1: \documentclass[aps,amsfonts,floatfix]{revtex4}
2:
3: \usepackage{latexsym,amssymb}
4: \usepackage[dvips]{graphics}
5: \usepackage[dvips]{color}
6: \usepackage{graphics,psfrag,graphicx,subfigure,bbm}
7: \usepackage{amsmath,amsthm,amscd}
8:
9: \newcommand{\Rset}{{\mathbb R}}
10: \newcommand{\Tset}{{\mathbb T}}
11: \newcommand{\ie}{{i.e.\ }}
12: \newcommand{\eg}{{e.g.\ }}
13: \newcommand{\bc}{{b.c.\ }}
14:
15: \newcommand{\beq}{\begin{equation}}
16: \newcommand{\eeq}{\end{equation}}
17: \newcommand{\bea}{\begin{eqnarray}}
18: \newcommand{\eea}{\end{eqnarray}}
19: \newcommand{\ba}{\begin{array}}
20: \newcommand{\ea}{\end{array}}
21: \newcommand{\bean}{\begin{eqnarray*}}
22: \newcommand{\eean}{\end{eqnarray*}}
23: \newcommand{\isubfig}[1]{\subfigure{#1}\nonumber}
24: \newcommand{\xyvect}[2]{\left(\begin{matrix} #1 \\ #2 \end{matrix}\right)}
25:
26: \begin{document}
27: \title{Renormalization and destruction of $1/\gamma^2$ tori in the
28: standard nontwist map}
29: \author{ A.~Apte, A.~Wurm and P.J.~Morrison}
30: \affiliation{Department of Physics and Institute for Fusion Studies\\
31: University of Texas at Austin, Austin, Texas 78712}
32: \date{\today}
33: \begin{abstract}
34: Extending the work of del-Castillo-Negrete, Greene, and Morrison,
35: Physica D {\bf 91}, 1 (1996) and {\bf 100}, 311 (1997) on the
36: standard nontwist map, the breakup of an invariant torus
37: with winding number equal to the inverse golden mean squared is
38: studied. Improved numerical techniques provide the greater accuracy
39: that is needed for this case. The new results are interpreted within
40: the renormalization group framework by constructing a renormalization
41: operator on the space of commuting map pairs, and by studying the
42: fixed points of the so constructed operator.
43: \end{abstract}
44: \maketitle
45:
46: {\bf
47: In recent years, area-preserving maps that violate the twist condition
48: locally in phase space have been the object of interest in several
49: studies in physics and mathematics. These \emph{nontwist} maps show
50: up in a variety of physical models. An important problem from the
51: physics point of view is the understanding of the breakup of invariant
52: tori, which show remarkable resilience in the region where the
53: twist condition is violated, called \emph{shearless tori}. In terms of
54: the physical system modelled, these tori represent transport barriers, and their
55: breakup corresponds to the transition to global chaos. Mathematically,
56: nontwist maps present a challenge since the standard proofs of
57: celebrated theorems in the theory of area-preserving maps rely heavily
58: on the twist condition. In this paper, we study the breakup of the
59: shearless torus with winding number \boldmath{$1/\gamma^2$}, where
60: \boldmath{$\gamma$} is the golden mean. This torus serves as a test
61: case for improved techniques we developed. At the point of breakup the
62: shearless torus exhibits universal scaling behavior which leads to a
63: renormalization group interpretation.}
64:
65: \section{Introduction}\label{sec:intro}
66: In this paper we consider the {\it standard nontwist map} (SNM) $M$, as
67: introduced in Ref.~\onlinecite{diego2}:
68: \bea
69: x_{n+1} & = & x_n + a \left(1-y^2_{n+1}\right)\nonumber\\[0.1in]
70: y_{n+1} & = & y_n - b \sin\left(2\pi x_n\right),
71: \label{eq:stntmap}
72: \eea
73: where $(x,y)\in\Tset\times\Rset$, $a\in(0,1)$, and
74: $b\in(-\infty,\infty)$. The map $M$ is {\it area-preserving} and
75: violates the {\it twist condition}
76: \beq
77: \frac{\partial x_{i+1}\left(x_i,y_i\right)}{\partial y_i}
78: \neq 0\qquad\qquad \forall (x_i,y_i),
79: \eeq
80: along a curve in phase space, which has been recently called the
81: {\it nonmonotone curve}.\cite{petrisor1} Traditionally, most studies of
82: area-preserving maps have dealt with the {\it twist} case, but in
83: recent years more and more research has been focused on the nontwist
84: case.
85:
86: Applications of nontwist maps occur in many fields, for example, the
87: study of magnetic field lines in toroidal plasma devices (see \eg
88: Refs.~\onlinecite{horton1,balescu}), in celestial mechanics,\cite{kyner}
89: fluid dynamics\cite{diego2} and atomic physics.\cite{chandre2}
90: It has been shown\cite{dullin,vander} that nontwist regions appear
91: generically in area-preserving maps that have a tripling bifurcation
92: of an elliptic fixed point.
93: In addition to these applications, the map is quite interesting from
94: a mathematical standpoint because many important theorems in the theory
95: of area-preserving maps assume the validity of the twist condition,
96: \eg the KAM theorem and the Poincare-Birkhoff theorem. The SNM can
97: serve as a model for the development of new proofs. Up to now, only a
98: few mathematical results exist for nontwist maps (see
99: \eg Ref.~\onlinecite{delshams,franks,petrisor1,simo}).
100:
101: We continue the work of del-Castillo-Negrete, Greene and
102: Morrison,\cite{diego4,diego5} who studied the breakup of the
103: shearless invariant torus with winding number $1/\gamma$, where
104: $\gamma=(1+\sqrt{5})/2$ is the golden mean. We present the analysis
105: of the breakup of the shearless invariant torus with winding number
106: (in continued fraction representation)
107: \beq
108: \omega=[0,2,1,1,\ldots]=1/\gamma^2.
109: \eeq
110: Because this winding number is a noble number (its continued fraction expansion
111: ends with $[1,1,1,\ldots]$), the behavior of the residues of the
112: approximating periodic orbits is expected to be the same as in the
113: $1/\gamma$ case, \ie we
114: should find the same fixed point of the renormalization group
115: operator with the same unstable eigenvalues that were found in
116: Ref.~\onlinecite{diego5}. But, the form of the renormalization group operator,
117: which is defined later in Sec.~\ref{sec:reng}, is
118: different from the $1/\gamma$ case.
119: Also, the region of parameter space we study is different.
120: Additionally, since the periods of approximating
121: periodic orbits are bigger than those for the $1/\gamma$ case, the
122: present work serves as a test case for improved numerical techniques
123: described later in Sec.~\ref{ssec:num}.
124:
125: A different approach, which yields rough parameter
126: values for the breakup of invariant tori, was used by Shinohara
127: and Aizawa in Ref.~\onlinecite{shin1}, who showed that a shearless invariant
128: torus crosses the $x$-axis at two points,\cite{note1}
129: $x_A = a/2-1/4$ and $x_B = a/2+1/4$.
130: \begin{figure}[!t]
131: \centering
132: \includegraphics[angle=270,width=0.48\textwidth]{shin.ps}
133: \caption{Parameter space around the critical point (marked by $\ast$)
134: of the $\omega=1/\gamma^2$-shearless curve, showing
135: the points for which shearless invariant tori exist.}
136: \label{fig:stntshin2}
137: \end{figure}
138: For a given $(a,b)$, a point on the shearless torus,
139: $(a/2+1/4,0)$, is iterated many times (we used $10^6$). If the $y$ value
140: stays below a threshold (we used $|y|<0.52$), it is assumed that the
141: shearless curve exists and the point is plotted. Figure
142: \ref{fig:stntshin2} depicts our duplication of their procedure.
143:
144: We see that the critical point for the
145: $1/\gamma^2$ shearless curve (indicated by $\ast$) lies on the
146: boundary in Fig.~\ref{fig:stntshin2}. Thus,
147: the boundary points of Fig.~\ref{fig:stntshin2} represent the {\it
148: critical function} for the SNM. This is a generalization of
149: the definition of the critical function for the standard twist map
150: (see \eg Ref.~\onlinecite{stark}), which has only one parameter \eg $k$. The
151: critical function in the twist case is then defined as
152: $k_c(\omega)$. Here, we have two parameters, but the shearless
153: invariant torus of a given winding number $\omega$ exists only for
154: parameter values belonging to a curve
155: $\left(a,b\left(a;\omega\right)\right)$ in the parameter space. Thus,
156: we can define the critical function by the critical points on each of
157: those curves by $\left(a_c,b\left(a_c;\omega\right)\right)$. By
158: finding the critical points for many other winding numbers (both
159: nobles and non-nobles), we hope to find a more accurate critical
160: function curve than the one shown in Fig.~\ref{fig:stntshin2}.
161:
162: In Sec.~\ref{sec:review}, we review some basic properties of the
163: SNM. The detailed breakup of the shearless invariant torus with
164: winding number $1/\gamma^2$ is presented in Sec.~\ref{sec:neww},
165: which also contains a discussion of the numerical procedures
166: involved. In Sec.~\ref{sec:reng}, we interpret the results within
167: the framework of the renormalization group. Section \ref{sec:concl}
168: a summary and some directions of future research.
