. In next section we shall prove
that any automorphism of the disk is a composition of such mappings and rotations. Moreover, ϕa is self-inverse, which means that
ϕ2a = idD or ϕa = ϕ−1
a .

Example 4.31

Find a conformal mapping from a cresent U := {z ∈ C : |z| < 2, |z − 1| > 1} onto a strip V := {w ∈ C : −1 < Re w < 0}.
78 CHAPTER 4. CONFORMAL MAPPINGS

Solution. We want a Möbius transformation that maps the circle |z − 1| = 1 to the imaginary axis and maps circle |z| = 2 to the line
Re w = −2. By geometric intuitions, this could be done by mapping 0, 2, −2 to 0, ∞, −2, as shown in Figure 4.2.
z 4z
Let f (z) = λ for some λ ∈ C. Since f (−2) = −2, we have λ = 4. The desired mapping is given by f (z) = .
z −2 z −2

Im Im

1+i

Re Re
f
O 1 2 −−−−−→ −2 O

Figure 4.2: Conformal mapping from a cresent area to a strip area.

Example 4.32
p p ª
Find a conformal mapping from a lozenge-shaped area z ∈ C : |z − i| < 2, |z + i| < 2 onto the first quadrant {z ∈ C : Re z >
©

0, Im z > 0}.

Solution. The lozenge area is bound by two arcs which intersects at −1 and 1. If we apply a Möbius transformation that maps −1 to
0 and 1 to ∞, then the two arcs will be mapped to two rays starting from 0, and the image of the lozenge would be a sector.
z +1
Let f (z) = λ for some λ ∈ C. Since Möbius transformation preserves angle at −1, we know that the angle of the sector is
z −1
exactly π/2. In order to map the lozenge onto the first quadrant, we may need a rotation, which is determined by λ ∈ ∂B (0, 1).
p
f maps i − 2 to a point on the real axis. Hence
à p !
p i− 2+1
Im f (i − 2) = 0 =⇒ Im λ p = 0 =⇒ Im (λ(i − 1)) = 0
i− 2−1
=⇒ λ = (i + 1)η for some η ∈ R

z +1
We choose η = 1. Then the desired mapping is given by f (z) = (i + 1) .
z −1

Im Im
i

−1 1 f
−−−−−→
Re Re

−i

Figure 4.3: Conformal mapping from a lozenge to a sector.

4.3. EXAMPLES OF CONFORMAL MAPPINGS 79

ea
n a−z o
eb Aut(D) = f (z) = eiθ : θ ∈ R, a ∈ D
1 − az

Annulus with a cut

f (z) = log z f (z) = exp z − π2 π

2
Vertical Semi-Infinite Strip
πi Unit Disk D
Affine
f (z) = sin z Transformations
Möbius
Transformations z −i
a b f (z) = f (z) = arcsin z
z +i
−πi z −1 Trigonometric
f (z) = i
z +1 Functions
πi
A Special Square Schwarz-
Christoffel
Mapping

Hyperbolic Horizontal Semi-Infinite Strip

Schwarz-Christoffel Functions
Mapping
Polygonal Area Upper Half Plane H

f (z) = exp z
µ ¶
1 1
az + b f (z) = − z +
½ ¾
Power
Aut(H) = f (z) = : a, b, c, d ∈ R, ad − bc = 1 Functions 2 z
cz + d
f (z) = z θ/π f (z) = log z
f (z) = z π/θ
Joukowski
Transformation

Crescent Area θ Upper Half Disk

Möbius
Transformations Infinite Sector
Power
f (z) = log z Functions Möbius Power
Transformations Functions
f (z) = exp z

θi

Horizontal Infinite Strip Quarter Plane

Circular Sector

Figure 4.4: Summary of Bihomomorphisms between Common Regions

80 CHAPTER 4. CONFORMAL MAPPINGS

4.3.2 Using other Elementary Functions.

Example 4.33. Power Functions.

Consider the power function f (z) = z n for n ∈ Z+ . f preserves angle everywhere except at z = 0, where angles are magnified
by a factor of n.

f maps the sector {z ∈ C : arg z ∈ (0, π/n)} to H, as shown in Figure 4.5. The inverse function f −1 (z) = z 1/n maps H to {z ∈ C :
arg z ∈ (0, π/n)} (we need to take a branch cut in this case).

Im Im

f : z 7→ z n
−−−−−−−−−−−−→
π/n

Re Re

Figure 4.5: Conformal mapping from a sector to H.

Example 4.34. Exponential Function.

The exponential function exp z is conformal everywhere. It maps the vertical line x = a to the circle |z| = ea , and maps the
horizontal line y = b to the ray arg z = c.

Therefore exp z maps the vertical strip {z ∈ C : Re z ∈ (a, b)} to the annulus A(0, ea , eb ) (this mapping is not injective!), and maps
the horizontal strip {z ∈ C : Im z ∈ (c, d )} to the sector {z ∈ C : arg z ∈ (c, d )}, as shown in Figure 4.6.

In reverse log z can maps a sector to a vertical strip, but again we should choose the branch cut carefully.

Im Im
eb

ea
z 7→ exp z
−−−−−−−−−−−−→
a b Re Re

Im d Im

c z 7→ exp z
−−−−−−−−−−−−→ d −c

Re Re

Figure 4.6: Conformal mapping of strips by exponential function.

Example 4.35. Trigonometric Functions.

The cosine function cos z = (eiz + e−iz )/2 is conformal everywhere. Consider the strip area U := {z ∈ C : Re z ∈ (0, π), Im z > 0}.
We would like to investigate how cos z maps the boundary lines. It is obvious that cos z maps (0, π) to (−1, 1) with orientation
reversed. For positive imaginary axis, notice that

cos(iy) = cosh y ∈ (1, ∞) for y > 0

4.3. EXAMPLES OF CONFORMAL MAPPINGS 81

Hence cos z maps the positive imaginary axis to (1, ∞). For the half-line {z ∈ C : Re z = π, Im z > 0}, notice that

cos(π + iy) = − cosh y ∈ (−∞, −1) for y > 0

Hence cos z maps the half-line to (−1, ∞). In conclusion, cos z maps ∂U to the entire real axis. However, it is still unclear
whether cos z maps U to the upper half plane or the lower half plane. So consider z = π/2 + iy for some y > 0. Then cos(π/2 +
iy) = − sin(iy) = −i sinh y has a negative imaginary part. Hence we conclude that cos z maps U to the lower half plane, as
shown in Figure 4.7.

Similarly, the hyperbolic function cosh z maps a horizontal semi-infinite strip to a half plane.

Im Im

Re −1 1 Re
O π z 7→ cos z
−−−−−−−−−−−−→

Im Im
z 7→ sin z
−−−−−−−−−−−→
Re −1 1 Re
−π/2 O π/2

Figure 4.7: Conformal mapping by trigonometric functions.

Example 4.36. Joukowsky Transformation.

The simplest form of Joukowsky Transformation is given by:

µ ¶
1 1
f (z) = z+
2 z
µ ¶
1 1
0
The derivative is given by f (z) = 1 − 2 . Hence the mapping is holomorphic on C\{0} and preserves angle everywhere
2 z
except at z = ±1. If we put w = f (z), then

w +1 z +1 2
µ ¶
2
2w z = z + 1 or =
w −1 z −1

z +1
We can regard the Joukowsky transformation as the composition g −1 ◦ h ◦ g , where g (z) := maps the left half plane to the
z −1
unit disk and h(z) := z 2 doubles the angles subtended at 0.

Suppose that z = r eiθ is mapped to w = u + iv. Then

1 1
u = (r + r −1 ) cos θ v = (r − r −1 ) sin θ
2 2
The circles {z ∈ C : |z| = ρ} (ρ > 0) are mapped to

u2 v2
1
+ 1
=1
4 (ρ + ρ ) 4 (ρ + ρ )
−1 2 −1 2

which are ellipses on the plane for ρ 6= 1. When ρ = 1, the Joukowsky transformation maps the unit circle to (−1, 1), as shown
in Figure 4.8. In particular, Joukowsky transformation maps the unit disk D to the "cut plane" C\[−1, 1], and maps the upper
unit disk {z ∈ C : |z| < 1, Im z > 0} to the lower half plane {z ∈ C : Im z < 0}.

The Joukowsky transformation is important in fluid dynamics as it maps certain circles to aerofoil shapes, as shown in Figure
4.8.
82 CHAPTER 4. CONFORMAL MAPPINGS

Im Im

z 7→ 12 z + 1z
¡ ¢
−1 1 Re −−−−−−−−−−−−−−−→ −1 1 Re

Im
Im

z 7→ 12 z + 1z
¡ ¢
−−−−−−−−−−−−−−−→

−1 Re
−1 Re

Figure 4.8: Joukowsky Transformation.

4.4 Automorphism Groups*

Definition 4.37. Automorphisms.

A conformal mapping from an open set U onto itself is called an automorphism on U . The set of all automorphisms on U
forms a group Aut(U ).

4.4.1 Automorphisms of the Unit Disk.

We begin with discussion of the automorphisms on the unit disk D.

Lemma 4.38. Schwarz’s Lemma.

Suppose that f : D → D is holomorphic with f (0) = 0. Then we have:

(i) | f (z)| É |z| for all z ∈ D;

(ii) | f 0 (0)| É 1;

(iii) if ∃ z 0 ∈ D\{0} | f (z 0 )| = |z 0 | or | f 0 (0)| = 1, then f is a rotation.

¡ ¢

That is, f (z) = eiθ z for some θ ∈ [0, 2π].

Proof. (i). Let g : D\{0} → D defined by g (z) = f (z)/z. Notice that z = 0 is a zero of f of multiplicity at least 1. So z = 0 is a removable
singularity of g . If we put g (0) = f 0 (0), then g is holomorphic on D. Fix r ∈ (0, 1), then by Corollary 2.11:

| f (z)| 1
sup |g (z)| = sup |g (z)| = sup É
z∈B (0,r ) z∈∂B (0,r ) |z|=r |z| r

We let r → 1 and obtain sup |g (z)| É 1. Hence | f (z)| É |z| for all z ∈ D.
z∈D

(ii). This is immediate as | f 0 (0)| = |g (0)| É 1.

(iii). First we assume that ∃ z 0 ∈ D\{0} | f (z 0 )| = |z 0 | . Then g attains maximum modulus 1 at some z 0 ∈ D. By Maximum
¡ ¢

Modulus Principle, g is constant. Hence f (z) = c z for some c ∈ C. Since |g (z 0 )| = |c| = 1, we can write c = eiθ and conclude
that f is a rotation.

Second, we assume that | f 0 (0)| = 1. This implies that g (0) = 1 and g attains maximum modulus 1 at z = 0. By the similar
argument, f is a rotation.
4.4. AUTOMORPHISM GROUPS* 83

Remark. We have shown that the Möbius transformation

a−z
ϕa (z) =
1 − āz
is an automorphism of D that swaps a ∈ D with 0. Schwarz’s Lemma can be generalized as follows:

Corollary 4.39. Schwarz-Pick Lemma.

Suppose that f : D → D is holomorphic with f (a) = b for some a, b ∈ D, then:

(i) |ϕb ◦ f (z)| É |ϕa (z)| for all z ∈ D;

1 − |b|2
(ii) | f 0 (a)| É ;
1 − |a|2
1 − |b|2
(iii) if ∃ z 0 ∈ D\{a} (|ϕb ◦ f (z 0 )| = |ϕa (z 0 )|) or | f 0 (a)| = , then f ∈ Aut(D).
1 − |a|2

Proof. Let g := ϕb ◦ f ◦ϕa . Then g is an automorphism of D and g (0) = ϕb ◦ f ◦ϕa (0) = ϕb ◦ f (a) = ϕb (b) = 0. We can apply Schwarz’s
Lemma to g .

(i). By Schwarz’s Lemma (i), |g (z)| É |z| =⇒ |ϕb ◦ f ◦ ϕa (z)| É |z| =⇒ |ϕb ◦ f (z)| É |ϕa (z)|.

(ii). By Schwarz’s Lemma (ii), |g 0 (0)| É 1, where

g 0 (0) = ϕ0b (b) · f 0 (a) · ϕ0a (0)

|b|2 − 1 |a|2 − 1
µ ¶ µ ¶
= f 0 (a) · ·
(1 − b̄z)2 z=b (1 − āz)2 z=0
1 − |a|2
= f 0 (a) ·
1 − |b|2

1 − |b|2
Hence | f 0 (a)| É .
1 − |a|2
1 − |b|2
(iii). If ∃ z 0 ∈ D\{a} (|ϕb ◦ f (z 0 )| = |ϕa (z 0 )|), then |g (z 0 )| = |z 0 |. If | f 0 (a)| =
, then |g 0 (0)| = 1. In either case we deduce
1 − |a|2
from Schwarz’s Lemma that g is an rotation. Hence f = ϕb ◦ g ◦ ϕa ∈ Aut(D).

Theorem 4.40. Automorphism Group of the Unit Disk.

n a−z o
Aut(D) = f (z) = eiθ : θ ∈ R, a ∈ D
1 − āz

Proof. Suppose that f is an automorphism of D. Then there exists a unique a ∈ D such that f (a) = 0. Let g := f ◦ ϕa . g is also an
automorphism of D. We have g (0) = f ◦ ϕa (0) = f (a) = 0 and g −1 (0) = 0. Apply Schwarz’s Lemma to g and g −1 :

∀z ∈ D : |g (z)| É |z|, |g −1 (z)| É |z|

=⇒∀ z ∈ D : |g (z)| É |z|, |z| É |g (z)|
=⇒∀ z ∈ D : |g (z)| = |z|

Hence by Schwarz’s Lemma (iii), g is a rotation. There exists θ ∈ [0, 2π] such that g (z) = eiθ z. Hence f (z) = g ◦ ϕ−1
a (z) =
iθ a − z
e as required.
1 − āz

Corollary 4.41

The only mappings in Aut(D) that fix the origin are the rotations.

Remark. We can see that Aut(D) acts transitively on D in the sense that for any a, b ∈ D, we have ϕa , ϕb ∈ Aut(D) and ϕb ◦ ϕa : a 7→ b.
84 CHAPTER 4. CONFORMAL MAPPINGS

4.4.2 Automorphisms of the Upper Half Plane.

We have proven in Example 4.28 and 4.29 that H and D are conformally equivalent. This leads to the group isomorphism Aut(D) ∼ =
Aut(H) via Γ : Aut(D) → Aut(H) defined by conjugation Γ(ψ) = f −1 ◦ ψ ◦ f , where f is any conformal mapping from H onto D. In the
next theorem, we shall give the elements of the group Aut(H) and show that it is isomorphic to PSL(2, R).

Definition 4.42. Projective Special Linear Group PSL(2, R).

The Special Linear Group SL(2, R) is the group of all linear operators in R2 with determinant 1. That is,
½µ ¶ ¾
a b
SL(2, R) := : a, b, c, d ∈ R; ad − bc = 1
c d

Just like PGL(2, C), PSL(2, R) is the group of all special projective transformations in RP1 :

PSL(2, R) = SL(2, R)/{I , −I }

Remark. We recall from Definition 4.16 that Mob ∼ = PGL(2, C). We should warn readers that, while PSL(2, C) = PGL(2, C), in fact
PSL(2, R) < PGL(2, R). This corresponds to RP1 being orientable, and PSL(2, R) only being the orientation-preserving transforma-
tions.

Theorem 4.43. Automorphism Group of the Upper Half Plane.

az + b
½ ¾
Aut(H) = f (z) = : a, b, c, d ∈ R, ad − bc = 1 ∼= PSL(2, R)
cz + d

Proof. Let SL(2, R) acts on H by

az + b
µ ¶
a b
M= 7 f M (z) =
→
c d cz + d
It is easy to check that f M (H) ⊆ H and that f M ◦ f N = f M N , f I = idH so that this is indeed a group action. In particular,
f M ∈ Aut(H) for every M ∈ SL(2, R) as it has a holomorphic inverse f M −1 .

First we shall prove that the group action is transitive. It suffices to prove that ∀ α ∈ H ∃ M ∈ SL(2, R) : f M (α) = i. Notice that

Re α −|α|2 α Re α − |α|2
µ ¶
M= ∈ SL(2, R), f M (α) = =i
Im α 0 α Im α

Next, for θ ∈ R, let M θ ∈ SL(2, R) defined by the rotation

cos θ − sin θ
µ ¶
M θ :=
sin θ cos θ

z −i
And let F := be a conformal mapping from H onto D. Then F ◦ f Mθ ◦ F −1 (z) = e−2iθ z is a rotation on D.
z +i
Finally, for any automorphism f ∈ Aut(H), suppose that f (i) = α. There exists N ∈ SL(2, R) such that f N (α) = i. Hence f N ◦
f (i) = i. Since F maps i to 0, F ◦ f N ◦ f ◦ F −1 ∈ Aut(D) fixes the origin. By Corollary 4.41, this is a rotation. Hence there exists
θ ∈ R such that F ◦ f N ◦ f ◦ F −1 = F ◦ f Mθ ◦ F −1 . Hence f = f N−1 ◦ f Mθ = f N −1 Mθ . There exists a, b, c, d ∈ R with ad − bc = 1 such
az + b
that f (z) = .
cz + d

4.4.3 Automorphisms of the Complex Plane.

Theorem 4.44. Automorphism Group of the Complex Plane.

Aut(C) = { f (z) = az + b : a, b ∈ C, a 6= 0} = Aff(C)

Proof. Suppose that f ∈ Aut(C). We first show that ∞ cannot be an essential singularity of f . Suppose for contradiction that ∞ is
an essential singularity. Let g (z) := f (1/z). Then g has an essential singularity at z = 0. By Casorati-Weierstrass Theorem,
g (D\{0}) is dense in C. Notice that the inversion z 7→ 1/z maps C\B (0, 1) to B (0, 1)\{0} = D\{0}. Therefore f (C\B (0, 1)) is dense
4.5. SCHWARZ REFLECTION PRINCIPLE* 85

in C. Since f is continuous and bijective, it maps the closure of C\B (0, 1) to the closure of f (C\B (0, 1)) 1 , which implies that
f (C\B (0, 1)) = C. Hence f cannot be injective, contradicting that f is an automorphism.

We know that ∞ is a removable singularity or a pole of f . By Proposition 4.9 (iii), f is a rational function. Moreover f must
be a polynomial since it is holomorphic on C. By injectivity of f let f −1 (0) = b ∈ C. Then f (z) = a(z −b)n where n = deg f and
a ∈ C. If n Ê 2, then f cannot be injective, as f (b + 1) = f (b + e2πi/n ) = a. Hence f = az + b for some a, b ∈ C.

Remark. Notice that we define Aut(C) to be the group of all biholomorphisms on C, It should not be confused with the group of all
(continuous) field automorphisms of C, which appears in field theory and sometimes shares the same notation.

4.4.4 Automorphisms of the Extended Complex Plane.

Theorem 4.45. Automorphism Group of the Extended Complex Plane.

az + b
Aut(C∞ ) = { f (z) = : a, b, c, d ∈ C, ad − bc 6= 0} = Mob
cz + d

Proof. Suppose that f ∈ Aut(C∞ ) and that f (∞) = ∞. Since f is bijective we have f |C ∈ Aut(C) (more rigorously f ◦ ι0 ∈ Aut(C)). By
Theorem 4.44 we have f (z) = az + b for some a, b ∈ C.

Suppose that f ∈ Aut(C∞ ) and that f (∞) = z 0 ∈ C. Then g (z) := 1/( f (z)− z 0 ) satisfies g (∞) = ∞. Hence g (z) = c z +d for some
c, d ∈ C.
1 1 az + b
cz + d = =⇒ f (z) = z 0 + =
f (z) − z 0 cz + d cz + d
where a := c z 0 and b := d z 0 + 1.

Remark. In this section we present four examples of automorphism groups: Aut(D), Aut(H), Aut(C) and Aut(C∞ ). We have Aut(D) = ∼
Aut(H), but they are not isomorphic to Aut(C) or Aut(C∞ ). They are all subgroups of Mob, the group of all Möbius Transformations.
The Uniformization Theorem 4.55 tell us that these groups can completely discribe any automorphism group of simply-connected
domain in C∞ up to isomorphism.

4.4.5 Automorphisms of an Annulus.

Theorem 4.46. Automorphism Group of Annulus.

Suppose that A = A(0, r, R) is an annulus, where 0 < r < R < ∞. Then we have

Rr
½ ¾
Aut(A) = f (z) = eiθ z or eiθ : θ∈R
z

We will present the proof in next section, after we prove the generalized Schwarz Reflection Principle.

4.5 Schwarz Reflection Principle*

In this section we will investigate holomorphic extension of functions and its application in constructing conformal mappings. The
key theorem is Schwarz Reflection Theorem and its generalization.

Lemma 4.47. Painlevé’s Theorem.

Suppose that U ⊆ C is a domain and γ : [0, 1] → C is a piecewise smooth path. Suppose that f : U → C is continuous on U and
holomorphic on U \γ∗ . Then f is holomorphic in the whole U .

Sketch of Proof. It suffices to prove that f is holomorphic in any open disk B (z 0 , r ) ⊆ U . By Morera’s Theorem, it suffices to prove
that Cauchy’s Theorem holds for any triangle T ⊆ B (z 0 , r ):
I
f (z) dz = 0
T
1 We know that if f : X → Y is a continuous mapping between topological spaces, and Z ⊆ X , then f (Z ) ⊆ f (Z ). Moreover, if f is bijective, then

Z = f −1 ( f (Z )) =⇒ f −1 ( f (Z )) ⊆ Z =⇒ f (Z ) ⊆ f (Z )
Hence we have f (Z ) = f (Z ).
86 CHAPTER 4. CONFORMAL MAPPINGS

γ∗ divides T into finitely many parts, each part is enclosed by a closed curve with a part of it on γ∗ . For each closed curve we
choose the positive orientation
I to integrate so that the integral on γ∗ cancels out. Apply Cauchy’s Theorem to each closed
path and we obtain that f (z) dz = 0 as required.
T

Theorem 4.48. Schwarz Reflection Principle.

Suppose that U ⊆ H is a domain with I = (a, b) ∈ R ∩ ∂U . Suppose that f : U ∪ I → C satisfies

1. f is continuous on U ∪ I ;

2. f is holomorphic in U ;

3. f is real-valued on I .

Then f can be holomorphically extended to U 0 ⊆ C, a domain symmetric to U with respect to the real axis, such that f (z) = f (z̄)
on U 0 .

Proof. We extend f to U 0 by defining f (z) = f (z̄) on U 0 . For x 0 ∈ I , the limit from the lower part:

lim f (z) = lim f (z̄) = f (x 0 ) = f (x 0 )

U 0 3z→x 0 U 3z→x 0

Hence f is continuous on U ∪ I ∪U 0 .

To prove that f is holomorphic in U 0 , for z, z 0 ∈ U 0 , we have z̄, z̄ 0 ∈ U . The Taylor expansion of f near z̄ 0 is given by f (z̄) =
a n (z̄ − z̄ 0 )n , which implies that f (z) = ā n (z − z 0 )n . Hence f is analytic in U ∪U 0 .
P P

Now by Painlevé’s Theorem, f is holomorphic on U ∪ I ∪U 0 .

We can replace R by any circle in C∞ . Recall that symmetric in Definition 4.24 we give the definition of a pair of symmetric points
with repsect to a circle or line. We say that U and V are symmetric with respect to the circle or line, if V contains all the symmetric
points of points in U and vise versa.

Theorem 4.49. Schwarz Reflection Principle for a Circle.

Suppose that γ(z 0 , r ) is a path whose image γ∗ = ∂B (z 0 , r ) is a circle in C. We denote the two connected components of C\γ∗
by Ω+ and Ω− . Suppose that U ⊆ Ω+ is a domain such that I := ∂U ∩ γ∗ is non-empty. Suppose that f : U ∪ I → C satisfies

1. f is continuous on U ∪ I ;

2. f is holomorphic in U ;

3. f (I ) ⊆ Γ∗ := ∂B (w 0 , ρ);

4. w 0 ∉ f (U ).

Then f can be holomorphically extended to U 0 ⊆ Ω− , a domain symmetric to U with respect to γ∗ , such that f maps a pair of
symmetric points with respect to γ∗ to a pair of symmetric points with respect to Γ∗ .

r2 ρ2
Proof. By Lemma 4.25, we know that z is symmetric to z 0 + with respect to B (z 0 , r ) and w is symmetric to w 0 + with
z̄ − z̄ 0 w̄ − w̄ 0
respect to B (w 0 , ρ). We define f on U 0 by
ρ2
f (z) = w 0 + µ
r2
¶
f z0 + − w̄ 0
z̄ − z̄ 0
f is well defined on U 0 as w 0 ∉ f (U ), which suggests that the denominator is never zero. Let w = f (z). Then we have

ρ2 r2
µ ¶
w0 + = f z0 +
w̄ − w̄ 0 z̄ − z̄ 0

ρ2 r2
Let φρ (w) := w 0 + and φr (z) := z 0 + . Then φρ ◦ f (z) = φρ (w) = f ◦ φr (z). Notice that φr and φρ are self-inverse.
w̄ − w̄ 0 z̄ − z̄ 0
For each open disk B (z 1 , δ) ⊆ U , we can expand f (z) into power series near φr (z 1 ) ∈ U :
0

µ 2 ¶n ¶n
∞ ∞ r r2 ∞ z̄ 1 − z̄
µ
f ◦ φr (z) = a n (φr (z) − φr (z 1 ))n = a n r 2n
X X X
an + =
n=0 n=0 z̄ − z̄ 0 z̄ 1 − z̄ 0 n=0 (z̄ − z̄ 0 )(z̄ 1 − z̄ 0 )
4.6. RIEMANN MAPPING THEOREM* 87

Hence
∞ ¶n ¶
z̄ 1 − z̄
µ µ
f (z) = φρ ◦ f ◦ φr (z) = φρ a n r 2n
X
n=0 (z̄ − z̄ 0 )(z̄ 1 − z̄ 0 )
2
ρ
= w0 + ∞ n
X (z 1 − z)
ā n r 2n − w̄ 0
n=0 (z − z 0 ) (z 1 − z 0 )n
n

Then f is holomorphic in U 0 .

Next we prove that f is continuous on I . Fix any ζ ∈ I . For z ∈ U 0 , φr (z) ∈ U . As z → ζ, we also have φr (z) → ζ. Since f is
continuous on U ∪ I , we have f ◦ φr (z) → f (ζ) ∈ Γ∗ . Then f (z) = φρ ◦ f ◦ φr (z) → f (ζ). Hence f is continuous at ζ ∈ I .

By Painlevé’s Theorem, f is holomorphic on U ∪ I ∪U 0 .

Remark. The theorem still holds if we replace γ∗ and Γ∗ by circles in C∞ (circles or lines in C). The proof is very similar and we are
not going to do it here.

Theorem 4.50. Conformal equivalence classes of annuli.

Let A(0, r, R) = {z : r < |z| < R} be an annulus with the smaller radius r and the larger radius R. In the case 0 < r i and R i < ∞,
A (0, r 1 , R 1 ) and A (0, r 2 , R 2 ) are conformally equivalent if and only if R 1 /r 1 = R 2 /r 2 .

For the degenerate cases, the annulus A(0, 0, ∞) = C\{0} is not conformally equivalent to any other annulus. The annuli
A(0, 0, R) and A(0, r, ∞) with r > 0 and R < ∞ are conformally equivalent to each other and not equivalent to any other annuli.

R2
Proof. We first consider the non-degenerate case. Suppose that R 1 /r 1 = R 2 /r 2 . Then f (z) = z maps A(0, r 1 , R 1 ) bijectively to
R1
A(0, r 2 , R 2 ). Conversely, suppose that f : A(0, r 1 , R 1 ) → A(0, r 2 , R 2 ) is a biholomorphism. We claim that f extends to a homeo-
morphism from A(0, r 1 , R 1 ) to A(0, r 2 , R 2 ). For the boundary behavior of biholomorphisms, we need to use the tools discussed
in Section 4.7. Since the boundary of A(0, r 1 , R 1 ) is the union of two disjoint simple closed paths, with some minor adaptation
of Proposition 4.62 and Proposition 4.63 we can prove that the claim is true. In particular, f maps ∂A(0, r 1 , R 1 ) continuously
and bijectively to ∂A(0, r 2 , R 2 ).

Suppose that f maps ∂B (0, r 1 ) to ∂B (0, r 2 ) and maps ∂B (0, R 1 ) to ∂B (0, R 2 ). By Schwarz Reflection Theorem for a circle, we can
"reflect" A(0, r!1 , R 1 ) across ∂B (0, r 1 ). Since R 1 is symmetric to !r 12 /R 1Ãwith respect
! to ∂B (0, r 1 ), we extend f holomorphically
2 2
r 22
à Ã
r1 r1
to A 0, , R 1 . Now f is a biholomorphism from A 0, , R 1 to A 0, , R 2 . We repeat the reflection. After n times, f is
R1 R1 R2
µ ¶2n µ ¶2n
r1 r2
µ ¶ µ ¶
a biholomorphism from A 0, R 1 , R 1 to A 0, R 1 , R 2 . Let n → ∞. Then f is biholomorphism from A(0, 0, R 1 )
R1 R2
to A(0, 0, R 2 ). In particular f maps 0 to 0. Then f is a biholomorphism from B (0, R 1 ) to B (0, R 2 ). Let φ1 (z) := z/R 1 and
R2
φ2 (z) := R 2 z. Then φ2 ◦ f ◦ φ1 ∈ Aut(D). By Theorem 4.40, there exists θ ∈ R such that f (z) = eiθ z. Hence we must have
R1
R 1 /r 1 = R 2 /r 2 .

Suppose that f maps ∂B (0, r 2 ) to ∂B (0, r 1 ) and maps ∂B (0, R 2 ) to ∂B (0, R 1 ). Then g (z) := R 2 r 2 / f (z) is a biholomorphism that
maps ∂B (0, r 1 ) to ∂B (0, r 2 ). We return to the previous case and conclude that R 1 /r 1 = R 2 /r 2 .
r2 R2
Now we consider the degenerate case. f (z) = z is a biholomorphism from A(0, r 1 , ∞) to A(0, r 2 , ∞); f (z) = z is a bi-
r1 R1
Rr
holomorphism from A(0, 0, R 1 ) to A(0, 0, R 2 ); f (z) = is a biholomorphism from A(0, 0, R) to A(0, r, ∞). The remaining
z
parts are trivial if we again invoke the boundary correspondence. Suppose that A(0, 0, R 1 ) are conformally equivalent to
A(0, r 2 , R 2 ) via biholomorphism f , where r 2 , R 1 , R 2 ∈ (0, ∞). Then f is a homeomorphism from ∂A(0, 0, R 1 ) = {0} ∪ ∂B (0, R 1 )
tp ∂A(0, r 2 , R 2 ) = ∂B (0, r 2 ) ∪ ∂B (0, R 2 ). Such f cannot be bijective, contradiction. Similarly we can prove that A(0, 0, ∞) is not
conformally equivalent to any other annuli.

Proof of Theorem 4.46. Suppose that f ∈ Aut(A). In the proof of Theorem 4.46, we have already shown that there are only two cases
Rr
about how f maps the boundary of A. They corresponds to f (z) = eiθ z and f (z) = eiθ respectively.
z
88 CHAPTER 4. CONFORMAL MAPPINGS

4.6 Riemann Mapping Theorem*

Theorem 4.51. Riemann Mapping Theorem.

Suppose that U ( C is a simply-connected domain and z 0 ∈ U . Then there exists a unique bijective holomorphic function f
which maps U onto the unit disk D such that f (z 0 ) = 0 and f 0 (z 0 ) > 0.

Remark. Riemann Mapping Theorem in fact implies that any two proper simply-connected domains in C are conformally equiv-
alent. This is a very profound result which establishes the connection between topological properties and holomorphic properties.
We regard this theorem as the cornerstone of the theory of simple-variable complex analysis.

Before proving Riemann Mapping Theorem we shall first introduce normal families and Montel’s Theorem.

Definition 4.52. Normal Families.

Suppose U ⊆ C is an open set. A set F of functions U → C is called a normal family, if any sequence in F has a subsequence
that converges uniformly on every compact subset of U (the limit is not necessarily in F ). The convergence is called normal
convergence.

Theorem 4.53. Montel’s Theorem.

Suppose that F is a set of holomorphic functions on U such that F is uniformly bounded on any compact subset of U , then
F is a normal family.

Proof. Since F is uniformly bounded, ther exists M > 0 such that | f (z)| É M for all z ∈ U and f ∈ F . Now we define a sequence of
compact subsets of U : ½ ¾ ½ ¾
1 1
K n := z ∈ U : dist(z, C\U ) Ê = z ∈ U : inf |z − w| Ê
n w∈C\U n
∞
It follows easily from definition that K n ⊆ K n+1 for n ∈ Z+ and that
[
K n = U . We say that {K n } is an exhaustion of U .
n=1

First we shall prove that F is equicontinuous. For each compact subset K n ∈ U , let r := 3 · dist(K n−1 , K n ) > 0. For z, w ∈ K n
such that |z − w| < r , we consider the circular path γ(z, 2r ) which lies entirely in K n+1 and apply Cauchy’s Intgeral Formula
to any given f ∈ F : µ ¶
1 1
I
f (z) − f (w) = f (ζ) − dζ
γ(z,2r ) ζ−z ζ−w
Hence
µ ¶¯
¯ 1 1
| f (z) − f (w)| É 2πr sup ¯ f (ζ) −
¯ ¯
ζ−z ζ−w
¯
ζ∈∂B (z,2r )
|z − w|
É 2πr M sup
ζ∈∂B (z,2r ) |ζ − z| |ζ − w|
|z − w|
É 2πr M = 2πM r −1 |z − w|
2r · r
Let L := 2πM r −1 be a constant which only depends on K . Then | f (z) − f (w)| É L|z − w| for all f ∈ F and all z, w ∈ K n that is
sufficiently closed. This is a uniform Lipschitz property of F and it implies that F is equicontinuous on K n .

Next, suppose { f n } ⊆ F is a sequence. By Arzelà-Ascoli Theorem 0.75, { f n } has a subsequence { f s(1,n) } that converges uni-
formly on K 1 (without loss of generality we may assume that K 1 6= ∅), where n 7→ s(1, n) is an injective increasing function on
Z+ .

By the same argument { f s(1,n) } has a subsequence { f s(2,n) } that converges uniformly on K 2 . Inductively we obtain a sequence
of subsequences { f s( j ,n) }. By the diagonal argument, { f s(n,n) } is a subsequence of { f n } which converges uniformly on every K n .
We further notice that any compact subset of U lies entirely in some K n . Hence we conclude that { f n } converges normally.
F is a normal family.

Remark. Montel’s Theorem guarantees the existence of the limit function, but it says nothing about the behavior of the limit func-
tion other than it is holomorphic. The following theorem gives a particular case which we can have some dichotomy information
about the limit function.
4.6. RIEMANN MAPPING THEOREM* 89

Theorem 4.54. Hurwitz’s Theorem.

Suppose that { f n } is a sequence of injective holomorphic functions on U . If f n converges normally to f on U , then f is either
injective or constant.

Proof. Suppose that f is non-constant. We argue for contradiction and suppose that f is not injective. There exists z 1 , z 2 ∈ U such
that f (z 1 ) = f (z 2 ). Let g n (z) := f n (z) − f n (z 2 ) and g (z) = f (z) − f (z 2 ). Since g is holomorphic, by identity theorem the roots
of g are isolated. There exists r > 0 such that g is non-zero in the deleted closed disk B (z 1 , r )\{z 1 }. We know that f n → f
normally on U . Therefore g n → g uniformly on B (z 1 , r ). Let

ε= inf |g (z)| > 0

z∈∂B (z 1 ,r )

By uniformly convergence, sup |g n (z) − g (z)| < ε for sufficiently large n.

z∈∂B (z 1 ,r )
It implies that |g (z)| > |g n (z) − g (z)| for all z ∈ ∂B (z 1 , r ). By Rouché’s Theorem, g (z) and g n (z) has the same number of zeros
in B (z 1 , r ). However, f n is injective and g n has no zeros in B (z 1 , r ), whereas g has one zero g (z 1 ) = 0 in B (z 1 , r ). This is a
contradiction. Hence f is injective.

Now we can proceed to the proof of Riemann Mapping Theorem.

Proof of Riemann Mapping Theorem.

Step 1: U is conformally equivalent to a bounded simply-connected domain.

Since U 6= C, we pick α ∈ C\U . Since U is simply-connected, by Proposition 1.38 (vii) we can define a holomorphic function
f (z) = log(z − α) on U , which has all the desired properties of logarithm. f is injective as exp ◦ f (z) = z − α. Fix any w ∈ U .
We claim that f (z) 6= f (w) + 2πi for all z ∈ U . If this is not the case, then f (z) = f (w) for some z ∈ U . Then

z − α = exp ◦ f (z) = exp( f (w) + 2πi) = w − α =⇒ z = w =⇒ f (z) = f (w)

which is a contradiction.

In fact, f (w) + 2πi ∉ f (U ). Suppose that there exists {z n } ⊆ U such that lim f (z n ) = f (w) + 2πi, Then lim z n = w as the
n→∞ n→∞
exponential function is continuous. But this implies that lim f (z n ) = f (w), which is a contradiction. Hence we can define
n→∞
F : U → C by
1
F (z) =
f (z) − ( f (w) + 2πi)
The above discussion suggests that F (U ) is bounded. Moreover, F is injective holomorphic. Hence U and F (U ) are confor-
mally equivalent. We have hence proven that any simply-connected domain is conformally equivalent to some bounded
simply-connected domain. So from now on we may assume that U is bounded.

We consider the following set of holomorphic function:

F = f : U → D ¯ f is injective holomorphic, f (z 0 ) = 0
© ¯ ª

Step 2: F is non-empty.

Since U is bounded, there exists R > 0 such that |g (z)| É R. We consider the mapping g (z) = (z − z 0 )/2R. Clearly f is injective
holomorphic; g (z 0 ) = 0; and |g (z)| = |z − z 0 |/2R É 1. Hence g ∈ F .

Step 3: F is a normal family, because it is uniformly bounded by 1 and we can use Montel’s Theorem.

Step 4: The maximizier of the functional f 7→ | f 0 (z 0 )| lies in F .

Since U is open and z 0 ∈ U , there exists r > 0 such that B (z 0 , r ) ⊆ U . For any f ∈ F , by Cauchy’s Inequality 1.22 we
have
1 1
| f 0 (z 0 )| É sup | f (z)| É
r z∈∂B (z0 ,r ) r
Hence M := sup | f 0 (z 0 )| < +∞. We can choose a sequence { f n } ⊆ F such that lim | f 0 (z 0 )| = M . Since F is a normal family,
f ∈F n→∞

{ f n } has a subsequence { f nk } that converges uniformly to f 0 on every compact subset of U . Then | f 00 (z 0 )| = M . By Hurwitz’s
Theorem, f 0 is either injective or constant. If f 0 is constant, then | f 00 (z 0 )| = M = 0. However, g (z) = (z − z 0 )/2R ∈ F and
|g 0 (z 0 )| = 1/2R > 0, which implies that M Ê 1/2R > 0. This is a contradiction. Hence f 0 is injective holomorphic. We have
f0 ∈ F .

Step 5: The maximizier f 0 is surjective.

90 CHAPTER 4. CONFORMAL MAPPINGS

Suppose for contradiction that f 0 (U ) 6= D. We shall construct explicitly h ∈ F such that |h 0 (z 0 )| > | f 0 (z 0 )|. We pick u ∈
D\ f 0 (U ). The following Möbius transformation maps u to 0:
z −u
ϕu (z) =
1 − ūz
Notice that ϕu ◦ f 0 (U ) is simply-connected and does not contain 0, we can define a holomorphic branch of the square root
function η(z) = z 1/2 on ϕu ◦ f 0 (U ).

Let v = η ◦ ϕu ◦ f 0 (z 0 ) and h = ϕv ◦ η ◦ ϕu ◦ f 0 . We observe that h is a compositon of

injective holomorphic functions and |h(z)| É 1. Moreover, h(z 0 ) = ϕv (v) = 0. Hence h ∈ F .
2
We write f 0 = ϕ−1
u ◦ λ ◦ ϕv ◦ h where λ(z) := z . Let Φ = ϕu ◦ λ ◦ ϕv . Notice that Φ maps D into D and that Φ(v) = 0. By
−1 −1 −1

Schwarz’s Lemma, |Φ (0)| < 1 because Φ is not a rotation. Hence

0

| f 00 (z 0 )| = |Φ0 (0) · h 0 (z 0 )| < |h 0 (z 0 )|

which contradicts that | f 00 (z 0 )| = sup | f 0 (z 0 )|. Hence f 0 is surjective.

f ∈F

We have obtained a bijective holomorphic function f 0 : U → D. To adjust the derivative at z 0 , we put

| f 00 (z 0 )|
f˜0 (z) = f 0 (z)
f 00 (z 0 )

so that f˜00 (z 0 ) = | f 00 (z 0 )| > 0. f˜0 is a conformal mapping U → D such that f˜0 (z 0 ) = 0 and f˜00 (z 0 ) > 0.

Step 6: f˜0 is unique.

Suppose that f˜1 is another function that satisfies the desired properties. Then ψ := f˜0 ◦ f˜1−1 is an automorphism of D. By
Theorem 4.40, there exists a ∈ D and θ ∈ [0, 2π] such that
z −a
ψ(z) = eiθ
1 − āz
Since ψ(0) = 0 and ψ0 (0) > 0, we must have θ = 0 and a = 0. Hence ψ = id and f˜0 = f˜1 as required.

Remark. The only property of simply-connected domain we use is the existence of holomorphic logarithm (existence of holo-
morphic square root follows directly). Thus we have finished the proof of Proposition 1.38 (vii)=⇒(i) by proving that the domains
satisfying (vii) are conformally equivalent (which is much stronger than homeomorphic) to the unit disk D.

In the language of extended complex plane (more rigorously Rimeann surfaces, or one-dimensional complex manifolds), Riemann
Mapping Theorem can be slightly strengthened as follows:

Corollary 4.55. Poincaré-Koebe Uniformization Theorem.

Suppose U ⊆ C∞ is a simply-connected domain. Then U is conformally equivalent to one of C∞ , C and D. More specifically,
if C∞ \U contains more than two points, then U is conformally equivalent to D; if C∞ \U contains exactly one point, then U
is conformally equivalent to C; if C∞ \U is empty, then U is conformally equivalent to C∞ . Furthermore, C∞ , C and D are not
conformally equivalent.

Proof. C∞ is not conformally equivalent to C or D, as they are not even homeomorphic. C is not conformally equivalent to D, as any
holomorphic function from C to D is constant by Liouville’s Theorem.

For U = C∞ , there is really nothing to prove. For U = C∞ \{w} and w ∈ C, the Möbius transformation f (z) = 1/(z − w) maps
U conformally onto C. For the case that C∞ \U contains more than two points, if ∞ ∈ C∞ \U , then this is just Riemann
Mapping Theorem; if ∞ ∉ C∞ \U , we can always use a Möbius transformation to map some w ∈ C∞ \U to ∞, which changes
the problem to the previous case.

Remark. The classification of the conformal equivalence classes of multiply-connected domains are much more complicated.
The conformal equivalence of annuli has been given in Theorem 4.50. The following theorem completely describes the conformal
equivalence class of doubly-connected domains. We are not going to give the proof here.

Theorem 4.56. Conformal equivalence of doubly-connected domains.

Suppose that U ⊆ C is a doubly-connected domain. Then U is conformally equivalent to an annulus with outer radius 1.
Moreover, the conformal mapping is unique up to translations, rotations, and inversion.
4.7. BOUNDARY CORRESPONDENCE* 91

Proof. See [Belyaev] Theorem 2.6.3.

4.7 Boundary Correspondence*

In the previous section, we have proven that any two proper simply-connected domains are conformally equivalent. Suppose that
f : U → V is a biholomorphism between two simply-connected domains. It is generally unclear if f can maps ∂U bijectively onto
∂V . The main result we are about to prove is as follows:

Theorem 4.57. Carathéodory Extension Theorem.

Suppose that U is a domain and f is a biholomorphism from U onto D. Then f can be continuous extended to a homeomor-
phism U → B (0, 1) if and only if ∂U is the image of a simple closed path.

The result has the following equivalent statement:

Theorem 4.58. Boundary Correspondence Theorem.

Suppose that U is a domain and f is a biholomorphism from U onto D. Then f maps ∂U bijectively to ∂D with orientation
preserved if and only if ∂U is the image of a simple closed path.

Definition 4.59. Accessible Points.

Suppose that U ⊆ C is open and α ∈ ∂U . We say that ζ is an accessible point of U , if for any sequence {z n } ⊆ U with lim z n = ζ
n→∞
there exists a path γ : [0, 1] → C such that:

γ(1) = ζ; γ([0, 1)) ⊆ U ;

and an increasing sequence {t n } ⊆ [0, 1) such that γ(t n ) = z n .

In other words, the path γ passes through all points in {z n } and lies entirely in U except for one of the end points.

Example 4.60. Inaccessible Points.

Let U := {z ∈ C : 0 < Re z < 1, 0 < Im z < 1} and I n := {x + 2−n i ∈ C : 0 É x É 1/2}. Let V := U \ ∞

¡S ¢
n=1 I n . Then [0, 1/2) are
inaccessible points of V , as shown in Figure 4.9. In other words, there are no paths in V that can approach points on [0, 1/2).

Im

Re

Figure 4.9: A domain with inaccessible boundary points.

The key result is that for accessible points, the limit of the mapping on the boundary exists. First we present a lemma due to Lindelöf
and Koebe.
92 CHAPTER 4. CONFORMAL MAPPINGS

Im Im
zn ζ
γn (u n )

1
π/m

Re γn (v n ) Re
εn γn

z n0 γn (u n )
ζ0

Figure 4.10: The dashed line on the second diagram is the reflection of γn about the real axis.

Lemma 4.61. Koebe’s Lemma

Let {z n }, {z n0 } ⊆ D be two sequences such that z n → ζ and z n0 → ζ0 , where ζ, ζ0 ∈ ∂D and ζ 6= ζ0 . Let γn be a simple path joining z n
and z n0 which lies in the annulus A(0, 1 − εn , 1) where εn → 0 as n → ∞.

Suppose that f : D → C is holomorphic and bounded. If sup | f (z)| → 0 as n → ∞, then f is identically zero on D.
z∈γ∗
n

Proof. Suppose that f is not identically zero. Without loss of generality we may assume that f (0) = 0. If f has a zero of multiplicity
n at 0, then we can replace f by f (z)/z n which satisfies all conditions of the lemma. By applying a rotation, we may further
assume that ζ = ζ0 . That is, ζ and ζ0 are symmetric about the real axis.

We can find a angle π/m < arg ζ and a sector S := {z ∈ C : arg z ∈ (−π/m, π/m)} such that there are infinitely many n with
γ∗n ∩ S 6= ∅ and z n , z n0 ∉ S, as shown in Figure 4.10.

Let u n ∈ [0, 1] be the largest number such that arg γn (u n ) = π/m and let v n > u n be the smallest number such that arg γn (v n ) =
0. Then γn restricted to [u n , v n ] is a path in the sector which joins γn (u n ) and γn (v n ). Let γ be the reflection of γ about the
real axis. Then we obtain a path joining γn (v n ) and γn (u n ), as shown in Figure 4.10. Let σn be the concatenation of the two
paths such that σn is a path joining γn (u n ) and γn (u n ) and is symmetric about the real axis.

Let T (z) := eπi/m z be the rotation by 2π/m. We consider the concatenation of successive rotations of σn :

η n := σn ? (T ◦ σn ) ? · · · ? T m−1 ◦ σn
¡ ¢

Hence we obtain a simple closed curve η n which lies entirely in the annulus A(0, 1 − εn , 1).

Finally, define F (z) := f (z) f (z̄) and

G(z) := F (z) · F ◦ T (z) · ... · F ◦ T m−1 (z)

Suppose that r n := sup | f (z)| and M := sup| f (z)|. By Schwarz Reflection Principle, f (z̄) is also holomorphic in D and bounded
z∈γ∗
n z∈D
by M . In particular F is bounded by r n M on σn . G is holomorphic in D. For any z ∈ η∗n , one of the factors of G is bounded by
r n M and the rest are bounded by M 2 . Hence G is bounded by r n M 2m−1 on η∗n . Suppose that f is non-constant. By Maximum
Modulus Principle, we have
| f (0)|2m = |G(0)| < sup |G(z)| É r n M 2m−1
z∈η∗n

Let n → ∞. Since r n → 0, we have f (0) = 0, which is a contradiction. Hence f is identically zero.

Proposition 4.62. Existence of Continuous Extension.

Suppose that f : U → D is a biholomorphism and ζ ∈ ∂U is accessible. If there exists r > 0 such that B (ζ, r ) ∩U is connected,
then lim f (z) exists and has modulus 1.
z→ζ

Proof. First, for any {z n } ⊆ U with lim z n = ζ, let γ be a path that satisfies the conditions in Definition 4.59. γ(t n ) = z n for each
n→∞
n ∈ Z+ . We claim that lim| f ◦ γ(t )| = 1. Suppose for contradiction that there exists ε > 0 and a subsequence {s n } of {t n } such
t →1
4.7. BOUNDARY CORRESPONDENCE* 93

that | f ◦ γ(s n )| É 1 − ε for all n ∈ Z+ . Let {u n } be a subsequence of {s n } such that w := lim f ◦ γ(u n ) exists. Then |w| É 1 − ε or
n→∞
w ∈ D. The inverse function f −1 : D → U maps

f −1 (w) = lim f −1 ◦ f ◦ γ(u n ) = lim γ(u n ) = γ(1)

n→∞ n→∞

which suggests that ζ = γ(1) does not lie on the boundary of U , contradiction.

Next, supppose for contradiction that lim f (z) does not exists. Since B (0, 1) is compact, there exists {a n }, {b n } ⊆ U such that
z→ζ
lim a n = lim b n = ζ but lim f (a n ) =: a 6= b := lim f (b n ). By connectivity, we can find a simple path γ : [0, 1] → U ∪ {ζ}
n→∞ n→∞ n→∞ n→∞
joining a 1 , b 1 , a 2 , b 2 , · · · and γ(1) = ζ. For each n ∈ Z+ , let γn be the restriction of γ on [γ−1 (a n ), γ−1 (b n )]. Then f ◦ γn is a
simple path joining f (a n ) and f (b n ).

Let g (z) := f −1 (z) − ζ defined on D. We apply Koebe’s Lemma to g . By the previous part, we know that lim| f ◦ γ(t )| = 1, which
t →1
implies that εn := inf∗ | f (z)| → 0 as n → ∞. In other words, f ◦ γn tends to the unit circle uniformly. Moreover, we have
z∈γn

sup |g (w)| = sup |z − ζ| → 0 as n → ∞

w∈( f ◦γn )∗ z∈γ∗
n

since γ is uniformly continuous and γ(1) = ζ. By Koebe’s Lemma, g is identically zero, which is obviously impossible. In
conclusion, lim f (z) exists and is unique. We can continuous extend f to ζ ∈ ∂U .
z→ζ

Remark. For the case that ζ ∈ ∂U has no connected neighborhood in U , the limit does not exists generally. However, if we extend
the definition of accessible boundary points, and associate each connected component of B (ζ, r ) ∩U with one "accessible point",
then the proposition still holds.

We should also point out that there are cases when B (ζ, r )∩U has no connected components. The formal definition of an accessible
point is an equivalence class of paths whose end point is ζ. This defintion works even if B (ζ, r ) ∩U has no connected components,
where ζ corresponds to infinitely many accessible points.

Proposition 4.63. Injectivity of Continuous Extension.

Suppose that f : U → D is a biholomorphism and ζ1 , ζ2 ∈ ∂U are two distinct accessible points. Suppose that f extends to ζ1 , ζ2
by continuity (existence proven in the previous proposition), then f (ζ1 ) 6= f (ζ2 ).

Proof. Suppose for contradiction that f (ζ1 ) = f (ζ2 ) = w 0 ∈ ∂D. By applying a rotation we may assume that w 0 = −1. Let γ1 : [0, 1] → C
and γ2 : [0, 1] → C be two paths such that γ1 (1) = ζ1 , γ2 (1) = ζ2 , and γ1 ([0, 1)), γ2 ([0, 1)) ⊆ U . Since ζ1 6= ζ2 , there exists t 0 ∈ (0, 1)
such that |γ1 (t ) − γ2 (t )| > |ζ1 − ζ2 |/2 for all t ∈ (t 0 , 1). There exists δ > 0 such that B (−1, δ) does not intersect with γ1 ([0, t 0 ]) or
γ2 ([0, t 0 ]).

Let A := D ∩ B (−1, δ). In the polar coordinates centered at −1, there exists a suitable function ϕ(r ) such that A = {−1 + r eiθ :
0 É r É δ, −ϕ(r ) É θ É ϕ(r )}. For each r ∈ (0, δ), let w 1 ∈ ( f ◦ γ1 )∗ ∩ ∂B (−1, r ) and w 2 ∈ ( f ◦ γ2 )∗ ∩ ∂B (−1, r ). Let g := f −1 be the
inverse function. Then we have |g (w 1 ) − g (w 2 )| > |ζ1 − ζ2 |/2.

Let η be the circular arc joining w 1 and w 2 . Then we have:

¯Z ¯ Z Z ϕ(r )
1
|ζ1 − ζ2 | < |g (w 1 ) − g (w 2 )| = ¯¯ g 0 (z) dz ¯¯ É |g 0 (z)| dz É |g 0 (−1 + r eiθ )|r dθ
¯ ¯
2 η η −ϕ(r )
µZ ϕ(r ) ¶2
1 2 0 iθ
=⇒ |ζ1 − ζ2 | < |g (−1 + r e )|r dθ
4 −ϕ(r )
µZ ϕ(r ) ¶2 µZ ϕ(r ) ¶2
É |g 0 (−1 + r eiθ )| dθ r dθ (Cauchy-Schwarz Inequality)
−ϕ(r ) −ϕ(r )
Z ϕ(r )
É π2 r 2 · |g 0 (−1 + r eiθ )|2 dθ
−ϕ(r )

Integrate with respect to r :

Z δ |ζ1 − ζ2 |2
Z δ Z ϕ(r ) Ï
iθ 2
dr É 0
|g (−1 + r e )| r dθ dr = |g 0 |2
0 4π2 r 0 −ϕ(r ) A

The left hand side diverges unless ζ1 = ζ2 , whereas the right hand side is equal to the area of g (A) and is finite. Contradiction.
Hence we must have f (ζ1 ) 6= f (ζ2 ).
94 CHAPTER 4. CONFORMAL MAPPINGS

Remark. By cosine theorem, ϕ(r ) is given by:

r 2 + 1 − δ2
µ ¶
ϕ(r ) = arccos
2r
We do not need this form in the proof of the proposition.

Proof of Theorem 4.57. "=⇒": If f : U → D can be extended to a homeomorphism f : U → B (0, 1), then f |∂U is a homeomorphism
from ∂U to the unit circle ∂D. Therefore ∂U is a simple closed path.

"⇐=": Suppose that ∂U is a simple closed path. In particular, every point on ∂U is an accessible point of U . By Proposition
4.62 and Proposition 4.63, there exists a continuous injective extension of f to ∂U which is in fact bijective, as f is invertible.
The continuity of the extension follows trivially.

The following converse of Boundary Correspondence Theorem is very useful:

Theorem 4.64. Converse of Boundary Correspondence Theorem.

Suppose that U → C is a simply-connected domain whose boundary ∂U is the image of a simple closed path. f : U → C is
holomorphic in U and continuous on U . If f maps ∂U bijectively to ∂D, then f is a biholomorphism from U to D.

Proof. Suppose that w 0 ∈ D. There exists a neighborhood V ⊆ U of ∂U such that f (z) 6= w 0 in V . For any simple closed curve
γ : [0, 1] → U , we consider the increment of the argument of f (z) − w 0 as z goes along γ:

∆γ arg( f (z) − w 0 )

which is invariant under homotopy. Suppose that Γ is a simple closed path with Γ∗ = ∂U . and Λ := γ(0, 1). By the statement
of the theorem we know that
∆Γ arg( f (z) − w 0 ) = ∆Λ arg(w − w 0 ) = 2π

The same relation also holds for a positively-oriented simple closed path γ : [0, 1] → V that is homotopic to Γ in V . By
Argument Principle, f (z) − w 0 has exactly one zero in D, the interior of γ. But f (z) 6= w 0 for z ∈ U \D ⊆ V . We conclude that
there exists exactly one z 0 ∈ U such that f (z 0 ) = w 0 .

Similarly, we can prove that no points in U is mapped to ∂D or C\B (0, 1). Hence f is a bijection between U and D.

4.8 Schwarz-Christoffel Mappings*

In this section, we aim to construct the explicit formula for a conformal mapping from the upper half plane to a polygonal area. We
say that P is a polygonal area, if ∂P is the image of a piecewise-linear simple closed path.

We denote the vertices of ∂P by w 1 , ..., w n . By Riemann Mapping Theorem, there exists a biholomorphism f : D → P . By Boundary
Correspondence Theorem, f extends to a bijection between ∂D and ∂P . On the other hand, we know that φ(z) := (z − i)/(z + i) is a
biholomorphism from H to D, which maps ∂H = R bijectively to ∂D\{1}. Hence f ◦ φ is a biholomorphism from H to P which maps
R bijectively to ∂P with one point removed.

We shall first give the Schwarz-Christoffel Intgeral and prove that it maps R to a polygon. Then we shall show that any biholomor-
phism between the upper half plane and a polygonal area can be written in the form of the Schwarz-Christoffel Intgeral. After that
we will investigate the behavior if we include the point of infinity. We will conclude the section with some examples of the use of the
Schwarz-Christoffel Mapping.

For a i ∈ R and βi < 1, the function (z − a i )−βi could be multi-valued. We can define a holomorphic branch of it by cutting the plane
along the ray {a i + iy ∈ C : y É 0}. For x ∈ R, we define:
(
−βi |x − a i |−βi x > ai
(z − a i ) = −βi −iπβi
|x − a i | e x < ai

Sn
In particular, s(z) := (z − a 1 )−β1 · · · (z − a n )−βn has a holomorphic branch in the cut plane C\ i =1
{a i + iy ∈ C : y É 0}. The cut plane is
simply-connected, so s(z) has a primitive, namely the Schwarz-Christoffel Integral:
4.8. SCHWARZ-CHRISTOFFEL MAPPINGS* 95

Definition 4.65. Schwarz-Christoffel Integral.

Sn
Suppose that −∞ < a 1 < · · · < a n < +∞ and β1 , · · · βn ∈ (−∞, 1). On the cut plane C\ i =1
{a i + iy ∈ C : y É 0}, the function
defined by the integral Z z
S(z) = (ζ − a 1 )−β1 · · · (ζ − a n )−βn dζ
z0

is called the Schwarz-Christoffel Integral, where z 0 is a fixed point and the intgeral is taken along any piecewise smooth path
from z 0 to z on the cut plane.

The condition βi < 1 implies that (z − a i )−βi is integrable near the singularity a i . Hence S(z) can be continuously extended on the
n
real line R. In addition, if βi > 1, then
X
i =1 Pn
βi
(|z| − a 1 )−β1 · · · (|z| − a n )−βn É c|z|− i =1

for sufficiently large |z|. It is not difficult to show that the integral S(z) converges as |z| → ∞. We denote the limit by w ∞ :=
lim S(z).
z→∞

Proposition 4.66

Let S(z) be the Schwarz-Christoffel Integral defined in Definition 4.65. Let w i := S(a i ) for i = 1, ..., n.
n
βi = 2, then S maps R to a ∂P \{w ∞ }, where ∂P is a n-sided polygon whose vertices are given in order by w 1 , ..., w n .
X
(i) If
i =1
The point w ∞ lies on the line segment between w n and w 1 . Moreover, the interior angle of ∂P at the vertex w i is π(1−βi ).
n
βi < 2, then S maps R to a ∂P \{w ∞ }, where ∂P is a (n + 1)-sided polygon whose vertices are given in order
X
(ii) If 1 <
i =1
byà w 1 , ..., w!n , w ∞ . Moreover, the interior angle of ∂P at the vertex w i is πβi , and interior angle at the vertex w ∞ is
n
π βi − 1 .
X
i =1

Proof. We can see that (i) is in fact a special case of (ii), where the interior angle at w ∞ is π. So we only need to prove (ii).

For i ∈ {0, ..., n} and x ∈ (a i −1 , a i ), we have

iY
−1 n n Pn
βi
S 0 (x) = (x − a k )−βk (x − a k )−βk = |x − a k |−βk e−iπ
Y Y
k=i
k=1 k=i k=1

n
βk is constant for x ∈ (a i −1 , a i ). As
X
Hence arg S 0 (x) = −π
k=i
Z x Z x
S(x) = S(a i ) + S 0 (t ) dt = w i + arg S 0 (x) |S 0 (t )| dt
ai ai

n
βi with the real
X
It suggests that S maps (a i −1 , a i ) to a line segment (w i −1 , w i ) on the complex plane, which makes angle −π
k=i
axis. For x > a n , we have arg S 0 (x) = 0. Hence S maps (a n , +∞) to a line segment (w n , w ∞ ) parallel to the real axis. For x < a 1 ,
n
arg S 0 (x) = −π βi . S maps (−∞, a 1 ) to a line segment (w ∞ , w 1 ).
X
k=1

For i ∈ {1, ..., n}, the interior angle θi at the vertex w i is given by:
à ! à !
n n
0 0
θi = π − lim− arg S (x) − lim arg S (x) = π − π βi − π βi = π(1 − βi )
X X
x→a i x→a i+ k=i k=i +1

The interior angle θ∞ at the vertex w ∞ is given by:

à !
n n
θ∞ = (n + 1)π − θk = π βi − 1
X X
k=1 i =1
96 CHAPTER 4. CONFORMAL MAPPINGS

Theorem 4.67. Schwarz-Christoffel Theorem.

Suppose that the open set P ⊆ C is a polygonal area whose boundary ∂P is a n-sided polygon with vertices (in order) w 1 , ..., w n .
Suppose that the interior angle of ∂P at the vertex w i is π(1 − βi ) where β1 ∈ (−1, 1). If f : H → P is a biholomorphism, then
there exists −∞ < a 1 < · · · < a n < ∞ such that
Z z
f (z) = C 1 (ζ − a 1 )−β1 · · · (ζ − a n )−βn dζ +C 2
z0

where z 0 , C 1 and C 2 are complex constants. Moreover, the extension of f to the homeomorphism from H to P implies that
f (a i ) = w i for i ∈ {1, ..., n}.

Proof. By Boundary Correspondence Theorem f extends to a homeomorphism f : H → P . Let a i := f −1 (w i ) for i = 1, ..., n. f maps
the real line bijectively to ∂P .

For i ∈ {2, ..., n − 1}, the interior angle at the vertex w i is π(1 − βi ). Hence we define g i : {z ∈ H : Re z ∈ (a i −1 , a i +1 )} → C by

g i (z) = ( f (z) − w i )1/(1−βi )

Then g i (a i ) = 0. And g i extends the angle subtended by two line segments near w 0 to π. As a result, g i maps the infinite
half-strip to an infinite half-strip. By Schwarz Reflection Principle, g i can be holomorphically extended across the real axis
and becomes a holomorphic function on the infinite strip a i −1 < Re z < a i +1 . On the upper half strip, we have:

f 0 (z) g 0 (z)
f (z) = w i + g i (z)1−βi =⇒ = (1 − βi ) i
f (z) − w i g i (z)

Since f is bijective, we have f 0 (z) 6= 0. Hence g i0 (z) 6= 0. The Schwarz Reflection Principle suggests that we also have g i0 (z) 6= 0
in the lower half strip. By some continuity argument we must have that g i is injective on the real interval (a i −1 , a i +1 ), which
implies that g i0 (z) 6= 0 on (a i −1 , a i +1 ). In conclusion, g i0 (z) 6= 0 on the whole strip a i −1 < Re z < a i +1 .

Next we shall prove that a i is a simple pole of f 00 / f 0 . The derivatives of f :

¢2
f 0 (z) = (1 − βi )g i (z)−βi g i0 (z) f 00 (z) = −βi (1 − βi )g i (z)−βi −1 g i0 (z) + (1 − βi )g i (z)−βi g i00 (z)
¡

Hence
f 00 (z) g 0 (z) g i00 (z)
= −βi i +
0
f (z) g i (z) g i0 (z)

Notice that the power series expansion of g i near a i is given by g i (z) = g i0 (a i )(z − a i ) + O((z − a i )2 ) and that g i is non-zero.
Then a i is a simple pole of g i0 /g i with residue Res(g i0 /g i , a i ) = 1. In addition, g i00 /g i0 is holomorphic in the strip. Therefore
there exists a holomorphic function q i on a i −1 < Re z < a i +1 such that

f 00 (z) 1
0
= −βi + q i (z)
f (z) z − ai

Similarly, for the infinite strip {z ∈ C : −∞ < Re z < a 2 } and {z ∈ C : a n−1 < Re z < +∞}, there exists holomorphic functions q 1
and q n defined on the respective strips such that

f 00 (z) 1 f 00 (z) 1
= −βi + q 1 (z) = −βi + q n (z)
f 0 (z) z − a1 f 0 (z) z − an

Next we investigate the behavior of f 00 / f 0 at the infinity. For sufficiently large R > 0, f maps (−∞, −R) ∪ (R, +∞) to a line
segment of w n −−w 1 . By Schwarz Reflection Principle, f is holomorphically extended to C\B (0, R). Since f maps ∞ ∈ C to
a point on the line segment w n −−w 1 , f is bounded and hence holomorphic at the infinity. There exists m ∈ N and c m 6= 0
such that
c m c m+1
f (z) = f (∞) + m + m+1 + · · ·
z z
The derivative of f :
cm c m+1 cm
f 0 (z) = −m − (m + 1) + · · · = −m m+1 + p(z)
z m+1 z m+2 z
where p is holomorphic on C\B (0, R) and p(∞) 6= 0. Hence we have

f 00 (z) m + 1 p 0 (z)
0
=− +
f (z) z p(z)
4.8. SCHWARZ-CHRISTOFFEL MAPPINGS* 97

f 00 (z)
Hence lim = 0. In particular f 00 / f 0 is bounded near the infinity.
z→∞ f 0 (z)
f 00 (z) Xn βi
Finally we let h(z) := + . Then h is a bounded entire function with lim h(z) = 0. By Liouville’s Theorem h(z) = 0
f (z) i =1 z − a i z→∞

on the whole plane. Hence for z ∈ H we have

f 00 (z) X n βi
0
=−
f (z) i =1 z − ai
We integrate the equation along any piecewise smooth curve from z 0 to z on the upper half plane. Then
n
log f 0 (z) = − βi log(z − a i ) + const
X
i =1

Hence
f 0 (z) = C 1 (z − a 1 )−β1 · · · (z − a n )−βn

Integrate again along any piecewise smooth curve from z 0 to z on the upper half plane:
Z z
f (z) = C 1 (ζ − a 1 )−β1 · · · (ζ − a n )−βn dζ +C 2
z0

which completes the proof.

Remark. If one of the vertices of ∂P is at the infinity, the mapping formula still applies. Suppose that w k = ∞, as shown in Figure
4.11. We pick w k0 on the line segment w k−1 −−w k and w k00 on the line segment w k −−w k+1 . We connect w k0 and w k00 by a line segment
and obtain a (n + 1)-sided polygon ∂P 0 : w 1 −− · · · −−w k−1 −−w k0 −−w k00 −−w k+1 −− · · · −−w n .

By Schwarz-Christoffel Theorem, the mapping from H to P 0 is given by

Z z
0 00
f (z) = C 1 (ζ − a 1 )−β1 · · · (ζ − a k0 )−βk (ζ − a k00 )−βk · · · (ζ − a n )−βn dζ +C 2
z0

where a k0 := f −1 (w k0 ) and a k00 := f −1 (w k00 ). The interior angle at the vertices w k0 and w k00 are π(1 − β0k ) and π(1 − β00k ) respectively.
0 00 0 00
As w k0 , w k00 → ∞, we have P 0 → P and a k0 , a k00 → a k . The factor (z − a k0 )−βk (z − a k00 )−βk → (z − a k )−βk −βk . If −π(1 − βk ) is the angle of
intersection of the line w k−1 −−w k0 and w k00 −−w k+1 , then we have π(1 − βk ) + π(1 − β0k ) − π(1 − β00k ) = π or −β0k − β00k = −βk . In this
case, the mapping formula is exactly same as the finite polygon case:
Z z
f (z) = C 1 (ζ − a 1 )−β1 · · · (ζ − a k )−βk · · · (ζ − a n )−βn dζ +C 2
z0

But notice that we give a slightly different definition of βk .

w k−1
−π(1 − βk )
π(1 − β0k )
w k0
w k+1

wk π(1 − β00
k
)

w k00

Figure 4.11: A polygonal area with one vertex at the infinity.

Schwarz-Christoffel Theorem ensures that any biholomorphism between the upper half plane and the polygonal area is expressed
in terms of Schwarz-Christoffel Integral. However, the points a 1 , ..., a n are often unknown when we want to find such mapping
to a given polygonal area. The next proposition demonstrates the uniqueness of the mapping if we fix three points on the real
line.
98 CHAPTER 4. CONFORMAL MAPPINGS

Proposition 4.68

Suppose that the open set P ⊆ C is a polygonal area whose boundary ∂P is a n-sided polygon (n Ê 3) with vertices (in order)
w 1 , ..., w n . Given three points on the real line −∞ < a 1 < a 2 < a 3 < +∞, there exists a unique biholomorphism f : H → P such
that f (a i ) = w i for i = 1, 2, 3.

Proof. We know that there exists a biholomorphism g : H → P . Suppose that for i ∈ {1, 2, 3}, b i = g −1 (w i ) ∈ R. By Proposition 4.21,
there exists a Möbius transformation T such that T (a i ) = b i for i = 1, 2, 3. Moreover, T ∈ Aut(H) by Theorem 4.43 since
a 1 , a 2 , a 3 , b 1 , b 2 , b 3 ∈ R. Therefore f := g ◦ T is a biholomorphism from H to P such that f (a i ) = w i .

Next we prove uniqueness. Suppose that f¯ is another biholomorphism from H to P such that f¯(a i ) = w i . Then f¯ ◦ f ∈
Aut(H). In particular, f¯ ◦ f is a Möbius transformation by Theorem 4.43 which fixes a 1 , a 2 , a 3 . By Proposition 4.21, there is
a unique Möbius transformation that fixes three points in C∞ , namely the identity mapping. Hence we must have f¯ = f as
required.

Next we look at the case when the biholomorphism maps the infinity to a vertex of the polygon. We shall prove that the formula is
obtained by deleting the last term (z − a n )−βn in the integral.

Theorem 4.69

Suppose that the open set P ⊆ C is a polygonal area whose boundary ∂P is a n-sided polygon with vertices (in order) w 1 , ..., w n .
Suppose that the interior angle of ∂P at the vertex w i is π(1 − βi ) where β1 ∈ (−1, 1). f : H → P is a biholomorphism. If there
are −∞ < a 1 < · · · < a n−1 < ∞ such that f (a i ) = w i for i = 1, ..., n − 1 and f (∞) = w n , then
Z z
f (z) = C 1 (ζ − a 1 )−β1 · · · (ζ − a n−1 )−βn−1 dζ +C 2
z0

where z 0 , C 1 and C 2 are complex constants.

Proof. We choose a < a 1 and consider the Möbius transformation T (z) = 1/(a − z) ∈ Aut(H). T maps a 1 , ..., a n−1 and a n = ∞ to
b 1 , ..., b n−1 and b n = 0. Now g := f ◦ T −1 is a biholomorphism from H to P which maps b 1 , ..., b n to w 1 , ..., w n . By Schwarz-
Christoffel Theorem, we have
Z z0
0
g (z ) = C 1 (η − b 1 )−β1 · · · η−βn dη +C 2
z 00

Change of variable:
1 1
η= dη = dζ
a −ζ (a − ζ)2
1 1 1 ζ − ai
Since b i = , we have η − b i = − = . The integral becomes
a − ai a − ζ a − a i (a − ζ)(a − a i )
Z T −1 (z 0 ) Pn
βi −2
f ◦ T −1 (z 0 ) = C 10 (ζ − a 1 )−β1 · · · (ζ − a n−1 )−βn−1 (a − ζ) i =1 dζ +C 2
T −1 (z 00 )

where C 10 = C 1 (a − a 1 )β1 · · · (a − a n−1 )βn−1 . Since

Pn
i =1 βi = 2, we have
Z T −1 (z 0 )
f ◦ T −1 (z 0 ) = C 10 (ζ − a 1 )−β1 · · · (ζ − a n−1 )−βn−1 dζ +C 2
T −1 (z 00 )

Put z = T −1 (z 0 ), z 0 = T −1 (z 00 ). We obtain the desired formula

Z z
f (z) = C 10 (ζ − a 1 )−β1 · · · (ζ − a n−1 )−βn−1 dζ +C 2
z0

Example 4.70. Trigonometric Functions Again.

Find a conformal mapping from H to the infinite half-strip U := {z ∈ C : −π/2 < Re z < π/2, Im z > 0}.
4.9. HARMONIC FUNCTIONS AND DIRICHLET PROBLEM 99

Solution. We can consider U as a polygonal area with vertices w 1 = −π/2, w 2 = π/2 and w 3 = ∞. The interior angle at each vertex
is π/2, π/2 and 0 respectively. Hence β1 = 1/2, β2 = 1/2, β3 = 1. We choose a 1 = −1, a 2 = 1 and a 3 = ∞ on the real line. By
Theorem 4.69, the mapping is given by
Z z Z z
1
f (z) = (ζ + 1)−1/2 (ζ − 1)−1/2 dζ +C 2 = p dζ +C 2 = C 1 arcsin z +C 2
0 0 ζ2 − 1

Since f (−1) = −π/2 and f (1) = π/2, we have C 1 = 1, C 2 = 0. Hence the desired mapping is given by f (z) = arcsin z. We can
compare this result with Example 4.35.

4.9 Harmonic Functions and Dirichlet Problem

In this section we shall investigate the properties of harmonic functions and solutions to Dirichlet boundary value problem with
the help of conformal mappings. For simply-connected domain U enclosed by simple closed curve ∂U , we can always transform the
problem into Dirichlet BVP on a unit disk, by Riemann Mapping Theorem and Boundary Correspondence Theorem. First, We repeat
the definition of harmonic functions here.

4.9.1 Harmonic Functions.

Definition 0.12: Laplacian, Harmonic Functions.

∂2 ∂2
The differential operator ∇2 := + 2 acting on twice differentiable functions in R2 is called the Laplacian. f : U → R is
∂x 2 ∂y
called a harmonic function if f ∈ ker ∇2 .

For Dirichlet boundary value problem on an open set U , we need to find a harmonic function with prescribed value on the boundary.
More formally:

Definition 4.71. Dirichlet Boundary Value Problem.

The Dirichlet boundary value problem consists of solving

∇2 u = 0 in U u=f on ∂U

for some given function f defined on ∂U .

Remark. Recall that the real and imaginary parts of a holomorphic function are harmonic. On the contrary, any harmonic function
is the real part of some holomorphic function.

Proposition 4.72

Suppose that U ⊆ C is a simply-connected domain and u ∈ C 2 (U ) is a harmonic function. Then there exists a holomorphic
function f : U → C such that Re f = u. Moreover, u is analytic.

∂u ∂u
Proof. Consider g : U → C defined by g (z) = − i . g is real-differentiable as u ∈ C 2 (U ) and it is easy to check that g satisfies
∂x ∂y
Cauchy-Riemann equations. Therefore by Proposition 0.11, g is holomorphic. Since U is simply-connected, by Proposition
1.38 g has a primitive G on U .

Suppose that G(z) = a(z) + ib(z). Then

∂a ∂a ∂u ∂u
−i = G0 = g = −i for z ∈ U
∂x ∂y ∂x ∂y

∂a ∂u ∂a ∂u
Hence = and = . In particular ∇(a − u) = 0. By chain rule, a and u differ by a constant on U . Then f (z) :=
∂x ∂x ∂y ∂y
G(z) + (a(z 0 ) − u(z 0 )) is a holomorphic function that satisfies Re f = u on U . By Theorem 1.19, f is analytic. It follows that
u = Re f is analytic.
100 CHAPTER 4. CONFORMAL MAPPINGS

Theorem 4.73. Mean Value Property of Harmonic Functions.

Suppose that u : B (z 0 , r ) → R is harmonic. Then

1
Z 2π
u(z 0 ) = u(z 0 + r eiθ ) dθ
2π 0

Proof. The mean value property of harmonic functions follows directly from the mean value property of holomorphic functions,
which is a direct corollary of Cauchy’s Integral Formula:

By Proposition 4.72, there exists a holomorphic function f : U → C such that Re f = u. By Cauchy’s Integral Formula, we have

f (z 0 + r eiθ )
Z 2π
1 f (z) 1 1 2π
I Z
iθ
f (z 0 ) = dz = ir e dθ = f (z 0 + r eiθ ) dθ
2πi γ(z0 ,r ) z − z 0 2πi 0 r eiθ 2π 0
Take the real part:
1
Z 2π
u(z 0 ) = u(z 0 + r eiθ ) dθ
2π 0

Remark. We say that u : U → R satisfies the mean value property, if for all B (z 0 , r ) ⊆ U we have

1 2π
Z
u(z 0 ) = u(z 0 + r eiθ ) dθ
2π 0
We shall prove a converse of Theorem 4.73 in Theorem 4.80, which states that any function satisfying the mean value property is
harmonic.

Theorem 4.74. Extreme Value Property of Harmonic Functions.

Suppose that U ⊆ C is a domain and u : U → C is harmonic and non-constant. Then u cannot attain maximum or minimum
value in U .

Proof. It suffices to prove that u cannot attain maximum in U . Suppose for contradiction that u attains maximum value at z 0 ∈ U .
Choose r > 0 such that B (z 0 , r ) ⊆ U . Since u is harmonic and satisfies mean value property, we have

1 2π
Z
u(z 0 ) = u(z 0 + r eiθ ) dθ É sup u(z 0 + r eiθ )
2π 0 θ∈[0,2π]

Hence u(z 0 ) = sup u(z 0 + r eiθ ) and u is constant on ∂B (z 0 , r ). This holds for any r > 0. By continuity u must be constant
θ∈[0,2π]
on B (z 0 , r ). Notice that B (z 0 , r ) has limit points in U . Since u is analytic, by identity theorem u is constant on the whole U ,
which is a contradiction.

Lemma 4.75

Suppose that f : U → V is holomorphic and u : V → R is harmonic. Then u ◦ f : U → R is also harmonic.

Proof. Being harmonic is a local property. It suffices to consider any B (z 0 , r ) ⊆ U . Suppose that w 0 := f (z 0 ). Fix ε > 0 such that
B (w 0 , ε) ⊆ V . By continuity of f we can find δ > 0 such that f (B (z 0 , δ)) ⊆ B (w 0 , ε). Since B (w 0 , ε) is simply-connected, we can
find a holomorphic function g : B (w 0 , ε) → C such that Re g = u. Hence on B (z 0 , δ) we have u ◦ f = Re(g ◦ f ). u ◦ f is harmonic
on B (z 0 , r ) and hence on the whole U .

4.9.2 Poisson Kernel.

Now we turn to the simplest case of Dirichlet BVP. We can express the value of a harmonic function in the unit disk in terms of its
value on the boundary by the so-called Poisson formula. All we need is the mean value property of harmonic functions.

Theorem 4.76. Poisson Formula for Harmonic Functions on the Unit Disk.

Suppose that u : B (0, 1) → C is harmonic on D. For z ∈ D, we have:

1 2π 1 − |z 0 |2
Z
u(z 0 ) = u(eiθ ) dθ
2π 0 | eiθ −z 0 |2
4.9. HARMONIC FUNCTIONS AND DIRICHLET PROBLEM 101

Proof. For any z 0 ∈ D, the automorphism

z0 − z
ϕz0 (z) =
1 − z¯0 z
of the unit disk exchanges 0 and z 0 . Let v := u ◦ ϕz0 . Then v is also harmonic on D and v(0) = u(z 0 ). By mean value property,
we have
1 2π
Z
v(0) = v(eiβ ) dβ
2π 0
Notice that ϕz0 maps ∂D to ∂D by
z 0 − eiβ z 0 − eiθ
eiθ = =⇒ eiβ =
1 − z¯0 eiβ 1 − z¯0 eiθ
Take the differential and modulus on both sides:

|z 0 |2 − 1 1 − |z 0 |2 1 − |z 0 |2
i eiβ dβ = i eiθ dθ =⇒ dβ = dθ = dθ
(1 − z¯0 eiθ )2 |1 − z¯0 eiθ |2 | eiθ −z 0 |2

Substitute back to the integral and we have:

1 2π 1 − |z 0 |2
Z
u(z 0 ) = u(eiθ ) dθ
2π 0 | eiθ −z 0 |2

1 − |z 0 |2
Remark. The factor is called the Poisson kernel of the unit disk:
| eiθ −z 0 |2

1 − |z|2 ζ+z
µ ¶
P (ζ, z) := = Re
|ζ − z|2 ζ−z

The factor (ζ + z)/(ζ − z) is called Schwarz kernel.

The Poisson Integral Formula is also written in the following form:

1
Z 2π
u(z) = P (ζ, z)u(ζ) dθ where ζ = eiθ
2π 0

Alternatively, in the polar coordinates, the Poisson kernel is given by

1−r2
P (ζ, z) = P (φ − θ, r ) = where ζ = eiθ , z = r eiφ
1 − 2r cos(φ − θ) + r 2

1 2π 1−r2
Z
u(r, φ) = u(1, θ) dθ
2π 0 1 − 2r cos(φ − θ) + r 2

Corollary 4.77. Properties of Poisson Kernel.

(i) P (ζ, z) > 0 for all z ∈ D and ζ ∈ ∂D;

Z 2π
(ii) P (ζ, z) dθ = 2π for all z ∈ D;
0
(iii) Fix ζ ∈ ∂D. Then P (ζ, z) is harmonic in D.

Proof. (i). Trivial by definition.

1 − |z 0 |2
Z 2π Z 2π Z 2π
(ii). P (ζ, z) dθ = dθ = dβ = 2π.
0 0 | eiθ −z 0 |2 0

ζ+z
µ ¶
(iii). P (ζ, z) = Re and the Schwarz kernel is holomorphic in D.
ζ−z
102 CHAPTER 4. CONFORMAL MAPPINGS

4.9.3 Dirichlet Boundary Value Problems.

Now we can discuss the existence and uniqueness of the solution to Dirichlet BVP on the unit disk.

Lemma 4.78

Suppose that f : ∂D → R is continuous. Then u : D → R defined by the Poisson integral

Z 2π
u(z) = P (ζ, z) f (ζ) dθ
0

is a bounded harmonic function. Moreover, u(z) → f (ζ) as z → ζ at any ζ ∈ ∂D.

Proof. Since ∂D is compact, and P (ζ, z) and f (ζ) are continuous, P (ζ, z) f (ζ) is bounded. Hence u(z) is bounded. u is harmonic
because µZ 2π ¶ Z 2π
∇2 u(z) = ∇2 P (ζ, z) f (ζ) dθ = f (ζ)∇2 P (ζ, z) dθ = 0
0 0

Now we fix ζ = e ∈ ∂D and ε > 0. Since f is continuous at ζ, there exists δ > 0 such that | f (eiθ ) − f (eiθ0 )| É ε whenever
iθ0

|θ − θ0 | É δ. Then
¯ ¯¯ Z 2π ´ ¯¯
iθ0 ¯ ¯ 1
¯ ³
iθ iθ iθ0
¯u(z) − f (e )¯ = ¯ P (e , z) f (e ) − f (e ) dθ ¯¯
¯
2π 0
1 1
Z ¯ ¯ Z ¯ ¯
É P (eiθ , z) ¯ f (eiθ ) − f (eiθ0 )¯ dθ + P (eiθ , z) ¯ f (eiθ ) − f (eiθ0 )¯ dθ
¯ ¯ ¯ ¯
2π |θ−θ0 |<δ 2π |θ−θ0 |Êδ
=: I 1 + I 2

For I 1 , since | f (eiθ ) − f (eiθ0 )| É ε, we have

1 1
Z ¯ ¯ Z
iθ iθ iθ0 ¯
I 1 := P (e , z) ¯ f (e ) − f (e )¯ dθ É εP (eiθ , z) dθ < ε
¯
2π |θ−θ0 |<δ 2π |θ−θ0 |<δ
For I 2 , for |θ − θ0 | Ê δ the Poisson kernel
1−r2 1−r2
P (eiθ , z) = P (φ − θ, r ) = É →0
1 − 2r cos(φ − θ) + r 2 1 − 2r cos δ + r 2

as r → 1. Hence for fixed ε and δ we can find η > 0 such that P (φ − θ, r ) < ε whenever |θ − θ0 | Ê δ and |z − eiθ0 | < η. Then
1 1
Z ¯ ¯ Z ¯ ¯
I 2 := P (eiθ , z) ¯ f (eiθ ) − f (eiθ0 )¯ dθ < ε ¯ f (eiθ ) − f (eiθ0 )¯ dθ < M ε
¯ ¯ ¯ ¯
2π |θ−θ0 |Êδ 2π |θ−θ0 |Êδ

where M := sup | f (eiθ )|. Hence ¯u(z) − f (eiθ0 )¯ < (1 + M )ε. The convergence hence follows.
¯ ¯
θ∈[0,2π]

Theorem 4.79. Dirichlet BVP on the Unit Disk.

Suppose that f : ∂D → R is continuous. Then u : D → R defined by the Poisson integral

2π 1 2π 1 − |z 0 |2
Z Z
u(z) = P (ζ, z) f (ζ) dθ = f (eiθ ) dθ
0 2π 0 | eiθ −z 0 |2

is the unique solution to the Dirichlet BVP:

∇2 u = 0 in D u=f on ∂D

Proof. By the previous lemma, the Poisson integral u(z) is a solution to the Dirichlet BVP. Suppose there is another solution v.
Then u − v is a harmonic with u − v = 0 on ∂D. By extreme value property u − v is constant in D and by continuity u = v as
required.

Remark. Very often we are dealing with piecewise continuous functions as boundary value condition instead. In this case the
uniqueness does not hold. However, if we restrict to bounded harmonic functions on D which agrees with f at those points on ∂D
where f is continuous, Dirichlet BVP still has unique solution.

Remark. As an application of Theorem 4.79, we can prove the converse of Theorem 4.73.
4.9. HARMONIC FUNCTIONS AND DIRICHLET PROBLEM 103

Theorem 4.80

Suppose that U ⊆ C is a domain and u : U → R satisfies the mean value property. Then u is a harmonic function.

Proof. For any B (z 0 , r ) ⊆ U , u satisfies the mean value property on B (z 0 , r ). By Theorem 4.79, (with an appropriate linear mapping)
there exists a unique harmonic function v : B (z 0 , r ) → R such that u = v on ∂B (z 0 , r ). Both u and v satisfies the mean value
property and hence the extrem value property. It follows that u = v in B (z 0 , r ). Since B (z 0 , r ) is arbitrary, u = v in U . Therefore
u is a harmonic function.

Remark. The Dirichlet BVP on any simply-connected domain with well-behaved boundary can be transformed to the Dirichlet BVP
on the unit disk. As an example we give the explicit formula for the solution to Dirichlet BVP on the upper half plane.

Theorem 4.81. Dirichlet BVP on the Upper Half Plane.

Suppose that f : R → R is continuous and both lim f (x) and lim f (x) exists and are finite. Then u : H → R defined by the
x→−∞ x→+∞
integral µ
1
Z +∞ f (t )
¶
u(z) = Re dt
πi −∞ t −z
is the unique solution to the Dirichlet BVP:

∇2 u = 0 in H u=f on ∂H

Proof. For any z 0 ∈ H, consider the biholomorphism

z − z0
ψ(z) =
z − z̄ 0
which maps H onto D, (−∞, +∞) onto ∂D\{1}, and z 0 to 0. Then ψ ◦ f is continuous on ∂D\{1} and bounded on ∂D. By
Theorem 4.79 and the remark after it, there exists a unique bounded harmonic function v : D → R such that v = ψ ◦ f on
∂D\{1}. Hence u := ψ−1 ◦ v is the unique solution to the Dirichlet BVP on H.

Now we turn to the computation of the explicit formula of the solution. By mean value property we have

1
Z 2π
v(0) = v(eiθ ) dθ
2π 0

Since ψ maps (−∞, +∞) onto ∂D\{1},

t − z0 t − z0
µ ¶
eiθ = =⇒ θ = −i log
t − z̄ 0 t − z̄ 0
Take the differential:
t − z̄ 0 z̄ 0 − z 0 z̄ 0 − z 0 2 Im z 0
µ ¶ µ ¶
2
dθ = −i · − dt = i dt = 2 dt = Re dt
t − z0 (t − z̄ 0 )2 (t − z 0 )(t − z̄ 0 ) t − 2t Re z 0 + |z 0 |2 i(t − z 0 )

Since v(0) = u(z 0 ), we have

1
Z +∞ µ
2
¶ µ Z +∞
1 f (t )
¶
u(z 0 ) = f (t ) Re dt = Re dt
2π −∞ i(t − z 0 ) πi −∞ t − z

Remark. In the Cartesian coordinates, the result can be written as

y +∞ f (t )
Z
u(x, y) = dt
π −∞ (t − x)2 + y 2