1: \documentclass[12pt]{iopart}
2:
3: \usepackage{iopams}
4: \usepackage{graphicx}
5: \usepackage{epsfig}
6: % \usepackage{amsmath}
7: % \usepackage{amssymb}
8:
9: \newcommand{\sgn}{\;\mbox{sgn}}
10: % \newcommand{\diag}{\;\mbox{diag}}
11:
12: \newcommand{\eps}{\varepsilon}
13: % \newcommand{\Tau}{{\cal T}}
14:
15: \newcommand{\la}{\langle}
16: \newcommand{\lla}{\left\langle}
17: \newcommand{\ra}{\rangle}
18: \newcommand{\rra}{\right\rangle}
19:
20: % \newcommand{\ts}{{\mbox{\tiny ts}}}
21: % \newcommand{\fs}{{\mbox{\tiny fs}}}
22: % \newcommand{\bs}{{\mbox{\tiny bs}}}
23: \newcommand{\mx}{{\mbox{\tiny max}}}
24: \newcommand{\SP}{{\mbox{\tiny SP}}}
25: \newcommand{\GOE}{{\mbox{\tiny GOE}}}
26: \newcommand{\Num}{{\mbox{\tiny Num}}}
27:
28: % \voffset1cm
29:
30: \begin{document}
31:
32: \bibliographystyle{unsrt}
33:
34: \title[Spectral properties of right triangles]
35: {Generic spectral properties of right triangle billiards}
36:
37: \author{Thomas Gorin}
38: \address{Centro de Ciencias Fisicas, UNAM, Cuernavaca, Morelos, Mexico}
39: \address{Centro Internacional de Ciencias, Cuernavaca, Morelos, Mexico}
40:
41: \ead{gorin@cicc.unam.mx}
42:
43: \begin{abstract}
44: This article presents a new method to calculate eigenvalues of right triangle
45: billiards. Its efficiency is comparable to the boundary integral method and
46: more recently developed variants. Its simplicity and explicitness however allow
47: new insight into the statistical properties of the spectra. We analyse
48: numerically the correlations in level sequences at high level numbers ($>10^5$)
49: for several examples of right triangle billiards. We find that the strength of
50: the correlations is closely related to the genus of the invariant surface of
51: the classical billiard flow. Surprisingly, the genus plays and important
52: r\^ ole on the quantum level also. Based on this observation a mechanism
53: is discussed, which may explain the particular quantum-classical correspondence
54: in right triangle billiards. Though this class of systems is rather small, it
55: contains examples for integrable, pseudo integrable, and non integrable
56: (ergodic, mixing) dynamics, so that the results might be relevant in a more
57: general context.
58: \end{abstract}
59:
60: \submitto{\JPA}
61: \pacs{03.65.GE, 03.65.Sq, 05.45.-a}
62: % 03.65.GE : Solutions of wave equations: bound states
63: % 03.65.Sq Semiclassical theories and applications
64: % 05.45.-a Nonlinear dynamics and nonlinear dynamical systems
65:
66: \maketitle
67:
68:
69: \section{\label{I} Introduction}
70:
71: Polygon billiards have been studied both classically and quantum mechanically
72: for roughly twenty years now \cite{RicBer81}. These systems are situated right
73: on the borderline between integrability and chaos. They are usually
74: divided into two classes: the rational polygon billiards where all vertex
75: angles are rational multiples of $\pi$, and the irrational ones where at least
76: one vertex angle is an irrational multiple of $\pi$.
77:
78: In the first case, there exist two constants of motion, so that one
79: would expect integrability. However, due to singularities in the billiard flow,
80: the invariant surface of the flow is not necessarily a torus (with genus
81: $g=1$), but may be of a more complicated topology ($1\le g<\infty$). This
82: produces a very complicated classical dynamics (see:
83: \cite{Gut86,Gut96,KenSmi00} and references therein). The systems are called
84: integrable if $g=1$ and pseudo integrable \cite{RicBer81} otherwise.
85:
86: In the second case (the irrational polygon billiards), there is no second
87: constant of motion. These systems are typically ergodic \cite{Gut86} and
88: probably weakly mixing \cite{ArtCas97,CasPro99}, though the Kolmogorov-Sinai
89: entropy \cite{LichLieb83} is always zero. \\
90:
91: Quantum and semiclassical calculations have been performed from the very
92: beginning \cite{RicBer81,Gau87,Shu93,Mil94}, but only recently
93: \cite{Bog99,CasPro99b} it became possible to calculate sufficiently large
94: level sequences at sufficiently high energies, such that correlation properties
95: could be analysed directly. There are fundamental open questions:
96: \begin{itemize}
97: \item[(i)]{Do the correlations in the spectra of polygon billiards eventually
98: become stationary at sufficiently high energy?}
99: \item[(ii)]{Are there families of polygon billiards with common statistical
100: properties (universality)?}
101: \item[(iii)]{What is the signature of classical pseudo integrability in the
102: quantum spectrum (quantum-classical correspondence)?}
103: \end{itemize}
104: On the one hand, there has been numerical evidence \cite{CasPro99b}, that
105: at very high energies the spectra of irrational triangle billiards are
106: statistically similar to spectra taken from the Gaussian Orthogonal Ensemble
107: (GOE). On the other hand, based on the numerical study of the spectra of
108: several rational right triangle billiards, it was proposed that pseudo
109: integrability implies so called ``intermediate statistics'' \cite{Bog99}. For
110: the nearest neighbour distribution \cite{Boh89} this means: linear increase at
111: small spacings (as in the GOE case) and exponential fall-off at large spacings
112: (as for a random Poissonian sequence). Intermediate statistics has also been
113: found in the context of disordered systems at the metal-insulator transition
114: point \cite{Shk93,BraMon98,VarBra00}, which might indicate some relationship
115: between both classes of systems. \\
116:
117: This paper is mainly concerned with question (iii). We consider the
118: one-parameter family of right triangle billiards, labeled by the value of the
119: smallest vertex angle $0< \alpha \le \pi/4$. For this class, a secular equation
120: is derived, which identifies the eigenvalues as zeros of the determinant of a
121: particular matrix $K(E)$. Though the matrix is infinite, its elements are given
122: explicitly by very simple expressions. This makes $K(E)$ an ideal point of
123: departure for numerical and analytical studies.
124:
125: The most obvious characteristic of rational polygon billiards is the genus $g$
126: of the invariant surface of the classical Hamiltonian flow (the irrational polygon
127: billiards can be included, setting $g=\infty$). Hence we will investigate in
128: detail the relation between $g$ and the correlations in the quantum spectra.
129: In the numerical part, level sequences are calculated at absolute level numbers
130: $>10^5$ for various examples of right triangle billiards. This provides valuable
131: complementary information to recent results from Bogomolny et al. \cite{Bog99}.
132: In the analytical part, the matrix $K(E)$ itself is considered. Though $K(E)$
133: is a pure quantum mechanical object, it is shown that $g$ and $\gamma$ (which
134: is closely related to $g$) play a crucial r\^ ole for iterated mappings of the
135: form $\Psi(n)= K^n(E) \Psi(0)$. Based on this observation, a mechanism is
136: proposed, which can explain the connection between the genus $g$ and the
137: correlation properties of the quantum spectrum. \\
138:
139: In section~\ref{SE} a secular equation is derived for the calculation of the
140: eigenvalues of right triangle billiards. It is used in section~\ref{LS} to
141: obtain and analyse the level spacing distributions for several right triangles.
142: In section~\ref{CM} we analyse the properties of the matrix $K(E)$ itself, and
143: we discuss the r\^ oles of the two classical parameters $g$ and $\gamma$ in
144: this context. The conclusions are presented in section~\ref{C}.
145:
146:
147:
148: \section{\label{SE} Secular equation}
149:
150: Our point of departure is the observation, that any right triangle can be
151: obtained from cutting an appropriate rectangle along its diagonal. This is
152: used to derive a secular equation of drastically reduced dimension for the
153: eigenvalues of the right triangle billiard.
154:
155: Let $H_0$ be the Hamiltonian for the rectangle billiard with sides $a$ and $b$.
156: Fixing the length scale by: $a^2+b^2= \pi^2$, the angle
157: $\alpha : \tan\alpha = b/a$ suffices to characterize the system completely.
158: Choosing an arbitrary corner of the rectangle billiard as the origin of a
159: Cartesian coordinate system, its eigenvalues and the corresponding eigenfunctions
160: may be written as follows:
161: \begin{eqnarray}
162: \eps(n,m) &= \frac{1}{2}\left(\frac{n^2}{\cos^2\alpha} +
163: \frac{m^2}{\sin^2\alpha}\right) \; , \quad n,m \ge 1 \label{SE_epsrec}\\
164: \Phi_{nm}(x,y) &= \frac{2}{\sqrt{ab}} \sin\!\!\left(\frac{\pi}{a} nx\right)
165: \sin\!\!\left(\frac{\pi}{b} my\right) \; .
166: \label{SE_Phirec}\end{eqnarray}
167: Consider the total Hamiltonian $H$:
168: \begin{equation}
169: H = H_0 + \eta \; W \; ,\quad
170: W = \delta\!\!\left( \frac{x}{a}-\frac{y}{b}\right) \; ,
171: \label{SE_Htri}\end{equation}
172: where the potential $\eta W$ is used to cut the rectangle billiard into two
173: congruent right triangle billiards (a similar cut potential, though in a
174: different context, has been used in \cite{Lew90}). As $\eta$ increases from $0$
175: to $\infty$, the spectrum of $H$ changes from the spectrum of the rectangle
176: billiard (\ref{SE_epsrec}) to the doubly degenerated spectrum of the two
177: triangle billiards. For any $\eta$, the Hamiltonian $H$ is invariant under
178: point reflection, so that the matrix representation of $H$ in the eigenbasis
179: of $H_0$ is block diagonal. One block is spanned by the odd basis states
180: $\{ \Phi_{nm} |\; n+m : {\rm odd} \}$ and the other by the even ones
181: $\{ \Phi_{nm} |\; n+m : {\rm even} \}$. Both blocks can be diagonalised
182: independently, leading to the same sequence of eigenvalues, which causes the
183: degeneracy mentioned above. \\
184:
185: In what follows we will work in the odd basis only. Let $q= n+m$ and
186: $p= n-m$, and order the states (\ref{SE_Phirec}) with increasing $q$, and for
187: equal $q$, with increasing $p$. Consider the subset of states with fixed $q$
188: and $p= -q+2,\ldots,q-2$ as one block. Then truncating the basis at a maximal
189: $q$-value $q_\mx$, one obtains $M= (q_\mx -1)/2$ blocks with $q-1$ states in each
190: block (note that $q$ and $p$ are odd). In total this gives $N= (q_\mx^2 -1)/4$
191: basis states. In this reordered basis, the matrix elements of $W$ are given by:
192: \begin{equation}
193: \eqalign{
194: W_{qp;q'p'} &= \int_0^a\rmd x\int_0^b\rmd y \; \Phi_{nm}(x,y) \Phi_{n'm'}(x,y)
195: \; \delta\left(\frac{x}{a}-\frac{y}{b}\right) \\
196: &= \; \frac{1}{2} \{ \delta(|p|-|p'|) + \delta(q-|p'|)
197: + \delta(q'-|p|) + \delta(q-q')\} \; ,
198: }
199: \end{equation}
200: where $n= (q+p)/2,\; m=(q-p)/2$ (and similarly for the primed indices). For
201: given $q_\mx$ the truncated matrix $W^{(N)}$ has only two distinct eigenvalues:
202: $0$ and $M+1$, and the eigenspace of the latter has dimension $M$ (in other
203: words: rank$[W^{(N)}] = M$). All eigenvectors with eigenvalue $M+1$ can be
204: calculated explicitly, and after proper normalization we collect them (as
205: column vectors) in the rectangular matrix $V$:
206: \begin{equation}
207: V_{k;qp} = \frac{1}{\sqrt{M+1}}
208: \left\{ \begin{array}{ll}
209: 0 \quad &: n_q < k \\
210: \sqrt{\frac{k+1}{k}} \quad &: n_q = k \\
211: \frac{1}{k(k+1)} \quad &: n_q > k \; ,\; n_p < k+1 \\
212: -\sqrt{\frac{k}{k+1}} \quad &: n_q > k \; ,\; n_p = k+1 \\
213: 0\quad &: n_q > k \; ,\; n_p > k+1 \end{array}\right. \; ,
214: \label{AVres}\end{equation}
215: where $k=1,\ldots,M$, $n_q= (q-1)/2$ and $n_p= (|p|+1)/2$. The truncated total
216: Hamiltonian may now be written as follows:
217: \begin{equation}
218: H^{(N)}= H_0^{(N)} + \eta (M+1) \; VV^T \; .
219: \label{SE_Htru}\end{equation}
220: Dividing the Schr\"odinger equation $(E-H^{(N)})\Psi =0$ by $E-H_0^{(N)}$, one
221: arrives after a few algebraic manipulations at the desired secular equation. It
222: determines the eigenvalues of $H^{(N)}$ as the zeros of the following determinant:
223: \begin{equation}
224: 0 = \det\left( 1 + \eta \tilde K^{(M)}(E) \right) \; , \qquad
225: \tilde K^{(M)}(E) = (M+1) \; V^T \frac{1}{E-H_0^{(N)}} V \; .
226: \label{SE_Sec}\end{equation}
227: Taking the limit $\eta\to\infty$, the unit matrix in the first equation of
228: (\ref{SE_Sec}) can be neglected, and one gets:
229: \begin{equation}
230: \det \tilde K^{(M)}(E) = 0 \; .
231: \label{SE_Sec2}\end{equation}
232: The advantage of this equation, is the reduced dimension $M\ll N= M(M+1)$.
233: Such a reduction is typical for a boundary integral method (see for example
234: \cite{VerSar95}). The matrix elements of $\tilde K^{(M)}$ are given by the
235: following expression:
236: \begin{equation}
237: \tilde K_{ij}^{(M)} = (M+1) \sum_{q=1}^{q_\mx} \sum_{p=-q+2}^{q-2}
238: \frac{V_{i,qp}\; V_{j,qp}}{E- \eps\left(\frac{q+p}{2},\frac{q-p}{2}\right)}
239: \; , \quad q,p:{\rm odd} \; .
240: \label{SE_tilK}\end{equation}
241: Being only interested in the zero eigenvalues of $\tilde K^{(M)}(E)$, any
242: (symmetric) similarity transformation $K^{(M)} = L^T \tilde K^{(M)} L$ may be
243: applied. The following choice for $L$ simplifies the problem considerably:
244: \begin{eqnarray}
245: L &=& {\rm diag}(1,\ldots,1/M) \, \left( \begin{array}{cccc}
246: 1 & -1 \\
247: & \ddots & \ddots \\
248: & & \ddots & -1 \\
249: & & & 1 \end{array}\right) \,
250: {\rm diag}\!\!\left(1,\ldots,\sqrt{M(M+1)}\right) \, . \nonumber\\
251: \quad &\quad &\quad
252: \end{eqnarray}
253: The resulting matrix $K^{(M)}(E)$ is defined in the same way as $K(E)$ in the
254: equations~(\ref{SE_Kfin1})-(\ref{SE_Kfin3}) below, but with the coefficients
255: $D_j= \sum_{i=0}^M d_{ij}$. Only in the limit $M\to \infty$, the expression for
256: $D_j$ simplifies to the formula~(\ref{SE_Kfin4}), as can be shown using the
257: partial fraction expansion of the {\it cot} function \cite{AbraSteg64}. \\
258:
259: To summarize, the right triangle spectrum is calculated using the secular
260: equation ${\rm det}[K(E)]= 0$, where $K(E)$ is constructed as follows:
261: \begin{equation}
262: K(E) = K_F(E) + K_D(E) \; .
263: \label{SE_Kfin1}\end{equation}
264: The matrix elements of $K_F$ are given by:
265: \begin{equation}
266: \eqalign{
267: [K_F]_{ij} = d_{ij} - d_{i,j+1} -d_{i-1,j} + d_{i-1,j+1} \; , \\
268: d_{ij} = d_{ij}^+ + d_{ij}^- \; , \qquad
269: d_{ij}^\pm = \frac{1}{e - q^2 - p^2 \pm 2qp \cos 2\alpha} \; ,
270: }
271: \label{SE_Kfin2}\end{equation}
272: where the scaled energy $e= E/(2\sin^2 2\alpha)$ is used, and $q= 2i+1$, and
273: $p= 2j-1$. Note, that $q^2+p^2\mp 2qp\, \cos(2\alpha) =
274: 2\sin^2(2\alpha)\;\eps[(q\pm p)/2,(q\mp p)/2]$, thus $q$ and $p$ may still be
275: regarded as auxiliary quantum numbers for the rectangle billiard $H_0$. The
276: matrix $K_D$ is tridiagonal:
277: \begin{equation}
278: [K_D]_{jj} = D_j + D_{j+1} \; , \qquad
279: [K_D]_{j,j+1} = [K_D]_{j+1,j} = - D_j \; ,
280: \label{SE_Kfin3}\end{equation}
281: with the coefficients $D_j$, given by
282: \begin{equation}
283: D_j = \frac{\pi \; \sin\pi \omega}{2\omega
284: \left( \cos\pi \omega + \cos\pi p \cos 2\alpha\right)}\; , \qquad
285: \omega = \sin 2\alpha \; \sqrt{2e- p^2} \; .
286: \label{SE_Kfin4}\end{equation}
287: Even though $\omega$ becomes imaginary for large values of $p$, the affected
288: functions: {\it sin} and {\it cos} convert into {\it sinh} and {\it cosh},
289: and finally the coefficient $D_j$ remains real. Its asymptotic behaviour for
290: large $j$ is: $D_j \sim \pi/(2|\omega|)$. \\
291:
292: This result is the basis for the analysis in section~\ref{LS}~and
293: section~\ref{CM}. The infinite matrix $K(E)$, as defined in
294: (\ref{SE_Kfin1})-(\ref{SE_Kfin4}), completely determines the spectrum
295: of any right triangle billiard as the set of zeros of its determinant. For
296: numerical purposes $K(E)$ must be truncated (see below), but one may get
297: important information also from an analysis of the infinite matrix $K(E)$
298: itself.
299:
300: For numerical calculations (section~\ref{LS}), $K(E)$ is truncated, keeping
301: only those elements $K_{ij}(E)$ for which $i,j \le M$. For meaningful results,
302: $M$ must be at least so large, that $p_\mx^2 > 2e,\; p_\mx= 2M-1$ [see the
303: definition of $p$ above equation~(\ref{SE_Kfin3})]. Experience shows, that for
304: accurate results (error less than $1\%$ of the mean level spacing), one should
305: increase the size of the matrix further by approximately 10\%. The zeros of
306: $\det[K(E)]$ are identified, calculating the smallest eigenvalue in magnitude
307: as a function of $E$. Using a standard root bracketing algorithm
308: \cite{NumRec92} we find those points at which the smallest eigenvalue of
309: $K(E)$ passes the zero axis. The eigenvalues of $K(E)$ are strictly decreasing
310: functions of $E$, and this facilitates the root finding considerably. It allows
311: to take rather large steps (of the order of the mean level distance), without
312: running the risk to loose any roots.
313:
314:
315:
316: \section{\label{LS} Level spacing distributions}
317:
318: In the case of polygon billiards, the genus $g$ of the invariant surface of the
319: Hamiltonian flow is the most obvious parameter to characterize the classical
320: dynamics \cite{Gut86}. Hence one may expect an influence of $g$ on the level
321: statistics of the corresponding quantum system. In this section we investigate
322: numerically whether the level statistics show a systematic dependence on $g$.
323: For several rational and one irrational right triangle, we calculate sequences
324: of $10^4$ levels starting at the absolute level number $10^5$ (Weyl's
325: law is used to determine the corresponding energy). Note that even in this
326: energy region the level statistics are usually not stationary. This has been
327: demonstrated in \cite{CasPro99b} for several examples of rational and
328: irrational triangle billiards. This should be kept in mind in the discussion
329: of the numerical results.
330:
331: In the case of right triangle billiards, there is another relevant parameter
332: intimately related to $g$ (see~\ref{AG}). This is $\gamma$, the smallest
333: integer such that $2\alpha\,\gamma/\pi \in \mathbb{N}$ (in the irrational case,
334: we set $\gamma = \infty$). It is shown in \ref{AG}, that $\gamma$ is the
335: smallest number of rhombuses which must be glued together to form the
336: invariant surface of the billiard flow. Moreover we find, that
337: $g= {\rm int}(\gamma/2)$. Hence $\gamma$ implies a finer classification of
338: the right triangle billiards than $g$ does. \\
339:
340: \begin{table}
341: \caption{All rational right triangle billiards with $g\le 7$, referenced by
342: their smallest vertex angle $\alpha/\pi = p/q$, and ordered with respect to
343: $g$ and $\gamma$. The first two entries for $g= g_a= 1$ are the only integrable
344: cases. The shaded entries refer to those cases analysed in \cite{Bog99}, and
345: the gray-scale corresponds to the value of $g_a$ (introduced there) as
346: indicated in the last column. \\ }
347: \begin{indented}
348: \item[] \includegraphics[scale=0.55]{gen4}
349: \end{indented}
350: \label{LS_gen2}\end{table}
351:
352: Up to an irrelevant energy scale, all right triangles may be labeled and
353: identified through their smallest vertex angle $0< \alpha \le \pi/4$. For
354: rational right triangles one may also use the pair of relatively prime integers
355: $p/q= \alpha/\pi$. This is done in table~\ref{LS_gen2}, where all rational
356: right triangles with $g\le 7$ are arranged with increasing $\gamma$ in the
357: vertical direction, and with increasing $\alpha$ in the horizontal direction.
358: The parameter $g_a$ is taken from \cite{Bog99}, where it is introduced as
359: ``arithmetical genus''. The entries under-laid with a gray shade have been
360: analysed there. The gray-scale corresponds to different values of $g_a$, as
361: indicated in the last column. Further below, we will compare our results with
362: those of \cite{Bog99}. \\
363:
364: \begin{figure}
365: \begin{center}
366: \input{deli12b.pstex_t}
367: \end{center}
368: \caption{Difference $\Delta I\, (s)$ of the integrated level spacing
369: distribution to the semi-Poisson case. $\Delta I\, (s)$ is plotted for various
370: values of $\alpha$ as indicated in (a). The abbreviation ``irr'' refers to
371: $\alpha= \pi(3-\sqrt{5})/4$. Whereas (b) shows the raw data, (a) shows the
372: corresponding smoothed curves (details in the text). The GOE expectation
373: ($N\to\infty$ limit) is plotted in (a) and (b) as a dotted line.}
374: \label{LS_f1}\end{figure}
375:
376: We analyse the level statistics by means of the nearest neighbour spacing
377: distribution $P(s)$ (the spacings are normalized to unit mean). Recently the
378: question has been raised, whether the rational right triangles show
379: intermediate statistics ({\it i.e.} a linear increase at small $s$ and
380: exponential fall-off at large $s$). A simple analytical example is the so
381: called ``semi-Poisson'' distribution \cite{Bog99}:
382: \begin{equation}
383: P_\SP(s) = 4s \; \rme^{-2s} \; .
384: \label{I_semiP}\end{equation}
385: Here, $P_\SP(s)$ is simply used as a conveniant reference to compare with.
386: The following quantity is plotted in figure~\ref{LS_f1}:
387: \begin{equation}
388: \Delta I\, (s) = \int_0^s\rmd s'\; \left\{ P_\Num(s') - P_\SP(s')\right\} \; .
389: \label{LS_DIs}\end{equation}
390: The theoretical curves in the figures~\ref{LS_f1}(a) and (b) show the result
391: for an infinite GOE spectrum, where $P_\Num(s')$ is replaced by the
392: corresponding level spacing distribution (taken from \cite{Haake91}). While
393: figure~\ref{LS_f1}(b) shows the raw numerical data curves for various right
394: triangle billiards, figure~\ref{LS_f1}(a) shows the corresponding smoothed
395: curves, in order to allow the identification of all the cases shown. For the
396: smoothing, ``natural smoothing splines'' have been used, as provided in
397: \cite{gnuplot99}.
398:
399: Let us first focus on the cases: $p/q= 1/8,1/5,1/12,1/7,3/16$ (which
400: correspond to a successive increase of $\gamma$ from $4$ to $8$), and
401: $\alpha/\pi = (3-\sqrt{5})/4$ (where $\gamma=\infty$). The $\Delta I$-curves
402: for these cases are plotted in figure~\ref{LS_f1}(a) with a solid line and
403: dashed lines of different dash lengths. Together with the results for the GOE
404: and the Poisson ensemble (uncorrelated random sequence), they roughly span a
405: one-parameter family of curves $\Delta I_\sigma(s)$. The parameter $\sigma$ may
406: be called ``correlation strength'' and it may be calibrated, requiring that
407: $\sigma=0$ gives the Poisson result (its graph is plotted in
408: figure~\ref{CM_f3}), $\sigma=1$ the GOE result, and $\sigma=1/2$ the
409: semi-Poisson result. Note, that $\Delta I_\sigma(s)$ is introduced solely to
410: facilitate the discussion of our results, so that it is not necessary to be
411: more specific.
412: % \cite{Note}.
413: % It is not meant as a proposition of a universal one-parameter
414: % family of level statistics.
415:
416: The $\Delta I$-curve for the irrational right triangle billiard comes closest
417: to the GOE result. However, it remains almost in the middle between the
418: semi-Poisson case and the GOE case ($\sigma \gtrsim 3/4$). Then follow the
419: cases $p/q= 3/16, 1/7$, and $1/12$, for which $\sigma$ decreases in
420: approximately equal steps. The $\Delta I$-curve for $p/q=1/12$ is closest to
421: the semi-Poisson result ($\sigma\approx 1/2$). The last two curves with
422: $p/q= 1/5$, and $1/8$ tend slightly towards the Poisson result. They are so
423: close to each other, that we would assign the same correlation strength to
424: both of them ($\sigma\lesssim 1/2$). In all we may say, that the correlation
425: strength $\sigma$ increases with increasing $\gamma$.
426:
427: Finally we included two more cases: $p/q= 1/10$ and $3/14$. In
428: figure~\ref{LS_f1}(a) the respective $\Delta I$-curves are plotted with dotted
429: lines. Thus we can compare the $\Delta I$-curves for the $1/5$- and
430: the $1/10$-triangle ($\gamma=5$), and the $\Delta I$-curves for the $1/7$-
431: and the $3/14$-triangle ($\gamma=7$). Both cases show, that even for right
432: triangles with the same value for $\gamma$, the respective $\Delta I$-curves
433: may differ considerably. The relation between $\gamma$ and the correlation
434: strength is apparently not very strict (at least not in the energy range
435: considered).
436:
437: In order to check, that our conclusions do not depend on the particular choice
438: of the correlation measure, we repeated the analysis above, using the number
439: variance $\Sigma^2(l)$ \cite{Boh89} instead of $\Delta I(s)$. The results
440: were perfectly compatible, so that a more detailed discussion is omitted. \\
441:
442: In the numerical analysis presented here, we concentrate on right triangle
443: billiards with small values for $g$ and $\gamma$. The main reason is, that
444: there is certainly an energy scale below which the quantum system cannot
445: possibly ``recognize'', whether the two hypotenuse angles $\alpha$ and $\beta$
446: are rational or not. Without any knowledge about this scale, small $g$
447: triangles are probably the more reliable examples, for the study of quantum
448: signatures of pseudo integrability. Hence the triangles for which we can
449: compare our results with those of \cite{Bog99} are only a few:
450: $p/q= 1/8,1/5,1/12$, and $1/7$.
451:
452: The results obtained in \cite{Bog99} agree with those presented here, only up
453: to a certain qualitative level. Beyond, we find that the correlation strength
454: has decreased considerably in almost all cases. This may be due to the higher
455: energy region considered here, which results in the rationality of the
456: hypotenuse angles being more important. However, the $\Delta I$-curves of the
457: first group of right triangles (with $4\le\gamma\le 6$) changed much less then
458: the others, such that the separation between both groups has decreased. It
459: seems that this separation was decisive for the introduction of the
460: arithmetical genus $g_a$. The fact that this separation has become much
461: smaller now, indicates that $g_a$ is probably not an appropriate alternative
462: for $g$. \\
463:
464: According to the numerical results presented in this section, it is possible
465: to order the right triangle billiards with respect to the strength of the
466: correlations found in their spectra, which coincides with that of increasing
467: $\gamma$. This finding confirms the general conjecture, that the genus of the
468: invariant surface of the classical billiard flow determines the strength of
469: the spectral correlations on the quantum level. Though the spectral
470: correlations are apparently not stationary at currently accessible energies,
471: the ordering seems to be energy independent, as long as the level sequences to
472: be compared, start with the same absolute level number.
473:
474:
475:
476: \section{\label{CM} The elliptic map}
477:
478: In the first part of this section it is shown, that the parameters $g$ and
479: $\gamma$ (see \ref{AG}) associated with the classical dynamics of right
480: triangle billiards, are important characteristics of the matrix $K(E)$ itself.
481: On the one hand this may be surprising, because $K(E)$ arose from a pure
482: quantum mechanical approach (see section~\ref{SE}), but on the other hand it
483: is a strong indication for the importance of classical pseudo integrability on
484: the quantum level. In the second part, we present a tentative explanation for
485: the dependence of the spectral statistics on $g$ and $\gamma$. \\
486:
487: \begin{figure}
488: \begin{center}
489: \includegraphics[scale=0.75]{f12d}
490: \end{center}
491: \caption{Portrait of the matrix $K(E)$ for $\alpha = \pi/5$, and
492: $E= 1.26\times 10^4$. The gray-scale corresponds to the absolute value of the
493: matrix elements.}
494: \label{CM_f1}\end{figure}
495:
496: In figure~\ref{CM_f1} the matrix $K(E)$ is portrayed for a typical case. The
497: gray-scale corresponds to the absolute value of the matrix elements. It can be
498: seen, that most of the matrix elements have vanishingly small absolute values.
499: Large absolute values can be found only along the diagonal and the first
500: off-diagonals, which are due to $K_D$, and on a ``moon''-like structure due to
501: $K_F$ [see equations~(\ref{SE_Kfin2})-(\ref{SE_Kfin4})]. The matrix elements
502: $[K_F]_{ij}$ become large, whenever the pair of integers $(i,j)$ is close to
503: the zero-line of one of the two functions
504: \begin{equation}
505: f_\pm(x,y)= e - 4\left( x^2 + y^2 \pm 2xy \cos 2\alpha \right) \; ,
506: \end{equation}
507: where $x,y$ are real, and positive, and $e$ is the scaled energy as used in
508: equation~(\ref{SE_Kfin2}).
509:
510: \begin{figure}[h]
511: \begin{center}
512: \input{kmatmap.pstex_t}
513: \end{center}
514: \caption{Schematic representation of the mapping $\vec y_1 = K(E) \vec y_0$
515: (details see text).}
516: \label{CM_fmap}\end{figure}
517:
518: The action of $K(E)$ on a localized state may be described schematically by a
519: double valued, symmetric map as shown in figure~\ref{CM_fmap}. The square in
520: the middle represents the matrix $K(E)$ (cf. figure~\ref{CM_f1}). An initial
521: state $\vec y_0$ localized at a given value $a_0$ is mapped to
522: $\vec y_1 = K(E) \vec y_0$ localized at $\{ a_0, a_{-1}, a_1\}$, where $a_{-1}$
523: and $a_1$ are the two solutions of the equation $f_\pm(a_0,x)=0$, for $x>0$.
524: Hence, the map $M$ associated with $K(E)$ may be defined as follows:
525: $\{ a_{-1}, a_1\} = M a_0$. Let us call it the ``elliptic map''. Due to
526: $f(x,y)=f(y,x)$, $a_0 \in M a_{-1}$ and $a_0 \in M a_1$. Consequently, an orbit
527: of such a map may be viewed as a doubly connected chain:
528: \begin{equation}
529: \cdots\!\!\!\parbox{7.4cm}{\input{mapflow.pstex_t}}\; \cdots
530: \label{CM_mapflow}\end{equation}
531: According to that picture, the $n$-fold image $\vec y_n = K(E)^n \; \vec y_0$
532: has localization peaks at the positions $\{a_{-n},\ldots,a_0,\ldots,a_n\}$.
533: Surprisingly, $M$ is isomorphic to the following, extremely simple map:
534: \begin{equation}
535: \varphi_{n\pm 1} = \varphi_n \pm 2\alpha \; ,
536: \label{CM_elli2}\end{equation}
537: where the result should be taken modulo $\pi$, such that it remains in the
538: interval $\left[ \right. -\pi/2,\pi/2\left. \right)$. This can be seen, using
539: the following parametrisation of the curve
540: $f_\pm(x,y) = 0$:
541: \begin{equation}
542: {x\choose y} = \sqrt{E}
543: {\sgn\left(\frac{\pi}{2}-2\alpha + \varphi\right) \; \cos (\varphi-2\alpha)
544: \choose \cos (\varphi)} \; , \quad
545: \varphi\in \left[\right. -\pi/2,\pi/2\left. \right) \; .
546: \label{CM_elli}\end{equation}
547: Replacing $y$ by an arbitrary point $a_n$ of the map $M$, one gets the
548: corresponding pair conjugated angles: $\cos(\pm\varphi_n)=a_n$. Replacing $x$
549: by $a_n$ one finds that $\pm\varphi_n$ must be mapped to
550: $\pm\varphi_n -2\alpha$ (mod $\pi$). It follows, that
551: \begin{eqnarray}
552: a_n &=& \cos(\pm\varphi_n) \nonumber\\
553: M a_n &=& \{a_{n-1},a_{n+1}\} =
554: \{\cos[\pm(\varphi_n-2\alpha)],\cos[\pm(\varphi_n+2\alpha)]\} \; .
555: \end{eqnarray}
556: From equation~(\ref{CM_elli2}) it follows that any orbit is restricted to a set
557: of $\gamma$ points, where $\gamma$ is the smallest integer such that
558: $2\alpha\,\gamma/\pi \in\mathbb{N}$. It is the same $\gamma$, which is
559: introduced in \ref{AG} as the number of rhombuses forming the invariant
560: surface of the billiard flow. Furthermore, the periodicity of the map $M$ is
561: ${\rm int}(\gamma/2)$ which is just the genus of that invariant surface. \\
562:
563: Here in the second part of this section we discuss a mechanism which can
564: explain the correspondence between the correlation properties of the quantum
565: spectrum and the classical parameter $\gamma$. The line of reasoning is as
566: follows:
567: \begin{itemize}
568: \item[1]{The correlation properties of the triangle spectrum at a given
569: energy $E$ are closely related to the correlation properties of the eigenvalues
570: of $K(E)$ in the vicinity of zero.}
571: \item[2]{At sufficiently high energy, $K_D$ (\ref{SE_Kfin1}) can be considered
572: as a random tridiagonal matrix with eigenstates which are typically localized.}
573: \item[3]{The matrix $K_F$ (\ref{SE_Kfin1}) has such a form, that
574: repeated multiplications of an initially localized state with $K(E)$, produce
575: an increasing number of copies at different positions. The positions are given
576: by the elliptic map $M$.}
577: \item[4]{If $\alpha$ is rational, all orbits of the elliptic map are
578: periodic with period $g$ and restricted to $\gamma$ points. This leads to an
579: approximate foliation of the Hilbert space into weakly coupled subspaces.
580: For any irrational $\alpha$, the elliptic map is ergodic, and all basis states
581: of the matrix $K(E)$ are strongly coupled.}
582: \end{itemize}
583: Point 3 has been treated in the first part of this section. The remaining
584: statements are discussed below.
585:
586: \begin{figure}
587: \begin{center}
588: \input{CM.pstex_t}
589: \end{center}
590: \caption{Difference of the integrated level spacing distribution to the
591: semi-Poisson case. In (a) $\Delta I\, (s)$ is plotted for the rational right
592: triangle billiard with $\alpha/\pi= 1/8$, in (b) for the irrational one with
593: $\alpha/\pi = (3-\sqrt{5})/4$. The solid lines show the result for
594: neighbored eigenvalues of the matrix $K(E)$, the dotted lines show the result
595: for the triangle spectrum. The dashed lines show the theoretical curves
596: for the Poisson case (long dashed lines) and for the GOE case (short dashed
597: lines).}
598: \label{CM_f3}\end{figure}
599:
600:
601: \subsection*{Correlation properties of the eigenvalues of the matrix $K(E)$}
602:
603: According to the secular equation derived in section~\ref{SE}, the triangle
604: eigenvalues are given by those energies, at which at least one eigenvalue of
605: $K(E)$ becomes zero. Therefore it seems plausible, that the correlation
606: properties of the eigenvalues of $K(E)$ close to zero and the triangle
607: eigenvalues are related. To verify this we calculate the spacing distribution
608: for those two neighbouring eigenvalues which have opposite signs (without
609: unfolding). The distribution is obtained from $10^4$ spacings, taken at
610: equi-distant energies, with the step size adjusted to the mean level spacing of
611: the corresponding triangle spectrum.
612:
613: In figure~\ref{CM_f3} we compare the results. In the rational case (a) as
614: well as in the irrational case (b), the $\Delta I(s)$-curves for the
615: eigenvalue pairs of $K(E)$ and for the triangle spectra differ remarkably,
616: though in (b) the agreement is somewhat better. However, at least
617: qualitatively, the results are as expected: In the rational case,
618: figure~\ref{CM_f3}(a), the eigenvalue statistics for $K(E)$ show indeed very
619: weak correlations, much weaker even than the corresponding triangle spectrum.
620: This can be seen from the $\Delta I(s)$-curve which clearly tends towards the
621: Poisson result (note also the behaviour at large $s$). In the irrational case,
622: figure~\ref{CM_f3}(b), both curves show relatively strong correlations.
623:
624: \begin{figure}
625: \begin{center}
626: \input{f14fin.pstex_t}
627: \end{center}
628: \caption{The eigenstates of $K_D(E)$ with eigenvalues close to zero for a
629: typical case. On the left, the absolute value of the eigenvector coefficients,
630: plotted as a function of $i$, the index for the basis of $K(E)$. On the
631: right, the corresponding eigenvalues plotted in a bar graph.
632: $\alpha= \pi/5, E = 1.5\times 10^5$.}
633: \label{LS_f6}\end{figure}
634:
635:
636: \subsection*{Localization of the eigenstates of $K_D$}
637:
638: The matrix $K_D$ is constructed in a simple manner from the coefficients $D_j$
639: [see equations~(\ref{SE_Kfin3}) and (\ref{SE_Kfin4})]. Those oscillate as
640: functions of $j$, the index of the basis of $K(E)$, more and more rapidly
641: while $p^2$ approaches $2e$. From a statistical point of view, it then seems
642: permissible to replace the arguments of the functions: {\it sin} and {\it cos}
643: by appropriate random variables. Even though the statistical properties of
644: the matrix elements are very complicated, one may expect Anderson localization
645: \cite{And58}.
646:
647: In figure~\ref{LS_f6} we show for a typical case, a series of eigenstates of
648: $K_D(E)$ ordered by their respective eigenvalues. Only those states with
649: eigenvalues in a small interval around zero are shown. Many eigenstates are
650: apparently localized. However, others are not, and spread over a wide range of
651: basis states. Usually those fluctuate only weakly and slowly and their
652: eigenvalues decrease very slowly with energy (not shown). Their r\^ ole is
653: still unclear, and will be the subject of future studies.
654:
655:
656: \subsection*{Foliation of the Hilbert space}
657:
658: Acting repeatedly with $K(E)$ on an initially localized state $\vec y_0$, the
659: first $g$ images will spread and localize at points
660: $\{a_0,\ldots,a_{\gamma-1}\}$ (here $a_{\gamma-1}$ is identical to $a_{-1}$),
661: as discussed in the first part of this section. Then, due to the periodicity of
662: the elliptic map, subsequent images spread only slowly away from these points.
663: In the ideal case the spreading would stop due to Anderson localization,
664: giving rise to an invariant subspace. In the same way, an initial state
665: localized in a different part of the Hilbert space, would lead to another
666: invariant subspace, and so on -- until possibly the whole Hilbert space would
667: have been decomposed into invariant subspaces. In the real system, such a
668: foliation of the Hilbert space occurs only approximately, and the subspaces
669: become weakly coupled. Nevertheless one may expect, that
670: correlations are to some extent suppressed due to this mechanism.
671:
672: For irrational $\alpha$, where the elliptic map is ergodic, an initially
673: localized state will spread out [by repeated multiplication with $K(E)$]
674: into the whole Hilbert space. No foliation of the Hilbert space can occur, and
675: one should expect correlations of similar strength as in the GOE case. \\
676:
677:
678:
679: \section{\label{C}Conclusions}
680:
681: We derived a new kind of secular equation for the determination of the spectra
682: of right triangle billiards. It involves the diagonalisation of the matrix
683: $K(E)$ which has a particularly simple and transparent structure. Based on
684: this equation we calculated spectra at level numbers $>10^5$ for various
685: examples of right triangle billiards, which shows the efficiency of the new
686: method.
687:
688: We found a clear correspondence between
689: the genus $g$ (or the related parameter $\gamma$) of the invariant surface of
690: the classical billiard flow and the strength of the correlations in the quantum
691: spectrum. While for small $g$ the spectral statistics is close to semi-Poisson
692: (with a slight tendency towards Poisson), it approaches the GOE
693: statistics when $g$ is increased. Our numerical results together with similar
694: studies \cite{Bog99,CasPro99b} suggest that the spectral correlations are not
695: stationary at currently accessible energies, but that the ordering with
696: increasing correlation strength and its correspondence to $g$ is conserved.
697:
698: In the second part of the paper, we found that the classical parameters $g$
699: and $\gamma$ are characteristic quantities for the matrix $K(E)$ itself. For
700: rational right triangle billiards, where $g$ is finite, one gets an approximate
701: foliation of the Hilbert space into invariant subspaces. The size of the
702: subspaces scales with $\gamma$. Based on this observation, we discussed a
703: mechanism which can explain the influence of $g$ and $\gamma$ on the level
704: statistics of right triangle billiards. \\
705:
706: The definition of $\gamma$ in \ref{AG} can be generalized to arbitrary polygon
707: billiards as follows: $\gamma$ is the smallest number of identical polygons
708: which must be glued together to form the invariant surface of the billiard
709: flow. In this general case, $g$ and $\gamma$ must possibly be considered as
710: independent parameters. Which of them is then more relevant for the
711: correlations in the quantum spectrum? Considering the r\^ oles of $g$ and
712: $\gamma$ in the description of the matrix $K(E)$, it seems that this is
713: $\gamma$ (the size of the approximate invariant subspaces of $K(E)$
714: scales with $\gamma$).
715: %
716: % Intermediate statistics has also been observed in the case of disordered
717: % systems at the metal-insulator transition point
718: % \cite{Shk93,BraMon98,VarBra00}.
719: % Also the matrix $K(E)$, as considered in section~\ref{CM} has certain
720: % similarities with tight binding models from the theory of disordered systems.
721: % Hence one might be able to learn something about critical phenomena in
722: % disordered systems from the study of polygon billiards. This is an important
723: % direction to follow in future research, and the present approach might be a
724: % suitable tool for this.
725:
726:
727: \ack
728: The author thanks T.~Prosen for providing numerical triangle spectra,
729: G.~Casati and T.~Prosen for making available their work prior to publication,
730: and F.~Leyvraz and T.~Prosen for valuable discussions.
731:
732:
733:
734: \appendix
735:
736: \section{ \label{AG} The invariant surface for the classical billiard flow}
737:
738: There is an elegant way to represent a trajectory moving in a polygon
739: billiard, which is particularly useful to construct the invariant surface of
740: the classical billiard flow. It consists in drawing the trajectory as a
741: straight line, and reflecting the billiard (instead of the trajectory) each
742: time the boundary is hit \cite{Gut86}. In the case of rational polygons all
743: possible trajectories can produce only a finite number of differently oriented
744: copies of the original polygon. Then there is a general recipe of how to glue
745: these copies together, in order to obtain the invariant surface.
746:
747: \begin{figure}
748: \begin{center}
749: \includegraphics[scale=0.55]{rosettes}
750: \end{center}
751: \caption{(a) Typical right triangle. (b) Invariant surface of the billiard flow
752: for the $1/8$-triangle. The dotted rhombus does not form part of the invariant
753: surface. (c) Invariant surface of the billiard flow for the $1/7$-triangle.
754: In (b) and (c): the initial rhombuses, with which the rosette constructions
755: are started are filled. Vertices are labeled by capital letters, edges by
756: lower case letters. However, only such edges or vertices are labeled, which
757: must be identified with one another, to obtain the invariant surface.}
758: \label{AP_fr}\end{figure}
759:
760: For rational right triangles one may follow a more explicit construction
761: scheme, which leads to a particularly simple invariant surface; the
762: ``rosette''. It is constructed as follows: Start with a right triangle as
763: depicted in figure~\ref{AP_fr}(a). Reflecting the triangle on the side $AC$,
764: the image on the side $BC$ and that image again on the side $AC$, produces a
765: rhombus [see figure~\ref{AP_fr}(b) and (c)]. This rhombus is rotated around
766: point $A$ by the angle $2\alpha$ (counter-clockwise) at each step. Stop one
767: step before arriving at the original rhombus or its point reflected image [see
768: figure~\ref{AP_fr}(b)]. Note, that the resulting surface may wind several times
769: around $A$. As it is shown below, the resulting surface can be closed by
770: identifying open edges with one another, which gives the invariant surface.
771:
772: The number of rotations by $2\alpha$ is just the smallest integer $\gamma$
773: such that $2\alpha\, \gamma/\pi\in\mathbb{N}$. We could equally well rotate
774: the rhombus step-wise by $2\beta$ around point $B$ [see figure~\ref{AP_fr}(b)
775: and (c)], resulting in a different representation of the invariant surface.
776: However, as shown below, the number of rotations (or rhombuses) necessary to
777: close the invariant surface is the same.
778:
779: Proof: Let $p/q= \alpha/\pi$ and $p'/q'= \beta/\pi$ with $p,q$ and $p',q'$
780: relatively prime.
781: \begin{equation}
782: \eqalign{
783: 2\gamma\; \frac{p}{q} = l =
784: 2\gamma \left( \frac{1}{2} - \frac{p'}{q'}\right) \; &\Rightarrow\;
785: 2\gamma\; \frac{p'}{q'} = \gamma - l \; , \\
786: 2\gamma'\; \frac{p'}{q'} = l' =
787: 2\gamma' \left( \frac{1}{2} - \frac{p}{q}\right) \; &\Rightarrow\;
788: 2\gamma'\; \frac{p}{q} = \gamma' - l' \; .
789: }
790: \label{Aproof}\end{equation}
791: Consider the first line of (\ref{Aproof}). As $2 p < q$ it follows
792: $\gamma - l > 0$. But $\gamma'$ is the smallest integer such that
793: $2\gamma'\; p'/q' \in\mathbb{N}$, hence: $\gamma' \le \gamma$. The same
794: argument applied to the second line of (\ref{Aproof}) shows:
795: $\gamma \le \gamma'$. Therefore: $\gamma = \gamma'$. \\
796:
797: It remains, to prove that the procedure above gives indeed a representation
798: of the invariant surface, and to calculate its genus $g$. To this end we show
799: that all free edges of the rosette can be identified with one another. Then the rhombuses define a triangularisation of the invariant surface, and counting all
800: faces $F$, edges $E$ and vertices $V$ of the triangularisation, we obtain the
801: genus via the Euler characteristic \cite{Gut86}: $g= 1-\chi/2, \chi= V-E+F$.
802:
803: The edges may be divided into inner edges, which are connected to the center of
804: the rosette, and outer edges, which are not connected to the center. Let us
805: label both groups counter-clockwise by e$_1$,\ldots and e$_1'$,\ldots
806: respectively, beginning with the lower edges of the initial rhombus (see
807: figure~\ref{AP_fr}, but note that the labels shown there are different, and
808: used only to identify different edges in the representation of the invariant
809: surface).
810:
811: Let us first discuss the case, where $q$ is odd. Then $\gamma = q$ and the
812: rosette winds $p$ times around its center before the last inner edge can be
813: identified with the first one [see figure~\ref{AP_fr}(c)]. In order to
814: identify all outer edges e$_1$,\ldots,e$_{2\gamma}$ pairwise with one
815: another, observe that for $j$ :odd, a trajectory leaving the surface crossing
816: e$_{j+3}$ would enter a triangle which is the parallel translation of the
817: triangle with the hypotenuse e$_j$. By consequence, both edges can be
818: identified. Hence one may identify the following edges: e$_1 \equiv$ e$_4$,
819: e$_3 \equiv$ e$_6$, \ldots, e$_{2\gamma-3} \equiv$ e$_{2\gamma}$, and there are
820: only two open outer edges left: e$_2$ and e$_{2\gamma-1}$ which can be
821: identified with one another on the same grounds.
822:
823: The vertices in the representation of the invariant surface have to be
824: identified taking into account that edges identified previously have the same
825: initial and end points (the triangle connected to the edge defines an
826: orientation). In this manner it is shown that for $q$ :odd, the rosette
827: (invariant surface) has $\gamma$ faces, $2\gamma$ edges and $3$ vertices.
828: Hence $\chi= 3-\gamma$ and $g= (\gamma-1)/2$.
829:
830: If $q$ is even, then $p$ (relatively prime) is odd and the rosette winds $p/2$
831: times around its center [see figure~\ref{AP_fr}(b)] which means, that we have
832: also two open inner edges: e$_1'$ and e$_{\gamma+1}'$. Identifying outer
833: edges as explained above, leaves two outer edges open: e$_2$ and
834: e$_{2\gamma-1}$. In this case we identify e$_1'$ with e$_{2\gamma-1}$ and
835: e$_2$ with e$_{\gamma+1}'$, which again closes the invariant surface. Note that
836: due to the identification of outer edges with inner edges, the central point of
837: the rosette must be identified with the outermost points. The remaining points,
838: must be identified as one single vertex $B$, if $\gamma$ is even [see
839: figure~\ref{AP_fr}(b)], otherwise they constitute two vertices $B$ and $B'$
840: (this case is not shown). Hence the rosette (invariant surface) has $\gamma$
841: faces, $2\gamma$ edges, and $2$ vertices if $\gamma$ is even and $3$ vertices
842: if $\gamma$ is odd. This gives $\chi= 2-\gamma \Rightarrow g= \gamma/2$ and
843: $\chi=3-\gamma \Rightarrow g= (\gamma-1)/2$ respectively.
844:
845: In all: $g=(\gamma-1)/2$ if $\gamma$ :odd, and $g=\gamma/2$ if $\gamma$
846: :even. Hence $g= {\rm int}(\gamma/2)$.
847:
848:
849:
850: \section*{References}
851:
852: \bibliography{../../Bib/np,../../Bib/rm,../../Bib/sk,../../Bib/pp,../../Bib/lit,../../Bib/su}
853:
854:
855: \end{document}
856: