1: \documentclass[aps,pre,twocolumn,showpacs]{revtex4}
2: \usepackage{amssymb,amsmath}
3: \usepackage[dvips]{graphicx}
4:
5: \begin{document}
6:
7: \title{Onset of Collective Oscillation in Chemical Turbulence
8: under Global Feedback}
9:
10: \author{Yoji Kawamura}
11: \email{kawamura@ton.scphys.kyoto-u.ac.jp}
12:
13: \author{Yoshiki Kuramoto}
14:
15: \affiliation{Department of Physics, Graduate School of Sciences,
16: Kyoto University, Kyoto 606-8502, Japan}
17:
18: \date{May 30, 2003}
19:
20: \pacs{05.45.-a, 47.27.-i, 82.40.-g}
21:
22: %%%%% abstract
23: \begin{abstract}
24: Preceding the complete suppression of chemical turbulence by means
25: of global feedback, a different universal type of transition, which
26: is characterized by the emergence of small-amplitude collective
27: oscillation with strong turbulent background, is shown to occur at
28: much weaker feedback intensity.
29: %%
30: We illustrate this fact numerically in combination with a phenomenological
31: argument based on the complex Ginzburg-Landau equation with global feedback.
32: \end{abstract}
33:
34: \maketitle
35:
36: %%%%% section 1
37: \section{Introduction} \label{sec:intro}
38:
39: Chemical turbulence in oscillatory reaction-diffusion systems
40: can be completely suppressed by means of global feedback
41: ~\cite{ref:battogtokh96,ref:battogtokh97,ref:kim01}.
42: %%
43: Theoretically, this fact was found in the complex Ginzburg-Landau equation
44: ~\cite{ref:battogtokh96,ref:battogtokh97}, i.e., the normal form
45: of oscillatory reaction-diffusion systems near the supercritical
46: Hopf bifurcation point~\cite{ref:kuramoto84}.
47: %%
48: Recent experiments on catalytic CO oxidation on Pt surface demonstrated
49: the same fact, revealing also a variety of wave patterns caused by the
50: effects of global delayed feedback~\cite{ref:kim01,ref:bertram03-1}.
51: %%
52: A theoretical model for this reaction system reproduced similar
53: behavior~\cite{ref:kim01,ref:bertram03-2}.
54:
55: In the present paper, we show that yet another transition of
56: universal nature can occur at a certain feedback intensity which
57: is much weaker than the critical intensity associated with the
58: complete suppression of turbulence.
59: %%
60: The new type of transition is characterized by the emergence of small-amplitude
61: collective oscillation out of the strongly turbulent medium without
62: long-range phase coherence.
63: %%
64: When the collective oscillation appeared, the system remains strongly turbulent,
65: while the effective damping rate of the uniform mode (i.e., the mean field) shows
66: a change of sign from positive to negative.
67: %%
68: Thus, the transition is interpreted as a consequence of a complete cancellation
69: of the effective damping of the mean field with the effect of its growth
70: produced by the global feedback.
71:
72: In Sec.~\ref{sec:langevin}, we start with the complex
73: Ginzburg-Landau equation (CGL) with global feedback.
74: %%
75: Then we derive phenomenologically a nonlinear Langevin equation governing
76: the mean field in the form of a noisy Stuart-Landau equation (SL).
77: %%
78: In order to clarify the nature of the transition of our concern,
79: some numerical results for the one- and two- dimensional CGL will
80: be compared in Secs.~\ref{sec:results} and \ref{sec:two},
81: respectively, with analytical results obtained from the noisy SL.
82: %%
83: Concluding remarks will be given in the final section.
84:
85:
86: %%%%% section 2
87: \section{Langevin equation for turbulent CGL as an effective equation}
88: \label{sec:langevin}
89:
90: One-dimensional complex Ginzburg-Landau equation with global feedback
91: is given by~\cite{ref:battogtokh96,ref:battogtokh97}
92: %%% eq.(1)
93: \begin{equation}
94: \partial_t W = W+\left(1+i c_1\right)\partial^2_x W
95: -\left(1+i c_2\right)\left|W\right|^2 W+\mu\bar{W},
96: \label{eq:cgl}
97: \end{equation}
98: %%% eq.(2)
99: \begin{equation}
100: \bar{W}\left(t\right) = \frac{1}{L}\int_0^{L}W\left(t, x\right)dx,
101: \label{eq:gf}
102: \end{equation}
103: %%
104: where $W$ is a complex field, and $L$ is the system size which
105: is supposed to be sufficiently large.
106: %%
107: The intensity $\mu$ of the global feedback is generally a complex number.
108: %%
109: It is known, however, that a suitable tuning of the delay time in the
110: feedback in the original system can control the phase of this parameter
111: ~\cite{ref:battogtokh96,ref:battogtokh97}.
112: %%
113: For the sake of simplicity, therefore, we shall confine our present analysis
114: to the case of real $\mu$, which corresponds to the situation where the delay
115: time in the feedback is fixed at a certain value but the feedback intensity
116: is allowed to vary.
117: %%
118: A brief comment will be made on the case of complex $\mu$ in the final section.
119:
120: We first consider the system without feedback ($\mu=0$), i.e., the usual
121: one-dimensional CGL~\cite{ref:aranson02,ref:shraiman92}.
122: %%
123: As is well known, uniform oscillations are linearly unstable and
124: turbulence develops when the Benjamin-Feir instability condition
125: %%% eq.(3)
126: \begin{equation}
127: 1+c_1 c_2<0
128: \label{eq:bf}
129: \end{equation}
130: %%
131: is satisfied.
132: %%
133: In what follows, we will fix the parameters $c_1$ and $c_2$ as $c_1=2.0$
134: and $c_2=-2.0$ so that the system may stay well within the turbulent regime.
135: %%
136: We confirmed that under this condition no collective oscillation exists,
137: i.e., $\bar{W}$ is randomly fluctuating on a ``microscopic'' scale around
138: the zero value without perceptible systematic motion.
139: %%
140: The core of our argument developed below depends little on the choice of
141: parameter values as far as the condition~(\ref{eq:bf}) is well satisfied.
142:
143: It is known that if the turbulence is sufficiently strong, which
144: is actually the case under the above parameter condition, the
145: system exhibits extensive chaos characterized by the property
146: ~\cite{ref:egolf94}
147: %%% eq.(4)
148: \begin{equation}
149: D_{\rm f} \propto L,
150: \label{eq:ec}
151: \end{equation}
152: %%
153: where $D_{\rm f}$ is the Lyapunov dimension of the high-dimensional
154: chaotic attractor describing the turbulence.
155: %%
156: Extensive chaos implies that the system can be imagined as composed
157: of a large number of cells of equal size such that the fluctuations
158: of some variables associated with the individual cells about their
159: mean value are statistically independent from cell to cell.
160: %%
161: Thus, the fluctuations of a macro-variable, i.e., a variable given by
162: a simple sum of cell variables over the entire system, are expected
163: to obey the central limit theorem.
164: %%
165: In particular, in the absence of long-range order, the characteristic
166: amplitude of $\bar{W}$ will scale like $1/\sqrt{L}$ for large system size.
167: %%
168: If we represent our continuous oscillatory medium with a long array of $N$
169: oscillators with sufficiently small but fixed separation between neighboring
170: oscillators, which we actually do in numerical simulations to be described below,
171: we expect asymptotically the property
172: %%% eq.(5)
173: \begin{equation}
174: \bar{W}=\frac{1}{N}\sum_{j=1}^N W_j=O\left(\frac{1}{\sqrt{N}}\right),
175: \label{eq:clt}
176: \end{equation}
177: %%
178: where $W_j$ is the complex amplitude of the $j$-th oscillator in the array.
179: %%
180: The simple nature of the extensive variable $\bar{W}$ mentioned above
181: also implies that its time-correlation function defined by
182: %%% eq.(6)
183: \begin{equation}
184: C\left(t\right) = \left\langle\bar{W}^{\ast}(0)\bar{W}(t)\right\rangle
185: \equiv\lim_{T\to\infty}\frac{1}{T}\int_0^T
186: \bar{W}^{\ast}(s)\bar{W}(s+t)ds
187: \label{eq:cor}
188: \end{equation}
189: obeys a simple exponential decay law for large $t$, or
190: %%% eq.(7)
191: \begin{equation}
192: %C(t) \propto e^{-\gamma t}.
193: C(t) \propto \exp\left(-\gamma t\right).
194: \label{eq:exp}
195: \end{equation}
196: %%
197: The effective damping coefficient $\gamma$ is a quantity which
198: could only be determined from the statistical mechanics of turbulent
199: fluctuations which is still far from being established.
200: %%
201: The above decay law is confirmed from our numerical simulation.
202: %%
203: Figure~\ref{fig:decay} shows numerically calculated $C(t)$ from
204: which the exponential law with $\gamma\simeq 0.22$ is confirmed except
205: for some initial transient.
206: %%
207: In this numerical simulation, and in all numerical simulations which
208: follow, we applied an explicit Euler integration scheme with a constant
209: time step $\varDelta t\leq 0.01$ and a fixed grid size $\varDelta x = 0.5$,
210: adopting periodic boundary conditions.
211: %%
212: The system size $N$ used ranges from $400$ to $200000$.
213: %%% fig.1
214: \begin{figure}
215: \centering
216: \includegraphics[width=8cm]{fig1.eps}
217: \caption{Numerically observed exponential decay of the time-correlation
218: function of the mean field in semi-logarithmic scales. The effective
219: damping coefficient $\gamma$, i.e., the mean tangent (the broken line)
220: of the curve with the initial transient excluded, is estimated to be $0.22$.
221: The system size is $N=200000$.}
222: \label{fig:decay}
223: \end{figure}
224:
225: We now introduce global feedback and study its effects on the dynamics
226: of the mean field. Let the complex amplitude be decomposed into Fourier
227: series as
228: %%% eq.(8)
229: \begin{equation}
230: W = \sum_{k=-\infty}^{\infty} \tilde{W_k}\, e^{i q_k x},
231: \label{eq:ft}
232: \end{equation}
233: where $q_k = 2\pi k/L$ ($k$ is an integer), and the Fourier amplitudes
234: are defined by
235: %%% eq.(9)
236: \begin{equation}
237: \tilde{W}_k = \frac{1}{L}\int_0^L W\, e^{-i q_k x}dx.
238: \label{eq:fa}
239: \end{equation}
240: The uniform amplitude $\tilde{W}_0$, which is identical with the mean
241: field $\bar{W}$ by definition, obeys the equation
242: %%% eq.(10)
243: \begin{equation}
244: \dot{\tilde{W_0}} = \tilde{W}_0
245: -\left(1+i c_2\right)\sum_{k_1, k_2}
246: \tilde{W}_{k_1}\tilde{W}_{k_2}\tilde{W}_{k_1+k_2}^{\ast}
247: +\mu\tilde{W}_0.
248: \label{eq:um}
249: \end{equation}
250: %%
251: One may wish to obtain an equation for the mean field in a closed form,
252: which would be a stochastic equation of the nonlinear Langevin type.
253: %%
254: However, deriving such an equation would be a formidable statistical
255: mechanical problem, so that in the present paper we will content
256: ourselves by simply assuming phenomenologically that the effective
257: exponential decay of the mean field as is seen in Fig.~\ref{fig:decay}
258: is a result of renormalization of the linear coefficient in Eq.~(\ref{eq:um})
259: by the nonlinear mode-coupling term.
260: %%
261: If we can neglect non-Markovian effects, the result of such renormalization
262: will generally take the form of a nonlinear Langevin equation
263: %%% eq.(11)
264: \begin{equation}
265: \dot{\bar{W}} =
266: \left[\left(\mu-\gamma\right)+i\omega\right]\bar{W}
267: +{\cal N}\left(\bar{W}\right)+f(t),
268: \label{eq:le}
269: \end{equation}
270: %%
271: where a nonlinear term ${\cal N}(\bar{W})$ with unknown specific form
272: has been included, and $f(t)$ represents random force with vanishing mean.
273:
274: The analysis of Eq.~(\ref{eq:le}) developed in the following section is
275: based on the simplifying assumption that $f(t)$ is white Gaussian with
276: the only non-vanishing second moment given by
277: %%% eq.(12)
278: \begin{equation}
279: \left\langle f(t)f^{\ast}(t')\right\rangle=4\Gamma\delta(t-t').
280: \label{eq:n2}
281: \end{equation}
282: %%
283: The Gaussian nature of the random force seems to hold due to the
284: aforementioned extensive nature of $\bar{W}$.
285: %%
286: The white-noise assumption also seems valid because we are particularly
287: concerned with the situation near the transition point where the
288: characteristic time scale of $\bar{W}$ becomes very long.
289: %%
290: We should also note that the damping coefficient $\gamma$ has been assumed
291: to be unchanged when the global feedback is introduced, which is actually the
292: property confirmed by our numerical experiments at least for real $\mu$.
