Complex analysis 2

Tero Kilpeläinen and Antti Vähäkangas

(AI-assisted English translation by Mikko Salo)

Lecture notes for Spring 2024

Foreword

The following pages contain lecture notes for the Complex Analysis 2 course. These
have been mostly adapted from lecture notes written by Tero Kilpeläinen to fit the
current 30-hour lecture series. The proof of the Riemann mapping theorem is based
on the lecture notes Complex Analysis II by Kari Astala and Eero Saksman.
Anna Tuhola wrote the first version of these notes in LATEX.
The course covers complex integration in very general regions of the complex plane,
power series representations of analytic functions, residue calculus, singularities, and
conformal mappings.
All references in the form [Link] point to section [Link] of the Complex Ana-
lysis 1 (Tero Kilpeläinen, ed. 2015) lecture notes.
 Sisällys

1. Cauchy’s theorem and integral formula — homological versions 1

1.1. Alternative proof of Cauchy’s integral formula . . . . . . . . . . . . . 12

2. Power series representation of analytic functions 14

2.1. On complex series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2. Power series expansion . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.3. Laurent series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

3. Singular points and residue theorem 35

3.1. Singular points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.2. Removable singularities . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.3. Poles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.4. Essential singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.5. Residue theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

4. Mapping properties of analytic functions 48

4.1. Analytic functions are angle-preserving . . . . . . . . . . . . . . . . . 48
4.2. Argument principle and Hurwitz’s theorem . . . . . . . . . . . . . . . 50
4.3. Open mapping theorem for analytic functions . . . . . . . . . . . . . 53
4.4. Introduction to conformal mappings . . . . . . . . . . . . . . . . . . . 55
4.5. Normal families and Montel’s theorem . . . . . . . . . . . . . . . . . 58
4.6. Proof of the Riemann mapping theorem . . . . . . . . . . . . . . . . 60

5. Extended complex plane and Möbius transformations 64

5.1. Extended complex plane . . . . . . . . . . . . . . . . . . . . . . . . . 64
5.2. Möbius transformations . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.3. Möbius transformations as conformal mappings . . . . . . . . . . . . 74

i
 1. Cauchy’s theorem and integral formula —
homological versions

To prove the general (homological) form of Cauchy’s theorem, we need some ad-
ditional concepts and a “reminder” of iterated integrals.

1.1. Definition. A cycle in the complex plane is a finite sequence σ = (γ1 , γ2 , . . . , γp ),

where γk are closed paths for all k = 1, 2, . . . , p.

γ3
γ1
γ2

1.2. Remark. The order of paths γk is irrelevant. Often, we identify, for example,
cycles (γ, β, α, ←
−, ←
γ −, β) and (←
γ − ∗ γ, β ∗ β, ←
γ −, α). We also identify a closed path γ
γ
with the cycle σ = (γ).
If σ = (γ1 , γ2 , . . . , γp ) is a cycle, then

|σ| = |γ1 | ∪ |γ2 | ∪ . . . ∪ |γp |,

and furthermore, σ is a cycle in the set A, if |σ| ⊂ A. Note! The trace |σ| of the
cycle σ is compact, as it is a finite union of compact sets.
If σ = (γ1 , γ2 , . . . , γp ) is a cycle and f : |σ| → C is continuous, then

Z p Z
X
f (z) dz = f (z) dz.
σ k=1 γ
k

In particular, the winding number of the cycle σ around the point z0 ∈

/ |σ| is
Z p
1 dz X
n(σ, z0 ) := = n(γk , z0 ).
2πi z − z0 k=1
σ

1
1.3 . Definition. Let σ be a cycle in an open set G ⊂ C. The cycle σ is null-
homologous in the set G, if

n(σ, z) = 0 for all z ∈ C \ G .

A closed path γ in an open set G is null-homologous in the set G, if the cycle

σ = (γ) is null-homologous in the set G.

α
γ

Two cycles σ0 = (γ1 , γ2 , . . . , γp ) and σ1 = (β1 , β2 , . . . , βq ) in the open set G are

homologous in the set G if the cycle
←
− ← − ←
−
σ = (γ1 , γ2 , . . . , γp , β1 , β2 , . . . , βq )

is null-homologous in the set G.

The paths λ0 and λ1 in the open set G are homologous in the set G if they
←
−
have common starting and ending points and the closed path λ = λ0 ∗ λ1 is null-
homologous in the set G.

1.4. Remark. Cycles σ0 and σ1 in the (open) set G are homologous in the set G if,
and only if,
n(σ0 , z) = n(σ1 , z) for all z ∈ C \ G .
This follows directly from the definition and the following calculation (notations as
above):
p q
X X ←−
n(σ, z) = n(γk , z) + n(βm , z)
k=1 m=1
p q
X X
= n(γk , z) − n(βm , z) = n(σ0 , z) − n(σ1 , z) ,
k=1 m=1

where z ∈ C \ G.

2
1.5 . Remark. All cycles in the disk B ⊂ C are null-homologous in the set B.
Let σ = (γ1 , γ2 , . . . , γp ) be a cycle in the disk B. Then |σ| ⊂ B and thus C \ B is
contained in the unbounded component of C \ |γk | for all k = 1, 2, . . . , p. By Lemma
CA1.5.4, for all z ∈ C \ B, we have

p
X
n(σ, z) = n(γk , z) = 0 .
k=1

We need an auxiliary result on the interchange of order of integration in iterated

(complex) integrals. If R = {z = x + iy : a ≤ x ≤ b, c ≤ y ≤ d} is a closed rectangle
and h : R → C is continuous, then by Fubini’s theorem,

Zd Zb Zb Zd



(1.1) h(t, s)dt ds = h(t, s)ds dt.
c a a c

Furthermore,

1.6. Lemma. Let γ : [a, b] → C and β : [c, d] → C be paths and g : |γ| × |β| → C
continuous.1 Then
Z Z Z Z



g(z, ζ)dz dζ = g(z, ζ)dζ dz.
β γ γ β

Proof: Since g is continuous on the compact set |γ|×|β|, g is uniformly continuous.

Thus, the mappings
Z Z
ζ 7→ g(z, ζ)dz and z 7→ g(z, ζ)dζ
γ β

are continuous (exercise). Hence, the integrals are well-defined.

For simplicity, let’s assume that β and γ are continuously differentiable — the
general case can be obtained by partitioning [a, b] and [c, d] into subintervals and
summing. Now, define the function h as follows,

h(t, s) := g(γ(t), β(s))γ ′ (t)β ′ (s),

1
Remember: |γ| × |β| = {(z, w) ∈ C2 : z ∈ |γ|, w ∈ |β|}.

3
which is continuous on the rectangle R = [a, b] × [c, d]. Therefore, from equation
(1.1), it follows that

Z Z Zd Zb
g(γ(t), β(s))γ ′ (t)dt β ′ (s)ds



g(z, ζ)dz dζ =
β γ c a

Zd Zb Zb Zd
′ ′


= g(γ(t), β(s))γ (t)β (s)dt ds = h(t, s)ds dt
c a a c
Zb Zd Z Z
′ ′



= g(γ(t), β(s))β (s)ds γ (t)dt = g(z, ζ)dζ dz.
a c γ β

1.7. Theorem (Global Cauchy theorem). Let σ be a cycle in an open set G. Then
Z
f (z) dz = 0 for all analytic functions f : G → C
σ

if and only if σ is null-homologous in G.

Proof: First, let’s prove the necessity of the condition. Let z0 ∈ C \ G. Then

1
f (z) =
z − z0

is analytic in G, so
Z Z
1 1 dz
0= f (z) dz = = n(σ, z0 ).
2πi 2πi z − z0
σ σ

Thus, σ is null-homologous in G.
Let’s then tackle the sufficiency of the condition. Let σ be a null-homologous cycle
in the set G. Define
V = {z ∈ C \ |σ| : n(σ, z) = 0}.
Since n(σ, ·) is constant on the components of the set C \ |σ| (details exercise—
remember Lemma CA1.5.4), V is a union of components of the set C \ |σ|. Thus,

4
V is open. Moreover, by Lemma CA1.5.4, V contains an unbounded component of
C \ |σ|. By assumption, C \ G ⊂ V . Hence, K := C \ V ⊂ G is closed and bounded,
i.e., compact. Furthermore, |σ| ⊂ K.
Let 0 < δ < dist(K, ∂G). Then B(z, δ) ⊂ G for all z ∈ K. Partition C into
disjoint coordinate-axis-aligned (closed) squares, whose edges lie on the lines

nδ nδ
x= and y= , n ∈ Z,
2 2

δ
i.e., each square’s side length is 2
and the four corners are at the origin.

The bounded set K intersects only finitely many of these squares. Let them be
Q1 , Q2 , . . . , Qr . From the construction, if zj is the center of square Qj , then Bj :=
B(zj , 2δ ) ⊂ G, since otherwise B(z, δ) ∩ ∂G ̸= ∅ for all z ∈ Qj ∩ K, which is a
contradiction. Furthermore, Qj ⊂ Bj .
Now let f be analytic in the set G.
Claim: Z
f (z) dz = 0 .
σ

Let k = 1, . . . , r and fix z ∈ int(Qk ). Now apply the local Cauchy integral formula
(Theorem CA1.5.5) in the disk Bk . We obtain
Z
1 f (ζ)dζ
f (z) = n(∂Qk , z) f (z) = ,
| {z } 2πi ζ −z
=1 ∂Qk

where the boundary is oriented counterclockwise. If j ̸= k, then from the local

Cauchy theorem (case z ̸∈ Bj ) or the integral formula (case z ∈ Bj ) in the disk Bj ,
it follows that Z
1 f (ζ)dζ
0 = n(∂Qj , z) f (z) = .
| {z } 2πi ζ −z
=0 ∂Qj

5
Thus,

r Z
1 X f (ζ)dζ
(∗) f (z) =
2πi j=1 ζ −z
∂Qj

r
S
for all z ∈ int(Qk ). Equation (∗) holds for all k, thus it holds for all z ∈ int(Qj ).
j=1

z
Qk

λq λ3

λ1 λ2

Let λ now be one of the directed edges forming the boundary ∂Qj . Now either
|λ| ∩ K = ∅ or |λ| ∩ K ̸= ∅.
If |λ|∩K ̸= ∅, then |λ| is also a side of another square Qk . In this case, the integral
←
−
on the right-hand side of equation (∗) traverses the edges λ and λ , which cancel
r
S
each other out when integrated. Thus, for all z ∈ Int Qj ,
j=1

q Z
1 X f (ζ)dζ
(∗∗) f (z) = ,
2πi k=1 ζ −z
λk

where λ1 , . . . , λq are the directed boundary edges of squares Qj whose traces do not
intersect set K. By Lemma CA1.5.6, we obtain that
Z
1 f (ζ)dζ
H(z) =
2πi ζ −z
λk

6
defines an analytic function, specifically a continuous function in C \ |λk |. Thus, the
right-hand side of equation (∗∗) defines a continuous function in the set
q
S
C\ |λk | .
k=1

Note that q
S
K ⊂C\ |λk | .
k=1

Since the left-hand side of equation (∗∗) is continuously analytic in the set G, (∗∗)
also holds on those parts of the edges of squares Qj that do not belong to the traces
of edges λj . In particular, (∗∗) holds for all z ∈ |σ|.
Finally, in the proof, let σ = (γ1 , γ2 , . . . , γp ). Now,
 
Z Z X q Z
(∗∗) 1 f (ζ)dζ 
f (z) dz =  dz
k=1
2πi ζ −z
σ σ λk
 
p q Z Z
1 X X
 f (ζ)dζ  dz
=
2πi l=1 k=1 ζ −z
γl λk

and since the integrand is continuous in the set

S
|γl | × |λk | ,
k,l

by Lemma 1.6, the above is equal to

 
q p Z Z
1 X X
 f (ζ)  dz dζ
2πi k=1 l=1 ζ −z
λk γl
 
q Z p Z
X X 1 dz 
= f (ζ)  dζ
k=1 λ l=1
2πi ζ − z
k γl
 
q Z p Z
X X 1 dz 
=− f (ζ)  dζ
k=1 l=1
2πi z−ζ
λk γl
q
XZ
=− f (ζ)n(σ, ζ) dζ = 0,
k=1 λ
k

since n(σ, ζ) = 0 for all ζ ∈ λk , which follows from the fact that since |λk | ∩ K = ∅,
then |λk | ⊂ V .

7
1.8. Corollary. Let f be analytic in an open set G, and let σ0 and σ1 be homologous
cycles or paths in G. Then
Z Z
f (z) dz = f (z) dz.
σ0 σ1

Proof: If σ0 = (γ1 , . . . , γp ) and σ1 = (β1 , . . . , βq ) are cycles and homologous

in G, then we apply Cauchy’s theorem in G to the null homologous cycle σ =
←
− ←
−
(γ1 , . . . , γp , β1 , . . . , βq ), where
Z Z Z
f (z) dz − f (z) dz = f (z) dz = 0.
σ0 σ1 σ

If σ0 and σ1 are paths, the claim follows similarly by applying Cauchy’s theorem
to the null homologous cycle (σ0 ∗ ←
σ−1 ).

Example. Calculate the integral

z2 + z + 1
Z
dz
z3 + z2
γ

along the path γ depicted in the figure below.

−1 0 1

We decompose the integrand into partial fractions to obtain

z2 + z + 1
Z Z   Z Z
1 1 dz dz
dz = + dz = + .
z3 + z2 z 2 z+1 z2 z+1
γ γ γ γ

8
Since F (z) = −z −1 is a primitive of the function f (z) = z −2 in C \ {0}, the first
integral is easy to compute:
Z i
dz 1
= = 1 + i.
z2 z 1
γ

Since γ is homologous to the line segment β = [1, i] in C \ {−1}, where the function
g(z) = (z + 1)−1 is also analytic, Corollary 1.8 yields
Z
dz
Z
dz √ πi ln 2 πi
= = Log(z + 1)|i1 = ln 2 + − ln 2 = − + .
z+1 z+1 4 2 4
γ β

Here, we utilize the fact that G(z) = Log(z + 1) is a primitive of the function g in
C \ (−∞, −1], which contains the line segment β = [1, i]. Thus, we obtain
Z 2
z +z+1 √ π
3 2
dz = 1 − ln 2 + i(1 + ).
z +z 4
γ

1.9. Theorem (Cauchy’s integral formula). Let f be analytic in the set G, and σ
be a null-homologous cycle in the set G. Then
Z
1 f (ζ)dζ
n(σ, z)f (z) =
2πi ζ −z
σ

for all z ∈ G \ |σ|.

Proof: Compare with the proof of Cauchy’s local version (Theorem CA1.5.5).
Let z ∈ G \ |σ| and define g : G → C,

 f (ζ) − f (z) ,

when ζ ̸= z
g(ζ) = ζ −z
 ′
f (z) , when ζ = z.
Then g is continuous in G and analytic in G \ {z}. Now by Theorem CA1.5.10, it
follows that f is analytic in G. Hence, by Cauchy’s theorem,
f (ζ) − f (z)
Z Z
0 = g(ζ)dζ = dζ
ζ −z
Zσ σ
f (ζ)
= dζ − 2πi n(σ, z)f (z).
ζ −z
σ

9
1.10. Theorem. Let f be analytic and σ be a null-homologous cycle in G. Then
for all k = 1, 2, . . . Z
(k) k! f (ζ)
n(σ, z)f (z) = dζ ,
2πi (ζ − z)k+1
σ

when z ∈ G \ |σ|.

Proof: Omitted (analogous in the proof of Theorem CA1.5.11).

Recall that a domain (or region) is a nonempty connected open set in C.

1.11. Definition. A domain D ⊂ C is simply connected if every closed curve γ (and

thus every cycle) in the set D is null-homologous in D.

1.12. Remark. The assertions of Cauchy’s theorem and integral formula hold for
all cycles in simply connected domains.

1.13. Remark. A bounded domain D is simply connected if and only if C \ D is

connected (exercise).
An unbounded domain D ̸= C is simply connected if and only if all components
of the set C \ D are unbounded (exercise).

1.14. Theorem. Let D ⊂ C be a domain. Every analytic function f : D → C has

a primitive in the domain D, if and only if D is simply connected.

Proof: Let D be a simply connected domain and f : D → C be analytic. Then

from Cauchy’s theorem it follows that
Z
f (z) dz = 0 for all closed paths γ , |γ| ⊂ D .
γ

10
Now by Theorem CA1.4.5 function f has a primitive in the domain D (in other
words, there exists an analytic function F : D → C such that F ′ (z) = f (z) for all
z ∈ D).
Now let γ be a closed path in the set D and z0 ∈ C \ D. Now the function

1
f (z) =
z − z0

is analytic in the domain D, thus it has a primitive there. Hence by Theorem

CA1.3.15 Z
1
0= f (z) dz = n(γ, z0 ).
2πi
γ

The path γ is therefore null-homologous, hence D is simply connected.

1.15. Theorem. Let D be a simply connected domain with 0 ∈ / D. There is an

analytic function g in D such that eg(z) = z for z ∈ D (i.e. there exists a branch of
the logarithm in D).

1
Proof: From Theorem 1.14, it follows that the function z
has a primitive f in the
domain D. Let z0 ∈ D be fixed and set

g(z) = f (z) − f (z0 ) + Log z0

F (z) = ze−g(z)

Now g and F are analytic and g is a primitive of the function 1

z
since g ′ (z) = z1 .
Furthermore,
1
F ′ (z) = e−g(z) (1 − z ) = 0.
z
Thus F is constant and since

z0
F (z0 ) = z0 e−g(z0 ) = z0 e− Log z0 = = 1,
z0

we have F (z) = 1 for all z ∈ D. In other words, z = eg(z) for all z ∈ D, i.e., g is a
branch of logarithm in the domain D.

11
1.1. Alternative proof of Cauchy’s integral formula

There are several different proofs of Cauchy’s theorem and integral formula. Here
we give another one, which proceeds by showing Cauchy’s integral formula using
several of the theorems proved in CA1.

Proof: Let f be analytic in G and let σ be a null-homologous cycle in G. We wish

to prove Cauchy’s integral formula
Z
1 f (w)
n(σ, z)f (z) = dw
2πi w−z
σ

for all z ∈ G \ |σ|. Recalling the definition of the winding number, this is equivalent
to proving that Z
h(z) := g(z, w) dw = 0, z ∈ G \ |σ|,
σ

where g(z, w) is defined for z, w ∈ G by

(
f (w)−f (z)
w−z
, z≠ w,
g(z, w) = ′
f (z), z = w.

The function g(z, w) is continuous in G × G (exercise) and hence h(z) is a well

defined function in G.
Let us prove that h is analytic in G. We first show that h is continuous in G. Let
z ∈ G and let zn → z in G. Then
Z Z
h(zn ) = g(zn , w) dw → g(z, w) dw = h(z).
σ σ

Here we used that g(zn , · ) converges uniformly to g(z, · ) in the compact set |σ|,
which follows from the uniform continuity of g on compact sets (exercise). Thus h
is continuous in G. We now apply Morera’s theorem: if ∆ is a triangle in G, then by
Fubini’s theorem (Lemma 1.6)
   
Z Z Z Z Z
h(z) dz =  g(z, w) dw dz =  g(z, w) dz  dw.
∆ ∆ σ σ ∆

For any fixed w ∈ G, the function z 7→ g(z, w) is analytic in G. (If z ̸= w this follows
from the definition of g, and if z = w this follows from the theorem of analytic

12
continuation to a point, Theorem CA1.5.10.)
R Then the inner integral vanishes by
Cauchy’s theorem for triangles. Thus ∆ h(z) dz = 0 for any triangle ∆ ⊂ G, which
implies that h is analytic in G by Morera’s theorem.
Next we wish to prove that there is an analytic function φ in C with φ|G = h.
To do this, let G1 = {z ∈ C \ |σ| : n(σ, z) = 0}. Motivated by the statement of
Cauchy’s integral formula, we define
Z
f (w)
h1 (z) := dw, z ∈ G1 .
w−z
σ

Then h1 is analytic in G1 by Lemma CA1.5.6. If z ∈ G ∩ G1 , the definition of G1

ensures that h(z) = h1 (z). We may thus define
(
h(z), z ∈ G,
φ(z) =
h1 (z), z ∈ G1 .

Since G1 contains the unbounded component of C \ |σ|, it also contains C \ G. It

follows that φ is an analytic function in C with φ|G = h.
Finally, we have
lim φ(z) = lim h1 (z) = 0.
[z|→∞ [z|→∞

Thus φ is a bounded analytic function in C (it is bounded in |z| ≥ R by the above

limit, and in |z| ≤ R by continuity). Liouville’s theorem implies that φ is a constant
function, and the above limit ensures that φ ≡ 0. This gives that h(z) = 0 for z ∈ G,
which concludes the proof.

13
 2. Power series representation of analytic functions

2.1. On complex series

2.1. Definition. Let zn ∈ C, n = 1, 2, . . .. A complex series

∞
X
zn
n=1

converges (to the number s ∈ C) if there exists a limit

k
X
lim zn (= s ∈ C).
k→∞
n=1

In this case, we denote

∞
X
s= zn
n=1
∞
P
and say that s is the sum of the series zn . In other words, if
n=1

k
X
sk = zn ,
n=1

then ∞
X
the series zn converges,
n=1

if and only if the sequence of partial sums (sk )∞

k=1 converges.
∞
P
If the series zn does not converge, it diverges.
n=1
∞
P
We say that the series zn converges absolutely if the series
n=1

∞
X
|zn | converges.
n=1

14
2.2. Remark (A few exercises).

∞
P ∞
P
- If zn converges absolutely, then zn converges.
n=1 n=1

∞
P ∞
P ∞
P
- If zn and wn converge and c ∈ C, then (czn + wn ) converges to the
n=1 n=1 n=1
∞
P ∞
P
number c zn + wn .
n=1 n=1

∞
P ∞
P ∞
P
- If zn and wn converge absolutely and c ∈ C, then (czn +wn ) converges
n=1 n=1 n=1
∞
P ∞
P
absolutely to the number c zn + wn .
n=1 n=1

∞
P
In the future, we will need other types of indexing besides . These, for example
n=1
∞
P
, are defined in a natural way.
n=−k

∞
z n , where z ∈ C. The partial sum
P
Example. Let’s examine the geometric series
n=0
of the series is
sk = 1 + z + · · · + z k .

Note! In this context and thereafter, we denote 00 = 1.

If z = 1, then sk = k + 1 → ∞ and the series diverges.
If z ̸= 1, then from the formula

1 − z k+1 = (1 − z)(1 + z + · · · + z k )

we get
1 − z k+1 1 z k+1
sk = = − .
1−z 1−z 1−z
If |z| < 1, then the last term
z k+1
→ 0.
1−z
If |z| > 1, then
z k+1
→ ∞.
1−z

15
If |z| = 1, z ̸= 1, then the last term rotates around the circle of radius 1/|1 − z|.
Thus, the limit does not exist when |z| = 1, z ̸= 1.
Therefore ∞
X
zn converges if and only if |z| < 1
n=0

and in this case ∞

X 1
zn = .
n=0
1−z

As an exercise, you can prove that the convergence is absolute.

Sequences of functions. In order to develop the theory of infinite series, we need

some facts regarding the (uniform) convergence of sequences of functions. We will
later study partial sums
k
X
sk (z) = fn (z) ,
n=1

where fn : A → C for each k ∈ N. First we consider general sequences (sk ) of

functions sk : A → C.

2.3. Definition. Let A ⊂ C and sk : A → C for each k ∈ N. A function sequence

sk converges uniformly on the set A (towards the function s : A → C) if for every
ϵ > 0 there exists N ∈ N such that

|sk (z) − s(z)| < ϵ

for all z ∈ A, when k ≥ N . Note! The number ϵ does NOT depend on the point
z ∈ A.

2.4. Remark. If sk : A → C are continuous and sk → s uniformly on the set A,

then s : A → C is continuous.

2.5. Remark. Remember the Cauchy criterion for uniform convergence:

A sequence sk : A → C converges uniformly on the set A (towards the function
s : A → C) if and only if for every ϵ > 0 there exists N = N (ϵ) ∈ N such that

sup |sk (z) − sm (z)| < ϵ

z∈A

when k, m ≥ N .

16
2.6. Definition. It is said that the function sequence sk : A → C converges locally
uniformly on the set A (towards the function s : A → C) if sk → s uniformly on
every compact subset K of the set A.

Example. Let’s consider functions sk (z) = z k , z ∈ B(0, 1). Let K ⊂ B(0, 1) be a

compact set, then K ⊂ B(0, r) for some 0 < r < 1. Assume ϵ > 0 and choose N
such that rN < ϵ. Now

|sk (z) − 0| = |z|k < rk ≤ rN < ϵ ,

for all z ∈ K, when k ≥ N . Thus sk → 0 uniformly on the set K. Since K was

an arbitrary compact subset of B(0, 1), then sk → 0 locally uniformly in the disk
B(0, 1). Note! The function sequence sk does not converge uniformly in the disk
B(0, 1) (exercise).

2.7. Remark. A singleton {z} is compact, so if sk → s locally uniformly, then

sk → s pointwise. In particular, the limit function of a locally uniformly convergent
sequence is unique.

2.8. Lemma. Let sk : G → C be continuous functions, k ∈ N, which converge

locally uniformly in an open set G (towards some function s : G → C). If γ is a
path in the set G, then

Z Z
s(z) dz = lim sk (z) dz .
k→∞
γ γ

Proof: Exercise; remember that the trace |γ| ⊂ G is compact.

Uniform convergence of analytic functions sk : G → C (locally) is so strong that

the order of differentiation and taking limits can be interchanged.

2.9 . Theorem. Assume that the sequence of analytic functions sk : G → C

converges locally uniformly in an open set G towards a function s : G → C. Then s
(n)
is analytic and sk → s(n) locally uniformly in the set G for all n = 0, 1, 2, . . . .

17
Proof: Since sk → s locally uniformly in the open set G, s is continuous in the set
G. From Lemma 2.8 and Cauchy’s theorem, it follows that
Z Z
s(z)dz = lim sk (z)dz = 0
k→∞
∂R ∂R

for all closed rectangles R in the set G. Therefore, by Morera’s theorem CA1.5.9,
s : G → C is analytic.
To prove the local uniform convergence of derivatives, it suffices to show that
s′k → s′ locally uniformly (the claim follows from this inductively). For this purpose,
it suffices to show that s′k → s′ uniformly in the disk B(z0 , r), where B = B(z0 , 2r) ⊂
G. (Why?) Next, we apply Cauchy’s estimate CA1.5.12 to the analytic function sk −s
in the disk B(z0 , 2r). So for all z ∈ B(z0 , r)

supζ∈B |s(ζ) − sk (ζ)|2r

|s′k (z) − s′ (z)| ≤
(2r − |z − z0 |)2
2 k→∞
≤ sup |s(ζ) − sk (ζ)| −−−→ 0,
r ζ∈B
| {z }
independent of z

since sk → s uniformly in the disk B ⊂ G. Since the last upper limit converges to
zero independently of the point z, s′k → s′ uniformly in the disk B(z0 , r).

2.10. Remark. Theorem 2.9 does not hold if we only assume that sk → s pointwise,
even if s is known to be analytic. If the sequence sk is locally uniformly bounded, then
by Cauchy’s estimate (Theorem CA1.5.12), it can be seen that pointwise convergence
implies local uniform convergence (exercise).

Series of functions. Let fn : A → C, n = 1, 2, . . . . We say that the function series

∞
P
fn converges (pointwise) in the set A, if the series
n=1

∞
X
fn (z) converges for every z ∈ A .
n=1

∞
P ∞
P
The series fn converges absolutely in the set A if |fn | converges in the set
n=1 n=1
A.

18
 ∞
P
We say that the function series fn converges uniformly in the set A if the
n=1
sequence of functions

k
X
sk = fn : A → C , k ∈ N,
n=1

uniformly converges in the set A.

∞
P
We say that the function series fn converges locally uniformly in the set A if
n=1
the sequence of functions sk (as above) converges locally uniformly in the set A.

∞
P
2.11. Remark. If the function series fn converges uniformly in the set A, then
n=1
the sequence of functions (fn )n∈N converges uniformly in the set A towards the zero
function. This follows from the uniform convergence criterion, i.e., Remark 2.5.

2.12. Lemma. Let fn : G → C be continuous, for which the series

∞
X
fn = f
n=1

converges locally uniformly in the open set G. If γ is a path in the set G, then

Z ∞ Z
X
f (z) dz = fn (z)dz.
γ n=1 γ

Pk
Proof: Let sk = n=1 fn for each k ∈ N. By assumption, the sequence sk of
continuous functions converges locally uniformly in the set G towards the function
f . Now, by the linearity of the integral and Lemma 2.8, we have

Z Z k Z
X ∞ Z
X
f (z) dz = lim sk (z) dz = lim fn (z) dz = fn (z) dz .
k→∞ k→∞
γ γ n=1 γ n=1 γ

19
2.13. Theorem. Let fn : G → C be analytic functions, such that the series

∞
X
fn
n=1

converges locally uniformly in the open set G. Then

∞
X
f= fn
n=1

∞
P (k)
is analytic in the set G, and for all k = 1, 2, . . . the series of derivatives fn
n=1
converges locally uniformly in the set G towards the derivative of f f (k) ,

∞
X
f (k) (z) = fn(k) (z) when z ∈ G .
n=1

Proof: Follow from Theorem 2.9 by applying linearity of differentiation (exercise).

The following test is often a useful way to verify uniform (and pointwise) conver-
gence of a function series.

2.14. Theorem (Weierstrass M-test). Let fn : A → C be functions, and suppose

that for all n ∈ N, there exists Mn < ∞ such that

|fn (z)| ≤ Mn

∞
P ∞
P
for all z ∈ A. If the series Mn converges, then the function series fn converges
n=1 n=1
uniformly and absolutely on the set A.
 k

P
Proof: Pointwise convergence is an easy exercise. Let ε > 0. Since Mn is
n=1 k
a Cauchy sequence, there exists a number N = N (ε) such that

k
X m
X
Mn − Mn < ε
n=1 n=1

20
when k ≥ m ≥ N . Let
k
X
sk = fn : A → C .
n=1

Now, when k > m ≥ N , for all z ∈ A

k
X
|sk (z) − sm (z)| = fn (z)
n=m+1
k
X k
X
≤ |fn (z)| ≤ Mn
n=m+1 n=m+1
k
X m
X
= Mn − Mn < ε .
n=1 n=1

Thus, the sequence of functions sk satisfies uniform convergence by the Cauchy

criterion (see Remark 2.5).

2.15. Definition. If an ∈ R, n = 1, 2, . . . , then define

lim sup an := lim (sup{an , an+1 , . . . }) = inf (sup{an , an+1 , . . . }) ∈ [−∞, ∞]

n→∞ n→∞ n

and

lim inf an := lim (inf{an , an+1 , . . . }) = sup(inf{an , an+1 , . . . }) ∈ [−∞, ∞].

n→∞ n→∞ n

2.16 . Remark. Instead of the interpretation of complex limits familiar from

Complex Analysis 1 course, which we usually use, the two limits appearing abo-
ve are taken on the extended real line R ∪ {±∞}. Since (for example)

bn = sup{an , an+1 , . . . }

is a decreasing sequence of numbers in the interval (−∞, ∞], it must have a limit (on
the extended real line) and thus lim sup is well-defined. Also lim inf is well-defined.

2.17. Remark. If the limit limn→∞ an exists on the extended real line, then

lim sup an = lim inf an = lim an .

n→∞ n→∞ n→∞

21
 Example. Let

an = i2n , then lim inf an = −1 < 1 = lim sup an .

n→∞ n→∞

2.18. Definition. Let z0 ∈ C. Series of the form

∞
X
(∗) an (z − z0 )n ,
n=0

a0 , a1 , · · · ∈ C, are called power series around point z0 (or Taylor series at point z0 ).
The numbers an are coefficients of the power series (∗). Let

1
ρ= p ,
lim supn→∞ n
|an |

then2 0 ≤ ρ ≤ ∞ is the radius of convergence of series (∗). In this case, the disk
B(z0 , ρ) is called the disk of convergence of series (∗). If ρ = ∞, then B(z0 , ρ) = C.

∞
an (z−z0 )n .
P
2.19. Theorem. Let ρ be the radius of convergence of the power series
n=0
Then the following holds:

∞
• The series an (z − z0 )n diverges for all z such that |z − z0 | > ρ.
P
n=0

∞
• If ρ > 0, then the series an (z − z0 )n converges absolutely and locally uni-
P
n=0
formly in the disk B(z0 , ρ) and thus the function

∞
X
f (z) = an (z − z0 )n
n=0

is analytic in the disk B(z0 , ρ). Moreover,

(2.1) f (n) (z0 ) = n!an .

2 1 1
We define 0 := ∞ and ∞ := 0.

22
2.20. Remark. Theorem 2.19 does not discuss the convergence of the series at the
boundary of the convergence disk B(z0 , ρ). The series may either converge or diverge
there (exercise).

Proof of Theorem 2.19: 1. Let’s consider

p a number z ∈ C such that |z − z0 | =
r > ρ. Now r−1 < ρ−1 = lim supn→∞ n |an |. This means there are infinitely many
indices n, for which p
n
|an | > r−1 or |an | > r−n ,
so
|an (z − z0 )n | = |an ||z − z0 |n > r−n r−n = 1
for infinitely many n. Especially,

|an (z − z0 )n | ̸→ 0 ,

∞
an (z − z0 )n diverges.
P
so the series
n=0
∞
an (z − z0 )n converges
P
2. Let ρ > 0 and 0 < r < ρ. It suffices to show that
n=0
uniformly and absolutely
p in the closed disk B̄r := B(z0 , r). (Why?) Let s ∈ (r, ρ).
Since lim supn→∞ |an | = ρ−1 < s−1 , there exists N ∈ N such that
n

p
n
|an | < s−1 , when n≥N.

Let c = max{1, |a0 |, |a1 |s, . . . , |aN |sN }, then

|an | ≤ cs−n for all n.

In other words,
 r n
|an (z − z0 )n | ≤ c =: Mn for all z ∈ B̄r .
s
∞
r
P
Since Mn converges, as s
< 1, by the Weierstrass M-test (2.14), it follows that
n=0
∞
an (z − z0 )n converges uniformly and absolutely in the disk B̄r .
P
the series
n=0

3. The claim regarding the analyticity of function f and its

Pkderivatives follows
from Theorem 2.13; note that the partial sum functions sk = n=0 an (z − z0 )n are
entire.

23
2.2. Power series expansion

An analytic function is locally given by a power series:

2.21. Theorem (Power series expansion). Let f be analytic in an open set G, and
B = B(z0 , r) ⊂ G. Then f has a power series expansion around the point z0 and f
determines the power series uniquely:
∞
X
f (z) = an (z − z0 )n , z ∈ B,
n=0

where
f (n) (z0 )
an = .
n!
Proof: Existence of power series expansion: Let

f (n) (z0 )
an =
n!
and z ∈ B. Let s be such that |z − z0 | < s < r. If ζ ∈ ∂B(z0 , s), then

z − z0 |z − z0 | s
= < = 1,
ζ − z0 s s

thus by the formula for the sum of a geometric series, we have

∞  n X ∞
f (ζ) f (ζ) 1 f (ζ) X z − z0 f (ζ)(z − z0 )n
= = = .
ζ −z ζ − z0 1 − z−z0
ζ−z0
ζ − z0 n=0 ζ − z0 n=0
(ζ − z0 )n+1

By the Weierstrass M -test, the last series converges (ζ-wise) locally uniformly in the
open set G \ B(z0 , |z − z0 |) ⊃ ∂B(z0 , s), thus if γ(t) = z0 + seit , 0 ≤ t ≤ 2π, then by
the Cauchy integral formula, we get
∞
f (ζ)(z − z0 )n
Z Z X
1 f (ζ) 1
f (z) = dζ = dζ .
2πi ζ −z 2πi n=0
(ζ − z0 )n+1
γ γ

By Lemma 2.12,
∞ ∞
f (ζ)(z − z0 )n
Z Z
1 X X 1 f (ζ) 
f (z) = n+1
dζ = (z − z0 )n dζ ,
2πi n=0 (ζ − z0 ) n=0
2πi (ζ − z0 )n+1
γ γ

24
and since by the Cauchy integral formula (1.10),

f (n) (z0 )
Z
1 f (ζ)
dζ = = an ,
2πi (ζ − z0 )n+1 n!
γ

we obtain
∞
X
f (z) = an (z − z0 )n .
n=0

Uniqueness of power series: Let

∞
X
f (z) = bn (z − z0 )n for all z ∈ B .
n=0

Then the convergence radius of the series is at least as large as r, and from Theorem
2.19, it follows that
f (n) (z0 )
bn = = an ,
n!
which proves the claim.

Example. Let’s find the power series expansion of the function f (z) = ez around
the origin.
Since f (n) (z) = ez for all z, we have f (n) (0) = 1 for all n, hence

∞ ∞
X f (n) (z0 ) z2 X zn
ez = zn = 1 + z + + ··· = for all z ∈ C .
n=0
n! 2! n=0
n!

As an application of power series expansion, we prove the following theorem of

uniqueness:

2.22. Theorem. Let f be analytic in domain D. Then the following conditions are
equivalent:

i) f ≡ 0 in domain D (i.e., f (z) = 0 for all z ∈ D).

25
 ii) The set
N = {z ∈ D : f (z) = 0}
has an accumulation point in D.

iii) There exists a point z0 ∈ D, such that

f (k) (z0 ) = 0 for all k = 0, 1, 2, 3, . . . .

Proof: Clearly, i) ⇒ ii).

Now we show that ii) ⇒ iii): Let z0 ∈ D be an accumulation point of the set N .
Since f is continuous, we have f (z0 ) = 0. If

f (k) (z0 ) ̸= 0 for some k ≥ 1

then let
k0 = min{k : f (k) (z0 ) ̸= 0} ∈ N.
In this case, the power series of f at the point z0 is
∞
X
an (z − z0 )n , where ak0 ̸= 0.
n=k0

If ∞
X
g(z) = an+k0 (z − z0 )n ,
n=0

then g is analytic in some neighborhood B of z0 and

f (z) = (z − z0 )k0 g(z) for all z ∈ B.

̸ 0, there exists a punctured neighborhood

Since g is continuous and g(z0 ) = ak0 =
∗
U of z0 where
g(z) ̸= 0 for all z ∈ U ∗ .
In particular,
0 ̸= (z − z0 )k0 g(z) = f (z) for all z ∈ U ∗ ,
which contradicts the assumption that z0 is an accumulation point of the set N .
Therefore, iii) follows.
Finally, we prove that iii) ⇒ i): Let

U = {z ∈ D : f (k) (z) = 0 for all k = 0, 1, 2, 3, . . . }.

26
Since all derivatives of the analytic function f are continuous (even analytic), the
set U is closed in D. Since z0 ∈ U , it suffices, by the connectedness of D, to show
that U is also open: Let w ∈ U and let
∞
X
f (z) = an (z − w)n
n=0

be the power series expansion of f , which converges in the disk B(w, r) ⊂ D. By

Theorem 2.21,
f (n) (w)
an = = 0 for all n = 0, 1, 2, 3, . . . ,
n!
so f (z) = 0 for all z ∈ B(w, r), and thus B(w, r) ⊂ U , which means that U is an
open set.

A function f is called discrete if for any w ∈ C the set f −1 ({w}) a discrete set,
i.e., it has no accumulation point in the domain of f . According to Theorem 2.22,
an analytic function in domain D is discrete unless it is a constant map.
By Theorem 2.22, the following definition makes sense.

2.23. Definition. Let f be analytic in an open set G and z0 ∈ G. If f (z0 ) = 0 and

f ̸≡ 0 in every neighborhood of z0 , then the number

k0 = min{k ∈ N : f (k) (z0 ) ̸= 0}

is called the order of vanishing at z0 .

2.24. Theorem. Assume that f is analytic in an open set G, z0 ∈ G, and k0 =

1, 2, 3, . . . . Then the following conditions are equivalent:

i) The point z0 is a zero of f of order k0 .

ii) There exists an analytic function g : G → C such that g(z0 ) ̸= 0 and

f (z) = (z − z0 )k0 g(z) for all z ∈ G .

Proof: Let z0 be a zero of f of order k0 , and let

∞
X
f (z) = an (z − z0 )n
n=0

27
be the power series expansion of f in the disk B(z0 , r) ⊂ G. Then an = 0 for all
n < k0 , thus

∞
f (z) −k0
X
= (z − z0 ) an (z − z0 )n
(z − z0 )k0 n=k 0
∞
X
= an+k0 (z − z0 )n , when z ∈ B ∗ (z0 , r) ,
n=0

where the right-hand side defines an analytic function in the disk B(z0 , r), and its
value at z0 is
f (k0 ) (z0 )
ak0 = ̸= 0 .
k0 !
Thus, the function

f (z)

 , when z ̸= z0
g(z) = (z − z0 )k0
ak 0 , when z = z0 ,


is the desired function.

Conversely, suppose g is an analytic function such that g(z0 ) ̸= 0 and

f (z) = (z − z0 )k0 g(z) for all z ∈ G .

Then f ̸≡ 0 in any neighborhood of the point z0 , and if k ≤ k0 , then

k  
(k)
X k k0 !
f (z) = (z − z0 )k0 +n−k g (n) (z) ,
n (k0 − k + n)!
n=0

where  
k k!
= .
n n!(k − n)!
Thus, (
(k) 0, when k < k0
f (z0 ) =
k0 !g(z0 ) ̸= 0 , when k = k0 ,

so z0 is a zero of f of order k0 .

28
2.3. Laurent series

For Laurent series, we need double series that are infinite in both directions:

∞
P
2.25. Definition. The double series zn converges if
n=−∞

∞
X ∞
X
both zn and z−n converge.
n=0 n=1

Then ∞ ∞ ∞
X X X
zn = zn + z−n .
n=−∞ n=0 n=1

∞
P ∞
P ∞
P
The series zn converges absolutely if both zn and z−n converge abso-
n=−∞ n=0 n=1
lutely.
P∞
We say that the series n=−∞ an 0n diverges if an ̸= 0 for some n < 0.
The convergence, absolute convergence, and (local) uniform convergence in a set
A of double function series are defined analogously to the corresponding concepts of
function series (see p. 16).

2.26. Definition. A Laurent series at point z0 ∈ C is of the form

∞
X
an (z − z0 )n , an ∈ C
n=−∞

as a double function series.

The number
1
ρO = p
lim sup n
|an |
n→∞
∞
an (z − z0 )n and
P
is the outer convergence radius of the Laurent series
n=−∞
p
n
ρI = lim sup |a−n |
n→∞

its inner convergence radius. When ρI < ρO , we say that

D = {z ∈ C : ρI < |z − z0 | < ρO }
is the convergence ring of the series. If ρI = 0 and ρO = ∞, then D = C \ {0}.

29
2.27. Remark. The Taylor series, i.e., the ordinary power series, is a Laurent series
when an = 0 for all n < 0. Its inner convergence radius is 0.

2.28. Theorem. Let ρO and ρI be the outer and inner convergence radii of the
∞
an (z − z0 )n . The series diverges if
P
Laurent series
n=−∞

|z − z0 | > ρO or |z − z0 | < ρI .

Additionally, the Laurent series has the following convergence properties:

∞
• If ρO > 0, then
P
an (z − z0 ) converges absolutely and locally uniformly in the
n=0
disk BO = B(z0 , ρO ) and thus defines an analytic function fO in the disk BO .
∞
• If ρI < ∞, then a−n (z − z0 )−n converges absolutely and locally uniformly
P
n=1
in the open set
∁BI = {z : |z − z0 | > ρI } ,

so the function
∞
X
fI (z) = a−n (z − z0 )−n
n=1

is analytic in the set ∁BI .

• If ρI < ρO , then the Laurent series under consideration converges absolutely

and locally uniformly in the annulus

D = {z : ρI < |z − z0 | < ρO }

and thus
∞
X
f (z) = an (z − z0 ) = fI (z) + fO (z)
n=−∞

is analytic in the set D. Moreover, for all n ∈ Z

Z
1 f (z)
an = dz ,
2πi (z − z0 )n+1
{|z−z0 |=r}

as long as ρI < r < ρO and {z : |z − z0 | = r} is oriented counterclockwise.

30
 ∞
an (z − z0 )n , it
P
Proof: Since ρO is the radius of convergence of the power series
n=0
follows from it and the statements concerning the function fO in Theorem 2.19.
Then consider the series
∞
X
a−n ζ n ,
n=1

which is a power series with convergence radius of 1/ρI . The series diverges if
|ζ| > ρ1I , and if ρI < ∞, it converges absolutely and locally uniformly in the disk
B(0, 1/ρI ). Denote
1
ζ= ,
z − z0
then from the local uniform convergence of the power series follows (exercise) that
the series
∞
X
a−n (z − z0 )−n
n=1

converges absolutely and locally uniformly in the set {z : |z −z0 | > ρI } and diverges
when |z − z0 | < ρI . The function fI becomes analytic in particular in the set ∁BI .
∞
an (z − z0 )n diverges, if
P
From these, it follows that the double series
n=−∞

either |z − z0 | > ρO or |z − z0 | < ρI .

If ρI < ρO i.e., ρO > 0 and ρI < ∞, then from the above considerations it follows
∞
an (z − z0 )n converges absolutely and locally uniformly in the open ring
P
that
n=−∞

D = BO ∩ ∁BI

and there
∞
X
f (z) = an (z − z0 )n = fO (z) + fI (z)
n=−∞

is analytic.
Let ρI < r < ρO and k ∈ Z. Denote

∞
f (z) X
= an (z − z0 )n−k−1 , z ∈ D.
(z − z0 )k+1 n=−∞

31
Let γ(t) = z0 + reit , t ∈ [0, 2π]. According to Lemma 2.12, the order of integration
and summation can be exchanged, because the series converges locally uniformly in
the set D ⊃ |γ|, and we obtain
Z Z ∞
1 f (z) 1 X
k+1
dz = an (z − z0 )n−k−1 dz
2πi (z − z0 ) 2πi n=−∞
γ γ
∞ Z
X an
= (z − z0 )n−k−1 dz
n=−∞
2πi
γ
= ak ,

since the function (z − z0 )n−k−1 has a primitive, except when n = k.

Conversely, for an analytic function in the annulus, the Laurent series expansion
is:

2.29. Theorem (Laurent series expansion of an analytic function). Assume that

0 ≤ a < b ≤ ∞. Let f be analytic in the annulus

D = {z ∈ C : a < |z − z0 | < b}.

Then f can be represented by a Laurent series in the set D,

∞
X
f (z) = an (z − z0 )n , z ∈ D.
n=−∞

The representation is unique: for all n ∈ Z,

Z
1 f (z) dz
an = ,
2πi (z − z0 )n+1
{|z−z0 |=r}

provided that a < r < b and the circle is {|z − z0 | = r} oriented counterclockwise.

Proof: Let r0 ∈ (a, b). Let

Z
1 f (z) dz
an = .
2πi (z − z0 )n+1
{|z−z0 |=r0 }

Note that, by Cauchy’s theorem, the choice of radius r0 does not matter because the
paths of two differently radiused contours are homologous in D.

32
 β
γ

z0
z
s r

Let z ∈ D. Choose r, s such that a < r < |z − z0 | < s < b. Let

γ(t) = z0 + reit , 0 ≤ t ≤ 2π
β(t) = z0 + seit , 0 ≤ t ≤ 2π.

Then the cycle σ = (β, ← −) is null-homologous in D and n(σ, z) = 1, so from Cauchy’s

γ
integral formula 1.9 follows that
Z Z Z
1 f (ζ) 1 f (ζ) 1 f (ζ)
(∗) f (z) = dζ = dζ − dζ.
2πi ζ −z 2πi ζ −z 2πi ζ −z
σ β γ

Now proceed as in the proof of the power series representation (see Theorem 2.21).
If ζ ∈ |β|, then
z − z0 |z − z0 |
= < 1,
ζ − z0 s
thus
∞
X f (ζ)(z − z0 )n
f (ζ) f (ζ) 1
= = .
ζ −z ζ − z0 1 − z−z
ζ−z0
0
n=0
(ζ − z0 )n+1

By the Weierstrass M -test, convergence (with respect to ζ) is locally uniform in

D \ B(z0 , |z − z0 |) ⊃ |β|, thus Lemma 2.12 implies that
 
Z ∞ Z
1 f (ζ) X
 1 f (ζ)
(∗∗) dζ = n+1
dζ (z − z0 )n
2πi ζ −z n=0
2πi (ζ − z0 )
β β
| {z }
=an
∞
X
= an (z − z0 )n .
n=0

33
 ζ−z0 r
Similarly, z−z0
= |z−z0 |
< 1 for all ζ ∈ |γ|, and

∞ ∞
f (ζ) f (ζ) 1 X f (ζ)(ζ − z0 )n X f (ζ)(ζ − z0 )n−1
=− ζ−z0
= − n+1
= − n
.
ζ −z z − z0 1 − z−z n=0
(z − z0 ) n=1
(z − z0 )
0

By virtue of the Weierstrass M -test, convergence is locally uniform with respect to

ζ in the set D ∩ B(z0 , |z − z0 |) ⊃ |γ|, so
 
Z ∞ Z
1 f (ζ) X
 1 f (ζ)
(∗ ∗ ∗) dζ = − −n+1
dζ (z − z0 )−n
2πi ζ −z n=1
2πi (ζ − z0 )
γ γ
| {z }
=a−n
∞
X
=− a−n (z − z0 )−n .
n=1

∞
an (z − z0 )n converges in the ring D, and from equations
P
Now the double series
n=−∞
(∗), (∗∗), and (∗ ∗ ∗), it follows that
∞ ∞
!
X X
f (z) = an (z − z0 )n − − a−n (z − z0 )−n
n=0 n=1
X∞
= an (z − z0 )n
n=−∞

for all z ∈ D.
Uniqueness follows from Theorem 2.28 as in the case of power series (exercise).

Example. Let f (z) = e1/z , then f is analytic in the set C \ {0}. Since
∞
z
X zn
e = for all z ∈ C ,
n=0
n!

we obtain
∞ 1 n ∞ 0


1
X
z
X z −n zn
X
e =
z = = .
n=0
n! n=0
n! n=−∞
(−n)!

This is the Laurent series of f , which is unique.

34
 3. Singular points and residue theorem

3.1. Singular points

A function f has an isolated singular point (or simply: singular point) z0 ∈ C if

there exists such r > 0 that f is analytic in the punctured disk

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

We say that f is analytic (in an open set) G except for isolated singular points,
if there is a set of isolated singular points E ⊂ G such that the set E has no
accumulation points in the set G and f is analytic in the set G \ E.
Let f have an isolated singular point z0 and let r > 0 be chosen such that f is
analytic in the punctured disk B ∗ = B(z0 , r) \ {z0 }. According to Theorem 2.29, the
function f has a unique Laurent series, that is
∞
X
f (z) = an (z − z0 )n
n=−∞

for all z ∈ B ∗ .
The singular points are classified as follows:

• z0 is a removable singularity of the function f if an = 0 for all n < 0.

• z0 is a pole of the function f if there exists k < 0 such that ak ̸= 0, but ak ̸= 0
for only finitely many k < 0.
• z0 is an essential singularity of the function f if ak ̸= 0 for infinitely many
k < 0.

Example. The origin (i.e., the point 0) is

sin z
a removable singularity of the function f (z) = ,
z
1
a pole of the function g(z) = , and
z
1
an essential singularity of the function h(z) = e z .

35
3.1. Remark (Important!). The point z0 is a removable singularity of the function
f if and only if (
f (z) , when z ̸= z0
f˜(z) =
a0 , when z = z0 ,

is well-defined and analytic in a neighborhood of point z0 ; here, the a0 is the 0th

coefficient of the Laurent series of f .

3.2. Remark. Let z0 be an isolated singular point of the function f . We say that

−1
X
S(z) = an (z − z0 )n
n=−∞

is the singular part (or principal part) of the Laurent series

∞
X
f (z) = an (z − z0 )n
n=−∞

Then S is analytic in the set C \ {z0 } (exercise).

Additionally, z0 is a singularity of the function f − S. However, it is removable
because ∞
X
f (z) − S(z) = an (z − z0 )n
n=0

in some punctured neighborhood of the point z0 .

3.3 . Remark. The coefficient a−1 appearing in the Laurent series of f plays a
special role. Suppose z0 is a singular point of the function f and let

a−1 =: Res(f, z0 )

denote the residue of the function f at point z0 . If r > 0 is sufficiently small, then
by Theorem 2.29, we have
Z
1
Res(f, z0 ) = f (z) dz ,
2πi
{|z−z0 |=r}

where the contour is oriented counterclockwise.

36
3.2. Removable singularities

Next, we characterize removable singularities. The following result is quite useful

in many situations.

3.4. Theorem (Riemann’s removable singularity theorem). Let f be a function

with an isolated singular point z0 . Then the following are equivalent:

i) z0 is a removable singularity.
ii) There exists r > 0 such that f is bounded in the punctured disk

B ∗ = {z ∈ C : 0 < |z − z0 | < r}.

iii) We have:
lim (z − z0 )f (z) = 0 .
z→z0

Proof: i) ⇒ ii). Exercise.

ii) ⇒ iii). Let M, r > 0 such that

|f (z)| ≤ M , when z ∈ B ∗ = {z ∈ C : 0 < |z − z0 | < r} .

Now
z→z
0
|z − z0 ||f (z)| ≤ M |z − z0 | −−−→ 0.

iii) ⇒ i). Define

(
(z − z0 )f (z) , when z ̸= z0
g(z) =
0, when z = z0 .

Since g is continuous in the disk B = B(z0 , r) for small r > 0 and analytic in the
punctured disk B ∗ (z0 , r), it follows from Theorem CA1.5.10 that g is analytic in the
entire disk B. Thus, when z ̸= z0 ,
∞
1 1 X
f (z) = g(z) = bn (z − z0 )n
z − z0 z − z0 n=0
∞ ∞
b0 =g(z0 )=0 1 X X
= bn (z − z0 )n = bn+1 (z − z0 )n ,
z − z0 n=1 n=0

which is a Laurent series with singular part equal to 0. Hence z0 is removable.

37
3.3. Poles

3.5. Definition. Let z0 be a pole of the function f and

∞
X
f (z) = an (z − z0 )n .
n=−∞

Then the number

k = − min{n < 0 : an ̸= 0}
is called the order of the pole at z0 .

3.6. Theorem. Let k ∈ N and f be analytic in the punctured disk B ∗ = B ∗ (z0 , r).
Then z0 is a pole of the function f of order k if and only if there exists an analytic
function g in the disk B(z0 , r) such that g(z0 ) ̸= 0 and

g(z)
f (z) = for all z ∈ B ∗ .
(z − z0 )k

Proof: First, let’s prove the necessity part. For all z ∈ B ∗

∞
X
f (z) = an (z − z0 )n , a−k ̸= 0.
n=−k

Define ∞
X
g(z) := an−k (z − z0 )n .
n=0

Then g is analytic in the disk B(z0 , r), as the series converges in the disk B(z0 , r).
Moreover, g(z0 ) = a−k ̸= 0. Now
∞
X ∞
X
n −k
f (z) = an (z − z0 ) = (z − z0 ) an−k (z − z0 )n
n=−k n=0
−k
= (z − z0 ) g(z).

Next, let’s prove the sufficiency part. When

∞
X
g(z) = bn (z − z0 )n , b0 ̸= 0 ,
n=0

38
and
f (z) = (z − z0 )−k g(z) ,

then
∞
X ∞
X
−k n
f (z) = (z − z0 ) bn (z − z0 ) = bn+k (z − z0 )n ,
n=0 n=−k

which is the Laurent series of the function f with −kth coefficient being b0 ̸= 0.

3.7. Remark (Important!). Let z0 be a pole of the function f of order k and g be

analytic in a neighborhood of point z0 such that for all z ∈ B(z0 , r) \ {z0 }

f (z) = (z − z0 )−k g(z).

Then
∞
X
−k
f (z) = (z − z0 ) bn (z − z0 )n
n=0
∞
X
= bn+k (z − z0 )n ,
n=−k

which is the Laurent series of the function f . Here, the coefficient of the term (z −
z0 )−1 is
g (k−1) (z0 )
b−1+k = bk−1 =
(k − 1)!
that is
g (k−1) (z0 )
Res(f, z0 ) = .
(k − 1)!
Since f is not defined at point z0 , it’s best to write this in the form: If z0 is a pole
of order k of the function f , then

1 dk−1
lim k−1 (z − z0 )k f (z) .


(3.1) Res(f, z0 ) =
(k − 1)! z→z0 dz

In particular, if z0 is a first-order pole, we have

Res(f, z0 ) = lim (z − z0 )f (z).

z→z0

39
 Example. Let
1
f (z) =
ez −1
Then
1 1 1 1
f (z) = z2
= z z2
= g(z),
z+ 2!
+ ... z 1+ 2!
+ 3! + . . . z

where g is analytic and g(0) ̸= 0. Thus the function f has a first-order pole at point
0 and
z 1
Res(f, 0) = lim zf (z) = lim z = d z = 1.
z→0 z→0 e − 1 (e )|z=0
dz

3.8. Example. Let

z2
f (z) = ,
1 + z4
then the function f has poles at
π 3π 5π 7π
a1 = e i 4 , a2 = ei 4 , a3 = ei 4 and a4 = e i 4 .

a2 a1

a3 a4

Clearly, the poles an are simple (exercise). Thus

z2
Res(f, a1 ) = lim (z − a1 )f (z) = lim
z→a1 z→a1 (z − a2 )(z − a3 )(z − a4 )
2
a1 √ 2i
= = 2
(a1 − a2 )(a1 − a3 )(a1 − a4 ) 2i · 2(1 + i) · 2i
√
2 1−i 1 iπ
= = √ = e− 4 .
4(i + 1) 4 2 4

Similarly
−1 − i 1 3iπ
Res(f, a2 ) = √ = e− 4 .
4 2 4
The residues at the poles a3 and a4 can be determined similarly.

40
3.9. Theorem. Let z0 be an isolated singularity of the function f . Then

i) point z0 is a pole of the function f if and only if limz→z0 |f (z)| = ∞.

ii) point z0 is a pole of order k if and only if k is the positive integer for which

lim |z − z0 |k |f (z)| ∈]0, ∞[.

z→z0

Proof: Let z0 be a pole of the function f of order k. Then there exists an analytic
function g in a neighborhood of point z0 such that g(z0 ) ̸= 0 and

g(z)
f (z) = .
(z − z0 )k

Thus
g(z)
lim |f (z)| = lim = ∞,
z→z0 z→z0 (z − z0 )k
because g(z0 ) ̸= 0. Also,

∞ ,
 if l < k
l l−k
|z − z0 | |f (z)| = |z − z0 | |g(z)| → |g(z0 )| ∈ (0, ∞) , if l = k

0, if l > k.


This proves the “only if” parts of both claims.

Conversely, suppose

1
lim |f (z)| = ∞ and h= .
z→z0 f

Now h is bounded (|h| ≤ 1) in a neighborhood of point z0 , so z0 is a removable

singularity for the function h and

h(z0 ) = lim h(z) = 0 .

z→z0

Since h ̸≡ 0, then if
∞
X
bn (z − z0 )n
n=0

is the power series expansion of the function h, then b0 = 0, but

k := min{n ∈ N : bn ̸= 0} ∈ {1, 2, 3, . . . }.

41
Thus the function
1
g(z) = P
∞
bn+k (z − z0 )n
n=0

is analytic in a neighborhood of point z0 and

1
g(z0 ) = ̸= 0 .
bk

Thus

1 1 1 g(z)
f (z) = = P
∞ = ∞ = .
h(z) P (z − z0 )k
bn (z − z0 )n (z − z0 )k bn+k (z − z0 )n
n=k n=0

According to Theorem 3.6, z0 is a pole of the function f with order k, which proves
the “if” part of the first claim.
Finally, if it is known that

lim |z − z0 |k |f (z)| = a ∈ (0, ∞)

z→z0

for some k ∈ N, then

|f (z)| → ∞ , when z → z0 ,
and the above proof shows that z0 is a pole of order k (note that 1/|bk | = a).

3.4. Essential singularities

The behavior of an analytic function around an essential singularity is wild. One

has Picard’s great theorem: If z0 is an essential singularity of an analytic function
f , then C \ f (B ∗ (z0 , r)) has at most one point, for any r > 0.
However, we will not prove this deep theorem, but instead a weaker (and easier)
result.

3.10. Theorem (Casorati-Weierstrass). If z0 is an isolated essential singularity of an

analytic function f in the punctured disk B ∗ (z0 , r), then the image set f (B ∗ (z0 , r))
is dense in C, i.e.,
f (B ∗ (z0 , r)) = C .

42
Proof: Suppose the image set is not dense, then in its complement there is a disk

B(w, ε) ∩ f (B ∗ (z0 , r)) = ∅ .

Then the function

1
g(z) =
f (z) − w
would be analytic in the punctured disk B ∗ (z0 , r). Moreover, g is bounded, because
1
|g(z)| =
|f (z) − w|
1
≤ ,
ε
so by Riemann’s removable singularity theorem, Theorem 3.4, z0 is a removable
singularity for g. Thus, since z0 has a limit for g, z0 is either a removable singularity
or a pole of 1/g. Hence, z0 is a singularity for the function
1
f (z) = w +
g(z)
which is either removable or a pole, which contradicts the assumption that z0 was
an essential singularity of f .

An entire function f is either a polynomial or it has an essential singularity at

infinity:

3.11. Theorem. Let f : C → C be an entire function. Then the function

1
g(z) = f ( )
z
has an essential singularity at the origin or f is a polynomial.

Proof: If f is not a polynomial, then f has a power series representation

∞
X
f (z) = an z n , z ∈ C,
n=0

where an ̸= 0 for infinitely many n. Thus, the Laurent series of the function g is
0
X
g(z) = bn z n , bn = a−n , z ∈ C \ {0} ,
n=−∞

so since bn = a−n ̸= 0 for infinitely many n < 0, 0 is its essential singularity.

43
 Example. Let f be analytic in the set B ∗ = B ∗ (z0 , r). Suppose that f is not
identically zero in B ∗ and that there exist points zn ⊂ B ∗ such that limn→∞ zn = z0
and f (zn ) = 0 for all n ∈ N. (As a concrete example, consider the function
 
1
f (z) = cos ,
z−1

which has an isolated singularity at z0 = 1.)

Then:

• z0 cannot be a removable singularity of the function f , since otherwise f would

be identically zero in the set B ∗ , see Theorem 2.22.
• Since limn→∞ |f (zn )| = 0 < ∞ and limn→∞ zn = z0 , then z0 cannot be a pole
of the function f , see Theorem 3.9.

Therefore, the point z0 must be an essential singularity (no other options are possible,
QED).

3.5. Residue theorem

The residue theorem provides an efficient tool for computing complex integrals.

3.12. Theorem (Residue theorem). Let σ be a null-homologous cycle in an open

set G, and let f be analytic in G except for an isolated set of singularities E, where
E ∩ |σ| = ∅. Then Z X
f (z) dz = 2πi n(σ, a) Res(f, a).
σ a∈E

3.13 . Remark. In the proof of Cauchy’s theorem, it was shown that if σ is a

null-homologous cycle in G, then

Kσ = |σ| ∪ {z ∈ C \ |σ| : n(σ, z) ̸= 0}

is a compact subset of G. Let D ⊂⊂ G be an open set such that Kσ ⊂ D. Then σ

is null-homologous in D. Since the set E has no accumulation points in G, the set
E ∩ D = {z1 , . . . , zp } is finite. Thus, the sum in the statement is well-defined:
p
X X
n(σ, a) Res(f, a) = n(σ, zj ) Res(f, zj ).
a∈E j=1

44
Proof: Let z1 , . . . , zp be as in the remark, and let Sk be the singular part of the
function f at the point zk , k = 1, . . . , p. Then Sk is analytic in C \ {zk } by Theorem
2.28, and thus the function f − Sk has a removable singularity at zk . Therefore,
g = f − S1 − S2 − · · · − Sp
is analytic in D except for removable singularities at z1 , . . . , zp . Since these are
removable, we can assume that g is analytic in the whole of D. Now, from Cauchy’s
theorem it follows that
Z Z p Z
X
0 = g(z)dz = f (z) dz − Sk (z)dz
σ σ k=1 σ

or
Z p Z
X
(∗) f (z) dz = Sk (z)dz.
σ k=1 σ

Now, if
∞
X
S(z) = a−n (z − z0 )−n
n=1

is the singular part of the function f at an arbitrary point z0 ∈ E, then the series S
converges locally uniformly in C \ {z0 }. In particular, it converges uniformly along
the cycle σ. Thus,
∞
Z Z X ! ∞ Z
a−n X dz
S(z)dz = n
dz = a −n n
,
n=1
(z − z 0 ) n=1
(z − z0 )
σ σ σ

where we used Lemma 2.12. Since the function (z −z0 )−n has a primitive when n > 1
(or by the Cauchy integral formula for derivatives),
Z
(z − z0 )−n dz = 0 , when n > 1 ,
σ

so we obtain
Z Z
dz
S(z)dz = a−1 = 2πi Res(f, z0 )n(σ, z0 ).
z − z0
σ σ

Now, from equation (∗) it follows that

Z p Z p
X X
f (z) dz = Sk (z)dz = 2πi n(σ, zk ) Res(f, zk ).
σ k=1 σ k=1

45
 Example. Let us show that

Z∞
x2 π
4
dx = √ .
1+x 2
−∞

First, we note that the integral converges, since

x2 1
4
≤ 2.
1+x x

Let
z2
f (z) = ,
1 + z4
so that the function f has poles at
π 3π 5π 7π
a1 = e i 4 , a2 = ei 4 , a3 = ei 4 and a4 = e i 4 .

Following the calculation in Example 3.8, we have

1 iπ 1 3iπ
Res(f, a1 ) = e− 4 , Res(f, a2 ) = e− 4 .
4 4

Let R > 1 and γR be the semicircle B(0, R) ∩ {z : Im(z) > 0} traversed coun-
terclockwise.

γR
a2 a1

−R 0 R

The residue theorem implies that

Z
1 i
(∗) f (z) dz = Res(f, a1 ) + Res(f, a2 ) = − √ .
2πi 2 2
γR

Since
ZR Zπ
x2 R3 e3it
Z
1 1 1
f (z) dz = 4
dx + dt,
2πi 2πi 1+x 2π 1 + R4 e4it
γR −R 0

46
we have
ZR Zπ
x2 (∗) π e3it
4
dx = √ − iR3 dt.
1+x 2 1 + R4 e4it
−R 0

Since t ∈ [0, π], we have |1 − 1 + R4 e4it | = R4 , so by the triangle inequality,

1 + R4 e4it ≥ R4 − 1,

when R > 1. Thus,

Zπ Zπ
3 e3it R3 πR3
R dt ≤ |e3it | dt = 4 → 0,
1 + R4 e4it R4 − 1 |{z} R −1
0 0 =1

as R → ∞. Therefore,

Z∞ ZR
x2 x2
dx = lim dx
x4 + 1 R→∞ x4 + 1
−∞ −R
Zπ
π e3it π
= √ − lim iR3 4 4it
dt = √ .
2 R→∞ 1+R e 2
0
| {z }
=0

3.14. Remark. When evaluating integrals as above by taking the limit R → ∞, it

must be ensured that the integrands converge absolutely! Compare this remark to
the one regarding double series.

47
 4. Mapping properties of analytic functions

4.1. Analytic functions are angle-preserving

Let U ⊂ R2 be open and let f : U → R2 be a C 1 map (not necessarily analytic).

In this section we will prove a geometric characterization of analytic functions: f is
analytic if and only if the map f preserves (infinitesimal) angles between curves.
We will identify vectors in R2 with points in C and write

z = (x, y) = x + iy.

Let z0 ∈ U and let γ : (−ε, ε) → U be a C 1 curve with γ(0) = z0 . Writing γ(t) =

(x(t), y(t)), the tangent vector of γ is given by

γ̇(t) = (ẋ(t), ẏ(t)) = ẋ(t) + iẏ(t).

If f = u + iv is a C 1 map U → C (not necessarily analytic), then (f ◦ γ)(t) is a C 1

curve through f (z0 ). Its tangent vector is given by the following formula.

4.1. Lemma. One has

(f ◦ γ)˙(t) = ∂f (γ(t))γ̇(t) + ∂f (γ(t))γ̇(t)

where the right hand side involves multiplication of complex numbers, and ∂f and
∂f are the Wirtinger derivatives

1 1
∂f (z) = (∂x f (z) − i∂y f (z)), ∂f (z) = (∂x f (z) + i∂y f (z)).
2 2
Proof: Write f = u + iv and γ(t) = (x(t), y(t)). Then the chain rule gives

(f ◦ γ)˙(t) = ∂t [u(x(t), y(t))) + i∂t (v(x(t), y(t))]

= ux (γ(t))ẋ(t) + uy (γ(t))ẏ(t) + i [vx (γ(t))ẋ(t) + vy (γ(t))ẏ(t)]
= (ux + ivx )ẋ(t) + (uy + ivy )ẏ(t)
= fx · x(t) + fy · y(t)
1 1
= (fx − ify )(ẋ(t) + iẏ(t)) + (fx + ify )(ẋ(t) − iẏ(t)).
2 2

48
4.2. Definition. Let f : U → C be a C 1 map. We say that f is angle-preserving,
or conformal, at z0 ∈ U if there are r(z0 ) > 0 and θ(z0 ) ∈ (−π, π] such that for any
C 1 curve γ(t) in U with γ(0) = z0 , one has

(f ◦ γ)˙(0) = r(z0 )eiθ(z0 ) γ̇(0).

Thus f is angle-preserving at z0 if the tangent vector of the curve (f ◦ γ)(t) at

t = 0 is obtained by scaling and rotating the tangent vector γ̇(0) by a factor r(z0 )
and angle θ(z0 ) that are independent of the curve γ. We require r(z0 ) > 0 to exclude
the degenerate case where the tangent vectors (f ◦ γ)˙(0) would always be zero.

4.3. Remark. With an appropriate definition of angle, one can prove (exercise)
that f is angle-preserving at z0 if and only if the angle between the tangent vectors
of (f ◦ γ1 )(t) and (f ◦ γ2 )(t) at t = 0 is equal to the angle between γ̇1 (0) and γ̇2 (0)
whenever γ1 (t) and γ2 (t) are C 1 curves through z0 . Thus an angle-preserving map
preserves the angles between tangent vectors. However, it may change the lengths
of tangent vectors (this happens when r(z0 ) ̸= 1).

4.4. Theorem (Analytic = conformal). Let f : U → C be a C 1 map and z0 ∈ U .

Then f is angle-preserving at z0 if and only f has a complex derivative at z0 and
f ′ (z0 ) ̸= 0. If γ is a C 1 curve with γ(0) = z0 , the tangent vector of (f ◦ γ)(t) satisfies

(f ◦ γ)˙(0) = f ′ (z0 )γ̇(0).

Proof: Writing f = u + iv, one has

1 1 1
∂f = (∂x f + i∂y f ) = (ux + ivx + i(uy + ivy )) = (ux − vy + i(uy + vx )).
2 2 2

Thus the Cauchy-Riemann equations ux = vy , uy = −vx are valid at z0 if and only if

∂f (z0 ) = 0.

Since the complex derivative is given by f ′ (z0 ) = ux (z0 ) + ivx (z0 ), we also have
∂f (z0 ) = f ′ (z0 ) whenever the complex derivative exists.
“⇐=” Suppose that f has a complex derivative at z0 and f ′ (z0 ) ̸= 0. Then the
Cauchy-Riemann equations are valid at z0 , so ∂f (z0 ) = 0. From Lemma 4.1 we
obtain that
(f ◦ γ)˙(0) = ∂f (z0 )γ̇(0)

49
whenever γ is a C 1 curve with γ(0) = z0 . Writing f ′ (z0 ) = ∂f (z0 ) = r(z0 )eiθ(z0 ) , we
see that f is angle-preserving at z0 . One has r(z0 ) > 0 since f ′ (z0 ) ̸= 0.
“=⇒” Suppose that f is angle-preserving at z0 . Then by definition there is w0 ∈ C,
w0 ̸= 0 with
(f ◦ γ)˙(0) = w0 γ̇(0)
for any C 1 curve γ(t) through z0 . On the other hand, Lemma 4.1 gives that

(f ◦ γ)˙(0) = ∂f (z0 )γ̇(0) + ∂f (z0 )γ̇(0).

Fix any w ∈ C and consider the curve γ(t) = z0 + tw. Then γ̇(0) = w, and the
previous two formulas imply

w0 w = ∂f (z0 )w + ∂f (z0 )w̄.

This can be rewritten as

(∂f (z0 ) − w0 )w + ∂f (z0 )w̄ = 0.

This is true for any w ∈ C. Choosing w = 1 and w = i yields

∂f (z0 ) = w0 , ∂f (z0 ) = 0.

Since ∂f (z0 ) = 0, the Cauchy-Riemann equations are valid at z0 and therefore f has
a complex derivative at z0 . One also has f ′ (z0 ) = ∂f (z0 ) = w0 ̸= 0.

Theorem 4.4 shows that indeed analytic functions can be characterized by the
property that they preserve (infinitesimal) angles. We will next begin to use the
facts proved in the previous chapters in order to study conformal mappings.

4.2. Argument principle and Hurwitz’s theorem

4.5. Theorem. (Argument principle). Let f be an analytic function in the domain

D, whose zeros are a1 , a2 , . . . , an ∈ D. Let kj ∈ N be the multiplicity of the zero aj
and σ be a null-homologous cycle in D. If aj ̸∈ |σ| for all j, then
n
f ′ (z)
Z
1 X
dz = kj n(σ, aj ) .
2πi f (z) j=1
σ

50
Proof: Let j ∈ {1, 2, . . . , n}. By Theorem 2.24, there exists an analytic function
g in the domain D and r > 0 such that g(aj ) ̸= 0 and f (z) = (z − aj )kj g(z) for all
z ∈ B(aj , r). Moreover, we can assume that g(z) ̸= 0 for all z ∈ B(aj , r), so

f ′ (z) kj (z − aj )kj −1 g(z) (z − aj )kj g ′ (z) kj g ′ (z)

= + = +
f (z) (z − z0 )kj g(z) (z − aj )kj g(z) z − aj g(z)

when z ∈ B ∗ (aj , r). Since the function g ′ /g is analytic in a neighborhood of point

aj , then Res(f ′ /f, aj ) = kj .
By the residue theorem,

n n
f ′ (z)
Z
1 X X
dz = n(σ, aj )Res(f ′ /f, aj ) = kj n(σ, aj )
2πi f (z) j=1 j=1
σ

as desired.

By applying Theorem 4.5 to the function f (z) − w0 , we obtain:

4.6 . Theorem. Let f be an analytic function in the domain D, and let

a1 , a2 , . . . , an ∈ D be the points where f (z) = w0 . Let kj ∈ N be the multiplicity of
the zero aj of w0 (i.e., the multiplicity of the zero of the function f − w0 ), and let σ
be a nullhomologous cycle in D. If aj ̸∈ |σ| for all j = 1, 2, . . . , n, then

n
f ′ (z)
Z
1 X
dz = kj n(σ, aj ) .
2πi f (z) − w0 j=1
σ

4.7. Example. If γ(t) = 2ei2πt , t ∈ [0, 1], then

Z
2z + 1
dz = 4πi ,
z2+z+1
γ

√
since both zeros of the denominator 21 (−1 ± i 3) are on the circumference of the
unit circle.

51
4.8 . Example. Let γ be a null-homologous closed path in the domain D, and
f : D → C be analytic. Let w0 ∈ C \ f (|γ|) and a1 , a2 , . . . , an be the points z ∈ D
such that f (z) = w0 . If kj ∈ N is the multiplicity of the zero w0 at aj and σ = f ◦ γ,
then σ is a closed path and

n
f ′ (z)
Z Z
1 1 1 X
n(σ, w0 ) = dw = dz = kj n(γ, aj ) ,
2πi w − w0 2πi f (z) − w0 j=1
σ γ

where the middle equality can be easily seen with a change of variables (exercise).

4.9. Theorem. (Hurwitz’s theorem). If the analytic functions fj , j ∈ N, converge

locally uniformly to a function f : D → C in the domain D, and fj (z) ̸= 0 for all
j ∈ N and z ∈ D, then either f ≡ 0 or f (z) ̸= 0 for all z ∈ D.

Proof: Assume that f is not identically zero in the domain D, but there exists a
zero of f of multiplicity k0 at z0 ∈ D. By Theorem 2.22, there exists r > 0 such that
B(z0 , 2r) ⊂ D and f does not attain the value zero in the set B(z0 , 2r) \ {z0 }. By
the argument principle (applied in the disk B(z0 , 2r)), we have

f ′ (z)
Z
dz = 2πik0 ̸= 0 .
f (z)
{|z−z0 |=r}

On the other hand, since fj does not vanish in the domain D, then

fj′ (z)
Z
dz = 0
fj (z)
{|z−z0 |=r}

for all j. Since the sequences fj and fj′ converge locally uniformly in D towards the
respective limit functions f and f ′ , and since the function f does not vanish in the
compact set {|z −z0 | = r}, then fj′ /fj converges uniformly to the limit function f ′ /f
in the set {|z − z0 | = r} (exercise). Thus,

fj′ (z) f ′ (z)

Z Z
0 = lim dz = dz = 2πik0 ̸= 0 ,
j→∞ fj (z) f (z)
{|z−z0 |=r} {|z−z0 |=r}

which is a contradiction.

52
 From Hurwitz’s theorem, we can derive the following result, which describes the
uniform limit of analytic injective functions.

4.10. Theorem. Suppose fj , j ∈ N are injective analytic functions in a domain D.

If the functions fj converge to the function f locally uniformly in the set D, then
either f is constant or it is injective in D.

Proof left as an exercise.

4.3. Open mapping theorem for analytic functions

From Theorem 2.22, it follows that a non-constant analytic function in the domain
D is discrete, i.e., the preimage of a point in D is always a discrete set, i.e., a set
without accumulation points. In this section, we analyze further the local behavior
of analytic functions and prove, among other things, that a non-constant analytic
function is an open mapping, i.e., it maps open sets to open sets.
According to the following theorem, an analytic function behaves around its zero
similarly as z n around the origin, where n is the multiplicity of the zero.

4.11. Theorem. Let f : B(z0 , R) → C be analytic and w0 = f (z0 ). If z0 is an n-fold

zero of the function f (z) − w0 , then there exist ε > 0 and 0 < δ < R such that when
0 < |w − w0 | < ε, the equation
f (z) = w
has exactly n simple solutions for z in the disk B(z0 , δ).

Proof: Since the multiplicity of z0 is finite, f is not constant. Since the zeros of
an analytic function are isolated (Theorem 2.24), there exists 0 < δ < 21 R such that
the equation
f (z) = w0
has no solutions in the punctured disk B ∗ (z0 , 2δ) and furthermore

f ′ (z) ̸= 0 for all z ∈ B ∗ (z0 , 2δ) .

Let
γ(t) = z0 + δe2πit , t ∈ [0, 1]
and σ = f ◦ γ. Since w0 ̸∈ |σ|, there exists ε > 0 such that

B(w0 , ε) ∩ |σ| = ∅.

53
Thus, B(w0 , ε) belongs to some component of C \ |σ|, so by Lemma CA1.5.4, for all
w ∈ B(w0 , ε),
n(σ, w0 ) = n(σ, w) .

Let w ∈ B ∗ (w0 , ε) and let aj be the kj -fold zeros of the function f (z) − w. According
to Example 4.8,

p
X
n = n n(γ, z0 ) = n(σ, w0 ) = n(σ, w) = kj n(γ, aj ) .
j=1

Since n(γ, aj ) is either = 0 or = 1 and since f ′ (z) ̸= 0 for all z ∈ B ∗ (z0 , δ), there are
exactly n simple solutions of the equation f (z) = w in aj ∈ B ∗ (z0 , δ).

4.12. Remark. Theorem 4.11 is known as the “branched covering principle”, which
refers to the fact that an analytic function covers the image locally n times.
By analyzing the proof, we see that any δ > 0 such that

f (z) ̸= w0 and f ′ (z) ̸= 0 when 0 < |z − z0 | ≤ δ .

is sufficient. Then, we only need to require ε > 0 such that

ε ≤ |w0 − f (z)| whenever |z − z0 | = δ .

A non-constant analytic function f is an open mapping, i.e., the image of an open

set G under f is always open:

4.13. Theorem (Open mapping theorem for analytic functions). Let f be a non-
constant analytic function in the domain D. Then f is an open mapping.

Proof: Let G ⊂ D be open and w0 ∈ f (G). Choose z0 ∈ G such that f (z0 ) = w0 .

Since f is non-constant, Theorem 4.11 can be applied and we find δ > 0 and ε > 0
such that B(z0 , δ) ⊂ G and for every w ∈ B(w0 , ε) there is (at least one) pre-image
z ∈ B(z0 , δ). Thus,


B(w0 , ε) ⊂ f B(z0 , δ) ⊂ f (G) ,

so f (G) is open.

54
4.14. Corollary. Let f : G → C be analytic and z0 ∈ G. If f ′ (z0 ) ̸= 0, then there
exists r > 0 such that


f |B(z0 ,r) : B(z0 , r) → f B(z0 , r)

is a homeomorphism.

Proof: Let w0 = f (z0 ). Since f ′ (z0 ) ̸= 0, Theorem 4.11 can be applied at the
point z0 in some neighborhood B(z0 , R) ⊂ G with n = 1. Thus, we find ε > 0 and
δ > 0 such that B = B(z0 , δ) ⊂ G and for every w ∈ B(w0 , ε) there is exactly
one pre-image z ∈ B (we can assume that there is exactly one pre-image for w0 as
well in the disk B; additional proof left as an exercise). By the continuity of the
function f , there exists 0 < r <
δ such that f (B(z0 , r)) ⊂ B(w0 , ε). We show that
f |B(z0 ,r) : B(z0 , r) → f B(z0 , r) is a bijection; note that this restriction is surjective
by definition.
For injectivity, let z ∈ B(z0 , r). Assume that ζ ∈ B(z0 , r) such that f (ζ) = f (z) =:
w ∈ B(w0 , ε). By the previous reasoning, z = ζ, otherwise point w would have (at
least) two
pre-images in the set B(z0 , r) ⊂ B(z0 , δ). Thus, f |B(z0 ,r) : B(z0 , r) →
f B(z0 , r) is a bijection; the continuity of the inverse map follows from the openness
of the mapping (Theorem 4.13).

4.15. Corollary. Let f : G → C be an injective analytic function. Then

f ′ (z) ̸= 0 for all z ∈ G .

Proof: Let z0 ∈ G and f (z0 ) = w0 . The point z0 is a simple root of the equation
f (z) = f (z0 ), otherwise we would find a point w ∈ f (G) (close to w0 ), which would
have multiple pre-image points in G (by Theorem 4.11), contradicting the fact that
f is an injection. Hence by the definition of multiplicity, f ′ (z0 ) ̸= 0.

4.4. Introduction to conformal mappings

Many physical phenomena, such as fluid flow, heat conduction, etc., are mathe-
matically modeled using Laplace’s equation

uxx + uyy = 0 , z = (x, y) ∈ C .

55
It is much easier to handle this equation if the domain of interest is either the
upper half-plane H or the unit disk B rather than a general domain G. Using the
Cauchy-Riemann equations, it can be seen that a “conformal change of variables”
preserves the Laplacian, i.e., if f : D → G is a conformal map, cf. Definition 4.16,
and u satisfies the Laplace equation in G, then u ◦ f satisfies the Laplace equation
in D. For this reason, it is important to find conformal mappings between different
domains.

4.16. Definition. A function f : G → C is a conformal map (univalent, schlicht)

in the domain G, if it is an analytic injection in the domain G.

4.17. Remark. Often, the phrase “f is conformal (at point z0 )” is used, meaning
only that f is locally injective (i.e., its derivative does not vanish). In our language,
a conformal map is a global injection. By Lemma 4.14, if f ′ (z0 ) ̸= 0, then there
exists a neighborhood U of z0 where f |U : U → f (U ) is a conformal map.
In the classical sense, conformality means that angles are preserved at an infi-
nitesimal scale: If γ and σ are regular paths passing through the point z0 and if
f ′ (z0 ) ̸= 0, then the angle between the curves γ and σ (tangent lines) = the angle
between the image curves f ◦ γ and f ◦ σ (tangent lines). This is easy to believe in
light of Remark CA1.2.3: both curves γ and σ rotate by the angle arg(f ′ (z0 )) under
the mapping f at the point z0 , and thus their angle remains preserved.

Example. The map f (z) = z 2 is conformal outside the origin, but it is not locally
injective at the origin.
On the other hand, the derivative of the exponential function never vanishes, thus
it defines a locally conformal map.

From the above remarks, we obtain:

4.18. Theorem. Let f : G → C be a conformal map. Then

(i) f defines a homeomorphism f : G → f (G).

(ii) f ′ (z) ̸= 0 for all z ∈ G.

(iii) The inverse function of f , f −1 : f (G) → G, is also a conformal map, and

1
(f −1 )′ (f (z)) = .
f ′ (z)

56
 Example. The conformal mapping between the domains G = {z ∈ C : |z| > 1}
and D = B ∗ (0, 1) is given by f (z) = z1 .

The following fundamental theorem states that every simply connected domain
(which is not the entire complex plane) is conformally equivalent to the unit disk.

4.19. Theorem (Riemann mapping theorem). Let D ̸= C be a simply connected

domain. Then there exists an analytic bijection f : D → B(0, 1).

The proof of the Riemann mapping theorem 4.19 is given in the following two
subsections. Using it, we can easily prove Lemma 4.21, which actually characterizes
the complex plane regions that are conformally equivalent to the disk; we only need
the following auxiliary result.

4.20. Lemma. Let D be a simply connected domain and f : D → C a conformal

map. Then f (D) is simply connected.

Proof: Since f is not constant, f (D) is a domain. Let w0 ∈ C \ f (D) and let σ be
a closed curve in f (D). It suffices to show that n(σ, w0 ) = 0.
Now γ = f −1 ◦ σ is a closed curve in D, where f −1 is the inverse map of f from
f (D) to D, which is also a conformal map. Moreover, σ = f ◦ γ. So

f ′ (z)
Z Z
1 dw 1
n(σ, w0 ) = n(f ◦ γ, w0 ) = = dz .
2πi w − w0 2πi f (z) − w0
f ◦γ γ

Since γ is a closed curve in the simply connected domain D and z 7→ f ′ (z)/(f (z)−w0 )
is analytic in D, the last integral is zero by Cauchy’s theorem.

4.21. Theorem. Let G be an open set. Then the following conditions are equivalent:

(A) G ̸= C and G is simply connected;

(B) there exists an analytic bijection f : G → B(0, 1).

Proof: The implication (A) ⇒ (B) follows from the Riemann mapping theorem.
Let G be an open set and let f : G → B(0, 1) be an analytic bijection; by Liouville’s
theorem, G ̸= C. Furthermore, since the inverse function f −1 : B(0, 1) → G is also
an analytic bijection, Lemma 4.20 shows that G = f −1 [B(0, 1)] is a simply connected
domain.

57
 The Riemann mapping theorem has many proofs, and the proof presented below
originates from Fejér and Riesz (1922). It is based on the ideas of Dirichlet and Rie-
mann, solving a suitably posed extremum problem. For this, we first briefly discuss
the theory of normal families.

4.5. Normal families and Montel’s theorem

Recall the following definition:

• A sequence fj : D → C, where D ⊂ C is a domain, converges locally uniformly

in the set D if and only if the functions fj converge uniformly on every compact
set K ⊂ D.

In this section, we provide an extremely useful criterion (Arzelà–Ascoli theorem),

which guarantees the existence of a locally uniformly convergent subsequence.

4.22. Definition. Let D be a domain and F ⊂ C(D, C) = {f : D → C continuous}

a family of continuous complex-valued functions. We say that the family F is nor-
mal on the domain D, if every sequence of functions from the family F contains a
subsequence that converges locally uniformly in the set D.

Note that the limit function of the (subsequence) does not necessarily belong to
the family F. However, the limit function is always continuous on D. We skip the
proof of the following lemma, which is typically presented in courses on topology or
functional analysis.

4.23 . Theorem (Arzelà–Ascoli). Let D ⊂ C be a domain. Then a family F is

normal in D if and only if conditions (1) and (2) hold:

(1) F is equicontinuous at every point, i.e., for every w ∈ D and ε > 0, there exists
δ > 0 such that

|f (z) − f (w)| < ε whenever z ∈ B(w, δ) and f ∈ F

(2) for every w ∈ D, the set A(w) = {f (w) : f ∈ F} is compact.

4.24 . Remark. Let F ⊂ C(D, C) be a normal family of analytic functions in

the domain D. Every sequence in the family F contains a subsequence that con-
verges uniformly in compact subsets of D. In particular, the limit function of such
subsequence is analytic by Lemma 2.9.

58
 We apply the Arzelà–Ascoli theorem to answer the following question: when is a
family F of analytic functions in an open set D ⊂ C normal?

4.25. Theorem (Montel’s theorem). Suppose D is a domain. A family F of analytic

functions in the set D is normal if and only if F is locally bounded, i.e., for every
compact subset K ⊂ D, there exists M = M (K) < ∞ such that

|f (z)| ≤ M whenever z ∈ K and f ∈ F .

Proof: Let F be a locally bounded family of analytic functions in the domain D.

First, we show that F is equicontinuous. Let w ∈ D and ε > 0. To construct δ > 0,
choose r > 0 such that K = B(w, r) ⊂ D. Consider a function f ∈ F, then by
assumption we have
sup |f (z)| ≤ M = M (K) < ∞ .
|z−w|≤r

Using Cauchy’s integral formula (Theorem 1.10), we have

Z
′ 1 f (ζ)
sup |f (z)| = sup dζ
|z−w|≤r/2 |z−w|≤r/2 2πi (ζ − z)2
∂B(w,r)
M 2πr 4M
≤ 2
= =: M ′ .
2π(r/2) r

Choose δ = min{ε/(2M ′ ), r/2} > 0, then

Z
sup |f (z) − f (w)| = sup f ′ (ζ) dζ
|z−w|≤δ |z−w|≤δ
[w,z]

≤ sup |z − w|M ′ ≤ δM ′ < ε .

|z−w|≤δ

Since δ does not depend on the function f , the family F is equicontinuous.

Furthermore, since {w} ⊂ D is compact, by assumption |f (w)| ≤ M ({w}) whe-
never w ∈ D and f ∈ F. In particular,

A(w) = {f (w) : f ∈ F}

is compact for every w ∈ D, since this set is always bounded.

The normality of the family F now follows from the Arzelà–Ascoli Theorem 4.23.
The necessity of Montel’s condition is left as an exercise.

59
4.6. Proof of the Riemann mapping theorem

We prove the Riemann mapping theorem (Theorem 4.19): if D ̸= C is a simply

connected domain, then there exists an analytic bijection f : D → B(0, 1). We
proceed with the proof. Fix some z0 ∈ D and define the family

F = {f : D → B(0, 1) : f is conformal and f (z0 ) = 0}.

The proof consists of three steps, roughly outlined as follows:

1. F ̸= ∅ ;

2. There exists f ∈ F such that |f ′ (z0 )| ≥ |g ′ (z0 )| for all g ∈ F;

3. If f ∈ F satisfies the condition in step 2, then f has the desired properties.

Step 1. The proof of the first step F ̸= ∅ is based on the assumption D ̸= C and
the simply connectedness of D; details are left as an exercise.
Step 2. Let s = sup{|g ′ (z0 )| : g ∈ F}. Choose a sequence fj indexed by j ∈ N
from the family F such that
lim |fj′ (z0 )| = s .
j→∞

Since F ̸= ∅ and the members of this family are conformal mappings that are
analytic injections in D, by Corollary 4.15 we have s > 0.
By definition, the family F is locally bounded (since for every compact set K ⊂ D,
we have |g(z)| ≤ 1 for all z ∈ K and g ∈ F). According to Montel’s theorem 4.25,
the family F is normal in D. Therefore, there exists a subsequence (fjk )k∈N of the
sequence (fj ) and a function f : D → C such that

k→∞
fjk −−−→ f locally uniformly in D .

By Theorem 2.9, the limit function f is analytic in D. By the same theorem, fj′k → f ′
locally uniformly as k → ∞. In particular,

0 < s = lim |fj′k (z0 )| = |f ′ (z)| ,

k→∞

so f is not constant in D. Since the functions fjk are analytic injections in D, by

Lemma 4.10, f is an injection from D to C. Furthermore, note that f (D) ⊂ B(0, 1)
and f is an open mapping by Lemma 4.13. Thus, f (D) ⊂ B(0, 1). Since fjk (z0 ) = 0
for all k ∈ N, we also have f (z0 ) = 0.

60
 From the above reasoning, it follows that f ∈ F and f satisfies the condition
|f (z0 )| ≥ |g ′ (z0 )| for all functions g ∈ F.
′

Step 3. We claim that f is an analytic bijection D → B(0, 1). After step 2, it

suffices to show that
f : D → B(0, 1) is surjective.

This is proved by contradiction: Suppose there exists w0 ̸= 0 such that

w0 ∈ B(0, 1) \ f (D) .

Using this assumption, we will construct a map ρ ∈ F such that |ρ′ (z0 )| > |f ′ (z0 )|.
This is a contradiction with the condition |f ′ (z0 )| ≥ |g ′ (z0 )| for all g ∈ F.
We need auxiliary mappings (details omitted). If a ∈ B(0, 1), then define

z−a
ϕa (z) = for z ∈ B(0, 1) .
1 − āz

The map ϕa is an analytic bijection from B(0, 1) to B(0, 1), and

1 − |a|2
(4.1) ϕ′a (z) = whenever z ∈ B(0, 1) .
(1 − āz)2

The composite mapping τ = ϕw0 ◦ f is a well-defined conformal mapping D →

B(0, 1). By Lemma 4.20, τ [D] ⊂ B(0, 1) is simply connected and 0 ̸∈ τ [D]. By
Lemma 1.15, there exists a branch of the logarithm log : τ [D] → C in the region
τ [D], which is an analytic injection satisfying elog(z) = z for all z ∈ τ [D]. Define

1
S(w) = e 2 log(w) , for w ∈ τ (D) .

Then S(w)2 = w for w ∈ τ (D), so S is a branch of the square root in τ (D). We will
show that S is a conformal mapping τ (D) → B(0, 1). For all w ∈ τ (D), we have

|S(w)|2 = |w| < 1, w ∈ τ (D).

Thus S(τ (D)) ⊂ B(0, 1). Moreover, if S(w) = S(w′ ) for some w, w′ ∈ τ (D), then

w = S(w)2 = S(w′ )2 = w′ .

This proves the injectivity of the mapping S.

61
 The map S ◦ ϕw0 ◦ f is then conformal and injective D → B(0, 1), but it maps z0
to S(ϕw0 (f (z0 ))) = S(−w0 ). We “normalize” this function by defining

ρ = ϕS(−w0 ) ◦ S ◦ ϕw0 ◦ f.

Then ρ is a conformal map D → B(0, 1) with ρ(z0 ) = 0. In particular, ρ ∈ F. To

prove the contradiction, it suffices to show that

(4.2) |ρ′ (z0 )| > |f ′ (z0 )| .

We still need to prove inequality (4.2). Observe that

ρ=G◦f

where G = ϕS(−w0 ) ◦ S ◦ ϕw0 : f (D) → B(0, 1) satisfies G(0) = 0. Then the chain
rule gives
ρ′ (z0 ) = G′ (0)f ′ (z0 ).

We will show that G′ (0) > 1. To estimate G′ (0) we use the chain rule and the
differentiation formula (4.1):

G′ (0) = ϕ′S(−w0 ) (S(−w0 ))S ′ (−w0 )ϕ′w0 (0)

1 − |S(−w0 )|2 ′
= S (−w0 )(1 − |w0 |2 )
(1 − |S(−w0 )|2 )2

Since S(w)2 = w, we have 2S(w)S ′ (w) = 1 and |S(w)|2 = |w|. Thus

1 − |w0 | 1 1 + |w0 |
G′ (0) = 2
p (1 − |w0 |2 ) = p .
(1 − |w0 |) 2 |w0 | 2 |w0 |

Finally, since
√
( x− √1 )2
1 + x2 x
=1+ >1 for all x ∈ (0, 1) ,
2x 2

one has G′ (0) > 1. Thus |ρ′ (z0 )| = |G′ (0)| |f ′ (z0 )| > |f ′ (z0 )|. The estimate (4.2)
follows.
This completes the proof of the Riemann mapping theorem.

62
4.26. Remark. The argument in Step 3 probably seems miraculous and it is not
clear where it came from. Here is a short explanation. The map G was conformal
f (D) → B(0, 1), and therefore it has a conformal inverse H = G−1 : B(0, 1) → f (D).
Since f (D) ⊂ B(0, 1) the map H is conformal from B(0, 1) to itself. By the Schwarz
lemma (Theorem 5.17), there are only two possibilities for such a map H:

• either H is a rotation and |H ′ (0)| = 1;

• or one has |H(z)| < |z| and |H ′ (0)| < 1.

Since G involves a square root S, the map H cannot be a rotation and one must
have |H ′ (0)| < 1. Therefore necessarily |G′ (0)| = 1/|H ′ (0)| > 1.
A deeper explanation for this phenomenon is that the Möbius transformations
ϕw0 and ϕS(−w0 ) preserve areas with respect to the hyperbolic metric in B(0, 1). The
square root function S must instead increase the area, because the square func-
tion w 7→ w2 shrinks the area. This last fact follows from a geometric interpreta-
tion of the Schwarz lemma in terms of the hyperbolic metric: any conformal map
H : B(0, 1) → B(0, 1) is either a Möbius transformation that preserves hyperbolic
distances, or it is a contraction that strictly decreases hyperbolic distances. This is
another explanation for the fact that |G′ (0)| > 1 in Step 3.

63
5. Extended complex plane and Möbius transformations

5.1. Extended complex plane

Often it is convenient to add the point at infinity ∞ ̸∈ C to the complex plane

C. The resulting set
Ĉ = C ∪ {∞}
is called the extended complex plane. Algebraic operations in Ĉ are defined similar
to the extended real number set:


 ∞ ± z = z ± ∞ = ∞, when z ∈ C

∞ · z = z · ∞ = ∞ , when z ∈ Ĉ \ {0}
z
 = 0, when z ∈ C
∞


z
0
= ∞, when z ∈ Ĉ \ {0} .

Note that operations

∞ 0
∞ − ∞, , and 0·∞
∞ 0
are not defined at all (and thus should not be used).
Traditionally, the extended complex plane is geometrically visualized as a three-
dimensional sphere, called the Riemann sphere, as follows: Let

S = {(u, v, w) ∈ R3 : u2 + v 2 + w2 = 1}

be the unit sphere in R3 , and let’s identify the complex plane C with the plane
{(x, y, 0) : x, y ∈ R}. Consider the straight lines L passing through the “north pole”
N = (0, 0, 1) of S. If L is not in the tangent plane of S, then it intersects the spherical
shell of S at exactly one point P = (u, v, w) ̸= N , and the plane C at exactly one
point
u v
π(P ) = π(u, v, w) = ( , , 0) .
1−w 1−w
This yields a bijection π : S \ {N } → C, called the stereographic projection. Setting
π(N ) = ∞ establishes correspondence between the extended complex plane Ĉ and
the sphere S. The inverse map π −1 can be easily determined:

2Re(z) 2Im(z) |z|2 − 1

π −1 (z, 0) = ( , , ).
|z|2 + 1 |z|2 + 1 |z|2 + 1

64
 As an exercise, it is fairly easy to verify that in the stereographic projection, arcs
passing through the north pole correspond to straight lines in the complex plane,
and circles in the complex plane correspond to circles on the spherical shell. For this
reason, circles and lines (extended to include the point at infinity ∞) in C are called
generalized circles of the extended complex plane Ĉ.
Half-planes and disks in C correspond to spherical caps on the Riemann sphere.
Additionally, the stereographic projection is ’conformal’, meaning it preserves angles.
Note that the point at infinity ∞ on the Riemann sphere S = Ĉ holds the same
position as any other point, so by rotating the point at infinity to another location,
one can get a good understanding of the local properties of the extended complex
plane. For our purposes, it suffices to define the ∞-centered disk as

1
B(∞, r) = ∞ ∪ {z ∈ C : |z| > } .
r

N = (0, 0, 1)

2y |z|2 −1
P = (u, v, w) = ( |z|2x
2 +1 , ,
|z|2 +1 |z|2 +1
)

(0, 1, 0)
y
(1, 0, 0)

u v
z = (x, y, 0) = ( 1−w , 1−w , 0)

Riemann sphere and stereographic projection.

5.2. Möbius transformations

5.1. Definition. A map f : Ĉ → Ĉ

az + b
f (z) = , where a, b, c, d ∈ C , ad − bc ̸= 0 ,
cz + d

65
 3
is called a Möbius transformation. Here we use the conventions:

az + b
f (z) = , when z ∈ C

If c = 0 , then d
f (∞) = ∞.
az + b

f (z) = cz + d ,

 when z ∈ C \ {− dc },
If c ̸= 0 , then f (− d ) = ∞

 c
f (∞) = ac .


One can easily prove:

5.2. Theorem. A Möbius transformation f : Ĉ → Ĉ is a homeomorphism and a

conformal map on C \ {− dc } (thus on C if c = 0).

5.3. Theorem. Möbius transformations

G = {f : Ĉ → Ĉ : f is a Möbius transformation}

form a group, where the operation is composition of maps.

Proof: Straightforward calculation (exercise). The inverse map of a Möbius trans-

formation
az + b
f (z) =
cz + d
is
dz − b
f −1 (z) = .
−cz + a

Note that the group of Möbius transformations is not commutative, i.e., there
exist Möbius transformations f, g for which f ◦ g ̸= g ◦ f (exercise).
3
A.F. Möbius, 1790–1868. Often Möbius transformations are referred to as linear fractional
transformations.

66
5.4. Remark. Möbius transformations
az + b
f (z) =
cz + d

and complex (2 × 2) matrices

 
a b
A= , det(A) = ad − bc ̸= 0 ,
c d

correspond to each other such that all matrices

 
λa λb
λA = , det(λA) = λ2 det(A) = λ2 (ad − bc) ̸= 0 ,
λc λd

correspond to the same Möbius transformation

az + b
f (z) = .
cz + d

The one-to-one correspondence can be achieved, for example, by normalizing (2 ×

2) matrices such that their determinant is = 1. Note that matrix multiplication
corresponds to composition of Möbius transformations (exercise).

5.5. Remark (Elementary Möbius transformations). A Möbius transformation

f (z) = z + w , w∈C

is called a translation. It translates the points of the complex plane C by the amount
w and keeps the point at infinity fixed f (∞) = ∞.
Another significant type of elementary Möbius transformations are the mappings

g(z) = λz , where λ ∈ C , λ ̸= 0 .

If |λ| = 1, it is a rotation of the complex plane: g rotates the points of the complex
plane by the angle arg(λ) and keeps the point at infinity fixed g(∞) = ∞.
If λ ∈ R and λ > 0, then the mapping g is a dilation: expansion if λ ≥ 1 and
contraction if 0 < λ ≤ 1.
The general case λ ∈ C, λ ̸= 0, is obtained by combining rotation and dilation, as

λ
λ = |λ| .
|λ|

67
The point at infinity remains fixed, g(∞) = ∞.
The third type of elementary Möbius transformations is inversion:

1
h(z) = .
z

Inversion swaps the point at infinity and the origin: h(∞) = 0 and h(0) = ∞.
Inversion can be obtained by reflecting the point z with respect to the circumference
∂B(0, 1) of the unit disk and then with respect to the real axis (taking the conjugate),
or vice versa:
1 z z̄
h(z) = = 2 = 2 ,
z |z| |z̄|
from which it follows that

1
|h(z)| = and arg(h(z)) = arg(z̄) = − arg(z) .
|z|

All Möbius transformations can be obtained by combining translation, dilation,

rotation, and inversion.

5.6. Theorem. Every Möbius transformation can be obtained by combining trans-

lation z 7→ z + w, mapping z 7→ λz, and inversion z 7→ z1 .

Proof: When ad − bc ̸= 0, then

bc − ad a

az + b

 d
+ , if c ̸= 0,
= c 2 (z + )
c
c
cz + d  az + b , if c = 0,
d d

which are of the required type.

5.7 . Theorem. Möbius transformations map generalized circles to generalized

circles.

Proof: Straight lines in the complex plane are of the form

(5.1) Bz + B̄ z̄ + c = 0 , where B ∈ C , c ∈ R,

68
which can be seen by substituting the equation of a line ax + by + c = 0 in R2 into
the equation z = x + iy, where 2B = a − ib. Furthermore, the equation of a circle

|z − z0 | = r

can be written as

(5.2) z z̄ + B̄z + B z̄ + c = 0 , where B = −z0 ∈ C , c = |B|2 − r2 ∈ R .

Clearly, lines are preserved as lines and circles as circles under the mappings z 7→
z + w and z 7→ λz. If z lies on the line (5.1), then its image point w = 1/z under
the inversion z 7→ 1/z satisfies the equation

cww̄ + B w̄ + B̄w = 0 ,

which, for c = 0, corresponds to a line passing through the origin. If c ̸= 0, then

according to equation (5.2), this is a circle with radius |B|/|c| and center −B/c.
Similarly, if a point z lies on the circle (5.2), then its image point w = 1/z under
the inversion satisfies the equation

cww̄ + Bw + B̄ w̄ + 1 = 0 ,

which, for c = 0 (meaning the circumference passes through the origin), corresponds
to the equation of a line. For other values of c, this is the equation of a circle.

Since Möbius transformations are homeomorphisms, we obtain:

5.8. Corollary. A Möbius transformation maps every open set that is either

- a disk B(z0 , r),

- the exterior of a disk Ĉ \ B̄(z0 , r), or

- a half-plane determined by a line in C,

to a set that is of one of the above types.

69
 A point z0 ∈ Ĉ for which f (z0 ) = z0 is called a fixed point of the transformation
f.

5.9. Theorem. A Möbius transformation f : Ĉ → Ĉ has either 1 or 2 fixed points

unless f is the identity map f (z) = z for all z ∈ Ĉ.

Proof: Exercise. Determine the number of solutions of the second degree equation

az + b
=z
cz + d

(be careful with the case z = ∞). Another way is to use Theorem 5.10 below.

In the definition of a Möbius transformation, the images of three points can be

specified, but not more.

5.10. Theorem. Let z1 , z2 , z3 ∈ Ĉ be three distinct points. If w1 , w2 , w3 ∈ Ĉ are

three distinct points, then there exists exactly one Möbius transformation f : Ĉ →
Ĉ, for which
f (z1 ) = w1 , f (z2 ) = w2 and f (z3 ) = w3 .

Proof: We can assume (why?) that

w1 = 1 , w2 = 0 and w3 = ∞ .

Existence: The sought transformation is (exercise: prove this!)

(z − z2 )(z1 − z3 )
f (z) =
(z − z3 )(z1 − z2 )

Note: if z1 , z2 , or z3 is ∞, then (respectively)

z − z2 z1 − z3 z − z2
f (z) = , or .
z − z3 z − z3 z1 − z2

Uniqueness: Let g be another such Möbius transformation. Then

az + b
f ◦ g −1 (z) =
cz + d

70
is a Möbius transformation, for which

−1
f ◦ g (0) = 0 ,
 implying b = 0
f ◦ g −1 (∞) = ∞ , implying c = 0,

f ◦ g −1 (1) = 1 , implying ad = 1 .


Thus,
f ◦ g −1 (z) = z for all z ∈ Ĉ ,

so f = g and the claim is proved.

The formula that defines the mapping f is called the cross ratio:

5.11. Definition. Let z1 , z2 , z3 , z4 ∈ Ĉ be four distinct points. The quantity

(z1 − z3 )(z2 − z4 )
[z1 , z2 , z3 , z4 ] =
(z1 − z2 )(z3 − z4 )

is called the cross ratio of the points z1 , z2 , z3 , z4 .4

The cross ratio is invariant under Möbius transformations.

5.12. Theorem. Let z1 , z2 , z3 , z4 ∈ Ĉ be four distinct points and f : Ĉ → Ĉ a

Möbius transformation. Then

[z1 , z2 , z3 , z4 ] = [f (z1 ), f (z2 ), f (z3 ), f (z4 )] .

Proof: Let
g(z) = [z1 , z2 , z3 , z]

be the unique Möbius transformation (Theorem 5.10), which maps the points z1 , z2 ,
and z3 (respectively) to points 1, 0, and ∞. Then g ◦ f −1 maps the points f (z1 ),
f (z2 ), and f (z3 ) (respectively) to points 1, 0, and ∞, so

[z1 , z2 , z3 , z4 ] = g(z4 ) = g ◦ f −1 (f (z4 )) = [f (z1 ), f (z2 ), f (z3 ), f (z4 )] ,

where the last equality follows from the uniqueness part of Theorem 5.10.
4
Warning: do not confuse this notation with a fraction. The cross ratio can be written in many
different orders, so always check from your source what the current definition is.

71
5.13 . Remark. With Theorem 5.12, we can quickly find the (unique) Möbius
transformation f that maps the given three distinct points z1 , z2 , z3 ∈ Ĉ to the
given three distinct points w1 , w2 , w3 ∈ Ĉ: solve for f (z) from the cross ratio

[w1 , w2 , w3 , f (z)] = [z1 , z2 , z3 , z] .

Example. Let’s construct a Möbius transformation f such that 0 7→ 0, 1 7→ 1, and

2 7→ i. Calculate

(1 − i)(−f (z)) (1 − 2)(−z)

[1, 0, i, f (z)] = [1, 0, 2, z] or = ,
(i − f (z)) 2−z

yielding
iz
f (z) = .
z(2 − i) − 2 + 2i

Note that in the light of Theorems 5.7 and 5.8, f maps the real axis to the circle
determined by the points 0, 1, and i. Furthermore, since f (−i) = −1/3, f maps
(bijectively) to the upper half-plane the disk bounded by the aforementioned circle

√
1 1 2
B( + i, ).
2 2 2

Let
C = {z ∈ C : |z − z0 | = r}

be a circle centered at z0 with radius r. If z ∈ Ĉ, then its reflection point with respect
to the circle C is
r2


 z0 + , if z ̸= z0 , ∞
∗
 z̄ − z¯0
z = ∞, if z = z0


if z = ∞ .

z0 ,

Sometimes, the points z and z ∗ are said to be symmetric with respect to the circle
C.

72
 z∗

C
z

z0 r

5.14. Theorem. Points z and w are symmetric with respect to the circle C if and
only if
[w, z1 , z2 , z3 ] = [z, z1 , z2 , z3 ] for all z1 , z2 , z3 ∈ C .

Proof: First, it is observed that

r2 (zj − z0 )(z̄j − z¯0 )

zj∗ = z0 + = z0 + = zj , when zj ∈ C ,
z̄j − z¯0 z̄j − z¯0

implying
r2 (z̄j − z̄)
z ∗ − zj = ,
(z̄j − z¯0 )(z̄ − z¯0
and therefore
[z ∗ , z1 , z2 , z3 ] = [z, z1 , z2 , z3 ] .

The proof of the converse direction is an exercise. Note that any three distinct
points z1 , z2 , z3 ∈ C uniquely determine the reflection point.

5.15 . Corollary. Let f be a Möbius transformation and C a generalized circle.

Then, the image points f (z) and f (z ∗ ) of symmetric points z and z ∗ with respect
to circle C are symmetric with respect to the generalized circle f (C).

73
 Example. Let’s construct a Möbius transformation f that maps the unit disk
B(0, 1) to itself, f (0) = 0, and f (1) = i. Note that since the origin remains fixed, its
symmetric point ∞ also remains fixed. Hence,

[i, 0, ∞, f (z)] = [1, 0, ∞, z] ,

implying
f (z) z
= or f (z) = iz .
i 1

5.3. Möbius transformations as conformal mappings

We demonstrate that conformal mappings of the disk to itself are restrictions of

Möbius transformations to the disk.
We start with the following result.

5.16. Theorem. Let z0 ∈ B(0, 1), λ ∈ C, |λ| = 1. Then the Möbius transformation

z − z0
f (z) = λ
1 − z z¯0

defines a (conformal) bijection f : B(0, 1) → B(0, 1) with f (z0 ) = 0.

Proof: Note that f is a Möbius transformation since

λ(1 − z0 z¯0 ) = λ(1 − |z0 |2 ) ̸= 0 .

Therefore, since f (z0 ) = 0, it suffices to check that

f (∂B(0, 1)) ⊂ ∂B(0, 1) .

Indeed, when |z| = 1, we have

|z − z0 | |z − z0 | |z − z0 | |z − z0 |
|f (z)| = |λ| = z̄ = = = 1.
|1 − z z¯0 | |z|| |z|2 − z¯0 | |z̄ − z¯0 | |z − z0 |

74
 We still need the Schwarz lemma:

5.17. Theorem. If f : B(0, 1) → B(0, 1) is analytic with f (0) = 0, then

|f ′ (0)| ≤ 1 and |f (z)| ≤ |z| for all z ∈ B(0, 1).

Moreover, if there exists z0 ∈ B(0, 1), z0 ̸= 0, such that |f (z0 )| = |z0 |, or if |f ′ (0)| =
1, then there exists λ ∈ C, |λ| = 1, such that

f (z) = λz for all z ∈ B(0, 1).

Proof: Proven in exercises: immediately follows when applying the maximum mo-
dulus principle CA1.5.16 to the function

 f (z)
, when z ̸= 0,
g(z) = z
f ′ (0) , when z = 0 ,

which is analytic in the entire disk B(0, 1).

5.18. Theorem. Let f : B(0, 1) → B(0, 1) be a conformal bijection and z0 = f −1 (0).

Then there exists λ ∈ C, |λ| = 1, such that

z − z0
f (z) = λ for all z ∈ B(0, 1) .
1 − z z¯0

Proof: Let
z − z0
g(z) = ,
1 − z z¯0
then by Theorem 5.16, g defines a conformal bijection of the disk to itself with
g(z0 ) = 0. Hence, h = f ◦ g −1 satisfies the assumptions of Schwarz lemma 5.17, so

|h(z)| ≤ |z| for all z ∈ B(0, 1) .

Similarly, h−1 satisfies the assumptions of Schwarz lemma 5.17, so

|h−1 (w)| ≤ |w| for all w ∈ B(0, 1) .

In particular, when z = h−1 (w), we have h(z) = w and

|h(z)| ≤ |z| = |h−1 (w)| ≤ |h(z)|

75
thus
|h(z)| = |z| for all z ∈ B(0, 1) .
Therefore, by a special case of Schwarz lemma, we have
h(z) = λz ,
and consequently
z − z0
f (z) = h(g(z)) = λg(z) = λ ,
1 − z z¯0
as claimed.

5.19. Corollary. If B is a disk, then every conformal mapping f : B → B is a

restriction of a Möbius transformation to the disk B.

Conformal mappings of the entire plane are affine mappings:

5.20. Theorem. Let f : C → C be an entire injective function. Then there exist

a, b ∈ C with a ̸= 0 such that
f (z) = az + b for all z ∈ C .
In particular, f is a Möbius transformation and a conformal bijection.

Proof: First, we show that f is a polynomial. If f is not a polynomial, then the

point at infinity is its essential singularity (Theorem 3.11), i.e., the origin is an
essential singularity of the function
1
z 7→ f ( )
z
By the Casorati-Weierstrass theorem 3.10, the set
f (C \ B(0, 1))
is dense in C, so by the open mapping theorem (Theorem 4.13), f (B(0, 1)) intersects
it,
f (C \ B(0, 1)) ∩ f (B(0, 1)) ̸= ∅ .
Thus, there exist points z1 ∈ B(0, 1) and z2 ̸∈ B(0, 1) such that contrary to the
injectivity of f ,
f (z1 ) = f (z2 ) .
Hence, f is a polynomial.
By the fundamental theorem of algebra (Theorem CA1.5.15) and Theorem 4.11,
an injective polynomial is of degree 1, so the claim follows.

76