169:
170: \section{Review of residue criterion and standard nontwist map}
171: \label{sec:review}
172: In this section, we give a brief review of some basic concepts of the theory of
173: area-preserving maps in the context of the SNM. For a more in-depth
174: discussion the reader is referred to Ref.~\onlinecite{diego4} and
175: references therein.
176:
177: \subsection{Periodic orbits and residue criterion}\label{ssec:po}
178: Since the pioneering work of Greene,\cite{greene2} periodic orbits
179: have proven to be very useful for studying the breakup of invariant
180: tori in area-preserving maps. Below are some standard definitions.
181:
182: An {\it orbit} of an area-preserving map $M$ is a sequence of points
183: $\left\{\left(x_i,y_i\right)\right\}$ such that $M\left(x_i,y_i\right)
184: = \left(x_{i+1},y_{i+1}\right)$.
185: The {\it winding number} $\omega$ of an orbit is defined as
186: $\omega = \lim_{i\to\infty} (x_i/i)$
187: when the limit exists. Here the $x$-coordinate is ``lifted'' from
188: $\Tset$ to $\Rset$.
189: A {\it periodic orbit} of period $n$ is a sequence of $n$ points
190: $\left\{\left(x_i,y_i\right)\right\}_{i=1}^n$, such that
191: $M^n \left( x_i, y_i\right) = \left( x_i+m, y_i\right)$
192: for all $i=1,\ldots,n$,
193: and $m$ is an integer. Periodic orbits have rational winding numbers
194: $\omega=m/n$.
195: An {\it invariant torus} is an orbit with irrational winding
196: number that covers densely a one-dimensional set in phase space. Of
197: particular importance are the invariant tori that wind around the
198: $x$-domain because, in two-dimensional maps, they act as transport
199: barriers.
200:
201: The linear stability of a periodic orbit is determined by the
202: value of its {\it residue},\cite{greene2} $R$, which is defined as
203: $R:=\left[2-\mbox{Tr}(L)\right]/4$. Here, $L$ is the map $M^n$
204: linearized about the periodic orbit of interest and Tr denotes the
205: trace. If $0<R<1$, the orbit is stable or elliptic; if $R<0$ or $R>1$,
206: it is unstable or hyperbolic; in the degenerate cases $R=0$ and $R=1$,
207: it is parabolic.
208:
209: Periodic orbits can be used to systematically approximate invariant
210: tori.\cite{greene2} The method is based on the observation that given
211: a sequence of rational numbers $\left\{m_i/n_i\right\}$ whose limit is
212: $\omega$, the sequence of periodic orbits with winding numbers
213: $\left\{m_i/n_i\right\}$ approaches the invariant torus with winding
214: number $\omega$ in phase space. It is important to find the ``best''
215: possible sequence, \ie the sequence that converges to $\omega$ the
216: fastest. The elements of the best possible sequence (see \eg
217: Ref.~\onlinecite{khinchin}) are the
218: convergents that are obtained from successive truncations of the
219: continued fraction expansion of $\omega$.
220:
221: The {\it residue criterion\/}\cite{greene2} can be stated as follows:
222: Consider an invariant torus with winding number $\omega$. Let $\{m_i/n_i\}$
223: be the sequence of convergents approximating $\omega$, and $R_i$ the
224: residues of their corresponding periodic orbits.
225: \begin{enumerate}
226: \item If $\lim_{i\to\infty} |R_i| = 0$, the invariant torus exists.
227: \item If $\lim_{i\to\infty} |R_i| =\infty$, the invariant torus is
228: destroyed.
229: \item At the boundary in parameter space between those two limits, the
230: invariant torus is at the threshold of destruction and the residues either
231: converge to a constant, non-zero value, or there are convergent subsequences.
232: \end{enumerate}
233:
234: This criterion is based on the idea that the destruction of an invariant
235: torus is caused by the de-stabilization of nearby periodic orbits.
236: The residue criterion has been used successfully in many cases to
237: predict with high precision the threshold for the destruction of
238: invariant tori. Several theorems have been proved
239: that lend mathematical support to the criterion.\cite{falc,mackay2}
240:
241: The numerical search for periodic orbits is difficult because, in principle,
242: it is a two-dimensional root finding problem. However, the task is
243: considerably simplified for {\it reversible
244: maps},\cite{greene2,devogel} which are maps that can be factored
245: as $M = I_1\circ I_0$,
246: where $I_{0,1}$ are {\it involution} maps that satisfy $I_1^2 =I_0^2 =1$.
247: The sets of fixed points of the involution maps,
248: $\Gamma_{0,1} = \left\{ (x,y)| I_{0,1} (x,y) = (x,y)\right\}$,
249: are one-dimensional sets, called {\it symmetry lines} of the map. Once
250: we know $\Gamma_{0,1}$, the search for periodic orbits is reduced to
251: a one-dimensional root finding problem, as explained below in
252: Sec.~\ref{ssec:spo}.
253:
254: \subsection{Standard nontwist map}\label{ssec:std_nt}
255:
256: The SNM is reversible. The symmetry lines $\Gamma_0$, composed of
257: fixed points of $I_0$ are $s_1 = \left\{(x,y) |x=0\right\}$ and
258: $s_2 = \left\{(x,y) |x=1/2\right\}$.
259: The symmetry lines $\Gamma_1$, composed of fixed points of $I_1$ are
260: $s_3 = \left\{(x,y) |x=a\left(1-y^2\right)/2\right\}$ and
261: $s_4 = \left\{(x,y) |x=a\left(1-y^2\right)/2+1/2\right\}$.
262:
263: A major difference between the standard nontwist map and twist
264: maps is that there are two periodic orbits, if they exist, with the
265: same winding number on each symmetry line. This can be seen easily
266: in the integrable case. For $b=0$, the $m/n$ periodic orbits on the
267: $s_1$ symmetry line are located at
268: \beq
269: (x,y) = \left( 0,\pm\sqrt{1-(m/n)/a}\right).
270: \label{eq:bzero}
271: \eeq
272: We will call the orbit with the bigger (smaller) $y$-coordinate the
273: {\it up} ({\it down}) periodic orbit.
274:
275: The SNM is also invariant with respect to the transformation
276: \beq
277: T\left(x,y\right) = \left(x+1/2,\,-y\right).
278: \label{eq:tdef}
279: \eeq
280: The coordinates of the up and down periodic orbits on the symmetry
281: lines $s_i$, denoted by $(x_{ui},y_{ui})$ and $(x_{di},y_{di})$
282: respectively, are related by this symmetry as follows:
283: \bea
284: (x_{d2},y_{d2}) = T\left((x_{u1},y_{u1})\right), \qquad \qquad
285: (x_{u2},y_{u2}) = T\left((x_{d1},y_{d1})\right), \nonumber\\
286: (x_{d4},y_{d4}) = T\left((x_{u3},y_{u3})\right), \qquad \qquad
287: (x_{u4},y_{u4}) = T\left((x_{d3},y_{d3})\right).
288: \label{eq:tmap}
289: \eea
290: Therefore, it is actually enough to compute periodic orbits on
291: $s_1$ and $s_3$, since the orbits along the other symmetry lines
292: can be obtained from (\ref{eq:tmap}).
293:
294: \subsection{Periodic orbit collisions and bifurcation curves}
295: \label{ssec:snmpocoll}
296:
297: Periodic orbits in the SNM can undergo a particular
298: form of bifurcation that
299: occurs when the up and down periodic orbits of the same winding number
300: meet on the symmetry line. These collisions were detected numerically
301: in Refs.~\onlinecite{stix,howard} and \onlinecite{diego4}.
302: Further studies of this bifurcation can be found in
303: Refs.~\onlinecite{petrisor1,simo}.
304:
305: From (\ref{eq:bzero}) it follows that, for a given $a$, only periodic
306: orbits with $m/n < a$ exist at $b=0$. As the value of $b$ increases,
307: the up and down orbits approach each other and at the bifurcation
308: value, they collide and annihilate each other. For higher values of $b$, both
309: orbits no longer exist. Figure \ref{fig:ybcurve} illustrates the
310: behavior of periodic orbits as we increase $b$ from $b=0$. Here the
311: $y$-coordinates of the $m/n = 3/8$ periodic orbits on $s_1$ is shown as
312: a function of $b$ for the fixed value of $a=0.4$.
313: \begin{figure}[ht]
314: \centering
315: \includegraphics[angle=270,width=0.48\textwidth]{yb38pocoll.ps}
316: \caption{Plot of behavior of the up and down periodic
317: orbit of winding number $3/8$ for increasing $b$-values at $a=0.4$. The
318: vertical axis shows the $y$-coordinates of the orbits along $s_1$.}
319: \label{fig:ybcurve}
320: \end{figure}
321: \begin{figure}[ht]
322: \center
323: \includegraphics[angle=270,width=.48\textwidth]{bifurcation-curves.ps}
324: \caption{\label{fig:bifurcurves} Bifurcation curves for several
325: convergents of $1/\gamma^2$.}
326: \end{figure}
327:
328: Based on these numerical observations, the notion of a bifurcation
329: curve in parameter space was defined in Ref.~\onlinecite{diego4}.
330: The {\it $m/n$-bifurcation curve
331: $b=\Phi_{m/n}(a)$} is the set of $(a,b)$ values for which the $m/n$ up
332: and down periodic orbits are at the point of collision.
333: The main property of this curve is that for $(a,b)$ values below
334: $b=\Phi_{m/n}(a)$, the $r/s$ periodic orbits, with $r/s < m/n$ exist.
335: Thus, $m/n$ is the maximum winding number for parameter values along
336: the $m/n$-bifurcation curve.
337:
338: The idea of approximating invariant tori with irrational winding
339: numbers by periodic orbits is used to define the bifurcation curve for an
340: invariant torus as follows:\cite{diego4}
341: The {\it $\omega$-bifurcation curve $b=\Phi_\omega(a)$} for an
342: irrational $\omega$ is the set of $(a,b)$ values such that
343: $b=\Phi_\omega(a)=\lim_{i\to\infty} \Phi_{m_i/n_i}(a)$, where
344: $\Phi_{m_i/n_i}(a)$ is the $m_i/n_i$-bifurcation curve and
345: $\{m_i/n_i\}$ are the convergents of $\omega$.
346: For $(a,b)$ points along the $\omega$-bifurcation curve the invariant
347: torus with irrational winding number $\omega$ is the curve of maximum
348: winding number and is called {\it shearless}.
349: Figure \ref{fig:bifurcurves} depicts the bifurcation curves for
350: several convergents of $1/\gamma^2$. This figure also makes it
351: plausible that the limit in the above definition exists.
352:
353: \section{Breakup of Torus with $\omega=1/\gamma^2$}\label{sec:neww}
354:
355: In this section, we present the analysis of the breakup of the
356: shearless invariant torus in the standard nontwist map with winding
357: number $\omega=1/\gamma^2$. Tables \ref{tab:conv} and \ref{tab:conv2}
358: list the convergents used for these calculations. For more details
359: see Ref.~\onlinecite{wurm}.
360:
361: \begin{table}[ht]
362: \begin{center}
363: \begin{tabular}{|rl||rl|}\hline
364: &&&\\[-2mm]
365: $[i]$ & $F_i/F_{i+2}$ & $[i]$ & $F_i/F_{i+2}$\\[2mm]
366: \hline
367: $[1]$ & 1/3 & $[19]$ & 6765/17711\\
368: $[3]$ & 3/8 & $[21]$ & 17711/46368\\
369: $[5]$ & 8/21 & $[23]$ & 46368/121393\\
370: $[7]$ & 21/55 & $[25]$ & 121393/317811\\
371: $[9]$ & 55/144 & $[27]$ & 317811/832040\\
372: $[11]$ & 144/377 & $[29]$ & 832040/2178309\\
373: $[13]$ & 377/987 & $[31]$ & 2178309/5702887\\
374: $[15]$ & 987/2584 & $[33]$ & 5702887/14930352\\
375: $[17]$ & 2584/6765 & $[35]$ & 14930352/39088169 \\
376: \hline
377: \end{tabular}
378: \caption{\label{tab:conv}Some of the convergents of
379: $\omega=[0,2,1,1,\ldots]$ for which the periodic orbits still
380: exist at criticality.}
381: \end{center}
382: \end{table}
383:
384: \begin{table}[ht]
385: \begin{center}
386: \begin{tabular}{|rl||rl|}\hline
387: &&&\\[-2mm]
388: $[i]$ & $F_i/F_{i+2}$ & $[i]$ & $F_i/F_{i+2}$\\[2mm]
389: \hline
390: $[2]$ & 2/5 & $[18]$ & 4181/10946\\
391: $[4]$ & 5/13 & $[20]$ & 10946/28657\\
392: $[6]$ & 13/34 & $[22]$ & 28657/75025\\
393: $[8]$ & 34/89 & $[24]$ & 75025/196418\\
394: $[10]$ & 89/233 & $[26]$ & 196418/514229\\
395: $[12]$ & 233/610 & $[28]$ & 514229/1346269\\
396: $[14]$ & 610/1597 & $[30]$ & 1346269/3524578\\
397: $[16]$ & 1597/4181 & $[32]$ & 3524578/9227465\\
398: \hline
399: \end{tabular}
400: \caption{\label{tab:conv2}Some of the convergents of
401: $\omega=[0,2,1,1,\ldots]$ for which the periodic orbits do not exist
402: at criticality.}
403: \end{center}
404: \end{table}
405:
406: \subsection{Numerical methods}\label{ssec:num}
407:
408: The computational steps necessary to find the critical point
409: and the residue behavior of the approximating periodic
410: orbits are as follows:
411: \begin{enumerate}
412: \item Find a good approximation to the $1/\gamma^2$-bifurcation
413: curve in $(a,b)$-space using the bifurcation curves for its
414: convergents.
415: \item Along this bifurcation curve, find the up and down
416: periodic orbits on the symmetry line $s_1$ that approximate
417: the invariant torus, and compute their residues.
418: \item Locate the $(a,b)$ point along the curve at which
419: the residues exhibit critical behavior.
420: \item Find the residues of the periodic orbits at
421: criticality along the remaining symmetry lines.
422: \item Find the eigenvalues of the unstable eigenmodes
423: of the renormalization group operator. The
424: details of how to do this depend crucially on the type of critical
425: scaling behavior that is exhibited by the residues.
426: \end{enumerate}
427:
428: \subsubsection{Searching for periodic orbits}\label{ssec:spo}
429:
430: Periodic orbits on the symmetry lines can be be computed relatively
431: easily for reversible maps using the following result:\cite{diego4}
432: If $(x,y) \in \Gamma_{0,1}$ then $M^n(x,y)=(x,y)$ if and only if
433: $M^{n/2}(x,y) \in \Gamma_{0,1}$ (for $n$ even) or $M^{(n\pm 1)/2}(x,y) \in
434: \Gamma_{1,0}$ (for $n$ odd).
435: Thus, for example, periodic orbits with odd period $n$
436: on the $s_1$ symmetry line can be obtained by looking for points
437: on $s_1$ that are mapped to $s_3$ or $s_4$ after
438: $(n+1)/2$ iterations. This can be implemented as a one-dimensional root
439: finding problem by considering the zeros of the function
440: $F(y) = \sin \left[ 2\pi\left(\hat{x}-a\left(1-\hat{y}^2\right)/2\right)
441: \right]$, where
442: $\left( \hat{x},\hat{y}\right):= M^{(n+1)/2}(0,y)$.
443: The sine function is used to eliminate the difference between $s_3$ and
444: $s_4$. Similar ideas can be applied to find other orbits.
445:
446: \subsubsection{Finding $m/n$-bifurcation curves}\label{sssec:fbif}
447:
448: Recall that the bifurcation curve $\Phi_{m/n}(a)$ of a periodic orbit of
449: winding number $m/n$ was defined in Sec.~\ref{ssec:snmpocoll} to be
450: the set of points $(a,b)$, at which the up and down periodic orbits of
451: winding number $m/n$ collide along the $s_1$ symmetry line. Thus, at a
452: given value of $a$, the function $F(y)$ has two
453: roots for $b<\Phi_{m/n}(a)$, no roots (locally) for $b>\Phi_{m/n}(a)$
454: and a single root, which is also an extremum, for $b=\Phi_{m/n}(a)$.
455: We thus search for the zero of the extremum of $F(y)$ as $b$ is varied.
456:
457: To find the whole (or large portions) of a bifurcation curve, we
458: use the monotonic nature of the curve (see Fig.~\ref{fig:bifurcurves})
459: as follows: Given a point $(a_1,b_1)$ on the
460: bifurcation curve \ie $b_1=\Phi_{m/n}(a_1)$, we increase $a$ by a fixed
461: amount to $a_2=a_1+a_{\mbox{\scriptsize step}}$. We then start at the point
462: $(a_2,b_1)$ and increase $b$ until we reach $b_2=\Phi_{m/n}(a_2)$.
463: To make sure that we are finding the correct bifurcation curve, we
464: start searching $(a,b)$-space at $(a,b)=(m/n,0)$. Even then,
465: the steps in $a$ cannot be taken to be too large. Experience has shown that
466: steps in $a$ of $1\times 10^{-5}$ or $1\times 10^{-6}$ are safe.
467: This method is unfortunately very slow because the part of the curve
468: at small $b$ values is very steep and the interesting (near critical)
469: part of the curve is far away from the $b=0$ limit.
470:
471: We managed to drastically improve the speed of these calculations by
472: using the following ideas:
473: \begin{enumerate}
474: \item Numerical evidence strongly suggests that a bifurcation curve
475: is smooth and monotonically increasing, although it is not proved
476: mathematically.\cite{note2} So we
477: use linear extrapolation from two previous points to find the new
478: value of $b$ around which to search for the bifurcation point. It was
479: found that any higher order extrapolation did not improve the
480: algorithm further.
481: \item To find bifurcation curves for periodic orbits
482: with very large periods (\eg of the order of several
483: million) the following procedure is used: Starting at the bifurcation
484: curve of a smaller period, we increase $b$ until the bifurcation curve
485: of the higher period is reached. The advantage of this procedure is
486: that we do not need to do the extremely time consuming calculations of
487: the bifurcation curves for very high period orbits starting at $b=0$,
488: but rather we can search for them near the region of interest.
489: \end{enumerate}
490:
491: \subsubsection{Finding $1/\gamma^2$-bifurcation curve}\label{sssec:wbif}
492:
493: Recall that the $1/\gamma^2$-bifurcation curve was defined as the limit of
494: $m_i/n_i$ bifurcation curves, where $m_i/n_i$ are convergents of
495: $1/\gamma^2$. It was numerically observed that close to criticality,
496: this limit is approached in accordance with the following
497: scaling relation:\cite{diego4}
498: \beq
499: \Phi_{[n+1]}(a) = \Phi_{1/\gamma^2}(a) + B_n(a)\; \nu_1^{n/12},
500: \label{eq:scalrel}
501: \eeq
502: where the $\Phi_{[n]}(a)$ denotes the bifurcation curve of the
503: periodic orbit with winding number $[n]=F_n/F_{n+2}$, $\nu_1$ is a
504: number to be determined later, and $B_n(a)$ is a period-twelve function,
505: \ie $B_{n+12}(a)=B_n(a)$.
506:
507: If Eq.~(\ref{eq:scalrel}) holds, it follows that for fixed $a$
508: \beq
509: \Phi_{1/\gamma^2} = \lim_{n\to\infty} \frac{\Phi_{[n+1]} \Phi_{[n+12]}-
510: \Phi_{[n]} \Phi_{[n+13]}}{\left(\Phi_{[n+1]}-\Phi_{[n]}\right)-
511: \left(\Phi_{[n+13]}- \Phi_{[n+12]}\right)}.
512: \label{eq:scalrel2}
513: \eeq
514: We obtained the $1/\gamma^2$-bifurcation curve using $n=19$ in
515: Eq.~(\ref{eq:scalrel2}), \ie using the bifurcation curves for $[32]$,
516: $[31]$, $[20]$ and $[19]$ (see Tables \ref{tab:conv} and
517: \ref{tab:conv2}).
518:
519: Now one can justify {\it a posteriori} the use of Eq.~(\ref{eq:scalrel}).
520: Solving Eq.~(\ref{eq:scalrel}) with $a=a_c$ for $\nu_1$ yields:
521: \beq
522: \nu_1 = \lim_{n\to\infty}
523: \left(\frac{\Phi_{[n+13]}\left(a_c\right)-b_c}
524: {\Phi_{[n+1]}\left(a_c\right)-b_c}\right),
525: \eeq
526: and
527: \beq
528: B_n(a_c) = \left( \Phi_{[n+1]}\left(a_c\right)-b_c\right)\; \nu_1^{-n/12},
529: \eeq
530: where $(a_c,b_c)$ is the critical point for breakup of the shearless
531: $1/\gamma^2$ invariant torus \ie $b_c=\Phi_{1/\gamma^2}(a_c)$.
532: We found that $\nu_1^{-1/12}=2.678$. Some numerical
533: evidence for the periodicity of $B_n(a_c)$ is given in
534: Table \ref{tab:bn}.
535:
536: \begin{table}[ht]
537: \begin{center}
538: \begin{tabular}{|rr|rr|}\hline
539: &&&\\[-2mm]
540: $n$ &$B_n(a_c)$ & $n$ & $B_n(a_c)$\\[2mm]
541: \hline
542: $15$ & -0.4865 & $27$ & -0.4865\\
543: $17$ & -0.7090 & $29$ & -0.7078\\
544: $18$ & 0.5019 & $20$ & 0.5028\\
545: $19$ & -0.3901 & $31$ & -0.3887\\
546: \hline
547: \end{tabular}
548: \caption{\label{tab:bn} Period-twelve behavior
549: of the scaling function $B_n(a_c)$.}
550: \end{center}
551: \end{table}
552:
553: \subsection{Results}\label{ssec:results}
554:
555: In this subsection, we present the results of our computations.
556:
557: \subsubsection{Residue behavior at criticality}
558:
559: We computed bifurcation curves up to $[32]=3524578/9227465$ and found the
560: critical points along them, \ie the parameter values along those curves
561: for which the residues of approximating periodic orbits neither
562: converge to zero nor diverge to infinity.
563: \begin{figure}[!ht]
564: \centering
565: \isubfig{
566: \includegraphics[angle=270,scale=2,width=.8\textwidth]{uporb2.ps}
567: }
568: \isubfig{
569: \includegraphics[angle=270,scale=2,width=.8\textwidth]{downorb2.ps}
570: }
571: \caption{Residue behavior of the up (top figure) and down (bottom
572: figure) periodic orbits on $s_1$ at the critical points on bifurcation
573: curves of $[20]$ ($+$), $[24]$ ($\circ$)and $[28]$ ($\ast$).}
574: \label{fig:updownorb2}
575: \end{figure}
576: Figure \ref{fig:updownorb2} shows the critical residue behavior of the
577: up and down periodic orbits on the symmetry line $s_1$ along several
578: different bifurcation curves. For lower period bifurcation curves, the
579: residues first show signs of a six-cycle (to be discussed later in
580: greater details), but then converge to $\left|R_i\right|\approx
581: 0.25$. This is because the invariant torus we are studying is not quite
582: shearless. Thus we see the same behavior of the
583: residues as in the case of a twist map. As we proceed to higher period
584: bifurcation curves, the behavior of the residues of the approximating
585: periodic orbits found along the $s_1$ symmetry line resembles more and
586: more a six-cycle. A renormalization group interpretation of these
587: results is given in Sec.~\ref{sec:reng}.
588:
589: Finally, we found the critical point $(a_c,b_c)$ along the
590: $1/\gamma^2$-bifurcation curve to be the following:
591: \beq
592: a_c= 0.425160543 \qquad,\qquad b_c= 0.9244636470355.
593: \eeq
594: At the critical parameter values $(a_c,b_c)$, the residues of the down
595: periodic orbits on $s_1$, which are equal to the residues of the up
596: periodic orbits on $s_2$ because of the symmetry of the map (see
597: Eq.~(\ref{eq:tmap})), converge to the six-cycle\cite{note3}
598: $\left\{ C_1, C_2, C_3, C_4, C_5, C_6\right\}$, where
599: \beq
600: \ba{ll} C_1 = -0.609\pm 0.005, & C_2= -1.288\pm 0.002,\\
601: C_3= 2.593\pm 0.005, & C_4= 1.584\pm 0.008,\\
602: C_5= 2.336\pm 0.006, & C_6= 2.593\pm 0.005.
603: \ea
604: \label{eq:cnum}
605: \eeq
606: The six-cycle can clearly be seen in Fig.~\ref{fig:updownorb} which
607: shows the residues of the up and down periodic orbits at the critical
608: point along the $s_1$ symmetry line.
609: \begin{figure}[ht]
610: \centering
611: \isubfig{
612: \includegraphics[angle=270,scale=2,width=.45\textwidth]{uporb.ps}
613: }
614: \isubfig{
615: \includegraphics[angle=270,scale=2,width=.45\textwidth]{downorb.ps}
616: }
617: \caption{Residue behavior of the up (left figure) and down (right
618: figure) periodic orbits on $s_1$ at the critical point on
619: $1/\gamma^2$-bifurcation curve.}
620: \label{fig:updownorb}
621: \end{figure}
622: The residues of the up periodic orbits on $s_1$ (and of the down periodic
623: orbits on $s_2$) converge to the six-cycle
624: $\left\{ D_1, D_2, D_3,D_4, D_5, D_6\right\}$, where
625: \beq
626: \ba{ll} D_1 = 1.584\pm 0.008, & D_2= -1.288\pm 0.002,\\
627: D_3= -2.630 \pm 0.006, & D_4= -0.609\pm 0.005,\\
628: D_5= 2.336 \pm 0.006, & D_6= -2.630 \pm 0.006.
629: \ea
630: \label{eq:bnum}
631: \eeq
632: The residue convergence for other symmetry lines is shown in Table
633: \ref{tab:respat} where we denote by $R_{u_i}$ and $R_{d_i}$ the
634: residues of the up and down periodic orbit on the symmetry line $s_i$.
635: \begin{table}[ht]
636: \begin{center}
637: \begin{tabular}{|r|cccc|}\hline
638: &&&&\\[-2mm]
639: $[i]\quad\qquad$ & $R_{u_1}=R_{d_2}$ & $R_{u_2} = R_{d_1}$ &
640: $R_{u_3}=R_{d_4}$ & $R_{u_4} = R_{d_3}$\\[2mm]
641: \hline
642: &&&&\\[-2mm]
643: $ [1]\textrm{\ ,\ }[13]\textrm{\ ,\ }[25]$ & $D_1$ & $C_1$ & $D_4$ & $C_4$\\[1mm]
644: $ [3]\textrm{\ ,\ }[15]\textrm{\ ,\ }[27]$ & $D_2$ & $C_2$ & $D_5$ & $C_5$\\[1mm]
645: $ [5]\textrm{\ ,\ }[17]\textrm{\ ,\ }[29]$ & $D_3$ & $C_3$ & $D_6$ & $C_6$\\[1mm]
646: $ [7]\textrm{\ ,\ }[19]\textrm{\ ,\ }[31]$ & $D_4$ & $C_4$ & $D_1$ & $C_1$\\[1mm]
647: $ [9]\textrm{\ ,\ }[21]\textrm{\ ,\ }[33]$ & $D_5$ & $C_5$ & $D_2$ & $C_2$\\[1mm]
648: $[11]\textrm{\ ,\ }[23]\textrm{\ ,\ }[35]$ & $D_6$ & $C_6$ & $D_3$ & $C_3$\\[1mm]
649: \hline
650: \end{tabular}
651: \caption{\label{tab:respat} Period-six convergence pattern of the
652: residues near criticality along the different symmetry lines.}
653: \end{center}
654: \end{table}
655: Note that the six-cycle $\{D_i\}$ of $R_{u_1}$ and $R_{d_2}$ (respectively, the
656: six-cycle $\{C_i\}$ of $R_{u_2}$ and $R_{d_1}$) is observed to be the same as
657: that of $R_{u_3}$ and $R_{d_4}$ (respectively, $R_{u_4}$ and
658: $R_{d_3}$) except it is shifted. The two six-cycles are related
659: because of the symmetry of the map as follows:
660: $D_1=C_4$, $D_2=C_2$, $D_4=C_1$, $D_5=C_5$, $C_3 =C_6$, and $D_3=D_6$.
661: It was numerically observed that $C_6\approx -D_6$, and therefore
662: $C_3\approx -D_3$. Using these relations we see that there are only five
663: independent residues which we take to be $C_1$,$C_2$,$C_3$,$C_4$, and $C_5$.
664:
665: We compared the values of the residues at three different
666: points along the $1/\gamma^2$-bifurcation curve, one point
667: below criticality, one at criticality, and one above
668: criticality: $(a_-,b_-)=(0.425160540,0.9244636195728)$,
669: $(a_c,b_c)=(0.425160543,0.9244636470355)$ and
670: $(a_+,b_+)=(0.425160545,0.9244636653440)$, respectively.
671: The numerical results for the $C_i$ are listed in Table \ref{tab:resnum1}.
672: \begin{table}[ht]
673: \begin{center}
674: \begin{tabular}{ | c || l | rrr | l | rrr | }\hline
675: {[n]} & & $z_-$ & $z_c$ & $z_+$ & & $z_-$ & $z_c$ & $z_+$ \\
676: \hline
677: {[01]} & $C_1$ & 0.565 & 0.565 & 0.565 & $C_4$ & 0.914 & 0.914 & 0.914 \\
678: {[07]} & & -0.702 & -0.702 & -0.702 & & 1.893 & 1.893 & 1.893 \\
679: {[13]} & & -0.601 & -0.601 & -0.601 & & 1.574 & 1.574 & 1.574 \\
680: {[19]} & & -0.611 & -0.611 & -0.612 & & 1.590 & 1.591 & 1.592 \\
681: {[25]} & & -0.610 & -0.612 & -0.614 & & 1.578 & 1.591 & 1.599 \\
682: {[31]} & & -0.566 & -0.612 & -0.644 & & 1.406 & 1.581 & 1.710 \\
683: \hline
684: {[03]} & $C_2$ & -0.752 & -0.752 & -0.752 & $C_5$ & 2.169 & 2.169 & 2.169 \\
685: {[09]} & & -1.328 & -1.328 & -1.328 & & 2.505 & 2.505 & 2.505 \\
686: {[15]} & & -1.286 & -1.286 & -1.286 & & 2.329 & 2.329 & 2.329 \\
687: {[21]} & & -1.289 & -1.290 & -1.291 & & 2.337 & 2.340 & 2.341 \\
688: {[27]} & & -1.273 & -1.289 & -1.300 & & 2.300 & 2.338 & 2.364 \\
689: {[33]} & & -1.161 & -1.249 & -1.276 & & 1.873 & 2.288 & 2.614 \\
690: \hline
691: {[05]} & $C_3$ & 3.450 & 3.450 & 3.450 & $C_6$ & 3.450 & 3.450 & 3.450 \\
692: {[11]} & & 2.534 & 2.534 & 2.534 & & 2.534 & 2.534 & 2.534 \\
693: {[17]} & & 2.598 & 2.598 & 2.598 & & 2.598 & 2.598 & 2.598 \\
694: {[23]} & & 2.588 & 2.594 & 2.598 & & 2.588 & 2.594 & 2.598 \\
695: {[29]} & & 2.498 & 2.588 & 2.650 & & 2.498 & 2.588 & 2.650 \\
696: \hline
697: \end{tabular}
698: \caption{Numerical values of the residue six-cycle $C_i$
699: at $z_-=(a_-,b_-)$, $z_c=(a_c,b_c)$, and $z_+=(a_+,b_+)$.}
700: {\label{tab:resnum1}}
701: \end{center}
702: \end{table}
703: We see that each element of the six-cycle tends to
704: zero for $\left(a_-,b_-\right)$, to infinity for
705: $\left(a_+,b_+\right)$, while it tends to the critical value at
706: $\left(a_c,b_c\right)$.
707: \begin{figure}[h!t]
708: \centering
709: \isubfig{
710: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c1res.ps}
711: }
712: \isubfig{
713: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c2res.ps}
714: }
715: \isubfig{
716: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c3res.ps}
717: }
718: \isubfig{
719: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c4res.ps}
720: }
721: \isubfig{
722: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c5res.ps}
723: }
724: \isubfig{
725: \includegraphics[angle=270,scale=2,width=.45\textwidth]{c6res.ps}
726: }
727: \caption{Residue Convergence for $C_1$ through $C_6$ at $z_1$ ($+$),
728: $z_2$ ($\circ$) and $z_3$ ($\times$) (see Table \ref{tab:resnum1}). }
729: \label{fig:res}
730: \end{figure}
731: Figure \ref{fig:res} clearly illustrates this behavior.
732:
733: A comparison with the results of Ref.~\onlinecite{diego4}
734: shows that, within numerical accuracy, we found the same values for
735: the residues of the six-cycle, but the sequence has shifted by two:
736: $C_1 = H_3$, $C_2= H_4$, $C_3=H_5$, $C_4= H_6$, $C_5= H_1$ and $C_6=
737: H_2$, where $H_i$ denote the residues for the $1/\gamma$ case found in
738: Ref.~\onlinecite{diego4}. A similar shift by two occurs for other
739: symmetry lines.
740:
741: \subsubsection{Spatial scaling at criticality}
742:
743: As expected, the shearless curve exhibits scale invariance at
744: criticality, which can be demonstrated explicitly by using {\it
745: symmetry line coordinates}\cite{diego4} $(\hat{x},\hat{y})$ defined by
746: $\hat{x}=x- a( 1-y^2)/2$ and $\hat{y}=y-y_s$.
747: In these coordinates, the $s_3$ symmetry line becomes a straight line
748: that intersects the shearless curve at the origin. We find that, in
749: symmetry line coordinates, the shearless $1/\gamma^2$ invariant
750: torus at criticality remains invariant under a scale change $(x,y) \to
751: (\alpha^{12} x, \beta^{12} y)$. This property is illustrated in
752: Fig.~\ref{fig:torusatcrit}.
753: \begin{figure}[h!t]
754: \centering
755: \isubfig{
756: \includegraphics[angle=270,scale=2,width=.45\textwidth]{critcurve1.ps}
757: }
758: \isubfig{
759: \includegraphics[angle=270,scale=2,width=.45\textwidth]{critcurve2.ps}
760: }
761: \caption{Invariance under rescaling of shearless $1/\gamma^2$ torus at
762: criticality.} \label{fig:torusatcrit}
763: \end{figure}
764:
765: As described in Ref.~\onlinecite{diego5}, we can find $y_s$ using
766: \beq
767: y_s = \lim_{i\to\infty} \frac{y_{[2i+1]}\; y_{[2i+11]} - y_{[2i-1]}\;
768: y_{[2i+13]}}{\left(y_{[2i+1]}-y_{[2i-1]}\right) - \left(
769: y_{[2i+13]}-y_{[2i+11]}\right)} \approx 0.47253494777 ,
770: \eeq
771: \noindent
772: where $y_{[n]}$ denotes the $y$-coordinate of the periodic orbit
773: $[n]$ along the $s_3$ symmetry line. To obtain the quoted value of
774: $y_s$ we used $i=10$. We then obtained $\alpha$ and $\beta$ as
775: follows:\cite{note4}
776: \beq
777: \alpha = \lim_{n\to\infty} \left|
778: \frac{\hat{x}_{[2i+1]}}{\hat{x}_{[2i+13]}}\right|^{1/12} \approx 1.61759
779: \label{eq:alphaus}
780: \eeq
781: and
782: \beq
783: \beta= \lim_{n\to\infty} \left|
784: \frac{\hat{y}_{[2i+1]}}{\hat{y}_{[2i+13]}}\right|^{1/12} \approx 1.65702,
785: \label{eq:betaus}
786: \eeq
787: where $\left( \hat{x}_{[n]},\hat{y}_{[n]}\right)$ are symmetry line
788: coordinates of the point of the periodic orbit $[n]$ that is
789: the closest to the origin. Within numerical accuracy, these values are
790: the same as in Ref.~\onlinecite{diego5}.
791:
792: Further numerical analysis shows that periodic orbits also exhibit
793: scaling behavior locally near the $s_3$ symmetry line.
794: Figure \ref{fig:periodscale} shows points of the periodic
795: orbit $[n]=[21]$ (in symmetry line coordinates) and points of the
796: periodic orbit $[33]$ with the $x$ and $y$ coordinates rescaled by
797: $\alpha^{12}$ and $\beta^{12}$ respectively. The result suggests that
798: periodic orbits remain invariant under a simultaneous spatial
799: rescaling and shifting of the winding number by twelve from $[n]$ to
800: $[n+12]$.
801: \begin{figure}[ht]
802: \centering
803: \includegraphics[angle=270,scale=2,width=0.9\textwidth]{periodscale.ps}
804: \caption{Invariance of periodic orbits under simultaneous rescaling
805: and shift of winding numbers by twelve. Here we show the periodic
806: orbits $[21]$ ($\times$) and $[33]$ ($\circ$) after rescaling of $x$
807: and $y$ coordinates by $\alpha^{12}$ and $\beta^{12}$ respectively.}
808: \label{fig:periodscale}
809: \end{figure}
810:
811: \subsection{Numerical accuracy}\label{sssec:erroran}
812:
813: We conclude this section with comments about the numerical accuracy of
814: the results.
815:
816: \begin{enumerate}
817: \item Points on the $m/n$-bifurcation curves were found
818: with an accuracy ranging between $10^{-12}$ and $10^{-15}$,
819: where the larger value corresponds to larger periods.
820: We obtain this measure of accuracy from the condition
821: \beq
822: F(y) =0,\quad F'(y)=0\quad \mbox{ and }\quad F''(y)\ne 0 ,
823: \eeq
824: as explained previously. The numbers quoted above are
825: the values of $F(y)$ obtained at the numerically found
826: minima in $y$.
827: \item Periodic orbits along the different symmetry lines
828: around the critical point where found with an accuracy
829: ranging between $10^{-15}$ and $10^{-17}$. Here, the criterion
830: is the difference between the winding number $m/n$ of
831: the periodic orbit of interest, and the winding number of
832: the orbit that results when
833: starting at the numerically found location of the periodic
834: orbit on the respective symmetry line, and then
835: iterating the map.
836: \item A criterion for the accuracy of the scaled bifurcation curve
837: $\Phi_{1/\gamma^2}$ is harder to find, since the location of
838: the actual curve is unknown. An upper bound
839: on the error, though, should be the distance between the
840: $\Phi_{1/\gamma^2}$ and $\Phi_{[37]}$, since the latter
841: definitely lies on the other side of $\Phi_{1/\gamma^2}$. This error was
842: found to be approximately $2\times10^{-13}$.
843: \item A criterion for the accuracy of the critical point
844: in parameter space, $(a_c,b_c)$ is even harder to
845: define, since we cannot actually find the residues of existing orbits
846: of {\it all } periods, which is required to check if the
847: six-cycle at that point continues {\it ad infinitum}. We believe that
848: the value for $a_c$ is accurate up to $1\times 10^{-9}$.
849: \item The uncertainties for the critical residues, quoted above in
850: Eq.~(\ref{eq:cnum}), were computed from the variation in numerical values
851: of residues at $a_c$ for the three or four highest period orbits found (\eg
852: for $C_1$, using residues of $[13]$, $[19]$, $[25]$ and $[31]$).
853: \item If we evaluate the residues of the up and down periodic orbits along
854: the $s_1$ and $s_3$ symmetry lines, then the residues on the other two
855: symmetry lines can be constructed using symmetry arguments (see
856: Sec.~\ref{ssec:po}). But, as a check of the numerical procedures,
857: we independently evaluated the residues on all the four symmetry lines
858: and confirmed the symmetry arguments.
859: \end{enumerate}
860:
861: \section{Renormalization group interpretation}\label{sec:reng}
862:
863: In this section, we interpret the above results within the
864: renormalization group framework. The analysis follows Refs.~\onlinecite{diego5}
865: closely, since, as expected, the residue behavior exhibits a six-cycle
866: at criticality. But, because of the different winding number (\ie
867: different from $1/\gamma$) the renormalization group operator
868: will have a different form.
869:
870: Renormalization ideas have been used fruitfully in area-preserving
871: maps and Hamiltonian flows. (See \eg Refs.~\onlinecite{shenker,mackay1,
872: greene4,escande,chandre3,abad,koch} and references
873: therein.) In contrast to mathematical KAM theory, which proves the
874: existence of dense sets of invariant tori in regions of phase space,
875: the renormalization group approach addresses the problem of the
876: destruction of an invariant torus with a {\it specific} winding number
877: under strong perturbation.
878:
879: The following renormalization approach (see \eg
880: Refs.~\onlinecite{greene6,diego5}) is based on the residue criterion
881: (Sec.~\ref{ssec:po}). To study the breakup of an invariant torus of
882: winding number $\omega$, we loosely represent the map $M$ as:
883: \beq
884: M = \left( R_1, R_2, R_3, \ldots\right) ,
885: \label{eq:lineofres}
886: \eeq
887: where the $\{R_i\}$ are the residues of the periodic orbits with
888: winding numbers $\{m_i/n_i\}$, the convergents of $\omega$. For
889: example, the integrable map will be represented by
890: $\left(0,0,0,\ldots\right)$ because all the orbits are parabolic in
891: that case. The key idea is to construct an operator ${\mathcal R}$
892: that explores the infinite tail of (\ref{eq:lineofres}) by mapping a
893: map given by (\ref{eq:lineofres}) to another map, ${\mathcal R}(M)$,
894: represented by ${\mathcal R}\left(M\right)= \left(
895: \hat{R}_1,\hat{R}_2,\ldots, \hat{R}_i,\ldots\right)$,
896: where $\hat{R}_i=R_{i+1}$. This operation can be interpreted as a
897: {\it time renormalization} since periodic orbits with large periods
898: are transformed into periodic orbits with smaller periods, which
899: amounts to a rescaling of time.
900:
901: The residue criterion can now be rephrased in this framework:
902: \begin{enumerate}
903: \item If $\lim_{n\to\infty} {\mathcal R}^n(M) =
904: \left( 0,0,0,\ldots\right)$, the invariant torus exists.
905: \item If $\lim_{n\to\infty} {\mathcal R}^n(M) =\left(\pm\infty,
906: \pm\infty,\ldots\right)$, the invariant torus is destroyed.
907: \item If $\lim_{n\to\infty} {\mathcal R}^n(M)$ is a map for
908: which the residues have finite, nonzero values, \ie a
909: map that is invariant under the action of ${\mathcal R}^m$ (a fixed
910: point of ${\mathcal R}^m$) for some $m>0$, the
911: invariant torus is at the threshold of destruction. Possible
912: scenarios are the convergence of the residues to a fixed value
913: or to a convergent subsequence.
914: \end{enumerate}
915: There are two kinds of fixed points: {\it simple} fixed points and
916: {\it critical} fixed points. In the case of area-preserving maps, we
917: come to the following interpretation.
918: A {\it simple} fixed point is an integrable map (all the residues are
919: zero), and its basin of attraction contains all the maps for which the
920: invariant torus exists.
921: A {\it critical} fixed point is a map for which the invariant torus
922: under consideration is at criticality. All the maps in its basin of
923: attraction exhibit the same universal behavior at the critical breakup.
924:
925: \subsection{Renormalization group operator}\label{sssec:operator}
926:
927: Following the discussion in Refs.~\onlinecite{mackay1} and
928: \onlinecite{diego5}, we use pairs of commuting maps because they
929: provide a simple way to define the renormalization operators for
930: invariant tori.
931:
932: A {\it pair of commuting maps} is an ordered pair of maps,
933: $(U,T)$, such that $UT=TU$. An {\it orbit} of a point $(x,y)$
934: generated by $(U,T)$ is the set of points $\{U^m T^n (x,y)\}$,
935: where $m$ and $n$ are integers. A {\it periodic orbit of period $m/n$}
936: is an orbit for which $U^{m}T^{n}(x_i,y_i) = (x_i,y_i)$.
937:
938: For the breakup of the invariant torus with winding number
939: $\omega=1/\gamma^2$, we define the renormalization group operator by
940: \beq
941: {\mathcal R} \left(\ba{c} U \\ T\ea\right) := B \left(\ba{c} U^{-1} T^{-1}\\
942: U\; T^2\ea\right) B^{-1}.\label{eq:rgodef}
943: \eeq
944:
945: \noindent
946: As for the case of $1/\gamma$,\cite{diego5} this operator contains both
947: time and space renormalization as follows:
948:
949: The {\it space renormalization} is represented by the operator $B$,
950: which rescales the $(x,y)$ coordinates, \ie $(x,y)\to B(x,y)$ where
951: \beq
952: B = \left(\ba{cc} r & 0\\ 0 & s \ea\right).
953: \eeq
954: At the critical fixed point studied in this paper, we see that
955: $r=\alpha$ and $s=\beta$ given by Eqs.~(\ref{eq:alphaus})-(\ref{eq:betaus}).
956:
957: The {\it time renormalization} is, again, accomplished by the specific
958: combination of the commuting maps.
959: If $(x,y)$ is a periodic orbit of $(U,T)$ with winding number
960: $F_{i}/F_{i+2}$, then $B (x,y)$ is a periodic orbit of
961: $(\hat{U},\hat{T}) = {\mathcal R}\left(U,T\right)$
962: with winding number $F_{i-1}/F_{i+1}$, as can be verified as follows:
963: \bean
964: \hat{U}^{F_{i-1}}\;\hat{T}^{F_{i+1}} B (x,y)
965: &=& B (U T)^{-F_{i-1}}\; \left(U T^2\right)^{F_{i+1}} (x,y)\\
966: &=& B U^{-F_{i-1}+F_{i+1}}\; T^{-F_{i-1}+2F_{i+1}} (x,y)\\
967: &=& B U^{F_i}\; T^{F_{i+2}} (x,y)\\
968: &=& B (x,y).
969: \eean
970: By induction, an orbit with winding number $F_{i}/F_{i+2}$ under
971: $(U,T)$ is transformed into an orbit of ${\mathcal R}^n(U,T)$ with
972: winding number $F_{i-n}/F_{i+2-n}$.
973:
974:
975: \subsection{Simple periodic orbit of ${\mathcal R}$}
976:
977: We can find the integrable period-two orbit $(U_\pm,T_\pm)$ of
978: the renormalization operator (\ref{eq:rgodef}) by requiring that
979: ${\mathcal R}(U_\pm,T_\pm)=(U_\mp,T_\mp)$. This two-cycle is given by
980: the following pairs of maps:
981: \beq
982: U_\pm\xyvect{x}{y} = \xyvect{x-\gamma^2\mp y^2/\gamma^2}{y}, \qquad
983: T_\pm\xyvect{x}{y} = \xyvect{x+1 \pm y^2}{y},
984: \eeq
985: where the rescaling of the coordinates is given by
986: \beq
987: B = \left(\ba{cc} -\gamma & 0\\ 0 & \pm\gamma \ea\right).
988: \eeq
989:
990: Using the definition $U_\pm^m T_\pm^n(x,y)=(x,y)$ of the periodic
991: orbits of period $m/n$, we get the rotation number as a function of $y$:
992: \bean
993: \omega_\pm(y) &=& -\frac{1\pm y^2}{-\gamma^2 \mp y^2/\gamma^2}
994: \;=\; \frac{1}{\gamma^2}\left(1\pm y^2\right)
995: \left(1\pm\frac{y^2}{\gamma^2}\right)^{-1}\\
996: &\approx& \frac{1}{\gamma^2}\left[1\pm
997: \left(1-\frac{1}{\gamma^4}\right)y^2 + \ldots \right].
998: \eean
999: Thus we see that the map $(U_-,T_-)$ is
1000: locally equivalent, under a change of coordinates, to the SNM with
1001: parameters $(a,b)=(1/\gamma^2,0)$.
1002:
1003: \subsection{Critical periodic orbit of ${\mathcal R}$}
1004:
1005: The next step is to analyze the critical periodic orbit of ${\mathcal R}$.
1006: Consider the nontwist map
1007: \beq
1008: {\mathcal C} = \left( C_1,-,C_2,-,C_3,-,C_4,-,C_5,-,C_6,-,C_1,-,C_2,
1009: \ldots\right),
1010: \eeq
1011: where the $C_i$ are the elements of the six-cycle computed earlier,
1012: and the ``-'' denote the periodic orbits that do not exist (see Table
1013: \ref{tab:conv2}). By construction,
1014: this map is a period-12 orbit of the renormalization group
1015: operator (a fixed point of ${\mathcal R}^{12}$), \ie
1016: \beq
1017: {\mathcal R}^{12}{\mathcal C} ={\mathcal C}.\label{eq:fixpt}
1018: \eeq
1019: In Sec.~\ref{ssec:results}, we found that the residues of the
1020: periodic orbits approximating the $1/\gamma^2$-shearless curve in the
1021: standard nontwist map exhibit
1022: convergence to
1023: the six-cycle $\left\{ C_i\right\}$.
1024: Assuming that we can fine-tune the results for
1025: $\left(a_c,b_c\right)$, we expect that
1026: $\lim_{n\to\infty} {\mathcal R}^n M \left(a_c,b_c\right) = {\mathcal C}$.
1027:
1028: If we are studying the breakup of the $1/\gamma^2$-shearless curve for
1029: parameter values along the bifurcation curve for one of the low-period
1030: convergents, then we start near the
1031: stable manifold of the critical periodic orbit of ${\mathcal R}$. But,
1032: under the action of ${\mathcal R}$, we are pushed along an
1033: unstable direction. Thus, we see parts of the six-cycle of residues
1034: (see Fig.~\ref{fig:updownorb2}), but the limiting residue behavior
1035: is observed to be $\lim_{i\to\infty} \left|R_i\right| \approx 0.25$,
1036: which is characteristic for the critical fixed point of {\it twist}
1037: maps (see \eg Ref.~\onlinecite{mackay1}). In renormalization group
1038: language, this means that part of the unstable manifold of the
1039: critical nontwist fixed point (maps for which $(a,b)$ is below
1040: $(a_c,b_c)$) is in the basin of attraction of the critical twist fixed
1041: point.
1042:
1043: \subsection{Eigenvalues}
1044:
1045: As shown in Ref.~\onlinecite{diego5}, it is possible to use our numerical
1046: data to draw further conclusions about the renormalization group
1047: operator ${\mathcal R}$, in particular to compute its unstable
1048: eigenvalues.
1049: The main difficulty in computing these eigenvalues is that the space
1050: of maps is infinite-dimensional whereas the $(a,b)$ parameter space
1051: has only two dimensions.
1052: But the fact that we can find an isolated point $\left(a_c,b_c\right)$
1053: in parameter space at which the map is at
1054: criticality means that the dimension of the
1055: unstable manifold is two.
1056: The map $M$ at $\left(a_c,b_c\right)$ is the intersection
1057: point of the two-parameter family of maps with
1058: the stable manifold of the fixed point, \ie
1059: values $a_c$ and $b_c$ describe the location
1060: of the critical fixed point of ${\mathcal R}^{12}$ in its unstable
1061: manifold.
1062:
1063: Below, we first compute the eigenvalues
1064: characterizing the approach to $\left(a_c,b_c\right)$
1065: in the $(a,b)$ parameter space using the numerical
1066: results from above.
1067: As shown in Ref.~\onlinecite{diego5}, based on the type of
1068: numerical data obtained, the two eigenvalues can be found by
1069:
1070: \beq
1071: \nu_1 = \lim_{n\to\infty} \left(
1072: \frac{\Phi_{[n+12]}\left(a_c\right)-b_c}{\Phi_{[n]}\left(a_c\right)-b_c}\right)
1073: \label{eq:nu1def}
1074: \eeq
1075: and
1076: \beq
1077: \nu_2 = \lim_{n\to\infty} \left( \frac{a_{c\,[2n+12]}-a_c}{a_{c\,[2n]}-a_c}\right).
1078: \label{eq:nu2def}
1079: \eeq
1080:
1081: The last step is to relate the values $\nu_i$ to the unstable
1082: eigenvalues $\delta_i$ of the renormalization group operator
1083: ${\mathcal R}$.
1084: The key idea is to study the behavior of the residues of the
1085: periodic orbits approximating $\omega$ at the $(a,b)$ values
1086: used in the computation of $\nu_i$. For details see Ref.~\onlinecite{diego5}.
1087:
1088: Denoting the unstable eigenvalues of ${\mathcal R}$ by $\delta_1$
1089: and $\delta_2$, we conclude that
1090: $\delta_i = \left(1/\nu_i\right)^{1/12}$.
1091: We find numerical values of
1092: \beq
1093: \delta_1 \approx 2.678\,,\qquad \delta_2 \approx 1.583.
1094: \eeq
1095: Comparing this to the values found in Ref.~\onlinecite{diego5} shows that within
1096: (assumed) numerical uncertainty these values are the same as those
1097: for $1/\gamma$, as predicted.
1098: The $a_c$ values used to determine $\delta_2$ were
1099: $a_{c[26]}$ and $a_{c[14]}$, which explains the larger
1100: discrepancy. Work is under way to improve this result.
1101:
1102: \section{Conclusion}\label{sec:concl}
1103:
1104: We have shown through numerical simulations that the critical
1105: residue values at the breakup of the $1/\gamma^2$-shearless curve
1106: in the standard nontwist map coincide with those of the
1107: $1/\gamma$-shearless curve. In addition, the critical scaling
1108: parameters and the unstable eigenvalues of the renormalization
1109: group operator were found to be the same for both cases.
1110: The main differences are the location of the respective critical
1111: point in parameter space and the detailed form of the
1112: renormalization group operator in terms of commuting maps pairs.
1113:
1114: Future work includes the search for the breakup values of
1115: more winding numbers to map out the details of the critical function
1116: depicted in Fig.~\ref{fig:stntshin2}. In addition, new fixed points of the
1117: renormalization group operator might be obtained by studying the breakup of
1118: shearless curves with non-noble winding numbers.
1119:
1120: \acknowledgments
1121: The authors would like to thank John Greene and Diego del-Castillo-Negrete
1122: for many helpful discussions.
1123: This research was supported in part by U.S. Department of Energy
1124: Contract No. DE-FG01-96ER-54346 and by an appointment of A.~Wurm
1125: to the U.S. Department of Energy Fusion Energy Postdoctoral Research
1126: Program administered by the Oak Ridge Institute for Science and
1127: Education.
1128:
1129: \begin{thebibliography}{99}
1130:
1131: \bibitem{diego2} D.~del-Castillo-Negrete and P.J.~Morrison,
1132: ``Chaotic transport by Rossby waves in shear flow,''
1133: Phys.~Fluids A {\bf 5}, 948 (1993).
1134:
1135: \bibitem{petrisor1} E.~Petrisor, ``Nontwist area preserving maps
1136: with reversing symmetry group,'' Int.~J.~Bif.~Chaos {\bf 11},
1137: 497 (2001).
1138:
1139: \bibitem{balescu} R.~Balescu, ``Hamiltonian nontwist
1140: maps for magnetic field lines with locally reversed
1141: shear in toroidal geometry,'' Phys.~Rev.~E {\bf 58},
1142: 3781 (1998).
1143:
1144: \bibitem{horton1} W.~Horton, H.-B.~Park, J.-M.~Kwon,
1145: D.~Strozzi, P.J.~Morrison and D.-I.~Choi, ``Drift wave
1146: test particle transport in reversed shear profile,''
1147: Phys.~Plasmas {\bf 5}, 3910 (1998).
1148:
1149: \bibitem{kyner} W.T.~Kyner, ``Rigorous and formal stability
1150: of orbits about an oblate planet,'' Mem.~Am.~Math.~Soc. {\bf 81},
1151: 1 (1968).
1152:
1153: \bibitem{chandre2} C.~Chandre, D.~Farrelly and T.~Uzer, ``Threshold
1154: to chaos and ionization for the hydrogen atom in rotating fields,''
1155: Phy.~Rev.~A {\bf 65}, 053402 (2002).
1156:
1157: \bibitem{dullin} H.R.~Dullin, J.D.~Meiss and D.~Sterling,
1158: ``Generic twistless bifurcations,'' Nonlinearity {\bf 13}, 202
1159: (2000).
1160:
1161: \bibitem{vander} J.P.~Van Der Weele, T.P.~Valkering, H.W.~Capel,
1162: T.~Post, ``The birth of twin Poincar\'e-Birkhoff chains near $1:3$
1163: resonance,'' Physica {\bf A 153}, 283 (1988); J.P.~Van Der Weele and
1164: T.P.~Valkering, ``The birth process of periodic orbits in non-twist
1165: maps,'' Physica {\bf A 169}, 42 (1990).
1166:
1167: \bibitem{delshams} A.~Delshams and R.~de~la~Llave, ``KAM
1168: theory and a partial justification of Greene's
1169: criterion for non-twist maps,'' SIAM~J.~Math.~Anal. {\bf 31},
1170: 1235 (2000).
1171:
1172: \bibitem{franks} J.~Franks and P.~Le~Calvez, ``Regions of instability
1173: for nontwist maps,'' preprint, math.DS/9910152, Los Alamos
1174: (1999).
1175:
1176: \bibitem{simo} C.~Sim\'o, ``Invariant curves of analytic
1177: perturbed nontwist area preserving maps,'' Regular and
1178: Chaotic Dynamics {\bf 3}, 180 (1998).
1179:
1180: \bibitem{diego4} D.~del-Castillo-Negrete, J.M. Greene and
1181: P.J.~Morrison, ``Area preserving nontwist maps: periodic orbits
1182: and transition to chaos,'' Physica~D {\bf 91}, 1 (1996).
1183:
1184: \bibitem{diego5} D.~del-Castillo-Negrete, J.M. Greene and
1185: P.J.~Morrison, ``Renormalization and transition to chaos in
1186: area preserving nontwist maps,'' Physica~D {\bf 100}, 311
1187: (1997).
1188:
1189: \bibitem{shin1} S.~Shinohara and Y.~Aizawa, ``The Breakup
1190: Condition of Shearless KAM Curves in the Quadratic Map,''
1191: Progr.~Theo.~Phys. {\bf 97}, 379 (1997).
1192:
1193: \bibitem{note1} Note, that the map used
1194: in Ref.~\onlinecite{shin1} is different from the map we use
1195: (Eq.~(\ref{eq:stntmap})), but they are related by a coordinate
1196: transformation, including rescaling of parameters.
1197:
1198: \bibitem{stark} J.~Stark, ``Determining he critical transition for
1199: circles of arbitrary rotation number,'' Phys.~Lett. {\bf A 163}, 258
1200: (1992).
1201:
1202: \bibitem{greene2} J.M.~Greene, ``A method for computing the
1203: stochastic transition,'' J. Math. Phys. {\bf 20}, 1183
1204: (1979).
1205:
1206: \bibitem{khinchin} A.Ya.~Khinchin, {\it Continued Fractions},
1207: 3rd. ed., University of Chicago Press, Chicago, IL (1964).
1208:
1209: \bibitem{falc} C.~Falcolini and R.~de~la~Llave, ``A rigorous
1210: partial justification of Greene's residue criterion,''
1211: J.~Stat.~Phys. {\bf 67}, 609 (1992).
1212:
1213: \bibitem{mackay2} R.S.~MacKay, ``On Greene's residue
1214: criterion,'' Nonlinearity {\bf 5}, 161 (1992).
1215:
1216: \bibitem{devogel} R.~de~Vogelaere, ``On the structure
1217: of symmetric periodic solutions of conservative
1218: systems, with applications,'' in: {\it
1219: Contributions to the Theory of Nonlinear Oscillations},
1220: Vol.~IV, ed. S.~Lefschetz, Princeton University Press,
1221: Princeton, New Jersey (1958), p.53.
1222:
1223: \bibitem{howard} J.E.~Howard and S.M.~Hohs, ``Stochasticity and
1224: reconnection in Hamiltonian systems,'' Phy.~Rev.~A {\bf
1225: 29}, 418 (1984); J.E.~Howard and J.~Humphreys,
1226: ``Nonmonotonic twist maps,'' Physica~D {\bf 80}, 256 (1995).
1227:
1228: \bibitem{stix} T.H.~Stix, ``Current penetration and
1229: plasma disruption,'' Phys.~Rev.~Lett. {\bf 36},
1230: 10 (1976).
1231:
1232: \bibitem{wurm} A.~Wurm, ``Renormalization Group Applications in Area-Preserving
1233: Nontwist Maps and Relativistic Quantum Field Theory,'' Ph.D. thesis,
1234: The University of Texas at Austin, Austin (May 2002).
1235:
1236: \bibitem{note2} There is a proof for the existence of smooth
1237: bifurcation curves for small values of $b$.\cite{delshams}
1238:
1239: \bibitem{note3} For comments
1240: about the numerical accuracy see Sec.~\ref{sssec:erroran}.
1241:
1242: \bibitem{note4} Note that these equations are derived {\it assuming}
1243: that the shearless curve scales at criticality. The validity of this
1244: assumption is demonstrated {\it a posteriori} by
1245: Fig.~\ref{fig:torusatcrit}.
1246:
1247: \bibitem{abad} J.J.~Abad and H.~Koch, ``A renormalization
1248: group for Hamiltonians: numerical results,'' Nonlinearity
1249: {\bf 11}, 1185 (1998); J.J.~Abad and H.~Koch, ``Renormalization and
1250: periodic orbits for Hamiltonian flows,'' Comm.~Math.~Phys.
1251: {\bf 212}, 371 (2000).
1252:
1253: \bibitem{chandre3} C.~Chandre and H.R.~Jauslin, ``Renormalization-group
1254: analysis for the transition to chaos in Hamiltonian systems,'' Phys.~Rep.
1255: {\bf 365}, 1 (2002).
1256:
1257: \bibitem{escande} D.F.~Escande and F.~Doveil, ``Renormalization
1258: method for computing the threshold of the large-scale stochastic
1259: instability in two degrees of freedom Hamiltonian systems,''
1260: J.~Stat.~Phys. {\bf 26}, 257 (1981).
1261:
1262: \bibitem{greene4} J.M.~Greene, ``How a swing behaves,''
1263: Physica~D {\bf 18}, 427 (1986).
1264:
1265: \bibitem{koch} H.~Koch, ``A renormalization group for
1266: Hamiltonians, with applications to KAM tori,''
1267: Ergod.~Th.~Dyn.~Sys. {\bf 19}, 475 (1999); H.~Koch, ``A
1268: renormalization group fixed point associated with the breakup of
1269: golden invariant tori,'' mp\_arc 02-175, Apr. 9 (2002).
1270:
1271: \bibitem{mackay1} R.S.~MacKay, {\it Renormalization in Area
1272: Preserving Maps}, Ph.D. thesis, Princeton (1982); R.S.~MacKay, ``A
1273: renormalization approach to invariant circles in area-preserving
1274: maps,'' Physica~D {\bf 7}, 283 (1983).
1275:
1276: \bibitem{shenker} S.J.~Shenker and L.P.~Kadanoff, ``Critical
1277: Behavior of a KAM surface: I. Empirical results,''
1278: J.~Stat.~Phys. {\bf 27}, 631 (1982).
1279:
1280: \bibitem{greene6} J.M.~Greene, ``The status of KAM theory
1281: from a physicist's point of view,'' in: {\it Chaos in
1282: Australia}, eds. G.~Brown and A.~Opie, World Scientific,
1283: Singapore (1993), p.8.
1284:
1285: \end{thebibliography}
1286:
1287: \end{document}
1288: