, , November 27, 2006

In [9] Keller proved for any non-flat

S

-unimodal map

f

the existence of an absolutely continuous invariant measure whenever there exists a constant

λ

such that for Lebesgue almost all

x

\begin{matrix} {limsup}_{n \to \infty} \frac{1}{n} \sum_{j = 0}^{n - 1} log | D f (f^{j} (x)) | \geq λ > 0 . \end{matrix}

(1)

This measure is called a physical or SRB (Sinai-Ruelle-Bowen) measure.

The result of Keller was somewhat generalized by Alves-Bonatti-Viana [4] in the context of non-flat maps with slow recurrence to the critical set defined on a compact Riemannian manifold

M

. For this, they assumed that

\begin{matrix} {liminf}_{n \to \infty} \frac{1}{n} \sum_{j = 0}^{n - 1} log (∥ (D f (f^{j} (x)))^{- 1} ∥^{- 1}) \geq λ > 0 \end{matrix}

(2)

for Lebesgue almost all

x \in M

. Of course, if the dimension of

M

is one we have

| D f (x) | = ∥ (D f (x))^{- 1} ∥^{- 1}

. Observe that

∥ (D f (x))^{- 1} ∥^{- 1} > 1

means that

D f (x)

expands in all directions, that is,

∥ (D f (x)) v ∥ \geq γ | v |

\forall v \in T_{x} M

(where

γ = ∥ (D f (x))^{- 1} ∥^{- 1}

). Systems satisfying ( 2 ) are called non-uniformly expanding, they generalize the expanding systems. As particular examples of this kind of systems we can mention one-dimensional maps with positive Lyapunov exponents ( like quadratic maps and, more in generally, non-flat multimodal maps [11] ), and in higher dimension, the Viana maps [12] .

The purpose of this paper is to generalize the result of Keller in the context of [4] with the far more weak condition of

limsup

as it appears in Keller's theorem, i.e.,

\begin{matrix} {limsup}_{n \to \infty} \frac{1}{n} \sum_{j = 0}^{n - 1} log (∥ (D f (f^{j} (x)))^{- 1} ∥^{- 1}) \geq λ > 0 . \end{matrix}

(3)

Moreover, we prove that the condition ( 3 ) indeed implies ( 2 ) and so, the definition of non-uniformly expanding map can be weakened.

Statement of results

Let

M

be a compact Riemannian manifold of dimension

d \geq 1

and

Leb

a normalized Riemannian volume form on

M

that we call Lebesgue measure.

A map

f : M \to M

will be called non-flat if

f

is local

C^{2}

diffeomorphism in the whole manifold except in a non-degenerate critical set

C \subset M

The definition of a non-degenerate critical set is given in the beginning of the section 2 ( If

d i m (M) = 1

this definition is the usual definition of non-flatness, see [11] ).

We say that

f

satisfies the slow approximation condition if given any

ε > 0

there exists

δ > 0

such that for Lebesgue almost every point

x \in M

we have

\begin{matrix} {limsup}_{n \to + \infty} \frac{1}{n} \sum_{j = 0}^{n - 1} - log {dist}_{δ} (f^{j} (x), C) \leq ε, \end{matrix}

(*)

where

d_{δ} (x, C)

denotes the

δ

-truncated distance from

x

C

defined as

d_{δ} (x, C) = d (x, C)

d (x, C) \leq δ

and

d_{δ} (x, C) = 1

otherwise.

We call the basin of some invariant measure

ν

the set

ℬ (ν)

of the points

x \in M

such that the average of Dirac measures along the orbit of converge in the weak

^{*}

topology to

ν

, that is,

{lim}_{n \to + \infty} \frac{1}{n} \sum_{j = 0}^{n - 1} φ (f^{j} (x)) = \int φ d ν, \forall φ \in C^{0} (M) .

Theorem 1. Let

f : M \to M

be a non-flat map satisfying the slow approximation condition. If

f

( or some fixed iterate ) satisfies

\begin{matrix} {liminf}_{n \to \infty} \frac{1}{n} \sum_{i = 0}^{n - 1} log (∥ (D f (f^{i} (x)))^{- 1} ∥) \leq - λ < 0 \end{matrix}

(4)

(or equivalently satisfies ( 3 )) for Lebesgue almost every point

x \in M

then there exists a finite collection of ergodic absolutely continuous invariant measures such that almost every point in

M

belongs to the basin of one of these measures.

Theorem 1 is in fact a consequence of the Main Theorem below which assures the existence of a global Markov structure with integrable time function. A global Markov structure is composted by a time function

x \mapsto R (x) \in N \cup {\infty}

, an induced map

F (x) = f^{R (x)} (x)

defined almost everywhere and a countably partition refining a finite triangulation of

M

. Each element of this partition is sent by

F

, with good properties ( see section 2 for details ), onto an element of the triangulation. One can prove that such

F

has an absolutely continuous invariant measure

ν

. Moreover, whenever

R

ν

-integrable it gives rise to an absolutely continuous

f

-invariant finite measure. We observe that the existence of a Markov structure allows a more deep study of the dynamical's properties the map

f

. For instance, it was used in [5, 7, 8, 13, 14] to study the decay of correlations and prove the Central Limit Theorem for a large class of maps.

Main Theorem. Every map satisfying the hypothesis of theorem 1 has a global Markov Structure with integrable time function.

The proof of the main theorem is a mix of the strategy adopted by Alves-Luzzatto-Pinheiro in [5] for non-uniform expanding maps ( inspired in Young's paper [13] ) and the proof of Keller's theorem in [11] . Here we are able to simplify the proof of the existence of the Markov Structure that appears in [5, 8, 13] and make it closer to the one dimensional case.

A map

f : M \to M

is called non-uniformly expanding if

f

is a non-flat map satisfying the slow approximation condition and it ( or some fixed iterate ) satisfies for Lebesgue almost every point

x \in M

the following condition,

\begin{matrix} {limsup}_{n \to \infty} \frac{1}{n} \sum_{i = 0}^{n - 1} log (∥ (D f (f^{i} (x)))^{- 1} ∥) \leq - λ < 0 \end{matrix}

(5)

(or equivalently satisfies ( 2 )).

Theorem 2. A map is non-uniformly expanding if and only if it satisfies the hypothesis of theorem 1 .

A crucial ingredient in the proof of these results is the existence, for almost all point

x \in M

, of moments

n = n (x) \in N

such that

f

looks like an expanding map on some neighborhood of

x

, that is, this neighborhood

V_{n} (x)

is sent by

f^{n}

, with good properties of expansion and distortion, to some definitively big ball

B (f^{x} (x))

centered in

f^{n} (x)

The last result of this paper is about hyperbolic times which are associated with theses moments of expansion mentioned above ( see Proposition 2.4 ). In [3] Alves-Araujo proved the existence of a SRB measure when the critical set is empty and the first hyperbolic time is Lebesgue integrable . Here, we were able to remove the hypothesis of integrability.

Theorem 3. Let

f : M \to M

be a

C^{2}

covering map (local diffeomorphism) on a compact manifold M. If the first hyperbolic time function is defined for Lebesgue almost every point of M then

f

is a non-uniformly expanding map.

2 Hyperbolic Times

Definition 2.1. Let

f : M \to M

be a

C^{2}

local diffeomorphism in the whole manifold except in a critical set

C \subset M

. We say that

C \subset M

is a non-degenerate critical set, more precisely, a

β

-non-degenerate critical set if

\exists B > 0

such that the following three conditions hold.

(1) $\frac{1}{B} d i s t (x, C)^{β} \leq \frac{∥ D f (x) v ∥}{∥ v ∥} \leq B$ for all $v \in T_{x} M$ .

For every

x, y \in M \ C

with

d i s t (x, y) < d i s t (x, C) / 2

we have

(2) $| log ∥ D f (x)^{- 1} ∥ - log ∥ D f (y)^{- 1} ∥ | \leq \frac{B}{d i s t (x, C)^{β}} d i s t (x, y)$ ;
(3) $| log | det D f (x) | - log | det D f (y) | | \leq \frac{B}{d i s t (x, C)^{β}} d i s t (x, y)$ .

Let us fix

0 < b < \frac{1}{2} min {1, 1 / β}

. If

f : M \to M

be a

C^{2}

local diffeomorphism outside a

β

-non-degenerate critical set

C

then, given

0 < σ < 1

and

δ > 0

, we will say that

n

is a

(σ, δ)

-hyperbolic time for a point

x \in M

if for all

1 \leq k \leq n

we have

\prod_{j = n - k}^{n - 1} ∥ (D f \circ f^{j} (x))^{- 1} ∥ \leq σ^{k} and

d i s t_{δ} (f^{n - k}, C) \geq σ^{b k}

. We denote the set of points of

M

such that

n \in N

(σ, δ)

-hyperbolic time by

H_{n} (σ, δ)

Proposition 2.2. [4] Let

f : M \to M

be a

C^{2}

non-flat map satisfying the slow approximation condition. Given

λ > 0

there exists

δ > 0

and

θ > 0

, depending only on

f

and

λ

, such that

# {1 \leq j \leq n ∥ x \in H_{j} (e^{- λ / 4}, δ)} \geq θ n,

whenever

\sum_{i = 0}^{n - 1} log ∥ (D f (f^{i} (x)))^{- 1} ∥^{- 1} > λ n

Corollary 2.3. Let

f : M \to M

be a

C^{2}

non-flat map satisfying the slow approximation condition. Given

λ > 0

there exists

δ > 0

and

θ > 0

, depending only on

f

and

λ

, such that if

A \subset M

with

limsup \frac{1}{n} \sum_{i = 0}^{n - 1} log ∥ (D f (f^{i} (x)))^{- 1} ∥^{- 1} > λ

for Lebesgue almost all

x \in A

then

{limsup}_{n} \frac{1}{n} # {1 \leq j \leq n ∥ x \in H_{j} (e^{- λ / 4}, δ)} \geq θ

for Lebesgue almost all

x \in A

We want to finish this section enunciating a proposition that assures a good behavior, with respect to

f^{n}

, of a neighborhood of a point

x

when

n

is a hyperbolic time to this point.

Proposition 2.4. Let

f : M \to M

be a

C^{2}

local diffeomorphism outside a non-degenerate critical set

C

. Given

σ < 1

and

δ > 0

, there exist

δ_{1}, ρ > 0

, depending only on

σ, δ

and on the map

f

, such that for any

x \in M

and

n \geq 1

(σ, δ)

-hyperbolic time for

x

, there exists a neighborhood

V_{n} (x)

x

with the following properties:

(1) $f^{n}$ maps $V_{n} (x)$ diffeomorphically onto the ball $B_{δ_{1}} (f^{n} (x))$ ;
(2) $d i s t (f^{n - j} (p), f^{n - j} (q)) \leq σ^{j / 2} dist (f^{n} (p), f^{n} (q))$ $\forall p, q \in V_{n} (x)$ and $1 \leq j < n$ ;
(3) $log | \frac{det D f^{n} (p)}{det D f^{n} (q)} | \leq ρ dist (f^{n} (p), f^{n} (q))$ $\forall p$ and $q \in V_{n} (x)$ .

We shall often refer to the sets

V_{n} (x)

as hyperbolic pre-balls and to their images

f^{n} (V_{n} (x))

as hyperbolic balls. Notice that the latter are indeed balls of radius

δ_{1}

Proof. For the proofs of items 1 and 2 see Lemma 5.2 in [4] . Let $p$ and $q \in V_{n} (x)$ . As $n$ is a hyperbolic time for $x$ , it follows from the item 2 above that $dist (f^{j} (q), C) \geq dist (f^{j} (x), C) - dist (f^{j} (x), f^{j} (q)) \geq σ^{b (n - j)} - δ_{1} σ^{(n - j) / 2} \geq (1 - δ_{1}) σ^{b (n - j)}$ . Now, using the condition 3 of the definition of a non-degenerate critical set we get $log | \frac{det D f^{n} (q)}{det D f^{n} (p)} | \leq \sum_{j = 0}^{n - 1} log | \frac{det D f (f^{j} (q))}{det D f (f^{j} (p))} | \leq$ $\leq \sum_{j = 0}^{n - 1} B \frac{σ^{(n - j) / 2} dist (f^{n} (q), f^{n} (p))}{((1 - δ_{1}) σ^{b (n - j)})^{β}} \leq$ $\leq \frac{B}{(1 - δ_{1})^{β} (1 - σ^{1 / 2 - b β})} dist (f^{n} (q), f^{n} (p))$ □

Markov Structures and The Partitioning Algorithm

In this section we will define what we mean by a Global Markov Structure and prove the first part of the Main Theorem. A Global Markov Structure for a map

f : M \to M

is a piecewise expanding Markovian map induced by it, that is, there exists a open subset

M^{'}

with Lebesgue total measure (

L e b (M \ M^{'}) = 0

) and a function

R : M^{'} \to N \cup {+ \infty}

(called time function) such that the induced map

F : M^{'} \to M

given by

F (x) = f^{R (x)} (x)

is a piecewise expanding Markovian map.

Piecewise expanding Markovian map

Let

F

be a map defined a open subset of

M

with full Lebesgue measure.

F

will be called a piecewise expanding Markovian map if there is a countable partition

P

of its domain into open sets and a finite collection of

{P_{1}, . . ., P_{n}}

of topological balls of

M

( for instance, a triangulation of M ) such that:

(1) Expansion: there is $0 < κ < 1$ such that for each $U \in P$ and $x \in U$ $∥ D F (x)^{- 1} ∥ < κ .$
(2) Bounded distortion: there is $K > 0$ such that for each $U \in P$ and $x, y \in U$ $log | \frac{det D F (x)}{det D F (y)} | \leq K dist (F (x), F (y)) .$
(3) Long branches:
- (a) Each $P_{j}$ have a piecewise $C^{2}$ boundary with a finite $(d - 1)$ -dimensional volume. Moreover, there is some $β \in (\frac{κ}{1 - κ}, 1]$ such that the $C^{2}$ components of the boundary of each $P_{j}$ meet at angles greater than $arcsin (β) > 0$ .
- (b) For each $U \in P$ , $F |_{U}$ is a $C^{2}$ diffeomorphism onto some element of ${P_{1}, . . ., P_{n}}$ .

The theorem below assures that every

C^{2}

piecewise expanding Markovian map

F : Δ \to Δ

has some absolutely continuous invariant measure

ν

whose density belongs to

L^{\infty} (L e b)

; see e.g. Lemma

4.4.1

of [1] .

Moreover, it is straightforward to check that if

R

ν

-integrable then

μ = \sum_{j = 0}^{\infty} f_{*}^{j} (ν | {R > j})

is an absolutely continuous

f

-invariant measure. Here

ν | {R > j}

denotes the measure given by

ν | {R > j} (A) = ν (A \cap {R > j})

, and

f_{*}^{j}

denotes the push-forward of the measure by

f^{j}

We will say that the Markov Structure has integrable time function if

R

is integrable with respect to any absolutely continuous

F

-invariant measures. As a consequence to the the theorem below, if we want to show that Markov Structure has integrable time function , we need only to verify the integrability of

R

with respect to the finite collection of ergodic absolutely continuous invariant measures gives by this theorem.

Theorem 4. [1, 2, 6] If

F : Δ \to Δ

is a

C^{2}

piecewise expanding Markovian map, then there exists a finite set of ergodic absolutely continuous invariant measures such that Lebesgue almost all point of

Δ

belongs to the basin of one of these measures. Moreover, the density of each of these measure with respect to Lebesgue is uniformly bounded by some constant.

The Partitioning Algorithm

Let

f : M \to M

be a

C^{2}

non-flat map satisfying the slow approximation condition and

λ_{0} > 0

. Suppose that

{limsup}_{n \to \infty} \frac{1}{n} \sum_{i = 0}^{n - 1} log ∥ (D f (f^{i} (x)))^{- 1} ∥^{- 1} > λ_{0}

for Lebesgue almost every

x \in M

. In this section we will show how one can construct a Markovian Structure on

M

Replacing, if is necessary,

f

by some

f^{n_{0}}

we may assume, without loss of generality, that

λ_{0} > 16 log 2

Let

λ = λ_{0} / 4

and

σ = exp (- λ) < \frac{1}{16}

be fixed from now on. Let

δ > 0

be given by the Proposition 2.2 and let

δ_{1} > 0

be the radius of the hyperbolic ball given by Proposition 2.4 . For short, let's put for

j \geq 1

H_{j} = H_{j} (σ, δ)

Remark 2.5. Observe that it follows from Corollary 2.3 the existence of a constant

θ > 0

such that

{limsup}_{n} \frac{1}{n} # {1 \leq j \leq n ∥ x \in H_{j}} \geq θ

for Lebesgue almost all

x \in M

Choose some

0 < δ_{0} < δ_{1} / 4

and a finite partition

P

M

(mod 0) generated by the interior of the elements of a triangulation of

M

with diameter smaller then

δ_{0}

, that is,

d i a m e t e r (P) < δ_{0}

\forall P \in P

. Given

U \subset M

and

0 < r

define the

r

-neighborhood of

U

B_{r} (U) = {x \in M ∥ d i s t (x, U) < r}

. For each element

P \in P

set

P^{1} = B_{δ_{0}} (P)

and consider the follows collections of rings. The first collection,

{I_{k}^{0} (P)}_{k \in N}

, is a partition (mod 0) of

B_{δ_{0}} (\partial P) \cap P

defined by

I_{k}^{0} (P) = {x \in P ∥ δ_{0} σ^{k / 2} \leq dist (x, \partial P) < δ_{0} σ^{(k - 1) / 2}} .

The second one,

{I_{k}^{1} (P)}_{k \in N}

, is a partition (mod 0) of

P^{1} \ P

established by

I_{k}^{1} (P) = {x \in P^{1} \ P ∥ δ_{0} σ^{k / 2} \leq dist (x, \partial P) < δ_{0} σ^{(k - 1) / 2}} .

Let us fix an element

P_{0} \in P

and set

Δ_{0} (P_{0}) = P_{0}

. Let

k_{0}

be the first integer such that

Δ_{0} (P_{0}) \ \cup_{k \geq k_{0}} I_{k}^{0} (P_{0}) \neq \emptyset

. For each

0 \leq k < k_{0}

we will set

A_{k} (P_{0}) = \emptyset

and

B_{k} (P_{0}) = Δ_{k} (P_{0}) = Δ_{0} (P_{0})

. Define the function

t_{0} : Δ_{0} (P_{0}) \to N

t_{0} (x) = j

x \in I_{j}^{0} (P)

. Inductively, set for every

0 \leq k < k_{0}

the functions

t_{k} : Δ_{k} (P_{0}) \to N

t_{k} (x) = t_{k - 1} (x) - 1

. We will define by induction the sets

Δ_{k} (P_{0})

A_{k} (P_{0})

B_{k} (P_{0})

and functions

t_{k} : Δ_{i} \to N

for every

k \in N

. For this, let us assume that for some

n \geq k_{0}

the sets

Δ_{i} (P_{0})

A_{i} (P_{0})

B_{i} (P_{0})

and functions

t_{i} : Δ_{i} \to N

are defined for all

i \leq n - 1

For Lebesgue almost every point

x \in A_{n - 1} (P_{0}) \cap H_{n}

there exists some

P \in P

such that

f^{n} (x) \in P

. As the diameter of

P^{1}

is smaller then

δ_{1} / 2

we also have

B_{δ_{1}} (f^{n}) (x)) \supset P^{1} \supset P

. Proposition 2.4 assures that there exists a hyperbolic pre-ball

V_{n} (x)

such that

f^{n} | V_{n} (x) : V_{n} (x) \to B_{δ_{1}} (f^{n} (x))

is a diffeomorphism with uniform bounded distortion ( not depending on

x

n

). Denote by

U_{n} (P_{0})

the collections of all sets

(f^{n} | V_{n} (x))^{- 1} (P^{1})

where

x \in A_{n - 1} (P_{0}) \cap H_{n}

and

f^{n} (x) \in P \in P

. It is easy to check that the elements of

U_{n} (P_{0})

are two by two disjoints.

Given

U \in U_{n} (P_{0})

, with

U = (f^{n} | V_{n} (x))^{- 1} (P^{1})

and

P \in P

, let's put

U^{0} = (f^{n} | V_{n} (x))^{- 1} (P)

and

U (j) = (f^{n} | V_{n} (x))^{- 1} (I_{j}^{1} (P))

. Now, let's set

Δ_{n} (P_{0}) = Δ_{n - 1} (P_{0}) \ ⋃_{U \in U_{n} (P_{0})} U^{0}

and define the function

t_{n} : Δ_{n} \to N

t_{n} (x) = {\begin{matrix} j & if x \in U (j) for some j \geq 1 and U \in U_{n} (P_{0}), \\ 0 & if x \in A_{n - 1} \ \cup_{U \in U_{n} (P_{0})} U, \\ t_{n - 1} (x) - 1 & if x \in B_{n - 1} \ \cup_{U \in U_{n} (P_{0})} U . \end{matrix}

With the function

t_{n}

we define

A_{n} (P_{0})

and

B_{n} (P_{0})

A_{n} (P_{0}) = {n \in Δ_{n} (P_{0}) ∥ t_{n} = 0}

and

B_{n} (P_{0}) = {x \in Δ_{n} (P_{0}) ∥ t_{n} > 0} .

At this point we have completely described the inductive construction restrict to

P_{0}

of the sets

A_{n} (P_{0})

B_{n} (P_{0})

and

Δ_{n} (P_{0})

. Doing the same construction to the other elements of

P

we obtain

A_{n} = ⋃_{P \in P} A_{n} (P),

B_{n} = ⋃_{P \in P} B_{n} (P)

and

Δ_{n} = ⋃_{P \in P} Δ_{n} (P) .

Let

M^{*} = {x \in M ∥ f^{n} (x) \in Δ_{0} \forall n \geq 0}

. As

f

is a diffeomorphism Lebesgue almost everywhere, we have

L e b (M \ M^{*}) = 0

. Finally, let's define a time function

R : M^{*} \ ⋂_{n} Δ_{n} \to N

putting

R (x) = n

when

x \in Δ_{n - 1} \ Δ_{n}

We can think of the set

A_{n}

as the set of allowed points of the stage

n + 1

, that is, the set of points

x \in M

which in the stage

n + 1

can be used, if

n + 1

is a hyperbolic time to

x

, as the center of a new element of the partition ( a component of

{R = n + 1}

). The set

B_{n}

is the forbidden points of the stage

n + 1

, i.e., if

x \in B_{n}

then, although

x

does not belong to a constructed element (

x / \in {R \leq n}

), we can not use

x

to be the “center” of a new element even if

x \in H_{n + 1}

. In fact,

B_{n}

in a union of “protection collars” associated to the components of

{R \leq n}

and we can not use the points of

B_{n}

in the stage

n + 1

because in this time the pre-image of the elements of the triangulation associated to some hyperbolic pre-ball

V_{n + 1} (x)

(with

x \in H_{n + 1}

) may be big compared with these collars and probably there will be overlaps of this pre-image with some ahead constructed element.

Remark 2.6. It is important to emphasize that we only constructed a new element of the partition as a pre-image, associated to a hyperbolic time, of some element of the triangulation.

That is, if

U

is a component of

{R = n}

, for some

n

, then

U \subset V_{n} (x)

for some

x \in A_{n - 1} \cap H_{n}

. In consequence,

U

is mapped diffeomorphically, with bounded distortion and good properties of expansion (see proposition 2.4 ), to some element of the triangulation. This control of distortion and expansion is fundamental to obtain the properties required from the Piecewise expanding Markovian map associated to the Markov structure.

The lemma below is important for preventing the overlaps on the sets of the partition. Indeed, in the stage

n

, associated to each component

U_{n - k}^{0}

{R = n - k}

, for some

k > 0

, there is a collar

Q_{k} \subset U_{n - k}^{1}

around it. By the lemma, the new components of

{R = n}

do not intersect “too much” the collar and in consequence, there will be no overlaps.

Lemma 2.7. If

U \in U_{n} = \cup_{P \in P} U_{n} (P)

then

U \cap {t_{n - 1} > 1} = \emptyset

Proof. Take some

k > 0

and let

U_{n - k}^{0}

be a component of

{R = n - k} = Δ_{n - k} \ Δ_{n - k - 1}

such that its collar

Q_{k}

(the part of

U_{n - k}^{1}

that is mapped by

f^{n - k}

onto

I_{k}^{1} (P)

for some

P \in P

) intersects

U = U_{n}^{1} \in U_{n}

. Recall that

Q_{k}

is precisely the collar around

U_{n - k}^{0}

on which

t_{n - 1}

takes the value 1. Letting

q_{1}

and

q_{2}

be any two points in distinct components of the boundary of

Q_{k}

, we have by Proposition 2.4

\begin{matrix} dist (f^{n - k} (q_{1}), f^{n - k} (q_{2})) \leq σ^{k / 2} dist (f^{n} (q_{1}), f^{n} (q_{2})) . \end{matrix}

(6)

We also have

\begin{matrix} dist (f^{n - k} (q_{1}), f^{n - k} (q_{2})) & \geq & δ_{0} σ^{(k - 1) / 2} - δ_{0} σ^{k / 2} \end{matrix}

\begin{matrix} = & δ_{0} σ^{k / 2} (σ^{- 1 / 2} - 1), \end{matrix}

which combined with ( 6 ) gives

dist (f^{n} (q_{1}), f^{n} (q_{2})) \geq δ_{0} (σ^{- 1 / 2} - 1) .

On the other hand, since

σ < 1 / 16

we conclude that

dist (f^{n} (q_{1}), f^{n} (q_{2})) > 3 δ_{0} > diameter (B_{δ_{0}} (P))

but this is impossible as

q_{1}

and

q_{2} \in B_{δ_{0}} (P)

. □

Corollary 2.8. If

U

and

V

are distinctly components of

{R = n}

and

{R = m}

then

U \cap V = \emptyset

Proof. Suppose that $U = (f^{n} |_{V_{n} (x_{0})})^{- 1} (P_{0})$ and $V = (f^{m} |_{V_{m} (x_{1})})^{- 1} (P_{1})$ , with $P_{0}$ and $P_{1} \in P$ .
As we observed during the presentation of the algorithm, if $n = m$ and $P_{0} = P_{1}$ we will have $U \cap V = \emptyset$ . Indeed, as $P_{0} = P_{1}$ are small compared with $B_{δ_{1}} (f^{n} (x_{1}))$ or $B_{δ_{1}} (f^{n} (x_{2}))$ ( $d i a m e t e r (P_{0}) < δ_{0} < δ_{1} / 4$ ) we will have $(f^{n} |_{V_{n} (x_{0})})^{- 1} |_{P_{0}} = (f^{n} |_{V_{n} (x_{1})})^{- 1} |_{P_{0}}$ and so, $U = V$ . Of course, if $P_{0} \neq P_{1}$ , with $n = m$ , than $f^{n} (U) \cap f^{n} (V) = P_{0} \cap P_{1} = \emptyset$ and so, $U \cap V = \emptyset$ . Finally, assuming that $m < n$ , it follows from the lemma above that $U \cap {t_{n - 1} > 1} = \emptyset$ . We know that $V$ is protected by some collar $Q \subset {t_{n - 1} > 1}$ , i.e., there is a connected component $C$ of $M \ Q$ such that $C \supset V$ and $C \cap A_{n - 1} = \emptyset$ . In consequence, as $U \cap A_{n - 1} \neq \emptyset$ and $Q \cap U = \emptyset$ , it follows from connectivity of $U$ that $V \cap U \subset C \cap U = \emptyset$ . □

Proposition 2.9 and Lemma 2.10 will be used to prove that the sum of the Lebesgue measure of all forbidden sets is finite (proposition 2.11 ).

This finiteness is fundamental to assure that almost all point of

M

will belongs to an construct element, i.e.,

L e b (M \ \cup_{n} {R = n}) = 0

Proposition 2.9. There exists

a_{1} > 0

such that for every

n \geq 1

Leb (B_{n - 1} \cap A_{n}) \geq a_{1} Leb (B_{n - 1})

Proof. It is enough to see this for each component of $B_{n - 1}$ at a time.
Let $C$ be a component of $B_{n - 1}$ and $Q$ be its outer ring corresponding to $t_{n - 1} = 1$ . Observe that by Lemma 2.7 we have $Q = C \cap A_{n}$ . Moreover, there must be some $k < n$ , an element $P$ of $P$ and a component $U_{k}^{0}$ of ${R = k}$ such that $f^{k}$ maps $C$ diffeomorphically onto some $\cup_{i = k}^{\infty} I_{i}^{1} (P)$ and $Q$ onto $I_{k}^{1} (P)$ , both with uniform bounded distortion (not depending $n$ ). Thus, it is sufficient to compare the Lebesgue measures of $\cup_{i = k}^{\infty} I_{i}^{1} (P)$ and $I_{k}^{1} (P)$ . As $P \in P$ is the interior of an element of a triangulation of $M$ , the boundary of $P$ is a finite union of smooth submanifolds with codimension one. Thus, for smalls $r_{2} > r_{1} > 0$ we have $L e b (B_{r_{2}} (P) \ B_{r_{1}} (P)) \approx r_{2} - r_{1}$ . As $P$ is a finite collection one can find $ζ > 0$ such that $\frac{1}{ζ} < \frac{L e b (B_{r_{2}} (P) \ B_{r_{1}} (P))}{r_{2} - r_{1}} < ζ$ $\forall P \in P$ and $\forall r_{2} > r_{1} > 0$ . In consequence, setting $C_{1} = exp (ρ diameter (M))$ (see proposition 2.4 ), we have $\frac{Leb (B_{n - 1} \cap A_{n})}{Leb (B_{n - 1})} \geq C_{1} \frac{Leb (I_{k}^{1} (P))}{Leb (\cup_{i = k}^{\infty} I_{i}^{1} (P))} \geq$ $\geq C_{1} \frac{1}{ζ^{2}} \frac{δ_{0} σ^{(k - 1) / 2} - δ_{0} σ^{k / 2}}{\sum_{i = k}^{\infty} (δ_{0} σ^{(i - 1) / 2} - δ_{0} σ^{i / 2})} = C_{1} \frac{1 - σ^{1 / 2}}{ζ^{2}} .$ □

Lemma 2.10. There exist a constant

C_{2} > 0

such that

L e b (A_{n - 1} \cap B_{n}) \leq C_{2} L e b ({R = n})

\forall n > N_{0}

Proof. By construction, each component $W$ of $A_{n - 1} \cap B_{n}$ is a ring given (mod 0) by the union of rings $U (j)$ that appears when construct the element $U^{0}$ of ${R = n}$ associated to an element $U$ of $U_{n}$ and can be written as $W = (f^{n} | V_{n} (x))^{- 1} (P^{1} \ P) = U \ U^{0} (mod 0),$ where $U = (f^{n} | V_{n} (x))^{- 1} (P^{1})$ , $U^{0} = (f^{n} | V_{n} (x))^{- 1} (P)$ , $P \in P$ and $x \in A_{n - 1} \cap H_{n}$ . Moreover, $W$ and $U^{0}$ are contained in the hyperbolic pre-ball $V_{n} (x)$ associated a which are mapped onto $B_{δ_{1}} (f^{n}) (x)) \supset P^{1} \supset P$ by $f^{n}$ . By the bounded distortion property, there is an universal $C_{1} > 0$ such that $L e b (W) = L e b (U \ U^{0}) \leq C_{1} \frac{L e b (P^{1} \ P)}{L e b (P)} L e b (U^{0})$ and by disjointness two by two of the elements $U_{n}$ we get $L e b (A_{n - 1} \cap B_{n}) = L e b (⋃_{U \in U_{n}} U \ U^{0}) = \sum_{U \in U_{n}} L e b (U \ U^{0}) \leq$ $\leq C_{2} \sum_{U \in U_{n}} L e b (U^{0}) = C_{2} L e b ({R = n}),$ where $C_{2} = C_{1} max {\frac{L e b (P^{1} \ P)}{L e b (P)} ∥ P \in P}$ . □

Proposition 2.11.

\sum_{n = 0}^{\infty} L e b (B_{n}) < \infty

Proof. We may write $L e b (B_{n - 1}) = L e b (B_{n - 1} \cap B_{n}) + L e b (B_{n - 1} \cap A_{n})$ which by Proposition 2.9 gives $L e b (B_{n - 1} \cap B_{n}) \leq (1 - a_{1}) L e b (B_{n - 1})$ .
Thus, using Lemma 2.10 , $L e b (B_{n}) = L e b (B_{n - 1} \cap B_{n}) + L e b (A_{n - 1} \cap B_{n}) \leq (1 - a_{1}) L e b (B_{n - 1}) + C_{2} L e b ({R = n})$ $\forall n > N_{0}$ . Summing in both sides of this inequality, we obtain for every $m \in N$ $\sum_{n = 1}^{m} L e b (B_{n}) \leq (1 - a_{1}) \sum_{n = 1}^{m} L e b (B_{n - 1}) + C_{2} \sum_{n = 1}^{m} L e b ({R = n}) \leq$ $\leq (1 - a_{1}) \sum_{n = 0}^{m} L e b (B_{n}) + C_{2} L e b (M)$ Thus, $a_{1} \sum_{n = 0}^{m} L e b (B_{n}) \leq L e b (B_{0}) + C_{2} L e b (M) = (1 + C_{2}) L e b (M) .$ □

Now, we are going to prove that this algorithm does indeed produce a partition mod 0 of

M

. As Corollary 2.8 assures that the elements constructed by this algorithm are two by two disjointed, we have only to check that

L e b (\cap_{n} Δ_{n}) = 0

(

Δ_{0} \supset Δ_{1} \supset . . .

is a nested sequence).

To prove this, first observe that, as

\sum_{n = 0}^{\infty} L e b (B_{n}) < \infty

, it follows from Fubini's argument that Lebesgue almost every point of

M

belongs only to a finite many

{B_{j}}^{'} s

, that is, defining the function

b : M \to N

\begin{matrix} b (x) = {\begin{matrix} sup {n \in N ∥ x \in B_{n - 1}} & if x \in Δ_{0} \\ + \infty & if x / \in Δ_{0} \end{matrix}, \end{matrix}

(7)

we will have

b (x) < + \infty

for Lebesgue almost all

x \in M

(because

\int b d L e b = \sum_{n} (B_{n})

). Moreover, as any Lebesgue generic point has infinity many hyperbolic times, one can find for almost all

x \in M^{*}

the first time

n > b (x)

such that

x \in H_{n}

. In this case, if

x

still belongs to

Δ_{n - 1}

we will have

x \in {R = n} = Δ_{n - 1} \ Δ_{n}

. In other words, almost every point of

M^{*}

( and in consequence of

M

) will belong to some

Δ_{n - 1} \ Δ_{n}

. So,

L e b (\cap_{n} Δ_{n}) = 0

Markov Structures

Let

F : Δ^{*} \to M

be given by

F (x) = f^{R (x)} (x)

, where

Δ^{*} = M^{*} \ \cap_{n} Δ_{n}

. By construction, the map

F

is a piecewise expanding Markovian map (see remark 2.6 ) and we conclude the proof of the first part of the Main Theorem.

3 Integrability of The Time Function

In this section we will finish the proof of the Main Theorem showing the integrability of the time function of the Markov Structure constructed in the previous section. We will use the objects and notation of the Partition Algorithm. Let

ν

be one of the

F

-invariant measure given by the Markov Structure and assume by contradiction that

\sum_{P \in P} R |_{P} ν (P) = \infty

. It follows from Birkhoff`s Ergodic Theorem that

\frac{1}{n} \sum_{j = 0}^{n - 1} R \circ F^{j} (x) ⟶ \int R d ν = \sum_{P \in P} R |_{P} ν (P) = \infty,

for

ν

-almost every point

x \in M

As the density of

ν

is uniformly bounded from above, its follows from proposition 2.11 that

\sum_{n = 0}^{\infty} ν (B_{n}) < \infty

. In consequence, for

ν

-generic points

x \in M

we get

\frac{1}{n} \sum_{j = 0}^{n - 1} b \circ F^{j} (p) ⟶ \int b d ν = \sum_{n} ν (B_{n}) < \infty

Let

M^{* *} = {x \in M^{*} ∥ F^{j} (x) \in D o m (F) \forall j \geq 0}

. It is clear that

ν (M \ M^{* *}) = 0

. Let

x \in M^{* *}

. Set, for every

i \in N

j_{i} = j_{i} (x) = \sum_{j = 0}^{i - 1} R \circ F^{j} (x)

, that is,

F^{i} (x) = f^{j_{i}} (x)

. Given

j \in N

there exists a unique integer

s = s (j) \geq 0

such that

j_{s} \leq j < j_{s + 1}

. Let's set

ℐ_{0} = {j_{1}, j_{2},, j_{3}, . . .}

and suppose that

x \in H_{j}

. In this case,

F^{s} (x) \in H_{m}

, where

m = j - j_{s}

. By construction, if

F^{s} (x) / \in {R = m}

( i.e.

m < R (F^{s} (x))

) is because

F^{s} (x) \in B_{m - 1}

. On the other hand,

F^{s} (x) \in {R = m}

implais that

m = j_{s + 1} - j_{s}

and so,

j = j_{s + 1} \in ℐ_{0}

As the number of integers

ℓ

between

j_{i}

and

j_{i + 1}

such that

x \in H_{ℓ}

is bounded by the number of integers

m

such that

F^{i} (x) \in B_{m - 1}

, we have

# {ℓ \in {j_{i} + 1, . . ., j_{i + 1} - 1} ∥ x \in H_{ℓ}} \leq b (F^{i} (x)) .

Thus, for each

n \in N

we can write

# {j \leq n ∥ x \in H_{j}} \leq # {j \leq n ∥ j \in ℐ_{0}} + \sum_{i = 0}^{s (n)} b (F^{i} (x)) \leq

\leq s (n) + \sum_{i = 0}^{s (n)} b (F^{i} (x))

Therefore,

\begin{matrix} \frac{1}{n} # {1 \leq j \leq n ∥ x \in H_{j}} \leq \frac{s (n)}{n} (1 + \frac{1}{s (n)} \sum_{j = 0}^{s (n)} b \circ F^{j} (x)) \end{matrix}

(8)

By construction, if

s (n) = i

, that is,

j_{i} \leq n < j_{i + 1}

, then

\begin{matrix} \frac{j_{i}}{i} \leq \frac{n}{s (n)} < \frac{j_{i + 1}}{i + 1} (1 + \frac{1}{i}) \end{matrix}

(9)

\frac{j_{i}}{s (j_{i})} = \frac{j_{i}}{i} = \frac{1}{i} \sum_{j = 0}^{i - 1} R \circ F^{j} (x) \to \infty

it follows from the equations 8 and 9

lim \frac{1}{n} # {1 \leq j \leq n ∥ x \in H_{j}} = lim \frac{s (n)}{n} = 0

. But this is an absurd as one can see in Remark 2.5 . So, we necessarily have the time function

R

ν

-integrable and so, we conclude the proof of the Main Theorem.

To prove theorems 1 , 2 and 3 let

{ν_{1}, . . ., ν_{s}}

be the set of ergodic absolutely continuous invariant measures given by Theorem 4 .

Proof of Theorem 1

As we observed before it is straightforward to check that as the time function

R

ν_{i}

-integrable, then

μ_{i} = \sum_{j = 0}^{\infty} f_{*}^{j} (ν_{i} | {R > j})

is an ergodic absolutely continuous

f

-invariant measure. Here

ν_{i} | {R > j}

denotes the measure given by

ν_{i} | {R > j} (A) = ν_{i} (A \cap {R > j})

, and

f_{*}^{j}

denotes the push-forward of the measure by

f^{j}

. Moreover, it follows from the theorem 4 that almost every point of

M

belongs to the basin of one of the

ν_{i}

. In consequence, almost all point

x \in M

also belongs to the basin of one of the measure

μ_{1}, . . ., μ_{s}

□

Remark 3.1. In the beginning of the Partitioning Algorithm we could have replaced, if necessary,

f

by some iterated

f^{n_{0}}

and so, the measures

μ_{1}, . . ., μ_{s}

may be only

f^{n_{0}}

invariant.

In such case we consider the induced

f

-invariant measures

μ_{1}^{'}, . . ., μ_{s}^{'}

given by

μ_{i}^{'} (A) = μ_{i} (A) + μ_{i} (f^{- 1} A) + . . . + μ_{i} (f^{- (n_{0} - 1)} A)

Proof of Theorem 2

As any non-uniformly expanding map satisfies the hypothesis of theorem 1 we have only to check that every non-flat map and with the slow recurrence satisfying the equation 3 for almost all

x \in M

also satisfies the equation 2 . This result is a consequence of the integrability of

x \mapsto log ∥ (D f (x))^{- 1} ∥

with respect to any absolutely continuous invariant measure ( see Lemma 3.2 below ). Indeed, as

log ∥ (D f)^{- 1} ∥^{- 1} = - log ∥ (D f)^{- 1} ∥

, if

log ∥ (D f)^{- 1} ∥ \in L^{1} (μ_{i})

it follows from Birkhoff 's Ergodic Theorem that we will have for

μ_{i}

almost all

x \in M

liminf \frac{1}{n} \sum_{j = 0}^{n - 1} log ∥ (D f (f^{n} (x)))^{- 1} ∥^{- 1} =

= limsup \frac{1}{n} \sum_{j = 0}^{n - 1} log ∥ (D f (f^{n} (x)))^{- 1} ∥^{- 1} .

As Lebesgue almost every point

x \in M

belongs to the basin of some

μ_{i}

we will conclude that the equality above between

liminf

and

limsup

holds to Lebesgue almost all

x \in M

□

Lemma 3.2. If

f : M \to M

is a

C^{2}

endomorphism then

log ∥ (D f)^{- 1} ∥

is integrable with respect to any absolutely continuous invariant measure.

Proof. In [10] (see remark 1.2 of [10] ) it was proved that $- \infty < \int log | det D f | d μ < + \infty$ with respect to any absolutely continuous invariant measure $μ$ . Let's show how this result implies the integrability of $log ∥ (D f)^{- 1} ∥$ . It is easy to see that $| λ | \leq ∥ A ∥$ for any eigenvalue $λ$ of the matrix $A \in R^{m \times m}$ .
Thus, $| det A | \leq ∥ A ∥^{m}$ and in consequence, $- \infty < \frac{1}{d i m (M)} \int log | det D f | d μ \leq \int log ∥ D f ∥ d μ < + \infty .$ We know that $\bar{A} A = det (A) I$ $\forall A \in R^{n \times n}$ , where $\bar{A}$ in the adjoint matrix and $I$ is the identity. So, $log | det D f | - log ∥ D f ∥ \leq log ∥ \bar{D f} ∥$ and in consequence $- \infty < \int log ∥ \bar{D f} ∥ d μ$ . Moreover, as $∥ \bar{D f} ∥$ is bounded, we get $- \infty < \int log ∥ \bar{D f} ∥ d μ < + \infty .$ Finally, from $(D f)^{- 1} = \frac{1}{det D f} \bar{D f}$ we get $- \infty < \int log ∥ (D f)^{- 1} ∥ d μ = \int log ∥ \bar{D f} ∥ d μ - \int log | det D f | d μ < + \infty .$ □

Proof of Theorem 3

Suppose that there exists

0 < σ < 1

and

δ > 0

such that Lebesgue almost every point

x \in M

has a

(σ, δ)

-hyperbolic time. So, we have

L e b (U) = 0

, where

U = M \ \cup_{j} H_{j} (σ, δ)

. Given a Lebesgue generic point

x \in M

we will have

f^{j} (x) / \in U

\forall j

. Let

n_{1}

be such that

x \in H_{n_{1}} (σ, δ)

and

m_{1}

such that

f^{n_{1}} (x) \in H_{m_{1}} (σ, δ)

. By the properties of hyperbolic times we get

x \in H_{n_{2}} (σ, δ)

where

n_{2} = n_{1} + m_{1} > n_{1}

Therefore, by induction, one can conclude that Lebesgue almost all point of

M

will have infinitely many

(σ, δ)

-hypebolic times. Remembering that

x \in H_{j} (σ, δ)

implies

\prod_{k = 0}^{j - 1} ∥ (D f \circ f^{k} (x))^{- 1} ∥ \leq σ^{j}

, we will have, for generic

x \in M

\frac{1}{j} \sum_{k = 0}^{j - 1} log ∥ (D f (f^{k} (x)))^{- 1} ∥ \leq log σ < 0

for infinitely many integers

j > 0

. In consequence,

liminf \frac{1}{j} \sum_{k = 0}^{j - 1} log ∥ (D f (f^{k} (x)))^{- 1} ∥ \leq log σ < 0

for Lebesgue almost all

x \in M

. As the critical set is empty, the map

f

satisfies the hypothesis of the theorem 1 . So, by the theorem 2 ,

f

is a non-uniformly expanding map.

□

References

J. Aaronson, An introduction to infinite ergodic theory. Math. Surv. Monographs 50, AMS, Providence R.I. US (1997).
J. F. Alves, SRB measures for non-hyperbolic systems with multidimensional expansion, Ann. Scient. Éc. Norm. Sup., $4^{e}$ série, 33 (2000), 1-32.
J. F. Alves, V. Araújo, Random perturbations of nonuniformly expanding maps, Astérisque 286 (2003), 25-62.
J. F. Alves, C. Bonatti, M. Viana, SRB measures for partially hyperbolic systems whose central direction is mostly expanding, Invent. Math. 140 (2000), 351-398.
J. F. Alves, S. Luzzatto, V. Pinheiro, Markov structures and decay of correlations for non-uniformly expanding dynamical systems, Annales Inst. Henri Poincaré (to appear), arXiv:math.DS/0205191.
J. F. Alves, M. Viana, Statistical stability for robust classes of maps with non-uniform expansion, Ergod. Th. & Dynam. Sys. 22 (2002), 1-32.
M. Benedicks, L.-S. Young, Markov extensions and decay of correlations for certain Hénon maps, Astérisque 261 (2000), 13-56.
S. Gouezel, Decay of correlations for nonuniformly expanding systems, arXiv:math.DS/0401184 v1 15 Jan 2004.
G. Keller, Exponents, attractors and Hopf decompositions for interval maps. Ergod. Th. & Dynam. Sys. (1990), 10, 717-744.
P.-D. Liu, Pesin's entropy formula for endomorphisms. Nagoya Math. J. Vol. 150 (1998), 197-209.
Melo, W. C., Strien, S. V. One Dimensional Dynamics, Springer-Verlag, 1993.
M. Viana, Multidimensional non-hyperbolic attractors, Publ. Math. IHES 85 (1997), 63-96.
L.-S. Young, Statistical properties of dynamical systems with some hyperbolicity, Ann. Math. 147 (1998), 585-650.
L.-S. Young, Recurrence times and rates of mixing, Israel J. Math. 110 (1999), 153-188.

Departamento de Matematica, Universidade Federal da Bahia, Av. Ademar de Barros s/n, 40170-110 Salvador, Brazil. E-mail address : viltonj@ufba.br

November 27, 2006

Vilton Pinheiro