1: \documentstyle[prl,aps]{revtex}
2: \draft
3: \begin{document}
4: \input{epsf}
5: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname@twocolumnfalse\endcsname
6: \title{Regularization of Synchronized Chaotic Bursts}
7: \author{Nikolai F. Rulkov}
8: \address{
9: Institute for Nonlinear Science, University of California, San
10: Diego, La Jolla, CA 92093-0402\\}
11:
12: \date{\today}
13: \maketitle
14: \begin{abstract}
15: The onset of regular bursts in a group of irregularly bursting
16: neurons with different individual properties is one of the most
17: interesting dynamical properties found in neurobiological systems.
18: In this paper we show how synchronization among chaotically
19: bursting cells can lead to the onset of regular bursting. In order
20: to clearly present the mechanism behind such regularization we
21: model the individual dynamics of each cell with a simple
22: two-dimensional map that produces chaotic bursting behavior similar
23: to biological neurons.
24:
25:
26: \end{abstract}
27: \pacs{PACS number(s): 05.45.+b, 87.22.-q}
28:
29: \narrowtext
30: \vskip1pc]
31:
32: Studies of cooperative behavior in coupled chaotic oscillators are
33: frequently based upon the analysis of the phenomenon of chaos
34: synchronization~\cite{synchr}. Different types of synchrony between
35: chaotic oscillators along with the various mechanisms responsible
36: for the onset of such synchronization have been
37: studied~\cite{sync}. These studies are mostly motivated by the
38: development of a theoretical framework that can explain
39: synchronization in neurobiological systems where complex, chaotic
40: behavior of neurons is quite typical~\cite{biosynch}.
41:
42: It has been observed in neurobiological experiments and in
43: numerical simulations that individual neurons may show irregular
44: bursts, while ensembles of such irregularly bursting neurons can
45: synchronize and produce regular, rhythmical bursting. This
46: regularization occurs despite the significant differences in the
47: individual dynamics of coupled neurons~\cite{CPG}. How can a group
48: of different neurons, whose individual dynamics are very sensitive
49: to the intrinsic parameters of the neuron, synchronize and produce
50: a robust rhythm of the bursts? How can such rhythmical bursting
51: become insensitive to the intrinsic parameters that control the
52: individual dynamics of the neurons? Answers to these questions may
53: be found through the study of models, which, on one hand, are
54: capable of demonstrating similar effects and, on the other hand,
55: are simple enough to reveal the mechanisms that lead to these
56: effects.
57:
58: In this communication we present an example that demonstrates the
59: effects of mutual synchronization and {\em chaos regularization} of
60: bursts in a group of chaotically bursting cells. We study $N$
61: oscillators, which are modeled with two-dimensional maps and
62: coupled to each other through the mean field~\cite{coupling}
63: \begin{eqnarray}\label{maps}
64: x(i,n+1)&=&\frac{\alpha}{(1+x(i,n)^2)}+y(i,n)+\frac{\epsilon}{N}\sum_{j=1}^{N}x(j,n)~,
65: \nonumber \\
66: y(i,n+1)&=&y(i,n)-\sigma x(i,n)-\beta~, \label{nmaps}
67: \end{eqnarray}
68: where $x(i,n)$ and $y(i,n)$ are, respectively, the fast and slow
69: dynamical variables of the $i$-th oscillator, $\epsilon$ is the
70: strength of global coupling. The slow evolution of $y(i,n)$ is due
71: to the small values of the positive parameters $\beta$ and
72: $\sigma$, which are on the order of 0.001. The value of parameter
73: $\alpha$ is selected in the region $\alpha>4.0$, where the map
74: produces chaotic oscillations in $x(i,n)$. When $\epsilon=0$,
75: depending on the value of parameter $\alpha$, each cell
76: demonstrates two qualitatively different regimes of chaotic
77: behavior: continuous chaotic oscillations and chaotic bursts, see
78: Fig.~\ref{fig1}a. Such dynamics resemble the bursting dynamics
79: measured in the experiments with biological neurons, see for
80: example~\cite{elson98}.
81:
82: \begin{figure}
83: \begin{center}
84: \leavevmode
85: \hbox{%
86: \epsfxsize=7.5cm
87: \epsffile{figure1.eps}}
88: \end{center}
89: \caption{The waveforms (left) and power spectra (right) of the chaotic
90: behavior of uncoupled cells, $\epsilon=0$, computed for different
91: values of $\alpha$, with $\sigma=\beta=0.001$. The power spectra
92: are computed for waveforms of length $L$, ($L$=50000 iterations).}
93: \label{fig1}
94: \end{figure}
95:
96: As one can see from Fig.~\ref{fig1}a, the mean duration of the
97: bursts is very sensitive to the value of $\alpha$. The chaotic
98: nature of the duration of the bursts can be clearly seen from the
99: power spectra calculated from $x(i,n)$. The low-frequency part of
100: the spectrum that corresponds to the frequency range of the bursts
101: is presented in Fig.~\ref{fig1}b. The introduction of coupling
102: between the cells results in the synchronization of the bursts.
103: While the bursts do become synchronized, the fast chaotic
104: oscillations of the cells during the bursts remain asynchronous.
105: The most interesting effect observed in this case is that the
106: synchronized bursts measured both in each chaotic cell and in the
107: mean field ($\sum_{j=1}^{N}x(j,n)/N$) become almost periodic, see
108: Fig~\ref{fig2}a. We define the formation of such a periodic
109: temporal pattern in the chaotic behavior of systems as {\em chaos
110: regularization}. The effect of chaos regularization can be clearly
111: seen in the power spectrum of $x(i,n)$ as the appearance of
112: periodic spectral components (compare Fig.~\ref{fig1}b and
113: Fig.~\ref{fig2}b). These periodic components correspond to the
114: period and the harmonics of the periodic temporal pattern formed in
115: the synchronized chaotic bursts. In the rest of the paper we will
116: study the mechanisms behind the chaos regularization of the bursts
117: in the model cells (\ref{nmaps}).
118:
119: \begin{figure}
120: \begin{center}
121: \leavevmode
122: \hbox{%
123: \epsfxsize=7.5cm
124: \epsffile{figure2.eps}}
125: \end{center}
126: \caption{The waveforms (left) and power spectra (right) of the
127: synchronized chaotic bursts of the coupled cells (\ref{nmaps}) with
128: $\epsilon=0.2$ and $N=256$. The values of the individual parameters
129: of the cells are randomly selected from the intervals: $\alpha$
130: from 4.1 to 4.9; $\beta$ and $\sigma$ from 0.0009 to 0.0011. The
131: value of $L$ is the same as in Fig.~\ref{fig1}}
132: \label{fig2}
133: \end{figure}
134:
135: {\em Chaotic Bursting}. The individual behavior of the cells is
136: described by the map of the form
137: \begin{eqnarray}
138: x_{n+1}&=&\frac{\alpha}{(1+x_n^2)}+y_n~,
139: \label{umapx} \\
140: y_{n+1}&=&y_n-\sigma x_n-\beta~. \label{umapy}
141: \end{eqnarray}
142: The slow evolution of $y_n$ for the next $m$ steps is given by
143: \begin{eqnarray}
144: y_{n+m}&=&y_n-m(\beta + \sigma \overline{x_{n,m}}), \label{mapy}
145: \end{eqnarray}
146: where $\overline{x_{n,m}}= (\sum_{j=n+1}^{n+m} x_j)/m$ is the mean
147: value of $x_n$ computed for $m$ consecutive iterations. It is clear
148: from (\ref{mapy}) that the value of $y_n$ slowly increases during
149: the next $m$ steps if $\sigma \overline{x_{n,m}} < -\beta$, and
150: decreases if $\sigma \overline{x_{n,m}} > -\beta$.
151:
152:
153: Since $y_n$ changes slowly, the dynamics of $x_n$ can be considered
154: independently of the map (\ref{umapy}) assuming that $y_n$ is a
155: parameter $\gamma = y_n$. Therefore, fast dynamics of the cell can
156: be understood from the analysis of the 1-D map
157: \begin{eqnarray}
158: x_{n+1}&=&F(x_n,\alpha,\gamma)\equiv\frac{\alpha}{(1+x_n^2)}+\gamma~.
159: \label{mapx}
160: \end{eqnarray}
161: The main dynamical properties of this map are clear
162: from~Fig~\ref{mplot}. The results of the bifurcation analysis of
163: the map are presented in Fig.~\ref{mbif}. If the parameter values
164: of the map are within the horn formed by the bifurcation curves
165: $L_{\tau12}$ and $L_{\tau23}$, given by the equations
166: $\alpha=-2((\gamma^2+9)\gamma-(\gamma^2-3)^{3/2})/27$ and
167: $\alpha=-2((\gamma^2+9)\gamma+(\gamma^2-3)^{3/2})/27$ respectively,
168: then the map~(\ref{mapx}) has three fixed points $x^*_1$, $x^*_2$
169: and $x^*_3$. The fixed point $x^*_1$ is stable, $x^*_2$ is unstable
170: and $x^*_3$ may change stability, see Fig.~\ref{mplot}.
171:
172: \begin{figure}
173: \begin{center}
174: \leavevmode
175: \hbox{%
176: \epsfxsize=4.2cm
177: \epsffile{figure3a.eps}
178: \epsfxsize=4.2cm
179: \epsffile{figure3b.eps}}
180: \end{center}
181: \caption{The shape of the function $F(x_n,\alpha,\gamma)$ plotted for
182: $\alpha=4.1$ and three different values of $\gamma$ (-2.65, -2.75
183: and -2.85) (a) and the chaotic trajectory of the map (b). $x^*_1$,
184: $x^*_2$ and $x^*_3$ are the fixed points of the map. $\gamma^*$
185: corresponds to the bifurcation value where $x^*_1$ and $x^*_2$
186: merge together.}
187: \label{mplot}
188: \end{figure}
189:
190:
191: When the parameter values of the map cross the curve $L_{\tau12}$
192: from left to right then the fixed points $x^*_1$ and $x^*_2$ merge
193: together and disappear. This bifurcation in (\ref{mapx})
194: corresponds to the beginning of the burst in system (\ref{umapx})
195: and (\ref{umapy}). Indeed, when $x_n$ is in the stable fixed point
196: $x^*_1$, then $\gamma
197: \equiv y_n$ slowly grows because $\overline{x_{n,1}}\approx x^*_1 <
198: -\beta/\sigma$, see~(\ref{mapy}). When the stable fixed point $x^*_1$
199: disappears, the trajectory of the fast map~(\ref{mapx}) goes to the
200: chaotic attractor corresponding to the chaotic pulsations during
201: the burst, see Fig.~\ref{mplot}.
202:
203: The end of the chaotic burst is due to the external crises of the
204: chaotic attractor in system (\ref{mapx}). When the trajectory of
205: this map is on the chaotic attractor, then
206: $\overline{x_{n,1}}>-\beta/\sigma$ (see Fig.~\ref{mplot}), and the
207: value of $\gamma \equiv y_n$ decreases until the chaotic trajectory
208: of the fast map arrives back to the stable fixed point $x^*_1$.
209: This becomes possible only after the parameter of the fast map goes
210: through the bifurcation value, which corresponds to the case when
211: the trajectory from the maximum of the map function
212: $x_{max}=F(0,\alpha,\gamma)$ maps into the unstable fixed point
213: $x^*_2$. Curve $L_h$, that corresponds to the bifurcation values
214: where $x_{max}$ iterates into a fixed point $x^*_2$ or $x^*_1$, is
215: given by the equation $\alpha=-(3\gamma
216: \pm\sqrt{\gamma^2-8})/2 $ and is shown in Fig.~\ref{mbif} . Note that
217: the external crises of the chaotic attractor is associated only
218: with the fixed point $x^*_2$. This bifurcation corresponds to the
219: low branch of the curve $L_h$ that starts from point $A12$.
220:
221: \begin{figure}
222: \begin{center}
223: \leavevmode
224: \hbox{%
225: \epsfxsize=3.9cm
226: \epsffile{figure4a.eps}
227: \epsfxsize=3.9cm
228: \epsffile{figure4b.eps}}
229: \end{center}
230: \caption{The bifurcation diagram of the uncoupled fast map (a), and
231: the enlarge portion of the diagram around the point $A12$ (b). The
232: bifurcation curves corresponding to the merging of the fixed points
233: $x^*_j$ and $x^*_m$ (curves $L_{\tau jm})$ and to the crisis of the
234: chaotic attractor (curve $L_h$). $g$ shows the gap between the
235: bifurcation curves.}
236: \label{mbif}
237: \end{figure}
238:
239:
240:
241: Therefore, the duration of the chaotic burst is determined by the
242: time interval that is required for the slow variable $y_n$ to move
243: from the bifurcation curve $L_{\tau12}$ to the curve $L_h$ plus the
244: additional time interval that is needed for the trajectory to map
245: below the unstable fixed point $x^*_2$. Both of these intervals
246: fluctuate because of the chaotic nature of the trajectory $x_n$,
247: and this is the reason for the chaotic fluctuations of the duration
248: of the bursts. After the chaotic firing terminates the trajectory
249: arrives at the stable fixed point $x^*_1$, and $\gamma$ slowly
250: moves towards $L_{\tau12}$, where the next burst starts. One can
251: see from Fig~\ref{mbif}b that the size of gap $g$ between the
252: bifurcation curves $L_{\tau12}$ and $L_h$ is very sensitive to the
253: value of parameter $\alpha$. That explains the high sensitivity of
254: the mean duration of the bursts to the internal parameter of the
255: cell $\alpha$.
256:
257: If the value of parameter $\alpha$ is selected above the point
258: $A12$, then the fixed points $x^*_1$ and $x^*_2$ appear within the
259: chaotic attractor (internal crisis). In this case $y_n$ fluctuates
260: in the vicinity of the bifurcation curve $L_{\tau 12}$ keeping the
261: cell in the regime of continuous chaotic oscillations, see
262: Fig.~\ref{fig1}a. We would like to note that the considered
263: mechanism of chaotic bursting is essentially the same as in the
264: Hindmarsh-Rose model of biological neuron, where the role of
265: parameter $\alpha$ is played by a hyperpolarization current, see
266: Ref.~\cite{HRModel}.
267:
268: {\em Synchronization}. Coupling between the cells (\ref{nmaps})
269: influences the fast dynamics of each cell (\ref{mapx}) by adding
270: the value $\epsilon \sum_{j=1}^{N}x(j,n)/N$ to the parameter
271: $\gamma$. When the $i$-th cell approaches the bifurcation values at
272: $L_{\tau12}$, being in the stable fixed point $x^*_1$, its behavior
273: becomes very sensitive to the influence of the other cells. The
274: cells, which are crossing this curve and starting to fire, sharply
275: increase their values of $x(j,n)$. As a result, the increased value
276: of $\gamma$ in the $i$-th cell pushes the cell over the bifurcation
277: value and triggers the burst. The greater the number of cells which
278: are involved in the simultaneous transition to the firing phase,
279: the larger the triggering impact will be that is experienced by the
280: remaining cell. Such an avalanche type of synchronous switching
281: takes place both at the beginning and at the end of each burst.
282: This threshold mechanism of synchronization is similar to the one
283: studied in ensembles of integrate-and-fire
284: oscillators~\cite{strogatz}.
285:
286: \begin{figure}
287: \begin{center}
288: \leavevmode
289: \hbox{%
290: \epsfxsize=4.65cm
291: \epsffile{figure5ab.eps}
292: \epsfxsize=3.5cm
293: \epsffile{figure5c.eps}}
294: \end{center}
295: \caption{The waveforms of the mean field (solid line) and of $x(i,n)$
296: for the cell with $\alpha=4.5$ (dots) computed for the group of
297: synchronously bursting cells with $\epsilon=0.1$ (a) and
298: $\epsilon=0.2$ (b). The bifurcation curves of the fast maps in the
299: case of synchronously bursting cells (c).}
300: \label{mbifc}
301: \end{figure}
302:
303:
304: {\em Regularization} The dynamical mechanism of regularization can
305: be also understood from the bifurcations of the fast map
306: (\ref{mapx}) taking into account the alternation of the two phases
307: in the waveform of the mean field $\epsilon \sum_{j=1}^{N}x(j,n)/N$
308: for synchronized bursts, see Fig.~\ref{mbifc}a,b. When the
309: synchronously bursting cells are in the non-firing phase, the fast
310: variable of each cell is located in the stable fixed point $x^*_1$
311: and the values of $y(i,n)$ and the $x$-coordinates of the fixed
312: point $x^*_1(i,n)$ slowly increase. In this case the evolution of
313: $x$ in each cell can be described by the map (\ref{nmaps}), where
314: the coupling term can be approximated by the value $\epsilon
315: x(i,n)$.
316: \begin{eqnarray}
317: x(i,n+1)&=&\frac{\alpha}{(1+x(i,n)^2)}+
318: \gamma+\epsilon x(i,n).
319: \label{mapix}
320: \end{eqnarray}
321: The bifurcation curve $L_{\tau12}$, which defines the threshold
322: value of $y(i,n)$ corresponding to the beginning of the burst, is
323: now given by the equation
324: $27\alpha(1-\epsilon)^2/2=-(\gamma^2+9(1-\epsilon)^2)
325: \gamma-(\gamma^2- 3(1-\epsilon)^2)^{3/2}$.
326: The location of this curve on the parameter plane $(\gamma,\alpha)$
327: depends upon the value of the coupling parameter $\epsilon$ and
328: shifts to the right from the original position of $L_{\tau12}$, see
329: Fig.~\ref{mbifc}c.
330:
331: When all synchronously bursting cells are in the phase of chaotic
332: firing their fast chaotic pulsations are uncorrelated. Therefore,
333: the value of the coupling term ($\epsilon \sum_{j=1}^{N}x(j,n)/N$)
334: during the burst can be approximated with the mean value of
335: $<x(j,n)>$. Due to the averaging, the fast chaotic fluctuation in
336: the coupling term disappears and the value $|\epsilon <x(j,n)>|$
337: slowly drifts in the interval between 0 and 0.08, see
338: Figs~\ref{mbifc} a,b. These values are significantly less than the
339: amplitude of the slow variable $|y_n|$, which is of the order of
340: one. Therefore, for simplicity, the contribution from the mean
341: field to the parameter $\gamma$ can be neglected during the firing
342: phase. As a result, the approximate position of the bifurcation
343: curves for the fast chaotic oscillations of synchronously firing
344: cells (\ref{nmaps}) can considered to be the same as for the
345: uncoupled map (\ref{mapx}).
346:
347: Fig.~\ref{mbifc}c presents the bifurcation curve $L_{\tau12}$
348: ($\epsilon>0$) that defines the beginning of a burst, and the
349: bifurcation curves $L_h$ (the part below $A12$) and $L_{\tau12}$
350: ($\epsilon=0$, the part above $A12$), which define the end of the
351: burst in each synchronously bursting cell. It follows from this
352: figure that the states of slow variable $y(i,n)$ corresponding to
353: the beginning and the end of the bursts become separated for all
354: values $\alpha$. This separation defines the period of the periodic
355: pattern formed in the chaotically bursting cells and helps to
356: synchronize cells with qualitatively different individual dynamics
357: whose bursting rates become insensitive to the parameter $\alpha$.
358:
359: Assuming that variables $y(i,n)$ drift with approximately the same
360: rate, the mean period of bursts, $T_B$, should be proportional to
361: the length of the most narrow gap $g$ between these two curves. One
362: can see from Fig.\ref{mbifc}c that the size of such a minimal gap
363: $g_{min}$ increases with the strength of the coupling. This
364: explains the change of bursts frequency in the waveforms shown in
365: Fig.\ref{mbifc}a,b.
366:
367: The presented analysis shows that chaos regularization in a group
368: of chaotically bursting cells has the following mechanism. When the
369: large enough fraction of the group of cells starts to burst
370: simultaneously due to the threshold synchronization, it leads to
371: the formation of relatively sharp transitions (ramps) in the mean
372: field (coupling term). These ramps create a gap between the states
373: of $y(i,n)$ corresponding to the beginning and the end of the
374: bursts for the rest of the cells, including the continuously firing
375: ones. All cells become bursting cells and it improves the
376: synchronization among them. Due to the formation of the gap, a
377: large portion of the group have about the same mean period which is
378: short enough to lead the bursting of the whole ensemble, see
379: Fig.\ref{mbifc}c. During the chaotic burst the chaotic components
380: of mean field are small because of averaging over the whole group
381: of the cell and, as the result, these components do not influence
382: the individual oscillations of the cell. When a large number of
383: firing cells get close to the threshold, the leading cells trigger
384: an avalanche of transitions. This avalanche sharply switches the
385: mean field to the non-firing phase and the rest of the cells are
386: forced to change their state independently of their chaotic
387: trajectory. This is how the synchronization makes the duration of
388: the bursts insensitive to the chaotic trajectory. It is clear that
389: the regularity of the bursts improves when the number of cells
390: increases. However, due to the formation of the gap, this type of
391: chaos regularization becomes noticeable even for small groups of
392: cells, including the case $N=2$.
393:
394: Finally, we would like to emphasize that this mechanism of chaos
395: regularization is due to the dynamical features of each cell at the
396: beginning and at the end of the chaotic burst. In the considered
397: map these features are characterized by bifurcations at
398: $L_{\tau12}$ and $L_h$, which are similar to a saddle-node
399: bifurcation and the appearance of a homoclinic orbit in more
400: realistic models of biological neurons, see for
401: example~\cite{HRModel}. Since the mechanism of chaos regularization
402: relies only upon the formation of sharp ramps in the coupling
403: forces, and the averaging down of the chaotic fluctuation during
404: the firing phase, this mechanism can occur for a number of
405: different types of coupling among the cells.
406:
407: The author is grateful to R. Elson, M.I. Rabinovich and H.D.I.
408: Abarbanel for helpful discussions. This work was supported in part
409: by U.S. Department of Energy (grant DE-FG03-95ER14516) and the U.S.
410: Army Research Office (MURI grant DAAG55-98-1-0269).
411:
412:
413: \begin{thebibliography}{99}
414:
415: \bibitem{synchr} See, for example, special focus issues devoted to
416: synchronization phenomena such as CHAOS {\bf 6}, No.3 (1996) and
417: CHAOS {\bf 7}, No.4 (1997)
418:
419: \bibitem{sync} Afraimovich, N.N. Verichev and M.I. Rabinovich,
420: Radiophys. Quant. Electr. {\bf 29}, 747 (1986); J.F. Heagy, L.M.
421: Pecora and T.L. Carroll, Phys. Rev. E {\bf 50}, 1874 (1994); N.F.
422: Rulkov {\em et al.}, Phys. Rev E {\bf 51}, 980 (1995); L. Kocarev
423: and U. Parlitz, Phys, Rev, Lett {\bf 76}, 1816 (1996); B.R. Hunt,
424: E. Ott and J.A. Yorke, Phys Rev. E {\bf 55}, 4029 (1997); A.
425: Pikovsky {\em et al.}, CHAOS {\bf 7}, 680 (1997); M.A. Zaks {\em et
426: al.}, Phys. Rev. Lett {\bf 82}, 4228 (1999).
427:
428: \bibitem{biosynch} L. Glass and and M.C. Mackey, {\em From clocks to
429: chaos: the rhythms of life} (Princeton, N.J. : Princeton University
430: Press, 1988); M.I. Rabinovich and H.D.I. Abarbanel, Neuroscience
431: {\bf 87}, 5 (1998); S.J. Schift {\em et al.}, Phys. Rev E {\bf 54},
432: 6708 (1996); R. Fitzgerald, Physics Today, March, 18 (1999);
433: Rabinovich, Radiophys. and Quant. Electr. {\bf 39}, 502 (1996);
434: H.D.I. Abarbanel {\em et al.,} Physics-Uspekhi, {\bf 39}, 337
435: (1996)..
436:
437: \bibitem{CPG} The Central Pattern Generator is a typical example of
438: such a system, see for example R.M. Harris-Warrick {\em et al.}, in
439: {\em Dynamics Biological Networks: The Stomatogastric Nervous
440: System}, edited by R.M. Harris-Warrick {\em et al.} (MIT Press,
441: Cambridge, MA,1992); M.I. Rabinovich {\em et al.} IEEE Tran. Circ.
442: Syst.-I {\bf 44}, 997 (1997);
443:
444: \bibitem{coupling} The dynamics of globally coupled nuerons with a
445: similar type of coupling were studied in a number of papers, see
446: for example D.Hansel and H. Sompolinsky, Phys. Rev. Lett., {\bf 68}
447: 718 (1992); M.I. Rabinovich {\em et al.}, Physica A, {\bf 263} 405
448: (1999); Y. Wang {\em et al.}, Phys. Rev. E, {\bf 61} 740 (2000),
449: and references overthere.
450:
451: \bibitem{elson98} R.C. Elson {\em et al.}, Phys. Rev. Lett. {\bf 81}
452: 5692 (1998).
453:
454: \bibitem{HRModel} J.L.Hindmarsh and R.M.Rose, Proc.R.Soc. London B
455: {\bf 221}, 87 (1984); X.-J. Wang, Physica D {\bf 6},263 (1993).
456:
457: \bibitem{strogatz} S.H. Strogatz and R.E. Mirollo, J. of Stat.
458: Phys., {\bf 63}, 613 (1991).
459:
460: \end{thebibliography}
461: \end{document}
462: