1: %
2: \documentclass[a4paper,12pt]{article}
3: %
4: \usepackage{epsfig}\usepackage{amssymb}
5: \newcommand\FIGURE[5]{\begin{figure}[htb]\centering\epsfig{file=#1.ps,
6: width=#2\linewidth,angle=#3}\caption{#4}\label{#5}\end{figure}}
7: \newcommand\FIGGER[3]{\begin{figure}[htb]\centering\epsfig{file=#1.ps,
8: width=#2\linewidth,angle=#3}\end{figure}}
9: %
10: \title{A Dynamical System with Two Strange Attractors.}
11: \author{R. C. Johnson\thanks{{\tt bob.johnson@dur.ac.uk}}
12: \\ {\em Department of Mathematical Sciences},\\ {\em University of Durham},
13: {\em Durham DH1 3LE}, {\em England}.}
14: %\renewcommand{\baselinestretch}{1.8}
15: \begin{document}
16: %
17: \maketitle
18: %
19: \begin{abstract}A six-dimensional R\"ossler-Lorenz hybrid
20: has two coexistent attractors. Both, either or neither may be
21: strange.\end{abstract}
22: %
23: \paragraph{Introduction.}
24: The R\"ossler \cite{rossler} and Lorenz \cite{lorenz} equations are textbook
25: examples of chaotic dynamical systems. Each is three-dimensional,
26: continuous-time, smooth and autonomous, with a single strange attractor for
27: certain parameter values.
28:
29: This paper draws attention to a six-dimensional hybrid with two coexistent
30: attractors, one or both of which may be strange.
31:
32: The attractors evidently stem from the constituent R\"ossler and Lorenz
33: systems, but each requires and occupies the larger phase space.
34:
35: The new system may well have potential for physical modelling, but its
36: discovery is the result of curiosity.
37: %
38: \paragraph{Construction.}
39: Start with the Lorenz equations \cite{lorenz}
40: \begin{eqnarray}\dot x&=&\sigma(y-x)\nonumber\\
41: \dot y&=&(r-z)x-y\label{eLRNZ}\\ \dot z&=&xy-\beta z\nonumber\end{eqnarray}
42: in a chaotic regime with
43: \begin{equation}\sigma=10,\quad r=28,\quad\beta=\textstyle{8\over3}.
44: \label{eLPAR}\end{equation}
45: Fig.~1(a) shows a corresponding solution trajectory on the familiar
46: \cite[p.~283]{geometry} two-winged strange attractor.
47:
48: Consider replacing $z$ on the right-hand side in the last component by some
49: external excitation $w(t)$:
50: \begin{equation}xy-\beta z\quad\to\quad xy-\beta w.\label{eZ}\end{equation}
51: The response may be regular or irregular, depending on $w$.
52:
53: For instance, with $w=\cos t$, there is steady oscillation about values
54: $$x=y=0,\quad z=r=28.$$If instead the excitation $w(t)$ is an isolated
55: pulse then the response is a sudden excursion of $z(t)$ followed later by
56: sudden relaxation --- each change accompanied by spikes of $x(t)$ and $y(t)$.
57:
58: Now, the chaotic R\"ossler system \cite{rossler}
59: \begin{eqnarray}\dot u&=&-v-w\nonumber\\
60: \dot v&=&u+av\label{eRSLR}\\
61: \dot w&=&b+(u-c)w\nonumber\end{eqnarray}
62: with parameters\begin{equation}a=0.2,\quad b=0.2,\quad
63: c=5.7\label{eRPAR}\end{equation}
64: has solution trajectories on a well-known \cite[p.~287]{geometry}
65: twisted-band strange attractor --- as Fig.~1(b) shows.
66:
67: Evidently its third component $w(t)$ provides an irregular sequence of
68: isolated pulses. Such a sequence, via eq.~(\ref{eZ}), provokes the Lorenz
69: $(x,y,z)$ variables into ragged oscillation about $(0,0,r)$.
70:
71: This chaotic ringing, however, does not have composite properties. For
72: delay-coordinate reconstruction \cite{ntsa} from $z(t)$ reveals a
73: R\"ossler-like twisted-band attractor in three embedding dimensions. Lorenz
74: adds nothing.
75:
76: Higher-dimensional and richer structure is possible --- but needs feedback
77: from Lorenz to R\"ossler. And one amoung many options is to mix the
78: oscillations of the $x$--$y$ pair into those of $u$ and $v$.
79:
80: Experiments eventually lead to linear driving of $v$ by $x$ and $y$ --- with
81: further linear coupling from $u$ to $x$ --- as follows:
82: \begin{eqnarray}\dot u&=&-v-w\nonumber\\
83: \dot v&=&x+ay\nonumber\\
84: \dot w&=&b+(u-c)w\nonumber\\
85: \dot x&=&\sigma(y-x)+\mu u\label{eROLO}\\
86: \dot y&=&(r-z)x-y\nonumber\\
87: \dot z&=&xy-\beta w.\nonumber\end{eqnarray}
88: This is the system of interest. With other parameters fixed, adjustments of
89: $\mu$ and $\sigma$ give one or two strange attractors, or a cycle plus a
90: strange attractor, or two cycles, or a single cycle.
91: %
92: \paragraph{Properties.}
93: If parameters $r$, $\sigma$, $\beta$, $a$, $b$, $c$ take the values in
94: eqs.~(\ref{eLPAR}) and (\ref{eRPAR}), phase-space volumes shrink rapidly.
95:
96: For $\sigma=10$, the system of eq.~(\ref{eROLO}) has two coexisting
97: attractors if $\mu$ is between about $0$ and $13$. They have the following
98: properties.\begin{itemize}
99: \item
100: The two attractors centre approximately around $u=x=y=0$ and $z=r=28$ but
101: are displaced in $v$ and $w$. See Figs.~2 and 3. Rectangular phase-space
102: boxes just sufficient to enclose them differ in volume by a factor of
103: typically $10^5$.
104: \item
105: The larger attractor exists for $\mu\lesssim13$, where it is always strange.
106: Its toadstool-like $xyz$ projection --- Fig.~2(a) --- is reminiscent of the
107: Lorenz attractor in Fig.~1(a). For this reason it is labelled $L$.
108: \item
109: The smaller one exists for $\mu\gtrsim0$ and is strange for
110: $\mu\lesssim3.26$ --- see Fig.~3. As $\mu$ increases beyond $3.26$ (through
111: 50 and beyond) it alternates between regular --- i.e., a cycle --- and
112: strange, bifurcating by period-doubling. This parallels the behaviour of the
113: Lorenz attractor as $r$ increases \cite{sparrow}.
114: \item
115: Its twisted-band $uvw$ projection --- Fig.~3(b) --- persists with changing
116: $\mu$. Because it is similar to the R\"ossler attractor in Fig.~1(b) this
117: attractor is labelled $R$.
118: \item
119: The two attractors have $v_L<v_R$ and $w_L>w_R$ and, where they coexist,
120: trajectories move much faster round $L$ than around $R$ --- see Fig.~4.
121: \item
122: Fig.~4 also shows more variability of period and amplitude on $R$ than
123: on $L$, and trajectories on $R$ and $L$ with $\mu=1$ are found to have
124: maximum Lyapunov exponent of about $1.4$ and $0.45$ respectively.
125: \item
126: Both attractors, where they are strange, are higher-dimensional objects. For
127: delay-coordinate reconstruction \cite{ntsa} of $R$ and $L$ from samples of
128: components $u,v,\dots,z$ shows that they need about 5 embedding dimensions.
129: \end{itemize}
130: These properties depend in different ways on $r$, $\beta$, $a$, $b$, $c$
131: and $\sigma$.
132: \begin{itemize}\item
133: A change of $r$, for instance, shifts $R$ and $L$ bodily along the
134: $z$-axis. And while minor variations in $a$, $b$, $c$ and $\beta$ affect
135: other numerical details, there are larger changes with $\sigma$.
136: \item With $\mu=1$, $L$ is a cycle for $2.8\lesssim\sigma\lesssim4.5$ ---
137: and then $R$ is also a cycle for $\sigma\approx3.7$ --- see Fig.~5.
138: \item As Fig.~5(a) suggests, both cycles bifurcate to strange attractors by
139: period-doubling. In this regime they overlap in all but their $v$ component
140: --- they still have $v_L<v_R$.
141: \end{itemize}
142: Note that when $R$ and $L$ vanish with parameter changes it is because their
143: basins of attraction shrink. No examples have been found where they
144: coalesce.
145:
146: Thorough exploration of six-dimensional phase space, varying seven
147: parameters, is a formidable task. However, investigation of the
148: $\mu$-$\sigma$ plane is promising, and some further illustrations are
149: available elsewhere \cite{www}.
150: %
151: \paragraph{Conclusion.}
152: As it stands, the system of eq.~(\ref{eROLO}) is no more than an interesting
153: construction without direct physical origin. It is just one in a range of
154: possible R\"ossler-Lorenz hybrids.
155:
156: Its immediate value is to show that six-dimensional phase space is roomy
157: enough to accommodate two attractors not connected by (e.g.) symmetry ---
158: and that all combinations of chaotic and regular oscillation may coexist.
159:
160: Dynamical models that use in the order of six effective dimensions occur
161: in for instance physiology (e.g. \cite{phys}), meteorology (e.g.
162: \cite{weath}) and economics (e.g. \cite{econ}) --- where the possibility of
163: such coexistent modes could be very interesting.
164: %
165: \paragraph{Acknowledgement.} The trajectory calculations used {\sl XPPAUT}
166: \cite{bard}. Lyapunov exponents and embedding dimension were determined
167: with {\sl NDT} \cite{ndt}. The authors are applauded for the functionality
168: and availability of their software.
169: %
170: \newpage
171: \begin{thebibliography}{99}
172: %
173: \bibitem{rossler}O. E. R\"ossler, Phys. Lett. A 57 (1976) 397.
174: %
175: \bibitem{lorenz} E. N. Lorenz, J. Atmos. Sci. 20 (1963) 130.
176: %
177: \bibitem{geometry} R. H. Abraham and C. D. Shaw, {\sl Dynamics: the
178: Geometry of Behaviour} (Addison-Wesley, Redwood City CA) 2nd ed. 1992.
179: %
180: \bibitem{ntsa} H. Kantz and T. Schreiber, {\sl Nonlinear Time-Series
181: Analysis} (CUP, Cambridge) 1997.
182: %
183: \bibitem{sparrow}C. Sparrow, {\sl The Lorenz Equations: Bifurcations,
184: Chaos, and Strange Attractors} (Springer, NY) Appl. Math. Sci. 41 (1982).
185: %
186: \bibitem{www}{\sl Ro-Lo Home Page}, {\tt maths.dur.ac.uk/\~{}dma0rcj/RoLo/}
187: %
188: \bibitem{phys}B. J. West, {\sl Fractal Physiology and Chaos in Medicine},
189: Studies of Nonlinear Phenomena in Life Science Vol. 1 (World Scientific
190: Publishing) 1990;\\
191: C. C. King, {\sl Fractal and Chaotic Dynamics in Nervous Systems},
192: Progress in Neurobiology 36 (1991) 279.
193: %
194: \bibitem{weath} The Met.Office, {\sl Numerical Weather Prediction},\\
195: {\tt www.met-office.gov.uk/sec5/sec5pg2.html}
196: %
197: \bibitem{econ}J. Benhabib, {\sl On Cycles and Chaos in Economics}
198: (Bibliography),\\
199: {\tt www.mitpress.com/e-journals/SNDE/001/abstracts/v1n1001.html};\\
200: A. Garratt, K. Lee, M. Hashem Pesaran and Y. Shin, {\sl A Long-Run
201: Structural Macroeconometric Model of the UK},
202: (University of Cambridge DAE Working Paper No.9812, revised 1999),\\
203: {\tt www.econ.cam.ac.uk/faculty/pesaran/ukm95.pdf}
204: %
205: \bibitem{bard}G. B. Ermentrout, {\sl XPPAUT}, {\tt www.pitt.edu/\~{}phase/}
206: %
207: \bibitem{ndt}J. Reiss, {\sl Nonlinear Dynamics Toolbox},\\
208: {\tt www.physics.gatech.edu/chaos/research/NDTOnline/aboutndt/ndt.htm}
209: \end{thebibliography}
210: %
211: %\end{document}
212: \newpage
213: \section*{Figure captions}
214: \begin{description}
215: \item[Figure 1](a) Solution of the Lorenz system eq.~(\ref{eLRNZ}) with
216: parameters as in eq.~(\ref{eLPAR}) --- a trajectory on its
217: two-winged strange attractor.
218: (b) Solution of the R\"ossler system eq.~(\ref{eRSLR}) with
219: parameters as in eq.~(\ref{eRPAR}) --- a trajectory on its
220: twisted-band strange attractor.
221: \item[Figure 2]System of eq.~(\ref{eROLO}) at $\mu=1$ with other
222: parameters as in eqs.~(\ref{eLPAR}) and (\ref{eRPAR}). The
223: $L$-attractor --- (a) $xyz$-projection of a trajectory;
224: (b) $uvw$-projection of a trajectory.
225: \item[Figure 3]System of eq.~(\ref{eROLO}) at $\mu=1$ with other
226: parameters as in eqs.~(\ref{eLPAR}) and (\ref{eRPAR}). The
227: $R$-attractor --- (a) $xyz$-projection of a trajectory;
228: (b) $uvw$-projection of a trajectory.
229: \item[Figure 4]System of eq.~(\ref{eROLO}) at $\mu=1$ with
230: other parameters as in eqs.~(\ref{eLPAR}) and (\ref{eRPAR}).
231: Component $z(t)$ of trajectories on (a) the $L$-attractor and
232: (b) the $R$-attractor. Compare the time scales and
233: variability of period and amplitude.
234: \item[Figure 5]System of eq.~(\ref{eROLO}) at $\mu=1$ with $\sigma=3.7$
235: and otherwise parameters as in eqs.~(\ref{eLPAR}) and (\ref{eRPAR}).
236: (a) The $L$-attractor --- $xyz$-projection of a trajectory.
237: (b) The $R$-attractor --- $uvw$-projection of a trajectory.
238: Both attractors are cycles, and $L$ is doubled.
239: \end{description}
240: %
241: %\end{document}
242: \newpage
243: % fig 1
244: \FIGGER{fig1a}{.5}{-90}
245: \FIGURE{fig1b}{.5}{-90}{(a) above, (b) below}{fONE}
246: % fig 2
247: \FIGGER{fig2a}{.5}{-90}
248: \FIGURE{fig2b}{.5}{-90}{(a) above, (b) below}{fTWO}
249: % fig 3
250: \FIGGER{fig3a}{.5}{-90}
251: \FIGURE{fig3b}{.5}{-90}{(a) above, (b) below}{fTHREE}
252: % fig 4
253: \FIGGER{fig4a}{.5}{-90}
254: \FIGURE{fig4b}{.5}{-90}{(a) above, (b) below}{fFOUR}
255: %
256: % fig 5
257: \FIGGER{fig5a}{.5}{-90}
258: \FIGURE{fig5b}{.5}{-90}{(a) above, (b) below}{fFIVE}
259: %
260: \end{document}
261: