cond-mat0312360/bif.tex
1: \documentclass[epj,final]{svjour}
2: 
3: \usepackage{graphicx}
4: \usepackage{amsmath}
5: \usepackage{amssymb}
6: \usepackage{amscd}
7: 
8: \begin{document} 
9:  
10: \title{Stability analysis of a noise-induced Hopf bifurcation}
11: \author{Kirone Mallick\inst{1} \and Philippe Marcq\inst{2}}                 
12: 
13: \institute{Service de Physique Th\'eorique, Centre d'\'Etudes de Saclay,
14:  91191 Gif-sur-Yvette Cedex, France \\
15: \email{mallick@spht.saclay.cea.fr}\and 
16: Institut de Recherche sur les Ph\'enom\`enes Hors \'Equilibre,
17: Universit\'e de Provence,\\
18: 49 rue Joliot-Curie, BP 146, 13384 Marseille Cedex 13, France\\
19:  \email{marcq@irphe.univ-mrs.fr}}
20: 
21: %\date{\today}
22: \date{August 8, 2003}
23: \abstract{
24: We study analytically and numerically  the  noise-induced transition 
25: between an absorbing and an oscillatory state in a Duffing oscillator 
26: subject to multiplicative, Gaussian white noise. We show in a non-perturbative
27:  manner that a stochastic   bifurcation  occurs when  the 
28: Lyapunov exponent of the linearised system becomes positive.
29:  We deduce from  a simple formula  for the Lyapunov exponent  
30:   the phase diagram of the stochastic Duffing oscillator.
31: The behaviour of  physical observables, such as the oscillator's mean energy, 
32: is studied both close to and far from  the bifurcation.
33: \PACS{
34:       {05.40.-a}{Fluctuation phenomena, random processes, noise, and 
35: Brownian motion}   \and
36:       {05.10.Gg}{Stochastic analysis methods (Fokker-Planck, Langevin, etc.)}
37:    \and      {05.45.-a}{Nonlinear dynamics and nonlinear dynamical systems}   
38:      } 
39: } 
40: 
41: \maketitle 
42: 
43: \section{Introduction}
44: \label{sec:intro}
45: 
46: 
47: 
48:      The stability   of a dynamical system can be strongly   
49: affected  by the presence of uncontrolled
50:  perturbations such as  random noise 
51:  \cite{lefever}.  Noise-induced bifurcations  are of primary
52:   interest in many fields of applied mathematics and physics and   occur 
53:  in various  experimental contexts,  as  for instance:
54:      electronic circuits  \cite{strato},
55:     mechanical oscillators \cite{landaMc},
56:    laser physics  \cite{lefever},
57:  surface waves \cite{fauve}, thermal and electrohydrodynamic
58:  convection in fluids  or liquid crystals \cite{kramer1,kramer2,behn} and 
59:  diffusion in  random media \cite{frisch}. In nonlinear dynamical systems,
60:  the  interplay of noise and  nonlinearity
61:   may produce  unusual  phenomena: 
62:  noise can shift   the  bifurcation  threshold 
63:  from one phase to another \cite{luecke},
64:  it can  induce new  phase-transitions in spatially-extended systems
65: \cite{vandenb1}, or create spatial patterns 
66: \cite{vandenb2} (for a recent review see \cite{toral}).
67: 
68:  One of the simplest systems that may serve  as a pa\-ra\-digm  for
69:  noise-induced transitions is the  nonlinear oscillator with
70:  parametric noise \cite{lindenberg,landa,drolet}, i.e., 
71: with a frequency that
72:  fluctuates randomly with time around  a given  mean value
73:  \cite{bourret,bourretFrisch}.  
74:  In a recent work \cite{philkir1,philkir2}, we studied  the time
75:   asymptotic behaviour  of such an oscillator in the small damping limit. 
76:  We showed  that  physical observables grow  algebraically  with time
77:   until the dissipative time scale is reached;
78:  then  the system  settles  into an oscillatory 
79:  state described by a stationary probability distribution function (PDF)
80:  of the energy $E$: the zero equilibrium  point  ($E = 0$) 
81:  is unstable in the small damping limit. In the present work,
82:  we investigate the stability of the origin for arbitrary 
83:  damping rate and noise strength.
84: 
85:  For  random   dynamical systems,  it   was recognized early on that
86:   various `naive'  stability criteria \cite{bourret,bourretFrisch}, 
87: obtained by linearizing the  
88: dynamical equation  around the  origin, lead to ambiguous results.  
89: (This is in  contrast with  the  deterministic  case    for which
90:  the bifurcation threshold  is  obtained  without ambiguity
91:  by studying the eigenstates of  the linearized  equations
92:   \cite{manneville}.) 
93:   In fact,   it was conjectured in \cite{bourretFrisch}
94:  and proved in  \cite{lindenberg3} that,   in a linear oscillator 
95: with arbitrarily small parametric noise, all moments beyond a certain 
96: order diverge in the long time limit.  Thus,  any   criterion based on 
97:  finite mean  displacement,  momentum,  or energy of the  linearized dynamical
98:  equation   is not  adequate   to insure global stability: the bifurcation 
99: threshold  of a  nonlinear random dynamical system cannot be determined  
100: simply   from  the moments of the linearized system. 
101:  In practice, the transition point is usually calculated 
102:  in a perturbative manner by using weak noise expansions
103:  \cite{landaMc,luecke,landa,drolet}.
104: 
105:  
106: In the present  work, we shall obtain in a non-per\-tur\-ba\-tive manner
107: the bifurcation threshold of a stochastic Duffing oscillator 
108: with multiplicative, Gaussian white noise. 
109: We use a technique described in \cite{philkir1} where
110: the stochastic dynamical equations are expressed
111: in terms of energy-angle variables. 
112: Thanks to a detailed analysis of the associated Fokker-Planck equation,
113: we shall  show  that the bifurcation occurs precisely
114: where  the Lyapunov exponent of the \emph{linear} oscillator changes sign.
115: This  finding is confirmed by numerical simulations. The stability of 
116: the fixed point $E=0$ of the  (nonlinear)  Duffing oscillator 
117: with parametric noise  is  deduced, in this sense, 
118: from the linearized  dynamics.
119: We shall also derive the scaling behaviour of physical 
120: observables in the vicinity of the
121: bifurcation with respect to the distance to threshold.
122: Finally, we study the behaviour of observables far 
123: from the bifurcation, in the small damping or strong noise limit.
124: 
125: 
126:     A rigorous mathematical theory of random dynamical systems
127:  has been recently    developed  and  theorems relating 
128:  the stability of the solution of a stochastic differential
129:  equation to the sign of Lyapunov exponents have
130:  been proved \cite{arnold}.  In the present work,  we do not
131:  rely upon the sophisticated tools of the general mathematical
132:  theory but follow an intuitive approach, easily accessible
133:  to physicists, based upon a factorisation hypothesis of the
134:  stationary PDF in the limit of small and large energies. 
135: 
136:  This article is organized  as follows. In Section~\ref{sec:Duffing},
137:  we  define the system considered, present the relevant phenomenology,
138: and introduce  energy-angle variables for the stochastic
139: Duffing oscillator.
140:   We   study the stationary PDF of the energy in Section~\ref{sec:pdf}.
141: In Section~\ref{sec:lyapunov}, we derive the stability criterion for 
142: the stochastic Duffing oscillator and obtain its phase diagram.
143:  In Section~\ref{sec:scaling}, we  perform a local analysis  
144:  of the  physical observables in the vicinity of and far from
145:  the bifurcation threshold.  Section~\ref{sec:disc} is devoted
146:  to some concluding remarks. In Appendix~\ref{sec:identity}, we  prove some  useful 
147:  identities  satisfied by the  Lyapunov exponent.
148:  In Appendix~\ref{sec:pdfangle}, an explicit formula for 
149:  the  Lyapunov exponent is derived.
150: 
151: \begin{figure}
152: \includegraphics*[width=0.75\columnwidth]{fig1a.eps}
153: \bigskip
154: 
155: \includegraphics*[width=0.75\columnwidth]{fig1b.eps}
156: \bigskip
157: 
158: \includegraphics*[width=0.75\columnwidth]{fig1c.eps}
159: 
160: \caption{Phase plane plots of a typical trajectory of the noisy Duffing
161: oscillator. Eq.~(\ref{dissipDuff}) is integrated numerically
162: with a time step $\delta t = 5 \ 10^{-4}$. (a) $\alpha = 0.5$,
163: $\Delta = 20$, $10 \le t \le 100$; (b) $\alpha = 2.0 $,
164: $\Delta = 20$, $t \le 100$; (c) $\alpha = 2.0$,
165: $\Delta = 60$, $10 \le t \le 100$. Early-time relaxation
166: towards noisy oscillations has been omitted for clarity in cases
167: (a) and (c). The trajectory spirals down towards the globally stable origin 
168: in case (b).}
169: \label{fig:traj} 
170: \end{figure}
171: 
172: \section{The Duffing oscillator with multiplicative white noise}
173: \label{sec:Duffing}
174: 
175: 
176: \subsection{Notations and phenomenology}
177: \label{sec:Duffing:phen}
178: 
179:  A nonlinear oscillator with a  randomly varying  frequency due  to 
180:   external noise can  be  described   
181: by the following equation 
182: \begin{equation}
183: \frac{\mathrm{d}^2 x}{\mathrm{d} t^2} + 
184:   \gamma \frac{\mathrm{d} x}{\mathrm{d} t}
185:   + ( \omega^2 + \xi(t)) x +
186:  \frac { \partial{\mathcal U}(x)}{\partial x}   = 0 \, ,
187:  \label{eqsto}
188: \end{equation}
189: where $x(t)$ is the position of the oscillator at time $t$,
190: ${\mathcal U}(x)$   an anharmonic confining potential and  $\gamma$
191:  the (positive) friction coefficient.
192:  The  linear frequency of the oscillator has a mean value 
193:  $\omega$  and its  fluctuations  are modeled   by 
194:  a  Gaussian white noise  $\xi(t)$ of zero mean-value
195:  and of amplitude  ${\mathcal D}$ 
196:  \begin{eqnarray}
197:        \langle \xi(t)  \rangle &=&   0   \, ,\nonumber \\
198:  \langle \xi(t) \xi (t') \rangle &=& {\mathcal D} \, \delta(t - t')   \, .  
199:  \label{statxi}
200: \end{eqnarray} 
201: In this work, all stochastic differential equations are interpreted
202: according to Stratonovich calculus.  We shall  study the   Duffing oscillator
203:  with multiplicative noise,    the confining
204:  potential ${\mathcal U}$  being given by 
205: \begin{equation}
206: {\mathcal U}(x) = \lambda \frac{x^4}{4} \,.
207: \end{equation} 
208:  By  rewriting  time  and amplitude in dimensionless units,
209:   $ t := \omega t$  and   $ x:= \lambda^{1/2} \omega^{-1} x,$
210:     respectively, 
211:   Eq.~(\ref{eqsto})  becomes 
212: \begin{equation}
213: \frac{\mathrm{d}^2 x}{\mathrm{d} t^2} + 
214: \alpha  \, \frac{\mathrm{d} x}{\mathrm{d} t} + x + x^3 = x \, \Xi(t) ,
215:  \label{dissipDuff}
216: \end{equation}  
217: where  $\Xi(t)$ is a delta-correlated Gaussian  variable:
218:   \begin{eqnarray}
219:        \langle \Xi(t)  \rangle &=&   0   \, ,\nonumber \\
220:  \langle \Xi(t) \Xi (t') \rangle &=& \Delta \, \delta(t - t')   \, .  
221:  \label{defXi}
222: \end{eqnarray} 
223:  The parameters 
224: \begin{equation}
225:   \alpha = \gamma/\omega \,\,\,\,\, \hbox{ and  }
226:       \,\,\,\,\,     \Delta = {\mathcal D}/\omega^3  \label{parametre}
227: \end{equation} 
228: correspond to dimensionless dissipation rate and to noise strength, 
229: respectively. In the rest of this work, all physical quantities
230: are expressed in dimensionless units.
231: 
232: In Fig.~\ref{fig:traj}, we present, for three values of the parameters,
233: a trajectory in the phase plane $(x, v = \dot x)$ characteristic
234: of the oscillator's behaviour. Here as well as for other
235: simulations presented in this article, Eq.~(\ref{dissipDuff}) is 
236: integrated numerically by using the one step collocation method advocated in 
237: \cite{mannella} and described in detail in \cite{philkir1}. 
238: Initial conditions are chosen far from the origin, with an 
239: amplitude of order $1$. Noisy oscillations are observed
240: for small values of the damping parameter $\alpha$
241: [See Fig.~\ref{fig:traj}.a]. A larger value of $\alpha$ turns
242: the origin into a global attractor for the dynamics 
243: [See Fig.~\ref{fig:traj}.b].
244: Increasing the noise amplitude $\Delta$ at constant $\alpha$
245: makes the origin unstable again [See Fig.~\ref{fig:traj}.c].
246: 
247: For all values of the parameters $(\alpha, \Delta)$,
248: the origin  $(x = \dot{ x} = 0)$  is a  fixed point of
249:  the random dynamical system~(\ref{dissipDuff}). 
250:  Our main  goal, in this work, is to investigate
251:   the stability properties  of  the fixed point at the origin 
252: when the control parameters vary.
253: 
254: 
255: 
256: \subsection{Energy-angle representation of the  Duffing oscillator}
257: \label{sec:Duffing:energyangle}
258: 
259:  We  now  follow the method presented
260:  in \cite{philkir1} and introduce the  energy-angle variables
261:  associated  with  the Duffing oscillator.
262:  The deterministic and undamped
263:  oscillator corresponding to Eq.~(\ref{dissipDuff})
264:  is an integrable dynamical system  because the mechanical energy $E$,
265:  defined as  
266: \begin{equation}
267:    E =  \frac{1}{2}\dot x^2 + \frac{1}{2} x^2 +
268:    \frac{1}{4} x^4 \, ,
269:  \label{DuffingE}
270:  \end{equation}
271:  is a conserved quantity. Introducing
272:  the  angular variable $\phi$, the constant
273:  energy  orbits of the Hamiltonian system 
274:   ${\ddot x} + \, x  +  \, x^3  = 0$   are parametrized as
275:   \begin{eqnarray}
276:           x  &=& \left(  \frac{ 4E^2}{ 4E + 1}\right) ^{1/4}  
277:  {\rm sd}\left( \phi \;  ;k  \right),
278:   \label{dufx}  \\                 
279:      \dot x  &=&  (2E)^{1/2}
280:  \frac{ {\rm cn}\left( \phi \;   ;k \right)}
281:       { {\rm dn}^2\left( \phi  \;  ; k  \right)} ,
282: \label{dufv}
283:  \end{eqnarray}
284: where  the functions sd, cn and dn are Jacobi elliptic functions
285: \cite{abram}. The elliptic modulus $k$ 
286:  is a function of  the energy and is  given by
287:   \begin{equation}
288:  k^2 = \frac{\sqrt{4E + 1} - 1}{2 \sqrt{4E + 1}} .
289:  \label{modulus}
290: \end{equation} 
291:  In the $(E,\phi)$ coordinate system, the dynamical
292:  equations for the  Duffing  oscillator,  
293:  without noise and dissipation, become  simply 
294:    \begin{eqnarray}
295:  \dot E &=&   0  \, , \nonumber \\
296:  \dot\phi &=& (4E + 1)^{\frac{1}{4}}  \, .
297: \end{eqnarray}
298: 
299:  We reintroduce  the multiplicative  noise and the dissipation terms,
300:   and  obtain a system of coupled stochastic equations
301:  that describes the evolution of the energy and angle variables:
302: %can be written formally as
303:   \begin{eqnarray}
304:  \dot E &=&  J_E(E,\phi) + D_E(E,\phi)  \,\Xi(t) \,\,,
305:  \label{duffEJD}    \\
306:  \dot \phi &=&  J_{\phi}(E,\phi) + D_{\phi}(E,\phi) 
307:   \,\Xi(t) \,\,,
308:  \label{duffangleJD}
309: \end{eqnarray}
310: where the drift and diffusion coefficients $J_E$, $J_\phi$, $D_E$ and 
311: $D_\phi$ are defined by (see \cite{philkir1} for details of the derivation):
312:   \begin{eqnarray}
313: &&J_E(E,\phi) =  - 2 \, \alpha \, E \, \frac{ {\rm cn}^2(\phi;k)}
314: { {\rm dn}^4(\phi;k)}\\
315: &&D_E(E,\phi) =
316:   \frac { 2E } { (4E + 1)^{\frac{1}{4}}} \,
317: \frac{{\rm sn}(\phi;k) {\rm cn}(\phi;k)}{{\rm dn}^3(\phi;k)}  \\
318: &&J_{\phi}(E,\phi) =  (4E + 1)^{\frac{1}{4}}
319: + \alpha \,  \frac{2E + 1}{4E + 1} \,
320: \frac{{\rm sn}(\phi;k) {\rm cn}(\phi;k)}{{\rm dn}^3(\phi;k)} \\
321: &&D_{\phi}(E,\phi) =
322:  -  \frac{(2E + 1)}{ (4E + 1)^{\frac{5}{4}}}
323: \, {\rm sd}^2(\phi;k) \nonumber  \\
324: && -   \frac{E}{ (4E + 1)^{\frac{7}{4}}} 
325: {\rm sd}^4(\phi;k)
326:  +   \frac{\alpha  \,E }{ (4E +1)^{\frac{3}{2}}} \,
327: \frac{ {\rm cn}(\phi;k) {\rm sd}^3(\phi;k)}{{\rm dn}^2(\phi;k)}
328: \nonumber\\
329: && +   \frac{E}{ (4E + 1)^{\frac{7}{4}}} \;
330:   \frac{ {\rm sn}(\phi;k) {\rm cn}(\phi;k)}{{\rm dn}^3(\phi;k)} \,
331:    \int_0^\phi  {\rm sd}^2(\theta;k) d\theta   \nonumber \\
332: && -  \frac{\alpha  \,E }{ (4E +1)^{\frac{3}{2}}} \,
333:  \frac{ {\rm cn}^2(\phi;k)}{{\rm dn}^4(\phi;k)} \,
334:    \int_0^\phi  {\rm sd}^2(\theta;k) d\theta   .  \label{duffangle}
335: \end{eqnarray}
336: 
337: 
338: \section{Analysis of the Fokker-Planck equation}
339: \label{sec:pdf}
340: 
341:   Using  the same notations, 
342:  the  Fokker-Planck equation that governs the time evolution of
343:  the probability distribution function $P_t(E,\phi)$ associated
344:  with the stochastic differential  system
345:  (\ref{duffEJD})-(\ref{duffangleJD}) reads \cite{vankampen}
346: \begin{eqnarray}
347:  \label{FPEphi}   
348: \partial_t P_t(E,\phi)  &=& - \partial_{E}\left(  J_E P \right)
349:    -  \partial_{\phi}\left(  J_{\phi} P  \right) \\
350:   +   \frac{ \Delta}{2} && \left[\right. 
351:  \partial_{E} \left( D_E\partial_{E} (D_E P ) \right) +
352:  \partial_{E}  \left( D_E \partial_{\phi}  
353: (D_{\phi} P )\right) \nonumber \\ 
354: && +   \partial_{\phi} 
355:   \left( D_{\phi}\partial_{E} (D_E P ) \right)
356:  + \partial_{\phi}
357:    \left( D_{\phi}\partial_{\phi}(  D_{\phi} P ) \right)
358:  \left. \right]. \nonumber
359: \end{eqnarray}
360: The goal of this section is to determine the conditions
361: under which there exist stationary 
362: solutions $P_{\rm{stat}}(E, \phi)$ of Eq.~(\ref{FPEphi})
363: with a non-zero mean energy. Such an extended PDF
364: describes an oscillatory asymptotic state.
365: 
366: 
367: Obtaining an analytical expression of $P_{\rm{stat}}(E, \phi)$ 
368: in the general case is a daunting task.
369: In the following, we shall perform a local analysis of  
370: Eq.~(\ref{FPEphi}) for small and large values of $E$.
371: In these limiting cases, the expressions of  the  drift and 
372: diffusion coefficients  take  simpler  forms. We may write, 
373: in full generality,
374: \begin{equation}
375:    P_{\rm{stat}}(E, \phi) = 
376: P_{\rm{stat}}(\phi | E) \, {\tilde P}_{\rm{stat}}(E)    \, ,
377: \label{probcond}
378: \end{equation}
379: where $P_{\rm{stat}}(\phi | E)$  represents, in the stationary state,
380: the conditional probability of the angle $\phi$ at a fixed value of 
381: the energy $E$. Assuming    the conditional probability  
382: $P_{\rm{stat}}(\phi | E)$  to be independent of $E$ when $E \ll 1$ and 
383: $E \gg 1$ (respectively Secs.~\ref{sec:pdf:small} and
384: \ref{sec:pdf:large}), we shall derive the local behaviour 
385: of a stationary  solution of the Fokker-Planck equation~(\ref{FPEphi}).
386: This solution is a legitimate PDF when it is normalizable: the normalizability
387: condition provides the location of the bifurcation \cite{graham}.
388: 
389: 
390:  
391: \subsection{Small $E$ limit: the linearized stochastic oscillator}
392: \label{sec:pdf:small}
393: 
394:   When the oscillator's mechanical energy $E$ is small,
395:   the position $x(t)$ and the velocity $\dot x(t)$ are also  small:
396:  nonlinear terms may be neglected. The Duffing oscillator simplifies
397:  to a  linear (harmonic) oscillator:
398:  \begin{equation}
399:    \ddot x + \alpha \dot x + x  = x \; \Xi(t) 
400: \label{lineaire}
401: \end{equation}
402: with $ \langle \Xi(t) \Xi (t') \rangle = \Delta \; \delta(t - t')$.
403: The energy now  reads  $E = \frac{1}{2} \, {\dot x}^2 + \frac{1}{2} \, x^2$,
404: and Eqs.~(\ref{dufx}) and (\ref{dufv}) reduce to 
405:  \begin{eqnarray}
406:          x &=& \sqrt{2E} \sin \phi  \, , \label{enanglelin1} \\
407:       \dot x &=& \sqrt{2E} \cos  \phi  \, . \label{enanglelin}
408:   \end{eqnarray}
409: In the small $E$ limit, the elliptic modulus $k$ goes to zero,  the
410: elliptic functions reduce to circular functions \cite{abram}, and 
411:  the drift and diffusion
412: coefficients simplify to
413: \begin{eqnarray}
414: J_E(E,\phi) &=& -2 \, \alpha \, E \, \cos^2\phi  \,    \label{lim2JE}\\
415: D_E(E,\phi) &=& 2\, E \, \sin\phi  \cos \phi   \, , \label{lim2DE} \\ 
416: J_{\phi}(E,\phi) &=& 1  + \alpha  \sin \phi  \cos \phi \, ,\label{lim2Jphi} \\ 
417: D_{\phi}(E,\phi) &=& - \sin^2 \phi \, . \label{lim2Dphi}
418: \end{eqnarray}
419: Substituting the  expressions (\ref{lim2JE})--(\ref{lim2Dphi})
420: in   Eq.~(\ref{FPEphi}) and  integrating 
421: over $E$,  we obtain an independent Fokker-Planck equation for 
422: the marginal PDF ${\tilde P}_{\rm{t}}(\phi)$:
423: \begin{eqnarray}
424:   \label{FPlin2}
425: \partial_t {\tilde P}_t(\phi) &=&  - \partial_{\phi} \left[
426:   (1 + \alpha \sin\phi \cos\phi)\,   {\tilde P}_t(\phi) \right] \nonumber\\
427: &&  +     \frac{\Delta}{2} \partial_{\phi} \left[ \sin^2\phi \, 
428:  \partial_{\phi}  (\sin^2\phi \, {\tilde P}_t(\phi)) \right] \,.
429: \end{eqnarray}
430:  An explicit formula for   the  stationary angular
431:   measure  ${\tilde P}_{\rm{stat}}(\phi)$, solution of
432:        Eq.~(\ref{FPlin2}), is  derived 
433: in Appendix~\ref{sec:pdfangle}.  
434: 
435: 
436:   We now make the  hypothesis
437:   that in the small $E$ limit,  the stationary  conditional  measure
438:    $P_{\rm{stat}}(\phi | E )$  becomes independent  of   $E$
439:  and  is  identical to  the stationary
440:  angular  distribution    ${\tilde P}_{\rm{stat}}(\phi)$
441:         of   the harmonic oscillator:  
442: \begin{equation}
443:   \hbox{ for }
444:    \,\,\,\,\,    E \ll  1 \,, \,\,\,\,\,    P_{\rm{stat}}(\phi | E ) = 
445:   {\tilde P}_{\rm{stat}}(\phi)   \, .    \label{hypfactor}
446: \end{equation}
447: Equation~(\ref{probcond}) becomes
448:  \begin{equation}
449:    P_{\rm{stat}}(E, \phi) =  {\tilde P}_{\rm{stat}}(E)
450:    {\tilde P}_{\rm{stat}}(\phi)   \, ,
451: \label{factor}
452: \end{equation}
453: i.e., we assume that the stationary PDF factorizes when $E \ll 1$.
454: An exactly solvable equation for  
455: ${\tilde P}_{\rm{stat}}(E)$ can now be obtained by
456: inserting expressions (\ref{lim2JE})--(\ref{lim2Dphi})
457: and the identity~(\ref{factor}) in the Fokker-Planck  
458: equation~(\ref{FPEphi}) and then averaging  over $\phi$:
459: \begin{eqnarray}
460: & \frac{\Delta}{2} \, \left\langle \sin^2\phi \cos^2\phi \right\rangle \,
461:   \partial_E \left( E  {\tilde P}_{\rm{stat}}(E) \right) =
462: \label{FPEmargpetit} \\
463: &\left( \frac{\Delta}{2}  \left\langle \sin^2\phi \,
464:  ( \sin^2\phi -  \cos^2\phi) \right\rangle  - \alpha
465:  \left\langle  \cos^2\phi \right\rangle  \right) {\tilde P}_{\rm{stat}}(E) 
466:  \nonumber \,.  
467: \end{eqnarray}
468: Expectation values denoted by brackets $\langle \, \rangle$ 
469: are calculated using the stationary angular measure
470:    ${\tilde P}_{\rm{stat}}(\phi)$.
471: Introducing the notations:
472:  \begin{eqnarray}
473:   \mu(\alpha, \Delta) &=&  \frac{\Delta}{2} \, 
474: \left\langle \sin^2\phi \cos^2\phi \right\rangle  \, ,  \label{defmu}   \\
475:   \Lambda(\alpha, \Delta)  &=&  \frac{\Delta}{2}  \left\langle \sin^2\phi \,
476:  ( \sin^2\phi -  \cos^2\phi) \right\rangle  - \alpha
477:  \left\langle  \cos^2\phi \right\rangle  \, ,
478:  \label{deflyap1}
479: \end{eqnarray} 
480: Eq.~(\ref{FPEmargpetit}) becomes:
481:  \begin{equation}
482: \mu \,  \partial_E \left( E  {\tilde P}_{\rm{stat}}(E) \right) 
483: =  \Lambda \, {\tilde P}_{\rm{stat}}(E) 
484:  \,,  \label{FPEmargpetit2}
485: \end{equation}
486: and admits the solution:
487: \begin{equation}
488:  \label{pdfEpetit}
489:   {\tilde P}_{\rm{stat}}(E) \propto   E^{ \frac{\Lambda }{\mu }
490:   - 1 }  \,\,\,\,\,    \hbox{ for }  \,\,\,\, \, E \ll  1 \,.
491: \end{equation}
492: Thus, for $E \ll 1$, the PDF ${\tilde P}_{\rm{stat}}(E)$ behaves as a 
493: power law of the energy.
494: 
495: The coefficient $\mu$ is always positive, but the sign of  $\Lambda$ 
496: is a function of the parameters $\alpha$ and $\Delta$. 
497:  For negative values of $\Lambda$,  the stationary solution
498:  (\ref{pdfEpetit}) is not normalizable and cannot represent a PDF;
499:  the only stationary  PDF solution  of (\ref{FPEphi})
500: is then a delta function centered  at  the origin:
501: \begin{equation}
502:  \label{pdfEpetitdelta}
503:   {\tilde P}_{\rm{stat}}(E) =   \delta(E)
504:    \,\,\,\,\,    \hbox{ for }  \,\,\,\, \, \Lambda < 0 \, .
505: \end{equation}
506:  Thus for negative values of  $\Lambda$,  the origin is a global
507: attractor for the stochastic nonlinear oscillator.
508: Non-trivial stationary solutions exist only for $\Lambda > 0$:
509: the bifurcation threshold corresponds to values of the parameters
510:  for which  $\Lambda(\alpha, \Delta) = 0$.
511: 
512: 
513: 
514:  
515: \subsection{Large  $E$ limit}
516: \label{sec:pdf:large}
517: 
518:  For very large values of $E$, the elliptic modulus $k$,
519:   defined in Eq.~(\ref{modulus}),   tends
520:   to   $1/\sqrt{2}$, and 
521:   the drift and diffusion coefficients become 
522:    \begin{eqnarray}
523: J_E(E,\phi) &=& -2 \, \alpha \, E \, \frac{\rm{cn}^2(\phi)}{\rm{dn}^4(\phi)},
524: \label{limJE}                  \\ 
525: D_E(E,\phi) &=& \sqrt{2} \, E^{3/4} \,
526: \frac{\rm{sn}(\phi) \, \rm{cn}(\phi)}{\rm{dn}^3(\phi)} ,
527: \label{limDE}                  \\ 
528: J_{\phi}(E,\phi) &=& (4E)^{1/4} + \frac{\alpha}{2}\,
529:   \frac{\rm{sn}(\phi) \, \rm{cn}(\phi)}{\rm{dn}^3(\phi)},
530: \label{limJphi}                  \\ 
531: D_{\phi}(E,\phi) &=& - \frac{1} {2\sqrt{2} E^{1/4}}  \, \rm{sd}^2(\phi),
532: \label{limDphi}
533:  \end{eqnarray}
534: in agreement with Eqs.~(3) and (4) of \cite{philkir2}
535:  (we have omitted the elliptic modulus $1/\sqrt{2}$ for
536:   sake of  simplicity).
537:   Using  expressions (\ref{limJE}) to (\ref{limDphi})
538:   for the drift and diffusion
539:  coefficients, the stochastic equations 
540:  (\ref{duffEJD}) and (\ref{duffangleJD})  reduce,
541:   in the large $E$ limit,  to 
542: \begin{eqnarray}
543:  \dot{E} & = & -2 \, \alpha \, E \,
544:   \frac{\rm{cn}^2(\phi)}{\rm{dn}^4(\phi)}
545:  + \sqrt{2} \, E^{3/4} \,
546: \frac{\rm{sn}(\phi) \,
547:   \rm{cn}(\phi)}{\rm{dn}^3(\phi)}\, \Xi(t)   \label{grandE}   \\ 
548:  \dot{\phi} & =  & (4E)^{1/4} + \frac{\alpha}{2} 
549:   \frac{\rm{sn}(\phi) \, \rm{cn}(\phi)}{\rm{dn}^3(\phi)}
550:     - \frac{\sqrt{2}} {4 E^{1/4}}  \rm{sd}^2(\phi)\,  \Xi(t) 
551:  \label{grandphi}
552:  \end{eqnarray} 
553: In terms of the variable  $\Omega = E^{1/4},$ 
554:   Eq.~(\ref{grandE})  becomes  linear and has a form 
555:  similar to  Ornstein-Uhlenbeck's  equation.
556:  Because  the elliptic functions are bounded,  
557:   $\Omega$    saturates to a value of the order
558:   of $\frac{\Delta}{\alpha}$. Thus, we deduce  from Eq.~(\ref{grandphi})  
559:   that  the phase $\phi$
560:  grows linearly with time. Hence,   $\phi$ is a fast variable
561:  as compared to $\Omega$.
562:  Integrating  out this  fast variable  from   the
563: Fokker-Planck equation leads to 
564:  an effective stochastic dynamics for $\Omega$, 
565:   as explained in detail in   \cite{philkir1}.  
566: An equivalent formulation   is to assume  
567:   that, in the large $E$ limit, 
568:  the  stationary conditional probability  $P_{\rm{stat}}(\phi | E)$, 
569:   defined in Eq.~(\ref{probcond}), becomes   uniform in $\phi$  and 
570:   independent  of  $E$, i.e.,
571:   \begin{equation}
572:  \hbox{ for }  \,\,\,\, \, E \gg 1 \, , \,\,\,\,\,
573: P_{\rm{stat}}(\phi \,  | \,  E)  \sim 1 \, .
574:   \end{equation}
575:  Averaging the dynamics  over $\phi$  
576:  yields  the following expression 
577:  for the stationary  PDF  \cite{lindenberg,philkir2}: 
578: \begin{equation}
579:   \hbox{ for }   \, E \gg 1 , \,
580:   {\tilde P}_{\rm{stat}}(E) \propto  E^{-1/4}
581: \exp \left\{ - \frac{8 \, \alpha \, E^{1/2}}{ 3 \,{\mathcal M} \, {\Delta}  }
582:  \right\} \, , 
583: \label{pdfEgrand}
584: \end{equation}
585:  with $ {\mathcal M} = \frac{8}{5 \pi^2} \Gamma\left( 3/4 \right)^4 \simeq
586:  0.3655$,
587:  $\Gamma$  being  the Eulerian factorial function.
588: A stationary solution of (\ref{FPEphi}) decays as a stretched exponential
589:  and is always integrable at infinity
590: irrespective of the values of $\alpha$ and $\Delta$.
591:    
592: \section{Phase Diagram of the Duffing oscillator}
593: \label{sec:lyapunov}
594: 
595: In the long time limit, the energy of a harmonic oscillator 
596: with parametric noise (\ref{lineaire})  varies exponentially with time 
597: \cite{bourret}. This behaviour is characterized by a 
598: Lyapunov exponent, defined as 
599: $\lim_{t \to \infty}  \; \frac{1}{2t} \,    \overline{ \log E  }$,
600: where the overline  means averaging over $P_t(E, \phi)$ \cite{hansel}.
601: [The factor $1/2$ is included for consistency with the usual definition 
602: of the Lyapunov exponent in terms of the position $x(t)$.]
603: 
604: In Appendix~\ref{sec:identity}, we prove that the coefficient  
605: $\Lambda(\alpha,\Delta)$ defined in  Eq.~(\ref{deflyap1})
606: is equal to the Lyapunov exponent:
607: \begin{equation}
608: \Lambda(\alpha,\Delta) =
609:  \lim_{t \to \infty}  \; \frac{1}{2t} \,    \overline{ \log E  } \, .
610:  \label{defLyap1}
611: \end{equation}
612: 
613: \begin{figure}[ht]
614: \includegraphics*[width=0.75\columnwidth]{fig2.eps}
615: \caption{\label{fig:lyap:diag}
616: Phase diagram of the Duffing oscillator with multiplicative noise.
617: The critical curve separates an absorbing and an oscillatory phase
618: in the $(\alpha,\Delta)$ plane. The dashed line represents the 
619: asymptotic approximation,  Eq.~(\ref{smalla}).}
620: \end{figure}
621: 
622: 
623: For the (nonlinear) Duffing oscillator, we found in Section~\ref{sec:pdf}
624: that a non-trivial stationary 
625: solution $P_{\rm{stat}}(E, \phi)$ of Eq.~(\ref{FPEphi})
626: exists only when the Lyapunov exponent  $\Lambda(\alpha, \Delta)$
627: of the associated linear oscillator is positive.
628: This extended PDF describes an oscillatory asymptotic 
629: state for the Duffing oscillator, as opposed to the
630: absorbing state associated with the delta function (\ref{pdfEpetitdelta}).
631: The location of the stochastic bifurcation threshold of the Duffing 
632: oscillator, which separates the two states, is given by the curve  
633: $\Lambda(\alpha, \Delta) = 0$. The following  explicit formula 
634:  is derived in Appendix~\ref{sec:pdfangle}:
635: \begin{equation}
636:   \Lambda(\alpha, \Delta)  = \frac{1}{2} \left\{  \frac {\int_0^{+\infty} 
637: \mathrm{d}u \; {\sqrt u} \;
638:   e^{ -\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha^2}{4} ) u
639:    + \frac{u^3}{12} \right) }  }
640:  {\int_0^{+\infty}  \frac{\mathrm{d}u}{\sqrt u}
641:   e^{-\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha^2}{4} ) u
642:    + \frac{u^3}{12} \right) } }
643:    -  \alpha \right\}   \, .       
644: \label{Lyapunov}
645: \end{equation} 
646: 
647: 
648:   The transition line is
649:   determined  by  the relation between $\alpha$ and $\Delta$ for which
650:  the Lyapunov exponent vanishes. For a given value
651:  of $\Delta$, the  critical value of the damping 
652: $\alpha = \alpha_c(\Delta)$  satisfies 
653: \begin{equation}
654: \alpha_c   =  {\displaystyle { \frac {\int_0^{+\infty} 
655: \mathrm{d}u \; {\sqrt u} \; 
656:   \exp\left\{-\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha_c^2}{4} ) u
657:    + \frac{u^3}{12} \right) \right\}  }
658:  {\int_0^{+\infty} \frac{\mathrm{d}u}{\sqrt u}
659:   \exp\left\{-\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha_c^2}{4} ) u
660:    + \frac{u^3}{12} \right) \right\}  } }. }
661: \label{alphac}
662: \end{equation} 
663: The critical curve 
664: $\alpha = \alpha_c(\Delta)$ [Eq.~(\ref{alphac})] 
665: is represented in Fig.~\ref{fig:lyap:diag}.
666: It separates two regions in parameter space: for $ \alpha <  \alpha_c$
667:  (resp. $\alpha >  \alpha_c$)
668:  the Lyapunov exponent is positive (resp. negative),
669: the stationary PDF is an extended function (resp. a delta
670: distribution) of the energy, the origin is unstable (resp. stable).
671: For a detailed analytical study, we must now distinguish three cases 
672:   corresponding
673:  to an underdamped ($\alpha_c < 2$), critically damped  ($\alpha_c =2$)
674: and overdamped ($ \alpha_c > 2$) oscillator.
675: 
676: 
677: 
678: 
679: 
680: \subsection{Underdamped oscillator}
681: \label{sec:lin:diag:under}
682: 
683: The case $\alpha_c < 2$  can formally  be 
684: mapped onto that of a linear oscillator with no damping.
685: In terms of the variables 
686: $v = u \left[1 - \frac{\alpha_c^2}{4} \right]^{-1/2}$ and 
687: $\Delta_< = \Delta \left[1 - \frac{\alpha_c^2}{4}\right]^{-3/2}$,
688: Eq.~(\ref{alphac}) reads
689: \begin{eqnarray}
690: \alpha_c &=&  \left[1 - \frac{\alpha_c^2}{4}\right]^{1/2} \;
691:     \frac {   \int_0^{+\infty} \mathrm{d}v \, {\sqrt v} \,
692:   \exp\left\{-\frac{2}{\Delta_<} \left( v+\frac{v^3}{12} \right) \right\} }
693:  {\int_0^{+\infty} \frac{\mathrm{d}v}{\sqrt v} \exp\left\{-\frac{2}{\Delta_<}  
694: \left(  v    + \frac{v^3}{12} \right) \right\}  }  \nonumber \\
695:    &=&  \frac{\Delta_< }{2} \;
696: \left[1 - \frac{\alpha_c^2}{4}\right]^{1/2}    \;
697:  \frac { \sum_{n = 0}^{\infty}  \frac{(-1)^n }{n!} \,
698: \Gamma\left(3n + \frac{3}{2}\right) \,
699: \left(\frac{\Delta_<^2 }{48}\right)^n}
700: {  \sum_{n = 0}^{\infty} \frac{(-1)^n}{n!} \,
701: \Gamma\left(3n + \frac{1}{2}\right) \,
702:  \left(  \frac{\Delta_<^2 }{48}\right)^n }    \, . \nonumber
703: %\label{alphac2}
704: \end{eqnarray} 
705:  Retaining   only  the lowest order term in this series,
706:   we obtain  the limiting   behaviour  of   the critical
707:  curve $\alpha_c(\Delta)$,  for small values of $\alpha$ and $\Delta$:
708:  \begin{equation}
709:         \alpha_c  = \frac{\Delta  }{4} .
710: \label{smalla}
711: \end{equation} 
712:  The change of variables used here breaks down at $\alpha_c = 2$,
713:  we shall therefore consider this case separately. 
714: 
715: 
716: 
717: \subsection{Critically damped oscillator}
718: \label{sec:lin:diag:crit}
719: 
720:   For   a critically damped oscillator,
721:  we substitute $ \alpha_c =  2 $  in Eq.~(\ref{alphac}) and  obtain
722:  \begin{equation}
723:  2 = \frac {\int_0^{+\infty} \mathrm{d}u \, {\sqrt u}\,
724:   \exp\left(-  \frac{u^3}{6 \Delta_c}  \right)   }
725:  {\int_0^{+\infty} \frac{\mathrm{d}u}{\sqrt u} \,
726:  \exp\left(-  \frac{u^3}{6 \Delta_c}  \right)  }
727:   = ( 6 \Delta_c )^{1/3} \, \frac {\Gamma\left( \frac{1}{2} \right)}
728: {\Gamma\left(\frac{1}{6} \right)} \, .
729: \label{dcrit}
730: \end{equation} 
731:  Thus,   the critical
732: noise amplitude $\Delta_c$   for  $ \alpha_c =  2 $ is 
733:   \begin{equation} 
734:  \Delta_c = \frac{4}{3} \;  \frac{ \Gamma\left(\frac{1}{6} \right)^3}
735: { \pi^{3/2}} \simeq   41.2969..., 
736:  \label{dcrit2}
737: \end{equation} 
738: and the expression~(\ref{Lyapunov}) for the Lyapunov exponent reduces to
739: \begin{equation}
740:   \label{eq:Lyap:alpha2}
741:   \Lambda =  \left(\frac{\Delta}{\Delta_c} \right)^{1/3} - 1 \, .
742: \end{equation}
743: 
744: 
745: 
746: \subsection{Overdamped oscillator}
747: \label{sec:lin:diag:over}
748: 
749:   For  $\alpha_c > 2$,  we define
750:   $v = u \left[ \frac{\alpha_c^2}{4} - 1 \right]^{-1/2} $  and 
751:  $\Delta_> = \Delta \left[ \frac{\alpha_c^2}{4} - 1 \right]^{-3/2}$. 
752:   Equation~(\ref{alphac}) then becomes 
753: \begin{eqnarray}
754: \alpha_c \, \left[\frac{\alpha_c^2}{4}- 1 \right]^{-1/2}   
755:   &=&    \frac {   \int_0^{+\infty} \mathrm{d}v \, {\sqrt v} \,
756: e^{-\frac{2}{\Delta_>} \left(- v +\frac{v^3}{12} \right) } }
757:  {\int_0^{+\infty} 
758: \frac{\mathrm{d}v}{\sqrt v} e^{-\frac{2}{\Delta_>}  
759: \left(  -v    + \frac{v^3}{12} \right) }  }  \label{alphacseries2} \\
760:    &=& (6{\Delta_> })^{1/3} \;
761:  \frac  { \sum_{n = 0}^{\infty} 
762: \frac{1}{n!} \Gamma\left(\frac{2n+3}{6}\right) 
763: \left(\frac{48}{\Delta_>^2 }\right)^{n/3 } }
764: {  \sum_{n = 0}^{\infty}  \frac{1}{n!} \,
765: \Gamma\left(\frac{2n+1}{6}\right) 
766: \left(\frac{48}{\Delta_>^2 }\right)^{n/3 } }  \nonumber  
767: \end{eqnarray}  
768: 
769:  When $\alpha_c \to \infty$, the left hand side  of  Eq.~(\ref{alphacseries2})
770: the previous equation  converges to 2  and 
771:    ${\Delta_>}$  tends to  a limiting value ${\cal K}$  that  
772: we computed numerically:  ${\cal K}  \simeq 28.337$.
773:  Thus, in the large  $\alpha_c$ limit, we   find  
774:  the following relation   between $\alpha_c$ and $\Delta$
775: \begin{equation}
776:  \Delta  \simeq  {\cal K} \,\left[\frac{\alpha_c^2}{4}- 1 \right]^{3/2} 
777:  \simeq \frac{  {\cal K} }{8} \,  \alpha_c^3 \simeq 3.54 \, \alpha_c^3 .
778: \label{largea}
779:   \end{equation} 
780: 
781: When $\alpha_c  \to  2^{+}$, the left hand side of Eq.~(\ref{alphacseries2})
782: increases to  infinity.  Retaining only   the leading order term 
783:   of the series, we obtain 
784: \begin{equation}
785: 2 \, \left[\frac{\alpha_c^2}{4}- 1 \right]^{-1/2}=
786:     (6{\Delta_> })^{1/3}  \frac{\Gamma(1/2)}{\Gamma(1/6)} \, .
787: \end{equation} 
788:  This equation  is consistent with Eq.~(\ref{dcrit2}).
789: 
790: 
791: \section{Physical observables}
792: \label{sec:scaling}
793: 
794: The goal of this section is  to determine how  physical observables,
795: such as the energy and the position of the oscillator,  scale
796: near  the  bifurcation threshold as well as far from it. 
797: The agreement between analytical calculations
798: and numerical simulations will validate the assumptions 
799: made in Section~\ref{sec:pdf} concerning the behaviour of the
800: conditional PDF $P_{\rm{stat}}(\phi | E)$ for small and large values of $E$.
801: 
802: Let us first of all review the results obtained in Section~\ref{sec:pdf}.
803: For parameter values such that $\Lambda < 0$, we demonstrated that
804: ${\tilde P}_{\rm{stat}}(E)  = \delta(E)$: the origin is a 
805: global attractor of the dynamics. Conversely, when  $\Lambda > 0$, 
806: the origin becomes unstable. The stationary measure of the energy
807: is then an extended function: it behaves as a power law 
808: for small values  of $E$  [see Eq.~(\ref{pdfEpetit})],
809: and decreases as a stretched exponential for large values of $E$
810: [see Eq.~(\ref{pdfEgrand})]:
811:    \begin{eqnarray}
812:     \hbox{ for }  \, \, E \ll  1 \, , \,\,\,
813:   {\tilde P}^{<}_{\rm{stat}}(E) &=&  A  
814:   E^{ \frac{ \Lambda }{ \mu }
815:   - 1 }      \,,    \nonumber \\
816:    \hbox{ for }  \, \, E \gg  1 \, , \,\,\,
817:   {\tilde P}^{>}_{\rm{stat}}(E) &=& B  E^{-\frac{1}{4}}
818: \exp \left\{ - \frac{8 \, \alpha \, E^{1/2}}{ 3 \,{\mathcal M} \, {\Delta}  }
819:  \right\} \, .
820:  \label{matching}
821:   \end{eqnarray}
822:   The precise values of $A$  and $B$ can only
823:  be obtained by solving  the full Fokker-Planck
824:   equation~(\ref{FPEphi}), and matching 
825:   the two asymptotic expressions in the 
826:   intermediate regime $E \simeq 1$. 
827: 
828: 
829: 
830: \subsection{Behaviour   in the vicinity of the bifurcation}
831: \label{sec:NL:bif}
832: 
833: 
834: \begin{figure}[th]
835: \includegraphics*[width=0.75\columnwidth]{fig3a.eps}
836: \bigskip
837: \includegraphics*[width=0.75\columnwidth]{fig3b.eps}
838: 
839:     \caption{\label{fig:bifn3} Scaling behaviour of physical
840: observables of the stochastic  Duffing  oscillator
841: in the vicinity of the bifurcation threshold as a function of
842: the reduced control parameter 
843: $\varepsilon = (\Delta - \Delta_c)/\Delta_c$. (a) Average energy; (b) 
844: Variance of the position. Dashed lines corresponding to a linear function
845: of $\varepsilon$ are drawn to guide the eye. The corresponding
846: linear-linear graphs are shown as insets. Eq.~(\ref{dissipDuff})
847: is integrated numerically for $\alpha = 2$, with a time step
848: $\delta t = 5 \ 10^{-4}$. The threshold value is given by Eq.~(\ref{dcrit2}):
849: $\Delta_c(2) \simeq 41.3$. Ensemble averages are computed over
850: $10^3$ realisations.}
851: \end{figure}
852: 
853: \begin{figure}[th]
854: \includegraphics*[width=0.75\columnwidth]{fig4.eps}
855: \caption{\label{fig:bifn} Scaling behaviour of the  mean   energy 
856: of  the general  nonlinear oscillator in the vicinity of the 
857: bifurcation threshold as a function of the reduced control 
858: parameter  $\varepsilon = (\Delta -\Delta_c)/\Delta_c$. Dashed lines
859: indicating a linear dependence in $\varepsilon$ are drawn to guide the eye.
860: Eq.~(\ref{eqsto}) is integrated numerically for $\alpha = 2$, 
861: and with ${\mathcal U}(x) = x^{2n}/(2n)$,   $n = 2$, $3$ and $4$. 
862: Data obtained for $n = 2$ are the same as in Fig.~\ref{fig:bifn3}. 
863: The time step is $\delta t = 5 \ 10^{-4}$; ensemble averages are computed over
864: $10^3$ realisations. }
865: \end{figure}
866: 
867: 
868:    For  a given value of the damping rate $\alpha$,
869:  let us choose the noise strength  $\Delta$  such
870:  that  $0 < \Delta -\Delta_c(\alpha) \ll 1$. In this case, 
871:    the parameters of the  system 
872:  are tuned just   above the bifurcation threshold  and 
873:     the Lyapunov exponent  is slightly positive:  
874:  $0 < \Lambda  \ll 1$.
875:   The stationary PDF is an extended function but,
876:   as can be inferred  from   Eq.~(\ref{matching}), 
877:  most of its  mass  is concentrated in  the vicinity 
878:  of $E=0$.  Therefore, a good approximation of 
879: the stationary PDF  is obtained by taking it equal to 
880:     ${\tilde P}^{<}_{\rm{stat}}(E)$  over  a finite
881:  interval in the vicinity of 0  (e.g.,   $0 \le E \le E_0$),  
882:  and  equal to zero     outside  this  interval.  
883: The mean value of the energy is given by:
884: \begin{equation}  
885:   \langle E \rangle_{\mathrm{stat}} \simeq 
886:  \frac{  \int_0^{E_0} E^{\frac{\Lambda }{\mu}} \mathrm{d} E   } 
887:       { \int_0^{E_0}  E^{ \frac{\Lambda }{\mu }  - 1 }  \mathrm{d} E  } 
888:  \propto  E_0 \, \frac { \Lambda }{ \mu} \, .   
889:   \label{multif1} 
890: \end{equation}
891:  If we now take into account that: $(i)$  
892: $ \Lambda(\alpha, \Delta_c(\alpha)) =0$; and
893:  $(ii)$  the Lyapunov  exponent and the coefficient $\mu$  are regular
894:  functions  of the noise  strength $\Delta$ when $\Delta \neq 0$, 
895:  we obtain from Eq.~(\ref{multif1})
896: \begin{equation}  
897:  \langle E \rangle_{\mathrm{stat}} 
898: \propto \Lambda(\alpha, \Delta) - \Lambda(\alpha, \Delta_c)
899: \propto (\Delta -\Delta_c)   \, , 
900:  \label{multif2}
901: \end{equation} 
902: where the last identity is obtained by a Taylor expansion 
903: of $\Lambda$ for $\Delta$ close to $\Delta_c$.
904: This result is confirmed by numerical simulations 
905: [See Fig.~\ref{fig:bifn3}.a]. A similar reasoning 
906: shows that $ \langle E \rangle_{\mathrm{stat}} $ scales
907: linearly with $\alpha - \alpha_c(\Delta)$ at fixed noise strength $\Delta$.
908: 
909: Let us now consider  the variance of the oscillator's
910: position close to the bifurcation.
911: Using Eqs.~(\ref{enanglelin1}) and (\ref{factor}), we find
912: \begin{eqnarray}
913: \langle x^2 \rangle_{\mathrm{stat}}  
914: &=& \iint P_{\mathrm{stat}}(E,\phi) \, 2 E \sin^2 \phi
915: \; \mathrm{d}E \, \mathrm{d}\phi \nonumber \\
916: &=& 2 \, \langle \sin^2 \phi \rangle \, 
917: \langle E \rangle_{\mathrm{stat}} \nonumber \\
918: &\propto& (\Delta -\Delta_c) \, .    \label{eq:scalingx2}
919: \end{eqnarray}
920: The variance of the position scales linearly with the distance to
921: threshold [See Fig.~\ref{fig:bifn3}.b]. 
922: Using a similar argument, one also finds that 
923: $\langle v^2 \rangle_{\mathrm{stat}} \propto (\Delta -\Delta_c)$.
924: 
925: 
926: 
927: 
928: 
929: As seen in  Fig.~\ref{fig:bifn}
930:  the linear scaling (\ref{multif2}) of the energy with respect
931:   to  the distance to threshold is observed
932: for  ${\mathcal U}(x) \propto x^{2n}  $, $n = 2$, $3$ and $4$,
933: where $\Delta_c(\alpha)$ is \emph{always} given by 
934:  $\Lambda(\alpha, \Delta_c(\alpha)) = 0$.
935: The threshold is independent of   the nonlinearity 
936:  involved in the confining  potential ${\mathcal U}$ appearing 
937:  in Eq.~(\ref{eqsto}), because the  local expression 
938:  ${\tilde P}^{<}_{\rm{stat}}(E)$  of the stationary PDF
939:  in the vicinity of $E =  0$ is obtained from the \emph{linearized}
940:  stochastic differential equation.   The observed dependence of the
941: constant of proportionality in Eq.~(\ref{multif2}) upon 
942: the stiffness of the anharmonic potential remains to be understood.
943: 
944: 
945: 
946: 
947: 
948: \begin{figure}[ht]
949: \includegraphics*[width=0.75\columnwidth]{fig5.eps}
950: \caption{\label{fig:loin} Scaling behaviour of the mean energy 
951: of  the general  nonlinear oscillator far from the bifurcation 
952: as a function of the parameter  $\Delta/\alpha$. 
953: Eq.~(\ref{eqsto}) is integrated numerically with 
954: ${\mathcal U}(x) = x^{2n}/(2n)$,   $n = 2$, $3$ and $4$. 
955: Dashed lines correspond to the predicted scaling (\ref{avgEtn}),
956: with a scaling exponent $n/(n-1) = 2$, $3/2$ and $4/3$.
957: The time step is $\delta t = 5 \ 10^{-4}$; ensemble averages are computed over
958: $10^3$ realisations. }
959: \end{figure}
960: 
961: 
962:  \subsection{Scaling  far from the bifurcation}
963: 
964:   We shall now study the scaling of the  physical observables far from the
965:  bifurcation point. This situation is encountered,
966:  for instance,
967:  when   the damping rate is vanishingly
968:  small for a given noise strength.  The Lyapunov exponent 
969:   is then a  finite, strictly positive, number. 
970:  When $\alpha \to 0$, 
971:  most of the mass of the stationary PDF is concentrated at large
972:   values of  $E$  as can be seen from Eq.~(\ref{matching}).
973:  Thus,  a satisfactory   approximation of 
974: the stationary PDF  is to assume it to be  identical to the  expression
975:     ${\tilde P}^{>}_{\rm{stat}}(E)$ for all $E$. 
976:  After a suitable normalisation, we find 
977:  \begin{equation}
978:    {\tilde P}_{\rm{stat}}(E) = \frac{1}{ 2 E \Gamma \left(3/2\right)}% \,
979: \left( \frac{8 \, \alpha  \, E^{1/2}}{ 3 \,{\mathcal M} \, {\Delta}   }
980: \right)^{3/2} 
981: \exp \left\{ - \frac{8 \, \alpha \, E^{1/2}}{ 3 \,{\mathcal M} \, {\Delta} }
982:  \right\} ,
983: \label{pdfstat}
984: \end{equation} 
985:  with  the numerical factor ${\mathcal M}$ 
986:    defined in Eq.~(\ref{pdfEgrand}).
987:  In this low damping regime,  we  recover,
988:  as expected, the  results derived in Refs.~\cite{philkir1,philkir2}.
989:  The average energy is readily obtained from Eq.~(\ref{pdfstat})
990:  \begin{equation}
991: \langle E \rangle_{\rm{stat}}  = \frac{135 }{256} \, {\mathcal M}^2 \,
992: \left( \frac{ {\Delta}   }{\alpha }\right)^2 
993:  \simeq 0.070  \left( \frac{ {\Delta}   }{\alpha }\right)^2  \, .
994: \label{avgEt}
995: \end{equation} 
996:  This result is  
997: at variance with the usual equilibrium fluc\-tua\-tion-dissipation theorem 
998: ($\langle E \rangle_{\mathrm{eq}} \propto {\Delta}/\alpha $);
999:  yet this is not  a contradiction since
1000:  Eq.~(\ref{dissipDuff}) models a system  out of  equilibrium.
1001: 
1002: 
1003: Far from the bifurcation threshold, the scaling  behaviour 
1004:  depends strongly  on   the stiffness of the potential ${\mathcal U}$. 
1005: If we take ${\mathcal U}(x) \propto x^{2n}  $, Eq.~(\ref{avgEt})
1006:  can be generalized to  \cite{philkir2} 
1007:  \begin{eqnarray}
1008: \langle E \rangle_{\rm{stat}}  = &&
1009: \frac{1}{2n} \;     \frac{\Gamma \left(\frac{3n+1}{2(n-1)} \right)} 
1010:   {\Gamma \left(\frac{n+1}{2(n-1)} \right)}  \label{avgEtn} \\
1011: &&  \left( ( n-1) \; 
1012:      \frac{\Gamma \left(\frac{3}{2n} \right)
1013:  \; \Gamma \left(\frac{3n+1}{2n} \right)} 
1014:   {\Gamma \left(\frac{1}{2n} \right) \; 
1015:  \Gamma \left(\frac{3n+3}{2n} \right)} 
1016: \right)^{\frac{n}{n-1}} \;
1017:  \left( \frac{ {\Delta}   }{\alpha }\right)^{\frac{n}{n-1}} \, .
1018:  \nonumber
1019:  \end{eqnarray}
1020: This prediction is again confirmed by numerical simulations
1021: [See Fig.~\ref{fig:loin}].
1022: 
1023: 
1024:  \section{Conclusion}
1025: \label{sec:disc}
1026: 
1027:     For a  deterministic Duffing  oscillator  with  linear damping,
1028:  the fixed point at the origin $(x = \dot{x} =0)$ is a 
1029:  global attractor: irrespective of the choice of initial conditions, 
1030:  the mechanical energy is totally  dissipated in the long time limit 
1031: and  the system  converges  to the origin.
1032: This picture is no longer valid once the oscillator is 
1033: subject to parametric noise. Indeed, the dissipative  Duffing  
1034: oscillator with parametric white  noise undergoes a noise-induced 
1035: bifurcation between: $(i)$ an absorbing state reminiscent of the 
1036: deterministic system;  and $(ii)$ an oscillatory state, resulting
1037: from a dynamical balance between energy dissipation and 
1038: (noise-induced) energy injection. Noisy oscillations appear
1039: when the fixed point at the origin becomes unstable: this transition 
1040: is thus referred to as a (forward) stochastic Hopf  bifurcation.
1041: Both states may be  characterized by the stationary probability
1042:  distribution function of the oscillator's mechanical energy $E$.
1043:  The PDF of the absorbing state is a  delta-function, whereas
1044:  the oscillatory  state corresponds to an extended PDF.
1045: 
1046: Our main result concerns the location of the transition point:
1047: a  noise-induced  bifurcation for the Duffing oscillator occurs precisely 
1048: where the Lyapunov exponent of the linear  oscillator with 
1049: parametric noise changes sign. We emphasize that this result
1050: is genuinely non-perturbative. It results from a simple hypothesis 
1051: on the behaviour of the  conditional probability of the angular variable 
1052: at fixed energy, valid for very small values of the energy 
1053:   and for \emph{arbitrary} noise amplitude.
1054: This assumption is physically plausible (i.e. mean-field-like) and
1055: justified by the consequences deduced from it.
1056:  Having identified the Lyapunov exponent  of the stochastic
1057: \emph{linear} oscillator as the important physical quantity, we  
1058: give a simple analytical expression of this exponent as a 
1059: function of the system's parameters, the damping rate and the noise strength.
1060: The location of the transition line that separates the absorbing and
1061: oscillatory states of the (nonlinear) Duffing  oscillator follows
1062: immediately. 
1063: 
1064: Although we have considered here only a cubic
1065:  oscillator with parametric Gaussian 
1066:   white noise, we believe that the relation
1067: between the sign of the Lyapunov exponent and the
1068:  location of the noise-induced transition holds 
1069:  irrespective  of the nonlinearity and of the nature
1070:  of the noise.   Finally, we demonstrate that
1071: all physical observables scale linearly with respect to the distance
1072: to threshold. Indeed, the argument we use to evaluate the mean energy
1073: $\langle E \rangle$ may be easily generalized to show that 
1074: all moments $\langle E^k \rangle$  also scale linearly,
1075:  suggesting that the concept of critical exponents
1076:  is not relevant here.
1077: One would certainly like to test experimentally whether 
1078: noise-induced transitions may indeed be characterized
1079: by a multifractal order parameter, at variance with the
1080: theory of equilibrium continuous transitions.
1081: 
1082: 
1083: 
1084: \appendix
1085: 
1086: 
1087: \section{Some useful identities for the  Lyapunov exponent}
1088: \label{sec:identity}
1089: 
1090: 
1091: In this appendix, we  prove that the Lyapunov exponent, defined as 
1092: $\lim_{t \to \infty}  \; \frac{1}{2t} \,    \overline{ \log E  }$,
1093: satisfies the relation~(\ref{defLyap1}). 
1094:   Using  the  `radial' variable  $\rho$  defined by 
1095: \begin{equation}
1096:   \rho = \frac{1}{2} \log \left( 2 E \right)  \,  , \label{defU}
1097: \end{equation}
1098:  we rewrite   Eq.~(\ref{lineaire}) as a
1099:   system of nonlinear  coupled stochastic equations
1100:   \begin{eqnarray}
1101:   \dot \rho &=&  -  \alpha   \cos^2 \phi +  \sin\phi \cos\phi  \, \Xi(t) 
1102:     \, , \label{eqUlin}   \\
1103:    \dot \phi &=&  1 + \alpha \sin\phi \cos\phi - \sin^2\phi\, \Xi(t)
1104:     \, .\label{eqthetalin}
1105:  \end{eqnarray}
1106:   The  Fokker-Planck
1107:   evolution  equation for the joint probability distribution
1108:  $P_t(\rho,\phi)$  is then given by  \cite{vankampen}
1109:  \begin{eqnarray}
1110:  \partial_t P_t& = & \label{FPlin3}
1111:   \alpha   \cos^2 \phi \, \partial_{\rho}  P_t   -  \partial_{\phi} 
1112: \left[ (1 + \alpha \sin\phi \cos\phi)\, P_t \right] \label{FPlin1}  \\ 
1113: && + \frac{\Delta}{2}  \left\{\right. 
1114:  \sin^2 \phi \cos^2\phi \,  \partial^2_{\rho\rho} P_t
1115:  - \sin\phi \cos\phi \, 
1116: \partial^2_{\rho \phi} \left[\sin^2\phi \, P_t \right]  \nonumber \\
1117: &&  - \partial_{\phi} \left[ \sin^3\phi \cos\phi \,  
1118: \partial_\rho P_t \right]   +   \partial_{\phi} \left[ \sin^2\phi \, 
1119:  \partial_{\phi}  (\sin^2\phi \, P_t \,) \right] \left.\right\} \nonumber
1120: \end{eqnarray}
1121:   From  Eq.~(\ref{FPlin1}) we derive the time evolution
1122:  of the mean value  $\overline{ \rho }$, where  the overline
1123:   indicates  averaging with respect to  
1124: $P_t(\rho,\phi)$. After  suitable integration by parts, we obtain
1125: \begin{equation}
1126:  \frac{ \rm{d}}{ \rm{d} t}\overline{ \rho }  =
1127:       -\alpha \;  \overline{  \cos^2\phi} \,  
1128:  +  \frac{\Delta}{2}  \overline{   \sin^2\phi\,  ( \sin^2\phi -  \cos^2\phi )}
1129:     \, .  \label{evolmoyU}
1130: \end{equation} 
1131:  Similarly,  we deduce the time evolution
1132:  of  $ \overline{ \log | \sin\phi |  }$
1133:  \begin{eqnarray}
1134:   \frac{ \rm{d}}{\rm{d} t}  \overline{ \log | \sin\phi |  }
1135: & = & \overline{ \mathrm{cotan}\, \phi } 
1136: + \alpha \; \overline{ \cos^2\phi }  \nonumber\\
1137: && - \frac{\Delta}{2} \; \overline{  \sin^2\phi\,  
1138: ( \sin^2\phi -  \cos^2\phi )}
1139:     \, .  \label{evolmoylog}
1140:  \end{eqnarray}
1141:  From Eqs.~(\ref{evolmoyU}) and (\ref{evolmoylog}), we obtain 
1142:  \begin{equation} 
1143:  \frac{ \rm{d}} { \rm{d} t}\overline{ \rho }   =
1144:  \overline{ \mathrm{cotan}\, \phi }    -  
1145:   \frac{ \rm{d}}{\rm{d} t}\overline{ \log | \sin\phi |  } 
1146:   \, .  \label{lienmoys} 
1147:  \end{equation} 
1148:  At large times, the PDF ${  P}_t(E, \phi)$ reaches
1149:  a stationary limit  ${ P}_{\rm{stat}}(E,\phi)$.
1150:  Thus,  when $ t \to \infty $,
1151:   the averages  $\overline{ \log | \sin\phi |  }$   and 
1152:  $\overline{ \mathrm{cotan}\, \phi }$   have  finite limits,
1153:  given by $\langle  \log | \sin\phi | \rangle$  and
1154:  $ \left\langle \mathrm{cotan}\, \phi  \right\rangle $, respectively,
1155:  where the expectation values  denoted by  brackets $\langle \rangle$ 
1156:  are calculated by using the marginal 
1157:  PDF,  ${\tilde P}_{\rm{stat}}(\phi)$.
1158:  Therefore 
1159:  \begin{equation}   
1160:  \lim_{t \to \infty } \frac{ \rm{d}} { \rm{d} t}  \overline{ \rho}    =  
1161:   \left\langle \mathrm{cotan}\, \phi  \right\rangle    \,.
1162: \label{meanrho}
1163: \end{equation}
1164:   From Eqs.~(\ref{defU})~and~(\ref{meanrho}), 
1165:  we derive
1166: \begin{equation}  
1167:  \lim_{t \to \infty}  \frac{1}{2 t} \overline{ \log E }
1168: = \lim_{t \to \infty}  \frac{1}{2} 
1169: \frac{\mathrm{d}}{\mathrm{d} t} \overline{ \log E }
1170: = \lim_{t \to \infty}  \frac{\mathrm{d}}{\mathrm{d} t} 
1171: \overline{ \rho} 
1172:  =    \left\langle \mathrm{cotan}\, \phi  \right\rangle   \, .
1173: \label{formLyap}
1174: \end{equation}
1175: Similarly,  we take the stationary limit of Eq.~(\ref{evolmoylog}) 
1176: and obtain
1177:  \begin{equation}
1178:    \left\langle \mathrm{cotan}\, \phi  \right\rangle  =
1179: - \alpha \; \left\langle \cos^2\phi  \right\rangle 
1180:  +  \frac{\Delta}{2} \;\left\langle  \sin^2\phi\,  
1181: ( \sin^2\phi -  \cos^2\phi ) \right\rangle    .
1182: \label{Lyapfin}
1183:  \end{equation}
1184: The right hand side of this equation is nothing but $\Lambda$ [See
1185: Eq.~(\ref{deflyap1})]. This 
1186: concludes the derivation of   Eq.~(\ref{defLyap1}) :
1187: \begin{equation}  
1188:  \lim_{t \to \infty}  \frac{1}{2 t} \overline{ \log E }
1189:  =    \left\langle \mathrm{cotan}\, \phi  \right\rangle  
1190:  =   \Lambda(\alpha, \Delta) \, .
1191: \label{idLyap}
1192: \end{equation}
1193: 
1194: 
1195: 
1196: 
1197: 
1198: 
1199:    \section{Calculation  of the  Lyapunov exponent}
1200: \label{sec:pdfangle}
1201: 
1202: 
1203:  Expressions for the  Lyapunov exponent  have been 
1204:   available   for  a long time   
1205:  in the  literature \cite{hansel},
1206:  but they  are rather unwieldy.
1207:  In  a recent  article 
1208:   about the quantum localisation problem \cite{tessieri},
1209:   an  expression for  the 
1210:  localisation length  is derived: it is shown to be identical
1211:  to  the inverse  Lyapunov exponent of the linear  oscillator
1212:   with parametric noise and zero damping. 
1213:  In this appendix, we generalize the calculation
1214:  of \cite{tessieri} to the dissipative case 
1215:  and derive a particularly  simple formula for the Lyapunov exponent.
1216: 
1217:   Since  $\Lambda = \langle \mathrm{cotan}\phi \rangle  $ 
1218:  [See Eq.~(\ref{idLyap})],  
1219:   we introduce  the auxiliary variable $z$ defined as
1220: \begin{equation}
1221:  z = \dot x/x = \mathrm{cotan}\, \phi  \, .
1222: \label{defz}
1223: \end{equation} 
1224:   Then, equation~(\ref{lineaire})  becomes  
1225: \begin{equation}
1226: \dot z + 1 + \alpha \; z  + z^2 = \Xi(t) \,.
1227: \label{eqz}
1228:  \end{equation} 
1229:  The  stationary Fokker-Planck equation  associated  with Eq.~(\ref{eqz})
1230:   is given by
1231:  \begin{equation}
1232:     \frac{\Delta}{2} \;  \frac{ \rm{d}} { \rm{d} z} P_{\rm{stat}}
1233:  = -( 1 +  \alpha z  +  z^2) P_{\rm{stat}}  + J \,,
1234: \label{StatFPz}
1235: \end{equation} 
1236: where we have introduced the stationary current $J$.
1237:  The solution of this equation   is found  by the method
1238: of the variation of constants:
1239: \begin{equation}
1240:  P_{\rm{stat}}(z) =  \frac{1}{N} \int_{-\infty}^z  \exp\left\{\frac{2}{\Delta} 
1241:   (\Phi(y) - \Phi(z) ) \right\} \mathrm{d}y,
1242:  \label{Pstatz}
1243: \end{equation} 
1244: where $\Phi(y) = y +\frac{\alpha}{2} y^2 + \frac{1}{3} y^3$.
1245: Writing  $\int_{-\infty}^{\infty} P_{\rm{stat}}(z) \; \mathrm{d}z = 1$, we
1246: obtain the normalisation constant $N$:
1247: \begin{equation}
1248:   N = \int_{-\infty}^{+\infty} \mathrm{d}z \int_{-\infty}^z  \mathrm{d}y
1249: \exp\left\{\frac{2}{\Delta}   (\Phi(y) - \Phi(z) ) \right\} .
1250: \label{formN1}
1251: \end{equation} 
1252:   Following \cite{tessieri}, we 
1253:  change the  $y$  variable to $u = z -y$ in Eq.~(\ref{formN1}):
1254: \begin{equation}
1255:    N  =  \int_{-\infty}^{+\infty} \mathrm{d}z \int_0^{+\infty}\mathrm{d}u  \, 
1256: e^{-\frac{2}{\Delta} \big(u  - \frac{\alpha}{2}u^2  + \frac{u^3}{3}
1257:  + z^2 u + z( \alpha u -  u^2) \big) }  \, .
1258: \end{equation} 
1259:   The integral over  $z$ is now Gaussian, and can be evaluated 
1260:  before the integral  over  $u$. Hence, we obtain 
1261: \begin{equation}
1262:    N  = \sqrt{\frac{\pi \Delta}{2}} \; \int_0^{+\infty} 
1263: \frac{\mathrm{d}u}{\sqrt u}  \,
1264:   \exp \left\{ -\frac{2}{\Delta} \,  \left(  ( 1 - \frac{\alpha^2}{4} ) u
1265:    + \frac{u^3}{12} \right) \right\} \, .
1266: \label{Norm}
1267: \end{equation} 
1268: 
1269:  The use of the identity
1270:  \begin{equation}
1271:   {\tilde P}_{\rm{stat}}(\phi)\rm{d}\phi  =  P_{\rm{stat}}(z) \rm{d}z \, ,
1272: \end{equation} 
1273:   yields the stationary angular measure
1274:  \begin{equation}
1275:   \label{probastatphi}
1276: {\tilde P}_{\rm{stat}}(\phi) = \frac{1}{N \sin^2\phi} 
1277:  \int_{-\infty}^{\rm{cotan}\;\phi}   
1278: \mathrm{d}y   \;
1279: e^{ \frac{2}{\Delta} \left(\Phi(y) - \Phi(\mathrm{cotan}\;\phi) \right)}
1280:  \, .
1281:     \end{equation}
1282:   We deduce that the 
1283:  Lyapunov exponent satifies
1284:  \begin{eqnarray}
1285: \Lambda &=&   
1286:  \int_0^\pi {\tilde P}_{\rm{stat}}(\phi) \, \mathrm{cotan} \phi \,
1287:  \mathrm{d}\phi   \label{eqlambdaz}  \\
1288: &=&  \frac{1}{N}\int_{-\infty}^{+\infty} \mathrm{d}z \; z 
1289:  \int_{-\infty}^z  \exp\left\{\frac{2}{\Delta} 
1290:   (\Phi(y) - \Phi(z) ) \right\} \mathrm{d}y    \, .\nonumber
1291:  \end{eqnarray}
1292:   We again  change the variables  from $(y,z)$ to $(u,z)$
1293: with $u = z -y$, and obtain 
1294:  \begin{equation}  
1295: \Lambda =   \frac{1}{N} 
1296: \int_{-\infty}^{+\infty} \mathrm{d}z \, z \int_0^{+\infty}\mathrm{d}u \,
1297: e^{-\frac{2}{\Delta} \left(u  - \frac{\alpha}{2}u^2  + \frac{u^3}{3}
1298:  + z( \alpha u -  u^2)  + z^2 u   \right) }  
1299: \end{equation} 
1300:   Evaluating the Gaussian integral in $z$  leads to
1301:  the     expression   of  the   Lyapunov exponent given in the text
1302:  \begin{equation}
1303:   \Lambda(\alpha, \Delta)  = \frac{1}{2} \left\{  \frac {\int_0^{+\infty} 
1304: \mathrm{d}u \; {\sqrt u} \;
1305:   e^{ -\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha^2}{4} ) u
1306:    + \frac{u^3}{12} \right) }  }
1307:  {\int_0^{+\infty}  \frac{\mathrm{d}u}{\sqrt u}
1308:   e^{-\frac{2}{\Delta} \left(  ( 1 - \frac{\alpha^2}{4} ) u
1309:    + \frac{u^3}{12} \right) } }
1310:    -  \alpha \right\}   .      
1311: \label{Lyapunovapp}
1312: \end{equation} 
1313:   This formula also appears in 
1314:   recent   mathematical   literature  \cite{imkeller}. 
1315:   It constitutes an exact result,
1316: obtained without approximations. We have computed numerically the Lyapunov
1317: exponent by measuring $\overline{\log E}$ in the time-asymp\-to\-tic regime
1318: and thus confirmed the validity of (\ref{Lyapunovapp}).
1319: 
1320: 
1321: 
1322: 
1323: 
1324: \begin{thebibliography}{article}
1325: \bibitem{lefever} H. Horsthemke and R. Lefever,  {\it Noise
1326: Induced Transitions}  (Springer-Verlag, Berlin, 1984).
1327: \bibitem{strato} R.L.~Stratonovich,  {\it Topics  on the
1328: Theory of Random Noise } (Gordon and Breach, New York, 1963 (Vol. 1), 
1329: 1967 (Vol. 2)).
1330: \bibitem{landaMc} P.S. Landa  and  P.V.E. McClintock, 
1331: Phys. Rep. {\bf 323},  1 (2000).
1332:  \bibitem{fauve} R. Berthet,  S. Residori,
1333:  B. Roman and  S. Fauve,  Phys. Rev. Lett. {\bf 33}, 557 (2002).
1334: \bibitem{kramer1} A.~Becker and L.~Kramer, Phys. Rev. Lett.
1335: {\bf 73}, 955 (1994); Physica D {\bf 90}, 408 (1996).
1336: \bibitem{kramer2} J. R\"oder, H. R\"oder  and L.~Kramer, 
1337:  Phys. Rev. E {\bf 55},  7068 (1997).
1338: \bibitem{behn} U.~Behn, A.~Lange and T.~John, 
1339:  Phys. Rev. E {\bf 58},  2047 (1998).
1340:   \bibitem{frisch} U. Frisch, in   {\em Probabilistic Methods
1341:   in Applied Mathematics}, edited by A.T.~Bharucha-Reid 
1342:   (Academic Press, New York, 1968).
1343: \bibitem{luecke} M.~L\"ucke and F.~Schank,  Phys. Rev. Lett.
1344:  {\bf 54}, 1465 (1985); M. L\"ucke, in  {\em Noise in Dynamical
1345:  Systems, Vol. 2: Theory of Noise-induced Processes in Special
1346:  Applications}, edited by F.~Moss and P.V.E.~Mc Clintock 
1347:  (Cambridge University Press, Cambridge, 1989).
1348: \bibitem{vandenb1} C. van den Broeck, J.M.R. Parrondo and R. Toral,
1349:   Phys. Rev. Lett {\bf 73}, 3395 (1994).
1350: \bibitem{vandenb2} J.M.R. Parrondo, C. van den Broeck, J. Buceta
1351:  and F. Javier de la Rubia, Physica {\bf 224} A, 153 (1996).
1352: \bibitem{toral} M. San Miguel and R. Toral, {\it Stochastic Effects
1353:  in Physical Systems}, in  {\em Instabilities and Nonequilibrium
1354: Structures }{\bf VI}, edited by E. Tirapegui and  W. Zeller  (Kluwer,
1355: Dordrecht, 1997).
1356: \bibitem{lindenberg}  V.  Seshadri, B. J. West and  K. Lindenberg,
1357:  Physica {\bf 107} A,  219 (1981).
1358: \bibitem{landa} P.S.~Landa and A.A.~Zaikin, Phys. Rev. E {\bf 54},
1359:   3535 (1996).
1360: \bibitem{drolet} F.~Drolet and J.~Vi\~nals, 
1361: Phys. Rev. E {\bf 57}, 5036 (1998).
1362: \bibitem{bourret} R.C.~Bourret, Physica   {\bf 54},  623 (1971).
1363: \bibitem{bourretFrisch} R.C.~Bourret, U.~Frisch and A.~Pouquet,
1364:  Physica   {\bf 65}, 303 (1973).
1365: \bibitem{philkir1} K.~Mallick and P.~Marcq, Phys. Rev. E 
1366: {\bf 66}, 041113 (2002).
1367: \bibitem{philkir2} K.~Mallick and P.~Marcq, 
1368:  Physica   {\bf 325} A, 213     (2003).
1369: \bibitem{manneville} P.~Manneville,  {\it Dissipative structures
1370:  and weak turbulence} (Academic Press, New York, 1990).
1371: \bibitem{lindenberg3} K.~Lindenberg,  V.~Seshadri and
1372:   B.J.~West,  Physica {\bf 105} A, 445  (1981).
1373: \bibitem{arnold} L.~Arnold, \emph{Random Dynamical Systems} 
1374: (Springer-Verlag, Berlin, 1998).
1375: \bibitem{mannella} R. Mannella, in  {\em Noise in Dynamical
1376: Systems, Vol. 3: Experiments and Simulations}, edited by
1377: F.~Moss and P.V.E.~Mc Clintock (Cambridge University Press, Cambridge, 1989).
1378: \bibitem{abram} M. Abramowitz and I. A. Stegun, {\em Handbook
1379: of Mathematical Functions} 
1380: (National Bureau of Standards, Washington, D.C., 1966).
1381: \bibitem{vankampen} N.G.~van Kampen, {\it Stochastic Processes in Physics  
1382: and Chemistry} (North-Holland, Amsterdam, 1992).
1383: \bibitem{graham} A.~Schenzle and H.~Brand, Phys. Rev. A {\bf 20}, 1628 (1979).
1384: \bibitem{hansel} D.~Hansel and J.F.~Luciani, 
1385: J. Stat. Phys. {\bf 54}, 971 (1989).
1386: \bibitem{tessieri} L.~Tessieri and F.M.~Izrailev, Phys. Rev. E
1387: {\bf 62},  3090 (2000).
1388: \bibitem{imkeller} P.~Imkeller and C.~Lederer, 
1389: Dyn. and Stab. Syst. {\bf 14},  385 (1999).
1390: 
1391: \end{thebibliography} 
1392: 
1393: 
1394: \end{document}
1395: