TRACE THEOREM FOR QUASI-FUCHSIAN GROUPS

arXiv:1703.05447v1 [math.OA] 16 Mar 2017

A. CONNES, F. SUKOCHEV, AND D. ZANIN

Dedicated to Dennis Sullivan.

Abstract. We complete the proof of the Trace Theorem in the quantized

calculus for quasi-Fuchsian group which was stated and sketched, but not
fully proved, on pp. 322-325 in the book “Noncommutative Geometry”of the
first author.

1. Introduction
We first recall how quasi-Fuchsian groups are obtained by Bers ([2]) from a pair
of cocompact Fuchsian groups Γ1 , Γ2 and a given group isomorphism α : Γ1 → Γ2 .
All required notations and notions used below are explained in Section 2. The
quasi-Fuchsian group G = G(Γ1 , Γ2 , α) is a discrete subgroup G ⊂ PSL(2, C) which
simultaneously uniformizes the compact Riemann surfaces Xj = D/Γj , j = 1, 2,
(where D is the unit disk in C) in the following sense ([6]):
(1) There is a Jordan curve C ⊂ C̄ = S 2 invariant under any g ∈ G and such
that the action of G on C is minimal (every orbit is dense).
(2) Let Σint and Σext be the connected components of the complement of C.
There are conformal diffeomorphisms Z : D → Σint , Z ′ : D → Σext and
group isomorphisms π : G → Γ1 , π ′ : G → Γ2 such that
g ◦ Z = Z ◦ π(g), g ◦ Z ′ = Z ′ ◦ π ′ (g), π ′ (g) = α(π(g)) ∀g ∈ G.
Furthermore, the group G = G(Γ1 , Γ2 , α) satisfies the following properties:
(i) G is finitely generated.
(ii) G does not contain elliptic or parabolic elements.
The Jordan curve C = Λ(G) is a quasi-circle whose Hausdorff dimension p is
strictly bigger than one except when the Γ1 and Γ2 are conjugate Fuchsian groups
([6], Theorem 2).
The main result of this paper is the following theorem appearing as Theorem 17
on p. 324 in [14]. It gives a formula for the p−dimensional geometric1 probability
measure on C = Λ(G) in terms of the quantized differential [F, Z] of the Riemann
mapping Z : D → Σint understood as a function on the circle S1 = ∂D (to which it
extends by continuity using the Caratheodory theorem ([26])). Here F is the Hilbert
transform on the circle; equivalently, F = 2P − 1, where P is the Riesz projection
and the algebra L∞ (∂D) is identified with its natural action on the Hilbert space
L2 (∂D) by pointwise multiplication. The basic formula depends on the fact that,
1A measure ν on C̄ is called p−dimensional geometric (relative to G) if d(ν ◦ g)(z) =
|g ′ |p (z)dν(z) for every g ∈ G. Here, g ′ is the complex derivative.
1
2

unlike for distributional derivatives, one can take the p-th power |[F, Z]|p of the
absolute value of the quantized differential [F, Z]. The nice geometric properties
of the quasi-Fuchsian groups G = G(Γ1 , Γ2 , α) are used crucially in the proof and
we formulate our result in a slightly greater generality and in more intrinsic terms
without reference to the joint uniformization.
Theorem 1.1. Let G be a finitely generated quasi-Fuchsian group without parabolic
elements. Let p > 1 be the Hausdorff dimension of C = Λ(G), and let ν be the
(unique) p−dimensional geometric probability measure on Λ(G). Then
(a) [F, Z] ∈ Lp,∞ .
(b) for every f ∈ C(Λ(G)) and for every bounded trace2 ϕ on L1,∞ , there exists a
constant c(G, ϕ) < ∞ such that
Z
p
(1.1) ϕ((f ◦ Z) · |[F, Z]| ) = c(G, ϕ) · f (t)dν(t).
Λ(G)

(c) for any Dixmier trace Trω , with ω power invariant, one has c(G, Trω ) > 0.
The statement (c) provides a large class of traces for which c(G, ϕ) > 0. The
notion of power invariance for the limiting process ω is explained in Section 7.
Theorem 1.1 was stated in [14] and the proof3 was sketched there after the
statement of the Theorem and using a number of lemmas but the reference [538]
was never published and the detailed proof is thus unpublished even if the various
steps were described in [14]. It is thus very valuable to make them available while
proving a more general result and introducing variants in the proposed proof in
[14]. The variants concern the estimate of the growth of the Poincaré series which
in [14] is attributed to Corollary 10 of [34] but the precise relation with the two
forms of the absolute Poincare series is assumed without a precise reference. This
relation is due to the convex co-compactness of the action of the quasi Fuchsian
group inside hyperbolic three space, but in this paper the same estimate is obtained
using a different method. The other important point not contained in [14] is the
proof of the Lemma 3.β.11, which is stated there without proof.
We are grateful to our colleagues Christopher Bishop, Magnus Goffeng, Denis
Potapov and Caroline Series for their help in the preparation of this paper.

2. Preliminaries
2.1. General notation. Fix throughout a separable infinite dimensional Hilbert
space H. We let L(H) denote the ∗−algebra of all bounded operators on H. It
becomes a C ∗ −algebra when equipped with the uniform operator norm (denoted
here by k · k∞ ). For a compact operator T on H, let λ(k, T ) and µ(k, T ) denote
its k-th eigenvalue and k-th largest singular value (these are the eigenvalues of |T |
arranged in the descending order). The sequence µ(T ) = {µ(k, T )}k≥0 is referred
to as the singular value sequence of the operator T. The standard trace on L(H) is
denoted by Tr. For an arbitrary operator 0 ≤ T ∈ L(H), we set
nT (t) := Tr(ET (t, ∞)), t > 0,
where ET (a, b) stands for the spectral projection of a self-adjoint operator T corre-
sponding to the interval (a, b). Fix an orthonormal basis in H (the particular choice
2in particular for every Dixmier trace
3which was joint work with D. Sullivan to whom the first author is indebted for his generosity
in sharing his geometric insight.
 3

of basis is inessential). We identify the algebra l∞ of bounded sequences with the

subalgebra of all diagonal operators with respect to the chosen basis. For a given
sequence α ∈ l∞ , we denote the corresponding diagonal operator by diag(α).
Similarly, let (X, κ) be a measure space (finite or infinite, atomless or atomic).
For a measurable function x on (X, κ), we write

n|x| (t) = κ({u : |x|(u) > t}), µ(s, x) = inf{t : n|x| (t) > s}.

2.2. Principal ideals Lp,∞ and infinitesimals of order p1 . For a given 0 <
p < ∞, we let Lp,∞ denote the principal ideal in L(H) generated by the operator
diag({(k + 1)−1/p }k≥0 ). Equivalently,

Lp,∞ = {T ∈ L(H) : µ(k, T ) = O((k + 1)−1/p )}.

These ideals, for different p, all admit an equivalent description in terms of spectral
projections, namely
1
(2.1) T ∈ Lp,∞ ⇐⇒ n|T | ( ) = O(np ).
n
We also have

(2.2) |T |p ∈ L1,∞ ⇐⇒ µp (k, T ) = O((k + 1)−1 ) ⇐⇒ T ∈ Lp,∞ .

The ideal Lp,∞ , 0 < p < ∞, is equipped with a natural quasi-norm4

kT kp,∞ = sup(k + 1)1/p µ(k, T ), T ∈ Lp,∞ .

k≥0

However, for 1 < p < ∞, it is technically convenient to use an equivalent norm

n
1 X
kT kp,∞ = sup(n + 1) p −1 µ(k, T ), T ∈ Lp,∞ .
n≥0
k=0

The following Hölder property (see [10] Section 6 of Chapter 11) is widely used
throughout the paper:
n n
Y 1 X 1
(2.3) Ak ∈ Lpm ,∞ , 1 ≤ m ≤ n, =⇒ Am ∈ Lp,∞ , = .
m=1
p m=1 pm

Similarly, let (X, κ) be a measure space (finite or infinite, atomless or atomic).

We define a function space
1
Lp,∞ (X, κ) = {x is κ−measurable : µ(t, x) = O(t− p )}.

In [14], a compact operator T ∈ L(H) is called an infinitesimal. It is said to

be of order α > 0 if it belongs to the ideal L α1 ,∞ . Equation (2.3) manifests the
fundamental fact that the order of the product of infinitesimals is the sum of their
orders.

4A quasinorm satisfies the norm axioms, except that the triangle inequality is replaced by
||x + y|| ≤ K(||x|| + ||y||) for some uniform constant K > 1.
4

2.3. Traces on L1,∞ .

Definition 2.1. If I is an ideal in L(H), then a unitarily invariant linear functional

ϕ : I → C is said to be a trace.

Since U −1 T U − T = [U −1 , T U ] for all T ∈ I and for all unitaries U ∈ L(H),

and since the unitaries span L(H), it follows that traces are precisely the linear
functionals on I satisfying the condition

ϕ(T S) = ϕ(ST ), T ∈ I, S ∈ L(H).

The latter may be reinterpreted as the vanishing of the linear functional ϕ on the
commutator subspace which is denoted [I, L(H)] and defined to be the linear span
of all commutators [T, S] : T ∈ I, S ∈ L(H).
It is shown in [25, Lemma 5.2.2] that ϕ(T1 ) = ϕ(T2 ) whenever 0 ≤ T1 , T2 ∈ I
are such that the singular value sequences µ(T1 ) and µ(T2 ) coincide. For p > 1, the
ideal Lp,∞ does not admit a non-zero trace while for p = 1, there exists a plethora
of traces on L1,∞ (see e.g. [18] or [25]). An example of a trace on L1,∞ is the
Dixmier trace introduced in [15] that we now explain.

Definition 2.2. The dilation semigroup on L∞ (0, ∞) is defined by setting

t
(σs x)(t) = x( ), t, s > 0.
s
In this paper a dilation invariant extended limit means a state on the algebra
L∞ (0, ∞) invariant under σs , s > 0, which vanishes on every function with bounded
support.

Dixmier trace. Let ω be a dilation invariant extended limit. Then the functional
Trω : L+
1,∞ → C defined by setting
5

t
1
 Z 
Trω (A) = ω t → µ(u, A)du , 0 ≤ A ∈ L1,∞ ,
log(1 + t) 0

is additive and, therefore, extends to a trace on L1,∞ . We call such traces Dixmier
traces.
These traces clearly depend on the choice of the functional ω on L∞ (0, ∞). Using
a slightly different definition, this notion of trace was applied in [14] in the setting of
noncommutative geometry. We also remark that the assumption used by Dixmier
of translation invariance for the functional ω is redundant (see [14, Section IV.2.β]
or [25, Theorem 6.3.6]).
An extensive discussion of traces, and more recent developments in the theory,
may be found in [25] including a discussion of the following facts.
(a) All Dixmier traces on L1,∞ are positive.
(b) All positive traces on L1,∞ are continuous in the quasi-norm topology.
(c) There exist positive traces on L1,∞ which are not Dixmier traces (see [33]).
(d) There exist traces on L1,∞ which fail to be continuous (see [18]).

5Here, singular value function is defined by the formula µ(A) = P

k≥0 µ(k, A)χ(k,k+1) .
 5

2.4. Kleinian groups. A Fuchsian (resp. Kleinian) group is Poincaré’s name for a
discrete subgroup of PSL(2, R) (resp. of PSL(2, C)). We are interested in Kleinian
groups which are obtained by deforming certain Fuchsian groups. A nice deforma-
tion of a Fuchsian group uniformizing a compact Riemann surface is called by Bers
a quasi-Fuchsian group ([2]). The corresponding action on the complex sphere C̄ is
topologically conjugate to the action of the Fuchsian group and Poincaré noticed
the deformation of the round circle of the Fuchsian group into a topological Jordan
curve with remarkable properties. This “so called curve” in the words of Poincaré
is now understood to have very nice conformally self-similar properties. We give
below the formal definitions (2.3, 2.4) of Kleinian, Fuchsian and quasi-Fuchsian
groups and work with intrinsic properties of the Kleinian groups with no mention
of the deformation.
We let SL(2, C) be the group of all 2 × 2 complex matrices with determinant 1.
We identify the group PSL(2, C) = SL(2, C)/{±1} and its action on the complex
sphere C̄ (see [26]) by fractional linear transformations. The element
 
g11 g12 g11 z + g12
g= ∈ SL(2, C) represents the mapping z → , z ∈ C̄.
g21 g22 g21 z + g22
The following definition of a Kleinian group is taken from [27] II.A. We refer the
reader to [27] for more advanced properties of Kleinian groups.
Definition 2.3. Let G ⊂ PSL(2, C) be a discrete subgroup. We say that
(a) G is freely discontinuous at the point z ∈ C̄ if there exists a neighborhood U ∋ z
such that g(U ) ∩ U = ∅ for every 1 6= g ∈ G.
(b) G is Kleinian if it is freely discontinuous at some point z ∈ C̄.
The set of all points z ∈ C̄ at which G is not freely discontinuous is called the
limit set of G and is denoted by Λ(G). This set is either infinite or consists of 0,
1 or 2 points. The latter 3 cases correspond to the so-called elementary Kleinian
groups, which are usually dropped from the consideration.
The definition below can be found in [27] on p. 103 and p. 192, respectively6.
Definition 2.4. A Kleinian group G is called
(a) Fuchsian (of the first kind) if its limit set is a circle.
(b) quasi-Fuchsian if its limit set is a closed Jordan curve.
It is known that a limit set of a finitely generated quasi-Fuchsian group (which
is not Fuchsian) has Hausdorff dimension strictly greater than 1 (see Corollary 1.7
in [12]).
It is known that (C̄\Λ(G))/G is a Riemann surface for an arbitrary Kleinian
group G. The following definition is taken from [12].
Definition 2.5. A Kleinian group G is called analytically finite if its Riemann
surface (C̄\Λ(G))/G is of finite type; i.e., a finite union of compact surfaces with
at most finitely many punctures and branch points.
We need the important notion of a p−dimensional geometric measure on C̄.
Definition 2.6. Let G be a Kleinian group. The measure ν on C̄ is called p−dimen-
sional geometric (relative to G) if d(ν ◦ g)(z) = |g ′ |p (z)dν(z) for every g ∈ G.
6More precisely what we call “quasi-Fuchsian”corresponds to “quasi-Fuchsian of the first kind”
6

An important condition for existence and uniqueness of geometric measures can

be found in [35] (see Theorem 1 there). Our proof of Theorem 1.1 (b) also delivers,
via the Riesz Representation Theorem, the existence of a p−dimensional geometric
measure concentrated on Λ(G) (for the case when p is the Hausdorff dimension of
Λ(G)).
A subgroup in G is called parabolic if it fixes exactly one point in C̄.
The notion of a fundamental domain F ⊂ C̄ of a Kleinian group G is defined in
[27], II.G. In particular, the sets {gF}g∈G , are pairwise disjoint.
We also need the notion of the Hausdorff dimension of a set X ⊂ C (applied to
the set Λ(G) in this text).
Definition 2.7. We say that the Hausdorff dimension of a set X ⊂ C does not
exceed q if there exist balls B(ai , ri ) such that
X q
X ⊂ ∪i B(ai , ri ), ri < ∞.
i
The infimum of all such q is called the Hausdorff dimension of a set X ⊂ C.
Remark 2.8. In what follows, we may assume without loss of generality that our
group G does not contain elliptic elements. By Selberg’s Lemma, there is a torsion-
free subgroup G0 ⊂ G which has finite index in G. The limit set of G0 is the limit set
of G. Since every finite index subgroup in a finitely generated group is itself finitely
generated (see p. 55 in [31]), it follows that the conditions of Theorem 1.1 hold for
the group G0 . The proof of this theorem constructs a geometric measure for the
subgroup of PSL(2, C) of invariance of the limit set of G0 and hence for the group
G. Moreover the uniqueness of the geometric measure for G0 implies uniqueness for
G. In addition to that, the group G0 does not contain elliptic elements. Indeed,
an elliptic element is conjugate in PSL(2, C) to a unitary element. Since G0 is
discrete, it follows that every elliptic element has finite order; since G0 is torsion
free, it follows that there are no elliptic elements.
This remark was written for the reason that some authors do not allow branches
in the Riemann surfaces. It is sometimes hard to check whether a particular paper
in the reference allows branches or not. The Riemann surface of a Kleinian group
without elliptic elements does not have branches, which makes it easier for the
reader.
2.5. Action of PSL(2, C) on hyperbolic space. Let us briefly recall how the
group PSL(2, C) acts on the three dimensional hyperbolic space. We refer the
reader to Section 1.2 in [19] for details.
By definition, the unit ball model B of hyperbolic space is the open unit ball of
R3 equipped with the following Riemannian metric.
(du0 )2 + (du1 )2 + (du2 )2
ds2 = 4 , u = (u0 , u1 , u2 ) ∈ B.
(1 − u20 − u21 − u22 )2
The Riemannian metric generates a distance in B. We do not need the (complicated)
distance formula, but only the fact that (see formula (2.5) on p. 10 in [19])
1 + |u|
(2.4) dist(u, 0) = log( ), u ∈ B.
1 − |u|
Here, u = (u0 , u1 , u2 ) is identified with the quaternion u0 +u1 i+u2j and |u| denotes
the norm of the quaternion (which coincides with the Euclidean norm of u).
 7

For a matrix g ∈ SL(2, C), consider the matrix π(g) of quaternions defined as
follows
a c′
     
1 1 −j 1 j
(2.5) π(g) = g = , |a|2 − |c|2 = 1.
2 −j 1 j 1 c a′

Here, the quaternions a and c are given by the following formulae.

1 1 1 1
(2.6) a= (g11 + ḡ22 ) + (g12 − ḡ21 )j, c= (g21 + ḡ12 ) + (g22 − ḡ11 )j.
2 2 2 2
Note that |a|2 − |c|2 = 1. The operation a → a′ is the inner automorphism imple-
mented by the quaternion k, it acts as follows

(a0 + a1 i + a2 j + a3 k)′ = a0 − a1 i − a2 j + a3 k, ∀aj ∈ R.

The action of the group SL(2, C) on B is given by the formula

(2.7) π(g) : u → (au + c′ )(cu + a′ )−1 , u ∈ B.

By Proposition 1.2.3 in [19], this action consists of isometries of B. Formulae (2.4),

(2.6) and (2.7) are crucially used in the proof of Lemma 3.1 below.

2.6. Bochner integration. The following definition of measurability can be found

e.g. in [22] (see Definition 3.5.4 there).

Definition 2.9. Let X be a Banach space. A function f : (−∞, ∞) → X is called

(a) strongly measurable if there exists a sequence of X-valued simple functions con-
verging to f almost everywhere.
(b) weakly measurable if the mapping s → hf (s), yi is measurable for every y ∈ X ∗ .

If the Banach space X is separable, then the Pettis Measurability Theorem (see
e.g. Theorem 3.5.3 in [22]) states the equivalence of the notions above.
A strongly measurable function f is Bochner integrable if
Z ∞
(2.8) kf (s)kX ds < ∞.
−∞

Theorem 3.7.4 in [22] states that there exists a sequence {fn }n≥0 of simple X-valued
functions such that
Z ∞
k(fn − f )(s)kX ds → 0, n → ∞.
−∞

The Bochner integral is now defined as

Z ∞ Z ∞
def
f (s)ds = lim fn (s)ds.
−∞ n→∞ −∞

Its key feature is that

Z ∞ Z ∞
f (s)ds ≤ kf (s)kX ds.


−∞ X −∞
8

2.7. Weak integration in L(H). The following definitions (and subsequent con-
struction of a weak integral) are folklore. For example, one can look at p. 77 in
[32] and put the topological space X there to be L(H) equipped with the strong
operator topology. Every functional on X can be written as a linear combination
of x → hxξ, ηi, ξ, η ∈ H.
Definition 2.10. A function s → f (s) with values in L(H) is measurable in the
weak operator topology if, for every vectors ξ, η ∈ H, the function
s → hf (s)ξ, ηi, s ∈ R,
is measurable.
For such functions, there is notion of weak integral. Note that the scalar-valued
mapping
s→ sup hf (s)ξ, ηi = kf (s)k∞ , s ∈ R,
kξk,kηk≤1

is measurable.
Let the function f : R → L(H) be measurable in the weak operator topology.
We say that f is integrable in the weak operator topology if
Z
(2.9) kf (s)k∞ ds < ∞.
R

Define a sesquilinear form

Z
(ξ, η) → hf (s)ξ, ηids, ξ, η ∈ H.
R

It is immediate that
Z
|(ξ, η)| ≤ kf (s)k∞ ds · kξkkηk, ξ, η ∈ H.
R

That is, for a fixed ξ ∈ H, the mapping η → (ξ, η) defines a bounded anti-linear
functional on H. It follows from the Riesz Lemma (description of the dual of a
Hilbert space) that there exists an element xξ ∈ H such that (ξ, η) = hxξ , ηi. The
mapping ξ → xξ is linear and bounded. The operator which maps ξ to xξ is called
the weak integral of the mapping s → f (s), s ∈ R.
The so-defined weak integral satisfies the following properties.
(a) If the mapping s → f (s) is integrable in the weak operator topology, then
Z ∞ Z ∞
f (s)ds ≤ kf (s)k∞ ds.


−∞ ∞ −∞

(b) If the mapping s → f (s) is integrable in the weak operator topology and if
A ∈ L(H), then s → A · f (s) is also integrable in the weak operator topology
and Z Z
A · f (s)ds = A · f (s)ds.
R R
(c) If the mapping s → f (s) is Bochner integrable in some Banach ideal in L(H),
then it is integrable in the weak operator topology. Its Bochner integral then
equals to the weak one.
 9

2.8. Double operator integrals. Here, we state the definition and basic proper-
ties of Double 0perator Integrals which were developed by Birman and Solomyak in
[7, 8, 9]. We refer the reader to [30] for the proofs and for more advanced properties.
Heuristically, the double operator integral TφX,Y , where X and Y are self-adjoint
operators and φ is a bounded Borel measurable function on Spec(X) × Spec(Y ), is
defined using the spectral decompositions:
ZZ
X,Y
Tφ (A) = φ(λ, µ)dEX (λ)AdEY (µ).

This formula defines a bounded operator from L2 to L2 . However, we want to

consider it as a bounded operator on other ideals — and this leads to difficulty
unless the function φ is good enough.
To specify the class of “good” functions, we use the integral tensor product of
[29], of L∞ (Spec(X), µX ) by L∞ (Spec(Y ), µY ) where the µ’s denote the spectral
measures. The integral projective tensor products were introduced in [29] where
it was proved that the maximal class of functions for which the double operator
integrals can be defined for arbitrary bounded linear operators coincides with the
integral projective tensor product of L∞ (Spec(X), µX ) by L∞ (Spec(Y ), µY ). Thus,
we consider only those functions φ which admit a representation
Z
(2.10) φ(λ, µ) = a(λ, s)b(µ, s)dκ(s),
Ω
where (Ω, κ) is a measure space and where
Z
(2.11) sup |a(λ, s)| · sup |b(µ, s)|dκ(s) < ∞.
Ω λ∈Spec(X) µ∈Spec(Y )

For those functions, we write

Z
(2.12) TφX,Y (A) = a(X, s)Ab(Y, s)dκ(s),
Ω
where the latter integral is understood in the weak sense (the integrand is measur-
able in the weak operator topology and the condition (2.9) holds thanks to (2.11)).
For the function φ from the integral tensor product, we have (see Theorem 4
in [30]) that TφX,Y : L1 → L1 and TφX,Y : L∞ → L∞ . In particular, we have that
TφX,Y : Lp,∞ → Lp,∞ for p > 1.
One of the key properties of Double Operator Integrals is that they respect
algebraic operations (see e.g. Proposition 2.8 in [28] or formula (1.6) in [11]).
Namely,
(2.13) TφX,Y
1 +φ2
= TφX,Y
1
+ TφX,Y
2
, TφX,Y
1 ·φ2
= TφX,Y
1
◦ TφX,Y
2
.
2.9. Fredholm modules. The following is taken from [14].
Definition 2.11. Let A be a ∗−algebra represented on the Hilbert space H. Let
F ∈ L(H) be self-adjoint unitary operator. We call a triple (F, H, A) Fredholm
module if [F, a] is compact for every a ∈ A.
The infinitesimal [F, a] is called the quantum derivative of the element a (see
Chapter IV in [14] for the studies of quantum derivatives).
A Fredholm module is called (p, ∞)−summable if [F, a] ∈ Lp,∞ for every a ∈ A.
Part (a) of Theorem 1.1 exactly states that the Fredholm module (F, L2 (S1 ), A)
is (p, ∞)−summable, where A is the ∗−algebra generated by Z.
10

3. Proof of Theorem 1.1 (a)

3.1. Growth of matrix coefficientsPin G. Let G be a Kleinian group. As stated
in Corollary II.B.7 in [27], the series g∈G |g ′ (z)|2 converges for a.e. z ∈ C̄ (with
respect to the Lebesgue measure). The critical exponent of G is defined7 (see e.g.
p. 323 in [14]) as follows
X
p = inf{q : |g ′ (z)|q < ∞ for a.e. z ∈ C̄}.
g∈G

Let kgk∞ denote the uniform norm of the matrix g ∈ SL(2, C) as an operator
on the Hilbert space C2 . Equip our countable group G with counting measure and
define lp,∞ (G) as in Subsection 2.2.
Lemma 3.1. Let G ⊂ PSL(2, C) be a Kleinian group. If p is its critical exponent,
then {kgk−2
∞ }g∈G ∈ lp,∞ (G).

Proof. By Corollary 5 in [34] (see also the right hand side estimate in Corollary 10
in [34]), we have
Card({g ∈ G : dist((π(g))(0), 0) ≤ r}) ≤ Cepr .
Using the formula (2.4) and denoting e−r by t, we arrive at
1 − |(π(g))(0)|
Card({g ∈ G : ≥ t}) ≤ Ct−p .
1 + |(π(g))(0)|
Since |(π(g))(0)| < 1, it follows that
Card({g ∈ G : 1 − |(π(g))(0)|2 ≥ 4t}) ≤ Ct−p .
Since |a′ | = |a| and |c′ | = |c|, it follows from (2.7) that
|c|2 1
(π(g))(0) = c′ (a′ )−1 and, therefore, 1 − |(π(g))(0)|2 = 1 − = 2.
|a|2 |a|
Thus,
1
Card({g ∈ G : ≥ 4t}) ≤ Ct−p .
|a|2
It is immediate from (2.6) that |a| ≤ 2kgk∞ . Therefore,
1
Card({g ∈ G : ≥ 4t}) ≤ Ct−p .
4kgk2∞
This concludes the proof. 
By Theorem II.B.5 in [27], g21 6= 0 for every 1 6= g ∈ G. This allows us to state
a stronger version of Lemma 3.1.
Lemma 3.2. Let G be a Kleinian group and let p be the critical exponent of G. If
∞ is not in the limit set of G, then {|g21 |−2 }16=g∈G ∈ lp,∞ (G).
Proof. By the assumption, ∞ ∈ / Λ(G). Hence, G is freely discontinuous at ∞.
It follows that {g(∞)}16=g∈G is a bounded set. Note that g(∞) = gg21 11
. Thus,
|g11 | = O(|g21 |).
Clearly,  
−1 g22 −g12
g = .
−g21 g11
7Sullivan uses a slightly different definition in [34], but they are equivalent.
 11

Applying the preceding paragraph to the element g −1 , we conclude that |g22 | =

O(|g21 |).
By Theorem II.B.5 in [27], the sequence {|g21 |}16=g∈G is bounded from below.
Thus,
g11 g22 − 1 |g11 | · |g22 | 1
|g12 | = | |≤ + = O(|g21 |) + O(1) = O(|g21 |).
g21 |g21 | |g21 |
Combining the estimates in the preceding paragraphs, we conclude that kgk∞ =
O(|g21 |). The assertion follows from Lemma 3.1. 
The following lemma provides the converse to Lemma 3.2 (under additional
assumptions on the group G).
Lemma 3.3. Let G ⊂ PSL(2, C) be as in Theorem 1.1. There exists C > 0 such
that n 1 o  
1 ≤ Cµ {|g21 |−2 }16=g∈G .
(k + 1) p k≥0
Proof. By Theorem 4 of [3] the group G is a quasiconformal deformation of a
Fuchsian group of the first kind. In particular, its limit set Λ(G) is a quasi-circle.
By Theorem 12 in [20], the Hausdorff dimension of Λ(G) is strictly less than 2. The
group G is finitely generated and thus by the Ahlfors Finiteness Theorem, G is
analytically finite. It follows now from Theorem 1.2 in [12] that G is geometrically
finite. Theorem 1 in [35] states that the critical exponent equals p. It is proved in
[5] that a geometrically finite Kleinian group without parabolic elements is convex
co-compact. Thus, the results of Section 3 in [34] are applicable.
By the left hand side estimate in Corollary 10 in [34]), we have
Card({g ∈ G : dist((π(g))(0), 0) ≤ r}) ≥ Cepr .
Using the formula (2.4) and denoting e−r by t, we arrive at
1 − |(π(g))(0)|
Card({g ∈ G : ≥ t}) ≥ Ct−p .
1 + |(π(g))(0)|
Since |(π(g))(0)| < 1, it follows that
Card({g ∈ G : 1 − |(π(g))(0)|2 ≥ t}) ≥ Ct−p .
Since |a′ | = |a| and |c′ | = |c|, it follows that
(2.7) |c|2 (2.5) 1
(π(g))(0) = c′ (a′ )−1 and, therefore, 1 − |(π(g))(0)|2 = 1 − = .
|a|2 |a|2
Thus,
Card({g ∈ G : |a|2 ≤ t−1 }) ≥ Ct−p .
We infer from (2.6) that
4|a|2 = |g11 + ḡ22 |2 + |g12 − ḡ21 |2 , 4|c|2 = |g21 + ḡ12 |2 + |g22 − ḡ11 |2 .
By the parallelogram rule, we have
8|a|2 ≥ 4|a|2 + 4|c|2 = 2|g11 |2 + 2|g22 |2 + 2|g12 |2 + 2|g21 |2 ≥ 2|g21 |2 .
It follows that
1
Card({g ∈ G : |g21 |2 ≤ t−1 }) ≥ Ct−p .
4
This concludes the proof. 
12

3.2. When does the quantum derivative fall into Lp,∞ ?. In this subsection,
we find a sufficient condition for the quantum derivative to belong to the ideal Lp,∞ ,
p > 1. A similar result for the ideal Lp is available as Theorem 4 and Proposition
5 on p. 316 in [14]. We get the required estimate by real interpolation.
Let α 6= −1 and let να be the measure on D defined by the formula
dνα (z) = |α + 1|(1 − |z|2 )α dm(z),
where m is the normalised Lebesgue measure on D. For α > −1, this is a finite
measure space; for α < −1, this is infinite measure space. Let Hol(D) be the space
of all holomorphic functions on D. The symbol [·, ·]θ,∞ denotes the functor of real
interpolation (see e.g. Definition 2.g.12 in [24]).
Lemma 3.4. If 1 < p0 < 2, then
[Lp0 (D, νp0 −2 ) ∩ Hol(D), L2 (D, ν0 ) ∩ Hol(D)]θ,∞ =
= [Lp0 (D, νp0 −2 ), L2 (D, ν0 )]θ,∞ ∩ Hol(D).
Proof. Clearly, L2 (D, ν0 ) ∩ Hol(D) is a closed subset in L2 (D, ν0 ). By Proposition
1.2 in [21], Lp0 (D, νp0 −2 ) ∩ Hol(D) is a closed subspace in Lp0 (D, νp0 −2 ), so that left
hand side is well defined.
The following map (see Proposition 1.4 in [21]) is called Bergman projection.
f (w)dν0 (w)
Z
(P0 f )(z) = 2
, z ∈ C.
D (1 − z w̄)
By Theorem 1.10 in [21], we have that
P0 : Lp0 (D, νp0 −2 ) → Lp0 (D, νp0 −2 ) ∩ Hol(D)
is a bounded mapping. Also, by Theorem 1.10 in [21], we have that
P0 : L2 (D, ν0 ) → L2 (D, ν0 ) ∩ Hol(D)
is a bounded mapping.
Therefore, for the left hand side of the equality in the statement of Lemma 3.4,
we have
LHS = [P0 (Lp0 (D, νp0 −2 )), P0 (L2 (D, ν0 ))]θ,∞ =
= P0 ([Lp0 (D, νp0 −2 ), L2 (D, ν0 )]θ,∞ ) = [Lp0 (D, νp0 −2 ), L2 (D, ν0 )]θ,∞ ∩ Hol(D).

The following lemma describes the class of functions f on the unit circle ∂D
for which its quantum derivative belongs to the weak ideal Lp,∞ , p > 1. Here, the
function space Lp,∞ (D, ν−2 ) is defined in Subsection 2.2.
Lemma 3.5. Suppose f : ∂D → C has an extension to an analytic function on D.
For p > 1, we have
k[F, f ]kp,∞ ≤ cp khkLp,∞ (D,ν−2 ) ,
where h(z) = (1 − |z|2 )|f ′ (z)|, z ∈ D.
Proof. Let Cp be the collection of all f : D → C such that the mapping z →
(1 − |z|2 )f (z), z ∈ D, belongs to the space Lp,∞ (D, ν−2 ). If p1 = 1−θ θ
p0 + 2 , then

[Lp0 (D, νp0 −2 ), L2 (D, ν0 )]θ,∞ = Cp .

Let 1
App = {f ∈ Hol(D) : f ′ ∈ Lp (D, νp−2 )}
 13

and let
Dp = {f ∈ Hol(D) : f ′ ∈ Cp }.
It follows from Lemma 3.4 that
1 1
[App00 , A22 ]θ,∞ = Dp .
By Theorem 4 and Proposition 5 on p. 316 in [14], we have
1
[F, f ] ∈ Lp ⇐= f ∈ App , 1 < p < ∞.
1 1
Applying real interpolation method to the Banach couples (App00 , A22 ) and (Lp0 , L2 ),
we infer
k[F, f ]kp,∞ = k[F, f ]k[Lp0 ,L2 ]θ,∞ ≤ cp kf k p
1 1 = kf kDp .
[Ap00 ,A22 ]θ,∞


3.3. Proof of Theorem 1.1 (a). We are now ready to prove the first part of our
main result.
Proof of Theorem 1.1 (a). As explained in the (first few lines of the) proof of Lemma
3.3, the group G is geometrically finite. By Theorem 1 in [35], the critical exponent
δ equals to the Hausdorff dimension p of Λ(G). Note that p > 1 by Theorem 2 in
[6].
Consider G acting on Σint . Let π be the action of G on the unit disk by the
formula
(3.1) g ◦ Z = Z ◦ π(g).
Since every π(g) is a conformal automorphism of the unit disk, it is automatically
fractional linear (see [26]). Thus, π(G) is a group of fractional linear transformations
preserving the unit circle, i.e. a Fuchsian group and it’s limit set is the unit circle
∂D, thus it is Fuchsian of the first kind. As a group, π(G) is isomorphic to G and
is, therefore, finitely generated.
We claim that the Fuchsian group π(G) does not contain parabolic elements.
Assume the contrary: let g ∈ G be such that π(g) is parabolic. Hence, there exists
a fixed point w0 ∈ ∂D of π(g) such that (π(g))n w → w0 as n → ±∞ for every
w ∈ D. Let w = Z(z), z ∈ Σint , and let w0 = Z(z0 ), z0 ∈ Λ(G). By (3.1), we have
that g n (z) → z0 as n → ±∞. Hence, g ∈ G is parabolic,8 which is not the case.
Since π(G) is finitely generated and of the first kind, it follows from Theorem
10.4.3 in [1] that the Riemann surface D/π(G) has finite area. Taking into account
that π(G) does not have parabolic elements, we infer from Corollary 4.2.7 in [23]
that the Riemann surface D/π(G) is compact. By Corollary 4.2.3 and Theorem
3.2.2 in [23], π(G) admits a fundamental domain F which is compactly supported
in D.
Step 1: We claim that there exists a finite constant such that for every g ∈ G,
const
sup (1 − |z|2 )|Z ′ (z)| ≤ .
z∈π(g)F |g21 |2

8 An element g ∈ PSL(2, C) is either parabolic or diagonalizable. If g is diagonalizable, then

(after conjugating g by a fractional linear transform), we have that g : z → az for every z ∈ C. If
|a| < 1, then g n (z) → 0 as n → ∞ and g n (z) → ∞ as n → −∞ for every 0 6= z ∈ C. If |a| > 1,
then g n (z) → 0 as n → −∞ and g n (z) → ∞ as n → ∞ for every 0 6= z ∈ C. If |a| = 1 and a 6= 1,
then the sequence {g n (z)}n∈Z diverges as n → ∞ and as n → −∞ for every 0 6= z ∈ C.
14

Indeed, we have z = π(g)w, where w ∈ F. We have9

1 − |z|2 = (1 − |w|2 )|(π(g))′ (w)|.
It follows from the chain rule that
(1 − |z|2 )|Z ′ (z)| = (1 − |w|2 ) · |Z ′ (π(g)w)| · |(π(g))′ (w)| = (1 − |w|2 ) · |(Z ◦ π(g))′ (w)|.
It follows from (3.1) and chain rule that
(3.1)
(3.2) (1 − |z|2 )|Z ′ (z)| = (1 − |w|2 ) · |(g ◦ Z)′ (w)| = |g ′ (Z(w))| · (1 − |w|2 )|Z ′ (w)|.
Since g ′ (u) = (g21 u + g22 )−2 and g −1 (∞) = − gg21
22
, it follows that
1 1 1
(3.3) |g ′ (Z(w))| = 2
= · .
|g21 Z + g22 | |g21 | |Z(w) − g −1 (∞)|2
2

Thus, for z ∈ π(g)F, we have, since g −1 (∞) stays in the unbounded component of
the complement of the limit set Λ(G) and thus |Z(w)−g −1 (∞)| ≥ dist(Z(F), Λ(G)),
1 1
(1 − |z|2 )|Z ′ (z)| ≤ 2
· 2 · sup(1 − |w|2 )|Z ′ (w)|.
|g21 | dist (Z(F), Λ(G)) w∈F
Since F is compact and Z ′ |F is continuous, the claim follows.
Step 2: Let h(z) = (1 − |z|2 )|Z ′ (z)| (see also the statement of Lemma 3.5). It
follows from Step 1 that
X 1
khkLp,∞(D,ν−2 ) ≤ khχF kLp,∞ (D,ν−2 ) + const · k χπ(g)F kLp,∞ (D,ν−2 ) .
|g21 |2
16=g∈G

Recall that F is compactly supported in D and, therefore, ν−2 (F) < ∞. Let ν−2 (F) =
a. Elements of the group π(G) are conformal automorphisms of the unit disk;
hence, isometries of the hyperbolic plane H2 . The measure ν−2 is a volume form
of H2 and is, therefore, invariant with respect to its isometries. Hence, ν−2 is
π(G)−invariant.10 It follows that
(3.4) ν−2 (π(g)F) = a, for every g ∈ G.
Thus,
X 1 n 1 o 
µ( χ
2 π(g)F
)=µ ⊗ χ(0,a) ,
|g21 | |g21 |2 16=g∈G
16=g∈G
where µ on the left hand side is computed in the measure space (D, ν−2 ) and µ on
the right hand side is computed in the algebra (G × (0, ∞), Card × m). Hence,
n 1 o
khkLp,∞(D,ν−2 ) ≤ khχF k∞ kχ(0,a) kp,∞ + const · ⊗ χ .

(0,a)
|g21 |2 16=g∈G p,∞

9This is a standard fact. Let k : w → αw+β , |α|2 − |β|2 = 1 be an arbitrary conformal

β̄w+ᾱ
automorphism of the unit disk. We have
|dk(w)| |β̄w + ᾱ|−2 |dw| |dw|
= |dw| = = .
1 − |k(w)| 2 |αw+β|2 | β̄w + ᾱ| 2 − |αw + β|2 1 − |w|2
1 − |β̄w+ᾱ|2

10This fact can also be seen directly as follows. Let k : z → αz+β , |α|2 − |β|2 = 1 be an
β̄z+ᾱ
arbitrary conformal automorphism of the unit disk. Its Jacobian is exactly |k ′ (z)|2 . Thus,
d(m ◦ k)(z) |k ′ (z)|2 dm(z)
d(ν−2 ◦ k)(z) = 2 2
= dm(z) = = dν−2 (z).
(1 − |k(z)| ) (1 − |k(z)|2 )2 (1 − |z|2 )2
This shows conformal invariance of the measure ν−2 .
 15

It follows now from Lemma 3.2 that h ∈ Lp,∞ (D, ν−2 ). The assertion follows now
from Lemma 3.5. 
The next lemma is the core part of the proof of Theorem 1.1 (c). Its proof is
similar to that of Theorem 1.1 (a).
Lemma 3.6. If G is as in Theorem 1.1, then
lim inf sk[F, Z]kp+s > 0.
s→0

Proof. Let h(z) = (1 − |z| )|Z ′ (z)|, z ∈ D. For every 1 6= g ∈ G, it follows from
2

(3.2) and (3.3) (in the proof of Theorem 1.1 (a)) that
1 1
inf (1 − |z|2 )|Z ′ (z)| ≥ · inf (1 − |w|2 )|Z ′ (w)| · inf ≥
z∈π(g)F |g21 |2 w∈F w∈F |Z(w) − g −1 (∞)|2
1 1 const
≥ · inf (1 − |w|2 )|Z ′ (w)| · ≥ .
|g21 |2 w∈F (kZk∞ + |g −1 (∞)|)2 |g21 |2
We have ν−2 (π(g)F) = a for every g ∈ G. Since the sets {π(g)F}g∈G are pairwise
disjoint, it follows that
khkLp+s(D,ν−2 ) ≥ const · k{|g21 |−2 }16=g∈G kp+s .
We infer from Lemma 3.3 that
1 const
k{|g21 |−2 }16=g∈G kp+s ≥ const · k{(k + 1)− p }k≥0 kp+s ≥ , s ↓ 0.
s
By Proposition 5 on p. 316 in [14], we have
const
kZk 1 ≥ cp khkLp+s(D,ν−2 ) ≥ , s ↓ 0.
p+s
Ap+s s
Since Z is an analytic function on D, it follows from Theorem 4 on p. 316 in [14]
that
const
k[F, Z]kp+s ≥ c−1
p kZk p+s1 = c−1
p kZk p+s 1 ≥ , s ↓ 0.
Bp+s Ap+s s
This completes the proof. 

4. Integration in (Lp,∞ )0 , p > 1.

Lemma 4.1. Let s → Z(s) be a bounded function from R to (Lp,∞ )0 . If it is
measurable in the weak operator topology, then it is weakly measurable11 in (Lp,∞ )0 .
Proof. Let γ be a bounded linear functional on (Lp,∞ )0 . By the noncommutative
Yosida-Hewitt theorem (see [17]), we have that γ extends to a normal functional
on Lp,∞ . Let Lq,1 be the Lorentz space which is the Köthe dual12 of Lp,∞ . There
exists x ∈ Lq,1 such that
γ(y) = Tr(xy), y ∈ Lp,∞ .
1
Fix n ∈ N and choose a finite rank operator xn such that kx − xn kq,1 < n. By
assumption, the scalar valued function
fn : s → Tr(xn Z(s)), s ∈ R,

11See Definition 2.9

12See [17] for the definition and basic properties of Köthe duals.
16

is measurable. On the other hand, we have

|f − fn |(s) ≤ kZ(s)kp,∞ kx − xn kq,1
and, therefore,
1
sup kZ(s)kp,∞ .
kf − fn k∞ ≤
n s∈R
Hence, fn converges to f uniformly. Since the limit of a sequence of measurable
functions is measurable, the weak measurability of the mapping s → Z(s) follows.

Lemma 4.2. Let s → Z(s) be a bounded function from R to (Lp,∞ )0 which is
measurable in the weak operator topology. If
Z
(4.1) kZ(s)kp,∞ ds < ∞,
R
then s → Z(s) is a Bochner integrable function from R to (Lp,∞ )0 . We have
Z
(4.2) Z(s)ds ∈ (Lp,∞ )0 .
R

Proof. By Lemma 4.1 the mapping s → Z(s) is weakly measurable from R to

(Lp,∞ )0 . Since (Lp,∞ )0 is separable, it follows from Theorem 3.5.3 in [22] that
the mapping s → Z(s) is strongly measurable from R to (Lp,∞ )0 (in the sense of
Definition 3.5.4 in [22]). Using Theorem 3.7.4 in [22] and (4.1), we obtain that the
mapping s → Z(s) is Bochner integrable from R to (Lp,∞ )0 . The inclusion (4.2)
follows now from the definition of Bochner integral (see Definition 3.7.3 in [22]). 
In what follows, we use the notation Az for the complex power of a positive
operator A ∈ L(H) defined as follows for z ∈ C of positive real part: ℜ(z) ≥ 0. Let
fz : [0, ∞) → C be the Borel function given by the formula
(
ez log(x) , x > 0
fz (x) =
0, x = 0.
We set Az = fz (A), where the right hand side is defined by means of the functional
calculus. In particular this defines the imaginary power Ais = fis (A) for s ∈ R.
′ ′
One has Az+z = Az Az for z, z ′ ∈ C of positive real part ℜ(z) ≥ 0, ℜ(z ′ ) ≥ 0.
One has fz (xy) = fz (x)fz (y) for ℜ(z) ≥ 0 and x, y ≥ 0. Thus using the conven-
tion 0z = 0 for z ∈ C, ℜ(z) ≥ 0, (in particular 0is = 0) one has the formula
λis Ais = (λA)is , s ∈ R, λ ≥ 0, 0 ≤ A ∈ L(H),
which is used repeatedly in Lemmas 5.1 and 5.2.
Lemma 4.3. Let A1 , A2 , A3 ∈ L(H) be positive and let X1 , X2 , X3 , X4 ∈ L(H).
The mapping
s → X1 Ais is is
1 X2 A2 X3 A3 X4 , s ∈ R,
is measurable in the weak operator topology.
Proof. For every bounded positive operator A, the mapping s → Ais is strongly
continuous. Indeed, let logfin be a Borel function on [0, ∞) defined by the formula
(
log(x), x > 0
logfin (x) = .
0, x = 0
 17

We have that logfin (A) is an unbounded self-adjoint operator. Thus, the mapping
s → Ais = exp(is logfin (A)) · EA (0, ∞)
is strongly continuous by Stone’s theorem.
Thus, for arbitrary vectors ξ, η ∈ H, the mapping
s → hX1 Ais is is
1 X2 A2 X3 A3 X4 ξ, ηi, s ∈ R,
is continuous. In particular, the latter scalar-valued mapping is measurable and
our vector-valued mapping is measurable in the weak operator topology. 

5. Proof of the key “commutator”estimate

This section contains a modification of Lemma 11 stated on p. 321 in [14]. The
proofs here were obtained with the help of Denis Potapov.
In this section, integrals are understood in the weak sense (see Subsection 2.7)
unless explicitly specified otherwise.
Lemma 5.1. For every p > 1, there exists a Schwartz function h such that, for
every 0 ≤ X, Y ∈ L(H) we have
Z
X − Y = V − X is V Y −is h(s)ds.
p p
R
Here, V = X p−1 (X − Y ) + (X − Y )Y p−1 .
Proof. Define a function g by setting
p p
e 2 t − e− 2 t  p
g(t) = 1 − , t ∈ R, t 6= 0, g(0) = 1 − .
t
− 2t
(e − e )(e
2 ( p−1
2 )t
+e ) −( p−1
2 )t 2
It is an even function of t, it is smooth at t = 0 with Taylor expansion
 p 1 3
p − 3p2 + 2p t2 + · · ·


g(t) = 1 − +
2 24
and one has
e2t − ept
g(t) = t
(e − 1) (ept + et )
so that g = 0 for p = 2, and g(t) is equivalent to e(1−p)t when t → ∞ for p < 2,
and to −e−t for p > 2. Similarly all derivatives of g have exponential decay at ∞.
Thus g is a Schwartz function. Set h to be the Fourier transform of g, so that h is
also a Schwartz function. Set
λ
φ1 (λ, µ) = g(log( )) ∀λ, µ > 0, φ1 (0, µ) = 0, ∀µ ≥ 0, φ1 (λ, 0) = 0, ∀λ ≥ 0.
µ
So that our function φ1 is defined on [0, ∞) × [0, ∞). Note that it is not continuous
at (0, 0). One has
λp − µp
(5.1) φ1 (λ, µ) = 1 − , λ, µ > 0, λ 6= µ.
(λ − µ)(λp−1 + µp−1 )
We claim that
Z
(5.2) φ1 (λ, µ) = h(s)λis µ−is ds, λ, µ ≥ 0.
R
Indeed, we have Z
g(t) = h(s)eist ds, t ∈ R.
R
18

For λ, µ > 0, we set t = log( λµ ) and obtain

Z Z
λ
is log( µ )
φ1 (λ, µ) = h(s)e ds = h(s)λis µ−is ds
R R
For λ = 0 or µ = 0, the left hand side of (5.2) vanishes by the definition of φ1 ,
while the right hand side vanishes due to the convention 0is = 0. Thus, formula
(5.2) holds for all λ, µ ≥ 0. Set
φ2 (λ, µ) = (λp−1 + µp−1 )(λ − µ), λ, µ ≥ 0.
This function is bounded on Spec(X) × Spec(Y ) and the same holds for
φ3 (λ, µ) = (λp−1 + µp−1 )(λ − µ) − (λp − µp ), ∀λ, µ ≥ 0.
The equality φ3 = φ1 φ2 holds on [0, ∞) × [0, ∞). Indeed this follows from (5.1)
for λ, µ > 0, λ 6= µ. For λ = µ > 0 one has φ1 (λ, λ) = 1 − p2 , φ2 (λ, λ) = 0 and
φ3 (λ, λ) = 0. If λ = 0 or µ = 0 one has φ1 (λ, µ) = 0 and φ3 (λ, µ) = 0.
It follows from the definition (2.12) of Double Operator Integrals and X, Y ≥ 0,
Z
X,Y
(5.3) Tφ1 (A) = h(s)X is AY −is ds.
R
Indeed, since h is a Schwartz function, the condition (2.11) holds and, therefore,
(2.12) reads as (5.3). Here, the integral on the right hand side is understood in
the weak sense. Measurability of the integrand is guaranteed by Lemma 4.3 and
condition (2.9) follows from the inequality
kh(s)X is AY −is k∞ ≤ |h(s)| · kAk∞ , s ∈ R,
and from the fact that h is a Schwartz (and, hence, integrable) function. In partic-
ular, TφX,Y
1
: L∞ → L∞ .
Using formulae (2.10) and (2.12), we obtain that TφX,Y
2
: L∞ → L∞ and
TφX,Y
2
(A) = X p A − X p−1 AY + XAY p−1 − AY p .
The function φ3 bounded on Spec(X) × Spec(Y ), TφX,Y
3
: L∞ → L∞ and
TφX,Y
3
(A) = (X p A − X p−1 AY + XAY p−1 − AY p ) − (X p A − AY p ).
We have φ3 = φ1 φ2 on Spec(X) × Spec(Y ), and thus
(2.13)
TφX,Y
1
(V ) = TφX,Y
1
(TφX,Y
2
(1)) = TφX,Y
3
(1) = V − (X p − Y p ).
The assertion follows now from (5.3). 
Lemma 5.2 below can be proved without any compactness assumption on the
operator B; however, the proof becomes much harder. We impose compactness
assumption due to the fact that B is compact in Lemma 5.3 (the only place where
we use Lemma 5.2).
Lemma 5.2. Let 0 ≤ A, B ∈ L(H). If 1 < p < ∞ and if B is compact, then
Z
1 1
B p Ap − (A 2 BA 2 )p = ”T (0)” − T (s)h(s)ds,
R
1 1
where we denote, for brevity, Y = A 2 BA 2 while
1 1
T (s) = B p−1+is [B, Ap+is ]Y −is + B p−1+is Ap− 2 +is [A 2 , B]Y −is +
1 1
+B is [B, A1+is ]Y p−1−is + B is A 2 +is [A 2 , B]Y p−1−is .
 19

and
1 1 1 1
”T (0)” := B p−1 [B, Ap ] + B p−1 Ap− 2 [A 2 , B] + [B, A]Y p−1 + A 2 [A 2 , B]Y p−1 .
P
Proof. By assumption, B is compact and, therefore, one can write B = j λj pj ,
P
where {pj } is a family of mutually orthogonal projections such that j pj = 1. We
have X X
B p Ap − Y p = pj (B p Ap − Y p ) = pj ((λj A)p − Y p ).
j j

Applying Lemma 5.1 to the expression in the brackets, we obtain13

X Z
(5.4) B p Ap − Y p = pj (Vj − (λj A)is Vj Y −is h(s)ds),
j R

where,
Vj = (λj A)p−1 (λj A − Y ) + (λj A − Y )Y p−1 = (λj A)p − (λj A)p−1 Y + λj AY p−1 − Y p .
p p p−1 p−1
A Y + BAY p−1 − Y p = ”T (0)”.
P
Therefore, we get j pj Vj = B A − B
Moreover we have
X X  
pj (λj A)is Vj = pj (λj A)p+is −(λj A)p−1+is Y +(λj A)1+is Y p−1 −(λj A)is Y p =
j j

pj λp+is pj λp−1+is
X X
= j · Ap+is − j · Ap−1+is Y +
j j
X X
+ pj λ1+is
j · A1+is Y p−1 − pj λis is p
j ·A Y .
j j
By the functional calculus, we have
X
pj (λj A)is Vj = B p+is Ap+is −B p−1+is Ap−1+is Y +B 1+is A1+is Y p−1 −B is Ais Y p =
j

= B p−1+is (BAp+is − Ap−1+is Y ) + B is (BA1+is − Ais Y )Y p−1 =

= B p−1+is [B, Ap+is ] + B p−1+is Ap−1+is (AB − Y )+
+B is [B, A1+is ]Y p−1 + B is Ais (AB − Y )Y p−1 =
1 1
= B p−1+is [B, Ap+is ] + B p−1+is Ap−1+is A 2 [A 2 , B]+
1 1
+B is [B, A1+is ]Y p−1 + B is Ais A 2 [A 2 , B]Y p−1 .
Substituting the last equality into (5.4) completes the proof. 

The following lemma is the main result of this section. It provides the key
estimate used in the proof of Theorem 1.1 (b). In [14], the corresponding Lemma
3.β.11 is stated without a proof.
1
Lemma 5.3. Let 0 ≤ A ∈ L∞ and let 0 ≤ B ∈ Lp,∞ , 1 < p < ∞. If [A 2 , B] ∈
(Lp,∞ )0 , then
1 1
B p Ap − (A 2 BA 2 )p ∈ (L1,∞ )0 .

13In this and subsequent formulae, imaginary powers are defined as in Section 4. The conven-
tion 0is = 0 is used.
20

1 1
Proof. Consider the formula for B p Ap − (A 2 BA 2 )p obtained in Lemma 5.2. We
have
1 1
B p Ap − (A 2 BA 2 )p = ”T (0)” − B p−1 · (I + II) − (III + IV ) · Y p−1 ,
where
Z Z
1 1
I= B is [B, Ap+is ]Y −is h(s)ds, II = B is Ap− 2 +is [A 2 , B]Y −is h(s)ds,
R R
Z Z
1 1
is 1+is −is
III = B [B, A ]Y h(s)ds, IV = B is A 2 +is [A 2 , B]Y −is h(s)ds.
R R
Step 1: We show that I ∈ (Lp,∞ )0 .
Without loss of generality, 0 ≤ A ≤ 1. For a fixed s ∈ R, the function x →
xp+is can be uniformly approximated by polynomials fm on the interval [0, 1]. It is
immediate that
[B, Ap+is ] − [B, fm (A)] = B(Ap+is − fm (A)) − (Ap+is − fm (A))B.
Thus,
k[B, Ap+is ] − [B, fm (A)]kp,∞ ≤ 2kBkp,∞ kAp+is − fm (A)k∞ → 0, m → ∞.
1
Due to the assumption [A , B] ∈ (Lp,∞ )0 , we have
2

1 1 1 1
[B, A] = A 2 [B, A 2 ] + [B, A 2 ]A 2 ∈ (Lp,∞ )0 .
Thus,
k−1
X
[B, Ak ] = Al [B, A]Ak−1−l ∈ (Lp,∞ )0 .
l=0
It follows that [B, fm (A)] ∈ (Lp,∞ )0 . Thus,
(5.5) [B, Ap+is ] ∈ (Lp,∞ )0 .
By hypothesis, one has B ∈ Lp,∞ . We infer from 0 ≤ A ≤ 1 that Ap+is is a
contraction for every s ∈ R. Hence, we have
(5.6) k[B, Ap+is ]kp,∞ ≤ 2kBkp,∞ kAp+is k∞ ≤ 2kBkp,∞ .
It follows from Lemma 4.3 that the mapping
s → B is [B, Ap+is ]Y −is h(s), s ∈ R,
is measurable in the weak operator topology. Combining Lemma 4.2 and (5.6), we
infer that I ∈ (Lp,∞ )0 .
Step 2: By Step 1, we have that I ∈ (Lp,∞ )0 . Repeating the argument in
1
Step 1 for III and using [A 2 , B] ∈ (Lp,∞ )0 , for II and IV , we obtain that also
II, III, IV ∈ (Lp,∞ )0 .
The next assertion is similar to (2.3) and it follows immediately from Corollary
2.3.16.b in [25]: if X ∈ (Lp,∞ )0 and 0 ≤ Z ∈ Lp,∞ , then XZ p−1 ∈ (L1,∞ )0 and
Z p−1 X ∈ (L1,∞ )0 . Since B, Y ∈ Lp,∞ , it follows that
B p−1 · (I + II) ∈ (L1,∞ )0 , (III + IV ) · Y p−1 ∈ (L1,∞ )0 .
Also, we have by Lemma 5.2
1 1 1 1
”T (0)” = B p−1 [B, Ap ] + B p−1 Ap− 2 [A 2 , B] + [B, A]Y p−1 + A 2 [A 2 , B]Y p−1 .
 21

Setting s = 0 in (5.5), we obtain that [B, Ap ] ∈ (Lp,∞ )0 . By the commutator

assumption and Leibniz rule, we have
1 1 1 1
[B, A] = [B, A 2 ]A 2 + A 2 [B, A 2 ] ∈ (Lp,∞ )0 .
Since B, Y ∈ Lp,∞ , it follows that ”T (0)” ∈ (L1,∞ )0 .
Combining these results, we complete the proof. 

6. Proof of Theorem 1.1 (b)

For a detailed study of commutator estimates for the absolute value function,
we refer the reader to [16] or [13].
Lemma 6.1. Let A, B ∈ L(H). If [A, B] ∈ (Lp,∞ )0 and [A, B ∗ ] ∈ (Lp,∞ )0 then
[A, |B|] ∈ (Lp,∞ )0 .
Proof. For a self-adjoint B, the assertion is proved in [16]. Let B ∈ L(H) be
arbitrary and set    
A 0 0 B
C= , D=
0 A B∗ 0
We have  
0 [A, B]
[C, D] = ∈ (Lp,∞ )0 .
[A, B ∗ ] 0
Since D is self-adjoint, it follows from Theorem 3.4 in [16] that [C, |D|] ∈ (Lp,∞ )0 .
However,  ∗ 
|B | 0
|D| =
0 |B|
Thus,
[A, |B ∗ |]
 
0
[C, |D|] =
0 [A, |B|]
This concludes the proof. 
The following lemma is Proposition 10, part (3) on p. 320 in [14].
Lemma 6.2. If T, S ∈ Lp,∞ are such that T − S ∈ (Lp,∞ )0 , then |T |p − |S|p ∈
(L1,∞ )0 .
The following lemma crucially uses Lemma 5.3 from the preceding section. Recall
the lightened notation: the algebra L∞ (∂D) is identified with its natural action on
the Hilbert space L2 (∂D) by pointwise multiplication.
Lemma 6.3. Let f ∈ C(S1 ) be such that [F, f ] ∈ Lp,∞ . Let g ∈ SL(2, C) be such
that the function u = gg21
11 f +g12
f +g22 is well defined and bounded. We have

(6.1) |[F, u]|p ∈ |[F, f ]|p · |g21 f + g22 |−2p + (L1,∞ )0 .

Proof. Since u is bounded, it follows that f is separated from − gg21
22
∈ C̄. Thus,
−1 1
v = (g21 f + g22 ) ∈ C(S ). If g21 = 0, then the assertion is trivial. Further, we
assume that g21 6= 0. Clearly, u = gg21
11
− g121 v. Thus,
1 1
[F, u] = − [F, v] = · v[F, g21 f + g22 ]v = v[F, f ]v.
g21 g21
Therefore, we have
[F, u] = [F, f ]v 2 + [v, [F, f ]] · v.
22

Since v ∈ C(S1 ), it follows from Theorem 8 (a) on p. 319 in [14] that

[F, u] ∈ [F, f ]v 2 + (Lp,∞ )0 .
By Lemma 6.2, we have (everywhere in the proof below, LHS means the left hand
side of (6.1))  p
LHS ∈ [F, f ]v 2  + (L1,∞ )0 .
 

Equivalently,  p
LHS ∈ |[F, f ]|v 2  + (L1,∞ )0 .
 

Since v 2 ∈ C(S1 ), it follows from Theorem 8 (a) (on p. 319 in [14]) that
h i
[F, f ], v 2 ∈ (Lp,∞ )0 .
By Lemma 6.1, we have
h i
(6.2) |[F, f ]|, v 2 ∈ (Lp,∞ )0 .
It follows from Lemma 6.2 that
 p
LHS ∈ v 2 |[F, f ]| + (L1,∞ )0 .
 

Equivalently,  p
LHS ∈ |v|2 |[F, f ]| + (L1,∞ )0 .
 

Since |v| ∈ C(S1 ), it follows from Theorem 8 (a) (on p. 319 in [14]) that
h i
[F, f ], |v| ∈ (Lp,∞ )0 .
By Lemma 6.1, we have
h i
(6.3) |[F, f ]|, |v| ∈ (Lp,∞ )0 .
We have
|v|2 |[F, f ]| = |v| · |[F, f ]| · |v| − |v| · [|[F, f ]|, |v|].
Thus,
|v|2 |[F, f ]| ∈ |v| · |[F, f ]| · |v| + (Lp,∞ )0 .
It follows from Lemma 6.2 that
 p  p
 2
|v| · |[F, f ]| ∈ |v| · |[F, f ]| · |v| + (L1,∞ )0 .
  

Thus,  p
LHS ∈ |v| · |[F, f ]| · |v| + (L1,∞ )0 .
 

Set A = |v|2 and B = |[F, f ]|. We have

1 1
LHS ∈ (A 2 BA 2 )p + (L1,∞ )0 .
1
On the other hand, the equality (6.3) reads as follows: [B, A 2 ] ∈ (Lp,∞ )0 . It follows
now from Lemma 5.3 that
LHS ∈ B p Ap + (L1,∞ )0 .
This is exactly (6.1) and the proof is complete. 
We also need the following auxiliary lemma. Page 314 in [14] mentions a cor-
responding assertion for the Dirac operator on the line and the action of SL(2, R).
Those settings (and results) are unitarily equivalent.
 23

Lemma 6.4. The mapping h → Uh , h ∈ SU(1, 1), defined by the formula

αz + β 1
(Uh ξ)(z) = ξ( ) , ξ ∈ L2 (∂D), |z| = 1,
β̄z + ᾱ β̄z + ᾱ
where  
α β
h= , |α|2 − |β|2 = 1,
β̄ ᾱ
is a unitary representation of the group SU(1, 1) on the Hilbert space L2 (∂D) which
commutes with F.
Proof. The fact that h → Uh is a homomorphism is simple and we omit the proof.
First, we show this representation is unitary. Indeed, we have
1 1
Z
hUh ξ, Uh ξi = |ξ ◦ h|2 (eit ) · dt.
2π ∂D |β̄e + ᾱ|2
it

On the circle ∂D, we have

def αeit + β
h : eit → eis = .
β̄eit + ᾱ
Thus,
ds 1 d(eis ) β̄eit + ᾱ 1 1
= e−is · = · · ieit = .
dt i dt i(αeit + β) (β̄eit + ᾱ)2 |β̄eit + ᾱ|2
Thus,
1
Z
hUh ξ, Uh ξi = |ξ|2 (eis )ds = hh, hi.
2π ∂D
Thus, Uh is indeed a unitary operator.
def
Let P+ = ED [0, ∞). Let en (z) = z n , |z| = 1, n ∈ Z. If n ≥ 0, then
(αz + β)n β̄
(Uh en )(z) = = (ᾱ)−n−1 (αz + β)n (1 + z)−n−1 =
(β̄z + ᾱ)n+1 ᾱ
∞  
X −n − 1 β̄ m
= (ᾱ)−n−1 (αz + β)n ( z) .
m=0
m ᾱ
The series converges uniformly on the unit circle S1 because |β| < |α|. The series
contains only positive powers of z and, therefore, P+ Uh en = Uh en .
It follows from the preceding paragraph that P+ Uh P+ = Uh P+ . Taking the ad-
joint, we obtain P+ Uh−1 P+ = P+ Uh−1 . Replacing h with h−1 , we obtain P+ Uh P+ =
P+ Uh . Thus, P+ Uh = Uh P+ . It follows that Uh commutes with F. 

We are now ready to prove our main result.

Proof of Theorem 1.1 (b). Consider the linear functional on C(Λ(G)) defined by
the formula
f → ϕ((f ◦ Z) · |[F, Z]|p ), f ∈ C(Λ(G)),
where ϕ is a continuous trace on L1,∞ .
It follows from boundedness of ϕ and (2.2) that
|ϕ((f ◦ Z) · |[F, Z]|p )| ≤ kϕkL∗1,∞ kf ◦ Zk∞ k[F, Z]kpp,∞ .
24

Thus, our functional is bounded and, by the Riesz Representation Theorem, it

admits a representation of the form
Z
(6.4) ϕ((f ◦ Z) · |[F, Z]|p ) = f (t)dκ(t), f ∈ C(Λ(G)).
Λ(G)

Here, κ is some Radon measure on Λ(G).

We claim that
Z Z
(6.5) (f ◦ g −1 )(t)dκ(t) = f (t)|g ′ (t)|p dκ(t), f ∈ C(Λ(G)), g ∈ G.
Λ(G) Λ(G)

To see this, let π(G) ⊂ SU(1, 1) be the Fuchsian group as in the proof of part
(a). Let h → Uh be its unitary representation given in Lemma 6.4. It is immediate
that
Uπ(g) (ξ · η) = (ξ ◦ π(g)) · Uπ(g) (η), ξ ∈ L∞ (∂D), η ∈ L2 (∂D).
Thus,
−1 −1
Uπ(g) ZUπ(g) = Z ◦ π(g) = g ◦ Z, (f ◦ g −1 ◦ Z) = Uπ(g) (f ◦ Z)Uπ(g) .
Since Uπ(g) commutes with F, it follows from the preceding formula that
−1
(f ◦ g −1 ◦ Z)|[F, Z]|p = Uπ(g) (f ◦ Z)Uπ(g) |[F, Z]|p =
−1 −1 p −1
= Uπ(g) (f ◦ Z)|Uπ(g) [F, Z]Uπ(g) | · Uπ(g) = Uπ(g) (f ◦ Z)|[F, g ◦ Z]|p · Uπ(g) .
It follows from the unitary invariance of the trace ϕ that
ϕ((f ◦ g −1 ◦ Z) · |[F, Z]|p ) = ϕ((f ◦ Z) · |[F, g ◦ Z]|p ).
By Lemma 6.3 with f = Z, we have
|[F, g ◦ Z]|p ∈ |[F, Z]|p · (|g ′ |p ◦ Z) + (L1,∞ )0 .
Since ϕ vanishes on (L1,∞ )0 , it follows that
ϕ((f ◦ g −1 ◦ Z) · |[F, Z]|p ) = ϕ((f ◦ Z) · |[F, Z]|p · (|g ′ |p ◦ Z)) =
Z
′ p p (6.4)
= ϕ(((f |g | ) ◦ Z) · |[F, Z]| ) = f (t)|g ′ (t)|p dκ(t).
Λ(G)
This proves (6.5). In other words, κ is a geometric measure.
As explained in the (first few lines of the) proof of Lemma 3.3, the group G is
geometrically finite. Theorem 1 in [35] states that geometric (probability) measure
on Λ(G) is unique. Setting c(G, ϕ) = κ(Λ(G)) completes the proof. 

7. Proof of Theorem 1.1 (c)

Let us introduce the power semigroup as follows.
(Ps x)(t) = x(ts ), t, s > 0.
If ω is an extended limit which is invariant under Ps (we say that it is power
invariant), then ω ◦ log is a state on L∞ (−∞, ∞) which is dilation invariant. This
state vanishes on every function whose support is bounded from above and is,
therefore, identified with a dilation invariant extended limit on L∞ (0, ∞).
In this section, we consider those extended limits which are dilation and power
invariant. The following assertion is available as Theorem 8.6.8 in [25]. For conve-
nience of the reader, we present a short proof here.
 25

Lemma 7.1. If ω is a dilation and power invariant extended limit, then

 1 1

Trω (A) = (ω ◦ log) t → Tr(A1+ t ) , 0 ≤ A ∈ L1,∞ .
t
Proof. We have
 1X 1 1

RHS = (ω ◦ log) t → (n + 1)−1− t · ((n + 1)µ(n, A))1+ t ).
t
n≥0

We have
1 1 1
|(n + 1)µ(n, A) − ((n + 1)µ(n, A))1+ t | ≤ sup{|x − x1+ t | : 0 ≤ x ≤ kAk1,∞ } = O( )
t
as t → ∞. Therefore,
 1X 1

RHS = (ω ◦ log) t → (n + 1)− t µ(n, A) .
t
n≥0

Set now
X
β= µ(n, A)χ(log(n+1),∞) .
n≥0

Clearly, β(u) = O(u) as u ↑ ∞. Using Theorem 8.6.7 in [25], we infer

β(t) h(t)
ω(t → ) = ω(t → ),
t t
where Z ∞
u 1
X
h(t) = e− t dβ(u) = (n + 1)− t µ(n, A).
0 n≥0

Thus,
 1 X  def  1 X 
RHS = (ω ◦ log) t → µ(n, A) = ω t → µ(n, A) .
t log(t) n+1<t
log(n+1)<t

Since A ∈ L1,∞ , it follows that

Z t X
µ(s, A)ds = µ(n, A) + O(1).
0 n+1<t

This completes the proof. 

Corollary 7.2. If ω is a dilation and power invariant extended limit, then c(G, Trω ) >
0.
Proof. Let T = |[F, Z]|p . It follows from Lemma 3.6 that
lim inf sTr(T 1+s ) > 0.
s→0

Therefore,
 1 1

(ω ◦ log) t → Tr(T 1+ t ) > 0.
t
The assertion follows now from Lemma 7.1. 
26

Remark 7.3. The existence of a Dixmier trace ϕ on L1,∞ such that ϕ(T ) 6= 0
follows from the weaker estimate lim sups→0 sTr(T 1+s ) > 0. Indeed, assume the
contrary, that is ϕ(T ) = 0 for every Dixmier trace ϕ. It follows from Theorem 9.3.1
in [25] that
lim sTr(T 1+s ) = 0,
s→0
which is not the case. Since ϕ(T ) = c(G, ϕ), the assertion follows.

References
[1] Beardon A. The geometry of discrete groups. Corrected reprint of the 1983 original. Graduate
Texts in Mathematics, 91. Springer-Verlag, New York, 1995.
[2] Bers L. Simultaneous uniformization. Bull. Amer. Math. Soc. 66 1960 94–97.
[3] Bers L. On boundaries of Teichmüller spaces and on Kleinian groups. I. Ann. of Math. (2)
91 (1970), 570–600.
[4] Biswas, I. Nag, S. Sullivan, D. Determinant bundles, Quillen metrics and Mumford isomor-
phisms over the universal commensurability Teichmuller space. Acta Math. 176 (1996), no.
2, 145–169.
[5] Bowditch B. Geometrical finiteness for hyperbolic groups. J. Funct. Anal. 113 (1993), no. 2,
245–317.
[6] Bowen R. Hausdorff dimension of quasi circles Inst. Hautes Etudes Sci. Publ. Math. No. 50
(1979), 11–25.
[7] Birman M.S. and Solomyak M.Z. Double operator Stieltjes integrals, Spectral theory and
wave processes, Problemy Mat. Fiz., vol. 1, Leningrad State University, Leningrad 1966, pp.
33–67.
[8] Birman M.S. and Solomyak M.Z. Double operator Stieltjes integrals. II, Spectral theory,
difraction problems, Problemy Mat. Fiz., vol. 2, Leningrad State University, Leningrad 1967,
pp. 26–60.
[9] Birman M.S. and Solomyak M.Z. Double operator Stieltjes integrals. III. Taking the limit
under the integral sign, Spectral theory and wave processes, Problemy Mat. Fiz., vol. 6,
Leningrad State University, Leningrad 1973, pp. 27–53.
[10] Birman M.S. and Solomyak M.Z. Spectral Theory of self-adjoint operators in Hilbert space,
D. Reidel Publishing Co, Dordrecht, 1987.
[11] Birman M.S. and Solomyak M.Z. Double operator integrals in a Hilbert space, Integral Equa-
tions Operator Theory 47 (2003), no. 2, 131-168.
[12] Bishop C., Jones P. Hausdorff dimension and Kleinian groups. Acta Math. 179 (1997), no.
1, 1–39.
[13] Caspers M., Potapov D., Sukochev F., Zanin D. Weak type estimates for the absolute value
mapping. J. Operator Theory 73 (2015), no. 2, 361–384.
[14] Connes A. Noncommutative geometry. Academic Press, Inc., San Diego, CA, 1994.
[15] Dixmier J. Existence de traces non normales, (French) C. R. Acad. Sci. Paris Ser. A-B 262
(1966) A1107–A1108.
[16] Dodds P., Dodds T., de Pagter B., Sukochev F. Lipschitz continuity of the absolute value and
Riesz projections in symmetric operator spaces. J. Funct. Anal. 148 (1997), no. 1, 28–69.
[17] Dodds P., de Pagter B. The non-commutative Yosida-Hewitt decomposition revisited. Trans.
Amer. Math. Soc. 364 (2012), no. 12, 6425–6457.
[18] Dykema K., Figiel T., Weiss G., Wodzicki M. Commutator structure of operator ideals. Adv.
Math. 185 (2004), no. 1, 1–79.
[19] Elstrodt J., Grunewald F., Mennicke J. Groups acting on hyperbolic space. Harmonic analysis
and number theory. Springer Monographs in Mathematics. Springer-Verlag, Berlin, 1998.
[20] Gehring F., Vaisala J. Hausdorff dimension and quasiconformal mappings. J. London Math.
Soc. (2) 6 (1973), 504–512.
[21] Hedenmalm H., Korenblum B., Zhu K. Theory of Bergman spaces. Graduate Texts in Math-
ematics, 199, Springer-Verlag, New York, 2000.
[22] Hille E., Phillips R. Functional analysis and semi-groups. Third printing of the revised edition
of 1957. American Mathematical Society Colloquium Publications, Vol. XXXI. American
Mathematical Society, Providence, R. I., 1974.
 27

[23] Katok S. Fuchsian groups. Chicago Lectures in Mathematics. University of Chicago Press,
Chicago, IL, 1992.
[24] Lindenstrauss J., Tzafriri L. Classical Banach spaces. II. Function spaces, Ergebnisse der
Mathematik und ihrer Grenzgebiete 97, Springer-Verlag, Berlin-New York, 1979.
[25] Lord S., Sukochev F., Zanin D. Singular Traces: Theory and Applications, de Gruyter Studies
in Mathematics, 46, de Gruyter, 2013.
[26] Markushevich A. Theory of functions of a complex variable. Vol. III. Chelsea Publishing Co,
1977.
[27] Maskit B. Kleinian groups. Grundlehren der Mathematischen Wissenschaften, 287.
[28] de Pagter B., Sukochev F., Witvliet H. Double operator integrals. J. Funct.Anal. 192 (2002),
no. 1, 52–111.
[29] Peller V.V., Hankel operators in the theory of perturbations of unitary and self-adjoint opera-
tors. Funktsional. Anal. i Prilozhen. 19:2 (1985), 37-51 (Russian), English version: Functional
Analysis and Its Applications, 1985, 19:2, 111–123.
[30] Potapov D., Sukochev F. Unbounded Fredholm modules and double operator integrals. J.
Reine Angew. Math. 626 (2009), 159–185.
[31] Rose J. A Course on Group Theory. Cambridge University Press, 1978.
[32] Rudin W. Functional analysis. Second edition. International Series in Pure and Applied
Mathematics. McGraw-Hill, Inc., New York, 1991.
[33] Semenov E., Sukochev F., Usachev A., Zanin D. Banach limits and traces on L1,∞ , Adv.
Math. 285 (2015), 568–628.
[34] Sullivan D. The density at infinity of a discrete group of hyperbolic motions. Inst. Hautes
Etudes Sci. Publ. Math. No. 50 (1979), 171–202.
[35] Sullivan D. Entropy, Hausdorff measures old and new, and limit sets of geometrically finite
Kleinian groups. Acta Math. 153 (1984), no. 3-4, 259–277.

College de France, 3 rue d’Ulm, Paris F-75005 France

E-mail address: alain@connes.org

School of Mathematics and Statistics, University of NSW, Sydney, 2052, Australia

E-mail address: f.sukochev@unsw.edu.au

School of Mathematics and Statistics, University of NSW, Sydney, 2052, Australia

E-mail address: d.zanin@unsw.edu.au