1: %% ****** Start of file template.aps ****** %
2: %%
3: %%
4: %% This file is part of the APS files in the REVTeX 4 distribution.
5: %% Version 4.0 of REVTeX, August 2001
6: %%
7: %%
8: %% Copyright (c) 2001 The American Physical Society.
9: %%
10: %% See the REVTeX 4 README file for restrictions and more information.
11: %%
12: %
13: % This is a template for producing manuscripts for use with REVTEX 4.0
14: % Copy this file to another name and then work on that file.
15: % That way, you always have this original template file to use.
16: %
17: % Group addresses by affiliation; use superscriptaddress for long
18: % author lists, or if there are many overlapping affiliations.
19: % For Phys. Rev. appearance, change preprint to twocolumn.
20: % Choose pra, prb, prc, prd, pre, prl, prstab, or rmp for journal
21: % Add 'draft' option to mark overfull boxes with black boxes
22: % Add 'showpacs' option to make PACS codes appear
23: % Add 'showkeys' option to make keywords appear
24: \documentclass[aps,pre,twocolumn,showpacs,groupedaddress]{revtex4}
25: %\documentclass[aps,pre,preprint,showpacs,superscriptaddress]{revtex4}
26: %\documentclass[aps,prl,twocolumn,groupedaddress]{revtex4}
27: %\documentclass[aps,prl,preprint,groupedaddress,draft,showpacs]{revtex4}
28:
29: \usepackage{graphicx}% Include figure files
30:
31: % You should use BibTeX and apsrev.bst for references
32: % Choosing a journal automatically selects the correct APS
33: % BibTeX style file (bst file), so only uncomment the line
34: % below if necessary.
35: %\bibliographystyle{apsrev}
36:
37: %\oddsidemargin=-0.8cm \evensidemargin=0cm
38: %\topmargin=-0.5cm \textheight=23.7cm
39:
40: %%%%% Personal Macros %%%%%%%%%%%%%%%%%%%
41:
42: \newcommand {\dip}{\displaystyle}
43: \newcommand {\s}{\dip \sqrt}
44: \newcommand {\f}{\dip \frac}
45: \newcommand {\integ}{\displaystyle \int}
46: \newcommand {\sq}{\sqrt}
47: \newcommand {\ig}{\includegraphics}
48:
49: \newcommand {\lan}{\langle}
50: \newcommand {\ran}{\rangle}
51: \newcommand {\lp}{\mbox {\boldmath $($}}
52: \newcommand {\rp}{\mbox {\boldmath $)$}}
53:
54: \begin{document}
55:
56: % Use the \preprint command to place your local institutional report
57: % number in the upper righthand corner of the title page in preprint mode.
58: % Multiple \preprint commands are allowed.
59: % Use the 'preprintnumbers' class option to override journal defaults
60: % to display numbers if necessary
61: %\preprint{}
62:
63: %Title of paper
64: \title{Spectral analysis and an area-preserving extension \\
65: of a piecewise linear intermittent map}
66:
67: % repeat the \author .. \affiliation etc. as needed
68: % \email, \thanks, \homepage, \altaffiliation all apply to the current
69: % author. Explanatory text should go in the []'s, actual e-mail
70: % address or url should go in the {}'s for \email and \homepage.
71: % Please use the appropriate macro foreach each type of information
72:
73: % \affiliation command applies to all authors since the last
74: % \affiliation command. The \affiliation command should follow the
75: % other information
76: % \affiliation can be followed by \email, \homepage, \thanks as well.
77: \author{Tomoshige Miyaguchi}
78: \email{tomo@nse.es.hokudai.ac.jp}
79: \affiliation{%
80: Meme Media Laboratory,
81: Hokkaido University,
82: Kita-Ku, Sapporo 060-0813, Japan
83: }
84: \author{Yoji Aizawa}
85: %\email{aizawa@waseda.jp}
86: \affiliation{%
87: Department of Applied Physics, School of Science and
88: Engineering, Waseda University, 3-4-1 Okubo, Shinjuku-ku, Tokyo,
89: 169-8555, Japan
90: }
91:
92: %
93: %\homepage[]{Your web page}
94: %\thanks{}
95: %\altaffiliation{}
96:
97: %Collaboration name if desired (requires use of superscriptaddress
98: %option in \documentclass). \noaffiliation is required (may also be
99: %used with the \author command).
100: %\collaboration can be followed by \email, \homepage, \thanks as well.
101: %\collaboration{}
102: %\noaffiliation
103:
104: \date{\today}
105:
106: \begin{abstract}
107: We investigate spectral properties of a 1-dimensional piecewise linear
108: intermittent map, which has not only a marginal fixed point but also a
109: singular structure suppressing injections of the orbits into
110: neighborhoods of the marginal fixed point. We explicitly derive
111: generalized eigenvalues and eigenfunctions of the Frobenius--Perron
112: operator of the map for classes of observables and piecewise constant
113: initial densities, and it is found that the Frobenius--Perron operator
114: has two simple real eigenvalues $1$ and $\lambda_d \in (-1,0)$, and a
115: continuous spectrum on the real line $[0,1]$. From these spectral
116: properties, we also found that this system exhibits power law decay of
117: correlations. This analytical result is found to be in a good agreement
118: with numerical simulations. Moreover, the system can be
119: extended to an area-preserving invertible map defined on the unit
120: square. This extended system is similar to the baker transformation,
121: but does not satisfy hyperbolicity. A relation between this
122: area-preserving map and a billiard system is also discussed.
123: % This extension is possible because the uniform density is invariant
124: % under the time evolution of the 1-dimensional map.
125: \end{abstract}
126:
127: % insert suggested PACS numbers in braces on next line
128: \pacs{05.45.-a }
129: % insert suggested keywords - APS authors don't need to do this
130: %\keywords{}
131:
132: %\maketitle must follow title, authors, abstract, \pacs, and \keywords
133: \maketitle
134:
135: % body of paper here - Use proper section commands
136: % References should be done using the \cite, \ref, and \label commands
137:
138: \section{Introduction}
139:
140: In the past decades, a lot of studies have been devoted to
141: investigations of the relations between microscopic chaos and
142: non-equilibrium behaviors such as relaxation and transport, and it has
143: been found that microscopic chaos plays essential roles in
144: non-equilibrium processes \cite{gaspard2,dorfman}.
145: For example, it is well known that, for fully chaotic (hyperbolic)
146: systems, correlation functions decay exponentially and their decay rates
147: are characterized by the discrete eigenvalues of its Frobenius--Perron
148: (FP) operator \cite{ruelle1,ruelle2,ruelle3,pollicott2,pollicott4}.
149: %
150: As one of the examples of the hyperbolic systems that permits detailed
151: calculations, the baker transformation has been studied extensively and
152: its spectral properties of FP operator are fully understood
153: \cite{hase1}. In addition, the baker map is considered as an
154: abstract model of chaotic Hamiltonian systems, because it has the
155: area-preserving property, which is an universal feature of the
156: Poincar\'e map of the Hamiltonian systems of 2 degrees of freedom
157: \cite{arnold1}. In fact, similarities of the baker map to the Lorentz
158: gas with finite horizon have been pointed out \cite{tel}.
159:
160: In contrast to hyperbolic systems, dynamics in generic Hamiltonian
161: systems is more complicated and diverse. When a phase space of a
162: Hamiltonian system consists of integrable (torus) and non-integrable
163: components (chaos), power law decay of correlations is frequently
164: observed \cite{karney,chirikov,meiss1,meiss2,geisel3,zaslavsky}.
165: Although such kind of systems, i.e., systems with mixed type phase
166: spaces, are more generic than the integrable or the fully chaotic
167: systems, the theoretical understanding of their statistical properties
168: is not enough;~for example the ergodic and mixing properties of chaotic
169: components of generic systems are still unclear from the theoretical
170: point of view.
171: % and there is a couple of candidates for origins of power
172: %law decay of correlations\cite{zaslavsky}.
173:
174: For understanding sub-exponential decay of correlation functions in
175: dynamical systems, non-hyperbolic 1-dimensional maps have been studied by
176: several authors
177: \cite{geisel1,geisel2,artuso1,artuso2,hase3,tasaki1,tasaki2,schuster,aizawa,aaronson}
178: and they have found power law decay of correlations in their models.
179: Therefore, it is natural to imagine connections of these non-hyperbolic
180: maps and mixed type Hamiltonian systems, however, extensions of these
181: maps to 2-dimensional area-preserving systems are unknown.
182: Thus in this paper, we introduce a modified version of the 1-dimensional
183: intermittent map studied in Ref.~\cite{tasaki1} and extend
184: it to an area-preserving system. This area-preserving map, which is
185: similar to the baker transformation, may be considered as an abstract
186: model of mixed type Hamiltonian systems.
187:
188: Our theoretical treatment is mainly based on Ref.~\cite{tasaki1}, where
189: a piecewise linear version of the Pomeau-Manneville map
190: \cite{pomeau1,pomeau2} is proposed and its generalized spectral
191: properties of the FP operator in a sense of
192: Refs.~{\cite{gelfand1,gelfand2}} have been elucidated. Their model has
193: a marginal fixed point and exhibits power law decay of correlation which
194: they have found to be an outcome of a continuous spectrum of the FP
195: operator. In addition to a marginal fixed point, the piecewise linear
196: map studied in the present paper has a singular structure, which
197: suppresses injections of the orbits into neighborhoods of the marginal
198: fixed point. Due to this property, the uniform density is invariant
199: under time evolution and the map can be extended to an area-preserving
200: map on the unit square. And it is shown that generalized eigenvalues of
201: the FP operator consists of two simple real eigenvalues $1$ and
202: $\lambda_d \in (-1,0)$, and a continuous spectrum on the real interval
203: $[0,1]$. It is also shown that correlation functions exhibit power law
204: decay due to the continuous spectrum.
205:
206: This paper is organized as follows. In Sec.~\ref{sec:defs}, we introduce
207: the piecewise linear map, and define the FP operator and functional
208: spaces of observables and initial densities. In Sec.~\ref{sec:sp}, we
209: derive the spectral decomposition of the FP operator. In
210: Sec.~\ref{sec:extn}, long time behaviors are analyzed and some numerical
211: results are displayed. We also discuss the extension of our model to an
212: area-preserving invertible map in Sec.~\ref{sec:extn}.
213: Section~\ref{sec:summary} is devoted to summary and remarks that include
214: comments about differences between our model and the one in
215: Ref.~\cite{tasaki1}, and about similarities to a billiard system.
216:
217: \begin{figure}
218: \vspace*{.cm}
219: \ig[width=8.6cm]{EPS/cPMmap.eps} \vspace*{-.2cm}
220: \caption{The piecewise linear map $\phi (x)$ (solid line) for $b=0.5$
221: and $\beta =1.3$. The circles indicate endpoints of the straight
222: line segments.}
223: \label{fig:cPMmap}
224: \end{figure}
225:
226: \section{\label{sec:defs}Piecewise linear map}
227: \subsection{Definition of the map}
228: The dynamical system we consider in this paper is the piecewise linear
229: map shown in Fig.~\ref{fig:cPMmap}. This map $\phi (x):[0,1] \to [0,1]$
230: consists of two parts:~the left $0 \leq x<b$ and the right part
231: $b \leq x<1$. Each part is formed by an infinite number of straight
232: line segments;~the circles plotted in Fig.~\ref{fig:cPMmap}
233: indicate endpoints of these segments. The map $\phi(x)$ is defined by
234: \begin{eqnarray}
235: \phi(x) =
236: \left\{
237: \begin{array}{l}
238: \eta_{k}^{-} (x - \xi_{k}^{-}) + \xi_{k-1}^{-} \\[.3cm]
239: ~~~~~{\rm for~} x \in [\xi_{k}^{-},\xi_{k-1}^{-}) ~~~(k=1,2,\cdots), \\[.5cm]
240: \eta_{k}^{+} (x - \xi_{k}^{+}) + \xi_{k-1}^{-} \\[.3cm]
241: ~~~~~{\rm for~} x \in [\xi_{k}^{+},\xi_{k-1}^{+}) ~~~(k=1,2,\cdots).
242: \end{array}
243: \right.
244: \label{eqn:1d-model}
245: \end{eqnarray}
246: In this definition, $\xi_k^-$~$(k=0,1,\cdots)$ represent the horizontal
247: coordinates of the endpoints of the segments on the left part ($x<b$)
248: and they are defined as
249: \begin{eqnarray}
250: \left\{
251: \begin{array}{l}
252: \xi_0^- = b, \\[.25cm]%~~~~~\xi_{-1}^- = 1,
253: \xi_{k-1}^{-} - \xi_k^- = \f b{\zeta (\beta)} \left( \f 1k \right)^{\beta}
254: ~~~{\rm for}~~k=1,2,\cdots,
255: \end{array}
256: \right.
257: \end{eqnarray}
258: where $\beta > 1$ is a parameter and
259: $\zeta (\beta) = \sum_{n=1}^{\infty} 1/n^{\beta}$ is the Riemann zeta
260: function. And $\eta_k^-$ is the slope of the
261: $k$-th segment and defined as
262: \begin{eqnarray}
263: \eta_k^- &=&
264: \f {\xi_{k-2}^{-} - \xi_{k-1}^{-}}{\xi_{k-1}^{-} - \xi_{k}^{-}}
265: \\[.25cm]
266: &=&
267: \left\{
268: \begin{array}{ll}
269: ~~\f {1-b}b \zeta (\beta) &~~~{\rm for}~~k=1, \\[.35cm]
270: \left( \f k{k-1} \right)^{\beta} &~~~{\rm for}~~k=2,3,\cdots,
271: \end{array}
272: \right.
273: \end{eqnarray}
274: where we define $\xi_{-1}^- = 1$ for convenience.
275: This definition of the left part ($0<x<b$) is the same as that of the
276: piecewise linear Pomeau-Manneville map proposed in the
277: Ref.~\cite{tasaki1}. The map $\phi(x)$ can be approximated as
278: $\phi (x) \sim x + C x^{\beta/(\beta - 1)}$ when $x \to 0+$,
279: where $C$ is a constant. Thus the origin $x=0$ is a marginal fixed
280: point.
281:
282: In the same way, $\xi_k^+$~$(k=0,1,\cdots)$ are the horizontal
283: coordinates of the endpoints of the segments on the right part
284: and defined as
285: \begin{eqnarray}
286: \left\{
287: \begin{array}{l}
288: \xi_1^+ = b\left( 1 + \f 1{\zeta (\beta)}\right),
289: ~~~~\xi_{0}^+ = 1, \\[.5cm]
290: \xi_{k-1}^{+} - \xi_k^+ = \f b{\zeta (\beta)}
291: \left( \f 1{(k-1)^{\beta}} - \f 1{k^{\beta}} \right) \\[.35cm]
292: ~~~{\rm for}~~k=2,3,\cdots,
293: \end{array}
294: \right.
295: \end{eqnarray}
296: and $\eta_k^+$ is the slope of the $k$-th segment and defined as
297: \begin{eqnarray}
298: \eta_k^+ &=&
299: \f {\xi_{k-2}^{-} - \xi_{k-1}^{-}}{\xi_{k-1}^{+} - \xi_{k}^{+}} \\[.25cm]
300: &=&
301: \left\{
302: \begin{array}{lll}
303: \f {1-b}{1-b \left[1 + 1/{\zeta (\beta)}\right]}
304: &~~~{\rm for}~~k=1, \\[.35cm]
305: \f {k^{\beta}}{k^{\beta} - (k-1)^{\beta}} &~~~{\rm for}~~k=2,3,\cdots.
306: \end{array}
307: \right.
308: \end{eqnarray}
309: This is the definition of the right part of the map $\phi (x)$,
310: $b \leq x < 1$. This part is different from the Pomeau-Manneville type maps,
311: and is similar to a map proposed by Artuso and Cristadoro \cite{artuso2}.
312: $\phi(x)$ behaves as $\phi (x) \sim (x-b)^{(\beta-1)/{\beta}}$ when
313: $x \to b+$. Therefore the derivative $\phi'(x)$ of the map is divergent
314: at $x=b+$.
315:
316: We assume that $\eta_1^+ > 0$, i.e.,
317: \begin{eqnarray}
318: b < \f {\zeta (\beta)}{ 1 + \zeta (\beta) }.
319: \label{eqn:condition}
320: \end{eqnarray}
321: Note that the uniform density on the interval $[0,1]$ is
322: invariant under the time evolution of this map because the relation
323: $1/{\eta^+_k} + 1/{\eta^-_k}=1$ is satisfied for $k=1,2,\cdots$.
324: Figure \ref{fig:cPMmap} shows the shape of the map $\phi(x)$ for
325: $\beta=1.3$ and $b = 0.5$. There is a singular structure near $x=b$,
326: which suppresses injections of the orbits into neighborhoods of the
327: marginal fixed point $x=0$. This system can be easily extended to a
328: 2-dimensional area-preserving map, whose dynamics of the expanding
329: direction is given by the map $\phi(x)$. This will be discussed in
330: Sec.~\ref{sec:extn}.
331:
332: \subsection{FP operator and functional spaces}
333: The FP operator $\hat P$ and its adjoint $\hat P^{\ast}$
334: are defined by
335: \begin{eqnarray}
336: \hat P \rho (x) &=&
337: \integ_0^1 dy \delta \lp x - \phi(y) \rp \rho (y), \\[.2cm]
338: \hat P^{\ast} A(x) &=& A \lp \phi(x) \rp,
339: \end{eqnarray}
340: respectively. We also define an inner product $(A, \rho)$ as the average
341: of an observable $A(x)$ with respect to a density $\rho (x)$,
342: \begin{equation}
343: (A, \rho) = \integ_0^1 dx A(x) \rho(x).
344: \end{equation}
345: Then, the average of $A(x)$ at time $t$ with respect to an initial
346: density $\rho(x)$ is given by
347: $(A, \hat P^t \rho) \equiv ( \hat P^{\ast t} A, \rho)$.
348:
349: Let us consider an observable $A(x)$ such as the inequality
350: \begin{equation}
351: |A(x) - a_0 - a_1 x| \leq K x^{\beta / (\beta - 1)}
352: \label{eqn:assm-obs}
353: \end{equation}
354: holds for some positive constant $K$, where the constants $a_0$ and
355: $a_1$ satisfy $a_0=A(0)$ and $a_1=A'(0)$, respectively \cite{tasaki1}.
356: This function is bounded on $[0,1]$ and smooth near the origin. And we
357: define a set $X_O$ as the functional space which consists of such
358: observables. This functional space is invariant under the action of the
359: adjoint of the FP operator $\hat P^{\ast}$, namely,
360: $
361: \hat P^{\ast} A(x) \in X_O ~~{\rm if}~~A(x) \in X_O.
362: $
363: If we define the norm as
364: \begin{eqnarray} \nonumber
365: || A(x) ||_O &=& |a_0| + |a_1| + \sup_x |A(x)| \\[.1cm]
366: && + \sup_x \f {|A(x) - a_0 - a_1 x|}{x^{\beta / (\beta - 1)}},
367: \end{eqnarray}
368: then this functional space becomes a Banach space with respect to this
369: norm. Note that this space is dense in the Hilbert space $L^2[0,1]$ of
370: the square integrable functions on $[0,1]$.
371:
372: Furthermore we restrict initial densities to be piecewise constant
373: \cite{tasaki2},
374: \begin{equation}
375: \rho (x) = \tilde \rho_k ~~~{\rm if}~~x \in [\xi^-_k, \xi^-_{k-1})
376: ~~~(k=0,1,2,\cdots).
377: \end{equation}
378: We also assume the following properties for $k=1,2,\cdots$,
379: \begin{eqnarray}
380: \tilde \rho_k = \dip \sum_{l=0}^{\infty} \rho_l \left(\f k{k+l} \right)^{\beta}
381: ~~~{\rm with}~~~
382: \dip \sum_{l=0}^{\infty} |\rho_l| \theta^l < + \infty,
383: \end{eqnarray}
384: where $\theta > 1$ is a constant. We also assume the normalization
385: condition
386: %\footnote{We assume the normalization condition Eq.~(\ref{eqn:normal})
387: %for clarity. But it is not essential in later calculations and almost
388: %the same result can be obtained without it.
389: %(This condition is used directly only in Eq.~(\ref{eqn:equilibrium}).)
390: %}
391: \begin{eqnarray}
392: \f b{\zeta (\beta)}
393: \dip \sum_{k=1}^{\infty} \f {\tilde \rho_k}{k^{\beta}}
394: +(1-b) \tilde \rho_0 = 1.
395: \label{eqn:normal}
396: \end{eqnarray}
397: In the following sections, we use this condition
398: [Eq.~(\ref{eqn:normal})] only for clarity of exposition. But it is not
399: essential and almost the same result can be obtained without it.
400:
401: We define a set $X_D$ as the functional space which consists
402: of such densities. This functional space is invariant under the action
403: of the FP operator;~namely,
404: $
405: \hat P \rho \in X_D ~~~{\rm if}~~~\rho \in X_D.
406: $
407: And if the norm of this space is defined as
408: \begin{equation}
409: || \rho ||_D = \sum_{l=0}^{\infty} |\rho_l| \theta^l,
410: \end{equation}
411: this functional space becomes a Banach space. The functional space $X_D$
412: is not dense in $L^2[0,1]$.
413:
414: In the above definition for initial densities, we assume that the
415: initial densities are constant on the interval $[b,1)$, i.e., the right
416: part of the map. Although it seems to be a strong restriction and it is
417: possible to extend to the densities piecewise constant also in the right
418: part, this extension does not make any changes for the long time
419: behaviors because relaxation to a constant density in the right part
420: finishes by only one iteration of the map $\phi(x)$.
421:
422:
423: \section{\label{sec:sp}Spectral analysis}
424: The purpose of this section is to derive a spectral decomposition of
425: the average $(A, \hat P^t \rho)$. First, we derive the matrix elements
426: of the resolvent operator of $\hat P$, then the average
427: $(A, \hat P^t \rho)$ is obtained by an integral transformation of the
428: matrix elements of the resolvent. Finally, deforming the integration
429: path, we have the spectral decomposition.
430:
431: \subsection{Matrix elements of the resolvent operator}
432:
433: Let us define the matrix elements of the resolvent of the
434: FP operator $\hat P$ as,
435: \begin{eqnarray} \nonumber
436: \left(
437: A, \f 1{z - \hat P} \rho
438: \right)
439: &=& \sum_{t=0}^{\infty} \f 1{z^{t+1}}
440: \left(
441: A, \hat P^t \rho
442: \right) \\[.cm]
443: &=& \sum_{k=1}^{\infty} \left[
444: \tilde \rho_k \hat B_k^- (z) + \tilde \rho_0 \hat B_k^+ (z)
445: \right],
446: \label{eqn:resolvent1}
447: \end{eqnarray}
448: where
449: %$(A, \hat P^t \rho)$ is the average of the observable $A(x)$ at
450: %time $t$ and
451: $\hat B_k^{\pm} (z)$ are defined below [see
452: Eq.~(\ref{eqn:hBk})]. Let us rewrite $(A, \hat P^t \rho)$ as
453: \begin{eqnarray} \nonumber
454: \left(
455: A, \hat P^t \rho
456: \right)
457: &=& \integ_0^1 dx A \circ \phi^t (x) \rho(x) \\[.1cm]
458: &=& \dip \sum_{k=1}^{\infty}
459: \left[
460: \tilde \rho_k B_k^- (t) + \tilde \rho_0 B_k^+ (t)
461: \right],
462: \end{eqnarray}
463: where $B_k^{\pm}(t)$ are defined for $k=1,2,\cdots$ by
464: \begin{eqnarray}
465: B_k^{\pm} (t) \equiv
466: \integ_{\xi_k^{\pm}}^{\xi_{k-1}^{\pm}} dx A \circ \phi^t (x).
