1: \documentstyle[prl,aps,floats,epsf,color]{revtex}

2: \addtolength{\topmargin}{2.0cm}

3: \begin{document}

4: \baselineskip=12pt

5: \def\be{\begin{equation}}

6: \def\ee{\end{equation}}

7: \def\bea{\begin{eqnarray}}

8: \def\eea{\end{eqnarray}}

9: \def\E{{\rm e}}

10: \def\bearst{\begin{eqnarray*}}

11: \def\eearst{\end{eqnarray*}}

12: \def\peleven{\parbox{11cm}}

13: \def\peffec{\peight{\bearst\eearst}\hfill\peleven}

14: \def\pspace{\peight{\bearst\eearst}\hfill}

15: \def\ptwelve{\parbox{12cm}}

16: \def\peight{\parbox{8mm}}

17: \twocolumn

18: [\hsize\textwidth\columnwidth\hsize\csname@twocolumnfalse\endcsname

19: 

20: 

21: \title

22: {  Regeneration of Stochastic Processes: An Inverse Method }

23: 

24: \author

25: { F. Ghasemi$^1$,  J. Peinke,$^2$, Muhammad Sahimi,$^3$ and  M.

26: Reza Rahimi Tabar,$^{1,4}$}

27: 

28: \vskip 1cm

29: 

30: \address

31: { $^1$Department of Physics, Sharif University of Technology, P.O.

32: Box

33: 11365-9161, Tehran 11365, Iran\\

34: $^2$Carl von Ossietzky University, Institute of Physics, D-26111

35: Oldenburg, Germany\\

36: $^4$Department of Chemical Engineering, University of Southern

37: California, Los Angeles, California 90089-1211, USA \\

38: $^3$CNRS UMR 6529, Observatoire de la C$\hat o$te d'Azur, BP 4229,

39: 06304 Nice Cedex 4, France}

40:  \maketitle

41: %\date{00/07/2000}

42: %\maketitle

43: 

44: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

45: %ABSTRACT

46: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

47: 

48: 

49: \begin{abstract}

50: 

51: We propose a novel inverse method that utilizes a set of data to

52: construct a simple equation that governs the stochastic process

53: for which the data have been measured, hence enabling us to

54: reconstruct the stochastic process. As an example, we analyze the

55: stochasticity in the beat-to-beat fluctuations in the heart rates

56: of healthy subjects as well as those with congestive heart

57: failure. The inverse method provides a novel technique for

58: distinguishing the two classes of subjects in terms of a drift and

59: a diffusion coefficients which behave completely differently for

60: the two classes of subjects, hence potentially providing a novel

61: diagnostic tool for distinguishing healthy subjects from those

62: with congestive heart failure, even at the early stages of the

63: disease development.

64: 

65: 

66: \pacs{ 05.10.Gg, 05.40.-a, 05.45.Tp }

67: \end{abstract}

68: \hspace{.3in}

69: \newpage

70: ]

71: 

72: \noindent{\bf 1. Introduction}

73: 

74: Many natural or man-made phenomena, as well as the morphology of

75: numerous physical systems, are charactertized by a degree of

76: stochasticity. Turbulent flows, fluctuations in the stocks prices,

77: seismic recordings, the internet traffic, pressure fluctuations in

78: chemical reactors, and the surface roughness of many materials and

79: rock [1,2] are but a few examples of such phenomena and systems. A

80: long standing problem has been the development of an effective

81: reconstruction method for such phenomena. That is, given a set of

82: data for certain characteristics of a phenomenon, one would like

83: to develop an effective equation that can reproduce the data with

84: an accuracy comparable to the measured data. If such a method can

85: be developed, one may utilize it to, (1) reconstruct the original

86: process with similar statistical properties, and (2) understand

87: the nature and properties of the stochastic process.

