Periodic Orbit Theory of Anomalous Diffusion
Extracted source text ยท arXiv chao-dyn/9302002
\documentstyle[openbib,12pt,fleqn]{article}
\oddsidemargin .4cm
\topmargin 0cm
\headsep .5cm
\textheight 21.5cm
\textwidth 14.9cm
\topskip .6cm
\mathindent 1cm
\renewcommand{\baselinestretch}{1.2}
\renewcommand{\arraystretch}{0.9}
% used to label figures in itemize environment:
\newcommand{\fg}[1]{\item {\label{#1}}}
\newcommand{\rb}[1]{\raisebox{1ex}{#1}}
% used to label references:
\newcommand{\rf}[1]{\raisebox{1ex}{\cite{#1}}}
\newcommand{\barr}{\begin{array}}
\newcommand{\bea}{\begin{eqnarray}}
\newcommand{\beq}{\begin{equation}}
\newcommand{\ear}{\end{array}}
\newcommand{\eea}{\end{eqnarray}}
\newcommand{\ceq}{\nonumber \\ & & }
\newcommand{\eeq}{\end{equation}}
\newcommand{\nnu}{\nonumber}
\newcommand{\pde}{\partial}
% from Gunnar Russberg, 8/2-89:
\newcommand{\continue}{\nonumber \\ }
\newcommand{\MinusA}[1]{- \nonumber \\ & & \mbox{\hspace{#1}} -}
\newcommand{\PlusA}[1]{+ \nonumber \\ & & \mbox{\hspace{#1}} +}
%\newcommand{\Div}[2]{ \frac{#1}{#2} }
%\newcommand{\Sqrt}[1]{\left[#1\right]^{1/2}}
\newcommand{\spao}[1]{\mbox{\hspace{#1}}}
\newcommand{\spav}[1]{\parbox{1mm}{\vspace*{#1}}}
\newcommand{\cmet}{\mbox{\hspace{1cm}}}
\newcommand{\picc}[1]{\mbox{\footnotesize $\dsty #1$}}
\newcommand{\dsty}{\displaystyle}
\newcommand{\ssty}{\scriptstyle}
\newcommand{\sssty}{\scriptscriptstyle}
\newcommand{\tsty}{\textstyle}
\newsavebox{\ipiu}
\newsavebox{\imen}
\renewcommand{\a}{\alpha}
\renewcommand{\b}{\beta}
\newcommand{\g}{\gamma}
\newcommand{\G}{\Gamma}
\renewcommand{\d}{\delta}
\newcommand{\D}{\Delta}
\newcommand{\ep}{\varepsilon}
\renewcommand{\L}{\Lambda}
\newcommand{\th}{\theta}
\newcommand{\vt}{\vartheta}
\newcommand{\Th}{\Theta}
\newcommand{\f}{\varphi}
\newcommand{\F}{\Phi}
\newcommand{\somm}[3]{\sum_{\stackrel{\scriptstyle #2}{#3}}^{#1}}
\newcommand{\prdd}[3]{\prod_{\stackrel{\scriptstyle #2}{#3}}^{#1}}
\sbox{\ipiu}{$\ssty i \sssty +1$}
\sbox{\imen}{$\ssty i \sssty -1$}
\begin{document}
\bibliographystyle{unsrt}
\begin{titlepage}
\centering \spav{1cm}\\
{\LARGE\bf Periodic Orbit Theory of\\ }
{\LARGE\bf Anomalous Diffusion\\ }
\spav{2cm}\\
{\large Roberto Artuso\\}
{\normalsize\em Dipartimento di Fisica dell'Universit\'a and I.N.F.N.\\
Via Celoria 16, I--20133 Milano, Italy\\}
{\large Giulio Casati\\}
{\normalsize \em Dipartimento di Fisica dell'Universit\'a and I.N.F.N.\\
Via Castelnuovo, I--22100 Como, Italy\\}
{\large and\\}
{\large Roberto Lombardi\\}
{\normalsize \em Dipartimento di Fisica dell'Universit\'a\\
Via Celoria 16, I--20133 Milano, Italy\\}
\spav{1mm}\\
\vfill
{\centerline \small PACS numbers: 05.45.+b,05.60.+w,05.40.+j}
\vfill
{\small\bf Abstract\\}
\spav{2mm}\\
{\small\parbox{13cm}{\spao{4mm}
We introduce a novel technique to find the asymptotic time behaviour
of deterministic systems exhibiting anomalous diffusion. The procedure
is tested for various classes of simple but physically relevant
1--D maps and possible relevance of our findings for
more complicated problems is briefly discussed.
}}
\spav{4mm}\\
\end{titlepage}
\setcounter{footnote}{0}
In the last few years the phenomenon of deterministic diffusion has
been widely investigated: as a matter of fact a major theoretical
challenge in the study of dynamical systems is to understand
thoroughly the generation of typical stochastic properties in purely
deterministic systems. Moreover even the simplest maps for which
deterministic diffusion has been observed are supposed to model
the behaviour of relevant
physical systems (like Josephson junctions in presence of a microwave
field, see e.g. {\cite{J}}).
A recent approach, introduced independently in
{\cite{DA}} and {\cite{DCEG}},
leads to an expression for
the diffusion coefficient in terms of the periodic orbits of the
system, in the form of a cycle
expansion{\rf{AAC1}}.
Cycle expansions have been applied in a number
of different contexts (see for instance {\cite{AAC2,CE}}): they work
remarkably well for low dimensional hyperbolic systems,
provided their topology (symbolic dynamics) is under control.
The problem of controlling the topology of the system is well
illustrated, in the context of diffusion properties via cycle
expansions,
when one deals with the
infinite Lorentz gas with bounded horizon (see {\cite{LGC}}).
If
one relaxes the hypothesis of pure hyperbolicity (absence of marginal
stability),
much more care has to be taken, and generally the
effectiveness of the expansions is regained only if one is able to sum
infinite contributions shadowing the marginal fixed point{\rf{AAC2}}.
We emphasize that this problem is of paramount relevance, as it
naturally arises when dealing with generic hamiltonian systems, in
which elliptic islands of stability and hyperbolic homoclinic webs
coexist.
Here we address explicitly the problem of marginal stability, which is
tightly connected to the appearance of anomalous diffusion (and also
represents a crucial feature in establishing a good theory for
diffusion in generic two--dimensional area preserving
maps{\rf{BSM}}).
We extend the theory developped in {\cite{DA,DCEG}} to study
deviations from normal diffusion, expressing the asymptotic time
behaviour in terms of properties of the periodic orbits of the system.
The method is then applied to a class of 1--D maps for which anomalous
diffusion has been previously observed.
We begin by recalling the essential features of the periodic orbit
theory of normal diffusion{\rf{DA,DCEG}}, by considering the simplest
context in which it may be applied.
We will consider lifts of one--dimensional circle maps
\beq
x_{t+1}=f(x_t)\,\,t\in {\bf N}\quad f(x+n)=n+f(x) \quad f(-x)=-f(x)
\label{tmap}
\eeq
together with corresponding torus maps $x_{t+1}=f(x_t)
|_{mod\,1}=\hat{f}(x_t)$. Normal diffusion means that asymptotically
$\sigma^2 (t)=<(x_t-x_0)^2>\sim 2Dt$, where the average is over
initial conditions (by symmetry $x_0$ may be taken in the unit
interval). The set of all periodic orbits of $\hat{f}$ will be denoted
by $\{p\}$: each orbit will be characterized by its period $n_p$
stability $\Lambda _p$ (product of the derivatives along the cycle) and
integer winding number $\sigma_p$ (such that for each cycle point $x_{i(p)}$
we have $f^{n_p} (x_{i(p)})=x_{i(p)}+\sigma_p$).
We focus our attention on the generating function
$<e^{\beta(x_t-x_0)}>=\Omega_t(\beta)$; it has been shown{\rf{DA,DCEG}} that
its asymptotic behaviour is
dominated by the leading eigenvalue of an appropriate transfer
operator: $\Omega_t(\beta)\sim z(\beta)^{-t}$, where
$z(\beta)$ is the smallest solution of
\[
\zeta_0^{-1}(z(\beta),\beta)\,=\,\prod_{\{p \} }\left( 1-
\frac{z^{n_p}\exp(\sigma_p \beta)}{|\Lambda_p|}\right)\,=\,0
\]
Expansion of the generating function for small $\beta$ thus yields
\[
D=-\frac{1}{2}\left. \frac{d^2}{d\beta^2} z(\beta) \right|_{\beta=0}
\]
Anomalous diffusion is associated to a vanishing or diverging $D$.
In order to generalize the theory we take into account that, up to time $t$,
$\Omega_t(\beta)$ gets contributions from periodic orbits with
$n_p\leq t${\rf{DA,DCEG}} (as it is connected
to the trace of the t--th power of the
transfer operator): we thus may write in general
\beq
\Omega_t(\beta)\sim \left[ z^{[t]}(\beta) \right]^{-t}
\label{conj}
\eeq
where
$z^{[t]}(\beta)$ is such that
$\zeta_{0[t]}^{-1}(z^{[t]}(\beta),\beta)=0$, $\zeta_{0[t]}^{-1}$
denoting the cycle expansion of $\zeta_0^{-1}$ truncated to finite order $t$.
From (\ref{conj}) we thus get the asymptotic behaviour
\beq
\sigma^2(t)\,\sim \, -t \left.\frac{\partial^2 /\partial \beta^2\,
\zeta_{0[t]}^{-1} (z,\beta)}{\partial / \partial z \,
\zeta_{0[t]}^{-1} (z,\beta)}\right|_{z=1,\beta=0}
\label{ast}
\eeq
where we took into account probability
conservation{\rf{AAC1}} ($z^{[t]}(0) \to 1$ as $t \to \infty$) as well as
symmetry of the map.
This procedure may be alternatively interpreted as follows: for a
sequence of large increasing times $t_k$ we approximate the system by
a sequence of hyperbolic (regularly diffusing) systems $S_k$ with
polynomial zeta functions
$\zeta_0^{-1}(z,\beta)_{S_k}=\zeta^{-1}_{0[t_k]}(z,\beta)$:
$\sigma^2(t)\sim 2D(t_k)\cdot t$ for $t\stackrel{<}{\sim} t_k$ (for both the
original system and $S_k$), thus corrections to normal diffusion are
connected with the asymptotic behaviour of $D(t)$.
To test the theory we consider a map of the form (\ref{tmap}) (see
fig. 1) characterized by the presence of a marginal
fixed point{\rf{GT,GNZ}}.
The corresponding map on the torus $\hat{f}$ is shown in fig. 2, and
consists of five complete branches.
The domain is accordingly partitioned into five subsets, which
we will denote respectively by $1,2,0,3,4$. Branches $1,2,3,4$ have a
constant absolute value of the slope $\Lambda$, while in region $0$ the
map takes the usual Manneville and Pomeau form{\rf{MP}}
$x_{n+1}=x_n+a\cdot x_n^{\gamma}$ ($\gamma >1$),
leading to intermittency.
This partition induces a good symbolic dynamics:
every possible combination of symbols is physically realized as a
trajectory of the map.
To apply the formalism we have to
accomplish two things: enumerate all possible cycles once the marginal
fixed point is pruned away (the effect of the marginal
fixed point is nonlinear and is probed by the infinity of cycles which
accumulate to it{\rf{AAC2}}), and specify $\sigma_p$ and
$|\Lambda_p|$ for each cycle. It is easy to see that the symbolic
dynamics corresponding to prohibiting an infinite repetition of $0$
is determined by
unrestricted grammar in the following (countable) alphabet $\{
0^k1,0^l2,0^n3,0^i4,1,2,3,4\}$ $(k,l,n,i=1,2,3\dots)$. As regards
winding numbers, if we denote by $\epsilon_j$ a generic letter of the
alphabet we have $\sigma_{\epsilon_1 \epsilon_2 \dots
\epsilon_n}=\sum_{i=1}^n\,\sigma_{\epsilon_i}$ where for each letter
including
symbols $1$ or $2$ $\sigma_{\epsilon}=1$, while for each letter
including
symbols $3$ or $4$ $\sigma_{\epsilon}=-1$. We now use the piecewise
linear approximation of Gaspard and Wang{\rf{INT}} to the Manneville
Pomeau system to estimate the stability of cycles shadowing the
marginal fixed point, so that
$|\Lambda_{\epsilon_1 \epsilon_2 \dots \epsilon_n}|=\prod_{j=1}^n
|\Lambda_{\epsilon_j}|$, and for the various letters we have
\beq \begin{array}{r@{\,=\,}l@{\;\;\;\;\;}l}
|\Lambda_{\epsilon}| & \Lambda &\epsilon=1,2,3,4 \\
|\Lambda_{\epsilon}| & \Lambda\cdot k^{\alpha+1}
&\epsilon=0^kL\,\,L=1,2,3,4
\end{array}
\label{W}
\eeq
with $\alpha=1/(\gamma-1)$.
As each curvature{\rf{PPRL}} is zero we get that the zeta function is
expressed in terms of the fundamental cycles{\rf{AAC1}} (which are determined by
the letters) and takes the following form
\[
\zeta_0^{-1}(z,\beta)=1-\frac{4 \cosh \beta}{\Lambda}
z \left(1+ \sum_{k=1}^{\infty} \frac{z^k}{k^{\alpha+1}}\right)
\]
In this way (\ref{ast}) reads
\[
\sigma^2(t)\sim t\frac{\sum_{k=1}^{t} k^{-(\alpha+1)}}{\sum_{k=1}^{t}
k^{-\alpha}}
\]
This leads to normal diffusivity for $\alpha > 1$, while for other
parameter values we get
\beq
\sigma^2(t) \sim
\left\{
\begin{array}{l@{\quad for \;\;}l}
t^{\alpha} & \alpha \in(0,1) \\
t/\ln t & \alpha =1
\end{array}\right.
\label{astast}
\eeq
in agreement with probabilistic estimates and numerical simulations
in {\cite{GT}}. We remark
that our findings are entirely based on metric and
topological properties of the system, thus the relation, expressed by
(\ref{astast}), between anomalous diffusion and exponents of power--law
stabilities does not rely on any
stochastic modelization of the system.
Our theory may also be applied when the diffusion is
accelerated{\rf{GNZ}}, that is when marginal fixed points appear in
the running branches (see fig. 3).
We will give a detailed treatment of the results elsewhere: here we
just quote the results (in agreement with {\cite{GNZ}}, except the
case $\alpha=1$)
\beq
\sigma^2(t)\,\sim \left\{
\begin{array}{l@{\quad for \;\;}l}
t & \alpha > 2 \\
t \ln t & \alpha =2 \\
t^{3-\alpha} & \alpha \in (1,2) \\
t^2/\ln t & \alpha =1 \\
t^2 & \alpha \in (0,1)
\end{array}\right.
\label{mm}
\eeq
where $\alpha$ is still $1/(\gamma-1)$.
We observe that topological intricacies of the system in its laminar
phase do not affect the asymptotic estimates (\ref{astast}) and
(\ref{mm}), which are exclusively determined by the sequence of cycles
accumulating to the marginal fixed points. It is precisely this
robustness with respect to fine details of the dynamics which let us
claim that our approach may be relevant also in much more complex
systems, like the Lorentz gas with infinite horizon. This may
be appreciated by a
simple argument: the weight of orbits travelling a time $t$ without
collisions in this system goes like $t^{-3}${\rf{BUNI}}, corresponding
to $\alpha=2$ in (\ref{W}). From (\ref{mm}) we thus get
$\sigma^2(t)\sim t \ln t$, which indeed seems to reproduce the correct
behaviour{\rf{UH}}.
In this paper we have proposed an extension of the periodic orbit theory of
deterministic diffusion{\rf{DA,DCEG}}, capable of predicting
the asymptotic time
behaviour of $\sigma^2(t)$ by using an appropriate
cycle expansion. In particular this technique has been tested with
classes of one--dimensional maps in which intermittency slows or
accelerates diffusion: the anomalous diffusion is completely
characterized by the exponent $\alpha$ ruling power--law stability of cycles
shadowing the marginal fixed points ($\alpha$ is in turn connected to
the
intermittency exponent $\gamma$).
Our results provide examples in which cycle expansions can
successfully deal with marginal stability: this supports the view that
this technique may represent a major tool in the analysis of generic
low--dimensional (both classical and quantum) hamiltonian systems.
{\bf Acknowledgements:}
{We warmly thank Predrag Cvitanovi\'c and Arkady Pikovsky for useful
comments.}
\vfill\eject
\renewcommand{\baselinestretch} {1}
\begin{thebibliography}{99}
{\small
\bibitem{J} B.A. Huberman, J.P. Crutchfield and N.H. Packard,
{\em Appl.Phys.Lett.}
{\bf 37} (1980) 750; M. Cirillo and N.F. Pedersen, {\em Phys.Lett.}
{\bf 90A} (1982) 150; E. Ben-Jacob, J. Goldhirsch, Y. Imry and S.
Fishman, {\em Phys.Rev.Lett.} {\bf 49} (1982) 1599
\bibitem{DA} R. Artuso, {\em Phys.Lett.} {\bf 160A} (1991) 528
\bibitem{DCEG} P. Cvitanovi\'c, J.-P. Eckmann and P. Gaspard, {\sf
Transport Properties of the Lorentz Gas in Terms of Periodic Orbits},
{\em Niels Bohr Institute preprint} (May 1991)
\bibitem{AAC1} R. Artuso, E. Aurell and P. Cvitanovi\'c, {\em
Nonlinearity} {\bf 3} (1990) 325
\bibitem{AAC2} R. Artuso, E. Aurell and P. Cvitanovi\'c, {\em
Nonlinearity} {\bf 3} (1990) 361
\bibitem{CE} P. Cvitanovi\'c and B. Eckhardt, {\em Phys.Rev.Lett.}
{\bf 63} (1989) 823; R. Mainieri, {\em Phys.Rev.} {\bf A45} (1992)
3580; G. Tanner, P. Scherer, E.B. Bogomolny, B. Eckhardt and D.
Wintgen, {\em Phys.Rev.Lett.} {\bf 67} (1991) 2410; F. Christiansen,
G. Paladin and H.H. Rugh, {\em Phys.Rev.Lett.} {\bf 65} (1990) 361
\bibitem{LGC} P. Cvitanovi\'c, P. Gaspard and T. Schreiber, {\em
Chaos} {\bf 2} (1992) 85
\bibitem{BSM} B. Eckhardt, {\sf Periodic Orbits and Diffusion in
Standard Maps}, {\em Marburg preprint} (July 1992)
\bibitem{GT} T. Geisel and S. Thomae, {\em Phys.Rev.Lett.} {\bf 52}
(1984) 1936
\bibitem{GNZ} T. Geisel, J. Nierwetberg and A. Zacherl, {\em
Phys.Rev.Lett.} {\bf 54} (1985) 616
\bibitem{MP} Y. Pomeau and P. Manneville, {\em Commun.Math.Phys.} {\bf
74} (1980) 189; P. Manneville, {\em J.Phys.} (Paris) {\bf 41} (1980)
1235
\bibitem{INT} P. Gaspard and X.-J. Wang, {\em Proc.Natl.Acad.Sci.
U.S.A.} {\bf 85} (1988) 4591; X.-J. Wang, {\em Phys.Rev.} {\bf A40}
(1989) 6647
\bibitem{PPRL} P. Cvitanovi\'c, {\em Phys.Rev.Lett.} {\bf 61} (1988)
2729
\bibitem{BUNI} L.A. Bunimovich, {\em Sov.Phys. JETP} {\bf 62} (1985)
842
\bibitem{UH} B. Friedman and R.F. Martin, Jr., {\em Phys.Lett.} {\bf
105A} (1984) 23; A. Zacherl, T. Geisel, J. Nierwetberg and G. Radons,
{\em Phys.Lett.} {\bf 114A} (1986) 317;
P.M. Bleher, {\em J.Stat.Phys.} {\bf 66} (1992) 315
}
\end{thebibliography}
\vfill\eject
{\bf {Figure captions}}
\vskip 20pt
\begin{enumerate}
\fg{f1}
Example of a circle map with a marginal fixed point.
\fg{f2}
Map on the torus associated to the one shown in fig. 1.
\fg{f3}
$\hat{f}$ for a model of accelerated diffusion: marginal fixed points
are situated in the running branches ($|\sigma|=1$).
\end{enumerate}
\end{document}