467: \label{eqn:Bk}
468: \end{eqnarray}
469: Then $\hat B_k^{\pm}(z)$ are defined for $k=1,2,\cdots$ by
470: \begin{eqnarray}
471: \hat B_k^{\pm}(z) \equiv \sum_{t=0}^{\infty} \f {B_k^{\pm} (t)}{z^{t+1}}.
472: \label{eqn:hBk}
473: \end{eqnarray}
474:
475: From Eq.~(\ref{eqn:Bk}), we have the following recursion relations for
476: $B_k^{\pm}(t)$:
477: \begin{eqnarray}
478: B_1^{\pm} (t+1) &=& \f 1{\eta_1^{\pm}} \sum_{k=1}^{\infty} B_{k}^+ (t),
479: \label{eqn:recB1}\\[.1cm]
480: B_k^{\pm} (t+1) &=& \f 1{\eta_k^{\pm}} B_{k-1}^- (t)
481: ~~~{\rm for}~~k=2,3,\cdots.
482: \label{eqn:recBk}
483: \end{eqnarray}
484:
485: \noindent
486: From Eqs.~(\ref{eqn:hBk})--(\ref{eqn:recBk}), we
487: have the recursion relations for $\hat B_k^{\pm} (z)$:
488: \begin{eqnarray}% checked
489: \hat B_1^{\pm} (z) &=& \f {B_1^{\pm} (0)}z
490: + \f 1{z \eta_1^{\pm}} \sum_{k=1}^{\infty} \hat B_{k}^+ (z),
491: \label{eqn:hatB1pm} \\[.25cm]
492: \hat B_k^{\pm} (z) &=& \f {B_k^{\pm} (0)}z
493: + \f 1{z \eta_k^{\pm}} \hat B_{k-1}^- (z)
494: ~~~{\rm for}~~k=2,3,\cdots.
495: \nonumber\\[-.3cm]
496: \label{eqn:hatBkpm}
497: \end{eqnarray}
498: Using these relations recursively, the following equation can be derived
499: for $k=1,2,\cdots$:
500: \begin{widetext}
501: \begin{eqnarray} %checked
502: \hat B_k^- (z)
503: %\dip \sum_{m=0}^{k-1}
504: %\f {B_{k-m}^- (0)}{z^{m+1} \eta_k^- \cdots \eta_{k-m+1}^-} +
505: %\f 1{z^{k} \eta_k^- \cdots \eta_{1}^-}
506: %\dip \sum_{m=1}^{\infty} \hat B_m^+ (z) \\[.5cm]
507: = \dip \sum_{m=0}^{k-1}
508: \f {B_{k-m}^- (0)}{z^{m+1}} \left( \f {k-m}{k} \right)^{\beta} +
509: \f 1{z^k k^{\beta}} \f {b}{(1-b) \zeta(\beta)}
510: \dip \sum_{m=1}^{\infty} \hat B_m^+ (z) .
511: \label{eqn:hatBk-}
512: %= \dip \sum_{l=1}^{k}
513: %\f {B_{l}^- (0)}{z^{k-l+1}} \left( \f {l}k \right)^{\beta} +
514: %\f 1{z^k k^{\beta}} \f {b}{(1-b) \zeta(\beta)}
515: %\dip \sum_{m=1}^{\infty} \hat B_m^+ (z)
516: \end{eqnarray}
517: Furthermore, from Eqs.~(\ref{eqn:hatBkpm}) and
518: (\ref{eqn:hatBk-}), we obtain for $k=2,3,\cdots$,
519: \begin{eqnarray} %checked
520: \hat B_k^+ (z)
521: &=& \f {B_k^+ (0)}z + \f 1{\eta_k^+}
522: \left[ \dip \sum_{m=0}^{k-2}
523: \f {B_{k-m-1}^- (0)}{z^{m+2}} \left( \f {k-m-1}{k-1} \right)^{\beta} +
524: \f 1{z^{k} (k-1)^{\beta}} \f {b}{(1-b) \zeta(\beta)}
525: \dip \sum_{m=1}^{\infty} \hat B_m^+ (z)
526: \right] .
527: \label{eqn:hatBk+}
528: \end{eqnarray}
529: \end{widetext}
530: Summing up Eq.~(\ref{eqn:hatB1pm}) and Eqs.~(\ref{eqn:hatBk+}), we have
531: \begin{equation}
532: \dip \sum_{k=1}^{\infty} \hat B_k^+ (z) = \f{ \Phi (z)}{ Z(z)},
533: \label{eqn:PhiZ}
534: \end{equation}
535: where $Z(z)$ and $\Phi (z)$ are defined by
536: \begin{eqnarray}%checked
537: Z (z) &\equiv& z - \f 1{\eta_1^+} - \f {b}{(1-b) \zeta(\beta)}
538: \dip \sum_{k=1}^{\infty} \f 1{\eta_{k+1}^+ z^{k} k^{\beta}} ,
539: \label{eqn:Z1} \\[.2cm]
540: %\Phi (z) &\equiv& \dip \sum_{k=1}^{\infty} B_k^+ (0)
541: %+ \dip \sum_{k=1}^{\infty} \dip \sum_{m=0}^{k-1}
542: % \f {B_{k-m}^- (0)}{\eta_{k+1}^+ z^{m+1}} \left( \f {k-m}{k} \right)^{\beta}.
543: \Phi (z) &\equiv&
544: \dip \sum_{k=1}^{\infty} B_k^+ (0)
545: + \dip \sum_{l=1}^{\infty} \dip \sum_{m=0}^{\infty}
546: \f {B_{l}^- (0)}{\eta_{m+l+1}^+ z^{m+1}} \left( \f {l}{m+l} \right)^{\beta}.
547: \nonumber
548: \\[-.2cm]
549: \label{eqn:Phi1}
550: \end{eqnarray}
551: The functions $Z (z)$ and $\Phi (z)$ are absolutely convergent for
552: $|z| > 1$ and thus analytic there. From Eqs.~(\ref{eqn:hatBk-}),
553: and (\ref{eqn:PhiZ})--(\ref{eqn:Phi1}), the matrix elements of the
554: resolvent [Eq.~(\ref{eqn:resolvent1})] can be rewritten as
555: \begin{eqnarray} \nonumber
556: \left(
557: A, \f 1{z - P} \rho
558: \right) &
559: =
560: \f {\Psi (z) \Phi (z)}{Z (z)}
561: + \dip \sum_{l=1}^{\infty} \dip \sum_{k=l}^{\infty}
562: \f {\tilde \rho_{k} B_{l}^- (0)}{z^{k-l+1}} \left( \f {l}k \right)^{\beta}
563: \\[.15cm]
564: %
565: %\hspace*{1.0cm} =
566: %\f {\Psi (z) \Phi (z)}{Z (z)}
567: %+ \dip \sum_{k=1}^{\infty}
568: % \f {\tilde \rho_{k} B_{1}^- (0)}{z^{k} k^{\beta}}
569: %+ \dip \sum_{l=2}^{\infty} B_{l}^- (0) l^{\beta} z^{l-1}\dip \sum_{k=l}^{\infty}
570: % \f {\tilde \rho_{k} }{z^{k} k^{\beta}} \\[.8cm]
571: %
572: &\hspace*{-2.2cm} =
573: \f {\Psi (z) \Phi (z)}{Z (z)}
574: + \f {(1-b) \zeta (\beta)}b
575: \left[
576: \Xi (z) - B_{1}^- (0) \tilde \rho_0
577: \right],
578: \label{eqn:resolvent2}
579: \end{eqnarray}
580: where we define the functions $\Psi (z)$ and $\Xi (z)$ as
581: \begin{eqnarray}
582: \Psi (z) \equiv \tilde \rho_0 + \f b{(1-b) \zeta (\beta)}
583: \sum_{k=1}^{\infty} \f {\tilde \rho_{k} }{z^{k} k^{\beta}},
584: \label{PSI1}
585: \end{eqnarray}
586: and
587: \begin{eqnarray} \nonumber
588: \Xi (z) &=&
589: B_{1}^- (0) \Psi (z)
590: + \dip \sum_{l=2}^{\infty} B_{l}^- (0) l^{\beta} z^{l-1} \\[.cm]
591: && \hspace*{-.3cm}
592: \times \left[
593: \Psi (z) - \tilde \rho_0 - \f b{(1-b) \zeta (\beta)}
594: \dip \sum_{k=1}^{l-1} \f {\tilde \rho_{k} }{z^{k} k^{\beta}}
595: \right],
596: \label{XI1}
597: \end{eqnarray}
598: respectively. These functions $\Psi (z)$ and $\Xi(z)$ are also
599: absolutely convergent and analytic for $|z| > 1$.
600:
601: \subsection{Analytic properties of individual functions}
602: For $ r > 1$, we obtain the average $(A, \rho_t)$
603: by an integral transformation of the resolvent
604: [Eq.~(\ref{eqn:resolvent2})] as follows:
605: \begin{eqnarray} \nonumber
606: \left(
607: A, \hat P^t \rho
608: \right)
609: &=& \dip \oint_{|z|=r} \f {dz}{2 \pi i} z^t
610: \left(
611: A, \f 1{z - P} \rho
612: \right) \\[.25cm]
613: &&\hspace*{-0.9cm}= \dip \oint_{|z|=r} \f {dz}{2 \pi i} z^t
614: \left[
615: \f {\Psi (z) \Phi (z)}{Z (z)} + \f {(1-b) \zeta (\beta)}b \Xi (z)
616: \right], \nonumber
617: \\[-.12cm]
618: \label{eqn:resolvent3}
619: \end{eqnarray}
620: where the integration path is taken in the counter clockwise direction.
621: In order to deform this integration path into the unit disk $|z|<1$ and
622: derive the spectral decomposition, we study the analytic properties of
623: the functions in the integrand of Eq.~(\ref{eqn:resolvent3}) in this
624: subsection. With the help of the identity
625: \begin{eqnarray}
626: \f 1{k^{\beta}} = \f 1{\Gamma(\beta)} \integ_0^{\infty} ds~s^{\beta -1} e^{-ks},
627: \end{eqnarray}
628: we have analytic continuations of the functions $Z(z)$, $\Phi(z)$
629: and $\Psi(z)$ into the unit disk $|z|<1$. For example, $\Phi(z)$ can be
630: analytically continued as follows:
631: \begin{widetext}
632: \begin{eqnarray}
633: \Phi (z) &=& \dip \sum_{k=1}^{\infty} B_k^+ (0)
634: + \f 1{\Gamma (\beta)} \dip \sum_{l=1}^{\infty}
635: \f {B_{l}^- (0) l^{\beta}}{z}
636: \integ_0^{\infty} ds s^{\beta -1} e^{-ls} (1-e^{-s})
637: \dip \sum_{m=0}^{\infty} (z^{-1}e^{-s})^m
638: \\[.35cm]
639: &=& \dip \sum_{k=1}^{\infty} B_k^+ (0) +
640: \f 1{\Gamma (\beta)} \sum_{l=1}^{\infty} B_l^- (0) l^{\beta}
641: \integ_0^{\infty} ds \f {s^{\beta -1} e^{-ls}}{z - e^{-s}} (1-e^{-s}),
642: \label{eqn:Phi}
643: \end{eqnarray}
644: where these calculations are justified because the convergence of the
645: summation is uniform in $s$ for $|z| > 1$. Note that each integral in
646: the right-hand side (rhs) of Eq.~(\ref{eqn:Phi}) is analytic except for
647: the real interval $[0,1]$. Then the function $\Phi(z)$ expressed as
648: Eq.~(\ref{eqn:Phi}) is analytic except for the cut $[0,1]$, because the
649: second infinite sum in Eq.~(\ref{eqn:Phi}) is absolutely convergent, i.e.,
650: \begin{eqnarray}
651: \f 1{\Gamma (\beta)} \sum_{l=1}^{\infty}
652: \left| B_l^- (0) l^{\beta}
653: \integ_0^{\infty} ds \f {s^{\beta -1} e^{-ls}}{z - e^{-s}} (1-e^{-s})
654: \right|
655: \leq \f {b \dip \sup_x |A(x)| }{\zeta (\beta) d(z,[0,1])},
656: \end{eqnarray}
657: where $d(z, [0,1])$ is the distance between $z$ and the real interval
658: $[0,1]$.
659:
660: In the same way, we obtain analytical continuations of the functions
661: $Z(z)$ and $\Psi(z)$:
662: \begin{eqnarray}
663: Z(z) &=& (z-1)
664: \left[
665: 1 + \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
666: \integ_0^{\infty} ds \f {s^{\beta -1} e^{-s}}{z - e^{-s}}
667: \right], \\[.3cm]
668: %
669: \Psi (z) &=& \tilde \rho_0 + \f b{(1-b)\zeta (\beta) \Gamma(\beta)}
670: \dip \sum_{l=0}^{\infty} \rho_l
671: \integ_0^{\infty} ds \f {s^{\beta -1} e^{-(l+1)s}}{z - e^{-s}}.
672: \end{eqnarray}
673: These expressions for the functions $Z(z)$ and $\Psi (z)$ are also
674: analytic except for the cut $[0,1]$.
675: %
676: And the function $\Xi (z)$ can be also analytically continued into the
677: unit disk $|z|<1$ in the same way. Here, however, we rewrite the
678: function $\Xi(z)$ in the following form:
679: \begin{eqnarray}
680: \Xi (z) = \Psi(z) \dip \sum_{l=1}^{\infty} l^{\beta} B_l^- (0) z^{l-1}
681: - \dip \sum_{l=2}^{\infty} l^{\beta} B_l^- (0) z^{l-1}
682: \left[
683: \tilde \rho_0 + \f b{(1-b)\zeta (\beta)}
684: \sum_{k=1}^{l-1} \f {\tilde \rho_k}{z^k k^{\beta}}
685: \right].
686: \end{eqnarray}
687: The infinite sums in this expression of the function $\Xi (z)$ are
688: absolutely convergent for $|z| < 1$. \vspace*{.3cm}
689: \end{widetext}
690:
691: %\begin{equation}
692: %\f 1{\Gamma (\beta)}
693: %\left|
694: %\integ_0^{\infty} ds \f {s^{\beta -1} e^{-s}}{z - e^{-s}}
695: %\right|
696: %\leq \f 1{d(z,[0,1])},
697: %\end{equation}
698: %\begin{eqnarray}
699: %\dip \sum_{l=2}^{\infty} B_{l}^- (0) l^{\beta} z^{l-1}
700: %\dip \sum_{k=1}^{l-1} \f {\tilde \rho_{k} }{z^{k} k^{\beta}}
701: %&\leq&
702: %C \dip \sum_{l=2}^{\infty}
703: %\dip \sum_{k=1}^{l-1}
704: %\f {z^{l-k-1} \tilde \rho_{k} }{k^{\beta}} \\[.7cm]
705: %&=&
706: %C \dip \sum_{l=2}^{\infty}
707: %\dip \sum_{m=1}^{l-1}
708: %\f {z^{m-1} \tilde \rho_{l-m} }{(l-m)^{\beta}} \\[.7cm]
709: %&=&
710: %C
711: %\dip \sum_{m=1}^{\infty}
712: %\dip \sum_{l=m+1}^{\infty}
713: %\f {z^{m-1} \tilde \rho_{l-m}}{(l-m)^{\beta}} \\[.7cm]
714: %\end{eqnarray}
715:
716: Let us consider the zeros of the function $Z (z)$. First we
717: define a function $\Omega (z)$ as $\Omega (z) \equiv {Z(z)}/(z-1)$.
718: %\begin{equation}
719: %\Omega (z) \equiv \f {Z(z)}{z-1}
720: %= 1 + \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
721: %\integ_0^{\infty} ds \f {s^{\beta-1} e^{-s}}{z - e^{-s}}
722: %\end{equation}
723: And if ${\rm Im} (z) \ne 0$, then
724: \begin{equation}
725: {\rm Im~}\Omega (z)
726: = - \f {b {\rm Im(z)}}{(1-b) \zeta (\beta) \Gamma (\beta)}
727: \integ_0^{\infty} ds \f {s^{\beta-1} e^{-s}}{|z - e^{-s}|^2} \ne 0.
728: \end{equation}
729: And also if ${\rm Im} z = 0$ and ${\rm Re} z > 1$, then $\Omega(z)>0$,
730: ~because $z-e^{-s} > 0$. Thus there is not zero of $Z(z)$ in these
731: regions.
732:
733: Next if ${\rm Im}z = 0$ and ${\rm Re} z < 0$, we have
734: \begin{eqnarray}
735: \Omega'(z) =
736: - \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
737: \integ_0^{\infty} ds \f {s^{\beta-1} e^{-s}}{(z - e^{-s})^2} < 0,
738: \end{eqnarray}
739: and
740: \begin{eqnarray}
741: \begin{array}{lll}
742: \Omega(-1) &=& 1 - \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
743: \integ_0^{\infty} ds \f {s^{\beta-1} }{e^s + 1} \\[.5cm]
744: %
745: &>& 1 - \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
746: \integ_0^{\infty} ds \f {s^{\beta-1} }{e^s} \\[.5cm]
747: %
748: &=& 1 - \f b{(1-b) \zeta(\beta)} > 0
749: \end{array}
750: \end{eqnarray}
751: The last inequality holds because of the
752: Eq.~(\ref{eqn:condition}). And it is easy to see that
753: $\Omega (z) \to - \infty$ as $z \to 0$.
754: Therefore, on the real interval $[-1,0]$, $\Omega(z)$ has the unique
755: zero $\lambda_d$ of order 1.
756:
757: \subsection{Spectral decomposition}
758:
759: From the results of the last subsection, the contour of the integration of
760: Eq.~(\ref{eqn:resolvent3}) can be deformed into the unit disk $|z| < 1$ as
761: \begin{eqnarray} \nonumber
762: \left(
763: A, \hat P^t \rho
764: \right)
765: &=& \dip \lim_{\eta \to 0} \dip \oint_{|z-\lambda_d|= \eta} \f {dz}{2 \pi i} z^t
766: \f {\Psi (z) \Phi (z)}{Z (z)}
767: \hspace*{2.cm} \\[.38cm]
768: &&
769: \hspace*{-1.5cm}
770: + \dip \lim_{\eta \to 0} \dip \oint_C \f {dz}{2 \pi i} z^t
771: \left[
772: \f {\Psi (z) \Phi (z)}{Z (z)} + \f {(1-b) \zeta (\beta)}b \Xi (z)
773: \right],
774: \label{eqn:average1}
775: \end{eqnarray}
776: where the integration path $C$ is defined by
777: \begin{eqnarray} \nonumber
778: C \equiv
779: \left\{
780: ~z~ |~ |z| = \eta,~|z-1|=\eta~
781: \right. \hspace*{2cm} \\[.28cm]
782: \left.
783: {\rm or}~z=\lambda \pm i0 ~(\eta < \lambda < 1 - \eta)
784: \right\}.
785: \end{eqnarray}
786: The first term of the rhs of Eq.~(\ref{eqn:average1}) corresponds to the
787: simple pole at $z= \lambda_d$ and thus can be calculated by the residue
788: theorem and Eq.~(\ref{eqn:average1}) is rewritten as
789: \begin{eqnarray}\nonumber
790: \left(
791: A, \hat P^t \rho
792: \right)
793: &=& {\lambda_d}^t
794: \f {\Psi (\lambda_d) \Phi (\lambda_d)}{(\lambda_d - 1) \Omega'
795: (\lambda_d)}
796: \hspace{3.7cm} \\[0.38cm]
797: && \hspace{-1.5cm}
798: + \dip \lim_{\eta \to 0} \dip \oint_C \f {dz}{2 \pi i} z^t
799: \left[
800: \f {\Psi (z) \Phi (z)}{Z (z)} + \f {(1-b) \zeta (\beta)}b \Xi (z)
801: \right].
802: \label{eqn:average2}
803: \end{eqnarray}
804:
805: Let us consider the integral of rhs of Eq.~(\ref{eqn:average2}).
806: Because of the facts (see the appendix)
807: \begin{eqnarray}
808: \dip \lim_{z \to 0} & z^{\alpha} \Xi (z) = 0,
809: \label{eqn:xito0}
810: % \\[.2cm]
811: \end{eqnarray}
812: and
813: \begin{eqnarray}
814: \dip \lim_{z \to 0} & z^{\alpha} \f {\Phi (z) \Psi (z)}{Z (z)} = 0,
815: \label{eqn:zto0}
816: \end{eqnarray}
817: for ${^{\forall}}\alpha > 0$, the contribution from the integration
818: around the origin, $\{ |z|=\eta \}$, vanishes.
819:
820: Similarly, we obtain (see the appendix)
821: \begin{eqnarray}
822: \lim_{z \to 1} (z-1) \Xi (z) = 0,
823: \label{eqn:xito1}
824: \end{eqnarray}
825: and then the contribution from the integral around $z=1$,
826: $\{|z-1|=\eta \}$, vanishes for the second term of the integrand. On the
827: other hand, we have
828: %\footnote{
829: %As is seen from Eqs.~(\ref{eqn:cut-asymp1}) and (\ref{eqn:cut-asymp2}),
830: %the contribution from the cut to the integration arround $z=1$ is of
831: %order $(1-\lambda)^{\beta -2}$ when $\lambda \approx 1$. Therefore the
832: %effect of the cut to the integral can be neglected as $\eta \to 0$.
833: %}
834: \begin{eqnarray} \nonumber
835: \dip \lim_{\eta \to 1} \integ_{|z-1|=\eta} dz
836: z^t \f {\Psi (z) \Phi (z)}{Z (z)}
837: \\[0.3cm] \nonumber
838: &&\hspace*{-3cm}=
839: \dip \lim_{z \to 1} (z-1) \f {\Phi (z) \Psi (z)}{Z (z)}
840: \\[0.3cm] \nonumber
841: &&\hspace*{-3cm}= \integ_0^{1} dx A(x)
842: \left[
843: (1-b) \tilde \rho_0 + \f {b}{\zeta (\beta)}
844: \sum_{k=1}^{\infty} \f {\tilde \rho_k}{k^{\beta}}
845: \right]
846: \\[.3cm]
847: &&\hspace*{-3cm}= \integ_0^{1} dx A(x), \\[-.05cm] \nonumber
848: \label{eqn:equilibrium}
849: \end{eqnarray}
850: where we have used the normalization condition Eq.~(\ref{eqn:normal}).
851: The rhs of Eq.~(\ref{eqn:equilibrium}) is the average of $A(x)$ with
852: respect to the invariant density which is uniform for the map
853: $\phi(x)$;~therefore Eq.~(\ref{eqn:equilibrium}) gives the average value
854: of the invariant state.
855:
856: Finally we have to evaluate the integral along the cut. For this
857: purpose, we define the new functions
858: $\hat \Omega (\lambda), ~f_{\lambda}(x)$
859: and $\nu_l(x)$ for $0< \lambda < 1$ as follows,
860: \begin{eqnarray}
861: %%%%%%%%%%%%%%% Z(z) %%%%%%%%%%%%%
862: \hat \Omega (\lambda) &\equiv&
863: 1 + \f {b}{(1-b) \zeta (\beta) \Gamma (\beta)}
864: \integ_0^{\infty} \hspace*{-.2cm}
865: ds {\mathcal P} \f {s^{\beta -1} e^{-s}}{\lambda -e^{-s}} ,
866: \hspace*{1.cm}\\[.5cm]
867: %%%%%%%%%%%%%%% ¦µ(z) %%%%%%%%%%%%%
868: \nonumber
869: f_{\lambda} (x) &\equiv&
870: \f 1{1-b}
871: \left[
872: \chi_0 (x) +
873: \f 1{\Gamma (\beta)} \sum_{l=1}^{\infty} l^{\beta} \chi_l (x)
874: \right.
875: \\[.3cm]
876: &&
877: \left.
878: \hspace*{1.3cm} \times \integ_0^{\infty} ds {\mathcal P}
879: \f {s^{\beta -1} e^{-ls}}{\lambda - e^{-s}} (1-e^{-s})
880: \right] , \\[.5cm]
881: %%%%%%%%%%%%%%% ¦·(z) %%%%%%%%%%%%%
882: \nu_l (\lambda) &\equiv&
883: \f 1{\zeta (\beta) \Gamma (\beta)}
884: \integ_0^{\infty} ds {\mathcal P} \f {s^{\beta -1} e^{-(l+1)s}}{\lambda- e^{-s}} ,
885: \end{eqnarray}
886: where ${\mathcal P}$ means the Cauchy's principle value and $\chi_l(x)$ is
887: defined, for $l=0,1,\cdots$, by
888: \begin{eqnarray}
889: \chi_l(x) =
890: \left\{
891: \begin{array}{ll}
892: 1 & ~~~{\rm for}~~\xi^-_l \leq x < \xi^-_{l-1}, \\[.5cm]
893: 0 & ~~~{\rm otherwise}.
894: \end{array}
895: \right.
896: \end{eqnarray}
897: Note that we have defined $\xi_{-1}=1$ in Sec.~{\ref{sec:defs}}. These
898: functions are related to the real parts of the functions
899: $\{Z(z), \Phi(z), \Psi(z)\}$ near the cut:~
900: ${\rm Re} Z(\lambda \pm i0) = (\lambda -1) \hat \Omega (\lambda)$,
901: ${\rm Re} \Phi(\lambda \pm i0) = (1-b)\int_0^1 dx A(x) f_{\lambda}(x)$, and
902: ${\rm Re} \Psi(\lambda \pm i0) =
903: \tilde \rho_0 + b/(1-b) \sum_{l=0}^{\infty} \rho_l \nu_l(\lambda)$.
904:
905: It can also be shown for the imaginary parts of the functions
906: $\{Z(z), \Phi(z), \Psi(z)\}$ near the cut that
907: \begin{eqnarray}
908: \f {{\rm Im} \Phi (\lambda - i0)}{{\rm Im} Z (\lambda + i0)}
909: &=& \f {(1-b) \zeta (\beta)}b
910: \sum_{l=1}^{\infty} l^{\beta} B_l^-(0) \lambda^{l-1}, \\[.5cm]
911: \f {{\rm Im} \Psi (\lambda - i0)}{{\rm Im} Z (\lambda + i0)}
912: &=& \f 1{1-\lambda}
913: \sum_{l=0}^{\infty} \rho_l \lambda^l.
914: \end{eqnarray}
915:
916: Using these functions, we can calculate the integral along the cut as
917: follows \cite{tasaki1}:
918: \begin{widetext}
919: \begin{eqnarray}\nonumber
920: \dip \lim_{\eta \to 0} && \hspace*{-.3cm}
921: \integ_{C \backslash \{|z|=\eta ~\cup ~|z-1|=\eta \}}
922: \f {dz}{2 \pi i} z^t
923: \left[
924: \f {\Psi (z) \Phi (z)}{Z (z)} + \f {(1-b)\zeta (\beta)}b \Psi (z)
925: \sum_{l=1}^{\infty} l^{\beta} B_l^- (0) z^{l-1}
926: \right] \\[0.3cm]
927: %
928: &=& \integ_0^1 \f {d \lambda}{\pi} \lambda^t
929: \f {{\rm Im} \left[ Z (\lambda + i0) \Phi (\lambda - i0) \right]
930: {\rm Im} \left[ Z (\lambda + i0) \Psi (\lambda - i0) \right] }
931: { |Z (\lambda + i0)|^2 {\rm Im}~Z(\lambda + i0) }
932: \\[.3cm]
933: %\nonumber
934: &=& \f b{(1-b) \zeta (\beta) \Gamma (\beta)}
935: \integ_0^1 d \lambda
936: \f { {\rm Im} \left[ Z (\lambda + i0) \Phi (\lambda - i0) \right] }
937: { {\rm Im}~Z(\lambda + i0) } ~
938: \f { \lambda^t (1 - \lambda) \left( \log \frac 1{\lambda} \right)^{\beta - 1} }
939: { |Z (\lambda + i0)|^2 }~
940: \f { {\rm Im} \left[ Z (\lambda + i0) \Psi (\lambda - i0) \right] }
941: { {\rm Im}~Z(\lambda + i0) } ~ \\[.3cm]
942: &=& \integ_0^1 d \lambda ~(A, F_{\lambda}) \lambda^t (\tilde F_{\lambda}, \rho),
943: \label{eqn:cut}
944: \end{eqnarray}
945: where we define the linear functional $(A, F_{\lambda})$ of the observables
946: $A(x)$ as
947: \begin{eqnarray}
948: (A, F_{\lambda}) \equiv
949: N (\lambda) \integ_0^1 dx \{ A(x) - A(0) \}
950: %
951: \left[
952: f_{\lambda} (x)
953: + \f {(\lambda - 1) \hat \Omega (\lambda) \zeta (\beta)}b
954: \dip \sum_{l=1}^{\infty} l^{\beta} {\lambda}^{l-1} \chi_l (x)
955: \right].
956: \label{eqn:l-func-ob}
957: \end{eqnarray}
958: The function $N(\lambda)$ in Eq.~(\ref{eqn:l-func-ob}) is given by
959: \begin{eqnarray}
960: N(\lambda) \equiv
961: \frac { \frac b{\zeta (\beta) \Gamma (\beta) (1-b)}
962: \left( \log \frac 1{\lambda} \right)^{\beta - 1} }
963: {
964: (1 - \lambda) \left\{
965: \hat \Omega^2 (\lambda) +
966: \left[
967: \frac {b}{\zeta (\beta) \Gamma (\beta) (1-b)}
968: \left( \log \frac 1 {\lambda} \right)^{\beta -1}
969: \right]^2 \right\} }
970: \end{eqnarray}
971: \end{widetext}
972: We also define in Eq.~(\ref{eqn:cut}) the linear functional
973: $(\tilde F_{\lambda}, \rho)$ of the initial densities $\rho (x)$ as
974: \begin{equation}
975: ( \tilde F_{\lambda}, \rho ) \equiv
976: (1-b) \tilde \rho_0 + b \dip \sum_{l=0}^{\infty} \rho_l
977: \left[
978: \nu_l (\lambda) - \f {1-b}b \hat \Omega (\lambda) \lambda^l
979: \right].
980: \end{equation}
981:
982: Similarly, the linear functionals associated to the eigenvalues
983: $z=\lambda_d$ and $z=1$, that have been obtained in
984: Eqs.~(\ref{eqn:average2}) and (\ref{eqn:equilibrium}), can be
985: expressed as
986: \begin{eqnarray}
987: (A, F_{d}) &\equiv& \f {1}{(\lambda_d - 1) \Omega'(\lambda_d)}
988: \integ_0^1 dx A(x) f_{\lambda_d}(x), \\[.1cm]
989: ( \tilde F_{d}, \rho ) &\equiv&
990: (1-b) \tilde \rho_0 +
991: b \dip \sum_{l=0}^{\infty} \rho_l \nu_l (\lambda_d),
992: \end{eqnarray}
993: and
994: \begin{eqnarray}
995: (A, F_{\rm in}) &\equiv& \integ_0^1 dx A(x), \\[.1cm]
996: ( \tilde F_{\rm in}, \rho ) &\equiv& 1,
997: \end{eqnarray}
998: respectively. It is easy to check that $(A, F_{\lambda}) = (A, F_{d}) = 0$~ if
999: $A(x)$ is a constant function and that
1000: $(\tilde F_{\lambda}, \rho) = (\tilde F_{d}, \rho) = 0$
1001: if $\rho (x) \equiv 1$.
1002: %\begin{eqnarray}
1003: %\dip \lim_{\lambda \to 1} ( \tilde F_{\lambda}, \rho )
1004: %%
1005: %= (1-b) \tilde \rho_0 + b \dip \sum_{l=0}^{\infty} \rho_l
1006: %\nu_l (1) - \dip \sum_{l=0}^{\infty} \rho_l
1007: %%
1008: %= ( \tilde F_{\rm in}, \rho ) - \dip \sum_{l=0}^{\infty} \rho_l
1009: %\end{eqnarray}
1010:
1011: Using these linear functionals, we can spectrally decompose
1012: the average $(A, \rho_t)$ [Eq.~(\ref{eqn:average2})] at time
1013: $t = 0, 1, 2, \cdots$ as,
1014: \begin{eqnarray} \nonumber
1015: (A,\hat P^t \rho) &=& (A, F_{\rm in})(\tilde F_{\rm in}, \rho)
1016: + {\lambda_d}^t (A, F_{d})(\tilde F_{d}, \rho) \hspace*{1cm} \\[.25cm]
1017: &&+ \integ_0^1 d \lambda \lambda^t
1018: (A, F_{\lambda})(\tilde F_{\lambda}, \rho).
1019: \label{eqn:sp-dc}
1020: \end{eqnarray}
1021: The left eigenfunctions
1022: $\{ F_{\rm in}, F_{\rm d}, F_{\lambda} \}$
1023: satisfy the relations
1024: \begin{eqnarray}
1025: \begin{array}{lllll}
1026: (A,\hat P F_{\rm in})
1027: &\equiv& (\hat P^{\ast} A, F_{\rm in})
1028: &=& (A, F_{\rm in}), \\[.25cm]
1029: (A,\hat P F_{\rm d})
1030: &\equiv& (\hat P^{\ast} A, F_{\rm d})
1031: &=& \lambda_d(A, F_{\rm d}), \\[.25cm]
1032: (A,\hat P F_{\lambda})
1033: &\equiv& (\hat P^{\ast} A, F_{\lambda})
1034: &=& \lambda (A, F_{\rm \lambda}).
1035: \end{array}
1036: \end{eqnarray}
1037: Therefore $\{ F_{\rm in}, F_{\rm d}, F_{\lambda} \}$ are
1038: eigenfunctions of the FP operator in a generalized sense
1039: \cite{gelfand1,gelfand2,tasaki1,tasaki2}.
1040: On the other hand, the right eigenfunctions
1041: $\{ \tilde F_{\rm in}, \tilde F_{\rm d}, \tilde F_{\lambda} \}$
1042: satisfy the relations
1043: \begin{eqnarray}
1044: \begin{array}{lllll}
1045: (\hat P^{\ast} \tilde F_{\rm in}, \rho)
1046: &\equiv& (\tilde F_{\rm in}, \hat P \rho)
1047: &=& (\tilde F_{\rm in }, \rho), \\[.25cm]
1048: (\hat P^{\ast} \tilde F_{\rm d}, \rho)
1049: &\equiv& (\tilde F_{\rm d}, \hat P \rho)
1050: &=& \lambda_d (\tilde F_{\rm d}, \rho), \\[.25cm]
1051: (\hat P^{\ast} \tilde F_{\lambda}, \rho)
1052: &\equiv& (\tilde F_{\lambda}, \hat P \rho)
1053: &=& \lambda (\tilde F_{\lambda}, \rho) .
1054: \end{array}
1055: \end{eqnarray}
1056: Therefore
1057: $\{ \tilde F_{\rm in}, \tilde F_{\rm d}, \tilde F_{\lambda} \}$
1058: are eigenfunctions of the adjoint of the FP operator in a generalized
1059: sense.
1060:
1061: \begin{figure*}
1062: \vspace*{-.cm}
1063: \ig[width=15.7cm,height=7.5cm]{EPS/PSD-EXP.eps} \vspace*{-.1cm}
1064: \caption{(a)~Power spectral densities $S(f)$ of time series $A(x(t))$
1065: (in log-log form) for five different values of the system parameter
1066: $\beta: \beta = $1.2~(the solid line), 1.4~(dotted), 1.6~(broken),
1067: 1.8~(dashed), 2.0~(dashed-dotted), 2.2~(dashed-two-dotted). These PSDs
1068: are obtained by averaging over 20000 initial conditions uniformly
1069: distributed in $[0,1]$.
1070: ~(b) The scaling exponent $\gamma$ of the PSD
1071: $S(f) \sim 1/f^{\gamma}$ as a function of $\beta$. The
1072: circles are the numerical results obtained by least square
1073: fitting in the low frequency region (below $f=10^{-4}$) of the PSDs
1074: $S(f)$;~and the dashed line is the theoretical prediction
1075: [Eq.~(\ref{eqn:psd})]. }
1076: \vspace{.cm}
1077: \label{fig:PSD-NHBmap}
1078: \end{figure*}
1079:
1080: Thus we obtain the spectral decomposition Eq.~(\ref{eqn:sp-dc}), where
1081: the spectrum consists of two discrete eigenvalues $1$ and $\lambda_d$,
1082: and a continuous spectrum on $[0,1]$. And Eq.~(\ref{eqn:sp-dc}) for
1083: $t = 0$ shows that these eigenfunctions are complete.
1084:
1085:
1086: \section{\label{sec:extn}
1087: Long time behaviors and an area-preserving extension}
1088:
1089: \subsection{Long time behaviors}
1090:
1091: In the previous section, we derive the spectral decomposition of the
1092: average $(A, \hat P^t \rho)$ and found that there is a continuous
1093: spectrum. Since long time behaviors are controlled by eigenvalues whose
1094: absolute values are close to 1, we consider the limit $\lambda \to 1$
1095: for the continuous spectrum. The eigenstate associated with the
1096: eigenvalue $\lambda_d$ does not contribute to long time behaviors,
1097: because this state decays exponentially. Under the assumption
1098: Eq.~(\ref{eqn:assm-obs}), the leading term of the left eigenstate $(A,
1099: F_{\lambda})$ when $\lambda \simeq 1$ is given by
1100: \begin{equation}
1101: (A, F_{\lambda}) \simeq K (1- \lambda)^{\beta - 2}
1102: \integ_0^1 dx [ A(x) -A(0) ]
1103: \label{eqn:cut-asymp1}
1104: \end{equation}
1105: where $K$ is a constant. Similarly, we obtain the leading term for the
1106: right eigenstate $(\tilde F_{\lambda}, \rho)$,
1107: \begin{eqnarray} \nonumber
1108: (\tilde F_{\lambda}, \rho) \simeq
1109: (\tilde F_{\rm in}, \rho) - \sum_{l=0}^{\infty} \rho_l
1110: = 1 - \sum_{l=0}^{\infty} \rho_l
1111: \label{eqn:cut-asymp2}
1112: \end{eqnarray}
1113: as $\lambda \simeq 1$. Using these facts and Eq.~(\ref{eqn:sp-dc}), we
1114: have for $t \to \infty$
1115: \begin{eqnarray} \nonumber
1116: (A, \hat P^t \rho) &\simeq&
1117: \integ_0^1 dx A(x) \\[.2cm]
1118: && \hspace*{-1.2cm}+ \f {K'}{t^{\beta - 1}}
1119: \left( 1 - \dip \sum_{l=0}^{\infty} \rho_l \right)
1120: \integ_0^1 dx [ A(x) - A(0) ],
1121: \label{eqn:correlation}
1122: \end{eqnarray}
1123: where $K'$ is a constant. Eq.~(\ref{eqn:correlation}) shows that the
1124: correlation functions decay algebraically.
1125: %
1126: From Eq.~(\ref{eqn:correlation}), it is found that the power spectral
1127: density (PSD) $S(f)$ behaves as \cite{schuster}
1128: \begin{eqnarray}
1129: S(f) \sim
1130: \left\{
1131: \begin{array}{lll}
1132: \f 1{f^{2-\beta}} &~~{\rm for}~~& 1 < \beta < 2, \\[.39cm]
1133: |\log f| &~~{\rm for}~~& \beta = 2, \\[.39cm]
1134: {\rm const.} &~~{\rm for}~~& \beta > 2.
1135: \label{eqn:psd}
1136: \end{array}
1137: \right.
1138: \end{eqnarray}
1139: %$S(f) \sim 1/f^{2-\beta}$ for $1 < \beta < 2$,
1140: %$S(f) \sim |\log f|$ for $\beta = 2$, and
1141: %$S(f) \sim {\rm const.}$ for $\beta > 2 ~$\cite{schuster}.
1142:
1143: Figure~\ref{fig:PSD-NHBmap}(a) shows PSDs $S(f)$ of time series
1144: $A(x(t))$, where the observable $A(x)$ is a step function defined on
1145: $x \in [0,1)$ as
1146: \begin{eqnarray}
1147: A(x) =
1148: \left\{
1149: \begin{array}{lll}
1150: -1 &~~{\rm for}~~& x \in [0, 1/2) \\[.25cm]
1151: 1 &~~{\rm for}~~& x \in [1/2, 1).
1152: \end{array}
1153: \right.
1154: \end{eqnarray}
1155: %$A(x) = -1$ for $x \in [0, 1/2)$ and $A(x) = 1$ for $x \in [1/2, 1)$.
1156: And $x(t)$ is produced by the successive iterations of the map $\phi(x)$:
1157: $x(t+1) = \phi (x(t))$. Each PSD is obtained by averaging over 20000
1158: initial conditions uniformly distributed in $[0,1]$.
1159:
1160: %\footnote
1161: %{
1162: % Using ensemble average, the autocorrelation function for the observable
1163: % $A(x)$ is defined by $C(t) = \int dx A(x_t)A(x_0) \rho_0(x_0)$. In our
1164: % case, the invariant density $\rho_0(x_0)$ is constant
1165: % $\rho_0(x_0)=1$. Then,
1166: %$
1167: %C(t) = \int dx A(x_t)A(x_0)
1168: % = \int dx \hat P^{\ast t}A(x_0) \{A(x_0) + 1 \}
1169: %$.
1170: %}.
1171:
1172: \begin{figure*}
1173: \ig[width=15.0cm,height=8.8cm]{EPS/nhb.eps} \vspace*{-.cm}
1174: \caption{ The area-preserving map $\psi(x,y)$ defined by
1175: Eq.~(\ref{eqn:2d-model}) for $b=0.5$ and $\beta=1.3$. The domains with
1176: labels \{A, B, C, a, b, c\} in the left cell (I) are mapped into the
1177: domains with the same labels in the right (II), respectively.
1178: Each domain is uniformly stretched in the horizontal direction and
1179: uniformly squeezed in the vertical direction. There are infinite number
1180: of such domains and each one is mapped in a similar way. The partition
1181: is displayed only for the region $\{x \in [\xi_{15}^-,\xi_{0}^-]\}$ and
1182: $\{ x \in [\xi_{6}^+,\xi_{0}^+]\}$ in the left cell (I); only for
1183: $x \in [\xi_{15}^-, \xi_0^+]$ in the right (II). The other regions
1184: are not displayed because the structures are too fine to see. }
1185: \label{fig:NHBmap}
1186: \vspace{-.cm}
1187: \end{figure*}
1188:
1189: In Fig.~\ref{fig:PSD-NHBmap}(a), the PSDs for $\beta < 2$ exhibit clear
1190: $1/f^{\gamma}$ scalings in low frequency regions. Figure
1191: \ref{fig:PSD-NHBmap} displays this scaling exponent $\gamma$ of the PSD
1192: $S(f) \sim 1/f^{\gamma}$ as a function of the system parameter
1193: $\beta$. And we show the theoretical prediction derived in the above
1194: [Eq.~(\ref{eqn:psd})] by the dashed line in
1195: Fig.~\ref{fig:PSD-NHBmap}(b). Obviously, the numerical results show a
1196: good agreement with the theoretical prediction.
1197: %For other choice of observables, see Ref.~\cite{tm3}.
1198:
1199: %\begin{widetext}
1200: %\onecolumngrid
1201: \subsection{Area-preserving extension}
1202:
1203: As stated in the introduction, the 1 dimensional map investigated in the
1204: present paper can be extended to an area-preserving 2-dimensional
1205: transformation defined on the unit square; this extension
1206: $\psi: [0,1)^2 \to [0,1)^2$ is defined as
1207: \begin{eqnarray}
1208: %(x', y') =
1209: \psi (x, y) =
1210: \left\{
1211: \begin{array}{l}
1212: \left(
1213: \eta_{k}^{-} (x - \xi_{k}^{-}) + \xi_{k-1}^{-},~
1214: \f y{\eta_k^-}
1215: \right) \\[.38cm]
1216: \hspace*{2.7cm}
1217: {\rm for} ~x \in [\xi_{k}^{-},\xi_{k-1}^{-}), \\[.33cm]
1218: %
1219: \left(
1220: \eta_{k}^{+} (x - \xi_{k}^{+}) + \xi_{k-1}^{-},~
1221: \f y{\eta_k^+} + \f 1{\eta_k^-}
1222: \right) \\[.38cm]
1223: %\hspace*{1.08cm}
1224: \hspace*{2.7cm}
1225: {\rm for} ~x \in [\xi_{k}^{+},\xi_{k-1}^{+}).
1226: \end{array}
1227: \right.
1228: \label{eqn:2d-model}
1229: \end{eqnarray}
1230: where $k=1,2,\cdots$. Obviously, the transformation for the horizontal
1231: coordinate $x$ is the same as the map $\phi(x)$ defined by
1232: Eq.~(\ref{eqn:1d-model}) and does not depend on the vertical coordinate
1233: $y$. This relation between 1-dimensional map $\phi(x)$ and its
1234: area-preserving extension $\psi(x,y)$ is the same as that between the
1235: Bernoulli and the baker transformations. Note that the map $\psi(x,y)$
1236: is area-preserving, because the Jacobian of this map equals to $1$
1237: everywhere. The phase space (i.e., the unit square) of this
1238: area-preserving map can be partitioned into infinite pieces like
1239: \begin{eqnarray} \nonumber
1240: [0,1)^2 &=&
1241: \bigcup_{k=1}^{ \infty} \{(x,y) ~|~ x \in [\xi_k^-, \xi_{k-1}^-), y \in [0,1) \}
1242: \hspace*{1cm} \\[.0cm]
1243: && \cup
1244: \bigcup_{k=1}^{ \infty} \{(x,y) ~|~ x \in [\xi_k^+, \xi_{k-1}^+), y \in [0,1) \}.
1245: \end{eqnarray}
1246: See Fig.~{\ref{fig:NHBmap}}(I). Each piece of this partition
1247: $\{ (x,y) ~|~ x \in [\xi_k^{\pm}, \xi_{k-1}^{\pm}), y \in [0,1)\}$
1248: is uniformly stretched in the horizontal direction, and uniformly
1249: squeezed in the vertical direction. The left pieces
1250: $\{ (x,y) ~|~ x \in [\xi_k^{-}, \xi_{k-1}^{-}), y \in [0,1) \}$
1251: are mapped to the bottom part of the unit cell and the right pieces
1252: $\{ (x,y) ~|~ x \in [\xi_k^{+}, \xi_{k-1}^{+}), y \in [0,1) \}$
1253: to the upper (Fig.~{\ref{fig:NHBmap}}(II)).
1254:
1255: %\end{widetext}
1256:
1257: Note that although our theoretical and numerical results are for the
1258: 1-dimensional map $\phi(x)$, these results are also true for this
1259: area-preserving map if an observable does not depend on the vertical
1260: coordinate $y$, namely $A(x,y) = A(x)$, because the horizontal
1261: coordinate $x$ of this area-preserving map is transformed by $\phi(x)$
1262: and independent of $y$. Therefore this area-preserving extension of
1263: $\phi(x)$ has also long time correlations with power law decay. This
1264: fact contrasts to the results for the baker transformation, which
1265: exhibits exponential decay of correlation functions.
1266:
1267: Here let us define some terms for later discussions. We define the
1268: $k$-th escape domain ${\mathcal D}_{k}$ as
1269: \begin{eqnarray}
1270: {\mathcal D}_{k}
1271: = \{ (x,y) ~|~ x \in [\xi_{k-1}^{-}, \xi_{k-2}^{-}), y \in [0,1)\}
1272: \end{eqnarray}
1273: for $k=1,2,\cdots$. For the points in $k$-th escape domain
1274: ${\mathcal D}_{k}$, it takes $(k-1)$ times of the mappings $\psi^n$ to
1275: escape from the left part $x<b$ to the right $x>b$.
1276: %
1277: We also define the $k$-th injection domain
1278: ${\mathcal D}_{k}^{\rm in}$ as
1279: \begin{eqnarray}
1280: {\mathcal D}_{k}^{\rm in}
1281: = \{ \psi(x,y) ~|~ x \in [\xi_k^{+}, \xi_{k-1}^{+}), y \in [0,1) \}
1282: \end{eqnarray}
1283: for $k=1,2,\cdots$. The injection domains are the upper part of the
1284: unit square, which are displayed in Fig.~{\ref{fig:NHBmap}}(II) by the
1285: labels $\{A, B, C, \cdots\}$. We denote the areas, namely the Lebesgue
1286: measures, of the $k$-th escape and injection domains as $S_k$ and
1287: $S_k^{\rm in}$, respectively. These areas obey scaling laws
1288: $S_k \sim 1/k^{\beta}$ and $S_k^{\rm in} \sim 1/k^{\beta+1}$.
1289: This latter power law gives the escape time distribution \cite{tm4}.
1290:
1291: \section{\label{sec:summary}Summary and remarks}
1292:
1293: In this paper, we have introduced a piecewise linear map and analyzed
1294: its spectral properties. We have derived the generalized eigenfunctions
1295: and eigenvalues explicitly for classes of observables and piecewise
1296: constant initial densities. Our model is a modified version of the map
1297: analyzed in Ref.~{\cite{tasaki1}}. A main difference of these two
1298: models is the normalizability of invariant densities. The invariant
1299: density of the model investigated in Ref.~{\cite{tasaki1}} is not
1300: normalizable for a parameter region. This is a typical property of
1301: dynamical systems with marginal fixed points
1302: \cite{tasaki1,aizawa,aaronson}, and is caused by divergence of invariant
1303: density at marginal fixed points.
1304:
1305: On the other hand, the uniform density is invariant for the map
1306: $\phi(x)$ discussed in the present paper;
1307: therefore the invariant density is normalizable for any values of the
1308: system parameters, even though our system has also a marginal fixed
1309: point. This is because the present model has the mechanism suppressing
1310: injections of the orbits into neighborhoods of the marginal fixed
1311: point and this property prevents divergences of the invariant density
1312: at the marginal fixed point. As a consequence of the normalizability,
1313: the present model does not exhibit non-stationarity, which is
1314: generically observed in maps with marginal fixed points
1315: \cite{tasaki1,aizawa,aaronson}.
1316:
1317: The spectral properties of the present model is similar to those of
1318: Ref.~{\cite{tasaki1}} in the locations of the discrete and the
1319: continuous spectra. There are two simple eigenvalues $1$ and
1320: $\lambda_d \in (-1,~0)$; the former corresponds to the invariant
1321: eigenstate and the latter to the oscillating one. The eigenstate
1322: associated to $\lambda_d$, however, does not contribute to the long time
1323: behaviors of the correlation functions because it decays exponentially
1324: fast. There is also the continuous spectrum on the real interval
1325: $[0,1]$; this continuous spectrum leads to power law decay of
1326: correlation functions. We have confirmed a good agreement between the
1327: theoretical prediction and the numerical result for scaling behaviors of
1328: the PSD $S(f) \sim 1/ f^{\gamma}$.
1329:
1330: Furthermore, the piecewise linear map $\phi(x)$ has been extended to an
1331: area-preserving invertible map on the unit square. In contrast to
1332: the baker transformation, which is hyperbolic and shows exponential
1333: decay of correlation functions, our model is non-hyperbolic and displays
1334: power law decay of correlations.
1335:
1336: As is well known, the mixed type Hamiltonian systems often exhibit power
1337: law decay of correlation functions. The area-preserving map $\psi(x,y)$
1338: introduced in this paper may be considered as an abstract model of the
1339: mixed type Hamiltonian systems in the following sense. Instabilities of
1340: the orbits of the map $\psi(x,y)$ [Eq.(\ref{eqn:2d-model})] is weak in
1341: neighborhoods of the line $x=0$, and the escape time from the left part
1342: $x<b$ to the right part $x > b$ diverges as $x \to 0$. In
1343: other words, the orbits stick to the line $x=0$ for long times.
1344: %Furthermore, injection of the orbits from the right part $x>b$ to the
1345: %vicinity of the line $x=0$ is suppressed; in other words, a region is
1346: %more sticky, the injection into there is lesser.
1347: This property seems to be similar to dynamics of Hamiltonian
1348: systems near torus, cantorus, and marginally unstable periodic orbits,
1349: where chaotic orbits stick for long times.
1350:
1351: And, in fact, similar dynamics is observed in a Poincar\'e map of the
1352: mushroom billiard \cite{tm4}, which has been proposed recently as a
1353: model of mixed type systems with sharply divided phase spaces
1354: \cite{bunimovich,bunimovich2,altmann,altmann2,shudo}. In
1355: Ref.~\cite{tm4}, it is found that an infinite partition can be
1356: constructed on a Poincar\'e surface using escape times from
1357: neighborhoods of the outermost tori, and that the area of the escape
1358: domains and the injection domains obey the scaling relations
1359: ${\mathcal D}_{k} \sim 1/k^2$ and
1360: ${\mathcal D}_{k}^{\rm in} \sim 1/k^3$, respectively. These relations
1361: correspond to the case $\beta = 2$ of the present model. Note that a
1362: correlation function of the Poincar\'e map of the mushroom billiard
1363: exhibits power law decay $C(n) \sim 1/n$ \cite{tm4}, and this is
1364: consistent with the analytical result of the present paper. This
1365: relation between the map $\psi(x,y)$ and a billiard system is similar to
1366: that of the baker map and the Lorentz gas \cite{tel}.
1367:
1368: Since the map $\psi(x,y)$ is an elementary model of conservative systems
1369: and can be treated analytically to some extent, this system may be
1370: important for understanding relationships between non-equilibrium
1371: phenomena, such as relaxation and transport, and underlying reversible
1372: dynamics; this is a fundamental problem in dynamical system theory and
1373: statistical mechanics
1374: \cite{gaspard2,dorfman}.
1375:
1376: %Since the microscopic instability of reversible Hamiltonian systems
1377: %causes the macroscopic irreversibility such as relaxation and
1378: %transport\cite{gaspard,dorfman}, it is important to characterize the
1379: %instability of microscopic chaos and its influence to the macroscopic
1380: %phenomenon and the model discussed in the present paper might be
1381: %considered as an elementary model of generic Hamiltonian systems.
1382:
1383: \begin{acknowledgments}
1384: The authors would like to thank Prof.~S.~Tasaki for helpful suggestions
1385: and discussions, and Prof.~A.~Shudo for valuable discussions and
1386: comments. This work is supported in part by Waseda University Grant for
1387: Special Research Projects (The Individual Research No.~2005B-243) from
1388: Waseda University.
1389: \end{acknowledgments}
1390:
1391: \appendix*
1392: \section{\label{sec:app}Asymptotic Behaviors}
1393:
1394: \subsection{When \boldmath $|z| \to 0$}
1395: In this appendix, we show the inequality,
1396: \begin{eqnarray}
1397: \left| \integ_0^{\infty} ds
1398: \f {s^{\beta -1} e^{-s}}{z - e^{-s}} \right|
1399: < C
1400: \left(
1401: \log \f 1{|z|}
1402: \right)^{\beta}
1403: \label{app:lem}
1404: \end{eqnarray}
1405: as $|z| \to 0$ and ${\rm arg}z \in [0,2\pi]$ is fixed, where $C$ is a
1406: positive constant. From this inequality, the Eqs.~(\ref{eqn:xito0}) and
1407: (\ref{eqn:zto0}) can be derived. Let $z = x+ iy$ in this subsection.
1408:
1409: First, let us assume $\arg z \in [3 \pi/4,~5\pi/4]$. Splitting the
1410: integral into two pieces, we have
1411: \begin{eqnarray} \nonumber
1412: \hspace*{-.7cm}\integ_0^{\infty} ds
1413: \f {s^{\beta -1} e^{-s}}
1414: {\left| z - e^{-s} \right|} &
1415: \\[.15cm]
1416: %\nonumber
1417: %&\hspace*{-2.7cm} = \integ_0^{\infty} ds
1418: % \f {s^{\beta -1} e^{-s}}{\s{ ({\rm Re} z - e^{-s})^2 + ({\rm Im} z)^2 }}
1419: %\\[.35cm]
1420: &\hspace*{-1.2cm} \leq \integ_0^{\log \frac 1{|x|}} ds s^{\beta-1}
1421: +\integ_{\log \frac 1{|x|}}^{\infty} ds
1422: \f {s^{\beta -1} e^{-s}} {|x|}.
1423: \label{eqn:asymp1}
1424: \end{eqnarray}
1425: Apparently, the first term of the rhs of Eq.~(\ref{eqn:asymp1}) has an
1426: upper bound $C_1\left( - \log {|z|}\right)^{\beta}$ for some positive
1427: constant $C_1$.
1428: %
1429: On the other hand, the second term %of the rhs of Eq.~(\ref{eqn:asymp1})
1430: has an upper bound $C'_1\left( -\log {|z|}\right)^{\beta-1}$. This is
1431: obtained by using an asymptotic expansion of the incomplete gamma
1432: function (see e.g., Ref.~\cite{henrici}). Thus, in this case the
1433: inequality Eq.~(\ref{app:lem}) holds.
1434:
1435: Second, we consider the case for $\arg z \in [\pi/4,~3 \pi /4]$ or
1436: $\arg z \in [5\pi/4,~7 \pi /4]$. Using similar calculations,
1437: we have
1438: \begin{eqnarray} \nonumber
1439: \integ_0^{\infty} ds
1440: \f {s^{\beta -1} e^{-s}}
1441: {\left| z - e^{-s} \right|} &\leq&
1442: \left( \log \f 1{|y|} \right)^{\beta}
1443: \integ_0^{1} ds~
1444: \f 1{\sqrt {H(s)} } \hspace*{1.5cm}\\[.3cm]
1445: &&\hspace*{1.2cm}+
1446: \integ_{\log \frac 1{|y|}}^{\infty} ds~
1447: \f {s^{\beta -1 }e^{-s}}{|y|},
1448: \end{eqnarray}
1449: where we define $H(s)$ as
1450: $
1451: H(s) \equiv
1452: \left( {x}/{|y|^s} - 1 \right)^2 + |y|^{2(1-s)}.
1453: $
1454: It is easy to check that $H(s) \geq 1/2$. Therefor, the first term has
1455: an upper bound $C_2 \left( -\log {|z|} \right)^{\beta}$.
1456: By using the asymptotic expansion of the incomplete Gamma function, we
1457: have an upper bound for the second term:
1458: $C'_2 \left( -\log {|z|} \right)^{\beta-1}$.
1459: Thus, this case also satisfies the inequality Eq.~(\ref{app:lem}).
1460:
1461: Finally, we consider the case $\arg z \in [0,~\pi /4]$ or
1462: $\arg z \in [7\pi/4,~2\pi]$. We split the integral as
1463: %\begin{widetext}
1464: %\begin{eqnarray}
1465: %\integ_0^{\infty} ds~ \f{s^{\beta-1} e^{-s}}{z - e^{-s}} =
1466: %%
1467: % \integ_{0}^{x/2} dt
1468: %\f {\left( {\log \frac 1t} \right)^{\beta -1} } {z - t}
1469: %+
1470: %\dip
1471: %\int_{0}^{x/2}
1472: %\f{ \left( {\log \frac 1{x - t }} \right)^{\beta -1} }
1473: % { t + i y} -
1474: %\f{ \left( {\log \frac 1{x + t }} \right)^{\beta -1} }
1475: % { t - i y} dt
1476: %%
1477: %+ \integ_{3 x /2}^1 dt
1478: %\f {\left( {\log \frac 1t} \right)^{\beta -1} } {z - t}.
1479: %\label{app:case3-2}
1480: %\end{eqnarray}
1481: %\end{widetext}
1482: \begin{widetext}
1483: \begin{eqnarray}
1484: \integ_0^{\infty} ds~ \f{s^{\beta-1} e^{-s}}{z - e^{-s}} =
1485: %
1486: \integ_{0}^{x/2} dt
1487: \f {G(t)} {z - t}
1488: +
1489: \dip
1490: \int_{0}^{x/2} dt
1491: \left(
1492: \f{ G(x-t) }{ t + i y} -
1493: \f{ G(x+t) }{ t - i y}
1494: \right)
1495: %
1496: + \integ_{3 x /2}^1 dt
1497: \f {G(t)} {z - t},
1498: \label{app:case3-2}
1499: \end{eqnarray}
1500: \end{widetext}
1501: where we define $G(t)$ as $G(t) \equiv \{\log (1/t)\}^{\beta-1}$.
1502: By using similar techniques as above, it can be shown that the
1503: absolute values of the first and the third terms of the rhs of
1504: Eq.~(\ref{app:case3-2}) have an upper bound
1505: $C_3 \left( -\log x\right)^{\beta}$.
1506: %\begin{eqnarray}
1507: %\begin{array}{lll} \dip
1508: %\lim_{\delta \to 0}
1509: %\left\{
1510: %\integ_{\lambda /2}^{\lambda - \delta} +
1511: %\integ_{\lambda + \delta}^{3\lambda /2} \right\}
1512: %\f {\left( {\log \frac 1t} \right)^{\beta -1} } {\lambda - t} dt
1513: %&=& \dip
1514: %\int_{0}^{1/2}
1515: %\f{
1516: %\left\{ {\log \frac 1{\lambda (1 - t) }} \right\}^{\beta -1} -
1517: %\left\{ {\log \frac 1{\lambda (1 + t) }} \right\}^{\beta -1}
1518: %}{t} dt. \\[.7cm]
1519: %\end{array}
1520: %\label{app:case3-3}
1521: %\end{eqnarray}
1522: For the imaginary part of the second term, it is easy to derive
1523: an upper bound of its absolute value:
1524: $C'_3 \left( -\log x \right)^{\beta-1}$.
1525: On the other hand, for the real part, we have an upper bound
1526: $C'_3 \left( -\log x \right)^{\beta-2}$, which can be derived through
1527: integration by parts after changing the variables as $t' = t/x$. This
1528: completes the proof of Eq.~(\ref{app:lem}).
1529:
1530: \subsection{When \boldmath $|z| \to 1$}
1531: The property Eq.~(\ref{eqn:xito1}) is also derived in the same way. We
1532: briefly mention about the derivation. Let us begin with an analytic
1533: continuation of $\Xi (z)$
1534: \begin{eqnarray}\nonumber
1535: \Xi(z) &=& B_1^-(0) \Psi(z) \hspace*{5.55cm}\\[.15cm]
1536: && \hspace*{-.3cm} + \sum_{j=0}^{\infty} \sum_{l=2}^{\infty}
1537: B_l^-(0) l^{\beta} \f {\rho_j}{\Gamma(\beta)}
1538: \int_0^{\infty} ds \f {s^{\beta -1} e^{-(j+l)s}}{z - e^{-s}}.
1539: \label{eqn:app-Xi1}
1540: \end{eqnarray}
1541: We consider the second term. The absolute value of the second term has
1542: an upper bound
1543: \begin{eqnarray}
1544: {\rm const.} \times \left|
1545: \sum_{j=0}^{\infty} \rho_j
1546: \int_0^{\infty} ds \f {s^{\beta -1} e^{-(j+2)s}}{(z - e^{-s})(1 - e^{-s})}
1547: \right|.
1548: \label{eqn:app-Xi2}
1549: \end{eqnarray}
1550: Therefore, we show that for $k=1,2,\cdots$, and $1<\beta <2$,
1551: \begin{eqnarray}
1552: \left|
1553: \int_0^{1} ds \f {s^{\beta -1} e^{-ks}}{(z - e^{-s})(1 - e^{-s})}
1554: \right|
1555: < C' |1-z|^{\beta -2},
1556: \label{app:lem2}
1557: \end{eqnarray}
1558: as $z \to 1$ and $\arg (z-1) \in [0, 2\pi)$ with a constant
1559: $C'$. Note that the integral in Eq.~(\ref{eqn:app-Xi2}) from $1$ to
1560: $\infty$ is convergent. For $\beta \geq 2$, it can be analyzed
1561: in a similar way but the rhs of Eq.~(\ref{app:lem2}) should be changed
1562: to a constant ($\beta>2$) or a log correction $-\log |1-z|$
1563: ($\beta=2$). Let $z = 1 + x + iy$ in the following.
1564:
1565: First, let us assume $\arg z \in [\pi/4,~3 \pi /4]$ or
1566: $\arg z \in [5\pi/4,~7 \pi /4]$, then we have
1567: \begin{eqnarray}
1568: \int_0^{1} ds
1569: \f {s^{\beta -1} e^{-ks}}
1570: {\left|
1571: (z - e^{-s})(1 - e^{-s})
1572: \right|
1573: }
1574: <
1575: e\int_0^1 ds
1576: \f {s^{\beta -2}}
1577: {\left|
1578: z - e^{-s}
1579: \right|
1580: },
1581: \label{app2:case1}
1582: \end{eqnarray}
1583: where we have used $1-e^{-s} \geq s/e$ for $s \in [0,1]$. The rhs of
1584: Eq.~(\ref{app2:case1}) can be estimated, by splitting the integral, as
1585: \begin{eqnarray}\nonumber
1586: \int_0^{1} ds \f {s^{\beta -2} }
1587: {\left|
1588: z - e^{-s}
1589: \right|}
1590: \leq \dip
1591: \int_0^{2e|y|} ds
1592: \f {s^{\beta -2}}{ |y| } +
1593: 2e \int_{2e|y|}^{1} ds {s^{\beta -3} }, \\[-.35cm]
1594: \label{app2:case1-2}
1595: \end{eqnarray}
1596: where we have used $1-e^{-s} \geq s/e$ again. It is obvious that the rhs
1597: of Eq.~(\ref{app2:case1-2}) is less than $C'_4 |1-z|^{\beta-2}$.
1598: Therefore the inequality Eq.~(\ref{app:lem2}) is satisfied in this case.
1599:
1600: Second, the case for $\arg z \in [0,~\pi /4]$ or
1601: $\arg z \in [7\pi/4,~2\pi)$ can be analyzed in the same way as the
1602: first case (But, in this case, split the integral in terms of $x$
1603: instead of $y$.). Thus we omit the detail.
1604:
1605: Finally, when $\arg z \in [3 \pi/4,~5\pi/4]$, we have
1606: \begin{widetext}
1607: \begin{eqnarray}
1608: \integ_0^{1} ds~ \f{s^{\beta-1} e^{-ks}}{(z - e^{-s})(1 -e^{-s})} =
1609: %
1610: \integ_{1/e}^{1-3\epsilon/2} dt
1611: \f {F(t)} {z - t}
1612: +
1613: \dip
1614: \int_{0}^{\epsilon/2} dt
1615: \left(
1616: \f{ F(1-\epsilon-t) }{ t + i y} -
1617: \f{ F(1-\epsilon+t) }{ t - i y}
1618: \right)
1619: %
1620: + \integ_{1 - \epsilon /2}^1 dt
1621: \f {F(t)} {z - t},
1622: \label{app2:case3-1}
1623: \end{eqnarray}
1624: \end{widetext}
1625: where we define $\epsilon > 0$ as $\epsilon= |x|$ and $F(t)$ as
1626: $F(t) \equiv \{\log(1/t)\}^{\beta-1} t^{k-1}/(1-t)$. The first and the
1627: third terms of the rhs can be estimated in the same way as
1628: Eq.~(\ref{app2:case1-2}), and we obtain an upper bound for their absolute
1629: values as $C_5 |1-z|^{\beta-2}$. For the imaginary part of the second
1630: term, we have easily a bound of its absolute value:~
1631: $C'_5 |1-z|^{\beta-2}$. On the other hand,
1632: for the real part, we also have a bound $C''_5 |1-z|^{\beta-2}$ through
1633: integration by parts after changing the variables as
1634: $t' = t/ \epsilon$. Thus we complete the proof of Eq.~(\ref{app:lem2}).
1635:
1636: From Eq.~(\ref{app:lem2}), it can be shown that
1637: $\Psi(z) \to \Psi(1) < \infty$,
1638: as $z \to 1$. Thus the first term of Eq.~(\ref{eqn:app-Xi1}) converges
1639: as $z \to 1$. Consequently, we have Eq.~(\ref{eqn:xito1}).
1640: %It can also be shown from Eq.~(\ref{app:lem2}) that
1641: %$\Omega(z) \to \Omega(1)$ and $\Phi(z) \to \Phi(1)$ as $z \to 1$.
1642: \vspace*{.2cm}
1643: %\twocolumngrid
1644: \bibliography{nhb}
1645:
1646:
1647:
1648: \end{document}
1649: %
1650: % ****** End of file template.aps ******
1651:
1652:
1653: % If in two-column mode, this environment will change to single-column
1654: % format so that long equations can be displayed. Use sparingly.
1655: %\begin{widetext}
1656: % put long equation here
1657: %\end{widetext}
1658:
1659: % Here is an example of the general form of a figure:
1660: % Fill in the caption in the braces of the \caption{} command. Put the label
1661: % that you will use with \ref{} command in the braces of the \label{} command.
1662: % Use the figure* environment if the figure should span across the
1663: % entire page. There is no need to do explicit centering.
1664:
1665: p% \begin{figure}
1666: % \includegraphics{}%
1667: % \caption{\label{}}
1668: % \end{figure}
1669:
1670: % Surround figure environment with turnpage environment for landscape
1671: % figure
1672: % \begin{turnpage}
1673: % \begin{figure}
1674: % \includegraphics{}%
1675: % \caption{\label{}}
1676: % \end{figure}
1677: % \end{turnpage}
1678:
1679: % Specify following sections are appendices. Use \appendix* if there
1680: % only one appendix.
1681: %\appendix
1682: %\section{}
1683:
1684: % If you have acknowledgments, this puts in the proper section head.
1685:
1686: % Create the reference section using BibTeX:
1687:
1688: % tables should appear as floats within the text
1689: %
1690: % Here is an example of the general form of a table:
1691: % Fill in the caption in the braces of the \caption{} command. Put the label
1692: % that you will use with \ref{} command in the braces of the \label{} command.
1693: % Insert the column specifiers (l, r, c, d, etc.) in the empty braces of the
1694: % \begin{tabular}{} command.
1695: % The ruledtabular enviroment adds doubled rules to table and sets a
1696: % reasonable default table settings.
1697: % Use the table* environment to get a full-width table in two-column
1698: % Add \usepackage{longtable} and the longtable (or longtable*}
1699: % environment for nicely formatted long tables. Or use the the [H]
1700: % placement option to break a long table (with less control than
1701: % in longtable).
1702: % \begin{table}%[H] add [H] placement to break table across pages
1703: % \caption{\label{}}
1704: % \begin{ruledtabular}
1705: % \begin{tabular}{}
1706: % Lines of table here ending with \\
1707: % \end{tabular}
1708: % \end{ruledtabular}
1709: % \end{table}
1710:
1711: % Surround table environment with turnpage environment for landscape
1712: % table
1713: % \begin{turnpage}
1714: % \begin{table}
1715: % \caption{\label{}}
1716: % \begin{ruledtabular}
1717: % \begin{tabular}{}
1718: % \end{tabular}
1719: % \end{ruledtabular}
1720: % \end{table}
1721: % \end{turnpage}