1: \documentclass{aa}
2: \usepackage{times}
3: \usepackage{graphics}
4: \begin{document}
5:
6: \thesaurus{2(02.03.1), 03.13.4, 05.03.1 }
7: \title{The power spectrum of geodesic divergences \\
8: as an early detector of chaotic motion}
9: \author{Ch.L. Vozikis, H. Varvoglis and K. Tsiganis}
10: \offprints{Ch.L. Vozikis}
11: \mail{chriss@astro.auth.gr}
12: \institute{Aristotle University of Thessaloniki, Department of
13: Physics, Section of Astrophysics, Astronomy and Mechanics,
14: \\ 54006 Thessaloniki, Greece}
15: \date{Received ......, 1999 / Accepted ......, 2000}
16:
17: \titlerunning{PSOD -- a new chaos detector}
18: \authorrunning{Vozikis et al.}
19: \maketitle
20:
21: \begin{abstract}
22: We propose a new method for determining the stochastic
23: or ordered nature of trajectories in non-integrable Hamiltonian dynamical systems.
24: The method consists of constructing a time-series from the divergence
25: of nearby trajectories
26: and then performing a power spectrum analysis of the series. Ordered
27: trajectories
28: present a spectrum that consists of a few spikes
29: while the spectrum of stochastic trajectories is continuous. A test of
30: the method with three
31: different systems, a 2-D mapping as well as a 2-D and a 3-D Hamiltonian,
32: shows that the method is fast and efficient, even in the case
33: of sticky trajectories.
34: The method is also applied to the motion of asteroids in the Solar
35: System.
36:
37:
38: \end{abstract}
39:
40:
41: \keywords{Chaos -- Methods: numerical -- Celestial Mechanics, stellar
42: dynamics}
43:
44: \section{Introduction}
45:
46: The problem of distinguishing a chaotic from an ordered trajectory in
47: a non-integrable Hamiltonian system has been a topic of active
48: investigation since the pioneering work of H\'{e}non and Heiles (1964).
49: Initially, when the study was restricted to 2-D
50: systems, the work was done through
51: surface of section plots. Later, when systems with more than
52: 2-D were considered, the method of choice was the calculation of the
53: Lyapunov Characteristic Numbers (LCNs) (Benettin et al. 1976,
54: Froeschl\'e 1984). Unfortunately both the above methods suffer from
55: the same drawback, namely they are not able to distinguish easily an
56: ordered from a ``sticky'' chaotic trajectory. Various methods have been
57: devised since then to address the above problem, namely the
58: distinction of an ordered from a chaotic trajectory using a relatively
59: short-time trajectory segment.
60:
61: These methods generally fall into two main classes: those that use
62: frequency or correlation analysis of a time-series, constructed by the
63: values of generalized co-ordinates (or
64: functions of them), and those that
65: use the geodesic
66: divergence of initially nearby trajectories. In the first
67: class belong the ``old'' method of the rotation
68: number (Contopoulos 1966),
69: the frequency map analysis developed by Laskar (Laskar et al. 1992,
70: Laskar 1993) and the power spectrum analysis of
71: quasi-integrals developed by
72: Voyatzis \& Ichtiaroglou (1992).
73: In the second belong the probability density function analysis of
74: stretching numbers
75: developed by Froeschl\'e et al. (1993) and Voglis \& Contopoulos
76: (1994) and
77: the Fast Lyapunov Indicators method developed by
78: Froeschl\'e et al. (1997). Each one of the above methods has its own
79: advantages and weaknesses; in particular some of them are more
80: suitable to test {\it large sets of trajectories}
81: rather than {\it single ones}, some are more efficient for {\it 2-D
82: systems} rather than for {\it N($>$2)-D systems}
83: and some perform better for {\it mappings} rather than {\it flows}.
84:
85: In 1997, Contopoulos
86: \& Voglis introduced a new method for
87: distinguishing chaotic from
88: ordered trajectories, which does not belong to any of the above
89: mentioned two classes but,
90: instead, may be classified as ``mixed''. This new method is based on the
91: analysis of the probability
92: density of {\it helicity} and {\it twist angles}. Voglis \&
93: Efthymiopoulos (1998) and subsequently Froeschl\'e \& Lega (1998) showed
94: that the twist angles method is very efficient in testing {\it phase space
95: regions}, at least in cases of
96: 2-D systems, where the twist angles can be easily calculated.
97: More recently Voglis et al. (1998, 1999) proposed two new and very
98: efficient methods, namely the method of ``Dynamical Spectral
99: Distance'' (DSD), which is particularly suitable for the characterization
100: of single trajectories in 4-D maps, and the method of ``Rotational Tori
101: Recognizer'' (ROTOR) which is very efficient for testing wide areas of the
102: phase space of 2-D maps.
103:
104: In the present paper we are introducing a new ``mixed" method, which we
105: show that it is at least as
106: sensitive as the other methods in the literature, may be applied
107: in a straightforward way to
108: dynamical systems
109: with more than two degrees of freedom and is equally efficient for single
110: trajectories as well as large sets of them.
111: The method consists in analyzing a time series
112: constructed by the values of the geodesic deviation of nearby
113: trajectories recorded at a properly selected frequency. It should be
114: noted that a method based on a similar technique has been proposed by
115: Lohinger and Froeschl\'{e} (1993).
116:
117: The paper is organized as follows. Section 2 describes the basic
118: features of the method, which, in Section 3, is tested upon three
119: dynamical systems of different types. A comparison of the results of
120: our method to those derived by various other methods is made in Section
121: 4. Section 5 treats the application of the proposed method to one of
122: the most important problems of solar system dynamics, the motion of
123: asteroids. Finally in Section 6 we present our conclusions.
124:
125:
126: \section{The method}
127:
128: In order to decide on the nature of a trajectory (chaotic or not), we
129: work as follows. We integrate numerically the ``main" trajectory together with
130: a nearby one, which at a time $t_0=0$ starts at an infinitesimal
131: distance in phase space, $d_0$, from the main, and we calculate their
132: distance, $d_1$, at a time $t = t_0 + \Delta t$. Let us denote by $q$ the
133: logarithm of the ratio of the two distances, $q = \ln (d_1/d_0)$.
134: We then renormalize the
135: nearby trajectory, so as to start from a new position in phase space, which
136: is at distance $d_0$ from the main trajectory in the direction of
137: $d_1$. The trajectories are
138: followed once again for a time interval $\Delta t$ and a new $q$ is
139: calculated. After following the trajectories for a time interval $T =
140: N~\Delta t$, we have constructed a time series consisting of the
141: consecutive $q$'s
142: \begin{equation}
143: q(t) = \ln \left[{d_t \over d_0}\right]_t~~~ {\rm or}~~~
144: q_k = \ln \left[{d_k \over d_0}\right]_{t_k}
145: \label{eq_qk}
146: \end{equation}
147:
148: An ordered trajectory of a N-D conservative dynamical system will lie on an
149: invariant torus, i.e. a N-D
150: manifold of the 2N-D phase space. Any such trajectory is, in general,
151: quasi-periodic and covers densely the
152: invariant set. If a nearby trajectory is started at an infinitesimal
153: distance from the previous one, this too would, in general, lie on an
154: invariant torus, so that the $q(t)$ time series should behave in a
155: quasi-periodic manner. On the other hand, a chaotic trajectory visits
156: different regions of phase space in a stochastic manner and $q(t)$ should also
157: be ``random''.
158:
159: Following the above considerations we calculate the {\it power
160: spectrum} $P(f)$ of the $q(t)$ time series. We first calculate
161: the discrete Fourier transform of the $q_k$ multiplied by a {\it
162: window function} $w_k$
163: \begin{equation}
164: Q_j=\sum_{k=0}^{N-1}q_k w_k e^{2\pi i j k /N} ~~~~~j=0..N-1
165: \end{equation}
166: Then the power spectrum, $P(f)$, is defined in $M=N/2+1$ frequencies as
167: \begin{eqnarray}
168: P(f_0)&=& \frac{1}{W} ~ \vert Q_0 \vert^2 \nonumber \\
169: P(f_j)&=& \frac{1}{W} ~ \left( \vert Q_j \vert^2 + \vert Q_{N-j} \vert^2
170: \right )~~~~~ j=1..(\frac{N}{2}-1) \\
171: P(f_c) &=& \frac{1}{W}~ \vert Q_{N/2} \vert^2 \nonumber
172: \end{eqnarray}
173: where
174: \begin{equation}
175: W = N \sum_{k=0}^{N-1}w_k^2
176: \end{equation}
177: and $f_c = f_{N/2}$ is the Nyquist frequency defined as
178: \begin{equation}
179: f_c=\frac{1}{2\Delta t}.
180: \end{equation}
181: The frequencies covered by the power spectrum are
182: \begin{equation}
183: f_j=f_c~\frac{j}{M}~~~~~~~j=0..M
184: \label{eq_freqs}
185: \end{equation}
186: In the present work we used the so called ``Hanning'' window. More details
187: on the calculation of the
188: power spectrum can be found in the book by Press et. al. (1992).
189:
190: \begin{figure}
191: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f01a.ps}}
192: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f01b.ps}}}
193: \vskip 0.1cm
194: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f01c.ps}}
195: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f01d.ps}}}
196: \caption{The successive points of the ``sticky'' trajectory (starting at
197: $J_0 = \pi $, $\theta_0 = 1.538~\pi$). Upper left, the first
198: 2048 points, upper right the first 4096, down left, the first 6144
199: and down right, the first 8192. The units in all axes are given as
200: multiples of $\pi$ }
201: \label{map_invar}
202: \end{figure}
203:
204: As a rule ``mixed" methods are expected to perform better in distinguishing
205: between ordered and chaotic trajectories. The reason for that
206: is that time series that are constructed from the geodesic divergence of
207: nearby orbits contain all the various characteristic frequencies that
208: locally affect the motion in the ``proper" ratio, i.e. the frequencies
209: corresponding to the different directions (degrees of freedom) are
210: properly weighted.In particular now, as far as our method is concerned,
211: the Power Spectrum Of Divergences (PSOD) of an ordered
212: trajectory is expected to posses only a few ``spikes" at specific
213: frequencies. The number of harmonics however depends on the system under
214: consideration as well as on the values of its ``controll'' parameters
215: (see below).
216: In contrast, the PSOD of a chaotic trajectory should appear
217: continuous, due to the random nature of the $q_k$ time series.
218: However, the above considerations lie behind all methods based on time
219: series analysis. What is really important, for the assessment of the
220: new method with respect to the
221: other ones appearing in the literature, is to evaluate (a) its
222: independence from the number
223: of degrees of freedom of the dynamical system, (b) its effectiveness
224: with respect to the object of
225: test (single trajectories or distributions of initial conditions
226: covering wide phase space regions) (c) its
227: sensitivity, i.e. the minimum length of the time series, necessary to
228: distinguish a sticky chaotic trajectory from an ordered one and (d) its
229: ability to produce a well-defined measure of chaos (as is the LCN).
230:
231: \begin{figure}
232: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f02a.ps}}
233: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f02b.ps}}}
234: \vskip 0.1cm
235: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f02c.ps}}
236: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f02d.ps}}}
237: \caption{The PSOD for the three trajectories of the 2-D mapping and
238: for N=1024 iterations.
239: Upper left : {\it map1} trajectory,
240: upper right : {\it map2} trajectory, down left : {\it map3} trajectory and
241: down right : {\it map3} trajectory but now with N=256 }
242: \label{map_PSOD}
243: \end{figure}
244:
245: \section{Evaluation of the method}
246:
247: We proceed in the assessment of the method using three different
248: dynamical systems, namely a
249: 2-D mapping as well as a 2-D and a 3-D Hamiltonian system. In each one
250: we evaluated the nature of a considerable number of trajectories. In the
251: following subsections
252: we present only three or four trajectories per system, which we think that
253: are typical examples of
254: the three different classes of trajectories, i.e. ordered, clearly
255: chaotic and sticky. The amplitudes of the
256: PSOD, in all figures, are normalized so that the highest has the value
257: one, while the frequency is given in cycles per time unit.
258:
259: \subsection{2-D mapping}
260:
261: \begin{figure}
262: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f03.ps}}}
263: \caption{The evolution of $\chi(t)$ for the sticky trajectory in the 2-D mapping }
264: \label{map_LCN}
265: \end{figure}
266:
267: We first test the method in the simple 2-D mapping
268: \begin{eqnarray}
269: J_{i+1} & = & J_i + k \cos(2 \theta_i)~~~\mathrm{mod}(2\pi) \nonumber\\
270: \theta_{i+1} & = & \theta_i + J_{i+1} ~~~~~~~~~~~~\mathrm{mod}(2\pi)
271: \end{eqnarray}
272: where the stochasticity parameter $k$ is taken equal to 0.7.
273:
274: We present here the
275: results of three trajectories, an ordered one ({\it map1}) starting at
276: $J_0 = \pi $, $\theta_0 = 1.4~\pi$,
277: a stochastic one ({\it map2}) starting at
278: $J_0 = \pi $, $\theta_0 = 1.5~\pi$ and a sticky one ({\it map3})
279: starting at
280: $J_0 = \pi $, $\theta_0 = 1.538~\pi$. The initial conditions of the
281: third one place it very close to the boundary
282: between the ordered region, surrounding the stable point
283: $J_0 = \pi $, $\theta_0 = 5\pi/4$, and the chaotic sea. Figure
284: \ref{map_invar}
285: shows the consequents ($\theta$, $J$) of the ``sticky'' trajectory at
286: various times (iterations).
287: As we can see, for up to 2\,048 iterations the trajectory
288: behaves like an ordered one. Around $i=4\,000$ it starts to
289: present some signs of irregularity and finally,
290: after $i=6\,000$, the chaotic nature of the trajectory becomes evident.
291: Using our method on these three trajectories, with the length of the
292: $q_k$ time
293: series being N=1024 iterations and defining $\Delta t = 1$, we obtain the
294: spectra shown in Fig.~\ref{map_PSOD}.
295:
296: The upper left frame of Fig.~\ref{map_PSOD} is the spectrum of the ordered
297: trajectory.
298: The spectrum consists of some basic frequencies while the
299: ``noise'' is at a very low level. On the contrary, the spectrum
300: of the chaotic trajectory, in the upper right frame, covers the whole
301: frequency range with comparable amplitudes,
302: i.e. a clearly continuous spectrum. Looking at the spectrum of
303: the ``sticky'' trajectory (lower left frame), we see a pattern almost
304: the same as that of the chaotic one. Some
305: high-amplitude spikes are evident
306: but the continuum noise level is again very high for the whole frequency
307: range, denoting the stochastic nature of the trajectory. Even with much less
308: iterations (N=256 - lower right frame of Fig.~\ref{map_PSOD}) we get
309: the same result. The spectrum is less dense, since it has a smaller
310: number of frequencies than before (see eq.(\ref{eq_freqs})), but the
311: basic features are the same. Note that for the LCN, which is the
312: limit of the function
313: \begin{equation}
314: \chi(t)= \frac{1}{N~\Delta t}\sum_{i=1}^{N} q_i
315: \end{equation}
316: as $N \rightarrow \infty$ (see Fig.~\ref{map_LCN}), or the plot
317: of the consequents of the mapping (Fig.~\ref{map_invar}), we need much
318: more than 1024 iterations in order to decide whether the trajectory
319: is chaotic or not.
320:
321: \subsection{An improved criterion}
322:
323: \begin{figure}
324: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f04a.ps}}
325: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f04b.ps}}}
326: \vskip 0.1cm
327: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f04c.ps}}
328: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f04d.ps}}}
329: \caption{The PSOD for the three
330: trajectories in the 2-D mapping shown in Fig.~\ref{map_PSOD}, the peaks
331: being sorted in descending order of amplitude.
332: The first three frames are for N=1024 iterations, while the fouth (
333: down right ) is {\it map3} trajectory with N=256 iterations.}
334: \label{map_PSOD_sort}
335: \end{figure}
336:
337: The above presented results show that it is, indeed, worth to consider
338: the new method as a useful tool in the assessment of the nature,
339: ordered or chaotic, of a trajectory. However the method, as it is,
340: does not entail a {\it clear} and {\it easy to apply} criterion for the
341: classification of a trajectory as ordered or chaotic. Here we try to
342: improve somehow the presentation of the results of our method, in order
343: to propose such a criterion. Note that this new criterion is
344: similar, graphically, to the criterion of the
345: FLI proposed by Froeschl\'e et al (1997).
346:
347: If the peaks appearing in the PSOD are plotted in descending order of
348: amplitude, we have a graphical representation of how many strong
349: frequencies the spectrum possesses. Figure~\ref{map_PSOD_sort} shows this
350: representation of the PSOD for the three
351: trajectories studied in the previous sub-section. Ordered trajectories
352: have only a few high amplitude frequencies
353: and the background is formed by peaks whose amplitudes are more than four
354: orders of magnitude smaller than that of the basic frequency.
355: On the other hand the stochastic trajectories
356: possess only a few high-amplitude frequencies and the largest part of the
357: spectrum consists of a ``continuum'' of frequencies with also considerable
358: amplitude. In this respect one may chose to represent the results by a signle
359: number (e.g. the number of peaks up to a certain amplitude), provided that
360: a certain ``threshold'' for the noise level is chosen (10$^{-8}$ in Fig.
361: 4). This value will, however, depend on the system at study.
362:
363: \subsection{Noise level}
364: \label{section_noise}
365:
366: \begin{figure}
367: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f05a.ps}}
368: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f05b.ps}}}
369: \vskip 0.1cm
370: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f05c.ps}}
371: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f05d.ps}}}
372: \vskip 0.1cm
373: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f05e.ps}}
374: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f05f.ps}}}
375: \caption{The PSOD for the regular (left) and the chaotic (right)
376: trajectories in the 2-D mapping with N=256 (top), N=4096 (middle) and
377: N=65\,536 (bottom).}
378: \label{noise1}
379: \end{figure}
380:
381:
382: \begin{figure}
383: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f06a.ps}}
384: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f06b.ps}}}
385: \caption{The sorted PSOD for the regular (left) and the chaotic (right)
386: trajectories in the 2-D mapping for different iteration number, N. Lines
387: from top to bottom correspond to N=256, 1024, 4096, 16\,384 and 65\,536. }
388: \label{noise2}
389: \end{figure}
390:
391: Any method of analysis of finite-sample time series is bound to suffer
392: from noise. Thus, even in the PSOD of a regular orbit a certain noise
393: level is expected. This is mainly due to ``leakage'' of power from
394: the frequency lobes, a side effect of the calculation of the power
395: spectrum using Discrete Fast Fourier Transform methods. When calculating
396: the power spectrum of a ``monochromatic'' signal, the power contained in
397: its basic frequency ``leaks'' into neighboring frequencies. The leakage
398: depends on the windowing function used. When a signal possesses two
399: closely spaced frequencies, all frequencies in between these two will
400: also gain considerable amplitudes, due to this phenomenon.
401: If the PSOD of a regular orbit has a large number of basic (strong)
402: frequencies, then this effect can lead us to falsely identify it as chaotic.
403: The problem can be tackled at the cost of taking more points in the sample.
404: In this way, the number of frequencies appearing in the spectrum is
405: increased but the frequency lobes become thinner. Thus, lobe overlapping is
406: reduced and the noise level drops.
407: For chaotic orbits on the other hand, the observed noisy pattern
408: is an inherent property of the spectrum and by increasing the number of
409: points one cannot alter the picture.
410:
411: The above can be seen in Fig.~\ref{noise1}, which shows the PSOD of a
412: regular (left column) and a
413: chaotic trajectory (right column) for three different values of N, i.e.
414: N=256 (top), N=4096 (middle) and N=65\,536 (bottom). As we see, the noise
415: level in the case of the regular orbit drops significantly when N
416: is increased, while in the case of the chaotic orbit it remains more or
417: less unchanged. In the N=65\,536 case the noise level for the ordered
418: trajectory drops below $10^{-20}$, becoming comparable to the
419: accuracy of the FFT calculation (double precision).
420:
421: This phenomenon is better seen if we use the amplitude-sorted
422: PSOD. Fig.~\ref{noise2} shows the amplitude-sorted PSOD of the regular
423: (left) and chaotic (right) trajectory with N=256, 1024, 4096, 16\,384 and
424: 65\,536. While the noise level in the PSOD of the ordered trajectory is
425: reduced when N is increased, it remains the same for the chaotic one.
426:
427: \subsection{2-D Hamiltonian system}
428:
429: \begin{figure}
430: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f07.ps}}}
431: \caption{The evolution of $\chi(t)$ of the four test trajectories in
432: the 2-D Hamiltonian model in the case $a=2$. The initial positions of
433: the trajectories are given in the text. Note that the $\chi(t)$ curves
434: of the two ordered trajectories have been lowered by one ({\it ord-2})
435: and half ({\it ord-2a)} orders of
436: magnitude in order to be clearly distinguished from the sticky trajectory}
437: \label{2d_a2_LCN}
438: \end{figure}
439:
440: \begin{figure}
441: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f08a.ps}}
442: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f08b.ps}}}
443: \vskip 0.1cm
444: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f08c.ps}}
445: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f08d.ps}}}
446: \caption{The PSOD, with N=2048, of the four test trajectories in the
447: 2-D Hamiltonian model for $a$ = 2. Upper left: the ordered {\it
448: ord-2}, upper right: the chaotic {\it ch-2}, lower left: the
449: sticky {\it st-2} and lower right: the ordered {\it ord-2a}. }
450: \label{2d_a2_PSOD}
451: \end{figure}
452:
453: \begin{figure}
454: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f09a.ps}}
455: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f09b.ps}}}
456: \vskip 0.1cm
457: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f09c.ps}}
458: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f09d.ps}}}
459: \caption{The surface of section $x$, $\dot x$ of {\it st-4} in
460: the 2-D Hamiltonian model. Upper left $t=3\,000$, upper right $t=8\,000$,
461: down left $t=15\,000$, down right $t=20\,000$}
462: \label{2d_inv}
463: \end{figure}
464:
465: \begin{figure}
466: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f10a.ps}}
467: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f10b.ps}}}
468: \vskip 0.1cm
469: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f10c.ps}}
470: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f10d.ps}}}
471: \caption{The PSOD, with N=4096, of the four test trajectories in the
472: 2-D Hamiltonian model for $a$ = 4. Upper left: the ordered {\it
473: ord-4}, upper right: the chaotic {\it ch-4} and lower left: the
474: sticky {\it st-4} and lower right: the ordered {\it
475: ord-4a}. }
476: \label{2d_PSOD}
477: \end{figure}
478:
479: \begin{figure}
480: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f11a.ps}}
481: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f11b.ps}}}
482: \vskip 0.1cm
483: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f11c.ps}}
484: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f11d.ps}}}
485: \caption{The sorted PSOD of Fig.~\ref{2d_PSOD} }
486: \label{2d_PSOD_sort}
487: \end{figure}
488:
489: We now proceed to test the method in a 2-D Hamiltonian system.
490: We selected the Hamiltonian used by Caranicolas \& Vozikis (1987),
491: \begin{equation}
492: H = {1 \over 2} \left( p_x^2 + p_y^2 \right) + x^4 + y^4 + 2
493: a x^2y^2 = h
494: \end{equation}
495: where the parameter $a$ is taken equal to 2.0 and the energy constant
496: $h=1.0$.
497:
498: Again we study three trajectories, one ordered ({\it ord-2}) starting at
499: $x(0) = 0.5$, one clearly stochastic ({\it ch-2}) starting at $x(0) = 0.001$
500: and one ``sticky'' ({\it st-2}) which starts at $x(0) = 0.1326$.
501: All three trajectories have also $y(0) = 0$ and $p_x(0) = 0$, while their
502: $p_y(0)$ is given by the energy integral.
503:
504: Using the PSOD with a renormalization time-step equal to $\Delta t = 1.5$,
505: we find that the spectra of the three test trajectories
506: present the same properties as those of the corresponding cases of the
507: mapping. In Fig.~\ref{2d_a2_PSOD} we see the PSOD of the three test
508: trajectories. The difference between the spectrum of the
509: chaotic trajectory {\it ch-2} and that of the ordered trajectory {\it ord-2}
510: is again obvious. Moreover, the ``sticky'' trajectory {\it st-2}
511: has a spectrum similar to that of {\it ch-2}. Note that we have used only
512: 2048 points (corresponding to $t= 3072$) while, if we look at the
513: evolution of the $\chi(t)$ (Fig.~\ref{2d_a2_LCN}), the trajectory
514: looks ordered for a time up to $t\approx 7\,000$. The fourth frame
515: (lower right) in Fig.~\ref{2d_a2_PSOD} corresponds to another regular orbit
516: ({\it ord-2a}) that starts at $x= 0.133$, i.e. very close to the sticky
517: trajectory. We see that, although the two trajectories start very
518: close to each other and, at least for the first 3\,000 times steps,
519: span approximately the same phase-space region, their PSODs are completely
520: different, clearly revealing the nature of each case.
521:
522: We decided to test also the case where the parameter $a$ is taken
523: equal to $4.0$. As Caranicolas \& Vozikis (1987) have shown, the
524: surface of section of this case has a completely different topology
525: from that of the $a=2$ case. The equipotential curves have negative
526: curvature along the $y=\pm x$ lines. This affects mainly the loop
527: orbits which appear ``squared''.
528:
529: Again we study four trajectories, one ordered ({\it ord-4}) starting at
530: $x(0) = 0.3$, one clearly chaotic ({\it ch-4}) starting at $x(0) = 0.7$,
531: one ``sticky'' ({\it st-4}) which starts at $x(0) = 0.36282$ and one
532: ordered ({\it ord-4a}) starting at $x(0)= 0.362$ very near to the sticky
533: one. The surface of section plot for the ``sticky'' trajectory is shown in
534: Fig.~\ref{2d_inv} at various times. Figures~\ref{2d_PSOD} and
535: \ref{2d_PSOD_sort} present the PSOD and the amplitude-sorted PSOD for
536: these four orbits (N=4096). An important characteristic seen in these two
537: figures is the presence of a high number of medium-amplitude frequencies
538: in the spectra of the regular orbits. However, a distinction between regular
539: and chaotic orbits can still be made. Due to frequency overlapping,
540: discussed in section \ref{section_noise}, individual frequencies cannot
541: be distinguished at an amplitude level smaller than $10^{-6}$. This makes
542: it very difficult to identify a sticky orbit with an amplitude level
543: around $10^{-6}-10^{-7}$. Nevertheless the noise level of the regular
544: orbits will be supressed if we take more points, while for a sticky
545: chaotic orbit it will remain more or less the same.
546:
547: \subsection{3-D Hamiltonian system}
548:
549: We apply our method to the model Hamiltonian used by
550: Magnenat (1982), Contopoulos \& Barbanis (1989),
551: Barbanis and Contopoulos (1995), Barbanis (1996), Varvoglis et
552: al. (1997), Barbanis et. al. (1999) and Tsiganis et al. (2000a)
553: \begin{eqnarray}
554: H &=& {1 \over 2} \left( p_x^2 + p_y^2 + p_z^2 \right) \nonumber \\
555: &&+ {1 \over 2} \left( A x^2 + B y^2 + C z^2 \right) - \epsilon x z^2 -
556: \eta y z^2 = h
557: \end{eqnarray}
558: where the parameters are taken as $A = 0.9$, $B = 0.4$, $C =
559: 0.225$, $\epsilon = 0.560$ and $\eta = 0.20$ and the energy
560: level is $h = 0.00765$
561:
562: \begin{figure}
563: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f12.ps}}}
564: \caption{The evolution of $\chi(t)$ of the four test trajectories in
565: the 3-D Hamiltonian
566: model. The initial positions of the trajectories are given in the text.}
567: \label{3d1}
568: \end{figure}
569:
570: We again test three trajectories, an ordered one ({\it reg01}) starting at
571: $\bar x =0.01$, $\bar y=0.032$, a chaotic one ({\it ch-01}) starting
572: at $\bar x=0$, $\bar y=0$ and a sticky one
573: ({\it st-01}) with initial conditions
574: $\bar x =0.01725$, $\bar y=0.032$, while $\bar z$, $p_x$, $p_y$
575: were taken equal to 0 and $p_z$ is calculated from the energy integral.
576: We use the variables $\bar x$, $\bar y$, $\bar z$ instead of $x$, $y$,
577: $z$ in order to be consistent with the previous publications.
578: The barred variables are defined as $\bar x= \sqrt{A}x$, $\bar y =
579: \sqrt{B} y$ and $\bar z = \sqrt{C} z$.
580:
581: In 3-D one cannot visualize a surface of section plot, in order to check
582: whether a
583: particular trajectory is ordered or chaotic. Therefore, if one is using
584: the traditional tools, he has to rely on the calculation of LCNs. It
585: should be pointed out that a positive LCN is a proof that the trajectory
586: under study is chaotic, while a monotonically decreasing value of $\chi
587: (t)$ is not a proof of order, since this behavior could very well
588: originate from ``stickiness". That is why we decided to test one
589: more trajectory ({\it reg00}) for which we can be almost certain that it is
590: ordered, as it has the same initial position as {\it reg01} but it belongs
591: to an almost integrable case of the model Hamiltonian, $\epsilon=0.01$ and
592: $\eta=0.01$.
593:
594: Figure~\ref{3d1} shows the calculation of the $\chi(t)$ for the four
595: trajectories. We can clearly see
596: that $\chi (t)$ of {\it st-01} is decreasing up to $t=7\,10^4$ and then
597: begins to saturate to a non-zero LCN value. The
598: stochasticity is even more evident after $t=2.2\,10^5$,
599: where we have a ``jump'' to a higher LCN value.
600:
601: \begin{figure}
602: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f13a.ps}}
603: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f13b.ps}}}
604: \vskip 0.1cm
605: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f13c.ps}}
606: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f13d.ps}}}
607: \caption{The PSOD of the four test trajectories in the 3-D Hamiltonian
608: model using 512 points. i.e. 7680 time steps. Upper left the ordered
609: {\it reg01}, upper right the chaotic {\it ch-01}, lower left the
610: sticky {\it st-01} and lower right
611: the ordered {\it reg00}.}
612: \label{3d2}
613: \end{figure}
614:
615: We calculate the PSOD using $\Delta t = 15$, a value approximately
616: equal to the time interval between two consecutive crossings of the
617: $\bar x - \bar y$ plane by the trajectory.
618: Figure~\ref{3d2} shows the PSOD of the four trajectories using
619: 512 points, i.e. for $t=7\,680$.
620: Note once again that we can decide that the ``sticky'' trajectory {\it st-01}
621: is actually stochastic well in advance of the LCN method.
622: The LCN shows the stochastic
623: behaviour of the trajectory only after $t=70\,000$, while with the PSOD we
624: need a modest $t=7\,680$.
625:
626:
627: \begin{figure}
628: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f14a.ps}}
629: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f14b.ps}}}
630: \vskip 0.1cm
631: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f14c.ps}}
632: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f14d.ps}}}
633: \caption{The PSOD of the four test trajectories in the 3-D Hamiltonian
634: model shown in Fig.~\ref{3d2}, the peaks being sorted in descending order
635: of amplitude.}
636: \label{3d3}
637: \end{figure}
638:
639: \section{Comparison with other methods}
640:
641: As we already mentioned in the introduction, three of the most recent
642: methods for distinguishing ordered from chaotic trajectories are the Fast
643: Lyapunov Indicators (FLI) (Froeschl\'e et al. 1997), the ``spectra'' of
644: stretching numbers and/or twist angles (Froeschl\'e et al. 1993, Voglis
645: \& Contopoulos 1994, Contopoulos \& Voglis 1997) and the ``spectral
646: distance'' (DSD) (Voglis et al. 1998, 1999). Out of these three methods
647: the fastest ones are the FLI and the DSD methods. In this
648: section we compare the PSOD with the FLI and the DSD methods by applying
649: it to the test models presented in the above mentioned papers. A
650: discussion concerning the spectra of stretching numbers follows.
651:
652: \begin{figure}
653: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f15a.ps}}
654: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f15b.ps}}}
655: \caption{The PSOD of the two trajectories
656: used by Froeschl\'e et al. (1997) to test the FLI method. In the
657: left frame is the PSOD of the chaotic trajectory ($x(0)=0.001$, $y(0)
658: =0.001$) and in the right frame is the PSOD of the ordered trajectory
659: ($x(0)=1$, $y(0)=0$).}
660: \label{flg-psod}
661: \end{figure}
662:
663: \subsection{Single trajectories}
664: \subsubsection{FLI}
665:
666: In the paper by Froeschl\'e et al. (1997) the authors tested the FLI method
667: on two trajectories of the standard map
668: \begin{eqnarray}
669: x_{i+1}&=& x_i + k \sin(x_i + y_i) ~~ \mathrm{mod}(2\pi) \nonumber \\
670: y_{i+1}&=& x_i + y_i~~~~~~~~~~~~~~~~~~~~~~ \mathrm{mod}(2\pi)
671: \end{eqnarray}
672: with $k=0.3$; one stochastic starting at
673: $x=0.001$, $y=0.001$ and one ordered starting at $x=1$, $y=0$. They
674: found that for the stochastic trajectory the FLI's drop very quickly down
675: to $10^{-20}$ (Fig.~2 in their paper - 200 iterations). On the contrary,
676: the function $\chi (t)$ levels only after about 10\,000
677: iterations. For the ordered trajectory the FLI's are slowly decreasing,
678: following $\chi (t)$. In Fig.~\ref{flg-psod} we
679: present the PSOD's of the stochastic (left frame) and the ordered
680: (right frame) trajectory,
681: calculated using only 256 iterations. It is obvious that the two
682: spectra clearly differentiate between the two types of trajectory.
683:
684: \subsubsection{Spectral distance (the 4D map)}
685: \begin{figure}
686: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f16a.ps}}
687: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f16b.ps}}}
688: \vskip 0.1cm
689: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f16c.ps}}
690: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f16d.ps}}}
691: \caption{The PSOD (upper frames) and the sorted PSOD (lower frames)
692: of the two test trajectories (left frames : regular A2, right frames
693: : chaotic A3) in the 4-D mapping for N=4096.}
694: \label{map4d}
695: \end{figure}
696:
697: In a recent paper Voglis et al. (1999) proposed, as a tool for the
698: distinction between chaotic and regular orbits in 4D maps, the use of the
699: ``spectral distance'' $D^2$.
700: The method is based on the property that the ``spectrum'' of
701: {\it stretching numbers} (as well as that of the {\it helicity angles})
702: of a
703: chaotic trajectory is independent of the initial orientation of the
704: deviation vector, while the spectrum of a regular trajectory is not.
705:
706: The ``spectral distance'' $D^2$ is a norm defined as
707: \begin{equation}
708: D^2 = \sum_q \left[ S_1(q) - S_2(q) \right]^2
709: \end{equation}
710: where the summation is for all $q$'s and $S_1$, $S_2$ are two spectra
711: of the same orbit but with two different initial deviation vectors.
712:
713: Voglis et al. (1999) applied their method to a 4-D mapping
714: consisting of two coupled 2-D standard maps, i.e.
715: \begin{eqnarray}
716: x_1'&=&x_1+x_2' \nonumber\\
717: x_2'&=&x_2+\frac{k}{2\pi}\sin(2\pi
718: x_1)-\frac{\beta}{\pi}\sin(2\pi(x_1-x_3))\\
719: x_3'&=&x_3+x_4'\nonumber \\
720: x_4'&=&x_4+\frac{k}{2\pi}\sin(2\pi
721: x_3)-\frac{\beta}{\pi}\sin(2\pi(x_3-x_1))\nonumber
722: \end{eqnarray}
723: where the $x_i$'s are defined in the interval [0,1) (i.e.$\mathrm{mod} 1$).
724:
725: We chose to test our method upon the two most interesting cases shown in
726: Voglis et al. (1999), namely trajectories A2 and A3, in their notation.
727: The A2 case has initial conditions ($x_1,x_2,
728: x_3,x_4$)=(0.55,0.1,0.62,0.2), $\beta=0.1$ and is a regular orbit, while
729: the A3 case has the same initial $x_i$'s, $\beta=0.3051$ and is a chaotic
730: orbit but with a very low value of LCN (around $4\,10^{-7}$).
731:
732: Fig.\ref{map4d} shows the PSOD of the two test trajectories. The left
733: panel corresponds to the regular orbit (A2) and the right panel
734: corresponds to the chaotic one (A3). The spectra were calculated using 4096
735: iterations for each orbit. The distinction between the regular and the
736: chaotic orbits is apparent. Note that our method gave the same result as the
737: $D^2$ method with almost the same computational effort. Of course, both
738: methods are much faster than the traditional LCN method.
739:
740: \subsubsection{``Spectrum'' of stretching numbers}
741: The ``spectrum'' of stretching numbers, $S(q)$, (Froeschl\'e et
742: al. 1993, Voglis \& Contopoulos 1994, Contopoulos \& Voglis 1997,
743: Dvorak et al. 1998) is a method also based on the divergence of
744: nearby trajectories.
745: It consists of the calculation of the probability density of
746: $q_k$ (eq.\~(\ref{eq_qk})), i.e.
747: \begin{equation}
748: S(q)=\frac{\Delta N(q)}{N dq}
749: \end{equation}
750: where $\Delta N(q)$ is the number of $q_k$ with values between $q$ and
751: $q+dq$. A quasi-periodic trajectory, which lies very close to a
752: periodic trajectory, has a ``U'' shaped distribution of stretching
753: numbers which is also symmetric around $q=0$ (Contopoulos et
754: al. 1997). As we move away from the periodic trajectory, this symmetry is
755: destroyed (Caranicolas \& Vozikis 1999) and the spectrum starts to
756: develop a greater number of maxima. On the
757: other hand if the trajectory is chaotic, the spectra have different shapes
758: and are not symmetric at all.
759: It should be pointed out, however, that, in order to obtain a well defined spectrum, one
760: needs to account for a large number of iterations (typically $N=10^5$ or
761: more). Therefore, we did not attempt to compare this method to our own.
762:
763: \subsection{Sets of trajectories}
764:
765: \begin{figure}
766: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f17.ps}}}
767: \caption{The logarithm of the average amplitude of PSOD in a cross
768: section of 1000 trajectories near the 1/6 resonance in
769: the standard map of eq.(6)}
770: \label{helios}
771: \end{figure}
772:
773: In order to circumvent the problem of calculating a trajectory for long times,
774: Contopoulos and Voglis (1997) proposed the use of the average value of $q$,
775: \begin{equation}
776: <q>=\frac{1}{N} \sum_{k=1}^N q_k
777: \end{equation}
778: If we keep N small, we can scan a wide area of initial conditions and map
779: its dynamical behavior. Trajectories in chaotic domains will have $<q>$
780: scattered around the value of the LCN of this domain, while trajectories
781: in the ordered domain will have $<q>$ near zero. The method is very fast
782: in distinguishing between ordered and chaotic domains. Fig.~9 of
783: Contopoulos \& Voglis (1997) is a very good example of the results obtained
784: by this method with only 10 iterations.
785:
786: However, although the method is very good in scanning wide areas of
787: phase space for locating islands of order, it cannot give reliable
788: results with so few iterations for a particular trajectory. In the case
789: of stochastic trajectories, $<q>$ for $N=10$ varies so much, that it may
790: yield a number as small as the one given for ordered trajectories. The
791: situation is even worse in the case of
792: sticky trajectories (i.e. in the borders of islands). In order to
793: decide on the character of such trajectories one needs considerably longer
794: calculations.
795:
796: Froeschl\'e \& Lega (1998) tested the FLI method along with the
797: method of twist angles (Contopoulos \& Voglis 1997), the
798: frequency map analysis method (Laskar et al. 1992, Laskar
799: 1993) and the sup-map method (Laskar 1990,
800: Froeschl\'e \& Lega 1996). Figs.~8a-d of their paper shows the results
801: of the four methods on a cross
802: section of 1\,000 trajectories near the hyperbolic point of the 1/6 resonance
803: of the standard map (Eq. 11) with $k=-1.3$. For the FLI method they
804: used 2\,000 iterations while for the other three methods 20\,000 iterations.
805: In order to produce unambiguous results, the trajectories in this test
806: are classified as ordered or chaotic by an appropriately selected
807: number/indicator. For the FLI method the authors have used as an indicator
808: the time necessary for the FLI to reach a value lower than $10^{-10}$.
809:
810: In order to compare our method with these results we need also an one-number
811: indicator derived from the PSOD. As such we have selected here the
812: average value of $<P/P_{max}>$. The averaging is performed not over
813: all values but by ignoring the highest 1/6th and the lowest 1/6th of the
814: amplitudes, the former being probably due to periodicities in sticky
815: trajectories and the latter probably coming from numerical errors in the
816: integration and the calculation of the power spectrum.
817: Fig.~\ref{helios} shows the same cross section with Fig.~8a-d of
818: Froeschl\'e \& Lega (1998), using the above indicator taken after
819: 8\,192 iterations. As we can see it gives essentially
820: the same information as the other four methods.
821: Note that the y-axis in Fig.~\ref{helios} is inverted for
822: easier comparison with Fig.~8a-d of Froeschl\'e \& Lega (1998).
823:
824: \section{Application to asteroidal motion}
825:
826: \begin{figure}
827: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f18a.ps}}
828: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f18b.ps}}}
829: \vskip 0.1cm
830: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f18c.ps}}
831: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f18d.ps}}}
832: \caption{The evolution of $\chi(t)$ of the trajectories of four
833: asteroids. Upper left, an
834: ordered trajectory with initial $a=3$, $e=0$; Upper right a highly
835: stochastic
836: trajectory with $a=4$, $e=0.2$; Lower left, a stochastic trajectory with
837: stickiness and with initial
838: $a=3.64$, $e=0.08$; Lower right, another sticky orbit with initial
839: $a=3.63$, $e=0.08$}
840: \label{3bp-lcn}
841: \end{figure}
842:
843: \begin{figure}
844: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f19a.ps}}
845: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f19b.ps}}}
846: \vskip 0.1cm
847: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f19c.ps}}
848: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f19d.ps}}}
849: \caption{The PSOD of the four asteroidal trajectories with N=512,
850: i.e. 6062 years}
851: \label{3bp-psod}
852: \end{figure}
853:
854: \begin{figure}
855: \resizebox{\hsize}{!}{\rotatebox{270}{\includegraphics*{9135f20a.ps}}
856: \hspace{1cm} \rotatebox{270}{\includegraphics*{9135f20b.ps}}}
857: \caption{The PSOD, with N=8192, of the ordered asteroidal trajectory
858: (left) and the chaotic Helga clone with $a=3.63$, $e=0.08$ (right)}
859: \label{3bp-psod2}
860: \end{figure}
861:
862: As an application of our method to a problem of physical importance, we
863: shall use it in order
864: to assess the chaotic or not nature of asteroidal trajectories. We use
865: a simplified model of the
866: solar system,
867: namely the planar restricted three body problem where the
868: Sun and Jupiter move on elliptic trajectories around their center of
869: mass and an asteroid of
870: infinitesimal mass moves
871: in the gravitational field of the two bodies. We calculate the PSOD
872: and the LCN using as
873: renormalization time, $\Delta t$, the
874: period of Jupiter, i.e. $\Delta t = T_J \simeq 11.86$ years.
875:
876: Fig.~\ref{3bp-lcn} shows the $\chi (t)$ of the four trajectories
877: tested. The upper
878: left frame belongs to an ordered trajectory starting with semi-major
879: axis $a=3$ and
880: eccentricity $e=0$,
881: the upper right to a stochastic trajectory with initial elements $a=4$,
882: $e=0.2$. The other two frames correspond to trajectories with initial
883: $a=3.64$, $e=0.08$ (lower
884: left) and $a=3.63$, $e=0.08$ (lower right) representing ``clones" of
885: the asteroid
886: 522 - Helga, which is a well-known example of ``stable chaos'' (Milani
887: \& Nobili 1992).
888:
889: The PSOD of the four trajectories (N=512) is shown in Fig.~\ref{3bp-psod}.
890: As we can see, a few Jovian periods are sufficient to decide if the
891: motion of a particular
892: asteroid is stochastic or ordered. In the case of the Helga clone
893: with $a=3.64$ (lower left of Fig.~\ref{3bp-psod})
894: the PSOD shows clearly a chaotic nature after only 512 Jovian periods
895: i.e. 6\,072 years, while even a rough calculation of the LCN
896: needs at least $10^5$ years (see lower right frame of
897: Fig.~\ref{3bp-lcn}). The PSOD of the $a=3.63$ Helga clone is rather
898: peculiar. Although it differs from that of the ordered orbit (upper
899: left), it is not clearly chaotic, unlike the PSOD of the other Helga
900: clone (lower left). Nevertheless, if we take more points in our time series
901: the chaotic nature of the orbit becomes apparent, as we can see in
902: Fig.~\ref{3bp-psod2}. However, the spectrum can still be described as
903: having a strong quasi-periodic component, something which is related to
904: the peculiar dynamical nature of this orbit as discussed in Tsiganis et al.
905: (2000b).
906:
907: \section{Conclusions -- Discussion}
908:
909: In the present paper we propose an alternative tool, which we call PSOD,
910: for the characterization of the chaotic or not nature of
911: trajectories in conservative dynamical systems. The method is based
912: on the frequency analysis of a time series, constructed by successive
913: records of the amplitude of the deviation vector of nearby
914: trajectories.
915: As discussed in Section 2, such a ``mixed"\footnote{Since the method uses
916: both the deviation vectors and frequency analysis it may be classified as
917: ``mixed"} method
918: is expected to have certain advantages. The reason is that the power
919: spectrum of such a time series will contain all the characteristic
920: frequencies of the motion in a properly ``weighted" ratio.
921:
922: The basic characteristic of the PSOD, seen in all three test cases (a 2-D
923: mapping, a 2-D and a 3-D Hamiltonian system), is that
924: \begin{itemize}
925: \item for ordered trajectories the spetrum possesses only a few
926: high-amplitude peaks, the exact number of which depends not only on
927: the system but also on the particular orbit. Of course a small
928: amplitude noise level, due to the numerical procedure, is superimposed
929: on the spectrum, which diminishes as the length of the time series is
930: increased;
931: \item for chaotic trajectories the spectrum has a noisy pattern.
932: For weakly chaotic orbits a few high-amplitude peaks are also present.
933: Increasing the length of the time series, the spectrum tends to
934: a white noise spectrum which remains
935: practically unchanged for any (large \- enough) number of points (see Figs.
936: 5 and 6).
937: \end{itemize}
938:
939: Like most methods existing in the literature, it seems that the method
940: performs better for maps than for flows. However, the results found for
941: the three Hamiltonian flows tested (including the three-boody problem)
942: show that the method can be applied to any system, no matter how many
943: degrees of freedom are involved. We believe that the results may be
944: significantly improved for flows, provided that proper analysis concerning
945: the renormalization time is conducted. The difference between maps
946: and flows is that, in the former case, isochronous records of dynamical
947: quantities also mean studying the system on a surface of section. This is
948: not the case for flows and, thus, a more refined analysis on how to
949: select a proper renormalization time has to be made.
950:
951: We have shown that the sensitivity of the PSOD in testing single
952: trajectories in maps is comparable to the FLI and the DSD methods.
953: This also makes the PSOD an efficient tool for scanning wide areas
954: of the phase space (see Section 4). For such purposes one would like
955: to have an one-number indicator for measuring chaos. The only uniquely
956: defined measure of chaos is of course the LCN. Any other
957: indicator should be in a one-to-one correspondence with the LCN, in
958: order to give the same information. There is no guarantee that such an
959: indicator can be based on the frequency content of the PSOD, as
960: chaotic orbits with similar LCNs may have a different frequency distribution.
961: Further analysis of the properties of the PSOD has to be performed in
962: order to decide whether such an indicator can be found.
963:
964: \begin{acknowledgements}
965: The authors would like to acknowledge the constructive comments of the
966: anonymous referee.
967: \end{acknowledgements}
968:
969: \begin{thebibliography}{}
970: \bibitem{bar3}
971: Barbanis, B., 1996, In: Proceedings of the 2nd Hellenic Astronomical
972: Conference, Contadakis M.E., Hadjidemetriou J.D., Mavridis L.N.,
973: Seiradakis J.H. (eds.), P. Ziti \& Co, Thessaloniki, p. 520-525
974: \bibitem{bar4}
975: Barbanis, B., Contopoulos, G., 1995, A\&A, 294, 33
976: \bibitem{bvv}
977: Barbanis, B., Varvoglis, H., Vozikis, Ch., 1999, A\&A, 344, 879
978: \bibitem{Ben76}Benettin, G., Galgani, L., Strelcyn J.M., 1976,
979: Phys. Rev. A, 14, 2338
980: \bibitem{CV87}
981: Caranicolas, N., Vozikis, Ch., 1987, Cel. Mech., 40, 35
982: \bibitem{CV99}
983: Caranicolas, N., Vozikis, Ch., 1999, A\&A, 349, 70
984: \bibitem{Cont66}
985: Contopoulos, G., 1966, In: Les Nouvelles M\'ethodes de la Dynamique
986: Stellaire'', Nahon, F. \& H\'enon, M. (eds.)
987: \bibitem{ContBar}
988: Contopoulos, G., Barbanis, B., 1989, A\&A, 222, 329
989: \bibitem{ContVog}
990: Contopoulos, G., Voglis, N., 1997, A\&A, 317, 73
991: \bibitem{Contetal1}
992: Contopoulos, G., Voglis, N., Efthymiopoulos, C., Froeschl\'e, C.,
993: Gonczi, R., Lega, E., Dvorak, R., Lohinger, E., 1997,
994: Cel. Mech. Dun. Astron., 67, 293
995: \bibitem{Dvorak}
996: Dvorak, R., Contopoulos, G., Efthymiopoulos, Ch., Voglis, N., 1998,
997: Planet. Space Sci., 46, 1567
998: \bibitem{cf1}
999: Froeschl\'e, C., 1984, Cel. Mech., 34, 95
1000: \bibitem{cfl1}
1001: Froeschl\'e, C., Lega, E., 1996, Cel. Mech. Dyn. Astron., 64, 21
1002: \bibitem{cfl2}
1003: Froeschl\'e, C., Lega, E., 1998, A\&A, 334, 355
1004: \bibitem{cfetal}
1005: Froeschl\'e, C., Froeschl\'e, Ch., Lohinger, E., 1993,
1006: Cel. Mech. Dyn. Astron., 51, 135
1007: \bibitem{cfetal2}
1008: Froeschl\'{e}, C., Lega, E. and Gonzi, R., 1997, Cel. Mech. Dyn. Astron.,
1009: 67, 41
1010: \bibitem{hh64}
1011: H\'{e}non M. and Heiles C., 1964, Astron. J., 69, 73
1012: \bibitem{lask1}
1013: Laskar, J., 1990, Icarus, 88, 266
1014: \bibitem{lask2}
1015: Laskar, J., 1993, Physica D, 67, 257
1016: \bibitem{lasketal}
1017: Laskar, J., Froeschl\'e, C., Celleti, A., 1992, Physica D, 56, 253
1018: \bibitem{legaF}
1019: Lega, E., Froeschl\'e, C., 1996, Physica D, 95, 97
1020: \bibitem{frloh}
1021: Lohinger, E. and Froeschl\'{e} C., 1993, Cel. Mech. Dyn. Astron., 57, 369
1022: \bibitem{Magnenat}
1023: Magnenat, P., 1982, Cel.Mech., 28, 319
1024: \bibitem{MN92}
1025: Milani A. and Nobili A., 1992, Nature, 357, 569
1026: \bibitem{numrec}
1027: Press W. H., Teukolsky S. A., Vetterling W. T. and Flannery B. P., 1992,
1028: {\it Numerical Recipes in Fortran -- The Art of Scientific Computing} 2nd
1029: edn (Cambridge: Cambridge University Press)
1030: \bibitem{TAV99}
1031: Tsiganis K., Anastasiadis A. and Varvoglis H., 2000a, Chaos Solit.
1032: \& Fract. (in press)
1033: \bibitem{TVH99}
1034: Tsiganis K., Varvoglis H. and Hadjidemetriou J.D., 2000b, Icarus
1035: (in press)
1036: \bibitem{VVB}
1037: Varvoglis, H., Vozikis, Ch., Barbanis, B. 1997, In: The Dynamical
1038: Behaviour of Our Planetary System, Henrard J., Dvorak R.
1039: (eds.), Kluwer, Dordrecht
1040: \bibitem{vogia}
1041: Voyatzis, G., Ichtiaroglou, S., 1992, J. Phys. A, 25, 5931
1042: \bibitem{Voglis1}
1043: Voglis, N., Contopoulos, G., 1994, J. Phys. A, 27, 4899
1044: \bibitem{Voglis2}
1045: Voglis, N., Contopoulos, G., Efthymiopoulos, C., 1998, Phys. Rev. E,
1046: 57, 372
1047: \bibitem{Voglis3}
1048: Voglis, N., Contopoulos, G., Efthymiopoulos, C., 1999,
1049: Cel. Mech. Dyn. Astron., 73, 211
1050: \bibitem{Voglis4}
1051: Voglis, N., Efthymiopoulos, C., 1998,
1052: J. Phys., A31, 2913
1053:
1054: \end{thebibliography}
1055:
1056: \end{document}
1057: