Complex Analysis I

Fall 2014

These are notes for the graduate course Math 5283 (Complex Analysis I) taught by
Dr. Anthony Kable at the Oklahoma State University (Fall 2014). The notes are taken
by Pan Yan ([email protected]), who is responsible for any mistakes. If you notice any
mistakes or have any comments, please let me know.

Contents
1 Complex Numbers (08/18) 4

2 Differentiability and the Cauchy-Riemann Equations (08/20) 5

3 Power Series (08/22) 7

4 Sin, Cos and Exp (08/25) 9

5 Preliminary Results on Holomorphic Functions I (08/27) 11

6 Preliminary Results on Holomorphic Functions II (08/29) 13

7 Preliminary Results on Holomorphic Functions III (09/03) 14

8 Preliminary Results on Holomorphic Functions IV (09/05) 17

9 Elementary Results on Holomorphic Functions I (09/08) 21

10 Elementary Results on Holomorphic Functions II (09/10) 23

11 Elementary Results on Holomorphic Functions III (09/15) 24

12 Elementary Results on Holomorphic Functions IV (09/17) 26

13 Logarithms, Winding Numbers and Cauchy’s Theorem I (09/19) 29

1
14 Logarithms, Winding Numbers and Cauchy’s Theorem II (09/22) 31

15 Logarithms, Winding Numbers and Cauchy’s Theorem III (09/24) 33

16 Logarithms, Winding Numbers and Cauchy’s Theorem IV (09/26) 36

17 Logarithms, Winding Numbers and Cauchy’s Theorem V (09/29) 38

18 Logarithms, Winding Numbers and Cauchy’s Theorem VI (10/1) 40

19 Logarithms, Winding Numbers and Cauchy’s Theorem VII (10/3) 42

20 Counting Zeroes and the Open Mapping Theorem I (10/6) 45

21 Counting Zeroes and the Open Mapping Theorem II (10/8) 46

22 Product Formula for Sine Function I (10/13) 49

23 Product Formula for Sine Function II (10/15) 52

24 Product Formula for Sine Function III (10/17) 54

25 Inverses of Holomorphic Functions I (10/20) 57

26 Inverses of Holomorphic Functions II (10/22) 58

27 Conformal Mappings I (10/27) 60

28 Conformal Mappings II (10/29) 61

29 Conformal Mappings III (11/1) 62

30 Conformal Mappings IV (11/3) 65

31 Conformal Mappings V (11/5) 65

32 Conformal Mappings VI (11/10) 68

33 Normal Families and the Riemann Mapping Theorem I (11/12) 72

34 Normal Families and the Riemann Mapping Theorem II (11/14) 72

35 Normal Families and the Riemann Mapping Theorem III (11/17) 73

36 Normal Families and the Riemann Mapping Theorem IV (11/19) 74

2
37 Normal Families and the Riemann Mapping Theorem V (11/21) 77

38 Normal Families and the Riemann Mapping Theorem VI (11/24) 78

39 Normal Families and the Riemann Mapping Theorem VII (12/1) 80

40 Topology of Uniform Convergence on Compact Subsets (12/3) 82

41 Determination of the Automorphism Groups Aut(A(0, r, R)) (12/5) 84

3
1 Complex Numbers (08/18)
The complex field C is R2 with a pair of operations that makes it a field:
(x1 , y1 ) + (x2 , y2 ) = (x1 + x2 , y1 + y2 ), (x1 , y1 ) · (x2 , y2 ) = (x1 x2 − y1 y2 , x1 y2 + x2 y1 ).
The point is that we have two elements: 1 = (1, 0), i = (0, 1), and
i2 = (0, 1) · (0, 1) = (−1, 0) = −1.
Any complex number can be expressed uniquely as x + yi where x, y ∈ R. If z = x + yi
with x, y ∈ R, then
p
Re(z) = x, Im(z) = y, |z| = x2 + y 2 , z = x − yi.
Conjugates are automorphism of the field C since
z + w = z + w, zw = z · w.
Notice that zz = |z|2 .
We use the metric d(z, w) = |z − w| on C to make it a metric space. Note this is the
standard d2 metric on R2 . So we already know that C is a complete metric space.
Example 1.1. Describe the set {z ∈ C : |z| + |z − 1| = 2}. It is an ellipse with foci at 0
and 1, major axis the real axis, and one vertex at 3/2.
If we switch to polar coordinates, then every non-zero complex number gets a unique
representation in the form (r cos(θ), r sin(θ)), where r > 0 and θ ∈ [0, 2π). By definition,
eit = cos(t) + i sin(t). So the polar representation can be expressed as z = reiθ with r > 0
and θ ∈ [0, 2π). This reveals some information about multiplication:
r1 eiθ1 · r2 eiθ2 = r1 r2 ei(θ1 +θ2 ) .
This tells us that the map Mw : C → C given by Mw (z) = wz consists of the composi-
tion of a homothety with factor |w| and a rotation anticlockwise by the angle θ such that
w = |w|eiθ . Note the map S : C → C given by S(z) = z is a reflection.
Example 1.2. Determine all solutions to the equation z 4 = i with z ∈ C. Write the
solutions in the form x + iy. Use polar representation, say z = reiθ , then z 4 = r4 e4iθ = i.
π
This tells us that r4 = 1 and r = 1. Thus e4iθ = i = e 2 i . Thus,
π
4θ = + 2kπ, k ∈ Z,
2
π 1
θ = + kπ, k ∈ Z.
8 2
The θ in the polar representation is an argument of the complex number. If z = reiθ
then arg(z) = θ. Note that arg is not a function — it has many values. In Complex Made
Simple (CMS), the principle argument is defined to be the unique argument that lies in
(−π, π]. It is denoted by Arg(z).

4
2 Differentiability and the Cauchy-Riemann Equations (08/20)
We consider functions f : C → C. These can be regarded as functions from R2 to R2 .

Definition 2.1. Let V ⊂ C be an open set, f : V → C, and p ∈ V . Then f is complex

differentiable at p if the limit
f (p + h) − f (p)
lim
h→0 h
exists.

If the limit exists, then it will be some complex number w. Then we will have

(2.1) f (z) = f (p) + w(z − p) + E(z)

where E(p) = 0, and limz→p E(z)z−p = 0. Conversely, if we have an expression like (2.1) then
f is differentiable at p and its derivative at p is w.
Recall that a linear map T : R2 → R2 can be represented by a 2 × 2 matrix [T ]. The
column of [T ] are T (e1 ) and T (e2 ) and it has the property that T (v) = [T ]v. What if we
think of R2 as C and consider the linear map Mw : C → C given by Mw (z) = wz? What
is [Mw ]?
Let w = a + bi. In C the standard basis is 1, i. Then

Mw (1) = w × 1 = w = a + bi,

Mw (i) = w × i = (a + bi)i = −b + ai,

and so  
a b
[Mw ] = .
−b a
2 2
Among all linear maps T : R → R, those that are complex linear are precisely the ones
a b
whose matrix has the form .
−b a
Say we have an open set V ⊂ C, p ∈ V , f : V → C. Let f = u + iv, where u : V → R
and v : V → R. Say that f is real differentiable at p. Then
   
x x
(2.2) f (x + yi) = f (p) + T ( − p) + E( )
y y

where E(p) = 0 and E is sublinear at p. We know that ux (p), uy (p), vx (p), vy (p) all exist.
Moreover,  
ux (p) uy (p)
[T ] =
vx (p) vy (p)
is the Jacobian matrix.

5
 Note that (2.2) would say that f is (complex) differentiable at p if T is actually complex

a −b
linear. We just found out that this happens when the matrix of T has the form
b a
and, in that case, it is multiplication by a + bi. This means that f is complex differentiable
at p if

(2.3) ux (p) = vy (p)

and

(2.4) uy (p) = −vx (p)

When this happens, f 0 (p) = ux (p) + vx (p)i. Equations (2.3) and (2.4) are called
Cauchy-Riemann equations (C-R equations) . We have the exact equivalence:
   
real differentiable at p and satisfy CR equations ⇔ complex differentiable at p

Recall from Advanced Calculus: If f = u + iv : V → C, p ∈ V , ux , uy , vx , vy all exist on

V , and are continuous at p, then f is real differentiable at p. In particular, if u, v : V → R
are C 1 and satisfy CR equations, then u + iv is complex differentiable at every point of V .

6
3 Power Series (08/22)
Definition 3.1. A power series is a series of the form
∞
X
cn (z − z0 )n .
n=0

Here c0 , c1 , c2 , · · · ∈ C, z, z0 ∈ C, and we have the convention that (z − z0 )0 means 1.

Theorem 3.2. Given a power series ∞ n
P
n=0 cn (z − z0 ) we define

R = sup{r ≥ 0 : (cn rn ) is bounded}.

The power series is uniformly absolutely convergent on any compact subset of the disk
D(zo , R). (Note that R ∈ [0, ∞) ∪ {∞}, D(zo , R) = {z ∈ C : |z − z0 | < R}.)
Proof. Let the compact set be K ⊂ D(z0 , R). Then there is a number 0 < r < R such
that K ⊂ D(z0 , r) (compact: each open cover has a finite subcover, pick the largest one),
unless R = 0. Note that R = 0 is trivial, and so we shall assume that R > 0 and choose
such a r. It suffices to show that the series converges uniformly absolutely on D(z0 , r).
There is a number ρ such that r < ρ < R and (cn ρn ) is bounded. Let M be a bound for
(cn ρn ). Thus |cn ρn | ≤ M for all n ≥ 0. Equivalently, |cn | ≤ M ρ−n for all n ≥ 0. Suppose
z ∈ D(z0 , r). Then
r r
|cn (z − z0 )n | ≤ |cn rn | = |cn ρn ( )n | ≤ M ( )n .
ρ ρ
r P∞ r n
Note 0 < ρ < 1. The (geometric) series n=0 M ( ρ ) converges. By the Direct Com-
parison Test, ∞
P n
P∞ n
n=0 |cn (z − z0 ) | converges uniformly on D(z0 , r). Thus n=0 cn (z − z0 )
converges uniformly absolutely on D(z0 , r).
We know that the sum of a power series on D(z0 , R) is continuous (Uniform Limit
Theorem: the uniform limit of any sequence of continuous functions P is continuous). In
fact, a power series is differentiable on D(z0 , R). Say that f (z) = ∞ c
n=0 n (z − z 0 )n with

z ∈ D(z0 , R). Suppose that w ∈ D(z0 , R). Then

∞
X
f (w) = cn (w − z0 )n
n=0
and so
∞
X
f (z) − f (w) = cn [(z − z0 )n − (w − z0 )n ]
n=0
X∞
cn [(z − z0 ) − (w − z0 )] (z − z0 )n−1 + (z − z0 )n−2 (w − z0 ) + · · · + (w − z0 )n−1
 
=
n=0
∞
X n−1
X
= (z − w) cn (z − z0 )n−1−j (w − z0 )j .
n=0 j=0

7
If z 6= w then we get
∞ n−1
f (z) − f (w) X X
= cn (z − z0 )n−1−j (w − z0 )j .
z−w
n=0 j=0

Let
∞
X n−1
X
Φ(z, w) = cn (z − z0 )n−1−j (w − z0 )j .
n=0 j=0

If we set z = w in Φ we get
∞
X ∞
X
Φ(z, z) = cn n(z − z0 )n−1 = cn n(z − z0 )n−1 .
n=0 n=1

To conclude that f is differentiable and that f 0 (z) = ∞ n−1 , all we need

P
n=1 cn n(z − z0 )
is that Φ : D(z0 , R) × D(z0 , R) → C is continuous. We can show that this P∞is so rbyn−1 using
the Weierstrass M-test. We can do this by comparison with a series n=1 n{ p } as
before.

Corollary 3.3. Say that f (z) = ∞ n

P
n=0 cn (z − z0 ) is a power series with positive radius
of convergence R. Then f is infinitely differentiable on D(z0 , R), we have

f (n) (z0 )
cn =
n!
for n ≥ 0, and f uniquely determines (cn ).

Definition 3.4. Let ∞

P P∞
n=0 an , n=0 bn be two complex series. The Cauchy product is
defined as
∞
X ∞
X ∞
X
cn = an bn
n=0 n=0 n=0
Pn
where cn = j=0 aj bn−j .

8
4 Sin, Cos and Exp (08/25)
For z ∈ C, we define exp(z) = ∞ 1 n 1 (n+1) | = |z| | 1 z n | and so
P
n=0 n! z . Note that | (n+1)! z n+1 n!
1 n 1 n
the sequence (| n! z |) is decreasing once n + 1 > |z|. It follows that (| n! z |) is always
bounded. Thus R = ∞ for this power series and so we get an infinitely differentiable
function exp : C → C. We have
∞
0
X n n−1
exp (z) = z
n!
n=1
∞
X 1
= z n−1
(n − 1)!
n=1
∞
X 1 n
= z
n!
n=0
= exp(z).

Fix w ∈ C and consider g : C → C given by g(z) = exp(−z) · exp(z + w). By the

Product and Chain Rules, g is infinitely differentiable. Also

g 0 (z) = −exp(−z) · exp(z + w) + exp(−z) · exp(z + w) = 0

for all z ∈ C. Two possibilities to show that g 0 (z) = 0 ⇒ g is constant: One is to use
the Mean Value Inequality (kφ(x) − φ(y)k ≤ supp∈[x,y] kDφ(p)k · kx − yk) from Advanced
Calculus, since g 0 (z) is just a special case of the Fréchet derivative. Second possibil-
ity is to show that g is given by  a power
 series with center 0. (Note: exp(z + w) =
P∞ 1 n
P ∞ 1 P n n j n−j = · · · ) and then use the Cauchy product
n=0 n! (z + w) = n=0 n! j=0 j z w

and the corollary. We deduce that g is constant. So g(z) = g(0) for all z. That is,
exp(−z) · exp(z + w) = exp(−0) · exp(0 + w) = exp(w) for all z, w ∈ C. Set w = 0,
we get exp(−z) · exp(z) = 1, and so exp(−z) = exp(z)−1 . Now multiply the identity
exp(−z) · exp(z + w) = exp(w) by exp(z) to obtain

exp(z + w) = exp(z)exp(w), ∀z, w ∈ C.

Remark 4.1. This implies that exp(z) 6= 0 for all z ∈ C.

9
 For z ∈ C,
∞
X 1
exp(iz) = (iz)n
n!
n=0
∞
X 1 n n
= i z
n!
n=0
∞ ∞
X 1 2m 2m X 1
= i z + i2m+1 z 2m+1
(2m)! (2m + 1)!
m=0 m=0
∞ ∞
X (−1)m X (−1)m 2m+1
= z 2m + i z
(2m)! (2m + 1)!
m=0 m=0
def
= cos(z) + i sin(z).

The series imply that cos is even, sin is odd, and so

eiz + e−iz eiz − e−iz

cos(z) = , sin(z) =
2 2i
for all z ∈ C. Note that if y ∈ R then

ei(iy) + e−i(iy) e−y + ey

cos(iy) = = = cosh(y),
2 2

ei(iy) − e−i(iy) e−y − ey e−y − ey ey − e−y

sin(iy) = = = −i =i = i sinh(y).
2i 2i 2 2
Previously, we define eiθ = cos(θ)+i sin(θ). Now this is true by definition of cos, sin, exp.
This justifies the polar representation that we used before. Note if z = reiθ with r > 0,
then z = eln(r) · eiθ = eln(r)+iθ . This tells us that exp(C) = C − {0}.

Remark 4.2. For z = x + iy ∈ C, | exp(z)| = exp(x).

10
5 Preliminary Results on Holomorphic Functions I (08/27)
Definition 5.1. If X is a metric space then a curve in X is a map γ : [a, b] → X for
some a ≤ b that is continuous. The interval [a, b] is the parameter interval and γ([a, b]) is
denoted γ ∗ and called the trace of γ.

Definition 5.2. A curve γ : [a, b] → C is smooth (better “piecewise smooth”) if there is a

partition a = t0 < t1 < · · · < tn = b such that
(1) γ|(tj−1 ,tj ) is C 1 for 1 ≤ j ≤ n;
(2) The one-sided limits of γ 0 exist at each tj (0 ≤ j ≤ n).

Definition 5.3. Let V ⊂ C be open and γ : [a, b] → V a piecewise smooth curve. Let
f : V → C be a continuous function. Then we define
Z Z b
f (z) dz = f (γ(t))γ 0 (t) dt.
γ a

This is the path or contour integral of f over γ.

Remark 5.4. Note that f (γ(t))γ 0 (t) is Riemann integrable because it is bounded and the
set of discontinuities is at worse a finite set (hence a zero set). Also, if a = t0 < t1 <
· · · < tn = b is a suitable partition for γ then
Z b n Z
X tj
0
f (γ(t))γ (t) dt = f (γ(t))γ 0 (t) dt.
a j=1 tj−1

Lemma 5.5 (Cauchy’s Theorem for Derivatives). Let V ⊂ C be an open set, f : V → C

continuous and suppose that there is an F : V → C such that F 0 = f . Let γ be a piecewise
smooth curve in V . Then
Z
f (z) dz = F (γ(b)) − F (γ(a))
γ
R
where [a, b] is the parameter interval for γ. In particular, γ f (z) dz = 0 if γ is closed
(means γ(a) = γ(b)).

Proof. By the Chain Rule for vector functions, we have

d
(F (γ(t))) = F 0 (γ(t)) · γ 0 (t) = f (γ(t)) · γ 0 (t)
dt
for all t where the derivative γ 0 exists. Let a = t0 < t1 < · · · < tn = b be a suitable

11
partition. Then
Z n Z
X tj
f (z) dz = f (γ(t))γ 0 (t) dt
γ j=1 tj−1
n Z tj
X d
= (F (γ(t))) dt
dt
j=1 tj−1
Xn
= [F (γ(tj )) − F (γ(tj−1 ))]
j=1

= F (γ(b)) − F (γ(a)).

Definition 5.6. If γ : [a, b] → C is a piecewise smooth curve then the length of γ is

Z b
L(γ) = |γ 0 (t)| dt.
a

Lemma 5.7 (ML Inequality). Let V ⊂ C be an open set, f : V → C be continuous, and

suppose that |f (z)| ≤ M for all z ∈ γ ∗ , where γ : [a, b] → V is a piecewise smooth curve.
Then Z
f (z) dz ≤ M L(γ).
γ

Proof. We have
Z Z b
f (z) dz = f (γ(t))γ 0 (t) dz
γ a
Z b
≤ |f (γ(t))| γ 0 (t) dz
a
Z b
≤ M γ 0 (t) dz
a
Z b
=M γ 0 (t) dz
a
= M L(γ).

12
6 Preliminary Results on Holomorphic Functions II (08/29)
Definition 6.1. Let S ⊂ C. Define A1 (S) to be the free abelian group on the set of
piecewise smooth curves in S. That is to say,
d
X
A1 (S) = { nj γj |nj ∈ Z, γj is a piecewise smooth curve in S}.
j=1
R
If f ∈ C(S) then we have defined γ f (z) dz ∈ C for each piecewise smooth path in S.
If Γ = dj=1 nj γj then we can define Γ f (z) dz = dj=1 nj γj f (z) dz ∈ C.
P R P R

Definition 6.2. B1 (S) is defined as

Z
B1 (S) = {Γ ∈ A1 (S)| f (z) dz = 0 for all f ∈ C(S)}.
Γ

Remark 6.3. This is a subgroup of A1 (S).

Definition 6.4. The chains in S are elements of A1 (S)/B1 (S).
R R
Remark 6.5. Two chains Γ1 = Γ2 if Γ1 f (z) dz = Γ2 f (z) dz for all f ∈ C(S).
Example 6.6. Define

γ : [0, 2π] → C, γ(t) = exp(it),

γ1 : [0, π] → C, γ(t) = exp(it),
γ2 : [π, 2π] → C, γ(t) = exp(it).

We have γ = γ1 +̇γ2 in the group of chains in C. To verify, let f ∈ C(C). Then

Z Z 2π
f (z) dz = f (exp(it))i exp(it) dt
γ 0
Z π Z 2π
= f (exp(it))i exp(it) dt + f (exp(it))i exp(it) dt
Z0 Z π

= f (z) dz + f (z) dz
γ1 γ2
Z
= f (z) dz.
γ1 +̇γ2

Thus γ = γ1 +̇γ2 in the group of chains in C.

There is a map ∂ : A1 (S) → A0 (S), where A0 (S) is the free group on S. It satisfies
∂(γ) = γ(b) − γ(a) if [a, b] is the parameter integral of γ.
The critical thing we need is that ∂(γ) = 0 if γ ∈ B1 (S). If so then we can define the
∂(γ) for γ a chain.

13
Definition 6.7. A cycle is a chain in the kernel of ∂.

In the example, ∂(γ) = 1−1 = 0, ∂(γ1 +γ2 ) = ∂(γ1 )+∂(γ2 ) = [(−1)−(1)]+[1−(−1)] =

0.
If z, w ∈ C, then [z, w] denotes the curve

γ : [0, 1] → C

such that γ(t) = (1 − t)z + tw.

Example 6.8. Show that [w, z] = −̇[z, w].

Solution. [w, z] is the curve

γ1 : [0, 1] → C such that γ1 (t) = (1 − t)w + tz.

−̇[z, w] is the curve

γ2 : [0, 1] → C such that γ2 (t) = −[(1 − t)z + tw].

Let f ∈ C(C), then

Z Z 1
f (z) dz = f ((1 − t)w + tz) · γ10 (t) dt
γ1 0
Z 1
= f ((1 − t)w + tz) · (z − w) dt
0
Z 0
= f (uw + (1 − u)z) · (z − w) · (−1) du (change of variable)
1
Z 1
=− f ((1 − t)z + tw) · (w − z) dt
Z 0
= f (z) dz.
γ2

Thus [w, z] = −̇[z, w].

Definition 6.9. If a, b, c ∈ C then define [a, b, c] to be the triangle with vertices a, b, c.

By definition, ∂[a, b, c] = [a, b]+̇[b, c]+̇[c, a].

7 Preliminary Results on Holomorphic Functions III (09/03)

Theorem 7.1 (Cauchy-Goursat). Let V ⊂ C be an open set, and R f : V → C a (complex)
differentiable function. Suppose that T ⊂ V is a triangle. Then ∂T f (z) dz = 0.

14
 R
Proof.
R Suppose not. Take a triangle T ⊂ R V such that ∂T f (z) dz 6=1 0. 2 Let 3 η =4
| ∂T f (z) dz| =
6 0. Call T T0 and note that | ∂T0 f (z) dz| = η. Write T0 = T0 ∪T0 ∪T0 ∪T0
and note that
Z Z Z Z Z
f (z) dz ≤ f (z) dz + f (z) dz + f (z) dz + f (z) dz .
∂T0 ∂T01 ∂T02 ∂T03 ∂T04

Thus there is at least one j ∈ {1, 2, 3, 4} such that

Z
1
f (z) dz ≥ η.
∂T0j 4

Call Toj with this property T1 . We have

Z
1
f (z) dz ≥ η.
∂T1 4
We repeat this process to construct a sequence (Tn )∞
n=0 of triangles such that

(1) T0 ⊃ T1 ⊃ T2 · · ·
(2) diam(Tn ) = 2−n diam(T0 )
(3) L(∂Tn ) = 2−n L(∂T0 )
Z
(4) f (z) dz ≥ 4−n η.
∂Tn

From a general fact about complete metric space, we know that ∩∞ n=0 Tn = {p} for some
0
p ∈ V . Now f is differentiable at p and so f (z) = f (p) + f (p)(z − p) + E(z). Note that
d f 0 (p)(z−p)2
d
) = f 0 (p)(z − p) and so ∂Tn f (z) dz = ∂Tn E(z) dz.
R R
dz (f (p)z) = f (p) and dz ( 2
Let  > 0. Then there is some δ > 0 such that |E(z)| ≤ |z − p| for all z such that
|z − p| < δ. Choose an n such that Tn ⊂ D(p, δ). Then
Z Z
f (z) dz = E(z) dz
∂Tn ∂Tn
Z
≤ |E(z)| dz
∂Tn
≤  · diam(Tn ) · L(∂Tn )
=  · 2−n · diam(T0 ) · 2−n · L(∂Tn )
=  · diam(T0 ) · L(∂Tn ) · 4−n .
We conclude that for any  > 0 there is an n such that
η · 4−n ≤  · diam(T0 ) · L(∂T0 ) · 4−n .
Thus
η ≤  · diam(T0 ) · L(∂T0 )
for all  > 0. This implies η = 0, a contradiction. Thus no such triangle exists.

15
Proposition 7.2. RLet V ⊂ C be a convex open set and f : V → C be a continuous
function such that ∂T f (z) dz = 0 for any triangle T ⊂ V . Then there is a differentiable
function F : V → C such that F 0 = f .
R
Proof. Let z0 ∈ V . Define F (z) = [z0 ,z] f (w) dw. Note [z0 , z] ⊂ V and so this is well
defined. For small enough h ∈ C, z + h ∈ V and so
Z
F (z + h) = f (w) dw
[z0 ,z+h]

and
Z
F (z + h) − F (z) = f (w) dw
[z0 ,z+h]+̇[z,z0 ]
Z
= f (w) dw
[z0 ,z+h]+̇[z+h,z]+̇[z,z0 ]−̇[z+h,z]
Z
= f (w) dw
−̇[z+h,z]
Z
= f (w) dw.
[z,z+h]

Thus Z
F (z + h) − F (z) − f (z)h = (f (w) − f (z)) dw.
[z,z+h]

Hence
F (z + h) − F (z)
Z
1
− f (z) = (f (w) − f (z)) dw
h h [z,z+h]

for all small enough h ∈ C − {0}. It suffices to show

Z
1
lim (f (w) − f (z)) dw = 0
h→0 h [z,z+h]

Let  > 0. Choose δ > 0 such that if 0 < |h| < δ then z + h ∈ V and |f (w) − f (z)| < 
for all w ∈ [z, z + h]. If |h| < δ then
Z
1 1
(f (w) − f (z)) dw ≤ ·  · |h| = .
h [z,z+h] |h|

1
R
That confirms that limh→0 h [z,z+h] (f (w) − f (z)) dw = 0.

16
8 Preliminary Results on Holomorphic Functions IV (09/05)
Theorem 8.1 (Cauchy’s Theorem for Convex Sets). Let V ⊂ C be convex R and open and
f : V → C be differentiable on V . Let γ be a closed path in V . Then γ f (z) dz = 0.
R
Proof. By Cauchy-Goursat Theorem,
R if T is a triangle, T ⊂ V , then ∂T f (z) dz = 0.
Since V is convex, the fact that ∂T f (z) dz = 0 implies that there is some F : RV → C such
that F 0 = f . By Cauchy’s Theorem for Derivatives, if γ : [a, b] → V then γ f (z) dz =
F (γ(b)) − F (γ(a)) = 0 because γ(a) = γ(b).

Theorem
R 8.2 (Morera’s Theorem). Suppose V ⊂ C is open, f : V → C is continuous. If
∂T f (z)dz = 0 for every triangle T ⊂ V , then f ∈ H(V ).

Theorem 8.3 (Cauchy’s Integral Formula). Let V ⊂ C be open, r > 0, z0 ∈ V and

D(z0 , r) ⊂ V and also let γ : [0, 2π] → V be the path γ(t) = z0 + reit . Suppose that
f : V → C is complex differentiable and z ∈ D(z0 , r). Then
Z
1 f (w)
f (z) = dw.
2πi γ w − z

r
z0
z

Figure 1:

Proof. Choose ρ > 0 so small that D(z, ρ) ⊂ D(z0 , r) (see Figure 2). Let γρ : [0, 2π] →
D(z0 , r) be γρ (t) = z + ρeit . Let g : V − {z} → C be g(w) = fw−z
(w)
. Then g is differentiable
on V − {z}. Note that σ1 +̇σ2 +̇σ3 = γ −̇γρ (see Figure 3). This tells us that
Z Z Z Z Z
g(w) dw − g(w) dw = g(w) dw + g(w) dw + g(w) dw.
γ γρ σ1 σ2 σ3

17
 γ

γρ r
z0
z

Figure 2:

σ1
σ3

σ2

Figure 3:

R
We know that σ1 g(w) dw = 0 by Cauchy’s Theorem for Convex Sets (See Figure 4).
R R R R
Similarly, σ2 g(w) dw = σ3 g(w) dw = 0, and so γ g(w) dw = γρ g(w) dw. Now,
Z Z
f (w)
g(w) dw = dw
γρ γρ w−z
2π
f (z + ρeit ) it
Z
= iρe dt
0 ρeit
Z 2π
=i f (z + ρeit ) dt.
0

18
 convex set

Figure 4:

This means that

Z Z 2π
f (w)
dw = lim i f (z + ρeit ) dt
γ w−z ρ→0+ 0
Z 2π
=i f (z) dt
0
= 2πif (z)

because f is continuous at z and so f (z + ρeit ) uniformly converges in t to f (z) as ρ →

0+ .

Remark 8.4. Cauchy’s Integral Formula implies that every (complex) differentiable func-
tion can be represented on small disks by a power series.
1 1
=
w−z (w − z0 ) − (z − z0 )
1
w−z0
= z−z0
1 − w−z 0
∞  n
1 X z − z0 z − z0
= (provided < 1)
w − z0 w − z0 w − z0
n=0
∞
X 1 z − z0
= n+1
(z − z0 )n (provided < 1).
(w − z0 ) w − z0
n=0

19
This tells us that
Z
1 f (w)
f (z) = dw
γ w−z
2πi
∞ Z
1 X f (w)
= (z − z0 )n dw
2πi γ (w − z0 )n+1
n=0
∞
X
= cn (z − z0 )n
n=0

with Z
1 f (w)
cn = dw.
2πi γ (w − z0 )n+1
Note then Z
(n) n! f (w)
f (z0 ) = dw.
2πi γ (w − z0 )n+1

20
9 Elementary Results on Holomorphic Functions I (09/08)
Theorem 9.1 (Cauchy’s Integral Formula for Derivatives). Let V ⊂ C be open, z0 ∈ V ,
r > 0 such that D(z0 , r) ⊂ V , f : V → C be differentiable. Let γ : [0, 2π] → V be
γ(t) = z0 + reit . Then Z
(m) m! f (w)
f (z) = dw
2πi γ (w − z)m+1
for all m ≥ 0 and z ∈ D(z0 , r).
Proof. Choose ρ > 0 so small that D(z, ρ) ⊂ D(z0 , r). Let γρ (t) = z + ρeit . By the
argument in the proof of Cauchy’s Integral Formula we get
Z Z
m! f (w) m! f (w)
dw = dw
2πi γ (w − z)m+1 2πi γρ (w − z)m+1
= f (m) (z)

by the fact that z is in the center of the circle γρ∗ .

Theorem 9.2 (Cauchy’s Estimates). Let V ⊂ C be open, f ∈ H(V ), z ∈ V , r > 0 such

that D(z, r) ⊂ V . Suppose that |f (w)| ≤ M for all w ∈ ∂D(z, r). Then
M m!
f (m) (z) ≤
rm
for all m ≥ 0.
Proof. Let γ : [0, 2π] → V be γ(t) = z + reit . Then
Z
m! f (w)
f (m) (z) = dw.
2πi γ (w − z)m+1

We assume that |f (z)| ≤ M for all w ∈ γ ∗ . Also, |w − z| = r for all w ∈ γ ∗ . Thus

f (w)
(w−z)m+1
M
≤ rm+1 for all w ∈ γ ∗ . The ML-inequality says that

m! M
f (m) (z) ≤ L(γ)
2π rm+1
m! M
= 2πr
2π rm+1
M m!
= m .
r

Notation. H(V ) is the set of all functions holomorphic on V .

Definition 9.3. The elements of H(C) are called entire functions.

21
Remark 9.4. If f is an entire function, and z0 ∈ C then
∞
X f (m) (z0 )
f (z) = (z − z0 )m
m!
m=0

for all z ∈ C.

Theorem 9.5 (Liouville’s Theorem). A bounded entire function is constant.

Proof. Let f ∈ H(C) be an entire function such that |f (z)| ≤ M for all z ∈ C. Then
∞
X f (m) (0) m
f (z) = z
m!
m=0

for all z ∈ C. Fix m ≥ 1. The Cauchy’s Estimate for D(0, r) says that

M m!
f (m) (0) ≤ .
rm

Note that this estimate is valid for r > 0. Let r → ∞ then f (m) (0) ≤ 0. Thus f (m) (0) =
0 for all m ≥ 1. Thus f (z) = f (0) for all z0 ∈ C. That is, f is constant.

Theorem 9.6 (Fundamental Theorem of Algebra). Let P ∈ C[z] be a polynomial with

complex coefficients. If P is non-constant, then P has a root in C.

Analysis: Say that P (z) = an z n + · · · + a0 with an 6= 0. Then the term an z n dominates

when |z| is large. For example,
n−1
X
n
|P (z)| ≥ |an z | − |aj z j |
j=0
 
n−1
X |aj | 
= |z n | · |an | −
|z n−j |
j=0
 
n−1
X |aj |
≥ |z n | · |an | −  (if |z| > 1).
|z|
j=0

Pn−1
j=0 aj
Thus P (z) has no roots in the set where |an | − |z| > 0 and |z| > 1. This is equivalent
Pn−1
j=0 aj 1
to |z| > max{1, |an | }. Also, it follows that lim|z|→∞ |P (z)| = 0.

Proof of Theorem 9.6. Suppose not. Then P1 ∈ H(C) and P1 is bounded because lim|z|→∞ 1
|P (z)| =
0. Thus P1 is constant by Liouville’s Theorem, contrary to our assumption.

22
10 Elementary Results on Holomorphic Functions II (09/10)
Proposition 10.1. Let Ω ⊂ C be a connected open set, f ∈ H(Ω), z0 ∈ Ω and suppose
that f (m) (z0 ) = 0 for all m ≥ 0. Then f is constant (in fact f ≡ 0).
Proof. Let

C = {z ∈ Ω|f (m) (z) = 0 for all m ≥ 0}

∞
\
= {z ∈ Ω|f (m) (z) = 0}.
m=0

Since each f (m) is continuous, C is closed. Now z0 ∈ C and so C 6= ∅. We claim that C is

also open. Let w ∈ C and r > 0 such that D(w, r) ⊂ Ω. For z ∈ D(w, r) we have
∞
X f (m) (w)
f (z) = (z − w)m = 0.
m!
m=0

That is, f |D (w, r) ≡ 0. Since D(w, r) is open, this implies that f (m) ≡ 0 on D(w, r) for all
m ≥ 0, and so D(w, r) ⊂ C. This confirms that C is open. Thus C = Ω (Ω is connected),
as required.

Definition 10.2. Let Ω ⊂ C be a connected open set, f ∈ H(Ω), z0 ∈ Ω such that

f (z0 ) = 0. Assume f is not constant. Then the order of z0 as a zero of f is the least N
such that f (N ) (z0 ) 6= 0. A zero of order one is called a simple zero.
Theorem 10.3. Let V ⊂ C be open, f ∈ H(V ), z0 ∈ V , and f has a zero of order n
at z0 . Then there is a holomorphic function g ∈ H(V ) such that g(z0 ) 6= 0 and f (z) =
(z − z0 )n g(z) for all z ∈ V .
Proof. Let Ω be the connected component P∞ of V that contains z0 . Choose r > 0 so such
that D(z0 , r) ⊂ Ω and write f (w) = c (w − z )m for w ∈ D(z , r). Note that
m=0 m 0 0
c0 = c1 = · · · = cn−1 = 0, cn 6= 0, and so
∞
X
f (w) = cm (w − z0 )m
m=n

for w ∈ D(z0 , r). Define g : V → C by

( f (z)
(z−z0 )n , if z 6= z0 ,
g(z) =
cn , if z = z0 .

Note that g|V −{z0 } ∈ H(V −{z0 }). Also, if w ∈ D(z0 , r)−{z0 }, then g(w) = ∞ m−n .
P
m=n cm (z − z0 )
This equation is also true if w = z0 . This tells us that g|D(z0 ,r) ∈ H(D(z0 , r)). Since
V − {z0 } and D(z0 , r) are open, g ∈ H(V ) as required. Note g(z0 ) = cn 6= 0 by assump-
tion.

23
Corollary 10.4. Let V ⊂ C be open and connected, f ∈ H(V ) non-constant, z0 ∈ V a
zero of f . Then there is a disk D(z0 , r) ⊂ V such that f (z) 6= 0 for all z ∈ D(z0 , r) − {z0 }.

Proof. Let n be the order of z0 as a zero of f . Write f (z) = (z − z0 )n g(z) with g ∈ H(V ),
g(z0 ) 6= 0. g is continuous and g(z0 ) 6= 0, so there is a disk D(z0 , r) ⊂ V such that g(z) 6= 0
for all z ∈ D(z0 , r). The identity f (z) = (z − z0 )n g(z) now implies that f (z) 6= 0 for all
z ∈ D(z0 , r) − {z0 }.

11 Elementary Results on Holomorphic Functions III (09/15)

If V ⊂ C is open and D0 (z, r) = D(z, r) − {z} ⊂ V then a function f ∈ H(V ) is said to
have an isolated singularity at z. It is enough to study H(D0 (z, r)).

Theorem 11.1 (Riemann’s Removable Singularity Theorem). If f ∈ H(D0 (z, r)) and f
is bounded then there is some g ∈ H(D(z, r)) such that g|D0 (z,r) = f .

Proof. Let h : D(z, r) → C be defined by

(
(w − z)2 f (w), if w ∈ D0 (z, r)
h(w) =
0, if w = z.

We claim that h ∈ H(D(z, r)). Certainly h is differentiable at every point of D0 (z, r).
Now if w ∈ D0 (z, r), then

h(w) − h(z) (w − z)2 f (w)

= = (w − z)f (w).
w−z w−z

Thus | h(w)−h(z)
w−z | = |(w − z)f (w)| ≤ M |w − z| where M is a bound for f . It follows that

h(w) − h(z)
lim = 0.
w→z w−z
Thus h is also differentiable at z and h0 (z) = 0. So h ∈ H(D(z, r)) and h has a zero of
order at least two at z. Let the order be N (if h is constantly zero then so is f and the
conclusion is easy). There is a ϕ ∈ H(D(z, r)) such that

h(w) = (w − z)N ϕ(w), ∀w ∈ D(z, r).

This implies that f (w) = (w − z)N −2 ϕ(w), ∀w ∈ D(z, r). We take g ∈ H(D(z, r)) to be
g(w) = (w − z)N −2 ϕ(w).

Next, what if f ∈ H(D0 (z, r)) is not bounded?

Definition 11.2. The function f has a pole at z if limw→z |f (w)| = ∞. It has an essential
singularity if limw→z |f (w)| does not exist (is not finite and is not ∞).

24
Example 11.3. (i) f : D0 (0, 1) → C given by f (z) = z1 has a pole at 0.
(ii) g : D0 (0, 1) → C given by g(z) = exp( z1 ) has an essential singularity at 0. (Note that
g(D0 (0, 1)) = C − {0} for any r > 0.)

Proposition 11.4. Let f ∈ H(D0 (z, r)) and suppose that f has a pole at z. Then there
ϕ(w) 0
is some N ≥ 1 and ϕ ∈ H(D(z, r)) such that f (w) = (w−z)N , ∀w ∈ D (z, r), ϕ(z) 6= 0.

Proof. Since limw→z |f (w)| = ∞, there is some 0 < ρ ≤ r such that |f (w)| ≥ 1 for all
w ∈ D0 (z, ρ). In particular, f has no zeros in D0 (z, ρ) and so g : D0 (z, ρ) → C defined by
1
g(w) = f (w) is holomorphic. Also, |g(w)| ≤ 1, ∀w ∈ D0 (z, ρ). It follows that g extends
to a holomorphic function on D(z, ρ). Note that g is not constant (for f is not) and
limw→z |g(w)| = 0. Thus g has a zero at z. Let N be the order of this zero. Then ∃
ψ ∈ H(D(z, ρ)) such that g(w) = (w − z)N ψ(w) and ψ(z) 6= 0. Thus there is a 0 < σ ≤ ρ
such that ψ(w) 6= 0 for all w ∈ D(z, σ). On D0 (z, σ) we have

1 ϕ(w)
f (w) = =
(w − z)N ψ(w) (w − z)N
1
if we let ϕ = ψ. Note that ϕ(z) 6= 0. Note that if we define ϕ : D(z, r) → C by letting it
be a function we already have on D(z, r) and ϕ(w) = (w − z)N f (w) on D(z, r) − D(z, σ2 )
then it follows that ϕ ∈ H(D(z, r)).

Theorem 11.5 (Identity Principle). If F, G ∈ H(V ), and F (w) = G(w) for w ∈ S ⊂ V

where S has a limit point in V . Then F = G on the connected component of V containing
this limit point.

Remark 11.6. If g ∈ H(V ) and g is not a polynomial then h : C − {0} → C given by

h(z) = g( z1 ) has an essential singularity at 0.

Remark 11.7. If f has a pole at z, then f1 has a removable singularity at z. This is because
∃r such that f ∈ H(D0 (z, r)) since limw→z |f (w)| = ∞, there is some 0 ≤ ρ ≤ r such that
|f (w)| ≥ 1 for w ∈ D0 (z, ρ). f has no zeroes in D0 (z, ρ) and so f1 ∈ H(D0 (z, ρ)). Also,
1
| f (w) | ≤ 1 for w ∈ D0 (z, ρ) and so z is a removable singularity by Riemann’s Removable
1
Singularity Theorem. In fact, f (z) = 0.

Example 11.8. Suppose that f ∈ H(D0 (z, r)). Can f 0 have a simple pole at 0?
A simple pole means “poles of order 1”. g has a pole of order 1 at 0 if the
(i) g(z) = h(z)
z with h ∈ H(D(0, 1)), h(0) 6= 0.
(ii) g(z) = Az + k(z) with k ∈ H(D(0, 1)).
(iii) limz→0 zg(z) exists and is nonzero.
(iv) g(z) = c−1
z + c0 + c1 z + · · · .

25
 If f 0 has a pole of order 1 then f 0 (z) = Az + k(z), k ∈ H(D(0, 1)), A 6= 0. Take
γ(t) = 12 eit , 0 ≤ t ≤ 2π. Then
Z Z Z
0 A
f (z) dz = dz + k(z) dz
γ γ z γ
Z 2π Z
A 1 it
= 1 it 2 ie dt + k(z) dz
0 2e Z γ

= 2πiA + k(z) dz.

γ

By Cauchy’s Theorem for Derivatives, we have γ f 0 (z) dz = 0. By Cauchy’s Theorem for

R
R
Convex Sets, we have γ k(z) dz = 0. Hence 0 = 2πiA + 0, and so A = 0. Contradiction.

Example 11.9. Suppose that f ∈ H(D0 (z, r)) has an essential singularity at z. Is it true
that f1 also has an essential singularity at z?
f has an essential singularity at z means: (i) f ∈ H(D0 (z, r)); (ii) limw→z |f (w)| does
not exist. If f ∈ H(D0 (z, r)), then f1 ∈ H(D0 (z, r)\Z(f )). f1 does not necessarily have an
essential singularity at z because f1 may not have an isolated singularity.
For example, f (z) = exp( z1 ) − 1, f ( 2πik
1
) = 0 for k ∈ Z\{0}, so f1 ∈ H(D(0, 1)\{{0} ∪
1
{ 2πik : k ∈ Z\{0}}}). 0 is not an isolated singularity.
1
Another example is f (z) = exp( z1 ). f (z) = exp( −1
z ) does not have essential singularity
at 0.

12 Elementary Results on Holomorphic Functions IV (09/17)

Parseval’s Formula
Let V R⊂ C be open, f ∈ H(V ), z0 ∈ V , r > 0 such that D(z0 , r) ⊂ V . Our aim is to
2π
calculate 0 |f (z0 + reit )|2 dt.
Note: (i) |f |2 is continuous
P on the set {zn : |z − z0 | = r} so the integral exists. (ii) We
have a power series f (z) = ∞ n=0 cn (z − z0 ) that converges uniformly absolutely to f on
D(z0 , r).

26
 We have
2
f (z0 + reit ) = f (z0 + reit )f (z0 + reit )
∞ ∞
!
X X
n int
= cn r e cm rm eimt
n=0 m=0
∞ ∞
!
X X
n int m −imt
= cn r e cm r e
n=0 m=0
X∞ X l
= cj rj eijt · cl−j rl−j e−i(l−j)t
l=0 j=0
∞
X l
X
= rl e−ilt cj cl−j e2ijt
l=0 j=0
∞
X l
X
= rl cj cl−j e(2j−l)it
l=0 j=0

and this converges absolutely and uniformly in t. So

Z 2π ∞
X l
X Z 2π
it 2 l
|f (z0 + re )| dt = r cj cl−j e(2j−l)it dt
0 l=0 j=0 0

X∞
= |ck |2 r2k 2π.
k=0

1
R 2π it 2
P∞ 2 2k
Hence 2π 0 R|f (z0 + re )| dt = k=0 |ck | r .
1 2π it 2 2
Note: 2π 0 |f (z0 +re )| dt is the average value of |f | over the circle {z : |z−z0 | = r}
since
Z 2π Z 2π Z 2π
1 it 2 1 it 2 1
|f (z0 + re )| ds = |f (z0 + re )| r dt = |f (z0 + reit )|2 dt.
2πr 0 2πr 0 2π 0
From
Z 2π ∞
1 it 2
X
|f (z0 + re )| dt = |ck |2 r2k
2π 0 k=0
∞
X
2
= |c0 | + |ck |2 r2k
k=1
∞
X
= |f (z0 )|2 + |ck |2 r2k
k=1

27
 1
R 2π
we conclude that |f (z0 )|2 ≤ 2π it 2
0 |f (z0 +R re )| dt with equality if and only if ck = 0 for
1 2π
k ≥ 1. This is equivalent to |f (z0 )|2 ≤ 2π it 2
0 |f (z0 + re )| dt with equality if and only if
f is constant on D(z0 , r).

Theorem 12.1 (Maximum Modulus Principle). Let V ⊂ C be a connected open set and
f ∈ H(V ) be non-constant. Then |f | does not achieve a maximum at any point in V .

Proof. Let z0 ∈ V and r > 0 be such that D(z0 , r) ⊂ V . Since f is not constant in V and
V is connected, f is not constant on D(z0 , r). Thus
Z 2π
2 1
|f (z0 )| < |f (z0 + reit )|2 dt.
2π 0

Thus there is some point w such that w − z0 = r and |f (w)|2 > |f (z0 )|2 . This implies that
|f (w)| > |f (z0 )|. Since r > 0 can be taken as small as desire, it follows that z0 is not a
local maximum point for |f |.

Remark 12.2. (i) If V ⊂ C is a connected open set and f ∈ H(V )∩C(V ) is non-constant,
then |f (z)| ≤ max{|f (w)| : w ∈ ∂V } for all z ∈ V .

Example 12.3. Let f (z) = exp(z). Let V = {z = x + iy|x > 0}. Notice that |f | = 1 on
∂V , and |f (z)| = exp(x). So the modulus |f | increases along the x-axis.

Example 12.4. Let f ∈ H(D(0, 1)) and f (D(0, 1)) ⊂ D(0, 1). Suppose that f ∈
C(D(0, 1)). Suppose also that f (0) = 0. Let g(z) = f (z)
z for z ∈ D0 (0, 1). Note that
g has a removable singularity at 0. So g ∈ H(D(0, 1)). We have |g(z)| = |f|z|
(z)| 1
≤ |z| =1
when |z| = 1. By Maximum Modulus Principle, |g(z)| ≤ 1 for all z ∈ D(0, 1). It follows
that |f (z)| ≤ |z| for any z ∈ D(0, 1).

28
13 Logarithms, Winding Numbers and Cauchy’s Theorem
I (09/19)
Theorem 13.1. Let V ⊂ C be Ran open set and f ∈ H(V ). Then there exists F ∈ H(V )
such that F 0 = f if and only if γ f (z) dz = 0 for all closed curve γ in V .

Proof. (⇒) We know that if ∃ F ∈ H(V ) with F 0 = f then γ f (z) dz = 0 for all closed
R

curves in V . (This is Cauchy’s

R Theorem for Derivatives.)
(⇐) Suppose that γ f (z) dz = 0 for all closed curves γ in V . We may write V as a
union of its connected components. It suffices to verify that ∃F ∈ H(Ω) such that F 0 = f
on Ω, for each connected component Ω of V . Note that the connected components are all
open sets – this is important. Now say that Ω is a connected component of RV . Then Ω
is path connected. Fix z0 ∈ Ω and define a function F : Ω → C by F (z) = γz f (w) dw
where γz is any smooth curve from z0 to z in Ω. Note that the hypothesis implies that
F (z) does not dependR on the choice of γz . Once again using the hypothesis, we have
F (z + h) − F (z) = [z,z+h] f (w) dw once h is small enough that it lies in a disk contained
in Ω and centered at z. This implies that
F (z + h) − F (z)
lim = f (z).
h→0 h
Thus F works on Ω.

Remark 13.2. z1 ∈ H(D0 (0, 1)), 1 1

R R
γ z dz can be non-zero. For example, γ z dz = 2πi
where γ(t) = 12 eit , t ∈ [0, 2π].

Now we turn to logarithms.

Definition 13.3. Let V ⊂ C be an open set. A function f ∈ H(V ) is said to be a branch

of logarithm in V if exp(f (z)) = z for all z ∈ V .

Remark 13.4. (i) If there is a branch of the logarithm in V then 0 6∈ V . This is because
if it were then exp(f (0)) = 0 which is impossible.
(ii) A branch of the logarithm in V is always one-to-one. This is because if f is a branch
of the logarithm in V , z, w ∈ V , and f (z) = f (w), then z = exp(f (z)) = exp(f (w)) = w.
(iii) If f is a branch of the logarithm in V , z ∈ V , and z = reiθ is a polar representation
of z then
f (z) = ln(r) + iθ + 2πik
for some k ∈ Z. This is because exp(f (z)) = z and exp(ln(r) + iθ) = reiθ = z. From
Assignment 1, this implies that

f (z) = ln(r) + iθ + 2πik

for some k ∈ Z.

29
Lemma 13.5. Suppose that V ⊂ C, f ∈ C(V ), and exp(f (z)) = z for all z. Then f is a
branch of the logarithm in V .
Proof. We have to show that f is differentiable. If z ∈ V and h 6= 0 is small enough, then
h
f (z+h)−f (z) is defined. (This is because f must be one-to-one by a similar argument to
one above). Also,
h (z + h) − z exp(f (z + h)) − exp(f (z))
= = .
f (z + h) − f (z) f (z + h) − f (z) f (z + h) − f (z)
Since f is continuous, limh→0 (f (z + h) − f (z)) = 0. This implies that
h exp(f (z + h)) − exp(f (z))
lim = lim
h→0 f (z + h) − f (z) h→0 f (z + h) − f (z)
= exp(f (z)) (since exp is differentiable)
= z.
As before, 0 6∈ V , and so
f (z + h) − f (z) 1
lim = .
h→0 h z
Thus, f is differentiable.

Theorem 13.6. Let V ⊂ C be open with 0 6∈ V . Then there is a branch of the logarithm
in V if and only if f (z) = z1 has an antiderivative in V .
Proof. (⇒) We just saw that the equation exp(F (z)) = z, ∀z ∈ V implies that F 0 (z) = z1
for all z ∈ V .
(⇐) Conversely, suppose that F ∈ H(V ) and F 0 (z) = z1 for all z ∈ V . Let G(z) =
z exp(−F (z)). Then G ∈ H(V ) and

G0 (z) = exp(−F (z)) + z exp(−F (z)) · (−F 0 (z))

1
= exp(−F (z)) + z exp(−F (z)) · (− )
z
= 0.
Thus G is constant on each connected component of V . That is, for each component Ω of
V , ∃ cΩ such that
exp(F (z)) = cΩ z.
Note that cΩ 6= 0 because exp(C) does not contain 0. Choose cΩ such that exp(cΩ ) = cΩ
and define L : V → C by L(z) = F (z) − cΩ . Then L ∈ H(V ) and exp(L(z)) = z. So L(z)
works.

Remark 13.7. V ⊂ C is an open set. ∃ f ∈ H(V ) such that ef (z) = z if and only if
(i) 0R6∈ V ;
(ii) γ z1 dz = 0 for all closed curve γ in V .

30
Remark 13.8. V ⊂ C is an open set, f ∈ H(V ). g ∈ H(V ) is a branch of logarithm of
f if eg(z) = f (z) for all z ∈ V . Such g exists if and only if
(i) Z(f ) ∩ V 6= ∅;
R 0 (z)
(ii) γ ff (z) dz = 0 for all closed curve γ in V .

Remark 13.9. We will see that if g ∈ H(V ) where

(i) V is simple connected,
(ii) g does not vanish in V ,
then g has a logarithm in V . That is to say, there exists f ∈ H(V ) such that ef (z) = g(z).
Then g has a m-th root h(z) = exp ( fm (z)
) ∈ H(V ) such that g(z) = h(z)m .

14 Logarithms, Winding Numbers and Cauchy’s Theorem

II (09/22)

σ̃(0) = t0
t0
R

p(t) = eit , p(t0 ) = σ(0)

∃ σ̃

σ(0)

σ
I

p(σ̃(t)) = σ(t), ∀t ∈ I
σ̃ is a lift of σ above p

Figure 5:

Let I = [0, 1]. Given a metric space X then define

π(X) = set of all continuous maps σ : I → X,

Ω(X) = set of all continuous loops σ : I → X, σ(0) = σ(1).

31
So Ω(X) ⊂ π(X), and π(X) is a metric space with

dπ (σ, ω) = sup dX (σ(t), ω(t)).

t∈I

Ω(X) is a subspace of π(X).

We want to study π(S) and Ω(S), where S = {z ∈ C : |z| = 1}.
Let S+ = {z ∈ S : Re(z) > 0}. Define ϕ : S+ → R by ϕ(z) = arctan( Im(z) Re(z) ). From
basis trigonometry, we have ϕ(1) = 0 and p(ϕ(z)) = z for all z ∈ S+ . ϕ is continuous
and if ψ : S+ → R such that ψ(1) = 0 and p(ψ(z)) = z for any z ∈ S+ , then ϕ = ψ.
Here is a non-elementary way to prove this. Let k(z) = ψ(z) − ϕ(z). Then exp(ik(z)) =
exp(i(ψ(z) − ϕ(z))) = z · z −1 = 1. So ik(z) ∈ 2πiZ. That is, k(z) ∈ 2πZ. k(z) is
continuous and k(1) = 0 and so k(S+ ) is a connected subset of 2πZ that contains 0. Thus
k(S+ ) = {0} and so ϕ = ψ.

Remark 14.1. ϕ : S+ → R is similar to “inverse functions” like arcsin, arccos, · · · . It is

an inverse once the domain is restricted.

More formally, let z0 ∈ S, let S+ (z0 ) = z0 S+ be the half-circle centered at z0 , choose

to ∈ R such that p(t0 ) = z0 . Then ψ : S+ (z0 ) → R given by ψ = Tt0 ◦ ϕ ◦ Rz −1 is an
0
inverse map for p such that ψ(z0 ) = t0 and it is unique among continuous maps with these
properties. Here Rw : S → S is Rw (z) = wz, Tτ : R → R is Tτ (t) = t + τ . Check the
properties of ψ: it is continuous because Tto , ϕ, Rz −1 are all continuous.
0

ψ(z0 ) = Tt0 (ϕ(Rz −1 (z0 ))) = Tt0 (ϕ(1)) = Tt0 (0) = t0 + 0 = t0

0

and
p(ψ(z)) = p(Tt0 (ϕ(Rz −1 (z))))
0

= p(t0 + ϕ(Rz −1 (z)))

0

= p(t0 ) · p(ϕ(Rz −1 (z)))

0

= z0 · Rz −1 (z)
0

= z0 · z0−1 ·z
= z.

Theorem 14.2 (Path Lifting for the Circle). Let σ ∈ π(S) and choose t0 ∈ R such
that p(t0 ) = σ(0). Then there is one and only one σ̃ = π(R) such that σ̃(0) = t0 and
p(σ̃(t)) = σ(t).

Proof. Let U = {zS+ |z ∈ S} be an open cover of S. Then ξ = {σ −1 (u)|u ∈ U } is an

open cover of I. Choose a Lebesgue number λ > 0 for ξ. This means that if [s1 , s2 ] ⊂ I

32
and s2 − s1 < λ then [s1 , s2 ] is entirely inside one element of ξ. Thus σ([s1 , s2 ]) is entirely
inside one half-circle in S. Partition I into

0 = s0 < s1 < s2 < · · · < sn−1 < sn = 1

such that sj −sj−1 < λ for all j. Then σ([s0 , s1 ]), σ([s1 , s2 ]), · · · , σ([sn−1 , sn ]) are all inside
half-circles in S. We can now use the inverse maps ψ previously constructed to write down
σ̃ on each interval. Finally, σ̃ is unique by connectedness argument.

All this says that every path σ : I → S has the form σ(t) = exp(iθ(t)) where θ : I → R
is continuous. Also θ is uniquely determined once θ(0) is chosen.
If σ is a loop (σ(0) = σ(1)) then θ(1) and θ(0) differ by a multiple of 2π.

exp(iθ(1)) = σ(1) = σ(0) = exp(iθ(0))

implies that θ(1) − θ(0) ∈ 2πZ.

Definition 14.3. We define the index of σ by
1
ind(σ) = (θ(1) − θ(0)).
2π
Remark 14.4. Note that although θ is not unique, once you have one θ, the others are
θk = θ + 2πk. Then
θk (1) − θk (0) = θ(1) − θ(0).
So index is well defined.
Index is easy to compute for explicit map.
1
Example 14.5. Say σ(t) = exp(4πit). For this loop, θ(t) = 4πt and so ind(σ) = 2π (θ(1)−
1
θ(0)) = 2π (4π − 0) = 2.

15 Logarithms, Winding Numbers and Cauchy’s Theorem

III (09/24)
Recall from last time: If σ : I → S is continuous then σ can be expressed as σ(t) =
exp(iθ(t)) where θ : I → R is continuous. θ is uniquely determined by θ(0). If σ is
smooth, so is θ. The correspondence σ θ is “continuous”. Given σ > 0, there is a δ > 0
such that if dπ(S) (σ, w) < δ and σ̃, w̃ are lifts of σ, w, then dπ(R) (σ̃ − σ̃(0), w̃ − w̃(0)) < σ.
Definition 15.1. We define ind : Ω(S) → Z by
1
ind(σ) = (θ(1) − θ(0))
2π
where σ(t) = exp(iθ(t)), θ continuous.

33
 In fact, ind is continuous (even uniformly continuous). There is δ > 0 such that if
σ, w ∈ Ω(S) and dΩ(S) (σ, w) < δ then ind(σ) = ind(w).
γ(t)
Definition 15.2. If γ : I → C − {0} then we define σ : I → S by σ(t) = |γ(t)| . We define
Ind(γ, 0) = ind(σ).

γ(t)

σ(t)

Figure 6:

The map γ 7→ σ is continuous. In fact,

γ1 (t) γ2 (t) |γ2 (t)|γ1 (t) − |γ1 (t)|γ2 (t)
− =
|γ1 (t)| |γ2 (t)| |γ1 (t)||γ2 (t)|
|γ2 (t)|(γ1 (t) − γ2 (t)) + |γ2 (t)|γ2 (t) − |γ1 (t)|γ2 (t)
=
|γ1 (t)||γ2 (t)|
|γ1 (t) − γ2 (t)| |γ2 (t)| − |γ1 (t)|
≤ +
|γ1 (t)| |γ1 (t)|
2|γ1 (t) − γ2 (t)|
≤ .
|γ1 (t)|
Given γ1 , One can choose η > 0 such that |γ1 (t)| ≥ η ∀t and then
γ1 (t) γ2 (t) 2
− ≤ · |γ1 (t) − γ2 (t)|.
|γ1 (t)| |γ2 (t)| η
γ
This shows that γ → |γ| is continuous (not uniformly so). This tells us that γ 7→ Ind(γ, 0)
from Ω(C − {0}) → Z is continuous. Finally, if a ∈ C and a 6∈ γ ∗ , then we define
Ind(γ, a) = Ind(γ − a, 0).

34
 If K ⊂ C is a compact set then C−K always has a unique unbounded component. Here
is the reason. Choose R > 0 such that K ⊂ D(0, R). Then C − D(0, R) = {w ∈ C||w| >
R} ⊂ C − K and is connected. The component that meets this set is unbounded because
it contains the set. Every other component does not meet this set and so is contained in
D(0, R) and hence is bounded.

Theorem 15.3. Let γ : I → C be a loop. Then the function a 7→ Ind(γ, a) from C−γ ∗ → Z
is constant on each component of C − γ ∗ and zero on the unbounded component.

Proof. The function a 7→ Ind(γ, a) is continuous, since it is the composition of some

continuous functions:
C − γ ∗ → Ω(C − {0}) → Ω(S) → Z
γ−a
a 7→ γ − a 7→ 7→ Ind(γ, a)
|γ − a|

Thus if w is a component of C − γ ∗ then the image of w under a 7→ Ind(γ, a) is a connected

subset of Z. This must be a singleton. We may choose R > 0 such that the unbounded
component of C − γ ∗ contains {w||w| > R}. Choose a ∈ C with |a| > R. Consider the
map [1, ∞) → Z given by
γ − sa
s 7→ Ind(γ, sa) = ind( ).
|γ − sa|
γ−sa a
As s → ∞, |γ−sa| → −sgn(a) = − |a| uniformly. The index thus converges to ind(constant map) =
0. Since Ind(γ, sa) is also constant, it equals 0.

Theorem 15.4. If γ is a smooth loop in C and a 6∈ γ ∗ then

Z
1 1
Ind(γ, a) = dz.
2πi γ z − a
γ−a
Proof. Let θ be a lift of |γ−a| and write ρ = |γ − a|. Then

γ(t) − a
γ(t) − a = |γ(t) − a| · = ρ(t) · exp(iθ(t)).
|γ(t) − a|

So
γ 0 (t) = ρ0 (t) · exp(iθ(t)) + ρ(t) · i exp(iθ(t)) · θ0 (t)
ρ0 (t)
= ρ(t) exp(iθ(t)) + iθ0 (t)ρ(t) exp(iθ(t))
ρ(t)
ρ0 (t)
= (γ(t) − a) + iθ0 (t)(γ(t) − a)
ρ(t)

35
and thus
γ 0 (t) ρ0 (t)
= + iθ0 (t).
γ(t) − a ρ(t)
Hence,
1
γ 0 (t)
Z Z
1 1 1
dz = dt
2πi γ z−a 2πi 0 γ(t) − a
Z 1 0 
1 ρ (t) 0
= + iθ (t) dt
2πi 0 ρ(t)
1
= (ln(ρ(1)) − ln(ρ(0)) + i(θ(1) − θ(0)))
2πi
1
= (θ(1) − θ(0))
2π
= Ind(γ, a).

16 Logarithms, Winding Numbers and Cauchy’s Theorem

IV (09/26)
Lemma 16.1. Let V1 , V2 ∈ C be open, Γ a formal sum of smooth paths in V2 , F : V1 ×V2 →
C is (jointly)
R continuous and holomorphic in the first variable. Define f : V1 → C by
f (z) = Γ F (z, w)dw. Then f ∈ H(V1 ).
Proof. First, f is continuous on V1 because F is continuous and Γ∗ is compact. Let T ⊂ V1
be a triangle, then
Z Z Z 
f (z) dz = F (z, w)dw dz
∂T ∂T Γ
Z Z 
= F (z, w)dz dw (by Fubini’s Theorem for Riemann integrals)
ZΓ ∂T

= 0 dw (by Cauchy-Goursat Theorem)

Γ
= 0.
By Morera’s Theorem, f ∈ H(V1 ).

Theorem 16.2 (Cauchy’s Integral Formula, Homology Version). Let V ⊂ C be open,

Γ ⊂ V a formal sum of closed curve, and suppose Ind(Γ, a) = 0 for all a ∈ C\V . Let
f ∈ H(V ), then Z
1 f (w)
dw = Ind(Γ, z)f (z)
2πi Γ w − z
for all z ∈ V \Γ∗ .

36
Proof. Let Ω = {w ∈ C\Γ∗ |Ind(Γ, w) = 0}. Then Ω is open (it is a union of connected
components of C\Γ∗ ) and Ω ⊃ C\V . Thus C = V ∪ Ω and also Γ∗ ∩ Ω = ∅. Consider
ϕ : V × V → C given by
(
f (z)−f (w)
z−w if z 6= w,
ϕ(z, w) = 0
f (z) if z = w.
Recall that we showed that ϕ can be written locally as a sum of a power series in z, w.
It follows that ϕ is continuous and holomorphic in each variable separately. We define
Φ1 : V → C by Z
1
Φ1 (z) = ϕ(z, w) dw.
2πi Γ
By Lemma 16.1, Φ1 ∈ H(V ). We define Φ2 : Ω → C by
Z
1 f (w)
Φ2 = − dw.
2πi Γ w − z
By Lemma 16.1, Φ2 ∈ H(V ).
Next, I want to check that if z ∈ V ∩ Ω, then Φ1 (z) = Φ2 (z). Let z ∈ V ∩ Ω, then
z 6∈ Γ∗ , Ind(Γ, z) = 0, and so
Z
1 1
dz = Ind(Γ, z) = 0.
2πi Γ w − z
Thus
Z
1
Φ1 (z) = ϕ(z, w) dw
2πi Γ
f (z) − f (w)
Z
1
= dw
2πi Γ z−w
Z Z
1 f (z) 1 f (w)
= dw − dw
2πi Γ z − w 2πi Γ z − w
Z
f (z) 1
= dw + Φ2 (z)
2πi Γ z − w
= −f (z)Ind(Γ, z) + Φ2 (z)
= Φ2 (z).
It follows that Φ : C → C given by
(
Φ1 (z), if z ∈ V
Φ(z) =
Φ2 (z), if z ∈ Ω
is holomorphic. That is, Ω is entire. We know that the (unique) unbounded component
of C\Γ∗ is contained in Ω. Thus
Z
1 f (w)
Φ(z) = dz
2πi Γ w − z

37
for all z with |z| large enough. It follows that lim|z|→∞ Φ(z) = 0. Liouville’s Theorem
says Φ is constant, and then Φ ≡ 0 because of the limit. So Φ1 (z) = 0 for all z. This
R f (z)−f (w)
1
says that 2πi Γ z−w dw = 0 for all z 6∈ Γ∗ , and this can be rearranged to give the
conclusion.

Remark 16.3. This proof is due to John Dixon (A brief Proof of Cauchy’s Integral The-
orem, 1971, Proceedings of the American Mathematical Society.).

17 Logarithms, Winding Numbers and Cauchy’s Theorem

V (09/29)
Theorem 17.1 (Cauchy’s Theorem, Homology Version). Let V ⊂ C be open, Γ ⊂ V be a
formal sum of closed curves. Suppose that Ind(Γ, a) = 0 for all a ∈ C\V . Let f ∈ H(V ),
then Z
f (w) dw = 0.
Γ

Proof. Choose z0 ∈ V \Γ∗ . Let g(z) = (z − z0 )f (z). Apply Cauchy’s Integral Formula,
Homology Version to g, we get
Z
1 g(w)
dw = Ind(Γ, z0 )g(z0 ) = 0.
2πi Γ w − z0
1
R g(w) 1
R R
On the other hand, 2πi Γ w−z0 dw = 2πi Γ f (w) dw and so Γ f (w) dw = 0.

Definition 17.2. Let 0 ≤ r < R ≤ ∞. We define, for a ∈ C,

A(a, r, R) = {z ∈ C|r < |z − a| < R}

and call this the annulus centered at a with radii r, R.

Remark 17.3. D0 (a, R) = A(a, 0, R), C − {a} = A(a, 0, ∞).

Theorem 17.4. Let f ∈ H(A(a, r, R)). Then there are functions f1 ∈ H(D(a, R)) and
f2 ∈ H(A(a, r, ∞)) such that f = f1 + f2 on the common domain and limz→∞ f2 (z) = 0.
Moreover, f1 , f2 are unique.

Proof. We may assume, after translating, that a = 0. For r < ρ < R, let γρ : [0, 2π] →
A(0, r, R) be γρ (t) = ρeit . Let r < r1 < R1 < R (see Figure 7), note that Ind(γR1 −̇γr1 , c) =
0 for all c 6∈ A(0, r, R). Thus we may apply Cauchy’s Integral Formula, Homology Version
to γR1 −̇γr1 . If we do so, then we get
Z
1 f (w)
Ind(γR1 −̇γr1 , z)f (z) = dw
2πi γR −̇γr1 w − z
1

38
 γR1
γr1

Figure 7:

∗ ∪ γ ∗ . In particular, if r < r < |z| < R < R, then

for all z 6∈ γR 1 r1 1 1
Z Z
1 f (w) 1 f (w)
f (z) = dw − dw.
2πi γR w − z 2πi γr1 w − z
1

If R1 < R2 < R and r < r1 < |z| < R1 < R then

Z Z
1 f (w) 1 f (w)
dw = dw
2πi γR w − z 2πi γR w − z
1 2

again by Cauchy’s Integral Formula, Homology Version. Similarly if r < r1 < r2 then
Z Z
1 f (w) 1 f (w)
dw = dw.
2πi γr1 w − z 2πi γr2 w − z

We now define f1 : D(0, R) → C by

Z
1 f (w)
f1 (z) = dw
2πi γR1 w−z

where R1 is any number such that |z| < R1 < R. We also define f2 : A(0, r, ∞) → C by
Z
1 f (w)
f2 (z) = − dw
2πi γr1 w − z

where r1 is any number such that r < r1 < |z|. These functions are well defined and holo-
morphic by Lemma 16.1. We already observed that f = f1 +f2 . We have limz→∞ fz−w (w)
=0
uniformly for w ∈ γr∗1 . Thus limz→∞ f2 (z) = 0.

39
 Finally, suppose f˜1 , f˜2 have all the same properties. Then f1 (z) + f2 (z) = f˜1 (z) + f˜2 (z)
for all z ∈ A(0, r, R). So f1 (z) − f˜1 (z) = f˜2 (z) − f2 (z) for all z ∈ A(0, r, R). It follows that
F : C → C given by
(
f1 (z) − f˜1 (z), if z ∈ D(0, r)
F (z) =
f˜2 (z) − f2 (z), if z ∈ A(0, r, ∞)

is entire. Moreover, limz→∞ F (z) = 0. Thus, F ≡ 0 by Liouville’s Theorem, and so

f1 (z) = f˜1 (z) and f2 (z) = f˜2 (z).

Theorem 17.5. Let f ∈ H(A(a, r, R)). Then there is a Laurent series

X
cn (z − a)n
n∈Z

that converges uniformly absolutely on compact subsets of A(a, r, R) to f . Moreover,

Z
1 f (w)
cn = dw
2πi γρ (w − a)n+1

for all n ∈ Z, r < ρ < R.

Proof. Once again, we may assume a = 0. Write f = f1 + f2 as in Theorem 17.4. Define
g : D0 (0, 1r ) → C by g(z) = f2 ( z1 ). Then limz→0 g(z) = 0 because limz→∞ f2 (z) = 0, and
so g extends to a holomorphic function on D(0, 1rP ) with g(0) = 0 by Riemann’s
P∞ Removable
∞ n and g(z) = n
Singularity Theorem. We know that f1 (z) = n=0 an z n=1 bn z with
uniformly absolute convergence on compact subsets. Thus
∞ ∞
1 X X 1 X
f (z) = f1 (z) + g( ) = an z n + bn ( )n = cn z n
z z
n=0 n=1 n∈Z

with uniformly absolute convergence on compact subsets. To get the formula for cn , note
that we are justified in integrating term by term.

18 Logarithms, Winding Numbers and Cauchy’s Theorem

VI (10/1)
Definition 18.1. Let V ⊂ C be open. If γ0 , γ1 are loops in V then γ0 , γ1 are said to be
homotopic in V , denoted as γ0 ∼ γ1 , if there is a path from γ0 to γ1 in Ω(V ), where Ω(V )
V
is the space of all loops in V with the sup-metric.
Example 18.2. Let V ⊂ C, γ0 (t) = sin(πt), t ∈ [0, 1] and γ1 (t) = 0, t ∈ [0, 1]. Define
τ : I → Ω(C) from γ0 to γ1 :
τ (s)(t) = (1 − s) sin(πt).

40
Now τ is continuous as a function of two variables, and

τ (0)(t) = sin(πt) = γ0 (t),

τ (1)(t) = 0 = γ1 (t),
τ (s)(0) = (1 − s) sin(π0) = (1 − s) sin(π1) = τ (s)(1).
Hence γ0 and γ1 are homotopic in V = C.

Remark 18.3. If X is path connected then there is a path through loops from a loop based
anywhere to a loop based at p ∈ X.

Suppose V ⊂ C open, γ0 , γ1 are loops in V , γ0 ∼ γ1 , a ∈ C\V . Then Ind(γ0 , a) =

V
Ind(γ1 , a). We know that γ 7→ Ind(γ, a) is continuous. Since γ0 ∼ γ1 there is a path σ
V
from γ0 to γ1 . The function s 7→ Ind(σ(s), a) is constant. So it is constant.

Remark 18.4. This implies that γ0 ∼ γ1 then Ind(γ1 −̇γ0 , a) = 0 for all a ∈ C\V . Thus
V

the Cauchy’s
R Theorem,
R Homology Version applies to the formal sum of paths Γ = γ1 −̇γ0 .
Thus γ0 f (z) dz = γ1 f (z) dz for any f ∈ H(V ).

Example 18.5. Let V = C − {−1, 1}, γ be the Barnes’ double circuit as in Figure 8. In
this case, Ind(γ, 1) = 0, Ind(γ, −1) = 0, γ ∼ constant path.
V

Definition 18.6. Let f ∈ H(D0 (z0 , r)). We know that f (z) = n∈Z cn (z − z0 )n with
P
uniformly absolute convergence on compact subsets of D0 (z0 , r). We define the residue of
f at z0 to be
Res(f, z0 ) = c−1 .

Let γ(t) = z0 + ρeit , t ∈ [0, 2π], 0 < ρ < r. Then

Z X Z
f (z) dz = cn (z − z0 )n dz
γ n∈Z γ

= 2πi · c−1
= 2πi · Res(f, z0 ).

41
 −1 1

Figure 8:

19 Logarithms, Winding Numbers and Cauchy’s Theorem

VII (10/3)
Theorem 19.1 (Residue Theorem). Let V ⊂ C be an open set. Suppose that S ⊂ V
is closed in V and every point of S is isolated in S. Let Γ ⊂ V \S be a formal sum of
closed curve such that Ind(Γ, a) = 0 for every a ∈ C\V . Let f ∈ H(V − S). Then the set
{p ∈ S|Ind(Γ, p) 6= 0} is finite and
Z
1 X
f (z) dz = Ind(Γ, p)Res(f, p).
2πi Γ
p∈S

Example 19.2. Let λ > 0 and f (z) = exp(iλz)

z 2 +1
. Then f ∈ H(C − {±i}). Γ = γ1 +̇γ2 where
it
γ1 (t) = t for t ∈ [−R, R], R > 1, γ2 (t) = Re for t ∈ [0, π]. (To apply Residue Theorem,
we should parametrize Γ using a single interval so that Γ is a closed curve. It makes no
difference.) Let S = {±i}. Note that Γ ⊂ C\{±i} provided R > 1. To use the Residue
Theorem, we need Ind(Γ, i) and Ind(Γ, −i). We know Ind(Γ, −i) = 0 because −i is in the
unbounded component of C\Γ∗ . It is a fact that Ind(Γ, i) = 1. One way would be to show
that Γ ∼ σ where σ is a small circle centered at i. Then Ind(σ, i) = 1 by computation,
C−S

42
so Ind(Γ, i) = 1. Next, we need Res(f, i).

e−λ
 
exp (iλz) exp(iλz) 1 1
f (z) = = · = + ··· · .
(z − i)(z + i) z+i z−i 2i z−i
e−λ
Thus Res(f, i) = 2i . By Residue Theorem,
Z
1
f (z) dz = Ind(Γ, i) · Res(f, i) + Ind(Γ, −i) · Res(f, −i).
2πi Γ

Thus Γ f (z) dz = πe−λ .

R

On the other hand,

Z Z Z
f (z) dz = f (z) dz + f (z) dz
Γ γ1 γ2
Z R Z
exp(iλt)
= dt + f (z) dz.
−R t2 + 1 γ2

Now
exp(iλz) exp(iλ(x + iy))
|f (z)| = 2
=
z +1 z2 + 1
exp(iλx − λy) e−λy
= =
z2 + 1 z2 + 1
1
≤ 2 ( for z such that y ≥ 0).
|z + 1|

Thus
Z  
1
f (z) dz ≤ max∗ 2
· πR ( ML-inequality)
γ2 z∈γ2 |z + 1|
1
= · πR.
R2 −1
RR exp(iλt)
So far, we have πe−λ = −R t2 +1
dt + F (R) where |F (R)| ≤ πR
R2 −1
. Let R → ∞ in this
equation, we get Z R
exp(iλt)
πe−λ = lim dt
R→∞ −R t2 + 1
because limR→∞ F (R) = 0.
Z ∞ Z R2
exp(iλt) exp(iλt)
dt = lim dt
−∞ t2 + 1 R1 →∞,R2 →∞ R1 t2 + 1

43
 RR RR
provided this exists, it equals limR→∞ −R exp(iλt)
t2 +1
dt. But −R might exist even when the
R∞ 1 R ∞ exp(iλt)
other doesn’t. By comparison with −∞ 1+t 2 dt, we know that −∞ t2 +1 dt converges.
Thus Z ∞ Z R
exp(iλt) exp(iλt)
2+1
dt = lim 2+1
dt = πe−λ .
−∞ t R→∞ −R t
R ∞ sin(λt) R∞
Note that −∞ t2 +1 dt = 0 because t2 +1 is odd. Thus, −∞ cos(λt)
sin(λt)
t2 +1
dt = πe−λ , for
R ∞ cos(−λt) R ∞ cos(λt) R∞ 1
λ > 0. Finally, −∞ t2 +1 dt = −∞ t2 +1 dt and −∞ t2 +1 dt = π, so we can remove the
restriction on λ.

44
20 Counting Zeroes and the Open Mapping Theorem I (10/6)
Let V ⊂ C be open and f ∈ H(V ) is not constant on any component of V . Say z0 ∈ Z(f ),
we can write f (z) = (z − z0 )n g(z) where g(z0 ) 6= 0, g ∈ H(V ). Then n = ord(f, z0 ) is the
order of zero at z0 . Then

f 0 (z) = n(z − z0 )n−1 g(z) + (z − z0 )n g 0 (z)

and so
f 0 (z) n g 0 (z)
= + .
f (z) z − z0 g(z)
This is valid as long as we stay close enough to z0 that g is non-zero, except for z0 . We
0 0
conclude that ff ∈ H(V − Z(f )) has a simple pole at z0 with residue Res( ff , z0 ) = n =
ord(f, z0 ).
Theorem 20.1 (The Argument Principle). Let V ⊂ C be open, f ∈ H(V ), and f is not
constant on any component of V . Let q ∈ C and define S = {z ∈ V |f (z) = q}. Suppose
that γ ⊂ V \S is a closed curve such that Ind(γ, a) = 0 for all a ∈ C\V and Ind(γ, a) = 0
or 1 for a ∈ C\γ ∗ . Let Ω = {z ∈ V \γ ∗ |Ind(γ, z) = 1}. Define γ̃ in C − {q} to be
γ̃ = f ◦ γ. Then the number of solutions to the equations f (w) = q with w ∈ Ω, counted
with multiplicity, is equal to Ind(γ̃, q).
Proof.
Z Z 1
1 1 1 1
Ind(γ̃, q) = dζ = γ̃ 0 (t) dt
2πi γ̃ ζ −q 2πi 0 γ̃(t) − q
Z 1
1 1
= · f 0 (γ(t)) · γ 0 (t) dt
2πi 0 f (γ(t)) − q
f 0 (γ(t))
Z 1
1
= · γ 0 (t) dt
2πi 0 f (γ(t)) − q
f 0 (z)
Z 1
1
= dz
2πi 0 f (z) − q
(f (z) − q)0
Z 1
1
= dz
2πi 0 f (z) − q
(f (z) − q)0
X  
= Ind(γ, w) · Res ,w
f (z) − q
w∈Z(f (z)−q)
(f (z) − q)0
X  
= Res ,w
f (z) − q
w∈Z(f (z)−q)∩Ω
X
= ord (f (z) − q, w)
w∈Z(f (z)−q)∩Ω

= number of solutions to f (w) = q in Ω counted with multiplicity.

45
 P
Remark 20.2. “Counted with multiplicity” means w∈Ω ord(f (w) − q, w).

Example 20.3 (Basic Case). V = D(0, 1), f ∈ H(V ) is f (z) = z k (k ∈ N), γ(t) =
1 it 1 1 it k 1 ikt
2 e , t ∈ [0, 2π]. Here Ω = D(0, 2 ), γ̃(t) = ( 2 e ) = 2k e . So Ind(γ̃, 0) = k. f has a
single zero in Ω, but the order of the zero is k. So the Argument Principle works in this
case. We know that q 7→ Ind(γ̃, q) is constant on components of C\γ̃. Thus Ind(γ̃, q) = k
for all q ∈ D(0, 21k ). Thus z k = q has k solutions counted with multiplicity, in Ω = D(0, 12 ).
Except for q = 0, these k solutions are all distinct.

Definition 20.4. A function ϕ : X → Y between two metric spaces is said to be an open

map if ϕ(U ) is open whenever U is open.

Theorem 20.5 (Open Mapping Theorem). Let V ⊂ C be a connected open set and
f ∈ H(V ) be non-constant. Then f : V → C is an open map.

Proof. Let U ⊂ V be a nonempty open set. Let q ∈ f (U ). We have to show that q is

an interior point of f (U ). Choose p ∈ U such that f (p) = q. Choose r > 0 such that
D(p, r) ⊂ U and f takes the value q on D(p, r) only at p. Let γ be ∂D(p, r). Then
the Argument Principle applies. It tells us that the number of solutions to f (w) = q
in D(p, r) is equal to Ind(γ̃, q) where γ̃ = f ◦ γ. Since f (p) = q, Ind(γ̃, q) 6= 0. Let
W be the component of C\γ̃ ∗ that contains q. Then Ind(γ̃, s) 6= 0 for all s ∈ W . By
the Argument Principle again, the equation f (w) = s has solutions in D(p, r). That is,
W ⊂ f (D(p, r)) ⊂ f (U ). Also, W is an open set. Thus, q ∈ int(f (U )), as required.

Remark 20.6. Open Mapping Theorem distinguishes complex differentiability and real
differentiability. For example, on the real line, the function f (x) = x2 is not an open map,
since it maps an open integral (−1, 1) to a non-open set [0, 1).

21 Counting Zeroes and the Open Mapping Theorem II

(10/8)
Say that γ0 and γ1 are two loops in C − {0}. Define γs for s ∈ [0, 1] by

γs (t) = (1 − s)γ0 (t) + sγ1 (t).

This is a homotopy from γ0 to γ1 in C. This is a homotopy in C − {0} provided γs (t) 6= 0

for all s, t. If γs (t) = 0 for some s, t then 0 is in the segment of [γ0 (t), γ1 (t)]. So γs is a
homotopy from γ0 to γ1 in C − {0} if and only if |γ1 (t) − γ0 (t)| < |γ1 (t)| + |γ0 (t)| for all
t. If |γ1 (t) − γ0 (t)| < |γ1 (t)| + |γ0 (t)| for all t, then Ind(γ0 , 0) = Ind(γ1 , 0).

Theorem 21.1 (Rouché’s Theorem). Let V ⊂ C be a connected open set. Let γ be a

closed curve in V such that Ind(γ, p) = 0 for all p ∈ C\V and Ind(γ, p) = 0 or 1 for

46
all p ∈ C\γ ∗ . Let Ω = {z ∈ C\γ ∗ : Ind(γ, z) = 1}. Let f, g ∈ H(V ) and suppose that
|f (z) − g(z)| < |f (z)| + |g(z)| for all z ∈ γ ∗ . Then f and g has same number of zeroes
counted with multiplicity in Ω.
Proof. If f and g are both nonconstant, then define γ̃f = f ◦ γ, γ̃g = g ◦ γ. We have

|γ̃f (t) − γ̃g (t)| < |γ̃f (t)| + |γ̃g (t)|

for all t by the hypothesis on f, g. So Ind(γ̃f , 0) = Ind(γ̃g , 0) by preliminary discussion.

By the Argument Principle, both of these count zeroes in Ω with multiplicity.
If f or g (or both) are constant then the claim follows by considering cases.

Example 21.2. Let f (z) = z n (z − 2) − 1. Take g(z) = z n (z − 2). Take γ to be ∂D(0, 1).
Then f (z) − g(z) = −1 and so to apply Rouché’s Theorem we need

1 < |z n (z − 2) − 1| + |z n (z − 2)|

for all z with |z| = 1. This is equivalent to

1 < |z n (z − 2) − 1| + |z − 2|

for all z with |z| = 1. Now 1 < |z − 2| for all z with |z| = 1 except z = 1. Note that the
other summand does not vanish at z = 1 and so we have the inequality. This implies that
f and g have the same number of zeroes counted with multiplicity, in Ω = D(0, 1). For
g this number is n. This means there are n zeroes of f in D(0, 1). By the Intermediate
Value Theorem, f has a zero slightly larger than 2.
Example 21.3. Let λ ∈ C − {0} and f (z) = z sin(z) − λ cos(z) and g(z) = z sin(z). I
claim that for a fixed λ, |f (z) − g(z)| ≤ |g(z)| on a rectangle provided that T and m are
large enough. This inequality is |λ| · | cos(z)| < |z| · | sin(z)| for any z in the rectangle (see
Figure 9). In general,

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

where z = x + iy. q q
(2m+1)π
On side 1 , we need |λ|·| sinh(y)| < |z| 1 + sinh2 (y). So |λ|·| sinh(y)| < 2 1 + sinh2 (y)
(2m+1)π
is sufficient. All that is necessary if |λ| ≤ 2 . This can be done by making m big
enough. Side 3 is similar.q q q
On side 2 , we need |λ|· cos2 (x) + sinh2 (T ) < |z| 1 + sinh2 (T ). So |λ|· cos2 (x) + sinh2 (T ) <
q q
T · sin2 (x) + sinh2 (T ) is enough. |λ| · 1 + sinh2 (T ) < T · sinh2 (T ) is enough. We can
do this by making T large. 4 is similar.
In the box, g has a double zero at 0, and a simple zero at ±π, ±2π, · · · , ±mπ. This is
2m + 2 zeroes inside the rectangle in total. Thus f also has this many once T and m are
large enough. The equation
z sin(z) = λ cos(z)

47
 iT 2

3 1

−(2m+1)π (2m+1)π
2 2

−iT 4

Figure 9:

is equivalent to
λ
tan(z) = .
z
This equation has 2m + 2 real solutions in the rectangle, so all solutions are real if λ > 0.
If λ < 0, then there are only 2m real solutions, so there must be 2 non-real solutions.
They are purely imaginary.

Theorem 21.4 (Hurwitz’s Theorem). Suppose that D is an open set, fn ∈ H(D), fn → f

uniformly on compact subsets of D, D(z, r) ⊂ D, and f has no zero on ∂D(z, r). Then
there exists N such that fn and f have the same number of zeroes in D(z, r) for all n > N .

Proof. Let  = minw∈∂D(z,r) |f (w)|, which exists and is non-zero since |f | is continuous
and ∂D(z, r) is compact. Since fn → f uniformly on compact subsets of D, there exists
N such that for w ∈ ∂D(z, r),

|fn (w) − f (w)| <  ≤ |f (w)| ≤ |f (w)| + |g(w)|

for all n > N . Then Rouché’s Theorem implies that fn and f have the same number of
zeroes in D(z, r) for all n > N .

48
22 Product Formula for Sine Function I (10/13)
The objective of this chapter is to prove the following two formulas
∞
Y z2
(22.1) sin(πz) = πz (1 − ) ∀z ∈ C,
n2
n=1

∞  
1 X 1 1
(22.2) π cot(πz) = + + ∀z ∈ C\Z.
z z−n z+n
n=1

(22.2) is basically a partial function decomposition of π cot(πz). The function has simple
0)
poles at every integer and the residue at each pole is 1 because π cot(πz) = (sin(πz)
sin(πz) (recall
f0
f has simple poles at zeroes of f with residue ord(f, p)). First aim is to prove (22.2).
Formula (22.2) comes from using Cauchy’s Integral Formula. The idea is to come up
with a function with π cot(πz) as a residue and the terms in the sum as residues. Then
show that as some parameter → ∞, the integral itself goes to zero.
The first thing I need is an estimate for the size of cot(πz) away from its poles.

Lemma 22.1. There is a constant M such that | cot(πz)| ≤ M if Im(z) ≥ 1 or Re(z) ∈

1
2 + Z.

Proof. Let z = x + iy. Then

| cos(πz)|2 cos2 (πx) + sinh2 (πy)

| cot(πz)|2 = = .
| sin(πz)|2 sin2 (πx) + sinh2 (πy)

If |y| ≥ 1 then

1 + sinh2 (πy) 1 1
| cot(πz)|2 ≤ 2 =1+ 2 ≤1+ .
sinh (πy) sinh (πy) sinh2 (π)

This shows that | cot(πz)| is bounded on the set |y| ≥ 1.

Now say x ∈ 12 + Z. Then

sinh2 (πy)
| cot(πz)|2 = ≤ 1.
1 + sinh2 (πy)
1
This shows | cot(πz)| is bounded on the set x ∈ 2 + Z.

Let N ≥ 1 be an integer. I will use the contour QN (see Figure 10). Fix z ∈ C\Z.
Assume that N is large enough that is inside the contour QN . Let f : C\(Z ∪ {z}) → C
be
π cot(πw)
f (w) = .
w−z

49
 ( 12 + N )i

−( 12 + N ) 1
2 +N

−( 12 + N )i

Figure 10:

The function f has simple pole at every point in Z ∪ {z}. The residues are
1
Res(f, n) = , ∀n ∈ Z,
n−z
Res(f, z) = π cot(πz).
Thus
Z N
1 X 1
f (w) dw = π cot(πz) +
2πi QN n−z
n=−N
1
R
by the Residue Theorem. If we could show limN →∞ 2πi QN f (w) dw = 0 then we get

N
X 1
π cot(πz) = lim
N →∞ z−n
n=−N

which is equivalent to what we wanted.

The required limit is true, but relies essentially on symmetry of QN .
To prove
N
X 1
π cot(πz) = lim ,
N →∞ z−n
n=−N

50
the only thing that remained was to show
Z
π cot(πw)
lim dw = 0.
N →∞ QN w−z
π cot(πw)
The key is to note that w 7→ w is even. Thus
Z
π cot(πw)
dw = 0
QN w
because contributions from opposite sides cancel. (Note: This is why the sum N 1
P
n=−N z−n
has to be symmetric.) Thus
Z Z   Z
π cot(πw) π cot(πw) π cot(πw) πz cot(πw)
dw = − dw = dw.
QN w − z QN w − z w QN w(w − z)

We know that |π cot(πw)| ≤ M on Q∗N by Lemma 22.1. So

M · |z|
Z
πz cot(πw)
dw ≤ · 4(2N + 1)
QN w(w − z) (N + 2 )(N + 12 − |z|)
1

1
(provided |z| < N + )
2
→ 0 (as N → ∞, as required).
So
N
X 1
π cot(πz) = lim
N →∞ z−n
n=−N
∞
X 
1 1 1
= + +
z z−n z+n
n=1
for z ∈ C\Z.
Remark 22.2. If we set z = ia in the formula, where a ∈ R\{0}, then we get
∞  
1 X 1 1
π cot(πia) = + + ,
ia ia − n ia + n
n=1
∞
cos(πia) 1 X 2ia
⇒π = + ,
sin(πia) ia −a2 − n2
n=1
∞
cosh(πa) 1 X 2ia
⇒π = + ,
i sinh(πa) ia −(a2 + n2 )
n=1
∞
cosh(πa) 1 X 2a
⇒π = + ,
sinh(πa) a a2 + n2
n=1
∞  
X 1 1 1
⇒ 2 2
= π coth(πa) − .
n +a 2a a
n=1

51
 For example,
∞
X 1 1
= (π coth(π) − 1) .
n2 + 1 2
n=1
Also, you can solve the Basel problem by letting a → 0. You could also P∞differentiate
with respect to a (needs justification). Then as a → 0, you could evaluate n=1 n14 .

23 Product Formula for Sine Function II (10/15)

Q∞ z2
The next task is to show that p(z) = πz n=1 (1 − n2
) is a holomorphic function.
Lemma 23.1. Let w1 , · · · , wn ∈ C. Then
 
n
Y Xn
1− (1 + wj ) ≤ exp  |wj | − 1.
j=1 j=1

Proof. Well,
n
Y n
X X X
1− (1 + wj ) = wj + wj 1 wj 2 + wj 1 wj 2 wj 3 + · · · + w1 w2 · · · wn
j=1 j=1 1≤j1 <j2 ≤n 1≤j1 <j2 <j3 ≤n
n
X X X
≤ |wj | + |wj1 wj2 | + |wj1 wj2 wj3 | + · · · + |w1 w2 · · · wn |
j=1 1≤j1 <j2 ≤n 1≤j1 <j2 <j3 ≤n
 2  3  n
n n n n
X 1 X 1 X 1 X
≤ |wj | +  |wj | +  |wj | + · · · +  |wj |
2 3! n!
j=1 j=1 j=1 j=1

(by multinomial theorem)

 
Xn
≤ exp  |wj | − 1.
j=1

Next, we want to think about convergence of a product

∞
Y
P = (1 + wj ).
j=1

We make the partial products

N
Y
PN = (1 + wj ).
j=1
We say P converges if PN → P .

52
 Suppose that (wj ) is a sequence of complex numbers and ∞
P
Lemma 23.2. Q∞ j=1 |wj | con-
verges. Then j=1 (1 + wj ) converges and it doesn’t converge to 0 unless one of the factors
is zero.

Proof. Let PN = N
Q
j=1 (1 + wj ). First, we show that (PN ) is bounded. Well,

N
Y
|1 − PN | = 1 − (1 + wj )
j=1
 
XN
≤ exp  |wj | − 1
j=1
 
X∞
≤ exp  |wj | − 1
j=1

and so  
∞
X
|PN | ≤ 1 + |1 − PN | ≤ exp  |wj | = B.
j=1

Say N1 < N2 . Then

   
N2
Y N2
X
|PN1 − PN2 | = |PN1 | · |1 − (1 + wj )| ≤ B · exp  |wj | − 1
j=N1 +1 j=N1 +1

by Lemma 23.1. Let  > 0, then we can choose P∞− 1) < . We can
PN2 η > 0 such that B(exp(η)
choose N such that if N ≤ N1 < N2 then j=N1 +1 |wj | < η because j=1 |wj | converges.
If N ≤ N1 < N2 then
|PN1 − PN2 | ≤ B · |exp(η − 1)| < .
Thus (PN ) is Cauchy and so converges to some P .
Lastly, we have to show that P 6= 0 provided that wj 6= −1 for all j. If M ≥ 1 is an
integer then  
∞
Y X∞
1− (1 + wj ) ≤ exp  |wj | − 1.
j=M j=M

By making M sufficiently large, we can make

 
∞
X 1
exp  |wj | − 1 < .
2
j=M

53
Thus for M this large,
∞
Y 1
1− (1 + wj ) <
2
j=M
Q∞ QM −1
and so j=M (1 + wj ) 6= 0. Also j=1 (1 + wj ) 6= 0 and so

M
Y −1 ∞
Y
P = (1 + wj ) · (1 + wj ) 6= 0
j=1 j=M

as required.

Remark
Q 23.3. This result applies P to products of functions uniformly.
If ∞ j=1 (1 + g j (z)) is such that ∞
j=1 |gj (z)| converges uniformly on some set then the
product also converges uniformly and the zeroes are the obvious ones.

24 Product Formula for Sine Function III (10/17)

P∞ z2
The series n=1 | n2 | converges uniformly on all compact subsets of C. Thus
∞
Y z2
p(z) = πz (1 − )
n2
n=1

converges uniformly on compact subsets of C and is zero only at integers. The partial
products
N
Y z2
pN (z) = πz (1 − 2 )
n
n=1

are polynomials. So they are all holomorphic. It follows that p is holomorphic on C.

Moreover,
p0N → p0
uniformly on compact subsets of C. It follows that on compact subsets of C\Z,

p0N uniformly p0
−→ .
pN p
0 f0
In general, ff is called the logarithmic derivative of f , denoted as D. That is, D(f ) = f .
Main thing about D:
D(f g) = D(f ) + D(g).

54
So
N
Y z2
D(pN ) = D(πz (1 − ))
n2
n=1
N
X z2
= D(π) + D(z) + D(1 − )
n2
n=1
N
1 X − n2z2
=0+ + z2
z
n=1 1 − n2
N
1 X −2z
= +
z n2 − z 2
n=1
N
1 X 2z
= +
z z2 − n2
n=1
N  
1 X 1 1
= + +
z z−n z+n
n=1
N
X 1
=
z−n
n=−N

for z ∈ C\Z. So
N
X 1
D(p) = lim D(pN ) = lim
N →∞ N →∞ z−n
n=−N
= π cot(πz)
= D(sin(πz)).

Now we know that D(p) = D(sin(πz)). Thus

 
sin(πz)
D = D(sin(πz)) − D(p) = 0
p

on C\Z. So  
d sin(πz)
=0
dz p
sin(πz)
and (since C\Z is connected) p is constant. There is some c ∈ C such that
∞
Y z2
sin(πz) = cπz (1 − ).
n2
n=1

55
So
∞
sin(πz) Y z2
= cπ (1 − 2 ).
z n
n=1

So
∞
sin(πz) Y z2
lim = cπ lim (1 − 2 ),
z→0 z z→0 n
n=1

hence
π = cπ,
and so
c = 1.
Thus
∞
Y z2
sin(πz) = πz (1 − )
n2
n=1

for all z ∈ C.

Example 24.1. For example, if we set z = 21 , we will get

∞
π π Y 1
sin( ) = (1 − 2 ),
2 2 4n
n=1
∞
Y 1 2
⇒ (1 − 2
)= ,
4n π
n=1
∞
Y 4n2 − 1 2
⇒ 2
= ,
4n π
n=1
∞
Y (2n − 1)(2n + 1) 2
⇒ 2
= ,
(2n) π
n=1
∞
Y (2n)2 π
⇒ = .
(2n − 1)(2n + 1) 2
n=1

56
25 Inverses of Holomorphic Functions I (10/20)
Theorem 25.1. Let V ⊂ C be an open set and f ∈ H(V ) be a one-to-one function. Then
f 0 (z) 6= 0 for all z ∈ V .
Proof. Suppose that p ∈ V and f 0 (p) = 0. Note that f 0 is not identically zero since f is
one-to-one. So p is an isolated zero of f 0 . Choose r > 0 so that D(p, r) ⊂ V and the only
zero of f 0 on D(p, r) is at p. Let γ = ∂D(p, r) and γ̃ = f ◦ γ. We know that Ind(γ̃, f (p))
is equal to the number of zeroes of f − f (p) in D(p, r). By hypothesis, f 0 (p) = 0 and so
Ind(γ̃, f (p)) ≥ 2. Thus Ind(γ̃, q) ≥ 2 for all q close enough to f (p). Choose q close to f (p)
and consider the solutions to f (z) = q with z ∈ D(p, r). These solutions are all simple
zeroes of f − q. Thus there must be at least two of them. That is, there is z1 , z2 ∈ D(p, r)
with f (z1 ) = q = f (z2 ). This contradicts f being one-to-one. So f 0 (z) 6= 0 at all points
z ∈V.

Remark 25.2. (i) The implication cannot be reversed. An example would be f (z) =
exp(z).
(ii) If f is one-to-one, then all solutions to f (z) = q are simple.
Theorem 25.3 (Inverse Function Theorem for Holomorphic Functions). Let V ⊂ C be
an open set, f ∈ H(V ) be one-to-one, W = f (V ) and g : W → V the inverse function.
Then g ∈ H(W ).
Proof. Differentiability is a local property, so it suffices to show that if q ∈ W then g is
differentiable on some open set containing q. Take q ∈ W and choose p ∈ V such that
f (p) = q and r > 0 such that D(p, r) ⊂ V . Let U = f (D(p, r)). By Open Mapping
Theorem, U is open. Also, it contains q. Let u ∈ U and consider the integral
wf 0 (w)
Z
1
dw = J.
2πi ∂D(p,r) f (w) − u
First, note that J makes sense since u 6∈ f (∂D(p, r)). Second, the integrand has one
singularity in D(p, r). This occurs at whichever point a ∈ D(p, r) satisfies f (a) = u. We
0 (w) 0
know that f − u has a simple zero at a. Thus f f(w)−u = (ff(w)−u)
(w)−u has a simple pole at a
wf 0 (w)
with residue equal to 1. (Slightly imprecisely - since a might be 0) we can say that f (w)−u
has a simple pole at a with residue a. Thus by Residue Theorem,
wf 0 (w)
Z
1
J= dw = a · 1 = a.
2πi ∂D(p,r) f (w) − u
Note a = g(u). That is,
wf 0 (w)
Z
1
g(u) = dw
2πi ∂D(p,r) f (w) − u
for all u ∈ U . By the lemma about integral being holomorphic, it follows that g|U is
holomorphic.

57
Lemma 25.4. Let V ⊂ C be a convex open set. Let f ∈ H(V ). Then if z1 6= z2 ,
z1 , z2 ∈ V , f (zz11)−f
−z2
(z2 )
lies in the closed convex hull of f 0 (V ).
Remark 25.5. If A ⊂ C is a set then there is a minimum closed convex set that contains
A. It is called the closed convex hull of A. One way to get it is to consider the intersection
of all closed convex sets that contain A.
Proof of Lemma 25.4. We know that
Z
f (z2 ) − f (z1 ) = f 0 (w) dw (note [z1 , z2 ] ⊂ V because V is convex)
[z1 ,z2 ]
Z 1
0
= f ((1 − t)z1 + tz2 )(z2 − z1 )dt
0
and so
1
f (z2 ) − f (z1 )
Z
= f 0 ((1 − t)z1 + tz2 )dt
z2 − z1 0
n
X j j 1
= lim f 0 ((1 − )z1 + z2 ) ·
n→∞ n n n
j=1
def
= lim Sn .
n→∞
Pn
But j=1 n1 = 1 and so Sn lies in the closed convex hull of f 0 (V ) for all n. Then limn→∞ Sn
also lies in the closed convex hull of f 0 (V ) because it is closed.

26 Inverses of Holomorphic Functions II (10/22)

Theorem 26.1. Let V ⊂ C be a convex open set, f ∈ H(V ) a non-constant holomorphic
function, K a closed convex set such that 0 6∈ int(K) and f 0 (z) ∈ K for all z ∈ V . Then
f is one-to-one.
Proof. We must first consider functions with constant derivative. If the derivative is
constant then f (z) = az + b with a 6= 0. This is indeed one-to-one.
Now suppose f is not a linear function. Fix z1 ∈ V and consider gz1 : V \{z1 } → C
defined by
f (z) − f (z1 )
gz1 (z) = .
z − z1
Then gz1 ∈ H(V \{z1 }) and gz1 is not constant. Thus gz1 (V \{z1 }) is an open set by Open
Mapping Theorem. Thus
f (z2 ) − f (z1 ) [
Ω={ |z1 , z2 ∈ V, z1 6= z2 } = gz1 (V \{z1 })
z 2 − z1
z1 ∈V

is an open set. By our previous result using convexity, Ω ⊂ K and so Ω ⊂ int(K). Thus
0 6∈ Ω and so f is one-to-one.

58
Remark 26.2. A typical example would be f ∈ H(D(0, 1)) such that Re(f 0 (z)) ≥ 0.

Theorem 26.3. Let V ⊂ C be open and f ∈ H(V ), z0 ∈ V , and f 0 (z0 ) 6= 0. Then there
is an open set U ⊂ V containing z0 such that f |U is one-to-one.

Proof. Note that f is not constant. Choose ρ > 0 such that 0 6∈ D(f 0 (z0 ), ρ). Choose
r > 0 such that f 0 (D(z0 , r)) ⊂ D(f 0 (z0 ), ρ). By Theorem 26.1 with V = D(z0 , r), K =
D(f 0 (z0 ), ρ), we can conclude that f |D(z0 ,r) is one-to-one.

Theorem 26.4. Let V ⊂ C be open, f ∈ H(V ), z0 ∈ V and suppose z0 is an m-fold

zero of f . Then there is an open set U ⊂ V containing z0 and a g ∈ H(U ) such that
f (z) = (g(z))m for all z ∈ U . Moreover, g has a simple zero at z0 .

Proof. We can write

f (z) = (z − z0 )m h(z)
where h ∈ H(V ) and h(z0 ) 6= 0. By choosing a small enough disk centered at z0 for U we
may choose
(1) U is simple-connected.
(2) h|U does not vanish.
(1), (2) imply that h has a logarithm on U . Thus h has an m-th root. We get k ∈ H(U )
such that k m = h. Then g(z) = (z − z0 )k(z) works.

59
27 Conformal Mappings I (10/27)
Definition 27.1. A Riemann sphere is a set C∞ = C ∪ {∞} where ∞ 6∈ C.
We want to make C∞ into a metric space. We consider the stereographic projection
2x 2y |z|2 − 1
S(z) = ( , , ) ∈ S2.
1 + |z|2 1 + |z|2 1 + |z|2
If we extend S : C → S 2 − {(0, 0, 1)} to S : C∞ → S 2 by

S(∞) = (0, 0, 1),

then S becomes one-to-one and onto.

(0, 0, 1)

S(z)

z = x + iy

Figure 11: Stereographic Projection

We can make C∞ into a metric space by saying d∞ (p, q) = kS(p) − S(q)k2 . This is
automatically a metric on C∞ .
Note that d∞ |C is not comparable to the usual distance. However, if B ⊂ C is bounded
then d∞ |B is comparable to the usual metric. This means that convergence of sequence
in C is unchanged provided the sequence is bounded. Explicitly, (zn ) ⊂ C∞ and z ∈ C∞
then zn → z in C∞ iff (1) zn ∈ C for all sufficiently large n and (2) zn → z in the usual
sense once ∞ terms are discarded. Also, zn → ∞ if |zn | → ∞ with the convention that
|∞| = ∞.
The function k : C∞ → C∞ given by k(z) = z1 if z ∈ C\{0}, k(0) = ∞, k(∞) = 0
is continuous, bijective, and its own inverse. One way to see this is to compute that the
corresponding map of S 2 is (u1 , u2 , u3 ) 7→ (u1 , −u2 , −u3 ). This is continuous, so k is.
Definition 27.2. Let V ⊂ C∞ be an open subset of C∞ . We say h : V → C∞ is
holomorphic if
(i) h is continuous.
1
(ii) The function h(z), h( z1 ), h(z) , h(11 ) are holomorphic (in the usual sense) on the set where
z
their argument lies in V ∩ C and they are finite.

60
Example 27.3. Define ϕ : C∞ → C∞ by ϕ(z) = z+2 z+1 if z 6∈ {−1, ∞}, ϕ(−1) = ∞, ϕ(∞) =
1. Then ϕ : C∞ → C∞ is holomorphic. To check the first condition, it is easy to see that
z+2
ϕ is continuous for most points. ϕ is continuous at −1 because limz→−1 z+1 = ∞, ϕ is
z+2
continuous at ∞ because limz→∞ z+1 = 1. To check the second condition, I need
(1) ϕ(z) = z+2z+1 is holomorphic on C\{−1},
(2) ϕ( z ) = 1+2z
1
1+z is holomorphic on C\{0, −1},
1 z+1
(3) ϕ(z) = z+2 is holomorphic on C\{−2},
(4) ϕ(11 ) = 1+2z
1+z
is holomorphic on C\{0, − 12 }.
z
This confirms that ϕ : C∞ → C∞ is holomorphic.
Suppose that V ⊂ C is an open set and f : V → C is holomorphic. What does this
mean more concretely? Say that V is connected. Then there are two possibilities. One
is f is constantly ∞. If this doesn’t happen, then the set of points at which f takes the
value ∞ is relatively closed and discrete in V . So there exists S ⊂ V that is closed in V ,
every point of S is isolated in S, f |V \S ∈ H(V \S), and f has a pole at each point of S.
This class of functions is denoted by M (V ) and they are called meromorphic functions on
V.

28 Conformal Mappings II (10/29)

Lemma 28.1. Let V ⊂ C∞ be open and connected, S ⊂ V be closed and nonempty, and
C be a component of V \S. Then clV (C) ∩ S 6= ∅.
Proof. Suppose clV (C) ∩ S = ∅. Then clV (C) ⊂ V \S is a connected set and C ⊂ clV (C)
and so C is closed in V . Since disks in C∞ are connected, C is also open in V \S and
hence in V . Since V is connected, either C = V or C = ∅. Components are never empty
and so C = V . But then S = ∅, contrary to our assumption.

Proposition 28.2. Let V ⊂ C∞ be open and connected, h : V → C∞ a holomorphic

function and Sp = {z ∈ V |h(z) = p}. If h is not constant, then every point of Sp is
isolated in Sp . Also Sp is closed in V .
Proof. Sp = h−1 ({p}) is a closed set in V because h is continuous. By replacing h by
h − p or h1 , we may assume that we are considering S0 . We know that S∞ is closed in V .
Suppose that S0 has a limit point in S0 . Let C be the component of V \S∞ that contains
this limit point. Both h(z) and h( z1 ) are holomorphic where they are defined on C. From
the Identity Principle, h is constantly zero on C. Thus h is identically zero on clV (C)
because h is continuous. If S∞ 6= ∅, then Lemma 28.1 implies that clV (C) ∩ S∞ 6= ∅ and
so h takes the value 0 and ∞ simultaneously at some point, which it can’t. Now, we know
that C = V and h ≡ 0 on V . It follows that if h is not constant then every point of S0 is
isolated in S0 .

Proposition 28.3. If f : C∞ → C∞ is holomorphic, then f is either constant or onto.

61
Proof. Let f : C∞ → C∞ be a holomorphic function and assume that f omits a value. We
1
may assume that it omits ∞. (If it omits q 6= ∞ then f (z)−q omits ∞). Thus f : C∞ → C.
Now f (C∞ ) is compact and so bounded. This means f |C is entire and bounded. Thus f |C
is constant by Liouville’s Theorem. Since f is continuous, f is constant.

Theorem 28.4. If f : C∞ → C∞ is holomorphic, then there are polynomials P, Q ∈ C[z]

P (z)
such that f (z) = Q(z) , ∀z ∈ C.

Proof. If f is constant, then this is clear. Assume f is not constant. Then S0 and S∞ are
closed subsets of C∞ in which every point is isolated. Thus they are both finite sets. At
each point of S0 ∩ C, f has a zero of some order. We can choose a polynomial P̃ having
zeroes of the same order at each of these points. Similarly, we may find a polynomial Q
such that Q1 has poles at each point of S∞ ∩ C with the same order as the pole of f at
Qf
this point. The function P̃
: C∞ → C∞ is holomorphic which does not take the zero
Qf
or ∞ on C. Since two values are missing on C, P̃
is not onto, thus it is constant. So
cP̃ P
f= Q = Q.

By definition, Aut(C∞ ) is the set of all holomorphic functions f : C∞ → C∞ that

are both one-to-one and onto. Note that if f ∈ Aut(C∞ ) then f −1 : C∞ → C∞ is
automatically holomorphic.

Definition 28.5. A continuous function ϕ : C∞ → C∞ is said to be a linear-fractional

transformation if there are a, b, c, d ∈ C such that ad − bc 6= 0 and
az + b
ϕ(z) =
cz + d
for all z ∈ C.

Theorem 28.6. Aut(C∞ ) consists precisely of the linear-fractional transformations.

Proof. Suppose f : C∞ → C∞ is an automorphism which is non-constant, so

P (z)
f (z) =
Q(z)

for some polynomials P, Q which may be assumed relatively prime. Thus S0 = Z(P )
and S∞ = Z(Q) both have one point. Moreover, p ∈ S0 and q ∈ S∞ are simple zeroes
and poles. Thus P, Q are linear. Thus f (z) = az+b
cz+d and ad − bc 6= 0 because f is not
constant.

29 Conformal Mappings III (11/1)

Proposition 29.1. Let ϕ ∈ Aut(C) then ϕ(z) = az + b for some a ∈ C∗ , b ∈ C.

62
Proof. Let ϕ ∈ Aut(C). Then ϕ−1 : C → C is holomorphic, and in particular, continuous.
If K ⊂ C is a compact set, then ϕ−1 (K) is also compact. This implies that limz→∞ ϕ(z) =
∞. If we define ϕ(∞) = ∞, then we extend ϕ to an element of Aut(C∞ ). Thus there
Az+B
exists A, B, C, D with AD − BC 6= 0 and ϕ(z) = Cz+D , ∀z ∈ C. Since ϕ(∞) = ∞, C = 0.
A B A
Thus D 6= 0 and so ϕ(z) = D z + D , ∀z ∈ C. Finally, D 6= 0 because ϕ is one-to-one.

Definition 29.2. Let U, V ⊂ C be open sets. A function f : U → V that is holomorphic,

one-to-one and onto is said to be biholomorphic. A biholomorphic map is also called a
conformal mapping between U and V .

Definition 29.3. A conformal map f : U → V is defined to be a map that is holomorphic

and has non-vanishing derivative.

Example 29.4. exp : C → C is a conformal map.

Remark 29.5. Lots of people use “conformal map” as a synonym for “conformal equiv-
alence”. If the conformal map is from one open set to another open set, then it actually
means conformal equivalence.

We want to consider

M = {ϕ : C∞ → C∞ |ϕ is fractional linear}.

Note that M is a group which acts on C∞ . The action is effective (namely, if ϕ(z) =
z, ∀z ∈ C∞ , then ϕ = e.)  
az+b a b
The form ϕ(z) = cz+d with ad − bc 6= 0 suggests considering the matrix A = .
c d
This is an invertible matrix by the condition ad − bc 6= 0. The group of all invertible
2 × 2 complex matrices is usually denoted by GL2 (C). One can check that the map
GL2 (C) → M given by A 7→ ϕA is a homomorphism. That is,

ϕAB = ϕA · ϕB .

This homomorphism is onto.

The next question is when is ϕA = id? If so then az+b cz+d = z for all z. So cz +
2

(d − a)z − b   all z ∈ C. Thus c = 0, b = 0 and a = d. Thus ϕA = id if and

= 0 for
a 0
only if A = = aI2 for some a ∈ C∗ . It follows that M ∼ = GL2 (C)/N where
0 a
  
a 0
N = Z(GL2 (C)) = : a ∈ C∗ .
0 a
Let C be the collection of all lines and circles in C (in C∞ , lines should include ∞).
Note that X ∈ C has an equation of the form

A(x2 + y 2 ) + Bx + Cy + D = 0.

63
This may be rewritten as
A|z|2 + Re(ζz) + D = 0
where ζ = B − Ci. I want to verify that if X ∈ C and ϕ ∈ M then ϕ(X) ∈ C. To achieve
this, I am going to find a generating set for M. If c = 0, then
az + b a b
= z+ .
cz + d d d
If c 6= 0, then
a
az + b + d) − ad
c (cz c +b
=
cz + d cz + d
ad
a b − c
= +
c cz + d
a −∆
= + (∆ = ad − bc)
c c(cz + d)
a −∆/c
= + .
c cz + d
These two expressions show that M is generated by

B = {Tw |w ∈ C} ∪ {Mw |w ∈ C× } ∪ {J}

where
Tw (z) = z + w,
Mw (z) = wz,
1
J(z) = .
z
What this means is that if ϕ ∈ M then ϕ may be expressed as a composition of elements
of B.
If
A|z|2 + Re(ζz) + D = 0
is the equation of an element of C and w = J(z) = z1 then z = 1
w and so the equation of
the image under J is
1 1
A| |2 + Re(ζ ) + D = 0,
w w
A ζw
⇒ + Re( 2 ) + D = 0,
|w|2 |w|
⇒ A + Re(ζw) + D|w|2 = 0,
⇒ A + Re(ξw) + D|w|2 = 0,
which is another element of C. To verify the corresponding thing for Tw and Mw , for Tw
use A(x2 + y 2 ) + Bx + Cy + D = 0 form, for Mw use A|z|2 + Re(ζz) + D = 0 form.

64
30 Conformal Mappings IV (11/3)
Now let’s consider the action of M on C∞ .
Definition 30.1. If G acts on X then G is m-transitive if given any two lists p1 , · · · , pm
and q1 , · · · , qm of distinct points, there is some g ∈ G such that gpj = qj for 1 ≤ j ≤ m.
The action is sharply m-transitive if the g is unique.
Theorem 30.2. The action of M on C∞ is sharply 3-transitive.
Proof. Given distinct points p1 , p2 , p3 , we can show that there is a ϕ ∈ M such that
ϕ(p1 ) = 0, ϕ(p2 ) = 1, ϕ(p3 ) = ∞. This will establish 3-transitivity. Build ϕ up by
1
composition. Start with ϕ1 which is the identity if p3 = ∞ and ϕ1 (z) = z−p 3
if not. Then
ϕ1 (p3 ) = ∞ and we let q1 = ϕ1 (p1 ) and q2 = ϕ1 (p2 ). Note that q1 and q2 are distinct and
not ∞. Let ϕ2 be ϕ2 (z) = z − q1 . Finally, take ϕ3 to be ϕ3 (z) = (q2 − q1 )−1 z, which is
possible since q1 6= q2 . Take ϕ = ϕ3 ◦ ϕ2 ◦ ϕ1 , then it works.
To show that ϕ is unique, we need only that if ψ ∈ M and ψ(0) = 0, ψ(1) = 1, ψ(∞) =
∞ then ψ = id. Let ψ(z) = az+b cz+d with ad − bc 6= 0. Since ψ(∞) = ∞, c = 0 and the
map may be written as ψ(z) = Az + B. ψ(0) = B = 0 and then ψ(1) = A = 1. So
ψ(z) = z.

Recall from geometry that given three distinct points p1 , p2 , p3 , there is a unique ele-
ment of C through all three. (If three points are co-linear, there is a unique line; if three
points are not co-linear, there is a unique circle.) We conclude that M acts transitively
on C. Take C1 , C2 ∈ C, choose three distinct points p1 , p2 , p3 on C1 and q1 , q2 , q3 on C2 .
Choose ϕ ∈ M such that ϕ(pj ) = qj (1 ≤ j ≤ 3). We must have ϕ(C1 ) = C2 .

31 Conformal Mappings V (11/5)

π + = {x + iy|y > 0} is conformally equivalent to D by a fractional linear transformation.
First, we construct ψ : π + → D.
z ∈ π + ⇔ |z − i| < |z + i|
z−i
⇔ <1
z+i
z−i
⇔ ∈ D.
z+i
 
z−i 1 −i
So ψ(z) = z+i is a conformal equivalence from π+ to D. The matrix of ψ is whose
1 i
   
1 i i i i
inverse is 2i ∼ . So
−1 1 −1 1
iz + i 1+z
ψ −1 (z) = =i· = ψc (z).
−z + 1 1−z

65
So ψc (z) : D → π + is a conformal equivalence, called the Cayley transform.
Next, we want to study the disk D in great detail. The first question is what is Aut(D)?
Theorem 31.1 (Schwarz Lemma). Let f : D → D be holomorphic with f (0) = 0. Then
|f (z)| ≤ |z| for all z ∈ D and |f 0 (0)| ≤ 1. Moreover, if |f (w)| = |w| for some w ∈ D − {0}
or |f 0 (0)| = 1 then f (z) = cz for some c ∈ C with |c| = 1.
Proof. Let g : D → C be (
f (z)
g(z) = z , if z 6= 0
0
f (0), if z = 0.
By Riemann’s Removable Singularity Theorem, g is holomorphic. On |z| = r, |g(z)| ≤ 1r
for 0 < r < 1. This means that |g(z)| ≤ 1r for all z with |z| ≤ r by Maximum Modulus
Theorem. It follows that |g(z)| ≤ 1 for all z ∈ D. This gives the first two conclusions.
For the “moreover”, g would achieve a maximum modulus in D and hence be constant.

We want to use Schwarz Lemma to determine Aut(D).

Proposition 31.2. If ϕ ∈ Aut(D) such that ϕ(0) = 0 then ϕ(z) = cz for some c ∈ C with
|c| = 1.
Proof. By the Schwarz Lemma, |ϕ0 (0)| ≤ 1. Let ψ : D → D be ϕ−1 . By the Schwarz
Lemma, |ϕ−1 (0)| ≤ 1. By the Chain Rule,

|ϕ0 (0)||ψ −1 (0)| ≤ 1.

Thus |ϕ0 (0)| = |ψ 0 (0)| = 1 and the conclusion follows from the Schwarz Lemma.

The next step is to show that Aut(D) acts transitively on D.

For α ∈ D let ϕα be the map
α−z
ϕα (z) = .
1 − αz
Note that ϕα is holomorphic on C\{ α1 }. In particular, ϕα is holomorphic on the disk D
because | α1 | = |α|
1
> 1.
Next,
α−z α−z
1 − |ϕα (z)|2 = 1 − ·
1 − αz 1 − αz
(1 − αz)(1 − αz) − (α − z)(α − z)
=
|1 − αz|2
1 − |α| − |z|2 + |α|2 · |z|2
2
=
|1 − αz|2
(1 − |α|2 )(1 − |z|2 )
= .
|1 − αz|2

66
Notice that if z ∈ D then it follows that 1 − |ϕα (z)|2 > 0 and so ϕα (z) ∈ D. Thus
ϕα : D → D. Note that ϕα (0) = α and ϕα (α) = 0.
Next, we calculate ϕα ◦ ϕα . Of course, you could do this directly, but there are other
ways. Here is the one.
α−z
ϕα (z) = .
1 − αz
So
−(1 − αz) − (α − z)(−α)
ϕ0α (z) =
(1 − αz)2
|α|2 − 1
= .
(1 − αz)2
|α|2 −1
In particular, ϕ0α (0) = |α|2 − 1 and ϕ0α (α) = (1−|α|2 )2
= 1
|α|2 −1
. Thus

(ϕα ◦ ϕα )0 (0) = ϕ0α (ϕα (0)) · ϕ0α (0)

= ϕ0α (α) · ϕ0α (0)
1
= · (|α|2 − 1)
|α|2 − 1
= 1.
Also,
(ϕα ◦ ϕα )(0) = ϕα (ϕα (0))
= ϕα (α)
= 0.
By the Schwarz Lemma, (ϕα ◦ ϕα )(z) = cz for some c ∈ C with |c| = 1, and actually c = 1
because (ϕα ◦ ϕα )0 (0) = 1. Thus ϕα−1 = ϕα . In particular, ϕα ∈ Aut(D). This shows that
Aut(D) acts transitively: If α, β ∈ D then (ϕα ◦ ϕβ )(β) = α.
Theorem 31.3. If ψ ∈ Aut(D) then ψ can be expressed uniquely in either of the forms
(31.1) ψ = Mc ◦ ϕα ,

(31.2) ψ = ϕ β ◦ Ma ,
where α, β ∈ D, a, c ∈ C with |a| = |c| = 1, and Ma , Mc are the maps given by multiplication
by a and c respectively.
Proof. Let ψ ∈ Aut(D). Let β = ψ(0). Consider ϕβ ◦ ψ ∈ Aut(D). We have (ϕβ ◦ ψ)(0) =
ϕβ (β) = 0 and so ϕβ ◦ ψ = Ma for some a ∈ C with |a| = 1. Thus ψ = ϕβ ◦ Ma . This
gives 31.2. If ψ = ϕβ ◦ Ma then ψ(0) = ϕβ (Ma (0)) = ϕβ (0) = β and so β is uniquely
determined as ψ(0). Then Ma = ϕβ ◦ ψ is also uniquely determined. Thus a = (ϕβ ◦ ψ)0 (0)
is uniquely determined. To get 31.1, apply 31.2 to ψ −1 .

67
 This tells us that as a set

Aut(D) ↔ D × {w ∈ C||w| = 1}.

Unfortunately, it is “hard” to compute the group operation in this representation:

(Mc1 ◦ ϕα1 ) ◦ (Mc2 ◦ ϕα2 ) = horrible expression.

32 Conformal Mappings VI (11/10)

α−u
Given w, z ∈ D define d(w, z) = |ϕw (z)|. (Recall ϕα (u) = 1−αu .) Actually, this is a metric
on D.
w−z
d(w, z) = .
1 − wz
This shows that d(w, z) ≥ 0 and d(w, z) = 0 only when z = w.

z−w |z − w| |w − z|
d(z, w) = = = = d(w, z).
1 − zw |1 − zw| |1 − wz|

The only thing that remains to check is the triangle inequality, which is postponed. d is
called the pseudohyperbolic metric.
Let U ⊂ C be an open set, γ : [0, 1] → U be a piecewise smooth curve. One can define
Z 1
L(γ) = |γ 0 (t)| · ρ(γ(t)) dt
0

where ρ : U → [0, ∞) (is a density function).

Theorem 32.1 (Pick’s Theorem, Schwarz-Pick Theorem, Invariant Schwarz Lemma). Let
f : D → D be a holomorphic function. Then
(i) d(f (w), f (z)) ≤ d(w, z), ∀w, z ∈ D;
|f 0 (z)| 1
(ii) 1−|f (z)|2
≤ 1−|z| 2 , ∀z ∈ D.

If (i) is an equality for any z 6= w or (ii) is an equality for any z ∈ D, then f ∈ Aut(D).
Conversely, (i) and (ii) are equalities if f ∈ Aut(D).

Proof. Let z ∈ D. Define α = f (z) and g = ϕα ◦ f ◦ ϕz . Note that g : D → D is

holomorphic, g(0) = 0, and g ∈ Aut(D) if and only if f ∈ Aut(D). From Schwarz Lemma
we know that |g(w)| ≤ |w| for all w ∈ D. Then

|ϕα (f (ϕz (w)))| ≤ |w|

for all w ∈ D and so

|ϕα (f (ϕz (ϕz (u))))| ≤ |ϕz (u)|

68
for all u ∈ D which implies that ϕα (f (u))| ≤ |ϕz (u)| for all u ∈ D (ϕ−1
z = ϕz .) Recall that
α = f (z) and so this inequality is

|ϕf (z) (f (u))| ≤ |ϕz (u)|

for all u ∈ D. That is,

d(f (z), f (w)) ≤ d(z, u).
This establishes (i).
If f ∈ Aut(D) then so is g and so we get |g(w)| = |w|. If you start from this point,
you get d(f (z), f (w)) = d(z, w) if f ∈ Aut(D). If d(f (z), f (w)) = d(z, w) for z 6= w then
|g(w)| = |w| for w 6= 0. So g ∈ Aut(D) and then f ∈ Aut(D) too.
To get (ii), we use |g 0 (0)| ≤ 1 with equality if and only if g ∈ Aut(D). The chain rule
tells us that
g 0 (0) = ϕ0α (α) · f 0 (z) · ϕ0z (0)
1
= · f 0 (z) · (1 − |z|2 )
1 − |α|2
1
= · f 0 (z) · (1 − |z|2 )
1 − |f (z)|2

and so
|f 0 (z)|
· (1 − |z|2 ) = |g 0 (0)| ≤ 1.
1 − |f (z)|2
This gives
|f 0 (z)| 1
2
≤ .
1 − |f (z)| 1 − |z|2
The case of equality follows as before.

Remark 32.2. Let f : D → D be holomorphic but not an automorphism. Suppose that

f (p) = p, f (q) = q for p 6= q in D. Then

d(f (p), f (q)) < d(p, q)

by Schwarz-Pick Lemma, which is equivalent to d(p, q) < d(p, q), a contradiction. Thus
such f have at most one fixed point in D.

We will have to verify that d satisfies the triangle inequality. We want

d(u, z) ≤ d(u, w) + d(w, z), ∀u, w, z ∈ D.

This is equivalent to

d(ϕw (u), ϕw (z)) ≤ d(ϕw (u), ϕw (w)) + d(ϕw (w), ϕw (z))

69
or
d(ϕw (u), ϕw (z)) ≤ d(ϕw (u), 0) + d(0, ϕw (z)).
It is actually enough to show that

d(z, w) ≤ d(z, 0) + d(0, w), ∀z, w ∈ D.

Now d(z, 0) = |z|, d(0, w) = |w|, so we need to show that

d(z, w) ≤ d(|z|, −|w|).

From here, the triangle inequality will follow.

On D we had defined dp (w, z) = |ϕw (z)|. We had reduced the triangle inequality for
dp to the inequality dp (w, z) ≤ |w| + |z|. Previously,

(1 − |α|2 )(1 − |z|2 )

1 − |ϕα (z)|2 = .
|1 − αz|2
Thus
(1 − |w|2 )(1 − |z|2 )
1 − |ϕw (z)|2 =
|1 − wz|2
(1 − |w|2 )(1 − |z|2 )
≥
(1 + |w||z|)2
(1 − |−|w||2 )(1 − ||z||2 )
=
(1 − (−|w|)(|z|))2
2
= 1 − ϕ−|w| (|z|)

and so |ϕw (z)| ≤ |ϕ−|w| (|z|)|. Thus dp (w, z) ≤ dp (−|w|, |z|) for all w, z ∈ D. This says

−|w| − |z| |w| + |z|

dp (w, z) ≤ = ≤ |w| + |z|.
1 − (−|w|) · |z| 1 + |w||z|

This verifies that (D, dp ) is a metric space.

 
1 1 + |ϕz (w)|
dH (z, w) = ln
2 1 − |ϕz (w)|
is the hyperbolic metric.
Next, I will show that (D, dp ) is a complete metric space.
|z−w|
First, let dE (z, w) = |z − w|. Note that dp (z, w) = |1−zw| > 12 dE (z, w). Second, say
dE
that (zn ) ⊂ D is a dp -Cauchy sequence. Then (zn ) ⊂ D is also dE -Cauchy. Thus zn → p for
some point p ∈ D. Third, I want to show that p 6∈ ∂D. Suppose p ∈ ∂D, then p = eiθ for
some θ. If I replace (zn ) by (e−iθ zn ) then this is still dp -Cauchy and now e−iθ zn → 1. I will

70
simply assume that p = 1. Since (zn ) is dp -Cauchy, I can find N such that if n > m ≥ N
then dp (zn , zm ) < 21 . That is,
|zn − zm | 1
<
|1 − z n zm | 2
dE
for all n > m ≥ N . Let n → ∞. We know that zn → 1 and so
|zn − zm | |1 − zm |
lim = = 1.
n→∞ |1 − z n zm | |1 − zm |

But
|zn − zm | 1
lim <
n→∞ |1 − z n zm | 2
and this is a contradiction. Thus p 6∈ ∂D. That is, p ∈ D. Finally, we want to show that
dp
zn → p.
|zn − p|
lim dp (zn , p) = lim =0
n→∞ n→∞ |1 − z n p|

because limn→∞ |zn − p| = 0 and limn→∞ |1 − z n p| = |1 − pp| = 1 − |p|2 > 0. This verifies
dp
that zn → p. So (D, dp ) is complete.

71
33 Normal Families and the Riemann Mapping Theorem I
(11/12)
Say (X, dX ) is a compact metric space. Say (Y, dY ) is a complete metric space (usually R
or C in the consideration of this course). Then you can make the space

C(X, Y ) = {f : X → Y |f is continuous}

into a metric space. The metric is

ρ(f, g) = max dY (f (x), g(x)).

x∈X

Convergence in this metric is the same as uniform convergence on X. That is, (fn ) ⊂
C(X, Y ) converges to f ∈ C(X, Y ) in the ρ-metric if and only if fn → f uniformly on X.

Remark 33.1. (C(X, Y ), ρ) is complete.

Can we characterize compact subsets of (C(X, Y ), ρ)?

Definition 33.2. A subset of a metric space is precompact if its closure is compact.

Lemma 33.3. A set is precompact if and only if every sequence in the set has a convergent
subsequence. (The convergent subsequence does not have to converge to a point in this set.)

Can we characterize precompact subsets of (C(X, Y ), ρ)? Yes – that is what the
Arzela-Ascoli Theorem does.

Remark 33.4. In compact metric space, point-wise continuity is the same as uniform
continuality. But we may consider point-wise continuity so that it can be applied to non-
compact metric space.

34 Normal Families and the Riemann Mapping Theorem II

(11/14)
Definition 34.1. A set F ⊂ C(X, Y ) is pointwise precompact if {f (p)|f ∈ F} is a
precompact subset of Y for all p ∈ X.

Definition 34.2. A set F ⊂ C(X, Y ) is equicontinuous if for all p ∈ X and all  > 0 there
is a δ > 0 such that if x ∈ X and dX (x, p) < δ then dY (f (x), f (p)) <  for all f ∈ F.

Theorem 34.3 (Arzela-Ascoli Theorem). Let X be a compact metric space and Y a

complete metric space. A set F ⊂ C(X, Y ) is precompact if and only if it is pointwise
precompact and equicontinuous.

72
 Here is the sketch to show that F is totally bounded.
Assume that F is pointwise precompact and equicontinuous. I want to show that F is
totally bounded. Let  > 0. For each p ∈ X choose δp > 0 such that δp works for 3 in the
definition of equicontinuous. {BX (p, δp )|p ∈ X} is an open cover and so we choose Qa finite
subcover {BX (p, δp )|p ∈ A} where A ⊂ X is finite. Consider the product y = Qp∈A yp ,
where yp = cl{f (p)|f ∈ F} ⊂ Y . By assumption,
Q yp is compact for all p ∈ A. So p∈A yp
with the
Q ∞ d -metric is also compact. So y∈A p is totally bounded. Choose
y a finite set
 Q
G1 ⊂ p∈A yp such that By (γ, 6 ), γ ∈ G1 cover y. There is a map F → p∈A yp given by
sending a function to the tuple of its values at each p ∈ A. Let G ⊂ G1 be the set of γ
such that By (γ, 6 ) intersects the image of this map. Choose fγ that maps into By (γ, 6 ) for
each γ ∈ G. We claim that {BC(X,Y ) (Fγ , )|γ ∈ G} covers F. If f ∈ F, then f maps into
Q  
p∈A yp and its image lands inside some By (γ, 6 ) for γ ∈ G. Thus dY (f (p), fγ (p)) < 3
for all p ∈ A. Also if x ∈ X then x ∈ BX (p, δp ) for some p ∈ A and so dY (f (x), f (p)) < 3
and dY (fγ (x), fγ (p)) < 3 . Thus dY (f (x), fγ (x)) <  for all x ∈ X.

35 Normal Families and the Riemann Mapping Theorem

III (11/17)
We have proved that if X is compact, Y is complete, then F ⊂ C(X, Y ) is precompact if
and only if it is pointwise precompact and equicontinuous.
What if X is not necessarily compact?
One reasonable type of convergence in C(X, Y ) where X need not be compact is
uniform convergence on compacta (u.c.c.).

Definition 35.1. A sequence (fn ) ⊂ C(X, Y ) converges uniformly on compacta to f ∈

C(X, Y ) if (fn |K ) converges uniformly to f |K for all compact K ⊂ X.

Definition 35.2. X is said to have a compact exhaustion if there is a sequence (Kn ) of

compact subsets of X such that
(i) Kn ⊂ int(Kn+1 ) for all n ≥ 1;
(ii) X = ∪∞
n=1 Kn .

Note that if X has a compact exhaustion, then X = ∪∞ n=1 int(Kn ). Further, if K ⊂ X

is any compact subset then K ⊂ int(Kn ) for some n. This follows because {int(Kn )|n ≥ 1}
is an open cover of X.
Any open set V ⊂ Rn has a compact exhaustion. One way to see this is to define
1
Kn = {x ∈ V |dist(x, ∂V ) ≥ and kxk ≤ n}.
n
With this definition, Kn is automatically
• a subset of V ,
• closed

73
• bounded.
So Kn is compact (closed, bounded ⇒ compact). Say that x ∈ Kn . Then kxk ≤ n and
dist(x, ∂V ) ≥ n1 . We need to find  > 0 such that B(x, ) ⊂ Kn+1 . Since V is open, we
can choose  > 0 small enough that B(x, ) ⊂ V . If y ∈ B(x, ), then

kyk = kx + (y − x)k
≤ kxk + ky − xk
≤n+

and so  < 1 will ensure that kyk ≤ n + 1. Then

disc(y, ∂V ) ≥ dist(x, ∂V ) − kx − yk
1
≥ −
n
and
1
disc(y, ∂V ) ≥
n+1
1
provided that  < n(n+1) . Combining these, we find  > 0 such that B(x, ) ⊂ Kn+1 . Thus
x ∈ int(Kn+1 ). (Another way to see it is to use the inverse limit, and define partial order
Kn < Kn+1 if Kn ⊂ int(Kn+1 ).)
Let X be a metric space with a compact exhaustion (Kn ). Suppose that you have a
family of functions (gn ) where gn ∈ C(Kn , Y ) for each n. Suppose that gn+1 |Kn = gn for
all n ≥ 1. Then there is a function g ∈ C(X, Y ) such that g|Kn = gn .
We can certainly define g : X → Y by g(x) = gn (x) for any n such that x ∈ Kn . Why
is this continuous? Let  > 0. Suppose that x ∈ Kn . Then x ∈ int(Kn+1 ) and so there
is δ1 > 0 such that B(x, δ1 ) ⊂ Kn+1 . Since gn+1 is continuous there is δ2 > 0 such that
if dX (x, y) < δ2 then dY (gn+1 (x), gn+1 (y)) < . Let δ = min{δ1 , δ2 }. If dX (x, y) < δ then
y ∈ Kn+1 and
dY (g(x), g(y)) = dY (gn+1 (x), gn+1 (y)) < .
(The condition Kn ⊂ int(Kn+1 ) is very critical here.)

36 Normal Families and the Riemann Mapping Theorem

IV (11/19)
Next, we want to consider all the space C(Kn , Y ) at once. One natural way is to consider
their product:
∞
Y
C(Kn , Y ).
n=1

74
Theorem 36.1. Suppose that (Zm , dm ) is a sequence of metric space. Let
∞
Y
Z= Zm .
m=1

Define d : Z × Z → R by
∞
X dm (z(m), w(m))
d(z, w) = 2−m ·
1 + dm (z(m), w(m))
m=1

Then (Z, d) is a metric space. Moreover, a sequence (zn ) in Z converges to z ∈ Z if and

only if zn (m) → z(m) for all m. Finally, if (zm , dm ) is compact for all m ≥ 1, then (Z, d)
is compact.
Proof. We give the main ingredients for the proof.
t
To verify the triangle inequality, observe that t 7→ 1+t is increasing on [0, ∞). If
z, w, u ∈ Z, then
dm (z(m), u(m)) dm (z(m), w(m)) + dm (w(m), u(m))
≤
1 + dm (z(m), u(m)) 1 + dm (z(m), w(m)) + dm (w(m), u(m))
dm (z(m), w(m))
=
1 + dm (z(m), w(m)) + dm (w(m), u(m))
dm (w(m), u(m))
+
1 + dm (z(m), w(m)) + dm (w(m), u(m))
dm (z(m), w(m)) dm (w(m), u(m))
≤ + .
1 + dm (z(m), w(m)) 1 + dm (w(m), u(m))
A sequence (bn ) in (Zm , dm ) converges to b ∈ Zm with respect to the metric dm if
dm
and only if it converges to b ∈ Zm with respect to the metric 1+d m
. The reason is that
dm dm
1+dm ≤ dm and dm ≤ 2 · 1+dm when dm ≤ 1. This suffices to show that the convergence
is the same.
The next claim is that a sequence (zn ) in Z converges to z ∈ Z if and only if zn (m) →
z(m) for all m ≥ 1.

z1 = (z1 (1), z1 (2), z1 (3), z1 (4), · · · )

z2 = (z2 (1), z2 (2), z2 (3), z2 (4), · · · )
..
.
zn = (zn (1), zn (2), zn (3), zn (4), · · · )

Say zn (m) → z(m) for all m. Let  > 0. Choose N such that
∞
X 
2−m < .
2
m=N +1

75
Then
∞
X dm (zn (m), z(m))
d(zn , z) = 2−m
1 + dm (zn (m), z(m))
m=1
N
X dm (zn (m), z(m)) 
< 2−m + .
1 + dm (zn (m), z(m)) 2
m=1

I can choose L such that n ≥ L. Then

dm (zn (m), z(m)) 
<
1 + dm (zn (m), z(m)) 2

for all 1 ≤ m ≤ N . If n ≥ L, d(zn , z) < . Thus zn → z in Z.

The inverse is also true.

If all (Zm , dm ) are compact, then (Z, d) is compact. Take a sequence (zn ) in (Z, d)
with (Zm , dm ) are compact. Find infinite set J1 ⊂ N such that (zn (1))n∈J1 converges to
w(1). From (zn (2))n∈J2 , choose infinite J2 ⊂ J1 such that (zn (2))n∈J2 converges to w(2).
Continue the process. Now make a subsequence (zn )n∈J . J contains the smallest element
of J1 , the smallest element of J2 which is larger than that, the smallest element of J3
which is larger than both, and so on.

Theorem 36.2 (Arzela-Ascoli Theorem, version 2). Let X be a metric space that has a
compact exhaustion. Let Y be a complete metric space. There is a metric on C(X, Y ) such
that convergence in this metric is the same as uniform convergence on compact subsets.
With this metric, F ⊂ C(X, Y ) is precompact if and only if it is pointwise precompact and
equicontinuous.

Proof. Choose a compact exhaustion (Km ). Let ρm be the metric on C(Km , Y ) that
corresponds to uniform convergence. Consider
∞
Y
Z= C(Km , Y )
m=1

with the metric

∞
X ρm (f, g)
d(f, g) = 2−m .
1 + ρm (f, g)
m=1

Note that there is a one-to-one function ι : C(X, Y ) → ∞

Q
m=1 C(Km , Y ) given by

ι(f ) = (f |Km )m≥1 .

We define the metric on C(X, Y ) to be v(f, g) = d(ι(f ), ι(g)) (the pull-back metric). This
metric on C(X, Y ) has the required property.

76
 Say F ⊂ C(X, Y ) is pointwise precompact and equicontinuous. Let Fm be the closure
of the image of F under ι at the m-th coordinate. That is, Fm ⊂ C(Km , Y ) is the closure
of the set {f |Km : f ∈ F}. {f |Km : f ∈ F} is pointwise precompact and equicontinuous
and so Fm is compact by the previous Arzela-Ascoli Theorem. Now ι(F) ⊂ ∞
Q
m=1 Fm
which shows that ι(F) is precompact. Last thing is to be careful about the image of ι.

37 Normal Families and the Riemann Mapping Theorem V

(11/21)
Definition 37.1. Let V ⊂ C be open and F ⊂ H(V ). Then F is a normal family if for
every compact subset K ⊂ V there is a constant M such that |f (z)| ≤ M for all z ∈ K
and f ∈ F.

Remark 37.2. To check that a family is normal, it suffices to show that if D(w, r) ⊂ V
then there is some M such that |f (z)| ≤ M for all z ∈ D(w, r) and f ∈ F.

Theorem 37.3 (Montel’s Theorem). Let V ⊂ C be an open set and F ⊂ H(V ) be a

normal family. Suppose that (fn ) is a sequence in F. Then (fn ) has a subsequence that
converges uniformly on compact subsets of V .

Proof. We have to verify that F is precompact in the topology of uniform convergence on

compact subsets. To do this, we want to apply the Arzela-Ascoli Theorem (Version 2).
Well, H(V ) ⊂ C(V, C), V has a compact exhaustion, and C is a complete metric space.
It remains to verify that F is pointwise precompact and equicontinuous. The precompact
subsets of C are the bounded subsets. If z ∈ V then {f (z)|f ∈ F} is bounded because
{z} is compact. So F is pointwise precompact. Let z ∈ V . I wish to show that F is
equicontinuous at z. Choose r > 0 such that D(z, 2r) ⊂ V . Suppose that w ∈ D(z, r).
Then
Z Z
1 f (ζ) 1 f (ζ)
|f (z) − f (w)| = dζ − dζ
2πi ∂D(z,2r) ζ − z 2πi ∂D(z,2r) ζ − w

z−w
Z
1
= f (ζ) dζ
2πi ∂D(z,2r) (ζ − z)(ζ − w)
|z − w|
Z
1
≤ |f (ζ)| |dζ|.
2π ∂D(z,2r) |ζ − z| · |ζ − w|

Choose M such that |f (ζ)| ≤ M , ∀f ∈ F, ∀ζ ∈ D(z, 2r). So

1 |z − w|
|f (z) − f (w)| ≤ M· · 2π(2r)
2π 2r · r
M
= · |z − w|
r

77
for all f ∈ F, w ∈ D(z, r). Let  > 0, choose δ > 0 such that δ < r and M
r δ < . Of f ∈ F
and |z − w| < δ then
M
|f (z) − f (w)| ≤ · |z − w| < .
r
Thus f is equicontinuous at z. Since z was arbitrary, F is equicontinuous. By Arzela-
Ascoli Theorem, it follows that F is precompact in the topology of uniform convergence
on compact subsets.

Here are a few remarks:

1. In the topology of uniform convergence on compact subsets, H(V ) is a closed
subset of C(V, C). To verify this, we have to show that if (fn ) ⊂ H(V ) and fn → f in
this topology, then f ∈ H(V ). We already know this. This means that the limit of the
subsequence in Montel’s Theorem is always a holomorphic function.
2. The map D : H(V ) → H(V ) given by D(f ) = f 0 is continuous for the uniform
convergence on compact subsets topology. We need to show that if fn → f in H(V ) then
fn0 → f 0 in H(V ). We also already know this.
3. The conditions in Montel’s Theorem are actually “if and only if” conditions. That
is, if F is precompact, then F has to be normal.
4. If F ⊂ H(V ) is a normal family then {|f 0 (z)| : f ∈ F} is bounded for each z ∈ U . In
fact, the sup is achieved by some function g ∈ cl(F) (because the closure of F is compact
if F is a normal family and apply the extreme value theorem).
5. If fn → f in H(V ) and each fn is one-to-one then f is either one-to-one or constant.
Why? Suppose f is not constant, but there are p, q ∈ V such that f (p) = f (q) but p 6= q.
There are some rp > 0 and some rq > 0 such that D(p, rp ), D(q, rq ) ⊂ V , f does not take
the value f (p) = f (q) on ∂D(p, rp ), ∂D(q, rq ) and D(p, rp ) ∩ D(q, rq ) = ∅. Now fn → f
uniformly on D(p, rp ) ∪ D(q, rq ). By Hurwitz’s Theorem, when n is large enough, fn takes
the value f (p) in D(p, rp ) and in D(q, rq ). But then we get z ∈ D(p, rp ) and w ∈ D(q, rq )
with fn (z) = f (p) = fn (w). This contradicts fn being one-to-one.

38 Normal Families and the Riemann Mapping Theorem

VI (11/24)
Theorem 38.1 (Riemann Mapping Theorem). Let V ⊂ C be a non-empty, open, proper,
simply connected, connected set. Take z0 ∈ V . There is one and only one conformal
equivalence f : V → D such that f (z0 ) = 0 and f 0 (z0 ) > 0.
Step 1: Uniqueness.
Suppose that f, g : V → D such that f (z0 ) = g(z0 ) = 0 and f 0 (z0 ) > 0, g 0 (z0 ) > 0. The
function ϕ = f ◦ g −1 : D → D is holomorphic, one-to-one, and onto. That is, ϕ ∈ Aut(D).
Also, ϕ(0) = (f ◦ g −1 )(0) = f (z0 ) = 0. From our knowledge of Aut(D), we know that
0 (z )
ϕ(w) = eiθ w for some θ. Now ϕ0 (0) = fg0 (z 0
0)
by chain rule. Thus ϕ0 (0) > 0. Well,
ϕ0 (0) = eiθ > 0 and so eiθ = 1 and ϕ = idD . This says that f = g.

78
Example 38.2. Let S = {x + iy| − 1 < x < 1} and f : S → D be the Riemann map (the
conformal equivalence) such that f (0) = 0 and f 0 (0) > 0. Show that f (−z) = −f (z) for
all z ∈ S.
One way is to consider g : S → D defined by g(z) = −f (−z). Note that g is one-to-one
and onto, holomorphic, g(0) = −f (−0) = −f (0) = 0, g 0 (0) = f 0 (0) > 0. Thus g = f by
the uniqueness in the Riemann Mapping Theorem and so f (−z) = −f (z).

Step 2.
If we could find f : V → D such that f (z0 ) = 0 then we could arrange to have
f 0 (z0 ) > 0. We know that f 0 (z0 ) 6= 0 because f is one-to-one. We may choose θ such that
eiθ f 0 (z0 ) > 0. Then g : V → D defined by g(z) = eiθ f (z) would satisfy g 0 (z0 ) > 0 and
g(z0 ) = 0. This means we only have to worry about f (z0 ) = 0.
We know that if V is simply connected, then a non-vanishing holomorphic function on
V has a holomorphic square-root.

Theorem 38.3 (Riemann Mapping Theorem, restated). Let V ⊂ C be a non-empty,

open, connected, proper, and such that every non-vanishing holomorphic function on V
has a holomorphic square root. Let z0 ∈ V . Then there is a unique conformal equivalence
f : V → D such that f (z0 ) = 0 and f 0 (z0 ) > 0.

Note that 38.3 implies 38.1.

Note that once we know this, we know that the square-root condition implies that V
is simply connected.

Step 3.
We need to show there are one-to-one holomorphic functions h : V → D such that
h(z0 ) = 0. Choose p ∈ C\V . Then z 7→ z − p is a non-vanishing holomorphic function on
V . Thus there is a holomorphic function g such that g(z)2 = z − p for all z ∈ V . Choose
q ∈ g(V ). We know that g is not constant. Thus g(V ) is open by Open Mapping Theorem.
Choose r > 0 such that D(q, r) ⊂ g(V ). We claim that g(V ) ∩ D(−q, r) = ∅. Suppose not
and choose w ∈ g(V )∩D(−q, r). Then −w ∈ D(q, r) ⊂ g(V ). Thus we may find z1 , z2 ∈ V
such that g(z1 ) = w, g(z2 ) = w. This implies z1 − p = g(z1 )2 = w2 = (−w)2 = z2 − p
and z1 = z2 . But then w = −w and so w = 0. So w = 0 6∈ g(V ), a contradiction. Thus
g(V ) ∩ D(−q, r) = ∅ and so |g(z) + q| ≥ r for all z ∈ V .
1
Also note that g is one-to-one. Thus g̃(z) = g(z)+q is one-to-one and bounded. Now
we may choose a ∈ C\{0}, b ∈ C such that

f (z) = ag̃(z) + b

satisfies f (z0 ) = 0 and f (V ) ⊂ D. (Just choose a small enough that the diameter of ag̃(V )
is less than 1 and b so that f (z0 ) = 0.)

79
39 Normal Families and the Riemann Mapping Theorem
VII (12/1)
Step 4.
We know that F = {f : V → D|f ∈ H(V ), f (z0 ) = 0, f one to one} is non-empty.
Next observation is that F is a normal family. By Montel’s Theorem, F is precompact
in the topology of uniform convergence on compact subsets of V . Thus cl(F) is a compact
set. The map g 7→ |g 0 (z0 )| is a continuous map from H(V ) to [0, ∞). Thus (Extreme
Value Theorem implies that) this map achieves a maximum value on cl(F). That is, there
is some f ∈ cl(F) such that |f 0 (z0 )| is a maximum over all functions in cl(F).

Step 5.
We know that there is some g ∈ F. Since g is one-to-one, g 0 (z0 ) 6= 0. Thus |g 0 (z0 )| > 0
and so |f 0 (z0 )| ≥ |g 0 (z0 )| > 0. This tells us that f is not constant. Since f is the limit of
sequence of functions in F, each of which is one-to-one, and f is not constant, it follows
that f is also one-to-one. Since g(z0 ) = 0 for all g ∈ F, f (z0 ) = 0. It is immediate (for
the same reason) that |f (z)| ≤ 1 for all z ∈ V . By the Maximum Modulus Principle, the
fact V is connected, and the fact that f is not constant, it follows that |f (z)| < 1 for all
z ∈ V . We have now shown that f ∈ F. It remains to show that F is onto D.

Step 6.
Say g : V → D with g(z0 ) = 0. Define Hg : V → [0, ∞) by
|g 0 (z)|
Hg (z) = .
1 − |g 0 (z)|
Note that Hg (z0 ) = |g 0 (z0 )|.
Extend Hg to all g : V → D for the next lemma.
Lemma 39.1. Let ψ : D → D be holomorphic. Then Hψ◦g (z) ≤ Hg (z), ∀z ∈ V with
equality if and only if ψ ∈ Aut(D) or g 0 (z) = 0.
Proof. By the Schwarz-Pick Lemma, we know that
|ψ 0 (w)| 1
2
≤
1 − |ψ(w)| 1 − |w|2
for all w ∈ D and we have equality if and only if ψ ∈ Aut(D). Substituting g(z) for w in
this inequality, we get
|ψ 0 (g(z))| 1
2
≤
1 − |ψ(g(z))| 1 − |g(z)|2
with equality if and only if ψ ∈ Aut(D). Multiplying both sides by |g 0 (z)| we get
|ψ 0 (g(z))g 0 (z)| |g 0 (z)|
≤
1 − |ψ(g(z))|2 1 − |g(z)|2

80
with equality if and only if ψ ∈ Aut(D) or |g 0 (z)| = 0. This says Hψ◦g (z) ≤ Hg (z) with
equality if and only if ψ ∈ Aut(D) or |g 0 (z)| = 0.

Now let’s go back to the main proof.

Step 7.
Suppose that our maximer is not onto D. Choose α ∈ D not in the image of f . Consider
ϕα ◦ f . This omits the value 0 and so there exists h ∈ H(V ) such that h2 = ϕα ◦ f . If
h(z0 ) = β, then define F = ϕβ ◦ h. Note that F is holomorphic, F (z0 ) = ϕβ (h(z0 )) =
ϕβ (β) = 0, F is one-to-one because f is, so ϕα ◦ f is, so h is, so F = ϕβ ◦ h is. This implies
that F 0 (z0 ) 6= 0. Since F (z0 ) = 0,

|F 0 (z0 )| = HF (z0 ) = Hϕ−1 ◦F (z0 ) (because ϕβ ∈ Aut(D))

β

= Hh (z0 ) > Hϕ◦h (z0 ) where ϕ : D → D is ϕ(w) = w2

(because ϕ ∈ Aut(D), h0 (z0 ) 6= 0 because h is one-to-one)
= Hϕα ◦f (z0 )
= Hf (z0 )
= |f 0 (z0 )| (f (z0 ) = 0).

This contradicts the assumption that f is the maximer. Thus f is onto.

This proves the Riemann Mapping Theorem, restated version and also the Riemann
Mapping Theorem.

Corollary 39.2. Let V ⊂ C be a connected, simple connected, open set. Then Aut(V )
acts transitively on V .

Proof. If V = ∅ it is trivial. If V = C, then the claim follows from our computation

of Aut(C). Otherwise, V is non-empty and proper and so there is a biholomorphic map
f : V → D. In addition, f is one-to-one and onto. We know that Aut(D) acts transitively
on D. The claim follows.

81
40 Topology of Uniform Convergence on Compact Subsets
(12/3)
The set of all holomorphic maps from D to D is precompact in the topology of uni-
form convergence on compact subsets. This implies, for example, that if (αn ) ⊂ D and
αn → 1 then (ϕαn ) must have a convergent subsequence. In fact, ϕαn converges to the
constant sequence 1 in the topology of uniform convergence on compact subsets. Note
that ϕαn (αn ) = 0 so (ϕαn (αn )) is not converging to 1! To show ϕαn → 1 in the topology
of uniform convergence on compact sets, let 0 < r < 1 and consider z ∈ D(0, r). Then

αn − z
|ϕαn (z) − 1| = −1
1 − αn z
(αn − z) − (1 − αn z)
=
1 − αn z
(αn − 1) + (αn − 1)z
=
1 − αn z
|αn − 1| + |αn − 1|r
≤
1−r
1+r
= |αn − 1| · .
1−r
Now I get ϕαn |D(0,r) → 1 uniformly. Thus ϕαn → 1 in the topology of uniform convergence
on compact sets because any compact K ⊂ D is contained in D(0, r) for some 0 < r < 1.
The metric space H(D) is not compact, but it is separable (that is, it has a countable
dense subset). We will verify that

P = {g|g polynomials on D with coefficients having rational real and imaginary parts}

is dense in H(D).
Choose a compact exhaustion (Kn ) of D. Specifically, we could choosePKn = D(0, rn )
where rn % 1 (increasingly approach to 1). Say f ∈ H(D). Let f (z) = ∞ m=0 am z
m be

the Maclaurin series for f . Since f ∈ H(D), we know that this power series has radius of
convergence at least 1. Also the power series converges uniformly to f on any compact
subset of D. In particular, the power series converges uniformly to f on Kn . In particular,
1
given n, I can choose some partial sum of the power series gn such that |gn (z) − f (z)| < 2n
PN
for all z ∈ Kn . Note gn ∈ C[z]. Next, say gn (z) = m=0 am z m . Then if z ∈ Kn , we have
N
X N
X N
X
m m
am z − bm z ≤ |am − bm |
m=0 m=0 m=0
≤ (N + 1) · max |am − bm |.
0≤m≤N

82
 PN m
I may choose bm ∈ Q(i) such that if hn = m=0 bm z then

1
|gn (z) − hn (z)| <
2n
for all z ∈ Kn . Then hn ∈ P and we have
1
|hn (z) − f (z)| <
n
for all z ∈ Kn . We hope that hn → f in the topology of uniform convergence on compact
sets.
Let K ⊂ D be compact and  > 0. Choose L such that K ⊂ KL and L1 < . If n ≥ L
then K ⊂ Kn also, and so
1 1
|hn (z) − f (z)| < ≤ < 
n L
for all z ∈ K. This verifies that hn → f uniformly on K. Hence hn → f in the topology
of uniform convergence on compact sets.
H(D) is also connected. Why – it is a vector space with the vector space operation
matching with the topology and so it is actually path connected. Given f, g ∈ H(D) define
Φ : [0, 1] → H(D) by Φ(t) = (1 − t)f + tg. This is path from f to g in H(D). The only
thing we need to check is that Φ is continuous.
Given N , let

ZN = {f ∈ H(D)|f is not constant and f has at least N zeroes in D

counted with multiplicity}.

Show that ZN ⊂ H(D) is open. It will suffice to show that if gn → f then gn is eventually
in ZN . First, choose a point w ∈ D where f 0 (w) 6= 0. We know that gn0 (w) → f 0 (w)
and so gn 0(w) 6= 0 for large enough 0. Thus gn is not constant once n is large enough.
Next, let w1 , · · · , wk be N zeroes of f counted with multiplicity. Then w1 , · · · , wk are
contained in D(0, r) for all large enough r < 1. Because the zeroes of f are isolated, we
may choose r large enough so that w1 , · · · , wk ∈ D(0, r) and f |∂D(0,r) does not vanish. Let
 = 12 minz∈∂D(0,r) |f (z)| > 0 and choose L such that if n ≤ L then |gn (z) − f (z)| <  for
all n ≥ L and all z ∈ D(0, r). By Rouché’s Theorem, gn has at least N zeroes in D(0, r)
and so in D. This is what we needed.

83
41 Determination of the Automorphism Groups Aut(A(0, r, R))
(12/5)
First, let 0 ≤ r < R. Consider a homeomorphism f : A(0, r, R) → A(0, r, R) (a continuous
function that has a continuous inverse). Then f induces a permutation of the set of
boundary components. To see this, let η > 0 be a small number and consider K =
A(0, r + η, R − η). Then f −1 (K) is a compact subset of A(0, r, R). Thus there is a λ > 0
such that
A(0, r, r + λ) ⊂ A(0, r, R)\f −1 (K).
Then
f (A(0, r, r + λ)) ⊂ A(0, r, R)\K = A(0, r, r + η) ∪ A(0, R − η, R)
and f (A(0, r, r + λ)) is connected. Thus f (A(0, r, r + λ)) belongs to one of the two sets
A(0, r, r + η) or A(0, R − η, R). For definiteness, say f (A(0, r, r + λ)) ⊂ A(0, r, r + η). Also,
there is a τ > 0 such that

A(0, r, r + τ ) ⊂ f (A(0, r, r + λ)) ⊂ A(0, r, r + η).

(To see this, repeat the argument with f −1 .) If this happens, then the permutation induced
by f is the identity permutation. The other alternative is that there is a τ > 0 such that
A(0, R −τ, R) ⊂ f (A(0, r, r +λ)) =⊂ A(0, R −η, R). If this happens, then the permutation
is the one that exchanges the two boundary components.
Example 41.1. The map f : D0 (0, 1) → D0 (0, 1) given in polar coordinates by

f (reiθ ) = (1 − r)eiθ , 0 < r < 1, 0 ≤ θ < 2π,

is a homeomorphism. This homeomorphism interchanges the boundary components. No-

tice f does not extend continuously to the closure of D0 (0, 1) in R2 !
Use this to compute Aut(A(0, r, R)). First assume that r > 0. Then f ∈ Aut(A(0, r, R))
is a homeomorphism and so it either interchanges or fixes the boundary components in
the sense discussed above. Define ψ : A(0, r, R) → A(0, r, R) by ψ(z) = rR z . Note if
r < |z| < R then
1 1 1
> >
r |z| R
and hence
rR
r< <R
|z|
. So ψ preserves the annulus. ψ ◦ ψ = id and ψ is holomorphic on A(0, r, R). This
automorphism interchanges the boundary components. By considering ψ ◦ f if necessary
we may assume that f fixes the boundary components.
Assume f fixes the boundary components. This means that lim|z|→r |f (z)| = r and
lim|z|→R |f (z)| = R. Thus |f | extends continuously to A(0, r, R) by giving it the value r on

84
the circle |z| = r and R on the circle |z| = R. Consider g : A(0, r, R) → C giving by g(z) =
f (z)
z . Note that g is holomorphic on A(0, r, R) and |g| extends continuously to A(0, r, R)
with value 1 on both boundary components. Since g does not have a zero in A(0, r, R),
we conclude from Maximum Modulus Theorem and Minimum Modulus Theorem that g
is constant. In fact, g(z) = c with |c| = 1. This tells us that f (z) = cz for some c with
|c| = 1. That is, f is a rotation about 0. So

Aut(A(0, r, R)) = {rotations} ∪ {rotations composed with ψ}.

We still have to look at r = 0. What is Aut(D0 (0, R))? Suppose f ∈ Aut(D0 (0, R)).
Note that f has a removable singularity at 0. So f extends to an element of H(D(0, R)).
Suppose that f interchanges the boundary components. Then f extends to an element of
H(D(0, R)) and |f | extends to a continuous function on D(0, R) with value 0 on the other
boundary. By Maximum Modulus Theorem, f ≡ 0 and this is impossible. Thus f does
not switch the boundary components. This tells us that the extended f satisfies f (0) = 0.
So Aut(D(0, R)) and hence f is a rotation.

85
Index
annulus, 38 Hurwitz’s Theorem, 48
argument, 4
Arzela-Ascoli Theorem, 72 Identity Principle, 25
Arzela-Ascoli Theorem, version 2, 76 index, 33
Invariant Schwarz Lemma, 68
biholomorphic, 63 Inverse Function Theorem for Holomorphic
branch of logarithm, 29 Functions, 57

Cauchy product, 8 length of a smooth curve, 12

Cauchy’s Estimates, 21 linear-fractional transformation, 62
Cauchy’s Integral Formula, 17 Liouville’s Theorem, 22
Cauchy’s Integral Formula for Derivatives,
21 m-transitive, 65
Cauchy’s Integral Formula, Homology Ver- Maximum Modulus Principle, 28
sion, 36 meromorphic function, 61
Cauchy’s Theorem for Convex Sets, 17 ML Inequality, 12
Cauchy’s Theorem for Derivatives, 11 Montel’s Theorem, 77
Cauchy’s Theorem, Homology Version, 38 Morera’s Theorem, 17
Cauchy-Goursat Theorem, 14
normal family, 77
Cauchy-Riemann equations, 6
Cayley transform, 66 open map, 46
chain, 13 Open Mapping Theorem, 46
compact exhaustion, 73 order, 23
complex differentiable, 5
conformal equivalence, 63 Parseval’s Formula, 26
conformal map, 63 path integral, 11
contour integral, 11 Path Lifting for the Circle, 32
curve, 11 Pick’s Theorem, 68
cycle, 14 piecewise smooth, 11
pointwise precompact, 72
entire function, 21 pole, 24
equicontinuous, 72 power series, 7
essential singularity, 24 precompact, 72
pseudohyperbolic metric, 68
free abelian group on the set of smooth curves
in a set, 13 residue, 41
Fundamental Theorem of Algebra, 22 Residue Theorem, 42
Riemann Mapping Theorem, 78
holomorphic (Riemann sphere), 60
Riemann Mapping Theorem, restated, 79
homotopy, 40
Riemann sphere, 60

86
Riemann’s Removable Singularity Theorem,
24
Rouché’s Theorem, 46

Schwarz Lemma, 66
Schwarz-Pick Lemma, 68
separable metric space, 82
sharply m-transitive, 65
simple zero, 23
stereographic projection, 60

The Argument Principle, 45

triangle, 14

uniform convergence on compacta, 73

87