1: \documentstyle[aps,psfig]{revtex}
2: \tightenlines
3: \newlength{\figwidth}
4: \setcounter{equation}{0}
5: \setlength{\figwidth}{3.4in}
6:
7:
8: \title{Exact and semiclassical Husimi distributions of Quantum Map
9: Eigenstates}
10: \author{M. Saraceno$^1$ and A. G. Monastra$^2$}
11: \address{$^1$Departamento de F\'{\i}sica, Comisi\'on Nacional de Energ\'{\i}a
12: At\'omica (CNEA), \\ Av. del Libertador 8250, 1429 Buenos Aires, Argentina
13: \\ $^2$Laboratoire de Physique Th\'eorique et Mod\`eles Statistiques,
14: B\^at. 100, \\ 91405 Orsay Cedex, France}
15:
16: \begin{document}
17: \maketitle {\begin{abstract} The projector onto single quantum map eigenstates
18: is written only in terms of powers of the evolution operator, up to half the
19: Heisenberg time, and its traces. These powers are semiclassically
20: approximated, by a complex generating function, giving the Husimi distribution
21: of the eigenstates. The results are tested on the Cat and Baker maps.
22: \end{abstract}}
23:
24:
25: \section{Introduction}
26: The relation between the classical invariant structures of a dynamical
27: system and its
28: quantum counterparts, the eigenenergies and eigenstates, is not
29: completely understood and
30: is an intense subject of investigation. When the system is integrable
31: the eigenenergies
32: are determined by the well-known Bohr-Sommerfeld or EBK quantization
33: rules and the
34: eigenfunctions have a very intuitive phase space structure: they are
35: essentially
36: localized on the quantized tori and they present characteristic quantum
37: features
38: --interference fringes for the Wigner function \cite{berry1} or zeros
39: for the Husimi
40: function \cite{voros}-- away from them. No such clear picture is
41: available for the
42: eigenstates of chaotic systems and much work
43: \cite{heller,bogo1,vergini,nonnenmacher} has gone
44: into trying to disentangle the purely statistical aspects of the
45: eigenfunctions from the
46: specific dynamical features such as scars and localization.
47:
48: Although a lot of recent work deals with the study of the statistical
49: aspects of chaotic
50: wavefunctions \cite{mirlin} we concentrate here on the
51: purely dynamical
52: aspects and explore how well the semiclassical formulae can reproduce
53: the details of
54: individual eigenfunctions.
55:
56: With regards to spectral properties the theoretical tool for this
57: analysis is the
58: Gutzwiller trace formula \cite{gutzw} relating the spectral density to
59: the periodic
60: orbits. As is well known, this relationship is fraught with
61: uncertainties both due to
62: the approximations involved, as to the fact that it diverges where it is
63: supposed to
64: locate the single eigenenergies and to the sheer complexity of adding
65: the contributions
66: of an exponentially increasing number of periodic orbits. Some of these
67: difficulties can
68: be overcome by resummation techniques \cite{bogo-keat} that limit the
69: exponential
70: increase by partially restoring the unitarity of the propagator, lost in
71: the
72: semiclassical approximation.
73:
74: In the present work we extend these resummation methods to the
75: calculation of
76: eigenfunctions of chaotic maps. We utilize Fredholm theory to construct
77: a projector on
78: single eigenstates whose expression is directly amenable to
79: semiclassical
80: approximations. These approximations share the same advantages - and the
81: drawbacks- of the resummation techniques available for spectral properties.
82:
83: In section 2 we construct explicitly the projector on single
84: eigenstates. This is
85: achieved exactly in terms of a finite sum on powers of the quantum map
86: propagator and
87: its traces. The cut-off time is half the Heisenberg time due to the
88: explicit imposition
89: of unitarity. In section 3 these formulae are implemented
90: semiclassically by a)
91: approximating the traces in terms of actions and instabilities of
92: periodic orbits in the
93: usual way, and b) by computing the propagator in the coherent state
94: representation in
95: terms of specially devised complex generating functions.
96:
97: The final result is a semiclassical expression for the Husimi function
98: of an eigenstate
99: in terms of properties of a finite set of periodic orbits.
100:
101: In section 4 we apply the method to the cat map --where the
102: semiclassical approximation
103: is exact -- and to the baker map where we can test the approximation.
104:
105: The formulae are here applied as a test to simple maps, but they can be
106: easily adapted
107: to more realistic Hamiltonian systems by selecting an appropriate
108: Poincar\'e section
109: \cite{sara-simo}.
110:
111:
112:
113: \section {Projector on single eigenstates}
114:
115: For a map, the density of Floquet states can be expressed in the usual
116: way in terms of
117: traces by the formula
118:
119: \begin{equation}
120: d(\epsilon) = \frac{2\pi}{N}+\sum_{n=1}^{\infty} [tr \hat U^n e^{- i
121: \epsilon n} ] +
122: C.C.
123: \end{equation}
124: $1/N$ plays the role of Plank's constant $h$, $n$ is the discrete time,
125: and $\hat U$ is
126: the unitary quantization of the map. The density of states results in a
127: $2\pi$ periodic
128: function with unit $\delta$-spikes at the eigenangles $\epsilon_k$,
129: defined by
130:
131: \begin{equation}
132: \hat U | \psi_k \rangle = e^{i \epsilon_k} | \psi_k \rangle \ .
133: \end{equation}
134:
135: Semiclassical approximations using this formula require the calculation
136: of traces of
137: powers of the map for large values of $n$ which are in turn related to
138: classical orbits
139: of the map by the Gutzwiller-Tabor \cite{gutzw} trace formula
140:
141: \begin{equation}
142: b_n = tr \hat U^n \approx \sum_{\gamma} A_{\gamma} \exp \left[
143: i(S_{\gamma}/\hbar -
144: \mu_{\gamma} \pi /2) \right] \label{gutz} \ ,
145: \end{equation}
146: where $\gamma$ label the $n$-periodic orbits of the map, $S_{\gamma}$ is
147: the action,
148: $\mu_{\gamma}$ is the Maslov index, and $A_{\gamma}$ is a coefficient
149: related to the
150: monodromy matrix. The approximation is valid for fixed $n$ in the limit
151: as $N
152: \rightarrow \infty$ (but not vice-versa).
153:
154: A more efficient scheme starts from the spectral determinant
155:
156: \begin{equation}
157: P(s) = \det (\hat I - s \hat U)
158: \end{equation}
159: which vanishes at $s = s_k=e^{ - i \epsilon_k}$ and which can be
160: expanded as
161:
162: \begin{equation}
163: P(s) = \sum_{n=0}^N \beta_n s^n \ .
164: \end{equation}
165: The coefficients $\beta_n$ can be computed recursively from the traces
166: as
167:
168: \begin{equation}
169: \beta_n = -\frac{1}{n} \sum_{j=1}^n \beta_{n-j} b_j \label{betarec}
170: \end{equation}
171: and therefore when computed semiclassically as in (3) they are expressed in
172: terms of linear
173: combinations of periodic orbits (pseudo orbits \cite{pseudo}) of period
174: up to $n$.
175:
176: The advantage over the calculation in term of traces is that only $N$
177: coefficients need
178: to be calculated. Moreover as a consequence of unitarity the symmetry
179:
180: \begin{equation}
181: \beta_{N-j} = (-1)^N \det \hat U ~ \bar{\beta}_j \label{rever}
182: \end{equation}
183: can be imposed, so that only $N/2$ coefficients are needed and therefore
184: only periodic
185: trajectories for times up to $N/2$ (i.e. half the Heisenberg time) are
186: involved for a
187: semiclassical calculation. This is now an optimal encoding of the
188: eigenvalues in terms
189: of traces, with the $N$ real phases $\epsilon_k$ expressed exactly in
190: terms of the $N/2$
191: complex coefficients $\beta_n$
192:
193: It is convenient to define another function \cite{bogo2}, proportional
194: to the spectral
195: determinant, which has the same roots. Utilizing the symmetry
196: (\ref{rever}), for an
197: even $N$, this function can be written
198:
199: \begin{equation}
200: Z(s) = \frac{P(s)}{\beta_{N/2} ~ s^{N/2}} = 1 + \frac {1}{\beta_{N/2}}
201: \sum_{n=0}^{N/2-1} \beta_n ~ s^{n-N/2} + \frac{1}{\bar{\beta}_{N/2}}
202: \sum_{n=0}^{N/2-1}
203: \bar{\beta}_n ~ s^{N/2-n} \ .
204: \end{equation}
205:
206: When $s$ is on the unit circle $1/s=\bar s$, and then the second sum is
207: the complex
208: conjugate of the first. Then the function $Z(s)$ becomes real on
209: $|s|=1$.
210:
211:
212: Unitarity also implies that the spectral determinant has all its zeros
213: on the unit
214: circle. However when the coefficients are computed semiclassically this
215: is no longer
216: guaranteed. If the symmetry (\ref{rever}) is imposed the resulting
217: spectral determinant
218: becomes self-inversive, which is a necessary -but by no means
219: sufficient- condition for
220: unitarity \cite{bbl}. Self-inversive polynomials have the following properties
221: \cite{marsden}:
222:
223: 1) All zeros are either on the unit circle or symmetric with respect to
224: it. Thus as a
225: function of a continuous parameter the zeros of a self-inversive
226: polynomial can only
227: leave the unit circle in pairs by degenerating on it.
228:
229: 2) If $P(s)$ has only single roots then $dP/ds$ has no zeros on the unit
230: circle.
231:
232:
233:
234:
235:
236:
237:
238: \subsection {Green Operator}
239:
240:
241: Very similar techniques can be implemented for the calculation of matrix
242: elements and
243: therefore of eigenfunctions. To obtain the eigenfunctions of the unitary
244: matrix $\hat
245: U$, we define a Green operator
246:
247: \begin{equation}
248: \hat G (s) = \frac {f(\hat U)}{\hat I - s \hat U} \ ,
249: \end{equation}
250: which, as an analytic function of $s$, has poles at $s = s_k = e^{ - i
251: \epsilon_k}$, and
252: the residues are proportional to the projectors $\hat P_k = |\psi_k
253: \rangle \langle
254: \psi_k |$. To obtain normalized projectors, we instead define a {\it
255: normalized} Green
256: operator
257:
258: \begin{equation}
259: \hat g (s) = \frac {\hat G (s)}{tr[\hat G (s)]} \ .
260: \end{equation}
261:
262: At the poles, the singularities cancel and we obtain $\hat g (s_k) =
263: \hat P_k$. $f(\hat
264: U)$ is left arbitrary for the moment, and specific choices -made below-
265: will be used to
266: obtain formulae with different properties.
267:
268: We can expand the preceding expression utilizing the transpose cofactor
269: matrix to write
270: the Green operator
271:
272: \begin{equation}
273: \hat g (s) = \frac {f(\hat U) ~ C^t (\hat I - s \hat U)}{ tr[f(\hat U) ~
274: C^t (\hat I - s
275: \hat U)]} \label{gs} \ .
276: \end{equation}
277: The transpose cofactor matrix can be written as a finite power series in
278: $s$
279:
280: \begin{eqnarray}
281: C^t (\hat I - s \hat U) &=& \sum_{n=0}^{N-1} s^n \hat X _n \ , \\
282: \hat X _n &=& \sum_{i=0}^n \beta_i ~ \hat U ^{n-i} \label{xn} \ .
283: \end{eqnarray}
284:
285: By the unitarity of $\hat U$, the $\hat X _n$ matrices satisfy the
286: following symmetry
287:
288: \begin{equation}
289: \hat X _{N-j} = - \hat U^{\dagger} (-1)^N \det \hat U ~ \hat X
290: _{j-1}^{\dagger} \ .
291: \label{reverx}
292: \end{equation}
293:
294:
295: Utilizing this symmetry two choices of $f(\hat U)$ lead to expansions
296: with useful
297: properties:
298:
299: \noindent a) For $f(\hat U)= \hat U$, for even $N$, the corresponding
300: Green function is
301:
302: \begin{equation}
303: \hat g_a (s) = \frac{ \sum_{n=0}^{N/2-1} s^n ~ \hat U \hat X_n - \det
304: \hat U ~ s^{N-1}
305: \sum_{n=0}^{N/2-1} s^{-n} ~ \hat X^{\dagger}_n } { \sum_{n=0}^{N/2-1}
306: s^n ~ tr
307: \left(\hat U \hat X_n \right) - \det \hat U ~ s^{N-1} \sum_{n=0}^{N/2-1}
308: s^{-n} ~ tr
309: \left(\hat X^{\dagger}_n \right)} \ .
310: \end{equation}
311:
312: Replacing the expansion of the $\hat X_n$ operators in powers of $\hat
313: U$, we obtain
314:
315: \begin{equation}
316: \hat g_a(s) = \frac{\sum_{i=0}^{N/2-1} c_i (s) ~ \hat U ^{i+1} - \det
317: \hat U ~ s^{N-1}
318: \sum_{i=0}^{N/2-1} \bar{c_i} (1/\bar s) ~ \hat U^{\dagger ~ i} }
319: {\sum_{i=0}^{N/2-1}
320: c_i (s) ~ tr \left(\hat U ^{i+1} \right) - \det \hat U ~ s^{N-1}
321: \sum_{i=0}^{N/2-1}
322: \bar{c_i} (1/\bar s) ~ tr \left(\hat U^{\dagger ~ i} \right) }
323: \label{greenfinal} \ ,
324: \end{equation}
325: in which the projector is expressed as a finite combination of the first
326: $N/2$ powers of
327: the propagator.
328:
329: The coefficients $c_i$ are obtained from the $\beta_n$
330:
331: \begin{equation}
332: c_i (s) = \sum_{n=i}^{N/2-1} \beta_{n-i} ~ s^n \label{cis} \ .
333: \end{equation}
334:
335: This normalized Green operator has no singularities as $s$ goes around
336: on the unit
337: circle. This can be demonstrated because the denominator is equal to
338: $-dP/ds$ (see
339: Appendix A). Then by the second property of the self-inversive
340: polynomials it is
341: guaranteed that this denominator never takes the zero value for $s$ over
342: the unit
343: circle. This property, guaranteed by unitarity, will remain true --even
344: when the traces
345: and propagator are calculated semiclassically-- as long as self
346: inversiveness is
347: maintained.
348:
349:
350: \noindent b) For $f(\hat U)= \hat I + s \hat U$ , $\hat G(s)$ in (9) is
351: the
352: Cayley transform of $\hat U$ and is therefore hermitian on the unit
353: circle. In
354: this case
355: \begin{equation}
356: \hat g_b (s) = \frac{\sum_{n=0}^N s^n ~ \hat Y_n}{\sum_{n=0}^N s^n ~ tr
357: \left(\hat Y_n
358: \right) }
359: \end{equation}
360: with
361: \begin{equation}
362: \hat Y_n = \hat X_n + \hat U \hat X_{n-1} = 2 \hat X_n - \beta_n \hat I
363: \label{yn} \ .
364: \end{equation}
365: These new operators satisfy a simpler symmetry relation than the
366: preceding $\hat X_n$,
367: analogous to the symmetry of the $\beta_n$ coefficients
368:
369: \begin{equation}
370: \hat Y_{N-j} = - (-1)^N \det \hat U ~ \hat Y^{\dagger}_j \ .
371: \label{revery}
372: \end{equation}
373: In this case, for even $N$, the normalized Green operator can be
374: written as
375:
376: \begin{equation}
377: \hat g_b(s) = \frac{\hat{\Sigma} (s) - \hat{\Sigma}^{\dagger} (1/\bar
378: s)} {tr\left[
379: \hat{\Sigma} (s) \right] - tr \left[ \hat{\Sigma}^{\dagger} (1/\bar s)
380: \right]}
381: \label{gb} \ ,
382: \end{equation}
383: where
384:
385: \begin{equation}
386: \hat{\Sigma} (s) = \frac{1}{\beta_{N/2}} \left[ \frac{1}{2} \hat Y_{N/2}
387: +
388: \sum_{n=0}^{N/2-1} s^{n-N/2} ~ \hat Y_n \right] \ .
389: \end{equation}
390:
391: In this expression, more symmetrical than (\ref{greenfinal}), the
392: numerator is
393: anti-hermitian on the unit circle, and the denominator is purely
394: imaginary. The basic
395: operator $\hat{\Sigma} (s)$ is a Fourier transform of the $\hat Y_n$
396: operators and their
397: traces. When utilizing this finite representation, they then become the
398: basic objects of
399: study in the time domain, in place of the propagator and its traces.
400:
401: When $\hat U$ is a finite matrix, $\hat g_a(s)$ and $\hat g_b(s)$ are
402: both rational
403: analytic functions of $s$. They are identical at the eigenvalues $s=s_k$
404: (on the unit
405: circle). However their zero and singularity structure is quite
406: different.
407:
408: In particular $\hat g_a(s)$, as noted, has a denominator which does not
409: vanish on the
410: unit circle, even when the coefficients are computed approximately, as
411: long as
412: self-inversiveness is imposed on them. The operator $\hat g_a(s)$ then,
413: has no
414: singularities in a small strip enclosing the unit circle and can be
415: followed continuously
416: from eigenvalue to eigenvalue. A disadvantage is that $\hat g_a(s)$ is
417: not hermitian
418: (except at the values $s=s_k$). On the contrary $\hat g_b(s)$ is
419: hermitian on the unit
420: circle, but its singularities (the zeroes of the denominator) are
421: precisely on the unit
422: circle, in between two successive eigenvalues.
423:
424: It should be noticed that, although for simplicity all the above formulae
425: are written for even $N$, they are easily adapted to the odd case, with
426: minimal changes.
427:
428: The preceding operator expressions for $\hat g_a(s)$ and $\hat g_b(s)$
429: provide a common structure that underlies all further semiclassical
430: approximations. Depending on the representation used to evaluate them,
431: they can yield results for transition probabilities, phase space
432: distributions, response functions or wave function correlations. This is
433: achieved by linking the spectral properties of the map to a {\it finite
434: number } of traces and matrix elements of the propagator. To apply them
435: one needs the semiclassical evaluation of the propagator traces as in
436: (\ref{gutz})- which are representation independent- and also of the
437: semiclassical propagator in the chosen representation.
438: For example if they are computed in the coordinate representation they
439: provide expressions for $|\langle x| \psi_k \rangle |^2$. If the propagator
440: is computed in the Weyl representation \cite{ozorio} they yield the Wigner
441: distribution of eigenfunctions \cite{fishman}.
442: In all cases the unitarity of the propagator - which is generally only
443: approximate in semiclassical treatments - is partially restored by the
444: imposition of the symmetries (\ref{rever},\ref{reverx},\ref{revery}).
445: In the next section the above formalism is developed in the coherent state
446: representation to obtain expressions for semiclassical Husimi distributions.
447:
448:
449: \section{Semiclassical approximation in the coherent state representation}
450:
451: The formulae derived in the previous section are exact and for maps can,
452: at first sight,
453: seem just a complicated way of looking at a simple algebraic eigenvalue
454: problem.
455: However, they are prepared in such a way as to make the semiclassical
456: transition as
457: clear as possible and to unveil the necessary approximations involved in
458: such a
459: transition,
460:
461: The coherent state mean value of $\hat g(s)$ is
462:
463: \begin{equation}
464: {\cal H} (s, z, \bar z) \equiv \frac{\langle z | \hat g(s) | z
465: \rangle}{\langle z | z \rangle} \ .
466: \end{equation}
467: This is a phase space distribution, analytic in $s$, which becomes the
468: Husimi
469: distribution of the eigenstates $|\psi_k \rangle$ at the values $s_k =
470: e^{- i
471: \epsilon_k}$.
472:
473: Its properties, reflecting the zeroes and singularities of $\hat g(s)$
474: can be followed
475: continuously in the $s$ complex plane. Its calculation involves the $N/2$
476: coherent state return amplitudes
477:
478: \begin{equation}
479: C_n (z, \bar z) \equiv \frac{\langle z | \hat U^n | z \rangle}{\langle z
480: | z \rangle}
481: \end{equation}
482: in addition of the $N/2$ traces
483:
484: \begin{equation}
485: b_n=tr \hat U^n = \int dz ~ d\bar z ~ C_n (z,\bar z) \ .
486: \end{equation}
487:
488: The first admits immediately a well known semiclassical approximation in
489: the time
490: domain, namely the Gutzwiller-Tabor trace formula (\ref{gutz}). The
491: stationary phase
492: approximations involved in the derivation are of course valid for fixed
493: time $n$ and for
494: $\hbar \rightarrow 0$, or equivalently, $N \rightarrow \infty$, and
495: their use in the
496: regime $n \approx N/2$ needed for the exact calculation is questionable.
497:
498: The semiclassical calculation of the return amplitudes $C_n (z, \bar z)$ can
499: be done by stationary phase calculation of the integral
500:
501: \begin{equation}
502: \langle z_f| \hat U^n |z_i \rangle = \int \int \langle z_f | q_f \rangle
503: \langle q_f |
504: \hat U^n | q_i \rangle \langle q_i | z_i \rangle ~ dq_i ~ dq_f
505: \label{intzq} \ ,
506: \end{equation}
507: where
508: \begin{equation}
509: \langle z| q \rangle = \frac{\exp \left\{ -\frac{1}{2 \hbar} \left[ z^2
510: + q^2 - 2 \sqrt
511: 2 z q \right] \right\} }{\sqrt [4]{\pi \hbar}}
512: \end{equation}
513: is the coherent state in coordinate representation, and the
514: propagator has the usual Van Vleck form
515: \begin{equation}
516: \langle q_f| \hat U^n |q_i \rangle = \frac{1}{\sqrt{2\pi i \hbar}}
517: \sum_{\gamma}
518: \sqrt{\frac{\partial ^2 S^{(n)}_{\gamma} (q_i,q_f) }{\partial q_i
519: \partial q_f}} ~ \exp \left[
520: ~ \frac{i}{\hbar} S^{(n)}_{\gamma} (q_i,q_f) \right] \label{vleck} \ .
521: \end{equation}
522: The sum must be made over all the paths in the phase space of the
523: corresponding
524: classical map that connect $q_i$ with $q_f$ in $n$ steps. Additional phases
525: connecting the different branches, the Maslov indices, must be added to
526: $S^{(n)}_{\gamma}$.
527:
528: The calculation of the integral (\ref{intzq}) with the Van Vleck
529: approximation
530: (\ref{vleck}) can be done as usual by stationary phase and the result
531: has the structure
532: of another Van Vleck formula with complex coordinates for a generating
533: function.
534:
535: \begin{equation}
536: \langle z_f | \hat U^n |z_i \rangle = \sum \sqrt{ -i \frac{\partial ^2
537: G_3}{\partial
538: \bar z_i \partial z_f} } ~ \exp \left[ -\frac{i}{\hbar} G_3(\bar
539: z_i,z_f) \right]
540: \label{zuzsem} \ .
541: \end{equation}
542:
543: This new classical generating function is a complex Legendre transform
544: of the action $S$
545:
546: \begin{equation}
547: G_3 = -S + \frac{1}{2} (p_f q_f - p_i q_i) + \frac{i}{4} (q_f^2 + p_f^2
548: + q_i^2 + p_i^2)
549: \label{gzz} \ ,
550: \end{equation}
551: or equivalently, in the complex coordinates
552:
553: \begin{eqnarray}
554: z &=& \frac{1}{\sqrt{2}} \left(q - i p \right) \\
555: \bar z &=& \frac{1}{\sqrt{2}} \left(q + i p \right) \ ,
556: \end{eqnarray}
557: it can be written as
558:
559: \begin{eqnarray}
560: G_3 = -S + \frac{i}{4} \left(z_f^2 + 2 z_f \bar z_f - \bar z_f^2
561: \right) - \frac{i}{4}
562: \left( z_i^2 - 2 z_i \bar z_i - \bar z_i^2 \right) \label{g1z} \ .
563: \end{eqnarray}
564:
565: This function depends explicitly on $\bar z_i,z_f$ variables, and yields
566: the equations
567: of motion
568:
569: \begin{equation}
570: z_i = -i \frac{\partial G_3}{\partial \bar z_i} \ , ~~~~~~~~~ \bar z_f = -i
571: \frac{\partial
572: G_3}{\partial z_f} \label{motion} \ .
573: \end{equation}
574:
575: These are algebraic equations that define the map implicitly provided
576: $\frac{\partial^2
577: G_3}{\partial \bar z_i \partial z_f} \neq 0$ where the sum must be made
578: over complex
579: paths that connect the point $\bar z_i$ to $z_f$. These formulae show
580: that the return amplitude $C_n$ has as exponent the function
581: $G_3(\bar z,z) -i \bar z z$ which using
582: (\ref{motion}) can be shown to be stationary at the periodic points.
583: Quadratic approximation of the generating function at these points yields
584: the coherent state return amplitudes as a sum of Gaussian peaks centered
585: on them
586:
587: \begin{equation}
588: C_n(z,\bar z) = \sum_{\gamma} \frac{1}{\sqrt{u_{\gamma}}} \exp \left[ -
589: \frac{1}{\hbar} {\cal G}^{\gamma} (z,\bar z) + \frac{i}{\hbar} S_{\gamma}
590: \right] \label{cn} \ .
591: \end{equation}
592:
593: \begin{equation}
594: {\cal G}^{\gamma} (z,\bar z) = \frac{1}{u_{\gamma}} \left[ \frac{\bar
595: v_{\gamma}}{2}
596: (\bar z - \bar z_{\gamma}) ^2 + (u_{\gamma}-1) (\bar z - \bar
597: z_{\gamma}) (z -
598: z_{\gamma}) - \frac{v_{\gamma}}{2} (z - z_{\gamma})^2 \right]
599: \label{gexp} \ ,
600: \end{equation}
601: where $z_{\gamma},\bar z_{\gamma}$ are the position of the periodic
602: points, $S_{\gamma}$
603: are the corresponding actions, and $u_{\gamma},v_{\gamma}$ and $\bar
604: v_{\gamma}$ are the
605: elements of the complex monodromy matrix defined by
606:
607: \begin{equation}
608: \left(
609: \begin{array}{cc}
610: \partial z_f / \partial z_i & \partial z _f / \partial \bar z_i \\
611: \partial \bar z_f / \partial z_i & \partial \bar z_f / \partial \bar z_i
612: \end{array}
613: \right) = \left(
614: \begin{array}{cc}
615: u & \bar v \\
616: v & \bar u
617: \end{array}
618: \right)
619: \end{equation}
620: calculated on the corresponding periodic point.
621:
622: Using (\ref{cn}) for the return amplitudes and (\ref{gutz}) for the
623: traces, the
624: Husimi distributions of eigenstates are explicitly written in terms of
625: properties of
626: periodic points of all periods up to half the Heisenberg time $N/2$.
627:
628: Although the semiclassical approximations are available in the time
629: domain for the
630: traces $b_n$ and the return amplitudes $C_n(z,\bar z)$ it is clear from
631: our formulae
632: that the essential ingredients for the calculations are the coefficients
633: $\beta_n$ and
634: the operators $\hat X_n$ (or $\hat Y_n$). These are related by
635: (\ref{betarec}) and
636: (\ref{xn}) to $b_n$ and $C_n$ and therefore can be recursively computed
637: from them. This
638: procedure yields the $\beta_n$ as linear combinations of {\it products}
639: --the so called
640: pseudo-orbits-- of terms involving many periodic orbits of different
641: periods and with
642: signs implying many possible cancellations between long and short
643: orbits. The same
644: situation prevails for the $\hat X_n$ (or $\hat Y_n$). The expressions
645: yield linear
646: combinations of return amplitudes at many different times. In fact $\hat
647: X_n(z,\bar z)$
648: give the phase space distribution of pseudo-orbits and it is easy to
649: show that
650: $\beta_n=(N-n) ~ tr \hat X_n$. In view of the cancellations involved it
651: would be very
652: desirable to have a direct semiclassical derivation --both for $\beta_n$
653: and $\hat
654: X_n$-- that would not go through the indirect procedure of approximating
655: $b_n$ and $\hat
656: U_n$, and then recursively computing $\beta_n$ and $\hat X_n$. However
657: to our knowledge
658: such a derivation is not available.
659:
660:
661:
662:
663:
664:
665: \section{Numerical calculations}
666:
667:
668: We will test the methods proposed on two simple well known maps: the cat
669: map, and the
670: baker map.
671:
672: For the cat map
673:
674: \begin{eqnarray}
675: \left(
676: \begin{array}{c}
677: q_f \\
678: p_f
679: \end{array}
680: \right) = \left(
681: \begin{array}{cc}
682: a & b \\
683: c & d
684: \end{array}
685: \right) \left(
686: \begin{array}{c}
687: q_i \\
688: p_i
689: \end{array}
690: \right) (mod ~ 1)
691: \end{eqnarray}
692: with $a,b,c$ and $d$ integer numbers, and $ad-bc=1$, the Van Vleck
693: formula in the
694: $q_i,q_f$ representation is exact \cite{hannay} and the corresponding
695: one in the
696: coherent state representation can be computed explicitly.
697:
698: The transformation to complex coordinates leads to the map
699:
700: \begin{eqnarray}
701: \left(
702: \begin{array}{c}
703: z_f \\
704: \bar z_f
705: \end{array}
706: \right) = \left(
707: \begin{array}{cc}
708: u & \bar v \\
709: v & \bar u
710: \end{array}
711: \right) \left(
712: \begin{array}{c}
713: z_i \\
714: \bar z_i
715: \end{array}
716: \right)
717: \end{eqnarray}
718: with
719:
720: \begin{eqnarray}
721: u &=& \frac{1}{2} \left [ (a+d) + i(b-c) \right ] \ , \\
722: v &=& \frac{1}{2} \left [ (a-d) + i(b+c) \right ] \ .
723: \end{eqnarray}
724: and its quantization with (\ref{vleck}) is obtained exactly by Gaussian
725: integration. The coherent state return amplitudes are computed with the
726: formulae (\ref{cn}) and (\ref{gexp}), with the
727: advantage that the coefficients of the monodromy matrix are the same for
728: all the
729: periodic points of a given period. Since the cat map is defined on the
730: torus we use
731: periodized coherent states \cite{voros}. With this result we can test
732: our formulae
733: keeping in mind that for this special case the semiclassical
734: approximation is exact.
735:
736: In Fig.1 we first look at quantities in the time domain, i.e. as a
737: function of $n$. The
738: correlations $|C_n(z, \bar z)|^2$ display for short times Gaussian peaks
739: at the
740: n-periodic points.
741:
742: However we have seen that the irreducible information about eigenstates
743: is carried by
744: more complicated operators such as the $\hat X_n$ in (\ref{xn}) or the
745: $\hat Y_n$ in
746: (\ref{yn}). These operators carry the phase space information that
747: corresponds to the
748: pseudo-orbits in the resummation of trace formula. In fact many
749: different powers of
750: $\hat U$ contribute to them and give rise to complex interference
751: structures in phase
752: space. The pseudo-orbit contributions --contained in the coefficient
753: $\beta_n$-- are
754: obtained from the fact that $\beta_n = (N-n) ~ tr \hat X_n$.
755:
756: In Fig.2 we have displayed the first few operators $\hat U \hat X_n$
757: appearing in the
758: expansion of $g_a(s)$. For very short times we notice the simple
759: superposition of the
760: patterns of $C_n(z,\bar z)$. In particular $\hat U \hat X_2$ shows peaks
761: at the period
762: 1, period 2 and period 3 periodic orbits. For longer times ($\hat U \hat
763: X_3$) the
764: interferences between orbits of different periods are very strong and
765: one can barely
766: notice the period 4 orbits, while the dominant structure is of period 1.
767: This is in
768: striking contrast with the regular appearance of periodic points in the
769: return amplitudes. Notice also the totally unexpected --and for us
770: unexplained-- appearance of a strong
771: periodicity (of period three) in the phase space patterns, which is
772: totally invisible in the $C_n$
773: coefficients. We cannot give an explanation to this behavior except
774: noting that a direct
775: semiclassical calculation of these coefficients would be very desirable
776: and might be
777: more effective than one which goes through the calculation with periodic
778: points.
779:
780: The smooth dependence of the operator $\hat g(s)$ on the quasi energy
781: variable allow us
782: to consider the exact response function as a distribution in phase space
783: depending
784: continuously on $s$, thus allowing an {\it exact} interpolation for all
785: values of $s$.
786: The Husimi eigen-distributions are particular values of this function
787: which
788: should be positive and display a pattern of $N$ zero values, in
789: accordance
790: with the general properties of Husimi distributions of pure states on
791: the
792: torus. \cite{voros}
793:
794: We display in Fig.3 the real part of ${\cal H}_a (e^{-i \epsilon}, z,
795: \bar z)$ for
796: values of $\epsilon$ near an eigenvalue. The distribution shows a very
797: stable pattern of
798: minima that gradually develops into $N=14$ zeros at
799: $\nu=\epsilon/2\pi=0.4167$. At this
800: point $Z(e^{-i \epsilon})=0$, the distribution is positive and we
801: verify that the
802: imaginary part vanishes for every phase space point. This is the signal
803: \cite{voros}
804: that a true projector has been reached and that the Husimi distribution
805: is the modulus
806: of an analytic function with $N$ zeros. Past the eigenvalue the
807: distribution can become
808: negative but the pattern of minima remains very stable. At the values
809: where $Z(e^{-i
810: \epsilon})$ has an extremum, in between zeros, we show in the appendix
811: that the
812: distribution becomes flat with value $1/N$. At this point the pattern of
813: minima changes
814: and picks up the structure of zeros of the next eigenfunction.
815:
816: For cat maps the semiclassical formulae are exact and therefore the above
817: results merely confirm that the whole scheme is consistent and accurate but
818: they do not really test the semiclassical approximation.
819: Towards this purpose we consider another simple linear map --the baker's--
820: where the semiclassics is not exact. The classical $T$-iteration of the
821: baker map from the $(q_0, p_0)$ to $(q_T, p_T)$ real coordinates
822: \cite{baker}, takes the complex form
823:
824: \begin{eqnarray}
825: z_T &=& u_T ~ z_0 + v_T ~ \bar {z_0} - w_T \\
826: \bar z_T &=& v_T ~ z_0 + u_T ~ \bar {z_0} - \bar w_T
827: \end{eqnarray}
828:
829: \begin{eqnarray}
830: u_T &=& \frac{2^T + 2^{-T}}{2} = \cosh (T \ln 2) \\
831: v_T &=& \frac{2^T - 2^{-T}}{2} = \sinh (T \ln 2) \\
832: w_T &=& \frac{1}{\sqrt{2}} \left( \gamma + i ~ 2^{-T} \gamma ^{\dagger}
833: \right)
834: \end{eqnarray}
835: $\gamma$ is the integer part of $2^T q_0$ and $\gamma^{\dagger}$ is the number
836: obtained after
837: inversion of their binary digits. Each possible $\gamma$ defines a periodic
838: point $z_{\gamma}=(\gamma-i\gamma^{\dagger})/\sqrt{2}(2^T-1)$. The return
839: amplitudes $C_n (z, \bar z)$ are expressed as a sum over them
840:
841: \begin{equation}
842: C_n(z,\bar z) = \frac{1}{\sqrt{u_T}} \sum_{\gamma=1}^{2^T-1} \exp \left[ -
843: \frac{1}{\hbar} {\cal
844: G}^{\gamma} (z,\bar z) + \frac{i}{\hbar} S_{\gamma} \right] \ ,
845: \end{equation}
846: where the complex generating function takes the standard form
847: \begin{equation}
848: {\cal G}^{\gamma} (z,\bar z) = \frac{1}{u_T} \left[ \frac{v_T}{2} (\bar z -
849: \bar z_{\gamma})
850: ^2 + (u_T-1) (\bar z - \bar z_{\gamma}) (z - z_{\gamma}) - \frac{v_T}{2} (z -
851: z_{\gamma})^2
852: \right]
853: \end{equation}
854: and
855: \begin{equation}
856: S_{\gamma} = \frac{\gamma ~ \gamma ^{\dagger}}{2^T-1}
857: \end{equation}
858: is the action of the corresponding periodic orbit.
859:
860: In Fig.4 we show the exact passage of the distribution through an
861: eigenvalue ($N=16$) where - as expected - the same general pattern as the
862: previous example can be observed.
863:
864: In Figs.5 and 6 we test the semiclassical
865: approximation by comparing both the spectrum and the eigendistributions with
866: the exact results. Fig.5 shows the $Z$ function for a small stretch of the
867: unit circle comprising three zeroes ($ N=20 $) and showing that the
868: semiclassical zeroes are approximated well within a fraction of the mean
869: eigenvalue separation.
870: In Fig.6 we show the three eigendistributions: the top three are the
871: exact ones obtained at the exact zeroes, while the bottom three are
872: the corresponding semiclassical ones (computed at the semiclassical zeroes).
873: Considering that the distributions are obtained by combining the classsical
874: information of $ 2^{N/2}\approx 1000 $ periodic points, it is remarkable
875: how well the approximation captures the essential structure
876: of the eigenfunctions. One important point should be noticed: in the
877: semiclassical case there is no guarantee that the distribution calculated at
878: a zero of $Z$ will correspond exactly to a one dimensional projector
879: (i.e. to a pure state) and therefore its Husimi distribution may not have
880: exactly $N$ zeroes and may be even negative. This is clearly the case in
881: the first semiclassical state at $\nu=0.4734$. The large white area has
882: negative values - and is not contoured - but corresponds very well with
883: the central area where the zeroes of the exact distribution are located.
884: The positive top of the distribution is roughly well located but it looks
885: like the whole distribution
886: had been shifted towards negative values, while mantaining a very accurate
887: correspondence of the maxima and minima. The second state at $\nu=0.5445$
888: reproduces very well the top part of the distribution, while now the minima
889: have become positive, but again maintaining a very accurate correspondence
890: with the exact pattern of zeroes. The third state at $\nu=0.6088$
891: reproduces in very fine detail the patterns of the exact state. Overall
892: we have observed that this state of affairs is quite typical: the position
893: of the maxima and minima of the semiclassical states are very well
894: reproduced but the relative heights of peaks and valleys are not always
895: accurate.
896:
897:
898:
899: \section{Conclusions}
900: We have presented a method that can, in principle, provide a reliable
901: way to compute
902: single Husimi eigendistributions and which has the same advantages --and
903: drawbacks-- as
904: the resummation methods based on the spectral determinant for the
905: calculation of single
906: eigenvalues. The final results (\ref{greenfinal}) and (\ref{gb}) for the
907: Green operators
908: when approximated semiclassically give approximations to the projectors
909: on single eigenstates as sums of
910: Gaussian centered on periodic points with periods up to half the
911: Heisenberg time. Similar
912: expressions, with contributions not so well localized, can be obtained
913: for Wigner
914: functions or probability distributions. The usual exponential
915: proliferation of periodic
916: orbits in a chaotic system is the main obstacle to applications in the
917: deep semiclassical
918: regime. These are however not limited to simple maps as the method has
919: been applied also
920: to the Bunimovich stadium in \cite{sara-simo}.
921:
922: We acknowledge many important discussions with F. Simonotti and P. Leboeuf,
923: and partial support from ANPCyT program PICT97-01015 and ECOS-Secyt A98E03.
924:
925:
926: \appendix
927: \section{Normalized Green operator at $Z$-function extremes}
928:
929:
930: To calculate the denominator of $\hat g_a$ we use a well known property
931: of the
932: parametrized matrices
933:
934: \begin{equation}
935: \frac{d}{ds} \det \left[ \hat M(s) \right] = \det \left[ \hat M(s)
936: \right] tr \left[
937: \frac{d}{ds} \left[ \hat M(s) \right] \hat M^{-1} (s) \right] \ .
938: \end{equation}
939:
940: If we use this property for $\hat M(s)=\hat I - s \hat U$, then its
941: determinant is the
942: characteristic polynomial, and we have
943:
944: \begin{equation}
945: \frac{d}{ds} P(s) = P(s) ~ tr \left[ - \hat U (\hat I - s \hat U)^{-1}
946: \right] \ .
947: \end{equation}
948: Then if $P(s)$ is introduced into the trace, we obtain the transpose
949: cofactor matrix,
950: and we demonstrate
951:
952: \begin{equation}
953: tr \left[ \hat U ~ C^t \left( \hat I - s \hat U \right) \right] = -
954: \frac{d}{ds} P(s) \ ,
955: \end{equation}
956: that it is exactly the denominator of $\hat g_a$. Recalling the definition
957: of $Z(s)$ function
958:
959: \begin{equation}
960: Z(s) = \frac{P(s)}{\beta_{N/2} s^{N/2}} \ ,
961: \end{equation}
962: we calculate its derivative respect to the real parameter $\epsilon$
963: that rounds the
964: unit circle $(s=e^{-i \epsilon})$
965:
966: \begin{equation}
967: \frac{dZ}{d \epsilon} = \frac{dZ}{ds} \frac{ds}{d \epsilon} =
968: \frac{i~s^{-N/2}}{
969: \beta_{N/2} } \left[ \frac{N}{2} P(s) - s \frac{d}{ds} P(s) \right] \ .
970: \end{equation}
971: Then at $s_{\alpha}=e^{-i \epsilon_{\alpha}}$ where $Z$, as a function
972: of $\epsilon$,
973: reach a maximum or a minimum, we have the following identity
974:
975: \begin{equation}
976: \frac{dP}{ds} (s_{\alpha}) = \frac{N}{2 s_{\alpha}} P(s_{\alpha}) \ .
977: \end{equation}
978:
979: The denominator of the $\hat g_a$ normalized Green operator is equal to
980: the derivative
981: of the spectral determinant. Then, in a diagonal basis of $\hat U$,
982: $\hat g_a$
983: specialized at a extreme of the $Z$ function is
984:
985: \begin{eqnarray}
986: \hat g_a(s_{\alpha}) &=& \frac{ \sum_{n=1}^N (1/s_n) \prod_{k \neq n}
987: (1-s_{\alpha}
988: /s_k) |\psi_n \rangle \langle \psi_n| } {-\frac{dP}{ds} (s_{\alpha})}
989: \nonumber \\
990: &=& - \frac{ \sum_{n=1}^N (1/s_n) \prod_{k \neq n} (1-s_{\alpha}/s_k)
991: |\psi_n \rangle
992: \langle \psi_n| } { \frac{N}{2s_{\alpha}} P(s_{\alpha})} \nonumber \\
993: &=& - \frac{2}{N} \frac{s_{\alpha}}{\prod_{k} (1-s_{\alpha} / s_k)}
994: \sum_{n=1}^N (1/s_n)
995: \prod_{k \neq n} (1-s_{\alpha}/s_k) |\psi_n \rangle \langle \psi_n| \ .
996: \end{eqnarray}
997: Introducing the spectral determinant of the denominator into the sum we
998: obtain
999:
1000: \begin{equation}
1001: \hat g_a(s_{\alpha}) = - \frac{2}{N} \sum_{n=1}^N \frac{(s_{\alpha}
1002: /s_n)}{1-(s_{\alpha}/ s_n) } |\psi_n \rangle \langle \psi_n| \ .
1003: \end{equation}
1004:
1005: We can separate the real and the imaginary part of the prefactor,
1006: knowing that both
1007: $s_{\alpha}$ and $s_n$ are on the unit circle.
1008:
1009: \begin{equation}
1010: \frac{(s_{\alpha}/s_n)}{1-s_{\alpha} /s_n} = -\frac{1}{2} + \frac{i}{2}
1011: \frac{ \Im
1012: (s_{\alpha}/s_n) }{\left[ 1- \Re (s_{\alpha}/s_n) \right]} \ .
1013: \end{equation}
1014:
1015: Contrary to the imaginary part, we see that the real part is independent
1016: of the index $n$ or $\alpha$. Then the
1017: normalized Green operator at the extremes of the $Z$ function can be
1018: written as
1019:
1020: \begin{eqnarray}
1021: \hat g(s_{\alpha}) &=& - \frac{2}{N} \sum_{n=1}^N \left( -\frac{1}{2} +
1022: \frac{i}{2}
1023: B_{{\alpha} n} \right) |\psi_n \rangle \langle \psi_n| \\
1024: &=& \frac{1}{N} \sum_{n=1}^N |\psi_n \rangle \langle \psi_n| -
1025: \frac{i}{N}
1026: \sum_{n=1}^N B_{{\alpha} n} |\psi_n \rangle \langle \psi_n|
1027: \end{eqnarray}
1028: and splits into an hermitian and antihermitian part. We have demonstrated
1029: here that the
1030: hermitian part is exactly the identity operator over $N$. Then, the real
1031: part of the
1032: expectation value of $\hat g(s_{\alpha})$ will be always $1/N$,
1033: independent of the
1034: quantum state.
1035:
1036:
1037: \begin{thebibliography}{99}
1038:
1039: \bibitem{berry1} M. V. Berry, Les Houches Lecture Notes, Summer School
1040: on Chaos and
1041: quantum Physics, M.-J. Giannoni, A. Voros, and J. Zinn-Austin, eds.,
1042: Elsevier Science
1043: Publishers B. V. (1991); {\sl Proc. Roy. Soc.} {\bf A 243}, 219 (1989)
1044: \bibitem{voros} P. Leboeuf and A. Voros, {\sl J. Phys.} {\bf A 23}, 1765
1045: (1990)
1046: \bibitem{heller} E. J. Heller, {\sl Phys. Rev. Lett.} {\bf 53}, 1515
1047: (1984); Wavepacket
1048: Dynamics and Quantum Chaology in {\sl Chaos and Quantum Physics}, M.-J.
1049: Giannoni, A.
1050: Voros, and J. Zinn-Austin, eds., Elsevier Science Publishers, Amsterdam
1051: (1990)
1052: \bibitem{bogo1} E. B. Bogomolny {\sl Physica} {\bf D 31}, 169 (1988)
1053: \bibitem{vergini} D. A. Wisniacki and E. Vergini, {\sl Phys. Rev. E}
1054: {\bf 62}, 4513
1055: (2000)
1056: \bibitem{nonnenmacher}S. Nonnenmacher, A. Voros {\sl J. Stat. Phys.}
1057: {\bf 92}, 431 (1998)
1058: \bibitem{mirlin} A. D. Mirlin, Lecture course given at the International
1059: School {\sl Enrico Fermi} on New Directions in Quantum Chaos, Varenna,
1060: July 1999 (cond-mat/0006421)
1061: \bibitem{gutzw} M. C. Gutzwiller, {\sl J. Math. Phys.} {\bf 8}, 1979
1062: (1967)
1063: \bibitem{bogo-keat} E. B. Bogomolny and J. P. Keating, {\sl Phys. Rev.
1064: Lett.} {\bf 77},
1065: 6091 (1994)
1066: \bibitem{agamfish} O. Agam and S. Fishman, {\sl Phys. Rev. Lett.} {\bf
1067: 73}, 806 (1994)
1068: \bibitem{sara-simo} M. Saraceno and F. Simonotti, {\sl Phys. Rev. E}
1069: {\bf 61}, 6527
1070: (2000)
1071: \bibitem{pseudo} M. V. Berry and J. P. Keating, {\sl J. Phys.} {\bf A
1072: 23}, 4839 (1990)
1073: \bibitem{bogo2} E. B. Bogomolny, {\sl Nonlinearity} {\bf 5}, 805 (1992)
1074: \bibitem{bbl} E. B. Bogomolny, O. Bohigas and P. Leboeuf, {\sl Phys. Rev.
1075: Lett.} {\bf 68}, 2726 (1992)
1076: \bibitem{marsden} M. Marsden, {\sl Geometry of Polynomials}, American
1077: Mathematical
1078: Society, Providence, Rhode Island (1966).
1079: \bibitem{berry-keat} M. V. Berry and J. P. Keating, {\sl Proc. Roy.
1080: Soc.} {\bf A 437},
1081: 151 (1992)
1082: \bibitem{ozorio}A. M. Ozorio de Almeida, {\sl Phys. Rep. }{\bf 295},265 (1998)
1083: \bibitem{fishman} S. Fishman, B. Georgeot and R. E. Prange, {\sl J.
1084: Phys.} {\bf A 29},
1085: 919 (1996); S. Fishman in {\sl Supersymmetry and Trace Formulae: Chaos
1086: and Disorder},
1087: Lerner {\sf et al.} Eds., NATO ASI {\bf B} Vol. 370, Kluwer Academic,
1088: New York (1999)
1089: \bibitem{hannay} J. H. Hannay and M. V. Berry, {\sl Physica} {\bf 1D},
1090: 267 (1980)
1091: \bibitem{baker} M. Saraceno and A. Voros, {\sl Physica D} {\bf 79}, 206
1092: (1994)
1093:
1094: \end{thebibliography}
1095:
1096: \newpage
1097:
1098: %*************************************************************
1099: \begin{figure}
1100: \centerline{\psfig{figure=fig1.epsi}}
1101: \vspace{0.2cm}
1102: \caption{\label{ucat} Modulo of the coherent state return amplitude for the
1103: first six powers of the Arnold cat map propagator. The torus is quantized
1104: with periodic boundary conditions, and $N=128$. Linear grey scale between
1105: 0 (white), and the maximum (black) of each picture.}
1106: \end{figure}
1107: %*************************************************************
1108: \begin{figure}
1109: \centerline{\psfig{figure=fig2.epsi}}
1110: \vspace{0.2cm}
1111: \caption{\label{uxcat} Modulo of the coherent state mean value for the
1112: $\hat U \hat X_n$ operators. Same parameters and scales as Fig. 1}
1113: \end{figure}
1114: %*************************************************************
1115: \begin{figure}
1116: \centerline{\psfig{figure=fig3.epsi}}
1117: \vspace{0.2cm}
1118: \caption{\label{gcat} Real part of ${\cal H}_a (s,z,\bar z)$, for $s = \exp
1119: (-i 2 \pi \nu)$.
1120: Linear grey scale between 0 (white), and $1/\sqrt{N}$ (black).
1121: Logarithmic level curves
1122: at $1/N, 1/N^2,1/N^3,...$. Arnold cat map with periodic boundary
1123: conditions, and $N=14$.
1124: There is an eigenvalue of the propagator near $\nu=0.4167$ (3rd.
1125: picture). The exact and
1126: semiclassical computation gives the same values.}
1127: \end{figure}
1128: %*************************************************************
1129: \begin{figure}
1130: \centerline{\psfig{figure=fig4.epsi}}
1131: \vspace{0.2cm}
1132: \caption{\label{gbak} The same as Fig. 3, but for the baker map, with
1133: anti-periodic boundary
1134: conditions, and $N=16$. There is an eigenvalue of the propagator near
1135: $\nu=0.4606$ (4th.
1136: picture).}
1137: \end{figure}
1138: %*************************************************************
1139: \begin{figure}
1140: \centerline{\psfig{figure=newfig5.eps}}
1141: \vspace{0.2cm}
1142: \caption{\label{zexsem} Exact $Z$ function (bold line), and its semiclassical
1143: approximation (dashed line), for the baker map, with anti-periodic boundary
1144: conditions, and $N=20$. The zeroes correspond to the eigenvalues}
1145: \end{figure}
1146: %*************************************************************
1147: \begin{figure}
1148: \centerline{\psfig{figure=newfig6.eps}}
1149: \vspace{0.2cm}
1150: \caption{\label{gbaksem} Exact and semiclassical eigenstates in correspondence
1151: with the zeroes in Fig.5. The gray scale and the contours are as in Fig.3. }
1152: \end{figure}
1153: %*************************************************************
1154:
1155:
1156:
1157:
1158: \end{document}
1159:
1160:
1161:
1162:
1163:
1164:
1165:
1166:
1167:
1168:
1169:
1170: