A Concise Course in Complex Analysis and Riemann Surfaces

A concise course in complex
analysis and Riemann surfaces

Wilhelm Schlag
Contents
Preface
Chapter 1. From i to z: the basics of complex analysis

1. The field of complex numbers
2. Differentiability and conformality
3. Mobius transforms
4. Integration
5. Harmonic functions
6. The winding number
7. Problems
Chapter 2.
1.
2.
3.
4.
5.
6.
7.
8.
From z to the Riemann mapping theorem: some finer points of basic

complex analysis
The winding number version of Cauchys theorem
Isolated singularities and residues
Analytic continuation
Convergence and normal families
The Mittag-Leffler and Weierstrass theorems
The Riemann mapping theorem
Runges theorem
Problems
1
1
3
7
12
19
21
24
27
27
29
33
36
37
41
44
46
Chapter 3. Harmonic functions on D

1. The Poisson kernel
2. Hardy classes of harmonic functions
3. Almost everywhere convergence to the boundary data
4. Problems
51
51
53
55
58
Chapter 4. Riemann surfaces: definitions, examples, basic properties

1. The basic definitions
2. Examples
3. Functions on Riemann surfaces
4. Degree and genus
5. Riemann surfaces as quotients
6. Elliptic functions
7. Problems
63
63
64
67
69
70
73
77
Chapter 5. Analytic continuation, covering surfaces, and algebraic functions

1. Analytic continuation
2. The unramified Riemann surface of an analytic germ
79
79
83
iii
iv
CONTENTS
3. The ramified Riemann surface of an analytic germ

4. Algebraic germs and functions
5. Problems
85
88
100
Chapter 6. Differential forms on Riemann surfaces

1. Holomorphic and meromorphic differentials
2. Integrating differentials and residues
3. The Hodge operator and harmonic differentials
4. Statement and examples of the Hodge decomposition
5. Problems
103
103
105
106
110
115
Chapter 7. Hodges theorem and the L2 existence theory

1. Weyls lemma and the Hodge decomposition
2. Existence of nonconstant meromorphic functions
3. Problems
119
119
123
128
Chapter 8. The Theorems of Riemann-Roch, Abel, and Jacobi

1. Homology bases, periods, and Riemanns bilinear relations
2. Divisors
3. The proof of the Riemann-Roch theorem
4. Applications and general divisors
5. The theorems of Abel and Jacobi
6. Problems
129
129
136
137
139
142
142
Chapter 9. The Dirichlet problem and Green functions

1. Green functions
2. The potential theory proof of the Riemann mapping theorem
3. Existence of Green functions via Perrons method
4. Behavior at the boundary
145
145
147
148
151
Chapter 10. Green functions and the classification problem

1. Green functions on Riemann surfaces
2. Hyperbolic Riemann surfaces admit Green functions
3. Problems
155
155
156
160
Chapter 11. The uniformization theorem

1. The statement for simply connected surfaces
2. Hyperbolic, simply connected, surfaces
3. Parabolic, simply connected, surfaces
161
161
161
162
Chapter 12. Review of some facts from algebra and geometry

1. Geometry and topology
2. Algebra
165
165
168
Bibliography
171
Preface
During their first year at the University of Chicago, graduate students in mathematics take classes in algebra, analysis, and geometry, one of each every quarter. The
analysis classes typically cover real analysis and measure theory, functional analysis, and
complex analysis. This book grew out of the authors notes for the complex analysis
class which he taught during the Spring quarter of 2007 and 2008. The course covered
elementary aspects of complex analysis such as the Cauchy integral theorem, the residue
theorem, Laurent series, and the Riemann mapping theorem with Riemann surface theory. Needless to say, all of these topics have been covered in excellent textbooks as
well as classic treatise. This book does not try to compete with the works of the old
masters such as Ahlfors [1], HurwitzCourant [20], Titchmarsh [39], AhlforsSario [2],
Nevanlinna [34], Weyl [41]. Rather, it is intended as a fairly detailed yet fast paced
guide through those parts of the theory of one complex variable that seem most useful
in other parts of mathematics. There is no question that complex analysis is a corner
stone of the analysis education at every university and each area of mathematics requires
at least some knowledge of it. However, many mathematicians never take more than an
introductory class in complex variables that often appears awkward and slightly outmoded. Often, this is due to the omission of Riemann surfaces and the assumption of
a computational, rather than geometric point of view. Therefore, the authors has tried
to emphasize the very intuitive geometric underpinnings of elementary complex analysis
which naturally lead to Riemann surface theory. As for the latter, today it is either not
taught at all or sometimes given a very algebraic slant which does not appeal to more
analytically minded students. This book intends to develop the subject of Riemann surfaces as a natural continuation of the elementary theory without which the latter would
indeed seem artificial and antiquated. At the same time, we do not overly emphasize the
algebraic aspect such as elliptic curves. The author feels that those students who wish
to pursue this direction will be able to do so quite easily after mastering the material in
this book. Because of such omissions as well as the reasonably short length of the book
it is to be considered intermediate.
Partly because of the fact that the Chicago first year curriculum covers topology and
geometry this book assumes knowledge of basic notions such as homotopy, the fundamental group, differential forms, co-homology and homology, and from algebra we require
knowledge of the notions of groups and fields, and some familiarity with the resultant of
two polynomials (but the latter is needed only for the definition of the Riemann surfaces
of an algebraic germ). However, only the most basic knowledge of these concepts is
assumed and we collect the few facts that we do need in Chapter 12.
Let us now describe the contents of the individual chapters in more detail. Chapter 1 introduces the concept of differentiability over C, the calculus of z , z, Mobius
(or fractional linear) transformations and some applications of these transformations to
v
vi
PREFACE
hyperbolic geometry. In particular, we prove the Gauss-Bonnet theorem in that case.

Next, we develop integration and Cauchys theorem in various guises, then apply this to
the study of analyticity, and harmonicity, the logarithm and the winding number. We
conclude the chapter with some brief comments about co-homology and the fundamental
group.
Chapter 2 refines the Cauchy formula by extending it to zero homologous cycles, i.e.,
those cycles which do not wind around any point outside of the domain of holomorphy.
We then classify isolated singularities, prove the Laurent expansion and the residue
theorems with applications. After that, Chapter 2 studies analytic continuation and
presents the monodromy theorem. Then, we turn to convergence of analytic functions
and normal families with application to the Mittag-Leffler and Weierstrass theorems in
the entire plane, as well as the Riemann mapping theorem. The chapter concludes with
Runges theorem.
In Chapter 3 we study the Dirichlet problem on the unit disk. This means that we
solve the boundary value problem for the Laplacian on the disk via the Poisson kernel. We
present the usual Lp based Hardy classes of harmonic functions on the disk, and discuss
the question of representing them via their boundary data both in the Lp and the almost
every sense. We then sketch the more subtle theory of homolomorphic functions in the
Hardy class, or equivalently of the boundedness properties of the conjugate harmonic
functions (with the F.& M. Riesz theorem and the notion of inner and outer functions
being the most relevant here).
The theory of Riemann surfaces begins with Chapter 4. This chapter covers the basic
definitions of such surfaces and the analytic functions on them. Elementary results such
as the Riemann-Hurwitz formula for the branch points are discussed and several examples
of surfaces and analytic functions defined on them are presented. In particular, we show
how to define Riemann surfaces via discontinuous group actions and give examples of this
procedure. The chapter closes with a discussion of tori and some aspects of the classical
theory of meromorphic functions on these tori (doubly periodic or elliptic functions).
Chapter 5 presents another way in which Riemann surfaces arise naturally, namely
via analytic continuation. Historically, the desire to resolve unnatural issues related
to multi-valued functions (most importantly for algebraic functions) lead Riemann
to introduce his surfaces. Even though the underlying ideas leading from a so-called
analytic germ to its Riemann surface are very geometric and intuitive (and closely related
to covering spaces in topology), their rigorous rendition requires some patience as ideas
such as analytic germ, branch point, (un)ramified Riemann surface of an analytic
germ etc., need to be defined precisely. This typically proceeds via some factorization
procedure of a larger object (i.e., equivalence classes of sets which are indistinguishable
from the point of view of the particular object we wish to construct). The chapter
also develops some basic aspects of algebraic functions and their Riemann surfaces. At
this point the reader will need to be familiar with the resultant of two polynomials.
In particular, we will see that every (!) compact Riemann surface is obtained through
analytic continuation of some algebraic germ. This uses the machinery of Chapter 5
together with a potential theoretic result that guarantees the existence of a non-constant
meromorphic function on every Riemann surface, which we prove in Chapter 7.
Chapter 6 introduces differential forms on Riemann surfaces and their integrals.
Needless to say, the only really important class are the 1-forms and we define harmonic,
holomorphic and meromorphic forms and the residues in the latter case. Furthermore,
PREFACE
vii
the Hodge operator appears naturally. We then present some examples that lead up to
the Hodge decomposition in the next chapter. This refers to the fact that every 1-form
can be decomposed additively into three components: a closed, co-closed, and a harmonic
form (the latter being characterized as being simultaneously closed and co-closed). In
this book, we follow the classical L2 -based derivation of this theorem. Thus, via Hilbert
space methods one first derives this decomposition with L2 -valued forms and then uses
Weyls regularity lemma (weakly harmonic functions are smoothly harmonic) to upgrade
to smooth forms.
The proof of the Hodge theorem is presented in Chapter 7. This chapter includes a
theorem on the existence of meromorphic differentials and functions on a general Riemann
surface. In particular, we derive the striking fact that every Riemann surface carries a
non-constant meromorphic function which is needed to complete the result on compact
surfaces being algebraic in Chapter 5.
Chapter 8 presents the well-known Riemann-Roch theorem which computes the dimension of certain spaces of meromorphic differentials from properties of the so-called
divisor and the genus of the underlying compact Riemann surface. Before proving the
theorem, there are a number of prerequisites to be dealt with, such as the Riemann
period relations and the definition of a divisor.
The remaining Chapters 9, 10 and 11 are devoted to the proof of the uniformization
theorem. This theorem states that the only simply connected Riemann surfaces (up
to isomorphisms) are C, D, and CP 1 . For the compact case, we deduce this from the
Riemann-Roch theorem. But for the other two cases we use methods of potential theory
which are motivated by the proof of the Riemann mapping theorem which is based on
the existence of a Green function. It turns out that such a function only exists for the
hyperbolic surfaces (such as D) but not for the parabolic case (such as C) or the compact
case. Via the Perron method, we prove the existence of a Green function for hyperbolic
surfaces, thus establishing the conformal equivalence with the disk. For the parabolic
case, a suitable substitute for the Green function needs to be found. We discuss this
in detail for the simply connected case, and also sketch some aspects of the non-simply
connected cases.
CHAPTER 1
From i to z: the basics of complex analysis

1. The field of complex numbers
The field C of complex numbers is obtained by adjoining i to the field R of reals. The
defining property of i is i2 + 1 = 0 and complex numbers z1 = x1 + iy1 and z2 = x2 + iy2
are added component-wise and multiplied according to the rule
z1 z2 = x1 x2 y1 y2 + i(x1 y2 + x2 y1 )
which follows from i2 + 1 = 0 and the distributional law. The conjugate of z = x + iy is
z = x iy and we have |z|2 := z
z = x2 + y 2 . Therefore every z 6= 0 has a multiplicative
1
2
inverse given by z := z|z| and C becomes a field. Since complex numbers z can be
represented as points or vectors in R2 in the Cartesian way, we can also assign polar
coordinates (r, ) to them. By definition, r = |z| and z = r(cos + i sin ). The addition
theorems for cosine and sine imply that
z1 z2 = |z1 ||z2 |(cos(1 + 2 ) + i sin(1 + 2 ))
which reveals the remarkable fact that complex numbers are multiplied by multiplying
their lengths and adding their angles. In particular, |z1 z2 | = |z1 ||z2 |. This shows that
power series behave as in the real case with respect to convergence, i.e.,
n=0
an z n converges on |z| < R and diverges for every |z| > R

1
R1 = lim sup |an | n

n
where the sense of convergence is relative to the length metric | | on complex numbers
which is the same as the Euclidean distance on R2 (the reader should verify the triangle
inequality); the formula for R of course follows from comparison with the geometric
series. Note that the convergence is absolute on the disk |z| < R and uniform on every
compact subset of that disk. Moreover, the series diverges for every
|z| > R as can be
an
seen by the comparison test. We can also write R = limn an+1 , provided this limit
exists. The first example that comes to mind here is
X
1
zn,
=
1 z n=0
|z| < 1.
Another example is of course

(1.1)
E(z) :=
X
zn
n=0
1
n!
1. BASIC COMPLEX ANALYSIS I
which converges absolutely and uniformly on every compact subset of C. Expanding

(z1 + z2 )n via the binomial theorem shows that E(z1 + z2 ) = E(z1 )E(z2 ). Recall the
definition of the Euler constant e: consider the ordinary differential equation (ODE)
y = y with y(0) = 1 which has a unique solution y(t) for all t R. Then set e := y(1).
Let us solve our ODE iteratively by the Picard method. Thus,
Z t
Z t
(t s)y(s) ds = . . .
y(s) ds = 1 + t +
y(t) = 1 +
0
n
X
tj
j=0
j!
(t s)n y(s) ds.
The integral on the right vanishes as n and we obtain

y(t) =
j
X
t
j=0
j!
which in particular yields the usual series expansion for e. Also, by the group property
of flows,
y(t2 )y(t1 ) = y(t1 + t2 )
which proves that y(t) = et for every rational t and motivates why we define
et :=
j
X
t
j=0
j!
t R.
Hence, our series E(z) above is used as the definition of ez for all z C. We have the
group property ez1 +z2 = ez1 ez2 , and by comparison with the power series of cos and sin
on R, we arrive at the famous Euler formula
ei = cos() + i sin()
for all R. This in particular shows that z = rei where (r, ) are the polar coordinates
of z. This in turn implies that
(cos + i sin )n = cos(n) + i sin(n)
for every n 1 (de Moivres formula). Now suppose that z = rei with r > 0. Then by
the preceding,
z = elog r+i or log z = log r + i.
Note that the logarithm is not well-defined since and + 2n for any n Z both have
the property that exponentiating leads to z. Similarly,
2ik n
1
=z 1kn
r n ei n e n
which shows that there are n different possibilities for n z. Later on we shall see how
these functions become single-valued on their natural Riemann surfaces. Let us merely
mention at this point that the complex exponential is most naturally viewed as the
covering map

C C := C \ {0}
z 7 ez
showing that C is the universal cover of C .
2. DIFFERENTIABILITY AND CONFORMALITY
But for now, we of course wish to differentiate functions defined on some open set
C. There are two relevant notions of derivative here and we will need to understand
how they relate to each other.
2. Differentiability and conformality
The first is the crucial linearization idea from multivariable calculus and the second
copies the idea of difference quotients from calculus. In what follows we shall either use
U or to denote planar regions, i.e., open and connected subsets of R2 . Also, we will
identify z = x + iy with the real pair (x, y) and will typically write a complex-valued
function as f (z) = u(z) + iv(z) = (u, v)(z) where u, v : C R.
Definition 1.1. Let f : C.
(a) We say that f C 1 () iff there exists df C(, M2 (R)), a 2 2 matrix-valued
function, such that
f (z + h) = f (z) + df (z)(h) + o(|h|)
h R2 , |h| 0
where df (z)(h) means the matrix df (z) acting on the vector h.

(b) We say that f is holomorphic on if
f (z) := lim
wz
f (w) f (z)
wz
exists for all z and is continuous on . We denote this by f H(). A function

f H(C) is called entire.
Note that (b) is equivalent to the existence of a function f C() so that
f (z + h) = f (z) + f (z)h + o(|h|)
|h| 0
where f (z)h is the product between the complex numbers f (z) and h. Hence, we
conclude that the holomorphic functions are precisely those functions in C 1 () in the
sense of (a) for which the differential df (z) acts as linear map via multiplication by a
complex number. Obvious examples of holomorphic maps are the powers f (z) = z n for
all n Z (if n is negative, then we exclude z = 0). They satisfy f (z) = nz n1 by
the binomial theorem. Also, since we can do algebra in C the same way we did over R
it follows that the basic differentiation rules like the sum, product, quotient, and chain
rules continue to hold for holomorphic functions. Let us demonstrate this for the chain
rule: if f H(), g H( ) and f : , then we know from the C 1 -chain rule that
(f g)(z + h) = (f g)(z) + Df (g(z))Dg(z)h + o(|h|)
|h| 0.
From (b) above we infer that Df (g(z)) and Dg(z) act as multiplication by the complex
numbers f (g(z)) and g (z), respectively. Thus, we see that f g H() and (f g) =
f (g)g . We leave the product and quotient rules to the reader.
It is clear that all polynomials are holomorphic functions. In fact, we can generalize
this to all power series within their disk of convergence. Let us make this more precise.
Definition 1.2. We say that f : C is analytic (or f A()) if f is represented
by a convergent power series expansion on a neighborhood around every point of .
Lemma 1.3. A() H()
Proof. Suppose z0 and

f (z) =
n=0
an (z z0 )n
z so that |z z0 | < r(z0 )
where r(z0 ) > 0. As in real calculus, one checks that differentiation can be interchanged
with summation and
X
nan (z z0 )n1
|z z0 | < r(z0 ).
f (z) =
n=0
In fact, one can differentiate any number of times and

f
(k)
(z) =
(n)k an (z z0 )nk
n=0
|z z0 | < r(z0 )
where (n)k = n(n 1) . . . (n k + 1). This proves also that an =
f (n) (z0 )
n!
for all n 0.
We note that with ez defined as above, (ez ) = ez from the series representation (1.1).
It is a remarkable fact of basic complex analysis that one has equality in Lemma 1.3,
i.e., A() = H(). In order to establish this equality, we need to be able to integrate;
see the section about integration below.
Recall that f = u + iv = (u, v) belongs to C 1 () iff the partials ux , uy , vx , vy exist
and are continuous on . If f H(), then by letting w approach z along the x or
ydirections, respectively,
f (z) = ux + ivx = iuy + vy
so that
u x = vy ,
uy = vx .
These relations are known as the CauchyRiemann equations. They are equivalent to
the property that

ux uy
df =
= A for some 0, A SO(2, R).
vx vy
In other words, at each point where a holomorphic function f has a nonvanishing derivative, its differential df is a conformal matrix: it preserves angles and the orientation
between vectors. Conversely, if f C 1 () has the property that df is proportional to a
rotation everywhere on , then f H(). Let us summarize these observations.
Theorem 1.4. A complex-valued function f C 1 () is holomorphic iff the CauchyRiemann system holds in . This is equivalent to df being the composition of a rotation
and a dilation (possibly by zero) at every point in .
Proof. We already saw that the Cauchy-Riemann system is necessary. Conversely,
since f C 1 (), we can write:
u(x + , y + ) = u(x, y) + ux (x, y) + uy (x, y) + o(|(, )|)
v(x + , y + ) = v(x, y) + vx (x, y) + vy (x, y) + o(|(, )|).

Using that ux = vy and uy = vx we obtain, with = + i,
f (z + ) f (z) = (ux + ivx )(z)( + i) + o(||)
2. DIFFERENTIABILITY AND CONFORMALITY
which of course proves that f (z) = ux (z) + ivx (z) = vy (z) iuy (z) as desired. The
second part was already discussed above.

The following notion is of central importance for all of complex analysis:
Definition 1.5. A function f C 1 () is called conformal if and only if df 6= 0 in
and df preserves the angle and orientation at each point.
Thus, the holomorphic functions are precisely those C 1 functions which are conformal
at all points at which df 6= 0. Note that f (z) = z belongs to C 1 (C) but is not holomorphic
since it reverses orientations. Also note that f (z) = z 2 doubles angles at z = 0 (in the
sense that curves crossing at 0 at angle get mapped onto curves intersecting at 0 at
angle 2), so conformality is lost there.
A particularly convenient as well as insightful way of distinguishing holomorphic
functions from C 1 functions is given by the z , z calculus. Assume that f C 1 (). Then
the real-linear map df (z) can be written as the sum of a complex-linear (meaning that
T (zv) = zT (v)) and a complex anti-linear transformation (meaning that T (zv) = zT (v));
see Lemma 6.2 below. In other words, there exist complex numbers w1 (z), w2 (z) such
that
df (z) = w1 (z) dz + w2 (z) d
z
where dz is simply the identity map and d
z the reflection about the real axis followed by
multiplication by the complex numbers w1 and w2 , respectively. We used here that all
complex linear transformations on R2 are given by multiplication by a complex number,
whereas the complex anti-linear ones become complex linear by composing them with a
reflection. To find w1 and w2 simply observe that
1
1
z ) + y f (dz d
z)
df (x) = x f dx + y f dy = x f (dz + d
2
2i
1
1
= (x f iy f ) dz + (x f + iy f ) d
z
2
2
=: z f dz + zf d
z.
In other words, f H() iff f C 1 () and zf = 0 in .
One can immediately check that zf = 0 is the same as the Cauchy-Riemann system. As
an application of this formalism we record the following crucial fact: for any f H(),
d(f (z) dz) = z f dz dz + zf d
z dz = 0
which means that f (z) dz is a closed differential form. This property is equivalent to the
homotopy invariance of the Cauchy integral that we will encounter below. We leave it
to the reader to verify the chain rules

z (g f ) = (w g) f z f + (w g) f z f

(1.2)
z(g f ) = (w g) f zf + (w g) f zf
2
as well as the representation of the Laplacean = 4 z

z . These ideas will be of particular
importance once we discuss differential forms on Riemann surfaces.
To continue our introductory chapter, we next turn to the simple but important idea
of extending the notion of analyticity to functions that take the value . In a similar
vein, we can make sense of functions being analytic at z = . To start with, we define
the one-point compactification of C, which we denote by C , with the usual basis of
the topology; the neighborhoods of are the complements of all compact sets. It is
intuitively clear that C S 2 in the homeomorphic sense. Somewhat deeper as well

as much more relevant for complex analysis is the fact that C S 2 \ {p} in the sense
of conformal equivalence where p S 2 is arbitrary. This is done via the well-known
stereographic projection; see the homework and Chapter 4 below as well as Figure 1.1.
x
as
If the circle in that figure is the unit circle, N = (0, 1), and X = (x, y), then Z = 1y
the reader will easily verify using similarity of triangles. This identifies the stereographic
projection as the map
x1 + ix2
: S 2 \ {(0, 0, 1)} C, X = (x1 , x2 , x3 ) 7
1 x3
The stereographic projection preserves angles as well as circles; see Problem 1.4. We will
see in Chapter 4 that
C S 2 CP 1
in the sense of conformal equivalences, and each of these Riemann surfaces are called
the Riemann sphere. Without going into details about the exact definition of a Riemann
N
Figure 1.1. Stereographic projection

surface, we mention in passing that C is covered by two charts, namely (C, z) and
(C \{0}, z 1 ), both of which are homeomorphisms onto C. On the overlap region C :=
C \ {0}, the change of charts is given by the map z 7 z 1 , which is of course a conformal
equivalence.
It is now clear how to extend the domain and range of holomorphic maps to
(1.3)
f : C C ,
f : C C,
f : C C .
First, we need to require that f is continuous in each case. This is needed in order to
ensure that we can localize f to charts. Second, we require f to be holomorphic relative
to the respective charts. For example, if f (z0 ) = for some z0 C, then we say that f
1
is holomorphic around z0 . To make sense of
is holomorphic close to z0 if and only if f (z)
f being analytic at z = with values in C, we simply require that f ( 1z ) is holomorphic

3. MOBIUS
TRANSFORMS
around z = 0. For the final example in (1.3), if f () = , then f is analytic around

z = if and only if 1/f (1/z) is analytic around z = 0. We remark that z 7 z1 is
conformal as a map from C C ; this is a tautology in view of our choice of chart at
z = . On the other hand, if we interpret C as the Riemann sphere then one needs
to use here that stereographic projection is a conformal map.
We shall see later in this chapter that the holomorphic maps f : C C are
constants (indeed, such a map would have to be entire and bounded and therefore constant by Liouvilles theorem, see Corollary 1.19 below). On the other hand, the maps
f : C C are precisely the meromorphic ones which we shall encounter in the next
chapter. Finally, the holomorphic maps f : C C are precisely the rational funcP (z)
where P, Q are polynomials. To see this1, one simply argues that any such f is
tions Q(z)
necessarily meromorphic with only finitely many poles in C and possibly a pole at z = .
3. M
obius transforms
If we now accept that the holomorphic, and thus conformal, maps C C are
precisely the rational ones, it is clear how to identify the conformal automorphisms (or
automorphisms) amongst these maps. Indeed, in that case necessarily P and Q both have
to be linear which immediately leads to the following definition. Based on the argument
of the previous paragraph (which the reader for now can ignore if desired), the lemma
identifies all automorphisms of C .
Lemma 1.6. Every A GL(2, C) defines a transformation

az + b
a b
,
A=
TA (z) :=
c d
cz + d
which is holomorphic as a map from C C . It is called a fractional linear or

Mobius transformation. The map A 7 TA only depends on the equivalence class of A
under the relation A B iff A = B, C . In other words, the family of all M
obius
transformations is the same as
(1.4)
P SL(2, C) := SL(2, C)/{Id}
obius transform is an
We have TA TB = TAB and TA1 = TA1 . In particular, every M
automorphism of C .
Proof. It is clear that each TA is a holomorphic map C C . The composition
law TA TB = TAB and TA1 = TA1 are simple computations that we leave to the
reader. In particular, TA has a conformal inverse and is thus an automorphism of C .
e SL(2, C), then
If TA = TAe where A, A
TA (z) =
ad bc
e
ade ebe
c
= TAe(z) =
2
e
(cz + d)
(e
cz + d)2
e under the assumption that

and thus cz + d = (e
cz + d)
ad bc = e
ade e
be
c=1
e are the same matrices in SL(2, C) possibly up to a choice of sign, which

Hence, A and A
establishes (1.4).

1The reader should not be alarmed in case he or she does not follow these arguments they will
become clear once this chapter and the next one have been read.
Fractional linear transformations enjoy many important properties which can be

checked separately for each of the following four elementary transformations. In particular, Lemma 1.7 proves that the group P SL(2, C) has four generators.
Lemma 1.7. Every M
obius transformation is the composition of four elementary
maps:
translations z 7 z + z0
dilations z 7 z, > 0
rotations z 7 ei z, R
inversion z 7 1z
Proof. If c = 0, then TA (z) = ad z + db . If c 6= 0, then
TA (z) =
bc ad 1
c2 z +
d
c
a
c
and we are done.
z1
take the right half-plane
The reader will have no difficulty verifying that z 7 z+1
on the disk D := {|z| < 1}. In particular, iR gets mapped on the unit circle. Similarly,
z 7 2z1
2z takes D onto itself with the boundary going onto the boundary. If we include all
lines into the family of circles (they are circles passing through ) then these examples
can serve to motivate the following lemma.
Lemma 1.8. Fractional linear transformations take circles onto circles.

Proof. In view of the previous lemma, the only case requiring an argument is the
inversion. Thus, let |z z0 | = r be a circle and set w = z1 . Then
0 = |z|2 2Re (
z z0 ) + |z0 |2 r 2
=
1
Re (wz0 )
2
+ |z0 |2 r 2
2
|w|
|w|2
If |z0 | = r, then one obtains the equation of a line in w. Note that this is precisely the
case when the circle passes through the origin. Otherwise, we obtain the equation

2
z0
r2

0 = w

|z0 |2 r 2
(|z0 |2 r 2 )2
which is a circle. A line is given by an equation
2Re (z
z0 ) = a
which transforms into 2Re (z0 w) = a|w|2 . If a = 0, then we simply obtain another line
through the origin. Otherwise, we obtain the equation |w z0 /a|2 = |z0 /a|2 which is a
circle.
An alternative argument uses the fact that stereographic projection preserves circles,
see homework problem #4. To use it, note that the inversion z 7 1z corresponds to a
rotation of the Riemann sphere about the x1 axis (the real axis of the plane). Since such
a rotation preserves circles, a fractional linear transformation does, too.

2
Since T z = az+b
obius transform T , we see
cz+d = z is a quadratic equation for any M
that T can have at most two fixed points unless it is the identity.
2Strictly speaking, this is a quadratic equation provided c 6= 0; if c = 0 one obtains a linear equation
with a fixed point in C and another one at z = .

3. MOBIUS
TRANSFORMS
It is also clear that every Mobius transform has at least one fixed point. The map
T z = z + 1 has exactly one fixed point, namely z = , whereas T z = z1 has two, z = 1.
Lemma 1.9. A fractional linear transformation is determined completely by its action
on three distinct points. Moreover, given z1 , z2 , z3 C distinct, there exists a unique
fractional linear transformation T with T z1 = 0, T z2 = 1, T z3 = .
Proof. For the first statement, suppose that S, T are Mobius transformations that
agree at three distinct points. Then S 1 T has three fixed points and is thus the identity.
For the second statement, let
T z :=
z z1 z2 z3
z z3 z2 z1
in case z1 , z2 , z3 C. If any one of these points is , then we obtain the correct formula
by passing to the limit here.

Definition 1.10. The cross ratio of four points z0 , z1 , z2 , z3 C is defined as
z0 z1 z2 z3
[z0 : z1 : z2 : z3 ] :=
z0 z3 z2 z1
This concept is most relevant for its relation to Mobius transformations.
Lemma 1.11. The cross ratio of any four distinct points is preserved under M
obius
transformations. Moreover, four distinct points lie on a circle iff their cross ratio is real.
Proof. Let z1 , z2 , z3 be distinct and let T zj = wj for T a Mobius transformation
and 1 j 3. Then for all z C,
[w : w1 : w2 : w3 ] = [z : z1 : z2 : z3 ] provided w = T z
This follows from the fact that the cross ratio on the left-hand side defines a Mobius
transformation S1 w with the property that S1 w1 = 0, S1 w2 = 1, S1 w3 = , whereas the
right-hand side defines a transformation S0 with S0 z1 = 0, S0 z2 = 1, S0 z3 = . Hence
S11 S0 = T as claimed. The second statement is an immediate consequence of the
first and the fact that for any three distinct points z1 , z2 , z3 R, a fourth point z0 has a
real-valued cross ratio with these three iff z0 R.

We can now define what it means for two points to be symmetric relative to a circle
(or line recall that this is included in the former).
Definition 1.12. Let z1 , z2 , z3 where C is a circle. We say that z and z
are symmetric relative to iff
[z : z1 : z2 : z3 ] = [z : z1 : z2 : z3 ]
Obviously, if = R, then z = z. In other words, if is a line, then z is the
reflection of z across that line. If is a circle of finite radius, then we can reduce matters
to this case by an inversion.
Lemma 1.13. Let = {|z z0 | = r}. Then for any z C ,
z =
r2
z z0
10
Proof. It suffices to consider the unit circle. Then

[z; z1 ; z2 : z3 ] = [
z : z11 : z21 : z31 ] = [1/
z : z1 : z2 : z3 ]
In other words, z = 1z . The general case now follows from this via a translation and
dilation.
Figure 1.2. Geodesics in the hyperbolic plane

Mobius transformations are important for several reasons. We already observed that
they are precisely the automorphisms of the Riemann sphere (though to see that every
automorphism is a Mobius transformation requires material from this entire chapter as
well as the next). In the 19th century there was much excitement surrounding nonEuclidean geometry and there is an important connection between Mobius transforms
and hyperbolic geometry: the isometries of the hyperbolic plane H are precisely those
Mobius transforms which preserve it. Let us be more precise. Consider the upper halfplane model of the hyperbolic plane given by
H = {z C : Im z > 0},
ds2 =
d
z dz
dx2 + dy 2
=
2
y
(Im z)2
It is not hard to see that the subgroup of P SL(2, C) which preserves the upper halfplane is precisely P SL(2, R). Indeed, z 7 az+b
cz+d preserves R := R {} if and only if
a, b, c, d R for some C . In other words, the stabilizer of R (as a set) is P GL(2, R)
which contains P SL(2, R) as an index two subgroup. The latter preserves the upper half
plane, whereas those matrices with negative determinant interchange the upper with
the lower half-plane. It is easy to check (see the home work problems) that P SL(2, R)
operates transitively on H and preserves the metric: for the latter, one simply computes
dw
dw
d
z dz
az + b
=
=
w=
2
cz + d
(Im w)
(Im z)2
In particular, the geodesics are preserved under P SL(2, R). Since the metric does not
depend on x it follows that all vertical lines are geodesics. We leave it to the reader to

3. MOBIUS
TRANSFORMS
11
verify that for all z0 H

{T (z0 ) | T Stab(z0 )} = SO(2, R)
which means that the stabilizer subgroup in P SL(2, R) at any point z0 in the upper half
plane acts on the tangent space at z0 by arbitrary rotations. Therefore, the geodesics
of H are precisely all circles which intersect the real line at a right angle (with the vertical
lines being counted as circles of infinite radius). From this it is clear that the hyperbolic
plane satisfies all axioms of Euclidean geometry with the exception of the parallel axiom:
there are many lines (i.e., geodesics) passing through a point which is not on a fixed
geodesic that do not intersect that geodesic. Let us now prove the famous Gauss-Bonnet
theorem which describes the hyperbolic area of a triangle whose three sides are geodesics
(those are called geodesic triangles). This is of course a special case of a much more
general statement about integrating the Gaussian curvature over a geodesic triangle on a
general surface. The reader should prove the analogous statement for spherical triangles.
Figure 1.3. Geodesic triangles
Theorem 1.14. Let T be a geodesic triangle with angles 1 , 2 , 3 . Then Area(T ) =

(1 + 2 + 3 ).
Proof. There are four essentially distinct types of geodesic triangles, depending on
how many of its vertices lie on R . Up to equivalences via transformations in P SL(2, R)
(which are isometries and therefore also preserve the area) we see that it suffices to
consider precisely those cases described in Figure 1.3. Let us start with the case in which
exactly two vertices belong to R as shown in that figure (the second triangle from the
right). Without loss of generality one vertex coincides with 1, the other with , and the
circular arc lies on the unit circle with the projection of the second finite vertex onto the
12
real axis being x0 . Then

Area(T ) =
1Z
x0
0
y(x)
dxdy
=
y2
x0
dx
1 x2
d cos
p
= 0 = 1
1 cos2 ()
as desired since the other two angles are zero. By additivity of the area we can deal
with the other two cases in which at least one vertex is real. We leave the case where no
vertex lies on the (extended) real axis to the reader, the idea is to use Figure 1.4.
C
B
Figure 1.4. The case of no real vertex

We leave it to the reader to generalize the Gauss-Bonnet theorem to geodesic polygons. Many interesting questions about Mobius transformations remain, for example how
to characterize those that correspond to rotations of the sphere, or how to determine all
finite subgroups of P SL(2, C). For some answers see Problem 4.9 as well as [22]. A
whole topic onto itself are the Fuchsian and Kleinian groups, see for example [25]. These
groups are of crucial importance for the uniformization theory of Riemann surfaces in
the non-simply connected case.
4. Integration
We now develop our complex calculus further. The following definition defines the
complex integral canonically in the sense that it is the only definition which preserves
the fundamental theorem of calculus for holomorphic functions.
Definition 1.15. For any C 1 -curve : [0, 1] and any compex-valued f C()
we define
Z 1
Z
f ((t)) (t) dt
f (z) dz =
If is a closed curve ((0) = (1)) then we also write
f (z) dz for this integral.
We remark that the integral is C-valued, and that f ((t)) (t) is understood as multiplication of complex numbers. From the chain rule, we deduce the fundamental fact that
the line integrals of this definition do not depend on any particular C 1 parametrization
of the curve as long as the orientation is preserved (hence, there is no loss in assuming
4. INTEGRATION
13
that is parametrized by 0 t 1). Again from the chain rule, we immediately obtain
the following: if f H(), then
Z 1
Z 1
Z
d
f ((t)) (t) dt =
f (z) dz =
f ((t)) dt = f ((1)) f ((0))
0 dt
0
for any as in the definition. In particular,

I
f (z) dz = 0
closed curves in
On the other hand, let us compute with r (t) := reit , r > 0,

Z 2
I
0
n 6= 1
n int
it
n
r e rie dt =
z dz =
(1.5)
2i
n = 1
0
r
n+1
In = C , the function f (z) = z n has the primitive Fn (z) = zn+1 provided n 6= 1.

This explains why we obtain 0 for all n 6= 1. On the other hand, if n = 1 we realize
from our calculation that 1z does not have a (holomorphic) primitive in C . This issue
merits further investigation (for example, we need to answer the question whether 1z
has a local primitive in C this is indeed the case and this primitive is a branch of
log z). Before doing so, however, we record Cauchys famous theorem in its homotopy
version. Figure 1.5 shows two curves, namely 1 and 2 , which are homotopic within the
1
2
Figure 1.5. Homotopy

annular region they lie in. The dashed curve is not homotopic to either of them within
the annulus.
14
Theorem 1.16. Let 0 , 1 : [0, 1] be C 1 curves3 with 0 (0) = 1 (0) and

0 (1) = 1 (1) (the fixed endpoint case) or 0 (0) = 0 (1), 1 (0) = 1 (1) (the closed case).
Assume that they are C 1 -homotopic4 in the following sense: there exists a continuous
map H : [0, 1]2 with H(t, 0) = 0 (t), H(t, 1) = 1 (t) and such that H(, s) is a C 1
curve for each 0 s 1. Moreover, in the fixed endpoint case we assume that H freezes
the endpoints, whereas in the closed case we assume that each curve from the homotopy
is closed. Then
Z
Z
f (z) dz = f (z) dz
1
for all f H(). In particular, if is a closed curve in which is homotopic to a point,

then
I
f (z) dz = 0
Proof. We first note the important fact that f (z) dz is a closed form. Indeed,
d(f (z) dz) = z f (z) dz dz + zf (z) d
z dz = 0
by the Cauchy-Riemann equation zf = 0. Thus, Cauchys theorem is a special case

of the homotopy invariance of the integral over closed forms which in turn follows from
Stokess theorem. Let us briefly recall the details: since a closed form is locally exact,
we first note that
I
f (z) dz = 0
for all closed curves which fall into sufficiently small disks, say. But then we can
triangulate the homotopy so that
Z
Z
XI
f (z) dz = 0
f (z) dz f (z) dz =
1
j
j
where the sum is over a finite collection of small loops which constitute the triangulation
of the homotopy H. The more classically minded reader may prefer to use Greens formula (which of course follows from the Stokes theorem): provided U is a sufficiently
small neighborhood which is diffeomorphic to a disk, say, one can write
I
I
u dx v dy + i(u dy + v dx)
f (z) dz =
U
ZZ
(uy vx ) dxdy + i
ZZ
(vy + ux ) dxdy = 0
where the final equality sign follows from the CauchyRiemann equations.
This theorem is typically applied to very simple configurations, such as two circles
which are homotopic to each other in the region of holomorphy of some function f . As
an example, we now derive the following fundamental fact of complex analysis which is
intimately tied up with the n = 1 case of (1.5).
3This can be relaxed to piece-wise C 1 , which means that we can write the curve as a finite concatenation of C 1 curves. The same comment applies to the homotopy.
4In light of commonly used terminology it is probably best to refer to this as homotopic through C 1
curves but for simplicity, we shall continue to abuse terminology and use C 1 homotopic.
4. INTEGRATION
15
Proposition 1.17. Let D(z0 , r) and f H(). Then

I
1
f ()
(1.6)
f (z) =
d where (t) = z0 + reit
2i
z
for all z D(z0 , r).

Proof. Fix any z D(z0 , r) and apply Theorem 1.16 to the region U := D(z0 , r) \
D(z, ) where > 0 is small. We use here that the two boundary circles of U are
homotopic to each other relative to the region . Then
Z
Z
f ()
1
f ()
1
d =
d
0=
2i
z
2i D(z0 ,r) z
U
Z
Z
1
f () f (z)
1
f (z)
d
d
2i D(z,)
z
2i D(z,) z
Z
f ()
1
d + O() f (z) as 0
=
2i D(z0 ,r) z
where we used the n = 1 case of (1.5) to pass to the third term of the last line.
We can now derive the astonishing fact that holomorphic functions are in fact analytic. This is done by noting that the integrand in (1.6) is analytic relative to z.
Corollary 1.18. A() = H(). In fact, every f H() is represented by a
convergent power series on D(z0 , r) where r = dist(z0 , ).
Proof. We have already observed that analytic functions are holomorphic. For the
converse, we use the previous proposition to conclude that
I
f ()
1
d
f (z) =
2i
z0 (z z0 )
1
=
2i
=
f () X z z0 n
d
z0
z0
X
1
2i
n=0
n=0
f ()
d (z z0 )n
( z0 )n+1
where the interchange of summation and integration is justified due to uniform and
absolute convergence of the series. Thus, we obtain that f is analytic and, moreover,
f (z) =
X
f (n)
n!
n=0
converges on |z z0 | < dist(z0 , ) with

(1.7)
for any n 0.
(n)
n!
(z0 ) =
2i
(z z0 )n
f ()
d
( z0 )n+1
16
In contrast to power
Pseries over R,n over C there is an explanation for the radius of
convergence: f (z) = n=0 an (z z0 ) has finite and positive radius of convergence R
iff f 6 H() for every which compactly contains D(z0 , R). We immediately obtain a
number of corollaries from this.
Corollary 1.19. (a) Cauchys estimates: Let f H() with |f (z)| M on .
Then
M n!
|f (n) (z)|
dist(z, )n
for every n 0 and all z .
(b) Liouvilles theorem: If f H(C) L (C), then f is constant. More generally,
if |f (z)| C(1 + |z|N ) for all z C, for some fixed integer N 0 and a finite constant
C, then f is polynomial of degree at most N .
Proof. (a) follows by putting absolute values inside (1.7). For (b) apply (a) to
= D(0, R) and let R . This shows that f (k) 0 for all k > N .

Part (b) has a famous consequence, namely the fundamental theorem of algebra.
Proposition 1.20. Every P C[z] of positive degree has a complex zero; in fact it
has exactly as many zeros over C (counted with multiplicity) as its degree.
Proof. Suppose P (z) C[z] is a polynomial of positive degree and without zero
in C. Then f (z) := P 1(z) H(C) and since |P (z)| as |z| , f is evidently
bounded. Hence f = const and so P = const contrary to the assumption of positive
degree. So P (z0 ) = 0 for some z0 C. Factoring out z z0 we conclude inductively that
P has exactly deg(P ) many complex zeros as desired.

Next, we show how Theorem 1.16 allows us to define local primitives. In particular,
we can clarify the characterization of the logarithm as the local primitive of z1 .
Proposition 1.21. Let be simply connected. Then for every f H() so that
f 6= 0 everywhere on there exists g H() with eg(z) = f (z). Thus, for any n 1
there exists fn H() with (fn (z))n = f (z) for all z . In particular, if C is
simply connected, then there exists g H() with eg(z) = z everywhere on .
Such a g
n
is called a branch of log z. Similarly, there exist holomorphic branches of any z on ,
n 1.
Proof. If eg = f , then g =
f
f
in . So fix any z0 and define

Z z
f ()
g(z) :=
d
z0 f ()
where the integration path joins z0 to z and consists a finite number of line segments
(say). We claim that g(z) does not depend on the choice of path. First note that
f
f H() due to analyticity and nonvanishing of f . Second, by the simple connectivity
of , any two curves with coinciding initial and terminal points are homotopic to each
other via a piece-wise C 1 homotopy. Thus, Theorem 1.16 yields the desired equality of
(z)
. Indeed,
the integrals. It is now an easy matter to check that g (z) = ff (z)
Z 1
g(z + h) g(z)
f (z + th)
f (z)
=
dt
as h 0
h
f (z)
0 f (z + th)
4. INTEGRATION
17
So g H() and (f eg ) 0 shows that eg = cf where c is some constant different from

zero and therefore c = ek for some k C. Hence, eg(z)k = f (z) for all z and we are
done.

Throughout, for any disk D, the punctured disk D denotes D with its center removed. The first part of the following result is known as the uniqueness theorem. The
name derives from the fact that two functions f, g H() are identical if {z :
f (z) = g(z)} has an accumulation point in . The second part establishes a normal
form for nonconstant analytic functions locally around any point, see (1.8). Amongst
other things, this yields the open mapping theorem for analytic functions.
Corollary 1.22. Let f H(). Then the following are equivalent:
f 0
for some z0 , f (n) (z0 ) = 0 for all n 0
the set {z | f (z) = 0} has an accumulation point in
Assume that f is not constant. Then at every point z0 there exist a positive integer
n and a holomorphic function h locally at z0 such that
f (z) = f (z0 ) + [(z z0 )h(z)]n ,
(1.8)
h(z0 ) 6= 0
In particular, there are disks D(z0 , ), D(f (z0 ), r) with the property that every w
D(f (z0 ), r) has precisely n pre-images under f in D(z0 , ) . If f (z0 ) 6= 0, then f
is a local C diffeomorphism. Finally, every nonconstant holomorphic map is an open
map (i.e., it takes open sets to open sets).
Proof. Let zn z0 as n , where f (zn ) = 0 for all n 1. Suppose
6= 0 for some m 0. Then
f (m) (z0 )
f (z) =
X
k=0
ak (z z0 )k = aN (z z0 )N (1 + O(z z0 )) as z z0
locally around z0 where N 0 is minimal with the property that aN 6= 0. But then it
is clear that f does not vanish on some disk D(z0 , r) , contrary to assumption. Thus,
f (n) (z0 ) = 0 for all n 0 and thus f 0 locally around z0 . Since is connected, it then
follows that f 0 on . This settles the equivalencies. If f does not vanish identically,
let us first assume that f (z0 ) 6= 0. We claim that locally around z0 , the map f (z) is a C
diffeomorphism from a neighborhood of z0 onto a neighborhood of f (z0 ) and, moreover,
that the inverse map to f is also holomorphic. Indeed, in view of Theorem 1.4, the
differential df is invertible at z0 . Hence, by the usual inverse function theorem we obtain
the statement about diffeomorphisms. Furthermore, since df is conformal locally around
z0 , its inverse is, too and so f 1 is conformal and thus holomorphic. If f (z0 ) = 0, then
there exists some positive integer n with f (n) (z0 ) 6= 0. But then
f (z) = f (z0 ) + (z z0 )n g(z)
with g H() satisfying g(z0 ) 6= 0. By Proposition 1.21 we can write g(z) = (h(z))n
for some h H(U ) where U is a neighborhood of z0 and h(z0 ) 6= 0, whence (1.8).
Figure 1.6 shows that case of n = 8. The dots symbolize the eight pre-images of some
point. Finally, by the preceding analysis of the n = 1 case we conclude that (z z0 )g(z)
is a local diffeomorphism which implies that f has the stated n-to-one mapping property.
The openness is now also evident.
18
1
0
00
11
00
11
00
11
1
0
00
11
11
00
00
11
00
11
11
00
00
11
Figure 1.6. A branch point
We remark that any point z0 for which n 2 is called a branch point. The
branch points are precisely the zeros of f in and therefore form a discrete subset of .
The open mapping part of Corollary 1.22 has an important implication known as the
maximum principle.
Corollary 1.23. Let f H(). If there exists z0 with |f (z)| |f (z0 )| for all
z , then f = const.
Proof. If f is not constant, then f () is open contradicting that f (z0 ) f (),
which is required by |f (z)| |f (z0 )| on .

The maximum principle has numerous important applications as well as variants and
extensions. In Problem 1.11, we present the simple but powerful Schwarz lemma as an
application, whereas for such extensions as the three lines and circle theorems, as well as
the Phragmen-Lindelof theorems we refer the reader to the classical literature, see [29]
and [39], as well as [37] (in fact, a version of the Phragmen-Lindelof principle is discussed
in Problem 3.6).
To conclude this chapter, we present Moreras theorem (a kind of converse to Cauchys
theorem) and (conjugate) harmonic functions. The latter is of central importance tp
complex analysis and Riemann surfaces. We begin with Moreras theorem.
Theorem 1.24. Let f C() and suppose T is a collection of triangles in which
contains all sufficiently small triangles5 in . If
I
f (z) dz = 0
T T
T
then f H().
5This means that every point in has a neighborhood in so that all triangles which lie inside that
neighborhood belong to T
5. HARMONIC FUNCTIONS
19
Proof. The idea is simply to find a local holomorphic primitive of f . Thus, assume
D(0, r) is a small disk and set
Z 1
Z z
f (tz) dt
f () d = z
F (z) :=
0
for all |z| < r. Then by our assumption, for |z| < r and h small,
Z 1
F (z + h) F (z)
f (z + ht) dt f (z)
=
h
0
as h 0. This shows that F H(D(0, r)) and therefore also F = f H(D(0, r)).
Hence f H() as desired.

Next, we introduce harmonic functions.
5. Harmonic functions
Definition 1.25. A function u : C is called harmonic iff u C 2 () and
u = 0.
Typically, harmonic functions are taken to be real-valued but there is no need to
make this restriction in general. The following result explains the ubiquity of harmonic
functions in complex analysis.
Proposition 1.26. If f H(), then Re (u), Im (v) are harmonic in .
Proof. First, u := Re (f ), v := Im (f ) C () by analyticity of f . Second, by the
CauchyRiemann equations,
uxx + uyy = vyx vxy = 0,
as claimed.
vxx + vyy = uyx + uxy = 0

This motivates the following definition.

Definition 1.27. Let u be harmonic on and realvalued. We say that v is the
harmonic conjugate of u iff v is harmonic and realvalued on and u + iv H().
Let us first note that a harmonic conjugate, if it exists, is unique up to constants:
indeed, if not, then we would have a real-valued harmonic function v on so that
iv H(). But from the CauchyRiemann equations we would then conclude that
v = 0 or v = const by connectedness of .
This definition of course presents us with the question whether every harmonic function on a region of R2 has a harmonic conjugate function. The classical example for the
failure of this is u(z) = log |z| on C ; the unique harmonic conjugate v with v(1) = 0
would have to be the polar angle which is not defined on C . However, in view of Proposition 1.21 it is defined on harmonic on every simply connected subdomain of C . As the
following proposition explains, this is a general fact.
Proposition 1.28. Let be simply connected and u real-valued and harmonic on
. Then u = Re (f ) for some f H() and f is unique up to an additive imaginary
constant.
20
Proof. We already established the uniqueness property. To obtain existence, we

need to solve the CauchyRiemann system. In other words, we need to find a potential
v to the vector field (uy , ux ) on , i.e., v = (uy , ux ). If v exists, then it is C 2 ()
and
v = uyx + uxy = 0
hence v is harmonic. Define
v(z) :=
z
z0
uy dx + ux dy
where the line integral is along a curve connecting z0 to z which consists of finitely many
line segments, say. If is a closed curve of this type in , then by Greens theorem,
ZZ
I
(uyy + uxx ) dxdy = 0
uy dx + ux dy =
where U = (this requires to be simply connected). So the line integral defining v

does not depend on the choice of curve and v is therefore well-defined on . Furthermore,
as usual one can check that v = (uy , ux ) as desired. A quick but less self-contained
proof is as follows: the differential form
:= uy dx + ux dy
is closed since d = u dx dy = 0. Hence, it is locally exact and by simple connectivity
of , exact on all of . In other words, = dv for some smooth function v on as
desired.

From this, we can easily draw several conclusions about harmonic functions. We
begin with the important observation that a conformal change of coordinates preserves
harmonic functions.
Corollary 1.29. Let u be harmonic in and f : 0 holomorphic. Then u f
is harmonic in 0 .
Proof. Locally around every point of there is v so that u+iv is holomorphic. Since
the composition of holomorphic functions is again holomorphic, the statement follows.
There is of course a direct way of checking this: since = 4z z one has from the chain
rule (1.2)
z (u f ) = (w u) f z f + (w u) f zf = (w u) f f
and thus furthermore
(u f ) = 4zz (u f ) = 4(w w u) f |f |2 = |f |2 (u) f

whence the result.
Next, we describe the well-known mean value and maximum properties of harmonic
functions. We can motivate them in two ways: first, they are obvious for the onedimensional case since then harmonic functions on an interval are simply the linear
ones; second, in the discrete setting (i.e., on the lattice Z2 and similarly on any higherdimensional lattice), the harmonic functions u : Z2 R are characterized by
1 X
u(m)
u(n) =
4
|nm|=1
6. THE WINDING NUMBER
21
where the sum is over the four nearest neighbors (thus, | | is the 1 (Z2 ) metric). The
reader will easily verify that this implies
1 X
1 X
u(m) =
u(m)
u(n) =
8
16
|nm|=2
|nm|=3
and so forth. In other words, the mean value property over the nearest neighbors extends
to larger 1 balls.
Corollary 1.30. Let u be harmonic on . Then u C (), u satisfies the meanvalue property
Z 1
(1.9)
u(z0 ) =
u(z0 + re2it ) dt
r < dist(z0 , )
0
and u obeys the maximum principle: if u attains a local maximum or minimum in ,

then
then u is constant. In particular, if is bounded and u C(),
min u u(z) max u
where equality can be attained only if u is constant.

Proof. Let U be simply connected, say a disk. By Proposition 1.28, u = Re (f )
where f H(U ). Since f C (U ), so is u. Moreover,
Z 1
I
f (z)
1
f (z0 + re2it ) dt
dz =
f (z0 ) =
2i z z0
0
Passing to the real part proves (1.9). For the maximum principle, suppose that u attains
a local extremum on some disk in . Then it follows from (1.9) that u has to be
constant on that disk. Since any two points in are contained in a simply connected
subdomain of , we conclude from the existence of conjugate harmonic functions on
simply connected domains as well as the uniqueness theorem for analytic functions that
u is globally constant.

It is not too surprising that both these properties by themselves, i.e., the mean value
property as well as the maximum property, already characterize harmonic functions.
6. The winding number
Let us apply the procedure of the proof of Proposition 1.28 to u(z) = log |z| on C .
Then, with r (t) = reit ,
Z 2
I
(sin2 (t) + cos2 (t)) dt = 2.
uy dx + ux dy =
r
This is essentially the same calculation as (1.5) with n = 1. Indeed, on the one hand,
the differential form
x
y
= 2 dx + 2 dy
r
r
pulls back to any circle as the form d this of course explains the appearance of 2.
On the other hand, the local primitive of z1 is (any branch of) log z. So integrating over
a loop that encircles the origin once, we create a jump by 2. On the one hand, this
property shows that log |z| does not have a conjugate harmonic function on C and on
the other, it motivates the following definition.
22
Lemma 1.31. Let : [0, 1] C be a closed curve. Then for any z0 C \ ([0, 1]) the
integral
I
1
dz
n(; z0 ) :=
2i z z0
is an integer. It is called the index or winding number of relative to z0 . It is constant
on each component of C \ ([0, 1]) and vanishes on the unbounded component.
Proof. Let
g(t) :=
d
ds
(s)
ds.
(s) z0
The integrand equals log((s) z0 ) for an arbitrary branch of log. In fact,

d g(t)
e
((t) z0 ) = eg(t) (t) + eg(t) (t) = 0
dt
which implies that
eg(1) ((1) z0 ) = eg(0) ((0) z0 ) = eg(0) ((1) z0 )
and thus eg(0)g(1) = 1; in other words, g(0) g(1) 2iZ as claimed. To establish the
constancy on the components, observe that
I
I
I
d
dz
dh 1 i
dz
dz = 0.
=
=
2
dz0 z z0
(z z0 )
dz z z0
Finally, on the unbounded component we can let z0 to see that the index vanishes.

PJ
This carries over to cycles of the form c = j=1 nj j where each j is a closed curve
and nj Z. If nj < 0, then nj j means that we take |nj | copies of j with the opposite
orientation. The index of a cycle c relative to a point z0 not on the cycle is simply
n(c; z0 ) :=
J
X
nj n(j ; z0 ).
j=1
Observe that
I
I
1
1
n(c; z0 ) =
d log( z0 ) =
dz0
2i c
2 c
where z0 is the argument relative to the point z0 . The real part of log( z0 ) does not
contribute since c is made up of closed curves. The differential form (we set z0 = 0)
y
x
d0 = 2 dx + 2 dy
r
r
is closed but not exact. In fact, it is essentially the only form with this property in the
domain C = R2 \ {0}. To understand this, note that a closed form on a domain is
exact if and only if
I
=0
closed curves c .
(1.10)
c
Indeed, it is clearly necessary; for the sufficiency set

Z z
f (z) :=
z0
6. THE WINDING NUMBER
23
where the integral is along an arbitrary path in connecting z0 to z. It is well-defined

due to the vanishing condition (1.10) and satisfies df = . Now let be an arbitrary
closed form on C and set
I
.
d0 , :=
e :=
2
[|z|=1]
Then de
= 0 and (1.10) holds due to the homotopy invariance of integrals of closed forms
(Stokess theorem). Finally, this implies that the map 7 is one-to-one on the space
closed forms
,
H1 (C ) :=
exact forms
so we have established the well-known fact that H1 (C ) R. Incidentally, it also follows
that
[c] 7 n(c; 0),
(1.11)
[c] 1 (C )
is an isomorphism of the fundamental group onto Z (for this, the cycles c have to be
rooted at some fixed base-point), see Problem 1.16.
Let us repeat this analysis on the space X := R2 \ {zj }kj=1 where zj C are distinct
and k 2. As before, let on X be a closed form and set
I
k
X
j
e :=
dz , j =
2 j
[|zzj |=j ]
j=1
where j > 0 is so small that the disks D(zj , j ) are all disjoint. Then we again conclude
1
0
0
1
1
0
1
0
1
0
1
0
1
0
Figure 1.7. Bouquet of circles

that (1.10) holds and thus that
e is exact. Since the map
7 {j }kj=1
is a linear map from all closed forms on X onto Rk with kernel equal to the exact forms,
we have recovered the well-known fact
H1 (X) Rk .
We note that any closed curve in X is homotopic to a bouquet of circles, see Figure 1.7;
more formally, up to homotopy, it can be written as a word
ai11 ai22 ai33 . . . aimm
where i {1, 2, . . . , k}, Z, and a are circles around z with a fixed orientation. It
is an exercise in algebraic topology to prove from this that 1 (X) = ha1 , . . . , am i, the
free group with m generators (use van Kampens theorem). Finally, the map
[c] 7 {n(c; zj )}m
j=1 ,
1 (X) Zm
24
is a surjective homomorphism, but is not one-to-one; the kernel consists of all curves
with winding number zero around each point. See Figure 1.8 for an example with k = 2.
z1
z2
Figure 1.8. Zero homologous but not homotopic to a point

7. Problems
Problem 1.1. (a) Let a, b C and k > 0. Describe the set of points z C which
satisfy
|z a| + |z b| k.
(b) Let |a| < 1, a C. The plane {z C} is divided into three subsets according to
whether
za
w=
1a
z
satisfies |w| < 1, |w| = 1, or |w| > 1. Describe these sets (in terms of z).
P
Problem 1.2. Let P (z) = nj=0 aj z j be a polynomial of degree n 1 with all roots
P
j z j . Show
inside the unit circle |z| < 1. Define P (z) = z n P (z 1 ) where P (z) = nj=0 a
that all roots of
P (z) + P (z) = 0
lie on the unit circle |z| = 1. Do the same for P (z) + ei P (z) = 0, with R arbitrary.
Problem 1.3. P
Suppose p0 > p1 > p2 > > pn > 0. Prove that all zeros of the
polynomial P (z) = nj=0 pj z j lie in {|z| > 1}.
1 +ix2
Problem 1.4. Let : S 2 C be the stereographic projection (x1 , x2 , x3 ) 7 x1x
.
3
(a) Give a detailed proof that is conformal. (b) Define a metric d(z, w) on C as
the Euclidean distance of 1 (z) and 1 (w) in R3 . Find a formula for d(z, w). In
particular, find d(z, ). (c) Show that circles on S 2 go to circles or lines in C under .
Problem 1.5. Find a M

obius transformation that takes |z i| < 1 onto |z 2| < 3.
Do the same for |z + i| < 2 onto x + y 2. Is there a M
obius transformation that takes
{|z i| < 1} {|z 1| < 1}
onto the first

quadrant? How about {|z 2i| < 2} {|z 1| < 1} and {|z
2} {|z + 3| < 2} onto the first quadrant?
3| <
7. PROBLEMS
25
Problem 1.6. Let {zj }nj=1 C be distinct points and mj > 0 for 1 j n. Assume
Pn
j=1 mj = 1 and define z =
j=1 mj zj . Prove that every line through z separates the
points {zj }nj=1 unless all of them are co-linear. Here separates means that there are
points from {zj }nj=1 on both sides of the line (without being on ).
Pn
Problem 1.7. (a) Suppose {zj }

{z C : Re z 0} is a given sequence. True
P
P j=1
P
2
2
or false: if both j=1 zj and j=1 zj converge, then
j=1 |zj | also converges.
(b) True or false: there are sequences of complex numbers {zj }
j=1 such that for each
P k
integer k 1 the infinite series j=1 zj converges, but fails to converge absolutely.
Problem 1.8. Find the holomorphic function f (z) = f (x + iy) with real part
and so that f (0) = 0.
x(1 + x2 + y 2 )
1 + 2x2 2y 2 + (x2 + y 2 )2
Problem 1.9. Discuss the mapping properties of z 7 w = 12 (z + z 1 ) on |z| < 1.

Is it one-to-one there? What is the image of |z| < 1 in the w-plane? What happens on
|z| = 1 and |z| > 1? What is the image of the circles |z| = r < 1, and of the half rays
Arg z = emanating from zero?
obius transformation.
Problem 1.10. Let T (z) = az+b
cz+d be a M
(a) Show that T (R ) = R iff we can choose a, b, c, d R. (b) Find all T such that
T (T) = T, where T = {|z| = 1} is the unit circle. (c) Find all T for which T (D) = D,
where D = {|z| < 1} is the unit disk.
Problem 1.11. Let f H(D) with |f (z)| < 1 for all z D.
a) If f (0) = 0, show that |f (z)| |z| on D and |f (0)| 1. If |f (z)| = |z| for some
z 6= 0, or if |f (0)| = 1, then f is a rotation.
b) Without any assumption on f (0), prove that
f (z ) f (z )
|z1 z2 |

1
2
z1 , z2 D
(1.12)

|1 z1 z2 |
1 f (z1 )f (z2 )
and
(1.13)
|f (z)|
1
2
1 |f (z)|
1 |z|2
z D.
Show that equality in (1.12) for some pair z1 6= z2 or in (1.13) for some z D implies
that f (z) is a fractional linear transformation.
Problem 1.12. a) Let f H() be one-to-one. Show that necessarily f (z) 6= 0
everywhere in , that f () is open (do you need one-to-one for this? If not, what do
you need?), and that f 1 : f () is also holomorphic. Such a map is called a biholomorphic map between the open sets and f (). If f () = , then f is also called
an automorphism.
b) Determine all automorphisms of D, H, and C.
Problem 1.13. Endow H with the Riemannian metric
1
1
dzd
z
ds2 = 2 (dx2 + dy 2 ) =
y
(Im z)2
26
and D with the metric
4
dzd
z
(1 |z|2 )2
These Riemannian manifolds, which turn out to be isometric, are known as hyperbolic
space. By definition, for any two Riemannian manifolds M, N a map f : M N is
called an isometry if it is one-to-one, onto, and preserves the metric.
(a) The distance between any two points z1 , z2 in hyperbolic space (on either D or H)
is defined as
Z
ds2 =
d(z1 , z2 ) = inf
k (t)k dt
where the infimum is taken over all curves joining z1 and z2 and the length of is
determined by the hyperbolic metric ds. Show that any holomorphic f : D D or
holomorphic f : H H satisfies
d(f (z1 ), f (z2 )) d(z1 , z2 )
for all z1 , z2 in hyperbolic space.

(b) Determine all orientation preserving isometries of H to itself, D to itself, as well
as from H to D.
(c) Determine all geodesics of hyperbolic space as well as its scalar curvature (we are
using the terminology of Riemannian geometry).
Problem 1.14. Let f H(D) be such that Re f (z) > 0 for all z D, and f (0) =
a > 0. Prove that |f (0)| 2a. Is this inequality sharp? If so, which functions attain it?
Problem 1.15. Prove Goursats theorem: if f is complex differentiable in (but
without assuming that f is continuous), then f H().
Problem 1.16. Show that two closed loops in C are homotopic if and only if they
have the same winding number around the origin. This proves that the map (1.11) is an
isomorphism.
CHAPTER 2
From z to the Riemann mapping theorem: some finer

points of basic complex analysis
1. The winding number version of Cauchys theorem
We now use the notion of winding number from the previous section to formulate
a more general version of Cauchys formula. We say that z c where c is a cycle iff z
lies on one of the curves that make up the cycle. In general, we write c both for the
cycle as well as the points on it. In what follows, we shall call a cycle c in a region
a 0homologous cycle in or relative to , iff n(c; z) = 0 for all z C \ . From the
discussion at the end of the previous section we know that such a cycle is not necessarily
homotopic to a point (via a homotopy inside , of course), see Figure 1.8. On the other
hand, it is clear that a cycle homotopic to a point is also homologous to zero.
Theorem 2.1. Let c be a 0homologous cycle in . Then for any f H(),
I
f (z)
1
dz
(2.1)
n(c; z0 ) f (z0 ) =
2i c z z0
for all z0 \ c. Conversely, if (2.1) holds for all f H() and a fixed z0 \ c, then
c is a 0homologous cycle in .
Proof. Define
(z, w) :=
f (z)f (w)
zw
if z 6= w
f (z)
if z = w
Then by analyticity of f , (z, w) is analytic in z and jointly continuous (this is clear for
z 6= w and for z close to w Taylor expand in z around w). The set
:= {z C \ c | n(; z) = 0}
is open, = C, and very importantly, . This property is due to c being zero

homologous as well as the winding number being constant on all components of C \ c.
The function
H
c (z, w) dw if z
g(z) :=
H f (w)
if z
wz dw
is therefore welldefined and g H(C); for the former, note that for any z ,
I
I
I
f (w) f (z)
f (w)
(z, w) dw =
(2.2)
dw =
dw 2if (z)n(c; z)
w
z
w
z
c
c
c
with n(c; z) = 0 for all z . The analyticity of g on is clear, whereas on it

follows from Fubinis and Moreras theorems. Finally, since g(z) 0 as |z| , we see
that g 0 on C. The theorem now follows from (2.2).
27
28
2. BASIC COMPEX ANALYSIS II
1
For the converse, fix any z1 C \ and apply (2.1) to f (z) = zz
. Then f H()
1
and therefore
I
1
1
dz
n(c; z0 )f (z0 ) =
2i c (z z0 )(z z1 )
I h
1
1 i
1
dz
= f (z0 )
2i c z z0 z z1
= f (z0 )n(c; z0 ) n(c; z1 )f (z0 )
whence n(c; z1 ) = 0 as claimed.
We can now derive the following more general version of Cauchys Theorem (cf. Theorem 1.16). As the reader will easily verify, it is equivalent to Theorem 2.1.
Corollary 2.2. With c and as in Theorem 2.1,
I
f (z) dz = 0
(2.3)
c
for all f H(). In particular, if is simply connected, then (2.3) holds for all cycles
in and f H().
Proof. Apply the previous theorem to h(z) = (z z0 )f (z) where z0 \ c. As for
the second statement, it uses the fact that C \ is connected if is simply connected
(and conversely). But then n(; z) = 0 for all z C \ by Lemma 1.31, and we are
done.

The final formulation of Cauchys theorem is the homotopy invariance. We say
that two C 1 cycles c1 and c2 are homotopic iff each closed curve from c1 (counted with
multiplicity) is C 1 homotopic to exactly one closed curve from c2 .
Theorem 2.3. Let c1 and c2 be two cycles in that are C 1 homotopic. Then
I
I
f (z) dz
f (z) dz =
c1
c2
for all f H(). In particular, if c is homotopic to a sum of points, then

for all f H().
c f (z) dz
=0
Proof. By summation, it suffices to consider closed curves instead of cycles. For

the case of closed curves one can apply Theorem 1.16 and we are done.

This is a most important statement, as it implies, for example, that the winding
number is homotopy invariant (a fact that we deduced from the homotopy invariance of
integrals of closed forms before); in particular, if a cycle c is 0homologous relative
to , then any cycle homotopic to c relative to is also 0homologous. As already noted
before the converse of this is false, see Figure 1.8.
We remark that Theorem 2.3 can be proven with continuous curves instead of C 1 . For
this, one needs to define the integral along a continuous curve via analytic continuation
of primitives. In that case, Theorem 2.3 becomes a corollary of the monodromy theorem;
see Theorem 2.16 below.
2. ISOLATED SINGULARITIES AND RESIDUES
29
2. Isolated singularities and residues

We now consider isolated singularities of holomorphic functions.
Definition 2.4. Suppose f H(\{z0 }) where z0 . Then z0 is called an isolated
singularity of f . We say that z0 is removable if f can be assigned a complex value at z0
that renders f holomorphic on . We say that z0 is a pole of f provided f (z) as
z z0 . Otherwise, z0 is called an essential singularity.
An example of a pole at z = 0 is exhibited by 1z , whereas an example of an essential
1
singularity at zero is given by e z . Indeed, simply consider the behavior of the latter
function as z 0 along the imaginary and real axes, respectively.
We will now give some criteria which allow one to characterize these different types
of isolated singularities. As usual, D denotes a disk D with the center removed.
Proposition 2.5. Suppose f H( \ {z0 }). Then there is the following mutually
exclusive trichotomy:
z0 is removable iff limzz0 (z z0 )f (z) = 0
z0 is a pole iff there exists a positive integer n 1 and h H() with h(z0 ) 6= 0
h(z)
such that f (z) = (zz
n
0)
z0 is essential iff for every > 0, the set f (D(z0 , ) ) is dense in C
Proof. Suppose limzz0 (z z0 )f (z) = 0. Then g(z) := (z z0 )f (z0 ) C() and
from Moreras theorem it follows that g H(). To apply Moreras theorem, distinguish
the cases where z0 lies outside the triangle, on the boundary of the triangle, or in the
interior of the triangle.
1
has a removable
Suppose that z0 is a pole. Then by the previous criterion, g(z) := f (z)
singularity at z0 (in fact, g(z) 0 as z z0 ). Hence, for some positive integer n there
is the representation g(z) = (z z0 )n ge(z) where e
g H() and e
g(z0 ) 6= 0. This implies
h(z)
that f (z) = (zz0 )n where h(z0 ) 6= 0 and h H(). Conversely, suppose that f (z) has
this form. Then f (z) as z z0 (which is equivalent to |f (z)| as z z0 ) and
z0 is a pole of f .
Finally, suppose f (D(z0 , )) D(w0 , ) = for some > 0 and w0 C, > 0. Then
1
f (z)w0 H(D(z0 , )) has a removable singularity at z0 which then further implies that
f (z) has a removable singularity or a pole at z0 . In the converse direction, the density
of the sets f (D(z0 , )) for every > 0 clearly precludes a removable singularity of pole
at z0 .

Let n be the integer arising in the previous characterization of a pole; then we say
that the order of the pole at z0 is n. We also remark that the characterization of essential
singularities in Proposition 2.5 is referred to as Casorati-Weierstrass theorem. The great
Picard theorem in fact states that for every > 0 such a function necessarily assumes
every value with one possible exception infinitely often on D(z0 , ) .
Definition 2.6. We say that f is a meromorphic function on iff there exists a
discrete set P such that f H( \ P) and such that each point in P is a pole of f .
We denote the field of meromorphic functions by M().
A standard and very useful tool in the study of isolated singularities are the Laurent
series.
30
Proposition 2.7. Suppose that f H(A) where

A = {z C | r1 < |z z0 | < r2 },
0 r1 < r2
is an annulus. Then there exist unique an C such that
X
an (z z0 )n
(2.4)
f (z) =
n=
where the series converges absolutely on A and uniformly on compact subsets of A.

Furthermore,
I
f (z)
1
dz
(2.5)
an =
2i |z|=r (z z0 )n+1
for all n Z and any r1 < r < r2 . The series (2.4) is called the Laurent series of f
around z0 , and
1
X
an (z z0 )n
n=
is its principal part.
Proof. Fix z . Let c be the cycle defined in by

c = r1 + r2
where r (t) := z0 + re2it , (t) := z + e2it , and r1 < r1 < |z z0 | < r2 < r2 and is
small. Then n(c; w) = 0 for all w C \ A and n(c; z) = 0. Hence, by the Cauchy formula
of Theorem 2.1,
I
I
I
f (w)
1
f (w)
1
f (w)
1
dw = 0 = f (z) =
dw
dw.
(2.6)
2i c w z
2i r w z
2i r w z
2
Now proceed as in the proof of Corollary 1.18 with contributing the an , n 0 as in

(2.5), and the inner curve r1 contributing an with n < 0 as in (2.5). Indeed, we simply
expand:
X
1
(z z0 )n
1
=
=
if |w z0 | > |z z0 |
wz
w z0 (z z0 )
(w z0 )n+1
r2
n=0
(w z0 )n
1
1
=
=
wz
w z0 (z z0 )
(z z0 )n+1
n=0
if |w z0 | < |z z0 |
Inserting these expansions into (2.6) and interchanging summation and integration yields
the desired representation (the interchange being justified by the uniform convergence
of these series on the integration curves). The absolute and uniform convergence of the
resulting series on compact sets follow as well. Note that these formulas, as well as
the uniqueness, follow from our previous calculation (1.5) (divide the Laurent series by
(z z0 ) and integrate).

Suppose now that r1 = 0 so that z0 becomes an isolated singularity. Amongst all
Laurent coefficients, a1 is the most important due to its invariance properties (this will
only become clear in the context of differential forms on Riemann surfaces). It is called
the residue of f at z0 and denoted by res(f ; z0 ). It is easy to read off from the Laurent
series which kind of isolated singularity we are dealing with:
2. ISOLATED SINGULARITIES AND RESIDUES
31
Corollary 2.8. Suppose z0 is an isolated singularity of f and suppose
X
an (z z0 )n
f (z) =
n=
is the Laurent expansion of f around z0 convergent on 0 < |z z0 | < for some > 0.
Then z0 is removable iff an = 0 for all n < 0. It has a pole iff there exists an integer
n0 < 0 such that an = 0 for all n < n0 but an0 6= 0 (and n0 is the order of the pole).
Otherwise, z0 is an essential singularity.
Proof. Simply apply Proposition 2.5 to the Laurent series of f at z0 .
Let us now clarify why holomorphic maps C C are necessarily rational. Recall
that this fact was already mentioned in Chapter 1 in connection with fractional linear
transformations.
Lemma 2.9. The analytic maps C C which are not identically equal to are
precisely the rational functions, i.e., all maps of the form PQ with P, Q polynomials over C
and Q 6 0.
Proof. All rational maps are analytic from the extended plane to itself. For the
converse, suppose f (z) C for all z C . Then f is entire and bounded and thus
constant. We can therefore assume that f (z0 ) = for some z0 C (consider f (1/z)
if necessary). By continuity of f the point z0 cannot be an essential singularity of f ,
cf. Proposition 2.5. In other words, z0 is a removable singularity or a pole. By the
uniqueness theorem, the poles cannot accumulate in C . Since the latter is compact,
there can thus only be finitely many poles. Hence, after subtracting the principal part
of the Laurent series of f around each pole in C from f , we obtain an entire function
which grows at most like a polynomial. By Liouvilles theorem, see Corollary 1.19, such
a function must be a polynomial and we are done.

A most useful result of elementary complex analysis is the residue theorem.
Theorem 2.10. Suppose f H(\{zj }Jj=1 ). If c is a 0homologous cycle in which
does not pass through any of the zj , then
I
J
X
1
n(c; zj )res(f ; zj ).
f (z) dz =
(2.7)
2i c
j=1
Proof. Let j := n(c; zj ) and define a new cycle

c := c
J
X
j j ,
j (t) := zj + e2it
j=1
where > 0 is small. Then n(c ; w) = 0 for all w C\ and n(c ; zj ) = 0 for all 1 j J.
The residue formula (2.7) now follows from Theorem 2.1 applied to \ {zj }Jj=1 .

The residue theorem can be used to evaluate definite integrals; see Problem 2.3. We
will apply it now to derive the argument principle. To motivate this principle, consider
f (z) = z n with n = 0. If r (t) = re2it is the circle of radius r around 0, then f r has
winding number n around 0. Hence, that winding number counts how many zeros of f
there are inside of r . If n < 0, then we obtain the order of the pole at 0 with a negative
sign.
32
1
0
11
00
11
00
c
3
11
00
Figure 2.1. An example of the cycle in the residue theorem

Proposition 2.11. Let c be a 0homologous cycle relative to . If f M() is
such that no zero or pole of f lie on c, then
X
X
(2.8)
n(f c ; 0) =
n(c; z)
n(c; )
z:f (z)=0
:f ()=
where zeros and poles are counted with multiplicity. In other words, the winding number
or increase in the argument of f along c counts zeros minus poles with multiplicity
and weighted by the winding number of c around the respective points.
Proof. We first point out the sum on the right-hand side of (2.8) only has finitely
many nonzero terms; indeed, zeros and poles can only cluster at the boundary where the
winding number necessarily vanishes. By definition,
I
I
f (z)
1
1
dw
=
dz.
(2.9)
n(f c ; 0) =
2i
w
2i
f (z)
f c
If f (z) = (z z0 )n g(z) with n 6= 0, g(z0 ) 6= 0, and g H(), then

f
; z0 = n
res
f
and the proposition follows by applying the residue theorem to (2.9).
It is clear that the argument principle gives another (direct) proof of the fundamental
theorem of algebra. Combining the homotopy invariance of the winding number (see
Theorem 2.3) with the argument principle yields Rouches theorem.
Proposition 2.12. Let c be a 0-homologous cycle in such that
{z C \ c : n(c; z) = 1} = 0
3. ANALYTIC CONTINUATION
33
has the property

{z C \ c : n(c; z) = 0} = C \ (0 c).
Let f, g H() and suppose that |g| < |f | on c. Then
#{z 0 | f (z) = 0} = #{z 0 | (f + g)(z) = 0}
where the zeros are counted with multiplicity.
Proof. The function (f + sg) c, 0 s 1 is a homotopy between the cycles f c

and (f + g) c relative to C with the property that Proposition 2.11 applies to each
sslice (note that in particular, f 6= 0 on c). Consequently,
X
X
n((f + sg) c; 0) =
n(c; z) =
1
z:(f +sg)(z)=0
z0 :(f +sg)(z)=0
= #{z 0 | (f + sg)(z) = 0}
does not depend on s and Rouches theorem follows.
Rouches theorem allows for yet another proof of the fundamental theorem of algebra:
If P (z) = z n + an1 z n1 + . . . + a1 z + a0 , then set f (z) := z n and g(z) := an1 z n1 +
. . . + a1 z + a0 . On |z| = R with R very large, |f | > |g| and Rouches theorem applies.
Many special functions, such as the Gamma and zeta functions, are defined by integral
or series representations in subdomains of the complex plane (such as a half-plane). The
question then arises whether these functions can be analytically continued outside of
this domain. Historically, this question turned out to be of fundamental importance
to complex analysis with many ramifications to other areas of mathematics. In fact,
Riemann surfaces appeared as the natural domains of analytic functions obtained by
analytic continuation of roots of algebraic equations.
In this chapter, we discuss the most elementary aspects of this theory and we begin with analytic continuation along curves. First, we define a chain of disks along a
continuous curve. Next, we will put analytic functions on the disks which are naturally
continuations of one another.
Definition 2.13. Suppose : [0, 1] is a continuous curve inside a region .
We say that Dj = D((tj ), rj ) , 0 j J, is a chain of disks along in iff
0 = t0 < t1 < t2 < . . . < tN = 1 and ([tj , tj+1 ]) Dj Dj+1 for all 0 j N 1.
For any and as in this definition there exists a chain of disks along in , by
uniform continuity of . Next, we analytically continue along such a chain.
Definition 2.14. Let : [0, 1] be a continuous curve inside . Suppose f
H(U ) and g H(V ) where U and V are neighborhoods of p := (0) and
q := (1), respectively. Then we say that g is an analytic continuation of f along iff
there exists a chain of disks Dj := D((tj ), rj ) along in where 0 j J, and fj in
H(Dj ) such that fj = fj+1 on Dj Dj+1 and f0 = f and fJ = g locally around p and q,
respectively.
In what follows, the only relevant information about f and g is their definition
locally at p and q, respectively, and not their domains of definition. This is equivalent
to saying that we identify f and g with their Taylor series around p and q, respectively.
34
11
00
00
11
00
11
11
00
00
11
1
0
1
0
11
00
11
00
Figure 2.2. A chain of disks

As expected, the analytic continuation of f along is unique whenever it exists. In
particular, it does not depend on the chain of disks along , but only on itself. This
follows from the uniqueness theorem, see Corollary 1.22 above.
Lemma 2.15. The analytic continuation g of f along as in Definition 2.14 only
depends on f and , but not on the specific choice of the chain of circles. In particular,
it is unique.
e k are two different chains of disks along with
Proof. Suppose that Dj and D
J
underlying partitions {tj }j=1 and {sk }K
k=1 , respectively. Denote the chain of analytic
functions defined on these disks by fj and gk . Then we claim that for any j, k with
tj1 sk tj ,
ek.
fj = gk on Dj D
Applying this claim to the end point of yields the desired uniqueness. To prove the
claim, one uses induction on j + k and the uniqueness theorem. As an exercise, supply
the details.

We have already encountered a special case of this: suppose that f H(). Then
locally around every point in there exists an anti-derivative (or primitive). Any such
primitive can be analytically continued along an arbitrary C 1 -curve : [0, 1] by
integration:
Z
F (z) := f () d
where (1) = z and (0) = z0 is kept fixed. This procedure, however, does not necessarily
lead to a global primitive F H(). The standard example = C and f (z) = z1
shows otherwise. On the other hand, it is clear from Theorem 2.3 that we do obtain a
global F if is simply connected. This holds in general for analytic continuations and is
known as the monodromy theorem.
35
Theorem 2.16. Suppose 0 and 1 are two homotopic curves (relative to some region
C) with the same initial point p and end point q. Let U be a neighborhood of p and
assume further that f H(U ) can be analytically continued along every curve of the
homotopy. Then the analytic continuations of f along j , j = 0, 1 agree locally around q.
Proof. Let H : [0, 1]2 be the homotopy between 0 and 1 which fixes the initial
and endpoints. Thus H = H(t, s) where 0 (t) = H(t, 0) and 1 (t) = H(t, 1), respectively.
Denote the continuation of f along H(, s) by gs . We need to prove that the Taylor series
of gs around q does not depend on s. It suffices to prove this locally in s. The idea is
of course to change s so little that essentially the same chain of disks can be used. The
details are as follows: let s (t) := H(t, s), fix any s0 [0, 1] and suppose {Dj }Jj=1 is
a chain of circles along s0 with underlying partition 0 = t0 < t1 < . . . < tN = 1
and functions fj on Dj defining the analytic continuation of f along s0 . We claim the
following: let Dj (s) denote the largest disk centered at s (tj ) which is contained in Dj .
There exists > 0 such that for all s [0, 1], |s s0 | < , the Dj (s) form a chain of disks
along s . In that case, we can use the same fj , which proves that for all |s s0 | < ,
the gs agree with gs0 locally around q. It remains to prove the claim. For this, we use
the uniform continuity of the homotopy H to conclude that there exists > 0 so that
for all |s s0 | < , each disk Dj (s) contains the -neighborhood of s ([tj1 , tj ]) for each
1 j J. This of course guarantees that {Dj (s)}Jj=1 is a chain of disks along s inside
as desired.

In particular, since any two curves with the same initial and end points are homotopic in a simply connected region, we conclude that under the assumption of simple
connectivity, analytic continuations are always unique. This of course implies all previous results of this nature (the existence of the logarithm etc.). Any reader familiar
with universal covers should be reminded here of the homeomorphism between a simply
connected manifold and its universal cover. Making this connection between the monodromy theorem and the universal cover requires the notion of a Riemann surface to
which we turn in Chapter 4.
An instructive example of how analytic continuation is performed in practice, such
as in the context of special functions, is furnished by the Gamma function (z). If
Re (z) > 0, then
Z
et tz1 dt
(z) :=
0
is holomorphic. This follows from Fubinis and Moreras theorems. One checks via
integration by parts that (n + 1) = n! and the functional equation
(z + 1) = z(z)
Re (z) > 0.
Since the left-hand side is defined for all Re (z) > 1, we set
(z + 1)
, Re (z) > 1
z
Note that z = 0 is a pole of first order. Iterating this identity yields, with k 0 an
arbitrary integer,
(z) :=
(z) =
(z + k + 1)
z(z + 1)(z + 2) . . . (z + k)
Re (z) > k 1.
36
This allows one to analytically continue as a meromorphic function to all of C; it has

n
simple poles at {n Z : n 0} with residues (1)
n! . For details as well as other examples
we refer the reader to Problems 2.10 and 2.13.
Returning to general properties, let us mention that any domain carries analytic
functions which cannot be continued beyond any portion of the boundary . Simply
let {zn }
n=1 be dense in and define
f (z) =
n=1
2n
1
z zn
which is analytic on . It follows from Corollary 1.18 that a power series with finite and
positive radius of convergence cannot be analytically continued across its entire circle of
convergence. A natural class of power series with R = 1 and which cannot be continued
across any portion of |z| = 1 are the gap series, see Problem 2.11.
4. Convergence and normal families
The next topic we address is that of convergence and compactness of sequences of
holomorphic functions. The concept of complex differentiability is so rigid that it survives
under uniform limits this is no surprise as Moreras theorem characterizes it by means
of the vanishing of integrals.
Lemma 2.17. Suppose {fn }
n=1 H() converges uniformly on compact subsets of
(k)
to a function f . Then f H() and fn f (k) uniformly on compact subsets of for
each k 1. Furthermore, suppose that
sup #{z | fn (z) = w} N < .
n1
Then either f w in or
#{z | f (z) = w} N.
The cardinalities here include multiplicity.
Proof. The first assertion is immediate from Moreras theorem. The second one
follows from Cauchys formula
I
fn (z)
k!
(k)
dz
(2.10)
fn (z0 ) =
2i |zz0|=r (z z0 )k+1
whereas the third is a consequence of Rouches theorem: assume that f 6 w and let
{z | f (z) = w} {zj }Jj=1
with J < . Since the set on the left-hand side is discrete in , there exist , > 0 small
so that |f (z) w| > on each circle |z zj | = , 1 j J, cf. (1.8). Now let n0 be so
large that |f fn | < on |z zj | = for all n n0 and each 1 j J. By Rouches
theorem, it follows that f has as many zeros counted with multiplicity as each fn with
n n0 inside these disks and we are done.

The following proposition shows that Lemma 2.17 applies to any bounded family
{fn }
n=1 H(), or at least subsequences thereof. This isMontels normal family theorem; one says that F = {f }A H() is a normal family provided for each compact
5. THE MITTAG-LEFFLER AND WEIERSTRASS THEOREMS
37
K one has
(2.11)
sup sup |f (z)| <
zK A
This is equivalent to local uniform boundedness, i.e., each point in has a neighborhood K for which (2.11) holds.
Proposition 2.18. Suppose F = {f }A H() is a normal family. Then there
exists a sequence fn in F that converges uniformly on compact subsets of .
(k)
Proof. By (2.10), we see that f with A is uniformly bounded on compact

subsets of for all k 0. By the cases k = 0 and k = 1, we see that {f }A is
in particular equi-continuous and bounded. By the Arzela-Ascoli theorem and a diagonal subsequence argument we thus construct a subsequence converging uniformly on all
compact subsets of as desired.

5. The Mittag-Leffler and Weierstrass theorems
Let us now use this machinery to answer some fundamental and classical questions
in complex analysis:
Can we find f M(C) so that f has poles exactly at a prescribed sequence
{zn } that does not cluster in C, and such that f has prescribed principal parts
at these poles (this refers to fixing the entire portion of the Laurent series with
negative powers at each pole)?
Can we find f H(C) such that f has zeros exactly at a given sequence {zn }
that does not cluster with prescribed orders n 1?
In both cases the answer is yes, as we can easily see now.
Theorem 2.19 (Mittag-Leffler). Given {zn }N
n=1 C, with |zn | if N = , and
polynomials Pn with positive degrees, there exists f M(C) so that f has poles exactly
at zn and
1
f (z) Pn
z zn
is analytic around zn for each 1 n N .
Proof. If N is finite, there is nothing to do: simply define
f (z) :=
N
X
n=1
Pn
1
.
z zn
If N = , then we need to guarantee convergence of this series on compact sets by

making at most aholomorphic
error. Let Dn := {|z| < |zn |/2} and Tn (z) be the Taylor

1
polynomial of Pn zzn of sufficiently high degree so that

1

Tn (z) < 2n .
sup Pn
z zn
zDn
Then
h

X
Pn
n=1
i
1
Tn (z)
z zn
38
converges on compact subsets of C \ {zn }

n=1 and thus defines a holomorphic function

1
there. Moreover, the zn are isolated singularities; in fact, they are poles with Pn zz
n
as principal parts.

As an example of this procedure, let f (z) =
2
.
sin2 (z)
n)2 .
zn = n Z and principal part hn (z) = (z

Clearly,
X
hn (z)
Then f M(C) with poles
nZ
converges uniformly on C \ Z to a function s(z) holomorphic there. Moreover, s and

g := f s are both 1periodic. In addition, g H(C). Finally, in the strip 0 Re z 1
we see that both f and s are uniformly bounded; in fact, they both tend to zero as
|Im z| . Hence, g H(C) L (C) is bounded and in fact vanishes identically. In
conclusion,
X
1
2
=
.
2
sin (z) n= (z n)2
Setting z =
1
2
shows that
n=1
n=1
1
2 X 1
2 X
=
and
thus
=
8
(2n + 1)2
6
n2
where the second series is obtained from the first by splitting into even and odd n.
As another example, consider f (z) = cot(z). It has simple poles at each n Z
1
with principal parts hn (z) = zn
. In this case we do require the Tn from the proof of
Theorem 2.19:
h
X
1
1i 1 Xh 1
1 i
1
= +
+
+
s(z) := +
z n= z n n
z
zn z+n
=
1
+
z
n=1
n6=0
X
n=1
2z
.
z 2 n2
By inspection, g := f s is 1-periodic and analytic on C. A simple estimate of both f

and s reveals that g is bounded. Hence g = const, and expanding around z = 0 shows
that in fact g 0. We have thus obtained the partial fraction decomposition
(2.12)
1 X 2z
cot(z) = +
.
z
z 2 n2
n=1
Let us now turn to the second question, the construction of an entire function with
prescribed zeros.
Q
Definition 2.20. Given {zn }

n=1 C , we say that
n=1 zn converges, iff
PN :=
N
Y
n=1
zn P C .
P
We say that this product converges absolutely iff
n=1 |1 zn | < . We shall also allow
Q
z
with
all
but
finitely
many
z
=
6
0.
In
the
case that some zn = 0, this product
n
n
n=1
5. THE MITTAG-LEFFLER AND WEIERSTRASS THEOREMS
39
is defined to be 0 provided the infinite product with all zn = 0 removed converges in the
previous sense.
Here is an elementary lemma whose proof we leave to the reader. Log z denotes the
principal branch of the logarithm, i.e.,
Log z := log |z| + iArg z,
Arg z [0, 2).
Q
P
Lemma 2.21. Let {zn }

(absolutely) iff
n=1 C . Then
n=1 zn converges Q
n=1 Log zn
converges (absolutely). The notion of uniform convergence of zn P
relative to some complex parameter is also reduced to the same question for the series
log zn .
We can now easily answer the question concerning entire functions with prescribed
zeros.
Theorem 2.22 (Weierstrass). Let {zn }n C be a sequence Z (finite or infinite) that
does not accumulate in C. Then there exists an entire function f that vanishes exactly
at zn to the order which equals the multiplicity of zn in Z.
Proof. We set
n
X
1 z i
z
exp
1
(2.13)
f (z) := z
zn
zn
n
=1
Q
where is the number of times 0 appears in Z, whereas is the product with all zn = 0
deleted. The mn 0 are integers chosen so that
mn

1 z
z X

+
< 2n
Log 1
zn
zn
Y
h
=1
1
2 |zn |.
Given any R > 0, all but finitely many zn satisfy |zn | 2R. By our
on |z| <
construction, the (tail of the) infinite product converges absolutely and uniformly on
every disk to an analytic function. In particular, the zeros of f are precisely those of the
factors, and we are done.
As in the case of the MittagLeffler theorem, one typically applies the Weierstrass
theorem to give entire functions. Here is an example: let f (z) = sin(z) with zero set
Z = Z. The zeros are simple. In view of (2.13), we define

Y
Y
z2
z z

en = z
1 2 .
1
g(z) := z
n
n
n=1
nZ
There exists an entire function h such that f (z) = g(z)eh(z) . In other words:

Y
z2
1 2
f (z) = sin(z) = zeh(z)
n
n=1
f (z)
f (z)
= cot(z) = h (z) +
1 X 2z
+
z n=1 z 2 n2
By (2.12), h 0 or h = const. Expanding everything around z = 0 we conclude that

eh = and we have shown that

Y
z2
1 2
sin(z) = z
n
n=1
40
Setting z =
1
2
in particular yields the Wallis formula
Y
(2n)2
=
2
(2n + 1)(2n 1)
n=1
We remark that the expression in brackets appearing in (2.13) is called the canonical
factor and denoted by Emn (z/zn ). In other words,
1 2
+...+ k1 z k
Ek (z) = (1 z)ez+ 2 z
k 0.
It is natural to ask under what circumstances the numbers mn in the proof of Theorem 2.22 remain bounded. By the proof of Weierstrass theorem, this questions is tied
up with the problem of analyzing the distribution of zeros of entire functions. More
precisely, we will need to control the number of zeros in the disk D(0, R) as R .
As evidenced by the Jensen formula (3.11), this number is related to the growth of the
entire function at infinity.
Definition 2.23. An entire function is of finite order provided
|f (z)| AeB|z|
zC
for some constants A, B and 0. The infimum of all possible is called the order
of f .
This class of entire functions satisfies the more precise Hadamard factorization theorem:
Theorem 2.24. Let f H(C) with order at most 0. Let k = . Then
Y
Ek (z/)
f (z) = eP (z) z
C :f ()=0
where P is a polynomial of degree at most k and 0 is the order of vanishing of f at

z = 0.
Proof. We note that we can control the number of zeros of such functions that fall
into a disk of radius r. Indeed, from the Jensen formula (3.11), assuming f (z) 6= 0 if
|z z0 | = r and f (z0 ) 6= 0,
Z 1
X
log |f (z0 + re())| d log |f (z0 )| =
log(r/|z z0 |)
0
|zz0 |<r,f (z)=0
one concludes that

#{|z| < r : f (z) = 0} Cr +
r1
for any > 0. Considering shells {2j < |z| < 2j+1 } with arbitrary j 0, one infers from
this estimate that
X
|z|b <
zC , f (z)=0
provided b > . But this of course implies that the canonical factor Ek suffices in the
Weierstrass theorem where k = (the largest integer which is ). FINISH THIS ; is
the degree of P = k or k+1?
6. THE RIEMANN MAPPING THEOREM
41
As an application of this theorem, we leave it to the reader to check that an entire

function with non-integral order must have infinitely many zeros. For further material
on entire functions of finite order we refer the reader to [39], [29], or [37].
To conclude the discussion of the Mittag-Leffler and Weierstrass theorems let us note
the following: given arbitrary disjoint sequences {zn } and {m } with no accumulation
point in C, there exists a function f M(C) with poles precisely at the {zn } and zeros
at the {m }. Moreover, the principal parts at the poles can be arbitrarily prescribed.
To see this, apply the Mittag-Leffler theorem which yields a function f0 M(C) with
the prescribed poles and principal parts. Then, divide out the zeros of f0 by means
of the Weierstrass function and then multiply the resulting function with another entire
function which has the prescribed zeros. In other words, there is no obstruction involving
zeros and poles. However, if we consider the compactified plane C , then there is an
important obstruction: first, we know that each (nonconstant) f M(C ) is a rational
function. It is easy to see from this representation that
#{z C : f (z) = 0} = #{z C : f (z) = }
where the zeros and poles are counted according to multiplicity (including a possible one
at ) and the cardinality here is finite. A better way of arriving at the same conclusion
is given by the argument principle:
n(f R ; 0) = #{z D(0, R) : f (z) = 0} #{z D(0, R) : f (z) = }
where R is a circle of radius R centered at the origin so that no zero or pole of f lies on
R . Taking the limit R now yields the desired conclusion. We leave it to the reader
to verify that this is the only obstruction to the existence of a meromorphic function
with the given number of zeros and poles. Of interest is also that in fact
#{z C | f (z) = w} = const
(counted with multiplicity) independently of w C in that case. We shall see later
that this is a general fact about analytic functions on compact Riemann surfaces (the
constancy of the valency or degree). We remark that both the MittagLeffler and Weierstrass theorems remain valid on regions C, see Conway [7], for example.
6. The Riemann mapping theorem
We now present the famous and fundamental Riemann mapping theorem. Later, it
will become part of the much wider uniformization theory of Riemann surfaces.
Theorem 2.25. Let C be simply connected and 6= C. Then there exists a
conformal homeomorphism f : D onto the unit disk D.
Proof. We first find such a map into D. Then we will maximize all such f to
select the desired homeomorphism. We may assume that 0 6 . By Proposition 1.21
e := (). Then is
there exists a branch of on which we denote by . Let
e
e
one-to-one and if w , then w 6 . Indeed, otherwise (z1 ) = w = (z2 ) with
e is open,
z1 , z2 would imply that z1 = z2 or w = w = 0 contrary to 0 6 . Since
we deduce that
e D(w0 , ) =
42
and observe that f is one-to-one

for some w0 C and > 0. Now define f (z) := (z)w
0
and into D. Henceforth, we assume that D and also that 0 (scale and translate).
Define
F := {f : D | f H() is one-to-one and f (0) = 0, f (0) > 0}.
Note in particular that f (0) is assumed to be real-valued. Then F 6= (since id F)

and F is a normal family; see Proposition 2.18. We claim that
s0 := sup f (0) > 0
f F
is attained by some f F. Indeed, let fn (0) s0 with fn F and fn f H()

(0) > 0 and, by the maximum
uniformly on compact subsets of . Then f (0) = 0, f
principle and the open mapping theorem, f : D since this map is clearly not
constant. Finally, from Lemma 2.17 we infer that f is also one-to-one and thus f F
as claimed.
g1
2
f
h
square root
F
4
Figure 2.3. The final step in the Riemann mapping theorem

It remains to prove that f is onto D. Suppose not, and let w0 D \ f () =: 1 .
Pick g1 Aut(D) (a Mobius transform) such that g1 (w0 ) = 0 and let 2 := g1 (1 ),
which
is simply connected.pIt therefore admits a branch of the square root, denoted by
. Let g2 Aut(D) take g1 (0) onto 0. By construction,
F := g2 g1 f
satisfies ei F F for suitable . The inverse of g2 g1 exists and equals the analytic
function
h(z) := g11 ((g21 (z))2 ) : D D
6. THE RIEMANN MAPPING THEOREM
43
which takes 0 to 0 and is not an automorphism of D. Hence, by the Schwarz lemma,

(0) which yields the desired
|h (0)| < 1. Since h F = f , we have h (0)F (0) = f
contradiction.

We refer to f as in the theorem as a Riemann map. It is clear that f becomes
unique once we ask for f (z0 ) = 0, f (z0 ) > 0 for any z0 (this can always be achieved).
It is clear that = C does not admit such a map (constancy of bounded entire functions).
Next, we address the important issue of the boundary behavior of the Riemann map.
Definition 2.26. We say that z0 is regular provided for all 0 < r < r0 (z0 )
{z C | |z z0 | = r} = {z0 + rei | 1 (r) < < 2 (r)}
for some 1 (r) < 2 (r) which are continuous in r. In other words, D(z0 , r) is an
arc for all small r > 0. We say that is regular provided all points of are regular.
This notion of regularity only applies to the Riemann mapping theorem (later we
shall encounter another potential theoretic notion of regularity at the boundary).
An example of a regular domain is a manifold with C 1 boundary and corners, see
below.
Theorem 2.27. Suppose is bounded, simply connected, and regular. Then any
D.
conformal homeomorphism as in Theorem 2.25 extends to a homeomorphism

Proof. Let f : D be a Riemann map. We first show that limzz0 f (z) exists
for all z0 , the limit being taken from within . Suppose this fails for some z0 .
Then there exist sequences {zn }

n=1 and {n }n=1 in converging to z0 and such that
f (zn ) w1 ,
f (n ) w2
as n . Here w1 6= w2 D. Let 1 be a continuous curve that connects the points
1
{f (zn )}
j
n=1 in this order and let 2 do the same with {f (n )}n=1 . Denote j := f
for j = 1, 2. Then j are continuous curves both converging to z0 . Let
zr D(z0 , r) 1 ,
r D(z0 , r) 2
where we identify the curves with their set of points. By regularity of z0 there exists an
arc cr D(z0 , r) with
Z
f (z) dz
f (zr ) f (r ) =
cr
which further implies that

2 Z
Z

f (z) dz
|f (zr ) f (r )|
2
cr
2r
2 (r)
1 (r)
(r)
(r)
|f (rei )| r d
2
|f (rei )|2 r d,
using Definition 2.26 and the Cauchy-Schwarz inequality. Dividing by r and integrating
over 0 < r < r0 (z0 ) implies that
ZZ
Z r0 (z0 )
dr
|f (z)|2 dz = area (D) <
|f (zr ) f (r )|2
r
44
contradicting that f (zr ) w1 and f (r ) w2 as r 0 where w1 6= w2 . Hence,

lim f (z)
zz0
D
of f . The continuity part
does exist and defines a continuous extension F :
here is implicit in the preceding; indeed, if it fails, then there would have to exist some
z0 and a sequence zn so that F (zn ) 6 F (z0 ) as n . Since we can find
zn which are arbitrarily close to zn this would then imply that f (zn ) 6 F (z0 ) even
though zn z0 . This contradicts the previous step and the continuity holds. Next,
apply the same argument to f 1 : D . This can be done since obviously D is regular
in the sense of Definition 2.26 and moreover, since any sequence zn converging to
z0 can be connected by a continuous curve inside indeed, use the continuity of
.
1 (r), 2 (r) in Definition 2.26. Therefore, f 1 extends to a continuous map G : D

Finally, it is evident that F G = IdD and G F = Id as desired.
The same statement applies to unbounded . In that case, we regard as a region
in C and call regular provided 0 is regular for 1 := {z 1 | z }. The following
obvious result gives some examples of regular domains.
is a C 1 manifold with boundary and
Lemma 2.28. Any region C so that
corners is regular. This means that for every z0 there exists a C 1 diffeomorphism
of a neighborhood U of z0 onto a disk D(0, r1 ) for some r1 = r1 (z0 ) > 0 and such that
( U ) = {rei | 0 < r < r1 , 0 < < 1 < 2}
A particularly important example of such regions are polygons. In that case the
conformal maps are given by the Schwartz-Christoffel formulas, see [37]. For example,
the map
Z z
d
Im z > 0
z 7
1
1 ,
0 2 ( 1) 2
takes the upper half-plane onto the half-strip
{z C | Re z 0, 0 < Im z < }
where 0 and 1 get mapped to the two finite vertices of the half-strip. In a similar spirit,
Z z
d
z 7
Im z > 0
1
1
1 ,
0 2 ( 1) 2 ( 2) 2
takes the upper half-plane onto a rectangle with 0, 1, and 2 being mapped onto three of
1
the vertices and the fourth vertex being the image of . The square roots (z a) 2 here
are defined to be positive when z > a and to take the upper half-plane into itself.
7. Runges theorem
We close this chapter with Runges theorem. It addresses the question as to whether
any f H() can be approximated on compact sets by a polynomial. Again, there is a
topological obstruction: f (z) = z1 cannot be approximated on 1 |z| 2 by polynomials
H
otherwise, |z|=1 dz
z = 0, which is false. However, on simply connected domains this
can be done. On general domains, it can be done by rational functions.
7. RUNGES THEOREM
45
Theorem 2.29. Let K C be compact. Any function holomorphic on a neighborhood

of K can be approximated uniformly on K by rational functions all of whose poles belong
to C \ K. In particular, if C \ K is connected, then the approximation is uniform by
polynomials.
Proof. Let f H(U ) where U K is open. Let > 0 be sufficiently small that all
squares with side-length which intersect K belong (together with their interior) entirely
to U . Now define a cycle cK by tracing out the boundaries of all square in an -grid with
the property that they intersect both K and U \ K. We can assume that there are no
degenerate configurations (squares that intersect K but not K). The cycle is made up
Figure 2.4. The cycle in Runges theorem
of finitely many curves which consist of finitely many line segments. Moreover, equipped
with the natural orientation, cK is 0-homologous relative to U , and the winding number
n(c; z) = 1 for all z K. It follows from Cauchys theorem that
I
f ()
1
d
z K.
f (z) =
2i c z
The integral on the righthand side can be approximated by a Riemann sum uniformly
on K and any such Riemann sum defines a rational function with poles on c and thus
in C \ K. To finish the theorem, we need to show that if C \ K is connected, then the
1
poles can be pushed to . In other words, we need to prove that f (z, ) := z
can be approximated uniformly by polynomials if 6 K. To this end, let C(K)
(a bounded linear functional on C(K)) that vanishes on all polynomials. We remark
that the polynomials are in general not dense in C(K) (see Proposition 2.18) so that
does not need to vanish. We claim, however, that (f (, )) = 0 for all C \ K. If
46
|| > supzK |z|, then this follows from

f (z, ) =

X
z n
n=0
z K.
Next, observe that (f (, )) is analytic in . Since C \ K is connected, the claim is

proved. By the HahnBanach theorem,
f (, ) span{p(z) | p C[z]} =: L
where the closure is with respect to C(K). Indeed, assume this fails. Then
(p + tf (, )) := t
p L, t C
defines a bounded linear functional on the span of L and f (, ) which vanishes on L

and does not vanish on f (, ). Extend it as a bounded functional to C(K) (without
increasing its norm which is (dist(f (, ), L))1 ) using the Hahn-Banach theorem.

Runges theorem has many deep consequences, such as the local Mittag-Leffler theorem, which we do not discuss here; for one application of Runges theorem see Problem 2.12.
8. Problems
Problem 2.1. (a) Suppose f H(D) satisfies |f (z)| M for all z D. Assume
further that f (z) vanishes at the points {zj }N
j=1 where 1 N . Prove that
m
Y
z zj

|f (z)| M

1 zj z
j=1
zD
for any 1 m N (or, if N = , then 1 m < N ).

(b) If N = and f 6 0, then show that
X
(1 |zj |) < .
j=1
Problem 2.2. Let z1 , z2 , . . . , zn C be distinct points. Suppose is a closed (large)

circle that contains these points in its interior and let f be analytic on a disk containing
. Then the unique polynomial P (z) of degree n 1 which satisfies P (zj ) = f (zj ) for all
1 j n is given by
I
1
f () () (z)
P (z) =
d
2i ()
z
provided (z) is a suitably chosen polynomial. Find and prove this formula.
Problem 2.3. In this exercise you are asked to use the residue theorem, Theorem 2.10
to evaluate several integrals.
8. PROBLEMS
47
First, compute the value of the following definite integrals.

Z
Z 2
Z
dx
x dx
dx
=?,
=?,
=?
2
4
4
1
+
x
1
+
x
1
+ x
Z
Z
sin x
1 cos x
dx =?,
dx =?
x
x2
0
0
Z
Z
Z
d
log x
(log x)3
=?
a
>
1,
=?,
dx =?
2
1 + x2
1 + x2
0 (a + cos )
0
0
Second, prove that
Z a1
Z
eat
x
dx =
dt =
,
0<a<1
t
1
+
x
1
+
e
sin
a
0
Z
2
2
R
ex e2ix dx = e
Z
2 sinh 2a
sin a
dx =
R, 0 < a < 1
e2ix
cosh
x
+
cos
a
sinh 2
Z
Z
cos x
x sin x
ea
dx
=
,
dx = ea , a > 0
2
2
2 + a2
x
+
a
a
x
Z 2ix
e
dx = (1 + 2||)e2|| ,
R
2 )2
(1
+
x
2
as well as
1 3 5 (2n 1)
dx
=
(1 + x2 )n+1
2 4 6 (2n)
d
2a
=
3 ,
2
(a + cos )
(a2 1) 2
d
2
=
,
2
a + b cos
a b2
a>1
a, b R, |a| > b
and finally, show that

Z
Z 2
log x
i
log |1 ae | d = 0,
dx =
log a,
2
2
x +a
2a
0
0
Z 1
log(sin x) dx = log 2
0
2
1
=
,
(u + n)2
(sin u)2
n=
Z
d
=
.
2
a 1
0 a + cos
Hint: use f (z) =
Problem 2.4. a) Prove that

Z
2
x d
=
2 + sin2
x
2 1 + x2
0
cot z
(u + z)2
x > 0.
a>0
48
b) Prove that
Z 2
1 r 1 2r 3r 2 n
2
(1 + 2 cos )n cos(n)
d =
1 r 2r cos
2r 2
1 2r 3r 2
0
for any 1 < r < 13 , n = 0, 1, 2, . . ..
and denote 0 = H. Suppose

Problem 2.5. a) Let be an open set in H
f H(0 ) C() with Im f (z) = 0 for all z H. Define
f (z) z
F (z) :=
f (
z ) z
where = {z : z }. Prove that F H( ).

so that |f (z)| = 1 on |z| = 1. If f does not vanish
b) Suppose f H(D) C(D)
anywhere in D, then prove that f is constant.
Problem 2.6. (a) Is there a bi-holomorphic map between the punctured disk {0 <
|z| < 1} and the annulus { 21 < |z| < 1}? If yes, then find it; if no, then prove that
it cannot exist.
(b) Prove that C is not conformally equivalent to any proper subdomain of itself.
Problem 2.7. Let f (z) = sinh(z),

n
1
1o
0 = z C : Re z > 0, < Im z <
,
2
2
and 1 = iH (the right half-plane). Prove that f : 0 1 is one-to-one, onto, and
bi-holomorphic (use the argument principle).
Problem 2.8. Let > 1. Show that the equation ez z = 0 has a unique zero
in the closed right half-plane Re z 0.
Problem 2.9. (a) Give the partial fraction expansion of r(z) =
(b) Let f (z) =
three annuli:
1
z(z1)(z2) .
z 2 +1
.
(z 2 +z+1)(z1)2
Find the Laurent expansion of f on each of the following
A1 = {0 < |z| < 1},
A2 = {1 < |z| < 2},
A3 = {2 < |z| < }
Problem 2.10. This exercise introduces and discusses some basic properties of the
Gamma function (z), which is of fundamental importance in mathematics:
(a) Show that
Z
et tz1 dt
(2.14)
(z) =
0
defines an analytic function in the half-plane Re z > 0. Also, verify the functional
equation (z + 1) = z(z) for all Re z > 0 as well as the identity (n + 1) = n! for all
integers n 0.
(b) Using the functional equation, show that there exists a unique meromorphic function on C which agrees with (z) on the right half-plane. Denoting this globally defined
function again by , prove that it has poles exactly at the nonpositive integers n with
n
for
n 0. Moreover, show that these poles are simple with residues Res(, n) = (1)
n!
all n 0.
8. PROBLEMS
49
(c) With (z) as in (a), verify the identity

Z
X
(1)n
et tz1 dt +
(2.15)
(z) =
n!(z + n)
1
n=0
for all Re z > 0. Now repeat part (b) using (2.15) instead of the functional equation.
(d) Verify that
Z a1
v
dv =
,
0 < Re a < 1.
(2.16)
1+v
sin a
0
Now apply this to establish that
sin z
as an identity
between meromorphic functions defined on C. In particular, we see that
( 12 ) = . Find an expression for |( 12 + it)| with t R.

To pass from (2.16) to (2.17), use the identity
Z
1x
(1 x) = y
euy ux du,
y>0
(z)(1 z) =
(2.17)
which holds for any 0 < x < 1.

(e) Check that
Z 1
()()
,
(1 t)1 t1 dt =
( + )
0
(f ) Prove that
Z
0
Re () > 0, Re () > 0.
tz1 cos t dt = (z) cos(z/2),
0 < Re z < 1
tz1 sin t dt = (z) sin(z/2),
1 < Re z < 1.
Deduce from this that
sin x
dx = ,
x
2
sin x
2.
dx
=
x3/2
(g) Let be a version of Hankels loop contour. This refers to a smooth curve
= (t) : R C \ (, 0] which approaches (, 0] from above as t and from
below as t . Moreover, it encircles w = 0 once in a positive sense. An example
0
Figure 2.5. A Hankel contour

of such a would be with : R (0, 1), (t) 0 as |t| , (t) = t i(t) for
< t < 1 and (t) = t + i(t) for 1 < t < as well as a circular arc (t) for
1 t 1 encircling w = 0 in a positive sense and joining the point 1 i(1) to the
point 1 + i(1).
50
Now prove that for all z C

(2.18)
1
2i
ew wz dw =
1
(z)
as an identity between entire functions. On the left-hand side wz = ezLog w where

Log w is the principal branch of the logarithm.
P
2n has radius of convergence R = 1.
Problem 2.11. Suppose that f (z) =
n=0 an z
Prove that f cannot be analytically continued to any disk centered at any point z0 with
|z0 | = 1 (assume that you can analytically continue to a neighborhood of z = 1 and
substitute z = aw2 + bw3 whereP
0 < a < 1 and a + b = 1).
nk where n
Generalize to other gap series
k+1 > nk for all k 1 with > 1
k=0 ak z
fixed.
Problem 2.12. Suppose K ( D, K 6= . Show that for any > 0 there exists a
polynomial P with P (0) = 1 and |P (z)| < on K (hint: use Runges theorem).
Problem 2.13. This exercise continues our investigation of the important Gamma
function.
(a) Prove that there is some constant A C such that
Z 1
dt
(z)
1 (1 t)z1
=
+A
Re z > 0.
(2.19)
(z)
t
0
Deduce from (2.19) that
(z) X 1
1
+A
=
(z)
n+1 n+z
n=0
(2.20)
as an identity between meromorphic functions on C.

(b) Derive the following product expansion from (2.20):

Y
z z
1
z
1+
e n
=e z
(2.21)
(z)
n
n=1
as an identity between entire functions. Here is the Euler constant

1
1
log N
= lim 1 + + . . . +
N
2
N
Alternatively, derive (2.21) from Hadamards theorem, see Theorem 2.24.
(c) Prove Gauss formula: if z C \ {n : n 0}, then
n! nz
(z) = lim
n z(z + 1)(z + 2) . . . (z + n)
CHAPTER 3
Harmonic functions on D
1. The Poisson kernel
There is a close connection between Fourier series and analytic (harmonic) functions
on the disc D. Heuristically speaking, a Fourier series can be viewed as the boundary
values of a Laurent series
X
an z n
n=
Let us use this observation to derive a solution formula for the following fundamental
problem: Given a function f on the boundary of D find a harmonic function u on D
which attains these boundary values.
Notice that this so-called Dirichlet problem is formulated too vaguely. In fact, much of
this as well as the following two chapters is devoted to a proper interpretation of what
we mean by attaining the boundary values and what kind of boundedness properties we
wish u to satisfy on all of D.
But for the moment, let us proceed heuristically. Starting with the Fourier series f () =
P
2i , we observe that one harmonic extension to the
nZ f (n)e(n) with e() := e

interior is given by
X
X
f(n)r n e(n),
z = re()
f(n)z n =
u(z) =
nZ
nZ
This is singular at z = 0, though, in case f(n) 6= 0 for one n < 0. Since both z n and z n
are (complex) harmonic, we can avoid the singularity by defining
(3.1)
u(z) =
f(n)z n +
1
X
f(n)
z |n|
n=
n=0
which at least formally is a solution

R 1 of our Dirichlet problem.
Inserting z = re() and f (n) = 0 e(n)f () d into (3.1) yields

Z 1
Z X
Pr ( ) f () d
r |n| e(n( ))f () d =:
u(re()) =
0
T nZ
where the Poisson kernel

Pr () :=
r |n| e(n) =
nZ
1 r2
1 2r cos(2) + r 2
via explicit summation. We start the rigorous theory by stating some properties of Pr .
51
52
3. HARMONIC FUNCTIONS ON D
1
1r
1r
Figure 3.1. The poisson kernel

Lemma 3.1. The function z = re() = Pr () is a positive harmonic function on D.
R1
It satisfies 0 Pr () d = 1 and for any (complex) Borel measure on T,
z = re() 7 (Pr )()
defines a harmonic function on D.
Proof. These properties are all either evident from the explicit form of the kernel
or via the defining series.

The behavior of the Poisson kernel close to the boundary can be captured by means
of the following notion.
Definition 3.2. A sequence {n }

n=1 L (T) is called an approximate identity
provided
R1
A1) 0 n () d = 1 for all n
R1
A2) supn 0 |n ()| d <
R
A3) for all > 0 one has |x|> |n ()| d 0 as n .
The same definition applies, with obvious modifications, to families of the form {t }0<t<1
(with n replaced by t 1).
A standard example is the box kernel

1
[,]
2
0<< 1
2
in the limit 0. Another example is the Fejer kernel from Fourier series. The relevant
example for our purposes is of course the Poisson kernel {Pr }0<r<1 , and we leave it to
the reader to check that it satisfies A1)A3). The significance of approximate identities
lies with their reproducing properties (as their name suggests).
Lemma 3.3. For any approximate identity {n }
n=1 one has
2. HARDY CLASSES OF HARMONIC FUNCTIONS
53
(1) If f C(T), then kn f f k 0 as n

(2) If f Lp (T) where 1 p < , then kn f f kp 0 as n .
These statements carry over to approximate identities t , 0 < t < 1 simply by replacing
n with t 1.
Proof. Since T is compact, f is uniformly continuous. Given > 0, let > 0 be
such that
sup sup |f (x y) f (x)| <
x |y|<
Then, by A1)A3),

Z

|(n f )(x) f (x)| = (f (x y) f (x))n (y) dy
T
Z
Z
|n (y)|2kf k dy
sup sup |f (x y) f (x)| |n (t)| dt +
xT |y|<
|y|
<C
provided n is large. For the second part, fix f Lp . Let g C(T) with kf gkp < .
Then
kn f f kp kn (f g)kp + kf gkp + kn g gkp

sup kn k1 + 1 kf gkp + kn g gk
n
where we have used Youngs inequality (kf1 f2 kp kf1 k1 kf2 kp ) to obtain the first term
on the right-hand side. Using A2), the assumption on g, as well as the first part finishes
the proof.

An immediate consequence is the following simple and fundamental result.
Theorem 3.4. Let f C(T). The unique harmonic function u on D, with u C(D)
and u = f on T is given by u(z) = (Pr f )(), z = re().
Proof. Uniqueness follows from the maximum principle. For the existence, we
observed before that u(z) := (Pr f )() with |z| < 1 is harmonic on D. By Lemma 3.3,
by
ku(re() f k 0 as r 1. This implies that we can extend u continuously to D
setting it equal to f on T.

2. Hardy classes of harmonic functions
Next, we wish to reverse this process and understand which classes of harmonic functions on D assume boundary values on T. Moreover, we need to clarify which boundary
values arise here and what we mean by assume. Particularly important classes known
as the little Hardy spaces are as follows:
Definition 3.5. For any 1 p define
Z

n

p
h (D) := u : D C harmonic sup
0<r<1 0
|u(re())|p d <
with norm
|||u|||p := sup ku(re())kLp (T)

0<r<1
54
By the mean value property, any positive harmonic function belongs to the space h1 (D).
Amongst those, the most important example is Pr () h1 (D). Observe that this function
has boundary values Pr 0 (the Dirac mass at = 0) as t 1 where the convergence
is in the sense of distributions. In what follows, M(T) denotes the complex-valued Borel
measures and M+ (T) M(T) the positive Borel measures.
Theorem 3.6. There is a one-to-one correspondence between h1 (D) and M(T) given
by M(T) 7 Fr () := (Pr )(). Under this correspondence, any M+ (T) relates
uniquely to a positive harmonic function. Furthermore,
kk = sup kFr k1 = lim kFr k1
(3.2)
0<r<1
r1
and the following properties hold:

(1) is absolutely continuous with respect to Lebesgue measure ( d) if and
only if {Fr } converges in L1 (T). If so, then d = f d where f = L1 -limit of
Fr .
(2) The following are equivalent for 1 < p : d = f d with f Lp (T)
{Fr }0<r<1 is Lp - bounded
{Fr } converges in Lp if 1 < p < and in weak- sense in L if

p = as r 1
(3) f is continuous F extends to a continuous function on D Fr converges

uniformly as r 1.
This theorem identifies h1 (D) with M(T), and hp (D) with Lp (T) for 1 < p .
Moreover, h (D) contains the subclass of harmonic functions that can be extended
continuously onto D; this subclass is the same as C(T). Before proving the theorem
we
present two simple lemmas. In what follows we use the notation Fr () := F re() .
Lemma 3.7.
(1) If F C(D) and F = 0 in D, then Fr = Pr F1 for any 0 r < 1.
(2) If F = 0 in D, then Frs = Pr Fs for any 0 r, s < 1.
(3) As a function of r (0, 1) the norms kFr kp are non-decreasing for any 1 p
.
Proof. 1.) is a restatement of Theorem 3.4. For 2.), rescale the disc sD to D and
apply the first property. Finally, by Youngs inequality
as claimed.
kFrs kp kPr k1 kFs kp = kFs kp
Lemma 3.8. Let F h1 (D). Then there exists a unique measure M(T) such
that Fr = Pr .
Proof. Since the unit ball of M(T) is weak- compact there exists a subsequence
rj 1 with Frj in weak- sense to some M(T). Then, for any 0 < r < 1,
Pr = lim (Frj Pr ) = lim Frrj = Fr
j
by Lemma 3.7. Let f C(T). Then, again by Lemma 3.7,
hFr , f i = hPr , f i = h, Pr f i h, f i
3. ALMOST EVERYWHERE CONVERGENCE TO THE BOUNDARY DATA
55
as r 1. This shows that, in the weak- sense,
(3.3)
= lim Fr
r1
which implies uniqueness of .
Proof of Theorem 3.6. If M(T), then Pr h1 (D). Conversely, given

F h1 (D) then by Lemma 3.8 there is a unique so that Fr = Pr . This gives the
one-to-one correspondence. Moreover, (3.3) and Lemma 3.7 show that
kk lim sup kFr k1 = sup kFr k1 = lim kFr k1 .
r1
r1
0<r<1
Since one also has

sup kFr k1 sup kPr k1 kk = kk ,
0<r<1
1
L (T) and
0<r<1
(3.2) follows. If f
d = f d, then Lemma 3.3 shows that Fr f in
L1 (T). Conversely, if Fr f in the sense of L1 (T), then because of (3.3) necessarily
d = f d which proves the first part. The other parts are equally easy and we skip the
detailssimply invoke Lemma 3.3, part 2.) for 1 < p < and Lemma23.3 part 1.) if
p = .

In passing we remark the following: an important role is played by the kernel Qr ()
which is the harmonic conjugate of Pr (). Recall that this means that Pr () + iQr () is
analytic in z = re() and Q0 = 0. In this case it is easy to find Qr () since

1+z
Pr () = Re
1z
and therefore

2r sin(2)
1+z
=
Qr () = Im
1z
1 2r cos(2) + r 2
Observe that {Qr }0<r<1 is not an approximate identity, since Q1 () = cot() which is
1
not the density of a measure it behaves like
close to = 0. The Hilbert transform
is the map which is formally defined as follows:

f 7 uf 7 u
ef 7 u
ef
T
where uf denotes the harmonic extension to D and u

ef its harmonic conjugate. From
the preceding, Q1 is the kernel of the Hilbert transform. It is a very important object,
especially for the role it played in the development of function theory. Similarly famously,
the Dirichlet kernel in Fourier series is not an approximate identity and the many efforts
in understanding its mapping properties have been of enormous importance in analysis.
But we will not pursue these topics any further here.
3. Almost everywhere convergence to the boundary data
Instead, we turn to the issue of almost everywhere convergence of Pr f to f as r 1.

The main idea here is to mimic the proof of the Lebesgue differentiation theorem. In
particular, we need the Hardy-Littlewood maximal function M f , which is defined as
follows:
Z
1
|f (y)| dy
M f (x) = sup
xIT |I| I
where I T is an (open) interval and |I| is the length of I. The most basic facts
concerning this (sublinear) operator are contained in the following result.
56
Proposition 3.9. M is bounded from L1 to weak L1 , i.e.,

3
kf k1
for all > 0. For any 1 < p , M is bounded on Lp .

mes[x T|M f (x) > ]
Proof. Fix some > 0 and any compact

K {x | M f (x) > }
(3.4)
There exists a finite cover {Ij }N

j=1 of T by open arcs Ij such that
Z
|f (y)| dy > |Ij |
(3.5)
Ij
for each j. We now pass to a more convenient sub-cover (this is known as Wieners
covering lemma): Select an arc of maximal length from {Ij }; call it J1 . Observe that any
Ij such that Ij J1 6= satisfies Ij 3 J1 where 3 J1 is the arc with the same center as
J1 and three times the length (if 3 J1 has length larger than 1, then set 3 J1 = T). Now
remove all arcs from {Ij }N
j=1 that intersect J1 . Let J2 be one of the remaining ones with
maximal length. Continuing in this fashion we obtain arcs {J }L
=1 which are pair-wise
disjoint and so that
L
N
[
[
3 J
Ij
j=1
=1
In view of (3.4) and (3.5) therefore,
mes(K) mes
L
3X
=1
L
[
=1
3 J
|f (y)| dy
L
X
mes(J )
=1
3
kf k1
as claimed. To prove the Lp statement, one interpolates the weak L1 bound with the
trivial L bound
kM f k kf k
by means of Marcinkiewiczs interpolation theorem.
We now introduce a class of approximate identities which can be reduced to the box
kernels. The importance of this idea is that it allows us to dominate the maximal function
associated with an approximate identity by the Hardy-Littlewood maximal function, see
Lemma 3.11 below.
Definition 3.10. Let {n }
n=1 be an approximate identity as in Definition 3.2. We
say that it is radially bounded if there exist functions {n }
n=1 on T so that the following
additional property holds:
A4) |n | n , n is even and decreasing, i.e., n (x) n (y) for 0 y x 12 ,
for all n 1. Finally, we require that supn kn k1 < .
Now for the domination lemma.
3. ALMOST EVERYWHERE CONVERGENCE TO THE BOUNDARY DATA
57
1
Lemma 3.11. If {n }
n=1 satisfies A4), then for any f L (T) one has
sup |(n f )(x)| M f (x) sup kn k1

n
for all x T.
Proof. It clearly suffices to show the following statement: let K : [ 12 , 12 ] R+ {0}
be even and decreasing. Then for any f L1 (T)
|(K f )(x)| kKk1 M f (x)
(3.6)
Indeed, assume that (3.6) holds. Then

sup |(n f )(x)| sup(n |f |)(x) sup kn k1 M f (x)
n
and the lemma follows. The idea behind (3.6) is to show that K can be written as an
average of box kernels, i.e., for some positive measure
(3.7)
K(x) =
1
2
[y,y] (x) d(y)
We leave it to the reader to check that

1
1
d = dK + K
2
2
is a suitable choice. Notice that (3.7) implies that
Z
K(x) dx =
0
1
2
2y d(y)
Moreover, by (3.7),
Z

|(K f )(x)| =
Z 21

1

[y,y] f (x) 2y d(y)
M f (x)2y d(y)
2y
0
0
= M f (x)kKk1
which is (3.6).
1
2
Finally, we can properly address the question of whether Pr f f in the almost

everywhere sense for f L1 (T). The idea is as follows: the pointwise convergence is clear
from Lemma 3.3 for continuous f . This suggests approximating f L1 by a sequence of
1
continuous ones, say {gn }
n=1 , in the L norm. Evidently, we encounter an interchange
of limits here, namely r 1 and n . As always in such a situation, we require
some form of uniform control. The needed uniform control is precisely furnished by the
HardyLittlewood maximal function.
1
Theorem 3.12. If {n }
n=1 satisfies A1)A4), then for any f L (T) one has
n f f almost everywhere as n .
58
Proof. Pick > 0 and let g C(T) with kf gk1 < . By Lemma 3.3, with
h = f g one has, with | | being Lebesgue measure,

x T lim sup | (n f ) (x) f (x)| >

n

x T lim sup |(n h)(x)| > /2 + x T |h(x)| > /2

n

x T sup |(n h)(x)| > /2 + x T |h(x)| > /2

n

x T CM h(x) > /2 + x T |h(x)| > /2
To pass to the final inequality we used Proposition 3.9 as well as Markovs inequality
(recall khk1 < ).

As a corollary we not only obtain the classical Lebesgue differentiation theorem,
but also almost everywhere convergence of of the Poisson integrals Pr f f for any
f L1 (T) as r 1. In view of Theorem 3.6 we of course would like to know whether a
similar statement holds for measures instead of L1 functions. It turns out, see Problem 3.2
below, that Pr f almost everywhere where f is the density of the absolutely
continuous part of in the Lebesgue decomposition. A most important example here
is Pr itself! Indeed, its boundary measure is 0 and the almost everywhere limit is
identically zero. Hence, in the almost everywhere limit we lose a lot of information
namely the singular part of the boundary measure. An amazing fact, known as the F. &
M. Riesz theorem, states that there is no such loss in the class h1 (D) H(D). Indeed, any
such function is the Poisson integral of an L1 function rather than a measure. Another
way of expressing this fact is as follows: if M(T) satisfies
(n) = 0 for all n < 0, then
is absolutely continuous with respect to Lebesgue measure on T. For this important
result we refer to the reader to the literature on Hardy spaces H p (D) := hp (D) H(D)
of holomorphic functions on D, see for example [17] and [35].
4. Problems
P
Problem 3.1. Let f (z) = n=0 an z n have radius of convergence R = 1. Problems

(a)(c) further explore the connection between the behavior of the series and the function
f at the boundary. They are completely elementary but a bit tricky.
(a) Suppose an R for all n 0 and sn = a0 + a1 + . . . + an as n . Prove
that f (z) cannot be analytically continued to any neighborhood of z = 1. Is it meaningful
to call z = 1 a pole of f ? Does the same conclusion hold if all an are real and |sn |
as n ?
P
(b) Suppose
n=0 an = s. Show that then f (z) s as z 1 inside the region
z K D, 0 < < arbitrary but fixed. Here K is a cone with tip at z = 1,
symmetric about the x-axis, opening angle , and with (, 1) K (this type of
convergence z 1 is called non-tangential convergence). Note that z = 1 can be
replaced by any z D.
(c) Now assume
P that nan 0 as n . If f (z) s as z 1 non-tangentially,
then prove that n=0 an = s. Note again that z = 1 can be replaced by any z D.
Problem 3.2. It is natural to ask whether there is an analogue of Theorem 3.12 for
measures M(T). Prove the following:
4. PROBLEMS
59
(a) If M(T) is singular with respect to Lebesgue measure (d), then for
a.e. x T (with respect to Lebesgue measure)
([x , x + ])
0 as 0
(b) Let {n }
n=1 satisfy A1)A4) from Chapter 3, and assume that the kernels {n }n=1
from Definition 3.10 also satisfy
sup |n ()| 0 as n
(3.8)
<||< 21
for all > 0. Under these assumptions show that for any M(T)
n f a.e. as n
where d = f d + ds is the Lebesgue decomposition, i.e., f L1 (T) and ds d.

Problem 3.3. (a) Let 0 r1 < r2 and suppose that u is a real-valued harmonic
function on the annulus A = {z C : r1 < |z| < r2 }. Prove that there exists some
unique k R and f H(A) such that
u(z) = k log |z| + Re f (z)
zA
Next, assume that r1 = 0. Prove that if u is bounded on A, then k = 0 and u extends to

a harmonic function throughout |z| < r2 .
(b) Suppose C is open and simply connected. Let z0 and suppose that
u \ {z0 } R is harmonic such that
u(z) log |z z0 |
remains bounded as z z0 . Show that there exists f H() such that f (z0 ) = 0,
u(z) = log |f (z)|, and f is one-to-one one some disk around z0 .
Problem 3.4. Suppose that u, v are harmonic in so that u and v never vanish
in (we call this non-degenerate). If f = u + iv is conformal (i.e., f H()), then we
know that the level curves u = const and v = const in are perpendicular to each other
(why?). This exercise addresses the converse:
(a) Suppose v, w are harmonic and non-degenerate in such that the level curves of
v and w coincide in . How are v and w related?
(b) Suppose u, v are harmonic and non-degenerate in , and assume their level curves
are perpendicular throughout . Furthermore, assume that |u(z0 )| = |v(z0 )| at one
point z0 . Prove that either u + iv or u iv is conformal in .
Problem 3.5. We say that u : [, ) is subharmonic (u sh()) provided
it is continuous and it satisfies the sub mean value property (SMVP): for every z0
and any 0 r < dist(z0 , ),
Z 2
1
u(z0 + rei ) d
(3.9)
u(z0 )
2 0
In addition, we require that u 6 . Establish the following ten properties:
(i) Maximum principle: if is bounded and u sh(), then
sup lim sup u(z) M < = u(z) M z
z
z
with equality being attained on the right-hand side for some z iff u = const.
60
(ii) Let u sh() on and suppose h is harmonic on some open disk K compactly
Further, assume that u h on K. Show that u h
contained in and h C(K).
on K. Further if u = h at some point in K, then u = h on K (this explains the name
sub-harmonic).
P
(iii) If u1 , . . . , uN sh(), then so is max(u1 , . . . , uN ) and N
j=1 cj uj with cj 0.
(iv) If h is harmonic on and : R R is convex, then h sh(). If v sh(),
and : [, ) R is continuous, non-decreasing, and convex, then v sh().
(v) This is a converse of (ii): suppose u : [, ) is continuous so that for
the harmonic majorization property holds: if h C(K)
is
every disk K with K
harmonic on K and satisfies u h on K then u h on K. Prove that u is subharmonic.
(vi) Prove that subharmonic functions are characterized already by the local SMVP:
for every z0 there exists 0 < (z0 ) dist(z0 , ) such that (3.9) holds for every
0 < r < (z0 ).
(vii) Suppose u sh(). Prove that for any z0 and any 0 < r1 < r2 <
dist(z0 , ),
Z 2
Z 2
1
1
i
u(z0 + r1 e ) d
u(z0 + r2 ei ) d
<
2 0
2 0
and
1
lim
r0+ 2
as well as
2
0
u(z0 + rei ) d = u(z0 )
|u(z0 + rei )| d <
for any 0 < r < dist(z0 , ).

(viii) Suppose u C 2 (). Then u sh() iff u 0 in .
(ix) Suppose f H(), f 6 0. Prove that the following functions are in sh():
log |f (z)|, |f (z)| for any > 0, e|f (z)| for any > 0.
(x) Show that in C() the harmonic functions are precisely those that satisfy the
mean value property. Use this to prove that the limit of any sequence {un }
n=1 of harmonic functions on which converges uniformly on every compact subset of is again
harmonic.
Problem 3.6. (a) Let 1 and let S be the sector

n
o

S := rei 0 < r < , || <
2
satisfy u M on S and u(z) < |z| in S where < . Prove
Let u sh(S) C(S)
that u M on S.
(b) Let u sh() where is a bounded domain. Further, suppose E := {zn }
n=1
has the property that
lim sup u(z) M
z\E
Prove that u M in .
Problem 3.7. Let u be subharmonic on a domain C.
(a) Prove that
hu, i 0 C
comp (), 0
4. PROBLEMS
61
where h, i denotes the standard L2 () pairing, and deduce from it that there exists a
unique positive Borel measure (called the Riesz measure) on such that
ZZ
(x)(dx)
hu, i =
C
comp ()
for all
(from this identity, (K) < for all compact K ). In other
words, even if a subharmonic function is not C 2 its distributional Laplacean is no worse
than a measure. Find for u = log |f | where f H().
(b) Show that with as in (a) and for any 1 compactly contained,
ZZ
log |z | (d) + h(z)
(3.10)
u(z) =
1
where h is harmonic on 1 . This is Rieszs representation of subharmonic functions.

Interpret its meaning for u = log |f | with f H(). Show that, conversely, any nonnegative Borel measure which is finite on compact sets of defines a subharmonic
function u via (3.10) (with h = 0) provided the integral on the right-hand side is continuous with values in [, ). Give an example of a where u is not continuous. But
show that (3.10) is always upper semicontinuous (usc), i.e,
u(z0 ) lim sup u(z)
zz0
for all z0 . Check that upper semicontinuous functions always attain their supremum
on compact sets. In fact, the theory of subharmonic functions which we have developed
so far applies to the wider class of usc functions satisfying the SMVP (try to see this)
basically unchanged.
(c) With u and as in (a), show that
Z r
Z 1
(D(z, t))
dt
u(z + re()) d u(z) =
(3.11)
t
0
0
for all D(z, r) (this is Jensens formula). In other words, measures the extent
to which the mean value property fails and really is a sub mean value property. Now
find an estimate for (K) where K is compact in terms of the pointwise size of u.
Finally, write (3.11) down explicitly for u = log |f | with f H().
Problem 3.8. This exercise introduces the important Harnack inequality and principle for harmonic functions.
2
be the Poisson kernel. Show that for any 0 < r < 1
(a) Let Pr () = 12r1r
cos +r 2
(3.12)
1r
1+r
Pr ()
1+r
1r
and deduce from this that for any nonnegative harmonic function u on D one has
sup u(z) C(r) inf u(z)
|z|r
|z|r
where C(r) < for 0 < r < 1. What is the optimal constant C(r)? Now show that for
any and K compactly contained in one has the inequality
sup u(z) C(K, ) inf u(z)
zK
zK
62
for nonnegative harmonic functions u on . Now prove that if u : R2 R is harmonic,

and bounded from one side (thus, u M in for some finite constant M or u M ),
then u is constant.
(b) Suppose u1 u2 u3 . . . are harmonic functions in . Let u = supn un . Then
either u or u is harmonic in .
Problem 3.9. Let u sh(D). Show that the following two properties are equivalent:
(i) u has a harmonic majorant on D, i.e., there exists h : D R harmonic such that
u h on D.
R1
(ii) sup0<r<1 0 u(re()) d < where e() = e2i .
We say that h0 is a least harmonic majorant of u iff h0 is a harmonic majorant of
u on D and if h h0 for every other harmonic majorant h of u.
Prove that if u has a harmonic majorant on D, then it has a least harmonic majorant.
Given an example of a u sh(D) that has no harmonic majorant.
Problem 3.10. Let f H(D), f 6 0. Then prove that the following two properties
are equivalent (here log+ x = max(0, log x)):
(i) log+ |f | has a harmonic majorant in D.
(ii) f = hg where g, h H(D) with |g| 1, 0 < |h| 1 in D.
Problem 3.11. You should compare this to Problem 2.1.
(a) Suppose Z = {zn }
n=0 D \ {0} satisfies
X
(1 |zn |) <
n=0
Prove that
B(z) =
Y
|zn | zn z
z 1 zn z
n=0 n
converges uniformly on every D(0, r) with 0 < r < 1 to a holomorphic function B H(D)
with |B(z)| 1 for all |z| < 1. It vanishes exactly at the zn (with the order of the zero
being equal to the multiplicity of zn in Z).
(b) We know that limr1 B(rei ) exists for almost every (after all, B h (D) so
Chapter 3 applies). Denote these boundary values by B(ei ). Prove that |B(ei )| = 1 for
almost every .
CHAPTER 4
Riemann surfaces: definitions, examples, basic properties

1. The basic definitions
The following definition introduces the concept of a Riemann surface. Needless to say,
this concept arose naturally in an attempt to understand algebraic functions (which are
the analytic continuations of the roots of a polynomial equation P (z, w) = 0 relative to
w, say) as well as other multi-valued analytic functions. We will discuss this important
construction in the following chapter, but for now start with the abstract definition.
Definition 4.1. A Riemann surface is a twodimensional, connected, Hausdorff
topological manifold M with a countable base for the topology and with conformal transition maps between charts. I.e., there exists a familiy of open set {U }A covering M
and homeomorphisms : U V where V R2 is some open set so that
: (U U ) (U U )
1
is biholomorphic (in other words, a conformal homeomorphism).

(U , ) as a chart and to A as an atlas of M .
U1
We refer to each
U2
V1
V2
Figure 4.1. Charts and analytic transition maps

The countability axiom can be dispensed with as it can be shown to follow from the
other axioms (this is called Rados theorem, see [15]), but in all applications it is easy to
check directly. Two atlases A1 , A2 of M are called equivalent iff A1 A2 is an atlas of M .
An equivalence class of atlases of M is called conformal structure and a maximal atlas of
63
64
4. RIEMANN SURFACES
M is the union of all atlases in a conformal structure. We shall often write (U, z) for a
chart indicative of the fact that p 7 z(p) takes U into the complex zplane. Moreover,
a parametric disk is a set D U where (U, z) is chart with z(D) a disk in C. We shall
always assume that z(D) z(U ) is compact. By a parametric disk D centered at p M
we mean that (U, z) is a chart with p U , z(p) = 0, and D = z 1 (D(0, r)) for some
r > 0.
We say that the Riemann surface M is an extension of the Riemann surface N iff
N M as an open subset and if the conformal structure of M restricted to N is exactly
the conformal structure that N carried to begin with.
Definition 4.2. A continuous map f : M N between Riemann surfaces is said to
be analytic iff it is analytic in charts. I.e., if p M is arbitrary and p U , f (p) V
where (U , z ) is a chart of M and (V , w ) is a chart of N , respectively, then w f z1
is analytic where it is defined. We say that f is a conformal isomorphism iff f is an
analytic homeomorphism. If N = C then one says that f is holomorphic, if N = C , it
is called meromorphic.
It is clear that the meromorphic functions on a Riemann surface form a field. One
refers to this field as the function field1 of a surface M .
2. Examples
In this section we discuss a number of examples of Riemann surfaces, some of which
will play an important role in the development of the theory. We begin with an obvious
class of examples which serve to illustrate that complex analysis as we have developed it
so far in this book, is really a special case of general Riemann surfaces (and the local
case).
1) Any open region C: Here, a single chart suffices, namely (, z) with z being
the identity on . The associated conformal structure consists of all (U, ) with U
open and : U C biholomorphic. Notice that an alternative, non-equivalent conformal
structure is (, z).
2) The Riemann sphere S 2 R3 , which can be described in three, conformally equivalent, ways: S 2 , C , CP 1 .
2a) We define a conformal structure on S 2 via two charts
(S 2 \ (0, 0, 1), + ),
(S 2 \ (0, 0, 1), )
where are the stereographic projections

+ (x1 , x2 , x3 ) =
x1 + ix2
,
1 x3
(x1 , x2 , x3 ) =
x1 ix2
1 + x3
from the north, and south pole, respectively, see Figure 1.1. If p = (x1 , x2 , x3 ) S 2 with
x3 6= 1, then
+ (p) (p) = 1
This shows that the transition map between the two charts is z 7 1z from C C .
2b) The one-point compactification of C denoted by C := C {}. The neighborhood base of in C is given by the complements of all compact sets of C. Again
1At least when M is compact this is commonly used terminology from algebraic geometry.
2. EXAMPLES
65
there are two charts, namely

1
(C \ {0}, )
z
in the obvious sense. The transition map is again given by z 7 1z .
2c) The one-dimensional complex projective space

CP 1 := [z : w] | (z, w) C2 \ {(0, 0)}
(C, z),
where the equivalence relation is (z1 , w1 ) (z2 , w2 ) iff z2 = z1 , w2 = w1 for some

C . Our charts are (U1 , 1 ) and (U2 , 2 ) where
z
U1 := {[z : w] CP 1 | w 6= 0}, 1 ([z : w]) =
w
w
U2 := {[z : w] CP 1 | z 6= 0}, 2 ([z : w]) =
z
1
Here, too, the transition map is z 7 z .
3) Any smooth, orientable two-dimensional submanifold of R3 is a Riemann surface

4) Any polyhedral surface S R3 is a Riemann surface: Such an S is defined to
be a compact toplogical manifold which can be written as the finite union of faces {fi },
edges {ej }, and vertices {vk }. Any fi is assumed to be an open subset of a plane in R3 ,
an edge is an open line segment and a vertex a point in R3 with the obvious relations
between them (the boundaries of faces in R3 are finite unions of edges and vertices
and the endpoints of the edges are vertices; an edge is where two faces meet etc.). To
Figure 4.2. Polyhedra are Riemann surfaces

define a conformal structure on such a polyhedral surface, proceed as follows: each fi
defines a chart (fi , i ) where i is Euclidean motion (affine isometry) that takes fi into
C = R2 = R3 where we identify R2 with the (x1 , x2 )plane of R3 , say. Each edge ej
defines a chart as follows: let fi1 and fi2 be the two unique faces that meet in ej . Then
(fi1 fi2 ej , j ) is a chart where j maps that folds fi1 fi2 ej at the edge so that
it becomes straight (piece of a plane) and then maps that plane isometrically into R2 .
Finally, at a vertex vk we define a chart as follows: for example, suppose three faces meet
at vk , say fi1 , fi2 , fi3 with respective angles 1 , 2 , and 3 . Let > 0 be defined so that
X
(4.1)
i = 2
66
4. RIEMANN SURFACES
and let the chart map these faces with their edges meeting at v into the plane in such a
way that angles get dilated by . It is easy to see that this defines a conformal structure
(for example, at a vertex, the transition maps are z where is as in (4.1)).
5) Covers and universal covers, and their quotients
6) Surfaces defined as smooth (projective) algebraic curves: Let P (z, w) be an irreducible polynomial such that dP 6= 0 on
S := {(z, w) C2 | P (z, w) = 0}
In other words, (z P, w P )(z, w) 6= (0, 0) when P (z, w) = 0 (such P are called nonsingular). Then S C2 is a Riemann surface imbedded in C2 , called an affine algebraic curve.
To defined the complex structure on S, one can use either z or w as local co-ordinates
depending on whether z P 6= 0 or w P 6= 0 on that neighborhood. The irreducibility
f (z)
where
of P implies that S is connected. By construction, any function of the form g(w)
f, g are meromorphic on C and g not identically zero, is a meromorphic function on S
(this of course raises the question as to what all meromorphic functions on S are). To
compactify S, we pass to the homogenized version of P : thus, let 1 be the minimal
integer for which
(4.2)
u P (z/u, w/u) =: Q(z, w, u)
has no negative powers of u. Then

(4.3)
Se := {[z : w : u] CP 2 | Q(z, w, u) = 0}
is well-defined and Se is called a smooth projective algebraic curve, whereas S = Se {[z :

e Assuming that Q is nonsingular, i.e.,
w : 1] : (z, w) C2 } is called the affine part of S.
e it follows just as before that Se is a Riemann surface which is compact as
dQ 6= 0 on S,
a closed subset of the compact space CP 2 . To be more precise, we use the three charts
{[1 : w : u] | (w, u) C2 }, {[z : w : 1] | (z, w) C2 }, {[z : 1 : u] | (z, u) C2 }
to cover CP 2 . If follows from Eulers relation dQ(z, w, u)(z, w, u) = Q(z, w, u) that on

Se one has dQ(z, w, u)(z, w, u) = 0. Consequently, if we set any one of the coordinates
equal to 1, then the polynomial in the two remaining variables is nonsingular in the
affine sense from above. This allows us to define complex structures on Se over each of
the three projective charts which are of course compatible with each other. The merof (z)
from above extend to meromorphic functions on Se provided f, g
morphic functions g(w)
are rational (in this case, too, we would like to characterize the meromorphic functions
e Being compact, Se has finite genus. An
on Se in other words, the function field on S).
example of (the affine part of) a curve of genus g 1 is given by
(4.4)
w2
N
Y
(z zj ) = 0
j=1
N
where {zj }N
j=1 C are distinct and N = 2g or N = 2g 1. For any z0 C \ {zj }j=1 one
has local coordinates
v
uN
uY
w(z) = t (z zj )
j=1
3. FUNCTIONS ON RIEMANN SURFACES
67
where the two signs correspond precisely to the two sheets locally near z0 ; note that
the square root is analytic. Near any z , 1 N one sets z := z + 2 so that
v
u N
u Y
w() = t
(z zj + 2 )
j=1, j6=
where the ambiguity of the choice of sign can be absorbed into (the square root is again
analytic). It is clear that the transition maps between the charts are holomorphic. The
reader will easily verify that the projective version of (4.4) with N 2, i.e.,
Q(z, w, u) := w2 uN 2
N
Y
(z uzj )
j=1
is nonsingular for N = 2, 3 but singular when N 4. Indeed, if N 4 one has dQ = 0 at

the point at infinity of Se (which means when u = 0). Returning to the affine curve, one
refers to the zj as branch points, since the natural projection w 7 z, which is a covering
map on C \ {zj }N
j=1 , ceases to be a covering map near each zj . Rather, one refers to
this case as a branched covering map and an algebraic curve is a branched cover of the
Riemann sphere. The specific algebraic curve we just considered is called elliptic curve if
g = 1 or hyper-elliptic curve if g > 1. A very remarkable fact of the theory of Riemann
surfaces is that any compact Riemann surface is the same as (i.e., conformally equivalent
to) an algebraic curve defined by some irreducible polynomial P C[z, w], see Chapter 5.
For this, we no longer assume that P is nonsingular and therefore proceed differently
with our construction of the Riemann surface of P it is defined via all possible analytic
continuations of a locally defined analytic solution w = w(z) of P (z, w(z)) = 0. The
difference in this procedure is that we do not immediately seek to imbed this Riemann
surface into CP 2 . In fact, in general this cannot be done when Q is singular (but is a
result of Riemann surface theory that every compact Riemann surface can be imbedded
into some CP d , in fact d = 3 suffices).
7) Riemann surfaces defined via analytic continuation of an analytic germ:
3. Functions on Riemann surfaces
We already defined analytic functions between Riemann surfaces and also introduced
the concept of a conformal isomorphism. Note that any conformal isomorphism has
an analytic inverse. In Example 1) above, the Riemann surfaces with with conformal
structures induced by (, z) and (, z), respectively, do not have equivalent conformal
structures but are conformally isomorphic (via z 7 z). As the reader may have guessed,
examples 3), 4), and 5) are isomorphic (we shall drop conformal when it is clearly
implied). As usual,
P SL(2, C) = SL(2, C)/{Id}
Theorem 4.3. The Riemann surfaces S 2 , C , and CP 1 are conformally isomorphic.
Furthermore, the group of automorphisms of these surfaces is P SL(2, C).
Proof. we leave it to the reader to write down the explicit isomorphisms between
these surfaces. As for the automorphism group, each

a b
A=
SL(2, C)
c d
68
4. RIEMANN SURFACES
defines an automorphism of CP 1 via

[z : w] 7 [az + bw : cz + dw]
Note that A define the same map (a Mobius transform). On the other hand, if f is
an automorphism of C , then composing with a Mobius transformation we may assume
that f () = . Hence, restricting f to C yields a map from Aut(C) which is of the
form (see Problem 1.12) f (z) = az + b and we are done.

We now state the important uniqueness and open mapping theorems for analytic
functions on Riemann surfaces.
Theorem 4.4 (Uniqueness theorem). Let f, g : M N be analytic. Then either
f g or {p M | f (p) = g(p)} is discrete in M .
Proof. Define
A := {p M | locally at p, f and g are identically equal}
B := {p M | locally at p, f and g agree only on a discrete set}
It is clear that both A and B are open subsets of M . We claim that M = A B which
then finishes the proof since M is connected. If p M is such that f (p) 6= g(p) then
p B. If, on the other hand, f (p) = g(p), then we see via the usual uniqueness theorem
in charts that {f = g} not discrete implies that f = g locally around p.

As an obvious corollary, note that for any analytic f : M N each level set
{f M | f (p) = q} with q N fixed, is either discrete or all of M (and thus f = const).
In particular, if M is compact and f not constant, then {p M | f (p) = q} is finite.
Theorem 4.5 (Open mapping theorem). Let f : M N be analytic. If f is not
constant, then f (M ) is an open subset of N . More generally, f takes open subsets of M
to open subsets of N .
Proof. By the uniqueness theorem, if f is locally constant around any point, then
f is globally constant. Hence we can apply the usual open mapping theorem in every
chart to conclude that f (M ) N is open.

Corollary 4.6. Let M be compact and f : M N analytic and nonconstant. Then
f is onto and N is compact.
Proof. Since f (M ) is both closed (since compact and N Hausdorff), and open by
Theorem 4.5, it follows that f (M ) = N as claimed.

It is customary to introduce the following terminology.
Definition 4.7. The holomorphic functions on a Riemann surface M are defined
as all analytic f : M C. They are denoted by H(M ). The meromorphic functions on
M are defined as all analytic f : M C . They are denoted by M(M ).
In view of the preceding the following statements are immediate.
Corollary 4.8. Let M be a Riemann surface. The the following properties hold:
i) if M is compact, then every holomorphic function on M is constant.
ii) Every meromorphic function on a compact Riemann surface is onto C .
iii) If f is a nonconstant holomorphic function on a Riemann surface, then |f | attains
neither a local maximum nor a positive local minimum on M .
4. DEGREE AND GENUS
69
To illustrate what we have accomplished so far, let us give a topological proof of

Liouvilles theorem: thus assume that f H(C) L (C). Then f (1/z) has a removable
singularity at z = 0. In other words, f H(C ) and thus constant. The analytical
ingredient in this proof consists of the uniqueness and open mapping theorems as well as
the removability theorem: the first two are reduced to the same properties in charts which
then require power series expansions. But instead of using expansions that converge on
all of C and Cauchys estimate we used connectivity to pass from a local property to a
global one.
It is a good exercise at this point to verify the following: the meromorphic functions
on C in the sense of standard complex analysis coincide exactly with M() \ {}
in the sense of Definition 4.7 (we need to discard the function which is constant equal to
infinity). In particular,
o
nP
| P, Q C[z], Q 6 0 {}
M(C ) =
Q
In other words, the meromorphic functions on C up to the const = are exactly the
rational functions. Note that we may prescribe the location of the finitely many zeros
and poles of f M(CP 1 ) arbitrarily provided the combined order of the zeros exactly
equals the combined order of the poles and provided the set of zeros is distinct from the
set of poles (construct the corresponding rational function).
4. Degree and genus
Definition 4.9. Let f : M N be analytic and nonconstant. Then the valency of f
at p M , denoted by f (p), is defined to be the unique positive integer with the property
that in charts (U, ) around p (with f (p) = 0) and (V, ) around f (p) (with (f (p)) = 0)
we have ( f 1 )(z) = (zh(z))n where h(0) 6= 0. If M is compact, then the degree of
f at q N is defined as
X
degf (q) :=
f (p)
p:f (p)=q
which is a positive integer.
Locally around any point p M with valency f (p) = n 1 the map f is n-to-one;
in fact, every point q close but not equal to q = f (p) has exactly n pre-images close to
p.
P
be a nonconstant rational function on C represented by a reduced
Let f = Q
fraction (i.e., P and Q are relatively prime). Then for every q C , the reader will
easily verify that degf (q) = max(deg(Q), deg(P )) where the degree of P, Q is in the sense
of polynomials. It is a general fact that degf (q) does not depend on q N .
Lemma 4.10. Let f : M N be analytic and nonconstant with M compact. Then
degf (q) does not depend on q. It is called the degree of f and denoted by deg(f ). The
isomorphisms from M to N are precisely those nonconstant analytic maps f on M with
deg(f ) = 1.
Proof. Recall that f is necessarily onto N . We shall prove that degf (q) is locally
constant. Let f (p) = q and suppose that f (p) = 1. As remarked before, f is then an
isomorphism from a neighborhood of p onto a neighborhood of q. If, on the other hand,
n = f (p) > 1, then each q close but not equal to q has exactly n preimages {pj }nj=1 and
70
4. RIEMANN SURFACES
f (pj ) = 1 at each 1 j n. This proves that degf (q) is locally constant and therefore globally constant by connectivity of N . The statement concerning isomorphisms is
evident.

We remark that this notion of degree coincides with the usual one from differentiable
manifolds, see Chapter 12. Let us now prove the RiemannHurwitz formula for branched
covers. If necessary, the reader should review the Euler characteristic and the genus on
compact surfaces, see Chapter 12. This simply refers to an analytic nonconstant map
f : M N from a compact Riemann surface M onto another compact surface N .
Theorem 4.11 (RiemannHurwitz). Let f : M N be an analytic nonconstant
map between compact Riemann surfaces. Define the total branching number to be
X
B :=
(f (p) 1)
pM
Then
1
gM 1 = deg(f )(gN 1) + B
2
where gM and gN are the genera of M and N , respectively. In particular, B is always
an even nonnegative integer.
(4.5)
Proof. Denote by B all p M with f (p) > 1 (the branch points). Let T be a
triangulation of N such that all f (p), p B are vertices of T . Lift T to a triangulation
Te on M . If T has V vertices, E edges and F faces, then T has nV B vertices, nE
edges, and nF faces where n = deg(f ). Therefore, by the EulerPoincare formula (12.1),
2(1 gN ) = V E + F
as claimed.
2(1 gM ) = nV B nE nF = 2n(1 gN ) B
5. Riemann surfaces as quotients

Many Riemann surfaces M are generated as quotients of other surfaces N modulo an
equivalence relation, i.e., M = N/ . A common way of defining the equivalence relation
is via the action of a subgroup G Aut(N ). Then q1 q2 in N iff there exists some
g G with gq1 = q2 . Let us state a theorem to this effect where N = C . Examples
will follow immediately after the theorem. First, let us recall the notion of a covering
map.
Definition 4.12. Let X, Y be topological spaces and f : Y X is called S
a covering
map if and only if every x X has a neighborhood U in X so that f 1 (U ) = j Vj with
open and disjoint Vj Y so that f : Vj U is a homeomorphism for each j.
In particular, a covering map is a local homeomorphism. For example exp : C C

is a covering map, as is z n : C C for each n 1. Note that if n 2, then the latter
example does not extend to a covering map C C; rather, we encounter a branch point
at zero and this extension is then referred to as a branched cover.
Theorem 4.13. Let C and G Aut(C ) with the property that
g() for all g G
for all g G, g 6= id, all fixed points of g in C lie outside of
5. RIEMANN SURFACES AS QUOTIENTS
71
let K be compact. Then the cardinality of {g G | g(K) K 6= } is finite.

Under these assumptions, the natural projection : /G is a covering map which
turns /G canonically into a Riemann surface.
Proof. By definition, the topology on /G is the coarsest one that makes continuous. In this case, is also open; indeed, for every open A ,
[
1 ((A)) =
g(A)
gG
is open since g(A) is open. Next, let us verify that the topology is Hausdorff. Suppose
(z1 ) 6= (z2 ) and define for all n 1,

r
An := z |z z1 | <
n

r
Bn := z |z z2 | <
n
where r > 0 is sufficiently small. Define K := A1 B 1 and suppose that An Bn 6=
for all n 1. Then for some an An and gn G we have
gn (an ) Bn
n1
Since in particular gn (K) K 6= , we see that there are only finitely many possibilities
for gn and one of them therefore occurs infinitely often. Let us say that gn = g G for
infinitely many n. Passing to the limit n implies that g(z1 ) = z2 or (z1 ) = (z2 ),
a contradiction. For all z we can find a small pre-compact open neighborhood of z
denoted by Kz , so that
(4.6)
g(Kz ) Kz =
g G, g 6= id
where we are using all three assumptions. Then : Kz Kz is the identity and therefore
we can use the Kz as charts. Note that the transition maps are given by g Aut(C )
(which
are Mobius transformations) and are therefore holomorphic. Finally, 1 (Kz ) =
S
1
1
gG g (Kz ) with pair wise disjoint open g (Kz ). The disjointness follows from (4.6)
and we are done.

We remark that any group G as in the theorem is necessarily discrete in the topological sense. First, this is meaningful as G Aut(C) = P SL(2, C) with the latter
carrying a natural topology; second, if G is not discrete, then the third requirement in
the theorem will fail (since we can find group elements in G as close to the identity as we
wish). There are many natural examples to which this theorem applies (in what follows,
we use hg1 , g2 , . . . , gk i to denote the group generated by these k elements) and as usual,
H is the upper half-plane.
1) The punctured plane and disk: C/hz 7 z + 1i C where the isomorphism is
given by the exponential map e2iz . Here = C, and G = hz 7 z + 1i. Similarly,
H/hz 7 z + 1i D .
2) The tori: Let 1 , 2 C be linearly independent over R. Then
C/hz 7 z + 1 , z 7 z + 2 i
is a Riemann surface. It is the same as C/ with the lattice

(4.7)
= {n1 + m2 | n, m Z}
In Figure 4.4 the lattice is generated by any distinct pair of vectors from {1 , 2 , 3 }
72
4. RIEMANN SURFACES
3
1
Figure 4.3. A lattice in C

which proves that a pair of generators is not unique. Furthermore, up to conformal
equivalence we may always assume that 1 = 1 and 2 = where Im ( ) > 0. A special
case is given by
C /hz 7 zi C/hz 7 z + 1, z 7 z +
1
log i
2i
where > 1. The same exponential map as in 1) induces the isomorphism here. A very
important question is to determine the so-called moduli space of tori, which is defined
to be the space of all conformal equivalence classes of tori. For this, see Problem 4.3.
3) The annuli: consider H/hz 7 zi with > 1. Then log z maps this onto
{w C : 0 < Im w < i, 0 Re w log }
with the sides Re w = 0, Re w = log identified. Next, send this via the conformal map
w
w 7 exp 2i
log
2
2
onto the annulus r := {r < |z| < 1} where log r = log
. We leave it to the reader
to check that no two r are conformally equivalent (hence, the moduli space of tori
{z C : r1 < |z| < r2 } with 0 < r1 < r2 is the same as all rr21 , i.e., (0, ).
This list of examples is important for a number of reasons. First, we remark that
we have exhausted all possible examples with = C. Indeed, we leave it to the reader
to verify that all nontrivial discrete subgroups of Aut(C) that have no fixed point are
either hz 7 z + i with 6= 0, or hz 7 z + 1 , z 7 z + 2 i with 1 6= 0, 2 /1 6 R,
see Problem 4.1. Second, C , D , r and C/ where is a lattice, is a complete list
of Riemann surfaces (up to conformal equivalence, of course) with nontrivial, abelian
fundamental group. This latter property is of course not so easy to see, cf. Chapter 11
as well as [16], Chapter IV.6.
6. ELLIPTIC FUNCTIONS
73
6. Elliptic functions
For the remainder of this section, we let
M = C/hz 7 z + 1 , z 7 z + 2 i
be the torus of Example 2). As usual, we refer to the group relative to which we are
factoring as the lattice . We first remark that 1 := a1 + b2 , 2 := c1 + d2
is another generator of the same lattice iff a, b, c, d Z and ad bc = 1. Thus, in
Figure 4.4 we can pass to other generators as well as other fundamental regions. The
latter here refers to any closed connected set P C with the property that
i) every point in z is congruent modulo the lattice to some point of P
ii) no pair of points from the interior of P are congruent
In Figure 4.4 the parallelogram spanned by 1 , 2 is one such fundamental region,
whereas the parallelogram spanned by 1 , 3 is another. An important, and in some
ways canonical choice of such a region is given by the Dirichlet polygon, see Problem 4.2.
Let us now turn to the study of meromorphic functions on the torus M . By definition
M(M ) = {f M(C) | f = f ( + 1 ) = f ( + 2 )}
where we ignore the function constant equal . These are called doubly periodic or elliptic
functions. First, since M is compact the only holomorphic functions are the constants.
Next, we claim that any nonconstant function f M(M ) satisfies deg(f ) 2. Indeed,
suppose deg(f ) = 1. Then, in the notation of RiemannHurwitz, B = 0 and therefore
1 = gM = gS 2 = 0, a contradiction. The reader should check that we can arrive at the
same conclusion by verifying that
I
f (z)
dz = 0
P f (z)
which implies that the sum of the residues inside P is zero (here P is a fundamental
domain so that f has neither zeros no poles on its boundary). Notice also, from Riemann
Hurwitz, that any elliptic function f with deg(f ) = 2 satisfies B = 4 and therefore has
exactly four branch points each with valency 2. An interesting question concerns the
existence of elliptic functions of minimal degree, viz. deg(f ) = 2. We shall now present
the classical Weierstrass function which is of this type. It is an important fact that all
elliptic functions can be expressed in terms of this one function, see Proposition 4.16.
Proposition 4.14. For any n 3, the series
X
f (z) =
(z + w)n
w
defines a function f M(M ) with deg(f ) = n. Furthermore, the Weierstrass function

X

1
(z + w)2 w2 ,
(z) := 2 +
z
is an elliptic function of degree two with as its group of periods. Here is as in (4.7)
and := \ {0}.
Proof. If n 3, then we claim that
f (z) =
(z + w)n
74
4. RIEMANN SURFACES
converges absolutely and uniformly on every compact set K C \ . Indeed, there exists
C > 0 such that
C 1 (|x| + |y|) |x1 + y2 | C(|x| + |y|)
for all x, y R. Hence, when z {x1 + y2 | 0 x, y 1}, then
|z + (k1 1 + k2 2 )| C 1 (|k1 | + |k2 |) |z| (2C)1 (|k1 | + |k2 |)
provided |k1 | + |k2 | is sufficiently large. Since

X
|k1 1 + k2 2 |n <
|k1 |+|k2 |>0
as long as n > 2, we conclude f H(C \ ). Since f (z) = f (z + 1 ) = f (z + 2 ) for all

z C \ , it is clear that f M(M ).
For the second part, we note that

C
|z||z + w|

(z + w)2 w2
2
2
|w| |z + w|
|w|3
provided |w| > 2|z| so that the series defining converges absolutely and uniformly on
compact subsets of C \ . For the periodicity of , note that
X
(z) = 2
(z + w)3
w
is periodic relative to the lattice . Thus, for every w ,

(z + w) (z) = C(w)
zC
with some constant C(w). Expanding around z = 0 yields C(w) = 0 as desired.
Another way of obtaining the function is as follows: let be defined as the Weierstrass product
Y
(4.8)
(z) := z
E2 (z/)
with canonical factors E2 as in Chapter 2. Then is entire with simple zeros precisely
at the points of . Now let
X h 1
(z)
1
1
z i
(z) =
= +
+ + 2
(z)
z
z
By inspection, =
The function has many remarkable properties, the most basic of which is the
following differential equation.
Lemma 4.15. With as before, one has
(4.9)
( (z))2 = 4((z) e1 )((z) e2 )((z) e3 )
where e1 = (1 /2), e2 = (2 /2), and e3 = ((1 + 2 )/2). Furthermore, one has

e1 + e2 + e3 = 0 so that (4.9) can be written in the form
( (z))2 = 4((z))3 g2 (z) g3
with constants gj (1 , 2 ), j = 2, 3.
6. ELLIPTIC FUNCTIONS
Proof. By inspection,
(z) = 2
75
(z + w)3
is an odd function M(M ) of degree three. Thus,
(1 /2) = (1 /2) = (1 /2) = 0
Similarly, (2 /2) = ((1 +2 )/2) = 0. In other words, the three points 1 /2, 2 /2, (1 +
2 )/2 are the three zeros of and thus also the unique points where has valency two
apart from z = 0. Denoting the right-hand side of (4.9) by F (z), this implies that
(z)
F (z) H(M ) and therefore equals a constant. Considering the expansion of (z) and
F (z), respectively, around z = 0 shows that the value of this constant equals 1, as
claimed. The final statement follows by observing from the Laurent series around zero
that
( (z))2 4((z))3 g2 (z)
for suitable g2 is analytic and therefore constant.

The previous proof shows that 0, 1 /2, 2 /2, (1 + 2 )/2 are precisely the branch
points of . We are now able to establish the following property of .
Proposition 4.16. Every f M(M ) is a rational function of and . If f is
even, then it is a rational function of alone.
Proof. Suppose that f is nonconstant and even. Then for all but finitely many
values of w C , the equation f (z) w = 0 has only simple zeros (since there are only
finitely many zeros of f ). Pick two such w C and denote them by c, d. Moreover, we
can ensure that the zeros of f c and f d are distinct from the branch points of .
Thus, since f is even and with 2n = deg(f ),
{z M : f (z) c = 0} = {aj , aj }nj=1
{z M : f (z) d = 0} = {bj , bj }nj=1

The elliptic functions
g(z) :=
and
h(z) :=
f (z) c
f (z) d
n
Y
(z) (aj )
(z) (bj )
j=1
have the same zeros and poles which are all simple. It follows that g = h for some
6= 0. Solving this relation for f yields the desired conclusion.
If f is odd, then f / is even so f = R() where R is rational. Finally, if f is any
elliptic function then
1
1
f (z) = (f (z) + f (z)) + (f (z) f (z))
2
2
is a decomposition into even/odd elliptic functions whence
f (z) = R1 () + R2 ()
with rational R1 , R2 , as claimed.
76
4. RIEMANN SURFACES
For another result along these lines see Problem 4.4. It is interesting to compare
the previous result for the tori to a similar one for the simply periodic functions, i.e.,
functions on the surface C/hz 7 z + 1i C . These can be represented via Fourier
series, i.e, infinite expansions in the basis e2iz which plays the role of in this case.
The reason we obtain infinite expansions rather than the finite ones in the case of tori
lies with the fact that the latter are compact whereas C is not.
We conclude our discussion of elliptic functions by turning to the following natural
question: to given disjoint finite sets of distinct points {zj } and {k } in M as well as
positive integers nj for zj and k for k , respectively, is there an elliptic function with
precisely these zeros and poles and the given orders? We remark that for the case of C
the answer was affirmative if and only if the constancy of the degree was not violated,
i.e.,
X
X
(4.10)
nj =
k
j
For the tori, however, there is a new obstruction:

X
X
(4.11)
nj zj
k k = 0
j
mod
This follows from the residue theorem since

I
X
X
f (z)
1
z
dz =
nj zj
k k
(4.12)
2i P f (z)
j
where P is the boundary of a fundamental domain P as in Figure 4.4 so that no zero or

pole of f lies on that boundary. Comparing parallel sides of the fundamental domain and
using the periodicity shows that the left-hand side in (4.12) is of the form n1 1 + n2 2
with n1 , n2 Z and thus 0 in .
It is a remarkable fact that (4.10) and (4.11) are also sufficient for the existence of
an elliptic function with precisely these given sets of zeros and poles.
Theorem 4.17. Suppose (4.10) and (4.11) hold. Then there exists an elliptic function which has precisely these zeros and poles with the given orders.
Proof. Listing the points zj and k with their respective multiplicities, we obtain
sequences zj and k of the same length, say n. Shifting by a lattice element if needed,
one has
n
n
X
X
k
zj =
j=1
k=1
Define, see (4.8) for ,

n
Y
(z zj )
f (z) =
(z j )
j=1
which has the desired zeros and poles. It remains to check the periodicity which we leave
to the reader.
7. PROBLEMS
77
7. Problems
Problem 4.1. Show that all nontrivial discrete subgroups of Aut(C) that have no
fixed point are either hz 7 z + i with 6= 0, or hz 7 z + 1 , z 7 z + 2 i with
1 6= 0, 2 /1 6 R.
Problem 4.2. Let C be the lattice (i.e., the discrete subgroup) generated by
1 , 2 which are independent over R. Show that the Dirichlet polygon
{z C : |z| |z |
is a fundamental region, cf. Chapter 4.

a b
Problem 4.3. Let A =
GL(2, C) and = hz 7 z + 1 , z 7 z + 2 i with
c d
1 , 2 linearly independent over R. Prove that = hz 7 z + 1 , z 7 z + 2 i where

1
1
=A
2
2
if and only if A GL(2, Z) with det(A) = 1. Now show that the moduli space of tori
as defined in Section 4 of Chapter 4 is H/P SL(2, Z). The group P SL(2, Z) is called
modular group, see for example [22] as well as many other sources.
Problem
4.4. Let M = C/ where is the lattice generated by 1 , 2 C with

1
Im 2 6= 0. As usual P denotes the Weierstrass function on M . Suppose that f

a b
M(M ) has degree two. Prove that there exists A =
SL(2, C) and w C such
c d
that
a(z w) + b
f (z) =
c(z w) + d
is compact and is a
Problem 4.5. Suppose N is a Riemann surface such that N
manifold with boundary. I.e., for every p N there exists a neighborhood U of p in
and a map : U R2+ such that takes U homeomorphically onto D {Im z 0}.
N
Moreover, we demand that the transition maps between such charts are conformal on
M where M is a Riemann surface. In other words, N
Im z > 0. Prove that then N
can be extended to a strictly larger Riemann surface.
Problem 4.6. Let M be a compact Riemann surface and S M discrete. Suppose
f : M \ S C is analytic and nonconstant. Show that the image of M \ S under f is
dense in C.
Problem 4.7. Fill in the missing details in Corollary 5.20. I.e., first check that
is indeed a homeomorphism from C/ onto E. Second, verify the integrals (5.11).
Problem 4.8. Let M, N be compact Riemann surfaces and suppose : M \ S
N \ S is an isomorphism where S, S are finite sets. Then extends to an isomorphism
from M N .
Problem 4.9. This exercise revisits fractional linear transformations.
(a) Prove that
n a b
o
G=
: a, b C, |a|2 |b|2 = 1
b a
78
4. RIEMANN SURFACES
is a subgroup of SL(2, C) (it is known as SU (1, 1)). Establish the group isomorphism
G/{I} Aut(D) in two ways: (i) by showing that each element of G defines a fractional
linear transformation which maps D onto D; and conversely, that every such fractional
linear transformation arises in this way uniquely up to the signs of a, b. (ii) By showing
that the map
i
i
e 2
z0 e 2
z
z
1|z0 |
1|z0 |
0
7
(4.13)
e2i
i
i
1 z0 z
z0 e 2 e
2
1|z0 |
1|z0 |
leads to an explicit isomorphism.

(b) We know from a previous problem that Aut(C ) is the group of all fractional
linear transformations, i.e.,
Aut(C ) = P SL(2, C) = SL(2, C)/{Id}
and that each such transformation induces a conformal homeomorphism of S 2 (indeed,

the stereographic projection is conformal). The purpose of this exercise is to identity the
subgroup Grig of those transformations in Aut(C ) which are isometries (in other words,
rigid motions) of S 2 (viewing C as the Riemann sphere S 2 ). Prove that
where
Grig SO(3) SU (2)/{I}

n

o
a b
SU (2) =
: a, b C, |a|2 + |b|2 = 1
b a
Find the fractional linear transformation which corresponds to a rotation of S 2 of angle
2 about the x1 -axis.

(c) Show that the quaternions can be viewed as the four-dimensional real vector space
spanned by the basis

1 0
i 0
0 1
0 i
e1 =
, e2 =
, e3 =
, e4 =
0 1
0 i
1 0
i 0
and with the algebra structure being defined via the matrix products of the ej s (typically,
one writes 1, i, j, k instead of e1 , e2 , e3 , e4 ). Show that in this representation the unit
quaternions are nothing but SU (2) and exhibit a homomorphism Q of the unit quaternions onto SO(3) so that ker(Q) = {1}.
Which rotation does the unit quaternion 1 + 2 i + 3 j + 4 k represent (i.e., what are
the axis and angle of rotation)?
CHAPTER 5
Analytic continuation, covering surfaces, and algebraic

functions
This chapter takes us back to the origins of Riemann surfaces as a way of explaining
multi-valued functions arising through analytic continuation. As an example, consider
the polynomial equation
(5.1)
w4 2w2 + 1 z = 0
It
has
a
solution
(z,
w)
=
(1,
2) and near z = 1 this gives rise to the function w1 (z) =
p
1 + z which solves (5.1); the convention here is that z > 0 if z > 0.

there
pHowever,
are other solutions, namely (z, w) = (1, 2) leading to w2 (z) = 1 + p

z as well
as (z, w) = (1, 0). The latter formally corresponds to the functions w(z) = 1 z
which are not analytic near z = 0. Thus, (1, 0) is referred to as a branch point and it is
characterized as a point obstruction to analytic continuation. The purpose of this chapter
is to put this example, as well as all such algebraic equations, on a solid foundation in
the context of Riemann surfaces.
Some of the material in this chapter may seem overly abstract due to the cumbersome definitions we will have to work through. The reader should therefore always try
to capture the simple geometric ideas underlying these notions. To begin with, we define
function elements or germs and their analytic continuations. There is a natural equivalence relation on these function elements leading to the notion of a complete analytic
function.
Definition 5.1. Let M, N be fixed Riemann surfaces. A function element is a pair
(f, D) where D M is a connected, open non-empty subset of M and f : D N is
analytic. We say that two function elements (f1 , D1 ) and (f2 , D2 ) are direct analytic
continuations of each other iff
D1 D2 6= ,
f1 = f2 on D1 D2
Note that by the uniqueness theorem on Riemann surfaces there is at most one f2
that makes (f2 , D2 ) a direct analytic continuation of (f1 , D1 ). This relation, denoted by
, is reflexive and symmetric but not transitive, cf. Figure 5.1. On the other hand, it
gives rise to an equivalence relation, denoted by , in the following canonical way:
e are analytic
Definition 5.2. We say that two function elements (f, D) and (g, D)
continuations of each other iff there exist function elements (fj , Dj ), 0 j J such that
e and (fj , Dj ) (fj+1 , Dj+1 ) for each 0 j < J.
(f0 , D0 ) = (f, D), (fJ , DJ ) = (g, D),
The complete analytic function of (f, D) is simply the equivalence class [(f, D)] of
this function element under .
79
80
5. COVERING SURFACES AND ALGEBRAIC FUNCTIONS
f1 = f 2
f1
?
f1 = f3
11111111
00000000
00000000
11111111
00000000
11111111
00000000
11111111
00000000
11111111
f2
111111
000000
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
000000
111111
f3
1111111
0000000
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000 =
1111111
f2
f3
Figure 5.1. Failure of transitivity

Our main goal for now is to analyze the complete analytic function in more detail. But
before doing so, we need to clarify the process of analytic continuation on a Riemann
surface some more. In particular, we wish to prove an analogue of the monodromy
theorem, Theorem 2.16, for Riemann surfaces. To begin with, we remark that the notion
of analytic continuation from Definition 5.2 agrees with analytic continuation along paths.
Lemma 5.3. Two function elements (f1 , D1 ) and (f2 , D2 ) are analytic continuations
of each other in the sense of Definition 5.2 iff there exists a continuous path : [0, 1] M
connecting a point of D1 to one of D2 with the following property: there exists a partition
0 = t0 < t1 < . . . < tm = 1 so that for each 0 i < m, ([ti , ti+1 ]) belongs to a single
parametric disk of M and analytic continuation relative to each of these local coordinates
in the sense of Definition 2.14 leads (necessarily uniquely) from f1 defined around (0)
to f2 defined around (1).
Proof. This is an immediate consequence of the path-connectedness of M and the
uniqueness theorem on Riemann surfaces. It there is such a path , then we let the
(fj , Dj ) in Definition 5.2 be given by evaluating the analytically continued function along
a sufficiently fine net of points on . Conversely, connecting points from the sets Dj by
continuous paths and partitioning the path into sufficiently many intervals shows that
f2 is obtained from f1 via continuation in local coordinates.

As an important corollary, we can now state the monodromy theorem.
Corollary 5.4. Let H : [0, 1]2 M be a homotopy with fixed endpoints p and q.
Suppose (f, D) is a function element with p D which can be analytically continued along
each path H(, s) of the homotopy leading to a function element (gs , Ds ) with q Ds ,
0 s 1. The g0 = g1 on D0 D1 . If H does not fix end points but each path
H(, s) is a closed curve, then the following holds: suppose (f0 , D0 ) is a function element
with H(0, 0) D0 which can be analytically continued along H(0, s) yielding function
elements (fs , Ds ) with H(0, s) Ds , 0 s 1. Assume further that each (fs , Ds ) can
e s ). If f0 = g0 on D0 D
e 0 then
be analytically continued along H(, s) yielding (gs , D
e
f1 = g1 on D1 D1 .
81
H(0,s)
q
H(t,0)
H(t,1)
Figure 5.2. A homotopy of closed curves

Proof. We begin with the fixed endpoint case. By the previous lemma, analytic
continuation can be carried out in local coordinates. More precisely, we can cover each
path by parametric disk and analytically continue locally in each of these disks. We
can use the same disks as long as we move the parameter s by very little. But then
the analytic continuation does not depend on s since it has the same property in local
coordinates. For the closed curves version, we argue analogously, covering by parametric
disks and changing s by very little. Figure 5.1 explains the meaning of the monodromy
statement for closed curves. The short arc connecting p and q is H(0, s) and we are
assuming that we are allowed to analytically continue fs along all curves t 7 H(t, s).
The point is that if we come back to the same function at p, then we have to come back
to the same function at q.

The monodromy theorem of course implies that on simply connected Riemann surfaces M any function element (f, D) that can be analytically continued everywhere on M
defines a global analytic function on M . Another way of arriving at the same conclusion
is furnished by the following sheaf theoretic device that can be a very useful tool. The
logic is as follows: if on a simply connected Riemann surface we can always glue function
elements together locally, then this can also be done globally.
Lemma 5.5. Suppose M is a simply connected Riemann surface and {D M :
A} a collection of domains (connected, open). Assume further that M = A D and
that for each A there is a family F of analytic functions f : D N such that if
f F and p D D then there is some g F so that f = g near p. Then, given
A and some f F there exists an analytic function F : M N so that F = f
on D .
Proof. Let
U := {(p, f ) | p D , f F , A}/
where (p, f ) (q, g) iff p = q and f = g in a neighborhood of p. Let [p, f ] denote the
equivalence class of (p, f ). As usual, ([p, f ]) := p. For each f F , let
:= {[p, f ] | p D }
D,f
82
Clearly, : D,f
D is bijective. We define a topology on U as follows: D,f
is
open iff () D is open. An arbitrary U is open iff D,f is open for each
) is the union of
D,g
, f F . This does indeed define open sets in U : since (D,f
connected components of D D by the uniqueness theorem (if it is not empty), it is
open in M as needed. With this topology, U is a Hausdorff space since M is Hausdorff
(we use this if the base points differ) and because of the uniqueness theorem (which we
use if the base points coincide).
D
D
Figure 5.3. Local gluing and simple connectivity implies global gluing
Note that by construction, we have made the fibers indexed by the functions in F
f is a connected
discrete in the topology of U . The main point is now to realize that, if M
f M is onto and in fact is a covering map. Let us check
component of U , then : M
f) M is open. Thus, let [p, f ] M
f and
that it is onto. First, we claim that (M
f 6= and since D , and thus

M
pick D with p D and f F . Clearly, D,f
f has to contain D
also D , are open and connected the connected component M
,f
,f
f) as claimed. Next, we need to check that M \ (M

f)
entirely. Therefore, D (M
f) and pick D so that p D . If D (M
f) = then we
is open. Let p M \ (M
f) and pick D containing q and some f F
are done. Otherwise, let q D (M
f (we are using the same nonempty intersection implies containment

with D,f M
argument as above). But now we can find g F with the property that f = g on a
f would have to contain D which
component of D D . As before, this implies that M
,g
f
is a contradiction. To see that : M M is a covering map, one verifies that
[
1 (D ) =
D,f
f F
The sets on the right-hand side are disjoint and in fact they are the connected components
f is homeomorphic to M , see Chapter 12.
of 1 (D ). Since M is simply connected, M
This property reduces to the existence of a globally defined analytic function which agrees
2. THE UNRAMIFIED RIEMANN SURFACE OF AN ANALYTIC GERM
83
with some f F on each D . By picking the connected component that contains any
given D,f
one can fix the sheet locally on a given D .

The reader should try to apply this proof to the case where M = C , D are all
possible disks in C , and F all possible branches of the logarithm on the disk D . What
f here, and is still a covering map?
are all possible M
Arguments as in the previous lemma are very powerful and allow one to base the theory of analytic continuations and thus the constructions of the following section entirely
on the theory of covering spaces (for example, the mondromy theorem then becomes the
well-known invariance of lifts under homotopies). For such an approach see [15]. The
author feels, however, that using these sheaves is somewhat less constructive (due to
arguments by contradiction as in the previous proof) and he has therefore chosen to
follow the more direct traditional approach based on analytic continuation along curves.
2. The unramified Riemann surface of an analytic germ
Heuristically, we can regard the complete analytic function from Definition 5.2 as a
f as follows: writing
analytic function F defined on a new Riemann surface M
[(f, D)] = {(f , D ) | A}
we regard each D as distinct from any other D (even if D = D ). Next, define f = f

on D . Finally, identify p D with q D iff (i) p = q when considered as points in
M and (ii) f = f near p. In other words, we let the functions label the points and only
identify if we have local agreement. You should convince yourselfthat this is precisely
the naive way in which we picture the Riemann surfaces of log z, z etc.
In the following lemma, we prove that this construction does indeed give rise to a
f and a function defined on it. Throughout, M, N will be fixed RieRiemann surface M
mann surfaces and any function element and complete analytic function will be defined
relative to them.
Lemma 5.6. (a) Given a complete analytic function A and p M , we define an
equivalence relation p on function elements in {(f, D) A | p D} as follows:
(f0 , D0 ) p (f1 , D1 ) f0 = f1 near p
We define [f, p] to be the equivalence class of (f, D), p D under p and call this a
germ. Then the germ [f, p] uniquely determines three things: the point p, the value f (p),
and the complete analytic function A.
(b) Let [f0 , p0 ] be a germ and let A = A(f0 , p0 ) be the associated complete analytic
function. Define
RS(M, N, f0 , p0 ) = {[f, p] | p D, (f, D) A}
and endow this set with a topology as follows: the base for the topology is
[f, D] = {[f, p] | p D, },
(f, D) A
With this topology, RS(M, N, f0 , p0 ) is a two-dimensional, arcwise connected, Hausdorff

manifold with a countable base for the topology.
(c) On RS(M, N, f0 , p0 ) there are two natural maps: the first is the canonical map
: RS(M, N, f0 , p0 ) M defined by ([f, p]) = p. The second is the analytic continuation of (f0 , p0 ), denoted by F , and defined as F ([f, p]) = f (p). The map is a local
homeomorphism and thus defines a complex structure on RS(M, N, f0 , p0 ) which renders
84
a local conformal isomorphism. Hence RS = RS(M, N, f0 , p0 ) is a Riemann surface,

called the unramified Riemann surface of the germ [f0 , p0 ] and F is an analytic function
RS N .
Proof. (a) It is clear that the germ determines p as well as the Taylor series at p.
(b) M is arcwise connected and RS(M, N, f0 , p0 ) is obtained by analytic continuation along curves so it, too, is arcwise connected. If two points [f, p] and [g, q]
in RS(M, N, f0 , p0 ) satisfy p 6= q, then use that M is Hausdorff. If p = q, then the germs
are distinct and can therefore be separated by open connected neighborhoods via the
uniqueness theorem. For the countable base, use that M satisfies this and check that
only countably many paths are needed to analytically continue a germ.
(c) The statements regarding and F are clear.

The map in part (c) does not need to be onto. Indeed, suppose f H(D) cannot
be analytically continued to any larger region than D. In that case, RS(C, C, f, 0) = D,
(z) = z and F = f . As another example, consider the function elements given by log z
1
or z n with integer n > 1 on a neighborhood of z = 1. Then we cannot analytically
continue into z = 0 which leads us to the notion of a branch point of the Riemann
surface constructed in the previous lemma. We remark that these two classes of examples
(logarithms and roots) are representative of all possible types of branch points and that
in the case of the roots there is a way to adjoin the branch point to RS(M, N, f0 , p0 )
and to make it essentially disappear, see below.
Definition 5.7. Let (U, ) be a chart at p1 M with (p1 ) = 0, (U ) = D. Let
(f, p) RS(M, N, f0 , p0 ) with p U \ {p1 }. If (f, p) can be analytically continued
along every path in U \ {p1 } but not into p1 itself, then we say that RS(U, N, f, p)
represents a branch point of RS = RS(M, N, f0 , p0 ) rooted at p1 . Under a branch point
p1 rooted at p1 M we mean an equivalence class as follows: suppose RS(U, N, f, p)
and RS(V, N, g, q) each represent a branch point of RS rooted at p1 . We say that they
are equivalent iff there is another such RS(W, N, h, r) with
RS(W, N, h, r) RS(U, N, f, p) RS(V, N, g, q)
The reader should convince himself or herself that RS(U, N, f, p) is not necessarily
the same as 1 (U ) (it can be smaller, see Figure 5.5). This is why we need to distinguish
between p1 and its root p1 in M . We now define the branching number at a branch point.
Definition 5.8. Let p1 M be the root of some branch point p1 and pick RS(U, N, f, p)
from the equivalence class of surfaces representing this branch point p1 as explained above.
Let (U ) = D, (p1 ) = 0 be a chart and let (t) = 1 ((p)e2it ) be a closed loop in
U around p1 . Then we let [fn , p] be the germ obtained by analytic continuation of [f, p]
along n = . . . (n-fold composition), n 1. We define the branching number at
p1 to be

iff [fn , p] 6= [f, p] n 1
(5.2)
B(p1 ) :=
min{n 1 | [fn , p] = [f, p]} 1 otherwise
If B(p1 ) = , then we say that p1 is a logarithmic branch point.
Figure 5.4 is a schematic view of a branch point with branching number equal to two.
We now need to check that these notions are well-defined. In what follows, we shall
freely use the simple fact that the winding number characterizes equivalence classes of
homotopic loops in D .
3. THE RAMIFIED RIEMANN SURFACE OF AN ANALYTIC GERM
85
Lemma 5.9. The branch number introduced in the previous definition is well-defined,
i.e., it does not depend on the representative RS(U, N, f, p). Furthermore, if RS(U, N, f, p)
is a representative of the branch point p1 rooted at p1 , then the following holds: let be
a closed loop in U \ {p1 }. Then [f, p] is invariant under analytic continuation along if
and only if
B(p1 ) + 1 divides n( ; 0)
Proof. Suppose RS(U, N, f, p) and RS(V, N, g, q) are equivalent in the sense of
Definition 5.7. Then there exists RS(W, N, h, r) with
RS(W, N, h, r) RS(U, N, f, p) RS(V, N, g, q)
In particular, the germ [h, r] is an analytic continuation of [f, p] as well as [g, q] along
paths in U \ {p} and V \ {q}, respectively. Let be the closed loop from Definition 5.8
for U , be the one for V , and the one for W . Since the winding number classifies
homotopy classes of closed curves in the punctured disk, we see that U in the sense
of homotopy relative to U := U \ {p1 }, as well as V relative to V := V \ {p1 }.
By the closed-curve version of the monodromy theorem, see Corollary 5.4, we conclude
that [f, p] and [h, r] yield the same number in (5.2) and by the same token, also [g, q] and
[h, r]. It therefore follows that B(p1 ) is well-defined on the equivalence class defining the
branch point.
For the final statement, suppose n0 := B(p1 ) + 1 does not divide n( ; 0). Then
n( ; 0) = kn0 + r0 ,
0 < r0 < n 0
Since a loop of winding number kn0 brings [f, p] back to itself, this implies that there
exists a loop of winding number r0 which does so, too. But this contradicts (5.2) and we
are done.

3. The ramified Riemann surface of an analytic germ
We now show that at each branch point p1 of RS with B(p1 ) < and for every representative RS(U ) = RS(U, N, f, p) of that branch point there is a chart defined globally
on RS(U ) (known as uniformizing variable) which maps RS(U ) bi-holomorphically onto
D . The construction is very natural and is as follows:
Lemma 5.10. With RS(U, N, f, p) representing a branch point, let : U D be
a chart that takes p1 7 0. Pick a path that connects [f, p] with an arbitrary [g, q]
1
RS(U ), and pick a branch 0 of the nth root z n locally around z0 = (p), n := B(p1 ) + 1.
Now continue the germ [0 , z0 ] analytically along the path to a germ [ , z] where
z = (q). Define ([g, q]) = (z). The map , once 0 has been selected, is well-defined.
Moreover, is analytic, and a homeomorphism onto D .
Proof. First, we check that the choice of does not affect . Let
e be another
path connecting [f, p] with [g, q] RS(U ). As usual,
e is the reversed path and e
is the composition. Analytically continuing [f, p] along e

then leads back to [f, p].
This implies that e

has winding number around zero which is divisible by n
(for otherwise we could obtain a smaller integer in (5.2)). Therefore, [0 , z0 ] is invariant
under analytic continuation along e
; in other words, (z) = (z) as was to
be shown. This also shows that is analytic since is a local homeomorphism as well
as a analytic map. is onto D by our standing assumption that analytic continuation
can be performed freely in U . Finally, we need to check that is one-to-one. Suppose
86
([g1 , q1 ]) = ([g2 , q2 ]). Then ([g1 , q1 ])n = ([g2 , q2 ])n which means that (q1 ) = (q2 )
and thus q1 = q2 . By construction, ([gj , qj ]) = j (z) where z = (q1 ). Since was
Figure 5.4. A schematic depiction of a uniformizing chart at a branch

point with branching number 2
1
obtained as analytic continuation of a branch of z n along , it follows that

n( 1 2 ; 0) = kn,
kZ
But then [g1 , q1 ] = [g2 , q2 ] by (5.2) and we are done.

1
Convince yourself that for the case of RS(C, C, z n , 1) one can think of 1 (z) = z n .
1
1
Obviously, in that case (1 (z)) n = (z n ) n = z for all z D . The point of our discussion
here is that locally at a branch point with finite branching number n 1 any unramified
Riemann surface behaves the same as the nth root. And, moreover, adjoining the branch
point p1 to the unramified surface yields another Riemann surface with a chart around
p1 that maps a neighborhood bi-holomorphically onto D. This is very relevant from the
point of view of analytically continuing the global function F into a branch point by
3. THE RAMIFIED RIEMANN SURFACE OF AN ANALYTIC GERM
87
means of the chart . This can indeed be done, at least in the algebraic case to which
we now turn.
Definition 5.11. We define the ramified Riemann surface by adjoining all algebraic
branch points. The latter are defined as being precisely those branch points with finite
branching number so that F (relative to the uniformizing variable ) has a removable
singularity at zero, i.e., F 1 : D N extends as an analytic function D N .
An example of a non-algebraic branch point with finite branching number is
1
RS(C , C , exp(z 2 ), 1)
Note that z = 0 is the root of an algebraic branch point (set z = 2 which yields e )
1
whereas z = is not algebraic since z = 2 with near zero leads to e which has
an essential singularity at = 0.
We can state precisely how to define the ramified Riemann surface.
Lemma 5.12. Let Pal be the set of algebraic branch points of RS(M, N, f0 , p0 ) and
define
g = RS(M,
g
RS
N, f0 , p0 ) := RS(M, N, f0 , p0 ) Pal
g is canonically a Riemann surface to which and F have analytic continuations
Then RS
g
g N , respectively. We call RS
g the ramified Riemann surface
e : RS M , and Fe : RS
(or just Riemann surface), and Fe the complete analytic function of the germ [f0 , p0 ]. At
each p Pal the branching number B(p) = (e
, p) 1 where is the valency as defined
earlier.
Proof. This is an immediate consequence of the preceding results and definitions,
in particular, of Lemma 5.10.

The term complete analytic function was introduced previously for the collection
of all function elements obtained via analytic continuation from a given one. However,
from now on we will use this term exclusively in this new sense. Next, we turn to the
important special case where the Riemann surface is compact.
g = RS(M,
g
Lemma 5.13. If RS
N, f0 , p0 ) is compact, then M is compact. Moreover,
there can only be finitely many branch points in R; we denote the set of their projections
g \ P M \ B (the restriction of
onto M by B and define P :=
e1 (B). The map
e1 : RS
g
e to RS \ P) is a covering map and the number of pre-images of this restricted map is

g and it equals the degree of
constant; this is called the number of sheets of RS
e. Finally,
the following Riemann-Hurwitz type relation holds:
1X
(5.3)
gRS
B(p)
g = 1 + S(gM 1) +
2 p
g
where gRS
g , gM are the respective genera, S is the number of sheets of RS, and the sum
runs over the branch points p in RS with B(p) being the respective branching numbers.
g that there can be only finitely many
Proof. It is clear from the compactness of RS
g
branch points. Note that RS \ P = RS \ P and
e1 : RS \ P M \ B is a local
homeomorphism which is also proper; i.e., the pre-images of compact sets in M \ B
g \ P is a covering map.
are compact. This then easily implies that
e restricted to RS
88
Figure 5.5 is a schematic depiction of a three-sheeted surface with two (finite) branch
g \ P in that case due to the fact that there are unbranched
points. Note that RS 6= RS
sheets covering the roots of branch points. It is clear that the cardinality of the fibres
branch points
roots
Figure 5.5. Three sheets

equals the degree of
e. Finally, the Riemann-Hurwitz relation follows from the general
Riemann-Hurwitz formula above applied to
e.

We remark that
e is what one calls a branched covering map.
4. Algebraic germs and functions
So far, our exposition has been very general in the sense that no particular kind
of function element was specified to begin with. This will now change as we turn to a
more systematic development of the ramified Riemann surfaces
of algebraic functions.
p
g of (z z1 ) (z zm ) where
For example, consider the ramified Riemann surface RS
e is compact, see below). In our notation,
zj C are distinct points (it is easy to see that R
we are looking at
p
g , C , (z z1 ) (z zm ), z0 ),
z0 6= zj 1 j m
RS(C
and with one of the two branches of the square root fixed at z0 . What is the genus of
g If m is even, then RS has M = m branch points, if m is odd, then is has M = m+1
RS?
branch points (the point at is a branch point in that case). In all cases, the branching
z1
sheet 1
z2
z4
z3
sheet 2
Figure 5.6. The case m = 2 with genus 1
4. ALGEBRAIC GERMS AND FUNCTIONS
89
numbers are one. The number of sheets is S = 2. Hence, in view of (5.3),

gRS
g =
M
1
2
In other words, the Riemann surface is a sphere with M

2 1 handles attached. This can
also be seen directly, by cutting the Riemann sphere along [z1 , z2 ], [z3 , z4 ], . . . , [zm1 , zm ]
if m is even. Then the surface consists of two copies of the Riemann sphere glued together
along these edges which results in genus m
2 1 as claimed. The reason for this cutting is of
sheet 1
sheet 2
1
Figure 5.7. The Riemann surface of

a global uniformizing map
+1
z 2 1 via the Jukowski map as
course to allow for the transition from one sheet to the other as we continue analytically
along a small loop centered at one of the zj . Note that we skipped the cut [z2 , z3 ] above
since analytic continuation along a loop surrounding the pair z1 , z2 but encircling none
of the other branch points does not change the sheet. If m is odd, then we also need to
introduce a cut from zm to . For the case of m = 2 it is possible to make all of this very
explicit. Indeed, setting z = 12 ( + 1 ) (this is the Jukowski map from Problem 1.9)
yields
p
1
(5.4)
(z 1)(z + 1) = ( 1 )
2
where we have made a choice of branch of the square root. Since the right-hand side
of (5.4) is analytic as a map from C to itself, we see immediately that the Riemann surface of the left-hand side is the Riemann sphere,
cf. Figure 5.7. Moreover, in the plane,
the two sheets of the Riemann surface of z 2 1 correspond precisely to the regions
{|| < 1} and {|| > 1}, respectively. Recall that 7 z is a conformal isomorphism from
each of these regions onto C \ [1, 1] with {|z| = 1} mapped onto [1, 1]. More precisely,
z = 1 are mapped onto 1, respectively, whereas each point z (1, 1) is represented
exactly twice in the form z = cos(), 0 < || < . Moreover, z () = 12 (1 2 ) = 0
exactly at = 1 and () 6= 0 at these points, which corresponds precisely to the simple branch points rooted at z = 1. This example serves to illustrate the uniformizing
charts from Lemma 5.10: at a branch point the ramified Riemann surface looks like every
other point in the plane. Finally, note that the map 7 1 switches these two sheets
and also changes the sign in (5.4) which needs to happen given the two different signs of
90
sheet 1
sheet 2
Figure 5.8. The two sheets form a Riemann sphere for m = 1

the square root. Clearly, the genus is zero, see also Figure 5.8. Another example of an
algebraic function is furnished by the ramified Riemann surface generated by any root of
P (w, z) = w3 3w z = 0 relative to the w variable. It is natural that the branch points
are given precisely by the failure of the implicit function theorem. In other words, by all
those pairs of (w, z) for which P = 0, w P = 0. This means that w = 1 and z = 2.
2 P 6= 0 at these points we conclude that the branching number equals two at
Since w
these points, whereas at z = the branching number equals three. Finally, the number
of sheets is three, so that the surface looks schematically like the one in Figure 5.5 (see
Figure 3.7.3 in Beardon [3] for a more artistic rendition of this surface). Finally, the
genus gRS
g = 1 from (5.3).
Throughout this section, M = N = CP 1 are fixed; in particular, note that analytic functions are allowed to take the value . As usual, we need to start with some
definitions:
Definition 5.14. An analytic germ [f0 , z0 ] is called algebraic iff there is a polynomial
P C[w, z] of positive degree so that P (f0 (z), z) = 0 identically for all z close to z0 .
The complete analytic function
1
g = RS(CP
g
Fe : RS
, CP 1 , f0 , z0 ) CP 1
generated by [f0 , z0 ] is called an algebraic function.
The following lemma develops some of the basic properties of algebraic functions. In
particular, we show that all branch points of an algebraic function are algebraic and that
g is compact.
RS
e be as in the previous definition. Then one has the following

Lemma 5.15. Let P and R
properties:
g In fact, the same is true with an irreducible factor
(a) P (f (z), z) = 0 for all [f, z] RS.
of P which is uniquely determined up to constant multiples.
(b) there is the following version of the implicit function theorem: Let P (w, z) C[w, z]
satisfy P (w1 , z1 ) = 0, Pw (w1 , z1 ) 6= 0. Then there is a unique analytic germ [f1 , z1 ] with
P (f1 (z), z) = 0 locally around z1 and with f1 (z1 ) = w1 .
91
P
(c) If P (w, z) = nj=0 aj (z)wj C[w, z] satisfies an 6 0, then up to finitely many z the
polynomial w 7 P (w, z) has exactly n simple roots.
(d) Given an algebraic germ [f0 , z0 ], there are finitely many points {j }Jj=1 CP 1 (called
critical points) such that [f0 , z0 ] can be analytically continued along every path in CP 1 \
{1 , . . . , J }. If the unramified Riemann surface
RS(CP 1 , CP 1 , f0 , z0 )
has a branch point at p, then p has to be rooted over one of the j . Furthermore, # 1 (z)
is constant on CP 1 \ {1 , . . . , J } and no larger than the degree of P (w, z) in w.
(e) All branch points of the unramified Riemann surface
RS(CP 1 , CP 1 , f0 , z0 )
generated by an algebraic germ are necessarily algebraic. In particular, the ramified

Riemann surface of an algebraic germ is compact.
Proof. (a) Since [f, z] is obtained from [f0 , z0 ] via analytic continuation, we conclude
g To pass to an
from the uniqueness theorem that P (f (z), z) = 0 for all [f, z] RS.
irreducible factor, we will need to use the resultant, see Chapter 12: given two relatively
prime polynomials P, Q C[w, z], there exist A, B C[w, z] such that
(5.5)
A(w, z)P (w, z) + B(w, z)Q(w, z) = R(z) C[z]
is a nonzero polynomial in z alone; it is called the resultant of P and Q. To conclude the

proof of (a), let P0 (w, z) be an irreducible factor of P for which P0 (f0 (z), z) = 0 near z0 .
In view of (5.5) this factor is unique up to constant factors, as claimed.
(b) First, there exist > 0 and > 0 such that
P (w, z) 6= 0 |z z1 | < , |w w1 | =
Next, from the residue theorem,

I
1
Pw (w, z)
dw = 1 |z z1 | <
2i |ww1|= P (w, z)
and one infers from this that P (w, z) = 0 has a unique zero w = f1 (z) D(w1 , ) for all
|z z1 | < . Finally, write
I
1
Pw (w, z)
f1 (z) =
w
dw
2i |ww1|= P (w, z)
again from the residue theorem, which allows one to conclude that f1 (z) is analytic in
D(z1 , ). The reader familiar with the Weierstrass preparation theorem will recognize
this argument from the proof of that result, see Problem 5.2.
(c) This follows from the (a) and (b) by considering the discriminant of P , which is
defined to be the resultant of P (w, z) and Pw (w, z).
(d) This is a consequence of (b) and (c). The constancy of the cardinality of 1 (z)
follows from the fact that this cardinality equals the degree of
e. It is clear that there
can be no more sheets than the degree of P in w specifies (there could be fewer sheets,
though, but as we shall see, only if P is reducible).
(e) Let p be a branch point of RS(CP 1 , CP 1 , f0 , z0 ). If p is rooted over , then
consider instead f0 ( +z 1 ) where is not the root of any branch point. So let us assume
that p is rooted over some point z1 C. We need to show that relative to the uniformizing
92
chart from Lemma 5.10 the analytic function F has a removable singularity or a pole
at z1 . For this it suffices to show that any solution f (z) of P (f (z), z) = 0 can grow at
most like |z z1 |k as z z1 for some k 0 (indeed, the role of is merely to remove
multi-valued issues but does not affect the polynomial growth as z z1 other than
by changing the power k). This follows easily from the fact that
n
X
aj (z)(f (z))j = 0
j=0
Indeed, suppose an (z1 ) 6= 0. Then let |an (z)| > > 0 for all |z z1 | < r0 small. Fix such
a z. Since
n1
X
aj (z)(f (z))j(n1)
f (z) = an (z)1
1
Pn1
j=0
either |f (z)| 1 or |f (z)|

j=0 |aj (z)|. If, on the other hand, an (z1 ) = 0, then
an (z) = (z z1 ) bn (z) where bn (z1 ) 6= 0. Thus |bn (z)| > for all |z z1 | < r0 . From this
we conclude that
|f (z)(z z1 ) | 1
n1
X
j=0
|aj (z)||z z1 |(n1j) |f (z)(z z1 ) |j(n1)
As before, this implies that f (z)(z z1 ) remains bounded and we have shown that all
branch points are indeed algebraic. Finally, since CP 1 is compact and due to Lemma 5.10
g can be covered by finitely many compact
it follows that that the ramified surface RS
g is itself compact and we are done.
sets. This of course implies that RS
Lemma 5.15 is not just of theoretical value, but also has important practical implications. Let us sketch how to build the ramified Riemann surface of some irreducible
polynomial equation P (w, z) = 0 which we solve for w. Suppose the degree of P in w is
n 1. First, let D(z) = R(P, Pw ) be the discriminant of P (thus, the resultant of P and
Pw ). First, we remove all critical points C C which are defined to be all zeros of D
and from C . It is clear that for all z0 C \ C there are n analytic functions wj (z)
defined near z0 so that P (wj (z), z) = 0 for 1 j n and such that all wj (z) are distinct.
Locally at each finite critical point the following happens: there is a neighborhood of ,
say U , so that on U at least one of the zeros wj ceases to be analytic due to the fact
that P (, ) = 0 has at least one multiple root. But in view of Lemma 5.10 we see that
locally around there is always a representation of the form (called Puiseux series)
wj (z) =
X
k=0
ajk (z ) j
with some j 1. If j = 1 then wj is of course analytic, whereas in all other cases it

is analytic as a function of which is defined via the uniformizing change of variables
z = + j . This is precisely the meaning of Lemma 5.10. It may seem clear from this
example that the number of sheets needs to be precisely n. However, this requires some
work as the reader will see below. For related examples see the problems of this chapter.
For an explicit example of a Puiseux series, consider the irreducible polynomial
P (w, z) = w4 2w2 + 1 z
93
and solve for w locally around (w, z) = (1, 0). This leads to
q
1
X
n
2 z2
w(z) = 1 + z =
n
n=0
which converges in |z| < 1.
Next, we turn to the following remarkable result which is in some sense a converse of
what we have done so far. More precisely, we will show how to generate an algebraic
equation from a nonconstant meromorphic function on an (abstract) compact Riemann
surface!
Proposition 5.16. Let z = z(p) be a meromorphic function of degree n 1 on a
compact Riemann surface M . If f : M CP 1 is any other nonconstant meromorphic
function, then f satisfies an algebraic equation
(5.6)
f n + 1 (z)f n1 + 2 (z)f n2 + . . . + n1 (z)f + n (z) = 0
of degree n, with rational functions j (z). In particular, if the ramified Riemann surface
1
g
RS(CP
, CP 1 , f0 , z0 )
is compact, where [f0 , z0 ] is any holomorphic germ, then [f0 , z0 ] is algebraic.

Proof. To prove the theorem, we proceed as follows: remove from C the point
, as well as the image z(p) of any branch point of the map p 7 z(p) (recall that a
branch point of an analytic function is defined as having valency strictly bigger than
one). Denote these finitely many points as C := {zj }Jj=0 and we refer to them as critical
points. If z C \ C, then let
{p1 (z), . . . , pn (z)}
be the n pre-images under z(p) in arbitrary order and define
X
j (z) :=
f (p1 (z)) . . . f (pj (z))
11 <2 <...<j n
with 0 = 1. Thus, the j are the elementary symmetric functions in f (p1 ), . . . , f (pn )
and they satisfy
n
n
Y
X
j
(w f (p (z)))
w nj (z) =
j=0
=1
By Lemma 5.5, each pj (z) is a holomorphic function on any simply connected subdomain
of C \ C (possibly after a renumbering of the branches). This implies that j is meromorphic on any such domain. Furthermore, analytic continuation of j along a small loop
surrounding an arbitrary point of C takes j back to itself as the different branches of pj
can only be permuted; however, j is invariant under such a permutation. This implies
that j has isolated singularities at the points of C and since f is meromorphic these singularities can be at worst poles. In conclusion, j is meromorphic on C and therefore
rational as claimed. Note that (5.6) holds by construction. For the final statement, let
1 , CP 1 , f , z ) CP 1 be the meromorphic function z(p) from the first part.
g
e : RS(CP
0 0
Thus, (f, p) is algebraic as desired.
94
Needless to say, Proposition 5.16 is a fundamental statement in the theory of algebraic

functions. It raises a number of important questions: Which Riemann surfaces carry
nontrivial meromorphic functions? Furthermore, what degrees can one achieve on a
given surface? These questions can be restated as follows: how can we realize a given
compact Riemann surface M as a branched cover of S 2 , and which restrictions exist for
the number of sheets of such a cover? The Riemann-Hurwitz formula provides one such
obstruction, albeit a rather weak one. A much more useful tool will later be given by
the Riemann-Roch theorem. Let us first observe the trivial fact that for M to admit a
meromorphic function of degree one, it is necessary and sufficient that M S 2 in the
sense of conformal isomorphism. We shall see later (in Chapter 8) that in fact every
compact simply connected Riemann surface is isomorphic to S 2 .
Before pursuing these matters any further, let us first derive a simple but important
corollary from Proposition 5.16. It will allow us to determine the number of sheets of
the Riemann surface determined by an irreducible polynomial.
Corollary 5.17. Let P C[w, z] be irreducible and suppose degw (P ) = n. Then
e generated by some germ [f0 , z0 ] with P (f0 (z), z) = 0
the ramified Riemann surface R
locally at z0 has n sheets and satisfies
e = R(C
e , C , f1 , z1 )
R
for every germ [f1 , z1 ] with P (f1 (z), z) = 0 for all z near z1 .
Proof. If the number of sheets were m < n, then by the previous result we could
find a polynomial Q of degree m with the property that Q(f0 (z), z) = 0 for z near z0 .
But then Q would necessarily need to be a factor of P which is impossible.

In other words, the ramified Riemann surface of a germ satisfying an irreducible
polynomial equation contains all germs satisfying this equation.
Let us now clarify the connection between the smooth, affine or projective, algebraic
curves which were introduced in the previous chapter, and the ramified Riemann surfaces
of an algebraic germ which we just constructed. In particular, the reader needs to recall
the notion of nonsingular polynomials as well as the homogenization of a polynomial, see
(4.2) and (4.3).
Lemma 5.18. Let P C[w, z] be an irreducible polynomial so that its homogenization
Q is nonsingular. Then the smooth projective algebraic curve Se defined by Q, see (4.3),
1 , CP 1 , f , z ) of any algebraic germ
e := RS(CP
g
is isomorphic to the Riemann surface R
0 0
g
[f0 , z0 ] defined by P . In particular, RS can be imbedded into CP 2 .
Proof. We will need to following fact, see Problem 4.8: Let M, N be compact Riemann surfaces and suppose : M \ C N \ C is an isomorphism where C, C are finite
e : M N.
sets. Then extends to an isomorphism
e
Remove from R all points at infinity, more precisely, all germs rooted at z = .
In other words, we consider
g
M := RS(C,
C , f0 , z0 )
for an arbitrary germ [f0 , z0 ] of P with z0 C. Similarly, remove from Se the line at
infinity, i.e.,
N := Se \ L ,
L := {[z : w : 0] | (z, w) C2 }
95
e The reader
In other words, consider the affine curve S instead of the projective one S.
e
should convince himself or herself that SL is always finite. By the aforementioned fact
concerning isomorphisms, it will suffice to show that M N . We will accomplish this by
showing that we can identify M and N as sets and also use the same charts. First, locally
at all non-critical points (z1 , w1 ) (which are defined via P (w1 , z1 ) = 0, w P (w1 , z1 ) 6= 0)
one has n = degw (P ) branches wj (z) defined and analytic near z1 . These branches define
n charts on both M and N on neighborhoods which we can identify provided they are
chosen small enough (so that there is no overlapping of values); indeed, we simply map
the germ [f, z] on M onto the pair (f (z), z) N .
Next, consider any critical point (z1 , w1 ). Since P is non-singular, one has z P (z1 , w1 ) 6=
0 which implies that there exists an analytic function Z1 (w) defined near w = w1 and
an analytic and nonvanishing function Q near (z1 , w1 ) so that
P (z, w) = (z Z1 (w))Q(z, w)
locally around (z1 , w1 ). This follows either from Lemma 5.15 or the Weierstrass prepaj
ration theorem, see Problem 5.2. Now suppose that w
P (w1 , z1 ) = 0 for all 0 j <
but w P (w1 , z1 ) 6= 0 with some 2. This is equivalent to
d Z
dj Z
(w
)
=
0
j
<
,
(w1 ) 6= 0
1
dwj
dw
Thus, the branching number of the branch point of M at (z1 , w1 ) is 1 1 and a
uniformizing chart as in Lemma 5.10 is given by z = Z1 (w) provided w = f (z) with [f, z]
a germ from the Riemann surface representing the branch point (z 6= z1 ). On the other
hand, by definition z = Z1 (w) is a chart on N near w1 . Hence, this again allows one to
identify M and N near (z1 , w1 ) with the same chart and we are done.

Proposition 5.16 raises the following natural and important question: Is M conformally equivalent to the Riemann surface associated with the algebraic equation (5.6)
over CP 1 ?. For this to hold (5.6) needs to be irreducible . We shall see later, as an
application of the Riemann-Roch theorem, that we can in fact find f to a given z such
that (5.6) is irreducible. This will then allow us to obtain an affirmative answer to
our question, which has a remarkable consequence: any compact Riemann surface which
carries a nonconstant meromorphic function is isomorphic to the Riemann surface associated to some irreducible polynomial P (w, z). Since we will later see that in fact
every Riemann surface carries a nonconstant meromorphic function, we can now state
the following truly remarkable conclusion.
Theorem 5.19. Every compact Riemann surface M is the ramified Riemann surface
of some algebraic germ.
Proof. As already explained, the proof hinges on two facts:

every Riemann surface carries a nonconstant meromorphic function z : M
CP 1
given a nonconstant meromorphic function on M , we can find another meromorphic function f on M such that (5.6) is irreducible
We will prove these two facts in Chapters 7, see Corollary 7.14, and 8, see Problem 8.5,
respectively. Assuming them for the moment, we can easily finish the argument by going
back to proof of Proposition 5.16. Indeed, let P (w, z) be the irreducible polynomial
96
g be the ramified Riemann surface of P (w, z). In view

associated with (5.6) and let RS
g has exactly n = deg(z) sheets. For any z0 CP 1 \ C there are n
of Corollary 5.17, RS
distinct pre-images {pj (z0 )}nj=1 with the property that locally around these points p 7 z
is an isomorphism onto some neighborhood of z0 . Moreover, these neighborhoods are
g To see this, simply note that each f (pj (z))
also isomorphic to neighborhoods of RS.
g But
is analytic close to z0 and therefore an unbranched function element from RS.
n
g
since RS has n sheets this implies that the germs {[f pj , z0 ]}j=1 are distinct and in
fact parametrize neighborhoods on distinct sheets. This in turn defines isomorphisms
between each of these neighborhoods and respective neighborhoods of the points pj (z0 )
on M : simply map [f pj , z] onto pj (z) for all z close to z0 . Now connect the points of C
by a piecewise linear path so that := C \ is simply connected. By Lemma 5.5
g that cover
these local isomorphisms extend to global ones between those sheets of RS
g
via the canonical projection
e : RS C as well as the respective components of M
g is
which cover via the map z. It remains to check that over each critical point RS
branched in exactly the same way as M . More precisely, one needs to verify that at each
critical point z0 C there is a one-to-one correspondence between the branch points
of M and RS, respectively, with equal branching numbers (recall that we may have
unbranched sheets over a given root, see Figure 5.5). However, this is clear from the
fact that the germs are given by f pj and thus inherit the branching numbers from the
map z: suppose analytic continuation of the germ [f pj , z1 ] with a fixed z1 = z0 + ei0
close to z0 along the loop := z0 + ei , 0 0 + 2 yields the germ [f pk , z1 ]
with j 6= k. Since we verified before that these germs are distinct, it follows that the lift
of to M under the branched cover z which starts at pj must end at pk . But this means
precisely that there is the desired one-to-one correspondence between the branch points
which preserves the branching number.

In other words, any compact Riemann surface is obtained by analytic continuation of
a suitable algebraic germ! Note that the proof gives an explicit conformal isomorphism:
p 7 (z(p), f (p))
where z and f are the meromorphic functions from the proof.

As the surface of the logarithm shows, this is not true in the non-compact case.
Another example is given by
1
RS(C , C , exp(z 2 ), 1)
It is clearly not algebraic and, in fact, is not compact since the branch point at z =
is not part of the ramified Riemann surface. The existence of a meromorphic function
on an (abstract) Riemann surface is a highly non-trivial issue; indeed, even though we
can of course define such functions locally on every chart, the challenge lies with the
extension of such a function beyond the chart as partitions of unity, say, would take us
outside of the analytic category (by the uniqueness theorem).
As already note before, the case n = 1 of Proposition 5.16 means precisely that
z : M CP 1 is an isomorphism and we recover the result that all meromorphic functions
on CP 1 are rational. The first interesting case is n = 2, and any compact M of genus
g > 1 carrying such a meromorphic function z(p) is called hyper-elliptic, whereas the
genus one case is typically referred to as elliptic (it will follow from the Riemann-Roch
theorem that every compact surface of genus one carries a meromorphic function of
97
degree two). Suppose now that n = 2 in Proposition 5.16. To proceed we quote a result
that we shall prove in Chapter 7, see Corollary 7.14: Let p, q M be two distinct points
on a Riemann surface M . Then there exists f M(M ) with f (p) = 0 and f (q) = 1.
In other words, the function field on M separates points. Apply this fact with
{p, q} = z 1 (z0 ) where z0 is not a critical value of z. By the preceding, f satisfies an
equation of the form
(5.7)
f 2 + 1 (z)f + 2 (z) = 0
with rational 1 , 2 . This equation needs to be irreducible. Indeed, otherwise f is a

rational function of z which would contradict that f takes different values at p and q.
Replacing f by f + 1 /2 shows that the Riemann surface generated by (5.7) is the same
as that generated by
f 2 + 3 (z) = 0,
3 := 2 + 12 /4
By irreducibility, 3 is not the square of a rational function. Let 3 = PQ where P, Q
are without loss of generality, relatively prime, monic polynomials. By the proof
q of
Theorem 5.19, we see that M is isomorphic to the Riemann surface generated by PQ .
A
moments reflection shows that this surface is isomorphic to the Riemann surface of
P Q, and moreover, we can assume that each linear factor in P Q appears only to the
first power. In other words, we infer that every elliptic or hyper-elliptic Riemann surface
is isomorphic to one of the examples which we already encountered above, viz.
p
1
g
RS(CP
, CP 1 , (z z1 ) . . . (z zm ), z0 )
where {zj }m
j=0 are all distinct. Let us now examine the elliptic surfaces more closely.
Corollary 5.20. Suppose M is a compact surface of genus one. Then M is isomorphic to the set of zeros in CP 2 of a cubic polynomial
(5.8)
E := {[z : w : u] CP 2 | Q (z, w, u) := w2 u z(z u)(z u) = 0}
where C \ {0, 1}. In other words, M can be imbedded into CP 2 . For the particular
case of tori C/ with := hz 7 z + 1 , z 7 z + 2 i and 1 , 2 independent over R, this
imbedding is given explicitly in terms of the Weierstrass function associated with the
lattice :

[(z) : (z) : 1] z 6= 0
(5.9)
: z C/ 7
[0 : 1 : 0]
z=0
Proof. We need to refer to the aforementioned fact (see Chapter 8, Corollary 8.14)
that every genus one compact surface can be realized as a two sheeted branched cover
of S 2 (in other words, it carries a meromorphic function of degree two). Note that for
the case of tori this is nothing but the existence of the Weierstrass function.
By our
qQ
m
previous discussion, M is therefore isomorphic to the Riemann surface of
j=1 (z zj )
with either m = 3 or m = 4 and distinct zj . However, these two cases are isomorphic so
it suffices to consider m = 3. Composing with a Mobius transform which moves z0 , z1 to
0, 1, respectively, and fixed , we now arrive at the the cubic polynomial
P := w2 z(z 1)(z ) = 0,
C \ {0, 1}
Not only is P non-singular but also its homogenization Q from (5.8). We leave this to
the reader to check. Hence, applying Lemma 5.18, we see that M is isomorphic to both
98
the Riemann surface of P from this chapter as well as the algebraic curve E as defined
in the previous chapter.
Finally, for the torus we proved in Chapter 4 that the Weierstrass function satisfies
the differential equation
( (z))2 = 4((z) e1 )((z) e2 )((z) e3 )
In other words, z 7 ((z), (z)) maps the torus into the set of zeros (where E stands
for elliptic curve)
E := {(, ) C2 : 2 4( e1 )( e2 )( e3 ) = 0}
We need to projectivize this in the usual way leading to (5.9). The choice of [0 : 1 : 0]
for z = 0 is the only possible one as can be seen from the Laurent expansion of and
at z = 0. The reader will easily verify that this map is a homeomorphism between the
torus and the projective version of E. Since the latter is in a canonical way the Riemann
surface of the irreducible polynomial 2 4( e1 )( e2 )( e3 ), we have obtained the
desired isomorphism. Historically, the inverse to the map : z 7 (, ) := ((z), (z))
was given by the elliptic integral
Z p
Z p
d
d
p
=
(5.10)
z(p) =
4 3 g2 g3

where p E where the latter is viewed as the Riemann surface of
2 4( e1 )( e2 )( e3 ) =: 2 (4 3 g2 g3 )
The integration here is along any path that avoids the branch points and the branch of
the square root in (5.10) is determined by analytic continuation along that path. Clearly,
the integral in (5.10) is invariant under homotopies, but the path is determined only up
to integral linear combinations of the homology basis a, b, see Figure 5.9. The dashed line
on b means that we are entering the other sheet. The torus is shown as E is topologically
equivalent to one. However, such an integral linear combination changes z(p) only by
b
b
a
Figure 5.9. The homology basis for E
(5.11)
d
+n
d
= m1 + n2
99
with n, m Z. We leave the evaluation of the integrals to the reader, see Problem 4.7.
Consequently, z(p) is well-defined as an element of the torus C/. Finally, since :
z 7 ((z), (z)) is clearly onto, we can write a path connecting to p as where
: [0, 1] C/ is a path connecting 0 to z0 . But then
Z 1
Z 1
d( )
(t) dt = (1) = z0
=
z(p) =
()
0
0
which shows that p 7 z(p) is indeed the inverse to .
Elliptic curves have many remarkable properties, of which we just mention the following one: the three distinct points {(zj )}3j=1 are colinear iff z1 + z2 + z3 = 0. In other
words,
(z1 ) (z1 ) 1
det (z2 ) (z2 ) 1 = 0 z1 + z2 + z3 = 0
(z3 ) (z3 ) 1
where the final equality is to be understood modulo . For the proof, as well as how to
use this fact to put a group structure on an elliptic curve (and many other properties of
these curves) we refer the reader to [11] and Chapter 3 of [22].
This result of course also raises a number of questions, for example: does any compact
Riemann surface of higher genus admit an imbedding into some projective space CP d ?
In fact, the answer is yes with d = 3. For more on these topics see Chapter 8 or [23].
To conclude this chapter, we illustrate the methods ofp
algebraic functions by means
of the (ramified) Riemann surfacedefined by the germ [ 1 + z, 1]. The convention

regarding the square root shall be x > 0 for x > 0. This germ gives rise to the unique
irreducible monomial P (w, z) = (w2 1)2 z. The system P = 0, w P = 0 has solutions
(w, z) {(0, 1), (1, 0), (1, 0)}
2 P (w, z) 6= 0 as well as P (z, w) 6= 0 at each of these points. Thus,
Note that w
z
Figure 5.10. The sheets of
1+
any branch point associated with them has branching number 1. Moreover,
if z = 0
then necessarily w = 1, whereas z = 1 yields w = 0 as well as w = 2. Finally,
z = is a branch point and analytic continuation around a large circle (more precisely,
any loop encircling z = 0, 1 once) permutes the four sheets cyclically. For this reason,
the sheets look schematically as shown in Figure 5.9 (the two sheets which branch over
z = 1 cannot branch again over z = 0 since this would contradict the aforementioned
100
cyclic permutation property at z = ). Let us now analyze the four sheets and their
permutation properties more carefully. First, on the simply connected region
:= C \ (, 0]
there exist four branches fj (z), 0 j 3 uniquely determined by the asymptotic

equalities
1
1
1
1
f0 (x) x 4 , f1 (x) ix 4 , f2 (x) x 4 , f3 (x) ix 4
as x . The enumeration here has been chosen so that analytic continuation along
any circle containing z = 0, 1 with positive orientation induces the cyclic permutation
f0 7 f1 7 f2 7 f3 7 f0
Next, for all x > 1 there are the explicit expressions

q
q
f0 (x) = 1 + x, f2 (x) = 1 + x
q
q
x 1, f3 (x) = i
x1
f1 (x) = i
Analytic continuation to yields
f1 (i0+) = f3 (i0) = f2 (0) = 1, f1 (i0) = f3 (i0+) = f0 (0) = 1
Recall that by our convention, z is analytic on with x > 0 if x > 0. Analytic

continuation around the loop z() = 1+ 2 ei , 0 2, with > 0 small leaves f0 and
f2 invariant, whereas f1 and f3 are interchanged. Similarly, analytic continuation around
z() = 2 ei (starting with > 0 small) takes f0 into f3 and f2 into f1 . This implies
that the monodromy group of this Riemann surface is generated by the permutations
(12)(03) and (13). The formal definition of the monodromy group is as follows: for any
c 1 (C \ {0, 1}) let (c) S4 (the group of permutations on four symbols) be defined
as the permutation of the four sheets which is induced by analytic continuation along
the closed loop c. For this we assume, as we may, that c has its base point somewhere
in and we pick the four germs defined on each branch over that base point for analytic
continuation. Note that is well-defined by the monodromy theorem. The map
: 1 (C \ {0, 1}) S4
is a group homomorphism and the monodromy group is the image (1 ). For example,
(13)(12)(03) = (0123) which is precisely the cyclic permutation at z = . The genus g
of this Riemann surface is given by the Riemann-Hurwitz formula as
1
g 1 = 4(1) + (1 + 1 + 1 + 3) = 4 + 3 = 1
2
which implies that g = 0.
5. Problems
Problem 5.1. Picture the unramified Riemann surfaces
(5.12)
RS(C, C, log z, 1),
RS(C, C, z n , 1),
n 2. Prove that they cover C . Compute the fundamental groups 1 (RS) of these
surfaces and prove that
RS(C, C, log z, 1) C
1
RS(C, C, z n , 1) C ,
n2
5. PROBLEMS
101
in the sense of conformal isomorphisms. Show that each of the surfaces in (5.12) has a
branch point rooted at zero.
Problem 5.2. Prove the Weierstrass preparation theorem: suppose f (z, w) is analytic in both variables1 and such that f (z, w) 6= 0 for all |ww0 | = r0 > 0 and |zz0 | < r1 .
Then there exists a polynomial
n
X
aj (z)wj
P (z, w) =
j=0
with aj analytic on |z z0 | < r1 so that
f (z, w) = P (z, w)g(z, w)
with g analytic and nonvanishing on D(z0 , r1 ) D(w0 , r0 ). Moreover, all solutions of

P (z, ) = 0 lie inside of D(w0 , r0 ) for all z D(z0 , r1 ). In particular, show the following:
f (z , w ) = 0 for all 0 < n, but n f (z , w ) 6= 0, then there exist r , r > 0 so
if w
0
0
0 1
0
0
w
that the previous statement applies.
Problem 5.3. Let A(z) be an n n matrix so that each entry Aij (z) is a polynomial
in z. Let the eigenvalues be denoted by j (z), 1 j n. Prove that around each point z0
at which j (z0 ) is a simple eigenvalue, j (z) is an analytic function of z. Furthermore,
if z1 is a point at which j (z1 ) has multiplicity k, then there is a local representation of
the form
X
n
an (z z1 )
j (z) =
n=0
with some 1 k (this is called a Puiseaux series). Now assume that A(z) is
Hermitian for all z R. Prove that each j is analytic on a neighborhood of R. In
other words, if z1 R then the Puiseux series is actually a power series. Check these
statements by means of the examples

0 z
0 z
A(z) =
,
A(z) =
1 0
z 0
Problem 5.4. For each of the following algebraic functions, you are asked to understand their Riemann surfaces by answering each of the following questions: Where are
the branch points on the surface (be sure to check infinity)? How many sheets does it
have? How are these sheets permuted under analytic continuation along closed curves
which avoid the (roots of the) branch points? What is its genus? You should also try to
obtain a sketch or at least some geometric intuition of the Riemann surface.
q
q
3
4
z 1 , w = 2 z + z + 1 , w3 3w z = 0
w=
p
p
3
w = (z z1 ) . . . (z zm ) , w = z 2 1
1This means that f is jointly continuous and analytic in each variable.
CHAPTER 6
Differential forms on Riemann surfaces

1. Holomorphic and meromorphic differentials
We already observed that every Riemann surface is orientable as a smooth twodimensional manifold. Next, we state another important fact. Throughout, M and N
will denote Riemann surfaces.
Lemma 6.1. Every tangent space Tp M is in a natural way a complex vector space.
In particular, M is orientable and thus carries a volume form. Moreover, if f : M N
is a C 1 map between Riemann surfaces, then f is analytic iff Df (p) is complex linear as
a map Tp M Tf (p) N for each p M .
Proof. First note that (~v , w)
~ is well-defined in Tp M . Simply measure this angle
in any chart because of conformality of the transition maps this does not depend on
the choice of chart. The sign of the angle is also well-defined because of the orientation
on M . Now let R be a rotation in Tp M by 2 in the positive sense. Then we define
i~v := R~v
It is clear that this turns each Tp M into a complex one-dimensional vector space. To
fix an orientation on M , simply define (v, iv) with v Tp M , v 6= 0, as the positive
orientation. Since f : U R2 with f C 1 (U ), U C open is holomorphic iff Df is
complex linear, we see via charts that the same property lifts to the Riemann surface
case.

As a smooth manifold, M carries k-forms for each 0 k 2. We allow these forms
to be complex valued and denote the respective spaces by
0 (M ; C),
1 (M ; C),
2 (M ; C)
By definition, 0 (M ; C) are simply C functions on M , whereas because of orientability

2 (M ; C) contains a 2-form denoted by vol which never vanishes; hence, every other
element in 2 (M ; C) is of the form f vol where f 0 (M ; C). This leaves 1 (M ; C) as
only really interesting object here. By definition, each 1 (M ; C) defines a real-linear
functional p on Tp M . We will be particularly interested in those that are complex
linear. We start with a simple observation from linear algebra.
Lemma 6.2. If T : V W is a R-linear map between complex vector spaces, then
there is a unique representation T = T1 + T2 where T1 is complex linear and T2 complex
2 (~v ).
anti-linear. The latter property means that T2 (~v ) = T
Proof. Uniqueness follows since a C-linear map which is simultaneously C-anti linear vanishes identically. For existence, set
1
1
T2 = (T + iT i)
T1 = (T iT i),
2
2
103
104
6. DIFFERENTIAL FORMS ON RIEMANN SURFACES
Then T1 i = iT1 and T2 i = iT2 , T = T1 + T2 , as desired.
As an application, consider the following four complex valued maps on U R2 where

U is any open set: 1 , 2 , z, z which are defined as follows
1 (x, y) = x, 2 (x, y) = y, z(x, y) = x + iy, z = x iy
Identifying the tangent space of U with R2 at every point the differentials of each of
these maps correspond to the following constant matrices

1 0
0 1
1 0
1 0
d1 =
, d2 =
, dz =
, d
z=
0 0
0 0
0 1
0 1
Let us now write 1 (M ; C) in local coordinates
1
1
z = u dz + v d
z
= a dx + b dy = (a ib) dz + (a + ib) d
2
2
Of course, this is exactly the decomposition of Lemma 6.2 in each tangent space. A very
important special case if = df with f 0 (M ; C) (we already encountered this in
Chapter 1). Then
df = z f dz + zf d
z
1
1
z f = (x f iy f ), zf = (x f + iy f )
2
2
If f H(M ), then df needs to be complex linear. In other words, zf = 0 which are
precisely the CauchyRiemann equations. In this notation, it is easy to give a one line
proof of Cauchys theorem: Let f H(U ) where U C is a domain with piecewise C 1
). Then
boundary and f C 1 (U
Z
Z
Z
zf d
z dz = 0
d(f dz) =
f dz =
U
We leave it to the reader the verify the chain rules

z (g f ) = (w g) f z f + (w g) f z f
z(g f ) = (w g) f zf + (w g) f zf
2
as well as the representation of the Laplacean = 4 z

z .
Definition 6.3. The holomorphic differentials on a Riemann surface M , denoted by

H1 (M ), are precisely those 1 (M ; C) so that = u dz in arbitrary local coordinates
with u holomorphic. The meromorphic differentials, denoted by M1 (M ), are all
H1 (M ; C \ S) where S M is discrete and so that in local coordinates around an
arbitrary point of M one has = u dz where u is meromorphic. The points of S will be
called poles of .
Note that we are assuming here that those points of S which are removable singularities of u have been removed. Obvious examples of holomorphic and meromorphic
differentials, respectively, are given by df where f H(M ) or f M(M ).
2. INTEGRATING DIFFERENTIALS AND RESIDUES
105
2. Integrating differentials and residues

Let us introduce the following terminology, with M, N Riemann surfaces as usual:
is compact and N is
Definition 6.4. We say that N M is a Stokes region if N
1
piecewise C . This means that N is the finite union of curves j : [0, 1] M which
are C 1 on the closed interval [0, 1].
The importance of a Stokes region lies with the fact that Stokes theorem applies to
it. We have the following simple but important properties:
R
Proposition 6.5. Suppose H1 (M ). Then d = 0. Thus, N
H = 0 for every
Stokes region N M . Moreover, for any closed curve the integral only depends
H
H
on the homology class of . In particular, = if and are homotopic closed
curves. Finally, if c is a curve with initial point p and endpoint q, then
Z
= f1 (q) f0 (p)
c
where df0 = locally around p and f1 is obtained via analytic continuation of f0 along c.
In particular, df1 = locally around q.
Proof. Since = u dz in a chart, one has
d = zu d
z dz = 0
as claimed. The other properties are immediate consequences of this via Poincares
c1
c3
c2
Figure 6.1. c1 and c2 are homologous to each other, but not to c3

lemma (closed means locally exact) and Stokes integral theorem. We skip the details.
For meromorphic differentials, we have the following facts.

Proposition 6.6. Suppose M1 (M ) with poles {pj }Jj=1 . Then at each of these
poles their order ord(, pj ) Z and residue res(, pj ) C are well-defined. In fact,
I
1
res(, pj ) =
2i c
106
where c is any small loop around pj . Given any Stokes region N M so that N does
not contain any pole of , we have
I
X
1
=
res(, p)
(6.1)
2i N
pN
Finally, if M is compact, then
res(; p) = 0
pM
for all M1 (M ). With N M a Stokes region and f M(M ),

I
1
df
= #{p N | f (p) = 0} #{p N | p is a pole of f }
2i N f
assuming that no zero or pole lies on N .

P
n
Proof. Let = u dz with u(p) =
n=n0 an z in local coordinates (U, z) around
pj with z(pj ) = 0 and an0 6= 0. Note that n0 does not depend on the choice of the chart
but the coefficients do in general. However, since
I
1
a1 =
2i
this coefficient does not depend on the chart and it is the residue. The residue theorem
(6.1) follows from Stokes applies to N := N \ Dj where Dj are small parametric disks
centered at pj N . Finally, if M is compact, then we triangulate M in such a way that
no edge of the triangulation passes through a pole (there are only finitely many of them).

This follows simply by setting = df
f in (6.1).
In the following chapter we shall prove that to a given finite sequence {pj } of points
and complex numbers {cj } adding up to zero at these points we can find a meromorphic
differential that has simple poles at exactly these points with residues equal to the cj .
This will be based on the crucial Hodge theorem to which we now turn.
3. The Hodge operator and harmonic differentials
Definition 6.7. To every 1 (M ; C) we associate a oneform defined as

follows: if = u dz + v d
z = f dx + g dy in local coordinates, then
:= iu dz + iv d
z = g dx + f dy
Moreover, if ,
1comp (M ; C)
(6.2)
(the forms with compact support), we set

Z

h, i :=
M
This defines an inner product on

by 12 (M ; C).
1comp (M ; C).
The completion of this space is denoted
Some comments are in order: first, is well-defined as can be seen from the change
of coordinates z = z(w). Then
= u dz + v d
z = uz dw + v z d
z
3. THE HODGE OPERATOR AND HARMONIC DIFFERENTIALS
107
and transforms the same way. Second, it is evident that (6.2) does not depend on
coordinates and, moreover, if , are supported in U where (U, z) is a chart, then with
= u dz + v d
z , = r dz + s d
z we obtain
and in particular,
r + v
s) dz d
z = 2(u
r + v
s) dx dy
= i(u
h, i = 2
ZZ
(u
r + v
s) dx dy
which is obviously a positive definite scalar product locally on U . By a partition of unity,

this shows that indeed (6.2) is a scalar product on 1comp (M ) and, moreover, that the
abstract completion 12 (M ) consists of all 1-forms which in charts have measurable
L2loc coefficients and which satisfy the global property kk22 := h, i < .
Let us state some easy properties of the Hodge- operator.
Lemma 6.8. For any , 12 (M ) we have
= ,
= ,
h , i = h, i
Proof. The first two identities follow immediately from the representation in local
coordinates, whereas the third is a consequence of the first two.

We now come to the very important topic of harmonic functions and forms. Recall
that the class of harmonic functions is invariant under conformal changes of coordinates,
see Corollary 1.29.
Definition 6.9. We say that f 0 (M ; C) is harmonic iff f is harmonic in every
chart. We say that 1 (M ; C) is harmonic iff d = d = 0, i.e., iff is both closed
and co-closed. We denote the harmonic forms on M by h(M ; R) if they are real-valued
and by h(M ; C) if they are complex-valued.
Let us state some basic properties of harmonic functions, mainly the important maximum principle.
Lemma 6.10. Suppose f 0 (M ; C) is harmonic with respect to some atlas. Then
it is harmonic with respect the any equivalent atlas and therefore, also with respect to the
conformal structure. Moreover the maximum principle holds: if such an f is real-valued
and the open connected set U M has compact closure in M , then
min f f (p) max f
U
pU
with equality being attained at some p U iff f = const. In particular, if M is compact,

then f is constant.
Proof. Under the conformal change of coordinates w = w(z) we have
2f
2f
= |w (z)|2
z z
w w
Thus, harmonicity is preserved under conformal changes of coordinates as claimed. For
the maximum principle, we note that f : U R is harmonic in a chart and attains a
local maximum in that chart, then it is constant on the chart by the maximum principle
for harmonic functions on open sets of C. But then f would have to be constant on all of
U by connectedness and the fact that harmonic functions in the plane that are constant
on some open subset of a planar domain have to be constant on the entire domain. Hence
108
we have shown that f cannot attain a local maximum on U . Finally, if M is compact,

then by taking U = M we are done.

For harmonic forms we have the following simple properties.
Lemma 6.11. Let h(M ; R) (or h(M ; C)). Then, locally around every point of M ,
= df where f is real-valued (or complex-valued) and harmonic. If M is simply connected, then = df where f is a harmonic function on all of M . Conversely, if f is a
harmonic function on M , then df is a harmonic 1form. If u is a harmonic (and either
real- or complex-valued) function, then the complex linear part of du is a holomorphic
differential. In other words, in local coordinates, = z u dz H1 (M ).
Proof. Since is locally exact, we have = df locally with f being either real- or
complex-valued depending on whether is real- or complex-valued. Then is co-closed
iff ddf = 0. In local coordinates, this is the same as
d(fy dx + fx dy) = (fxx + fyy ) dx dy = 0
which is the same as f being harmonic. This also proves the converse. If M is simply
connected, then f is a global primitive of . For the final statement, note that
zz u = 0
since u is harmonic.
Here is a useful characterization of harmonic differentials. We will omit the field R

or C from our notation if this choice makes no difference.
Lemma 6.12. Let 1 (M ) and suppose that = a dx + b dy in some chart (U, z).
Then is harmonic iff f := a ib is holomorphic on z(U ).
Proof. Since
d = (ay + bx ) dx dy,
d = (by + ax ) dx dy
we see that is harmonic iff a, b satisfy the CauchyRiemann system on z(U ) which
is equivalent to a ib being holomorphic on z(U ).

Next, we make the following observation linking holomorphic and harmonic differentials.
Lemma 6.13. Let 1 (M ; C). Then
(1) is harmonic iff = + where , H1 (M )
(2) H1 (M ) iff d = 0 and = i iff = + i where h(M ; R)
In particular, every holomorphic differential is harmonic and the only real-valued holomorphic differential is zero.
Proof. Write = u dz + v d
z in local coordinates. For (1), observe that
d = (zu + z v) dz d
z
d = i(zu + z v) dz d
z
both vanish identically iff zu = 0 and z v = 0. In other words, iff = u dz and = v dz

are both holomorphic differentials.
For (2), is holomorphic iff v = 0 and zu = 0 iff d = 0 and = i. If
= + i with harmonic, then d = 0 and = i = i. For the converse,
). Then = 2i ( +
), + i = , and h(M ; R) as desired.
set = 12 ( +
3. THE HODGE OPERATOR AND HARMONIC DIFFERENTIALS
109
Finally, it is clear from (1) that every holomorphic differential is also harmonic. On
the other hand, if H1 (M ) and real-valued, then we can write
= + i = = i
or = 0 which is the same as = 0.
In the simply connected compact case it turns out that there are no non zero harmonic
or holomorphic differentials.
Corollary 6.14. If M is compact and simply connected, then
h(M ; R) = h(M ; C) = H1 (M ) = {0}
Proof. Any harmonic 1form can be written globally on M as = df with f
harmonic, see Lemma 5.5. But then f = const by the maximum principle and so = 0.
Consequently, the only harmonic 1form is also zero.

The obvious example for this corollary is of course M = CP 1 . Let us now consider
some examples to which Corollary 6.14 does not apply. In the case of M C simply
connected we have in view of Lemmas 6.116.13,
H1 (M ) = {df | f H(M )}
(6.3)
h(M ; C) = {df + dg | f, g H(M )}
h(M ; R) = {a dx + b dy | a = Re (f ), b = Im (f ), f H(M )}
= {df + df | f H(M )}
In these examples harmonic (or holomorphic) 1forms are globally differentials of harmonic (or holomorphic) functions.
For a non-simply connected example, take M = {r1 < |z| < r2 } with 0 r1 < r2
. In these cases, a closed form is exact iff
I
= 0,
r (t) = re2it
r
for one (and thus every) r M (or any closed curve in M that winds around 0). This
implies that every closed can be written uniquely as
I
1
, f C (M )
= k d + df, k =
2 r
and, with being any branch of the polar angle,

x
y
d := 2 dx + 2 dy, r 2 = x2 + y 2
r
r
We remark that d h(M ; R) since (any branch of) the polar angle is harmonic. This
is of course is in agreement with the de Rham chomology fact H 1 (R2 \ {0}) R. The
conclusion is that
h(M ; R) = {df + k d | f harmonic and R-valued on M , k R}
(6.4)
(6.5)
h(M ; C) = {df + k d | f harmonic and C-valued on M , k C}

dz
H1 (M ) = fz dz + ik
| f harmonic, R-valued on M , k R
z

dz
= dg + | g H(M ), C
z
110
The representation (6.4) follows from Lemma 6.13, whereas for (6.5) we note that
H1 (M ) is exact iff
I
=0
r
Hence, (6.5) follows by setting
I
1
=
2i r
Rz
which then allows us to define g(z) := 1 along an arbitrary curve. To reconcile (6.5)
with (6.4), we first observe that
I
I
I
fz d
z=
fz dz =
fz dz iR
r
since f is real-valued. Conversely, if a R, then there exists f real-valued and harmonic

on M such that
I
fz dz = 2ia
r
Indeed, simply set f (z) = a log |z| for which fz (z) dz = a rz2 dz. This explains why it
suffices to add ik dz
z with k R in (6.4).
As a final example, let M = C/h1, i where h1, i Aut(C) is the group generated by
z 7 z + 1, z 7 z + and Im > 0. Then any h(M ; R) lifts to the universal cover
of M which is C. Thus, we can write = a dx + b dy where a ib is an analytic function
on M and thus constant. Hence,
dimR h(M ; R) = dimC h(M ; C) = 2 = 2 dimC H1 (M )
Any reader familiar with Hodges theorem will recognize the statement here that H 1 (M )
R2g where M is a compact surface of genus g.
4. Statement and examples of the Hodge decomposition
In Chapter 7 we shall prove the following version of Hodges theorem:
12 (M ; R) = E E h2 (M ; R)
(6.6)
Here 12 (M ; R) are the square integrable, real-valued, one forms from Definition 6.7,
and
h2 (M ; R) := h 12 (M ; R),

0
(M ; R) ,
E := df | d comp

0
E := df | d comp
(M ; R)
where the closure is meant in the sense of 12 (M ). Figure 6.2 describes the subspaces
appearing in Hodges theorem (the reason for the (co)closed planes will become clear
0
(M ; R)
later, see Lemma 7.2). Let us first clarify that E E: thus, let f, g comp
and compute
Z
Z
Z
d(f dg) = 0
df dg =
df dg =
M
by Stokes. Hence,
12 (M ; R) = E E (E (E) )
and the main issue then becomes equating the intersection at the end with h2 (M ; R).
This is nontrivial, since we will need to prove that all forms in E (E) are smooth
4. STATEMENT AND EXAMPLES OF THE HODGE DECOMPOSITION
111
E
1111111111111111111
0000000000000000000
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
closed
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
*E
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
coclosed
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
000000000000000000000000000000000000000000
111111111111111111111111111111111111111111
0000000000000000000
1111111111111111111
harmonic
Figure 6.2. A schematic view of the Hodge decomposition

while at first sight they only have L2 coefficients. It will turn out that it is easy to see
from the definition that E (E) are weakly harmonic1 so that the issue then is to
prove that weakly harmonic forms are strongly harmonic. This of course is the content
of the basic Weyls lemma which is an example of elliptic regularity theory.
The purpose of this section is to establish the Hodge decomposition (6.6) on four
manifolds: C, S 2 , C/Z2 , and D. In each case we shall determine the space of harmonic
differentials and exhibit the role they play in (6.6). Heuristically speaking, the space h2
is nonzero due to either a nontrivial cohomology or a large boundary of M . The former
is revealed by the torus C/Z2 , whereas the latter arises in the case of D. More technically
speaking, in each of the four examples we will prove the Hodge decomposition by solving
the Poisson equation and the harmonic forms arise either because of an integrability
condition or due to a boundary condition.
Example 1: M = C
Pick any h2 (C). In view of (6.3), = a dx + b dy with a, b being harmonic and
2
L bounded:
ZZ
(|a|2 + |b|2 ) dxdy <
R2
We claim that necessarily a = b = 0. Indeed, from the mean-value theorem

ZZ
2

1

2
a() dd
|a(z)| =
|D(z, r)|
D(z,r)
ZZ
1
kak22
|a()|2 dd
0
|D(z, r)|
|D(z, r)|
D(z,r)
as r . So h(M ) = {0} in that case. Note that while h(C; R) is a huge space (since
there are many entire functions by the Weierstrass theorem), the L2 condition only leaves
the zero form. This can be thought of the fact that the boundary of C is tiny, and in fact
it consists only of the point at infinity in S 2 . Indeed, heuristically the previous argument
1This will be made precise in the next chapter.
112
can be thought of as follows: any form in h2 can be continued to a harmonic form on S 2

which then has to vanish.
Hence, by (6.6) every L2 form is the sum of an exact and a co-exact form (more
precisely, up to L2 closure). Let us understand this first for smooth, compactly sup (R2 ). Then we seek f, g C (R2 )
ported . Thus, let = a dx + b dy with a, b Ccomp
with
= df + dg
(6.7)
Since C is simply connected, this is equivalent to writing = + where d = 0 and

d = 0. This in turn shows that (6.7) is equivalent to writing a smooth, compactly
supported vector field as the sum of a divergence-free field and a curl-free field. To find
f, g, we apply d and d to (6.7) which yields
(6.8)
f = ax + by ,
g = ay + bx
(R2 ). A solution
We therefore need to solve the Poisson equation f = h with h Ccomp
to this equation is not unique; indeed, we can add linear polynomials to f . On the
other hand, solutions that decay at infinity are necessarily unique from the maximum
principle. To obtain existence, we invoke the fundamental solution of the Laplacian on
1
log |z|. This means that = 0 in the sense of distributions
R2 , which is (z) = 2
and we therefore expect to find a solution via f = h. We now derive the solution to
Poissons equation in the smooth setting; much weaker conditions suffice but we do not
wish to dwell on that for now.
(R2 ). Then the function
Lemma 6.15. Let h Ccomp
ZZ
ZZ
1
1
h() log |z | dd =
h(z ) log || dd
(6.9)
f (z) :=
2
2
R2
R2
(with = + i) satisfies f C and solves f = h. Moreover,

(6.10)
where hhi :=
the form
f (z) =
R
R2
1
hhi log |z| + O(1/|z|) as |z|
h dd is the mean of h. In fact, f is the unique solution which is of
(6.11)
f (z) = k log |z| + o(1) as |z|
for some constant k R.

Proof. Differentiating under the integral sign yields
ZZ
1
z h(z ) log || dd
f (z) =
2
R2
ZZ
1
=
lim
h(z ) log || dd
2 0 |z|>
Now apply Greens identity (which follows from Stokes theorem on manifolds)
Z
Z
v
u
u ) d
(v
(vu uv) dd =
n
n
4. STATEMENT AND EXAMPLES OF THE HODGE DECOMPOSITION
113
to conclude that (and using that log || is harmonic away from zero)
ZZ
log || h(z ) dd
|z|>
ZZ
h
i
log || h(z ) h(z ) log || dd
|z|>
Z
h
i
log ||
=
h(z ) h(z )
log || d
n
n
|z|=
where n is the outward pointing norm vector relative to the region |z | > . Thus,
1
log || =
n
||
In conclusion, letting 0 yields f = h as desired. We remark that in dimensions

d 3 the fundamental solutions of are cd |x y|2d with a dimensional constant cd
by essentially the same proof. In d = 1 a natural choice is x+ := max(x, 0) or anything
obtained from this by adding a linear function. Inspection of our solution formula (6.9)
now establishes (6.10); indeed, simply expand the logarithm
log |z | = log |z| + log |1 /z|
To establish the uniqueness, one uses that the only harmonic function of the form (6.11)
vanishes identically (use the mean value property). In particular, the solution f of
f = h and h as above decays at infinity if and only if hhi = 0.

Returning to our discussion of Hodges decomposition, recall that h is given by the
right-hand sides of (6.8). Evidently, in that case hhi = 0 so that (6.9) yields smooth
functions f, g decaying like 1/|z| at infinity and which solve (6.8). It remains to check
that indeed
= df + dg = (fx gy )dx + (fy + gx )dy
To this end we simply observe that
ZZ
1
a() log |z | dd = a(z)
(fx gy )(z) =
2
2
Z ZR
1
(fy + gx )(z) =
b() log |z | dd = b(z)
2
R2
To obtain the final two equality signs no calculations are necessary; in fact, since a
vanishes at infinity, the only decaying solution to f = a is f = a and the same holds
for b. To summarize: we have shown that every compactly supported 1 (C; C) is
the sum of an exact and a co-exact smooth 1form and each of these summands typically
decay only like |z|2 (as differentials of functions decaying like |z|1 ). It is important to
note that this rate of decay is square integrable at infinity relative to Lebesgue measure
in the plane; if this were not so, then one could not derive the Hodge decomposition in
the L2 setting. In particular, we see here this is crucial that we are dealing with forms
and not functions, as the harmonic functions themselves only decay at the rate |z|1
which is not square integrable in the plane.
It is important to now that our proof extends to the L2 setting; that is, given a, b L2
there exist f, g H 1 (R2 ) (which is the Sobolev space of L2 functions with an L2 weak
derivative) so that a dx + b dy = df + dg as an equality between L2 functions. As usual,
this can be obtain from the smooth, compactly supported case which we just discussed
114
since those functions are dense in H 1 (R2 ). The required uniform control required to pass
to the limit is given by the L2 boundedness of the double Riesz transforms Rij . We leave
the the details of this to the reader, see Problem 6.1.
Example 2: M = S 2 or M is any compact, simply connected Riemann surface.
Since we already observed that h(M ) = {0} in this case, we see that trivially h2 (M ) =
{0} so that Hodges decomposition (6.6) again reduces to
12 (M ) = E E
Indeed, it is easy to prove that every 1 (M ) is of the form

= df + dg,
f, g C (M )
Applying d to this yields d = M g vol where vol is a suitably normalized volume

form on M and M is the LaplaceBeltrami operator on M . Recall that the Laplace
Beltrami operator M on any compact orientable manifold M has discrete spectrum
and L2 (M ) has an orthonormal basis consisting of (smooth) eigenfunctions Yn of M
with eigenvalues {n }
n=0 , 0 = 0 < 1 2 . . .. The lowest eigenvalue 0 is simple,
i.e., Y0 is necessarily a constant as follows from the maximum principle. Finally, from
Weyls law the eigenvalue n grow at some power rate in n. Hence, continuing in this
degree ofRgenerality, one concludes that M f = h with f C (M ), say, has a solution
iff hhi = M h dv = 0 (v is the volume form on M ). In that case the solution is given by
X
f=
2
n hh, Yn iYn
n1
which converges rapidly

R since h is smooth. Returning to d = M g vol, one sees from
Stokes theorem that M d = 0 so that the integrability condition holds. This yields a
smooth solution g whence dg is a closed form on M . Since M is simply connected, it
is also exact and thus = df + dg for some smooth f . Note that the simple connectivity
of M entered only at the final step. In the following example, we shall see how the
harmonic functions precisely eliminate the obstruction to the exactness of closed forms
in the genus one case.
Example 3: M = T2 = C/Z2 = C/hz 7 z + 1, z 7 z + ii

In view of our previous discussion of the harmonic forms in this case, (6.6) reduces
the following: any = a dx + b dy with smooth, Z2 periodic functions a, b can be written
as
a dx + b dy = df + dg + c1 dx + c2 dy
(6.12)
where f, g are smooth, Z2 periodic functions and suitable constants c1 , c2 . In fact, it will
turn out that
Z Z
Z Z
1
c1 =
a(x, y) dxdy,
b(x, y) dxdy
c2 =
As in the discussion of the whole plane, finding f, g reduces to a suitable Poisson equation.
Hence, let us first understand how to solveRfR = h on T2 with smooth h. Integrating
1 1
over T2 shows that the vanishing condition 0 0 h(x, y) dxdy = 0 is necessary. It is also
sufficient for solvability; indeed, any such smooth h has a convergent Fourier expansion
X
1 , n2 )e(xn1 + yn2 )
h(n
h(x, y) =
n1 ,n2
5. PROBLEMS
115
0) = 0 (and with e(x) := e2ix ). The solution to f = h is therefore given by

where h(0,
f (x, y) =
n1 ,n2
1 , n2 )
h(n
e(xn1 + yn2 )
4 2 (n21 + n22 )
which is again smooth. We write this schematically as f = 1 h. As in the case

of M = C, solving (6.12) reduces to solving (6.8) on T2 . Notice that our vanishing
condition is automatically satisfied and we therefore obtain smooth solutions f, g. In
order to conclude that = df + dg it remains to check that a = 1 a and the same
for b. This is true iff a and b have vanishing means and confirms our choice of c1 , c2
above. Here we see an example where h(M ) = h2 (M ) plays a topological role. This is
typical of the compact case but not of the non-compact case.
Example 3: M = D
Our fourth example is the disk D (which is the same as the upper half-plane or
any other simply connected true subdomain of C). In this case there is not only an
abundance of harmonic and holomorphic one-forms, but also of square integrable ones.
First, we remark that
E = {df | f H01 (D)}
where H01 (D) is the usual Sobolev space with vanishing trace on D (see [12], for example). Second, let us reformulate (6.6) as an equivalent fact for vector fields ~v = (v1 , v2 )
L2 (D) rather than forms: there exist f, g H01 (D), as well as ~ = (1 , 2 ) smooth and
both divergencefree and curlfree, and with 1 , 2 L2 (D) so that
~ +
~ g + ~
~v = f
~ g := (gy , gx ). To find f , we need to solve

where
f = div ~v H 1 (D),
f H01 (D)
whereas for g, we need to solve

g = div ~v H 1 (D),
g H01 (D)
where div~v = y v1 + x v2 . This can be done uniquely with f, g H01 (D) via the usual
machinery of weak solutions for elliptic equations, see the chapter on elliptic equations
in [12]. For the uniqueness, suppose that g = 0 and g H01 (D). Then g H 1 (D) =
H01 (D) and
Z
|g|2 dxdy
0 = h g, gi =
D
which implies that g is constant and therefore zero. This shows that any E which
is also harmonic is zero. Notice the importance of the boundary condition in this
regard which was built into the space E (coming from the compact support condition).
Of course there are many (nonzero) harmonic differentials which are also in L2 (D), but
(D) only if they vanish identically.
they are limits of differentials df with f Ccomp
5. Problems
Problem 6.1. Provide the details for the L2 extension of our proof of the Hodge
decomposition in the plane.
116
Problem 6.2. Discuss Hodges theorem for the Riemann surfaces M = {r1 < |z| <
r2 } where 0 r1 < r2 . Identify 12 (M ), h2 (M ) as well as E for each of these cases
and show directly that
12 (M ) = E E h2 (M )
Problem 6.3. Let M be a simply connected Riemann surface and let u be a harmonic
function on M . Then u has a global harmonic conjugate on M .
Problem 6.4. This exercise introduces and studies Bessel functions Jn (z) with n
Z, z C: they are defined as the coefficients in the Laurent expansion

z
X
1
Jn (z) n , 0 < || <
(6.13)
exp ( ) =
2
n=
(a) Show from the generating function (6.13) that for each n Z the function Jn (z)
is entire and satisfies
Z
1
(6.14)
Jn (z) =
cos(n z sin ) d
0
as an identity between entire functions. Also, prove that Jn = (1)n Jn .
(b) Using (6.14) prove that for each n Z, w = Jn (z) satisfies Bessels equation
(6.15)
z 2 w (z) + zw (z) + (z 2 n2 )w(z) = 0
This equation, which arises frequently in both mathematics and physics (as well as other
applications), is the reason why Bessel functions are so important.
In what follows, C := C \ {0}. Also, for the remainder of (b) we allow n = C
in (6.15). Prove that for any z0 C as well as w0 , w1 C arbitrary there exists a
unique function w(z) defined and analytic locally around z = z0 with the property that
w(z0 ) = w0 , w (z0 ) = w1 and so that (6.15) holds on the domain of w (use power series
around z = z0 ). We refer to such a solution as a local solution around z0 . What happens
at z0 = 0? Show that any local solution around an arbitrary z0 C can be analytically
continued to any simply connected domain C containing z0 . Moreover, show that
for any simply connected domain C there exist two linearly independent solutions
W0 , W1 H() of (6.15) so that any local solution w around an arbitrary z0 is
a linear combination of W0 , W1 (such a pair is referred to as a fundamental system of
solutions on ).
Given a local solution w(z) around an arbitrary z0 C , set f () = w(e ) which is
defined and analytic around any 0 with e0 = z0 . Derive a differential equation for f and
use it to argue that f can be analytically continued to an entire function (in the language
of Riemann surfaces this shows that e uniformizes the Riemann surface of any local
solution of Bessels equation; loosely speaking, the worst singularity that a solution of
Bessels equation can have at z = 0 is logarithmic).
(c) Using either (6.13) or (6.14) prove that the power series expansion of Jn (z)
around zero is
X
(1)k (z/2)2k
n
(6.16)
Jn (z) = (z/2)
k!(n + k)!
k=0
provided n 0. What is the power series of Jn forPn < 0?

k
(d) Suppose the formal power series wn (z) =
k=0 ak,n z satisfies the ordinary differential equation (6.15) with some fixed integer n 0. Derive a recursion relation for
5. PROBLEMS
117
the coefficients ak,n and show that up to a multiplicative constant the formal power series
equals (6.16), i.e., wn is a multiple of Jn . In particular, Jn is the only solution of (6.15)
(up to multiples) which is analytic around z = 0.
(e) Find a fundamental system of solutions of Bessels equation (6.15) with n = 0
on G = C \ (, 0] (it will help if you remember from (b) that the worst singularity
at z = 0 of any solution of (6.15) is logarithmic). Of course you need to justify your
answer. What about general n 0?
(f ) We now use (6.16) to define J for C by the formula
(6.17)
X
(1)k (z/2)2k
J (z) = (z/2)
k! ( + k + 1)
k=0
where we select the principal branch of

= eLog z for definiteness. Hence, we view (6.17)
as an element of H(C \ (, 0]). Check that
2 1
2
sin z
J 1 (z) =
2
z

2 1 sin z
2
J 3 (z) =
cos z
2
z
z
Show that (6.17) agrees with the previous definition for all integer , including negative
ones. Argue that J thus defined solves (6.15) with n replaced by C, and prove that
this property fails if we were to define J by replacing n with C in (6.14) (which
explains why we used the power-series instead). It is worth noting that for any C,
the function J (e ) is entire in (why?). This is in agreement with our abstract result
from part (b).
(g) Prove that for all C with Re > 21 there is the representation
Z 1
1
(z/2)
J (z) =
eizt (1 t2 ) 2 dt
1
( + 2 ) 1
Z
(z/2)
=
cos(z cos ) sin2 d
1
( + 2 ) 0
Check directly that these integral representations satisfy the Bessel equation (6.15) with
n = .
(h) Using (6.17) show that for any C
Z

z 2 dw
(z/2)
exp w
J (z) =
2i
4w w+1
(6.18)
Z
z
d
1
exp
( 1 ) +1
=
2i
2
where is a Hankel contour (see (2.18)). In both cases the powers involving are principal branches. (6.18) should of course remind you of our starting point (6.13). Indeed,
check that for = n Z the representation (6.18) is nothing but the integral computing
the nth Laurent coefficient of (6.13). Note that the power +1 in the denominator is
single-valued if and only if Z.
Finally, deduce from (6.18) that
Z
1
ez sinh d
(6.19)
J (z) =
2i e
118
where
e = Log is the (principal) logarithm of a Hankel contour , see Figure 12.1.
(i) Use (6.19) to prove the recursion relation of the Bessel functions
J1 (z) + J+1 (z) = (2/z)J (z)
and from these that
J1 (z) J+1 (z) = 2J (z)

J+1 (z) = (/z)J (z) J (z)
J1 (z) = (/z)J (z) + J (z)

In particular, J0 (z) = J1 (z).
CHAPTER 7
Hodges theorem and the L2 existence theory

1. Weyls lemma and the Hodge decomposition
In this chapter we develop some of the basic potential theory which is essential for
various existence theorems on Riemann surfaces. Not only are we going to obtain the
Hodge decomposition this way, but we shall also be able to prove that every Riemann
surface carries a nonconstant meromorphic function. This important result is one of the
ingredients in the proof of Theorem 5.19 on compact Riemann surfaces.
We shall now prove Hodges representation (6.6). Recall that 12 (M ; R) is the space
of real-valued 1forms with measurable coefficients and such that
Z
<
kk2 =
M
Furthermore,

0
(M ; R) ,
E := df | d comp

0
E := df | d comp
(M ; R)
where the closure is in the sense of 12 (M ; R).
Theorem 7.1. Let h2 (M ; R) := h(M ; R) 12 (M ; R). Then

12 (M ; R) = E E h2 (M ; R)
We begin with the following observation.
Lemma 7.2. Let 12 (M ; R) and smooth. Then E iff d = 0 and (E)

iff d = 0. In particular, E (E) and E E .
Proof. First,
Moreover,
E df | f Ccomp
(M )
Z
df
0 = h, df i = h, df i =
M
Z
Z
f d
d(f ) f d =
=
M
(M ) is the same as d = 0. Thus, is co-closed. The calculation for

for all f Ccomp
(E) is essentially the same and we skip it.
This lemma implies that

12 (M ; R) = E E (E (E) )
and our remaining task is to identify the intersection on the right. Note that Figure 6.2
becomes clear when compared to Lemma 7.2. It is clear from Lemma 7.2 that
E (E) h2 (M ; R)
119
120
7. THE HODGE THEOREM
It remains to show equality here. This is very remarkable in so far as the intersection thus
consists of smooth 1-forms. The required (elliptic) regularity ingredient in this context
is the so-called Weyl lemma, see Lemma 7.4 below. Note that the following lemma
concludes the proof of Theorem 7.1.
Lemma 7.3.
E (E) = h2 (M ; R)
Proof. Take E (E) . Then by Lemma 7.2,
f Ccomp
(M )
h, df i = h, df i = 0,
and f complex-valued (say). With = u dz + v d

z in local coordinates (U, z), z = x + iy
and with f supported in U , we conclude that
Z
h, df i = 2 (u fz + v fz ) dxdy
Z
h, df i = 2i (u fz v fz ) dxdy
This system is in turn equivalent to
Z
u
fz dxdy = 0,
vfz dxdy = 0
for all such f . Now setting f = gz and f = gz , respectively, where g is supported in U ,

yields
Z
Z
u
g dxdy = vg dxdy = 0
which implies by Weyls lemma below that u, v are harmonic and thus smooth in U . In
view of Lemma 7.2, is both closed and co-closed and therefore harmonic.

Lemma 7.4. Let V C be open and u L1loc (V ). Suppose u is weakly harmonic,
i.e.,
Z
(V )
u dxdy = 0 Ccomp
V
Then u is harmonic, i.e., u C (V ) and v = 0.
Proof. As a first step, we prove this: suppose {un }

n=1 C (V ) is a sequence of
1
harmonic functions that converges in the sense of Lloc to u . Then u C (V ) and
u is harmonic.
This follows easily from the mean-value property. Indeed, for each n, and each disk
D(z, r) V ,
ZZ
1
un () dd
un (z) =
|D(z, r)|
D(z,r)
Hence, by the assumption of L1loc convergence, {un }n is a Cauchy sequence in C(V ) and
therefore converges uniformly on compact subsets of V to u which is thus continuous.
Moreover, it inherits the mean value property
Z
1
u (z) =
u () d()
2r |z|=r
and is thus harmonic. Indeed, simply compare u to that harmonic function u

e which takes
values u on the boundary of some disk D (we know that u
e exists from Chapter 3 since
1. WEYLS LEMMA AND THE HODGE DECOMPOSITION
121
we can solve the Dirichlet problem for the Laplacean on a disk). Now observe that u
eu
both satisfy the mean-value property and therefore also the maximum principle. Since
u
e u = 0 on the boundary of some disk D, necessarily also u
e u = 0 on D. See also
Problem 3.5, part (x).
To conclude the proof, we let u L1loc (V ) be weakly harmonic and define
un (z) := (u n )(z)
z V 1
n2 (n)
where
for all n 1 with 0 a smooth bump function, supp() D,
R n () :=
and = 1. Furthermore,
Vn := {z V | dist(z, V ) > 1/n}
Since u is weakly harmonic, un = 0 on Vn and un u in L1loc (V ) by Lemma 3.3 (which

applies to approximate identities in R2 equally well). By the previous paragraph, u is
smooth and harmonic as claimed.

Let us now explain why there is always a nonzero harmonic form on a compact surface
with positive genus. This goes through the standard loop-form construction. Thus, let
c : [0, 1] M be a smooth, closed curve where M is an arbitrary Riemann surface. For
1
f
M
c1
c2
Figure 7.1. The function f giving rise to the loop-form c

simplicity, we shall also assume that c([0, 1]) is an imbedded onedimensional manifold
and we put the natural orientation on it, i.e., c(t) is oriented according to increasing t
e N be neighborhoods to the
(such a closed curve will be called loop). Then let N
left of c obtained by taking the finite union of left halves (relative to c) of parametric
e are
disks centered at points of c. Furthermore, the disks used in the construction of N
assumed to be compactly contained in those for N . Then let f be a smooth function
e , f = 0 on N \ N
e and f = 0 on M \ N , see Figure 7.1.
on N with f = 1 on N
While f is not smooth on M , the loop-form of c defined as
c := df 1 (M ; R)
is smooth and compactly supported. By construction, dc = 0 and the cohomology class

of c is uniquely determined by the homology class of c, cf. Figure 7.1. First, if f1 and f2
122
(M )
are smooth functions constructed to the left of c as explained, then f1 f2 Ccomp
so that df1 df2 is exact. Second, if c1 c2 is the boundary of a compact submanifold
of M , then c1 c2 = dh for some smooth, compactly supported function on M . Finally,
if we had defined loop-forms ec = dg with a function g that equals one on a neighborhhod
to the right of c, then c ec is exact. In other words, reversing the orientation of c
merely changes the sign of (the cohomology) class of c . The importance of loop-forms
can be seen from the following simple but crucial fact:
Lemma 7.5. Let 1 be closed. Then

h, c i =
Proof. Compute
Z
df
c =
N
Z
Z
Z
f =
d(f ) =
=
h, c i =
as claimed.
This property of loop-forms immediately allows us to characterize all exact forms

amongst the closed ones as well as give a general topological criterion which guarantees
that non-zero harmonic forms exist.
Corollary 7.6. 1) Let 1 (M ; R). Then is exact iff h, i = 0 for all co-closed
1 (M ; R) of compact support.
2) Let E be smooth. Then is exact, i.e., = df for some real-valued f
C (M ).
3) Supposed the closed loop c does not separate M , i.e., M \c([0, 1]) is connected. Then
there exists a closed form 1 (M ; R) which is not exact. In particular, h2 (M ; R) 6=
{0}.
Proof. If = df is exact, then
Z
Z
df =
h, i =
f d = 0
for any as in 1). Conversely, c is co-closed and compactly supported for any loop. It
follows that
Z
0 = h, c i =
c
for any loop c. Thus is exact as claimed.

Property 2) follows from 1) via Lemma 7.2. Finally, for 3), let c be a closed curve
in M that crosses c transversally. See Figure 7.2. This exists since M \ c is connected.
Hence,
Z
c = 1
and c is closed but not exact. From Theorem 7.1,

c = + ,
E,
h2 (M ; R)
Since and c is smooth, so is . By 2), is exact so 6= 0 as desired.
2. EXISTENCE OF NONCONSTANT MEROMORPHIC FUNCTIONS
123
c*
M
c
Figure 7.2. The curves c and c1

2. Existence of nonconstant meromorphic functions
We shall now derive some very important consequences from Hodges theorem. More
precisely, we answer the fundamental question: does a general Riemann surface carry
a nonconstant meromorphic function? We saw that in general this cannot be done
with holomorphic functions, since compact surfaces do not allow this. However, deeper
questions will still elude us here such as: which Riemann surfaces carry a meromorphic
function with exactly one simple pole? In the following chapter we shall present the
machinery (known as the Riemann-Roch theorem) needed for this purpose.
The basis for our entire existence theory will be the following result. It should be
thought of as an answer to the following question: Let p M . Can we find a function
u harmonic on M \ {p} so that in some parametric disk centered at p, u has a given
singularity at p such as 1z or log |z|?
Or, more generally: let D be a parametric disk on a Riemann surface M centered at
\ {p}. Can
p M and suppose h is a harmonic function on D \ {p}, differentiable on D
we find u harmonic on M \ {p} with u h harmonic on all of D?
These questions are part of potential theory and we shall address them by means of
the Hodge theorem. As usual, M and N are arbitrary Riemann surfaces.
M, N
compact with smooth boundary. Fix p0 N and
Proposition 7.7. Let N
1
) and h = 0 on N where n is some normal
h harmonic on N \ {p0 } with h C (N
n
vector field on N that never vanishes.
Then there exists u harmonic in M \ {p0 }, u h harmonic on N , and u 12 (M \ K)
for any compact neighborhood K of p0 . Also, u is unique up to constants.
Proof. For the existence part, take a C function on N which agrees with h on
N \ K where K is an arbitrary but fixed (small) compact neighborhood of p0 . Then
extend to M simply by setting it = 0 outside of N and define d = 0 on M \ N . Note
that d 12 (M ; C). By Hodge,
d = + ,
E, E
124
(M ), then
If Ccomp
hd, di = h, di,
h, di = 0
First, suppose that supp() M \ K. Then from
h
n
= 0 and d dh = 0 on N , we
Figure 7.3. The sets in Proposition 7.7

obtain that
hd, di =
=
d dh
i (dh) = 0
N
where i : N M is the inclusion map. The main point to note here is that i (dh) is
h
= 0.
proportional to n
Hence, is harmonic on M \ K. On the other hand, if supp() N , then
hd , di = 0,
hd , di = 0
so that d is harmonic on N . In particular, is smooth on M and thus = df with

f smooth. Now set
u=f +h
By inspection, u has all the desired properties.

Finally, if v had the same properties as u, then u v would be harmonic on M and

d(u v) 12 . In conclusion, d(u v) E h2 = {0}, so u v = const.
We remark that if h were harmonic on all of N , then h = const because of the
h
= 0 on N . Indeed, this is merely the fact that
Neumann condition n
Z
Z
2
=0
h i (dh)
dh dh =
kdhkL2 (N ) =
N
where i : N M is the inclusion and i the pull-back as in the proof of the previous
proposition.
Furthermore, we remark that the exact same proof allows for several exceptional
points p0 , . . . , pk N . The statement is as follows:
125
M, N
compact with smooth boundary. Fix finitely many
Corollary 7.8. Let N
k
) and h = 0 on N
points {pj }j=0 N and h harmonic on N \ {pj }kj=0 with h C 1 (N
n
where n is some normal vector field on N .
Then there exists u harmonic in M \ {pj }kj=0 , u h harmonic on N , and u
12 (M \K) for any compact neighborhood K of {pj }kj=0 . Also, u is unique up to constants.
Proof. The proof is essentially the same as that or Proposition 7.7 and we leave
the details to the reader.

We can now collect a number of corollaries:
Corollary 7.9. Given n 1 and a coordinate chart (U, z) around p0 in M with
z(p0 ) = 0 there is u harmonic on M \ {p0 } with u z n harmonic on U and du
12 (M \ K) for any compact neighborhood K of p0 .
and define
Proof. Simply let without loss of generality z(U ) D
h(z) = z n + z n |z| 1
= z 1 (D)
since h = 0 on |z| = 1.
The theorem applies with N
n
Next, we would like to place a log |z| singularity on a Riemann surface. To apply
h
= 0. This amounts to
Proposition 7.7 we need to enforce the Neumann condition n
solving the Neumann problem
u
u = 0 in |z| < 1,
= log r = 1 on |z| = 1
n
r
But this has no solution since the integral of 1 around |z| = 1 does not vanish (necessary
by the divergence theorem). Now let us also note that with M = C the function u(z) =
log |z| satisfies
z
1 dz 1 d
+
du(z) =
2 z
2 z
which is not in L2 around |z| = (it barely fails). Finally, this calculation also shows
that if we could place a log |z| singularity on M then this would produce a meromorphic
differential = du + i du with exactly one simple pole. If M is compact, then this
violates the fact that the sum of the residues would have to vanish.
What all of this suggests is that we should try with two logarithmic singularities (in
other words, instead of using a point charge, we use a dipole). This is indeed possible:
Corollary 7.10. Let p0 , p1 M be distinct and suppose z and are local coordinates
around p0 and p1 , respectively. Then there exists u harmonic on M \ {p0 , p1 } with
u log |z| and u + log || harmonic locally around p0 , p1 , respectively. Moreover, du
12 (M \ K) where K is any compact neighborhood of {p0 , p1 }.
Proof. For this one, assume first that p0 , p1 are close together. Then let (U, z) be
Define
a coordinate chart with z(p0 ) = z0 D \ {0}, z(p1 ) = z1 D \ {0} and z(U ) D.
(z z )(z z )

0
0
h(z) = log

(z z1 )(z z1 )
where z0 , z1 are the reflections of z0 , z1 across D (i.e., zj = zj1 ). Then check that
|(z zj )(z zj )| = |z|2 |(z zj )(z zj )|
126
h
which gives h(z ) = h(z) (why?). This in turn implies the Neumann condition n
= 0.
Hence, by Corollary 7.8, there exists u with all the desired properties. If p0 and p1 do
not fall into one coordinate chart, then connect them by a chain of points that satisfy
this for each adjacent pair. This yields finitely many functions u0 , u1 , u2 etc. The desired
function is the sum of all these.
To conclude our existence theory, we now state some simple but most important
corollaries on meromorphic differentials.
Corollary 7.11. (a) Given n 1 and p0 M there exists a meromorphic differdz
holomorphic locally around p0 (here z are local coordinates at p0 ).
ential with z n+1
1
Moreover, 2 (M \ K) for every compact neighborhood K of p0 .
(b) Let p0 , p1 M . There exists meromorphic on M with dz
z holomorphic
d
around p0 and + holomorphic around p1 , respectively (with z, local coordinates).
Moreover, 12 (M \ K) for every compact neighborhood K of {p0 , p1 }.
Proof. With u as in Corollary 7.9 and 7.10, respectively, we set = du. In the first

case, = 1
2n ( + i ), whereas in the second, = + i .
As a reality check, take M = C and p0 = 0, say. Then for (a) we would simply obtain
dz
= z n+1
. Note that the L2 condition holds when n 1 but not for n = 0. For (b),
dz
dz
zp
. This has all the desired properties, including the L2
we would take = zp
0
1
condition at z = . In the classical literature, meromorphic differentials all of whose
residues vanish (as in (a)) are called differentials of the second kind, whereas those with
simple poles are called differentials of the third kind (the holomorphic differentials are
called abelian or of the first kind).
From the existence of differentials of the second kind we can easily derive the following
specical case of the uniformization theorem:
Corollary 7.12. Let M be compact and simply connected. Then M carries a meromorphic function of degree one. In particular, M S 2 . In other words, up to conformal
isomorphisms, there is exactly one compact simply connected Riemann surface, namely
CP 1 S 2 C .
Proof. Let H1 (M \ {p0 }) where p0 M is arbitrary and such that =
local coordinates around p0 . Then set
Z p
f (p) :=
dz
z2
in
p1
where the integration path connects an arbitrary but fixed p1 M \ {p0 } with p without
passing through p0 . Since M is simply connected and since res(; p0 ) = 0, it follows that
f is well-defined. Clearly, f has a simply pole at p0 and is holomorphic elsewhere. Since
deg(f ) = 1, this map induces an isomorphism M C as desired.

This result of course raises many questions, such as: what can one say about the
non-simply connected case? More precisely, on a compact surface of genus g, what is the
minimal degree that a nonconstant meromorphic function can achieve?, How many (in
the sense of dimension) meromorphic functions are there on a compact Riemann surface
which have poles at finitely many points p M with orders at most s (a positive
integer)?
127
The Riemann-Roch theorem in the following chapter attempts to answer these questions, at least it relates the dimension of a certain space of meromorphic functions to the
dimension of a space of meromorphic differentials.
For now, let us mention that the proof of Corollary 7.12 gives the following, more
precise, statement: Define the linear space
V := {f M(M ) : f H(M \ {p0 }), ord(f, p0 ) 1}
Then dim(V ) = 2.
The reader will have no difficulty verifying this from the previous proof. Note that
we need dimension 2 here since the constants are in V . The Riemann-Roch theorem will
generalize these dimension counts to arbitrary divisors on compact surfaces, see the
following chapter for the definition of a divisor. In Problem 7.4 we give an extension of
our observation concerning dim(V ) which is a special case of Riemann-Roch for genus
zero.
Finally, we can now state and prove the following very satisfactory result which
applies every Riemann surface.
P
Theorem 7.13. Let {pj }Jj=1 M , J 2, and cj C with Jj=1 cj = 0. Then there
exists a meromorphic differential , holomorphic on M \ {p1 , p2 , . . . , pJ } so that has a
simple pole at each pj with residue cj .
Proof. Pick any other point p0 M and let j be meromorphic with simple poles
PJ
at p0 , pj and residues cj , cj , respectively. The differential =
j=1 j has all the
desired properties.
1
1
p2
p1
p3
Figure 7.4. The construction of a meromorphic function with a pole at

p0 (and possibly other poles)
From here we immediately conclude the following remarkable result.
Corollary 7.14. Every Riemann surface carries a nonconstant meromorphic function. In fact, given any distinct points p0 , p1 , p2 M , there exists a meromorphic function f on M for which f (p1 ) = 1, and so that f has a simple pole at p0 and a simple
128
zero at p2 . In particular, the function field M(M ) separates points1 on every Riemann
surface M .
Proof. Take three points p0 , p1 , p2 M and let 1 be a meromorphic one-form with
simple poles at p0 , p1 and residues 1, 1, respectively and holomorphic everywhere else.
Similarly, let 2 be a meromorphic one-form with simple poles at p1 , p2 and residues 1, 1,
respectively and holomorphic everywhere else. See Figure 7.4. Now set f = 21 where the
division is well-defined in local coordinates and defines a meromorphic function. Clearly,
f (p1 ) = 1 and f (p2 ) = 0, and the function f is not constant, meromorphic with a pole
at p0 .

In the following chapter we shall study the vector space (or rather, its dimension) of
meromorphic functions and differentials with zeros and poles at prescribed points.
3. Problems
Problem 7.1. Show that every compact Riemann surface admits a triangulation.
Problem 7.2. Show that the meromorphic functions on every Riemann surface separate points in the following strong sense: given distinct points {pj : 1 j n} on
some Riemann surface M , and n distinct values zj C , there exists a meromorphic
function on M which takes the value cj at pj (of course f (pj ) = means that f has a
pole at pj ).
Problem 7.3. Let M be a Riemann surface. Here you are asked to give an alternative
proof for the existence of a meromorphic differential which has poles of order 2 at
prescribed points p1 , . . . , pn M with vanishing residues and is holomorphic everywhere
else by the following strategy: by linearity, n = 1. Pick a parametric disk (U, z) around
where is a smooth cut-off function which is
p1 with z(p1 ) = 0. Then let := (z) dz
z2
supported in the unit disk and so that = 1 near zero. Then is a smooth 1-form on
M \ {p1 } with the property that := i is smooth on M , and in fact, = 0 near p1 .
Apply the Hodge decomposition theorem to write, with a harmonic form ,
= df + dg +
where f, g are smooth (since is smooth prove it). Now use this to define a meromorphic differential with the desired properties.
Problem 7.4. Let2 M be a compact Riemann surface of genus zero. Select finitely
many points {p }n=1 M as well as a positive integer s for each 1 n. Define
V := {f M(M ) | f H(M \ {p }n=1 ), the pole of f at p has order at most s }
P
Prove that dim(V ) = 1 + n=1 s .
1This means that to two distinct points there exists a meromorphic function taking distinct values
at these points. Note that we cannot separate points by means of holomorphic functions, in general, as
they may be constant (as on a compact surface).
2This problem is exactly the genus zero case of the Riemann-Roch theorem for integral divisors.
CHAPTER 8
The Theorems of Riemann-Roch, Abel, and Jacobi

1. Homology bases, periods, and Riemanns bilinear relations
We will now turn to the following much deeper question: what kind of nonconstant
meromorphic functions does a given Riemann surface admit? More precisely, if M is
compact and of genus g, what can we say about the minimal degree of a meromorphic
function on M ? Answering this question will lead us to the Riemann-Roch theorem. To
appreciate this circle of ideas, note the following: Suppose M is compact and admits a
meromorphic function f of degree one. Then M defines an isomorphism between M and
CP 1 . This implies that no such function exists if M has genus one or higher! On the
other hand, we will show (from Riemann-Roch) that for the simply connected compact
case there is such a function. In this chapter, the reader will need to know about basic
topology of compact surfaces: homology, the canonical homology basis, the fundamental
polygon of a compact surface of genus g. These topics are briefly discussed in Chapter 12.
In addition, we will need the loop-forms from the previous chapter.
Let M be a compact Riemann surface of genus g. The intersection numbers between
two closed curves 1 , 2 are defined as
Z
1 2 = h1 , 2 i
1 2 =
M
which is always an integer. Recall that for an (oriented) loop is = df where f

is a smooth function of compact support with f = 1 on a small neck to the left of .
The homology class of determines the cohomology class of . Hence, the intersection
number is well-defined as a product between homology classes in H1 (M ; Z). Note that
b a = a b,
(a + b) c = a c + b c
for any classes a, b, c.

Pick a (canonical) homology basis for the 1-cycles, and denote it by {Aj }2g
j=1 . Here
Aj = aj if 1 j g and Aj = bjg if g + 1 j 2g where
aj bk = jk ,
aj ak = 0,
bj bk = 0
for each 1 j, k g. Next, we define a dual basis {k }2g

k=1 for the cohomology. It is
simply
k = bk , 1 k g
k = akg , g + 1 k 2g
and satisfies the duality relation
(8.1)
k = jk .
Aj
129
130
8. THE THEOREMS OF RIEMANN-ROCH, ABEL, AND JACOBI
We also record the important fact

nZ
o2g
(8.2)
j k
j,k=1
0 I
I 0
=: J
This is due to the fact that the entries of this matrix are all possible intersection numbers
of the curves aj , bk .
Lets collect some important properties (a form is called real if = ):
1
Lemma 8.1. The real one-forms {j }2g
j=1 are a basis of H (M ; R) (de-Rham space of
one forms). Let j denote the orthogonal projection of j onto the harmonic forms (from
the Hodge theorem). Then j is a real one-form, and {j }2g
j=1 is a basis of both h(M ; R)
and h(M ; C), the real and complex-valued harmonic forms, respectively. In particular,
dimR h(M ; R) = dimC h(M ; C) = 2g

R
The relation (8.2) holds also for M j k .
Proof. Since {Aj } is a basis of H1 (M ; Z), a closed form is exact iff

Z
= 0 1 j 2g
Aj
Hence the linear map
H 1 (M ; R)
R2g
Aj
2g
j=1
is injective. Because of (8.1) this map is also onto and is thus an isomorphism. It is called
the period map. The exact same argument also works over C. Since every cohomology
class has a unique harmonic representative, we obtain the statements about h. To check
that j is real, write
j = dfj + j
j = j = dfj + j
so that
j + j = d(fj fj )
is both harmonic and exact, and thus zero.
Next, we find a basis for the holomorphic one-forms H1 (in the classical literature,
these are called differentials of the first kind).
Lemma 8.2. With j as above, define the holomorphic differential j = j + i j .
Then {j }gj=1 is a basis in H1 . In particular, dimC H1 = g.
Proof. The dimension statement is immediate from
h(M ; C) = H1 H1
To see the statement about the basis, we express as a matrix relative to the basis
{j }2g
j=1 . This is possible, since preserves the harmonic forms. Also, note that it
preserves real forms. Hence, with jk R,

2g
X
1 2
j =
jk k , A = GA, G =
GL(2d, R)
3 4
k=1
1. HOMOLOGY BASES, PERIODS, AND RIEMANNS BILINEAR RELATIONS
131
where A is the column vector with entries 1 , . . . , 2g . From = Id we deduce

G 2 = I2d . We expect 2 , 3 to be invertible since heuristically should correspond to
a switch between the aj and the bk curves. Indeed, we have
Z
2g
2g
X
X
k
jk
jk hk , i =
hj , i = h j , i =
k=1
k=1
or, in matrix notation,
= GJ =
2 1
4 3
where = {hj , i}2g

j,=1 is a positive definite matrix (since it is the matrix of a positive
definite scalar product) and J is as above, see (8.2). Hence 2 > 0 and 3 > 0; in
particular, these matrices are invertible.
Suppose there were a linear relation
c1 1 + . . . + cg g = (v t + iwt ) (A1 + i A1 ) = 0

1
where v, w Rg are column vectors, and A = A
A2 . By the preceding paragraph,
A1 = 1 A1 + 2 A2 so that

[v t A1 wt (1 A1 + 2 A2 )] + i wt A1 + v t (1 A1 + 2 A2 )] = 0
Since the terms in brackets are real one-forms, it follows that they both vanish. Therefore,
we obtain the following relation between the linearly independent vectors A1 , A2 of oneforms:
(v t1 w)t A1 = (2 w)t A2 , (t1 v + w)t A1 = (2 v)t A2
which finally yields 2 w = 2 v = 0, and thus v = w = 0 as desired.

To proceed, we need the following remarkable identity:
Lemma 8.3. Let , e be closed one-forms. Then

Z
Z
Z
Z
g Z
X
e
e =
(8.3)
M
aj
bj
bj
aj
j=1
In particular, if is harmonic, then

Z
Z
Z
g Z

X
2
kk2 =
aj
j=1
bj
bj
aj
Proof. The integral on the left-hand side of (8.3) only depends on the cohomology
e respectively. Thus, we can write
classes of and ,
=
2g
X
j j + df,
j=1
where j =
by (8.2).
Aj
,
ej =
Aj
e =
e It follows that
.
2g
X
j,k=1
j
ek
e =
2g
X
j k =
j=1
ej j + dfe
g
X
j=1
j
ej+g j+g
ej

132
We now state two important corollaries:

Corollary 8.4. Suppose H1 . Assume that either
R
all aperiods vanish, i.e., aj = 0 for all 1 j g
or all periods of are real
Then = 0.
Proof. From the previous lemma, since = i,
Z
Z
2
= i
kk =

M
=i
g Z
X
aj
j=1
bj
bj
aj
which vanishes under either of our assumptions.

Corollary 8.5. The map
(
R

R
,
.
.
.
,
a1
ag
H1 Cg
is a linear isomorphism. In particular, there exists a unique basis {j }gj=1 of H1 for

R
which aj k = jk .
Proof. This is an immediate consequence of the previous corollary and the fact that
dimC H = g.

R
This result raises the question what the other periods jk := bj k look like. Before
proceeding, consider the simplest example with g = 1, i.e., M = C/h1, i where h1, i is
the group generated by the translations z 7 z + 1, z 7 z + with Im > 0. It is clear
that in this case the basis of the previous corollary reduces to = dz with a-period 1,
and b-period . Here we chose the a-loop to be the edge given by z 7 z + 1, and the
b-loop as the edge z 7 z + .
Returning to the general case, given a basis {j }gj=1 of H1 , we call the g 2g matrix
whose j th row consists of the periods of j , the period matrix of the basis.
Lemma 8.6. Riemanns bilinear relations: The period matrix of the basis {j }gj=1
from above has the form
(I, ),
t = ,
I = Idgg ,
Im > 0
Proof. If , e are holomorphic, then e = 0. Hence,

Z
Z
Z
g Z
X
e
0=
aj
j=1
bj
bj
aj
In particular, setting = k , e = ,
Z
Z
Z
g Z

X
k

k = k k
0=
j=1
aj
bj
aj
bj
Next, we have
hk , j i =
=
k
M
g Z
X
=1
j = i
k
133
k j
Z
k
b

j = 2 Im kj
which implies that Im > 0.
In the case of the torus from the previous example, = and Im = Im > 0 by
construction. In other words, the periods 1, define a lattice in C = Cg (since g = 1
here). This turns out to have a generalization to higher genera, leading to the Jacobian
variety. Thus, let {k }gk=1 be any basis of the holomorphic differentials H1 (M ). It
turns out that the columns of the periodic matrix of this basis are linearly independent
over the reals:
Lemma 8.7. Let {k }gk=1 be any basis of the holomorphic differentials H1 (M ) and
g
let {Aj }2g
j=1 be the canonical homology basis as above. Then the 2g vectors in C (the
columns of the period matrix)
Z
Z
t
pj :=
1 , . . . ,
g
Aj
Aj
for 1 j 2g are linearly independent over R. In particular,

L(M ) :=
2g
nX
j=1
Cg .
n j pj | n j Z
is a lattice in
The quotient J(M ) := Cg /L(M ) is called the Jacobian variety and it
is a compact, commutative complex Lie group of dimension g.
Proof. Suppose the columns are linearly dependent over R. Then there exist j
R, 1 j 2g for which
Z
2g
X
= 0
1g
j
j=1
Aj
In other words,
h ,
Thus, for any H1 (M ),
2g
X
j=1
j Aj i = 0
h,
1g
2g
X
j Aj i = 0
2g
X
j Aj i = 0
j=1
and by the reality of the 1-form in the second slot here,

h,
j=1
134
P
for any harmonic differential . But this implies that 2g
j=1 j Aj = 0 and so j = 0 for
all 1 j 2g. That L(M ) is a lattice simply means that L(M ) is a discrete subgroup
of Cg which is clear, as are the stated properties of J(M ).

In the case of torus M = C/h1, i we see that J(M ) M . In fact, there is always
a (canonical) holomorphic map that takes a compact Riemann surface into its Jacobian
variety.
Lemma 8.8. The map : M J(M ) defined as
Z p t
Z p
p 7
1 , . . . ,
g
p0
p0
where p0 M is an arbitrary point, is a well-defined holomorphic mapping with maximal

rank one (over C). Here the integration runs over any path that connects p0 to p provided
we chose the same path for each of the g entries of .
Proof. Let and
e be two paths connecting p0 to p. Then for all 1 k g,
Z
Z
Z
2g
X
k
nj
k = k +
Aj
j=1
with nj Z. It follows that the two realizations of (p) only differ by an element of
L(M ), as claimed. It is clear that d(p) = 0 iff all j vanish at p. But this is impossible
by Corollary 8.15 below. Hence d has rank one as claimed.

This result of course raises a number of questions, such as: for g = 1, is M J(M )
in the sense of conformal isomorphisms?, or is an imbedding for g 2?. We will return
to these questions later in the Section on Abels and Jacobis theorems.
For now we return to the investigation of bilinear relations. The name bilinear
relation in Riemann surface theory refers to any relation that originates by applying
e Next, we wish to obtain bilinear relations for
Lemma 8.3 to a specific choice of , .
meromorphic differentials, and not just holomorphic ones. In order to do so, we first
re-prove this lemma in a somewhat more intuitive fashion via Stokes theorem.
Second proof of Lemma 8.3. We use the fundamental polygon of the Riemann
1
surface M , which is the polygon F bounded by the curves a1 , b1 , a1
1 , b1 , a2 , . . . and
with appropriate identifications on the boundary. Since F is simply connected, = df
on F. With some z0 F,
Z z
f (z) =
z0
Note that f does not necessarily agree at identified points. By Stokes,

Z
Z
Z
Z
e
e
e
f e
d(f ) =
df =
=
(8.4)
g Z
X
j=1
aj
f e +
bj
f e +
a1
j
f e +
b1
j

f e
To proceed, let z, z F be identified points on aj , a1

j , respectively. Then
135
bj
c
aj
aj
~
c
1
bj
Figure 8.1. How the bilinear relations arise

Z
aj
f e +
a1
j
f e =
Z Z
z0
aj
Z Z
aj
bj
z0

e
Z
Z

e
e
aj
z .
In Figure 8.1 the curve c depicts the curve joining z to

the bj , b1
j integrals. Plugging this into (8.4) yields (8.3).
bj
A similar formula holds for

The importance of this method of proof lies with the fact that it applies to the case
when e is a meromorphic differential as well. In that case we also pick up residues when
applying Stokes theorem. Heres an important example, which uses our L2 existence theory. In what follows, it will be understood automatically that the a, bloops representing
a homology basis do not pass through any pole of a meromorphic form. In particular,
we regard them as fixed loops now rather than as homology classes.
Lemma 8.9. Fix some p M and a parametric disk (U, z) at p with z(p) = 0, and
(n)
let n 2. Denote by p the unique meromorphic differential, holomorphic on M \ {p},
with singularity1 zdzn locally at p, and with vanishing a-periods. Then
Z
2i
p(n) =
(8.5)
,n2 1 g
n1
b
where ,k denotes the Taylor coefficients of locally at p, i.e.,
X

(z) =
,k z k dz
k=0
(n)
in the same local coordinates z in which p
has singularity
(n)
dz
zn .
Proof. To start, note that p = + where is as in Corollary 7.11 and

(n)
is holomorphic and chosen in such a way that the a-periods of p vanish. To prove
1This means that (n)
p
dz
zn
is holomorphic around p.
136
(8.5), pick a small positively oriented loop around p and let F denote the fundamental
polygon F with the disk bounded by deleted. Let = df on F and thus also on F .
Then, by the second proof of Lemma 8.3 presented above,
Z
Z
Z
Z
g Z
Z
X
(n)
(n)
(n)
f p(n)
p
p =
(8.6)
0=
F
aj
bj
bj
aj
j=1
The vanishing of the left-hand side is the fact that e = 0 for holomorphic one-forms.
In local coordinates,
Z
Z
X
,k k+1 dz
2i
f p(n) =
z
,n2
=
n
k+1
z
n1
|z|=
k=0
whereas
g Z
X
j=1
aj
bj
p(n)
bj
aj
p(n)
In view of (8.6) we are done.
p(n)
This lemma is needed in the proof of the Riemann-Roch theorem to which we now
turn.
2. Divisors
The following terminology is standard in the field:
Definition 8.10. A divisor D on M is a finite formal sum D =
p M are distinct and s Z. The degree of D is the integer
X
deg(D) =
s
s p
where
If s 0 P
for all then D is called integral. We write D D for two divisors iff
D D = s p is integral. If f is a nonconstant meromorphic function on M , then

we define the divisor of f as
X
(f ) =
ord(f ; p ) p
where the sum runs over the zeros and poles of f with the sign being chosen depending
on whether p is a zero or pole, respectively. If f = const (but neither 0 nor ), then
(f ) = 0. In the same way, we define the divisor of a non-zero meromorphic differential:
X
() =
ord(; p ) p
where the sign is chosen again via the zero/pole dichotomy. Given a divisor D, we define
the C-linear space
L(D) = {f M(M ) | (f ) D or f = 0}
where M(M ) are the meromorphic functions. Analogously, we define the space
(D) = { M1 (M ) | () D or = 0}
where M1 (M ) are the meromorphic differentials.
We collect some simple observations about these notions (all dimensions here are
over C):
3. THE PROOF OF THE RIEMANN-ROCH THEOREM
137
Lemma 8.11. In what follows, Div(M ) denotes the additive free group of divisors
on M , a compact Riemann surface of genus g.
(1) deg : Div(M ) Z is a group homomorphism.
(2) The map f 7 (f ) is a homomorphism from the multiplicative group M(M ) of
the field M(M ) (which excludes f ) of meromorphic functions to Div(M ).
The image under this map is called the subgroup of principal divisors and
the quotient Div(M )/(M(M ) ) is called divisor class group and the conjugacy
classes are called divisor classes. The homomorphism deg factors through to the
divisor class group.
(3) The divisors of non-zero meromorphic differentials always belong to the same
divisor class (called the canonical class K).
(4) If D D , then L(D) L(D ).
(5) L(0) = C and L(D) = {0} if D > 0.
(6) If deg(D) > 0, then L(D) = {0}.
(7) dim L(D) and dim (D) only depend on the divisor class of D. Moreover,
dim (D) = dim L(D K) where K is the canonical class.
(8) (0) = H1 Cg
Proof. 1) is obvious. For 2), note that (f g) = (f ) + (g) and deg(f ) = 0 for any
f, g M(M ) . For 3), observe that for any non-zero 1 , 2 M1 the quotient f = 21
is a non-zero meromorphic function. Since (1 ) (2 ) = (f ), the statement follows. 4)
is clear. For 5), note that f L(D) with D 0 implies that f is holomorphic and thus
constant. For 6), observe that (f ) D implies that 0 = deg((f )) deg(D). For 7),
suppose that D = D + (h) where h is nonconstant meromorphic. Then f 7 f h takes
L(D ) Clinearly isomorphically onto L(D). In particular, dim L(D) = dim L(D ). The
map 7 takes (D) isomorphically onto L(D K) where K = () is the canonical
class, whence the dimension statement. Finally,
dim (D) = dim L(D K) = dim L(D K) = dim (D )
For 8), simply note that (0) consists of all holomorphic differentials.
3. The proof of the Riemann-Roch theorem

We now state the main result of this chapter for integral divisors.
Theorem 8.12 (Riemann-Roch). Let D be an integral divisor. Then
(8.7)
dim L(D) = deg(D) g + 1 + dim (D)
= deg(D) g + 1 + dim L(D K)
Proof. By Lemma 8.11,

P(8.7) holds for D = 0. Hence, we can assume that deg(D) >
0. Thus, assume that D = n=1 s p with s > 0. To expose the ideas with a minimum
of technicalities, we let s = 1 with p distinct for all 1 n. Let us also first assume
that g 1.
S
If (f ) D, then df M1 is holomorphic on M \ {p } with ord(df, p ) 2;
clearly, df has zero periods and residues. Conversely, if M1 has all these properties,
then
Z q
f (q) =
p
138
is well-defined where p M is fixed and the integration is along an arbitrary curve

avoiding the p . It satisfies df = and (f ) D. Hence,
dim L(D) = dim V + 1
n
V := M1 | has vanishing periods and residues,
o
[
is holomorphic on M \ {p }, and ord(, p ) 2
To compute dim V , we define for any t := (t1 , . . . , tn )

t :=
n
X
t p(2)
n=1
(2)
where p is as in Lemma 8.9. By construction, t has vanishing a-periods and vanishing

residues. Second, we define the map as
nZ
og
t
: t 7
b
=1
Every V satisfies = t for some unique t but not every t V ; in fact, V = ker
(2)
under this identification since the a-periods of p vanish by construction. With { }g=1
the basis from above,
Z
n
X
t = 2i
t ,0 (p )
b
=1
see (8.5), where
(z) =
hX
j=0
i
,j (p ) z j dz
locally around p . Thus, is defined by the matrix
1,0 (p1 ) . . . 1,0 (pn )
...
2i
g,0 (p1 ) . . . g,0 (pn )
The number of linear relations between the rows of this matrix equals
dim{ H1 | (p ) = 0 1 n}
which in turn equals dim L(D K). For the latter equality, fix any non-zero M1 .
Then f L(D K) iff
(f ) D () (f ) D
iff = f H1 (M ) with (p ) = 0 for all . In summary,
dim L(D) = dim V + 1 = dim ker + 1
= n rank + 1 = n (g dim L(D K)) + 1
= deg(D) g + 1 + dim L(D K)
as claimed. Finally, if g = 0 then periods do not arise and for integral D with deg(D) > 0
one simply has dim V = n = deg(D) and dim L(DK) = 0 so that dim L(D) = n+1 =
deg(D) + 1 as desired.
4. APPLICATIONS AND GENERAL DIVISORS
139
The case of integral D which is not the sum of distinct points, the proof is only
notationally
more complicated. We again consider the case g 1 first. Then, with
P
D = s p and n = deg(D), consider
t :=
+1
X sX
t,k p(k)
,
k=2
t = {t,k }2ks +1
Every in the linear space

n
V := M1 | has vanishing periods and residues,
o
[
is holomorphic on M \ {p }, and ord(, p ) s 1
Cn .
satisfies = t for some t

As before, we have dim L(D) = dim V + 1. With
as above, t V iff (t ) = 0 so that dim V = dim ker . From Lemma 8.9 we compute
the b-periods as
Z
X X
,k2 (p )
t = 2i
t,k
k1
b
2ks +1
For the purposes of computing dimensions, we may evidently replace t,k with
the same argument as in the case s = 1 one now concludes that
t,k
k1 .
By
dim ker = n (g dim(D K))

and (8.7) follows. Finally, the case g = 0 was already treated in Problem 7.4 in fact,
our argument for the g 1 case only requires minimal changes to pass to g = 0 which
we leave to the reader.

4. Applications and general divisors
As an immediate corollary, we obtain the following remarkable result which we already obtained before, see Corollary 7.12. Here we phrase the proof in terms of the
Riemann-Roch theorem, whereas in the previous chapter we essentially gave the proof
of the Riemann-Roch theorem in this special case (see also Problem 7.4).
Corollary 8.13. Let M be compact and simply connected. Then M CP 1 in the
sense of conformal isomorphism.
Proof. Let D = p0 where p0 M is arbitrary. Since g = 0, Theorem 8.12 implies
that
dim L(D) 2
Thus, there exists a nonconstant f M(M ) with a simple pole at p0 and no other poles.
Hence deg(f ) = 1 and f is an isomorphism between M and CP 1 .

The generalization of this proof method to higher genus yields the following statement
concerning branched covers of compact Riemann surfaces.
Corollary 8.14. Let M be compact with genus g 1. Then M can be represented
as a branched cover of CP 1 with at most g + 1 sheets. In other words, there exists a
nonconstant meromorphic function on M of degree at most g + 1.
140
Proof. Take D = (g + 1)p0 . Then dim L(D) 2 by Riemann-Roch. Hence,

there exists f : M CP 1 meromorphic, f holomorphic everywhere on M \ {p0 } with
ord(f, p0 ) g 1. Evidently, such an f defines the branched covering.

This shows in particular that every compact Riemann surface of genus one admits
a two-sheeted branched cover. This is the missing piece in the proof of Corollary 5.20
which states that every compact Riemann surface of genus one is an elliptic curve. As
the example of hyper-elliptic curves shows, there exist compact surfaces of arbitrarily
large genus which can still be written as two-sheeted branched covers.
Let us give one last illustration of this proof method; to motivate the result, recall
that on a simply connected surface M any holomorphic differential vanishes identically.
If g 1, we now observe completely different behavior: there is no point p0 M at
which all holomorphic differentials vanish.
Corollary 8.15. Let M be compact with g 1. Then there is no point in M at
which all H1 vanish.
Proof. Suppose p0 M were such a point and set D = p0 . Then
H1 (M ) = (0) = (D) = L(K) = L(D K)
By Riemann-Roch therefore
dim L(D) = 1 g + 1 + dim L(K) = 2 g + g = 2
As in the proof of Corollary 8.13 we could now conclude that M is conformally equivalent
to CP 1 , which cannot be. For example, this would contradict the Riemann-Hurwitz
formula.

The significance of the previous result has to do with projective imbeddings. Indeed,
for M as in the corollary, let {j }gj=1 be a basis of H1 (M ) and define
M : M CP g , p 7 [f1 (p) : . . . : fg (p)]
where j = fj dz in local coordinates. This is clearly well-defined as changes of coordinates only multiply the entries by a nonzero factor; moreover, at least one fj does
not vanish. Thus, is well-defined and, in fact, analytic. In Problem 8.6 the reader
will study the question as to when this map M is in fact an imbedding. Again, the
Riemann-Roch theorem is the crucial tool.
Finally, we need to prove the Riemann-Roch theorem for arbitrary divisors, and not
just integral ones. The proof of this extension requires the following intermediate step
which is of course an interesting result in its own right.
Lemma 8.16. The degree of the canonical class is given by
deg(K) = 2g 2 = (M )
Proof. If g = 0, then take M = C and = dz in the chart z C. Under the
. Hence, deg(K) = 2. If g 1,
change of variables z 7 1z , this transforms into = dz
z2
1
pick any nonzero H (M ) (which can be done since this space has dimension g).
Then () = K is integral and by Theorem 8.12,
dim L(K) = deg(K) g + 1 + dim L(0) = deg(K) g + 2
whereas L(K) = H1 (M ). Hence dim L(K) = dim H1 (M ) = g, and deg(K) =

2g 2 as claimed.
4. APPLICATIONS AND GENERAL DIVISORS
141
An alternative proof based on the Riemann-Hurwitz formula is as follows: By Theorem 8.12 there exists a meromorphic function f with n simple poles for some integer
n 2 and holomorphic elsewhere. In particular, deg(f ) = n. Take = df . Suppose
that p M is a branch point of f . Then p is not a pole of f and
ord(; p) = bf (p)
where bf (p) is the branch number of f at p. If p is a pole of f , then ord(; p) = 2 so
that
X
deg(()) = 2n +
bf (p)
pM
By the Riemann-Hurwitz formula,
2(g 1) = 2n +
bf (p)
pM
and we are done.
Combining this with Theorem 8.12 now yields the full Riemann-Roch theorem.
Theorem 8.17. Equation (8.7) holds for all divisors D.
Proof. We already covered the case where D is equivalent to an integral divisor.
Suppose D is such that K D is equivalent to an integral divisor. Then, from Theorem 8.12,
dim L(D K) = deg(K D) g + 1 + dim L(D) = deg(D) + g 1 + dim L(D)
which is the desired statement. Suppose therefore that neither D nor KD are equivalent
to an integral divisor. Then
dim L(D) = dim L(D K) = 0
It remains to be shown that deg(D) = g 1. For this we write D = D1 D2 where D1
and D2 are integral and have no point in common. Clearly, deg(D) = deg(D1 ) deg(D2 )
with both degrees on the right-hand side positive. By Theorem 8.12,
dim L(D1 ) deg(D1 ) g + 1 = deg(D) + deg(D2 ) g + 1
If deg(D) g, then dim L(D1 ) deg(D2 ) + 1 and there exists a function f L(D1 )
which vanishes at all points of D2 to the order prescribed by D2 . Indeed, this vanishing
condition imposes deg(D2 ) linear constraints which leaves us with one dimension in
L(D1 ) (for example, if deg(D2 ) = 1 then we use the constant function to make any
nonconstant meromorphic f with (f ) D1 vanish at the point given by D2 ). For this
f,
(f ) + D D1 + D2 + D = 0
so that f L(D) contrary to our assumption. This shows that deg(D) g 1.

Similarly,
deg(K D) = 2g 2 deg(D) g 1 = deg(D) g 1
and we are done.
142
5. The theorems of Abel and Jacobi

We shall now study the Jacobian variety J(M ) = Cg /L(M ) in more detail. Let us
first recall a result for the tori M = C/h1 , 2 i which we proved in Chapter 4 in the
context of elliptic functions. That is, recall Theorem 4.17 which establishes the existence
of a meromorphic function on M with zeros zj and poles k iff conditions (4.10) and (4.11)
hold. Using the language of divisors, this is in agreement with the following more general
result, known as Abels theorem. As above, M is a compact Riemann surface and we let
{j }gj=1 be an arbitrary basis of H1 (M ), g the genus of M .
Proposition 8.18. Let D be a divisor on the compact Riemann surface M . There
exists a meromorphic function f on M with (f ) = D iff the following two conditions
hold:
deg(D) = 0P
with2 D = n=1 (p q ) one has
Z
n Z q
X
1 , . . . ,
(8.8)
(D) :=
=1
q
p

g 0
mod L(M ))
Proof. If g = 0, then only deg(D) = 0 is relevant. Thus, we are reduced to the fact
that a meromorphic function exists on S 2 with poles and zeros at prescribed locations
and with prescribed orders as long as the combined order of the zeros is the same as the
combined order of the poles.
Let us therefore assume that g 1. Clearly, deg(D) = 0 is necessary. For (8.8),
consider the map
: [1 : 2 ] 7 ((1 f + 2 ))
e : CP 1
from CP 1 J(M ). Clearly, is continuous and lifts to a continuous map
g
C which is, moreover, holomorphic (i.e., each component is). We conclude from the
maximum principle that each component is constant whence
0 = ([0 : 1]) = ([1 : 0]) = (D)
as claimed. For an alternative proof of the necessity of (8.8).
For the sufficiency, we use Theorem 7.13.
6. Problems
Problem 8.1. Determine the divisor class group for S 2 .
Problem 8.2. Show that every compact surface M of genus two is a hyper-elliptic
surface, i.e., it carries a meromorphic function of degree two.
Problem 8.3. Given an example of a compact Riemann surface of genus g 3 which
is not hyper-elliptic, in other words, it cannot be written as a two-sheeted branched cover
of S 2 .
2The p do not need to be pairwise distinct, and the same holds for the q . However, p 6= q for
any 1 , n.
6. PROBLEMS
143
Problem 8.4. Let M be compact of genus g 1 and let {j }gj=1 be a basis of

H1 (M ). Prove that
g
X
2
|j (p)|2 on Tp M
ds (p) :=
j=1
with p M arbitrary defines a positive definite metric on M . Show that it has nonpositive
curvature. Discuss the possible vanishing of the curvature.
Problem 8.5. In this problem, you are asked to improve on Proposition 5.16 in
the following way: Given z of degree n as in that proposition, prove that there exists a
meromorphic function f on M which renders the polynomial in (5.6) irreducible. Note
that this concludes the proof of Theorem 5.19.
Problem 8.6. Here we discuss the question when the canonical map M : M CP g
defined above is in fact an imbedding.
CHAPTER 9
The Dirichlet problem and Green functions

1. Green functions
In the notes on Hodge theory we encountered the fundamental problem of solving
the so called Poisson equation
(9.1)
u = f
2
(R2 ). Such a u is not unique (add any linear function). However, we
when f Ccomp
singled out the solution
Z
1
log |z | f () dd
(9.2)
u(z) =
2 R2
where = + i. The reader will easily verify that it is the unique solution of (9.1)
with the property that u(z) = k log |z| + o(1) as |z| for some constant k. In fact,
necessarily k = hf i.
1
The function (z, ) = 2
log |z | is of great importance. It is called the fundamental
solution of which means that G(, ) = in the sense of distributions. We are now
led to ask how to solve (9.1) on a bounded domain C (for example on = D).
To obtain uniqueness from the maximum principle we impose a Dirichlet boundary
condition u = 0 on . By a solution of
(9.3)
u = f in ,
u = 0 on
C 2 ()C()
with f C() we mean a function u

sense. To solve (9.3), we try to emulate (9.2).
which satisfies (9.3) in the pointwise
Definition 9.1. We say that C admits a Green function if there exists G with
the following properties:
G(, ) C( \ {z = })
The function h(z, ) := G(z, ) (z, ) is harmonic on in the first variable
for all , and jointly continuous on
G(z, ) = 0 for all (z, )
It is important to note that this definition applies to unbounded C, but in that

case we view C so that . In particular, we require vanishing at infinity
in that case. It is clear that if G exists, then it is unique. Also, G < 0 in by the
maximum principle.
Lemma 9.2. If admits a Green function, then (9.3) has a unique solution for every1
2
() given by
f Ccomp
Z
G(z, )f () d
z
u(z) =
1The assumptions on f in the previous lemma can be relaxed, but this does not concern us here.
145
146
9. DIRICHLET AND GREEN
Proof. Uniqueness follows from the maximum principle. By the continuity assumptions on G, u is continuous on and satisfies u = 0 on . Moreover, with = + i,
we can write
Z
Z
(z, )f () dd
u(z) = [G(z, ) (z, )]f () dd +
=: u1 (z) + u2 (z)
The second integral on the right-hand side, which we denoted by u2 , satisfies u2 = f .

2
Indeed, setting f = 0 outside of leads to f Ccomp
(R2 ) and we can apply the discussion
centered around (9.2). As for u1 , we have
Z
h(, )f () dd C()
u1 () =
by continuity of h. Moreover, the mean value property holds:

Z 1
u1 (z0 + re2i ) d = u1 (z0 )
0
for all z0 and small r > 0. This follows from Fubinis theorem since h(, ) is
jointly continuous and satisfies the mean value property in the first variable. Thus, u1
is harmonic and u = f as desired.

So which C admit a Green function? For example, take = D. Then G(z, 0) =
log |z| does the trick for = 0. Next, we map to 0 by the Mobius transformation
z
T (z) = 1z
. This yields
z
1

log
GD (z, ) =

2
1 z
1
2
as our Green function for D. It clearly satisfies Definition 9.1. Moreover, by inspection,
GD (z, ) = GD (, z) z, D
Now let be the disk with n 1 points removed, i.e., = D \{z1 , . . . , zn }. If G were
a Green function on , then for all , z 7 G(z, ) would need to be continuous in
a neighborhood of zj for all 1 j n and harmonic away from zj . Then each zj would
constitute a removable singularity and G(z, ) (see Problem 3.3) therefore be harmonic
in a disk around each zj . In other words, G would be the Green function of D and
therefore negative at each zj violating the vanishing condition. In conclusion, does not
admit a Green function in the sense of Definition 9.1.
Finally, any simply connected C for which the Riemann mapping f : D
extends continuously to admits a Green function (for this it suffices to assume that
consists of finitely many C 1 arcs). Indeed, observe that
G (z, ) := GD (f (z), f ())
satisfies Definition 9.1. This procedure applies to unbounded , for example = H. It
automatically enforces the vanishing condition at infinity required by the fact that we
view C .
2. THE POTENTIAL THEORY PROOF OF THE RIEMANN MAPPING THEOREM
147
2. The potential theory proof of the Riemann mapping theorem

This discussion raises the following interesting question: Is it possible to construct
the Riemann mapping D from the Green function on (assuming it exists)? The
answer, which was found by Riemann, is yes.
Theorem 9.3. Suppose C is simply connected and admits a Green function.
Then G gives rise to a bi-holomorphic mapping f : D.
Proof. The idea is simply to write, with fixed,
2G(z, ) = log |z | + Re F (z)
where F H() (this can be done since is simply connected). Then we set
f (z) := (z ) exp(F (z)) H()
Note that f is unique up to a unimodular number. By construction,
|f (z)| = exp(log |z | + Re F (z)) = exp(2G(z, )) < 1
for all z so that f : D; furthermore, |f (z)| = 1 for all z and |f | extends
as a continuous mapping to all of .
We claim that f () = D. By analyticity and since f is obviously not constant, f ()
is open. To show that it is closed, suppose that f (zn ) w D. Then zn z (if
needed, pass to a subsequence of zn , which we call zn again; recall that we are viewing
C which is compact). If z , necessarily |w| = 1 which is a contradiction. So
z and f (zn ) f (z ) = w which shows that f () is closed.
It remains to show that f is one-to-one. Locally around this is clear (why?), but
not globally on . We also remark that f (z) = 0 iff z = 0. In view of this, the logic is
now as follows: suppose f is one-to-one for any . Then T := f f1 Aut(D)
is a Mobius transformation which takes f () to 0. This suggests we prove the following
converse for arbitrary \ {}.
Claim: Let f () = w and T (w) = 0, T Aut(D). Then |T f | = |f |.
If the claim holds, then we are done: assume that f () = f (e
) = w D and let
T (w) = 0 were T Aut(D). Then
|T f | = |f | = |fe|
so that fe() = 0 implies = e as desired. To prove the claim, note that for any
0 < < 1, and some integer k 1,
log |T f (z)| = k log |z | + O(1) log |z | + O(1)
2(1 )G(z, )
as z . Moreover,
lim sup log |T f (z)| 0,
z
G(z, ) 0 as z
Hence, on \ D(, ) for all > 0 small, we see that the harmonic function 2G(, )
dominates the subharmonic function log |T f ()| on \ {}. In conclusion,
log |T f ()| 2G(, )
148
In particular, since T (z) =

(9.4)
zw
1z w
and thus T (0) = w,
2G(, ) = log |f ()| = log |w| = log |T (0)|

= log |T f ()| 2G(, )
whence G(, ) G(, ) which implies

(9.5)
G(, ) = G(, )
This is the well-known symmetry property of the Green function. It follows that we have
equality in (9.4)
log |T f ()| = 2G(, )
from which we conclude via the strong maximum principle on \ {} that

log |T f ()| = 2G(, ) = log |f ()|
as claimed. This finishes the proof.
The importance of this argument lies with the fact that it extends from domains
C to simply connected Riemann surfaces M , at least to those that admit a Green
function see the following chapter for the exact definition of this concept on Riemann
surfaces (we caution the reader that a Green function G on a Riemann surfaces will not
necessarily conform to Definition 9.1 above in case M C).
3. Existence of Green functions via Perrons method
Let us now consider the important problem of finding the Green function on bounded
domains C. Fix and solve if possible the Dirichlet problem
(9.6)
u(z) = 0 in ,
u(z) = log |z | on
Then G(z, ) := u(z) + log |z | is the Green function. This was Riemanns original
approach, but he assumed that (9.6) always has a solution via the so-called Dirichlet
principle. In modern terms this refers to the fact that the variational problem, with
f C 1 () and being C 2 regular,
Z
|u|2 dxdy
inf
uA
A := {u H 1 () | u f H01 ()}
has a (unique) minimizer u0 A (minimizer here means that u0 attains the infimum).
Here
H 1 () = {u L2 () | u L2 ()}
is the Sobolev space where u is the distributional derivative, and H01 () H 1 () is

the subspace of vanishing trace. It a standard fact, see Evanss book, that Dirichlets
principle holds and that the minimizer u0 is a harmonic function so that u0 f C()
vanishes on as desired. Riemann did not have this Hilbert spaces machinery at his
disposal, however.
Instead, we will use an elegant method due to Perron based on subharmonic functions.
It requires less on the boundary than the variational approach. We first need to lift the
concept of subharmonic functions to a general Riemann surface M .
3. EXISTENCE OF GREEN FUNCTIONS VIA PERRONS METHOD
149
Definition 9.4. A function u : M [, ) is subharmonic iff it is continuous

and subharmonic in every chart. We denote the class of subharmonic functions on the
Riemann surface M by sh(M ).
Since subharmonicity is preserved under conformal transformations, this definition
is meaningful. From our definition it is clear that subharmonicity is a local property.
Hence, properties that can be checked in charts immediately lift from the planar case to
Riemann surfaces. Here are two examples:
If u C 2 (M ) then u is subharmonic iff u 0 in every chart on M .
P
If u1 , . . . , uk sh(M ), then max(u1 , . . . , uk ) sh(M ) and kj=1 aj uj sh(M )
for any aj 0.
The following lemma collects several global properties of this class which mirror those
in the planar case. We begin with the maximum principle.
Lemma 9.5. The following properties hold for subharmonic functions:
(1) If u sh(M ) attains its supremum on M , then u = const.
(2) Let h be harmonic on M and u sh(M ). If u h on M then either u < h or
u = h everywhere on M .
(3) Let M be a domain with compact closure in M . Suppose h is harmonic on
and continuous on . If
(9.7)
lim sup u(p) h(q)

pq
p
then u h in . Equality here can only be attained in if u = h throughout .

Proof. For 1), assume that u u(p0 ) with p0 M . Then
E = {p M | u(p) = u(p0 )}
is both open (since that is a local property and follows by considering charts) and closed.
Hence E = M , as desired.
For 2), apply 1) to u h sh(M ).
For 3), it suffices to consider the case h = 0 (otherwise consider u h). If u(p) > 0
for any p , then u attains its supremum on and is therefore constant. But this
would contradict (9.7). Hence, u 0 on with equality being attained at one point
forcing constancy by 2).

We remark that property 2) above characterizes sh(M ) and gives a nice way of
defining subharmonic functions intrinsically on Riemann surfaces. For future purposes
we denote property (9.7) by u h on . We are now going to describe Perrons method
for solving the following Dirichlet problem on a Riemann surface M :
Let M be a domain with compact and suppose : R is continuous. Find
u C() so that u is harmonic on and u = on .
The first step towards the solution is furnished by showing that the upper envelope of all
subharmonic functions v on with v is harmonic. The second step then addresses
how to ensure that the boundary data are attained continuously. For the first step we
need to following easy result.
150
Lemma 9.6. Let D be a parametric disk and suppose f sh(M ) is real-valued. Let
h be the harmonic function on D which has f as boundary values on D. The function

f on M \ D
fD :=
h on D
satisfies fD sh(M ) and fD f .
Proof. It is clear that fD is continuous. By the maximum principle, fD f . It is

clear that fD is subharmonic on M \ D. However, if p D, then we see from fD f
that the sub-mean value property holds at p for sufficiently small circles (relative to some
chart at p). Therefore, fD sh(M ) as claimed.

Now for the first step in Perrons method.
Proposition 9.7. Let be any bounded function on . Then
(9.8)
is harmonic on .
u = sup{v | v sh(), v on }
Proof. Denote the set on the right-hand side of (9.8) by S . First note that any
v S satisfies
sup v sup <
Moreover, replacing any v S by max(v, inf ), we can assume that all v S are
bounded below. Take any p and a sequence of {vn }
n=1 S so that vn (p) u(p).
Replacing the sequence by
v1 , max(v1 , v2 ), max(v1 , v2 , v3 ), . . .
{vn }
n=1
we may assume that

is non-decreasing. In addition, by Lemma 9.6 each vn can
be assumed to be harmonic on on some parametric disk D centered at p. By Harnacks
inequality on D, vn v uniformly on compact subsets of D with v harmonic on D.
It remains to check that u = v on D. Take any q D and let {wn }
n=1 S with
wn (q) u(q). As before, we can assume that wn is harmonic and increasing on D. In
fact, we can also assume that wn vn for each n. We conclude that wn w uniformly
on compact sets with w harmonic on D and w (p) = v (p). Since w v in D, it
follows that w = v on D. In particular, u(q) = v (q) and we are done.

It is worth noting that this proof has little to do with . In fact, it applies to any
Perron family which we now define.
Definition 9.8. A family F of real-valued subharmonic functions on a Riemann
surface M is called Perron family iff
for any f, g F there is h F with h max{f, g}
for any parametric disk D M and any v F there exists w F with w
harmonic on D and w v
Then we have the following immediate corollary of the proof of Proposition 9.7:
Lemma 9.9. For any Perron family F on a Riemann surface M the function
u := sup v
vF
is either or harmonic on M .
4. BEHAVIOR AT THE BOUNDARY
151
4. Behavior at the boundary

Next, we turn to the boundary behavior. For a standard example of what can go
wrong on the boundary consider = D \ {0}. Setting (0) = 0 and (z) = 1 for |z| = 1
we see that Perrons method yields u = const = 1; indeed, for any > 0 the function
v(z) = 1 + log |z| S .
Definition 9.10. A barrier at a point p is defined to be a function with the
following properties:
sh()
C() and 0 on with > 0 on \ {p}, and (p) = 0
Any point q which admits a barrier is called regular and is called regular iff all
of its points are regular.
It turns out that regularity of a boundary point is a mild condition:
Lemma 9.11. Suppose p satisfies an exterior disk condition, i.e., there exists a
disk D(z0 , ) in local coordinates (U, z) around p so that
z(U ) D(z0 , ) = {p}
Then p is regular. In particular, any C 2 boundary is regular. Moreover, suppose that

p is accessible, i.e.,
z( U ) D \ (1, 0]
in some chart (U, z) with z(p) = 0. Then p is regular.
Proof. For the exterior disk condition, we define for all q ,

log(|z(q) z0 |/) if |z(q) z0 |
(q) =
log(/)
if |z(q) z0 | >
where > > 0 are sufficiently small. If p is accessible, then we map D \ (1, 0]
conformally onto a sector of angle which guarantees the exterior disk condition
at p.

An obvious example of a non-accessible boundary point is p = 0 for = D \ {0}.
The importance of barriers lies with the following fact:
Proposition 9.12. Suppose p is regular and a bounded function on which
is continuous at p. Then the function u from Proposition 9.7 satisfies
lim u(q) = (p)
qp
q
In particular, if is regular and : R is continuous, then u is a solution of

Dirichlets problem on with boundary data .
Proof. Let S be as in Proposition 9.7. Recall that
(9.9)
inf v sup
for any v S . We now claim the following: given > 0 there exists C = C() such that
(9.10)
v(q) C(q) (p) + q
152
for any v S . To prove this, we let D be a small parametric disk centered at p. It can
be chosen so that
sup v C (p) +
(D)
due to the continuity of , the positivity

min > 0,
\D
and provided C is large enough. The maximum principle now shows that (9.10) holds
on D. On \ D, we let C be so large that (9.10) holds due to (9.9).
In conclusion,
lim sup u(q) (p) +
qp
For the lower bound, we observe by the same arguments that

Hence,
so that
C + (p) S
u C + (p)
lim inf u(q) (p)
qp
as desired.
We remark that the regularity of is also necessary for the solvability of the
Dirichlet problem for general continuous boundary data; indeed, the boundary data
f (p) = |p p0 | yields a barrier.
Let us make another remark concerning solving the Dirichlet problem outside some
compact set K M . As the example K = D 2D shows, we cannot expect unique
solvability of the Dirichlet problem with data on K. However, the Perron method always
yields existence of bounded harmonic functions. The following result is a corollary of the
proof of the preceding proposition.
Corollary 9.13. Let K M be compact and K regular. Then for any : K
R continuous and any constant A maxK there exists a harmonic function u on
:= M \ K with u C(), u = on and
min u A
on .
Proof. Define
(9.11)
u := sup{v | v sh(), v on , v A}
The set on the right-hand side is a non-empty Perron family and u is harmonic on and
satisfies (9.11). Let 0 be a barrier at p . Let D, D be parametric disks centered
compact. Then for > 0 sufficiently small,
at p and D compactly contained in D and D
the function
:= min{0 , } on D
is superharmonic on D with the property that = on D \ D . This shows that we
can extend to all of by setting
= on \ D
4. BEHAVIOR AT THE BOUNDARY
153
The point is that we have constructed a barrier at p which is uniformly bounded away
from zero on \ D (this is another expression of the fact that being regular is a local
property around a point). Since u is bounded from above and below, the reader will
have no difficulty verifying that the exact same proof as in Proposition 9.12 applies in
this case.

In the following two chapters it will become clear that the solution constructed in
Corollary 9.13 is unique iff M does not admit a negative nonconstant subharmonic function (or in the terminology of the following chapter, if M is not hyperbolic). An example
would be M = C (the reader is invited to establish uniqueness in Corollary 9.13 in that
case). Note that this uniqueness is clear (as is the existence from Proposition 9.12) if
M is compact. From the classification that we develop in the following two chapters it
will become clear that uniqueness in Corollary 9.13 with M not compact holds iff M is
conformally equivalent to C while it does not hold iff M is conformally to D.
To summarize, we have solved the Dirichlet problem for all domains M with
compact closure and regular boundary. In particular, if M = C , any such domain
admits a Green function. Moreover, if C is simply connected, then G gives rise
to a biholomorphic f : D. This latter fact (the Riemann mapping theorem) we
proved earlier in a completely different way which did not require any information on
the boundary.
CHAPTER 10
Green functions and the classification problem

1. Green functions on Riemann surfaces
We would like to generalize the proof of the Riemann mapping theorem from the previous
chapter to any Riemann surface M which admits a Green function. But what is the
correct definition of a Green function G on M ? Since there is no boundary, at least in
the topological sense, we need to find a substitute for the the crucial vanishing condition
at the boundary. One option would be to require vanishing at infinity, i.e.,
inf
sup |G(p, q)| = 0
K compact pM \K
However, this turns out to be too restrictive. As an example, consider M = D \ {0}.

Infinity here is the set {0}{|z| = 1} but we cannot enforce vanishing at {0}. However,
the Green function on D is in a very precise sense also the Green function of D\{0}.
In fact, uniquely so, as we shall see. The issue here is that a single point is negligible
(more generally, sets of zero logarithmic capacity are negligible). While it is of course true
that this M is not simply connected, it would be unwise to introduce simple connectivity
into the concept of the Green function.
As often in analysis, the correct definition of a Green function on M imposes a
minimality condition on G. Following a time-honored tradition, we will consider positive
Green functions rather than negative ones. Of course, this just amounts to switching the
sign. In addition, we drop the factor of 2.
Definition 10.1. By a Green function with singularity at q M we mean a realvalued function G(p, q) defined on M \ {q} such that
G(p, q) + log |z| is harmonic locally around p = q where z are local coordinates
near q with z(q) = 0
p 7 G(p, q) is harmonic and positive on M \ {q}
if g(p, q) is any other function satisfying the previous two conditions, then
g(, q) G(, q) on M \ {q}.
It is evident that G is unique if it exists. Also, if f : N M is a conformal isomorphism, then it is clear that G(f (p), f (q)) is the Green function on N with singularity
at q. By the maximum principle, if G is a Green function as in the previous chapter,
then G satisfies Definition 10.1. We remark that no compact surface M admits such
a Green function (since G(, q) would then be a negative subharmonic function on M
and therefore constant by the maximum principle). Note that M = C does not admit a
Green function either:
Lemma 10.2. Suppose u < is a subharmonic function on C with some constant
< . Then u = const.
155
156
10. GREEN FUNCTIONS AND CLASSIFICATION
Proof. Let us first observe the following: suppose v is subharmonic and negative
on 0 < |z| < 2 and set v (z) := v(z) + log |z| where 0 < < 1. Then v is subharmonic
on 0 < |z| < 1. Moreover, v (z) = v(z) for all |z| = 1 and v (z) as z 0. It
follows from the maximum principle that v (z) max|z|=1 v(z) < 0 for all 0 < |z| < 1.
Now send 0 to conclude that v(z) max|z|=1 v(z) < 0 for all 0 < |z| 1.
To prove the lemma, we may assume that u < 0 everywhere and supC u = 0. Consider
u(1/z) on 0 < |z| < 2. It is subharmonic and negative and therefore by the preceding
paragraph
sup u(z) < 0.
|z|1
It follows that sup|z|1 u(z) = 0 which is impossible.
2. Hyperbolic Riemann surfaces admit Green functions

So which M do admit Green functions? As we saw, D and thus any domain in C conformally equivalent to it. Note that these surfaces obviously admit negative nonconstant
subharmonic functions. This suggests a classification:
Definition 10.3. A Riemann surface M is called hyperbolic iff it carries a nonconstant negative subharmonic function. If M is not hyperbolic and noncompact, then it is
called parabolic.
The logic here is as follows: using the exact same proof idea as in the Theorem 2
of the previous chapter we will show that the hyperbolic, simply connected surfaces are
conformally equivalent to D, whereas Riemann-Roch showed that the compact simply
connected ones are conformally equivalent to CP 1 . This leaves the simply connected
parabolic surfaces, and you have guessed it they are conformally equivalent to C.
In conclusion, every simply connected Riemann surface is conformally equivalent to
either D, C or CP 1 . In the non-simply connected case, one then passes to the universal
f, and proves that M is obtained
covering of M , which is again a Riemann surface M
f
from M by factoring by the covering group. All of these facts constitute the so-called
uniformization theorem. It is of course a famous and central result of the field.
Now for the hard work of constructing Green functions as in Definition 10.1. We will do
this via a Perron-type argument by setting
(10.1)
G(p, q) := sup v(p),

vGq
the supremum being taken over the family Gq that we now define.
Definition 10.4. Given any q M we define a family Gq of functions as follows:
any v in Gq is subharmonic on M \ {q}
v + log |z| is bounded above on U where (U, z) is some chart around q
v = 0 on M \ K for some compact K M
Since 0 Gq we have Gq 6= . Note that if G(p, q) is a Green function on some domain
C in the sense of the previous chapter, then
(G(p, q) )+ Gq
for any > 0. As another example, let M = C and q = 0. Then log (|z|/R) G0
for any R > 0. This shows that G(p, 0) as defined in (10.1) satisfies G(p, 0) = for all
2. HYPERBOLIC RIEMANN SURFACES ADMIT GREEN FUNCTIONS
157
p C. As we shall see shortly, this agrees with the fact that C does not admit a negative
nonconstant subharmonic function. In general, one has the following result.
Theorem 10.5. Let q M be fixed and let G(p, q) be defined as in (10.1). Then
either G(p, q) = for all p M or G(p, q) is the Green function of M with singularity
at q. Moreover,
inf G(p, q) = 0
pM
Proof. Observe that Gq is a Perron family. Hence, by the methods of the previous
chapter, either G(, q) = identically or it is harmonic.
Next, we need to check that p 7 G(p, q) + log |z(p)| is harmonic locally near p = q. In
fact, it suffices to check that
G(p, q) = log |z| + O(1) as p q
(10.2)
where z = z(p) since G(p, q) + log |z(p)| then has a removable singularity at p = q as a
harmonic function. If v Gq , then locally around q and for any > 0,
v(p) + (1 + ) log |z(p)|
is subharmonic and tends to as p q. Therefore, by the maximum principle, for

any p z 1 (D \ {0}),
v(p) + (1 + ) log |z(p)| sup v sup G(, q) =: k(q)
z 1 (D)
z 1 (D)
Hence, locally around q,

G(p, q) log |z(p)| + k(q)
For the reverse direction, simply note that
v(p) = log+ (1/|z(p)|) Gq
Let = inf pM G(p, q) 0. If v Gq , then outside some compact set K, and with > 0
arbitrary
v = 0 G(p, q)
whereas
(1 )v(p) G(p, q) as p q
By the maximum principle,
(1 )v G(, q) on M \ {q}
Letting 0 and by the definition of G, G(, q) G(, q) which implies that 0

and thus = 0 as claimed.
Finally, suppose that g(, q) satisfies the first two properties in Definition 10.1. Then
for any v Gq , and any 0 < < 1,
(1 )v g(, q)
by the maximum principle. It follows that G g as desired.
Next, we establish the connection between M being hyperbolic and M admitting a

Green function. This is subtle and introduces the important notion of harmonic measure.
Theorem 10.6. For any Riemann surface M the following are equivalent:
M is hyperbolic
the Green function G(, q) with singularity at q exists for some q M
158
the Green function G(, q) with singularity at q exists for each q M

Proof. We need to prove that a hyperbolic surface admits a Green function with
an arbitrary singularity. The ideas are as follows: we need to show that G(p, q) < if
p 6= q which amounts to finding a lid for our family Gq . In other words, we need to find
a function, say wq (p), harmonic or superharmonic on M \ {q} and positive there, and so
that wq (p) = log |z(p)| + O(1) as p q. Indeed, in that case we simply observe that
v w for every v Gq . Of course, G itself is such a choice if it exists so realistically we
can only hope to make wq superharmonic. So we need to find a subharmonic function
v1 (which would be wq ) which is bounded from above and has a log |z(p)| + O(1) type
singularity as p q. By assumption, there exists a negative subharmonic function
v0 on M . It, of course, does not fit the description of v1 since it does not necessarily
have the desired logarithmic singularity. So we shall need to glue log |z(p)| in a chart
around q to a subharmonic function like v0 which is bounded from above. However, it
is hard to glue subharmonic functions. Instead, we will produce a harmonic function
u that vanishes on the boundary of some parametric disk D and which is positive1 on
M \ D (by solving the Dirichlet problem outside of D). The crucial property of u is its
positivity on M \ D and this is exactly where we invoke the nonconstancy of v0 .
The details are as follows. Pick any q M and a chart (U, z) with z(q) = 0. We
may assume that D2 := z 1 (2D) and its closure are contained in U . Set D1 := z 1 (D).
1 ) with v 0 on D1 and such that
Consider the Perron family F of all v sh(M \ D
0 v 1 on M \ D1 . By Corollary 9.13, u := supvF v is continuous on M \ D1 with
u = 0 on D1 and 0 u 1. We claim that u 6 0. To this end, let v0 < 0 be a
nonconstant subharmonic function on M and set = maxD 1 v0 . Then < 0 and
v0
v0
=1+
F
1
||
Hence, || + v0 ||u. By non-constancy of v0 ,
max v0 >
2
D
1 . We shall now build

so that u > 0 somewhere and therefore u > 0 everywhere on M \ D
a subharmonic function v1 , globally defined on M and bounded above, and such that v1
behaves like log |z| around q. In fact, define
|z| 1
log |z|
max{log |z|, ku} 1 |z| 2
v1 :=
ku
z M \ D2
Here the constant k > 0 is chosen such that
ku > log 2 |z| = 2
Due to this property, and the fact that u = log |z| = 0 on |z| = 1, v1 C(M ). Moreover,
checking in charts reveals that v1 is a subharmonic function off the circle |z| = 1. Since
the sub-mean value property holds locally at every |z| = 1 we finally conclude that v1 is
subharmonic everywhere on M .
We are done: Indeed, any v Gq (see Definition 10.4) satisfies
v v1
11 u is called the harmonic measure of D relative to M \ D.
2. HYPERBOLIC RIEMANN SURFACES ADMIT GREEN FUNCTIONS
Hence, G(p, q) v1 (p) < for any p M \ {q}.
159
The previous proof shows that if some compact parametric disk admits a harmonic
measure, then M is hyperbolic. Let us now elucidate the important symmetry property
of the Green function. We already encountered it in the previous chapter as part of the
Riemann mapping theorem. However, it has nothing to do with simple connectivity as
we will now see.
We begin with the following simple observation.
Lemma 10.7. Let M be hyperbolic and N M be a sub-Riemann surface with piece compact. Then N is hyperbolic, GN G, and GN (p, q) =
wise C 2 boundary2 and N
GN (q, p) for all p, q N .
and with
Proof. Fix any q N and let uq be harmonic on N , continuous on N
boundary data G(, q). This can be done by the results of the previous chapter. Then
GN (p, q) := G(p, q) + uq (p)
is the Green function on N . It follows from the maximum principle that GN G. To

prove the symmetry property, fix p 6= q N and let N = N \ D1 D2 where D1 , D2 N
are parametric disks around p, q, respectively. Define
u = GN (, p),
v = GN (, q)
Then, by Greens formula on N ,

Z
Z
u dv v du =
0=
D1 D2
u dv v du
Again by Greens formula, but this time on D1 with local coordinates z, centered at p
(z(p) = 0),
Z
Z
(u + log |z|) dv v d(u + log |z|)
u dv v du =
D1
D1
Z
log |z| dv v d log |z|
D1
= GN (p, q)
and similarly
as desired.
D2
u dv v du = GN (q, p)
To obtain the symmetry of G itself we simply take the supremum over all N as in
the lemma. We will refer to those N as admissible.
Proposition 10.8. Let M be hyperbolic. Then the Green function is symmetric:
G(p, q) = G(q, p) for all q 6= p M .
Proof. Fix q M and consider the family
Fq = {GN (, q) | q N, N is admissible}
2This means that we can write the boundary as a finite union of C 2 curves : [0, 1] M .
160
where we extend each GN to be zero outside of N . This extension is subharmonic on

M \ {q} and Fq is a Perron family on M \ {q}:
max{GN1 (, q), GN2 (, q)} GN1 N2 (, q)
and GN D (, q) GN (, q) for any parametric disk D M \{q} with GN D (, q) harmonic

on D. Note that both N1 N2 and N D are admissible. Let
g(, q) := sup v G(, q)
vFq
Moreover, it is clear that

g(, q) sup v = G(, q)
vGq
Indeed, simply use that every compact K M is contained in an admissible N (take N to

be the union of a finite open cover by parametric disks). In conclusion, g(p, q) = G(p, q)
which implies that GN (q, p) = GN (p, q) G(p, q) for all admissible N . Hence, taking
suprema, G(q, p) G(p, q) and we are done.

3. Problems
Problem 10.1. Prove that C \ {zj }Jj=1 is not hyperbolic in the sense of Chapter 10.
Show that C \ (D {zj }Jj=1 ) is hyperbolic.
CHAPTER 11
The uniformization theorem

1. The statement for simply connected surfaces
We can now state and prove the following remarkable classification result.
Theorem 11.1. Every simply connected surface M is conformally equivalent to either
CP 1 , C, or D. These correspond exactly to the compact, parabolic, and hyperbolic cases,
respectively.
The compact case was already done via Riemann-Roch, whereas for the hyperbolic
case we will employ the Green function technique that lead to a proof of the Riemann
mapping theorem in Chapter 9.
2. Hyperbolic, simply connected, surfaces
Proof of Theorem 11.1 in the hyperbolic case. Let q M and G(p, q) be
the Green function with singularity at q. Then there exists fq holomorphic on M with
|fq (p)| = exp(G(p, q))
qM
with the understanding that fq (p) = 0. This follows from gluing technique of Lemma 5.5
since such a representation holds locally everywhere on M (alternatively, apply the monodromy theorem). Now fq : M D with fq (p) = 0 iff p = q. It remains to be shown
that fq is one-to-one since then fq (M ) is a simply connected subset of D and therefore,
by the Riemann mapping theorem, conformally equivalent to D.
We proceed as in the planar case, see Theorem 9.3. Thus, let p M with q 6= p and
T a Mobius transform with T (fq (p)) = 0. We claim that |T fq | = |fp |. This will then
show that fq is one-to-one (suppose fq (p) = fq (p ), then by the claim, |fp | = |fp | and
thus fp (p ) = 0 and p = p ).
To prove the claim, we observe that
wq := log |T fq | G(, q) on M \ {q}
Indeed, locally around p = q we have, for some integer k 1,
log |T fq (p)| k log |z(p) z(q)| + O(1) as p q
where z are any local coordinates around q. In addition, wq > 0 everywhere on M . From
these properties we conclude via the maximum principle that
wq v
v Gq = wq G(, q)
as desired (here Gq is the family from Definition 10.4). Since
G(p, q) = log |fq (p)| = log |T (0)| = log |(T fq )(q)| G(q, p) = G(p, q)
we obtain from the maximum principle that wq = G(, p) whence the claim.
161
162
11. UNIFORMIZATION
3. Parabolic, simply connected, surfaces

It remains for us to understand the parabolic case. The logic is as follows with M
simply connected: in the compact case we established the existence of a meromorphic
function with a simple pole. This followed from the Riemann-Roch theorem which in
turn was based (via a counting argument) on the existence of meromorphic differentials
singularity at a point.
with a prescribed dz
z2
In the hyperbolic case, we were able to place a positive harmonic function on M
with a log |z| singularity at a given point in fact, the hyperbolic M are precisely the
surfaces that allow for this. Amongst all such harmonic function we selected the minimal
one (the Green function) and constructed a conformal equivalence from it.
For the parabolic case we would like to mimic the compact case by constructing a
meromorphic function as before. In view of the fact that we are trying to show that
M is equivalent to C, and therefore compactifiable by the addition of one point this
is reasonable. Assuming therefore that f : M CP 1 is meromorphic and one-to-one,
note that f cannot be onto as otherwise M would have to be compact. Without loss
of generality, we can thus assume that f : M C. If f were not onto C, then by the
Riemann mapping theorem we could make f (M ) and thus M equivalent to D. But this
would mean that M is hyperbolic. So it remains to find a suitable meromorphic function
on M for which we need several more technical ingredients. The first is the maximum
principle outside a compact set which establishes uniqueness in Corollary 9.13.
Proposition 11.2. Let M be a parabolic Riemann surface and K M compact.
Suppose u is harmonic and bounded above on M \ K, and u 0 on K. Then u 0
on M \ K.
Proof. We have already encountered this idea in the proof of Theorem 10.6. There
K was a parametric disk and we proved that a hyperbolic surface does admit what is
called a harmonic measure of K here we are trying to prove the non-existence of a
harmonic measure. Since the latter was shown there to imply the existence of a Green
function, we are basically done. The only issue here is that K does not need to be a
parametric disk so we have to be careful when applying the Perron method because of
K not necessarily being regular. However, this is easily circumvented.
Suppose the proposition fails and let u > 0 somewhere on M \ K. Extend u to M by
setting u = 0 on K. Then K {u < } is an open neighborhood of K for every > 0.
This implies that for some > 0 we have u0 := (u )+ sh(M ) and u0 > 0 somewhere.
for some parametric disk D. Now define
Moreover, {u0 = 0} D
| v 0 on D, 0 v 1}
F =: {v sh(M \ D)
It is a Perron family, and from Chapter 9 we infer that
w := sup v
vF
with w = 0 on D and 0 w 1 on M \ D. Since u0 F we

is harmonic on M \ D
further conclude that w > 0 somewhere and thus everywhere on M \ D. As in the proof
of Theorem 10.6 we are now able to construct a Green function with singularity in D,
contrary to our assumption of M being parabolic.

It is instructive to give an independent proof of this fact for the case of C: we may
assume that 0 K. Then v(z) := u(1/z) is harmonic on := {1/z | z C \ K} and
3. PARABOLIC, SIMPLY CONNECTED, SURFACES
163
v 0 on \ {0}. Moreover, v is bounded above on . Given > 0 there exists R large

so that for all > 0
v(z) log(|z|/R) z
Sending 0 and then 0 shows that v 0 on as claimed.

Now let us make another observation concerning any u harmonic and bounded on
C \ K with K compact: for any R > 0 with K {|z| < R} we have
Z
u
d = 0
(11.1)
n
|z|=R
Indeed, v(z) = u(1/z) is bounded and harmonic on 0 < |z| < R1 + for some > 0. By
Problem 3.3, v is necessarily harmonic on a neighborhood of zero. Hence,
ZZ
I
v
d
0=
v dxdy =
n
1
|z| R
1
|z|= R
and (11.1) follows. An analogous result holds on any parabolic Riemann surface, but
the previous proof in C does not generalize to that setting. Let us give one that does
generalize: Without loss of generality K D. For any R > 1 denote by R the harmonic
function so that = 1 on {|z| = 1} and = 0 on {|z| = R} (harmonic measure). This
exists of course by Perron but we even have an explicit formula:
(z) =
log(R/|z|)|
,
log R
1 |z| R
Then, by Stokes theorem,

0=
|z|=1
Since
1
r log R ,
u
d +
r
|z|=R
d
r
d
r
|z|=1
it follows upon sending R that

I
u
d
0=
r
|z|=1
as desired. This proof can be made to work on a general parabolic Riemann surface and
we obtain the following result.
Lemma 11.3. Let D be a parametric disk on a parabolic
surface M and suppose u is

harmonic and bounded on M \ D. If u C 1 M \ D , then
Z
du = 0
D
is compact, D
N , and N is
Proof. We say that N M is admissible if N
2
so that = 1 on
piecewise C . Then by N we mean the harmonic function on N \ D
D and = 0 on N . We claim that
F := {N | N admissible}
164
11. UNIFORMIZATION
where we set each N = 0 on M \ N . In this way, each N

is a Perron family on M \ D
To verify that F is a Perron family, observe that from
becomes subharmomic on M \ D.
the maximum principle,
max{N1 , N2 } N1 N2
(N )K N K
is any parametric disk in the second line. Since N1 N2 and N K

where K M \ D
are again admissible, F is indeed such a family and
:= sup v
vF
is harmonic on M \ D with 0 1 and = 1 on D. Apply the maximum

principle for parabolic surfaces as given by Proposition 11.2 to 1 yields = 1
everywhere on M \ D.
Returning to any admissible N as above, we infer from Stokes theorem that
Z
N du u dN
0=
(N \D)
or, with suitable orientations,

Z
Z
Z
(11.2)
du = u dN +
u dN
D
It is clear that dN is of definite sign on both D and N . Indeed, on these boundaries

this differential form, evaluated at a tangent vector ~e to the boundary, is the directional
derivative of N along ~e (with a fixed sense of orientation along the boundary). Furthermore, again by Stokes,
Z
Z
dN =
dN
In view of (11.2) and the boundedness of u it therefore suffices to show that

Z

dN = 0
inf

N admissible
However, this follows immediately from the fact that = 1 everywhere on M \ D.
CHAPTER 12
Review of some facts from algebra and geometry

1. Geometry and topology
12.1. Classification of compact topological manifolds. A topological manifold
M is a second countable, connected, Hausdorff space, which is locally homeomorphic
to open subsets of the plane. A curve in M is a continuous map c : [0, 1] M and
a closed curve satisfies c(0) = c(1). On those curves, we define a boundary operator
c := c(1) c(0) which characterizes the closed curves c as all curves satisfying c = 0.
The difference of points
P here is a formal construction: we define a 0-cycle to be an
arbitrary finite sum j mj pj where mj Z and pj M . A 1-cycle is a formal finite
P
sum of closed curves with integer coefficients c := j nj cj where all cj are closed curves.
P
The boundary operator extends via linearity, i.e., c = j nj cj . We say that c is
closed iff c = 0 and exact iff c = where is a 2-cycle.
b2
b1
a1
a2
merge
Figure 12.1. A homology basis

The homology groups of M , denoted by H k (M ; Z) are
12.2. Differential forms. For the material here, see do Carmo [8], and Madsen,
Tornehave [30]. Every differentiable manifold M carries vector fields (defined as smooth
sections of the tangent bundle) and differential 1-forms (defined as smooth sections of
the co-tangent bundle). In local coordinates (x1 , x2 , . . . , xn ) with n = dim(M ), a 1-form
is given by
n
X
aj dxj =: aj dxj
=
j=1
(summation convention) with smooth aj on some open set. If we change co-ordinates

from x to y, then
xj
= aj k dy k
y
165
166
12. REVIEW OF SOME FACTS FROM ALGEBRA AND GEOMETRY
a1
1
b1
b1
a2
a1
b2
a1
2
b2
Figure 12.2. The fundamental polygon of a compact surface of genus g

This is referred to as a co-variant transformation rule of the coefficients. In contrast,
vector fields transform contra-variantly
v = vj
j y
v
=
v
xj
xj y k
More generally, the k-forms k (M ) are smooth sections of the bundle k (M ) of alternating k-forms on M . In this book, dim(M ) = 2 and only k = 0, 1, 2 are relevant. However,
0 (M ) = C (M ) and 2 (M ) are the so-called volume forms. This refers to the fact
that
Z
f
M
C (M )
is well-defined for every f

with compact support and 2 (M ). If M is
orientable (as a Riemann surface is), then there exists a volume form 0 so that at every
point p M there are local co-ordinates (x, y) with (x, y)(p) = (0, 0) and
0 = f (x, y) dx dy
with f (0, 0) 6= 0. This then implies that 2 (M ) C (M ) since every 2-form is a

multiple of 0 . There are two operations on forms that we shall need: the exterior
product (denoted by ) and the exterior differentiation (denoted by d). If 1 1 (M )
and 2 2 (M ), then
as well as
1 2 1 +2 (M ),
1 2 = (1)1 2 2 1
d(1 2 ) = (d1 ) 2 + (1)1 1 (d2 ),
d(f ) = df + f d
for any f 0 (M ). In local coordinates,

aj
f
dxj ,
d(aj dxj ) =
dxk dxj
df =
j
x
xk
From the chain rule one verifies that exterior product and differentiation commute with
pullbacks. I.e., if f : M N is a smooth map between differentiable manifolds, then
the pullback f satisfies
f ( ) = f f ,
d(f ) = f (d)
1. GEOMETRY AND TOPOLOGY
167
for any differential forms , on N . By equality of mixed partial d2 f = 0, and d2 = 0

for any 1 because d2 = {0} (in fact, d2 = 0 always). We say that 1 (M ) is
closed iff d = 0 and is exact if = df for some f 0 (M ). By the preceding, exact
forms are closed, but the converse is not true (consider the form d in R2 \ {(0, 0)} where
is the polar angle). The Poincare lemma says that locally all closed forms are exact.
From Poincares lemma and Stokes theorem one obtains that for any closed 1-form
Z
Z
=
c2
c1
if c1 , c2 have the same end-points and are homotopic. Similarly,

I
I
=
c2
c1
if c1 , c2 are closed homotopic curves.
12.3. de Rham cohomology. Of great topological importance are the de Rham

spaces, which are characterized as closed forms modulo exact forms, i.e., for each k 1,
H k (M ) := {f k (M ) : df = 0}/{dg : g k1 (M )}
If M is compact, then these spaces have finite dimension. The dimensions are the Betti
numbers k (M ) and the de Rham theorem says that they agree with the dimensions
(over Z) of the homology groups. For us the only really relevant case is k = 1, whereas
for k = 2 one has H 2 (M ) R due to the orientability of M . Finally, we note that the
pull back via smooth maps is well-defined on the cohomology since it commutes with the
exterior differentiation. Moreover, the pull back map on the cohomology is the same for
any two smooth functions which are homotopic.
12.4. The degree. Next, we recall the notion of degree from topology and check
that it coincides with the degree defined in Chapter 4 for Riemann surfaces. For the
sake of this paragraph alone, let M, N be n-dimensional smooth orientable, connected
compact manifolds. Then integration defines a linear isomorphism
Z
n
H (M ) R, [] 7
M
H n (M )
where
is the de Rham space of n-forms modulo exact n-forms. Let f : M N
be a smooth map and f : H n (N ) H n (M ) the induced map defined via the pull-back.
There exists a real number denoted by deg(f ) such that
Z
Z
H n (N )
f () = deg(f )
M
Since the pull-back map on the cohomology spaces H n only depends on the homotopy
class, so does deg(f ). It is also easy to see that it is multiplicative with regard to
composition. Changing variables in charts, it is easy to verify that for any regular value
q N (which means that Df (p) : Tp M Tq N is invertible for every p with f (p) = q)
X
deg(f ) =
Ind(f ; p)
pM : f (p)=q
where Ind(f ; p) = 1 depending on whether Df (p) preserves or reverses the orientation.

In particular, deg(f ) Z and deg(f ) 6= 0 implies that f is onto. By Sards theorem,
the regular values are always dense in N (an excellent reference for all of this is [30]).
168
12. REVIEW OF SOME FACTS FROM ALGEBRA AND GEOMETRY
Returning to Riemann surfaces, we see that this notion of degree coincides exactly with
the one from Chapter 4 since every analytic f : M N necessarily preserves the
orientation.
12.5. Euler characteristic. Every topological twodimensional compact manifold
M has an integer (M ) associated with itself, called the Euler characteristic. It is defined
as
V E + F = (M ), V = vertices, E = edges, F = faces
relative to an arbitrary triangulation of M (this is the homological characterization of
Figure 12.3. A sphere with k handles has genus k

(M )); this is well-defined, in other words, V E + F does not depend on the particular
choice of triangulation. Another important theorem relates the Euler characteristic with
the genus g of M : if we realize M as S 2 with g handles attached, then we have the
EulerPoincare formula
(12.1)
(M ) = 2 2g
Finally, let us recall the cohomological characterization of (M ): let M be a compact,

smooth twodimensional manifold and let H k (M ) denote the de Rham spaces of closed
k forms modulo exact forms, 0 k 2. Then, with k := dim H k (M ),
(M ) = 0 1 + 2
If M is orientable (as in the case of a Riemann surface), then it is easy to see that
0 = 2 = 1. Thus, 1 = 2g where g is the genus.
2. Algebra
For the material here, see for example [28]. Given two relatively prime polynomials
P, Q C[w, z], there exist A, B C[w, z] such that
A(w, z)P (w, z) + B(w, z)Q(w, z) = R(z) C[z]
is a nonzero polynomial in z alone; it is called the resultant of P and Q. The proof of

this fact is Euclids algorithm carried out in the polynomials in K(z)[w] where K(z) is
2. ALGEBRA
169
the quotient field of C[z], i.e., the field of rational functions of z. The resultant has many
interesting properties, for example, if both P and Q have leading coefficient 1, then
Y
R(z) =
(j (z) k (z))
j ,k
where j runs over all zeros of P (w, z) and k runs over all zeros of Q(w, z) in w, respectively. Thus, R(z0 ) = 0 iff P (w, z0 ) and Q(w, z0 ) have a common zero in w. Moreover,
with
m
n
X
X
aj (z)wj ,
Q(w, z) =
bk (z)wk
P (w, z) =
j=0
k=0
it follows that R(z) equals the following determinant in the coefficients aj , bk

n
m

z
}|
{
}|
{
z

ak
0
bm
0
0

ak1 ak
bm1 bm

ak2 ak1
bm2 bm1

(12.2)

a0
a1

0
a0

0
0
Bibliography
[1] Ahlfors, L. Complex analysis. An introduction to the theory of analytic functions of one complex
variable. Third edition. International Series in Pure and Applied Mathematics. McGraw-Hill Book
Co., New York, 1978
[2] Ahlfors, L., Sario, L. Riemann Surfaces, Princeton University Press, 1960.
[3] Beardon, A. F. A primer on Riemann Surfaces, London Mathematical Society Lecture Notes Series
78, Cambridge University Press, Cambridge, 1984.
[4] Beardon, A. F. The geometry of discrete groups. Corrected reprint of the 1983 original. Graduate
Texts in Mathematics, 91. Springer-Verlag, New York, 1995.
[5] Beardon, A. F. Algebra and Geometry. Cambridge University Press, Cambridge, 2005.
[6] Chandrasekharan, K. Elliptic functions. Grundlehren der Mathematischen Wissenschaften, 281.
Springer-Verlag, Berlin, 1985.
[7] Conway, J. B. Functions of one complex variable, Springer, Second Edition, 1978.
[8] do Carmo, M. P. Differential forms and applications. Universitext. Springer-Verlag, Berlin, 1994.
[9] Dubrovin, B. A. Theta-functions and nonlinear equations. (Russian) With an appendix by I. M.
Krichever. Uspekhi Mat. Nauk 36 (1981), no. 2(218), 1180.
[10] Dubrovin, B. A., Krichever, I. M., Novikov, S. P. Topological and algebraic geometry methods in
contemporary mathematical physics. Classic Reviews in Mathematics and Mathematical Physics, 2.
Cambridge Scientific Publishers, Cambridge, 2004
[11] Ekedahl, T. One semester of elliptic curves. EMS Series of Lectures in Mathematics. European
Mathematical Society (EMS), Z
urich, 2006.
[12] Evans, L. C. Partial Differential equations, AMS, Graduate Studies in Mathematics, vol. 19, 1998.
[13] Fay, John D. Theta functions on Riemann surfaces. Lecture Notes in Mathematics, Vol. 352.
Springer-Verlag, Berlin-New York, 1973.
[14] Feldman, J., Kn
orrer, H., Trubowitz, E. Riemann surfaces of infinite genus. CRM Monograph Series,
20. American Mathematical Society, Providence, RI, 2003.
[15] Forster, O. Lectures on Riemann surfaces. Reprint of the 1981 English translation. Graduate Texts
in Mathematics, 81. Springer-Verlag, New York, 1991.
[16] Farkas, H. M., Kra, I. Riemann surfaces. Second edition. Graduate Texts in Mathematics, 71.
Springer-Verlag, New York, 1992.
[17] Garnett, John B. Bounded analytic functions. Revised first edition. Graduate Texts in Mathematics,
236. Springer, New York, 2007.
[18] Gesztesy, F., Holden, H. Soliton equations and their algebro-geometric solutions. Vol. I. (1 + 1)dimensional continuous models. Cambridge Studies in Advanced Mathematics, 79. Cambridge University Press, Cambridge, 2003.
[19] Griffiths, P., Harris, J. Principles of algebraic geometry. Reprint of the 1978 original. Wiley Classics
Library. John Wiley & Sons, Inc., New York, 1994.
[20] Hurwitz, A., Courant, R. Vorlesungen u
ber allgemeine Funktionentheorie und elliptische Funktionen.
Interscience Publishers, Inc., New York, (1944).
[21] J
anich, K. Einf
uhrung in die Funktionentheorie, zweite Auflage, Springer, 1980.
[22] Jones, G., Singerman, D. Complex functions. An algebraic and geometric viewpoint. Cambridge
University Press, Cambridge, 1987.
[23] Jost, J. Compact Riemann surfaces. An introduction to contemporary mathematics. Third edition.
Universitext. Springer-Verlag, Berlin, 2006.
[24] Jost, J. Riemannian geometry and geometric analysis. Fourth edition. Universitext. Springer-Verlag,
Berlin, 2005.
171
172
BIBLIOGRAPHY
[25] Katok, S. Fuchsian groups. Chicago Lectures in Mathematics. University of Chicago Press, Chicago,
IL, 1992.
[26] Lang, S. Elliptic functions. With an appendix by J. Tate. Second edition. Graduate Texts in Mathematics, 112. Springer-Verlag, New York, 1987.
[27] Lang, S. Complex analysis. Fourth edition. Graduate Texts in Mathematics, 103. Springer-Verlag,
New York, 1999.
[28] Lang, S. Algebra. Revised third edition. Graduate Texts in Mathematics, 211. Springer-Verlag, New
York, 2002.
[29] Levin, B. Ya. Lectures on Entire Functions, Translations of Mathematical Monographs, Volume 150,
AMS 1996.
[30] Madsen, I., Tornehave, J. From calculus to cohomology. de Rham cohomology and characteristic
classes. Cambridge University Press, Cambridge, 1997.
[31] McKean, H., Moll, V. Elliptic curves. Function theory, geometry, arithmetic. Cambridge University
Press, Cambridge, 1997.
[32] Mumford, D. Tata lectures on theta I, II, III, reprints of the original edition, Birkhaeuser, 2007.
[33] Narasimhan, R. Compact Riemann surfaces. Lectures in Mathematics ETH Z
urich. Birkh
auser Verlag, Basel, 1992.
[34] Nevanlinna, R. Uniformisierung, Die Grundlehren der mathematischen Wissenschaften in Einzeldarstellungen, Band 64, Springer, zweite Auflage, 1967.
[35] Rosenblum, M., Rovnyak, J. Topics in Hardy Classes and Univalent Functions, Birkh
auser Advanced
Texts, Basel, 1994.
[36] Springer, G. Introduction to Riemann surfaces. Addison-Wesley Publishing Company, Inc., Reading,
Mass. 1957.
[37] Stein, E., Shakarchi, R. Complex analysis, Princeton Lectures in Analysis II, Princeton University
Press, 2003.
[38] Teschl, G. Jacobi operators and completely integrable nonlinear lattices. Mathematical Surveys and
Monographs, 72. American Mathematical Society, Providence, RI, 2000.
[39] Titchmarsh, E. C. The theory of functions, Oxford, second edition, 1939.
[40] Titchmarsh, E. C. The theory of the Riemann zeta-function. Second edition. The Clarendon Press,
Oxford University Press, New York, 1986.
[41] Weyl, H. Die Idee der Riemannschen Fl
ache. F
unfte Auflage. B. G. Teubner, Stuttgart, 1974.

A Concise Course in Complex Analysis and Riemann Surfaces

Uploaded by

Copyright:

Available Formats

A Concise Course in Complex Analysis and Riemann Surfaces

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Concise Course in Complex Analysis and Riemann Surfaces

Uploaded by

Copyright:

Available Formats

What topics does the book cover?

What topics does the book cover?

What are Riemann surfaces and how are they defined?

What are Riemann surfaces and how are they defined?

A concise course in complex

analysis and Riemann surfaces

Chapter 1. From i to z: the basics of complex analysis

From z to the Riemann mapping theorem: some finer points of basic

Chapter 3. Harmonic functions on D

Chapter 4. Riemann surfaces: definitions, examples, basic properties

Chapter 5. Analytic continuation, covering surfaces, and algebraic functions

3. The ramified Riemann surface of an analytic germ

Chapter 6. Differential forms on Riemann surfaces

Chapter 7. Hodges theorem and the L2 existence theory

Chapter 8. The Theorems of Riemann-Roch, Abel, and Jacobi

Chapter 9. The Dirichlet problem and Green functions

Chapter 10. Green functions and the classification problem

Chapter 11. The uniformization theorem

Chapter 12. Review of some facts from algebra and geometry

hyperbolic geometry. In particular, we prove the Gauss-Bonnet theorem in that case.

From i to z: the basics of complex analysis

an z n converges on |z| < R and diverges for every |z| > R

R1 = lim sup |an | n

Another example is of course

1. BASIC COMPLEX ANALYSIS I

which converges absolutely and uniformly on every compact subset of C. Expanding

(t s)n y(s) ds.

The integral on the right vanishes as n and we obtain

2. DIFFERENTIABILITY AND CONFORMALITY

where df (z)(h) means the matrix df (z) acting on the vector h.

exists for all z and is continuous on . We denote this by f H(). A function

1. BASIC COMPLEX ANALYSIS I

Proof. Suppose z0 and

z so that |z z0 | < r(z0 )

In fact, one can differentiate any number of times and

where (n)k = n(n 1) . . . (n k + 1). This proves also that an =

v(x + , y + ) = v(x, y) + vx (x, y) + vy (x, y) + o(|(, )|).

f (z + ) f (z) = (ux + ivx )(z)( + i) + o(||)

2. DIFFERENTIABILITY AND CONFORMALITY

as well as the representation of the Laplacean = 4 z

1. BASIC COMPLEX ANALYSIS I

intuitively clear that C S 2 in the homeomorphic sense. Somewhat deeper as well

Figure 1.1. Stereographic projection

around z = 0. For the final example in (1.3), if f () = , then f is analytic around

which is holomorphic as a map from C C . It is called a fractional linear or

P SL(2, C) := SL(2, C)/{Id}

e under the assumption that

e are the same matrices in SL(2, C) possibly up to a choice of sign, which

1. BASIC COMPLEX ANALYSIS I

Fractional linear transformations enjoy many important properties which can be

and we are done.

Lemma 1.8. Fractional linear transformations take circles onto circles.

with a fixed point in C and another one at z = .

1. BASIC COMPLEX ANALYSIS I

Proof. It suffices to consider the unit circle. Then

Figure 1.2. Geodesics in the hyperbolic plane

verify that for all z0 H

Figure 1.3. Geodesic triangles

Theorem 1.14. Let T be a geodesic triangle with angles 1 , 2 , 3 . Then Area(T ) =