MAT3253 Complex Variables

Lecture Notes
Kenneth Shum

Spring 2024

Contents
0.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

1 Definitions and arithmetic of complex numbers 1

1.1 Axioms and models of complex field . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Complex numbers as points on a plane . . . . . . . . . . . . . . . . . 3
1.1.2 Complex numbers as square matrices . . . . . . . . . . . . . . . . . . 4
1.1.3 Complex numbers as polynomials . . . . . . . . . . . . . . . . . . . . 5

2 Geometry of complex numbers 7

2.1 Polar form and complex conjugate . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Complex numbers in Python . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 DeMoivre formula and roots of complex numbers . . . . . . . . . . . . . . . 13

3 Linear fractional transformation, Riemann sphere 16

3.1 Geometry on the complex plane . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2 Linear fractional transformations . . . . . . . . . . . . . . . . . . . . . . . . 19
3.3 Riemann sphere and spherical representation of complex numbers . . . . . . 21
3.4 Appendix: Details of stereographic projection . . . . . . . . . . . . . . . . . 22

4 Mapping properties of linear fractional transformation 27

4.1 Composition of linear fractional transformations . . . . . . . . . . . . . . . 27

i
 4.2 Mapping circles to circles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Construction of linear fractional transformation . . . . . . . . . . . . . . . . 32
4.4 Appendix: Complex projective space . . . . . . . . . . . . . . . . . . . . . . 37

5 Complex exponential function and Euler’s formula 39

5.1 Complex plane as a metric space . . . . . . . . . . . . . . . . . . . . . . . . 39
5.2 Complex series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Complex exponential function . . . . . . . . . . . . . . . . . . . . . . . . . . 46
5.4 Euler’s formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

6 Complex trigonometric and log function 51

6.1 Complex trigonometric functions . . . . . . . . . . . . . . . . . . . . . . . . 51
6.2 Complex log function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.3 Complex power function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6.4 Inverse of complex trigonometric functions . . . . . . . . . . . . . . . . . . . 59

7 Complex differentiability 61
7.1 Limit and continuity of complex function . . . . . . . . . . . . . . . . . . . 61
7.2 Complex differentiable function . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.3 Cauchy-Riemann equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7.4 Appendix: A necessary and sufficient condition for complex differentiability
at a point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

8 Holomorphic functions 70
8.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
8.2 Differentiation rules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

9 Properties of Holomorphic functions 76

9.1 Conformal property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
9.2 Phase portrait . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
9.3 Harmonic functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.4 Application to Dirichlet’s problem . . . . . . . . . . . . . . . . . . . . . . . 84
9.5 Additional proof . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
9.6 Appendix: Physical interpretation . . . . . . . . . . . . . . . . . . . . . . . 89

ii
10 Complex differential form 90
10.1 Tangent space and differential form . . . . . . . . . . . . . . . . . . . . . . . 90
10.2 Complex differential form . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
10.3 Partial differential operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

11 Multi-valued functions 102

11.1 Parametric curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
11.2 The discontinuity of the principal branch of the argument function . . . . . 104
11.3 Computing the angle within a half-plane . . . . . . . . . . . . . . . . . . . . 105
11.4 The angle function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

12 Winding number, derivative of inverse function 111

12.1 Winding number, branch point, branch cut . . . . . . . . . . . . . . . . . . 111
√
12.2 Riemann surface of function z . . . . . . . . . . . . . . . . . . . . . . . . . 112
12.3 Derivative of inverse function . . . . . . . . . . . . . . . . . . . . . . . . . . 115
12.4 Appendix: A more rigorous definition of branch point . . . . . . . . . . . . 120

13 Contour integral 123

13.1 Review of line integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
13.2 Integration of complex-valued function defined on the real line . . . . . . . 124
13.3 Contour integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
13.4 Appendix: Definition of line integral and contour integral using differential
form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

14 Calculus of contour integral 132

14.1 Independence of parameterization . . . . . . . . . . . . . . . . . . . . . . . . 132
14.2 Varying the paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
14.3 Fundamental theorem of calculus for complex integral . . . . . . . . . . . . 135
14.4 Independence of path . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

15 Cauchy-Goursat theorem 141

15.1 Cauchy theorem version 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
15.2 The length of a contour and ML inequality . . . . . . . . . . . . . . . . . . 142
15.3 Cauchy-Goursat theorem for triangular domains . . . . . . . . . . . . . . . 144

iii
16 Local primitive function and Cauchy theorem for regions of various shapes149
16.1 Application to the evaluation of real integral . . . . . . . . . . . . . . . . . 149
16.2 Existence of local primitive function . . . . . . . . . . . . . . . . . . . . . . 151
16.3 Defining a branch of log function by integral . . . . . . . . . . . . . . . . . . 157

17 Cauchy integral formula 160

17.1 Cauchy integral formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
17.2 Liouville theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
17.3 Fundamental theorem of algebra . . . . . . . . . . . . . . . . . . . . . . . . 166

18 Taylor series expansion 167

18.1 Convergence of power series . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
18.2 Differentiation of power series . . . . . . . . . . . . . . . . . . . . . . . . . . 170
18.3 Existence of power series expansion for holomorphic functions . . . . . . . . 173
18.4 Uniqueness of Taylor series . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
18.5 Application to generating functions . . . . . . . . . . . . . . . . . . . . . . . 179

19 Identity theorem 182

19.1 Cauchy integral formula for higher-order derivatives . . . . . . . . . . . . . 182
19.2 Identity theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186

20 Maximum modulus principle, zero of a function 190

20.1 Maximum modulus principle . . . . . . . . . . . . . . . . . . . . . . . . . . 190
20.2 Zeros of a function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
20.3 Classification of singular points . . . . . . . . . . . . . . . . . . . . . . . . . 194
20.4 Appendix: Casorati-Weierstrass theorem . . . . . . . . . . . . . . . . . . . . 196

21 Laurent series 197

21.1 Removable singularity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
21.2 Pole order . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
21.3 Laurent series and residue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
21.4 Appendix: Existence of Laurent series expansion . . . . . . . . . . . . . . . 208

22 Residues 212
22.1 Uniqueness of Laurent series coefficients . . . . . . . . . . . . . . . . . . . . 212

iv
 22.2 Calculation of residues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

23 Evaluation of real integral 220

23.1 Semi-circular contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
23.2 Jordan lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

24 Indented contour and keyhole contour 227

24.1 Indented contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
24.2 Keyhole contour . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
24.3 Appendix: Differentiation and residue at the point of infinity . . . . . . . . 236

25 Argument principle 242

25.1 Analytic definition of winding number . . . . . . . . . . . . . . . . . . . . . 242
25.2 Argument principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
25.3 Rouché theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
25.4 Appendix: Open mapping theorem . . . . . . . . . . . . . . . . . . . . . . . 251

26 Gamma function and Riemann zeta function 253

26.1 Morera’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
26.2 Weierstrass’s theorem on convergence of sequence of functions . . . . . . . . 256
26.3 Analytic continuation by power series and open discs . . . . . . . . . . . . . 263
26.4 Appendix: Schwarz reflection principle . . . . . . . . . . . . . . . . . . . . . 268
26.5 Appendix: Hurwitz theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 272

References
[Ahlfors] Ahlfors, Complex Analysis – An Introduction to the Theory of Analytic Functions
of One Complex Variable, 3rd edition, McGraw-Hill, New York, 1979.

[BakNewman] J. Bak and D. J. Newman, Complex analysis, third edition, Springer, 2010.

[BrownChurchill] J. W. Brown and R. V. Churchill, Complex Variables and Applications,

9th edition, McGraw-Hill, New York, 2014.

[Freitag] E. Freitag and R. Busam, Complex analysis, third edition, Springer, 2005.

v
[Miranda] R. Miranda, Algebraic curves and Riemann surface, Graduate Studies in Math-
ematics, vol. 5, American Mathematical Society, 1995.

[Rudin] W. Rudin, Principles of mathematical analysis, 3rd edition, McGraw-Hill, 1976.

[Spivak] M. Spivak, Calculus on manifolds – a modern approach to classical theorems of

advanced calculus, Addison-Wesley, 1965.

[Stein] E. M. Stein and R. Shakarchi, Complex analysis, Princeton Lectures in Analysis,

Princeton University Press, New York, 2003.

[StewartTall] I. Stewart and D. Tall, Complex analysis, 2nd edition, Cambridge University
Press, 2018.

vi
 0.1 Notation
N Set of integers larger than or equal to 1

Z Set of integers

Q Set of rational numbers

R Set of real numbers

sup, inf Supremum and infinum

lim sup Limit superior

C Set of complex numbers

z̄, z ∗ Conjugate of complex number z

Re(z) Real part of z

Im(z) Imaginary part of z

|z| The modulus/absolute value/magnitude of z

arg(z) The argument/angle of z

Arg(z) Principal argument of z

D(z, r) Open disc with radius r centered at z

(df )p Differential of f at a point p

f ′ , df /dz Complex derivative of function f

∂ ∂
∂z , ∂ z̄ Two differential operators
R
C f dz Complex integral of f along a curve C

Res(f ; z0 ) Residue of function f at the point z0

n(C; z0 ) Winding number of a closed curve C relative to the point z0

vii
1 Definitions and arithmetic of complex numbers
Summary:

ˆ Axioms of complex numbers

ˆ Models of complex field

We begin with the rational numbers. The set of rational numbers is denoted by Q. A
rational number can be written in the form a/b, where a and b are integer. Formally, we
can represent it by a pair (a, b) of integers. Two pairs (a, b) and (c, d) are said to be equal if
ad = bc. So strictly speaking, a rational number is an equivalence class of pairs of integers.
Addition and multiplication are performed according to the following rules:

(a, b) + (c, d) ≜ (ad + bc, bd)

(a, b) · (c, d) ≜ (ac, bd).

For example, 1/2 and 1/3 are represented by (1, 2) and (1, 3), respectively, and their sum
is computed by (1 · 3 + 2 · 1, 2 · 3) = (5, 6). The numbers in the form (k, 1) behave in the
same way as the integers. We see that integers are embedded in this number system. This
number system has the feature that we can take the reciprocal of any non-zero number. If
a ̸= 0 ̸= b, we have (a, b) · (b, a) = (ab, ab), which is equivalent to (1, 1).
The rational number is not closed under taking supremum and infimum. The supremum
of a set of numbers A is defined as the number s, such that s is larger than any number
in A, and if s′ is any other upper bound of A, we must have s ≤ s′ . In other words, the
supremum of a set A is the least upper bound. Likewikse, the infimum of a set A is defined
as the greatest lower bound of A. It is well known that the set

{x ∈ Q : x2 < 2}

does not have a smallest upper bound in Q. Indeed, we can find the supremum of this
√
set in the set of real numbers, namely, the irrational number 2. The real numbers R is
constructed to guarantee that every subset of R has supremum and infimum.

1.1 Axioms and models of complex field

The rational numbers Q and the real numbers R have a common algebraic structure.

1
 Definition 1.1. A number system (F, +, ·) is called a field if it satisfies the following
conditions.

1. (closed) a + b ∈ F for all a, b ∈ F .

2. (associative) (a + b) + c = a + (b + c), for all a, b, c ∈ F .

3. (commutative) a + b = b + a, for all a, b ∈ F .

4. (existence of zero) ∃0 ∈ F such that 0 + a = a + 0 = a, for all a ∈ F .

5. (additive inverse) for all a ∈ F , ∃a′ ∈ F such that a + a′ = 0.

6. (closed) a · b ∈ F for all a, b ∈ F .

7. (associative) (a · b) · c = a · (b · c), for all a, b, c ∈ F .

8. (commutative) a · b = b · a, for all a, b ∈ F .

9. (existence of one) ∃1 ∈ F such that 1 ̸= 0 and 1 · a = a · 1 = a, for all a ∈ F .

10. (multiplicative inverse) for all a ∈ F \ {0}, ∃a′′ ∈ F such that a · a′′ = 1.

11. (distributive) a · (b + c) = a · b + a · c, for all a, b, c ∈ F .

A subset K of a field F is called a subfield of F if the elements in K satisfy the all

axioms of a field, and F is called an extension of K.

Example 1.1. The set of rational numbers Q and the set of real numbers R are fields.
Moreover, Q is a subfield of R.

Definition 1.2. A complex field is a field that extends R and contains a special
number i such that i2 = −1, and is generated by R and i. A complex field is denoted
by the symbol C.

Remark. The phrase “generated by R and i” means that there is no proper subfield that
contains all real numbers and the special number i. Consider the rational functions in the
form f (T )/g(T ), where f (T ) and g(T ) are polynomial in T with complex coefficients. The

2
set of all complex rational functions is a field, and it contains all real numbers and the
number i. However, it is too large to be considered a complex field. If we mix the real
numbers and the special number i, we will never see the transcendental variable T .

1.1.1 Complex numbers as points on a plane

We can implement a complex field by regarding a point (x, y), where x and y are real
numbers, as a number. The set of all numbers form a plane called the complex plane or
Argand diagram. The point (x, y) can also be regarded as a vector. In this implementation,
addition is just defined by vector addition.
Let
F1 ≜ {(a, b) : a, b ∈ R}. (1.1)

Two pairs (a, b) and (c, d) are equal if a = c and b = d.

The addition and multiplication operators for two points in F1 are defined by

(a, b) + (c, d) ≜ (a + c, b + d),

(a, b) · (c, d) ≜ (ac − bd, ad + bc).

The real numbers are embedded in F1 by x 7→ (x, 0). By identifying x1 ∈ R with (x1 , 0)
and x2 ∈ R with (x2 , 0), the sum and product of x1 and x2 are respectively

(x1 , 0) + (x2 , 0) = (x1 + x2 , 0), and

(x1 , 0) · (x2 , 0) = (x1 x2 , 0).

Geometrically, it is just saying that the horizontal axis in the complex plane is the real
number line. We can show that this number system satisfies all the axioms of field.
The number (1, 0) is the multiplicative identity, as

(1, 0)(c, d) = (1 · c − 0 · d, 1 · d + 0 · c) = (c, d).

The special number (0, 1) has the property

(0, 1) · (0, 1) = (0 · 0 − 1 · 1, 0 · 1 + 1 · 0) = (−1, 0).

This is the number i required in the definition of complex field.

3
 Every number (x, y) in this system can be expressed as a linear combination of 1 = (1, 0)
and i = (0, 1),
(x, y) = x(1, 0) + y(0, 1) = x + iy (1.2)

We next show how to divide two complex numbers in principle. Suppose we want to
divide the complex number (a, b) by complex number (c, d), where a, b, c and d are real
numbers. We want to find a complex number (e, f ) such that

(e, f ) · (c, d) = (a, b).

If we multiply the two numbers on the left, and equate the corresponding components on
both sides, we obtain a system of linear equations
" #" # " #
c −d e a
= .
d c f b

So, the computation of division can be done by solving a system of two linear equations,
provided that the determinant of the matrix in the above equation is non zero. As the
determinant is equal to c2 + d2 , this can be done when c and d are not both zero.

1.1.2 Complex numbers as square matrices

In the second model, a complex number is represented by a 2 × 2 matrix.

The set of all complex numbers in the second model is
" #
n a −b o
F2 ≜ : a, b ∈ R . (1.3)
b a

Addition and multiplication are performed using the usual matrix addition and multipli-
cation. The additive and multiplicative identities are the zero matrix and identity matrix,
respectively. The arithmetic is exactly the same as in the first model, but now it is expressed
in matrix format:
" # " # " #
a −b c −d a + c −b − d
+ =
b a d c b+d a+c
" # " # " #
a −b c −d ac − bd −(bc + ad)
· = .
b a d c bc + ad ac − bd

4
The matrices in this model constitute a field.
A real number x is represented by a diagonal matrix.
" #
x 0
.
0 x

The subset of diagonal matrices is isomorphic to the real number system

" # " # " #
x1 0 x2 0 x1 + x2 0
+ =
0 x1 0 x2 0 x1 + x2
" # " # " #
x1 0 x2 0 x1 x2 0
· = .
0 x1 0 x2 0 x1 x2
"
#
0 −1
We let i denote the special matrix . We can easily check that
1 0
" #" # " #
0 −1 0 −1 −1 0
= .
1 0 1 0 0 −1

A complex number in matrix form can be decomposed as

" # " # " #
x −y 1 0 0 −1
=x +y . (1.4)
y x 0 1 1 0

1.1.3 Complex numbers as polynomials

A polynomial in R is in the form

a0 + a1 T + a2 T 2 + · · · + ad T d

where a0 , . . . , ad are real numbers, and d is called the degree of the polynomial. The symbol
T is an indeterminate. The set of all polynomials in T with real coefficients is denoted by
R[T ].
We impose an equivalence relation on R[T ] by regarding two polynomials f (T ) and
g(T ) to be equivalent if f (T ) − g(T ) is divisible by T 2 + 1. Each equivalence class can be
represented by a polynomial in the form a + bT , for some choice of real numbers a and b.
Formally, the resulting collection of equivalence class is denoted by R[T ]/(T 2 + 1).

5
 We can take R[T ]/(T 2 + 1) as the carrier set, and perform addition and multiplication
using the usual polynomial addition and multiplication, modulo the polynomial T 2 + 1.
The resulting algebraic system is a complex field.

All three representations are isomorphic to each other. The first construction emphasizes
that a complex number is a pair of real numbers. The second construction emphasizes that
complex multiplication is the same as multiplying matrices in a special form. The one-to-
one correspondence is given by
" #
a −b
(a, b) ←→ ←→ a + bT
b a

Definition 1.3. The special number i that satisfies i2 = −1 is called the imaginary
unit. For a complex number z = a + bi in C, define the real and imaginary part of
z as
Re(z) ≜ a and Im(z) ≜ b.

In the three models of complex numbers we described above, the imaginary units are
" #
0 −1
(0, 1), , and T
1 0

respectively.
We will use the notation
a + bi
to represent a complex number. We thus see that the complex numbers form a vector space
of dimension 2 over the real numbers. In (1.2) and (1.4), we use1 and i as a basis over R.
With this notation, the multiplication of complex numbers can be written as

(a + bi)(c + di) = ac − bd + i(ad + bc). (1.5)

We have a coercion of real numbers into the set of complex numbers

r 7→ r + 0i.

A complex number with zero imaginary part is regarded as a real number. On the other
hand, a complex number with zero real part is said to be purely imaginary.

6
2 Geometry of complex numbers
Summary:

ˆ Polar form and conjugate

ˆ DeMoivre formula

ˆ Geometry on the complex plane

2.1 Polar form and complex conjugate

A nonzero complex number, like a vector in R2 , can be written in polar form. Given any
nonzero complex number x + iy, we can write it in the form

x + iy = r(cos θ + i sin θ)

for some nonnegative integer r and angle θ. The number r signifies the length of the vector
corresponding to x + iy, and θ is the angle from the positive axis to the vector x + iy.
With the second construction of complex numbers (1.3), we can write the 2 × 2 matrix
corresponding to a + bi as
" # " #
a −b cos θ − sin θ
=r .
b a sin θ cos θ
" #
cos θ − sin θ
The matrix is a rotation matrix that represents the action of rotating
sin θ cos θ
counter-clockwise by θ.

Proposition 2.1. Given z1 = r1 cos θ1 + ir1 sin θ1 and z2 = r2 cos θ2 + ir2 sin θ2 in
polar form, the product of z1 and z2 can be computed by

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

Proof. The proof follows from the definition of complex multiplication and basic trigono-

7
metric identities,

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

 
= r1 r2 cos θ1 cos θ2 − sin θ1 sin θ2 + i(cos θ1 sin θ2 + sin θ1 cos θ2 )
= r1 r2 [cos(θ1 + θ2 ) + i sin(θ1 + θ2 )].

The geometric meaning of multiplication by a + bi = r(cos θ + i sin θ) is thus, (i) first

rotate by θ counter-clockwise, then (ii) scale up (or down) by a factor of r.

Definition 2.2. The argument of a nonzero complex number z is defined as the

angle from the positive real axis to the straight line from 0 to z. We write arg(z) to
denote the argument of z. The argument of z = 0 is not defined. The argument of
z is also called the phase or the angle.
The modulus of z is defined as
p
|z| = a2 + b2 .

The modulus of z is also called the absolute value, the magnitude or the radius.

The argument of a complex number is defined up to integral multiple of 2π, because we

get the same point on the complex plane if we rotate around the origin by 2π radian. The
argument function is thus a multi-function. The argument of 0 is undefined.
A nonzero complex number z = x + iy can be written in polar form

z = r cos θ + ir sin θ = r(cos θ + i sin θ),

where r > 0 is the modulus of z and θ is an argument of z.

Example 2.1. The number −1 in polar form can be written as

−1 = 1 · (cos(π + 2πk) + i sin(π + 2πk)),

where k is an integer. The argument of −1 can be any number in the set {π + 2πk : k ∈ Z}.
We may write
arg(−1) = π + 2πk, for k ∈ Z.

8
 To eliminate the ambiguity in the argument function, one may pick a value in a pre-
defined range. The common choices of this range is (i) (−π, π], or (ii) [0, 2π). The unique
value that falls within the chosen range is called the principal argument. We remark that
the notion of principal argument is relative to the choice of the range, and hence is a matter
of convention, with the purpose of simplifying notation.

Definition 2.3. Suppose we choose the interval (−π, π] as the range. Given a
nonzero complex number z, we define the principal argument of z as the unique angle
θ0 (in radian) in (−π, π] such that z = |z|(cos(θ0 ) + i sin(θ0 )), and write Arg(z) = θ0 .

In this notes we write Arg(z) to denote the principal argument, following the use of
notation in [BrownChurchill]. The argument function arg(z) starting with small letter a is
a multi-valued function.
Example 2.2. If we take (π, π] as the range, the principal argument of −1 − i is −3π/4.

Definition 2.4. Define the complex conjugate of a complex number z = a + bi by

z̄ ≜ z ∗ ≜ a − bi.

Geometrically, the complex conjugate of a complex number z is the reflection of z along

the real axis. The modulus is the distance between the origin and the point z in the complex
plane.

Proposition 2.5. (i) (z ∗ )∗ = z for any z ∈ C.

(ii) z ∗ = z if and only if z is real.
(iii) |z|2 = zz ∗ .
(iv) Given any two complex numbers z1 and z2 in C,

(z1 + z2 )∗ = z1∗ + z2∗ and (z1 z2 )∗ = z1∗ z2∗ .

Part (iv) in Prop. 2.5 says that the reflection of the sum (resp. product) of two complex
numbers is the same as the sum (resp. product) of the two points obtained by reflection.
The proof is simple and is omitted. Using Prop. 2.5, we can show that the modulus is a
multiplicative function.

9
 Proposition 2.6. For any two complex numbers z1 , z2 ∈ C,

|z1 z2 | = |z1 | |z2 |.

Proof. Use the fact that |z|2 = zz ∗ for any z ∈ C, and complex multiplication is commuta-
tive
|z1 z2 |2 = (z1 z2 )(z1 z2 )∗ = z1 z2 z1∗ z2∗ = z1 z1∗ z2 z2∗ = |z1 |2 |z2 |2 .

If we substitute z1 by a + bi and z2 by c + di in the proposition, we recover a classical

identity about sum of squares:

(ac − bd)2 + (ad + bc)2 = (a2 + b2 )(c2 + d2 ). (2.1)

The relationship between the real part, imaginary part and complex conjugate are
z + z∗ z − z∗
Re(z) = and Im(z) = . (2.2)
2 2i
We can use complex conjugate to calculate division efficiently. Suppose we want to
divide z1 = a + bi by z2 = c + di, where c and d are not zero. We multiply and divide by
the conjugate of z2 ,
z1 a + bi c − di ac + bd bc − ad
= · = 2 +i 2 . (2.3)
z2 c + di c − di c + d2 c + d2
We can also do complex division using the 2×2 representation of complex numbers. Division
is the same as taking matrix inverse,
" #" #−1 " #" # " #
a −b c −d a −b c d 1 1 ac + bd ad − bc
= = 2 .
b a d c b a −d c c2 + d2 c + d2 bc − ad ac + bd
"#
c −d
Remark. We note that c2 + d2 is the determinant of the matrix and is the same as
d c
the square of the absolute value of c + di. The conjugate operation in matrix representation
is just the transpose operation.

10
2.2 Complex numbers in Python
Complex number is a built-in type in Python. The imaginary unit is represented by the
symbol j.
# Basic arithemtic of complex numbers
z1 = 2 + 3 j
z2 = -1 + 0.2 j

print ( f " The sum of { z1 } and { z2 } is { z1 + z2 }. " )

print ( f " The difference of { z1 } and { z2 } is { z1 - z2 }. " )
print ( f " The product of { z1 } and { z2 } is { z1 * z2 }. " )
print ( f " The quotient of { z1 } and { z2 } is { z1 / z2 }. " )

The program output is:

The sum of (2+3 j ) and ( -1+0.2 j ) is (1+3.2 j ) .
The difference of (2+3 j ) and ( -1+0.2 j ) is (3+2.8 j ) .
The product of (2+3 j ) and ( -1+0.2 j ) is ( -2.6 -2.6 j ) .
The quotient of (2+3 j ) and ( -1+0.2 j ) is
( -1.346153846153846 -3.269230769230769 j ) .

Commands for the real part, imaginary part and conjugate is part of the data structure.
# Real part , imaginary part , and conjugate

z = 3.2 -5.3 j
print ( f " Real Part of { z } = { z . real } " )
print ( f " Imaginary Part of { z } = { z . imag } " )
print ( f " Conjugate of { z } = { z . conjugate () } " )

If we run this program, we get

Real Part of (3.2 -5.3 j ) = 3.2
Imaginary Part of (3.2 -5.3 j ) = -5.3
Conjugate of (3.2 -5.3 j ) = (3.2+5.3 j )

The argument and modulus are called absolute value and phase in Python.

11
# Absolute value and phase
from cmath import phase
from numpy import abs

z = 12+34 j
print ( f " z = { z } " )
print ( f " Argument of z = { phase ( z ) } " )
print ( f " Magnitude of z = { abs ( z ) } " )

The command phase and abs are imported from library cmath and numpy, respectively.
The returned value of phase() is in radian.
z = (12+34 j )
Argument of z = 1 .2 31 50 371 23 40 85 19
Magnitude of z = 36.05 551275 463989

We can check Proposition 2.1 by the Python program below.

from cmath import phase
from numpy import abs , sin , cos

z1 = 12+34 j
z2 = 3 -6 j
r1 = abs ( z1 ) # absolute value of z1
r2 = abs ( z2 ) # absolute value of z2
theta1 = phase ( z1 ) # argument of theta1
theta2 = phase ( z2 ) # argument of theta2
print ( f " z1 = { z1 } , z2 = { z2 } " )
print ( f " z1 * z2 = { z1 * z2 } " ) # compute the product directly

w = r1 * r2 *( cos ( theta1 + theta2 ) + sin ( theta1 + theta2 ) *1 j )

print ( f " The above product should be equal to { w } " )

The variables z1 and z2 can be any complex numbers. The result of z1*z2 and the value
of w are the same, up to numerical error.

12
z1 = (12+34 j ) , z2 = (3 -6 j )
z1 * z2 = (240+30 j )
The above product should be equal to
( 24 0 + 3 0 . 0 0 0 0 0 0 0 0 0 0 0 0 0 0 7 j )

2.3 DeMoivre formula and roots of complex numbers

The DeMoivre formula describes what happens to the argument of a complex number when
we raise it to some power.

Theorem 2.7 (DeMoivre formula). For any n ∈ Z and θ ∈ R, we have

(cos θ + i sin θ)n = cos(nθ) + i sin(nθ). (2.4)

Proof. The formula is obviously true when n = 0 or n = 1. We apply mathematical

induction to establish (2.4) for all positive integers n and for all real numbers θ. Suppose
(cos θ + i sin θ)n = cos(nθ) + i sin(nθ) holds for some positive integer n. Then

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

= (cos nθ + i sin nθ)(cos θ + i sin θ)
= (cos nθ cos θ − sin nθ sin θ) + i(cos nθ sin θ + sin nθ cos θ)
= cos(n + 1)θ + i sin(n + 1)θ.

For negative n, we let m = −n and re-use the result we just proved for positive integers.
1
(cos θ + i sin θ)n =
(cos θ + i sin θ)m
1 cos mθ − i sin mθ
= ·
cos mθ + i sin mθ cos mθ − i sin mθ
= cos mθ − i sin mθ
= cos nθ + i sin nθ.

This proves the DeMoivre formula for all integers.

13
 Using the matrix representation of complex numbers, the DeMoivre’s formula can be
stated as " #n " #
cos θ − sin θ cos(nθ) − sin(nθ)
= .
sin θ cos θ sin(nθ) cos(nθ)
Geometrically speaking, this says that rotating n times by an angle θ is the same as rotating
once by angle nθ.
√
Example 2.3. Compute (−1 + i 3)8 .
√
The complex number −1 + i 3 in polar form is
√
2(−1/2 + i 3/2) = 2(cos(2π/3) + i sin(2π/3)).

Hence, with the use of DeMoivre’s formula, we get

√
(−1 + i 3)8 = 28 (cos(8 · 2π/3) + i sin(8 · 2π/3)) = 256(cos(4π/3) + i sin(4π/3)).
√
In Cartesian form, the answer is 128(−1 − i 3).

Example 2.4. Compute (−1 + i)20 .

√
Express −1 + i in polar form 2(cos(3π/4) + i sin(3π/4)). By DeMoivre’s formula,

(−1 + i)20 = 220/2 (cos(20 · 3π/4) + i sin(20 · 3π/4))

= 1024(cos π + i sin π)
= −1024.

Example 2.5. Express sin(5θ) as a polynomial in sin(θ).

By DeMoivre’s formula,

(cos(5θ) + i sin(5θ)) = (cos θ + i sin θ)5

= cos5 θ + 5i cos4 θ sin θ − 10 cos3 θ sin2 θ − 10i cos2 θ sin3 θ + 5 cos θ sin4 θ + i sin5 θ.

Equating the imaginary parts, we obtain

sin(5θ) = 5 cos4 θ sin θ − 10 cos2 θ sin3 θ + sin5 θ

= 5(1 − sin2 θ)2 sin θ − 10(1 − sin2 θ) sin3 θ + sin5 θ
= 5 sin θ − 10 sin3 θ + 5 sin5 θ − 10 sin3 θ + 10 sin5 θ + sin5 θ
= 5 sin θ − 20 sin3 θ + 16 sin5 θ.

14
 In Python, we can compute the the n-th root of a complex number by the command
z**(1/n). Raising to a rational power m/n can be computed by z**(m/n), for integers m
and n. Python will return one of the n possible answers. Compare the following calculations
in Python.
# Computation of roots of complex numbers in Python
z = -1+1 j
w1 = ( z **3) **(1/4) # raise to the power 3 and then take
fourth root
w2 = z **(3/4) # raise to the power 3/4
print ( f " w1 = { w1 } " )
print ( f " w2 = { w2 } " )

print ( f " \ nThey are both fourth roots of { z * z * z }. " )

print ( f " w1 ^4={ w1 **4} " )
print ( f " w2 ^4={ w2 **4} " )

The number w1 is obtained by first raising z to the third power and then taking the
fourth root. The number w2 is computed by directly raising z to the power 3/4. Mathe-
matically, they are supposed to be the same, but the actually results are not the same. It
is because there are four possible answers when we take the fourth root. The fourth power
of both of them are equal to (−1 + i)3 = 2 + 2i.
w1 = ( 1 . 2 7 1 9 2 1 1 4 6 2 4 6 3 9 0 9 + 0 . 2 5 3 0 0 0 8 4 6 3 2 0 1 1 7 8 j )
w2 = ( - 0 . 2 5 3 0 0 0 8 4 6 3 2 0 1 1 7 7 7 + 1 . 2 7 1 9 2 1 1 4 6 2 4 6 3 9 0 9 j )

They are both fourth roots of (2+2 j ) .

w1 ^ 4 = ( 1 . 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 1 + 1 . 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 1 j )
w2 ^ 4 = ( 1 . 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 3 + 1 . 9 9 9 9 9 9 9 9 9 9 9 9 9 9 8 7 j )

15
3 Linear fractional transformation, Riemann sphere
Summary:

ˆ Inversion function

ˆ Linear fractional transformation

ˆ Riemann sphere

In Lecture 2, we delve into the geometry of complex multiplication. Now, we shift our
focus to complex division. Geometrically, on the complex plane, the inversion function
f (z) = 1/z can be understood as the composition of a circle inversion function and reflec-
tion. A circle inversion maps a point z = r(cos θ + i sin θ) to the point (1/r)(cos θ + i sin θ).
This mapping preserves the angle but takes the reciprocal of the radius. By taking the
complex conjugate of (1/r)(cos θ + i sin θ), we obtain:
1 1
(1/r)(cos θ − i sin θ) = = .
r(cos θ + i sin θ) z

Hence, the function z 7→ 1/z and the circle inversion function differ only by a complex
conjugate.
Reciprocals can be taken for all complex numbers except zero. We may define 1/0 as a
special symbol, denoted by ∞. The complex plane, together with this additional symbol,
is referred to as the extended complex plane. In contrast, the extended real number system
is obtained by adding two special points to the real number system, i.e., R ∪ {∞, −∞}. In
the real case, we have both +∞ and −∞ due to the total ordering of the real number line.
However, since there is no ordering in the complex case, we only have one point at infinity.

3.1 Geometry on the complex plane

We start with the simple but useful observation that

arg(z1 /z2 ) = arg(z1 ) − arg(z2 ) mod 2π.

This device allows us to compute the angle between two line segments in terms of the
argument function.

16
 We can test whether three points z1 , z2 and z3 lies on the same line in terms of complex
operations, without going to the real domain. The idea is: if we divide z1 − z2 by z3 − z2 ,
the argument of the quotient is the angle between the line segment between z1 and z2 , and
the line segment between z3 and z2 . They are collinear if and only if the angle is 0 or π,
z − z 
1 2
arg = 0 or π.
z 3 − z2
Equivalently, this condition can be written as
z − z 
1 2
Im = 0.
z3 − z2

# Determine whether three points lie on a straight line

# Input three tuples : pt1 = ( x1 , y1 ) , pt2 = ( x2 , y2 ) , pt3 =
( x3 , y3 )
# return True if the three points pt1 , pt2 , pt3 are collienar
def test_collinear ( pt1 , pt2 , pt3 ) :
z1 = complex ( pt1 [0] , pt1 [1]) # convert to complex numbers
z2 = complex ( pt2 [0] , pt2 [1])
z3 = complex ( pt3 [0] , pt3 [1])
return abs ((( z1 - z2 ) /( z3 - z2 ) ) . imag ) <= 1e -8 # test whether
the imaginary part equals 0 , up to numerical error

We can use this Python program as follows.

In [1]: test_collinear ((1 ,2) ,(2 ,3) ,(3 ,4) )
Out [1]: True

In [2]: test_collinear ((1 ,2) ,(2 ,3) ,(3 ,3) )

Out [2]: False

Similarly, we can derive a condition to test whether three complex points form an
equilateral triangle. If z1 , z2 and z3 are the vertices of an equilateral triangle, then the line
segments z1 − z2 and z3 − z2 must have the same length, and the angle between them is
equal to ±π/3. Hence,
√
z1 − z2 1 ± 3i
= cos(±π/3) + i sin(±π/3) = .
z3 − z2 2

17
This condition is equivalent to
√
z1 − z2 1 ± 3i
− =
z3 − z2 2 2
√
⇔ (2z1 − z2 − z3 ) = ± 3i(z3 − z2 )
⇔ (2z1 − z2 − z3 )2 = −3(z3 − z2 )2
⇔ z12 + z22 + z32 = z1 z2 + z1 z3 + z2 z3 .

This gives a concise criterion to determine whether z1 z2 z3 form an equilateral triangle.

We can express the equation of a circle in the complex plane using the modulus. Given
a point z0 in the complex plane and a positive real number, the circle with radius r and
center z0 has equation
|z − z0 | = r.

Using the fact that |z|2 = zz ∗ , we can write the equation in an alternate form:

(z − z0 )(z − z0 )∗ = r2 .

We know that a straight line in the x-y plane can be represented by equation

ax + by = c,

for some real numbers a, b and c. Using the relation x = Re(z) = (z + z ∗ )/2 and y =
Im(z) = (z − z ∗ )/(2i), we can write the equation as

z + z∗ z − z∗
a +b = c.
2 2i
After some algebraic manipulation, this can be simplified to

ai(z + z ∗ ) + b(z − z ∗ ) = 2ic

(b + ai)z − (b − ai)z ∗ = 2ic (3.1)

If we define α = b + ai, then we can further simplify it to

Im(αz) = c. (3.2)

In words, only the points on the straight line satisfies Im(αz) = c.

18
 Actually, there is nothing special about the imaginary part in the equation for straight
line. We can use the real part as well. We simply make an alternate substitution, by making
a substitution β = a − ib in (3.1), we will otain

Re(βz) = c. (3.3)

Hence, a straight line in the complex plane can be represented by (3.2) or (3.3).

3.2 Linear fractional transformations

We first treat the special case of affine transformations, which can be classified into three
types.
A translation is a function f : C → C in the form

fw (z) = z + w

for some complex constant w. Geometrically it is just moving horizontally by Re(w) and
vertically by Im(w).
A scaling function is simply expanding by a constant real factor. It can be written as

ga (z) = az

for some real constant a. This is also called a dilation.

A rotation by angle θ can be realized by multiplication by (cos θ + i sin θ),

hθ (z) = (cos θ + i sin θ)z.

We can understand rotation in matrix form. If the real and imaginary part of z are x and
y, respectively, the R.H.S. in the above equation is the same as a matrix multiplication,
" #" #
cos θ − sin θ x −y
sin θ cos θ y x

The 2 × 2 matrix on the left is a rotation matrix. Multiplication by this matrix represents
an action of rotation by θ radian counter-clockwise.
All three types of mappings preserve the shape of a geometric object in the domain. For
example, if we draw a triangle on the complex plane and apply a translation or a dilation
to it, we will get a triangle that is similar to the original one. In particular, straight line

19
is mapped to a straight line. If we have two intersecting lines, the angle between them is
invariant under translation, scaling and rotation.
If we apply a series of translation, dilation and rotation, the overall effect can be ex-
pressed in terms of an affine function

z 7→ az + b

where a is a nonzero complex number and b is a complex number. This transformation is

also called an affine transformation. We note that the composition of two affine functions
is also affine.
If we mix affine functions with the inversion function, using function composition, we
then obtain the class of linear fractional transformations.

Definition 3.1. A linear fractional transformation (or Möbius transformation, or

bilinear transformation) is a function in the form

az + b
f (z) = (3.4)
cz + d
where a, b, c and d are complex numbers with ad − bc ̸= 0.

If b = 0, c = 0 and d = 1, a linear fractional transformation reduces to a rotation

followed by a dilation, i.e., f (z) = az.
If a = 1, c = 0 and d = 1, a linear fractional transformation is a translation f (z) = z +b.
If a = 0, b = 1, c = 1 and d = 0, a linear fractional transformation becomes the inversion
function f (z) = 1/z.
We can verify that any linear fractional transformation can be expressed as a composi-
tion of translation, scaling, rotation, and inversion. If c = 0 and d ̸= 0, then (az + b)/d is
just an affine transformation, and hence can be computed by first multiplying by a/d and
then adding the constant b/d. When c ̸= 0, by applying a partial fraction expansion, we
can see that a linear fractional transformation can be written as:
az + b a bc − ad 1
= + . (3.5)
cz + d c c cz + d
From the above expression, we observe that the condition ad − bc ̸= 0 ensures that the
linear fractional transformation is a non-constant function. In fact, we may scale all four
parameters a, b, c and d by a nonzero complex constant such that ad − bc = 1.

20
 Figure 1: Riemann sphere and stereographic projection

3.3 Riemann sphere and spherical representation of complex numbers

When z = −d/c, the function f (z) = (az + b)/(cz + d) has division by zero. To remedy
this inconvenience, we extend the domain to the Riemann sphere.

Definition 3.2. We identify a complex number x + iy with a point (x, y) in R2 .

Consider the sphere

{(x, y, z) ∈ R3 : x2 + y 2 + (z − r)2 = r2 },

which has radius r and touches x-y plane at one point. The stereographic projection
is a function that maps a complex number x + iy to the point P (x, y) on the sphere
such that (x, y), P (x, y) and the north pole (0, 0, 2r) lie on a straight line. The north
pole is called the point at infinity, and is denoted by the symbol ∞. The sphere is
referred to as the Riemann sphere.

There is no easy way to perform addition and multiplication on the Riemann sphere.
The main purpose of introducing the Riemann sphere is to bring in the point of infinity,
which can be treated as an ordinary point like the other complex numbers, and making the
resulting Riemann sphere a compact set. Hence, the Riemann sphere is often called the
one-point compactification of the complex plane.

21
 Definition 3.3. The extended complex number system, as a set, is defined as C∪{∞}.
It is also called the extended complex plane. We will use the notation C̄. Other
common notation for the extended complex plane includes Ĉ and C∞ .

The behavior of a function f (z) near the point at infinity can be seen by making a
change of variable w = 1/z, and considering the function g(w) = f (1/w). For example, the
values of the linear fractional transformation f (z) = (az + b)/(cz + d), for z ∈ C, is the
same as the values of the function
a/w + b
g(w) =
c/w + d
for w = 1/z. If w is close to 0 but not equal to 0, we can write g(w) as a+bw
c+dw . When the
complex variable w approaches 0, the function g(w) approaches a/c. Hence, we can say
that the original function f (z) approaches a/c as z tends to infinity.
We can now refine the definition linear fractional transformation and define it as a
function on the extended complex plane.

Definition 3.4. Given complex numbers a, b, c and d with ad − bc ̸= 0, we extend

the domain of function f (z) = (az + b)/(cz + d) to the extended complex plane as
follows. When c = 0, we define

(a/d)z + (b/d) if z ̸= ∞,
f (z) =
∞ if z = ∞.

When c ̸= 0, define

 az+b
 if z ̸= −d/c, z ̸= ∞
 cz+d

f (z) = ∞ if z = −d/c


a/c if z = ∞.


3.4 Appendix: Details of stereographic projection

We identify a complex number x + iy on the horizontal plane in a 3D space by x + iy 7→
(x, y, 0). Write a general point in the 3D space as (α, β, γ) (See Fig. 1). Place a sphere S

22
of radius r centered at (0, 0, r) in the 3D space. The equation of S is

S : α2 + β 2 + (γ − r)2 = r2 . (3.6)

The sphere S is called the Riemann sphere. We call N = (0, 0, 2r) the north pole of S.
Given any point (x, y, 0), let P (x, y) be the point of S such that (0, 0, 2r), (x, y, 0) and
P (x, y) are collinear.
There is a one-to-one correspondence between the points on the complex plane and the
points on the punctured sphere S \ {(0, 0, 2r)}.

Proposition 3.5. The functions P : R2 → S \ {(0, 0, 2r)} described in the previous

paragraph is given by
 4r2 x 4r2 y 2r(x2 + y 2 ) 
P (x, y) = , , .
4r2 + x2 + y 2 4r2 + x2 + y 2 4r2 + x2 + y 2

This gives a bijection between the complex plane R2 and the punctured sphere S \
{(0, 0, 2r)}.

We may alternately write the projection function P with z = x + iy as input,

 4r2 Re z 4r2 Im z 2r|z|2 
P (x + iy) = , , .
4r2 + |z|2 4r2 + |z|2 4r2 + |z|2

When the radius of the sphere is equal to r = 1, we can further simplify it to

 4 Re z 4 Im z 2|z|2 
P (x + iy) = , , .
4 + |z|2 4 + |z|2 4 + |z|2

Proof. Let N denote the north pole (0, 0, 2r) of S. The triangle with vertices N , (0, 0, 0)
and (x, y, 0) and the triangle with vertices N , (0, 0, γ), P (x, y) are similar. This gives
x y 2r
= = . (3.7)
α β 2r − γ
2r−γ 2r−γ
Substitute α = 2r x and β = 2r y into the equation of sphere S (Equation 3.6), we get

2r − γ
2 2
(x + y 2 ) + (γ − r)2 = r2
2r
(2r − γ)(x2 + y 2 ) = 4r2 γ.

23
In the last step, we cancel the factor 2r −γ from both sides. This is possible because γ < 2r.
We can express γ in terms of x and y

2r(x2 + y 2 )
γ= .
4r2 + x2 + y 2
Hence,
8r3
2r − γ = .
4r2 + x2 + y 2
Substituting the expression of 2r − γ back to (3.7) gives

4r2 x 4r2 y
α= and β = .
4r2 + x2 + y 2 4r2 + x2 + y 2
This proves that P (x, y) can be written as in the proposition.
We now verify that P is a bijection by checking that P has an inverse function
2rα 2rβ

Q(α, β, γ) ≜ , . (3.8)
2r − γ 2r − γ
For any point (α, β, γ) on S \ {N }, we can compute
 2rα 2rβ 
P (Q(α, β, γ)) = P ,
2r − γ 2r − γ
 2r(2r − γ) 2r(2r − γ) 2r(α2 + β 2 ) 
= α , β , .
(2r − γ)2 + α2 + β 2 (2r − γ)2 + α2 + β 2 (2r − γ)2 + α2 + β 2
(3.9)

From the equation of S, we have

α2 + β 2 = 2rγ − γ 2 . (3.10)

The demoninators in (3.9) can be simplified to

(2r − γ)2 + α2 + β 2 = 4r2 − 4rγ + γ 2 + α2 + β 2 = 4r2 − 2rγ.

The point in (3.9) can then be simplified to (α, β, γ). This proves that the composite
function P ◦ Q is the identity function on the punctured sphere.
In the other direction, we want to show that
 4r2 x 4r2 y 2r(x2 + y 2 ) 
Q(P (x, y)) = Q , ,
4r2 + x2 + y 2 4r2 + x2 + y 2 4r2 + x2 + y 2

24
is equal to the point (x, y). To this end, we compute
4r x 2
2rα 2r 4r2 +x 2 +y 2
= 2r(x2 +y 2 )
=x
2r − γ 2r − 4r2 +x2 +y2

and
4r y2
2rβ 2r 4r2 +x 2 +y 2
= 2 2 ) = y.
2r − γ 2r − 4r2 +x+y
2r(x
2 +y 2

This proves that the composite function Q◦P is the identity function on the complex plane.
Because Q ◦ P and P ◦ Q are identity functions, we conclude that P is a bijection.

The stereographic projection is not only a bijection, it preserves circle and straight line.
Suppose that a plane with equation

aα + bβ + cγ = d (3.11)

for some constants a, b, c and d intersects with the sphere S. We know that the cross-section
is circular in shape. We derive the equation that describes the stereographic projection of
the cross-section as follows.
Consider a plane Π that contains the north pole of S. This plane intersects the sphere
S at a circle as long as it is not a horizontal plane, i.e., when a and b are not both zero.
Using the fact that the vector (a, b, c) is a normal vector of Π, we can write the equation
of Π as

[(α, β, γ) − (0, 0, 2r)] · (a, b, c) = 0

aα + bβ + γc = 2rc.

We note that when the plane Π passes through the north pole, the constant d in (3.11)
is equal to d = 2rc. The intersection of Π and the horizontal plane is the straight line
described by the equation aα + bβ = 2rc.
Next, suppose aα + bβ + cγ = d is the equation of a plane Π that cut through the sphere
S but does not pass through the north pole N . Let (x, y) be a point on the horizontal plane

25
such that the point (α, β, γ) = P (x, y) lies on the plane Π. We obtain the equation

4r2 x 4r2 y 2r(x2 + y 2 )

a + b + c =d
4r2 + x2 + y 2 4r2 + x2 + y 2 4r2 + x2 + y 2
4r2 ax + 4r2 by + 2rc(x2 + y 2 ) = d(4r2 + x2 + y 2 )
(d − 2rc)(x2 + y 2 ) − 4r2 ax − 4r2 by + 4r2 d = 0.

This equation defines a circle provided that d ̸= 2rc, which happens if and only if the plane
Π does not contain the north pole. By completing the squares we see that the center of the
circle is located at
2r2
(x, y) = (a, b)
d − 2rc
and the radius is
2r p
r2 (a2 + b2 ) − d(d − 2rc). (3.12)
|d − 2rc|
The number r2 (a2 + b2 ) − d(d − 2rc) should be positive, because the radius is supposed
to be positive. We can show this algebraically. We are assuming that the perpendicular
distance between the plane and the center of S is less than r, otherwise the plane Π and
the sphere S has no intersection, or intersect at a single point. We express this condition
using the dot product
(a, b, c)
(α, β, γ − r) · √ < r.
a2 + b2 + c2
This condition can be simplified to

(αa + βb + γc − rc)2 < r2 (a2 + b2 + c2 )

which can be further simplified to

d2 − 2drc < r2 (a2 + b2 ).

This is equivalent to the condition that r2 (a2 + b2 ) − d(d − 2rc) is positive.

26
4 Mapping properties of linear fractional transformation
Summary:

ˆ Bijective on Riemann sphere

ˆ Map circle to circle

ˆ 3-transitive

ˆ Cross ratio

In this lecture we study the mapping properties of linear fractional transformation,

whose domain and range include the point at infinity.

4.1 Composition of linear fractional transformations

From (3.5), we know that a linear fractional transformation can be expressed as the com-
position of translation, dilation, rotation, and inversion.
az + b a bc − ad 1
f (z) = = + .
cz + d c c cz + d
We adopt the following convention for the computation with the point at infinity.

ˆ For any complex number a, we define a + ∞ = ∞ + a = ∞.

ˆ For any b ∈ C̄ \ {0}, we define b · ∞ = ∞ · b = ∞

ˆ For any c ∈ C \ {0}, we define c/∞ = 0 and c/0 = ∞.

ˆ ∞ + ∞ and 0 · ∞ is undefined.

We can evaluate f (z) at the point at infinity. When c ̸= 0, we have

az + b a bc − ad 1
f (∞) = = + .
cz + d c c c·∞+d
With the above convention, we evaluate c∞ ˙ + d to ∞. The fraction 1/(c · ∞ + d) is equal
to 0. As a result, f (∞) is equal to a/c when c is not zero.
When z = −d/c, we have 1/(c(−d/c) + d) = ∞ because we define 1/0 = ∞. We see
that the above rules of calculations with ∞ is consistent with the definition f (−d/c) = ∞.

27
 Theorem 4.1. The composition of two linear fractional transformations is a linear
fractional transformation.

Proof. (sketch) In the followings, f (z) is a linear fractional transformation f (z) = (az +
b)/(cz + d). It suffices to consider three cases.
(i) The function f (z) + α is a linear fractional transformation for any complex constant α.
(ii) The function βf (z) is a linear fractional transformation for any nonzero complex con-
stant β.
(iii) The function 1/f (z) is a linear fractional transformation.

We next consider the inverse of a linear fractional transformation. We can use the fact
that a linear fractional transformation can be written as the composition of four types of
functions — translation, rotation, dilation, and inversion. Because the inverse of translation
(resp. rotation, dilation, inversion) is a translation (resp. rotation, dilation, inversion), we
have the following theorem.

Theorem 4.2. The inverse of a linear fractional function exists, and is a linear
fractional transformation.

We give a second, and more direct proof of this theorem below, by computing the inverse
of a linear fractional transformation, and showing that the inverse is a linear fractional
transformation.

Proof. Let f (z) be a linear fractional transformation defined by f (z) = (az + b)/(cz + d).
We consider two cases.
Suppose that c = 0. In this case, we must have a ̸= 0 and d ̸= 0 (because we assume
ad − bc ̸= 0). The function f (z) can be written in the form az + b, by writing a/d as a and
b/d as b. The inverse of f (z) is g(w) = (w − b)/a. We also have f (∞) = ∞ in this case.
The inverse of f (z) is 
(w − b)/a if w ̸= ∞,
g(w) =
∞ if w = ∞.
We can check that g ◦ f and f ◦ g are identity functions.

28
 Next suppose c ̸= 0. From w = (az + b)/(cz + d), we can apply the identity in (3.5) and
write
cw − a 1
= .
bc − ad cz + d
If w ̸= a/c, we can solve for z in C
1  bc − ad  b − dw
z= −d = .
c cw − a cw − a
If w = a/c, we have f (∞) = w = a/c. The inverse function of f (z) is

b−dw
 if ∞ ≠ w ̸= a/c
 cw−a


g(w) = ∞ if w = a/c


−d/c if w = ∞.


Since g(f (z)) and f (g(w)) are identity functions from Riemann sphere to itself, we conclude
that f is a bijection.

Example 4.1. Consider the function f (z) = (z + 1)/(z − i). When z = i, we define f (i) = ∞
and when z = ∞, we define f (∞) = 1. To compute the inverse function of f (z), we solve
for z in the equation w = f (z),
z+1
w=
z−i
wz − iw = z + 1
(w − 1)z = 1 + iw
1 + iw
z= .
w−1
̸ 1. The inverse of f (z) is thus
The last line requires w =

1+iw
 if w ̸= 1, w ̸= ∞,
 w−1


g(w) = ∞ if w = 1,


i if w = ∞.


Because the inverse of a linear fractional transformation always exists, we can conclude
that a linear fractional transformation is a bijection on the extended complex plane.

Theorem 4.3. A linear fractional transformation defines a bijection on the Riemann

sphere.

29
4.2 Mapping circles to circles
Linear fractional mapping has the special property of transforming a circle to a circle or a
straight line, and transforming a straight line to a circle or a straight line. It is obvious that
a translation, dilation and rotation preserve shape, it remains to show that the inversion
function transforms circle to circle. In fact, circles and straight lines on the complex plane
are considered as the same type of objects on the Riemann sphere. Both of them corresponds
to a circle on the Riemann sphere. We may think of a straight line on the complex plane
as a circle that passes through the point at infinity.

Theorem 4.4. The inversion function f (z) = 1/z transforms circles and straight
lines to circles and straight lines.

Instead of giving a formal proof, we illustrate this theorem by examples.

Example 4.2. Find the image of the circle |z − 2i| = 1 under the transformation f (z) = 1/z.
Let w = 1/z.
1
− 2i = 1
w
⇔ |1 − 2wi|2 = |w|2
⇔ 3|w|2 − 2wi + 2w∗ i = −1.
By completing square, the last equation is equivalent to
2i 2 1
w+ = ,
3 9
which is the equation of a circle with radius 1/3 and center −2i/3.

30
Example 4.3. Consider a circle C : |z − z0 | = |z0 | that passes through the origin. Find the
image of C under the transformation f (z) = 1/z.
Define a variable w = 1/z.
1
− z0 = |z0 |
w
⇔ |1 − wz0 |2 = |wz0 |2
⇔ 1 − wz0 − w∗ z0 ∗ = 0
⇔ Re(wz0 ) = 1/2.

This equation defines a straight line in the w-plane. Indeed, if we write w = u + iv and
z0 = x0 + iy0 , then
Re(wz0 ) = 1/2 ⇐⇒ ux0 − vy0 = 0.5.

Hence, the image of the circle |z − z0 | = |z0 | is a straight line with equation ux0 − vy0 = 0.5.

To prove Theorem 4.4, we can show that, for circles and straight lines in general po-
sition, the procedures in the previous two examples always go through . If we can prove
Theorem 4.4, we can extend it to linear fractional transformation as an immediate corollary.

Theorem 4.5. A linear fractional transformation f (z) = (az+b)/(cz+d) transforms

circles and straight lines to circles and straight lines.

31
4.3 Construction of linear fractional transformation
Motivated by the geometric result in the previous section, one may ask the question: Given
two circles (or straight lines) on the complex plane, can we fine a linear fractional trans-
formation that maps the first circle to the second circle. The answer is yes, and we can
establish a stronger result. It is known that a circle is uniquely determined by three points.
In the next theorem, we prove that we can find a linear fractional transformation that maps
three given points to three other given points.

Theorem 4.6. Given any three distinct points z1 , z2 and z3 and any three distinct
points w1 , w2 and w3 , all in the extended complex plane C̄, we can find a linear
fractional transformation that maps z1 to w1 , z2 to w2 , and z3 to w3 .

Proof. We first consider the special case w1 = 0, w2 = 1, and w3 = ∞. We claim that we

can find a linear fractional transformation that maps z1 to 0, z2 to 1 and z3 to ∞, for any
choice of z1 , z2 , and z3 . For example, when z1 , z2 and z3 are all in C, the function
z − z1 z2 − z 3
f (z) = · (4.1)
z − z 3 z2 − z 1
is a linear fractional transformation that satisfies the requirements.
There are three boundary cases to consider. When z1 = ∞ and z2 and z3 are finite, we
can map ∞ to 0, z2 to 1, and z3 to ∞ by
z2 − z 3
f (z) = .
z − z3
When z2 = ∞ and z1 and z3 are finite, we can map z1 to 0, ∞ to 1, and z3 to ∞ by
z − z1
f (z) = .
z − z3
Finally, when z3 = ∞ and z1 and z2 are finite, we can map z1 to 0, z2 to 1, and ∞ to ∞ by
z − z1
f (z) = .
z2 − z 1
This completes the proof of the claim.
To prove the existence of linear fractional transformation in Theorem 4.6, we can con-
sider a linear fractional transformation f (z) that maps z1 to 0, z2 to 1, and z3 to ∞, and

32
another linear fractional transformation g(w) that maps w1 to 0, w2 to 1, and w3 to ∞.
Then the composite function g −1 ◦ f is a linear fractional transformation that satisfies the
required mapping properties.
For explicit calculations, we can set up an equation
z − z 1 z2 − z3 w − w1 w2 − w3
· = · (4.2)
z − z 3 z2 − z1 w − w3 w2 − w1
and write w as a function of z. The resulting function of z will be a linear fractional
transformation that equals g −1 (f (z)).

Example 4.4. Find a linear fractional transformation that maps 0 to 0, 1 to 1, and z0 to

∞, where z0 is a complex number different from 0 and 1.
Because we want to take z0 to ∞, the linear fractional transformation has the form
az + b
f (z) =
z − z0
for some choice of complex constants a and b. Because we want f (0) = 0, the only choice
of constant b is b = 0. From the requirement f (1) = 1, we derive that a is equal to 1 − z0 .
The answer is
z
f (z) = (1 − z0 ) .
z − z0
Consider the case that z0 is equal to i. The three points 0, 1 and i in the domain of
the transformation lie on the circle with equation |z − (1 + i)/2|2 = 1/2. The three points
0, 1, ∞ in the range lie on the line Im(z) = 0 (the real axis plus the point at infinity). In
this case, the fractional linear transformation
(1 − i)z
z 7→
z−i
√
maps the circle with and radius 1/ 2 and center at (1 + i)/2 to the real axis.
Example 4.5. In this example we construct a linear fractional transformation that maps
three points z1 = 1, z2 = 2, and z3 = 3 on the real axis to three points w1 = i, w2 = 3i,
and w3 = 5i on the imaginary axis. We apply the equation in (4.2),
z−1 2−3 w − i 3i − 5i
· = · .
z−3 2−1 w − 5i 3i − i
After some algebraic calculations, we can simplify it to

w = 2iz − i.

33
The desired linear fractional transformation is

f (z) = 2iz − i = i(2z − 1).

Example 4.6. Consider the linear fractional transformations f1 (z) = 1/z and f2 (z) = 1 − z.
Both of them induce a permutation on the three points 0, 1, ∞ on the Riemann sphere.

f1 (0) = ∞, f1 (1) = 1, f1 (∞) = 0,

f2 (0) = 1, f2 (1) = 0, f2 (∞) = ∞.

In this example we want to exhaust all function that can be obtained by composing f1 and
f2 recursively. For example, using f1 and f2 , we can obtain functions in the form f1 (f2 (z)),
f2 (f1 (z)), f1 (f1 (f2 (z))), f2 (f1 (f2 (z))), etc. How many functions can we get?
First, we see that we can get the identity function from

f1 (f1 (z)) = f2 (f2 (z)) = z.

We can discover two new functions from

1
f1 (f2 (z)) = ,
1−z
1 z−1
f2 (f1 (z)) = 1 − = .
z z
If we apply the inversion function to f2 (f1 (z)), we get
z
f1 (f2 (f1 (z))) = .
z−1
We claim that these six functions
1 z−1 z
f0 (z) = z, f1 (z) = 1/z, f2 (z) = 1 − z, f3 (z) = , f4 (z) = , f5 (z) =
1−z z z−1
are all we can get. We can organize the above data in a composition table.

◦ f0 f1 f2 f3 f4 f5
f0
f1 f0 f3 f5
f2 f4 f0
f3
f4
f5

34
The non-empty cells correspond to the equations we know so far: f0 = f1 ◦ f1 = f2 ◦ f2 ,
f3 = f 1 ◦ f2 , f4 = f2 ◦ f1 , f5 = f1 ◦ f4 = f1 ◦ f2 ◦ f1 .
We complete the table, and note that the six functions are closed under composition.

f0 (z) = f1 (z) = f2 (z) = f3 (z) = f4 (z) = f5 (z) =

1 1 z−1 z
◦ z z 1−z 1−z z z−1
1 1 z−1 z
f0 (z) = z z z 1−z 1−z z z−1
1 1 1 z z−1
f1 (z) = z z z 1−z 1−z z−1 z
z−1 z 1 1
f2 (z) = 1 − z 1−z z z z−1 z 1−z
1 1 z 1 z−1
f3 (z) = 1−z 1−z z−1 z z z 1−z
z−1 z−1 z 1 1
f4 (z) = z z 1−z z−1 z 1−z z
z z 1 z−1 1
f5 (z) = z−1 z−1 1−z z z 1−z z

In every column and every row, each of the six functions appears once and only once.
As a matter of fact, these six functions correspond to the permutation of three points
0, 1 and ∞ on the Riemann sphere. Because there are six permutations in total, and the
action on three points determines a linear fractional transformation, we obtain six functions
accordingly.
The linear fractional transformation in (4.1) is a special one. It maps the three complex
points z1 , z2 and z3 to the real axis (plus the point at infinity). The image of a complex
number z0 under this function is called a cross ratio. We use the notation,
z0 − z1 z 2 − z3
[z0 , z1 , z2 , z3 ] ≜ · .
z0 − z3 z 2 − z1
Because linear fractional transformation is bijective and maps circle to circle/straight line,
the point z0 lie on the circle that contains z1 , z2 and z3 if and only if [z0 , z1 , z2 , z3 ] lies on
the real axis. We thus obtain a criterion that determine whether four distinct points lie on
a circle. We say that a set of points are concyclic if they are on the same circle. We have
the following test for concyclicity:

Im([z0 , z1 , z2 , z3 ]) = 0 iff z0 , z1 , z2 and z3 are concyclic. (4.3)

This gives a convenient test for concyclic points.

Example 4.7. Derive the equation of the circle that passes through (0, 0), (1, 0) and (0, 2)
in the x-y plane.

35
 We take z1 = 0, z2 = 1 and z3 = 2i. Let z = x + iy be a point that lie on the circle that
contains z1 , z2 and z3 . Applying the criterion in (4.3), we obtain the condition
 x + iy 1 − 2i 
Im · = 0,
x + iy − 2i 1
which can then be simplified to

x2 + y 2 − x − 2y = 0,

or equivalently
5
(x − 1/2)2 + (y − 1)2 = .
4
√
This is the equation of a circle with center at (1/2, 1) and radius 5/2
We implement a test of concyclicity in Python.
# Determine whether four distinct points lie on a circle
# Input four tuples : pt1 = ( x1 , y1 ) , pt2 = ( x2 , y2 ) , pt3 =
( x3 , y3 ) , pt4 =( x4 , y4 )
# return True if the four points pt1 , pt2 , pt3 , pt4 are
concyclic , False if not concyclic

def test_concyclic ( pt0 , pt1 , pt2 , pt3 ) :

z0 = complex ( pt0 [0] , pt0 [1]) # convert to complex numbers
z1 = complex ( pt1 [0] , pt1 [1])
z2 = complex ( pt2 [0] , pt2 [1])
z3 = complex ( pt3 [0] , pt3 [1])
return abs ((( z0 - z1 ) *( z2 - z3 ) /(( z0 - z3 ) *( z2 - z1 ) ) ) . imag ) < 1e -8

For example, we can use this Python function to check that four points (0, 0), (1, 0),
(1, 2), and (0, 2) lie on the circle in the previous example.
In [1]: test_concyclic ((0 ,0) , (1 ,2) , (0 ,2) , (1 ,0) )
Out [1]: True

36
4.4 Appendix: Complex projective space
There is another construction of the extended complex plane called the complex projective
space. This is analogous to real projective line.
In this 2
" #construction, we consider C , which consists of pairs that we write as a column
z1
vector , with both z1 and z2 being complex numbers and not both z1 and z2 are zero.
z2
" # " #
z1 z′
Define an equivalence relation by declaring ∼ 1′ if there exists a complex number
z2 z2
λ such that z1′ = λz1 and z2′ = λz2 . Two pairs are thus equivalent if they lie on the same
complex line that passes through the origin in C2 .
The complex projective line is defined as the set that consists of all equivalence classes
defined in the previous paragraph. A common notation for the complex projective line is
CP1 . In an equivalence class, the exact values of z1 and z2 are not relevant, but only the
fraction z1 /z2 is relevant.
" #When z2 is nonzero, the fraction z1 /z2 is a complex number, and
z
we can pick the vector , with z = z1 /z2 , as the representative of this equivalence class.
1
" #
1
If z2 = 0, we can select the vector as the representative.
0
We thus have a bijection between Ĉ" and # CP1 . A finite complex number z corresponds
z
to the equivalence class that contains . The point at infinity is the equivalence class
1
" #
1
that contains .
0
" #
z1
We can perform a linear fractional transformation by multiplying from the left by
z2
a 2 × 2 matrix, " #" #
a b z1
c d z2
" #
z1
One can show that it does not depend on the choice of in an equivalence class. If
z2
z2 ̸= 0, the resulting vector is equivalent to
" #
az+b
cz+d
1

37
where z = z1 /z2 . " #
1
If we multiply with the point at infinity , we obtain
0
" #" # " #
a b 1 a
∼
c d 0 c

If we multiply with the projective point that corresponds to the complex number −d/c,
" #" # " #
a b −d/c a(−d/c) + b
∼ ,
c d 1 0

which is equivalent to the point at infinity (using the assumption that ad − bc ̸= 0.) There-
fore, the definition using matrix multiplication is isomorphic to Definition 3.4.
" We# also observe that the condition ad − bc ̸= 0 is the same as requiring that the matrix
a b
is non-singular.
c d

38
5 Complex exponential function and Euler’s formula
Summary:

ˆ Complex sequence and series

ˆ Complex power series

ˆ Complex exponential function

ˆ Euler’s formula

5.1 Complex plane as a metric space

Complex numbers can be interpreted as points on a 2-dimensional plane. The topology of
the complex plane is the same as the Euclidean space R2 , because we have the same metric
function. As a result, the convergence concepts and limits in the real case and the complex
case are basically the same. The main difference is that we now have an extra point at
infinity.
The Euclidean distance between two complex numbers z1 and z2 can be expressed using
the modulus function,
d(z1 , z2 ) ≜ |z1 − z2 |.

We verify that d(z1 , z2 ) is indeed a metric function. It is clear that d(z1 , z2 ) ≥ 0 with
equality if and only if z1 = z2 , and d(z1 , z2 ) = d(z2 , z1 ). It remains to prove the triangle
inequality.

Proposition 5.1 (Triangle inequality).

|z1 + z2 | ≤ |z1 | + |z2 |

for any two complex numbers z1 and z2 .

Although the triangle inequality is geometrically obvious, we give an algebraic proof,

and show that it is compatible with the arithmetic of complex numbers.

39
Proof. Take the square of the left-hand side,

|z1 + z2 |2 = (z1 + z2 )(z1∗ + z2∗ )

= |z1 |2 + 2 Re(z1 z2∗ ) + |z2 |2
≤ |z1 |2 + 2| Re(z1 z2∗ )| + |z2 |2 . (5.1)

Because the real part of any complex number has absolute value less than or equal to
the modulus of the complex number, we obtain

| Re(z1 z2∗ )| ≤ |z1 z2∗ | = |z1 | |z2∗ | = |z1 | |z2 |. (5.2)

Substituting (5.2) into (5.1), we get

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

The concepts of point-set topology in R2 can be directly imported to the complex case.
For example, open disc is defined as follows.

Definition 5.2. The open disc of radius r centered at z0 is the set

D(z0 , r) ≜ {z ∈ C : |z − z0 | < r}.

A subset A in C is called an open set for every point z0 ∈ A, we can find a sufficiently
small radius r so that D(z0 , r) is a subset of A. A set B is said to be closed if the complement
of B in C is open, i.e., given any point w0 that is not in B, there is radius r such that
D(w0 , r) and B are mutually disjoint.
As for the “open discs” centered at the point at infinity, we take the inversion function
w = f (z) = 1/z. A circle centered at the origin in the w-plane {w ∈ C : |w| < ϵ}
corresponds to a neighborhood of ∞ in extended complex plane in the z variable. In the
Riemann sphere, it is a cap at the top of the sphere. Hence, a small “open disc” centered
at the point at infinity has the form

{z ∈ C : |z| > R}

for some large radius R.

40
Notation: A sequence of complex numbers z1 , z2 , z3 , . . . is denoted by (zk )∞
k=1 or simply
{zk }.

Definition 5.3. Given a complex sequence (zn )∞

n=1 , we say that zn converges to a
complex number w ∈ C if

∀ϵ > 0 ∃N, s.t. |zn − w| < ϵ, ∀n ≥ N.

Geometrically, it is equivalent to saying that for any small ϵ, the points in the com-
plex sequence will eventually stay inside the open disc D(w, ϵ). A sequence that is
divergent if it is not converging to any complex number in C.
We write
lim zn = w
n→∞

if (zn )∞
n=1 converges to the limit w.

Example 5.1. The complex sequence {zn } defined by

1 i
zn = + 2
n n
converges to 0, because |zn | → 0 as n → ∞.
n
Example 5.2. Compute limn→∞ n+i .
We can first make a guess that the limit should be 1, because when n is large, adding i
to n has negligible effect. To make the argument rigorous, we write
n n+i−i i
= =1− .
n+i n+i n+i
Then
n i 1
−1 = =√ → 0, as n → ∞.
n+i n+i n2 + 1
Therefore n/(n + i) → 1 as n → ∞.
If we consider the complex sequence as points on the Riemann sphere, then it may
converge to the point at infinity, even though it does not converge to any point in the
complex plane.

41
 Definition 5.4. A sequence of complex numbers {zk }∞ k=1 is said to converge to the
point at infinity if for any positive number r, there exists a sufficiently large integer
N such that |zk | > r for all k ≥ N .

When the sequence {zk }∞

k=1 converges to the point at infinity, we write zn → ∞ as
n → ∞, or
lim zn = ∞.
n→∞

Example 5.3. The sequence (n(1 + i))∞

n=1 is divergent, but is converging to the point at
infinity on the Riemann sphere.
Example 5.4. Let {zk }∞
k=1 be a sequence of complex numbers such that

lim zn = ∞.
n→∞

Consider another sequence of complex numbers defined by wk = −zk . The sequence

{wk }∞
k=1 converges to the point at infinity as well,

lim wn = lim (−zn ) = ∞.

n→∞ n→∞

Definition 5.5. A sequence (zn )∞ n=1 is called a Cauchy sequence if for all ϵ > 0,
there exists an integer N such that

|zm − zn | ≤ ϵ whenever m, n ≥ N.

We note that the absolute value in the above definition is the absolute value for complex
numbers. It is easy to show that a complex sequence is Cauchy if and only if the real and
imaginary parts are real Cauchy sequences. As a result, the basic property of Cauchy
sequence for real numbers extends to the complex case readily. For example, the followings
properties of real sequences are also valid for complex sequences.

Proposition 5.6.
(i) A complex sequence (zn )∞ ∞
n=1 converges if and only if both (Re(zn ))n=1 and
(Im(zn ))∞
n=1 converge.
(ii) A complex sequence (zn )∞ ∞
n=1 converges if and only if (zn )n=1 is Cauchy.

42
 Part (ii) of this proposition says that the set of complex numbers inherits the complete-
ness property of real numbers.

5.2 Complex series

Definition 5.7. Given a sequence of complex numbers (zk )∞

k=1 , we define the infinite
P∞
series k=1 zk by
X∞ Xn
zk ≜ lim zk
n→∞
k=1 k=1

if the limit exists.

As in calculus, the convergence of infinity series is the same as the convergence of

sequence of partial sums.
Example 5.5. (Complex geometric series) Evaluate ∞ k
P
k=1 (0.5i) .
For any finite n, we have
n
X (0.5i)n+1 − 0.5i
(0.5i)k = .
0.5i − 1
k=1

We take limit as n → ∞,
∞
X (0.5i)n+1 − 0.5i −0.5i −1 + 2i
(0.5i)k = lim = = .
n→∞ 0.5i − 1 0.5i − 1 5
k=1

The next two proposition provides two tests for convergence. The first one is a test for
divergence, and the second one is a test for convergence.

Proposition 5.8. (n-th term test) If ∞

P
k=1 zk converges, then |zn | → 0 as n → ∞.
In other words, if (|zn |)n=1 does not converge to 0, then ∞
∞
P
k=1 zk does not converge.

Proof. Suppose ∞
P Pn ∞
k=1 zk converges to w. The sequence of partial sums ( k=1 zk )n=1 con-
verges, and is thus a Cauchy sequence (Prop. 5.6 part (ii)). By the definition of Cauchy
sequence, given any ϵ > 0, there exists an integer N such that
m
X n
X
zk − zk ≤ ϵ
k=1 k=1

for any n, m ≥ N . By picking m = n + 1, we obtain |zn+1 | ≤ ϵ for all n ≥ N .

43
 Proposition 5.9. (Absolute convergence test) Given a sequence of complex numbers
P∞ P∞
(zk )∞
k=1 , if the real infinite series k=1 |zk | converges, then k=1 zk also converges.

Proof. Suppose ∞
P
k=1 |zk | converges. Since | Re(zk )| ≤ |zk | for all k ≥ 1, the real series
P∞ P∞
k=1 | Re(zk )| converges. By the absolute convergence property for real series, k=1 Re(zk )
P∞ P∞
converges. Similarly, k=1 Im(zk ) is convergent. The convergence of k=1 zk then follows
by applying part (i) of Prop. 5.6.

Example 5.6. The series

∞
X 2 + i/n
(5.3)
n2 + 2ni
n=1
is convergent because we can upper bound the magnitude of each term
2 + i/n
≤ c/n2
n2 + 2ni
for some real constant c. Because the series ∞ −2 is convergent, by Prop. 5.9, the
P
n=1 n
series in (5.3) converges.
In view of Prop. 5.9, we make the following definition.

Definition 5.10. We say that a series ∞

P P∞
k=1 zk is absolutely convergent if k=1 |zk |
P∞ P∞
is convergent. If k=1 zk converges but k=1 |zk | does not converge, then we say
that it converges conditionally.

Proposition 5.11. (limit ratio test) Suppose ak , for k = 1, 2, 3, . . . are complex

numbers and limk→∞ |ak+1 /ak | exists and is equal to L.
P∞
1. If L > 1, then k=1 ak diverges;
P∞
2. If L < 1, then k=1 ak converges absolutely.

Proof. First consider the case L > 1. By the meaning of converging to L, which is larger
than 1 by assumption, there exists an integer N such that |ak+1 /ak | > 1 for all k ≥ N .
Therefore
|aN | < |aN +1 | < |aN +2 | < · · ·

44
 P
By applying the n-th term test, we see that ak is divergent.
Next consider the case L < 1. Let ϵ be a small positive real number such that L <
1 − ϵ < 1. By the definition of convergence, there exists an integer N such that
ak+1
<1−ϵ<1
ak
for all k ≥ N . Then, we have

|aN +1 | < |aN |(1 − ϵ)

|aN +2 | < |aN +1 |(1 − ϵ) < |aN |(1 − ϵ)2 ,

and in general
|aN +k | < |aN |(1 − ϵ)k
for k ≥ 1. Therefore, for each positive number m, we have
m
X m
X
|aN +j | ≤ |aN | (1 − ϵ)j .
j=0 j=0
P∞
Since j=0 (1 − ϵ)j is finite, the sequence
m
X
∞
|aN +j | m=0
j=0

is a bounded and monotonically increasing sequence. Hence it must be convergent. By

Prop. 5.9, ∞
P P∞
j=0 aN +j is also convergent. This proves that k=1 ak is convergent.

We quote two properties of absolutely convergent series.

Proposition 5.12. If ∞
P P∞
k=0 ak and k=0 bk are absolutely convergent series, then

∞
X ∞
 X  X∞
ak bk = ck ,
k=0 k=0 k=0

where
ck = a0 bk + a1 bk−1 + · · · + ak−1 b1 + ak b0 .

Proposition 5.13. If a series converges absolutely, then a series obtained by re-

arranging the terms converges and converges to the same limit.

45
 The proofs can be found in [Rudin, Theorem 3.50] and [Rudin, Theorem 3.55]. Although
the corresponding theorems in [Rudin] are stated for real series, but the proofs works for
the complex case in a straightforward manner.

5.3 Complex exponential function

Definition 5.14. A complex power series centered at the origin is a series in the
form ∞ k
P
k=0 ak z , where ak ∈ C.

We investigate a relationship between complex conjugate and complex power series.

P∞ k
Theorem 5.15. Suppose complex power series k=0 ck z converges at z = z0 . Then
∞
X ∗ ∞
X
ck z0k = c̄k z¯0 k .
k=0 k=0

Proof. Let L be the limit of ∞ k

P
k=0 ck z0 . The value L satisfies the defining property: for
any positive real number ϵ, there exists a positive integer N such that
n
X
ck z0k − L ≤ ϵ
k=0

for all n ≥ N . We can take the conjugate of the expression inside the absolute value and
see that
Xn
c̄k z̄0k − L̄ ≤ ϵ
k=0

holds for any n ≥ N . Hence by the definition of limit, we can say that
∞
X ∞
X ∗
∗
c̄k z̄0k =L = ck z0k .
k=0 k=0

46
 Definition 5.16. Define the complex exponential function ez by the complex power
series
∞
X zk z z2 z3
ez ≜ exp(z) ≜ =1+ + + + ···
k! 1! 2! 3!
k=0

for z ∈ C.

When z is a real number, the above definition is the same as the power series of the
exponential function with real domain. The complex exponential function is well-defined
for all z ∈ C, i.e., the power series converges for all z, because for any z ̸= 0, we can apply
the limit ratio test and check that the ratio of two consecutive terms
z n+1 /(n + 1)! |z|
n
=
z /n! n+1
converges to 0 as n → ∞.
We establish below a fundamental property of the function exp(z).

Theorem 5.17. For any complex numbers z1 and z2 ,

exp(z1 + z2 ) = exp(z1 ) exp(z2 ).

Proof. Given any z1 and z2 in C, the complex power series that determine ez1 and ez2 are
absolutely convergent by the ratio test in Prop. 5.11. We can also see this by taking the
absolute value of each term
|z1 |2 |z1 |3 |z1 |4 |z1 |5
1 + |z1 | + + + + + ···
2! 3! 4! 5!
and observe that this is equal to e|z1 | .
Apply Prop. 5.12 to ak = z1k /k! and bk = z2k /k!. For k ≥ 0, the coefficient ck is
k
X z1ℓ z2k−ℓ
ck =
ℓ! (k − ℓ)!
ℓ=0
k  
1 X k ℓ k−ℓ
= z z
k! ℓ 1 2
ℓ=0
1
= (z1 + z2 )k .
k!

47
Hence
∞
X (z1 + z2 )k
exp(z1 ) exp(z2 ) = ≜ exp(z1 + z2 ).
k!
k=0

Using this theorem we can derive some immediate corollaries.

Corollary 5.18. For any z ∈ C,

ˆ e−z = (ez )−1 ;

ˆ ez ̸= 0 for all z ∈ C.

Proof. In the previous proposition, let z1 = z and z2 = −z,

ez e−z = ez−z = e0 = 1.

Therefore (ez )−1 = e−z .

To see that ez ̸= 0 for all z ∈ C, we note that ez = 1/e−z and e−z is a complex number.
Hence ez cannot be equal to zero for all z ∈ C.

The next corollary is about the modulus of exponential function.

Corollary 5.19. For any complex number z = x + iy,

ˆ |ez | = ex ;

ˆ |eiy | = 1.

Proof.

|ez |2 = exp(z) · exp(z)∗ = exp(z) · exp(z ∗ ) = exp(z + z ∗ ) = exp(2x) = (ex )2 .

We can take square root of both sides, because |ez | and ex are both real numbers, and prove
that |ez | = ex .
We can now let z = iy to get |eiy | = e0 = 1.

48
 Although the complex exponential function is available from the numpy library, we can
implement it from scratch using power series. The function my exp below add the first 50
terms in the power series that defines the complex exponential function, starting from the
higher degree to the lower degree.
# Computing complex exponential function by power series
def my_exp ( z ) :
answer =1
for k in range (50 ,0 , -1) :
answer = answer * z / k +1
return answer

5.4 Euler’s formula

The main relationship between the complex exponential and the sinusoidal functions is the
next theorem.
Theorem 5.20 (Euler’s formula). For any real number θ, we have

eiθ = cos(θ) + i sin(θ). (5.4)

Proof. The proof of Euler’s formula is by re-arranging the terms in the power series that
defines eiθ . This is permissible because
iθ i2 θ2 i3 θ3 i4 θ4 i5 θ5
eiθ = 1 + + + + + + ···
1 2! 3! 4! 5!
is an absolutely convergent power series. By separating the real and imaginary parts by
terms re-arrangement, we obtain
 θ2 θ4   θ3 θ5 
eiθ = 1 − + − ··· + i θ − + − ···
2! 4! 3! 5!
the first term is precisely the power series expansion of cos(θ), and the second term is i
times the power series expansion of sin(θ).

Combining Theorems 5.17 and 5.20, we can evaluate the complex exponential function
in terms of the real exponential function and the real sine and cosine functions,

ex+iy = ex (cos(y) + i sin(y)). (5.5)

49
 √
Example 5.7. We can compute e 2+i by
√ √
2+i 2
e =e (cos(1) + i sin(1)).

Example 5.8. When z = πi, we obtain

eπi = cos(π) + i sin(π) = −1.

Remark. The Euler’s formula is a vast generalization of DeMoivre formula. We can recover
DeMoivre formula from Euler’s formula by considering einθ , which can be written as

einθ = cos(nθ) + i sin(nθ).

By Prop. 5.17, we can also write it as

einθ = (eiθ )n = (cos θ + i sin θ)n .

Application to engineering and physics. We can use the complex exponential function
to represent a sinusoidal function. Given amplitude A, frequency ω and phase θ, we can
write
A cos(ωt + θ) = Re(Aei(ωt+θ) ) = Re(eiωt z0 ),

where z0 is a complex number z0 = Aeiθ .

Example 5.9. Consider n sinusoidal functions with the same frequency ω,

Ak cos(ωt + θk ),

where amplitudes Ak and phase θk , for k = 1, 2, . . . , n, are real numbers. The sum of them
is another sinusoidal function with the same frequency ω. To find the amplitude and phase
of the sum, we can simply add the n complex numbers Ak eiθk , for k = 1, 2, . . . , n,
n
X
Ak eiθk ≜ z0 .
k=1

The modulus and argument of z0 is then the amplitude and the phase of the resulting
sinusoidal function
n
X n
X
Ak cos(ωt + θk ) = Re eiωt Ak eiθk = Re(eiωt z0 ) = |z0 | cos(ωt + arg(z0 )).

k=1 k=1

50
6 Complex trigonometric and log function
Summary

ˆ Complex trigonometric functions

ˆ Complex log function

ˆ Complex power function

6.1 Complex trigonometric functions

We extend the definition of trigonometric function and hyperbolic trigonometric functions
to the complex plane.

Definition 6.1. For z ∈ C, define the complex sine complex cosine, complex hyper-
bolic sine, and complex hyperbolic cosine functions by
∞
X z 2n+1
sin(z) ≜ (−1)n ,
(2n + 1)!
n=0
∞
X z 2n
cos(z) ≜ (−1)n ,
(2n)!
n=0
∞
X z 2n+1
sinh(z) ≜ ,
(2n + 1)!
n=0
∞
X z 2n
cosh(z) ≜ .
(2n)!
n=0

By the ratio test, they all converges absolutely on the whole complex plane. They
reduce to the real trigonometric and real hyperbolic trigonometric function when z is a real
number. In terms of the complex sine and complex cosine function, we can express the
Euler’s formula as
Theorem 6.2. For z ∈ C,

eiz = cos(z) + i sin(z). (6.1)

51
Proof. The proof is basically the same as the proof of Theorem 5.20. We just need to apply
the definition of cos(z) and sin(z) for complex number z.
Starting from the power series expansion of eiz

z 2 iz 3 z 4 iz 5 z 6
eiz = 1 + iz − − + + − + ···
2! 3! 4! 5! 6!
because of absolute convergence, we can re-arrange the terms and separate the terms with-
out i and the terms with i,
∞
X (iz)n
eiz ≜
n!
n=0
∞ ∞
X z 2k X z 2k+1
= (−1)k +i (−1)k
(2k)! (2k + 1)!
k=0 k=0

= cos(z) + i sin(z).

The last equality is true by the definition of complex sine and complex cosine functions.

Using Theorem 6.2, we can express sine function and cosine function in terms of the
complex exponential function.

Theorem 6.3. For any z ∈ C,

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

Proof. The complex cosine function is an even function, i.e., cos(−z) = cos(z), because
the power series that define cos(z) consists of even powers. Similarly, because all terms in
the power series that define sin(z) have odd powers, the complex sine function is an odd
function, i.e., sin(−z) = − sin(z). This gives

e−iz = cos(−z) + i sin(−z) = cos(z) − i sin(z) (6.2)

By adding (6.2) to (6.1), we get

eiz + e−iz = 2 cos(z).

52
By subtracting (6.2) from (6.1), we get

eiz − e−iz = 2i sin(z).

The result in Theorem 6.3 holds for any complex numbers. In particular, when we
restrict z to a real number θ, this yields

eiθ + e−iθ
cos(θ) = ,
2
eiθ − e−iθ
sin(θ) = .
2i
Remark. Some people take the two identities in Theorem 6.3 as the definition of complex
sine and complex cosine, and then deduce Definition 6.1 as a corollary. Which definitions
we adopt is a matter of taste. The resulting theory is the same.
Example 6.1. Evaluate sin(i). By Theorem 6.3,

ei·i − e−i·i e−1 − e1 e1 − e−1

sin(i) = = =i = i sinh(1).
2i 2i 2
The trigonometric functions in the numpy library can compute complex number. We can
verify the calculations in this example as follows.
In [0]: from numpy import sin , sinh
In [1]: sin (1 j )
Out [1]: 1. 17 52 01 19 364 38 01 4 j

In [2]: sinh (1)

Out [2]: 1. 17 52 01 19 364 38 01 4

Example 6.2. For any real number a, we can evaluate cos(π + ai) by using the formula in
Theorem 6.3,

ei(π+ai) + e−i(π+ai) e−a+iπ + ea−iπ e−a + ea

cos(π + ai) = = =− = − cosh(a).
2 2 2
The next theorem states the relationship between trigonometric functions and hyper-
bolic functions.

53
 Theorem 6.4. For all complex number z ∈ C,

sin(iz) = i sinh(z),
cos(iz) = cosh(z).

Proof. The first identity can be proved by considering the power series expansion of sin(iz),
∞ 2k+1 ∞ 2k 2k+1 ∞
X
k (iz)
X
k i z
X z 2k+1
sin(iz) ≜ (−1) =i (−1) =i = i sinh(z).
(2k + 1)! (2k + 1)! (2k + 1)!
k=0 k=0 k=0

The proof of the second identity is similar.

In particular, for any real number x, we have

sin(ix) = i sinh(x), and cos(ix) = cosh(x).

Some trigonometric identities in the real case continue to hold when the variables are
complex numbers.

Theorem 6.5. For any complex numbers z and w, the following hold:

ˆ cos2 z + sin2 z = 1.

ˆ sin(z + w) = sin z cos w + sin w cos z.

ˆ cos(z + w) = cos z cos w − sin z sin w.

Proof. We prove the first one. The proofs of the rest are similar.

L.H.S. = cos2 z + sin2 z

 eiz + e−iz 2  eiz − e−iz 2
= +
2 2i
1 iz
= [(e + e−iz )2 − (eiz − e−iz )2 ]
4
1
= [e2iz + 2 + e−2iz − e2iz + 2 − e−2iz ]
4
= 1 = R.H.S.

54
6.2 Complex log function
The complex exponential function has a complex period 2πi; that is, for any z ∈ C,

ez+2πki = ez , for k ∈ Z.

As a result, the inverse function of ez is multi-valued.

Notation: We will use the notation ln x for the natural log function with positive real
number as input.
Example 6.3. Find z ∈ C such that ez = 1 + i.
The complex number 1 + i in polar coordinates is
√
2(cos(π/4) + i sin(π/4)).

We want to find all complex numbers x + iy such that

√
ex+iy = ex (cos y + i sin y) = 2(cos(π/4) + i sin(π/4)).

By comparing the moduli and arguments, we have

1
x= ln(2)
2
y = π/4 + 2πk, for k = 0, ±1, ±2, . . .

Hence
1
log(1 + i) = ln(2) + i(π/4 + 2πk)
2
for k ∈ Z.
In general, given a complex number r(cos θ + i sin θ) in polar form, the complex log of
r(cos θ + i sin θ) can take values
ln r + i(θ + 2πk)

for any integer k. In general, we have the following

Definition 6.6. For nonzero w ∈ C, the complex log function is defined as

log(w) ≜ ln |w| + i(arg(w) + 2πk)

where k = 0, ±1, ±2, . . .

55
 The function arg(w) in Definition 6.6 is the multi-valued argument function. If we
specify a range for the angle by take the principal argument (Definition 2.3), we have a
uniquely defined function.

Definition 6.7. The function

Log(w) ≜ ln |w| + i Arg(w),

is called the principal complex log function. We will denote the principal log function
by Log(z).

The complex log function extends the domain of the real log function to the whole
complex plane except the origin. We note that it is not possible to define the log(0) even
if we extend the number system to complex numbers, because the complex exponential
function is never equal to zero. One way to see it is by the following general fact

∀z ∈ C, ez e−z = ez−z = e0 = 1.

This will yield a contradiction if ez = 0.

Using the complex log function we can compute the log of a negative number.
Example 6.4. Compute the principal log of −10, using a principal argument function with
range [0, 2π).
We first write −10 as
−10 = 10(cos π + i sin π)

If ex+iy = 10(cos π + i sin π), we must have 10 = ex , giving x = ln 10. The imaginary part
y equals π, which is the unique argument in the range [0, 2π). Therefore, the principal log
of −10 is
Log(−10) = ln(10) + iπ.

The log function from the numpy library in Python is able to compute the complex log
function. However, we need to activate the mode of complex computation by providing a
complex number as the input. If we directly ask Python to compute log(−10), we will get
a warning, saying that there is a type mismatch.
In [0]: from numpy import log
In [1]: log ( -10)

56
RuntimeWarning : invalid value encountered in log
Out [1]: nan

One way to make it works is to manually convert the input to a complex number. Then
we can compare with log(10) + iπ, and verify that the answer is correct.
In [2]: from math import log , pi
In [3]: log ( complex ( -10) )
Out [3]: ( 2 . 3 0 2 5 8 5 0 9 2 9 9 4 0 4 6 + 3 . 1 4 1 5 9 2 6 5 3 5 8 9 7 9 3 j )

In [4]: log (10) +1 j * pi

Out [5]: ( 2 . 3 0 2 5 8 5 0 9 2 9 9 4 0 4 6 + 3 . 1 4 1 5 9 2 6 5 3 5 8 9 7 9 3 j )

An alternate way is to use the complex log function in the cmath library. It automatically
converts the input value to the complex number data type.
In [2]: from cmath import log
In [3]: log ( -10)
Out [3]: ( 2 . 3 0 2 5 8 5 0 9 2 9 9 4 0 4 6 + 3 . 1 4 1 5 9 2 6 5 3 5 8 9 7 9 3 j )

6.3 Complex power function

We can define the complex power function via the complex log function.

Definition 6.8. Given a nonzero complex number z and a complex number w, we

define the complex power z w by

z w ≜ ew log(z)

where log(z) is the complex log function.

Example 6.5. Compute 2i .

2i ≜ exp(i log 2) = exp(i(ln 2 + i2πk))

= exp(−2πk + i log 2)
= e−2πk (cos(ln 2) + i sin(ln 2)),

57
where k can take any integer as its value. We can consider the value when k = 0 as the
principal value. If we want to write down just one answer, we can write

2i = cos(ln 2) + i sin(ln 2).

The following Python commands compute the principal value of 2i .

In [0]: from numpy import cos , sin , log
In [1]: 2**1 j
Out [1]: ( 0 . 7 6 9 2 3 8 9 0 1 3 6 3 9 7 2 1 + 0 . 6 3 8 9 6 1 2 7 6 3 1 3 6 3 4 8 j )

In [2]: cos ( log (2) ) +1 j * sin ( log (2) )

Out [2]: ( 0 . 7 6 9 2 3 8 9 0 1 3 6 3 9 7 2 1 + 0 . 6 3 8 9 6 1 2 7 6 3 1 3 6 3 4 8 j )

We can use the complex power function to compute the n-th roots of a complex numbers.
We can compare this method with the DeMoivre formula in (2.4).
Example 6.6. Calculate the cube roots of 2i by (2i)1/3 ≜ exp(log(2i)/3).
We compute


log(2i) = ln 2 + i π/2 + 2πk , for k ∈ Z
log(2i) ln 2

= + i π/6 + 2πk/3 .
3 3
Exponentiating both sides, we get
log(2i)
√ 3
 π 2πk
π 2πk

exp = 2 cos + + i sin +
3 6 3 6 3
√
for k = 0, 1, 2. There are three answers, all with magnitude 3 2.
The next example has complex base and complex exponents at the same time.
Example 6.7. Compute ii by calculating ei log i .
 π 
exp(i log i) = exp i(log 1 + i( + 2πk))
2
π

= exp − ( + 2πk) ,
2
where k ∈ Z. All possible values are real numbers. By taking k = 0, we can say that the
principal value of ii is e−π/2 .
We double check the answer by Python.

58
In [0]: from numpy import exp , pi
In [1]: 1 j **1 j
Out [1]: ( 0 . 2 0 7 8 7 9 5 7 6 3 5 0 7 6 1 9 3 + 0 j )

In [2]: exp ( - pi /2)

Out [2]: 0 . 20 7 8 79 5 7 63 5 0 76 1 9 3

6.4 Inverse of complex trigonometric functions

Inverse trigonometry function can be computed by solving quadratic equation.
Example 6.8. Find all complex numbers z such that cos(z) = 2.
Recall from Theorem 6.3 that we can express cos(z) as
eiz + e−iz
cos(z) = .
2
Let w = eiz . This amounts to solve
w + w−1
= 2.
2
√
This is a quadratic equation w2 − 4w + 1 = 0. The roots are 2 ± 3. Hence
√
eiz = 2 ± 3.

By taking the complex log function, we obtain

 √  √
z = −i ln(2 ± 3) + 2πik = −i ln(2 ± 3) + 2πk,

where k ∈ Z.
The arccos function in Python is able to compute the complex arc cosine function. We
√
can check that it returns −i log(2 + 3) as the answer.
In [0]: from numpy import arccos , log
In [1]: arccos (2+0 j )
Out [1]: -1.3169578969248166 j

In [2]: - log (2+ sqrt (3) ) *1 j

Out [2]: ( -0 -1.3169578969248166 j )

59
 √
However, as we have shown in the example, the other solution is −i log(2 − 3), which
differs from the first one by a minus sign. We can verify this by the following Python script.
In [3]: - log (2 - sqrt (3) ) *1 j
Out [3]: 1. 31 69 57 89 692 48 16 4 j

In [4]: cos ( - log (2 - sqrt (3) ) *1 j )

Out [4]: (1.9999999999999996 -0 j )
In [5]: cos ( - log (2+ sqrt (3) ) *1 j )
Out [5]: (1.9999999999999998 -0 j )

Example 6.9. Solve cot(z) = 2i for z ∈ C.

This is equivalent to solving
eiz + e−iz 2i
= 2i.
2 eiz − e−iz
By letting w = eiz , we can simplify it to
w + w−1
=2
w − w−1
w2 + 1
= 2.
w2 − 1
√
The values for w is ± 3. Then, we can obtain z by taking complex log
√ √ √
z = −i log(± 3) = −i(ln 3 + iπk) = πk − i ln 3,

where k is any integer.

We double check the answer by Python. Because the arc cotangent function is not
available in Python, we import the arc tangent function instead.
In [0]: from numpy import log , arctan

In [1]: -1 j * log (3**(1/2) )

Out [1]: -0.5493061443340548 j

In [2]: arctan (1/2 j )

Out [2]: -0.5493061443340549 j

60
7 Complex differentiability
Summary

ˆ Limit and continuity

ˆ Approximation by linear function

ˆ Cauchy-Riemann equations

In order to compute the derivative of a function at a point, we need to make a small

difference in the domain and investigate the resulting change in the function value. In
order to be able to have enough elbow room, we usually assume that the complex function
is defined in an open set. In complex analysis, the term domain/region is reserved to mean
an open and connected set in C.
We can understand the term “connected” as “path-connected”, i.e., any two points in
the domain are connected by a path.

7.1 Limit and continuity of complex function

Definition 7.1. A function f : C → C is said to be continuous at z0 if for any ϵ > 0,

there exists δ > 0 such that

|z − z0 | < δ ⇒ |f (z) − f (z0 )| < ϵ.

A function f is said to be continuous in a domain D if f is continuous at every point

in D.

By consider the real and imaginary part separately, we can prove the following

Theorem 7.2. A complex function f is continuous if and only if the real and imag-
inary parts are continuous.

The proof of this theorem is straightforward and is omitted.

Example 7.1. Show that f (z) = 1/z is continuous in the domain C \ {0}.
Suppose z = x + iy and z ̸= 0.
1 1 x − iy
= = 2
z x + iy x + y2

61
The real part of f (z) is x/(x2 + y 2 ) and the imaginary part is −y/(x2 + y 2 ). Both of them
are continuous functions in the domain C \ {0}. Hence f (z) is continuous by the previous
theorem.
Another way to understand continuity is by means of sequence. If a complex function
f is continuous at z0 , and if (zk )∞
k=1 is a sequence of complex numbers converging to z0 ,
∞
then the image points (f (zk ))k=1 should converge to f (z0 ). We can approach z0 from any
direction, and the limit of the sequence should equal f (z0 ).
The third way to define continuity of a function is by limit of functions. Given a complex
function g(z) and a complex number z0 , we say that the following limit

lim g(z)
z→z0

and is equal to L if for any ϵ > 0, there exists δ > 0 such that

0 < |z − z0 | < δ =⇒ |g(z) − L| < ϵ.

We note that in this definition the function g(z) need not be defined at z0 . With this
notation, we have

Lemma 7.3. A complex function f (z) is continuous at z0 iff

(i) f (z0 ) is well defined,
(ii) the limit limz→z0 f (z) exists, and
(iii) limz→z0 f (z) = f (z0 ).

The limit for complex function behaves in the same as the limit of vector function.
Hence, complex limit has the same properties as in multi-variable calculus.

7.2 Complex differentiable function

By regarding a complex function f (z) as a two-dimensional vector field,

f (x + iy) = u(x, y) + iv(x, y).

When we say that f (x+iy) is real differentiable, if the vector-valued function (u(x, y), v(x, y))
is differentiable as in multivariable calculus. By definition, if (u(x, y), v(x, y)) is real differ-
entiable at a point (x0 , y0 ), we can approximate it by linear function,
" # " # " #" #
u(x0 + ∆x, y0 + ∆y) . u(x0 , y0 ) ux (x0 , y0 ) uy (x0 , y0 ) ∆x
= + (7.1)
v(x0 + ∆x, y0 + ∆y) v(x0 , y0 ) vx (x0 , y0 ) vy (x0 , y0 ) ∆y

62
when (∆x, ∆y) is small. The entries in the 2 × 2 matrix are the partial derivatives of u and
.
v evaluated at (x0 , y0 ). The symbol “=” means that the difference between the left-hand
side and the right-hand side of (7.1) is converging to zero faster than the linear term, i.e.,

∥Difference between L.H.S and R.H.S. of (7.1)∥

lim p = 0.
(∆x,∆y)→(0,0) ∆x2 + ∆y 2

Example 7.2. The function f (z) defined by x2 y + i(x + y) is real differentiable. Partial
derivatives of the real part u(x, y) = x2 y and imaginary part v(x, y) = x + y exist, and we
have " # " # " #" #
(x0 + ∆x)2 (y0 + ∆y) . x20 y0 2x0 y0 x20 ∆x
= + (7.2)
x0 + ∆x + y0 + ∆y x 0 + y0 1 1 ∆y
for any x0 and y0 .

Definition 7.4. A complex function f is said to be complex differentiable at z0 if

the limit
f (z0 + ∆z) − f (z0 )
lim (7.3)
∆z→0 ∆z
exists. This is equivalent to requiring that
.
f (z0 + ∆z) = f (z0 ) + w0 · ∆z, (7.4)

were w0 is the complex constant that equals to the limit in (7.3). The limit in (7.3)
is denoted by f ′ (z0 ), if the limit exists.

.
The symbol = in (7.4) means that

f (z0 + ∆z) = f (z0 ) + w0 · ∆z + o(∆z),

where o(∆z) is a function such that lim∆z→0 |o(∆z)|/|∆z| = 0.

Example 7.3. Consider the function

f (z) = z 3 = (x + iy)3 = (x3 − 3xy 2 ) + i(3x2 y − y 3 ).

The real and imaginary parts are u(x, y) = x3 − 3xy 2 and v(x, y) = 3x2 y − y 3 , respectively.

63
The partial derivatives of u and v are

ux (x, y) = 3x2 − 3y 2 ,
uy (x, y) = −6xy,
vx (x, y) = 6xy,
vy (x, y) = 3x2 − 3y 2 .

Suppose we fix a base point (x0 , y0 ) = (2, 1). The linear approximation in (7.1) at (2, 1)
can be written as
" # " # " #" #
u(2 + ∆x, 1 + ∆y) . 2 9 −12 ∆x
= + .
v(2 + ∆x, 1 + ∆y) 11 12 9 ∆y
For general z = x + iy, the 2 × 2 matrix consisting of the partial derivatives is
" #
3x2 − 3y 2 −6xy
.
6xy 3x2 − 3y 2
We can use complex arithmetic to realize the linear approximation by

f (z + ∆z) = f (z) + (3x2 − 3y 2 + i(6xy)) · (∆x + i∆y)

= f (z) + (3z 2 )∆z.

Hence, we can conclude that z 3 is complex differentiable and the complex derivative is 3z 2 .

As in the real case, a function that is complex differentiable at a point is continuous at

that point (See Lemma 7.3).

Theorem 7.5. If f (z) is complex differentiable at z0 , then f (z) is continuous at z0 .

Proof. Suppose f (z) is complex differentiable at z0 . By the definition of complex differen-

tiability, we can find a complex number w0 such that

f (z0 + h) = f (z0 ) + w0 · h + o(h),

where h is a complex variable and o(h) is a function with the property that |o(h)/h| → 0
as h → 0. If we do not divide by h, the function o(h) converges to zero as well, because
o(h) o(h)
lim o(h) = lim h = lim · lim h = 0 · 0 = 0.
h→0 h→0 h h→0 h h→0

64
Therefore
lim f (z0 + h) = f (z0 ) + lim w0 · h + lim o(h) = f (z0 ).
h→0 h→0 h→0

7.3 Cauchy-Riemann equations

We derive a necessary condition for complex differentiability.

Theorem 7.6 (Cauchy-Riemann equations). Suppose f (x + iy) = u(x, y) + iv(x, y)

is complex differentiable at z0 = x0 + iy0 (see Definition 7.4), where z0 is in the
domain of f (z). Then the partial derivatives ux , uy , vx and vy at z0 exist, and they
satisfy
ux (x0 , y0 ) = vy (x0 , y0 ), and uy (x0 , y0 ) = −vx (x0 , y0 ).

Proof. Let f be a complex function that is complex differentiable at z0 = x0 + iy0 . The

limit in (7.3) that defines the complex derivative does not depend on how we approach the
point z0 . We can approach z0 horizontally or vertically, and the results must be the same
if the function is complex differentiable.
Let ∆z = ∆x and take ∆x → 0.
u(x0 + ∆x, y0 ) + iv(x0 + ∆x, y0 ) − u(x0 , y0 ) − iv(x0 , y0 )
lim
∆x→0 ∆x
u(x0 + ∆x, y0 ) − u(x0 , y0 ) v(x0 + ∆x, y0 ) − v(x0 , y0 )
= lim + i lim
∆x→0 ∆x ∆x→0 ∆x
∂u ∂v
= (x0 , y0 ) + i (x0 , y0 ).
∂x ∂x
Next suppose ∆z = i∆y and take ∆y → 0.
u(x0 , y0 + ∆y) + iv(x0 , y0 + ∆y) − u(x0 , y0 ) − iv(x0 , y0 )
lim
∆y→0 i∆y
u(x0 , y0 + ∆y) − u(x0 , y0 ) v(x0 , y0 + ∆y) − v(x0 , y0 )
= lim + i lim
∆y→0 i∆y ∆y→0 i∆y
∂u ∂v
= −i (x0 , y0 ) + (x0 , y0 ).
∂y ∂y
By equating real and imaginary parts of the two limits, we get ux (x0 , y0 ) = vy (x0 , y0 ) and
uy (x0 , y0 ) = −vx (x0 , y0 ).

65
 The function in Example 7.2 is real differentiable everywhere but not complex differen-
tiable. As a matter of fact, the function in Example 7.2 fails the Cauchy-Riemann equation
for all real numbers x0 and y0 , and thus is not complex differentiable anywhere.
A necessary and sufficient condition for complex differentiability at a fixed point z0
is given in Theorem 7.10 in the appendix of this lecture. It states that a function is
complex differentiable at a point if and only if (i) it is real differentiable at this point
and (ii) the Cauchy-Riemann equations are satisfied. Checking whether a real function is
differentiable from definition is not an easy task. We can apply the following basic result
from multivariable calculus (See Theorem 9.21 in [Rudin]).

Theorem 7.7. Suppose f⃗(x, y) = (u(x, y), v(x, y)) is a two-dimensional vector field.
A sufficient condition for f⃗ to be real differentiable at (x0 , y0 ) is (i) partial derivatives
ux , uy , vx and vy exists in a neighborhood of (x0 , y0 ), and (ii) the partial derivatives
ux , uy , vx and vy are continuous at (x0 , y0 ).

This theorem gives a sufficient condition for real differentiability. Combining this the-
orem with Theorem 7.10, we have the following sufficient condition for complex differen-
tiability. The proof idea is, given conditions 1 and 2 in Theorem 7.7, the vector function
f⃗(x, y) = (u(x, y), v(x, y)) is real differentiable at (x0 , y0 ). Because the "Cauchy-Riemann
#
a −b
equations are satisfied at (x0 , y0 ), the 2 × 2 matrix in (7.1) is in the form . We can
b a
thus write the linear approximation in (7.1) in terms of complex multiplication.

Theorem 7.8 (A sufficient condition for complex differentiability). A complex func-

tion f is complex differentiable at z0 if

1. The partial derivatives ux , uy , vx , vy exists in a neighborhood of z0 ;

2. ux , uy , vx , vy are continuous at z0 ;

3. Cauchy-Riemann equations are satisfied at z0 .

From the proof of Theorem (7.6), the complex derivative of f (z) can be computed using
the partial derivatives of the real and imaginary part of f (z).

66
 Corollary 7.9. If a complex function f (z) = u(x, y) + iv(x, y) is complex differen-
tiable at z0 = x0 + iy0 , the complex derivative at z0 can be computed by

ux (x0 , y0 ) + ivx (x0 , y0 ),

or
vy (x0 , y0 ) − iuy (x0 , y0 ).

7.4 Appendix: A necessary and sufficient condition for complex differ-

entiability at a point

Theorem 7.10. Suppose f (z) = u(x, y) + iv(x, y) is a complex function defined on

a domain D which contains z0 = x0 + iy0 . Then f (z) is complex differentiable at
z0 if and only if the real vector field (u(x, y), v(x, y)) is real-differentiable at (x0 , y0 )
and the partial derivatives satisfy the Cauchy-Riemann equations

ux = vy , uy = −vx

at (x0 , y0 ).

Proof. (⇒) Suppose f (z) is complex differentiable at z0 . This means that

f (z0 + h) − f (z0 )
lim
h→0 h
exists. Let the limit be a + ib. When the magnitude of h is small, we can approximate
f (z0 + h) by
f (z0 + h) = f (z0 ) + (a + ib) · h + o(h), (7.5)

where o(h) represents a complex function of h that satisfies

o(h)
lim = 0.
h→0 h

By expressing f (z) in terms of the real and imaginary parts and h = ∆x + i∆y, we can

67
write (7.5) as

u(x0 + ∆x, y0 + ∆y) = u(x0 , y0 ) + a∆x − b∆y + Re(o(∆x + i∆y)),

v(x0 + ∆x, y0 + ∆y) = v(x0 , y0 ) + a∆y + b∆x + Im(o(∆x + i∆y)).

In terms of matrix, we have

" # " # " #" # " #
u(x0 + ∆x, y0 + ∆y) u(x0 , y0 ) a −b ∆x Re(o(∆x + i∆y))
= + + . (7.6)
v(x0 + ∆x, y0 + ∆y) v(x0 , y0 ) b a ∆y Im(o(∆x + i∆y))
p
The length of the error term approaches to zero faster than ∆x2 + ∆y 2 , because
p
Re(o(∆x + i∆y))2 + Im(o(∆x + i∆y))2 |o(h)| o(h)
p = =
2
∆x + ∆y 2 |h| h

converges to zero as h → 0. This proves that the vector field (u(x, y), v(x, y)) is real
differentiable at (x0 , y0 ). " #
a −b
The fact that the 2 × 2 matrix in (7.6) is in the special form implies that the
b a
Cauchy-Riemann equations hold at (x0 , y0 ).
(⇐) In the reverse direction, we suppose that the vector field (u(x, y), v(x, y)) is real-
differentiable at (x0 , y0 ), and the Cauchy-Riemann equations are satisfied at (x0 , y0 ). From
the definition of real-differentiability, the quantity
" # " # " #" #
1 u(x0 + ∆x, y0 + ∆y) u(x0 , y0 ) ux uy ∆x
p − −
∆x2 + ∆y 2 v(x0 + ∆x, y0 + ∆y) v(x0 , y0 ) vx vy ∆y

approaches zero as the vector (∆x, ∆y) approaches the origin. We thus have a vector
function g(∆x, ∆y) with variables ∆x and ∆y, satisfying equation
" # " # " #" #
u(x0 + ∆x, y0 + ∆y) u(x0 , y0 ) ux uy ∆x
= + + g(∆x, ∆y) (7.7)
v(x0 + ∆x, y0 + ∆y) v(x0 , y0 ) vx vy ∆y

and
∥g(∆x, ∆y)∥
p →0
∆x2 + ∆y 2
p
as ∆x2 + ∆y 2 → 0.

68
 Moreover, from the Cauchy-Riemann equation, we can write (7.7) as
" # " # " #" #
u(x0 + ∆x, y0 + ∆y) u(x0 , y0 ) a −b ∆x
= + + g(∆x, ∆y)
v(x0 + ∆x, y0 + ∆y) v(x0 , y0 ) b a ∆y

for some real numbers a and b.

If we identify the vector function g(∆x, ∆y) with a complex-valued function e(∆x+i∆y)
whose real and imaginary parts equal the first and second component of g(∆x, ∆y), we get

g(∆x, ∆y) = (Re(e(∆x + i∆y), Im(e(∆x + i∆y)),

and
f (z0 + h) = f (z0 ) + (a + ib) · h + e(h).

Since limh→0 |e(h)|/|h| = 0, we can conclude that f (z) is complex differentiable at z0 =

x0 + iy0 .

69
8 Holomorphic functions
Summary:

ˆ Example of holomorphic functions

ˆ Differentiation rules

It turns out that being differentiable at an isolated point (as defined in Definition 7.4)
is not a very useful property. Much more structure can be seen if we require a function to
be complex differentiable in an open neighborhood of a point.

Definition 8.1. A complex function f (z) is said to be holomorphic at a point z0 if

there is a neighborhood of z0 such that f is complex differentiable at every point in
the neighborhood. A function is said to be entire if it is complex differentiable at
df
every point in C. We denote the complex derivative of f by dz when f is holomorphic.

If we can show that a function is complex differentiable at every point in the domain of
definition, then it is holomorphic in the domain.
Remark. In this notes, we will use the term “analytic” to refer to complex functions that
are locally representable by complex power series. We will show later that this is equivalent
to the condition of complex differentiable at every point in the domain.
Using Theorem 7.8, we can check that a complex function is holomorphic if it satisfies
the Cauchy-Riemann equations and has continuous first partial derivatives throughout the
domain of definition.

8.1 Examples
The following examples illustrate how to determine whether a function is complex differen-
tiable (or the contrary). We start with the simplest example.
Example 8.1. For any complex constant a, the function f (z) = az is an entire function, i.e.,
it is complex differentiable at every point in the complex plane, and the derivative is equal
to a. We can see this by checking the definition of complex derivative.

(a(z0 + h)) − az0

− a = 0.
h

70
Trivially, we have limh→0 a(z0 +h)−az
h
0
= a.
Likewise, we can see that the derivative of a translation function z 7→ z + b is complex
differentiable, and the derivative is equal to 1.
Example 8.2. The conjugate function f (z) = z̄ is not complex differentiable anywhere. It
is because ux = 1 and vy = −1, and ux ̸= vy at any point in C. It fails the CR equations
everywhere.
Example 8.3. The square function f (z) = z 2 is entire with derivative 2z. The real and
imaginary parts are u(x, y) = x2 − y 2 and v(x, y) = 2xy, respectively. We check that the
partial derivatives.
ux = 2x, uy = −2y, vx = 2y, vy = 2x
exist and are continuous at every point in C. This check conditions 1 and 2 in Theorem 7.8.
Furthermore, the Cauchy-Riemann equalities hold, because

ux = vy = 2x, and uy = −vx = −2y.

By Theorem 7.8, f (z) = z 2 is entire, and the complex derivative is ux + ivx = 2z.
Example 8.4. The function f (z) = 1/z is defined in the domain C \ {0}. It is complex
differentiable everywhere in the domain because, for z ̸= 0,
1 1
z+h − z 1  z − (z + h) 
=
h h (z + h)z
1
=− .
z(z + h)
1
When h → 0, the limit of − z(z+h) is −1/z 2 . Therefore f (z) = 1/z is complex differentiable
for z ∈ C \ {0}, and the complex derivative is −1/z 2 .
Example 8.5. The function f (z) = |z|2 = x2 + y 2 has zero imaginary part. As a real-valued
function it is real differentiable. However it is complex differentiable only at z = 0. We see
this by computing the partial derivatives

ux = 2x, vx = 0,
uy = 2y, vy = 0.

The Cauchy-Riemann equations are satisfied only at z = 0. Therefore it is not complex

differentiable if z ̸= 0. By Theorem 7.8, it is complex differentiable only at the origin z = 0.
This function is not holomorphic at any point.

71
Example 8.6. We have seen in (5.5) that the complex exponential function can be written
as
exp(x + iy) = ex cos(y) + iex sin(y).

This is an entire function because the Cauchy-Riemann equations

ux = ex cos(y) = vy
uy = −ex sin(y) = −vx

are satisfied at all points, and the partial derivatives of u and v are continuous everywhere.
By Theorem 7.8, the complex exponential function is complex differentiable everywhere,
and the derivative is

exp(z)′ = ux + ivx = ex cos(y) + i(ex sin y) = exp(z).

Example 8.7. Define a complex function


 x3 −y3 +i(x3 +y3 ) if x + iy ̸= 0
x2 +y 2
f (z) =
0 if x + iy = 0.

This function is continuous at z = 0, satisfies the Cauchy-Riemann equations at z = 0, but

fail to be complex differentiable at z = 0.
We first show that f (z) is continuous at z = 0. Take the absolute value of f (z),
1
|f (z)| = |x3 − y 3 + i(x3 + y 3 )|
x2
+ y2
p
2(x6 + y 6 )
= .
x2 + y 2

We simplify and upper bound x6 + y 6 as

(x2 + y 2 )(x4 − x2 y 2 + y 4 ) ≤ (x2 + y 2 )(x4 + 2x2 y 2 + y 4 ) = (x2 + y 2 )3 .

We then get
p √
|f (z)| ≤ 2(x2 + y 2 ) = 2|z|

If we take |z| → 0, we have |f (z)| → 0. Therefore f (z) approaches 0 as z approaches 0.

The function f (z) is a continuous function.

72
 Table 1: Summary of the examples

Not differentiable Real differentiable Holomorphic

but not holomorphic in a domain
Ex. 8.7 Ex. 8.2, 8.5 Ex. 8.1, 8.3, 8.4, 8.6

The real and imaginary parts of f (z) are

x3 − y 3 x3 + y 3
u(x, y) = , v(x, y) = ,
x2 + y 2 x2 + y 2
respectively. We next show that the Cauchy-Riemann equations hold at z = 0,
∂x ∂y
ux (x, 0) = = 1, vy (0, y) = = 1,
∂x ∂y
∂(−y) ∂x
uy (0, y) = = −1, vx (x, 0) = = 1.
∂y ∂x
Therefore, we have ux (0, 0) = vy (0, 0) and uy (0, 0) = −vx (0, 0).
However, this function is not complex differentiable at z = 0, (and hence is not real
differentiable by Theorem 7.10). If this function were complex differentiable, the complex
derivative would be
ux (0, 0) + ivx (0, 0) = 1 + i
by Corollary 7.9. But if we approach z = 0 in a straight line from the point 1 + i, we will
get
1 1   1 2t3 1+i
lim (f (t + it) − f (0)) = lim u(t, t) + iv(t, t) = i 2 = .
t→0 t(1 + i) t→0 t(1 + i) t(1 + i) 2t 2
Obviously 1 + i is not equal to (1 + i)/2. Therefore f (z) is not complex differentiable at
z = 0.

8.2 Differentiation rules

The usual rules for differentiation continue to hold for holomorphic functions. We note that
df
another common notation for complex derivative is dz . This is a complex function whose
′
value at z0 is f (z0 ).

73
 The followings are the basic differentiation formulae. Suppose f (z) and g(z) are holo-
morphic in a domain D.

(f (z) + g(z))′ = f ′ (z) + g ′ (z),

(ag(z))′ = ag ′ (z), where a ∈ C,
(f (z)g(z))′ = f ′ (z)g(z) + f (z)g ′ (z),
f ′ (z)g(z) − f (z)g ′ (z)
(f (z)/g(z))′ = , provided that g(z) ̸= 0.
g(z)2
If the co-domain of g(z) is inside the domain of f , and g, f are both holomorphic, we have
the chain rule
d
f (g(z)) = f ′ (g(z))g ′ (z).
dz
The proof is the same as in calculus and omitted. We illustrate the differentiation rules
by two examples.
Example 8.8. Show that (z n )′ = nz n−1 for all z ∈ C.
It is trivially true when n = 1. We prove the rest by induction. Suppose that it is true
for n = k. When n = k + 1, we apply the product rule to see that

(z k )′ = (zz k−1 )′ = (1)z k−1 + z(k − 1)z k−2 = kz k−1 .

Example 8.9. By combining the previous example with the sum rule, we see that polyno-
mials are entire (complex differentiable at all point), and
d
a0 + a1 z + a2 z 2 + · · · + an z n = a1 + 2a2 z + · · · + nan z n−1 .


dz

Example 8.10. Show that (sin z)′ = cos z.

eiz −e−iz
We differentiate both sides of the identity sin z = 2i to get
 eiz − e−iz ′ (eiz − e−iz )′ ieiz − (−i)e−iz eiz + e−iz
(sin z)′ = = = = = cos z.
2i 2i 2i 2
Example 8.11. We can show that a linear fractional function f (z) = (az + b)/(cz + d) is
complex differentiable and has nonzero derivative at any point in the domain of definition.
Using the quotient rule for differentiation, we obtain
 az + b ′ a(cz + d) − c(az + b) ad − bc
= =
cz + d (cz + d)2 (cz + d)2

74
As part of the definition of LFR, the four constant a, b, c and d satisfies ad − bc ̸= 0. The
complex derivative of LFR is thus nonzero as long as z is not equal to −d/c.
Remark. We can consider the complex derivative of LFR at the point at infinity by making
a change of variable w = 1/z and investigate the function behavior near w = 0. Suppose
f (z) = (az + b)/(cz + d) is an LFR. Define g(w) = f (1/w).

a w1 + b wb + a
g(w) = 1 = .
cw + d wd + c

The derivative of g(w) equals

b(wd + c) − d(wb + a) bc − da
g ′ (w) = 2
= ,
(wd + c) (wd + c)2

which is nonzero at w = 0.
Remark. The complex number −d/c is mapped to the point at infinity by the LFR f (z) =
(az + b)/(cz + d). We can study the behavior new the point −d/c by considering the
reciprocal h(z) ≜ 1/f (z). The function h(z) is an LFR, and it maps −d/c to the origin.
We can compute the derivative of h(z) as
 cz + d ′ c(az + b) − a(cz + d) bc − ad
h′ (z) = = 2
=
az + b (az + b) (az + b)2

The value of h′ (z) at z = −d/c is

bc − ad c2 (bc − ad) c2
h′ (−d/c) = = =
(a(−d/c) + b)2 (−ad + bc)2 bc − ad

If the coefficient c is nonzero, we see that the function f (z) has nonzero derivative at the
point −d/c.
The remaining case c = 0 corresponds to affine function, which maps ∞ to ∞. The
behavior of an affine function at the point at infinity is an exercise.

75
9 Properties of Holomorphic functions
Summary:

ˆ Conformal property

ˆ Phase portrait

ˆ Harmonic functions

9.1 Conformal property

Suppose f (z) is complex differentiable at a given point z0 in C, and suppose f ′ (z0 ) is
nonzero. One can show that the function f is conformal, or angle-reserving at z0 .
Draw two parametric curves γ1 (t) and γ2 (t) through z0 . By “parametric curve” it means
a smooth map from an interval in R to C. Let the range of the parameter t in γ1 (t) be
[a, b], where a < b, and γ1 (t1 ) = z0 , for some t1 in the range [a, b]. Likewise, let the range
of t in γ2 (t) be [c, d], where c < d and γ2 (t2 ) = z0 , for some t2 in [c, d]. In the domain of f ,
the angle between the tangent line of γ1 at z0 and the tangent line of γ2 at z0 is
 γ ′ (t ) 
2 2
θ = arg .
γ1′ (t1 )

Both γ1′ (t1 ) and γ2′ (t2 ) are nonzero, so that the argument is well defined. (See Fig. 2)
If we apply the function f to γ1 (t) and γ2 (t), we obtain a pair of curves intersecting at
the point w0 = f (z0 ),
w0 = f (γ1 (t1 )) = f (γ2 (t2 )).

By applying the chain rule, we get the angle between the two tangent lines
 (f ◦ γ )′ (t )   f ′ (γ (t )) · γ ′ (t )   γ ′ (t ) 
2 2 2 2 2 2 2 2
arg = arg ′ = arg =θ
′
(f ◦ γ1 ) (t1 ) ′
f (γ1 (t1 )) · γ1 (t1 ) γ1′ (t1 )

In particular, if the two curves are perpendicular to each other in the domain, then it
continues to be perpendicular to each other in the co-domain.
The conformal property explains why the conjugate function f (z) = z̄ is not complex
differentiable anywhere. The conjugate function is a reflection geometrically, and reflection
reverses the orientation of angles.

76
 Figure 2: Angle-preserving property of holomorphic function

Definition 9.1. We say that two regions R1 and R2 in C are conformally equivalent
if we can find a bijective holomorphic map f (z) from R1 to R2 .

Example 9.1. The upper half-plane is conformally equivalent to the unit disc. We can
apply the linear fractional transformation f (z) = (z − i)/(z + i). Since a linear fractional
transformation maps a circle/straight line to circle/straight line, we can check the image of
three specific points on the real axis.

1−i (1 − i)2 −2i

f (1) = = = = −i
1+i 2 2
0−i −i
f (0) = = = −1
0+i i
−1 − i (−1 − i)2
f (−1) = = =i
−1 + i 2
The unique circle that passes through the three image points −i, −1 and i is the unit circle.
Therefore, the function f (z) maps the real axis to the unit circle. We further check that
f (i) = 0, i.e., the point i in the upper half plane is mapped to the origin. The above
argument shows that f (z) maps the upper half-plane to the interior of the unit disc.
Since linear fractional transformation is always bijective and conformal, we can say that

77
 Figure 3: Mapping property of the complex exponential function.

the upper half-plane and the unit disc are conformally equivalent.
Example 9.2. The complex exponential function f (z) = ez maps the horizontal strip {z :
−π/2 < Im(z) < π/2} conformally and bijectively to the right half plane {z : Re(z) > 0}.
Infinite horizontal strips are conformally equivalent to the unit open disk.
The complex exponential function also maps

{z : −π < Im(z) < π} → C \ {x + iy : y = 0 and x ≤ 0}

conformally and bijectively (See Fig. 3).

9.2 Phase portrait

Phase portrait, which is also known as domain coloring, plots the argument of a complex
function on a plane, by associating the numbers in the range [0, 2π) with the color spectrum.
The brightness corresponds to the magnitude of the complex function. A dark point means
the function is zero. A bright color indicates the area where function magnitude is large.
We take the phase portrait of the identity function as the reference (Fig. 4).

78
 Figure 4: Phase portrait of f (z) = z

The phase portrait of some complex functions are shown in Fig. 5. The picture for
the function f (z) = 1/z has a bright point near the origin, because this function has large
magnitude near the origin. We also note that as we go around the origin, we see the color
in the reverse order. It is because the function f (z) = 1/z multiply the argument by −1.
In the phase portrait of f (z) = z 2 , we see a zero of order 2 at the origin. As we go around
the origin, we meet all the colors twice.
Example 9.3. The inverse function f (z) = 1/z maps the open disc |z − 1| = 1 to the
half-plane Re(w) > 1/2.

9.3 Harmonic functions

Another necessary condition for being holomorphic is that the real and imaginary parts are
both harmonic functions. A twice-differentiable real-valued function F (x, y) is said to be
harmonic if it satisfies the Laplace equation

Fxx + Fyy = 0,

where Fxx and Fyy are the second partial derivatives of F with respect to x and y, respec-
tively.
Suppose that u(x, y) and v(x, y) are twice differentiable, and the second-order deriva-
tives are continuous functions, so that uxy = uyx and vxy = vyx . Under these assumptions,

79
 (a) Phase portrait of f (z) = 1/z (b) Phase portrait of f (z) = z 2

(c) Phase portrait of

f (z) = (z 2 + 9)/(z − 2) (d) Phase portrait of f (z) = tan(z)

Figure 5: Phase portraits.

80
we can deduce that the real and imaginary part of a holomorphic function are both harmonic
functions. (We will show in a later lecture that the additional smoothness requirements
always hold.) In general, we say that a function is in C 2 if the second partial derivatives
exist and are continuous.

Theorem 9.2. Let f (z) be a holomorphic function on a region R, and the real and
imaginary parts are in C 2 . Then

uxx + uyy = 0 = vxx + vyy

in the region R.

Proof. The complex function f (z) is holomorphic by assumption, hence must satisfy the
Cauchy-Riemann equations ux = vy and uy = −vx . Take partial derivatives with respect
to x and y, we obtain uxx = vyx and uyy = −vxy . Because we can exchange the order of
the partial derivatives, we get

uxx + uyy = vyx − vxy = vxy − vxy = 0.

Similarly, we can show that v(x, y) is harmonic by first deriving uxy = vyy and uyx =
−vxx from the Cauchy-Riemann equation, and then

vxx + vyy = −uyx + uxy = −uxy + uxy = 0.

In the followings, we will assume that u(x, y) and v(x, y) are C 2 functions.
Motivated by the previous theorem we define the notion of harmonic conjugates

Definition 9.3. Given a harmonic function u(x, y), we say that v(x, y) is a harmonic
conjugate of u(x, y) if u(x, y) + iv(x, y) is holomorphic.

Equivalently, we can define that v(x, y) is a harmonic conjugate to u if u and v satisfy

the Cauchy-Riemann equations. It is obvious that harmonic conjugate to a given harmonic
functions, if exists, is not unique. It is because, if v(x, y) is a harmonic conjugate to u(x, y),
then for any complex constant C, the function v(x, y) + C is also a harmonic conjugate.

81
 The existence of harmonic conjugate depends on the topology of the domain on which
the function u(x, y) is defined. It is known that if the domain D is simply connected, then
a harmonic function u(x, y) defined on D has a harmonic conjugate. We can formulate the
problem in terms of differential form. We want to find a function v(x, y) whose differential
is given by
dv = vx dx + vy dy = −uy dx + ux dy.
Since uy and ux are given, the problem of finding a harmonic conjugate to u(x, y) reduces
to solving the differential equation above. We prove in the appendix of this section that
solution always exists if the region of definition D is simply connected (The proof relies an
application of the Green’s theorem).
Example 9.4. We consider the example

u(x, y) = sin(x) cosh(y)

. This is harmonic because

uxx = − sin(x) cosh(y), and uyy = sin(x) cosh(y).

We can set out to find its harmonic conjugate v(x, y). From the condition vx = −uy =
− sin(x) sinh(y), we can obtain
Z
v(x, y) = − sin(x) sinh(y) dx = cos(x) sinh(y) + C(y),

where C(y) is a function of y. By applying the other Cauchy-Riemann equation vy = ux ,

we get
vy = cos(x) cosh(y) + C ′ (y) = ux = cos(x) cosh(y).
Hence the function C(y) should be a constant. The harmonic conjugate of u(x, y) can be
written as
v(x, y) = cos(x) sinh(y) + C
for some constant C. The complex holomorphic function is
eix − e−ix ey + e−y  eix + e−ix ey − e−y 
sin(x) cosh(y) + i cos(x) sinh(y) = · +i ·
2i 2 2 2
1 ix
= [(e − e−ix )(ey + e−y ) − (eix + e−ix )(ey − e−y )]
4i
ei(x+iy) − e−i(x+iy)
=
2i
= sin(z).

82
Example 9.5. Find a harmonic conjugate of the function

u(x, y) = −2x2 + x3 + 2y 2 − 3xy 2 .

The function u(x, y) is defined on the whole complex plane. We check that it is harmonic:

ux = −4x + 3x2 − 3y 2
uxx = −4 + 6x
uy = 4y − 6xy
uyy = 4 − 6x.

Hence uxx + uyy is identically equal to zero.

We now proceed to find a harmonic conjugate. We first integrate

vx = −uy = −4y + 6xy

with respect to x and get

Z
v(x, y) = −4y + 6xy dx = −4xy + 3x2 y + C(y),

where C(y) is a constant that may involve y. Differentiate the above with respect to y,

vy = −4x + 3x2 + C ′ (y).

After comparing with ux , we see that C ′ (y) = −3y 2 , and hence C(y) = −y 3 + C for some
constant C. The answer is

v(x, y) = −4xy + 3x2 y − y 3 + C,

where C is any constant.

When u(x, y) is a polynomial in x and y, there is a slick method to directly obtain the
complex function f (z) whose real part is u(x, y). We can write z = x + iy and exploit the
following equalities

z + z̄ z − z̄
f (z) + f (z) f (z) + f (z̄)

u , = u(x, y) = Re(f (z)) = = .
2 2i 2 2
In the last step we can replace f (z) by f (z̄), because f is a polynomial in z.

83
 We can re-do Example 9.5 by using this method.
z + z̄ z − z̄
z + z̄
2 z + z̄
3 z − z̄
2 z + z̄
z − z̄
2
u , = −2 + +2 −3
2 2i 2 2 2i 2 2i
2 2 3 3
= −z − z̄ + z /2 + z̄ /2.

By comparing it with (f (z) + f (z̄))/2, we obtain f (z) = −2z 2 + z 3 .

One can further streamline this method by directly computing
z z

2u , − u(0, 0). (9.1)
2 2i
(We subtract u(0, 0) in order to preserve the constant term.) This will be a complex
differentiable function f in variable z only and the real part is precisely u(x, y). The
complex function in Example 9.5 is thus.
z z

2u , − u(0, 0)
2 2i
z z z z z

= 2 − 2( )2 + ( )3 + 2( )2 − 3( )( )2
2 2 2i 2 2i
2 3
= −2z + z .

9.4 Application to Dirichlet’s problem

The Dirichlet problem involves solving the Laplace equation in a region with a given bound-
ary condition. This equation holds significant importance in physics, as it can be used to
determine the steady-state temperature distribution in a given region. However, solving
the Dirichlet problem on a complex or irregularly shaped region can be quite challenging.
To simplify the problem, a technique called problem reduction can be employed. This
technique involves transforming the original Dirichlet problem on a complex region into a
Dirichlet problem on a standard and simpler shape, such as the unit circle. By utilizing a
conformal map to transform the region to a circle, we can focus our attention on solving
the Dirichlet problem specifically on the circle. Consequently, solving the Dirichlet problem
on the original complex region boils down to finding an appropriate conformal map.
We note that Laplace equation and the physical application are not directly related to
complex variables. However, it has been discovered that using the language of complex
variables can greatly simplify matters. Let’s consider a scenario where we have a region
denoted as R with its boundary described by a closed curve γ1 . We identify a point (x, y)

84
with a complex number z = x + iy. Suppose we can transform this region to the unit disc
in the w-plane using a holomorphic transformation

T (z) = u(x, y) + iv(x, y)

that has a non-vanishing derivative throughout the domain R. The boundary condition on
R is transformed to a boundary condition on the unit circle.
We first show that the transformation T is locally one-to-one. Considering T as a
multi-variable function with variables x and y, we can express its Jacobian matrix as:
" #
ux uy
vx vy

Since we assume that the Cauchy-Riemann equations are satisfied, we can rewrite the
determinant as

ux vy − vx uy = ux (ux ) − vx (−vx ) = u2x + vx2 = |f ′ (z)|2 ,

which is non-zero based on our assumption. This demonstrates that the Jacobian is not
equal to zero.
Next, we will demonstrate that a solution to the Dirichlet problem on the unit circle
corresponds to a solution in the original region R. Let F (u, v) be a real-valued function
that is twice differentiable and defined on the unit circle. We define the composite function
G as
G(x, y) = F (u(x, y), v(x, y)).

We claim that
Gxx + Gyy = (Fxx + Fyy )|T ′ (z)|2 (9.2)

Consequently, if T ′ (z) is nonzero, we have ∆G = 0 if and only if ∆F = 0. The equation

in (9.2), often referred to as the “transfer lemma,” establishes a relationship between the
Laplacian in the z-domain and the Laplacian in the w-domain.
The proof of equation (9.2) involves applying multivariable calculus while assuming the
following conditions: (1) u(x, y) is harmonic, (2) v(x, y) is harmonic, and (3) the Cauchy-
Riemann equations are satisfied.
Taking the derivative with respect to x again, and applying the product rule, we have

Gx = Fu ux + Fv vx .

85
 Figure 6: Transforming the domain of Dirichlet problem

Differentiating with respect to x again, and apply the product rule

Gxx = Fu ux + Fv vx x
= (Fu )x ux + Fu uxx + (Fv )x vx + Fv vxx .

Here, (Fu )x represents the partial derivative of Fu with respect to x, which can be computed
as
(Fu )x = Fuu ux + Fuv vx .

Likewise, the partial derivative of Fv with respect to x is

(Fv )x = Fvu ux + Fvv vx .

By combining these results, we obtain

Gxx = Fuu u2x + Fuv ux vx + Fu uxx + Fvu ux vx + Fvv vx2 + Fv vxx .

By repeating the same calculations with x replaced by y, we can derive the expression
for Gyy ,
Gyy = Fuu u2y + Fuv uy vy + Fu uyy + Fvu uy vy + Fvv vy2 + Fv vyy .

86
 By substituting the expressions for Gxx and Gyy into the Laplacian ∆G, we obtain a
lengthy expression:

Gxx + Gyy = Fuu (u2x + u2y ) + Fvv (vx2 + vy2 ) + (Fuv + Fvu )(ux vx + uy vy )
+ Fu (uxx + uyy ) + Fv (vxx + vyy ).

Given that u and v are assumed to be harmonic, the terms in the second line of the
expression vanish. Moreover, applying the Cauchy-Riemann equations allows us to simplify
the term (ux vx + uy vy ) as follows,

ux vx + uy vy = ux (−uy ) + uy (ux ) = 0.

By Cauchy-Riemann again, we have vx2 + vy2 = u2x + u2y . We can simplify the Laplacian of
G to
Gxx + Gyy = (Fuu + Fvv )(u2x + u2y ) = (Fuu + Fvv )|f ′ (z)|2 .

This completes the derivation of (9.2).

9.5 Additional proof

In this optional section we prove that harmonic conjugate of a harmonic function exists if
the domain of definition is simply connected. The main step is the Green’s theorem from
multi-variable calculus. In the followings, the functions u and v must be sufficiently smooth
so that the Green’s theorem can be applied.
Consider the differential form

dv = vx dx + vy dy = −uy dx + ux dy.

We claim that the line integral

Z (x1 ,y1 )
−uy dx + ux dy (9.3)
(x0 ,y0 )

does not depend on path. We can see it from Green’s theorem, using the assumption that
D is simply connected and u(x, y) satisfies the Laplace equation. If C is a closed path
enclosing a region S, we have
I ZZ ZZ
∂ ∂
−uy dx + ux dy = ux − (−uy ) ds = uxx + uyy ds = 0.
C S ∂x ∂y S

87
 We can thus define a function v(x, y) by first fixing a point (x0 , y0 ) in the domain D of
u(x, y), and define a function v(x, y) by
Z (x,y)
v(x, y) ≜ −uy dx + ux dy (9.4)
(x0 ,y0 )

for any point (x, y) in D. This is well-defined because the line integral is independent of
path, and there is at least one path connecting (x0 , y0 ) and (x, y).
We now check that the function v(x, y) so defined is a harmonic conjugate of u(x, y).
Let (a, b) be any point in D. The partial derivative of v with respect to x at this point is
v(a + ∆x, b) − v(a, b)
vx (a, b) = lim
∆x→0 ∆x
Z (a+∆x,b)
1
= lim −uy dx + ux dy
∆x→0 ∆x (a,b)
Z (a+∆x,b)
1
= lim −uy dx
∆x→0 ∆x (a,b)

= −uy (a, b).

In the last step we have used the assumption that uy is continuous at (a, b).
Likewise, by consider the partial derivative of v with respect to y, we get
v(a, b + ∆y) − v(a, b)
vy (a, b) = lim
∆y→0 ∆y
Z (a,b+∆y)
1
= lim −uy dx + ux dy
∆y→0 ∆y (a,b)
Z (a,b+∆y)
1
= lim ux dy
∆y→0 ∆y (a,b)

= ux (a, b).

We have used the continuity of ux in the last step. This proves that the Cauchy-Riemann
equations are satisfied at the point (a, b). By the sufficient condition in Theorem 7.8, we
conclude that u(x, y) + iv(x, y) is a holomorphic function in the domain D.
We can now conclude that a harmonic conjugate of u(x, y) can be written as in (9.4)
when D is simply connected. This provides the existence proof of harmonic conjugate, and
is the theoretical basis for the computational examples in this lecture.
In general, the integration in (9.3) is not tractable. The purpose of the proof is to give
a sufficient condition under which harmonic conjugate exists.

88
9.6 Appendix: Physical interpretation
Given a complex function f (z) = u(x, y)+iv(x, y) that is complex differentiable in a domain
D, we define a real vector field (M (x, y), N (x, y)) by

M (x, y) ≜ u(x, y), N (x, y) ≜ −v(x, y).

This is the vector field corresponding to the conjugate of f (z). This is sometime called the
Polya’s vector field of f (z).

Theorem 9.4. If f (z) is holomorphic in D, then the vector field (M (x, y), N (x, y))
has zero 2D curl and zero divergence in D.

Proof. The 2-dimensional curl of (M (x, y), N (x, y)) is the usual 3-dimensional curl function
applied to (M (x, y), N (x, y), 0), with the third component set identically to zero. The curl
of (M (x, y), N (x, y), 0) is

⃗i ⃗j ⃗k
∂ ∂
∂
∂x
∂
∂y
∂
∂z
= ⃗k ∂x ∂y
= ⃗k(−vx − uy ).
M (x, y) N (x, y)
M (x, y) N (x, y) 0

By Cauchy-Riemann equation, vx + uy is equal to ⃗0 throughout the domain D.

On the other hand, the divergence of (M (x, y), N (x, y)) is

∂ ∂
M (x, y) + N (x, y) = ux + (−v)y
∂x ∂y
which is also identically zero by Cauchy-Riemann equation.

We thus see that a holomorphic function could be interpreted as the reflection of an

irrotational and incompressible vector field. This is the basis of the application of complex
analysis to solving some 2-dimensional partial differential equations.

89
10 Complex differential form
Summary:

ˆ Tangent space

ˆ Linear functional and real differential

ˆ Complex differential

ˆ ∂¯ operator

The main reference for this lecture is [Spivak, Chapter 5] and [Miranda, Chapter IV].
In calculus, a differential dx is usually regarded as an infinitesimal quantity, representing
a very small change in the x variable. The total differential of a bi-variate function f (x, y)
is
∂f ∂f
df = dx + dy.
∂x ∂y
However, the treatment of differential form in calculus is usually not very rigorous.
In this lecture we study complex differentiability by first considering multi-variable real
functions and give a more rigorous definition of differential dx and dy. Then we extend
it to the complex case and give a concrete meaning of the symbol dz. In the following
sections we will borrow some notation from real manifolds. However, we do not need the
full generality but only need the special case of dimension 2 and when the manifold is flat.

10.1 Tangent space and differential form

We first consider the Euclidean space R2 in the real case.
We define the tangent space at a given a point p in R2 as the set of all vectors that
emanate from p. The center point is represented by a position vector p. To emphasize the
starting point of a tangent vector, we write

(p, v)

to represent a tangent vector emanating from the point p.

The first component, p, indicates that p is the initial point. We can add two vectors in
Tp as follows,
(p, v1 ) + (p, v2 ) ≜ (p, v1 + v2 )

90
and multiply by a scalar
a(p, v) ≜ (p, av).

However, if p and q are two distinct vector, the addition of (p, u) and (q, v) are not defined.
(See Fig. 7)
Formally, given a point p, the tangent space at p is defined as the set

Tp ≜ {(p, v) : v ∈ R2 }.

For each point p, the tangent space Tp is a two-dimensional vector space. We will use e1
and e2 to denote the standard basis. We will also use the shorter notation vp instead of
(p, v). For concreteness, we can consider the example

2(4e1 − e2 )p + (e1 + 3e2 )p = (9e1 + e2 )p .

Dot product of two vectors in Tp is done in the expected way

(p, u) · (p, v) ≜ (p, u · v). (10.1)

A vector field is a mapping that associates, for each point p, a tangent vector vp , in a
continuous and smooth manner. For instance, we can represent a vector in each tangent
space using a formula, such as

((3x + y)e1 + (x − y)e2 )(x,y)

for (x, y) ∈ R2 . In simple terms, the expression above says that at the point (x, y), we select
the vector (3x+y)e1 +(x−y)e2 from the associated tangent space T(x,y) . To further simplify
notation, very often we skip the subscript (x,y) . In physics notation, the two unit vectors
are commonly denoted as ⃗ı and ⃗ȷ. Therefore, if we write (3x + y)⃗ı + (x − y)⃗ȷ, it represents a
vector emanating from the point (x, y), with horizontal and vertical components equal to
3x + y and x − y, respectively. The graph of this vector field is depicted in Fig. 8.
The cotangent space at p, denoted as Tp∗ , is the dual space of Tp . It consists of all linear
functionals, also known as covectors, that take a tangent vector at p as input and return a
real number. A differential form is a continuous selection of a covector at each cotangent
space.
An example of covector is a function that returns the x-coordinate of a vector.

ϕ1 (xe1 + ye2 ) = x. (10.2)

91
Figure 7: Tangent space at point p and tangent space at q

Figure 8: Plot of vector field (3x + y)⃗ı + (x − y)⃗ȷ

92
Another simple example is a function that returns the y-coordinate

ϕ2 (xe1 + ye2 ) = y. (10.3)

In fact, a family of examples can be derived from directional derivatives. Given a

differentiable function f (x, y) defined at a specific point p = (x0 , y0 ) and a vector vp =
(re1 + se2 )p , the directional derivative of f at p in the direction of vp is given by

f (x0 + rh, y0 + sh) − f (x0 , y0 ) ∂f (x0 , y0 ) ∂f (x0 , y0 )

lim = r+ s.
h→0 h ∂x ∂y
We denote the partial derivative with respect to the first coordinate by fx , and the partial
derivative with respect to the second coordinate by fy . Using the dot product notation for
tangent vectors in (10.1) and the gradient

(∇f )p = (fx (x0 , y0 ), fy (x0 , y0 ))p ,

we can express the directional derivative as ∇fp · vp . This define a linear functional ϕf :
Tp → R by
ϕf (vp ) = ∇fp · vp .

Example 10.1. Let f (x, y) = x2 y 3 . At the point p = (−1, 2), the function f gives rise to a
linear functional

ϕf (vp ) = (2xy 3 e1 + 3x2 y 2 e2 )p · vp

x=−1,y=2

= (−16e1 + 12e2 )p · vp

If we change the point to p′ = (2, 4), then the linear functional defined by f at p′ is

ϕf (vp′ ) = (2xy 3 e1 + 3x2 y 2 e2 )p′ · vp′

x=2,y=4
8 6
= (2 e1 + 3 · 2 · e2 )p′ · vp′

This leads to the next definition.

Definition 10.1. Given a real differential function f (x, y), we let df be the differ-
ential form that chooses the linear functional (df )p at p that take a tangent vector
vp as input and returns ∇fp · vp .

93
 We can express (df )p as

(df )p = fx p
· ϕ1 + fy p
· ϕ2

If we consider f (x, y) = x, then (df )p is the function that return the first coordinate of
the input vector vp . This is equivalent to the covector ϕ1 defined in (10.2),

dx = ϕ1 .

Similarly, if we take f (x, y) = y, then (dy)p is the same as ϕ2 in (10.3),

dy = ϕ2 .

Hence, we can write df in the convenient form as

df = fx dx + fy dy,

but it is important to note that both sides of the equation are linear functionals. They
become numbers only when evaluated at a specific tangent vector.
In calculus textbooks, it is often stated that we can assign any values to dx and dy,
suggesting their linearly independence. This perspective makes sense when treating dx and
dy as numbers. However, after defining differentials as functions, we are able to assert that
dx and dy are linearly independent just like in linear algebra. More precisely, suppose there
exists constant a and b such that a · dx + b · dy is equal to the zero function. We apply this
functional to the basis vector e1 to get

(a · dx + b · dy)(e1 ) = a · 1 + b · 0 = a,

which should be equal to zero as a · dx + b · dy is a zero function. Likewise, by evaluating

the functional a · dx + b · dy to e2 , we obtain b = 0. This proves the following theorem.

Theorem 10.2. The two differential dx and dy are linearly independent, and hence
form a basis of the cotangent space Tp∗ .

2
Example 10.2. Let f (x, y) = ex sin(y). The differential of f is
2 2
df = 2xex sin(y) dx + ex cos(y) dy.

94
At the point p = (1, π), (df )p is the linear functional

(df )(1,π) = −e · ϕ2 .

This is a function that take a tangent vector with initial point (1, π) and return (−e) times
the second coordinate of the tangent vector.
Since dx and dy form a basis of the cotangent space, we can write a real differential
form as
a(x, y)dx + b(x, y)dy

for some real-valued functions a(x, y) and b(x, y). If a point p = (x0 , y0 ) is given, the above
differential form will pick up the differential

a(x0 , y0 )(dx)p + b(x0 , y0 )(dy)p

∗
in the cotangent space T(x .
0 ,y0 )

Remark. In this section, we have focused on defining the differential of a real-valued func-
tion. However, for vector-valued functions, such as g : Rm → Rn , the differential of g can
be represented by an m × n matrix. This matrix is constructed using the partial derivatives
of the component functions of g. You can find more information on this topic in the book
by Rudin [Rudin].

10.2 Complex differential form

We complexify the cotangent space by allowing complex numbers as the coefficients. This
means that when forming a linear combination a(dx)p + b(dy)p , the value of a and b can
be complex numbers1 .
As an example, consider the complex differential idy. It takes a vector (r, s)p as input
and returns the value i · s. Here, i represents the imaginary unit. This shows that the
complex differential idy assigns complex numbers to tangent vectors, allowing for a more
general and flexible framework.
This complexification process can be formally realized by tensoring Tp∗ with C over R. In mathematical
1

symbols, the complex differential lives in the space C ⊗R Tp∗ .

95
 Definition 10.3. Define dz and dz̄ to be the complex differential form whose com-
ponent at a point p is

(dz)p ≜ (dx)p + i(dy)p

(dz̄)p ≜ (dx)p − i(dy)p .

The symbol “(dz)p ” represents a function, that takes a vector, say (α, β)p with an initial
point p, and return the complex number α + iβ. Similarly, the function (dz̄)p accepts a
vector (α, β)p as input and return the complex number α − iβ.
As in the real case, we will drop the subscript p in order to simplify notation.
The next example analyze a complex function f (z) by treating it as a multi-variable
real function.
Example 10.3. Find the differential of the real and imaginary parts of f (z) = z 2 .
The real and imaginary parts are u(x, y) = x2 −y 2 and v(x, y) = 2xy. Their differentials
are

du = 2x dx − 2y dy
dv = 2y dx + 2x dy.

We can put du and dv together to form a complex differential,

du + idv = (2x dx − 2y dy) + i(2y dx + 2x dy)

= 2(x + iy)(dx + idy)
= 2zdz.

We will call it the differential of the complex-valued function f .

In general, for a complex function f (z) = u(x, y) + iv(x, y), we define the complex
differential df as
du + idv.

Despite the relationship between z and z̄, it is important to note that the two complex
differentials, dz and dz̄, are linearly independent. This fact can be formulated in the
following theorem.

96
 Theorem 10.4. Suppose w1 dz + w2 dz̄ = 0 for some complex numbers w1 and w2 .
Then w1 = w2 = 0.

Proof. The condition in the theorem says that w1 dz + w2 dz̄ = 0 is a constant zero function.
Thus, we can write it as

w1 (dx + idy) + w2 (dx − idy) = (w1 + w2 )dx + i(w1 − w2 )dy.

This expression should be equivalent to the constant zero function. We can evaluate it at
the standard basis vectors e1 and e2 , and the results should be zero. This leads to the
following equations:

w1 + w2 = 0
i(w1 − w2 ) = 0.

Solving the system of linear equations, we easily get w1 = 0 and w2 = 0.

The previous theorem establishes that dz and dz̄ are linearly independent over C. There-
fore, the transformation from dx and dy to dz and dz̄ corresponds to a change of basis in
the complexified cotangent space.
The relationship between dx, dy, dz, and dz̄ can be expressed as follows:
dz + dz̄
dx = , (10.4)
2
dz − dz̄
dy = . (10.5)
2i
This implies that dx and dy can be expressed as linear combinations of dz and dz̄. This
change of basis allows us to work with dz and dz̄, which are more convenient when dealing
with complex functions and complex analysis.

10.3 Partial differential operators

Suppose the real and imaginary parts of a complex function

f (x + iy) = u(x, y) + iv(x, y)

97
are real differentiable as functions of x and y. We want to express the complex differential
df using the basis dz and dz̄.
We start by writing the differentials of u and v as follows:

du = ux dx + uy dy
dv = vx dx + vy dy,

where ux , uy , vx and vy are first-order partial derivatives of u and v. We can then express
du + idv as

du + idv = (ux dx + uy dy) + i(vx dx + vy dy)

= (ux + ivx )dx + (uy + ivy )dy.

To simplify notation, we define

∂f
≜ ux + ivx
∂x
∂f
≜ uy + ivy .
∂y
Continuing the derivation, we have
∂f ∂f
df = du + idv = dx + dy
∂x ∂y
∂f  dz + dz̄  ∂f  dz − dz̄ 
= +
∂x 2 ∂y 2i
1  ∂f ∂f  1  ∂f ∂f 
= −i dz + +i dz̄.
2 ∂x ∂y 2 ∂x ∂y
This expression shows
 that the  differential
 df can
 be expressed in terms of dz and dz̄
1 ∂f ∂f 1 ∂f ∂f
using the coefficients 2 ∂x − i ∂y and 2 ∂x + i ∂y .

Definition 10.5. Define

∂f 1  ∂f ∂f 
≜ −i
∂z 2 ∂x ∂y
∂f 1  ∂f ∂f 
≜ +i .
∂ z̄ 2 ∂x ∂y
 
The operator 1
2
∂
∂x
∂
+ i ∂y is called the ∂¯ operator .

98
 We now relate the ∂¯ operator to the Cauchy-Riemann equations.

Theorem 10.6. For any complex function f (z) = u(x, y)+iv(x, y), the two Cauchy-
Riemann equations ux = vy and uy = −vx are equivalent to ∂f /∂ z̄ = 0. Moreover,
when the Cauchy-Riemann equations are satisfied, we have
∂f
= f ′ (z),
∂z
∂f
where ∂z is defined in Definition 10.5.

Proof. To expand the partial derivative ∂f ∂ z̄ with respect to z̄ in terms of the partial deriva-
tives of the real and imaginary parts of f (z), we start with the expression
∂f 1  ∂f ∂f  1
≜ +i = (ux + ivx + i(uy + ivy ))
∂ z̄ 2 ∂x ∂y 2
1
= (ux − vy + i(vx + uy )).
2
If ∂f /∂ z̄ is zero, then the real and imaginary parts on the right-hand side of the equation
must both be zero. This implies that the Cauchy-Riemann equations hold. Conversely, if
the Cauchy-Riemann equations are satisfied, the real and imaginary parts of the expression
above are both zero, and thus ∂f /∂ z̄ = 0.
Now, assuming that the Cauchy-Riemann equations hold, we can expand the expression
for ∂f /∂z as follows:
∂f 1  ∂f ∂f  1
≜ −i = (ux + ivx − i(uy + ivy ))
∂z 2 ∂x ∂y 2
1
= (ux + ivx − i(−vx + iux ))
2
= (ux + ivx )

By Corollary 7.9, we see that ∂f /∂z = f ′ (z).

The ∂¯ operator behaves like taking derivative with respect to z̄. We can evaluate ∂¯z̄
¯ as follows,
and ∂z
∂ z̄ 1 ∂ ∂ 
∂¯z̄ = = +i (x − iy) = 1
∂ z̄ 2 ∂x ∂y
¯ = ∂z = 1 ∂ + i ∂ (x + iy) = 0.
 
∂z
∂ z̄ 2 ∂x ∂y

99
On the other hand, we have
∂ z̄ 1 ∂ ∂ 
= −i (x − iy) = 0
∂z 2 ∂x ∂y
∂z 1 ∂ ∂ 
= −i (x + iy) = 1.
∂z 2 ∂x ∂y
Example 10.4. Apply the theory to the square function

f (z) = z 2 = (x2 − y 2 ) + i(2xy).

The complex differential df is

df = d(x2 − y 2 ) + id(2xy) = 2xdx − 2ydy + 2i(ydx + xdy)

= 2(x + iy)dx + 2i(x + iy)dy.

Using (10.4) and (10.5), we can perform a change of basis and write df in terms of dz and
dz̄,
dz + dz̄ dz − dz̄
df = 2z + 2iz
2 2i
= z(dz + dz̄) + z(dz − dz̄)
= 2zdz.

The complex differential d(z 2 ) is thus equal to 2zdz. It only depends on z and dz, but does
not depend on dz̄.
Example 10.5. The function f (x+iy) = x+ix2 is continuous and smooth as a real function.
We can express it in terms of z and z̄ as
z + z̄ z + z̄
2
x + ix2 = +i
2 2
1
= 2z + 2z̄ + iz 2 + 2iz z̄ + iz̄ 2 .


4
This is a function of both z and z̄. We can check that ∂f /∂ z̄ is equal to

∂f 1 ∂ ∂ 
= +i (x + ix2 )
∂ z̄ 2 ∂x ∂y
1
= + ix.
2

100
The function f (x + iy) = x + ix2 is not complex differentiable because ∂f /∂ z̄ is not equal
to zero at any point z.
Anyway we can still express d(x + ix2 ) in terms of the basis dz and dz̄

dz + dz̄ 1 + 2ix 1 + 2ix

d(x + ix2 ) = (1 + 2ix) dx = (1 + 2ix) = dz + dz̄.
2 2 2
For a complex function f (z) that is complex differential in a domain D, we have the
following results.

ˆ If we apply the ∂¯ operator to f (z), the resulting function ∂f

∂ z̄ is identically zero in D.
We can interpret this as “f is a function of z only, but not a function of z̄.”

ˆ Since ∂f /∂ z̄ is zero, we can write the differential df in the form of

∂f
df = dz.
∂z
We thus obtain a relationship between the complex differential, the partial derivative
of f with respect to z, and the complex derivative f ′ (z):

∂f
df = dz = f ′ (z)dz.
∂z

101
11 Multi-valued functions
Summary

ˆ Parametric curve

ˆ Multi-valued function

ˆ The angle function

√
In the context of real-valued functions, the square root function x serves as a funda-
mental example of a multi-valued function. For any positive input value, there exist two
possible choices of square roots. By convention, we typically select the positive root as the
default choice.
Similarly, inverse trigonometric functions are also multi-valued. Take, for instance, the
arc tangent function tan−1 (x). In this case, we usually choose the function value that falls
within the range of −π/2 to π/2.
Formally, a multi-valued function, or simply a multi-function, is a function that maps
a point in the domain to a set of values. For example, we can consider the square root
√ √
function as a multi-function and assign the set { x, − x} to a positive real number x.
When viewing the arc tangent function tan−1 (x) as a multi-function, it maps a real number
x to a set in the form of {θ + πk : k ∈ Z}, where θ is a number such that tan θ = x.

11.1 Parametric curves

We formally define a path as follows.

Definition 11.1. parametric curve, or a path is defined as a continuous function γ,

that maps a closed interval [a, b] to the complex numbers C:

γ : [a, b] → C. (11.1)

A path is differentiable if γ is a differentiable function, i.e., γ ′ (t) exists. We say that a

path is piece-wise differentiable) if it is a concatenation of finitely many differentiable
paths.

The constants a and b represent the start and end times, respectively.

102
 Figure 9: An example of parametric curve that has self-intersections.

For any t within the interval [a, b], the complex number γ(t) indicates the robot’s location
at time t. Self-intersection points are allowed in the parametric curve, and it possesses a
sense of direction. An example of such a curve is depicted in Figure 9.

Definition 11.2. The parametric curve in (11.1) is said to be

ˆ simple if there is no self-intersection, i.e., γ(t1 ) ̸= γ(t2 ) for any a ≤ t1 < t2 ≤ b,

except when t1 = a and t2 = b.

ˆ closed if the start point is the same as the end point, i.e., γ(a) = γ(b).

From multi-variable calculus, we know how to compute the length of the path using line
integral. If the x and y coordinates of the point γ(t) is x(t) and y(t), respectively, then we
can compute the length of the path γ(t) by
Z bp
(x′ (t))2 + (y ′ (t)2 ) dt,
a

provided that both x′ (t) and y ′ (t) are continuous functions. In the followings we need to
assume that the length of the path represented by γ(t) is finite. For this purpose, we will
assume that the real and imaginary parts of γ(t) are both continuously differentiable.

103
 Figure 10: Going around the origin.

11.2 The discontinuity of the principal branch of the argument function

To illustrate the problem we aim to solve, we will consider a concrete scenario. In Python,
we can calculate the principal argument of a complex number x + iy using the atan2(y,x)
function. This function computes the angle between the point x + iy and the origin, with
a range of (−π, π].
For instance, let’s compute the angle of the point z = −1 along the negative real axis.
from math import atan2

x , y = -1 , 0
a = atan2 (y , x )
print ( f " The angle of { x }+ i ({ y }) is { a } " )

We can see that the angle is π.

The angle of -1+ i (0) is 3 .14159 265358 9793

If we move the point slightly below the negative real axis, the angle is very close to −π.
We can run

104
 from math import atan2

x , y = -1 , -0.0000001
a = atan2 (y , x )
print ( f " The angle of { x }+ i ({ y }) is { a } " )

This will give

The angle of -1+ i ( -1e -07) is -3.1415925535897933

There is a sudden change of function value is due to the discontinuity of the function
atan2(y,x) at the negative real axis.
Imagine a robot moving within a complex plane, where our objective is to measure
the angle along its path. However, the robot can only compute the angle based on its
location. It possesses GPS to determine its coordinates and has some computing capability.
Nevertheless, the robot lacks intelligence and can only execute the instructions provided to
it.
Suppose the location of the robot is given by γ(t) for time t in the interval [a, b].
The value of γ(t) is a complex number expressed as x + iy, where x and y represent the
coordinates of the robot at time t.
If we compute the angle using the atan2 function on the robot, and the robot traverses a
loop that includes the origin, we will observe a discontinuity jump as illustrated in Figure 10.
This discontinuity occurs when the robot crosses the negative real axis. However, it is
desirable to have a continuous measurement of the angle.

11.3 Computing the angle within a half-plane

One solution is to select a designated region in the complex plane as the domain and restrict
the robot’s movement within that domain. A suitable choice for the domain is a half-plane
with its boundary passing through the origin. For instance, we can consider the right half-
plane H ≜ {x+iy ∈ C : x > 0}. To construct a continuous angle function on the half-plane
H, we can express a complex number z = x + iy and define

−1 y
 
A(z) ≜ tan . (11.2)
x

105
 Figure 11: The right half-plane and the angle function.

The output range of this function lies within the open interval (−π/2, π/2) (refer to Fig-
ure 11). Here, we leverage the property that the x-coordinate is strictly positive in the
right half-plane. The mapping (x, y) 7→ tan−1 (y/x) is well-defined and avoids division by
zero errors. As long as the point x + iy remains within the right half-plane, there will be
no discontinuity in the angle measurement.
In reality, the robot has the capability to travel to any location. Therefore, we propose
a solution for measuring the change of angle without restricting the robot to the right half-
plane. The idea is to cover the entire complex plane (except the origin) using an infinite
number of half-planes. For any given angle α, let Hα denote the half-plane with a boundary
defined by a straight line passing through the origin, where the normal vector makes an
angle of α radians with the positive real axis. An illustrative example is depicted in Fig. 12.
We can define an angle function on Hα by utilizing the existing A function as follows:

Aα ≜ A(e−iα z) + α. (11.3)

Geometrically, this process involves rotating the half-plane Hα to the right half-plane H0 ,
applying the function A, and then rotating it back to Hα . We record this construction in
the following proposition.

106
 Figure 12: The half plane Hα .

Proposition 11.3. We can define a continuous angle function in an open half plane
in the complex plane, with the boundary being a straight line passing through the
origin.

When |α1 − α2 | ≤ π/2, i.e., when the difference between the two angles α1 and α2 is
within π/2, the functions Aα1 and Aα2 are identical when restricted to the intersection of
Hα1 and Hα2 .
We do not have to use the arc tangent function in computing the angle. We may use
other trigonometric functions instead. For example, if the domain is the upper half-plane

Hπ/2 = {x + iy ∈ C : y > 0},

we may use the arc cosine function

 x 
Aπ/2 (x + iy) = cos−1 p .
x2 + y 2
It define a continuous angle function on the upper half-plane, and its range is (0, π)
If the half-plane is
Hπ/4 = {x + iy ∈ C : x + y > 0},

107
whose boundary is obtained by rotating counter-clockwise 45 degrees from the imaginary
axis, we can compute the angle function as
√ 
−1 (x − y)/ 2

Aπ/4 (x + iy) = cos p .
x2 + y 2
The range of this function is (−π/4, 3π/4).

11.4 The angle function

In this section we will describe how to select a branch of the argument function on a path
(which may or may not be a simple path)
Since the robot’s path may not be confined within a single half-plane, we can divide
the path into small portions and cover each portion with a suitable half-plane. We apply
the corresponding function Aα as long as the input point remains within the respective
half-plane.
We make the following assumptions: firstly, the robot never reaches the origin where the
angle function is not defined, and secondly, the robot is capable of measuring the distance
traveled. Let γ(t) represent the robot’s movement on a path for 0 ≤ t ≤ b. We can employ
the following algorithm:
The robot starts at a designated point called the base point z0 . Since z0 is non-zero,
the distance to the origin d0 ≜ |z0 | is positive. Let α0 = arg z0 denote the initial argument.
At the beginning, we utilize the function Aα0 defined on the half-plane Hα0 . We know
that the robot’s location will remain within the half-plane Hα0 as long as it stays within
a distance d0 from the initial point. In the first segment of the trajectory with a length of
d0 , we utilize the function Aα0 to measure the angle. Suppose that at time t1 , the robot
has traveled a distance of d0 . For 0 ≤ t ≤ t1 , we define

ϕ(t) ≜ Aα0 (γ(t)), for 0 ≤ t ≤ t1 .

We mark the point z1 at time t1 . The angle at this point is Aα0 (z1 ), and the distance
to the origin is d1 ≜ |z1 |. At this moment, we change the base point from z0 to z1 , and
use the function Aα1 , which is defined on the half-plane Hα1 . Let t2 be the time when the
robot travel a distance of d1 along the contour γ(t) from the point z1 . For time t between
t1 and t2 , we continue the angle measure by

ϕ(t) ≜ Aα1 (γ(t)) − Aα1 (z1 ) + ϕ(t1 ), for t1 ≤ t ≤ t2 .

108
Figure 13: Computing the change of angle by dividing a curve into smaller parts

109
 At time t2 , we update the base point to z2 ≜ γ(t2 ). The distance to the origin becomes
d2 = |z2 |, and the corresponding argument is ϕ(t2 ). Let t3 denote the time instance when
the robot has traveled a distance of d2 from the point z2 . For t2 ≤ t ≤ t3 , we define

ϕ(t) ≜ Aα2 (γ(t)) − Aα2 (z2 ) + ϕ(t2 ), for t2 ≤ t ≤ t3 .

This process can be repeated in a similar manner. The resulting function ϕ(t) is con-
tinuous and measures the angle between the robot and the origin. When the robot comes
to a stop at time tn = b, the total change in angle can be computed as
n−1
X
Aαk (γ(tk+1 )) − Aαk (γ(tk )). (11.4)
k=0

See Fig. 13 for a sample of division of path.

In conclusion, the procedure described in this section is able to obtain a continuously-
varying angle function on any path, provided that the path does not pass through the
origin, and the length of the curve is well-defined and finite.

110
12 Winding number, derivative of inverse function
Summary:

ˆ Winding number
√
ˆ Riemann surface of z

ˆ Derivative of inverse function

12.1 Winding number, branch point, branch cut

The previous section demonstrates that it is possible to define an angle function along a
continuous path, even if the path is complex and may encircle the origin multiple times.
We refer to such a function as a “choice” of the angle function. It is important to note that
the choice is not unique, as we can add any integral multiple of 2π to the function.
However, regardless of the specific choice made, it is geometrically evident that the
change in angle for any closed curve is always an integral multiple of 2π. With this in
mind, we can now introduce the definition of the winding number.

Definition 12.1. We define the winding number (also known as the index ) of a
closed parametric curve not passing through the origin by (1/2π) times the change
of angle.

The above definition of winding number is relative to the origin. If we want to study
how many times a curve revolves around another point, we can translate the curve suitably
and reduce the problem to the above definition.
The winding number is an integer, and its sign represents the orientation of the closed
path. For instance, the change in angle for the closed curve depicted in Figure 13 is 0.
Additional examples are provided in Fig. 14, illustrating different winding numbers.”
The need for a more involved process to compute the winding number arises due to
the fact that the angle is a multi-function, and the origin serves as a branch point for this
multi-function. We can provide a working definition of a branch point as follows:
Consider a general multi-function F defined on the complex plane. For each z in the
domain, F (z) represents a set of possible values. Now, let’s imagine traveling around a
specific point z0 within a small circle while continuously varying the function value. After

111
 (a) Winding number = −1 (b) Winding number = 1 (c) Winding number = 0

Figure 14: Winding number of a closed curve relative to the origin.

completing one full cycle around z0 , we may observe that the function value does not return
to its initial value. If this phenomenon persists for arbitrarily small circles centered at z0 ,
then we consider z0 to be a branch point of F .
A branch of the multi-function F refers to an ordinary single-valued function that selects
one value from the set F (x) of possible values.
In order to make a continuous choice of value, one approach is to designate a branch
cut and define a branch at every point in the complex plane except for the points located
on the branch cut. The branch cut should be carefully selected in a manner that prevents
any closed cycle from encircling a branch point

√
12.2 Riemann surface of function z
Instead of introducing a branch cut, another approach is to extend the domain to a Riemann
surface. Let’s illustrate the construction of a Riemann surface for the square root function.
The typical domain used in the definition of the complex square root function is given
by
D = C \ {x + iy ∈ C : y = 0, x ≤ 0},
which involves removing the negative real axis from the complex plane. This removed half-
line is referred to as a branch cut. To define a single-valued continuous square root function
in the domain D, we can express z ∈ D in polar coordinates as z = reiθ with r > 0 and
−π < θ < π. We then define
√
f (z) = reiθ/2 .

112
 √
Figure 15: A model for the Riemann surface of the function z.

113
 Figure 16: Image of the curve in Fig. 15 under the square root function.

Under this function, the image of D is the right half-plane. However, the drawback of this
definition is that the square root of points on the negative real axis is not well-defined.
A better approach involves constructing a special surface and using it as the domain
for the square root function. This surface is a two-sheeted covering of the complex plane.
We take two copies of the complex plane and position one on top of the other, such that
each complex number (except the origin) corresponds to two points—one in sheet 1 and
one in sheet 2. The two origins are identified as the same point. We introduce a cut along
the negative real axis and connect the two sheets as shown in Fig. 15. For any positive
real number r, the point −r in sheet 1 is connected to the point −r − ϵi in sheet 2, where
ϵ represents an infinitesimally small real number. Similarly, the point −r in sheet 2 is
connected to −r − ϵi in sheet 1. In this model, we do not remove the negative real axis;
instead, the negative real axis is replicated in both sheet 1 and sheet 2.
A double loop is illustrated in Fig. 15. This loop encircles the origin twice. After
traveling along this curve once, the change in angle is 4π. Under the square root function,
the corresponding image is a closed loop that winds around the origin once (see Fig. 16). It
is important to note that both point A and point D lie on the positive part of the real axis.
However, they belong to different sheets of the Riemann surface. We select the positive
square root for point A in sheet 1, and the negative square root for point D in sheet 2.
When we travel from point A to point D on the Riemann surface, the change in angle is
2π. After applying the square root function, the change in angle is divided by 2, resulting

114
in an angle change of π.

12.3 Derivative of inverse function

We can study the inverse of a complex function by treating it as a vector field and reduce
the problem to real analysis. We will identify a complex functions f (z) and g(z) as a
vector-valued functions f (x, y) and g(x, y) whose domain and range are subsets of R2 .
The following theorem from real analysis is called the inverse function theorem.

Theorem 12.2 ([Rudin] Thm 9.24). Let g be a function from R2 to R2 , and suppose
g is continuously differentiable in a domain D. If g′ is invertible at (x0 , y0 ) ∈ D, then
there exist an open set U ⊆ D that contains (x0 , y0 ), and an open set V ⊆ R2 such
that g is a bijection from U to V and the inverse of g is continuously differentiable
in V .

An equivalent way to state the conclusion is to say that the function g is locally invertible
at the point (x0 , y0 ).
We recall that when g(x, y) is a vector-valued function with components u(x, y) and
v(x, y), then the derivative of g at a point (x0 , y0 ), denoted by g′ (x0 , y0 ), is the 2 × 2 matrix
" #
′ ux (x0 , y0 ) uy (x0 , y0 )
g (x0 , y0 ) = ,
vx (x0 , y0 ) vy (x0 , y0 )

which is also called the Jacobian of g at (x0 , y0 ). Another common notation for this matrix
is
∂(u, v)
∂(x, y)
Suppose that g(z) is a holomorphic function in some domain D and the real and imagi-
nary parts are both continuously differentiable. (This assumption can be relaxed to differ-
entiable, using a corollary of the Cauchy-Gousat theorem.)
The following theorem is the complex analog of this theorem.

115
 Theorem 12.3. Let g(z) be a complex function that is continuously differentiable in
D. If g ′ (z0 ) ̸= 0 at a point z0 ∈ D, then

1. there exist open sets U ⊆ D and V ⊆ C, with z0 ∈ U , such that g is a bijection

from U to V ,

2. the inverse g is continuous differentiable (as a real function) in V , and complex

differentiable in V .

Proof. Suppose the real and imaginary parts of the complex function g(x + iy) are u(x, y)
and v(x, y), respectively. Identify the complex function g(z) with vector-valued function
" #
u(x, y)
g(x, y) = .
v(x, y)

Let z0 be the complex number x0 + iy0 . The derivative g′ at (x0 , y0 ) is an invertible matrix,
because its determinant
ux (x0 , y0 ) uy (x0 , y0 ) ux (x0 , y0 ) −vx (x0 , y0 )
= = ux (x0 , y0 )2 + vx (x0 , y0 )2 = |g ′ (z0 )|2
vx (x0 , y0 ) vy (x0 , y0 ) vx (x0 , y0 ) ux (x0 , y0 )

is nonzero by the assumption g ′ (z0 ) ̸= 0.

By the inverse function theorem from real analysis, we know that g is real-differential
and locally invertible at (x0 , y0 ). Precisely, it means that there exists an open set U that
contains the point (x0 , y0 ), and an open set V in the image of g such that the image of U
under the function g is exactly equal to V , and furthermore, g is injective on U . Hence,
we know that there is a function f that maps from V to U and satisfies g ◦ f = idV and
f ◦ g = idU .
f
V U
g

It remains to show that the inverse of g is complex-differentiable at the point g(x0 , y0 ).

Suppose the image of the point (x0 , y0 ) under g is (r0 , s0 ). We apply the chain rule to
g ◦ f = idV , and get
g′ (x0 , y0 ) · f ′ (r0 , s0 ) = I2×2

116
where I2×2 denotes the 2 × 2 identity matrix. Hence, f ′ (r0 , s0 ) is the inverse of g′ (x0 , y0 )
Because the matrix g′ (x0 , y0 ) is in the form
" #
a −b
b a

for some constant a and b, the matrix f ′ (r0 , s0 ) can be written as

" #
1 a b
,
a2 + b2 −b a

which is a matrix in the form " #

c −d
d c
for some constants c and d, representing the complex number g ′ (z0 )−1 .

Theorem 12.3 shows that if g ′ (z) does not vanish at z = z0 , then the inverse of g is
holomorphic, and the derivative of the inverse function at the point g(z0 ) is

(g ′ (z0 ))−1 .

Example 12.1. The log function is the inverse function of exponential function. To dif-
ferentiate the log function, we have to first pick a branch, so that log(z) is single-valued.
Suppose we take the negative real axis as the branch cut (first part of Fig. 17). For

z ∈ C \ {x + iy : y = 0, x ≤ 0},

we define log(z) as a complex number u + iv such that eu+iv = z and −π < v < π. This
defines a one-to-one onto mapping to the horizontal strip

{u + iv : −π < v < π}.

Differentiate both side of

exp(log(z)) = z,

we obtain
exp(log(z))′ · log(z)′ = 1.

117
Figure 17: Two different branches of complex log function.

118
Using the fact that exp(w)′ = exp(w), we get
1 1
log(z)′ = = .
exp(log(z)) z
We have thus derive
d 1
log(z) =
dz z
for z in the chosen branch C \ {x + iy : y = 0, x ≤ 0}.
We may take another choice of branch cut of the log function. If we pick the positive
real axis as the branch cut, the corresponding area in the z-plane is shown in the second
part of Fig. 17. We can repeat the same calculations and show that the derivative is also
equal to 1/z for this choice of branch cut.
Example 12.2. In this example we compute the derivative of the square root function. Since
the square root function is a multi-valued, we need to first make it single-valued by selecting
a branch cut. Suppose we select the negative real axis as the branch cut. For each z not
√
on the negative real axis, we define z as the unique complex number w on the right half
√
plane such that w2 = z. Hence, we impose the condition that z has positive real part.
(Refer to the first half of Fig. 18.)
√
With this choice of branch cut, z is a bijection from C \ {x + iy : y = 0, x ≤ 0}
to {u + iv : u > 0}. Let z be a complex number not lying on the negative real axis and
g(z) = z 2 denote the squaring function. Differentiate both side of
√
g( z) = z,

we obtain
√ √
g ′ ( z) · ( z)′ = 1.
Since g is the squaring function, we know that g ′ (w) = 2w. Therefore
√ 1
( z)′ = √ .
2 z
As a numerical example, we select the point z = −i, and compute the complex derivative
√ 1 1+i
( z)′ z=−i
= = √ .
2e−iπ/4 2 2
Suppose we pick another branch cut for the square root function. Instead of the negative
real axis, we may choose the positive real axis as the branch cut. The resulting single-
valued square root function is mapping C \ {x + iy : x ≥ 0, y = 0} to the upper half

119
 Figure 18: Two different branches of square root function.

plane {u + iv : v > 0}. Given a complex number z0 not lying on the positive real axis,
√
we now define z as the unique complex number u + iv in the upper half plane such that
√
(u + iv)2 = z. With this choice of branch cut, we have −i = e3πi/4 . So
√ 1 −1 − i
( z)′ z=−i
= = √ .
2e3πi/4 2 2
We thus see that the derivative of the square root function has two possible values, and
the function value depends on the choice of the branch cut.

12.4 Appendix: A more rigorous definition of branch point

We continue with the discussion of branch point.
Suppose we have a path γ(t) defined as a continuous function on the interval [0, b] in
C. At each point along the curve γ(t), we select a value from the multi-function in such a
way that the function values vary continuously as we travel along the curve. Let v(t) be a
continuous function of t for 0 ≤ t ≤ b, such that v(t) ∈ F (γ(t)) for all t.

120
 We are interested in understanding the change in the function value after traversing the
curve. We define the difference in function values as

δ(F, γ, v0 ),

where F represents the multi-function, γ is the path, and v0 denotes the initial function
value, i.e., v0 = v(0) is an element in F (γ(0)).
It is desirable that the change in function value does not depend on the shape of the
curve but only on the positions of the initial and final points, as well as the choice of the
function value at the initial point. To capture this property, we introduce a name. We say
that the multi-function F is independent of path in a region R if, for any closed path γ
that lies entirely within the region R, and no matter how we select function values for each
point on the curve such that the values vary continuously along the curve, the difference in
function value after completing one loop around the path is zero.

Theorem 12.4. If F is independent of path in a region R, then we can select a

branch of F that is continuous on R (without any branch cut).

Proof sketch. We want to construct a branch g of F on the region R such that g(z) ∈ F (z)
for all z ∈ R and g(z) is a continuous function on R.
Let’s choose an arbitrary point z0 from R. We select a single value v0 from the set
F (z0 ). For any other point z in R, we can draw a continuous path from z0 to z. Along
this path, we continuously vary the function value starting from the initial value v0 . (The
assumption that R is path-connected guarantees the existence of such a path.)
We define g(z) as the function value at the end of the path. To complete the proof, we
need to show that g(z) is well-defined, meaning it is independent of the specific shape of
the path from z0 to z, and that it is a continuous function.

Given this preliminary, we can give a formal definition of branch point.

121
Definition 12.5. A point z0 is called a regular point of a multi-function F if there
exists a small disc D centered at z0 such that F (z) is independent of path in the
region D.
A point z0 is called a branch point if, for every open disk D centered at z0 , there
exists a closed path around z0 such that the change in function value δ(F (z), γ, v0 )
after traversing the closed path is nonzero, while every other point in D \ {z0 } is a
regular point.

122
13 Contour integral
Summary

ˆ Integration on the real number line

ˆ Definition of contour integral

In this lecture we first consider integrating a complex-valued function defined on a real

interval, and then extend it to contour integral.

13.1 Review of line integral

In this section we recall the meaning of line integral for real-valued functions.
Let F(x, y) denote a vector-valued function

F(x, y) = (M (x, y), N (x, y))

where M (x, y) and N (x, y) are continuous functions.

We represent a curve parametrically using a vector-valued function r(t) = (x(t), y(t))
for t ∈ [a, b]. Typically, we assume that x(t) and y(t) are both continuously differentiable,
and the tangent vector (x′ (t), y ′ (t)) is nonzero for all t ∈ [a, b].
Given these data, we can define the line integral of F over the curve C represented by
Z Z b
F · dr ≜ (M (x(t), y(t))x′ (t) + N (x(t), y(t))y ′ (t)) dt. (13.1)
C a

Physically, it represents the work done by the force field F on a particular moving along
the curve C.
Mathematically, we interpret the integrand as the dot product of the vector field
(M (x(t), y(t)), N (x(t), y(t))) and the tangent vector r′ (t) = (x′ (t), y ′ (t)). This explains
R
the notation C F · dr. To numerically approximate this integral, we can divide the interval
[a, b] into sub-intervals by creating a partition:

a = t0 < t1 < t2 < · · · tn−1 < tn = b.

We then select a sample point t∗k between tk and tk+1 for k = 0, 1, . . . , n − 1, and compute
the sum
n−1
X h i
F(x(t∗k ), y(t∗k )) · r′ (t∗k )(tk+1 − tk ) .
k=0

123
Under the usual assumption of Riemann integrability, the above sum will converge as the
width of the partition converges to 0. (The width is defined as the maximum of tk+1 − tk
over all indices k.

13.2 Integration of complex-valued function defined on the real line

In this course integration is defined in the sense of Riemann integral. When f (t) is complex-
valued, we define the integral of f (t) as a complex number whose real part is the integral
of the real part of f (t), and the imaginary is the integral of the imaginary part of f (t).

Definition 13.1. Let f be a complex-valued function defined on the interval [a, b]

on the real number line, with u(t) and v(t) as the real and imaginary parts,

f (t) = u(t) + iv(t)

for t ∈ [a, b]. We define the integral of f as

Z b Z b Z b
f (t) dt ≜ u(t) dt + i v(t) dt, (13.2)
a a a

provided that both u(t) and v(t) are Riemann integrable over [a, b].
When the function f (t) is defined on the whole real number line, we define the
improper integral as double limit,
Z ∞ Z b
f (t) dt ≜ lim f (t) dt.
−∞ a→−∞ a
b→∞

Example 13.1. Consider the complex function f (x) = (x − 1) + ix2 , for −1 ≤ x ≤ 1. The
integral of this function on the interval [−1, 1] is
Z 1 Z 1 Z 1
2i
f (x) dx = (x − 1) dx + i x2 dx = −2 + .
−1 −1 −1 3

Example 13.2. We can integrate the complex exponential function by considering the real
and imaginary parts separately.
Z π Z π Z π
it
e dt = cos(t) dt + i sin(t) dt = 2i.
0 0 0

124
 Integral transforms such as Fourier and Laplace transform are one of the major appli-
cations of complex analysis in science and engineering. For instance, the textbook [Stein]
contains a whole chapter on Fourier transform.
Example 13.3. Given any function f (t), the Fourier transform of f (t) is defined as
Z ∞
F (ω) ≜ f (t)e−iωt dt,
−∞

where ω is a real number. If f (t) is a probability distribution function, i.e., if f (t) ≥ 0

R∞
for all t ∈ R and −∞ f (t) dt = 1, then the characteristic function of the corresponding
probability distribution is defined by
Z ∞
ϕ(ω) ≜ f (t)eiωt dt,
−∞

which is basically the same as Fourier transform, except that there is a sign change in the
input variable ω.
Example 13.4. The (two-sided) Laplace transform of f (t) can be viewed as a generalization
of the Fourier transform, Z ∞
ˆ
f (s) ≜ f (t)e−st dt.
−∞
The variable s is complex number. When s is purely imaginary, it reduces to Fourier
transform.
Example 13.5. The Mellin transform of a function f (t) is
Z ∞
f (t)ts−1 dt,
0

where s is a complex number. The Mellin transform of exponential function is called the
Gamma function Z ∞
Γ(s) = e−t ts−1 dt.
0

13.3 Contour integral

The notion of parametric curve in the complex plane is almost the same as in the real case.

125
 Definition 13.2. A parametric curve C represented by

z : [a, b] → C
z(t) = x(t) + iy(t)

is said to be smooth if
(i) x(t) and y(t) are continuously differentiable on [a, b], and
(ii) the vector (x′ (t), y ′ (t)) is not equal to the zero vector for t ∈ [a, b].

We note that the condition in (ii) means that the tangent vector is well-defined at all
points on the curve. We write z ′ (t) = x′ (t) + iy ′ (t) as a tangent vector at t. To emphasize
the direction/orientation of the curve from t = a to t = b, we sometime use the word
“contour” instead of “path”. If a curve can be divided into finitely many parts and each
part is smooth, then we call it a piece-wise smooth curve.
Remark. In the rest of the lecture notes, all curves are piece-wise smooth curve. Because
a piece-wise smooth curve can be treated as a concatenation of smooth curves, we will
describe the computation of a complex integral on a smooth curve.
One way to motivate contour integral of function f (z) is to divide a smooth curve C in
Definition 13.2 into many small parts, and consider the Riemann sum
n−1
X
f (zk∗ ) · ∆zk
k=0

where zk∗ is a sample point in the k-th part, and ∆zk is the difference of the k-th part.
If f is a continuous function, Riemann sum will converge to a constant when we take n
approaching infinity. The limit is then called the contour integral of f (z) over C. (See
Fig. 19)

126
 Figure 19: Division of a contour in constructing a Riemann sum

Definition 13.3. Given a continuous complex-valued function f : C → C and a

smooth curve C, the complex integral (or contour integral ) of f over C is defined as
Z Z b
f (z) dz ≜ f (z(t)) · z ′ (t) dt (13.3)
C a

with the right-hand side defined as in Definition 13.1.

If C can be decomposed into n parts C1 , C2 , . . . , Cn , such that each part is a smooth
contour, the integral of f (z) over C is defined as the sum
Z n Z
X
f (z) dz ≜ f (z) dz.
C k=1 Ck

Remark. Definition 13.3 is a generalization of Definition 13.1. The complex integral for a
function f : [a, b] → C can be regraded as a contour integral with the contour on the real
axis from the complex point z = a to z = b. If a < b, we can parametrize the curve by
z(t) = t + 0i for t ∈ [a, b].
Notation: We write the integral as
I
f (z) dz
C

if we want to emphasize that the curve C is closed.

127
Example 13.6. Integrate f (z) = z̄ over the line segment from z1 = 0 to z2 = 2i. We
parameterize the line segment using the y coordinate as the parameter,

C : z(y) = 0 + iy

for 0 ≤ y ≤ 2. The tangent vector is constant i.

Z Z 2 Z 2 h y 2 i2
z̄ dz = 0 + iy · (i) dy = (−iy)i dy = = 2.
C 0 0 2 0

Example 13.7. Consider the same complex function f (z) = z̄ as in the previous example,
but now we integrate over a semi-circle C ′ with center i and radius 1, from the point z1 = 0
to the point z2 = 2i. We represent the semi-circle by angle θ in the range [−π/2, π/2],

z(θ) = cos θ + i(1 + sin θ).

The tangent vector is

z ′ (θ) = − sin θ + i cos θ.

Substitute the above data to the definition of contour integral, and compute
Z Z π/2
f (z) dz = (z(θ))∗ · z ′ (θ) dθ
C −π/2
Z π/2
= (cos θ − i(1 + sin θ)) · (− sin θ + i cos θ) dθ
−π/2
Z π/2
= cos θ + i(1 + sin θ) dθ
−π/2

= 2 + iπ.

The answer is not the same as in the previous example.

Examples 13.6 and 13.7 illustrate that if we change the contour, the contour integral
will be different in general.
Example 13.8. Integrate f (z) = z 2 over the curve C = {x + iy ∈ C : y = x2 } from 0 to
1 + i. We use the representation z(x) = x + ix2 for 0 ≤ x ≤ 1, with the variable x as the

128
parameter. The derivative of z(x) is z ′ (x) = 1 + 2xi.
Z Z 1
2
z dz = (x + ix2 )2 · (1 + 2xi) dx
C 0
Z 1
= (x2 − 5x4 ) + i(4x3 − 2x5 ) dx
0
 x3 1 x6 1
− x5 0 + i x4 −

=
3 3 0
2 2
= − + i.
3 3

We can interpret contour integral in terms of line integrals. The first one is through
line integral for real vector field. Suppose

f (x + iy) = u(x, y) + iv(x, y)

and the curve C is parameterized by z(t) = x(t) + iy(t) for t ∈ [a, b]. We can decompose
the contour integral into two parts.
Z Z b
f (z) dz = (u(x(t), y(t)) + iv(x(t), y(t))) · (x′ (t) + iy ′ (t)) dt
C a
Z b
= u(x(t), y(t))x′ (t) − v(x(t), y(t))y ′ (t) dt
a
Z b
+i u(x(t), y(t))y ′ (t) + v(x(t), y(t))x′ (t) dt
Z a Z
= udx − vdy + i udy + vdx.
C C
We are actually computing two line integral simultaneously. Indeed, some people take the
above as the definition of contour integral.
Recall from Section 9.6 that there is a physical interpretation by considering the vector
field (M (x, y), N (x, y)) = (u(x, y), −v(x, y)), which is called Polya’s vector field of f (z). If
we make a substitution u(x, y) = M (x, y) and v(x, y) = −N (x, y), we can write the contour
integral as
Z Z Z
(M (x, y) − iN (x, y)) dz = M dx + N dy + i M dy − N dx.
C C C
The real and imaginary parts are precisely the work integral and flux integral of the vector
field (M (x, y), N (x, y)). This is a physical interpretation of the contour integral.

129
13.4 Appendix: Definition of line integral and contour integral using
differential form
Another interpretation of line is through the differential form

ω = M (x, y)dx + N (x, y)dy.

At the point (x0 , y0 ), the linear functional ω(x0 ,y0 ) maps a tangent vector v = (v1 , v2 ) in
T(x0 ,y0 ) to
ω(x0 ,y0 ) (v) = M (x0 , y0 )v1 + N (x0 , y0 )v2 .

(See Section 10.1) If the curve r(t) passes through the point (x0 , y0 ) at time t0 , we can
obtain the tangent vector at (x0 , y0 ) as (x′ (t0 ), y ′ (t0 )). We apply the differential form
ω = M (x, y)dx + N (x, y)dy to the tangent vector (x′ (t0 ), y ′ (t0 )) to obtain a scalar function

ω(x(t),y(t)) (r′ (t)), (13.4)

which is a function of parameter t.

The integral of differential form M (x, y)dx + N (x, y)dy along the curve r(t) is defined
as the integral of the function in (13.4),
Z Z b
ω≜ ω(x(t),y(t)) (r′ (t)) dt. (13.5)
C a

If we expand it in terms of the component functions M and N , we see that this is the same
as the line integral in (13.1).
We can also define contour integral using complex differential form. Similar to (13.5),
we can regard
ω = f (z)dz = f (x + iy)(dx + idy)

as a mathematical object. It defines a (complex) linear functional ωx0 +iy0 on the point
x0 + iy0 , mapping a tangent vector τ ≜ τ1 + iτ2 in the tangent space Tx0 +iy0 to

ωx0 +iy0 (τ ) = f (x0 + iy0 ) · (τ1 + iτ2 )

Suppose the contour γ(t) is represented by z(t) = x(t) + iy(t) and it passes through the
point x0 + iy0 at t = t0 , we can obtain the tangent vector at this point by

γ ′ (t0 ) = x′ (t0 ) + iy ′ (t0 ).

130
We obtain a complex-valued function

f (x(t) + iy(t)) · γ ′ (t). (13.6)

The integral of the complex-valued function in (13.6) is exactly the same as the contour
integral
Z b Z b
f (x(t) + iy(t)) · (x′ (t) + iy ′ (t)) dt = f (z(t)) · z ′ (t) dt.
a a
It is now obvious that the line integral and contour integral belong to the same class of
objects through differential form.

131
14 Calculus of contour integral
Summary:

ˆ Independence of parameterization

ˆ Path concatenation

ˆ Primitive function / anti-derivative

14.1 Independence of parameterization

There are more than one way to represent a curve in parametric form, but the value of the
contour does not depend on the parameterization.

Theorem 14.1. The answer obtained from the definition of contour integral does
not depend on the parameterization of the curve.

Proof. Suppose a curve C is represented in two different ways. The first parameterization
is z(t) for a ≤ t ≤ b, and the second parameterization is w(t) for c ≤ t ≤ d. We want to
show that Z b Z d
f (z(t))z ′ (t) dt = f (w(τ ))w′ (τ ) dτ. (14.1)
a c
Because both z(t) and w(τ ) represent the same curve, there is a monotonically increasing
function α : [c, d] → [a, b] such that

w(τ ) = z(α(τ )).

We can illustrate the relationship among w, z and α by the diagram

w z
[c, d] C [a, b]

Apply chain rule and differentiate with respect to τ ,

w′ (τ ) = z ′ (α(τ ))α′ (τ ).

132
We make a change of variable t = α(τ ), and substitute dt by dt = α′ (τ )dτ ,
Z b Z d
f (z(t))z ′ (t) dt = f (z(α(τ )))z ′ (α(τ )) · α′ (τ ) dτ
a c
w′ (τ ) ′
Z d
= f (w(τ )) ′ · α (τ ) dτ
c α (τ )
Z d
= f (w(τ ))w′ (τ ) dτ.
c

This proves the equality in (14.1).

The contour integral satisfies the usual operational rule as for line integral. For example,
for a fixed curve C, we have linearity property
Z Z Z
af (z) + bg(z) dz = a f (z) dz + b g(z) dz,
C C C

where a and b are complex constants, and f and g are complex functions.

14.2 Varying the paths

Even though contour integral does not depend on parameterization, it does depend on the
direction of the path.

Definition 14.2. The negation or the reverse of a curve C is the curve with the
same locus with reverse direction. The negation of C is denoted by −C.

Without loss of generality, we may suppose the range of the parameter starts from 0. If
a contour C is represented by z(t), for t going from 0 to b, then −C can be represented by
z(b − t). Similar to the proof of Theorem 14.1, we can apply the chain rule to obtain the
following relationship between the integrals on a contour in opposite direction.
Z Z
f (z) dz = − f (z) dz. (14.2)
−C C

This is analogous to the formula

Z b Z a
g(x) dx = − g(x) dx.
a b

in calculus.

133
 i

C3
C1

C2 1

Figure 20: The three contour in Example 14.1.

For integer n and contour C, we use the notation nC to represent going through the
curve C n times. This notation is particular convenient if C is a closed curve and we want
to say “go around the curve C multiple times. For example, given a closed curve C, the
symbol “2C” means going around C counter-clockwise 2 times. Thus, we have
Z Z
f (z) dz = 2 f (z) dz.
2C C

Example 14.1. Consider the function f (z) = xz = x2 + ixy. Compute the complex integral
from i to 1 via (i) a vertical line segment from i to 0 and then a horizontal line segment
from 0 to 1, (ii) a single line segment from i to 1.
Let C1 , C2 , and C3 be paths as illustrated in Fig. 20.
The function f (z) is zero on the path C1 . Therefore we can directly see that
Z
f (z) dz = 0.
C1

For contour C2 , we can represent it by C2 using x as the parameter

z(x) = x, z ′ (x) = 1.

Z Z 1
f (z) dz = (x2 ) · (1) dx = 1/3.
C2 0

134
 Therefore, the contour integral along C1 and C2 is
Z
1 1
f (z) dz = 0 + = .
C1 +C2 3 3

If we go directly from i to 1 through C3 , we can parameterize the contour by the x

coordinate,
z(x) = x + i(1 − x)

for 0 ≤ x ≤ 1. The tangent vector is constantly equal to 1 − i. We then compute the

contour integral along C3 by
Z Z 1
f (z) dz = (x2 + ix(1 − x))(1 − i) dx
C3 0
1 i i
= (1 − i)( + − )
3 2 3
1
= (1 − i)(2 + i)
6
3−i
= .
6
If we go around the triangle counter-clockwise, along the curves C1 , C2 and −C3 , the
integral of f (z) is
−1 + i
I
1 1 i
f (z) dz = − + = .
C1 +C2 +(−C3 ) 3 2 6 6

14.3 Fundamental theorem of calculus for complex integral

We define the notion of anti-derivative as in calculus.

Definition 14.3. A function F (z) is called a primitive or anti-derivative of f (z) if

F ′ (z) = f (z) for all z in the domain of definition of f (z).

Implicit in the definition is that F (z) is a holomorphic function. Because the derivative
of a complex function is continuous, it also implicit in the definition that f (z) is a continuous
function.
Remark. Recall from the end of Section 10.3 that when F (z) is holomorphic, the differential
of F equals
dF = F ′ (z)dz = f (z)dz.

135
In general, if there exists a holomorphic function F (z) such that f (z)dz, we say that f (z)dz
is an exact differential form. Hence, the condition in Definition 14.3 is the same as saying
that the complex differential f (z)dz is exact.
If f (z) has a primitive, then the calculation of contour integral becomes much easier.

Theorem 14.4 (Fundamental theorem of calculus for complex integration). Suppose

f (z) is a continuous complex function defined on a domain D. If f has primitive
F (z) in D, then for any smooth contour C in D from z1 to z2 , we have
Z
f (z) dz = F (z2 ) − F (z1 ).
C

We note that the domain D in Theorem 14.4 needs not be simply-connected. The
domain could be multiply connected. We can check that the proof below does not require
that the domain is simply connected.

Proof. Suppose C is a smooth curve in D parameterized by z(t) for a ≤ t ≤ b. We want to

calculate
Z Z b
f (z) dz ≜ f (z(t))z ′ (t) dt
C a
Z b
d
= F (z(t)) dt.
a dt

We have applied the chain rule of differentiation in the last step. Hence, we see that the
integrand can be written as the derivative of some function. To facilitate notation, we may
define g(t) = F (z(t)).
The remaining steps is to apply the fundamental theorem of calculus to the real and
imaginary part. Suppose real and imaginary parts of g(t) are u(t) and v(t), respectively,
such that
g ′ (t) = u′ (t) + iv ′ (t).

136
Integrate real and imaginary parts separately,
Z Z b Z b
′
f (z) dz = u (t) dt + i v ′ (t)
C a a
= u(b) − u(a) + i(v(b) − v(a))
= g(b) − g(a)
= F (z(b)) − F (z(a))
= F (z2 ) − F (z1 ).

This proves the fundamental theorem for contour integral.

Theorem 14.4 can be extended easily to piece-wise smooth curve. Suppose a contour C
is the concatenation of n smooth contours,

C = C1 + C2 + · · · + Cn ,

and for j = 1, 2, . . . , n, Cj starts at zj−1 and ends at zj . If f (z) has a primitive function
F (z) in a domain that contains all Cj ’s, then
Z Z Z Z


f (z) dz = + +··· + f (z) dz
C C1 C2 Cn

= (F (z1 ) − F (z0 )) + (F (z2 ) − F (z1 )) + · · · + (F (zn ) − F (zn−1 ))

= F (zn ) − F (z0 ).

14.4 Independence of path

R
If f (z) has a primitive, then the value of C f (z) dz only depends on the two end points of
the contour and the anti-derivative F (z), but does not depend on the shape of the contour.

Definition 14.5. Given a continuous complex function f (z) on a domain D, we say

R
that the integral C f (z) dz is independent of path in domain D if for any curve C,
R
the contour integral C f (z) dz only depends on the start and end points, but does
not depend on the shape of the curve C.

Theorem 14.4 says that the existence of primitive implies independence of path.

137
Example 14.2. We denote the circle with radius r, positive orientation, centered at the
origin, by Cr . For nonegative integer n, the integral
I
z n dz
Cr

n+1
is equal to 0, because z n has primitive zn+1 throughout the complex plane.
When n = −1, we cannot apply Theorem 14.4, even though we have (log z)′ = 1/z
for all z ̸= 0. The reason we cannot apply Theorem 14.4 is that the log function fails
to be continuous, and hence is not holomorphic, in the punctured complex plane C \ {0}.
R
(We need a branch cut to make it holomorphic). Indeed the integral C 1/z dz is not
equal to zero. We can compute the integral by parameterizing the circle with radius r by
γ(t) = reit , for t ∈ [0, 2π]. The derivative of γ(t) with respect to t is γ ′ (t) = rieit . Using
this parameterization, we have
Z Z 2π Z 2π
−1 −iθ iθ
1/z dz = r e (rie ) dθ = i dθ = 2πi.
Cr 0 0

When n ≤ −2, the function f (z) = z n is holomorphic in the punctured plane C \ {0},
n+1
and have primitive function zn+1 for z ∈ C \ {0}. By Theorem 14.4, the integral Cr z n dz
H

is zero.
We can double check the integral of z n along Cr is
Z Z 2π
z n dz = rn eint (rieit ) dt
Cr 0
Z 2π
= rn+1 iei(n+1)t dt.
0

When the value of n + 1 in the exponent is not equal to zero, we can evaluate the integral
using the anti-derivative,
Z h ei(n+1)t i2π
z n dz = rn+1 i = 0.
Cr i(n + 1) 0

We summarize this example as a theorem.

138
 Theorem 14.6. Let Cr denote the circle centered at the origin, with radius r and
counter-clockwise orientation.

I 2πi if n = −1,
z n dz =
Cr 0 if n ∈ Z \ {−1}.

As in multi-variable calculus, we have an equivalent condition for path independence.

Theorem 14.7. Suppose f (z) is a complex-valued continuous function defined on a

domain D. The followings are equivalent:
R
(a) For any piece-wise smooth curve C in domain D, the integral C f (z) dz only
depends on the start and end point of C, but not on the shape of C.
H
(b) C f (z) dz = 0 for any closed contour C.

R
Proof. (a) ⇒ (b). Suppose the integral C f (z) dz is independent of path. Let C be a closed
contour. We pick two distinct points z1 and z2 on C, and divide C into two parts. The
first part C1 is the portion of C that goes from z1 to z2 . The second part C2 is the portion
of C that goes from z2 back to z1 .
I Z Z
f (z) dz = f (z) dz + f (z) dz
C C1 C2
Z Z
= f (z) dz − f (z) dz.
C1 −C2

Since C1 and −C2 have the same start point and end point, the two integral in the last line
have the same value. Therefore, their difference must be zero.
H
(b) ⇒ (a). Suppose C f (z) dz is equal to 0 for any closed contour C. Let C1 and C2
be two contours with the same start point and end point. Concatenate C1 with the reverse
path −C2 . The resulting contour C1 + (−C2 ) is a closed, and by assumption, the integral

139
on C1 + (−C2 ) is zero. We thus obtain
Z Z
f (z) dz + f (z) dz = 0
C1 −C2
Z Z
f (z) dz − f (z) dz = 0
C1
ZC2 Z
f (z) dz = f (z) dz.
C1 C2

140
15 Cauchy-Goursat theorem
Summary

ˆ Arc length of a smooth curve.

ˆ ML inequality

ˆ Cauchy-Goursat theorem for triangle

15.1 Cauchy theorem version 1

If we assume that f (z) is sufficiently smooth, then we can proof the Cauchy theorem easily
as a consequence of Green theorem.

Theorem 15.1 (Cauchy theorem version 1). If C is a simple closed curve enclosing
a simply connected region D and f (z) is continuously differentiable in D, then
I
f (z) dz = 0.
C

Proof. Suppose u and v are the real and imaginary parts of f , respectively. The integral
H
C f (z) dz can be written as
I I I
f (z) dz = u dx − v dy + i u dy + v dx.
C C C

By Green’s theorem, the first integral on the right-hand side equals

ZZ  
− vx − uy dxdy

and the second integral equals ZZ  

ux − vy dxdy.

Because Riemann-Cauchy equations holds for functions u and v, both of them are equal
to 0.

The assumption that the partial derivatives are continuous is redundant. We can prove
the conclusion in Theorem 15.1 by assuming that f (z) is complex differentiable in the
domain D.

141
15.2 The length of a contour and ML inequality
We first prove a few useful results.

Theorem 15.2 (Triangle inequality for complex integral). Suppose g(t) is a contin-
uous complex function from [a, b] to C. Then
Z b Z b
g(t) dt ≤ |g(t)| dt. (15.1)
a a

Rb
Proof. Let α denote the complex integral α ≜ a g(t) dt. If α = 0, then (15.1) is obviously
true, because the left-hand side is zero. Hence, we can suppose α ̸= 0. Write α in polar
form α = reiθ . (θ is defined because α ̸= 0.)
Then,
Z b
−iθ
e g(t) dt = r
a
is a real number. We can re-write it as
Z b
e−iθ g(t) dt.
a

Let u(t) and v(t) be the real and imaginary parts of e−iθ g(t), respectively. We get
Z b Z b
u(t) dt = r and v(t) dt = 0.
a a

But
p p
u(t) ≤ u2 (t) ≤ u2 (t) + v 2 (t) = |g(t)|.

By the monotonic property for real integration,

Z b Z b Z b
g(t) dt = r = u(t) dt ≤ |g(t)| dt.
a a a

To compute the arc length of a path that is parameterized by z(t) = x(t) + iy(t), we
can apply a result from multi-variable calculus:
Z bp
x′ (t)2 + y ′ (t)2 dt.
a

142
We express the above integral using complex variable, by observing that the integrand can
be written as
x′ (t)2 + y ′ (t)2 = |x′ (t) + iy ′ (t)| = |z ′ (t)|.
p

Hence, the arc length of a curve C can be formally defined as follows.

Definition 15.3. The length of a smooth curve C, represented by z(t) for a ≤ t ≤ b,

Rb
is defined as a |z ′ (t)| dt.

Remark. We can also define the length of a smooth curve using complex differential dz =
dx + idy: Z
|dz|.
C
p
The symbol |dz| represents the differential dx2 + dy 2 , which maps a tangent vector v =
p
v1 + iv2 to v12 + v22 . We now apply the integral in (13.5).
The ML inequality gives a bound on the absolute value of a contour integral in terms
of the length of the path and the largest absolute value of the integrand on the path.

Theorem 15.4 (ML inequality). If |f (z)| ≤ M for z on a smooth curve C and the
length of C is equal to L, then
Z
f (z) dz ≤ M L.
C

Proof. Apply the triangle inequality (15.1) to get

Z Z b Z b
′
f (z) dz = f (z(t))z (t) dt ≤ f (z(t))z ′ (t) dt.
C a a

This yields
Z Z b Z b
′
f (z) dz ≤ |f (z(t))| · |z (t)| dt ≤ M |z ′ (t)| dt = M L.
C a a

The next example illustrates an application of Theorem 15.4 in proving an inequality.

The original inequality does not involve any integral, but the proof is facilitated by complex
integral.

143
Example 15.1. For any complex numbers z1 and z2 with negative real part, prove that

|ez2 − ez1 | ≤ |z2 − z1 |.

We note that ez2 − ez1 is the same as C ez dz for any path from z1 to z2 . We take C to be
R

the direct path from z1 to z2 in the form of a straight line. The integrand ez has modulus
less than 1, because z1 and z2 have negative real parts. The length of the line segment
between z1 and z2 is precisely |z2 − z1 |. Hence, by the ML inequality,
Z
z2 z1
|e − e | = ez dz ≤ 1 · |z2 − z1 |.
C

15.3 Cauchy-Goursat theorem for triangular domains

We need the following theorem, which is usually called “Cantor’s intersection theorem”
from point-set topology.

Theorem 15.5. Suppose K1 , K2 , K3 , . . . are nonempty compact sets (in some metric
space). If K1 , K2 , K3 , . . . is a decreasing sequence, i.e.,

K1 ⊇ K2 ⊇ K3 ⊇ · · ·

then ∩∞
j=1 Kj is not empty.

We note that the conclusion of this theorem will fail without the compactness assump-
tion. For example, consider a sequence of intervals

A1 ⊇ A2 ⊇ A3 ⊇ · · ·

defined by Ak ≜ [k, ∞) for k = 1, 2, 3, . . .. The sets are unbounded, and hence are not
compact. Their intersection is empty.
If we let Bk be the open interval (0, 1/k), for k = 1, 2, . . ., we have a decreasing sequence
of open intervals, and the intersection is empty.
The following version of Cauchy theorem is due to Goursat.

144
 Im

C3

C4
C1 C2

Re

Figure 21: Division of a triangular regions into four equal parts.

Theorem 15.6 (Cauchy-Goursat for triangles). Suppose f (z) is holomorphic in a

domain D and T is a triangle contained inside D. Then
I
f (z) dz = 0,
∂T

where ∂T denotes the boundary of triangle T .

Remark. It is important to note that the theorem only assumes that the derivative of f (z)
exists. We do not need to assume that f ′ (z) is a continuous function of z.

Proof. Let C be the boundary of triangle T with counter-clockwise orientation. Let L be

the perimeter of triangle T .
R
Suppose C f (z) dz = I. We want to show that |I| = 0. The following figure depicts
the triangle T on which f (z) is holomorphic.
Divide the triangle T into four equal parts. The four sub-triangles are similar to each
other, and have the same area and perimeter. Let Ci be the boundary of the i-th sub-
triangle, for i = 1, 2, 3, 4, all in the counter-clockwise direction (See Fig. 21).

145
 The integrals on the internal lines cancel. Hence,
Z Z Z Z
I= f (z) dz + f (z) dz + f (z) dz + f (z) dz.
C1 C2 C3 C4

By triangle inequality,
Z Z Z Z
|I| ≤ f (z) dz + f (z) dz + f (z) dz + f (z) dz .
C1 C2 C3 C4

Suppose T (1) is the sub-triangle that has the largest integral in magnitude,
Z Z
f (z) dz = max f (z) dz
∂T (1) j=1,2,3,4 Cj

(Here ∂T (1) means the boundary of T (1) .)

This implies
|I|
Z
≤ f (z) dz
4 ∂T (1)

by the choice of T (1) .

Recursively, for k ≥ 1, divide T (k) into four equal parts, and let T (k+1) be the sub-
triangle that has the largest integral in absolute value. Thus,

T ⊇ T (1) ⊇ T (2) ⊇ T (3) ⊇ T (4) ⊇ · · ·

and
|I|
Z
≤ f (z) dz
4k ∂T (k)

for k = 1, 2, 3, . . . The perimeter of T (k) is L/2k , for k ≥ 1.

By Cantor intersection theorem (Theorem 15.5), there is a point z0 that lies inside T (k)
for all k. Since f is assumed to be holomorphic in T , it is complex differentiable at the point
z0 . Suppose that the derivative is f ′ (z0 ). By the definition of complex differentiability, we
have
f (z) = f (z0 ) + f ′ (z0 )(z − z0 ) + ϵ(z)(z − z0 ) (15.2)

where ϵ(z) is a function such that |ϵ(z)| → 0 as z → z0 .

For any k, we can integrate both sides of (15.2) on the boundary of T (k) .
Z Z
f (z) dz = f (z0 ) + f ′ (z0 )(z − z0 ) + ϵ(z)(z − z0 ) dz.
∂T (k) ∂T (k)

146
Because f (z0 ) + f ′ (z0 )(z − z0 ) is a linear function in z, it has an anti-derivative. By
Theorem 14.4, ∂T (k) f (z0 ) + f ′ (z0 )(z − z0 ) dz is zero. So,
R

Z Z
f (z) dz = ϵ(z)(z − z0 ) dz
∂T (k) ∂T (k)

for any k = 1, 2, 3, . . .
We upper bound |z − z0 | by the diameter of T (k) . In general, the diameter of a set S in
C is the supremum of the Euclidean distance between two points in S,

diam(S) ≜ sup {distance between p and p′ }.

p,p′ ∈S

For example, the diameter of a circle as defined above is the diameter in the usual sense.
The diameter of an ellipse is the length of the major axis.
Let d denote the diameter of original triangle T . We know that the diameter of d(k)
equals d/2k , for k = 1, 2, 3, . . ., because in each subdivision step we scale down the triangle
by a factor of 2. By the very definition of diameter, we have |z − z0 | ≤ d(k) for all z ∈ T (k) .
We now fix a small and positive real number δ, and choose a sufficiently large k such
that |ϵ(z)| ≤ δ for all z ∈ T (k) . This is possible as ϵ(z) → 0 as z → z0 . For each k, we can
bound |ϵ(z)(z − z0 )| by δd/2k . Apply ML inequality (Theorem 15.4),
Z
1
|I| ≤ ϵ(z)(z − z0 ) dz
4k ∂T (k)
d L
≤ δ k · k.
2 2
Hence
|I| ≤ δdL.

Since d and L are constant and δ can be arbitrarily small, we must have I = 0.

We can immediately extend Cauchy theorem to domain with polygonal shape.

Theorem 15.7 (Cauchy theorem for rectangle). Suppose f (z) is holomorphic in an

open rectangle R, then I
f (z) dz = 0.
∂R

147
Proof. Pick one of the diagonal of the rectangle R, and consider the two triangles obtained
by dividing R along this diagonal. The integral along the boundary of R is the sum of the
two integrals along the two triangles. Since the integrals on the triangles are both 0, the
sum is also 0.

With the same argument as in the proof of Theorem 15.7, we can show that Cauchy-
Goursat theorem is valid for polygons, such as parallelogram, trapezoid, etc.

148
16 Local primitive function and Cauchy theorem for regions
of various shapes
Summary
ˆ Construction of local primitive function

ˆ Cauchy theorem for convex and star-shape regions

ˆ Application to the definition of complex log function

If a primitive function exists, Theorem 14.4 guarantees that the integral of f (z) depends
only on the initial and final points of the contour, regardless of the shape of the contour.
However, it is not always the case that a primitive function exists for every function f (z).
As we observed in Example 14.2, the integral of the function 1/z over a circle centered at
the origin is not zero. However, if we introduce a branch cut and prohibit closed paths from
R
encircling the origin, we can conclude that C 1/z dz = 0 for any closed curve C within the
domain C \ {x + iy :, y = 0, x ≤ 0}. This motivates us to consider a more restricted notion
of a primitive function, as the existence of such a function may be possible within a smaller
domain.

16.1 Application to the evaluation of real integral

We illustrate an application of Cauchy theorem to the evaluation of Fourier transform.
Consider the probability distribution function of the standard normal distribution
1 2
f (x) = √ e−x /2 .
2π
Our objective is to demonstrate that for any t ∈ R, the Fourier transform of f (x) is given
by Z ∞
1 2 2
√ e−x /2 e−itx dx = e−t /2 .
−∞ 2π
It is sufficient to consider the case where t > 0. We examine the complex function
2
f (z) = e−z /2 and integrate it along the rectangular contour depicted in Fig. 22.
2
Since e−z /2 is an entire function, we can apply Cauchy theorem (Theorem 15.7) to see
that the integrals along C1 , C2 , C3 and C4 are all zero,
Z Z Z Z
+ + + = 0.
C1 C2 C3 C4

149
 Im

C3
−R + it R + it

C4 C2
Re
−R R
C1

Figure 22: Closed contour used in the derivation of the Fourier transform of the Gaussian
probability distribution function.

The integral along C1 from −R to R on the real axis equals

Z Z R
−z 2 /2 2 /2
e dz = e−x dx.
C1 −R
√
This integral tends to 2π as R → ∞.
The contour C2 is a vertical straight line segment from R to R + it. We parameterize
C2 by z(y) = R + iy for 0 ≤ y ≤ t. The derivative is constant z ′ (y) = i. Apply the triangle
inequality (Theorem 15.2) to obtain an upper bound the integral over C2 ,
Z Z t
2 /2
f (z) dz = e−(R+iy) · i dy
C2 0
Z t
2 −y 2 )/2
≤ e−(R dy
0
−(R2 −t2 )/2
≤e · t.

For any fixed t, the upper bound approaches 0 as R approaches infinity. Therefore the
integral of f over contour C2 tends to 0 as R tends to ∞.
Likewise, we can show that the integral over contour C4 approaches 0 as R → ∞.
Finally we consider the contour C3 . We represent the reversed contour −C3 using the
parameterization z(x) = x + it, where x ranges from −R down to R. The derivative is

150
z ′ (x) = 1. Using this parameterization, we have
Z Z R
2 /2
f (z) dz = e−(x+it) dx
−C3 −R
Z R
t2 /2 2 /2
=e e−x e−ixt dx.
−R

By Cauchy theorem for rectangle, for each R > 0, we obtain the equation
Z Z Z Z
− f (z) dz = f (z) dz + f (z) dz + f (z) dz.
C3 C1 C2 C4

Taking limit as R → ∞, we see that the integrals over C2 and C4 approaches zero, and the
√
integral over C1 approaches 2π. Thus, we have

2
Z ∞
2 √
et /2 e−x /2 e−ixt dx = 2π.
−∞

After performing some algebraic operations, we obtain the desired formula for t > 0:
Z ∞
1 2 2
√ e−x /2 e−ixt dx = e−t /2 .
2π −∞
The proof for t < 0 follows the exact pattern, except that the contour is modified to
lies below the real axis.

16.2 Existence of local primitive function

Definition 16.1. We say that a function f (z) has a primitive at a given domain D
if we can find a holomorpic function F (z) defined on D such that F ′ (z) = f (z) for
all z ∈ D. We will sometime call F (z) a local primitive of f (z) in the domain D.

We note the domain D in the above definition may not necessarily be the entire domain
of definition of f (z). Instead, it can be a small open set that contains a specific point or a
restricted region within the overall domain of f (z).
There are several sufficient conditions that can guarantee the existence of local primitive
function. The key part of the argument is contained in the following theorem.
Consider a continuous complex function f (z) defined on a domain D (which is assumed
to be open and connected, but may not be simply connected). We pick a base point z0

151
 Figure 23: Constructing local primitive function by integral.

in D. Since D is connected, there is at least one path from z0 to z that lies completely
inside D. For each point z ∈ D, we select a path from z0 to z, and denote it by Cz . We
then define a candidate function for the primitive function as
Z
F (z) ≜ f (z) dz. (16.1)
Cz

Furthermore, we assume that the paths are selected in a way that if z1 and z2 are two
points connected by a line segment lying inside the domain D, then the following equation
holds: Z
F (z2 ) = F (z1 ) + f (z) dz. (16.2)
z1 →z2
Here, we introduce a non-standard notation
Z
f (z) dz,
z1 →z2

which denote the contour integral of f (z) on the line segment from z1 to z2 . (See Fig. 23)
If the line segment from z1 and z2 does not lie entirely inside D, then there is nothing to
R
check in (16.2). The integral z1 →z2 f (z) dz makes no sense in this case anyway.

Theorem 16.2. Assuming that f (z) is a continuous function defined on a domain

D, and that F (z) defined in (16.1) satisfies the condition in (16.2), then F (z) is
holomorphic in D with derivative f (z).

We note that the axiom of choice is used in defining F (z).

152
Proof. Let z be a point in domain D. Since D is an open set (as required by the definition
of a domain), we can draw an open disc D(z, r) centered at z with radius r such that D(z, r)
is contained inside D. A point in D(z, r) can be written as z + h, where h is a complex
number with modulus less than r.
Let h be a nonzero complex number with modulus less than r. We want to compute
the difference F (z + h) − F (z) and divide it by h, and show that the result tends to f (z)
as h → 0. According to the assumption on how the function F (z) is defined, the difference
F (z + h) − F (z) equals the integral obtained by integrating f (z) from z to z + h along the
line segment connecting z and z + h. This is valid because the line between z and z + h is
contained within the open disc D(z, r) and, therefore, also within the domain D.
By assumption, we have
Z
F (z + h) = F (z) + f (w) dw.
z→z+h

We re-write it as
F (z + h) − F (z)
Z
1
= f (w) dw.
h h z→z+h
On the other hand, if we integrate the constant function f (z) along the same line segment,
we get Z
f (z) dw = hf (z),
z→z+h
This gives
F (z + h) − F (z)
Z Z
1 1
− f (z) = f (w) dw − f (z) dw
h h z→z+h h z→z+h
Z
1
= (f (w) − f (z)) dw.
h z→z+h
Since f (z) is continuous at z by assumption, for any small positive real number ϵ0, we
can find a sufficiently small positive real number δ such that |f (w) − f (z)| < ϵ for all w
satisfying |w − z| < δ. Then, for a complex number h with |h| < min(r, δ), we can apply
the ML inequality to obtain
F (z + h) − F (z) 1
− f (z) ≤ ϵ · (length of line segment from z to z + h) = ϵ
h |h|
because the length of the line segment from z to z + h is exactly equal to |h|. Since ϵ can be
arbitrarily chosen, we have proved that for any given real number ϵ > 0, we can guarantee

153
that
F (z + h) − F (z)
− f (z) ≤ ϵ
h
for all complex numbers h with |h| ≤ min(δ, r), where δ is determined from ϵ through the
continuity of f (z). This is equivalent to saying that

F (z + h) − F (z)
lim = f (z)
h→0 h
This completes the proof that f (z) has a primitive function F (z) in the domain D.

Using this technical result, we can now prove that the existence of local primitive func-
tion is logically equivalent to independent of path.

Theorem 16.3. Let f be a continuous complex-valued function defined in a region

R
region D. Then integral C f dz is independent of path in region D if and only if
f (z) has a primitive function F (z) in the domain D.

Proof. (⇐) This is Theorem 14.4.

(⇒) We apply Theorem 16.2 to prove that (b) ⇒ (c). Let us arbitrarily choose a base
R
point z0 in D. Since it is assumed that the integral C f (z) dz is independent of path, we
can define F (z) by integrating f (z) along any path from z0 to z. The condition in (16.2)
automatically holds due to the assumption of path independence. Thus, the function F (z)
serves as the primitive of f (z) in the region D.

We summarize what we know about path independence and existence of primitive.

Let f be a continuous complex-valued function defined in a region D. The followings are
equivalent:
H
(a) The contour integral C f (z), dz around any closed contour C in the region D is zero.
R
(b) The integral C f dz is path independent for any piece-wise smooth curve C in the
region D.

(c) Function f (z) has a primitive function F (z) in the domain D.

(d) the differential form f (z) dz is exact, i.e., there exists function F (z) defined on D
such that dF = f dz.

154
 We now apply Theorem 16.2 to the case where D is a convex set and f (z) is holomorphic.

Theorem 16.4 (Existence of primitive in a convex domain). If f (z) is complex

differentiable at every point inside a convex domain D, then f has a primitive in D.

Proof. We arbitrarily choose a point z0 in the domain D. Since D is convex, the straight
line from z0 to z lies entirely inside D for any point z ∈ D. We define a function F (z) as
Z
F (z) ≜ f (z) dz.
z0 →z

The condition in (16.2) holds by appealing to the Cauchy-Goursat theorem for triangles.
Consider two points z1 and z2 in the convex domain D. The three points z0 , z1 and z2
form a triangle that lies inside D. By applying the Cauchy-Goursat theorem for triangles
(Theorem 15.6), we know that the integral along the boundary of the triangle is zero
Z Z Z
f (z) dz + f (z) dz + f (z) dz = 0
z0 →z1 z1 →z2 z2 →z0

See Fig. 24. In terms of the function F (z), this can be expressed as
Z
F (z2 ) = F (z1 ) + f (z) dz.
z1 →z2

Hence, all the conditions in Theorem 16.2 are satisfied. We can now apply Theorem 16.2
to conclude that F (z) is a primitive of f (z) in the domain D.

Because the complex plane is convex, we can apply Theorem 16.4 to the whole complex
plane to get

Corollary 16.5. If f (z) is entire, then f (z) has a primitive defined on the whole
complex plane.

Example 16.1. Using this theorem, we now know that the following functions have primitives
in the complex plane:

sin(cos(ez )), cos32 (z) sin53 (z),

because they are entire function.

155
Figure 24: Proof of Theorem 16.4 about construction of primitive function for convex
region.

Example 16.2. Whether a function has primitive depends on the domain. The inverse
function 1/z has primitive in the open disc

D(3, 3) ≜ {z ∈ C : |z − 3| < 3},

because 1/z is defined and differentiable in this open disc. The primitive is given by a
branch of the complex log function. For example, we can take the principal branch Log(z).
However, if we consider the punctured disc

{z ∈ C : 0 < |z| < 3},

we cannot apply Theorem 16.4, because the punctured disc is not convex. In this case, it
is true that no primitive function exists for 1/z in the punctured disc D(0, 3). It is because
the integral of 1/z on a circle centered at 0 with radius r < 3 is 2πi, which is nonzero. By
Theorem 16.3, no such primitive function exists.
Example 16.3. Consider the inverse square function 1/z 2 in the punctured complex plane C\
{0}. The punctured complex plane is not convex, and hence we cannot apply Theorem 16.4.
Nevertheless, we do have a primitive function for 1/z 2 , namely −1/z, which is well defined
and differentiable in the punctured complex plane.
We have proved that we can always construct a primitive for a given holomorphic
function defined in a convex domain. Even though a lot of common shapes, such as circle,

156
semicircle, rectangle, and parallelogram, etc., are convex, we want to extend it further to
include nonconvex sets.
In fact, the proof of Theorem 16.4 does not require the full force of convexity. It only
requires that the domain D has a star shape, which means that we can find a point z0 in
D such that the line segment between z0 and any point in D lies entirely inside D (See
Fig. 25). Given a star-shaped domain S and the special point z0 , we can define a primitive
function F (z) as in the proof of Theorem 16.4.

Theorem 16.6 (Existence of primitive in a star-shaped domain). If f (z) is complex

differentiable at every point inside a domain D that has a star shape, then f has a
primitive in D.

Proof. The proof is exactly the same as in the proof of Theorem 16.4.

A more general form of the Cauchy theorem is stated in terms of simply connected
region

Theorem 16.7 (Cauchy theorem for simply connected regions). Suppose D is a

simply connected region. Then any holomorphic function f (z) defined in the domain
D has a primitive, and therefore
I
f (z) dz = 0
C

for any closed curve C in the domain D.

A typical proof relies on the fact that any simple closed curve in a simply connected
domain can be continuously shrunk to a point. However, providing a rigorous proof requires
a deeper understanding of topology and is beyond the scope of this notes.

16.3 Defining a branch of log function by integral

We provide an alternate definition of the complex log function by contour integral. This
definition is analogous to the definition of the real-valued log function through integration
in some calculus books, Z x
1
ln(x) = du,
1 u

157
 Figure 25: Star-shaped region

for x > 0.
Let D be the set defined as

D ≜ C \ {x + iy ∈ C : x ≤ 0, y = 0},

which consists of all complex numbers except for the negative real axis. This set represents
a star-shaped domain. We choose the point z0 = 1 as the base point. Every point z in
D can be reached through a straight line from 1 to z. Because 1/z is holomorphic in the
domain D, we can define Log(z) using the same argument as in the proof of Theorem 16.4:
Z
1
Log(z) ≜ dz.
1→z z

Here, the contour in the integral is the line segment from 1 to z. The function Log(z) in
this way serves as a primitive of 1/z in the domain D.
Now, consider a point z not on the negative real axis. We can express it as z = reiθ ,
where −π < θ < π. Since a primitive exists, the integral of 1/z is independent of path (as
long as the path does not cross the branch cut). We can choose a more convenient path for
calculation purposes. Consider a contour C from 1 to z that first goes horizontally from 1
to r, and then travels along an arc from r to reiθ .

158
 reiθ

θ
0 1 r

We can compute the integral of 1/z from 1 to z in two parts. First, consider the integral
from 1 to r: Z r Z r
1 1
dz = dx = ln r.
1 z 1 x
This integral simplifies to ln r.
The integral from r to reiθ along the circular arc is
Z reiθ Z θ
1
dz = r−1 e−iθ (ireiθ ) dθ
r z 0
= iθ.

The second integral simplifies to iθ.

Combining the results, we can express this branch of the complex log function as

Log(z) = ln(r) + iθ.

This matches the definition of principal log function in Definition 6.7.

159
17 Cauchy integral formula
Summary

ˆ Cauchy integral formula

ˆ Deformation theorem

ˆ Liouville theorem

17.1 Cauchy integral formula

The Cauchy integral formula is a landmark in complex analysis.

Theorem 17.1 (Cauchy integral formula). Consider a domain D and a circular

path C with counter-clockwise orientation inside D. If function f (z) is holomorphic
in D, then for any z in the interior of C,
Z
1 f (w)
f (z) = dw. (17.1)
2πi C w − z

The Cauchy integral formula provides an integral representation of a holomorphic func-

tion. The integral on the right-hand side is taken over a circle with an arbitrary radius,
provided that the function f (z) is holomorphic within the circle. However, the integrand
fails to be holomorphic at only one point, namely, at the point z.
A remarkable feature about the Cauchy integral formula is that, in the calculation of
the contour integral, we only require the value of the function on the circle C. The fraction
1/(w − z) acts as a weighting factor. By using different weighting factors, we can determine
the value of function f at any point z inside the circle. Hence, the value of f within the
circle can be completely determined by its values on the boundary.
The proof of Theorem 17.1 relies on a theorem commonly known as the “deformation
theorem.” The following statement represents a particular case of the general deformation
theorem, which is adequate for deriving the Cauchy integral formula for circle.

160
 C2

C1

Figure 26: The two circular paths in Theorem 17.2.

Theorem 17.2 (Deformation theorem for circles). Let C1 and C2 be two circular
contours with the same orientation, such that C2 is contained inside C1 (See Fig. 26).
Suppose f (z) is holomorphic at all points that are inside C1 and on the curve C1 ,
with the possible exception at the center of circle C2 , then
Z Z
f (z) dz = f (z) dz.
C1 C2

Proof. Since f (z) is holomorphic at every point on the outer circle C1 , there exists a disc
slightly larger than C1 in which f (z) is holomorphic, except possibly at the center of C2 .
We divide the region between the circles into four parts by drawing a horizontal and a
vertical line passing through the centers of the inner circle C2 (See Fig. 27). Each closed
path is contained within a convex set where the function f (z) is holomorphic. Consequently,
the integral of f (z) over each closed path is equal to zero.
On the other hand, the sum of the integrals over the four closed paths is equivalent to
the difference between the integrals over C1 and C2 . This demonstrates that the integrals
of f (z) over C1 and C2 are equal.

161
 Figure 27: Proof of Theorem 17.2.

Proof of Cauchy integral formula. Consider an arbitrary point z within the interior of the
circle C. Draw another circle, denoted as C ′ , centered at z with a radius ρ. Choose the
radius ρ such that C ′ lies entirely inside C (see Fig. 28). The circular contour C ′ is traversed
in a counterclockwise direction.
The fraction f (w)/(w − z) is holomorphic at every point on the curves C and C ′ as well
as the points enclosed between them. We can apply Theorem 17.2 to obtain the following
equality, Z Z
f (w) f (w)
dw = dw.
C w−z C′ w − z
Hence, it is sufficient to prove
Z
f (w)
dw = 2πif (z). (17.2)
C′ w−z

We start by writing
f (w) = f (z) + [f (w) − f (z)]

162
 C
C′
z

Figure 28: Proof of Cauchy integral formula.

and then decompose the integral in (17.2) into two parts:

f (w) − f (z)
Z Z Z
f (w) f (z)
dw = dw + dw. (17.3)
C′ w − z C′ w − z C′ w−z

To evaluate the first integral on the right-hand side of (17.3), we can employ the same
calculations as in Example 14.2. We parameterize C ′ as z + ρeiθ , where θ ranges from 0 to
2π. Then we have
Z 2π
iρeiθ
Z Z
f (z) 1
dw = f (z) dw = f (z) dθ = 2πif (z).
C′ w − z C′ w − z 0 ρeiθ

To simplify notation, we introduce g(w) ≜ f (w) − f (z). On the circle C ′ , the modulus
|g(w)| can be expressed as
|g(z + ρeiθ )|

for θ in the interval [0, 2π]. Since [0, 2π] is a closed and bounded interval, and |g(z + ρeiθ )|
is a continuous function of θ, we can conclude that a maximum exists and is finite. We
define
Mρ ≜ sup |g(z + ρeiθ )| = max′ |g(w)|.
θ∈[0,2π] w∈C

By ML inequality (Theorem 15.4), we can upper bound the modulus of the integral
Z
g(w)
dw.
C′ w − z

163
by Z
g(w) Mρ
dw ≤ 2πρ = 2πMρ .
C′ w−z ρ
Since the function g is continuous, Mρ approaches 0 as ρ approaches 0.The integral in (17.1)
is thus equal to 0. This proves that the right-hand side of (17.3) is equal to 2πif (z).
z
Example 17.1. Calculate C z 2e+1 dz when C is the contour (i) |z − i| = 1, (ii) |z + i| = 1,
R

(iii) |z − 2| = 1, (iv) |z| = 4, all with positive orientation.

The denominator can be factorized as

z 2 + 1 = (z + i)(z − i).

The function ez /(z 2 + 1) is not holomorophic at the points i and −i.

(i) Let the circle centered at i with radius 1 and positive orientation be denoted by C1 .
It encloses the point z = i but not z = −i.
ez
ez  ez 
Z Z
z+i
dz = dz = 2πi = πei .
C1 z2 + 1 C1 z−i z+i z=i

(ii) Let C2 be the circular contour centered at −i with radius 1 and positive orientation.
It encloses the point z = −i but not z = i.
ez
ez  ez 
Z Z
z−i
dz = dz = 2πi = −πe−i .
C2 z2 + 1 C2 z+i z−i z=−i

(iii) The circle centered at z = 2 with radius 1 does not contain the singular points ±i,
and hence is holomorphic inside and on the boundary of the circle. By Cauchy theorem,
the integral is equal to 0.
(iv) If we integrate along the circle centered at the origin with radius 4 counter-clockwise,
the answer is the sum of the integrals in parts (i) and (ii),

πei − πe−i = π(ei − e−i ) = 2πi sin(1).

17.2 Liouville theorem

Using the Cauchy estimate, we can establish the Liouville theorem. Recall that an entire
function is a complex function that is holomorphic at every point in the complex plane. We
can motivate the Liouville theorem by first looking at some examples.

164
 (1) The functions ez , sin(z) and cos(z) are entire but not bounded. It is important
to note that the function sin(z) and cos(z) are not bounded, because | sin(z)| approaches
infinity when we move along the vertical line x + iy, with a constant x and y → ∞.
(2) The function e−|z| defined for all z ∈ C is bounded, but is not entire.
(3) The inverse function 1/z is unbounded, and is not defined at the origin.

Theorem 17.3 (Liouville theorem). An entire and bounded function must be a con-
stant function.

Proof. Suppose f (z) is entire and there exists a real number M such that |f (z)| ≤ M for
all z ∈ C. We want to show that f (z1 ) = f (z2 ) for any complex numbers z1 and z2 .
The idea of proof is to draw a large circle centered at the origin. The radius R is selected
such that
R ≥ 2 max(|z1 |, |z2 |).

(The exact value of the factor 2 is not very important. We can replace it by any real number
larger than 1.)
For any complex number w on the circle with radius R and centered at the origin, we
have
R = |w| = |w − zk + zk | ≤ |w − zk | + |zk | ≤ |w − zk | + R/2.

for k = 1, 2. Hence,
R R
|w − z1 | ≥ and |w − z2 | ≥ (17.4)
2 2
for all w on the circle |w| = R. Apply Cauchy integral formula to express f (z1 ) and f (z2 )
as integrals over the circle, and write the difference as
Z
1 f (w) f (w)
f (z1 ) − f (z2 ) = − dw.
2πi |w|=R w − z1 w − z2

Simplify the integrand on the right to

z 1 − z2
Z
f (w)
f (z1 ) − f (z2 ) = dw.
2πi |w|=R (w − z1 )(w − z2 )

By the ML inequality, and the two assumptions in (17.4), we get

|z1 − z2 | M 1
|f (z1 ) − f (z2 )| ≤ · 2πR = O( ).
2π (R/2)2 R

165
Because f (z) is entire, we can enlarge the circle |w| = R by increasing the radius R, and the
above argument is still valid. Because R can be arbitrarily large, we see that |f (z1 ) − f (z2 )|
must be equal to 0.

17.3 Fundamental theorem of algebra

We have a quick proof of the fundamental theorem of algebra using Liouville theorem.

Theorem 17.4 (Fundamental theorem of algebra). A polynomial with complex co-

efficient of degree at least 1 has a root in C.

Proof. Suppose p(z) is a polynomial

c0 + c1 z + c2 z 2 + · · · cd z d

with degree d ≥ 1. Without loss of generality, we can assume that cd = 1.

We will prove this by contradiction. Assume that p(z) ̸= 0 for all z ∈ C. Then 1/p(z)
is a well-defined and holomorphic for all z ∈ C, i.e., 1/p(z) is an entire function.
Consider the ratio
p(z) c0 c1
d
= d + d−1 + · · · + 1.
z z z
As |z| → ∞, this ratio approaches 1. Hence, there exists a sufficiently large real number R
such that
p(z) 1
d
>
z 2
for all z with modulus larger than R. In other words, we have
1 2 2
< d < d
p(z) |z | R

for all |z| > R.

On the other hand, for |z| ≤ R, the modulus |1/p(z)| is bounded, because 1/|p(z)| is a
continuous function of z and |z| ≤ R is a compact set.
As a result, |1/p(z)| is bounded by a constant for all z ∈ C. By Liouville theorem, 1/p(z)
is a constant, contradicting that the assumption that the degree of p(z) is at least 1.

166
18 Taylor series expansion
Summary

ˆ Power series, radius of convergence, and term-wise differentiation

ˆ Power series expansion of holomorphic function

ˆ Cauchy integral formula for higher-order derivatives

ˆ Uniqueness of power series expansion

18.1 Convergence of power series

We study power series in more details in the remainder of this lecture. For ease of notation,
we assume that the center of the power series is the origin. All results below hold in the
general case when the center is not the origin.

Proposition 18.1. Suppose ∞ k

P
k=0 ak z converges at z0 ̸= 0, then it converges abso-
lutely for all z with |z| < |z0 |.

Proof. By the n-th term test (Prop. 5.8) we know that |ak z0k | → 0 as k → ∞. Fix ϵ > 0.
there is a sufficiently large N such that

|ak z0k | < ϵ, ∀k ≥ N.

Hence, for all k ≥ 0, we have

|ak z0k | ≤ max(ϵ, |a0 |, |a1 z0 |, |a2 z02 |, . . . , |aN −1 z0N −1 |) ≜ C.

Let z be a complex number with |z| < |z0 |. Let r be the ratio |z|/|z0 |. For each k, re-write
|ak z k | as
|z|k
|ak z k | = |ak z0k | ≤ Crk ,
|z0 |k
Because r < 1 and k Crk is a convergent geometric series, we can apply comparison test
P

and conclude that k ak z k is convergent absolutely for |z| < |z0 | (See Prop. 5.9).
P

167
 By taking the contrapositive, we can re-write Prop. 18.1 as

ak z k diverges at z1 , then ak z k diverges whenever |z| >

P P
Proposition 18.2. If k k
|z1 |.

Proof. We prove by contradiction. Suppose k ak z k converges for some point z = z2 with

P

modulus |z2 | strictly larger than |z1 |. By the previous proposition, the power series must
converge (absolutely) at z = z1 . This contradicts the assumption that k ak z k diverges
P

at z1 .

In view of the previous two proposition, we can make the following definition.
P∞ k,
Definition 18.3. Given a power series k=0 ak z let
X
ak z k converges .

R ≜ sup |z| : (18.1)
k

The value of R in (18.1) is called the radius of convergence of ∞ k

P
k=0 ak z . The region
|z| < R is called the region of convergence. If k ak z k converges for all z, then we
P

write R = ∞, and say that the radius of convergence is infinite. If k ak z k converges

P

only at z = 0, then R = 0.

Example 18.1. The power series that define the complex functions exp(z), sin(z), cos(z),
sinh(z) and cosh(z) have radius of convergence R = ∞.
Example 18.2. If we expand the function 1/(1 + z 2 ) as a power series, the resulting power
series is
1 − z2 + z4 − z6 + · · ·

The radius of convergence is 1.

An explicit expression for calculating the radius of convergence is given by the Hadamard
formula.

168
 Theorem 18.4 (Hadamard formula for radius of convergence). Given a complex
power series ∞ k
P
k=0 ak z , the radius of convergence can be computed by

1
R= .
lim supk |ak |1/k

Proof. Suppose z is a complex number with |z| < R. Pick a real number R1 between |z|
and R, i.e., |z| < R1 < R. By the definition of R in the theorem,
1
> lim sup |ak |1/k .
R1
Using the property of limsup, there exists a sufficiently large N such that for all k ≥ N ,

|ak |1/k < 1/R1 ⇒ |ak | < 1/R1k .

This yields
|z|k
|ak z k | < .
R1k
Since |z|k /R1k < 1, the geometric series k k
P
k |z| /R1 is convergent. By comparison test,
k
P
k ak z is convergent.
Now suppose that |z| > R. Pick a real number R2 such that |z| > R2 > R. Taking
reciprocal,
1
< lim sup |ak |1/k .
R2
By the defining property of limsup,
1
< |ak |1/k for infinitely many k.
R2
k
Hence |ak z k | > Rz2 > 1 for infinitely many k. By the nth-term test (Prop. 5.8)), the
power series diverges whenever |z| > R.

We recall that by uniform convergence, we mean that for all ϵ > 0, there exists an
integer N (ϵ) that depends only on ϵ but not on the value of z, such that
n
X ∞
X
k
ak z − ak z k < ϵ
k=0 k=0

169
for all n ≥ N (ϵ) and for all z ∈ S. The number N (ϵ) should be a function of ϵ only inde-
pendent of z. The following theorem gives a sufficient condition for uniform convergence.
The proof idea is basically Weierstrass’s M -test.

Theorem 18.5. Suppose a complex power series ∞ k

P
k=0 ak z converges absolutely at
P∞
z = z0 , then k=0 ak z k converges uniformly in the closed disc S = {z : |z| ≤ |z0 |}.

ak z k converges absolutely for all z ∈ S. It is because

P
Proof. We first note that the series k

∞
X ∞
X
|ak ||z|k ≤ |ak ||z0 |k ,
k=0 k=0

and the series on the right converges by assumption.

For k ≥ 0, let Mk ≜ |ak ||z0 |k . By assumption, ∞
P
k=0 Mk converges. Given any ϵ > 0,
there exists an integer N (ϵ) such that
∞
X
Mk ≤ ϵ, for all n ≥ N (ϵ).
k=n+1

So, for this choice of N (ϵ), we have

n
X ∞
X ∞
X ∞
X ∞
X
k k k k
ak z − ak z = ak z ≤ |ak ||z| ≤ Mk ≤ ϵ
k=0 k=0 k=n+1 k=n+1 k=n+1
P∞
ak z k is uniformly convergent to k
P
for all z ∈ S. This proves that k k=0 ak z for z ∈ S.

18.2 Differentiation of power series

P∞ k
Using the property of uniform convergence, we can prove that the function k=0 ak z is
complex differentiable within the region of convergence.

170
 Theorem 18.6 (Term-wise differentiation of power series). Suppose power series
P∞ k
k=0 ak z has radius of convergence R > 0.

1. The series ∞ k−1 obtained by term-wise differentiation has the same

P
k=1 kak z
radius of convergence R.
P∞ k
2. The function defined by k=0 ak z is complex differentiable at every z with
|z| < R, and the complex derivative is given by ∞ k−1 .
P
k=1 kak z

Proof. We first use the Hadamard formula in Theorem 18.4 to show that k kak z k−1 has
P

radius of convergence equal to R. We note that the radius of convergence of k kak z k−1
P

is the same as the radius of convergence of k kak z k , and the radius of convergence of the
P

latter series can be computed by

1 1 1
1/k
= √ = = R.
lim supk |kak | k
lim supk k|ak |1/k lim supk |ak |1/k
√
In the second equality, we have used the fact that limk→∞ k k = 1.
To prove the second part, we pick a complex number z0 inside the region of convergence,
i.e., |z0 | < R. Let δ be a real number such that |z0 | + δ < R. The choice of δ is made to
ensure that the open disc {z : |z − z0 | < δ} lies inside the region of convergence.
We set up some notation. For n ≥ 0, let fn (z) denote the n-th partial sum
n
X
fn (z) ≜ ak z k ,
k=0

and let f (z) denote the limit

∞
X
f (z) ≜ ak z k .
k=0

By Theorem 18.5, fn (z) converges to f (z) uniformly in the closed disk S = {z : |z| ≤
|z0 | + δ}.
By part 1 and Theorem 18.5, the series ∞ k−1 converges absolutely and uni-
P
k=1 kak z
formly in the closed disk S.

171
 For each n ≥ 0, define
n
fn (z0 + h) − fn (z0 ) X (z0 + h)k − z0k
φn (h; z0 ) ≜ = ak (18.2)
h h
k=0

for nonzero h with |h| ≤ δ. When n → ∞, φn (h; z0 ) converges to

f (z0 + h) − f (z0 )
.
h
The convergence is uniform for |h| ≤ δ. Indeed, we can use the identity

αk − β k = (α − β)(αk−1 + αk−2 β + αk−3 β 2 + · · · + αβ k−2 + β k−1 ),

to upper bound the k-th term in the summation in (18.2),

(z0 + h)k − z0k |(z0 + h)k − z0k |

ak = |ak |
h |h|
X k−1 
= |ak | |z0 + h|j |z0 |k−1−j
j=0

≤ |ak |k(|z0 | + δ)k−1 .

Because the series ∞ k−1 is convergent and does not depend on the value
P
k=1 |ak |k(|z0 | + δ)
of h, we can see that the convergence φ(h; z0 ) is uniform over h.
Consider the double limit  
lim lim φn (h; z0 ) .
n→∞ h→0
Pn k−1
For each n, fn (z) is a polynomial. Hence limh→0 φn (h; z0 ) = k=1 kak z0 . This gives
  n
X ∞
X
lim lim φn (h; z0 ) = lim kak z0k−1 = kak z0k−1 . (18.3)
n→∞ h→0 n→∞
k=1 k=1

(Convergence of the last summation is justified by part 1.)

On the other hand, since φn converges uniformly for |h| ≤ δ, we can exchange the order
of the two limits to get
  f (z0 + h) − f (z0 )
lim lim φn (h; z0 ) = lim . (18.4)
h→0 n→∞ h→0 h

172
Since (18.3) and (18.4) are equal, the limit in (18.4) exists and is equal to the power series
in (18.3). We thus obtain
∞
X
f ′ (z0 ) = kak z0k−1 .
k=1

This proves that a power series is complex differentiable, and we can differentiate it term-
wise.

By repeatedly applying Theorem 18.6, we can differentiate a power series arbitrarily

many times.
P∞ k
Theorem 18.7. Suppose the radius of convergence of power series k=0 ak z is
positive. For any positive integer j, the function f (z) defined by a power series
P∞ k
k=0 ak z can be differentiated j-th times at any point in the interior of the region
of convergence, and the j-th derivative can be computed by the power series
∞
X
(j)
f (z) = k(k − 1)(k − 2) · · · (k − j + 1)ak z k−j .
k=j

In real analysis, we have example of infinitely differentiable functions that does not have
power series expansion. The following real continuous function

0 if x ≤ 0
f (x) ≜ 4
e−1/x if x > 0

is infinitely smooth. However, the derivatives of all orders at x = 0 are all equal to zero.
The power series expansion of f (x) at x = 0 is the constant function 0, which is not the
same as the original function f (x). This function fails to have power series expansion at
x = 0.
The main result in the next section says that we do not have this kind of example in
the complex case.

18.3 Existence of power series expansion for holomorphic functions

We have shown in the previous section that a power series is holomorphic. It turns out
that the converse is also true, and it is the next objective.

173
 Definition 18.8. A complex function is called analytic at a point z0 if we can draw
a small circle centered at z0 such that the function in this circle is the same as a
power series centered at z0 . We say that a function is analytic if it is analytic at any
point in the domain of definition.

The next theorem guarantees that if a complex function f (z) is holomorphic at z0 , then
f (z) is analytic at z0 .

Theorem 18.9 (Taylor expansion). Suppose f (z) is a complex-valued function that

is complex differentiable in a domain that contains the closed disc |z − z0 | ≤ r. Then
f (z) has Taylor series expansion
∞
X
f (z) = an (z − z0 )n
n=0

in the open disc |z − z0 | < r. Moreover, for n = 0, 1, 2, 3, . . ., the coefficient an is

given by I
1 f (w)
an = dw. (18.5)
2πi |z−z0 |=r (w − z0 )n+1

Recall that when we say that a function f (z) is holomorphic at a point z0 , it means
that f (z) is complex differentiable in an open disc |z − z0 | < ρ, for some positive radius ρ.
Thus, the above theorem applies to every point at which the function is holomorphic.
A heuristic argument is as follows. Consider a contour C defined as |z − z0 | = r with a
counter-clockwise orientation. By Cauchy integral formula (Theorem 17.1), for any complex
number z inside the contour C, we can express f (z) as
I
1 f (w)
f (z) = dw.
2πi C w − z
The proof idea of this theorem is to expand 1/(w − z) using a geometric series,
1 1
=
w−z w − z0 − (z − z0 )
1  1 
= z−z0
w − z0 1 − w−z
0

1 z − z0 (z − z0 )2 (z − z0 )3
= + + + + ··· (18.6)
w − z0 (w − z0 )2 (w − z0 )3 (w − z0 )4

174
 w

z0 C
z

Figure 29: Proof of Taylor expansion of holomorphic function.

This geometric series converges because |z − z0 | < |w − z0 |. If we can justify the exchange
of infinite summation and contour integral, then we get
∞
(z − z0 )n
I X
1
f (z) = f (w) dw
2πi C (w − z0 )n+1
n=0
∞  I
X 1 f (w) dw 
= (z − z0 )n .
2πi C (w − z0 )n+1
n=0

This is a Taylor series centered at z0 .

Proof. To make the above argument rigorous, we can utilize the property of uniform con-
vergence. Let ρ be a real number in the range 0 < ρ < r. Consider a complex number z
that lies inside the circle |z − z0 | < ρ (See Fig. 29).
For each positive integer n, consider the partial sum
n
X
ak (z − z0 )k ,
k=0

where ak is defined in (18.5). We want to show that it converges to f (z) as n → ∞, for

|z − z0 | < ρ.
Because |(z − z0 )/(w − z0 )| < 1, we can use geometric series to simply the infinite sum
∞
X (z − z0 )k (z − z0 )n+1 1 1 (z − z0 )n+1
= z−z0 = .
(w − z0 )k+1 (w − z0 )n+2 1 − w−z (w − z) (w − z0 )n+1
k=n+1 0

We can thus write 1/(w − z) as

1 1 z − z0 (z − z0 )n 1 (z − z0 )n+1
= + + · · · + + .
w−z w − z0 (w − z0 )2 (w − z0 )n+1 (w − z) (w − z0 )n+1

175
Multiply each term by f (w) and integrate over the curve C. (Since this is a finite sum,
there is no problem in exchanging finite summation and integration.)
n 
f (w) (z − z0 )n+1
Z Z Z
1 f (w) X 1 f (w) dw  k 1
dw = (z − z 0 ) + dw.
2πi C w − z 2πi C (w − z0 )k+1 2πi C (w − z) (w − z0 )n+1
k=0

Using the definition of coefficients an in the theorem, we can write the right-hand side of
the above equation as
n
f (w) (z − z0 )n+1
Z
X
k 1
ak (z − z0 ) + dw.
2πi C (w − z) (w − z0 )n+1
k=0

This holds for any n ≥ 0. The summation is a partial sum of the desired Taylor series
expansion. The integral on the right is the corresponding error term.
To bound the error term, we observe that

|z − z0 | < ρ
|w − z| ≥ r − |z − z0 |
|w − z0 | = r.

Because |f (w)| is a continuous function, the value of |f (w)| on the circle C is bounded by
some constant M . We now apply the ML inequality (Theorem 15.4),
f (w) (z − z0 )n+1 ρn+1
Z
1 1 M Mρ  ρ n
dw ≤ · 2πr = .
2πi C (w − z) (w − z0 )n+1 2π r − |z − z0 | rn+1 r − |z − z0 | r
Take n → ∞, the modulus of the remainder term approach 0. (This step depends on the
assumption that ρ < r). Consequently, the Taylor series is convergent at z, and converges
to the intended value f (z).

Because a convergent power series is differentiable, we have proved a key feature of

holomorphic functions

Theorem 18.10. A complex function is holomorphic in a domain if and only if it

is analytic. Moreover, a holomorphic is infinitely differentiable, i.e., the n-th order
derivative exists for all positive integer n.

Because of this connection, in the remainder of this lecture notes we will use the two
terms “holomorphic” and “analytic” interchangeably.

176
 We summarize what we know about holomorphic function below. Let f (z) denote a
continuous complex-valued function in a domain D. Then the followings are equivalent:

(a) f (z) is holomorphic in D, i.e., at each point z0 ∈ D, the limit

f (z0 + h) − f (z0 )
lim
h→0 h
exists.

(b) f (z) is real-differentiable as a multi-variable function, and satisfies the Cauchy-Riemann

equations

(c) f (z) is analytic in D, i.e., at each point z0 ∈ D, f (z) can be expanded locally as a
power series centered at z0 .

Theorem 18.5 also sheds light on the radius of convergence. The radius of convergence of
a function f (z) at the center point z0 is the largest radius r such that there is no singularity
within the open disc |z − z0 | < r.
1
Example 18.3. Consider the function f (z) = z(z−2) . It is complex differentiable everywhere
except two points, namely 0 and 2. Suppose we want to find the Taylor series function at
the point i. We can draw a circle with radius 1, centered at i, so that f (z) is holomorphic
inside the circle. This is the largest circle that does not contain any of the two problematic
points 0 and 2. By Theorem 18.5, the power series expansion of f (z) with center i has
radius convergence equal to 1.
Now, suppose we change the center to 1 + i. The distance from 1 + i to the two points
√
0 and 2 are both 2. Hence, the radius of convergence of the power series with center 1 + i
√
is 2.

18.4 Uniqueness of Taylor series

The identity theorem is another surprising result in complex analysis. It says that if a
holomorphic function is identically zero in a small open disk, then the function must be
identically zero throughout the domain. This is related to the uniqueness of the coefficients
of Taylor series. We first recall the notion of accumulation point.

177
 Definition 18.11. An accumulation point (a.k.a. cluster point or limit point), of a
set S of points is a point p such that every neighborhood of p contains at least one
point of S not equal to p.

It is known that in a point p in a metric space is an accumulation point of S if and only

if every neighborhood of p contains infinitely many points of S.

Theorem 18.12 (Uniqueness of coefficients in Taylor series). Suppose

∞
X ∞
X
n
f (z) = an (z − z0 ) and g(z) = bn (z − z0 )n
n=0 n=0

are functions defined by power series, both centered at z0 , and both of them have
positive radii of convergence. If we can find a set S of points in the common region
of convergence, such that f (w) = g(w) for all w ∈ S and z0 is an accumulation point
of S, then we must have ak = bk for all k.

Proof. From the hypothesis in the theorem, we can find a sequence (zj )∞ j=1 in S that
converges to z0 . For example, we can let zj to be a point in S with 0 < |zj − z0 | < 1/j. By
construction, this sequence converges to z0 , and zj ̸= z0 for all j.
Let h(z) be the difference of the two functions f (z) and g(z). By assumption we have
h(zj ) = 0 for all j = 1, 2, 3, . . .. Since h(z) is holomorphic at z0 , we can write h(z) has a
power series centered at z0 ,
X∞
cn (z − z0 )n ,
n=0

where cn = an − bn is the coefficient of the n-th term.

We claim that cn = 0 for all n.
Suppose the claim is false and suppose there is a nonzero coefficient. Let m be the
smallest integer such that cm ̸= 0. If m = 0, i.e., c0 ̸= 0, then by the continuity of
power series, the value of h(z) is nonzero in a small neighborhood of z0 , contradicting the
hypothesis that there exists a sequence of points zj converging to z0 and h(zj ) = 0 for all j.
Suppose m ≥ 1. We can write h(z) as

h(z) = (z − z0 )m cm + cm+1 (z − z0 ) + cm+2 (z − z0 )2 + · · · .

 

178
Because cm ̸= 0 by assumption, the power series inside the square bracket is nonzero for z
in a sufficiently small neighborhood of z0 . The first factor (z − z0 )m is nonzero whenever
z ̸= z0 . As a result, we see that h(z) is nonzero in a punctured disc centered at z0 . This
yields the same contradiction as in the previous paragraph.
Therefore the claim is true, and we must have cn = 0 for all n. This proves that an = bn
for all n.

18.5 Application to generating functions

The method of generating function is a powerful way to solve difference equation and
linear recurrence relation. This is a topic in combinatorics, computer science and discrete
mathematics. In general, a linear recurrence relation of order k is a recursive definition in
the form
an = c1 an−1 + c2 an−2 + · · · + ck an−k ,

where c1 , . . . , ck are constants. If k initial values a0 , a1 , . . . , ak−1 are given, we can use the
above equation to define an infinite sequence. The question is to study this sequence, e.g.
by giving some asymptotic estimates of the numbers when n is large. In signal processing,
it is the same as z-transform.
The method of generating function is to consider a power series, whose coefficients are
the numbers in the sequence (ak )∞ k=1 ,
∞
X
f (z) = ak z k .
k=0

Provided that the sequence does not increase too fast (say no faster than exponential
growth), then we can study the associated complex-valued function f (z) and try to obtain
information about the sequence (ak )∞k=0 from the function.
In the following we consider the standard example: Fibonacci sequence. This example
contains all the ingredient in the method of generating function. The Fibonacci numbers
are recursively defined by F0 = F1 = 1, and

Fk = Fk−1 + Fk−2

for k ≥ 2. The first few numbers in this sequence are

1, 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, 144, . . .

179
 Let
f (z) = F0 + F1 z + F2 z 2 + F3 z 3 + F4 z 4 + F5 z 5 + · · ·
= 1 + z + 2z 2 + 3z 3 + 5z 4 + 8z 5 + 13z 6 + · · ·
The k-th coefficient (for k = 0, 1, 2, . . .) is the k-th Fibonacci number. The power series has
positive radius of convergence, because the k-th Fibonacci number is upper bounded by
2k . (This can be proved by mathematical induction) Hence, f (z) has a non-trivial region
of convergence with radius at least 1/2.
By comparing f (z), zf (z) and z 2 f (z), we can see that the function f (z) satisfies the
following equation
f (z) − zf (z) − z 2 f (z) = 1.
In the region of convergence, f (z) is equal to a rational function in z:
1
f (z) = .
1 − z − z2
We will write the function f (z) in a more convenient form,
1
f (z) =
(1 − αz)(1 − βz)
where α and β are the reciprocal root of 1 − z − z 2 . To this end, we consider the reciprocal
polynomial, which is commonly known as the characteristic polynomial:
P (u) = u2 − u − 1,
whose coefficients are the same as in f (z) but in reverse order. In this example, one of the
roots is the famous Golden ratio. We denote the two roots by
√ √
1+ 5 1− 5
φ1 = , φ2 = .
2 2
Hence, we have
1
f (z) = .
(1 − φ1 z)(1 − φ2 z)
We can apply partial fraction expansion, and expand the two resulting fraction by
geometric series,
√ √
φ1 / 5 φ2 / 5
f (z) = −
1 − φ1 z 1 − φ2 z
∞  k+1
X φ1 φk+1 
= √ − √ 2
zk .
k=0
5 5

180
By the uniqueness of power series expansion, the k-th Fibonacci number is equal to

φk+1
1 φk+1
2
Fk = √ − √ .
5 5
Because the Fibonacci numbers are integers, and the absolute value of φ2 is less than 1, we
can further simplify the expression to
 φk+1 
1
Fk = round √
5
where round() is the function that rounds a real number to the nearest integer.
We thus see that Fibonacci grows exponentially fast, and asymptotically the ratio be-
tween two consecutive numbers is the Golden ratio.
In general, the asymptotic growth rate of a sequence defined by linear recurrence relation
is controlled by the largest root of the characteristic function.

181
19 Identity theorem
Summary

ˆ Cauchy integral formula for higher-order derivatives

ˆ Cauchy estimate

ˆ Identity theorem

The fact that a holomorphic function has power series expansion has many surprising
consequences. The scenario is very different from real-differentiable function.

19.1 Cauchy integral formula for higher-order derivatives

At this point, we know that the derivative of a holomorphic function of any order exists.
There is a formula for the higher-order derivatives that is analogous to the original Cauchy
integral formula.

Theorem 19.1. If a function f is holomorphic in a domain D, then f is infinitely

differentiable. The n-th derivative at z = z0 can be evaluated by
I
(n) n! f (w)
f (z0 ) = dw (19.1)
2πi C (w − z0 )n+1

where C is a circle centered at z0 such that f (z) is holomorphic at every point inside
and on the circle C. In particular, we have
I
′ 1 f (w)
f (z0 ) = dw.
2πi C (w − z0 )2

Remark. In some books, the formula in (19.1) is proved directly from Cauchy theorem,
without mentioning power series. In this notes, we derive it as an easy corollary of Taylor
series expansion. The difficult part has already been done in the proof of Theorem 18.9.

Proof. From Theorem 18.9 we know that the Taylor series

f (z) = a0 + a1 (z − z0 ) + a2 (z − z0 )2 + a3 (z − z0 )3 + · · · (19.2)

182
converges in a neighborhood of z = z0 , with coefficient an equal to
I
1 f (w)
an = dw.
2πi C (w − z0 )n+1

If we substituting z = z0 into (19.2), we get

I
1 f (w)
f (z0 ) = a0 = dw,
2πi C w − z0
which is the same as Cauchy’s integral formula in Theorem 17.1. Since Taylor series can
be differentiated term-wise, we can differentiate both sides of (19.2) and substitute z = z0
to get Z
′ 1 f (w)
f (z0 ) = a1 = dw.
2πi C (w − z0 )2
If we differentiate n times and substitute z = z0 , we obtain (19.1).

Based on (19.1) one can estimate the modulus of the derivative of a holomorphic func-
tion. This is commonly known as the Cauchy estimate for derivatives

Corollary 19.2 (Cauchy estimate). If |f (z)| ≤ M for z on the circle

{z : |z − z0 | = r}

and f is holomorphic in a neighborhood of the disc |z − z0 | ≤ r, then

n!
|f (n) (z0 )| ≤ M
rn
for n = 0, 1, 2, 3, . . .

Proof. Take absolute value on both sides of (19.1) and apply ML inequality (Theorem 15.4)
Z
(n) n! f (w) n! M n!M
|f (z0 )| = n+1
dw ≤ n+1
2πr = n .
2πi C (w − z0 ) 2π r r

An application of the formula in (19.1) is to evaluate integral in the form

I
f (z)
dz
C (z − z0 )m

183
where m is an integer larger than or equal to 2, and z0 is a complex number contained in
the interior of contour C. Here, it is assumed that C has simple shape, e.g. a circle or a
square, such that f (z) is holomorphic in the region bounded by C and the Cauchy integral
formula applies.
Example 19.1. Suppose we want to integrate eiωz /(z − z0 )2 , for some real constant ω and
complex constant z0 ,
For simplicity, we take z0 = 1, and C is a circle that contains 1 in the interior. We can
evaluate the integral
eiωz
I
2
dz
C (z − 1)

by applying the equation in (19.1),

eiωz
I
d
dz = 2πi eiωz = 2πi(iω)eiωz = −2πωeiω .
C (z − 1)2 dz z=1 z=1

Example 19.2. We apply the Cauchy integral formula for higher derivatives to evaluate a
trigonometric integral: Z 2π
1
dθ (19.3)
0 (a + b cos θ)2
where a and b are positive constants satisfying a > b. The assumption a > b ensures that
the integrand never vanishes for any θ. Integrals of this type can be transformed to a
contour integral on the unit circle by making a substitution z = eiθ , and use the formula

eiθ + e−iθ
cos θ = .
2
We transform the trigonometric integral in (19.3) to
I
1 1
−1 2
· dz
C (a + b(z + z )/2) iz

where C is the unit circle with counter-clockwise orientation. We single out the integrand
and simplify it as follows:
1 1 4z
−1 2
· =
(a + b(z + z )/2) iz i(bz + 2az + b)2
2

4z
= 2 2 .
ib (z + 2(a/b)z + 1)2

184
Denote the root of the quadratic polynomial z 2 + 2(a/b)z + 1 in the demoninator as
√ √
−a + a2 − b2 −a − a2 − b2
α≜ , β≜ .
b b
We need to determine whether α and β are inside the unit circle. We first use the fact
that the product αβ = 1 to see that exactly one of the roots α and β is inside the unit
circle. Since β is negative and less than −1, and hence is outside the unit circle. we know
that the first root α must be inside the unit circle. We express the contour integral in the
form of the Cauchy integral formula,
I I
1 1 4 z
−1 2
· dz = 2 dz.
C (a + b(z + z )/2) iz ib C (z − α) (z − β)2
2

To apply Cauchy integral formula, we consider the function

z
f (z) = .
(z − β)2
We note that the function f (z) is holomorphic inside the unit circle C. Compute the
derivative of f (z)

1 · (z − β)2 − z(2(z − β)) β+z

f ′ (z) = =− ,
(z − β)4 (z − β)3
and evaluate it at z = α. This gives

β+α −2a/b ab2

f ′ (α) = − = − = .
(α − β)3 8(a2 − b2 )3/2 /b3 4(a2 − b2 )3/2
The final answer is
Z 2π
1 4
2
dθ = 2 2πif ′ (α)
0 (a + b cos θ) ib
4 ab2
= 2 2πi
ib 4(a2 − b2 )3/2
a
= 2π 2 .
(a − b2 )3/2
We can double check the answer by noting that the answer should be a real and positive
number. In the degenerate case of b = 0, the answer should be equal to 2π/a2 . Another
sanity check is: if increase both a and b by a factor of k, then the integral should decrease
by a factor of k 2 . We see that the answer passes all checks.

185
19.2 Identity theorem
Another peculiar property of holomorphic function: if we know the values of a holomorphic
function in a small disk inside the domain, then the function values in the whole domain are
uniquely determined. As long as it has a nonempty interior so that we can find a sequence
of points converging to an accumulation point, then there is only one way to extend the
function to a larger domain. This property is content of the identity theorem.
In the proof below, we need a result about connected set. In point-set topology, a set
is said to be disconnected if it can be decomposed as the union of two nonempty disjoint
open sets. A set is connected if it is not disconnected, i.e., if it cannot be decomposed as
the union of two nonempty disjoint open subsets. A set that is open and closed is often
called a clopen set. It is easy to see that empty set is a clopen set. The complement of the
empty set, which is the whole space, is also clopen.

Theorem 19.3. The subsets of a path-connected open set S in an Euclidean space

(e.g. R2 ) that are clopen (in the relative topology) are precisely the empty set and S.

Sketch of proof. One can prove that a path-connected set is always connected2 . In the
other direction, we know that an open and connected set in an Euclidean space is path-
connected3 .
It remains to prove that the empty set and the whole set are the only clopen sets in
a connected set. This can be easily done, because if A is a clopen set in S that is not
the empty set and not the entire set S, then A and Ac are nonempty and disjoint subsets,
whose union is the whole space. The complement Ac is open (in the relative topology of
S) because A is assumed to be closed. We can thus write S as a union of two nonempty
disjoint open sets, violating that assumption that S is connected.

2
A proof that a path-connected set is connected
3
A proof from mathonline.com

186
 Theorem 19.4 (Identity Theorem). Suppose f and g are holomorphic in a do-
main D. The followings are equivalent:

1. f (z) = g(z) for all z ∈ D.

2. f (z) = g(z) on a sequence of distinct points in D that has a limit point in D.

3. There exists a point w in D such that f (n) (w) = g (n) (w) for all n = 0, 1, 2, 3, . . ..

Proof. The two directions “(1) implies (2)”, and “(1) implies (3)” are trivial. To show
that the two reverse directions are also valid, it is sufficient to consider the difference
h(z) ≜ f (z) − g(z), and show that h(z) = 0 for all z ∈ D.
(2) ⇒ (1). Suppose h(z) = 0 on a sequence of distinct points (zn )∞ n=1 , which is a
sequence of distinct points with a limit point in D. Denote the limit point by z0 . We have
h(zn ) = 0 for all n = 1, 2, 3, . . .
Since z0 is in the interior of the domain D and h(z) is holomorphic at z0 , we can find
a radius r > 0 such that h is complex differentiable in the open disc |z − z0 | < r. By
Theorem 18.9, we can expand h as a power series at the point z0 ,
∞
X
h(z) = ck (z − z0 )k , (19.4)
k=0

where ck are complex numbers, and the equality in (19.4) holds for all z in the open disc
|z − z0 | < r.
Because the sequence (zn )∞ n=1 converges to z0 , it will eventually fall inside this open disc
|z − z0 | < r. So, we can find an integer N such that |zn − z0 | < r for all n ≥ N . Obviously
this subsequence also converges to z0 . By Theorem 18.12, there is a unique choice of
coefficients ck ’s such that h(zn ) = 0 for all n ≥ N . But in fact, the power series with all
zero coefficients is one such possibility. Therefore, we can conclude that the coefficients ck
in (19.4) must be all equal to zero. This proves that h(z) is identically equal to zero in the
open disc |z − z0 | < r.
We now consider the subset in D in which h(z) is equal to 0,

{z ∈ D : h(z) = 0}.

187
We know that it has nonempty interior. Let

U ≜ {z ∈ D : h(z) = 0}◦ ,

where ◦ denote the interior of a set. By the argument in the previous paragraph, U is
nonempty, as it contains the open disc |z − z0 | < r.
By the meaning of “interior”, U is an open set by construction. We claim that U is also
a closed set. We can see why it is closed by checking that any accumulation point of U is
in U . Let w0 be such an accumulation point. There is a sequence of points w1 , w2 , w3 , . . .
in U converging to w0 . By the continuity of function h(z), we have

h(w0 ) = h(lim wk ) = lim h(wk ) = 0.

k k
The point w0 is in the domain D, and by the hypothesis in the theorem, h is holomorphic
in a small open disc centered at w0 . Using the same argument as in the first part of the
proof, h(z) is identically zero in this disc. Hence w0 is in the interior of {z ∈ D : h(z) = 0},
and thus is in U .
Since D is connected, by the fact stated in Theorem 19.3, any open and closed set in
D is either the empty set or the set D. But U is non-empty. The only possibility is that
U = D. Therefore, all points in D is in the set U , and hence satisfy h(z) = 0. This proves
that h(z) is identically equal to 0 in domain D.
(3) ⇒ (1). There is an open disc with positive radius, centered at w, such that the
function h is holomorphic inside this open disc. By hypothesis, the n-th derivative h(n) (w)
is equal to 0 for all positive integer n. Therefore, using the same idea in the proof of “(2)
implies (1)”, we can say that h(z) is equal to zero for every point z within this open disc,
and therefore h(z) = 0 for all points z in D.

The identity theorem gives a constraint on the vanishing locations of holomorphic func-
tion.
Corollary 19.5. Suppose f (z) is an analytic function defined on a domain D that
is not constantly equal to 0. Then the set

{z ∈ D : f (z) = 0}

has no accumulation point in domain D (See Definition 18.11 for the meaning of
accumulation point).

188
 From the proof of the identity theorem, we have proved that a nonzero holomorphic
function cannot vanish on an open disc. This gives another way to think about the identity
theorem. Suppose we are given a holomorhpic function f and we want to change the
function values f (z) for some points z inside a small set (perhaps we would like to increase
the modulus of the function value), without modifying the function value outside this set.
The identity theorem says that this kind of local operations are impossible, if we wish
to maintain the analyticity of the function. We say that a holomorphic function is rigid
precisely when we are referring to this property.
We record this fact below.

Corollary 19.6. If f (z) is analytic in a domain D and the values of f (z) is zero
within an open disc in D, then f is necessarily the zero function in D.

Example 19.3. There is only one way to extend the real exponential function ex , for real
value x, to an analytic function that is defined on the whole complex plane. Suppose g(z)
is an entire function with the property that g(x + i0) = ex for all real number x. we can
find a sequence of points (xk )∞
k=1 on the real axis that approaches the origin x = 0. By the
identity theorem, there is only one analytic function such that the value of xk is exk for all
k, and the complex exponential function defined by (5.16) is one such choice. Therefore, the
complex exponential function defined in (5.16) is the unique analytic function that agrees
with the real exponential function.
Likewise, we can see that there is only one way to extend the sine, cosine, and tangent
functions to the complex domain.
Remark. A common proof technique in differential geometry is partition of unity. This
method involves a family of “bump functions”, each of which is constantly equal to 1
within a compact set, and drops rapidly to zero outside this compact set, so that the sum
of the functions in this family is the constant function that identically equal to 1. In view
of the identity theorem, thee is no partition of unity in complex analysis if we want each
bump function to be an analytic function.

189
20 Maximum modulus principle, zero of a function
Summary

ˆ Maximum modulus principle

ˆ Zeros of a function

ˆ Classification of isolated singularity

20.1 Maximum modulus principle

Another special feature of holomorphic function is the maximum modulus principle.

Theorem 20.1 (Maximum modulus principle). Let f (z) be a holomorphic function

defined on a domain D. If |f (z)| attains a maximum value at an interior point in
D, then f (z) is a constant function.

Remark. Note that the maximum value mentioned in the theorem should be located in the
interior of D, because D is assumed to be open and does not include any boundary point.
An equivalent form of the maximum modulus principle is: if f is a nonconstant analytic
function, then the maximum value of |f (z)| over a domain D occurs on the boundary of D.

Proof. Suppose the modulus of f (z) attains maximum at an interior point z0 of D. We can
find a sufficiently small ϵ > 0 such that all points z in the open disc |z − z0 | < ϵ lie inside
the domain D. By assumption, we have

|f (z)| ≤ |f (z0 )| whenever |z − z0 | < ϵ.

Pick any real number r that is between 0 and ϵ. By Cauchy integral formula (Theorem 17.1),
Z
1 f (z)
f (z0 ) = dz.
2πi |z−z0 |=r z − z0

Represent the circle |z − z0 | = r by z = z0 + reiθ , for 0 ≤ θ ≤ 2π. Then

Z 2π
1 f (z0 + reiθ ) iθ
f (z0 ) = ire dθ
2πi 0 reiθ
Z 2π
1
= f (z0 + reiθ ) dθ.
2π 0

190
The above equation says that the value f (z0 ) at the center of the circle equals the mean of
the function values on the circle.
Taking modulus on both sides, we get
Z 2π
1
|f (z0 )| ≤ |f (z0 + reiθ )| dθ
2π 0
Z 2π
1
≤ |f (z0 )| dθ
2π 0
≤ |f (z0 )|.

By the sandwich argument, all of the above equalities hold, and we obtain
Z 2π Z 2π
1 1
|f (z0 )| dθ = |f (z0 + reiθ )| dθ,
2π 0 2π 0
or equivalently Z 2π h i
|f (z0 )| − |f (z0 + reiθ )| dθ = 0.
0
Since |f (z0 )| ≥ |f (z0 +reiθ )| by hypothesis, by the monotone property of integral, we obtain

|f (z0 )| = |f (z0 + reiθ )|

for all 0 ≤ θ ≤ 2π.

Because the radius r can be any positive real number less than ϵ, we obtain

|f (z0 )| = |f (z0 + reiθ )|

for all r ∈ (0, ϵ) and θ ∈ [0, 2π]. Thus, the modulus |f (z)| is equal to a constant on the open
disc |z − z0 | < ϵ. By considering the Cauchy-Riemann equations of f (z), we can see that
f (z) must be a constant function in the open disc |z − z0 | < ϵ, and the constant is f (z0 ).
Finally, by the Identity Theorem (Theorem 19.4), f (z) is constant throughout D.

Example 20.1. Find the maximum value of |z 2 − z| in |z| ≤ 1.

By the maximum modulus principle, we only need to consider the boundary |z| = 1.
On the boundary the function to be maximized is

|z 2 − z| = |z| · |z − 1| = |z − 1|.

The maximum value occurs at the point on the unit circle that is farthest away from z = 1.
The maximum value is thus |(−1)2 − (−1)| = 2.

191
Example 20.2. The maximal modulus principle implies the corresponding statement for
harmonic function: if u(x, y) is harmonic inside a region and if D is a closed disc within the
region, then the maximum value of u(x, y) over the closed disc D is attained somewhere on
the boundary of D.
We can see this by considering a harmonic conjugate v(x, y) of u(x, y), and the complex
function
f (z) = eu(x,y)+iv(x,y) .

Because u(x, y) + iv(x, y) and exp are both holomorphic, so is their composite function.
The modulus of f is eu(x,y) . By the monotonicity of the real exponential function, the
modulus |f (z)| attains the maximum value exactly at the location where u(x, y) attains the
maximum value. Hence, by maximum modulus principle, the maximum value of u(x, y) in
the disc D can only occur at the boundary of D.

20.2 Zeros of a function

Definition 20.2. A complex number z0 is called a zero of a function f if f (z0 ) = 0.

Notation: We say that “f has a zero at z0 ” if z0 is a zero of f .

Example 20.3. For a polynomial f (z) = a0 + a1 z + · · · ad z d , there are at most d zeros. The
set of zeros is a finite set.
Example 20.4. The complex exponential function has no zero, because the equation ez = 0
has no solution. However, if c is a nonzero complex constant, the equation ez = c has
infinitely many solutions. The zero of ez − c are precisely the complex log of c.

Theorem 20.3. The zeros of a nonzero function are isolated.

More precisely, the theorem says that if f (z) is a nonzero holomorphic function and z0
is a zero of f , then we can find a small radius r such that the function f (z) is nonzero in
the punctured disk 0 < |z − z0 | < r.
Example 20.5. The function sin(π/z) is equal to zero for z in the set
1
{ : n = ±1, ±2, ±3, . . .}.
n

192
The origin is an accumulation point of this set. However, the function sin(π/z) is not zero.
It does not violate the Theorem 20.3 because the domain of this function does not include
the origin, and the origin is not a zero of the function.
The proof follows directly from Corollary 19.5, which states that the set of zeros of a
nonzero function cannot have an accumulation point in the domain of the function. Another
way to express the same idea is that there are finitely many zeros in a closed and bounded
region.

Corollary 20.4. Suppose f (z) is a nonzero holomorphic function defined in a do-

main D. Then in any closed and bounded region in D, the function f (z) has finitely
many zeros.

Proof. We prove by contradiction. Suppose B is a closed and bounded region in D such that
f (z) has infinitely many zeros. Because B is closed and bounded, we can apply Bolzano-
Weierstrass theorem and conclude that the set of zeros of f (z) in B has an accumulation
point. The accumulation point is a point in D. Hence, by the identity theorem, the
function f (z) is a constant zero function, violating the assumption that f (z) is a nonzero
function.

Definition 20.5. If f is not identically equal to zero, the order of a zero is the
exponent of the first nonzero term in the power series expansion of f at z0 . In other
words, f (z) has a zero of order m at z0 if the power series of f (z) at z0 is

am (z − z0 )m + am+1 (z − z0 )m+1 + am+2 (z − z0 )m+2 + · · ·

with am ̸= 0. A zero of order 1 is called a simple zero. A zero order 2 is called a

double zero.

The order of zero is also referred to as the order of vanishing. We know that a holo-
morphic function can always be expanded as a power series at any point in its domain, and
the coefficients are uniquely determined (Theorem 18.12). This ensures that the order of a
zero is well defined.
Because power series can be differentiated term-wise, the order of zero is the smallest
integer m such that f ′ (z0 ), f ′′ (z0 ), . . . , f (m−1) (z0 ) are all equal to zero but f (m) (z0 ) is

193
nonzero.
The behavior of function f (z) near the point z0 can be approximated by the power
function (z − z0 )m . If z0 is a zero of order m and we travel around the point z0 on a small
circle centered at z0 , the image through the function f will cycle m times around the point
f (z0 ).

20.3 Classification of singular points

Definition 20.6. A point z0 in C is called a singular point of f (z) if f (z) is not

complex differentiable at z0 . It is called an isolated singular point if f (z) is not
holomorphic at z0 but is holomorphic in a punctured disc 0 < |z − z0 | < r for some
radius r.

Example 20.6. The function f (z) = 1/(z 2 + 1) has two isolated singular points located at
z = i and z = −i.
1
Example 20.7. The function sin(π/z) has a non-isolated singular point at z = 0. It is because
∞
the sequence {1/n}n=1 converges to z = 0, and this function is singular at z = 1/n, for
n = 1, 2, 3, 4, . . ..
Example 20.8. The principal log function with domain

C \ {x + iy : y = 0, x ≤ 0}

has a non-isolated singularity at the origin.

Isolated singularity can classified into three categories.

Definition 20.7. An isolated singular point z0 of f (z) is called

(i) a removable singularity if |f (z)| is bounded in a neighborhood of z0 ;

(ii) a pole if |f (z)| → ∞ as z → z0 ;

(iii) an essential singularity otherwise.

We note that the three cases exhaust all possibilities; an isolated singular point that is
not removable and is not a pole is by definition an essential singularity. Because essential
singularity is more complicated, we will mainly consider singularity of the first two types.

194
 Figure 30: Phase portrait of e1/z near the origin.

Definition 20.8. A complex function is said to be meromorphic if the singular

points are isolated, and all of them are removable or poles, i.e., there is no essential
singularity within the domain.

Example 20.9. The function f (z) = sin(z)z has a removable singularity at z = 0. This
function is basically the same as the function defined by the power series
1 z3 z5 z2 z4
(z − + − ···) = 1 − + − ···
z 3! 5! 3! 5!
The power series has infinite radius convergence, and hence is an entire function.
n
−1
Example 20.10. For any positive integer n, the function zz−1 has a removable singularity
at z = 1. The function is not defined at z = 1. But if we take limit as z approach 1, the
function value has limit equal to n. The behavior of this function is essentially the same as
the polynomial z n−1 + z n−2 + · · · + z + 1.
1
Example 20.11. The function f (z) = z(z+1) has two poles located at z = 0 and z = −1.
Example 20.12. The complex trigonometric functions tan(z) and cot(z) have infinitely many
poles. The poles of tan(z) are the same as the zeros of cos(z), and they are located at ±π/2,
±3π/2, ±5π/2, . . .

195
Example 20.13. The function f (z) = e1/z has an essential singularity at z = 0. We can see
that it is an essential singular point by approaching z = 0 from the right and from the left.
For real variable x, if we take x → 0 from the right, the value of f (z) tends to positive
infinity. Hence it is not bounded near z = 0. On the other hand if we take x approaching 0
from the negative real axis, then f (z) tends to 0. The modulus is not approaching infinity.
See Fig. 30 for the phase portrait of this function. (Compare with the other phase portraits
in Fig. 5.)
The complex function in 20.7 is not meromophic, because it has non-isolated singularity.
The function e1/z in Example 20.13 are not meromorphic because it contains an essential
singularity.

20.4 Appendix: Casorati-Weierstrass theorem

The behavior near an essential singularity is complicated. The Casorati-Weierstrass theo-
rem captures part of the complications.

Theorem 20.9 (Casorati-Weierstrass). Suppose f (z) is homophobic in U \ {z0 },

where U is an open set and z0 is an essential singularity. For any positive real
numbers ϵ and δ, and for any choice of complex number w, there exists a complex
number z U such that

0 < |z − z0 | < δ, and |f (z) − w| < ϵ.

An equivalent but more compact way to express the conclusion in the theorem is: If N
is any neighborhood of z0 contained in U , then the image of N \ {z0 } under the function f
is dense in C.
We can formulate this theorem in terms of a game. In this game a complex function f (z)
and a point z0 in the domain of f (z) are given. Player A first selects a small neighborhood
of z0 and a complex number w. Then, Player B tries to find a sequence of points in the
neighborhood such that the image of the sequence under the function f is convergent with
limit w. Player B wins if he/she can find such a sequence, otherwise Player A wins.
If z0 is an essential singular point, then Player B will always win.
The proof of Casorati-Weierstrass theorem is skipped in this note. We will focus more
on the other two types of singularity.

196
21 Laurent series
Summary:

ˆ Riemann’s theorem on removable singularity

ˆ Laurent series

ˆ Residue

The following theorem is an extension of Theorem 17.2 to multiple holes.

Theorem 21.1 (Cauchy theorem for function with isolated singularity). Suppose
complex function f (z) defined on a domain D has isolated singularity, i.e., f (z) is
holomorphic in D except for some isolated points. Let C be a simple closed curve in
D with counter-clockwise orientation, so that no singular point of f (z) lie on C. Let
z1 , z2 , . . . , zm denote the singular points inside C. For k = 1, 2, . . . , m, let Ck be a
positively-oriented circular contour centered at zk , with the radius rk is chosen such
that f (z) is holomorphic in the punctured disc 0 < |z − zk | < rk . Then
I m I
X
f (z) dz = f (z) dz.
C k=1 Ck

We note that the assumption of isolated singularity is used in justifying why we can
choose a sufficiently small radius rk so that there is only one singular point in the open disc
D(zk , rk ). Fig. 31 illustrates the notation in the theorem.
We will use a property of simple closed curve: the contour C divides the complex plane
into two parts: interior and exterior. Moreover, for any point α in the interior, the winding
number of C relative to α is 1, and for any point β on in the exterior, the winding number
of C relative to β is equal to 0. (Why it is true is the content of the Jordan curve theorem.)

Proof. We first illustrate the proof when C is a circle and there are two isolated singular
points inside C. We divide the ares bounded between the outer circle and the inner circles
into six parts. We first draw a straight line that joins the two singular points z1 and z2 .
Then we draw perpendicular lines at z1 and at z2 . We obtain six closed paths. The sum

197
 Figure 31: Cauchy theorem with isolated singularity.

of these six closed paths is the sum of the three circles C, −C1 and −C2 , because all the
internal line segments cancel.
Note that we take we take the negation of C1 and C2 . See Fig. 32.
For each cycle, the area bounded within a closed path is not convex. However, each
closed path is contained in a convex region, that avoids the two singular points z1 and z2 .
Since f (z) is holomorphic in the convex region, the function f (z) has a primitive, and hence
the integral along the closed path is equal to 0.
Each of the six closed paths is contained inside a convex region, in which f (z) is holo-
morphic. Therefore the integral of all six closed paths equal 0.
The proof idea can be modified to incorporate more than two inner circles. The shape
of circle is not crucial. The theorem is valid for other simple shapes such as ellipse, square,
rectangle, etc. The crucial point is that the winding number of the curve C relative to each
interior point is 1.

21.1 Removable singularity

The next theorem justifies the name “removable singularity”.

198
Figure 32: Dividing the area into six parts. Each part is contained in a convex set in which
the function to be integrated is holomorphic.

Theorem 21.2 (Riemann’s theorem on removable singularity). Suppose f is holo-

morphic in a domain that contains a punctured disc {z : 0 < |z − z0 | < ϵ}. If f (z)
is bounded in this punctured disc, then we can re-defined f at z0 so that f becomes
holomorphic at z0 .

More precisely, the theorem is saying that there exists a function f˜(z) that is holomor-
phic for |z − z0 | < ϵ such that f˜(z) = f (z) whenever 0 < |z − z0 | < ϵ.

Proof. Suppose f is holomorphic and bounded in a punctured disc {z ∈ C : 0 < |z−z0 | < ϵ}.
We want to prove that we can define f (z0 ) appropriately to make f (z) continuous at z0 .
Let D(z0 , ϵ) denote the open disc |z −z0 | < ϵ and let D(z0 , ϵ)\{z0 } denote the punctured
disc. By the hypothesis in the theorem, there is a real number M such that |f (z)| ≤ M for
z in the punctured disc D(z0 , ϵ) \ {z0 }. Define
I
1 f (w)
f˜(z) ≜ dw,
2πi C w − z
where C is the circle |z − z0 | = ϵ, with counter-clockwise orientation, and z is any point
inside the circle. We want to show that (i) f˜(z) = f (z) for all z ∈ D(z0 , ϵ) \ {z0 }, and (ii)

199
f˜ is analytic in D(z0 , ϵ).

C0
z0 C1
ϵ z

Let z be a point in the puncture disc 0 < |z −z0 | < ϵ. Draw a small circle C0 of radius r0
with center at z0 , such that C0 is inside the circle C but does not contain z. Draw another
small circle C1 of radius r1 with center at z such that C1 is inside the circle C but does not
contain z0 .
By Theorem 21.1, we have
Z Z Z
˜ 1 f (w) 1 f (w) 1 f (w)
f (z) = dw = dw + dw. (21.1)
2πi C w − z 2πi C0 w − z 2πi C1 w − z

Since f (z) is analytic inside the circle C1 , we can apply Cauchy integral formula (Theo-
rem 17.1) to obtain Z
1 f (w)
dw = f (z).
2πi C1 w − z

We next prove that | C0 fw−z (w)

R
dw| can be made arbitrarily small. For complex number w
on C0 , the distance |w−z| is lower bounded by |z−z0 |−r0 . By ML inequality (Theorem 15.4)
Z
f (w) M
dw ≤ 2πr0 ,
C0 w − z |z − z0 | − r0

which approaches 0 as r0 → 0. Because (21.1) holds for all sufficiently small radius r1 of
C1 , we can conclude that f˜(z) = f (z). Since this holds for all z ∈ D(z0 , ϵ) \ {z0 }, this
proves the part (i).
For part (ii), we just need to show that f˜(z) is holomorphic at the center z0 , because
f˜(z) is equal to f (z), which is assumed to be holomorphic in D(z0 ; ϵ) \ {z0 }. We first use

200
the definition of f˜ at z0 to write

f˜(z0 + h) − f˜(z0 )
I
1 f (w) f (w)
= − dw
h 2πih C w − z0 − h w − z0
Z
1 f (w)
= dw
2πi C (w − z0 − h)(w − z0 )
where C is the circle |z − z0 | = ϵ, with counter-clockwise orientation. We then show that it
converges as h → 0. Because we already has a target limit, we calculate

f˜(z0 + h) − f˜(z0 )
Z Z
1 f (w) 1 f (w)h
− 2
dw = dw
h 2πi C (w − z0 ) 2π C (w − z0 )2 (w − z0 − h)
1 M |h|
≤ 2
2πϵ.
2π ϵ (ϵ − |h|)
The last inequality is the ML inequality. For fixed ϵ, it converges to 0 as |h| approaches 0.
Therefore f ′ (z0 ) exists and is equal to
Z
′ 1 f (w)
f (z0 ) = dw.
2πi C (w − z0 )2
This proves part (ii).

In view of Theorem 21.2, we can ignore any removable singularity, because we can
always re-define the function appropriately so that it is no longer a singularity. We note
that Theorem 21.2 fails for real functions. The real-valued function f (x) = |x| is not
differentiable at x = 0 and is bounded near 0. It is continuous at x = 0, but is not analytic
at x = 0.

21.2 Pole order

Notation: We sometime say that “f (z) has a pole at z0 ” if z0 is a pole of f (z).
Suppose f (z) has a pole at z0 . By definition we have |f (z)| → ∞ as z → z0 , and f (z)
is nonzero in a small neighborhood of z0 . Consider the reciprocal function 1/f (z). It is
bounded in a neighborhood of z0 . Indeed, by the meaning of “diverging to infinity”, it
means that for any positive positive M , we can find a sufficiently small ϵ > 0, such that
|f (z)| > M for all 0 < |z − z0 | < ϵ. In other words, we have |1/f (z)| < 1/M for all
|z − z0 | < ϵ. We can take M to be an arbitrary positive real number, and see that |1/f (z)|
is bounded by 1/M in a punctured disc.

201
 The singular point z0 is a removable singularity of 1/f (z). Since |f (z)| → ∞ as z → z0 ,
the domain of function 1/f (z) can be extended to an open disc |z − z0 | < ϵ by defining

1/f (z) if 0 < |z − z | < ϵ
0
g(z) ≜
0 if z = z . 0

By the theorem of removable singularity, the new function g(z) is holomorphic at z0 , can
thus be expressed as a power series centered at z0 . The constant term must be equal to zero
because g(z0 ) = 0. However, the coefficients of the power series expansion of g(z) cannot
be all zero, because g(z) has an isolated zero at z0 . Let m be the order of the zero z0 of
g(z). (The order m must be finite by Corollary 20.4.) We can write

g(z) = am (z − z0 )m + am+1 (z − z0 )m+1 + am+2 (z − zm )m+2 + . . .

where am is a nonzero coefficient. We can write f (z) as

1  1 
f (z) = · (21.2)
(z − z0 )m am + am+1 (z − z0 ) + am+2 (z − z0 )2 + · · ·
Because am is a nonzero constant, the denominator is nonzero within a small neighborhood
of z0 . Hence, the fraction enclosed in the parentheses is holomorphic at z0 . We can thus
write the fraction in the parentheses as a power series (Theorem 18.9),

f (z) = (z − z0 )−m (c0 + c1 (z − z0 ) + c2 (z − z0 )2 + · · ·

= c0 (z − z0 )−m + c1 (z − z0 )−m+1 + c2 (z − z0 )−m+2 + · · ·

We note that the first coefficient c0 is nonzero, because am is nonzero. Indeed, we have
c0 = 1/am .
This leads to define the order of a pole as the zero order of the reciprocal function.

Definition 21.3. The order of a pole z0 of f (z) is the smallest positive integer m
such that limz→z0 (z − z0 )m f (z) is a nonzero constant. A pole of order 1 is called a
simple pole. A pole of order 2 is called a double pole.

(z−1)(z−2)
Example 21.1. Consider a rational function f (z) = (z−3)(z+i) 2 . This function f (z) has two

zeros of order 1 at z = 1 and z = 2. There is a simple pole at z = 3, because

1 h (z − 1)(z − 2) i
f (z) =
z−3 (z + i)2

202
 (z−1)(z−2)
and (z+i)2
is nonzero when evaluated at z = 3. We can formally check this by calculating
(z − 1)(z − 2) (3 − 1)(3 − 2)
lim (z − 3)f (z) = lim 2
= ̸= 0.
z→3 z→3 (z + i) (3 + i)2
Another pole of f (z) is located at z = −i. This is a double pole because
(z − 1)(z − 2)
lim (z + i)f (z) = lim = ∞,
z→−i z→−i (z − 3)(z + i)
but
(z − 1)(z − 2) (−i − 1)(−i − 2)
lim (z + i)2 f (z) = lim = ̸= 0.
z→−i z→−i z−3 −i − 3
Example 21.2. The function f (z) = sin(z)zn has a pole of order n − 1 at the origin. It is
because sin(z)/z has a removable singularity at z = 0. If we multiply f (z) by z n−1 , the
product f (z)z n−1 has a finite limit as z approaches 0. On the other hand, the reciprocal
1/f (z) has poles at z = ±kπ, for k = 1, 2, 3, . . .

21.3 Laurent series and residue

We have seen that at a pole z0 of order m, we can express the function as an infinite sum
∞
X
ck (z − z0 )k ,
k=−m

where c−m is a nonzero coefficient, in a neighborhood of z0 .

In the case of essential singularity, we can also expand the function as an infinite series.
For example, we can expand the function e1/z at z = 0 as
1 1 1
e1/z = 1 + + 2
+ + ···
z 2!z 3!z 3
which converges whenever z ̸= 0 (see Example 20.13).
This motivates the definition of Laurent series.
Definition 21.4. A Laurent series is an infinite sum in the form
∞
X
cn (z − z0 )n . (21.3)
n=−∞

The first summation ∞ n

P
n=0 cn (z − z0 ) is called the analytic part, or the regular part.
P∞
The second summation n=1 c−n (z − z0 )−n is called the principal part. We say that
the Laurent series converges if both the analytic and principal parts converge.

203
 If the principal part is zero, a Laurent series reduces to a Taylor series. Sometime we
write a Laurent series in the form
X∞ ∞
X
an (z − z0 )n + bn (z − z0 )−n
n=0 n=1

to emphasize the analytic part and the principal part.

Remark. Laurent series is closely related to Fourier series. Suppose the coefficients an ’s in
(21.3) are conjugate symmetric, i.e., c−n = c∗n , and the Laurent series converges on the unit
circle. If we substitute z by e2πit , for some real number t, then
∞
X
cn e2πint
n=−∞

is a real function with period 1. In fact we can re-write it as

∞
X
c0 + (cn e2πint + c−n e−2πint )
n=1
X∞
= c0 + Re(cn ) cos(2πnt) − Im(cn ) sin(2πnt),
n=1

which is in the form of a Fourier series. When the domain of the Laurent series is restricted
to the unit circle, it is the same as complex Fourier series.
In order to simplify notation, we consider z0 = 0 below. The first structural result is
that the region of convergence has annulus shape.
P∞ n
Theorem 21.5. The region of convergence of a Laurent series n=−∞ an z has
the shape of an annulus R1 < |z| < R2 . (If R1 > R2 , the region of convergence is
empty.)

Proof. By definition, Laurent series converges if and only if the analytic part and the
principal parts converge. The analytic part is a power series, and hence is convergent inside
an open disc, say with radius R2 . The value of R2 is given by the Hadamard formula
(Theorem 18.4).
For the principal part, we make a substitution u = 1/z,
∞
X ∞
X
−n
bn z = bn un .
n=1 n=1

204
This is a power series with variable u, and hence will converge in an open disc, say with
radius 1/R1 .
Combining the two parts, we see that this Laurent series converges if |z| < R2 and
1/|z| = |u| < 1/R1 . Therefore, the region of convergence is in the form

R1 < |z| < R2 .

The next example illustrates the dependency on the region of convergence.

1
Example 21.3. Let f (z) denote the function (1−z)(3−z) . Find the Laurent series expansion
at three regions (i) |z| < 1, (ii) 1 < |z| < 3, and (iii) 3 < |z|.
We first obtain the partial fraction expansion of f (z),
1 1 1 1
f (z) = − .
2 (1 − z) 2 (3 − z)

(i) When |z| < 1, we can use geometric series to expand both terms in the partial
fraction expansion,
1 1 1 1
= (1 + z + z 2 + z 3 + · · · ) − · (1 + (z/3) + (z/3)2 + · · ·
(1 − z)(3 − z) 2 2 3
∞
1X
= (1 − (1/3)k+1 )z k .
2
k=0

This Laurent series converges for |z| < 1, and is indeed a power series.
(ii) If 1 < |z| < 3, we write the first fraction in the partial fraction expansion as
−1 1
2 1 − (1/z)

We can then expand f (z) in the annulus 1 < |z| < 3 as

1 −1 1 1 1
= −
(1 − z)(3 − z) 2 1 − (1/z) 2 · 3 1 − (z/3)
1 1 1 1 z z2
= − (1 + + 2 + · · · ) − (1 + + 2 + · · · ).
2 z z 6 3 3
The principal part of this Laurent series is non-zero and contains infinitely many terms.

205
 (iii) When 3 < |z|, the Laurent series expansion has zero analytic part.
1 −1 1 1 1
= +
(1 − z)(3 − z) 2 1 − (1/z) 2z 1 − 3/z
1 1 1 1 3 32
= − (1 + + 2 + · · · ) + (1 + + 2 + · · · )
2 z z 2z z z

Remark. In contrast to power series, it is not true in general that cn = f (n) (0)/n! for
Laurent series. Nevertheless, the coefficients can be computed using contour integrals that
is similar to the case of power series.

Theorem 21.6 (Existence and uniqueness of Laurent series). Suppose f (z) analytic
in an annulus R1 < |z − z0 | < R2 . where R1 and R2 are positive real constants.
There exist coefficients (ak )∞
k=−∞ such that

∞
X
f (z) = ak (z − z0 )k
k=−∞

for all z in the annulus R1 < |z − z0 | < R2 .

The coefficients ak is uniquely determined by contour integral
I
1 f (z)
ak = dz
2πi C (z − z0 )k+1

for k ∈ Z, where C is a simple contour in the annulus R1 < |z − z0 | < R2 going

around z0 with positive orientation.

The proof idea is similar to the analogous theorem for power series (Theorem 18.9).
The existence part of the theorem is relegated to the appendix at the end of this section.
We note that the contour C in the theorem must lie within the annulus R1 < |z − z0 | < R2 .
The computation of the coefficients ak ’s depend on the region of convergence.
We summarize the relationship between principal part of Laurent series and pole order:
Suppose z0 is an isolated singularity of f (z). Expand f (z) using a Laurent series with
center at z0 .

1. If z0 is a removable singularity, then the principal part is zero.

2. If z0 is a pole of f (z), then the principal part has finitely many terms. The pole order

206
 is the integer m such that a−m (z − z0 )−m is the smallest-degree term with a nonzero
coefficient.

3. If z0 is an essential singularity, then the principal part has infinitely many terms.

We define the notion of residue in terms of the coefficients of a Laurent series.

Definition 21.7. Consider a complex function f (z) that is analytic in a domain

except some isolated singular points. The residue of f (z) at a point z0 is defined as
the coefficient b1 in the Laurent series expansion
∞
X ∞
X
f (z) = an (z − z0 )n + bn (z − z0 )−n ,
n=0 n=1

with convergence in a small punctured disc 0 < |z − z0 | < ϵ centered at z0 , for some
ϵ > 0.

It is important to specify the region of convergence 0 < |z − z0 | < ϵ in this definition,

because the coefficients depend on the choice of the region of convergence.
Notation: There are several notation for residue, e.g. Res(f ; z0 ), Resz0 (f ), and Res (f ).
z=z0
When f (z) is analytic at z0 , the residue at z0 is zero.
From the formula for computing the coefficients of a Laurent series in Theorem 21.6,
we have the following equation
Z
1
Res(f ; z0 ) = f (z) dz, (21.4)
2πi C

where C is a small contour around z0 that contains no singular point except z0 . In fact, we
can take (21.4) as an alternate definition of of residue.
Example 21.4. The residue of e1/z at z = 0 is 1, because the coefficient of 1/z in
1 1
e1/z = 1 + + + ···
z 2!z 2
is equal to 1.
Theorem 21.1 can be expressed in terms of residues.

207
 Theorem 21.8 (Residue theorem). Suppose f is analytic in a domain D except for
some isolated singularities. If C is a simple closed curve enclosing singular points
z1 , z2 , . . . , zn in the interior, then
Z n
X
f (z) dz = 2πi Res(f ; zk ).
C k=1

Proof. For k = 1, 2, . . . , n, we draw a small circle Ck centered at zk so that the circle does
not contain the other singular points. By Theorem 21.1,
Z n Z
X
f (z) dz = f (z) dz.
C k=1 Ck

1
R
Since the residue of f at zk is equal to the integral 2πi Ck f (z) dz, the integral around the
contour Ck can be replaced by the corresponding residue.
Z n
X
f (z) dz = 2πi Res(f ; zk ).
C k=1

21.4 Appendix: Existence of Laurent series expansion

For notation convenience, we assume that the center z0 is the origin.
We set up the proof by picking two finite real numbers r1 and r2 such that

R1 < r1 < r2 < R2 .

In general, R2 may equal to infinity. In this case r2 can be any real number larger than r2 .
We consider the region with modulus less than r2 and larger than r1 , by drawing a circle
C1 with radius r1 centered at origin, and a circle C2 with radius r2 centered at the origin,
with counter-clockwise orientation.
Let z be a complex number in the region between C1 and C2 . Our setup ensures that
function f (z) is well-defined on C1 and C2 , and the input value z stays away from the two
boundaries.

208
 C2
z
r2
C3

w r1 C1

Figure 33: Proof of the existence of Laurent series expansion

Draw a positively-oriented circle C3 centered at z with radius r3 such that C3 is within

the area between C1 and C2 . The notation is illustrated in Figure 33.
We can extend the argument in the deformation theorem (Theorem 17.2) to more than
one circles. The integral over the outer circle C2 is the same as the sum of integrals over
the two inner circle C1 and C3 ,
Z Z Z
f (w) f (w) f (w)
dw = dw + dw.
C2 w − z C1 w − z C3 w −z

By Cauchy integral formula, f (z) can be expressed as a contour integral along the circle
C3 . Therefore, we can write
Z
1 f (w)
f (z) = dw (21.5)
2πi C3 w − z
Z Z
1 f (w) 1 f (w)
= dw − dw. (21.6)
2πi C2 w − z 2πi C1 w − z

Because f is continuous, and hence bounded on C1 and C2 , the contour integrals

Z Z
f (w) f (w)
dz and dz
C1 w − z C2 w −z

209
are well-defined and finite. We will not have ∞ − ∞.
We claim that the first integral can be expanded as a convergent Taylor series
Z ∞
1 f (w) X
dw = an z n , (21.7)
2πi C2 w − z
n=0

and the second integral can be represented as

Z ∞
1 f (w) X
dw = bn z −n . (21.8)
2πi C1 w − z
n=1

The proof of (21.7) is similar to the proof of Theorem 18.9 and is skipped. We note that
the derivation only requires that f is holomorphic on the contour C2 , so that we can apply
the ML inequality in one of the steps. We do not require that f (z) is analytic throughout
the circular region enclosed by C2 .
In the remainder of the proof we consider (21.8). We write
1 1 1 w w2 w3

= = 1 + + 2 + 3 + ···
z−w z(1 − w/z) z z z z
and observe that it converges for |w| < |z|. Since the location of z is outside C1 , we have
convergence for all w ∈ C1 ,
∞ N ∞ N
1 X wn X wn−1 X wn−1 X wn−1 wN
n
= n
+ n
= n
+ N .
z z z z z z (z − w)
n=0 n=1 n=N +1 n=1

Thus, when |w| < |z|, we have

N
X f (w)wn−1
f (w) f (w)wN
− = + .
w−z zn z N (z − w)
n=1

Integrate the above along the curve C1 , and exchange the order of summation and integra-
tion to obtain
N Z
f (w) wN
Z Z
f (w) X
n−1

−n
− dw = f (w)w z + N
dw.
C1 w − z n=1 C1 C1 z − w z

The summation is the partial sum of the principal part in the Laurent series. We next show
that remainder term approach zeros as N approaches infinity. This is another standard
application of the ML inequality. For w ∈ C1 and z outside C1 , we have

|w| = r1 , and |z − w| ≥ |z| − r1 .

210
Furthermore, |f (w)| is bounded by some constant M for w ∈ C1 , because C1 is a compact
set and |f (w)| is a continuous function on C1 . By ML inequality (Theorem 15.4),

f (w) wN r1N
Z
M
dw ≤ 2πr1 .
C1 z − w zN (|z| − r1 ) |z|N

r1 r1N
Because |z| < 1, we have |z|N
→ 0 as N → ∞. This proves the equality in (21.8) with
Z
1
bn = f (w)wn−1 dw.
2πi C1

We have thus proved that, for r1 < |z| < r2 , the complex function f (z) can be expanded
as
∞ Z ∞ Z
1 X f (w) 1 X
f (w)wn−1 dw z −n .

n

dw z +
2πi C2 wn+1 2πi C1
n=0 n=1

By the deformation theorem, we can replace the contour C1 and C2 in the above integrals
by a simple contour C with positive orientation in the annulus r1 < |z| < r2 , going around
the origin once. We can then simplify it to
∞  Z
X 1 f (w) 
f (z) = n+1
dw zn.
n=−∞
2πi C w

Finally, since r1 can be any number larger than R1 , and r2 can be any number less
than R2 (satisfying r1 < r2 ), we have a convergent Laurent series for any z in the annulus
R1 < |z| < R2 .

211
22 Residues
Summary:

ˆ Uniqueness of Laurent series

ˆ Computation of residues

The residue of a function f (z) at a given point is defined as the coefficient of (z − z0 )−1
in the Laurent expansion of f (z) in a punctured disc 0 < |z − z0 | < r. We first show
that it is well defined, but proving that the coefficients of Laurent series expansion are
uniquely determined by the function values. We then derive formula for the computations
of residues.

22.1 Uniqueness of Laurent series coefficients

Suppose that a function f (z) is represented by a Laurent series
∞
X
f (z) = ck (z − z0 )k
k=−∞

in an annulus R1 < |z − z0 | < R2 , for some nonzero positive constants R1 and R2 . We can
uniquely recover the coefficients from the function values. For notational convenience we
assume that the center z0 is 0.
The idea is to choose a contour C that lie within the annulus with winding number 1
(respect to the center of the circle). We first show that the integral
I
f (z) dz
C

is independent of the choice of C. Suppose there are two closed curves C1 and C2 in the
annulus, each of them revolves around the center once counter-clockwise (Fig. 34). The
integral of f (z) over the two curves must be the same. We can see this by dividing them
into several parts, by drawing line segments from the center of the circles. In each part, we
connect a portion of C1 and a portion by C2 by two line segments. The resulting closed
curve is included inside a convex set (highlighted in as a shaded aea in Fig. 35), in which
the function f (z) is holomorphic. Therefore, by Theorem 16.4, the integral of f (z) over
every closed curve inside this domain is zero. We note that the closed curve may have

212
 Figure 34: Two closed curves in an annulus

self-intersection, but it is allowed in Theorem 16.4. The sum of the integral over all parts
is equal to the integral over C1 minus the integal over C2 , and the difference is equal to 0.
This proves that we can take any closed curve in the annulus with winding number when
we compute the contour integral.
A non-rigorous proof of uniqueness may proceed a follows: Assume that we can exchange
the order of contour integral and infinite sum. By using the basic integrals in Theorem 14.6,
we obtain 
I 2πi if k = −1,
z k dz =
C 0 otherwise.

213
 Figure 35: Division of the two contours into parts

This gives
I I ∞
X
f (z) dz = ck (z − z0 )k dz
C C k=−∞
∞ I
X
= ck (z − z0 )k dz
k=−∞ C

= 2πic−1 .

Therefore c−1 is equal to (2πi)−1 C f (z) dz. The value of other coefficients cn , for n ̸= −1,
H

can be obtained by shifting the Laurent series so that cn is the coefficient of the term
with power −1. This can be done by multiplying f (z) by z −n−1 and repeating the above
argument,
I I X ∞
−n−1
f (z)z dz = ck (z − z0 )k−n−1 = 2πicn .
C C k=−∞

214
 To make this argument rigorous, we divide the Laurent series into three parts,
∞ ∞
X
−k c−1 X
f (z) = c−k z + + ck z k (22.1)
z
k=2 k=0

Interchanging the order of integral and summation that consists of finitely many terms is
always justified. We only need to assume that the integral of each term exists. Our task
is to show that the integral of the first term and the third term on the right-hand side of
(22.1) are both zero.
We recall that, when we say that the Laurent series converges, it means that the analytic
part and the principal part converge. The third term is a power series ∞ k
P
k=0 ck z , and
by assumption, it is convergent in the annulus R1 < |z| < R2 . Because the region of
convergence is an open disc, the power series indeed converges at all point in the disc
|z| < R2 , and is a holomorphic function of z for |z| < R2 . Because an open disc is convex,
by Theorem 16.4, the power series has a primitive function in the open disc, and hence
the integral of this power series is zero if the contour is a closed curve inside the region of
convergence. Therefore
I X ∞ 
ck z k dz = 0.
C k=0

We now consider the first summation in (22.1). We want to show that it has anti-
derivative in the region |z| > R1 . We first make a substitute w = 1/z. Let g(w) denote the
Taylor series
∞
X
g(w) ≜ c−k wk .
k=2

Since f (z) converges in the annulus R1 < |z| < R2 , we know that the power series g(w)
is convergent in the region R2−1 < |w| < R1−1 . However, if g(w) converges at any point w
in this annulus, it also converges for all point with modulus less than |w| (using the fact
that power series is absolute convergent). We thus see that the power series g(w) indeed
converges in the open disc |w| < R1−1 .
Consider the function h(w) represented by the series
∞
X c−k k−1
h(w) ≜ w .
k−1
k=2

215
This has the same radius of convergence as g(w) by the Hadamard formula. We differentiate
with respect to variable w,
∞
d X
h(w) = c−k wk−2 .
dw
k=2

Here, exchange of derivative and infinite sum is permitted because it is a power series
(Theorem 18.6). We get the relationship between power series h(w) and power series g(w),

d
h(w) = w−2 g(w). (22.2)
dw
We substitute w by 1/z and apply the chain rule of differentiation,

d d  1 ′
h(1/z) = h(w) ·
dz dw w=1/z z
−2
= (1/z) g(1/z) · (−1/z 2 )
= −g(1/z).

This proves that ∞ −k has anti-derivative −h(1/z) in the region |z| > R . Because
P
k=2 c−k z 1
it has anit-derivative, if we integrate it over any closed contour in the region |z| > R1 , the
result is equal to 0.
H
Finally, by Cauchy integral formula (Theorem 17.1), the integral of C c−1 /z dz is the
same as 2πic−1 .
With all these preparation, we now integrate both sides of (22.1),
I ∞
I X I ∞
I X
−k
 c−1 k

f (z) dz = c−k z dz + dz + ck z dz
C C C z C
k=2 k=0

= 0 + 2πic−1 + 0 = 2πic−1 .

We thus obtain I
1
c−1 = f (z) dz, (22.3)
2πi C

The other coefficients in the Laurent series can be obtained by dividing f (z) by z n+1 ,
for some integer n, and integrate the result along the curve C. The coefficient that is
associated with the term with degree −1 in f (z)/z n+1 is cn . Therefore, by repeating the
above argument, we have I
1 f (z)
cn = dz.
2πi C z n+1

216
for all integers n.
In summary, the coefficients in the Laurent series expansion is uniquely determined
by the function value within the annulus of convergence. This means that if we apply two
different methods to derive Laurent series of a given function, the two answers are the same.
In particular, we see that the residue of a function at given point is uniquely determined.

22.2 Calculation of residues

For pole with small order, the residue can be computed efficiently. If z0 is a pole of f (z)
with order m, then
bm b1
f (z) = + ··· + + a0 + a1 (z − z0 ) + · · ·
(z − z0 )m z − z0
(z − z0 )m f (z) = bm + · · · + b1 (z − z0 )m−1 + a0 (z − z0 )m + · · · .

We can extract the coefficient b1 by differentiating it m − 1 times, taking limit as z→ z0 ,

and dividing by (m − 1)!. In particular, for pole with order 1,

Res(f ; z0 ) = lim (z − z0 )f (z),

z→z0

and for pole with order 2,

d
Res(f ; z0 ) = lim (z − z0 )2 f (z).
z→z0 dz
We summarize in the following theorem.

Theorem 22.1. Suppose f (z) has a pole of order m at the point z0 . The residue
can be computed by

1 dm−1
(z − z0 )m f (z) .


Res(f ; z0 ) = lim m−1
(m − 1)! z→z 0 dz

There is a useful formula for computing the residue of function in the form of P (z)/Q(z).

P (z)
Proposition 22.2. If f (z) = Q(z) , where Q(z) has a simple zero at z0 and P (z) is
holomorphic at z0 , then
P (z0 )
Res(f ; z0 ) = .
Q′ (z0 )

217
Proof. Suppose Q(z) has a simple zero at z0 . By definition, it means that the power series
expansion of Q(z) at z0 has the form
Q(z) = a1 (z − z0 ) + a2 (z − z0 )2 + a3 (z − z0 )3 + · · ·
where a1 is a nonzero constant. We factor out the common factor (z − z0 ),
Q(z) = (z − z0 )(a1 + a2 (z − z0 ) + a3 (z − z0 )2 + · · · ).
Denote the power series in the parenthesis by g(z), so that
Q(z) = (z − z0 )g(z).
The function g(z) is holomorphic in a neighborhood of z0 (Theorem 18.6), and has the
property that g(z0 ) = a1 ̸= 0.
The residue of f (z) at z0 equals
P (z)
lim (z − z0 )f (z) = lim (z − z0 )
z→z0 z→z0 (z − z0 )g(z)
P (z)
= lim
z→z0 g(z)
P (z0 )
= .
g(z0 )
By product rule for differentiation,
Q′ (z) = g(z) + (z − z0 )g ′ (z).
We see that g(z0 ) is the same as Q′ (z0 ). We obtain
P (z0 ) P (z0 )
Res(f ; z0 ) = = ′ .
g(z0 ) Q (z0 )

Example 22.1. The function f (z) = (z 2 + 1)/(z 3 − 1) has simple pole at the cube roots of
unity. Let ω = e2πi/3 . We compute the residue of f (z) at 1, ω and ω 2 using the formula in
Prop. 22.2. In the calculations below, we use the relation 1 + ω + ω 2 = 0.
12 + 1 2
Res(f ; 1) = 2
=
3(1) 3
2
ω +1 −ω ω
Res(f ; ω) = 2
= 2
=
3(ω) 3ω 3
2 2
(ω ) + 1 ω+1 −ω 2 ω
Res(f ; ω 2 ) = = = =− .
3(ω 2 )2 3ω 3ω 3

218
Example 22.2. Compute Z
dz
C z(z − 1)(z − 2)
with C being the contour |z| = 1.5 with counter-clockwise orientation.
The contour C contains two poles at z = 0 and z = 1. The residues at these two poles
are
 1  1 1
Res ; 0 = lim z =
z(z − 1)(z − 2) z→0 z(z − 1)(z − 2) 2
 1  1
Res ; 1 = lim (z − 1) = −1.
z(z − 1)(z − 2) z→1 z(z − 1)(z − 2)
Apply residue theorem (Theorem 21.8),
Z
dz 1
= 2πi( − 1) = −πi.
|z|=1.5 z(z − 1)(z − 2) 2

Example 22.3. Repeat Example 22.2 with contour C replaced a circle of radius 3 centered
at the origin (counter-clockwise).
The contour now contains all three poles at 0, 1 and 2. The residue at z = 2 is
 1  1 1
Res ; 2 = lim (z − 2) = .
z(z − 1)(z − 2) z→2 z(z − 1)(z − 2) 2
By residue theorem (Theorem 21.8), the answer is
Z
1 1 1
dz = 2πi − 1 + = 0.
|z|=3 z(z − 1)(z − 2) 2 2

Example 22.4. Evaluate Z

dz
C z(z − 1)2
over the contour C : |z| = 2 with counter-clockwise orientation.
The contour C encloses the simple pole at z = 0 and the double pole at z = 1.
 1  1
Res 2
; 0 = lim z =1
z(z − 1) z→0 z(z − 1)2
 1  dh 2 1 i
Res ; 1 = lim (z − 1) = −1.
z(z − 1)2 z→1 dz z(z − 1)2
By residue theorem, the integral is equal to 2πi(1 + (−1)) = 0.

219
23 Evaluation of real integral
Summary

ˆ Semi-circular contour

ˆ Jordan lemma

23.1 Semi-circular contour

Example 23.1. Derive Z ∞
cos(bx) π
2
dx = b for b > 0.
−∞ 1+x e
We complexify the integral into a contour integral
Z ∞
eibz
2
dz.
−∞ 1 + z

In this integral we integrate along the real axis. The original integral is the real part of this
complex integral. The imaginary part is zero, because sin(x)/(1 + x2 ) is an odd function.
To be more accurate, we emphasize that what we are computing is the principal value
R
eibz
Z
lim dz,
R→∞ −R 1 + z2
RR
so that the imaginary part limR→∞ −R sin(x)/(1 + x2 ) dx is indeed equal to zero. The
problem reduces to showing
Z ∞
eibz π
2
dz = b for b > 0.
−∞ 1 + z e

Let C1 and C2 be the contours in the figure below.

C2

−R C1 R

220
 We compute the complex integral
eibz
Z
dz
C1 +C2 1 + z2
along the closed path formed by C1 and C2 . When R is large enough, the contour includes
the simple pole at z = i. By residue theorem (Theorem 21.8), the complex integral is equal
to
eibz eibz e−b
Z
π
2
dz = 2πi Res( 2
; i) = 2πi = b.
C1 +C2 1 + z 1 + z 2i e
We next show that the integral long the semi-circular path C2 tends to zero as the
radius increases. When z is on C2 , we can write z as Reiθ , for 0 ≤ θ ≤ π. The modulus of
eiz on C2 is upper bounded by 1, because

|eibz | = |eibR(cos θ+i sin θ) | = e−bR sin θ ,

which is less than 1 when 0 ≤ θ ≤ π. Let R be a sufficiently larger number so that

|1 + z 2 | > R2 − 1 for all z on the circle |z| = R. By ML inequality (Theorem 15.4),
eibz
Z
1
2
dz ≤ 2 (πR).
C2 1 + z R −1
Putting all the calculations together, we obtain
eibz eibz eibz
Z Z Z
2
dz − 2
dz = 2
dz
C1 +C2 1 + z C1 1 + z C2 1 + z
Z R
π cos(bx) πR
b
− 2
dx ≤ 2 .
e −R 1 + x R −1
This holds whenever R > 1. Since R/(R2 − 1) has limit zero as R → ∞, we have
Z R
cos(bx) π
lim dx = b .
R→∞ −R 1 + x2 e
We check that when b = 0, the integral is simplified to a standard result in calculus:
Z ∞
1
2
dx = π.
∞ x +1

We mentioned in Example 13.4 that Fourier transform and Laplace transform are im-
portant examples of complex integrals. In this section we illustrate how to evaluate some
real integrals that are related to Fourier transform. The Jordan lemma is a useful tool for
this purpose.

221
 CR

−R R

Figure 36: The semi-circular contour in Jordan lemma.

23.2 Jordan lemma

The Jordan lemma provides a convenient way to check that the integral on the semi-circular
arc tends to zero as the radius tends to infinity.

Lemma 23.1 (Jordan lemma). Consider the semi-circular contour CR in Fig. 36.
Suppose that for all sufficiently large R, f (z) is holomorphic on CR , and we can find
a function MR of R that approaches 0 as R approaches ∞, such that |f (z)| ≤ MR
for all z on CR . Then for any positive real constant a,
Z
lim f (z)eiaz dz = 0.
R→∞ CR

Proof. The idea of proof is to represent the circular path by z = Reiθ , for θ ∈ [0, π].
Substituting z by Reiθ in eiaz , we obtain

eiaR(cos θ+i sin θ) = e−aR sin θ+iaR cos θ .

The magnitude is e−aR sin θ . Hence it is of interest to give a good estimate to the integral
Z π
e−aR sin θ dθ.
0

We prove the following claim. For any positive real number a and radius R, we have
the inequality Z π
π
e−aR sin θ dθ < . (23.1)
0 aR

222
 Figure 37: Proof of Jordan inequality

For θ in the range 0 ≤ θ ≤ π/2, the value sin θ is larger than or equal to 2θ/π. We can see
this by drawing a line segment from the origin to the point (π/2, 1), and see that the graph
of sin θ is below the graph of sin θ in this range. (See Fig. 37) We thus get an upper bound
Z π/2 Z π/2 h −π iπ/2 π π
e−aR sin θ dθ ≤ e−aR(2θ/π) dθ = e−2bθ/π = (1 − e−aR ) < .
0 0 2aR 0 2aR 2aR

By using the symmetry of the graph of sine function, the same argument apply to the
second part of the interval from π/2 to π,
Z π
π
e−aR sin θ dθ < .
π/2 2aR

This proves (23.1) for any positive real numbers a and R.

We apply this inequality to derive an upper bound
Z Z π
iaz
f (z)e dz ≤ f (Reiθ ) e−aR sin θ · |Rieiθ | dθ
CR 0
Z π
≤ RMR e−aR sin θ dθ
0
π
= MR .
a
Since a is a constant and MR approaches 0 as R → ∞, the upper bound approaches 0 as
R approaches ∞.

223
 Jordan lemma is useful in evaluating integral of type
Z ∞ Z ∞
P (x) P (x)
cos(x) dx or sin(x) dx
−∞ Q(x) −∞ Q(x)

where P (x) and Q(x) are polynomials and deg Q − deg P ≥ 1.

Example 23.2. Evalaute the integral
Z ∞
cos x
dx,
0 (x2 + a2 )(x2 + b2 )

where a and b are positive constants with a > b.

Because the function to be integrated is an even function, we can transform the problem
to the evaluation of Z R
cos x
2 + a2 )(x2 + b2 )
dx,
−R (x
and take limit as R approaching infinity. This integral above is the same as the contour
integral
eiz
Z
2 2 2 2
dz
LR (z + a )(z + b )

on the line segment LR from −R to R. Let CR denote the semi-circular path in Fig. 36
from R to −R along the circle |z| = R on the upper-half plane. By residue theorem
(Theorem 21.8), the integral

eiz eiz
Z Z
+
LR (z + a )(z 2 + b2 )
2 2
CR (z 2 + a2 )(z 2 + b2 )

is equal to 2πi time the sum of residue of the integrand inside the semi-circle bounded by
LR and CR .
The integrand has two simple poles in the upper half plane at z = ai and z = bi. The
residue at the pole ai equals

eiz eiz e−a

Res( ; ai) = = .
(z 2 + a2 )(z 2 + b2 ) (z + ai)(z 2 + b2 ) z=ai 2ai(b2 − a2 )

The other residue at bi is

eiz eiz e−b
Res( ; bi) = = .
(z 2 + a2 )(z 2 + b2 ) (z + bi)(z 2 + a2 ) z=bi 2bi(a2 − b2 )

224
 On the other hand, because
1 1
=
(z 2 + a2 )(z 2 + b2 ) (|z|2 + a2 )(|z|2 + b2 )

approaches 0 as |z| → ∞, by applying Jordan lemma, we can see that the integral on CR
approaches 0 as R approaches infinity. Therefore

eiz  e−a e−b 

lim dz = 2π +
R→∞ (z 2 + a2 )(z 2 + b2 ) 2ai(b2 − a2 ) 2bi(a2 − b2 )
π  e−b e−a 
= − .
2(a2 − b2 ) b a
The final answer is
Z ∞  e−b e−a 
cos x π
dx = −
0 (x2 + a2 )(x2 + b2 ) 4(a2 − b2 ) b a
Example 23.3. Show that
Z ∞
cos x π −√3/2 1 π

4 2
dx = √ e sin +
0 x +x +1 3 2 6
We first note that the integrand is an even function. It is sufficient to show that
Z ∞
cos x π −√3/2 1 π

4 2
dx = 2 √ e sin + .
−∞ x + x + 1 3 2 6
cos x
Since x4 +x2 +1
is an odd function. We can show this by evaluating the principal value
∞
eiz
Z
dz.
−∞ z4 + z2 + 1

The roots of z 4 + z 2 + 1 are sixth roots of unity, because

(z 2 − 1)(z 4 + z 2 + 1) = z 6 − 1.

There are two roots that lie on the upper-half plane, namely
√ √
iπ/6 1 3i iπ/3 1 3i
α≜e = + , and β ≜ e =− + .
2 2 2 2
We let
eiz
f (z) =
z4 + z2 + 1

225
 Let LR denote the line segment on the real axis from −R to R, and CR be the semi-circle
in the upper half plane from R to −R, centered at the origin. By residue theorem, we know
that for all R > 1,
Z Z
f (z) dz + f (z) dz = 2πi(Res(f, α) + Res(f, β)). (23.2)
LR CR

The first integral is equal to the real integral

Z R
cos x
dx.
−R x4 + x2 + 1

By Jordan lemma, the second integral tends to 0 as R tends to ∞.

Using Theorem 22.2, we calculate the two residues below:
√
eiα ei/2− 3/2
Res(f ; α) = 3
= √
4α + 2α −3 + 3i
√
eiα e−i/2− 3/2
Res(f ; β) = = √
4β 3 + 2β 3 + 3i
The sum of residues at α and β equal
√
√
− 3/2
h ei/2 e−i/2 i √
− 3/2 3(e
i/2 − e−i/2 ) + 3i(ei/2 + e−i/2 )
e √ + √ =e
−3 + 3i 3 + 3i 3i2 − 9
√
√
− 3/2 6i sin(1/2) + 2 3i cos(1/2)
=e
−12
−
√
3/2
√
ie  3 1 
=− √ sin(1/2) + cos(1/2)
3 2 2
√
ie− 3/2
=− √ sin(1/2 + π/6).
3
Substituting it back to (23.2), and taking limit as R approaching infinity, we get
√
∞
2πe− 3/2
Z
cos x
dx = √ sin(1/2 + π/6).
−∞ x4 + x2 + 1 3

226
24 Indented contour and keyhole contour
Summary

ˆ Indented contour

ˆ Keyhole contour

24.1 Indented contour

In some examples the contour contains one or more poles of the integrand. We can approx-
imate the contour integral by adding some dents on the contour.
A common contour is shown in Fig. 38. The indented contour is contained in a star-
shaped domain. For example, we can consider the whole complex plane with the negative
imaginary axis removed.
C \ {x + iy : y ≤ 0, x = 0}.
This is a star-shaped domain. We may take the point i as the center of the star shape. If a
complex function has a pole at the origin, and is holomophic everywhere else, by invoking
the Cauchy theorem for star-shaped region (Theorem 16.6), we can see that any closed
contour within this domain has zero integral. So in particular, the integral of this function
over the indented circular path is zero.
Example 24.1. In this example we derive the Dirichlet integral
Z ∞
sin x π
dx = .
0 x 2
The function sin(x)/x has a removable singularity at x = 0, as we can see from the
power series expansion:
1
(x − x3 /3! + x5 /5! − · · · ) = 1 − x2 /3! + x4 /5! − · · ·
x
The function to be integrated is a continuous function.
We want to replace sin x by eiz in the integral, so that sin x is the imaginary part of eix
when x is real number. We note that the function eiz /z has a simple pole at the origin.
Since sinx x is an even function and cosx x is odd, it is sufficient to prove
 Z −ϵ Z R  eiz
lim + dz = πi.
R→∞ −R ϵ z
ϵ→0

227
 CR

Cϵ

L1 0 L2

Figure 38: Indented semi-circular path. The outer semi-circle has radius R and the inner
semi-circle has radius ϵ.

iz
Note that we avoid the pole of ez at the origin by not including the interval [−ϵ, ϵ]. We
consider the indented contour in Fig. 38.
We have
eiz
Z
dz = 0.
L1 +L2 +CR +Cϵ z
iz
By Jordan lemma (Lemma 23.1), the integral of ez along CR decreases to 0 as R → ∞.
R∞
The integrals on the real axis approaches −∞ eiz /z dz as R → ∞ and ϵ → 0. The problem
reduces to proving
eiz
Z
lim dz = −πi.
ϵ→0 Cϵ z

The integrand can be represented by Laurent series

eiz 1 z iz 2
= +i− − + ···
z z 2 6
2
For small enough ϵ > 0, the analytic part i − z2 − iz6 + · · · is bounded (because it converges
and is continuous at z = 0). (Another way to see this is by observing (eiz − 1)/z has a
removable singularity at z = 0.)
By ML inequality (Theorem 15.4),
z iz 2
Z
i− − + · · · dz → 0
Cϵ 2 6
as ϵ → 0. The integral of 1/z on Cϵ is equal to
Z 0
iϵeiθ
Z
1
dz = iθ
dθ = −πi.
Cϵ z π ϵe

228
This proves that
eiz eiz z iz 2
Z Z Z Z
1
dz + dz = − dz − i− − + · · · dz.
L1 +L2 z Cϵ z CR z Cϵ 2 6
Take absolute value of both sides and take limit as R → ∞ and ϵ → 0,
eiz eiz z iz 2
Z Z Z
dz − πi ≤ dz + i− − + · · · dz .
L1 +L2 z CR z Cϵ 2 6
As the two terms on the right-hand side tends to zero as R → ∞ and ϵ → 0, we obtain
eiz
Z
lim dz = πi.
ϵ→0 L +L z
1
R→∞2

This completes the derivation of the Dirichlet integral.

We extract a useful result that is used in the previous example.

Theorem 24.1. Suppose f (z) has a simple pole at z0 with residue

Res(f ; z0 ) = lim (z − z0 )f (z).

z→z0

Let Cr denote a part of a circle with radius r and positive orientation, and with
angle θ1 to θ2 , relative to the center z0 . That is, it is parameterized by z0 + reiθ for
θ1 ≤ θ ≤ θ2 . Then
Z
lim f (z) dz = i(θ2 − θ1 ) Res(f ; z0 ).
r→0 Cr

Proof. See the notes of Tutorial 10.

Example 24.2. In this example we compute the real integral

Z ∞ 1/3
x
dx
0 x2 + 1
using complex analysis.
√
The complex cube root function 3 z is a multi-valued function. We put the branch cut
√
a the negative imaginary axis in the computation of 3 z. More precisely, given a complex
number that does not lie on the negative imaginary axis, we write it in polar form as

z = reiθ

229
 √ √
for r > 0 and −π/2 < θ < 3π/2, we define the branch of 3 z by 3 reiθ/3 .
√
With the branch of 3 z so defined, we integrate the function
√3
z
f (z) = 2
z +1
on the indented contour in Fig. 38. Note that we avoid the branch point at the origin. By
the residue theorem (Theorem 21.8), the integral is equal to the residue at the pole at i
multiplied by 2πi. We calculate the pole at i by the formula in Prop. 22.2.
√ √
3
z 3
z eiπ/6
Res( 2 ; i) = = .
z +1 2z z=i 2i
By decomposing the indented contour into four parts, we obtain
Z √3
Z √ 3
Z √ 3
Z √ 3
z dz z dz z dz z dz
2
+ 2
+ 2
+ 2
= πeiπ/6 .
CR z + 1 L1 z + 1 Cϵ z + 1 L2 z + 1

The integral of f (z) on L1 equals

Z √ 3
Z R √
3
zdz x
2
= dx.
L1 z +1 ϵ x2 +1
We represents the line segment L2 by parameter x for x from −R to −ϵ. The integral
on L2 is
Z √ Z −ϵ p Z R √
3
zdz 3
|x|eiπ/3 iπ/3
3
x
2
= 2
dx = e 2
dx.
L2 z + 1 −R x +1 ϵ x +1
The integral on CR approaches 0 as R → 0, because
Z √ √
3
3
zdz R
2
≤ 2
(πR)
CR z + 1 |R − 1|
1+1/3
by ML inequality. When R > 1, the ratio RR2 −1 has limit 0 as R → ∞.
Lastly, we consider the integral over Cϵ for 0 < ϵ < 1,
Z √ 3
√
3
zdz ϵ
2
≤ (πϵ).
Cϵ z + 1 1 − ϵ2
1+1/3
The ratio ϵ1−ϵ2 tends to 0 as ϵ tends to 0.
By taking limits as ϵ → 0 and R → ∞, we obtain
Z ∞ √ 3
x
(1 + eiπ/3 ) 2
= πeiπ/6 .
0 x + 1

230
Divide both side by (1 + eiπ/3 ),
∞ √
eiπ/6
Z 3
x
= π
0 x2 + 1 1 + eiπ/3
π
= iπ/6
e + e−iπ/6
π π
= =√ .
2 cos(π/6) 3

24.2 Keyhole contour

A keyhole contour has shape as shown in Fig. 39. We can consider the keyhole contour
when the integrand is a multi-value function. We can pick a branch of the function so that
the branch cut lie outside the keyhole contour. The keyhole contour consists of four parts.
The outer and inner circular part has radius R and ϵ, respectively. The line segments L1
and L2 lie above and below the positive real axis. Let 2δ denote the distance between L1
and L2 . The methodology is to first evaluate the contour on this contour, and then take
limit as ϵ → 0, δ → 0, and R → ∞.
The region
C \ {x + iy : y = 0, x ≥ 0}

is a star-shaped region. We can thus apply the Cauchy theorem and the residue theorem
to the keyhole contour.
Example 24.3. Suppose a is a constant between 0 and 1. We can prove that
Z ∞ a−1
x π
dx =
0 1+x sin(aπ)

by converting it to a complex integral on the keyhole contour.

Consider the complex function

z a−1
f (z) = .
1+z
The numerator is a multi-valued function. We take the branch with branch cut located on
the positive real axis. For complex number z that is not on the positive real axis, we write
z = reiθ , where r > 0 and 0 < θ < 2π, and calculate z a−1 by

z a−1 = (reiθ )a−1 = ra−1 eiθ(a−1) . (24.1)

231
 CR

L1
Cϵ
L2

Figure 39: Keyhole contour

Because there is a simple pole inside the keyhole contour at z = −1, the integral of f (z)
is equal to
Z Z Z Z 
+ + + f (z) dz = 2πi Res(f (z); −1).
CR L2 Cϵ L1

The residue at z = −1 is

Res(f (z); −1) = lim (1 + z)f (z)

z→−1

= lim z a−1 .
z→−1

Using the branch of z a−1 in (24.1), we obtain

Res(f (z); −1) = eiπ(a−1) .

The integral of f (z) on the horizontal line L1 has limit

Z Z ∞ a−1
x
lim f (z) dz = dx
R→∞ L1 0 1+x
ϵ→0
δ→0

232
Because L2 is on the other side of the branch cut, the variable of integration x is multiplied
by a factor of e2πi ,
∞
xa−1 e2πi(a−1)
Z Z
lim f (z) dz = − dx
R→∞ L2 0 1 + xe2πi
ϵ→0
δ→0
∞
xa−1
Z
2π(a−1)i
= −e dx.
0 1+x

Using similar argument as in Example 24.2, we can show that the integral of f (z) on CR
and Cϵ approaches 0.
Putting all the data together, we have
Z ∞ a−1 Z ∞ a−1
x x
dx − e2π(a−1)i dx = 2πieiπ(a−1)
0 1 + x 0 1 + x
Z ∞ a−1
x 2πieiπ(a−1)
dx =
0 1+x 1 − e2π(a−1)i
2πi
= −iπ(a−1)
e − eiπ(a−1)
π
=−
sin((a − 1)π)
π
= .
sin(aπ)

This completes the proof in this example.

Remark. The integral formula in the previous example in fact is valid if a is replaced by a
complex number with real part between 0 and 1, i.e., if

0 < Re(a) < 1.

This condition is required to show that the integrals over the two circular paths approach 0.
We can use the keyhole contour to evaluate real integral of the form
Z ∞
P (x)
dx,
0 Q(x)

where P (x) and Q(x) are polynomials with deg Q ≥ deg P + 2 and Q(x) ̸= 0 for x ≥ 0. The
assumption that Q(x) ̸= 0 for x ≥ 0 ensures that there is no division by 0 in the integral.
We demonstrate the procedure using the following example.

233
Example 24.4. Evaluate Z ∞
1
dx.
0 x3 +1
Note that the integrand is no an even function. We cannot evaluate it through the integral
R∞ 3
−∞ 1/(x + 1) dx.
The trick to multiply the function to be integrated by a log function, and consider the
complex integral Z
log z
3
dz,
C z +1
over the keyhole contour C as shown in Fig. 39. For the complex log function we take the
nonnegative real axis as the branch cut, i.e., for complex number in polar form reiθ , with
0 < θ < 2π, the log function is evaluated as

log r + iθ for 0 < θ < 2π.

The contour C consists of four parts. The outer circle CR has radius R and positive
orientation. The inner circle Cϵ has radius ϵ and negative orientation. The distance between
L1 and L2 is 2δ. When R → ∞, ϵ → 0 and ϵ → 0, the integrals along L1 and L2 have limits
Z Z ∞
log z log x
3+1
dz → 3+1
dx
z x
ZL1 0
Z ∞
log z log x + 2πi
3
dz → − dx.
L2 z + 1 0 x3 + 1

The integral over Cϵ has modulus upper bounded by

Z
log z
3
dz ≤ 2πϵMϵ max | log z|
Cϵ z + 1 |z|=ϵ

where Mϵ denotes the maximum of |1/(z 3 +1)| on the circle |z| = ϵ. Since it is assumed that
Q(z) is defined at z = 0, Mϵ can be upper bounded by another constant independent of ϵ.
In this example Mϵ is approach 1 as ϵ approaches 0, and hence we can say that Mϵ < 2 for
all sufficiently small ϵ. The modulus of log(z) is no more than the modulus of log(ϵ) + i2π.
Hence, as ϵ → 0, the modulus of the integral of Cϵ is upper bounded by a constant times
ϵ| log ϵ + 2πi|, which decreases to zero as ϵ → 0.
For complex number z on CR , the modulus | log(z)/(z 3 + 1)| is upper bounded by a
constant times | logRR+2πi|
3 −1 . The integral over CR approaches 0 as R approaches infinity.

234
 Therefore, Z Z ∞
log(z) 1
lim dz = −2πi dx.
ϵ→0 z3 + 1 x3 + 1
δ→0 L1 +L2 +CR +Cϵ 0
R→∞

We can re-write the above equation as

Z ∞ Z
1 1 log(z)
3
dx = − lim dz.
0 x +1 ϵ→0 2πi L +L +C +Cϵ z 3 + 1
1 2
δ→0 R
R→∞

The polynomial z 3 +1 has three roots, namely, −1, eπi/3 and e5πi/3 . By residue theorem
(Theorem 21.8), we can compute the integral by
Z ∞
1 h log(z)
log(z) πi/3
log(z) 5πi/3
i
dx = − Res ; −1 + Res ; e + Res ;e .
0 x3 + 1 z3 + 1 z3 + 1 z3 + 1

Since the pole of log(z)/(z 3 +1) are all simple roots, we can use the formula in Prop. 22.2.
Evaluate the residues at −1, eπi/3 and e5πi/3 by

log(z)
log(z) 1
Res ; −1 = = πi
z3 + 1 3z 2 −1 3
log(z) πi/3
log(z) πi e−2πi/3
Res 3 ;e = =
z +1 3z 2 eπi/3 3 3
log(z) 5πi/3
log(z) 5πi e−10π/3
Res 3 ;e = = .
z +1 3z 2 e5πi/3 3 3
Adding the three residues, we get
√ √ √
iπ  −1 − 3i −1 + 3i  2 3
1+ +5 =− π,
3 6 6 9
and √
Z ∞
1 2 3
dx = π.
0 x3 + 1 9
Example 24.5. We give another derivation of the integral in the previous example, using a
different contour shown in Fig 40. The function f (z) has a simple pole at eπi/3 when the
radius R is larger than 1. By Residue Theorem, for all sufficiently larger R, we have
Z Z Z
f (z) dz + f (z) dz + f (z) dz = 2πi Res(f (z), eπi/3 )
L1 L2 CR

235
 We compute the residue of f (z) at the sixth root of unity eπi/3 by using the formula in
Proposition 22.2,
1 1
Res(f (z), eπi/3 ) = 2 = − eπi/3 .
3x z=eπi/3 3
The integral of f (z) along the line segment L1 is
Z Z R
1
f (z) dz = dx.
L1 0 x3

On the other hand, the integral of f (z) from 0 to Re2πi/3 , along the opposite of L2 , is
Z Z R Z R
1 2πi/3 1
f (z) dz = e dt = e2πi/3 dt.
−L2 0 1 + (e2πi/3 t)3 0 1 + t3

The integral on CR approaches 0 as R → ∞, by using ML inequality.

Therefore, by taking limit as R → ∞, we obtain
Z ∞
2πi/3 1 1
(1 − e ) 3
dx = 2πi (−eπi/3 )
0 1+x 3

The answer is equal to

∞
2πieπ/3
Z
1
dx = −
0 1 + x3 3(1 − e2πi/3 )
2i
=π
3(eπ/3 − e−πi/3 )
π
=
3 sin(π/3)
√
2π 2 3π
= √ = .
3 3 9

24.3 Appendix: Differentiation and residue at the point of infinity

Definition 24.2. Given a complex function f (z), we say that it is continuous at ∞

if the function g(w) ≜ f (1/w) is continuous at w = 0. Likewise, the function f (z) is
said to be complex differentiable at ∞ if g(w) = f (1/w) is complex differentiable at
w = 0, and is holomorphic at ∞ if g(w) is holomorphic at w = 0.

236
 Figure 40: A contour with the shape of a fan.

Example 24.6. Consider the function f (z) = (z + 1)/(z − 1). It is differentiable at ∞. The
function g(w) = f (1/w) is
1
+1 1+w
g(w) = w1 = .
w − 1 1−w
As w → 0, we have g(w) → 1. The derivative of g(w) is
2
g ′ (w) =
(1 − w)2

At the point w = 0, the derivative of g(w) is g ′ (0) = 2.

To understand the behavior of a function f (z) at the point at infinity, we make a change
of variable w = 1/z. The variable w = 1/z is called the local parameter /uniformizer at ∞.

Definition 24.3. Given a complex function f (z), make a change of variable and
define a new function g(w) = f (1/w). We say that the function f (z) is analytic at
z = ∞ if g(w) is analytic at w = 0. The point at infinity is said to be a removable
singularity (resp. pole, or essential singularity) if g(w) has a removable regularity
(resp. pole, or essential singularity) at w = 0.

Example 24.7. The function f (z) = z + 8 has a simple pole at z = ∞, because g(w) =
f (1/w) = w1 + 8 has a simple pole at w = 0.

237
Example 24.8. The function f (z) = (z + 1)/z 3 has a doube zero at z = ∞, because
g(w) = f (1/w) = w2 + w3 has a double zero at w = 0.
Example 24.9. The function f (z) = ez has an essential singularity at z = ∞, because
1 1
g(w) = exp(1/w) = 1 + + + ···
w 2w2
has an essential singularity at w = 0.

The residue at the point at infinity is a little bit more complicated. We take the
definition by contour integral as the definition of residue at the point at infinity.

Definition 24.4. Suppose f (z) has finitely many singular points in the complex
plane, so that f (z) is holomorphic in the domain |z| > R for some R. The residue
at ∞ of f (z) is defined as
I
1
Res(f ; ∞) ≜ f (z) dz
2πi C ′

where C ′ is a circle containing all singular points in the interior, with clockwise
orientation.

The assumption that f (z) has finitely many singular points is the same as assuming
that the point at infinity is an isolated singular point.
By making a change of variable w = 1/z, dw = −1/z 2 dz, we get
−1 1
 −1 1  
Z Z
1 1
f (z) dz = f dw = Res f ;0
2πi C ′ 2πi C w2 w w2 w

where C is the image of C ′ under the transformation w = 1/z. In the w-plane, the function
−1
f w1 has a isolated singularity at w = 0. This proves the following theorem.


w2

Theorem 24.5. Suppose f has finitely many singular points and C is a contour with
counter-clockwise orientation, containing all singular points in the interior. Then
I
1 1 1


f (z) dz = Res 2 f ;0 .
2πi C z z

We can compare with (22.2), where we also have a factor of 1/w2 . The explanation for
the appearance of this factor is that the notion residue is actually defined for differential

238
form f (z) d(z − z0 ), where z − z0 is the local parameter at z0 . In the finite complex plane,
the differential d(z − z0 ) is the same for all z0 . So, for simplicity we usually say “residue of
a function f (z) at z0 ”. However, for the point at infinity, the local parameter is 1/z, and
the relationship between d(1/z) and dz is given by

d(1/z) = −z −2 dz.

So, the differential f (w)dw with local parameter w = 1/z is converted to a usual differential
by
dw
f (w) dw = f (1/z) dz = f (1/z)(−z −2 ) dz.
dz
Example 24.10. We can use residue at infinity to do Example 22.3. To find the residue of
1
f (z) =
z(z − 1)(z − 2)

at the point at infinity, we make a substitute w = 1/z and calculate

1 1 w
f (1/w)(1/w2 ) = 1 1

1

2
= .
w w −1 w −2 w (1 − w)(1 − 2w)

This function is holomorphic at w = 0. Therefore, the residue of f (z) at the point at

R
infinity is 0, and hence the integral C f (z) dz is zero. This conforms with the answer in
Example 22.3.
Example 24.11. Evaluate I
4z + 1
dz.
|z|=2 z(z − 1)
There are two simple poles at z = 0 and z = 1. Both of them are inside the contour. We
compute the answer by three methods.
Method 1. Using the residue at infinity, we can calculate the complex integral by
I
4z + 1  1 4(1/w) + 1 
dz = 2πi Res ; 0
|z|=2 z(z − 1) w2 (1/w)((1/w) − 1)
 4+w 
= 2πi Res ;0
w(1 − w)
= 2πi · 4 = 8πi.

239
 Method 2. By the residue theorem (Theorem 21.8), The integral is equal to 2πi times
the sum of the residues of the integrand at z = 0 and z = 1. Let f (z) = (4z + 1)/(z(z − 1)).
The residues of f (z) at z = 0 and z = 1 are, respectively,
4z + 1
Res(f ; 0) = lim zf (z) = lim = −1
z→0 z→0 z − 1
4z + 1
Res(f ; 1) = lim (z − 1)f (z) = lim = 5.
z→1 z→1 z
The integral is thus equal to 2πi(5 − 1) = 8πi.
Method 3. Expand the function f (z) using partial fraction expansion. Write f (z) as

4z + 1 A B
= +
z(z − 1) z z−1

for some constants A and B. The constants A and B are solutions to

A+B =4
−A = 1.

So, we have A = −1 and B = 5, and

I I I
1 1
f (z) dz = − dz + 5 dz = (−1)(2πi) + 5(2πi) = 8πi.
|z|=2 |z|=2 z |z|=2 z − 1

Example 24.12. Evaluating a contour integral from the perspective from the point of infinity
can be useful when most of the singular points are inside the circle. Consider the example
of contour integral I
1
dz
C (z − 1)(z − 2)(z − 3) · · · (z − 10)
over the contour C : |z| = 9. The integrand has ten poles. Nine of them lie inside C and
one of them outside C. A direct application of the Residue Theorem requires the evaluation
of the residues at the nine singular point inside the contour.
An alternate method is to use the residue at the point at infinity. However, we cannot
apply Theorem 24.5, but we can adopt the proof idea. Make a change of variable z = 1/w,
and rewrite the integral as
I
1
−1 −1 −1 −1
(−w−2 ) dw
C′ (w − 1)(w − 2)(w − 3) · · · (w − 10)

240
over a contour C ′ : |w| = 1/9 with clockwise orientation. The integral can be simplified as

−w8
I
dw.
C ′ (1 − w)(1 − 2w)(1 − 3w) · · · (1 − 10w)

There are two poles inside the contour C ′ . Let

w8
g(w) = .
(1 − w)(1 − 2w)(1 − 3w) · · · (1 − 10w)

The residue of g(w) at w = 0 and w = 0.1 are

Res(g; 0) = 0
w8
Res(g; 0.1) =
(1 − w)(1 − 2w)(1 − 3w) · · · (1 − 9w) w=0.1
0.18 10
= =
(1 − 0.1)(1 − 0.2) · · · (1 − 0.9) 9!

respectively. The answer is equal to 20πi/9!

241
25 Argument principle
Summary:

ˆ Winding number

ˆ Argument principle

ˆ Rouché theorem

25.1 Analytic definition of winding number

The technical part of winding number is contained in the following theorem. This is about
a closed but not necessarily simple smooth curve.

Theorem 25.1. If C is a closed piece-wise smooth curve not passing through a point
z0 , then I
1 1
dz
2πi C z − z0
is an integer.

Proof. Represent the curve by

γ : [0, 1] → C \ {z0 }.

Write γ(t) − z0 in polar form

γ(t) − z0 = r(t)eiθ(t) ,

for t between 0 and 1.

We denote the distance between γ(t) and z0 by r(t), and the angle with respect to the
given point z0 by θ. Since it is assumed that γ(t) does not pass through z0 , we have r(t) > 0
for all t.
Differentiate γ(t) with respect to t to get

γ ′ (t) = [r′ (t) + ir(t)θ′ (t)]eiθ(t) .

242
Using the definition of complex integral, we obtain

[r′ (t) + ir(t)θ′ (t)]eiθ(t)

1
I Z
1 1 1
dz = dt
2πi γ z − z0 2πi 0 r(t)eiθ(t)
Z 1 ′ Z 1
1 r (t) 1
= dt + θ′ (t) dt
2πi 0 r(t) 2π 0
1 1
= [log(r(1)) − log(r(0))] + [θ(1) − θ(0)].
2πi 2π
Since the curve is closed, we have r(0) = r(1), and θ(1) − θ(0) = 2πk for some integer k.
The first term above is thus equal to zero, and the second term is an integer, which is the
number of times the curve γ goes around the point z0 .

In view of the previous theorem, we make the following definition.

Definition 25.2. The winding number of a closed curve C around a point z0 is

defined as I
1 1
n(C; z0 ) ≜ dz.
2πi C z − z0
Other notation for the winding number includes: w(C, z0 ), ind(C, z0 ). When the
curve is parameterized by γ(t), we also write it as n(γ; z0 ).

If γ is not closed, we can use the integral

Z
1 1
dz
2πi γ z − z0

to compute the change of angle relative to the point z0 . (cf. Section 11.4).

25.2 Argument principle

1
R f ′ (z)
The argument principle is about the integral 2πi C f (z) dz when C is simple closed curve.
This integral is always equal to an integer, because it can be interpreted as the winding
number of some curve around the origin.

243
 Theorem 25.3. Suppose C is a simple closed curve parameterized by γ(t), for 0 ≤
t ≤ 1, and f (z) is a complex function that does not vanish on the curve C. Then

f ′ (z)
I
1
dz = n(f (γ(t)); 0).
2πi C f (z)

It the same as the number of times the curve f (γ(t)) goes around the origin (in the
counter-clockwise direction).

The non-vanishing of the function f (z) on C ensures that the integral is well defined.
Hence, the curve f (γ(t)) does not pass through the origin. We note that if we set the
function f (z) to f (z) = z − z0 , it reduces to Theorem 25.1.

Proof. Consider the curve C ′ that is parameterized by f (γ(t)), for t in [0, 1], and let g(t) =
f (γ(t)) for 0 ≤ t ≤ 1. The winding number of C ′ around 0 is
I
′ 1 1
n(C ; 0) = dz
2πi C ′ z
Z 1 ′
1 g (t)
= dt
2πi 0 g(t)
Z 1 ′
1 f (γ(t))γ ′ (t)
= dt
2πi 0 f (γ(t))
f ′ (z)
I
1
= dz.
2πi C f (z)

The previous proof shows that (2πi)−1 C f ′ /f dz is always an integer. In fact, it is

R

counting the number of zeros minus the number of poles of f (z) (counted with multiplicity)
inside C.
Theorem 25.4 (Argument principle). Suppose C is the boundary of a simply con-
nected region and f is a nonzero meromorphic function defined in the interior of C.
Then
f ′ (z)
I
1
dz
2πi C f (z)
is equal to the number of zeros minus the number of pole of f (counted with multi-
plicity) inside C.

244
Proof. Suppose z1 , z2 , . . . , zn are the zeros or poles of f in the interior of C. There could
be only finitely many such points. Otherwise, either the zeros will have a limit point, or the
poles will have a limit point, which contradicts that f is a nonzero function (Theorem 19.4).
For each k = 1, 2, . . . , n, we write f (z) as

f (z) = amk (z − zk )mk + amk +1 (z − zk )mk +1 + amk +2 (z − zk )mk +2 + · · ·

The integer mk is the smallest exponent such that the coefficient is nonzero. It is positive
if zk is a zero, and is negative if zk is a pole. (Note that we use the assumption that there
is no essential singularity here.) (mk cannot be zero because zk is either a pole or zero.)
By construction, the coefficient amk is nonzero.
The derivative is

f ′ (z) = mk amk (z − zk )mk −1 + (mk + 1)amk +1 (z − zk )mk + (mk + 2)amk +2 (z − zk )mk +1 + · · · .

The ratio f ′ /f can be written as

mk amk (z − zk )mk −1 h 1 + c1 (z − zk ) + c2 (z − zk )2 + · · · i
amk (z − zk )mk 1 + d1 (z − zk ) + d2 (z − zk )2 + · · ·

where cj and dj are some complex numbers. The fraction inside the square bracket is a
holomorphic function in a neighborhood of zk and the value at zk is 1. Therefore, we can
represent it as a power series with constant term 1. As a result, for each k = 1, 2, . . . , n, we
can write f ′ /f as
f ′ (z) mk
= + an analytic function of z.
f (z) z − zk
By consider the residues of f ′ (z)/f (z) at z1 , z2 , . . . , zn together, we obtain

f ′ (z)
I
1
dz = m1 + m2 + · · · + mn .
2πi C f (z)

For k = 1, 2, . . . , n, the integer mk is either positive or negative. It is positive (resp.

negative) when zk is a zero (resp. pole) of f (z), and is equal to the order of zero (resp. pole)
of zk . The sum m1 + m2 + · · · + mn is thus the number of zeros minus the number of poles
in the interior of C, counted with multiplicity.

We can combine Theorem 25.3 and Theorem 25.4. Suppose the contour C and complex
functionf (z) satisfy the conditions in Theorem 25.4. The change of argument of f (z), after

245
traveling the contour C once, is equal to 2π times the difference between the number of zeros
and the number of poles inside the contour C. This is the explanation why Theorem 25.4 is
called the argument principle. In some textbooks, such as [BrownChurchill], the argument
principle is formulated as
1
Z −P = (∆C arg f (z)),
2π
where Z denotes the number of zeros of f (z) inside C, P denotes the number of poles of
f (z) inside C, and ∆C arg f (z) is the change of the argument of f (z) as we travel along the
contour C. (cf. Equation (11.4).) Using the formulation in this notes, the connection is

f ′ (z)
I
1 1
n(f ◦ γ; 0) = (∆C arg f (z)) = dz = Z − P.
2π 2πi C f (z)

Example 25.1. Consider the function f (z) = (z − 1)(z − i)(z − 3), and a circle C centered at
the origin with radius 2. Suppose that the orientation is counter-clockwise. The function is
holomorphic. The number of zeros inside the circle C can be obtained using the argument
principle by

f′ (z − i)(z − 3) + (z − 1)(z − 3) + (z − 1)(z − i)

Z Z
1 1
dz = dz
2πi C f 2πi C (z − 1)(z − i)(z − 3)
Z Z Z
1 h 1 1 1 i
= dz + dz + dz
2πi C z − 1 C z−i C z−3
1
= [2πi + 2πi + 0] = 2.
2πi

25.3 Rouché theorem

From the fundamental theorem of algebra we know that a polynomial of degree d has d
roots, counted with multiplicities. Rouché theorem provides a method to estimate the
locations of the roots. Rouché theorem applies to holomorphic functions in general, not
just to polynomials.

246
 Theorem 25.5 (Rouché theorem). Suppose C is a simple closed curve with posi-
tive orientation. If f and g are functions that are holomorphic in a neighborhood
containing C and its interior, and

|f (z)| > |g(z)|

for all z ∈ C, then the number of zeros of f + g inside C is the same as the number
of zeros of f inside C. (A zero of order m is counted as m zeros.)

We note that the assumption |f (z)| > |g(z)| implies (i) f (z) ̸= 0 for all z ∈ C, and (ii)
f (z) + g(z) ̸= 0 for all z ∈ C.

Proof. Parameterize the curve C by γ(t), for t ∈ [0, 1]. The main idea of proof is that the
curve 1 + g(γ(t))/f (γ(t)), for t ∈ [0, 1] lies strictly inside the circle |z − 1| = 1 with radius
1 and center z = 1. The winding number of the curve (1 + fg ) ◦ γ (t) around the origin is

thus zero. By Theorem. 25.3, this is saying that

Z 1 + g
′
1 f
dz = 0. (25.1)
2πi γ 1 + fg

Figure 41 shows a closed curve in the circle |z − 1| = 1. No matter what the shape is, the
winding number around the origin is 0.
By the argument principle (Theorem 25.4), the number of zeros of f + g inside C can
be computed by
f ′ + g′
Z
1
dz,
2πi C f + g
while the number of zeros of f inside C can be computed by

f′
Z
1
dz.
2πi C f

247
 Figure 41: A closed curve inside the circle |z − 1| = 1.

(1+g/f )′ f ′ +g ′ f′
We need to find a relationship among 1+g/f , f +g and f . Indeed, one can verify that
g
′
1+ f f′ g ′ f − gf ′ f ′
+ = +
1 + fg f f (f + g) f
g ′ f − gf ′ + f ′ f + f ′ g
=
f (f + g)
′
f +g ′
= .
f +g
Integrating both sides and divide by 2πi,
Z 1 + g
′
f′ f ′ + g′
Z Z
1 f 1 1
g dz + dz = dz.
2πi C 1 + f 2πi C f 2πi C f + g

We have shown in (25.1)) that the first integral is zero. Therefore

f′ f ′ + g′
Z Z
1 1
dz = dz.
2πi C f 2πi C f + g
This proves that the number of zeros of f + g inside C is the same as the number of zeros
of f inside C (counted with multiplicity).

248
Example 25.2. Using Rouché theorem, we can show that the polynomial z 10 + 3z 3 + 1 has
exactly 3 complex roots inside the unit circle. For |z| = 1, we check that

|3z 3 | = 3, but |z 10 + 1| < 2.

Apply Rouché theorem with f (z) = 3z 3 and g(z) = z 10 +1. The number roots of z 10 +3z 3 +1
inside the unit circle is the same as the number of roots of 3z 3 inside the unit circle. Since
3z 3 has a triple root at z = 0, z 10 + 3z 3 + 1 has three roots inside the unit circle.
Besides root counting, Rouché also has other applications. We give two examples of
different sorts below.
Example 25.3. Evaluate the integral
6z 5 + 3z 2 − 8z
Z
dz
|z|=1 z 6 + z 3 − 4z 2
over the unit circle, with counter-clockwise orientation.
We note that the numerator of the integrand is the derivative of the denominator. By
the argument principle (Theorem 25.4), the integral is equal to 2πi times the number of
zeros of z 6 + z 3 − 4z 2 inside the unit circle. By Rouché theorem, z 6 + z 3 − 4z 2 and −4z 2
has the same number of zeros inside the unit circle. The answer is thus 2πi(2) = 4πi.
Example 25.4. Consider a polynomial of degree n with leading coefficient equal to 1,

h(z) = z n + c1 z n−1 + c2 z n−2 + · · · + cn .

Show that there is some point z on the unit circle such that |h(z)| ≥ 1.
Suppose on the contrary that |h(z)| < 1 for all z on the unit circle. Apply the Rouché
theorem with f (z) = z n and g(z) = −h(z). We can check that the condition |f (z)| > |g(z)|
is satisfied on the unit circle, i.e.,

|f (z)| = |z n | = 1 > | − h(z)| = |g(z)| for all z with |z| = 1.

By Rouché theorem, the function f and f + g have the same number of zeros inside the
unit circle. On one hand, f (z) = z n has exactly n zeros, namely, n repeated roots at z = 0,
inside the unit circle. However

f (z) + g(z) = c1 z n−1 + c2 z n−2 + · · · + cn

has at most n − 1 zeros. The polynomial f + g cannot have n zeros inside the unit circle.
This contradiction shows that there must be some point z with |z| = 1 such that |h(z)| ≥ 1.

249
 We give a proof of the fundamental theorem of algebra as an application of Rouché
theorem.

Theorem 25.6. A polynomial over C of degree d ≥ 1 has d roots in C (counted with

multiplicity).

Proof. Consider a polynomial in the form

p(z) = ad z d + ad−1 z d−1 + + · · · a1 z + z0

with ad ̸= 0. The idea of proof is to compare the polynomial with a simpler polynomial
ad z d . This is a nonzero polynomial as ad is nonzero.
Let f (z) be the polynomial consisting of a single term ad z d , and let

g(z) ≜ ad−1 z d−1 + + · · · a1 z + z0

be the difference of polynomial p(z) and f (z).

We choose a real number R that is strictly larger than

|ad−1 | + |ad−2 | + · · · + |a0 |

.
|ad |

On the circle |z| = R, we have

|f (z)| = |ad |Rd .

The magnitude of g(z) on the circle |z| = R is strictly less than the magnitude of f (z),
because, when |z| = R,

g(z) ≤ |ad−1 |Rd−1 + |ad−2 Rd−2 + · · · + |a0 |

≤ Rd−1 (ad−1 + ad−2 + · · · + |a0 |)
< |ad |Rd = |f (z)|.

We can thus apply Rouché theorem to conclude that p(z) and ad z d has the same number
of roots. Because ad z d has d roots, counted with multiplicity, so does p(z).

250
 Figure 42: Proof of open mapping theorem

25.4 Appendix: Open mapping theorem

We give a proof of the open mapping theorem as an application of Rouché theorem. In
topology, a continuous function is defined as a function such that the inverse image of any
open set is open. The open mapping theorem in complex analysis is a statement in the
other direction.

Theorem 25.7 (Open mapping theorem). The image of an open set under a non-
constant holomorphic map is open.

Proof. Let f (z) be a nonconstant holomorphic function, and let S be an open subset inside
the domain of f (z). We want to show that the image of S under f is an open set.
Let z0 be a point in S. The function f (z) − f (z0 ) has a zero at z0 . Since f (z) is
nonconstant, the zeros of the function f (z)−f (z0 ) are isolated. We can choose a sufficiently
small radius r so that the open disc D(z0 , r) belongs to S, and f (z) − f (z0 ) is equal to 0
only at z = z0 for all |z − z0 | ≤ r (See Fig. 42).
Let g(z) denote the function f (z) − f (z0 ). Draw a circle C : |z − z0 | = r centered at z0
with radius r. Since C is compact and |g(z)| is continuous, we can take the minimum

m ≜ min |g(z)| = min |f (z) − f (z0 )|

z∈C z∈C

of |g(z)| over the circle C, and the minimum value m is positive.

251
 We claim that the open disc |w − f (z0 )| < m/2 is contained in the image of f (z). That
is, we want to show that any point in the open disc D(f (z0 ), m/2) has a pre-image in S
under the function f .
Let w be any complex number satisfying |w − f (z0 )| < m/2. For any z on the circle C,
we have

|f (z) − w| = |f (z) − f (z0 ) − (w − f (z0 ))|

≥ |f (z) − f (z0 )| − |w − f (z0 )|
> m − m/2 = m/2.

Therefore, |f (z) − w| is strictly larger than |w − f (z0 )| for all z on the circle |z − z0 | = r. By
Rouché theorem, f (z) − z0 and f (z) − f (z0 ) − (w − f (z0 )) = f (z) − w has the same number
of zeros for z in the open disc D(z0 , r). But by our choice of r, f (z) − f (z0 ) has exactly one
zero in this open disc. Therefore, there exists exactly one point z in the open disk D(z0 , r)
such that f (z) = w. This point z is in S, because D(z0 , r) is in S. This proves that every
point in the open disc D(f (z0 ), m/2) is in the image of S.
We thus show that for every point z0 in S, we can find a small radius rz0 , such that
D(f (z0 ), rz0 ) is contained inside the image of S. The image of S can thus be written as a
union
[
D(f (z0 ), rz0 ).
z0 ∈S

Since arbitrary union of open set is open, we prove that S is an open set.

The open mapping theorem has the same strength as the maximal modulus principle.
In fact, we can prove the maximal modulus principle from the open mapping theorem.
Suppose z0 is a maximum point of a non-constant function f (z) such that z0 lies in the
interior of the domain. From the proof of Theorem 25.7, the image f (z0 ) of z0 is contained
in a disc D(f (z0 ), rz0 ) that is completely inside the range of f . The complex numbers
f (z0 ) + ϵ(±1 ± i) are in the range of f for a small but nonzero real number ϵ. One of them
will have modulus larger than f (z0 ). This contradicts the assumption that maximum is
attained at z = z0 .

252
26 Gamma function and Riemann zeta function
Summary:

ˆ Local uniform convergence of sequence of functions

ˆ Analytic continuation by power series

ˆ Gamma function and Riemann zeta function

The Gamma function and Riemann’s zeta functions are originally defined as a function
with real numbers as the input value.
For positive real number x, the Gamma function Γ(x) is defined by an integral
Z ∞
Γ(x) ≜ tx−1 e−t dt. (26.1)
0

It is easy to verify that Γ(1) = 1, and Γ(n + 1) = nΓ(n) for positive integer n. Hence, for
√
positive integer n, Γ(n) = (n − 1)!. With more effort, one can calculate that Γ(1/2) = π.
A graph of the Gamma function is shown in Fig. 43.
The Riemann’s zeta function is defined for real number x that is strictly larger than 1,
∞
X 1
ζ(x) ≜ . (26.2)
nx
n=1

The convergence can be verified by the p-test. When x = 1, the series is the harmonic series
and hence is divergent. Special values of ζ function include ζ(2) = π 2 /6 and ζ(4) = π 4 /90.
A plot of the Riemann zeta function for real input x is shown in Fig. 44.
The main objective of this lecture is to show that the domains of Gamma function and
zeta function can be extended to the complex plane.

26.1 Morera’s theorem

The basic Cauchy theorem says that given a function f (z) that is holomorphic in a convex
or star-shaped domain, then the integral of f (z) over any closed curved in D is zero. This
statement remains valid if the domain is simply connected. However, if the domain D is
multiply connected, the conclusion of Cauchy theorem may fail.
Morera theorem provide a sufficient condition under which a function f (z) is holomor-
phic. In the next theorem, the domain D is connected but may be multiply connected.

253
Figure 43: Plot of the gamma function with real input.

Figure 44: Plot of the zeta function with real input.

254
 Theorem 26.1 (Morera theorem). Consider a continuous complex function f in a
domain D. If I
f (z) dz = 0
C
for all closed curves in D, then f is holomorphic in D.

Proof. The theorem is proved by exhibiting an anti-derivative of f . Choose and fix a point
z0 in D. For each other point z ∈ D, we can find a path from z0 to z, because the domain
R
D is assumed to be connected. Since C f dz is zero for all closed path C, the integral from
z0 to z is independent of path. We can thus define a function F by
Z z
F (z) ≜ f (w) dw,
z0

with the integral taken over any smooth curve from z0 to z. The function F (z) so defined
is differentiable and derivative F ′ (z) is equal to f (z) for all z ∈ D. (See the proof of
Theorem 16.2) Therefore F (z) is holomorphic in D.
In general, the derivative of a holomorphic function is differentiable (see Theorem 18.10),
i.e., F ′ (z) is complex differentiable at every point z ∈ D. Since F ′ (z) = f (z), we conclude
that f (z) is complex differentiable at every point z ∈ D.

We note that the condition in Morera theorem is sufficient but not necessary. It is
possible that a complex function f (z) is holomorphic in a region but fails to have zero
integral over all closed curve. The complex log function defined on the punctured complex
plane C \ {0} is such an example.
When we have more information about the shape of the domain D, we have a stronger
result that only checks the contours with triangular shape.

Theorem 26.2. Suppose f is a complex function that is continuous on a convex

R
set D. Then f is holomorphic in D if and only if T f (z) dz = 0 for all triangular
contours T .

Proof. The forward direction in the theorem is the Cauchy-Goursat theorem for triangle
(Theorem 15.6).

255
 To prove the backward direction, we choose an arbitrary base point z0 in the open
disc D. Define a function F (z) by the integral
Z
f (w) dw.
z0 →z
R
Here, the notation z0 →z f (w) dw stands for the integral of f on the line segment from z0
to z. The line segment lies within the domain D because it is assumed that D is convex.
By the assumption that the integral of f (z) over any triangle is zero, we have
Z
F (z2 ) = F (z1 ) + f (w) dw
z1 →z2

for any z1 and z2 in the domain. We can invoke Theorem 16.2 and conclude that F (z) is
a holomorphic function with derivative f (z). Since F (z) is infinitely differentiable (Theo-
rem 18.10), the derivative of f (z) exists for all z ∈ D.

26.2 Weierstrass’s theorem on convergence of sequence of functions

One way to construct holomorphic function is by taking the limit of a sequence of functions,
such as power series. We list below a few notion of convergence for sequence of functions.

Definition 26.3. Let (fk )∞

k=1 denote a sequence of functions defined in a common
domain D.

ˆ The sequence (fk )∞k=1 converges to f , if for any point z0 in the domain D and
ϵ > 0, there is an integer N (which may depend on z0 ) such that

|fk (z0 ) − f (z0 )| < ϵ for all k ≥ N.

ˆ The sequence (fk )∞ k=1 converges uniformly to f , if for any ϵ > 0, there exists a
sufficiently large integer N (which only depends on ϵ) such that

|fk (z) − f (z)| < ϵ for all k ≥ N and z ∈ D.

ˆ The sequence (fk )∞k=1 converges locally uniformly to f if at any point z0 in the
domain D, there is a neighborhood U containing z0 such that the sequence
(fk )∞
k=1 converges to f uniformly in U .

256
 The logical relationship among these three mode of convergence is:

Converge uniformly ⇒ Converge locally uniformly ⇒ Converge

“Locally uniform convergence” is somewhere between “convergence” and “uniform conver-

gence”.
Remark. We note that locally uniform convergence is equivalent to compact convergence.
This means that a sequence of functions (fk )∞
k=1 converges locally uniformly in a domain D
∞
if and only if (fk )k=1 converges on any compact subset of D. In [Stein], the authors state
the theorems using compact convergence in place of local uniform convergence.
We quote two properties about convergence of functions from real analysis.

Proposition 26.4. Suppose (fk )∞ k=1 is a sequence of continuous functions that con-
verge to f locally uniformly. Then the function f is continuous.

Proposition 26.5 (Weierstrass M test). Suppose (fk (z))∞ k=1 is a sequence of

complex-valued function defined on a domain D. If we can find positive real numbers
Mk , for k ≥ 1, such that
(i) |fk (z)| ≤ Mk for all z ∈ D, and
(ii) ∞
P
k=1 Mk converges,
P∞
then the infinite sum k=1 fk (z) converges absolutely and uniformly on the do-
main D.

The proofs of these two propositions can be found in a real analysis book.
We can exchange the order of contour integration and limit if we have locally uniform
convergence.

Theorem 26.6. Suppose fk (z) is a continuous complex-valued function for k =

1, 2, 3, . . . If (fk (z))∞
k=1 converges locally uniformly to f (z) in domain D, then
Z Z
lim fk (z) dz = f (z) dz
k→∞ C C

for any piece-wise smooth curve C in the domain D.

257
Proof. We first not that the limit function f (z) is continuous, by Prop. 26.4.
The proof relies on the fact that the points in C form a compact set. Suppose the curve
C is represented by γ(t), for t in the interval [a, b]. Because the interval [a, b] is closed and
bounded, the image under a continuous function γ(t) is also closed and bounded.4 Because
we assume that the derivative γ ′ (t) is continuous, except finitely many points, the modulus
Rb
of g ′ (t) is finite for all t ∈ [a, b]. Hence the length of C = a |γ ′ (t)| dt is finite.
Because C is a compact set, we can apply the hypothesis that fk (z) converges to f (z)
for z on the curve C. Let ϵ be an arbitrarily small positive real numbers. There exists
integer N such that |fk (z) − f (z)| < ϵ for all z ∈ C and for all k ≥ N . We now apply the
ML estimate (Theorem 15.4)
Z Z Z
fk (z) dz − f (z) dz = fk (z) − f (z) dz ≤ length of C · ϵ.
C C C
R R
This is the same as saying that C fk (z) dz is converging to C f (z) dz as k → ∞.

As a consequence of Morera theorem, we can prove

Theorem 26.7. If fk (z), for k = 1, 2, . . ., are holomorphic functions defined on a

region R, converging to locally uniformly to f (z), then f (z) is holomorphic in R.

Proof. Since it is assumed that the function fk is holomorphic for any k, the function fk
is continuous in the domain R for any k. The basic property of uniform convergence in
Prop. 26.4 says that the pointwise limit f (z) is a continuous function.
Let z0 denote a point in R, and D be an open disc centered at z0 . We choose the radius
to be sufficiently small so that the open disc lies completely inside region R. We want to
show that f (z) is holomorphic in open disc D.
Consider a triangle T in D. By Theorem 26.6, we have
Z Z
fk (z) dz → f (z) dz.
T T

By appealing to the Cauchy-Goursat theorem for triangle (Theorem 15.6), the integral
R R
T fk (z) dz is equal to zero for each k, because fk is holomorphic. Therefore, T f (z) dz = 0.
Since this holds for any triangular contour T in D, by Theorem 26.2, the limit function f
is holomorphic in D.
4
https://proofwiki.org/wiki/Continuous Image of Compact Space is Compact

258
 The argument in the previous paragraph holds for any point in the region R. This
proves that f is holomorphic.

Remark. The analogous statement of Theorem 26.7 in real analysis is false. It is possible
to construct a sequence of smooth functions that converges pointwise to a function that
is not differentiable at some point. The Weierstrass approximation theorem asserts that
given any continuous real-valued function on a compact set, we can approximate it by a
sequence of polynomials, and the convergence is uniform. If we take the function f (x) =
|x| to start with, then by Weiestrass approximation theorem, we can find a sequence of
polynomial functions, which are all infinitely smooth, that converge to a function that is
not differentiable at x = 0.
The next theorem is due to Weierstrass.

Theorem 26.8 (Weierstrass). Suppose (fk )∞ k=1 is a sequence of holomorphic func-

tions that converges to a limit function f locally uniformly. Then f is holomorphic
and the derivatives fk′ converges to f ′ as k → ∞.

Proof. Let C be a simple closed curve in D. It encloses a compact set in the domain. By
Heine-Borel theorem and locally uniform convergence, the functions fk converge uniformly
on the area enclosed by C. By Theorem 26.7, the limit function f is holomorphic in D.
It remains to prove that fk′ converges to f ′ pointwise as k → ∞.
Let z0 be a point in D. Select a radius r such that the open disc |z − z0 | < r is contained
in D. Let C be the boundary |z − z0 | = r with counter-clockwise orientation. We use the
Cauchy integral formula for the first derivative,
Z
′ 1 fk (w)
fk (z) = dw,
2πi C (w − z)2

which holds for integers k = 1, 2, 3, . . . From the assumption that fk (w)’s converge locally
uniformly to f (w) on C, we have fk (w)/(w − z)2 converges to f (w)/(w − z)2 on C as well.
Therefore Z Z
1 fk (w) 1 f (w)
2
dw → dw as k → ∞.
2πi C (w − z) 2πi C (w − z)2
This proves that fk′ (z) → f ′ (z) as k → ∞.

259
 Theorem 26.9. We can extend the Riemann zeta function, which is originally de-
fined for real x > 1, to a holomorphic function in the half plane Re(s) > 1, by
∞
X 1
ζ(s) ≜ . (26.3)
ns
n=1

Proof. In the followings we write the real and imaginary parts of complex number s as σ
and t, respectively, so that s = σ + it. Recall from Definition 6.8 that, for complex s, the
meaning of ns is
ns ≜ e(ln n)s .

The modulus of ns is equal to

|ns | = |e(ln n)(σ+it) | = |e(ln n)σ | = nσ .

We choose a real number a that is strictly larger than 1, and consider the half plane
Re(s) > a. For complex number s in this region, we have the bound
1 1 1
= σ ≤ a
ns n n
for all integer n ≥ 1.
Because it is known that the summation ∞ −a converges for a > 1, the series in
P
n=1 n
(26.3) converges absolutely by Weierstrass M test. More precisely, this means for each
complex number s in the half plane Re(s) > a, the partial sum
m
X 1
ns
n=1

is convergent as m → ∞, and the convergent is uniform (because absolute convergence

implies uniform convergence). Since we can take the constant a to be any real number
larger than 1, we can now apply Theorem 26.7 to conclude that the infinite summation in
(26.3) is a holomorphic function in the half plane Re(a) > 1.

It requires more effort to extend the domain of the Gamma function. Because the
integral in the definition of real gamma function in (26.1) is an imporper integral.

260
 Theorem 26.10. We can extend the domain of the real Gamma function to the
right-half plane Re(s) > 0. The extended function is given by the integral
Z ∞
Γ(s) ≜ xs−1 e−x dx
0

and is holomorphic in the domain Re(s) > 0.

Proof. The proof follows [Stein, Chapter 6, Prop. 1.1]. Let ϵ and M denote two real numbers
such that ϵ < M . We first show that the integral in the definition of Γ(s) is absolutely
convergent for s in the vertical strip δ < Re(s) < M .
We continue to use the customary notation in the literature and denote the real part of
s by σ. For complex number s with real part σ, we have

|xs−1 e−x | = xσ−1 e−x

for all positive real numbers x. We treat the improper integral in the theorem as the limit
Z ∞ Z k
xs−1 e−x dx = lim xs−1 e−x dx.
0 k→0 1/k

We introduce the notation Z k

fk (s) ≜ xs−1 e−x dx.
1/k

We first show that for each complex number s with real part σ between δ and M ,
the functions fk (s) converge uniformly as k → ∞. Indeed, we can show that it converges
absolutely. We divide the integral into two parts.
Z 1 Z k
xs−1 e−x dx, and xs−1 e−x dx.
1/k 1

In the first integral, if we take the absolute value of the integrand, we can upper bound it

261
by
Z 1 Z 1
s−1 −x
|x e | dx ≤ xσ−1 dx
1/k 1/k
h xσ i 1
≤
σ 1/k
1
1 − k −σ


=
σ
1 1
≤ ≤ .
σ δ
To bound the second integral, we choose a constant Cσ , for δ < σ < M , such that

xσ−1 ≤ Cσ ex/2

for x in the range [1, ∞). We choose the constants Cσ such that Cσ ≤ Cτ whenever σ ≤ τ .
For example, we can take Cσ to be the maximum of xσ−1 e−x/2 over x ∈ [1, ∞).
Then
Z k Z k
σ−1 −x
x e dx ≤ Cσ ex/2 e−x dx
1 1
k
= Cσ − 2x−x/2 1


= 2Cσ (e−1/2 − e−1/k )

≤ 2CM e−1/2 .

Therefore, we obtain an upper bound

Z k
1
|xs−1 e−x | dx ≤ + 2CM e−1/2 ,
1/k δ
which is independent of k and s. This shows that the integral
Z k
xs−1 e−x dx
1/k

converges absolutely, for all complex number s in the vertical strip δ < Re(s) < M , as
k → ∞. This finish the proof of the claim that fk (s) converges uniformly in any vertical
strip δ < Re(s) < M .
Because δ and M can be any real numbers, subject to the constraint δ < M , the
convergence of the sequence (fk (s))∞
k=1 is locally uniform. By Theorem 26.7, the limit
function Γ(s) is holomorphic.

262
 Theorem 26.11. For Re(s) > 0, the complex Gamma function satisfies

Γ(s + 1) = sΓ(s).

Proof. The proof is done by integration by parts for complex integral. Let a and b to be
real numbers with a < b. If we integrate from a to b, we can write
Z b b
Z b
xs e−x dx = − xs e−x a + sxs−1 e−x dx.

a a

We then take limit as b → ∞ and a → 0.

26.3 Analytic continuation by power series and open discs

A method of extending the domain of an analytic function is by power series expansion.
Suppose we have a converging power series with center z0 and radius of convergence r0 . It
defines an analytic function
X∞
f0 (z) = ak (z − z0 )k
k=0

in the open disc D(z0 , r0 ). Let z1 be a point inside this open disc. Expand the function
f0 (z) with center z1 ,
∞
X
f1 (z) = bk (z − z1 )k
k=0

and let the radius of convergence is r1 . If the open disc D(z1 , r1 ) is not completely inside
D(z0 , r0 ) (see Fig. 45), we can extend the domain of f0 (z) to the union of the two open
disc, 
f (z) if z ∈ D(z , r ),
0 0 0
g(z) =
f (z) if z ∈ D(z , r ) \ D(z , r ).
1 1 1 0 0

This definition makes sense because the two functions f0 (z) and f1 (z) have the same value
for z in the intersection. The function with the extended domain is analytic, because it
has power series representation at every point in open discs D(z0 , r0 ) and D(z1 , r1 ). By
the Identity Theorem (Theorem 19.4), this is the only way to extend f0 (z) to D(z0 , r0 ) ∪
D(z1 , r1 ).

263
 Figure 45: Extending the domain of a function by power series.

1
Example 26.1. As a toy example we consider the power series expansion of z−1 centered at
the origin. This defines a function

f0 (z) = −1 − z − z 2 − z 3 − z 4 − . . .

This defines a function with the open disc D(0, 1) as the domain.
We can extend the domain beyond the open disc D(0, 1) by taking one point z1 in
D(0, 1) and consider the power series of f0 (z) at z1 . For instance suppose we take z1 = i/2
and expand the function f0 (z) at z1 . This process is possible by Theorem 18.9. Since
we know that the function has closed-form expression 1/(z − 1), we can leverage on this
knowledge and compute the power series centered at i/2 by
1 1
=
z−1 −1 + i/2 + z − i/2
1 1
= ·
−1 + i/2 1 + z−i/2
−1+i/2
1 h z − i/2 z − i/2
2 z − i/2
3 i
= 1− + − + ···
−1 + i/2 −1 + i/2 −1 + i/2 −1 + i/2
√
This power series converges when z is in the open disc D(i/2, 5/2). We can thus define
another function by this power series
1 h z − i/2 z − i/2
2 z − i/2
3 i
f1 (z) = 1− + − + ···
−1 + i/2 −1 + i/2 −1 + i/2 −1 + i/2

264
 √
for |z − i/2| < 5/2. We can combine f0 (z) and f1 (z) and define a function g(z) on the
√
union of D(0, 1) and D(i/2, 5/2) by

f (z) if z ∈ D(0, 1),
0
g(z) ≜
f (z) if z ∈ D(i/2, √5/2).
1

If z is in the intersection of the two open disc, we can define g(z) by either f0 (z) or f1 (z).
Take another point in the new domain and consider the power series expansion on this
point. If this power series converges in some region outside the old domain, we can further
extend the domain. We can continue this process indefinitely. This will recover the original
function 1/(z − 1) on the complex plane.
We remark that this method will fail in the strange case of natural boundary as in
Example 26.2. It is not possible to analytically continue the function in Example 26.2 to a
larger domain.
Example 26.2. (natural boundary) Consider the lacunary power series
∞
n
X
f (z) = z 2 = z + z 2 + z 4 + z 8 + z 16 + · · · .
n=0

(“Lacunary” means there are a lot gaps in the series.) This series converges absolutely for
|z| < 1 by comparing with a simple geometric series,
∞
X
|z| + |z|2 + |z|4 + |z|8 + |z|16 + · · · ≤ |z|k .
k=1

However, this series diverges at all (2n )-th roots of unity, for all positive integer n. For
example,

f (−1) = −1 + 1 + 1 + 1 + · · · = ∞,
f (i) = i − 1 + 1 + 1 + · · · = ∞,
f (−i) = −i − 1 + 1 + 1 + · · · = ∞,
f (e2πi/8 ) = e2πi/8 + e2πi/4 + e2πi/2 + 1 + 1 + · · · = ∞.

The set of singular points of f (z) on the unit circle is countable and dense, and hence are
not isolated. It is not possible to analytically extend the domain to a larger open set. The
unit circle is a natural boundary of the domain of definition.
The singular points on the unit circle are plotted in Fig. 33.

265
 Figure 46: The singular points of a lacunary series are dense on the unit circle.

Using the recursive formula, Γ(s + 1) = sΓ(s), which also holds for complex number s
in the right-half plane Re s > 0, we can define the Gamma function for α in the vertical
strip −1 < Re(s) ≤ 0, except s = 0, by the same formula,

Γ(s + 1)
Γ(s) ≜ , if − 1 < Re(s) ≤ 0, s ̸= 0.
s
We can further extend the domain to complex numbers in the vertical strip −2 < Re(s) ≤
−1, except s = −1, by the formula

Γ(s + 1) Γ(s + 2)
Γ(s) ≜ = , if − 2 ≤ Re(s) ≤ −1, except s = −1.
s s(s + 1)

Repeating this process, we can define Γ(s) for s in the complex plane except 0, −1, −2, . . .
(See Fig. 47).
Using the method of analytic continuation using power series, in theory, we can extend
the domain to all other complex numbers. A more systematic way is to consider the
auxiliary function
ξ(s) ≜ π −s/2 Γ(s/2)ζ(s)

and establishing the functional equation

ξ(1 − s) = ξ(s).

We refer the reader to other books in analytic number theory for the derivation.

266
Figure 47: Analytic continuation in defining Gamma function for Re(s) ≤ 0.

267
 It turns out that we can extend the domain of the zeta function to the whole complex
plane except the point s = 1. Because of the uniqueness of analytic continuation, the values
of the zeta function on the other part of the complex plane is uniquely defined by the values
on the right-half plane {x + iy : x > 1}. It turns out that ζ(−1) is equal to −1/12, and we
can thus write it as the following sum
1
1 + 2 + 3 + 4 + ··· = − .
12
The value of ζ(z) at −2, −4, −6, etc. are zeros. These are called the trivial zeros of the
Riemann zeta function.
It was Riemann’s vision that the values of the zeta function on the complex plane
contain important information about the distribution of primes. For example, the fact that
s = 1 is a pole is equivalent to the infinitude of primes. The prime number theorem can
be deduced from the property that the zeta function does not vanish on the vertical line
Re s = 1.
The famous Riemann hypothesis is the conjecture about the location of the zeros in the
critical strip 0 < Re(s) < 1.

Conjecture 26.12 (Riemann hypothesis). Let ζ(s) denote the complex function
obtained by analytic continuation of the zeta function defined for positive real numbers
larger than 1. All zeros in the critical strip 0 < Re(s) < 1 lie on the vertical line
Re(s) = 1/2.

We remark that, by the Identity Theorem, the values of ζ(s) in the strip 0 < Re(s) < 1
(and other part of the complex plane), are uniquely determined by ζ(x) for real x > 1.
There is no ambiguity in determining the locations of the zeros of ζ(s).

26.4 Appendix: Schwarz reflection principle

To motivate the idea of reflection principle, we consider a symmetric region D about the
real axis, i.e., D = {z̄ : z ∈ D}, and suppose f is a complex function that is known to be
analytic on D. As D is a connected domain, the domain D must intersect the real axis
in at least one point. Moreover, since any domain is open by assumption, the intersection
with the real axis must contain at least one open interval with positive length (See Fig. 48).
We note that the domain D need not be convex.

268
 Let f (z) be an analytic function on D. By checking with the Cauchy-Riemann equations,
one can show that the function f (z̄) is analytic in the same domain.

Theorem 26.13. If f (z) is analytic in a symmetric domain D along the real axis,
then the function f (z̄) is analytic in D.

Proof. We first note that because of the symmetry assumption, f (z̄) is well-defined, i.e., z̄
is in D whenever z is in D. Suppose f (z) = u(x, y) + iv(x, y). We want to show that

g(z) = f (z̄) = u(x, −y) − iv(x, −y)

is analytic in D as well.
Let p(x, y) and q(x, y) denote the real and imaginary parts of g(z), respectively. We
check that
∂
px (x, y) = u(x, −y) = ux (x, −y)
∂x
∂
qy (x, y) = (−v(x, −y)) = vy (x, −y)
∂y
∂
py (x, y) = u(x, −y) = −uy (x, −y)
∂y
∂
qx (x, y) = (−v(x, −y)) = −vx (x, −y).
∂x
Since u(x, y) and v(x, y) satisfy the Cauchy-Riemann equations, p(x, y) and q(x, y)
also satisfy the Cauchy-Riemann equations. Because f (z) is assumed to be analytic, the
real-valued functions p(x, y) and q(x, y) are continuously differentiable. By the sufficient
condition for complex differentiability (Theorem 7.7), f (z̄) is analytic in D.

Furthermore, suppose that f (z) is real-valued on the intersection of D and the real
axis. For examples, f (x) could be a convergent power series with real coefficients. Then
the difference f (z) − f (z̄) is analytic, and is equal to zero for z on the real axis. By the
identity theorem (Theorem 19.4), f (z)−f (z̄) must be identically equal to zero on the whole
domain. This proves that f (z) = f (z̄) throughout the domain D.
The reflection principle is a kind of converse to this result.

269
 Figure 48: Symmetric domain

Theorem 26.14 (Reflection principle). Suppose D is a nonempty domain symmetric

along the real axis. Let D+ and L denote the intersection of D with the upper half
plane and real axis, respectively. If f is a complex function defined on D+ ∪ L such
that (i) f is analytic in D+ , (ii) continuous in D+ ∪L, and (iii) real-valued on L, then
f (z) has a unique extension to D, and the extended function satisfies f (z) = f (z̄)
for z ∈ D.

Proof. Let D− be the intersection of D and the lower half plane. We define a new function
g(z) on domain D by 
f (z) if z ∈ D+ ∪ L
g(z) =
f (z̄) if z ∈ D− .

By the same proof as in the proof of Prop. 26.13, g(z) is analytic within D− . Moreover,
g(z) is continuous in the symmetric domain D.
We show that g(z) is analytic throughout the domain D using Morera theorem (Theo-
rem 26.2). Suppose T is a triangular contour. If T completely lies inside D+ , or completely
lies inside D− , then the integral T g(z) dz = 0. So, it remains to consider triangle T that
R

overlaps with the real axis.

270
 Figure 49: Triangular contour in a symmetric region.

Denote the vertices of T by points A, B and C, and suppose A is in D+ , and B and C

are in D− (See Fig. 49). Let B ′ and C ′ be two points on line AB and AC, respectively, so
that B ′ C ′ is parallel and is above the real axis. Similarly let B ′′ and C ′′ be two points on AB
and BC so that B ′′ C ′′ is below and parallel to the real axis. Select the points such that the
parallel distance between B ′ C ′ and B ′′ C ′′ is ϵ. The contour integral of g(z) on the triangle
T can be decompose into three parts, namely, AB ′ C ′ , B ′ B ′′ C ′′ C ′ , and B ′′ BCC ′′ . The
two contours that are inside D+ or D− have zero integral. The contour along B ′ B ′′ C ′′ C ′
approaches 0 as ϵ approaches 0 by the continuity of g(z). This shows that the integral of
g(z) along the triangle T is zero.
The same argument applies to other triangles in domain D. Hence, we have
Z
g(z) dz = 0
T

for any triangle T in D. By Morera theorem for triangular contour, we conclude that g(z)
is analytic inside D.

Actually we have established a more general principle in the previous theorem. The
assumption that the line of reflection has zero slope is not crucial. The function can also
be complex-valued on the line of reflection. We single out the key part in the following
theorem.

271
 Theorem 26.15 (symmetry principle). Let D+ and D− be two regions that are
symmetric along the common line L with positive length. We have a holomorphic
function f + defined on D+ , and a holomorphic function f − defined on D− . If f +
and f − can be extended to the common line of intersection and f + (z) = f − (z) for
all z ∈ L, then the function g(z) defined by



 f + (z) if z ∈ D+

g(z) ≜ f − (z) if z ∈ D−


f + (z) or f − (z) if z ∈ L


is holomorphic in the combined region D+ ∪ L ∪ D− .

The proof is the same as in the proof of Theorem 26.14.

26.5 Appendix: Hurwitz theorem

The theorem in this section depicts another special property of holomorphic function. It
is about the zeros of a sequence of functions that locally uniformly, and the zeros of the
limiting function. The proof is based on Rouché theorem.

Theorem 26.16 (Hurwitz theorem). Let (fk )∞ k=1 be a sequence of holomorphic func-
tion defined on a domain D. Suppose that the functions fk (z) do not vanish in D,
and converges locally uniformly to f (z). Then, unless f (z) is identically zero, f (z)
has no zero in D.

Proof. By Theorem 26.8, we know that the pointwise limit f (z) is holomorphic in D.
Suppose f (z) is not identically zero, but there is a point z0 in the domain D such that
f (z0 ) = 0. We will derive a contradiction.
Consider a small circle |z − z0 | = ϵ with radius ϵ so that f (z) ̸= 0 for all z on this circle.
This can be done because the zeros of a nonzero holomorphic function are isolated. Let m
be the minimum value of f (z) on this circle. Because f (z) is not zero on this circle, we
have m > 0. The minimum exists and is positive because f (z) is continuous on the circle
and the circle is a compact set.

272
 Because the sequence of functions fk (z) converges locally uniformly, together with an
application of Heine-Borel theorem, for any ϵ > 0, we can find a sufficiently large integer
N such that |fN (z) − f (z)| < ϵ for all z on this circle. We choose an ϵ that is strictly less
than m.
Given this data, we apply Rouché theorem (Theorem 25.5) and compare the values of
the functions f (z) and fN (z) − f (z) on the circle |z − z0 | = ϵ. Because the modulus of
fN (z) − f (z) is less than the modulus of f (z) for all z on the circle, Rouché theorem says
that f (z) and f (z) + (fN (z) − f (z)) has the same number of zeros inside the open disc
|z − z0 | < ϵ. However, it is assumed that fN (z) does not vanish anywhere in the domain.
This contradicts that z0 is a zero of f (z).

Hurwitz theorem is false for real functions. We can consider a sequence of real functions
fn (x) = x2 +1/n, for n = 1, 2, 3, . . ., for x in the interval [−1, 1]. When n approaches infinity,
this sequence converges uniformly to f (x) = x2 , which as a zero at x = 0.
Using the same type of argument as in the proof of Theorem 26.16, uniform limit of
one-to-one holomorphic functions is also one-to-one.

Theorem 26.17. Let (fk )∞ k=1 be a sequence of complex functions that is holomorphic
and injective in a domain D. If this sequence converges locally uniformly, then the
limit function is either injective in D or a constant function.

Proof. The steps in this proof are in parallel to the proof of Theorem 26.16. We prove by
contradiction and suppose that f (z) is not injective. Let z0 and z1 be two points in the
domain of f (z) such that f (z0 ) = f (z1 ) = w0 . Pick a sufficiently small radius ϵ such that
f (z) − w0 does not vanish on the two circles |z − z0 | = ϵ and |z − z1 | = ϵ. Let m be the
minimum of f (z) over all z on these two circles. We have m > 0.
Because fk converges locally uniformly, we can find a large enough integer N such that
|fN (z) − f (z)| < m on both circles. By Rouché theorem, the two functions f (z) − w0 and
f (z) − w0 + (fN (z) − f (z)) = fn (z) − w0 has the same number of zeros inside the circles
|z − z0 | = ϵ and |z − z1 | = ϵ. Since fN (z) is assumed to be an injective function, if we
subtract w0 from fN (z), the function fN (z) − w0 has at most one zero inside the two circles.
It contradictss that f (z) − w0 has two zeros.

273
 In Theorem 26.17, it is possible that a sequence of injective functions converges uni-
formly to a constant function. For example, the functions fk (z) = z/k, defined for z in a
compact set, converge uniformly to the zero function, which is not injective.

274