EMIS/ELibM Electronic Journals

Outdated Archival Version

These pages are not updated anymore. They reflect the state of 20 August 2005. For the current production of this journal, please refer to http://www.math.psu.edu/era/.

%_ **************************************************************************
%_ * The TeX source for AMS journal articles is the publishers TeX code     *
%_ * which may contain special commands defined for the AMS production      *
%_ * environment.  Therefore, it may not be possible to process these files *
%_ * through TeX without errors.  To display a typeset version of a journal *
%_ * article easily, we suggest that you retrieve the article in DVI,       *
%_ * PostScript, or PDF format.                                             *
%_ **************************************************************************
% Author Package file for use with AMS-LaTeX 1.2
\dateposted{September 11, 2000}
\PII{S 1079-6762(00)00082-2}



\newtheorem*{casex}{Case when $T$ is not invertible}
\newtheorem*{casexx}{Case when $T$ is invertible}


\newcommand{\R}{\mathbb R}
\newcommand{\N}{\mathbb N}

\newcommand{\Ze}{\mathbb Z}

\newcommand{\Z}{\mathcal Z}
\newcommand{\W}{\mathcal W}
\newcommand{\V}{\mathcal V}

\newcommand{\B}{\mathcal B}

\newcommand{\myS}{\mathcal S}

\newcommand{\tauT}{\tau_T}        %return time for a cylinder in the base


\newcommand{\myu}{\textup u}
\newcommand{\myv}{\textup v}




\newcommand{\tauphi}{\tau_\Phi}   %return time for a ball in the 
%suspension space


\newcommand{\liminfn}{\liminf_{n\rightarrow \infty}} 


\newcommand{\eqdef}{\stackrel{\scriptscriptstyle\rm def}{=}}

%\newcommand{\tauphi}{\tau_\Phi}   %return time for a ball in the 
%suspension space


\title{Local dimensions for Poincar\'e recurrences}

\author{Valentin Afraimovich}
\address{IICO-UASLP, A. Obregon 64, San Luis Potosi SLP,
78210 Mexico}

\author{Jean-Ren\'e Chazottes}
\address{IICO-UASLP, A. Obregon 64, San Luis Potosi SLP,
78210 Mexico}

\author{Beno\^{\i}t Saussol}
\address{Departamento de Matem\'atica, Instituto Superior T\'ecnico, 
1049-001 Lisboa, Portugal}

\subjclass[2000]{Primary 37C45, 37B20}

\date{March 31, 2000}
\revdate{May 8, 2000}

\commby{Svetlana Katok}

\copyrightinfo{2000}{American Mathematical Society}

\begin{abstract} Pointwise dimensions and spectra for measures
associated with Poincar\'e recurrences are calculated for
arbitrary weakly specified subshifts with positive entropy
and for the corresponding special flows.  It is proved that the
Poincar\'e recurrence for a ``typical'' cylinder is asymptotically its
length. Examples are provided which show that this is not true for
some systems with zero entropy.  Precise formulas for dimensions of
measures associated with Poincar\'e recurrences are derived, which
are comparable to Young's formula for the Hausdorff dimension of
measures and Abramov's formula for the entropy of special flows.



Poincar\'e recurrences are main indicators and characteristics of the 
of behavior of dynamical systems in time. A traditional approach is to 
statistical properties of the quantity $\tau_{U}(x)$, the first return 
of the orbit through $x$ into a set $U$. 
We adopt another point of view: instead of looking at the mean return 
or at the return time of points, we are going to study $\tau(U)$, the 
smallest possible return
time into $U$, that is we define the {\it Poincar\'e recurrence for a 
set}, as the
infimum over all return times of the points inside the set \cite{A}.
Poincar\'e recurrences for a set $U$ can be very different from return 
times $\tauU(x)$. 
If $U=\xi^{n}(x)$ is a cylinder of length $n$, then
the return time $\tau_{\xi^{n}(x)}(x)$ of a $\mu$-generic point behaves 
$\exp(n\textup{h}_{\mu}(T,\xi))$ \cite{OW} (where $T$ is the map generating
the dynamical system, and $\textup{h}_{\mu}(T,\xi)$ is
the entropy of $\mu$, with respect to $T$ and $\xi$), whereas
the Poincar\'e recurrence for $\xi^{n}(x)$ is typically of order $n$,
provided that $\mu$ is ergodic, $T$ is weakly specified and 
(Theorem \ref{limtau/n} below). Let us emphasize that this result does 
not depend on
a particular choice of a map $T$, a partition $\xi$ and an ergodic 
measure $\mu$.
Since we deal with a function of sets (namely $U\mapsto\tau(U)$), 
it is natural to use
ideas and methods of dimension theory \cite{P}. We define and calculate
pointwise dimensions for Poincar\'e recurrences (Theorems 
\ref{form-dim-loc} and
\ref{form-dim-loc-Phi}) to
obtain spectra for 
measures (Theorems \ref{spectrum} and \ref{spectrum-Phi}). These 
quantities reflect the balance
between times needed for the return to the ball $B(x,\eps)$ (of radius 
$\eps$ centered
at $x$) and $\eps$ for almost
every point $x$ with respect to an arbitrary ergodic measure, provided 
that $\eps$ is small
This provides a new insight into the
nature of recurrences.
We remark that the positivity of the entropy is an unavoidable
assumption (Theorem~\ref{taun-rotations}). 

This work is part of the manuscript \cite{ACS}.

\section*{Setup for maps}

In this work we shall deal with dynamical systems $(X,T)$ which are 
weakly specified subshifts.
This means that there exists a finite set $\Sigma$, called the 
alphabet, and $X$ is a closed 
subset of $\Sigma^\N$ (non-invertible case) or $\Sigma^\Ze$ (invertible 
case) which is 
invariant under the shift map $T$ defined by $(Tx)_i=x_{i+1}$. 
Given $a\in \Sigma$ let $[a]=\{x\in X: x_0=a\}$, and let 
$\xi=\{[a]:a\in\Sigma\}$ denote the
partition into 1-cylinders. We endow $X$ with the product topology, 
which makes $X$ a compact
metrizable space.
Our results concern only measures with positive entropy, so we will 
assume that $(X,T)$ has
positive topological 
entropy. We also assume that $(X,T)$ is weakly specified; see \cite{K}. 
We now define a
metric on $X$ equivalent to the product topology (see 

Denote by $\xi^{n}$ the dynamical partition, that is: 
$\xi^{0}\eqdef\{X,\emptyset\}$. Then $\xi^{n}(x)$ will
be the atom of the refined partition $\xi^{n}$ that contains $x$ and 
will be referred to as
the $n$-cylinder about $x$. Given a continuous function 
$\myu:X\to(0,\infty)$ we endow $X$ with
the metric $\dX$ defined by
\dX(x,y)\eqdef \e^{-\myu(\xi^{n}(x))}
whenever $y\in\xi^n(x)$ and $y\not\in\xi^{n+1}(x)$, where 
\myu(\xi^n(x))=\sup_{k\le n} \sup_{z\in\xi^k(x)} \big(\myu(z)+\myu(Tz)+\cdots+
\myu(T^{k-1}z)\big),\ n=1,2,\dots\,.
Remark that the standard metric is recovered when one chooses $\myu\equiv 
1$. If one
chooses $\myu(x)=-\log\lambda(x_0)$, which is a constant on every atom of 
\dX(x,y)= \prod_{\ell=0}^{n-1}\lambda(x_\ell),
\textup{diam}\,\xi^{n}(x) = \prod_{\ell=0}^{n-1}\lambda(x_\ell),
i.e. we have a situation similar to that encountered in Moran-like 
geometric construction.
More generally, if one chooses a H\"older continuous function $\myu$, 
then one gets the
distance used to generate Cantor-like sets in $\R^{d}$ \cite{P,AUUS} 
modeled by subshifts.
Denote by $\xi_m^n$ the dynamical
partition, that is:
$\xi_{m}^{n}\eqdef T^{m}\xi\vee T^{m-1}\xi\vee\cdots\vee T^{-n+1}\xi$,
$\xi_{0}^{0}\eqdef\{X,\emptyset\}$, where  $m\geq 0$, $n\geq 0$.
Then $\xi_{m}^{n}(x)$ will
be the atom of the refined partition $\xi_{m}^{n}$ that contains $x$ 
and will be referred to as
the $(m,n)$-cylinder about $x$. 
Given two continuous functions $\myu,\myv:X\to(0,\infty)$ such that
$\myu(x)=\myu(y)$ whenever $\xi_0^n(x)=\xi_0^n(y)$ for every $n\geq 0$ and
$\myv(x)=\myv(y)$ whenever $\xi_m^0(x)=\xi_m^0(y)$ for every $m\geq 0$, we 
endow $X$ with the metric
$\dX$ defined by~\eqref{defmetric} below.
For an arbitrary pair $x,y\in X$, there is a unique pair $(m,n)$ such 
that $y\in\xi_{m}^{n}(x)$
and $y\notin(\xi_{m+1}^{n}(x)\cup\xi_{m}^{n+1}(x))$.
\myu(\xi_{0}^{n}(x))=\sup_{k\le n}\sup_{z\in\xi_{0}^{k}(x)} \big(\myu(z)+
\myu(Tz)+\cdots+\myu(T^{k-1}z)\big),\ n=1,2,\dots,
\myv(\xi_{m}^{0}(x))=\sup_{k\le m}\sup_{z\in\xi_{k}^{0}(x)} \big(\myv(z)+
\myv(T^{-1}z)+\cdots+\myv(T^{-k+1}z)\big),\ m=1,2, \dots\,.
If one chooses $\myu(x)=-\log\lambda(x_0),\;\myv(x)=-\log\gamma(x_{-1})$, 
which are
constants on every atom of $\xi_0^1$ and $\xi_1^0$, respectively,  then
\textup{diam}\,\xi_{m}^{n}(x) = 
\max\Big\{ \prod_{\ell=1}^{m}\gamma(x_{-\ell}),\;\prod_{\ell=0}^{n-1}%

Such a situation occurs, for example, in the case of a piecewise linear 
Smale horseshoe.
In the general case of basic axiom A sets on surfaces, there exist
an associated subshift of finite type $(X,T)$ and some functions 
$\myu,\myv$ satisfying the above
(i.e. depending only on forward, respectively backward, itineraries) 
giving rise to a metric
$\dX$ which is ``adapted'' to the initial system.

Given $x\in X$ and $\eps\ge 0$ we denote by $B(x,\eps)$ the open ball 
of radius $\eps$
centered at $x$. The proof of the following lemma will be omitted.
$(X,\dX)$ is an ultra-metric space, and for any $x\in X$ and $\eps>0$ 
we have
$B(x,\eps) = \xi^{n_{x,\eps}}(x)$, where we set
$n_{x,\eps}=\min\{n\in\N:\e^{-\myu(\xi^n(x))}<\eps\}$ in the 
non-invertible case.
$B(x,\eps) = \xi_{m_{x,\eps}}^{n_{x,\eps}}(x)$, where we set
$n_{x,\eps}=\min\{n\in\N:\e^{-\myu(\xi_0^n(x))}<\eps\}$ and
in the invertible case.
The topology generated by $d$ is equivalent to the product topology.
\section*{Poincar\'e recurrences of sets and local dimensions}
For any set $U\subset X$, one can define the first return time of a 
point $x\in U$ into $U$:
\tauU(x)\eqdef\inf\{k\geq 1:T^{k}x\in U\}\,.
By convention we put this return time to be infinite if the point $x$ 
never comes back
to $U$. 

Let $U$ be a subset of $X$. Then,
\tauT(U)\eqdef\inf\{\tauU(x):x\in U\}\;.

In the following lemma 
we collect various basic properties
of the Poincar\'e recurrence of sets. The proof will be omitted. 
Let $(X,T)$ be a dynamical system and $U\subset X$ any set. Then the 
following properties hold:
\item $\tauT(U)=\inf\{k>0:T^{k}U\cap 
U\neq\emptyset\}=\inf\{k>0:T^{-k}U\cap U\neq\emptyset\}$.
\item $\tauT(U)=\tauT(T^{-1}U)$. If $T$ is invertible, then 
\item Monotonicity: $A\subset B\Rightarrow \tauT(A)\geq\tauT(B)$.
\subsection*{Spectra for Poincar\'e recurrences}

For any $A\subset X$, any $\al\in\R$ and any $q\in\R$, we define
where the infimum is taken over all finite or countable collections 
$(x_i,\eps_i)$ such that $\bigcup_{i} B(x_{i},\eps_i)\supseteq A$. 
The limit $\M^{T}(A,\al,q)\eqdef \lim_{\eps\rightarrow 
exists by monotonicity and we give the
following definition (which was first stated in \cite{A}):
For any non-empty $A\subset X$ and any $q\in\R$,
\al_{T}(A,q)\eqdef \left\{\begin{array}{ll}
\inf \{\al:\M^{T}(A,\al,q)=0\} & \text{if $q\ge 0$,} \\
\sup\{\al:\M^{T}(A,\al,q)=\infty\} & \text{if $q<0$,}
is called the spectrum for Poincar\'e recurrences for the map
$T$ of the set $A$.
It is easy to see that whenever $\al_{T}(A,q)$ is finite one has 
\al_{T}(A,q) = \inf \{\al:\M^{T}(A,\al,q)=0\}=
Note that $\al_{T}(A,0)$ is nothing but the Hausdorff dimension of the 
set $A$.
Since spectra for
Poincar\'e recurrences are just Carath\'eodory dimensions for 
Carath\'eodory structures, we may define, following \cite{P}, the 
following quantity.

Let $\alpha_{T}(\cdot,q)$ be the spectrum defined above. Let 
$\mu$ be a Borel probability measure on $X$. Then 
\al^{\mu}_{T}(q)\eqdef \inf\{\al_{T}(Y,q):Y\subset X, \,\mu(Y)=1\}\;.

We will call it the spectrum for the measure $\mu$.

We emphasize that the family of sets used to define 
\eqref{def-stat-sum} (here, balls) is very
important to get non-trivial results for the spectrum of the measure. A 
similar quantity
(for $\alpha=0$) is studied in \cite{psv}, using covers by open sets
in \eqref{def-stat-sum}. Basically it is proven therein that the 
spectrum of an invariant
ergodic measure is always trivial.

\subsection*{Pointwise dimensions for Poincar\'e recurrences}

Following the ideas of \cite{P}, we define the following quantities.
\begin{definition}[Lower and upper pointwise dimensions]
$\!\!$The lower and upper pointwise dimensions of $\mu$ at a point 
$x$ are defined by
\liminfsup_{\eps\to 0}
\frac{\log\mu(B(x,\eps))+q\tauT(B(x,\eps))}{\log \eps}\;.

This definition is not exactly as in \cite{P}. However, by adopting 
such a definition
we may show directly that spectrum for a measure coincides with this 
quantity for almost
every point $x$ (Theorems \ref{form-dim-loc} and \ref{spectrum}).
The relationship between Definition \ref{pointdim} and Pesin's 
definition of pointwise
dimension \cite{P} was studied in \cite{CS} in the case of the 
general Carath\'eodory
\section*{Main results for maps}

To find formulas for pointwise dimensions and to establish their 
with the spectrum for a measure, we need to know Poincar\'e recurrences
for ``typical'' cylinders.

\subsection*{Local rate of Poincar\'e recurrences for cylinders}

Lower and upper local rates of
Poincar\'e recurrences for cylinders are defined respectively for 
and invertible transformations by
\liminfsup_{n+m\rightarrow\infty}\frac{\tauT(\xi_m^n(x))}{m+n} \;.

Weak specification property (see \cite{K}) immediately implies the 
following result (we omit the
If the system  $(X,T)$ is weakly specified, then 
$\overline{\Ret}_{\xi}(x)\leq 1$.

The following result established in the non-invertible case in 
\cite{stv}, by using 
ideas Kolmogorov's complexity, will be crucial in what follows.
We shall give a direct proof based on the Shannon-McMillan-Breiman

Let $(X,\mathfrak{B},\mu)$ be a probability space where $\mu$ is ergodic
with respect to a measurable transformation $T:X\to X$. 
If $\xi$ is a finite or countable measurable partition with strictly 
entropy $h_\mu(T,\xi)$, then the lower rate
of Poincar\'e recurrences for cylinders is almost surely bigger than one:
\underline{\Ret}_{\xi}(x) \ge 1 \quad\text{for $\mu$-a.e. $x\in X$}.

Coming back to our setup and putting together Proposition 
\ref{limsuptau/n} and Theorem \ref{liminftau/n},
we obtain the following theorem:
Let $\mu$ be an ergodic measure of positive entropy on the weakly 
specified subshift 
$(X,T)$, and $\xi$ the finite partition of $X$ defined in the setup. 
\quad\text{for $\mu$-a.e. $x\in X$}.

The following examples show that for systems with zero entropy the 
statement of Theorem~\ref{liminftau/n}
and \emph{a fortiori} the one of Theorem~\ref{limtau/n} may not hold.

\subsection*{Examples in the case of zero entropy systems}
A first example is given by the dyadic adding machine. In this case, it 
is simple
to show that $\Ret\equiv 0$.

We now consider a more interesting situation, namely a rotation 
$f_{\om}:x\mapsto x\!-\!\om\;\textup{mod 1}$
(i.e. $f_{\om}^{-1}x=x+\om\;\textup{mod 1}$), on the
circle $\myS^{1}=\R/\Ze$, where $0<\om<1$ is an irrational number. The 
number $\om$ can
be approximated by rational numbers $p/q$ ($p$ and $q$ are relatively 
prime) in such a way that
for some value $\beta$ and some pair $(p,q)$. Let $\beta(\om)\eqdef 
\sup\beta$, where the supremum
is taken over all $\beta$ for which inequality (\ref{approx-om}) has 
infinitely many solutions
$(p,q)$ with $q>0$. Assume that $\beta(\om)<\infty$, i.e. $\om$ is a 
Diophantine number. Then
for every $\delta\in(0,1)$ the inequality
holds for infinitely many relatively prime pairs $(p_{i},q_{i})$, with 
as $i\rightarrow\infty$. Consider the partition $\xi$ of $\myS^{1}$ made 
up of two intervals
$[0,\omega)$ and $[\omega,1)$.
The rotation is metrically isomorphic to the subshift $\clos(\pi\myS^1)$,
where the coding map $\pi:\myS^1\to \{0,1\}^\N$ is defined in the obvious 
by $\pi(x)_n = 0$ if $f_\om^n(x)\in [0,\om)$ and $\pi(x)_n= 1$ if  
$f_\om^n(x)\in [\om,1)$.

We now state the following theorem:
If $\beta(\om)>3$ then
for almost every $x$ with respect to Lebesgue measure on $\myS^{1}$.
\subsection*{Existence of pointwise dimensions and formulas of spectra}

One can find formulas for pointwise dimensions and spectra for measures.
Under the assumptions of Theorem \ref{limtau/n}, for any $q\in\R$ 
and for $\mu$-a.e. $x\in X$, one has:
\frac{\h_{\mu}(T)-q}{\int \myu\, d\mu}
\quad\text{in the non-invertible case, and}
 (\h_{\mu}(T)-q)\left(\frac{1}{\int \myu\,d\mu}+\frac{1}{\int \myv\, 
\quad\text{in the invertible case}.

{}From this result, we can deduce the expression for the spectrum for 
the measure:
Under the assumptions of Theorem \ref{limtau/n}, for any $q\in\R$ one has
\al^{\mu}_{T}(q)=\frac{\h_{\mu}(T)-q}{\int \myu\,d\mu}
\quad\text{in the non-invertible case, and}
\al^{\mu}_{T}(q)=(\h_{\mu}(T)-q)\left(\frac{1}{\int \myu\,d\mu}+
\frac{1}{\int \myv\, d\mu}\right)
\quad\text{in the invertible case}.

One could see these relations as an analog of Young's formula 
\cite{young} for dimensions for Poincar\'e
recurrences. It is explicit if $q=0$. Note also that the spectrum for 
the set $X$, $\al(X,q)$, 
was obtained for certain subshifts in \cite{AUUS}, where it has been 
shown that it satisfies 
a non-homogeneous Bowen's equation. 

The value of $q$ for which $\alpha_{T}^{\mu}$ vanishes is equal to
$h_{\mu}(T)\eqdef q_{0}^{T}$.
\section*{Setup for special flows}

Let $\varphi$ be a strictly positive continuous function, and
$(X,T)$ a compact metric space with a distance $d_{X}$.

Define the special space and the special flow as follows:
X^{\varphi}\eqdef\{(x,t):x\in X,0\leq t\leq\varphi(x)\}\;,
where we identify the points $(x,\varphi(x))$ and $(Tx,0)$ for each
$x\in X$, and
Assume for a moment that $\varphi(x)\equiv 1$. Let us recall the
definition of the distance on $X^{1}$ \cite{walters}. Consider the subset
$X\times\{t\}$ of $X\times [0,1]$ and let $\rho_{t}$ denote the
metric on $X\times\{t\}$ defined by
$\rho_{t}((x,t),(y,t))\eqdef(1-t)\dX(x,y)+t\dX(Tx,Ty)$, $x,y\in X$. Thus,
$\rho_{0}((x,0),(y,0))=\dX(x,y)$ and $\rho_{1}((x,1),(y,1))=\dX(Tx,Ty)$.
Consider a chain $x=w_{0},w_{1},\dots,w_{n}=y$ between $x$ and $y$
where for each $i$ either $w_{i}$ and $w_{i+1}$ belong to $X\times\{t\}$
for some $t$ ($[w_{i},w_{i+1}]$ is said to be a horizontal segment and
1},t))$) or
$w_{i}$ and $w_{i+1}$ are on the same orbit ($[w_{i},w_{i+1}]$ is said
to be a vertical segment and $\textup{length}([w_{i},w_{i+1}])$ is
the shortest temporal distance between $w_{i}$ and $w_{i+1}$ along the
orbit). The length of the chain is defined to be the sum of the lengths 
its segments. Then $d_{X^{1}}((x,s),(y,t))$ is defined to be the
infimum of the lengths of all finite chains between $(x,s)$ and $(y,t)$.

We now show that $X^{\varphi}$ and $X^{1}$
are homeomorphic provided that $\varphi(x)>0,\;x\in X$. Indeed, set
$h(x,s)\eqdef(x,s\varphi(x))$ for any $0\leq s<1$ and $(x,s)\in X^{1}$ 
$h(x,1)=(x,\varphi(x))$. This map is continuous and one-to-one 
Now, introduce the following distance on $X^{\varphi}$:
d_{X^{\varphi}}((x,s),(y,t))\eqdef d_{X^{1}}(h^{-1}(x,s),h^{-1}(y,t))\;.

Given $(x,s)\in X^{1}$, $\eps>0$, let
\Big\{(y,t)\in X^{1}:
\vert t-s\vert<\frac{\eps}{2}, 
\Phi_{s-t}(y,t)\in\Seg((x,s),\eps),\;\textup{if}\; s\geq t,
i.e. $\tilde{B}((x,s),\eps)$ is the set of pieces of temporal length 
of (semi-)orbits that pass through the ``horizontal ball''
It is clear that if $\tilde{B}((x,s),\eps)\cap 
X\times(\{0\}\cup\{1\})=\emptyset$, then
$\tilde{B}((x,s),\eps)$ is homeomorphic to the direct product of 
and an interval. 

Set $B((x,s),\eps)\eqdef h^{-1}\tilde{B}((x,s),\eps)$. We shall use 
to cover $X^{\varphi}$; see below.

If $\mu$ is a $T$-invariant probability measure on $X$, then we define
a $\Phi$-invariant probability measure $\overline{\mu}\eqdef 
by putting, for any $F$ continuous on $X^{\varphi}$:
That is, $\overline{\mu}$ is the normalization on $X^{\varphi}$ 
obtained by taking
the direct product
of $\mu$ with Lebesgue measure on $\R$. (It can be proved that every 
probability measure on $X^{\varphi}$ can be obtained in this way from a 
probability measure on $X$ by Fubini's theorem.)

We can now define the first return time of the set $B((x,t),\eps)$:
For any $x\in X$, $t>0$ and $\eps>0$, the Poincar\'e recurrence for the 
set $B((x,t),\eps)$ is
\tauphi(B((x,t),\eps))\eqdef \inf\{s>\eps:\Phi_{s}(B((x,t),\eps))\cap 
\subsection*{Spectra for Poincar\'e recurrences}
For each $Z\subset X^{\varphi}$, each $\al\in\R$ and each $q\in\R$, we 
where $B((x_{i},t_{i}),\eps_{i})$ is defined above and
$\bigcup_{i}B((x_{i},t_{i}),\eps_{i})\supseteq Z$. Then the limit
$\M^{\Phi}(Z,\al,q)\eqdef \lim_{\eps\rightarrow 
exists by monotonicity and we introduce the
following definition:
\begin{definition}[Spectrum for flows]
For any $Z\subset X^{\varphi}$ and any $q\geq 0$,
\al_{\Phi}(Z,q)\eqdef \sup\{\al:\M^{\Phi}(Z,\al,q)=\infty\}
is called the spectrum for Poincar\'e recurrence for the flow $\Phi$ of 
the set $Z$.

The quantity $\al_{\Phi}(Z,0)$ coincides with the Hausdorff
dimension of $Z$ for the flow.

Now we proceed to the definition of the spectra for measures. 
\begin{definition}[Spectra for measures]
Let $\alpha_{\Phi}(\cdot,q)$ be the spectra defined above. Let 
$\bmu$ be a Borel probability
measure on $X^{\varphi}$. Then define the following spectrum: 
\al_{\Phi}^{\overline{\mu}}(q)\eqdef \inf\{\al_{\Phi}(V,q):
V\in X^{\varphi},\,\overline{\mu}(V)=1\}.
\subsection*{Pointwise dimensions for Poincar\'e recurrences}

\begin{definition}[Lower and upper $\Phi$-pointwise dimensions]
The lower and upper $\Phi$-pointwise dimensions of $\bmu$ at the point 
$(x,t)$ are
defined by
\underline{\overline{\lim}}_{\eps\rightarrow 0}

\section*{Main results for special flows}

We shall call a $B$-set $B((x,s),\eps)$ a ``good'' one
if $B((x,s),\eps)=h^{-1}\tilde{B}((x,s),\eps)$ and there exists a 
cylinder $\xi_{n}(y)$ such that
$\Seg((x,s),\eps)=\Phi_{s}(\xi_{n}(x))$. In this case, we will denote 
$\xi_{n}(y)$ by
$\eta_{n}(B((x,s),\eps))$. Let us denote by $n_{\eps}$ the index $n$ 
for the cylinder
where $B((x,s),\eps)$ is a good $B$-set. For convenience,
we shall write $\eta_{n_{\eps}}(x)$ instead of 
Then, the asymptotic behavior of the
Poincar\'e recurrence $\tauphi(B((x,t),\eps))$ can be described as 
For $\bmu$-almost every point $(x,s)\in X^{\varphi}$, 
\lim_{\eps\to 0} \frac{\tauphi((B(x,s),\eps))}{n_\eps}=
\int \varphi\, d\mu\,.

By using this proposition, we prove the following results.
Let $\bmu$ be the measure on $X^{\varphi}$ induced by the ergodic measure
$\mu$ on $X$. 
The pointwise dimension exists $\bmu$-a.e. and is equal to
d_{\bmu,q}^{\Phi}(x,t) = 1 +
\frac{1}{\int \myu\,d\mu}\left(\h_{\mu}(T)-q\int\varphi\,d\mu\right)
\;\textup{in the non-invertible case, and}
d_{\bmu,q}^{\Phi}(x,t) = 1 +
\left(\frac{1}{\int \myu\,d\mu}+\frac{1}{\int \myv\,d\mu}\right)
\left(\h_{\mu}(T)-q\int\varphi\,d\mu\right)\;\textup{in the invertible 

Now we can state the following theorem:
The spectrum for the measure $\bmu$ is given for all $q\in\R$ by
\alpha_\Phi^{\bmu}(q)=1 +
\frac{1}{\int \myu\,d\mu}\left(\h_{\mu}(T)-q\int\varphi\,d\mu\right)
\;\textup{in the non-invertible case, and}
\alpha_\Phi^{\bmu}(q) = 1 +
\left(\frac{1}{\int \myu\,d\mu}+\frac{1}{\int \myv\,d\mu}\right)
\left(\h_{\mu}(T)-q\int\varphi\,d\mu\right)\;\textup{in the invertible 

We have
1 + \alpha_{T}^{\mu}(q)\;.

The value of $q$ for which $\al^{\bmu}_{\Phi}$ vanishes is equal to
\frac{\h_{\mu}(T)+\int \myu\,d\mu}{\int\varphi\,d\mu}
\eqdef q_{0}^{\Phi}=\frac{q_{0}^{T} + \int \myu\,d\mu}{\int\varphi\,d\mu}
\;\textup{ in the non-invertible case, and}
\frac{\h_{\mu}(T)+\frac{\int \myu\,d\mu\times\int\myv\,d\mu}{\int(\myu+
{\int\varphi\,d\mu}\eqdef q_{0}^{\Phi}=
\frac{q_{0}^{T}+\frac{\int \myu\,d\mu\times\int\myv\,d\mu}{\int(\myu+
{\int\varphi\,d\mu}\;\textup{ in the invertible case}.
\section*{Proof of Theorem \ref{liminftau/n}}

For the sake of definiteness, we write the proof for the case of 
invertible $T$. 
The case of non-invertible $T$ can be obtained
in a similar way after evident simplifications.

It suffices to prove the theorem for finite partitions; 
the case of countable $\xi$ will follow easily.
More precisely, if $\xi=\{B_1,B_2,\ldots\}$ is countable, then for some 
$m<\infty$ the 
finite partition $\hat\xi = \{B_1,B_2,\dots,B_m, \bigcup_{l>m}B_l\}$ will 
have positive entropy. 
In addition, $\xi$ is finer than $\hat\xi$, hence $\tauT(\xi_m^n(x)) 
\geq \tauT(\hat\xi_m^n(x))$,
and the statement follows.

Assume now that $\xi$ is finite. Observe that $h\eqdef h_\mu(T,\xi)$ 
is non-zero and finite. Fix $\eps\in(0,h/3)$. 
By the Shannon-McMillan-Breiman theorem for $\mu$-a.e. $x$,
there exists $N(x)$ such that if $n+m>N(x)$ then 
\left|\frac{1}{n+m}\log\mu(\xi_m^n(x))+h\right| \leq \eps.
By Egoroff's theorem, if $M=M(\eps)$ is sufficiently large then 
$E_M\eqdef\{x\in X: N(x)1-\eps$.
We can choose $c$ so large that for any $x\in E_{M(\eps)}$ and any 
positive integers $n,m$ 
c^{-1} e^{[-(n+m)h-(n+m)\eps]} \leq \mu(\xi_m^n(x)) \leq c e^{[-(n+m)h+
We now write $E=E_{M(\eps)}$. Let $\delta=1-\frac 3h\eps$ and set
\A_m^n\eqdef\{x\in E: \tauT(\xi_m^n(x))\leq \delta(n+m)\}.
Obviously $\A_m^n=\bigcup_{k=1}^{\delta(n+m)} R_m^n(k)$  where
R_m^n(k) \eqdef\{x\in E: \tauT(\xi_m^n(x))=k\}.
We shall prove that $\sum_{n,m}\mu(\A_m^n)<\infty$.
Let $n,m$ be positive integers and $0\leq k\leq n+m$.
If the return time of the cylinder
$C=[a_{-m}a_{-m+1}\cdots a_{0}\cdots a_{n-1}]\in\xi_{m}^{n}$
is equal to $k$, i.e. $\tauT(C)=k$, then it can be readily
checked that $a_{j+k}=a_{j}$, for all $-m\leq j\leq n-k-1$.
This means that any block made with $k$ consecutive symbols
completely determines the cylinder $C$. In particular, since
there exists $p\geq 0$ such that $p\leq m$ and $0\leq k-p\leq n$,
we can choose the cylinder $Z=\xi_p^{k-p}(x)\supset\xi_m^n(x)$.
\Z=\{ \xi_p^{k-p}(x):x\in R_m^n(k)\}.
Because of the structure of cylinders under consideration, for any 
cylinder $Z\in\Z$
there exists a (unique) cylinder $C_Z\in\xi_m^n$ such that 
$C_Z\subset Z$ and $Z\cap R_m^n(k)\subset C_{Z}$. This implies
\mu(R_m^n(k)) = \sum_{Z\in \Z} \mu( Z\cap R_m^n(k) ) \leq \sum_{Z\in 
\Z} \mu(C_Z).
By definition, for each $Z\in \Z$, there exists $x\in E$ such that
$Z=\xi_p^{k-p}(x)$ and $C_Z=\xi_m^n(x)$. Using \eqref{smb} we get
\mu(\xi_m^n(x)) \leq  c \e^{[-(n+m)h+(n+m)\eps]}
1 \leq c \mu(\xi_p^{k-p}(x)) \e^{[kh+k\eps]}.
Multiplying these inequalities we get 
\mu(C_Z) \leq c^2 \exp[-(n+m)h+(n+m)\eps] \exp[kh+k\eps] \mu(Z).
Summing up on $Z\in \Z$ we get (recall that $k\leq n+m$)
\mu(R_m^n(k)) \leq c^2 \exp[-(n+m-k)h+2(n+m)\eps].
This implies that 
&= \sum_{k=1}^{[\delta(n+m)]} \mu(R_m^n(k)) \\
&\leq c^2 \frac{e^h}{e^h-1} \exp\left[-(n+m)(h-\delta h -2\eps)\right].
Since $h-\delta h-2\eps= h -(1-\frac 3h\eps)h -2\eps=\eps>0$, we get that
\sum_{m\geq 1,n\geq 1} \mu(\A_m^n) < +\infty.
In view of the Borel-Cantelli lemma, we finally get that for $\mu$-almost 
every $x\in E$,
$\tauT(\xi_m^n(x))\geq (1-\frac 3h\eps)(n+m)$ except for finitely many 
pairs  of integers $(n,m)$. 
Since in addition $\mu(E)>1-\eps$, the arbitrariness of $\eps$ implies 
the desired result


V. A. is supported by CONACYT
through its program of ``C\'atedras Patrimoniales II''under contract
485100-2.  J.-R.~C. and B.~S. are grateful to the Instituto de
Investigaci\'on en Comunicaci\'on Optica UASLP for hospitality
during their visit to San Luis Potosi.  B.~S. is supported by FCT's
Funding Program and by the Center for Mathematical Analysis, Geometry,
and Dynamical Systems.


V. Afraimovich,
\emph{Pesin's dimension for Poincar\'e recurrences}, 
Chaos~\textbf{7} (1997), 12--20.
V. Afraimovich, J.-R. Chazottes, and B. Saussol,
\emph{Pointwise dimensions for Poincar\'e recurrences associated with 
maps and special flows},
in preparation.
V. Afraimovich, J. Schmeling,  E. Ugalde, and J. Ur\'{\i}as,
\emph{Spectra of dimensions for Poincar\'e recurrences}, to appear in
Discrete and Continuous Dynamical Systems~(2000).
R. Bowen and P. Walters,
\emph{Expansive one-parameter flows}, J. Diff. Equas. \textbf{12} (1972),
J.-R. Chazottes and B. Saussol,
\emph{Sur les dimensions locales et les dimensions des mesures}, 
preprint (2000).
A. Katok and B. Hasselblatt,
\emph{Introduction to the modern theory of dynamical systems}, 
Encyclopedia of Math. and its Applications~\textbf{54}, 
Cambridge University Press, Cambridge, 1995.
D. S. Ornstein and B. Weiss,
\emph{Entropy and data compression schemes}, IEEE Transactions on 
Information Theory~\textbf{39} (1993), 78--83.
V. Penn\'e, B. Saussol, and S. Vaienti, 
\emph{Dimensions for recurrence times: topological and dynamical 
Discrete and Continuous Dynamical Systems~\textbf{5} (1999), 783--798. 
Ya. B. Pesin,
\emph{Dimension theory in dynamical systems, contemporary views and 
Chicago Lectures in Mathematics, 1997.
B. Saussol, S. Troubetzkoy, and S. Vaienti, In preparation.
L.-S. Young,
\emph{Dimension, entropy and Lyapunov exponents},
Ergod. Th. \& Dyn. Syst \textbf{2} (1982), 109--124.