1: % 16/12/2004 -- PRE-final
2: \documentclass[aps,pre,twocolumn,superscriptaddress,showpacs]{revtex4}
3: %\documentclass[aps,pre,preprint,superscriptaddress,showpacs]{revtex4}
4: \usepackage{amsfonts,amssymb,amsmath,latexsym,epsfig}
5: \baselineskip 25pt
6:
7: \begin{document}
8:
9: [Phys. Rev. E {\bf 71}, 036215 (2005)]
10:
11: \title{Effective dynamics in Hamiltonian systems with mixed phase space}
12:
13: \author{Adilson E. Motter}
14: \email{motter@mpipks-dresden.mpg.de}
15: \affiliation{Max Planck Institute for the Physics of Complex Systems,
16: N\"othnitzer Strasse 38, 01187 Dresden, Germany}
17:
18: \author{Alessandro P. S. de Moura}
19: \affiliation{Instituto de F\'{\i}sica, Universidade de S\~ao Paulo,
20: Caixa Postal 66318, 05315-970, S\~ao Paulo, SP, Brazil}
21:
22: \author{Celso Grebogi}
23: \affiliation{Max Planck Institute for the Physics of Complex Systems,
24: N\"othnitzer Strasse 38, 01187 Dresden, Germany}
25: \affiliation{Instituto de F\'{\i}sica, Universidade de S\~ao Paulo,
26: Caixa Postal 66318, 05315-970, S\~ao Paulo, SP, Brazil}
27:
28: \author{Holger Kantz}
29: \affiliation{Max Planck Institute for the Physics of Complex Systems,
30: N\"othnitzer Strasse 38, 01187 Dresden, Germany}
31:
32: \date{\today}
33:
34: \begin{abstract}
35:
36: An adequate characterization of the dynamics of Hamiltonian systems at
37: physically relevant scales has been largely lacking. Here we investigate this
38: fundamental problem and we show that the finite-scale Hamiltonian dynamics is
39: governed by effective dynamical invariants, which are significantly different
40: from the dynamical invariants that describe the asymptotic Hamiltonian dynamics.
41: The effective invariants depend both on the scale of resolution and the region
42: of the phase space under consideration, and they are naturally interpreted
43: within a framework in which the nonhyperbolic dynamics of the Hamiltonian system
44: is modeled as a chain of hyperbolic systems.
45:
46: \end{abstract}
47:
48: \pacs{05.45.-a, 05.45.Df}
49:
50: \maketitle
51:
52: \section{Introduction}
53: \label{1}
54:
55: A comprehensive understanding of Hamiltonian dynamics is a long outstanding
56: problem in nonlinear and statistical physics, which has important applications
57: in various other areas of physics. Typical Hamiltonian systems are
58: nonhyperbolic as they exhibit mixed phase space with coexisting regular and
59: chaotic regions. Over the past years, a number of ground-breaking works
60: \cite{chirikov1,asymp,meiss1,greene1,pikovsky1,christiansen1,lau1,zasl,uptodate} have
61: increasingly elucidated the asymptotic behavior of such systems and it is now
62: well understood that, because of the stickiness due to Kolmogorov-Arnold-Moser
63: (KAM) tori, the chaotic dynamics of typical Hamiltonian systems is fundamentally
64: different from that of hyperbolic, fully chaotic systems. Here ``asymptotic''
65: means in the limit of large time scales and small length scales. But in
66: realistic situations, the time and length scales are limited. In
67: the case of hyperbolic systems, this is not a constraint because the
68: (statistical) self-similarity of the underlying invariant sets guarantees the
69: fast convergence of the dynamical invariants (entropies, Lyapunov exponents,
70: fractal dimensions, escape rates, etc) and the asymptotic dynamics turns out to
71: be a very good approximation of the dynamics at finite scales. In nonhyperbolic
72: systems, however, the self-similarity is usually lost because the invariant sets
73: are not statistically invariant under magnifications. As a result, the
74: finite-scale behavior of a Hamiltonian system may be fundamentally different
75: from the asymptotic behavior considered previously, which is in turn hard to
76: come by either numerically \cite{uptodate,fluid} or experimentally \cite{nh_h}.
77:
78: The aim of this paper is to study the dynamics of Hamiltonian systems at
79: finite, physically relevant scales. To the best of our knowledge, this problem
80: has not been considered before. Herewith we focus on Hamiltonian chaotic
81: scattering, which is one of the most prevalent manifestations of chaos in open
82: systems, with examples ranging from fluid dynamics \cite{fluid,nh_h} to
83: solid-state physics \cite{sol_stat} to general relativity \cite{relat}. We show
84: that the finite-scale dynamics of a Hamiltonian system is characterized by {\it
85: effective} dynamical invariants (e.g., effective fractal dimension), which: (1)
86: may be significantly different from the corresponding invariants of the
87: asymptotic dynamics; (2) depend on the resolution but can be regarded
88: as constants over many decades in a given region of the phase space; and (3) may
89: change drastically from one region to another of the {\it same} dynamically
90: connected (ergodic) component. These features are associated with the slow and
91: nonuniform convergence of the invariant measure due to the breakdown of
92: self-similarity in nonhyperbolic systems. To illustrate the mechanism behind
93: the properties of the effective invariants, we introduce a simple deterministic
94: model which we build on the observation that a Hamiltonian system can be
95: represented as a chain of hyperbolic systems.
96:
97: The paper is organized as follows. We start, in Sec. \ref{s2}, with the
98: analysis of the invariant measure and the outline of the transport structures
99: underlying its convergence.
100: Our chain model is introduced and analyzed in Sec. \ref{3}. The
101: effective fractal dimension is defined in Sec. \ref{4} and its properties are
102: verified for a specific system in Sec. \ref{5}. Conclusions are presented
103: in the last section.
104:
105: \section{Invariant Measure}
106: \label{s2}
107:
108: For concreteness, consider a two-dimensional area preserving map with a major
109: KAM island surrounded by a chaotic region. One such map captures all the main
110: properties of a wide class of Hamiltonian systems with mixed phase space. When
111: the system is open (scattering), almost all particles initialized in the chaotic
112: region eventually escape to infinity. We first study this case with a diffusive
113: model for the transversal motion close to the main KAM island, obtaining an
114: analytical expression for the probability density $\rho(x,t)$
115: of particles remaining in the scattering region at time $t$ and distance $x$
116: from the island [see APPENDIX]. We find that, in the case of chaotic scattering, a singularity
117: develops and the invariant measure, given by
118: $\lim_{t\rightarrow\infty}\rho(x,t)$, accumulates on the outermost KAM torus of
119: the KAM island [APPENDIX]. Physically, this corresponds to the tendency
120: of nonescaping particles to concentrate around the regular regions.
121: Dynamically, the stickiness due to KAM tori underlies two major features of
122: Hamiltonian chaotic scattering, namely the algebraic decay of the survival
123: probability of particles in the scattering region
124: \cite{asymp,meiss1,greene1,pikovsky1,christiansen1} and the integer
125: dimension of the chaotic saddle \cite{lau1}, and distinguishes this phenomenon
126: from the hyperbolic chaotic scattering characterized by exponential decay and
127: noninteger fractal dimension. However, the convergence of the measure is rather
128: slow and highly nonuniform, as shown in Fig.~\ref{fig1} for typical parameters,
129: which is in sharp contrast with the fast, uniform convergence observed in
130: hyperbolic systems. Our main results are ultimately related to this slow and
131: nonuniform convergence of the invariant measure.
132:
133: Previous works on transport in Hamiltonian systems have used stochastic models,
134: where invariant structures around KAM islands are smoothened out and the
135: dynamics is given entirely in terms of a diffusion equation
136: \cite{chirikov1,greene1} or a set of transition probabilities (Markov chains or
137: trees) \cite{meiss1,other_chains}. The stochastic approach is suitable to
138: describe transport properties (as above), but cannot be used to predict the
139: behavior of dynamical invariants such as Lyapunov exponents and fractal
140: dimensions. Here we adopt a deterministic approach where we use the Cantori
141: surrounding the KAM islands to split the nonhyperbolic dynamics of the
142: Hamiltonian system into a chain of hyperbolic dynamical systems.
143:
144:
145: Cantori are invariant structures that determine the transversal transport close
146: to the KAM islands \cite{asymp,meiss1}. There is a hierarchy of infinitely many
147: Cantori around each island. Let $C_1$ denote the area of the
148: scattering region outside the outermost Cantorus, $C_2$ denote the annular area
149: in between the first and second Cantorus, and so on. As $j$ is increased, $C_j$
150: becomes thinner and approaches the corresponding island. For simplicity, we
151: consider that there is a single island \cite{hier} and that, in each iteration,
152: a particle in $C_j$ may either move to the outer level $C_{j-1}$ or the inner
153: level $C_{j+1}$ or stay in the same level \cite{meiss1}. Let $\Delta_j^{-}$ and
154: $\Delta_j^{+}$ denote the transition probabilities from level $j$ to $j-1$ and
155: $j+1$, respectively. A particle in $C_1$ may also leave the scattering region,
156: and in this case we consider that the particle has escaped. The escaping region
157: is denoted by $C_0$. The chaotic saddle is expected to have points in $C_j$ for
158: all $j\geq 1$. It is natural to assume that the transition probabilities
159: $\Delta_j^{-}$ and $\Delta_j^{+}$ are constant in time. This means that each
160: individual level can be regarded as a hyperbolic scattering system, with its
161: characteristic exponential decay and noninteger chaotic saddle dimension.
162: Therefore, a nonhyperbolic scattering is in many respects similar to a sequence
163: of hyperbolic scatterings.
164:
165:
166:
167: % FIGURE 1:
168: \begin{figure}[pt]
169: \begin{center}
170: \epsfig{figure=fig1.eps,width=6.0cm}
171: \caption{Snapshots of the probability density $\rho$ as a
172: function of $x$, for $\rho(x,0)=\delta(x-x_0)$,
173: $x_0=1$, $x_1=2$, $\alpha=3$, and the outermost torus of
174: the KAM island at $x=0$ [APPENDIX].
175: The time $t$ is indicated in the figure.}
176: \label{fig1}
177: \end{center}
178: \end{figure}
179:
180:
181:
182: \section{Chain Model}
183: \label{3}
184:
185: We now introduce a simple deterministic model that incorporates
186: the above elements and reproduces essential features of the Hamiltonian dynamics.
187: Our model is depicted in Fig.~\ref{fig2} and consists of a semi-infinite
188: chain of 1-dimensional ``$/\backslash/$-shaped'' maps,
189: defined as follows:
190: \begin{equation}
191: M_j(x) = \left \{
192: \begin{array}{lll}
193: &\xi_j x, & 0\leq x<1/\xi_j\\
194: - &\xi_j (x - \Delta_j^{-}) +2, & 1/\xi_j< x - \Delta_j^{-} < 2/\xi_j\\
195: &\xi_j (x-1)+1, & -1/\xi_j < x -1 \leq 0 ,
196: \end{array} \right. \nonumber
197: \end{equation}
198: where $\xi_j> 3$ and $0< \Delta_j^{-} < 1-3/\xi_j\;$
199: ($j=1,2,\ldots$). If $x$ falls in the interval $ 1/\xi_j\leq x \leq
200: 1/\xi_j + \Delta_j^{-}$, where $M_j$ is not defined, the ``particle''
201: is considered to have crossed a Cantorus to the ``outer level''
202: $j-1$. This interval is mapped uniformly to $[0,1]$, and the iteration
203: proceeds through $M_{j-1}$. Symbolically, this is indicated by
204: $j\rightarrow j-1$. Similarly, if $x$ falls into $ 1-1/\xi_j -
205: \Delta_j^{+}\leq x\leq 1-1/\xi_j$, where $\Delta_j^{+} = 1-3/\xi_j -
206: \Delta_j^{-}$, the particle goes to the ``inner level'', and
207: $j\rightarrow j+1$.
208: Particles
209: that reach $ 1/\xi_1\leq x
210: \leq 1/\xi_1 + \Delta_1^{-}$ are considered to have escaped.
211: The domain of $M_j$ is denoted by $I_j$ and is analogous to $C_j$ in a Hamiltonian system,
212: where $\Delta_j^{-}$ and $\Delta_j^{+}$ represent the transition probabilities.
213: The transition rate ratios
214: $\mu = \Delta_j^{+}/\Delta_j^{-}$ and $\nu = \Delta_{j+1}/\Delta_j$ are taken in
215: the interval $0 <\mu < \nu <1$ and are set to be independent of $j$, where
216: $\Delta_{j}= \Delta_j^{+} + \Delta_j^{-}$.
217: The parameter $\mu$ is a measure of the fraction of particles
218: in a level $j$ that will move to the inner level $j+1$ when leaving level $j$, while $\nu$
219: is a measure of how much longer it takes for the particles in the inner level to escape.
220: The nondependence on $j$ corresponds to
221: the approximate scaling of the Cantori suggested by the renormalization theory \cite{meiss1}.
222: Despite the hyperbolicity of each map, the entire chain behaves as a nonhyperbolic system.
223: For a uniform initial distribution in $I_1$,
224: it is not difficult to show \cite{details} that
225: the number of particles remaining in the chain after a long time $t$
226: decays algebraically as $Q(t)\sim t^{-\ln \mu /\ln \nu}$,
227: and that the initial conditions of never escaping particles form a zero Lebesgue
228: measure fractal set
229: with box-counting dimension 1.
230: However, the finite-scale behavior may deviate considerably from these
231: asymptotics, as shown in Fig.~\ref{fig3}.
232:
233:
234: % FIGURE 2:
235: \begin{figure}[bt]
236: \begin{center}
237: \epsfig{figure=fig2.eps,width=5.0cm}
238: \caption{Semi-infinite chain of hyperbolic maps $M_j$, $j=1,2,\ldots$}
239: \label{fig2}
240: \end{center}
241: \end{figure}
242:
243:
244: In Fig.~\ref{fig3}(a) we show the survival probability $Q$ as a function of
245: time. For small $\mu$ and $\nu$, the curve is composed of a discrete sequence
246: of exponentials with scaling exponents $\ln (1-\Delta_j^-)$,
247: which decrease (in absolute value) as we go forward in the sequence.
248: The length of each exponential segment is of the
249: order of $\mu$ in the decay of $Q$ and $-\ln\nu$ in the variation of $\ln t$.
250: This striking behavior is related to the time evolution of the density of
251: particles inside the chain. This is shown in Fig.~\ref{fig3}(b), where we plot
252: the average position $\langle j\rangle$ of an ensemble of particles initialized
253: in $I_1$ (i.e., $j=1$). The transitions between successive exponentials in the
254: decay of $Q$ [Fig.~\ref{fig3}(a)] match the transitions from a level $j$ to the
255: next in the average position of the remaining particles [Fig.~\ref{fig3}(b)].
256: In a Hamiltonian system, the increase of $\langle j\rangle$ in time is related
257: to the development of the singular invariant measure anticipated in our
258: diffusion analysis [see Fig.~\ref{fig1}]. The piecewise exponential behavior
259: of $Q$ is smoothened out for large $\mu$ and $\nu>\mu$ [Figs.~\ref{fig3}(a) and
260: \ref{fig3}(b)].
261:
262: In Fig.~\ref{fig3}(c) we show the fractal dimension of the set
263: of initial conditions of never escaping particles as computed from the
264: uncertainty algorithm \cite{uncert}, which consists in measuring the scaling
265: of the fraction $f(\varepsilon)$ of {\it $\varepsilon$-uncertain} points (initial points
266: whose escaping time is different from the escaping time
267: of points taken $\varepsilon$ apart).
268: The scaling is statistically well defined
269: over decades and the exponent $\alpha=\Delta \ln f(\varepsilon)/\Delta \ln
270: \varepsilon$ can be computed accurately.
271: However, the resulting dimension $1-\alpha$ is not only significantly smaller
272: than 1 but also depends critically
273: on the region $L$ of the phase space where it is computed. The convergence of
274: the dimension is indeed so slow that it can only be noticed when observed over
275: very many decades of resolution, as shown in Fig.~\ref{fig3}(d) where data of
276: Fig.~\ref{fig3}(c) is plotted over 35 decades! Initially smaller, the dimension
277: measured for $L=I_1$ approaches the dimension measured for $L=I_2$ as the scale
278: $\varepsilon$ is reduced beyond $10^{-15}$ (i.e. the corresponding curves in
279: Fig.~\ref{fig3}(d) become parallel). As shown in Fig.~\ref{fig3}(d), this
280: behavior is related to a transition in the average innermost level $\langle
281: j_{max} \rangle$ reached by the particles launched from $\varepsilon$-uncertain
282: points. As $\varepsilon$ is further reduced, new transitions are expected. The
283: dimension measured in between transitions is mainly determined by the dimension
284: $D=\ln 3/\ln \xi_k$, $k=\langle j_{max} \rangle$, of the corresponding element
285: of the chain. For given $j$ and $\varepsilon$,
286: the measured dimension is larger when
287: $L$ is taken in a denser part of the invariant set,
288: such as in the subinterval of $I_1$ first mapped into $I_2$ [Fig.~\ref{fig3}(c); diamonds],
289: because $\langle j_{max} \rangle$ is larger in these regions.
290: In some regions, however, the measured dimension is quite different
291: from the asymptotic value even at scales as small as $\varepsilon =10^{-30}$.
292: This slow convergence of the dimension is due to the slow increase of $\langle j_{max} \rangle$,
293: which in a Hamiltonian system is related to the slow convergence of the invariant
294: measure [Fig.~\ref{fig1}].
295: The convergence is even slower for smaller $\mu$ and larger $\nu$.
296: Incidentally, the experimental measurements of the fractal dimension are usually
297: based on scalings over less than two decades \cite{avnir1}. Therefore, at
298: realistic scales the dynamics is clearly not governed by the asymptotic
299: dynamical invariants.
300:
301: % FIGURE 3:
302: \begin{figure}[pt]
303: \begin{center}
304: \epsfig{figure=fig3.eps,width=8.0cm}
305: \caption{Chain model for $\xi_1=4.1$.
306: (a) Survival probability $Q$ and (b) average position $\langle j\rangle$ as a function of time
307: for $\mu=0.01$ and $\nu=0.02$ (full line), $\mu=0.01$ and $\nu=0.1$ (dashed, bottom),
308: and $\mu=0.08$ and $\nu=0.1$ (dashed, top).
309: (c) Fraction $f(\varepsilon)$ of uncertain points as a
310: function of the scale $\varepsilon$ for points taken from $L=I_1$ (circles), $L=I_2$ (squares),
311: and the subinterval $L$ of $I_1$ first mapped into $I_2$ (diamonds),
312: where $\mu=0.01$ and $\nu=0.1$.
313: Circles in (c) are shifted vertically upward for clarity.
314: (d) The same as in (c) for $\varepsilon \geq 10^{-35}$ and $L=I_1$ (circles), $L=I_2$ (squares),
315: and $L=I_3$ (triangles). Dashed line (right-side axis): average maximum $j$ of
316: orbits started from $\varepsilon$-uncertain points, for $L=I_1$.
317: }
318: \label{fig3}
319: \end{center}
320: \end{figure}
321:
322:
323: \section{Effective Dynamical Invariants}
324: \label{4}
325:
326: Our results on the chain model motivate us to introduce the concept of
327: effective dynamical invariants. As a specific example, we consider the
328: {\it effective} fractal dimension, which, for the intersection of a fractal set $S$
329: with a $n$-dimensional region $L$, we define as
330: \begin{equation}
331: \left. D_{eff}(L;\varepsilon)= n-\frac{d \ln f(\varepsilon')}{d\ln \varepsilon'}\right|_{\varepsilon'=\varepsilon},
332: \label{2}
333: \end{equation}
334: where $f(\varepsilon')= N(\varepsilon')/N_0(\varepsilon')$, and
335: $N(\varepsilon')$ and $N_0(\varepsilon')$ are the number of cubes of edge length
336: $\varepsilon'$ needed to cover $S\cap L$ and $L$, respectively \cite{prev_work}.
337: We take $L$ to be a generic segment of line [i.e., $n=1$ in Eq. (\ref{2})] intersected
338: by $S$ on a fractal set.
339: In the limit $\varepsilon\rightarrow 0$, we recover the usual box-counting
340: dimension $D=1-\lim_{\varepsilon\rightarrow 0}\Delta \ln f(\varepsilon)/\Delta
341: \ln \varepsilon$ of the fractal set $S\cap L$, which is known to be 1 for all
342: our choices of $L$. However, for any practical purpose, the parameter
343: $\varepsilon$ is limited and cannot be made arbitrarily small (e.g., it cannot
344: be smaller than the size of the particles, the resolution of the experiment, and
345: the length scales neglected in modeling the system). At scale $\varepsilon$ the
346: system behaves as if the fractal dimension were $D_{eff}(L;\varepsilon)$
347: (therefore ``effective'' dimension). In particular, the final state sensitivity
348: of particles launched from $L$, with the initial conditions known within
349: accuracy $\varepsilon^*$, is determined by $D_{eff}(L;\varepsilon^*)$ rather
350: than $D$: as $\varepsilon$ is variated around $\varepsilon^*$, the fraction of
351: particles whose final state is uncertain scales as
352: $\varepsilon^{1-D_{eff}(L;\varepsilon^*)}$, which is different from the
353: prediction $\varepsilon^{1-D}$. This is important in this context because, as
354: shown in Fig.~\ref{fig3} (where the effective dimension is given by $1-\alpha$),
355: the value of
356: $D_{eff}(L;\varepsilon)$
357: may be significantly different from the
358: asymptotic value $D=1$ even for unrealistically small $\varepsilon$ and may also
359: depend on the region of the phase space. Similar considerations apply to many
360: other invariants as well.
361:
362: We now return to the Hamiltonian case. Consider a scattering process in which
363: particles are launched from a line $L$ transversal to the stable manifold $W_s$
364: of the chaotic saddle. Based on the construction suggested by the chain model,
365: it is not difficult to see that $W_s\cap L$ exhibits a hierarchical structure
366: which is not self-similar and is composed of infinitely many nested Cantor sets,
367: each of which is associated with the dynamics inside one of the regions $C_j$.
368: As a consequence, the effective dimension $D_{eff}(L;\varepsilon)$ in
369: Hamiltonian systems is expected to behave similarly to the effective dimension
370: in the chain model [Figs.~\ref{fig3}(c) and \ref{fig3}(d)]. In particular,
371: $D_{eff}(L;\varepsilon)$ is expected to display a strong dependence on $L$ and a
372: weak dependence on $\varepsilon$.
373:
374:
375: \section{Numerical Verification}
376: \label{5}
377:
378: We test our predictions on the area preserving H\'enon map: $f(x,y)=
379: (\lambda -y -x^2,x)$, where $\lambda$ is the bifurcation parameter. In this
380: system, typical points outside KAM islands are eventually mapped to infinity.
381: Because of the symmetry $f^{-1}=g\circ f\circ g$, where $g(x,y)=(y,x)$,
382: the stable and unstable manifolds of the chaotic saddle are obtained from each other
383: by exchanging $x$ and $y$.
384: For $\lambda=0.05$, the system displays a period-one and a period-four major island, as shown in
385: Fig.~\ref{fig4}(a). In the same figure we also show the complex invariant structure
386: around the islands, the stable manifold of the chaotic saddle, and three different
387: choices for the line of starting points: a large interval away from the islands
388: ($L_a$), a small subinterval of this interval where the stable manifold
389: appears to be denser ($L_b$), and an interval closer to the islands ($L_c$).
390: The corresponding effective dimensions are computed
391: for a wide interval of $\varepsilon$. The results are shown in Fig.~\ref{fig4}(b):
392: $D_{eff}(L_a;\varepsilon)= 0.84$,
393: $D_{eff}(L_b;\varepsilon)= 0.90$, and
394: $D_{eff}(L_c;\varepsilon)= 0.97$ for $10^{-8}<\varepsilon< 10^{-5}$.
395: These results agree with
396: our predictions that the effective fractal dimension has the following
397: properties: $D_{eff}$ may be significantly different from the asymptotic
398: value $1$ of the fractal dimension; $D_{eff}$ depends on the
399: resolution $\varepsilon$ but is nearly constant over decades; $D_{eff}$ depends
400: on the region of the phase space under consideration and, in particular,
401: is larger in regions closer to the islands and in regions where the stable
402: manifold is denser.
403: Similar results are expected for any typical Hamiltonian system with mixed phase space.
404:
405:
406: % FIGURE 4:
407: \begin{figure}[pt]
408: \begin{center}
409: \epsfig{figure=fig4a.eps,width=5.0cm}
410: \epsfig{figure=fig4b.eps,width=5.0cm}
411: \caption{(a) KAM islands (blank), stable manifold (gray),
412: and the lines of initial conditions ($L_b$ is a subinterval of $L_a$).
413: (b) Effective dimension for $L=L_a$ (circles), $L=L_b$ (squares),
414: and $L=L_c$ (triangles). The data in (b) are shifted vertically for clarity.}
415: \label{fig4}
416: \end{center}
417: \end{figure}
418:
419:
420: \section{Conclusions}
421:
422: %In summary,
423:
424: We have shown that the finite-scale dynamics of Hamiltonian systems,
425: relevant for realistic situations, is governed by effective dynamical
426: invariants. The effective invariants are not only different from the asymptotic
427: invariants but also from the usual hyperbolic invariants because they strongly
428: depend on the region of the phase space. Our results are generic and expected
429: to meet many practical applications. In particular, our results are expected to
430: be relevant for fluid flows, where the advection dynamics of tracer particles is
431: often Hamiltonian \cite{fluid}. In this context, a slow nonuniform convergence
432: of effective invariants is expected not only for time-periodic flows, capable of
433: holding KAM tori, but also for a wide class of time-irregular incompressible
434: flows with nonslip obstacles or aperiodically moving vortices.
435:
436: \acknowledgements
437:
438: This work was supported by MPIPKS, FAPESP, and CNPq.
439: A. E. M. thanks Rainer Klages for illuminating discussions.
440:
441:
442:
443: \appendix*
444: \section{}
445: %\section{Diffusion Model}
446: \label{appendix}
447:
448: The diffusion model is: $\partial_t P(x,t) = \partial_x [x^{\alpha}\partial_x
449: P(x,t)]$, where $P$ is the probability density of all particles, $x\ge 0$, and
450: $\alpha>2$ \cite{chirikov1}. The outermost torus of the KAM island is at $x=0$,
451: where the diffusion rate (proportional to $x^{\alpha}$) vanishes. In a chaotic
452: scattering process the initial distribution of particles is localized apart from
453: the confining islands. We take $P(x,0)=\delta (x-x_0)$, $x_0>0$, and consider a
454: particle to escape when it reaches $x\geq x_1$. Under the approximation that
455: for large $x_1$ the return of particles can be neglected, we disregard the
456: boundary condition $P(x_1,t)=0$ and we take the solution to be the corresponding
457: Green function: $P(x,t)=(\alpha-2) (xx_0)^{-1/2} yy_0 \exp(-y^2-y_0^2)
458: I_{\beta}(2yy_0)$, where $y=(\alpha-2)^{-1}t^{-1/2}x^{-(\alpha-2)/2}$, $y_0$ is
459: $y$ at $x=x_0$, $\beta=(\alpha-1)/(\alpha-2)$, and $I_{\beta}$ is the modified
460: Bessel function, which scales as $I_{\beta}\sim (2yy_0)^{\beta}$ for small
461: $2yy_0$ \cite{greene1}. For any fixed $x>0$, we can show that the distribution
462: for large $t$ decreases as $P(x,t)\sim t^{-\beta-1}$, where
463: $\beta=(\alpha-1)/(\alpha-2)$. On the other hand, as shown in Ref.~\cite{pikovsky1},
464: the fraction of particles in the interval $x<x_1$ decays
465: algebraically as $Q(t)\equiv\int_0^{x_1} P(x,t)$d$x\sim t^{-\beta}$. Combining
466: these two results, it follows that the normalized probability density
467: $\rho(x,t)\equiv P(x,t)/Q(t)$ decreases as $\rho(x,t) \sim t^{-1}$ at each fixed
468: $x\in (0,x_1)$ for large enough $t$ and diverges arbitrarily close to $x=0$.
469:
470:
471:
472: \begin{references}
473:
474: \bibitem{chirikov1}
475: B. V. Chirikov, in {\it Lecture Notes in Physics} {\bf 179} (Springer, Berlin, 1983), pp. 29-46.
476:
477: \bibitem{asymp}
478: C. F. F. Karney, Physica D {\bf 8}, 360 (1983);
479: R. S. MacKay, J. D. Meiss, and I. C. Percival, Physica D {\bf 13}, 55 (1984);
480: B. V. Chirikov and D. L. Shepelyansky, Physica D {\bf 13}, 395 (1984).
481:
482: \bibitem{meiss1}
483: J. D. Meiss and E. Ott, Physica D {\bf 20}, 387 (1986).
484:
485: \bibitem{greene1}
486: J. M. Greene, R. S. Mackay, and J. Stark, Physica D {\bf 21}, 267 (1986).
487:
488: \bibitem{pikovsky1}
489: A. S. Pikovsky, J. Phys. A: Math. Gen. {\bf 25}, L477 (1992).
490:
491: \bibitem{christiansen1}
492: F. Christiansen and P. Grassberger, Phys. Lett. A {\bf 181}, 47 (1993).
493:
494: \bibitem{lau1}
495: Y.-T. Lau, J. M. Finn, and E. Ott, Phys. Rev. Lett. {\bf 66}, 978 (1991).
496:
497: \bibitem{zasl}
498: G. M. Zaslavsky, M. Edelman, and B. A. Niyazov, Chaos {\bf 7}, 159 (1997);
499: and references therein.
500:
501: \bibitem{uptodate}
502: M. Weiss, L. Hufnagel, and R. Ketzmerick, Phys. Rev. E {\bf 67}, 046209 (2003).
503:
504:
505: \bibitem{fluid}
506: C. Jung, T. T\'el, and E. Ziemniak, Chaos {\bf 3}, 555 (1993).
507:
508: \bibitem{nh_h}
509: J. C. Sommerer, H.-C. Ku, and H. E. Gilreath, Phys. Rev. Lett. {\bf 77}, 5055 (1996).
510:
511:
512: \bibitem{sol_stat}
513: R. Ketzmerick, Phys. Rev. B {\bf 54}, 10841 (1996)
514:
515: \bibitem{relat}
516: A. E. Motter and P. S. Letelier, Phys. Lett. A {\bf 285}, 127 (2001).
517:
518: %\bibitem{motter:long}
519:
520: \bibitem{other_chains}
521: Chain models have also been used to describe power laws generated by a hierarchy
522: of repellers in the motion towards the Feigenbaum attractor [P. Grassberger and
523: M. Scheunert, J. Stat. Phys. {\bf 26}, 697 (1981)] and to study
524: deterministic diffusion [R. Klages and J. R. Dorfman, Phys. Rev. Lett.
525: {\bf 74}, 387 (1995)].
526:
527: \bibitem{hier}
528: Generalization to a full hierarchy of KAM
529: islands is straightforward and does not change the essence of our results.
530:
531: \bibitem{details} The escape rate follows from the same argument used in [J.
532: D. Meiss and E. Ott, Physica D {\bf 20}, 387 (1986)]. The fractal dimension
533: follows from the observation that the dimension of the invariant set of map
534: $M_j$ goes to $1$ as $j\rightarrow \infty$.
535:
536:
537: \bibitem{uncert}
538: C. Grebogi, S. W. McDonald, E. Ott, and J. A. Yorke, Phys. Lett. A {\bf 99}, 415 (1983).
539:
540: \bibitem{avnir1}
541: D. Avnir, O. Biham, D. Lidar, and O. Malcai, Science {\bf 279}, 39 (1998).
542:
543:
544: \bibitem{prev_work}
545: Different effective invariants have been introduced in Refs.~\cite{finite1}.
546: See also Refs.~\cite{finite2}.
547:
548:
549: \bibitem{finite1}
550: A. E. Motter, Y.-C. Lai, and C. Grebogi, Phys. Rev. E {\bf 68}, 056307 (2003);
551: A. P. S. de Moura and C. Grebogi, to be published.
552:
553:
554: \bibitem{finite2}
555: Y.-C. Lai, M. Ding, C. Grebogi, and R. Bl\"umel, Phys. Rev. A {\bf 46}, 4661 (1992);
556: W. Breymann, Z. Kov\'acs, and T. T\'el, Phys. Rev. E {\bf 50}, 1994 (1994);
557: P. Gaspard and J. R. Dorfman, Phys. Rev. E {\bf 52}, 3525 (1995);
558: V. Constantoudis and C. A. Nicolaides, Phys. Rev. E {\bf 64} 056211 (2001).
559:
560:
561: \end{references}
562:
563: %\newpage
564:
565:
566:
567:
568:
569:
570: \end{document}
571: