Вы находитесь на странице: 1из 108

Università degli Studi di Trieste

Dipartimento di Matematica e Geoscienze


Corso di Laurea Magistrale in Matematica

THE UNIFORMIZATION PROBLEM


FOR THE N-PUNCTURED SPHERE

Relatore: Tesi di Laurea di:


Prof. Boris Dubrovin Lorenzo Guerini

Anno Accademico 2015-2016


ii

A Francesco e Matteo
Contents

1 Uniformization Theorem 1
1.1 Riemann Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Elementary properties of holomorphic mapping . . . . . . 2
1.1.2 Covering maps . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Uniformization theorem . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.1 Fuchsian groups . . . . . . . . . . . . . . . . . . . . . . . 8

2 Liouville equation 11
2.1 Liouville equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Monodromy and deck transformation group . . . . . . . . . . . . 15
2.3 Punctured Riemann surfaces . . . . . . . . . . . . . . . . . . . . 18
2.3.1 Properties of deck transformation group . . . . . . . . . . 18
2.3.2 Canonical connection of a punctured Riemann surface . . 20
2.3.3 Deck transformation group of the first kind . . . . . . . . 24

3 The 3-punctured sphere 27


3.1 Accessory parameters . . . . . . . . . . . . . . . . . . . . . . . . 27
3.2 Properties of hypergeometric equation . . . . . . . . . . . . . . . 28
3.2.1 Local solutions at the origin . . . . . . . . . . . . . . . . . 28
3.2.2 Local solution at 1 and infinity . . . . . . . . . . . . . . . 30
3.2.3 Monodromy of the hypergeometric equation . . . . . . . . 31
3.3 Analytic continuation of local solution . . . . . . . . . . . . . . . 33
3.4 Theta-functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 The modular lambda function . . . . . . . . . . . . . . . . . . . . 38
3.6 Solution of Liouville equation . . . . . . . . . . . . . . . . . . . . 45

4 Solutions of Liouville equation 49


4.1 Asymptotic behavior of the solutions of Liouville’s equation . . . 49
4.1.1 Uniqueness of solutions to Liouville equation . . . . . . . 54
4.1.2 Solution with periodic condition . . . . . . . . . . . . . . 55
4.2 The punctured torus . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2.1 Accessory parameter of a 1-punctured torus . . . . . . . . 59
4.3 Liouville equation as Euler Lagrange equation of a functional . . 64

5 Teichmüller spaces 67
5.1 Beltrami equation . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.1.1 Quasiconformal mapping . . . . . . . . . . . . . . . . . . 67
5.2 Teichmüller spaces . . . . . . . . . . . . . . . . . . . . . . . . . . 73

iii
iv CONTENTS

5.2.1 Teichmüller space of a Fuchsian group . . . . . . . . . . . 75


5.2.2 Simultaneous uniformization . . . . . . . . . . . . . . . . 79
5.2.3 Bers embedding . . . . . . . . . . . . . . . . . . . . . . . . 80
5.2.4 The infinitesimal approach . . . . . . . . . . . . . . . . . 82
5.3 A generating function for the accessory parameters . . . . . . . . 86
5.3.1 Complex analytical covering of the space of punctured
spheres . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.3.2 Zograf–Takhtadzhyan theorem . . . . . . . . . . . . . . . 89
Introduction

The Riemann mapping theorem was first stated in the PhD thesis of Riemann
[16] in the 1851. In the original version the theorem asserts that every non
empty simply connected open subset U of the complex plane, such that ∂U is
piecewise smooth, is conformally equivalent to the upper half plane H. It is well
known that Riemann’s proof is incomplete, as a matter of fact Riemann took
for granted the so-called Dirichlet principle, which is not always valid. Only in
1899 Hilbert gave a rigorous justification of Dirichlet principle, under the same
assumption on the boundary used by Riemann.
Meanwhile Poincaré (1882) and Klein (1883) conjectured the uniformization
theorem for simply connected Riemann surfaces of algebraic curves. The state-
ment of their conjecture is that if the Riemann surface of an algebraic curve
is simply connected then it is conformally equivalent to the complex sphere
Ĉ, the complex plane or the upper half plane H. Later on in 1884 Poincaré
[14] showed that the differential equation characterizing the uniformization of
a Riemann surface of an algebraic curve of genus g with n branch points is de-
scribed by 3g − 3 + n complex parameters, which are called accessory parameters.
However the proof of the existence of these parameters, not to speak of their
explicit computation, turned out to be a very difficult problem. Therefore he
abandoned this approach. Finally in 1907 Poincaré [15] and Koebe [10], using
the so-called potential theory, proved the uniformization theorem for simply
connected Riemann surfaces in the modern formulation.

Theorem (Uniformization theorem – simply connected case). Every simply


connected Riemann surface is conformally equivalent to Ĉ, C or H.

Using the theory of covering maps we easily prove the general version of
uniformization theorem.

Theorem (Uniformization theorem). Given a Riemann surface R the universal


covering J : R̂ → R of R is an analytic covering map J such that R̂ is a simply
connected Riemann surface. If

Deck(J) = {γ ∈ Aut(R̂)| J ◦ γ = J},

then R is conformally equivalent to R̂/Deck(J).

We know that R̂ is conformally equivalent to Ĉ, C or H, hence the group


Deck(J) is a group of Möbius transformation, which acts properly discontinuously
on R̂. In the case R̂ = H the subgroups of Aut(H) acting properly discontinuously
are known as Fuchsian groups.

v
vi CONTENTS

Even if we know that an uniformization of the Riemann surface R exists, it


is very difficult to find its explicit structure, i.e. the map J and Deck(J), except
for the few cases in which the universal covering space is not H.
It turns out that the approach proposed by Poincaré in [14] helps us to find
out the explicit uniformization of the Riemann surface R.
We briefly describe it in the case of a n-punctured sphere. Let X = C \
{z1 , . . . , zn , 0, 1}, then Liouville equation

∆ϕ = e2ϕ ,

is uniquely solvable in the class of C 2 function on X with boundary conditions


(
− log |z − zi | − log | log |z − zi ||, z → zi ,
ϕ(z, z̄) = (1)
− log |z| − log log |z|, z → ∞,

this unique solution is denoted as ϕX . It is possible to prove that the function


ωX = ∂z2 ϕX − (∂z ϕX )2 is meromorphic on Ĉ, that every branch of J −1 satisfies
the differential equation
2
(J −1 )000 (J −1 )00

3
− = 2ωX
(J −1 )0 2 (J −1 )0

and that Deck(J) is isomorphic to the monodromy group of the second order
differential equation u00 + ωX u = 0. Moreover ωX is described it term of n − 3
parameters, which are indeed the accessory parameters.

Theorem. Let X = C \ {z1 , . . . , zn−3 , 0, 1}. The meromorphic function ωX has


the following form
n−1  
X 1 ci 1 cn 1
ωX (z) = + = 2 + 3 +o z → ∞.
i=1
4(z − zi ) 2 2(z − zi ) 4z 2z z3

The {ci } are complex numbers called accessory parameters. They satisfy
n−1 n−1 n−1
X X n X
ci = 0, ci zi = 1 − , zi (1 + ci zi ) = cn
i=1 i=1
2 i=1

The accessory parameters are easily calculated in the case of the 3-punctured
sphere, while there is still no way to calculate them for a general Riemann surface
of genus g and with n punctures, when 3g + n − 3 ≥ 1.
In the case X = C \ {0, 1} we find that

Theorem. The universal covering map of the 3-punctured sphere C \ {0, 1} is


the modular lambda function

θ24 (0, τ )
λ(τ ) = ,
θ34 (0, τ )

where θi (z, τ ) : C × H → C are the so-called theta functions (cfr. [20]).


CONTENTS vii

All those results were proved at the beginning of the XX century. The
approach to uniformization based on Liouville equation and accessory parameters
has attracted the attention of specialists in quantum field theory during the ’80s
in connection with Liouville quantum gravity. In particular on the class of C ∞
functions with asymptotics conditions (1) we define Liouville action as
Z   
1 πn
S[ψ] = lim |ψz |2 + e2ψ dxdy + (log r + 2 log | log r|) .
r→0
Xr 4 2

Polyakov conjectured the following theorem, which was proved in 1988 by Zograf
and Takhtadzhyan [18]
Theorem. Let S(z1 , . . . , zn−3 ) = S[ϕ(z; z1 , . . . , zn−3 )] where (z1 , . . . , zn−3 ) ∈
Wn = {z ∈ Cn−3 |zi 6= zj 6= 0, 1}, and ϕ is the solution of Liouville equation on
C \ {z1 , . . . , zn−3 }. Then
2 ∂S
ci = − .
π ∂zi
The present thesis is organized in the following way.
In the first chapter we present some basic notions necessary to explain the
uniformization theorem. The result for simply connected surfaces is not proved
because it would take too long. At the end of the chapter we define what is it a
Fuchsian group and we prove some useful facts related to those type of groups.
In the second chapter we present the Liouville approach to uniformization
in the case of a generic Riemann surface. In particular we study the case
of a punctured Riemann surface and we prove the existence of the accessory
parameters.
The third chapter is dedicated to the study of the 3-punctured sphere.
Essentially we prove that λ is the universal covering map of C \ {0, 1}. In order
to prove this result we present first the theory of hypergeometric equations and
the theory of theta functions.
In the fourth chapter we study the solutions of the Liouville equation with
boundary behavior (1). Eventually this analysis is helpful in order to establish a
connection formula between the accessory parameter c of the 4-punctured sphere
and the accessory parameter E of the 1-punctured torus.
In the last chapter we prove the result of Zograf and Takhtadzhyan. The
proof of this theorem requires a good background of the theory of Teichmüller
spaces. Hence in the first two sections the reader is introduced to the theory of
quasiconformal mappings and then to the theory of Teichmüller spaces.
viii CONTENTS
Introduzione

Il teorema della mappa di Riemann è stato enunciato per la prima volta nella tesi
di PhD di Riemann [16] nel 1851. Nella versione originale il teorema asserisce
che ogni insieme aperto e semplicemente connesso U del piano complesso, con
bordo C ∞ a tratti, è conformemente equivalente al semipiano superiore H. È ben
noto che la dimostrazione di Riemann è incompleta in quanto da per scontato
il cosidetto principio di Dirichlet, il quale non è vero in generale. Solamente
nel 1899 Hilbert diede una rigorosa giustificazione di tale principio per i domini
studiati da Riemann.
Nel frattempo Poincaré (1882) e Klein (1883) congetturarono che se la
superfice di Riemann di una curva algebrica è semplicemente connessa, allora è
conformemente equivalente ad uno fra la sfera complessa Ĉ, il piano complesso o il
semipiano superiore H. Nel 1884 Poincaré [14] provò che l’equazione differenziale
caratterizzante l’uniformizzazione di una superficie algebrica di genere g con n
branch points è descritta da 3g − 3 + n parametri complessi, chiamati parametri
accessori. Tuttavia la prova dell’esistenza di questi parametri, per non parlare
del loro calcolo, risultò essere un problema molto difficile. Di conseguenza questo
approccio fù abbandonato. In fine nel 1907 Poincarè [15] e Koebe [10], utilizzando
la cosidetta potential theory, provarono la formulazione moderna del teorema di
uniformizzazione per superfici di Riemann semplicemente connesse.

Theorem (Teorema di uniformizzazione – caso semplicemente connesso). Ogni


superfice di Riemann semplicemente connessa è conformemente equivalente a Ĉ,
C o H.

Utilizzando invece la teoria delle mappe di rivestimento, si dimostra facilmente


la versione generale del teorema di uniformizzazione.

Theorem (Teorema di uniformizazione). Data una superficie di Riemann R,


sia J : R̂ → R la mappa di rivestimento universale di R, dove R̂ é una superficie
di Riemann semplicemente connessa e J é una mappa di rivestimento analitica.
Se
Deck(J) = {γ ∈ Aut(R̂)| J ◦ γ = J},
allora R é conformemente equivalente a R̂/Deck(J).

Dal momento che R̂ puó essere solamente Ĉ, C o H, allora il gruppo Deck(J)
é un gruppo di trasformazioni di Möbius. Inoltre questo gruppo agisce in
maniere propriamente discontinua su R̂. Se R̂ = H allora i sottogruppi di Aut(H)
che agiscono in maniera propiamente discontinua sono conosciuti come gruppi
Fuchsiani.

ix
x CONTENTS

Il più grosso problema del teorema di uniformizzazione è che, sebbene


un’uniformizzazione di R esiste, non siamo in grado di trovare la sua strut-
tura esplicita, ovvero la mappa J e Deck(J), ad eccezioni dei pochi casi in cui
lo spazio di rivestimento universale non è H.
Quello che si scopre è che l’approccio proposto da Poincaré in [14] può essere
utile per trovare l’uniformizzazione esplicita di R.
Descriviamo ora brevemente questo approccio nel caso della sfera con n punture.
Sia X = C \ {z1 , . . . , zn , 0, 1}, allora l’equazione di Liouville

∆ϕ = e2ϕ ,

ammette un’unica soluzione nella classe di funzioni C 2 su X con condizioni al


bordo (
− log |z − zi | − log | log |z − zi ||, z → zi ,
ϕ(z, z̄) =
− log |z| − log log |z|, z → ∞,
questa particolare soluzione è denotata come ϕX . Si può provare che ωX =
∂z2 ϕX − (∂z ϕX )2 è meromorfa su Ĉ, che ogni branch di J −1 soddisfa l’equazione
differenziale 2
(J −1 )000 3 (J −1 )00

− = 2ωX
(J −1 )0 2 (J −1 )0
e che Deck(J) è isomorfo al gruppo di monodromia dell’equazione differenziale
di secondo ordine u00 + ωX u = 0. Inoltre la funzione ωX é scritta in termini di
n − 3 parametri, i cosiddetti parametri accessori.

Theorem. Sia X = C \ {z1 , . . . , zn−3 , 0, 1}. La funzione meromorfa ωX ha la


forma seguente
n−1  
X 1 ci 1 cn 1
ωX (z) = + = 2 + 3 +o z → ∞.
i=1
4(z − zi )2 2(z − zi ) 4z 2z z3

Dove {ci } sono numeri complessi chiamati parametri accessori, i quali soddisfano
n−1 n−1 n−1
X X n X
ci = 0, ci zi = 1 − , zi (1 + ci zi ) = cn
i=1 i=1
2 i=1

I parametri accessori sono calcolati facilmente nel caso della sfera con 3
punture, mentre non si conosce ancora un modo di calcolarli per una generica
superficie di Riemann di genere g e con n punture, nel caso in cui 3g + n − 3 ≥ 1.
Se X = C \ {0, 1} é possibile provare che

Theorem. La mappa di rivestimento universale di C\{0, 1} é la funzione lambda


modulare
ϑ4 (0, τ )
λ(τ ) = 42 ,
ϑ3 (0, τ )
dove ϑi (z, τ ) : C × H → C sono le cosiddette funzioni theta (cfr. [20]).

Tutti questi risultati sono stati provati all’inizio del XX secolo. L’approccio
all’uniformizzazione basato sull’equazione di Liouville ha attirato l’attenzione
degli specialisti durante gli anni 80, in relazione alla Liouville quantum gravity. In
CONTENTS xi

particolare sullo spazio delle funzioni C ∞ con condizioni asintotiche (1), l’azione
di Liouville viene definita come
Z   
2 1 2ψ πn
S[ψ] = lim |ψz | + e dxdy + (log r + 2 log | log r|) .
r→0
Xr 4 2

Polyakov ha congetturato il seguente teorema, il quale è stato provato nel 1988


da Zograf e Takhtadzhyan [18]

Theorem. Sia S(z1 , . . . , zn−3 ) = S[ϕ(z; z1 , . . . , zn−3 )] dove (z1 , . . . , zn−3 ) ∈


Wn = {z ∈ Cn−3 |zi 6= zj 6= 0, 1} e ϕ è la soluzione dell’equazione di Liou-
ville su C \ {z1 , . . . , zn−3 }. Allora

2 ∂S
ci = − .
π ∂zi
Questa tesi è organizzata nella maniera seguente.
Nel primo capitolo sono presentate alcune nozioni necessarie a spiegare il teorema
di Uniformizzazione. Non viene provato il risultato per superfici semplicemente
connesse, in quanto troppo lungo. Alla fine del capitolo diamo la definizioni di
gruppo Fuchsiano e dimostriamo alcuni utili proposizioni che riguardano questi
gruppi.
Nel secondo capitolo viene presentato l’approccio di Liouville al problema di
uniformizzazione nel caso di una superficie di Riemann generica. In particolare
viene studiato il caso delle superfici di Riemann puntate a viene dimostrata
l’esistenza dei parametri accessori.
Il terzo capitolo viene dedicato allo studio della sfera con tre punture. In
sostanza viene provato che λ è la mappa di rivestimento universale di C \ {0, 1}.
Al fine di provare questo risultato è necessario introdurre la teoria delle equazioni
ipergeometriche e la teoria delle funzioni theta.
Nel quarto capitolo vengono studiate le soluzioni dell’equazione di Liouville
con condizioni al contorno (1). Questa analisi permette di stabilire una formula di
connessione fra il parametro accessorio c della sfera con 4 punture e il parametro
accessorio E del toro con 1 punture.
Infine nell’ultimo capitolo viene provato il risultato di Zograf e Takhtadzhyan.
La dimostrazione del teorema richiede un background sulla teoria degli spazi di
Teichmüller. Di conseguenza nelle prime due sezioni saranno presentate la teoria
delle mappe quasiconformi e appunto la teoria degli spazi di Teichmüller.
xii CONTENTS
Chapter 1

Uniformization Theorem

In this chapter we present the uniformization theorem for Riemann surfaces.


In the first section we introduce the theory of Riemann surfaces. We refer the
interested reader to the book of Forster [7] for additional information about this
argument.
In the second section we present and prove the uniformization theorem,
together with some results connected with Fuchsian groups, i.e. groups of
Möbius transformation acting properly discontinuously on the upper half plane.

1.1 Riemann Surfaces


A Riemann surface is a one dimensional complex manifold, i.e. a two dimensional
real manifold that carries a complex analytic structure. The rigorous definition
is the following.
Definition A Riemann surface is a connected Hausdorff space R together with
a collection of charts {Uα , zα } that satisfies the following properties:
• The collection of sets Uα form an open covering of R.
• Each zα is a homeomorphic mapping of Uα onto an open subset of the
complex plane C.
• If Uα ∩ Uβ 6= ∅ then fαβ := zβ ◦ zα−1 is holomorphic on zα (Uα ∩ Uβ ).

Remark The determinant of dfαβ is positive on Uα ∩ Uβ , therefore a Riemann


surface is an oriented manifold.

Definition Suppose R and S are two Riemann surfaces. A continuous mapping


f : R → S is called holomorphic, if for every pair of charts ψ1 : U1 → V1 on R
and ψ2 : U2 → V2 on S with f (U1 ) ⊂ U2 , the mapping

ψ2 ◦ f ◦ ψ1−1 : V1 → V2

is holomorphic.
A mapping f : R → S is called biholomorphic if it is bijective and both f
and f −1 are holomorphic. Two Riemann surfaces are conformally equivalent if
there exists a biholomorphic map between them.

1
2 Chapter 1. UNIFORMIZATION THEOREM

The family of holomorphic functions between a Riemann surface R and C is the


C-algebra of holomorphic functions R, which is denoted as O(R).
In the same way the family of holomorphic functions between R and the complex
sphere Ĉ = C ∪ {∞} is the C-algebra of the meromorphic functions, which is
denoted as M (R).

1.1.1 Elementary properties of holomorphic mapping


Since an holomorphic function between two Riemann surfaces, is locally an
analytic function between an open subset of C and C, many properties of
holomorphic function are true even for Riemann surfaces.
We will list the most important.

Theorem 1.1.1 (Riemann’s removable singularities theorem). Let U be an open


subset of a Riemann surface and let a ∈ U . Suppose the function f ∈ O(U \ {a})
is bounded in some neighborhood of a. Then f can be extended uniquely to a
function f˜ ∈ O(U ).

Theorem 1.1.2 (Identity theorem). Suppose R, S are Riemann surfaces and


f1 , f2 : R → S are two holomorphic mappings. If the two functions coincide on
a set A ⊂ R which has an accumulation point a ∈ R, they are identically equal.

Proof. Let G be the set defined as

G = {x ∈ R| ∃ U 3 x open set s.t. f1 |U ≡ f2 |U }

G is an open set.
Let b ∈ Ḡ, we consider a chart (U, ϕ) containing b and a chart (U 0 , ψ) in S
containing f1 (b) = f2 (b). Here the equality follows from the continuity of the
two functions.
Identity theorem for holomorphic functions implies that

ψ ◦ f1 ◦ ϕ−1 = ψ ◦ f2 ◦ ϕ−1

in a neighborhood Ũ 3 ϕ(b) therefore f1 ≡ f2 in a neighborhood of b. This


proves that G is close.
Since A has a limiting point we can apply again the Identity theorem and prove
that G is not empty. Therefore since the manifold is connected G = R.

Theorem 1.1.3 (Local behavior of holomorphic mappings). Suppose R and S


are Riemann surfaces, and f : R → S is a non-constant holomorphic mapping.
For every a ∈ R, there exists a chart (U, ϕ) centered at that point and a chart
(U 0 , ψ) of S centered in f (a), with f (U ) ⊂ U 0 and such that the map

F (z) := ψ ◦ f ◦ ϕ−1 (z) = z k

for a certain integer k ≥ 1.

Proof. Let’s take any couple of charts (U, ϕ) of R centered in a and a chart
(U 0 , ψ) centered in f (a), such that f (U ) ⊂ U 0 . Let V = ϕ(U ) and V 0 = ψ(U 0 ),
the map
g := ψ ◦ f ◦ ϕ−1 : V → V 0
1.1. Riemann Surfaces 3

is not constant, since if it wasn’t from identity theorem f would be constant too.
Since g(0) = 0 there exist k ≥ 1 such that g(z) = z k h(z), where h is holomorphic
and satisfies h(0) 6= 0. In addition to this there exists a neighborhood V1 ⊂ V
and an holomorphic function h1 such that hk1 = h.
Let α(z) = zh1 (z), we define U1 = ϕ−1 (V1 ) and ϕ1 = α ◦ ϕ. The couple of charts
(U1 , ϕ1 ) and (U 0 , ψ) satisfies the hypothesis of the theorem.
Corollary 1.1.4 (Open mapping theorem). Let R and S be Riemann surfaces
and let f : R → S be a non-constant holomorphic mapping. Then f is open and
discrete, this means that for any b ∈ S the fiber f −1 ({b}) is a discrete subset in
R.
Proof. We already proved that f is open in the proof of the previous theorem.
Suppose f is not discrete, then there exist b ∈ S such that f −1 ({b}) is a set with
an accumulation point, From identity theorem it follows that f is constant.
Corollary 1.1.5 (Maximum principle). Suppose R is a Riemann surface and
f : R → C is a non-constant holomorphic function. Then the absolute value of
f does not attain its maximum.
Proof. Let M = sup{|f (x)| : x ∈ R} and let BM be the closed ball of radius M
in C, since f (R) is open f (R) ∩ ∂B = ∅ therefore the value M is not attained at
any point in R.
Corollary 1.1.6. Suppose R and S are two Riemann surfaces with R compact.
If f : R → S is an holomorphic mapping, then S is compact and f is surjective.

1.1.2 Covering maps


Definition Let X, Y and Z be three topological spaces and let p : X → Z,
f : Y → Z be two continuous maps. A lifting of f with respect to p is a map
g : Y → X such that f = p ◦ g.
Definition Suppose X and Y are topological spaces. A surjective mapping
p : Y → X is called a covering map if the following property holds.
Every point x ∈ X has an open neighborhood U such that its preimage
p−1 (U ) can be represented as
[
p−1 (U ) = Vj
j∈J

where the Vj , are disjoint open subsets of Y , and all the mappings P |Vj : Vj → U
are homeomorphism. The space Y is called a covering space of X.
Remark A covering map is a local homeomorphism.
Given a connected topological space X, there is one covering space of X, which
satisfies an universal property. This space is called the universal covering of X.
Definition Suppose X and Y are connected topological spaces and p : Y → X
is a covering map. p : Y → X is called the universal covering of X if it satisfies
the following universal property. For every covering map q : Z → X, with Z
connected, and every choice of points y0 ∈ Y , z0 ∈ Z with p(y0 ) = q(z0 ) there
exists exactly one continuous fiber-preserving mapping f : Y → Z, i.e a map
such that q ◦ f = p, that satisfies f (y0 ) = z0 .
4 Chapter 1. UNIFORMIZATION THEOREM

Remark A connected topological space X has at most one universal covering,


up to homeomorphism.
If p : Y → X and q : Z → X are two universal covering of X, there exists z0 ∈ Z
and y0 ∈ Y such that q(z0 ) = p(y0 ). Since both q and p are universal covering
maps there exist two functions f : Y → Z and g : Z → Y with f (y0 ) = z0 and
g(z0 ) = y0 , such that q ◦ f = p and p ◦ g = q.
If we consider the map f ◦ g : Z → Z, we obtain the equation q ◦ f ◦ g = q,
therefore f ◦ g is a fiber-preserving mapping, moreover f ◦ g(z0 ) = z0 . But
the map idZ : Z → Z has the same two properties, therefore by universality
f ◦ g = idZ . In the same way g ◦ f = idY . This proves that Y and Z are
homeomorphic.

Let’s now consider the case in which X and Y are two connected Hausdorff
spaces.
Proposition 1.1.7. Suppose X and Y are Hausdorff spaces and p : Y → X is a
covering map. Furthermore, suppose Z is a simply connected, pathwise connected
and locally pathwise connected topological space and f : Z → X is a continuous
mapping. Then for every choice of points z0 ∈ Z and y0 ∈ Y with f (z0 ) = p(y0 )
there exists precisely one lifting fˆ : Z → Y such that fˆ(z0 ) = y0 .

A continuous manifold X is an Hausdorff space such that every point of it


has an open neighborhood homeomorphic to Rn . If a continuous manifold X
is connected then it is also pathwise connected and locally pathwise connected.
Moreover if X is a continuous manifold and p : Y → X is a covering of X, then
Y is a continuous manifold too.
Theorem 1.1.8. Suppose X and Y are connected continuous manifolds, such
that Y is simply connected and p : Y → X is a covering map. Then p is the
universal covering of X.
Proof. This follows directly from the definition and from Theorem (1.1.7).
It remains to prove that given an arbitrary Riemann surface it admits an
universal covering. We proved that a simply connected covering space is a
universal covering space, hence we only have to check that there exists a simply
connected covering.
Theorem 1.1.9. Suppose X is a connected manifold. Given two points x0 , x ∈ X
the set π(x0 , x) denotes the set of homotopy classes of curves having initial point
x0 and end point x. Let

X̃ := {(x, α) : x ∈ X, α ∈ π(x0 , x)}.

Define the mapping p : X̃ → X by p(x, α) := x then there exist a topology on X̃


in which it becomes a simply connected continuous manifold and in which p is a
covering map
1.2. Uniformization theorem 5

1.2 Uniformization theorem


First of all we would like to endow the universal covering of the manifold R with
a complex structure.
Theorem 1.2.1. Suppose X is a Riemann surface, Y is a Hausdorff topological
space and p : Y → X is a local homeomorphism. Then there is a unique complex
structure on Y such that p is holomorphic.
Proof. Let ϕ : U → V ⊂ C be a chart of the complex structure of X. We may
assume there exists an open subset U 0 such that p|U 0 : U 0 → U a homeomorphism.
Hence ϕ0 = ϕ ◦ p : U 0 → V is a complex chart on Y . Let U be the set of all
complex charts on Y obtained in this way. U turns out to be a compatible atlas
for Y .
If Y is endowed with this atlas, the projection p is locally biholomorphic and
therefore holomorphic.
Uniqueness may be proved as follows. Suppose we have another complex structure
U 0 on Y such that the mapping p : (Y, U 0 ) → X is holomorphic and thus locally
biholomorphic. The identity mapping (Y, U ) → (Y, U 0 ) is locally biholomorphic
and therefore it is a holomorphic mapping. Hence the two atlas define the same
complex structure.
Since the universal covering map J is a local homeomorphism, R̂ admits a
complex structure such that it is a simply connected Riemann surface and J is a
locally biholomorphic map.
The following theorem was proved by Poincaré and Koebe. It gives a de-
scription of all the simply connected Riemann surfaces. This Theorem is a
generalization of the Riemann mapping theorem. Since its proof is very long and
difficult, we skip it and refer the interested reader to the following article [4].
Theorem 1.2.2 (Uniformization theorem-simply connected case). Every simply
connected Riemann surface is conformally equivalent to the upper half plane, the
complex plane, or the Riemann sphere
Remark The three simply connected surfaces presented in the theorem are
not conformally equivalent one with another. As a matter of fact the sphere is
not even homeomorphic to the complex plane and the upper half plane, while
Liouville theorem implies there is no surjective map between the complex plane
and the unit disk (which is conformally equivalent to the upper half plane).

Let’s study the automorphism groups of H, C and Ĉ


Lemma 1.2.3.
(i) Every element of Aut(Ĉ) has the form
az + b
γ(z) = ,
cz + d
where a, b, c, d ∈ C and ad − bc = 1.
(ii) Every element of Aut(C) has the form
γ(z) = az + b,
where a, b ∈ C and a 6= 0.
6 Chapter 1. UNIFORMIZATION THEOREM

(iii) Every element of Aut(H) has the form


az + b
γ(z) = ,
cz + d
where a, b, c, d ∈ R and ad − bc = 1.
Proof. Firstly we determine the form of an element γ ∈ Aut(Ĉ). If γ(∞) =
∞ then the function γ does not have any other pole in C. Moreover γ is
meromorphic and therefore it is a polynomial of degree one. If γ(∞) = z0 6= ∞,
we put γ1 (z) = 1/(z − z0 ). Now γ1 ◦ γ ∈ Aut(Ĉ) and γ1 ◦ γ(∞) = ∞. Thus
γ1 ◦ γ(z) = a1 z + b1 , with a1 =
6 0. Therefore γ(z) is a Möbius transformation
and it is possible to choose the coefficients in order to have ad − bc = 1.
We already proved the second point so it remains to find the Automorphism
group of the upper half plane. It is convenient to study first the automorphism
group of the unit disk ∆ and then the one of H.
Let γ ∈ Aut(∆) be such that γ(0) = β. The Möbious transformation
z−β
γ1 (z) =
1 − β̄z
belongs to Aut(∆). Hence γ1 ◦ γ ∈ Aut(∆) and it fix the origin. Schwarz’s
lemma implies it is a rotation, therefore γ1 ◦ γ(z) = eiθ z. Therefore γ has the
form
az + b
γ(z) =
b̄z + ā
with |a|2 − |b|2 = 1. Since all this transformations are automorphism of ∆ we
found the automorphism group of ∆.
Finally we consider the biholomorphic map T : H → ∆
z−i
T (z) = .
z+i
If γ ∈ Aut(H), then γ1 = T ◦ γ ◦ T −1 ∈ Aut(∆). This proves that γ is a Möbious
transformation. Then
az + b
γ = T −1 ◦ γ1 ◦ T = ,
cz + d
with a, b, c, d ∈ R. Hence we choose the coefficients in order to have that
ad − bc = 1.
The following Lemma will give more information on the structure of the
group Deck(J).
Lemma 1.2.4. Given a Riemann surface R and its universal covering J : R̂ → R
the Deck transformation group, i.e.
n o
Deck(J) := γ ∈ Aut(R̂)| J ◦ γ = J ,

satisfies the following properties:


(i) For any p, q ∈ R̂ with J(p) = J(q) there exist a unique element γ ∈ Deck(J)
such that γ(p) = q.
1.2. Uniformization theorem 7

(ii) For every p ∈ R̂ there exists a neighborhood U 3 p such that γ(U ) ∩ U = ∅


for all γ ∈ Deck(J) \ {id}. In particular every element in Deck(J) \ {id}
has no fixed points.
(iii) Deck(J) acts properly discontinuously on R̂, i.e. for any compact subset
K of R̂, there are at most finitely many elements γ ∈ Deck(J) such that
γ(K) ∩ K 6= ∅.
Proof. (i) follows from the definition of universal covering space.
In order to prove (ii), let’s take a point p ∈ R̂ and let’s consider p0 = J(p).
Since J is a covering map, there exists a neighborhood U 0 3 p0 such that its
preimage is a collection of disjoint open sets and the restriction of J to those
sets is a biholomorphism. Let U be the element of this collection containing p.
Let’s suppose that there exists γ ∈ Deck(J) \ {id} such that γ(U ) ∩ U 6= ∅. Take
p1 = γ(q1 ) in the intersection, since J|U is a biholomorphism and J = J ◦ γ it
follows that p1 = q1 . Hence from the definition of universal covering, γ = id.
Finally in order to verify (iii) assume that there exists a sequence {γn }∞ n=0
consisting of mutually distinct elements of Deck(J) such that γn (K) ∩ K 6= ∅ for
all n. For every n there exists a couple of point qn , rn ∈ K such that rn = γn (qn ).
Since K is compact we may assume that rn → r0 and qn → q0 .
Since J(qn ) = J(rn ) , from the continuity of J it follows that J(q0 ) = J(r0 ).
Take a neighborhood U 0 of J(q0 ) such that J −1 (U 0 ) is a collection of disjoint
open set, each one biholomorphic to U 0 . Denote with U and V the connected
components of J −1 (U 0 ) containing qo and r0 respectively. Since {γn (qn )}∞ n=0
converges to r0 , for n big enough γn (U ) ∩ V 6= ∅. Since J ◦ γn (U ) = U 0 it follows
that γn (U ) = V . Therefore γn (U ) = γn+1 (U ) which is in contradiction with the
point (ii).
The Uniformization theorem for a generic Riemann surface is the obvious
consequence of the machinery we built in the previous sections. We can state it
in the following way
Theorem 1.2.5. Let R be a Riemann surface and let J : R̂ → R is the universal
covering map of R. The Riemann surface R is conformally equivalent to

R̂/Deck(J).

Proof. In order to prove the theorem we only have to endow R̂/Deck(J) with a
complex structure.
Given a point p̂ ∈ R̂, we take a neighborhood Up̂ 3 p̂ such that γ(Up̂ ) ∩ Up̂ = ∅
for every element in Deck(J) \ {id}. One may assume that there exists local
coordinate zp̂ on Up̂ . Hence, let π : R̂ → R̂/Deck(J) be the projection, if we put
p = π(p̂), Up = π(Up̂ ) and zp = zp̂ ◦ π −1 we conclude that {(Up , zp )}p∈R̂/Deck(J)
defines a complex structure on R̂/Deck(J) and the map

[p] 7→ J(p),

is a well defined biholomorphic map.

Fundamental domain Given a Riemann surface R and its universal covering


map J : R̂ → R, an open set of H is a fundamental domain if it satisfies the
following conditions:
8 Chapter 1. UNIFORMIZATION THEOREM

(i) γ(F ) ∩ F = ∅ for every γ ∈ Deck(J) with γ 6= id;


(ii) If F̄ is the closure of F in H, then
[
H= γ(F̄ ).
γ∈Deck(J)

(iii) The relative boundary ∂F of F has measure zero with respect to the
two-dimensional Lebesgue measure.

In order to find a fundamental domain for the surface R we proceed in the


following way. First we cut smooth path on R to get a simply connected domain
R0 . Let F be a connected component of the inverse image J −1 (R0 ). By the
construction of the universal covering we see that F is a fundamental domain
for Deck(J).

1.2.1 Fuchsian groups


Definition Given a subset G ⊆ P SL(2, R), a point z ∈ Ĥ is a limit point of G
if there exist w ∈ H and a subsequence gn in G such that gn (w) → z. The limit
set of G is the set of all limit points and it is denoted by Λ(G)

Fuchsian group Let Γ be a subgroup of P SL(2, R) then the following are


equivalent
(i) Γ is discrete;
(ii) Γ acts properly discontinuously on H.
A subgroup which satisfies one of those three property is called Fuchsian. A
Fuchsian group is of the first kind if Λ(Γ) = R̂.

The proof of the equivalence between these equivalent definitions can be found
in [9].

Lemma 1.2.6. The limit set Λ(Γ) of a Fuchsian group is closed in R̂.
Proof. We say that the group Γ acts discontinuously at a point x if there exist
a neighborhood of the point U such that {g ∈ Γ|U ∩ g(U ) 6= ∅} is finite. It
is easy to see that if a group acts properly discontinuously on H then it act
discontinuously at every point in H. Moreover the set of point in Ĉ where the
action is discontinuous is open.
It is a simple exercise to check that

z ∈ Λ(Γ) ⇔ Γ acts continuously at z.

Therefore Λ(Γ) is a closed set.

Definition A family of holomorphic function F , defined on a domain Ω, is


normal if every sequence of function fn in F admits a subsequence which
converges uniformly on compact sets of Ω.

The interested reader can found the proof of the following theorem in [17].
1.2. Uniformization theorem 9

Theorem 1.2.7 (Montel). A F of holomorphic functions defined on a domain


Ω which omits two points in C is normal
Lemma 1.2.8. Let γn be a sequence of element in P SL(2, R) which converges
uniformly on compact subsets of H to a holomorphic function f defined in H.
Here f can be a constant function with value ∞. Then either one of the following
holds:

(i) f is an element of P SL(2, R);

(ii) f is a constant function c with c ∈ R̂.


Proof. Since H and D are biholomorphic we can replace H with D and P SL(2, R)
with Aut(D). Clearly |f | ≤ 1, then for the maximum principle either |f | < 1 or
f = c with |c| = 1. In the first case from Montel’s theorem it follows that γn−1 is
a normal family. Hence there exist a subsequence of it that converges uniformly
on compact subset of D to a holomorphic function g, which turns out to be the
inverse of f . Hence f ∈ Aut(D)
Proposition 1.2.9. Let Γ = Deck(J) be a Fuchsian group of the first kind.
For an arbitrary point ζ ∈ R̂, there exists a sequence γn ∈ Γ such that γn (z0 )
converges to ζ for any z0 ∈ H.
Proof. Since Γ is a Fuchsian group of the first kind, given ζ ∈ R̂, there exists a
sequence γn and z ∈ H such that γn (z) → ζ. The sequence γn is normal, hence
if we restrict to a subsequence, γn converges uniformly on compact subsets of H
to an holomorphic function f defined in H. From lemma above f must be the
constant function ζ.
10 Chapter 1. UNIFORMIZATION THEOREM
Chapter 2

Liouville equation

Liouville equation was first studied by Poincaré during the last two decades
of the XIX century. The main difficulty related to this equation is related to
the problem of finding some quantities called accessory parameters. If those
parameters are known, their knowledge allows us to find an explicit uniformizing
map for the given Riemann surface.

2.1 Liouville equation


Let R be a Riemann surface. Let’s suppose that the universal covering space of
R is H and let J : H → R be the universal covering map. In this case we say
that R is an hyperbolic Riemann surface. Firstly we consider the case in which
R ⊂ Ĉ.
Let
ds2 = e2ϕ dzdz̄
be a metric on the surface, where ϕ : R → R.
We want to find a function ϕ such that the universal covering map J is a
local isometry between H, endowed with the standard metric dzdz̄/(=z)2 and
the manifold R.
The standard metric on H has constant curvature K = −1, hence the metric ds2
has the same curvature. Therefore Liouville equation holds

∆ϕ = e2ϕ , (2.1)

more information about Liouville equation can be found in [5].


If the Riemann surface R is not contained in C, if {(Uα , zα )} is an atlas of R,
on Uα then any metric on R can be written as

ds2 |Uα = e2ϕα (zα ,z̄α ) dzα dz̄α

for some ϕα : zα (Uα ) → R.


In this case if gαβ : zβ (Uα ∩ Uβ ) → zα (Uα ∩ Uβ ) is the transition function
zα = gαβ (zβ ), we obtain that
0
ds2 |Uα ∩Uβ = e2(ϕα ◦gαβ (zβ ,z̄β )+|gαβ (zβ )|) dzβ dz̄β

11
12 Chapter 2. LIOUVILLE EQUATION

It follows that on zβ (Uα ∩ Uβ )


0
ϕβ (z, z̄) = ϕα ◦ gαβ (z, z̄) + log |gαβ (z)|.
Definition Given an hyperbolic Riemann surface R and an atlas {(Uα , zα )}α∈I ,
a solution of Liouville equation is a set of function {ϕα }α∈I with ϕα : zα (Uα ) → R
such that
(i)
∆ϕα (z, z̄) = e2ϕα (z,z̄) z ∈ zα (Uα ).
(ii)
0
ϕβ (z, z̄) = ϕα ◦ gαβ (z, z̄) + log |gαβ (z)| on zβ (Uα ∩ Uβ ).
This collection of function induce a metric ds2 , such that
ds2 |Uα = e2ϕα (zα ,z̄α ) dzα dz̄α ,
the metric has constant curvature K = −1.
Remark If we replace the atlas {(Uα , zα )} with a compatible atlas {(Uα0 , zα0 )},
any solution of Liouville equation relative to the old atlas corresponds to a
solution in the new atlas. This correspondence is obtained using the connection
formula of the definition. This implies that the choice of the atlas of R is
irrelevant.
Proposition 2.1.1. Let R be an hyperbolic Riemann surface with an atlas
{(Uα , zα )}α∈I . For any α, given a point p ∈ Uα , we consider a branch of J −1
defined in a neighborhood Vp ⊆ Uα of the point. Then the collection of functions
|(J −1 ◦ zα−1 )0 (z)|
ϕα,R (z, z̄) = log z ∈ zα (Uα ), (2.2)
=(J −1 ◦ zα−1 (z))
is a solution of Liouville’s equation on R and does not depend from the particular
choice of J −1 . The collection of functions {ϕα,R }α∈I is called the canonical
solution of Liouville equation.
Proof. Let ds2H be the standard metric of H, the pullback of ds2H |J −1 (Vp ) via
J −1 ◦ zα−1 doesn’t depend from the particular choice of J and of the branch of
J −1 . As a matter of fact for any other possible J˜−1 there exist γ ∈ P SL(2, R)
such that J˜−1 = γ ◦ J −1 . Since γ is an isometry of (H, ds2 )
(J˜−1 ◦ zα−1 )∗ ds2H |J˜−1 (Vp ) = (J −1 ◦ zα−1 )∗ γ ∗ ds2H |γ◦J −1 (Vp ) = (J −1 ◦ zα−1 )∗ ds2H |J −1 (Vp ) .
Thus the function ϕα,R is well defined at any point p ∈ Uα and it doesn’t depend
from the choice of J −1 .
With a direct calculation we find that ∆ϕα,R = e2ϕα,R .
Finally on Uα ∩ Uβ we obtain that
|(J −1 ◦ zβ−1 )0 (zβ )|
ϕβ,R (zβ , z̄β ) = log
=(J −1 ◦ zβ−1 (zβ ))
|(J −1 ◦ zβ−1 ◦ gβα )0 (zα )| 0
= log − log |gβα (zα )|
=(J −1 ◦ zβ−1 ◦ gβα (zα ))
0
= ϕα,R ◦ gαβ (zβ , z̄β ) + log |gαβ (zβ )|
2.1. Liouville equation 13

Proposition 2.1.2. Let R be an hyperbolic Riemann surface, let {(Uα , zα )}α∈I


be an atlas of R and let {ϕα }α∈I be a solution of Liouville Equation. For every
α ∈ I we define the holomorphic function ωα : zα (Uα ) → C as

ωα = ∂z2 ϕα − (∂z ϕα )2 .

On zβ (Uα ∩ Uβ ) the connection relation

1
0
ωβ (z) = ωα ◦ gαβ (gαβ )2 + S (gαβ ),
2
holds true, where S (f ) is the so-called Schwartzian derivative
2
f 000 f 00

3
S (f ) := 0 − .
f 2 f0

{ωα }α∈I is a holomorphic projective connection. The connection {ωα,R }α∈I


coming from the canonical solution of Liouville equation (2.2) is called the
canonical connection of the Riemann surface .

Proof. We prove first that ωα is holomorphic for every α. Since φα satisfies (2.1)
it follows that

∂z̄ ωα = ∂zzz̄ ϕα − 2∂z ϕα ∂zz̄ ϕα


1 1
= ∂z ( e2ϕα ) − 2∂z ϕα e2ϕα
4 4
= 0.

On zβ (Uα ∩ Uβ ) we know that


0
ϕβ = ϕα ◦ gαβ + log |gαβ (z)|,

hence
000  00 2 !
0 2 00 1 gαβ gαβ
∂zz ϕβ = ∂zz ϕα · (g ) + ∂z ϕα · g
αβ αβ + 0
− 0
2 gαβ gαβ
 00 2
0 00 1 gαβ
(∂z ϕβ )2 = (∂z ϕα )2 (gαβ )2 + ∂z ϕα · gαβ + 0
.
4 gαβ

In the end we obtain that


1
0
ωβ = ωα ◦ gαβ (gαβ )2 + S (gαβ ).
2

Remark The canonical solution of Liouville equation for the upper half plane is

ϕH (z) = − log(=(z))

In this case
ωH = 0.
14 Chapter 2. LIOUVILLE EQUATION

Suppose that the canonical connection {ωα,R }α∈I of the hyperbolic Riemann
Surface R is known, then it is possible to find a differential equation connecting
the inverse of the universal covering map J −1 and {ωα,R }.

Proposition 2.1.3. Let R be an hyperbolic Riemann surface, let {(Uα , zα )}α∈I


be an atlas of R and let {ωα,R }α∈I be the canonical connection of R. The following
differential equation is true for every α ∈ I

S (J −1 ◦ zα−1 ) = 2ωα,R z ∈ zα (Uα ).

Proof. The canonical solution of Liouville equation can be written as

ϕα,R (z, z̄) = ϕH (J −1 ◦ zα−1 (z)) + log |(J −1 ◦ zα−1 )0 (z)|.

Elementary calculations, similar to the ones of the previous proof give us that
1
ωα,R (z) = ωH ◦ J −1 ◦ zα−1 (z) · [(J −1 ◦ zα−1 )0 (z)]2 + S (J −1 ◦ zα−1 ).
2
Since ωH = 0 we obtain the desired result.

The problem of finding a solution of a differential equation of the form


S (f ) = Q is reduced to a simpler problem thanks to the following proposition.

Proposition 2.1.4. Let u1 and u2 be two independent solution of the equation


1
u00 (z) + Qu = 0.
2
u1
The function f = u2
satisfies the equation

S (f ) = Q.

Proof. One easily obtain that

u01 u1 u0
f0 = − 22
u2 u2
0
2u
f 00 = − 1 f 0
u2
(u0 )2
f = Qf 0 + 6 22 f 0 .
000
u2

If we evaluate the Schwartzian derivative of f we obtain the desired result.

The canonical projection {ωα,R } determines the second order differential


equation on the Riemann surface R

u00α + ωα,R uα = 0.

The solution to this equation is a (multi-valued) differential of order −1/2, i.e.


a collection of functions {uα }α∈I , with uα : zα (Uα ) → C, together with the
connection formula
0
uβ (z) = uα ◦ gαβ (z) · (gαβ (z))−1/2 .
2.2. Monodromy and deck transformation group 15

The formal expression


u = uα (zα )(dzα )−1/2 ,
shows why it is called a differential of order −1/2.
One can notice that if u is a solution of the second order differential equation,
on zβ (Uα ∩ Uβ )
1
u00β = u00α ◦ gαβ (gαβ
0 0
)3/2 − uα ◦ gαβ (gαβ )−1/2 S (gαβ ).
2
Therefore

u00β + ωβ,R uβ = (u00α + ωα,R uα ) ◦ gαβ (gαβ


0
)3/2
= 0.

2.2 Monodromy and deck transformation group


Given an hyperbolic Riemann surface R endowed with an atlas {(uα , zα )}α∈I ,
we observed that the canonical projection {ωα,R }α∈I determines a second order
differential equation, and that it’s solution is a differential of the form.

u = uα (zα )(dzα )−1/2 .

Let’s fix a point p0 ∈ R and let’s consider a closed curve η with base point p0 .
Since p0 ∈ Uα for some α, we can find two independent solution of the equation
u00α + ωα,R uα = 0 defined in a neighborhood of zα (p0 ). Therefore we can find two
differentials of order −1/2, u1 and u2 which solve the second order differential
equation near p0 .
One can move the differentials u1 , u2 on the curve η. In particular when the
0
curve goes to another coordinate open set Uβ , uα → uβ = uα ◦ gαβ (gαβ )−1/2 and
00
we know that uβ solves the equation uβ + ωβ,R uβ = 0 on zβ (Uβ ).
When η comes back to p0 ũ1 , ũ2 will be a linear combination of u1 and u2

ũ1 = au1 + bu2


ũ2 = cu1 + du2 .
 
a b
The matrix is called the Monodromy matrix relative to the path η.
c d η

One can prove that the Monodromy matrix doesn’t depend from the path
itself but only from the equivalence class in π1 (p0 , R).
We define the Monodromy group
(  )
a b

M on(R) := ∈ M at(2 × 2, R) η ∈ π1 (p0 , R) .

c d η

Lemma 2.2.1. Given a hyperbolic Riemann surface R and a point p0 ∈ R, the


monodromy group M on(R) is a subgroup of SL(2, R).
Proof. Given a closed curve η : [0, 1] → R, we define the function

w(t) = W (u1,α , u2,α ) ◦ zα ◦ η(t), for η(t) ∈ Uα ,


16 Chapter 2. LIOUVILLE EQUATION

here the function W (u, v) = u0 v − uv 0 is the Wronskian function.


Firstly we have to prove that w(t) is well defined. As a matter of fact if
η(t) ∈ Uα ∩ Uβ we obtain that

u01,β u2,β − u1,β u02,β = (u01,α u2,α − u1,α u02,α ) ◦ gα,β .

If we take the derivative of w(t) we easily see that w0 (t) = 0, but we know that
when t = 1, u1 , u2 are transformed into

ũ1 = au1 + bu2


ũ2 = cu1 + du2 .

Hence we obtain that


w(1) = (ad − bc)w(0).
0
Since w (t) = 0 and w(0) 6= 0, then ad − bc = 1, which means that M on(R) ⊆
SL(2, R).
Theorem 2.2.2. Let R be an hyperbolic Riemann surface. If we identify every
element of M on(R) with its corresponding Möbius transformation, the group we
find coincides, up to a conjugation with an automorphism of H, with Deck(J).
Proof. If J is the covering map of R, and {(Uα , zα )} is an atlas of R, on zα (Uα )
we have that
S (J −1 ◦ zα−1 ) = 2ωα,R .
given p0 ∈ Uα , we consider u1 and u2 two independent solutions of the equation

u00α + ωα,R uα = 0,

defined in a simply connected neighborhood U ⊆ Uα . We recall that the solutions


of this equation are differentials u = uα (zα )(dzα )−1/2 .
If we consider the function f := u1 /u2 on U , it follows from Proposition
(2.1.4) that S (f ◦ zα−1 ) = 2ωα,R on zα (U ). If we take any branch of J −1 defined
in U , using the composition property of the Schwarzian derivative, we find that,
on J −1 (U )

S (f ◦ J) = S (f ◦ zα−1 ◦ zα ◦ J)
= S (f ◦ zα−1 ) ◦ zα ◦ J(zα ◦ J)02 + S (zα ◦ J)
= 2ωα,R ◦ zα ◦ J(zα ◦ J)02 − S (J −1 ◦ zα−1 ) ◦ zα ◦ J(zα ◦ J)02
= 0.

Therefore f ◦ J is a linear fractional transformation of J −1 (U ), i.e.


az + b
f ◦J = with a, b, c, d ∈ C, ad − bc = 1.
cz + d
Moreover, up to taking a linear combination of {u1 , u2 } as basis, we can suppose
that
f ◦ J = Id|J −1 (U ) ,
Moreover since J is a surjective

J ◦ f = Id|U .
2.2. Monodromy and deck transformation group 17

Let’s take a close path η : [0, 1] → R with base point p0 . We take the analytic
continuation of the two solutions u1 and u2 along the path. The ratio u1 /u2 is a
multi-valued analytic function defined in a neighborhood of the curve. For any t
we have that
J ◦ ft = Id|Ut ,

where ft is the analytic continuation of f = u1 /u2 and Ut is the neighborhood


of ηt where it is defined. It is easy to see that ft = u1,t /u2,t .
After a loop around the closed path, the two solutions u1 , u2 are transformed
into     
u1,t=1 a b u1
=
u2,t=1 c d u2
 
a b
where Mη = is the monodromy matrix associated to the path η.
c d
It follows that, on some neighborhood Ũ of p0

au1 + bu2
f1 = = γ ◦ f0 .
cu1 + du2

On Ũ we obtain
J ◦ γ ◦ f = J ◦ f.

Therefore on some neighborhood of f (p0 )

J ◦ γ = J.

The equality is true on the whole H. As a matter of fact we can take any point
w on the upper half plane and we set p = J(w). Now we consider a path η1
joining p0 and p and we put η̃ = η1 · η · η1−1 . We obtain that J ◦ γ = J in a
neighborhood of w. This proves that γ ∈ Deck(J).

On the other hand we must prove that for any γ ∈ Deck(J) there exists a
close path in R which induces a monodromy conjugated with γ.
Let consider a path between w0 ∈ H and γ(w0 ) and let project it on R using the
universal covering map J. In this way we obtain a closed path on the manifold.
We chose two solutions of the differential equation u00α + ωα,R uα = 0 such that
their ratio evaluated in p0 gives w0 . It is easy to check that the monodromy
induced by the closed path corresponds to an element γ̂ ∈ Deck(J) such that

γ(w0 ) = γ̂(w0 ).

This implies that γ̂ ◦ γ −1 (w0 ) = w0 . However the only element of Deck(J) with
a fixed point is the identity, therefore

γ(z) = γ̂(z).

In the end we notice that we choose u1 and u2 arbitrarily, and not in such a
way that f ◦ J = Id, the resulting group of linear fractional transformations
induced by the monodromy of the differential equation, is equal to Deck(J) up
to a conjugation with an element of P SL(2, C).
18 Chapter 2. LIOUVILLE EQUATION

2.3 Punctured Riemann surfaces


Definition A Riemann surface with punctures is a Riemann Surface R such
˜ such that every Ui is
that there exist a sequence of open set Ui ⊂ R, with i ∈ I,
conformally equivalent to the punctured disk

zi : Ui −→ D \ {0} = {z ∈ C| 0 < |z| < 1}

Given a Riemann surface with punctures, we can extend the homeomorphism zi


to the whole D.
ẑi : Ûi = Ui ∪ pi → D,

where ẑi (pi ) = 0.


A complex atlas for the Riemann surface
[
R̂ = R ∪ {pi }
i

is given by the union of the atlas A of R together with the charts (Ûi , ẑi ).

Definition A compact Riemann surface with punctures is a Riemann surface


with punctures such that R̂ is compact.

2.3.1 Properties of deck transformation group


If we consider a punctured hyperbolic Riemann surface and we take its universal
covering J : H → R, the Deck transformation group Deck(J) satisfies some
remarkable properties.
All Möbius transformations in P SL(2, R) can be classified in the following
way,

Definition Let γ be a real Möbius transformation which is not the identity. We


say that

(i) γ is parabolic if and only if it has a sole fixed point on R̂.

(ii) γ is elliptic if and only if it has two distinct fixed points z0 , z1 such that
z¯0 = z1 .

(iii) γ is hyperbolic if and only if it has two distinct fixed points on R.

An elementary calculation shows that

Proposition 2.3.1. Let γ be a real Möbius transformation which is not the


identity. Then

(i) γ is parabolic if and only if (a + d)2 = 4.

(ii) γ is elliptic if and only if 0 ≤ (a + d)2 < 4.

(iii) γ is hyperbolic if and only if (a + d)2 > 4.


2.3. Punctured Riemann surfaces 19

Theorem
S 2.3.2. Let R be an hyperbolic punctured Riemann surface R = R̂ \
i {p i }. Given a puncture pi and zi : Ui → D\{0}, let Hi be one of the connected
components of J −1 (Ui ), then Hi is simply connected.
If γ ∈ Deck(J) either γ(Hi ) = Hi or γ(Hi ) ∩ Hi = ∅. Moreover the subgroup
of Deck(J) given by

Γi = {γ ∈ Deck(J)|γ(Hi ) = Hi }

is cyclic and it is generated by a parabolic element of Deck(J).


Proof. Let Hi be one of the connected components of J −1 (Ui ). Let’s consider a
domain Ω ∈ H such that its boundary ∂Ω is a closed curve in J −1 (Ui ). Then
J(Ω) ⊂ Ui , if this was not true, there would be a sequence of points {zn } ∈ Ω
such that |zi ◦ J(zn )| → 1 but this is impossible since

sup |zi ◦ J(z)| ≤ sup |zi ◦ J(zn )| = ρ < 1,


z∈Ω z∈∂Ω

by the maximum modulus principle.


Therefore Ω ⊂ Hi . Thus Hi has no holes, hence it is simply connected.
Let γ ∈ Deck(J). From theorem (3.2.1) it follows that γ corresponds to an
element of the monodromy group of the equation u00α + ωα,R uα = 0. Let z0 ∈ Hi ,
if η is a curve connecting z0 and γ(z0 ), then J ◦ η is a closed curve in R. Now if
J ◦ η is homotopic to a curve entirely contained in Ui we can join z0 and γ(z0 )
with a curve η 0 contained in Hi , therefore γ(z0 ) ∈ Hi . Otherwise, if J ◦ η is not
homotopic to a curve contained in Ui , then γ(z0 ) ∈ / Hi .
It is easy to prove that this fact doesn’t depend from the choice of z0 but only
from the choice of γ, therefore either γ(Hi ) ∩ Hi = ∅ or γ(Hi ) = Hi . The last
equality follows from the fact that if γ(Hi ) ⊂ Hi then also γ −1 (Hi ) ⊂ Hi .
Since zi ◦ J : Hi → D \ {0} is a covering map and Hi is simply connected,
there exists a biholomorphic map ϕ : H → Hi such that

zi ◦ J ◦ ϕ(z) = e2πiz ,

as a matter of fact the map J˜ : H → D \ {0} given by J(z)


˜ = e2πiz is the universal
covering map of D \ {0}, hence by the definition of universal covering there exists
such ϕ.
Let γ be an element of Deck(J) such that γ(Hi ) = Hi , the map γ̃ = ϕ−1 ◦ γ ◦ ϕ
belongs to Deck(J).˜ We obtain the following commutative diagram

ϕ zi ◦J
H Hi D \ {0}
γ̂ γ
zi ◦J
ϕ
H Hi

Therefore the group

Γi = {γ ∈ Deck(J)| γ(Hi ) = Hi }
˜ Now Γi is cyclic and generated by a parabolic
is conjugated with Deck(J).
˜
element since also Deck(J) has those properties.
20 Chapter 2. LIOUVILLE EQUATION

Remark Once we fix the puncture pi it is possible to choose an universal


covering such that the group Γi = hz + 1i.
As a matter of fact if γ0 ∈ P SL(2, R), then J ◦ γ0 is another universal covering
of the Riemann surface, and Deck(J ◦ γ0 ) is equal to γ0 ◦ Deck(J) ◦ γ0−1 . 
a b
Let γ be the parabolic Möbius transformation that generates Γi and let
c d
be the corresponding element in SL(2, R) having trace a + d = 2. In order to
find an element γ0 such that γ0 ◦ γ ◦ γ0−1 (z) = z + 1, we must solve the equation
     
x y a b 1 1 x y
= ,
z w c d 0 1 z w
which is equivalent to the linear system
    
a−1 c −1 0 x 0
 b
 d − 1 0 −1   y  0
  =  .
 0 0 a−1 +c   z  0
0 0 b d−1 w 0
This system has determinant det(A) = (ad − bc − d − a + 1)2 = 0, thus such a
γ0 exists.
Let’s now study the n-punctured sphere. First of all if n ≥ 3 we can choose
the holes such that three of them are 0, 1, ∞. This can be done using a Möbius
transformation that send three of the holes into 0, 1, ∞.
Corollary 2.3.3. Let X = C \ {z1 , . . . , zn−3 , 0, 1} and J : H → X be its
universal covering. There exists a system of parabolic generators S1 , . . . , Sn
which satisfies the single relation S1 · · · Sn = 1 and with fixed points ẑ1 , . . . , ẑn .
Under a conjugation in P SL(2, R) we will assume that ẑn−2 = 0, ẑn−1 = 1 and
ẑn = ∞.
The elements S1 , . . . , Sn are represented in the form
   
1 + λi ẑi −λi ẑi2 1 λn
Si = and Sn =
λi 1 − λi ẑi 0 1
with λi ∈ R.
Proof. If the element Si corresponds to the monodromy matrix of a loop around
the point zi , we know that Si is parabolic and that that the set of Si generates
Deck(J), since π1 (X, p0 ) ∼ M on(X) ∼ Deck(J) is generated by loops around
the missing points zi . The relation S1 · · · Sn = 1 comes from the fact that
π1 (X, po ) ∼ Deck(J).

2.3.2 Canonical connection of a punctured Riemann sur-


face
Let R be a punctured Riemann surface, we consider an atlas of R such that it
contains the charts zi : Ui → D \ {0} i ∈ I˜ corresponding to the punctures. In
this case the canonical connection {ωα,R }α∈I on those charts is a meromorphic
function.
We consider a puncture pi , we may assume that Γi defined previously, is
generated by the parabolic element γ = z + 1, in this case the following lemma
holds true
2.3. Punctured Riemann surfaces 21

Lemma 2.3.4. The domain Hi , defined in Proposition (2.3.2), contains an


upper half plane Hc := {z ∈ C|=(z) > c}.
Proof. Let’s take a vertical line η(t) = c + it, with c ∈ R and t > 0. Since Hi is
connected and Hi + 1 = Hi , the line intersects Hi . Let t0 = inf t {t > 0| η(t) ∈
Hi } and t1 = supt {t ∈ R|η(]t0 , t]) ∈ Hi }. Since Hi is simply connected and
translationally invariant, η(t) ∈/ Hi for t > t1 . We would like to prove that
t1 = ∞ for every c.
Let’s suppose t1 < ∞, we consider the curve E(t) = zi ◦ J ◦ η(t) :]t0 , t1 [→ D \ {0}.
This curve tends to the border of D \ {0} for t → t0 and t → t1 , but the limit
point cannot be z = 0. As a matter of fact if the curve tends to z = 0 for
t → t0 then there exists a sequence wn of point in H, converging to a point
w = η(t0 ) ∈ H, such that J(wn ) tends to the puncture pi , this implies that
J(w) = pi which is absurd. The same is true for t → t1 .
Let’s take a smooth curve ξ(t) ⊂ Hi joining ŵ = η(t̂) and ŵ + 1, where t ∈]t0 , t1 [,
such that the curve has the same direction in the two ending points and doesn’t
intersect η(t) and η(t) + 1. The curve Ξ(t) = zi ◦ J ◦ ξ(t) ⊂ D \ {0} is a C 1 closed
curve and intersect E(t) in an even number of points, this is absurd since it is
supposed to intersect the curve only in zi ◦ J(ŵ). This proves that t1 = ∞, a
similar argument proves that E(t) → 0 for t → ∞.
As a consequence of this, ∂Hi is a continuous periodic curve in H and Hi
lays above this curve. Let c be the maximum value of the imaginary part of the
curve, then Hc ⊂ Hi .

Theorem 2.3.5. Let R be an hyperbolic Riemann surface with punctures. We


consider an atlas {(Uα , zα )} containing the charts zi : Ui → D \ {0} with i ∈ I˜
corresponding to the punctures of R. Every element the subset {ωi,R }i∈I˜ of the
canonical connection is a meromorphic function on zi (Ui ) = D \ {0}. It can be
written as
1 ci
ωi,R (z) = 2 + + h(z),
4z 2z
where h(z) is an holomophic function on D.
Proof. From theorem (2.3.2) it follows that for every i ∈ I, ˜ there exist a simply
connected domain Hi which is a connected component of J −1 (Ui ). Moreover we
know that the elements γ ∈ Deck(J) such that γ(Hi ) = Hi form the cyclic group
Γi . We choose J in such a way that Γi = hτ + 1i. Let Hc = {z ∈ C| =(z) > c} be
an upper half plane contained in Hi , the existence of it is guaranteed by previous
lemma.
One can show that the closed curve zi ◦ J(t + ic) contains the origin, therefore
the open set D0 = zi ◦ J(Hc ) ⊂ D \ {0} is such that D = D0 ∪ {0} is simply
connected and its boundary is the curve zi ◦ J(t + ic). Riemann mapping theorem
implies that D is conformally equivalent to the unit disk D. We can choose the
biholomorphism φ : D → D in such a way that φ(0) = 0.
We obtained the following diagram

H H
f =z+ic JD\{0} =exp(2πiz)

φ◦zi ◦J
Hc D \ {0}
22 Chapter 2. LIOUVILLE EQUATION

0
Since JD\{0} is a covering map between H and the punctured disk, by the
definition of universal covering map, there exist a Möbious transformation γ which
closes the diagram above. Let γ 0 = γ ◦ f −1 , then γ 0 is a Möbious transformation.
We have that
zi ◦ J = φ−1 ◦ JD\{0} ◦ γ 0 .
Therefore

2ωi,R = S (J −1 ◦ zi−1 )
= S γ 0−1 ◦ JD\{0}
−1

◦φ
= S JD\{0}
−1

◦φ
= S JD\{0}
−1
◦ φ · (φ0 )2 + S (φ).


One has that


1
ωi (z) = φ0 (z)2 + S (φ)(z)
4φ(z)2

Since φ : D → D is biholomorphic and φ(0) = 0, we can expand it near the


origin. We obtain that
φ(z) = φ0 (0)z + z 2 h(z),
with h(z) holomorphic near the origin and φ0 (0) 6= 0. It remains to observe that
1
ωi,R (z) = (1 + z h̃(z)) + c,
4z 2
with h̃(z) holomorphic and c = S (φ)(0).

Remark Let S (Uα , zα ) be another chart containing one or more puncture. Then
˜ On zα (Ui ∩ Uα ) we know that
zα (Uα ) = Ω \ i∈I 0 {zi } ⊂ C with I 0 ⊂ I.

1
0
ωα,R (z) = ωi,R ◦ giα (z) · (giα (z))2 + S (giα )(z).
2
Since giα is bounded near z = zi , it can be extended to an holomorphic map
from zα (Uα ∩ Ui ) ∪ {zi } to zi (Uα ∩ Ui ) ∪ {0} and giα (zi ) = 0.
Therefore
1 1
ωα,R (z) = (g 0 (z))2 + S (giα )(z)
4giα (z) iα 2
1 ˜
= (g 0 (zi ) + r̃(z)(z − zi ))2 + r̃(z)
4(giα (zi )) (z − zi )2 + (z − zi )4 r(z)) iα
0 2

1 ci
= + + h(z),
4(z − zi )2 2(z − zi )

with h(z) holomorphic near zi .

Let’s consider a punctured complex sphere X = C \ {z1 , . . . , zn−3 , 0, 1}, with


finite punctures. Using a Möbious transformation, we can choose the punctures
in such a way that three of them coincide with 0, 1 and ∞. This Riemann surface
can be described with a single chart, therefore the canonical connection {ωα,X }
is simply a meromorphic function on C.
2.3. Punctured Riemann surfaces 23

Theorem 2.3.6. Let X = C \ {z1 , . . . , zn−3 , 0, 1}. The canonical connection


{ωα,X } is a meromorphic function on C, it can be written as
n−1
1 X 1 ci
ωX (z) = QX (z) = + , (2.3)
2 i=1
4(z − zi ) 2 2(z − zi )

with the asymptotic condition


 
1 cn 1
ωX (z) = 2
+ 3 +o for z → ∞.
4z 2z z3
The {ci } are complex numbers called accessory parameters. They satisfy the
following conditions
n−1 n−1 n−1
X X n X
ci = 0, ci zi = 1 − , zi (1 + ci zi ) = cn (2.4)
i=1 i=1
2 i=1

Proof. Near every puncture ωX (z) has a pole of order two of the type
1 ci
ωX (z) = + + hi (z).
4(z − zi )4 2(z − zi )
Therefore there exist an entire function h(z) such that
n−1
1 X 1 ci
ωX (z) = QX (z) = + + h(z).
2 i=1
4(z − zi ) 2 2(z − zi )

Let’s now study the behavior of ωX at infinity. We consider the map f :


Ĉ\{z1 , . . . , zn−3 , 0, 1, ∞} → X 0 = Ĉ\{1/z1 , . . . , 1/zn−3 , ∞, 1, 0} given by f (z) =
1/z. Let J : H → X and J 0 : H → X 0 be the universal covering maps of X and
X 0.
Since f ◦ J is a covering map, there exist a Möbious transformation γ : H → H
which closes the following commutative diagram.

J
H X
γ f (z)
0
J
H X0

Keeping in mind that 2ωX = S (J −1 ), we write that


2ωX = S (γ −1 ◦ (J 0 )−1 ◦ f )
= S ((J 0 )−1 ◦ f )
= 2ωX 0 ◦ f (f 0 )2 .
The canonical connection of R0 near z = 0 can be written as ωX 0 (z) = 1/(4z 2 ) +
cn /(2z) + ĥ(z), with ĥ(z) holomorphic near the origin. Therefore
 2 
z cn z 1
ωX (z) = + + ĥ(z)
4 2 z4
 
1 cn 1
= 2 + 3 +O .
4z 2z z4
24 Chapter 2. LIOUVILLE EQUATION

This implies that the function h(z) = 0


1
In the end, if we expand f (z) = z−zi
near ∞, we get that

z2
 
1 zi 1
f (z) = + 2 + i3 + O .
z z z z4

If we substitute this into equation (2.3), we find that


Pn−1 n−1
Pn−1 Pn−1  
i=1
ci 2
+ i=1
ci zi i=1
zi (1 + ci zi ) 1
ωX (z) = + 2
+ +O .
2z 2z 2z 3 z4

A comparison with the asymptotic formula for ωX near ∞ gives the desired
relations for the ci .

2.3.3 Deck transformation group of the first kind


Finally we prove that the Deck transformation group of a compact hyperbolic
Riemann surface with finite punctures is of the first kind, i.e. Λ(Deck(J)) = R̂,
where

Λ(Deck(J)) = {z ∈ R̂| ∃z0 ∈ H, ∃γn ∈ Deck(J) s.t. γn (z0 ) → z}.

Firstly, given a Fuchsian group Γ and z0 ∈ H, we define the Dirichlet Region

Dz0 (Γ) = {z ∈ H| ∀γ ∈ Γ dH (z, z0 ) < dH (z, γ z0 )}

Lemma 2.3.7. If R is an hyperbolic Riemann surface and Γ = Deck(J) then


for every z0 ∈ H, the Dirichlet region Dz0 (Γ) is a fundamental domain. It’s
border is made of a countable union of lines Ci , those lines can be either geodesics
of H or intervals on the real line.

Proof. The first two points in the definition of fundamental domain are easily
verified.
In order to verify the last point, we notice that the Dirichlet region can be
written as \
Dz0 (Γ) = Hp (γ),
γ∈Γ\id

where
Hp (γ) = {z ∈ H| dH (z, p) < dH (z, γ p)}.
We easily check that the points in H such that dH (z, p) = dH (z, γ p) form a
geodetic, hence this geodetic divides the upper half plane in two parts and Hp (γ)
is the part containing p. Since Γ is numerable when we intersect all the Hp (γ)
we obtain that ∂Dp (Γ) is made of a countable union of geodesics and intervals
on the real line. Hence the third point in the definition of fundamental domain
is satisfied.

Given a domain Ω ⊂ H we define the hyperbolic area of Ω as


Z
dzdz̄
A(Ω) = dsH , dsH = .
Ω (=z)2
2.3. Punctured Riemann surfaces 25

Lemma 2.3.8. Let R be an hyperbolic compact Riemann surface of genus g and


n punctures. Let Γ = Deck(J), if we consider a point z0 ∈ H, the border of the
Dirichlet region Dz0 (Γ) is only made of of geodesics and the hyperbolic area of
the region is finite.

Proof. Firstly we prove that the border of Dz0 (Γ) doesn’t contain an interval of
the extended real line.
Since the Dirichlet region is a fundamental domain, the analytic covering map
J : Dz0 (Γ) → R can be extend continuously to the border of Dz0 (Γ) ⊂ Ĉ, this
extension turn out to be a bijective map Jˆ between Dz0 (Γ) and the compactifica-
tion R̂ of R, which is a compact Riemann surface of genus g. Let’s suppose that
one of the lines in the border of Dz0 (Γ) is an interval I contained in r̂. Since the
punctures of R are isolated, then J 0 (I) = pi where pi is one of the puncture.
Let’s consider the map zi : Ui → D \ {0}, corresponding to the puncture pi . The
map zi ◦ J defined on Dz0 ∩ J −1 (Ui ) can be extended using Schwartz reflection
principle to an holomorphic function defined on a domain containing I. This
function is constant on I and this is not possible. Therefore the border of the
Dirichlet region doesn’t contain any interval of the extended real line.
We notice that the set of points in which Dz0 (Γ) touches R̂ is in bijection
with the punctures of R. Therefore they are finite. We take ε > 0 and the
subset Hε = {z ∈ H| ε ≤ =z ≤ 1/ε}. It is easy to show that, if we choose ε
small enough, then Hεc ∩ Dz0 (Γ) has a number of connected component equal
to the number of punctures of R. With an elementary calculation we find that
the hyperbolic area of this connected components and the hyperbolic area of
Hεc ∩ Dz0 (Γ) are finite.

Lemma 2.3.9. For every point z0 ∈ H the Dirichlet region Dz0 (Γ) is locally
finite, i.e if K is any compact set contained in H then the number of elements
γ ∈ Γ such that
γ(Dz0 (Γ)) ∩ K 6= ∅
is finite

Proof. Suppose K ∩ γi (Dz0 (Γ)) 6= ∅ for some infinite sequence of γi ⊂ Γ. Let


σ = supz∈K dH (z, z0 ). Since K is compact σ is finite. Let wj ∈ K ∩ γi (Dz0 (Γ))
then wj = γi (zi ) for some zi ∈ Dz0 (Γ). Hence

dH (z0 , γi (z0 )) ≤ dH (z0 , wi ) + dH (wi , γi (z0 ))


= dH (z0 , wi ) + dH (zi , z0 )
≤ dH (z0 , wi ) + dH (wi , z0 ) ( from the definition of Dz0 (Γ))
≤ 2σ.

Thus the infinite set of points γi (z0 ) belongs to the hyperbolic ball of center
z0 and radius 2σ, which contradicts the fact that the action of Γ is properly
discontinuous.

Lemma 2.3.10. Let γn be any sequence in Γ, and F be a locally finite funda-


mental region for Γ. Then the euclidean diameter of γn (F ) goes to zero.

Proof. Let’s assume that there the diameter doesn’t go to zero. Hence there
exists a subsequence γnk and points wk , zk ∈ γnk (F ) such that the diameter of
26 Chapter 2. LIOUVILLE EQUATION

γnk (F ) doesn’t go to zero and wk → w, zk → z.


By local finiteness both z and w belong to R̂. Then γk (F ) accumulates on the
geodesic which connects w and z and this is absurd by local finiteness.
Proposition 2.3.11. If R is a punctured hyperbolic Riemann surface of genus
g with n punctures and J : H → R is its universal covering, then Γ = Deck(J)
is a fuchsian group of the first kind.

Proof. If z0 ∈ H we proved that the Dirichlet region Dz0 (Γ) has finite area.
Let ζ ∈ R and let’s consider an euclidean disk Uε1 of radius ε1 > 0. From the
lemmata above it follows that there exist γ1 ∈ Γ such that γ1 (Dz0 (Γ)) ⊂ Uε1 .
Hence if we take a sequence of εn → 0 we find a sequence of γn ∈ Γ such that
γn (z0 ) → ζ. Hence ζ ∈ Λ(Γ) and since Λ(Γ) is closed in the extended real line,
also ∞ ∈ Λ(Γ).
Chapter 3

The 3-punctured sphere

It is possible to find an explicit form for the universal covering map of the sphere
with three missing point, i.e the domain C \ {0, 1}. In the first section of the
chapter we show that the equation
u00 + ωC\{0,1} u = 0
is equivalent to the so-called hypergeometric equation
Then we introduce the theory of hypergeometric equation. In particular we
find local solutions and we study the monodromy of the equation.
After this we find an analytic continuation of the local solutions of the
hypergeometric equation connected to the uniformization problem of C \ {0, 1}.
In the last part of this chapter we present the modular lambda function and we
prove that it is the universal covering map of C \ {0, 1}.

3.1 Accessory parameters


In the case of C \ {0, 1} we can easily find an expression for the canonical
connection. Using relations (2.4), we find that c0 = 1/2, c1 = −1/2 and
c∞ = 1/2. The equation S (J −1 ) = 2ωC\{0,1} becomes
z2 − z + 1
S (J −1 )(z) = ,
2z 2 (z − 1)2
and the associated O.D.E is a second order rational equation with poles at 0, 1
and ∞. It’s possible to transform this equation into an hypergeometric equation.
The form of the generic hypergeometric equation is the following
z(z − 1)w00 + [c − (a + b + 1)z]w0 − abw = 0 (3.1)
Proposition 3.1.1. For any choice of the parameters a, b, c, all the solutions
of (3.1) are given by
u
w = −c/2 ,
z (1 − z)(c−a−b−1)/2
where u is a solution of
z 2 [1 − (a − b)2 ] + z[2c(a + b − 1) − 4ab] + c(2 − c)
u00 + u = 0.
4z 2 (1 − z)2
This is the so called Q-form of the hypergeometric equation.

27
28 Chapter 3. THE 3-PUNCTURED SPHERE

Proof. We substitute w(z) with u(z)v(z). The hypergeometric equation becomes

z(z − 1)(u00 v + 2u0 v 0 + uv 00 ) + [c − (a + b + 1)z](u0 v + uv 0 ) − abuv = 0.

Now we choose v in such a way that the sum of the term containing u0 goes to 0.
We obtain the following equation for v

2z(1 − z)v 0 + [c − (a + b + 1)z]v = 0.

Any solution of this equation is a multiple of

v(z) = z −c/2 (1 − z)(c−a−b−1)/2 .

Substituting v in the previous expression completes the proof.

We want to find coefficients a, b, c such that the Q-form of the hypergeometric


equation is
z2 − z + 1
u00 + 2 u = 0.
4z (z − 1)2
The only possible choice for those coefficients is a = 1/2, b = 1/2 and c = 1.
The resulting hypergeometric equation is
1
z(z − 1)w00 + (1 − 2z)w − w = 0. (3.2)
4

3.2 Properties of hypergeometric equation


The content of this section is mainly taken from the article of Toshio Oshima[13].
We will present the general theory of hypergeometric equation, with the exception
of some degenerate cases.

3.2.1 Local solutions at the origin


Let α, β and γ be three complex numbers. We consider the hypergeometric
equation
z(1 − z)w00 + (γ − (α + β + 1)z)w0 − αβw = 0.
If we set
d d
∂ := , θ := z
dz dz
the equation becomes
Pα,β,γ w = 0
where P is

Pα,β,γ = z(1 − z)∂ 2 + (γ − (α + β + 1)z)∂ − αβ


= θ(θ + γ − 1) − (θ + α)(θ + β)
P∞ P∞
Let w = n=0 cn z n , since θw = n=0 ncn z n , we obtain the following equation
for cn
(n − 1 + α)(n − 1 + β) (α)n (β)n
cn = cn−1 = · · · = c0 ,
n(n + γ − 1) (γ)n n!
3.2. Properties of hypergeometric equation 29

where
n−1
Y
(a)n = (a + k) = a(a + 1) . . . (a + n − 1).
k=0

If γ ∈
/ {0, −1, −2, . . . } the series converges for |z| < 1. Hence one solution of
hypergeometric equation is given by

X (α)n (β)n
w[α,β,γ] = z n = F (α, β, γ; z),
n=0
(γ)n n!

F is called the hypergeometric function.

We want to find a second independent solution defined in a neighborhood of


the origin.
Given a function h and a linear differential operator P we define a new differential
operator, namely
w
Ad(h)(P )w := hP .
h
It is easy to check that

h0
Ad(h)(∂) = ∂ − Ad(xλ )(θ) = θ − λ.
h
Thus we have
Ad(z γ−1 )(zPα,β,γ ) = zPα−γ+1,β−γ+1,2−γ .

Since the two equations Pα,β,γ w = 0 and Ad(z γ−1 )(zPα,β,γ )z γ−1 w = 0 are
equivalent, a second solution of (3.1) near the origin is given by

ŵ[α,β,γ] (z) = z 1−γ F (α − γ + 1, β − γ + 1, 2 − γ; z).

/ {2, 3, 4, . . . }1 .
for γ ∈

If γ = 1, w[α,β,γ] and ŵ[α,β,γ] coincide. Therefore we must define ŵ[α,β,1] in a


different way.
For |ε| < 1, we set

1
k[α,β,1−ε] := (ŵ[α,β,1−ε] − w[α,β,1−ε] )
ε
z ε F (α + ε, β + ε, 1 + ε; z) − F (α, β, 1 − ε; z)
=
ε
zε − 1 F (α + ε, β + ε, 1 + ε; z) − F (α, β, 1 − ε; z)
= F (α + ε, β + ε, 1 + ε; z) +
ε ε
= F1,ε (z) + F2,ε (z).

For ε → 0 , the first term of the expression tends to

F1,ε (z) → log(z)F (α, β, γ; z).


1 It is possible to find two solutions to the hypergeometric equation even if γ is an integer

different from 1, since we are not interested in this case we will not treat it
30 Chapter 3. THE 3-PUNCTURED SPHERE

While for the second term we obtain that


∞  
1 X (α + ε)n (β + ε)n (α)n (β)n
F2,ε (z) = − zn
ε n=0 (1 + ε)n n! (1 − ε)n n!
∞  
1 X (α)n (β)n n (α + ε)n (β + ε)n n! n!
= z − .
ε n=0 (n!)2 (α)n (β)n (1 + ε)n (1 − ε)n

We have that
  
(α + ε)n  ε ε ε
= 1+ 1+ 1+ ...
(α)n α α+1 α+2
n−1
X 1
=1+ε + O(ε2 ),
k=0
α + k

hence
∞ n−1  
X (α)n (β)n X 1 1 2
F2,ε (z) = zn + − + O(ε).
n=0
(n!)2 k=0
α+k β+k k+1

In this way we found a second independent solution of the hypergeometric


equation, defined near the origin. It has the following form
∞ n−1  
X (α)n (β)n n X 1 1 2
ŵ[α,β,1] = log(z)F (α, β, γ; z) + z + − .
n=0
(n!)2 k=0
α+k β+k k+1

3.2.2 Local solution at z = 1 and z = ∞


By the change of variables
T0↔1 : z 7→ 1 − z
We obtain the transformation T0↔1 (∂) = −∂, hence
T0↔1 Pα,β,γ = Pα,β,α+β−γ+1 .
A basis of solutions defined for |1 − z| < 1 is
w1 (z) = w[α,β,α+β−γ+1] (1 − z)
ŵ1 (z) = ŵ[α,β,α+β−γ+1] (1 − z)
By the change of variables
1
T0↔∞ : z 7→
z
we obtain the transformation T0↔1 (θ) = θ. One can prove that
zAd(z −α ) ◦ T0↔∞ (zPα,β,γ ) = −zPα,α−γ+1,α−β+1 .
The function u is a solution of zAd(z −α ) ◦ T0↔∞ (zPα,β,γ )u = 0 if and only if
T0↔∞ Pα,β,γ (z −α u) = 0. A basis of solutions defined for |z| > 1 is
 
−α 1
w∞ (z) = z w[α,α−γ+1,α−β+1]
z
 
1
ŵ∞ (z) = z −α ŵ[α,α−γ+1,α−β+1]
z
3.2. Properties of hypergeometric equation 31

3.2.3 Monodromy of the hypergeometric equation


Any solution of the hypergeometric equation defined in some subdomain U ⊂
X := C\{0, 1} can be analytically continued along any path starting from a point
z0 ∈ U . It continuation defines a holomorphic function on the universal covering
of X, hence if we fix a simply connected domain in X such that U ∩ X = 6 ∅,
the analytic continuation of the solution on this domain defines an holomorphic
function on it.
Let η0 , η1 and η∞ be three closed path starting from z0 and making a loop
in counterclockwise direction around the points 0, 1 and ∞ respectively.
If we set a basis of solution (wz0 , ŵz0 ), near z0 we see that for p = 0, 1, ∞, there
exists Mp ∈ GL(2, C) such that
   
ηp wz0 wz0
= Mp
ηp ŵz0 ŵz0

The three matrices satisfy the relation

M∞ M1 M0 = Id2 .

If we choose a different basis of solution (wz0 0 , ŵz0 0 ) we obtain a new set of


generators (gM0 g −1 , gM1 g −1 , gM∞ g −1 )
Theorem 3.2.1. Suppose that

α∈
/ {0, −1, −2, . . . };
β∈
/ {0, −1, −2, . . . }.

Under those conditions there exist a basis of solutions in which the monodromy
matrices are written as
   2πi(γ−α−β) 
1 0 e a1
(M0 , M1 ) = , .
a0 e−2πiγ 0 1

The coefficients a0 and a1 satisfy the following condition

a0 a1 = −4e−πi(β+α) sin(π(γ − β)) sin(π(γ − α))

Proof. The functions

w0 (z) = F (α, β, γ; z)
w1 (z) = F (α, β, α + β − γ + 1; 1 − z),

are two solutions of the hypergeometric equation defined in a neighborhood of 0


and 1 respectively. We take the analytic continuations of the two solutions on
X 0 := C \ {0, 1}. They are denoted as w0 and w1 .
Suppose that they are linearly dependent. Since the hypergeometric function is
holomorphic in a neighborhood of the origin, w0 is be an entire function on C.
The function w0 near ∞ can be written as a linear combination of w∞ and
ŵ∞ , therefore, thanks to the form of those two functions there exist N ∈ N such
that
w0 (z)
→0 for z → ∞.
zN
32 Chapter 3. THE 3-PUNCTURED SPHERE

The only class of entire functions with this property is the class of polynomials.
If α and β are not in the set {0, −1, −2, . . . } then

(α)n (β)n
6= 0 ∀n ∈ N,
(γ)n n!

hence w0 is not a polynomial near the origin. Therefore we proved that w0 and
w1 are independent solutions of hypergeometric equation.
We can evaluate the monodromy in the basis (w0 , w1 ). Firstly we consider a
loop η0 around z0 .
Since w0 is a holomorphic function defined near 0 we easily obtain that

η0 w0 = w0 .

In order to find η0 w1 we proceed in this way. Firstly we find two complex


numbers a and b such that near the origin w1 = aw0 + bŵ0 . Now we must
distinguish the case γ 6= 1 and γ = 1. If γ 6= 1 the function ŵ0 takes the form

ŵ0 = z 1−γ F (α − γ + 1, β − γ + 1, 2 − γ; z).

Taking η0 (t) = e2πit , we find that

η0 ŵ0 = e−2πiγ ŵ0 .

Hence

η0 w1 = aw0 + e−2πiγ bŵ0


= a0 w0 + e−2πiγ w1 .

If γ = 1 the function ŵ0 takes the form


∞ n−1  
X (α)n (β)n n
X 1 1 2
ŵ0 = log(z)F (α, β, γ; z) + z + − .
n=0
(n!)2 k=0
α+k β+k k+1

In this case we have that


η0 ŵ0 = 2πw0 + ŵ0 .
Therefore
η0 w1 = 2πbw0 + w1 .
In both cases we proved that the matrix M0 is written as
 
1 0
M0 = .
a0 e−2πiγ

In the same way we prove that


 2πi(γ−α−β) 
e a1
M1 = .
0 1
−1
We know that M1 M0 = M∞ . Therefore we obtain the equation
−1
T r(M1 M0 ) = T r(M∞ ).
3.3. Analytic continuation of local solution 33

Since T r(gAg −1 ) = T r(A) the trace of M∞ doesn’t depend from the basis in
−1
which we evaluate the monodromy. In the basis w∞ , ŵ∞ , T r(M∞ ) = e−2πiα +
e−2πiβ , hence

e−2πiγ + a0 a1 + e2πi(γ−α−β) = e−2πiα + e−2πiβ ,

therefore

a0 a1 = −e−2πiγ − e2πi(γ−α−β) + e−2πiα + e−2πiβ


= e−2πiβ −e2πi(β−γ) − e2πi(γ−α) + e2πi(β−α) + 1


= e−2πiβ 1 − e2πi(β−γ) 1 − e2πi(γ−α)


 
 πi(γ−β)
− eπi(β−γ)
  πi(α−γ)
− eπi(γ−α)

e e
= 4e−2πiβ eπi(β−γ) eπi(γ−α)
2 2
= 4e−πi(β+α) sin(π(γ − β)) sin(π(α − γ))

Remark If γ − α ∈ / Z and γ − β ∈
/ Z, we can choose a couple of independent
solutions such that

a0 = 2e−πi(α+β) sin π(β − γ)


a1 = 2 sin π(γ − α).

As a matter of fact in this case a0 a1 6= 0, therefore if we take w̃1 = kw1 we


obtain that
η0 w̃1 = ka0 w0 + e−2πiγ w̃1
Since a0 6= 0 we can choose k in such a way that ã0 = e−πi(α+β) sin π(γ − β).

3.3 Analytic continuation of local solution


Using the theory of hypergeometric equation we developed in last section, we
find that the monodromy group of (3.2) is
   
1 0 1 2
Γ= , .
2 1 0 1
Therefore the Deck transformation group of J is
 
z
Deck(J) = z + 2, .
2z + 1
Thus we obtain the isomorphism

C \ {0, 1} = H/Deck(J).

Now we would like to find two independent solution of the equation u00 +
ωC\{0,1} u = 0 defined on the subset C \ (] − ∞, 0] ∪ [1, ∞[).
We know that the independent solutions u0 and u1 can be taken as
(
u0 = v · F (1/2, 1/2, 1; z) |z| < 1
u1 = v · F (1/2, 1/2, 1; 1 − z) |1 − z| < 1
34 Chapter 3. THE 3-PUNCTURED SPHERE

where
v(z) = z −1/2 (1 − z)−1/2 .

It is possible to find an explicit expression of the analytic continuation of the


function F (1/2, 1/2, 1; z) on the domain C \ [1, ∞[.
The complete elliptic integral of the first kind is defined as
Z 1
1
K(k) = √ √ dt.
0 1−t 2 1 − k 2 t2


The function H(z) = K( z) satisfies the following property

Proposition 3.3.1. The functions


Z 1
1
H(z) = √ √ dt (3.3)
0 1−t 2 1 − zt2

is a holomorphic function on the domain C \ [1, +∞[. On the unitary disk we


have the equality
π
H(z) = F (1/2, 1/2, 1; z) .
2
Proof. ∀z ∈ C \ {1} the integral
Z 1
1
H(z) = √ √ dt
0 1−t 2 1 − zt2

exists and it is finite. In order to prove that the function H is continuous in


any point z ∈ C \ [1, +∞[ we take a sequence zn converging to z. We chose the
sequence zn in such a way that none of its elements lays on the line [1, +∞[.
Hence there exists a constant C such that

1 < C√ 1

sup √ √ ∀n ∈ N.
t∈[0,1] 1 − t 1 − zn t
2 2 1 − t2

We found a dominant function in L1 [0, 1], hence

lim H(zn ) = H(z).


n→∞

The complex derivative of H(z) takes the form


1
t2
Z
1
∂z H(z) = √ dt
2 0 1 − t2 (1 − zt2 )3/2

We proved above that this integral exists and it is continuous ∀z ∈ C\[1, +∞[.
Moreover ∂z̄ H(z) = 0, therefore the function H(z) is holomorphic on C \ [1, +∞[

In order to prove that H(z) coincide up to a multiplication with a constant


with the hypergeometric function, we expand H(z) in Taylor series.
3.4. Theta-functions 35

The value of the evaluation at z = 0 of the n-th derivative of H(z) is

(2n − 1)!! 1 t2n


Z
H (n) (0) = √ dt
2n 0 1 − t2
(2n − 1)!! 1 xn−1/2
Z
= √ dx
2n+1 0 1−x
(2n − 1)!!
= β(n + 1/2, 1/2)
2n+1
(2n − 1)!! Γ(n + 1/2)Γ(1/2)
=
2n+1 Γ(n + 1)
(2n − 1)!!2 π
= ,
22n+1 n!

where (2n − 1)!! = 1 · 3 · 5 · · · (2n − 1).


Therefore H(z) on the unit disk can be written as

π X (2n − 1)!!2 n
H(z) = z
2 n=0 22n n!2
π
= F (1/2, 1/2, 1; z).
2

3.4 Theta-functions
Definition The Theta-function ϑ4 (z, τ ) : C × H → C is defined as

2
X
ϑ4 (z, τ ) = (−1)n eiπτ n +2niz

−∞

It is a quasi doubly-periodic function with respect to the variable z, i.e

ϑ4 (z + π, τ ) = ϑ4 (z, τ )
ϑ4 (z + πτ, τ ) = −e−iπτ −2iz ϑ4 (z, τ )

1 and −e−iπτ −2iz are the periodicity factor with respect to the periods π and πτ .

Proof. Firstly we notice that, if the imaginary part of τ is greater than 0 we


have |eiπτ | = a < 1, therefore
2 2
|(−1)n eiπτ n +2niz
| ≤ an e2n|z| .

Then ϑ4 is well defined and it is an entire function for fixed τ .

Let’s study the quasi-periodicity of ϑ4 . For the first quasi-period it’s obvious
that
ϑ4 (z + π, τ ) = ϑ4 (z, τ ).
36 Chapter 3. THE 3-PUNCTURED SPHERE

For the second quasi-period we find



2
X
ϑ4 (z + πτ, τ ) = (−1)n eiπτ n +2niπτ +2niz
−∞

2
X
= −e−iπτ −2iz (−1)n+1 eiπτ (n+1) +2(n+1)iz

−∞

= −e−iπτ −2iz ϑ4 (z, τ )

Definition The Theta function ϑ1 is defined as


 
1 πiτ
iz+ 4 1
ϑ1 (z, τ ) = −ie ϑ4 z + πτ, τ
2

1 2
(−1)n eiπτ (n+ 2 ) e(2n+1)iz
X
= −i
n=−∞

ϑ2 is defined as
 
1
ϑ2 (z, τ ) = ϑ1 z + π
2

X iπτ n+ 1 2
= e ( 2 ) e(2n+1)iz
n=−∞

ϑ3 is defined as
 
1
ϑ3 (z, τ ) = ϑ4 z + π, τ
2

2
X
= eiπτ n e2niz .
n=−∞

Remark It is common to hide the dependence from τ and to write

ϑi (z) = ϑi (z, τ )
ϑi = ϑi (0, τ )

Remark The Theta-functions are all well defined function from C × H → C.


They are all quasi doubly-periodic function, with periods π and πτ . The
periodicity factors are summarized in the following table.
ϑ1 (z, τ ) ϑ2 (z, τ ) ϑ3 (z, τ ) ϑ4 (z, τ )
π -1 -1 1 1
πτ -N N N -N

where N= e−iπτ −2iz

If a point z0 is a zero of ϑi (z), any point of the form z0 + mπ + nπτ is a zero


of ϑi (z). Therefore it suffices to find the zeroes of ϑi (z) inside the cell C with
corners t, t + π, t + πτ, t + π + πτ .
3.4. Theta-functions 37

Proposition 3.4.1. For any i ∈ 1, 2, 3, 4 the function ϑi (z) has a simple 0 in


the cell C.
Proof. It suffices to prove the statement in the case of ϑ4 (z).
Since the function ϑ4 (z) is entire, the number of zeroes inside C is given by
Z 0
1 ϑ4 (z)
N= dz.
2πi C ϑ4 (z)
Hence
t+π t+πτ
ϑ04 (z) ϑ04 (z + πτ ) ϑ04 (z) ϑ04 (z + π)
Z   Z   
1
N= − dz − − dz
2πi t ϑ4 (z) ϑ4 (z + πτ ) t ϑ4 (z) ϑ4 (z + π)
It is easy to prove that
ϑ04 (z) ϑ04 (z + πτ ) ϑ04 (z) ϑ04 (z + π)
− = 2i − = 0,
ϑ4 (z) ϑ4 (z + πτ ) ϑ4 (z) ϑ4 (z + π)
therefore Z t+π
1
N= 2i = 1
2πi t

Remark The zeros of ϑ1 (z) are the points congruent to 0, hence the zeros of
ϑ1 (z), ϑ2 (z), ϑ3 (z), ϑ4 (z) are respectively the points congruent to 0, π2 , π2 + πτ
2
, πτ
2

Lemma 3.4.2. The Theta functions satisfy the following equations

ϑ22 (z)ϑ24 = ϑ24 (z)ϑ22 − ϑ21 (z)ϑ23


(3.4)
ϑ23 (z)ϑ24 = ϑ24 (z)ϑ23 − ϑ21 (z)ϑ22

Proof. The functions ϑ2i (z) are all quasi doubly-periodic function with respect
to the periods π and πτ , with periodicity factors 1 and e−2iπτ −4iz , therefore the
functions
aϑ21 (z) + bϑ24 (z) a0 ϑ21 (z) + b0 ϑ24 (z)
,
ϑ22 (z) ϑ23 (z)
are doubly periodic functions with periods π and πτ . We can choose the constants
a, b, a0 , b0 such that the two function above have only one simple pole inside C,
this implies that the two functions are constant.
We can suppose that those two functions are equal to 1, therefore the constants
a, b, a0 , b0 satisfy the following relations

aϑ21 (z) + bϑ24 (z) = ϑ22 (z) a0 ϑ21 (z) + b0 ϑ24 (z) = ϑ23 (z).
πτ
If we set z = 0 and z = 2
we obtain that

ϑ23 = aϑ24 , ϑ22 = bϑ24 ; ϑ22 = −a0 ϑ24 , ϑ23 = b0 ϑ24 .

This concludes the proof of the theorem.


Corollary 3.4.3. The following equality holds

ϑ42 + ϑ44 = ϑ43


38 Chapter 3. THE 3-PUNCTURED SPHERE

π
Proof. If we set z = 2
in the second equation of (3.4) we obtain the desired
result.
Lemma 3.4.4. Let τ 0 = − τ1 then

ϑ3 (τ 0 ) = (−τ )1/2 ϑ3 (τ )
(3.5)
ϑ2 (τ 0 ) = (−iτ )1/2 ϑ4 (τ )
2
Proof. The Fourier transform of eiπτ x is
h 2
i 0 2
Fx eiπτ x = (−iτ )−1/2 eiτ πk .

Using Poisson summation formula



X ∞
X
f (n) = fˆ(k),
n=−∞ k=−∞

and the definition of ϑ3 we obtain the first formula. The second formula can be
proved with analogous methods.

3.5 The modular lambda function


Now we can write the universal covering map of C \ {0, 1}.
Definition Let λ : H −→ C be defined as
ϑ42 (τ )
λ(τ ) = .
ϑ43 (τ )

The function λ(τ ) is called modular lambda function.


The following theorem holds true
Theorem 3.5.1. The function λ : H → C \ {0, 1} is the universal covering map
of the sphere with three missing points.
This entire section is dedicated to the proof of the theorem
We remind that the Deck transformation group of C \ {0, 1} is
 
τ
Deck(J) = τ + 2, .
2τ + 1

Hence λ(τ ) should satisfy the following relations

λ(τ + 2) = λ(τ ),
 
τ
λ = λ(τ ).
2τ + 1

The first one is obvious from the definitions of the Theta-functions. The second
one follows from the fact that

λ(τ 0 ) = 1 − λ(τ ). (3.6)


3.5. The modular lambda function 39

Which follows from(3.5).


In order to prove that λ(τ ) is the universal covering map of the three
punctured sphere, we have to prove first that λ(τ ) is a surjective map from H
onto C \ {0, 1}.
We must solve the Inversion problem

λ(τ ) − c = 0

with c ∈ C \ {0, 1}. This problem is solved by the following theorem.

Theorem 3.5.2. Let λ(τ ) be the modular lambda function

(i) λ(τ ) : D ⊂ H −→ C \ (] − ∞, 0] ∪ [1, ∞[) is bijective.

(ii) λ(τ ) : D̄ ⊂ H −→ C \ (] − ∞, 0] ∪ [1, ∞[) is surjective

−1 0 1

Figure 3.1: The domain D

Proof. λ(τ ) is analytic on H, hence the number of solution of the equation


λ(τ ) − c = 0 inside a generic domain Ω is given by
Z
1 dλ(τ )
dτ. (3.7)
∂Ω λ(τ ) − c dτ

We define the function h(τ ) and g(τ ) as

ϑ44 (τ )
g(τ ) =
ϑ43 (τ )
λ(τ )
h(τ ) = − .
g(τ )

If τ 0 = −1/τ the following equalities hold true

λ(τ + 2) = λ(τ ), g(τ + 2) = g(τ ), λ(τ ) + g(τ ) = 1


f (τ + 1) = h(τ ), λ(τ 0 ) = g(τ ), g(τ 0 ) = λ(τ ).

Take the contour ABCDEF E 0 D0 C 0 B 0 A shown in the figure. Let’s suppose


temporarily that λ(τ ) − c has no zero on ∂D. We will analyze later this case
The contour is constructed in the following way:

• F E is drawn parallel to the real axis, at a large distance from it.


40 Chapter 3. THE 3-PUNCTURED SPHERE

• AB is the inverse of F E with respect to the circle |τ | = 1.

• D taken on the line <z = <E and = D = = A

• BC is the inverse of ED with respect to |τ | = 1.

• CD is the reflection of the arc AB with respect to the line < τ = 1/2.

• The left-hand half of the figure is the reflection of the right-hand half with
respect to the line < τ = 0.

Now we evaluate (3.7) on the border of the domain we have constructed.


Since λ(τ + 2) = λ(τ ), we have
Z Z 
1 dλ(τ )
+ dτ = 0.
DE E 0 D0 λ(τ ) − c dτ

Moreover, as τ describes BC and B 0 C 0 , τ 0 = −1/τ describes E 0 D0 and ED

E’ F E

D’ A D
C’ B’ B C

-1 0 1

Figure 3.2: the integration path ABCDEF E 0 D0 C 0 B 0 A


3.5. The modular lambda function 41

respectively

dg(τ 0 )
Z Z  Z Z 
1 dλ(τ ) 1
+ dτ = + dτ
BC C 0 B0 λ(τ ) − c dτ BC C 0 B0 g(τ ) − c dτ
0

dg(τ 0 ) 0
Z Z 
1
= + dτ
BC C 0 B0 g(τ ) − c dτ 0
0

= 0.

The last equality follows from the fact that g(τ 0 + 2) = g(τ 0 ).
Since λ(τ ± 1) = h(τ ), we have
Z Z  Z 
1 dλ(τ ) 1 dh(τ )
+ dτ = dτ ;
0
D C 0 CD λ(τ ) − c dτ 0
B AB h(τ ) − c dτ

but, as τ 0 describes B 0 AB, τ describes EE 0 , and so the integral around the


complete contour reduces to

dh(τ 0 ) dλ(τ 0 )
Z  
1 dλ(τ ) 1 1
+ + dτ
EE 0 λ(τ ) − c dτ h(τ 0 ) − c dτ λ(τ 0 ) − c dτ
Z  
1 dλ(τ ) 1 dh(τ ) 1 dg(τ )
= − + dτ
EE 0 λ(τ ) − c dτ h(τ ){1 − c h(τ )} dτ g(τ ) − c dτ
(3.8)

Now as EE 0 moves off to infinity, λ(τ ) − c → −c 6= 0, g(τ ) − c → 1 − c 6= 0.


Therefore the limit of the integral is
Z
1 d
− lim {log h(τ )}dτ
EE 0 1 − c h(τ ) dτ
=E→+∞

1 −πiτ
It is easy to see that if =τ → +∞, the functions λ1 (τ ) = 16 e λ(τ ) and g(τ )
tend to unity, uniformly with respect to <τ , when −1 ≤ <τ ≤ 1; and their
derivatives tend uniformly to zero in the same way.
Since 1 − c h(τ ) → 1 the limit of the integral is given by
Z  
1 d log λ1 (τ ) d log g(τ )
lim πi + − dτ = 2πi.
EE 0 1 − c h(τ ) dτ dτ
=E→+∞

If we choose EE 0 to be initially so far from the real axis that λ(τ ) − c, 1 − c h(τ )
and g(τ ) − c have no zeroes when τ is above EE 0 , then the contour will pass over
no zeros of λ(τ ) − c as EE 0 moves off to infinity and the radii of the arcs CD,
D0 C 0 , B 0 AB diminish to zero. The integral will not change while the contour is
modified, and so the original contour integral will be 2πi, therefore the number
of zeroes of λ(τ ) − c inside the original contour is precisely one.
It remains to discuss what happens if z0 ∈ ∂D is a zero of λ(τ ) − c. In this
case it is possible to modify the shape of the contour of integration in such a
way that z0 ∈ D0 E 0 is the unique zero of λ(τ ) − c inside the new contour. For
example if z0 ∈ D0 E 0 we can make an indentation in D0 E 0 and a corresponding
one in DE. In this case the integral along the indentations cancel in virtue of
the relation λ(τ ) = λ(τ + 2).
42 Chapter 3. THE 3-PUNCTURED SPHERE

Let J be the universal covering map of C \ {0, 1}. In the previous chapter
we proved that J −1 satisfies the equation S (J −1 ) = 2ωC\{0,1} , where is the
canonical projective connection of Liouville equation on the manifold.
Let’s consider the function
H(1 − z)
f (z) = i , z ∈ C \ (] − ∞, 0] ∪ [1, ∞[).
H(z)
Since f (z) is the ratio of two independent solutions of (3.2), it will satisfies the
equation S (f ) = 2ωC\{0,1} . Hence f coincides, up to a conjugation with an
element of P SL(2, C), to any branch of J −1 . Let’s consider a branch of J −1 , if
we set V = J −1 (C \ (] − ∞, 0] ∪ [1, ∞[)), we obtain that
f ◦ J|V = γ|V (z), γ ∈ P SL(2, C).

Let’s suppose that we are able to prove that


f ◦ λ|D = idD ;
λ|D ◦ f = idC\(]−∞,0]∪[1,∞[) .
Therefore f : C\] − ∞, 0] ∪ [1, ∞[ is bijective. Moreover we notice that γ|V :
V → D, and that
λ|D = J|V ◦ γ|−1
V (3.9)

Lemma 3.5.3. The functions λ, J : H → C \ {0, 1} admit no extension to any


domain Ω ) H.
Proof. The only property of the functions λ and J required in order to prove
the theorem is that ∀γ ∈ Γ = hτ + 2, 2ττ+1 i, we have that
λ(γ(z)) = λ(z), J(γ(z)) = J(z).
We will prove the lemma only for the function λ, then the proof is equal also for
J.
If the function λ admitted an extension defined in an open neighborhood
Ω ) H there would be a point ζ ∈ R̂ such that ζ ∈ Ω. From proposition (2.3.11)
it follows that Γ is a Fuchsian group of the First kind, thus proposition (1.2.9)
implies that there exists a sequence γn ∈ Γ such that, for every z ∈ H, γn (z) → ζ.
Let’s call λ̄ the extension of λ on Ω. Since λ̄|H = λ it follows that
λ̄(ζ) = lim λ(γn (z)) = λ(z),
n→∞

this imply that λ is a constant function, which is absurd.


We notice that the R.H.S. of (3.9) can be expanded on the domain γ(H) to
the function J ◦ γ −1 .
Lemma above implies that γ(H) ⊂ H. In the same way we prove that
γ −1 (H) ⊂ H, this prove that γ ∈ P SL(2, R). Therefore equation (3.9) can be
rewritten as
λ ◦ γ = J,
Which means that λ is the universal covering of C \ {0, 1}.
It remains to prove that
3.5. The modular lambda function 43

Proposition 3.5.4. The function f : C \ (] − ∞, 0] ∪ [1, ∞[), defined as

H(1 − z)
f (z) = i
H(z)

is bijective onto D and its inverse is λ|D .

Lemma 3.5.5. The Theta-functions satisfy the following differential equation


 
d ϑ1 (z) ϑ2 (z) ϑ3 (z)
= ϑ24 .
dz ϑ4 (z) ϑ4 (z) ϑ4 (z)

If we set ξ(z) = ϑ1 (z)/ϑ4 (z) it satisfies also the following differential equation
 2
dξ(z)
= (ϑ22 − ξ 2 ϑ23 )(ϑ23 − ξ 2 ϑ22 )
dz

Proof. The function ϑ1 (z)/ϑ4 (z) is quasi doubly-periodic with periodicity factors
1 and −1 with respect to the periods π and πτ . The derivative of it is

ϑ01 (z)ϑ4 (z) − ϑ1 (z)ϑ04 (z)


,
ϑ24 (z)

it has the same periodicity factors of ϑ1 (z)/ϑ4 (z).


We easily see that
ϑ2 (z)ϑ3 (z)
ϑ24 (z)
has again periodicity factors −1 and 1.
If we define the function
ϑ01 (z)ϑ4 (z) − ϑ1 (z)ϑ04 (z)
φ(z) = ,
ϑ2 (z)ϑ3 (z)

it will be a doubly-periodic function with period π and πτ , and the only possible
poles of φ(z) are poles congruent to the points π2 and π2 + πτ 2
If we consider
φ(z + πτ /2), from the relations
 πτ  πiτ
 πτ  πiτ
ϑ1 z + = ie− 4 −iz ϑ4 (z), ϑ4 z + = ie− 4 −iz ϑ1 (z)
2 2
 πτ  πiτ
 πτ  πiτ
ϑ2 z + = e− 4 −iz ϑ3 (z), ϑ3 z + = e− 4 −iz ϑ2 (z)
2 2
we easily see that  πτ 
φ z+ = φ(z).
2
Therefore φ(z) is doubly periodic with respect to the periods π and πτ 2
and the
only possible poles are single poles at the points congruent to π2 . Therefore φ(z)
must be a constant function.
If we send z → 0, using the following proposition, which can be found in [20]

ϑ01 (0) = ϑ2 (0)ϑ3 (0)ϑ4 (0)


ϑ04 (0) = 0,
44 Chapter 3. THE 3-PUNCTURED SPHERE

we obtain that
φ(z) = ϑ24 .

This implies that


 
d ϑ1 (z) ϑ2 (z) ϑ3 (z)
= ϑ24 .
dz ϑ4 (z) ϑ4 (z) ϑ4 (z)
. The second part of the proposition follows easily from relations (3.4)

Let’s write
ξϑ3 /ϑ2 = y zϑ23 = u

Under this change of variables, if τ ∈ D we obtain that y(u) satisfies the following
differential equation
 2
dy
= (1 − y 2 )(1 − λ|D (τ )y 2 ) (3.10)
du

From previous lemma we see that one particular solution of this differential
equation is given by
ϑ3 ϑ1 (uϑ−2
3 )
y= .
ϑ2 ϑ4 (uϑ−2
3 )

A common notation used for this function is the following

y(u) = sn(u, k) = sn(u).

From equation (3.10) the inverse function u(y) can be written as


Z y(u)
1
u= √ p dt.
0 1−t 2 1 − λ|D (τ )t2

If u0 = π2 ϑ23 it follows that

π

ϑ 3 ϑ1 2
y(u0 ) = π
 = 1,
ϑ 2 ϑ4 2

Therefore
Z 1
π 2 1 π
ϑ (τ ) = √ dt = H(λ|D (τ )).
2 3
p
0 1−t 2 1 − λ|D (τ )t 2 2

If we replace τ with τ 0 = −1/τ . Using relations (3.5) and (3.6) we obtain that

−iτ ϑ23 = H(1 − λ|D (τ )),

thus
τ = f ◦ λ|D (τ ).

Since λ|D is bijective it follows that λ|D is the inverse of f .


3.6. Solution of Liouville equation 45

3.6 Solution of Liouville equation


In the last section of this chapter we write the canonical solution of Liouville
equation on C \ {0, 1}.
Firstly we will prove that the function function H(z) we defined in (3.3) has a
logarithmic behavior near z = 1.

Lemma 3.6.1. For every natural number n the following inequality holds true

1 1 (2n − 1)!! 1
p <p < <√
π(n + 1) π(n + 1/2) (2n)!! πn

Proof. Let’s consider the real-valued function


Z π/2
I(n) = sinn θdθ, n ∈ N.
0

We have that
Z π/2
I(n + 2) = sinn θ(1 − cos2 θ)dθ
0
π/2
d sinn+1 θ
Z
= I(n) − cos θ dθ
0 (n + 1)dθ
1
= I(n) − I(n + 2).
n+1
Therefore we have that
n+1
I(n + 2) = I(n).
n+2
Iterating the formula we obtain that

(2n − 1)!! (2n − 1)!! π


I(2n) = I(0) = ,
(2n)!! (2n)!! 2

and
(2n)!! (2n)!!
I(2n + 1) = I(1) = · 1.
(2n + 1)!! (2n + 1)!!

Therefore
1 π
I(2n + 1)I(2n) = , ∀n ∈ N.
2n + 1 2

Since 0 < sin θ < 1 for 0 < θ < π2 , it follows that

I(2n − 1) < I(2n) < I(2n + 1).

Consequently,
1 π 1 π
< I 2 (2n) < .
2n + 1 2 2n 2
46 Chapter 3. THE 3-PUNCTURED SPHERE

Lemma 3.6.2. The function H(z) can be written in a neighborhood of z = 1 as


1
H(z) = − log(1 − z) + h1 (z) log(1 − z) + h2 (z),
2
where h1 (z) and h2 (z) are holomorphic functions near z = 1 and h1 (z) = 0.
Proof. H(z) is a solution of the hypergeometric equation with coefficients α =
β = 1/2 and γ = 1, defined near z = 1. A basis of solution near this point is
given by

u1 (z) = F (1/2, 1/2, 1; 1 − z),


u2 (z) = log(1 − z)F (1/2, 1/2, 1; 1 − z) + ψ(z),

with ψ(z) holomorphic near z = 1.


Therefore H(1 − z) can be written as a linear combination

H(1 − z) = au1 (z) + bu2 (z).

Let z ∈]0, 1[. On this segment, using the previous lemma, we find that
∞  2 ∞
π X (2n − 1)!! 1X1 n 1
H(z) = zn ≤ z = − log(1 − z),
2 n=0 (2n)!! 2 n=0 n 2

and that
∞  2 ∞
π X (2n − 1)!! 1X 1 1
H(z) = zn ≥ z n = − z log(1 − z),
2 n=0 (2n)!! 2 n=0 n + 1 2

Therefore b = −1/2. Since F (1/2, 1/2, 1; 0) = 1 + h1 (z), with h1 (z) = 0 the


lemma is proved.
Theorem 3.6.3. The canonical solution of Liouville equation on the three
punctured sphere C \ {0, 1} is given by
 
1
ϕC\{0,1} (z, z̄) = log .
π|z||1 − z||<(H(z)H(1 − z̄))|
Proof. A branch of J −1 defined on C\] − ∞, 0] ∪ [1, ∞[ is given by
H(1 − z)
J −1 (z) = i .
H(z)
Applying formula (2.2) we find that
 
|∂z [H(1 − z)]H(z) − ∂z H(z)H(1 − z)| 1
ϕC\{0,1} (z, z̄) = log    
|H(z)|2 H(1−z)
< H(z)
 
|∂z [H(1 − z)]H(z) − ∂z H(z)H(1 − z)|
= log .
|<(H(z)H(1 − z̄))|
We recall that H(1 − z) and H(z) solve the differential equation
1
z(1 − z)u00 + (1 − 2z)u0 − u = 0.
4
3.6. Solution of Liouville equation 47

We define the Wronskian function w(z) := ∂z H(1 − z) · H(z) − ∂z H(z) · H(1 − z).
Now we substitute the asymptotic expression of H(z) found in the lemma into
w(z). Then, near z = 1, w(z) takes the form
π
w(z) = − + log(1 − z) · h1 (z) + h2 (z),
4(1 − z)

where h1 and h2 are holomorphic near z = 1. Moreover w(z) solves the differential
equation
2z − 1
w0 = w.
z(1 − z)
This equation can be integrated by part, hence
1
w(z) = w0 .
z(1 − z)

Comparing the two expression of w(z), we obtain that w0 = −π/4, therefore


 
π
ϕC\{0,1} (z, z̄) = log .
4|z||1 − z||<(H(z)H(1 − z̄))|
48 Chapter 3. THE 3-PUNCTURED SPHERE
Chapter 4

Solutions of Liouville equation

S punctured Riemann surface R, i.e. a


In this chapter we prove that on a compact
punctured Riemann surface R = R̂ \ i {pi }, there exists a unique solution of
Liouville equation that on zi (Ui ) = D \ {0} is written as

ϕi,R (z, z̄) = − log |z| − log | log |z|| + o(1), z → 0,

where zi : Ui → D is the chart relative to the puncture pi .


This solution coincide with the canonical solution of Liouville equation defined
in (2.2).
Using this result the problem of finding the accessory parameter on the one
punctured torus T \ {0} is proved to be equivalent to the problem of finding
the accessory parameter on the four punctured sphere C \ {0, 1, 1/a} where
1/a ∈ C \ {0, 1}.
Moreover it is proved that the solution of Liouville equation on the punctured
sphere minimize a functional defined on the space of functions with the asymptotic
behavior near the punctures described above.

4.1 Asymptotic behavior of the solutions of Liou-


ville’s equation
Let R be an hyperbolic punctured Riemann surface with atlas {(Uα , zα )}α∈I ,
containing the charts zi : Ui → D \ {0}. Let J : H → R be the universal
covering map of it. We want to prove that the canonical solution of Lioville
equation {ϕα,R }α∈I , on the charts {ϕi,R }i∈I˜ corresponding to the punctures, has
the asymptotic behavior

ϕi,R (z, z̄) = − log |z| − log | log |z|| + o(1) z → 0.

Given a solution of Liouville equation {ϕα }α∈I we introduce the collection of


functions
λα (z) = eϕα (z) , λα : zα (Uα ) → ]0, ∞].
λα (z) satisfies a modified version of Liouville equation of the type

∆ log(λα (z)) = λα (z)2 . (4.1)

49
50 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

On zβ (Uα ∩ Uβ ) we have the connection formula


0
λβ (z, z̄) = λα ◦ gα,β (z, z̄) · |gαβ (z)|.

Hence we can write √


λ = λα,R (zα , z̄α )|dzα | = ds2 .
We will first prove a lemma, due to Ahlfors, which prove that the standard
metric on the unit disk, i.e the metric
4
ds2D = dzdz̄
(1 − |z|2 )2
is connected with the maximal solution of Liouville’s equation on the unit disk.
The proof of this theorem together with other ideas contained in this chapter
was taken from the article [11].
Lemma 4.1.1 (Ahlfors’ Lemma). Let λ(z) be a solution of equation (4.1) on
the unit disk.
For any z ∈ D
λ(z) ≤ λD (z),
where
2
λD (z) = .
1 − |z|2
Proof. Let’s take R < 1, the function λ(z) satisfies equation (4.1) on DR = {z ∈
C| |z| < R} and is continuous on the boundary of DR .
If we set
2R
λDR (z) = 2 ,
R − |z|2
then the function
 
λ(z)
u(z) := log , z ∈ DR ,
λDR (z)
tends to −∞ as |z| → R, hence it attains its maximal value at some point
z0 ∈ DR . Since u is of class C 2 near z0 , its laplacian evaluated in z0 is non
positive. Therefore

0 ≥ ∆u(z0 )
= ∆ log λ(z0 ) − ∆ log λDR (z0 )
= λ(z0 )2 − λDR (z0 )2 .

This implies that


λ(z0 )2
 
1
u(z0 ) = log ≤ 0.
2 λDR (z0 )2
Since z0 was a maximum point for u(z), it follows that λ(z) < λDR (z) on the
disk DR .
If we take any point in D, there exist R such that z ∈ DR . sending R → 1,
we obtain that
λ(z) ≤ λD (z).
4.1. Asymptotic behavior of the solutions of Liouville’s equation 51

Corollary 4.1.2. Let R be an hyperbolic Riemann with atlas {(Uα , zα )}α∈I . If


{ϕα (z, z̄)} is a solution of Liouville equation, then

ϕα (z, z̄) ≤ ϕα,R (z, z̄) ∀α ∈ I,

where {ϕα } is the canonical solution of Liouville equation on R.


Proof. Given a solution Liouville equation {ϕα (z, z̄)}, we consider the collection
of function λ(z, z̄)α = eϕα (z,z̄) , if we pullback the metric

ds2 = λα (zα , z̄α )2 dzα dz̄α ,

using the universal covering map J : D → R, we obtain a metric of constant


curvature K = −1 on D. Therefore the function

λ̃(z, z̄) = λα ◦ zα ◦ J(z, z̄)|(zα ◦ J)0 (z)|

is well defined on D and solves the equation (4.1) on D. Moreover if we consider


the collection of function λα,R (z, z̄) = eϕα,R (z,z̄) , we easily prove that λD (z, z̄) =
λα,R ◦ zα ◦ J(z, barz)|(zα ◦ J)0 (z)|, Ahlfors’ lemma implies that on J −1 (Uα )

λα ◦ zα ◦ J(z, z̄)|(zα ◦ J)0 (z)| ≤ λα,R ◦ zα ◦ J(z, z̄)|(zα ◦ J)0 (z)|.

since |(zα ◦ J)0 (z)| =


6 0 we conclude that

λα (z, z̄) ≤ λα,R (z, z̄) z ∈ zα (Uα ),

hence ϕα (z, z̄) ≤ ϕα,R (z, z̄).


Corollary 4.1.3. Let R and S be two hyperbolic Riemann surfaces such that
there exist an embedding i : R ,→ S. Let’s take an atlas {(Uα , zα )}α∈I on S
and the atlas {(i−1 (Uα ), zα ◦ i)}α∈I on R. We consider the canonical solutions
of Liouville equation {ϕα,R } and {ϕα,S } on R and on S respectively. On zα ◦
i(i−1 (Uα ) ⊂ zα (Uα ) we have that

ϕα,S (z, z̄) ≤ ϕα,R (z, z̄).

Proof. We easily see that if we restrict every ϕα,S to zα ◦ i(i−1 (Uα ) we obtain
a solution of Liouville equation on R . The result follows from the corollary
above.
Let’s now come back to the case in which R ⊆ C. It is convenient to find
the canonical solution of Liouville equation on the punctured disk D0z0 ,R := {z ∈
C| 0 < |z − z0 | < R} and on the twice punctured plane C \ {z0 , z1 }.
Lemma 4.1.4. The canonical solutions of Liouville equation for the punctured
disk and the twice punctured plane are respectively:
1
ϕD0R (z, z̄) = log ,
|z − z0 | log |R/(z − z0 )|
and
π|z − z |
tϕC\{z0 ,z1 } (z, z̄) = log h0  1   i .
4|z − z0 ||z − z1 | < H zz−z 0
H z̄−z̄1

1 −z0 z̄0 −z̄1

52 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

Where
∞  
π X (2n − 1)!! n
H(z) = z
2 n=0 (2n)!!
(
1 n = 2m + 1
and n!! = n(n − 2)(n − 4) · · ·
2 n = 2m

Proof. For what regards the punctured disk, we know that the universal covering
map J : H → D0R is JD0z ,R (z) = z0 + Re2πiz . Therefore if we take any branch of
0
J −1 and we use the formula (2.2), we obtain the desired expression for ϕD0R .
In the previous chapter we found that the canonical solution of Liouville
equation on C \ {0, 1} is given by
π
ϕC\{0,1} = log .
4|z||1 − z||<[H(z)H(1 − z̄)]|

If one considers the map f : C \ {z0 , z1 } → C \ {0, 1} given by f (z) = (z −


z0 )/(z1 − z0 ) and he takes the pullback of eϕC\{0,1} dzdz̄ via f , he finds the desired
formula for C \ {z0 , z1 }.

Remark If R = D0R the canonical solution of Liouville equation (2.2) is given


by

ϕD0z (z, z̄) = − log |z − z0 | − log | log |z − z0 | − log R|


0 ,R

log R
= − log |z − z0 | − log | log |z − z0 || − log 1 −
log |z − z0 |
= − log |z − z0 | − log | log |z − z0 || + o(1),

As z → 0.
Let’s now study the twice punctured plane. We recall that H(0) = π/2 and that,
near z = 1, H(z) can be written as

1
H(z) = − log(1 − z) + h1 (z) log(1 − z) + h2 (z),
2
with h1 and h2 holomorphic and h1 (1) = 0.
Therefore the canonical solution of Liouville equation ϕC\{z0 ,z1 } near z = z0 can
be written as

ϕC\{z0 ,z1 } =
    
z − z1 4 z̄ − z̄1 z − z0
= − log |z − z0 | − log
− log <
H H
z0 − z1 π z̄0 − z̄1 z1 − z0
= − log |z − z0 | − log | log |z − z0 || + o(1).

Proposition 4.1.5. Let R ⊂ C be an hyperbolic Riemann surface with punctures


{zi }i∈I˜. Let ϕR be the canonical solution of Liouville equation on R, then near
every puncture, ϕR has the asymptotic behavior

ϕR (z, z̄) = − log |z − zi | − log | log |z − zi || + o(1), as z → zi .


4.1. Asymptotic behavior of the solutions of Liouville’s equation 53

Proof. Let’s suppose that R is not a punctured disk nor a twice punctured plane.
In this case if we consider the puncture zi , there exists R > 0 and z1 ∈ C such
that D0z0 ,R ⊂ R ⊂ C \ {z0 , z1 }.
Using corollary (4.1.3) we obtain that

ϕC\{z0 ,z1 } (z, z̄) ≤ ϕR (z, z̄) ≤ ϕD0z (z, z̄).


0 ,R

Therefore ϕR (z, z̄) has the asymptotic behavior

ϕR (z, z̄) = − log |z − z0 | − log | log |z − z0 || + o(1).

The same theorem holds true in the case of a generic Riemann surface with
punctures

Theorem 4.1.6 (Asymptotic Behavior near the puncture). Let R be an hy-


perbolic Riemann surface with punctures. Let {(Uα , zα )}α∈I be an atlas of R
containing the charts {(Ui , zi )}i∈I˜ corresponding to the punctures.
Then if {ϕα,R }α∈I is the canonical solution of Liouville equation on R, we have
that
ϕi,R (z, z̄) = − log |z| − log | log |z|| + o(1), ˜
∀i ∈ I.
S
Proof. Let R̂ = R ∪ i∈I˜{pi } be the Riemann surface obtained adding the
punctures to R. We S consider  the universal covering Jˆ : S → R̂.
0 −1 0
Let S = S \ J i∈I˜
{pi } , S is an hyperbolic Riemann surface  with punctures.
As a matter of fact since J is a covering map J −1
S
{p } is a collection of
S i∈I˜ i
isolated points in S, therefore for every zi ∈ J −1 i∈I˜
{p i } , there exists R > 0
such that Dz0 i ,R ⊂ S 0 . If S 0 was not hyperbolic then the universal covering
map JS 0 : T → S 0 would have a universal covering space T 6= H. In this case
ˆ S 0 ◦ JS 0 : T → R is the universal covering map of R and since R is hyperbolic
J|
T = H.
Let JS 0 : H → S 0 be the universal covering map of S 0 , one can chose it in
such a way that JR = J| ˆ S 0 ◦ JS 0 is the universal covering map of R.
Let’s consider the puncture pi on R and the corresponding map zi : Ui → D \ {0}.
The canonical solution of Liouville equation (2.2) relative to the chart (Ui , zi ) is
given by

|(JR−1 ◦ zi−1 )0 (z)|


ϕi,R (z, z̄) = log
=(JR−1 ◦ zi−1 )(z)
|(J −1 ˆ −1
0 ◦ J| 0 ◦ zi
−1 0
) (z)|
S S
= log
=(J −1 ˆ −10 ◦ zi−1 )(z)
◦ J|
S0 S
ˆ −10 ◦ zα−1 )(z, z̄) + log |(J|
= ϕS 0 ◦ (J| ˆ −10 ◦ zi−1 )0 (z)|,
S S

where ϕS 0 is the canonical solution of Liouville equation on S 0 .


The map f = J|−1 −1 ˆ −10 can be extended to an holomorphic
S 0 ◦ zi : D \ {0} → J| S
map defined in D by setting f (0) := zi = Jˆ−1 (pi ). From the previous proposition
we know that

ϕS 0 (z, z̄) = − log |z − zi | − log | log |z − zi || + o(1).


54 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

Since f (z) = zi + f 0 (0)z + h(z)z 2 , we obtain

ϕi,R (z, z̄) =


= − log |f 0 (0)z + z 2 h(z)| − log | log |f 0 (0)z + z 2 h(z)|| + log |f 0 (0) + z h̃(z)|
0
f (0)z + z 2 h(z)
= − log
− log | log |z| + log |f 0 (0) + zh(z)|
f 0 (0) + z h̃(z)
0 0

− log 1 + log |f (0) + zh(z)|
f (0) + zh(z)
= − log |z| − log | log |z|| − log
f 0 (0) + z h̃(z) log |z|
= − log |z| − log | log |z|| + o(1).

Remark If another chart (Uα , zα ) contains one or more punctures, then if zi is


the corresponding puncture in zα (Uα ) one obtains that near zi

ϕα,R = − log |z − zi | − log | log |z − zi | + o(1).

This follows from the connection relation


0
ϕα,R = ϕi,R ◦ giα (z) + log |giα (z)|, z ∈ zα (Ui ∩ Uα ).

As a matter of fact giα (z) can be extended to an holomorphic function in zi


0
by setting giα (zi ) = 0. Therefore we can write it as giα (z) = giα (zi )(z − zi ) +
2
(z − zi ) h(z). If we insert this expression in the formula for ϕα,R we obtain the
desired asymptotic behavior for ϕα,R .

4.1.1 Uniqueness of solutions to Liouville equation


Now we prove S
that on a compact punctured Riemann surface, i.e. a surface such
that R̂ = R ∪ i∈I˜{pi } is compact, the solution of Liouville equation is unique

Theorem 4.1.7. Let R be an hyperbolic compact Riemann surface with punc-


tures. Let {(Uα , zα )}α∈I be an atlas of R containing the charts {(Ui , zi )}i∈I˜
corresponding to the punctures. In this case there exists a unique solution to
Liouville equation on R, such that for all i ∈ I˜ it has the asymptotic behavior

ϕi (z, z̄) = − log |z| − log | log |z|| + o(1), z → 0.

Proof. We already know that the canonical solution of Liouville equation satisfies
the required asymptotic behavior.
We want to prove that this solution is unique. Let {ϕα }α∈I and {ψα }α∈I be two
solutions of Liouville equation with the required asymptotic behavior. We set
uα (p) = ϕα (zα , z̄α )(p) − ψα (zα , z̄α )(p) for every α ∈ I. We see that uα : Uα → R,
and on Uα ∩ Uβ one has that

uβ (p) = [ϕβ (zβ , z̄β ) − ψβ (zβ , z̄β )](p)


0 0
= [ϕα (zα , z̄α ) + log |gαβ | ◦ zβ − ψα (zα , z̄α ) − log |gαβ | ◦ zβ ](p)
= uα (p).
4.1. Asymptotic behavior of the solutions of Liouville’s equation 55

Hence u defines a twice differentiable


S function on R. Moreover it defines a
continuous function on R̂ = R ∪ i∈I˜{pi }. As a matter of fact, near the
punctures
u|Ui (p) = − log |zi (p)| − log | log |zi (p)|| + o1 (zi (p))
+ log |zi (p)| + log | log |zi (p)|| − o2 (zi (p))
=o1 (zi (p)) − o2 (zi (p))
Therefore pi is a removable singularity for u and u(pi ) = 0.
Since R̂ is compact the function u attains its maximum and minimum at some
points pM and pm . Let’s suppose that u(pM ) > 0, then pM ∈ Uα for some α ∈ I,
then zM = zα (pM ) is a maximum of the function
vα (z, z̄) = uα ◦ zα−1 (z, z̄) = ϕα (z, z̄) − ψα (z, z̄), z ∈ zα (Uα ).
Since zM is a maximum and vα |zM > 0, then zM is not a puncture and vα is
twice differentiable in zM . In this case ∆vα (zM , z̄M ) ≤ 0, then
0 ≥ ∆vα |zM
= (∆ϕα − ∆ψα )|zM
= (e2ϕα − e2ψα )|zM
> 0.
Therefore maxp∈R̂ u(p) = 0. In a similar way we can prove that minp∈R̂ u(p) = 0,
from this it follows that u(p) = 0 and therefore ϕα (z, z̄) = ψα (z, z̄) for every
α ∈ I.

4.1.2 Solution with periodic condition


We proved that on an hyperbolic compact punctured Riemann surface there
exist a unique solution of Liouville equation with asymptotic behavior near the
punctures of the type ϕi = − log |z| − log | log |z|| + o(1), therefore this solution
coincide with the canonical solution of Liouville equation defined in (2.2).
In order to find such a solution one should in principle consider an atlas for
R and work locally on every chart. This procedure can be very complicated,
therefore we want to simplify it.
GivenSthe Riemann surface R we consider the compact Riemann surface
R̂ = R ∪ {pi } and S it’s universal covering Jˆ : S → R̂.
ˆ ˆ
Let S = S \ J ( i pi ) then J|S 0 : S 0 → R is an analytic covering map for R.
0 −1

One can notice that for every Möbius transformation γ ∈ Deck(J) ˆ it’s restriction
0 0 0
γ|S 0 : S → S is a covering map of S .
Theorem 4.1.8. Let R be an hyperbolic compact punctured Riemann surface
and let S 0 be defined as above. Then there exists a unique solution of Liouville
equation on S 0 which satisfies both the periodic condition
ϕ ◦ γ(z, z̄) = ϕ(z, z̄) − log |γ 0 (z)|, ˆ
∀z ∈ S 0 , ∀γ ∈ Deck(J), (4.2)
and the boundary conditions
!
[
ϕ(z, z̄) = − log |z − zi | − log | log |z − zi || + o(1), ∀zi ∈ Jˆ−1 {pi } .
i
56 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

This solution coincides with the canonical solution of Liouville equation ϕS 0 on


S0.
Proof. Firstly we prove that the canonical solution of Liouville equation satisfies
the conditions of the theorem.
Let JS 0 : H → S 0 be the universal covering map of S 0 , then γ −1 ◦ J is an universal
covering of S 0 . In proposition (2.1.1) we proved that the definition of canonical
solution of Liouville equation doesn’t depend from the specific choice of the
universal covering map. Since S 0 ⊆ Ĉ, ϕS 0 is given by
|(J˜−1 )0 (z)|
ϕS 0 (z, z̄) = log ,
=(J˜−1 (z))
where J˜ is any universal covering map of S 0 .
Thus one has that
|(JS−1 0
0 ◦ γ) (z)|
ϕS 0 (z, z̄) = log −1 = ϕS 0 ◦ γ(z, z̄) + log |γ 0 (z)|.
=(JS 0 ◦ γ(z))
Since we already know that the canonical solution of Liouville equation has the
desired boundary behavior near the punctures, this prove the existence of a
solution with the desired behavior.
For what regards the uniqueness, one can notice that if ϕ(z, z̄) is a solution
of Liouville equation on S 0 with the desired properties, then it is connected with
the canonical solution of Liouville equation on R.
Let’s take any branch of J| ˆ −10 and let {(Uα , zα )}α∈I be an atlas of R. The
S
collection of functions
ˆ −10 ◦ zα + log |(J|
ϕ̃α = ϕ ◦ J| ˆ −10 ◦ zα )0 |,
S S

is a solution of Liouville equation on R.


The expression of ϕ̃α doesn’t depend from the choice of the branch of J|−1 S 0 . If
J2−1 and J2−1 are two different branches of J|−1 ˆ
S 0 , then there exist a γ ∈ Deck(J)
such that
J2−1 = γ ◦ Jˆ1−1 .
Thus one has that
ϕ ◦ J2−1 ◦ zα + log |(J2−1 ◦ zα )0 | =
= ϕ ◦ γ ◦ Jˆ1−1 ◦ zα + log |(γ ◦ Jˆ1−1 ◦ zα )0 |
((
= ϕ ◦ Jˆ1−1 ◦ zα − ( 0 (ˆ
◦ zα | + log |(Jˆ1−1 ◦ zα )0 |.
0 (( ((
log(|γ( ◦ J1−1( ◦ zα | + (
log(|γ( −1(
◦ J(
1

On the chart corresponding to the punctures this solution behaves like ϕi (z, z̄) =
− log |z| − log | log |z|| + o(1), therefore since there is unique solution of Liouville
equation on R with that asymptotic behavior ϕ̃α = ϕα,R for all α ∈ I.
Let’s suppose there is another solution ψ(z, z̄) of Liouville equation on S with
the same properties of ϕ, then
ˆ −10 ◦ zα + log |(J|
ϕα,R = ψ ◦ J| ˆ −10 ◦ zα )0 |,
S S

ˆ −10
Therefore for every α ∈ I and every branch of J|S

ˆ −10 ◦ zα = ϕ ◦ J|
ψ ◦ J| ˆ −10 ◦ zα ,
S S

This means that ϕ = ψ on S 0 .


4.2. The punctured torus 57

4.2 The punctured torus


In this section we study the Riemann surface R = T \ {z1 . . . zn }. In this case
instead of working with local charts, it is more convenient working on the
universal covering of R̂ = T.
We will first discuss briefly the complex structure of a generic torus.
Proposition 4.2.1 (Complex Tori). Let ω1 and ω2 be two complex numbers
which are linearly independent on the real field. We define the lattice

Λω1 ,ω2 = {mω1 + nω2 | m, n ∈ Z}.

We say that two complex numbers z and ẑ are equivalent mod Λω1 ,ω2 if and
only it z − ẑ ∈ Λω1 ,ω2 . The quotient Tω1 ,ω2 = C/Λω1 ,ω2 is a compact Riemann
surface, homeomorphic to a torus.
Proof. First of all we must endow Tω1 ,ω2 with a topology. The natural choice is
to use the quotient topology. More in detail if π : C → Tω1 ,ω2 is the projection
π(z) = [z], then we say that U is open in Tω1 ,ω2 if π −1 (U ) is an open set in
C. The map S π is open, as a matter of fact if V is an open set in C then
π −1 (π(V )) = ω∈Λω ,ω (V + ω) is open, and therefore π(V ) is open in the
1 2
quotient topology.
Secondly we have to give Tω1 ,ω2 a complex structure. This can be done in
the following way. Let 0 < ε < min{|ω1 |, |ω2 |} and for every zα ∈ C consider an
open disk Dα centered in zα with radius ε. The map π|Dα : Dα → π(Dα ) is an
homeomorphism. Let ϕα : π(Dα ) → Dα be it’s inverse. One can easily check
that {(π(Dα ), ϕα )}zα ∈C defines a complex structure on the complex torus.
From its construction Tω1 ,ω2 is homeomorphic to a topological torus. It is
compact because it is covered by the image under π of the compact set

{αω1 + βω2 | α, β ∈ [0, 1]}.

The projection map π : C → Tω1 ,ω2 given by π(z) = [z] is an analytic covering
map with a simply connected domain, therefore it is the universal covering map
of Tω1 ,ω2 . The Deck transformation group is Deck(J) = hz + ω1 , z + ω2 i.
Any automorphism of C is of the form f (z) = az + b, if we consider only the
one such that f (0) = 0, therefore the ones with b = 0, we obtain that the map
f˜ : Tω1 ,ω2 → Taω1 ,aω2 given by f˜([z]) = [az] is a conformal equivalence between
the two tori. If one take τ = ωω21 such that =(τ ) > 0 the Torus Tτ := T1,τ is
isomorphic to Tω1 ,ω2 . Moreover the following proposition holds true
Proposition 4.2.2. Let τ, τ 0 ∈ H, then Tτ and Tτ 0 are conformally equivalent
if and only if
aτ 0 + b
 
a b
τ= 0 , ∈ SL(2, Z).
cτ + d c d
Proof. Firstly we assume that there exist a biholomorphic map f : Tτ → Tτ 0 .
From the definition of Universal covering, there exist a map f˜ : C → C such that
f ◦ Jτ = Jτ 0 ◦ f˜. Thus f˜ = αz + β for some α, β ∈ C. Moreover, we may assume
that β = 0.
58 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

Since f [z]τ = [αz]τ 0 , we have that [ατ ]τ 0 = [α]τ 0 = [0]τ 0 . Therefore there exist
a, b, c, d ∈ Z such that

ατ = aτ 0 + b,
α = cτ 0 + d.

Hence
aτ 0 + b
τ= ,
cτ 0 + d
Applying the same argument to f −1 we find that

âτ + b̂
τ0 = ,
ĉτ + dˆ
Therefore ad − bc = ±1, the sign must be positive in order to have that both
0
=(τ ) > 0 and =(τ 0 ) > 0. Conversely, if τ = aτ +b
cτ 0 +d
, then a biholomorphic map
between the two tori is given by f [z] = [(cτ + d)z].
One can identify the space of all complex tori with the quotient space

H/SL(2, Z).

Let T = Tω1 ,ω2 \ {p1 , . . . , pn } be a punctured Torus. In view of what we said in


the last section, we consider the universal covering of T̂ = Tω1 ,ω2 given by (C, π).
The canonical solution of Liouville equation on S = C \ π −1 ({p1 , . . . , pn }) is
doubly periodic. As a matter of fact every γ ∈ Deck(π) has the form γ(z) =
z + mω1 + nω2 . Using formula (4.2) we obtain that

ϕS 0 (z + mω1 + nω2 ) = ϕS 0 (z).

The canonical connection ωS 0 is a doubly periodic meromorphic function on


C with poles of order 2 at every point of π −1 ({p1 , . . . , pn }). Since it is doubly
periodic, there exist a set of accessory parameters {c1 , . . . , cn } such that for every
zi ∈ π −1 (pi ) the canonical connection, near zi has the asymptotic behavior
1 ci
ωS 0 (z) = + + O(1)
4(z − zi ) 2 2(z − zi )
In addition to this there is another one accessory parameter, that we will denote
with E, such that the canonical connection is written as
n
X X  1 ci

ωS (z) =
0 + + E,
i=1 2
4(z − zi − mω1 − nω2 )2 2(z − zi − mω1 − nω2 )
m,n∈Z

where the zi are representatives for π −1 (pi ). This is not a function, since the
summation does not converge, but the formula gives the idea of the asymptotic
behaviors of the canonical connection near the punctures of S 0 .
It is possible to prove that there is a relation on the accessory parameters ci .
Proposition 4.2.3. The accessory parameters of the punctured torus satisfy
the relation n
X
ci = 0.
i=1
4.2. The punctured torus 59

Proof. Let’s consider a contour C which is a parallelogram with vertexes ẑ, ẑ +


ω1 , ẑ + ω2 and ẑ + ω1 + ω2 such that it has no poles on his edges. The contour
C contains one representative for every set π −1 (pi ). The integral
Z
ωS 0 (z)dz = 0,
C

since the function is doubly periodic.


Moreover using the Residue Formula, one obtain that
Z n
X
0= ωS 0 (z)dz = 2πi ci .
C i=1

If the torus has a unique puncture, one can choose the universal covering of T̂
in such a way that the poles of ωS 0 are located on the lattice {mω1 + nω2 | m, n ∈
Z}. In this case the relation on the accessory parameters reduces to c1 = 0,
therefore the function ωS 0 has the form
1
ωS 0 = ℘(z) + E,
4
where the function ℘(z) is the Weierstrass elliptic function, which has the form

1 X0  1 1

℘(z) = 2 + − . (4.3)
z ω∈Λ
(z − ω)2 ω2
ω1 ,ω2

4.2.1 Accessory parameter of a 1-punctured torus


In this section we establish a relation between the accessory parameter of the
1-punctured torus and the accessory parameter of the 4-punctured sphere.
The canonical connection of the sphere with 4 missing point X = C\{0, 1, ∞, 1/a}
with a ∈ C \ {0, 1} is
3
X 1 ci
ωX (z) = + ,
i=1
4(z − zi )2 2(z − zi )

where z1 = 0, z2 = 1 and z3 = 1/a.


Using relations (2.4) we find that
 
1
c1 = 1 + c −1 ,
a
c
c2 = −1 − ,
a
c3 = c.

Since a 6= 0, 1 there exist τa ∈ H such that λ(τa ) = a, where λ is the modular


lambda function, which is the universal covering map of C\{0.1}. One can notice
that the lattice Λ1,τ doesn’t depend from the particular choice of τa ∈ λ−1 (a)
Let’s take any two complex numbers ω1 , ω2 such that ω1 /ω2 = τ and let’s consider
60 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

the Weierstrass elliptic function (4.3) on the lattice Λω1 ,ω2 .


Let ω  ω   
1 2 ω1 + ω2
e1 = ℘ e2 = ℘ e3 = ℘ ,
2 2 2
It is possible to prove that the modular lambda function can be written as
e3 − e2
λ(τ ) = ,
e1 − e2
thus if τa ∈ λ−1 (a)
e3 − e2
a= .
e1 − e2 τ =τa
Let ω1 = 2 and ω2 = 2τa . Let ℘2 (z) be the Weierstrass elliptic function with
periods ω1 and ω2 , and ℘1 (z) be the Weierstrass elliptic function with periods
ω1 /2 and ω2 /2. If we consider the function
e1 − e2
f (z) = ,
℘2 (z) − e2
defined on S 0 = C \ Λ1,τ , it takes values in the punctured sphere X.
Lemma 4.2.4. The function f : S 0 → X is a covering map.
Proof. Since ℘2 is a doubly periodic function, it can be seen as a function defined
from
℘2 : C/Λ2,2τ → Ĉ.
This function attains every complex value two time. Since the derivative of ℘2 (z)
satisfies the relation
(℘02 (z))2 = 4(℘2 (z) − e1 )(℘2 (z) − e2 )(℘2 (z) − e3 ),
and the point [0] is the only pole and it is of order two, the point on the torus
with multiplicity equal to 2 are only [0], [1], [τ ], [1 + τ ].
The function f is a combination between ℘2 and an automorphism of Ĉ, therefore
it can be seen as a function defined on the torus f : C/Λ2,2τ → Ĉ. Also f attains
every complex value two time and the only points with multiplicity 2 are [0], [1], [τ ]
and [1+τ ]. If we consider the restriction of f on T 0 = C/Λ2,2τ \{[0], [1], [τ ], [1+τ ]},
we find that
f |T 0 : T 0 → X = C \ {0, 1, 1/a},
and that every value in X is attained 2 times. Hence f is surjective.
In order to see that it is a covering map, it suffices to check that the derivative
of f evaluated in every point of S 0 is a complex number different from 0.
The derivative of f is given by
e1 − e2
f 0 (z) = ℘0 (z)
(℘2 (z) − e2 )2 2
the square of f 0 is
(℘2 (z) − e1 )(℘2 (z) − e3 )
(f 0 (z))2 = (e1 − e2 )2 ,
(℘2 (z) − e2 )3
thus we see that it never vanishes on S 0 nor it goes to infinity. Therefore f is a
covering map.
4.2. The punctured torus 61

It is possible to establish a connection between the accessory parameter E of


the 1-punctured torus and the accessory parameter c of the 4-punctured sphere.
Lemma 4.2.5. Let τ ∈ H and let ℘1 and ℘2 be the Weierstrass elliptic function
with periods {1, τ } and {2, 2τ } respectively. We have that
3
S (℘2 )(z) = − ℘1 (z).
2
Proof. Let’s consider the relation

(℘02 (z))2 = 4(℘2 (z) − e1 )(℘2 (z) − e2 )(℘2 (z) − e3 ),

and let’s take the derivatives of both sides. We obtain that


X
℘002 (z) = 2 (℘2 (z) − ei )(℘2 (z) − ej ).
1<i<j<3

If we derive another time both sides, we obtain that

℘000 0 0
2 (z) = 12℘2 (z)℘2 (z) − 4℘2 (z)(e1 + e2 + e3 ).

It is well known that e1 + e2 + e3 = 0 therefore


℘000
2 (z)
= 12℘2 (z).
℘02 (z)
Moreover
2
℘002 (z)

3 3 X (℘2 (z) − ei )(℘2 (z) − ej )
− =− − 12℘2 (z).
2 ℘02 (z) 2 1<i<j<3 ℘2 (z) − ek

hence
3 X (℘2 (z) − ei )(℘2 (z) − ej )
S (℘2 )(z) = − + 3℘2 (z).
2 1<i<j<3 ℘2 (z) − ek

The function we obtained is a meromorphic elliptic function with periods {2, 2τ }


and poles on the lattice Λ1,τ . The poles at the points of sublattice Λ2,2τ are of
order two and the series expansion of S (℘2 )(z) near those points has the form
3 1 1
S (℘2 )(z) = − + h(z),
2 (z − zi )2 z − zi
where h(z) is an holomorphic function near zi ∈ Λ2,2τ .
Near the point of the sublattice 1/2 + Λ2,2τ the only term which goes to infinity
is
3 (℘2 (z) − e2 )(℘2 (z) − e3 )
p(z) = − .
2 ℘2 (z) − e1
If we expand the denominator near 1/2 we find that
1 00
℘2 (z) − e1 = ℘ (1/2)(z − 1/2)2 + o[(z − zi )2 ].
2 2
Since ℘002 (1/2) = 2(e1 − e2 )(e1 − e3 ), the expansion of p near z = 1/2 is given by
3 1 1
p(z) = − + h̃(z) .
2 (z − 1/2) 2 z − 1/2
62 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

Therefore S (℘2 )(z) has an asymptotic behavior of the type − 2(z−z 3


i)
2 at the

points zi ∈ 1/2 + Λ2,2τ .


In the same way we prove that S (℘2 )(z) has the same asymptotic behavior at
the points of the sublattices τ /2 + Λ2,2τ and (1 + τ )/2 + Λ2,τ . We obtain that
the Schwartzian derivative of ℘2 can be written as
3
S (℘2 )(z) = − ℘1 (z) + H(z),
2
where H(z) is a meromorphic elliptic function with poles of degree at most 1,
therefore H(z) is constant on C.
In order to find the explicit value of H = H(z) we must find first a better
expansion of S (℘2 )(z) near z = 1/2.
Since ℘2 is even its expansion near z0 = 1/2 is given by

℘002 (z0 ) ℘(4) (z0 )


℘2 (z) = e1 + (z − z0 )2 + 2 (z − z0 )4 + o[(z − z0 )4 ]
2 4!
We set ζ = z − z0 , using the fact that e1 = −e2 − e3 , we obtain that

(℘2 (z) − e2 )(℘2 (z) − e3 )


p̃(z) =
℘2 (z) − e1
 
℘00
2 (z0 ) ℘00
2 (z0 )
(e1 − e2 + 2
ζ 2 + o(ζ 2 )) e1 − e3 + 2
ζ 2 + o(ζ 2 )
= (4)
℘00
2 (z0 ) ℘2 (z0 )
2
ζ2 + 4!
ζ 4 + o(ζ 4 )
℘00
2 (z0 ) ℘00
2 (z0 )
2
+ (2e1 − e3 − e ) ζ 2 + o(ζ 2 )
2 2
=  (4) 
℘00
2 (z0 ) 2 ℘2 (z0 ) 2 2
2
ζ 1 + 12℘ 00 (z ) ζ + o(ζ )
0 2

1 3e1 ζ 2 + o(ζ 2 )
=  (4) +  (4) .
℘2 (z0 ) ℘2 (z0 ) 2
ζ2 1 + 12℘00
ζ 2 + o(ζ 2 ) ζ 2 1 + 12℘ 2
00 (z ) ζ + o(ζ )
2 (z0 ) 0 2

If we derive the formula ℘000 0


2 (z) = 12℘2 (z)℘2 (z) we get

℘(4) 0 2 00
2 (z) = 12 (℘2 (z) + ℘(z)℘ (z)) ,

(4)
℘ 0 (z )
therefore 12℘ 00 (z ) = ℘2 (z0 ) = e1 .
2
2 0
Thus one obtains that
1 1
p̃(z) = 2
− e1 + 3e1 + o(1) = 2 + 2e1 + o(1).
ζ ζ

In the end, if we substitute this expression into the one of S (℘2 )(z) we get
3 1
S (℘2 )(z) = − −
3e
1 +
3e
1 + o(1),
2 (z − z0 )2

therefore H = 0, hence the lemma is proved.


4.2. The punctured torus 63

Theorem 4.2.6 (Connection formula). Let X = C \ {0, 1, 1/a, ∞} and S 0 =


C \ Λ1,τa , with λ(τa ) = a. Let ωX and ωS 0 be the canonical connection of the two
Riemann surface. Let f (z) = (e1 − e2 )/(℘2 (z) − e2 ), then
1
ωS 0 (z) = ωX ◦ f (z)(f 0 (z))2 + S (f )(z).
2
If c is the accessory parameter of X and E is the accessory parameter of S 0 we
obtain the relation
(e1 − e2 )(e1 − e3 )
E = 2e1 + 2c .
e3 − e2
Proof. Let JS 0 : H → S 0 be the universal covering map of S 0 . Since f is a
covering map, the function JX = f ◦ JS 0 is a covering map of X. Thus if we
chose the branches of JS−1 0 and of JX−1
in such a way that JS−1 −1
0 = JX ◦ f , the

canonical solution of Liouville equation on S 0 can be written as


|(JS−1 0
0 ) (z)|
ϕS 0 (z, z̄) = log −1
=(JS 0 (z))
−1
|(JX ◦ f )0 (z)|
= log −1
=(JX ◦ f (z))
= ϕX ◦ f (z, z̄) + log |f 0 (z)|.
Therefore the canonical connection of S 0 and X are connected by the equation
1
ωS 0 (z) = ωX ◦ f (z)(f 0 (z))2 + S (f )(z).
2
Since f (z) = γ ◦ ℘2 (z) and γ = (e1 − e2 )/(z − e2 ) is a Möbius transformation,
from lemma above
1 1 3
S (f )(z) = S (℘2 )(z) = − ℘1 (z).
2 2 4
The canonical connection of S 0 takes the form ωS0 = 41 ℘1 (z) + E, thus
E = ωX ◦ f (z)(f 0 (z))2 − ℘1 (z).
Since E is a constant, we evaluate the limit of the RHS for z → 0. The canonical
connection ωX is written as
2
X 1 ci
ωX = + ,
i=1
4(z − zi ) 2 2(z − zi )
The square of the derivative of f is given by
(e1 − e2 )2 0 2
(f 0 (z))2 = ℘ (z)
(℘(z) − e2 )4 1
(℘(z) − e1 )(℘(z) − e3 )
= 4(e1 − e2 )2
(℘(z) − e2 )3
Thus f 0 (z) → 0 when z → 0. Therefore we have that
2  
X 1 ci
E = lim + f 0 (z)2 − ℘1 (z)
z→0
i=1
4(f (z) − z i)
2 2(f (z) − z i)

1 c1
= lim f 0 (z)2 − ℘1 (z) + f 0 (z)2
z→0 4f (z)2 2f (z)
64 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION

The first term in the limit can be written as


1 (℘1 (z) − e1 )(℘1 (z) − e3 )
(f 0 (z))2 =
4f (z)2 ℘1 (z) − e2

If we expand it at the origin we obtain that

(℘1 (z) − e1 )(℘1 (z) − e3 ) 1


z4
− ze12 − ze32 + O(1)
= 1
℘1 (z) − e2 z2
(1 − e2 z 2 + O(z 4 ))
 
1
= − e1 − e3 + O(z ) (1 + e2 z 2 + O(z 4 ))
2
z2
1
= − e1 − e3 + e2 + o(1)
z2
1
= 2 + 2e2 + o(1).
z
When we take the limit as z → 0 the 1/z 2 cancels with the function ℘1 (z). It
remains to find
c1 (℘1 (z) − e1 )(℘1 (z) − e3 )
lim (f 0 (z))2 = lim 2c1 (e1 − e2 )
z→0 2f (z) z→0 (℘1 (z) − e2 )2
  
1
=2 1+c −1 (e1 − e2 ).
a

Hence we obtained that


  
1
E = 2e2 + 2 1 + c −1 (e1 − e2 )
a
(e1 − e2 )(e1 − e3 )
= 2e1 + 2c .
e3 − e2

4.3 Liouville equation as Euler Lagrange equa-


tion of a functional
Let’s consider a n-punctured sphere X = C \ {z1 , . . . , zn3 , 0, 1}. Let’s consider
the space of function on X
( ( )
− log |z − zi | − log | log |z − zi || + o(1) z → zi
A = ψ ∈ C (X) ψ(x) =

− log |z| − log log |z| + o(1) z→∞

This is an affine space over the vector field

V = {ψ ∈ C ∞ (X, R) | ψ(x) = o(1) z → zi }

On A we define the functional


Z   
2 1 2ψ πn
S[ψ] = lim |ψz | + e dxdy + (log r + 2 log | log r|) , (4.4)
r→0
Xr 4 2
4.3. Liouville equation as Euler Lagrange equation of a functional 65

where !
n−1 
[ 1
Xr = X \ {z : |z − zi | < r} ∪ z : |z| > .
i=1
r

We may notice that the term πn 2


(log r + 2 log | log r|) is a corrective term that
guarantees the convergence of the functional.
Theorem 4.3.1. The extremals of the functional S[ψ] are all and only the
functions in A which satisfy Liouville equation

∆ϕ = e2ϕ .

Proof. Let δψ ∈ V , then

δS[ψ] = S[ψ + δψ] − S[ψ]


Z
1 2ψ+2δψ 
= lim |ψz + δψz |2 − |ψz |2 + e − e2ψ dxdy
r→0 4
ZXr
1 2ψ+2δψ 
= lim ψz δψz̄ + δψz ψz̄ + δψz δψz̄ + e − e2ψ dxdy
r→0
Xr 4

Using integration by part and the fact that δψ|δXr → 0, we obtain that
Z  
1
δS[ψ] = lim −2ψzz̄ + e2ψ δψ + O(δ 2 ψ)dxdy.
r→0
Xr 2
δS[ψ]
If ϕ is an extremal of S it must satisfy δψ
= 0. Thus ∆ϕ = 4ϕzz̄ = e2ϕ .
Liouville equation admits a unique solution in A , therefore the functional S
admits a unique extremal, and it is ϕX .
66 Chapter 4. SOLUTIONS OF LIOUVILLE EQUATION
Chapter 5

Teichmüller spaces

In the last chapter of this thesis we present a theorem proved in the article [18]
by Zograf and Takhtadzhyan. Those two mathematicians where able to find a
potential for the accessory parameters, written using the functional (4.4).
In order to prove this result, it is required some knowledge of the theory of
quasiconformal mapping and Teichmüller spaces. In the first part of this chapter
we present them. For more detailed information we refer to [1] or [8].

5.1 Beltrami equation


In this section we show that, given a certain µ ∈ L∞ (C) with kµk∞ < 1, there
exists a unique solution of the so-called Beltrami equation

fz̄ = µfz

fixing 0, 1 and ∞.
The function f belongs to the set of quasiconformal mappings.

5.1.1 Quasiconformal mapping


Definition Let f be an homeomorphism of a domain Ω onto C which preserves
orientation. We say that f is a quasiconformal mapping on Ω if it satisfies the
two following conditions:
(i) The distributional partial derivatives of f with respect to z and z̄ can be
represented by locally integrable functions fz and fz̄ on Ω.
(ii) There exists a constant k with 0 ≤ k < 1 such that

|fz̄ | ≤ k|fz | a.e. on Ω.

Let K = (1 + k)/(1 − k), we say that f is K − quasiconf ormal or equivalently


K − qc on Ω.

Notice that K is not unique since it depends from the choice of k. If kf is the
minimum of the k for which fz̄ ≤ kfz a.e on Ω, then Kf is the smallest value for
which f is Kf − qc, moreover f is K − qc for all K ≥ Kf .

67
68 Chapter 5. TEICHMÜLLER SPACES

Proposition 5.1.1. Let g be a conformal mapping from Ω to Ω0 , and let f be a


K − qc mapping on Ω0 . Then f ◦ g is a K − qc mapping on Ω.

Proof. The distributional derivatives of f ◦ g with respect to z and z̄ exist and


are given by

(f ◦ g)z = (fz ◦ g)g 0 ,


(f ◦ g)z̄ = (fz̄ ◦ g)g¯0 .

Therefore f ◦ g is a K − qc mapping on Ω.

Proposition 5.1.2 (Weyl’s lemma). Let f be a continuous function on Ω whose


distributional derivative fz̄ is locally integrable on Ω. If fz̄ = 0 in the sense of
distributions on Ω, then f is holomorphic on Ω.

Proof. Let Ω1 be a relatively compact subdomain of Ω. We can construct a


sequence {fn }∞ ∞
n=1 ⊂ C0 (Ω) such that on every compact subset K of Ω, fn
converges uniformly to f and such that
Z
lim |(fn )z̄ − fz̄ |dx dy = 0,
n→∞
ZK
lim |(fn )z − fz |dx dy = 0.
n→∞
K

As a matter of fact let η ∈ C0∞ (Ω) be a function identically equal to 1 in a


neighborhood of K. Then the sequence of function fn = ϕn ∗ (ηf ), where
(  
1
C · exp − 1−|z|2 |z| < 1
ϕ(z) = ,
0, |z| ≥ 1

and ϕn (z) = n2 ϕ(nz), satisfies the desired properties.


(ηf )z̄ = 0 in some neighborhood of K, then if n is big enough

(fn (z))z̄ = ϕn ∗ (ηf )z̄ = 0, on int(K).

Thus fn is holomorphic on K for n big enough, and it converges uniformly to f .


This implies that f is holomorphic on int(K), hence since K was arbitrary, f is
holomorphic on Ω.

Corollary 5.1.3. A 1 − qc mapping is conformal.

The following proposition give us more information on the inverse and the
composition of K − qc mappings.

Proposition 5.1.4. The following are true

• The inverse of a K − qc mapping is again a K − qc mapping.

• For every K1 − qc mapping f : Ω 7→ Ω0 and every K2 − qc mapping


g : Ω0 7→ C the composed mapping is K1 K2 − qc.
5.1. Beltrami equation 69

Proof. Firstly we notice that the division by fz ◦ f −1 makes sense. As a matter


of fact the set
E = {ζ ∈ Ω| fz (ζ) = 0},
is measurable. Moreover on every measurable set D
Z
A(f (D)) = (|fz |2 − |fz̄ |2 )dx dy,
D

as a matter of fact the determinant of the jacobian of f is |fz |2 − |fz̄ |2 and it is


positive.
If fz vanishes a.e on E so does fz̄ , therefore A(f (E)) = 0. This means that
the division by fz ◦ f −1 makes sense.
One can notice that
fz̄ ◦ f −1 −1
(f −1 )z̄ = − (f )z
fz ◦ f −1
Therefore if there exists 0 ≤ k < 1 such that |fz̄ | ≤ k|fz | a.e, then a.e is true
that
|fz̄−1 | ≤ k|fz−1 |.
In order to prove the second part of the proposition we will show that the
following inequality is true a.e. in Ω
|(f ◦ g)z̄ (z0 )| + |(f ◦ g)z (z0 )|
≤ K1 K2 . (5.1)
|(f ◦ g)z̄ (z0 )| − |(f ◦ g)z (z0 )|

Let ϕ : Ω̃ → Ω be a conformal mapping such that ϕ : 0 7→ z0 . It is easy to see


that |(f ◦ g ◦ ϕ)z̄ (0)| = |(f ◦ g)z̄ (z0 )||ϕ0 (0)|, and that a similar expression holds
true for the derivative with respect to z. Therefore (5.1) is valid for f ◦ g in z0 if
and only if it is valid for f ◦ g ◦ ϕ in 0.
If h : Ω 3 0 → C is a complex differentiable map, one can always choose two
constant θ1 and θ2 such that if h̃ = eiθ1 h(eiθ2 ) then h̃z (0) and h̃z̄ (0) are both
real positive numbers.
We consider the constants θ1 and θ2 in the case of f ◦ g ◦ ϕ. If we put
F (z) = eiθ1 f (z) and G(z) = g ◦ ϕ(eiθ2 z), by proposition (5.1.1), F is K1 − qc
and G is K2 − qc. Now we prove prove (5.1) for F ◦ G.
Firstly using the derivation rule for complex derivatives we obtain that

(F ◦ G)z̄ = Fz ◦ G · Gz̄ + Fz̄ ◦ G · Gz


(F ◦ G)z = Fz ◦ G · Gz + Fz̄ ◦ G · Gz̄

From those formulae it follows that

((F ◦ G)2z − (F ◦ G)2z̄ )|0 = (|Fz ◦ G|2 − |Fz̄ ◦ G|2 )(|Gz |2 − |Gz̄ |2 )|0

this proves that (F ◦ G)z (0) − (F ◦ G)z̄ (0) > 0 .


Hence it is easy to prove that

(F ◦ G)z (0) + (F ◦ G)z̄ (0) ≤ (|Fz ◦ G| + |Fz̄ ◦ G|)(|Gz | + |Gz̄ )|0


(F ◦ G)z (0) + (F ◦ G)z̄ (0) ≥ (|Fz ◦ G| − |Fz̄ ◦ G|)(|Gz | − |Gz̄ )|0 .
70 Chapter 5. TEICHMÜLLER SPACES

From this it follows


(F ◦ G)z (0) + (F ◦ G)z̄ (0)
≤ K1 K2 .
(F ◦ G)z (0) − (F ◦ G)z̄ (0)

Therefore (5.1) is true for almost every point z0 ∈ Ω. This proves that f ◦ g
is a K1 K2 − qc mapping.

Remark In the previous proposition we saw that if

E = {ζ ∈ Ω| fz (ζ) = 0},

then f (E) is a zero measure set.


Since f −1 is a K − qc mapping, E is a zero measure set too.

It is possible to give a geometrical definition of quasiconformal mapping and to


prove that it is equivalent to the analytic one. The proof of the equivalence of
the two definitions is omitted.

Geometrical definition of quasiconformal mapping Let f be a homeo-


morphism of a domain Ω into C which preserves orientation and let K ≥ 1. We
say that f is a K−quasiconformal mapping if

µ(f (Q)) ≤ Kµ(Q),

for every quadrilateral (Q; a, b, c, d) contained in Ω.

If f is a quasiconformal map such that kf is the infimum of the k such that


|fz̄ | < k|fz |, then since µ(E) = 0, the function

fz̄
µf =
fz

is a bounded measurable function on Ω, and it satisfies

ess sup |µf | ≤ kf < 1


z∈Ω

We call µf the complex dilatation of f on Ω.

Proposition 5.1.5. For every quasiconformal mapping f and g on a domain


Ω, the complex dilatation

fz µg − µf
µg◦f −1 ◦ f = a.e. on Ω. (5.2)
fz 1 − µf µg

Proof. From Proposition (5.1.4) it follows that g ◦ f −1 is quasiconformal on f (Ω),


therefore the expression µg◦f −1 makes sense.
If we apply the chain rule on g = g ◦ f −1 ◦ f , we obtain that

gz = (g ◦ f −1 )w ◦ f · fz + (g ◦ f −1 )w̄ ◦ f · f¯z ,
gz = (g ◦ f −1 )w ◦ f · fz + (g ◦ f −1 )w̄ ◦ f · µf fz
5.1. Beltrami equation 71

and

gz̄ = (g ◦ f −1 )w ◦ f · fz̄ + (g ◦ f −1 )w̄ ◦ f · fz ,


µg gz = (g ◦ f −1 )w ◦ f · µf fz + (g ◦ f −1 )w̄ ◦ f · fz

We can show, as we did in Proposition (5.1.4), that fz (z), gz (z) and (g ◦ f −1 )w (z)
are different from 0 for almost every z ∈ Ω. From the equalities above we obtain
that
gz (1 − µf µg )
(g ◦ f −1 )w ◦ f =
(|µf |2 − 1)fz
gz (µf − µg )
(g ◦ f −1 )w̄ ◦ f = ,
(|µf |2 − 1)fz

hence the ratio of the two expression gives the desired result.
For every quasiconformal mapping f , defined on a domain Ω, there exist a
bounded measurable function µf which satisfies kµf k∞ < 1, where

kµf k∞ := ess sup |µ(z)|.


z∈Ω

Let B(D)1 = {µ ∈ L∞ (Ω)| kµk∞ < 1}, be the space of the Beltrami Differential.
Given µ ∈ B(D)1 we would like to find a quasiconformal mapping f : Ω → C
having µ as complex dilatation, i.e. we would like to solve the so-called Beltrami
equation
fz̄ = µfz .
The solution of this equation is essentially unique.
Proposition 5.1.6. Let µ ∈ B(D)1 , if f is a solution of Beltrami equation for
µ, then for every h : f (Ω) → C conformal, the map h ◦ f is again a solution of
Beltrami equation.
Conversely, if g is another solution of Beltrami equation, then g ◦ f −1 is a
conformal mapping.
Proof. We notice that

(h ◦ f )z̄ = hz ◦ f · fz̄
= µ · h z ◦ f · fz
= µ · (h ◦ f )z .

In order to prove the second part of the proposition, we use formula (5.2), to
obtain that
µg◦f −1 ◦ f = 0,
therefore (g ◦ f −1 )z̄ = 0. From Weyl’s lemma it follows that the function g ◦ f −1
is conformal
In order to find a unique solution to Beltrami equation we must set some
boundary conditions. If Ω = C there exist a unique solutions of Beltrami
equation, with some normalization condition. The proof of this theorem can be
found in [8].
72 Chapter 5. TEICHMÜLLER SPACES

Theorem 5.1.7 (Solutions of Beltrami equation). For every Beltrami coefficient


µ ∈ B(C)1 , there exists a homeomorphism f : Ĉ → Ĉ which is a quasiconformal
mapping of C with complex dilatation µ.
This solution is unique if we impose the following normalization conditions

f (0) = 0, f (1) = 1, f (∞) = ∞.

The function f is called the canonical µ − qc mapping of Ĉ or the canonical


quasiconformal mapping of Ĉ with complex dilatation µ and it is denoted as f µ .
A similar theorem hold true in the case of Ω = H.
Proposition 5.1.8. Let µ be an arbitrary element of B(H)1 . Then there exists
a unique homeomorphism of H = H ∪ R̂ onto itself, satisfying the following
normalization conditions:

w(0) = 0, w(1) = 1, w(∞) = ∞.

Moreover its restriction to H is a quasiconformal mapping of H onto H having


complex dilatation µ.
Proof. The uniqueness follows from the fact that if g is another map with the
same properties, then g ◦ f −1 is an analytic automorphism of H, therefore a
Möbious transformation with three fixed points. Therefore g ◦ f −1 = id.
In order to prove the existence of f µ we take µ̂ ∈ B(C)1 as

µ(z), z ∈ H

µ̂(z) = 0, z∈R

µ(z̄), z ∈ H∗ := C \ H.

Let’s consider the canonical µ̂-qc mapping f µ̂ : Ĉ → Ĉ. We can prove that the
map
g(z) := f µ̂ (z̄),
solve Beltrami equation for µ̂ and fixes 0, 1 and ∞, therefore

f µ̂ (z) = f µ̂ (z̄).

in particular f µ̂ (R̂) = R̂. Since f µ̂ preserves the orientation the restriction


f = f µ̂ |H satisfies the conditions required by the theorem.
The last part of this section is dedicated to the study of the canonical solutions
of Beltrami equation for a certain family of Beltrami coefficients µ(t).
Theorem 5.1.9. Let {µ(t)} ∈ B(C)1 be a family of Beltrami coefficients de-
pending on a real or a complex parameter t. Suppose that kµ(t)k∞ → 0 as t → 0,
and that µ(t) is written in the form

µ(t)(z) = tν(z) + tε(t)(z),

for certain ν ∈ L∞ (C) and ε(t) ∈ L∞ (C) such that kε(t)k∞ → 0 as t → 0. Then
µ(t)
f (ζ) − ζ
f˙[ν](ζ) = lim
t→0 t
5.2. Teichmüller spaces 73

exists for every ζ ∈ C. The convergence is locally uniform on C. Moreover


 
ζ −1
Z
1 1 ζ
f˙[ν](ζ) = − ν(z) − + dxdy. (5.3)
π C z−ζ z−1 z
There is a similar formula when
µ(t)(z) = µ(z) + tν(z) + tε(t)(z), z ∈ C,
where µ ∈ B(Γ01 . In this case
µ(t)
f (ζ) − ζ
f˙µ [ν](ζ) = lim ,
t→0 t
can be written as
f µ (ζ)(f µ (ζ) − 1)(fzµ (z))2
Z
1
f˙µ [ν](ζ) = − ν(z) dxdy. (5.4)
π C f (z)(f µ (z) − 1)(f µ (z) − f µ (ζ))
µ

The proof of these assertions can be found in [8].

5.2 Teichmüller spaces


Let R be an hyperbolic Riemann surface and let J : H → R be the universal
covering map of R. From now on we adopt the notation Γ = Deck(J).
Firstly we introduce the definition of Teichmüller space of the surface R.

Teichmüller space Let R be a Riemann surface. We consider all the pairs


(S, f ) where S is a Riemann surface and f : R → S is a sense-preserving
q.c. mapping. We say that (S1 .f1 ) ∼ (S2 , f2 ) if f2 ◦ f1−1 is homotopic to a
conformal mapping of S1 on S2 . The Teichmüller space of R is the space of all
the equivalence classes [S, f ]. The point [R, idR ] is the initial point of T (R).
Every quasiconformal mapping f : R → S can be lifted to a quasiconformal
mapping f˜ : H → H which satisfies JS ◦ f˜ = f ◦ JR , where JS and JR are the
universal covering maps of S and R. One may assume that 0, 1 and ∞ are fixed
points for some elements of both ΓR \ {id} and ΓS \ {id}. In this case it is possibe
to prove that there is a unique lifting which fixes 0, 1 and ∞. This lifting is
called the canonical lifting of f .
If γ ∈ ΓR then
JS ◦ f˜ ◦ γ ◦ f˜−1 = f ◦ JR ◦ γ ◦ f˜−1
= f ◦ JR ◦ f −1
= JS .
Moreover
0 = ∂z̄ (JS ◦ f˜ ◦ γ ◦ f˜−1 )
= J 0 (f˜ ◦ γ ◦ f˜−1 )∂z̄ (f˜ ◦ γ ◦ f˜−1 ).
S

Since J 6= 0 it follows that f˜ ◦ γ ◦ f˜−1 ∈ Aut(H). Therefore every element in


0
S
T (R) defines an injective homomorphism
ρf˜ : Γ → P SL(2, R),

where ρf˜(γ) = f˜ ◦ γ ◦ f˜−1 .


74 Chapter 5. TEICHMÜLLER SPACES

Lemma 5.2.1. Two quasiconformal mappings f1 , f2 : R → S are homotopic if


and only if ρf˜1 = ρf˜2 .

Proof. Let’s consider the continuous family of functions ft : R → S, with


1 ≤ t ≤ 2, that realize the homotopy between f1 and f2 . Let f˜1 be the canonical
lift of f1 , then there exists a unique continuous lift F̃t such that F̃1 = f˜1 . The
family of function F̃t gives an homotopy between f˜1 and a lift F̃2 of f2 .
Let’s fix an element γ ∈ Γ arbitrarily. The two continuous family of functions
{F̃t ◦ γ| t ∈ [1, 2]} and {f˜1 ◦ γ ◦ f˜1−1 ◦ F̃t | t ∈ [1, 2]} have the same initial point
f˜1 ◦ γ. Moreover they are both lifts of the family of functions {ft }. Since they
have the same initial point, they coincide for all t ∈ [1, 2]. Thus for t = 2 one
has that
ρf˜1 = ρF̃2 .
Let’s consider an hyperbolic element γ0 such that 0 is an attractive fixed point
of γ0 . Then ρF̃2 is still hyperbolic and has 0 as attractive fixed point. Hence
F̃2 (0) = 0. The same is true if γ0 is parabolic. Moreover we can prove that F̃2
fixes also 1 and ∞. This means that F̃2 = f˜2 , then ρf˜1 = ρf˜2 .
Conversely let ρ = ρg̃1 = ρg̃2 . For every γ ∈ Γ we have that

g̃j ◦ γ = ρ ◦ g̃j , j = 1, 2.

For every t ∈ [0, 1] and every z ∈ H, letting hz be the geodesic connecting g̃1 (z)
and g̃2 (z), we denote by g̃(z, t) the point which divides hz in the ratio t : 1 − t.
g̃t (z) = g̃(z, t) is an homotopy between g̃1 and g̃2 . Hence we already proved that

g̃t ◦ γ = ρ(γ) ◦ g̃t , γ ∈ Γ, t ∈ [0, 1].

Therefore g̃t can be projected into a continuous mapping gt of R onto S2 and g1


is homotopic to g2 .

Lemma 5.2.2. Two points [S1 , f1 ], [S2 , f2 ] ∈ T (R) satisfy [S1 , f1 ] = [S2 , f2 ] if
and only if ρf˜1 = ρf˜2 , where the mappings f˜i are the canonical lifts of fi .

Proof. If [S1 , f1 ] = [S2 , f2 ], then f2 ◦ f1−1 is homotopic to a conformal mapping


ϕ : S1 → S2 . Then ϕ ◦ f1 : R → S2 is homotopic to f2 . From previous lemma it
follows that
ρ^ −1
= ρf˜2 .
ϕ◦f1

−1
The canonical lift of ϕ ◦ f is the composition of the canonical lift of ϕ which
1

is the identity, and the canonical lift f˜1−1 , hence ϕ^


◦ f1−1 = f˜1−1 . Therefore
ρf˜1 = ρf˜2 .
Conversely, suppose that ρf˜1 = ρf˜2 = ρ. Let Γ0 = ρ(Γ), then S1 and S2
are conformally equivalent to H/Γ0 , hence there exists a biholomorphic map
ϕ : S1 → S2 .
The maps ϕ ◦ f1 and f2 are quasiconformal mappings from R to S2 , and the
canonical lift of ϕ ◦ f1 is equal to f˜1 . Therefore they are homotopic. This implies
that (S1 , f1 ) ∼ (S2 , f2 ).

One property of quasiconformal mappings is the so called convergence property.


The proof of the following proposition can be found in [1].
5.2. Teichmüller spaces 75

Theorem 5.2.3 (Convergence Property). Suppose that fn is a sequence of


Kn − qc mappings of Ĉ such that
(a) every fn fixes 0, 1 and ∞;
(b) Kn < K < ∞;
Then there exist a subsequence of fn which is uniformly convergent on compacts
to a quasiconformal homeomorphism
Teichmüller metric Let x1 = [S1 , f1 ] and x2 = [S2 , f2 ] be two points in T (R),
we define a the function dT (R) : T (R) × T (R) → R as
dT (R) (x1 , x2 ) := inf log Kg ,
g∈Ff1 ,f2

where Ff1 ,f2 is the space of all quasiconformal mapping from S1 on S2 homotopic
to f2 ◦ f1−1 , and Kg is the smallest value for which g is K − qc. The function
dT (R) defines a distance on T (R) called the Teichmüller distance.
Proof. Firstly we prove that the definition doesn’t depend from the choice of the
representatives (Si , fi ). If (Si0 , fi0 ) ∼ (Si , fi ) then the maps fi ◦ (fi0 )−1 : Si0 → Si
are homotopic to two conformal maps ϕ1 and ϕ2 . For every map g 0 ∈ Ff10 ,f20 , the
map ϕ2 ◦ g 0 ◦ ϕ−1
1 ∈ Ff1 ,f2 . Moreover since ϕi are holomorphic, it follows that
Kg0 = Kϕ2 ◦g0 ◦ϕ−1 ,
1

this proves that the Teichmüller distance is well defined.


We must prove that the Teichmüller metric is a distance.
Simmetry follows from the fact that Kg = Kg−1 and triangle inequality follows
from the fact that Kg1 ◦g2 ≤ Kg1 Kg2 .
Finally suppose that d(x1 , x2 ) = 0, then there exist a sequence of map gn all
homotopic to f2 ◦ f1−1 such that gn is Kn − qc and Kn → 0. The lift of this
sequence of map is a collection of K̃n − qc map g̃n , and every element in the
sequence fix 0, 1 and ∞. Moreover K̃n → 0 too. We can apply the convergence
property, therefore there exist a subsequence g̃n0 → σ uniformly on compact
sets. We can project this map to a conformal map ϕ : S1 → S2 , this map will be
arbitrarily close to some gn for n big enough. Therefore it is homotopic to gn for
some n, which is homotopic to f2 ◦ f1−1 . This proves that [S1 , f1 ] = [S2 , f2 ].

5.2.1 Teichmüller space of a Fuchsian group


Every Riemann surface can be written as S/Γ where S is a simply connected Rie-
mann surface and Γ is a Fuchsian group. We now define what is the Teichmüller
space of a Fucshian group, which is denoted by T (Γ). We will prove that, if Γ
is the Deck transformation group of the universal covering map of a Riemann
surface R , T (Γ) and T (R) are isometric.
Definition A Beltrami differential is a function µ ∈ L∞ (H) such that

γ 0 (z)
µ(γz) = µ(z), ∀γ ∈ Γ.
γ 0 (z)
We denote the space of Beltrami differentials with B(Γ). B(Γ)1 is the open ball
of radius 1 in B(Γ).
76 Chapter 5. TEICHMÜLLER SPACES

The term Beltrami differential comes from the fact that the expression µ(z)dz̄/dz
is invariant with respect the action of Γ.
For every element µ ∈ B(Γ)1 , there exist a unique quasiconformal function
f µ such that
fz̄µ = µfzµ ,
and which satisfies the normalization condition f µ (0) = 0, f µ (1) = 1 and
f µ (∞) = ∞.
For every γ ∈ Γ the function f µ ◦ γ ◦ (f µ )−1 is an automorphism of the upper
half plane. As a matter of fact

∂z̄ (f µ ◦ γ ◦ (f µ )−1 ) =
= fzµ ◦ γ ◦ (f µ )−1 γ 0 ◦ (f µ )−1 [(f µ )−1 ]z̄ + fz̄µ ◦ γ ◦ (f µ )−1 γ 0 ◦ (f µ )−1 [(f µ )−1 ]z
= fzµ ◦ γ ◦ (f µ )−1 [(f µ )−1 ]z (−γ 0 ◦ (f µ )−1 µ ◦ (f µ )−1 + µ ◦ γ ◦ (f µ )−1 γ 0 ◦ (f µ )−1 )
= fzµ ◦ γ ◦ (f µ )−1 [(f µ )−1 ]z (γ 0 ◦ (f µ )−1 µ ◦ (f µ )−1 + µ ◦ (f µ )−1 γ 0 ◦ (f µ )−1 )
= 0,

then by Weyl’s lemma f µ ◦ γ ◦ (f µ )−1 ∈ Aut(H).


The group Γµ := f µ Γ(f µ )−1 is a Fuchsian group of the same type of Γ.
Each element of B(Γ)1 induces an injective homomorphism ρf µ : Γ → P SL(2, R).
Definition A representation of a group G on a vector space V over a field K is
a group homomorphism
ρ : G → GL(V ).
A projective representation is a group homomorphism

ρ : G → P GL(V ) := GL(V )/K ∗ .

A representation or a projective representation is faithful if it is an injective


group homomorphism.
Two representations (or projective representations) ρ and π on the vector spaces
V and W , are equivalent if there exist a group isomorphism α : V → W such
that
ρ(g) = α ◦ π(g) ◦ α−1 , ∀g ∈ G.
Thus we see that ρf µ is a projective representation over R2 .
Teichmüller space of a Fuchsian group The Teichmüller space T (Γ) of the
Fuchsian group Γ is the space of the representation ρf µ : Γ → P SL(2, R).
The function µ 7→ ρf µ defines a projection

Φ : B(Γ)1 → T (Γ).

From now on we assume that the Fuchsian group we are considering is of the
first kind, i.e. that the limit set of Γ is the real line. This happens for example
when we consider the deck transformation group of a compact Riemann surface
of genus g with n punctures.
Lemma 5.2.4. Let Γ be a Fuchsian group of the first type. Let µ, ν ∈ B(Γ)1
then either Φ(µ) = Φ(ν) or Φ(µ) is not equivalent to Φ(ν).
Moreover if Φ(µ) = Φ(ν) then f ν = f µ on the extended real line.
5.2. Teichmüller spaces 77

Proof. Let’s suppose that Φ(µ) is equivalent to Φ(ν), hence there exists α ∈
P SL(2, R) such that

f µ ◦ γ ◦ (f µ )−1 = α ◦ f ν ◦ γ ◦ (α ◦ f ν )−1 , ∀γ ∈ Γ.

Hence
(f ν )−1 ◦ α−1 ◦ f µ ◦ γ = γ ◦ (f ν )−1 ◦ α−1 ◦ f µ , ∀γ ∈ Γ.

The functions f µ , f ν and α send R̂ into itself and are continuous on it, hence
h = (f ν )−1 ◦ α−1 ◦ f µ has the same property. Given ζ ∈ R̂ since Γ is of the first
kind, from proposition (1.2.9) it follows that there exist a sequence γn contained
in Γ which converge uniformly on compact subset to the constant function ζ.
Hence, since h ◦ γn = γn ◦ h, taking the limit for n → ∞ we obtain that

h(ζ) = ζ.

Hence h|R̂ = id and


f µ (ζ) = α ◦ f ν (ζ) ∀ζ ∈ R̂.
The functions f ν and f µ fix 0, 1 and ∞, hence α fixes those points too. Therefore
α = idH . This implies that Φ(µ) = Φ(ν) and that f µ = f ν on the extended real
line, thus the proposition is proved.

Teichmüller metric of a Fuchsian group Let ρ1 , ρ2 ∈ T (Γ) we define the


function dT (Γ) : T (Γ) × T (Γ) → R as

dT (Γ) (ρ1 , ρ2 ) = inf ess sup dD (µ1 , µ2 ),


µi ∈Φ−1 (ρi ) H

where
|1 − z̄w| + |z − w|
dD (z, w) = log
|1 − w̄z| − |z − w|
is the Poincaré distance on the unit disk. This function defines a distance on
T (Γ).

If we endow T (Γ) with the metric topology coming from the the projection Φ is
continuous.

Theorem 5.2.5. Given a Riemann surface R, the Teichmüller space of the


surface T (R) and the Teichmüller space T (Γ) of the group Γ are isometric.

Proof. Given µ ∈ B(Γ)1 , the canonical solution of Beltrami equation f µ can be


projected into a function g µ : R → H/Γµ . From lemma (5.2.2) it follows that if
µ0 is another Beltrami differential such that Φ(µ) = Φ(µ0 ), then [g µ , H/Γµ ] =
0 0
[g µ , H/Γµ ]. Hence there exists a map F : T (Γ) → T (R) such that the map
G : µ 7→ [g µ , H/Γµ ] can be factorized as G = F ◦ Φ.
Lemma (5.2.2) implies that the map F is injective. We can prove that it is
surjective in the following way. First of all we consider an element [f, S] ∈ T (R),
if f˜ : H → H is the canonical lift of f , then

ρf˜ ◦ f˜ = f˜ ◦ γ, ∀ γ ∈ γ.
78 Chapter 5. TEICHMÜLLER SPACES

Then for almost every z ∈ H

(ρf˜(γ)0 ◦ f˜) · f˜z = (f˜z ◦ γ) · γ 0


(ρf˜(γ)0 ◦ f˜) · f˜z̄ = (f˜z̄ ◦ γ) · γ 0 ,

Thus we obtain that


µf˜ = (µf˜ ◦ γ)γ 0 /γ 0 .
Since f˜ is quasiconformal, it follows that µf˜ ∈ B(Γ)1 . The map F sends the
element Φ(µf˜) ∈ T (Γ) in [f, S], hence it is a bijection from T (Γ) onto T (R).
Using formula (5.2) we notice that, if f, g are quasiconformal mapping from
the upper half plane to itself, then
    
µf − µg µf − µg
log Kf ◦g−1 = ess sup log 1 + 1 −
H 1 − µ̄g µf 1 − µ̄g µf
= ess sup dD (µf , µg )
H

where dD is the Poincare metric on the disk.


If f : R → S is a quasiconformal mapping and f˜ is its canonical, then

f ◦ JR = JS ◦ f˜,

where JR and JS are the universal covering maps of R and S respectively.


Therefore one has that
J0
µf ◦ JR · R0 = µf˜,
JR
and then Kf = Kf˜.
Given ρ1 , ρ2 ∈ T (Γ) if one consider µ1 , µ2 ∈ B(Γ)1 such that ρi = Φ(µi ), then
if Si = H/Γµi and g µi : R → Si are the projections of f µi , we have

log Kgµ2 ◦(gµ1 )−1 = ess sup dD (µ1 , µ2 ).


H

0
If we take a different pair of µ0i ∈ π −1 (ρi ), we know that (Si0 , g µi ) ∼ (Si , g µi ).
0 0
Moreover there exist two conformal maps ϕ1 , ϕ2 such that ϕ1 ◦g µ2 ◦(g µ1 )−1 ◦ϕ−1 2 ∈
Fgµ1 ,gµ2 , moreover Kϕ ◦gµ02 ◦(gµ01 )−1 ◦ϕ−1 = Kgµ02 ◦(gµ01 )−1 , hence
1 2

inf log Kg ≤ inf ess sup dD (µ1 , µ2 ),


g∈Fgµ1 ,gµ2 µi ∈π −1 [ρi ] H

and so dT (R) (F (ρ1 ), F (ρ2 )) ≤ dT (Γ) (ρ1 , ρ2 ).


Let g ∈ Fgµ1 ,gµ2 and let f : H → H be its canonical lift. f and f µ2 ◦ (f µ1 )−1 are
homotopic, hence ρf = ρf µ2 ◦(f µ1 )−1 . The function f20 = f ◦ f µ1 projects into a
function g20 = g ◦ g1 : R → S2 and ρf20 = ρf µ2 . The Beltrami coefficient µ02 = µf20
is contained in Φ−1 (ρ2 ), therefore if we consider the couple µ1 , µ02

log Kg = ess sup dD (µ1 , µ02 ).


H

If we take the infimum in Fgµ2 ◦(gµ1 )−1 we obtain that dT (R) (F (ρ1 ), F (ρ2 )) ≥
dT (Γ) (ρ1 , ρ2 ). Therefore F : T (Γ) → T (R) is an isometry.
5.2. Teichmüller spaces 79

Let Aut∗ (Γ) be the group of proper automorphism of Γ which carry parabolic
elements into parabolic elements. Given an element Φ(µ) ∈ T (Γ) we obtain a
representation
α(Φ(µ)) = Φ(µ) ◦ α.
We will prove that α(Φ(µ)) ∼ Φ(µα ) for a Beltrami differential µα ∈ B(Γ)1 .
If µ0α ∈ B(Γ)1 is such that α(Φ(µ)) ∼ Φ(µ0α ), from lemma (5.2.4) it follows
that Φ(µα ) = Φ(µ0α ). Hence Aut∗ (Γ) defines an action on T (Γ) which sends
Φ(µ) → Φ(µα ). The proof of the following theorem is contained in [6].
Theorem 5.2.6. Let α : Γ → Γ be a proper automorphism which carry parabolic
elements into parabolic elements, then there exist a quasiconformal map fα :
H → H such that
fα ◦ γ = α(γ) ◦ fα , γ ∈ Γ.
Proof. OMITTED
Given the map fα , we have that

α(Φ(µ))(γ) = Φ(µ)(α(γ))
= (f µ ◦ fα ) ◦ γ ◦ (f µ ◦ fα )−1
= ρf µ ◦fα (γ),

We consider an element σ ∈ P SL(2, R) such that gµ,α := σ ◦ f µ ◦ fα fixes 0, 1


and ∞, hence µα = (gµ,α )z̄ /(gµ,α )z is a Beltrami differntial and Φ(µα ) = ρgµ,α is
equivalent to α(Φ(µ)).
Teichmüller modular space The Teichmüller modular space of the Fuchsian
group Γ is defined as the quotient space M od(Γ) := Aut∗ (Γ)/Inn(Γ) where
Inn(Γ) is the group of inner automorphism of Γ, i.e. the proper automorphism
of the form α(γ) = σ ◦ γ ◦ σ −1 , with σ ∈ P SL(2, R).

Remark Every element of Inn(Γ) sends parabolic elements of Γ into parabolic


elements, as a matter of fact T r(σ ◦ γ ◦ σ −1 ) = T r(γ). Using proposition (2.3.1),
one finds that Inn(Γ) ⊂ Aut∗ (Γ). Therefore the definition is well posed.
Moreover the elements of Inn(Γ) act as the identity over T (Γ).

Moduli space The moduli space of the Fuchsian group Γ is the quotient space
R(Γ) = T (Γ)/M od(Γ).

5.2.2 Simultaneous uniformization


It is very difficult to understand when µ, ν ∈ B(Γ)1 give the same Teichmüller
point in T (Γ). This problem can be simplified in the following way.
Given an element µ ∈ B(Γ)1 we consider the Beltrami coefficient
(
µ(z), z ∈ H
µ̃(z) =
0, z ∈ C \ H.

We denote the solution of Beltrami equation which fixes 0, 1 and ∞ as fµ . We


can define the map

ρfµ (γ) = fµ ◦ γ ◦ (fµ )−1 , γ ∈ Γ,


80 Chapter 5. TEICHMÜLLER SPACES

and prove that ρfµ : Γ → P SL(2, C).


Thus we obtain a subgroup Γµ := ρfµ (Γ) of P SL(2, C) which acts properly
discontinuously on both Hµ = fµ (H) and H∗µ = fµ (H∗ ).
fµ induces a quasiconformal mapping from R = H/Γ onto Rµ = Hµ /Γµ and
a biholomorphic mapping from R∗ = H∗ /Γ onto H∗µ /Γµ . The Riemann surfaces
Rµ and R∗ are uniformized simultaneously by a single discrete group Γµ . This
uniformization is called Bers’ simultaneous uniformization. More information
on it can be found in [3].
Lemma 5.2.7. Let µ, ν ∈ B(Γ)1 , the following conditions are equivalent:
(i) f µ = f ν on R;
(ii) fµ = fν on H∗ .
Proof. If f µ = f ν on R, the map
(
(f µ )−1 ◦ f ν (z), z ∈ H,
f :=
z, z ∈ Ĉ \ H∗ ,

is quasiconformal. The map g = fµ ◦ f ◦ (fν )−1 is conformal, hence it is a Möbius


transformation. Since g fixes 0, 1 and ∞ it must be the identity. Therefore
fµ = fν .
Conversely if fµ = fν on H∗ , the equality holds true on the whole Ĉ \ H.
Thus we obtain a quasiconformal mapping h = f µ ◦ (fµ )−1 ◦ fν ◦ (f ν )−1 : H → H.
By the same argument as above, the function h must be the identity, therefore
f µ = f ν on R
Proposition 5.2.8. Let µ, ν ∈ B(Γ)1 , Φ(µ) = Φ(ν) if and only if f µ = f ν on
R.
Proof. If f µ = f ν on the real axis then for every γ ∈ Γ we have that Φ(µ)(γ) =
Φ(ν)(γ) on the real line and hence they are equal on the whole Ĉ. Therefore
Φ(µ) = Φ(ν).
The converse was already proved in lemma (5.2.4).
Corollary 5.2.9. let µ, ν ∈ B(Γ)1 , Φ(µ) = Φ(ν) if and only if fµ = fν on H∗ .

5.2.3 Bers embedding


Given a Fuchsian group Γ we defined the Teichmüller space T (Γ). It is possible
to put a complex structure on it. This structure is the one in which the map
Φ : B(Γ)1 → T (Γ) is analytic.
Firstly we embed T (Γ) in the space of the Holomorphic Quadratic Differentials
Holomorphic quadratic differentials Given a Fuchsian group Γ, the space
of cusp form of weight 4 for the group Γ is the space of all holomorphic function
on H which are regular and transform according to the rule

q(γz)γ 0 (z)2 = q(z), ∀γ ∈ Γ.

It is a normed space with norm kqk = supz∈H (=z)2 |q(z)|. This space is denoted
as Q(Γ) and its elements are called holomorphic quadratic differentials.
5.2. Teichmüller spaces 81

We can also define the space of Holomorphic quadratic differentials on the


lower half plane H∗ . This space is denoted as Q∗ (Γ).

Given an element µ ∈ B(Γ)1 we consider fµ . For every element γ ∈ Γ, γµ =


fµ ◦γ ◦fµ−1 is a Möbius transformation. Therefore if one consider the Schwartizian
derivative of fµ ◦ γ = γµ ◦ fµ it obtains that

S (fµ )(z) = S (fµ ) ◦ γ(z)γ 0 (z)2 .

The function ψµ = S (fµ )|H∗ is an element in Q∗ (Γ). Hence

ϕµ (z) = ψµ (z̄)

is an element in Q(Γ).
We notice that if ν and µ project into the same Teichmüller point, since
fµ = fν on H∗ , then the elements ϕν and ϕµ coincide in Q(Γ). The map

Φβ : B(Γ)1 → Q(Γ),

is a continuous map called Bers projection, it induces a map

ιβ : T (Γ) → Q(Γ).

which is a continuous injection called the Bers embedding.


If we identify T (Γ) with ∆(Γ) = ιβ (T (Γ)) ⊂ Q(Γ), we have that Φ = Φβ .
There are some remarkable properties of Bers embedding. Now we make a
list of them, without any proof. The interested reader can find these proofs on
the book of Ahlfors [1].

Theorem 5.2.10. The image of Bers embedding ∆(Γ) = β(B(Γ)1 ) is contained


in the closed ball of radius 3/2 of Q(Γ).
Moreover every holomorphic quadratic differential φ with kφk ≤ 1/2 is contained
in ∆(Γ).

Theorem 5.2.11. ∆(Γ) is an open subset of Q(Γ).

Using Riemann-Roch Theorem we can calculate the complex dimension of


the vector space Q(Γ), in the case Γ is the Deck transformation group of the
Universal covering of a Riemann surface of type (g, n), i.e. a Riemann surface of
genus g with n punctures. The proof of this can be found in [12].

Proposition 5.2.12 (Riemann-Roch Theorem). Let R be an hyperbolic Riemann


surface of genus g with n punctures. The space Q(Γ) is a complex vector space
of dimension 3g + n − 3.

It is interesting to notice that all the hyperbolic Riemann surfaces satisfy


3g + n − 3 ≥ 0, dim(Q(Γ)) = 0 only in the case of the 3 punctured sphere.
In conclusion given an hyperbolic Riemann surface R of genus g and with n
punctures, its Teichmüller space T (R) is equal to T (Γ) which is identified with
an open subset ∆(Γ) ⊂ Q(Γ) ∼ = C 3g+n−3 . Bers’ embedding allows us to put a
complex structure on T (R). With this complex structure Bers’ embedding and
Bers’ projection are analytic mapping.
82 Chapter 5. TEICHMÜLLER SPACES

5.2.4 The infinitesimal approach


We would like to study now the derivative of Bers’ projection Φ in the direction
ν ∈ B(Γ) at µ ∈ B(Γ)1 .Those derivatives are denoted by Φ̇µ [ν].
Given µ and ν as above, if t is small enough µ + tν =: µt ∈ B(Γ)1 . Thus we
define
1
Φ̇µ [ν] = lim (Φ(µt ) − Φ(µ)),
t→0 t

here the convergence is intended in the norm of Q(Γ).


We study the case in which µ = 0, hence µt = tν belongs to B(Γ)1 if |t| < 1/kνk∞ .

Theorem 5.2.13. For every ν ∈ B(Γ)1 the derivative Φ̇0 [ν] exists and it is
given by
Z
6 µ(ζ)
Φ̇0 [ν] = − dxdy, µ ∈ B(Γ)
π H (ζ̄ − z)4

Proof. If |t| ≤ 1/kνk∞ we consider the function ftν , which is the canonical
solution of Beltrami equation on C with Beltrami coefficient
(
tν z ∈ H
tν̃(z) = .
0 z ∈ Ĉ \ H

Using formula (5.3), we obtain that

ftν (z) = z + tf˙[ν̃] + o(t),

where  
z−1
Z
1 1 z
f˙[ν̃](z) = − ν(ζ) − + dxdy.
π H ζ −z ζ −1 ζ

The function gt = (ftν − z − tf˙[ν̃])/t is holomorphic on H∗ and converges


uniformly to 0 on compact subset of H∗ . Hence its derivatives gt(n) → 0 uniformly
on compact subsets of H∗ .
0
ftν = 1 + tf˙[ν̃]0 + o(t),
f 00 = tf˙[ν̃]00 + o(t),

000
ftν = tf˙[ν̃]000 + o(t).

All the residues divided by t converge to 0 uniformly on compact subset of H∗


as t → 0. Thus we obtain that
1
S (ftν ) = f˙[ν̃]000 + o(1).
t
In the end
1
Φ(tν)(z) = f˙[ν̃]000 (z̄) + o(1),
t
uniformly on compact set of H. The function f˙[ν̃](z) is holomorphic in H∗ ,
therefore if z ∈ H∗ and we take a close path around z, using Cauchy integral
5.2. Teichmüller spaces 83

formula we obtain that


ξ(ξ − 1)
Z Z
˙ (n) n! 1 1
f [ν̃] (z) = − ν(ζ) dxdydξ
2πi π γ H ζ(ζ − 1)(ζ − ξ) (ξ − z)n+1
ξ(ξ − 1)
Z Z
1 n! 1
=− ν(ζ) dξdxdy
π H 2πi γ ζ(ζ − 1)(ζ − ξ) (ξ − z)n+1
 
z(z − 1)
Z
1
=− ν(ζ)∂zn dxdy.
π H ζ(ζ − 1)(ζ − z)

Hence Z
6 µ(ζ)
f˙[ν̃]000 (z̄) = − dxdy.
π H (ζ̄ − z)4
From lemma (5.2.10) it follows that |Φ(tν)(z)y 2 | ≤ 3/2 for |t| ≤ kνk∞ , moreover
if z is fixed, since the complex derivative of fz (t) = Φ(tν)(z)y 2 with respect to t
is obtained from formula (5.4), it follows that fz (t) is an holomorphic function
with respect to t. Thus if δ = kνk∞ and t < δ using Cauchy’s differentiation
formula we obtain that, for ε small enough
Z
00
1 fz (τ )
|fz (t)| =

πi |t|=kνk∞ −ε (τ − t) 3
Z
1 3

π |t|=kνk∞ −ε 2(kνk∞ /2 − ε)3
6π(kνk∞ − ε)
≤ .
2π(kνk∞ /2 − ε)3

If ε → 0 we have that |fz00 (t)| ≤ Cν for t ≤ δ, where Cν is a constant which


depends only by ν.
We obtain that

|fz (t) − tfz0 (0)| ≤ t2 ,
2
µ(ζ)
R
We already know that fz0 (0) = − π6 H (ζ̄−z) 2
4 dxdy · y . In conclusion

Z
Φ(tν) 6 µ(ζ) Cν
+ dxdy ≤t ,

t π H (ζ̄ − z)4 2


Q(Γ)

hence if we send t → 0 we obtain the desired formula for Φ̇0 [ν]


This theorem implies that the tangent space at the point Φ(0) ∈ T (Γ) is
isomorphic to
T0 (T (Γ)) ∼
= B(Γ)/N (Γ),
where N (Γ) ⊂ B(Γ) is the kernel of Φ̇0 .
Let’s consider the map
Z
12 µ(ζ)
Λ(µ)(z) = dxdy, µ ∈ B(Γ),
π H (ζ̄ − z)4

and the map


Λ∗ (q)(z) = (=z)2 q(z), q ∈ Q(Γ).
84 Chapter 5. TEICHMÜLLER SPACES

Lemma 5.2.14. Let µ ∈ B(Γ) then Λ(µ) ∈ Q(Γ). Conversely if q ∈ Q(Γ) then
Λ∗ (q) ∈ B(Γ).
Proof. If µ ∈ B(Γ) then
Λ(µ) = −2Φ̇[µ],
the function on the right hand side belongs to Q(Γ) and so does Λ(µ).
The proof of the second part of the lemma follows from the fact that if γ ∈ Γ
then
1
=(γz)2 = − (γz − γ z̄)2 = (=z)2 γ 0 (z)γ 0 (z),
4
thus
1 γ 0 (z) ∗
Λ∗ (q)(γz) = (=z)2 γ 0 (z)γ 0 (z) q(z) = Λ (q)(z).
γ 0 (z)2 γ 0 (z)

Lemma 5.2.15. Suppose φ is holomorphic in H and


sup |φ(z)|(=z)2 < ∞,
z∈H

then
φ(ζ)y 2
Z
12
φ(z) = dxdy.
π H (ζ̄ − z)4
Proof. Firstly we notice that
y2 1 (ζ̄ − ζ)2
=−
(ζ̄ − z)4 4 (ζ̄ − z)4
2(ζ − z) (ζ − z)2
 
1 1
=− − +
4 (ζ̄ − z)2 (ζ̄ − z)3 (ζ̄ − z)4
1 (ζ − z)2
 
1 ∂ 1 ζ −z
=− − + − .
4 ∂ z̄ ζ̄ − z (ζ̄ − z)2 3 (ζ̄ − z)3

Assume first that φ is analytic on H, i.e it is analytic in some neighborhood of


H. Integration by parts gives that
φ(ζ)y 2
Z Z
12 1 φ(ζ)
dxdy = dζ = φ(z).
π H (ζ̄ − z) 4 2πi R ζ − z

If φ is not analytic on H, we consider the function φ(z + iε), for ε > 0, then
φ(ζ + iε)y 2
Z
12
φ(z + iε) = dxdy.
π H (ζ̄ − z)4
Now we take the limit keeping z fixed. The RHS converges to φ(z). In the LHS
we notice that
φ(ζ + iε)y 2 c
(ζ̄ − z)4 ≤ |ζ̄ − z|4 ,

which is an integrable function on H. Using dominate convergence theorem we


conclude the proof.
Corollary 5.2.16. Let Λ : B(Γ) → Q(Γ) and Λ∗ : Q(Γ) → B(Γ) be defined as
above, then we have that
Λ ◦ Λ∗ = idQ(Γ)
5.2. Teichmüller spaces 85

Tangent and Cotangent space of T (Γ)


We already saw that T0 (T (Γ) = B(Γ)/N (Γ). Since ΛΛ∗ = idQ(Γ) the function Λ
is surjective. Then the following sequence is exact
Λ
0 → N (Γ) ,→ B(Γ) −
→ Q(Γ) → 0.

This enable us to realize B(Γ)/N (Γ) as the subspace H (Γ) = Λ∗ (Q(Γ)) of B(Γ).
The space H (Γ) is the space of the so-called harmonic Beltrami differentials.
Proposition 5.2.17. Given µ ∈ B(Γ)1 the tangent space at T (Γ) at the point
Φ(µ) is isomorphic to
TΦ(µ) (T (Γ)) ∼
= H (Γµ ),
where Γµ = f µ Γ(f µ )−1 .
Proof. We already proved the proposition at the base point Φ(0).
Given another µ ∈ B(Γ)1 we consider the map

F : T (Γ) → T (Γµ )

given by
F (ρ)(γ 0 ) = ρ((f µ )−1 ◦ γ 0 ◦ f µ ), ∀γ 0 ∈ Γµ .
This map is well defined, Moreover if ρ = ρµ = Φ(µ), it follows that

F (Φ(µ)) = idΓµ = Φµ (0),

Where Φµ : B(Γµ ) → Q(Γµ ) is the usual Bers projection.


The map F is biholomorphic, it follows that

TΦ(µ) (T (Γ)) ∼
= TF (Φ(µ)) (T (Γµ )) = H (Γµ ).

The cotangent space at the point Φ(µ) can be described using the pairing
Z
(µ, q) = µ(z)q(z)dxdy, µ ∈ B(Γ), q ∈ Q(Γ).
H/Γ

This pairing is well defined on the Riemann surface H/Γ, and the following
lemma holds true
Lemma 5.2.18. The subspace where the pairing (µ, q) is degenerate is exactly
the subspace N (Γ) where Φ̇0 vanishes
The pairing (µ, q) induces a non degenerate pairing between B(Γ)/N (Γ) ∼
=
TΦ(0) (T (Γ)) and Q(Γ). Therefore

TΦ(0) (T (Γ)) ∼
= Q(Γ).

In the end one obtains the following proposition


Proposition 5.2.19. Given µ ∈ B(Γ)1 the cotangent space at T (Γ) at the point
Φ(µ) is isomorphic to

TΦ(µ) (T (Γ)) ∼
= Q(Γ).
86 Chapter 5. TEICHMÜLLER SPACES

5.3 A generating function for the accessory pa-


rameters
Let’s consider an hyperbolic Riemann surface R = H/Γ. For each µ ∈ B(Γ)1 we
consider the canonical solution f µ of Beltrami equation. We project the map
fµ : H → H to the map
F µ : R → Rµ = H/Γµ ,
in such a way that the following diagram commutes


H H
J Jµ


R Rµ

The function F µ satisfies Beltrami equation

Fz̄µ = M Fzµ ,

where M = µ ◦ J −1 (J −1 )0 /(J −1 )0 is well defined on X. Then F µ is a quasicon-


formal homeomorphism.
It is easy to prove that Rµ is a Riemann surface. Moreover the following
proposition holds true
Proposition 5.3.1. If R is an hyperbolic (compact punctured) Riemann surface
of genus g and with n punctured, and f : R → S is quasiconformal, then S is
a Riemann surface of the same type of R, i.e. has genus g and n punctured.
Moreover the map f can be extended to an homeomorphism between R̂ onto Ŝ
This proposition easily follows from the theorem below. The interested reader
can find the proof of this theorem in the book [19].
Theorem 5.3.2 (Removable singularities of a qc mapping). Suppose that d :
D → D0 and that b is an isolated point of ∂D. Then f has a limit b0 at b, and
b0 is an isolated point of ∂D0 . Defining f ∗ (b) = b0 and f ∗ |D = f we obtain a qc
mapping f ∗ : D ∪ {b} → D0 ∪ {b0 }.
Let’s consider a n-punctured sphere X = C\{z1 , . . . , zn−3 , 0, 1}. The Riemann
surface X µ is again a n-punctured sphere. It is possible to choose J µ in such
a way that X µ = C \ {z1µ , . . . , zn−3
µ
, 0, 1} and that the extension F̂ µ of F µ fixes
0, 1 and ∞.
Lemma 5.3.3. The function µ 7→ (z1µ , . . . , zn−3
µ
) from B(Γ)1 to

Wn = {(z1 , . . . , zn−3 ) ∈ C n−3 | zi 6= zj , zi 6= 0, 1}

induces a function
Ψ : T (Γ) → Wn .
Proof. If ν is Teichmüller equivalent to µ then there exist a conformal map
ϕ : X µ → X ν such that ϕ is homotopic to F ν ◦ (F µ )−1 . Let Ft be a function
which realize the homotopy between the two maps.
Both ϕ and F ν ◦ (f µ )−1 can be extended to functions defined from Ĉ into itself.
5.3. A generating function for the accessory parameters 87

Let’s take a loop η around the point ziµ in X µ , if ziν = F ν ◦ (F µ )−1 (ziµ ), then Ft
maps the loop η into a loop ηt around ziν for every t ∈]0, 1[. Therefore ϕ(ziµ ) = ziν .
This means that ϕ is a biholomorphic map which fixes 0, 1 and ∞, hence it is
the identity and (z1µ , . . . , zn−3
µ
) = (z1ν , . . . , zn−3
ν
).
Suppose that µ = εν then the function f εν is the canonical solution of
Beltrami equation with beltrami coefficient

εν(z) z ∈ H,

εν̃(z) = 0 z ∈ R,

εν(z̄) z ∈ H∗ ,

hence Z Z
1 1
f˙[ν](z) = − ν(ζ)R(ζ, z)dxdy − ν(ζ)R(ζ̄, z)dxdy,
π H π H

where
1 z−1 z
R(ζ, z) = + − .
ζ −z ζ ζ −1
In turn, since the projection of f εν , which is denoted by F εν solves Betrami
equation on C with complex dilatation εN = εν ◦ J −1 (J −1 )0 /(J −1 )0 , we have
that Z
1
Ḟ [N ](z) = − N (ζ)R(ζ, z)dxdy
π C

5.3.1 Complex analytical covering of the space of punc-


tured spheres
We want to prove that the function Ψ : T (Γ) → Wn is a complex analytical
covering.
Let D2 (X) be the space of rational functions on Ĉ of order O(|z|−3 ) as z → ∞
and with at worst simple poles at the other punctures of X. The mapping

q 7→ Q = q ◦ J −1 (J −1 )02 ,

is a well defined linear isomorphism of the spaces Q(Γ) and D2 (Γ). It’s inverse
is given by
Q 7→ q = Q ◦ JJ 02 .
On Q(Γ) we define the Petersson inner product as
Z
hq1 , q2 iQ = q1 (z)q2 (z)(= z)2 dxdy.
H/Γ

Setting hP, QiD2 = hP ◦ JJ 02 , Q ◦ JJ 02 iQ we obtain that


Z
hP, QiD2 = P ◦ J(z)Q ◦ J(z)|J 0 (z)|4 (= z)2 dxdy
X
=(J −1 (z))2
Z
= P (z)Q(z) −1 0 dxdy
|(J ) (z)|2
ZX
= P (z)Q(z)e−2ϕX (z,z̄) dxdy,
X
88 Chapter 5. TEICHMÜLLER SPACES

where ϕX is the canonical solution of Liouville equation we defined in (2.2).


Let  
1 1 zi − 1 zi 1
Pi (z) = − + − = − R(z, zi ),
π z − zi z z−1 π
those functions are linearly independent and generate D2 (X). We denote by
Qi the basis of D2 (X) bihortogonal to Pi , i.e a basis of D2 (X) such that
hPi , Qj iD2 = δij . We take a basis of H (Γ) of the form

νi (z) = (= z)2 qi (z),

where qi = Qi ◦ JJ 02 .
Similarly we define the basis νiµ of the space H (Γµ ). Then νiµ can be seen as a
vector field on T (Γ).
Lemma 5.3.4. The mapping Ψ : T (Γ) → Wn is a complex analytic covering
and

dΨΦ(µ) (dΦµ νiµ ) = .
∂zi
Proof. It suffices to prove the theorem at the point Φ(0). Let’s take ν ∈
H (Γ) ∩ B(Γ)1 , then
n−3
X
ν= εi νi .
i=1

In the coordinates (ε1 , . . . , εn−3 ) the map Ψ ◦ Φ is given by

(ε1 , . . . , εn−3 ) 7→ (F ν(ε1 ,...,εn−3 ) (z1 ), . . . , F ν(ε1 ,...,εn−3 ) (zn−3 ).

Then
∂(Ψ ◦ Φ)j
= Ḟ [Ni ](zj ).
∂εi 0

From the definition of νi we have that

(J −1 )0
Ni = νi ◦ J −1
(J −1 )0
(J −1 )0
= (=J −1 )2 qi ◦ J −1
(J −1 )0
= e−2ϕX Qi .

Therefore, from the particular choice of Pi we made, we obtain that

Ḟ [Ni ](wj ) = hPj , Qi i = δi,j .

This proves that Ψ is a local diffeomorphism at the point Φ(0). As a matter of


fact

dΨΦ(0) (dΦ0 (νi )) =
∂zi
and since dΦ0 (νi ) form a basis for TΦ(0) (T (Γ)) the differential is invertible.
Given α ∈ Aut∗ (Γ) from theorem (5.2.6) we know that there exist a quasicon-
formal map fα : H → H such that fα ◦ γ = α(γ) ◦ fα for all γ ∈ Γ. Hence it
induces a quasiconformal Fα : X → X. The map Fα can be extended as a self
5.3. A generating function for the accessory parameters 89

quasiconformal map on the complex sphere that send the puncture of X into
the punctures of X.
Therefore every element α ∈ Aut∗ (Γ) induces an element sα ∈ Symm(n). This
map induces an epimorphism of the group M od(Γ) onto the symmetric group
Symm(n) the kernel of this epimorphism is denoted by G(Γ). The map Ψ
is invariant under G(Γ) and Wn ∼ = T (Γ)/G(Γ). Hence Ψ is a covering with
automorphism group G(Γ).

5.3.2 Zograf–Takhtadzhyan theorem


Let’s consider the solution of Liouville equation on the punctured sphere C \
{z1 , . . . , zn−3 , 0, 1}. We consider it as a function ϕ(z; z1 , . . . , zn−3 ) defined on
Wn+1 .
The basis ν1 , . . . , νn−3 of H (Γ) is uniquely determined by the element
(z1 , . . . , zn−3 ) ∈ Wn . Thus the functions Ḟ [Ni ] are well defined on Wn+1 . Let
Ḟ i (z; z1 , . . . , zn−3 ) := Ḟ [Ni ](z), i = 1, . . . , n − 3.
The following lemma was contained in an article of Ahlfors [2].
Lemma 5.3.5. For any ν ∈ H (Γ) we have that
∂ |fzεν |2
= 0,
∂ε ε=0 (=f εν )2

here the derivative is taken for ε ∈ R.


Proof. Firstly we notice that
f εν
f˙z̄ [ν] = lim z̄
ε→0 ε
ενfzεν
= lim
ε→0 ε
= ν,
here we were allowed to exchange the limit in ε → 0 and the derivatives since
the convergence is uniform on compact subset as ε → 0.
If ν ∈ H (Γ) then ν = Λ∗ (q) for a certain q ∈ Q(Γ), thus
ν = Λ∗ Λ(ν).
We know that
Z Z
1 1
f˙[ν](z) = − ν(ζ)R(ζ, z)dxdy − ν(ζ)R(ζ̄, z)dxdy,
π H π H

hence, since f˙z̄ [ν] = ν, the function


Z
2
Θ[ν] = f˙[ν] + if˙[iν] = − ν(ζ)R(ζ̄, z)dxdy,
π H

is holomorphic in H. Since Θ[ν] is holomorphic we can prove, as we did in


theorem (5.2.13), that its third derivative is given by
Z
000 12 ν(ζ)
Θ [ν] = dxdy = −Λ(ν)
π H (ζ̄ − z)4
90 Chapter 5. TEICHMÜLLER SPACES

Therefore, since Λ∗ (q) = (=z)2 q(z),

ν = −Θ̄000 [ν](z)(= z)2

After an easy calculation, we obtain


 
1
∂z̄ −Θ̄ [ν](=z) + Θ̄ [ν]i(=z) + Θ̄[ν] = −Θ̄000 [ν](z)(= z)2 .
00 2 0
2

Using again the fact that f˙[ν]z̄ = ν, we arrive to the following expression
1
f˙[ν] = −Θ̄00 [ν](=z)2 + Θ̄0 [ν]i(=z) + Θ̄[ν] + F (z)
2
where F (z) is an holomorphic function.
All the functions f εν attain real values on R. Therefore =(f˙[ν]) = 0 as
=z = 0.
One can easily prove that |Θ00 [ν](=z)| and |Θ0 [ν]| are bounded in H, hence
Θ00 [ν](=z)2 = 0 and Θ0 [ν](=z) = 0 as =z = 0. Thus if z ∈ R̂,
   
˙ 1 1
0 = =(f [ν])(z) = = Θ̄[ν](z) + F (z) = = − Θ[ν](z) + F (z) .
2 2

The function H(z) = F (z) − 12 Θ[ν](z) is an holomorphic function on H which is


continuous on the real line, hence it can be extended to an holomorphic function
on C. From the definition of f˙[ν] and the fact that f εν fixes 0, 1 and ∞, it
follows that f˙[ν] vanishes at 0, 1, ∞ and so does Θ[ν]. Therefore the function
H(z) vanishes at those three point. The extension of H(z) is an entire function
on C and since it vanishes at ∞ it must be bounded, hence it is constantly 0.
Finally we obtained the representation
1 1
f˙[ν] = −Θ̄00 [ν](=z)2 + Θ̄0 [ν]i(=z) + Θ̄[ν] + Θ[ν].
2 2
Let’s consider the derivative with respect to z,
1 i
∂z (f˙[ν](=z)−2 ) = (Θ0 [ν] − Θ̄0 [ν]) + (Θ[ν] + Θ̄[ν])(=z)3 ,
2 2
hence
< ∂z (f˙[ν](=z)−2 ) = 0.
In the end we get that

∂ |fzεν |2 f˙[ν]z + f˙[ν]z i


= + (f˙[ν] − f˙[ν])
∂ε ε=0 (=f εν )2 (=z)2 (=z)3

!
f˙[ν]z i
=< + (f˙[ν])
(=z)2 (=z)3
= < ∂z (f˙[ν](=z)−2 )
= 0.
5.3. A generating function for the accessory parameters 91

Lemma 5.3.6. The function ϕ is continuously differentiable on Wn+1 and

2ϕzi + 2ϕz Ḟ i + Ḟzi = 0.

Proof. Let Γ be the Fucshian group which uniformize the surface X = C \


{z1 , . . . , zn−3 , 0, 1}. Let ν = ε1 ν1 + . . . εn−3 νn−3 ∈ H (Γ) Then f ν is a smooth
function in z ∈ H and if the εi are small enough it is smooth also in the variables
εi .
The projection F ν (z) : X → X ν , of f ν , is differentiable with respect to z ∈ X
and it depends analytically on the coordinates εi .
Since J ν ◦ f ν = F ν ◦ J, it follows that J ν and (J ν )0 are continuously differ-
entiable in the variables εi and that we can choose branches of (J ν )−1 and of
[(J ν )−1 ]0 that are continuously differentiable outside the singular point of X.
Using Lemma (5.3.4) we write that

ϕ(z; z1 , . . . , zn−3 ) = d(Ψ ◦ Φ)0 (νi )(ϕ))
∂zi
= νi (ϕ(z; Ψ ◦ Φ(µ))
ϕ(z, Ψ ◦ Φ(ενi )) − ϕ(z, Ψ ◦ Φ(0))
= lim
ε→0 ε

= ϕ(z, Ψ ◦ Φ(ενi ))
∂ε 0
ϕ(z, Ψ ◦ Φ(ενi )) is the canonical Liouville equation of the surface X ενi , hence it
can be written as
|[(J ενi )−1 ]0 (z)|
ϕ(z, Ψ ◦ Φ(ενi )) = log ,
=[(J ενi )−1 (z)]
This expression is differentiable in ε and so ϕ is differentiable with respect to wi .
Using the expression of the canonical solution of Liouville equation (2.2) and
the fact that J εν ◦ f εν = F εν ◦ J we obtain that
|fzεν |2 εν εν

εν 2
= e2ϕ(J ◦f ,Ψ◦Φ(εν)) |(J εν ◦ f εν )z |2
(=f )
εν
◦J,Ψ◦Φ(εν))
= e2ϕ(F |Fzεν ◦ J|2 |J 0 |2

Now using this expression together with the previous lemma and the fact that
the function F εν it’s holomorphic in ε, we obtain that

2 ϕ(z, Ψ ◦ Φ(εν) + 2ϕz Ḟ [N ] + Ḟ [N ]z = 0
∂ε ε=0
If we put ν = νi we obtain the desired result.
Let S[ψ] be the action defined as
Z   
1 πn
S[ψ] = lim |ψz |2 + e2ψ dxdy + (log r + 2 log | log r|)
r→0
Xr 4 2
The function
S(z1 , . . . , zn ) := S[ϕ(z; z1 , . . . , zn )]
is defined on Wn .
92 Chapter 5. TEICHMÜLLER SPACES

Theorem 5.3.7 (Zograf–Takhtadzhyan). The function S is continuously differ-


entiable on Wn and
2 ∂S
ci = − , i = 1, . . . , n − 3.
π ∂zi
Proof. Let
Z
πn
Sr (z1 , . . . , zn−3 ) = |ψz |2 dxdy + (log r + 2 log | log r|).
Xr 2

For any r > 0 the function Sr is continuously differentiable on Wn . The Riemann


metric on X is given by e2ϕ dzdz̄, and using Gauss–Bonnet formula
Z
e2ϕ dxdy = −2πχ(X) = 2π(n − 2),
X

we find that if r → 0 then Sr → S − π(n−2) 2


on Wn pointwise.
π
We first show that ∂Sr /∂zi → − 2 ci as r → 0 in the sense of pointwise
convergence
Z
∂Sr ∂
= |ϕz |2 dxdy
∂zi ∂zi Xr
Z Z
i i
= ∂zi |ϕz |2 dz ∧ dz̄ + |ϕz |2 dz̄,
2 Xr 2 Cri

where the circle Cri = {z| |z −wi | = r} is oriented as a component of the boundary
∂Xr . It follows from the previous lemma that

∂zi |ϕz |2 = ϕzzi ϕz̄ + ϕz̄zi ϕz


 
1 i 1
=− ϕz̄ Ḟzz + ϕz Ḟziz̄ + ϕ2z Ḟz̄i − ∂z (|ϕz |2 Ḟ i )
2 2
1 1
= (ϕzz − ϕ2z )Ḟz̄i − ∂z (ϕz Ḟz̄i ) + ∂z̄ (ϕz Ḟzi ) − ∂z (ϕz̄ Ḟzi ) − ∂z (|ϕz |2 Ḟ i ).
2 2
By Stokes’ formula,
Z Z Z Z
2 2 i 1 i
∂zi |ϕz | dz ∧ dz̄ = (ϕzz − ϕ )Ḟ dz ∧ dz̄ −
z z̄ ϕz Ḟ dz̄ − z̄ ϕz Ḟzi dz
Xr Xr ∂Xr 2 ∂Xr
Z Z
1 i 2 i
− ϕz̄ Ḟz dz̄ − |ϕz | Ḟ dz̄
2 ∂Xr ∂Xr

= I1 + I2 + I3 + I4 + I5 .

We compute each integral separately. The following formulae hold true

Ḟ i (z) =δi,j + (z − zj )Ḟzi (zj ) + o(|z − zj |), j 6= n, z → wj ,


i i
Ḟ (z) = z Ḟ (∞) + o(|z|),
z z → ∞,
i −2ϕ
Ḟ = e
z̄ Q̄i ,

they follow from the definition of Ḟ i , the integral representation of Ḟ i and the
definition of the Qi .
5.3. A generating function for the accessory parameters 93

Let’s begin with the integral I1 , the function ω = ϕzz − ϕ2z is holomorphic,
therefore by Stokes’ theorem
Z
I1 = = (ϕzz − ϕ2z )Ḟz̄i dz ∧ dz̄
Xr
Z
=− ω Ḟ i dz
∂Xr
n−1 Z  
X 1 cj
=− + + . . . (δij + (z − zj )Ḟzi (zj ) + . . . )dz
j=1 Cr
j 4(z − zj )2 2(z − zj )
Z  
1 cn
− 2
+ 3 + . . . (z Ḟzi (∞) + . . . )dz
n
Cr 4z 2z
n−1
πi X πi
= πici + Ḟz (zj ) − Ḟzi (∞) + o(1), r → 0.
2 j=1 2

Here the circle Crn = {z| |z| = 1/r} is oriented as a component of the boundary
∂Xr . The integrals I2 , I3 and I4 all vanishes as r → 0,
Z Z
1
I2 = − ϕz Ḟz̄i dz̄ = ∂z (e−2ϕ(z) )Qi (z)dz̄ = o(1) r→0
∂Xr 2 ∂Xr
Z
I3 + I4 = − Ḟzi (ϕz dz + ϕz̄ dz̄) = o(1) r → 0.
∂Xr

We know that near the punctures ϕ ∼ − log |z − zi | − log | log |z − zi ||, hence the
partial derivative with respect to z behaves like
(
1
+ . . . , z → zi
ϕz = 2(z−z
1
i)

2z
+ . . ., z → ∞.

For more detail, we refer to the article [18]. Hence the last integral I5 is estimated
by
Z
I5 = − |ϕz |2 Ḟ i dz̄
∂Xr
Z n−1 Z  
2
X 1
=− |ϕz | dz̄ − + ... ((z − wi )Ḟzi (zj ) + . . . )dz̄
Cri j=1
j
Cr 4|z − zj |2
Z  
1
− + ... (z Ḟzi (∞) + . . . )dz̄
Crn 4|z|2
Z n−1
πi X i πi
=− |ϕz |2 dz̄ − Ḟz (zj ) + Ḟzi (∞) + o(1), r → 0.
Cri 2 j=1 2

Finally we obtain that


Z Z
2
∂zi |ϕz | dz ∧ dz̄ = πici − |ϕz |2 dz̄ + o(1), r → 0.
Xr Cri

Hence
∂Sr 2
= − ci + o(1), r → 0. (5.5)
∂zi π
94 Chapter 5. TEICHMÜLLER SPACES

It remains to show that the reminder of this formula can be estimated uniformly in
a neighborhood of an arbitrary point (z1 , . . . , zn−3 ) ∈ Wn . Let Γ be a group which
uniformizes the surface X = C\{z1 , . . . , zn−3 , 0, 1}. If ν = ε1 ν1 +· · ·+εn−3 νn−3 ∈
H (Γ), the function J ν is continuously differentiable with respect to εi . Hence
using the formula 2.2 we see that ϕX ν is a continuously differentiable function
with respect to the variables εi . Therefore the convergence 5.5 is uniform in a
neighborhood of any point in Wn . This concludes the proof of the theorem.
Bibliography

[1] L.V. Ahlfors. Lectures on quasiconformal mappings. Van Nostrand Mathe-


matical studies, 1966.
[2] L.V. Ahlfors. “Some remarks on Teichmüller space of Riemann surfaces”.
In: (1961), pp. 171–191.
[3] L. Bers. “Simultaneous uniformization”. In: (1960), pp. 94–97.
[4] K. T. Chan. “Uniformization of Riemann surfaces”. In: (2004).
[5] B.A. Dubrovin, S.P Novikov, and A.T. Fomenko. Modern Geometry-
Methods and Applications. Part I. The Geometry of Surfaces, Transforma-
tion Groups, and Fields. Springer Verlang, 1992, pp. 117–120.
[6] W. Fenchel and J. Nielsen. “On discontinuous groups of isometric transfor-
mations of the non-euclidean plane”. In: (1948).
[7] O. Forster. Lectures on Riemann surfaces. Springer.
[8] Y. Imayoshi and M Taniguchi. An Introduction to Teichmüller Spaces.
Springer, 1992.
[9] S. Katok. Fuchsian Groups. The University of Chigago press, 1992, pp. 23–
46.
[10] P. Koebe. “Über die Uniformisierung beliebiger analytischer Kurven”. In:
(1907), pp. 633–669.
[11] D. Kraus and O. Roth. “Conformal Metrics”. In: (2008). url: https :
//arxiv.org/pdf/0805.2235v1.pdf.
[12] J. Lehner. Discontinuous groups and automorphic functions. American
Mathematical Society, 1964.
[13] T. Oshima. “An elementary approach to the Gauss hypergeometric func-
tion”. In: (2010). url: http://akagi.ms.u- tokyo.ac.jp/~oshima/
josaixref.pdf.
[14] H. Poincaré. “Sur les groupes des équations linéaires”. In: (1884), pp. 201–
312.
[15] H. Poincaré. “Sur l’uniformisation des fonctions analytiques”. In: (1907),
pp. 1–63.
[16] B. Riemann. “Grundlagen f?r eine allgemeine Theorie der Functionen einer
ver?nderlichen complexen Gr?sse”. PhD thesis. G?ttingen, 1851.
[17] J.L. Schiff. Normal Families. Springer, 1993, pp. 54–56.

95
96 BIBLIOGRAPHY

[18] L.A. Takhtajan and P.G. Zograf. “On Liouville’s equation, Accessory
parameters, and the geometry of Teichmüller space for Riemann surfaces
of genus 0”. In: (1988).
[19] J. Vaisala. Lectures on n-dimensional quasiconformal mappings. Lecture
Notes in Mathematics. Springer, 1971, p. 229.
[20] E.T. Whittaker and G.N. Watson. A Course of Modern Analysis. Cam-
bridge University Press, 1927, pp. 462–485.

Вам также может понравиться