PhD FSTC-2018-48

The Faculty of Science, Technology and Communication

### DISSERTATION

### Presented on 14/06/2018 in Luxembourg

### to obtain the degree of

### DOCTEUR DE L’UNIVERSITÉ DU

### LUXEMBOURG

### EN MATHÉMATIQUES

### by

### Andrea Tamburelli

born on 15 April 1991 in Como (Italy)

### ANTI-DE SITTER GEOMETRY: CONVEX

### DOMAINS, FOLIATIONS AND VOLUME

### Dissertation Defence Committee

### Dr. Schlenker Jean-Marc, dissertation supervisor

Professor, Université du Luxembourg

### Dr. Wiese Gabor, Chairman

Professor, Université du Luxembourg

### Dr. Guéritaud François

Chargé de Recherche, CNRS Université de Lille 1

### Dr. Kassel Fanny

Chargée de Recherche, IHES

### Dr. Michael Wolf, Deputy Chairman

## Abstract

We study various aspects of the geometry of globally hyperbolic anti-de Sitter 3-manifolds.

For manifolds with convex space-like boundaries, homeomorphic to the product of
a closed, connected and oriented surface of genus at least two with an interval, we
prove that every couple of metrics with curvature less than_{−1 on the surface can be}
realised on the two boundary components.

For globally hyperbolic maximal compact (GHMC) anti-de Sitter manifolds, we study various geometric quantities, such as the volume, the Hausdorff dimension of the limit set, the width of the convex core and the Hölder exponent of the manifold, in terms of the parameters that describe the deformation space of GHMC anti-de Sitter struc-tures.

Moreover, we prove existence and uniqueness of a foliation by constant mean cur-vature surfaces of the domain of dependence of any quasi-circle in the boundary at infinity of anti-de Sitter space.

## Acknowledgement

First of all, I would like to thank my advisor Jean-Marc Schlenker for introducing me to the world of anti-de Sitter geometry and his invaluable advice throughout these three years. It was an honour for me working under his supervision and being part of his active and stimulating research group.

I am also truly grateful to my future post-doc mentor Mike Wolf: despite his very busy schedule, he has always found the time to talk about mathematics and share his ideas during my two visits at Rice. I am confident the next 2+1 years will be as productive and exciting.

Part of this thesis would not have been possible without the helpful suggestions of Francesco Bonsante and Andrea Seppi, with whom I hope to continue a fruitful collaboration over the years.

I thank Thierry Barbot and Sara Maloni for accepting to report this thesis, and François Guéritaud, Fanny Kassel and Gabor Wiese for being part of the defence committee.

This experience in Luxembourg would not have been so pleasant without all the people I have met during these three years: my office mates Gilles Becker, Qiyu Chen and Filippo Mazzoli with whom I shared daily the excitment and the frustration of doing research in math, and all the Ph.D. students, post-docs and professors that animated the department.

Finally, I dedicate this work to my family: although they cannot be present physically at the defence, I know they will be thinking of me. I would not have achieved this without their support.

## Contents

Introduction ix

1 Anti-de Sitter geometry 1

1.1 The Klein model . . . 1

1.2 Anti-de Sitter space as Lie group . . . 4

1.3 Surfaces in anti-de Sitter manifolds . . . 6

1.4 The universal cover of anti-de Sitter space . . . 8

1.5 GHMC anti-de Sitter three manifolds . . . 9

2 Prescribing metrics on the boundary of anti-de Sitter 3-manifolds 13 2.1 Definition of the problem and outline of the proofs . . . 13

2.2 Equivariant isometric embeddings . . . 16

2.3 Definition of the maps φ± . . . 18

2.4 Topological intersection theory . . . 20

2.5 Some properties of the maps φ± . . . 22

2.6 Prescription of an isometric embedding and half holonomy . . . 30

2.7 Proof of the main result . . . 32

3 Constant mean curvature foliation of domains of dependence 37 3.1 Definition of the problem and outline of the proofs . . . 37

3.2 Quasi-symmetric and quasi-conformal maps . . . 38

3.3 Existence of a CMC foliation . . . 40

3.4 Study of the principal curvatures of an H-surface . . . 46

3.5 Uniqueness of the CMC foliation . . . 52

3.6 Application . . . 53

4 The volume of GHMC anti-de Sitter 3-manifolds 57 4.1 Definition of the problem and outline of the proofs . . . 57

4.2 Volume and length of earthquake laminations . . . 61

4.2.1 Volume of the complement of the convex hull. . . 62

4.2.2 Length of earthquake laminations . . . 64

4.3 Holomorphic energy . . . 66

4.3.2 Holomorphic energy and Schatten energy . . . 69

4.4 L1-energy between hyperbolic surfaces . . . 71

4.5 Thurston’s asymmetric distance . . . 74

4.5.1 The general upper bound . . . 74

4.5.2 A negative result . . . 76

4.5.3 Discussion of the optimality . . . 77

4.6 Weil-Petersson distance . . . 81

4.6.1 Weil-Petersson metric on Teichmüller space . . . 81

4.6.2 A negative result . . . 82

4.6.3 A lower bound on the volume . . . 83

4.7 Gradient of length function . . . 86

4.7.1 Riera’s formula . . . 88

4.7.2 Estimates in the thick part of the hyperbolic surface . . . 89

4.7.3 Estimates in the thin part of the hyperbolic surface . . . 93

4.7.4 Conclusion of the proof and an application . . . 98

5 Entropy degeneration 101 5.1 A parameterisation using maximal surfaces . . . 101

5.1.1 Relation with Mess’ parameterisation . . . 103

5.2 Hölder exponent . . . 103

5.3 Entropy . . . 107

5.3.1 Lorentzian Hausdorff dimension . . . 107

5.3.2 Entropy of the maximal surface . . . 108

5.3.3 Estimates for the induced metric on the maximal surface . . . 109

5.3.4 Asymptotics of the Lorentzian Hausdorff dimension . . . 113

5.4 Width of the convex core . . . 115

6 Perspectives and future work 117 6.1 Prescription of metrics and measured laminations . . . 117

6.2 Convexity of volume and energy . . . 118

6.3 Special foliation of Lorentzian 3-manifolds . . . 119

## Introduction

This thesis studies various aspects of anti-de Sitter geometry and its relation with Teichmüller theory. The strong link between the two subjects was first discovered in 1990 by Geoffrey Mess, who described, in his pioneering work [Mes07], the deformation space of (2 + 1)-space-times with compact space-like slices, and found a new, and somehow simpler, proof of the celebrated Thurston’s Earthquake Theorem, exploiting the geometry of globally hyperbolic anti-de Sitter manifolds. Since then, this has become a very active area of research: namely, anti-de Sitter geometry turned out to be a convenient setting where to study earthquakes between hyperbolic surfaces ([BS12]), possibly with boundary ([BKS11], [Ros17]) and with conical singularities ([BS09]), quasi-conformal extensions of quasi-symmetric homeomorphisms of the circle ([BS16], [Sep17]) and polyhedra inscribed in quadrics ([DMS14]). This thesis fits into this framework and tries to enrich the existing literature about the description of the geometry of anti-de Sitter manifolds.

Three-dimensional anti-de Sitter space can be thought of as the analog of
hy-perbolic space in Lorentzian geometry. It can be defined as the set of time-like
vectors of R4 endowed with a bilinear form of signature (2, 2), and it is the local
model for Lorentzian manifolds with constant sectional curvature_{−1. In this thesis,}
we are interested in a special class of manifolds, locally isometric to anti-de Sitter
space, called globally hyperbolic. Those are characterised by the existence of a
space-like surface S, named Cauchy surface, that intersects any causal curve in exactly
one point. This property puts strong restrictions on the topology of these manifolds,
being them necessarily diffeomorphic to a product S_{× R ([Ger70]); nonetheless their}
geometry is very rich. Once he fixed the topological type of the surface S, Mess
studied the possible anti-de Sitter structures that can be defined on S_{× R, up to}
diffeomorphisms isotopic to the identity, that are maximal in the sense of isometric
inclusions. If S is supposed to be closed, connected and oriented, of genus τ _{≥ 2,}
Mess parameterised the deformation space of globally hyperbolic maximal anti-de
Sitter structures on S_{× R by two copies of the Teichmüller space of S. This result}
can be interpreted as the analog of Bers’ double Uniformisation Theorem ([Ber74])
for hyperbolic quasi-Fuchsian three-manifolds. The similarity between hyperbolic
quasi-Fuchsian manifolds and globally hyperbolic maximal anti-de Sitter manifolds

goes further. In both we can find a convex core, that is the smallest convex subset
onto which the manifold retracts, which has a very interesting geometry. If the
man-ifold is Fuchsian, namely the two parameters of Bers’ and Mess’ parameterisations
coincide, the convex core is a totally geodesic hyperbolic surface; otherwise it is a
three dimensional domain, homeomorphic to S_{× I, the two boundary components}
being naturally endowed with hyperbolic structures and pleated along measured
laminations. Some aspects about the geometry of the convex core are still to be
understood. In particular, two main conjectures by Thurston and Mess remain open:
Conjecture. (Thurston) The space of quasi-Fuchsian three-manifolds can be
parameterised either by the induced metrics on the boundary of the convex core or
by the two geodesic measured laminations.

Conjecture. (Mess) The space of globally hyperbolic maximal anti-de Sitter
structures on S _{× R can be parameterised either by the induced metrics on the}
boundary of the convex core or by the two geodesic measured laminations.

In both setting it is known that every couple of hyperbolic metrics ([Lab92a], [Dia13]) and every couple of filling measured laminations ([BO04], [BS12]) can be realised, but uniqueness is still unknown.

However, one can ask similar questions for other compact, convex domains
that contain the convex core. In fact, it is possible to talk about the induced metrics
on the two boundary components and the role of the measured geodesic laminations
is replaced in a natural sense by the third fundamental forms. In the hyperbolic
setting, this has been first studied by Labourie ([Lab92a]) who proved that any
smooth metrics on S with curvature at least _{−1 can be realised on the boundary}
of a compact, convex hyperbolic three-manifold, and later by Schlenker ([Sch06]),
who proved the uniquess part of this question and extended this result to the third
fundamental form, as well.

In Chapter 2, we address the existence part of this problem in the anti-de Sitter setting. More precisely, we prove the following:

Theorem A. ([Tam18]) Let g± be two smooth metrics with curvature less

than_{−1 on a closed, connected, oriented surface S of genus τ ≥ 2. Then there exists}
a compact, convex, globally hyperbolic anti-de Sitter manifold with convex boundary
M ∼= S× I such that the metrics induced on the two boundary components are g±.

A similar result holds also for the third fundamental forms.

xi

M ∼= S× R, with induced metrics g±on the two boundary components. Then, using

Mess’ parameterisation, for every smooth metric g on S with curvature less than_{−1,}
we construct a smooth map φ_{g} from the space of equivariant isometric embeddings
of (S, g) into anti-de Sitter space to Teich(S) _{× Teich(S), which associates to}
an isometric embedding the holonomy representation of the globally hyperbolic
maximal anti-de Sitter manifold in which (S, g) is contained. Therefore, the proof
of Theorem A follows by showing that the images of φg+ and φg− are never disjoint,
if g± are any two smooth metrics with curvature less than−1. This is accomplished

by proving it directly for a specific couple of metrics g±, and by then verifying that

the intersection persists when deforming one of the two metrics.

The geometry of globally hyperbolic maximal anti-de Sitter manifolds can also be understood using special foliations by space-like surfaces. In case of closed Cauchy surfaces, this theory was developed by Barbot, Béguin and Zeghib, who proved that every such manifold can be foliated uniquely by constant mean curvature surfaces ([BBZ07]) and constant Gauss curvature surfaces ([BBZ11]). These results have been recently generalised in different directions: when conical singularities of angle less than π along time-like geodesics are allowed ([CS16], [QT17]), and when there is no co-compact action of a surface group ([BS16],[Tam16]). Chapter 3 focuses on the latter problem for constant mean curvature surfaces. If we identify the universal cover of a globally hyperbolic maximal anti-de Sitter manifold M ∼= S× R with a domain of dependence in anti-de Sitter space, the foliation by constant mean curvature surfaces is lifted to a foliation by discs of constant mean curvature of the domain of dependence. These discs intersect the boundary at infinity of anti-de Sitter space in a curve, called quasi-circle, that can be interpreted as the graph of the quasi-symmetric homeomorphism of the circle that conjugates the two Fuchsian representations in Mess’ parameterisation. It is thus natural to ask if a foliation by constant mean curvature surfaces exists for more general domains of dependence, whose closure intersect the boundary at infinity of anti-de Sitter space in a general quasi-circle.

Theorem B. ([Tam16]) Let φ : S1 _{→ S}1 be a quasi-symmetric
homeomo-prhism of the circle and let c_{φ} be the corresponding quasi-circle in the boundary
at infinity of anti-de Sitter space. Then there exists a unique foliation by constant
mean curvature surfaces of the domain of dependence of cφ.

The proof relies on an approximation argument. Namely, every quasi-circle cφcan be seen as a limit in the Hausdorff topology of quasi-circles cnthat are graphs

of quasi-symmetric homeomorphisms that conjugate two Fuchsian representations. We show that the sequence of constant mean curvature surfaces with boundary at infinity cnconverges to a constant mean curvature surface asymptotic to cφand that

family of quasi-conformal extensions of φ: a surface with constant mean curvature H and boundary at infinity cφ provides a quasi-conformal extension ΦH of φ with

the following property. The map Φ_{H} can be decomposed uniquely as Φ_{H} = f2◦ f1−1

where f1 and f2 are harmonic maps of the hyperbolic plane with Hopf differential

Hopf(f1) = e2iθHopf(f2), and θ =− arctan(H) +π_{2}.

In the second part of the thesis, we address the general question of
describ-ing the geometry of a globally hyperbolic maximal anti-de Sitter manifold with
compact Cauchy surface in terms of the two points in Teichmüller space provided
by Mess’ parameterisation. The first interesting geometric quantity that we study
in Chapter 4 is the volume of the convex core. An analogous question for hyperbolic
quasi-Fuchsian manifolds was investigated by Brock ([Bro03]), who showed that
the volume of the convex core is roughly equivalent to the Weil-Petersson distance
between the two corresponding points in Bers’ parameterisation. It turns out that
a similar result does not hold in this Lorentzian setting, as we are able to construct
a sequence of globally hyperbolic maximal anti-de Sitter manifolds such that the
volume of their convex core diverges, but the Weil-Petersson distance between the
two Mess’ parameters remains bounded. However, we find a quantity that
ap-proximates the volume of the convex core up to multiplicative and additive constants:
Theorem C. ([BST17]) Let M_{h,h}0 be the globally hyperbolic maximal anti-de
Sitter manifold corresponding to (h, h0) _{∈ Teich(S) × Teich(S) in Mess’ }
parameter-isation. Then the volume of the convex core of M_{h,h}0 is coarsely equivalent to the
L1-energy between the hyperbolic surfaces (S, h) and (S, h0).

The L1-energy between hyperbolic surfaces is defined as the infimum, over all C1 maps f isotopic to the identity, of the L1-norm of the differential of f . Very few is known about this quantity: in contrast with the more studied L2-energy that is realised by the L2-norm of the differential of the unique harmonic map isotopic to the identity, we do not know, for instance, if the infimum is attained. As a consequence of Theorem C, we shed some light about the behaviour of the L1-energy: Corollary D. ([BST17]) Let h, h0 be two hyperbolic metrics on S and suppose that h0 is obtained from h by an earthquake along a measured geodesic lamination λ. Then, the L1-energy between (S, h0) and (S, h0) is roughly equivalent to the length of λ. By Thurston’s Earthquake Theorem, we have two possible choices for λ, de-pending on whether we perfom a left or right earthquake. Our techniques show that the lengths of these two laminations are comparable, being their difference bounded by an explicit constant that depends only on the topology of the surface.

xiii

Lorentzian Hausdorff dimension of the limit set and the Hölder exponent. As
we discussed above, the limit set of a globally hyperbolic maximal anti-de Sitter
manifold M with compact Cauchy surface can be identified with the graph c_{φ} of a
quasi-symmetric homeomorphism φ of the circle. This is the Lorentzian analog of
the quasi-circle that appears as limit set of a quasi-Fuchsian group acting on the
three-dimensional hyperbolic space. However, while the Hausdorff dimension of the
limit set of a quasi-Fuchsian group varies between 1 and 2 ([Sul84]) and is equal
to 1 if and only if the group is Fuchsian ([Bow79]), the Hausdorff dimension of c_{φ}
is always 1. In particular, it does not distinguish if a representation is Fuchsian.
Glorieux and Monclair introduced a notion of Lorentzian Hausdorff dimension
that fits this issue ([GM16]): roughly speaking, they replaced Euclidean balls with
Lorentzian ones in the classical definition of Hausdorff dimension, obtaining thus a
quantity that is always bounded by 1 and is equal to 1 if and only if the manifold
is Fuchsian. The Hölder exponent of M is also related to the homeomorphism φ: it
is the minimum between the best Hölder exponent of φ and φ−1. We provide an
explicit formula for this that depends only on the holonomy representation of M .
The main results of Chapter 5 concern the asymptotic behaviour of these
quantities. More precisely, we use the parameterisation of the deformation space of
globally hyperbolic maximal anti-de Sitter structures on S _{× R by the cotangent}
bundle to the Teichmüller space of S ([KS07]) and study the asymptotic behaviour
along rays of quadratic differentials:

Theorem E. ([Tam17]) Let M_{t} be the family of globally hyperbolic maximal
anti-de Sitter manifolds associated to the ray (h, tq) _{∈ T}∗Teich(S). Then the
Lorentzian Hausdorff dimension of the limit set and the Hölder exponent of Mt tend

to 0 when t goes to +_{∞.}

In order to explain this result, let us first recall how the parameterisation by
T∗Teich(S) works: to a point (h, q) _{∈ T}∗Teich(S) one associates the globally
hyperbolic maximal anti-de Sitter manifold M ∼= S × R which has an embedded
maximal surface with induced metric conformal to h and with second fundamental
form determined by the real part of q. The existence of such a manifold is obtained
by solving a quasi-linear PDE. By studying carefully this differential equation, we
are able to provide estimates for the induced metric on the maximal surface along
rays of quadratic differentials, and prove that its volume entropy converges to 0
when t goes to +_{∞. The proof of the first part of Theorem E then follows from the}
fact that the Lorentzian Hausdorff dimension is bounded from above by the entropy
of the maximal surface.

compactification of Teichmüller space ([Wol89]) with our explicit formula for the Hölder exponent, we deduce the second part of Theorem E.

### Outline of the thesis

The thesis is organised as follows. Chapter 1 introduces anti-de Sitter geometry and reviews the main classical results in this field. In Chapter 2 we study globally hyperbolic anti-de Sitter manifolds with convex space-like boundary and the problem of finding a manifold with prescribed metric on the boundary. The material of this chapter can be found in:

[Tam18] Tamburelli, A. "Prescribing metrics on the boundary of anti-de Sit-ter 3-manifolds". InSit-ternational Mathematics Research Notices, Volume 2018, Issue 5, pp. 1281-1313, 2018.

In Chapter 3 we prove the existence and uniqueness of a foliation by con-stant mean curvature surfaces of the domain of dependence of a quasi-circle in the boundary at infinity of anti-de Sitter space. The content of this chapter has been published in:

[Tam16] Tamburelli, A. "Constant mean curvature foliation of domains of de-pendence in anti-de Sitter space". To appear in Transactions of the AMS.

Chapter 4 deals with the volume of globally hyperbolic maximal anti-de Sit-ter manifolds: we compare the volume of the convex core and the volume of the entire manifold and find coarse estimates in terms of the L1 energy, the Weil-Petersson distance and Thurston’s asymmetric distance between the two points in Teichmüller space given by Mess’ parameterisation. These results can be found in: [BST17] Bonsante, F., Seppi, A., Tamburelli A. "On the volume of anti-de Sitter maximal globally hyperbolic three-manifolds". Geometric and Functional Analysis, Volume 27, Issue 5, pp. 1106-1160, 2017.

In Chapter 5, we use the parameterisation of globally hyperbolic anti-de Sit-ter structures by the cotangent of the Teichmüller space to describe the behaviour of the entropy of the maximal surface and the Lorentzian Hausdorff dimension of the limit set along rays of quadratic differentials. The material covered here has appeared in the preprint:

### Chapter 1

## Anti-de Sitter geometry

In this chapter we introduce the protagonist of the thesis, i.e the three-dimensional anti-de Sitter space. The material covered here is classical, the main objective being fixing the notation and recalling the well-established results in the field.

### 1.1

### The Klein model

Let us denote with R2,2 the vector space R4 endowed with the bilinear form of signature (2, 2):

hx, yi2,2= x0y0+ x1y1− x2y2− x3y3 .

We define

d

AdS3={x ∈ R2,2 | hx, xi2,2 =−1} .

The restriction of the bilinear form _{h·, ·i}_{2,2} to the tangent space of dAdS3 induces

a Lorentzian metric on dAdS3 with constant sectional curvature −1. Given a point

p_{∈ d}AdS3 and a tangent vector v ∈ TpAdSd_{3}, we will say that
• v is space-like, if hv, vi2,2> 0;

• v is light-like, if hv, vi2,2= 0;

• v is time-like, if hv, vi2,2 < 0.

Similarly, we say that a geodesic γ in dAdS3 is space-like (resp. light-like or

time-like) if ˙γ is space-like (resp. light-like or time-time-like). It is straightforward to verify that geodesics are obtained by intersecting planes through the origin of R2,2 with

d

AdS3. The causal type of the geodesic can be understood from the signature of the

restriction of the bilinear form_{h·, ·i}2,2 to the plane:

• if it is degenerate and the intersection with dAdS3 is non-empty, we obtain a

light-like geodesic;

• if it has signature (0, 2) we obtain a time-like geodesic.

Analogously, totally geodesic planes are obtained by intersecting dAdS3 with

hyper-planes of R2,2. Given a totally geodesic plane P , we say that • P is space-like if the induced metric on P is positive definite; • P is light-like if the induced metric on P is degenerate; • P is time-like if the induced metric on P is Lorentzian.

Again, the induced metric on P can be easily deduced by studying the signature of
the restriction of the bilinear form_{h·, ·i}_{2,2} on the hyperplane that defines P .

We endow dAdS3 with the orientation induced by the standard orientation of R4. A

time-orientation is the choice of a never-vanishing time-like vector field X on dAdS3.

The isometry group of orientation and time-orientation preserving isometries of dAdS3

is the connected component of SO(2, 2) containing the identity.

We define anti-de Sitter space AdS_{3} as the image of the projection of dAdS3 into

RP3. More precisely, if we denote with π : R4\ {0} → RP3 the canonical projection, anti-de Sitter space is

AdS3 = π({x ∈ R2,2 | hx, xi2,2< 0}) .

It can be easily verified that π : dAdS3 → AdS3is a double cover, hence we can endow

AdS3 with the unique Lorentzian structure that makes π a local isometry. This is

called the Klein model of anti-de Sitter space, in analogy with the more familiar Klein model of hyperbolic geometry. It follows from the definition and the above discussion that geodesics and totally geodesics planes are obtained by intersecting AdS3 with projective lines and planes.

In order to better visualise anti-de Sitter space, it is convenient to consider the
intersection with an affine chart. Let U_{3} = _{{[x}0, x1, x2, x3] ∈ RP3 | x3 6= 0}. The

map ϕ3 : U3 → R3 [x]7→ x0 x3 ,x1 x3 ,x2 x3

1.1. The Klein model 3

spacelike geodesic timelike geodesic

lightlike geodesic

Figure 1.1: Geodesics in AdS_{3}.

It is natural to define the boundary at infinity of anti-de Sitter space as ∂∞AdS3 = π({x ∈ R4 | hx, xi2,2= 0}).

It can be easily verified that ∂∞AdS3coincides with the image of the Segre embedding

s : RP1× RP1→ RP3 ,

hence the boundary at infinity of anti-de Sitter space is a double-ruled quadric
home-omorphic to a torus. We will talk about left and right ruling in order to distinguish
the two rulings. This homeomorphism can be also be described geometrically in the
following way. Fix a totally geodesic plane P_{0} in AdS_{3}. The boundary at infinity of
P0 is a circle. Let ξ ∈ ∂∞AdS3. There exists a unique line of the left ruling lξ and

a unique line of the right ruling r_{ξ} passing through ξ. The identification between
∂∞AdS3 and S1× S1 induced by P0 associates to ξ the intersection points πl(ξ) and

πr(ξ) between lξ and rξ and the boundary at infinity of P0. These two maps

πl : ∂∞AdS3→ S1 πr: ∂∞AdS3 → S1

are called left and right projections, respectively. In the affine chart U3the boundary

at infinity of AdS_{3} coincides with the quadric of equation x2+ y2_{− z}2 = 1.

The action of orientation and time-orientation preserving isometries of AdS3extends

Given a map φ : RP1 → RP1_{, the identification between ∂}

∞AdS3 and RP1× RP1

described above allows us to represent the graph of φ as a curve c_{φ}on the boundary
at infinity of anti-de Sitter space. If φ is an orientation-preserving homeomorphism,
the curve cφ turns out to be weakly space-like, i.e for every ξ ∈ cφ the curve cφ

is contained in the region bounded by the lines through ξ in the left and right
ruling which is connected to ξ by like paths. Viceversa, every weakly
space-like curve Γ _{⊂ ∂}∞AdS3 can be obtained as a graph of an orientation-preserving

homeomorphism of the circle.

Given a weakly space-like curve Γ on the boundary at infinity of anti-de Sitter space, we define two objects that will play a fudamental role in the theory of globally hyperbolic manifolds outlined in Section 1.5:

• the convex hull of Γ is the smallest closed convex subset of AdS3 with boundary

at infinity Γ and it will be denoted withC(Γ);

• the domain of dependence D(Γ) of Γ is the set of points p ∈ AdS3 ⊂ RP3 such

that the plane p∗, which is the projective dual of p, is disjoint from Γ. Domains of dependence are always contained in an affine chart and admit only light-like support planes.

### 1.2

### Anti-de Sitter space as Lie group

Let gl(2, R) be the vector space of 2-by-2 matrices with real coefficients. The quadratic form

q(A) =− det(A)

induces, by polarisation, a scalar product η on gl(2, R), which in the basis consisting of elementary matrices can be represented by

η(X, Y ) = Xt
0 0 0 _{−}1_{2}
0 0 1_{2} 0
0 1_{2} 0 0
−1
2 0 0 0
Y .

It follows easily that η has signature (2, 2).

Let us now consider the submanifold SL(2, R) = {A ∈ gl(2, R) | q(A) = −1}. We claim that SL(2, R) endowed with the restriction of η is a 3-dimensional Lorentzian manifold. Since η is invariant by left- and right- multiplication by elements of SL(2, R) (because q is), it is sufficient to check this at Id∈ SL(2, R). Now

TIdSL(2, R) = sl(2, R) ={A ∈ gl(2, R) | trace(A) = 0} .

In the basis of sl(2, R) given by

1.2. Anti-de Sitter space as Lie group 5

the scalar product η is represented by the matrix

η|_{sl(2,R)}(X, Y ) = Xt
1 0 0
0 0 1_{2}
0 1_{2} 0
Y .

Hence (SL(2, R), η) is a Lorentzian manifold that we denote by dAdS3. From the

above computation, it follows also that η(X, Y ) = 1

2trace(XY )
for every X, Y _{∈ sl(2, R).}

The group SL(2, R)× SL(2, R) acts on dAdS3 as

(A, B)_{· X := AXB}−1

by isometries. _{In particular, η induces a Lorentzian structure on PSL(2, R) ∼}=
SL(2, R)/{±Id}, which we identify with anti-de Sitter space AdS3.

Remark 1.2.1. An explicit isometry between dAdS3 as introduced in Section 1.1 and

(SL(2, R), η) is given by the restriction of the map R4 → dAdS3

(x0, x1, x2, x3)7→

x0+ x1 x3+ x2

x2− x3 x0− x1

where (x0, x1, x2, x3)∈ R4are coordinates with respect to a basis in which the bilinear

form of signature (2, 2) is

hx, yi2,2 =−x0y0+ x1y1+ x2y2− x3y3 .

If we see AdS3 ⊂ Pgl(2, R), we can define the boundary at infinity of AdS3 as

∂∞AdS3 = P({A ∈ gl(2, R) \ {0} | q(A) = 0}) ,

namely the projectivisation of rank 1-matrices. This can then be identified with RP1× RP1 by

∂∞AdS3 → RP1× RP1

[M ]_{7→ ([Im(M)], [Ker(M)]) .}

It is easy to check that the action of PSL(2, R)× PSL(2, R) on AdS3 extends to

the boundary at infinity and, in the above identification, coincides with the obvious
action of PSL(2, R)× PSL(2, R) on RP1_{× RP}1_{.}

Geodesics of AdS3 are obtained by intersection of projective planes with AdS3.

• a space-like geodesic at x with tangent vector v is lifted to
exp_{x}(tv) = cosh(t)ˆx + sinh(t)ˆv ;

• a time-like geodesic at x with tangent vector v is the projectivisation of
exp_{x}(tv) = cos(t)ˆx + sin(t)ˆv ;

• a light-like geodesic at x with tangent vector v lifts to
exp_{x}(tv) = ˆx + tˆv .

In particular, geodesics through [Id]_{∈ PSL(2, R) are 1-parameter subgroups.}

### 1.3

### Surfaces in anti-de Sitter manifolds

In this section we describe the theory of immersions of surfaces in anti-de Sitter space, which is a straightforward adaptation of the classical theory for Euclidean space.

Let us denote with_{∇}AdS the Levi-Civita connection of the Lorentzian metric g_{AdS}
of AdS_{3}. Given a smooth immersion σ : ˜S _{→ AdS}3, the first fundamental form is

the pull-back of the induced metric

I(V, W ) = gAdS(dσ(V ), dσ(W )) V, W ∈ Γ(T ˜S) .

We say that σ( ˜S) is space-like if the first fundamental form is a Riemannian metric. From now on, we will always suppose that the immersion is space-like.

We denote with N the future-directed unit normal vector field on σ( ˜S). Since the
immersion is space-like, N is a time-like vector. The Levi-Civita connection _{∇}I of
the first fundamental form I of ˜S is defined from the relation:

∇AdS

V W =∇IVW + II(V, W )N ,

and II(V, W ) is called the second fundamental form of the immersion. The shape
operator B_{∈ End(T ˜}S) of ˜S is defined as

B(V ) =_{−∇}AdS_{V} N .

It turns out that B is I-self-adjoint and the second fundamental form is related to the shape operator by

II(V, W ) = I(B(V ), W ) .

1.3. Surfaces in anti-de Sitter manifolds 7 • the anti-de Sitter version of the Guass equation:

det(B) =_{−1 − K}I ,

where we have denoted with K_{I}the Gaussian curvature of the first fundamental
form;

• the Codazzi equation

d∇IB = 0 ,

where d∇I : Ω1(T ˜S)→ Ω2_{(T ˜}_{S) is the operator defined by:}

(d∇IB)(V, W ) =_{∇}I_{V}(B(W ))_{− ∇}I_{W}(B(V ))_{− B([V, W ]) .}

As for Euclidean space, the embedding data I and B of a simply connected surface determines the immersion uniquely up to global isometries of AdS3:

Theorem 1.3.1 (Fundamental theorem of surfaces in anti-de Sitter space). Let ˜S be a simply connected surface. Given a Riemannian metric I and an I-self-adjoint operator B : T ˜S → T ˜S, satisfying the Gauss-Codazzi equations

det(B) =−1 − KI

d∇IB = 0

there exists a smooth immersion σ : ˜S _{→ AdS}3 such that the first fundamental

form is I and the shape operator is B. Moreover, σ is uniquely determined up to post-composition with an isometry of AdS3.

We can also define the third fundamental form of ˜S as III(V, W ) = I(B(V ), B(W )) .

We notice that if ˜S is strictly convex, i.e. the determinant of B is strictly positive at
every point, then the third fundamental form is a Riemannian metric. Notice that,
by the Gauss equation, this is equivalent to say that the curvature of the induced
metric is strictly smaller than _{−1.}

The third fundamental form is linked to a duality between convex surfaces in anti-de Sitter space. More precisely, the projective duality between points and planes in RP3 induces a duality between convex space-like surfaces in AdS3: given a convex

space-like surface, the dual surface ˜S∗ is defined as the set of points which are dual to the support planes of ˜S. The relation between ˜S and ˜S∗ is summarised in the following proposition:

Proposition 1.3.2 ([BBZ11]). Let ˜S _{⊂ AdS}3 be a smooth space-like surface with

curvature κ <_{−1. Then}

• the pull-back of the induced metric on ˜S∗ through the duality map is the third fundamental form of ˜S;

• if κ is constant, the dual surface ˜S∗ has curvature κ∗ =_{−}_{κ+1}κ .

### 1.4

### The universal cover of anti-de Sitter space

As the careful reader might have noticed from the description of the Klein model in Section 1.1, anti-de Sitter space in not simply-connected, being it diffeomorphic to a solid torus. It is sometimes convenient to work in the Universal cover, especially when dealing with space-like embeddings of surfaces into AdS3.

Let us denote with H2 the hyperbolic plane. In this section we will always think of H2as one connected component of the two-sheeted hyperboloid in Minkowksy space. The map

F : H2× S1 → dAdS3

(x0, x1, x2, eiθ)7→ (x0cos(θ), x1, x2, x0sin(θ))

is a diffeomorphism, hence H2× S1 _{is isometric to anti-de Sitter space, if endowed}

with the pull-back metric

(F∗gAdS3)(x,eiθ)= (gH2)x− x 2 0dθ2 .

We easily deduce that the Universal cover of anti-de Sitter space can be realised as g

AdS3∼= H2× R endowed with the Lorentzian metric:

(g
d
AdS)(x,t) = (gH2)x− x
2
0dt2 .
We will denote
χ2 =_{−}
∂
∂t
2
and
gradt =− 1
χ2
∂
∂t .

The Universal cover is particularly useful to study embedded space-like surfaces. In fact, space-like surfaces in gAdS3 are graphs of functions ([BS10, Proposition 3.2])

u : H2 → R
x_{7→ u(x) .}

Moreover, the space-like condition provides a uniform bound on the gradient of u.
For instance, let us consider the function ˆ_{u on H}_{× R given by}

ˆ

1.5. GHMC anti-de Sitter three manifolds 9

The correspondent space-like surface is defined by the equation ˆu(x)− t = 0. This surface is space-like if and only if the normal vector at each point

ν =−χ2_{gradt}_{− grad(ˆu)}

is time-like. We deduce the uniform bound kgrad(u)k2< 1

χ2

on the gradient of the function u. In particular, space-like surfaces are graphs of Lipschitz functions.

### 1.5

### GHMC anti-de Sitter three manifolds

A 3-dimensional anti-de Sitter space-time is a manifold N locally isometric to AdS3

with a fixed orientation and time-orientation. This means that N is endowed with an atlas of charts taking values on AdS3 so that the transition functions are restrictions

of elements in PSL(2, R)× PSL(2, R).

We are actually interested in a special class of anti-de Sitter manifolds.

Definition 1.5.1. An anti-de Sitter manifold N is Globally Hyperbolic Maximal Compact (GHMC) if it satisfies the following conditions:

1. Global Hyperbolicity: N contains a space-like surface, called Cauchy-surface, that intersects every inextensible causal curve in exactly one point;

2. Maximality: if N0 is another globally hyperbolic AdS_{3} space-time and φ : N _{→}
N0is any isometric embedding sending a Cauchy surface into a Cauchy surface,
then φ is a global isometry;

3. Spacial Compactness: if the Cauchy surface is compact.

The first condition implies that N must be diffeomorphic to S_{× R ([Ger70]), where}
S is homeomorphic to the Cauchy surface of N . We will always assume that S is a
closed, connected, oriented surface of genus τ _{≥ 2. We will denote with GH(S) the}
deformation space of GHMC anti-de Sitter structures on S_{× R. By the pioneering}
work of Mess, the deformation theory of GHMC anti-de Sitter structures is strongly
related to Teichmüller theory. This becomes evident from the following result:
Theorem 1.5.2 ([Mes07]). GH(S) is parameterised by Teich(S) × Teich(S).
The parameterisation goes as follows. First, recall that the Teichmüller space of S is
identified to a certain connected component in the space of representations of π1(S)

on S, and the desired connected component is given by the subset of representations with maximal Euler class, called Fuchsian ([Gol80]):

Teich(S) ∼={ρ0 : π1(S)→ PSL(2, R) : e(ρ0) =|χ(S)|}/PSL(2, R) .

Mess proved that for every GHMC AdS_{3} manifold M , the holonomy representation
ρ = (ρl, ρr) : π1(S)→ PSL(2, R) × PSL(2, R)

satisfies e(ρl) = e(ρr) =|χ(S)|, and therefore ([ρl], [ρr]) defines a point in Teich(S)×

Teich(S). The representations ρland ρrare called left holonomy and right holonomy.

Example 1.5.3. If h is a hyperbolic metric on S, then one can define the following
metric on M = S_{× (−π/2, π/2), where t is the “vertical” coordinate:}

gh=−dt2+ cos2(t)h . (1.1)

It turns out that g_{h} has constant sectional curvature _{−1, that S × {0} is a totally}
geodesic Cauchy surface, and that (M, g_{h}) is maximal globally hyperbolic. It can be
verified that, in this case, ρl = ρr. The maximal globally hyperbolic manifolds for

which [ρ_{l}] = [ρr]∈ Teich(S) are called Fuchsian and correspond to the diagonal in

GH(S) ∼= Teich(S)× Teich(S) .

Equivalently, they contain a totally geodesic spacelike surface isometric to H2/ρ0(π1(S)), where ρ0 := ρl= ρr.

Going back to Theorem 1.5.2, Mess explicitly constructed an inverse of the map GH(S) → Teich(S) × Teich(S) we have just defined. Given a couple (ρl, ρr) of

Fuch-sian representation, there exists a unique orientation-preserving homeomorphism
φ : RP1 → RP1 _{such that}

φ_{◦ ρ}l(γ) = ρr(γ)◦ φ

for every γ_{∈ π}1(S). As explained in Section 1.1, we can see the graph of φ as a curve

cφ on the boundary at infinity of anti-de Sitter space. It turns out that ρ(π1(S)) =

(ρl(π1(S)), ρr(π1(S))) acts properly discontiously on the domain of dependence of

cφ and the quotient is a GHMC anti-de Sitter manifold, with holonomy ρ. We will

denote this manifold by

Mhl,hr :=D(φ)/(ρl, ρr)(π1(S)) ,

where h_{l} and h_{r} _{are the hyperbolic metrics of S induced by H}2/ρl(π1(S)) and

H2/ρr(π1(S)) respectively. We will often refer to hl and hr as the left and right

metric. It follows from Mess’ proof that the class of Mhl,hr in GH(S) only depends
on the isotopy classes of h_{l} and h_{r}.

The quotient of the convex hull of cφ

1.5. GHMC anti-de Sitter three manifolds 11

is called the convex core of Mhl,hr, and it is the smallest compact, convex subset
homotopy equivalent to M_{h}_{l}_{,h}_{r}. The convex core has an interesting geometry: if
it is not a totally geodesic space-like surface, which happens only if h_{l} = hr, its

boundary consists of two space-like surfaces homeomorphic to S, naturally endowed with hyperbolic metrics and pleated along measured laminations.

Moreover, the left and right hyperbolic metrics corresponding to the left and right
representations can be constructed explicitly starting from space-like surfaces
embed-ded in M_{h}_{l}_{,h}_{r}. Mess gave a description in a non-smooth setting using the upper and
lower boundary of the convex core of M_{h}_{l}_{,h}_{r} as space-like surfaces. More precisely, if
m± are the hyperbolic metrics on the upper and lower boundary of the convex core
and λ± are the measured geodesic laminations along which they are pleated, the left
and right metrics h_{l} and h_{r} are related to m± by an earthquake along λ±:

hl = Eλ + l (m+) = Eλ − r (m−) hr= Eλ + r (m+) = Eλ − l (m−) .

Mess obtained in this way a new proof of Thurston’s Earthquake Theorem:

Theorem 1.5.4 (Earthquake theorem). Given two hyperbolic metrics h, h0 on a
closed oriented surface S, there exists a unique pair of measured laminations λ_{l}, λ_{r}
such that

Eλl

l (h) = h

0 _{and} _{E}λr

r (h) = h0 .

Later, this description was extended ([KS07]), thus obtaining explicit formulas for
the left and right metric, in terms of the induced metric I, the complex structure J
and the shape operator B of any strictly negatively curved smooth space-like surface
S embedded in Mhl,hr. The construction goes as follows. We fix a totally geodesic
space-like plane P_{0}. Let ˜S _{⊂ AdS}3 be the universal cover of S. Let ˜S0 ⊂ U1AdS3

be its lift into the unit tangent bundle of AdS3 and let p : ˜S0 → ˜S be the canonical

projection. For any point (x, v) _{∈ ˜}S0, there exists a unique space-like plane P in
AdS3 orthogonal to v and containing x. We define two natural maps Π∞,l and Π∞,r

from ∂∞P to ∂∞P0, sending a point x∈ ∂∞P to the intersection between ∂∞P0 and

the unique line of the left or right foliation of ∂∞AdS3containing x. Since these maps

are projective, they extend to hyperbolic isometries Π_{l}, Πr : P → P0. Identifying P

with the tangent space of ˜S at the point x, the pull-backs of the hyperbolic metric on P0 by Πl and by Πr define two hyperbolic metrics on ˜S

hl= I((E + J B)·, (E + JB)·) and hr= I((E− JB)·, (E − JB)·) .

The isotopy classes of the corresponding metrics on S do not depend on the choice
of the space-like surface S and their holonomies are precisely ρ_{l} and ρ_{r}, respectively
([KS07, Lemma 3.6]).

By applying this construction to the unique maximal surface (i.e. with vanishing mean curvature) S embedded in a GHMC AdS3 manifold, Krasnov and Schlenker

Definition 1.5.5. An orientation-preserving diffeomorphism m : (S, h) _{→ (S, h}0)
is minimal Lagrangian if it is area-preserving and its graph is a minimal surface in
(S_{× S, h ⊕ h}0).

It is known [BS10, Proposition 1.3] that minimal Lagrangian diffeomorphisms are
characterized by having a decomposition m = (f0)_{◦f}−1, where f and f0are harmonic
maps from a Riemann surface (S, X) with opposite Hopf differential.

It turns out that in this case Π_{l,r} induce harmonic diffeomorphisms between (S, I)
and (S, h_{l,r}), which have opposite Hopf differential. Hence we obtain a minimal
Lagrangian diffeomorphism between (S, hl) and (S, hr) that factors through the

### Chapter 2

## Prescribing metrics on the

## boundary of anti-de Sitter

## 3-manifolds

In this chapter we prove that given two metrics g_{+} and g− with curvature κ <−1

on a closed, oriented surface S of genus τ _{≥ 2, there exists an AdS}_{3} manifold N
with smooth, space-like, strictly convex boundary such that the induced metrics on
the two connected components of ∂N are equal to g_{+} and g−. Using the duality

between convex space-like surfaces in AdS_{3}, we obtain an equivalent result about
the prescription of the third fundamental form.

### 2.1

### Definition of the problem and outline of the proofs

As it should be clear from Chapter 1, the 3-dimensional anti-de Sitter space AdS3 is

the Lorentzian analogue of hyperbolic space, and globally hyperbolic maximal com-pact AdS3 manifolds share many similarities with hyperbolic quasi-Fuchsian

man-ifolds. As a consequence, it is possible to formulate many classical questions of quasi-Fuchsian manifolds even in this Lorentzian setting. The question we address here is the following. Let K be a compact, convex subset with two smooth, strictly convex, space-like boundary components in a GHMC AdS3 manifold. By the Gauss

formula, the boundaries have curvature κ <_{−1. We can ask if it is possible to realise}
every couple of metrics, satisfying the condition on the curvature, on a surface S via
this construction. The analogous question has a positive answer in a hyperbolic
set-ting ([Lab92a]), where even a uniqueness result holds ([Sch06]). In this chapter, we
will follow a construction inspired by the work of Labourie ([Lab92a]), in order to
obtain a positive answer in the anti-de Sitter world. The main result of the chapter
is thus the following:

Corollary 2.3.3. For every couple of metrics g+ and g− on S with curvature less

than _{−1, there exists a globally hyperbolic convex compact AdS}3 manifold K ∼= S×

[0, 1], whose induced metrics on the boundary are exactly g±.

Using the duality between space-like surfaces in anti-de Sitter space, we obtain an analogous result about the prescription of the third fundamental form:

Corollary 2.3.4. For every couple of metrics g+ and g− on S with curvature less

than _{−1, there exists a globally hyperbolic convex compact AdS}_{3} manifold K ∼= S×
[0, 1], such that the third fundamental forms on the boundary components are g+and

g−.

We outline here the main steps of the proof for the convenience of the reader.
The first observation to be done is that Corollary 2.3.3 is equivalent to proving
that there exists a GHMC AdS_{3} manifold M containing a future-convex space-like
surface isometric to (S, g−) and a past-convex space-like surface isometric to (S, g+).

Adapting the work of Labourie ([Lab92a]) to this Lorentzian setting, we prove that the space of isometric embeddings I(S, g±)± of (S, g±) into a GHMC AdS3 manifold

as a future-convex (or past-convex) space-like surface is a manifold of dimension 6τ−6. On the other hand, by the work of Mess ([Mes07]), the space of GHMC AdS3

structures is parameterised by two copies of Teichmüller space, hence a manifold of
dimension 12τ_{− 12. This allows us to translate our original question into a question}
about the existence of an intersection between subsets in Teich(S)_{× Teich(S). More}
precisely, we will define in Section 2.3 two maps

φ±_{g}_{±} : I(S, g±)±→ Teich(S) × Teich(S)

sending an isometric embedding of (S, g±) to the holonomy of the GHMC AdS3

manifold containing it. Corollary 2.3.3 is then equivalent to the following:

Theorem 2.3.2. For every couple of metrics g_{+} and g− on S with curvature less

than _{−1, we have}

φ+_{g}_{+}(I(S, g+)+)∩ φ−g−(I(S, g−)
−_{)}

6= ∅ .

In order to prove this theorem we will use tools from topological intersection theory,
which we recall in Section 2.4. For instance, Theorem 2.3.2 is already known to
hold under particular hypothesis on the curvatures ([BMS15]), hence we only need
to check that the intersection persists when deforming one of the two metrics on the
boundary, as the space of smooth metrics with curvature less than_{−1 is connected}
(see e.g. [LS00, Lemma 2.3]). More precisely, given any smooth paths of metrics g±_{t}
with curvature less than _{−1, we will define the manifolds}

W± = [

t∈[0,1]

2.1. Definition of the problem and outline of the proofs 15

and the maps

Φ± : W± _{→ Teich(S) × Teich(S)}

with the property that the restrictions of Φ± to the two boundary components co-incide with φ±

g±_{0} and φ
±

g_{1}±. We will then prove the following:

Proposition 5.1. The maps Φ± are smooth.

Hence, we will have the necessary regularity to apply tools from intersection theory.
In particular, we can talk about transverse maps and under this condition we can
define the intersection number (mod 2) of the maps φ+_{g}

+ and φ
−
g− as the cardinality
(mod 2), if finite, of (φ+_{g}
+× φ
−
g−)
−1_{(∆), where}

φ+_{g}_{+}_{× φ}−_{g}_{−} : I(S, g+)+× I(S, g−)−→ (Teich(S))2× (Teich(S))2

and ∆ is the diagonal in (Teich(S))2_{× (Teich(S))}2. We will compute explicitly this
intersection number (see Section 2.7) under particular hypothesis on the curvatures
of g+ and g−: the reason for this being that the transversality condition is in general

difficult to check when the metrics do not have constant curvature. It turns out that in that case the intersection number is 1.

We then start to deform one of the two metrics and check that an intersection persists. Here, one has to be careful that, since the maps are defined on non-compact manifolds, the intersection does not escape to infinity. This is probably the main technical part of the proof and requires results about the convergence of isometric embeddings (Corollary 2.5.5), estimates in anti-de Sitter geometry (Lemma 2.5.12) and results in Teichmüller theory (Lemma 2.5.11). In particular, applying these tools, we prove

Proposition 2.5.13. For every metric g− and for every smooth path of metrics {g+

t }t∈[0,1] on S with curvature less than −1, the set (Φ+× φ−_{g}−)−1(∆) is compact
This guarantees that when deforming one of the two metrics the variation of the
intersection locus is always contained in a compact set. The proof of Theorem 2.3.2
then follows applying standard argument of topological intersection theory.

In Section 2.6, we study the map

p1◦ Φ+: W+→ Teich(S) ,

where p1 : Teich(S)× Teich(S) → Teich(S) is the projection onto the left factor.

The main result we obtain is the following:

Proposition 2.6.1. Let g be a metric on S with curvature less than _{−1 and let h}
be a hyperbolic metric on S. Then there exists a GHMC AdS_{3} manifold M with left
metric isotopic to h containing a past-convex space-like surface isometric to (S, g).
This is proved by showing that p1 ◦ φ+g is proper of degree 1 (mod 2). Again, we

and the general statement then follows since for any couple of metrics g and g0 with
curvature less than _{−1, the maps p}1 ◦ φg and p1 ◦ φg0 are connected by a proper
cobordism.

### 2.2

### Equivariant isometric embeddings

Let S be a connected, compact, oriented surface of genus τ _{≥ 2 and let g be a}
Riemannian metric on S with curvature κ less than _{−1. An isometric equivariant}
embedding of S into AdS_{3} is given by a couple (f, ρ), where f : ˜S _{→ AdS}3 is an

isometric embedding of the universal Riemannian cover of S into AdS3 and ρ is a

representation of the fundamental group of S into PSL(2, R)× PSL(2, R) such that f (γx) = ρ(γ)f (x) ∀ γ ∈ π1(S) ∀ x ∈ ˜S .

The group PSL(2, R)× PSL(2, R) acts on a couple (f, ρ) by post-composition on
the embedding and by conjugation on the representation. We denote by I(S, g)
the set of equivariant isometric embeddings of S into AdS_{3} modulo the action of
PSL(2, R)× PSL(2, R).

Also in an anti-de Sitter setting, an analogue of the Fundamental Theorem for sur-faces in the Euclidean space holds:

Theorem 2.2.1. There exists an isometric embedding of (S, g) into an AdS3

man-ifold if and only if it is possible to define a g-self-adjoint operator b : T S _{→ T S}
satisfying

det(b) =_{−κ − 1} Gauss equation

d∇b = 0 Codazzi equation

Moreover, the operator b determines the isometric embedding uniquely, up to global isometries.

This theorem enables us to identify I(S, g) with the space of solutions of the Gauss-Codazzi equations, which can be studied using the classical techniques of elliptic operators.

Lemma 2.2.2. The space I(S, g) is a manifold of dimension 6τ_{− 6.}

Proof. We can mimic the proof of Lemma 3.1 in [Lab92a]. Consider the sub-bundle Fg ⊂ Sym(T S) over S of symmetric operators b : T S → T S satisfying the Gauss equation. We prove that the operator

2.2. Equivariant isometric embeddings 17

is elliptic of index 6τ _{− 6, equal to the dimension of the kernel of its linearization.}
Let J0 be the complex structure induced by g. For every b∈ Γ∞(Fg), the operator

J = J0b pdet(b)

defines a complex structure on S. In particular we have an isomorphism
F : Γ∞(Fg)_{→ A}

b_{7→} J0b
pdet(b)

between smooth sections of the sub-bundle Γ∞(Fg_{) and the space} _{A of complex}

structures on S, with inverse

F−1 :A → Γ∞(Fg)

J _{7→ −}√_{−κ − 1J}0J .

This allows us to identify the tangent space of Γ(_{F}g) at b with the tangent space of
A at J, which is the vector space of operators ˙J : T S_{→ T S such that ˙}J J + J ˙J = 0.
Under this identification the linearization of d∇ is given by

L( ˙J ) =−J0(d∇J ) .˙

We deduce that L has the same symbol and the same index of the operator∂, sending
quadratic differentials to vector fields. Thus L is elliptic with index 6τ_{− 6.}

To conclude we need to show that its cokernel is empty, or, equivalently, that its adjoint L∗ is injective. If we identify Λ2T S⊗ T S with T S using the metric g, the adjoint operator L∗ is given by (see Lemma 3.1 in [Lab92a] for the computation)

(L∗ψ)(u) =−1

2(∇J0uψ + J∇J0J uψ) .

The kernel of L∗ consists of all the vector fields ψ on S such that for every vector field u

J_{∇}uψ =−∇J0J J0uψ .

We can interpret this equation in terms of intersection of pseudo-holomorphic curves:
the Levi-Civita connection _{∇ induces a decomposition of T (T S) into a vertical V}
and a horizontal H sub-bundle. We endow V with the complex structure J , and
H with the complex structure _{−J}0J J0. In this way, the manifold T S is endowed

Similarly, we obtain the following result:

Lemma 2.2.3. Let_{{g}_{t}_{}}_{t∈[0,1]} be a differentiable curve of metrics with curvature less
than _{−1. The set}

W = [

t∈[0,1]

I(S, gt)

is a manifold with boundary of dimension 6τ _{− 5.}

Proof. Again we can mimic the proof of Lemma 3.2 in [Lab92a]. Consider the
sub-bundle _{F} _{⊂ Sym(T S) over S × [0, 1] of symmetric operators, whose fiber over a}
point (x, t) consists of the operators b : T S _{→ T S, satisfying the Gauss equation}
with respect to the metric g_{t}. The same reasoning as for the previous lemma shows
that

d∇: Γ∞(F )→ Γ∞(Λ2T S_{⊗ T S)}

is Fredholm of index 6τ _{− 5. Since W = (d}∇)−1(0), the result follows from the
implicit function theorem for Fredholm operators.

Let N be a GHMC AdS_{3} manifold endowed with a time orientation, i.e. a nowhere
vanishing time-like vector field. Let S be a convex embedded surface in N . We say
that S is past-convex (resp. future-convex), if its past (resp. future) is geodesically
convex. We will use the convention to compute the shape operator of S using the
future-directed normal. With this choice if S is past-convex (resp. future-convex)
then it has strictly positive (resp. strictly negative) principal curvatures.

Definition 2.2.4. We will denote with I(S, g)+ and I(S, g)− the spaces of equivari-ant isometric embeddings of S as a past-convex and future-convex surface, respec-tively.

### 2.3

### Definition of the maps

### φ

±The parameterisation of GHMC anti-de Sitter structure described in Section 1.5 enables us to formulate our original question about the prescription of the metrics on the boundary of a compact AdS3manifold in terms of existence of an intersection

of particular subsets of Teich(S)_{× Teich(S).}

2.3. Definition of the maps φ± 19

induced on S+and S− respectively, by lifting the embeddings σ±: (S±, g±)→ N to

the Universal cover, we obtain an element of I(S, g+)+ and an element of I(S, g−)−.

Viceversa, if N is a GHMC AdS_{3} manifold, by cutting N along a past-convex
space-like surface and a future-convex space-space-like surface we obtain a convex, compact,
globally hyperbolic anti-de Sitter manifold with convex boundary. Thus, the question
of prescribing the metrics on the boundary components of a compact, convex, globally
hyperbolic anti-de Sitter manifold with strictly convex boundary is equivalent to the
question of finding a future-convex and a past-convex isometric embedding into the
same GHMC AdS_{3} manifold.

This suggests the following construction:

Definition 2.3.1. Let g be a metric on S with curvature κ < _{−1. We define the}
maps

φ±_{g} : I(S, g)±→ Teich(S) × Teich(S)

b7→ (hl(g, b), hr(g, b)) := (g((E + J b)·, (E + Jb)·), g((E − Jb)·, (E − Jb)·))

associating to every isometric embedding of (S, g) the left and right metric of the GHMC AdS3 manifold containing it.

We recall that we use the convention to compute the shape operator using always the future-oriented normal. In this way, the above formulas hold for both future-convex and past-convex surfaces, without changing the orientation of the surface S.

We will prove (in Section 2.7) the following fact, which is the main theorem of the chapter:

Theorem 2.3.2. For every couple of metrics g+ and g− on S with curvature less

than _{−1, we have}

φ+_{g}_{+}(I(S, g+)+)∩ φ−g−(I(S, g−)
−

)6= ∅ .

Therefore, there exists a GHMC AdS_{3} manifold containing a past-convex
space-like surface isometric to (S, g+) and a future-convex space-like surface isometric to

(S, g−). We deduce from this the answer to our original question:

Corollary 2.3.3. For every couple of metrics g_{+} and g− on S with curvature less

than _{−1, there exists a globally hyperbolic convex compact AdS}3 manifold K ∼= S×

[0, 1], whose induced metrics on the boundary are exactly g±.

If we apply the previous corollary to the dual surfaces, we obtain an analogous result about the prescription of the third fundamental form:

Corollary 2.3.4. For every couple of metrics g+ and g− on S with curvature less

### 2.4

### Topological intersection theory

As outlined in Section 2.1, the main tool used in the proof of the main theorem is the intersection theory of smooth maps between manifolds, which is developed for example in [GP74]. We recall here the basic constructions and the fundamental results.

If not otherwise stated, all manifolds considered in this section are non-compact without boundary.

Let X and Z be manifolds of dimension m and n, respectively and let A be a closed
submanifold of Z of codimension k. Suppose that m_{− k ≥ 0. We say that a smooth}
map f : X _{→ Z is transverse to A if for every z ∈ Im(f)∩A and for every x ∈ f}−1(z)
we have

df (TxX) + TzA = TzZ .

Under this hypothesis, f−1(A) is a submanifold of X of codimension k.

When k = m and f−1(A) consists of a finite number of points we define the inter-section number between f and A as

=(f, A) := |f−1(A)_{|} (mod 2) .

Remark 2.4.1. When A is a point p_{∈ Z, f is transverse to p if and only if p is a}
regular value for f . Moreover, if f is proper, f−1(p) consists of a finite number of
points and the above definition coincides with the classical definition of degree (mod
2) of a smooth and proper map.

We say that two smooth maps f : X _{→ Z and g : Y → Z are transverse if the map}
f _{× g : X × Y → Z × Z}

is transverse to the diagonal ∆_{⊂ Z × Z. Notice that if Im(f) ∩ Im(g) = ∅, then f}
and g are transverse by definition.

Suppose now that 2 dim X = 2 dim Y = dim Z. Moreover, suppose that the maps
f : X → Z and g : Y → Z are transverse and the preimage (f × g)−1_{(∆) consists of}

a finite number of points. We define the intersection number between f and g as =(f, g) := =(f × g, ∆) = |(f × g)−1(∆)| (mod 2) .

It follows by the definition that if_{=(f, g) 6= 0 then Im(f) ∩ Im(g) 6= ∅.}

One important feature of the intersection number that we will use further is the
invariance under cobordism. We say that two maps f_{0} : X0 → Z and f1 : X1 → Z

are cobordant if there exists a manifold W and a smooth function F : W _{→ Z such}
that ∂W = X0∪ X1 and F|_{Xi} = fi.

Proposition 2.4.2. Let W be a non-compact manifold with boundary ∂W = X_{0}_{∪X}_{1}.
Let H : W _{→ Z be a smooth map and denote by h}i the restriction of H to the

boundary component Xi for i = 0, 1. Let A ⊂ Z be a closed submanifold. Suppose

2.4. Topological intersection theory 21

(i) codimA = dim Xi;

(ii) H is transverse to A;
(iii) H−1(A) is compact.
Then_{=(h}0, A) ==(h1, A).

Proof. By hypothesis the pre-image H−1(A) is a compact, properly embedded
1-manifold, i.e. it is a finite disjoint union of circles and arcs with ending points on
a boundary component of W . This implies that h−1_{0} (A) and h−1_{1} (A) have the same
parity.

In particular, we deduce the following result about the intersection number of two maps:

Corollary 2.4.3. Let W be a non-compact manifold with boundary ∂W = X0∪ X1.

Let F : W _{→ Z be a smooth map and denote by f}i the restriction of F to the boundary

component X_{i} for i = 0, 1. Let g : Y _{→ Z be a smooth map. Suppose that}
(i) 2 dim Xi = 2 dim Y = dim Z;

(ii) F and g are transverse;
(iii) (F _{× g)}−1(∆) is compact.
Then_{=(f}0, g) ==(f1, g).

Proof. Apply the previous proposition to the map H = F_{× g : W × Y → Z × Z and}
to the submanifold A = ∆, the diagonal of Z_{× Z.}

The hypothesis of transversality in the previous propositions is not restrictive, as it is always possible to perturb the maps involved on a neighbourhood of the set on which transversality fails:

Theorem 2.4.4 (Theorem p.72 [GP74]). Let h : W _{→ Z be a smooth map between}
manifolds, where only W has boundary. Let A be a closed submanifold of Z. Suppose
that h is transverse to A on a closed set C _{⊂ W . Then there exists a smooth map}
˜

h : W _{→ Z homotopic to h such that ˜h is transverse to A and ˜h agrees with h on a}
neighbourhood of C.

Now the question arises whether the intersection number depends on the particular perturbation of the map that we obtain when applying Theorem 2.4.4.

and coincide with h outside the interior part of a compact set B containing h−1(A). Then

=(˜h, A) = =(˜h0, A) .

Proof. Let ˜H : W = X× [0, 1] → Y be an homotopy between ˜h and ˜h0 _{such that}

for every x _{∈ (X \ B) × [0, 1] we have ˜}H(x, t) = h(x) . Notice that ˜H−1(A) is
compact. Up to applying Theorem 2.4.4 to the closed set C = (X_{\ B) × [0, 1] ∪ ∂W ,}
we can suppose that ˜H is transverse to A. By Proposition 2.4.2, we have that
=(˜h, A) = =(˜h0, A) as claimed.

Moreover, in particular circumstances, we can actually obtain a 1_{−1 correspondence}
between the points of h−1_{0} (A) and h−1_{1} (A). The following proposition will not be used
for the proof of the main result of the chapter, but it might be a useful tool to prove
the uniqueness part of the question addressed in this chapter, as explained in Remark
2.7.3.

Proposition 2.4.6. Under the same hypothesis as Proposition 2.4.2, suppose that the cobordism (W, H) between h0 and h1 satisfies the following additional properties:

(i) W fibers over the interval [0, 1] with fiber X_{t};

(ii) the restriction ht of H at each fiber is tranverse to A .

Then_{|h}−1_{0} (A)_{| = |h}−1_{1} (A)_{|.}

Proof. It is sufficient to show that in H−1(A) there are no arcs with ending points in
the same boundary component. By contradiction, let γ be an arc with ending point
in X_{0}. Define

t0 = sup{t ∈ [0, 1] | γ ∩ Xt6= ∅} .

A tangent vector ˙γ at a point p_{∈ X}_{t}_{0}_{∩ γ is in the kernel of the map}
dpH : T(p,t0)W → Tq(Z× Z)/Tq(A) ,

where q = H(p). The contradiction follows by noticing that on the one hand ˙γ
is contained in the tangent space T_{p}Xt0 by construction but on the other hand
dpht0 : TpXt0 → Tq(Z× Z)/Tq(A) is an isomorphism by transversality.

A similar reasoning works when γ has ending points in X1.

### 2.5

### Some properties of the maps

### φ

±2.5. Some properties of the maps φ± 23

AdS3manifold M the class in Teichmüller space of the left and right metrics of M . It

follows easily from Lemma 2.2.3 that for any couple of metrics g and g0with curvature
less than_{−1 the maps φ}±_{g} and φ±_{g}0 are cobordant through a map Φ±. In this section
we will define the maps Φ± and will study some of its properties, which will enable
us to apply the topological intersection theory described in the previous section.
More precisely, the first step will consist of proving that all the maps involved are
smooth. This is the content of Proposition 2.5.1 and the proof will rely on the fact
that the holonomy representation of a hyperbolic metric depends smoothly on the
metric. Then we will deal with the properness of the maps Φ±(Corollary 2.5.8) that
will follow from a compacteness result of isometric embeddings (Corollary 2.5.5).
This will allow us also to have a control on the space where two maps φg and φg0
intersect: when we deform one of the two metrics the intersection remains contained
in a compact set (Proposition 2.5.13).

Recall that given a smooth path of metrics_{{g}_{t}_{}}_{t∈[0,1]} on S with curvature less than
−1, the set

W± = [

t∈[0,1]

I±(S, gt)

is a manifold with boundary ∂W± = I(S, g0)± ∪ I(S, g1)± of dimension 6τ − 5

(Lemma 2.2.3). We define the maps

Φ± : W± → Teich(S) × Teich(S)

bt7→ (hl(gt, bt), hr(gt, bt)) := (gt((E + J bt)·, (E + Jbt)·), gt((E− Jbt)·, (E − Jbt)·))

associating to an equivariant isometric embedding (identified with its Codazzi
op-erator b_{t}) of (S, g_{t}) into AdS3 the class in Teichmüller space of the left and right

metrics of the GHMC AdS3 manifold containing it. We remark that the restrictions

of Φ± to the boundary coincide with the maps φ±_{g}
0 and φ

±

g1 defined in Section 2.3. We deal first with the regularity of the maps.

Proposition 2.5.1. The functions Φ± : W± _{→ Teich(S) × Teich(S) are smooth.}
Proof. Let_{M}_{S} be the set of hyperbolic metrics on S. We can factorise the map Φ±
as follows:
W± Φ
0±
−−→ MS× MS
π
−→ Teich(S) × Teich(S)

where Φ0± associates to an isometric embedding of (S, gt) (determined by an

op-erator b_{t} satisfying the Gauss-Codazzi equation) the couple of hyperbolic metrics
(gt((E + Jtbt)·, (E +Jtbt)·), gt((E−Jtbt)·, (E −Jtbt)·)), and π is the projection to the

hyperbolic plane and we fix a unitary frame_{{w}1, w2} of T0H2. We can realise every
element of the fundamental group of S as a closed path passing through p. Let
γ be a path passing through p and let _{{U}i}i=0,...n be a finite covering of γ such

that every Ui is homeomorphic to a ball. We know that there exists a unique map

f0: U0→ B0⊂ H2 such that
f0(p) = 0
dpf0(vi) = wi
f_{0}∗g_{H}2 = h .

Then, for every i _{≥ 1 there exists a unique isometry f}_{i} : Ui → Bi ⊂ H2 which

coincides with fi−1on the intersection Ui∩Ui−1. Let q = fn(p)∈ H2. The holonomy

representation sends the homotopy class of the path γ to the isometry Iq: H2→ H2

such that I_{q}(q) = 0. Moreover, its differential maps the frame _{{u}i = dfn(vi)} to

the frame w_{i}. The isometry I_{q} depends smoothly on q and on the frame u_{i}, which
depend smoothly on the metric because each fi does.

The next step is about the properness of the maps Φ±. This will involve the study of sequences of isometric embeddings of a disc into a simply-connected spacetime, which have been extensively and profitably analysed in [Sch96]. In particular, the author proved that, under reasonable hypothesis, a sequence of isometric embeddings of a disc into a simply-connected spacetime has only two possible behaviours: it converges C∞, up to subsequences, to an isometric embedding, or it is degenerate in a precise sense:

Theorem 2.5.2 (Theorem 5.6 [Sch96]). Let ˜fn: D→ X be a sequence of uniformly

elliptic1 immersions of a disc D in a simply connected Lorentzian spacetime (X, ˜g).
Assume that the metrics ˜f_{n}∗g converge C˜ ∞ towards a Riemannian metric ˜g∞ on

D and that there exists a point x ∈ D such that the sequence of the 1-jets j1_{f}˜_{n}_{(x)}
converges. If the sequence ˜fndoes not converge in the C∞topology in a neighbourhood

of x, then there exists a maximal geodesic γ of (D, ˜g∞) and a geodesic arc Γ of (X, ˜g)

such that the sequence ( ˜fn)|γ converges towards an isometry ˜f∞: γ→ Γ.

We start with a straightforward application of the Maximum Principle, which we recall here in the form useful for our purposes (see e.g. [BBZ11, Proposition 4.6]). Proposition 2.5.3 (Maximum Principle). Let Σ1 and Σ2 two future-convex

space-like surfaces embedded in a GHMC AdS_{3} manifold M . If they intersect in a point
x and Σ1 is in the future of Σ2 then the product of the principal curvatures of Σ2 is

smaller than the product of the principal curvatures of Σ1.

Proposition 2.5.4. Let Σ be a future-convex space-like surface embedded into a
GHMC AdS_{3} manifold M . Suppose that the Gaussian curvature of Σ is bounded

1