nlin0010022/scar.tex
1: \documentclass[dvips,12pt]{article}
2: \usepackage{a4,graphics,psfig,t1enc}
3: \usepackage{epsf,amsmath,amssymb,latexsym}
4: \textheight 22.5 cm
5: \topmargin -0.5 cm
6: \textwidth 16.0 cm
7: \oddsidemargin -0.0 cm
8: %\linespread{1.6}
9: %\renewcommand{\baselinestretch}{2.0}
10: 
11: \DeclareMathOperator{\rpart}{Re}
12: \DeclareMathOperator{\impart}{Im}
13: 
14: \begin{document}
15: 
16: \setcounter{page}{1}
17: 
18: {\small{\date{\today}}}
19: \title{Orbit bifurcations and the scarring of wavefunctions\footnote{Short
20:  title: Bifurcations and scarring}}
21: \author{J.~P.~Keating$^{1}$ and S.~D.~Prado$^{2}$ \\
22: {\small $1$ School of Mathematics, University of Bristol, Bristol} \\ 
23: {\small BS8 1TW, UK, and BRIMS, Hewlett-Packard Laboratories,} \\
24: {\small Filton Road, Stoke Gifford, Bristol BS34 8QZ, UK.}\\
25: {\small $2$ Instituto de F{\'\i}sica, Universidade Federal do Rio
26:  Grande do Sul} \\
27: {\small P.O. Box 15051, 91501-970 Porto Alegre, RS, Brazil.}
28: }
29: \maketitle
30: \begin{abstract}
31: We extend the semiclassical theory of scarring of quantum 
32: eigenfunctions $\psi_{n}({\bf{q}})$ by classical periodic orbits to 
33: include situations where these orbits undergo generic bifurcations.
34: It is shown that $\left |\psi_{n}({\bf{q}}) \right |^{2}$, averaged locally 
35: with respect to position ${\bf q}$ and the energy spectrum $\{E_{n}\}$, has
36: structure around bifurcating periodic orbits with an amplitude and 
37: length-scale whose $\hbar$-dependence is determined by the bifurcation
38: in question. Specifically, the amplitude scales as $\hbar^\alpha$ and the 
39: length-scale as $\hbar^\omega$, and values of the {\em scar exponents}, 
40: $\alpha$ 
41: and $\omega$, are computed for a variety of generic bifurcations.  In each 
42: case, the scars are semiclassically wider than those associated with isolated 
43: and unstable periodic orbits; moreover, their amplitude is 
44: at least as large, and in 
45: most cases larger.  In this sense, bifurcations may be said to give rise to 
46: {\em superscars}.  The competition between the contributions from different
47: bifurcations to determine the moments of the averaged 
48: eigenfunction amplitude is
49: analysed.  We argue that there is a resulting universal $\hbar$-scaling 
50: in the semiclassical asymptotics of these moments for irregular states in 
51: systems with a mixed phase-space dynamics.  
52: Finally, a number of these predictions are illustrated by numerical 
53: computations for a family of perturbed cat maps. 
54: \end{abstract}
55: 
56: \newpage
57: 
58: \section{Introduction}
59: 
60: \hspace{\parindent}
61: 
62: One of the main goals in quantum chaology has been to determine the 
63: link between classical periodic orbits and quantum spectral 
64: fluctuations in the semiclassical limit. In fully chaotic systems, 
65: where the periodic orbits are isolated and unstable, this connection
66: is embodied in Gutzwiller's trace formula (Gutzwiller 1971), and in
67: integrable systems by a corresponding expression involving the phase-space
68: tori (Berry \& Tabor 1976). These formulae fail, by diverging, when
69: periodic orbits bifurcate; that is, when combinations of stable and 
70: unstable orbits collide and transmute, or annihilate, as a system 
71: parameter varies --  phenomena that characterize dynamics in the mixed
72: regime. They must then be replaced by transitional or uniform 
73: approximations which interpolate through the bifurcation (Ozorio de
74: Almeida \& Hannay 1987; Tomsovic {\em et al.} 1995; Ullmo {\em et al.} 1996;
75:  Sieber 1996; 
76: Schomerus \& Sieber 1997; Schomerus 1998; Sieber \& Schomerus 1998).
77: 
78: That individual orbit bifurcations can have an important, and sometimes
79: dominant influence on spectral statistics was pointed out by Berry 
80: {\em et al.} (1998), and demonstrated for a particular example, the 
81: perturbed cat maps. More generally, Berry {\em et al.} (2000) developed
82: a semiclassical theory for the competition
83: between the various generic bifurcations found in Hamiltonian systems
84: to determine the moments of the quantum energy level counting function. This
85: suggests that these moments diverge in a universal way, characterized
86: by certain {\em twinkling exponents}, as $\hbar \rightarrow 0$.
87: 
88: A second major goal of quantum chaology has been to understand the 
89: influence of classical periodic orbits on quantum wavefunctions in the 
90: semiclassical limit. It was first noticed by McDonald (McDonald 1983) 
91: that individual eigenfunctions can have enhanced intensity along short
92: periodic orbits in classically chaotic systems. This phenomenon was 
93: later studied systematically by Heller (Heller 1984), who called
94: such structures {\em scars}. He developed a theory of scarring, based
95: on wavepacket dynamics, which has subsequently been extended to 
96: describe a variety of statistical properties of quantum chaotic
97: eigenfunctions (Kaplan 1999).
98: 
99: An alternative theory of scarring, based on an approach closely related
100: to the trace formula, was initiated by Bogomolny (Bogomolny 1988). In
101: this, the semiclassical approximation to the energy-dependent Green
102: function is used to show that for quantum eigenfunctions
103:  $\psi_n({\bf q})$  corresponding to energy
104: levels $E_{n}$ in a fixed energy range, 
105: $\left < \left | \psi_n({\bf q})\right |^{2} \right >$, where 
106: $\left < \cdots \right >$ denotes an average over the states in question
107: and locally over position ${\bf q}$, has complex-Gaussian fringes with, in 
108: two-degree-of-freedom systems, amplitude and length-scale of
109: the order of $\hbar^{1/2}$ around unstable periodic orbits. A 
110: corresponding theory for Wigner functions was developed by Berry (1989).
111: 
112: We emphasize two limitations of the theories mentioned above. First, 
113: they only describe scarring in eigenfunctions that have been averaged 
114: over an energy
115: interval which, semiclassically, contains a large number of states. 
116: Resummation techniques have been applied to provide
117: some information about individual eigenfunctions (Agam \& Fishman 1994;
118: Fishman {\em et al.} 1996), but a detailed understanding of the 
119: phenomenon in this case remains to be developed. Second, they 
120: concentrate specifically  on the influence of periodic
121: orbits which are isolated and unstable. (Semiclassical theories 
122: describing the connection between quantum wavefunctions and phase-space
123: tori in classically integrable systems have also been developed; see, 
124: for example, Berry 1983 for a detailed review.)
125: 
126: Our purpose here is to address the second of these limitations. 
127: Specifically, our aim is to show how Bogomolny's theory can be extended
128: to include the description of semiclassical structures in quantum 
129: eigenfunctions associated with generic classical periodic orbit 
130: bifurcations in systems with two degrees of freedom. We focus in 
131: particular on the $\hbar$-dependence of the amplitude and length-scale
132: of the fringes corresponding to 
133: those identified by Bogomolny. Our main result is that the amplitude
134: is of the order of $\hbar^{\alpha}$, and the length-scale is of the order
135: of $\hbar^{\omega}$, where $\alpha$ and $\omega$ are bifurcation-dependent
136: {\em scar exponents} whose values we calculate in a number of 
137: different cases. Crucially, $\omega<1/2$ for all the bifurcations
138: studied, and for most $\alpha<1/2$ as well. In this sense, 
139: bifurcations may be said to give rise to {\em superscars}. In order 
140: to quantify this, we determine the way in which bifurcating orbits 
141: contribute, via a competition, to the semiclassical asymptotics of the 
142: moments of 
143: $\left < \left | \psi_{n}(q)\right |^{2} \right >$, in the same way
144: as was done for spectral fluctuations by Berry {\em et al.} (2000).  It is 
145: argued that this competition results in universal $\hbar$-scalings of the 
146: moments for the irregular eigenfunctions of systems with mixed
147: phase-space dynamics.
148: Finally, as an example, we apply some of the techniques developed to 
149: study the influence of one particular bifurcation on the eigenfunctions
150: of a family of quantum perturbed cat maps.   
151: 
152: \section{Scar Formulae}
153: 
154: Our aim in this section is to derive semiclassical scar formulae for
155: bifurcating periodic orbits which generalize those obtained in Bogomolny (1988)
156:  for unstable orbits far from bifurcation.
157: 
158: We begin, following Bogomolny, with the energy-dependent Green function
159: 
160: \begin{equation}
161: G({\bf q'},{\bf q};E)=\sum_{n}\frac{\psi^{\ast}_{n}({\bf q'})\psi_{n}({\bf q})}{E-E_{n}},
162: \label{f1}
163: \end{equation}
164: where $\psi_{n}({\bf q})$ is the eigenfunction of the quantum Hamiltonian 
165: corresponding to the energy level $E_{n}$. The identity we seek to 
166: exploit follows from setting ${\bf q'}={\bf q}$:
167: 
168: \begin{equation}
169: \sum_{n}\left | \psi_{n}({\bf q}) \right |^{2}\delta_{\varepsilon}
170: \left (E-E_{n} \right )=-\frac{1}{\pi}\impart G({\bf q},{\bf q};E+i\varepsilon).
171: \label{f2}
172: \end{equation}
173: 
174: \noindent Here, $\delta_{\varepsilon}(x)$ is a normalized, 
175: Lorentzian-smoothed $\delta$-function of width $\varepsilon$. (It is
176: straightforward to transform (\ref{f2}) to give differently smoothed
177: $\delta$-functions, for example Gaussians.) The left-hand side of 
178: (\ref{f2}) thus corresponds to a sum over eigenstates for which 
179: $E_{n}$ lies within a range of size of the order of $\varepsilon$ centred on 
180: $E$. Semiclassically, it is approximately the average of $|\psi_{n}({\bf q})|^{2}$ 
181: with respect to these states multiplied by $\overline{d}(E)$, the
182: mean level density. For systems with two-degrees-of-freedom
183: \begin{equation}
184: \overline{d}(E) \sim \frac{V(E)}{(2\pi\hbar)^{2}}
185: \label{f3}
186: \end{equation}
187: as $\hbar \rightarrow 0$, where 
188: \begin{equation}
189: V(E)=\int \delta (E-H({\bf p},{\bf q})) d^{2}{\bf q} d^{2}{\bf p}
190: \label{f4}
191: \end{equation}
192: and $H({\bf p},{\bf q})$ is the classical Hamiltonian.
193: 
194: The connection with classical mechanics is achieved using the 
195: semiclassical approximation to the Green function. For systems with 
196: two-degrees-of-freedom, this is 
197: \begin{equation}
198: G({\bf q'},{\bf q};E)\approx \frac{1}{i\hbar \sqrt{2\pi i \hbar}}
199: \sum_{\gamma}\sqrt{\left | D_{\gamma} \right |} \exp{ \left \{\frac{i}{\hbar}
200: S_{\gamma}({\bf q'},{\bf q};E)-\frac{i\pi}{2}\nu_{\gamma} \right \}},
201: \label{f5}
202: \end{equation}
203: where the sum includes all classical trajectories from ${\bf q}$ to ${\bf q'}$
204: at energy $E$, $S_{\gamma}$ is the action along the trajectory labelled
205: $\gamma$,
206: \begin{equation}
207: D_{\gamma}=\det{ \left (
208: \begin{array}{ll}
209: \frac{\partial^{2}S_{\gamma}}{\partial {\bf q'}\partial {\bf q}} &
210: \frac{\partial^{2}S_{\gamma}}{\partial {\bf q'}\partial E} \\
211: \frac{\partial^{2}S_{\gamma}}{\partial E \partial {\bf q}} &
212: \frac{\partial^{2}S_{\gamma}}{\partial E^{2}}
213: \end{array} \right )},
214: \label{f6}
215: \end{equation}
216: and $\nu$ is the Maslov index (Gutzwiller 1990). When ${\bf q'}={\bf q}$,
217: the sum in (\ref{f5}) is clearly over closed orbits.
218: 
219: We note in passing that it follows from (\ref{f2}) that
220: \begin{equation}
221: \sum_{n}\delta_{\varepsilon}(E-E_{n})=-\frac{1}{\pi}\impart \int G({\bf q},{\bf q};E+
222: i\varepsilon) d^{2}{\bf q}.
223: \label{f7}
224: \end{equation}
225: Substituting in the closed orbit sum for $G({\bf q},{\bf q};E+i\varepsilon)$ and
226: integrating term-by-term using the method of stationary phase leaves 
227: contributions from the periodic orbits. Assuming these are all isolated,
228: as is the case for hyperbolic systems, the result is the trace 
229: formula (Gutzwiller 1971)
230: 
231: \begin{multline}
232: \sum_{n}\delta_{\varepsilon}(E-E_{n})\approx \overline{d}(E) +  
233: \frac{1}{\pi \hbar} \sum_{p}\sum_{r=1}^{\infty} \frac{T_{p}}
234: {\sqrt{\left | \det{ ({\bf M}^{r}_{p}-{\bf I})}\right |}} \times \\
235: \cos{\left (
236: \frac{rS_{p}}{\hbar}- r\nu_p \frac{\pi}{2}\right ) }\exp{ \left [-\varepsilon 
237: \frac{rT_{p}}
238: {\hbar}\right ] },
239: \label{f8}
240: \end{multline}
241: where $p$ labels primitive periodic orbits with period $T_{p}$ and
242: monodromy matrix ${\bf M}_{p}$, and $r$ labels repetitions. As noted
243: in the Introduction, this formula fails at bifurcations, where
244: $\det{({\bf M}^{r}_{p}-{\bf I})}=0$. Assuming that the periodic orbits lie in
245: families which form tori in phase space gives the corresponding
246: expression for integrable systems (Berry \& Tabor 1976).
247: 
248: Bogomolny's scar formula follows not from integrating over all 
249: positions ${\bf q}$, as in (\ref{f7}), but from performing a {\em local 
250: average} of (\ref{f2}) with respect to ${\bf q}$ (we postpone specifying the
251: size of the averaging range until after the result has been stated), which we
252: take to be smooth (convolution with a normalized Gaussian, for example). On
253: the left-hand side this gives, approximately, $\overline{d}(E)\left <
254:  \left | \psi_{n}({\bf q}) \right |^{2} \right >$, where 
255: $\left < \cdots \right >$ denotes a combination of the spectral average
256: described above and the local ${\bf q}$-average.  On the
257:  right-hand side, the ${\bf q}$-average selects from the closed orbits
258: those that are close to periodic orbits (i.e. for which the change
259: in momentum after return is appropriately small). These can then be 
260: described by linearizing about the periodic orbits. Essentially, this
261: corresponds to expanding the action up to terms which are quadratic
262: in the distance from the periodic orbit. The result is that
263: \begin{multline}
264: \sum_{n} \left < \left | \psi_{n}({\bf q})\right |^{2}\right >_{\bf q} 
265: \delta_{\epsilon}(E-E_{n})\approx 
266: \frac{1}{(2\pi \hbar)^{2}}\Omega({\bf q};E) - \\
267: \frac{1}{\pi\hbar^{3/2}}\impart 
268: \frac{1}{i\sqrt{2\pi i}}\sum_{p}\sum_{r=1}^{\infty} 
269: \frac{1}{\left |\dot{z}\right |
270: \sqrt{\left [ {\bf M}^{r}_{p} (z) \right ]_{12}}} \times \\
271:  \exp{
272: \left [\frac{i}{\hbar} \left ( r S_{p}+\frac{1}{2}
273: \frac{\det{\left ({\bf M}^{r}_{p}-I \right )}}
274: {\left [{\bf M}^{r}_{p}(z)\right ]_{12}}y^2-r\nu_{p}\frac{\pi}{2} \right ) 
275: \right ]} 
276: \exp{\left [-\varepsilon \frac{rT_{p}}{\hbar}\right ]},
277: \label{f9}
278: \end{multline}
279: where $z$ is a coordinate along a given periodic orbit and $y$ is a
280: coordinate transverse to it, 
281: $\left [ {\bf M}^{r}_{p} (z) \right ]_{ij}$ denotes
282: the elements of the monodromy matrix (which are functions of $z$), 
283: $\dot{z}$ is the velocity along the periodic orbit, and
284: \begin{equation}
285: \Omega({\bf q};E)=\int \delta(E-H({\bf p},{\bf q}))d^{2}{\bf p}.
286: \label{f10}
287: \end{equation}
288: This in turn implies that
289: \begin{multline}
290: \left < \left | \psi_{n}({\bf q}) \right |^{2} \right > \approx
291: \frac{\Omega({\bf q};E)}{V(E)} - \frac{4\pi\sqrt{\hbar}}{V(E)} \impart 
292: \frac{1}{i\sqrt{2\pi i}}\sum_{p} \sum_{r=1}^{\infty}
293: \frac{1}{\left | \dot{z}\right |
294: \sqrt{\left [ {\bf M}^{r}_{p}(z) \right ]_{12}}} \times \\
295: \exp{
296:  \left [
297:  \frac{i}{\hbar} \left (
298:  r S_{p} +\frac{1}{2}\frac{\det{\left ( 
299: {\bf M}^{r}_{p}-{\bf I} \right )}}
300: {\left [ {\bf M}^{r}_{p}(z) \right ]_{12}}
301: y^2-r\nu_{p}\frac{\pi}{2}\right )
302:  \right ] }   
303:  \exp{\left [ -\varepsilon \frac{ r T_{p}}{\hbar} \right ]  }.
304: \label{f11}
305: \end{multline}
306: 
307: Equation (\ref{f11}) is Bogomolny's scar formula. In classically ergodic
308: systems, the first term represents
309: the quantum-ergodic limit of the eigenfunction 
310: probability density (Shnirelman 1974, 
311: Colin de Verdi\`ere 1985, Zelditch 1987).  The second  describes complex
312: Gaussian fringes (the $y$-dependent part), with length-scale and 
313: amplitude both of the order of $\hbar^{1/2}$, associated with each 
314: periodic orbit. This structure will be resolved if the local 
315: ${\bf q}$-average is over regions whose dimensions 
316: are small compared to the length-scale
317: of the fringes; that is, over regions whose dimensions scale 
318: as $\hbar^{\delta}$,
319: where $\delta > 1/2$. In order for the near-to-periodic-orbit 
320: approximation to be valid, we must also have $\delta<1$; that is, the 
321: dimensions of the 
322: averaging range must be large compared to a de Broglie wavelength.
323: 
324: The trace formula (\ref{f8}) can be recovered from (\ref{f9}) by 
325: integrating over $z$ and $y$. The $z$-integral gives the period in the
326: amplitude of the periodic orbit contributions, and the $y$-integral
327: gives the determinant. Note that the power of $\hbar$ in the trace
328: formula amplitude is the amplitude exponent in (\ref{f9}), $-3/2$
329: (which in turn is equal to the amplitude exponent in (\ref{f11}) minus
330: two -- the exponent in $\overline{d}$), plus the length-scale exponent
331: of the fringes, $1/2$.
332: 
333: The approximations (\ref{f9}) and (\ref{f11}) break down in two ways. First,
334: at self-focal points along an orbit 
335: $\left [ {\bf M}^{r}_{p} \right (z)]_{12}=0$ and the amplitude diverges. 
336: This can be remedied straightforwardly using Maslov's method, and we 
337: will not concern 
338: ourselves further with it here. Second, when an orbit bifurcates
339: $\det{({\bf M}^{r}_{p}-{\bf I})}=0$, and so the formulae become
340: $y$-independent. Essentially, this means that the fringes are 
341: infinitely wide (it is this infinity which, upon integration with 
342: respect to $y$, transfers itself to the amplitude in the trace 
343: formula). Our purpose in this paper is to show how to correct (\ref{f9})
344: and (\ref{f11}) in this case. 
345: 
346: It might be thought that the scar formulae
347: for bifurcating orbits could be obtained easily by expanding the 
348: action in (\ref{f5}) to higher order than quadratic. For some of the 
349: simpler bifurcations (e.g. the codimension-one bifurcations of orbits 
350: with $r=1$ and $r=2$) this is correct (see the example in Section 4).
351: However, for more complicated bifurcations it is incorrect, because
352: for these the linearized map ${\bf M}^{r}_{p}$ is equal to the identity,
353: which cannot be generated by the action $S({\bf q'},{\bf q};E)$. Thus it is
354: difficult to build into the semiclassical expression for the 
355: ${\bf qq'}$-representation of the Green function a well-behaved description of
356: the nonlinear dynamics which the linearized map approximates. The
357: solution to this problem, originally proposed in Ozorio de Almeida
358: \& Hannay (1987), is to transform the Green function to a mixed
359: position-momentum representation, and this is the approach we now take.
360:      
361: The Ozorio de Almeida-Hannay method involves, first, Fourier 
362: transforming $G({\bf q'},{\bf q};E)$ with respect to ${\bf q'}$. 
363: This gives the 
364: Green function in the ${\bf q}$ ${\bf p'}$-representation, 
365: $\tilde{G}({\bf p'},{\bf q};E)$
366: (${\bf p'}$ is the momentum conjugate to ${\bf q'}$). The semiclassical 
367: approximation to $\tilde{G}$ takes the same form as (\ref{f5}),
368: except that $S({\bf q'},{\bf q};E)$ is replaced by the 
369: ${\bf q}$ ${\bf p'}$-generating 
370: function $\tilde{S}({\bf p'},{\bf q};E)$. $G({\bf q'},{\bf q};E)$ 
371: may then be rewritten,
372: semiclassically, as the Fourier transform of this expression with 
373: respect to ${\bf p'}$. (For an alternative approach leading to the same
374: final answer, see Sieber 1996). The result, for the semiclassical 
375: contribution to $G({\bf q},{\bf q};E)$ from closed orbits in the neighbourhood
376: of a bifurcating periodic orbit, takes the following form.
377: 
378: Consider the case of a codimension-$K$ bifurcation of a periodic 
379: orbit with repetition number $r$. 
380: As before, let $z$ be a coordinate along the orbit at
381: bifurcation, let $y$ be a coordinate transverse to it, and let $p_{y}$
382: be the momentum conjugate to $y$, so that $y$ and $p_{y}$ are local
383: surface of section coordinates. Let $\Phi_{r,K}(y,  p_{y},{\bf x})$
384: be the normal form which corresponds to the local (reduced) generating
385: function in the neighbourhood of the bifurcation (Arnold 1978;
386: Ozorio de Almeida 1988), where ${\bf x}=(x_{1},x_{2},\cdots,
387: x_{K})$
388: are parameters controlling the unfolding of the bifurcation.
389: Then, up to irrelevant factors, the contribution to $G({\bf q},{\bf q};E)$ is
390: \begin{equation}
391: G_{r,K}(y;{\bf x})=\frac{1}{\hbar^2}\int 
392: \exp{\left [\frac{i}{\hbar}
393: \Phi_{r,K}(y,p_{y},{\bf x})\right ]} dp_{y}
394: \label{f12}
395: \end{equation}
396: (as already stated, we are here interested in determining the
397: $\hbar$-dependence of the amplitude and length-scale of the associated
398: fringes, and so have neglected terms in (\ref{f12}), such as an 
399: $\hbar$-independent factor in the integrand, which do not influence
400: these). 
401: 
402: Before proceeding further, we make three remarks about (\ref{f12}). 
403: First, the power of $\hbar$ outside the exponential arises from adding
404: $1/2$, which comes from the Fourier transform, to the exponent in 
405: (\ref{f5}), $3/2$. Second, the representation used in Berry {\em et
406: al.} (2000)
407: for the fluctuating part of the spectral counting function may be
408: derived from (\ref{f12}) by taking the trace of $G_{r,K}$, which
409: involves
410:  integrating the right-hand side of (\ref{f12}) with respect to 
411: $y$ (the $z$-integral is trivial, as before), and then integrating
412: with
413: respect to $E$ resulting in a further multiplication by $\hbar$.  
414: Likewise, the formulae of Ozorio de Almeida and Hannay (1987) 
415: correspond to taking the trace of $G_{r,K}$, keeping the terms we
416: have neglected. Third, the $\hbar$-dependence of the fringes in 
417: Bogomolny's scar formula for unstable periodic orbits far from
418: bifurcation can be recovered using the appropriate normal form:
419: \begin{equation}
420: \Phi_{r,0}=p^{2}_{y}+y^{2},
421: \label{f13}
422: \end{equation}
423: which corresponds to a particular, $\hbar$-independent choice of units
424: for $p_{y}$ and $y$. Evaluating the integral then gives
425: \begin{equation}
426: G_{r,0}({\bf q})\propto \frac{1}{\hbar^{3/2}} \exp{\left (
427: i\frac{y^2}{\hbar} \right )},
428: \label{f14}
429: \end{equation}
430: as in (\ref{f9}).
431: 
432: Equation (\ref{f12}) is the starting point for the analysis of 
433: bifurcating orbits. Our strategy is essentially the same as that used
434: in Berry {\em et al.} (2000) to study the related fluctuations in the 
435: spectral counting function (see also Berry 2000 for a review of 
436: applications to other areas in wave physics): first, rescale $y$ and
437: $p_y$ to remove the $1/\hbar$ factor from the 
438: dominant term (germ) of $\Phi_{r,K}$ in the exponent, and then apply a 
439: compensating rescaling of the parameters $x_{1},x_{2},\cdots, x_{K}$
440: to remove the $\hbar$-dependence from the other terms which do not
441: vanish as $\hbar \rightarrow 0$. This  will lead to 
442: \begin{equation}
443: G_{r,K}(y;{\bf x};\hbar)=\frac{1}{\hbar^{2-\alpha_{r,K}}}G_{r,K}
444: \left (\frac{y}{\hbar^{\omega_{r,K}}},\left
445: \{ \frac{x_{n}}{\hbar^{\sigma_{n},r,K}}\right \},1 \right ).
446: \label{f15}
447: \end{equation}
448: The exponent $\alpha$ describes the semiclassical amplitude of the fringes in
449: $\left < \left | \psi_{n}({\bf q}) \right |^{2} \right >$ associated with
450: the 
451: bifurcation, and the exponent $\omega$ describes the
452: $\hbar$-dependence
453:  of their length-scale, or width. We call these the {\em scar
454: exponents}. Note that the corresponding amplitude exponent in
455:  $\sum_{n}\left < \left | \psi_{n}({\bf q}) \right |^{2}\right
456: >_{q}\delta_\varepsilon (E-E_{n})$ is $2-\alpha$. The exponents $\sigma$
457: describe the range of influence of the bifurcation in the different
458: unfolding directions $x_{n}$. Their sum
459: \begin{equation}
460: \gamma_{r,K}=\sum_{n=1}^{K}\sigma_{n,r,K}
461: \label{f16}
462: \end{equation}
463: describes the $\hbar$-scaling of the $K$-dimensional
464: ${\bf x}$-space
465: hypervolume affected by the bifurcation.
466: 
467: We now calculate these exponents in a variety of examples. Consider
468: first the $r=1$ bifurcations which correspond to cuspoid (i.e. corank
469: 1) catastrophes. For these, the normal forms are (Berry {\em et al.}
470: 2000)
471: \begin{equation}
472: \Phi_{1,K}(y,p_{y};{\bf x})=p^{2}_{y}+y^{K+2}+\sum_{n=1}^{K}
473: x_{n}y^{n}.
474: \label{f17}
475: \end{equation}
476: Substituting this into (\ref{f12}) and evaluating the integral then
477: gives
478: \begin{equation}
479: G_{1,K}(y;{\bf x}) \propto \frac{1}{\hbar^{3/2}}\exp{\left [\frac{i}{\hbar}
480: \left (
481: y^{K+2}+\sum^{K}_{n=1}x_{n}y^{n} \right ) \right]} 
482: \label{f18}
483: \end{equation}
484: Making the rescalings $\tilde{y}=y/\hbar^{1/(K+2)}$, 
485: $\tilde{x}_{n}=x_{n}/\hbar^{1-n/(K+2)}$ removes the $\hbar$-dependence
486: of the exponent, and so for the cuspoids we have
487: \begin{equation}
488: \alpha_{1,K}=\frac{1}{2}, {\mbox{\hspace{1cm}}}  \omega_{1,K}=\frac{1}{K+2}, 
489: {\mbox{\hspace{1cm}}} 
490: \sigma_{n,1,K}=1-\frac{n}{K+1}
491: \label{f19}
492: \end{equation}
493: and
494: \begin{equation}
495: \gamma_{1,K}=\frac{K(K+3)}{2(K+2)}.
496: \label{f20}
497: \end{equation}
498: Analogous expressions can be written down for the $r=1$ bifurcations
499: corresponding to the more complicated case of catastrophes of corank 
500: $2$ (see, for example, Berry 2000) in the same way.
501: 
502: When $r>1$, the generic bifurcations with $K=1$ have been classified
503: by Meyer (Meyer 1970, 1986; Arnold 1978; Ozorio de Almeida 1988), and
504: those with $K=2$ by Schomerus (1998). The relevant parts of the 
505: corresponding normal forms, taken from Berry (2000) (to which readers
506: are referred for further details), are summarized in Table 1 (in the
507: expressions given, we are retaining only those terms which affect the
508: exponents we seek to calculate).
509: 
510: \begin{table}[htb] 
511: \begin{center}
512: \begin{tabular}{|c|c|} \hline\hline
513: $r$ & $\Phi_{r,2}$ \\ \hline
514: $2$ & $p^{2}_{y}+y^6+x_{1}y^2+x_{2}y^{4}$ \\ 
515: $3$ & $(p^{2}_{y}+y^{2})^2+x_{1}(p^{2}_{y}+y^{2})+x_{2}
516: \rpart{\left[ (p_{y}+ i y)^3\right ]}$ \\ 
517: $4$ & $p^{2}_{y}y^{2}+x_{1}(p^{2}_{y}+y^{2}) + 
518: x_{2}(p^{2}_{y}-y^{2})^2$\\ 
519: $5$ & $\rpart{\left [(p_{y}+i
520: y)^{5}\right ] }+x_{1}(p^{2}_{y}+y^2)+x_{2}(p^{2}_{y}+y^{2})^{2}$ \\ 
521: $\geq 6$ &
522: $(p^{2}_{y}+y^{2})^{3}+x_{1}(p^{2}_{y}+y^{2})+x_{2}(p^{2}_{y}+
523: y^{2})^{2}$ \\ \hline\hline
524: \end{tabular}
525: \end{center}
526: \caption{The relevant parts of the normal forms for $K=2$ bifurcations
527: of period-$r$ orbits (taken from Berry 2000). The corresponding 
528: expressions for $K=1$ bifurcations, $\Phi_{r,1}$, follow from settling
529: $x_{2}=1$.}
530: \label{table1}
531: \end{table}
532: 
533: 
534: The corresponding scar exponents, and the hypervolume exponents
535: $\gamma$
536: are given in Table 2 $(K=1)$ and Table 3 ($K=2$).
537: 
538: \begin{table}[htb]
539: \begin{center}
540: \begin{tabular}{|c|c|c|c|} \hline\hline 
541: $r$      & $\alpha_{r,1}$ & $\omega_{r,1}$ & $\gamma_{r,1}$ \\ \hline 
542: $2$      & $1/2$     &  $ 1/4$        &     $ 1/2$     \\ 
543: $3$      & $1/3$     &  $ 1/3$        &     $ 1/3$     \\ 
544: $\geq 4$ & $1/4$     &  $ 1/4$        &     $ 1/2$     \\ \hline\hline
545: \end{tabular}
546: \end{center}
547: \caption{Scar exponents for generic, codimension-1 bifurcations.}
548: \label{table2}
549: \end{table}
550: 
551: \begin{table}[htb]
552: \begin{center}
553: \begin{tabular}{|c|c|c|c|} \hline\hline
554: $r$      & $\alpha_{r,2}$ & $\omega_{r,2}$ & $\gamma_{r,2}$ \\ \hline
555: $2$      & $1/2$     &   $1/6$        &      $1$       \\ 
556: $3$      & $1/4$     &   $1/4$        &      $3/4$     \\ 
557: $4$      & $1/4$     &   $1/4$        &      $1/2$     \\ 
558: $5$      & $1/5$     &   $1/5$        &      $4/5$     \\ 
559: $\geq 6$ & $1/6$     &   $1/6$        &      $1$       \\ \hline\hline 
560: \end{tabular}
561: \end{center}
562: \caption{Scar exponents for generic, codimension-2 bifurcations.}
563: \label{table3}
564: \end{table}
565: 
566: Finally, we consider bifurcations of orbits for which $r\geq 2K+2$.
567: In this case, the relevant terms in the normal forms are (Berry {\em
568: et
569: al.} 2000)
570: \begin{equation}
571: \Phi_{r,K}(y,p_{y};{\bf x})=I^{K+1}+\sum^{K}_{n=1}x_{n}I^{n}+
572: {\cal{O}}(I^{K+2}),
573: \label{f21}
574: \end{equation}
575: where
576: \begin{equation}
577: I=y^2+p^{2}_{y}.
578: \label{f22}
579: \end{equation}
580: Expressing $\Phi$ in terms of $y$ and $p_{y}$, we find
581: \begin{equation}
582: \alpha_{r,K}=\omega_{r,K}=\frac{1}{2(K+1)}
583: \label{f21a}
584: \end{equation}
585: and 
586: \begin{equation}
587: \gamma_{r,K}=\frac{K}{2}
588: \label{f21b}
589: \end{equation}
590: (c.f. the $r\geq 4$ exponents in Table 2, and the $r\geq 6$ exponents in Table
591: 3).
592: 
593: The main point we wish to draw attention to is that, in all the cases
594: listed above, $\omega <1/2$ and $\alpha\leq 1/2$, and that in most cases
595: $\alpha<1/2$. Recall that $\omega=\alpha=1/2$ for periodic orbits far from
596: bifurcation. In this sense, bifurcations may be said to give rise to
597: {\em superscars}; that is, to scars that are semiclassically wider,
598: and
599:  often greater in amplitude than those associated with non-bifurcating
600: orbits. We shall demonstrate this with an explicit example in Section
601: 4.
602: 
603: The width exponent $\omega$ determines the scale for the ${\bf q}$-average in
604: $\left < \left | \psi_{n}({\bf q}) \right |^{2} \right >$ which allows the
605: fringe structure to be resolved.  Specifically, if the dimensions of the 
606: averaging range
607: scale semiclassically as $\hbar^\delta$, the fringe structure will be 
608: resolved if $\delta > \omega$, but not if $\delta < \omega$.  Recall 
609: that for the approximation to hold in which the main contribution to the
610: average comes from closed orbits in the neighbourood of periodic orbits, we 
611: must also have $\delta<1$; that is, the average must extend over many de 
612: Broglie wavelengths.
613: 
614: 
615: We also note that the scar exponents satisfy
616: \begin{equation}
617: \beta=1-\alpha-\omega,
618: \label{f24}
619: \end{equation}
620: where $\beta$ is the amplitude exponent of the fluctuations in the 
621: spectral counting function associated with the bifurcation in
622: question
623: (Berry {\em et al.} 2000). This follows from a comparison of
624: (\ref{f12}) with the corresponding expression for the counting
625: function, which, as already noted, corresponds to integrating 
626: (\ref{f12}) with respect to $y$ and multiplying by $\hbar$. It
627: generalizes the connection discussed above between the power of 
628: $\hbar$ in the trace formula for $d(E)$ and the amplitude and width
629: exponents of Bogomolny's fringes for non-bifurcating orbits.
630: 
631: \section{Moment asymptotics}
632: 
633: One way to quantify scarring effects is in terms of the moments of the 
634: wavefunctions.  Consider the case when all periodic orbits are isolated and unstable.  We define
635: \begin{equation}
636: C_{2m}(\hbar)=\frac{1}{\Delta q}\int \left (
637: \left < \left | \psi_{n}({\bf q'}) 
638: \right |^{2} \right >-\frac{\Omega({\bf q'};E)}{V(E)} \right)^{2m}d^2{\bf q'}
639: \label{f26}
640: \end{equation}
641: where the ${\bf q'}$-integral is over an $\hbar$-independent volume 
642: $\Delta q$.  Note that these are the moments not of the
643: amplitude of the wavefunction itself, but of the amplitude averaged with
644: respect to position (over a region which shrinks as $\hbar \rightarrow 0$, 
645: but which contains an increasing number of de Broglie wavelengths) and 
646: energy (semiclassically many levels).  
647: 
648: Assuming that the wavefunctions are quantum ergodic on the scale of the
649: local ${\bf q}$-average implies that $C_{2m}(\hbar) \rightarrow 0$ as 
650: $\hbar \rightarrow 0$ when $m>0$.  It follows from the fact that the 
651: fringes in (\ref{f11}) have scar 
652: exponents $\alpha=\omega=1/2$ that their individual contributions 
653: to the moments scale as
654: $\hbar^{m+1/2}$ in this limit.  The corresponding contribution from a 
655: bifurcating orbit is of the order of $\hbar^{2m\alpha + \omega}$, and so is 
656: semiclassically larger.  
657: For the bifurcations of periodic orbits with $r=1$ corresponding to the
658: cuspoid catastrophes we have that
659: \begin{equation}
660: {\tilde{\mu}}_{m,1,K}=2m\alpha_{1,K}+\omega_{1,K}=m+\frac{1}{K+2}.
661: \label{f26a}
662: \end{equation}
663: The 
664: values of these exponents for the generic bifurcations with $K=1$  and $K=2$ 
665: when $r>1$ are given in Table 4 and Table 5, for $m \le 3$.
666: 
667: 
668: \begin{table}[htb] 
669: \begin{center}
670: \begin{tabular}{|c|c|c|c|c|c|c|} \hline\hline
671: $r$ & ${\tilde{\mu}}_{1,r,1}$ & ${\overline{\mu}}_{1,r,1}$
672:     & ${\tilde{\mu}}_{2,r,1}$ & ${\overline{\mu}}_{2,r,1}$
673:     & ${\tilde{\mu}}_{3,r,1}$ & ${\overline{\mu}}_{3,r,1}$  \\ \hline
674: $2$ & $5/4$ & $7/4$ & $9/4$ & $11/4$ & $13/4$ & $15/4$  \\ 
675: $3$ & $1$   & $4/3$ & $5/3$ &  $6/3$ &  $7/3$ &  $8/3$  \\  
676: $\geq 4$    & $3/4$ & $5/4$ &  $5/4$ &  $7/4$ &  $7/4$ & $9/4$ \\
677:     \hline\hline
678: \end{tabular}
679: \end{center}
680: \caption{Values of ${\tilde{\mu}}_{m,r,1}=2m\alpha_{r,1}+\omega_{r,1}$
681: and ${\overline{\mu}}_{m,r,1}=2m \alpha_{r,1}+\omega_{r,1}+\gamma_{r,1}$
682: for the codimension-1 scar exponents listed in Table 2.}
683: \label{table4}
684: \end{table}
685: 
686: \begin{table}[htb] 
687: \begin{center}
688: \begin{tabular}{|c|c|c|c|c|c|c|} \hline\hline
689: $r$ & ${\tilde{\mu}}_{1,r,2}$ & ${\overline{\mu}}_{1,r,2}$
690:     & ${\tilde{\mu}}_{2,r,2}$ & ${\overline{\mu}}_{2,r,2}$
691:     & ${\tilde{\mu}}_{3,r,2}$ & ${\overline{\mu}}_{3,r,2}$  \\ \hline
692: $2$ & $7/6$ & $13/6$ & $13/6$ & $19/6$ & $19/6$ & $25/6$  \\ 
693: $3$ & $3/4$ &  $3/2$ & $5/4$ &     $2$ &  $7/4$ &  $5/2$  \\  
694: $4$ & $3/4$ & $5/4$ &  $5/4$ &  $7/4$ &  $7/4$ & $9/4$ \\
695: $5$ & $3/5$ & $7/5$ &    $1$ &  $9/5$ &  $7/5$ & $11/5$ \\
696: $\geq 6$ & $1/2$ & $3/2$ &  $5/6$ &  $11/6$ &  $7/6$ & $13/6$ \\ \hline\hline
697: \end{tabular}
698: \end{center}
699: \caption{Values of ${\tilde{\mu}}_{m,r,2}=2m\alpha_{r,2}+\omega_{r,2}$
700: and ${\overline{\mu}}_{m,r,2}=2m \alpha_{r,2}+\omega_{r,2}+\gamma_{r,2}$
701: for the codimension-2 scar exponents listed in Table 3.}
702: \label{table5}
703: \end{table}
704: 
705: 
706: 
707: 
708: The moments defined by (\ref{f26}) are implicitly functions of the system
709: parameters ${\bf x}$.  Averaging them with respect to ${\bf x}$ produces an
710: opportunity for a competition between the various generic bifurcations.  The
711: contribution of each bifurcation must be weighted by the associated hypervolume
712: in ${\bf x}$-space, and so scales as $\hbar^{2m\alpha + \omega + \gamma}$,
713: provided that the ${\bf x}$-average of the $2m$th power of the 
714: $\hbar$-independent term in (\ref{f15}) exists (see 
715: Berry 1977 for a discussion of this subtle point).
716: For the $r=1$ bifurcations corresponding to the cuspoid catastrophes,
717: \begin{equation}
718: {\overline{\mu}}_{m,1,K}=2m \alpha_{1,K}+\omega_{1,K}+\gamma_{1,K}=
719: m+\frac{K+1}{2}.
720: \label{f27a}
721: \end{equation}
722: When $r>1$, the values of these exponents for the generic 
723: bifurcations with $K=1$ and $K=2$
724: are also listed in Table 4 and Table 5.  
725: The bifurcation that wins the competition, and hence which determines the 
726: rate at which the ${\bf x}$-averaged moments tend to zero in the 
727: semiclassical limit, is the one for which $2m \alpha +\omega + \gamma$ is
728: minimized; that is,
729: \begin{equation}
730: \frac{1}{\hbar^{\mu_m}}\left< 
731: C_{2m}(\hbar) \right >_{\bf x}=o(\hbar^\epsilon)
732: \label{f27}
733: \end{equation}
734: for any $\epsilon < 0$, with
735: \begin{equation}
736: \mu_m={\rm min}(2m\alpha+\omega+\gamma)
737: \label{f28}
738: \end{equation}
739: where $\left < C_{2m}(\hbar) \right >_{\bf x}$ denotes the ${\bf x}$-averaged
740: moments and the minimum is with respect to the generic bifurcations.  This,
741: of course, assumes that the minimum exists.  We now argue that it does.
742: 
743: Our reasoning is based directly on that of Berry {\em et al.} (2000), where
744: the analogous problem of the moments of fluctuations in the level counting 
745: function was considered.  First, we note that if the competition is 
746: restricted to bifurcations with $r\geq 2K+2$, then
747: \begin{equation}
748: {\rm min}(2m\alpha+\omega+\gamma)={\rm min}_K 
749: \left (\frac{2m+1}{2(K+1)}+\frac{K}{2} \right ),
750: \label{f29}
751: \end{equation}
752: which exists for any $m$ and can be calculated straightforwardly.  
753: Second, it was
754: shown by Berry {\em at al.} that each bifurcation with $r\geq 2K+2$ has a
755: counterpart with $r < 2K+2$ with the property that the counterpart has 
756: a normal form with the 
757: same germ, and so the same $\alpha$ and $\omega$ exponents, but a larger
758: $\gamma$ exponent.  Hence (\ref{f29}) represents the minimum with respect to
759: all of the generic bifurcations and so
760: \begin{equation}
761: \mu_m={\rm min}_K \left (\frac{2m+1}{2(K+1)}+\frac{K}{2} \right ).
762: \label{f30}
763: \end{equation}
764: For example, $\mu_1=5/4$ (coming from $K=1$), $\mu_2=7/4$ (also coming
765: from $K=1$), and $\mu_3=13/6$ ($K=2$).  In general $\mu_m \approx 
766: \sqrt{2m+1}-1/2$.
767: 
768: It is natural to compare $\mu_m$ to the corresponding exponent for Gaussian
769: random functions, which are often taken as models of quantum chaotic 
770: wavefunctions.  In that case, the moments (\ref{f26}) are semiclassically
771: of the order of $\hbar^m$.  This follows from the results of Section IIIB
772: of Eckhardt {\em et al.} 1995, if the operator considered there is the 
773: characteristic function of the region over which the local ${\bf q}$-average
774: in (\ref{f26}) extends.  The same rate of vanishing also holds for the 
775: eigenvectors of random hermitian matrices (see Section IIIA of Eckhardt 
776: {\em et al.}).  (Readers are referred to B\"acker {\em et al.} 1998 for a 
777: detailed review of the rate of quantum ergodicity, and its characterization
778: by moments analogous to those defined by (\ref{f26})).  Crucially, we note that
779: $\mu_m \le m$ for $m>1$, and so then, if the background to the scars due to 
780: individual periodic orbits is modelled by a Gaussian random function, 
781: bifurcations 
782: dominate the semiclassical asymptotics.  The contributions from individual
783: non-bifurcating orbits are always subdominant.
784: 
785: To summarize, the exponents $\mu_m$, which are analogous to the {\em twinkling exponents} 
786: of Berry {\em et al} (2000), determine the asymptotic scaling of the
787: parameter-averaged moments $C_{2m}$ in the limit as $\hbar \rightarrow
788: 0$ when $m>1$.  Note that they are universal, that is, system independent.  Note also
789: that they are determined solely by generic bifurcation processes.  As pointed
790: out in the Introduction, these processes are characteristic of mixed
791: phases-space dynamics, and so one might expect the
792: exponents (\ref{f30}) 
793: to describe the semiclassical deviations of the irregular (in 
794: the sense of Percival 1973) eigenfunctions in mixed systems 
795: from their ergodic limit.  
796: (They do not describe the regular eigenfunctions, for which
797: the corresponding moments have a different origin, and can be calculated 
798: using the results of Berry {\em et al.} 1983.)
799: 
800: 
801: \section{Perturbed cat maps}
802: 
803: We now illustrate some of the general ideas described in the previous sections 
804: by focusing on a particular example: a family of perturbed cat maps.
805: 
806: The maps we consider are of the form
807: \begin{equation}
808: \left ( \begin{array}{l} 
809:          q_{n+1} \\
810:          p_{n+1} 
811:         \end{array} \right )=
812: \left ( \begin{array}{ll}
813:          2 & 1 \\
814:          3 & 2
815:         \end{array} \right ) \left ( \begin{array}{l}
816:                                       q_n \\
817:                                       p_{n}
818:                                      \end{array} \right )+
819: \frac{\kappa}{2 \pi} \cos(2\pi q_{n}) \left (\begin{array}{l}
820:                                   1 \\
821:                                   2
822:                                  \end{array} \right ) \mbox{mod 1},
823: \label{f40}
824: \end{equation}
825: where $q$ and $p$ are coordinates on the unit two-torus, 
826: and are taken to be a position and its conjugate momentum.  
827: These maps are Anosov systems for $\kappa \leq \kappa_{\rm 
828: max}=(\sqrt{3}-1)/\sqrt{5}\approx 0.33$; that is, for 
829: $\kappa$ in this range they are completely hyperbolic and 
830: their orbits are conjugate to those of the map with 
831: $\kappa=0$ (i.e. there are no bifurcations).  Outside this range, 
832: bifurcations occur, stable islands are created, and the dynamics becomes mixed 
833: (see, for example, Berry {\em et al.} 1998, where these systems were used 
834: to demonstrate the influence of periodic orbit bifurcations on 
835: long-range spectral statistics).
836: 
837: The quantization of maps like (\ref{f40}) was developed by 
838: Hannay \& Berry (1980), when $\kappa=0$, and Bas\'ilio de 
839: Matos \& Ozorio de Almeida (1995) for non-zero $\kappa$.  
840: The quantum kinematics associated with a phase space that 
841: has the topology of a two-torus restricts Planck's constant 
842: to taking inverse integer values.  The integer in question, 
843: $N$, is the dimension of the Hilbert space of admissible 
844: wavefunctions.  With doubly  periodic boundary conditions 
845: (see, for example, Keating {\em et al.} 1999), these 
846: wavefunctions in their position representation have support 
847: at points $q=Q/N$, where $Q$ takes integer values between 1 
848: and $N$.  They may thus be represented by $N$-vectors with 
849: complex components.  The quantum dynamics is then generated 
850: by an $N \times N$ unitary matrix ${\bf U}$ whose action on 
851: the wavefunctions reduces to (\ref{f40}) in the classical 
852: limit; for example
853: \begin{equation}
854: U_{Q_1,Q_2}=\frac{1}{\sqrt{iN}}\exp \left [ \frac{2\pi i} 
855: {N} (Q_1^2-Q_1Q_2+Q_2^2)+\frac{iN}{2\pi}\kappa\sin (2\pi Q_1/N) \right ].
856: \label{f41}
857: \end{equation}
858: This matrix plays the role of the Green function of the 
859: time-dependent Shr\"odinger equation for flows.  
860: 
861: Denoting the eigenvalues of ${\bf U}$ by 
862: ${\rm e}^{i\theta_n}$, and the corresponding eigenfunctions by 
863: $\Psi_n(Q)$, we have that
864: \begin{equation}
865: \sum_{n}\left |\Psi_{n}(Q)
866: \right |^{2}\delta_{\varepsilon}(\theta-\theta_{n}) 
867: =1+\rpart \sum^{\infty}_{k=1}U_{Q,Q}^{k}\exp{(-i\theta k 
868: -\varepsilon k)}
869: \label{f42}
870: \end{equation}
871: where
872: \begin{equation}
873: \delta_{\varepsilon}(x)=\frac{1-e^{-\varepsilon}\cos{x}}{1+e^{-2\varepsilon}-2e^{-\varepsilon}\cos{x}}
874: \label{f43}
875: \end{equation}
876: is a periodized, Lorentzian-smoothed $\delta$-function of 
877: width $\varepsilon$ (Keating 1991).  Equation (\ref{f42}) is the analogue 
878: for quantum maps of (\ref{f2}).  The left-hand side 
879: corresponds, approximately, to $N$ times the local 
880: $n$-average (over a range of size of the order of 
881: $\varepsilon$) of $\left | \Psi_n(Q) \right |^2$, and so, 
882: dividing both sides by $N$ and averaging smoothly with 
883: respect to $Q$ (for example, taking the convolution with a 
884: normalized Gaussian) over a range large compared to a de Broglie 
885: wavelength ($\Delta Q=1$) but small compared to $N$,
886: \begin{equation}
887: \left < |\Psi_{n}(Q)|^{2} \right > \approx \frac{1}{N}+\frac{1}{N}
888: \rpart \sum^{\infty}_{k=1} \left< U_{Q,Q}^{k} \right >_Q\exp{(-i\theta k 
889: -\varepsilon k)}.
890: \label{f44}
891: \end{equation}
892: Here $\left< \dots \right >$ denotes a combination of the 
893: $n$-average and the $Q$-average $\left < \dots \right >_Q$.  
894: 
895: In our computations we took $\varepsilon$ large enough so that the 
896: dominant contributions to (\ref{f42}) and (\ref{f44}) come 
897: from the $k=1$ terms in the sums on the right, and so, for example, we 
898: may substitute (\ref{f41}) directly into (\ref{f44}).  In 
899: the semiclassical limit, as $N \rightarrow \infty$, the 
900: $Q$-average selects regions close to stationary points of 
901: the phase of (\ref{f41}), which we denote by $Q/N=q_f$.  
902: These stationary points coincide with the positions of the 
903: fixed points of the classical map (\ref{f40}); that is, 
904: $q_f$ satisfies
905: \begin{equation}
906: q_f=\frac{1}{2} \left ( j-\frac{\kappa}{2\pi}\cos (2\pi q_f) \right)
907: \label{f45a}
908: \end{equation}
909: for integers $j$ such that $0 \leq q_f < 1$ (see, for 
910: example, Boasman \& Keating 1995).  Expanding the phase of (\ref{f41})
911: around $q_f$ gives, up to cubic terms,
912: \begin{equation}
913: U_{Q,Q}\approx \frac{1}{\sqrt{iN}}\exp \left [ 2\pi i S_f+ \pi iN
914: (2-\kappa \sin (2\pi q_f))y^2 -\frac{2\pi^2 iN}{3}\kappa \cos (2\pi q_f) 
915: y^3 \right ],
916: \label{f45}
917: \end{equation}
918: where
919: \begin{equation}
920: y=\frac{Q}{N}-q_f
921: \label{f46}
922: \end{equation}
923: and $2\pi S_f$ denotes the phase evaluated at $q_f$.
924: 
925: Provided that $2-\kappa \sin (2\pi q_f) \ne 0$, this approximation is 
926: dominated by
927: the quadratic term in the exponent when $y$ is small.  It thus describes
928: complex-Gaussian fringes around the classical fixed points with a length-scale
929: (in terms of $y$)
930: of the order of $N^{-1/2}$.  These are the analogues of Bogomolny's fringes.
931: They will be resolved if the local $Q$-average is over a range that is small
932: compared to $N^{1/2}$ (but which still grows as $N \rightarrow \infty$).
933: 
934: For the example being considered here, when $\kappa < \kappa^{*} \approx 
935: 5.94338$ 
936: the two values of $j$ in (\ref{f45a}), $j=0$ and $j=1$, each give rise 
937: to a single unstable fixed
938: point for which the condition $2-\kappa \sin (2\pi q_f) \ne 0$ is satisfied.
939: In Figure 1, we plot the left hand side of (\ref{f42}) when $\kappa=3$, with
940: $\varepsilon=2.2$ and for $N=1597$.  The structure around the fixed points
941: is clearly visible, and is most easily seen by applying the local $Q$-average
942: (in this case, making a convolution with a normalized Gaussian of width 
943: $0.02$).  
944: 
945: It is at bifurcations that $2-\kappa \sin (2\pi q_f) = 0$.  Then the
946: quadratic term in (\ref{f45}) vanishes, and the fringe structure comes instead
947: from the cubic term.  It thus has a $y$-length-scale of the order of 
948: $N^{-1/3}$.  The amplitude is the
949: same as in the case of isolated fixed points ($N^{-1/2}$ in the contribution to
950: $\left < |\Psi_{n}(Q)|^{2} \right >$).  In the language of Section 2,
951: this corresponds to a codimension-one bifurcation of a periodic orbit with
952: $r=1$ (a tangent bifurcation).
953: 
954: In our example, the first bifurcation occurs when $\kappa=\kappa^{*}$.  At this
955: parameter value, two new degenerate solutions of (\ref{f45a}) 
956: appear,  for both $j=0$ and $j=1$, corresponding to the birth of a 
957: pair of fixed points, one stable and 
958: the other unstable.  In Figure 2 we plot the left hand side of (\ref{f42}) with
959: $\varepsilon=2.2$ and $N=1597$, as above, but now with $\kappa=\kappa^{*}$.
960: It is apparent that the scars around the two bifurcations, at positions 
961: $q=0.05$ and $q=0.44$, are wider than those around the two non-bifurcating
962: fixed points, at positions $q=0.69$ and $q=0.81$, and that 
963: they are also wider than those shown in Figure 1.  
964: In Figure 3 we plot the left hand side of 
965: (\ref{f42}) close to the bifurcation point at $q=0.44$, together with the
966: approximation (\ref{f45}), which clearly captures the details of
967: the associated fringe structure.
968: 
969: It is straightforward now to deduce the semiclassical 
970: scaling with $N$ of the moments
971: \begin{equation}
972: C_{2m}(N)=N^{2m}\sum_{Q=1}^N 
973: \left ( \left < |\Psi_{n}(Q) |^{2} \right > - \frac{1}{N} \right )^{2m}.
974: \label{f47}
975: \end{equation}
976: The arguments of Section 3 suggest that 
977: $C_{2m}$ is of the order of $N^{-m+w}$, where $w=1/2$ away from 
978: bifurcations and $w=2/3$ at the bifurcation ($w$ is one plus the width 
979: exponents deduced from (\ref{f45}), because those were for $y$ rather than
980: $Ny$, as we need here).  In Figure 4 we plot $\log C_{2}$ against $\log N$ when
981: $\kappa=3$.  The fact that the points lie on a straight line confirms that
982: there is a power-law scaling; furthermore, the gradient 
983: is close to $-1/2$, as expected.  In
984: Figure 5, we make the same plot for $\kappa=\kappa^{*}$.  In this case the
985: gradient is close to the expected value of $-1/3$ (a possible explanation for 
986: the deviation is that for the range of values of $N$ shown, the 
987: bifurcation exponent is
988: contaminated by the contributions from the non-bifurcating periodic orbits).
989: Finally, in Figure 6 we plot
990: \begin{equation}
991: g(m)=\lim_{N\rightarrow \infty}\frac{\log{C_{2m}(N)}}{\log N},
992: \end{equation}
993: calculated numerically from the gradients of best-fitting straight lines
994: to plots like those in Figures 4 and 5.  For both $\kappa=3$ and $\kappa=
995: \kappa^{*}$ the results are in accord with the
996: scaling law suggested above. 
997: 
998: We emphasize that these numerical computations illustrate the influence of
999: one individual bifurcation only.  They do not test the competition which
1000: would result from averaging over a parameter range that contains many 
1001: different generic bifurcations, and which the analysis of Section 3 suggests
1002: has a universal outcome for the moment exponents.
1003: 
1004: \section{Acknowledgements}
1005: 
1006: It is a pleasure to acknowledge stimulating discussions with Arnd B\"acker,
1007: John Hannay and Jens Marklof, and comments on the manuscript by Sir Michael
1008: Berry.  SDP wishes to thank BRIMS for financial support, and 
1009: BRIMS and the School of Mathematics at the University of Bristol for 
1010: hospitality during the period when this work was carried out.
1011: 
1012: 
1013: %\begin{thebibliography}
1014: 
1015: \ \
1016: 
1017: \noindent {\Large References}
1018: 
1019: \ \
1020: 
1021: \noindent Agam, O. \& Fishman, S. 1994 Semiclassical criterion for 
1022: scars in wave-functions of chaotic systems.  {\em Phys. Rev. Lett} 
1023: {\bf 73}, 806-809.
1024: 
1025: \ \ 
1026: 
1027: \noindent Bas\'ilio de Matos, M. \& Ozorio de Almeida, A.M. 1995 Quantization
1028: of Anosov maps.  {\em Ann. Phys.} {\bf 237}, 46-65.
1029: 
1030: \ \
1031: 
1032: \noindent Arnold, V.I. 1978 {\em Mathematical Methods in Classical
1033: Mechanics.} Springer.
1034:  
1035: \ \
1036: 
1037: \noindent B\"acker, A., Schubert, R. \& Stifter, P. 1998 Rate of quantum 
1038: ergocity in Euclidean billiards.  {\em Phys. Rev.} E {\bf 57}, 5425-5447.
1039: 
1040: \ \
1041: 
1042: \noindent Berry, M.V. 1977 Focusing and twinkling: critical exponents from
1043: catastrophes in non-Gaussian random short waves.  {\em J. Phys.} A {\bf 10},
1044: 2061-2081.
1045: 
1046: \ \
1047: 
1048: \noindent Berry, M.V. 1983 Semiclassical mechanics of regular and irregular 
1049: motion.  In {\em Les Houches Lecture Series} 
1050:  (ed. G. Iooss, R.H.G. Helleman \& R. Stora), vol. 36, pp. 171-271 
1051: Amsterdam: North Holland.
1052: 
1053: \ \
1054: 
1055: \noindent Berry, M.V. 1989, Quantum scars of classical closed orbits in phase 
1056: space. {\em Proc. R. Soc. Lond.} A  {\bf 243}, 219-231.
1057: 
1058: \ \
1059: 
1060: \noindent Berry, M.V. 2000 Spectral twinkling. {\em New Directions in
1061: Quantum Chaos.} Proceedings of the International School of Physics
1062: ``Enrico Femi'', 45-63. Italian Physical Society.
1063: 
1064: \ \
1065: 
1066: \noindent Berry, M.V., Hannay, J.H. \& Ozorio de Almeida, A.M. 1983 
1067: Intensity moments of semiclassical wavefunctions.  {\em Physica} D {\bf 8},
1068: 229-242.
1069: 
1070: 
1071: \ \
1072:  
1073: \noindent Berry, M.V,  Keating, J.P \& Prado, S.D. 1988 Orbit bifurcations and
1074: spectral statistics. {\em J. Phys.} A {\bf 31}, L245-254.
1075: 
1076: \ \
1077: 
1078: \noindent Berry, M.V., Keating, J.P. \& Schomerus, H. 2000 Universal twinkling
1079: exponents for spectral fluctuations associated with mixed chaology. 
1080: {\em Proc. R. Soc. Lond.} A {\bf 456}, 1659-1668.
1081: 
1082: \ \
1083: 
1084: \noindent Berry, M.V. \& Tabor, M. 1976 Closed orbits and the regular bound 
1085: spectrum. {\em Proc. R. Soc. Lond.} A {\bf 349}, 101-123.
1086: 
1087: \ \
1088: 
1089: \noindent Boasman, P.A. \& Keating, J.P. 1995 Semiclassical asymptotics of
1090: perturbed cat maps.  {\em Proc. R. Soc. Lond.} A {\bf 449}, 629-653.
1091: 
1092: \ \
1093: 
1094: \noindent Bogomolny, E.B. 1988 Smoothed wavefunctions of chaotic quantum systems.
1095: {\em Physica} D {\bf 31}, 169-189.
1096: 
1097: \ \
1098: 
1099: \noindent Colin de Verdi\`ere, Y. 1985 Ergodicit\'e et fonctions propres du 
1100: laplacien.  {\em Commun. Math. Phys.} {\bf 102}, 497-502.
1101: 
1102: \ \
1103: 
1104: \noindent Eckhardt, B., Fishman, S., Keating, J.P., Agam, O., Main, J., 
1105: \& M\"uller, K. 1995 Approach to ergodicity in quantum wave functions.  
1106: {\em Phys. Rev.} E {\bf 52}, 5893-5903.
1107: 
1108: \ \
1109: 
1110: \noindent Fishman, S., Georgeot, B. \& Prange, R. E. 1996 
1111: Fredholm method for scars.  {\em J. Phys.} A {\bf 29}, 919-937.
1112: 
1113: \ \
1114: 
1115: \noindent Gutzwiller, M.C. 1971 Periodic orbits and classical quantization 
1116: conditions. {\em J. Math. Phys.} {\bf 12}, 343-358.
1117: 
1118: \ \
1119: 
1120: \noindent Gutzwiller, M.C. 1990 {\em Chaos in Classical and Quantum Mechanics} (New York: Springer).
1121: 
1122: \ \
1123: 
1124: \noindent Hannay, J.H. \& Berry, M.V. 1980 Quantization of linear maps on
1125: the torus -- Fresnel diffraction by a periodic grating.  {\em Physica} D
1126: {\bf 1}, 267-290.
1127: 
1128: \ \
1129: 
1130: \noindent Heller, E.J. 1984 Bound state eigenfunctions of classically 
1131: chaotic Hamiltonian systems - scars of periodic orbits.  
1132: {\em Phys. Rev. Lett.} {\bf 53}, 1515-1518.
1133: 
1134: \ \
1135: 
1136: \noindent Kaplan, L. 1999 Scars in quantum chaotic wavefunctions.  
1137: {\em Nonlinearity} {\bf 12}, R1-R40.
1138: 
1139: \ \
1140: 
1141: \noindent Keating, J.P. 1991 The cat maps: quantum mechanics and classical
1142: motion.  {\em Nonlinearity} {\bf 4}, 309-341.
1143: 
1144: \ \
1145: 
1146: \noindent Keating, J.P., Mezzadri, F. \& Robbins, J.M. 1999 Quantum 
1147: boundary conditions for torus maps.  {\em Nonlinearity} {\bf 12}, 579-591.
1148: 
1149: \ \
1150: 
1151: \noindent McDonald, S.W. 1983 {\em Lawrence Berkeley Laboratory Report} LBL -
1152:  14837.
1153: 
1154: \ \
1155: 
1156: \noindent Meyer, K.R. 1970 Generic bifurcations of periodic points.
1157: {\em Trans. Am. Math. Soc.} {\bf 149}, 95-107.
1158: 
1159: \ \
1160: 
1161: \noindent Meyer, K.R. 1986 Bibliographic notes on generic bifurcations
1162: in Hamiltonian Systems.
1163: {\em Contemp. Math.} {\bf 56}, 373-381.
1164: 
1165: \ \
1166: 
1167: \noindent Ozorio de Almeida, A.M. 1988 {\em Hamiltonian Systems: Chaos
1168: and Quantization.} Cambridge University Press.
1169: 
1170: \ \
1171: 
1172: \noindent Ozorio de Almeida, A.M. \& Hannay, J.H. 1987 Resonant periodic orbits
1173: and the semiclassical energy spectrum. {\em J. Phys} A {\bf 20}, 
1174: 5873-5883.
1175: 
1176: \ \
1177: 
1178: \noindent Percival, I.C. 1973 Regular and irregular spectra.  
1179: {\em J. Phys.} B {\bf 6}, L229-L232.
1180: 
1181: \ \
1182: 
1183: \noindent Schomerus, H. 1998 Periodic orbits near bifurcations of codimension
1184: two: Classical mechanics, semiclassics and Stokes transitions. {\em
1185: J. Phys.} A {\bf 31}, 4167-4196.
1186: 
1187: \ \
1188: 
1189: \noindent Schomerus, H. \& Sieber, M. 1997 Bifurcations of periodic orbits
1190: and uniform approximations. {\em J. Phys.} A {\bf 30}, 4537-4562.
1191: 
1192: \ \
1193: 
1194: \noindent Shnirelman, A.I. 1974 Ergodic properties of eigenfunctions (in 
1195: Russian).  {\em Usp. Math. Nauk} {\bf 29}, 181-182.
1196: 
1197: \ \
1198: 
1199: \noindent Sieber, M. 1996 Uniform approximation for bifurcations of 
1200: periodic orbits
1201: with high repetition numbers. {\em J. Phys.} A {\bf 29}, 4715-4732. 
1202: 
1203: \ \
1204: 
1205: \noindent Sieber, M. \& Schomerus, H. 1998 Uniform approximations for 
1206: period-quadrupling bifurcations. {\em J. Phys.} A {\bf 31}, 165-183.
1207: 
1208: \ \
1209: 
1210: \noindent Tomsovic, S., Grinberg, M. \& Ullmo, D. 1995 Semiclassical trace 
1211: formulas of near-integrable systems: Resonances. {\em Phys. Rev. Lett.}
1212: {\bf 75}, 4346-4349.
1213: 
1214: \ \
1215: 
1216: \noindent Ullmo, D., Grinberg, M. \& Tomsovic, S. 1996 Near-integrable systems:
1217: Resonances and semiclassical trace formulas. {\em Phys. Rev.} E 
1218: {\bf 54}, 136-152.
1219: 
1220: \ \
1221: 
1222: \noindent Zelditch, S. 1987 Uniform distribution of eigenfunctions on compact 
1223: hyperbolic surfaces.  {\em Duke Math. J.} {\bf 55}, 919-941.
1224: 
1225: %\end{thebibliography}
1226: 
1227: \newpage 
1228:  
1229: \centerline{FIGURES}
1230: 
1231: 
1232: \begin{figure}[htb]
1233:   \begin{center}
1234:    \leavevmode
1235:    \psfig{figure=wave_k3.eps,width=16.0cm,angle=270}
1236:   \end{center}
1237: \caption{$\sum_{n}\left | \psi_{n}(Q) \right |^{2}\delta_{\varepsilon}
1238: \left (\theta-\theta_{n} \right )-1$, with $\varepsilon=2.2$, $\kappa=3$
1239: and $N=1597$ (circles connected by dotted lines).  Also shown is a
1240: convolution of the data with a normalized Gaussian of width 0.02 (bold line).
1241: The positions of the fixed points are $q=0.65$ and $q=0.85$.}
1242: \label{fig:k3}
1243: \end{figure}
1244: 
1245: \begin{figure}[htb]
1246:   \begin{center}
1247:    \leavevmode
1248:    \psfig{figure=wave_kbif.eps,width=16.0cm,angle=270}
1249:   \end{center}
1250: \caption{$\sum_{n}\left | \psi_{n}({Q}) \right |^{2}\delta_{\varepsilon}
1251: \left (\theta-\theta_{n} \right )-1$, with 
1252: $\varepsilon=2.2$, $\kappa=\kappa^{*}$
1253: and $N=1597$ (circles connected by dotted lines).  Also shown is a
1254: convolution of the data with a normalized Gaussian of width 0.02 (bold line).
1255: There are unstable fixed points at $q=0.69$ and $q=0.81$, and bifurcations
1256: at $q=0.05$ and $q=0.44$.}
1257: \label{fig:kbif}
1258: \end{figure}
1259: 
1260: \begin{figure}[htb]
1261:   \begin{center}
1262:    \leavevmode
1263:    \psfig{figure=local.eps,width=16.0cm,angle=270}
1264:   \end{center}
1265: \caption{$\sum_{n}\left | \psi_{n}({Q}) \right |^{2}\delta_{\varepsilon}
1266: \left (\theta-\theta_{n} \right )-1$, with $\varepsilon=2.2$, $\kappa=\kappa^{*}$
1267: and $N=1597$, as in Figure 2, in the neighbourhood of the bifurcation at 
1268: $q=0.44$ (circles connected by dotted lines).  Also shown is the local 
1269: approximation obtained by substituting (\ref{f45}) into (\ref{f42}) 
1270: (bold line).}
1271: \label{fig:local}
1272: \end{figure}
1273: 
1274: \begin{figure}[htb]
1275:   \begin{center}
1276:    \leavevmode
1277:    \psfig{figure=slope3.eps,width=16.0cm,angle=270}
1278:   \end{center}
1279: \caption{$\log C_2$, calculated using a local $Q$-average of size 
1280: $0.02N^{1/2}$, 
1281: plotted against $\log N$ when $\kappa=3$ (circles). Also shown is a 
1282: best-fitting straight line, which has gradient -0.50.}
1283: \label{fig:slope3}
1284: \end{figure}
1285: 
1286: \begin{figure}[htb]
1287:   \begin{center}
1288:    \leavevmode
1289:    \psfig{figure=slopebif_s.eps,width=16.0cm,angle=270}
1290:   \end{center}
1291: \caption{$\log C_2$, calculated using a local $Q$-average of size 
1292: $0.02N^{1/2}$, 
1293: plotted against $\log N$ when $\kappa=\kappa^{*}$ (circles). Also shown is a 
1294: best-fitting straight line, which has gradient -0.29.}
1295: \label{fig:slopebif}
1296: \end{figure}
1297: 
1298: \begin{figure}[htb]
1299:   \begin{center}
1300:    \leavevmode
1301:    \psfig{figure=higherm.eps,width=16.0cm,angle=270}
1302:   \end{center}
1303: \caption{$g(m)$ plotted against $m$, for $\kappa=3$ (full circles), and
1304: $\kappa=\kappa^{*}$ (open circles).  Also shown are the best fitting straight
1305: lines: $g=-0.96m+0.57$ and $g=-1.01m+0.72$, respectively.}
1306: \label{fig:higherm}
1307: \end{figure}
1308: 
1309: \end{document}
1310: 
1311: 
1312: 
1313: 
1314: 
1315: 
1316: