nlin0110010/lpv.tex
1: %% This document created by Scientific Word (R) Version 3.0
2: \documentclass[12pt]{article}
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: \usepackage{amsfonts}
6: \usepackage{amssymb}
7: %TCIDATA{OutputFilter=latex2.dll}
8: %TCIDATA{CSTFile=LaTeX article (bright).cst}
9: %TCIDATA{Created=Wed Jul 18 09:34:45 2001}
10: %TCIDATA{LastRevised=Fri Jul 20 14:51:33 2001}
11: %TCIDATA{<META NAME="GraphicsSave" CONTENT="32">}
12: %TCIDATA{<META NAME="DocumentShell" CONTENT="General\Blank Document">}
13: %TCIDATA{Language=American English}
14: \renewcommand{\baselinestretch}{2}
15: \newtheorem{theorem}{Theorem}
16: \newtheorem{acknowledgement}[theorem]{Acknowledgement}
17: \newtheorem{algorithm}[theorem]{Algorithm}
18: \newtheorem{axiom}[theorem]{Axiom}
19: \newtheorem{case}[theorem]{Case}
20: \newtheorem{claim}[theorem]{Claim}
21: \newtheorem{conclusion}[theorem]{Conclusion}
22: \newtheorem{condition}[theorem]{Condition}
23: \newtheorem{conjecture}[theorem]{Conjecture}
24: \newtheorem{corollary}[theorem]{Corollary}
25: \newtheorem{criterion}[theorem]{Criterion}
26: \newtheorem{definition}[theorem]{Definition}
27: \newtheorem{example}[theorem]{Example}
28: \newtheorem{exercise}[theorem]{Exercise}
29: \newtheorem{lemma}[theorem]{Lemma}
30: \newtheorem{notation}[theorem]{Notation}
31: \newtheorem{problem}[theorem]{Problem}
32: \newtheorem{proposition}[theorem]{Proposition}
33: \newtheorem{remark}[theorem]{Remark}
34: \newtheorem{solution}[theorem]{Solution}
35: \newtheorem{summary}[theorem]{Summary}
36: \newenvironment{proof}[1][Proof]{\textbf{#1.} }{\ \rule{0.5em}{0.5em}}
37: 
38: 
39: \begin{document}
40: 
41: 
42: \title{\bf Markov-chain approach to a process with long-time memory}
43: 
44: \author{Guglielmo Lacorata\\Dipartimento di Fisica, Universit\'{a} di Roma 
45: ``La Sapienza'' Italy, and\\
46:   Dipartimento di Fisica, Universit\'{a} de L'Aquila, Italy\\
47: \and Rub\'{e}n A. Pasmanter\\Royal Dutch Meteorological Institute, 
48: Netherlands\\
49:  \and Angelo Vulpiani\\Dipartimento di Fisica, INFM and SMC,\\ 
50: Universit\'{a} di Roma ``La Sapienza'', Italy\\
51: }
52: 
53: 
54: \maketitle
55: 
56: 
57: \thispagestyle{empty}
58: \newpage
59: \setcounter{page}{1} 
60: \pagestyle{plain}
61: 
62: \begin{abstract}
63: \noindent 
64: We show that long-term memory effects, present in the chaotic
65: dispersion process generated by a meandering jet model, can be 
66: nonetheless taken into account 
67: by a first order Markov process, provided that the 
68: states of the phase space partition, chosen to describe the
69: system, be appropriately defined.  
70: \end{abstract}
71: 
72: 
73: 
74: \section{Introduction}
75: 
76: Geophysical processes often exhibit an irregular behavior\ that is due to
77: deterministic chaos and/or to the presence of many relevant degrees of freedom.
78: Striking examples are climate dynamics (Fraedrich 1988, Nicolis et al. 1997) 
79: and
80: transport processes in the atmosphere and in the oceans where
81: many different characteristic temporal scales are involved (Samelson 1992, 
82: Yang 1996). 
83: After Lorenz's
84: seminal work (Lorenz 1963), it is now well recognized that even deterministic
85: systems with as few as three degrees of freedom can have complex temporal
86: evolution  similar to genuine stochastic processes 
87: (Ott 1993, Beck and Schl\"ogl 1993). For
88: example, the large-scale and long-time dispersion characteristics of drifters
89: deterministically advected by a non-stochastic velocity field, giving rise to
90: Lagrangian chaos, can, sometimes, be well described by random walk models
91: (Ottino 1989, Crisanti et al. 1991). 
92: On the other hand, a simple probabilistic description, e.g., one in
93: terms of autoregressive models and Markov chains (Vautard et al. 1990, 
94: Nicolis et al. 1997), 
95: is often not able to catch the
96: richness of most geophysical phenomena. 
97: 
98: In a recent paper 
99: (Cencini et al. 1999),
100: mixing and transport in a meandering jet were investigated and compared with a
101:  Markov chain. It was found that, due to the presence of very
102: different characteristic time scales, this simple approach was not fully
103: satisfactory. In fact, the ubiquitous presence of a hierarchy of time scales
104: forces one to develop nonstandard techniques for data analysis and for their
105: modeling. 
106: 
107: The main aim of this Note is to show how a suitable Markov chain
108: with a small number of states can reproduce in a proper way the main
109: statistical features of systems similar to the one that is considered below.
110: The basic point is an appropriate definition of the states of the system, 
111: such that long-time memory effects are straightforwardly taken into 
112: account in a Markov chain approximation of the dynamics.  
113: We expect this kind of analysis to be of interest in the more general context
114: of systems with multiple time scales.
115: 
116: \section{Transport in a meandering jet}
117: 
118: The test case under consideration consists of a 2D meandering jet, formerly
119: introduced as a model for studying Lagrangian transport across the Gulf Stream
120:  (Bower 1991, Samelson 1992). 
121: In this model, the steady flow is described by the
122: following stream function:
123: \begin{equation}
124: \Psi(x,y)=-\tanh\left\{  k\frac{y-B\cos\left(  kx\right)  }{\sqrt{1+B^{2}k^{2}
125: \sin^{2}\left(  kx\right)  }}\right\}  +cy
126: \label{eq:steadyflow}%
127: \end{equation}
128: where $x$ and $y$ are the fluid particle zonal and meridional coordinates, 
129: respectively, $k$ is the
130: wavenumber of the flow along the $x$-direction, $B$ controls the amplitude of
131: the meanders and $c$ is the retrograde, counter stream, velocity component far
132: from the jet's core which is centered on $y=0$. The advective time, 
133: $\tau_{adv}$, is defined 
134: as the time needed for a particle, traveling close to the jet core, 
135: to move along the $x$ direction  a distance 
136: equal to the wavelength of the flow 
137: $\lambda = 2 \pi / k$. For the typical values of the parameters used in 
138: our calculations, the value of this characteristic time is   
139: $\tau_{adv} \simeq 4 \pi$. The recirculation time, $\tau_{rec}$, is defined 
140: as the period of a trajectory moving on a closed orbit next to the 
141: boundary of a gyre. This time is 
142: $\tau_{rec} \sim 10 \cdot \tau_{adv}$.   
143: The 2D incompressible velocity
144: field $(u,v)$ and the stream function $\Psi(x,y)$ are related by,
145: \begin{equation}
146: u=-{\frac{\partial\Psi}{\partial y}},\;\;\;\;v={\frac{\partial\Psi}{\partial
147: x}}.
148: \label{eq:vel}%
149: \end{equation}
150: The flow is invariant under north-south inversion ($y$ direction) 
151: and is periodic along
152: the $x$-axis with period $\lambda$. 
153: In the absence of perturbations, passive particles follow the
154: streamlines at constant $\Psi$. The steady flow corresponding to this stream
155: function consists of open streamlines, to be identified as ``the jet'', and
156: closed streamlines, to be called ``the gyres'', see Figure~
157: \ref{fig:flow}. Non-trivial behavior appears as soon as a perturbation is
158: added to the above-described flow. In particular, cross-stream transport,
159: i.e., from gyre to gyre, occurs either by the action of a random forcing or
160: due to non-stochastic, time dependent terms inserted in the stream function
161: which lead to the onset of Lagrangian chaos. In the stochastic model the
162: position of a passive particle evolves according to
163: \begin{equation}
164: \frac{dx}{dt}=u(x,y)+w_{1}(t),\;\;\;\;\frac{dy}{dt}=v(x,y)+w_{2}%
165: (t)\label{eq:noise}%
166: \end{equation}
167: where the stochastic fields $w_{i}(t)$ have red-noise correlations, i.e.,
168: \begin{equation}
169: <w_{i}(t)w_{j}(t^{\prime})>=2\sigma^{2}\delta_{ij}e^{-|t-t^{\prime}|/\tau
170: },\;i,j={1,2,}\label{eq:rednoise}%
171: \end{equation}
172: with $\sigma^{2}$ the variance of the stochastic fields and $\tau$ their
173: correlation time. The deterministic, chaotic model is obtained by making the
174: amplitude of the meanders, $B$ in equation (\ref{eq:steadyflow}), time
175: dependent,
176: \begin{equation}
177: B\Rightarrow B(t)=B_{0}+\epsilon\cos(\omega t+\phi)\label{eq:modB}%
178: \end{equation}
179: where $B_{0}$ is the mean meander amplitude and 
180:  $\epsilon$ and $\omega$ are appropriately 
181: chosen so as to generate large-scale
182: mixing, as in Cencini et al. (1999), 
183: i.e., leading to an overlap of resonances (Chirikov 1979); 
184: $\phi$ is an arbitrary
185: phase. 
186: 
187: The symbolic dynamics (Beck and Schl\"ogl 1993) 
188: of the noisy model can be satisfactorily approximated by
189: a simple minded, first order Markov process. 
190: In the sequel we
191: show that, in order to describe the chaotic dynamics via a first order Markov
192: chain, it is necessary to introduce  
193:  a more suitable definition of ``state'' which 
194: take into account the non trivial dynamical behavior of the system. 
195: This is made clear by comparing the symbolic dynamics generated by two
196: different partitions of the phase space, both partitions having four cells.
197: 
198: The simple minded partition, call it $\Pi$, is defined by the southern gyre
199: (symbol $1$), the southern half jet (symbol $2$), the northern half jet
200: (symbol $3$) and the northern gyre (symbol $4$), see
201: Figure~\ref{fig:partition1}. The partition takes into account the north-south
202: symmetry as well as the periodicity of the system. The position of the
203: advected particles is sampled millions of times at a time interval
204: $T=2\pi/\omega.$ At each sampling time, one associates one of the symbols
205: according to the partition cell in which the particle is located at that time.
206: In this way, a symbolic sequence is generated from each trajectory. Once this
207: has been done, one computes the transition matrix $W$, giving the
208: state-to-state transition probabilities in the time interval $T.$ 
209: 
210: While the precise nature of the perturbation, i.e.,
211: stochastic or purely deterministic, does not affect the gross qualitative
212: features of the dispersion process, it does lead to very different correlation
213: functions. In the deterministic, chaotic case, the transport process shows
214: persistent, nontrivial long term correlations which are suppressed when
215: stochastic noise is present. These differences are clearly seen in Figure~ 
216: \ref{fig:noise_chaos_sig}.
217: The $W_{ij}$ 
218: matrix element is the probability to observe state $j$ at time $(n+1)T$ 
219: knowing that the state $i$ occurs at time $nT$ (where $n$ is integer).   
220: In the
221: present case, using partition $\Pi,$ this analysis leads to,%
222: 
223: \begin{equation}
224: W=\left(
225: \begin{array}
226: [c]{cccc}%
227: 0.851 & 0.147 & 0.002 & 0.000\\
228: 0.247 & 0.535 & 0.215 & 0.003\\
229: 0.003 & 0.215 & 0.535 & 0.247\\
230: 0.000 & 0.002 & 0.147 & 0.851
231: \end{array}
232: \right)  .\label{eq:Wsc1}%
233: \end{equation}
234: The symmetries in the transition matrix $W$ reflect the dynamical and
235: statistical equivalence of the northern and southern halves of the system. For
236: future reference, we also quote the corresponding steady probability vector
237: $P=(P_{1},P_{2},P_{3},P_{4})$ which satisfies the equation  
238: \begin{equation}
239: \sum_{i=1}^4 P_i W_{ij} = P_j
240: \label{eq:eigenvec}
241: \end{equation} 
242: One finds: in the
243: gyres, \ $P_{1}=P_{4}=0.315,$ and in the jet, $P_{2}=P_{3}=0.185.$
244: 
245: In contraposition to the stochastic model defined by Eqs. (\ref{eq:noise}) and
246: (\ref{eq:rednoise}), the deterministic evolution with (\ref{eq:modB}) is
247: characterized by long-term correlations which are poorly reproduced by the 
248: Markov chain generated from $W.$
249: Essentially, the matrix elements $W_{23}$ and $W_{32}$ overestimate the
250: coupling between the two\ halves of the systems. This failure of the simple 
251: probabilistic model based on $W$ is
252: illustrated in Figure~\ref{fig:chaos_corr_smarkov} where the actual
253: autocorrelation functions of the states 1 and 2 are compared with the
254: corresponding correlation functions generated by the first order Markovian
255: process described by $W.$ 
256: 
257: As one can see, in the actual dynamics there are basically two
258: characteristic times, $\tau_f$ and $\tau_s$, 
259: that correspond to the inverse of the two exponential
260: decay rates at short and at long times, respectively. As order of magnitude, 
261: the shorter decay time is $\tau_f \sim T$, and the 
262: longer one is $\tau_s \sim 10^2 \cdot T$.    
263: 
264: A way of
265: estimating a lower bound for the order of the Markov chain which would be
266: necessary in order to adequately reproduce the system's statistical features,
267: see, e.g., (Khinchin 1957), is based on the block entropies $H_{n}$
268: which are defined by
269: \[
270: H_{n}=-\sum_{C_{n}}P(C_{n})\ln P(C_{n}),
271: \]
272: where $P(C_{n})$ is the probability of observing a sequence $C_{n}$
273: $=(i_{t+1},i_{t+2},.....,i_{t+n})$ of $n$ successive symbols generated by the
274: dynamical system, with $i_{m}\in\{1,...,4\}$ the partition cell visited at the
275: time $t=mT$. From the block entropies $H_{n}$ one computes next the quantities
276: $h_{n}=H_{n}-H_{n-1}$ which represent the average additional information
277: needed to specify the $n$-th symbol $i_{n}$ given the sequence $C_{n-1}$. The
278: limit of $h_{n}$ for $n\rightarrow\infty$ gives the Shannon entropy $h_{S}$ of
279: the infinite sequence $i_{1},i_{2},......$ For a Markov process of order 
280: $\nu$,
281: one has $h_{n}=h_{S}$ for all $n\geq \nu+1$, (Khinchin 1957). 
282:  In agreement with what was found in the paper by 
283:  Cencini et al. (1999), 
284: at least in order to reproduce the entropic properties, one would
285: need a Markov process of very high order. Recall that this procedure
286: gives only a lower bound to the order of the Markov chain.
287: It is clear that the strategy consisting of
288: simply increasing the order of the Markov process is not practical at all.
289: Surprisingly, we were able to give a non-obvious choice of the partition
290: that allows us to work with a first order Markov chain where now the time
291: memory effects are satisfactorily described. 
292: This is explained in the following. 
293: The new partition, denoted by $\Pi^{\ast},$ is a 4-state partition similar to
294: the previous $\Pi$ but the jet is partitioned in a different way: the jet
295: corresponds to symbol $2$ if the last previously visited cell is the southern
296: gyre $1$ or the cell $2$ itself; if the last previously visited cell is the
297: northern gyre $4$ or the cell $3$ itself, then the jet corresponds to symbol
298: $3$, see Figure~\ref{fig:partition2}. 
299: Accordingly, transitions between the two jet states $2$ and $3$ are not 
300: possible. 
301: This partition of the jet preserves
302: part of the long time memory of the gyre-to-current transitions, a feature of
303: the deterministic chaotic model which is impossible to reproduce with the
304: previous partition $\Pi$. In other words, this partition introduces some
305: memory even if formally the model remains of first order. The time signal
306: generated by the $\Pi^{\ast}$ Markov process is compared to the actual time
307: signal in Figure~\ref{fig:chaos_sig}. The memory effects are visible through
308: the existence of ``blocks'' describing the fast back and forth gyre-to-current
309: transitions in one half of the system, before moving to the other half. The
310: new transition matrix $W^{\ast}$ is found to be,
311: \begin{equation}
312: \;W^{\ast}=\left(
313: \begin{array}
314: [c]{cccc}%
315: 0.850 & 0.150 & 0.000 & 0.000\\
316: 0.243 & 0.749 & 0.000 & 0.008\\
317: 0.008 & 0.000 & 0.749 & 0.243\\
318: 0.000 & 0.000 & 0.150 & 0.850
319: \end{array}
320: \right)  .\label{eq:Wsc2}%
321: \end{equation}
322: The corresponding time-independent probability vector $P^{\ast}$ is equal to
323: the one found for the $\Pi$ partition, $P^{\ast}=P,$ as it should be.
324: Notice that, because the transitions between states $1$ and $2$ ($3$ and $4$) 
325: are much more frequent than those from state $3$ to $1$ (from $2$ to $4$), 
326: the matrix $W^{\ast}$ is not far from being block-diagonal. 
327:  
328: The improvements introduced by the $\Pi^{\ast}$ partition show up in the
329: correlation functions it generates, as can be seen in
330: Figure~\ref{fig:chaos_corr_hmarkov}, and in the fast convergence of the
331: $h_{n}$ entropies shown in Figure~\ref{fig:shannon}. 
332: Moreover, with the $\Pi^{\ast}$
333: partition the two characteristic times are well estimated.
334: 
335: \section{Conclusions}
336: 
337: We have shown that the long-term memory effects which are present in the
338: chaotic dispersion processes generated by the flow defined by Eqs.
339: (\ref{eq:steadyflow}) and (\ref{eq:modB}) can be reproduced by a 
340: Markov chain, provided that the partition of the phase space is performed in a
341: special way. 
342: It is important to remark that first order in a Markov process generated with 
343: the $\Pi^*$ partition corresponds to a very large order, 
344: say $\sim O(\tau_s/T)$, 
345: in a Markov process generated from the $\Pi$ partition. 
346: We hope that the ideas presented in this Note will be of use in a
347: more general context and we plan to apply the above discussed technique to the
348: analysis of other geophysical phenomena. Needless to say, when more than two
349: characteristic times are relevant, partitions with a larger number of cells
350: will be required.
351: 
352: \section{Acknowledgements}
353: 
354: R.A.P. acknowledges the hospitality of the Roman TNT group at University 
355: ``La Sapienza''. 
356: A.V. acknowledges support 
357: from the INFM {\em Center for Statistical Mechanics and Complexity}.
358: 
359:  
360: 
361: 
362: 
363: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
364: \newpage
365: \begin{thebibliography}{99}
366: 
367: 
368: \item[]
369: Beck, C. and F. Schl\"ogl, 1993: Thermodynamics of chaotic systems. 
370: ({\it Cambridge University Press}, Cambridge UK).
371: 
372: \item[]
373: Bower, A.S., 1991: A simple kinematic mechanism for mixing fluid parcels 
374:  across a meandering jet. 
375: {\it J. Phys. Oceanogr.}, {\bf 21}, 173-180. 
376: 
377: \item [] Cencini, M., G. Lacorata, A. Vulpiani 
378: and E. Zambianchi, 1999:  
379: Mixing in a meandering jet: a Markovian approximation. 
380: {\it J. Phys. Oceanogr.}, {\bf 29}, 2578-2594.
381: 
382: \item[]
383: Chirikov, B.V., 1979:
384: A universal instability of many-dimensional oscillator
385: systems.
386: {\it Phys. Rep.}, {\bf 52}, 263-379.
387: 
388: 
389: \item[]Crisanti, A., M. Falcioni, G. Paladin and A. Vulpiani, 1991:
390: Lagrangian Chaos: Transport, Mixing and Diffusion in Fluids.
391: {\it La Rivista del Nuovo Cimento}, {\bf 14}, 1-80.
392: 
393: 
394: 
395: \item[]
396: Fraedrich, K., 1988: El Ni\~no-Southern Oscillation predictability. 
397: {\it Mon. Wea. REv.}, {\bf 116}, 1001-1012.
398: 
399: \item[]
400: Khinchin, A.I., 1957: Mathematical foundations of Information Theory.
401: {\it Dover}.  
402: 
403: 
404: 
405: \item[]
406: Lorenz, E.N., 1963: Deterministic non-periodic flow. 
407: {\it J. Atmos. Sci.}, {\bf 20}, 130-141.
408: 
409: \item[]
410: Nicolis, C., W. Ebeling and C. Baraldi, 1997: Markov processes, dynamic 
411: entropies and statistical prediction of mesoscale weather regimes. 
412: {\it Tellus}, {\bf 49A}, 108-118.
413: 
414: \item[]
415: Ott, E., 1993: Chaos in dynamical systems. 
416: ({\it Cambridge University Press}, Cambridge UK).
417: 
418: \item[] Ottino, J.M., 1989:  
419: The kinematics of mixing: stretching, chaos and transport.
420: ({\it Cambridge University Press}, Cambridge UK).
421: 
422: 
423: 
424: \item[]
425: Samelson, R.M., 1992: Fluid exchange across a meandering jet. 
426: {\it J. Phys. Oceanogr.}, {\bf 22}, 431-440.
427: 
428: 
429: Vautard, R., K. C. Mo, and M. Ghil, 1990:
430:   Statistical significance test for transition matrices of atmospheric
431: Markov chains.
432: {\it J. Atmos. Sci.}, {\bf 47},  1926-1931.
433: 
434: \item[]
435: Yang, H., 1996: Chaotic transport and mixing by ocean gyre circulation. 
436: {\it Stochastic modelling in Physical Oceanography}, 
437: R.J. Adler, P. Muller and B.L. Rozovskii, Eds. Birkh\"auser, 439-466.
438: 
439: 
440: 
441: 
442: 
443: 
444: \end{thebibliography}
445: 
446: \newpage
447: 
448: \centerline{FIGURE CAPTIONS}
449: 
450: 
451: \noindent 
452: FIGURE~\ref{fig:flow}. 
453: Schematic diagram of the circulation in the meandering jet model. 
454: Arrows indicate the direction of particle motion: ``westerly'' advection 
455: in the jet current, clockwise and anti-clockwise recirculation in the 
456: southern and northern gyres, respectively. Spatial coordinates $X$ and $Y$ 
457: are given in unit of the wavelength $\lambda=2\pi/k$ and $B_0$ is 
458: set equal to $0.16 \lambda$, see eq. 
459: (\ref{eq:steadyflow}). 
460: 
461: 
462: \noindent 
463: FIGURE~\ref{fig:partition1}. 
464: $\Pi$
465: partition of the steady meandering jet: 1) south gyre; 2) south half-jet; 3)
466: north half-jet; 4) north gyre.
467: 
468: \noindent 
469: FIGURE~\ref{fig:noise_chaos_sig}. 
470: Comparative plot of the two typical symbolic sequences obtained from:
471:  a) the stochastic model and b) the chaotic model, with the parameters
472:  in eq. (\ref{eq:modB}) being $\epsilon = B_0 / 4$ and 
473: $\omega \simeq 2 \pi / \tau_{adv}$. The parameters of the red noise in 
474: eq. (\ref{eq:rednoise}) are $\sigma^2=6 \cdot 10^{-2} (\lambda/\tau_{adv})^2$ 
475: and $\tau=T/4$. 
476: 
477: 
478: \noindent 
479: FIGURE~\ref{fig:chaos_corr_smarkov}. 
480: Autocorrelation functions for the states $1$ and $2$ of the $\Pi
481: $ partition, see Figure~\ref{fig:partition1}: actual autocorrelations
482: from the chaotic model (full lines) 
483: and those generated by the transition-rate matrix 
484: $W$ (dotted lines). Parameters of
485: the deterministic perturbation: as in Figure~\ref{fig:noise_chaos_sig}. 
486: 
487: \noindent 
488: FIGURE~\ref{fig:partition2}. 
489: $\Pi
490: ^{\ast}$ partition of the steady meandering jet: 1) south gyre; 2) and 3) jet;
491: 4) north gyre. The central jet corresponds to state $2$ if the last visited
492: gyre is state $1$, and to state $3$ if the last visited gyre is state $4$.
493: 
494: 
495: 
496: \noindent 
497: FIGURE~\ref{fig:chaos_sig}
498: Comparative plot of: a) the chaotic model symbolic dynamics using 
499: the partition  
500: $\Pi^*$ and b) the symbol sequence simulated on the basis of the 
501: transition-rate matrix 
502: $W^{*}$.  
503: 
504: \noindent 
505: FIGURE~\ref{fig:chaos_corr_hmarkov}. 
506: Autocorrelation functions for the states $1$ and $2$ of the 
507: $\Pi^{*}$
508:  partition, see Figure~\ref{fig:partition2}: actual autocorrelations from
509: the chaotic model (full lines) and from the Markovian approximation 
510:  generated with $W^{*}$ (dotted lines).  
511: 
512: \noindent 
513: FIGURE~\ref{fig:shannon}. 
514: Differential Block Entropy $h_n = H_{n+1}-H_{n}$, as a function 
515: of the sequence-length $n$, for 
516: the $\Pi$ ($\star$) and the $\Pi^*$ ($\square$) partitions. 
517: For each length $n$ approximately 
518:  $\sim 10^7$ sequences were used.
519: Notice the very different rate of convergence of $h_n$ to its limit 
520: value (Shannon entropy) $h_S$ in the two cases. 
521: 
522: 
523: 
524: \newpage
525: 
526: \begin{figure}[ptb]
527: \includegraphics[angle=-90, width=1.0\textwidth]{flow.ps}
528: \caption{
529: %Schematic diagram of the circulation in the meandering jet model. 
530: %Arrows indicate the direction of particle motion: ``westerly'' advection 
531: %in the jet current, clockwise and anti-clockwise recirculation in the 
532: %southern and northern gyres, respectively. Spatial coordinates $X$ and $Y$ 
533: %are given in unit of the wavelength $\lambda=2\pi/k$ and $B_0$ is 
534: %set equal to $0.16 \lambda$, see eq. 
535: %(\ref{eq:steadyflow}). 
536:  }%
537: \label{fig:flow}%
538: \end{figure}
539: 
540: \pagebreak[4]
541: 
542: \begin{figure}[ptb]
543: \includegraphics[angle=-90, width=1.0\textwidth]{oldpart.ps}
544: \caption{
545: %$\Pi$
546: %partition of the steady meandering jet: 1) south gyre; 2) south half-jet; 3)
547: %north half-jet; 4) north gyre. 
548: }
549: \label{fig:partition1}%
550: \end{figure}
551: 
552: \pagebreak[4]
553: 
554: \begin{figure}[ptb]
555: \centerline{a)}
556: \includegraphics[angle=-90, width=0.75\textwidth]{noisysig.ps}%
557: 
558: \centerline{b)}
559: \includegraphics[angle=-90, width=0.75\textwidth]{chaosig.ps}%
560: \caption{
561: %Comparative plot of the two typical symbolic sequences obtained from:
562: % a) the stochastic model and b) the chaotic model, with the parameters
563: % in eq. (\ref{eq:modB}) being $\epsilon = B_0 / 4$ and 
564: %$\omega \simeq 2 \pi / \tau_{adv}$. The parameters of the red noise in 
565: %eq. (\ref{eq:rednoise}) are $\sigma^2=6 \cdot 10^{-2} (\lambda/\tau_{adv})^2$ 
566: %and $\tau=T/4$. 
567: }
568: \label{fig:noise_chaos_sig}%
569: \end{figure}
570: 
571: \pagebreak[4]
572: 
573: \begin{figure}[ptb]
574: \includegraphics[angle=-90, width=1.0\textwidth]{oldcorr.ps}%
575: \caption{
576: %Autocorrelation functions for the states $1$ and $2$ of the $\Pi
577: %$ partition, see Figure~\ref{fig:partition1}: actual autocorrelations
578: %from the chaotic model (full lines) 
579: %and those generated by the transition-rate matrix 
580: %$W$ (dotted lines). Parameters of
581: %the deterministic perturbation: as in Figure~\ref{fig:noise_chaos_sig}. 
582: }
583: \label{fig:chaos_corr_smarkov}%
584: \end{figure}
585: 
586: \pagebreak[4]
587: 
588: \begin{figure}[ptb]
589: \includegraphics[angle=-90, width=1.0\textwidth]{newpart.ps}
590: \caption{
591: %$\Pi
592: %^{\ast}$ partition of the steady meandering jet: 1) south gyre; 2) and 3) jet;
593: %4) north gyre. The central jet corresponds to state $2$ if the last visited
594: %gyre is state $1$, and to state $3$ if the last visited gyre is state $4$. 
595: }
596: \label{fig:partition2}%
597: \end{figure}
598: 
599: 
600: \pagebreak[4]
601: 
602: 
603: \begin{figure}[ptb]
604: \centerline{a)}
605: \includegraphics[angle=-90, width=0.75\textwidth]{chaosig_2.ps}%
606: 
607: \centerline{b)}
608: \includegraphics[angle=-90, width=0.75\textwidth]{hmarkovsig.ps}%
609: \caption{
610: %Comparative plot of: a) the chaotic model symbolic dynamics using 
611: %the partition  
612: %$\Pi^*$ and b) the symbol sequence simulated on the basis of the 
613: %transition-rate matrix 
614: %$W^{*}$.  
615: }
616: \label{fig:chaos_sig}%
617: \end{figure}
618: 
619: 
620: \pagebreak[4]
621: 
622: \begin{figure}[ptb]
623: \includegraphics[angle=-90, width=1.0\textwidth]{newcorr.ps}%
624: \caption{
625: %Autocorrelation functions for the states $1$ and $2$ of the 
626: %$\Pi^{*}$
627: % partition, see Figure~\ref{fig:partition2}: actual autocorrelations from
628: %the chaotic model (full lines) and from the Markovian approximation 
629: % generated with $W^{*}$ (dotted lines).  
630: }
631: \label{fig:chaos_corr_hmarkov}%
632: \end{figure}
633: 
634: \pagebreak[4]
635: 
636: \begin{figure}[ptb]
637: \includegraphics[angle=-90, width=1.0\textwidth]{shannon.ps}%
638: \caption{
639: %Differential Block Entropy $h_n = H_{n+1}-H_{n}$, as a function 
640: %of the sequence-length $n$, for 
641: %the $\Pi$ ($\star$) and the $\Pi^*$ ($\square$) partitions. 
642: %For each length $n$ approximately 
643: % $\sim 10^7$ sequences were used.
644: %Notice the very different rate of convergence of $h_n$ to its limit 
645: %value (Shannon entropy) $h_S$ in the two cases. 
646: }
647: \label{fig:shannon}%
648: \end{figure}
649: 
650: \end{document}
651: