Horseshoes with infinitely many branches and a

characterization of Sinai-Ruelle-Bowen measures

Fernando José Sánchez-Salas
October 01, 2000

Abstract
We prove that the dynamics of a C 2 diffeomorphism of compact mani-
fold preserving a hyperbolic measure µ with positive entropy can be reduced,
up to return times, to a generalized horseshoe with positive measure in-
duced on suitable neighborhoods of hyperbolic points; some consequences
are drawn for the characterization of Sinai-Ruelle-Bowen measures

1 Introduction
Following Ledrappier [6] we will say that a Borel probability µ preserved by a
C 2 diffeomorphism of a compact riemannian manifold is a Sinai-Ruelle-Bowen
(SRB) measure if it is smooth along the unstable lamination. SRB measures are
the best known examples of physically observable measures P in that the ergodic
n−1
basin, that is, the set of points x ∈ M m such that 1 / n k=0 δf k (x) −→ µ
vaguely, has positive volume [12]. A remarkable and well known theorem due
to Ledrappier and Young says that µ is SRB if and only if it satisfies the Pesin
entropy formula
Z X
hµ (f ) = χ+i (x) dim Ei (x) dµ(x). (1)
χ+
i >0

See [6] and [7]. Recently, so called horseshoes with infinitely many branches and
variable return times were introduced in [15] so giving an unified approach to
several results on the existence of SRB measures and their statistical properties.
This paper is aimed at proving that the presence of that generalized horseshoes
in phase space is indeed a necessary and sufficient condition for the existence of
SRB measures. The following is our main result.
Theorem 1.1. Let f be a C 2 diffeomorphism of a compact riemannian manifold
M m (m ≥ 2) and µ an ergodic hyperbolic Borel probability preserved by f . Then
the following conditions are equivalent:

1. µ satisfies the Pesin entropy formula;

1
 2. the conditional measures of µ respect to any measurable decomposition
subordinated to the unstable Pesin’s lamination are absolutely continuous
respect to the riemannian volume on the leaves;
3. for every point p ∈ supp µ there is a regular neighborhood R(p) and a
horseshoe with infinitely many branches Ω∗ ⊂ R(p) induced by a countable
collection of suitable returns such that:
u
(a) Lebu (Wloc (x) ∩ Ω∗ ) > 0 for µ-a.e. x ∈ Ω∗ and
R
(b) N (x) dLebu (x) < +∞,
u
where Lebu is the Lebesgue measure class of Wloc (x).

Hyperbolic dynamical systems were the first examples where existence and
statistical properties of SRB measures became well understood. See [8, Chapter
III] and [4, Part 4 ]. However, to the best of my knowledge in spite of several
outstanding results appeared in the past decade proving the existence of SRB
measures for some non hyperbolic systems, to give effective verifiable conditions
for the existence of SRB measures in non uniformly hyperbolic systems remains
by large an open question.
The quadratic family and Hénon map are outstanding examples of non hy-
perbolic systems with an SRB measure. In 1981 Jakobson proved the existence
of a set of positive measure of chaotic quadratic maps by inducing a hyperbolic
transformation with infinitely many expanding branches and bounded distor-
tion for a special set of parameters. Later, further developments on Benedicks-
Carlesson approach to the dynamics of Hénon maps led to the remarkable dis-
covering of a set of positive Lebesgue measure of parameters displaying persis-
tently an strange attractor and to the proof that these invariant sets support
an ergodic SRB measure. Cf. [14] and [16].
The idea to reduce Hénon attractors to a sort of topological Markov chain
with countably many states has been around for some time. See for example
[3]. Actually, as pointed to me by Viana, we should expect to reduce Hénon
attractors to some compact set with hyperbolic product structure and non trivial
unstable Cantor sets with positive Lebesgue measure. This is due to homoclinic
tangencies. Unfortunely, as far a I know, no complete exposition of this fact
seems to be yet available. We refer to [15] for an outline of that construction.
In this paper we stand up in a broader, perhaps more conceptual set up, by
using Pesin theory techniques to induce large hyperbolic sets in non-uniformly
hyperbolic systems. This lead us to revisit several classical results of this the-
ory in an unified setting, including absolute continuity properties of the stable
lamination and Ledrappier-Young theorem on the inversion of Pesin entropy
formula. Cf. [6], [10].
For this we use methods of [5] and [9] to get large hyperbolic sets using,
among other things, pseudo-Markov property of returns to hyperbolic Pesin
sets, bounded geometry of admissible manifolds, graph transforms properties,
expansion along unstable admissible manifolds, volume estimates and tempered
corrections due sub-exponential fluctuations along the orbits of the system.

2
 Results in this paper were strongly influenced by the work of Mendoza [9] and
L.S. Young’s [14]. In fact, at some point of this investigation it becames clear
that to get sequences of horseshoes with strong measure-theoretical properties
as claimed in [9, Theorem 4.1] we should consider infinitely many hyperbolic
returns to a single regular neighborhood, so getting a compact set with positive
measure with the same structure of generalized horseshoes as described at [14].

Acknowledgments: This work outgrew from firsts drafts of my doctoral dis-

sertation at IMPA. I would like to thank my adviser, prof. Jacob Palis, for
his support and encouragement during that period. Thanks are also due to M.
Viana and Gustavo T. A. Moreira from IMPA for many suggestions and remarks
on the subject.

2 Statement of results
We say that Ω has a hyperbolic product structure if there are two continuous
laminations F s and F u of discs of complementary dimension such that:
1. each F s (x) is an stable invariant manifold, i.e. distances of positive iter-
ates of points in F s (x) are contracted exponentially; similarly so negative
iterates of points on leaves of F u ;
2. stable leaves intersects transversally unstable leaves at an angle unifomrly
bounded from below;
3. Ω = F s ∩ F u .
S S

Throughout this work f : M m −→ M m will be a C 2 diffeomorphism leaving

invariant an ergodic Borel probability µ without zero Lyapunov exponents. We
shall suppose in addition that µ has positive entropy. This excludes the trivial
case of a uniform distributed mass along a hyperbolic periodic orbit. We will
prove that given a point p in the support of µ we can find a regular neighborhood
R (see below) containing p and a compact subset Ω∗ ⊆ R of relative large
measure with hyperbolic product structure which is maximal for that property.
For this we will use pseudo-Markov property of returns to hyperbolic Pesin sets.
The proof of Theorem 1.1 is in reality a gathering between dimension theory
of dynamical systems and Pesin theory. Actually, as we shall see below, the
characterization of SRB measures follows from measure-theoretical regularities
of certain dynamically defined Carathéodory’s structure introduced by Pesin in
[10]. We recall this construction. Compare also [11, Chapter 1], [11, Chapter 4]
and [13].
Let us start remarking that each stable leave F s (x) comes endowed with
nested sequences of stable Pesin cylinders Sn (x) such that F s (x) = n>0 Sn (x).
T
The reunion of these cylinders forms a family ℘∗ which generates the Borel stable
sets on R, that is, the Borel subsets which are reunion of stable leaves. We use

3
this to construct as follows a dynamically defined Carathéodory structure for
Ω∗ .
Now, given an admissible unstable manifold γ u we define ℘∗ (γ u ) = {P ∩ γ u :
P ∈ ℘∗ }. A countable family U = {Ui }i>0 U ⊆ ℘∗ with diam (Ui ) < δ is called
a δ-covering by ℘∗ (γ u )-sets. Now, given a > 0 we set :
+∞
X
Da,℘∗ ,γ u (X) = lim+ inf Vol γ u (Ui )a , (2)
δ→0 U
i=1

where infimum is taken over δ-covering of ℘∗ (γ u )-sets. We get in this way a well
defined and non-trivial measure class Dα,γ u and a dimensional characteristic
dimγ u ,℘ defined by

dimγ u ,℘∗ (X) = inf{Dα,γ u (X) = 0} = sup{Dα,γ u (X) = +∞}. (3)

We call these the dynamical measure and the unstable dimension respectively.
The following is our main technical result
Theorem 2.1. Let 0 < ρ < 1 be given. Then, for every p ∈ supp µ there is a
regular neighborhood Q and a compact set with hyperbolic product structure Ω∗
contained in a slightly larger regular neighborhood R ⊃ Q such that µ(Ω∗ ∩ Q) ≥
(1 − ρ)µ(Q) and
hµ (f )
dimu (Ω∗ ) = R . (4)
ln J u f (x) dµ(x)
Moreover, the following holds:
1. the stable lamination F s of Ω∗ is absolutely continuous wrt the dynamical
measures {Dα,γ u }γ u ∈Γu with Jacobian
"+∞ #α
d H ∗ Dα,γ u Y J(T | γ u )(T i (x))
(x) = . (5)
d Dα,γ u i=1
J(T | γ u )(T i (H(x)))

H is the holonomy map of the local stable lamination F s defined by unsta-

ble admissible unstable sections γ u , γ u , T is the induced first return map
to Ω∗ and α = dimu (Ω∗ ) is the unstable dimension;
2. we can find C > 1 such that, for every geometrical cylinder P ∈ ℘∗ we
have
µ(P )
C −1 ≤ ≤ C; (6)
(Vol P )dimu (Ω∗ )
moreover, the projection of µ | Ω∗ along stable leaves onto an unstable
manifold F u (z) will be equivalent to Dα,F u (z) so

µF u (z) (B)  Dα,F u (z) (B), (7)

bounded in [C −1 , C] by a suitable constant C > 1, for every Borel subset

B ⊂ Ω∗ .

4
Corollary 2.1. There exists a sequence Ω∗n ⊂ Ω∗ of ordinary horseshoes with
finitely many branches and Gibbs measures µ∗n supported on Ω∗n such that:
1. Ω∗ = n Ω∗n ;
S

2. µ∗n −→ µ∗ is the weak topology;

3. dimu (Ω∗n ) −→ dimu (Ω∗ );
4. there is a constant C > 1 such that for every P ∈ ℘∗ and n > 0
µ∗n (P )
C −1 ≤ ∗ ≤ C,
(Vol P )dimu (Ωn )

where µ∗ is µ | Ω∗ normalized to a probability.

The projection µF u (x) , sometimes called the transversal measure at x, is a
measure on the unstable Cantor sets cutted by a local unstable manifold of Ω∗ :
!
[
µF u (x) (A) = µ s
F (z) for every Borel subset A ⊂ F u (x) ∩ Ω∗ .
z∈A

By absolute continuity of the stable lamination of Ω∗ these measures are equiva-

lent, so they yield a well defined measure class on the space of local stable leaves.
The absolute continuity of the stable lamination implies that dimγ u ,℘∗ (Ω∗ ) does
not depend on γ u , so the unstable dimension of Ω∗ , dimu (Ω∗ ) = dimW,℘∗ (Ω∗ )
for W ∈ F u , is well defined.
The plan of the exposition is as follows. In Section 3 we compute the unstable
dynamical dimension of Ω∗ and then we prove Theorem1.1 using Theorem 2.1.
In Section 4 we construct the induced horseshoe Ω∗ . Section 5 is dedicated to get
some useful bounded distortion estimates which shall be used later. In Section
6 we recall what is the dynamical measure and the dynamical dimension of a
horseshoe and show that the unstable Cantor sets Ω∗ ∩ F u (x) have non trivial
dynamical measure. This is sufficient to see that F s is absolutely continuous
with respect to the dynamical measure classes {Dα,W }W ∈F u , using the same
arguments of [13, Section 6]. Bounded distortion estimates of Section 4 will be
used to prove that unstable Cantor sets are regular with respect to the dynamical
measure class, concluding the proof of Theorem 2.1.

3 Proof of Theorem 1.1

We shall prove our main result using Theorem . Items (1), (3), (4) and (5) in
Theorem will be proved in Section 6, after a review of Pesin’s theory which shall
give us the necessary bounded distortion estimates. So, let us start proving the
following
Lemma 3.1.
hµ (f )
dimu (Ω∗ ) = R (8)
ln J u f (x) dµ(x)

5
Proof. Let T : Ω∗ be the first return map. According to inequalities in
Theorem 2.1, for µ almost every x ∈ Ω∗ :
ln µ(Sn (x))
dimu (Ω∗ ) = lim .
n→+∞ ln Vol (Sn (x))
Let µ∗ be the restriction of µ to Ω∗ normalized to a probability. µ∗ is a T-
invariant Borel probability. Using a local entropy approach we prove

lim −1 / n ln µ∗ (Sn (x)) = hµ∗ (T ), µ − a.e.

n→+∞

On the other hand, by bounded distortion properties stated and proved in

Section 3 and Section 5, we can find a constant C > 1 such that C −1 ≤
Vol (Sn (x))J u T n (z) ≤ C for every z ∈ Sn (x)∩F u (x), where J u T is the unstable
Jacobian of T : J u T = |det DT | E u (x)| (see below). So, almost surely:
Z
lim −1 / n ln Vol (Sn (x)) = ln J u T (z)dµ∗ (z).
n→+∞

On the other hand, let us define Rn∗ = {x ∈ Ω∗ : N (x) = x}, where N = N (x)
is the first return time of x to Ω∗ . Then, we have
Z +∞ Z
X n−1
X
ln J u T (z)dµ∗ (z) = ln J u f (f k (z))dµ∗ (z)
Ω∗ n=0
∗
Rn k=0
+∞ n−1
X XZ Z
∗
= u
ln J f (z)dµ (z) = ln J u f (z)dµ∗ (z).
∗)
f k (Rn Mm
n=0 k=0

By a theorem of Abramov [1, Theorem 1] we have hµ (f ) = µ(Ω∗ )hµ∗ (T ). There-

fore, the unstable dimension dimu (Ω∗ ) is equal to
hµ∗ (T ) µ(Ω∗ )hµ∗ (T ) hµ (f )
R
u ∗
= R
u
=R .
ln J T (z)dµ (z) Mm
ln J f (z)dµ(z) Mm
ln J u f (z)dµ(z)

This is simply what Billingsley did in [2, pp. 136 - pp. 145] in a simpler set
up for dynamically defined Cantor sets in the real line.

Proof of Theorem 1.1

Let T : Ω∗ be a C 2 generalized horseshoe induced in a regular neighborhood
according to Theorem 2.1. Then, the following statements are equivalent:
1. dimu (Ω∗ ) = 1;
2. Vol F u (Ω ∩ F u (x)) > 0 for some local unstable manifold F u (x);
3. the volume Vol F u (Ω) of the unstable Cantor sets Ω ∩ F u (x) is uniformly
bounded away from zero and F s is absolutely continuous wrt Lebesgue
with uniformly bounded Jacobians;

6
 4. µF u , the projection of µ onto F u (x) is absolutely continuous wrt the
riemannian volume along the local unstable manifolds.
In addition, if any of the above conditions hold the stable invariant lamination
F s is absolutely continuous wrt Lebesgue and it has a bounded Jacobian, namely
+∞
d H ∗ Lγ u Y J(T | γ u )(T i (x))
(x) = (9)
d Lγ u i=1
J(T | γ u )(T i (H(x)))

for every pair of admissible unstable manifolds γ u , γ u ∈ Γu . Here Lγ u is the

Lebesgue measure class of γ u .
Indeed, D1,γ u represents the Lebesgue measure class of γ u . This is a straight-
forward consequence of the definitions and the fact that ℘∗ generates the sta-
ble lamination. In particular, Vol γ u (X) > 0 implies dimu (X) = 1. Thus,
dimu (Ω∗ ) = 1 implies that µW , the projection of µ along F s onto the local un-
stable manifolds, is equivalent to the restriction of the Lebesgue measure class
Lebu to F u (x) ∩ Ω∗ , bounded by constants not depending on x ∈ Ω∗ neither on
F u (x). In particular, the volume of the unstable Cantor sets F u (x) ∩ Ω∗ is uni-
formly bounded away from zero, using the absolute continuity of the lamination
and that Jacobians of the holonomy are uniformly bounded. On the other hand,
if Lebu (F u (x) ∩ Ω∗ ) > 0 for some local unstable manifold then dimu (Ω) = 1, so
the conditions are equivalent.
Now, observe that the family of local unstable manifolds {F u (x)}x∈Ω∗ de-
fines a measurable partition and that µF u (x) are precisely the conditional mea-
sures of µR respect to that decomposition. On the other hand, according to Kac’s
formula N (x)dµ(x) = 1 / µ(Ω∗ ). Therefore, N = N (x) is integrable with re-
spect to µF u (x) almost surely. So, the return time is integrable respect to Lebu ,
the measure class of F u (x) for µ-a.e. x ∈ Ω∗ , since Lebu is equivalent to µF u (x)
almost surely.
Let ξ be a any measurable partition subordinated to the unstable lamination.
We can suppose that ξ is increasing i.e. ξ refines f ∗ ξ [7]. µ satisfies the Pesin
entropy formula (1) if and only if dimu (Ω∗ ) = 1. This is equivalent to µξ(z) be
absolutely continuous respect to the riemannian volume for almost every z ∈ Ω∗ .
The set of point satisfying this condition is f-invariant since ξ is increasing, so
it has total measure, by ergodicity , proving that µ is smooth along unstable
leaves. Conversely, by definition if µ is SRB then the conditional measures µξ are
absolutely continuous respect to any measurable decomposition ξ subordinated
to the unstable lamination. So, let Ω∗ be an induced generalized horseshoe. As
Ω∗ has positive measure then µξ(x) (Ω∗ ) > 0 almost surely, so Lebu (Ω∗ ∩F ( x)) >
0 for µ-a.e. x ∈ Ω∗ implying that dimu (Ω∗ ) = 1 and we are done.

4 Inducing fat horseshoes

We will recall for further use some notions of Pesin theory. We refer to [5] for
details.

7
 As related to slowly fluctuations on some quantities defined along orbits of
a non-uniformly hyperbolic system, the following terminology will be frequently
below: give  > 0 we say that a Borel measurable and positive function K
defined over M m is -tempered if (1 + )−1 ≤ K(x) / K(f (x)) ≤ (1 + ) almost
surely. The Tempering Kernel Lemma [5, Lemma S.2.12] says that given a
positive measurable function K such that limn→∞ 1 / n ln(K(f n (x))) = 0, then
for every  > 0 there is an -tempered function K bounding K from above.
Given a hyperbolic measure preserved by a C 2 diffeomorphism of a com-
pact manifold we can perform an -reduction ([5, Theorem S.2.10]) to make the
system uniformly hyperbolic on a set of total measure. This is done by con-
structing a Lyapunov metric defined by linear coordinate changes defined by
a Borel measurable function C : M m −→ GLn (R) such that kC k has slowly
subexponential fluctuations √ along the orbit. Further, there is an -tempered
function L such that 1 / 2kukp ≤ kuk∗p ≤ L (p)kukp . L measures the linear
distortion of the Lyapunov metric with respect to the riemannian metric <, >
in M m .
Then define Lyapunov charts Ψp : B(0, r (p)) ⊆ Rm −→ M m as Ψp =
expp ◦ C (p). The local representative of the diffeomorphism in Lyapunov charts
is  C 1 near of Ap ∈ GLn (R), which represents the linear part of f at p. The geo-
metric distortion due to the introduction of these coordinate changes is bounded
by a tempered function D . See [5, Theorem S.3.1].
We define regular neighborhoods of size h as follows:

R(p, h) = Ψp (B s (0, hr(p)) ⊕ B u (0, hr(p))),

where B u (0, r) ⊆ Ru and B s (0, r) ⊆ Rs are the open balls of center 0 and
radius r > 0 in the standard euclidean metric. A point admitting a regular
neighborhood as above is called completely regular. The set of completely regular
points of µ has full measure. Cf. [5, pp-673].
We use Lyapunov charts on regular neighborhoods to define stable and un-
stable cones of width γ [4, Definition 6.2.9].For example, we can define Kuγ (x)
at every point in Ψp (x) ∈ R setting Kuγ (x) ≡ DΨp (x){(u, v) ∈ E u × E s :
kvk ≤ γ kuk}. Likewise Ksγ (x). Then we introduce families of admissible sta-
ble and unstable submanifolds of dispersion γ [5, Definition S.3.4] denoted Γu
and Γs respectively. γ u ∈ Γu is the image under the Lyapunov chart Ψp of a
graph γ u = Graph(φ) of a C 1 map φ : B u (0, r(p)) −→ B s (0, r(p)) whose tan-
gent spaces lies inside the unstable cones in Rm . Every γ u ∈ Γu and γ s ∈ Γs
intersect transversally at a single point x with an angle bounded from below.
The geometry of graphs with dispersion γ is bounded. In fact, let g ij be the
riemannian metric of γ u , a graph with dispersion γ, then g ij (x) = δij + o(γ 2 ).
The property of bounded geometry of graphs with small dispersion is transferred
to admissible manifolds γ u = Ψp (γ u ) with slowly varying corrections along the
orbit, due to properties of Lyapunov charts.
Lemma 4.1. Let  > 0 be a positive number, p a completely regular point and
u
R(p) a regular neighborhood, γ u ∈ Γu (p) and γ u ∈ Γ (0, p) such that γ u =

8
Ψp (γ u ). Then there is an -tempered positive function K0 (p) = K0 (p, γ, ) > 1
such that the ratios:
g ij (x) V γ u (x) dist γ u (Ψp (x), Ψp (y))
, , and (10)
gij (Ψp (x)) Vγ u (Ψp (x)) kx − yk

are bounded in the interval [ K0−1 (p) , K0 (p) ] for every x, y ∈ γ u , where gij
is the intrinsic riemannian metric and Vγ u the intrinsic volume given by the
inmersion in (M m , g).
The above result is an example of the following useful principle of tem-
pered bounded geometry of admissible manifolds: properties proved for
u
graphs with small dispersion γ u ∈ Γ (0, p) translate into properties of admis-
sible manifolds in M m , γ u ∈ Γu (p, h), with corrections bounded by tempered
functions.
Admissible submanifolds associated to a suitable -reduction are preserved
by iterations under f if their dispersion is small enough and if they are near
enough to a hyperbolic orbit. Indeed, we can find a universal constant 0 < γ0 <
1 such that for every γ ∈ (0, γ0 ] there exists  = (γ) ∈ (0, 1) depending on γ and
an -reduction such that for every 0 < h ≤ 1 and for every γ u ∈ Γu (p, h/2) the
image f (γ u ) ∩ R(f (p), h/2) is an admissible unstable manifold with dispersion
< γ. This gives a well defined graph transform f∗ : Γu (p, h/2) −→ Γ(f (p), h/2).
So we can fix γ ∈ (0, γ0 ], choose an -reduction and then we find a constant
λ > 1 depending on the rate of hyperbolicity of µ, γ and  such that for every
completely regular point p and for every admissible manifold γ u ∈ Γu (p, h/2) we
can find a disc Du ⊂ γ u such that f : Du −→ f∗ γ u is an expanding diffeomor-
phism with expansion coefficient bounded from below by λ, with a correction
due the tempered bounded geometry of admissible manifolds. Compare [5,
Proposition S.3.5] and [13, Lemma 3.2].
Now, we endow Γu (p, h) (resp. Γs (p, h)) with an structure of Banach space
in the obvious way, getting a 0 < θ < 1 only depending on the -reduction,
γ and the expansion coefficient along unstable admissible manifolds such that
dist C 1 (f∗ (γ1u ), f∗ (γ2u )) ≤ θ distC 1 (γ1u , γ2u ). So the graph transform is a contrac-
tion. This the Aleeksev-Moser lemma. Cf. [4, Lemma 6.2.17].
We can use Luzin-Egoroff’s theorem to get large hyperbolic non-invariant
sets. They will be called Pesin sets. A δ-Pesin set is a Pesin set with µ(Λδ ) ≥
1 − δ. We choose Λδ in order that the linear Lyapunov charts C : M −→
GL(m, R), the size of regular neighborhoods, the geometrical distortion intro-
duced by Lyapunov charts and the splittings into stable and unstable subspaces
to be continuous. Cf. [5, Theorem S.4.3]). Let Lδ , Dδ be the maximum of L and
D respectively on Λδ and rδ the minimum radius of Lyapunov charts defined
over points in the Pesin set. We denote R̃(p, h) := Ψp (B s (0, hrδ ) ⊕ B u (0, hrδ ))
and call it the h-reduced regular neighborhood centered at p.
Lyapunov charts depends continuously on points in a given hyperbolic Pesin
set (cf. [5, Proposition S.4.5]). Thus, although changes in the center of a
Lyapunov coordinate system may increase the dispersion, this deformation can
be controlled when p runs in a hyperbolic Pesin set. Namely,

9
Lemma 4.2. ([5, Corollary S.4.6]) Let Λδ be given and 0 < h < 1. Then, given
1 < β ≤ 2 there exists κ = κ(δ, β, h) > 0 such that if p, , q ∈ Λδ , d(p, q) < κ and
γ u ∈ Γu (p, h) then Ψp (Ψ−1 u
q (γ )) ∩ R̃(q, h/2)) is part of an admissible (u, βγ)-
submanifold near p.
The above facts permits to prove the following

Lemma 4.3. (Hyperbolic Return Lemma) Let Λδ be a hyperbolic Pesin set.

Then, for every 0 < h ≤ 1 there exists κ = κ(δ, h) > 0 with the following
property.
Given p , q ∈ Λδ and x ∈ Λδ returning to Λδ between m iterates and such that
d(x, p) < κ and d(f m (x), q) < κ, then there are stable and unstable cylinders
S ⊆ R̃(p, h) and U ⊆ R̃(q, h) containing x and f m (x) respectively such that:
1. f m : S −→ U is hyperbolic;
2. diam (f i (S)) ≤ Dδ max{diam (R̃(p, h))θi , diam (R̃(f m (p), h))θm−i } for -
every i = 0, · · · , m;

3. for every γ u ∈ Γu (p)) it holds that f m : γ u ∩S(x) −→ f∗n γ u is an expanding

diffeomorphism, that is:

du (f m (z), f m (z 0 )) ≥ Kδ λm du (z, z 0 )

for every z, z 0 ∈ γ u ∩ S(x), being du the intrinsic riemannian metric along

admissible unstable manifolds.
S(x) is the connected component of R(p, h) ∩ f −m (R(q, h) containing x and U (x)
is the the connected component of f m (R(p, h)) ∩ R(q, h) containing f m (S(x)).
A compact connected subset subset S ⊆ R(p) is an admissible stable cylinder
if it admits a foliation by admissible stable submanifolds and if its unstable
sections S ∩ γ u are convex sets for every γ u ∈ Γu (R). Unstable cylinders are
defined similarly.
A map C 1 T : S −→ U between stable and unstable cylinders is hyper-
bolic if it preserves strictly the cone families: Ks (T (x)) ⊆ int DT (x)Ks (x) and
DT (x)Ku (x) ⊆ int Ku (T (x)).
Previous Lemma explains what is the pseudo-Markov property of hyperbolic
Pesin sets covered by regular neighborhoods. See [Definition S.4.15] and [Theo-
rem S.4.16] in [5]. Hyperbolic Return Lemma follows since expansion along ad-
missible unstable overflows the slowly subexponential corrections of Lyapunov
charts, when one return to a Pesin set so that small pieces of admissible unstable
manifolds can be uniformly expanded onto admissible manifolds of uniformly
bounded size in finitely many iterates.
The following is a consequence of the pseudo-Markokv property and it will
be used below. See [5, Theorem S.4.13].

Katok’s Closing Lemma Let Λδ be a hyperbolic Pesin set and let  > 0

10
be given. Then there exists β = β(, δ) < κ with the followinbg property: sup-
pose that x and f m (x) belongs to Λδ and that dist (x, f m (x)) < β, then there
is a hyperbolic periodic point z ∈ M m such that dist (f j (x), f j (z)) <  for
j = 0, · · · , m.

Proposition 4.1. There is a maximal compact set Ω∗ ⊆ R such that µ(Ω∗ ) ≥

(1 − ρ)µ(Q) with a hyperbolic product structure, that is:
1. there are two continuous laminations F s and F u such that dim F s (x) and
dim F u (x) are admissible manifolds for every x ∈ Ω∗
2. F u (f n (x)) ⊂ f n (F u (x)) and f n (F s (x)) ⊂ F s (f n (x)) whenever x ∈ Ω∗
and f n (x) ∈ Ω∗ ; further, d(f n (p), f n (q)) −→ 0+ exponentially fast for
every pair of points p, q ∈ F s (x), x ∈ Ω∗ ; likewise for backward orbits of
points in F u -leaves;
3. F s (x) intersects F u (y) transversally at a unique point with angle uni-
formly bounded away from zero, for every x and y in Ω∗ ;
4. the familiy of stable subspaces, x 7−→ Exs = Tx F s , satisfies a Hölder con-
dition with an exponent 0 < θ < 1 and a norm C > 0 only depending on
Λδ . Similarly so E u ;
5. Ω∗ = F s ∩ F u .
S S

To prove this proposition we take a hyperbolic Pesin set Λδ for an ergodic

hyperbolic Borel probability µ, 0 < h < 1, p ∈ Λδ a point in the support
of the measure µ and R̃(p, h) an h-reduced regular neighborhood. Let also
h0 = h0 (κ) be such that R̃(p, h0 ) ⊆ R̃(p, h) and diam R̃(p, h0 ) < κ. These regular
neighborhoods will be held fixed and shall denoted by R = R̃(p, h) and Q =
R̃(p, h0 ), respectively. Now, let 0 < ρ < 1 and choose δ > 0 and 0 < h < 1 such
that µ(Λδ ∩ Q) ≥ (1 − ρ)µ(Q).
We can get a countable dense subset of hyperbolic periodic points {zi } such
that Q ∩ Λδ ⊆ {z1 , z2 , · · · } using Katok’s Closing Lemma. For this we let ℘n be
a sequence of decompositions ℘n of Q into compact sets with non overlapping
interiors such that diam (℘n ) −→ 0+ as n goes to infinite (take for example
images by Lyapunov charts of decompositions into dyadical cubes in Rm ). Using
Katok’s closing lemma we find a hyperbolic periodic point z ∈ ℘n (x), for every
x ∈ Q ∩ Λδ and n > 0 such that µ(℘n (x)) > 0. Let {z1 , z2 , · · · } be the set
of hyperbolic periodic points obtained in this way, indexed in increasing order
according to their prime period. Clearly {z1 , z2 , · · · } is dense in Q ∩ Λδ .
Then we use the following fact: let E ⊆ Q be a finite subset of hyperbolic pe-
riodic points, then there exists a finite disjoint collection of cylinders {S(z)}z∈E
and {U (z)}z∈E and hyperbolic branches Tz : S(z) −→ U (z) generating a hyper-
bolic horseshoe Λ∗ (E) which contains E. This is an straightforward consequence
of the Hyperbolic Return Lemma and the definition of a horseshoe.
∗
Given n > 0 one construct a horseshoe with finitelyS ∗ many branches Ωn such
∗
that {z1 , · · · , zn } ⊆ Ωn using the above reamrk. n Ωn satisfies conditions (1),

11
(2), (3) and (4) in the proposition. In fact, every F s -leave (resp.F u -leave) is
in the C 1 -topology, of a suitable sequence of local stable manifolds of
the limit, S
points in n Ω∗n . Hyperbolic Return Lemma and properties of the graph trans-
form implies that these local stable manifolds have the claimed properties. Then
we can find a maximal element among the family of sets in R with hyperbolic
product structure whose invariant leaves are admissible. Ω∗ is the homoclinic
closure of all transversal homoclinic points related to points x at the hyperbolic
Pesin set Λδ whose local invariant manifolds passing by R are admissible.
Lemma 4.4. There is a countable collection of disjoint sets Ω∗i ⊆ Ω∗ such that:
1. µ(Ω∗ − i Ω∗i ) = 0;
S

2. for each Ω∗i there is a return time ni such that f ni (Ω∗i ) ⊆ Ω∗ ; further, for
every n > 0, # {i : ni = n} is finite;
3. for every x ∈ Ω∗i we have W s (x, R) ∩ Ω∗ ⊆ Ω∗i and W u (f ni (x), R) ∩ Ω∗ ⊆
f ni (Ω∗i ), that is, Ω∗i is saturated in the stable direction and f ni (Ω∗i ) is
saturated in the unstable direction.
Proof. As µ(Ω∗ ) > 0 we can decompose it into disjoint subsets according to their
first past return time to Ω∗ , namely: Rn is the set of points x ∈ Ω∗ such that
f −n (x) ∈ Ω∗ and f −k (x) 6∈ Ω∗ for 0 < k < n. Each Rn decomposes into a finite
family of disjoint subsets saturated in the stable direction. Indeed, let Ωs =
s ∗
S
x∈Ω∗ F (x) be the reunion of all the stable leaves passing by points in Ω and
−n s
let n > 0 a return time. Then f Ω is a compact set and is a disjoint reunion
of stable leaves, since f −n F s (x) overflows F s (f −n x). In fact, as f −n R ∩ R
decomposes into finitely many connected components, say Sin , i = 1, · · · , Nn ,
which are stable cylinders and such that f n maps Sin hyperbolically onto an
unstable cylinder we can define Ω∗i = Sin ∩ Ω∗ concluding that f −n Ωs ∩ Ω∗ =
SNn
i=1 Ωi .

The above subsets Ω∗i and f ni (Ω∗i ) are called s-sets and u-sets respectively,
according to Young’s terminology. They provide a decomposition of the hyper-
bolic product structure in Ω∗ satisfying axioms (P1) and (P2) in [15]. We also
notice for later use that Ω∗n = ni ≤n Ωi is an ordinary horseshoe with finitely
S
many hyperbolic branches. So we have the following

Corollary 4.1. There isSa sequence of horseshoes with finitely many branches
Ω∗n ⊆ Ω∗ such that Ω∗ = n Ω∗n .

5 Bounded distortion estimates

The following bounded distortion estimates will be used below.

Lemma 5.1. Let 0 < h < 1 be given. Then, for every  > 0 there is an
-tempered correction K2 > 1 with the following properties:

12
 1. let γ u ∈ Γu (p, h/2) be an admissible unstable manifold, m > 0 and Dum ⊂
γ u such that f m : Dum −→ f∗m γ u is an expanding map. Then,

Jac (f n | γ u )(z)
≤ exp[K2 (f n (p)) du (f n (z), f n (z 0 ))] ∀ z, z 0 ∈ Dum ;
Jac (f n | γ u )(z 0 )
(11)

2. let γ u (x), γ u (y) ∈ Γu (p, h/2) be two admissible unstable manifolds passing
by x and y and suppose that they are in the same local stable manifold.
We define
+∞
Y Jac(f | γ u (x))(f i (x))
h(x, y) = .
i=1
Jac(f | γ u (y))(f i (y))

Then, ln h(x, y) ≤ K2 (p) ds (x, y) and for every x0 ∈ γ u (x) and every
y 0 ∈ γ u (y) whcih are in the same local stable manifold it holds

h(x, y)
ln ≤ K2 (p) max{du (x, x0 ), du (y, y 0 )}; (12)
h(x0 , y 0 )

3. for every z, z 0 ∈ M m resting in the same local unstable manifold we have

+∞
Y J u f (f −i (z))
∆(z, z 0 ) = ≤ exp(K2 (p)du (z, z 0 )). (13)
i=1
J u f (f −i (z 0 ))

Jacobians and distances du along unstable admissible manifolds are computed

with the intrinsic riemannian metric.

Proof of (1): we shall prove part (1) considering graphs with small dispersion.
Next we shall use tempered bounded geometry principle to draw our conclu-
sions. For this we let γ u = Graph(φ) a graph in B u (0, r(p)) × B s (0, r(p)) whose
dispersion is bounded by 0 < γ < 1. We let also fn : Rm −→ Rm be the
representative of f in Lyapunov charts at f n (p), that is:

fn (x, y) = Ψ−1
f n+1 (p) ◦ f ◦ Ψf (p) (x, y)
n

whenever (x, y) ∈ B u (0, r(f n (p))) × B s (0, r(f n (p))). We can write fn (x, y) =
(Asn · x + φsn (x, y), Aun · y + φun (x, y)), where An is the representative of the
linear part is the Lyapunov coordinates. Now let Fn = fn ◦ · · · ◦ f0 . We have
a well defined graph transform (Fn )∗ γ u and a disc Dun ⊂ γ u such that Fn :
Dun −→ (Fn )∗ γ u is an expanding diffeomorphism. We can write the restriction
of Fn to γ u as Fn (x) = Fn (x, φ(x)), for some Dn (0) ⊂ B u (0, r(p)). Therefore,
Fn (x) = Ln ·x+ψn (x), for a matrix Ln with kLn k ≥ λn and ψn and a non-linear
part such that kDψn (x)k is bounded from above by a constant only depending
on γ, the expansion λ along admissible manifolds and the -reduction. Now,

13
let Jn (x) = det(Ln + Dψn (x)) be the expansion coefficient of the volume under
iterations by Fn , then:

Jn (x) det(Id + L−1

n Dψn (x))
= ≤ 1 + C1 kL−1 −1
n Dψn (x) − Ln Dψn (y)k,
Jn (y) det(Id + L−1
n Dψn (x))

for a suitable C1 = C1 (γ, λ). We can write [fn | Dun ](x) = Mn · x + ξn (x) for a
matrix Mn and a non linear perturbation ξn , for x ∈ Dn (0), so
n−1
Y n−1
X n−i−1
Y
Fn (x) = Mn−k · x + Mn−k · ξi+1 ◦ Fi (x).
k=0 i=1 k=0

Pn−1
Thus, kL−1
n (Dψn (x) − Dψn (y)k ≤ i=1 λ−n λn−i−2 γkFi (x) − Fi (y)k and then
n−1
X
kL−1
n (Dψn (x) − Dψn (y)k ≤ λ−(i+2) γλ−(n−i) kFn (x) − Fn (y)k
i=1
−(n+2)
≤ γ nλ kFn (x) − Fn (y)k ≤ C2 (γ, λ)kFn (x) − Fn (y)k,

for a suitable constant C2 . From this we conclude

Jn (x)
≤ exp(K 0 (f n (p))du (F n (x), F n (y)), (14)
Jn (y)

where K 0 = K 0 (p) is a correction due to the tempered bounded geometry of

small dispersion graphs.

Proof of (2): take a point p in a hyperbolic Pesin set Λδ and consider the
continuous lamination of stable local manifolds passing by x ∈ R(p) ∩ Λδ . We
define, for every x ∈ R(p) ∩ Λδ , a set W(x) formed by the u-planes E ⊂ Ku (y)
contained in the unstable cones at points y ∈ W s (x, R(p)) . We introduce
W(p, δ) the reunion of all W(x) passing by points x ∈ R(p) ∩ Λδ . This is a
compact subset of Gu (p) the Grassmannian of u planes passing by points in
R(p). Now let us consider J(E) = det|Df | E|. In fact, J : Gu (p) −→ R
is a C 1 smooth function so its derivative DJ is bounded and we can define
K 00 (p) = sup{kD ln J(E)k : E ∈ Gu (p)}. Further, K 00 is a tempered function,
i.e. limn→+∞ 1/n ln K 00 (f n (p)) = 0. This can be seen as follows: let γ u an
admissible unstable manifold tangent to E ∈ Gup (p). Using Lyapunov local co-
ordinates we can check that J(E) comparable to det < ∂xi α, ∂xj α >, bounded
by a tempered correction, where α(x) = (x, φ(x)) is the natural parametriza-
tion of γ u the small dispersion graph which represents γ u in these coordinates.
Observe also that < ∂xi α, ∂xj α >= δij + < ∂xi φ, ∂xj φ >. A further elementary
computation shows that kD ln J(E)k is comparable with r(p), the radius of the
Lyapunov chart, bounded by a tempered correction.
Now, by graph transform contraction property, there is 0 < θ < 1 such that
d(f∗ E, f∗ E 0 ) ≤ θ d(E, E 0 ) for every E, E 0 ∈ W(x). So, |J(f∗ E) − J(f∗ E 0 )| ≤

14
K 00 (f (p))d(f∗ E, f∗ E 0 ) ≤ K 00 (f (p))θ d(E, E 0 ).Hence,
+∞ +∞
Y Jac(f | γ u (x))(f i (x)) X 00 i
ln ≤ K (f (p))θn d(Tx γ u (x), Ty γ u (y)).
i=1
Jac(f | γ u (y))(f i (y)) i=1

Take  > 0 such that (1 + )θ ≤ θ0 < 1 and K00 an -tempered upper bound
function for K 00 . We conclude that ln h(x, y) ≤ K 000 (p)θ0 (1 − θ0 )−1 ds (x, y), for
some tempered correction K 000 . Here ds is, as usual, the distance along the local
s
stable manifold Wloc passing by x and y with the intrinsic riemannian metric.
Then, ln h(x, y) ≤ K 000 (p)θ0 (1 − θ0 )−1 ds (x, y) and

h(x, y)
ln ≤ K 000 (p)θ0 (1 − θ0 )−1 [ds (x, y) − ds (x0 , y 0 )]
h(x0 , y 0 )
≤ K 000 (p)θ0 (1 − θ0 )−1 max{du (x, x0 ), du (y, y 0 )}, (15)

using triangular inequality.

u
To prove (3) we argue as we did above using the fact that Wloc are backward in-
1
variant C submanifolds. Compare [6, Proposition 3.1]. To complete the proof
of the above lemma we take the maximum of tempered corrections at (14) and
(15). QED

Lemma 5.2. There is a tempered function K3 = K3 (p, , γ) > 1 such that for
every stable cylinder S ⊂ R(p) ∩ f −n R(f n (p)) it holds
1. K3−1 ≤ Vol γ u (S)Jac(f n | γ u )(x) ≤ K3 , for every γ u ∈ Γu (p, h/2), n > 0
and x ∈ γ u ∩ S;
2. K3−1 ≤ Vol γ u (x) (S) / Vol γ u (y) (S) ≤ K3 for any pair of unstable admissible
manifolds γ u (x) and γ u (y) passing by points x, y ∈ S in the same stable
local manifold, i.e. W s (x, R) = W s (x, R);
3. K3−1 ≤ Vol (S)Jac(f n | γ u )(x) ≤ K3 for every γ u ∈ Γu (p, h/2), x ∈
S ∩ γu.
Proof. As f n : S ∩ γ u −→ f∗n γ u is onto, for any γ u ∈ Γu (p) so we get

inf Vol u (γ u ) ≤ Vol γ u (f n (S ∩ γ u )) ≤ sup Vol u (γ u ).

γ u ∈Γu (f n (p)) γ u ∈Γu (f n (p))

Also, by the above lemma, the ratios

Jac(f n | γ u )(z)dVol γ u (z)

R
S∩γ u
Vol γ u (S)Jac(f n | γ u )(x)

are bounded between

K2 (f n (p))−1 diam γ u (R(f n (p)) and K2 (f n (p))diam γ u (R(f n (p)).

15
Hence, the following tempered function is an upper bound for the products
Vol γ u (S)Jac(f n | γ u )(x):

supγ u ∈Γu (f n (p)) Vol u (γ u )

K30 = K2 (f n (p))diam γ u (R(f n (p)) .
inf γ u ∈Γu (f n (p)) Vol u (γ u )

A similar computation gives a tempered lower bound. To prove (2) we first

observe that for every x, y ∈ S in the same stable local manifold, the ratios
Vol γ u (x) (S) / Vol γ u (y) (S) are bounded in the interval

Jac(f n | γ u )(y)
[ (K30 )−1 (p) , K30 (p) ],
Jac(f n | γ u )(x)

and that, by Lemma 5.1(2) we see that

Jac(f n | γ u )(y)
≤ h(x, y) ≤ K2 (p) diam s (R(p)),
Jac(f n | γ u )(x)

where diam s is the diameter along the stable manifold W s (x, R) computed with
the intrinsic riemannian metric. (3) is proved in a similar way observing that
Vol (S) is comparable with the volume of unstable sections S∩γ u up to tempered
corrections. K3 is the maximum of the above corrections.
We collect in a single tempered function K = K(p, , γ) the several correc-
tions introduced along this Section. That gives us a control on the geometry and
forward and backward volume distortion on iterates of admissible and locally
invariant unstable manifolds through a single tempered correction parameter.
Lemma 5.3. (Bounded Distortion Lemma) For any given δ there is a hyper-
bolic Pesin set Λδ and Kδ > 1 such that or every p ∈ Λδ :
1. for every γ u ∈ Γu (p, h/2)) and m > 0 such that f m (p) ∈ Λδ

Jac (f n | γ u )(z)
ln ≤ Kδ du (f n (z), f n (z 0 )),
Jac (f n | γ u )(z 0 )

for every z, z 0 ∈ Dum , where Dum ⊂ γ u is such that f m : Dum −→ f∗m γ u is

an expanding map;
2. ln h(x, y) ≤ Kδ ds (x, y), for every γ u (x), γ u (y) ∈ Γu (p, h) such that x and
y lies in the same stable manifold, i.e. W s (x, R(p)) = W s (y, R(p)). Also,
for every x0 ∈ γ u (x) and y 0 ∈ γ u (y) such that

y 0 = W s (x0 , R(p, h)) ∩ γ u (y)

it holds that
h(x, y)
ln ≤ Kδ max{du (x, x0 ), du (y, y 0 )};
h(x0 , y 0 )

16
 3. ln ∆(z, z 0 ) ≤ Kδ du (z, z 0 );
4. let S ⊂ R(p) ∩ f −n R(f n (p)) be an stable cylinder produced some hyperbolic
return, then the have the following bounds:
(a) Kδ−1 ≤ Vol γ u (S)Jac(f n | γ u )(x) ≤ Kδ ;
(b) Kδ−1 ≤ Vol γ u (S) / Vol γ u (S) ≤ Kδ ;
(c) Kδ−1 ≤ Vol (S) / Vol γ u (S) ≤ Kδ and
(d) Kδ−1 ≤ Vol (S)Jac(f n | γ u )(x) ≤ Kδ
for every pair of admissible unstable sections γ u , γ u ∈ Γu (p), n > 0 and x ∈
Dun = γ u ∩ S a disc such that f n : Dun −→ f∗n γ u is an expanding diffeomorphism.

6 The dynamical measure of a generalized horse-

shoe
As Ω∗ has positive measure we can induce a first return map T : Ω , defined
by T (x) = f N (x) (x) where f k (x) 6∈ Ω∗ for 0 < k < N (x). T extends to the
invariant laminations. Further, for every stable leave F s (x), x ∈ ΩT∗ there is a
nested sequence of stable cylinders Sn (x) ⊂ R such that F s (x) = n>0 Sn (x).
The reunion of these stable cylinders forms a family ℘∗ which generates the
Borel stable sets on R. So, given an admissible unstable manifold γ u ∈ Γu we
can define a C-structure as we did in Section 2.
Now, let us outline the mian points in Theorem 2.1 proof. First we show
that the dynamical measure is non trivial. Then we use the same arguments
in [13, Section 6] to prove that F s is absolutely continuous and to compute its
Jacobian. We use the absolute continuity of the stable lamination, bounded
distortion estimates to show that τ : Ω∗ /F s −→ Ω∗ /F s is a measure-theoretical
expanding transformation according to definitions in [8, Chapter III], when we
endow the space of stable leaves with the dynamical measure class. Therefore
we can find a unique τ -invariant ergodic measure µ∗W for every W ∈ F u which
is equivalent to Dα,W , being α = dimu (Ω∗ ). This equilibrum state is in fact the
measure of maximal dimension and it satisfies the estimations (6) in Theorem
2.1. Finally we show that µ∗W is the transversal measure of µ. This completes
the proof of Theorem 2.1.
Lemma 6.1. The dynamical measure class is non trivial.
Proof. First observe that there is a sequence Ω∗n ⊂ Ω∗ of ordinary horseshoes
with finitely many branches and ergodic measures µn such that:
1. Ω∗ = n Ω∗n ;
S

2. there is a constant C > 1 such that, for every local unstable manifolds W =
W u (x, R) and for every P ∈ ℘∗ (Ω∗n ), the family of stable cylinder gener-
∗
ating the stable subsets of Ω∗n , it holds that µ∗n,W (P )  Vol W (P )dimu (Ωn )
bounded by C. Here µn,W denotes the projection of µ∗n the natural equi-
librium state of µ∗n onto W along the stable lamination.

17
 Indeed, let µα the equilibrium state of the potential −α ln J u T of an ordinary
C horseshoe and µW the projection of µα onto W = W u (x, R) along the stable
2

leaves. We proved in [13, Theorem 2.4] that µW is equivalent to the dynamical

measure Dα,W . Moreover, we found a constant C > 1 only depending on bounds
of the non linear distortion of the volume along unstable leaves, the expansion
coefficient of Ω and the bounded geometry of admissible manifolds such that

µW (P )
C −1 inf u Vol (W ) ≤ ≤ C sup Vol (W ), (16)
W ∈F (Vol W P )dimu (Ω) W ∈F u

for every stable cylinder P ∈ ℘.

Now, let Ω∗n the sequence of ordinary horseshoes constructedS in Corol-
lary 4.1. Ω∗ is the topological limit of these Ω∗n , that is, Ω∗ = n Ω∗n . By
the Bounded Distortion Lemma 5.3, bounds for the non linear volume dis-
tortion of Ω∗n are independent of n > 0. So, let µ∗n the Gibbs measure of
φ = − dimu (Ω∗n ) ln J u T . µ∗n is the measure which maximizes the unstable
∗
dimension and, by (16) µ∗n,W (P )  (Vol W P )dimu (Ωn ) for every P ∈ ℘(Ω∗n ),
bounded in [C −1 , C] for a suitable constant C = C(δ, λ, Γ) > 1. Notice also
that ℘(Ωn ) ⊂ ℘(Ωn+1 ).
Now let α = limn→+∞ dimD,W (Ωn ), These limits exists since dimD,W (Ωn )
is monotone and bounded. Let µ∗ be a limit point of µ∗n . µ∗ is ergodic since it
is a limit of ergodic measures and for every P ∈ ℘∗ it holds
∗
µ∗ (P ) ≥ lim sup µ∗n (P ) ≥ C −1 lim sup(Vol W P )dimD,W (Ωn )
n→+∞ n→+∞

since P ∈ ℘∗ is closed. Thus µ∗ (P ) ≥ C −1 (Vol W P )α for every P ∈ ℘∗ , con-

cluding that Dα,W (Ω∗ ) < +∞, using Frostman’s lemma argument.
Furthermore, µ∗ (B) ≥ C −1 · Dα,W (B) for every Borel subset B ⊆ Ω∗ so it
is absolutely continuous respect to the dynamical measure class. Similarly so,

µ∗ (int P ) ≤ lim inf µ∗n (int P ) ≤ C lim inf (Vol W P )dimD,W (Ωn ) ,
n→+∞ n→+∞

since Vol (∂ P ) = 0 for every P ∈ ℘. Then, µ∗ (int P ) ≤ C Vol W (P )α and this

imply Dα,W (Ω) > 0.

Given F u (x) and F u (T (x)) we define τ (x) = H ◦ T (x), where H is the

holonomy transformation defined by these unstable sections. Let Ω∗ ∩ F u (z)
be the unstable Cantor set intersected by F u (x). Let us define d∗ (x, y) =
dist H (F s (x), F s (y)), where dist H is the Hausdorff distance of sets. As it is
easy to see (Ω∗ ∩ F u (z), d∗ ) is a compact metric space. We endow this space
with the Borel measure Dα,F u (x) .
The dynamical measure class is conformal in the following sense: let h :
W −→ Rm be a C 1+ embedding. Suppose in addition that 0 < Dα,W (Ω∗ ) < ∞.
Then Z
Dα,W (h(B)) = (Jh(x))α dDα,W (x),
B

18
for every Borel subset B ⊆ Ω∗ ∩ W . Jh is the Jacobian of h wrt intrinsic
riemannian volume in W. Cf. [13, Lemma 6.2].
We use this property to prove that τ has a Jacobian wrt Dα,F u (x) . Indeed,

d H ∗ Dα,F u (x)
Jα τ (x) = (T (x))[J u T (x)]α . (17)
d Dα,F u (T (x))

Lemma 6.2. There exists a constant C > 1 such that for every n > 0

Jα τ n (x)
− 1 ≤ C d∗ (τ n (x), τ n (y)) (18)
Jα τ n (y)

whenever x, y ∈ Dun , where T n : Dun ⊂ F u (x) −→ F u (T (x)) is expanding.

Proof. Bounded Distortion Lemma will be used to prove (18). In fact let us
denote
+∞
Y J u (T i z)
J u (x) = Jac (T | F u (x))(x) and h(z) = .
i=1
J u (T i H(z))

With this notation we can write

"n−1 #α
Jα τ n (x) Y h(T (τ k (x))) J u (τ k (x))
= · . (19)
Jα τ n (y) h(T (τ k (y))) J u (τ k (y))
k=0

Using the Bounded Distortion Lemma we see that

n−1
X
| ln J u (τ k (x)) − ln J u (τ k (y))|
k=0
n−1
X
≤ Kδ λ−n+k d∗ (τ n (x), τ n (y))
k=0
≤ Kδ (1 − λ−1 )d∗ (τ n (x), τ n (y)).

Also, by Bounded Distortion Lemma ln h(T (τ k (x))) / h(T (τ k (y))) is bounded

from above Kδ max {du (T (τ k (x)), T (τ k (y))) , du (τ k+1 (x), τ k+1 (y))}, and this,
in turn, is limited by Kδ2 d∗ (τ k+1 (x), τ k+1 (y)), since du is comparable with d∗
with a correction bounded by Kδ . Thus, for every n > 0 we have
"n−1 # n−1
Y h(T (τ k (x))) X
ln k
≤ K 2
δ λ−n−k−1 d∗ (τ n (x), τ n (y))
h(T (τ (y)))
k=0 k=0
≤ Kδ2 (1 − λ−1 )d∗ (τ n (x), τ n (y)).

Therefore, Jα τ n (x) / Jα τ n (y) ≤ Cd∗ (τ n (x), τ n (y)) absorbing different constants

into C > 1. The conclusion follows wasily from here.

19
 The above Lemma shows that

τ : (Ω∗ ∩ F u (z), Dα,F u (z) ) −→ (Ω∗ ∩ F u (z), Dα,F u (z) )

satisfies hypotheses of [8, Chapter III, Theorem 1.3], so there is a unique ergodic
τ invariant measure µ∗F u (z) such that

Dα,F u (τ −n B)
µ∗F u (B) = lim
n→+∞ Dα,F u (Ω∗ )

for every Borel subset B. µF u has a density bounded away from zero and
infinite, so it is equivalent to the dynamical measure class. Next, we use absolute
continuity or arguments in [8, Theorem 2.1] to pull back µF u to a T-invariant
ergodic Borel probability µ∗ supported on Ω∗ .
Proposition 6.1.
µ | Ω∗
µ∗ = .
µ(Ω∗ )
Proof. We shall prove that there is a sequence of horseshoes Ω∗n ⊂ Ω∗ and
equilibrium states µ∗n supported on Ω∗n such that µ∗n −→ M∗ , the restriction
of µ to Ω∗ normalized to a probability. For this we let {φi }i>0 a enumerable
dense subset of continuous functions on Ω∗ . Then we define Gρ,n,N a set of
quasi-generic points for M∗ as the set of points x ∈ R such that
m−1 Z
1 X
φi (T i (x)) − φi (z) dM∗ (z) < ρ i = 1, · · · , n ∀ m ≥ N. (20)
m i=0

Let also  > 0 such that |φi (x)−φi (y)| < ρ for i = 1, · · · , n whenever d(x, y) < .
We claim that there exists N = N (ρ, n, ) > 0 such that Gρ,n,m is saturated
in the stable direction for m ≥ N . In fact, for every  > 0 there exists N0 =
N0 () > 0 such that d(T i (x), T i (y)) <  for every i ≥ N0 whenever F s (x) =
F s (y). Here we use the rate of contraction along stable leaves provided by the
hyperbolic product structure in Ω∗ . Thus, suppose that y ∈ F s (x), then

m−1 m−1
1 X j 1 X 2N0 max1≤i≤n kφi k∞ m − N0
φi (T (x)) − φi (T j (y)) ≤ + ρ.
m j=0 m j=0 m m

So, for N = N (ρ, n, ) > 0 we have

m−1 m−1
1 X 1 X
φi (T j (x)) − φi (T j (y)) ≤ 2ρ ∀ m ≥ N. (21)
m j=0 m j=0

Thus, if x ∈ Gρ,n,N implies that F s (x) ⊂ G3ρ,n,N .

Now, for given ρ and n it holds M∗ (Gρ,n,m ) −→ 1 as m −→ +∞. This is
true since µ restricted to Ω∗ is T invariant and ergodic. So, let 0 < δ < 1 and

20
choose N = N (δ) the least integer such that Gρ,n,m is saturated in the stable
SN
direction and M∗ (Gρ,n,m ) ≥ 1 − δ for every m ≥ N and Ω∗N,ρ = i=1 Ωi has
M∗ (Gρ,n,N ∆Ω∗N,ρ ) = 0, where ∆Sdenotes the set-theoretic symmetric difference.
This is possible since µ(Ω∗ − i Ωi ) = 0. Ω∗N,ρ is a horseshoe with finitely
many hyperbolic branches. Further, µ∗N,ρ , the Gibbs measure for the potential
φ∗ = − dimu (Ω∗N,ρ )J u T , is the (unique) asymptotic measure the class of the
dynamical measure. Now, take a generic point z ∈ Ω∗N,ρ for µ∗N,ρ and choose
m ≥ N such that

m−1 Z
1 X
φi (T (z)) − φi (x) dµ∗N,ρ (x) < ρ
j
for i = 1, · · · , n,
m j=0

and z 0 ∈ Gρ,n,N such that F s (z) = F s (z 0 ). Therefore, using (20) and (21) we
get for N large enough and for every i = 1, · · · , n
Z Z
φi (z) dM∗ (z) − φi (z) dµ∗N,ρ (z) < 4ρ. (22)

From the above arguments we get a sequence of horseshoes Ω∗n with finitely many
branches and equilibrium states µ∗n such that µ∗n −→ M∗ vaguely. We conclude
that M∗ is absolutely continuous with respect to the dynamical measure class,
using as we did before, a Frostman’s Lemma type argument. Also we get α =
dimu (Ω∗ ). Therefore, M∗ is absolutely continuous respect to µ∗ , so they are
equal, because they are ergodic.

Departamento de Matemáticas y Computación, Facultad de Cien-

cias. Universidad del Zulia

email : [email protected]

References
[1] Abramov, L.M. : On the entropy of an induced map. Translations of the
AMS 49, 167 -170 (1966).
[2] Billingsley, P.: Ergodic Theory and Information. Wiley & Sons. 1965.

[3] Newhouse, S. - Jakobson, M. : On the structure of non hyperbolic attractors.

Preprint.
[4] Katok, A. and Hasselblat, B. : Introduction to the Modern Theory of Dy-
namical Systems, Encyclopedia of Math. and its Applications. Vol. 54 . Cam-
bridge University Press, 1995.

21
[5] Katok, A. - Mendoza, L. : Dynamical systems with non-uniformly hyperbolic
behavior. Supplement to Introduction to the Modern Theory of Dynamical
Systems, by Katok, A. and Hasselblat, B..
[6] Ledrappier, F. : Propietés ergódiques des mesures de Sinai. Publ. Mathe-
matiques IHES, 59 (1984).

[7] Ledrappier, F. - L.-S. Young : Entropy of diffeomorphisms I, II. Ann. of

Math., 1985.
[8] Mañé, R : Ergodic Theory and Differentiable Dynamics. Springer-Verlag.
1987.

[9] Mendoza, L. : Ergodic attractors for diffeomorphisms of surfaces. Journal

of the London Math. Soc. , 1988.
[10] Pesin, Ya. : On the notion of dimension with respect to a dynamical system.
Erg. Th. and Dyn. Sys. 4, (1984).

[11] Pesin, Ya. : Dimension Theory and Dynamical Systems. Chicago Lectures
in Mathematics.Chicago, 1997.
[12] Pugh, C. - Shub, M. : Ergodic Attractors. Transactions of the AMS. Vol.
312, No. 1, 1989.
[13] Sánchez-Salas, Fernando J. : Some Geometric Properties of Ergodic At-
tractors. Informes de Matemática, Série F-112-Janeiro/99, IMPA.
[14] Young, L. -S.: Ergodic Theory of Attractors. Proc. ICM-94, v.2, 1221-1229.
Birkhäuser 1995.
[15] Young, L. -S.: Statistical Properties of Dynamical Systems with Some Hy-
perbolicity. Annals of Math. 147 (1998), 585 - 650.
[16] Viana, M. : Proc. ICM-94, v.2, 1221-1229. Birkhäuser 1995.

22