293: %%
294: Our final remark is that in the above argument about Eq.~(\ref{eq:le})
295: we did not explicitly refer to spatial dimension.
296: %%
297: Therefore, there seems to be no reason why Eq.~(\ref{eq:le}) should
298: not be applied to systems of two or higher dimensions.
299:
300: Equation~(\ref{eq:le}) tells that a transition occurs when
301: the global feedback intensity $\mu$ becomes equal to the effective
302: damping coefficient $\gamma$ of the mean filed.
303: %%
304: We denote this value of $\mu$ as $\mu_{\rm c}$, or
305: %%% eq.(13)
306: \begin{equation}
307: \mu_{\rm c} = \gamma.
308: \label{eq:muc}
309: \end{equation}
310: %%
311: Figure~\ref{fig:over} shows numerically obtained long-time averages
312: of the mean field amplitude, denoted as $\langle r\rangle$, as a function
313: of the feedback intensity $\mu$ over a wide range of $\mu$.
314: %%
315: Although not very clear from these data, there is an indication of transition from
316: vanishing to non-vanishing $\langle r\rangle$ at some value of $\mu$ much smaller
317: than those giving rise to a hysteresis between turbulent and non-turbulent
318: uniform states.
319: %%
320: In fact, the whole analysis in the rest of this paper is devoted to finding
321: out unambiguous evidence for the existence of a transition through a closer
322: investigation of small-$\mu$ region.
323: %%% fig.2
324: \begin{figure}
325: \centering
326: \includegraphics[width=8cm]{fig2.eps}
327: \caption{Long-time averages of the mean field amplitude
328: $\langle r\rangle$ as a function of the feedback intensity $\mu$
329: over a wide range of $\mu$. The dotted line corresponds to
330: uniform oscillations. The system size is $N=10000$.}
331: \label{fig:over}
332: \end{figure}
333:
334: In Fig.~\ref{fig:stp}, two space-time plots of the
335: complex amplitude modulus for $\mu=0.1$ (weaker feedback)
336: and $\mu=0.3$ (stronger feedback) are indistinguishable.
337: %%
338: In Fig.~\ref{fig:phase}, however, two phase portraits
339: in the complex amplitude plane each obtained for $\mu=0.1$
340: and $\mu=0.3$ are contrasted with each other.
341: %%
342: While the distribution of the representative points of the local
343: oscillators looks almost isotropic for the case of $\mu=0.1$,
344: implying the absence of collective oscillations, such symmetry is
345: obviously lost when $\mu=0.3$, implying the existence of collective
346: oscillations.
347: %%
348: Qualitative difference between the two situations is further confirmed
349: from Fig.~\ref{fig:orbit} where a trajectory of the mean field
350: over a long time at each value of $\mu$ is displayed.
351: %%
352: It is clear that when the feedback is weak the mean field is non-oscillatory,
353: simply fluctuating (presumably due to the finite-size effects described above)
354: around the origin, whereas for stronger feedback the same quantity clearly
355: exhibits a closed orbit with some amplitude fluctuation again due to the
356: finite-size effects.
357: %%
358: Thus, if there is a transition somewhere between these $\mu$-values, it is
359: presumably characterized by a noisy Hopf bifurcation.
360: %%% fig.3
361: \begin{figure}
362: \centering
363: \begin{tabular}{c c c}
364: & (a) & (b) \\[1mm]
365: \rotatebox{90}{\hspace{1.2cm}{\Large time $\rightarrow$}} &
366: \includegraphics[width=4cm]{fig3a.eps} &
367: \includegraphics[width=4cm]{fig3b.eps}
368: \end{tabular}
369: \caption{Space (horizontal) - time (vertical) plot of the complex
370: amplitude modulus $|W|$ for $\mu=0.1$ (a) and $\mu=0.3$ (b).
371: The system size is $N=400$.}
372: \label{fig:stp}
373: \end{figure}
374: %%% fig.4
375: \begin{figure}
376: \centering
377: \includegraphics[width=8cm]{fig4a.eps} \\
378: \includegraphics[width=8cm]{fig4b.eps}
379: \caption{Phase portraits for $\mu=0.1$ (a) and $\mu=0.3$ (b)
380: at a given time. The system size is $N=40000$.}
381: \label{fig:phase}
382: \end{figure}
383: %%% fig.5
384: \begin{figure}
385: \centering
386: \includegraphics[width=8cm]{fig5.eps}
387: \caption{Trajectories of the mean field in the complex plane
388: for $\mu=0.1$ and $\mu=0.3$. The mean field is fluctuating
389: around the origin when $\mu=0.1$, while it forms a closed
390: orbit with small amplitude fluctuations when $\mu=0.3$.
391: The fluctuations come from the finiteness of the system size.
392: The system size is $N=40000$.}
393: \label{fig:orbit}
394: \end{figure}
395:
396: The nonlinear Langevin equation~(\ref{eq:le}) actually predicts
397: the occurrence of a noisy Hopf bifurcation.
398: %%
399: From the symmetry of our system and the smallness in amplitude of the collective
400: oscillation near its onset, the nonlinear effects ${\cal N}(\bar{W})$ in
401: Eq.~(\ref{eq:le}) will be dominated by a cubic term.
402: %%
403: Then the mean field obeys a noisy Stuart-Landau equation
404: %%% eq.(14)
405: \begin{equation}
406: \dot{\bar{W}}
407: =\left[\left(\mu-\mu_{\rm c}\right)+i\omega\right]\bar{W}
408: -\left(\eta+i\alpha\right)\left|\bar{W}\right|^2\bar{W}+f(t),
409: \label{eq:sl}
410: \end{equation}
411: %%
412: where the coefficients $\mu_{\rm c}$, $\omega$, $\eta(>0)$, and $\alpha$ are
413: real and depend generally on $c_1$ and $c_2$.
414: %%
415: Since the above equation can be handled analytically, it would be interesting
416: to compare some of the results from its analysis with our direct numerical
417: simulations on Eq.~(\ref{eq:cgl}) and its two-dimensional extension,
418: which are the subjects of Secs.~\ref{sec:results} and \ref{sec:two},
419: respectively.
420:
421:
422: %%%%% section 3
423: \section{Predicted critical behavior and comparison with numerical results}
424: \label{sec:results}
425:
426: The nonlinear Langevin equation~(\ref{eq:sl}) is equivalent
427: with the Fokker-Planck equation of the following form
428: ~\cite{ref:risken89,ref:kampen81}:
429: %%% eq.(15)
430: \begin{align}
431: \frac{\partial}{\partial t} P\left(r,\phi,t\right)
432: &=\frac{1}{r}\frac{\partial}{\partial r}
433: \left\{\left[-\left(\mu-\mu_{\rm c}\right)r^2+\eta r^4\right]P
434: +\Gamma r \frac{\partial}{\partial r}P\right\} \nonumber \\
435: &+\left[\left(-\omega+\alpha r^2\right)
436: \frac{\partial}{\partial \phi}P
437: +\frac{\Gamma}{r^2}\frac{\partial^2}{\partial \phi^2}P\right].
438: \label{eq:fp}
439: \end{align}
440: %%
441: Here $P(r,\phi,t)$ is the probability density for the amplitude $r$
442: and the phase $\phi$ of the mean field at time $t$.
443: %%
444: The above equation admits a stationary solution independent of $\phi$
445: given by
446: %%% eq.(16)
447: \begin{equation}
448: P_{\rm st}\left(r\right) = \exp \left[-\frac{\eta}{4\Gamma}
449: \left(r^2-\frac{\mu-\mu_{\rm c}}{\eta}\right)^2\right].
450: \label{eq:st1}
451: \end{equation}
452: Various moments of $r$ defined by
453: %%% eq.(17)
454: \begin{equation}
455: \left\langle r^n\right\rangle =
456: \frac{\displaystyle \int_0^\infty r^n P_{\rm st}\left(r\right)r dr}
457: {\displaystyle \int_0^\infty P_{\rm st}\left(r\right)r dr},
458: \label{eq:rn}
459: \end{equation}
460: %%
461: and those of the fluctuation $\delta r\equiv r-\langle r\rangle$
462: can be calculated~\cite{ref:risken65}.
463: %%
464: In particular, in the limit of weak random force, the mean field amplitude
465: $\langle r\rangle$ near the transition is found to depend on $\mu-\mu_{\rm c}$ as
466: %%% eq.(18)
467: \begin{equation}
468: \left\langle r\right\rangle=
469: \begin{cases}
470: A\left(\mu-\mu_{\rm c}\right)^{1/2} &
471: \quad\left(\mu>\mu_{\rm c}\right), \\
472: 0 & \quad\left(\mu<\mu_{\rm c}\right),
473: \end{cases}
474: \label{eq:r1}
475: \end{equation}
476: %%
477: where $A$ is a constant.
478: %%
479: Similarly, the mean field fluctuation $\langle (\delta r)^2\rangle$ is given by
480: %%% eq.(19)
481: \begin{equation}
482: \left\langle(\delta r)^2\right\rangle\propto\left|\mu-\mu_{\rm c}\right|^{-1},
483: \label{eq:dr2}
484: \end{equation}
485: %%
486: which holds for $\mu\gtrless\mu_{\rm c}$.
487: %%
488: It is clear that the critical exponents associated with $\langle r\rangle$
489: and $\langle(\delta r)^2\rangle$ obey the classical law or the mean field theory,
490: reflecting the fact that the transition is caused by the applied mean field and
491: not by the development of local order to a macroscopic scale.
492: %%
493: Still the transition is in some sense statistical in nature unlike bifurcations
494: in deterministic dynamical systems.
495: %%
496: This is because the loss of long-range order is solely due to turbulent fluctuations,
497: so that a full theoretical understanding of the transition phenomenon would be
498: impossible without statistical mechanics of chemical turbulence.
499:
500: Long-time averages of $r$ and $(\delta r)^2$ were obtained as a function
501: of the feedback intensity from numerical simulation of Eqs.~(\ref{eq:cgl})
502: and (\ref{eq:gf}) with $N=200000$, and the results are shown in
503: Figs.~\ref{fig:exponent}(a) and (b), respectively.
504: %%
505: By assuming that the long-time averages are identical with ensemble averages,
506: these numerical data were fitted with the theoretical curves given by
507: Eqs.~(\ref{eq:r1}) and (\ref{eq:dr2}), respectively, where a scale
508: factor is used as the only adjustable parameter.
509: %%
510: Note that $\mu_{\rm c}$ is not an adjustable parameter, but is a constant
511: given by Eq.~(\ref{eq:muc}).
512: %%% fig.6
513: \begin{figure}
514: \centering
515: \includegraphics[width=8cm]{fig6a.eps} \\
516: \includegraphics[width=8cm]{fig6b.eps}
517: \caption{$\langle r\rangle^2$ vs.\ $\mu$ (a).
518: $N^{-1}\langle(\delta r)^2\rangle^{-1}$ vs.\ $\mu$ (b).
519: The open circles and the solid lines are the numerical data
520: and the theoretical curves, respectively. In each of (a) and (b),
521: a suitable scale factor is used as an adjustable parameter to
522: achieve a good fitting between the theory and numerical experiments.
523: The system size is $N=200000$.}
524: \label{fig:exponent}
525: \end{figure}
526:
527: In obtaining Eq.~(\ref{eq:r1}) for the mean field amplitude,
528: we considered the limit of weak random force.
529: %%
530: Let our argument be generalized to include the dependence of
531: $\langle r\rangle$ on the noise intensity as well as on $\mu-\mu_{\rm c}$.
532: %%
533: Because the origin of the random force driving the mean field is the
534: finiteness of the system size, the noise intensity should be inversely
535: proportional to the system size, or
536: %%% eq.(20)
537: \begin{equation}
538: \Gamma=\frac{D}{N},
539: \label{eq:fse}
540: \end{equation}
541: %%
542: where $D$ is a constant independent of $N$.
543: %%
544: One may thus write the stationary distribution in the form
545: %%% eq.(21)
546: \begin{equation}
547: P_{\rm st}(r)=\exp\left[-\frac{N\eta}{4D}
548: \left(r^2-\frac{\varepsilon}{\eta}\right)^2\right],
549: \label{eq:st2}
550: \end{equation}
551: %%
552: where $\varepsilon=\mu-\mu_{\rm c}$. Applying the finite-size scaling
553: law developed in Ref.~\cite{ref:pikovsky99} to the average amplitude
554: of the mean field, we obtain a scaling form
555: %%% eq.(22)
556: \begin{equation}
557: \left\langle r\right\rangle=N^{-1/4} F\left(\varepsilon N^{1/2}\right),
558: \label{eq:fss}
559: \end{equation}
560: %%
561: where $F$ is a function (called the scaling function) depending on
562: $\varepsilon$ and $N$ only through $\varepsilon N^{1/2}$.
563: %%
564: Equation~(\ref{eq:fss}) is a generalization of Eq.~(\ref{eq:r1}).
565: %%
566: Numerically calculated $\langle r\rangle$ for various $\varepsilon$ and $N$
567: confirms this scaling law.
568: %%
569: Figure~\ref{fig:finite}(a) shows the dependence of the long-time
570: average of the mean field amplitude on feedback intensity for some
571: different values of $N$.
572: %%
573: As is seen from Fig.~\ref{fig:finite}(b), all these data come to lie
574: on an identical universal curve after the rescalings of $\langle r\rangle$
575: and $\varepsilon$ by $N^{1/4}$ and $N^{1/2}$, respectively.
576: %%
577: In this way, the finite-size scaling law~(\ref{eq:fss}) is confirmed,
578: providing unmistakable evidence for a phase transition.
579: %%% fig.7
580: \begin{figure}
581: \centering
582: \includegraphics[width=8cm]{fig7a.eps} \\
583: \includegraphics[width=8cm]{fig7b.eps}
584: \caption{$\langle r\rangle$ vs.\ $\mu$ for some different values of $N$ (a).
585: Rescaling of the data in (a) according to the finite-size scaling
586: given by Eq.~(\ref{eq:fss}) produces a universal curve independent
587: of $N$ (b).}
588: \label{fig:finite}
589: \end{figure}
590:
591:
592: %%%%% section 4
593: \section{Two-dimensional case} \label{sec:two}
594:
595: The one-dimensional reaction-diffusion systems which we have
596: numerically studied in Secs.~\ref{sec:langevin} and
597: \ref{sec:results} are not very realistic.
598: %%
599: Especially, surface chemical reactions, such as catalytic
600: CO oxidation on Pt, occur in two dimensions.
601: %%
602: As mentioned in the foregoing sections, the transition considered here
603: is the mean field type, so that the nature of the transition is expected
604: to be the same as that in one-dimensional systems.
605: %%
606: We carried out numerical simulations on the two-dimensional complex
607: Ginzburg-Landau equation with global feedback, and obtained a clear
608: indication of transition, although the corresponding value of
609: $\mu_{\rm c}$ is considerably smaller than that of the one-dimensional
610: case under the same parameter condition.
611: %%
612: Figure~\ref{fig:two} summarizes numerical results for
613: various $\varepsilon$ and $N$, exhibited in a similar manner
614: to Fig.~\ref{fig:finite}(b), i.e., in the form of
615: $\langle r\rangle N^{1/4}$ vs.\ $\varepsilon N^{1/2}$
616: for different values of $N$.
617: %%
618: These data form almost an identical curve again, which we take as
619: evidence for a transition similar to that in one-dimensional systems.
620: %%% fig.8
621: \begin{figure}
622: \centering
623: \includegraphics[width=8cm]{fig8.eps}
624: \caption{Similar to Fig.~\ref{fig:finite}(b) but in two space dimensions.
625: Parameter values are the same as in the one-dimensional case, i.e., $c_1=2.0$
626: and $c_2=-2.0$, which gives $\mu_{\rm c}\simeq 0.10$.
627: This value is used in rescaling of the numerical data.}
628: \label{fig:two}
629: \end{figure}
630:
631:
632: %%%%% section 5
633: \section{Concluding remarks} \label{sec:remarks}
634:
635: Preceding the transition at which the turbulence is completely
636: suppressed and uniform oscillations set in, a different type of
637: transition characterized by the emergence of collective oscillations
638: was shown to exist in the one- and two- dimensional complex
639: Ginzburg-Landau equations with global feedback.
640: %%
641: The transition is well described phenomenologically with the noisy
642: Stuart-Landau equation governing the mean field.
643: %%
644: Since the noise there comes from the finite-size effects, the transition
645: becomes infinitely sharp in the limit of infinite system size.
646: %%
647: The critical exponents of this transition obey the mean field theory
648: because the origin of cooperativity is nothing but the mean field produced
649: by the global feedback.
650: %%
651: This also gives the reason why spatial dimension one is sufficient for
652: giving rise to the transition.
653: %%
654: We also confirmed that the two-dimensional complex Ginzburg-Landau equation
655: with global feedback which is more realistic also exhibits a transition of
656: the same type.
657:
658: Throughout the present paper, our analysis was confined to the case
659: that the feedback intensity $\mu$ is real.
660: %%
661: From our ongoing study, it is being confirmed that a transition of the same
662: nature persists over some range of complex $\mu$.
663: %%
664: The nonlinear Langevin equation~(\ref{eq:le}) also seems to remain valid.
665: %%
666: Unlike the case of real $\mu$, however, the effective damping coefficient
667: $\gamma$ appearing in the Langevin equation~(\ref{eq:le}) seems to
668: depend on $\mu$, which raises an interesting theoretical problem to be
669: tackled in the future.
670:
671: Our analysis~\cite{ref:kawamura06} also suggests that the realistic dynamical
672: model for the catalytic CO oxidation~\cite{ref:kim01,ref:bertram03-2}
673: under experimentally accessible parameter conditions also exhibit a
674: similar transition.
675: %%
676: We strongly hope for its experimental verification.
677:
678:
679: %%%%% acknowledgments
680: \begin{acknowledgments}
681: The authors thank H.~Nakao for useful discussions.
682: %%
683: Numerical computation of this work was carried out with the Computer
684: Facility at Yukawa Institute for Theoretical Physics, Kyoto University.
685: \end{acknowledgments}
686:
687:
688: %%%%% references
689: \begin{thebibliography}{99}
690:
691: \bibitem{ref:battogtokh96}
692: D.~Battogtokh and A.~S.~Mikhailov,
693: Physica D {\bf 90}, 84 (1996).
694:
695: \bibitem{ref:battogtokh97}
696: D.~Battogtokh, A.~Preusser, and A.~S.~Mikhailov,
697: Physica D {\bf 106}, 327 (1997).
698:
699: \bibitem{ref:kim01}
700: %M.~Kim {\em et al.},
701: M.~Kim, M.~Bertram, M.~Pollmann, A.~von Oertzen,
702: A.~S.~Mikhailov, H.~H.~Rotermund, and G.~Ertl,
703: Science {\bf 292}, 1357 (2001).
704:
705: \bibitem{ref:kuramoto84}
706: Y.~Kuramoto,
707: {\em Chemical Oscillations, Waves, and Turbulence}
708: (Springer, New York, 1984; Dover, New York, 2003).
709:
710: \bibitem{ref:bertram03-1}
711: %M.~Bertram {\em et al.},
712: M.~Bertram, C.~Beta, M.~Pollmann,
713: A.~S.~Mikhailov, H.~H.~Rotermund, and G.~Ertl,
714: Phys. Rev. E {\bf 67}, 036208 (2003).
715:
716: \bibitem{ref:bertram03-2}
717: M.~Bertram and A.~S.~Mikhailov,
718: Phys. Rev. E {\bf 67}, 036207 (2003).
719:
720: \bibitem{ref:aranson02}
721: I.~S.~Aranson and L.~Kramer,
722: Rev. Mod. Phys. {\bf 74}, 99 (2002).
723:
724: \bibitem{ref:shraiman92}
725: %B.~I.~Shraiman {\em et al.},
726: B.~I.~Shraiman, A.~Pumir, W.~van Saarloos,
727: P.~C.~Hohenberg, H.~Chat\'e, and M.~Holen,
728: Physica D {\bf 57}, 241 (1992).
729:
730: \bibitem{ref:egolf94}
731: D.~A.~Egolf and H.~S.~Greenside,
732: Nature {\bf369}, 129 (1994).
733:
734: \bibitem{ref:risken89}
735: H.~Risken,
736: {\em The Fokker-Planck Equation}
737: (Springer, Berlin, 1989).
738:
739: \bibitem{ref:kampen81}
740: N.~G.~van Kampen,
741: {\em Stochastic Process in Physics and Chemistry}
742: (North-Holland, Amsterdam, 1981).
743:
744: \bibitem{ref:risken65}
745: H.~Risken,
746: Z. Phys. {\bf 186}, 85 (1965).
747:
748: \bibitem{ref:pikovsky99}
749: A.~Pikovsky and S.~Ruffo,
750: Phys. Rev. E {\bf 59}, 1633 (1999).
751:
752: \bibitem{ref:kawamura06}
753: Y.~Kawamura and Y.~Kuramoto,
754: Prog. Theor. Phys. Suppl. {\bf 161}, 216 (2006).
755:
756: \end{thebibliography}
757:
758: \end{document}
759: