Frances PDF
Frances PDF
Frances PDF
Charles Frances
4 December 2003
Abstract. In this article, we introduce some basic tools for the study of
Lorentzian Kleinian groups. These groups are discrete subgroups of the
Lorentzian Möbius group O(2, n), acting properly discontinuously on some
non empty open subset of Einstein’s universe, the Lorentzian analogue of
the conformal sphere.
Mathematics Subject Classification: 22E40, 30F40, 53A30, 53C50.
1 Introduction
To understand an hyperbolic manifold Hn+1 /Γ (Hn+1 denotes here the
(n+1)-hyperbolic space and Γ ⊂ O(1, n+1) is a discrete group of hyperbolic
isometries), a nice and powerful tool is the dynamical study of the conformal
action of Γ on the sphere Sn . This deep relationship between hyperbolic and
conformally flat geometry has a counterpart in Lorentzian geometry, often
quoted by physicists as AdS/CFT correspondence. Let us first recall what
is the Lorentzian analogue of the pair (Hn+1 , Sn ). The (n + 1)-dimensional
Lorentzian model space of constant curvature −1 is called anti-de Sitter
space, denoted AdSn+1 (precisely, we are speaking here of the quotient of
the simply connected model AdS ^ n+1 by the center of its isometry group, see
[O’N], [Wo]). This space, like the hyperbolic space, has a conformal bound-
ary. It is called Einstein’s universe, denoted Einn , and it can be defined,
up to a two-sheeted covering, as the product S1 × Sn−1 endowed with the
conformal class of the metric −dt2 × gS n−1 . From the conformal viewpoint,
Einstein’s universe has a lot of properties reminiscent of those of the sphere.
In particular, the group O(2, n) of isometries of AdSn+1 turns out to be also
the group of conformal transformations of Einn . The understanding of an
anti-de Sitter manifold AdSn+1 /Γ thanks to the conformal dynamics of Γ
on Einn is one of the motivations for studying Lorentzian Kleinian groups,
that we define by analogy with the classical theory, as discrete subgroups
1
of O(2, n) acting freely properly discontinuously on some non empty open
subset of Einn .
Since the works of Poincaré and Klein at the end of the nineteenth cen-
tury, the classical theory of Kleinian groups has generated a great amount
of works, and progressed very far (we refer the reader to [A], [Ka], [Ma],
[MK] for an historical account and good expositions on the subject).
Other notions of Kleinian groups also appeared in other geometric con-
texts, such as complex hyperbolic and projective geometry (see for example
[Go], [SV]).
To our knowledge, nothing systematic has been done for studying Loren-
tzian Kleinian groups, so that the aim of this article is to lay some basis for
the theory. In particular, our first task is to build and study nontrivial
examples of such groups.
The first part of the paper (sections 3 and 4) is devoted to what could be
called Lorentzian Möbius dynamics, namely the dynamical study of diver-
gent sequences of O(2, n), acting on Einn . This dynamics appears reacher
than that of classical Möbius transformations on the sphere. This is essen-
tially due to the fact that O(2, n) has rank two, and the different ways to
reach infinity in O(2, n) induce different dynamical patterns for the action
on Einn . These patterns, which are essentially three, are described in sec-
tion 3, propositions 3, 4 and 5. Let us mention here two new phenomena
(with respect to the Riemannian context) illustrating the dynamical com-
plications we are confronted with. Firstly, the Lorentzian Möbius group
O(2, n) is not a convergence group for its action on Einn (roughly speaking,
a group G acting by homeomorphisms on a manifold X is a convergence
group if any sequence (gi ) of G tending to infinity admits a subsequence
with a ”north-south” dynamics, i.e a dynamics with an attracting pole p+
and a repelling one p− . See for example [A] p.40 for a precise definition).
Secondly, a discrete subgroup Γ ⊂ O(2, n) does not always act properly on
AdSn+1 .
In spite of these differences with respect to the classical theory, it is
still possible to define the limit set of a discrete subgroup Γ ⊂ O(2, n) (see
section 4). This is a closed Γ-invariant subset ΛΓ ⊂ Einn , such that the
action on the complement ΩΓ is proper. It is moreover a union of lightlike
geodesics, so that it defines naturally a Γ-invariant closed subset Λ̂Γ of Ln ,
the space of lightlike geodesics of Einn (this space is described in section
2.5). Unfortunately, the nice properties of the limit set in the classical case
of groups of conformal transformations of the sphere, are generally no longer
satisfied in our situation. For example, the limit set that we define is not, in
general, a minimal set for the action of Γ on Einn (although Λ̂Γ is sometimes
minimal for the action of Γ on Ln , see theorem 1 below). Groups Γ ⊂ O(2, n)
acting properly on AdSn+1 are those whose behaviour is closest to that of
classical Kleinian groups. They will be called groups of the first type. For
them, we get nice properties for the limit set:
2
Theorem 1. Let Γ be a Kleinian group of the first type and ΛΓ its limit
set.
(i) The action of Γ is proper on ΩΓ ∪ AdSn+1 ⊂ Einn+1 .
(ii) ΩΓ is the unique maximal element among the open sets Ω ⊂ Einn
such that Γ acts properly on Ω ∪ AdSn+1 .
(iii) If moreover Γ is Zariski dense in O(2, n), then ΩΓ is the unique
maximal open subset of Einn on which Γ acts properly, and Λ̂Γ is a minimal
set for the action of Γ on Ln .
3
2.1 Projective model for Einstein’s universe
Let R2,n be the space Rn+2 , endowed with the quadratic form q 2,n (x) =
−x21 − x22 + x23 + ... + x2n+2 . The isotropic cone of q 2,n is the subset of R2,n
on which q 2,n vanishes. We call C 2,n this isotropic cone, with the origin
removed. Throughout this article, we will denote by π the projection from
R2,n minus the origin, on RP n+1 . The set π(C 2,n ) is a smooth hypersurface
Σ of RP n+1 . This hypersurface turns out to be endowed with a natural
Lorentzian conformal structure. Indeed, for any x ∈ C 2,n , the restriction
of q 2,n to the tangent space Tx C 2,n , that we call q̂x2,n , is degenerate. Its
kernel is just the kernel of the tangent map dx π. Thus, pushing q̂x2,n by
dx π, we get a well defined Lorentzian metric on Tπ(x) Σ. If π(x) = π(y) the
two Lorentzian metrics on Tπ(x) Σ obtained by pushing q̂x2,n and q̂y2,n are in
the same conformal class. Thus, the form q 2,n determines a well defined
conformal class of Lorentzian metrics on Σ. One calls Einstein’s universe
the hypersurface Σ, together with this canonical conformal structure.
The intersection of C 2,n with the euclidean sphere defined by x21 + x22 +
... + x2n+2 = 1 is a smooth hypersurface Σ̂ ⊂ R2,n . One can check that
q 2,n has lorentzian signature when restricted to Σ̂, and in fact, (Σ̂, q 2,n ) is
|Σ̂
isometric to the product (S1 × Sn−1 , −dt2 + gSn−1 ). Now Einstein’s universe
is conformally equivalent to the quotient of (S1 × Sn−1 , −dt2 + gSn−1 ) by an
involution (induced by the map x 7→ −x of R2,n ).
4
p = π(u), with u some isotropic vector of R2,n , then C(p) is just π(P ∩C 2,n ),
where P is the degenerate hyperplane P = u⊥ (the orthogonal is taken for
the form q 2,n ). The lightcones are not smooth submanifolds of Einn . The
only singular point of C(p) is p, and C(p)\{p} is topologically R × Sn−2 .
- Minkowski components.
Given a point p ∈ Einn , the complement of C(p) in Einn is an homogeneous
open subset of Einn , which is conformally equivalent to Minkowski space
R1,n−1 . We say that this is the Minkowski component associated to p. In
fact, we have an explicit formula for the stereographic projection identifying
Einn \C(p) and R1,n−1 (see [CK] [Fr1]).
5
C(p) with vertex p. Since ∆ is a lightlike geodesic, we have ∆ ⊂ C(p).
Now, the intersection of C(p) with Ω∆ is a degenerate hypersurface of Ω∆ ,
diffeomorphic to Rn−1 . We call it H(p). If p 6= p0 , H(p) and H(p0 ) only
intersect along ∆, and the leaves of the foliation H∆ are just the H(p) for
p ∈ ∆.
6
decomposition of the group SO(2, n) (compare [B], [IW]). Such a decom-
position also exists for O(2, n), with K a compact set of O(2, n). Moreover,
for every g ∈ O(2, n), there is a unique a(g) ∈ A+ such that g ∈ Ka(g)K.
The element a(g) is called the Cartan projection of g. As a matrix, it is
written:
eλ(g)
eµ(g)
1
a(g) =
..
.
1
e−µ(g)
e−λ(g)
The reals λ(g) ≥ µ(g) ≥ 0 are called the distortions of the element g
(associated with the given Cartan decomposition).
7
The interest of this definition for the study of actions of discrete groups
can be illustrated by the following: let Γ be a discrete group of Homeo(X)
acting on some open subset Ω ⊂ X. Then, one proves easily:
Proposition 1. The group Γ acts properly on Ω iff no two points of Ω are
dynamically related.
Assuming that the action of Γ on Ω is proper, we also have:
Proposition 2. If the action of Γ on Ω has compact quotient, then every
x ∈ ∂Ω must be dynamically related to some point y of Ω (depending on x).
Now, let (gk ) be a divergent sequence of O(2, n). We define λk = λ(gk ),
µk = µ(gk ) and δk = λk − µk . We say that the sequence (gk ) tends simply
to infinity when:
a) the three sequences (λk ), (µk ) and (δk ) converge respectively to some
λ∞ , µ∞ and δ∞ in R.
b) compact factors in the Cartan decomposition of (gk ) both admit a
limit in K.
Of course, every sequence tending to infinity admits some subsequence
tending simply to infinity, so that we will restrict our study to these last
ones. The sequences tending simply to infinity split into three categories:
(i) Sequences with balanced distortions
This name denotes the sequences (gk ) for which λ∞ = µ∞ = +∞ and δ∞ is
finite.
(ii) Sequences with bounded distortion
This denotes the sequences (gk ) for which µ∞ 6= +∞.
(iii) Sequences with mixed distortions
This denotes the sequences (gk ) for which λ∞ = µ∞ = δ∞ = +∞.
8
3.2.1 Dynamics with balanced distortions
Proposition 3. Let (gk ) be a sequence of O(2, n) with balanced distor-
tions. Then we can associate to (gk ) two lightlike geodesics ∆+ and ∆−
called attracting and repelling circles of (gk ), and two submersions
π+ : Einn \∆− → ∆+ (resp. π− : Einn \∆+ → ∆− ), whose fibers are
the leaves of H∆− (resp. H∆+ ), such that :
For every compact subset K of Einn \∆− (resp. Einn \∆+ ), D(gk ) (K) =
π+ (K) (resp. D(g−1 ) (K) = π− (K)).
k
Remark 1. Before begining the proof, let us remark that if (gk ) has balanced
distortions (resp. bounded distortion, resp. mixed distortions), it will be so
(1) (2) (1)
for any compact perturbation of (gk ), i.e any sequence (lk gk lk ) for (lk )
(2)
and (lk ) two converging sequences of O(2, n). In the same way, the conclu-
sions of the above proposition are not modified by a compact perturbation,
even if of course, π± and ∆± are. So in the following (and also for sections
3.2.2 3.2.3), we will restrict the proofs to the case where (gk ) is a sequence
of A+ .
9
of lightlike geodesics of C − = C(p− ) in the space of lightlike geodesics of
C + = C(p+ ), conformal with respect to the natural conformal structure of
these two spaces, such that :
(i) For all compact subset K inside Einn \C − , we have D(gk ) (K) = {p+ }.
(ii) For a lightlike geodesic ∆ ⊂ C − and a point x of ∆ distinct from
p− , D(gk ) (x) is the lightlike geodesic ĝ∞ (∆).
(iii) The set D(gk ) (p− ) is the whole Einn .
10
3.2.3 Mixed dynamics
Proposition 5. Let (gk ) be a sequence of O(2, n) with mixed distortions.
Then we can associate to (gk ) two points p+ and p− called attracting and
repelling poles of the sequence, as well as two lightlike geodesics ∆+ et
∆− (called attracting and repelling circles), with the inclusions p+ ∈
∆+ ⊂ C + = C(p+ ) and p− ∈ ∆− ⊂ C − = C(p− ), such that the following
properties hold:
(i) For every compact subset K inside Einn \C − , the set D(gk ) (K) is
{p+ }.
(ii) If x is a point of C − , not on ∆− , then D(gk ) (x) is the lightlike
geodesic ∆+ .
(iii) If x is a point of ∆− , distinct from p− , then D(gk ) (x) is the attracting
cone C + .
(iv) The set D(gk ) p− is the whole Einn .
11
4 About the limit set of a Lorentzian Kleinian
group
4.1 Definition of the limit set
Given a Kleinian group Γ on a manifold X, it is quite natural to ask if there
is in some sense a “canonical” open set Ω ⊂ X on which Γ acts properly.
For example, any Kleinian group Γ on the sphere Sn admits a limit set ΛΓ
and the open set ΩΓ = Sn \ΛΓ is distinguished, since it is the only maximal
open subset on which Γ acts properly. The nice properties of the limit set
of a Kleinian group on Sn rest essentially on the fact that the Möbius group
O(1, n + 1) is a convergence group on Sn . We just saw in the previous
section that O(2, n) is quite far from being a convergence group on Einn
but nevertheless, we would like to define a limit set ΛΓ associated to a given
discrete group Γ ⊂ O(2, n). We require that such a limit set have at least
the two following properties:
(i) ΛΓ is a Γ-invariant closed subset of Einn .
(ii) The action of Γ on ΩΓ = Einn \ΛΓ is properly discontinuous.
(1)
[
ΛΓ = ∆+ (γk ) ∪ ∆− (γk )
(γk )∈SΓ
and
(2)
[
ΛΓ = C + (γk ) ∪ C − (γk )
(γk )∈TΓ
12
4.2 Lorentzian Kleinian groups of the first and the second
type
Until now, we didn’t focus on a fundamental difference between the action
of O(1, n + 1) on Sn and that of O(2, n) on Einn . Although any discrete
group Γ ⊂ O(1, n + 1) automatically acts properly on Hn+1 , it is not true
in general that a discrete Γ ⊂ O(2, n) does so on AdSn+1 . This motivates
the following distinction between subgroups of O(2, n):
Definition 4. A discrete group Γ of O(2, n) is of the first type if it acts
properly on AdSn+1 . If not, it is said to be of the second type.
Notice that this terminology has no connection with the denomination
first kind and second kind for the standard Kleinian groups on the sphere.
The previous dichotomy has a nice translation into dynamical terms
thanks to the:
Proposition 6. A Kleinian group Γ of O(2, n) is of the first type if and
only if it doesn’t admit any sequence (γk ) with bounded distortion.
Proof : We endow R2,n+1 with the quadratic form q 2,n+1 (x) = −2x1 xn+2 +
2x2 xn+1 + x23 + ... + x2n + x2n+3 and call e1 , ..., en+3 the canonical basis.
The subgroup of O(2, n + 1) leaving invariant the subspace spaned by the
first n + 2 basis vectors can be canonically identified with O(2, n). This
identification defines an embedding j from O(2, n) into O(2, n + 1). The
action of j(O(2, n)) on Einn+1 leaves invariant a codimension 1 Einstein
universe that we call Einn . As we saw in the introduction, the complement
of Einn in Einn+1 is conformally equivalent to the anti-de Sitter space
AdSn+1 .
Let us consider some g in O(2, n). In the basis e1 , ..., en+3 , j(g) =
g
, so that when we perform the Cartan decomposition of j(g), we
1
find the same distortions as for g.
Suppose now that Γ admits some sequence (γk ) with bounded distortion.
By the remark above, j(γk ) has also bounded distortion as a sequence of
O(2, n+1). We call C + and C − its attracting and repelling cones in Einn+1 .
By proposition 4, D(gk ) (C − ∩ AdSn+1 ) = C + ∩ AdSn+1 . Therefore, we can
find two points of AdSn+1 which are dynamically related, so that the action
of (γk ) on AdSn+1 can not be proper (proposition 1).
Conversely, let us consider some sequence (γk ) tending simply to infinity
and with balanced or mixed distortions. Then the sequence j(γk ) has the
same properties. Let us call ∆+ and ∆− the attracting and repelling circles
of this latter sequence. Looking at the matrix expressions, it is clear that
∆+ ⊂ Einn and ∆− ⊂ Einn . By propositions 3 and 5, D(gk ) (x) ⊂ Einn
for any point x ∈ AdSn+1 . So, if we assume that Γ has no sequence with
bounded distortion, we get DΓ (x) ⊂ Einn for any point x ∈ AdSn+1 . Using
proposition 1, we get that Γ acts properly on AdSn+1 .
13
4.3 Limit set of a group of the first type: proof of Theorem
1
Since Γ is of the first type, ΛΓ is also the limit set of Γ, seen as a subgroup
of O(2, n + 1) acting on Einn+1 . The complement of this limit set in Einn+1
is precisely ΩΓ ∪ AdSn+1 , so that point (i) of the theorem is clear.
To prove point (ii), let us suppose that Γ acts properly on some Ω ∪
AdSn+1 with Ω not included in ΩΓ . Then there is a sequence (γk ) of Γ
(with balanced or mixed distortions) such that ∆− (γk ) meets Ω.
Proof. Suppose on the contrary that for some (γk ) with balanced distor-
tions, we have ∆+ (γk ) ∩ Ω 6= ∅. From proposition 3, we infer that the set
D(γk ) (∆+ (γk ) ∩ Ω) contains a lightlike geodesic ∆ in its interior. So, there
is a tubular neighbourhood U of ∆ contained in Ext(Ω) (Ext(Ω) denotes
the complement of Ω in Einn ). But we also infer from proposition 3 that
for any ∆ not meeting ∆− (γk ), we have limk→+∞ γk (∆) = ∆+ (γk ). As
a consequence, any lightlike geodesic of Ext(Ω) has to cut ∆− (γk ). Since
all the lightlike geodesics included in U can’t all meet ∆− (γk ), we get a
contradiction.
The lemma above tells us that the sequence (γk ) has mixed distortions.
For any point x ∈ ∆− (γk ) ∩ Ω, we have D(γk ) (x) = C + (γk ). Since C + (γk )
meets AdSn+1 , we get pairs of points in Ω ∪ AdSn+1 which are dynamically
related, and the action can’t be proper, by proposition 1.
14
both g(∆− (γk )) and ∆− (γk ) is contained in a two dimensional Einstein’s
universe, which have to be fixed by Γ: a contradiction with the Zariski
density of Γ.
If g(∆− (γk )) and ∆− (γk ) meet in one point p, then any lightlike geodesic
meeting both g(∆− (γk )) and ∆− (γk ) has to contain p. Indeed, due to the
fact that the quadratic form q 2,n can’t have some 3 dimensional isotropic
subspace, there is no nontrivial triangle of Einn , whose edges are pieces of
lightlike geodesics. We infer that Γ has to fix the lightcone C(p) and we get
once again a contradiction.
We can now prove that ΩΓ is the maximal open set on which the action
of Γ is proper. Suppose that Γ acts properly on Ω which is not included
in ΩΓ . We call Λ the complement of Ω in Einn . Since ΛΓ 6⊂ Λ, there is a
sequence (γk ) tending simply to infinity in Γ with ∆+ (γk ) ∩ Ω 6= ∅.
15
tient R1,n−1 /Γ or AdS
] n /Γ, where Γ is a dicrete group of Lorentzian isome-
tries. This deep theorem was first proved for the case of curvature zero
by Carrière in [Ca], and generalized by Klingler in [Kl] (note that compact
Lorentzian manifolds can not have curvature +1). Another result, known
as finiteness of level (see [KR], [Ze]), ensures that any compact quotient
] n /Γ̃ (where Γ̃ is a discrete group of isometries) is in fact, up to finite
AdS
cover, a quotient AdSn /Γ. Since we saw in section 2 that R1,n−1 and AdSn
both embed conformally into Einn , and thanks to theorem 4, we get that
any compact Lorentzian structure with constant curvature is (up to finite
cover) uniformized by a Lorentzian Kleinian groups. Moreover, in this case,
the structure of the groups involved is fairly well understood, thanks to
[CaD], [Sa] and [Ze].
Example 1.
16
We write each z ∈ Cn+1 as z = (x, y) with x and y in Rn . We identify the
real hyperbolic space HnR with the set of points (x, 0) with −x21 + x22 + ... +
x2n+1 = −1 and x1 > 0. If (x, y) is moreover in the unit tangent bundle of
HnR , it satisfies the two extra equations:
−x1 y1 + x2 y2 + ... + xn yn = 0
−y12 + y22 + ... + yn+1
2 =1
Projectivising, we get an open subset Ω̂ ⊂ S2n−1 . In fact Ω̂ is precisely
S2n−1 minus a (n − 1)-dimensional sphere Σ (the projection on S2n−1 of the
set {z = (x, 0)| − x21 + x22 + ... + x2n+1 = 0}).
Now, the subgroup G = O(1, n) of real matrices in U (1, n) acts on S2n−1 ,
and preserves Ω̂. Identifying Ω̂ with T 1 HnR , we get that G acts properly and
transitively on Ω̂. As a consequence, we have the following:
Fact 1. Any discrete group Γ in O(1, n) acts properly discontinuously on
Ω̂. Seen as a subgroup of O(2, 2n) it yields a Kleinian group acting on Ein2n .
The Kleinian manifold Ω/Γ obtained in this way are circle bundles over
T 1 (N ), where N is the hyperbolic manifold HnR /Γ.
Example 2.
Inside U (1, n), there is a group G isomorphic to the Heisenberg group of
dimension 2n − 1. The group G fixes a point p∞ on S2n−1 and acts simply
transitively on the complement of this point. By proposition 7, any discrete
group in G will yield a Lorentzian Kleinian group, acting properly on the
complement of a lightlike geodesic. The Kleinian manifolds obtained in this
way will be circle bundles over nilmanifolds.
Lemma 3. We can write Einn as a union Ω1 ∪Ω2 ∪Σ. The set Ω1 (resp. Ω2 )
is open, G-invariant, homogeneous under the action of G, and conformally
equivalent to the product dSr × Hs (resp. Hr × dSs ). Σ is a singular,
degenerate G-invariant hypersurface.
17
w = π2 (u) are isotropic gives the hypersurface Σ. We will say more about
it later.
The vectors u = (v, w) for which neither v nor w is isotropic are of two
kinds.
- Those for which q 2,n (v) > 0.
Since we work projectively, we can suppose that q 2,n (v) = 1 and q 2,n (w) =
−1. In a further quotient by −Id, these vectors project on the product
dSr × Hs . They constitute the open set Ω1 .
- Those for which q 2,n (v) < 0.
These vectors project on a product Hr × dSs , and constitute the open set
Ω2 .
Example 3.
Let us take a discrete group Γ̂ inside O(1, r) and any representation ρ of Γ̂
inside O(1, s). We call Γρ = Graph(Γ̂, ρ) = {(γ̂, ρ(γ̂))|γ̂ ∈ Γ̂}. Then Γρ is a
Lorentzian Kleinian group of O(2, n). Indeed, its action on Ω2 = Hr × dSs
is clearly proper. Let us say a little bit more about the limit set of these
groups. We call ΛΓ̂ the limit set of the group Γ̂ on the sphere Σ1 .
18
p− (γ̂k ) and p− (ρ(γ̂k ))). In particular, the limit set ΛΓρ is a closed subset of
Σ (strictly included in Σ if ΛΓ̂ 6= Σ1 ).
An interesting subcase arises when we take for Γ̂ a cocompact lattice in
O(1, 2), and a quasi-fuchsian representation ρ : Γ̂ → O(1, s) (s ≥ 2). The
limit set of ρ(Γ̂) on Σ2 is a topological circle, and we get for the limit set
ΛΓρ a topological torus. One can prove moreover (but we don’t do it here)
that the action of Γρ is cocompact on the complement of its limit set.
19
S
(ii) The group Γ is Kleinian. More precisely, Ω = γ∈Γ γ(D) is an open
subset of X, and Γ acts properly discontinuously and cocompactly on Ω, with
fundamental domain D.
Proof : We do the proof for two groups Γ1 and Γ2 , the final result being then
obtained by induction. Let γ = γs γs−1 ...γ2 γ1 be a word of Γ such that γi ∈
Gji (ji ∈ {1, 2}) and ji 6= ji+1 . Then, the first condition on the fundamental
domains yields the inclusions γs γs−1 ...γ2 γ1 (D) ⊂ γs γs−1 ...γ2 (Ext(Dj1 )) ⊂
... ⊂ γs (Ext(Djs−1 )) ⊂ Ext(Djs ). So, for any non trivial reduced g, γ(D) ∩
D = ∅. This proves that γ can’t be the identity and the point (i) follows.
In the same way, we prove that γ(D) ∩ D = ∅ as soon as s > 1. Since D is
compact in Ω1 and Ω2 and the action of Γ1 and Γ2 is proper, we get the:
Lemma 4. The intersection γ(D) ∩ D is empty for all but a finite number
of γ’s.
20
that for i sufficiently large, xi must be in D ∪ γi1 (D) ∪ ... ∪ γim (D), and yi
in γ0 (D) ∪ γ0 γi1 (D) ∪ ... ∪ γ0 γim (D)). But then, lemma 4 implies that the
sequence (γi ) takes its values in a finite set, a contradiction with the fact
that (γi ) tends to infinity in Γ.
21
We now take some g ∈ Conf(Einn ) as in the lemma above. Let us choose
V1 (resp. V2 ) an open tubular neighbourhood of ∆1 (resp. ∆2 ) such that
V 1 ⊂ D1 (resp. V 2 ⊂ D2 ). The complement of Vi (i = 1, 2) in Einn is
denoted by Ext(Vi ). It follows from proposition 5 (points (i) and (ii)) that
the set dynamically associated to Ext(V2 ) with respect to (g k ) is included in
∆+ . Since Ext(V2 ) contains a lightlike geodesic, it is exactly ∆+ . Hence, for
k0 sufficiently large, g k0 (Ext(V2 )) ⊂ V1 . We call Γ02 = g k0 Γ2 g −k0 . The group
Γ02 is a cocompact Lorentzian Kleinian group, with fundamental domain
D20 = g k0 (D2 ). But g k0 (D2 ) contains g k0 (Int(V2 )), and as we just saw,
Ext(V1 ) ⊂ g k0 (Int(V2 )). So Ext(D20 ) ⊂ D1 . We can then apply theorem
5, and we get that the group generated by Γ02 and Γ1 is still Kleinian,
cocompact, and isomorphic to Γ1 ∗ Γ02 , i.e Γ1 ∗ Γ2 .
Example 4
All the cocompact Lorentzian Kleinian groups of the examples 1 and 2 of
section 5 satisfy the hypothesis of theorem 2. This is also the case of most
instances of example 3, when ρ is injective with discrete image. Thus, such
groups can be combined and give new examples. Notice that in the proof of
theorem 2, the gluing element g can be chosen in many ways. In particular,
starting from two groups of the examples 1, 2 or 3, suitable choices of g will
give combined groups which are Zariski dense in O(2, n).
22
Remark 6. This surgery process is reminiscent of Kulkarni’s construction
of a conformally flat Riemannian structure on the connected sum of two con-
formally flat Riemannian manifolds ([K1]). We don’t know if the connected
sum of two conformally flat Lorentzian manifolds can still be endowed with
a conformally flat Lorentzian structure.
23
S
For each k ∈ N, we call Fk = |γ|≤k γ(D), with the convention F0 =
S
T not difficult to check that Fk−1 ⊂ Fk , and Ω = k∈N Fk . So,
D. It’s
Λ = k∈NT Ext(Fk ). For each k, we set Λk = Ext(Fk ), and thus, we also
have Λ = k∈N Λk . The set Λk is a disjoint union of exactly 2g.(2g − 1)k
connected components, in one to one correspondence with the words of
k+1
length k + 1 in Γ. For example, to the word si11 ...sik+1 corresponds the
component si11 ...sikk (U ik+1
k+1
) of Λk . We can now state:
Lemma 7. There is an homeomorphism K between the boundary ∂Γ and the
space of connected components of Λ (endowed with the Hausdorff topology
for the compact subsets of Einn ).
Proof : Let γ∞ = si11 ....sikk ... be an element of ∂Γ. We call γk = si11 ....sikk and
k−1
we look at the decreasing sequence of compact subsets K(γk ) = si11 ....sik−1 (U ikk ).
This decreasing sequence of compact sets tends to a limit compact set K(γ∞ )
for the Hausdorff topology. Since the Ui± are connected, so are the K(γk ),
and K(γ∞ ) is itself connected. Let us remark that if γ∞ and γ∞ 0 are distinct
0
in ∂Γ, then K(γk ) and K(γk ) are disjoint for k large (they represent two
distinct components of Λk ), so that K(γ∞ ) and T K(γ∞0 ) are disjoint.
Proof : Let us consider γ∞ = si11 ...sikk ... in the boundary of Γ. We know that
k−1
K(γ∞ ) is the limit of the sequence si11 ...sik−1 (U ikk ). Since the sequence is
24
decreasing, the limit remains the same if we consider a subsequence. Thus,
we can make the extra assumption that K(γ∞ ) is the limit of a sequence
j
γk (U j00 ), such that (γk ) tends simply to infinity and the first and last letters
j
of γk are always the same, namely si11 and sj11 . Let us precise that j1 j1 6=
−j0 j0 . We are going to discuss the different possible dynamics for (γk ), and
we first prove that (γk ) can’t have bounded distortion.
Suppose that it is the case. We call p+ (resp. p− ) and C + (resp. C − ) the
attracting (resp. repelling) pole and cone of (γk ). If x is a point of D, then
−j
for all k ∈ N, γk (x) ∈ Ui11 and γk −1 (x) ∈ Uj1 1 . So, we must have p+ ∈ Ui11
−j j
and p− ∈ Uj1 1 . In particular, p− is not in Uj0 0 . On the other hand,
this is a general fact that in Einn , any lightlike cone meets any lightlike
geodesic (just because degenerate hyperplanes always meet null 2-planes
j j
in R2,n ). In particular, the cone C − meets ∆j00 , and thus Uj0 0 . We call
j j j
Vj0 0 = C − ∩ Uj0 0 . Since Uj0 0 does not contain p− , we infer from proposition
j j
4 (points (i) and (ii)) that K(γ∞ ) = D(γk ) (V j00 ). More precisely, if V̂j0 0 is
j
the set of lightlike geodesics of C − meeting Vj0 0 , then K(γ∞ ) is the closure
j
of the union the lightlike geodesics of γ̂∞ (V̂j0 0 ) (see proposition 4 for the
notation γ̂∞ ). In particular, K(γ∞ ) contains a lightlike geodesic. Now, some
j
lightlike geodesic of C − does not meet Vj0 0 . Indeed, if it is not the case,
the point (ii) of proposition 4 ensures that K(γ∞ ) = C − . But if we take
γ∞0 6= γ , K(γ 0 ) contains some lightlike geodesic by the remark above, and
∞ ∞
since any lightlike geodesic meets C − , we get a contradiction with the fact
that K(γ∞ ) and K(γ∞ ) have to be disjoint.
j −j
Now, let us perturb slightly the sets Uj0 0 and Uj0 0 into some sets Uj00 j0
and Uj00 −j0 , in order to get another fundamental domain D0 , very close to
0
D. Since it isS very near D,SD is included in some Fk for k sufficiently
0
large, and so γ∈Γ γ(D ) = γ∈Γ γ(D). We prove as above that the limit
of the compact sets γk (Uj00 j0 ) is still a connected component of Λ, and
consequentely of the form K(γ∞ 0 ). We just saw that some lightlike geodesics
j
of C − don’t meet V j00 , so that V̂j0 0 is not the whole Sn−2 . It is thus possible
j
j j
to choose U 0 j0 in such a way that some points of Vˆ0 0 are not in V̂ 0 . But
j0 j0 j0
then, K(γ∞ 0 ) and K(γ ) will be two different components, hence disjoint.
∞
j
On the other hand, since the intersection of Uj0 0 and Uj00 j0 is not empty
j 0 ) must have some common points. We
(∆j00 is inside), K(γ∞ ) and K(γ∞
thus get a contradiction.
It remains to deal with the case where (γk ) has mixed or balanced dis-
tortions. Once again, if x is a point of D then for all k ∈ N, γk (x) ∈ Ui11 and
−j1
γk −1 (x) ∈ Uj1 . Hence, the attracting circle ∆+ is in Ui11 and the repelling
−j j
one ∆− is in Uj1 1 . In particular U j00 does not meet ∆− . We infer from
j j
proposition 5 and proposition 3 that limk→∞ γk (U j00 ) ⊂ ∆+ , but since U j00
25
j
contains a lightlike geodesic, we have the equality limk→∞ γk (U j00 ) = ∆+ .
We finally obtain that K(γ∞ ) = ∆+ .
Now, we claim that the equality ΛΓ = Λ holds. Indeed, for any sequence
(γk ) of Γ tending simply to infinity, (γk ) tends to ∆+ (γk ). We thus see
that ΛΓ ⊂ Λ. Now, it is a general fact that if a group Γ acts properly co-
compactly on some open set Ω, it can’t act properly on some open set Ω0
strictly containing Ω. So, Ω can’t be strictly contained in Einn \ΛΓ and we
get ΛΓ = Λ.
26
For the proof of (iv), we refer to the theorem 5 of [Fr2] (in fact, in [Fr2],
we considered only particular cases of Schottky groups, but the proof of
theorem 5 includes the general case).
References
[A] B.N.Apanasov - Conformal geometry of discrete groups and
manifolds. de Gruyter Expositions in Mathematics, 32. Walter
de Gruyter and Co., Berlin, 2000.
27
[Fr2] C.Frances - Sur les variétés lorentziennes dont le groupe con-
forme est essentiel. (2003) . To appear in Mathematische Annalen.
28
[Ma] B.Maskit - Kleinian groups. Grundlehren der Mathematischen
Wissenschaften , 287. Springer-Verlag, Berlin, 1988.
Charles FRANCES
Laboratoire de Topologie et Dynamique
Université de Paris-Sud, Bât. 430
91405 ORSAY
FRANCE
email: Charles.Frances@math.u-psud.fr
29