NOTES FOR MATH 520: COMPLEX ANALYSIS

KO HONDA
1. Complex numbers
1.1. Denition of C. As a set, C = R
2
= (x, y)[ x, y R. In other words, elements of
C are pairs of real numbers.
C as a eld: C can be made into a eld, by introducing addition and multiplication as
follows:
(1) (Addition) (a, b) + (c, d) = (a +c, b +d).
(2) (Multiplication) (a, b) (c, d) = (ac bd, ad +bc).
C is an Abelian (commutative) group under +:
(1) (Associativity) ((a, b) + (c, d)) + (e, f) = (a, b) + ((c, d) + (e, f)).
(2) (Identity) (0, 0) satises (0, 0) + (a, b) = (a, b) + (0, 0) = (a, b).
(3) (Inverse) Given (a, b), (a, b) satises (a, b) + (a, b) = (a, b) + (a, b).
(4) (Commutativity) (a, b) + (c, d) = (c, d) + (a, b).
C(0, 0) is also an Abelian group under multiplication. It is easy to verify the properties
above. Note that (1, 0) is the identity and (a, b)
1
= (
a
a
2
+b
2
,
b
a
2
+b
2
).
(Distributivity) If z
1
, z
2
, z
3
C, then z
1
(z
2
+z
3
) = (z
1
z
2
) + (z
1
z
3
).
Also, we require that (1, 0) ,= (0, 0), i.e., the additive identity is not the same as the multi-
plicative identity.
1.2. Basic properties of C. From now on, we will denote an element of C by z = x + iy
(the standard notation) instead of (x, y). Hence (a +ib) + (c +id) = (a +c) +i(b +d) and
(a +ib)(c +id) = (ac bd) +i(ad +bc).
C has a subeld (x, 0)[x R which is isomorphic to R. Although the polynomial x
2
+ 1
has no zeros over R, it does over C: i
2
= 1.
Alternate descriptions of C:
1. R[x]/(x
2
+ 1), the quotient of the ring of polynomials with coecients in R by the ideal
generated by x
2
+ 1.
1
2 KO HONDA
2. The set of matrices of the form
_
a b
b a
_
, a, b R, where the operations are standard
matrix addition and multiplication.
HW 1. Prove that the alternate descriptions of C are actually isomorphic to C.
Fundamental Theorem of Algebra: C is algebraically closed, i.e., any polynomial a
n
x
n
+
a
n1
x
n1
. . . a
0
with coecients in C has a root in C.
This will be proved later, but at any rate the fact that C is algebraically closed is one of the
most attractive features of working over C.
Example: Find a square root of a + ib, i.e., z = x + iy such that z
2
= a + ib. Expanding,
we get x
2
y
2
= a and 2xy = b. Now, (x
2
+y
2
)
2
= (x
2
y
2
)
2
+ (2xy)
2
= a
2
+b
2
, so taking
square roots (over R), x
2
+y
2
=

a
2
+b
2
. (Here we take the positive square root.) Then
x
2
=
a +

a
2
+b
2
2
, y
2
=
a +

a
2
+b
2
2
.
Now just take square roots.
1.3. C as a vector space over R. We will now view C as a vector space over R. An
R-vector space is equipped with addition and scalar multiplication so that it is an Abelian
group under addition and satises:
(1) 1z = z,
(2) a(bz) = (ab)z,
(3) (a +b)z = az +bz,
(4) a(z +w) = az +aw.
Here a, b R and z, w C. The addition for C is as before, and the scalar multiplication
is inherited from multiplication, namely a(x +iy) = (ax) +i(ay).
C is geometrically represented by identifying it with R
2
. (This is sometimes called the
Argand diagram.)
1.4. Complex conjugation and absolute values. Dene complex conjugation as an R-
linear map C C which sends z = x +iy to z = x iy.
Properties of complex conjugation:
(1) z = z.
(2) z +w = z +w.
(3) z w = z w.
Given z = x + iy C, x is called the real part of C and y the imaginary part. We often
denote them by Re z and Im z.
NOTES FOR MATH 520: COMPLEX ANALYSIS 3
Re z =
z +z
2
, Im z =
z z
2i
.
Dene [z[ =
_
x
2
+y
2
. Observe that, under the identication z = x + iy (x, y), [z[ is
simply the (Euclidean) norm of (x, y).
Properties of absolute values:
(1) [z[
2
= zz.
(2) [zw[ = [z[[w[.
(3) (Triangle Inequality) [z +w[ [z[ +[w[.
The rst two are staightforward. The last follows from computing
[z +w[
2
= (z +w)(z +w) = [z[
2
+[w[
2
+ 2Re zw [z[
2
+[w[
2
+ 2[zw[ = ([z[ +[w[)
2
.
4 KO HONDA
2. Day 2
2.1. Some point-set topology.
Denition 2.1. A topological space (X, T ) consists of a set X, together with a collection
T = U

of subsets of X, satisfying the following:

(1) , X T ,
(2) if U

, U

T , then U

T ,
(3) if U

T for all I, then

I
U

T . (Here I is an indexing set, and is not

necessarily nite.)
T is called a topology for X, and U

T is called an open set of X.

Example: R
n
= R R R (n times). If x = (x
1
, . . . , x
n
), we write [x[ =
_
x
2
1
+ +x
2
n
. U R
n
is open i x U > 0 and B(x, ) = y R
n
[[yx[ < U.
In particular, the topology on C is the topology on R
2
.
The complement of an open set is said to be closed.
Denition 2.2. A map : X Y between topological spaces is continuous if U Y open

1
(U) = x X[f(x) U open.
Restricting to the case of R
n
, we say that f(x) has limit A as x tends to a and write
lim
xa
f(x) = A if for all > 0 there exists > 0 so that 0 < [x a[ < [f(x) A[ < .
Let C be an open set and f : C be a (complex-valued) function. Then f is
continuous at a if lim
xa
= f(a).
HW 2. Prove that f is a continuous function i f is continuous at all a .
HW 3. Prove that if f, g : C are continuous, then so are f + g, fg and
f
g
(where the
last one is dened over x[g(x) = 0).
2.2. Analytic functions.
Denition 2.3. A function f : C (here is open) is dierentiable at a if the
derivative
f

(a)
def
= lim
xa
f(x) f(a)
x a
exists. If f is dierentiable at all a , then f is said to be analytic or holomorphic on .
Suppose f, g : C are analytic. Then so are f + g, fg,
f
g
(where the last one is dened
over x[g(x) = 0.
Example: f(z) = 1 and f(z) = z are analytic functions from C to C, with derivatives
f

(z) = 0 and f

(z) = 1. Therefore, all polynomials f(z) = a

n
z
n
+ +a
1
z +a
0
are analytic,
with f

(z) = na
n
z
n1
+ +a
1
.
NOTES FOR MATH 520: COMPLEX ANALYSIS 5
Fact: An analytic function is continuous.
Proof. Suppose f : C is analytic with derivative f

(z) = lim
h0
f(z+h)f(z)
h
. Then
lim
h0
(f(z +h) f(z)) = f

(z) lim
h0
h = 0.
2.3. The Cauchy-Riemann equations. Write f(z) = u(z) + iv(z), where u, v : R
are real-valued functions. Suppose f is analytic. We compare two ways of taking the limit
f

(z):
First take h to be a real number approaching 0. Then
f

(z) =
f
x
=
u
x
+i
v
x
.
Next take h to be purely imaginary, i.e., let h = ik with k R. Then
f

(z) = limk 0
f(z +ik) f(z)
ik
= i
f
y
= i
u
y
+
v
y
.
We obtain:
f
x
= i
f
y
,
or, equivalently,
u
x
=
v
y
and
v
x
=
u
y
.
The equations above are called the Cauchy-Riemann equations.
Assuming for the time being that u, v have continuous partial derivatives of all orders (and
in particular the mixed partials are equal), we can show that:
u =

2
u
x
2
+

2
u
y
2
= 0, v =

2
v
x
2
+

2
v
y
2
= 0.
Such an equation u = 0 is called Laplaces equation and its solution is said to be a harmonic
function.
6 KO HONDA
3. Day 3
3.1. Geometric interpretation of the Cauchy-Riemann equations. Let f : C
C be a holomorphic function, i.e., it has a complex derivative f

(z) = lim
h0
f(z+h)f(z)
h
at all
z . If we write z = x +iy and view f(z) as a function (u(x, y), v(x, y)) : R
2
R
2
,
then u, v satisfy the Cauchy-Riemann equations:
u
x
=
v
y
,
v
x
=
u
y
.
Recall from multivariable calculus that the Jacobian J(x, y) is a linear map R
2
R
2
given
by the matrix
_
u
x
u
y
v
x
v
y
_
. Using the Cauchy-Riemann equations, we can write
J(x, y) =
_
u
x

v
x
v
x
u
x
_
.
Namely, J(x, y) is of the form
_
a b
b a
_
, which looks suspiciouly like the second alternative
description of C from Day 1.
The Jacobian J(x, y) maps the tangent vector (1, 0) based at (x, y) to the vector (
u
x
,
v
x
)
based at (u(x, y), v(x, y)). Likewise it maps (0, 1) to (
u
y
,
v
y
) = (
v
x
,
u
x
). The thing to notice
is that J(x, y) maps the pair (1, 0), (0, 1) of orthogonal vectors to the pair (
u
x
,
v
x
), (
v
x
,
u
x
)
of orthogonal vectors. Moreover, we can show the following:
HW 4. Prove that every linear transformation : R
2
R
2
given by a matrix of the form
_
a b
b a
_
(a, b not both zero) is conformal, namely it the angle between any two vectors
v, w R
2
is the same as the angle between vectors (v), (w). (Hint: First show that
v, w) = C(v), (w)), where , ) is the standard Euclidean inner product and C is a
constant which does not depend on v, w.)
Thus, f : C is often called a conformal mapping.
Example: Consider f(z) = z
2
. Then u(x, y) = x
2
y
2
, v(x, y) = 2xy. First we look
at the level curves u = u
0
and v = v
0
. x
2
y
2
= u
0
and 2xy = v
0
are both mutually
orthogonal families of hyperbolas. (Notice that since f is conformal, f
1
, where dened and
dierentiable, is also conformal.) Next, consider x = x
0
. Then u = x
2
0
y
2
, v = 2x
0
y, and
we obtain v
2
= 4x
2
0
(x
2
0
u). If y = y
0
, then v
2
= 4y
2
0
(y
2
0
+u). They give orthogonal families
of parabolas.
Theorem 3.1. f(z) = u(z) +iv(z) is analytic with continuous derivative f

(z) i u, v have
continuous rst-order partial derivatives which satisfy the Cauchy-Riemann equations.
NOTES FOR MATH 520: COMPLEX ANALYSIS 7
Proof. We already proved one direction. Suppose u, v have continuous rst-order partials
satisfying the Cauchy-Riemann equations. Then
u(x +h, y +k) u(x, y) =
u
x
h +
u
y
k +
1
,
v(x +h, y +k) v(x, y) =
v
x
h +
v
y
k +
2
,
where

1
h+ik
0 and

2
h+ik
0 as (h, k) 0. Now,
f(x +h, y +k) f(x, y) =
_
u
x
+i
v
x
_
(h +ik) +
1
+i
2
.
Therefore,
lim
h+ik0
f(x +h, y +k) f(x, y)
h +ik
=
u
x
+i
v
x
.

3.2. Harmonic functions. Recall that if f = u + iv is analytic, then u =

v
= 0, i.e.,
u, v are harmonic. If u, v satisfy the Cauchy-Riemann equations, then v is said to be the
conjugate harmonic function of u.
Remark: If v is the conjugate harmonic function of u, then u is the conjugate harmonic
function of v.
Example: Compute the conjugate of u = 2xy. Then
u
x
= 2y and
u
y
= 2x and we verify
that u = 0. Next,
v
x
= 2x and
v
y
= 2y. Then v(x, y) = x
2
+ (y) and

(y) = 2y.
Therefore, (y) = y
2
+c, and v(x, y) = x
2
+y
2
+c. u +iv = iz
2
.
8 KO HONDA
4. Day 4
4.1. More on analytic and harmonic functions. Continuing our discussion from last
time:
Proposition 4.1. If u : R
2
R is a harmonic function and u is of class C

, then there
is a harmonic function v : R
2
R satisfying
v
x
=
u
y
and
v
y
=
u
x
.
This follows immediately from the following lemma:
Lemma 4.2. Suppose
v
x
= f,
v
y
= g, and
f
y
=
g
x
, then v exists.
Proof. Dene v(x) =
_
x
0
f(t, y)dt +(y). Then clearly
v
x
= f. Now,
v
y
=
_
x
0
f
y
(t, y)dt +

(y) =
_
x
0
g
x
(t, y)dt +

(y) = g(x, y) g(0, y) +

(y).
If we set (y) =
_
y
0
g(0, t)dt, then were done.
Example: Consider f(z) = z = x iy. This is not analytic, as we can check the Cauchy-
Riemann equations:
u
x
= 1,
v
y
= 1, and they are not identical!!
One formal way of checking is to write:
f
z
def
=
1
2
_
f
x
i
f
y
_
,
f
z
def
=
1
2
_
f
x
+i
f
y
_
.
Claim. f is analytic i
f
z
= 0.
The proof is immediate from the Cauchy-Riemann equations.
Observe:
z
(z) = 1 and
z
(z) = 0, whereas
z
(z) = 0 and
z
(z) = 1.
Claim. If p(z, z) is a polynomial in two variables z, z, then p(z, z) is analytic i there are
no terms involving z.
HW 5. Prove the claim!
4.2. Geometric representation of complex numbers. View C as R
2
.
Addition: Since z
1
+z
2
corresponds to (x
1
, y
1
) + (x
2
, y
2
) = (x
1
+x
2
, y
1
+y
2
), the addition
is standard vector addition.
Multiplication: Write z in polar form r cos + ir sin = r(cos + i sin ), where r > 0
(hence [z[ = r). Then:
z
1
z
2
= r
1
(cos
1
+i sin
1
)r
2
(cos
2
+i sin
2
) = r
1
r
2
(cos(
1
+
2
) +i sin(
1
+
2
)).
(1) The norm (r) of a product is the product of the lengths of the factors.
(2) The argument () of a product is the sum of the arguments of the factors.
NOTES FOR MATH 520: COMPLEX ANALYSIS 9
We will often write z = r(cos +i sin ) = re
i
, whatever this means. This will be explained
later when we actually dene e
z
, but for the time being its not unreasonable because you
expect:
z
1
z
2
= r
1
e
i
1
r
2
e
i
2
= r
1
r
2
e
i(
1
+
2
)
,
using rules of exponentiation.
Remark: Notice that were using properties of trigonometric functions when they havent
been dened yet.
Ignoring such rigorous considerations for the time being, we will compute powers and roots
of complex numbers.
1. If z = ae
i
, then z
n
= a
n
e
in
. This is often called de Moivres formula, and can be used
for computing cos n and sin n in terms of sin and cos .
2. If z = ae
i
, then its nth roots are
a
1
n
e
i(

n
+k
2
n
)
,
for k = 0, 1, . . . , n1. Also, when a = 1, the solutions to z
n
= 1 are called nth roots of unity.
If we write = cos

n
+i sin

n
, then the other roots of unity are given by 1, ,
2
, . . . ,
n
.
Example: Consider the analytic function f(z) = z
2
. f maps rays =
0
to = 2
0
. Hence
f is a 2:1 map away from the origin. The unit circle [z[ = 1 winds twice around itself under
the map f. [Describe how the lines y = const get mapped to parabolas under f.]
10 KO HONDA
5. Polynomials and rational functions
5.1. Polynomials. Let P(z) = a
0
+a
1
z + +a
n
z
n
be a polynomial in z with coecients
in C. By the Fundamental Theorem of Algebra (which we will prove later), P(z) admits
a factorization (z
1
)P
1
(z). By repeated application of the Fundamental Theorem, we
obtain a complete factorization:
P(z) = a
n
(z
1
)(z
2
) . . . (z
n
),
where
1
, . . . ,
n
are not necessarily distinct. The factorization is unique except for the order
of the factors. (Why?)
If exactly h of the
i
s coincide, then their common value is a zero of order or multiplicity
h. We write P(z) = a
n
(z )
h
P
h
(z) with P
h
() ,= 0.
Observe: P

() = = P
(h1)
() = 0 but P
(h)
() ,= 0.
5.2. Rational functions. Consider the rational function R(z) =
P(z)
Q(z)
which is the quotient
of two polynomials. We assume that P(z) and Q(z) have no common factors. Then the
zeros of Q(z) will be called the poles of R(z). The order of the pole is the multiplicity of
z in Q(z).
We will now explain how to extend
R : Cpoles C
to
R : C C .
C is called the extended complex plane, obtained by adding the point at to C.
At this point C is not even a topological space, but later when we discuss Riemann
surfaces, well explain how R is a holomorphic map S
2
S
2
between Riemann surfaces.
First dene R(pole) = . (The reason for this is that as z pole, [R(z)[ .)
Next, R() is dened as follows: if
R(z) =
a
0
+a
1
z + +a
n
z
n
b
0
+b
1
z + +b
m
z
m
,
then
R
_
1
z
_
= z
mn
_
a
0
z
n
+a
1
z
n1
+. . .
b
0
z
m
+b
1
z
m1
+. . .
_
.
(The reason for doing is that as
1
z
, z 0.) If m > n, then R has a zero of order mn
at ; if m < n, then R has a pole of order n m at ; if m = n, then R() =
an
bm
.
Let p = max(m, n); this is called the order of the rational function.
NOTES FOR MATH 520: COMPLEX ANALYSIS 11
Observe: A rational function R(z) of order p has exactly p zeros and p poles. Indeed, if
m n, then there are m poles in C and no pole at ; also there are n zeros in C and
(mn) zeros at .
Example: The simplest rational functions are the fractional linear transformations S(z) =
z+
z+
with det
_


_
,= 0. Special cases are S(z) =
1
z
(inversion), and S(z) = z+1 (parallel
translation).
5.3. Partial fractions. We will explain how to write any rational function R(z) as
R(z) = G(z) +

j
G
j
_
1
z
j
_
,
where G, G
j
are polynomials and
j
are the poles of R.
Example:
1
z
2
1
=
1
(z1)(z+1)
=
1/2
z1
+
1/2
z+1
.
First write R(z) = G(z) + H(z), where G(z) is a polynomial without a constant term and
H(z) has degree of denominaor degree of numerator, i.e., H(z) is nite at .
If
j
is a pole of R(z), then substituting z =
j
+
1

( =
1
z
j
), we obtain:
R
_

j
+
1

_
= G
j
() +H
j
(),
where G
j
is a polynomial and H
j
() is nite at = .
Then take R(z) (G(z) +

G
j
(
1
z
j
)). There are no poles besides and
j
. At each
z =
j
, the only innite terms cancel out, and the dierence is nite. Hence the dierence
must be a constant. By placing the constant inside G(z) (for example), we have shown that
R(z) admits a partial fraction expansion as above.
12 KO HONDA
6. Riemann surfaces and holomorphic maps
6.1. The extended complex plane. Today we try to make sense of the extended complex
plane C , which is also called the Riemann sphere. As a topological space, we take
S
2
= x
2
1
+ x
2
2
+ x
2
3
= 1 R
3
. (Its topology is the induced topology from R
3
, namely the
topology is T = W S
2
[ W is open in R
3
.) Then consider the stereographic projection
from the north pole (0, 0, 1) to the x
1
x
2
-plane, which we think of as C. The straight line
passing through (0, 0, 1) and (x
1
, x
2
, x
3
) S
2
intersects the x
1
x
2
-plane at (
x
1
1x
3
,
x
2
1x
3
, 0).
(Check this!) This gives us a continuous map
: S
2
(0, 0, 1) C,
(x
1
, x
2
, x
3
) z =
x
1
+ix
2
1 x
3
.
It is not hard to see that is 1-1, onto, and inverse is also continuous.
HW 6. Prove that : S
2
(0, 0, 1) C is a homeomorphism, i.e., is invertible and
,
1
are both continuous.
The above stereographic projection misses (0, 0, 1). If we want a map which misses the
south pole (0, 0, 1), we could also do a stereographic projection from (0, 0, 1) to the
x
1
x
2
-plane.
HW 7. Compute the stereographic projection from (0, 0, 1) to the x
1
x
2
-plane.
For our purposes, we want to do something slightly dierent: First rotate S
2
by along the
x
1
-axis, and then do stereographic projection from (0, 0, 1). This has the eect of mapping
(x
1
, x
2
, x
3
) (x
1
, x
2
, x
3
)
x
1
ix
2
1 +x
3
.
6.2. Riemann surfaces.
Denition 6.1. A Riemann surface , also called a 1-dimensional complex manifold, is a
topological space (, T ) together with a collection / = U

of open sets (called an atlas of

) such that:
(1) U

= , i.e., / is an open cover of .

(2) For each U

there exist a coordinate chart

: U

C, which is a homeomorphism
onto its image.
(3) For every U

,= ,

(U

(U

) is a holomorphic map.
(These are called transition functions.)
(4) (Technical condition 1) is Hausdor, i.e., for any x ,= y there exist open sets
U
x
and U
y
containing x, y respectively and U
x
U
y
= .
(5) (Technical condition 2) is second countable, i.e., there exists a countable subcollec-
tion T
0
of T and any open set U T is a union (not necessarily nite) of open sets
in T
0
.
NOTES FOR MATH 520: COMPLEX ANALYSIS 13
Example: The extended complex plane S
2
= C is a Riemann surface. We have two
open sets U = S
2
(0, 0, 1) and V = S
2
(0, 0, 1), and U V = S
2
. We dened
homeomorphisms
: U C, (x
1
, x
2
, x
3
)
x
1
+ix
2
1 x
3
,
and
: V C, (x
1
, x
2
, x
3
)
x
1
ix
2
1 +x
3
.
UV = S
2
(0, 0, 1), (0, 0, 1). The transition function is then given by
1
: C0
C 0, z =
x
1
+ix
2
1x
3
w =
x
1
ix
2
1+x
3
. We compute that
1
z
=
1x
3
x
1
+ix
2
x
1
ix
2
x
1
ix
2
=
x
1
ix
2
1+x
3
= w, using
x
2
1
+x
2
2
+x
2
3
= 1. Therefore, the transition function is z
1
z
, which is indeed holomorphic!
HW 8. Prove that taking the stereographic projection from (0, 0, 1) and the stereographic
projection
0
from (0, 0, 1) would not have given us a holomorphic transition function!
6.3. Holomorphic maps between Riemann surfaces. Having dened Riemann sur-
faces, we now describe the appropriate class of maps between Riemann surfaces.
Denition 6.2. A map f :
1

2
between Riemann surfaces is holomorphic if for all
x
1
there exist coordinate charts U x and V f(x) s.t. composition
(U)

1
U
f
V

(V )
is holomorphic.
Example: Given a rational function R(z), we described it as a function from S
2
S
2
last
time.
HW 9. Prove that the extension of R(z) to the extended complex plane S
2
= C is a
holomorphic map S
2
S
2
.
Although the general case is left for HW, Ill explain some simpler cases:
Case 1: R(z) = z
2
. Use coordinates z
1
, w
1
=
1
z
1
for the source and z
2
, w
2
=
1
z
2
for the target.
R(z) = z
2
means with respect to coordinates z
1
and z
2
, z
1
z
2
= z
2
1
. This is a perfectly
holomorphic function! Now, with respect to w
1
and z
2
, we have: w
1

1
w
2
1
, which is not
dened for w
1
= 0. Therefore, we switch to coordinates w
1
, w
2
, and write: w
1
w
2
= w
2
1
,
which is holomorphic.
Case 2: R(z) =
za
zb
, a ,= b. z
1
z
2
=
z
1
a
z
1
b
and is holomorphic for z
1
,= b. Near z
1
= b
we use coordinates z
1
, w
2
and write z
1
w
2
=
z
1
b
z
1
a
. This is holomorphic for z
1
,= a. Now,
w
1

1/w
1
a
1/w
1
b
=
1aw
1
1bw
1
, which is holomorphic near w
1
= 0. Moreover, R() = R(w
1
= 0) = 1.
Hopefully in the framework of Riemann surfaces and maps between Riemann surfaces, the
ad hoc denitions for extending rational functions to C now make more sense!
14 KO HONDA
7. Fractional linear transformations
7.1. Group properties. Recall that a fractional linear transformation is a rational function
of the form S(z) =
az+b
cz+d
. From the discussion on Riemann surfaces, S is a holomorphic map
of the Riemann sphere S
2
to itself.
Let GL(2, C) be the group of 22 complex matrices with nonzero determinant (= invertible
2 2 complex matrices). GL(2, C) is called the general linear group over C. (Verify that
GL(2, C) is indeed a group!)
GL(2, C) acts on S
2
as follows: Given
_
a b
c d
_
GL(2, C), it maps z
az+b
cz+d
. It is not hard
to see directly that id(z) = z and (S
1
S
2
)(z) = S
1
(S
2
(z)).
However, well use homogeneous coordinates in order to see that we have a group action.
Write z =
z
1
z
2
and w =
w
1
w
2
, then w = S(z) can be written as:
w
1
= az
1
+bz
2
,
w
2
= cz
1
+cz
2
.
Equivalently,
_
w
1
w
2
_
=
_
a b
c d
__
z
1
z
2
_
.
With this notation, it is clear that the composition S
1
(S
2
(z)) corresponds to the product of
the two matrices corresponding to S
1
and S
2
.
Observe that there is some redundancy, namely S and S give rise to the same transformation
on S
2
, if C

= C 0. Hence we dene the projectivized general linear group to

be PGL(2, C) = GL(2, C)/C

, where the equivalence relation is given by S S for all

S GL(2, C) and C

. Another way of describing PGL(2, C) is to take the special linear

group SL(2, C) consisting of 2 2 complex matrices with determinant 1, and quotienting by
the subgroup id. [PGL(2, C) is also called PSL(2, C).]
CP
1
: We will now describe 1-dimensional complex projective space CP
1
. As a set, it is
C
2
(0, 0)/ , where (z
1
, z
2
) (z
1
, z
2
) for C

= C0. We have local coordinate

charts
1
: U
1
= z
1
,= 0 C which maps (z
1
, z
2
) (1,
z
2
z
1
)
z
2
z
1
and
2
: U
2
= z
2
,= 0
C which maps (z
1
, z
2
) (
z
1
z
2
, 1)
z
1
z
2
.
HW 10. Prove that CP
1
can be given the structure of a Riemann surface and that CP
1
is
biholomorphic to S
2
. Here two Riemann surfaces X, Y are biholomorphic if there is a map
: X Y such that both and
1
are holomorphic.
In view of the above HW, it is clear that the natural setting for PGL(2, C) to act on S
2
is
by viewing S
2
as CP
1
!
NOTES FOR MATH 520: COMPLEX ANALYSIS 15
Examples:
_
1 b
0 1
_
is called a parallel translation;
_
k 0
0 1
_
is called a homothety, with
special case [k[ = 1 a rotation;
_
0 1
1 0
_
is called an inversion.
7.2. The cross ratio. We consider fractional linear transformations (FLTs) S which take
z
2
, z
3
, z
4
into 1, 0, in that order. (We assume that z
2
, z
3
, z
4
are distinct and are not .)
One such FLT is:
S(z) =
z
2
z
4
z
2
z
3

z z
3
z z
4
.
Claim: There is a unique FLT (the one above) which takes z
2
, z
3
, z
4
to 1, 0, , in that order.
Proof. It suces to prove that there is a unique FLT S which takes 1, 0, to 1, 0, , in that
order; the FLT is the identity map S(z) = z. If there are two FLTs S
1
, S
2
sending z
2
, z
3
, z
4
to 1, 0, , then S
2
S
1
1
sends 1, 0, to itself, and S
1
= S
2
.
Let S(z) =
az+b
cz+d
. Then S(0) = 0 implies that
b
d
= 0, and hence b = 0. Likewise, S() =
implies that c = 0. Now S(z) =
a
d
z and S(1) = 1 implies that S(z) = z.
We can generalize the above claim:
Claim: There is a unique FLT which takes z
2
, z
3
, z
4
to z

2
, z

3
, z

4
. (Assume both triples are
distinct.)
Denition 7.1. The cross ratio of a 4-tuple of distinct complex numbers (z
1
, z
2
, z
3
, z
4
) is
S(z
1
), where S is the FLT which maps z
2
, z
3
, z
4
to 1, 0, .
In other words, (z
1
, z
2
, z
3
, z
4
) =
z
2
z
4
z
2
z
3

z
1
z
3
z
1
z
4
.
Fact: If z
1
, z
2
, z
3
, z
4
are distinct points on the Riemann sphere S
2
and T is any FLT, then
(Tz
1
, Tz
2
, Tz
3
, Tz
4
) = (z
1
, z
2
, z
3
, z
4
).
Proof. Suppose S maps z
2
, z
3
, z
4
to 1, 0, . Then ST
1
maps Tz
2
, Tz
3
, Tz
4
to 1, 0, . The
fact follows from observing that Sz
1
= (ST
1
)Tz
1
.
Now we come to the key property of FLTs. First we dene a circle to be either a circle in
C or a line in C. A line passes through , so is a circle in the Riemann sphere S
2
.
Theorem 7.2. FLTs take circles to circles.
Proof. We will prove that an FLT S maps the real axis to a circle. (Why does this
quickly imply the theorem?) The image of the real axis satises the equation Im S
1
z = 0.
Equivalently, S
1
z = S
1
z. Writing S
1
(z) =
az+b
cz+d
, we have:
az +b
cz +d
=
a z +b
c z +d
.
16 KO HONDA
Cross multiplying gives:
(ac ac)[z[
2
+ (ad bc)z + (bc ad)z + (bd bd) = 0.
If ac ac = 0, then we have Im ((ad bc)z bd) = 0, which is the equation of a line.
Otherwise, we can divide by r = ac ac and complete the square as follows:
zz +
ad bc
r
z +
bc ad
r
z =
bd bd
r
,
_
z +
bc ad
r
__
z +
ad bc
r
_
=
bd bd
r
+
bc ad
r
ad bc
r
,

z +
ad bc
r

=
(ad bc)(ad bc)
r
2
=

ad bc
r

.
Here we note that r is purely imaginary. The equation is the equation of a circle.
Corollary 7.3. The cross ratio (z
1
, z
2
, z
3
, z
4
) is real i the four points lie on a circle.
NOTES FOR MATH 520: COMPLEX ANALYSIS 17
8. Power series
Today we study the convergence of power series
a
0
+a
1
z +a
2
z
2
+ +a
n
a
n
+. . .
where a
i
are complex and z is complex.
8.1. Review of series.
Denition 8.1. A sequence a
n

n=1
has limit A if for all > 0 there is an integer N > 0
such that n N implies that [a
n
A[ < . If a
n
has a limit, then the sequence is convergent
and we write lim
n
a
n
= A.
Fact: A sequence a
n
is convergent i a
n
is Cauchy, i.e., for all > 0 there is N s.t.
m, n N implies [a
m
a
n
[ < .
lim
n
sup a
n
: Let A
n
= supa
n
, a
n+1
, . . . . A
n
is a nonincreasing sequence and its limit is
lim
n
sup a
n
. It may be a nite number or . lim
n
inf a
n
is dened similarly. Note
that if lim
n
a
n
exists, then it is the same as limsup and liminf.
An innite series a
1
+ a
2
+ + a
n
+ . . . converges if the sequence of partial sums S
n
=
a
1
+ +a
n
converges.
Absolute convergence: If [a
1
[ + [a
2
[ +. . . converges, then so does a
1
+ a
2
+. . . , and the
sequence is said to be absolutely convergent.
8.2. Uniform convergence. Consider a sequence of functions f
n
(x), all dened on the
same set E.
Denition 8.2. f
n
converges to f uniformly on E if > 0 N s.t. n N [f
n
(x)
f(x)[ < for all x E.
Proposition 8.3. The limit f of a uniformly convergent sequence of continuous functions
is continuous.
Proof. Given > 0, N s.t. n N [f
n
(x) f(x)[ <

3
for all x E. Also, since f
n
is
continuous at x
0
, s.t. [x x
0
[ < [f
n
(x) f
n
(x
0
)[ <

3
. Adding them up, we have:
[f(x) f(x
0
)[ [f(x) f
n
(x)[ +[f
n
(x) f
n
(x
0
)[ +[f
n
(x
0
) f(x
0
)[ <

3
+

3
+

3
= .

Cauchy criterion: f
n
converges uniformly on E i > 0 N s.t. m, n N implies
[f
m
(x) f
n
(x)[ < for all x E.
18 KO HONDA
8.3. Power series. The convergence of a
0
+a
1
z +a
2
z
2
+. . . is modeled on the convergence
of the geometric series
1 +z +z
2
+z
3
+. . . .
Take partial sums
S
n
(z) = 1 +z +. . . z
n1
=
1 z
n
1 z
.
Then if [z[ < 1, then z
n
0 and the series converges to
1
1z
. If [z[ > 1, then [z[
n
, so
diverges. Finally, [z[ = 1 is the hardest. If z = 1, then 1 + 1 + . . . diverges. If z ,= 1, then
we have
e
in
1
e
i
1
, and e
in
wanders around the unit sphere and does not approach any single
point.
HW 11. Carefully treat the case [z[ = 1.
Radius of convergence: In general, dene the radius of convergence R as follows:
1
R
= lim
n
sup
n
_
[a
n
[.
Example: For the geometric series 1 +z +z
2
+. . . , lim
n
sup
n

1 = 1.
Example: For the derivative

n=1
nz
n1
of the geometric series, we have lim
n
sup
n1

n =
lim
n
n

n. (Why can we do this?)

Claim: lim
n
n

n = 1.
Proof. Indeed, if
n

n = 1 + , then n = (1 + )
n
= 1 + n +
n(n1)
2

2
+ > 1 +
n(n1)
2

2
.
Hence n 1 >
n(n1)
2

2
and
_
2
n
> . As n goes to , goes to zero.
Now we describe our main theorem:
Theorem 8.4 (Abel). Consider the series a
0
+a
1
z +a
2
z
2
+. . . .
(1) The series converges absolutely for every z with [z[ < R. If 0 < < R, then the
convergence is uniform on E = [z[ .
(2) The series diverges for [z[ > R.
Proof. The proof is by comparison with the geometric series.
(1) If [z[ < R, then there is > 0 so that [z[ < < R. Hence
1

>
1
R
. By denition of
limsup,
n
_
[a
n
[ <
1

for all suciently large n. This means that [a

n
[ <
1

n
[a
n
z
n
[ <
|z
n
|

n
,
and
[a
n
z
n
[ +[a
n+1
z
n+1
[ + <
[z[
n

n
+
[z[
n+1

n+1
+. . . .
The RHS is a geometric series which converges, so the LHS is convergent; hence the original
series is absolutely convergent. For uniform convergence, take <

< R. By repeating the

NOTES FOR MATH 520: COMPLEX ANALYSIS 19
above with

instead of , we obtain
[a
n
z
n
[ +[a
n+1
z
n+1
[ + <
_

_
n
+
_

_
n+1
+. . . ,
and the RHS is nite and independent of z.
(2) If [z[ > R, then > 0 s.t. R < < [z[
1
R
>
1

. Hence there exist arbitrarily large

n s.t.
n
_
[a
n
[ >
1

[a
n
z
n
[ >
|z|
n

n
. The RH term goes to , hence the series diverges.
Remark: We are not making any statements about [z[ = R.
20 KO HONDA
9. More Series
9.1. Analyticity. Let f(z) = a
0
+a
1
z +a
2
z
2
+ =

a
n
z
n
, with a radius of convergence
R > 0 dened by
1
R
= lim
n
sup
n
_
[a
n
[.
Theorem 9.1. f(z) is analytic for [z[ < R with derivative f

(z) =

na
n
z
n1
. The deriva-
tive also has the same radius of convergence R.
Proof. We prove the theorem in two steps.
Step 1:

na
n
z
n1
has the same radius of convergence R as

a
n
z
n
.
Indeed, lim
n
sup
n

na
n
= lim
n
n

n
1
R
, and weve already shown that lim
n
n

n = 1.
(Note weve also shifted terms by one....)
Step 2: Write f(z) = S
n
(z) +R
n
(z), where S
n
(z) =

n1
i=0
a
i
z
i
is the nth partial sum. For
the time being, write g(z) =

na
n
z
n1
. Then

f(z) f(z
0
)
z z
0
g(z
0
)

S
n
(z) S
n
(z
0
)
z z
0
S

n
(z
0
)

R
n
(z) R
n
(z
0
)
z z
0

+[(S

n
(z
0
) g(z
0
))[ .
For any > 0, s.t. the rst term on the right is <

3
, since S
n
is a polynomial, hence
analytic. The second term is bounded as follows:

k=n
a
k
(z
k
z
k
0
)
z z
0

k=n
[a
k
(z
k1
+z
k2
z
0
+ +z
k1
0
)[

k=n
a
k
k
k1
,
where [z[, [z
0
[ < < R. Since the series converges, by taking n suciently large we may
bound the second term by

3
. Finally, the third term is [

k=n
a
k
kz
k1
0
[, which is bounded
by

3
in the same way taking n suciently large.
Corollary 9.2. If f(z) =

a
n
z
n
with radius of convergence R > 0, then it has derivatives
f

, f

, f

, etc., and their radius of convergence is also R.

It is also not hard to see (by repeated dierentiation) that
f(z) = f(0) +f

(0)z +
f

(0)
2!
z
2
+
f

(0)
3!
z
3
+. . . ,
namely we have the familiar Taylor series, provided we assume that f(z) admits a power
series expansion!
9.2. Abels Limit Theorem.
Theorem 9.3 (Abels Limit Theorem). Consider the power series f(z) =

a
n
z
n
. Assume
WLOG (without loss of generality) that the radius of convergence R = 1. If

a
n
converges
(i.e., f(1) exists), then f(z) approaches f(1), provided z approaches 1 while keeping
|1z|
1|z|
bounded.
NOTES FOR MATH 520: COMPLEX ANALYSIS 21
One way of interpreting
|1z|
1|z|
is as follows: Let z be on a circle of radius r, where r is very
close to but smaller than 1. Then [1 z[ is the distance from z to 1, and 1 [z[ = 1 r
is the distance from the circle of radius r to 1. When z is close to 1, the ratio
|1z|
1|z|
is very
close to
1
cos
of the angle made by z 1 and r 1. Hence there exists > 0 such that if we
write z = 1 +e
i
with 0 2, then (

2
+,
3
2
).
Proof. WLOG

a
n
= 0. Write s
n
= a
0
+a
1
+ +a
n
. Rewrite S
n
(z) = a
0
+a
1
z +a
2
z
2
+
+a
n
z
n
as:
S
n
(z) = (1 z)(a
0
+ (a
0
+a
1
)z + (a
0
+a
1
+a
2
)z
2
+ + (a
0
+ +a
n1
)z
n1
) + (a
0
+ +a
n
)z
n
= (1 z)(s
0
+s
1
z + +s
n1
z
n1
) +s
n
z
n
.
Here, we are taking [z[ < 1, and s
n
0, so s
n
z
n
0. Therefore,
f(z) = (1 z)

s
n
z
n
.
Now, we write
[f(z)[ [1 z[

m1

k=0
s
k
z
k

+[1 z[

k=m
s
k
z
k

.
Given > 0, there exists k suciently large (for example, k m) such that [s
k
[ . Hence
the second term on the RHS is dominated by the sum [1 z[

1|z|
of the geometric series.
Using our assumption, this in turn is dominated by K for some predetermined constant K.
Now the rst term on the RHS is a product of a nite number of terms, so can be made
arbitrarily close to 0 by taking z 0. This proves the theorem.
9.3. Exponential functions. Dene the exponential function
e
z
= 1 +
z
1!
+
z
2
2!
+ +
z
n
n!
+. . . .
The radius of convergence of e
z
is , i.e., e
z
converges on the whole plane, since
n
_
1
n!
0.
HW 12. Prove that
n

n! . [Hint: given any real number x, show that x

n
< n! for n
suciently large.]
The exponential function is the unique function which is a power series in z and is a solution
of the dierential equation
f

(z) = f(z),
with initial condition f(0) = 1. In fact, if we dierentiate f(z) = a
0
+a
1
z + +a
n
z
n
+. . . ,
we obtain f

(z) = a
1
+ 2a
2
z + + na
n
z
n1
+ . . . , and equating the coecients we obtain
a
0
= 1, a
1
= 1, a
2
=
1
2
, and so on. [Why is it legitimate to equate the coecients?]
22 KO HONDA
10. Exponential and trigonometric functions; Arcs, curves, etc.
10.1. Exponential functions. Last time we dened e
z
= 1 + z +
z
2
2!
+ +
z
n
n!
+ . . . .
f(z) = e
z
is the unique function such that f

(z) = f(z) and f(0) = 1.

Lemma 10.1. e
a+b
= e
a
e
b
Proof.

z
(e
z
e
cz
) = e
z
e
cz
e
z
e
cz
= 0, and hence e
z
e
cz
is a constant. By setting z = 0,
we nd that e
z
e
cz
= e
c
. Finally, if we write c = a +b and z = a, then were done.
Remark: When restricted to R, e
z
is the standard exponential function on R. Hence we
can think of e
z
: C C as the extension of e
x
: R R.
10.2. Trigonometric functions. We dene:
cos z =
e
iz
+e
iz
2
, sin z =
e
iz
e
iz
2i
.
Then we compute that
cos z = 1
z
2
2!
+
z
4
4!
. . . ,
sin z = z
z
3
3!
+
z
5
5!
. . . .
It is easy to prove the following properties:
(1) cos z +i sin z = e
iz
.
(2) cos
2
+sin
2
= 1.
(3)

z
(cos z) = sin z.
(4)

z
(sin z) = cos z.
If z = x +iy, then we can write e
z
= e
x
e
iy
= e
x
(cos y +i sin y), where e
x
, cos y, sin y are all
functions of one real variable.
f(z) has period c if f(z + c) = f(z) for all z. By writing in polar form as above, it is clear
that the smallest period of f(z) = e
z
is 2i.
We view e
z
geometrically: e
z
maps the innite strip 0 y 2 to C0. (Notice that e
z
is never zero.) When y = 0, e
z
= e
x
, and the image is x > 0 R. The line y = maps to
the ray which makes an angle with the positive x-axis. Since e
z+2i
= e
z
, the lines y =
and y = + 2 map to the same ray.
10.3. The logarithm. We want to dene the logarithm log z as the inverse function of e
z
.
The problem is that each w C 0 has innitely many preimages z such that e
z
= w;
they are all given by exp
1
(w) = log [w[ +i(arg w+2n), n Z. Here log [w[ is the real
logarithm, and 0 arg w 2. We will dene one branch as follows:
log w := log [w[ + i arg w.
NOTES FOR MATH 520: COMPLEX ANALYSIS 23
The important thing to remember is that, although we chose 0 arg w < 2, there is
nothing canonical (natural) about this choice.
Also dene a
b
= e
b log a
, when a, b C.
10.4. Arcs and curves. We now change topics and give the basics of arcs, curves, etc.
Denition 10.2. An arc or a path in C is a continuous map : [a, b] C. (Here a < b.)
(a) is called the initial point of the arc and (b) is the terminal point of the arc.
Remark: Two arcs are the same i they agree as maps. It is not sucient for them to have
the same image in C.
: [a, b] C is:
(1) of class C
1
(called dierentiable in the book), if
d
dt
=

(t) = x

(t) +iy

(t) exists and

is continuous. (Here dierentiability at a point means dierentiability on some open
set containing that point.)
(2) simple if (t
1
) = (t
2
) t
1
= t
2
.
(3) a closed curve if (a) = (b).
(4) a simple closed curve if is a closed curve and (t
1
) = (t
2
) t
1
= t
2
away from
the endpoints.
Dene : [b, a] C by (t) = (t). This traces the image of in the opposite
direction, and is called the opposite arc of .
10.5. Conformality revisited. Suppose : [a, b] C is an arc and f : C C is an
analytic function. Let w = f((t)). If

(t) exists, then w

(t) exists, and:

w

(t) = f

((t))

(t).
Suppose z
0
= (t
0
) and w
0
= f(z
0
). If

(t
0
) ,= 0 and f

(z
0
) ,= 0, then w

(t
0
) ,= 0. We also
have
arg w

(t
0
) = arg f

(z
0
) + arg

(t
0
).
Here, we are considering the argument to be mod 2. This implies that the angle between

(t
0
) and w

(t
0
) is equal to f

(z
0
). If
0
(t
0
) =
1
(t
0
) = z
0
, then the angle made by

0
(t
0
) and

1
(t
0
) is preserved under composition with f. Recall we called this property conformality.
Also observe that
[w

(t
0
)[ = [f

(z
0
)[ [

(t
0
)[,
in other words, the scaling factor is also constant.
24 KO HONDA
11. Inverse functions and their derivatives
Let C be an open set and f : C be an analytic function. Suppose f is 1-1,
f

(z) ,= 0 at all z , and f is an open mapping, i.e., sends open sets to open sets. If z
0

and w
0
= f(z
0
), then:
Claim. f
1
is analytic and (f
1
)

(w
0
) =
1
f

(z
0
)
.
Observe that f is an open mapping f
1
is dened on an open set and f
1
is continuous.
Hence, given > 0, there exists > 0 such that [ww
0
[ < implies [f
1
(w) f
1
(w
0
)[ < .
Proof.
lim
ww
0
f
1
(w) f
1
(w
0
)
w w
0
= lim
f
1
(w)f
1
(w
0
)
f
1
(w) f
1
(w
0
)
w w
0
= lim
zz
0
z z
0
w w
0
=
1
f

(z
0
)
.
Here, by the above discussion, w w
0
implies f
1
(w) f
1
(w
0
).
Fact: The condition that f be an open mapping is redundant.
The Open Mapping Theorem (used in the proof of the inverse/implicit function theorems)
states that if f

(z
0
) ,= 0 and f is in class C
1
(dierentiable with continuous derivative), then
there is an open neighborhood U z
0
on which f is an open mapping. [It will turn out that
if f is analytic, then f has derivatives of all orders.]
We will not use this fact for the time being its proof will be given when we discuss
integration.
Example: Consider w =

z. This is naturally a multiple-valued function, since there is
usually more than one point w such that w
2
= z. To make it a single-valued function (what
we usually call function), there are choices that must be made. The choices are usually
arbitrary. (The procedure of making these choices is often called choosing a single-valued
branch.)
Let the domain be C x + iy[x 0, y = 0, i.e., the complement of the negative
real axis (and the origin). We think of as obtained from C by cutting along the negative
real axis, hence the name branch cut. Write z = re
i
, where < < . Then dene
w =

re
i/2
. Geometrically, gets mapped onto the right half-plane

= x > 0.
We verify that w =

z is continuous: Write w = u+iv and w
0
= u
0
+iv
0
. If [z z
0
[ < ,
then [ww
0
[[w+w
0
[ < . Now w, w
0
are in the right half-plane, so [w+w
0
[ u +u
0
u
0
.
Hence [w w
0
[ < if = u
0
.
Now that we know w =

z is continuous, it is analytic with derivative
w
z
=
1
z/w
=
1
2w
=
1
2

z
.
NOTES FOR MATH 520: COMPLEX ANALYSIS 25
Example: Consider w = log z. Again, take the domain to be = Cx+iy[x 0, y = 0.
Then choose a branch as follows: Write z = re
i
, with < < . Then map z w =
log [r[ +i. The image is the innite strip

= i < y < i.
If w = log z is continuous, then
w
z
=
1
z/w
=
1
e
w
=
1
z
,
as expected from calculus.
It remains to prove the continuity of w = log z. Write w = u+iv and w
0
= u
0
+iv
0
. Dene
a closed set A to be a box in the range which is bounded to the left by u = u
0
log 2, to the
right by u = u
0
+ log 2, below by v = and above by v = , with the disk [w w
0
[ <
(with small) removed. The continuous function [e
w
e
w
0
[ attains a minimum on A (its
a minimum, not an inmum, since A is closed). Moreover, > 0, since the only times when
the minimum is zero are when w = w
0
+ 2in, which are not in A.
Dene = min(,
1
2
e
u
0
).
We claim that no point of [z z
0
[ < maps into A: if [z z
0
[ < maps into A, then
[e
w
e
w
0
[ , a contradiction.
We also claim that no point of [zz
0
[ < maps into u < u
0
log 2 (to the left of A) or into
u > u
0
+log 2 (to the right of A). If z is mapped to the right of A, then [e
w
e
w
0
[ e
u
e
u
0

2e
u
0
e
u
0
= e
u
0
. If mapped to the left, then [e
w
e
w
0
[ e
u
0
e
u
e
u
0

1
2
e
u
0
=
1
2
e
u
0
. In
either case, we obtain a contradiction.
This implies that [z z
0
[ < has nowhere to go but [w w
0
[ < , proving the continuity.
Example: w =

1 z
2
. Dene to be the complement in Cof two half-lines y = 0, x 1
and y = 0, x 1. Consider
+
= y 0. This is mapped to Cy = 0 and x 1
under z
2
. Multiplying by 1 rotates this region by about the origin to Cy = 0, x 1.
Adding 1 shifts to C y = 0, x 0, and squaring gives x > 0. The map

1 z
2
on

+
with the exception of points on the x-axis which map to y = 0, 0 x 1 in the image.
Now

= y 0 is also mapped to the same right half-plane. Observe that the map
from is a 2-1 map everywhere except for z = 0 which corresponds to w = 1.
Example: w = cos
1
z. If we write z = cos w =
e
iw
+e
iw
2
and solve for e
iw
in terms of z, we
get a quadratic equation
e
2iw
2ze
iw
+ 1 = 0
with solution e
iw
= z

z
2
1. Hence w = i log(z

z
2
1). Since z

z
2
1 are
reciprocals, we can write
w = i log(z +

z
2
1).
We take the following branch: w = i log(z + i

1 z
2
). Take the domain to be the same
as in the previous example. One can check that z + i

1 z
2
maps to the upper half-plane
(minus the x-axis): Since z i

1 z
2
are reciprocals, if one of them is real, then z, i

1 z
2
26 KO HONDA
are both real. But if z is on the real axis with 1 < x < 1, then i

1 z
2
is not real. Hence
we can map using log as dened above to the innite strip 0 < y < i.
NOTES FOR MATH 520: COMPLEX ANALYSIS 27
12. Line integrals
12.1. Line integrals. If f = (u, v) : [a, b] C is a continuous function, then
_
b
a
f(t)dt
def
=
_
b
a
u(t)dt +i
_
b
a
v(t)dt,
where the right-hand side terms are the standard Riemann integrals on R.
Properties:
(1)
_
b
a
c f(t)dt = c
_
b
a
f(t)dt, c C.
(2)
_
b
a
(f(t) +g(t))dt =
_
b
a
f(t)dt +
_
b
a
g(t)dt.
(3) [
_
b
a
f(t)dt[
_
b
a
[f(t)[dt.
Let : [a, b] C be a piecewise dierentiable arc. Given f : C continuous, dene:
_

f(z)dz
def
=
_
b
a
f((t))

(t)dt.
This is not so unreasonable on a formal level. Indeed, if z = (t), then dz =

(t)dt.
Key Property: If is an increasing, piecewise dierentiable function [, ] [a, b], then
_

f(z)dz =
_

f(z)dz.
Assuming everything in sight is dierentiable (not just piecewise), we have:
_
b
a
f((t))

(t)dt =
_

f((()))

(())

()d =
_

f( ())( )

()d.
The rst equality follows from the change of variables formula in integration, and the latter
follows from the chain rule. Here we write t = ().
Orientation Change: We have

f(z)dz =
_

f(z)dz.
Indeed, writing = t, we have:
_

f(z)dz =
_
a
b
f((t))(

(t))dt
=
_
b
a
f((t))(

(t))dt
=
_
b
a
f(())

()(d) =
_

f(z)dz.
28 KO HONDA
The integral
_
f(z)dz over does not depend on the actual parametrization it only depends
on the direction/orientation of the arc. If we change the direction/orientation, then
_
f(z)dz
acquires a negative sign.
Notation: We often write dz = dx +idy and dz = dx idy. Then
_

fdx =
_

f((t))
dx
dt
dt,
_

fdy =
_

f((t))
dy
dt
dt, and
_

f(z)dz =
_

(u(z) + iv(z))(dx + idy) =

_

(udx vdy) +
i
_
(vdx +udy).
More Notation: A subdivision of can be written as
1
+ +
n
, and
_

1
++n
fdz
def
=
_

1
fdz + +
_
n
fdz.
Path Integrals: Dene
_

f[dz[
def
=
_
b
a
f((t))[

(t)[dt, with : [a, b] . Then the

following are easy to see:
_

f[dz[ =
_

f[dz[,

fdz

[f[[dz[.
12.2. Exact dierentials. A dierential = pdx+qdy, where p, q : C are continuous,
is called an exact dierential if there exists U : C such that
U
x
= p and
U
y
= q. [It
turns out (but we will not make use of the fact) that if U has continuous partials, then it is
dierentiable and of class C
1
.]
Independence of path:
_

depends only on the endpoints of if we have

_

1
=
_

2
,
whenever
1
and
2
have the same initial and terminal points.
Claim.
_

depends only on the endpoints i

_

= 0 for all closed curves .

