nlin0503051/paper.tex
1: \documentclass[twocolumn,aps,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
2: %\documentclass[preprint,aps,showpacs,preprintnumbers,amsmath,amssymb]{revtex4}
3: \usepackage{graphicx}
4: \usepackage{dcolumn}
5: \usepackage{bm}
6: 
7: \begin{document}
8: \linespread{1.6}
9: 
10: \title{Resemblances and differences in mechanisms of noise-induced resonance} 
11: 
12: \author{R.~Centurelli$^{1}$, 
13:         S.~Musacchio$^{1,2}$,
14:         R.A.~Pasmanter$^{3}$ and     
15:         A.~Vulpiani$^{1,2,4}$}
16: 
17: \affiliation{$^{1}$ Department of Physics, 
18: University La Sapienza, P.le A. Moro 2, I-00185 Roma, Italy.  }
19: \affiliation{$^{2}$ INFM (UdR and CSM) - Unit\`a di Roma La Sapienza.} 
20: \affiliation{$^{3}$ K.N.M.I., P.O.Box 307, 3730 AE, De Bilt, The Netherlands}
21: \affiliation{$^{4}$ INFN - Sezione di Roma La Sapienza.}
22: 
23: \date{\today}
24: 
25: \begin{abstract}
26: 
27: Systems showing stochastic resonance (SR) 
28: or coherent resonance (CR) share some features, 
29: in particular the nearby periodic character of the signal. 
30: We show that in spite of this resemblance the different underlying 
31: dynamics can be detected in experimental data
32: by studying the histogram of inter-spikes
33: times and some statistical properties like two-times correlation 
34: functions.
35: We discuss the possible relevance for climate modeling. 
36: 
37: \end{abstract}
38: 
39: \pacs{PACS number(s)\,: 05.10.Gg, 92.60.Ry }
40: 
41: \maketitle
42: 
43: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
44: \section{Introduction} 
45: 
46: The mechanism of stochastic resonance (SR) was initially introduced as a possible
47: explanation of climate changes on long time-scales~\cite{1}.
48: During the last two decades it has been applied to a wide class of systems
49: ranging from analog circuits, neurobiology, ring lasers, systems
50: with colored noise, etc; for a review see Ref.~\cite{2}.
51: 
52: The prototypical system showing SR, which is also the original one used to
53: model climate changes, is the stochastic differential
54: equation
55: \begin{equation}
56:   \frac{dx}{dt}=
57:  -\frac{\partial V(x,t)}{\partial x}
58:  +\sqrt{2 D}\eta \;,
59: \label{eq:1.1}
60: \end{equation}
61: where $\eta$ is a Gaussian, white noise with
62: $\langle \eta(t) \rangle = 0 $ and  
63:  $\langle \eta(t) \eta (t') \rangle = \delta (t - t')$,
64: $D$ measures the noise intensity
65: and $V(x,t)$ a is double well potential with a time periodic term
66: \begin{equation}
67: V(x,t) = \frac{x^4}{4} -\frac{x^2}{2} + A x \cos(2\pi t/T) \,\, .
68: \label{eq:1.2}
69: \end{equation}
70: In the case of a stationary potential, i.e., $A=0$, the jumps between the
71: two minima at $x=-1$ and $x=1$ are independent events whose probability
72: distribution is approximately Poissonian~\cite{R1}.
73: Using simple arguments based on the Kramers exit-time formula~\cite{3},
74: it can be shown that there is range of values of $D, \, T$
75: and $A$ where SR is present, i.e., the jumps between 
76: the two minima (close to $-1$ and $+1$ if $A$ is sufficiently small)
77: are strongly synchronized with the forcing
78: and that 
79: the probability distribution function (PDF) of the jumping 
80: time $\tau$ has a relatively sharp
81: peak around $T$ ~\cite{1,2}.
82: 
83: The phenomenon of SR provides one example of the nontrivial
84: role that noise can play in dynamical systems with an external periodic 
85: forcing.
86: Besides SR, there exist other 
87: examples of the ``constructive role'' of noise, e.g., 
88: one can have a synchronization of trajectories generated 
89: by different initial conditions and 
90: the same noise realization~\cite{R2}. Our interest will focus on cases 
91: where noise can enhance periodic behavior, e.g., the so-called
92: coherent resonance (CR) and the noise-induced dynamics in systems
93: with time delay (ND).
94: 
95: The phenomenon of CR~\cite{4}  has been found in models describing excitable
96: systems that occur in different fields like chemical reactions, neuronal
97: and other biological processes~\cite{5,5b}. 
98: The prototypical stochastic differential
99: equation used in this case is the FitzHugh-Nagumo system defined by:
100: \begin{eqnarray}
101:  \epsilon \frac{dx}{dt} & = &  x-\frac{x^3}{3} -y 
102:  \label{eq:1.3} \\
103:  \frac{dy}{dt}  & = &  x+a+\sqrt{2D}\eta \,\, 
104: \label{eq:1.4}
105: \end{eqnarray}
106: with $\epsilon \ll 1$ so that the time evolution of $x$ is much faster
107: than that of $y.$
108: For $|a|>1$ there is a stable fixed point, for $|a|<1$ 
109: there is an unstable fixed point and a limit cycle.
110: The cycle consists of two pieces of slow motion connected
111: by a fast jump.
112: If $|a|$ is slightly larger than $1$ the system is excitable~\cite{4} , i.e.,
113: small deviations from the fixed point may generate large pulses (also
114: called spikes)~\cite{6}.
115: Moreover, in this case, one finds that there is a range of values
116: of the noise intensity $D$ such that CR appears, i.e., roughly periodic
117: noise-excited oscillations are present, resembling the SR oscillations~\cite{4}.
118: 
119: The prototypical example for ND~\cite{7} is the over-damped particle motion
120: in the double-well potential $V(x(t),x(t-T))$:
121: \begin{eqnarray}
122:   { \frac{dx(t)}{dt} } & = & 
123:  -{ \frac{\partial V(x(t),x(t-T))}{\partial x(t)} }
124:  +\sqrt{2 D}\eta \nonumber \\
125:  & = & x(t) - x(t)^3 - A x(t-T) +\sqrt{2D}\eta
126: \label{eq:1.5}
127: \end{eqnarray}
128: where $T$ is the delay.
129: It is not very difficult to realize 
130: that the delay term $Ax(t-T)$ has a role similar to that of
131: the periodic forcing in Eq.~(\ref{eq:1.1} , \ref{eq:1.2}).
132: Accordingly, in a certain range of parameters' values , there is a
133: sort of periodic motion
134: with period $T$ or $2T$ (this depends upon the sign of $A$).
135: This ND equation had been proposed as a model for 
136: some climate changes~\cite{R3}. 
137: 
138: %%%%%%%%%%%%%%%%%%%%%%%%
139: \begin{figure}[t]
140: \begin{center}
141: \includegraphics[scale=0.7]{Figure1.eps}
142: \caption{PDF of the inter-spikes time 
143: in the system described by Eq.~(\ref{eq:1.1}) 
144: showing stochastic resonance.
145: The external periodic force has period $T=100$
146: and amplitude $A = -0.15$. The noise intensity is at the
147: corresponding optimal value $D^* = 0.10$. 
148: }
149: \label{fig:PdfSR}
150: \end{center}
151: \end{figure}
152: %%%%%%%%%%%%%%%%%%%%%%%%
153: %%%%%%%%%%%%%%%%%%%%%%%%
154: \begin{figure}[t]
155: \begin{center}
156: \includegraphics[scale=0.7]{Figure2.eps}
157: \caption{PDF of the inter-spikes time 
158: in the system described by Eq.~(\ref{eq:1.5}).
159: The amplitude of the delay term is $A = - 0.15$, 
160: and the delay time is $T = 100$. 
161: The noise intensity is at the corresponding 
162: optimal value $D^* = 0.10$. 
163: }
164: \label{fig:PdfND}
165: \end{center}
166: \end{figure}
167: %%%%%%%%%%%%%%%%%%%%%%%%
168: 
169: Although SR, CR and ND, are similar phenomena,
170: in the sense that their time evolution is nearly periodic,
171: there are also some important differences. For example:
172: a)
173: due to the presence of the term $A x(t-T),$ Eq.~(\ref{eq:1.5})  is
174: in fact an infinite dimensional system since in order to determine
175: $x(t)$ for $t>0$ one has to specify $x(t')$ with $-T \le t' \le 0$.
176: On the contrary for Eq.~(\ref{eq:1.1}) it is sufficient to know $x(0)$.
177: b)
178: in the case of SR the periodicity is due to the external forcing
179: while in the CR case the periodicity has an internal origin, i.e.,
180: the periodic motion is due to the intrinsic dynamics
181: and, at variance with SR, its period cannot be changed
182: by tuning external control parameters. This difference can play an
183: important role in, e.g., the context of climate changes and glaciation,
184: for more details, refer to the last Section.
185: 
186: The aim of this paper is to analyse the differences among
187: SR, ND and CR and their possible relevance to applications.
188: In Sect II we will briefly review some properties of the PDF 
189: of the inter-spikes times for SR, ND and CR.
190: In particular, we recover a recent result~\cite{8} showing that
191: CR and SR are not conflicting or excluding mechanisms, i.e.,
192: the same  periodically driven system, 
193: e.g., the one given by Eqs.~(\ref{eq:1.3},\ref{eq:1.4})
194:  with a time-dependent parameter
195: $a(t)  = a_0 + a_1 cos(2 \pi t /T)$ can present a transition
196: from SR to CR behavior when the noise intensity $D$ is increased .
197: In Sect III we show that, in spite of some resemblance, SR, ND and
198: CR exhibit different statistical features which, at least in principle,
199: can be detected in experimental data. In particular we have that 
200: for SR the correlation function $C(\tau)$, after a transient period,
201: is periodic and does not relax to zero. 
202: On the contrary, for the CR and ND cases $C(\tau)$ shows damped 
203: oscillations.
204: Sect IV is devoted to general remarks and conclusions. In
205: particular we deal with the potential relevance of the differences
206: between SR, CR and ND to climate modeling.
207: 
208: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
209: \section{Statistics of inter-spikes times}
210: \label{secII}
211: 
212: One basic feature shared by the three models introduced in the
213: previous Section,  
214: is the presence of two characteristic states: two equilibria in the case
215: of SR and ND, a rest state and an excited one in the FitzHugh-Nagumo model. 
216: Jumps between these states are made possible by the noise.
217: Moreover, it turns out that there exists an optimal value of the noise
218: intensity such that this jumping becomes approximately periodic, 
219: i.e., the typical time between two consecutive transitions is 
220: roughly constant. 
221: 
222: According to the terminology of biological systems, where CR 
223: was originally introduced, we will refer to the time interval between 
224: consecutive transitions as the ``inter-spikes time $\tau$''. 
225: In the case of the system~(\ref{eq:1.3},\ref{eq:1.4}) where 
226: the dynamical variables show well defined maxima, 
227: the definition of the inter-spikes time
228:  $\tau$ is rather natural. 
229: In the case of the other two systems~(\ref{eq:1.1},\ref{eq:1.2}) 
230: and~(\ref{eq:1.5}) one can define 
231: $\tau$ as $ \tau =  t_{n+1} -t_n$ where $t_n$ is the $n-$th crossing time, 
232: i.e.,
233: $x(t_n) = 0$ and $ \dot{x} (t_n) > 0 $.  
234: A measure of the signal's periodic character is provided by 
235: the normalized variance $NV$ of inter-spikes times,  
236: $NV = \sqrt{Var(\tau)} / \langle \tau \rangle $.
237: For generic noise intensity the transitions occur at random times, 
238: and $p(\tau)$, the PDF of inter-spikes times, is weakly localized, 
239: i.e., $NV$ is of order 1.  
240: We define the resonant or optimal value of the noise intensity, and
241: denote it by $D^*$, as the value of $D$
242: for which the system has minimal normalized variance $NV$.  
243: 
244: 
245: %%%%%%%%%%%%%%   Fig. (3)     %%%%%%%%%%%%%%%%%%%%%
246: \begin{figure}
247: \begin{center}
248: \includegraphics[scale=0.7]{Figure3.eps}
249: \caption{
250: PDF of the inter-spikes time in the
251: FitzHugh-Nagumo model Eqs.~(\ref{eq:1.3},\ref{eq:1.4})
252: with parameters' values
253: $a_0 = 1.05$,
254: $a_1 = 0.04$,
255: $\epsilon = 0.01$
256: and
257: $T=10 > T_*=3.7$
258: for different noise intensities. From top to bottom,
259: $D = 2.5 \times 10^{-5}$,
260: $D = 4.3 \times 10^{-5}$,
261: $D = 1.1 \times 10^{-4}$,
262: $D = 2.3 \times 10^{-4}$
263: and
264: $D = 4.3 \times 10^{-3}$. 
265: }
266: \label{fig:PdfFN}
267: \end{center}
268: \end{figure}
269: %%%%%%%%%%%%%%%%%%%%%%%%
270: 
271: %%%%%%%%%%%%%%%%%%%%%%%%
272: \begin{figure}
273: \begin{center}
274: \includegraphics[scale=0.7]{Figure4.eps}
275: \caption{
276: As in Fig.~(\ref{fig:PdfFN} )
277: but with 
278: $T=1 < T_*=3.7$.    
279: }
280: \label{fig:isto1}
281: \end{center}
282: \end{figure}
283: %%%%%%%%%%%%%%%%%%%%%%%%
284: 
285: 
286:  
287: In Figure~{\ref{fig:PdfSR}} we show the 
288: PDF of the inter-spikes time interval in the case of SR
289: for the system described by Eq.~(\ref{eq:1.1})
290: at the optimal noise intensity $D^*=0.1$. We see that $p(\tau)$ 
291: is peaked around $\tau = nT$ with 
292: $ n=1,2,.....  $, 
293: and that the envelope is approximately exponential.
294: This feature can be easily explained as follows. 
295: Consider the trajectories $x(t)$, starting at $t=0$ 
296: from the favored well, i.e. $x(0)= 1$ if $A < 0$. 
297: In the case of SR, at $t$ close to $T/4$ many of the trajectories 
298: will jump onto the other minimum at $x= -1$ and after half period they 
299: will jump back again onto $x = 1$. However, a fraction of the trajectories
300: remains in the ``wrong position'' (i.e. in the unfavored well)
301: for $t$ close to $T$.
302: Calling $P$ the probability of this event, we have that  
303: the integral of $p(\tau)$ around $T$, say for $\tau \in [ 0.5 T , 1.5 T ]$, is $(1-P)$.
304: The events with inter-spikes time $\tau \sim nT$ 
305:  correspond to trajectories $x(t)$ 
306: which are in the the ``wrong'' minimum at $T, 2T, ....,nT$.
307: Taking into account the periodicity of the forcing and assuming that
308: the system's memory is much shorter than the external period $T$
309: we have that the probability to have $x(t)$ in the ``wrong'' minimum
310: at  $t \sim kT$ under the condition that $x(t)$ was in  
311: the the ``wrong'' minimum
312: at  $t \sim (k-1)T$, does not depend on the behavior for $t<(k-1)T$.
313: Therefore  the  integral of $p(\tau)$ around $nT$ is 
314: $(1-P)P^{n-1} \sim e^ {-c n}$ with $c=- \ln P$. The envelop
315: of the inter-spikes-interval histogram has recently 
316: been computed by Berglund and Gentz~\cite{BG}
317: in a more general and rigorous setting.
318: 
319: In the case of CR (not shown) and ND, see Figure~\ref{fig:PdfND},
320:  the PDF of the inter-spikes 
321: time interval is peaked around $T$ where $T$ is the characteristic internal 
322: time of the system, namely the delay time in the ND case, 
323: and the period of the limit 
324: cycle in the CR case. 
325: The parameters of the memory term in the
326: ND-model~(\ref{eq:1.5}) have been
327: chosen in order to emphasize the similarities with the SR-model,
328: i.e., the amplitude $A$ 
329: and delay time $T$ are identical to the amplitude 
330: and period of the external forcing in Eq.~(\ref{eq:1.1}).
331: One consequence of this choice is that the intensity of the optimal noise coincides with that in the SR-model. 
332: Notice that, 
333: at variance with the SR case, no peaks are present at multiples of $T$. 
334: 
335: 
336: 
337: In summary, in all three systems there 
338: exists an optimal noise intensity which 
339: %``regularize'' their behavior, i.e.
340: produces a roughly periodic signal $x(t)$. 
341: At this intensity a sharp peak appears in the 
342: inter-spikes time PDF. 
343: In the case of CR and ND the PDF has only one maximum at 
344: $\tau = T$. On the other hand, in the case of SR 
345: other maxima appear at $nT$. This effect can be considered 
346: as one of the distinctive marks of the stochastic resonance.
347: In contraposition to the CR and ND cases, 
348: the periodicity of the signal in the SR case is 
349: induced by an external periodic force, which triggers 
350: the jumps at fixed times $t=nT$ in such a way that 
351: the system synchronizes with this external ``clock".  
352: 
353: %%%%%%%%%%%%%%%%%%%%%%%%
354: \begin{figure}[t!]
355: \begin{center}
356: \includegraphics[scale=0.7,draft=false]{Figure5.eps}
357: \caption{Conditional average 
358: $\langle x(t)|x_0 \rangle$
359: (solid line)
360: and conditional variance
361: $\sigma^2(t) $
362: (dashed line) 
363: in the case of SR for the system described by Eq.~(\ref{eq:1.1}). 
364: The initial position is $x_0 = 1$, i.e., in the favored well.  
365: The external force has period $T=100$
366: and amplitude $A = -0.15$. The noise intensity is at its 
367: optimal level $D = 0.10$.
368: The average is taken over $N=10^4$ realizations of the noise.  
369: } 
370: \label{fig:XSR}
371: \end{center}
372: \end{figure}
373: %%%%%%%%%%%%%%%%%%%%%%%%
374: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
375: % \subsection{Coexistence of CR and SR}
376: 
377: The combined effects of noise and nonlinearity can result 
378: in more complex behavior~\cite{yacomotti,mendez}. 
379: In particular it has recently been shown in~\cite{8}
380: that SR and CR can coexist in the same system.
381: This behavior occurs, e.g., in the model~(\ref{eq:1.3}-\ref{eq:1.4}) 
382: when small oscillations are imposed on the control parameter 
383: $ a(t) = a_0 + a_1 \cos({2 \pi t /T})$. 
384: We consider only $a_0$ and oscillation amplitudes $a_1$
385: such  that the control parameter $a(t)$ never crosses the critical value, 
386: i.e. $ a(t) > 1, \forall t$.  
387: The presence of these small oscillations determines privileged 
388: times $t_n = (2n+1)T/2$  
389: at which the system is closer to the excited state, 
390: i.e., $a(t_n)$ approaches 1 from above,
391: and a noise induced transition is facilitated. Analogously to the case 
392: of the double well system with periodic forcing, 
393: there exists an optimal noise intensity for which a regular,
394: quasi-periodic behavior emerges. More precisely:
395: at low noise intensity, $p(\tau)$ 
396: has rather sharp peaks at $\tau = nT$ with an 
397: approximately exponential envelope, 
398: showing all the features of SR, 
399: see the two upper panels in Figure~{\ref{fig:PdfFN}}.
400: At higher noise intensity, as in the two lower panels in
401: Figure~{\ref{fig:PdfFN}}, the small oscillations
402: in $a(t)$ become irrelevant and the system shows CR 
403: as if the control parameter were fixed at its mean value $a_0$.
404: The shape of $p(\tau)$ behaves accordingly
405: as the noise increases: the maxima of SR diminish and a  
406: single peak with an exponential tail appears at $\tau = T_*$ where $T_*$ is the
407: period of the system's limit cycle. 
408: 
409: Moreover, when the external 
410: force period $T$ is shorter than the internal, limit-cycle period $T_*$ 
411: there are peaks only for $nT \ge T_*$ because once the excited state 
412: is reached the system needs at least a time
413: $T_*$ in order to relax back into the excitable, rest state and restart
414: the whole cycle (see fig.~\ref{fig:isto1}). 
415: 
416: %%%%%%%%%%%%%%%%%%%%%%%%
417: \begin{figure}[t!]
418: \begin{center}
419: \includegraphics[scale=0.7,draft=false]{Figure6.eps}
420: \caption{Same as in Fig.~\ref{fig:XSR}
421: in the case of CR for the system described by 
422: Eqs.~(\ref{eq:1.3},\ref{eq:1.4}) 
423: with $a=1.05$, $\epsilon = 0.01$ and 
424: optimal noise intensity $D = 2.5 \times 10^{-3}$. 
425: The initial position is $x_0 = 2, y_0 = 0.8$, i.e., in the 
426: excited state.  
427: The average is taken over $N=10^4$ realizations of the noise.
428: }  
429: \label{fig:XCR}
430: \end{center}
431: \end{figure}
432: %%%%%%%%%%%%%%%%%%%%%%%%
433: 
434: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
435: \section{Conditional averages and correlation functions}
436: 
437: If one would observe just a single trajectory, SR, CR and ND
438: would appear rather similar since the three cases would
439: present us with a
440: nearly periodic $x(t)$.
441: An analysis based on Fourier spectra, as it is often done,
442: would reinforce this picture.
443: In the previous Section we contrasted the multi-peaked PDF
444: of inter-spikes time in the SR case with the one-peaked PDF
445: in the CR and ND cases. 
446: In this Section we bring to the fore some statistical properties
447: which are present in the SR case but are absent both in the CR and 
448: in the ND cases.
449: 
450: Consider an ensemble of $N$ trajectories
451: $\{ x^{(n)}(t), n=1,...., N \gg 1 \},$  sharing the same initial conditions
452: $x^{(n)}(0)= x_0$, but with different realizations of the noise
453: $\eta (t)$
454: and compute from Eqs.~(\ref{eq:1.1}),~(\ref{eq:1.3}-\ref{eq:1.4}) 
455: and~(\ref{eq:1.5}) 
456: the conditional average
457: $\langle x(t)|x_0 \rangle$, 
458: \begin{equation}
459: \langle x(t)|x_0 \rangle = \frac{1}{N} \sum_{n=1}^N x^{(n)}(t) \;,
460: \label{eq:3.1}
461: \end{equation}
462: and the conditional variance
463: \begin{eqnarray}
464: \sigma^2(t) & = & \langle x^2(t)|x_0 \rangle - \langle x(t)|x_0 \rangle ^2 \nonumber \\
465: & = & \frac{1}{N} \sum_{n=1}^N [x^{(n)}(t) - \langle x(t)|x_0 \rangle ]^2 \;,
466: \label{eq:3.2}
467: \end{eqnarray}
468: 
469: %%%%%%%%%%%%%%%%%%%%%%%%
470: \begin{figure}[t]
471: \begin{center}
472: \includegraphics[scale=0.7]{Figure7.eps}
473: \caption{The periodic correlation function $C(\tau)$ 
474: of the SR model Eq.~(\ref{eq:1.1}).
475: Parameters' values as in Fig.~\ref{fig:PdfSR}
476: } 
477: \label{fig:CorrSR}
478: \end{center}
479: \end{figure}
480: %%%%%%%%%%%%%%%%%%%%%%%%
481: 
482: 
483: In Figs.~\ref{fig:XSR} and \ref{fig:XCR} we show 
484: $ \langle x(t)|x_0 \rangle $ and $\sigma^2(t)$
485: as functions of the time $t$, for the SR case and the CR case
486: respectively. In both cases the noise-intensity 
487: values $D$ are the optimal ones. 
488: In the SR case $\langle  x(t)|x_0 \rangle $ does 
489: not relax to zero at large times $t$
490: and, ignoring the initial transient, it is periodic.
491: The conditional variance $\sigma^2(t)$ reaches its minima
492: when the absolute value of the  conditional  average 
493: $ \langle x(t)|x_0 \rangle $
494: reaches its maxima. The largest values for $\sigma^2(t)$ 
495: are achieved
496: when  $\langle x(t)|x_0 \rangle $ is around zero.
497: In other words, the main uncertainty in the process occurs
498: around $T/4, 3T/4, 5T/4 $ and so on, i.e., when the jumps between
499: the two minima take place.
500: 
501: As it can be seen in Fig~\ref{fig:XCR} the behavior found in the CR
502: case is different.
503: Even with the noise intensity at its optimal value, 
504: after a few damped oscillations both the conditional average and variance 
505: relax to the constant values 
506: $ \langle \langle x(t) \rangle \rangle $ and 
507: $\langle \langle x^2(t) \rangle \rangle  
508: - \langle \langle  x(t)  \rangle \rangle^2$, 
509: where $\langle \langle \cdots \rangle \rangle $ indicates a 
510: time average. 
511: 
512: This behavior underlines the intrinsic difference between the SR and 
513: CR mechanisms. 
514: In the case of SR, the presence of an external synchronizing
515:  force, makes the 
516: transitions from state $+1$ to state $-1$ to occur around 
517: preferred times. 
518: A set of independent replicas, 
519: initially localized in one of the two wells, will therefore 
520: quickly reach a periodic configuration, with the 
521: maximum probability localized in the time-dependent favored well.
522: On the contrary, in the CR case, there are no externally defined
523: preferred times for the transitions, 
524: therefore each replica quickly loses its initial 
525: synchronization with the other ones and after a few periods the 
526: jumps occur at different times for different replicas. 
527: 
528: The correlation function
529: \begin{equation}
530: C(\tau)= 
531: { 
532: \frac 
533: {\langle \langle x(t+\tau)x(t) \rangle \rangle 
534: - \langle \langle x \rangle \rangle^2} 
535: {\langle \langle x^2 \rangle \rangle - 
536: \langle \langle x \rangle \rangle ^2}
537: },
538: \label{eq:3.3}
539: \end{equation}
540: behaves similarly to
541: $\langle x(t)|x_0 \rangle$: in the CR case it relaxes to zero while 
542: in the SR case it remains periodic with non decreasing amplitude, 
543: see Figs.~(\ref{fig:CorrSR}) and (\ref{fig:CorrCR}).
544: If one defines a correlation time $\tau_c$ as
545: \begin{equation}
546: \tau_c = \int_0^{\infty} C(\tau)^2 d \tau ,
547: \label{eq:3.4}
548: \end{equation}
549: one finds that $\tau_c$
550: diverges in the SR case while it remains finite in the CR 
551: case and that,
552: as a function of the noise intensity, 
553: it attains its maximum $\tau_c = 1.03$
554: at the optimal noise intensity value.
555: The behavior in the ND case (not shown) is 
556: qualitatively very similar to that obtained in the CR case. 
557: 
558: %%%%%%%%%%%%%%%%%%%%%%%%
559: \begin{figure}[t!]
560: \begin{center}
561: \includegraphics[scale=0.7]{Figure8.eps}
562: \caption{Correlation function $C(\tau)$ 
563: of the $y$ variable for the FitzHugh-Nagumo 
564: model described 
565: by Eqs.~(\ref{eq:1.3} , \ref{eq:1.4}).
566: Parameters' values as in Fig.~\ref{fig:XCR}
567: }
568: \label{fig:CorrCR}
569: \end{center}
570: \end{figure}
571: %%%%%%%%%%%%%%%%%%%%%%%%
572: 
573: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
574: \section{Conclusions and Discussion}
575: 
576: The numerical study of SR, CR and ND presented in this paper
577: is focused on contrasting statistical features
578: which would be difficult or impossible to detect by simply
579: looking at the spectrum since
580: the spectrum shows a peak that achieves
581: a maximal sharpness for the optimally chosen noise intensities
582: in all the cases we have studied.
583: As we have seen, it is possible to distinguish between the
584: different underlaying dynamical mechanisms by studying 
585: the PDF of inter-spikes times, conditional 
586: averages and time-delayed correlation functions.
587: 
588: We believe that these results are of interest not only in the
589: context of dynamical systems but also in the study of certain
590: climate phenomena.
591: For time scales of order $O(10^5 ys)$ and larger,
592: Milankovich\cite{Milan}
593: has proposed that Earth's climate has been determined by the
594: influx of solar energy to such an extent that the fluctuations
595: in, e.g., the global mean temperature and seasonality 
596: must have been closely
597: correlated with the variations in the incoming energy flux due
598: to the periodicities in Earth's orbit.
599: 
600: Another possible issue, for which the presented results are
601: potentially interesting, is
602: the so-called Dansgaard-Oeschger(DO) events~\cite{DO}
603: which have been inferred from the study of
604: Late-Pleistocene ice cores and marine sediments.
605: These measurements show rapid warmings
606: of the atmosphere followed by a much
607: slower decay back into the average glacial conditions.
608: The warmings took place on a time scale of a few decades while
609: the relaxation back into glacial temperatures lasted some centuries
610: up to millennia.
611: They seem to have occurred at intervals of $1.500 \pm 200$ years
612: or integer multiples hereof~\cite{Schulz}.
613: They were absent during the Holocene, i.e., during the $10^4$ 
614: years before present.
615: The discovery of these rapid warmings led to proposing
616: a number of possible explanations,
617: some of them favoring the internal origin of the period
618: approximately equal to $1.500$ years~\cite{Broecker,Winton,Sakai},
619: while other explanations assume a similar period due to an external
620: astronomical forcing~\cite{Ganopolski}, i.e. a SR-type scenario.
621: In particular, it was shown that in an ocean-circulation box model 
622: and within an appropriate parameters' range,
623: the purely deterministic system has a fixed point and does not show
624: any time dependence while the addition of noise
625: leads to the generation of spikes with a well defined
626: inter-spikes time interval~\cite{Axel},
627: i.e., that coherence resonance is present in this ocean circulation
628: model.
629: As discussed in Section II, now we know that, at least
630: in some systems, it is possible to observe 
631: either CR or SR behavior depending upon the noise
632: intensity.
633: 
634: 
635: We have shown that it is
636: possible to distinguish between the SR and CR, e.g., by looking
637: at time-delayed correlation functions and at the PDF of the inter-spikes
638: times.
639: Needless to say, in order to compute such correlation functions,
640: or the PDF, a sufficiently long and accurate
641: series of measurements is required.
642: From the limited information about the
643: DO events that has been extracted from the geological record
644: it is difficult to decide in favor of one scenario or the other.
645: Indeed, the PDF for the inter-spikes time is qualitatively
646: in agreement with the one observed in the SR case.
647: On the other hand  the time values at which one observes
648: peaks of the PDF do not correspond to known astronomical
649: periods, accordingly, a CR scenario would seem
650: more appropriate.
651: 
652: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
653: \section*{Acknowledgments}
654: We thank A.~Timmermann and A.S.~Pikovsky for stimulating discussions and useful comments.
655: This work has been supported by
656: MIUR-COFIN03 "Complex Systems and Many-Body Problems"
657: (2003020230)  and the  INFM (Statistical Mechanics 
658: and Complexity Center, Rome). 
659: 
660: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
661: 
662: \begin{thebibliography}{99}
663: 
664: \bibitem{1} 
665:      R. Benzi, A. Sutera and A. Vulpiani,
666:      J. Phys. A {\bf 14}, L453 (1981);
667:      C.Nicolis
668:      Tellus {\bf 34}, 1 (1982);
669:      R. Benzi, G. Parisi, A. Sutera and A. Vulpiani,
670:      Tellus {\bf 34}, 10 (1982);
671:      R. Benzi, G. Parisi, A. Sutera and A. Vulpiani,
672:      SIAM J. Appl Math {\bf 43}, 565 (1983).
673: 
674: \bibitem{2} 
675:      K. Wiesenfeld and F. Moss, 
676:      Nature {\bf 373},33 (1995);
677:      L. Gammaitoni, P. H\"anggi, P. Jung, and F. Marchesoni,
678:      Rev. Mod. Phys. {\bf 70}, 223 (1998). 
679: 
680: \bibitem{R1} 
681:      C.~W.~Gardiner
682:      {\it Handbook of Stochastic Methods} (Springer-Verlag, Berlin 1990)  
683: 
684: \bibitem{3}
685:      S. Chandrasekhar,
686:      Rev. Mod. Phys. {\bf 15}, 1 (1943).
687: 
688: \bibitem{R2} 
689:      K. Matsumoto and I. Tsuda, 
690:      J. Stat. Phys. {\bf 31}, 87 (1983).
691: 
692: \bibitem{4}
693:      A.S. Pikovsky and J. Kurths,
694:      Phys. Rev. Lett. {\bf 78}, 775 (1997). 
695: 
696: \bibitem{5}
697:     X. Pei, K. Bachmann and F. Moss,
698:     Phys. Lett. A {\bf 206}, 61 (1995).
699: 
700: \bibitem{5b}
701:     J.J. Collins, C.C. Chow and T.T. Imhoff,
702:     Phys. Rev. E {\bf 52}, R3321 (1995).
703: 
704: \bibitem{6}
705:      A.C. Scott Rev. Mod. Phys. {\bf 47},487 (1975).
706: 
707: \bibitem{7}
708:      L.S. Tsimring and A. Pikovsky, 
709:      Phys.  Rev. Lett. {\bf 87}, 250602 (2001).
710: 
711: \bibitem{R3}
712:      J. P. Pelletier, 
713:      J. Geophys. Res. Atmos. {\bf 108} (D20), 4645 (2003). 
714: 
715: \bibitem{8}
716:      GerardoJ. Escalera Santos, M. Rivera, P. Parmananda,
717:      Phys. Rev. Lett. {\bf 92}, 230601 (2004).
718: 
719: \bibitem{BG}
720:      N.~Berglund and B.~Gentz,
721:      http://www.arxiv.org/abs/cond-mat/0408321 (2004).
722: 
723: \bibitem{yacomotti}
724:     A. M. Yacomotti, M. C. Eguia, J. Aliaga, O. E. Martinez, 
725:     G. B. Mindlin and A. Lipsich, 
726:     Phys. Rev. Lett. {\bf 83}, 292 (1999).
727: 
728: \bibitem{mendez}
729:     J. M. Mendez, J. Aliaga and G. B. Mindlin, 
730:     Phys. Rev. Lett. {\bf 89}, 160601 (2002).
731: 
732: \bibitem{Milan}
733:     W. S. Broecker et al., Science {\bf 159}, 297--300 (1968);
734:     R. A. Kerr, Science {\bf 235}, 973 (1987).
735: 
736: \bibitem{DO}
737:     Dansgaard W., S. J. Johnson,
738:     H. B. Clauser, D. Dahl-Jensen, N. Hammer and C. U. Oeschger,
739:     (1984), Geophys. Monogr. no. 29, Amer. Geophys. Union, 288-298. 
740: 
741: \bibitem{Schulz}
742:     M. Schulz, Paleoceanogr. {\bf 17}, doi:10.1029/2000PA000571 (2002). 
743: 
744: \bibitem{Broecker} 
745:     W. S. Broecker et al., Paleoceanogr. {\bf 5}, 469, (1990) 
746: 
747: \bibitem{Winton} 
748:     M. Winton, in {\it Ice in the climate system}, pp. 417, 
749:     ed. W. R. Peltier, (Springer-Verlag, Berlin 1993).
750: 
751: \bibitem{Sakai}  
752:     K. Sakai and W. R. Peltier, J. Climate {\bf 10}, 949 (1997).
753: 
754: \bibitem{Ganopolski} 
755:     A. Ganopolski, and Rahmstorf, Nature {\bf 409}, 153 (2001). 
756: 
757: \bibitem{Axel}   
758:       Schulz M., A. Paul and A. Timmermann
759:       (2002), Geophys. Res. Letters, 29, doi:10.1029/2002GL016144;
760:       Timmermann A., H. Gildor and E. Tziperman, (2003),
761:       J. Climate, {\bf 16}, 2569 -- 2585. 
762: 
763: 
764: \end{thebibliography}
765: 
766: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
767: 
768: \end{document}
769: