nlin0207043/q4pap.tex
1: 
2: %   q4pap.tex (v3)
3: 
4: %   ``Analytical perturbative approach to periodic orbits 
5: %     in the homogeneous quartic oscillator potential''
6: 
7: %     by M Brack, S N Fedotkin, A G Magner and M Mehta
8: 
9: %   accepted by J. Phys. A on December 4, 2002
10: 
11: %   revised and extended version, resubmitted December 3, 2002
12: 
13: %   corrected version including ``Note added in proof'', 19.12.02
14: 
15: \documentclass[12pt,a4paper]{article}
16: \usepackage{a4,epsf,here}
17: \setlength\textwidth{16.6cm}
18: \setlength\textheight{24.0cm}
19: \setlength{\unitlength}{1cm}
20: \addtolength\evensidemargin{-2.85cm}
21: \addtolength\oddsidemargin{-0.55cm}
22: \addtolength\topmargin{-1.8cm}
23: 
24: %my own macros:
25: \renewcommand{\arraystretch}{1.05}
26: \newcommand{\der}[2]{{\frac{d #1}{d #2}}}
27: \def\K{\hbox{\bf K}}
28: \def\E{\hbox{\bf E}}
29: \def\sn{{\rm sn}}
30: \def\dn{{\rm dn}}
31: \def\cn{{\rm cn}}
32: \def\Ec{{\rm Ec}}
33: \def\Es{{\rm Es}}
34: \def\En{{\rm E}_n}
35: \def\Fn{{\rm F}_n}
36: \def\Eps{{\cal E}}
37: \def\SA{S_{\rm A}}
38: \def\TA{T_{\rm A}}
39: \def\M{{\rm M}}
40: \def\MA{{\rm M}_{\rm A}}
41: \def\trMA{{\rm tr}\,\MA}
42: \def\trM{{\rm tr}\,\M}
43: \def\be{\begin{equation}}
44: \def\ee{\end{equation}}
45: \def\bea{\begin{eqnarray}}
46: \def\eea{\end{eqnarray}}
47: \def\eq#1{(\ref{#1})}
48: \def\bs{\bigskip}
49: \def\ms{\medskip}
50: \def\fig#1{figure \ref{#1}}
51: \def\tab#1{table \ref{#1}}
52: \def\etal{{\it et al}}
53: \def\eps{\epsilon}
54: \def\siml{\,\hbox{\kern.1em \lower.6ex \hbox{$\sim$} \kern-1.12em
55:           \raise.6ex \hbox{$<$} }}
56: \def\simg{\,\hbox{\kern.1em \lower.6ex \hbox{$\sim$} \kern-1.12em
57:           \raise.6ex \hbox{$>$} }}
58: \newcommand{\Figurebb}[9]{
59: \begin{figure}[H]
60: \leavevmode
61: \epsfysize=#7cm
62: \epsfbox[#2 #3 #4 #5]{#6}
63: \par
64: \parbox{#8cm}{
65: \caption[figure]{\renewcommand{\baselinestretch}{0.8} \small
66:                                            \hspace{-0.3truecm}#9}
67: \label{#1}}
68: \end{figure}
69: }
70: \newcommand{\Table}[4]{
71: \begin{table}[H]\begin{center}{#3}
72: \parbox{#2cm}{
73: \vspace{0.5cm}
74: \caption[table]{\renewcommand{\baselinestretch}{0.8} \small
75:                                            \hspace{-0.3truecm}#4}
76: \label{#1}}
77: \end{center}
78: \end{table}
79: }
80: %end of my own macros
81: 
82: \begin{document}
83: 
84: \baselineskip 14pt
85: 
86: \centerline{\bf \Large Analytical perturbative approach to periodic orbits}
87: 
88: \bs
89: 
90: \centerline{\bf \Large in the homogeneous quartic oscillator potential}
91: 
92: \bs
93: \bs
94: \bs
95: 
96: \centerline{\bf M Brack$^1$, S N Fedotkin$^{1,2}$, A G Magner$^{1,2}$ 
97:                 and M Mehta$^{1,3}$}
98: 
99: \ms
100: \bs
101: 
102: {\small
103: 
104: \centerline{$^1$Institute for Theoretical Physics, University of
105: Regensburg, D-93040 Regensburg, Germany}
106: 
107: \centerline{$^2$Institute for Nuclear Research, 252028 Prospekt 
108: Nauki 47, Kiev-28, Ukraine}
109: 
110: \centerline{$^3$Harish-Chandra Research Institute, Chhatnag Road, Jhusi,
111: Allahabad, 211019 India}
112: 
113: \bs
114: 
115: \centerline{\today~~(v3)}
116: 
117: \bs
118: 
119: \noindent
120: {\bf \large Abstract}
121: 
122: \ms
123: \noindent
124: We present an analytical calculation of periodic orbits in the 
125: homogeneous quartic oscillator potential. Exploiting the properties of 
126: the periodic Lam{\'e} functions that describe the orbits bifurcated 
127: from the fundamental linear orbit in the vicinity of the bifurcation 
128: points, we use perturbation theory to obtain their evolution away from 
129: the bifurcation points. As an application, we derive an analytical 
130: semiclassical trace formula for the density of states in the separable 
131: case, using a uniform approximation for the pitchfork bifurcations 
132: occurring there, which allows for full semiclassical quantization. For 
133: the non-integrable situations, we show that the uniform contribution of 
134: the bifurcating period-one orbits to the coarse-grained density of states
135: competes with that of the shortest isolated orbits, but decreases with 
136: increasing chaoticity parameter $\alpha$.
137: 
138: \section{Introduction}
139: 
140: The homogeneous quartic oscillator potential $V(x,y)=a\,x^4+b\,y^4+
141: (\alpha/2)\,x^2y^2$ has been the object of both classical, semiclassical 
142: and quantum-mechanical studies \cite{q4ls,q4po,btu,erda}. Lakshminarayan 
143: \etal~\cite{lakh} have investigated the fixed points in Poincar{\'e} 
144: surfaces of section corresponding to orbits of period four and determined 
145: empirically some of their scaling properties. Due to the homogeneity of 
146: the potential in the coordinates, orbits at different energies are related 
147: to each other through a simple scaling of coordinates and momenta. We may 
148: therefore fix the energy $E$ at an arbitrary value. The nonlinearity 
149: parameter that regulates the dynamics is the parameter $\alpha$. The 
150: system possesses periodic straight-line orbits along both axes which 
151: undergo stability oscillations under variation of $\alpha$. Infinite 
152: sequences of new periodic orbits bifurcate from each of these straight-line 
153: orbits and their repetitions, leading to {\it almost} completely chaotic 
154: dynamics \cite{daru} in the limit $\alpha\to\infty$.
155: 
156: In this paper we specialize to the symmetric case $a=b=1/4$ in which the 
157: potential has C$_{4v}$ symmetry; this potential shall in the following be 
158: denoted as the Q4 potential:
159: \be
160: V_{Q4}(x,y)=\frac14\left(x^4+y^4\right)+\frac{\alpha}{2}\,x^2y^2\,. 
161: \label{q4xy}
162: \ee
163: The straight-line orbits along the $x$ and $y$ axes then obey identical
164: equations of motion;
165: we denote them as the A orbits. The dynamics of the Q4 potential \eq{q4xy} 
166: is invariant under the symmetry operation (cf \cite{erda}) 
167: $\alpha\to (3-\alpha)/(1+\alpha)$, which corresponds to a rotation 
168: in the $(x,y)$ plane about 45 degrees and a simultaneous stretching of 
169: coordinates and time by a factor $[2/(1+\alpha)]^{1/4}$. The limit 
170: $\alpha\to\infty$ therefore is equivalent to the limit 
171: $\alpha\to -1$. There are three values of $\alpha$ for which the
172: potential is integrable: 1) $\alpha=0$, giving separability in $x$ and $y$; 
173: 2) $\alpha=1$, which is the fixed point of the above symmetry 
174: operation, giving the isotropic quartic oscillator $V(r)=\frac14\,r^4$ with 
175: $r^2=x^2+y^2$; and 3) $\alpha=3$, giving separability after rotation about 
176: 45 degrees.
177: 
178: \Figurebb{q4trm}{-30}{55}{795}{490}{q4papf1.eps}{8}{16.6}{
179: Stability discriminant $\trM$ of period-one orbits in the Q4 potential,
180: plotted versus $\alpha$. Shown are the curves for the primitive orbits
181: A, B and C, and the orbits L and S bifurcating from orbits A and B,
182: respectively. Subscripts indicate the Maslov indices $\sigma_{po}$ 
183: appearing in \eq{dgtf}. The vertical dashed lines at $\alpha=0$, 1 and 
184: 3 correspond to the integrable situations.
185: }
186: 
187: In a recent paper \cite{lamp}, the period-one and period-two orbits 
188: bifurcating from the A orbits have been classified completely in terms of 
189: periodic Lam{\'e} functions. The motion of the primitive A orbit along the 
190: $y$ axis is given analytically by 
191: \be
192: x_A(t) = 0\,,\quad y_A(t) = y_0\,\cn(y_0t,\kappa)\,, 
193:          \qquad y_0=(4E)^{1/4}, \qquad \kappa^2=1/2\,,
194: \label{yaq4}
195: \ee
196: with the period $\TA = 4\K/y_0$, where $\K=K(\kappa)=F\left(\frac{\pi}{2},
197: \kappa\right)$ is the complete elliptic integral of the first kind with 
198: modulus $\kappa$, and $\cn(z,\kappa)$ is one of the Jacobi elliptic 
199: functions \cite{grry}. The turning points are $\pm y_0$. Note that this 
200: solution does not depend on the value of $\alpha$. The stability of the 
201: orbit A, however, does depend on $\alpha$. The linearized equation of 
202: motion in the transverse $x$ direction yields, after transformation to 
203: the scaled time variable $z=y_0t$, the Hill equation
204: \be
205: x''(z) + \alpha\,[1-\sn^2(z,\kappa)]\,x(z) = 0\,.
206: \label{lameq4}
207: \ee
208: This is a special case of the Lam{\'e} equation \cite{erde}
209: \be
210: \Lambda''(z) + \left[h - n(n+1)\,\kappa^2\,{\rm \sn}^2(z,\kappa)\right]\!
211:                \Lambda(z) 
212:        = 0 \qquad \hbox{with} \quad \kappa^2=\frac12\,, 
213:            \quad h = \alpha_n = \frac12 \,n\,(n+1)\,.
214: \label{lameq}
215: \ee
216: We therefore know here analytically the eigenvalues $h=\alpha_n$ of the 
217: Lam{\'e} equation, which correspond to the bifurcation points $\alpha_n$ 
218: of the A orbit \eq{yaq4}. This agrees with the analytical result for its 
219: stability discriminant, given by the trace of the stability matrix M, 
220: which has been derived long ago by Yoshida \cite{yosh}:
221: \be
222: \trMA = 4\,\cos\left(\frac{\pi}{2}\sqrt{1+8\alpha}\right) + 2\,.
223: \label{trmaq4}
224: \ee
225: It is easily seen that the bifurcation condition $\trMA=+2$ leads exactly 
226: to the values $\alpha_n$ in \eq{lameq}. 
227: 
228: In \fig{q4trm} we show the stability discriminant $\trM$ for some period-one 
229: orbits of the Q4 potential in the interval $-1\leq\alpha\leq 4$. The
230: integrable situations are indicated by the vertical dashed lines. The upper 
231: horizontal dotted line is the bifurcation line $\trM=+2$. The orbits B are
232: the two straight-line orbits along the diagonals $y=\pm x$, which are mapped 
233: onto the A orbits under the above-mentioned symmetry operation. Their motion 
234: is given by
235: \be
236: \pm\,x_B(t) = y_B(t) = y_0\,[2(1+\alpha)]^{-1/4}\,\cn(y_\alpha t,\kappa)\,, 
237:                        \qquad    
238:    y_\alpha = y_0\,[(1+\alpha)/2]^{1/4},
239: \ee
240: their period is $T_{\rm B}=\TA\,[2/(1+\alpha)]^{1/4}$, and their stability 
241: discriminant $\trM_{\rm B}$ is found from $\trMA$ in \eq{trmaq4} by 
242: replacing $\alpha\to (3-\alpha)/(1+\alpha)$. The orbit C is a rotational 
243: orbit which has a discrete degeneracy of two because of time reversal 
244: symmetry; its solutions $x_C(t)$ and $y_C(t)$ could not be found analytically
245: for arbitrary values of $\alpha$ (see section 3 for the case $\alpha=0$). 
246: L and S are the first librating orbits born from A and B, respectively, in 
247: isochronous (period-one) pitchfork bifurcations. Not shown are the period-one
248: orbits bifurcating from B for $\alpha\leq-3/7$. The results for $\trM$ for 
249: the C, L and S orbits were obtained numerically. The subscripts of all the 
250: orbits shown in \fig{q4trm} denote their Maslov indices $\sigma_{po}$ used 
251: in the semiclassical trace formula \eq{dgtf} in section 3.
252: 
253: The periodic solutions of the Lam{\'e} equation \eq{lameq} are the Lam{\'e} 
254: functions \cite{erde,inc1} Ec$_n^m(z,\kappa)$ and Es$_n^m(z,\kappa)$ which 
255: are even and odd functions of $z$, respectively, with $m$ zeros in 
256: $z\in[0,2\K)$. For integer $n$ they are polynomials of degree $n$ in the
257: Jacobi elliptic functions sn, cn and dn. Those with even $m$ have the 
258: period $T=2\K$, those with odd $m$ have $T=4\K$. For the special case 
259: $\kappa^2=1/2$, $m$ is fixed by $m={\rm int}[(n+1)/2]$ and there exists 
260: only one type of Lam{\'e} polynomial, Ec or Es, for each value of $n$. 
261: It is therefore sufficient here to denote these polynomials by E$_n(z)$ 
262: with $n=0,1,2,\dots$ Their explicit expressions up to $n=15$ have been 
263: given in \cite{lamp}. As there, we use from now on the short notation 
264: $\cn(z)=\cn(z,\kappa)$, $\sn(z)=\sn(z,\kappa)$ etc, keeping in mind that 
265: $\kappa^2=1/2$.
266: 
267: Nontrivial period doublings of the orbits A occur when $\trMA=-2$ (cf the 
268: lower horizontal line in \fig{q4trm}), which leads with \eq{trmaq4} to the 
269: critical values $\alpha_p = 2\,p\,(p+1)+3/8$ with $p = 0,1,2,\dots$ The 
270: corresponding solutions of \eq{lameq} with $n=(4p+1)/2$ are algebraic 
271: Lam{\'e} functions \cite{inc2,erd2} of period $8\K$ with $m=(2p+1)/2$. 
272: They are discussed in \cite{lamp} in connection with the period-two orbits 
273: born at the corresponding (island-chain type) bifurcations.
274: 
275: The purpose of this short paper is to demonstrate the use of the 
276: simple properties of the Lam{\'e} functions for analytical 
277: classical and semiclassical studies involving the periodic orbits
278: of the Q4 potential. In section 2 we shall formulate a perturbation
279: expansion for describing the evolution of the bifurcated orbits away 
280: from the bifurcation points $\alpha_n$, and in section 3 we apply some of
281: its results to semiclassical calculations of the density of states.
282: 
283: \section{Perturbation expansion around bifurcation points}
284: 
285: As shown in \cite{lamp}, the transverse motion of the orbits bifurcated 
286: from A is, infinitesimally close to the bifurcation values $\alpha_n$, 
287: exactly described by the Lam{\'e} functions. In \cite{lamp,mbgu} similar 
288: bifurcation cascades were investigated for H{\'e}non-Heiles type 
289: potentials. For these it was possible to determine the amplitude of the 
290: transverse motion of the bifurcated orbits from the conservation of the 
291: total energy, exploiting the asymptotic separability of these systems 
292: near the saddle energy into motions parallel and transverse to the 
293: bifurcating orbits. However, due to the scaling property of the Q4 
294: potential \eq{q4xy}, a variation of the energy does not affect the 
295: stability of the A orbits and hence the energy conservation cannot be 
296: exploited in the same way. In order to find the evolution of the new 
297: orbits away from their bifurcations, we propose here a perturbative 
298: series expansion of the equations of motion around the bifurcation points 
299: $\alpha_n$, leading to successive analytical expressions for the 
300: corrections to the simple (lowest-order) Lam{\'e} solutions.
301: 
302: The exact equations of motion for the Hamiltonian 
303: \be
304: H(p_x,p_y,x,y) = \frac12 \left(p_x^2+p_y^2\right) + V_{Q4}(x,y)
305: \label{ham}
306: \ee
307: with the Q4 potential \eq{q4xy} are, in the Newtonian form,
308: \be
309: \ddot x + x\,(x^2+\alpha\,y^2) = 0\,, \qquad
310: \ddot y + y\,(y^2+\alpha\,x^2) = 0\,.                    \label{eom}
311: \ee
312: In the following we expand the solutions of these equations around an
313: arbitrary bifurcation point $\alpha_n$ into Taylor series
314: \be
315: x(t) = x_A(t) + \epsilon\,x_1(t) + \epsilon^2 x_2(t) + \dots\,,
316: \qquad 
317: y(t) = y_A(t) + \epsilon\,y_1(t) + \epsilon^2 y_2(t) + \dots\,,
318: \label{exp}
319: \ee
320: whereby the small dimensionless expansion parameter $\eps$ is chosen, 
321: with support from numerical evidence, as
322: \be
323: \epsilon = \sqrt{|\alpha - \alpha_n|} \,.                \label{eps}
324: \ee
325: As zero-order solution, we have taken $x_A(t)=0$ and $y_A(t)=y_0\,{\rm 
326: E}_1(z)$ as given in \eq{yaq4} for the A orbit, since all bifurcated 
327: orbits are degenerate with the A orbit at the bifurcation points. 
328: Whereas this unperturbed solution is the same for all bifurcations, the 
329: corrections $x_k(t)$, $y_k(t)$ with $k=1,2,\dots$ will depend explicitly 
330: on the value $\alpha_n$ of the chosen bifurcation point. 
331: 
332: We now rewrite the exact equations of motion \eq{eom}, singling out the 
333: value of $\alpha_n$ and replacing the difference $\alpha-\alpha_n$ by 
334: $\epsilon^2$:
335: \be
336: \ddot x + \alpha_n\,x\,\,y^2 + \epsilon^2 x\,y^2 + x^3 = 0\,,\qquad
337: \ddot y + y^3 + \alpha_n\,y\,x^2 + \epsilon^2 y\,x^2 = 0\,.\label{eomp}
338: \ee 
339: Hereby we have chosen $\alpha\geq\alpha_n$ which holds for all bifurcated 
340: orbits with $n\geq 1$. The orbit L$_3$ ($n=0$) and the orbits F$_6$, P$_7$ 
341: ($n=1/2$, see \cite{lamp}) bifurcate towards smaller values of $\alpha$, ie 
342: $\alpha\leq\alpha_0=0$ and $\alpha\leq\alpha_{1/2}=3/8$, respectively; for 
343: these cases the sign in front of $\epsilon^2$ in the equations \eq{eomp} 
344: must be reversed.
345: 
346: Next we insert \eq{exp} into \eq{eomp} and extract the equations obtained 
347: separately at each order in $\epsilon$. This leads to a recursive sequence 
348: of {\it linear} second-order differential equations in the scaled time 
349: variable $z$:
350: \be
351: x_k''(z) + \alpha_n\,\cn^2(z)\,x_k(z) = h_k(z)\,, \qquad
352: y_k''(z) + 3\,\cn^2(z)\,y_k(z)        = g_k(z)\,, \qquad
353:            k=1,2,3,\dots                                   \label{keq}
354: \ee
355: where the inhomogeneities on the rhs contain nonlinear combinations of 
356: $x_{k'}(t)$ and $y_{k'}(t)$ with $k'<k$. The homogeneous parts of these 
357: equations are identical to \eq{lameq4}, \eq{lameq} and have the periodic 
358: Lam{\'e} polynomials $\En(z)$ as solutions. According to their general 
359: theory \cite{erde,inc1} the second, linearly independent solutions are 
360: non-periodic for integer $n$; we shall in the following denote them by 
361: $\Fn(z)$. We normalize them such that their Wronskians with the $\En$ 
362: become unity: $W\!\left\{\En(z),\Fn(z)\right\}=1$. The solutions $\En$ 
363: and $\Fn$ are then related to each other by \cite{wiwa}
364: \be
365: \Fn(z) = \En(z) \int_0^z \frac{dz'}{[ \En(z')]^2}\,,   \qquad
366: \En(z) = \Fn(z) \int_z^0 \frac{dz'}{[ \Fn(z')]^2}\,.
367: \label{fsol}
368: \ee
369: In \tab{EnFn} we give the solutions $\En(z)$ and $\Fn(z)$ for the lowest
370: integer $n$. Hereby we have defined
371: \be
372: \Eps(z) = 2E(z)-z = \int_0^z \!\cn^2(u)\,du\,.
373: \label{Eps}
374: \ee
375: $E(z)$ is the incomplete elliptic integral of second kind, related
376: to that of the first kind $F(\varphi,\kappa)$ by
377: \be
378: E(z) = E(\varphi,\kappa)
379:      = \int_0^\varphi [1-\kappa^2\sin^2\theta]^{1/2}\,d\theta\,,\qquad 
380:    z = F(\varphi,\kappa)\,
381:      = \int_0^\varphi [1-\kappa^2\sin^2\theta]^{-1/2}\,d\theta\,. 
382: \ee
383: The function $\Eps(z)$ in \eq{Eps} is nonperiodic; its periodic part
384: is given, with $\E=E(\kappa)=E(\frac{\pi}{2},\kappa)$, by
385: \be
386: \mathrm{per}\{\Eps(z)\} = 2E(z) - z - \frac{\pi}{2\K^2}\,z
387:                         = 2E(z) - \frac{2\E}{\K}\,z\,.  \label{Epsper}
388: \ee
389: The last equality above follows from a known relation \cite{grry} 
390: between the complete elliptic integrals $\K$ and $\E$ for $\kappa^2=1/2$.
391: 
392: \Table{EnFn}{16.6}{
393: \begin{tabular}{|r|r|l|l|}
394: \hline
395: $n$ & $\alpha_n$ & E$_n(z)$ & F$_n(z)$ \\
396: \hline 
397: 0 & 0 & E$_0(z)$ = Ec$_0^0(z)$ = \,1 & F$_0(z)$ = $z$ \\
398: 1 & 1 & E$_1(z)$ = Ec$_1^1(z)$ = \,cn$(z)$                    
399:       & F$_1(z)$ = $2\,\sn(z)\,\dn(z) - \cn(z)\,\Eps(z)$ \\
400: 2 & 3 & E$_2(z)$ = Es$_2^1(z)$ = \,dn$(z)$\,\sn$(z)$          
401:       & F$_2(z)$ = $z\,\dn(z)\,\sn(z) - \cn(z)$ \\
402: 3 & 6 & E$_3(z)$ = Es$_3^2(z)$ = \,cn$(z)$\,dn$(z)$\,\sn$(z)$ 
403:       & F$_3(z)$ = $2 - 3\,\cn^4(z)-3\,\cn(z)\,\sn(z)\,\dn(z)\,\Eps(z)$\\
404: \hline
405: \end{tabular}
406: }{~The first four pairs of orthogonal solutions E$_n(z)$, F$_n(z)$ of 
407: the Lam{\'e} equation \eq{lameq}. We use the short notation
408: $\cn(z)=\cn(z,\kappa)$, $\sn(z)=\sn(z,\kappa)$ etc; here $\kappa^2=1/2$.
409: }
410: 
411: \vspace*{-0.5cm}
412: 
413: For half-integer $n$ -- appearing at the period-doubling bifurcations of 
414: the A orbits leading to the algebraic Lam{\'e} functions \cite{inc2,erd2} 
415: -- the linearly independent solutions $\Fn(z)$ are also periodic; the 
416: development given below must then be modified at some points. For 
417: simplicity, we limit ourselves in the following to the integer-$n$ cases 
418: occurring at the bifurcations of the primitive A orbits.
419: 
420: The general solutions of the equations \eq{keq} are of the standard form
421: \be
422: x_k(z) = c_k\,\En(z)  + d_k\,\Fn(z)  + H_k(z)\,,          \qquad
423: y_k(z) = a_k\,{\rm E}_2(z) + b_k\,{\rm F}_2(z) + G_k(z)\,,\label{ksol}
424: \ee
425: where the particular solutions of the inhomogeneous equations are given by
426: \bea
427: H_k(z) & = & \int_0^z h_k(z')\,[\,\Fn(z)\,\En(z')
428:                             - \Fn(z')\,\En(z)\,]\,dz'\nonumber\\
429:        & = & \En(z)\int_0^z du\, \frac{1}{[\En(u)]^2}
430:              \int_0^u h_k(w)\,\En(w)\,dw \;\;\;\;
431: \label{xkinh}
432: \eea
433: and
434: \bea
435: G_k(z) & = & \int_0^z g_k(z')\,[\,{\rm F}_2(z)\,{\rm E}_2(z')
436:                            - {\rm F}_2(z')\,{\rm E}_2(z)\,]\,dz'\nonumber\\
437:        & = & {\rm E}_2(z)\int_0^z du\, \frac{1}{[{\rm E}_2(u)]^2}
438:              \int_0^u g_k(w)\,{\rm E}_2(w)\,dw\,.\quad
439: \label{ykinh2}
440: \eea
441: The second parts of the above equations are useful for analytical
442: computations. The coefficients $a_k$, $b_k$, $c_k$ and $d_k$ in \eq{ksol} 
443: are determined recursively by requiring $x_k(z)$ and $y_k(z)$ to be 
444: periodic and to have the same symmetries as the lowest-order solutions 
445: $y_0(z)=y_A(z)$ and $x_1(z)=c_1\En(z)$, since these symmetries cannot be 
446: changed by varying $\alpha$ away from the bifurcations. The latter 
447: requirement leads immediately to $a_k=0$ for all $k$. Requiring $x_k(z)$ 
448: to be periodic allows us for $k\geq3$ to determine the constant $c_{k-2}$, 
449: appearing in different powers on the rhs of \eq{keq}, and (for any $k$) 
450: the constant $d_k$. Periodicity of $y_k(z)$ determines $b_k$ in terms of 
451: the $c_{k'}$ with $k'<k$. We find that $y(z)$ and $x(z)$ are overall even 
452: and odd functions of $\eps$, respectively, so that 
453: $c_{2k}=d_{2k}=b_{2k+1}=0$ and hence $x_{2k}(z)\equiv y_{2k+1}(z)\equiv 0$ 
454: for all $k=0,1,2,\dots$ Most of the integrals can be done analytically.
455: 
456: Up to this point, we have not respected the fact that the new bifurcated 
457: orbits develop their own periods which deviate from $\TA$ when moving 
458: away from the bifurcation points $\alpha_n$. The solutions $y_k(z)$ and 
459: $x_k(z)$ outlined above have, indeed, all the same period $\TA$ as the A 
460: orbit. A consequence of this is that these solutions do not conserve the
461: total energy $E$ as a function of $\alpha$. In the conventional perturbation 
462: theory, these unwanted effects are avoided by expanding the frequencies 
463: (or periods) of the perturbed system in powers of $\epsilon$ consistently 
464: along with the coordinates \eq{exp}. In our present case, this would have 
465: introduced more unknown parameters at each order, and the procedure of 
466: their determination would have become rather tedious. We have therefore
467: used an alternative approach by an {\it a posteriori} rescaling of the 
468: dimensionless argument $z=y_0t$ in the above solutions. At a given order 
469: of the perturbation expansion, we set $z=wt$ and determine the value of 
470: $w$ by the expansion of the new periods
471: \be
472: T = \frac{4\K}{w} = \TA \left[1 + \epsilon\,\tau_1 + \epsilon^2\tau_2 + 
473:                     \dots\right],                          \label{period}
474: \ee
475: leading to
476: \be
477: w = y_0\!\left[ 1 + \epsilon\,\tau_1 + \epsilon^2\,\tau_2 + \dots
478:     \right]^{-1}.                                             \label{wk}
479: \ee
480: The coefficients $\tau_k$ can be determined by writing the total energy 
481: in terms of the series \eq{exp} as
482: \be
483: E = \frac12\, ({\dot x}^2+{\dot y}^2)
484:   + \frac14\, (x^4+y^4) + \frac{\alpha}{2}\, x^2 y^2
485:   = E_0 + \epsilon\,E_1 + \epsilon^2\,E_2 + \dots\,,         \label{eq4}
486: \ee
487: inserting the above solutions for $x_k(wt)$ and $y_k(wt)$ with $w$ given
488: by \eq{wk}, and imposing that the energy $E$ be independent of $\epsilon$ 
489: (ie, of $\alpha$), which means that $E_0=E$ and $E_k=0$ for $k>0$. Whereas 
490: the lowest-order equation $E_0=E$ is trivially fulfilled for the solutions 
491: $x_A$, $y_A$ in \eq{yaq4}, the $E_k$ evaluated in terms of the above
492: solutions $x_k(y_0t)$, $y_k(y_0t)$ for $k>0$ are, indeed, not all equal to 
493: zero. However, in terms of the rescaled solutions $x_k(wt)$, $y_k(wt)$ we 
494: can impose $E_k=0$ for all $k>0$ successively by choosing the coefficients 
495: $\tau_k$ appropriately. An alternative way of deriving the same results 
496: (and thereby to check the algebra) is to consider the perturbation 
497: expansion \eq{eq4} with new ``variables'' (constants of the motion) $E_k$ 
498: which specify the total energy $E$. The energy conservation equation 
499: \eq{eq4} then leads at each order $k$ to a first-order differential 
500: equation which can be integrated analytically with respect to the 
501: perturbative solutions $y_k$, with the help of \eq{keq} for the $x_k$. 
502: From their periodicity conditions we obtain the $E_k$, and hence
503: the $\tau_k$ valid for all values of $\alpha$ and $E$. 
504: We found for all orbits investigated here that $\tau_1=\tau_2=\tau_3=0$ 
505: and $\tau_4\neq0$, so that $T$ varies with $\alpha$ like 
506: \be
507: T(\alpha) = \TA\!\left[1+\tau_4\,(\alpha-\alpha_n)^2+\dots\right],
508: \label{period4}
509: \ee 
510: whereby $\tau_4$ is an energy-independent constant. Table \ref{t4tab}
511: contains the values of $\tau_4$ of the first eight stable orbits born 
512: at the pitchfork bifurcations of A. They will be discussed further in 
513: section 3.2. We see that the L$_3$ orbit plays a special role; note that 
514: this orbit only exists at $\alpha\leq\alpha_0=0$, whereas all the other 
515: orbits given in the table exist only at $\alpha\geq\alpha_n$.
516: 
517: \Table{t4tab}{16.6}{
518: \begin{tabular}{|r|r|l|l|}
519: \hline
520: $n$ & $\alpha_n$ & O$_n$ & $\tau_4(n)$ \\
521: \hline 
522: 0  & 0   & L$_3$    & $3\pi^2\!/16\K^4$ \\
523: 3  & 6   & R$_5$    & $-$11/1260 \\
524: 5  & 15  & L$_7$    & $-$475/354816 \\
525: 7  & 28  & R$_9$    & $-$4807/12627300 \\
526: 9  & 45  & L$_{11}$ & $-$160425/1092591808 \\
527: 11 & 66  & R$_{13}$ & $-$75981/1115482060 \\
528: 13 & 91  & L$_{15}$ & $-$4626964303/129311102954880 \\
529: 15 & 120 & R$_{17}$ & $-$56892225/2767492640836 \\
530: \hline
531: \end{tabular}
532: }{~Lowest non-vanishing perturbation term $\tau_4$ in the period
533: \eq{period4} of the first eight stable orbits O$_n$ born at the 
534: bifurcations of the primitive orbit A. 
535: }
536: 
537: \vspace*{-0.5cm}
538: 
539: As an illustration of our method, we present the results for the orbit R$_5$ 
540: born at $\alpha_3=6$ and existing only for $\alpha\geq6$. We obtain the 
541: following analytical solutions up to order $\epsilon^4$ (with $z=wt$)
542: \bea
543: x_1(t) & = & c_1\,\sn(z)\,\dn(z)\,\cn(z)\,,\qquad
544:   x_3(t) = -\,c_1\,\sn(z)\,\dn(z)\,\cn(z)\left[\frac{25}{132} +
545:              \frac{7}{180}\,\cn^4(z)\right],\nonumber\\
546: y_0(t) & = & w\,\cn(z) \; = \; y_A(z)\,,\quad \qquad\! 
547:   y_2(t) = -\,\frac{c_1^2}{4w}\,\cn(z)\,[1+\cn^4(z)]\,,\nonumber\\
548: y_4(t) & = & \frac{c_1^2}{w}\,\cn(z)\left\{\frac{1}{10\,080}
549:                \left[-\,797-206\,\cn^4(z)+111\,\cn^8(z)\right]
550:                +\frac{5}{264}\left[1+\cn^4(z)\right]\right\},\label{r5sol}
551: \eea
552: where $c_1 = w\sqrt{11/45}$, and $\tau_4=-11/1260$ so that 
553: $w=y_0/[1-(11/1260)\eps^4]$. 
554: 
555: \newpage
556: 
557: \Figurebb{y4R5}{42}{50}{760}{515}{q4papf2.eps}{10.7}{16.6}{
558: Coordinates $x(t)$ and $y(t)$ of the orbit R$_5$, evaluated at $\alpha=7$, 
559: ie $\epsilon=1$ (left side) and $\alpha=8$, ie $\epsilon=\sqrt{2}$ (right 
560: side). Solid lines: numerical results. Heavy dotted lines: perturbation 
561: series \eq{exp} including the terms in \eq{r5sol} up to 3rd and 4th order 
562: for $x(t)$ and $y(t)$, respectively. Other lines give partial 
563: contributions as labeled by the inserts. (Units such that $\hbar=1$; 
564: $E=1/6$.)
565: }
566: 
567: In figure \ref{y4R5} we compare the results obtained by summing the above
568: analytical results according to \eq{exp} up to order $\eps^4$, shown by 
569: the heavy dotted lines, with the results obtained by numerical solution 
570: of the exact equations of motion \eq{eom}, shown by the solid lines. The 
571: single contributions obtained at each order in $\eps$ are shown by the 
572: other lines, as labeled by the inserts. The left panels are calculated
573: for $\alpha=7$ where $\eps=1$, and the right panels for $\alpha=8$ where 
574: $\eps=\sqrt{2}$. The convergence is surprisingly good even when the 
575: expansion parameter $\eps$ is larger than unity. This is due to the 
576: rapidly decreasing amplitudes of the $x_k(z)$ and $y_k(z)$ with increasing 
577: $k$. Note that for $y(t)$, where the agreement is perfect for $\eps=1$ and 
578: still very good for $\eps=\sqrt{2}$, we have included three terms, whereas 
579: $x(t)$ only contains two terms. Obtaining $x_5(z)$ would have required to 
580: calculate both $y_6(z)$ and $x_7(z)$ and to make them periodic, which -- 
581: though analytically possible -- would have been rather cumbersome. But 
582: the fact that the remaining errors in $x(t)$ are of the same order as 
583: $\eps^4y_4(z)$ suggests that adding the third term $\eps^5x_5(z)$ would 
584: lead to an equally good convergence for $x(t)$. Similar results were 
585: also obtained for other bifurcated orbits. 
586: 
587: For the period-two orbits born at period-doubling bifurcations, the 
588: algebraic Lam{\'e} functions have to be used. The repeated integrations 
589: arising in the perturbation expansion then become more difficult and we 
590: could not do all of them analytically. Resorting to numerical 
591: integrations, however, whereby the coefficients $b_k$, $c_k$ and $d_k$ 
592: in \eq{ksol} were determined numerically by iteration, we could reach a 
593: similar convergence of the perturbation series.
594: 
595: \newpage
596: 
597: \section{Semiclassical trace formulae for the density of states}
598: 
599: In this section we shall apply our perturbative results to investigate
600: the role of the pitchfork bifurcations of the A orbit in semiclassical 
601: calculations of the density of states. We coarse-grain both the 
602: exact and the semiclassical density of states by a convolution with a 
603: normalized Gaussian $\exp\{-(E/\gamma)^2\}/\gamma\sqrt{\pi}$. The 
604: quantum-mechanical coarse-grained density of states is then given by 
605: \be
606: g_{qm}(E) = \frac{1}{\gamma\sqrt{\pi}}\sum_n \exp\{-[(E-E_n)/\gamma]^2\}  
607: \label{gqm}
608: \ee
609: in terms of the exact quantum spectrum $\{E_n\}$ which we have obtained 
610: by diagonalization of \eq{q4xy} in a harmonic oscillator basis. For 
611: larger values of $\gamma$, the prominent gross-shell structure in the 
612: density of states is emphasized while finer details of its oscillations 
613: are suppressed. In the limit $\gamma\to 0$ the sum of delta functions 
614: is recovered. The oscillating part of \eq{gqm} is defined as 
615: \be
616: \delta g_{qm}(E) = g_{qm}(E)-g_{TF}(E)\,,                  \label{dgqm}
617: \ee 
618: where $g_{TF}(E)$ is the average part obtained in the Thomas-Fermi 
619: approximation which for the Q4 potential can be calculated analytically:
620: \bea
621: g_{TF}(E)&=&\frac{2}{\pi\hbar^2}\,K\!\left(\!\sqrt{(1-\alpha)/2}\,\right)\!
622:             \sqrt{E}\hspace{3.6cm}\hbox{for }-1<\alpha\leq +1\,,\nonumber\\
623:          &=&\frac{2}{\pi\hbar^2}\,\sqrt{\frac{2}{\alpha+1}}\,
624:             K\!\left(\!\sqrt{(\alpha-1)/(\alpha+1)}\,\right)\!\sqrt{E}
625:             \qquad\quad\hbox{for }\;\alpha\geq +1\,.          \label{gtf}
626: \eea
627: Higher-order $\hbar$ corrections to the average density of states 
628: \cite{book} are negligible in the present system. 
629: 
630: The semiclassical trace formula for the density of states has the general 
631: form
632: \be
633: \delta g_{sc}(E) = \frac{1}{\pi\hbar}\sum_{po}{\cal A}_{po}(E)\,
634:                    f_\gamma(T_{po})\,
635:                    \cos\left[\frac{1}{\hbar}\,S_{po}(E)
636:                    -\sigma_{po}\frac{\pi}{2}\right].           \label{dgtf}
637: \ee
638: The sum goes over all periodic orbits ($po$) of the classical system. 
639: $S_{po}(E)$ are the action integrals $\oint {\bf p}\cdot d{\bf q}$ along 
640: the periodic orbits and $\sigma_{po}$ the so-called Maslov indices. The 
641: amplitudes ${\cal A}_{po}(E)$ depend on the number of constants of motion 
642: of the system. The factor $f_\gamma(T_{po})$ is given by \cite{book}
643: \be
644: f_\gamma(T_{po}) = \exp\left[{-(\gamma\,T_{po}/2\hbar)^2}\right],
645: \ee
646: where $T_{po}=dS_{po}(E)/dE$ are the periods of the orbits. This 
647: coarse-graining factor favours the contributions of the shortest
648: orbits to the gross-shell structure obtained with larger values of 
649: $\gamma$. When the system possesses no other constant of the motion
650: besides the energy, all orbits are isolated and Gutzwiller's original 
651: form \cite{gutz} of the amplitudes ${\cal A}_{po} (E)$ applies. In the 
652: presence of continuous symmetries, and for integrable systems in general, 
653: other forms must be used (see \cite{book} for a survey of trace formulae 
654: and the calculation of ${\cal A}_{po}$ and $\sigma_{po}$). 
655: 
656: In the presence of bifurcations, the Gutzwiller amplitudes ${\cal A}_{po}
657: (E)$ for isolated orbits cannot be used. However, the uniform 
658: approximation for generic pitchfork bifurcations \cite{ozha,ssun} can 
659: be applied to describe the isochronous bifurcations of the A orbit 
660: discussed in this paper. The factor 2 appearing in the period doubling 
661: for the generic case here plays the role of the extra degeneracy factor 
662: 2 of the bifurcated orbit pairs, which is due either to their reflection 
663: symmetry at the symmetry axis containing the A orbit (for the librating
664: orbits L$_n$) or to the time reversal symmetry (for the rotating orbits 
665: R$_n$). Adapting the results of \cite{ssun} to the present system, the 
666: combined contribution to the semiclassical density of states from all 
667: the orbits participating in the bifurcation at $\alpha_n$ (including 
668: their degeneracy factors) is given by
669: \be
670: \delta g_{sc}^{bif}(E,\alpha_n) = \frac{\TA\,\Gamma(1/4)}
671:                 {\pi\sqrt{2\pi}\,\hbar^{5/4}|a_n|^{1/4}k^{3/4}}\,
672:                 \cos\left[\frac{k\SA}{\hbar}-k\nu_n\frac{\pi}{2}
673:                 -(-1)^n\frac{\pi}{4}-{\rm sign}(a_n)\frac{\pi}{8}\right]. 
674: \label{levbif}
675: \ee
676: Here $k$ is the repetition number of the orbits and $\nu_n$ the Maslov 
677: index, corresponding to $\sigma_{po}$ in \eq{dgtf}, of the unstable 
678: primitive orbit involved. The parameter $a_n$ stems from the normal form 
679: of the action function $S(q',p)$ used in the phase-space representation 
680: of the trace integral at the $n$-th bifurcation of the A orbit. From the
681: expansions given in \cite{ssun} for the properties of the periodic orbits 
682: near the bifurcation, and using our results in the previous section, we 
683: can determine $a_n$ to be 
684: \be
685: a_n = \frac{1}{4S_{\rm A}}\left(\frac{4\pi}{2n+1}\right)^2
686:       \frac{1}{\tau_4(n)}\,.                            \label{an}
687: \ee
688: 
689: In the following, we want to examine the importance of the contribution 
690: \eq{levbif} in relation to that of other non-bifurcating orbits of the 
691: system. Quite generally, in an integrable two-dimensional system the 
692: leading families of degenerate orbits have an amplitude proportional to
693: $\hbar^{-3/2}$ (except, eg, for harmonic oscillators with rational frequency 
694: ratios, where the leading amplitudes go like $\hbar^{-2}$, see \cite{brja}). 
695: They are therefore expected to dominate over the bifurcating orbits which 
696: here have an amplitude proportional to $\hbar^{-5/4}$ as seen in \eq{levbif}. 
697: In section 3.1 below we will test the relative importance of the latter 
698: for the particular integrable situation given in our system for $\alpha=0$. 
699: On the other hand, the standard Gutzwiller amplitudes for isolated orbits 
700: in a non-integrable system go like $\hbar^{-1}$, so that the bifurcating 
701: orbits can be expected to play a larger relative role. As we will see in 
702: section 3.2 their relative weight is, however, subject to a rather subtle 
703: balance between the stability of the shortest isolated orbits and the 
704: $\alpha$ dependence of the factor $|a_n|^{-1/4}$ in the amplitude of 
705: \eq{levbif}. 
706: 
707: \subsection{The separable case $\alpha=0$}
708: 
709: For $\alpha=0$ the potential \eq{q4xy} is separable and thus defines
710: an integrable system in which the majority of the periodic orbits live 
711: on a 2D torus in phase space. The EBK quantization can therefore be 
712: applied to obtain a semiclassical quantum spectrum from which a trace 
713: formula is readily derived \cite{beta}. However, the system possesses  
714: also the isolated orbit A which undergoes a bifurcation at $\alpha=0$. 
715: As we see from \fig{q4trm} the orbit A, which is stable (A$_3$) for 
716: $0<\alpha<1$, becomes unstable (A$_2$) for $\alpha<0$; the orbit L$_3$ 
717: bifurcating from it exists only for $\alpha\leq0$ and is stable for 
718: $-0.5\siml\alpha<0$. This gives a nontrivial contribution to the 
719: density of states whose role will be investigated numerically below.
720: 
721: Straightforward EBK quantization of the 2D torus gives the approximate 
722: spectrum
723: \be
724: E^{\rm EBK}_{n_xn_y} = (1/4)\,(6\pi\hbar/4\K)^{4/3}        
725:                        \left[(n_x+1/2)^{4/3}+(n_y+1/2)^{4/3}\right].
726:                        \qquad (n_x,n_y=0,1,2,\dots)      \label{ewkb} 
727: \ee
728: From this we obtain in the standard way \cite{beta,crli} the following 
729: Berry-Tabor type trace formula to leading order in $\hbar$:
730: \be
731: \delta g_{sc}^{2D}(E) = \left(\frac{4\K}{2\pi\hbar}\right)^{\!3/2}\!
732:                          (4E)^{1/8}\; 2\!\!\! \sum_{n,m=1}^\infty(-1)^{n+m}\,
733:                          \frac{nm}{(n^4+m^4)^{5/8}}\,f_\gamma(T_{nm})\,
734:                          \cos\left[\frac{1}{\hbar}\,S_{nm}(E)
735:                                    -\frac{\pi}{4}\right]. \label{dgsc2}
736: \ee
737: The classical actions and periods
738: \be
739: S_{nm}(E)=(4\K/3)\,(4E)^{3/4}\,(n^4+m^4)^{1/4}, \qquad
740: T_{nm}(E)=4\K\,(4E)^{-1/4}\,(n^4+m^4)^{1/4}        \label{snm}
741: \ee
742: of the 2D rational tori ($n,m$) correspond to the periodic orbits
743: \be
744: x(t)=w_x\,\cn(w_xt+\phi)\,, \qquad y(t)=w_y\,\cn(w_yt)\,, \quad\;\,
745: \ee
746: with
747: \be
748: w_x=(4E)^{1/4}\,n\,(n^4+m^4)^{-1/4}, \qquad 
749: w_y=(4E)^{1/4}\,m\,(n^4+m^4)^{-1/4}.
750: \ee
751: These orbits form degenerate families described by the parameter
752: $\phi\in[0,4\K)$. The $(n,n)$ resonances are the families containing
753: the $n$-th repetitions of the orbits B (for $\phi=0$ or $2\K$) and C 
754: (for $\phi=\K$ or $3\K$). Summation over all resonances $(n,m)$ with 
755: $n,m>0$ yields the trace formula \eq{dgsc2}.
756: 
757: \newpage
758: 
759: However, the system also contains the isolated resonances $(n,0)$ and 
760: ($0,m$) with $n,m=1,2,3,\dots$ which correspond to the (repeated) 
761: one-dimensional A orbits in the $x$ and $y$ direction. With the uniform
762: approximation \eq{levbif} discussed above, we can include them in the
763: trace formula, together with the orbits L$_3$ born at the bifurcation
764: A$_3\to$ A$_2$ + L$_3$. This gives the following common contribution of 
765: all A and L orbits and their $k$-th repetitions
766: \be
767: \delta g_{sc}^{\rm AL}(E)= \frac{(4\K)^{3/4}}{(\pi\hbar)^{5/4}}\,
768:                            (4E)^{-1/16} \sum_{k=1}^\infty (-1)^k
769:                            \frac{1}{k^{3/4}}\,f_\gamma(k\TA)\,
770:                            \cos\left[\frac{k}{\hbar}\SA(E)-\frac{3\pi}{8}
771:                            \right]\!,                       \label{dgsc1} 
772: \ee
773: where $\SA(E)=(4\K/3)\,(4E)^{3/4}$ is the action of the primitive A orbit. 
774: Hereby we have used the value $\tau_4=3\pi^2\!/16\K^4$ for the L$_3$ orbit 
775: as given in \tab{t4tab}, and the relation \cite{grry} 
776: $\Gamma(1/4)=(4\sqrt{\pi}\,\K)^{1/2}$.
777: 
778: The total oscillating part of the semiclassical density of states at 
779: $\alpha=0$ then is given by the sum of the contributions \eq{dgsc2} and 
780: \eq{dgsc1}. We observe that the latter is of order $\hbar^{1/4}$ relative 
781: to the leading-order contributions of the 2D torus families. (For isolated 
782: orbits, this relative factor would be $\hbar^{1/2}$ as mentioned above.) 
783: We therefore expect that the bifurcating orbits have a non-negligible 
784: influence on the density of states, at least for low energies where the 
785: negative power of $E$ in the amplitude of \eq{dgsc1} does not suppress 
786: their contribution too much. 
787: 
788: In \fig{dg_1} we show the coarse-grained level densities obtained with 
789: $\gamma=1$ (in units such that $\hbar=1$); the periodic orbit sums in 
790: \eq{dgsc2} and \eq{dgsc1} can here be limited to $n_{max}=m_{max}=
791: k_{max}=2$. In the upper panel, we show separately the semiclassical 
792: results \eq{dgsc2} of the 2D tori (solid line) and \eq{dgsc1} of the 
793: bifurcating orbits (dotted line). Both are seen to have a monotonously 
794: increasing amplitude of the oscillations, whereas the quantum result 
795: (dashed line) exhibits a pronounced beating structure. When adding both 
796: semiclassical contributions, as shown in the lower panel, the quantum 
797: result is nicely reproduced. The quantum beat is clearly the result of 
798: the interference between the shortest orbits of the torus family on one 
799: hand and the bifurcating isolated orbits A and L$_3$ on the other hand.
800: 
801: \Figurebb{dg_1}{10}{35}{795}{485}{q4papf3.eps}{9}{16.6}{
802: Oscillating part of the coarse-grained density of states 
803: at $\alpha=0$ (units such that $\hbar=1$; Gaussian smoothing width 
804: $\gamma=1$). Dashed lines: quantum-mechanical results \eq{dgqm}. 
805: Solid and dotted lines: semiclassical results with $n_{max}=m_{max}=
806: k_{max}=2$. Upper panel: separate contributions \eq{dgsc2} of 2D tori
807: and \eq{dgsc1} of bifurcating orbits A + L$_3$. Lower panel: the solid
808: line is here the sum of both semiclassical contributions.
809: }
810: 
811: \newpage
812: 
813: \Figurebb{dg_06}{20}{35}{795}{500}{q4papf4.eps}{9.2}{16.6}{
814: The same as \fig{dg_1}, but for the total densities of states including 
815: the average part \eq{gtf}, with $\gamma=0.06$. Periodic orbit sums are 
816: cut at $n_{max}=m_{max}=k_{max}=30$.
817: }
818: 
819: The effect of the bifurcating orbits is less dramatic when we choose a 
820: finer resolution of the energy spectrum, obtained with a smaller value
821: of $\gamma$. In \fig{dg_06} we show the total density of state including 
822: the average part \eq{gtf}, obtained for $\gamma=0.06$. Hereby periodic 
823: orbits up to $n_{max}=m_{max}=k_{max}\simeq 30$ contribute to the 
824: semiclassical results. We have normalized $g\,(E)$ here by the factor 
825: $\gamma\sqrt{\pi}$ so that the quantum result exhibits the correct 
826: degeneracies 1 or 2. Some apparently wrong higher degeneracies are badly 
827: resolved accidental (near-)degeneracies. At first glance there is not 
828: much of a difference between the upper and lower panels of \fig{dg_06}. 
829: However, a closer inspec\-tion reveals that the semiclassical degeneracies 
830: are wrong, with errors of up to $\sim 20\%$, in the upper part where only 
831: the 2D tori are included. They come much closer to the exact ones when 
832: including the bifurcating orbits, as seen in the lower panel. There also 
833: the bottom regions between the peaks are imp\-roved, the remaining small 
834: oscillations being numerical noise. Note that the inclusion of the 
835: bifurcating orbits does not affect the semiclassical peak positions which 
836: are exactly those of the spectrum $E_{nm}^{\rm EBK}$ given in \eq{ewkb}. 
837: The shifts in the peaks seen at the lowest energies are due to the typical 
838: errors inherent in the EBK approximation, which rapidly decrease with 
839: increasing energy.
840: 
841: \subsection{The non-integrable case $\alpha=6$}
842: 
843: As an example of a non-integrable situation we choose $\alpha=6$ where the 
844: rotating orbit R$_5$ is born at the bifurcation A$_5\to$ A$_6$ + R$_5$. In 
845: \fig{poinc6} we show a Poincar{\'e} surface of section $(y,p_y)$ taken at 
846: $x=0$. The large regular island in the middle contains the orbits A and 
847: R$_5$ at the origin. At its border we see an island chain of eight pairs 
848: of stable and unstable orbits born from a period-quadrupling bifurcation 
849: of the A orbit at $\alpha=5.4306$; the scaling properties of these fixed 
850: points were investigated in \cite{lakh}. The four drop-like small regular 
851: islands around $(y,p_y)=(\pm 0.431,\pm 0.362)$ contain the stable librating 
852: orbits S$_3$ that are born from the B orbits in the isochronous pitchfork 
853: bifurcation B$_3\to$ B$_2$ + S$_3$ at $\alpha=3$, as seen in \fig{q4trm}. 
854: [This is the same scenario as A$_3 \to$ A$_2$ + L$_3$ at $\alpha=0$, 
855: obtained after the transformation $\alpha\to(3-\alpha)/(1+\alpha)$ 
856: discussed in the introduction.] The orbits S$_3$ are still stable at 
857: $\alpha=6$. Since their common periods and actions are smaller than those 
858: of the orbit A, we expect them to influence the gross-shell structure in 
859: the density of states.
860: 
861: \Figurebb{poinc6}{-61}{37}{811}{520}{q4papf5.eps}{8.6}{16.6}{
862: Poincar{\'e} surface of section (taken at $x=0$) for $\alpha=6$.
863: See text for a discussion of the visible regular islands.
864: }
865: 
866: Clearly, there is no hope of obtaining full semiclassical quantization
867: by summing over all periodic orbits in this mixed phase-space system, 
868: where many orbits of longer periods will also be close to bifurcations. 
869: However, we may try to obtain the coarse-grained density of states using 
870: only the shortest periodic orbits, as in \fig{dg_1} above, and thus
871: using a cut-off $k_{max}$ of the sum over the repetition number $k$. The 
872: contribution of the four S$_3$ orbits to the semiclassical trace formula
873: has the standard form for a stable isolated orbit \cite{gutz,mill}:
874: \be
875: \delta g_{sc}^{S_3}(E)=\frac{2}{\pi\hbar}\,T_{{\rm S}_3}(E)\sum_{k=1}^{k_{max}}
876:             \frac{(-1)^k}{\sin(k\chi_{{\rm S}_3}/2)}\,f_\gamma(kT_{{\rm S}_3})
877:                          \sin\left[\frac{k}{\hbar}\,S_{{\rm S}_3}(E)\right], 
878:                                                             \label{dgS3}
879: \ee
880: whereby the period $T_{{\rm S}_3}(E)$, the action integral $S_{{\rm 
881: S}_3}(E)$ and the stability angle $\chi_{{\rm S}_3}$ have been determined 
882: numerically. The contribution from the bifurcating A and R orbits is 
883: given by \eq{levbif} with $\nu_n=6$ and using \eq{an} with $\tau_4=
884: -11/1260$ according to \tab{t4tab} for $n=3$.
885: 
886: Figure \ref{dg6} shows the coarse-grained density of states obtained
887: with a Gaussian smoothing width of $\gamma=1.0$, including only the
888: $k_{max}=2$ lowest harmonics of the shortest periodic orbits. The upper
889: panel gives the separate contributions of the isolated S$_3$ orbits 
890: (solid line) and of the bifurcating A and R orbits (dotted line); the
891: lower panel gives their sum. The dashed lines give the quantum-mechanical 
892: result in both panels. As in the case shown in \fig{dg_1}, each of the
893: single contributions give a monotonically increasing amplitude of the
894: oscillations in $\delta g_{sc}(E)$, and only their superposition yields a 
895: beating result that reproduces the quantum-mechanical result fairly well. 
896: The remaining discrepancies must be due to the missing contributions 
897: from other orbits with longer periods, of which there are already many at 
898: $\alpha=6$. 
899: 
900: In spite of the fact that their amplitude in \eq{levbif} is larger by a 
901: relative factor $\hbar^{-1/4}$ than that of the isolated orbits, the 
902: overall contribution of the bifurcating orbits to the coarse-grained
903: density of states is seen in the upper part of \fig{dg6} to be lower. 
904: This has two reasons. First, the shortest isolated orbits $S_3$ have a 
905: period that is substantially smaller than that of the orbits A and R$_5$, 
906: which leads to a larger value of the coarse-graining factor $f_\gamma$ 
907: in the trace formula. Second, the orbits $S_3$ are stable and have 
908: therefore a stability denominator of order unity, whereas the amplitude 
909: in \eq{levbif} has a larger overall denominator. Still, the uniform
910: A + R$_5$ contribution is strong enough to cause the beating interference 
911: pattern seen in the total $\delta g_{sc}(E)$.
912: 
913: \Figurebb{dg6}{-20}{35}{795}{505}{q4papf6.eps}{9.25}{16.6}{
914: Oscillating part of density of states as in \fig{dg_1}, but at $\alpha=6$.
915: Upper panel: separate contributions of the stable isolated orbit S$_3$ 
916: (solid line) and of the bifurcating orbits A + R$_5$ (dotted line). 
917: Lower panel: sum of orbits S$_3$ + A + R$_5$ (solid line). Truncation 
918: of the trace formulae \eq{levbif} and \eq{dgS3} at $k_{max}=2$, Gaussian 
919: averaging width $\gamma=1.0$. The dashed line in both panels gives the 
920: quantum-mechanical result.
921: }
922: 
923: It is now interesting to speculate about the relative contribution of
924: the bifurcating orbits when the parameter $\alpha$ is further increased
925: on the route towards chaos. On one hand, the average stability of the 
926: shortest isolated orbits can be expected to decrease. On the other hand, 
927: \fig{uniamp} reveals that the uniform amplitude in \eq{levbif} of the 
928: period-one ($k=1$) orbit A with its offsprings at the bifurcation 
929: 
930: \Figurebb{uniamp}{-120}{70}{795}{560}{q4papf7.eps}{8.3}{16.6}{
931: Amplitude of the uniform approximation \eq{levbif} for the bifurcating 
932: A orbit ($k=1$) to the semiclassical density of states ($E=1/6$). The 
933: crosses give the values obtained using the values of $\tau_4$ in 
934: \tab{t4tab} for $n>0$ at the bifurcation points $\alpha_n$; the solid 
935: line is an asymptotic fit for large $\alpha$.
936: }
937: 
938: \noindent
939: points also decreases with increasing $\alpha_n$. (The crosses give the 
940: analytical values obtained using the $\tau_4$ values from \tab{t4tab},
941: and the solid line is a numerical fit showing that the amplitude decays 
942: asymptotically like $\alpha^{-1/4}$.) Taking furthermore account of the 
943: fact that more and more new pairs of isolated orbits are created from 
944: the bifurcations along the route to chaos (and new orbits bifurcate from 
945: these again), even the number of shortest (period-one) isolated orbits 
946: increases very fast. We therefore tend to conclude that the relative
947: weight of the bifurcating orbits in the coarse-grained density of 
948: states will decrease with $\alpha$. It would, however, require a major 
949: numerical effort to study this balance in more detail and, in particular, 
950: to investigate the contributions from higher repetitions to the finer 
951: details of the density of states.
952: 
953: \section{Summary}
954: 
955: We have formulated a perturbative scheme to calculate analytically the 
956: shapes of periodic orbits bifurcating from the straight-line orbits in 
957: the quartic oscillator potential, exploiting the nice properties of the 
958: periodic Lam{\'e} functions in terms of Jacobi elliptic functions. In 
959: order to simplify the recursive determination of the unknown constants
960: in the perturbation expansion, we have used an unconventional {\it a 
961: posteriori} rescaling of the perturbed solutions that takes care of the 
962: varying periods of the bifurcated orbits.
963: 
964: For stable period-one orbits we are able to give analytical expressions 
965: of the resulting perturbative series. Even when the perturbation 
966: parameter $\eps$ is greater than unity, satisfactory convergence 
967: to the numerically obtained solutions can be reached by going up to 
968: order $\eps^4$ where the necessary algebraic work is still not too 
969: demanding.
970: 
971: We have studied semiclassical trace formulae for approximating the 
972: quantum-mechanical density of states of the quartic oscillator. In the 
973: separable case $\alpha=0$ we could give a completely analytical trace
974: formula, consisting of the contributions \eq{gtf}, \eq{dgsc2} and 
975: \eq{dgsc1}. It was shown numerically to give a very good approximation 
976: to the exact quantum-mechanical density of states. Its low-frequency 
977: part, extracted with an energy coarse-graining parameter $\gamma=1$, 
978: reveals a strong quantum beat. This beat can only be reproduced 
979: semiclassically when the shortest periodic orbit families are allowed 
980: to interfere with the isolated orbits A and L$_3$ taking part in a 
981: pitchfork bifurcation. The contribution of the latter was calculated 
982: using a slightly modified uniform approximation for generic pitchfork 
983: bifurcations given in \eq{dgsc1}. The high-resolution spectrum is 
984: dominated by the Berry-Tabor part \eq{dgsc2} which yields exactly the 
985: peaks corresponding to the EBK spectrum. The quantum degeneracies are, 
986: however, substantially affected by the contributions from the 
987: bifurcating orbits. 
988: 
989: In the non-integrable situation $\alpha=6$, which exhibits a strongly 
990: mixed phase space, we could also reproduce the coarse-grained quantum 
991: density of states semiclassically, including besides the bifurcating 
992: period-one orbits A and R$_5$ also the shortest stable isolated orbits 
993: S$_3$ which had to be obtained numerically. The two types of orbits 
994: interfere, again leading to a beat structure. Finally, we have shown 
995: the uniform contribution of the bifurcating A orbits and its offsprings 
996: to decrease like $\alpha^{-1/4}$ for further increasing values of the 
997: chaoticity parameter $\alpha$.
998: 
999: With these results we have demonstrated the possibility of accessing 
1000: analytically the periodic orbits in a system with mixed classical 
1001: dynamics. We hope to stimulate further research in this direction,
1002: in particular towards semiclassical studies of the role of periodic 
1003: orbits and their bifurcations in connection with spectral statistics 
1004: \cite{q4ls,spec}, which would have exceeded the scope of our present 
1005: short report. We also aim at further improving our understanding of 
1006: the uniformization of semiclassical trace formulae, including  
1007: bifurcations of higher codimensions and -- as a long-term (and perhaps 
1008: too ambitious?) goal -- including complete bifurcation cascades.
1009: 
1010: \newpage
1011: 
1012: \noindent
1013: {\Large \bf Note added in proof}
1014: 
1015: \bs
1016: 
1017: We have confirmed the result \eq{dgsc1} in an alternative way,
1018: without making use of any normal form. We start from the Poisson 
1019: summation-integration of the density of states 
1020: $\sum_{n_x,n_y}\delta(E-E_{n_x,n_y}^{EBK})$ using the 
1021: EBK spectrum \eq{ewkb}, like in the standard derivation \cite{beta}
1022: of \eq{dgsc2}. The first integration can be done exactly exploiting 
1023: the delta function. For the edge contributions corresponding to the 
1024: A and L$_3$ orbits, the usual stationary phase approximation cannot 
1025: be used and the second integration must be done more carefully. Its 
1026: asymptotic evaluation, valid here for $k\SA/\hbar\simg 2$, leads 
1027: precisely to the analytical result \eq{dgsc1}. 
1028: 
1029: \bs
1030: \bs
1031: 
1032: \noindent
1033: {\Large \bf Acknowledgements}
1034: 
1035: \bs
1036: 
1037: \noindent
1038: We are grateful to J{\"o}rg Kaidel for helpful discussions and Christian 
1039: Amann for providing us with a program for the numerical computation of 
1040: the quantum spectrum. SNF, AGM and MM acknowledge the hospitality of the
1041: University of Regensburg during their research visits and financial 
1042: support by the Deutsche Forschungsgemeinschaft.
1043: 
1044: \newpage
1045: 
1046: \begin{thebibliography}{31}
1047: 
1048: \setlength{\itemsep}{-0.25ex}
1049: 
1050: \bibitem{q4ls} Zimmermann Th, Meyer H-D, K{\"o}ppel H and Cederbaum L S
1051:                1986 {\it Phys.\ Rev.} A {\bf 33} 4334\\
1052:                Meyer H-D 1986 {\it J. Chem.\ Phys.} {\bf 84} 3147
1053: 
1054: \bibitem{q4po} Eckardt B 1988 {\it Phys.\ Rep.} {\bf 163} 205\\
1055:                Eckardt B, Hose G and Pollak E 1989 {\it Phys.\ Rev.} A 
1056:                {\bf 39} 3776  
1057: 
1058: \bibitem{btu}  Bohigas O, Tomsovic S and Ullmo U 1993 {\it Phys.\ Rep.} 
1059:                {\bf 223} 43
1060: 
1061: \bibitem{erda} Eriksson A B and Dahlqvist P 1993 {\it Phys.\ Rev.} E 
1062:                {\bf 47} 1002
1063: 
1064: \bibitem{lakh} Lakhshminarayan A, Santhanam M S and Sheorey V B 1996 
1065:                {\it Phys.\ Rev.\ Lett.} {\bf 76} 396
1066: 
1067: \bibitem{daru} Dahlqvist P and Russberg G 1990 {\it Phys.\ Rev.\ Lett.} 
1068:                {\bf 65} 2837
1069: 
1070: \bibitem{lamp} Brack M, Mehta M and Tanaka K 2001 {\it J. Phys.} A
1071:                {\bf 34} 8199  
1072: 
1073: \bibitem{grry} Gradshteyn I S and Ryzhik I M 1994 {\it Table of
1074:                Integrals, Series, and Products} (Academic Press, New
1075:                York, 5th edition) ch 8.1
1076: 
1077: \bibitem{erde} Erd{\'e}lyi A {\it et al} 1955 {\it Higher Transcendental 
1078:                Functions} vol 3 (New York: McGraw-Hill) ch 15
1079: 
1080: \bibitem{yosh} Yoshida H 1984 {\it Celest.\ Mech.} {\bf 32} 73 
1081: 
1082: \bibitem{inc1} Ince E L 1940 {\it Proc.\ R. Soc.} {\bf 60} 47 
1083: 
1084: \bibitem{inc2} Ince E L 1940 {\it Proc.\ R. Soc.} {\bf 60} 83
1085: 
1086: \bibitem{erd2} Erd{\'e}lyi A 1941 {\it Phil.\ Mag.} {\bf 32} 348
1087: 
1088: \bibitem{mbgu} Brack M 2001 {\it Foundations of Physics} vol 31, 
1089:                ed A Inomata {\it et al} p 209 
1090:                (Festschrift in honor of the 75th birthday 
1091:                of Martin Gutzwiller)\\
1092:                (Brack M 2000 LANL {\it preprint} nlin.CD/0006034)
1093: 
1094: \bibitem{wiwa} Whittacker E T and Watson G N 1969 {\it A course of modern
1095:                analysis} 4th edn (Cambridge: University Press)
1096: 
1097: \bibitem{gutz} Gutzwiller M C 1971 {\it J. Math.\ Phys.} {\bf 12} 343
1098: 
1099: \bibitem{book} Brack M and Bhaduri R K 1997 Semiclassical Physics 
1100:                {\it Frontiers in Physics} vol 96 (Reading, MA: 
1101:                Addison-Wesley) Revised paperback edition to appear in 
1102:                January 2003 (Boulder, CO: Westview Press)
1103: 
1104: \bibitem{ozha} Ozorio de Almeida A M and Hannay J H 1987 {\it J.\ Phys.} 
1105:                A {\bf 20} 5873
1106: 
1107: \bibitem{ssun} Schomerus H and Sieber M 1997 {\it J. Phys.} A {\bf 30} 4537,
1108:                and earlier references quoted therein
1109: 
1110: \bibitem{brja} Brack M and Jain S R 1995 {\it Phys.\ Rev.} A {\bf 51}, 3462
1111: 
1112: \bibitem{beta} Berry M V and Tabor M 1976 {\it Proc.\ R. Soc.} {\bf 349}, 101
1113: 
1114: \bibitem{crli} Creagh S C and Littlejohn R G 1992 {\it J. Phys.} 
1115:                A {\bf 25}, 1643
1116: 
1117: \bibitem{mill} Miller W H 1975 {\it J. Chem.\ Phys.} {\bf 63} 996
1118: 
1119: \bibitem{spec} Bogomolny E B and Keating J P 1996 {\it Phys.\ Rev.\ Lett.} 
1120:                {\bf 77} 1472\\
1121:                Berry M V, Keating J P and Prado S D 1998 {\it J. Phys.} A 
1122:                {\bf 31} L245\\
1123:                Sieber M and Richter K 2001 {\it Physica Scripta} {\bf T 90}
1124:                128 
1125: 
1126: \end{thebibliography}
1127: 
1128: \end{document}
1129: 
1130: