ASYMPTOTIC SCHUR ORTHOGONALITY IN HYPERBOLIC GROUPS

WITH APPLICATION TO MONOTONY

ADRIEN BOYER AND ŁUKASZ GARNCAREK

arXiv:1610.06429v2 [math.GR] 19 Nov 2016

A BSTRACT. We prove a generalization of Schur orthogonality relations for certain

classes of representations of Gromov hyperbolic groups. We apply the obtained
results to show that representations of non-abelian free groups associated to the
Patterson-Sullivan measures corresponding to a wide class of invariant metrics
on the group are monotonous in the sense introduced by Kuhn and Steger. This
in particular includes representations associated to harmonic measures of a wide
class of random walks.

1. I NTRODUCTION
Given an irreducible unitary representation π of a group Γ on a Hilbert space H,
one may consider its matrix coefficients, i.e. the functions on the group of the form
g 7→ hπ ( g)v, wi with v, w ∈ H. If Γ is a finite group, then H is finite-dimensional,
and the Schur orthogonality relations tell us that
dim H
|Γ| ∑ hπ ( g)v1, w1 ihπ ( g)v2, w2 i = hv1, v2 ihw1, w2 i.
g∈Γ

In this form, the ortogonality relations do not make sense for infinite groups and
infinite-dimensional representations. However, the sum in the left-hand side of
the above equation could be replaced by some kind of average over either a ball,
or an annulus, and we may consider the limit of the obtained expression as the
radius goes to infinity.
The aim of the paper is to make that statement precise for a certain class of rep-
resentations of Gromov-hyperbolic groups, studied in [16], generalizing the rep-
resentations of fundamental groups of negatively curved manifolds introduced in
[4]. If Γ is a non-elementary Gromov hyperbolic group endowed with an invariant
metric d satisfying some regularity assumptions, the Gromov boundary ∂Γ of Γ can
be endowed with a Patterson-Sullivan measure µ PS corresponding to d. The class of
µ PS is invariant under the action of Γ, yielding a quasi-regular representation π on
L2 (∂Γ, µ PS ), given by
s
dg∗ µ
[π ( g)v](ξ ) = ( ξ ) v ( g −1 ξ ) ,
dµ
which we call the boundary representation associated to the metric d. The main result
of our paper, the Asymptotic Orthogonality Theorem, reads as follows.
Theorem. Let π be a boundary representation of a hyperbolic group Γ corresponding
to the metric d with Patterson-Sullivan measure µ PS . Let A R = { g ∈ Γ : d(1, g) ∈

2010 Mathematics Subject Classification. Primary 20C15, 20F65, 22D10, 22D40 Secondary 22D25,
37A25, 37A30, 37A55.
Key words and phrases. hyperbolic group, boundary representation, Schur orthogonality, Property
RD, Patterson-Sullivan measure, Gibbs measures, equidistribution.
Adrien Boyer was partially supported by ERC Grant 306706. Łukasz Garncarek was partially sup-
ported by Narodowe Centrum Nauki grant 2012/06/A/ST1/00259.
1
2 ADRIEN BOYER AND ŁUKASZ GARNCAREK

[ R − h, R + h]}, where h is sufficiently large. Then there exists a one-parameter family of

probability measures µ R supported in A R , satisfying µ R ( g) ≤ C/ | A R | for some C > 0
independent of R, such that for any v1 , v2 , w1 , w2 ∈ L2 (∂Γ, µ PS ) and f 1 , f 2 ∈ C (Γ ∪ ∂Γ)

µ R ( g ) f 1 ( g ) f 2 ( g −1 )
lim
R→∞
∑ hπ ( g)1∂Γ , 1∂Γ i2
hπ ( g)v1, w1 ihπ ( g)v2, w2 i =
g∈ A R

= h f 2 |∂Γ v1 , v2 ihw1 , f 1 |∂Γ w2 i.

In the case of a convex cocompact group of isometries of a CAT(-1) space with a

non-arithmetic spectrum, we refine this result further, obtaining µ R ( g) = 1/ | A R |.
Moreover, apart from Patterson-Sullivan measures, there is another important fam-
ily of measures, defined on the boundaries of fundamental groups of negatively
curved manifolds, called the Gibbs streams. We prove a variant of the Asymptotic
Orthogonality Theorem also for the quasi-regular representations associated with
these measures, studied in [9]. Finally, by [5], the harmonic measures of a large
family of random walks are in fact Patterson-Sullivan measures corresponding to
metrics satisfying our assumptions, and therefore the Asymptotic Orthogonality
Theorem holds also for the quasi-regular representations corresponding to these
harmonic measures.
As an application of the Asymptotic Orthogonality Theorem, we extend the re-
sults of Kuhn and Steger about monotony of free group representations. In [20],
they propose a method of classifying the unitary representations of a free group
Γ, which are weakly contained in the regular representation, using the notion of
boundary realization. Roughly speaking, a boundary realization of a representation
π on a Hilbert space H is a Γ-equivariant isometric embedding of H into a space
of the form L2 (∂Γ, µ), acted upon by the quasi-regular representation πµ corre-
sponding to some measure µ, in such a way that the image is a cyclic subspace
for the multiplication representation of C (∂Γ). If this embedding is surjective, the
realization is said to be perfect.
If a representation π of a free group is weakly contained in the regular repre-
sentation, then it admits a boundary realization, and Kuhn and Steger distinguish
three different behaviors of the class of boundary realizations of π, inspired by
the case of restrictions of irreducible unitary representations of SL2 (R ) to a free
lattice [20, Section 5]. They say that π is monotonous if it admits a unique boundary
realization which is perfect, odd if it admits a unique boundary realization which
is imperfect, and duplicitous if it admits exactly two perfect boundary realizations,
and any imperfect realization is a combination of these two, in a certain precise
sense. Then they conjecture the following.

Conjecture. A unitary representation of a free group, weakly contained in the regular

representation, is either monotonous, duplicitous, or odd.

To motivate the conjecture, they prove that a certain class of representations of

free groups, associated to some explicitly constructed measures on their bound-
aries, are monotonous. As free groups are hyperbolic, the construction of [16] pro-
vides them with a supply of boundary representations, and we manage to show
the following.

Theorem. Boundary representations of free groups are monotonous.

Our techniques combine functional analytic tools, namely property RD [12],

with equidistribution results in the spirit of Margulis [21] and Roblin [24], which
we prove in the general context of Gromov hyperbolic groups.
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 3

1.1. Organization. The paper is organized as follows. In Section 2 we discuss all

the preliminaries, explain the necessary facts involving hyperbolic groups, their
boundaries and Patterson-Sullivan measures. We also discuss a slightly stronger
variant of property RD satisfied by hyperbolic groups. Section 3 deals with the
proof of the Asymptotic Orthogonality Theorem. Sections 4 and 5 deal with the
variants of the main theorem for convex cocompact groups of isometries of CAT(-
1) spaces, and representations associated to Gibbs streams, respectively. Finally, in
Section 6 we present the application of our results to the phenomenon of monotony
of representations of free groups.
1.2. Acknowledgments. We would like to thank Uri Bader for numerous helpful
discussions and Christophe Pittet for his interest to this work. We are grateful to
Gabriella Kuhn and Tim Steger for several explanations concerning the concepts of
monotony, duplicity and oddity. Finally, we would like to thank Nigel Higson for
informing us about Kuhn and Steger’s conjecture of classifications of irreducible
unitary representations of the free group.

2. T HE SETTING

We begin by briefly discussing the setting of hyperbolic groups, the Gromov

boundary and the Patterson-Sullivan measure in the generality of roughly geo-
desic metrics. A more detailed explanation of this setting can be found in the ar-
ticle [16], where boundary representations of hyperbolic groups are defined, and
their irreducibility and classification are studied.
2.1. Conventions for writing estimates. Throughout the article we will follow
some conventions regarding various kinds of estimates, allowing us to hide the
constants whose values are inessential.
First of all, we will always consider a group Γ endowed with a metric d from
a certain class, and we will treat both these objects as fixed. In particular, if we
refer to some quantity as constant, we still allow it to depend on the choice of Γ
and d—but not on the other parameters, unless explicitly stated otherwise.
We will write
(2.1) a >x,y,... b
to indicate that there exists C = C ( x, y, . . .), possibly dependent on the parameters
x, y, . . . (and also on Γ and d, as already mentioned) such that
(2.2) a ≤ b + C.
If both a >x,y,... b and a ? x,y,... b hold, we write a ≈ x,y,... b. Analogous notation will
be used for multiplicative estimates; we write
(2.3) a ≺ x,y,... b
if a ≤ Cb for some C > 0, possibly depending on x, y, . . ., and if both a ≺ x,y,... b,
and a ≻ x,y,... b hold, then we use the symbol a ≍ x,y,... b.
2.2. Hyperbolicity. A metric space ( X, d) is said to be Gromov hyperbolic, or hyper-
bolic for short, if for any x, y, z ∈ X and some/any1 basepoint o ∈ X one has
(2.4) ( x, y)o ? min{( x, z)o , (z, y)o },
where ( x, y)o stands for the Gromov product of x and y with respect to o, that is
1
(2.5) ( x, y)o = (d( x, o ) + d(y, o ) − d( x, y)).
2
1 if the condition holds for some o and δ, then it holds for any o and 2δ
4 ADRIEN BOYER AND ŁUKASZ GARNCAREK

If we restrict to the class of geodesic metric spaces, hyperbolicity becomes invari-

ant under quasi-isometries, which is not the case for arbitrary metric spaces [5]. In
particular, if Γ is a finitely generated group, all its Cayley graphs associated with
finite generating sets are pairwise quasi-isometric geodesic metric spaces, so either
none or all of them are hyperbolic. In the latter case we say that Γ is a hyperbolic
group.
Let Γ be a hyperbolic group. As the basepoint we take 1 ∈ Γ and omit it from
notation of Gromov product. We will also write | g| for d(1, g). We denote by D(Γ)
the class of all metrics d on Γ satisfying the following three conditions:
(1) d is quasi-isometric to a word metric d S on Γ, i.e. there exist C ≥ 0 and
L ≥ 1 such that
1
(2.6) d ( g, h) − C ≤ d( g, h) ≤ LdS ( g, h) + C,
L S
(2) d is left-invariant, i.e. d(kg, kh) = d( g, h),
(3) d is hyperbolic.
By the discussion in [16, Section 3.1], for any d ∈ D(Γ) the metric space (Γ, d) is
roughly geodesic. This means that for any x, y ∈ Γ there exists a roughly geodesic
segment joining x and y, i.e. a map γ : [0, ℓ] → Γ, such that
(2.7) d(γ(s), γ(t)) ≈ |s − t| , γ(0) = x, γ(ℓ) = y,
where the estimate constant depends only on d ∈ D(Γ). Rough geodesics and
roughly geodesic rays are defined by a similar condition. When speaking about
roughly geodesic segments/rays/lines we will always assume that they satisfy
the uniform estimate (2.7) with the fixed constant depending only on d.
2.3. The Gromov boundary. Let ( X, d) be a roughly geodesic hyperbolic metric
space. Two roughly geodesic rays γ1 , γ2 : [0, ∞) → X are asymptotic if the distance
d(γ1 (t), γ2 (t)) is bounded. The quotient of the set of all roughly geodesic rays in
X by the asymptoticity relation relation is called the Gromov boundary of X and
denoted by ∂X. It carries a family of visual metrics, depending on d and a real
parameter ǫ > 0. We say that dǫ is a visual metric on ∂X if
(2.8) dǫ (ξ, η ) ≍ e−ǫ( ξ,η ) ,
where the extension of the Gromov product to X ∪ ∂X is defined as
(2.9) (ξ, η ) = inf{lim inf(γ1 (t), γ2(t)) : ξ = [γ1 ], η = [γ2 ]},
t→∞
assuming that elements of X are represented by corresponding constant functions.
It is important to note that the diameter of set under the infimum operator is
bounded by some constant, so if we are interested in the value of the Gromov
product only up to an additive constant, instead of infimum we may just take the
limit over some arbitrarily chosen representatives [10, Remark 3.17]. Such a metric
always exists for sufficiently small ǫ. Moreover, as a topological space, the Gromov
boundary does not depend on the choice of ǫ, and the metric d ∈ D(Γ).
Finally, if for xn ∈ X, and ξ ∈ ∂X we define lim xn = ξ if and only if lim( xn , ξ ) =
∞, then X = X ∪ ∂X becomes a compactification of X.
Now, let Γ be a hyperbolic group, and let d ∈ D(Γ). By the above discussion,
this allows us to endow Γ with its boundary ∂Γ, independent from d as a topolog-
ical space, and a visual metric on ∂Γ. The action of Γ on itself by left translations is
isometric, so it extends to an action on the set of roughly geodesic rays, compatible
with the relation of asymptoticity, and descends to the quotient ∂Γ. The action of
Γ on ∂Γ satisfies
(2.10) |( gξ, gη ) − (ξ, η )| ≤ Cg
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 5

for some constant Cg depending only on g, and therefore Γ acts on (∂Γ, d ǫ ) contin-
uously, by bi-Lipschitz maps.
2.4. The Patterson-Sullivan measure. The space (∂Γ, d ǫ ) is a compact metric space,
and therefore it admits a Hausdorff measure of dimension D. It is nonzero, finite,
and finitely-dimensional, and in the context of hyperbolic geometry it is known as
the Patterson-Sullivan measure. We will denote it by µ PS , and normalize it so that
µ PS (∂Γ) = 1. Usually all measures of the form ρdµ PS with 0 < c ≤ ρ(ξ ) ≤ C < ∞
for some c, C are called Patterson-Sullivan measures. This class of measures is in-
dependent of the choice of ǫ, but different metrics d ∈ D(Γ) usually give rise to
mutually singular measures. Moreover, the Hausdorff dimension changes with ǫ,
although the relationship is very simple, namely
(2.11) eǫD = ω,
where ω is defined by
(2.12) ω = lim |{ g ∈ Γ : | g| ≤ R}|1/R .
R→∞
Since Γ acts on its boundary by bi-Lipschitz maps, the Patterson-Sullivan mea-
sure µ PS is quasi-invariant. It actually satisfies a stronger condition of quasi-con-
formality, namely
dg∗ µ PS
(2.13) (ξ ) ≍ ω 2( g,ξ )−|g| .
dµ PS
It is also Ahlfors regular of dimension D, i.e. for r ≤ diam ∂Γ we have the follow-
ing estimate for the volumes of balls:
(2.14) µ PS ( B∂Γ (r, ξ )) ≍ r D .
Finally, the Patterson-Sullivan meaure is ergodic for the action of Γ.
The theory of Patterson-Sullivan measures for geodesic hyperbolic spaces was
developed in [13], and it was extended to the roughly geodesic case in [5].
2.5. Boundary representations of hyperbolic groups. Given a measure space ( X, µ)
with a measure class preserving action of a group G, we may define a unitary rep-
resentation πµ of G on the Hilbert space L2 ( X, µ) with the formula
 1/2
dg∗ µ
(2.15) [πµ ( g)v]( x ) = (x) v ( g −1 x )
dµ
for g ∈ G, v ∈ L2 ( X, µ), and x ∈ X. It is often referred to as the quasi-regular
representation or Koopman representation. Its unitary equivalence class depends only
on the equivalence class of µ. If we apply this construction to the action of Γ on
(∂Γ, µ PS ), where µ PS is the Patterson-Sullivan measure corresponding to a metric
d ∈ D(Γ), we obtain the boundary representation of Γ corresponding to the metric d.
In [16] irreducibility and equivalence of boundary representations of hyperbolic
groups were studied, leading to the following results.
Theorem 2.1 ([16, Theorem 6.2 and 7.4]). For any hyperbolic group Γ which is not
virtually cyclic, and any metric d ∈ D(Γ), the corresponding boundary representation is
irreducible. Moreover, if d, d ′ ∈ D(Γ) are two metrics giving rise to Patterson-Sullivan
measures µ PS and µ′PS , and boundary representations π and π ′ , then the following condi-
tions are equivalent.
(1) the metrics d and d′ are roughly similar, i.e. there exists a constant α > 0 such
that the difference |d − αd′ | is uniformly bounded,
(2) the Patterson-Sullivan measures µ PS and µ′PS are equivalent,
(3) the boundary representations π and π ′ are unitarily equivalent.
6 ADRIEN BOYER AND ŁUKASZ GARNCAREK

2.6. Shadows. Let Γ be a hyperbolic group endowed with a metric d ∈ D(Γ). We

will now introduce the setting used in [16]. First, for any g ∈ Γ we may choose
ĝ ∈ ∂Γ such that ( g, ĝ) ≈ | g|, where the symbol ≈ means that the difference of
both sides is bounded by a constant (depending only on the metric group (Γ, d)).
This defines an ‘approximate’ projection of Γ onto its boundary, which despite the
arbitrary choices in its definition, turns out to be very nice in the topological sense,
as the next lemma explains.
Lemma 2.2. The map p : Γ → ∂Γ given by
(
x̂ if x ∈ Γ,
(2.16) p( x ) =
x if x ∈ ∂Γ
is a continuous retraction.
Proof. Since Γ is discrete, p is clearly continuous at every point of Γ, so we just
need to check continuity at every ξ ∈ ∂Γ. Let x n ⊆ Γ be a sequence converging to
ξ. We have, for xn ∈ Γ,
(2.17) ( p( xn ), ξ ) ? min{( p( xn ), xn ), ( xn , ξ )} ? min{| xn | , ( xn , ξ )} = ( xn , ξ ).
But xn → ξ is equivalent to ( xn , ξ ) → ∞, which implies ( p( xn ), ξ ) → ∞, i.e.
p( xn ) → ξ. For x n ∈ ∂Γ we have p( xn ) = xn , so the convergence is clear. 
Let us now denote ǧ = gd −1 . In [16] the double shadows Σ2 ( g, ρ ) ⊆ ∂Γ × ∂Γ of
elements of Γ were introduced. They are neighborhoods of ( ĝ, ǧ) defined as
   
(2.18) Σ2 ( g, ρ) = B∂Γ ĝ, e−ǫ(|g|/2−ρ) × B∂Γ ǧ, e−ǫ(|g|/2−ρ) .
If for R, h ≥ 0 we introduce the annulus A R,h of radius R and thickness 2h (where
d refers to the metric used, and will be omitted from notation whenever a metric
is fixed by the context) defined by
(2.19) A R,h = { g ∈ Γ : | g| ∈ [ R − h, R + h]},
we can formulate following fundamental result.
Proposition 2.3 ([16, Proposition 4.5]). For sufficiently large ρ and h the family of
double shadows
(2.20) {Σ2 ( g, ρ) : g ∈ A R,h }
is a cover of ∂Γ × ∂Γ for all R > 0.
Henceforth, we will fix ρ and h sufficiently large for this result to hold, and such
that the estimate
 
(2.21)  A R,h  ≍ ω R
is satisfied. We will omit them from notation, by writing A R , and Σ2 ( g).
2.7. Property RD. Let G be a finitely generated group. Denote by | x | the word-
length of x ∈ G corresponding to a fixed generating set. For f ∈ C [ G ] denote
(2.22) r ( f ) = max{| x | : f ( x ) 6= 0}.
This quantity measures how far the support of f is from the unit element of G.
The multiplication in the group algebra C [ G ] extends to an action of C [ G ] on its
completion ℓ2 ( G ) by convolution. The standard definition of the property of rapid
decay, abbreviated to ‘property RD’ reads as follows. We say that G has property
RD if there exists a polynomial P with real coefficients, such that for any f ∈ C [ G ]
and g ∈ ℓ2 ( G ) we have
(2.23) k f ∗ gk2 ≤ P(r ( f )) k f k2 k gk2 .
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 7

Clearly, this does not depend on the choice of the generating set of G, and the
polynomial P can be chosen to be of the form P(t) = C (1 + t)n for some C > 0.
Property RD can be reformulated in terms of matrix coefficients of unitary rep-
resentations of G. First, we say that a unitary representation π : G → U (H) is
weakly contained in the regular representation, if for all f ∈ C [ G ] we have
(2.24) kπ ( f )kop ≤ k f kop ,
where π is uniquely extended to C [ G ] by linearity, and on the right we treat f
as its corresponding convolution operator, i.e. λ( f ), where λ is the left regular
representation. Now, a finitely generated group G has property RD if and only
if there exist constants C, s > 0 such that for every representation π : G → U (H)
weakly contained in the regular representation, and for every ξ, η ∈ H the estimate
!1/2
|hπ ( x )ξ, η i|2
(2.25) ∑ (1 + |x |)s ≤ C k ξ k kη k
x∈G
is satisfied.
The crucial fact we will be relying on is the following strengthening of the stan-
dard property RD for hyperbolic groups, where thanks to restricting to the an-
nulus, one can decrease the exponent (without this restriction, hyperbolic groups
satisfy (2.25) with exponent s = 3).
Proposition 2.4. Let Γ be a hyperbolic group endowed with a metric d ∈ D(Γ), and let
π be a unitary representation of Γ on a Hilbert space H, weakly contained in the regular
representation. Then for v, w ∈ H the estimate
 1/2
(2.26)  ∑ |hπ ( g)v, wi|2  ≺h (1 + R) kv k kw k
g ∈ A R,h

holds uniformly in R.
In a less general setting of a group acting on a negatively curved manifold, this
result appears in [17, Section 3.2], and the proof presented therein requires only
some technical adjustments to work for a general hyperbolic group Γ equipped
with a metric d ∈ D(Γ). We will include the adapted proof for the sake of self-
containment. Note that a proof of the fact that hyperbolic groups satisfy property
RD can be fund in [15].
Before we proceed to the proof, note that since by [1] the action of Γ on ∂Γ is
amenable, and by [18], ergodic amenable actions lead to quasi-regular represen-
tations which are weakly contained in the regular representation, Proposition 2.4
applies to the matrix coefficients of boundary representations.
We begin the proof of Proposition 2.4 with several technical lemmas.
Lemma 2.5. Let g ∈ Γ. Then
(2.27) diam{ x ∈ A R,h : x −1 g ∈ A R′ ,h′ } >h,h′ max{0, R + R′ − | g|}.
In particular, for fixed h and h′ the cardinality of this set is bounded uniformly in R +
R ′ − | g |.
Proof. Assume that the considered set is non-empty, and contains two elements
x1 , x2 . We have
1
(2.28) ( x1 , x2 ) ? min{( x1 , g), ( x2, g)} ≈h,h′ ( R + | g| − R′ ),
2
and therefore
(2.29) d( x1, x2 ) = | x1 | + | x2 | − 2( x1 , x2 ) ≈h,h′ R + R′ − | g| . 
8 ADRIEN BOYER AND ŁUKASZ GARNCAREK

Lemma 2.6. Suppose that φ, ψ ∈ C [Γ] are supported in A R,h and A R′ ,h′ , respectively.
Then for g ∈ Γ we have
(2.30) |φ ∗ ψ( g)|2 ≺h,h′ ,R+ R′ −| g| |φ|2 ∗ |ψ|2 ( g).
Proof. We have
(2.31) φ ∗ ψ ( g ) = ∑ φ ( x ) ψ ( x −1 g ) ,
x

where x runs through the set { x ∈ A R,h : x −1 g ∈ A R′ ,h′ }, whose cardinality is, by
Lemma 2.5, bounded by some C depending only on h, h′ , and R + R′ − | g|. Hence,
(2.32) |φ ∗ ψ( g)|2 ≤ C ∑ | φ( x )|2 |ψ( x −1 g)|2 = C |φ|2 ∗ |ψ|2 ( g). 
x

Lemma 2.7. Let g ∈ A R,h A R′ ,h′ = {kk′ : k ∈ A R,h , k′ ∈ A R′ ,h′ } for some R, R′ , h, h′ ≥
0, and put p = ( R + R′ − | g|)/2. Then there exist u g , v g ∈ Γ such that
(1) u g v g = g,
(2) |u g | ≈h,h′ R − p, and |v g | ≈h,h′ R′ − p,
(3) for any decomposition g = uv with u ∈ A R,h , and v ∈ A R′ ,h′ , we have |u− 1
g u| =
|vv− 1
g | ≈h,h ′ p.
′
  we have | g| ?h,h′ R − R, so R − p ?h,h′ 0. We
Proof. By the triangle inequality
 
may thus pick u g satisfying u g ≈h,h′ R − p on some roughly geodesic segment
from 1 to g, and define v g = u− 1
g g. Then
   
(2.33) v g  ≈ | g| − u g  ≈h,h′ R′ − p.
To verify condition (3), choose a decomposition g = uv with u ∈ A R,h , and v ∈
A R′ ,h′ , and observe that u− 1 −1 −1 −1
g u = ( vv g ) , so we have to estimate only | u g u|. We
have
 
(2.34) u g  ≥ (u g , u) ? min{(u g , g), ( g, u)},
where
 
(2.35) (u g , g) ≈ u g  ≈h,h′ R − p,
and
1
(2.36) ( g, u) = (| g| + |u| − d( g, u)) ≈h,h′ R − p,
2
so
(2.37) (u g , u) ≈h,h′ R − p.
It follows that
(2.38) |u− 1
g u| = | u g | + | u| − 2( u g , u) ≈h,h ′ p. 
Lemma 2.8. Suppose that φ, ψ ∈ C [Γ] are supported in A R,h and A R′ ,h′ , respectively.
Then, the restriction of their convolution to the annulus A R′′ ,h′′ satisfies the estimate


(2.39) φ ∗ ψ| A R′′ ,h′′ ≺h,h′ ,h′′ k φk2 kψk2 .
2

Proof. Pick g ∈ A R′′ ,h′′ such that φ ∗ ψ( g) 6= 0; in particular, g ∈ A R,h A R′ ,h′ . Let
p = ( R + R′ − R′′ )/2, and let g = u g v g be the decomposition from Lemma 2.7.
Then there exists a constant κ ≥ 0 depending only on h, h′ and h′′ such that
 
 
(2.40) |φ ∗ ψ( g)| ≤ ∑ |φ(u)ψ(v)| ≤ ∑ φ(u g k)ψ(k−1 v g ) ≤ φ p (u g )ψ p (v g ),
uv = g k ∈ A p,κ
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 9

where
 1/2  1/2
 2
 
(2.41) φ p ( u) =  ∑ |φ(uk)|2  and ψp (v) =  ∑ ψ(k−1 v) 
k ∈ A p,κ k ∈ A p,κ

are defined on the annuli of the form A R− p,r and A R′ − p,r ′ respectively, where r and
r ′ depend only on h, h′ , h′′ . Therefore, we have
2
2
φ ∗ ψ| A R′′ ,h′′ = ∑ |φ ∗ ψ( g)| ≤ ∑ φ p (u g )2 ψ p (v g )2 ≤
2
g ∈ A R′′,h ′′ g ∈ A R′′,h ′′
(2.42)
2
≤ ∑ (φ p ∗ ψ p ( g)) .
g ∈ A R′′,h ′′

Since R − p + R′ − p ≈h,h′ ,h′′ R′′ , by Lemma 2.6 we have

2 2
(2.43) ∑ (φ p ∗ ψ p ( g))2 ≺h,h′ ∑ φ2p ∗ ψ2p ( g) = φ p 2 ψ p 2 .
g ∈ A R′′,h ′′ g ∈ A R′′,h ′′

But by Lemma 2.5

2
(2.44) φp =
2 ∑ ∑ |φ(uh)|2 ≺r,κ kφk22 ,
u ∈ A R− p,r h ∈ A p,κ

and similarly with ψ p , thus finishing the proof. 

Using the lemmas above, we may prove the following estimate.

Proposition 2.9. Let Γ be a hyperbolic group endowed with a metric d ∈ D(Γ), and let
R, h > 0. Then for φ ∈ C [Γ] supported in A R,h and ψ ∈ ℓ2 (Γ) we have
(2.45) k φ ∗ ψ k2 ≺ h ( 1 + R ) kφ k 2 k ψ k2
Proof. Without loss of generality we may assume that φ and ψ take positive real
values. If we denote by χn the characteristic function of the annulus { g ∈ Γ : | g| ∈
[n, n + 1)}, and ψn = ψχn , then for n ∈ Im = [m − R − h − 1, m + R + h] we have
by Lemma 2.8
(2.46) k(φ ∗ ψn )χm k2 ≺h k φk2 k ψn k2 ,
and for n outside this interval the corresponding norm is 0. Hence
k(φ ∗ ψ)χm k2 ≤ ∑ k(φ ∗ ψn )χm k ≺h ∑ k φ k 2 k ψn k 2 ≺ h
n ∈ Im n ∈ Im
(2.47) !1/2
≺h (1 + R) 1/2
k φ k2 ∑ kψn k22 ,
n ∈ Im

and therefore
kφ ∗ ψk22 = ∑ k(φ ∗ ψ)χm k22 ≺h (1 + R) kφk2 ∑ ∑ k ψn k22 ≺h
m m n ∈ Im

≺h (1 + R) kφ k2 ∑ 2
k ψn k22 = (1 + R)2 kφk22 kψk22 . 
n

The proof of Proposition 2.4 now becomes a simple calculation.

Proof of Proposition 2.4. Denote φ( g) = hπ ( g)v, wi for g ∈ A R,h , and φ( g) = 0

otherwise. The expression we are to estimate is then the ℓ2 -norm of φ. We get,
10 ADRIEN BOYER AND ŁUKASZ GARNCAREK

using Proposition 2.9

kφk22 = ∑ φ( g)hπ ( g)v, wi = hπ (φ)u, vi ≤

(2.48) g∈Γ
≤ kπ (φ)kop kvk kwk ≺h (1 + R) kφk2 k vk k wk ,
and consequently k φk2 ≺h (1 + R) kvk k wk. 

3. E QUIDISTRIBUTION AND ASYMPTOTIC S CHUR ORTHOGONALITY

Let Γ be a hyperbolic group endowed with a metric d ∈ D(Γ), and the corre-
sponding visual metric dǫ on the boundary ∂Γ, where ǫ > 0 is fixed. Denote by
µ PS the associated Patterson-Sullivan measure, and let H = L2 (∂Γ, µ PS ).
In this section we first deal with the distribution of the points of the form ( ĝ, ǧ)
in ∂Γ × ∂Γ. We show that in a certain sense, they are approximately uniformly
distributed. Namely, to the points ( ĝ, ǧ) with g ∈ A R , we may assign weights in a
controlled manner, in such a way that the weighted average of the corresponding
Dirac masses gives a good approximation of µ PS ⊗ µ PS .
Using this equidistribution result, we prove that the matrix coefficients of a
boundary representation of Γ satisfy certain asymptotic Schur orthogonality relations
resembling Schur orthogonality relations.

3.1. Equidistribution. Define the maps J : Γ → Γ × Γ and J ′ : Γ → ∂Γ × ∂Γ by

(3.1) J ( g) = ( g, g−1) and J ′ ( g) = ( ĝ, ǧ).
By Lemma 2.2 these maps are both continuous.
We begin by showing that two ways of approximating a measure of the form
µ ⊗ µ on ∂Γ × ∂Γ are equivalent.
Lemma 3.1. Let {µ R : R ≥ 0} be a one-parameter family of probability measures on Γ,
and let µ ∈ Prob(∂Γ) be atom-free. Then the following conditions are equivalent
(1) in the space C (Γ × Γ )∗ endowed with the weak-* topology we have
(3.2) lim J∗ µ R = µ ⊗ µ.
R→∞

(2) in the space C (∂Γ × ∂Γ)∗ endowed with the weak-* topology we have
(3.3) lim J∗′ µ R = µ ⊗ µ.
R→∞

Proof. First, assume (1), and let F ∈ C (∂Γ × ∂Γ). Define Fe ∈ C (Γ × Γ ) using the
retraction p : Γ → ∂Γ from Lemma 2.2, as Fe( x, y) = F ( p( x ), p(y)). By (1), and the
fact that p is identity on ∂Γ, we have
Z Z Z
(3.4) lim F dJ∗′ µ R = lim Fe dJ∗ µ R = F (ξ, η ) dµ(ξ )dµ(η ).
R→∞ R→∞

Let us now prove the other implication. Assume (2), and take F ∈ C (Γ × Γ).
Define Fe in the same manner as before. We then have
Z Z  Z
 
 ′  e
(3.5)  F dJ∗ µ R − F dJ∗ µ R  ≤ | F − F | dJ∗ µ R ,

but G = | F − Fe| vanishes on ∂Γ × ∂Γ, and we may estimate the above integral
separately on some large ball B = BΓ (1, r ) ⊆ Γ, and its complement, by writing
Z Z Z
(3.6) G dJ∗ µ R = G ◦ J dµ R + G ◦ J dµ R .
B Γ\ B
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 11

The first term can be bounded by

Z
(3.7) G ◦ J dµ R ≤ k G k∞ µ R ( B).
B
To see that limR→∞ µ R ( B) = 0, observe that since µ is atom-free, and Γ is a compact
Hausdorff space, for any λ > 0 there exists a continuous function H : Γ × Γ → [0, 1]
such that H ( ĝ, ǧ) = 1 for g ∈ B, and
Z
(3.8) H dµ2 < λ.

Then
Z Z
(3.9) µR ( B) = H ( ĝ, ǧ) dµ R ( g) ≤ H dJ∗′ µ R ,
B
and therefore
Z Z
(3.10) lim µ R ( B) ≤ lim H dJ∗′ µ R ( g) = H dµ2 < λ
R→∞ R→∞
for any ball B ⊆ Γ.
To bound the second term in (3.6), observe that for any λ > 0 the set
(3.11) Uλ = {( x, y) ∈ Γ × Γ : G ( x, y) < λ}
is an open neighborhood of ∂Γ × ∂Γ, and since ∂Γ × ∂Γ can be written as an inter-
section of closed sets
\

(3.12) ∂Γ × ∂Γ = (Γ \ BΓ (1, r )) × (Γ \ BΓ (1, r )) ,
r >0
by compactness every Uλ contains one of these sets. This means that for suffi-
ciently large r
Z
(3.13) G ◦ J dµ R < λµ R (Γ \ B) ≤ λ.
Γ\ B
This gives estimates in (3.6), and proves that (2) implies (1). 
Now, we may proceed to the equidistribution theorem, asserting that for a hy-
perbolic group, the product measure µ PS ⊗ µ PS can be approximated in the way
described in Lemma 3.1 by a reasonably well-behaved family of measures on Γ.
Theorem 3.2. For any hyperbolic group Γ endowed with a metric d ∈ D(Γ), there exists
a family of measures {µ R : R ≥ 0} ⊆ Prob(Γ) such that
(1) lim R→∞ J∗ µ R = µ PS ⊗ µ PS in the weak-* topology on C (Γ × Γ)∗
(2) there exists h > 0 such that every µ R is supported in the annulus A R,h ,
(3) the estimate µ R ({ g}) ≺ ω − R is satisfied uniformly in R.
Notice, that conditions (2) and (3) above are as close as one can get to uniform
distribution on the annulus. Indeed, in the context of general hyperbolic groups,
from the very beginning all equalities and inequalities are approximate.
Proof of Theorem 3.2. The construction of measures µ R will proceed analogously to
the proof of [16, Proposition 5.4]. Using Proposition 2.3, we may take ρ and h
sufficiently large for the double shadows Σ2 ( g, ρ) with g ∈ A R,h to cover ∂Γ × ∂Γ.
Fix R and define a family { Eg : g ∈ A R,h } of subsets of ∂Γ × ∂Γ by putting an
arbitrary linear order on A R,h , and taking inductively
[
(3.14) E g = Σ2 ( g ) \ Eh ⊆ Σ 2 ( g ) .
h<g

These sets form a measurable partition of ∂Γ × ∂Γ, and we may put

(3.15) µ R ({ g}) = µ2PS ( Eg ) ≤ µ2 (Σ2 ( g)).
12 ADRIEN BOYER AND ŁUKASZ GARNCAREK

Since µ PS is normalized to be a probability measure, conditions (2) and (3) are

satisfied.
It remains to verify condition (1), or equivalently, condition (2) of Lemma 3.1.
Take F ∈ C (∂Γ × ∂Γ). The product ∂Γ × ∂Γ is a compact metric space when en-
dowed with the ℓ∞ -product d2ǫ of the visual metrics dǫ on the factors, and there-
fore F is uniformly continuous, and admits a modulus of continuity, i.e. a non-
decreasing continuous function m : [0, ∞) → [0, ∞) such that m(0) = 0, and
 
(3.16)  F (ξ, η ) − F (ξ ′ , η ′ ) ≤ max{m(d ǫ (ξ, ξ ′ )), m(dǫ (η, η ′ ))}.

We now have
Z Z Z
(3.17) F dJ∗′ µ R = F ◦ J ′ dµ R = ∑ F ( ĝ, ǧ)µ R ({ g}) = ∑ F ( ĝ, ǧ) dµ2PS ,
Γ Γ g∈ A R g ∈ A R Eg

and since for g ∈ A R and ζ, ζ ′ ∈ Eg ⊆ Σ2 ( g, ρ) we have d2ǫ (ζ, ζ ′ ) ≤ Ce−ǫR/2 for

some constant C, we finally get
Z Z 
 
 F dµ2PS − F dJ∗′ µ R  ≤
∂Γ × ∂Γ Γ
Z
≤ ∑ | F (ξ, η ) − F ( ĝ, ǧ)| dµ2PS (ξ, η ) ≤
g∈ A R Eg

≤ ∑ µ2PS ( Eg )m(Ce−ǫR/2 ) ≺ | A R |−1 ∑ m(Ce−ǫR/2 ) =

g∈ A R g∈ A R
−ǫR/2
= m(Ce ) −−−→ 0. 
R→∞

3.2. Asymptotic Schur orthogonality relations. Let us stop for a moment and
recall the classical Schur orthogonality relations. Let G be a finite group, and let
π be its irreducible representation on a finite-dimensional Hilbert space V. The
matrix coefficients of π are functions Duv ∈ C [ G ], indexed by u, v ∈ V, defined by
(3.18) Duv ( g) = hπ ( g)u, viV .
As G is finite, C [ G ] is naturally a Hilbert space with the normalized inner product
1
(3.19) hφ, ψi =
|G| ∑ φ( g)ψ( g).
g∈ G

With this notation, Schur orthogonality (for a single representation) can be rephrased
as the identity

(3.20) dim V · h Duv, Du′ v′ i = hu, u′ ihv, v′ i.

Now, let us return to our original setting. Let π : Γ → U (H) be the boundary
representation of Γ associated to the metric d ∈ D(Γ). Also, define
1
(3.21) e( g) =
π π ( g ).
hπ ( g)1, 1i
The matrix coefficient
(3.22) Ξ( g) = hπ ( g)1, 1i ∈ R +
appearing in the above normalization is often referred to as the Harish-Chandra
function of the representation π. Recall, that by [16, Lemma 5.1] it can be estimated
by

(3.23) Ξ( g) ≍ ω −|g|/2 (1 + | g|)

 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 13

uniformly in g ∈ Γ. We clearly have πe ( g −1 ) = π

e ( g)∗ . If we denote, following [16],
 1/2
e 1 dg∗ µ
(3.24) Pg (ξ ) = (ξ ) ,
Ξ( g) dµ
then
(3.25) [π eg (ξ )v( g−1ξ ).
e ( g)v](ξ ) = P
With this notation, we can formulate our Asymptotic Orthogonality Theorem
as follows.
Theorem 3.3 (Asymptotic Orthogonality). Let {µ R : R > 0} be a family of probabil-
ity measures on Γ, satisfying conditions (1)-(3) of Theorem 3.2. Then for all continuous
functions f 1 , f 2 ∈ C (Γ), and vectors v1 , v2 , w1, w2 ∈ L2 (∂Γ, µ) we have
Z
(3.26) lim f 1 ( g) f 2 ( g−1 )hπ e( g)v2, w2 i dµ R ( g) =
e( g)v1, w1 ihπ
R→∞
= h f 2 |∂Γ v1 , v2 ihw1 , f 1 |∂Γ w2 i.
To see how it relates to the Schur orthogonality relations, let us specify to the
case where f 1 = f 2 = 1Γ . If we forget for a moment about the limit, then the left-
hand side is an analogue of the scalar product of matrix coefficients of π restricted
to an annulus, with some weights added. By passing to the limit, we get an as-
ymptotic version of the orthogonality relation between the matrix coefficients.
The proof of Theorem 3.3 will be divided into two lemmas. But first, we will
introduce some auxiliary notation. Fix f 1 , f 2 ∈ C (Γ), and denote the expression
under the limit in the left-hand side of (3.26) as
Z
(3.27) Φ R ( v 1 , v 2 , w1 , w2 ) = f 1 ( g) f 2 ( g−1 )hπ e( g)v2, w2 i dµ R ( g).
e( g)v1, w1 ihπ

It can be regarded as a quadrilinear form on H × H ∗ × H ∗ × H.

Lemma 3.4. For any f 1 , f 2 ∈ C (∂Γ) the corresponding quadrilinear forms Φ R are uni-
formly bounded with respect to the norm
(3.28) kΦk = inf{C > 0 : Φ(v1 , v2 , w1 , w2 ) ≤ C kv1 k k v2 k kw1 k kw2 k}.
Proof. Without loss of generality we may assume that k f i k∞ ≤ 1. Then for v1 , v2 , w1 , w2 ∈
H we get
Z
|Φ R (v1 , v2 , w1 , w2 )| ≤ e( g)v1, w1 ihπ
|hπ e( g)v2, w2 i| dµ R ( g) =
µR ( g)
(3.29)
= ∑ Ξ ( g )2
|hπ ( g)v1, w1 ihπ ( g)v2, w2 i| ≺
g∈ A R
1
≺
(1 + R)2 ∑ |hπ ( g)v1, w1 ihπ ( g)v2, w2 i| .
g∈ A R

But by Proposition 2.4,

∑ |hπ ( g)v1,w1 ihπ ( g)v2, w2 i| ≤

g∈ A R
!1/2 !1/2
(3.30) 2 2
≤ ∑ |hπ ( g)v1, w1 i| ∑ |hπ ( g)v2, w2 i| ≺
g∈ A R g∈ A R

≺ ( 1 + R ) 2 k v 1 k k w1 k k v 2 k k w2 k ,
which concludes the proof. 
14 ADRIEN BOYER AND ŁUKASZ GARNCAREK

Lemma 3.5. For v i , wi ∈ Lip(∂Γ, d ǫ ), we have the convergence

(3.31) lim Φ R (v1 , v2, w1 , w2 ) = h f 2 |∂Γ v1 , v2 ihw1 , f 1 |∂Γ w2 i.
R→∞

Proof. Define F ∈ C (Γ × Γ) by
(3.32) F ( g, h) = f 1 ( g) f 2 (h)v1 ( p(h))w1( p( g))v2( p(h))w2 ( p( g)),
and put
e ( g ) v i , w i i,
αi ( g) = hπ
(3.33) β i ( g) = vi ( ǧ)wi ( ĝ),
dνR ( g) = | f 1 ( g) f 2 ( g−1 )|dµ R ( g).
By definition of µ R we have
Z Z
(3.34) lim F dJ∗ µ R = F dµ2 = h f 2 |∂Γ v1 , v2 ihw1 , f 1 |∂Γ w2 i,
R→∞
and we need to estimate the difference
 Z 
 
( v ,v ,
Φ R 1 2 1 2 w , w ) − F dJ∗ µ R  ≤
(3.35) Z  
 
≤ α1 ( g)α2 ( g) − β 1 ( g) β 2 ( g) dνR ( g).
By [16, Lemma 5.3], for v, w ∈ Lip(∂Γ, dǫ ) we have
 
 
(3.36) hπe( g)v, wi − v( ǧ)w( ĝ) ≺v,w (1 + | g|)−1/D ,
so for g ∈ A R , with estimates depending on vi and wi ,
 
 
α1 ( g)α2 ( g) − β 1 ( g) β 2 ( g) ≤
   
   
≤ α1 ( g)(α2 ( g) − β 2 ( g)) + (α1 ( g) − β 1 ( g)) β 2 ( g) ≺
(3.37)
≺ (|α1 ( g)| + | β 2 ( g)|)(1 + R)−1/D ≺
≺ (| β 1 ( g)| + | β 2 ( g)| + (1 + R)−1/D )(1 + R)−1/D .
But
(3.38) | β i ( g)| ≤ kvi k ∞ kwi k ∞ ,
hence the differences in (3.37) converge to 0 uniformly as R → ∞, and since νR (Γ)
is bounded by k f 1 k∞ k f 2 k ∞ , so is the difference in (3.35). 
Proof of the Asymptotic Orthogonality Theorem 3.3. By Lemma 3.4, the forms Φ R are
uniformly bounded. Hence, to show the desired convergence, it suffices to do
so for vectors belonging to some dense subspaces of H. The space of Lipschitz
functions is such a space, and by Lemma 3.5, the convergence holds. We are thus
done with the proof. 
As an immediate corollary, we deduce an ergodic theorem à la Bader-Muchnik
for general hyperbolic groups. For f ∈ C (∂Γ) let m( f ) denote the correspond-
ing multiplication operator on H, and let P be the orthogonal projection onto the
subspace spanned by 1∂Γ .
Corollary 3.6. Let Γ be a hyperbolic group endowed with a metric d ∈ D(Γ), and let
µ R ∈ Prob(Γ) be a family of measures satisfying the conditions (1)-(3) of Theorem 3.2.
Then for any function f ∈ C (Γ) we have
(3.39) lim
R→∞ g∈Γ
∑ µR ( g) f ( g)πe( g) → m( f |∂Γ ) P
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 15

in the weak operator topology.

Note that this ergodic theorem can be seen as a special case of [16, Proposition
5.4], and is sufficient for proving the irreducibility of boundary representations
associated with Patterson-Sullivan measures of hyperbolic groups.

4. C ONVEX COCOMPACT GROUPS OF ISOMETRIES OF CAT(-1) SPACES

CAT(-1) spaces are a class of very well-behaved hyperbolic metric spaces. Their
geometry is more precise, and many inequalities are in fact equalities, yielding
better estimates in the asymptotic Schur orthogonality relations. This is essentially
due to an equidistribution result of Roblin, deduced from the mixing property of
the geodesic flow. The CAT(-1) setting is the perfect receptacle to develop the phe-
nomenon of mixing of the geodesic flow as well as several of its consequences,
and to generalize some deep results of ergodic theory of discrete groups in nega-
tive curvature [24]. In this section we survey the geometry of CAT(-1) spaces. We
freely rely on [6] where the reader could consult for further details.
4.1. CAT(-1) spaces and groups. Let ( X, d) be a proper (i.e. the balls are relatively
compact) geodesic metric space. It is said to be CAT(-1) if every geodesic trian-
gle in X is thinner than its comparison triangle in the hyperbolic plane, see [10,
Introduction]. This implies that X is hyperbolic, and since it is geodesic, the dis-
cussion of Section 2 applies to X with genuine geodesics used in place of rough
geodesics. Moreover, if an → ξ ∈ ∂X, and bn → η ∈ ∂X, then the correspond-
ing Gromov products converge without the additive error, which appears in the
general hyperbolic case, namely we have
(4.1) (ξ, η ) x = lim ( an , bn ) x .
n→∞
Finally, the formula
(4.2) d x (ξ, η ) = e−( ξ,η ) x
already defines a metric on ∂X (we set d x (ξ, ξ ) = 0). This is due to M. Bourdon,
we refer to [6, 2.5.1 Théorème] for more details.
Now, let Γ be a non-elementary discrete group of isometries of X. The limit set
of Γ, denoted by ΛΓ is the set of all accumulation points of a fixed orbit of Γ in ∂X,
i.e. ΛΓ = Γx ∩ ∂X, where the closure is taken in X. It is independent of the choice
of x ∈ X. The geodesic hull GH (ΛΓ ) is defined as the union of all geodesics in
X with both endpoints in ΛΓ , and the convex hull of ΛΓ , denoted by CH (ΛΓ ), is
the smallest subset of X containing GH (ΛΓ ) with the property that every geodesic
segment between any pair of points x, y ∈ CH (ΛΓ ) also lies in CH (ΛΓ ). We say
that Γ is convex cocompact if it acts cocompactly on CH (ΛΓ ). It is a well known fact
that convex cocompact groups are hyperbolic, see [6, Corollaire 1.8.3]. In this case,
∂Γ = ΛΓ .
4.2. Bowen-Margulis-Sullivan measures. We return to the general case of a dis-
crete group of isometries of X, possibly non-hyperbolic. It turns out that one can
still perform a construction similar to the one yielding the Patterson-Sullivan mea-
sures. First, one defines the critical exponent α(Γ) of Γ by
n o
(4.3) α(Γ) = inf s ∈ R ∗+ : ∑ e−sd( gx,x ) < ∞
g∈Γ

(in case of a hyperbolic group, we have α = Dǫ = log ω). Now, a Γ-invariant

conformal density of dimension α is a family of positive finite measures {µ x : x ∈ X }
on X such that
(1) for g ∈ Γ we have g∗ µ x = µ gx ,
16 ADRIEN BOYER AND ŁUKASZ GARNCAREK

(2) for x, y ∈ X the measures µ x and µy are equivalent, and moreover

dµy
(4.4) (ξ ) = ω β ξ ( x,y),
dµ x

where β ξ ( x, y) is the horospherical distance of x and y relative to ξ, defined as

the limit

(4.5) β ξ ( x, y) = lim d( x, γ(t)) − d(y, γ(t))

t→∞

where γ is any geodesic ray in X representing ξ.

In case of a discrete group Γ of isometries of a CAT(-1) space, there always exists a
conformal density supported in the limit set ΛΓ . For a proof, see [11] and [6].
In [25], D. Sullivan constructed measures on the unit tangent bundle of the n-
dimensional real hyperbolic space, and proved some striking results for this new
class of measures. We will now briefly recall the definitions of analogous measures
in CAT(-1) spaces. We follow [24, Chapitre 1C] where the reader could find more
details.
Let SX be the set of isometries from R to ( X, d) endowed with the topology
of uniform convergence on compact subsets of R. In other words, SX is the set
of geodesics of X parametrized by R. We have a canonical projection γ 7→ γ(0)
from SX to X, playing the role of the projection from the unit tangent bundle of
a manifold to the manifold. Notice that in the setting of CAT(-1) spaces this map
may be non-surjective, as the geodesics need not be bi-infinite.
The trivial flow on R induces a continuous flow (Φt )t∈R on SX, called the geo-
desic flow , given by Φt (γ)(s) = γ(s + t). For γ ∈ SX, we will denote by γ± ∈ ∂X
the endpoints of γ, i.e. the limits limt→±∞ γ(t). Let

(4.6) ∂2 X = ∂X × ∂X \ {( x, x ) : x ∈ ∂X } .

If we fix a basepoint x ∈ X, we may identify SX with ∂2 X × R via

(4.7) γ 7→ (γ− , γ+ , β γ− (γ(0), x )).

Under this identification, called the Hopf parametrization, the geodesic flow can be
written as

(4.8) Φt (ξ, η, s) = (ξ, η, s + t),

and it commutes with the action of Γ on ∂2 X × R by

(4.9) g · (ξ, η, s) = ( gξ, gη, s + β ξ ( x, g−1 x )).

Using a Γ-invariant conformal density of dimension α, we define the Bowen-

Margulis-Sullivan measure m on SX, often referred to as the BMS measure, by

dµ x (ξ )dµ x (η )ds
(4.10) dm(ξ, η, s) = ·
d x (ξ, η )2α

It is invariant under both the geodesic flow Φt , and the action of Γ. The latter
action admits a measurable fundamental domain, which can be identified with
the quotient SX/Γ. We denote by mΓ the measure on SX/Γ corresponding to the
restriction of m to the fundamental domain, and say that Γ admits a finite BMS
measure if mΓ is finite.
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 17

4.3. Equidistribution and asymptotic Schur orthogonality. The translation length

of an element g ∈ Γ is defined as

(4.11) t( g) := inf d( x, gx ),
x∈X

and the spectrum of Γ is the subgroup of R generated by t( g) where g ranges over

the hyperbolic isometries in Γ. If it is a discrete subgroup, we say that Γ has arith-
metic spectrum. We are interested in discrete groups with a non-arithmetic spec-
trum, because they guarantee mixing properties of the induced geodesic flow on
SX/Γ. This condition is satisfied for isometry groups of Riemannian surfaces, hy-
perbolic spaces, and CAT(-1) spaces with a non-trivial connected component in
their limit set. We refer to [14] and to [24, Proposition 1.6, Chapitre 1] for more
details. The proof of mixing can be found in [3, Proposition 7.7] in the case of
negatively curved manifolds, and in [24, Chapitre 3] for general CAT(-1) spaces.
Denote by δx the unit Dirac mass centered at x. Using the mixing property of the
induced geodesic flow on (SX/Γ, mΓ ), in [24, Théorème 4.1.1, Chapitre 4], Roblin
proved the following result.

Theorem 4.1 (T. Roblin). Let Γ be a discrete group of isometries of a CAT(-1) space X
with a non-arithmetic spectrum. Assume that Γ admits a finite BMS measure associated
to a Γ-invariant conformal density µ of dimension α = α(Γ). Then for all x, y ∈ X we
have

(4.12) lim αe−αn kmΓ k ∑ δg−1 x ⊗ δgy = µ x ⊗ µy

n→∞
g∈Γ
d ( x,gy )< n

in the weak* topology of C ( X × X )∗ .

Let A R ( x, y) = { g ∈ Γ| R − h ≤ d( x, gy) ≤ R + h} for some sufficiently large

h > 0. We now have the following.

Corollary 4.2 (T. Roblin). Under the assumptions of Theorem 4.1, for all x, y ∈ X we
have
1
(4.13) lim
R→∞ | A R ( x, y)| ∑ δg−1 x ⊗ δgy = µ x ⊗ µy
g ∈ A R ( x,y )

in the weak* topology on C ( X × X )∗ .

Now, suppose again that Γ is convex cocompact, and thus hyperbolic. Fix a
basepoint o ∈ X and denote dΓ ( g, h) = d( go, ho ). If the orbit map g 7→ go is
not injective, this is only a pseudometric, which can be remedied by applying a
bounded perturbation
(
+ 0 if g = h,
(4.14) dΓ ( g, h) =
dΓ ( g, h) + 1 otherwise,

yielding a roughly isometric metric d + Γ ∈ D( Γ). Recall that we may identify ∂Γ

with ΛΓ , and then the measure µo of the Γ-invariant conformal density is then
in the Patterson-Sullivan class corresponding to d+ Γ . If A R,h is the annulus corre-
sponding to d+Γ , then by Corollary 4.2, for sufficiently large h > 0 we have
1
(4.15) lim   ∑ δgo ⊗ δg−1 o = µo ⊗ µo ,
R→∞ A R,h  g ∈ A R,h
18 ADRIEN BOYER AND ŁUKASZ GARNCAREK

and since the orbit map g 7→ go extends to a continuous map Γ → X identifying

∂Γ with ΛΓ , we also have
1
(4.16) lim   ∑ δg ⊗ δg−1 = µo ⊗ µo ,
R → ∞  A R,h 
g∈ A R,h

obtaining a family of measures on Γ yielding a sharper version of the Equidistri-

bution Theorem 3.2.
From this, we deduce the asymptotic Schur orthogonality relations in CAT(-1)
spaces. Let π be the boundary representation associated to the metric d +Γ , and
recall that
π ( g)
(4.17) e( g) =
π
Ξ( g)
is the normalization with the Harish-Chandra function.
Theorem 4.3 (Asymptotic Schur Orthogonality). Assume that Γ is a convex cocom-
pact group of isometries of a CAT(-1) space with a non-arithmetic spectrum. Then for suf-
ficiently large h > 0, all continuous functions f 1 , f 2 ∈ C ( X ), and vectors v1 , v2 , w1 , w2 ∈
L2 (∂X, µo ) we have
1
(4.18) lim   ∑ f 1 ( go ) f 2 ( g−1 o )hπ e( g ) v 2 , w2 i =
e( g)v1, w1 ihπ
R→∞ A R,h  g ∈ A R,h

= h f 2 |∂X v1 , v2 ihw1 , f 1 |∂X w2 i.

Note that this theorem implies Theorem A of [7], an ergodic theorem à la Bader-
Muchnik.
Remark 4.4. We may wonder whether this convergence holds for non-uniform lat-
tices in SL(2, R ) (or more generally in rank one semisimple Lie groups) for bound-
ary representations associated to the Lebesgue measure on the boundary. It turns
out that, since the Lebesgue measure satisfies Ahlfors regularity (with respect to
the visual metric associated to the Riemannian metric of the symmetric space) the
convergence holds for Lipschitz functions on the boundary. But it is important
to note that the convergence does not extend to L2 functions because it requires
property RD with respect to the Riemannian metric.
Remark 4.5. Let Γ be a non-abelian free group acting on its Cayley graph X corre-
sponding to the standard generators. Although in this case the spectrum is arith-
metic, by [8] we still have an equiuistribution theorem, leading to asymptotic or-
thogonality relations
1
(4.19) lim ∑ f 1 ( g) f 2 ( g−1 )hπ
e( g)v1, w1 ihπ
e( g ) v 2 , w2 i =
n→∞ | n|
S g ∈ Sn

= h f 2 |∂Γ v1 , v2 ihw1 , f 1 |∂Γ w2 i,

where f 1 , f 2 ∈ C (Γ ∪ ∂Γ), and Sn = { g ∈ Γ : | g| = n}.

5. S IMPLY CONNECTED MANIFOLDS AND G IBBS MEASURES

5.1. The Gibbs stream. We briefly recall the theory of Gibbs measures in negative
curvature, following [23] where the reader could consult for more details. Let X
be a complete simply connected Riemannian manifold of dimension at least 2,
and pinched sectional curvature −b2 ≤ K ≤ −1 with b ≥ 1, equipped with its
Riemannian distance denoted by d, and acted upon by a non-elementary discrete
group of isometries Γ. Such manifolds form a particular class of CAT(-1) spaces in
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 19

which the Patterson-Sullivan measures can be further generalized. We will assume

that X = M e is the universal cover of a manifold M with fundamental group Γ, and
denote q : X → M the covering map.
Let p : T 1 X → X be the unit tangent bundle of X, and for v ∈ T 1 X denote by γv
the unique geodesic in X with γv′ (0) = v. We may equip T 1 X with the following
metric
Z ∞
1 2
(5.1) d T1 X (v, w) = √ d(γv (t), γw (t))e−t /2 dt.
π −∞
Now, for a Hölder-continuous map F : T 1 M → R, called a potential, let Fe = F ◦ q
be its lift to a Γ-invariant potential on T 1 X. We will require that Fe is symmetric, i.e.
invariant under the antipodal map v 7→ −v.
For all x, y ∈ X, define
Z d ( x,y )
(5.2) d F ( x, y) = Fe(γ′ (t)) dt
0

where γ is the geodesic segment from x to y. A priori, d F is not non-negative and

is far from being a metric, nevertheless the symmetry of Fe implies that
(5.3) d F ( x, y) = d F (y, x ).
Now, for ξ ∈ ∂X define the Gibbs cocycle as
(5.4) CξF ( x, y) = lim d F (y, ξ t ) − d F ( x, ξ t ),
t →+ ∞

where ξ t is any geodesic ray representing ξ. Observe that if Fe = −1 the Gibbs

cocycle is nothing but the horospherical distance β ξ ( x, y).
Let σ ∈ R. A family of positive finite measures {µ xF : x ∈ X } on X is a Gibbs
stream of dimension σ for (Γ, F ) if
(1) For all x and y in X, µ xF and µyF are equivalent, and

dµyF CξF −σ ( x,y )

(5.5) (ξ ) = e ,
dµ xF
(2) For all g ∈ Γ, and x ∈ X we have g∗ µ xF = µ gx
F .

In this context, the critical exponent of (Γ, F ) is defined as

1 F ( x,gx )
(5.6) σΓ,F = lim sup ∑ ed ,
n→∞ n g ∈ Sn,h

where Sn,h = { g ∈ Γ : n − h ≤ d( gx, x ) ≤ n}, and h > 0 is sufficiently large.

Proposition 5.1. (S-J. Patterson, [22]) If σΓ,F < ∞, then there exists at least one Gibbs
stream of dimension σΓ,F with support exactly equal to ΛΓ .
5.2. Equidistribution and asymptotic Schur orthogonality. For all x ∈ X and
ξ, η ∈ ∂X define the gap map as
d F ( x, ξ t ) + d F ( x, ηt ) − d F (ξ t , ηt )
(5.7) Dx,F (ξ, η ) = exp lim .
t→∞ 2
When F = −1, the gap map Dx,F is just the visual metric d x from Section 4.1. Also,
note the Γ-invariance of Dx,F, namely
(5.8) Dgx,F ( gξ, gη ) = Dx,F (ξ, η )
for all g ∈ Γ and ξ, η ∈ ∂X.
20 ADRIEN BOYER AND ŁUKASZ GARNCAREK

Let µ F be a Gibbs stream for (Γ, F ) of dimension σ = σ( Γ,F ). Once we have fixed
a base point x ∈ X and used the Hopf parametrization, define on T 1 X the Gibbs
measures associated with µ F by
dµ xF (η )dµ xF (ξ )dt
(5.9) dm(ξ, η, t) = .
Dx,F −σ(η, ξ )2
Similarly as in Section 4.2, this measure descends to a measure m F on the quotient
T 1 X/Γ = T 1 M, called the Gibbs measure on T 1 M associated with µ F . It is invari-
ant under the geodesic flow, and in the case of convex cocompact groups, always
finite.
We are now ready to state the equidistribution theorem for the Gibbs stream
due to Paulin, Policott and Shapira [23].
Theorem 5.2 (Paulin-Pollicott-Schapira). Assume that σ is finite and positive. If m F
is finite and mixing under the geodesic flow on T 1 X, then for all x, y ∈ X and sufficiently
large h > 0
σ km F ke−σn F ( x,gy )
(5.10) lim ∑ ed δg−1 x ⊗ δgy = µ xF ⊗ µyF
n → ∞ 1 − e− cσ
g ∈ Sn,h

in the weak* topology of C ( X × X )∗ .

In [9], the irreducibility and classification of the quasi-regular representations
associated with Gibbs streams are proved. In particular, the Harish-Chandra func-
tions have been studied, and some useful estimates were obtained.
Pick a basepoint x ∈ X and let µ = µ xF be the measure in the Gibbs stream
associated to x. Denote by π the corresponding quasi-regular representation of Γ
on L2 ( X, µ), and consider its Harish-Chandra function Ξ( g) = hπ ( g)1, 1i. Again,
e ( g) = π ( g)/Ξ( g). We have
let π
Proposition 5.3. Assume that Γ is convex cocompact and that F is symmetric. Then we
have
F − σ ( x,gx ) /2
(5.11) Ξ( g ) ≍ ed (1 + d( x, gx )).
In the proof of Theorem 3.3, a crucial role is played by [16, Lemma 5.3], asserting
that for v, w ∈ Lip(∂Γ, d ǫ ) we have
 
 
(5.12) eµ PS ( g)v, wi − v( ǧ)w( ĝ) ≺v,w (1 + | g|)−1/D .
hπ
In the context of Gibbs stream we have the following analogous result.
Lemma 5.4. Assume that Γ is convex cocompact and that F is symmetric. Then for any
ǫ > 0 there exists Cǫ > 0 such that for all v, w ∈ Lip(∂X, d x )
  Cǫ
 
(5.13) e( g)v, wi − v( ǧ)w( ĝ) ≺v,w ǫ +
hπ
d( x, gx )
Proof. Following the proof of [16, Lemma 5.3] we have
  Z CξF −σ ( x,gx ) /2
  e
e( g)v, wi − v( ǧ)w( ĝ) ≤ kvk∞
hπ |w(ξ ) − w( ĝ)| dµ(ξ )
∂X Ξ( g)
(5.14)
Z CξF −σ ( x,g −1 x ) /2
e
+ k wk∞ |v(ξ ) − v( ǧ)| dµ(ξ ).
∂X Ξ ( g −1 )
Both terms are similar, so it is enough to study only the first one. Take ǫ > 0 and
define B = B( ĝ, ǫ). We will estimate the integrals over B and ∂X \ B separately.
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 21

Since w is in Lip(∂X, do ), we have

Z CξF −σ ( x,gx ) /2
e
(5.15) |w(ξ ) − w( ĝ)| dµ(ξ ) ≺w ǫ.
B Ξ( g)
The integral over ∂X \ B is controlled thanks to the estimate [9, Lemma 5.1]. More
precisely we have
Z CξF −σ ( x,gx ) hπ ( g)1, 1∂X \ B i
e Cǫ
(5.16) |w(ξ ) − w( ĝ)| dµ(ξ ) ≺w ≤ ,
∂X \ B Ξ( g) Ξ( g) d( x, gx )
where the last inequality follows from [9, Lemma 5.1] and complete the proof. 
We now obtain the following asymptotic Schur orthogonality relations for quasi-
regular representations associated with Gibbs streams. Again, the assumption of
non-arithmeticity guarantees the mixing property of the geodesic flow in the con-
text of Gibbs measures, see [23, Theorem 8.1]. By taking f 2 = 1X and v2 = w2 =
1∂X we recover Theorem A in [9], another ergodic theorem à la Bader-Muchnik,
used to prove irreducibility of the boundary representation.
Theorem 5.5. Assume that Γ is convex cocompact with a non-arithmetic spectrum, and
that the potential F is symmetric. Then for sufficintly large h > 0, all f 1 , f 2 ∈ C ( X ) and
all v1 , v2 , w1 , w2 ∈ L2 (∂X, µ) we have
F ( x,gx )
(5.17) lim Ce−σR ∑ ed f 1 ( g) f 2 ( g−1 )hπ e ( g ) v 2 , w2 i =
e( g)v1, w1 ihπ
R→∞ g ∈ S R,h

= h f 2 |∂X v1 , v2 ihw1 , f 1 |∂X w2 i,

where
σ km F k
(5.18) C= .
1 − e−hσ
The proof is a straightforward adaptation of the proof of the Asymptotic Or-
thogonality Theorem 3.3. To show the uniform boundedness part, one just needs
to observe that for g ∈ SR,h
F F ( x,gx )− σd( x,gx) F −σ
Ce−σR ed ( x,gx ) Ced Ced ( x,gx ) 1
(5.19) 2
≍h = ≍ .
Ξ( g) Ξ ( g )2 Ξ ( g )2 (1 + R)2
For the convergence part, one uses Lemma 5.4.

6. A PPLICATION TO MONOTONY

Here we will present an example of application of our results. We will deal with
the phenomenon of monotony of representations, described by Kuhn and Steger in
[20].
Let Γ be a non-abelian free group. By a generalized boundary representation (called
boundary representation in [20]) of Γ on a Hilbert space H we will understand a pair
(σ, ρ), where
(1) σ is a unitary representation of Γ on H,
(2) ρ is a representation of the C ∗ -algebra C (∂Γ) on H,
(3) σ and ρ satisfy the condition σ ( g)ρ( f )σ ( g−1) = ρ( f ◦ g −1 ) for all g ∈ Γ
and f ∈ C (∂Γ).
In other words, this is a representation of the crossed product C (∂Γ) ⋊ Γ, where Γ
acts on C (∂Γ) via g · f = f ◦ g−1 .
Now, let π be a unitary representation of Γ on a Hilbert space H. A boundary
realization of π on a Hilbert space H ′ is a triple (ι, σ, ρ), where (σ, ρ) is a generalized
22 ADRIEN BOYER AND ŁUKASZ GARNCAREK

boundary representation of Γ on H ′ , and ι : H → H ′ is a Γ-equivariant isometric

embedding, such that ρ(C (∂Γ))ι(H) is dense in H ′ . If ι is bijective, the realization
is said to be perfect.
Boundary representations studied herein come equipped with tautological per-
fect realizations. Indeed, if π is a boundary representation on H = L2 (∂Γ, µ), then
the pair (π, m), where m is the multiplication action m( f )φ = f φ of C (∂Γ) on H,
is a generalized boundary representation of Γ, and together with the identity map
of H it yields a perfect boundary realziation of π.
The fundamental idea of Kuhn and Steger in [20] is the use of boundary realiza-
tions to classify representations of the free group weakly contained in the regular
representation. They propose three classes of representations, monotonous, duplic-
itous, and odd, distinguished by the behavior of their boundary realizations, and
inspired by the case where Γ is a lattice in SL2 (R ).
Let π be a unitary representation of Γ, and let (ι, σ, ρ) and (ι′ , σ ′ , ρ′ ) be two
boundary realizations of π, on Hilbert spaces H, and H ′ , respectively. We will
consider them equivalent, if there exists a unitary isomorphism U : H → H ′ inter-
twining (σ, ρ) with (σ ′ , ρ′ ), such that Uι = ι′ .
We say that the representation π satisfies monotony, if it admits a unique (up to
equivalence) boundary realization, and moreover this realization is perfect. If Γ is
a lattice in SL2 (R ), then the restrictions of the endpoint spherical principal series
representations to Γ are monotonous.
The representation π is said to satisfy duplicity, if it admits exactly two non-
equivalent perfect boundary realizations (ι1 , σ1 , ρ1 ) and (ι2 , σ2 , ρ2 ), and moreover
any imperfect realization is, up to equivalence, a combination of these two, given
by the embedding ι(v) = t1 ι1 (v) ⊕ t2 ι2 (v), where t1 , t2 > 0, and t21 + t22 = 1. If Γ
is a lattice in SL2 (R ), this happens for the restrictions of non-endpoint principal
series representations.
Finally, a representation π of Γ on a Hilbert space H satisfies oddity, if it admits a
unique boundary realization (ι, σ, ρ) on H ′ , this realization is imperfect, and more-
over if K is the orthogonal complement of ι(H) in H ′ , then the restriction ρ|K is
irreducible, not equivalent to π, and (idK , ρ, σ ) is its unique boundary realization.
In case where Γ is a lattice in SL2 (R ), this situation happens for the restrictions of
the mock discrete series representations.
Conjecture (Kuhn-Steger). Any irreducible unitary representation of Γ weakly con-
tained in the regular representation is monotonous, or duplicitous, or odd.
Examples of representations of non-abelian free groups satisfying monotony
were defined in [19], and the proof of their monotony can be found in [20]. The
authors mention that “monontony is a generic condition”.
Remark 6.1. Monotony of a boundary representation can be thought of as a rigidity
phenomenon. Indeed, given a quasi-regular representation πµ associated with a
measure µ on the boundary ∂Γ of the free group, there is a unique (up to equiva-
lence) way for the algebra of continuous functions C (∂Γ) to act on L2 (∂Γ, µ) while
satisfying the cross product relation with πµ , namely the multiplication action. In
other words, there is an unique way to extend the unitary representation πµ of Γ
to a representation of the cross product C (∂Γ) ⋊ Γ.
Whereas we do not investigate in this paper the phenomenon of duplicity or
oddity, one of the consequences of our Asymptotic Orthogonality Theorem is a
new strategy for proving monotony of a wider class boundary representations. It
allows us to obtain the following result, strengthening the speculation of Kuhn
and Steger about monotony as a generic condition.
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 23

Theorem 6.2. Let Γ be a non-abelian free group, and let µ be the Patterson-Sullivan
measure on ∂Γ associated with a metric d ∈ D(Γ). Then the associated boundary repre-
sentation π satisifes monotony.
Our method will be still relying on the Good Vector Bound (GVB) introduced by
Kuhn and Steger. Endow Γ with its word-length dw associated to its free basis,
and denote by Sn ⊆ Γ the set of elements of length n. A representation π of Γ on a
Hilbert space H satisfies GVB if there exists a non-zero vector w ∈ H such that for
all v ∈ H we have
(6.1) ∑ |hπ ( g)v, wi|2 ≤ C k vk22 .
g ∈ Sn

The key result is the following.

Proposition 6.3 ([20, Proposition 2.7 and Corollary 2.8]). If a unitary representation
of Γ does not satisfy the Good Vector Bound then it satisfies monotony.
Proof of Theorem 6.2. We shall prove that π cannot satisfy GVB. Suppose to the con-
trary that (6.1) holds for some nonzero w ∈ H. Since the metrics d and dw are
quasi-isometric, for all ǫ, R > 0, and v ∈ H we obtain
|hπ ( g)v, wi|2 |hπ ( g)v, wi|2
∑ (1 + R)1+ ǫ
≺h ∑ 1+ ǫ
≺
g ∈ A R,h g ∈ Γ (1 + d ( g ))
(6.2)
|hπ ( g)v, wi|2
≺ ∑ 1+ ǫ
≺ǫ,w kvk2 .
g ∈ Γ (1 + d w ( g ))

On the other hand, for sufficiently large h > 0, and a suitable family of measures
{µ R }, by the Asymptotic Orthogonality Theorem 3.3 we have
Z
|hπ ( g)v, wi|2
(6.3) k vk2 kwk2 = lim e( g)v, wi|2 dµ R ≺h lim inf
|hπ ∑ ,
R→∞ R→∞
g ∈ A R,h
(1 + R)2

so for large R
|hπ ( g)v, wi|2
(6.4) ∑ ≻ h ( 1 + R ) 1 − ǫ k v k2 k w k 2 .
g ∈ A R,h (1 + R)1+ ǫ

If we choose ǫ < 1, this contradicts (6.2). 

Thanks to the work of [5], Theorem 6.2 covers also quasi-regular representations
associated with harmonic measures of random walks. Fix a word metric on Γ
corresponding to the free basis, and let ν be a symmetric random walk on Γ. We
say that ν has exponential moment, if for some λ > 0 we have
(6.5) ∑ eλ|g| ν( g) < ∞.
g∈Γ

Let F ( g, h) denote the probability that a trajectory of the random walk starting at
g passes through h. We will assume that for any r > 0 there exists C (r ) such that
(6.6) F ( g, h) ≤ C (r ) F ( g, k) F (k, h)
for all k within distance r from a geodesic joining g and h in the Cayley graph of Γ.
Both this condition, and having exponential moment are satisfied by any finitely
supported ν, and condition (6.6) is due to Ancona [2]. Together they guarantee
that the Green metric
(6.7) dν ( g, h) = − log F ( g, h)
24 ADRIEN BOYER AND ŁUKASZ GARNCAREK

is in the class D(Γ). By [5, Theorems 1.1(ii) and 1.5], the harmonic measure µ on ∂Γ
associated to the random walk ν is equivalent to the Patterson-Sullivan measure
corresponding to the Green metric dν . Hence the following.

Corollary 6.4. Let Γ be a non-abelian free group, and let ν be a symmetric random walk
on Γ with an exponential moment, satisfying condition (6.6). If µ is the associated har-
monic measure on ∂Γ, then the corresponding quasi-regular representation πµ satisfies
monotony.

R EFERENCES
[1] S. Adams. Boundary amenability for word hyperbolic groups and an application to smooth dy-
namics of simple groups. Topology, 33(4):765–783, 1994.
[2] A. Ancona. Théorie du potentiel sur les graphes et les variétés, Ecole d’été de Probabilités de Saint-
Flour XVIII—1988. Lecture Notes in Math. Springer, Berlin, 1427:1–112, 1990.
[3] Martine Babillot. On the mixing property for hyperbolic systems. Israel J. Math., 129:61–76, 2002.
[4] Uri Bader and Roman Muchnik. Boundary unitary representations—irreducibility and rigidity. J.
Mod. Dyn., 5(1):49–69, 2011.
[5] Sébastien Blachère, Peter Haı̈ssinsky, and Pierre Mathieu. Harmonic measures versus quasicon-
formal measures for hyperbolic groups. Ann. Sci. l’École Norm. Supérieure, 44(4):683–721, 2011.
[6] Marc Bourdon. Structure conforme au bord et flot géodésique d’un CAT(−1)-espace. Enseign.
Math. (2), 41(1-2):63–102, 1995.
[7] Adrien Boyer. Equidistribution, ergodicity and irreducibility in CAT(−1) spaces. arXiv, 1412.8229,
2014.
[8] Adrien Boyer and Antoine Pinochet Lobos. An ergodic theorem for the boundary representations
of free group. Accepted to Bulletin of the Belgian Mathematical Society, arXiv, 1601.00668, 2016.
[9] Adrien Boyer and Dustin Mayeda. Equidistribution, ergodicity and irreducibility associated with
Gibbs measures. Accepted to Commentarii Mathematici Helvetici, arXiv, 1601.02275, 2016.
[10] Martin R. Bridson and André Haefliger. Metric Spaces of Non-Positive Curvature. Springer, 1999.
[11] M. Burger and S. Mozes. CAT(−1)-spaces, divergence groups and their commensurators. J. Amer.
Math. Soc., 9(1):57–93, 1996.
[12] Indira Chatterji. Introduction to the Rapid Decay property. arXiv, 1604.06387, 2016.
[13] Michel Coornaert. Mesures de Patterson-Sullivan sur le bord d’un espace hyperbolique au sens
de Gromov. Pacific J. Math., 159(2):241–270, 1993.
[14] Françoise Dal’bo. Remarques sur le spectre des longueurs d’une surface et comptages. (french)
[remarks on the length spectrum of a surface, and counting]. Bol. Soc. Brasil. Mat. (N.S.) 30, no. 2,
199-221, 1999.
[15] Pierre de la Harpe. Groupes hyperboliques, algèbres d’opérateurs et un théorème de Jolissaint. C.
R. Acad. Sci. Paris Sér. I Math., 307(14):771–774, 1988.
[16] Łukasz Garncarek. Boundary representations of hyperbolic groups. arXiv, 1404.0903, 2014.
[17] Paul Jolissaint. Rapidly Decreasing Functions in Reduced C∗ -Algebras of Groups. Trans. Am. Math.
Soc., 317(1):167–196, 1990.
[18] M. Gabriella Kuhn. Amenable Actions and Weak Containment of Certain Representations of Dis-
crete Groups. Proc. Am. Math. Soc., 122(3):751–757, 1994.
[19] M. Gabriella Kuhn and Tim Steger. More irreducible boundary representations of free groups.
Duke Math. J., 82(2):381–436, 1996.
[20] M. Gabriella Kuhn and Tim Steger. Monotony of Certain Free Group Representations. J. Funct.
Anal., 179(1):1–17, 2001.
[21] Grigory Margulis. On some aspects of the theory of anosov systems. Springer Monographs in Math-
ematics, Springer-Verlag, Berlin, With a survey by Richard Sharp: Periodic orbits of hyperbolic flows,
Translated from the Russian by Valentina Vladimirovna Szulikowska, 2004.
[22] S. J. Patterson. The limit set of a Fuchsian group. Acta Math., 136(3-4):241–273, 1976.
[23] Frédéric Paulin, Mark Pollicott, and Barbara Schapira. Equilibrium states in negative curvature.
Astérisque, (373):viii+281, 2015.
[24] Thomas Roblin. Ergodicité et équidistribution en courbure négative. Mém. Soc. Math. Fr. (N.S.),
(95):vi+96, 2003.
[25] Dennis Sullivan. The density at infinity of a discrete group of hyperbolic motions. Inst. Hautes
Études Sci. Publ. Math., (50):171–202, 1979.
 ASYMPTOTIC ORTHOGONALITY IN HYPERBOLIC GROUPS 25

(A. Boyer) W EIZMANN I NSTITUTE OF S CIENCE , H ERZL S T 234, R EHOVOT, 7610001, I SRAEL
E-mail address: aadrien.boyer@gmail.com

(Ł. Garncarek) W EIZMANN I NSTITUTE OF S CIENCE , H ERZL S T 234, R EHOVOT, 7610001, I SRAEL
E-mail address: lukgar@gmail.com