1: % ****** Start of file apssamp.tex ******
2: %
3: % This file is part of the APS files in the REVTeX 4 distribution.
4: % Version 4.0 of REVTeX, August 2001
5: %
6: % Copyright (c) 2001 The American Physical Society.
7: %
8: % See the REVTeX 4 README file for restrictions and more information.
9: %
10: % TeX'ing this file requires that you have AMS-LaTeX 2.0 installed
11: % as well as the rest of the prerequisites for REVTeX 4.0
12: %
13: % See the REVTeX 4 README file
14: % It also requires running BibTeX. The commands are as follows:
15: %
16: % 1) latex apssamp.tex
17: % 2) bibtex apssamp
18: % 3) latex apssamp.tex
19: % 4) latex apssamp.tex
20: %
21:
22: \documentclass[twocolumn,showpacs,preprintnumbers,amsmath,amssymb,nobibnotes,prl]{revtex4}
23: %\documentclass[preprint,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
24:
25: % Some other (several out of many) possibilities
26: %\documentclass[preprint,aps]{revtex4}
27: %\documentclass[preprint,aps,draft]{revtex4}
28: %\documentclass[prb]{revtex4}% Physical Review B
29:
30: \usepackage{graphicx}% Include figure files
31: \usepackage{dcolumn}% Align table columns on decimal point
32: \usepackage{bm}% bold math
33:
34: %\nofiles
35:
36: \begin{document}
37:
38: %\preprint{APS/123-QED}
39:
40: \title{Analysis of Firing Behaviors in Networks of Pulse-Coupled Oscillators\\
41: with Delayed Excitatory Coupling}% Force line breaks with \\
42:
43: \author{Wei Wu}
44: \email{051018023@fudan.edu.cn}
45: \author{Tianping Chen}%
46: \altaffiliation{Corresponding author}%Lines break automatically or can be forced with \\
47: \email{tchen@fudan.edu.cn}
48: \affiliation{Key Laboratory of Nonlinear Mathematics Science, School
49: of Mathematical Sciences, Fudan University, Shanghai 200433,
50: P.~R.~China}
51:
52: %\author{Ann Author}
53: % \altaffiliation[Also at ]{Physics Department, XYZ University.}%Lines break automatically or can be forced with \\
54: %\author{Second Author}%
55: % \email{Second.Author@institution.edu}
56: %\affiliation{%
57: %Authors' institution and/or address\\
58: %This line break forced with \textbackslash\textbackslash
59: %}%
60:
61: %\author{Charlie Author}
62: % \homepage{http://www.Second.institution.edu/~Charlie.Author}
63: %\affiliation{
64: %Second institution and/or address\\
65: %This line break forced% with \\
66: %}%
67:
68: \date{\today}% It is always \today, today,
69: % but any date may be explicitly specified
70:
71: \begin{abstract}
72: For networks of pulse-coupled oscillators with delayed excitatory
73: coupling, we analyze the firing behaviors depending on coupling
74: strength and transmission delay. The parameter space consisting of
75: strength and delay is partitioned into two regions. For one region,
76: we derive a low bound of interspike intervals, from which three
77: firing properties are obtained. However, this bound and these
78: properties would no longer hold for another region. Finally, we show
79: the different synchronization behaviors for networks with parameters
80: in the two regions.
81: \end{abstract}
82:
83: \pacs{05.45.¨Ca, 05.45.+b}% PACS, the Physics and Astronomy
84: % Classification Scheme.
85: %\keywords{Suggested keywords}%Use showkeys class option if keyword
86: %display desired
87: \maketitle
88:
89: For decades, complex networks have been focused on by scientists
90: from various fields, for instance, sociology, biology, chemistry and
91: physics, etc. In particular, networks of pulse-coupled oscillators,
92: as an important class of interconnected dynamical systems, have
93: gained increasing attentions because of their intimate relationship
94: to natural systems as diverse as cardiac pacemaker cells, flashing
95: fireflies, chirping crickets, biological neural networks, and
96: earthquakes (cf.
97: \cite{relationship-to-natual-systems,Buck1988,Mirollo1990}). A
98: pioneering work on modeling and analyzing pulse-coupled units was
99: done by Mirollo and Strogatz \cite{Mirollo1990}. Inspired by
100: Peskin's model for self-synchronization of the cardiac pacemaker,
101: they proposed a pulse-coupled oscillator model with undelayed
102: excitatory coupling to explain the synchronization of huge
103: congregations of South East Asian fireflies. With the framework of
104: the Mirollo-Strogatz model, many theoretical and numerical results
105: on pulse-coupled networks have been obtained
106: \cite{Vreeswijk1993-Chen1994-Corral1995,Mathar1996-Goel2002,Ernst1995,
107: Ernst1998,Timme2002-1,Timme2002-2,Kim2004,Wu2007}.
108:
109: Pulse-coupling is difficult to handle mathematically because it
110: introduces discontinuous behavior into the otherwise continuous
111: model and so stymies most of the standard mathematical techniques
112: \cite{Strogatz1983}. Particularly for delayed pulse-coupling, the
113: mathematical analysis of collective dynamics of networks becomes a
114: challenging problem. Past research experience indicates that some
115: underlying facts and assumptions about firing behaviors play a
116: crucial role in mathematical analysis
117: \cite{Mirollo1990,Mathar1996-Goel2002,Ernst1995,Ernst1998,Timme2002-1,Timme2002-2,Wu2007}.
118: For example, in \cite{Mirollo1990,Mathar1996-Goel2002},
119: synchronization was proved by making use of the fact that the firing
120: order of oscillators is always preserved for complete and undelayed
121: pulse-coupling; in \cite{Ernst1998}, an assumption about firing
122: times made the analysis easier by reducing the number of case
123: distinctions; in \cite{Wu2007}, the proof of desynchronization was
124: essentially due to a low bound of interspike intervals.
125:
126: In this Letter, networks of pulse-coupled oscillators with delayed
127: excitatory coupling are studied. We analyze the firing behaviors
128: depending on coupling strength and transmission delay. The parameter
129: space consisting of strength and delay is partitioned into two
130: regions. For one region, we give a low bound of interspike
131: intervals. By using the bound, three firing properties are derived,
132: which would be very helpful for discussing synchronization of
133: networks and stability of periodic solutions. Unfortunately, these
134: properties no longer hold for another region. Furthermore, the
135: different synchronization behaviors for networks with parameters in
136: the two regions are presented.
137:
138: We consider a system of $N$ identical oscillators which are
139: pulse-coupled in a delayed excitatory manner. As in
140: \cite{Timme2002-1}, the coupling structure is specified by the sets
141: $\mathrm{Pre}(i)$ of presynaptic oscillators that send pulses to
142: oscillator $i$, or the sets $\mathrm{Post}(i)$ of postsynaptic
143: oscillators that receive pulses from oscillator $i$. A phase
144: variable $\phi_i(t)\in[0,1]$ is used to characterize the state of
145: the oscillator $i$ at time $t$. In the case of no interaction, the
146: dynamics of $\phi_i$ is given by
147: \begin{eqnarray}
148: \mathrm{d}\phi_i(t)/\mathrm{d}t=1,\label{law1}
149: \end{eqnarray}
150: namely, the cycle period of the free oscillator is 1. When $\phi_i$
151: reaches the threshold $\phi_i=1$, the oscillator $i$ fires and
152: $\phi_i$ jumps back instantly to zero, after which the cycle
153: repeats. That is,
154: \begin{eqnarray}
155: \phi_i(t)=1\Rightarrow\phi_i(t^{+})=0.\label{law2}
156: \end{eqnarray}
157: Because of the transmission delay, the oscillators interact by the
158: following form of pulse-coupling: if oscillator $i$ fires at time
159: $t$, it emits a spike instantly; after a delay time $\tau$, the
160: spike reaches all postsynaptic oscillators $j\in\mathrm{Post}(i)$
161: and induces a phase jump according to
162: \begin{eqnarray}
163: \phi_j(t+\tau)=
164: f^{-1}(\min[1,f(\phi_j((t+\tau)^{-}))+\varepsilon_{ij}]),\label{law3}
165: \end{eqnarray}
166: where $\varepsilon_{ij}>0$ is the coupling strength from oscillator
167: $i$ to oscillator $j$; and the function $f$ is twice continuously
168: differentiable, monotonously increasing, $f'>0$, concave down,
169: $f''<0$, and satisfies $f(0)=0$, $f(1)=1$. For a more detailed
170: introduction of the model, see
171: \cite{Mirollo1990,Ernst1995,Ernst1998,Timme2002-1,Timme2002-2,Wu2007}.
172: In this Letter, we further assume the following: (i) The coupled
173: system starts at time $t=0$ with a set of initial phases
174: $0<\phi_i(0)\leq1$; (ii) there is no self-interaction, i.e.,
175: $i\not\in\mathrm{Pre}(i)$ for any oscillator $i$; (iii) $0<\tau<1$;
176: and (iv) the coupling strengths are normalized such that for all
177: oscillator $i$,
178: $\sum_{j\in\mathrm{Pre}(i)}\varepsilon_{ji}=\varepsilon$ with
179: $0<\varepsilon<1$.
180:
181: We partition the parameter space
182: $\mathcal{A}=\{(\tau,\varepsilon)|\,0<\tau<1,0<\varepsilon<1\}$ into
183: two regions
184: \begin{eqnarray*}
185: &&\mathcal{A}_1=\{(\tau,\varepsilon)\in\mathcal{A}|\,f(\tau)+\varepsilon<1\},\\
186: &&\mathcal{A}_2=\{(\tau,\varepsilon)\in\mathcal{A}|\,f(\tau)+\varepsilon\geq1\}.
187: \end{eqnarray*}
188: First of all, we use ``proof by contradiction" to prove that no
189: oscillator can fire twice in a time window of length $\tau$, if
190: parameters $(\tau,\varepsilon)\in\mathcal{A}_1$. Let $t_1$ and $t_2$
191: with $t_1<t_2$ be two successive firing times of oscillator $i$.
192: Suppose $t_2-t_1\leq\tau$. We claim that if $t_2>\tau$, there must
193: be some oscillator $i'\in\mathrm{Pre}(i)$ which fires more than once
194: in the time interval $(t_1-\tau,t_2-\tau]\cap[0,\infty)$. In fact,
195: this comes from the monotony and concavity assumption of the
196: function $f$. Since $f'>0$ and $f''<0$, we have that for any
197: $0<\delta<1$, if $0\leq\theta_1<\theta_2\leq f^{-1}(1-\delta)$, then
198: $f^{-1}(f(\theta_1)+\delta)-\theta_1<f^{-1}(f(\theta_2)+\delta)-\theta_2$,
199: namely the property (A7) in \cite{Ernst1998}. It implies that in the
200: same circle, the later the spike arrives, the larger the induced
201: phase jump is \cite{Ernst1995,Ernst1998}. Therefore, if all the
202: presynaptic oscillators $j\in\mathrm{Pre}(i)$ fire at most once in
203: $(t_1-\tau,t_2-\tau]\cap[0,\infty)$, then in the time interval
204: $(t_1,t_2]$ the sum of the phase jumps of oscillator $i$ is not more
205: than $f^{-1}(f(t_2-t_1)+\varepsilon)-(t_2-t_1)$, i.e., the sum
206: reaches its maximum if all spikes arrive at time $t_2$
207: simultaneously. It means $\phi_i(t_2)\leq
208: f^{-1}(f(t_2-t_1)+\varepsilon)\leq f^{-1}(f(\tau)+\varepsilon)<1$,
209: which contradicts that oscillator $i$ fires at $t_2$. Thus, there
210: exists some oscillator $i'\in\mathrm{Pre}(i)$ firing more than once
211: in $(t_1-\tau,t_2-\tau]\cap[0,\infty)$. Let
212: $t_3,t_4\in(t_1-\tau,t_2-\tau]\cap[0,\infty)$ with $t_3<t_4$ be two
213: successive firing times of oscillator $i'$. From $t_2-t_1\leq\tau$,
214: it follows that $t_4-t_3\leq\tau$. Similarly as above, if
215: $t_4>\tau$, then there must be some oscillator
216: $i''\in\mathrm{Pre}(i')$ which fires more than once in the time
217: interval $(t_3-\tau,t_4-\tau]\cap[0,\infty)$. Repeating the
218: derivation leads to a finite sequence of pairs of firing times:
219: \begin{eqnarray}
220: \{t_1,t_2\}\rightarrow\{t_3,t_4\}\rightarrow\cdots\rightarrow\{t_{2n-1},t_{2n}\}
221: \label{sequence}
222: \end{eqnarray}
223: which satisfies
224: $[t_{2k+1},t_{2k+2}]\subseteq(t_{2k-1}-\tau,t_{2k}-\tau]\cap[0,\infty)$
225: for $k=1,\ldots,n-1$; $t_{2n}\leq\tau$, $t_{2k}>\tau$ for
226: $k=1,\ldots,n-1$; and each term of (\ref{sequence}) is two
227: successive firing times of some oscillator. Particularly, $t_{2n-1}$
228: and $t_{2n}$ are two successive firing times of some oscillator
229: $i_0$. However, similarly as the argument of $\phi_i(t_2)<1$,
230: according to $t_{2n}\leq\tau$, $f(\tau)+\varepsilon<1$ and the
231: assumption that the coupled system starts at time $t=0$, we can get
232: $\phi_{i_0}(t_{2n})<1$. It contradicts that oscillator $i_0$ fires
233: at $t_{2n}$. This contradiction comes from our hypothesis
234: $t_2-t_1\leq\tau$. For a more detailed proof, see \cite{remark1}. As
235: a consequence, we get
236:
237: {\it Theorem 1:} If parameters $(\tau,\varepsilon)\in\mathcal{A}_1$,
238: all interspike intervals of each oscillator in the coupled system
239: must be longer than the delay time $\tau$.
240:
241: Here and throughout, ``interspike interval" is referred to as the
242: time between two successive firing activities of an oscillator.
243: However, as opposed to Theorem 1, at each
244: $(\tau,\varepsilon)\in\mathcal{A}_2$, the coupled system has
245: solutions in which some interspike intervals are not longer than
246: $\tau$. The simplest example is that the oscillators with initial
247: phases $\phi_1(0)=\cdots=\phi_N(0)$ fire synchronously with a period
248: $t=\tau$, if $(\tau,\varepsilon)\in\mathcal{A}_2$. In the rest of
249: the Letter, one will see that this can cause significantly different
250: dynamical behaviors of the system at
251: $(\tau,\varepsilon)\in\mathcal{A}_1$ and at
252: $(\tau,\varepsilon)\in\mathcal{A}_2$, especially the different
253: firing behaviors. Before discussing the difference of firing
254: behaviors, let us give some definitions and notations. Denote
255: $t^i_m$ the time at which oscillator $i$ fires its $m$-th time.
256: Clearly, the firing times $t^i_m$, $i=1\ldots,N$, $m\geq1$, are
257: determined by initial phases. For a given set of initial phases
258: $[\phi_1(0),\ldots,\phi_N(0)]$, the solution
259: $[\phi_1(t),\ldots,\phi_N(t)]$ is said to be a period-$d$ solution
260: if there exist a $\Delta t_0>0$, and positive integers $M$ and $d$
261: such that the firing times of arbitrary oscillator $i$ satisfies
262: $t^i_{m+d}-t^i_{m}=\Delta t_0$ for all $m\geq M$. For a given set of
263: initial phases $[\phi_1(0),\ldots,\phi_N(0)]$, the solution
264: $[\phi_1(t),\ldots,\phi_N(t)]$ is said to be a completely
265: synchronized solution if there exists a $T\geq0$ such that the phase
266: variables of arbitrary oscillators $i$ and $j$ satisfy
267: $\varphi_i(t)=\varphi_j(t)$ for all $t\geq T$. For the convenience
268: of later use, we let $\varepsilon_{ij}=0$ for
269: $j\not\in\mathrm{Post}(i)$. Then, the phase jump (\ref{law3}) also
270: holds for $j\not\in\mathrm{Post}(i)$.
271:
272: By using Theorem 1, we conclude that if
273: $(\tau,\varepsilon)\in\mathcal{A}_1$, any solution of the coupled
274: system possesses the following properties:
275:
276: {\it Property 1:} For oscillators $i$ and $j$ satisfying
277: $\varepsilon_{ij}=\varepsilon_{ji}$ and
278: $\varepsilon_{ki}=\varepsilon_{kj}$ for all $k\in
279: K_{ij}:=\{1,2,\ldots,N\}\setminus\{i,j\}$, if $t^i_{m_i}\leq
280: t^j_{m_j}$, then $t^i_{m_i+1}\leq t^j_{m_j+1}$, i.e., the firing
281: order of $i$ and $j$ is always preserved.
282:
283: {\it Property 2:} For oscillators $i$ and $j$ satisfying
284: $\varepsilon_{ij}=\varepsilon_{ji}$ and
285: $\varepsilon_{ki}=\varepsilon_{kj}$ for all $k\in K_{ij}$, if
286: $t^i_{m_i}=t^j_{m_j}$, then $\phi_i(t)=\phi_j(t)$ for all $t\geq
287: t^i_{m_i}$.
288:
289: {\it Property 3:} If $[\phi_1(t),\ldots,\phi_N(t)]$ is a completely
290: synchronized solution, then it is a period-one solution.
291:
292: In fact, Theorem 1 implies that in the case of
293: $(\tau,\varepsilon)\in\mathcal{A}_1$, a spike of oscillator $i$ must
294: reach oscillators $j\in\mathrm{Post}(i)$ before $i$ emits the next
295: spike. Thus, for oscillators $i$ and $j$ satisfying
296: $\varepsilon_{ij}=\varepsilon_{ji}$ and
297: $\varepsilon_{ki}=\varepsilon_{kj}$ for all $k\in K_{ij}$, the
298: instantaneous synchronization $t^i_{m_i}=t^j_{m_j}$ can lead to
299: $\phi_i(t)=\phi_j(t)$ for all $t\geq t^i_{m_i}$ (Property 2). For
300: the same reason, if $[\phi_1(t),\ldots,\phi_N(t)]$ is a completely
301: synchronized solution, then
302: $t^i_{m+1}-t^i_{m}=1-[f^{-1}(f(\tau)+\varepsilon)-\tau]$ for all $i$
303: and $t^i_{m}\geq T$. That is to say, any completely synchronized
304: solution is a period-one solution with the final interspike interval
305: being $1-[f^{-1}(f(\tau)+\varepsilon)-\tau]$ (Property 3).
306:
307: Due to space limitations, here we prove Property 1 for the case of
308: $N=2$. By Property 2, we only need to prove that if
309: $t^1_{m_1}<t^2_{m_2}$, then $t^1_{m_1+1}\leq t^2_{m_2+1}$. The proof
310: is divided into four cases:
311:
312: \noindent {\bf Case 1}: $t^2_{m_2}\geq t^1_{m_1+1}$.
313:
314: In this case, we have $t^1_{m_1+1}\leq t^2_{m_2}<t^2_{m_2+1}$.
315:
316: \noindent {\bf Case 2}: $t^1_{m_1}+\tau\leq t^2_{m_2}<t^1_{m_1+1}$.
317:
318: In this case, since $(t^2_{m_2},t^1_{m_1+1}]\subset
319: (t^1_{m_1}+\tau,t^1_{m_1+1}+\tau)$, oscillator $2$ cannot receive
320: any spikes from oscillator $1$ in the time interval
321: $(t^2_{m_2},t^1_{m_1+1}]$. This, combined with
322: $0=\phi_2((t^2_{m_2})^{+})<\phi_1(t^2_{m_2})<1$, leads to
323: $\phi_2(t)<\phi_1(t)$ for all $t\in(t^2_{m_2},t^1_{m_1+1}]$. It
324: implies $t^1_{m_1+1}<t^2_{m_2+1}$.
325:
326: \noindent {\bf Case 3}: $t^2_{m_2}<t^1_{m_1}+\tau$ and
327: $\phi_1(t^1_{m_1}+\tau)>\phi_2(t^1_{m_1}+\tau)$.
328:
329: Since $(t^1_{m_1}+\tau,t^1_{m_1+1}]\subset
330: (t^1_{m_1}+\tau,t^1_{m_1+1}+\tau)$ and
331: $\phi_2(t^1_{m_1}+\tau)<\phi_1(t^1_{m_1}+\tau)$, similarly as Case 2
332: we can get $\phi_2(t)<\phi_1(t)$ for all
333: $t\in(t^1_{m_1}+\tau,t^1_{m_1+1}]$. It implies
334: $t^1_{m_1+1}<t^2_{m_2+1}$.
335:
336: \noindent {\bf Case 4}: $t^2_{m_2}<t^1_{m_1}+\tau$ and
337: $\phi_1(t^1_{m_1}+\tau)\leq\phi_2(t^1_{m_1}+\tau)$.
338:
339: Since $(t^2_{m_2},t^1_{m_1}+\tau)\subset(t^1_{m_1},t^1_{m_1}+\tau)$,
340: by Theorem 1 oscillator $2$ cannot receive any spikes from
341: oscillator $1$ in $(t^2_{m_2},t^1_{m_1}+\tau)$. This, combined with
342: $0=\phi_2((t^2_{m_2})^{+})<\phi_1(t^2_{m_2})<1$, leads to
343: $\phi_2((t^1_{m_1}+\tau)^{-})<\phi_1((t^1_{m_1}+\tau)^{-})$. Let
344: $f_0=f(\phi_1((t^1_{m_1}+\tau)^{-}))-f(\phi_2((t^1_{m_1}+\tau)^{-}))$.
345: Because the spike emitted by oscillator $1$ at $t^1_{m_1}$ reaches
346: oscillator $2$ at $t^1_{m_1}+\tau$, we have
347: $f(\phi_2(t^1_{m_1}+\tau))-f(\phi_1(t^1_{m_1}+\tau))=\varepsilon_{12}-f_0$.
348: It can be claimed that
349: $f(\phi_2((t^2_{m_2}+\tau)^{-}))-f(\phi_1((t^2_{m_2}+\tau)^{-}))<\varepsilon_{12}-f_0$.
350: Indeed, this comes from the property (A5) in \cite{Ernst1998}:
351: $f(\theta_2)-f(\theta_1)>f(\theta_2+\delta)-f(\theta_1+\delta)$, if
352: $\theta_1<\theta_2$ and $0<\delta\leq1-\theta_2$. Denoting $\Delta
353: t=t^2_{m_2}-t^1_{m_1}$, from the property (A5) in \cite{Ernst1998}
354: we get
355: $f(\phi_2((t^2_{m_2}+\tau)^{-}))-f(\phi_1((t^2_{m_2}+\tau)^{-}))
356: =f(\phi_2(t^1_{m_1}+\tau)+\Delta t)-f(\phi_1(t^1_{m_1}+\tau)+\Delta
357: t)
358: <f(\phi_2(t^1_{m_1}+\tau))-f(\phi_1(t^1_{m_1}+\tau))=\varepsilon_{12}-f_0$.
359: Because the spike emitted by oscillator $2$ at $t^2_{m_2}$ reaches
360: oscillator $1$ at $t^2_{m_2}+\tau$, we have
361: $f(\phi_1(t^2_{m_2}+\tau))=\min[1,f(\phi_1((t^2_{m_2}+\tau)^{-}))+\varepsilon_{21}]$.
362: So, if $f(\phi_1(t^2_{m_2}+\tau))=1$, then from Theorem 1 it follows
363: that $f(\phi_2(t^2_{m_2}+\tau))<1=f(\phi_1(t^2_{m_2}+\tau))$; if
364: $f(\phi_1(t^2_{m_2}+\tau))<1$, then from the above claim it follows
365: that $f(\phi_2(t^2_{m_2}+\tau))-f(\phi_1(t^2_{m_2}+\tau))=
366: f(\phi_2((t^2_{m_2}+\tau)^{-}))-f(\phi_1((t^2_{m_2}+\tau)^{-}))-\varepsilon_{21}
367: <\varepsilon_{12}-f_0-\varepsilon_{21}=-f_0<0$. It implies
368: $t^1_{m_1+1}<t^2_{m_2+1}$.
369:
370: In fact, we proved that for the case of $N=2$, if
371: $t^1_{m_1}<t^2_{m_2}$, then $t^1_{m_1+1}<t^2_{m_2+1}$. For the case
372: of $N>2$, the proof is similar, and also can be divided into the
373: above four cases. The distinction is that when $N>2$,
374: $t^1_{m_1+1}=t^2_{m_2+1}$ may happen in Cases 2-4. This derives from
375: the fact that two oscillators are likely to desynchronize, while the
376: other oscillators try to synchronize them \cite{Ernst1998}.
377:
378: Numerical analysis shows that from any initial phases, the coupled
379: system approaches a period solution with groups of synchronized
380: oscillators \cite{Ernst1995,Ernst1998,Timme2002-1,Wu2007,Kim2004}.
381: In larger networks, the oscillators can be divided into groups in a
382: combinatorial number of ways, and exponentially many periodic
383: solutions are present \cite{Timme2002-1}, which greatly increases
384: the complexity of firing behaviors. Properties 1-3 indicate that the
385: firing behaviors of the coupled system at
386: $(\tau,\varepsilon)\in\mathcal{A}_1$ are relatively simple. However,
387: when parameters $(\tau,\varepsilon)\in\mathcal{A}_2$, there may be
388: some solutions, which do not possess some or all of Properties 1-3.
389: It makes firing behaviors more complicated. Whether or not such
390: solutions exist depends on parameters $(\tau,\varepsilon)$ and
391: coupling strengths $\varepsilon_{ij}$. For the system with
392: $(\tau,\varepsilon)\in\mathcal{A}^0_2:
393: =\{(\tau,\varepsilon)\in\mathcal{A}|f(\tau)+\varepsilon>1\}$ and
394: $\varepsilon_{ij}=\varepsilon/(N-1)$, $i\neq j$ (hereinafter
395: referred to as all-to-all coupling), such solutions always exist.
396: Furthermore, for any such solution, there must be some interspike
397: intervals not exceeding the delay time $\tau$. Otherwise, by
398: previous arguments, the solution possesses Properties 1-3. As an
399: example, we simulate a network of $N=4$ all-to-all coupled
400: oscillators with $\tau=0.9$, $\varepsilon=0.6$. We use for $f$ an
401: example of the leaky integrate-and-fire model
402: \begin{eqnarray}
403: \frac{\mathrm{d}f(\phi)}{\mathrm{d}\phi}=-\bigg(\ln\frac{I}{I-1}\bigg)\cdot
404: f(\phi)+I\cdot\ln\frac{I}{I-1} \label{IF}
405: \end{eqnarray}
406: where $I=1.05$. In Fig. \ref{fig:fig1}(a), a period-four solution is
407: given. In this solution, the firing order of oscillators $1,2$ (or
408: $3,4$) is not preserved, e.g., $t^1_3<t^2_4$ but $t^1_5>t^2_6$; and
409: the instantaneous synchronization $t^i_{m_i}=t^j_{m_j}$ does not
410: mean $\phi_i(t)=\phi_j(t)$ for all $t\geq t^i_{m_i}$, e.g.,
411: $t^1_4=t^2_5$ but $t^1_5>t^2_6$. In Fig. \ref{fig:fig1}(b), a
412: period-two completely synchronized solution is given. In addition,
413: one can see that in Fig. \ref{fig:fig1}(a) and (b), most interspike
414: intervals of the oscillators are shorter than the delay time
415: $\tau=0.9$.
416:
417: \setlength{\unitlength}{1in}
418: \begin{figure}[!t]
419: \includegraphics[width=2.75in]{fig_1a.eps}
420: \put(-3,0.8){\footnotesize (a)}
421:
422: \vspace{1ex}
423: \includegraphics[width=2.75in]{fig_1b.eps}
424: \put(-3,0.8){\footnotesize (b)}
425:
426: \caption{\label{fig:fig1} Firing times of four all-to-all
427: pulse-coupled oscillators with $\tau=0.9$ and $\varepsilon=0.6$.
428: Vertical dashed lines are used to indicate the boundaries of
429: periods. (a) Initial phases
430: $[\phi_1(0),\phi_2(0),\phi_3(0),\phi_4(0)]=[0.1766,0.4298,0.4079,0.7061]$.
431: Two periods (i) and (ii) are presented. (b) Initial phases
432: $[\phi_1(0),\phi_2(0),\phi_3(0),\phi_4(0)]=[0.4974,0.2492,0.8932,0.8501]$.
433: Four periods (i), (ii), (iii) and (iv) are presented.}
434: \end{figure}
435:
436: \setlength{\unitlength}{1in}
437: \begin{figure}
438: \includegraphics[width=2.25in]{fig_2.eps}
439: \put(-2.5,1.85){\footnotesize(a)}
440: \put(-2.5,1.2){\footnotesize(b)}
441: \put(-2.5,0.55){\footnotesize(c)}
442:
443: \vspace{-1ex}\caption{\label{fig:fig2} Dependence of
444: $P_N(\tau,\varepsilon)$ on network size $N$. (a) $\tau=0.55$,
445: $\varepsilon=0.4$. (b) $\tau=0.7$, $\varepsilon=0.35$. (c)
446: $\tau=0.8$, $\varepsilon=0.3$.}
447:
448: \vspace{3ex}
449: \includegraphics[width=2.25in]{fig_3.eps}
450: \vspace{-1ex}\caption{\label{fig:fig3} Prevalence of completely
451: synchronized solutions for $(\tau,\varepsilon)\in\mathcal{A}_2$.
452: Parameters with $P_{100}(\tau,\varepsilon)>0$ are marked in black.
453: The dashed curve represents
454: $\{(\tau,\varepsilon)\in\mathcal{A}|f(\tau)+\varepsilon=1\}$.}
455: \end{figure}
456:
457: Completely synchronized solutions, as a special type of periodic
458: solutions, have been widely studied
459: \cite{Mirollo1990,Mathar1996-Goel2002,Ernst1995,Ernst1998,Timme2002-2,Kim2004,Wu2007}.
460: The following analysis demonstrates the different synchronization
461: behaviors for networks with $(\tau,\varepsilon)\in\mathcal{A}_1$ and
462: $(\tau,\varepsilon)\in\mathcal{A}_2$. In \cite{Wu2007}, we proved
463: that under the assumption $f(2\tau)+\varepsilon<1$, from any initial
464: phases (other than $\phi_1(0)=\cdots=\phi_N(0)$), all-to-all
465: pulse-coupled oscillators with delayed excitatory coupling cannot
466: achieve complete synchronization. In fact, we can extend this result
467: to the case of $(\tau,\varepsilon)\in\mathcal{A}_1$ (see
468: \cite{remark2}). Interestingly, we found that when parameters
469: $(\tau,\varepsilon)\in\mathcal{A}_2$, completely synchronized
470: solutions become prevalent. In order to exhibit this, for networks
471: with all-to-all coupling, we numerically estimate the fraction
472: $P_N(\tau,\varepsilon)$ of the phase space
473: $\mathbf{\Phi}:=\{(\phi_1,\ldots,\phi_N)|0<\phi_i\leq1\}$ occupied
474: by initial phases of completely synchronized solutions. We still use
475: (\ref{IF}) for $f$. Fig. \ref{fig:fig2}(a)-(c) show the dependence
476: of $P_N(\tau,\varepsilon)$ on $N$ for $\tau=0.55$,
477: $\varepsilon=0.4$; $\tau=0.7$, $\varepsilon=0.35$; and $\tau=0.8$,
478: $\varepsilon=0.3$, respectively. More generally, we observed that
479: $P_N(\tau,\varepsilon)$ converges to a constant depending on
480: $(\tau,\varepsilon)$ as $N$ goes to infinity. For networks of
481: $N=100$, Fig. \ref{fig:fig3} shows the region of parameter space
482: $\mathcal{A}$ where completely synchronized solutions appear
483: ($P_{100}(\tau,\varepsilon)>0$). For completely synchronized
484: solutions, there must be some interspike intervals not exceeding the
485: delay time $\tau$. Otherwise, the system cannot be completely
486: synchronized (see \cite{Wu2007,remark2}). The performance of Fig.
487: \ref{fig:fig3} is supported by the observation \cite{Buck1988} of
488: flashing patterns of two firefly species {\it Photinus pyralis} and
489: {\it Pteroptyx malaccae}. For the species {\it P. pyralis}, the
490: normalized delay (neural delay/endogenous flashing period) is
491: $\approx0.03$. The whole group of the species rarely synchronizes
492: flashing; instead, wave, chain or sweeping synchrony has been
493: reported. For the species {\it P. malaccae}, the normalized delay is
494: $\approx0.36$, and perfect synchrony is usually achieved.
495:
496: In summary, our analysis demonstrates different dynamics for
497: pulse-coupled networks with $(\tau,\varepsilon)\in\mathcal{A}_1$ and
498: $(\tau,\varepsilon)\in\mathcal{A}_1$. For the region
499: $\mathcal{A}_1$, we derive a low bound of interspike intervals and
500: three firing properties, which provide a basis for future researches
501: addressing the dynamics in networks, e.g., stability of periodic
502: solutions. The difference of synchronization presented at the end of
503: the Letter is useful for understanding and interpreting
504: synchronization phenomena in some natural systems.
505:
506: This work was supported by the National Science Foundation of China
507: under Grants 60574044 and 60774074.
508:
509: \begin{thebibliography}{99}
510: \bibitem{relationship-to-natual-systems}
511: T. J. Walker, Science {\bf 166}, 891 (1969); C. S. Peskin,
512: Mathematical Aspects of Heart Physiology (Courant Institute of
513: Mathematical Sciences, New York, 1975); L. F. Abbott and C. van
514: Vreeswijk, Phys. Rev. E {\bf 48}, 1483 (1993); A. V. M. Herz and J.
515: J. Hopfield, Phys. Rev. Lett. {\bf 75}, 1222 (1995); A. V. M. Herz
516: and J. J. Hopfield, Phys. Rev. Lett. {\bf 75}, 1222 (1995).
517:
518: \bibitem{Buck1988}
519: J. Buck, Q. Rev. Biol. {\bf 63}, 265 (1988).
520:
521: \bibitem{Mirollo1990}
522: R. E. Mirollo and S. H. Strogatz, SIAM J. Appl. Math. {\bf 50}, 1645
523: (1990).
524:
525: \bibitem{Strogatz1983}
526: S. H. Strogatz and I. Stewart, Scientific American {\bf 269}, 68
527: (1983).
528:
529: \bibitem{Vreeswijk1993-Chen1994-Corral1995}
530: C. van Vreeswijk and L. F. Abbott, SIAM J. Appl. Math. {\bf 53}, 253
531: (1993); C. C. Chen, Phys. Rev. E {\bf 49}, 2668 (1994); A. Corral
532: {\it et al}, Phys. Rev. Lett. {\bf 74}, 118 (1995).
533:
534: \bibitem{Mathar1996-Goel2002}
535: R. Mathar and J. Mattfeldt, SIAM J. Appl. Math. {\bf 56}, 1094
536: (1996); P. Goel and B. Ermentrout, Physica D {\bf 163}, 191 (2002).
537:
538: \bibitem{Ernst1995}
539: U. Ernst, K. Pawelzik, and T. Geisel, Phys. Rev. Lett. {\bf 74},
540: 1570 (1995).
541:
542: \bibitem{Ernst1998}
543: U. Ernst, K. Pawelzik, and T. Geisel, Phys. Rev. E {\bf 57}, 2150
544: (1998).
545:
546: \bibitem{Timme2002-1}
547: M. Timme, F. Wolf, and T. Geisel, Phys. Rev. Lett. {\bf 89}, 154105
548: (2002).
549:
550: \bibitem{Timme2002-2}
551: M. Timme, F. Wolf, and T. Geisel, Phys. Rev. Lett. {\bf 89}, 258701
552: (2002).
553:
554: \bibitem{Kim2004}
555: D. E. Kim, BioSystems {\bf 76}, 7 (2004).
556:
557: \bibitem{Wu2007}
558: W. Wu and T. P. Chen, Nonlinearity {\bf 20}, 789 (2007).
559:
560: \bibitem{remark1}
561: In Lemma 1 of \cite{Wu2007}, we prove that if the coupling is
562: all-to-all and parameters $\tau$, $\varepsilon$ satisfy
563: $f(2\tau)+\varepsilon<1$, then all interspike intervals of each
564: oscillator in the coupled system are longer than $2\tau$. For
565: normalized coupling and parameters
566: $(\tau,\varepsilon)\in\mathcal{A}_1$, the proof is simalar.
567:
568: \bibitem{remark2}
569: The proof is almost the same as that in \cite{Wu2007}. Here, we
570: briefly describe the proof process. By Theorem 1 in this Letter,
571: Lemma 1 in \cite{Wu2007} becomes that if an oscillator fires at time
572: $t_1$ and $t_2$ with $t_1\neq t_2$, then $|t_1-t_2|>\tau$. Although
573: the result of Lemma 1 is weakened, Lemmas 2-7 and Theorem 1 in
574: \cite{Wu2007} still hold. Moreover, all the derivations need not be
575: changed except that of Lemma 7. For the proof of Lemma 7, an
576: additional but straightforward case distinction is required.
577:
578: \end{thebibliography}
579:
580: \end{document}
581: