An overview of

Patterson-Sullivan theory
J.-F. Quint

1 Introduction
1.1 Geometry, groups and measure
Let M be a complete Riemannian manifold with negative sectional curvature.
Then the universal cover X of M is diffeomorphic to a Euclidean space and
may be geometrically compactified by adding to it a topological sphere ∂X.
The fundamental group Γ of M acts by isometries on X and this action
extends to the boundary ∂X.
In [13], S.-J. Patterson discovered, in case X is the real hyperbolic plane,
how to associate to a point x of X a probability measure νx on the boundary
∂X that is Γ-quasi-invariant. In some sense, this measure gives the pro-
portion of elements of the orbit Γx that goes to a given zone in ∂X. The
construction of these measures was extended to all hyperbolic spaces by D.
Sullivan in [15]. What’s more, Sullivan discovered deep connections between
the measures νx and harmonic analysis and ergodic theory of the geodesic
flow of M . Today, we know how to construct Patterson-Sullivan measures
when X is any simply connected complete Riemannian manifold with nega-
tive curvature.
If X is a symmetric space, that is if X possesses a very large group of
isometries (we shall have a precise definition later), as real or complex hy-
perbolic space, and if Γ is cocompact, that is if M is compact, the Patterson-
Sullivan measures νx are invariant measures for some subgroups of the group
of isometries that act transitively on the boundary and most of the results
of the theory are consequences of theorems about harmonic analysis in the
group of isometries. So the power of Patterson-Sullivan theory is to allow to
draw an analogy between the case where X is symmetric and Γ cocompact
and the general case.

1
 In these notes, we shall focus on this analogic point of view. Therefore in
section 2 we will briefly recall the definition of classical symmetric spaces and
their basic geometric properties. In section 3 we will show how to construct
measures on the boundary of symmetric spaces with good geometric proper-
ties, only by group-theoretic methods. Finally, in section 4, which is the core
of these notes, we will explain how to construct measures in general mani-
folds that have close properties to the ones appearing in the homogeneous
situation.
But first of all, we begin by giving an example of boundary measures
coming from classical analysis. It will later appear to be related to our
geometric problems.

1.2 Harmonic measures in the disk

Let us recall some well-known facts on the integral representation of harmonic
functions.
We equip R2 with the canonical scalar product, that is, for x = (x1 , x2 )
in R2 , we put kxk2 = x21 + x22 . We denote by D the open unit disk {x ∈
R2 | kxk < 1} and by S1 the unit circle ∂D = {x ∈ R2 | kxk = 1}. We let σ be
the uniform probability measure on S1 , that is its measure as a Riemannian
submanifold of R2 , normalized in such a way that σ(S1 ) = 1. For x in D and
ξ in S1 , we define P (x, ξ) by

1 − kxk2
P (x, ξ) = .
kξ − xk2
The function P is known as the Poisson kernel. For x in D, we set νx =
P (x, .)σ: it is a probability measure on S1 . We call it the harmonic measure
associated to x. Note that ν0 = σ.
A C 2 function ϕ : D → C is said to be harmonic if ∆ϕ = 0 where
∂2 ∂2
∆ = ∂x 2 + ∂x2 denotes the Laplace operator. A direct calculation shows
1 1
1
that, for ξ in S , the function P (., ξ) is harmonic. Therefore, for each f in
L∞ (S1 ) its Poisson transform Pf , defined by
Z Z
∀x ∈ D Pf (x) = f (ξ)dνx (ξ) = f (ξ)P (x, ξ)dσ(ξ),
S1 S1

is a bounded harmonic function on D.

We have the following

2
 Figure 1: Geodesics in the hyperbolic disk

Theorem 1.1. The map f 7→ Pf gives an isomorphism between L∞ (S1 ) and

the space of bounded harmonic functions on D. It preserves the L∞ -norm.

Let us give an other interpretation of the harmonic measures and the Pois-
son kernel in terms of the hyperbolic geometry in the disk. The hyperbolic
Riemannian metric of the disk is defined by
4
∀x ∈ D gx = ge
2
2 x
1 − kxk

where g e is the Euclidean metric; that is, the hyperbolic length of a C 1 curve
γ : [0, 1] → D is
Z 1
2 0
2 kγ (t)k dt.
0 1 − kγ(t)k

This metric is complete and the complete geodesics are the diameters of D
and the arcs of circles orthogonal to S1 . Any geodesic thus possesses two
limit points in S1 .
Let ξ be in S1 . A horocycle with center ξ is the intersection of D with a
Euclidean circle (of radius < 1) passing through ξ. The horocycles of D are
the curves which are orthogonal to the geodesics having ξ as a limit point.
For x and y in D and ξ in S1 , define the Busemann function bξ (x, y) as the
hyperbolic distance between x and the point where the geodesic line passing
through x and having ξ as a limit point hits the horocycle centered at ξ and
passing through y, this distance being counted positively if this hitting point
lies between x and ξ and negatively else (see figure 2). The geodesics and
horospheres appearing in this definition being orthogonal, if x0 (resp. y 0 ) lies

3
 ξ
x
bξ (x, y)

Figure 2: Horocycles and Busemann function

in the same horocycle centered at ξ as x (resp. y), one has bξ (x, y) = bξ (x0 , y 0 ).
In particular, the Busemann function satisfies the cocyle identity:

∀x, y, z ∈ D bξ (x, z) = bξ (x, y) + bξ (y, z).

From the definition of the metric, the Busemann function may be com-
puted explicitly; we then get:
 
1 P (y, ξ)
∀ξ ∈ S ∀x, y ∈ D bξ (x, y) = log .
P (x, ξ)
In other words, the harmonic measures associated to points of D satisfy:
dνy
∀ξ ∈ S1 ∀x, y ∈ D (ξ) = e−bξ (y,x) .
dνx
Note in particular that this relation implies the cocycle identity.
In the sequel, we will be interested in finding measures satisfying analo-
gous properties in the context of manifolds of negative curvature. We begin
by studying the ones possessing a large group of isometries.

2 Rank one symmetric spaces of noncompact

type
2.1 Symmetric spaces
We recall material from [6] and [10]. Given a connected Riemannian manifold
M , consider a point x of M and a symmetric neighbourhood U of M in

4
the tangent space Tx M possessing the property that the exponential map
Expx : U → M is well-defined and is a diffeomorphism onto its image V .
The symmetry u 7→ −u of U then induces a map sx on V . We shall call sx
the local geodesic symmetry centered at x.
We say that M is a Riemannian locally symmetric space if, for any x
in M , the local symmetry sx is a local isometry of M . We say that M is
globally symmetric if, for any x, this isometry may be extended (necessarily
uniquely) to M . A complete simply connected locally symmetric space is
globally symmetric. Globally symmetric spaces are complete spaces which
possess a very large group of isometries (in particular their group of isometries
is transitive).
Let M be a connected Riemannian manifold, with isometry group G. If
x is a point of G, consider the map G → T∗ x M ⊗ TM, g 7→ (gx, dg(x)): it
is injective and has closed image. From this we deduce that the group of
isometries of M is separable, locally compact and second countable for the
compact-open topology. In case M is globally symmetric, G can be proved to
carry a structure of Lie group compatible with this topology. The structure
of Riemannian globally symmetric spaces is therefore intrinsically linked with
the theory of Lie groups. In particular, globally symmetric spaces have been
classified by E. Cartan.
It turns out that every globally symmetric space M is the Cartesian
Riemannian product of three globally symmetric spaces M0 , M+ and M− ,
where M0 is isometric to some Rk , with its canonical Euclidean structure,
and M− (resp. M+ ) has nonpositive (resp. nonnegative) curvature, but may
not be written as a product of R with some other Riemannian manifold.
Spaces of the form M− (resp. M+ ) are said to be of noncompact (resp.
compact) type. In the sequel, we shall be concerned by symmetric spaces of
noncompact type.
A (connected) Lie group is said to be semisimple if it has no non-trivial
abelian connected normal closed subgroup. In other words, a Lie group is
semisimple if its Lie algebra is semisimple. If G is a semisimple Lie group it
possesses maximal compact subgroups: these subgroups are all conjugate to
each other and equal to their normalizer. Therefore, if K is such a maximal
compact subgroup, the manifold G/K may be seen as the set of maximal
compact subgroups of G. Let g be the Lie algebra of G and k the one of K.
As K is compact, its adjoint action preserves a scalar product on the vector
space g/k which is naturally identified to the tangent space at K of G/K.
This scalar product then induces a G-invariant metric on G/K. This metric

5
can be proved to make G/K a globally symmetric space of noncompact type
and we have the following theorem, due to E. Cartan:
Theorem 2.1. The Riemannian globally symmetric spaces of noncompact
type are the spaces of the form G/K equipped with a G-invariant metric,
where G is a connected semisimple Lie group and K a maximal compact
subgroup of G.
As every complete Riemannian manifold with nonpositive curvature, all
these spaces are diffeomorphic to some Rk .

Example 2.1. The Lie group SLn (R) can be checked to be semisimple (this
is in fact the generic example of a semisimple group). As every compact group
of linear automorphism of Rn preserves a scalar product, the group SO(n) of
orthogonal matrices with determinant 1 is a maximal compact subgroup of
SLn (R) and every maximal compact subgroup of SLn (R) is conjugate to it.
The tangent space to SLn (R)/SO(n) at SO(n) may be SO(n)-equivariantly
identified with the vector space of symmetric matrices. On this space, the
bilinear form (A, B) 7→ Tr(AB) is a SO(n)-invariant scalar product. We
therefore get a SLn (R)-invariant Riemannian metric on SLn (R)/SO(n). The
automorphism g 7→ (g −1 )t (where t denotes the transpose matrix) of SLn (R)
fixes SO(n) and induces on SLn (R)/SO(n) an isometry which extends the
local geodesic symmetry centered at SO(n).
Consider the particular case n = 2. The group SL2 (R) acts on the Rie-
mann sphere P1C by projective automorphisms and preserves the circle P1R . As
this group is connected,
 it preserves the upper half  H = {z ∈ C| Im z >
 plane
a b a b az+b
0}: for ∈ SL2 (R) and z ∈ H, we get · z = cz+d . This ac-
c d c d
tion is transitive, as upper triangular matrices already act transitively, and
the stabilizer of i is SO(2). Therefore it identifies SL2 (R)-equivariantly H
and SL2 (R)/SO(2). The SL2 (R)-invariant metric g we defined above can be
written gx+iy = y22 gx+iy
e
where g e is the Euclidean metric: in other word it
is (up to a scalar multiple) the upper half plane model for hyperbolic plane
(see paragraph 2.2).

A totally geodesic submanifold of a globally symmetric space M is neces-

sarily itself a globally symmetric space. If M is of noncompact type, totally
geodesic submanifolds have nonpositive curvature and, thus, don’t have com-
pact type factors. In that case, we say that M has rank k if it contains a flat

6
totally geodesic submanifold of dimension k and if every other flat totally
geodesic submanifold has rank ≤ k. Maximal flat subspaces can be shown
to be conjugate under the group of isometries. As M contains geodesics, its
rank is ≥ 1. A symmetric space has rank one if and only if it has negative
curvature, that is its sectional curvature as a function on the Grassmannian
bundle G 2 M of tangent 2-planes of M , is everywhere negative.

Example 2.2. The rank of SLn (R)/SO(n) is n − 1. More precisely, if

A is the group of diagonal matrices with positive entries in SLn (R), the set
F = ASO(n) is a flat totally geodesic submanifold of SLn (R)/SO(n). The
other maximal flat subspaces are of the form gF for some g in SLn (R). In
particular, for n = 2, the hyperbolic plane has rank one and the geodesics
are the curves of the form t 7→ g · et i for some g in SL2 (R).

2.2 Rank one symmetric spaces

The classification of globally symmetric spaces of noncompact type is the
same as the classification of semisimple Lie groups. As often in Lie group
theory, the classification contains a finite number of infinite lists (as the one
of special linear groups SLn (R), n ≥ 2), the so-called classical groups, and a
finite set of “exceptional” examples.
For rank one symmetric spaces, there are three lists of classical spaces:
real, complex and quaternionic hyperbolic spaces. There is only one excep-
tional one, the Cayley hyperbolic plane, which we will not describe here.

2.2.1 Real hyperbolic spaces

Fix an integer n ≥ 1 and equip Rn+1 with the quadratic form q(x0 , . . . , xn ) =
x20 − x21 − . . . − x2n of signature (1, n). Denote by HnR the set {x ∈ Rn+1 |q(x) =
1 and x0 > 0}: this is one of the two connected components of the set
{q = 1}. For x in HnR , the tangent space at x of HnR identifies with its q-
orthogonal hyperplane. On that space, by Sylvester’s theorem, the restriction
of q is negative definite. Denote by gx its opposite: the field of bilinear forms
g is a Riemannian metric on HnR . We call this Riemannian manifold real
hyperbolic space of dimension n.
The other classical models for hyperbolic space may be recovered from
this one, which, as we shall soon see, is the most practical one to describe the
group of isometries. First of all, we can identify HnR with the set of vector lines

7
 x0

x2

x1

Figure 3: The hyperbolic hypersurface

containig a q-positive vector, as such a line hits the hyperbolic hypersurface at

only one point: we then get the Klein model of the hyperbolic space, which is
seen as an open subset in PnR . We can then project by stereographic projection
from the point (−1, 0, . . . , 0) the set HnR onto the ball Bn = {x ∈ Rn | kxk < 1}
(where the norm is the canonical Euclidean norm of Rn ) which we view as
a subset of {0} × Rn . We get the ball model for real hyperbolic space, that
4 e e
is the metric x 7→ 2 g , where, as usual, g denotes the Euclidean
(1−kxk2 ) x
metric. Finally, we can apply to the ball √ model the Euclidean inversion of
n
R with center (−1, 0, . . . , 0) and radius 2: we then get the upper half space
model, that is the set {(x1 , . . . , xn ) ∈ Rn |x1 > 0} equipped with the metric
x 7→ x12 gxe .
1
Return now to the original model. We will exhibit strong properties of
transitivity of some groups of isometries HnR . We shall need the following

Lemma 2.2. Let V be a vector subspace of Rn+1 containing an element x of

HnR . Then V ∩ HnR is a totally geodesic submanifold of HnR . It is isometric to
real hyperbolic space of dimension dim V − 1. Every complete totally geodesic
submanifold of HnR is of this form. In particular, complete geodesics of HnR
are the nonempty intersections of HnR with planes of Rn+1 .

Proof. As x is a q-anisotropic vector we have Rn+1 = Rx⊕x⊥ (where ⊥ refers

to orthogonality with respect to q). As V contains x, we get V = Rx ⊕ (x⊥ ∩
V ). Since the restriction of q to x⊥ is negative definite, the restriction of q

8
 (−1, . . . , 0)

Figure 4: Stereographic projection

to V is nondegenerate and has signature (1, dim V − 1). Therefore V ∩ HnR

is isometric to hyperbolic space of dimension dim V − 1. Let us show that
it is totally geodesic. As the restriction of q to V is nondegenerate, we have
Rn+1 = V ⊕ V ⊥ . Let s denote the q-orthogonal reflection with respect to V ,
that is the linear automorphism that is y 7→ y on V and y 7→ −y on V ⊥ : s
is a q-isometry and, as it fixes x, stabilizes HnR , where it therefore induces an
isometry. The fixed point set of this isometry is exactly V ∩ HnR . Thus this
set is a totally geodesic submanifold, by the local uniqueness of geodesics.
Finally, let M ⊂ HnR be a complete totally geodesic submanifold and let x be
a point of M . If W is the tangent space to M at x and V = Rx ⊕ W , M and
V ∩ HnR are complete totally geodesic submanifolds having the same tangent
space at x and are therefore equal.
Denote by O(1, n) the orthogonal group of q, by SO(1, n) the special
orthogonal group and by SO◦ (1, n) the connected component of the identity
in SO(1, n): it has index 2 in SO(1, n) as each element of O(1, n) either
stabilizes HnR or exchanges HnR and −HnR . Let (e0 , e1 , . . . , en ) be the canonical
basis of Rn+1 . We shall write K for the subgroup of SO◦ (1, n) consisting
of isometries of the form (x0 , x1 , . . . , xn ) 7→ (x0 , g(x1 , . . . , xn )) where g lies
in SO(n): it is the stabilizer of e0 in SO◦ (1, n) and e0 is the unique fixed
point of K in HnR . Let us denote by (f0 , f1 , . . . , fn ) the basis of Rn+1 such
that f0 = e0√+e2 n , f1 = e1 , . . . , fn−1 = en−1 , fn = e0√−e2 n and y0 , . . . , yn the
coordinates in this base. We have q = 2y0 yn − y12 − . . . − yn−1 2
. For t ∈ R,

9
the linear operator at whose matrix with respect to (f0 , f1 , . . . , fn ) is
 t 
e 0 0
 0 In−1 0 
0 0 e−t

is a q-isometry; its matrix with respect to (e0 , e1 , . . . , en ) is

 
cosh t 0 − sinh t
 0 In−1 0 .
sinh t 0 cosh t

In particular the curve t 7→ at e0 is a unit speed geodesic. We write A for the

subgroup {at |t ∈ R} of SO◦ (1, n).
Let ∂HnR denote the boundary of HnR as a subset of projective space, that
is the projective image of the isotropic cone of q. It is diffeomorphic to the
sphere Sn−1 .

Lemma 2.3. The group SO◦ (1, n) acts transitively on HnR and the group K
acts transitively on ∂HnR . The sectional curvature of HnR has constant value
−1.

Proof. Let x = (x0 , . . . , xn ) be in HnR . By applying an element of K, we can

suppose (x2 , . . . , xn−1 ) = 0. Then x lies in Ae0 . The proof is analogous for
isotropic vectors.
As SO(n) acts transitively on 2-planes, the group K acts transitively
on 2-planes of Te0 HnR and SO◦ (1, n) acts transitively on the Grassmannian
bundle G 2 HnR of 2-planes tangent to HnR . Therefore, the sectional curvature
of HnR is constant. We postpone the calculus of its value to appendix A.
Let N be the subgroup of elements of SO◦ (1, n) whose matrix with respect
to the basis (f0 , f1 , . . . , fn ) is of the form
2
1 u − kuk

2
nu = 0 In−1 −ut 
0 0 1

where u is some line vector in Rn−1 and ut designs the transpose column
vector. The map u 7→ nu is an isomorphism from Rn−1 onto N . The group
N is normalized by A and one has at nu a−t = net u . Finally let M be the

10
subgroup of elements of SO◦ (1, n) whose matrix with respect to the basis
(f0 , f1 , . . . , fn ) has the form
 
1 0 0
mg = 0 g
 0
0 0 1

for some g in SO(n − 1): it normalizes N and one has mg nu mg−1 = ngu . Let
P = M AN .
The links between all these subgroups are explained in the following:
Lemma 2.4. One has SO◦ (1, n) = KP = KAN and P is the stabilizer of
ξ0 = Rf0 in SO◦ (1, n). For every ξ in ∂HnR , the stabilizer Pξ of ξ is conjugate
to P and one has SO◦ (1, n) = KPξ .
Proof. Let us show that SO◦ (1, n) = KAN : it suffices to show that AN
acts transitively on HnR let (y0 , . . . , yn ) be the coordinates in (f0 , f1 , . . . , fn )
of an element y of HnR . Then n(y2 ,...,yn−1 ) y belongs to Ae0 and this implies the
result.
Let g be an element of SO◦ (1, n) stabilizing Rf0 . From the preceding, we
can suppose that g fixes e0 . Therefore, it stabilizes the vector plane spanned
by f0 and fn . As the only isotropic lines of this plane are Rf0 and Rfn , it
stabilizes both these lines. Since it fixes e0 , the eigenvalue on these lines must
be 1. Now g must stabilize the q-orthogonal space of the plane spanned by
f0 and fn , hence g belongs to M .
As we shall see in the next section these objects will play a key role in
the description of harmonic measures on ∂HnR . We will now generalize their
construction to the other classical rank one symmetric spaces.

2.2.2 Complex hyperbolic spaces

We now consider on Cn+1 the hermitian quadratic form q(x0 , x1 , . . . , xn ) =
|x0 |2 − |x1 |2 − . . . − |xn |2 of signature (1, n) corresponding to the hermitian
sesquilinear form hx, yi = x0 y0 − x1 y1 − . . . − xn yn . Consider the open set
U = {q > 0} in Cn+1 . On U , we define a (complex) subbundle E of the
tangent bundle as follows: for each x, we take Ex as being the q-orthogonal
space of x. Then, on Ex , by Sylvester’s theorem, the restriction of q is
1
negative definite. We define gx to be the restriction to Ex of − q(x) q: this
hermitian metric on U is invariant by multiplication by complex scalars. Let

11
HnC be the open subset of PnC which is the image of U and let π : U → HnC be
the natural map. Then HnC is diffeomorphic to the ball B2n . The differential
dπ : E → THnC is surjective on fibers. As the metric g is invariant by
multiplication by scalars, it induces a hermitian metric on HnC : we call this
space complex hyperbolic space of dimension n.
We know the structure of the complex hyperbolic line:

Lemma 2.5. The space H1C is isometric to the hyperbolic plane H2R equipped
with the metric which is equal to 14 times the usual one.

Proof. The map C → P1C , z 7→ [1, z] induces a (holomorphic) diffeomorphism

from the disk D = {z ∈ C| |z| < 1} onto H1C . The pulled back metric may be
1
written z 7→ g e where g e denotes the canonical Euclidean metric, that
2 2 z
( 1−|z| )
is 41 times the hyperbolic metric in the ball model and the lemma follows.
We say that a R-subspace of Cn+1 is Lagrangian if it is totally isotropic for
the skew-symmetric R-bilinear form (x, y) 7→ Imhx, yi. The following result
is an analogue of lemma 2.2:

Lemma 2.6. Let V be a complex vector subspace of Cn+1 containing an

element x such that q(x) > 0. Then P (V ) ∩ HnC is a totally geodesic complex
submanifold of HnC . It is isometric to complex hyperbolic space of dimension
dimC V − 1. Every complete totally geodesic complex submanifold of H nC is of
this form.
Let V be a Lagrangian real vector subspace of Cn+1 containing an element
x such that q(x) > 0. Then the image of V in PnC intersects HnC on a totally
geodesic real submanifold. It is isometric to real hyperbolic space of dimension
dimR V − 1.
Every complete totally geodesic submanifold of HnC is of one of these forms.
In particular, complete geodesics of HnC are the image in HnC of Lagrangian
R-planes of Cn+1 .

Proof. The proof that the image of a complex subspace of Cn+1 in HnC is
totally geodesic and that every totally geodesic complex submanifold is of
this form is analogous to the proof of lemma 2.2.
Suppose now V is a Lagrangian R-subspace of Cn+1 containing a vector
x such that q(x) > 0. As V is Lagrangian, x⊥ ∩ V = {y ∈ V | Rehx, yi = 0}
is a R-hyperplane of V and the restriction of Reh., .i to V is a nondegenerate
R-bilinear form of signature (1, dimR V − 1). Since V is Lagrangian, we have

12
V ∩ iV = {0} and the projection map {y ∈ V |q(y) = 1} → HnC induces an
isometry of its image with real hyperbolic space of dimension dimR V − 1.
Finally let us choose a maximal Lagrangian R-subspace W ⊃ V of Cn+1 .
Then one has W ⊕ iW = Cn+1 and conjugation with respect to this decom-
position (that is x + iy 7→ x − iy) induces an anti-q-isometry of Cn+1 and an
isometry of HnC , with fixed points set the image of W in HnC . Therefore this
image is totally geodesic and isometric to HnR . Hence the image of V , which
is contained in the one of W , is totally geodesic by lemma 2.2.
The classification of totally geodesic submanifolds of complex hyperbolic
space will be achieved in appendix A.
Denote by U(1, n) the unitary group of the form q and by PU(1, n) its
projective image: these are connected groups. As before, we shall denote
by (e0 , e1 , . . . , en ) the canonical basis of Cn+1 and write K for the image
in PU(1, n) of the group of isometries of q of the form (x0 , x1 , . . . , xn ) 7→
(x0 , g(x1 , . . . , xn )) where g lies in U(n): it is the stabilizer of Ce0 in PU(1, n)
and Ce0 is the unique fixed point of K in HnC . Let us still denote by
(f0 , f1 , . . . , fn ) the basis ( e0√+e2 n , e1 , . . . , en−1 , e0√−e2 n ) and y0 , . . . , yn the coor-
dinates in this base. We have q = 2 Re(y0 yn ) − |y1 |2 − . . . − |yn−1 |2 . For
t ∈ R, the linear operator at whose matrix with respect to (f0 , f1 , . . . , fn ) is
 t 
e 0 0
 0 In−1 0 
0 0 e−t
is a q-isometry with matrix with respect to (e0 , e1 , . . . , en ):
 
cosh t 0 − sinh t
 0 In−1 0 ,
sinh t 0 cosh t
and the curve t 7→ at Ce0 is a unit speed geodesic. We write A for the image
of the group {at |t ∈ R} in PU(1, n).
Let ∂HnC denote the boundary of HnC as a subset of projective space. It
is the projective image of the isotropic cone of q and is diffeomorphic to the
sphere S2n−1 .
Lemma 2.7. The group PU(1, n) acts transitively on HnC and the group K
acts transitively on ∂HnC . The sectional curvature of HnC lies everywhere be-
tween −4 and −1 and reaches the value −1 exactly on Lagrangian real 2-
planes and the value −4 exactly on complex lines, viewed as real 2-planes.

13
Proof. The first part is proved the same way as for lemma 2.3. The compu-
tation of the curvature will be achieved in appendix A.
Let now N be the the image in PU(1, n) of the group of elements of
SU(1, n) whose matrix with respect to the basis (f0 , f1 , . . . , fn ) is of the form
2
1 u is − kuk

2
n(u,s) = 0 In−1 −ut 
0 0 1

where u is some line vector in Cn−1 and s is a real number. The group N is
(2n − 1)-dimensional Heisenberg group, that is its Lie algebra is isomorphic
to R2n−2 × R equipped with the Lie bracket defined by [(U, S), (V, R)] =
(0, ω(U, V )) where ω is some skew-symmetric nondegenate bilinear form on
R2n−2 . As before, this group is normalized by A and one has at n(u,s) a−t =
n(et u,e2t s) . Finally let M be the subgroup of elements of PU(1, n) which are
images of q-isometries with matrix with respect to the basis (f0 , f1 , . . . , fn )
of the form  
1 0 0
mg =  0 g 0 
0 0 1
for some g in PU(n − 1): it normalizes N and one has mg n(u,s) mg−1 = n(gu,s) .
Let P = M AN .
We now have an analogue of lemma 2.4:

Lemma 2.8. One has PU(1, n) = KP = KAN and P is the stabilizer of

ξ0 = Cf0 in PU(1, n). For every ξ in ∂HnC , the stabilizer Pξ of ξ is conjugate
to P and one has PU(1, n) = KPξ .

2.2.3 Quaternionic hyperbolic space

We let Q be the set of quaternionic numbers and, as usual, i, j, k be three
fixed elements of Q such that i2 = j 2 = k 2 = −1 and ij = k, jk = i √and
ki = j. We write x 7→ x for the quaternionic conjugation, x 7→ |x| = xx
for the quaternionic module and Q0 for the space of pure quaternions, that
is those satisfying x = −x (it is the R-vector space spanned by i, j, k).
If E is a right quaternionic vector space, a (quaternionic) hermitian form
on E is a R-bilinear map ϕ : E × E → Q such that, for α, β in Q, for
x, y in E, one gets ϕ(xα, yβ) = αϕ(x, y)β and ϕ(y, x) = ϕ(x, y). For such

14
maps, the analogue of Sylvester’s theorem holds: we have classification by
signature for nondegenerate forms. The unitary group of such a form is the
group of Q-linear automorphisms of E which preserve it. The unitary group
of the standard form of signature (p, q) on Qp+q , hx, yi = x1 y1 + . . . + xp yp −
xp+1 yp+1 − . . . − xq yq , which is the polar form of q(x) = |x1 |2 + . . . + |xp |2 −
|xp+1 |2 − . . . − |xq |2 , is denoted by Sp(p, q). It is a real form of the symplectic
group Sp2(p+q) (C).
Let us now fix the standard form hx, yi = x0 y0 − x1 y1 − . . . − xn yn of
signature (1, n) on Qn+1 , which we consider as a right vector space, and set
q(x) = |x0 |2 − |x1 |2 − . . . − |xn |2 . As before, we let HnQ denote the open
set which is the image of {q > 0} in the set PnQ of one-dimensional right
Q-subspaces of Qn+1 ; it is diffeomorphic to the ball B4n . For each x with
1
q(x) > 0 the form − q(x) q is positive definite on the orthogonal of x. This
defines a hermitian metric on HnQ . We call this space quaternionic hyperbolic
space of dimension n.
As the complex hyperbolic line, the quaternonic hyperbolic line has a
special structure: the same proof as the one of lemma 2.5 gives the

Lemma 2.9. The space H1Q is isometric with the hyperbolic hyperspace H4R
equipped with the metric which is equal to 41 times the usual one.

Let pR be the R-projection x 7→ 21 (x − x) of Q onto Q0 . A real subspace V

of Qn+1 is said to be Lagrangian if the restriction to V of the skew-symmetric
R-bilinear map (x, y) 7→ pR (hx, yi) is trivial. In the same way, if K is a
subfield of Q which is isomorphic to C, there exists a R-subspace W of Q,
supplementary to K, which is invariant by both left and right multiplication
by elements of K: this is the set of y in Q such that, for any x in K, one
has xy = yx (for K = R[i], one has W = Rj ⊕ Rk). It is contained in
Q0 . We let pK denote the R-projection onto W with kernel K: it is left
and right K-linear. In particular, if E is a right Q-vector space and ϕ a
hermitian form on E, then pK ◦ ϕ is a skew-symmetric K-bilinear map on E.
A right K-subspace of Qn+1 is said to be Lagrangian if it is totally isotropic
for the skew-symmetric K-bilinear map (x, y) 7→ pK (hx, yi). We still have an
analogous result to lemmas 2.2 and 2.6:

Lemma 2.10. Let V be a quaternionic vector subspace of Qn+1 containing

an element x such that q(x) > 0. Then P (V ) ∩ HnQ is a totally geodesic
quaternionic submanifold of HnC . It is isometric to quaternionic hyperbolic

15
space of dimension dimC V − 1. Every complete totally geodesic quaternionic
submanifold of HnC is of this form.
Let K be a subfield of Q which is isomorphic to C and let V be a La-
grangian K-vector subspace of Qn+1 containing an element x such that q(x) >
0. Then the image of V in PnQ intersects HnQ on a totally geodesic submanifold.
It is isometric to complex hyperbolic space of dimension dim K V − 1.
Every complete totally geodesic submanifold of HnQ either is of the first
form, or is contained in a manifold of the first form and of Q-dimension 1,
or is contained in a submanifold of the second form. In particular, complete
geodesics of HnQ are the image in HnQ of Lagrangian R-planes of Qn+1 .

Denote by PSp(1, n) the projective image of Sp(1, n), that is its quotient
by the group of diagonal real matrices. Let still (e0 , e1 , . . . , en ) be the canoni-
cal basis of Qn+1 and (f0 , f1 , . . . , fn ) the basis ( e0√+e2 n , e1 , . . . , en−1 , e0√−e2 n ) (with
coordinates y0 , . . . , yn in such a way that q = 2 Re(y0 yn )−|y1 |2 −. . .−|yn−1 |2 ).
Write K for the image in PSp(1, n) of the group of isometries of q of the form
(x0 , x1 , . . . , xn ) 7→ (αx0 , g(x1 , . . . , xn )) where α is a unit modulus quaternion
and g lies in Sp(n): it is the stabilizer of e0 Q in PSp(1, n) and e0 Q is the
unique fixed point of K in HnQ . As before, for t ∈ R, the linear operator at
whose matrix with respect to (f0 , f1 , . . . , fn ) is
 t 
e 0 0
 0 In−1 0 
0 0 e−t

is a q-isometry and the curve t 7→ at e0 Q is a unit speed geodesic. We write

A for the image of the group {at |t ∈ R} in PSp(1, n).
Let ∂HnQ denote the boundary of HnQ as a subset of projective space. It
is the projective image of the isotropic cone of q and is diffeomorphic to the
sphere S4n−1 . We have a quaternionic version of lemmas 2.3 and 2.7:

Lemma 2.11. The group PSp(1, n) acts transitively on HnQ and the group
K acts transitively on ∂HnQ . The sectional curvature of HnQ lies everywhere
between −4 and −1 and reaches the value −1 exactly on Lagrangian real 2-
planes and the value −4 exactly on real 2-planes that are K-lines for some
maximal commutative subfield K of Q.

Let now N be the the image in PSp(1, n) of the group of elements of

PSp(1, n) whose matrix with respect to the basis (f0 , f1 , . . . , fn ) is of the

16
form 2
1 u s − kuk

2
n(u,s) = 0 In−1 −ut 
0 0 1
where u is some line vector in Qn−1 and s lies in Q0 . The group N has
Lie algebra isomorphic to Qn−1 × Q0 equiped with the Lie bracket defined
by [(U, S), (V, R)] = (0, pR (hU, V i)), where h., .i denotes the standard Q-
hermitian scalar product on Qn−1 . This group is still normalized by A and
one has at n(u,s) a−t = n(et u,e2t s) . Finally let M be the subgroup of elements
of PSp(1, n) who are images of q-isometries with matrix with respect to the
basis (f0 , f1 , . . . , fn ) of the form
 
α 0 0
m(α,g) =  0 g 0 
0 0 α

for some unit quaternion α and some g in PSp(n − 1): it normalizes N and
one has m(α,g) n(u,s) m(α−1 ,g−1 ) = n(gu,αsα−1 ) . Let P = M AN .
We still have an analogue of lemma 2.4 and 2.8:

Lemma 2.12. One has PSp(1, n) = KP = KAN and P is the stabilizer of

ξ0 = f0 Q in PSp(1, n). For every ξ in ∂HnQ , the stabilizer Pξ of ξ is conjugate
to P and one has PSp(1, n) = KPξ .

3 Homogeneous harmonic measures for rank

one classical symmetric spaces
Here we shall develop the formalism of harmonic measures for the spaces
introduced above. We begin by a fact from general group theory.

3.1 A Haar measure computation

Let us recall some general notions from Haar measure theory. Let G be a
separable, locally compact and second countable topological group. Then up
to homothety G possesses a unique Radon measure which is invariant by left
(resp. right) translations by elements of G: such a measure is called a left
(resp. right) Haar measure for G. Let λ be a left Haar measure. Then, for

17
each g in G, the image of λ by right translation by g −1 is still a left Haar
measure for G. It is therefore of the form ∆G (g)λ for some ∆G (g) in R∗+ ,
which does not depend on the choice of λ. The function ∆G is a continuous
homomorphism from G into the multiplicative group R∗+ ; it is called the
modular function of G. The measure ∆1G λ is a right Haar measure for G
Rand, for
−1
every continuous
R −1
function ϕ with compact support on G, one gets
G
ϕ(g )dg = G
∆ G (g) ϕ(g)dg.
The group G is said to be unimodular if ∆G = 1, that is if it possesses
a bi-invariant Radon measure. Compact groups are unimodular: they don’t
possess any non-trivial homomorphism into R∗+ (as this latter group doesn’t
have any non-trivial compact subgroup). Abelian groups are unimodular as
for them both left and right multiplication coincide and, by an induction ar-
gument, this implies that nilpotent groups are unimodular. Discrete groups
are unimodular as their Haar measure is the counting measure which is in-
variant under any bijection. The groups Pξ from the preceding section are
not unimodular (we shall soon compute their modular function).

Lemma 3.1. The groups SO◦ (1, n), PU(1, n) and PSp(1, n) are unimodular.

Proof. Choose one of these groups, denote it by G and let X be the associated
hyperbolic space. Let us keep the notations of the previous section and let
o be the point of X associated to e0 . Then, as K acts transitively on the
unit sphere of To X, every point x of X belongs to Kar o where r = d(o, x).
Therefore, we have G = K{at |t ≥ 0}K, and, as K is compact, the modular
function ∆G is determined by its restriction to A. But, for t in R, at belongs
to Ka−t K and, therefore, ∆G (at ) = 0. The conclusion follows.
Let H be a closed subgroup of G. Then the homogeneous space G/H
possesses a G-invariant measure if and only if the modular functions of G
and H are equal on H. Such a measure is then unique. If H is compact,
there exists an invariant measure on G/H: it is the projection of some Haar
measure of G onto G/H. In particular, if G is compact, this measure is finite
and can therefore be uniquely normalized to have total mass 1.
Let us focus on the situation provided by lemmas 2.4, 2.8 and 2.12. We
therefore fix a unimodular group G, a compact subgroup K of G and a closed
subgroup P , this last one with modular function ∆. We fix some left Haar
measures dg, dk and dp on G, K and P .

18
Lemma 3.2. Suppose we have G = KP . Then the Haar measures can be
normalized in such a way that, for any continuous function ϕ with compact
support on G one gets:
Z Z
ϕ(g)dg = ∆(p)−1 ϕ(kp)dpdk.
G K×P

Proof. Consider the topological group K × P and let it act on G in such a

way that (k, p) · g = kgp−1 . By the hypothesis, this action is transitive. It
induces an homeomorphism from G onto (K ×P )/H where H is the compact
subgroup {(h, h)|h ∈ K ∩ P }. As the Haar measure of G is right P -invariant
and left K-invariant, it induces a K × P invariant measure on (K × P )/H,
which is the projection of some Haar measure of K × P . By normalizing
suitably, we therefore get, for a continuous function ϕ with compact support
on G:
Z Z Z
−1
ϕ(g)dg = ϕ(kp )dpdk = ∆(p)−1 ϕ(kp)dpdk.
G K×P K×P

Suppose G = KP . Then K acts transitively on G/P and thus preserves

a unique probability measure on G/P . The action of G on this set preserves
the measure class of this measure and we will give an explicit description of
the associated Radon-Nikodym cocycle.
Choose once for all a Borel section s : G/P ≡ K/(K ∩ P ) → K, that
is a Borel map such that, for every g in G, g belongs to s(g)P (such a
map always exists for quotients of separable, locally compact and second
countable topological groups). For every g in G and ξ in G/P , there exists
a unique σ(g, ξ) in P such that gs(ξ) = s(gξ)σ(g, ξ). The Borel function
σ : G × (G/P ) → P clearly satisfies te cocycle identity:

∀g, h ∈ G ∀ξ ∈ G/P σ(gh, ξ) = σ(g, hξ)σ(h, ξ).

As the restriction of ∆ to K ∩ P is trivial, the R∗+ -valued cocycle θ = ∆ ◦ σ

doesn’t depend on the choice of the section s and is continuous.
We now can prove the general formula we will later use in the context of
hyperbolic spaces:
Proposition 3.3. Suppose G = KP and let ν be the unique K-invariant
measure on G/P . Then, for every g in G, one gets g∗ ν = θ(g −1 , .)ν.

19
Proof. From the definition of s, the action of G on itself is Borel equivalent
to its action on (G/P ) × P defined by g · (ξ, p) = (gξ, σ(g, ξ)p). From lemma
3.2, we know that, under this equivalence, the Haar measure of G may be
written is the measure λ on (G/P ) × P defined by
Z Z
ϕ(ξ, p)dλ(ξ, p) = ∆(p)−1 ϕ(ξ, p)dpdν(ξ),
(G/P )×P (G/P )×P

for every continuous function ϕ with compact support on G × (G/P ). As

this measure is G-invariant, for such a ϕ, we get, for every g in G:
Z Z
−1
∆(p) ϕ(ξ, p)dpdν(ξ) = ∆(p)−1 ϕ(gξ, σ(g, ξ)p)dpdν(ξ)
G×(G/P ) (G/P )×P
Z Z
= θ(g, ξ) ∆(p)−1 ϕ(gξ, p)dpdν(ξ).
(G/P ) P

Therefore, for every continuous function ϕ on G/P and every g in G we get:

Z Z
ϕdν = θ(g, ξ)ϕ(gξ)dν(ξ)
G/P G/P

and the conclusion follows.

3.2 Harmonic densities

We return to the study of hyperbolic spaces. We let X be HnR , HnC or HnQ
and we denote by ∂X the boundary introduced in paragraph 2.2. We set
G = SO◦ (1, n), PU(1, n) or PSp(1, n), following the nature of X and we
conserve the notations A, t 7→ at , N , M , P and ξ0 introduced above. In
order to apply proposition 3.3, we need to compute the modular function of
P:

Lemma 3.4. Let ∆ be the modular function of P . Then, for t in R, n in N

and m in M , we get

(i) if X = HnR , ∆(mat n) = e−(n−1)t .

(ii) if X = HnC , ∆(mat n) = e−2nt .

(iii) if X = HnQ , ∆(mat n) = e−2(2n+1)t .

20
Proof. For a Lie group H the modular function is h 7→ det(Adh )−1 where
Ad denotes the adjoint action of H on its Lie algebra. The result easily
follows.
Let us now introduce the Busemann function. Denote by o the fixed point
of K in X. In projective space, we have at o −−−→ ξ0 . Let us describe the
t→∞
links between the boundary ∂X and the geodesics of X:
Lemma 3.5. Let σ :] − ∞, ∞[→ X be a geodesic. Then σ has two distinct
limit points σ(+∞) and σ(−∞) in ∂X. Conversely, if ξ 6= η are two points
of ∂X, there exists up to parameter translation a unique geodesic σ such that
σ(+∞) = ξ and σ(−∞) = η. If x is a point of X, there exists a unique
geodesic σ such that σ(+∞) = ξ and σ(0) = x.
The proof requires an intermediate lemma.
Let K be R, C or Q, following the nature of X and let n be the K-
dimension of X. Recall that we wrote h., .i for the hermitian product of
signature (1, n) on Kn+1 that allowed us to define the metric of X and q for
the form x 7→ hx, xi. We write ⊥ for orthogonality with respect to q.
Lemma 3.6. Let ξ be in ∂X, that is ξ is a q-isotropic line. Then the form
induced by q on the quotient vector space ξ ⊥ /ξ is negative definite.
Proof. Let v be a non-zero element of ξ and let w be a vector such that
hv, wi 6= 0. As v is isotropic, if V is the plane spanned by v and w, the
restriction of q to V is nondegenerate and has signature (1, 1). Therefore,
the restriction of q to V ⊥ is negative definite. The result follows, since we
have ξ ⊥ = ξ ⊕ V ⊥ .
Proof of lemma 3.5. The first point is true for the geodesic t 7→ at o and
therefore for a general geodesic as G acts transitively on the set of geodesics,
since it acts transitively on points of X and K acts transitively on the unit
sphere of To X. The third point is true for x = o as K acts transitively on
∂X and therefore for any x, as, for every ξ in ∂X, its stabilizer Pξ in G acts
transitively on X by lemmas 2.4, 2.8 and 2.12.
For the second point, let ξ and η be distinct points of the boundary and
choose non-zero vectors v and w in the K-lines ξ and η. There exists α in K
such that pR (hv, wαi) = 0. Let P be the Lagrangian R-plane spanned by v
and wα. Then the restriction of q to P is a real quadratic form which has
isotropic vectors and is nondegenerate by lemma 3.6. Therefore, it contains

21
positive vectors and, by lemmas 2.2, 2.6 and 2.10, its image in X is a geodesic
with limit points ξ and η. It is clearly unique.
We shall need the following:

Lemma 3.7. For every n in N , one gets d(no, at o) − t −−−→ 0.

t→∞

Proof. For any t, we have

|d(no, at o) − t| = |d(no, at o) − d(o, at o)|

= d(o, n−1 at o) − d(o, at o)
 

≤ d(n−1 at o, at o)
= d(o, a−t nat o) −−−→ 0,
t→∞

as a−t nat −−−→ e in G.

t→∞

Let x, y be in X and ξ be in ∂X the Busemann function bξ (x, y) is the

limit limt→∞ t − d(y, r(t)) where r : [0, ∞[→ X is the unique geodesic ray
sutch that r(0) = x and r(t) −−−→ ξ. Such a limit always exists, as we shall
t→∞
see later, by general arguments on negative curved spaces. However, we can
proove its existence and describe its value by a group-theoretic method:

Lemma 3.8. Let g in G, s in R and n in N be such that ξ = gξ0 , x = go

and y = gas no. Then bξ (x, y) = s.

Note that such g, s and n always exist by both transitivities of the action
of K on ∂X ≡ G/P and of the action on AN on X ≡ G/K. As in paragraph
3.1, this formula implies b to satisfy the cocycle identity:

∀x, y, z ∈ X ∀ξ ∈ ∂X bξ (x, z) = bξ (x, y) + bξ (y, z).

Proof. As the Busemann function is invariant under the natural action of G,

it suffices to prove the lemma for g = e. Then, the geodesic ray going from
o to ξ0 is t 7→ at o and we get t − d(as no, at 0) = t − d(no, at−s o) −−−→ s, by
t→∞
lemma 3.7, what should be proved.
We can now come to the extension to hyperbolic spaces of some of the
objects appearing in paragraph 1.2:

22
Proposition 3.9. For each x in X, let νx be the unique probability measure
on ∂X which is invariant under the stabilizer of x in G. Then, for x, y in
X, νx and νy are equivalent and one has

dνy
∀ξ ∈ ∂X (ξ) = e−δX bξ (y,x) .
dνx
where δX = n − 1 if X is HnR , δX = 2n if X is HnC and δX = 2(2n + 1) if X
is HnQ .

Proof. The measures are clearly equivalent as they all belong to the Lebesgue
class of the manifold ∂X. By the equivariance of the formula, it suffices to
show that, for every g in G and ξ in ∂X, dν go
dνo
(ξ) = dgdν∗ oνo (ξ) = e−δX bξ (go,o) .
Let us write ξ = kξ0 and k −1 g ∈ as nK, for some k in K, s in R and
n in N . Then, by lemma 3.8, we get bξ (o, go) = s. On the other hand,
we have g −1 k ∈ Kn−1 a−s and, thus, with the notations of paragraph 3.1,
θ(g −1 , ξ) = ∆(n−1 a−s ) = eδX s , the value of the modular function being given
by lemma 3.4. The result now follows from proposition 3.3.

3.3 The space of geodesics

We will now describe the connection between the measures (νx )x∈X and the
homogeneous invariant measures for geodesic flows of compact manifolds with
universal cover isometric to X.
Consider the homogeneous space G/M . The group M is the stabilizer
in K of the unit vector tangent at o to the geodesic t 7→ at o. As K acts
transitively on the set of unit vectors in To X, the unit tangent bundle of
X identifies G-equivariantly with G/M and the geodesic flows reads as the
action of A by right translations on G/M .
Set ∂ 2 X = ∂X × ∂X − {(ξ, ξ)|ξ ∈ ∂X}. By lemma 3.5, the map which as-
signs to a geodesic its limit points in +∞ and in −∞ induces a G-equivariant
surjection onto ∂ 2 X. In particular, G acts transitively on ∂ 2 X (of course,
you can show this last point directly by proving that P acts transitively on
∂X − {ξ0 }). Let ξ0∨ be the limit in X ∪ ∂X of at o as t goes to −∞ (that is
fn K). Then M A both fixes ξ0 and ξ0∨ and, thus, the surjection G/M → ∂ 2 X
identifies with the natural map G/M → G/M A. In other terms, G/M is the
set of pointed oriented complete geodesics of X and G/M A ≡ ∂ 2 X is the set
of oriented geodesics up to parameter translation.

23
 As M A is an unimodular group, G preserves a measure on G/M A; this
measure has Lebesgue class. For x in X, the measure νx ⊗ νx has Lebesgue
class on ∂ 2 X. Therefore it is equivalent to the invariant measure. In other
terms, there exists a Borel function fx : ∂ 2 X → R∗+ such that for (ξ, η) in
∂ 2 X and g in G, one gets fx (ξ, η) = e−δX (bξ (gx,x)+bη (gx,x)) fx (g −1 ξ, g −1 η). We
shall make such a function explicit; this can be done by a geometric way as
we shall see later. In this paragraph, we will use an algebraic method.
Let x be in X, a point which we view as a vector line in Kn+1 . We denote
by h., .ix the hermitian scalar product for which x and x⊥ are orthogonal and
such that h., .ix = h., .i on x and h., .ix = −h., .i on x⊥ and we let k.kx be
the hermitian norm associated to h., .ix . For g in G and v in Kn+1 , we have
kvkgx = kg −1 vkx . In particular, k.kx is invariant by the stabilizer of x in G.
Let us give a way of computing the Busemann function:

Lemma 3.10. Let x, y be in X and ξ be in ∂X. Then, if v is a non-zero

element of the vector line ξ, we get
kvkx
bξ (x, y) = log .
kvky

Proof. As usual, it suffices to check this formula for ξ = ξ0 , x = o, y = as no,

with s in R and n in N . Then we have kf0 kx = 1, kf0 ky = es and, by lemma
3.8, bξ (x, y) = s.
Let ξ and η be two different points of ∂X and let v and w be non-zero
vectors in the vector lines ξ and η. By lemma 3.6, if ξ 6= η, the q-isotropic
vector w cannot belong to ξ ⊥ , that is we have hv, wi 6= 0. If x is a point of
X, we set s
|hv, wi|
dx (ξ, η) = .
kvkx kwkx
Although we shall not use it, we prove incidentally the following

Lemma 3.11. For any x in X, the function dx : ∂X × ∂X → R+ is a

distance.

Proof. Symmetry is evident and, as pointed above, separation follows from

lemma 3.6. Let us prove the triangle inequality. For this take ξ, η, ζ in
∂X and u, v, w non-zero vectors in ξ, η, ζ. Choose a vector a in x such that
q(a) = kak2x = 1. Then we can write, for some α, β, γ in K and some u0 , v0 , w0

24
in x⊥ , u = aα + u0 , v = aβ + v0 and w = aγ + w0 . As u, v, w are q-isotropic
vectors, we have |α| = ku0 kx , |β| = kv0 kx ,|γ| = kw0 kx . By multiplying u, v, w
by suitable elements of K, we can suppose that ku0 kx = kv0 kx = kw0 kx = 1,
that hu0 , w0 ix ∈ R and that α = β = γ. Then, we have:
p p
dx (ξ, ζ) = |hu, wi| = |1 − hu0 , w0 i|
s 
 1 2 2 2


= 1 −
 ku0 − w0 kx − ku0 kx − kw0 kx 
2
1 1 1
= √ ku0 − w0 kx ≤ √ ku0 − v0 kx + √ kv0 − w0 kx .
2 2 2
But
ku0 − v0 k2x = 2(1 − Re(hu0 , w0 ix )) ≤ 2 |1 − hu0 , w0 ix | = 2 |hu0 , v0 ix |
(where, for t in K, Re(t) stands for 21 (t + t)). Therefore we have
1 1
√ ku0 − v0 kx ≤ dx (ξ, η) and, similarly, √ kv0 − w0 kx ≤ dx (η, ζ)
2 2
and the result follows.
By lemma 3.10, if y is another point of X, we get
1
dy (ξ, η) = e 2 (bξ (x,y)+bη (x,y)) dx (ξ, η).
Summarizing the discussion above, we have the
Proposition 3.12. For x in X, the measure dx−2δX νx ⊗ νx on ∂ 2 X doesn’t
depend on x. It’s up to homothety the unique G-invariant measure on ∂ 2 X.

3.4 Geodesic flows

Suppose Γ is a discrete subgroup of G. If Γ doesn’t contain any element of
finite order, then Γ\X is a manifold. With its quotient metric, it is a locally
symmetric space. The unit tangent bundle of this manifold is Γ\G/M and
the geodesic flow on this space reads as the action of A by right translation.
Let µ be a Radon measure on G/M A. For every continuous function ϕ
with compact support on G/M , set
Z Z Z
ϕdµ̃ = ϕ(gat )dtdµ(gM A).
G/M G/M A R

25
The correspondence µ 7→ µ̃ establishes a G-equivariant bijection between the
set of Radon measures on G/M A and the set of right A-invariant Radon
measures on G/M . In the same way, given a measure λ on Γ\G/M , define a
measure λ̃ on G/M by setting, for any continuous function ϕ with compact
support on G/M ,
Z Z X
ϕdλ̃ = ϕ(gγ)dλ(gΓ).
G/M Γ\G/M γ∈Γ

This defines a G-equivariant bijection between the set of Radon measures on

Γ\G/M and the set of left Γ-invariant Radon measures on G/M . Thus, there
is a natural one-to-one correspondence between the set of left Γ-invariant
Radon measures on G/M and the set of invariant measures for the geodesic
flow on Γ\G/M .
We shall now concentrate on the case where all measures we consider are
in fact G-invariant. A discrete subgroup of G is said to be a lattice if the
G-invariant measure on Γ\G is finite. A cocompact subgroup is a lattice,
but there exists both cocompact and non-cocompact lattices. We have the
following first step in the ergodic theory of geodesic flows of finite volume
locally symmetric spaces:

Theorem 3.13. Let Γ be a lattice in G and normalize the Haar measure of

G in such a way that the associated measure µ on Γ\G/M has total mass
one. Then the action of A on this measure is mixing, that is
Z Z Z
2
∀ϕ, ψ ∈ L (µ) ϕ(x)ψ(xat )dµ(x) −−−→ ϕdµ ψdµ.
Γ\G/M |t|→∞ Γ\G/M Γ\G/M

Proof. This is a consequence of the Howe-Moore theorem about unitary rep-

resentations of G, for which we refer, for example, to [2].

4 Conformal densities
We now come to the core of our subject, that is the construction of conformal
densities on general manifolds with negative curvature. We begin by recalling
general notions about these spaces.

26
4.1 Manifolds with negative curvature
The reader may find a more detailed exposition in [5] and [8].
Let X be a complete simply connected Riemannian manifold with nonpos-
itive sectional curvature, that is, as a function on the Grassmannian bundle
G 2 X of tangent 2-planes of X, the sectional curvature is everywhere nonpos-
itive. Then for every point x of X, the exponential map Expx : Tx X → X is
a diffeomorphism. The space X is thus diffeomorphic to an open ball. There
exists a compactification X̄ = X ∪ ∂X which extends this diffeomorphism to
an homeomorphism with the closed ball. To describe it, consider the set of
geodesic rays r : [0, ∞[→ X. Two rays r1 and r2 are said to be asymptotic
if and only if the function t 7→ d(r1 (t), r2 (t)) is bounded. The boundary ∂X
is the set of equivalence classes of geodesic rays. If r is a geodesic ray, we
denote by r(∞) its equivalence class. If ξ lies in ∂X and x in X there exists
exactly one geodesic ray with origin x such that r(∞) = ξ. We therefore put
on X ∪ ∂X the topology inherited from the sphere compactification of Tx X:
it doesn’t depend on the base point x. If σ : R → X is a complete geodesic,
it has to limit points in ∂X which we denote by σ(+∞) and σ(−∞).
For x, y, z in X, set bz (x, y) = d(x, z) − d(y, z). Then, for any ξ in ∂X
bz (x, y) has a limit as z goes to ξ. We still denote this limit by bξ (x, y). The
function b : X̄ ×X ×X → R is continuous. We call it the Busemann function
of X. For x, y, z in X and ξ in ∂X, we have bξ (x, z) = bξ (x, y) + bξ (y, z). If
σ is a complete unit speed geodesic such that σ(+∞) = ξ, for any s, t in R,
one has bξ (σ(s), σ(t)) = t − s. For x in X and ξ in ∂X, the horosphere with
center ξ based at x is the set of y in X such that bξ (x, y) = 0.
These results are based on the fact that the triangles of X are finer
than the ones of Euclidean space: let x, y, z be points of X and let x0 , y0 , z0
be points of R2 , equipped with its canonical Euclidean structure, such that
d(x, y) = d(x0 , y0 ), d(x, z) = d(x0 , z0 ) and d(y, z) = d(y0 , z0 ). For s, t in [0, 1],
let u (resp. v) be the point of the unique geodesic joining x to y (resp. z) such
that d(x, u) = sd(x, y) (resp. d(x, v) = td(x, z)) and let u0 = (1 − s)x0 + sy0
and v0 = (1 − t)x0 + tz0 (see figure 5). Then we have d(u, v) ≤ d(u0 , v0 ).

Example 4.1. If X is Rn equipped with the Euclidean structure associated

with the canonical scalar product h., .i, the boundary ∂X naturally identifies
with the unit sphere Sn−1 . If u is a unit vector and x and y are two points
of Rn , we have bu (x, y) = hu, y − xi. If u and v are unit vectors, there exists
a complete geodesic with limit points u and v if and only if v = −u.

27
 
z 
z 0

v 

 
v0

 
 y


u y

  u 0
0

x x0

Figure 5: Comparison of distances

Suppose now the sectional curvature of X is bounded above by some

negative constant −c for some c > 0. By normalizing the metric, we can
suppose c = 1. The triangles of X are now finer than those of real hyperbolic
plane. For any two points ξ 6= η, there exists an unique complete geodesic
with limit points ξ and η.
Let ξ 6= η be in the boundary and let x be in X. The Gromov product
(ξ|η)x of ξ and η viewed from x is the quantity 21 (bξ (x, y)+bη (x, y)) where y is
any point of the geodesic with limit points ξ and η; it does not depend on y.
The map dx = e−(.|.)x is a distance on ∂X (with the convention dx (ξ, ξ) = 0,
of course). It clearly satisfies
1
∀x, y ∈ X ∀ξ, η ∈ ∂X dy (ξ, η) = e 2 (bξ (x,y)+bη (x,y)) dx (ξ, η)

(compare with paragraph 3.3). In particular, if g is an isometry of X, its

action extends to the boundary and, for x in X and ξ, η in ∂X, one has
dx (gξ, gη) = dg−1 x (ξ, η) so that dx (gξ, gη) ≤ ed(x,gx) dx (ξ, η).
The compact-open topology makes the group of isometries of X a separa-
ble, locally compact and second countable topological group (see paragraph
2.1). Let us recall the usual classification of its elements. An isometry is
said to be elliptic if it fixes a point in X. It is then contained in a compact
group of isometries. An isometry is said to be parabolic if it fixes exactly one
point in ∂X. It then stabilizes every horosphere centered at its fixed point.
Finally, a non-elliptic isometry is said to be hyperbolic if it fixes two points
in ∂X.

Remark 4.1. All the objects introduced above come from CAT(−1)-

28
 ξ

Figure 6: Shadow

geometry and the theory developed below can in fact be extended to the study
of isometric actions of discrete groups on CAT(−1)-spaces (and especially on
trees). The reader may refer to [4] and [14].

4.2 Shadows and isometries

Let always X be complete simply connected with curvature ≤ −1. If x and y
are points of X and r is a positive real number, we define the shadow Or (x, y)
to be the set of ξ in ∂X such that the geodesic ray issued from x with limit
point ξ hits the closed ball of center y with radius r.
We shall use the following

Lemma 4.1. Let x, y be in X and r > 0. For ξ in Or (x, y), one has

d(x, y) − 2r ≤ bξ (x, y) ≤ d(x, y).

Proof. Let t 7→ xt be the geodesic ray such that x0 = x and x∞ = ξ and let
z be a point of that geodesic ray being at distance ≤ r to y. For t ≥ 0, we
have

d(y, xt ) ≤ d(y, z) + d(z, xt ) = d(y, z) + d(x, xt ) − d(x, z)

≤ 2d(y, z) + d(x, xt ) − d(x, y) ≤ 2r + d(x, xt ) − d(x, y)

and thus bξ (x, y) = limt→∞ d(x, xt ) − d(y, xt ) ≥ d(x, y) − 2r. The lemma
follows, the other inequality being obvious.
Note that shadows may in some sense be very large:

29
 x

Figure 7: Large shadow

Lemma 4.2. Let x be in X. For every y 6= x let ηy be the limit point of the
geodesic going from x to y Then as r goes to ∞, one has
sup dx (ξ, ηy ) −−−→ 0.
y6=x r→∞
ξ6∈Or (y,x)

Proof. Suppose to the contrary there are ε > 0, rn → ∞, yn in X and ξn

in X − Orn (yn , x) such that dx (ξ, ηyn ) ≥ ε for each n. Then after eventually
choosing a subsequence, we can suppose that, for some ξ and η in ∂X, one
has ξn → ξ and yn → η. We then have ηyn → η and dx (ξ, η) ≥ ε. Therefore
there exists a ball with center x that hits the geodesic line from η to ξ, what
contradicts the fact that rn → ∞.
Shadows will allows us to find pieces of the boundary where isometries
contract the distances:
Lemma 4.3. Let g be an isometry of X, x a point and r > 0. Then for ξ, η
in Or (g −1 x, x), one has
dx (gξ, gη) ≤ e2r−d(x,gx) dx (ξ, η).
Proof. We have
1 −1 x)+b −1 x))
dx (gξ, gη) = dg−1 x (ξ, η) = e 2 (bξ (x,g η (x,g
dx (ξ, η).
By lemma 4.1, we have
bξ (g −1 x, x) ≥ d(x, gx) − 2r and bη (g −1 x, x) ≥ d(x, gx) − 2r,
what implies the result.

30
 Let us use this lemma to give more details on hyperbolic isometries:
Proposition 4.4. Let g be a hyperbolic isometry of X, ξ 6= η two fixed points
of g in ∂X and x a point of the geodesic D joining ξ to η. Suppose that gx
lies between x and ξ. Then l(g) = d(x, gx) > 0 doesn’t depend on x on D
and for any y in X, one has l(g) = bξ (y, gy) ≤ d(y, gy). For any ζ 6= η in
∂X, one has dx (g n ζ, ξ) = O(e−nl(g) ), uniformly for ζ bounded away from η.
In particular, ξ and η are the unique fixed points of g in ∂X.
The quantity l(g) is called the translation length of g. In the sequel, we
shall denote by g + and g − the attractive and repulsive fixed points of g.
Proof. As g fixes ξ and η, it stabilizes D and induces a translation on it. Since
we assumed g were not elliptic, this translation is not trivial and l(g) doesn’t
depend on x and is not trivial. Let y be in X and z be the unique point
of D such that bξ (y, z) = 0. Then one has bξ (y, gy) = bξ (y, z) + bξ (z, gz) +
bξ (gz, gy). As g fixes ξ, bξ (gz, gy) = bξ (z, y) and thus bξ (y, gy) = bξ (z, gz) =
d(z, gz) = l(g). Finally let r be a positive number. Then, by lemma 4.3, for
every ζ in Or (g −1 x, x), one has dx (gζ, ξ) ≤ e2r−l(g) dx (ζ, ξ). The result follows
now by lemma 4.2.

4.3 Groups of isometries

Let Γ be a discrete group of isometries of X. We will say that Γ is non-
elementary if it does not stabilize any finite subset of X ∪ ∂X. If Γ is such
a subgroup, its limit set ΛΓ is the set of limit points of Γx in ∂X where x is
any point of X (by definition of the boundary, this set doesn’t depend on x).
The exponent of growth of Γ is the exponent of convergence of the Dirichlet
series X
e−sd(x,γx) (s ∈ R),
γ∈Γ
that is the quantity
1
lim sup log(card{γ ∈ Γ|d(x, γx) ≤ r}).
r→∞ r

We will assume this quantity to be finite. The next lemma shows that it is
the case in most interesting examples. Let m denote the Riemannian volume
of X and δX its volume entropy
1
δX = lim sup log(m(B(x, r))) < ∞.
r→∞ r

31
Lemma 4.5. Assume δX < ∞. Let Γ be a discrete group of isometries of X.
Then one has δΓ ≤ δX < ∞. If Γ is cocompact, then δΓ = δX and ΛΓ = ∂X.

For symmetric spaces, δX is finite and is the number appearing in section

3. More generally, δX is finite, if the sectional curvature of X is negatively
pinched, that is if it lies between two negative constants. It is the case when
X possesses a cocompact group of isometries.
Proof. As Γ is discrete, there exists a real number s > 0 such that, for every
γ in Γ, B(γx, s) ∩ B(x, s) 6= ∅ ⇒ γx = x. Let n be the number of elements
of Γ that fix x. For r > 0, we have

card{γ ∈ Γ|d(x, γx) ≤ r}m(B(x, s)) ≤ nm(B(x, r + s))

Γ ≤ δX . Suppose now Γ is cocompact. There exists s > 0

and therefore δS
such that X = γ∈Γ B(γx, s). Thus, for r ≥ 0, we have
[
B(x, r) ⊂ B(γx, s)
γ∈Γ
d(x,γx)≤r+s

and
m(B(x, r)) ≤ card{γ ∈ Γ|d(x, γx) ≤ r + s}m(B(x, s))
so that δX ≤ δΓ . Finally, let ξ be a point of ∂X and (xn ) a sequence of
points of X converging to ξ. There exists (γn ) in Γ such that, for each n,
d(xn , γn x) ≤ s. Therefore γn x → ξ and ξ belongs to ΛX .
Let us see how to construct non-elementary discrete groups of isometries.
This is the classical example of Schottky groups.

Lemma 4.6. Let g and h be two hyperbolic isometries of X with no common

fixed point and let x in X. Then, after eventually having replaced g and h by
powers of themselves, the group Γ of isometries generated by g and h is free
and discrete and there exists 0 < k < inf(d(x, gx), d(x, hx)) such that, for γ
in Γ, if γ = g1 . . . gn is the decomposition of γ as a reduced word in g and h,
one has
d(x, γx) ≥ d(x, g1 x) + . . . + d(x, gn x) − nk.
In particular 0 < δΓ < ∞.

32
Proof. Fix ε > 0 such that the dx -balls of radius ε centered at the fixed
points of g and h are all disjoint and their union does not cover ∂X. By
proposition 4.4, g −n x −−−→ g − so that, after having replaced g by a power,
n→∞
by lemma 4.2, we can find an r > 0 such that Or (g −1 x, x) contains all
points of ∂X − Bx (g − , ε). Then, by lemma 4.3, for ξ and η in this shadow,
we have dx (gξ, gη) ≤ e2r−d(x,gx) dx (ξ, η) ≤ e2r−l(g) dx (ξ, η). After again hav-
ing replaced g by some power, we can suppose l(g) > 2r and g(∂X −
Bx (g − , ε)) ⊂ Bx (g + , ε). Doing the same job for g −1 , h and h−1 , we fi-
nally get on one hand ∂X − Bx (g + , ε) ⊂ Or (gx, x), ∂X − Bx (h− , ε) ⊂
Or (h−1 x, x) and ∂X −Bx (g + , ε) ⊂ Or (gx, x) and on the other hand g −1 (∂X −
Bx (g + , ε)) ⊂ Bx (g − , ε), h(∂X − Bx (h− , ε)) ⊂ Bx (h+ , ε) and h−1 (∂X −
Bx (h+ , ε)) ⊂ Bx (h− , ε).
Let now γ = g1 . . . gn be a reduced word in g and h, that is each gi is g,
h, g −1 or h−1 and gi+1 6= gi−1 . We have to show that γ, as an isometry of
X, is far away from the identity. Take ξ in ∂X which does not belong to
the union of the four balls constructed above. Then gn ξ belongs to the ball
centered at the attractive fixed point of gn and, by induction, γξ belongs to
the ball centered at the attractive fixed point of g1 , what implies the result.
In the preceding construction, for each 1 ≤ i ≤ n, we have gi+1 . . . gn ξ ∈
Or (gi−1 x, x) so that, by lemma 4.1, bgi+1 ...gn ξ (gi−1 x, x) ≥ d(gi−1 x, x) − 2r =
d(x, gi x) − 2r. Therefore we have
n
X
−1 −1
d(x, γx) = d(γ x, x) ≥ bξ (γ x, x) = bξ (gn−1 . . . gi−1 x, gn−1 . . . gi+1
−1
x)
i=1
n
X
= bgi+1 ...gn ξ (gi−1 x, x)
i=1
n
X
≥ d(x, gi x) − 2nr,
i=1

what should be proved. The estimates on δΓ now follow from exponential

growth of the free group.
We can now say something about general discrete subgroups:
Proposition 4.7. Let Γ be a non-elementary discrete group of isometries of
X. Then Γ contains hyperbolic isometries. The set ΛΓ is the closure of the
set of fixed points of hyperbolic elements of Γ and is the smallest nonempty
closed Γ-invariant subset of ∂X. The exponent δΓ is positive.

33
Proof. Let x be a point of X, ξ a point of ∂X and (γn ) be a sequence of
elements of Γ such that γn x → ξ. For each n, let ξn be the limit point of
the geodesic ray joining x to γn x and ηn the limit point of the ray joining x
to γn−1 x. Then, ξn → ξ and after extracting a subsequence, we can suppose
that ηn → η for some η. As Γ is non-elementary, there exists f in Γ such
that dx (ξ, f η) > 0 and after replacing (γn ) by (γn f −1 ), we can suppose that
η 6= ξ. Choose 0 < ε ≤ 15 dx (ξ, η). By lemma 4.2, there exists an r > 0
such that, for any y 6= x in X, ∂X − Or (y, x) is contained in the dx -ball
of radius ε with center the limit point of the geodesic ray joining x to y.
Then, for sufficiently large n, we have ∂X − Or (γn−1 x, x) ⊂ Bx (η, 2ε) and
∂X − Or (γn x, x) ⊂ Bx (ξ, 2ε). What’s more, by lemma 4.3, γn is e2r−d(x,γn x) -
Lipschitz on Or (γn−1 x, x). As γn Or (γn−1 x, x) = Or (x, γn x) 3 ξn , for n suffi-
ciently large, γn possesses an attractive fixed point in Bx (ξn , ε) ⊂ Bx (ξ, 2ε).
In the same way, γn−1 possesses an attractive fixed point in Bx (η, 2ε). Thus
γn is hyperbolic and dx (γn+ , ξ) ≤ ε, what should be proved.
Let now F be a closed Γ-invariant nonempty set in ∂X and let γ be a
hyperbolic element of Γ. Let ξ be an element of F . Then there exists f in
Γ such that f ξ 6= γ − . We then have γ n f ξ → γ + , thus γ + belongs to F . As
this is true for any hyperbolic isometry γ in Γ, we have ΛΓ ⊂ F .
Finally, let γ be an hyperbolic element of Γ. There exists an element f of
Γ such that f γ + 6= γ + , f γ − 6= γ + , f γ + 6= γ − and f γ − 6= γ − . In other words,
the isometries g = f γf −1 and h = γ satisfy the hypothesis of lemma 4.6.
Therefore, by this lemma, Γ contains a subgroup Γ0 with δΓ0 > 0. Hence we
have δΓ > 0.

4.4 Patterson construction

Let always Γ be a (non-elementary) discrete group of isometries of X and let
β be a real number. A Γ-conformal density of dimension β is a map x 7→ νx
from X to the set of Radon measures on ∂X which is Γ-equivariant, that is
γ∗ νx = νγx for γ in Γ and x in X, and such that, for each x, y in X, νy and
νx are equivalent and we have
dνy
∀ξ ∈ ∂X (ξ) = e−βbξ (y,x) .
dνx
Fixing a base point x in X, the data of a conformal density of dimension β
is equivalent to the one of a Radon measure νx such that, for each γ in Γ,
one has γ∗ νx = e−βb. (γx,x) .

34
Example 4.2. From proposition 3.9, we know that if X is a hyperbolic
space, there exists a conformal density of dimension δX which is equivariant
under the full group of isometries.

Conformal densities have been originally constructed by Patterson for

isometries of the real hyperbolic plane but this construction extends to our
general situation:

Theorem 4.8. Let Γ be a non-elementary discrete group of isometries of X.

Then there exists a Γ-conformal density of dimension δΓ with support ΛΓ .
−δΓ d(x,γx)
P
Proof. Fix x in X and suppose first that γ∈Γ e = ∞. Set, for
−sd(x,γx)
P
s > δΓ , Φ(s) = γ∈Γ e and

1 X −sd(x,γx)
νs = e Dγx
Φ(s) γ∈Γ

where, for y in X, Dy is the Dirac measure at y. Then, for s > δΓ , νs may

be seen as a probability measure on the compact space X ∪ ∂X. Therefore,
there exists a sequence sn → δΓ such that νsn converges weakly to some
probability measure ν. As Φ(sn ) → ∞, for each r ≥ 0, νsn (B(x, r)) → 0 and
ν is concentrated on ∂X. As the support of each νs is Γx ∪ ΛΓ , ν has support
⊂ ΛΓ . Let now θ be an element of Γ. For s > δΓ , we have
1 X −sd(x,γx)
θ∗ νs = e Dθγx
Φ(s) γ∈Γ
1 X −sd(x,θ−1 γx)
= e Dγx
Φ(s) γ∈Γ
1 X −s(d(θx,γx)−d(x,γx)) −sd(x,γx)
= e e Dγx .
Φ(s) γ∈Γ

Consider the function ϕ : X ∪ ∂X → R such that ϕ(y) = d(θx, y) − d(x, y)

for y in X and ϕ(ξ) = bξ (θx, x) for ξ in ∂X. This function is continuous and
for each s > δΓ , we have θ∗ νs = e−sϕ νs . Therefore θ∗ ν = e−δΓ b. (θx,x) and we
have a Γ-conformal density of dimension δΓ , by the remark above. Finally, as
ν is Γ-quasi-invariant its support is Γ-invariant. As this support is contained
in ΛΓ , it is ΛΓ by proposition 4.7.

35
 Let now γ∈Γ e−δΓ d(x,γx) be < ∞. Then we can make the same construc-
P

tion by setting Φ(s) = γ∈Γ h(d(x, γx))e−sd(x,γx) and

P

1 X
νs = h(d(x, γx))e−sd(x,γx) Dγx ,
Φ(s) γ∈Γ

where h is the
P function provided by the following lemma applied to the
measure λ = γ∈Γ Dd(x,γx) on R+ .
Lemma 4.9. Let λ be a Radon measure on R+ , such that the Laplace trans-
form of λ Z
e−st dλ(t) (s ∈ R)
R+

has critical exponent δ ∈ R. Then there exists a nondecreasing function

h : R+ → R∗+ with the following properties:
(i) one has Z
h(t)e−δt dλ(t) = ∞.
R+

(ii) for every ε > 0, there exists t0 ≥ 0 such that, for any u ≥ 0 and
t ≥ t0 , one has
h(u + t) ≤ eεu h(t).

In particular, the Laplace transform of hλ has critical exponent δ.

Proof. Choose a decreasing sequence (εn )n∈N of positive real numbers, going
to 0. By induction, we will construct an increasing sequence of real numbers
(tn )n∈N with t0 = 0 and a function h : R+ → R∗+ such that, for n in N, the
logarithm of h will be affine with slope εn on [tn , tn+1 ] and that
Z
h(t)e−δt dλ(t) ≥ 1.
[tn ,tn+1 [

Such a function will clearly satisfy the conclusions of the lemma.

Let therefore n be a nonegative integer and suppose t0 < . . . < tn and
h : [t0 , tn ] → R∗+ as above are constructed. Let hn : [tn , ∞[→ R∗+ the function
which is logarithmically affine with slope εn and such that hn (tn ) = h(tn ).
As Z
e−(δ−εn )t dλ(t) = ∞,
R+

36
we have Z
hn (t)e−δt dλ(t) = ∞
[tn ,∞[

and, thus, there exists a real number tn+1 > tn such that
Z
hn (t)e−δt dλ(t) ≥ 1.
[tn ,tn+1 [

We then set, for t in ]tn , tn+1 ], h(t) = hn (t) and this construction may be
pursued by induction.

4.5 Sullivan’s shadow lemma

We shall now emphasize the connection between growth of groups and densi-
ties. The key point there is the following lemma due to Sullivan, which allows
to estimate the measure of certain subsets of the boundary with respect to
conformal densities:
Lemma 4.10. Let Γ be a non-elementary group of isometries of X, ν a
Γ-conformal density of dimension β and x a point of X. Then there exists
r0 > 0 such that, for every r ≥ r0 , there exists C > 0 such that, for any γ in
Γ, one has
1 −βd(x,γx)
e ≤ νx (Or (x, γx)) ≤ Ce−βd(x,γx) .
C
Proof. As Γ is non-elementary, the support of νx is not a point and, for every ξ
in ∂X, one has νx ({ξ}) < νx (∂X). Therefore, by compacity of the boundary,
there exists  > 0 such that, for any ξ in ∂X, νx (∂X − Bx (ξ, ε)) ≥ ε. By
lemma 4.2, there exists r0 > 0 such that, for r ≥ r0 , for y in X, one has
∂X − Bx (ξ, ε) ⊂ Or (y, x). Let γ be in Γ. We have

νx (Or (x, γx)) = νx (γOr (γ −1 x, x)) = νγ −1 x (Or (γ −1 x, x))

Z
−1
= e−βbξ (γ x,x) dνx (ξ).
Or (γ −1 x,x)

Thus, by lemma 4.1, if β ≥ 0,

εe−βd(x,γx) ≤ νx (Or (x, γx)) ≤ νx (∂X)e2βr e−βd(x,γx) ,

and the lemma follows, the case β < 0 being handled similarly (and being
empty, as we shall see below !)

37
 θx

x
γx
ξ

Figure 8: Shadow lemma and covering

From this lemma we deduce the

Theorem 4.11. Let Γ be a discrete group of isometries of X. If there exists

a Γ-conformal density of dimension β, then one has β ≥ δΓ .

Proof. Let x be a point of X and r > 0 and C > 0 as in lemma 4.10. For n
in N, let Γn be the set of elements γ in Γ such that n ≤ d(x, γx) < n + 1 and
an the cardinal of Γn . Then we have δΓ = lim supn→∞ n1 log an . Let γ and θ
be in Γn and suppose Or (x, γx) ∩ Or (x, θx) 6= ∅. Let ξ be in this set. Then
the geodesic ray going from x to ξ contains a point y such that d(y, γx) ≤ r.
As n ≤ d(x, γx) ≤ n + 1, one has n − r ≤ d(x, y) ≤ n + 1 + r. In the same
way, choose a point z on the geodesic ray from x to ξ such that d(z, θx) ≤ r,
and hence n − r ≤ d(x, z) ≤ n + 1 + r (see figure 8). As y and z ly in the same
geodesic ray, we have d(y, z) ≤ 1 + 2r and, therefore, d(γx, θx) ≤ 1 + 4r. In
other words, if p is be the number of γ in Γ such that d(x, γx) ≤ 1 + 4r, for
every ξ in ∂X, one has card{γ ∈ Γn |ξ ∈ Or (x, γx)} ≤ p. Hence, if β ≥ 0, for
any n, we have
!
[
νx (∂X) ≥ νx Or (x, γn x)
γ∈Γn
1 X
≥ νx (Or (x, γn x))
p γ∈Γ
n

1 X −βd(x,γn x) e−β(n+1)
≥ e ≥ an ,
pC γ∈Γ pC
n

that is an ≤ pCνx (∂X)eβ(n+1) . Therefore β ≥ δΓ . The case β < 0 is handled

similarly and reveals to be empty.

38
4.6 Geodesic flows
Conformal densities have two major applications. The first one is the inves-
tigation of the geodesic flow of the quotient orbifold Γ\X, in analogy with
what has been done for lattices in classical symmetric spaces at paragraphs
3.3 and 3.4.
As every geodesic in X has two limit points in ∂X, the set of oriented
geodesics of X naturally identifies with ∂ 2 X = ∂X × ∂X − {(ξ, ξ)|ξ ∈ ∂X}.
Let T1 X be the unit tangent bundle of X. Then the action of the geodesic
flow (g t ) in T1 X is proper and the quotient of T1 X by this action is the
set of oriented geodesics ∂ 2 X. As a consequence, there is a natural bijection
between the set of invariant Radon measures for the geodesic flow in T1 X and
the set of Radon measure on ∂ 2 X. As the action of the group of isometries of
X on T1 X commutes with the geodesic flow, the bijection above is equivariant
under this action.
Let Γ be a non-elementary discrete group of isometries. If no non-trivial
element of γ has a fixed point in X, that is if Γ doesn’t contain any element
of finite order, then Γ\X has a natural structure of Riemannian manifold.
Its geodesic flow is the quotient action of (g t ) on Γ\T1 X. In any case,
this action exists. We still denote it by (g t ). The discussion above shows
that there exist a natural bijection between the set of (g t )-invariant Radon
measures on Γ\T1 X and the set of Γ-invariant Radon measures on ∂ 2 X.
In the same way, there is a bijection between the set of closed (g t )-
invariant subsets of Γ\T1 X and the set of closed Γ-invariant subsets of ∂ 2 X.
In particular, let FΓ = (ΛΓ × ΛΓ ) ∩ ∂ 2 X and let EΓ be the corresponding
(g t )-invariant subset of Γ\T1 X: EΓ is the image of the set of unit vectors
whose tangent geodesic has both its limit points in ΛΓ . Let γ be a hyperbolic
element of Γ. Then (γ + , γ − ) belongs to FΓ . As the geodesic going from γ − to
γ + is stable by Γ, the image of any of its unit tangent vectors in Γ\T1 X has
closed orbit under the flow (g t ). Conversely, every closed orbit is obtained
this way.

Lemma 4.12. The set EΓ is the closure of the union of closed (g t )-orbits in
Γ\T1 X. Equivalently, the points of the form (γ + , γ − ), where γ is a hyperbolic
element of Γ, is dense in FΓ .

Proof. Let ξ 6= η be elements of ΛΓ . By proposition 4.7, there exists se-

quences of hyperbolic elements (γn ) and (ηn ) in Γ such that γn+ → ξ and
θn− → η. After extracting subsequences, we can suppose that γn− → ζ and

39
that θn+ → ς for some ζ and ς in ΛΓ . Let f be an element of Γ such that
f ς 6= ζ. Let V and W be open neighborhoods of ξ and η that do not in-
tersect. Then, for large n, the isometry ρn = γn f θn acts on ∂X − W as a
contraction with image ⊂ V , and ρ−1
n acts on ∂X − V as a contraction with
image ⊂ W , that is ρn is hyperbolic and ρ+ −
n → ξ and ρn → η, what should
be proved.
Consider now a Γ-conformal density ν of dimension β and let x be a point
of X. As Γ is non-elementary, νx is not a Dirac measure and therefore the
restriction to ∂ 2 X of the measure νx ⊗ νx is non-trivial. On ∂X, the product
of this measure by the function d−2βx is Γ-invariant and does not depend on
x. To this measure, we associate a (g t )-invariant measure µ on Γ\T1 X: we
say that µ is the Bowen-Margulis measure associated to ν. Note that µ is
not necessarily finite (compare with paragraph 3.3).
Let G be a locally compact topological group and let (E, λ) be a σ-finite
measure space with a measure preserving Borel action of G. Choose a right
Haar measure on G. A measurable subset A of E is said to be wandering if,
for λ-allmost every x in A, the set {g ∈ G|gx ∈ A} has finite Haar measure.
There exists up to sets of measure 0 a unique partition E = EC ∪ ED into G-
invariant measurable subsets such that ED is a countable union of wandering
sets and every wandering set is contained in ED . The set ED is called the
dissipative part of E and EC its conservative part. The system is said to be
conservative (resp. dissipative) if λ(ED ) = 0 (resp. λ(EC ) = 0).
The following theorem has a long story. One of its ingredients is the
original proof of the ergodicity of the geodesic flow of surfaces by Hopf. This
method has been extended to the study of conformal densities by Sullivan.
The general form we give here is due to Roblin in [14], where the reader may
find its proof.

Theorem 4.13. Let Γ be a non-elementary discrete group of isometries of

X and let ν be a Γ-conformal density of dimension β with associated Bowen-
Margulis measure µ. Fix an arbitrary point x of X. Then either one has
simultaneously
−βd(x,γx)
P
(i) γ∈Γ e < ∞.

(ii) the action of Γ on (∂ 2 X, νx ⊗ νx ) is dissipative.

(iii) the action of (g t ) on (Γ\T1 X, µ) is dissipative.

40
or

e−βd(x,γx) = ∞.
P
(i) γ∈Γ

(ii) the action of Γ on (∂ 2 X, νx ⊗ νx ) is conservative and ergodic.

(iii) the action of (g t ) on (Γ\T1 X, µ) is conservative and ergodic.

Note that if the second condition is true one has necessarily β = δΓ thanks
to theorem 4.11.
A group Γ such that γ∈Γ e−δΓ d(x,γx) = ∞ is said to be of divergence type.
P
The groups constructed in lemma 4.6 can be showed to be of divergence type.
P 
Remark 4.2. Let E be a countable set and y∈E p(x, y)Dy a family
x∈E
of transition probabilities on E (where D stands for Dirac measure). Then,
if x is a point
P∞ of E, the associated Markov chain starting at x is recurrent if
and only if n=0 pn (x, x) = ∞, with
X
pn (x, x) = p(x, x1 )p(x1 , x2 ), . . . , p(xn−1 , x)
x1 ,...,xn−1 ∈E

for n ≥ 0. This fact relies on the Borel-Cantelli lemma. A version of the

Borel-Cantelli lemma appears in the proof of theorem 4.13 too.

Corollary 4.14. Let Γ be a non-elementary discrete group of isometries

of X. If Γ is of divergence type, there exists up to homothety a unique Γ-
conformal density of dimension δΓ . It is concentrated on ΛΓ .

Proof. The existence is provided by theorem 4.8. For the uniqueness, con-
sider a density ν of dimension δΓ .
Let us show that ν has no atoms. By theorem 4.13, the associated Bowen-
Margulis measure µ is (g t ) conservative and therefore, if x is a point of X,
and r and C are as in lemma 4.10, for νx -almost ξ in ΛΓ , if t 7→ xt is the
geodesic ray joining x to ξ, there exists a sequence tn → ∞ and a sequence
(γn ) of elements of Γ such that d(γn x, xtn ) ≤ r. One then has d(x, γn x) → ∞
and, for any n, ξ ∈ Or (x, γn x). Therefore, for any n, νx ({ξ}) ≤ Ce−δΓ d(x,γn x) ,
what implies νx ({ξ}) = 0, what should be proved.
As νx has no atoms, the set {(ξ, ξ)|ξ ∈ ∂X} has νx ⊗ νx -measure 0 and
the projection on the first component from ∂ 2 X onto ∂X maps νx ⊗ νx to νx .

41
In particular the action of Γ on (∂X, νx ) is ergodic. As this is true for any
density of dimension δΓ and as a convex combination of two densities of the
same dimension is again a density, there exists only one density of dimension
δΓ .
For cocompact groups, the measure µ is finite and therefore conservative
by Poincaré recurrence theorem. We therefore have the

Corollary 4.15. Suppose X possesses a discrete cocompact group of isome-

tries. Then there exists up to homothety a unique map x 7→ νx from X to
the set of Radon measures on ∂X such that

(i) for any x, y in X, νx and νy are equivalent and

dνy
∀ξ ∈ ∂X (ξ) = e−δX bξ (y,x) .
dνx

(ii) for any isometry g of X one has g∗ νx = νgx .

Proof. Let Γ be a discrete cocompact group of isometries. Uniqueness comes

from the fact that such a map is a Γ-conformal density of dimension δX
and that Γ is of divergence type. For the existence, consider a Γ-conformal
density ν of dimension δX and fix a point x in X. Let G be the group of
isometries of X. Then Γ is a discrete cocompact subgroup of G and, as G
possesses such a subgroup, it is unimodular. Fix a Haar measure dg on G
and still denote by dg the induced finite measure on G/Γ.
R Let ϕ be a continuous function on ∂X. Consider the function ψ(ϕ) : g 7→
δX bgξ (gx,x)
∂X
e ϕ(gx)dνx (ξ). For γ in Γ, one has
Z
ψ(ϕ)(gγ) = eδX bgξ (gγx,x) ϕ(gξ)d(γ∗ νx )(ξ)
Z∂X
= eδX (bgξ (gγx,x)−bξ (γx,x)) ϕ(gξ)dνx (ξ)
Z∂X
= eδX (bgξ (gγx,x)−bgξ (gγx,gx)) ϕ(gξ)dνx (ξ) = ψ(ϕ)(g).
∂X

We define
R a measure
R ν̃x on ∂X by setting, for any continuous function ϕ on
∂X, ∂X ϕdν̃x = G/Γ ψ(ϕ)(g)dg. Let h be in G and let us compute h∗ ν̃x .

42
For ϕ a continuous function on ∂X, for g in G, we have
Z
ψ(ϕ ◦ h)(g) = eδX bgξ (gx,x) ϕ(hgx)dνx (ξ)
Z∂X
= eδX (bhgξ (hgx,x)−bhgξ (hx,x)) ϕ(hgx)dνx (ξ)
∂X
= ψ e−δX b· (hx,x) ϕ (hg)

and therefore
Z Z Z
ϕd(h∗ ν̃x ) = ψ(ϕ ◦ h)(g)dg = ψ(e−δX b· (hx,x) ϕ)(hg)dg
∂X G/Γ G/Γ
Z Z
−δX b· (hx,x)
= ψ(e ϕ)(g)dg = e−δX bξ (hx,x) ϕ(ξ)dν̃x (ξ),
G/Γ ∂X

what is to say h∗ ν̃x = e−δX b· (hx,x) ν̃x . By setting, for y in X, ν̃y = e−δX b· (y,x) ν̃x ,
we get a map satisfying the properties required in the corollary. Note that,
by uniqueness, we have in fact ν̃ = ν!
We finally may ask if the analogy with the homogeneous theory may
be pursued and if we have an analogous result to theorem 3.13. If Γ is
cocompact and has no finite order element, the geodesic flow of Γ\T1 X is
an Anosov contact flow and the Bowen-Margulis measure associated to its
unique conformal density of dimension δX is the measure of maximal entropy
of this flow (this fact explains the terminology). It is therefore mixing. The
reader may refer to [9] for the general notions involved in the discussion
above.
To extend theorem 3.13, we however need a supplementary hypothesis on
Γ: we say that Γ has non-arithmetic spectrum if the subgroup of R generated
by the numbers l(γ), where γ is an hyperbolic element of Γ, is dense. This
may be shown to always hold if Γ is cocompact or if X is a symmetric space.
Note that we don’t have any example where we can show this to be false.
Suppose that for some Γ-conformal density the associated Bowen-Margu-
lis measure has finite mass. By theorem 4.13 and Poincaré recurrence the-
orem, the dimension of the density is δΓ and, by corollary 4.14, it is the
unique conformal density of dimension δΓ . We then say that the associated
Bowen-Margulis measure is the Bowen-Margulis measure of Γ and that Γ has
finite Bowen-Margulis measure.
We have the following theorem, due to Roblin ([14]) in this general form:

43
Theorem 4.16. Let Γ be a non-elementary discrete group of isometries of
X with non-arithmetic spectrum and finite Bowen-Margulis measure. Then
this measure is mixing under the action of the geodesic flow.

4.7 Conformal densities and potential theory

Conformal densities appear to be a powerful tool in some rigidity problems
connected to the barycenter method. As an example, of such a connection,
we give here a characterization of symmetric spaces due to Ledrappier [12]
and Besson, Courtois and Gallot [3]. We suppose here the curvature of X is
pinched.
Let ∆ be the Laplace-Beltrami operator on X. A C 2 function ϕ : X → C
is said to be harmonic if ∆ϕ = 0. There exists an extension of theorem 1.1,
which is due to Anderson and Schoen [1]:

Theorem 4.17. There exists a map x 7→ µx from X to the set of probability

measures on ∂X with the following properties:

(i) The measures µx , for x in X, are all equivalent and, for x, y in X,

dµy
dµx
is a continuous function on ∂X.

(ii) Set, for any essentially bounded function f on ∂X with respect

R to the
measure class of the (µx )x∈X and for any x in X, Pf (x) = ∂X f dµx .
Then the correspondence f 7→ Pf establishes a L∞ -norm preserving
isomorphism between L∞ (∂X, (µx )x∈X ) and the space of bounded har-
monic functions on X.

The map µ : x 7→ µx is called the harmonic density of X.

Suppose that X possesses a discrete cocompact group of isometries. On
the boundary we therefore have the harmonic density µ of X and the confor-
mal density ν of dimension δX provided by corollary 4.15. If X is a symmetric
space we have µ = ν (see appendix B). This is the only case where it can
happen:

Theorem 4.18. Suppose there exists x in X such that νx and µx are pro-
portional. Then X is a symmetric space.

44
A Curvature and submanifolds of symmetric
spaces
We finish here the proofs of lemmas 2.3, 2.6, 2.7, 2.10 and 2.11 by deter-
mining the sectional curvature of hyperbolic spaces and the totally geodesic
submanifolds of complex and quaternionic hyperbolic spaces.
We go back to notations of paragraph 2.2.

A.1 Curvature of real hyperbolic space

We refer to [7] for general notions of Riemannian geometry.
Denote by D the covariant derivative for HnR . Let x be a point of HnR , X
and Y two vector fields defined in an Rn+1 -neighbourhood of x. Then (DX Y )x
is simply the q-orthogonal projection on x⊥ = Tx HnR of the Euclidean Rn+1 -
derivative at x of Y in the direction Xx at x: indeed this defines a connection
on HnR , which clearly satisfies the Levi-Civita axioms.
Let R be the curvature tensor of HnR . We have the following:
Proposition A.1. Let v and w be unit orthogonal tangent vectors of HnR at
some point x. Then we have R(v, w)v = −w. In particular, HnR has constant
sectional curvature −1.
Proof. By transitivity of the group of isometries and by lemma 2.2, it suffices
to prove the result for n = 2, x = (1, 0, 0), v = (0, 0, 1) and w = (0, 1, 0). For
 geodesic c : t 7→
this, we shall construct a Jacobi vector field on the unit speed 
1 0 0
(cosh t, 0, sinh t). As for each s in R, the matrix ks = 0 cos s sin s 
0 − sin s cos s
◦
belongs to SO (1, n), the map
H : (s, t) 7→ (cosh t, sin s sinh t, cos s cosh t) = ks c(t)
is a geodesic variation with H(0, .) = c. Therefore, the vector field

∂H(s, t) 
X(t) = = (0, sinh t, 0)
∂s s=0
is a Jacobi vector field on c, that is is satisfies Dc20 X = −R(c0 , X)c0 where R is
the curvature tensor. But, by the remark above, the vector field t 7→ (0, 1, 0)
is parallel along c and, therefore, we have Dc20 X = X so that R(c0 , X)c0 = −X
and the sectional curvature has constant value −1.

45
A.2 Curvature and submanifolds of complex hyper-
bolic space
Let still R denote the curvature tensor. Note that, by lemma 2.5 and propo-
sition A.1, the complex hyperbolic line H1C has constant sectional curvature
−4. We can use the computation of the curvature of real hyperbolic space
and complex hyperbolic line to prove the following
Proposition A.2. Let v and w be unit tangent vectors of HnC at some point
x such that Rehv, wix = 0. If hv, wix = 0, we have R(v, w)v = −w. If
w = iv, we have R(v, w)v = −4w. In particular, HnC has sectional curvature
lying everywhere between −4 and −1 and the value −1 is reached exactly on
Lagrangian tangent real 2-planes and the value −4 on tangent complex lines,
viewed as real 2-planes.
Proof. Let v and w be as above and let u be a unit vector such that hv, uix = 0
and w = siv + tu for some s, t in R with s2 + t2 = 1. Then, u and v span
a Lagrangian plane and hence, by lemma 2.2, they are tangent to a totally
geodesic submanifold which is isometric to real hyperbolic plane. Therefore,
by proposition A.1, we have R(v, u)v = −u. In the same way, v and iv
are tangent to a totally geodesic submanifold which is isometric to complex
hyperbolic line and, therefore, by lemma 2.5, we have R(v, iv) = −4iv. Thus,
we have R(v, w)v = −4siv − tu and the lemma follows.
Now we can use these informations to determine the totally geodesic
submanifolds. To do this, we must first complete the information provided
by proposition A.2:
Lemma A.3. Let v and w be unit tangent vectors of HnC at some point x such
that hv, wix = 0. Then we have R(v, iv)w = −2iw and R(v, w)(iv) = −iw.
Proof. The sectional curvature of HnC being known, we just have to use classi-
cal Riemannian geometry machinery. Note that, by proposition A.2, for t in
R, one has R(v + tw, i(v + tw))(v + tw) = −4i(1 + t2 )(v + tw). Identifying the
coefficient of t in this equation, we get R(w, iv)v + R(v, iw)v + R(v, iv)w =
−4iw. By proposition A.2, R(v, iw)v = −iw, thus
R(v, iv)w − R(iv, w)v = −3iw.
Exchanging the roles of v and iv, we have
R(v, iv)w + R(v, w)(iv) = −3iw.

46
Finally, the Bianchi identity for the three vectors v, iv and w reads

R(v, iv)w + R(iv, w)v − R(v, w)(iv) = 0.

Solving this system of equations, we get the result.

We now have the following proposition, which achieves the proof of lemma
2.6:

Proposition A.4. Let M be a totally geodesic real submanifold of HnC and

let x be a point of M . Then the real tangent space of M at x is either a real
Lagrangian subspace or a complex subspace.

The key point is the

Lemma A.5. Let v, w be non-zero tangent vectors of M at x that are R-

independent. Then we are in one of the following mutually excluding situa-
tions:

(i) we have Imhv, wix = 0 and the intersection of Tx M with the C-plane
spanned by v and w is the R-plane spanned by v and w.

(ii) the C-subspace spanned by v and w is contained in Tx M .

Proof. We can suppose that v and w are unitary and that Rehv, wix = 0.
Let V be the tangent space to M at x. We have R(v, w)w ∈ V . Let u be a
vector of v ⊥ such that w = siv + tu, for some s, t in R. Then, by lemma A.3,
we have R(v, w)v = −(4isv, tu).
If s and t are both non-zero, then iv and u both belong to V and, therefore,
as by lemma A.3 iu = − 12 R(v, iv)u, iu belongs to V and we are in the second
case.
If s = 0 (and t = 1), then Rv ⊕ Ru ⊂ W . But, by lemma A.3, R(v, u)
has no stable R-line in Riv ⊕ Riu and, therefore, V = Cv ⊕ Cu, that is we
are in the second case, or W = Rv ⊕ Ru and we are in the first case.
If t = 0 (and s = 1), then Cv ⊂ V and as, by lemma A.3, R(v, iv) has no
stable R-line in Cu, we have V = Cv ⊕ Cu or W = Cv: we are in the second
case.
Proof of proposition A.4. Suppose first there exists two tangent vectors v
and w of M at x that are R-independent and such that the C-subspace V
spanned by v and w is contained in Tx M . Then, if u is a vector of Tx M that

47
does not belong to V , as Cv is contained in Tx M , by lemma A.5, we have
Cu ⊂ Tx M and Tx M is a complex subspace. Suppose now that for every
pair v, w of R-independent vectors in Tx M , the C-subspace they span is not
included in Tx M . Then, by lemma A.5, the space Tx M is Lagrangian, what
should be proved.

A.3 Curvature and submanifolds of quaternionic hy-

perbolic space
From lemma 2.9, we know that the quaternionic hyperbolic line H1Q has
constant sectional curvature −4.
Let x be in Q. Then x is contained in a maximal commutative subfield
of Q. Therefore, if V is a right quaternionic vector space and v and w are
vectors in V , there exists a maximal commutative subfield K of Q and a K-
subspace of V that contains v and w. From this and lemma 2.10, we deduce
the

Proposition A.6. Let v and w be unit tangent vectors of HnQ at some point
x such that Rehv, wix = 0. If hv, wix = 0, we have R(v, w)v = −w. If
w ∈ vQ0 , we have R(v, w)v = −4w. In particular, HnQ has sectional curvature
lying everywhere between −4 and −1 and the value −1 is reached exactly on
Lagrangian tangent real 2-planes and the value −4 on tangent planes that are
K-lines for some maximal commutative subfield K of Q.

Knowing the curvature tensor, we can complete the proof of lemma 2.10
by the

Proposition A.7. Let M be a totally geodesic real submanifold of HnQ and

let x be a point of M . Then the real tangent space of M at x is either
contained in a quaternionic line or in a Lagrangian K-subspace, for some
maximal commutative subfield K, or is a quaternionic subspace.

As in the complex case, this relies on a technical lemma:

Lemma A.8. Let v, w be non-zero tangent vectors of M at x that are R-

independent. Then we are in one of the following mutually excluding situa-
tions:

(i) v and w are contained in the same Q-line.

48
 (ii) v and w are Q-independent, hv, wix ∈ R and the intersection of
Tx M with the Q-plane spanned by v and w is the R-plane spanned by
v and w.
(iii) v and w are Q-independent, and there exists a maximal commu-
tative subfield K 3 hv, wix such that the intersection of Tx M with the
Q-plane spanned by v and w is the K-plane spanned by v and w.
(iv) v and w are Q-independent and Tx M contains the Q-plane spanned
by v and w.
Proof. As usual, we suppose v and w are unit vectors and hv, wix ∈ Q0 .
Choose a unit vector u with hv, uix = 0, s in Q0 and t in R such that
w = vs + ut. Then, by proposition A.6, we have R(v, w)v = −v4s − ut and,
therefore, vs and ut belong to V . Suppose now v and w are Q-independent.
Then t 6= 0 and u belongs to V . If V ∩ vQ ⊕ wQ = vR ⊕ wR, then s = 0,
and we are in the second case. Else, suppose V contains an element of the
form vq + ur for some q and r in Q0 that are not both zero. By lemma A.6,
we have R(vq + ur, v)v = v4q + ur and, therefore, vq and ur belong to V .
By lemma A.3, we have R(ur, u)v = v2r and R(ur, u)vq = v2qr and, thus v,
vq, vr and vqr belong to V . In the same way, u, uq, ur and uqr belong to V .
Hence, either V contains an element of the form uq +vr for some q and r who
does not belong to the same real line of Q0 and, then, uQ ⊕ vQ ⊂ V , or there
exists a maximal commutative subfield K such that uQ ⊕ vQ ∩ V = uK ⊕ vK.
In particular, we then have hv, wix ∈ K and the claim follows.
Proof of proposition A.7. Let the tangent space to M at x not be contained
in a Q-line. Let v be a non-zero vector in Tx M . Then, by lemma A.8, there
exists a subfield K of Q such that vQ ∩ Tx M = vK and that, for every w in
Tx M − vQ, one has hv, wix ∈ K. As this is true for w in vQ ∩ Tx M too, the
result follows.

B Busemann functions of symmetric spaces

Let K be R, C or Q and X = HnK . We shall prove here that the functions of
the form y 7→ e−δX bξ (y,x) on X are harmonic. Let us freely use the notations
of paragraphs 3.2 and 3.3.
Observe that the definition of ∂X given in paragraph 3.2 coincides we the
one of paragraph 4.1, by lemma 3.3. Therefore Busemann functions defined

49
in paragraph 3.2 coincide with the ones of paragraph 4.1. What’s more, note
that, as the distances dx , for x in X, defined in both contexts satisfy the
same homogeneity properties, it suffices to check that they coincide for a
particular choice of x and of two points in the boundary. For this, take x
to be the image of e0 in X, and ξ and η to be the images of the vectors f0
and fn . Then, on one hand, the Gromov product of ξ and η viewed from
x is 0 as x ly in the geodesic from ξ to η, and, on the other hand, we have
hf0 , fn i = kf0 kx = kfn kx = 1: the distances are the same.
We now can prove the
Proposition B.1. For every x in X and ξ in ∂X, the function X → R, y 7→
e−δX bξ (y,x) is harmonic.
Proof. Note that a function f is harmonic if and only if, for every x in X and
r > 0, if S(x, r) denotes the sphere with center x and radius rRand σx,r the
normalized Riemannian measure on this sphere, one has f (x) = S(x,r) f dσx,r .
Let o be our usual base point in X, that is the image of e0 in X. Then, as K
acts transitively on the sphere of the tangent space at o, for any r > 0, one
has S(o, r) = Kar o (where t 7→ at is as in paragraph 3.2) and the measure σo,r
is the unique K-invariant measure on that subset. What’s more, for every
x in X, there exists g in G such that go = x and, thus, S(x, r) = gKar o.
Therefore,
R a function f on X is harmonic if and only if, for any g, h in G, we
have K f (gkho)dk = f (go).
Let us check this equation for the functions we are studying. Let so g, h
be in G and consider the measure µ on ∂X such that, for any continuous
function ϕ on ∂X, we have
Z Z Z
ϕdµ = ϕ(ghkξ)dνo (ξ)dk.
∂X K ∂X

Then this measure is gKg −1 -invariant and therefore it is equal to νgo . In

other words, for any continuous function ϕ on ∂X, we have
Z Z Z Z
−bξ (go,o)
e ϕ(ξ)dν0 (ξ) = ϕdνgo = ϕ(ghkξ)dνo (ξ)dk
∂X ∂X K ∂X
Z Z 
−bξ (gkho,o)
= e dk ϕ(ξ)dνo (ξ),
∂X K
R
that is e−bξ (go,o) = K e−bξ (gkho,o) dk for νo -almost ξ and, hence, for every ξ as
the functions are continuous. The claim follows.

50
References
[1] M. Anderson, R. Schoen, Positive harmonic functions on complete mani-
folds of negative curvature, Annals of mathematics 121 (1985), 429-461.

[2] B. Bekka, M. Mayer, Ergodic theory and topological dynamics of group

actions on homogeneous spaces, London Mathematical Society Lecture
Notes Series 269, Cambridge University Press, Cambridge, 2000.

[3] G. Besson, G. Courtois, S. Gallot, Entropies et rigidités des espaces lo-

calement symétriques de courbure strictement négative, Geometric and
functional analysis 5 (1995), 731-799.

[4] M. Burger, S. Mozes, CAT(−1)-spaces, divergence groups and their com-

mensurators, Journal of the American Mathematical Society 9 (1996),
57-93.

[5] M. Bourdon, Structure conforme au bord et flot géodésique d’un

CAT(−1)-espace, L’enseignement mathématique 41 (1995), 63-102.

[6] P. Eberlein, Geometry of nonpositively curved manifolds, Chicago lectures

in mathematics, Chicago, 1996.

[7] S. Gallot, D. Hulin, J. Lafontaine, Riemannian geometry, Universitext,

Springer Verlag, Berlin, 1993.

[8] E. Ghys, P. de la Harpe (eds), Sur les groupes hyperboliques, d’après

Mikhael Gromov, Progress in mathematics 83, Birkhäuser, Boston, 1990

[9] B. Hasselblatt, A. Katok, Introduction to the modern theory of dynamical

systems, Encyclopedia of mathematics and its applications 54, Cambridge
university press, Cambridge, 1995.

[10] S. Helgason, Differential geometry, Lie groups and symmetric spaces,

Pure and applied mathematics 80, Academic press, San Diego, 1978.

[11] S. Helgason, Groups and geometric analysis, Pure and applied mathe-
matics 113, Academic press, San Diego, 1984.

[12] F. Ledrappier, Harmonic measures and Bowen-Margulis measures, Israel

Journal of Mathematics 71 (1990), 275-287.

51
[13] S.-J. Patterson, The limit set of a fuchsian group, Acta mathematica
136 (1976), 241-273.

[14] Th. Roblin, Ergodicité et équidistribution en courbure négative, Mé-

moires de la SMF 95 (2003).

[15] D. Sullivan, The density at infinity of a discrete group of hyperbolic

motions, Publications mathématiques de l’IHES 50 (1979), 171-202.

[16] R. Zimmer, Ergodic theory and semisimple groups, Birkhäuser, Boston,

1984.

52