Journal of Geometry and Physics 128 (2018) 99–117

Contents lists available at ScienceDirect

Journal of Geometry and Physics

journal homepage: www.elsevier.com/locate/geomphys

Constant scalar curvature hypersurfaces in

(3 + 1)-dimensional GHMC Minkowski spacetimes
Graham Smith
Instituto de Matemática, UFRJ, Av. Athos da Silveira Ramos 149, Centro de Tecnologia - Bloco C, Cidade Universitária - Ilha do Fundão,
Caixa Postal 68530, 21941-909, Rio de Janeiro, RJ, Brazil

article info a b s t r a c t
Article history: We prove that every (3 + 1)-dimensional flat GHMC Minkowski spacetime which is not
Received 4 April 2017 a translation spacetime or a Misner spacetime carries a unique foliation by spacelike
Received in revised form 19 September hypersurfaces of constant scalar curvature. In other words, we prove that every such
2017
spacetime carries a unique time function with isochrones of constant scalar curvature.
Accepted 10 November 2017
Furthermore, this time function is a smooth submersion.
Available online 6 December 2017
© 2017 Elsevier B.V. All rights reserved.
MSC:
53C21
53C50
53C42
52A20
35J60

Keywords:
Minkowski spacetimes
GHMC
Scalar curvature

1. Introduction

1.1. Introduction

Let Rd,1 denote (d + 1)-dimensional Minkowski space, that is Rd+1 furnished with the semi-riemannian metric

ds2 := dx21 + · · · + dx2d − dx2d+1 .

For the purposes of this paper, a (d + 1)-dimensional Minkowski spacetime is a semi-riemannian manifold X which is
everywhere locally isometric to Rd,1 . Minkowski spacetimes are of interest in cosmology as the simplest possible solutions
of Einstein’s equations. However, since the pioneering work [1] of Mess, they have also found deep and broad applications
in the study of Teichmüller theory and its higher-dimensional analogous.
In this paper, we will be concerned with time functions defined over certain types of Minkowski spacetimes known as
GHMC Minkowski spacetimes (see below). Here, a time function is defined to be a real-valued submersion all of whose level
sets are spacelike. It is often possible to construct time functions which have interesting geometric properties and these,
in turn, can be useful in studying the physics or the geometry of the ambient spacetime. For example, in [2], Andersson,
Barbot, Béguin & Zeghib study smooth time-functions over constant curvature1 GHMC spacetimes with level sets of constant

E-mail address: [email protected].

1 That is, de-Sitter, Minkowski and anti de-Sitter.

https://doi.org/10.1016/j.geomphys.2017.11.006
0393-0440/© 2017 Elsevier B.V. All rights reserved.
100 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

mean curvature, thus proving the existence over such spacetimes of so-called ‘‘York’’ time functions, which are known to
be of considerable use in general relativity (c.f. [3]). Likewise, in [4] and [5], Barbot, Béguin & Zeghib construct smooth time
functions over (2 + 1)-dimensional GHMC Minkowski and anti de-Sitter spacetimes with level sets of constant extrinsic
curvature, and in [6] and [7], Bonsante, Mondello & Schlenker use these time functions to provide fascinating new insights
into the earthquake and grafting maps of classical Teichmüller theory. We refer the interested reader to our review [8] for
more details of these and other related constructions.
In contrast to the above mentioned results, in [9], Bonsante & Fillastre show that smooth time functions with level sets
of constant extrinsic curvature do not always exist for higher dimensional ambient spacetimes. This can be understood as
a manifestation of the subtle manner in which constant extrinsic curvature is actually a uniformly elliptic condition for
surfaces, but is no longer so in higher dimensions. For this reason, in [10], we introduced the so-called ‘‘special lagrangian
curvature’’ which, by involving the Calabi–Yau structure of the tangent bundle of the ambient space, yields an alternative
generalisation of extrinsic curvature which continues to be uniformly elliptic even in higher dimensions. In particular, when
the ambient spacetime is (3 + 1)-dimensional, the ‘‘special lagrangian curvature’’ of any spacelike hypersurface is none other
than its scalar curvature which, we recall, is defined over a riemannian manifold M by
−1
S := g ik Rijk j ,
m(m − 1)
where m here denotes the dimension of M, g denotes its metric, R denotes its Riemann curvature tensor, and the summation
convention is implied.2 Using this curvature notion, in [10], we partially complemented the work [2] of Andersson, Barbot,
Béguin & Zeghib by constructing smooth time functions over open subsets of GHMC de-Sitter spacetimes with level sets of
constant scalar curvature. In the present work, we extend this result to the Minkowski case, that is, we construct smooth
time functions over (3 + 1)-dimensional GHMC Minkowski spacetimes with level sets of constant scalar curvature.
Before stating the main result of this paper, it is necessary to properly introduce the class of GHMC Minkowski spacetimes.
Although this class arises from fairly natural physical considerations, it is nonetheless rather unfamiliar to most geometers,
and its definition therefore requires a brief detour. We begin with the concept of causality. A tangent vector of X is said to be
spacelike, timelike or null according to whether its norm-squared is positive, negative or null respectively, and is said to be
causal whenever it is either timelike or null. A continuously differentiable, embedded curve in X is said to be causal whenever
all of its tangent vectors are causal. Finally, the spacetime X is itself said to be causal whenever it contains no closed causal
curve. This condition reflects the physical requirement that it is not possible to move into the past by travelling into the
future.
We henceforth suppose that X is oriented, time-oriented and causal. The past of any given point x of X is then defined to
be the closed set of all initial points of future-oriented, causal curves terminating in x. Likewise, the future of that point is
defined to be the closed set of all terminal points of future-oriented, causal curves starting at x. The spacetime X is said to be
globally hyperbolic whenever the intersection of the past of any point with the future of any other point is compact (c.f. [11]
and [12]).
A Cauchy hypersurface in X is a spacelike hypersurface Σ which intersects every inextensible causal curve exactly
once. In [11] and [12], Bernal & Sanchez show that an oriented and time-oriented spacetime is globally hyperbolic if and
only if it contains a smooth Cauchy hypersurface. Although the Cauchy hypersurface is trivially not unique, all smooth
Cauchy hypersurfaces of a given globally hyperbolic spacetime are diffeomorphic to one another, and the spacetime itself is
diffeomorphic to the Cartesian product of any such hypersurface with R. A globally hyperbolic spacetime is said to be Cauchy
compact whenever its Cauchy hypersurface is compact.
Finally, a globally hyperbolic spacetime X is said to be maximal whenever there exists no isometric embedding e of X into
a strictly larger spacetime X̃ such that the image under e of a Cauchy hypersurface in X is also a Cauchy hypersurface in X̃ .
In order to understand this property, consider the future cone in Rd,1 ,

Cd,1 := x ∈ Rd,1 | ∥x∥2 < 0, xd+1 > 0 .

{ }

This space is globally hyperbolic and maximal, even though it isometrically embeds into the strictly larger space Rd,1 . Indeed,
its Cauchy hypersurface is given by

Σ d := x ∈ Rd,1 | ∥x∥2 = −1, xd+1 > 0 ,

{ }

which is not a Cauchy hypersurface of Rd,1 .

We say that an oriented and time oriented spacetime is GHMC whenever it is Globally Hyperbolic, Maximal and Cauchy-
compact. Building on the work [1] of Mess, a classification of all GHMC Minkowski spacetimes was initiated by Bonsante
in [13] and completed by Barbot in [14]. Within this classification, two exceptional families stand out, namely the translation
spacetimes and the Misner spacetimes. A GHMC translation spacetime X is one whose universal cover is the whole of Rd,1 . Up
to a finite cover, such a spacetime has the form

X = T d × R,

2 The normalisation is chosen here so that the scalar curvature of the unit sphere in Euclidean space is equal to 1.
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 101

furnished with the metric

g = dx21 + · · · + dx2d − dx2d+1 ,

where T d is the quotient of Rd by some cocompact lattice Λ, and it carries the foliation (T d × {t })t ∈R , all of whose leaves have
vanishing scalar curvature. A GHMC Misner spacetime Y is, up to reversal of the temporal orientation, one whose universal
cover is Rd−1 × C 1,1 , where C 1,1 is the future cone in R1,1 . Up to a finite cover, every such spacetime has the form

Y = T d × ]0, ∞[,

furnished with the metric

g = dx21 + · · · + dx2d−1 + t 2 dx2d − dx2d+1 ,

where T d is again the quotient of Rd by some cocompact lattice Λ. However, in this case, it is worth observing that the
projection of the xd -axis need not necessarily yield a closed curve in T d . Such a spacetime carries the foliation (T d × {t })t >0 ,
all of whose leaves also have vanishing scalar curvature.
In both of the above cases, it follows from the geometric maximum principle that there exist no immersed spacelike
hypersurfaces of constant non-zero scalar curvature. For all other GHMC Minkowski spacetimes, we have

Theorem 1.1.1. Let X be a (3 + 1)-dimensional GHMC Minkowski spacetime which is not a translation spacetime or a Misner
spacetime. There exists a unique smooth submersion T : X → ]0, ∞[ such that, for all t ∈ ]0, ∞[, the level set T −1 ({t }) is a
convex Cauchy hypersurface of constant scalar curvature equal to (−t 2 ).

Remark. By the classification [14] of Barbot, every GHMC Minkowski spacetime which is not a translation spacetime or a
Misner spacetime is, up to reversal of the temporal orientation and up to a finite cover, a twisted product of what we choose
to call a ‘‘Bonsante spacetime’’ with a Euclidean torus (c.f. Section 3.1). Theorem 1.1.1 is then a straightforward consequence
of the corresponding result for Bonsante spacetimes, and thus follows immediately from Theorem 3.4.1.

Remark. In fact, an analogous results holds in all dimensions for submersions with level hypersurfaces of constant ‘‘special
lagrangian curvature’’. In particular, when the ambient space is (2 + 1)-dimensional, we recover the result [5] of Barbot,
Béguin & Zeghib. However, in higher dimensions, the ‘‘special lagrangian curvature’’ no longer has an elementary expression
in terms of classical curvature notions and, for this reason, we shall not discuss this further here.

2. Special legendrian geometry

2.1. Real special lagrangian structures

The proof of Theorem 1.1.1 is based on a compactness result for families of hypersurfaces of constant scalar curvature. This
result in turn depends on the special legendrian structure of the bundle of future-oriented, unit, timelike tangent vectors over
Rd,1 . In order to understand this structure, we first consider the fully integrable case. Thus, denote by ⟨·, ·⟩ the Minkowski
metric over Rd,1 , that is

⟨X , Y ⟩ = X 1 Y 1 + · · · + X d Y d − X d+1 Y d+1 ,
and consider the following structures defined over the cartesian product Rd,1 × Rd,1 .

ω((Xr , Xi ), (Yr , Yi )) := ⟨Xr , Yi ⟩ − ⟨Yr , Xi ⟩,

J(Xr , Xi ) := (−Xi , Xr ), and
R(Xr , Xi ) := (Xr , −Xi ).
Observe that ω is antisymmetric and non-degenerate,

J2 = −Id, and
R2 = Id.
The form ω is the symplectic structure, and the maps J and R are, respectively, the complex and real structures. They are
related to one another by

ω(J·, J·) = ω(·, ·),

ω(R·, R·) = −ω(·, ·), and
JR + RJ = 0.
We call the structure defined by the triple (ω, J, R) a real special lagrangian structure. Its symmetry group is given by the
action of O(d, 1) on Rd,1 × Rd,1 which, for any matrix M, maps the pair (Xr , Xi ) to the pair (MXr , MXi ).
102 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

The real special lagrangian structure yields various auxiliary structures which play a key role in the sequel. First, there is
a semi-riemannian metric of signature (2d, 2), given by
g((Xr , Xi ), (Yr , Yi )) := ω((Xr , Xi ), J(Yr , Yi )) = ⟨Xr , Yr ⟩ + ⟨Xi , Yi ⟩.
Next, there is another semi-riemannian metric, this time of signature (d + 1, d + 1), given by
m((Xr , Xi ), (Yr , Yi )) := −ω((Xr , Xi ), R(Yr , Yi )) = ⟨Xr , Yi ⟩ + ⟨Yr , Xi ⟩.
In particular, over every lagrangian subspace of Rd,1 × Rd,1 , m satisfies
m((Xr , Xi ), (Yr , Yi )) := 2⟨Xr , Yi ⟩ = 2⟨Yr , Xi ⟩.
Finally, there is, up to a choice of sign, a unique complex (d + 1)-form (complex with respect to the structure J) whose
restriction to the real subspace Rd,1 × {0} coincides with the volume form of the metric g. This form, which we denote by
Ω̂ is also described explicitly as follows. Let e1 , . . . , ed+1 be an orthonormal basis3 of Rd,1 and, for all k, denote fk := Jek . Let
e1 , . . . , ed+1 , f 1 , . . . , f d+1 be the basis dual to e1 , . . . , ed+1 , f1 , . . . , fd+1 and, for all k, denote
dz k := ek + if k .
It follows from the definitions that
Ω̂ = ±dz 1 ∧ · · · ∧ dz d+1 .
In particular, Ω̂ is independent of the orthonormal basis of Rd,1 chosen.

Lemma 2.1.1. (1) The restriction of g to any lagrangian subspace of Rd,1 × Rd,1 has signature (d, 1).
(2) If e1 , . . . , ed+1 is an orthonormal basis of some lagrangian subspace of Rd,1 × Rd,1 , then
⏐ ⏐
⏐Ω̂ (e1 , . . . , ed+1 )⏐ = 1.
⏐ ⏐

Proof. Let E be a lagrangian subspace of Rd,1 × Rd,1 and denote F := JE. Observe that E and F are mutually orthogonal with
respect to g. It follows that Rd,1 × Rd,1 = E ⊕ F and that the restrictions of g to E and F are non-degenerate. Furthermore,
since J restricts to an isometry from E to F with respect to g, the restrictions of g to these two subspaces both have the same
signature. In particular, if this signature is equal to (p, q) then, by orthogonality of E and F again, (2p, 2q) = (2d, 2), and (1)
follows.
Now let A be any isometry sending E into the real subspace Rd,1 × {0}. This map extends to a unique unitary map of
R × Rd,1 to itself. In particular,
d,1
⏐ ⏐ ⏐ ⏐
⏐Ω̂ (e1 , . . . , ed+1 )⏐ = ⏐Ω̂ (Ae1 , . . . , Aed+1 )⏐ .
⏐ ⏐ ⏐ ⏐

However, if e1 , . . . , ed+1 is an orthonormal basis of E, then Ae1 , . . . , Aed+1 is an orthonormal basis of Rd,1 × {0} so that, by
definition,
Ω̂ (Ae1 , . . . , Aed+1 ) = ±1,
and (2) follows. □

Corollary 2.1.2. If L is a lagrangian subspace over which

( )
Im eiθ Ω̂ =0
for some θ then, for any orthonormal basis e1 , . . . , ed+1 of L,
( )
Re eiθ Ω̂ (e1 , . . . , ed+1 ) = ±1.

2.2. The real special legendrian structure

Let U+ Rd,1 denote the bundle of future-oriented, unit, timelike tangent vectors over Rd,1 , let M := M d denote its total
space, that is
M := M d := U+ Rd,1 = (x, y) | ∥y∥2 = −1, yd+1 > 0 .
{ }

The tangent space to M at the point (x, y) is given by

T(x,y) M = {(Xr , Xi ) | ⟨Xi , y⟩ = 0} .

3 Since the metric of Rd,1 has signature (d, 1), we take this to mean that e , . . . , e are spatial and e
1 d d+1 is temporal.
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 103

Observe now that the symplectic form ω defined in Section 2.1 is the exterior derivative of the Liouville form

λ(x,y) (Xr , Xi ) := ⟨Xr , y⟩.

The restriction of this form to TM makes M into a contact manifold. We denote the resulting contact bundle by α . Its fibre at
the point (x, y) is given by

α(x,y) := {(Xr , Xi ) | ⟨Xr , y⟩ = ⟨Xi , y⟩ = 0} .

Since the maps J and R both preserve α they define, together with ω := dλ, a real special lagrangian structure over every
fibre of the bundle. We say that the manifold M carries a real special legendrian structure. Of particular interest to us will be
the real and imaginary subspaces of α(x,y) given by

R(x,y) := {X | X = RX } = {(X , 0) | ⟨X , y⟩ = 0} , and

(2.1)
I(x,y) := {X | X = −RX } = {(0, X ) | ⟨X , y⟩ = 0} .
From the point of view of the unitary bundle, R(x,y) and I(x,y) are respectively the horizontal and vertical subspaces of α(x,y) .
Furthermore, each of these spaces naturally identifies with the orthogonal complement of y in Rd,1 and, in particular, they
each identify with one another.
The forms g and m are defined over each fibre of α as before. Significantly, over each fibre, g now defines a riemannian
metric, and m defines a metric of signature (d, d). Let Ω denote the unique (up to choice of sign) complex d-form which
coincides with the volume form of the metric g over the subspace

R(x,y) = {(X , 0) | ⟨X , y⟩ = 0} .

It is straightforward to show that at every point (x, y) of M,

Ω(x,y) = ±c(y,0) Ω̂ ,
where c(y,0) here denotes the operator of contraction by the vector (y, 0).
Let D denote the restriction to M of the standard differentiation operator over Rd,1 × Rd,1 . Let ∇ denote its composition
with orthogonal projection onto the distribution α . Let X := (Xr , Xi ) and Y := (Yr , Yi ) be tangent vector fields over M taking
values in α . Recalling that (y, 0) and (0, y) are temporal vectors, and therefore have negative norm-squared, we have

∇ X Y = DX Y + g ((y, 0), DX Y ) (y, 0) + g ((0, y), DX Y ) (0, y)

= DX Y − g (Y , DX (y, 0)) (y, 0) − g (Y , DX (0, y)) (0, y)
= DX Y − ⟨Yr , Xi ⟩(y, 0) − ⟨Yi , Xi ⟩(0, y).
The shape operator of α is defined to be the difference between D and ∇ , that is

AX Y := DX Y − ∇ X Y = ⟨Yr , Xi ⟩(y, 0) + ⟨Yi , Xi ⟩(0, y).

The curvature of the distribution α is then given by

g RXY Z , W = g AX W , AY Z − g AX Z , AY W
( ) ( ) ( )

= −⟨Wr , Xi ⟩⟨Zr , Yi ⟩ − ⟨Wi , Xi ⟩⟨Zi , Yi ⟩ (2.2)

+ ⟨Zr , Xi ⟩⟨Wr , Yi ⟩ + ⟨Zi , Xi ⟩⟨Wi , Yi ⟩,
where X , Y , Z and W are vectors in α . This formula will play a key role in the computations that follow.
Finally, straightforward computations show that for all vectors X in α ,

(∇ J)X = (∇ R)X = 0,

for all vectors X and Y in α ,

(∇ g)(X , Y ) = (∇ω)(X , Y ) = (∇ m)(X , Y ) = 0,

and for all vectors X , Y1 , . . . , Yd in α ,

(∇ X Ω )(Y1 , . . . , Yd ) = 0.

2.3. Special legendrian submanifolds

Let Σ̂ ⊆ M be a d-dimensional embedded submanifold. We say that Σ̂ is legendrian whenever its tangent space is always
contained in α and

ω|T Σ̂ = 0.
104 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

Let ∇ be the Levi-Civita covariant derivative of Σ̂ . The shape operator  of Σ̂ is defined by

ÂX Y := ∇ X Y − ∇X Y ,

where X and Y are tangent vector fields over Σ̂ . Observe, in particular, that ÂX Y is always normal to Σ̂ . The second
fundamental form of Σ̂ is then given by

II(X , Y , Z ) := g(ÂX Y , JZ ) = −ω(ÂX Y , Z ).

In particular, since Σ̂ is legendrian, II is symmetric in all three variables. Furthermore, we have the following Codazzi–
Mainardi equations.

Lemma 2.3.1. For all vectors X , Y , Z and W tangent to Σ̂ ,

(∇W II)(X , Y , Z ) = (∇X II)(W , Y , Z ) + g(RWX Y , JZ ). (2.3)

Proof. Indeed, consider a point p ∈ Σ̂ and tangent vector fields X , Y , Z and W which are all parallel at this point. Then

(∇W II)(X , Y , Z ) = W II(X , Y , Z )

= −W ω(∇ X Y , Z )
= −ω(∇ W ∇ X Y , Z ) − ω(∇ X Y , ∇ W Z ).

Since both Y and Z are parallel at p, both ∇ X Y and ∇ W Z are elements of the lagrangian subspace JTp Σ̂ , and the second term
on the right hand side therefore vanishes. Since both X and W are both parallel at p, their commutator [W , X ] vanishes at
this point, and the result now follows by symmetry. □

For θ ∈ R, we say that the legendrian submanifold Σ̂ is θ -special legendrian whenever

Im(eiθ Ω )|T Σ̂ = 0,

and we call θ the special legendrian angle of Σ̂ . Observe, in particular, that if T Σ̂ is the graph of the linear map B : R → I ,
then Σ̂ is θ -special lagrangian if and only if

Im(eiθ Det(Id + iB)) = 0. (2.4)

We will make good use of this property in the sequel.

Lemma 2.3.2. If Σ̂ is θ -special legendrian for some θ then, for any orthonormal basis e1 , . . . , ed of T Σ̂ ,
d
∑
II(·, ea , ea ) = 0.
a=1

Remark. Observe that this is not the same as minimality of Σ̂ , since it is still possible for the mean curvature vector of this
submanifold to be non-trivial in the direction of the Reeb vector field of M.

Proof. Indeed, consider a point p in Σ̂ . Let e1 , . . . , ed be an orthonormal basis of Tp Σ̂ and extend this to a frame of Σ̂ which
is parallel at p. For any other tangent vector X to Σ̂ at p,
d
∑
X Ω (e1 , . . . , ed ) = (∇ X Ω )(e1 , . . . , ed ) + Ω (e1 , . . . , ∇ X ea , . . . , ed )
a=1
d
∑
= Ω (e1 , . . . , ÂX ea , . . . , ed ).
a=1

Since ÂX ea is normal to Tp Σ̂ for all a, it follows that

d
∑
X Ω (e1 , . . . , ed ) = Ω (e1 , . . . , g(ÂX ea , Jeb )Jeb , . . . , ed )
a,b=1
( d
)
∑
=i II(X , ea , ea ) Ω (e1 , . . . , ed ),
a=1
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 105

where the second equality follows since Ω is C-multilinear. Taking the imaginary parts now yields
( d )
iθ
II(X , ea , ea ) Re(eiθ Ω (e1 , . . . , ed ))
∑
X Im(e Ω (e1 , . . . , ed )) =
a=1
d
∑
=± II(X , ea , ea ),
a=1

where the second equality here follows by Corollary 2.1.2. The result follows. □

2.4. Positive special legendrian submanifolds

We say that a special legendrian submanifold Σ̂ is positive whenever

m|T Σ̂ ≥ 0.

Although this may seem like a strong restriction, positive special legendrian submanifolds arise naturally as lifts of certain
hypersurfaces of Rd,1 . As such, they will play a key role in our study of constant scalar curvature hypersurfaces in (3 + 1)-
dimensional Minkowski space. Indeed, recall that, by the Gauss–Codazzi equations, the scalar curvature of an embedded
hypersurface in R3,1 is given by
1
S := − (λ1 λ2 + λ1 λ3 + λ2 λ3 ),
3
where λ1 , λ2 and λ3 its principal curvatures. We now have

Lemma 2.4.1. Let Σ be an embedded spacelike hypersurface in R3,1 , let N : Σ → H3 be its future-oriented, unit, normal vector
field and define the embedded hypersurface Σ̂ by

Σ̂ := {(x, N(x)) | x ∈ Σ } .
(1) Σ̂ is legendrian;
(2) if Σ has constant scalar curvature equal to (−1/3), then Σ̂ is π/2-special legendrian;
(3) if Σ is complete, then so too is Σ̂ ; and
(4) if Σ is locally strictly convex, then Σ̂ is positive.

Remark. In fact, an analogous result holds in all dimensions. For example, when d = 2, special legendrian submanifolds
correspond to surfaces of constant extrinsic curvature. However, in higher dimensions, the relevant notion of curvature is
more involved, and we refer the reader to [15] for details.

Proof. Consider a point x of Σ and denote y := N(x). Let R and I denote respectively the real and imaginary subspaces
of α(x,y) as defined in Eq. (2.1). Recall that both R and I identify with the orthogonal complement of y in R3,1 , that is, the
tangent space to Σ at x. In particular, via this identification, Tp Σ̂ itself identifies with the graph of the shape operator of Σ
at x, which we now denote by A. For X and Y tangent to Σ , we now have

ω((X , AX ), (Y , AY )) = ⟨X , AY ⟩ − ⟨AX , Y ⟩ = 0,
so that Σ̂ is legendrian. This proves (1). Next, we have

Det(Id + iA) = 1 + iTr(A) − σ2 (A) − iDet(A),

so that Im(eiπ /2 Det(Id + iA)) vanishes whenever σ2 (A) = 1. This proves (2). The metric over Σ̂ is given by

g ((X , AX ), (Y , AY )) = ⟨X , Y ⟩ + ⟨AX , AY ⟩,

so that Σ̂ is complete whenever Σ is. This proves (3). Finally,

m((X , AX ), (Y , AY )) = 2⟨X , AY ⟩,

so that m is positive definite whenever A is positive definite. This proves (4). □

Significantly, the hypothesis of positivity makes it straightforward to prove strong compactness results for special
legendrian submanifolds. In order to state these results, we first require some terminology. Thus, a pointed, embedded
submanifold is defined to be a pair (Σ̂ , p) where Σ̂ is an embedded submanifold and p is a point of Σ̂ . A sequence (Σ̂m , pm )
of complete, pointed, embedded submanifolds is said to converge towards the complete, pointed, embedded submanifold
(Σ̂∞ , p∞ ) whenever there exists a sequence (φm ) of smooth maps from Σ̂∞ into M with the following properties.
106 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

(1) φm (p∞ ) = pm for all m; and

for every compact subset K of Σ̂∞ , there exists m0 ≥ 0 such that
(2) for all m ≥ m0 , φm defines an embedding over K whose image is contained in Σ̂m ; and
(3) the subsequence (φm )m≥m0 converges in the C ∞ sense over K to the identity map.
In [15], we prove the following compactness theorem.

Theorem 2.4.2. Let (θm ) be a sequence of real numbers converging to θ∞ . For all m, let (Σ̂m , pm ) be a complete, pointed, positive,
θm -special legendrian submanifold of M. If there exists a compact subset K of M such that pm ∈ K for all m, then there exists a
complete, pointed, positive, θ∞ -special legendrian submanifold (Σ̂∞ , p∞ ) towards which the sequence (Σ̂m , pm ) subconverges in
the sense described above.

2.5. Degenerate submanifolds

Recall now that, whereas the compactness result of Theorem 2.4.2 concerns special legendrian submanifolds of M, what
we actually require is a compactness result for constant scalar curvature, spacelike hypersurfaces in Rd,1 . Bearing in mind
Lemma 2.4.1, such a compactness result will follow once we have identified all positive special legendrian submanifolds of
M which do not project down to constant scalar curvature, spacelike hypersurfaces in Rd,1 . However, these are precisely the
positive special legendrian submanifolds over which the restriction of m is degenerate at some point. We now study how
this property affects the global structure of such submanifolds.
Consider first a point p ∈ M, and a lagrangian subspace E of the fibre αp . We say that X ∈ E is an eigenvector of m over E
with eigenvalue λ whenever
m(X , Y ) = λg(X , Y )
for all other Y in E. In particular, when 0 is an eigenvalue of m over E, we define the nullity of m over E to be the multiplicity of
this eigenvalue, and we define it to be equal to 0 otherwise. Consider now a positive, special legendrian submanifold Σ̂ of M
so that, in particular, Tp Σ̂ is a lagrangian subspace of the fibre αp at every point p ∈ Σ̂ . In this and the following section, we
will show that, in the case of interest to us, the nullity of m is constant over Σ̂ . To this end, for all p ∈ M, for any lagrangian
subspace E of αp , and for all 1 ≤ k ≤ d, define
k
∑
φ k (E) := Inf m(Xa , Xa ),
(X1 ,...,Xk )
a=1

where (X1 , . . . , Xk ) ranges over all orthonormal k-tuples of vectors in E. By abuse of notation, for all 1 ≤ k ≤ d, define also
φ k : Σ̂ → [0, ∞[ by
φ k (p) := φ k (Tp Σ ).
The following lemma will prove useful.

Lemma 2.5.1. Let p be a point in M and let E be a lagrangian subspace of the fibre αp . There exists an orthonormal basis of E
which simultaneously diagonalises both m(·, ·) and m(·, J ·).

Proof. Let A, B : E → E be the unique linear maps such that, for all X , Y ∈ E,

m(X , Y ) = g(X , AY ), and

m(X , JY ) = g(X , BY ).
It suffices to show that A and B commute. Observe first that these linear maps are given by

A = π JR, and
B = π JRJ,
where π : αp → E here denotes the orthogonal projection with respect to g. Consequently,

[A, B] = π JRπ JRJ − π JRJπ JR

= π RJπ RJ2 + π RJ2 π JR
= −π R(Jπ + π J)R.
However, since E is lagrangian, its orthogonal complement in αp with respect to g is given by

E ⊥ = JE ,
so that
Jπ + π J = J,
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 107

and

[A, B] = −π RJR = π R2 J = π J = 0,

as desired. □

Lemma 2.5.2. If Σ̂ is positive then, for all 1 ≤ k ≤ d, there exists a continuous function f k over Σ̂ such that
k
∑
∆φ k + f k φ k ≤ −Tr(m) ⟨Xa,i , Xa,i ⟩m(Xa , JXa ) (2.5)
a=1

in the viscosity sense, where (X1 , . . . , Xk ) is any orthonormal k-tuple of vectors in E realising φ k and, for each a, Xa =: (Xa,r , Xa,i ).

Proof. Choose p ∈ Σ̂ . Bearing in mind Lemma 2.5.1, let e1 , . . . , ed be an orthonormal basis of joint eigenvectors of m(·, ·)
and m(·, J·) over Tp Σ̂ , chosen in such a manner that m(e1 , e1 ) ≤ · · · ≤ m(ed , ed ). We extend e1 , . . . , ed to an orthonormal
frame field over Σ̂ in a neighbourhood of p which is parallel along geodesics leaving p. In particular,
k
∑
φ k (p) = m(ea (p), ea (p)),
a=1

and for all other q near p,

k
∑
φ k (q) ≤ m(ea (q), ea (q)).
a=1

Choose 1 ≤ a ≤ k and, for ease of presentation, set X := ea . Consider now the function m(X , X ). We have
d
∑
∆m(X , X ) = eb eb m(X , X )
b=1
d
∑
2eb m ∇ eb X , X
( )
=
b=1
d
∑
2m ∇ eb ∇ eb X , X + 2m ∇ eb X , ∇ eb X
( ) ( )
=
b=1
d ( )
∑
2m ∇eb ∇eb X + Âeb ∇eb X + ∇ eb Âeb X , X + 2m ∇ eb X , ∇ eb X .
( )
=
b=1

Since X is parallel along geodesics leaving p, this yields

d ( ) ( )
∑
∆m(X , X ) = 2m ∇ eb Âeb X , X + 2m Âeb X , Âeb X .
b=1

Since Σ̂ is positive, m is non-positive over JTp Σ̂ , and so

d ( )
∑
∆m(X , X ) ≤ 2m ∇ eb Âeb X , X .
b=1

Since m(·, ·) and m(·, J·) are both diagonal with respect to the basis e1 , . . . , ed , this becomes
d ( ) ( )
∑
∆m(X , X ) ≤ 2g ∇ eb Âeb X , X m(X , X ) + 2g ∇ eb Âeb X , JX m(JX , X ).
b=1
( ) ( )
Since g ÂY X , Z and g ÂY X , JÂW Z always vanish, this yields

d ( )
∑
∆m(X , X ) ≤ (−2)g Âeb X , Âeb X m(X , X ) + 2(∇eb II)(eb , X , X )m(JX , X ).
b=1
108 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

Finally, applying the Codazzi–Mainardi equations (2.3) yields

d [ ( )
∑
∆m(X , X ) ≤ (−2)g Âeb X , Âeb X m(X , X )
b=1
(2.6)
+ 2(∇X II)(eb , eb , X )m(JX , X ) ]
+ 2g(Reb X eb , JX )m(JX , X ) .

Since m is non-negative over Tp Σ̂ , the first term on the right hand side is absorbed into f k φ k , and since the second vanishes
by Lemma 2.3.2, it only remains to study the contribution of the third. However, by (2.2),
d d [
∑ ∑
g(Reb X eb , JX )m(JX , X ) = (−1) ⟨(JX )r , eb,i ⟩⟨eb,r , Xi ⟩
b=1 b=1 ]
+ ⟨(JX )i , eb,i ⟩⟨eb,i , Xi ⟩ m(JX , X )
[
+ ⟨eb,r , eb,i ⟩⟨(JX )r , Xi ⟩
]
+ ⟨eb,i , eb,i ⟩⟨(JX )i , Xi ⟩ m(JX , X )

Thus, using again the fact that e1 , . . . , ed diagonalises both m(·, ·) and m(·, J·),
d [
∑
g(Reb X eb , JX )m(JX , X ) = ⟨Xi , Xi ⟩⟨Xr , Xi ⟩
b=1 ]
− ⟨Xr , Xi ⟩⟨Xi , Xi ⟩ m(JX , X )
d
∑
− ⟨eb,r , eb,i ⟩⟨Xi , Xi ⟩m(JX , X )
b=1
d
∑
+ ⟨eb,i , eb,i ⟩⟨Xr , Xi ⟩m(JX , X ),
b=1

so that
d
∑ 1
g(Reb X eb , JX )m(JX , X ) = − Tr(m)⟨Xi , Xi ⟩m(JX , X )
2
b=1
d
1∑
+ ⟨eb,i , eb,i ⟩m(JX , X )m(X , X ).
2
b=1

The second term on the right hand side is again absorbed into f k φ k , and the result now follows upon summing over all a. □

2.6. The refined special lagrangian angle

We now show how positive special legendrian submanifolds divide into disjoint families, one of which will have the
properties that we require. Indeed, consider a point p ∈ M, let R and I be respectively the real and imaginary subspaces
of the fibre αp as defined in (2.1), and let E be a positive, lagrangian subspace of the fibre αp . Upon perturbing E slightly if
necessary, we may suppose that it is the graph over R of some symmetric matrix A : R → I . Bearing in mind (2.4), we see
that E is θ -special lagrangian whenever
d
∑
Θ (E) := Arctan(λi ) = −θ mod π Z,
i=1

where 0 < λ1 ≤ · · · ≤ λd are the eigenvalues of A. The function Θ extends continuously to all positive special lagrangian
subspaces, including those that are not graphs. We call Θ the refined special lagrangian angle of E. Of particular interest to us
will be the case where Θ = π/2.

Lemma 2.6.1. Let E be a positive lagrangian subspace of the fibre αp with refined special lagrangian angle equal to π/2. For all
1 ≤ k ≤ (d − 1), there exists an orthonormal k-tuple (X1 , . . . , Xk ) of unit vectors in E realising φ k (E) such that, for all a,
m(Xa , JXa ) ≥ 0.
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 109

Proof. Indeed, upon perturbing E slightly if necessary, we may suppose that it is the graph of a positive definite, symmetric
matrix A. If 0 < λ1 ≤ · · · ≤ λd are the eigenvalues of this matrix, then
d
∑
0< Arctan(λa ) = Θ (E) = π/2.
a=1

In particular,
π
Arctan(λd−1 ) ≤ ,
4
and
π
Arctan(λd−1 ) ≤ Arctan(λd ) ≤ − Arctan(λd−1 ),
2
so that

λd−1 ≤ 1,
and
1
λd−1 ≤ λd ≤ .
λd−1
Furthermore, λd = 1/λd−1 if and only if d = 2 and λ1 = λ2 = 1. Suppose now that (X1 , . . . , Xk ) is an orthonormal k-tuple of
vectors in E which realises φ k (E). Then, for all 1 ≤ a ≤ k, we may suppose that
1
Xa = (Xa′ , µa Xa′ ),
1 + µ2a
for some unit eigenvector Xa′ of A with eigenvalue µa . However, for each a,
2µa
m(X , X ) = .
1 + µ2a
On the other hand, a straightforward calculation shows that
2λd 2λd−1 2/λd−1
≥ = ,
1+λ 2
d 1+λ 2
d−1 1 + 1/λ2d−1
and for all 1 ⩽ b ⩽ (d − 1),
2λb 2λd−1
≤ .
1+λ 2
b 1 + λ2d−1
It follows that

µa ≤ λd−1 ≤ 1,
so that
1
m(X , JX ) = m((Xa′ , µa Xa′ ), (−µa Xa′ , Xa′ ))
1 + µ2a
1 − µ2a
=
1 + µ2a
≥ 0,
as desired. □

Theorem 2.6.2. If Σ̂ is a positive, π/2-special legendrian submanifold of M with refined special lagrangian angle equal to π/2,
then m has constant nullity over Σ̂ .

Proof. First choose 1 ≤ k ≤ (d − 1). By Lemmas 2.5.2 and 2.6.1, there exists a continuous function f k : Σ̂ → R such that

∆φ k + f k φ k ≤ 0.
It follows by the strong maximum principle (c.f. Theorem 3.5 of [16], and the subsequent discussion) that if φ k vanishes at a
single point, then it vanishes identically. We conclude that either m has constant nullity over Σ̂ , or that its nullity is at least
(d − 1) at every point of this submanifold. Finally, since Σ̂ is special legendrian with Θ = π/2, the nullity of m cannot be
equal to (d − 1), so that the latter case only occurs when m vanishes identically. This completes the proof. □
110 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

2.7. The geometry of curtain submanifolds

Following [17], we call curtain submanifolds those positive, special legendrian submanifolds over which the restriction of
′ ′
m has non-trivial nullity. We now describe their geometry. Observe first that, for all d′ < d, the product Rd−d × M d naturally
′ ′
embeds into M d and, if Σ̂ is a positive special legendrian submanifold of M d , then R(d−d ) × Σ̂ is a positive special legendrian
submanifold of M d with the same refined special lagrangian angle. We now show that, up to the action of an element of
O(d, 1), such products account for all curtain submanifolds. To this end, the following construction will prove useful. Let Hd
denote the upper component of the unit pseudosphere in Rd,1 , that is,
Hd = y | ∥y∥2 = −1, yd+1 > 0 .
{ }

Observe that Hd is isometric to d-dimensional hyperbolic space. Consider now an embedded submanifold S in Hd and denote
its normal bundle in Hd by NS, that is
NS := (x, y) | y ∈ S , x ⊥ y, x ⊥ Ty S .
{ }

Given a smooth function ξ : S → Rd,1 , we now define

N ξ S := (x + ξ (y), y) | y ∈ S , x ∈ Ny S .
{ }

In particular, we may assume that ξ is everywhere normal to Ny S. We now establish under which conditions N ξ S is positive
special legendrian.

Lemma 2.7.1. Let ξ : S → Rd,1 be such that ξ (y) is orthogonal to Ny S for all y ∈ S.
(1) N ξ S is a legendrian submanifold of M d if and only if
ξ (y) = ∇φ (y) − φ (y)y
for some smooth function φ : S → R;
(2) N ξ S is positive if and only if S is totally geodesic; and
(3) N ξ S is positive special legendrian if and only if
Im(eiθ Det(Hess(φ ) + (i − φ )Id)) = 0,
for some θ ∈ R, where φ : S → R is the function defined in (1).

Remark. In particular, if N ξ S is positive special legendrian then, upon applying an element of O(d, 1), we may suppose that
S is an open subset of the totally geodesic subspace

({0} × Rd ,1 ) ∩ Hd .
′

It then follows that

N ξ S = R(d−d ) × Σ̂
′

′
for some d′ < d and for some d′ -dimensional submanifold Σ̂ of M d . In particular, it is straightforward to verify that Σ̂ is
also positive special legendrian, so that N ξ S is one of the products described above.

Proof. Let 0 ≤ m < d be the dimension of S. Consider a point (x, y) ∈ NS. Let e1 , . . . , em be an orthonormal basis of Ty S and
extend this to an orthonormal basis e1 , . . . , ed of Ty Hd . In particular em+1 , . . . , ed is an orthonormal basis of Ny S. A basis of
T(x+ξ (y),y) N ξ S is now given by

(A(x, y)ei + Dei ξ (y), ei ) if 1 ≤ i ≤ m, and

{
fi :=
(ei , 0) if m < i ≤ d,
where, for all (x, y) ∈ NS, A(x, y) : Ty S → Ty S here denotes the shape operator of S at the point y with respect to the normal
vector x. For all 1 ≤ i, j ≤ m, denote
aij (x, y) := ⟨A(x, y)ei + Dei ξ (y), ej ⟩.
Suppose now that N ξ S is legendrian. In particular, T(x+ξ (y),y) N ξ S is contained in the fibre α(x+ξ (y),y) so that, for all i,
⟨Dei ξ (y), y⟩ = 0.
Thus, setting φ := ⟨ξ (y), y⟩, we have
⟨∇φ (y), ei ⟩ = Dei ⟨ξ (y), y⟩ = ⟨ξ (y), Dei y⟩ = ⟨ξ (y), ei ⟩,
and since y has negative norm-squared, it follows that
ξ (y) = ∇φ (y) − φ (y)y.
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 111

Conversely, if ξ has the above form, then we readily verify that T(x+ξ (y),y) N ξ S is contained in the fibre α(x+ξ (y),y) and that aij is
symmetric, so that N ξ S is legendrian. This proves (1). Observe now that N ξ S is positive if and only if the matrix aij is positive
definite for all x and for all y. However, since A is linear in x, this holds if and only if A(x, y) vanishes identically, and this
proves (2). Finally, observe that the Hessian of φ over Σ is given by

Hess(φ )(y)(ei , ej ) = ⟨Dei ∇φ (y), ej ⟩ = ⟨Dei ξ (y), ej ⟩ + φ⟨ei , ej ⟩.

However, as in (2.4), N ξ S is special legendrian if and only if

Im(eiθ Det(a + iId)) = 0,

for some θ , and since A(x, y) vanishes identically, this proves (3). □

Theorem 2.7.2. If Σ̂ is a complete curtain submanifold of M d with refined special lagrangian angle equal to π/2 then, up to the
action of an element of O(d, 1),
′
Σ̂ = R(d−d ) × Σ̂ ′ ,
′
where d′ < d and Σ̂ ′ is a complete, positive, special legendrian submanifold of M d with refined special lagrangian angle equal to
π/2.

Remark. Observe that the only positive special legendrian submanifolds of M 1 with refined special lagrangian angle equal
to π/2 are simply the fibres of this bundle over R1,1 . In this manner, we recover the structure of 2-dimensional curtain
submanifolds studied by Labourie in [17].

Proof. By Lemma 2.7.1, it suffices to show that every point p of Σ̂ has a neighbourhood Ω of the form

Ω = N ξ S,
for some embedded submanifold S of Hd and some smooth function ξ : S → Rd,1 . To this end, observe first that

Ker(m) = (T Σ̂ ∩ R) ⊕ (T Σ̂ ∩ I ).

Furthermore, since Σ̂ has refined special lagrangian angle equal to π/2, the distribution T Σ̂ ∩ I has dimension at most 1,
and is non-trivial if and only if m vanishes identically over Σ̂ . In particular, the distribution T Σ̂ ∩ R has constant non-zero
dimension. Observe now that T Σ̂ ∩ R also coincides with Ker(Dπ ), where π : M d → Hd is the projection onto the second
factor. However, considered as a 1-form taking values in Rd,1 , we have

dDπ = 0,

and it follows that T Σ̂ ∩ R is integrable. Let F denote the smooth foliation of Σ̂ that it defines, and observe that π is constant
over every leaf of this foliation.
Consider now a point p ∈ Σ̂ , let Ŝ be a smooth submanifold passing through p which is transverse to F , and let S be its
image under the projection π . Upon reducing Ŝ if necessary, we may suppose that S is an embedded submanifold of Hd , and
that π restricts to a diffeomorphism of Ŝ onto S. Let Ω be the neighbourhood of p consisting of the union of all leaves of F
which pass through Ŝ. We show that Ω has the desired form. Indeed, let q̂ be another point of Ω , denote q := π (q̂), let L be
the leaf of F passing through q̂, and let X := (Xr , 0) be a tangent vector to L at this point. Since Tq̂ Σ̂ is contained in αq̂ , we
have

⟨Xr , q⟩ = 0,
and since m is non-negative semi-definite over Tq̂ Σ̂ , for all other tangent vectors Y := (Yr , Yi ) to Σ̂ at q̂, we have

m(X , Y )2 ≤ m(X , X )m(Y , Y ) = 0,

so that, for all tangent vectors Yi to S at q, we have

⟨X r , Y i ⟩ = 0 .
It follows that

Tq̂ = Nq S ,

and since Σ̂ is complete, upon integrating we find that there exists a point ξ ∈ Rd,1 such that

Lq̂ = ξ + Nq S ,

as desired. □
112 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

3. Hypersurfaces in Minkowski spacetimes

3.1. Affine deformations and GHMC Minkowski spacetimes

Let PSO(d, 1) denote the group of linear isometries of Rd,1 which preserve the spatial and temporal orientations. Consider
now the upper unit pseudo-sphere in Rd,1 ,

Hd := x ∈ Rd,1 | ∥x∥2 = −1, xd+1 > 0 ,

{ }

and recall that the semi-riemannian metric of Rd,1 restricts to a complete hyperbolic metric over this submanifold. Since
PSO(2, 1) preserves Hd , it identifies in this manner with the group of orientation preserving isometries of hyperbolic space.
A subgroup Γ of PSO(d, 1) is said to be kleinian whenever it is discrete, cocompact and torsion free (c.f. [18]). In particular,
the quotient of Hd by a kleinian subgroup is a compact hyperbolic manifold and, conversely, the fundamental group of any
compact hyperbolic manifold identifies with some kleinian subgroup.
The group of affine isometries of Rd,1 which preserve the spatial and temporal orientations is given by the semidirect
product PSO(d, 1) ⋉ Rd,1 , where the group law is given by

(α, x) · (β, y) := (αβ, x + α (y)).

Given a kleinian subgroup Γ , an affine deformation is a homomorphism ρ : Γ → PSO(d, 1) ⋉ Rd,1 of the form

ρ (α ) = (α, τ (α )),
for some map τ : Γ → Rd,1 . By abuse of notation, we denote the image of ρ in PSO(d, 1) ⋉Rd,1 by Γ ⋉ τ . The homomorphism
property of ρ is equivalent to the cocycle condition on τ , namely

τ (αβ ) = τ (α ) + ατ (β ),
for all α, β ∈ Γ . It follows that the set of affine deformations of Γ naturally identifies with the set of Rd,1 -cocycles over this
group, which itself constitutes a finite-dimensional vector space.
GHMC Minkowski spacetimes are parametrised by affine deformations as follows. First, recall that the future cone in Rd,1
is given by

C := x ∈ Rd,1 | ∥x∥2 ≤ 0, xd+1 ≥ 0 .

{ }

We say that a closed, convex subset K of Rd,1 is future-complete whenever

K + C := {x + y | x ∈ K , y ∈ C } ⊆ K .

In [13], building on the work [1] of Mess, Bonsante shows that, for any affine deformation Γ ⋉ τ in PSO(d, 1) ⋉ Rd,1 , there
exists a unique future-complete, closed, convex subset K of Rd,1 such that
(1) K is invariant under the action of Γ ⋉ τ ,
(2) Γ ⋉ τ acts properly discontinuously on the interior of K , and
(3) K is maximal with respect to inclusion amongst all future-complete, closed, convex subsets of Rd,1 which satisfy (1)
and (2).
The quotient K o /Γ ⋉ τ is a GHMC Minkowski spacetime. Throughout the sequel, we will refer to the set K as Bonsante’s
domain for τ , and we will refer to GHMC spacetimes that arise in this manner as Bonsante spacetimes. In [14], Barbot shows
that, up to reversal of the temporal orientation, every GHMC Minkowski spacetime which is not a translation spacetime or a
Misner spacetime is, up to a finite cover, a twisted product of a Bonsante spacetime with a euclidean torus. In particular, the
general result readily follows from the result for Bonsante spacetimes, and we therefore restrict our attention henceforth to
this case.
Bonsante’s construction can also be interpreted as follows. Given a fixed kleinian subgroup Γ , there exists a set valued
function K which maps every Rd,1 -cocycle τ over Γ to Bonsante’s domain K (τ ) for τ . It can be deduced from [13] that this
function is continuous with respect to the local Hausdorff topology, that is, if (τm ) is a sequence of cocycles converging to τ∞
then, for all r > 0,

K (τm ) ∩ Br (0) → K (τ∞ ) ∩ Br (0)

in the Hausdorff sense, where Br (0) here denotes the open ball of (euclidean) radius r about the origin in Rd,1 . Furthermore,
all supporting hyperplanes to K (τ ) are either spacelike or null and the intersection of K (τ ) with any spacelike hyperplane is
compact. It follows, in particular, that if (xm ) is a divergent sequence of boundary points of K (τ ) and if, for all m, Nm is a unit,
timelike supporting normal at the point xm , then the sequence (Nm ) also approaches the light cone as m tends to infinity.
At this stage, it is worthwhile to verify the existence of non-trivial examples of Rd,1 -cocycles and affine deformations.
Indeed, although, in the (2 + 1)-dimensional case, large families of cocycles are obtained via the natural identification of
R2,1 -cocyles with tangent vectors to the Teichmüller space of the surface H2 /Γ (c.f. [19]), in higher dimensions, Cartan–
Weyl local rigidity makes the problem of constructing non-trivial examples a good deal more subtle. One nice technique,
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 113

however, involves interpreting cocycles as infinitesimal variations of the hyperbolic manifold Hd /Γ within the space of flat
conformal manifolds. Indeed, consider the canonical injection

θ0 : Γ → PSO(d, 1) → PSO(d + 1, 1),

and recall that any homomorphism θ : Γ → PSO(d + 1, 1) which is sufficiently close to θ0 is the holonomy of some flat
conformal structure over the manifold Hd /Γ (c.f. [20] and [10]). Suppose now that θ0 extends to a non-trivial smooth family
(θt )t ∈]−ϵ,ϵ [ of homomorphisms of Γ into PSO(d + 1, 1). Then τ̃ := θ0−1 (∂t θ )0 defines a cocycle taking values in the Lie algebra
so(d + 1, 1). However, when the space so(d + 1, 1) is considered as a representation of PSO(d, 1) via the adjoint action, it
naturally decomposes as

so(d + 1, 1) = so(d, 1) ⊕ Rd,1 ,

and since the first component of τ̃ vanishes by Cartan–Weyl local rigidity, this map is entirely determined by its second
component, which we readily verify to be the desired cocycle in Rd,1 . Finally, there are various known constructions of non-
trivial, smooth families of flat conformal structures. The simplest involves bending a compact hyperbolic manifold around a
totally geodesic hypersurface, whenever such a hypersurface exists (c.f. [21] and [22]). We refer the reader to [23,24] and [25]
for more details of this and other constructions.
We conclude this section by studying the case where the cocycle vanishes, which is known as the fuchsian case, and
forms the starting point of our construction. In particular, Theorem 1.1.1 trivially holds in this case. Indeed, observe first
that Bonsante’s domain for the trivial cocycle τ = 0 coincides with the future cone C . Now, given k ∈ ]0, ∞[, consider the
hypersurface

Σk := x ∈ C | ∥x∥2 = −1/k2 ,
{ }

and observe that this hypersurface has constant scalar curvature equal to (−k2 ). Furthermore, the family (Σk )k∈]0,∞[
constitutes a smooth foliation of the interior of C . Indeed, if we define the smooth submersion φ : C o → ]0, ∞[ by
−1
φ (y) := ,
∥y∥2
then, for all k, Σk is the level subset of φ at height k. Observe finally that (Σk ) tends to ∂ C in the local Hausdorff sense as
k tends to +∞. This foliation, which we henceforth refer to as the fuchsian foliation, will be of use at various stages in the
sequel.

3.2. Stability

Let Γ ⊆ PSO(3, 1) be a kleinian subgroup, let τ : Γ0 → R3,1 be a cocycle, and let K := K (τ ) be Bonsante’s domain for τ .
For k > 0, let Σk ⊆ K be a spacelike, locally strictly convex hypersurface in K of constant scalar curvature equal to (−k2 ) and
invariant under the action of Γ ⋉ τ . In this section, we study infinitesimal perturbations of Σk corresponding to infinitesimal
variations of the cocycle τ .
First recall that, by the Gauss–Codazzi equations, the scalar curvature of Σk is given by
1
S := − (λ1 λ2 + λ1 λ3 + λ2 λ3 ),
3
where λ1 , λ2 and λ3 are its principal curvatures. The Jacobi operator of Σk then measures the infinitesimal variation of scalar
curvature resulting from an infinitesimal normal perturbation of this surface. More formally, let N : Σk → R3,1 be the
future-oriented, unit normal vector field over Σk , and for f ∈ C0∞ (Σk ), and for t ∈ R, define

Φf ,t (x) := x + tf (x)N(x).
For any given x, and for sufficiently small t, Φf ,t is also an embedding near x. In particular, letting Sf ,t (x) denote its scalar
curvature at the point x, we define
∂
⏐
Sf ,t (x)⏐⏐ ,
⏐
(Jf )(x) :=
∂t t =0

and we call J the Jacobi operator of Σk .

Now let A be the shape operator of Σk . Since Σk is locally strictly convex, A is everywhere positive definite. Define B by
1
B := (Tr(A)Id − A) .
3
It is straightforward to see that B is also everywhere positive definite, and that

Tr(BA) = 2k2 .
114 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

Lemma 3.2.1. The Jacobi operator of Σk is given by

Jf = Bij A2ij f − Bij f;ij ,
where the summation convention is implied and f;ij here denotes the Hessian of f along Σk .
Applying the maximum principle and the Fredholm alternative immediately yields

Corollary 3.2.2. The Jacobi operator of Σk is invertible.

Proof of Lemma 3.2.1. The following calculation is standard in the riemannian setting. We repeat it here in the lorentzian
setting as care is required with signs that are different in certain places. Fix f ∈ C ∞ (Σk ). Define I : Σk × R → R3,1 by
I(x, t) = x + tf (x)N(x).
Denote by g the pull-back through I of the Minkowski metric over R3,1 . We extend N to a vector field over Σk × ] − ϵ, ϵ [
such that, for all (x, t), N(x, t) is normal to the hypersurface, Σk × {t }. Observe that, by definition, for all x
f (x, 0)N(x, 0) = ∂t .
We claim that, along Σk × {0},
∇∂t N = ∇ Σ f .
Indeed,
∂t ⟨N , N ⟩ = 0
⇒ ⟨∇∂t N , N ⟩ = 0.
Likewise, for any vector field U tangent to Σk and independent of t,
∂t ⟨N , U ⟩ = 0
⇒ ⟨∇∂t N , U ⟩ = −⟨N , ∇∂t U ⟩
= −⟨N , ∇U ∂t ⟩
= −⟨N , ∇U fN ⟩
= ⟨∇ Σ f , U ⟩,
where the last equality follows from the fact that ⟨N , N ⟩ = −1. This proves the assertion.
We now define the endomorphism field A over Σk × ] − ϵ, ϵ [ such that, for all (x, t), (AN)(x, t) = 0, and the restriction of
A(x, t) to the tangent space of Σk × {t } coincides with the shape operator of this hypersurface at this point. We are interested
in the covariant derivative of A in the time direction. Bearing in mind that R3,1 is flat, for any vector field U tangent to Σk
and independent of time, we have
(∇∂t A)U = ∇∂t (AU) − A∇∂t U
= ∇∂t ∇U N − A∇U ∂t
= ∇U ∇∂t N − A∇U fN
= ∇Σ Σ
U ∇ f − fA U .
2

Finally, the scalar curvature of Σk is given by

1(
S = σ (A) := − Tr(A)2 − Tr(A2 ) .
)
6
The derivative of the function σ at A is given by
DS(A)M = −Bij Mij ,
where the summation convention is implied, and the result now follows by the chain rule. □

Lemma 3.2.3. Let (τt )t ∈]−ϵ,ϵ [ be a smoothly varying family of R3,1 -cocycles such that τ0 = τ . Upon reducing ϵ if necessary, there
exists a unique, smoothly varying family (Σk,t )t ∈]−ϵ,ϵ [ of spacelike, locally strictly convex hypersurfaces such that, for all t, Σk,t is
of constant scalar curvature equal to (−k2 ) and is invariant under the action of Γ ⋉ τt .

Proof. We first define a smooth family (Σk′ ,t )t ∈]−ϵ,ϵ [ of spacelike, locally strictly convex hypersurfaces such that, for all t, the
hypersurface Σk′ ,t is invariant under the action of Γ ⋉ τt but does not necessarily satisfy the curvature condition. The desired
family will then be obtained via a perturbation argument. In what follows, we identify Γ with the action of Γ ⋉ τ0 over Σ0 .
Let N : Σk → R3,1 be the future-oriented, unit, normal vector field over Σk . Let φ ∈ C0∞ (Σk ) be a smooth, positive function
of compact support such that, for all x ∈ Σk ,
∑
φ (γ (x)) = 1.
γ ∈Γ
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 115

Define e : ] − ϵ, ϵ [ × Σk → R3,1 by
∑
φ (γ (x)) τt (γ −1 ) − τ0 (γ −1 ) ,
( )
et (x) := e(x) +
γ ∈Γ

and, for all t, denote et := e(t , ·). By construction, for all t, et (Σk ) is invariant under the action of Γ ⋉ τt and, upon reducing
ϵ if necessary, we may suppose furthermore that it is embedded. This yields the desired family.
For all t, let Nt : Σk → R3,1 be the future-oriented, unit, normal vector field over et . Define Φ : C ∞ (Σk ) × ] − ϵ, ϵ [ →
C (Σk , R3,1 ) by
∞

Φf ,t (x) = et (x) + f (x)Nt (x).

Observe that if f is Γ -invariant, then Φf ,t is Γ ⋉ τt equivariant for all t. Furthermore, for sufficiently small (f , t), Φf ,t is
an embedding. Let Cinv ∞
(Σk ) denote the space of smooth functions over Σk that are Γ -invariant and define S : Cinv ∞
( Σk ) ×
] − ϵ, ϵ [ → Cinv (Σk ) such that, for all sufficiently small (f , t), and for all x, Sf ,t (x) is the scalar curvature of the embedding
∞
l,α
Φf ,t at the point x. Now, given (l, α ), by defining the Hölder space Cinv (Σk ) in a similar manner, we see that the functional S
l+2,α l,α
extends continuously to a smooth map from Cinv (Σk ) × ] − ϵ, ϵ [ into Cinv (Σk ). Furthermore, its partial derivative with
respect to the first component at the point (0, 0) is simply the Jacobi operator J. Since J is invertible, it follows by the
implicit function theorem for maps between Banach manifolds that, upon reducing ϵ if necessary, there exists a smooth
l+2,α
map φ : ] − ϵ, ϵ [ → Cinv (Σk ) such that, for all t, Φφ (t),t (Σk ) has constant scalar curvature equal to (−k2 ). Furthermore, by
elliptic regularity, these surfaces are smooth for all t, and this completes the proof. □

3.3. Compactness

We first obtain an elementary result concerning the position of a constant scalar curvature hypersurface inside a given
GHMC Minkowski spacetime. As before, let Γ ⊆ PSO(3, 1) be a kleinian subgroup, let τ : Γ → R3,1 be a cocycle, and let
K := K (τ ) be Bonsante’s domain for τ . For r > 0, let K r denote the set of all points of K lying at a (timelike) distance of at
least 1/r from the boundary ∂ K . For k > 0, let Σk ⊆ K be a spacelike, locally strictly convex hypersurface of constant scalar
curvature equal to (−k2 ) which is invariant under the action of Γ ⋉ τ , and let Σk+ denote the future-complete, convex set
bounded by Σk .

Lemma 3.3.1. K k ⊆ Σk+ ⊆ K .

Proof. Let x be a point of ∂ K . For all ϵ > 0, denote x(ϵ ) := x + ϵ e4 , where e4 here denotes the fourth canonical basis vector
of R3,1 , and let Cx(ϵ ) be the future cone based on x(ϵ ). Define φ : Cx(
o
ϵ ) → ]0, ∞[ by

−1
φ (y) = ,
∥y − x(ϵ )∥2
and observe that the level sets of φ are simply the leaves of the fuchsian foliation of Cx(
o
ϵ ) which was introduced in Section 3.1.
In particular, for all y ∈ Cx(ϵ ) , the level set of φ passing through the point y has constant scalar curvature equal to φ (y).
o

Consider now a fundamental domain Ω of Σk and observe that, for all ϵ > 0, there are only finitely many elements α of
Γ ⋉ τ such that α (Ω ) has non-trivial intersection with Cx(ϵ ) . From this it follows that Σk ∩ Cx(ϵ ) is a relatively compact open
subset of Σk , and so φ attains a minimum value at some point y, say, of this intersection. At this point, Σk is an interior
tangent to the leaf of curvature φ (y) so that, by the geometric maximum principle, k2 ≤ φ (y). In particular, Σk has trivial
intersection with the set φ −1 (]0, k[), and the result follows by letting ϵ tend to 0 and taking the union over all x ∈ ∂ K . □
Consider now a sequence (km ) of positive real numbers, and a sequence (τm ) of cocycles. For all m, let Km := K (τm ) be
Bonsante’s domain for τm , and let Σk,m ⊆ Km be a spacelike, locally strictly convex hypersurface of constant scalar curvature
equal to (−k2m ) which is invariant under the action of Γ ⋉ τm . Suppose that (km ) and (τm ) converge to k∞ and τ∞ respectively.
In particular (Km ) converges in the local Hausdorff sense to K∞ := K (τ∞ ).

Lemma 3.3.2. For every compact subset X of R3,1 , there exists a compact subset Y of H3 such that, for any m, and for any point
x of Σm ∩ X , if N(x) is the future-oriented, unit normal vector of Σm at this point, then N(x) is an element of Y .

Proof. Suppose the contrary. There exists a sequence (xm ) such that, for all m, xm ∈ Σm ∩ X but such that (Nm (xm )) diverges.
Since Γ is cocompact, upon composing with suitable elements of Γ ⋉ τm , we may suppose instead that the sequence (Nm (xm ))
km k
remains within some fixed compact set, but that the sequence (xm ) diverges. For all m, denote K m := Km − xm , K m := Kmm − xm
+ km
and Σ m := Σm
+
− xm . Since (Km ) converges towards (K∞ ), we may suppose that both (K m ) and (K m ) converge in the local
km +
Hausdorff sense to the future side of the same null hyperplane H, say. Furthermore, since ⊆ Σ m ⊆ K m , the sequence
Km
+
(Σ m ) also converges to the future side of H. However, by compactness, we may suppose that Nm converges towards some
+
limit N∞ , say. In particular, this vector is a supporting normal at the origin to the limit of (Σ m ), that is, the future side of H.
This is absurd, since H is null, and the result follows. □
116 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117

Lemma 3.3.3. There exists a spacelike, locally strictly convex hypersurface Σ∞ towards which Σm subconverges in the Cloc
∞
sense.

Proof. Upon rescaling, we may suppose that km = −1/3 for all m. Furthermore, by Lemma 3.3.2, the sequence (Σm ) is
uniformly spacelike over every compact set. Since the scalar curvature is an elliptic curvature function (c.f. [26] and [27]), it
now suffices to show that for every compact subset X of Rd,1 , there exists B > 0 such that, for all m, the shape operator Am
of Σm satisfies
1
≤ Am (x) ≤ B,
B
for all x ∈ Σm ∩ X . Furthermore since, for all m, Σm has constant scalar curvature equal to (−1/3), it suffices to prove the
lower bound. Now suppose the contrary, so that there exists a sequence of points (xm ) contained within some compact set
X , say, and a sequence (λm ) of positive real numbers converging to zero such that, for all m, xm is an element of Σm and λm
is an eigenvalue of the shape operator of Σm at this point.
For all m, let Nm be the future-oriented, unit, normal vector field over Σm , and define

Σ̂m := {(x, Nm (x)) | x ∈ Σm } , and

x̂m := (xm , Nm (x)).

By Lemma 2.4.1, for all m, Σ̂m is a complete, positive, special legendrian submanifold of M := U+ R3,1 with refined special
lagrangian angle equal to π/2. Furthermore, by Lemma 3.3.2, the sequence (x̂m ) is contained within a compact subset of M. It
follows by Theorem 2.4.2 that there exists a complete, pointed, positive, special legendrian submanifold (Σ̂∞ , x̂∞ ) towards
which (Σ̂m , x̂m ) subconverges.
Since the least eigenvalue of the shape operator of Σm at the point xm tends to zero, it follows by Theorem 2.6.2 that Σ̂∞
is a curtain submanifold. In particular, by Theorem 2.7.2 its projection onto R3,1 is foliated by complete, spacelike geodesics,
which must all be contained in K∞ . This is absurd, since the intersection of K∞ with any spacelike hyperplane is compact.
The result follows. □

3.4. Existence and uniqueness

We now prove the main result of this paper.

Theorem 3.4.1. Let Γ ⊆ PSO(3, 1) be a kleinian subgroup. Let τ : Γ → Rd,1 be an R3,1 -cocycle. Let K := K (τ ) be Bonsante’s
domain for τ . For all k > 0, there exists a unique, spacelike, locally strictly convex hypersurface Σk in K which is of constant scalar
curvature equal to (−k2 ) and which is invariant under the action of Γ ⋉ τ . Furthermore, the family (Σk )k>0 constitutes a smooth
foliation of the interior of K .

Proof. We first prove existence. Thus, let I ⊆ [0, 1] be the set of all t ∈ [0, 1] such that the existence part of the result holds
for the cocycle t τ . By Lemmas 3.2.3 and 3.3.3, I is both open and closed. The fuchsian case described at the end of Section 3.1
shows that 0 is an element of I so that, by connectedness, 1 is also an element of I, and existence follows.
In fact, it follows from Lemmas 3.2.1 and 3.3.3 that the space of such hypersurfaces is discrete and compact and is therefore
finite. Furthermore, one can then show via an elementary degree theoretic argument (c.f. [27]) that the number of such
hypersurfaces is independent of both k and τ . However, uniqueness can also be proven more directly as follows. Let Σk be a
spacelike, locally strictly convex hypersurface in K of constant scalar curvature equal to (−k2 ) which is invariant under the
action of Γ ⋉ τ . For t ∈ R, define Σk,t := Σk + te4 , where e4 here denotes the fourth canonical basis element of R3,1 . That
is, Σk,t is obtained by translating Σk vertically upwards by a distance of t. Observe now that Σk is asymptotic to ∂ K in the
sense that the vertical distance between the two tends to 0 at infinity. Now let Σk′ be another hypersurface with the same
properties as Σk . Since Σk′ is also asymptotic to ∂ K , there exists t ∈ R such that Σk′ is an interior tangent to Σk,t at some
point, and it follows by the strong geometric maximum principle that these two hypersurfaces coincide. Finally, since both
Σk and Σk′ are asymptotic to one another, it follows that t = 0, and this proves uniqueness.
We now prove that the family (Σk )k>0 smoothly foliates the interior of K . Consider k ∈ R. For l sufficiently close to k, Σl
is the normal graph of some function fl over Σk . Consider now the partial derivative,
∂
g := fl | .
∂ l l=k
By Lemma 3.2.1,

Bij g;ij = 1 + φ g ,
 G. Smith / Journal of Geometry and Physics 128 (2018) 99–117 117

for some positive-definite matrix Bij and some strictly positive function φ . It follows by the maximum principle that g is
strictly negative and there therefore exists ϵ > 0 such that (Σl )l∈]k−ϵ,k+ϵ [ smoothly foliates a neighbourhood of Σk . Since
this holds for all k, it follows that (Σk )k>0 smoothly foliates some open subset of K .
It remains to show that this foliation covers the whole of the interior of K . However, by Lemma 3.3.1, Σk converges to
∂ K as k tends to +∞. On the other hand, for t ∈ R let Ct be the future cone based on the point te4 , where e4 again denotes
the fourth canonical basis vector of R3,1 . For t sufficiently large and negative, X is contained in Ct . Now let (Σk′ )k>0 be the
fuchsian foliation of Ct constructed at the end of Section 3.1. By the geometric maximum principle, for all k, Σk lies above
Σk′ . From this it follows that (Σk )k>0 foliates the whole of the interior of K , and this completes the proof. □

Acknowledgements

The author is grateful to François Fillastre for long and enlightening conversations on the subject of semi-riemannian
geometry, without which this work would not have been realised. The author is also grateful to Thierry Barbot for helpful
comments concerning earlier drafts of this paper. This paper was written as part of the project MATH AMSUD 2017, Project
No. 38888QB - GDAR.

References

[1] G. Mess, Lorentz spacetimes of constant curvature, Geom. Dedicata 126 (2007) 3–45.
[2] L. Andersson, T. Barbot, F. Béguin, A. Zeghib, Cosmological time versus CMC time in spacetimes of constant curvature, Asian J. Math. 16 (1) (2012)
37–88.
[3] S. Carlip, Quantum gravity in 2 + 1 dimensions, in: Cambridge Monographs of Mathematical Physics, Cambridge University Press, Cambridge, 1998.
[4] T. Barbot, F. Bonsante, J.-M. Schlenker, Collisions of particles in locally AdS spacetimes I. Local description and global examples, Comm. Math. Phys.
308 (1) (2011) 147–200.
[5] T. Barbot, F. Béguin, A. Zeghib, Prescribing gauss curvature of surfaces in 3-dimen-sional spacetimes, application to the minkowski problem in the
minkowski space, Ann. Instit. Fourier. 61 (2) (2011) 511–591.
[6] F. Bonsante, G. Mondello, J.-M. Schlenker, A cyclic extension of the earthquake flow, Geom. Topol. 17 (2013) 157–234.
[7] F. Bonsante, G. Mondello, J.-M. Schlenker, (2015) A cyclic extension of the earthquake flow II, Ann. Sci. 48 (2015) 811–859.
[8] F. Fillastre, G. Smith, Group actions and scattering problems in Teichmueller theory, arXiv:1605.04563.
[9] F. Bonsante, F. Fillastre, The equivariant Minkowski problem in Minkowski space, Ann. Inst. Fourier, in press.
[10] G. Smith, Moduli of flat conformal structures of hyperbolic type, Geom. Dedicata 154 (1) (2011) 47–80.
[11] A.N. Bernal, M. Sánchez, On smooth Cauchy hypersurfaces and Geroch’s splitting theorem, Comm. Math. Phys. 243 (3) (2003) 461–470.
[12] A.N. Bernal, M. Sánchez, Globally hyperbolic spacetimes can be defined as ‘causal’ instead of ‘strongly causal’, Class. Quant. Grav. 24 (3) (2007) 745–749.
[13] F. Bonsante, Flat spacetimes with compact hyperbolic Cauchy surfaces, J. Differential Geom. 69 (3) (2005) 441–521.
[14] T. Barbot, Flat globally hyperbolic spacetimes, J. Geom. Phys. 53 (2) (2005) 123–165.
[15] G. Smith, Special lagrangian curvature, Math. Ann. 335 (1) (2013) 57–95.
[16] D. Gilbarg, N.S. Trudinger, Elliptic partical differential equations of second order, in: Die Grundlehren der mathemathischen Wissenschaften, Vol. 224,
Springer-Verlag, Berlin, New York, 1977.
[17] F. Labourie, Un lemme de morse pour les surfaces convexes, Invent. Math. 141 (2) (2000) 239–297.
[18] M. Kapovich, Kleinian Groups in Higher Dimensions, in Geometry and Dynamics of Groups and Spaces, Progress in Mathematics, 265 487–564.
[19] W.M. Goldman, Discontinuous groups and the Euler class (Ph.D. thesis), University of California, Berkeley, 1980, p. 138.
[20] R.S. Kulkarni, U. Pinkall, A canonical metric for Möbius structures and its applications, Math. Z. 216 (1) (1994) 89–129.
[21] D. Johnson, J.J. Millson, Deformation spaces associated to compact hyperbolic manifolds, in: Discrete groups in geometry and analysis (New Haven,
Conn., 1984), in: Progr. Math., vol. 67, Birkhäauser Boston, Boston, MA, 1987, pp. 48–106.
[22] C. Kourouniotis, Deformations of hyperbolic structures, Math. Proc. Cambridge Philos. Soc. 98 (2) (1985) 247–261.
[23] B.N. Apanasov, Deformations of conformal structures on hyperbolic manifolds, J. Differential Geom. 35 (1) (1992) 1–20.
[24] M. Kapovich, Deformations of representations of discrete subgroups of SO(3, 1), Math. Ann. 299 (2) (1994) 341–354.
[25] K.P. Scannell, Infinitesimal deformations of some so(3, 1) lattices, Pacific J. Math. 194 (2) (2000) 455–464.
[26] L. Caffarelli, L. Nirenberg, J. Spruck, Nonlinear second-order elliptic equations. V. The Dirichlet problem for Weingarten hypersurfaces, Comm. Pure
Appl. Math. 41 (1) (1988) 47–70.
[27] H. Rosenberg, G. Smith, Degree Theory of Immersed Hypersurfaces, Mem. Amer. Math. Soc., in press.