Proof. If is a closed curve, then , have the same endpoints. Hence
_

=
_

, which implies that

_

= 0. If
1
,
2
have the same endpoints, then
_

1
=
_

2
,
and
_

2
= 0.
Theorem 12.1. = pdx + pdy, with p, q continuous, is exact i
_

only depends on the

endpoints of .
Proof. Suppose pdx +qdy is exact. Then
_

pdx +qdy =
_

_
U
x
dx +
U
y
dy
_
=
_
b
a
d
dt
U(x(t), y(t))dt = U(b) U(a).
Suppose
_

only depends on the endpoints of . We assume WLOG that is connected,

i.e., if = A . B (disjoint union) with A, B open, then A or B is .
NOTES FOR MATH 520: COMPLEX ANALYSIS 29
HW 13. Prove that an open set C is connected i it is path-connected, i.e., for any
p, q C there exists a continuous path from p to q. Moreover, we may take to be a path
which is piecewise parallel either to the x-axis or the y-axis.
Pick z
0
. Let be a path from z
0
to z which is piecewise of the form x = a or
y = b. Then dene U(z) =
_

pdx + qdy. Since the integral only depends on the endpoints,

U(z) is well-dened. If we extend horizontally to a path from z
0
to z + h, h R, then
U(z +h) U(z) =
_
z+h
z
pdx +qdy, and
U
x
= p. Similarly,
U
y
= q.
12.3. Complex dierentials. Given a dierentiable function F : C, we write dF =
F
x
dx+
F
y
dy; in other words, an exact dierential is always of the form dF for some function
F.
An alternative way of writing dF is dF =
F
z
dz +
F
z
dz.
HW 14. Verify this!
The latter term vanishes if F is analytic. Hence we have the following:
Fact: If F is analytic, then F

(z)dz =
F
z
dz is exact.
Example:
_

(z a)
n
dz = 0 for n 0, if is closed. This follows from the fact that
F(z) =
(za)
n+1
n+1
is analytic with derivative (z a)
n
.
Example:
_
C
1
za
dz = 2i, where C is the circle parametrized by z = a + e
i
, [0, 2].
This proves that there is no analytic function F(z) on Ca such that F

(z) =
1
za
. (In
other words, it is not possible to dene log(z a) on all of Ca.)
30 KO HONDA
13. Cauchys theorem
Let R be the rectangle [a, b] [c, d]. Dene its boundary R to be the simple closed curve

1
+
2
+
3
+
4
, where
1
: [a, b] C maps t t + ic,
2
: [c, d] C maps t b + it,

3
: [b, a] C maps t t +id, and
4
: [d, c] C maps t ait. In other words,
R is oriented so that it goes around the boundary of R in a counterclockwise manner.
We prove the following fundamental theorem:
Theorem 13.1 (Cauchys theorem for a rectangle). If f(z) is analytic on R, then
_
R
dz = 0.
Later, a more general version of the theorem will be given it will be valid for more
general regions.
Proof. Dene (R)
def
=
_
R
f(z)dz. Subdivide R into 4 congruent rectangles R
(1)
, R
(2)
,
R
(3)
, and R
(4)
. [Draw a line segment from the midpoint of
1
to the midpoint of
3
and
a segment from the midpoint of
2
to the midpoint of
4
.] Then it is easy to verify that
R = R
(1)
+R
(2)
+R
(3)
+R
(4)
and
(R) = (R
(1)
) +. . . (R
(4)
).
Taking R
(i)
with the largest [(R
(i)
)[, we have [(R
(i)
)[
1
4
[(R)[. By continuing the subdi-
vision, there exist rectangles R R
1
R
2
. . . such that [(R
n
)[
1
4
n
[(R)[.
Now, we can take a sequence of points z
n
R
n
; by the Cauchy criterion, this sequence
converges to some point z

. (Why?) Hence we have R

n
[z z

[ < for suciently

large n. If f is analytic, then for any > 0 there exists > 0 such that [z z

[ <

f(z)f(z

)
zz

(z)

< , or [f(z) f(z

) f

(z

)(z z

)[ < [z z

[.
Recall from last time that
_
Rn
dz = 0 and
_
Rn
zdz = 0, since 1 and z have antiderivatives
z and
z
2
2
, and hence dz and zdz are exact. Now,
(R
n
) =
_
Rn
f(z)dz =
_
Rn
(f(z) f(z

) f

(z

)(z z

)) dz.
Therefore, [(R
n
)[
_
Rn
[z z

[[dz[. If we write d for the diagonal of R, then an upper

bound for [z z

[ is
1
2
n
d, the diagonal of R
n
. Next, if L is the perimeter of R, then
1
2
n
L is
the perimeter of R
n
. We obtain [(R
n
)[ (
1
2
n
d)(
1
2
n
L) = dL
1
4
n
.. Since we may take to
be arbitrarily small, we must have (R
n
) = 0 and hence (R) = 0.
A Useful Generalization: If f is analytic on R nite number of interior points, and
lim
z
(z )f(z) = 0 for any singular point , then
_
R
f(z)dz = 0.
Proof. By subdividing and using Cauchys theorem for a rectangle, we may consider a small R
with one singular point at the center. We may shrink R if necessary so that [(z)f(z)[ <
for z R [f(z)[ <

|z|
for z ,= R.
NOTES FOR MATH 520: COMPLEX ANALYSIS 31
Therefore, [(R)[
_
R

|z|
[dz[. It remains to estimate the right-hand term. If R is a
square whose sides have length r, then [z [ is bounded below by
1
2
r on R. Hence an
upper bound for the right-hand term is
4r
1/2r
= 8. Since was arbitrary, we are done.
Cauchys theorem on a rectangle can be used to prove a stronger version of the theorem,
valid for any closed curve on an open disk, not just the boundary of a rectangle.
Theorem 13.2 (Cauchys theorem for a disk). If f(z) is analytic on an open disk D =
[z a[ < , then
_

f(z)dz = 0 for every closed curve in D.

Proof. Let
0
be a path from a to z = x + iy, consisting of a horizontal line segment,
followed by a vertical line segment. Then dene F(z) =
_

0
f(z)dz. By Cauchys theorem
on a rectangle, F(z) =
_

1
f(z)dz, where
1
is a path from a to z, consisting of a vertical line
segment, followed by a horizontal one. Using one of the two types of paths, we compute:
F
x
=
d
dh
_
x+h
x
f(z)dx = f(z),
F
y
=
d
dk
_
y+k
y
f(z)idy = if(z).
Hence F satises the Cauchy-Riemann equations. Therefore, f(z)dz = F

(z)dz is exact,
and
_

f(z)dz = 0 for all closed curves (by Theorem 12.1). [Observe that the fact that F
satises the CR equations and that it has continuous partials implies that F itself is analytic.
This follows from Theorem 3.1.]
32 KO HONDA
14. The winding number and Cauchys integral formula
14.1. The winding number. Let : [, ] C be a piecewise dierentiable closed curve.
Assume does not pass through a, i.e., a , Im , where the image of is written as Im .
Denition 14.1. The winding number of about a is n(, a)
def
=
1
2i
_

dz
z a
.
Properties of the winding number:
1. n(, a) = n(, a). Proof. Follows from Orientation Change property from Day 12.
2. If C is a circle centered at a and oriented in the counterclockwise direction, then n(C, a) =
1
2i
_
C
dz
za
= 1. Proof. Direct computation.
3. n(, a) is always an integer.
Proof. (See the book for a slightly dierent proof.) Subdivide : [, ] C into
1
+ +
n
,
where each
j
: [
j1
,
j
] C is contained in a sector C arg(z a) C + for small .
Here =
0
<
1
< <
n
= . (Give proof!)
Now dene a single-valued branch of log(z a) on each sector so that arg(
j1
(
j1
)) =
arg(
j
(
j1
)). Then
_

j
dz
z a
=
_

j
d log(z a) =
_

j
d log [z a[ +i
_

j
d arg(z a).
Therefore,
_

dz
z a
=
_

d log [z a[ +i
n

j=1
d arg(z a).
Since the initial and terminal points of are the same, the rst term on the right is zero
and the second is a multiple of 2i.
4. If Im D
2
= [z b[ < , then n(, a) = 0 if a , D
2
.
Proof. If a , D
2
, then
1
za
is holomorphic on D
2
. Therefore, by Cauchys Theorem,
_

dz
za
=
0.
5. As a function of a, n(, a) is constant on each connected component of CIm .
Proof. Any two points a, b can be joined by a polygonal path in . We may reduce to
the case where a, b are connected by a straight line segment in . (The general case follows
by inducting on the number of edges of the polygon.) Observe that the function
za
zb
is real
and negative i z is on the line segment from a to b. (Check this!) Therefore, there is a
single-valued branch of log
za
zb
which can be dened on the complement of the line segment.
NOTES FOR MATH 520: COMPLEX ANALYSIS 33
Hence,
_

_
1
z a

1
z b
_
dz =
_

d log
z a
z b
= 0.
This proves that n(, a) = n(, b).
14.2. Cauchy Integral Formula.
Theorem 14.2. Suppose f(z) is analytic on the open disk D and let be a closed curve in
D. If a , Im , then
n(, a) f(a) =
1
2i
_

f(z)dz
z a
.
Proof. Consider F(z) =
f(z)f(a)
za
. Suppose rst that a D. Then F is analytic on D a
and satises lim
za
(z a)F(z) = 0. Hence Cauchys Theorem holds:
_

f(z) f(a)
z a
dz = 0.
Rewriting this equation, we get:
f(a) n(, a) = f(a)
_

dz
z a
=
_

f(z)dz
z a
.
On the other hand, if a , D, then F is analytic on all of D, and we can also apply Cauchys
Theorem. (In this case, both sides of the equation are zero.)
Application: When n(, a) = 1, e.g., is a circle oriented in the clockwise direction, we
have
f(a) =
1
2i
_

f(z)dz
z a
.
We can rewrite this as:
f(z) =
1
2i
_

f()d
z
.
(Here were thinking of z as the variable.)
Key Observation: The point of the Cauchy Integral Formula is that the value of f at z D
can be determined from the values of f on D, using some averaging process. (Assume f is
analytic in a neighborhood of D.)
34 KO HONDA
15. Higher derivatives, including Liouvilles theorem
15.1. Higher Derivatives. Let f : C be an analytic function, and let D = [z a[ <
. Let D be the boundary circle, oriented counterclockwise. Then
f(z) =
1
2i
_
D
f()d
z
for all z D, by the Cauchy Integral Formula.
The Cauchy Integral Formula allows us to dierentiate inside the integral sign, as follows:
Theorem 15.1.
f

(z) =
1
2i
_
D
f()d
( z)
2
, f
(n)
(z) =
n!
2i
_
D
f()d
( z)
n+1
.
In particular, an analytic function has (complex) derivatives of all orders.
The theorem follows from the following lemma:
Lemma 15.2. Let : [, ] C be a piecewise dierentiable arc and : Im C be a
continuous map. Then F
n
(z) =
_

()d
( z)
n
is holomorphic on C Im and its derivative
is F

n
(z) = nF
n+1
(z).
We will only prove the continuity and dierentiability of F
1
(z). The rest is similar and is
left as an exercise.
Continuity of F
1
(z): Suppose z
0
C Im . Since C Im is open, there is a

> 0
such that D

(z
0
)
def
= [z z
0
[ <

CIm . Let z D

/2
(z
0
). We write
F
1
(z) F
1
(z
0
) =
_

_
()
z

()
z
0
_
d = (z z
0
)
_

()d
( z)( z
0
)
.
We estimate that [ z[ >

2
and [ z
0
[ >

2
, and
[F
1
(z) F
1
(z
0
)[ [z z
0
[
4
[

[
2
_

[()[[d[.
If we further shrink D

/2
(z
0
), we can make [F
1
(z) F
1
(z
0
)[ arbitrarily small.
Dierentiability of F
1
(z): We write
(1)
F
1
(z) F
1
(z
0
)
z z
0
=
_

()d
( z)( z
0
)
.
Now, use the above step with
()
z
0
instead of (). [Note that
()
z
0
is continuous on Im
since z
0
avoids Im .] By the above step, Equation 1 is continuous on C . Hence as
NOTES FOR MATH 520: COMPLEX ANALYSIS 35
z z
0
we have
_

()d
( z)( z
0
)

_

()d
( z
0
)
2
.
Therefore F

1
is holomorphic and F

1
(z) = F
2
(z).
15.2. Corollaries of Theorem 15.1.
15.2.1. Moreras Theorem.
Theorem 15.3. If f : C is continuous and
_

fdz = 0 for all closed curves : [, ]

, then f is analytic on .
Proof. We already showed that f = F

for some analytic function F. f is then itself analytic

by Theorem 15.1.
15.2.2. Liouvilles Theorem.
Theorem 15.4. Any analytic function f : C C which is bounded is constant.
Proof. Let D = [z a[ < r. Suppose [f(z)[ M for all z C. If we apply the integral
formula to D, then
f
(n)
(a)
n!
2
M 2r
r
n+1
=
n!M
r
n
.
But r is arbitrary, so (taking n = 1) f

(a) = 0 for all a. Hence f is constant.

15.2.3. Fundamental Theorem of Algebra.
Theorem 15.5. Any polynomial P(z) of degree > 0 has a root.
Proof. Suppose not. Then
1
P(z)
is analytic on the whole plane. As z ,
1
P(z)
0, so
1
P(z)
is bounded. By Liouville,
1
P(z)
is constant, which is a contradiction.
36 KO HONDA
16. Removable singularities, Taylors theorem, zeros and poles
16.1. Removable singularities.
Theorem 16.1. Let a . If f : a C is analytic, then f can be extended to an
analytic function C i lim
za
(z a)f(z) = 0.
Proof. If f is analytic on , then it is continuous on . Hence lim
za
(za)f(z) = lim
za
(z
a) lim
za
f(z) = 0 f(a) = 0.
Suppose lim
za
(z a)f(z) = 0. Take a small disk D centered at a. Consider the
function g(z) =
1
2i
_
D
f()d
z
. This is an analytic function on all of the interior of D. We
claim that, away from z = a, f(z) = g(z). Indeed, in the proof of the Cauchy Integral
Formula, we used G() =
f()f(z)
z
. There are two points b = z and b = a, where we need
to check that lim
b
( b)G() = 0 (in order to use Cauchys Theorem). If b = z, then
we have lim
z
(f() f(z)) = 0 by the continuity of f at z. If b = a, then we have
lim
a
f()f(z)
az
( a) = 0 by the condition lim
za
(z a)f(z) = 0.
For example, we can use the removable singularities theorem if f(z) is bounded in a neigh-
borhood of z = a.
16.2. Taylors Theorem. Consider the function F(z) =
f(z)f(a)
za
. If f is analytic on , then
F is analytic on a. By the removable singularities theorem, F extends to an analytic
function on all of . By continuity of F at a, we must have F(a) = lim
za
F(z) = f

(a).
We can therefore write
f(z) = f(a) + (z a)f
1
(z),
where f
1
= F. Recursively we let f
1
(z) = f
1
(a) + (z a)f
2
(z), and so on, to obtain
f(z) = f(a) +f
1
(a)(z a) +f
2
(a)(z a)
2
+ +f
n1
(a)(z a)
n1
+ (z a)
n
f
n
(z).
We can dierentiate f(z) and set z = a to obtain:
Theorem 16.2 (Taylors theorem). Suppose f : C is an analytic function. Then there
is an analytic function f
n
: C such that
f(z) = f(a) +f

(a)(z a) +
f

(a)
2!
(z a)
2
+ +
f
(n1)
(a)
(n 1)!
(z a)
n1
+ (z a)
n
f
n
(z).
The remainder term f
n
(z) can be written as:
f
n
(z) =
1
2i
_
D
f()d
( a)
n
( z)
,
where D is a small disk centered at a.
NOTES FOR MATH 520: COMPLEX ANALYSIS 37
Proof. We only need to prove the expression for the remainder. We prove the statement for
n = 1, leaving the general case as an exercise.
f
1
(z) =
1
2i
_
D
f
1
()d
z
=
1
2i
_
D
(f() f(a))d
( a)( z)
=
1
2i
_
D
f()d
( a)( z)

f(a)
2i
_
D
d
( a)( z)
Now, we can write
1
(a)(z)
=
1
az
_
1
a

1
z
_
. Since
_
D
d
a
=
_
D
d
z
= 2i (a and z
are in the same connected component of CD), the second term in the equation vanishes.

Remark: We will discuss Taylor series next time.

16.3. Zeros.
Proposition 16.3. Let f : C be an analytic function, a , and be connected. If
f(a) = f

(a) = = f
(n)
(a) = 0 for all n, then f is identically zero.
Proof. Suppose f
(n)
(z) = 0 for all n. Then f(z) = f
n
(z)(z a)
n
for all n. Using the above
expression for f
n
(z), we write:
[f
n
(z)[
1
2
_
D
M[d[
R
n
(R [z a[)
=
1
2
M
R
n
2R
(R [z a[)
=
M
R
n1
(R [z a[)
.
Here D is a disk of radius R centered at a, [f()[ M for on D, and z int(D). This
implies:
[f(z)[
_
[z a[
R
_
n
MR
R [z a[
.
As n ,
_
|za|
R
_
n
0, and were done.
By Proposition 16.3, all the zeros of an analytic function have nite order, i.e., there exist
an integer h > 0 and an analytic function f
h
such that f(z) = (z a)
h
f
h
(z) and f
h
(a) ,= 0.
We say that z = a is a zero of order h.
Corollary 16.4.
(1) The zeros of an analytic function f , 0 are isolated.
(2) If f, g are analytic on and if f(z) = g(z) on a set which has an accumulation point
in , then f g. (Here, f g means f(z) = g(z) for all z .)
Proof. Follows from observing that f
h
(a) ,= 0 and f
h
is continuous f(z) ,= 0 in a neigh-
borhood of a, provided z ,= a.
38 KO HONDA
In other words, an analytic function f : C is uniquely determined by f[
X
(f restricted
to X), where X is any set with an accumulation point. For example, if f, g agree on
a nontrivial subregion of or agree on a nontrivial (= does not map to a point) arc, then
f g.
NOTES FOR MATH 520: COMPLEX ANALYSIS 39
17. Analysis of isolated singularities
Denition 17.1. An analytic function f has an isolated singularity at a if f is analytic on
0 < [z a[ < for some > 0.
We already discussed removable singularities, i.e., isolated singularities where lim
za
(z
a)f(z) = 0. In such a case f can be extended across a. (The isolated singularity was a
singularity only because of lack of information.)
17.1. Poles. Suppose a and f is analytic on a. If lim
za
f(z) = , then z = a
is a pole of f(z).
Claim. A holomorphic function f : a C with a pole at z = a can be extended to a
holomorphic function f : S
2
.
Proof. Change coordinates on S
2
from z-coordinates (about 0) to w =
1
z
-coordinates (about
). To distinguish these coordinates from the z-coordinates on , we write them as z
2
and w
2
. Then z
2
= f(z) w
2
= g(z) =
1
f(z)
. Since lim
za
f(z) = , it follows that
lim
za
g(z) = 0. Hence, by the removable singularities theorem, we can extend g(z) holo-
morphically across z = a.
The order of the pole z = a is h if g(z) = (z a)
h
g
h
(z), g
h
(a) ,= 0. Note that, from the
perspective of the claim, zeros and poles are completely analogous: one counts the order of a
preimage under f of 0 S
2
and the other counts the order of a preimage under f of S
2
.
If the pole z = a of f is of order h, then we can write
(z a)
h
f(z) = a
h
+a
h1
(z a) + +a
1
(z a)
h1
+(z)(z a)
h
,
where we have expanded (z a)
h
f(z) using Taylors theorem. Here the a
i
are constants and
is an analytic function. This implies that
f(z) =
a
h
(z a)
h
+
a
h1
(z a)
h1
+ +
a
1
(z a)
+(z).
Denition 17.2. A holomorphic function f : S
2
is said to be meromorphic on .
17.2. Essential singularities. In order to analyze isolated singularities, consider the fol-
lowing conditions:
(1) lim
za
[z a[

[f(z)[ = 0. Here R.
(2) lim
za
[z a[

[f(z)[ = .
Proposition 17.3. Given an isolated singularity z = a there are three possibilities:
A. f 0 and (1) holds for all .
B. There is an integer h such that (1) holds for > h and (2) holds for < h.
C. Neither (1) nor (2) holds for any .
40 KO HONDA
Proof. Suppose (1) holds for some . If (1) holds for some , it holds for larger . Therefore,
take to be a suciently large integer m. We may assume (z a)
m
f(z) has a removable
singularity and vanishes at z = a. Therefore, we can write (z a)
m
f(z) = (z a)
k
g(z), with
g(a) ,= 0, provided f is not identically zero. Therefore, f(z) = (z a)
km
g(z), and (1) holds
for > mk and (2) holds for < mk.
The case where (2) holds for some is similar.
In case B, h is called the algebraic order of f at z = a. If case C holds, then z = a is
called an essential singularity.
At rst glance, essential singularities are not easy to understand or visualize they have
rather peculiar properties.
Theorem 17.4 (Casorati-Weierstrass). An analytic function f comes arbitrarily close to
any complex value in every neighborhood of an essential singularity.
Proof. We argue by contradiction. Suppose there exists A such that [f(z) A[ > > 0 for
all z in a neighborhood of a. Then it follows that
lim
za
[z a[

[f(z) A[ =
for < 0. Therefore, a is not an essential singularity for f(z) A. Using Proposition 17.3,
it follows that there is >> 0 so that
lim
za
[z a[

[f(z) A[ = 0.
Now, [z a[

[A[ 0, so [z a[

[f(z)[ 0 as well, a contradiction.

Remark: The converse is also true. If z = a is an isolated singularity, and, for any disk
[za[ < about a, f(z) comes arbitrarily close to any complex value A, then a is an essential
singularity. (It is easy to verify that B cannot hold under the above conditions.)
Example: f(z) = e
z
has an essential singularity at . (Equivalently, by changing to w =
1
z
coordinates, g(w) = e
1/w
has an essential singularity at w = 0.) Indeed, given any small
open disk about z = , i.e., D = [z[ > R for R large, if we take C >> R, then
the innite annulus C Im z C + 2 D maps onto C0 via f. In this case, the
essential singularity misses all but one point, namely the origin.
Remark: According to the big Picard theorem, near an essential singularity z = a, f
takes every complex value A, with at most one exception! (Moreover, the set f(z) = A is
innite.) We will not prove this fact in this course.
NOTES FOR MATH 520: COMPLEX ANALYSIS 41
18. Local mapping properties
Let f be an analytic function on an open disk D. Suppose f , 0.
Theorem 18.1. Let z
j
be the zeros of f. If is a closed curve in D which does not pass
through any z
j
, then

j
n(, z
j
) =
1
2i
_

(z)
f(z)
dz.
Proof. Suppose f has a nite number of zeros z
1
, . . . , z
n
on D. We assume that a zero is
repeated as many times as its order. Then we can write
f(z) = (z z
1
) (z z
n
)g(z),
where g(z) is analytic and ,= 0 on D. We compute that
f

(z) =

j
(z z
1
)

(z z
j
) (z z
n
)g(z) + (z z
1
) (z z
n
)g

(z),
where a hat indicates the term is omitted. Hence we obtain
f

(z)
f(z)
=
1
z z
1
+ +
1
z z
n
+
g

(z)
g(z)
.
Remark: This is what we would get if log f(z) was a well-dened single-valued function. If
so, we could write (log f(z))

=
f

(z)
f(z)
and log f(z) = log(z z
1
) + +log(z z
n
) +log g(z),
so we would have the above equation.
Now,
1
2i
_

(z)
f(z)
dz =
1
2i
__

dz
z z
1
+ +
_

dz
z z
n
+
_

(z)
g(z)
dz
_
= n(, z
1
) + +n(, z
n
),
observing that
_

(z)
g(z)
dz = 0 by Cauchys Theorem, since g(z) ,= 0 on D
Now suppose there are innitely many zeros of f on D. Since f is not identically zero, the
zeros do not accumulate inside D. Take a smaller disk D

D which contains Im . Then

there exist nitely many zeros of f on D

, and the previous considerations apply.

Interpretation: If f maps from the z-plane to the w-plane, i.e., w = f(z), then we can
write
n(f , 0) =
_
f
dw
w
=
_

(z)
f(z)
dz.
Hence we can interpret Theorem 18.1 as follows:
n(f , 0) =

f(z
i
)=0
n(, z
i
).
42 KO HONDA
More generally, we can write
n(f , a) =

f(z
i
)=a
n(, z
i
),
provided a , Im .
Theorem 18.2. Suppose f(z) is analytic near z
0
, f(z
0
) = w
0
, and f(z) w
0
has a zero of
order n at z
0
. For a suciently small > 0, there exists > 0 such that for all w ,= w
0
in
D

(w
0
), f(z) = w has n distinct roots on D

(z
0
).
Remark: If f(z
0
) = w
0
, then z
0
is a zero of f(z) w
0
of order 1 i f

(z
0
) ,= 0. (Why?)
Proof. Consider a small disk D

(z
0
) centered at z
0
which does not contain any other zeros
of f(z) w
0
. By taking > 0 suciently small, we may assume that f

(z) ,= 0 for z ,= z
0
in D

(z
0
). [If f

(z
0
) ,= 0, then it is clear such a disk exists; if f

(z
0
) = 0, then such a disk
exists since the zeros of f

are isolated.] By the above remark, all the zeros of f(z) A are
of order 1, if z ,= z
0
D

(z
0
). Let = D

(z
0
).
Now take a small disk D

(w
0
) about w
0
= f(z
0
) which misses Im f . Then for w
D

(w
0
)

f(z
i
)=w
n(, z
i
) = n(f , w) = n(f , w
0
) = n,
and the preimages z
i
of w are distinct by the previous paragraph.
Corollary 18.3 (Open Mapping Theorem). An analytic function f , 0 maps open sets to
open sets, i.e., is an open mapping.
Corollary 18.4. If f is analytic at z
0
and f

(z
0
) ,= 0, then f is 1-1 near z
0
and its local
inverse f
1
is analytic.
Proof. By Theorem 18.2, f is 1-1 near z
0
if f

(z
0
) ,= 0; by the previous corollary, f is an
open mapping. It follows from the Claim from Day 11 that the local inverse f
1
is analytic.

Factorization of maps: Suppose f(z)w

0
has a zero of order n at z
0
, i.e., f(z)w
0
= (z
z
0
)
n
g(z) with g(z
0
) ,= 0. Provided [g(z) g(z
0
)[ < [g(z
0
)[, i.e., g(z) is contained in a disk of
radius [g(z
0
)[ about g(z
0
), h(z)
def
=
n
_
g(z) exists, and we can write f(z)w
0
= ((zz
0
)h(z))
n
.
We can then factor w = f(z) into maps z = (z z
0
)h(z), followed by w = w
0
+
n
.
The rst map is locally 1-1 and the second is a standard nth power.
NOTES FOR MATH 520: COMPLEX ANALYSIS 43
19. Maximum principle, Schwarz lemma, and conformal mappings
19.1. Maximum Principle. A corollary of the Open Mapping Theorem is the following:
Theorem 19.1 (Maximum Principle). If f(z) is analytic and nonconstant on an open set
, then [f(z)[ has no maximum on .
Proof. Suppose [f(z)[ attains a maximum at z
0
. Then there is an open disk D

(z
0
)
which maps to an open set about f(z
0
). In particular, [f(z
0
)[ is not maximal.
A reformulation of the Maximum Principle is the following:
Theorem 19.2. If f is dened and continuous on an closed bounded set E and analytic on
int(E), then max [f(z)[ occurs on the boundary of E. (Here bdry(E)
def
= E int(E).)
Proof. Since E is compact, max [f(z)[ is attained on E. It cannot occur on int(E), so must
occur on bdry(E).
19.2. Schwarz Lemma.
Theorem 19.3 (Schwarz Lemma). Suppose f is analytic on [z[ < 1 and satises [f(z)[ 1
and f(0) = 0. Then [f(z)[ [z[ and [f

(0)[ 1. If [f(z)[ = [z[ for some z ,= 0, or if

[f

(0)[ = 1, then f(z) = cz for [c[ = 1.

We can view the Schwarz Lemma as a Contraction Mapping Principle, namely either
the distance from z to 0 is contracted under f for all z D, or else f is a rotation.
Proof. By comparison with g(z) = z. Take f
1
(z) =
f(z)
g(z)
=
f(z)
z
. Since f(0) = 0, f
1
(z) is
analytic and f
1
(0) = f

(0). On [z[ = r 1, [f
1
(z)[
1
r
, so [f
1
(z)[
1
r
on [z[ r by the
Maximum Principle. By taking the limit r 1, we have [f
1
(z)[ 1 on [z[ < 1. Hence
[f(z)[ [z[ and [f

(0)[ 1.
If [f
1
(z)[ = 1 on [z[ < 1, then f
1
is constant. Hence f(z) = cz with [c[ = 1.
19.3. Automorphisms of the open disk and the half plane.
Terminology: A holomorphic map f :
1

2
between two Riemann surfaces is biholo-
morphism if it has a holomorphic inverse.
1
an
2
are then biholomorphic. If
1
=
2
= ,
then f is often called an automorphism. We denote by Aut() the group of automorphisms
of .
As an application of the Schwarz Lemma, we determine all automorphisms f of the open
unit disk D. By the results from last time, if f is analytic and f : D D is 1-1 and onto,
then f
1
is also analytic.
Theorem 19.4. Let f : D D be an automorphism satisfying f(0) = 0. Then f(z) = e
i
z
for some R.
HW 15. Use the Schwarz Lemma to prove Theorem 19.4.
44 KO HONDA
To understand the general case, i.e., f() = , we need an automorphism g

: D D which
maps to 0. Consider g

(z) =
z
1z
, where D.
HW 16. Verify that g

is a fractional linear transformation which takes D to itself.

If f : D D is an automorphism which takes to , then g

f g
1

maps 0 to 0,
hence is e
i
z. This implies that the group Aut(D) of automorphisms of D consists of f(z) =
g
1

(e
i
g

(z)). In particular, automorphisms of the open unit disk are all fractional linear
transformations.
We now consider automorphisms of the upper half plane H = Im z > 0. Let g : C C
be the fractional linear transformation z
zi
z+i
.
HW 17. Prove that g maps H biholomorphically onto D.
Observe that f is an automorphism of H i g f g
1
is an automorphism of D. (We say f
is conjugated by g.) Hence, Aut(H) = g
1
Aut(D)g. Alternatively, we can use the following
HW to show that Aut(H) = PSL(2, R).
HW 18. The fractional linear transformations that preserve H are given by PSL(2, R).
19.4. Conformal mappings. Towards the end of this course, we will prove the following
fundamental result:
Theorem 19.5 (Riemann Mapping Theorem). Any simply connected (connected) region
C is biholomorphic to the open unit disk D.
Remark: If = C, then any analytic function f : C D is bounded, and hence constant.
Hence, C and D are not biholomorphic!
We will dene simple connectivity in due course, but for the time being, it roughly means
that there are no holes in . It turns out that the only simply connected Riemann surfaces
are D, S
2
, and C, and we have the following theorem:
Theorem 19.6.
(1) Aut(D) Aut(H) PSL(2, R).
(2) Aut(S
2
) PSL(2, C).
(3) Aut(C) az +b [ a, b C.
The second and third assertions are left for HW. Also observe that Aut() in the Riemann
Mapping Theorem is conjugate to (and hence isomorphic to) Aut(D).
Examples: We now give examples of simply connected regions that are biholomorphic
to D. We emphasize that all the regions are open! By a succession of biholomorphisms, all
the regions that appear here are biholomorphic to D.
1. H D by z
zi
z+i
as before.
2. (First quadrant) x > 0, y > 0 H by sending z z
2
.
3. (Quarter disk) x > 0, y > 0, [z[ < 1 (half disk) [z[ < 1, y > 0 via z z
2
.
NOTES FOR MATH 520: COMPLEX ANALYSIS 45
4. (Half disk) [z[ < 1, y > 0 (rst quadrant) x > 0, y > 0 via z
1+z
1z
.
5. (Half disk) [z[ < 1, y > 0 (semi-innite strip) 0 < y < , x < 0 via z log z.
6. H (innite strip) 0 < y < via z log z.
7. (Half plane with slit) x > 0 y = 0, 0 < x 1 x > 0 by a succession of maps
z z
2
, z z 1, z

z, whose composition is z

z
2
1.
8. He
i
[0 < < a < x > 0 y = 0, 0 < x b via z
z1
z+1
.
9. 0 < y < x = 0, 0 < y < a He
i
[0 < < a < via z e
z
.
46 KO HONDA
20. Weierstrass theorem and Taylor series
20.1. Weierstrass theorem. Consider the sequence f
n
, where f
n
is analytic on the open
set
n
. Suppose in addition that
1

2

n
. . . and =
n

n
.
Theorem 20.1. Suppose f
n
converges to f on , uniformly on every compact subset of
. Then f is analytic on . Moreover, f

n
converges uniformly to f

on every compact subset

of .
We will often write UCOCS to mean uniform convergence on compact sets.
Remark: Every compact set E is covered by
n
, so must be contained in some
n
and hence all
n
for n

n.
Proof. Suppose D = [z a[ r . (Here D will denote the open disk.) By the Cauchy
integral formula,
f
n
(z) =
1
2i
_
D
f
n
()d
z
,
for all z D. As n ,
f(z) =
1
2i
_
D
f()d
z
,
so f is analytic on D. [Since f
n
(z) f(z) uniformly on D,
_
D
fn()d
z

_
D
f()d
z
as
n ; this is a standard property of Riemann integrals under uniform convergence.]
Similarly, f

n
(z) =
1
2i
_
D
fn()d
(z)
2
, so f

n
(z) f

(z) UOCS.
Corollary 20.2. If a series f(z) = f
1
(z) +f
2
(z) + . . . , with f
i
analytic on , has UCOCS
of , then f is analytic on , and its derivative can be dierentiated term-by-term.
Remark: It suces to prove uniform convergence on the boundary of compact sets E, by
the maximum principle. [f
n
(z) f
m
(z)[ on E i [f
n
(z) f
m
(z)[ on E.
Theorem 20.3. If f
n
is analytic and nowhere zero on , and f
n
f UOCS, then f(z) is
either identically 0 or never zero on .
Proof. Suppose f is not identically zero. If f(z
0
) = 0, then there exists > 0 such that
f(z) ,= 0 on D

(z
0
) z
0
. Then f
n
f uniformly on D

(z
0
) and also f

n
f

uniformly on D

(z
0
). Hence we have
lim
n
1
2i
_
D

(z
0
)
f

n
(z)
f
n
(z)
dz =
1
2i
_
D

(z
0
)
f

(z)
f(z)
dz.
For suciently large n, f
n
(z) ,= 0 on D since f ,= 0 on D; therefore the left-hand side
makes sense. Now, the LHS is zero, but the RHS is nonzero, a contradiction.
NOTES FOR MATH 520: COMPLEX ANALYSIS 47
20.2. Taylor series.
Theorem 20.4. Let f be an analytic function on and a . Then
f(z) = f(a) +f

(a)(z a) + +
f
(n)
(a)
n!
(z a)
n
+. . . .
The power series converges in the largest open disk D centered at a.
Proof. Recall that f(z) has a Taylor expansion:
f(z) = f(a) +f

(a)(z a) +
f

(a)
2!
(z a)
2
+ +
f
(n1)
(a)
(n 1)!
(z a)
n1
+ (z a)
n
f
n
(z),
with remainder term:
f
n
(z) =
1
2i
_
D
f()d
( a)
n
( z)
.
As before, we estimate
[(z a)
n
f
n
(z)[
_
[z a[
R
_
n
MR
R [z a[
,
where R is the radius of D. On a slightly smaller disk D

(a) = [z a[ , where < R,

the remainder term [(z a)
n
f
n
(z)[ 0 uniformly. This proves the theorem.
One often refers to this theorem as a holomorphic function (one that has a complex deriv-
ative) is analytic (can be written as a power series).
48 KO HONDA
21. Plane topology
21.1. Chains and cycles. Let C be an open region.
Denition 21.1. A chain is a formal sum = a
1

1
+ + a
n

n
, where a
i
Z and

i
: [
i
,
i
] is piecewise dierentiable. In other words, a chain is an element of the free
abelian group generated by piecewise dierentiable arcs in .
Write C
0
() for the free abelian group generated by points in , and C
1
() for the free
abelian group generated by piecewise dierentiable arcs in .
Consider the map : C
1
() C
0
(), which maps an arc : [a, b] to (b) (a). It is
extended linearly to elements of C
1
(), i.e., formal linear combinations of arcs. Elements of
ker are called cycles of .
Let be a dierential pdx+qdy or f(z)dz. Consider the integration map C
1
()

C, which
sends an arc to
_

, and is extended linearly to C

1
(). We observe the following:
(1) If : [a, b] and
1
= [
[a,c]
,
2
= [
[c,b]
, where a < c < b, then
_

=
_

1
+
_

2
.
(2) If

is an orientation-preserving reparametrization of , then

_

=
_

.
(3) If is the opposite arc of , then
_

=
_

.
Therefore, as far as integration is concerned, we may take equivalence relations
1
+
2
,

, and () 1().
Remark: Given a cycle =

a
i

i
, it is equivalent to a sum

1
+ +

k
, where each

i
is a closed curve. First, by reversing orientations if necessary, we can write

i
,
where
i
may not be closed. Since =

i
= 0, given some
i
: [
i
,
i
] , there exists

j
: [
j
,
j
] such that
i
(
i
) =
j
(
j
). We can then concatenate
i
and
j
to obtain a
shorter chain. Repeat until all the
i
are closed curves.
21.2. Simple connectivity. We give three alternate denitions for a connected open set
C to be simply connected, and prove their equivalence.
Denition 21.2. A connected open region C is simply connected if one of the following
equivalent conditions holds:
(1) C is connected.
(2) n(, a) = 0 for all cycles in and a , .
(3) Every continuous closed curve in is contractible.
Denition (3) is the usual denition of simple connectivity.
Roughly speaking, a simply connected region is a region without holes. Examples of simply
connected are C, the open unit disk, the upper half plane, and 0 < y < 2.
Two continuous closed curves
0
,
1
: [a, b] are homotopic in if there is a continuous
function : [a, b] [0, 1] such that (s, 0) =
0
(s) and (s, 1) =
1
(s). is a homotopy
NOTES FOR MATH 520: COMPLEX ANALYSIS 49
from
0
to
1
. A continuous closed curve : [a, b] is contractible if is homotopic to a
closed curve that maps to a point in .
(1)(2). Suppose C is connected. C Im is open and n(, a) is constant for all a
in a given component of C Im . Since C is connected, the component of C Im
containing must also contain C . Now n(, a) = 0 for all a in the component of
C Im containing .
(2)(1). Suppose C is not connected. We can write C = A . B, where A and B
are open sets (and hence closed sets), and B. A and B are both closed and bounded
in C, hence compact. Hence d(A, B) = min
xA,yB
d(x, y) exists and is positive.
HW 19. Given compact subsets A, B of a complete metric space X, prove that d(A, B) =
min
xA,yB
d(x, y) exists and is positive if A and B are disjoint.
Now cover C with a net of squares whose side is <
d(A,B)

2
. Let Q
i

i=1
be the set of (closed)
squares. Observe that each Q
i
intersects at most one of A or B. Then take =

Q
i
A=
Q
i
.
We claim that if we simplify by canceling pairs of edges, then does not meet A and B,
hence . Not meeting B is immediate, since Q
i
A ,= implies Q
i
B = . If a A is
contained in an edge of , then a is in adjacent Q
i
and Q
j
whose common edge is subjected
to a cancellation, a contradiction. [A similar argument holds in case a is a corner of Q
i
.]
Now, n(Q
i
, a) = 1 and n(Q
j
, a) = 1 if i ,= j, for a int(Q
i
). Hence n(, a) = 1 for all
a int(Q
i
), where the union is over all Q
i
A ,= . We could have taken the net so that
a A is at the center of one of the squares, so we have obtained on and a A so that
n(, a) = 1, a contradiction.
50 KO HONDA
22. The general form of Cauchys theorem
22.1. Simple connectivity, contd. We continue the proof of the equivalence of the three
denitions of simple connectivity.
(3)(2). We rst dene n(, a), a Im , for a continuous closed curve as follows: Divide
into subarcs
1
, . . . ,
n
so that Im
i
is contained in a disk which does not intersect a.
(This is possible by compactness.) Replace
i
by a line segment
i
with the same endpoints
as
i
. Then =
1
+ +
n
, and dene n(, a) = n(, a).
HW 20. Prove that this denition of n(, a) does not depend on the choice of
j
, and agrees
with the standard denition of n(, a) for piecewise smooth closed curves.
HW 21. Prove that if two closed curves
0
and
1
are homotopic through a homotopy that
does not intersect a, then n(
0
, a) = n(
1
, a).
Given a piecewise dierentiable closed curve , n(, a) = n(pt, a) = 0, by the above HW, if
a , . Hence n(, a) = 0 for all on and a , .
(1),(2)(3). We will cheat slightly and appeal to the Riemann Mapping Theorem, i.e., a
simply connected, connected region C is biholomorphic to the open disk. (We will
prove the Riemann Mapping Theorem later, using only denitions (1) and (2) of simple
connectivity.) Any closed curve on D can easily be contracted to a point and any closed
curve on C can also be contracted to a point.
22.2. General form of Cauchys Theorem. A cycle on is homologous to zero (or
nullhomologous) in if n(, a) = 0 for all a C. Two cycles
1
and
2
are homologous
if
1

2
is nullhomologous. We will write [] to denote the equivalence class of cycles
homologous to . In particular, [] = 0 means is nullhomologous.
Remark: This is not the usual denition of a nullhomologous .
Theorem 22.1. If f(z) is analytic on , then
_

f(z)dz = 0 for all cycles which are

nullhomologous in .
Corollary 22.2. If f(z) is analytic on a simply connected , then
_

f(z)dz = 0 for all

cycles on .
Proof of Corollary. If is simply connected, then any cycle on is nullhomologous by
condition (2).
Suppose is simply connected and f is analytic on . Since
_
f(z)dz is independent of the
choice of path, f(z) is the complex derivative of an analytic function F(z) which is dened
on all of .
Corollary 22.3. If f(z) is analytic and nowhere zero on a simply connected , then log f
and
n

f admit single-valued branches on .

NOTES FOR MATH 520: COMPLEX ANALYSIS 51
Proof. Let F(z) be the analytic function on whose derivative is
f

(z)
f(z)
. This is basically
log f(z), but we need to check that it satises e
F(z)
= f(z). First we verify that f(z)e
F(z)
has
derivative 0, hence f(z)e
F(z)
is constant. Choosing z
0
, we set f(z)e
F(z)
= f(z
0
)e
F(z
0
)
,
and hence
e
F(z)F(z
0
)log f(z
0
)
= f(z).
We take F(z) F(z
0
) log f(z
0
) to be log f. We also take
n

f = e
1
n
log f
.
Remark: This corollary will become crucial in the proof of the Riemann Mapping Theorem.
We will prove the following more general result:
Theorem 22.4. Let = pdx + qdy be a locally exact dierential on , i.e., it is exact in
some neighborhood of each point of . Then
_

= 0 for all nullhomologous cycles in .

Proof.
Step 1: Let [] = 0. We may assume that the cycle is a nite sum of closed curves by
the Remark from last time. We now replace by a piecewise linear one with each piece
a horizontal or vertical line segment: Subdivide (which we assume WLOG to be a single
closed curve) into subarcs
i
as above, so that each Im
i
lies in a small disk . Replace

i
by a horizontal arc, followed by a vertical one, which we call
i
.
i
and
i
have the same
endpoints. Since is exact on this disk (if suciently small),
_

i
=
_

i
. Also one can
easily verify that [] = [].
Step 2: Extend the horizontal and vertical line segments of to lines in C. The lines cut
up C into rectangles R
i
and unbounded regions. Pick a
i
int(R
i
) and form

0
=

n(, a
i
)R
i
.
We claim that and
0
become equivalent after cancelling pairs of edges. Indeed, if a
int(R
i
), then n(
0
, a) =

n(, a
i
)n(R
i
, a) = n(, a
i
) = n(, a). Moreover n(
0
, a) =
n(, a) = 0 for a in the unbounded regions. Suppose the reduced expression of
0
contains the multiple c
ij
, where
ij
is a common side of rectangles R
i
and R
j
. Then
n(
0
cR
i
, a
i
) = n(
0
cR
i
, a
j
), but then the LHS is c and the RHS is 0, hence
implying c = 0. Now

n(, a
i
)R
i
as far as integration is concerned.
Step 3: If n(, a
i
) ,= 0, then we claim that R
i
. First, int(R
i
) , since a int(R
i
)
n(, a) = n(, a
i
) ,= 0. Next, if a lies on an edge of R
i
, then either that edge is in Im (and
hence in ), or else the edge is not contained in Im , in which case n(, a) ,= 0 and a .
Hence, if n(, a
i
) ,= 0, then
_
R
i
= 0. This proves the theorem.
52 KO HONDA
23. Multiply connected regions and residues
23.1. Multiply connected regions. A (connected) region which is not simply connected
is said to be multiply connected. has connectivity n if C has n connected components
and innite connectivity if C has innitely many connected components.
We consider the case of nite connectivity. Then there are n connected components of
C, which we write as A
1
, . . . , A
n
, with A
n
.
Lemma 23.1. There exist closed curves
1
, . . . ,
n1
such that n(
i
, a
j
) =
ij
, if a
j
A
j
.
Proof. Use the proof of (2)(1) from Day 21. In other words, take a suciently ne net of
squares Q
k
that covers A
i
(and A
i
only), and let
i
=

Q
k
A
i
=
Q
k
.
Lemma 23.2. Every cycle on is homologous to a (unique) linear combination c
1

1
+
+c
n1

n1
.
Proof. If c
i
= n(, a
i
) for a
i
A
i
, then n(

c
i

i
, a
j
) = 0 for all a
j
A
j
, i = 1, . . . , n1.
A
n
contains , so n(

c
i

i
, a
n
) = 0 as well. Hence [] = [

c
i

i
].
Next to prove uniqueness, if [

a
i

i
] = [

b
i

i
], then [

(a
i
b
i
)
i
] = 0, and evaluation
on the
j
gives a
i
= b
i
.
We say that
1
, . . . ,
n1
form a homology basis for .
By Cauchys Theorem, we have
_

f(z)dz = c
1
_

1
f(z)dz + +c
n1
_

n1
f(z)dz.
We call
_

i
f(z)dz the periods of the dierential fdz.
Example: Let be the annulus r
1
< [z[ < r
2
. has connectivity 2, and has a homology
basis consisting of one element, the circle C of radius r, r
1
< r < r
2
. Then any cycle is
homologous to nC for some n Z, and
_

f(z)dz = n
_
C
f(z)dz.
23.2. Residues. Let f(z) be an analytic function on the region , with the exception of iso-
lated singularities. Suppose for the moment that there are nitely many isolated singularities
a
1
, . . . , a
n
. We assume that they are either poles or essential singularities.
Let C
j
be a small circle about a
j
which contains no other singularities in the interior.
If we let R
j
=
1
2i
_
C
j
f(z)dz, then f(z)
R
j
za
j
has vanishing period about a
j
. R
j
is said to
be the residue of f(z) at z = a
j
, and is written Res
z=a
j
f(z).
Theorem 23.3 (Residue Formula). If f(z) is analytic on with the exception of isolated
singularities, then
1
2i
_

f(z)dz =

j
n(, a
j
)Res
z=a
j
f(z)
for any cycle on which is nullhomologous on and avoids the singular points.
NOTES FOR MATH 520: COMPLEX ANALYSIS 53
Proof. Since [] = 0 on , n(, a) = 0 for all a , . Hence n(

j
n(, a
j
)C
j
, a) = 0 for
all a ,

= a
1
, . . . , a
n
. In other words, [

j
n(, a
j
)C
j
] = 0 in

. Hence we have:
1
2i
_

f(z)dz =

j
n(, a
j
)Res
z=a
j
f(z).
Observe that there are only nitely many isolated singularities in the bounded connected
components of CIm .
23.3. Computations of residues. In principle, the isolated singularities can be essential
singularities, but in practice we will only treat poles.
Lemma 23.4. If a meromorphic function f is written as
B
k
(za)
k
+ +
B
1
za
+g(z) near z = a,
where g(z) is analytic, then Res
z=a
f(z) = B
1
.
Proof. Observe that all the other terms besides
B
1
za
admit antiderivatives. (g(z) by Cauchys
theorem on the disk, and the others explicitly.) Hence
1
2i
_
C
f(z)dz only leaves
1
2i
_
C
B
1
dz
za
,
where C is a small circle about z = a.
In particular, if f has a simple pole at z = a, then f(z) =
B
1
za
+ g(z) and (z a)f(z) =
B
1
+ (z a)g(z). Evaluating (z a)f(z) at z = a gives the residue B
1
.
Example: f(z) =
cos z
(z1)(z2)
. Since z = 1, 2 are not zeros of cos z, they are simple zeros of
f(z). Res
z=1
f(z) = cos 1 and Res
z=2
f(z) = cos 2. If is a closed curve which winds once
around z = 1 and z = 2 (for example, if is the circle [z[ = 3, oriented counterclockwise),
then
_

f(z)dz = n(, 1)Res

z=1
f(z) +n(, 2)Res
z=2
f(z) = cos 1+cos 2. If winds around
z = 1 only, say is the circle [z[ =
3
2
, oriented counterclockwise, then
_

f(z)dz = cos 1.
Example: f(z) =
sinz
z
4
. The only possible pole is z = 0. Expand sin z using Taylors
theorem: sin z = z
z
3
3!
+ z
5
g(z). Then f(z) =
1
z
3

1
6
1
z
+ zg(z), and Res
z=0
f(z) =
1
6
.
Similarly, f(z) =
sinz
z
5
has Res
z=0
f(z) = 0.
54 KO HONDA
24. Residues
24.1. Application: The argument principle. Suppose f is meromorphic on , with
zeros a
i
and poles b
i
. We will compute n(f , 0) =
_

(z)
f(z)
dz, using the Residue Formula
from last time. (This is a generalization of Theorem 18.1 from Day 18.)
Theorem 24.1. n(f , 0) =

j
n(, a
j
)

j
n(, b
j
), where a
j
and b
j
are repeated as
many times as their orders.
Proof. Near a zero z = a of order h, we have f(z) = (z a)
h
g(z), where g(z) is a nonzero
analytic function. Therefore,
f

(z)
f(z)
=
h
z a
+
g

(z)
g(z)
,
and Res
z=a
f

(z)
f(z)
= h, which is the order of the zero. Similarly, near a pole z = b of order k,
f(z) = (z a)
k
g(z), and Res
z=b
f

(z)
f(z)
= k, which is minus the order of the pole.
A corollary of the Argument Principle is the following theorem:
Theorem 24.2 (Rouches Theorem). Suppose [] = 0 on and n(, z) = 0 or 1 for all
z Im . Also suppose that f(z), g(z) are analytic on and satisfy [f(z)g(z)[ < [f(z)[
on . Then f(z) and g(z) have the same number of zeros enclosed by .
Proof. Take
g(z)
f(z)
. Then on we have [
g(z)
f(z)
1[ < 1. Therefore, Im
g
f
is contained in the
disk of radius 1 about z = 1, and hence n(
g
f
, 0) = 0. This implies that

j
n(, a
j
) =

k
n(, b
k
), where a
j
are the zeros of
g
f
and b
k
are the poles of
g
f
. Upon some thought,
one concludes that f and g must have the same number of zeros enclosed by . [Here z is
enclosed by means n(, z) ,= 0.]
Example: Consider g(z) = z
8
5z
3
+ z 2. Find the number of roots of g(z) inside the
unit disk [z[ 1.
Let be [z[ = 1, oriented counterclockwise. Also let f(z) = 5z
3
. Then on we have
[f(z)g(z)[ = [z
8
+z2[ [z
3
[+[z[+[2[ 4, whereas [g(z)[ = 5. Hence [f(z)g(z)[ < [g(z)[,
and the number of zeros of g in [z[ < 1 is equal to the number of zeros of f in [z[ < 1, which
in turn is 3 (after counting multiplicities).
24.2. Evaluation of denite integrals.
A.
_
2
0
R(cos , sin )d, where R is a rational function of two variables.
Example:
_
2
0
1
a+sin
d, where a is real and > 1. We change coordinates z = e
i
and
integrate over the closed curve = [z[ = 1, oriented counterclockwise. Then dz = ie
i
d,
and d = i
dz
z
. Also we have
1
2
(z +
1
z
) =
1
2
(e
i
+e
i
) = cos and
1
2i
(z
1
z
) = sin .
NOTES FOR MATH 520: COMPLEX ANALYSIS 55
We substitute
_
2
0
d
a + sin
= 2
_

dz
z
2
+ 2iaz 1
.
If we write z
2
+2iaz1 = (z)(z), then = i(a

a
2
1) and = i(a+

a
2
1).
Observe that is in the unit disk, while is not. The residue of
1
(z)(z)
at z = is
1

,
and
2
_

dz
z
2
+ 2iaz 1
= 2(2i)Res
z=
1
(z )(z )
=
2

a
2
1
.
B.
_

f(x)dx, where f(z) is meromorphic, has a nite number of poles, has no poles on
R, and satises [f(z)[
B
|z|
2
for [z[ >> 0. If f is a rational function, then the last condition
means that the degree of the denominator is at least two larger than the degree of the
numerator.
Example:
_

1
x
4
+1
dx. First consider the integral of
1
z
4
+1
over the closed curve consisting
of an arc C
1
from R to R on R, followed by a counterclockwise semicircle C
2
= [z[ =
R, Im z 0.
We rst observe that

_
C
2
1
z
4
+ 1
dz

_
C
2
1
[z[
4
[dz[
R
R
4
=

R
3
0
as R .
Hence,
_

1
x
4
+ 1
dz =
_

dz
z
4
+ 1
.
The fourth roots of 1 are e
i/4
, e
3i/4
(and are simple, in particular). Only two of them
e
i/4
and e
3i/4
are contained in the region bounded by . Therefore,
_

dz
z
4
+ 1
= 2i(Res
z=e
i/4 +Res
z=e
3i/4).
A convenient way of computing the residues is:
HW 22. Prove that the residue of
1
f(z)
at z = a is
1
f

(a)
if a is a simple pole.
Hence
_

dz
z
4
+ 1
= 2i
_
1
4e
3i/4
+
1
4e
9i/4
_
.
[Of course, this can be further simplied to give a real expression....]
56 KO HONDA
25. Day 25
25.1. Evaluation of denite integrals, Day 2.
C. (Fourier Transforms) Integrals of the form
_

f(x)e
iax
dx, where f(z) is meromorphic,
has a nite number of poles (none of them on the x-axis), and [f(z)[
B
|z|
for [z[ >> 0.
[Observe that we only need f(z) to have a zero of order at least 1 at . Compare with B,
where f(z) was required to have a zero of order at least 2 at .]
Example:
_

e
iax
x
2
+1
dx, a > 0. Integrate on the contour which is the counterclockwise
boundary of the rectangle with vertices X
1
, X
2
, X
2
+iY, X
1
+iY , where X
1
, X
2
, Y >> 0.
Call the edges of the rectangle C
1
, C
2
, C
3
, C
4
in counterclockwise order, where C
1
is the edge
on the x-axis from X
1
to X
2
. Using the bound [
1
z
2
+1
[
B
|z|
, we obtain:

_
C
2
e
iaz
z
2
+ 1
dz

B
X
2
_
Y
0
e
ay
dy =
B
X
2
_
1
a
_
(1 e
aY
),

_
C
4
e
iaz
z
2
+ 1
dz

B
X
1
_
1
a
_
(1 e
aY
),

_
C
3
e
iaz
z
2
+ 1
dz

B
Y
_
X
2
X
1
e
aY
[dx[ =
B
Y
e
aY
(X
1
+X
2
).
If we take X
1
, X
2
to be large, and then let Y , then all three integrals go to zero. Hence,
_

e
iax
x
2
+ 1
dx =
_

e
iaz
z
2
+ 1
dz = 2i Res
z=i
_
e
iaz
z
2
+ 1
_
= 2i
e
iaz
z +i

z=i
= e
a
.
Example:
_

sinx
x
dx. In this problem, we want to compute
_

e
ix
x
dx, which has a pole
at x = 0. Instead, we take the Cauchy principal value p.v.
_

def
= lim
0
(
_

+
_

).
Strangely enough,
p.v.
_

e
ix
x
dx =
_
p.v.
_

cos x
x
dx
_
+i
__

sin x
x
dx
_
,
since
sinx
x
is actually continuous at x = 0 and theres no need for p.v.s.
Take to be the boundary of RD, where R is the rectangle as in the previous example,
and D is the disk of radius about z = 0. There is one pole inside R D, which gives
2i Res
z=0
e
iz
z
= 2i. Now,
2i =
_

e
iz
z
dz = p.v.
_

e
iz
z
dz + lim
0
_
C
e
iz
z
dz,
where C is the lower semicircle of D. Writing e
iz
=
1
z
+ g(z), where g(z) is analytic (and
hence continuous) near z = 0, we nd that
_
C
e
iz
z
dz
_
C
1
z
dz = i as 0. We therefore
NOTES FOR MATH 520: COMPLEX ANALYSIS 57
obtain
p.v.
_

e
iz
z
dz = i,
_

sin x
x
dx = .
D. (Mellin Transforms) Integrals of the form
_

0
f(x)x

dx, where 0 < < 1. Here f(z) is

meromorphic, has a nite of poles (none of them on the x-axis), and [f(z)[
B
|z|
2
for [z[ >> 0
and [f(z)[
B
|z|
for [z[ near 0.
We take the contour = C
1
+ C
2
+ C
3
+ C
4
, where C
1
= [z[ = R
1
, oriented clockwise,
C
3
= [z[ = R
2
, oriented counterclockwise, C
2
is the line segment from z = R
1
to z = R
2
,
and C
4
is the line segment from z = R
2
to z = R
1
. Here R
1
< R
2
. [More rigorously, were
interested in the boundary of the region R
1
< r < R
2
, < < 2 (in polar coordinates)
and were letting 0.]
Then

_
C
1
f(z)z

dz

B
R
1
1
2R
1
= 2BR

1
0, as R
1
0;

_
C
3
f(z)z

dz

B
R
2
2
2R
2
=
2B
R
1
2
0, as R
2
.
On the other hand, since z

= e
log z
, the values of log z on C
2
and C
4
dier by 2i and
those of z

dier by e
2i
. Hence,
_
C
4
f(z)z

dz =
_
C
2
f(z)e
2i
z

dz,
(1 e
2i
)
_
C
2
f(z)z

dz = 2i

Res
z=a
j
(f(z)z

),
_

0
f(x)x

dx = 2i

Res
z=a
j
(f(z)z

),
where the sum is over all residues in the plane.
This technique can be used to compute integrals such as
_

0
x

x
2
+1
dx.
25.2. Harmonic functions. A function u : R is harmonic if u =

2
u
x
2
+

2
u
y
2
= 0. For
the time being, assume u has continuous partials up to second order.
Recall: If f = u +iv and f is analytic, then u and v are harmonic.
Example: The simplest harmonic functions are u(x, y) = ax +by.
Example: If f(z) = log z, then f(z) = log r + i, where z = re
i
. (Suppose we are
restricting attention to a domain where is single-valued and continuous.) u(x, y) =
log r = log
_
x
2
+y
2
and v(x, y) = are harmonic.
Question: Given a harmonic form u on , is there a harmonic conjugate v : R, such
that f = u +iv is analytic on ?
58 KO HONDA
Let du =
u
x
dx +
u
y
dy. Then we dene
du =
u
x
dy
u
y
dx.
( is called the Hodge -operator.)
Lemma 25.1. If u is harmonic, then g(z) =
u
x
i
u
y
is analytic.
This g(z) is the (complex) derivative of the function f(z) we are looking for.
Proof. We compute that g(z) satises the Cauchy-Riemann equations
g
x
= i
g
y
:
g
x
=

2
u
x
2
i

2
u
xy
,
g
y
=

2
u
xy
+i

2
u
x
2
.

NOTES FOR MATH 520: COMPLEX ANALYSIS 59

26. More on harmonic functions
26.1. Existence of harmonic conjugate. Let u : R be a harmonic function. We are
looking for its harmonic conjugate.
Lemma 26.1.
_

du = 0 if [] = 0 (i.e., is nullhomologous).
Proof. Dene g(z) =
u
x
i
u
y
. Since
g(z)dz =
_
u
x
i
u
y
_
(dx +idy) = du +i du,
we have
_

g(z)dz =
_

du +i
_

du.
The integral over g(z)dz is zero by Cauchys Theorem, if [] = 0. The integral over du is
always zero, since du is exact. Hence,
_

du = 0 whenever [] = 0.
Remark: This also follows from remarking that = du is a closed dierential 1-form.
A closed 1-form = pdx + qdy satises d = (
q
x

p
y
)dxdy = 0. It is known (Poincare
Lemma) that closed forms are locally exact.
Theorem 26.2. There is a single-valued harmonic conjugate v : R if is simply
connected. Moreover, v is uniquely determined up to a constant.
Proof. By the above lemma,
_

du = 0 for all cycles on a simply-connected . Hence u

is exact on . To show uniqueness, if f
1
= u + iv
1
and f
2
= u +iv
2
are both analytic, then
f
1
f
2
= i(v
1
v
2
) is also analytic. Recall that a nonconstant analytic function is an open
mapping. Therefore, v
1
v
2
, which has image on the y-axis, must be constant.
In general, the obstacles to the existence of a single-valued harmonic conjugate v are the
periods
_

i
du where
i
are basis elements of the homology of .
26.2. Mean value property.
Theorem 26.3. Let u : R be a harmonic function and D
r
(z
0
) be a closed disk.
Then u(z
0
) =
1
2
_
2
0
u(z
0
+re
i
)d.
This is called the mean value property, since the value of u at the center of a disk is the
average of the values on the boundary of the disk.
Proof. On the disk D
r
(z
0
) the harmonic function u has a harmonic conjugate v. Hence
f = u +iv is analytic and by the Cauchy integral formula we have:
f(z
0
) =
1
2i
_
Dr(z
0
)
f(z)dz
z z
0
=
1
2
_
2
0
f(z
0
+re
i
)d.
Taking the real part of the equation yields the theorem.
60 KO HONDA
Corollary 26.4. u does not attain its maximum or minimum in the interior of .
Theorem 26.5. Let u be a harmonic function dened on the annulus A = r
1
< [z[ < r
2
.
Then
1
2
_
|z|=r
ud = log r +, where , are constant. Here r
1
< r < r
2
.
In view of the mean value property, if u is dened on D = [z[ < r
2
, then = 0 and
= u(0).
Proof. The proof we give has elements of analytic continuation, which we will discuss in more
detail later. Consider subsets U
1
= 0 < < , U
2
=

2
< <
3
2
, U
3
= < < 2,
and U
4
=
3
2
< <
5
2
(of A). Since each U
j
is simply-connected, on each U
j
there exists
a harmonic conjugate v
j
of u. Having picked v
1
, pick v
2
so that v
1
= v
2
on U
1
U
2
. (Recall
they initially dier by a constant. Then modify v
2
so it agrees with v
1
.) Likewise, pick v
3
so
that v
2
= v
3
on U
2
U
3
, and pick v
4
.... Write f
j
= u +v
j
.
Suppose f
4
f
1
= iC, where C is a constant. Consider the functions F
j
= f
j

C
2
log z,
where branches of log z are chosen so that f
j

C
2
log z agrees with f
j+1

C
2
log z. Now,
log z for F
4
and log z for F
1
dier by 2i and F
4
F
1
= iC
C
2
2i = 0. Therefore, the F
j
glue to give a globally dened analytic function F. Thus,
1
2i
_
|z|=r
F(z)
z
dz =
1
2
_
2
0
F(re
i
)d
is constant and its real part is:
1
2
_
|z|=r
ud +
1
2
_
|z|=r
C
2
log rd =
1
2
_
|z|=r
ud +
C
2
log r.

26.3. Poissons formula. We now give an explicit formula for which expresses the values
of a harmonic function u on a disk in terms of the values of u on the boundary of the disk.
This is analogous to the Cauchy integral formula (and is indeed a consequence of it).
Theorem 26.6. Suppose u(z) is harmonic on [z[ < R and continuous for [z[ R. Then
u(a) =
1
2
_
|z|=R
R
2
[a[
2
[z a[
2
u(z)d =
1
2
_
|z|=R
Re
z +a
z a
u(z)d,
if [a[ < R.
Sometimes we will use the center expression and at other times we will use the RH ex-
pression.
Proof. For simplicity, well assume that R = 1 and u is harmonic on all of [z[ R. The
Poisson formula is just a restatement of the mean value property.
First observe that the fractional linear transformation S(z) =
z+a
az+1
sends the unit disk
D = [z[ < 1 to itself and 0 to a. (Recall our discussion of automorphisms of D.) Then
NOTES FOR MATH 520: COMPLEX ANALYSIS 61
u S is a harmonic function on D and by the mean value property we have:
(2) u(a) = u S(0) =
1
2
_
2
0
u(S(z))d arg z.
We now rewrite the RH integral in terms of w = S(z). Inverting, we obtain z = S
1
(w) =
wa
aw+1
and
d arg z = i
dz
z
= i
_
1
w a
+
a
aw + 1
_
dw =
_
w
w a
+
aw
aw + 1
_
d.
By using w =
1
w
on the unit circle, we can rewrite the RHS as:
_
w
w a
+
a
w a
_
d =
1 [a[
2
[w a[
2
d,
or alternatively as
1
2
_
w +a
w a
+
w +a
w a
_
d = Re
w +a
w a
d.

By using the RH expression in the Poisson integral formula (in Theorem 26.6), it follows
that u(z) is the real part of the analytic function
f(z) =
1
2i
_
|z|=R
+z
z
u()
d

+iC.
(f is analytic by Lemma 15.2 from Day 15.)
62 KO HONDA
27. Schwarz reflection principle
27.1. Schwarzs Theorem.
Theorem 27.1. Given a piecewise continuous function U() on 0 2, the Poisson
integral
P
U
(z) =
1
2
_
2
0
Re
e
i
+z
e
i
z
U()d
is harmonic for [z[ < 1 and lim
ze
i
0
P
U
(z) = U(
0
), provided U is continuous at
0
.
Proof. By Lemma 15.2, f(z) =
1
2i
_
||=1
+z
z
U()
d

is analytic on [z[ < 1, and hence P

U
(z)
is harmonic for [z[ < 1.
P is a linear operator which maps piecewise continuous functions U on the unit circle to
harmonic functions P
U
on the open unit disk. It satises P
U
1
+U
2
= P
U
1
+P
U
2
and P
cU
= cP
U
for c constant. We can easily deduce that P
c
= c and that m U M implies m P
U
M.
(You can verify this using Equation 2.)
WLOG assume that U(
0
) = 0; otherwise we can take U U(
0
) instead (and still call it
U). (Observe that we can do this because P
c
= c.) Take a short arc C
2
D (here D is the
unit disk) containing
0
in its interior, such that [U()[ < on C
2
. (This is possible by the
continuity of U at
0
.) Let C
1
be its complement in D. Let U
i
, i = 1, 2, equal U on C
i
and
zero elsewhere. Then U = U
1
+U
2
.
P
U
1
is harmonic away from C
1
; in particular it is harmonic (and continuous) in a neigh-
borhood of e
i
0
. Moreover, by the center expression in Theorem 26.6, we can verify that P
U
1
is zero on C
2
. On the other hand, P
U
2
satises [P
U
2
[ < since [U
2
[ < . Adding up P
U
1
and
P
U
2
we see that [P
U
[ can be made arbitrarily small in a neighborhood of e
i
0
.
Therefore, there is a 1-1 correspondence between continuous functions on the unit circle and
continuous functions on the closed unit disk that are harmonic on the open unit disk. The
correspondence is given by U P
U
, and its inverse map is just the restriction P
U
[
D
.
27.2. Schwarz reection principle.
Goal: Extend/continue an analytic function f : C to a larger domain. The ultimate
goal is to nd the maximal domain on which f can be dened.
Observation: If f(z) is analytic on , then f(z) is analytic on

= z[z .
If f(z) is an analytic function, dened on a region which is symmetric about the x-axis,
and f(z) = f(z), then f(z) is real on the x-axis. We have the following converse:
Theorem 27.2. Let be a symmetric region about the x-axis and let
+
= Im z > 0,
= Im z = 0. If f(z) is continuous on
+
, analytic on
+
, and real for all z ,
then f(z) has an analytic extension to all of such that f(z) = f(z).
Theorem 27.2 follows from the following result for harmonic functions:
NOTES FOR MATH 520: COMPLEX ANALYSIS 63
Theorem 27.3. Suppose v(z) is continuous on
+
, harmonic on
+
, and zero on .
Then v has a harmonic extension to satisfying v(z) = v(z).
Proof. Dene V (z) to be v(z) for z
+
, 0 for z , and v(z) for z

= Im z < 0.
We want to prove that V (z) is harmonic. For each z
0
, take an open disk D

(z
0
) .
Then dene P
V
to be the Poisson integral of V with respect to the boundary D

(z
0
). By
Theorem 27.1 above, P
V
is harmonic on D

(z
0
) and continuous on D

(z
0
). It will be shown
that V = P
V
.
On the upper half disk, V and P
V
are both harmonic, so V P
V
is harmonic. V P
V
= 0
on the upper semicircle, since V (z) = v(z) by denition and P
V
(z) = v(z) by the continuity
of P
V
(here z is on the semicircle). Also, V P
V
= 0 on D

(z
0
), since v(z) = 0 by
denition and P
V
(z) =
1
2
_
||=

2
|z|
2
|z|
2
V ()d, and we note that the contributions from the
upper semicircle cancel those from the lower semicircle.
Summarizing, V P
V
is harmonic on the upper half disk D

(z
0
)
+
, continuous on its
closure, and zero on its boundary. Therefore, V = P
V
on the upper half disk.
Proof of Theorem 27.2. Given f(z) = u(z) + iv(z) on
+
, extend f(z) to

by dening
f(z) = f(z) = u(z) iv(z) for z

. From above, v(z) is extended to a harmonic function

V (z) on all of as above. Since u(z) is the harmonic conjugate of v(z) on
+
, we dene
U(z) to be a harmonic conjugate of V (z) (at least in a neighborhood D

(z
0
) of z
0
).
Adjust U(z) (by adding a constant) so that U(z) = u(z) on the upper half disk.
We prove that g(z) = U(z) U(z) = 0 on D

(z
0
). Indeed, U(z) = U(z) on , so
g
x
= 0
on . Also,
g
y
= 2
u
y
= 2
v
x
= 0 on . Therefore, the analytic function
g
x
i
g
y
vanishes
on the real axis, and hence is constant. Since g(z) = 0 on , g(z) is identically zero. This
implies that U(z) = U(z) on all of D

(z
0
), hence proving the theorem.
I want to emphasize that the symmetry about the x-axis is simply a normalization, and that
the reection principle is applicable in far greater generality.
64 KO HONDA
28. Normal families, Arzela-Ascoli
28.1. Normal families. Let T be a collection (called family) of functions f : C.
(Much of what we discuss hold for functions f : X, where X is a metric space, but
well keep things simple.)
Denition 28.1. T is equicontinuous on E if > 0 > 0 such that [z z

[ < ,
z, z

E [f(z) f(z

)[ < for any f T.

Basically we are asking for all f T to be simultaneously uniformly continuous with the
same constants.
Denition 28.2. T is normal if every sequence f
1
, f
2
, . . . in T has a subsequence which
converges uniformly on compact subsets.
Remark: The limit f of a convergent sequence in T does not need to be in T. (In other
words, the closure T must be compact.)
28.2. Arzela-Ascoli Theorem.
Theorem 28.3 (Arzela-Ascoli). A family T of continuous functions f : C is normal i
(i) T is equicontinuous on each compact E and (ii) for any z the set f(z)[f T
is compact.
Roughly speaking, if the functions are f : R R with usual coordinates y = f(x) and were
imagining the graph of f, then (i) controls the vertical spread in values of a single function
f[
E
(the vertical spread is uniform for all f T) and (ii) controls the vertical spread at a
single point z E as we vary f.
Proof. Suppose T is normal. We prove (i). If T is not equicontinuous on a compact subset E,
then there are sequences z
n
, z

n
E and f
n
T such that [z
n
z

n
[ 0 but [f
n
(z
n
)f
n
(z

n
)[
. Since E is compact, there is a subsequence (we abuse notation and also call it z
n
) such that
z
n
z E. Also, since z
n
and z

n
are close, z

n
z E. (We still have [f
n
(z
n
)f
n
(z

n
)[ .)
On the other hand, since T is normal, there is a subsequence (still called f
n
) f
n
T which
converges to a continuous function f on E. (Why is f continuous?) But then
[f
n
(z
n
) f
n
(z

n
)[ [f
n
(z
n
) f(z
n
)[ +[f(z
n
) f(z

n
)[ +[f(z

n
) f
n
(z

n
)[,
and the rst and third terms on the RHS 0 by uniform convergence, and the second term
0 as z
n
and z

n
get arbitrarily close (by continuity of f). This contradicts [f
n
(z
n
)f
n
(z

n
)[
.
Next we prove (ii). Any sequence f
n
has a convergent subsequence (which we still call
f
n
), so given any z , f
n
(z) f(z). Since any sequence in f(z)[f T has a convergent
subsequence, the set is compact.
Now suppose (i) and (ii) hold. Given f
1
, . . . , f
n
, T, take a sequence
1
,
2
, . . .
which is everywhere dense in (e.g., the set of points with rational coordinates). Take a
subsequence of f
n
which converges at
1
. Denote it by indices n
11
< n
12
< < n
1j
< . . . .
Next, take a subsequence of it which converges at
2
. Denote it by indices n
21
< n
22
< <
NOTES FOR MATH 520: COMPLEX ANALYSIS 65
n
2j
< . . . . Continuing in this manner, we then take the diagonal n
11
< n
22
< n
33
< . . . .
Then f
n
jj
converges at all the points
i
.
We now claim that f
n
jj
converges uniformly on any compact E . Indeed, given any
z E,
[f
n
ii
(z) f
n
jj
(z)[ [f
n
ii
(z) f
n
ii
(
k
)[ +[f
n
ii
(
k
) f
n
jj
(
k
)[ +[f
n
jj
(
k
) f
n
jj
(z)[,
where
k
is within a distance of z. Since f
n
ii
and f
n
jj
are equicontinuous, given > 0,
there is > 0 so that the rst and last terms on the RHS are <

3
. Since f
n
ii
(
k
) converges,
there is an N such that n
ii
> N implies that the middle term is <

3
.
28.3. Montels theorem. We now apply the Arzela-Ascoli theorem in the setting of a
family T of analytic functions.
Theorem 28.4 (Montel). A family T of analytic functions f : C is normal i functions
f T are uniformly bounded on each compact set E , i.e., there exists a constant M
such that [f(z)[ < M for all z E and f T.
Proof. Suppose T is normal. Cover E with a nite number of disks D
i
of radius (this is
possible by compactness). On each disk D
i
centered at z
i
, if f

, f

T and z D
i
we have:
[f

(z) f

(z
i
)[ [f

(z) f

(z
i
)[ +[f

(z
i
) f

(z
i
)[,
and rst term on the RHS is bounded above by equicontinuity and the second term is
bounded since f(z
i
)[f T is compact. Since there are only nitely many balls, we have
uniform boundedness.
Now suppose T is uniformly bounded on E. It is sucient to prove equicontinuity, in
view of Arzela-Ascoli. If D
r
(z
0
) is a disk of radius r about z
0
and z, z

E D
r/2
(z
0
), then
f(z) f(z

) =
1
2i
_
D
i
_
f()
z

f()
z

_
d =
z z

2i
_
D
1
( z)( z

)
f()d,
and
[f(z) f(z

)[
[z z

[
2
2r
(r/2)(r/2)
M
4M
r
[z z
0
[.
This proves equicontinuity on D
r/2
(z
0
).
Now, cover E with nitely many disks of radius
r
2
. If z, z

E and [z z

[ <
r
4
, then there
is some disk D
r/2
(z
0
) which contains both z and z

. Given > 0, pick = min(

r
4
,
r
4M
).
Then [z z

[ < [f(z) f(z

)[ < .
66 KO HONDA
29. Riemann mapping theorem
Theorem 29.1. A simply connected, connected, open C is biholomorphic to the open
unit disk D.
Step 1: Let a C . Since simply connected, there is a single-valued branch of
f(z) =

z a on . (Recall that f(z) = e

1
2
log(za)
and log(z a) can be dened as an
antiderivative of
1
za
. This is because, according to Cauchys Theorem,
_

dz
za
= 0 for all
closed curves in a simply connected .) f gives a biholomorphism of onto its image. For
simplicity take a = 0. Then consider D

(b) Im f. We have D

(b) Im f = . (Observe
that D

(b) is D

(b) which has been reected across the origin. This means that they would
have the same image under the map w w
2
.) Now take an FLT which maps D

(b) to the
complement of the open unit disk. From now on we assume that is a subset of D, and also
that 0 . (Recall there is an FLT which is an automorphism of D and takes any interior
point of D to 0.)
Step 2: Consider a holomorphic map f : D which satises:
(1) f is 1-1,
(2) f(0) = 0,
(3) f

(0) > 0 (this means f

(0) is real and positive).

Let T be the family of all such holomorphic maps. In particular, id is one such map, so T
is nonempty.
If f is not onto, then we nd h T with larger h

(0). Let a Df(). We take g to be

the composition g
3
g
2
g
1
of the following maps:
(1) FLT g
1
, an automorphism of D which sends a to 0,
(2) g
2
(z) =

z, which is single-valued on g
1
f() (since it is simply connected),
(3) FLT g
3
, an automorphism of D which sends g
2
g
1
(0) to 0 and satises (g
3
g
2
g
1
)

(0) > 0.
(The latter condition can be achieved by composition with an appropriate rotation
about the origin.)
We claim that (g f)

(0) > f

(0). Indeed g
1
can be viewed as a map from D to itself which
sends 0 0 such that (g
1
)

(0) > 0. By the Schwarz lemma, (g

1
)

(0) < 1 since g

1
is not
a rotation. This means that g

(0) > 1, and hence (g f)

(0) > f

(0).
Step 3: Now consider T. T is uniformly bounded (since all f map to D), so take a sequence
f
1
, . . . , f
n
, . . . in T for which f

i
(0) M = sup
fF
f

(0). Then there is a subsequence of

f
n
converging UCOCS to a holomorphic map f : D by Montels theorem. (This in
particular implies that M < .)
It remains to show that f is 1-1. We argue using the argument principle. By uniform
convergence,
_

i
(z)
f
i
(z)
dz
_

(z)
f(z)
dz,
NOTES FOR MATH 520: COMPLEX ANALYSIS 67
for closed curves that avoid the zeros of f. But these represent winding numbers this
implies that the winding numbers do not change in the limit, and that f is 1-1. Hence f T.
By Step 2, f must be onto D. This completes the proof of the Riemann Mapping Theorem.
Enhancements:
1. Let and

be open regions. If f :

is a homeomorphism and z
n
is a sequence
that tends to , then f(z
n
) tends to

. (Proof left as exercise.)

We dene = int() (the closure of minus (the interior of) ). We also say that
z
n
tends to if for all z , D

(z) and N > 0 such that z

n
, D

(z) if n > N.
Observe that if K is compact, then by covering K with open disks, there exists N > 0
such that z
n
, K if n > N.
2. The Uniformization Theorem is a generalization of the Riemann Mapping Theorem which
says that a simply connected (=any closed curve can be contracted to a point), connected
Riemann surface (without boundary) is biholomorphic to one of C, S
2
, or D.
3. A multiply connected region C of connectivity n + 1 (recall that this means that
C has n + 1 connected components) is biholomorphic to the annulus 1 < [z[ < (for
some > 1) with n 1 arcs, each of which is a subarc of [z[ =
i
for some 1 <
i
< ,
removed.
68 KO HONDA
30. Analytic continuation
30.1. Riemann mapping theorem and boundary behavior. Recall that last time we
proved that a simply connected C is biholomorphic to the open unit disk D. We give
more enhancements:
4. If contains a free one-sided analytic arc , then the biholomorphism f :

D has
an analytic extension to , and is mapped onto an arc of D.
An analytic arc : [a, b] is an arc which, in a neighborhood of each t
0
[a, b], is given
by a Taylor series
(t) = a
0
+a
1
(t t
0
) +a
2
(t t
0
)
2
+ . . . ,
with a nonzero radius of convergence.
A free one-sided boundary arc is a regular (i.e.,

(t) ,= 0 on [a, b]), simple (i.e., 1-1) arc

such that there are neighborhoods C of t
0
[a, b] and

C of (t
0
) such that
Im z > 0 gets mapped onto

and Im z < 0 gets mapped to C.

Proof. Use the Schwarz reection principle, after changing coordinates so that f looks like
a map from a subset of the upper half plane to itself.
5. If is simply connected and D is given by a simple (continuous) closed curve, then the
biholomorphism f : D extends to a homeomorphism f : D.
The proof is omitted.
30.2. Analytic continuation. Denote by (f, ) an analytic function f dened on an open
region . (We assume, as usual, that is connected.)
Denition 30.1. (f
1
,
1
) and (f
2
,
2
) are direct analytic continuations of each other, if

1

2
,= and f
1
= f
2
on
1

2
.
Observe that if (f
2
,
2
) and (g
2
,
2
) are direct analytic continuations of (f
1
,
1
) to
2
, then
f
2
= g
2
, since f
2
[

2
= g
2
[

2
. (Recall that if two analytic functions with the same
domain agree on a set with an accumulation point, then they are identical.)
Denition 30.2. If there exists a sequence (f
1
,
1
), . . . , (f
n
,
n
) such that (f
i+1
,
i+1
) is a
direct analytic continuation of (f
i
,
i
), then (f
n
,
n
) is an analytic continuation of (f
1
,
1
).
Remark: It is possible that
1
=
n
but f
1
,= f
n
.
Example: Consider f(z) = log z. Dene
j
=
j
2
< <
j
2
+ . Then on
j
, we dene
f
j
(z) = log [z[ +i arg(z), where
j
2
< arg(z) <
j
2
+. We have
0
=
4
, but f
4
= f
0
+ 2i.
30.3. Germs and sheaves. Now consider a pair (f, ), where C and f is analytic in
a neighborhood of . We view (f
1
,
1
) and (f
2
,
2
) as equivalent i
1
=
2
and f
1
= f
2
on
NOTES FOR MATH 520: COMPLEX ANALYSIS 69
some neighborhood of
1
=
2
. Such an equivalence class is called a germ of a holomorphic
function. Notice that (f, ) gives rise to a germ (f, ) at each .
If D is an open set in C, then the set of all germs of holomorphic functions (f, ) with
D is called the sheaf of germs of holomorphic functions over D, and will be denoted T
D
or T, if D is understood. There is a projection map : T D, which sends (f, ) .

1
() is called the stalk at , and is also denoted T

.
Theorem 30.3. The set T can be given the structure of a Hausdor topological space such
that : T D becomes a local homeomorphism (i.e., for each (f, ) T there is an open
neighborhood U whose image (U) is open and homeomorphic to U).
Proof. A set V T is open i for every (f, ) V there exists a function element (f, )
with (which restricts to (f, )) such that all (f,

) V for

. (Verify that this

satises the axioms of a topology!)
Now, given (f, ) T, take some function element (f, ) which restricts to (f, ). (Here
were assuming that .) Then take V to be the set (f,

)[

of restrictions of f
to all points in . It is clear that maps V homeomorphically onto .
It remains to show that T is Hausdor. Given (f
1
,
1
) and (f
2
,
2
), if
1
,=
2
, then thats
easy. (Take open sets
1
and
2
about
1
and
2
, respectively, on which f
1
and f
2
can be
dened, and (f
1
,
1
) and (f
2
,
2
) give rise to disjoint open sets, as in the previous paragraph.)
Now suppose
1
=
2
. Let be an open set containing
1
=
2
on which f
1
and f
2
are both
dened. If (f
1
,

) = (f
2
,

) for any

, then f
1
= f
2
on all of . Therefore the open sets
V
1
and V
2
corresponding to (f
1
, ) and (f
2
, ) do not intersect.
Remark: : T D is not quite a covering space, if you know what that means. This
is because not every component of
1
U (for an open set U D) is homeomorphic to U.
(Why?)
70 KO HONDA
31. Analytic continuation
Last time we dened T, the sheaf of germs of holomorphic functions, and showed that
it can be given the structure of a Hausdor topological space such that the projection
: T C (today, the base space is C instead of D) is a local homeomorphism. A basis for
the topology of T is the set of U
(f,)
= (f, )[ , where f is an analytic function on an
open set C.
31.1. Riemann surface of a function. Given a function element (f, ), take its corre-
sponding open set U
(f,)
in T, and the connected component of T which contains U
(f,)
.
Call this
(f,)
, or simply , if (f, ) is understood.
Claim. can be given the structure of a Riemann surface, where the holomorphic coordinate
charts are given by the local homeomorphism : C.
will be called the Riemann surface of (f, ). Note that is the set of all (g,

) for which
there is an analytic continuation from (f, ) to (g,

) with

. (Eectively, we have
pasted together such (g,

) to obtain .)
There also is a holomorphic map (often called a global analytic function) f : C obtained
by setting (f, ) f(). (Verify that this is holomorphic!) We refer to as the Riemann
surface of f and write =
f
.
Remark:
f
is Hausdor since T is. Observe that we havent shown that
f
is second
countable. (See the denition of a Riemann surface.) The verication is not trivial but
wont be done here.
Example: Riemann surface of f(z) = log z. Above each point of C0, there are innitely
many sheets, corresponding to f
i
(z) = log [z[ +i arg(z), where
j
2
< arg(z) <
j
2
+. As we
move around the origin, we can move from one sheet to another. Observe that an alternate
way of obtaining
f
is to start with
j
given above, and glue
j
to
j+1
along the region
where f
j
and f
j+1
agree.
Example: Riemann surface of f(z) =

z. Above each point of C0 there are two sheets
corresponding to re
i

re
i/2
. Notice that as we continue one choice of

z around the
origin, we reach the other choice, and circling twice around the origin gives the original
function element.
Remark: We havent dealt with the branch points, e.g., z = 0 for f(z) =

z. There is a
reasonable way to ll in the point z = 0 to give a genuine Riemann surface (one without any
singularities). Note that the usual picture of a Riemann surface of f(z) =

z is rather
misleading, since it exhibits what looks like a singular point at the origin.
31.2. Analytic continuation along arcs and the monodromy theorem. Let : [a, b]
C be a continuous arc. Consider a connected component
f
of T. An arc : [a, b]
f
NOTES FOR MATH 520: COMPLEX ANALYSIS 71
is an analytic continuation of the global analytic function f along , if = . (This is
called a lift of to
f
in topological jargon.)
Lemma 31.1. Two analytic continuations
0
and
1
of a global analytic function f along
are either identical, or
0
(t) ,=
1
(t) for all t.
Proof. It follows from the Hausdor/local homeomorphism property that the set of t for
which
0
(t) =
1
(t) is open and the set for which
0
(t) ,=
1
(t) is also open.
Let be a region in C. Suppose f is a global analytic function which can be continued
along all continuous arcs in and starting at any (f, )
f
. Let
0
and
1
be two
continuous arcs in which are homotopic in relative to their endpoints, i.e., there is a
continuous map : [a, b] [0, 1] so that:
(1) (x, 0) =
0
(x) and (x, 1) =
1
(x),
(2) (a, t) and (b, t) do not depend on t.
Then we have the following:
Theorem 31.2 (Monodromy theorem). Suppose
0
and
1
are homotopic relative to their
endpoints and f is as in the preceding paragraph. Given a germ of f at the common initial
point of
0
and
1
, their continuations along
0
and
1
lead to the same germ at the common
terminal point.
Proof. It suces to prove this theorem when the homotopy
t
(x)
def
= (x, t) ranges over a
small subinterval t [t
0
, t
0
+] [0, 1]. We can therefore subdivide [a, b] into a = x
0
< x
1
<
x
2
< < b = x
k
so that the germ of f at each
t
0
(x
i
) is dened in a neighborhood U
i
, and

t
(x) U
i
for all t [t
0
, t
0
+ ] and x [x
i
, x
i+1
]. By using an argument similar to that of
Lemma 31.1, were done.
Corollary 31.3. If f is a global analytic function which can be continued along all arcs in
a simply connected , then f is a single-valued analytic function.
This gives another proof of the fact that f(z) =

z a admits a single-valued analytic

branch on a simply-connected , if a , .
72 KO HONDA
32. Universal covers and the little Picard theorem
32.1. Universal cover of C0, 1.
Denition 32.1. A covering space of X is a topological space

X and a map :

X X such
that the following holds: there is an open cover U

of X such that
1
(U

) is a disjoint
union of open sets, each of which is homeomorphic to U

via .
Denition 32.2. A universal cover :

X X is a covering space of X which is simply
connected.
Theorem 32.3. Any reasonable space X has a universal cover

X. Moreover, the universal
cover is unique, in the sense that given any other

:

X

X there exists a homeomorphism

:

X

X

such that

= .
As a set, the universal cover

X is given as follows: Pick a basepoint x
0
X. Then

X is
the quotient of the set of paths with initial point x
0
, by the equivalence relation which
identies and

if they have the same endpoints and are homotopic arcs relative to their
endpoints.
Well presently construct the universal cover of C0, 1. Let = 0 < Re z < 1, [z
1
2
[ >
1
2
, Im z > 0 be a region of the upper half plane H. has three boundary components
C
1
= Re z = 0, Im z > 0, C
2
= [z
1
2
[ =
1
2
, Im z > 0, and C
3
= Re z = 1, Im z > 0
(together with points 0, 1). Using the Riemann mapping theorem (and its enhancements),
there exists a biholomorphic map : H, which extends to a map from C
1
to Im z =
0, Re z < 0, C
2
to Im z = 0, 0 < Re z < 1, and C
3
to Im z = 0, 1 < Re z. Using the
Schwarz reection principle, we can reect along each of the C
j
. For example, if we reect
across C
1
, goes to

= 1 < Re z < 0, [z +
1
2
[ >
1
2
, Im z > 0, and extends to a
holomorphic map on

C
1
, where

is mapped to the lower half plane. Continuing

in this manner, we dene a map : H C 0, 1. It is not hard to verify that this is a
covering map. Since H D is simply-connected, is the universal covering map.
32.2. Little Picard Theorem.
Theorem 32.4. Let f(z) be a nonconstant entire function. Then C Im f consists of at
most one point.
Proof. Suppose f : C C misses at least two points. By composing with some fractional
linear transformation, we may assume that f misses 0 and 1.
We now construct a lift

f : C H of f to the universal cover, i.e., nd a holomorphic
map

f satisfying

f = f. [This is called the homotopy lifting property, and is always
satised if the domain is a reasonable simply connected topological space X.]
Given z
0
C, there is a neighborhood V f(z
0
) and W H so that W

V is a
biholomorphism. By composing with
1
, we can dene

f : U W H, where U is a
suciently small neighborhood of z
0
. Take some lift

f in a neighborhood of a reference point
z = 0. Let : [0, 1] C be a continuous arc in C with (0) = 0, (1) = z. There exist
NOTES FOR MATH 520: COMPLEX ANALYSIS 73
0 = t
0
< t
1
< < t
n
= 1 such that each [t
i1
, t
i
] is mapped into V
i
C 0, 1, where

1
(V
i
) is the disjoint union of biholomorphic copies of V
i
(by the covering space property).
Suppose we have extended

f along up to t
i
. The pick the component W of
1
(V
i+1
) which
contains

f((t
i
)). Continue the lift to [t
i
, t
i+1
] by composing f with
1
: V
i+1

W.
Since C is simply connected, the monodromy theorem tells us that the value of

f(z) does
not depend on the choice of path . This gives a holomorphic map

f : C H. Now,
composing with the biholomorphism H

D, we obtain a bounded entire function. Since
we know that a bounded entire function is a constant function, it follows that

f is constant.

Enhancements:
1. (Montel) Let C be an open set and T be a family of analytic maps C. If each
f T misses the same two points a, b, then T is normal.
2. (Big Picard) Suppose f is holomorphic on z
0
and has an essential singularity at
z
0
. If U is any (small) neighborhood of z
0
, then f assumes all points of C innitely
many times in U z
0
, with the possible exception of one point. [Big Picard implies Little
Picard.]