1: \documentstyle[aps,pre,rotate,epsf]{revtex}

2: \begin{document}

3: \draft

4: \title{Comment on ``Analysis of chaotic motion and its shape dependence in a

5: generalized piecewise linear map''}

6: \author{R. Klages\cite{em}}

7: \address{Max Planck Institute for Physics of Complex Systems,

8: N\"othnitzer Str. 38,  D-01187 Dresden, Germany}

9: \date{\today}

10: \maketitle

11: \begin{abstract}

12: Rajagopalan and Sabir [Phys.\ Rev.\ E {\bf 63}, 057201 (2001)] recently

13: discussed deterministic diffusion in a piecewise linear map using an approach

14: developed by Fujisaka et al. We first show that they rederived the random walk

15: formula for the diffusion coefficient, which is known to be the exact result

16: for maps of Bernoulli type since the work of Fujisaka and Grossmann [Z. Physik

17: B {\bf 48}, 261 (1982)]. However, this correct solution is at variance to the

18: diffusion coefficient curve presented in their paper. Referring to another

19: existing approach based on Markov partitions, we answer the question posed by

20: the authors regarding solutions for more general parameter values by recalling

21: the finding of a fractal diffusion coefficient. We finally argue that their

22: model is not suitable for studying intermittent behavior, in contrast to what

23: was suggested in their paper.

24: \end{abstract}

25: \pacs{PACS numbers: 05.45.-a, 05.45.Ac, 05.60.-k, 05.40.-a}

26: The study of deterministic diffusion in simple chaotic maps on the line

27: appears to have originated about twenty years ago (see, e.g., Ref.\

28: \cite{RKdiss} and further references therein). Already in the seminal work by

29: Fujisaka and Grossmann \cite{GF2}, a variety of piecewise linear models was

30: defined and analyzed by means of stochastic modeling. All these maps are of

31: the form $x_{n+1}=M_h(x_n)$, where $h\in N$ is a control parameter, and $x_n$

32: is the position of a point particle at discrete time $n$. $M_h(x)$ is

33: continued periodically beyond the interval $[0,1)$ onto the real line by a

34: lift of degree one, $M_h(x+1)=M_h(x)+1$. The map defined in Ref.\

35: \cite{RaSa01}, which is sketched again in Fig.\ \ref{fig1}, provides a

36: straightforward generalization of the one introduced in Ref.\ \cite{GrTh83},

37: which is recovered at $h=1/2$. For this type of maps, indeed a vast literature

38: exists on how to obtain exact analytical results at specific cases of

39: parameter values; Refs.\ \cite{RKdiss,ddlit} sumarize some of these methods,

40: with more complete lists of references therein. It is furthermore well-known

41: that the calculations are particularly simple if the parameter is such that

42: the map exhibits the Bernoulli property \cite{Schu}.

43: 

44: We first wish to present a considerable shortcut to the diffusion coefficient

45: calculations published in Ref.\ \cite{RaSa01}. Based on a theory which appears

46: to be a precursor of what was called ``Fujisaka's characteristic function

47: method'' in Ref. \cite{RaSa01}, Fujisaka and Grossmann have shown \cite{GF2}

48: that the diffusion coefficient formula

49: \begin{equation}

50: D=\frac{<j^2(x_n)>}{2} \label{eq:drw}

51: \end{equation}

52: provides the exact solution for types of maps as the one studied in Ref.\

53: \cite{RaSa01}, i.e., if they share the Bernoulli property. Here $j(x_n)$ is

54: the jump velocity defined as $j(x_n):=[x_{n+1}]-[x_n]$ with $[x]$ being the

55: largest integer less than $x$, and $<\ldots>$ denotes the average over the

56: invariant probability density. This expression is just identical to the

57: familiar random walk formula for diffusion on a one-dimensional lattice, where

58: the length of jumps squared is weighted with the probability to perform such

59: jumps (see Ref.\ \cite{dcrc} and further references therein). Evaluating this

60: equation for the map under consideration leads to

61: \begin{equation}

62: D(h,r)=\sum_{j=1}^hj^2\delta(j,r) \label{eq:dk1}

63: \end{equation}

64: with $h\in N , 0<r<1$, where $2\delta(j,r)$ denotes the probability

65: to jump over a distance of $j$ steps and is easily calculated to

66: \begin{equation}

67: \delta(j,r)=\frac{2}{m_j} \;, \label{eq:dk2}

68: \end{equation}

69: $m_j$ being the slopes of the map. Combining the above two equations yields

70: Eq.\ (17) of Ref.\ \cite{RaSa01}. We conclude that Rajagopalan and Sabir have

71: confirmed again Eq.\ (\ref{eq:drw}) of Fujisaka and Grossmann as applied to

72: their specific map. We now focus on the author's special case of the map

73: defined by the relation for the slopes

74: \begin{equation}

75: m_0=3+\frac{4(1-r^h)}{r^h(1-r)}

76: \end{equation}

77: with $m_i/m_{i-1}=r$. Solutions for Eqs.\ (\ref{eq:dk1}), (\ref{eq:dk2}) under

78: this constraint are shown in Fig.\ \ref{fig2} for different $h$. This figure

79: corrects the erroneous result shown in Fig.\ 3 of Ref.\ \cite{RaSa01}, which

80: only includes a few data points and appears to indicate a rather irregular

81: curve for the diffusion coefficient at $h=2$. Below we will explain why all

82: the curves shown in Fig.\ \ref{fig2} must indeed be simple functions of $r$.

83: 

84: However, first we would like to recall a second method which is not restricted

85: to special cases of parameters such as integer heights, in contrast to the one

86: outlined in Refs.\ \cite{GF2,RaSa01}. The basic idea of this method is to

87: directly solve the Frobenius-Perron equation of the dynamical system,

88: \begin{equation}

89: \rho_{n+1}(x) = \int dy \; \rho_n(y) \; \delta(x-M_h(y)) \; ,

90: \end{equation}

91: where $\rho_n(x)$ is the probability density for points on the real

92: line. There exists a dense set of parameter values $h$ for which one can

93: construct Markov partitions of the map, and for each of these parameter values

94: this equation can be written as a matrix equation

95: \cite{RKdiss,RKD},

96: \begin{equation}

97: \mbox{\boldmath $\rho$}_{n+1}=\, T(h,r) \, \mbox{\boldmath

98: $\rho$}_n\;.

99: \end{equation}

100: $\mbox{\boldmath $\rho$}_n$ represents a column vector of the probability

101: densities defined on each part of the Markov partition at time $n$, and

102: $T(h,r)$ is a topological transition matrix which can be constructed from the

103: Markov partition. This setup provides two ways of solution: one way is to

104: solve the eigenvalue problem of $T(h,r)$ and to relate the diffusion

105: coefficient to its eigenvalues. As is shown in detail in Refs.\ \cite{RKD},

106: in special cases all calculations can be performed analytically. For the

107: simple map defined in Ref.\ \cite{RaSa01} these calculations are

108: straightforward and confirm again Eqs.\ (\ref{eq:dk1}), (\ref{eq:dk2}).  For

109: more general cases, the matrix equation can simply be iterated

110: \cite{RKdiss,dcrc} yielding numerically exact solutions for the probability

111: density vector $\mbox{\boldmath $\rho$}_n$ at any time step $n$, as well as

112: for any other dynamical quantity based on probability density averages. Both

113: such methods were previously applied to various examples of piecewise linear

114: maps \cite{RKdiss,dcrc,RKD}. Fig.\ \ref{fig1} presents analogous

115: results for the map studied in Ref.\ \cite{RaSa01} at $h=2$ and $r=0.5$, cp.\

116: to Fig.\ 3.1 on p.54 of Ref.\ \cite{RKdiss}. The probability density is a

117: Gaussian on a coarse scale, whereas the fine scale is determined by the

118: invariant density of the map on the unit interval. These deviations from an

119: exact Gaussian can quantitatively be evaluated, e.g., by calculating the

120: curtosis of the respective map density; for a more detailed discussion of such

121: aspects we refer to Chapter 3 of Ref.\ \cite{RKdiss}. This interplay between

122: fine and coarse structure of the probability densities was furthermore

123: discussed in terms of the spectrum of eigenmodes of the Frobenius-Perron

124: operator, see \cite{RKdiss,RKD}.  These known results appear to be

125: recovered in Ref.\ \cite{RaSa01} by a respective analysis of the fluctuation

126: spectrum, which provides an alternative way to look at the probability density

127: of the map. 

128: 

129: In their outlook to further work, the authors of Ref.\ \cite{RaSa01} raised

130: the question of how to compute the diffusion coefficient for maps with

131: fractional heights $h$, and how it may look like. The application of the

132: arsenal of methods outlined above has already given a full answer to this

133: problem. As a central result, it was found that the diffusion coefficient for

134: these maps is a fractal function of the parameter $h$. To present an example,

135: Fig.\ \ref{fig3} depicts the result for a mirrored zigzag-map with uniform

136: slope, which has some similarities with the one studied in Ref.\

137: \cite{RaSa01}; for details see Refs.\ \cite{RKdiss,dcrc}. Knowing these

138: results, it is straightforward to conclude that, for arbitrary height $h$, the

139: map studied in Ref.\ \cite{RaSa01} will just yield another fractal diffusion

140: coefficient; further evidence for that statement is provided by the numerical

141: and analytical data presented in Ref.\ \cite{Tse94}.

142: 

143: So why can the diffusion coefficient of the map in Ref.\ \cite{RaSa01} not be

144: fractal as a function of $r$ at integer values of the height? One way to look

145: at this problem is to inquire how the topology of the map is affected by

146: parameter variation. A fundamental tool providing detailed information about

147: the topology of a dynamical system are Markov partitions. Varying $r$ at

148: integer heights does not change the Markov partition, thus the topology of the

149: map does not change, and any quantity resulting from an average over the

150: invariant density is a simple function of the parameter

151: \cite{StSt97}. However, changing the height changes the Markov partition in a

152: complicated way and reflects the topological instability of the map under this

153: type of parameter variation. This topological instability results in fractal

154: transport coefficients.

155: 

156: Finally, we comment on the conclusion of Rajagopalan and Sabir that the map

157: studied in their paper is ``suited in describing diffusion systems showing

158: intermittency''. In this aspect the authors appear to follow Ref.\

159: \cite{GrTh83}, where the map shown in Fig.\ \ref{fig1} at $h=1/2$ was

160: introduced for the purpose of modeling ``strong correlations between

161: successive steps... as realized in Brownian motion with directional

162: persistence''. Indeed, Grossmann and Thomae revealed a persistent dynamics

163: which they characterized as ``intermittent-{\em like}'' behavior. They linked

164: these correlations to deviations from a pure Gaussian probability density such

165: as the ones discussed above.

166: 

167: In the following we use the term ``intermittency'' in the sense of Pomeau and

168: Manneville (see, e.g., Ref.\ \cite{Schu} for a tutorial about their

169: results). Particularly, we wish to distinguish it from the denotation

170: ``intermittency-like'' in the sense of Grossmann and Thomae. Extensive studies

171: of diffusion in one-dimensional intermittent maps led to the conclusion that,

172: generally, in this case a diffusion coefficient does not exist

173: \cite{anom}. Furthermore, all maps studied in these references are inherently

174: nonlinear. For piecewise linear expanding maps which are uniquely ergodic if

175: restricted to compact spaces, such as the one of Refs.\ \cite{RaSa01,GrTh83},

176: there is no evidence for intermittency nor for anomalous diffusion

177: \cite{RKdiss}. Applying the concept of conjugacy enables to transform

178: piecewise linear maps onto nonlinear ones. However, the diffusive dynamics is

179: invariant under conjugacy \cite{GrTh83}, thus the corresponding nonlinear map

180: is again non-intermittent and normal diffusive. To our knowledge the only

181: piecewise linear map exhibiting intermittency was introduced in Ref.\

182: \cite{GaWa88}, and it belongs to a very different class than the one of Refs.\

183: \cite{RaSa01,GrTh83}.

184: 

185: In summary, by relating the piecewise linear map studied in Ref.\

186: \cite{RaSa01} to intermittent behavior the authors confuse the meaning of

187: intermittency, in the sense of Pomeau and Manneville, with the existence of

188: intermittent-{\em like} behavior, in the sense of persistence in the diffusive

189: motion. Intermittency generally leads to anomalous diffusion, whereas

190: persistence in piecewise linear maps shows up in form of local extrema of the

191: fractal diffusion coefficient at integer and half-integer heights, see Fig.\

192: \ref{fig3}. We conclude that the analysis of chaotic motion and its shape

193: dependence as performed in Ref.\ \cite{RaSa01} has nothing to do with

194: intermittency, but instead recovers features of the parameter-dependent normal

195: diffusion coefficient as studied in Refs.\

196: \cite{RKdiss,GF2,GrTh83,ddlit,Schu,dcrc,RKD,Tse94,StSt97}.

197: 

198: The author thanks N.Korabel and J.R.Dorfman for helpful remarks.

199: 

200:   

201: \vspace*{-0.5cm}  

202: 

203: \begin{thebibliography}{10}

204:   

205: \vspace*{-1.5cm}

206: 

207: \bibitem[*]{em}

208: Electronic address: rklages@mpipks-dresden.mpg.de.

209: 

210: \bibitem{RKdiss}

211: R. Klages, {\em Deterministic diffusion in one-dimensional chaotic dynamical

212:   systems} (Wissenschaft \& Technik-Verlag, Berlin, 1996).

213: 

214: \bibitem{GF2}

215: H. Fujisaka and S. Grossmann, Z. Physik B {\bf 48},  261  (1982).

216: 

217: \bibitem{RaSa01}

218: S. Rajagopalan and M. Sabir, Phys. Rev. E {\bf 63},  057201  (2001).

219: 

220: \bibitem{GrTh83}

221: S. Grossmann and S. Thomae, Phys. Lett. A {\bf 97}, 263 (1983).

222: 

223: \bibitem{ddlit}

224: P. Gaspard, {\em Chaos, Scattering, and Statistical Mechanics} (Cambridge

225: University Press, Cambridge, 1998); J.R. Dorfman, {\em An Introduction to

226: Chaos in Nonequilibrium Statistical Mechanics} (Cambridge University Press,

227: Cambridge, 1999); R. Artuso, {\em Classical and quantum zeta-functions and

228: periodic orbit theory}; in: G. Casati, I. Guarneri, and U. Smilansky, Editors,

229: {\em New Directions in Quantum Chaos}, (IOS Press, Amsterdam, 2000);

230: P. Cvitanovi{\'c} et al., {\em Classical and Quantum Chaos} (Niels Bohr

231: Institute, Copenhagen, 2001).

232: 

233: \bibitem{Schu}

234: H.G. Schuster, {\em Deterministic Chaos}, 2nd ed. (VCH Verlagsgesellschaft

235: mbH, Weinheim, 1989).

236: 

237: \bibitem{dcrc}

238: R. Klages and J.R. Dorfman, Phys. Rev. E {\bf 55},  R1247  (1997).

239: 

240: \bibitem{RKD}

241: R. Klages and J.R. Dorfman, Phys. Rev. Lett. {\bf 74}, 387 (1995);

242: Phys. Rev. E {\bf 59}, 5361 (1999); P. Gaspard and R. Klages, Chaos {\bf 8},

243: 409 (1998).

244: 

245: \bibitem{Tse94}

246: H.-C. Tseng {\it et~al.}, Phys. Lett. A {\bf 195},  74  (1994).

247: 

248: \bibitem{StSt97}

249: R. Stoop and W.-H. Steeb, Phys. Rev. E {\bf 55},  7763  (1997).

250: 

251: \bibitem{anom}

252: T. Geisel and S. Thomae, Phys. Rev. Lett. {\bf 52}, 1936 (1984); T. Geisel, J.

253: Nierwetberg, and A. Zacherl, Phys. Rev. Lett. {\bf 54}, 616 (1985); R. Artuso,

254: G. Casati, and R. Lombardi, Phys. Rev. Lett. {\bf 71}, 62 (1993); X.-J. Wang

255: and C.-K. Hu, Phys. Rev. E {\bf 48}, 728 (1993).

256: 

257: \bibitem{GaWa88}

258: P. Gaspard and X.-J. Wang, Proc. Nat. Acad. Sci. USA {\bf 85}, 4591 (1988).

259: 

260: \end{thebibliography}

261:   

262: \begin{figure}[b]

263: \epsfxsize=7cm

264: \centerline{\rotate[r]{\epsfbox{zzfig1.ps}}}

265: %\vspace*{0.3cm} 

266: \caption{Time-dependent probability density $\rho_n(x)$ for the map sketched

267: in the figure as it evolves starting from a uniform density in a box situated

268: around $x=0$. The results have been obtained from iterating transition

269: matrices as explained in the text. Included are Gaussian solutions from the

270: ordinary diffusion equation corresponding to the exact diffusion coefficient

271: $D(h,r)$ of the map, where $h=2, r=0.5$. These dashed lines are almost

272: indistinguishable from the map densities, however, they are lacking the

273: step-like fine structure. From above to below, the time steps are $n=20$,

274: $n=50$, $n=200$. The quantities plotted in this and in the following figures

275: are dimensionless.}

276: \label{fig1}

277: \end{figure}

278: 

279: 

280: \begin{figure}[b]

281: \epsfxsize=10cm

282: \centerline{\rotate[r]{\epsfbox{zzfig2.ps}}}

283: %\vspace*{0.3cm} 

284: \caption{The diffusion coefficient $D(h,r)$ for the map shown in Fig.\

285: \ref{fig1} according to Eqs.\ (\ref{eq:dk1}), (\ref{eq:dk2}). Solutions are shown for the

286: values of the height $h=2,3,4,5$ starting from below. The case $h=2$ corrects

287: the erroneous result in Fig.\ 3 of Ref.\ [3].}

288: \label{fig2}

289: \end{figure}

290: 

291: \begin{figure}[b]

292: \epsfxsize=10cm

293: \centerline{\rotate[r]{\epsfbox{zzfig3.ps}}}

294: %\vspace*{0.3cm} 

295: \caption{Fractal diffusion coefficient $D(h)$ for the mirrored zigzag-map

296: sketched in the figure as a function of the height $h$. Shown are 13376 data

297: points. The data is from Refs.\ [1,5].}

298: \label{fig3}

299: \end{figure}

300: 

301: \end{document}

302: 

303: 

304: 

305: