Introduction to Hyperbolic Geometry

Julien Paupert
Spring 2016

1
Contents

1 Geometry of real and complex hyperbolic space 3

1.1 The hyperboloid model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Isometries of H n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Reflections . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Geodesics and totally geodesic subspaces . . . . . . . . . . . . . . . . . . . . . . 9
1.5 The ball model and upper half-space model . . . . . . . . . . . . . . . . . . . . . 10
1.5.1 The ball model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.2 The upper half-space model . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.5.3 Inversions and Möbius transformations . . . . . . . . . . . . . . . . . . . 11
1.5.4 Dimensions 2 and 3: SL(2, R) and SL(2, C) . . . . . . . . . . . . . . . . . 15
1.5.5 Geodesics, totally geodesic subspaces and metric . . . . . . . . . . . . . . 18
1.6 The Klein model and Hilbert metric . . . . . . . . . . . . . . . . . . . . . . . . . 19
1.7 Curvature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.8 Boundary at infinity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
1.9 Complex hyperbolic space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.10 Totally geodesic subspaces of HnC . . . . . . . . . . . . . . . . . . . . . . . . . . 28
1.11 Bisectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

2 Discrete subgroups and lattices in Isom(HnR ) and Isom(HnC ) 33

2.1 Discontinuous group actions and fundamental domains . . . . . . . . . . . . . . 33
2.2 Hyperbolic manifolds and orbifolds . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.3 Structure theorems for finitely generated linear groups and hyperbolic lattices . 40
2.3.1 Finitely generated linear groups . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.2 Discrete hyperbolic isometry groups . . . . . . . . . . . . . . . . . . . . . 41
2.3.3 Hyperbolic lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.4 Arithmetic lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5 Arithmeticity vs. non-arithmeticity . . . . . . . . . . . . . . . . . . . . . . . . . 47
2.6 Hyperbolic Coxeter groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.7 Some non-arithmetic lattices in Isom(H2C ) . . . . . . . . . . . . . . . . . . . . . . 52
2.7.1 Mostow’s lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7.2 Configuration spaces of symmetric complex reflection triangle groups . . 52
2.7.3 Sporadic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
2.7.4 Arithmeticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.7.5 Commensurability classes . . . . . . . . . . . . . . . . . . . . . . . . . . 55
2.7.6 Discreteness and fundamental domains . . . . . . . . . . . . . . . . . . . 56

2
Chapter 1

Geometry of real and complex

hyperbolic space

1.1 The hyperboloid model

Let n > 1 and consider a symmetric bilinear form of signature (n, 1) on the vector space Rn+1 ,
e. g. the standard Lorentzian form:

hX, Y i = x1 y1 + ... + xn yn − xn+1 yn+1 , (1.1.1)

where: X = (x1 , ..., xn+1 )T and Y = (y1 , ..., yn+1 )T .

Consider the following hypersurface H ⊂ Rn+1 , which we call the hyperboloid (more pre-
cisely, H is the upper sheet of the standard 2-sheeted hyperboloid in Rn+1 ):

H = X = (x1 , ..., xn+1 )T ∈ Rn+1 | hX, Xi = −1 and xn+1 > 0 .



Lemma 1.1.1 (1) H is a smooth (oriented) hypersurface.

(2) for any X ∈ H, TX H = X ⊥ = {Y ∈ Rn+1 | hX, Y i = 0}.
(3) H is (connected and) simply-connected.

Proof. (1) Let f : Rn+1 −→ R be the function defined by f (X) = hX, Xi. Then f is
everywhere differentiable and −1 is a regular value of f . Indeed, noting that f (X + Y ) =
f (X) + 2hX, Y i + hY, Y i, we see that the differential of f at X is dX f : Y 7→ 2hX, Y i (as the
term 2hX, Y i is linear in Y whereas hY, Y i is quadratic - we identify here the vector space Rn+1
with its tangent space at X). This has rank 1 everywhere except at X = 0.
(2) This also tells us that, for X ∈ H, TX H = Ker dX f = X ⊥ .
(3) The projection Rn+1 −→ Rn onto the n first coordinates induces a homeomorphism from
H to Rn , as the last coordinate of a point X = (x1 , ..., xn+1 )T ∈ H is given by the first n
coordinates via xn+1 = (1 + x21 + ... + x2n )1/2 . 

Note that, given X ∈ H, the restriction of the ambient bilinear form h., .i to X ⊥ is positive
definite (because hX, Xi < 0). Obviously this product varies smoothly with the point X, as h., .i
is a smooth function of 2 variables. In fancier terms, we say that h., .i induces a Riemannian

3
metric on H. We will denote H n the resulting Riemannian manifold. We will sometimes
(seldom) use the quadratic notation for this Riemannian metric:

ds2 = dx21 + ... + dx2n − dx2n+1 .

Recall that a Riemannian metric (an infinitesimal object) on a manifold M induces a (global)
distance function on M as follows, by integrating the metric along paths (like in Calculus). The
length of a path γ : [0, 1] −→ M (relative to ds) is:
Z 1
l(γ) := ds(γ 0 (t)).dt
0

(Strictly speaking, this only makes sense for paths for which this integral is well-defined; we
call such paths rectifiable). This defines a distance d on M by letting, for X, Y ∈ M :
d(X, Y ) = Inf(l(γ)), where the infemum is taken over all (rectifiable) paths γ with γ(0) = X
and γ(1) = Y .
We will see that there is in fact a unique geodesic segment connecting any 2 given points
X, Y ∈ H, i. e. a unique path realizing the infemum above. In fact, there is a very simple
and convenient formula to compute distances in H, which we state without proof for now
(we will prove this formula once we know more about the isometries and subspaces of H, see
Proposition 1.4.1).

Proposition 1.1.1 [Distance formula] Let X, Y ∈ H. Then d(X, Y ) = cosh−1 (−hX, Y i).

1.2 Isometries of H n
We start by recalling the definitions of isometries in the setting of metric spaces and Riemannian
manifolds.
A (metric) isometry from a metric space (X, dX ) to another metric space (Y, dY ) is a bijec-
tion f : X −→ Y preserving distances, meaning that:

dY (f (x), f (y)) = dX (x, y) for all x, y ∈ X.

Given two Riemannian manifolds (M, g) and (N, h) (where g, h denote the Riemannian
metrics, i.e. inner products on the tangent spaces), a (Riemannian) isometry from (M, g)
to (N, h) is a diffeomorphism f : M −→ N (i.e. a smooth bijection with smooth inverse)
preserving the Riemannian metrics, meaning that:

h(dx f (X), dx f (Y )) = g(X, Y ) for all x ∈ M and X, Y ∈ Tx M.

In our setting, we will use the word isometry to mean a Riemannian isometry; note that
these are in particular isometries for the associated distance functions. Given an (oriented)
Riemannian manifold M , we will denote Isom(M ) the group of isometries from M to itself,
and Isom+ (M ) the subgroup of orientation-preserving isometries of M . (These groups also
come with a topology, in general terms the so-called compact-open topology, but in the case
of hyperbolic spaces we will identify their isometry groups with matrix groups with their usual
topology).

4
 We now return to the hyperboloid. Denote O(n, 1) the group of linear transformations of
n+1
R preserving the ambient bilinear form. In symbols:
O(n, 1) = A ∈ GL(n + 1, R) | (∀X, Y ∈ Rn+1 ) hAX, AY i = hX, Y i .


Note that H and −H are the 2 connected components of {X ∈ Rn+1 | hX, Xi = −1}, so
that elements of O(n, 1) either preserve each of H and −H or exchange them. We denote by
O+ (n, 1) the index 2 subgroup of O(n, 1) consisting of those transformations which preserve H:
O+ (n, 1) = {A ∈ O(n, 1) | A(H) = H} .
Since the metric on H is defined entirely in terms of the form h., .i, we have:

Lemma 1.2.1 O+ (n, 1) ⊂ Isom(H n ) and SO+ (n, 1) ⊂ Isom+ (H n ).

We will show that these two inclusions are in fact equalities in Theorem 1.3.1, after discussing
reflections; for now this already gives us enough isometries of H n to show the following prop-
erties.

Proposition 1.2.1 (1) H n is a homogeneous and isotropic space. In other words, Isom(H n )
acts transitively on the unit tangent bundle U T H n : given any 2 points x, y ∈ H n and unit
tangent vectors u ∈ Tx H n , v ∈ Ty H n , there exists an isometry f of H n such that f (x) = y and
dx f (u) = v.
(2) The stabilizer in O(n, 1) of any point of H is isomorphic to O(n).
(3) H n is a (Riemannian) symmetric space: given any point x ∈ H n there exists an isometry
ix of H n such that ix (x) = x and dx ix = −Id.

Proof. (1) Let x ∈ H. Complete x to a (Lorentzian) orthonormal basis (v1 , ..., vn , x) and the
define Ax ∈ GL(n + 1, R) to be (the matrix in the standard basis of) the linear transformation
sending the standard basis (e1 , .., en+1 ) to (v1 , ..., vn , x). Then Ax ∈ O(n, 1) as both bases are
h., .i-orthonormal. Moreover, Ax (en+1 ) = x so A preserves H and Ax ∈ O+ (n, 1). Then, given
any 2 points x, y ∈ H, Ay A−1 + n
x ∈ O (n, 1) sends x to y. The claim that H is isotropic will
follow from (2).
(2) If A ∈ O+ (n, 1) fixes a point x ∈ H, it also preserves x⊥ (because it preserves the form
h., .i). In particular, the stabilizer of en+1 in O+ (n, 1) consists of all block matrices A of the
form:  
A
e 0
e ∈ O(n).
A= with A
0 1
Note that a such matrix A acts on Ten+1 H ' e⊥ n+1 = Span(e1 , ..., en ) by the usual action of A on
e
Rn , so den+1 A ' A.
e Since O(n) acts transitively on unit vectors in Rn , StabO+ (n,1) (en+1 ) acts
transitively on unit tangent vectors in Ten+1 H. The same is then true at any other point of H:
since O+ (n, 1) acts transitively on H, the stabilizer of any point is conjugate to the stabilizer
of en+1 .
(3) Define ien+1 ∈ O+ (n, 1) in block form as above with A e = −Id. Then: ien+1 (en+1 ) = en+1
and den+1 ien+1 = −Id. Now, given any point x ∈ H and Ax ∈ O+ (n, 1) defined as above,
ix = Ax ien+1 A−1
x satisfies: ix (x) = x and dx ix = −Id. 

5
1.3 Reflections
We start with a digression on general bilinear algebra. Let V be a (finite-dimensional) real vector
space and h., .i a non-degenerate symmetric bilinear form (this means that V ⊥ = {0}: the only
vector perpendicular to all vectors is 0). Let W ⊂ V be a subspace such that V = W ⊕ W ⊥
(i.e. W ∩ W ⊥ = {0}). Attached to such a subspace W are 2 linear maps denoted pW (the
orthogonal projection onto W ) and rW (the ”reflection” across W), defined as follows. (We
write ”reflection” with quotes because we will reserve that term for the case where W has
codimension 1). For any v ∈ V , pW (v) is defined as the unique w ∈ W such that v − w ∈ W ⊥ ,
and:
rW (v) = 2pW (v) − v. (1.3.1)

Lemma 1.3.1 (a) pW (v) = v ⇐⇒ v ∈ W , and: pW (v) = 0 ⇐⇒ v ∈ W ⊥ .

(b) rW (v) = v ⇐⇒ v ∈ W , and: rW (v) = 0 ⇐⇒ v = 0 (i.e. rW ∈ GL(V )).
(c) rW has order 2, i.e. rW (rW (v)) = v for all v ∈ V .
(d) rW ∈ O(h., .i), i.e. hrW (u), rW (v)i = hu, vi for all u, v ∈ V .

Proof. (a) follows from the definition of pW .

(b) From (a), if v ∈ W then rW (v) = 2v − v = v. Conversely, if v ∈ V satisfies rW (v) = v then
2pW (v) = 2v so v = pW (v) ∈ W .
(c) For any v ∈ V :

rW (rW (v)) = 2pW (rW (v)) − rW (v)

= 2pW (pW (v) − v) − (2pW (v) − v)
= 4pW (v) − 2pW (v) − 2pW (v) + v
= v.

(d) Recall that pW (v) ⊥ (v − pW (v)), i.e. hpW (v), vi = hpW (v), pW (v)i. Then:

hrW (v), rW (v)i = h2pW (v) − v, 2pW (v) − vi

= 4hpW (v), pW (v)i − 4hpW (v), vi + hv, vi
= hv, vi.

The result follows by polarization (a symmetric bilinear form B(u, v) is determined by its
associated quadratic form q(v) = B(v, v)). 

The case that will be of particular interest to us is when W = e⊥ for some e ∈ V with
he, ei 6= 0. Then V = Re ⊕ e⊥ , and re⊥ is called the reflection across the hyperplane e⊥
(sometimes called ”parallel to e”). In that case there is a simple explicit formula for re⊥ which
we will use later:

Lemma 1.3.2 Let e ∈ V such that he, ei =

6 0. Then, for any v ∈ V :

hv, ei
re⊥ (v) = v − 2 e. (1.3.2)
he, ei

6
 hv,ei
Proof. Denote W = e⊥ , and let w = v − he,ei
e. Then w ∈ W , because:

hv, ei hv, ei
hw, ei = hv − e, ei = hv, ei − he, ei = 0.
he, ei he, ei

Moreover, v − w ∈ Re = W ⊥ . Therefore w = pW (v), and rW (v) = 2w − v. 

Proposition 1.3.1 Let V be a finite-dimensional real vector space and h., .i a non-degenerate
symmetric bilinear form on V . Then O(V, h., .i) is generated by reflections.

Proof. By induction on the dimension of V .

Let V be a real vector space of dimension n + 1, h., .i a non-degenerate symmetric bilinear form
on V , A ∈ O(V, h., .i) and v ∈ V satisfying hv, vi =
6 0.

• First assume that hAv −v, Av −vi = 6 0, and consider the reflection ρ across the hyperplane
⊥
(Av − v) . Note that v = 2 (Av + v) − 12 (Av − v) with h(Av + v), (Av − v)i = 0, therefore
1
1
2
(Av + v) = p(Av−v)⊥ (v) and ρ(v) = 2p(Av−v)⊥ (v) − v = Av.
Then ρ ◦ A(v) = v, so ρ ◦ A preserves v ⊥ , and by the induction hypothesis (ρ ◦ A)|v⊥ is a
product of reflections in v ⊥ , say (ρ◦A)|v⊥ = r10 ◦...◦rk0 . Extending each ri0 to a reflection ri
in O(V, h., .i), we get that ρ◦A and r1 ◦...◦rk both preserve the decomposition V = Rv⊕v ⊥
(as they are both in O(V, h., .i)) and agree on v ⊥ , therefore they are equal (again, because
they preserve h., .i). Thus A = ρ ◦ r1 ◦ ... ◦ rk .

• Now, if by bad luck hAv − v, Av − vi = 0, we replace A by −A in the preceding argument.

Indeed, noting that hAv, vi = hv, vi, we have:

h−Av − v, −Av − vi = hAv, Avi + 2hAv, vi + hv, vi = 4hv, vi =

6 0.
Therefore, by the preceding argument, −A is a product of reflections, and so is A because
−Id is always a product a reflections by the following:

Lemma 1.3.3 If v1 , ..., vn+1 is an orthonormal basis of V then rv1⊥ ◦ ... ◦ rvn+1
⊥ = −Id.

Recall that orthonormal means that hvi , vj i = 0 if i =

6 j and ±1 if i = j (with the number of
+1’s and −1’s determined by the signature of h., .i). The lemma follows by noting that:

hvi , vj i
rv⊥j (vi ) = vi − 2 vj = vi if i 6= j and − vi if i = j.
hvj , vj i
This concludes the proof of the proposition. 

We now return to the case of the hyperboloid, with V = Rn+1 and h., .i the standard
Lorentzian form (1.1.1).

Lemma 1.3.4 Given e ∈ Rn+1 with he, ei =

6 0, the reflection re⊥ preserves H if he, ei > 0, and
exchanges H and −H if he, ei < 0.

7
 Proof. Recall that, by lemma 1.3.1 (d), re⊥ is in O(n, 1) so it either preserves each of H and
−H or exchanges them. The lemma follows by noting that e⊥ intersects H ∪ −H if and only
if he, ei > 0 (as the restriction of h., .i to e⊥ has signature (n − 1, 1) in that case, as opposed to
(n, 0)). 

Proposition 1.3.2 O+ (n, 1) is generated by reflections.

Note that this means that O+ (n, 1) is generated by the reflections it contains, i.e. by the
reflections of the form re⊥ with he, ei > 0.
Proof. Let A ∈ O+ (n, 1). Since O(n, 1) is generated by reflections, A = ρ1 ◦ ... ◦ ρk where
each ρi is a reflection in O(n, 1), i.e. ρi = re⊥i for some ei ∈ Rn+1 with hei , ei i 6= 0 (but not
necessarily hei , ei i > 0). For each ρi = re⊥i which is in O(n, 1) but not in O+ (n, 1), we replace
ρi with a product of reflections which are each in O+ (n, 1) as follows. Since hei , ei i < 0, we
can complete ei to an orthonormal basis (ei , v1 , ..., vn ) with hvj , vj i > 0. Then each σj = rv⊥j
is a reflection in O+ (n, 1), and from lemma 1.3.3 we have that: ρi ◦ σ1 ◦ ... ◦ σn = −Id, so
ρi = −σ1 ◦ ... ◦ σn . After replacing each such ρi in this way, we get A = ±(ρ01 ◦ ... ◦ ρ0l ) where
each ρ0j is a reflection in O+ (n, 1). Since A and each ρ0j preserve H, the sign must be a +. 

Theorem 1.3.1 Isom(H n ) = O+ (n, 1) and Isom+ (H n ) = SO+ (n, 1). In particular, Isom(H n )
is generated by reflections.

We will prove this using the following fact that a Riemannian (local) isometry is entirely
determined by the image of a single point and its differential at that point:

Proposition 1.3.3 Let M, N be two Riemannian manifolds of the same dimension, with M
connected. If φ1 : M −→ N and φ2 : M −→ N are two local isometries such that φ1 (x) = φ2 (x)
and dx φ1 = dx φ2 for some x ∈ M , then φ1 = φ2 .

Proof. Let S = {y ∈ M | φ1 (y) = φ2 (y) and dy φ1 = dy φ2 }. Then S is closed in M, and

non-empty by assumption. We now prove that S is also open in M , which implies that S = M
because M is connected.
Let x ∈ S. Imφ1 and Imφ2 are open in N because M, N have the same dimension and
φ1 , φ2 are local diffeomorphisms. Therefore there exist neighborhoods U1 , U2 of x in M and V
of φ1 (x) = φ2 (x) in N such that φi : Ui −→ V is bijective (and an isometry) for i = 1, 2. We use
the existence of of a normal neighborhood U of x in M , which is a neighborhood such that the
restriction of the exponential map to some neighborhood W of 0 in Tx M is a diffeomorphism
W −→ U . (Recall that the exponential map expx at a point x ∈ M is defined as the map:

expx : Tx M −→ M
v 7−→ γv (1)

where γv is the (unique) geodesic arc with γv (0) = x and γv0 (0) = v.) By shrinking U if necessary
we may assume that U ⊂ U1 . Consider then f = (φ2|U2 )−1 ◦ φ1|U1 . Then f is an isometry from
U1 onto U2 , f (x) = x and dx f = Id. Therefore, for any geodesic arc γ starting at x, f ◦ γ is

8
also a geodesic arc starting at x, with tangent vector (f ◦ γ)0 (0) = dx f (γ 0 (0)) = γ 0 (0). Thus
f ◦ γ = γ; since this holds for any geodesic arc γ starting at x we have: f ◦ (expx|W ) = expx|W ,
hence f|U = Id|U (as expx|W : W −→ U is a diffeomorphism). Therefore: φ1|U = φ2|U (in
particular, dp φ1 = dp φ2 for any p ∈ U ), so that U ⊂ S. Therefore S is open in M . 

Proof of Theorem 1.3.1. Let ϕ ∈ Isom(H n ) and let x ∈ H n . Consider the linear map
A ∈ GL(n + 1, R) given in block form by:
 
dx ϕ 0
A=
0 1
in the decompositions x⊥ ⊕ Rx (at the source) and ϕ(x)⊥ ⊕ Rϕ(x) (at the range) of Rn+1 . Since
dx ϕ sends h., .i|x⊥ ×x⊥ to h., .i|ϕ(x)⊥ ×ϕ(x)⊥ (as ϕ is an isometry) and hϕ(x), ϕ(x)i = −1 = hx, xi,
A preserves the form h., .i so A ∈ O(n, 1). In fact A ∈ O+ (n, 1) as A(x) = ϕ(x). Therefore A
is an isometry of H n , with A(x) = ϕ(x) and dx A = dx ϕ, so by the previous result A = ϕ. 

1.4 Geodesics and totally geodesic subspaces

We now determine the geodesics and totally geodesic subspaces of H n in the hyperboloid
model. Recall that a smooth parametrized curve γ(t) in a Riemannian manifold (M, g) is
geodesic ifR it is locally length-minimizing, i.e. if for all close enough x = γ(t1 ) and y = γ(t2 ),
t
d(x, y) = t12 g(γ 0 (t)).dt.
Proposition 1.4.1 Let x, y be two distinct points in H.
(a) There exists a unique geodesic segment [xy] connecting x and y; it is contained in the (bi-
infinite) geodesic line H ∩ Span(x, y).
(b) Given a unit tangent vector v ∈ Tx H, the unit-speed parametrization of the unique geodesic
γ satisfying γ(0) = x and γ 0 (0) = v is given by:
γ(t) = cosh(t).x + sinh(t).v (t ∈ R). (1.4.1)
(c) Distance formula: cosh(d(x, y)) = −hx, yi.
Proof. (a) Denote (xy) = H ∩ Span(x, y). As isometries of H n are restrictions to H of linear
maps of Rn+1 , any isometry fixing x and y fixes (xy) pointwise. As (xy) is a 1-dimensional
submanifold of H, it is the unique geodesic line containing x and y. (Here we use the fact that
geodesics always exist locally).
(b) The same argument as above tells us that, as a set, γ(R) = H ∩ Span(x, v). Since H n is
homogeneous and isotropic, we may normalize by an isometry so that: x = en+1 and v = en .
Then a parametrization of H ∩ Span(x, v) is: γ(t) = cosh(t).en+1 + sinh(t).en ; this satisfies
γ(0) = x and γ 0 (0) = v. Moreover γ 0 (t) = sinh(t).en+1 + cosh(t).en , so hγ 0 (t), γ 0 (t)i = cosh2 (t) −
sinh2 (t) = 1.
(c) Normalize again so that x = en+1 and y ∈ γ as above, so that y = cosh(T )en+1 + sinh(T )en
RT
for some T > 0. Then: d(x, y) = 0 hγ 0 (t), γ 0 (t)i.dt = T and hx, yi = −cosh(T ). 

As a corollary of (a) and the Hopf-Rinow theorem we get:

9
Corollary 1.4.1 H n is complete.

Recall that a submanifold N of a Riemannian manifold M is totally geodesic if any geodesic

ray starting in N and tangent to N remains in N , in other words if for any geodesic arc
γ : γ(0) = x ∈ N and γ 0 (0) ∈ Tx N ⇒ (∀t) γ(t) ∈ N . If M is uniquely geodesic (i.e. any 2
points in M are connected by a unique geodesic segment), then this is equivalent to: for any
x, y ∈ N , the geodesic segment [xy] is contained in N .
A (k + 1)-dimensional linear subspace W of Rn+1 is called hyperbolic (resp. parabolic, resp.
elliptic) if the restriction to W of the form h., .i has signature (k, 1) (resp. (k, 0), resp (k +1, 0)).

Proposition 1.4.2 Let N be a totally geodesic subspace of H n .

(a) N is of the form H ∩ W for some hyperbolic linear subspace W of Rn+1 .
(b) In particular, N is an isometrically embedded copy of H k in H n (with k = dim N ).
(c) N is the fixed-point locus of an isometry of order 2.

Proof. (a) follows from Proposition 1.4.1 (a).

(b) follows from the construction of the hyperboloid model.
(c) Take rW as defined in Section 1.3; the fact that W is hyperbolic ensures that rW preserves
H. 

1.5 The ball model and upper half-space model

1.5.1 The ball model
Denote Rn+1,+ = {(x1 , ..., xn+1 ) ∈ Rn+1 | xn+1 > 0}, and let π : Rn+1,+ −→ Rn × {0} denote
radial projection from the point −en+1 . Explicitly, in coordinates:

(x1 , ..., xn , 0)
π(x1 , ..., xn+1 ) = .
1 + xn+1

Lemma 1.5.1 π induces a diffeomorphism H −→ B, where B ' B × {0} is the open unit ball
in Rn .

Proof. We can write the inverse map g explicitly as g(y1 , ..., yn ) = (λy1 , ..., λyn , λ − 1) with
λ chosen so that the latter point lies in H, namely λ = 2/(1 − y12 − ... − yn2 ). 

We then carry over the metric on H to a metric on B by this diffeomorphism (more for-
mally, we consider the pullback of the metric on H by π −1 ), and denote B n the corresponding
Riemannian manifold, which is by construction isometric to H n .

1.5.2 The upper half-space model

√ √
Let S = S(−en , 2) ∈ Rn be the sphere centered at −en with radius 2, and consider the
inversion i = iS in S (see below if you don’t remember what an inversion is in Rn ). Explicitly,
in coordinates:

10
 i : Rn \ {−en } −→ Rn
x+en
x 7−→ 2 ||x+e n ||
2 − en ,

where ||.|| denotes the Euclidean norm on Rn .

Lemma 1.5.2 i induces a diffeomorphism from the (open) unit ball B to the upper half-space
U = {(x1 , ..., xn ) ∈ Rn | xn > 0}.
Proof. This follows from general properties of inversions (see below), by noting that the unit
sphere ∂B is a sphere containing the center −en of S, so i(∂B) is a hyperplane, which must be
Rn−1 × {0} (as the latter contains the (n − 2)-sphere ∂B ∩ S). The image of B is then one of
the 2 half-spaces bounded by Rn−1 × {0}, which must be U as i(0) = en . 

Again, we carry over the metric on B to a metric on U by this diffeomorphism (more formally,
we consider the pullback of the metric on B by i−1 = i), and denote U n the corresponding
Riemannian manifold, which is by construction also isometric to H n .

1.5.3 Inversions and Möbius transformations

The isomeries in these 2 models are best described using inversions, so we now recall their
definition and main properties.
Given a center x0 ∈ Rn and a radius R > 0, the inversion iS in the sphere S = S(x0 , R) -
also denoted i(x0 ,R) - is the map:

iS : Rn \ {x0 } −→ Rn
R2 (1.5.1)
x 7−→ x0 + ||x−x0 ||2
(x − x0 ),
where ||.|| denotes the Euclidean norm on Rn . We extend iS to the one-point compactification
cn = Rn ∪ {∞} by setting iS (∞) = x0 and iS (x0 ) = ∞. For the convenience of statements
R
relating to inversions, we define a generalized sphere in R cn to be either a (top-dimensional)
n
Euclidean sphere in R or a hyperplane (affine, i.e. not necessarily containing the origin). Note
that a generalized sphere is a hyperplane if and only if it contains ∞. The following proposition
summarizes the main geometric properties of inversions:
Proposition 1.5.1 Let S = S(x0 , R) be a sphere and i = iS its inversion.
(a) iS is an involution, fixing S pointwise and exchanging interior and exterior of S.
(b) iS induces a conformal diffeomorphism Rn \ {x0 } −→ Rn \ {x0 }.
(c) If S 0 is a generalized sphere, then i(S 0 ) is also a generalized sphere.
(d) If S 0 is a generalized sphere, then: i(S 0 ) = S 0 ⇐⇒ S = S 0 or S ⊥ S 0 .
Recall that a smooth map f : U −→ Rn (U ⊂ Rn open) is called conformal if it preserves
angles; more precisely this means that its differential at each point preserves dot-products up
to a muliplicative factor:
(∀x ∈ U )(∃K(x) ∈ R)(∀u, v ∈ Tx U ) dx f (u) · dx f (v) = K(x) u · v.
We will use the following lemma, describing the composite of 2 inversions with common
center; D(x0 , R) denotes the dilation with center x0 and ratio R:

11
 R2
Lemma 1.5.3 If i1 = i(x0 ,R1 ) and i2 = i(x0 ,R2 ) then i2 ◦ i1 = D(x0 , R22 ).
1

 
R12 R22
Proof of lemma. i2 ◦ i1 (x) = i2 x0 + ||x−x0 ||2
(x − x0 ) = R12
(x − x0 ) + x0 . 

Proof of proposition. (a) By the lemma, iS is an involution. The second statement follows
by noting that: ||x − x0 ||2 < R2 ⇐⇒ ||i(x) − x0 ||2 > R2 (likewise with the 2 inequalities
reversed, respectively replaced by equalities).
(b) Note that iS = Tx0 ◦ D(0, R2 ) ◦ i0,1 ◦ Tx−1
0
, where Tx0 denotes translation by x0 . Translations
and dilations are conformal, therefore it suffices to show that the inversion i(0,1) in the unit
sphere is conformal. Let x ∈ Rn \ {0} and u ∈ Rn \ {x}. We claim that:
1
dx i(0,1) (u) = r ⊥ (u),
||x||2 x
which gives the result as reflections are conformal. In order to see this, note that:
1 1 x·u
= − 2 + o(||u||),
||x + u||2 ||x||2 ||x||4
1 x·u
because the derivative of the function x 7−→ ||x||2
is u 7−→ −2 ||x|| 4 . Therefore:

x+u x+u x·u(x+u)

i(0,1) (x + u) = ||x+u||2
= ||x|| 2 − 2 ||x||4
+ o(||u||)
x u x·ux
= ||x||2 + ||x||2 − 2 ||x||4 + o(||u||)
1
= i(0,1) (x) + ||x|| 2 rx⊥ (u) + o(||u||).

This proves the claim. Note that in general, it follows by composing with the appropriate
translations and dilation that:
R2
dx i(x0 ,R) = r
||x − x0 ||2
for some (Euclidean) reflection r.
(c) Note that this statement contains 4 statements about Euclidean subspaces, by our conven-
tions regarding generalized spheres and the point ∞. Namely, given a hyperplane H and a
sphere S 0 :
(1) if x0 ∈ H then i(H) = H.

/ H then i(H) = S 00 for some sphere S 00 with x0 ∈ S 00 .

(2) if x0 ∈

(3) if x0 ∈ S 0 then i(S 0 ) = H 0 for some hyperplane H 0 with x0 ∈

/ H 0.

/ S 0 then i(S 0 ) = S 00 for some sphere S 00 with x0 ∈

(4) if x0 ∈ / S 00 .
We now prove these four statements separately.
(1) follows from the definition, as x0 , x and i(x) are collinear for all x.
For the remaining statements we assume that i = i(0,1) (as the results are invariant under
translations and dilations).
(2) Let H be a hyperplane not through 0, which we write in the form H = h + h⊥ (where

12
 h 1
h ∈ Rn \ {0} is the orthogonal projection of 0 onto H). Let C = 2||h|| 2 and R = 2||h||
. Then, for
x 6= 0:
i(x) ∈ S(C, R) ⇐⇒ ||i(x) − C||2 = R2
 2
 x h  1
⇐⇒  ||x|| 2 − 2||h||2 
= 4||h|| 2
1 x·h
⇐⇒ ||x||2 − ||x||2 ||h||2 = 0
⇐⇒ h · h − x · h = 0
⇐⇒ x ∈ H = h + h⊥ .
Therefore i(H) = S(C, R), a sphere containing 0.
(3) follows from (2) as i is an involution.
(4) Let S = S(C, R) be a sphere not containing 0 (i.e. ||C||2 6= R2 ). Then, for x 6= 0:
 2
 x
i(x) ∈ S(C, R) ⇐⇒  ||x||2 − C  = R2

1 x·C 2 2
⇐⇒ ||x|| 2 − 2 ||x||2 + ||C|| = R

⇐⇒ ||C||21−R2 − 2 ||C||x·C 2
2
2 + ||x|| = 0
 2 −R 2
⇐⇒ x − ||C||C2 −R2  = (||C||||C|| 1 R2
2 −R2 )2 − ||C||2 −R2 = (||C||2 −R2 )2
 
 
⇐⇒ x ∈ S ||C||C2 −R2 , ||C||R2 −R2 .

||x|| 2
(in the third step we muliplied both sides by ||C|| 2 −R2 ). The latter is a sphere not through 0,

proving the statement. Note that the center of the image sphere is not the image of the center.

(d) First assume that iS (S 0 ) = S 0 . Then S ∩ S 0 6= ∅ (or else S 0 would be entirely in-
side/outside of S, and i(S 0 ) on the opposite side). Let x ∈ S ∩ S 0 , and let ~nS , ~nS 0 denote normal
vectors at x to S, S 0 respectively. Then dx i(~nS ) = −~nS , and dx i(~nS 0 ) = λ~nS 0 for some λ 6= 0,
because i fixes x and preserves S 0 . In fact, λ = ±1 because dx i is a reflection, with eigenspaces
R~nS corresponding to −1 and ~n⊥ S corresponding to 1. If λ = −1 then ~ nS 0 is collinear to ~nS so
S, S 0 are tangent and S = S 0 (again, or else S 0 would be entirely inside/outside S). If λ = 1
then ~nS ⊥ ~nS 0 and S ⊥ S 0 .
Conversely, assume that S ⊥ S 0 . If S 0 is a hyperplane then S 0 contains the center of S so
i(S 0 ) = S 0 by (a). If S 0 is a sphere, denote C its center and let x ∈ S ∩ S 0 . Then i(S 0 ) is a
sphere, centered on the line (Cx0 ) (by symmetry), containing x (which is not on (Cx0 )) and
with normal vector −~nS 0 , so i(S 0 ) = S 0 . 

We are now ready to characterize isometries of the ball model and upper half-space model
in terms of inversions. A Möbius transformation of R cn is a product of inversions and reflections.
(From now on we will use the word ”inversion” to mean generalized inversion, i.e. inversion or
reflection). We denote Möb(R cn ) the group of all Möbius transformations of R cn , and for any
subset V ⊂ R cn , Möb(V ) the group of all Möbius transformations preserving V . For example, by
part (d) of the previous proposition, Möb(B) is generated by inversions in generalized spheres
S 0 such that S 0 ⊥ ∂B, and likewise Möb(U ) is generated by inversions in generalized spheres
S 0 such that S 0 ⊥ ∂U .

Theorem 1.5.1 Isom(B n ) = Möb(B) and Isom(U n ) = Möb(U ).

13
Recall that we passed from the hyperboloid model H to the ball model B via the radial pro-
jection:

π: H −→ B
(x1 , ..., xn+1 ) 7−→ (x1+x
1 ,...,xn )
n+1
.
and recall that Isom(H n ) = O+ (n, 1) is generated by reflections. The first part of the theorem
then follows from the following:

Proposition 1.5.2 If r ∈ O+ (n, 1) is a reflection, then ρ = π ◦ r ◦ π −1 is (the restriction to B

of ) an inversion.

The second part of the theorem then follows by noting that we passed from the ball model B
to the upper half-space model U by a reflection, which implies that Isom(B n ) and Isom(U n )
are conjugate inside Möb(R cn ).
n+1
Proof of proposition 1.5.2. We will use coordinates on
 R of the form  (X, t), with X ∈ Rn
2
X
and t ∈ R, so that: π : (X, t) 7−→ 1+t and π −1 : Y 7−→ 1−||Y
2Y
, 1+||Y || = (Y 0 , u).
||2 1−||Y ||2

Recall that a reflection r ∈ O+ (n, 1) is of the form: rv⊥ : x 7−→ x − 2 hx,vi

hv,vi
v for some v ∈ Rn+1
with hv, vi > 0. Normalizing v so that hv, vi = 1, this becomes simply: rv⊥ (x) = x − 2hx, viv.
With coordinates x = (X, t) as above, and v = (V, s) (normalized with hv, vi = 1 = V · V − s2 ),
this becomes: r(X, t) = (X, t) − 2(X · V − st)(V, s).
Let ρ = π ◦ r ◦ π −1 . Then:

ρ(Y ) = π ◦ r(Y 0 , u) = π ((Y 0 , u) − 2(Y 0 · V − su)(V, s))

= π ((Y 0 − 2(Y 0 · V − su)V, u − 2(Y 0 · V − su)s))
Y 0 −2(Y 0 ·V −su)V
= 1+u−2(Y 0 ·V −su)s
Y −(2Y ·V −s(1+Y ·Y ))V
= 1−(2Y ·V −s(1+Y ·Y ))s

where we factored out u = 1−Y2 ·Y from the numerator and denominator in the last step. We
now simplify the numerator and denominator separately. First the numerator:

Y − (2Y · V − s(1 + Y · Y ))V = Y − V /s + (V /s − 2(Y · V )V + sV + s(Y · Y )V )

= Y − V /s + s ||Y − V /s||2 V

V ·V 1+s2 1
because: s
= s
= s
+ s. Now the denominator:

1 − (2Y · V − s(1 + Y · Y ))s = 1 − 2sY · V + s2 (1 + Y · Y ) = s2 ||Y − V /s||2 .

Y −V /s
Putting the pieces back together gives: ρ(Y ) = V /s+ s2 ||Y −V /s||2
= iS (Y ) where S = S (V /s, 1/s).
Note that when s = 0, ρ is in fact a reflection (across a hyperplane containing 0 ∈ B). 

14
1.5.4 Dimensions 2 and 3: SL(2, R) and SL(2, C)
We’ve seen that, in all dimensions, the group Isom(H n ) can be identified with the matrix group
O+ (n, 1) < GL(n + 1, R). In dimensions 2 and 3 we also have the following ”coincidences”:

Proposition 1.5.3 Isom+ (U 2 ) ' PSL(2, R) and Isom+ (U 3 ) ' PSL(2, C).

Recall that PSL(2, R) = SL(2, R)/{±Id} and PSL(2, C) = SL(2, C)/{±Id}. We first show the
following:

Proposition 1.5.4 Möb+ (R

c2 ) ' PSL(2, C).

We identify R2 with C and consider the action of SL(2, C) on C by fractional linear transfor-
mations:
 
a b
g= ∈ SL(2, C)
c d
acts on C by z 7→ g(z) = az+b
cz+d
. This action is extended to Cb = C ∪ {∞} in the obvious way
(i.e. letting g(∞) = ac and g(− dc = ∞)). Note that ±Id are the only elements acting trivially
on C, therefore this gives a faithful action of PSL(2, C) (allowing us to identify PSL(2, C) with
a subgroup of the bijections of C).
b
Proposition 1.5.4 follows from the following lemma, which shows that PSL(2, C) corresponds
under this action to the group generated by products of an even number of inversions of R c2 ,
+ c
namely Möb (R2 ).

Lemma 1.5.4 (a) PSL(2, C) acts transitively on generalized circles in C.

b
−1
(b) If g ∈ PSL(2, C) and i is an inversion, gig is also an inversion.
(c) Any element of PSL(2, C) is a product of 2 or 4 inversions.
(d) Any product of 2 inversions is in PSL(2, C).

Proof. (a) First note that PSL(2, C) contains:

• all dilations Dr : z 7−→ rz (with r > 0), taking as matrix representative:

 √ 
r 0√
Dr =
0 1/ r

• all translations Tv : z 7−→ z + v, taking as matrix representative:

 
1 v
Tv =
0 1

• all rotations Rθ : z 7−→ eiθ z, taking as matrix representative:

 iθ/2 
e 0
Rθ =
0 e−iθ/2

15
 • and the involution I : z 7−→ −1/z, taking as matrix representative:
 
0 −1
I=
1 0
.
Now the group generated by all dilations, translations and rotations acts transitively on the
set of circles (resp. lines) in C. Moreover, I exchanges the line {Im z = 1} and the circle
C(i/2, 1/2), because for x ∈ R:

 2
1 x−i  i  1 2 1
I(x + i) = − =− 2 and I(x + i) − = ||i − x|| = .
x+i x +1 2   4(x2 + 1)2 2
z
(b) First consider i0 = i(0,1) : z 7−→ |z|2
= z̄1 , and let:
 
a b
g= ∈ SL(2, C)
c d

be arbitrary. We claim that gi0 g −1 is an inversion. To see this, we compute:

 
gi0 g −1 (z) = gi0 −cz+a dz−b
= g −c̄z̄+ā


¯ b̄
dz̄−
a( −c̄z̄+ā )+b ¯
2 +bdz̄−|b| 2
= −ac̄z̄+|a| ez̄+f
¯ b̄
dz̄−
= c −c̄z̄+ā 2 z̄+āc+|d|2 z̄−b̄d = hz̄−ē
¯ b̄ )
( dz̄− +d −|c|

with e = bd¯ − ac̄, f = |a|2 − |b|2 and h = |d|2 − |c|2 . If h = 0 then gi0 g −1 is a reflection across
2 1/2
 
a line; if h 6= 0 let z0 = he and R = fh + |e| h 2 . Then:

gi0 g −1 (z) = ez̄+f

hz̄−ē
= ez̄/h+f
z̄−ē/h
/h

z0 z̄−|z0 |2 +R2 R2
= z̄−z¯0
= z0 + |z−z0 |2
(z − z0 ) = i(z0 ,R) (z).
Now, if i is any inversion then i = hi0 h−1 with h = T D ∈ SL(2, C) for some translation T
and dilation D (because translations and dilations act transitively on circles). Therefore, for
any g ∈ SL(2, C), gig −1 = ghi0 h−1 g −1 is an inversion by the previous computation.

(c) From linear algebra we know that any g ∈ SL(2, C) is conjugate to a matrix h of the
form:  
λ 0
D λ2 = for some λ ∈ C∗ ,
0 1/λ
if it is diagonalizable, or to:  
1 1
P = T1 =
0 1
if not. Note that, writing λ = reiθ with r > 0, we have: Dλ2 = Dr2 R2θ . We know that R2θ
and T1 are products of 2 reflections across lines, and that Dr2 is a product of 2 (concentric)
inversions. Therefore h is a product of 2 or 4 inversions, and by conjugation (using part (b)),
so is g.

16
(d) Let i1 , i2 be 2 (generalized) inversions. If they are both reflections across lines then their
product is a translation or a rotation, so lies in PSL(2, C). If not, by (a) we may assume that
i1 = i(0,1) , i.e. i1 : z 7−→ 1/z̄. If i2 is a reflection across a line we may write it as: i2 : z 7−→ az̄ +b
(with a 6= 0). Then:  
bz + a b a
i2 ◦ i1 (z) = a/z + b = = z
z 1 0
and rescaling this matrix (by 1/a1/2 ) identifies i2 ◦ i1 with an element of PSL(2, C). If i2 is
R2
(strictly speaking) an inversion, we may write it as: z 7→ z0 + z̄− z¯0
for some z0 ∈ C and R > 0.
Then:
R2 R2 z (R2 − |z0 |2 )z + z0
 2 
R − |z0 |2 z0
i2 ◦ i1 (z) = z0 + = z0 + = = z
1/z − z¯0 −z¯0 z + 1 −z¯0 z + 1 −z¯0 1
and rescaling this matrix (by 1/R) identifies i2 ◦ i1 with an element of PSL(2, C). 

Poincaré extension: This concludes the proof of Proposition 1.5.4, namely the fact
that: Möb+ (R
c2 ) ' PSL(2, C). The second statement of Proposition 1.5.3, which stated that
+ 3
Isom (U ) ' PSL(2, C), follows from the following general observation:

Proposition 1.5.5 For all n > 1, Isom(U n+1 ) ' Möb(R

cn ).

Proof. As the intersection of any generalized sphere in Rn+1 perpendicular to Rn × {0} is a

generalized sphere in Rn × {0}, by taking restrictions we get a map from the set of all inversions
of Rn+1 preserving U n+1 to the set of inversions of Rn × {0}. This map is in fact a bijection,
because each generalized sphere in Rn × {0} is the trace of a unique generalized (half-) sphere
in U n+1 , perpendicular to Rn × {0}. Taking the group generated by the inversions on both
sides gives the result. 

The process of extending an element of Möb(R cn ) to an isometry of U n+1 is called Poincaré

extension.
Remark: We have seen that inversions are conformal; therefore, denoting Conf(R cn ) the
group of all conformal transformations of Rcn , we have: Möb(R cn ) ⊂ Conf(Rcn ). It can be shown
in fact that this is an equality (we won’t prove this; see for example chapter 1 of [BP]). Then
the above correspondence can be stated in the following way:
• every isometry of U n acts as a conformal map on ∂U n , and
• every conformal map of ∂U n extends to a (unique) isometry of U n .
(The same statements hold for the other models of H n , and in fact they can be made without
refernce to a specific model once we have an intrinsic description of the boundary ∂H n .)
Finally, the first half of Proposition 1.5.3, stating that Isom+ (U 2 ) ' PSL(2, R), follows by
noting that SL(2, R) is the stabilizer of the real axis in SL(2, C) (acting again on C by fractional
linear transformations).
Classification of isometries: In the course of the above proofs we have come across
different types of conjugacy classes in SL(2, C), which have different geometric and dynamical
properties and are accordingly separated into 3 types. An element g ∈ PSL(2, C) \ {Id} is
called:

17
  
eiθ 0
• ellipitic if it is conjugate to for some θ,
0 e−iθ
 
1 1
• parabolic if it is conjugate to , and
0 1
 
λ 0
• loxodromic if it is conjugate to for some λ ∈ C with |λ| =
6 1.
0 λ
(after possibly muliplying g by −Id). As a special case, if g is loxodromic with λ ∈ R, it is
called hyperbolic. Note that these 3 cases can be differentiated by traces, namely:

• g is elliptic ⇐⇒ Tr(g) ∈ (−2, 2),

• g is parabolic ⇐⇒ Tr(g) = ±2, and

• g is loxodromic ⇐⇒ Tr(g) ∈
/ [−2, 2] (and hyperbolic ⇐⇒ Tr(g) ∈ R \ [−2, 2]).

The more general definition (which we will see later, in all dimensions and for K = R, C or
H) is according to the number and location of fixed points, namely g ∈ Isom(HnK ) is called:

• elliptic if it has a fixed point in HnK ,

• parabolic if it has no fixed point in HnK and a single fixed point on ∂HnK , and

• loxodromic if it has no fixed point in HnK and exactly 2 fixed points on ∂HnK .

1.5.5 Geodesics, totally geodesic subspaces and metric

From the decription of reflections in the ball model B n and upper half-space model U n as
(restrictions of) inversions, we get the following description of totally geodesic subspaces:

Proposition 1.5.6 The totally geodesic subspaces of U n (resp. B n ) are the (intersections with
U n , resp. B n of ) generalized spheres orthogonal to ∂U n (resp. B n ).

In particular:

• geodesics in U n are either vertical half-lines or half-circles orthogonal to ∂U n ,

• geodesics in B n are either line segments through the origin or arcs of circles orthogonal
to ∂B n .

We now give an explicit formula for the Riemannian metric on B n and U n (in the infinitesi-
mal quadratic form ds2 ) which will be useful later. We denote as usual ||.|| the Euclidean norm,
and as before use coodinates (X, t) for U n with X ∈ Rn and t > 0. This metric on the the unit
ball B is called the Poincaré metric.
 2
2
Proposition 1.5.7 (1) Ball model: For Y ∈ B n and v ∈ TY B n : ds2Y (v) = 1−||Y ||2
||v||2 .
||v||2
(2) Upper half-space model: For (X, t) ∈ U n and v ∈ T(X,t) U n : ds2(X,t) (v) = t2
.

18
Proof. The metrics on U n and B n are defined as pullbacks of the metric on H. The brute force
computation can be a little unpleasant, but we will only compute the pullback at a single point
(namely O ∈ B n and (O, 1) ∈ U n ) and move it around by using isometries.
(1) The pullback metric on B n is defined by (in bilinear form notation):
(u, v)Y = hdY π −1 (u), dY π −1 (v)iπ−1 (Y ) ,
where π was defined as follows:
π: H −→ B
X
(X, t) 7−→ 1+t .
Then: d(O,1) π = 21 IdRn (identifying T(O,1) H ' Rn and TO B ' Rn ), so dO π −1 = 2IdRn and
ds2O (v) = 4||v||2 for v ∈ TO B n .   1/2 
Now, for Y 6= O in B n , consider the sphere S = S ||YY |2 , ||Y1||2 − 1 . Then S ⊥ ∂B
(by the Pythagorean theorem, as the radius R of S satisifies R2 + 1 = ||Y1||2 , the square of the
distance between the 2 centers). Therefore iS induces an isometry of B n , and:
Y (1/||Y ||2 ) − 1 (−Y )
iS (O) = + . = Y.
||Y ||2 ||Y /||Y ||2 ||2 ||Y ||2
2
|| )−1
Moreover, as we saw when we studied inversions, dO iS = (1/||Y ||Y ||2 /||Y ||4
r = (1 − ||Y ||2 )r for some
(Euclidean) reflection r, giving as claimed:
 2
2 2
dsY (v) = ||v||2 .
1 − ||Y ||2
n n
(2) Recall that we passed from the √ ball model B to the upper half-space model U by the
inversion iS , where S = S(−en , 2). As above, dO iS is of the form 2r, with r a Euclidean
reflection, so the metric at the point iS (O) = (O, 1) ∈ U n is given by ds2(O,1) (v) = ||v||2 (i.e. it
coincides with the Euclidean metric). Now dilations Dλ : (X, t) 7−→ (λX, λt) are isometries of
U n (because they are products of 2 inversions in concentric half-spheres orthogonal to ∂U n ),
2
and d(X,t) Dλ = λId, so ds2(O,λ) (v) = ||v||
λ2
. The result follows for all points (X, λ) by noting that
horizontal translations are also isometries of U n (as they are products of 2 reflections across
hyperplanes orthogonal to ∂U n ). 

1.6 The Klein model and Hilbert metric

We now briefly decribe another model for H n with underlying set the unit ball in Rn but with
a different metric from the Poincaré metric described earlier. This model is usually called the
Klein (or Klein-Beltrami) model. It is again obtained from the hyperboloid by radial projection
onto a hyperplane, but this time the radial projection is from the origin O, and the hyperplane
is {xn+1 = 1}. Explicitly, consider the radial projection:
π0 : H −→ B 0
(x1 , .., xn+1 ) 7−→ (x1 /xn+1 , ..., xn /xn+1 , 1)

19
Lemma 1.6.1 π 0 induces a diffeomorphism from H to the unit ball B 0 ⊂ Rn × {1}.

Proof. Note that:

(x1 , ..., xn+1 ) ∈ H ⇐⇒ x21 + ... + x2n − x2n+1 = −1

⇐⇒ x21 /x2n+1 + ... + x2n /x2n+1 = 1 − 1/x2n+1 ,

therefore if X ∈ H then π 0 (X) ∈ B 0 , and the inverse map is given by g(y1 , ..., yn ) = (λy1 , ..., λyn , λ)
 1/2
1
with λ chosen so that this lies in H, namely λ = 1−||Y || 2 . 

We denote B 0n the Riemannian manifold B 0 with the pullback of the metric on H by π 0−1 .

Proposition 1.6.1 (1) The totally geodesic subspaces of B 0n are (intersections with B 0 of )
affine subspaces of Rn . In particular, geodesics are Euclidean line segments.
(2) For X, Y ∈ B 0 , d(X, Y ) = 21 ln [X, Y ; P, Q] where P, Q are the endpoints on ∂B 0 of the line
segment connecting X and Y (so that P, X, Y, Q appear in that order), and [X, Y ; P, Q] denotes
the cross-ratio of (affine coordinates of ) X, Y, P, Q.

Recall that the cross-ratio of 4 points z1 , ..., z4 ∈ C

b is defined by:

z4 − z1 z3 − z2
[z1 , z2 ; z3 , z4 ] = · .
z4 − z2 z3 − z1
In fact the formula defining the distance in (2) makes sense for pairs of points in any convex
subset of Rn , and defines a distance function on any such set, called the Hilbert metric. Thus
hyperbolic space (in the Klein model) is a special case of a convex subset of Rn with the Hilbert
metric.
Proof. (1) From the geometric decription of π 0 it is clear that it sends linear subspaces of
Rn+1 to affine subspaces of the hyperplane {xn+1 = 1}; in more formal terms this follows because
π 0 is a collineation, i.e. an element of PGL(n + 1, R) acting by fractional linear transformations
- in fact it is the image in PGL(n + 1, R) of Id ∈ GL(n + 1, R).
(2) Normalize X and Y so that X = (0, ..., 0) and Y = (t, 0, ..., 0) for some t > 0. Then
P = (−1, 0, ..., 0) and Q = (1, 0, ..., 0) so that [X, Y ; P, Q] = 1+t
1−t
.
On the other hand, we lift X and Y to points X̃ and Ỹ in H and compute their distance there
by the distance formula, Proposition (1.1.1). This gives: X̃ = (0, ..., 0, 1) and Ỹ = (λt, 0, ..., 0, λ)
with λ = (1 − t2 )−1/2 . Then:

2 −1/2 ed + e−d
hX̃, Ỹ i = −(1 − t ) =−
2
by the distance formula, with d = d(X̃, Ỹ ) = d(X, Y ). Write y = ed + e−d and x = ed ; then by
1+t 1/2
p
the quadratic formula: x = y/2 + y 2 /4 − 1 = (1 − t2 )−1/2 + t(1 − t2 )−1/2 = 1−t


. 

20
1.7 Curvature
We first give a brief qualitative overview of the negative curvature properties of hyperbolic
space. We will then define these properties more precisely/quantitatively and prove them.
Hyperbolic space H n (with n > 2)) has negative curvature in any of the following senses:
• (a) Local Riemannian: H n has negative sectional curvature; this is the most classical
meaning of (negative) curvature. In fact, H n has constant negative sectional curvature κ
(equal to −1); here constant means both constant at all points and in all (pairs of) direc-
tions. Moreover, H n is the model space of constant negative curvature in the sense that
it is the unique simply-connected (complete) Riemannian manifold of constant curvature
−1. (Likewise, Euclidean space E n and the sphere S n with its standard round metric are
the model spaces of constant null, respectively positive sectional curvature).
• (b) Global Riemannian: The angle sum in geodesic triangles is < π. In fact, we will
see that any geodesic triangle T with internal angles α, β, γ (is contained in a copy of
H 2 and) satisfies:
Area(T ) = π − (α + β + γ). (1.7.1)
We will see that one can recover from this property the fact that κ ≡ −1.
• (c) Metric: Distance functions are strictly convex. More precisely, this means that,
given a point P and a geodesic γ(t) not containing P (resp. 2 geodesics γ1 (t), γ2 (t)),
parametrized with unit speed, (1) the function t 7→ d(P, γ(t)) is strictly convex and (2)
the function t 7→ d(γ1 (t), γ2 (t)) is strictly convex. In fact these properties are shared by
Euclidean space, except (2) in the case of intersecting or parallel lines (in which case the
distance function is linear).
In fact there are more general notions of negative (or non-positive) curvature for metric
spaces which are modelled on the properties of H n . These properties have played an increasingly
important role in geometric group theory since Gromov introduced them in the 1980’s (following
a long line of ideas from people like E. Cartan, A.D. Alexandrov and V.A. Toponogov - hence
the letters CAT).
• CAT(κ) spaces: A metric space is said to be CAT(κ) if its triangles are at least as
thin as in the correponding model space of constant curvature κ (the most common cases
are κ = 0 and −1). More precisely, given a geodesic triangle T with vertices x, y, z in a
(geodesic) metric space X, we define a comparison triangle to be any geodesic triangle
with the same side-lengths as T in X 0 , the model space of constant curvature κ (this
requires additional conditions on the 3 side-lengths if κ > 0). One then compares the
thinness of the 2 triangles by comparing the distances between pairs of points along their
various sides. More precisely, given a vertex and the pair of sides of T emanating from
that vertex, parametrized with unit speed from the vertex as γ1 (t1 ) and γ2 (t2 ), with γ10 (t1 ),
γ20 (t2 ) the unit-speed parametrizations of the 2 corresponding sides of T 0 in X 0 , we say
that T is at least as thin as T 0 along that pair of sides if d(γ1 (t1 ), γ2 (t2 )) 6 d0 (γ10 (t1 ), γ20 (t2 ))
for all t1 , t2 in the appopriate intervals. We say that T is at least as thin as T 0 if this
holds for all 3 pairs of sides, or that T satisfies the CAT(κ) inequality. Finally, we say
that X is a CAT(κ) space if all of its geodesic triangles satisfy the CAT(κ) inequality.

21
 • δ-hyperbolic spaces: A metric space is said to be δ-hyperbolic (for some fixed δ > 0)
if its geodesic triangles are uniformly δ- thin. This means that, in any geodesic triangle,
each side is entirely contained in a δ-neighborhood of the union of the 2 other sides.

Examples: Hyperbolic space H n is both CAT(−1) (by definition) and δ-hyperbolic for
some δ > 0 (Exercise: prove this and in particular compute a suitable δ). Trees are CAT(κ)
for any κ 6 0 and are 0-hyperbolic (triangles are tripods). Further examples include Cayley
graphs of certain finitely generated groups - e.g. free groups corresponding to trees - and
higher-dimensional cell complexes such as CAT(0) cube complexes.
We now prove properties (a), (b) and (c) announced above, starting with (c):
(c) Convexity of distance functions:
Recall that a function f : R 7−→ R is convex ⇐⇒ f ( t1 +t2
2
) 6 f (t1 )+f
2
(t2 )
for all t1 , t2 ∈ R. We
n
claimed above that, given a point O ∈ H and a geodesic γ(t) not containing O (parametrized
with unit speed), the function t 7→ d(O, γ(t)) is convex. This can be restated as follows,
denoting P1 = γ(t1 ), P2 = γ(t2 ) and M = γ( t1 +t
2
2
) (the condition about unit speed guarantees
that M is the midpoint of [P1 P2 ] - constant speed would suffice).

Lemma 1.7.1 Let O, P1 , P2 be 3 points in H n , and M the midpoint of [P1 P2 ]. Then:

2d(O, M ) < d(O, P1 ) + d(O, P2 ).

Proof. Let s denote the central involution at M. Then s(P1 ) = P2 ; denote O0 = s(O). By the
triangle inequality: 2d(O, M ) = d(O, O0 ) < d(O, P1 ) + d(P1 , O0 ) = d(O, P1 ) + d(O, P2 ). 

The second convexity property of distance functions is the following:

Proposition 1.7.1 Let γ1 (t), γ2 (t) be 2 distinct geodesics in H n , parametrized with unit speed.
Then the function t 7→ d(γ1 (t), γ2 (t)) is strictly convex.

Proof. Let t1 6= t2 ∈ R, P1 = γ1 (t1 ), Q1 = γ1 (t2 ), M1 = γ1 ( t1 +t

2
2
), and likewise P2 = γ2 (t1 ),
t1 +t2
Q2 = γ2 (t2 ), M2 = γ2 ( 2 ). The goal is to show that:

2d(M1 , M2 ) < d(P1 , P2 ) + d(Q1 , Q2 ).

Consider g = sM2 ◦ sM1 (where sx denotes the central involution at x), and let M = g(M1 ) =
sM2 (M1 ). Then M2 is the midpoint of [M1 M ], so d(M1 , M ) = 2d(M1 , M2 ). Let P = g(P1 ) =
sM2 (Q1 ); then:
d(P1 , P ) 6 d(P1 , P2 ) + d(P2 , P ) = d(P1 , P2 ) + d(Q1 , Q2 ),
so it suffices to show that d(M1 , M ) < d(P1 , P ), in other words: d(M1 , g(M1 )) < d(P1 , g(P1 ))
(geometrically, this says that points on the axis of g are moved a smaller distance than points off
the axis). We pass to the upper half-space model U n to show the latter inequality, normalizing
so that M1 , M2 lie on the vertical axis {O} × R+ . Then g = sM2 ◦ sM1 is a dilations Dλ for some
λ > 0 (as can be seen by writing each of sM1 , sM2 as a product of 3 inversions/reflections, 2 of
them across orthogonal vertical hyperplanes being common to both products, and the third in
each product being an inversions in a sphere centered at O ∈ ∂U n ). By assumption, P1 does
not lie on the vertical axis, and the distance d(P1 , g(P1 )) is unchanged by dilations so we may

22
asume that P1 is at the same height (the second factor in U n = Rn × R+ ) as M1 , from which
it follows that g(P1 ) is also at the same height as g(M1 ). It then
R follows from the explicit
R 2 form
n 2 2
of the metric in U that d(P1 , g(P1 )) > d(M1 , g(M1 )) because (dx1 + ... + dxn ) > dxn along
any non-vertical curve. 

(a) Riemannian sectional curvature I:

We will not define the Riemannian sectional curvature κx (X, Y ) at a point x ∈ H n in
the direction of the 2-plane in Tx H n spanned by X, Y , but we will use 2 of its characteristic
properties (see below).

Proposition 1.7.2 The sectional curvature at any point x ∈ H n in the direction of any 2-plane
in Tx H n is -1.

Proof. First observe that κx (X, Y ) is independent of the point x and the direction Span(X, Y ),
as Isom(H n ) acts transitively on H n and doubly transitively on orthonormal pairs (X, Y ) ∈
(Tx H n )2 , because O(n) (' Stab(x)) acts transitively on orthogonal pairs of vectors in the unit
sphere in Rn .
We now compute the value of κ at the point O in the ball model of H 2 , using the classical
fact (see e.g. [GHL]) that if Ω is a domain in R2 with metric of the form ds2x = α(x)2 ds2E 2 , then
1
the sectional curvature of Ω is given by: κ(x) = − α(x) 2 (∆logα) (x).
2 2
In the ball model B , we’ve seen that: α(x) = 1−||x|| 2 so α(O) = 4, ∆logα(O) = 4 and

κ(O) = −1. 

(b) Angle-sum in geodesic triangles: First observe the following:

Lemma 1.7.2 Any m points in H n (with m 6 n) belong to an isometrically embedded copy of

H m.

Proof. This is clear in the description of totally geodesic subspaces in the hyperboloid model,
as any m points in Rn+1 belong to a linear subspace of dimension m. 

In particular, any 3 points (and the geodesics connecting them) lie in a copy of H 2 . Therefore
we only need to analyze gedesic triangles in H 2 :

Proposition 1.7.3 Any geodesic triangle T in H 2 with internal angles α, β, γ satisfies:

Area(T ) = π − (α + β + γ).

In particular, α + β + γ < π.

Proof. Consider the upper half-plane model U 2 . Note that this is a conformal model (the metric
is a multiple of the Euclidean one), so the angles we ”see” are the intrinsic (Riemannian) angles
betwen geodesic rays.

23
 First assume that T has a vertex at ∞; we may assume (after maybe applying an isometry
of U 2 ) that the other 2 vertices lie on the unit cicle, say P = (cos(π − α), sin(π − α)) and
Q = (cos(β), sin(β) with β < π − α. Then:

T = {(cos(θ), y) | β 6 θ 6 π − α, y > sin(θ)},

so that:
ZZ Z π−α Z ∞ Z π−α
dxdy dy
Area(T ) = = sin θdθ · = dθ = π − α − β.
T y2 β sinθ y2 β

Now, if T has all 3 vertices P, Q, R in H 2 , we ”subdivide” to express T as an algebraic sum

of triangles with a vertex at ∞ as follows:

Area(T ) = Area(P Q∞) + Area(QR∞) − Area(P R∞)

= (π − α − α0 − β1 ) + (π − β2 − γ − γ 0 ) − (π − α0 − γ 0 )
= π − α − β − γ.


(a’) Riemannian sectional curvature II: (by parallel transport around boundaries of
triangles)
It can be shown (see e.g. [GHL]) that on an oriented Riemannian surface S, for any em-
bedded closed disk with piecewise linear boundary:
Z
κ(x)dA = (v,\T∂D (v))
D

for any p ∈ ∂D and v ∈ Tp S, where T∂D denotes parallel transport along ∂D, counterclockwise.
Denote by φ(∂D) the angle on the right-hand side (which is, by this fact, independent of p and
v). By this formula, given a point x ∈ S, if (Dn ) is a decreasing sequence of disks as above,
◦
φ(∂Dn )
with x ∈ Dn for all n and Diam(Dn ) −→ 0, then: κ(x) = lim Area(D n)
. Using a decreasing
sequence of triangles Tn , the formula for the area of a triangle gives: κ(x) = −1. Indeed,
parallel transport along the geodesic sides does not contribute to the angle φ(∂Tn ); only the
external angles at the vertices do, giving: φ(∂Tn ) = αn + βn + γn − π.

1.8 Boundary at infinity

Recall that a geodesic ray is a half-infinite geodesic line; a geodesic ray γ is entirely determined
by a point p ∈ H n and a unit tangent vetor v ∈ Tp H n , namely γ is then parametrized with
unit speed by γ(t) = expp (tv) for t > 0.
Let R = {geodesic rays in H n parametrized with unit speed}, and define an equivalence
relation ∼ on R by:
γ1 ∼ γ2 if Sup d(γ1 (t), γ2 (t)) < ∞.
t∈R+

24
Two rays equivalent under this relation are called asymptotic. The (Gromov) boundary ∂H n
of H n is then defined as R/ ∼. We denote by [γ]∼ the equivalence class of γ ∈ R under ∼.
We then define a topology on H n = H n ∪ ∂H n by requiring that the inclusion H n ,→ H n is
a homeomorphism onto its image, and that a basis of neighborhoods for any point p ∈ ∂H n is
of the form (U (x, V, r))(x,V,r) where each U (x, V, r) is defined, given a choice of representative
γ (such that p = [γ]∼ ), x = γ(0), V a neighborhood of γ 0 (0) in U Tx H n , and r > 0 by:
[ [
U (x, V, r) = γ((r, ∞)) ∪ [γ]∼ ,
where each union is taken over rays γ such that γ(0) = x and γ 0 (0) ∈ V . (The verification that
this family forms a neighborhood basis is straightforward and is left to the reader).
This definition of the boundary is intrinsic, i.e. it only depends on the Riemannian manifold
H n and not on a particular choice of model; we now shows that it agrees with the obvious
boundary in the ball model (and hence also in the upper half-space model). We denote as
before the unit ball in Rn by B.
Proposition 1.8.1 ∂H n is homeomorphic to ∂B = S n−1 , and H n is homeomorphic to B.
This will easily follow from the following description of asymptotic rays in the upper half-
space model. Given a geodesic ray γ in U n , we denote by γ(∞) the endpoint on ∂U n =
(Rn−1 × {0}) ∪ {∞} of the line segment/arc of circle γ.
Lemma 1.8.1 Let γ1 , γ2 be 2 geodesic rays in U n . Then: γ1 ∼ γ2 ⇐⇒ γ1 (∞) = γ2 (∞).
Proof. First assume that γ1 (∞) = γ2 (∞) and normalize this common endpoint to be ∞. Then
γ1 , γ2 are 2 vertical half-lines, with unit speed parametrizations of the form γ1 (t) = (X1 , et−t1 )
(with t > 0) and γ2 (t) = (X2 , et−t2 ) (with t > 0), for some X1 , X2 ∈ Rn−1 and t1 , t2 ∈ R. Then,
for all t > 0:




d γ1 (t), γ2 (t) 6 d (X1 , et−t1 ), (X1 , et−t2 ) + d (X1
, et−t2 ), (X2 , et−t2 )
6 |t1 − t2 | + d (X1 , e−t2 ), (X2 , e−t2 )
which is a uniform bound, independent of t.
Conversely, assume that γ1 (∞) 6= γ2 (∞), and normalize so that γ1 (∞) = ∞. Then γ1 is a
vertical half-line as above, and γ2 is an arc of circle, terminating on Rn−1 × {0}, whose unit-
speed parametrizations are of the form: γ1 (t) = (X1 , f1 (t)) and γ2 (t) = (X2 (t), f2 (t)) where f1
increases to ∞ and f2 eventually decreases to 0. Therefore, for large enough t:
Z f1 (t)

dxn+1
d γ1 (t), γ2 (t) > −→ ∞.
f2 (t) xn+1 t→∞


Proof of Proposition 1.8.1. By the lemma, the map:

φ : ∂H n −→ ∂B
[γ]∼ 7−→ γ(∞)
(where we’ve identified ∂B ' ∂U ) is well-defined and a bijection. It induces a homeomorphism
from H n to B, because the basis of neighborhoods of any point p ∈ ∂H n is sent by φ to a basis
of neighborhoods of φ(p) in ∂B. 

25
1.9 Complex hyperbolic space
We now describe complex hyperbolic space HnC in the projective model. Start with the Hermitian
vector space Cn,1 which is Cn+1 endowed with a Hermitian form of signature (n, 1), for example
the standard form:

hX, Y i = x1 y1 + ... + xn yn − xn+1 yn+1 , (1.9.1)

T T
where: X = (x1 , ..., xn+1 ) and Y = (y1 , ..., yn+1 ) .
Let V− = {X ∈ Cn,1 | hX, Xi < 0} and V0 = {X ∈ Cn,1 | hX, Xi = 0}. As a set, we define:
HnC = π(V− ) where π : Cn+1 \ {0} −→ CP n denotes projectivization (then we will have as before
∂HnC = π(V0 )). For example, taking the standard Hermitian as above, this gives in the affine
chart {[z1 : ... : zn+1 ] | zn+1 6= 0} of CP n :

HnC = (z1 , ..., zn , 1) | |z1 |2 + ... + |zn |2 < 1 ,



the unit ball BCn in Cn . Note that we cannot directly use the hyperboloid:

H = X ∈ Cn+1 | hX, Xi = −1


as a model for HnC as it is not a complex manifold (it has real dimension 2n + 1). However, we
may still use H to define a (Riemannian or Hermitian) metric on HnC as follows.
Note that π|H : H −→ HnC is a submersion (in fact it is a fiber bundle, induced from the
π
so-called tautological bundle C∗ −→ Cn+1 \ {0} −→ CP n ), and that for X ∈ H:

TX H = Y ∈ Cn,1 | Re(hX, Y i) = 0 ⊃ X ⊥


(this was an equality in the real case).

Lemma 1.9.1 The subspace X ⊥ of TX H is transverse to the fiber of π at X, therefore dX π|H
∼
induces an isomorphism X ⊥ −→ Tπ(X) HnC .
Proof. Denote F = π −1 (π(X)) the fiber of π at X. Then: F = C∗ X so TX F ' CX. Writing,
for  ∈ C: h(1 + )X, (1 + )Xi = (1 + )(1 + ¯)hX, Xi = −1 − 2Re() − ||2 , we see that
TX F ∩ TX H = iRX and TX F ∩ X ⊥ = {0}. 

Now, as before X ⊥ is equipped with a positive definite Hermitian form, namely the re-
striction of the ambient form h., .iX ⊥ ×X ⊥ (which induces a positive definite quadratice form,
Reh., .iX ⊥ ×X ⊥ ). By the lemma these forms carry over to Tπ(X) HnC , inducing a Hermitian met-
ric (and a Riemannian metric) on HnC . (Concretely, given x ∈ HnC , lift x to X ∈ Cn,1 with
hX, Xi = −1; then Tx HnC ' X ⊥ and the Hermitian, resp. Riemannian, metric at x is given by
h., .iX ⊥ ×X ⊥ , resp. Reh., .iX ⊥ ×X ⊥ .)
As in the real hyperbolic case, since the metric is entirely defined in terms of the form h., .i,
the group:

U(n, 1) = A ∈ GL(n + 1, C) | hAX, AY i = hX, Y i (∀X, Y ∈ Cn,1 )



acts by (holomorphic) isometries on HnC , i.e. PU(n, 1) ⊂ Isom+ (HnC ), where PU(n, 1) =
U(n, 1)/{λId | λ ∈ U(1)} and Isom+ (HnC ) now denotes the subgroup of holomorphic isometries

26
(not to be confused with orientation-preserving – for even n all isometries of HnC preserve orienta-
tion, whereas for odd n orientation-preserving isometries coincide with holomorphic isometries).
In fact it can be shown that Isom+ (HnC ) = PU(n, 1), and the full isometry group Isom(HnC )
is generated by PU(n, 1) and a single antiholomorphic isometry, such as the standard real
reflection σ0 : (z1 , ..., zn ) 7−→ (z1 , ..., zn ) in ball coordinates.
Analogously to the real case (Proposition 1.2.1), the isometries of HnC given by PU(n, 1) are
enough to show that HnC enjoys the following properties:
Proposition 1.9.1 (1) U(n, 1) acts transitively on V− , in particular HnC is homogeneous.
(2) The stabilizer in PU(n, 1) of any point of HnC is isomorphic to U (n).
(3) PU(n, 1) acts transitively on UTHnC .
(4) HnC is a symmetric space.
The proof is analogous to the real case and is left to the reader. Note that a more precise
statement of (2) is that the stabilizer of Cen+1 in U(n, 1) is U(n) × U(1), so
that the stabilizer
of O = π(en+1 ) (the origin in the ball model) in PU(n, 1) is P U(n) × U(1) ' U(n).
The main difference from the real case is that now U(n) acts transitively on the unit sphere
in Cn ' R2n , but not transitively on pairs of orthogonal unit vectors (contrary to O(n)).
Geometrically, this has several manifestations in HnC , for example in the fact that sectional
curvature is not constant in HnC (all unit tangent vectors are equivalent, but not all directions
spanned by 2 unit tangent vectors); one can show that it is in fact pinched between the constants
−1 (in the directions of complex projective lines) and −1/4 (in the directions of totally real
planes – see below). A similar phenomenon appears more globally, in the fact that triangles in
HnC (triples of points and the geodesic segments connecting them) are not in general contained in
a 2-dimensional totally geodesic subspace (as they are in H n , where every triangle is contained
in a copy of H 2 ). What remains true is that any triangle in HnC is conatined in a copy of H2C ,
but this has (real) dimension 4.
Note that, as a consequence of part (2) of the proposition, if an element g ∈ PU(n, 1) is
elliptic (i.e. has a fixed point in HnC ) then any of its lifts to U(n, 1) is diagonalizable, with unit
modulus eigenvalues. The following are special kinds of elliptic isometries:
• Central involutions are elements of PU(n, 1) with a lift conjugate to the diagonal matrix
Diag(−1, ..., −1, 1) ∈ U(n, 1).
• More generally, complex reflections in a point are elements of PU(n, 1) with a lift conjugate
to a diagonal matrix of the form Diag(eiθ , ..., eiθ , 1) ∈ U(n, 1) for some θ ∈ R.
• Complex reflections are elements of PU(n, 1) with a lift conjugate to a diagonal matrix
of the form Diag(eiθ , 1, ..., 1) ∈ U(n, 1).
Note that these special isometries all act on Cn,1 according to the following formula:
hX, ei
X 7−→ X + (eiθ−1 ) e, (1.9.2)
he, ei
for some e ∈ Cn,1 with he, ei 6= 0, reminiscent of the formula in Equation (1.3.2). Here, when
he, ei < 0 this formula gives the complex reflection in the point π(e) through angle −θ; when
he, ei > 0 the formula gives the complex reflection across the (complex) hyperplane e⊥ , through
angle θ.

27
1.10 Totally geodesic subspaces of HnC
In this section we determine the totally geodesic submanifolds of complex hyperbolic space
HnC . As in the real hyperbolic case these will be projective images of linear subspaces of the
Hermitian vector space Cn,1 , but the situation is more subtle due to the interaction between
the real and complex structures on this space. In order to state the main result we need a few
definitions.
Definitions: An R-linear subspace W of Cn,1 is called totally real if hX, Y i ∈ R for all
X, Y ∈ W (in other words, if the restriction of h., .i to W × W is a quadratic form). A C-
linear (resp. totally real) subspace W of Cn,1 is called hyperbolic if the restriction of h., .i to
W × W has signature (k, 1) for some k > 1; elliptic if it is positive definite; parabolic if it is
positive semidefinite (but not definite). Note that if W is a hyperbolic C-linear subspace of
Cn,1 with dimC W = k + 1 (resp. a hyperbolic totally real subspace with dimR W = k + 1),
then π(W ∩ V− ) ⊂ HnC is a copy of HkC (resp. HkR ) – in fact we will see that these copies are
isometrically embedded. Such subspaces π(W ∩ V− ) are called Ck -planes, resp. Rk -planes. The
main result is the following:

Theorem 1.10.1 The totally geodesic submanifolds of HnC are exactly the Ck -planes and Rk -
planes with 0 6 k 6 n.

This has the following consequence, which in the context of discrete groups of isometries makes
the construction of fundamental polyhedra more interesting, but alo more challenging than in
constant curvature:

Corollary 1.10.2 There do not exist totally geodesic real hypersurfaces in HnC when n > 2.

We will prove Theorem 1.10.1 in several steps, the main tool being the following:

Lemma 1.10.1 Let X be a symmetric space, and M ⊂ X a submanifold. Then, denoting sx

the central symmetry at a x ∈ X: M is totally geodesic ⇐⇒ sx (M ) = M for all x ∈ M .

Recall that M is totally geodesic if it contains every geodesic line which is tangent to it (in
particular, totally geodesic submanifolds are geodesically complete).
Proof. The direction ⇒ follows from the definition of the central symmetry sx . More
explicitly, assume that M is totally geodesic in X and let x ∈ M . Then, for any geodesic line γ
through x and tangent to M at x, sx (γ) = γ is contained in M by hypothesis. Therefore sx (M )
is contained in the union of such geodesic lines γ, which is contained in M . But sx (M ) is a
geodesically complete submanifold of X of the same dimension as M , therefore sx (M ) = M .
Conversely, assume that sx (M ) = M for all x ∈ M . We use the equivalent formulation that
M is totally geodesic if and only if, locally, geodesic segments connecting pairs of points in M
are in M . Let x0 ∈ M and let N be a neighborhood of x0 in X such that any 2 points of N
are connected by a unique geodesic segment in X; likewise, let N1 be a neighborhood of x0 in
M such that any 2 points of N1 are connected by a unique geodesic segment in M . We may
assume that N1 ⊂ N (after shrinking it if necessary).
Let p, q ∈ N1 , and denote by α(t) with t ∈ [0, a] (resp. β(t) with t ∈ [0, b]) the unique
geodesic segment in X (resp. M ) connecting p to q, parametrized with unit speed. Let m =

28
β(b/2) ∈ M be the midpoint of the geodesic segment β; then sm preserves β (as it is an
isometry preserving M ) and exchanges p and q. Since sm exchanges p and q, it also preserves
α and therefore also fixes the midpoint m0 = α(a/2) of α. But central symmetries have an
isolated fixed point, so (after shrinking N if necessary) m = m0 , i.e. α(a/2) = β(b/2). By
iterating this process, we get that α(ta) = β(tb) for any t of the form k/2n (for some n ∈ N
and 1 6 k 6 2n − 1). But such points are dense in [0, 1], therefore by continuity α(ta) = β(tb)
for all t ∈ [0, 1], giving (a = b and) α = β. 

We now state and prove the easier half of Theorem 1.10.1:

Proposition 1.10.1 If W is a hyperbolic C-linear or totally real subspace of Cn,1 , then π(W ∩
V− ) is a totally geodesic subspace of HnC .

Proof. Let M = π(W ∩ V− ) and p ∈ M ; lift p to P ∈ W with hP, P i = −1. Then the central
symmetry sp lifts to U(n, 1) as SP : X 7−→ X − 2hX, P iP , which preserves W (because W is
C-linear or totally real). Therefore sp (M ) = M , and by Lemma 1.10.1 M is totally geodesic.


Before proving the second half of Theorem 1.10.1 we use this to give a concrete description
of geodesics in HnC :

Proposition 1.10.2 (1) Geodesics through the origin O in the ball model are exactly straight
line segments.
(2) PU(n, 1) acts transitively on geodesics in HnC .
(3) There exists a unique geodesic connecting any 2 distinct points in HnC .
(4) Given 2 distinct points p, q ∈ HnC with lifts P, Q to Cn,1 normalized

so that hP, Qi ∈ R, the
geodesic line (pq) containing p and q is π SpanR (P, Q) ∩ V− .

Note: In practice, part (4) allows us to parametrize (pq) \ {q} by π(P + tQ) (with t in the
interval giving hP + tQ, P + tQi < 0), and likewise (pq) \ {p} by π(tP + Q). These are however
not unit speed parametrizations. The most convenient way to parametrize a unit speed geodesic
in HnC is to use its endpoints p∞ , q∞ ∈ ∂HnC (or, as in the real hyperbolic case, paramatrize a
standard geodesic then carry it over by an isometry). Namely, it can easily be checked that
given lifts P∞ , Q∞ ∈ V0 , normalized so that hP∞ , Q∞ i = −1/2, the following is a unit speed
parametrization of the geodesic line connecting p∞ and q∞ :

γ(t) = e−t P∞ + et Q∞ .


Proof. (1) Take W = SpanR (1, 0, ..., 0, 1), (−1, 0, ..., 0, 1) ⊂ Cn,1 . This is a hyperbolic
totally real subspace of Cn,1 , of real dimension 2, so by Proposition 1.10.1 γ0 = π(W ∩ V− ) is a
totally geodesic subspace of dimension 1, i.e. a geodesic line. Now any straight line through O
is in the orbit of γ0 under StabPU(n,1) (O) because, as we’ve seen, StabPU(n,1) (O) acts transitively
on unit tangent vectors at O. Conversely any geodesic through O is a straight line, as there
can only be one geodesic through a given point with a given tangent vector.
(2) and (3) follow from transitivity of the action of PU(n, 1) on HnC .
(4) Any geodesic line γ in HnC is the intersection of a complex line and a totally real plane,

29
because γ0 has this property and this property

is invariant undern,1
the action of PU(n, 1). This
means exactly that: γ = π SpanR (P, Q) ∩ V− for some P, Q ∈ C with hP, Qi ∈ R. 

We now state and prove the more difficult half of Theorem 1.10.1:

Proposition 1.10.3 Any totally geodesic submanifold of HnC is a Ck -plane or an Rk -plane for
some k with 0 6 k 6 n.

Proof. We break down the proof into a sequence of lemmas. Let M be a totally geodesic
submanifold of HnC in the ball model B ⊂ Cn ; by applying an isometry we may assume that M
contains the origin O.

Lemma 1.10.2 M is of the form W ∩ B for some R-linear subspace W of Cn .

Proof of Lemma 1.10.2: We know that the geodesic lines through O are exactly the straight
line segments through O contained in B. Therefore, M = W ∩ B where W = RM is a subset
of Cn closed under multiplication by real numbers. Now let u, v ∈ W and assume in fact (after
maybe rescaling) that u, v ∈ W ∩ B. Then, since W is totally geodesic, by the same dyadic
subdivision argument as in the proof of Lemma 1.10.1, the whole line segment [u, v] is contained
in M . In particular, u + v ∈ W and W is an R-linear subspace of Cn . 

In what follows, given a subspace W of Cn , we identify W with W × {0} ⊂ Cn . Given

z ∈ V− , recall from Equation (1.9.2) that the central symmetry sπ(z) at π(z) ∈ HnC lifts to
Sz ∈ U(n, 1) acting on Cn+1 by:

hz, wi
Sz (w) = w − 2 z.
hz, zi

Lemma 1.10.3 Let W be an R-linear subspace of Cn . Then: W ∩ B is totally geodesic ⇐⇒

Sz (W + Ren+1 ) ⊂ C · (W + Ren+1 ) for all z ∈ V− ∩ (W + Ren+1 ).

Proof of Lemma 1.10.3: This follows directly from Lemma 1.10.1 and the fact that W =
π(W + Ren+1 ) = π(C · (W + Ren+1 )) (again, identifying W with W × {0} ⊂ Cn ). 

Lemma 1.10.4 Let W be an R-linear subspace of Cn such that W ∩ B is totally geodesic, and
identify as above W with W × {0} ⊂ Cn . For any w ∈ W \ {0}, define:

F (w) = λ ∈ C | λw ∈ W

(a) For any w ∈ W \ {0}, F (w) = R or C.

(b) For any w, w0 ∈ W \ {0}, hw, w0 i ∈ F (w) ∩ F (w0 ).
(c) If λ ∈ F (w) then λhw, w0 i ∈ F (w0 ).
(d) If hw, w0 i ∈ R then F (w) = F (w0 ).
(e) Either F (w) = R for all w ∈ W \ {0}, or F (w) = C for all w ∈ W \ {0}.

30
 Proof of Lemma 1.10.4: (a) For any w ∈ W \ {0}, F (w) is an R-linear subspace of C,
containing R (because W is an R-linear subspace). Therefore F (w) = R or C.

(b) Let w, w0 ∈ W \ {0}, λ = hw, w0 i and assume (after possibly rescaling by a real number)
that hw, wi < 1. This guarantees that x = w + en+1 ∈ V− ∩ (W + Ren+1 ), because:

hx, xi = hw, wi + 2Rehw, en+1 i + hen+1 , en+1 i = hw, wi − 1.

Given any r ∈ R, denote y = rw0 + en+1 . Then by Lemma 1.10.3, sx (y) ∈ C · (W + Ren+1 ), i.e.
hx,yi
y − νx = µ(w00 + ten+1 ) for some µ ∈ C and t ∈ R, with ν = 2 hx,xi .
0
In terms of w, w , r this becomes:

rw0 + en+1 − ν(w + en+1 ) = µ(w00 + ten+1 ), or :

rw0 − νw + (1 − ν)en+1 = µw00 + tµen+1 .

Since en+1 and W are (C-)linearly independent, this gives:

1 − ν = tµ
(1.10.1)
rw − νw = µw00
0

The first line gives R(ν) = R(µ), and on the other hand R(ν) = R(λ) if r 6= 0 (because
hx, xiν/2 = hx, yi = hw + en+1 , rw0 + en+1 i = rhw, w0 i − 1), so that R(µ) = R(λ).
Combining the 2 lines of (1.10.1) gives: w00 = µ−1 rw0 + (t − µ−1 )w ∈ W . Since tw ∈ W , this
gives: µ−1 (rw0 − w) ∈ W , so µ−1 ∈ F (rw0 − w), hence λ ∈ F (rw0 − w) (because R(µ) = R(λ)).
In other words, λ(rw0 − w) ∈ W . Taking r = ±1, we get λw, λw0 ∈ W as claimed.

(c) By part (b), hλw, w0 i = λhw, w0 i ∈ F (w0 ) if λ ∈ F (w).

(d) First assume that hw, w0 i = r ∈ R \ {0}. By part (c), for any λ ∈ F (w) we have
rλ ∈ F (w0 ) so λ ∈ F (w0 ), giving F (w) ⊂ F (w0 ). Likewise, F (w0 ) ⊂ F (w), reversing the roles
of w and w0 . Now, if hw, w0 i = 0 then hw, w + w0 i = hw, wi =6 0 (as h., .i is positive definite on
W ). Then by the previous argument F (w) = F (w + w ) and likewise F (w0 ) = F (w + w0 ) so
0

that F (w) = F (w0 ).

(e) Let w, w0 ∈ W \ {0} and λ = hw, w0 i. If λ = 0 then by part (d) F (w) = F (w0 ). If not,
then hλ−1 w, w0 i = 1 so by part (d) F (λ−1 w) = F (w0 ). But λ ∈ F (w) by part (b), and F (w) is
a field by part (a). Therefore F (w) = F (w0 ). 

This concludes the proof of Proposition 1.10.3, except the bound 0 6 k 6 n for the dimension
of an Rk -plane in HnC . Note that Rk -planes through O in the ball model are of the form W ∩ B
for some totally real R-linear subspace of Cn . R-linear subspaces of Cn may have any real
dimension 6 2n, but the bound on the dimension of totally real subspaces of Cn comes from
the following observation:

Lemma 1.10.5 Let W be an R-linear subspace of Cn . Then: W is totally real if and only if
W ⊥Reh.,.i iW .

31
Proof of Lemma 1.10.5: W is totally real ⇐⇒ hX, Y i ∈ R for all X, Y ∈ W , which is
equivalent to RehX, iY i = 0 for all X, Y ∈ W , as hX, iY i = −ihX, Y i. 

Now Reh., .i is positive definite on Cn , therefore if W is totally real Lemma 1.10.5 implies
W ∩ iW = {0}, so that W has (real) dimension at most n. 

1.11 Bisectors
Following Mostow, we call equidistant hypersurfaces in HnC bisectors. Explicitly, given 2 distinct
points p1 , p2 ∈ HnC , the bisector equidistant from p1 , p2 is:

B(p1 , p2 ) = p ∈ HnC | d(p, p1 ) = d(p, p2 ) .



The complex spine Σ of B = B(p1 , p2 ) is the complex line spanned by p1 , p2 ; the real spine σ
of B is the real geodesic B ∩ Σ.
As we’ve seen, bisectors cannot be totally geodesic (as they would be in constant curvature
spaces) because they are real hypersurfaces (Corollary 1.10.2), but they admit 2 foliations by
totally geodesic subspaces as follows:

Proposition 1.11.1 Let p1 , p2 be 2 distinct points in HnC , B = B(p1 , p2 ) the bisector equidistant
from p1 and p2 , Σ the complex spine of B and σ its real spine. Then:
(1) B = πΣ−1 (σ), where πΣ denotes orthogonal projection onto Σ, and
(2) B is the union of all Rn -planes containing σ.

Part (1) tells us that B is foliated by the Cn−1 -planes πΣ−1 {x} for x ∈ σ; these are

called the slices of B; the Rn -planes containing σ are called the meridians of B (the meridian
decompoisition is not literally a foliation as all meridians share the real spine σ). Note also
that part (1) tells us that the real and complex spines only depend on the bisector itself (and
not on the pair of points used to define it). Also, given a bisector B, the pairs of points from
which B is equidistant are much more constrained than in constant curvature, namely they are
exactly pairs of points in Σ \ σ which are symmetric with repsect to σ.
We will skip the proof of Proposition 1.11.1 for the sake of time. Part (1) is due to Mostow
and part (2) to Goldman. The following gives a complete description of the geodesic non-
convexity of bisectors; it is also due to Goldman and we will also skip its proof:

Proposition 1.11.2 Given a bisector B in HnC and 2 distinct points p, q ∈ B, the geodesic (pq)
is contained in B iff p, q are in a common slice or meridian of B.

Our motivation to study these equidistant hypersurfaces comes from the world of discrete groups
and fundamental domains, where they arise as bounding hypersurfaces for Dirichlet domains,
which we define and study in the next part.

32
Chapter 2

Discrete subgroups and lattices in

Isom(HnR) and Isom(HnC)

2.1 Discontinuous group actions and fundamental do-

mains
We start with a few general definitions about group actions on metric spaces. Let Γ be a group
acting by isometries on a metric space X. The definitions and setup are in complete generality,
however most of the properties that we are interested in will require X to be a proper metric
space, meaning that all closed metric balls in X are compact (note that such a space is in
particular locally compact).
Definitions: We will say that Γ acts properly discontinuously on X if for any compact
K ⊂ X, the set γ ∈ Γ | γK ∩ K 6= ∅ is finite.
A simply-connected subset D ⊂ X with non-empty interior will be called a fundamental
domain for the action of Γ on X if:
[
(a) γD = X, and
γ∈Γ

◦ ◦
(b) γ1 6= γ2 ∈ Γ =⇒ γ1 D ∩ γ2 D = ∅.
When these conditions are satisfied, we say that the images of D under Γ tessellate X. Note
that traditionally, a domain is a simply-connected open set, but we do not require D to be
open, just to have non-empty interior. There is a related notion that is a bit more technical
but useful in practice: D as above is called a fundamental domain for the action of Γ modulo
a subgroup H < Γ if it satisfies condition (a) and the following condition in place of (b):
◦ ◦ ◦ ◦
(b0 ) For all γ1 , γ2 ∈ Γ, either γ1 D ∩ γ2 D = ∅, or γ1 D = γ2 D and γ1 γ2−1 ∈ H.
The following is a general construction of fundamental domains for groups acting properly
discontinuously on a (proper) metric space. Given a point p0 ∈ X, the Dirichlet domain for Γ
centered at p0 is defined as:

DΓ (p0 ) = x ∈ X | d(x, p0 ) 6 d(x, γp0 ) ∀γ ∈ Γ . (2.1.1)

33
Note: We will be mostly interested in the case where the group Γ is infinite, in which case it is
not clear whether or not DΓ (p0 ) has non-empty interior (or even whether or not it is reduced to
{p0 }), and if so, whether or not it has finitely many faces, i.e. whether or not a finite collection
of elements γ ∈ Γ suffice in (2.1.1).
For any two distinct points x, y ∈ X, the space X is divided into the following 3 regions:

B(p1 , p2 ) = x ∈ X | d(x, p1 ) = d(x, p2 ) (2.1.2)
−

B (p1 , p2 ) = x ∈ X | d(x, p1 ) < d(x, p2 ) (2.1.3)
+

B (p1 , p2 ) = x ∈ X | d(x, p1 ) > d(x, p2 ) (2.1.4)

We will say that X is balanced if for any two distinct points p1 , p2 ∈ X, every point of B(p1 , p2 )
is an accumulation point of both B(p1 , p2 )− and B(p1 , p2 )+ .

Proposition 2.1.1 Let Γ be a group acting by isometries on a proper balanced metric space
X.
(1) If Γ acts properly discontinuously on X, then for all p0 ∈ X, DΓ (p0 ) contains an open ball
centered at p0 .
(2) Assume moreover that (a) Γ is a finitely generated linear group acting faithfully on X, and
(b) for any isometry f 6= Id of X, Fix(f ) is nowhere dense in X (respectively, X admits a
nonzero measure m such that for any isometry f 6= Id of X, m(Fix(f )) = 0). If for all p0 ∈ X
DΓ (p0 ) contains an open ball centered at p0 , then Γ acts properly discontinuously on X.
(3) If DΓ (p0 ) contains an open ball centered at p0 then DΓ (p0 ) is a fundamental domain for the
action of Γ modulo StabΓ (p0 ).

We start by stating and proving the following auxiliary results:

Lemma 2.1.1 (a) DΓ (p0 ) is closed.

◦
(b) If x ∈ DΓ (p0 ) then d(x, p0 ) < d(x, γp0 ) for all γ ∈ Γ such that γp0 6= p0 .
(c) For any γ0 ∈ Γ, γ0 DΓ (p0 ) = DΓ (γ0 p0 ).
◦ ◦
(d) For γ1 , γ2 ∈ Γ, if γ1 p0 6= γ2 p0 then γ1 DΓ (p0 ) ∩ γ2 DΓ (p0 ) = ∅.

Proof. When the dependence on p0 is not relevant (as it is in (c)) we will write D for DΓ (p0 ).

(a) For each γ ∈ Γ, x ∈ X | d(x, p0 ) 6 d(x, γp0 ) is closed, therefore D is an intersection
of closed sets and is closed.

(b) Let x ∈ DΓ (p0 ) and assume that d(x, p0 ) = d(x, γp0 ) for some γ ∈ Γ such that γp0 6= p0 .
Then x ∈ B(p0 , γp0 ), and since X is assumed balanced any ball centered at x contains points
of both B − (p0 , γp0 ) and B + (p0 , γp0 ). Therefore x is not an interior point of DΓ (p0 ).

(c) Note that:

x ∈ DΓ (p0 ) ⇐⇒ (∀γ ∈ Γ) d(γ0 x, γ0 p0 ) 6 d(γ0 x, γ0 γp0 )

⇐⇒ γ0 x ∈ DΓ (γ0 p0 ),

34
where the first equivalence follows from the fact that γ0 is an isometry and the second from the
fact that γ 7−→ γ0 γ is a bijection Γ −→ Γ.
◦
(d) From parts (b) and (c), for i = 1, 2: if x ∈ γi D then for all γ ∈ Γ such that γγi p0 6= γi p0 ,
◦ ◦
d(x, γi p0 ) < d(x, γγi p0 ). Assume that there exists a point x ∈ γ1 D ∩ γ2 D. Then, taking
γ = γ2 γ1−1 in the above condition for i = 1 gives d(x, γ1 p0 ) < d(x, γ2 p0 ). Likewise, taking
taking γ = γ1 γ2−1 in the above condition for i = 2 gives d(x, γ2 p0 ) < d(x, γ1 p0 ), a contradiction.


Proof of Proposition 2.1.1: (1) Assume Γ acts properly discontinuously  on X, let p0 ∈ X and
r > 0 and denote B = B(p0 , r). Then B̄ is compact by properness of X, so γ ∈ Γ | γ B̄ ∩ B̄ 6= ∅
is finite. Denote γ1 , ..., γn = γ ∈ Γ | γ B̄ ∩ B̄ 6= ∅ and γp0 6= p0 . After possibly shrinking r,
we may assume that p0 ∈ / γ1 B̄ ∪ ... ∪ γn B̄. Note that:

• if γ B̄ ∩ B̄ = ∅, then B̄ ⊂ x ∈ X | d(x, p0 ) 6 d(x, γp0 ) , and

• if γ B̄ ∩ B̄ ⊂ ∂B, then B ⊂ x ∈ X | d(x, p0 ) 6 d(x, γp0 ) .


Therefore it suffices to take ε 6 Min d(p0 , γi B̄) to get B(p0 , ε) ⊂ DΓ (p0 ).
16i6n

(2) Assume the action of  Γ on X is not properly

discontinuous. Then there exists a compact
subset K of X such that γ ∈ Γ | γK ∩ K 6= ∅ is infinite. This provides a sequence (γn ) of
distinct elements of Γ and a sequence (xn ) of points of K such that xn ∈ γn K ∩ K for all n.
Since K is compact, we may assume after passing to a subsequence that the sequence (xn )
(resp. (γn−1 xn )) converges, say to x0 (resp. x1 ). By continuity of the action Γ × X −→ X,
(γn−1 x0 ) converges to x1 . If this sequence is eventually constant then StabΓ (x0 ) is infinite; if not
then the orbit Γx0 has an accumulation point. We treat these 2 cases separately.

Lemma 2.1.2 If the orbit Γx0 has an accumulation point then DΓ (x0 ) does not contain any
open ball centered at x0 .

Proof of Lemma 2.1.2: Assume that D = DΓ (x0 ) contains an open ball B(x0 , ε) centered at x0 ,
and let (γn x0 ) be a sequence of distinct points converging to some point p ∈ X, where γn ∈ Γ
for all n. Then, for m, n large enough: d(γm x0 , γn x0 ) 6 ε. On the other hand, γm D (resp. γn D)
◦ ◦
contains B(γm x0 , ε) (resp. B(γn x0 , ε)), which would imply that γm D ∩ γn D 6= ∅, contradicting
Lemma 2.1.1 (d), as γm x0 6= γn x0 . 

Lemma 2.1.3 If some point of X has an infinite stabilizer, then Γ has an orbit with an accu-
mulation point.

Proof of Lemma 2.1.3: Assume that |Γp | = ∞ for some p ∈ X, denoting Γp = StabΓ (p).

Claim: Γp has an infinite orbit.

35
 Lemma 2.1.3 then follows from the claim, as the orbits of Γp are contained in spheres
centered at p, which are compact by properness of X (as they are closed subsets of closed balls,
which are assumed to be compact).
We now prove the claim. Assume that all orbits of Γp are finite. Then in particular:

(∀x ∈ X) (∀γ ∈ Γp ) (∃n > 1) γ n x = x. (2.1.5)

n
S γ ∈ Γp , and for n > 1 let Xn = Fix(γn ) = {x ∈ X | γ x = x}. Then by (2.1.5):
Fix
Xn = X. But by assumption, each Xn is either all of X, or nowhere dense (resp. has
n>1
measure 0). Assuming that Xn 6= X for all n, this would imply by the Baire Category Theorem
(as X is locally compact Hausdorff) that X is nowhere dense (resp. that X has measure 0 by
subadditivity of measure), a contradiction. Therefore Xn = X for some n > 1, hence γ has
finite order as the action is faithful. This proves that all elements of Γp have finite order.
Problem: Is it true that a (finitely generated) group Γ all of whose elements have finite
order is finite?
This is a famous question, known as the Burnside problem. Burnside proved that the an-
swer is yes if Γ is linear (i.e. a subgroup of GL(n, C) for some n) and the orders of all elements
are uniformly bounded; Schur then extended this result by removing the bounded orders as-
sumption. (The answer is no in general: Golod and Shafarevich constructed counter-examples
in the 1970’s, with unbounded orders, and subsequently Olshanskii and Ivanov constructed
counter-examples with bounded orders).
In our case, under the assumption that Γ is a finitely generated linear group, we cannot
directly apply Schur’s theorem to Γp (as it could be infinitely generated). However, if we look
more closely at the proof of Schur’s theorem we see that the result also applies to subgroups
of finitely generated linear groups (see e.g. Curtis-Reiner). Alternatively, we could appeal to
Selberg’s lemma (Theorem 2.3.1), which says that Γ has a finite-index torsion-free subgroup
H; then H ∩ Γp has finite index in Γp and is trivial, so Γp is finite.
This proves the claim and concludes the proof of Proposition 2.1.1(2).

Proposition 2.1.1(3) follows from Lemma 2.1.1(d) and the following:

Lemma 2.1.4 If DΓ (p0 ) contains an open ball centered at p0 , then DΓ (p0 ) contains an element
of each orbit.

Proof of Lemma 2.1.4: Let x ∈ X and d = Inf d(x, γp0 ) | γ ∈ Γ .

• If d is attained, i.e. d = d(x, γ0 p0 ) for some γ0 ∈ Γ, then x ∈ DΓ (γ0 p0 ) which by

Lemma 2.1.1(c) is equal to γ0 DΓ (p0 ) .

• If not, there exists a sequence (γn ) of distinct elements of Γ such that d(x, γn p0 ) decreases
to d. Then, for n sufficently large γn p0 ∈ B(p0 , 2d) which is compact by properness of
X, hence (γn p0 ) has a converging subsequence. Now by assumption DΓ (p0 ) contains an
open ball B(p0 , ε) centered at p0 . Then, for m, n large enough we have d(γm p0 , γn p0 ) 6 ε,
◦ ◦
which implies that γm DΓ (p0 ) ∩ γn DΓ (p0 ) 6= ∅, contradicting Lemma 2.1.1(d). 

36
2.2 Hyperbolic manifolds and orbifolds
We now describe the structure that arises on a quotient of the form X/Γ where X is a proper
metric space and Γ < Isom(X) is a discrete subgroup. Here discreterefers to the compact-open

topology on Isom(X), which is generated by sets of the form UK,V = g ∈ Isom(X) | g(K) ⊂ V ,
where K (resp. V ) is a fixed compact (resp. open) subset of X.

Proposition 2.2.1 Let X be a proper metric space and Γ < Isom(X). Then:
(1) Γ is discrete (in the compact-open topology) ⇐⇒ Γ acts properly discontinuously on X.
(2) If Γ is discrete then X/Γ is Hausdorff. If moreover Γ acts freely (i.e. without fixed points)
then p : X 7−→ X/Γ is a covering map and a local isometry.

Notes: (a) When as in part (2) Γ is discrete and acts freely on X, if X is a (topological, resp.
smooth,...) manifold then so is X/Γ, and X/Γ is locally homeomorphic (resp. diffeomorphic,...)
to X (as p is then also a local diffeomorphism). Loosely speaking, such an X/Γ inherits any ”ge-
ometric structure” that X has. We now make this more precise in the case where X = HnR or HnC .

(b) When Γ < Isom(HnR ) (resp. Isom(HnC )) is discrete and acts freely the manifold X/Γ
is locally isometric to HnR (resp. HnC ). Such manifolds are called hyperbolic manifolds (resp.
complex hyperbolic manifolds). More generally, a hyperbolic orbifold (resp. complex hyperbolic
orbifold ) is a quotient space HnR /Γ (resp. HnC /Γ) where Γ is a discrete subgroup of Isom(HnR )
(resp. Isom+ (HnC )). Moreover, it can be shown that the compact-open topology on Isom(HnR )
(resp. Isom+ (HnC )) coincides with the usual matrix topology on O+ (n, 1) (resp. the quotient
topology on PU(n, 1)) – this is left to the reader as an exercise.

(c) In particular, if M = HnR /Γ is a hyperbolic manifold then M is a complete Riemannian

manifold with constant negative curvature −1. It can be shown that HnR is the only complete,
simply-connected Riemannian n-manifold with constant negative curvature −1, from which it
follows that if M is a complete Riemannian manifold with constant negative curvature −1,
then M is isometric to HnR /Γ for some discrete Γ < Isom(HnR ) acting freely on HnR , i.e. M is
a hyperbolic manifold as we’ve defined it. This gives an equivalent definition of hyperbolic
manifolds which is the standard definition in differential geometry.
We will use the following lemmas in the proof of Proposition 2.2.1:

Lemma 2.2.1 Γ < G = Isom(X) is discrete ⇐⇒ any sequence of elements of Γ converging

to Id is eventually constant.

Proof : Γ is discrete ⇐⇒ (∀γ ∈ Γ) {γ} is open in Γ

⇐⇒ (∀γ ∈ Γ) (∃Uγ a neighborhood of γ in G) Γ ∩ Uγ = {γ}
⇐⇒ (∀γ ∈ Γ) any sequence of elements of Γ converging to γ
is eventually constant
⇐⇒ any sequence of elements of Γ converging to Id
is eventually constant.
Note for the last point that for any γ, the map g 7−→ γg is a self-homeomorphism of G. 

37
Lemma 2.2.2 The compact-open topology on G = Isom(X) is the topology of pointwise con-
vergence. In other words, given a sequence (gn ) of elements of G and g ∈ G:
gn −→ g for the compact-open topology on G ⇐⇒ (∀x ∈ X) gn (x) −→ g(x).

Proof. We will use the following general fact about the compact-open topology: if Y is a metric
space (and X any topological space) then the compact-open topology on C(X, Y ) = {continuous
functions X −→ Y } is the topology of uniform convergence on compact subsets of X (see e.g.
[]). This immediately gives the forward implication, namely if gn −→ g for the compact-open
topology on G then gn (x) −→ g(x) for all x ∈ X.
Conversely, assume that (∀x ∈ X) gn (x) −→ g(x) and let K ⊂ X be compact. Assuming
that the convergence is not uniform on K, there would exist
ε > 0, a subsequence (gni ) of (gn )
and a sequence of points xi ∈ K such that d gni (xi ), g(xi ) > ε for all i. But, K being compact,
we may assume (after possibly passing to a subsequence)
that (xi ) converges, say to x. Then,
for i large enough, d(xi , x) < ε/4 and d gni (x), g(x) < ε/2, giving:





d gni (xi ), g(xi ) 6 d gni (xi ), gni (x) + d gni
(x), g(x) + d g(x), g(xi )
= 2d(xi , x) + d gni (x), g(x) < ε,

a contradiction. (Note that we used in the last step that g and the gni are isometries of X.) 


Lemma 2.2.3 Let K1 , K2 ⊂ X be compact. Then: GK1 ,K2 = g ∈ G | g(K1 ) ⊂ K2 is compact
(for the compact-open topology). In particular, StabIsom(X) (x) is compact for all x ∈ X.

Proof. This follows immediately from the Arzela-Ascoli theorem, as GK1 ,K2 is:

(a) equicontinuous (as all its elements are isometries),

(b) pointwise relatively compact, i.e. (∀x ∈ K1 ) GK1 ,K2 is relatively compact as it is contained
in K2 , and

(c) closed, by continuity of the orbit map (g, x) 7−→ gx. 

Proof of Proposition 2.2.1: (1) First assume that Γ is not discrete in G = Isom(X). Then,
by Lemma 2.2.1 there exists a sequence (γn ) of distinct elements of Γ converging to Id. Let
0 ∈ X and r > 0; then
x for n large enough γn x0 ∈ B(x0 , r) = B by Lemma 2.2.2, in particular
γ ∈ Γ | γB ∩ B 6= ∅ is infinite. But B is compact by properness of X, so the action of Γ on
X is not properly discontinuous.
Conversely, assume that the action of Γ on X is not properly discontinuous. As previously,
this gives either an orbit with an accumulation point, or a point with an infinite stabilizer.
In the latter case, Γ is not discrete as point stabilizers in G are compact by Lemma 2.2.3. If
Γ has an orbit with an accumulation point, there exists a sequence of distinct points (γn x0 )
converging to a point p ∈ X (for some
γn ∈ Γ). Fix r > 0; then for n large enough we
have d(γn x0 , p) 6 r hence γn B(x0 , r) ⊂ B(p, 2r) (indeed, if y ∈ X satisfies d(y, γn x0 ) 6 r
then d(y, p) 6 d(y, γn x0 ) + d(γn x0 , p) 6 2r). In other words, for n large enough we have
γn ∈ GB(x0 ,r),B(p,2r) which is compact by Lemma 2.2.3 and by properness of X. Therefore (γn )

38
has a converging subsequence, therefore Γ is not discrete as the γn are distinct.

(2) First assume that Γ is discrete; then by (1) Γ acts properly discontinuously on X. Let
x, y ∈ X with p(x) 6= p(y), i.e. x ∈ / Γy (and equivalently,
 y∈/ Γx). Let U, V be neighborhoods

U , V are compact. Then γ ∈ Γ | γ(U ∪ V ) ∩ (U ∪ V ) 6= ∅ is finite, say
of x, y inX such that
equal to γ1 , ..., γr . Now by assumption x 6= γi (y) (resp. y 6= γi (x)) for i = 1, ..., r, so we may
assume after possibly
Sr / γi (V ) (resp. y ∈
shrinking U (resp.S V ) that x ∈ / γi (U )) for i = 1, ..., r.
0 0 r 0
Let U = U \ i=1 γi (V ) and V = V \ i=1 γi (U ). Then U is a neighborhood of x in X not
intersecting ΓV 0 , and V 0 is a neighborhood of y in x not intersecting ΓU 0 , therefore p(U 0 ) and
p(V 0 ) are disjoint neighborhoods of p(x) and p(y) in X/Γ, hence X/Γ is Hausdorff.

Now assume moreover that Γ acts freely on X. The fact that the quotient map p is a
covering map then follows from the following:

Lemma 2.2.4 Given x ∈ X, there exists a neighborhood U of x in X such that:

∀γ ∈ Γ \ {Id} γU ∩ U = ∅.

Proof. This follows essentially the same lines as the previous argument. Let V be a neighbor-
hood of x in X with compact closure; by proper discontinuity of the action of Γ on X, the set
γ ∈ Γ | γV ∩V 6= ∅ is finite, say equal to Id, γ1 , ..., γr . Note that for i = 1, ..., r, γi−1 (x) 6= x.
 

Let W ⊂ V be another neighborhood

Sr of x with compact closure, such that γi−1 (x) ∈ / W for
i = 1, ..., r. Let U = W \ i=1 γi−1 (W ). Then U is a neighborhood of x inX, and: (a)
γi U ∩ U = ∅ for i = 1, ..., r by definition of U , and (b) γU ∩ U = ∅ for γ ∈ Γ \ Id, γ1 , ..., γr
by definition of W . 

Finally, the fact that the covering map p is also a local isometry follows from the fact that
each element of Γ is an isometry. This concludes the proof of Proposition 2.2.1. 

Moral: Given a discrete group Γ < Isom(HnK ) for K = R or C (resp. a discrete group acting
freely), we get a hyperbolic orbifold (resp. manifold) HnK /Γ. This raises the following question:

Question: How can one find/construct discrete subgroups of Isom(HnK ) for K = R or C?

We will investigate this question in detail in the remaining sections, where we give an
overview of two main families of constructions: arithmetic constructions, leading to so-called
arithmetic lattices, and more geometric constructions such as Coxeter groups and more generally
reflection groups.
We finish this section by pointing out that for discrete subgroups of Isom(HnK ) for K = R
or C, acting freely is nothing more than being torsion-free (i.e. having no nontrivial element of
finite order):

Lemma 2.2.5 Let Γ < Isom(HnK ) be discrete (with K = R or C). Then, for γ ∈ Γ \ {Id}:
γ has a fixed point in HnK ⇐⇒ γ has finite order. In particular:
Γ acts freely on HnK ⇐⇒ Γ is torsion-free.

39
Note that this is not true in general, for example in Isom(S n ) the antipodal map x 7→ −x has
order 2 but has no fixed point in S n .

Proof. If γ has a fixed point p then hγi = γ n | n ∈ Z ⊂ StabG (p) ∩ Γ is finite as StabG (p)
is compact and Γ discrete, hence γ has finite order.
Conversely, if γ has finite order then γ has a fixed point in HnK . This follows from the
classification of isometries (into elliptic, parabolic and loxodromic isometries) by noting that
parabolic and loxodromic isometries have infinite order. 

2.3 Structure theorems for finitely generated linear groups

and hyperbolic lattices
In this section we give a brief overview of structure theorems for finitely generated linear groups
and hyperbolic lattices, namely: the Selberg lemma and Tits alternative for f.g. linear groups,
the Margulis lemma fro discrete groups of hyperbolic isometries, and Mostow rigidity and finite
presentation of hyperbolic lattices.

2.3.1 Finitely generated linear groups

Theorem 2.3.1 (Selberg’s lemma) If Γ is a finitely generated subgroup of GL(n, C) for
some n, then Γ has a torsion-free subgroup Γ0 of finite index.
Using Proposition 2.2.1 and Lemma 2.2.5, this can be rephrased as follows for discrete groups
of hyperbolic isometries (which we have seen to be linear):
Corollary 2.3.2 If Γ is a finitely generated subgroup of Isom(HnK ), the hyperbolic orbifold HnK /Γ
is finitely covered by a hyperbolic manifold HnK /Γ0 .
Note: In Selberg’s lemma one can in fact require Γ0 to be normal in Γ as well, giving a normal
finite cover in the Corollary.
Idea of Proof: Reduce entries mod. maximal ideals of the ring generated (over Z) by the
entries of the generators of Γ – since Γ is finitely generated this ring is finitely generated. For
example, if Γ < GL(n, Z), consider Γp = Γ ∩ Kerφp , where φp : GL(n, Z) 7−→ GL(n, Z/pZ) is
reduction mod. a prime p. Then Γp (is normal and) has finite index in Γ, and one can show
that for p large enough Γp is torsion-free.

For the next statement we recall the definitions of solvable and virtually solvable groups. A
group G is solvable if its derived series terminates, i.e. if there exists n > 1 such that Gn = Id,
where the derived series (Gn ) is defined inductively as follows:

G0 = G
Gn+1 = [Gn , Gn ],
where [H, K] denotes the subgroup generated by all commutators of the form [h, k] = hkh−1 k −1
with h ∈ H and k ∈ K. The group G is called virtually solvable if it has a solvable subgroup
of finite index.

40
Theorem 2.3.3 (Tits alternative) If Γ is a subgroup of GL(n, C) for some n, then either Γ
contains a non-abelian free group, or Γ is virtually solvable.

Idea of Proof: The main ingredient is what Tits called the Ping-Pong lemma, which is a
reformulation of Klein’s combination theorem, and which in this formulation is very flexible
and can be applied in many different contexts (see for example [dlH]):

Lemma 2.3.1 (Ping-Pong lemma) Let Γ be a group acting on a set X, and let a1 , a2 ∈ Γ.
− −
Assume that there exist 4 non-empty, pairwise disjoint subsets A+ +
1 , A1 , A2 , A2 ⊂ X such that,
for i = 1, 2:
ai (X \ A− +

i ) ⊂ Ai
a−1 + −
i (X \ Ai ) ⊂ Ai .

Then ha1 , a2 i ' F2 .

k
Sketch of Proof of Ping-Pong lemma: Let g = ak11 ak22 ...a1r−1 ak2r be a (non-trivial) reduced
word in ha1 , a2 i, with all ki 6= 0 except possibly k1 and kr . Assume for instance that k1 6= 0
and kr 6= 0. Note that ak1 (A± ± k ± ± ± ±
2 ) ⊂ A1 and a2 (A1 ) ⊂ A2 , therefore g(A2 ) ⊂ A1 , hence g 6= Id.
The other cases are handled similarly. 

Idea of proof of Tits alternative: One can show that, unless Γ is virtually solvable, Γ contains
2 non-commuting matrices a1 , a2 , each with at least 2 eigenvalues of distinct absolute values.
Then the eigenspaces for the eigenvalues with extremal absolute values, λmax i /λmin
i correspond
n−1
to attracting/repulsing fixed points for ai acting on CP . One can then apply the Ping-Pong
lemma, taking A± i to be suitable neighborhoods of the attracting/repulsing fixed points of ai .

2.3.2 Discrete hyperbolic isometry groups

We first introduce an important invariant of isometries of metric spaces. Given an isometry γ
of a metric space X, the displacement (or translation length) of γ is defined as:

dγ = Inf d(x, γx) | x ∈ X .

This allows to roughly classify all isometries of any metric space into 3 types. An isometry γ
of X is called:

• hyperbolic if dγ > 0 and is attained, i.e. (∃x ∈ X) d(x, γx) = dγ ,

• elliptic if dγ = 0 and is attained, i.e. γ has a fixed point in X, and

• parabolic if dγ is not attained.

If X is negatively curved then in the latter case one can show that dγ = 0. It is left to the
reader as an exercise that this coincides with the previous classification of isometries in the case
of X = HnK , where we defined γ to be loxodromic (rather than hyperbolic) if it has no fixed
point in X and exactly 2 fixed points on ∂X, and parabolic if it has no fixed point in X and
exactly 1 fixed point on ∂X.

41
 The following result describes the structure of hyperbolic isometry groups for which some
point is not moved very much by the generators; informally one can think of it as giving a
uniform lower bound on the size of an open ball contained in any hyperbolic orbifold of a given
dimension. It is usually called the Margulis lemma, but is essentially due to Kazhdan-Margulis,
based on earlier work by Zassenhaus.
For the statement we recall the definitions of nilpotent and virtually nilpotent groups. A
group G is nilpotent if its lower central series terminates, i.e. if there exists n > 1 such that
G0n = Id, where the lower central series (G0n ) is defined inductively as follows:
 0
G0 = G
G0n+1 = [G0n , G],
where as before [H, K] denotes the subgroup generated by all commutators of the form [h, k] =
hkh−1 k −1 with h ∈ H and k ∈ K. The group G is called virtually nilpotent if it has a nilpotent
subgroup of finite index.
Theorem 2.3.4 (Margulis lemma) Given K = R or C and n > 1, there exists a constant
µ = µ(K, n) such that, for any discrete subgroup Γ < Isom(HnK ), if there exists a point x ∈ HnK
and a generating set S for Γ such that d(x, γx) 6 µ for all γ ∈ S, then Γ is virtually nilpotent.

2.3.3 Hyperbolic lattices

If G is a Lie group (or more generally, a locally compact topological group), a subgroup Γ < G
is called a lattice in G if Γ is discrete in G and G/Γ carries a finite G-invariant measure (where
G acts on G/Γ by left translation). This holds in particular if G/Γ is compact.
For a Lie group G with a Haar measure (a regular measure invariant under left- or right-
translation), this is equivalent to requiring that the Haar measure on G descends to a finite
measure on G/Γ. For a connected semisimple Lie group G with associated symmetric space
X = K\G (e.g. G = Isom+ (HnK ) and X = HnK ), this is in turn equivalent to the condition that
Γ acts on X with finite covolume, i.e. that X/Γ has finite volume.

Examples: The most basic example of a lattice is Z in R, and more generally Zn in Rn .

We’ll see many other examples in the remaining sections, including the following: SL(2, Z) is a
lattice in SL(2, R), more generally SL(n, Z) is a lattice in SL(n, R) and this is the prototype of
what is called an arithmetic lattice. We will see other examples provided by reflection groups
(which also generalize SL(2, Z), in a different direction).
The following notion gives a convenient way of showing that a group acting on a topological
space is finitely generated or finitely presented. Assume Γ is a group acting properly discontin-
uously on a topological space X. A subset D ⊂ X is called a coarse fundamental domain for
the action of Γ on X if: [
(a) γD = X, and
γ∈Γ

(b) γ ∈ Γ | γD ∩ D 6= ∅ is finite.
The fact that Γ is finitely generated or finitely presented can be obtained via the structure of
a coarse fundamental domain as follows (see [WM] for this formulation, and ideas going back
to Weil):

42
Proposition 2.3.1 Let Γ be a group acting properly discontinuously on a topological space X.
(1) If X is connected and there exists an open coarse fundamental domain for Γ acting on X,
then Γ is finitely generated.
(2) If X is simply-connected and there exists a connected open coarse fundamental domain for
Γ acting on X, then Γ is finitely presented.

The idea for part (1) is to show that Γ is generated by the elements γ such that γD ∩ D 6= ∅.
The second part is more complicated, and in fact for hyperbolic lattices (more precisely, lattices
in rank-1 semisimple real Lie groups), Garland and Raghunathan proved the following result
by producing a general construction of fundamental domains with finitely many faces for such
groups:

Theorem 2.3.5 Lattices in Isom(HnK ) (K = R, C, n > 1) are finitely presented.

The following result is fundamental for the study of the geometry and topology of hyperbolic
manifolds. It states that finite-volume hyperbolic manifolds are determined up to isometry by
their fundamental group, in dimensions > 3. This is a strong rigidity result, implying in
particular that one cannot deform the hyperbolic structure on a hyperbolic manifold with fixed
topology – except in the case of surfaces, see below (the latter statement is known as local
rigidity). In general it is not even known if (closed, aspherical) manifolds with isomorphic
fundamental groups are necessarily homeomorphic; this is known as the Borel conjecture. The
following result is originally due to Mostow, in the case where HnK /Γ is compact, and was
extended by Prasad to the case where HnK /Γ is non-compact bu thas finite volume.

Theorem 2.3.6 (Mostow rigidity) Let Γ1 , Γ2 be two lattices in Isom(HnK ) (with n > 3 if
K = R and n > 2 if K = C). If Γ1 , Γ2 are isomorphic as groups then there exists g ∈ Isom(HnK )
such that Γ2 = gΓ1 g −1 ; in particular the orbifolds HnK /Γ1 and HnK /Γ2 are isometric.

The restriction on dimension is necessary because it is well-known that, in the case of H2R ' H1C ,
hyperbolic surfaces can be deformed (in fact, they have a rich deformation theory). More
precisely, given a closed srface Σg with g > 2, there exists a (6g − 6)-dimensional space of
pairwise non-isometric hyperbolic structures on Σg , called the Teichmüller space of Σg (and
this space is path-connected, in fact homeomorphic to R6g−6 , so any 2 such structres can be
defomed to each other). Succinctly, this can be seen by decomposing the surface into 2g−2 pairs
of pants (spheres minus 3 disks, with geodesic boundary). The hyperbolic structure on eacg
such pair of pants is determined by 3 real numbers (the lengths of the boundary components,
or cuffs); when we glue the various pieces by isometries we require the lengths of the cuffs that
we glue to be equal, but there is an extra ”twist” parameter at each such gluing. The 6g − 6
length and twist parameters are called Fenchel-Nielsen coordinates on Teichmüller space.

2.4 Arithmetic lattices

We start by giving examples of arithmetic lattices, in families of increasing complexity, before
giving the general definition of arithmetic lattices in linear algebraic groups and semisimple real

43
Lie groups.

(1) First examples:

√
If we take a discrete subring such as Z ⊂ R or Od ⊂ C (the ring of integers of Q[ −d]
for some square-free integer d, e.g. Z[i] when d = 1 and Z[e2iπ/3 ] when d = 3), we get dis-
crete matrix groups by taking entries in the discrete subring, e.g. SL(n, Z) < SL(n, R) and
SL(n, Od ) < SL(n, C), or in the hyperbolic world SO(n, 1, Z) < SO(n, 1) and SU(n, 1, Od ) <
SU(n, 1). Discreteness of such subgroups is obvious (because the topology on the matrix group
is the same as the product topology on the entries, i.e. a sequence of matrices converges to
Id if and only if each entry converges to the correponding entry of Id). It turns out that all
of the subgroups above are in fact lattices in the corresponding Lie group; in full generality
this follows from the Borel–Harish-Chandra theorem (Theorem 2.4.1 below), but some cases
were previously known. The case of SL(2, Z) < SL(2, R) is essentially due to Gauss, and Siegel
proved that Γ = SL(n, Z) has finite covolume in G = SL(n, R) by exhibiting an explicit region
(in coordinates in the KAN decomposition, now called a Siegel region) with finite Haar mea-
sure in G, whose cosets under Γ cover G.

(2) The Galois trick, part I:

√
The subgroup Z[ 2] is dense in R (this follows from √
the fact that additive subgroups of R
are either cyclic or dense). However, one can realize Z[ 2] as a discrete subgroup (in fact a
lattice) of R2 , by the following embedding:
√
ϕ: Z[√ 2] −→ R2 √ √
(a + b 2) 7−→ (a + b 2, a − b 2).
√
The fact that Im ϕ is a lattice in R2 can be easily seen by writing it as Z(1, 1)T ⊕ 2Z(1, −1)T .
Note that the projection of Im ϕ to each factor is dense.
√ √ √
Analogously, denoting σ : a + b 2 7−→ a − b 2 the non-trivial Galois conjugation in Z[ 2]
and σ M the matrix obtained from a matrix M by applying σ to all entries, the embedding:
√
ϕ : SL(2, Z[ 2]) −→ SL(2, R) × SL(2, R)
M 7−→ (M, σ M )
√ √

realizes SL(2, Z[ 2]) as a lattice Γ = ϕ SL(2, Z[ 2]) in SL(2, R) × SL(2, R) (this will follow
from the Borel–Harish-Chandra theorem). Such quotients (H2R × H2R )/Γ are called Hilbert mod-
ular surfaces.

More generally, for any totally real number field E with Gal(E/Q) = {σ1 , ..., σr } (and
n > 1), we have an embedding:

ϕ : SL(n, OE ) −→ SL(n, R) × ... × SL(n, R)

M 7−→ (σ1 M, ..., σr M ).

44


The Borel–Harish-Chandra theorem will tell us that ϕ SL(n, OE ) is a lattice in
SL(n, R) × ... × SL(n, R). This is perfectly fine, but this construction produces lattices in
higher-rank Lie groups, whereas the hyperbolic isometry groups we are interested in have rank
1. More precisely, the real rank of a linear Lie group G < GL(n, C) is the maximal dimension
of an R-split torus, i.e. a closed, connected subgroup of G all of whose elements are simultane-
ously diagonalizable over R. For example, a maximal R-split torus in SL(n, R) is the subgroup
of diagonal matrices, so SL(n, R) has real rank n − 1. Geometrically, if G is semisimple the
real rank of G can be seen in the symmetric space X = G/K associated to G (where K is a
maximal compact subgroup of G), namely the real rank of G is then the maximal dimension
of a flat in X, which is an isometrically embedded, totally geodesic Euclidean subspace. This
shows that the hyperbolic spaces HnK have real rank 1, as do their isometry groups which are
(up to finite index) SO(n, 1) and SU(n, 1). Note that real rank is additive under products, so
even in the case of SL(2, R) (which has rank 1), when we take products of more than one copy
we get a higher-rank Lie group. However the Galois trick can be upgraded to produce lattices
in rank 1 Lie groups, as we now illustrate.

(3) The Galois trick, part II:

When considering Lie groups definied by a quadratic or Hermitian form such as SO(n, 1)
and SU(n, 1), the second part of the Galois trick is to use a form which changes under the
Galois conjugations
√ of the number field under consideration.
√ We start again with the example
of E = Q[ 2] with its ring of integers OE = √ Z[ 2]. The quadratic form on Rn+1 given
by the symmetric matrix Q = Diag(1, ..., 1, − 2) has signature (n, 1), so the group SO(Q)
is conjugate √ to SO(n, 1) in GL(n + 1, R), whereas the quadratic form corresponding to σ Q =
Diag(1, ..., 1, 2) has signature (n + 1, 0), so the group SO(σ Q) is conjugate to SO(n + 1)
and in particular
√
is compact. As above, the Borel–Harish-Chandra theorem will tell us that
ϕ SO(Q, Z[ 2]) is a lattice in SO(Q) × SO(σ Q), where as before:
√
ϕ : SO(Q, Z[ 2]) −→ SO(Q) × SO(σ Q)
M 7−→ (M, σ M ).

But now the projection to the first factor p1 : SO(Q) × SO(σ Q) −→ SO(Q) has compact kernel,
hence it maps discrete sets to discrete sets, and lattices to lattices, by the following lemma,
whose proof is left to the reader as an excercise.

Lemma 2.4.1 Let G,H be 2 (locally compact) topological groups, and φ : G −→ H a continuous
homomorphism with compact kernel. Then, if Γ < G is a discrete subgroup (resp. a lattice),
then φ(Γ) < H is a discrete subgroup (resp. a lattice).
√
Therefore SO(Q, Z[ 2]) is a lattice in SO(Q) ' SO(n, 1). More generally, the same argu-
ment gives the following:

Proposition 2.4.1 Let E be a totally real number field and Q a quadratic form of signature
(n, 1) with coefficients in E. If for all σ ∈ Gal(E/Q) \ {Id} the form σ Q is definite then
SO(Q, OE ) is a a lattice in SO(Q) ' SO(n, 1).

45


Sketch of proof: The Borel–Harish-Chandra theorem will tell us as before that ϕ SO(Q, OE )
is a lattice in SO(Q) × SO(σ1 Q) × ... × SO(σr Q), denoting Gal(E/Q) = {Id, σ1 , ..., σr }, via the
embedding:
ϕ : SO(Q, OE ) −→ SO(Q) × SO(σ1 Q) × ... × SO(σr Q)
M 7−→ (M,σ1 M, ..., σr M ).
But the projection to the first factor p1 : SO(Q) × SO(σ1 Q) × ... × SO(σr Q) −→ SO(Q) has
compact kernel, by the assumption that all the nontrivial Galois conjugates σ Q are positive
definite or negative definite, hence SO(Q, OE ) is a a lattice in SO(Q) by Lemma 2.4.1. 

The analogous statement in the unitary setting is the following:

Proposition 2.4.2 Let E be a purely imaginary quadratic extension of a totally real number
field F and H a Hermitian form of signature (n, 1) with coefficients in E. If for all σ ∈
Gal(E/Q) not inducing the identity on F the form σ H is definite then SU(H, OE ) is a a lattice
in SU(H) ' SU(n, 1).
This is in fact a characterization of all arithmetic lattices in GL(n + 1, C) with entries in OE
and preserving a Hermitian form of signature (n, 1) defined over E, see Proposition 2.7.4 below.

(4) General arithmetic lattices:

Recall that a real linear algebraic group defined over Q is a subgroup G of GL(n, R) for
some n, such that the elements of G are precisely the solutions of a set of polynomial equations
in the entries of the matrices, with the coefficients of the polynomials lying in Q; one denotes
G(R) = G and G(Z) = G ∩ GL(n, Z).

Theorem 2.4.1 (Borel–Harish-Chandra) If G is a linear algebraic group defined over Q

then G(Z) is a lattice in G(R).

This result implies that any real semisimple Lie group has infinitely many (distinct com-
mensurability classes of) lattices, both cocompact and non cocompact, see [WM].
One then obtains the general definition by extending this notion to all groups equivalent to
such groups G(Z) in the following sense:

Definition: Let G be a semisimple Lie group, and Γ a subgroup of G. Then Γ is an

arithmetic lattice in G if there exist an algebraic group S defined over Q and a continuous
homomorphism φ : S(R)0 −→ G with compact kernel such that Γ is commensurable to φ(S(Z)∩
S(R)0 ).
Note that by definition, if Γ1 , Γ2 < G are commensurable then Γ1 is an arithmetic lattice in
G if and only if Γ2 is. (We say that 2 subgroups Γ1 , Γ2 < G are commensurable if Γ1 ∩ gΓ2 g −1
has finite index in Γ1 and in gΓ2 g −1 for some g ∈ G).
The fact that a subgroup Γ as in the definition is a lattice in G follows from the Borel–
Harish-Chandra theorem. We now briefly explain why the groups constructed with the Galois
tricks, parts I and II above, are instances of this general definition.
The main ingredient is restriction of scalars; in its simplest form, this is the fact that in a
field extension E/F , E naturally has the structure of a vector space over F . In the particular

46
case where E is a number field (i.e. a finite extension of Q), denoting d = [E : Q], selecting
a basis for E over Q gives an embedding i : E ,→ GL(d, Q) by sending any element x ∈ E to
multiplication by x (seen as a Q-linear endomorphism of the Q-vector space E). Now if we
choose this basis to be a Z-basis for OE , then additionally: i(OE ) = i(F ) ∩ GL(d, Z).
The groups appearing in the Galois trick part I are of this form, taking SL rather than
GL which is obviously a linear algebraic group defined over Q (as the polynomial defining
it is the determinant, which is an integer polynomial in the entries of a matrix). For the
groups appearing in the Galois trick, part II, we only need to argue additionally that the group
S = SO(Q) × SO(σ1 Q) × ... × SO(σr Q) (respectively S = SU(H) × SU(σ1 H) × ... × SU(σr H)) is
a linear algebraic group defined over Q; the continuous homomorphism φ : S(R)0 −→ G with
compact kernel appearing in the definition of an arithmetic lattice is then simply projection
onto the first factor of S. (As pointed out above this has compact kernel by the assumption
that all non-trivial Galois conjugates of Q (resp. H) are definite).
The fact that such groups S are linear algebraic groups defined over Q follows again from
restriction of scalars. More precisely, a product of the form S = SO(Q)×SO(σ1 Q)×...×SO(σr Q)
can be embedded in block form into SL((n + 1)(r + 1), R) (where (n + 1) is the rank of Q and
(r + 1) the number of factors). The condition of being a block matrix of a given shape is
given by polynomial equations over Q in the entries (namely, that the appropriate entries are
0). Now for each block Ai ∈ SL(n + 1, R), the condition of being in SO(σi Q) is given by the
equation ATi σi QAi = σi Q, which corresponds to (n + 1)2 polynomial equations over E in the
entries of Ai . Now as above, choosing a Q-basis for E allows us to write each of these equations
as a system of d polynomial equations over Q. Putting the pieces back together, we see that
S = SO(Q)×SO(σ1 Q)×...×SO(σr Q) is a subgroup of SL((n+1)(r +1), R) given by a collection
of polynomial equations over Q, i.e. S is a linear algebraic group defined over Q.

2.5 Arithmeticity vs. non-arithmeticity

It turns out that the arithmetic construction is the only way to produce lattices in higher-rank
simple Lie groups; this is the following deep and celebrated result of Margulis, for which he was
awarded a Fields medal in 1978 (see Section 2.4 (2) for the definition of real rank).

Theorem 2.5.1 (Margulis arithmeticity theorem) If G is a simple real Lie group of real
rank at least 2, then any lattice in G is arithmetic.

This result also holds when G is a semisimple real Lie group of real rank at least 2, under
the additional assumption that the lattice is irreducilble, i.e. its projection onto each simple
factor of G is dense. This excludes cases like Γ × Γ < G × G with Γ a non-arithmetic lattice in
a rank-1 simple group G, e.g. G = PSL(2, R).
We now examine to which extent the simple Lie groups of real rank 1, where the Margulis
arithmeticity theorem does not apply, are known to admit non-arithmetic lattices. The first
observation is the following:

Observation: SL(2, R) has uncountably many non-arithmetic lattices.

47
 Indeed, as pointed out in the discussion of Mostow rigidity, most lattices in SL(2, R) (in-
cluding all closed surface groups) have families of continuous deformations to isomorphic but
non-conjugate lattices (the Teichmüller space of the surface). But there are only countably
many arithmetic lattices up to conjugacy in any given Lie group (finitely many defined over
any given number field); this follows for example from Tits’ classification of linear algebraic
groups defined over Q ([Ti]), and in the case of SL(2, R) it follows from a result of Borel
([Borel]) that there are in fact only finitely many arithmetic lattices in the Teichmüller space
of a closed surface with fixed genus.
From the classification of simple real Lie groups (due to Killing and E. Cartan), the com-
plete list of simple real Lie groups of real rank 1 is the following. For simplicity, we list groups
G which are subgroups of GL(n, C) for some n; they are not simple but have finite center,
the correpsonding simple group is G/Z(G) which is finitely covered by G. We also list the
corresponding symmetric spaces G/K (with K < G a maximal compact subgroup), which are
exactly the hyperbolic spaces over R, C, H and the exotic octave plane.

Simple real Lie groups of rank 1: (up to finite index)

(−20)
Group: SO(n, 1) SU(n, 1) Sp(n, 1) F4

Symmetric space: HnR HnC HnH H2O

The Margulis arithmeticity theorem was extended as follows to the groups corresponding
to quaternions and octaves by Corlette and Gromov–Schoen:
(−20)
Theorem 2.5.2 (Corlette, Gromov–Schoen) Any lattice in Sp(n, 1) or F4 is arith-
metic.

In real hyperbolic space, beyond dimension 2, Makarov and Vinberg gave the first exam-
ples of non-arithmetic lattices in SO(n, 1) for n = 3, 4, 5 in the 1960’s. Their examples were
hyperbolic Coxeter groups, whose construction will be discussed in detail in the next section.
Subsequently, Gromov and Piatetskii-Shapiro gave in [GPS] a construction producing non-
arithmetic lattices in SO(n, 1) for all n > 2 (and in fact, infinitely many, non-commensurable,
both cocompact and non-cocompact in each dimension).

Theorem 2.5.3 (Gromov–Piatetskii-Shapiro) There exist non-arithmetic lattices in SO(n, 1)

for all n > 2.

We briefly sketch the idea of the construction. The Gromov–Piatetski-Shapiro construction,

which they call interbreeding of 2 arithmetic lattices (often referred to as hybridation), produces
a lattice Γ < PO(n, 1) from 2 lattices Γ1 and Γ2 in PO(n, 1) which have a common sublattice
Γ12 < PO(n − 1, 1). Geometrically, this provides 2 hyperbolic n-manifolds V1 = Γ1 \HRn and
V2 = Γ2 \HRn with a hyperbolic (n − 1)-manifold V12 which is isometrically embedded in V1
and V2 as a totally geodesic hypersurface. This allows one to produce the hybrid manifold V
by gluing V1 − V12 and V2 − V12 along V12 . The resulting manifold is also hyperbolic because
the gluing took place along a totally geodesic hypersurface, and its fundamental group Γ is
therefore a lattice in PO(n, 1). The point is then that if Γ1 and Γ2 are both arithmetic but
non-commensurable, their hybrid Γ is non-arithmetic.

48
 Much less is known in the remaining case of complex hyperbolic space, corresponding to
SU(n, 1). The first constructions of complex hyperbolic lattices (beyond dimension 1) go back to
Picard, in [Pic1] and [Pic2]. In fact it turns out that some of the groups constructed in [Pic2] are
non-arithmetic lattices, but this was only discovered later. Mostow constructed the first known
non-arithmetic lattices in SU(2, 1) in [Mos1]; we will decsribe his construction in more detail in
the last section. Deligne and Mostow subsequently extended Picard’s construction (which was
based on monodromy groups for so-called hypergeometric functions) up to (complex) dimension
5 in [DM] (Mostow pushed the construction to dimension 9 in [Mos2]). It turns out that, beyond
Picard’s groups, there is only one non-arithmetic lattice in the Deligne–Mostow construction
in dimension 3 (it is non-cocompact). We summarize this in the following:
Theorem 2.5.4 (1) (Mostow) There exist non-arithmetic lattices in SU(2, 1).
(2) (Deligne–Mostow) There exists a non-arithmetic lattice in SU(3, 1).
In the 2-dimensional case, the Picard and Mostow groups each provided 7 non-arithmetic
lattices; considerations of trace fields show that these fell in between 7 and 9 distinct commen-
surability classes, and it was shown in [KM] that there are in fact 9. Recently, Deraux, Parker
and the author have extended Mostow’s construction to produce 5 additional commensurability
classes of non-arithmetic lattices in SU(2, 1) (see [DPP1], [DPP2] and the last section).
The following outstanding questions remain open in the complex hyperbolic case:

Open questions:
(1) Do there exist non-arithmetic lattices in SU(n, 1) when n > 4?
(2) For fixed n > 2, do there exist infinitely many non-commensurable non-arithmetic lattices
in SU(n, 1)?

2.6 Hyperbolic Coxeter groups

Let X denote one of the 3 model spaces of constant curvature, X = Sn , En or Hn . A polyhedron
in X is a subset P of X which is bounded by finitely many hyperplanes, in other words P is
the intersection of finitely many closed half-spaces. A wall of P is a hyperplane H such that
P ∩ H has dimension n − 1. In that case we will say that P ∩ H is a (maximal) face of P ; in
turn this polyhedron of dimension n − 1 has faces, and inductively this defines the set of all
faces of P . (Alternatively, in such a constant curvature space one can also construct P from
the bottom up, by taking convex hulls of vertices). To avoid all confusion we will call k-face of
P any of its faces of dimension k.
A Coxeter polyhedron in X is a polyhedron P all of whose dihedral angles are submultiples
of π. (This means that if Hi and Hj are 2 walls of P , then either they are disjoint or at
any intersection point their outward-pointing normals form an angle of measure π − π/mij
where mij ∈ N ∪ {∞}). The point of this condition is that the reflections across the walls of
such a polyhedron generate a discrete group of isometries of X, for which P is a fundamental
polyhedron. More precisely, we have the following result which is due to Poincaré when X = H2 ,
to Coxeter when X = S n or E n , and in general follows from the more general version of what
is now called the Poincaré polyhedron theorem (see section 2.7.5 and [EP]).

49
 Figure 2.1: Coxeter diagram for H5 , a simplex in H4R .

Theorem 2.6.1 Let P be a Coxeter polyhedron in X = Sn , En or Hn , denote H1 , ..., Hk the

walls of P , r1 , ..., rk the reflections across these walls and π/mij the dihedral angle between the
walls Hi and Hj (where mij ∈ N ∪ {∞}, with the convention that mii = 2 for all i). Then:
(1) Γ = hr1 , ..., rk i is a discrete subgroup of Isom(X),
(2) P is a fundamental domain for the action of Γ on X, and
(3) Γ admits the presentation hr1 , ..., rk |(ri rj )mij = 1 (for 1 6 i 6 j 6 k)i.
An abstract group given by a presentation of the form in part (3) is called a Coxeter group.

A convenient way to encode the information about such a Coxeter group Γ is by means
of the associated Coxeter diagram DΓ ([Co]) which is a graph with labeled edges, constructed
as follows. The vertices of DΓ correspond to the generating reflections r1 , ..., rk ; two distinct
vertices corresponding to ri and rj are joined by an edge if (a) mij > 3, in which case the edge
is labeled mij (alternatively, an edge of multiplicity mij − 2 is drawn), or (b) by a bold (resp.
dotted) edge if the walls of ri and rj are parallel (resp. ultraparallel), in the hyperbolic case.
The point of this convention is that connected components of DΓ correspond to irreducible
factors of Γ.
The numbers mij also determine a quadratic form via a symmetric matrix called the Gram
matrix of the polyhedron or group. Namely, if Γ is generated by the k reflections r1 , ..., rk
with mij as above, then the Gram matrix of Γ is the symmetric k × k matrix with 1’s on the
diagonal and (i, j)-entry equal to − cos(π/mij ) (resp. − cosh d(Hi , Hj ) if the walls Hi and Hj
are parallel or ultraparallel). This is the Gram matrix of the outward-pointing normal vectors
to the walls of P . One can then define invariants such as the determinant, rank, signature of
P (or Γ or DΓ ) as the corresponding invariant of its Gram matrix. Accordingly, we will say
that DΓ is hyperbolic if its Gram matrix is of signature (n, 1) for some n > 1, parabolic if it is
positive semidefinite (but not definite), and elliptic if it is positive definite.
The Gram matrix represents a quadratic form on Rk which is preserved by the so-called
“geometric representation” (see e.g. [H]). More relevant from our point of view is the repre-
sentation that we started with (of Γ in Isom(Hn ) < PGL(n + 1, R) arising from the polyhedron
P in Hn ). These representations coincide only when the polyhedron is a simplex, that is when
the Gram matrix has maximal rank.
Elliptic and parabolic Coxeter diagrams were classified by Coxeter in 1934 ([Co], see pp.
202–203 of [V1] and [H]), whereas no such classification is known for hyperbolic diagrams,
except in dimensions 2 (Poincaré) and 3 by Andreev’s theorem ([An]). We now state some
partial results, see [V1] for more details. Vinberg has proved the following general existence
result for acute-angled hyperbolic polyhedra (Theorem 2.1 of [V2]):
Theorem 2.6.2 (Vinberg) Let G be an indecomposable symmetric matrix of signature (n, 1)
with 1’s on the diagonal and non-positive entries off it. Then there exists a convex polyhedron
in Hn whose Gram matrix is G, and this polyhedron is unique up to isometry.
The following results are in striking contrast to the Euclidean and spherical cases, for which
compact Coxeter polyhedra exist in all dimensions:

50
 Figure 2.2: Coxeter diagram for Bugaenko’s compact polyhedron in H8R .

Theorem 2.6.3 (Vinberg) There do not exist compact Coxeter polyhedra in Hn when n > 30.

Theorem 2.6.4 (Prokhorov-Khovanskij) There do not exist finite-volume Coxeter polyhe-

dra in Hn when n > 996.

There is a large gap between the dimension bounds given in these theorems and the dimensions
where compact/finite-volume Coxeter polyhedra are known to exist. The largest dimension of
a known compact hyperbolic Coxeter polyhedron is 8 (due to Bugaenko, [Bu]), whose Coxeter
diagram is pictured in Figure 2.2; the largest dimension of a known finite-volume hyperbolic
Coxeter polyhedron is 21 (due to Borcherds, [Borch]). Both of these are arithmetic (i.e. the
corresponding reflection group is an arithmetic lattice). Non-arithmetic compact Coxeter poly-
hedra are known to exist in Hn for n 6 5 (the first examples were due to Makarov [Mak] for
n = 3, others can be found among the simplex groups); non-arithmetic finite-volume Coxeter
polyhedra are known to exist in Hn for n 6 10 (Ruzmanov, [Ru]).
We now review some results concerning finiteness of finite-volume Coxeter polyhedra/groups
in any given dimension. Allcock has proved the following ([Al]):

Theorem 2.6.5 (Allcock) For all 3 6 n 6 19, there exist infinitely many non-isometric
finite-volume Coxeter polyhedra in Hn .

The idea of Allcock’s construction is to start with a given Coxeter polyhedron, such that
all the dihedral faces along a given face are right angles; such a polyhedron is called doublable:
one can reflect it along that face to obtain a new Coxeter polyhedron. Allcock finds polyhedra
which are redoublable in the sense that they satisfy the right-angled condition along 2 disjoint
faces, which allows to double it successively to produce an infinite sequence of pairwise distinct
Coxeter polyhedra. Note that by construction, all of the doubled/redoubled polyhedra are
commensurable to each other, and it is still unknown whether or not there exist infinitely many
distinct commensurability classes of finite-volume Coxeter polyhedra in Hn for n > 4 (for n = 2
triangle groups provide infinitely many such groups; for n = 3 it follows from Andreev’s theorem
[An] that there exist infinitely many such groups). However, it is known that the answer is
negative for arithmetic Coxeter groups; the following result is due to Borel for n = 2 with
bounded genus ([Borel]), Agol for n = 3 ([Ag]) and independently Agol–Belolipetsky–Storm–
Whyte ([ABSW]) and Nikulin ([N]) for n > 4:

Theorem 2.6.6 For any n > 2, there exist only finitely many maximal arithmetic finite-volume
Coxeter polyhedra in Hn .

51
2.7 Some non-arithmetic lattices in Isom(H2C)
In this section we give a brief overview of Mostow’s construction of non-arithmetic lattices in
SU(2, 1) from [Mos1], followed by a related construction by Deraux, Parker and the author from
[ParPau], [DPP1] and [DPP2].

2.7.1 Mostow’s lattices

Notation: Γ(p, t) < SU(2, 1), where p = 3, 4 or 5 and t is a real parameter.

The Γ(p, t) are symmetric complex reflection triangle groups, i.e.:

• Γ = hR1 , R2 , R3 i where each Ri is a complex reflection of order p.

• symmetric means that there exists an isometry J of order 3 such that JRi J −1 = Ri+1 , or
equivalently J(Li ) = Li+1 where Li = Fix(Ri ).

Moreover Mostow imposes the braid relation Ri Rj Ri = Rj Ri Rj .

Facts:
• For fixed p there is a 1-dimensional family of such groups (hence the t).
• Only finitely many of the Γ(p, t) are discrete; the discrete ones are lattices.
The second statement is the hardest, and is obtained by a numerical analysis of Dirichlet do-
mains for the groups Γ(p, t) acting on complex hyperbolic space. See [De] for a discussion of the
complexity of the corresponding Ddirichlet domains, and see [DFP] for a simpler construction
of fundamental domains for Mostow’s lattices. This produces 15 lattices in SU(2, 1), 7 of which
are non-arithmetic (see the Vinberg–Mostow arithmeticity criterion below). For concreteness,
the following are presentations for the Mostow lattices (more precisely, of the group hJ, R1 i, see
below) obtained in [DFP]:

Γ̃(p, t) = hJ, R1 , R2 | J 3 = R1p = R2p = J −1 R2 JR1−1 = R1 R2 R1 R2−1 R1−1 R2−1

= (R2 R1 J)k = ((R1 R2 )−1 J)l = Ii,
where p = 3, 4 or 5, k = ( 41 − 1
2p
+ 2t )−1 and l = ( 14 − 1
2p
− 2t )−1 .

2.7.2 Configuration spaces of symmetric complex reflection triangle

groups
We now work with Γ̃ = hR1 , Ji which contains Γ with index 1 or 3. We now drop the braid
relation.
Fact: (For fixed p) the space of such groups has dimension 2. More precisely:

Proposition 2.7.1 Let R1 be a complex reflection and J a regular elliptic isometry of order 3
in PU(2, 1). Then Γ̃ = hR1 , Ji is determined up to conjugacy by the conjugacy class of R1 J.

52
 In concrete terms, we parametrize the conjugacy class of R1 J by either:

• τ := TrR1 J (good for arithmetic), or

• the angle pair {θ1 , θ2 } of R1 J when elliptic (good for geometry).

Recall that the angle pair of an elliptic element of PU(2, 1) is the pair of angles of the eigen-
values of the correpsonding element of U(2).

Figure 2.3 illustrates the admissible angle pairs {θ1 , θ2 } for elliptic products R1 J when R1
is a complex reflection through angle 2π/p and J a regular elliptic isometry of order 3, for
p = 3, .., 10. For each fixed value of p the admissible region is the polygon bounded by the blue
line segments; the angle pairs corresponding to Mostow’s lattices (resp. to normal subgroups of
Mostow’s lattices) comprise the green (resp. red) segments. The crosses indicate the sporadic
groups, see below.
Notation: We denote Γ(2π/p, τ ) = hR1 , Ji, where R1 is a complex reflection through angle
2π/p, J a regular elliptic isometry of order 3, and τ := TrR1 J. For concreteness we give explicit
matrices for the generators and Hermitian form, denoting ψ = 2π/p:

 
0 0 1
J = 1 0
 0 (2.7.1)
0 1 0
 2iψ/3
−eiψ/3 τ

e τ
R1 =  0 e−iψ/3 0  (2.7.2)
−iψ/3
0 0 e

These preserve the Hermitian form hz, wi = w∗ Hτ z where

2 sin(ψ/2) −ie−iψ/6 τ

ieiψ/6 τ


Hτ =  ieiψ/6 τ 2 sin(ψ/2) −ie−iψ/6 τ  . (2.7.3)

−ie−iψ/6 τ ieiψ/6 τ 2 sin(ψ/2)

2.7.3 Sporadic groups

The following result ([ParPau]) gives a necessary condition for dicscreteness, under the assump-
tion that 2 given words in the group are elliptic or parabolic (see [Sc] for a discussion of why
these words are relevant), and provides the motivation to study the so-called sporadic groups.
Theorem 2.7.1 Let R1 be a complex reflection and J a regular elliptic isometry of order 3 in
PU(2, 1). Suppose that R1 J and R1 R2 = R1 JR1 J −1 are elliptic. If the group Γ = hR1 , Ji is
discrete then one of the following is true:

• Γ is one of Mostow’s lattices.

• Γ is a subgroup of one of Mostow’s lattices.

• Γ is one of the sporadic groups listed below.

53
 Figure 2.3: Polygons of configurations of Γ(2π/p, τ ) for p = 3, ..., 10

0
0 1 2 3 4 5 6

Mostow’s lattices correspond to τ = eiφ for some angle φ; subgroups of Mostow’s lattices to
τ = e2iφ + e−iφ for some angle φ, and sporadic groups are those for which τ takes one of the 18
values {σ1 , σ1 , ..., σ9 , σ9 } where the σi are given in the following list:

σ1 := eiπ/3 + e−iπ/6 2 cos(π/4) σ2 := eiπ/3 + e−iπ/6 2 cos(π/5)

σ3 := eiπ/3 + e−iπ/6 2 cos(2π/5) σ4 := e2πi/7 + e4πi/7 + e8πi/7
σ5 := e2πi/9 + e−iπ/9 2 cos(2π/5) σ6 := e2πi/9 + e−πi/9 2 cos(4π/5)
σ7 := e2πi/9 + e−iπ/9 2 cos(2π/7) σ8 := e2πi/9 + e−iπ/9 2 cos(4π/7)
σ9 := e2πi/9 + e−iπ/9 2 cos(6π/7).
Therefore, for each value of p > 3, we have a finite number of groups to study, the Γ(2π/p, σi )
and Γ(2π/p, σi ) which are hyperbolic (i.e. preserve a form of signature (2,1)). We determined
exactly which sporadic groups are hyperbolic; notably these exist for all values of p, and more
precisely:

Proposition 2.7.2 For p > 4 and τ = σ1 , σ2 , σ3 , σ4 , σ5 , σ6 , σ7 , σ8 or σ9 , Γ(2π/p, τ ) is hyper-

bolic.

2.7.4 Arithmeticity
The symmetric complex reflection triangle groups are all integral groups preserving a form, in
the following sense (recall that τ = Tr(R1 J)):

54
Proposition 2.7.3 The generators R1 , R2 and R3 , as well as the Hermitian form, may be
conjugated within SU(2, 1) and scaled so that their matrix entries lie in the ring Z[τ, τ , e±2iπ/p ].

The following result gives a concrete criterion to determine whether or not a given integral
group is contained in an arithmetic lattice. This formulation is essentially due to Mostow
(Lemma 4.1 of [Mos1]), following the Vinberg criterion from [V2] (see p. 226 of [V1]).

Proposition 2.7.4 (Vinberg, Mostow) Let E be a purely imaginary quadratic extension of

a totally real field F , and H a Hermitian form of signature (2,1) defined over E.

1. SU(H; OE ) is a lattice in SU (H) if and only if for all ϕ ∈ Gal(F ) not inducing the identity
on F , the form ϕ H is definite. Moreover, in that case, SU(H; OE ) is an arithmetic lattice.

2. Suppose Γ ⊂ SU(H; OE ) is a lattice. Then Γ is arithmetic if and only if for all ϕ ∈ Gal(F )
not inducing the identity on F , the form ϕ H is definite.

Note that when the group Γ as in the Proposition is non-arithmetic, it necessarily has
infinite index in SU (H, OK ) (which is non-discrete in SU (H)). Using this criterion we get the
following (Theorem 3.1 of [Pau]):

Theorem 2.7.2 For any p > 3 and τ ∈ {σ1 , σ1 , ..., σ, σ9 }, Γ(2π/p, τ ) is contained in an arith-
metic lattice in SU(2, 1) if and only if p = 3 and τ = σ4 .

2.7.5 Commensurability classes

Proposition 2.7.5 (Deligne-Mostow) Q[TrAdΓ] is a commensurability invariant.

This gives a concrete tool to distinguish commensurability classes; for unitary groups (in fact,
of any signature), traces under the adjoint representation are easily computed by the following
(see ”Lemma 4.2” of [Mos1] and [Pau]):

Lemma 2.7.1 For γ ∈ SU(n, 1), Tr(Adγ) = |Tr(γ)|2 .

√
Proposition 2.7.6 For Γ = Γ(2π/p, σ4 ), Q[TrAdΓ] = Q[cos 2π
p
, 7 sin 2π
p
].

Corollary 2.7.3 (1) The 6 groups Γ = Γ(2π/p, σ¯4 ) with p = 3, 4, 5, 6, 8, 12 lie in different
commensurability classes.

(2) The 6 groups Γ = Γ(2π/p, σ¯4 ) with p = 3, 4, 5, 6, 8, 12 are not commensurable to any
Mostow or Picard lattice.

55
2.7.6 Discreteness and fundamental domains
The following is the main result of [DPP2]:

Theorem 2.7.4 Let p > 3, R1 ∈ SU(2, 1) be a complex reflection through angle 2π/p √ and
J ∈ SU(2, 1) be a regular elliptic map of order 3 Suppose that τ = σ4 = Tr(R1 J) = −(1+i 7)/2.
Define c = 2p/(p − 4) and d = 2p/(p − 6).
The group hR1 , Ji is a lattice whenever c and d are both integers, possibly infinity, that is
when p = 3, 4, 5, 6, 8, 12.
Moreover, writing R2 = JR1 J −1 and R3 = JR2 J −1 = J −1 R1 J, this group has presentation

* R1p = J 3 = (R1 J)7 = id, +

R2 = JR1 J −1 , R3 = J −1 R1 J,

R1 , R2 , R3 , J  .

(R1 R2 )2 = (R2 R1 )2 ,
(R1 R2 )2c = (R1 R2 R3 R2−1 )3d = id

Note that the first of these groups is the arithmetic lattice from Theorem 2.7.2; the other
5 groups are non-arithmetic lattices, not commensurable to each other or to any Mostow or
Picard lattice by Corollary 2.7.3.
Strategy: Construct a polyhedron in H2C and use the Poincaré Polyhedron Theorem (The-
orem 2.7.5 below) to prove that it is a fundamental domain for the action of Γ.

Key ingredients:

• a polyhedron D in H2C with side-pairings;

• for each orbit of 2-faces, local tessellation conditions around these 2-faces must be satisfied.

We now give a slightly simplified exposition of the Poincaré Polyhedron Theorem (Theo-
rem 2.7.5 below) following [DFP] in the compact case. For a more general class of polyhedra,
including noncompact ones, see section 3.2 of [DPP2]; see also [EP] for background on this
theorem and an exposition in the constant curvature case.
Definition: A polyhedron is a cellular space homeomorphic to a compact polytope. In
particular, each codimension two cell is contained in exactly two codimension one cells. Its
realization as a cell complex in a manifold X is also referred to as a polyhedron. We will say a
polyhedron is smooth if its faces are smooth.

Definition: A Poincaré polyhedron is a smooth polyhedron D in X with codimension one

faces Ti such that

1. The codimension one faces are paired by a set ∆ of isometries of X which respect the cell
structure (the side-pairing transformations). We assume that if γ ∈ ∆ then γ −1 ∈ ∆.

2. For every γij ∈ ∆ such that Ti = γij Tj then γij D ∩ D = Ti .

56
 Remark: If Ti = Tj , that is if a side-pairing maps one side to itself then we impose,
moreover, that γij be of order two and call it a reflection. We refer to the relation γij2 = 1 as a
reflection relation.

Cycles: Let T1 be an (n − 1)-face and F1 be an (n − 2)-face contained in T1 . Let T10 be

the other (n − 1)-face containing F1 . Let T2 be the (n − 1)-face paired to T10 by g1 ∈ ∆ and
F2 = g1 (F1 ). Again, there exists only one (n − 1)-face containing F2 which we call T20 . We
define recursively gi and Fi , so that gi−1 ◦ · · · ◦ g1 (F1 ) = Fi .

Definition: Cyclic is the condition that for each pair (F1 , T1 )(an (n − 2)-face contained in
an (n − 1)-face), there exists r > 1 such that, in the construction above, gr ◦ · · · g1 (T1 ) = T1
and gr ◦ · · · g1 restricted to F1 is the identity. Moreover, calling g = gr ◦ · · · ◦ g1 , there exists a
positive integer m such that g1−1 (P ) ∪ (g2 ◦ g1 )−1 (P ) ∪ · · · ∪ g −1 (P ) ∪ (g1 ◦ g)−1 (P ) ∪ (g2 ◦ g1 ◦
g)−1 (P ) ∪ · · · ∪ (g m )−1 (P ) is a cover of a closed neighborhood of the interior of F1 by polyhedra
with disjoint interiors.
The relation g m = (gr ◦ · · · g1 )m = Id is called a cycle relation.

Theorem 2.7.5 (Poincaré Polyhedron Theorem) Let D be a compact Poincaré polyhe-

dron in HnC with side-pairing transformations ∆ satisfying condition Cyclic. Let Γ be the group
generated by ∆. Then Γ is a discrete subgroup of Isom(HnC ), D is a fundamental domain for Γ
and Γ has presentation:
Γ = h∆ | cycle relations, reflection relations i

Description of the domains D and E: We construct 2 related polyhedra in H2C . D will

be a fundamental domain for the lattice Γ, and E will be a fundamental domain for the action
of Γ modulo hP i, where P = R1 J has order 7.

E is constructed as follows: start with 4 bisectors R± and S ± , with R1 (R+ ) = R− and

S1 (S + ) = S − . (S1 is a special element in Γ - namely S1 = P 2 R1 P −2 R1 P −2 - and is related to
an obvious complex reflection in Γ by P 2 S1 = R2 R3−1 R2−1 ).

E is then defined as the intersection of the 28 half-spaces bounded by P k (R± ), P k (S ± )

(k = 0, ..., 6) and containing OP , the isolated fixed point of P .
Proposition 2.7.7 (E, ∂E) is homeomorphic to (B 4 , ∂B 4 )(with some vertices removed when
Γ is NC).
Typical results: Given an element g ∈ h side-pairings i, one wishes to prove that g(D) ∩ D
is exactly a certain k-face of D (where 0 6 k 6 3). One way to do this is to prove that g(D)
and D are on opposite sides of well-chosen bisectors. We now illustrate the arguments that
come into play.
Bisector intersections: Pairwise intersections of bisectors can be nasty, e.g. disconnected,
singular... (Goldman). However, if the bisectors are coequidistant (i.e. their complex spines
intersect, away from their real spines), then their intersection is nice:

57
 Figure 2.4: A view of the domain E for p = 12

Theorem 2.7.6 (Giraud, 1934) If B1 and B2 are 2 coequidistant bisectors, then B1 ∩ B2

is a (non-totally geodesic) smooth disk. Moreover, there exists a unique bisector B3 6= B1 , B2
containing it.

Proposition 2.7.8 All 2-faces of E are contained in Giraud disks or complex lines.

Bad projections of Giraud disks: In order to show that g(D) and D are on opposite sides
of a certain bisector B, we argue on the successive k-skeleta of D (for k = 0, 1, 2, 3). The bisector
is given by a certain equation which can be written as a real quadratic equation in the real and
imaginary parts of the ball coordinates of points of H2C . Checking that the vertices of D and
g(D) lie on the appropriate sides of B amounts to checking a finite number of such inequalities.
For the 1-skeleton, it turns out that all 1-faces of our polyhedra are geodesic segments, which
makes matters simpler as the intersection of a geodesic and a bisector is governed by a single
quadratic polynomial (of 1 variable); in other words we only need to compute the corresponding
discriminant for each 1-face of D and g(D).
The most delicate arguments are along the 2-skeleton; one cannot argue that the 2-skeleton
is automatically on the same side of the bisector as the 1-skeleton. While this is true for 2-faces
contained in complex lines, it may fail for 2-faces contained in Giraud disks, as Figures 2.5 and
2.6 illustrate. Recall that the bisector B is the preimage of its real spine under projection to
its complex spine, therefore the fact that a subset of H2C lies on one side or the other of B can
be seen by projecting to the complex spine. Figure 2.5 shows a Giraud disk whose projection
to a complex line has 2 self-intersections; Figure 2.6 shows a Giraud disk whose projection to
a complex line ”spills over” the projection of its boundary. (Both figures courtesy of Martin

58
 Figure 2.5: Projection of a Giraud disk with pinching

Deraux). In particular, if B is the bisector whose real spine is the horizontal axis in Figure 2.6,
then there are points of the interior of the face which lie on the opposite side of the bisector than
the edges of the face (if the face is large enough so that its edges are close to the boundary of the
Giraud disk). We must argue that this does not happen in the cases that we consider; in [DPP2]
we argue by computing critical points of the distance functions for which the corresponding
bisectors are level sets, and showing that these critical points do not lie in the interior of the
corresponding Giraud face.

59
Figure 2.6: Projection of a Giraud disk with spilling

60
Bibliography

[Ag] I. Agol; Finiteness of arithmetic Kleinian reflection groups. International Congress of

Mathematicians Vol. II, pp. 951–960, Eur. Math. Soc. , Zürich, 2006.

[Al] D. Allcock; Infinitely many hyperbolic Coxeter groups through dimension 19. Geometry
and Topology 122 (2006), 159–169.

[An] E.M. Andreev; On convex polyhedra in Lobachevskij spaces. Math. USSR Sb. 10 (1970),
413–440.

[ABSW] I. Agol, M. Belolipetsky, P. Storm, K. Whyte; Finiteness of arithmetic hyperbolic

reflection groups. Groups Geom. Dyn. 2 (2008), 481–498.

[BGS] W. Ballmann, M. Gromov, V. Schroeder; Manifolds of Nonpositive Curvature.

Birkhaüser Progress in Math. vol. 61, 1985.

[Be] A. Beardon; The geometry of discrete groups. Graduate Texts in Mathematics Vol. 91.
Springer-Verlag, New York (1983).

[BP] R. Benedetti, C. Petronio; Lectures on Hyperbolic Geometry, Universitext, Springer-

Verlag (1992).

[Borch] R. Borcherds; Automorphism groups of Lorentzian lattices. J. Algebra 111 (1987),

133–153.

[Borel] A. Borel; Commensurability classes and volumes of hyperbolic 3-manifolds. Ann. Scuola
Norm. Sup. Pisa Cl. Sci. (4) 8 (1981), no. 1, 1–33.

[Bu] V.O. Bugaenko; Arithmetic crystallographic groups generated by reflections, and reflective
hyperbolic lattices. Adv. Sov. 8 (1992), 33–56.

[ChGr] S. Chen, L. Greenberg; Hyperbolic spaces, in Contributions to Analysis. Academic

Press, New York (1974), 49–87.

[Co] H.S.M. Coxeter. Discrete groups generated by reflections. Ann. Math. 35 (1934), 588–621.

[dlH] P. de la Harpe; On Hilbert’s metric for simplices. Geometric group theory, Vol. 1 (Sussex,
1991), London Math. Soc. Lecture Note Ser. 181, Cambridge Univ. Press, Cambridge (1993),
97–119.

61
[DM] P. Deligne & G.D. Mostow; Monodromy of hypergeometric functions and non-lattice
integral monodromy. Publ. Math. I.H.E.S. 63 (1986), 5–89.
[De] M. Deraux; Dirichlet domains for the Mostow lattices. Experiment. Math. 14 (2005), 467–
490
[DFP] M. Deraux, E. Falbel, J. Paupert; New constructions of fundamental polyhedra in com-
plex hyperbolic space. Acta Math. 194 (2005), 155–201.
[DPP1] M. Deraux, J.R. Parker, J. Paupert; Census of the complex hyperbolic sporadic groups.
Experiment. Math. 20 (2011), 467–486.
[DPP2] M. Deraux, J.R. Parker, J. Paupert; New non-arithmetic complex hyperbolic lattices.
Preprint 2013.
[EP] D.B.A. Epstein, C. Petronio; An exposition of Poincaré’s polyhedron theorem. Enseign.
Math. (2) 40 (1994), no. 1-2, 113–170.
[FalPar] E. Falbel, J. Parker; The geometry of the Eisenstein-Picard modular group. Duke
Math. J. 131 (2006) 249–289.
[FZ] E. Falbel, V. Zocca; A Poincaré’s polyhedron theorem for complex hyperbolic geometry.
J. Reine Angew. Math. 516 (1999), 138–158.
[GHL] S. Gallot, D. Hulin, J. Lafontaine;
[G] W.M. Goldman; Complex Hyperbolic Geometry. Oxford Mathematical Monographs. Ox-
ford University Press (1999).
[GPS] M. Gromov, I. Piatetski-Shapiro; Non-arithmetic groups in Lobachevsky spaces. Publ.
Math. IHES 66 (1987), 93–103.
[H] J.E. Humphreys; Reflection Groups and Coxeter Groups. Cambridge Studies in advanced
mathematics vol. 29. Cambridge University Press (1990).
[KM] A. Kappes, M. Moeller; Lyapunov spectrum of ball quotients with applications to com-
mensurability questions. Preprint 2012. arXiv:1207.5433.
[L] M. Lackenby; Hyperbolic Manifolds. Lecture notes (Oxford 2000). Available at:
http://people.maths.ox.ac.uk/lackenby/
[MR] C. Maclachlan & A.W. Reid; The arithmetic of hyperbolic 3-manifolds. Graduate Texts
in Mathematics 219. Springer-Verlag, New York (2003).
[Mak] V.S. Makarov; On a class of discrete groups in Lobachevskij space having infinite fun-
damental domain of finite measure. Sov. Math. Dokl. 7 (1966), 328–331.
[McR] D.B. McReynolds; Arithmetic lattices in SU(n, 1). Monograph, preliminary version.
[Mos1] G.D. Mostow; On a remarkable class of polyhedra in complex hyperbolic space. Pacific
J. Maths. 86 (1980), 171–276.

62
[Mos2] G.D. Mostow; Generalized Picard lattices arising from half-integral conditions.
Publ. Math. I.H.E.S. 63 (1986), 91–106.

[MSW] D. Mumford, C. Series, D. Wright (with cartoons by Larry Gonick); Indra’s pearls :
the vision of Felix Klein, Cambridge University Press (2002).

[N] V.V. Nikulin; Finiteness of the number of arithmetic groups generated by reflections in
Lobachevsky spaces. Izv. Math. 71 (2007), no. 1, 53–56.

[Par1] J.R. Parker; Unfaithful complex hyperbolic triangle groups I: Involutions. Pacific J.
Maths. 238 (2008), 145–169.

[Par2] J.R. Parker; Complex hyperbolic lattices, in Discrete Groups and Geometric Structures.
Contemp. Maths. 501 AMS, (2009) 1–42.

[ParPau] J.R. Parker & J. Paupert; Unfaithful complex hyperbolic triangle groups II: Higher
order reflections. Pacific J. Maths. 239 (2009), 357–389.

[Pau] J. Paupert; Unfaithful complex hyperbolic triangle groups III: arithmeticity and com-
mensurability. Pacific J. Maths. 245 (2010), 359–372.

[Pic1] E. Picard; Sur une classe de groupes discontinus de substitutions linéaires et sur les
fonctions de deux variables indṕendantes restant invariables par ces substitutions. (French)
Acta Math. 1 (1882), no. 1, 297–320.

[Pic2] E. Picard; Sur les fonctions hyperfuchsiennes provenant des séries hypergéométriques de
deux variables. Ann. Sci. École Norm. Sup. (3) , 2 (1885), 357–384.

[Ra] J.G. Ratcliffe; Foundations of Hyperbolic Manifolds, second edition. Graduate Texts in
Mathematics Vol.149. Springer (2006).

[Ru] O.P. Ruzmanov; Examples of non-arithmetic crystallographic Coxeter groups in the n-

dimensional Lobachevskij space for 6 6 n 6 10. Vopr. Teor. Grupp. Gomologicheskoy Algebry
9 (1989), 138–142 (Russian).

[Sc] R.E. Schwartz; Complex hyperbolic triangle groups. Proceedings of the International
Congress of Mathematicians Vol II (Beijing, 2002), 339–349. Higher Ed. Press, Beijing, 2002.

[Th] W.P. Thurston; The Geometry and Topology of Three-Manifolds. Lecture notes (Prince-
ton 1980). Available at: http://library.msri.org/books/gt3m/

[Ti] J. Tits; Classification of algebraic semisimple groups. In: 1966 Algebraic Groups and
Discontinuous Subgroups (Proc. Sympos. Pure Math., Boulder, Colo., 1965), Amer. Math.
Soc. (1966), 33–62.

[To] D. Toledo; Representations of surfaces groups in complex hyperbolic space. J. Differential

Geom. 29 (1989), 125–133.

[V1] E.B. Vinberg (ed.); Geometry II, in: Encyclopedia of Mathematical Sciences, vol. 29,
Springer–Verlag (1991).

63
[V2] E. B. Vinberg; Hyperbolic reflection groups. Russian Math. Surveys 40:1 (1985), 29–66.

[WM] D. Witte Morris; Introduction to Arithmetic Groups. Available at:

http://people.uleth.ca/∼dave.morris/books/IntroArithGroups.html

64