88: 

89: In this paper we use a novel method to address this general

90: problem. The proposed method utilizes a set of data for a

91: phenomenon which contains a degree of stochasticity and constructs

92: a simple equation that governs the phenomenon. As we show below,

93: in addition to being highly accurate, the method is quite general;

94: it is capable of providing a rational explanation for complex

95: features of the phenomenon; it requires no scaling feature, and it

96: enables us to accomplish the tasks listed above. As an example, we

97: apply the method to analyze cardiac interbeat intervals which

98: normally fluctuate in a complex manner. We show that the

99: application of the method to the analysis of interbeat

100: fluctuations in the heart rates may potentially lead to a novel

101: method for distinguishing healthy subjects from those with

102: congestive heart failure (CHF).

103: 

104: \bigskip

105: \noindent{\bf 2. The Method}

106: 

107: \bigskip

108: We begin by describing the steps that lead to the development of a

109: stochastic equation, based on the (stochastic) data set, which is

110: then utilized to reconstruct the original data, as well as an

111: equation that describes the phenomenon.

112: 

113: (1) As the first step we check whether the data follow a Markov

114: chain and, if so, estimate the Markov time (length) scale $t_M$.

115: As is well-known, a given process with a degree of randomness or

116: stochasticity may have a finite or an infinite Markov time

117: (length) scale. The Markov time (length) scale is the minimum time

118: interval over which the data can be considered as a Markov process

119: [3-6]. To determine the Markov scale $t_M$, we note that a

120: complete characterization of the statistical properties of

121: stochastic fluctuations of a quantity $x$ in terms of a parameter

122: $t$ requires the evaluation of the joint probability distribution

123: function (PDF) $P_n(x_1,t_1;\cdots;x_n,t_n)$ for an arbitrary $n$,

124: the number of the data points. If the phenomenon is a Markov

125: process, an important simplification can be made, as the $n$-point

126: joint PDF, $P_n$, is generated by the product of the conditional

127: probabilities $p(x_{i+1},t_{i+1}|x_i,t_i)$, for $i=1,\cdots,n-1$.

128: A necessary condition for a stochastic phenomenon to be a Markov

129: process is that the Chapman-Kolmogorov (CK) equation [7],

130: \begin{equation}

131: p(x_2,t_2|x_1,t_1)=\int\hbox{d}(x_3)\;p(x_2,t_2|x_3,t_3)\;p

132: (x_3,t_3|x_1,t_1)\;,

133: \end{equation}

134: should hold for any value of $t_3$ in the interval $t_2<t_3<t_1$.

135: One should check the validity of the CK equation for different

136: $x_1$ by comparing the directly-evaluated conditional probability

137: distributions $p(x_2,t_2|x_1,t_1)$ with the ones calculated

138: according to right side of Eq. (1). The simplest way to determine

139: $t_M$ for stationary or homogeneous data is the numerical

140: calculation of the quantity,

141: $S=|p(x_2,t_2|x_1,t_1)-\int\hbox{d}x_3

142: p(x_2,t_2|x_3,t_3)\,p(x_3,t_3|x_1,t_1)|$, for given $x_1$ and

143: $x_2$, in terms of, for example, $t_3-t_1$ and considering the

144: possible errors in estimating $S$. Then, $t_M=t_3-t_1$ for that

145: value of $t_3-t_1$ for which $S$ vanishes or is nearly zero

146: (achieves a minimum).

147: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

148: \begin{figure}

149: \epsfxsize=7truecm\epsfbox{1.eps}

150: \epsfxsize=7truecm\epsfbox{2.eps}

151:  \narrowtext \caption{ Interbeats fluctuations of healthy subjects

152: (top), and those with congestive heart failure (bottom).}

153:  \end{figure}

154: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

155: 

156: (2) Deriving an effective stochastic equation that describes the

157: fluctuations of the quantity $x(t)$ constitutes the second step.

158: The CK equation yields an evolution equation for the change of the

159: distribution function $P(x,t)$ across the scales $t$. The CK

160: equation, when formulated in differential form, yields a master

161: equation which takes the form of a Fokker-Planck equation:

162: \begin{equation}

163: \frac{d}{dt}P(x,t)=\left[-\frac{\partial}{\partial x}D^{(1)}(x,t)+

164: \frac{\partial^2}{\partial x^2}D^{(2)}(x,t)\right]P(x,t)\;.

165: \end{equation}

166: The drift and diffusion coefficients, $D^{(1)}(x,t)$ and

167: $D^{(2)}(x,t)$, are estimated directly from the data and the

168: moments $M^{(k)}$ of the conditional probability distributions:

169: \begin{eqnarray}

170: && D^{(k)}(x,t)=\frac{1}{k!}\hskip .2cm lim_{\Delta t\to

171: 0}M^{(k)},

172: \cr\nonumber\\

173: && M^{(k)}=\frac{1}{\Delta t}\int dx'(x'-x)^k p(x',t+\Delta

174: t|x,t).

175: \end{eqnarray}

176: We note that this Fokker-Planck equation is equivalent to the

177: following Langevin equation [8]:

178: \begin{equation}

179: \frac{d}{dt}x(t)=D^{(1)}(x)+ \sqrt{D^{(2)}(x)}\;\; f(t)\;,

180: \end{equation}

181: where $f(t)$ is a random force with zero mean and Gaussian

182: statistics, $\delta$-correlated in $t$, i.e., $\langle

183: f(t)f(t')\rangle=2\delta(t-t')$. Note that such a reconstruction

184: of a stochastic process does {\it not} imply that the data do not

185: contain any correlations, or that the above formulation ignores

186: the correlations.

187: 

188: (3) Regeneration of the stochastic process constitutes the last

189: step. Equation (4) enables us to regenerate a stochastic quantity

190: which is similar to the original one {\it in the statistical

191: sense}. The stochastic process is regenerated by iterating Eq. (4)

192: which yields a series of data {\it without memory}. To compare the

193: regenerated data with the original ones, we must take the spatial

194: (or temporal) interval in the numerical discretization of Eq. (4)

195: to be unity (or renormalize it to unity). However, the Markov

196: length or time is typically greater than unity. Therefore, we

197: should correlate the data over the Markov length or time scale.

198: There are a number of methods to correlate the generated data in

199: this interval [8-12]. Here, we propose a new technique which we

200: refer to as the {\it kernel} method, according to which one

201: considers a kernel function $K(u)$ that satisfies the condition

202: that,

203: \begin{equation}

204: \int_{-\infty}^\infty K(u)du=1\;,

205: \end{equation}

206: such that the data are determined by

207: \begin{equation}

208: x(t)=\frac{1}{nh}\sum_{i=1}^nx(t_i)K\left(\frac{t-t_i}{h}\right)\;,

209: \end{equation}

210: where $h$ is the window width. For example, one of the most useful

211: kernels is the standard normal density function,

212: $K(u)=(2\pi)^{-1/2}\exp(-\frac{1}{2} u^2)$. In essence, the kernel

213: method represents the data as a sum of `bumps' placed at the

214: observation points, with its function determining the shape of the

215: bumps, and its window width $h$ fixing their width. It is evident

216: that, over the scale $h$, the kernel method correlates the data to

217: each other.

218: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

219: \begin{figure}

220: \epsfxsize=7truecm\epsfbox{fig2.eps} \narrowtext \caption{ Test of

221: Chapman-Kolmogorov equation for $x_1=-5$, $x_1=0$ and $x_1=5$. The

222: bold and open symbols represent, respectively, the

223: directly-evaluated PDF and the integrated PDF. The PDFs are

224: shifted in the vertical directions for better presentation. Values

225: of $x$ are measured in units of the standard deviation. }

226:  \end{figure}

227: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

228: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

229: \begin{figure}

230: \epsfxsize=7truecm\epsfbox{D1.eps}

231: \epsfxsize=7truecm\epsfbox{D2.eps} \narrowtext \caption{ The drift

232: and diffusion coefficients $D^{(1)}(x)$ and $D^{(2)}(x)$,

233: estimated by Eq. (3). For the healthy subjects (triangles)

234: $D^{(1)}(x)$ and $D^{(2)}(x)$ follow linear and quadratic behavior

235: in $x$, while for patients with CHF (squares) they follow third-

236: and fourth-order behavior in $x$.}

237:  \end{figure}

238: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

239: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

240: \begin{figure}

241: \epsfxsize=7truecm\epsfbox{fig4.eps} \narrowtext \caption{ The

242: curves show, from top to bottom, the actual interbeat data (for a

243: healthy subject), the regenerated data using the corresponding

244: Langevin equation, and the regenerated data using the kernel

245: method. The time series are shifted in the vertical directions for

246: better presentation.}

247:  \end{figure}

248: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

249: \vskip 3cm

250: \bigskip

251: \noindent{\bf 3. Application to Fluctuations in Human Heartbeats}

252: 

253: \bigskip

254: We now apply the above method to reconstruction of the

255: fluctuations in the human heartbeats of both healthy and ill

256: subjects by taking $h\simeq t_M$. Recent studies [13-18] reveal

257: that under normal conditions, beat-to-beat fluctuations in the

258: heart rate might display extended correlations of the type

259: typically exhibited by dynamical systems far from equilibrium. It

260: has been shown [14], for example, that the various stages of sleep

261: may be characterized by extended correlations of heart rates

262: separated by a large number of beats. We show that the Markov time

263: scale $t_M$, and the drift and diffusion coefficients of the

264: interbeat fluctuations of healthy subjects and patients with

265: congestive heart failure (CHF) have completely different

266: behaviour, when analyzed by the method we propose in this paper,

267: hence helping one to distinguish the two groups of the subjects.

268: 

269: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

270: \begin{figure}

271: \epsfxsize=7truecm\epsfbox{fig5.eps} \narrowtext \caption{

272: Logarithmic plot of the second moment of the height-difference

273: versus $m$, for the actual data (circles) and the samples

274: regenerated by the kernel method (squares). The corresponding time

275: series are plotted in Fig. 4.}

276:  \end{figure}

277: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

278: We analyze both daytime (12:00 pm to 18:00 pm) and nighttime

279: (12:00 am to 6:00 am) heartbeat time series of healthy subjects,

280: and the daytime records of patients with CHF. Our data base

281: includes 10 healthy subjects (7 females and 3 males with ages

282: between 20 and 50, and an average age of 34.3 years), and 12

283: subjects with CHF, with 3 females and 9 males with ages between 22

284: and 71, and an average age of 60.8 years). Figure 1 presents the

285: typical data.

286: 

287: We first estimate the Markov time scale $t_M$ of the data for the

288: interbeat fluctuations. For the healthy subjects we find the

289: Markov time scale for the daytime data to be (all the values are

290: measured in units of the average time scale for the beat-to-beat

291: times of each subject), $t_M=3,3,3,1,2,3,3,2,3$ and 2. The

292: corresponding results for the nighttime records are, $t_M$ are

293: $3,3,1,3,3,2,3,3,2$ and $3$, respectively, comparable to those for

294: the daytime. On the other hand, for the daytime records of the

295: patients with CHF, the estimated Markov time scales are,

296: $t_M=151,258,760,542,231,257,864,8,366,393,385$, and 276.

297: Therefore, the healthy subjects have $t_M$ values that are much

298: smaller than those of the patients with CHF, hence providing an

299: unambiguous quantity for distinguishing the two.

300: 

301: We then check the validity of the CK equation for several $x_1$

302: triplets by comparing the directly-evaluated conditional

303: probability distributions $p(x_2,t_2|x_1,t_1)$ with the ones

304: calculated according to right side of Eq. (1). Here, $x$ is the

305: interbeat and for all the samples we define, $x\equiv (x-\bar

306: x)/\sigma$, where $\bar x$ and $\sigma$ are the mean and standard

307: deviations of the interbeats data. In Figure 2, the two

308: differently-computed PDFs are compared. Assuming the statistical

309: errors to be the square root of the number of events in each bin,

310: we find that the two PDFs are {\it statistically identical}.

311: 

312: The corresponding drift and diffusion coefficients $D^{(1)}(x)$

313: and $D^{(2)}(x)$ are displayed in Figure 3. We find that, in

314: addition to the Markov time scale $t_M$, the two coefficients

315: provide another important indicator for distinguishing the ill

316: from the healthy subjects: For the healthy subjects the drift

317: $D^{(1)}$ and the diffusion coefficient $D^{(2)}(x)$ are,

318: respectively, a linear and a quadratic function of $x$, whereas

319: the corresponding coefficients for patients with CHF follow a

320: third- and fourth-order equations in $x$. The analysis of the data

321: yields the following approximants for the healthy subjects,

322: \begin{eqnarray}

323: & & D^{(1)}(x)= -0.12 x\;, \cr \nonumber\\

324: & & D^{(2)}(x)=0.05 -0.042 x + 0.07 x^2\;,

325: \end{eqnarray}

326: whereas for the patients with CHF we find that,

327: \begin{eqnarray}

328:  D^{(1)}(x)&=&-0.0026 x - 0.0018 x^2 - 0.0007 x^3\;, \cr \nonumber \\

329: D^{(2)}(x)&=&0.0006 - 0.0007 x + 0.0005 x^2 \cr \nonumber \\

330: &+& 0.0003 x^3 + 0.0002x^4\;.

331: \end{eqnarray}

332: 

333: Equations (7) and (8) present the drift and diffusion coefficients

334: for a typical healthy subject and one with CHF. We note that the

335: final result for the Langevin equation is the same as the results

336: obtained in Ref. [18]. For other data measured for other patients

337: the functional dependence of $D^{(1)}$ and $D^{(2)}(x)$ would be

338: the same but with different numerical coefficients. The order of

339: magnitude of the coefficients is the same for all the healthy

340: subjects, and likewise for those with CHF (see also Ref. [19]).

341: Moreover, if we analyze different parts of the time series

342: separately, we find, (1) almost the same Markov time scale for

343: different parts of the time series, but with some differences in

344: the numerical values of the drift and diffusion coefficients, and

345: (2) that the drift and diffusion coefficients for different parts

346: of the time series have the same {\it functional forms}, but with

347: {\it different coefficients} in equations such as (7) and (8).

348: Hence, one can distinguish the data for sleeping times from those

349: for when the subjects are awake [20].

350: 

351: We also find another important difference between the heartbeat

352: dynamics of the two classes of subjects: Compared with the healthy

353: subjects, the drift and diffusion coefficients for the patients

354: with CHF are very small (reflecting, in some sense, the large

355: Markov time scale $t_M$). Hence, we suggest that one may use the

356: Markov time scales, the dependence of the drift and diffusion

357: coefficients on $x$, as well as their comparative magnitudes, for

358: characterizing the dynamics of human heartbeats and their

359: fluctuations, and to distinguish healthy subjects from those with

360: CHF. To our knowledge, this proposal is novel. Given its relative

361: simplicity, it would be most interesting to study whether this

362: proposal can be developed into a diagnostic tool for early

363: detection of congestive heart failure. Work in this direction is

364: in progress.

365: 

366: We compare in Figure 4 the original time series $x(n)$ with those

367: reconstructed by the Langevin equation [by, for example, using

368: Eqs. (4) and (7)] and the kernel method. While both methods

369: generate series that look similar to the original data, the kernel

370: method appears to better mimic the behavior of the original data.

371: To demonstrate the accuracy of Eq. (6), we compare in Figure 5 the

372: second moment of the stochastic function,

373: $C_2(m)=\langle[x(0)-x(m)]^2 \rangle$, for both the measured and

374: reconstructed data using the kernel method. The agreement between

375: the two is excellent. However, it is well-known that the agreement

376: between the second moments of a stochastic time series and its

377: reconstructed version is not sufficient for proving the accuracy

378: of the reconstruction method. Hence, we have also checked the

379: accuracy of the higher-order structure function,

380: $S_n=\langle|x(t_1)-x(t_2)|^n\rangle$ [21]. We find that the

381: agreement between $S_n$ for the original and reconstructed time

382: series for $n\leq 5$ is excellent, while the difference between

383: higher-order moments of the two times series, which are related to

384: the tails of the PDF of the $x-$increments, increases.

385: 

386: \bigskip

387: \noindent{\bf 4. Summary}

388: 

389: \bigskip

390: We have analyzed the interbeat fluctuations in the heart rates of

391: healthy subjects, as well as those with congestive heart failure,

392: by an inverse method for reconstruction of the stochastic process

393: that governs the fluctuations. The method, which is quite general

394: and can regenerate a stochastic process with high precision, is

395: based on utilizing measured data to estimate a drift and a

396: diffusion coefficients to be used in a Fokker-Planck, or an

397: equivalent Langevin, equation that describes the stochastic

398: process. The analysis of the times series for human heartbeat

399: dynamics using the new method, for both healthy subjects and those

400: with CHF, not only demonstrates the accuracy of the method, but

401: also potentially provides a novel technique for distinguishing the

402: heartbeat dynamics of the two classes of subjects.

403: 

404: We should point out that Stanley and co-workers [13,15-17,20,21]

405: analyze the type of data we considered in this paper by a method

406: different from what we present in the present paper. Their

407: analysis indicates that there may be long-range correlations in

408: the data, which might be characterized by self-affine fractal

409: distributions, such as the fractional Brownian motion or other

410: types of stochastic processes that give rise to such correlations.

411: They distinguish healthy subjects from those with CHF in terms of

412: the type of correlations that might exist in the data (negative as

413: opposed to positive correlations). The method proposed in the

414: present paper is different from that of Stanley and co-workers in

415: that, we analyze the data in terms of Markov processes. Although

416: our analysis does indicate the existence of correlations in the

417: data but, as is well-known in the theory of Markov processes, such

418: correlations, though extended, eventually decay. We distinguish

419: the healthy subjects from those with CHF in terms of the

420: differences between the drift and diffusion coefficients of the

421: Fokker-Plank equation which, in our view, provides a clearer and

422: more physical way of understanding the differences between the two

423: groups of the subjects. In addition, our method provides an

424: unambiguous way of reconstructing the data, hence providing a

425: means to predict the behavior of the data over periods of time

426: that are on the order of the Markov time scale. Although it

427: remains to be tested, we believe that our method is more sensitive

428: to small differences between the data for the two groups of the

429: subjects and, therefore, might eventually provide a diagnostic

430: tool for early detection of CHF in humans.

431: 

432: \bigskip

433: \noindent{\bf Acknowledgment}

434: 

435: \bigskip

436: We would like to thank Armin Bunde for useful comments on the

437: manuscript.

438: 

439: 

440: 

441: 

442: 

443: \begin{thebibliography}{99}

444: 

445: \bibitem{1} S. Torquato, {\it Random Heterogeneous Materials} (Springer,

446: New York, 2002); C.L.Y. Yeong and S. Torquato, Phys. Rev. E {\bf

447: 57}, 495 (1998); {\it ibid.} {\bf 58}, 224 (1998).

448: 

449: \bibitem{2} M. Sahimi, {\it Heterogeneous Materials}, Volume II

450: (Springer, New York, 2003).

451: 

452: \bibitem{3} R. Friedrich and J. Peinke, Phys. Rev. Lett. {\bf 78}, 863

453: (1997).

454: 

455: \bibitem{4} R. Friedrich, J. Peinke, and C. Renner, Phys. Rev. Lett.

456: {\bf 84}, 5224 (2000).

457: 

458: \bibitem{5} R. Friedrich, Th. Galla, A. Naert, J. Peinke and Th.

459: Schimmel, in {\it A Perspective Look at Nonlinear Media}, edited

460: by J. Parisi, S.C. Muller, and W. Zimmermann, (Springer, Berlin,

461: 1997), p. 313; R. Friedrich, {\it et al.}, Phys. Lett. A {\bf

462: 271}, 217 (2000).

463: 

464: \bibitem{6} M. Siefert, A. Kittel, R. Friedrich, and J. Peinke, Euro.

465: Phys. Lett. {\bf 61}, 466 (2003); S. Kriso, {\it et al.}, Phys.

466: Lett. A {\bf 299}, 287 (2002); S. Siegert, R. Friedrich, and J.

467: Peinke, Phys. Lett. A {\bf 243}, 275 (1998).

468: 

469: \bibitem{7} H. Risken, {\it The Fokker-Planck Equation} (Springer,

470: Berlin, 1984).

471: 

472: \bibitem{8} G. R. Jafari, S. M.

473: Fazlei, F. Ghasemi, S. M. Vaez Allaei, M. Reza Rahimi Tabar, A.

474: Iraji Zad, and G. Kavei, Phys. Rev. Lett. {\bf 91}, 226101 (2003).

475: 

476: \bibitem{9} C. Renner, J. Peinke, and R. Friedrich, J. Fluid Mech. {\bf

477: 433}, 383 (2001).

478: 

479: \bibitem{10} M. Ragwitz and H. Kantz, Phys. Rev. Lett. {\bf 87}, 254501

480: (2001).

481: 

482: \bibitem{11} R. Friedrich, C. Renner, M. Siefert, and J. Peinke, Phys.

483: Rev. Lett. {\bf 89}, 149401 (2002).

484: 

485: \bibitem{12} J. Davoudi and M. Reza Rahimi Tabar, Phys. Rev. Lett. {\bf

486: 82}, 1680 (1999).

487: 

488: \bibitem{13} C.-K. Peng, J. Mietus, J. M. Hausdorff, S. Havlin, H. E.

489: Stanley, and A. L. Goldberger, Phys. Rev. Lett. {\bf 70}, 1343

490: (1993).

491: 

492: \bibitem{14} A. Bunde, S. Havlin, J. W. Kantelhardt, T. Penzel, J.-H.

493: Peter, and K. Voigt, Phys. Rev. Lett. {\bf 85}, 3736 (2000).

494: 

495: \bibitem{15} P. Bernaola-Galvan, P. Ch. Ivanov, L. N. Amaral, and H. E.

496: Stanley, Phys. Rev. Lett. {\bf 87}, 168105 (2001).

497: 

498: \bibitem{16} V. Schulte-Frohlinde, Y. Ashkenanzy, P. Ch. Ivanov, L.

499: Glass, A. L. Goldberger, and H. E. Stanley, Phys. Rev. Lett. {\bf

500: 87}, 068104 (2001).

501: 

502: \bibitem{17} Y. Ashkenazy, P. Ch. Ivanov, Shlomo Havlin, C-K. Peng, A. L.

503: Goldberger, and H. E. Stanley, Phys. Rev. Lett. {\bf 86}, 1900

504: (2001).

505: 

506: \bibitem{18} T. Kuusela, Phys. Rev. E {\bf 69}, 031916 (2004).

507: 

508: \bibitem{19} M. M. Wolf, G. A. Varigos, D. Hunt, and J. G. Sloman, Med.

509: J. Aust {\bf 2}, 52 (1978).

510: 

511: \bibitem{20} P. Ch. Ivanov, A. Bunde, L. A. N. Amaral, S. Havlin, J.

512: Fritsch-Yelle, R. M. Baevsky, H. E. Stanley, and A. L. Goldberger,

513: Europhys. Lett. {\bf 48}, 594 (1999).

514: 

515: \bibitem{21} P. Ch. Ivanov, L. A. N. Amaral, A. L. Goldberger, S.

516: Havlin, M. G. Rosenblum, Z. Struzik, and H. E. Stanley, Nature

517: (London) {\bf 399}, 461 (1999).

518: 

519: 

520: \end{thebibliography}

521: 

522: \end{document}

523: