You are on page 1of 7

MATH 566 LECTURE NOTES 5: THE RIEMANN MAPPING THEOREM

TSOGTGEREL GANTUMUR
1. Introduction
We start with a denition.
Denition 1. A mapping : is called biholomorphic onto , or conformal, if O()
and is invertible with
1
O(). If there is a conformal mapping between and , then
they are said to be conformally equivalent domains.
Biholomorphic mappings are important because they can be used to plant complex analy-
sis of one domain onto another domain: in the setting of the preceding denition, if f O()
then f O(), and if g O() then g
1
O(). Conformal equivalence is an
equivalence relation on the space of all domains (or more generally, Riemann surfaces), and
so in principle it suces to study one representative from each of those equivalence classes
(called conformal classes). Therefore, identifying conformal classes is a fundamental problem
in complex analysis. The Riemann mapping theorem characterizes the conformal class of the
unit disk D = z C : [z[ < 1, which turns out to be the collection of all simply connected
domains (except C itself). In the following, we discuss the Riemann mapping theorem and
some other topics related to it.
2. Mapping properties of holomorphic functions
Let f be a nonconstant holomorphic function on a neighbourhood of 0 C, with f(0) = 0.
Obviously we can write f(z) = z
n
g(z) with g(0) ,= 0, and n 1. Moreover, there is > 0
so small that the both functions f and f

are nowhere vanishing in the punctured closed disk


D

0. Let w C

be such that [w[ < min


|z|=
[f(z)[. Then, by Rouches theorem, the function
z f(z) w has exactly n zeroes in D

. Furthermore, those zeroes must be in D

because
w ,= 0, and since f

does not vanish in D

, the zeroes must be simple. We conclude that


each point in the disk D
R
of radius R = min
|z|=
[f(z)[ has exactly n distinct points in D

as its
preimage set under f. Since f is continuous, f
1
(D
R
) D

is an open neighbourhood of 0,
so f is n-to-1 on a neighbourhood of 0. In particular, recalling that n = 1 is equivalent to
f

(0) ,= 0, local injectivity implies nonvanishing of the rst derivative and vice versa. This
substantially claries local mapping properties of holomorphic functions.
Theorem 2. Let f be a holomorphic function on a neighbourhood of c C.
(a) If f is injective on a neighbourhood of c, then f

(c) ,= 0.
(b) Conversely, if f

(c) ,= 0, then f is injective on a neighbourhood of c.


(c) More generally, if f is not constant, then on a neighbourhood of c, it can be written as
f(z) = f(c) + (z)
n
,
where is a holomorphic injection and n = Ord(f

, c) + 1.
Date: December 3, 2010.
1
2 TSOGTGEREL GANTUMUR
Proof. We may assume, without loss of generality, that c = 0 and f(0) = 0. Claims (a) and
(b) have been proven above. In the notations of the paragraph preceding this theorem, since
g is nonvanishing on a neighbourhood of 0, there is a holomorphic such that (z)
n
= g(z)
on a neighbourhood of 0. Then with (z) = z(z), we have

(0) = (0) ,= 0.
Theorem 3 (Inverse function theorem). With some disk D, let f O(D) be injective, and
let = f(D). Then the inverse function f
1
: D is holomorphic, and we have
f
1
(w) =
1
2i

D
zf

(z)dz
f(z) w
, w .
In particular, if f O() is injective on a domain , then its inverse f
1
is holomorphic.
Proof. Let D, and let w = f(). Since f is injective, z f(z) w has one simple zero at
z = in D. We apply the generalized argument principle with g(z) = z, to obtain
=
1
2i

D
zf

(z)dz
f(z) w
,
establishing the formula. By holomoprhy of integral then we get the holomorphy of f
1
.
For the second part of the theorem, for each w f(), we apply the rst part of the theorem
to a disk centred at f
1
(w) whose closure lies in , to conclude that f
1
is holomorphic on
a neighbourhood of w.
Holomorphic mappings generally preserve angles and orientations, in the sense we discuss
below. By denition, a curve : [1, 1] C is dierentiable at the parameter value 0 if there
exists C such that
() = (0) + + o([[),
and in this situation is called the tangent of at (0). Now if is a holomorphic function
on a neighbourhood of (0), then there is C such that
(z + h) = (z) + h + o([h[),
and moreover is a curve passing through and dierentiable at ((0)). By combining
the preceding two denitions we can write
(()) = ((0)) + + o(),
revealing that the tangent of at ((0)) is . In particular, if

((0)) ,= 0, or
equivalently if is injective on a neighbourhood of (0), then under all tangents are rotated
by the same angle, so the angle between any two curves intersecting at (0), together with its
orientation, is preserved under . This is almost the general picture, since the derivative

is holomorphic and therefore its zeroes form a discrete set, provided that is not constant.
What happens at the zeroes of

is also simple to understand. We know from Theorem


2(c) that at the zeroes of

of order m, is equal to a holomorphic injection followed by


the polynomial map z z
m+1
, hence the angles are simply multiplied by m + 1. Let us
summarize these observations into the following informal remark, where we call the zeroes of

the critical points of .


Remark 4. Near any of its non-critical points, a holomorphic mapping is one-to-one, and angle-
and orientation preserving. Near a critical point, the mapping is n-to-1 for some integer n
(that may vary from critical point to critical point), and at the critical point itself, the angles
are multiplied by n. The critical points form a discrete set.
THE RIEMANN MAPPING THEOREM 3
3. Automorphism groups
Denition 5. Given a connected open set , its automorphism group, denoted by Aut , is
the set of biholomorphic self-maps of .
One can easily see that ane maps, that are the maps of the form z az +b with a C

and b C, are in the automorphism group of the complex plane C. In fact, those are the only
maps in Aut C.
Lemma 6. Aut C = z az + b : a C

, b C.
Proof. Let Aut C. Then (D) is an open set, and by injectivity, near , does not take
any values from (D). So it follows from the Casorati-Weierstrass theorem that

C is not
an essential singularity, which means that is a polynomial. But since

has no zero in C,
we have

= const ,= 0, which concludes the proof.


Let us turn to the automorphisms of

C. Given a 2 2 complex matrix A =

a b
c d

, dene
the function F
A
:

C

C by F
A
(z) =
az+b
cz+d
. Its derivative is given by F

A
(z) =
adbc
(cz+d)
2
, so a
necessary condition for F
A
to be locally injective is ad bc det A ,= 0. Such functions F
A
with det A ,= 0 are called fractional linear transformations. If c = 0 (and so ad ,= 0) these
reduce to ane maps, which now can be thought of as fractional linear transformations that
leave

C xed. In general, it holds that F
A
() = a/c and F
A
(d/c) = . Moreover,
we have F
A
(1/w) = F
A
(w) where A

is the matrix obtained from A by interchanging its


columns, and 1/F
A
(z) = F
A
(z) where A

is the matrix obtained from A by interchanging


its rows. Since det A

= det A

= det A, it follows that F


A
is locally injective everywhere
in

C. One can easily check that the inverse of F
A
is equal to F
A
1, thus implying that
F
A
Aut

C. In fact, recalling that GL(2, C) = A C
22
: det A ,= 0, the mapping
A F
A
is a group homomorphism between GL(2, C) and Aut

C with the kernel given by
nonzero complex multiples of the identity matrix. In other words, we have F
AB
= F
A
F
B
and F
sA
= F
A
for s C

. The following lemma says that this homomorphism is surjective,


i.e., the automorphisms of

C are exactly the fractional linear transformations.
Lemma 7. Aut

C = F
A
: A GL(2, C).
Proof. Let Aut

C. So must be a rational function, which can be written in the form
(z) =
a (z
1
) (z
n
)
(z
1
) (z
m
)
,
with
j
,=
k
for all j, k. Then we have (
j
) = 0 for all j, and (
k
) = for all k. Since
is invertible, this forces n = 1 and m = 1.
One useful property of fractional linear transformations is that one can freely specify the
images of any three distinct points. Namely, if , , are three distinct points of

C, then
(z) =
(z )( )
(z )( )
,
maps to 0, to 1, and to . Now by composing this with the inverse of another such
mapping, the triple (, , ) can be mapped to any triple (

) as long as

are
distinct.
4 TSOGTGEREL GANTUMUR
4. Schwarzs lemma
Lemma 8 (Schwarz). Let : D D be holomorphic with (0) = 0. Then we have [(z)[ [z[
for z D and

(0) 1. Moreover, if either [(z)[ = [z[ for some z ,= 0 or [

(0)[ = 1, then
(z) = z for some D.
Proof. This is a direct application of the maximum principle. Let
g(z) =

(z)/z, if z ,= 0,

(0), if z = 0.
Then we have g O(D), and [g[
1
1
on D
1
for > 0. The maximum principle gives
[g[
1
1
in D
1
, and letting 0, we infer [g[ 1 in D. This is the rst claim of the
lemma. In terms of g, the hypothesis of the second claim is simply that [g(z)[ = 1 for some
z D. But this implies by the maximum principle that g = const = with [[ = 1, which
establishes the second claim.
As a nontrivial application of Schwarzs lemma, below we characterize the automorphisms of
the unit disk. Our initial candidates will be a special class of fractional linear transformations,
namely, maps of the form
B

(z) =
z
1 z
, with [[ < 1.
These maps are called the Blaschke factors. Since [1z [ > 0 for z D, we have B

O(D).
Moreover, from that
[B

(z)[ =

z
1 z

z
z
1 z

1 z
1 z

= 1, for z D,
and that B

is not constant, we infer by the maximum principle that [B

(z)[ < 1 for z D.


Clearly B

has the inverse B


1

= B

on D, hence B

Aut D. Let us also make a note of


the simple properties B

() = 0 and B

(0) = .
Lemma 9. Aut D = z B

(z) : D, D.
Proof. First we prove that an automorphism of D that leaves 0 xed must be a rigid rotation
around 0. Let Aut D with (0) = 0. Then Schwarzs lemma applied to and to

1
implies that [(z)[ [z[ and [
1
(z)[ [z[ for z D. The latter can be written
as [z[ = [
1
((z))[ [(z)[ for z D, which, combined with the former, implies that
[(z)[ = [z[ for z D. Now yet another application of Schwarzs lemma guarantees that
(z) = z for some D.
For general Aut D, we consider the map = B

with =
1
(0). Then we have
Aut D and (0) = (B

(0)) = () = 0, and so by the preceding paragraph is a


rotation around 0. The proof is complete upon recalling = B
1

= B

.
5. The Riemann mapping theorem
When is an open set C conformally equivalent to the unit disk D? One can exclude the
case = C right away because by Liouvilles theorem, any holomorphic function : C D
must be constant. Also, since biholomorphic mappings are in particular homeomorphisms,
must be homeomorphic to D, so a necessary condition is that must be simply connected.
The all-important Riemann mapping theorem says that these conditions are also sucient.
Namely, any simply connected, open proper subset of C is conformally equivalent to D. Before
stating the Riemann mapping theorem, we introduce a denition.
Denition 10. A connected open set C is said to have the square-root property, if for
any zero-free function f O(), there exists g O() such that g(z)
2
= f(z) for z .
THE RIEMANN MAPPING THEOREM 5
We have seen that if is simply connected, then it has the above-dened property.
Theorem 11. Let C have the square-root property, and let . Then there is a
unique biholomorphic mapping : D such that () = 0 and

() > 0.
Proof. We explain here the main structure of the proof, deferring certain specic results to
the three subsections below. Let us dene the set
= : D injective, holomorphic, and () = 0.
That this set is nonempty is proven in 5.1. Then we x a point with ,= , and look
for maximizers of [()[ among . Let B = sup[()[ : , and let
n
be a
sequence such that [
n
()[ B as n . Since the functions
n
map into D, we have
|
n
|

1 for all n. By Montels theorem that is proven in 5.2 below,


n
has a subsequence
that converges locally uniformly on . Let us denote this limit function by : C, which
by construction satises () = B. The Weierstrass convergence theorem guarantees that
is holomorphic, and the Hurwitz injection theorem implies that is injective, and moreover
that the range is in D, i.e., that () D. Furthermore, by the inverse function theorem,

1
: () is holomorphic. We prove that () = D in 5.3, so that the existence of a
biholomorphism : D is established. The uniqueness part of the theorem is demonstrated
in the lemma that follows.
Lemma 12 (Poincare). Let C be an open set, and let . Assume that : D
and : D are biholomorphic mappings onto D such that () = () = 0 and that both

() and

() are positive real numbers. Then = .


Proof. Since
1
Aut D and (
1
(0)) = 0, we have
1
is a rotation around 0. In
other words, (z) = (z) with some D. Therefore

() =

(), and so = 1.
5.1. Koebes square-root trick. Recall from the proof of Theorem 11 that is the set of
holomorphic injections : D that satisfy () = 0. In this subsection we establish the
nonemptiness of , by explicitly constructing an element of .
The complexity of the construction depends on . If is bounded, a translation (that
sends to 0) followed by a dilation will do. If is unbounded and if there is an open disk
D
r
(c) C , then the transformation z
1
zc
will render bounded, and so the problem
reduces to the previous case.
In general, since ,= C there is c C . Let f(z) = z c and let g O() be such that
g()
2
= f(). The existence of g is guaranteed by the square-root property of . We have
g(z
1
) = g(z
2
) z
1
c = g(z
1
)
2
= g(z
2
)
2
= z
2
c z
1
= z
2
,
meaning that g is injective. On the other hand, the same argument gives
g(z
1
) = g(z
2
) z
1
= z
2
,
meaning that if 0 ,= w g(), then w , g(). Now g() is an open set, so there exists an
open disk D g() with D , 0. Then the disk D = w : w D is in the complement of
g(), i.e., D C g(), which reduces the problem to the previous case. Let us state what
we have proved here as a lemma.
Lemma 13. Let C be a connected, open set having the square-root property, and let
. Then there is a holomorphic injection : D that satises () = 0.
6 TSOGTGEREL GANTUMUR
5.2. Montels theorem. In this subsection, we discuss compactness properties of . First
we recall the celebrated Arzel`a-Ascoli theorem, in a form convenient for our purposes.
Theorem 14 (Arzel`a-Ascoli). Let C be open and connected, and let f
n
: C be a
sequence that is locally equicontinuous and locally equibounded. Then there is a subsequence
of f
n
that converges locally uniformly.
That the sequence f
n
is locally equibounded means that for any compact set K one
has sup
n
|f
n
|
K
< . Similarly, that the sequence f
n
is locally equicontinuous means that
for any compact set K the sequence f
n
is (uniformly) equicontinuous on K. It turns
out that if f
n
is a sequence of holomorphic functions, then the local equicontinuity condition
can be dropped from the Arzel`a-Ascoli theorem.
Theorem 15 (Montel). Let C be open and connected, and let f
n
O() be a locally
equibounded sequence. Then there is a subsequence of f
n
that converges locally uniformly.
Proof. In view of the Arzel`a-Ascoli theorem, it suces to show local equicontinuity of f
n
.
We will prove here that f
n
is equicontinuous on any closed disk D , and the general
case follows by a covering argument. Let D = D

(c) and D

= D
+
(c) be two concentric
disks such that D

and > 0. Then the Cauchy estimates give


[f

n
(z)[
1

max
D

(z)
[f
n
[, for z D, and so |f

n
|
D

1

|f
n
|
D
.
For z, w D, we have
[f
n
(z) f
n
(w)[ = [f

n
, [zw])[ [z w[ |f

n
|
[zw]
[z w[
1

|f
n
|
D
.
Since |f
n
|
D
is bounded uniformly in n, the sequence f
n
is equicontinuous on D.
5.3. Caratheodory-Koebe expansion map. Recall from the proof of Theorem 11 that
is the set of holomorphic injections : D that satisfy () = 0, and that is an
element of satisfying [()[ = sup[()[ : , where , and ,= . The goal of
this subsection is to show that : D is surjective.
Let = (). We say that a holomorphic injection : D is an expansion if (0) = 0
and [(z)[ > [z[ for any z 0. Suppose that there exists an expansion : D. Then
we have , and
[[ ]()[ = [(())[ > [()[,
so from the maximality of [()[ we conclude that there does not exist an expansion : D.
But we show that there does exist an expansion : D if ,= D, implying that = D.
Lemma 16. Let D be a connected, open set having the square-root property, with 0 .
Then there exists an expansion : D.
Proof. For D, let us dene

(z) =
z
z 1
= B

(z),
where B

Aut D is the Blaschke factor dened in 4. So we have

Aut D, and
1

.
We also dene the following maps sending D into D:
j(z) = z
2
, and

2 j

,
and so in particular we have

(0) =

2(

(0)
2
) =

2(
2
) = 0.
THE RIEMANN MAPPING THEOREM 7
Moreover, since j : D D is not an automorphism,

is not a rotation, hence by Schwarzs


lemma we have
[

(z)[ < [z[, for all z D 0.


We see that

is a contraction as opposed to an expansion, but in what follows we will


construct its inverse on which is an expansion.
Let D be such that
2
, . This is possible since D. Then

2 : D is
zero-free and satises

2(0) =
2
. By the square-root property, there is O() such that
(z)
2
=

2(z) for all z , or equivalently,

2 = j on , and that (0) = . Since

2() D, we have () D. Now we dene =

. Then () D, (0) = 0, and

2 j

2 j =

2 = id

,
where id

: is the identity mapping on . This implies that : D is injective,


and in particular that if z ,= 0 then (z) ,= 0. Finally, we infer
[z[ = [

((z))[ < [(z)[, for all z 0,


from the contraction property of

, which completes the proof.

You might also like