1: % last change 19.6.2006
2: \documentclass[aps,prl,twocolumn,groupedaddress,showpacs,amsmath,
3: amssymb,amsfonts]{revtex4}
4: \usepackage{graphicx}
5: \usepackage{bm}
6: \newcommand{\rhovec}{\boldsymbol{\rho}}
7: \newcommand{\alphav}{\boldsymbol{\alpha}}
8: \begin{document}
9: \title{Can one count the shape of a drum?}
10: \author{Sven Gnutzmann$^{3,1}$, Panos D. Karageorge$^2$
11: and Uzy Smilansky$^{1,2}$} \email[]{uzy.smilansky@weizmann.ac.il}
12: \affiliation{$^1$Department of Physics of Complex Systems, The
13: Weizmann Institute of Science, Rehovot 76100, Israel}
14: \affiliation{$^2$School of Mathematics, Bristol University,
15: Bristol BS81TW, England, UK.} \affiliation{$^3$Institut f\"ur
16: Theoretische Physik, Freie Universit\"at Berlin, Arnimallee 14,
17: 14195 Berlin, Germany}
18:
19: \date{\today}
20: \begin{abstract}
21: Sequences of nodal counts store information on the geometry
22: (metric) of the domain where the wave equation is considered. To
23: demonstrate this statement, we consider the eigenfunctions of the
24: Laplace-Beltrami operator on surfaces of revolution. Arranging the
25: wave functions by increasing values of the eigenvalues, and
26: counting the number of their nodal domains, we obtain the nodal
27: sequence whose properties we study. This sequence is expressed as
28: a trace formula, which consists of a smooth (Weyl-like) part which
29: depends on global geometrical parameters, and a fluctuating part
30: which involves the classical periodic orbits on the torus and
31: their actions (lengths). The geometrical content of the nodal
32: sequence is thus explicitly revealed.
33: \end{abstract}
34:
35: \pacs{02.30.Zz,03.65.Ge,03.65.Sq, 05.45.Mt}
36:
37: \maketitle
38:
39: Eigenfunctions of the Schr\"odinger and other wave equations can
40: be characterized by the number of their nodal domains - a nodal
41: domain being a maximally connected region where the eigenfunction
42: has a constant sign. The intimate connection between the spectra
43: of wave equations and the corresponding sequences of nodal counts
44: is well known and frequently used in various branches of physics
45: and mathematics. Sturm's oscillation theorem states that in one
46: dimension the $n$-th eigenfunction has exactly $n$ nodal domains.
47: In higher dimensions Courant proved that the number of nodal
48: domains $\nu_n$ of the $n$-th eigenfunction cannot exceed $n$
49: \cite{CH}. Recently, it was shown that the fluctuations in the
50: nodal sequence $\{\nu_n\}_{n=1}^{\infty}$ display universal
51: features which distinguish clearly between integrable (separable)
52: and chaotic systems \cite{BGS}. Their study also leads to
53: surprising connections with percolation theory \cite{Bogoschmit}.
54: Moreover, the nodal sequences of several \emph{isospectral} (yet
55: not isometric) domains were recently shown to differ in a
56: substantial way \cite{GnutzSS, BandSS}. The later observations
57: suggest that the nodal sequence stores information about the
58: domain geometry, and this information is not equivalent to the one
59: stored in the spectrum. Here, we provide further evidence by
60: deriving an asymptotic trace formula for the \emph{nodal counting
61: function}
62:
63: \begin{equation}
64: C(K)= \sum_{n=1}^{[\![ K ]\!]} \nu_n \ , \ K>0 \ ;\ \ [\![ \cdot
65: ]\!] = \text{the integer part.}
66: \label{eq:nodalcounting_def}
67: \end{equation}
68: The trace formula (see
69: \eqref{eq:torus_cKfluct} and \eqref{eq:nodaltrace} ) shows the explicit dependence of the
70: nodal sequence on the geometry of the surface in both the smooth
71: (Weyl-like) and the fluctuating parts. Thus, the nodal trace
72: formula is similar in structure to the corresponding spectral
73: trace formula \cite{BerryTabor,cdv, bleher94}. Kac's famous
74: question ``Can one hear the shape of a drum?'' was triggered by
75: the study of the progenitors of the \emph{spectral} trace formulae
76: \cite{kac}. The trace formula for the nodal counts leads us to the
77: title of this letter in which ``count'' replaces ``hear''. We will
78: consider here two particular classes of systems, namely, the wave
79: equation on convex smooth surfaces of revolution and on simple
80: two-dimensional tori. Generalization to other Riemannian manifolds
81: in two or more dimensions are possible, provided the wave equation
82: is separable.
83:
84:
85: The nodal counting function \eqref{eq:nodalcounting_def} is well
86: defined if the spectrum is free of degeneracies, $E_n > E_{n-1}$.
87: In case of degeneracies we represent the wave functions in the
88: unique (real) basis in which the wave functions appear in product
89: form. This, however, does not suffice to set a unique order
90: within the degenerate states, which consequently introduces
91: ambiguities in the nodal sequence. To circumvent this problem, we
92: modify the definition of the nodal counting function: First define
93: $\tilde{c}(E)=\sum_{n=1}^\infty \nu_n \Theta(E - E_n)$. This
94: function is based on information obtained from the nodal sequence
95: and the spectrum. To eliminate the dependence on the latter, we
96: use the ($\epsilon$-smoothed) spectral counting function
97: $\mathcal{N}_{\epsilon}(E)= \sum_{n=1}^\infty
98: (\frac{1}{2}+\frac{1}{\pi} \arctan \frac{E - E_n}{\epsilon})$,
99: which for finite $\epsilon$ is monotonic and can be inverted. We
100: define $E_{\epsilon}(K)$ as the solution of
101: $\mathcal{N}_{\epsilon}(E)=K$ and the modified nodal counting
102: function is
103: \begin{equation}
104: c(K)=\lim_{\epsilon\rightarrow 0} \tilde{c}( E_{\epsilon} (K) )\ .
105: \label{eq:modified_def}
106: \end{equation}
107: If there are no degeneracies, $c(K)$ is equivalent to
108: \eqref{eq:nodalcounting_def} up to a shift $K \rightarrow
109: K-\frac{1}{2}$. A $g$-times degenerate eigenvalue $E_{n}=E_{n+1}
110: =\dots=E_{n+g-1}$ contributes a single step function
111: $\Theta\big(K-(n-1+\frac{g}{2})\big)\sum_{s=1}^g \nu_{n+s-1}$
112: where the nodal counting function increases by the sum of the
113: nodal counts within the degeneracy class. We will derive a trace
114: formula for this modified nodal counting function below (and omit
115: the `modified' in the sequel).
116:
117:
118:
119: We start with the simpler case of a 2-dim torus represented as a
120: rectangle with side lengths $a$ and $b$ and periodic boundary
121: conditions. The eigenvalues take the values
122: $E_{n,m}=(2\pi)^2\left[\frac{n^2}{a^2}+\frac{m^2}{b^2}\right]$
123: where $m,n\in \mathbb{Z}$. The corresponding wavefunctions for
124: $m,n \ge 0$ are $\psi_{n,m}(x,y)=\cos(2\pi n x/a)\cos(2\pi m y/b)$
125: and the cosine is replaced by a sine for negative $m$ or $n$. The
126: number of nodal domains in the wavefunction $\psi_{n,m}$ is
127: \begin{equation}
128: \nu_{m,n}= (2 |n| +\delta_{n,0})(2|m|+\delta_{m,0})\ .
129: \end{equation}
130: The aspect ratio $\tau=a/b$ is the only free parameter in this
131: context because the number of nodal domains is invariant to
132: re-scaling of the lengths.
133:
134: Using Poisson's summation formula $\sum_{n=n_1}^{n_2} f(n)=
135: \sum_{N=-\infty}^\infty \int_{n_1-1/2}^{n_2+1/2} {\rm d}
136: n\ f(n) e^{2\pi
137: i Nn}$ the leading asymptotic trace formula for the spectral
138: counting function $\mathcal{N}(E)=\sum_{m,n} \Theta(E-E_{n,m})$
139: for a torus, is,
140: \begin{eqnarray}
141: \mathcal{N}(E)&=&
142: \mathcal{A} E \\
143: &+&\sqrt{\frac{8}{\pi}}\mathcal{A}
144: E^{\frac{1}{4}}\sum_{\textbf{r}} \frac{
145: \sin(L_{\textbf{r}}\sqrt{E}-\frac{\pi}{4})}{L_{\textbf{r}}^{\frac{3}{2}}}+
146: \mathcal{O}(E^{-\frac{3}{4}}) \nonumber
147: \label{eq:torus_speccount}
148: \end{eqnarray}
149: Here, $\mathcal{A}=ab/(4\pi)$, and the sum is over the winding
150: numbers $ \textbf{r}=(N,M)\in\mathbb{Z}^2\backslash(0,0)$ (in the sequel
151: every sum over $\textbf{r}$ will not include $(0,0)$ unless stated
152: otherwise). $L_{\textbf{r}}=\sqrt{(Na)^2+(Mb)^2}$ is the length
153: of a periodic geodesic (periodic orbit) with $ \textbf{r}=(N,M)$.
154:
155: Our goal is to derive a similar trace formula for the leading
156: asymptotic behavior of the nodal counting function. Again, using
157: Poisson re-summation and the saddle point approximation we get for
158: $\tilde{c}(E)$
159: \begin{eqnarray}
160: \tilde{c}(E)&=&
161: \frac{2 \mathcal{A}^2}{\pi^2} E^2 \\
162: &+&
163: E^{\frac{5}{4}} \frac{
164: 2^{\frac{11}{2}}\mathcal{A}^3}{\pi^{\frac{1}{2}}}
165: \sum_{\textbf{r}} \frac{|MN|}{L_{\textbf{r}}^{\frac{7}{2}}}
166: \sin(L_{\textbf{r}}\sqrt{E}-\frac{\pi}{4})+ \mathcal{O}(E)\ .
167: \nonumber \label{eq:torus_cE}
168: \end{eqnarray}
169: To express
170: the counting function as a function of the index $K$, we formally
171: invert the spectral counting function to order $\mathcal{O}(K^0)$
172: \begin{equation}
173: E(K)=\frac{K}{\mathcal{A}}
174: -K^{\frac{1}{4}}\frac{2^{\frac{3}{2}}}{\mathcal{A}
175: \pi^{\frac{1}{2}} }\sum_{\textbf{r}}
176: \frac{\sin(l_{\textbf{r}}\sqrt{K}-\frac{\pi}{4})}{l_{\textbf{r}}^{\frac{3}{2}}}
177: \label{eq:invert}
178: \end{equation}
179: where $l_{\textbf{r}}=L_{\textbf{r}}/\sqrt{\mathcal{A}}$ is the
180: re-scaled (dimensionless) length of a periodic orbit. This formal
181: inversion needs a proper justification which makes use of the
182: fact that we actually invert the smooth and monotonic ${\cal
183: N}_{\epsilon}(E)$. However, a detailed discussion of this point
184: goes beyond the scope of the present note. The numerical tests
185: which we provide here, supports the validity of this formal
186: manipulation. Replacing $E$ by $E(K)$ in \eqref{eq:torus_cE} and
187: keeping only the leading order terms, we get the nodal counting
188: function, which we write as a sum
189: $c(K)=\overline{c}(K)+c_{\mathrm{osc}}(K)$ of a smooth part
190: $\overline{c}$ and an oscillatory part $c_{\mathrm{osc}}$:
191: \begin{eqnarray}
192: \label{eq:torus_cKfluct}
193: \overline{c}(K)&=&\frac{2}{\pi^2} K^2 + \mathcal{O}(K)\ ,\nonumber
194: \\
195: c_{\mathrm{osc}}(K)&=&K^{\frac{5}{4}} \sum_{{\textbf{r}}}
196: a_{\textbf{r}}
197: \sin(l_{{\textbf{r}}}\sqrt{K}-\frac{\pi}{4})
198: +\mathcal{O}(K) \\
199: a_{\textbf{r}}&=&\frac{2^{\frac{7}{2}}}{\pi^{\frac{5}{2}}
200: l_{{\textbf{r}}}^{\frac{3}{2}}} \left(\frac{4\pi^2
201: |NM|}{l_{{\textbf{r}}}^2}-1\right)\nonumber
202: \end{eqnarray}
203: While the smooth part is independent of the geometry of the torus,
204: the oscillating part depends explicitly on the aspect ratio $\tau$
205: and can distinguish between different geometries. The main
206: difficulty in computing higher order corrections to the leading
207: behavior of the nodal counting function is that products of sums
208: over periodic orbits appear already in the terms of order $K$.
209:
210: Turning now to more general surfaces we consider analytic, convex
211: surfaces of revolution $\mathcal{M}$ which are created by the
212: rotation of the line $y=f(x), x\in I\equiv [-1,1]$ about the $x$
213: axis. To get a smooth surface, $f(x)$ in the vicinity of $x= \pm
214: 1$, should behave as $f^2(x) \approx a_{\pm}(1 \mp x)$, with
215: $a_{\pm}$ positive constants. Convexity is achieved by requiring
216: the second derivative of $f(x)$ to be strictly negative, and
217: $f'(x_{max})=0$, $x_{max}\in I$, where $f$ reaches the value
218: $f_{\mathrm{max}}$. We consider the wave equation $-\Delta
219: \psi(x,\theta)=E \psi(x,\theta)$ and the Laplace-Beltrami operator
220: for a surface of revolution is
221: \begin{equation}
222: \Delta = \frac{1}{f(x)\sigma(x)}\frac{ \partial}{ \partial
223: x} \frac{f(x)}{\sigma(x)}\frac{ \partial }{ \partial x}+
224: \frac{1}{f(x)^2}\frac{ \partial^2 }{ \partial \theta^2}\ .
225: \label{eq:LapBel}
226: \end{equation}
227: Here, $\sigma(x)= \sqrt {1+f'(x)^2}$ and $\theta$ is the azimuthal
228: angle. The domain of $\Delta$ are the doubly differentiable,
229: $2\pi$ periodic in $\theta$ and non singular functions on $ [I
230: \times S^1]$. Under these conditions, $\Delta$ is self adjoint
231: and its spectrum is discrete. $\Delta $ is separable and the
232: general solution can be written as a product $\Psi(x,\theta) =
233: \exp(i m \theta)\ \phi_m(x)$ where $m\in \mathbb{Z}$. For any $m$,
234: \eqref{eq:LapBel} reduces to an ODE of the Sturm-Liouville type,
235: with eigenvalues $E_{m,n}$ (doubly degenerate when $m\ne0$) and
236: eigenfunctions $\phi_{n,m}(x)$ with $n=0,1,2,\dots$ nodes. The
237: eigenfunctions corresponding to the eigenvalue $E_{n,m}$ can be
238: written as linear combinations of $\cos(m\theta) \phi_{n,m}(x)$
239: and $\sin(m\theta) \phi_{n,m}(x)$. To be definite, we chose these
240: two functions as the basis for the discussion, and associate the
241: former with positive values of $m$ and the later with the negative
242: values of $m$. The nodal pattern is that of a checkerboard typical
243: to separable systems and contains
244: \begin{equation}
245: \nu_{n,m} = (n+1)(2|m| + \delta_{m,0})
246: \label{eq:nu}
247: \end{equation}
248: nodal domains. The semi-classical spectrum is constructed by
249: using the Bohr-Sommerfeld approximation \cite{cdv},
250: \begin{equation}
251: E_{n,m}^{\mathrm{scl}} = H(n+\frac{1}{2},m) + \mathcal{O}(1) \qquad ,
252: n \in
253: \mathbb{N},\ m\in \mathbb{Z}\ .
254: \label{eq:BSspectrum}
255: \end{equation}
256: where $H(n,m)$ is the classical Hamiltonian defined in terms of
257: the action variables $m$ and $n$, where $m$ is the momentum
258: conjugate to the angle $\theta$ and $n$ is
259: \begin{equation}
260: \begin{split}
261: n(E;m) =& \frac{1}{2\pi} \oint p_x(E,x) \ {\rm d}x =
262: \frac{1}{\pi}\int_{x_{-}}^{x_{+}}
263: p_x(E,x) {\rm d}x \\
264: p_x(E,x)=& \sqrt{(E f^2(x)-m^2)(1+f'(x)^2)}/f(x)\ .
265: \end{split}
266: \label{eq:naction}
267: \end{equation}
268: $x_{\pm}$ are the classical turning points $E f^2(x_{\pm})-m^2
269: =0$, with $x_{-}\le x_{max}\le x_{+}$. The hamiltonian is obtained
270: by expressing $E$ as a function of $n,m$ using the implicit
271: expression \eqref{eq:naction}. $H(n,m)$ is a homogenous function
272: of order 2: $H(\lambda n,\lambda m)=\lambda^2 H(n,m) $. It
273: suffices therefore to study the function $n(m)\equiv n(E=1,m)$
274: which defines a line $\Gamma$ in the $(n,m)$ plain. $n(m)$ is defined
275: for $|m| \le f_{\mathrm{max}} $ and is even in $m$, $n(m)=n(-m)$.
276: The function $n(m)$ is monotonically decreasing from its maximal
277: value $n(0)$ to $n(m=f_{\mathrm{max}})=0$. All relevant
278: information on the geodesics on the surface can be derived from
279: $n(m)$. Periodic geodesics appear if the angular velocities
280: $\omega_n=\frac{\partial H(n,m)}{\partial n}$ and
281: $\omega_m=\frac{\partial H(n,m)}{\partial m}$ are rationally
282: related. Since $\frac{dn(m)}{dm}=-\frac{\omega_m}{\omega_n}$ this
283: is equivalent to the condition
284: \begin{equation}
285: M+N \frac{dn(m)}{dm}=0
286: \label{eq:periodic_motion}
287: \end{equation}
288: for $M,N \neq 0$. The integers $\textbf{r} =(M,N)\in
289: \mathbb{Z}^2\backslash(0,0)$ are the winding numbers in the $\theta$ and
290: $x$ directions. The classical motion is considerably simplified if
291: the \emph{twist condition} \cite{bleher94} $n''(m) \equiv
292: \frac{d^2 n(m)}{dm^2}\neq 0$ for $0<m\le f_{\mathrm{max}}$ is
293: fulfilled. This excludes, for example, the sphere but includes all mild
294: deformations of an ellipsoid of revolution. We will assume the
295: twist condition for the rest of this letter. It guarantees that
296: there is a unique solution to \eqref{eq:periodic_motion} which we
297: will call $m_{\textbf{r} }$. Note, that $n'(m)$ has a finite range
298: $\Omega$ and a solution only exists if $-M/N \in \Omega$. The
299: cases $N=0$, $M\neq 0$ or $M\neq 0$, $N=0$ are not described by
300: solutions of \eqref{eq:periodic_motion}. They describe a pure
301: rotation in the $\theta$ direction at constant
302: $x=x_{\mathrm{max}}$ where $m_{0,\pm|M|}=\pm f_{\mathrm{max}}$
303: ($N=0$) or a a periodic motion through the two poles at fixed
304: angle $\theta\ \mathrm{mod}\ \pi$ ($M=0$) such that $m_{|N|,0}=0$.
305: The length of a periodic geodesic is given by
306: \begin{equation}
307: L_{\textbf{r} }=2 \pi \left|N n(m_{\textbf{r} }) + M
308: m_{\textbf{r} }\right|\ .
309: \end{equation}
310: Returning to the spectrum, we note that the leading terms in the
311: trace formula for the spectral counting function $N(E)=\sum_{m,n}
312: \Theta(E-E_{n,m})$ can be obtained by using (\ref{eq:BSspectrum})
313: and Poisson's summation formula \cite{bleher94}.
314: \begin{equation}
315: \mathcal{N}(E)= \mathcal{A} E + E^{\frac{1}{4}}
316: \sum_{\textbf{r}} \mathcal{N}_{\textbf{r}}(E)
317: \label{eq:bleher}
318: \end{equation}
319: where
320: \begin{equation}
321: \mathcal{A}
322: = \int_{-f_{\mathrm{max}}}^{f_{\mathrm{max}}} n(m) {\rm d}m
323: =||\mathcal{M}||/4\pi
324: \end{equation}
325: and $||\mathcal{M}||$ is the area of the surface. The oscillating
326: parts contain integrals $\propto
327: \int_{-f_{\mathrm{max}}}^{f_{\mathrm{max}}} {\rm d}m \ e^{2\pi
328: iE^{\frac{1}{2}}(N n(m) +M m)}$ which can be calculated to leading
329: order in $E^{\frac{1}{2}}$ using the stationary phase
330: approximation. The points of stationary phase are identified as
331: the classically periodic tori \eqref{eq:periodic_motion} with
332: $m=m_{\textbf{r}}$. This restricts the range of contributing
333: $\textbf{r}$ values to the classically accessible domain $-M/N\in
334: \Omega$. Thus,
335: \begin{equation}
336: \mathcal{N}_{\textbf{r}}(E)=(-1)^{N}
337: \frac{\sin(L_{\textbf{r}}E^{\frac{1}{2}}+
338: \sigma \frac{\pi}{4})}{2\pi |N^3
339: n''_{\textbf{r}}|^{\frac{1}{2}}}
340: +\mathcal{O}(E^{-\frac{1}{2}} )
341: \end{equation}
342: where $n''_{\textbf{r}}=n''(m=m_{\textbf{r}})$ and
343: $\sigma=\mathrm{sign}(n''_{\textbf{r}})$ which is the same for all
344: values of $\textbf{r}$. The contributions of the terms with either
345: $N=0$ or $M=0$ or with $-M/N \notin \Omega$ are of higher order
346: in $1/E$ and will not be considered here.
347:
348: We are now ready to derive the asymptotic trace formula for
349: \begin{eqnarray}
350: c(K)&=&\tilde{c}(E(K))=\sum_{n=0}^\infty\sum_{m=-\infty}^\infty
351: \nu_{mn}\ \theta\left (E(K)-E_{m,n}\right)\nonumber \\
352: E(K)&=&\frac{K}{\mathcal{A}}-\left(\frac{K}
353: {\mathcal{A}}\right)^{\frac{1}{4}} \sum_{\textbf{r}} \frac{
354: \mathcal{N}_{\textbf{r}}(\frac{K}{\mathcal{A}})}
355: {\mathcal{A}}+\mathcal{O}(K^0)\ .
356: \end{eqnarray}
357: The second equation was obtained by inverting $K=\mathcal{N}(E)$
358: to the desired order using the trace formula \eqref{eq:bleher}. We
359: follow the same approach as for $\mathcal{N}$ and expand the
360: result in $\delta E= E(K)-K/\mathcal{A}$ such that
361: $c(K)=\tilde{c}(K/\mathcal{A})+\tilde{c}'(K/\mathcal{A})\delta E
362: +\mathcal{O}(\tilde{c}'' \delta E^2)$ which is consistent if we
363: neglect all orders smaller than $\mathcal{O}(K)$. The result can
364: be expressed as a sum $c(K)=\overline{c}(K)+c_{\mathrm{osc}}(K)$
365: of a smooth part $\overline{c}$ and an oscillatory part,
366: $c_{\mathrm{osc}}$, in complete analogy to
367: (\ref{eq:torus_cKfluct}).
368: \begin{eqnarray}
369: \overline{c}(K)&=&2
370: \frac{\overline{mn}}{\mathcal{A}} K^2
371: + \frac{\overline{m}}{\mathcal{A}^{\frac{1}{2}}} K^{\frac{3}{2}}
372: +\mathcal{O}(K) \nonumber \\
373: c_{\mathrm{osc}}(K)&=&
374: K^{\frac{5}{4}}
375: %\left(\frac{K}{\mathcal{A}}\right)^{\frac{5}{4}}
376: \!\!\!\!\!\sum_{\textbf{r}:-\frac{M}{N}\in \Omega}\!\!\!\!\!\!
377: a_{\textbf{r}}
378: \sin(l_{\textbf{r}}K^{\frac{1}{2}}+\frac{\sigma\pi}{4}
379: +\mathcal{O}(K)\nonumber\\
380: a_{\textbf{r}}&=&(-1)^{N} \frac{m_{\textbf{r}} n(m_{\textbf{r}})
381: -2 \overline{mn}}{\mathcal{A}^{\frac{5}{4}}\pi
382: |N^3 n''_{\textbf{r}}|^{\frac{1}{2}}}
383: \nonumber \\
384: l_{\textbf{r}}&=&L_{\textbf{r}}/\sqrt{\mathcal{A}}\\
385: \overline{m^pn^q}&=&\frac{1}{\mathcal{A}}\int_{E(m,n)<1}
386: {\rm d}m {\rm d}n\ |m|^p n^q \nonumber \label{eq:nodaltrace}
387: \end{eqnarray}
388: where $l_{\textbf{r}}$ is the re-scaled length of a periodic
389: geodesic and $a_{\textbf{r}}$ is the amplitude contributed by the
390: (classically allowed) $\textbf{r}$ torus. For $m_r=0$ or $m_r=\pm
391: f_{\mathrm{max}}$ only one half of the stationary phase integral
392: contributes and the amplitude $a_{r}$ has to be multiplied by
393: $1/2$.
394:
395: \begin{figure*}
396: \begin{center}
397: \includegraphics[width=.95\linewidth,height=6.8cm]{fig.eps}
398: \caption{Numerical checks of the fluctuating parts of trace
399: formulae for the two ellipsoids (\ a) and b), see text) and the
400: two tori (\ c) and d), see text). \textbf{left}: Double
401: logarithmic plot of the integrated squared fluctuations (\ref{rk})
402: (arbitrary scale), the full line has slope $7/2$. \textbf{Right}:
403: Length spectra of the nodal counting functions (\ref{sl}). The
404: full line is obtained from the trace formulae
405: \eqref{eq:nodaltrace} and \eqref{eq:torus_cKfluct} and the points
406: represent the numerical data.}
407: \end{center}
408: \label{figure}
409: \end{figure*}
410:
411: The approximations involved in the above calculation have been
412: tested on an extensive numerical data base for two ellipsoids of
413: revolution defined by the equation $f(x)=R\sqrt{1-x^2}$ ($R=2$ in
414: data set a), and $R=1/2$ in data set b)) and for two different
415: tori ($\tau^2=2$ in data set c), and $\tau^2=\sqrt{2}$ in data set
416: d)). The spectral interval used for the numerical tests included
417: the first $10^5$ eigenvalues for the ellipsoids, and the first
418: $4\times10^6$ eigenvalues for the tori. The numerically computed
419: $c(K)$ were fitted to a fourth order polynomial in
420: $\kappa=\sqrt{K}$ and in all cases, the agreement of the two
421: leading coefficients with the asymptotic theory was better than a
422: percent. The oscillating part has been obtained numerically by
423: subtracting the best polynomial fit from the exact $c(K)$. The
424: fluctuating parts of the trace formulae were tested in two ways.
425: The integral of the squared oscillatory part
426: \begin{equation}
427: R(K)\approx \int_0^K {\rm d}K'\ c_{\mathrm{osc}}(K')^2\ .
428: \end{equation}
429: was computed as a function of $K$, and compared with the
430: theoretical expression which consists of a double sum over
431: periodic geodesics. Simplifying this expression by considering
432: only its diagonal part, one gets the estimate
433: \begin{equation}
434: R(K)=\frac{2}{7}K^{\frac{7}{2}}
435: \sum_{\textbf{r}} |a_{\textbf{r}}|^2
436: \label{rk}
437: \end{equation}
438: which scales like $K^{7/2}$. This scaling has been tested and the
439: results are shown in the left part of figure \ref{figure}.
440: Clearly, the expected power law is reached for sufficiently large
441: values of the counting index $K$. A more stringent test of the
442: trace formula is provided by computing the length spectrum,
443: defined by the properly scaled Fourier transform of $c_{osc}(K)$
444: with respect to $\kappa=\sqrt{K}$.
445: \begin{equation}
446: S(l)=l^{3/2}\! \int_0^\infty {\rm d} \kappa\ \kappa^{-5/2}
447: c_{\mathrm{osc}}(K=\kappa^2)
448: e^{-\!\frac{(\kappa-\kappa_0)^2}{\omega}+i \kappa l}
449: \label{sl}\ .
450: \end{equation}
451: Gaussian windows centered at $\kappa=\kappa_0$ and with a width
452: $\propto \sqrt{\omega}$ restricted the data used to be well within
453: the semiclassical domain. The trace formula for the nodal counts
454: predicts pronounced peaks at the scaled lengths $l=l_{\textbf{r}}$
455: of the periodic geodesics. The right frame in Figure \ref{figure}
456: shows a remarkable agreement of the numerical data with the
457: theoretical predictions. This excellent agreement provides further
458: support for the validity of the approximations which were used in
459: the derivation of the two versions of the nodal counts trace
460: formula.
461:
462: Recent studies \cite{GnutzSS,BandSS} have shown that the nodal
463: sequences of \emph{isospectral} domains are distinct, and can be
464: used to resolve isospectrality. Thus, the geometrical information
465: stored in the nodal sequence is not equivalent to the one stored
466: in the spectral sequence. This result together with the trace
467: formula obtained here motivates further research of the nodal
468: sequence as a tool in spectral analysis.
469:
470:
471:
472: \begin{acknowledgments}
473: This work was supported by the Minerva Center for non-linear
474: Physics and the Einstein (Minerva) Center at the Weizmann
475: Institute, and by grants from the GIF (grant I-808-228.14/2003),
476: and EPSRC (grant GR/T06872/01).
477: \end{acknowledgments}
478: \begin{thebibliography}{}
479: \bibitem {CH}
480: R.~Courant and D.~Hilbert,
481: \textit{Methods~of~Mathematical~Physics}, Vol I pp 451-465
482: (Interscience, New York, 1953).
483: \bibitem {BGS} G.~Blum, S.~Gnutzmann
484: and U.~Smilansky, Phys.~Rev.~Lett.\textbf{88}, 114101 (2002).%3
485: \bibitem {Bogoschmit}E.~Bogomolny and C.~Schmit,Phys.~Rev.~Lett.
486: \textbf{88}, 114102 (2002).
487: \bibitem {GnutzSS} S.~Gnutzmann, U.~Smilansky and N.~Sondergaard,
488: J.~Phys.~A \textbf{38} 8921 (2005).
489: \bibitem {BandSS} R.~Band, T.~Shapira, and U.~Smilansky, to be
490: published. %6
491: \bibitem {BerryTabor} M. Berry and M. Tabor,
492: Proc.~Roy.~Soc.~Lond.~Ser.~A
493: \textbf{356}, 375 (1977).
494: \bibitem{cdv} Y. Colin de Verdiere, Math.~Z.\textbf{171}, 51 (1980).%8
495: \bibitem{bleher94} P.M. Bleher, Duke~Math.~J. \textbf{74}, 45 (1994).%9
496: \bibitem{kac} M. Kac, Amer.~Math.~Monthly
497: \textbf{73}, 1 (1966).%10
498: \end{thebibliography}
499: \end{document}
500:
501: