1: 

2: \section{Properties of edge state spectra}

3: \label{chap:stat}

4: 

5: 

6: 

7: We apply the weighted spectral densities discussed in the previous

8: chapter to analyze two aspects of the interior and the exterior edge

9: spectra. First we perform a statistical analysis comparing the

10: results to the predictions of random matrix theory. Second, we check

11: the validity of the semiclassical trace formula taking the disk

12: billiard as an example.

13: 

14: \subsection{Universal auto-correlations}

15: \label{sec:auto}

16: 

17: 

18: 

19: 

20: One of the central goals in the field of quantum chaos is to

21: understand how the statistical properties of the quantum spectrum

22: reflect the nature of the underlying classical dynamics

23: \cite{LesHouches91}. We extend these studies to magnetic billiards by

24: making use of the spectral measure of edge states introduced in the

25: previous chapter. It was constructed to focus on the non-trivial part

26: of phase space which is determined by the billiard boundary map

27: \eref{eq:Birmap}.

28: 

29: As a first point we  check whether the edge spectra of both interior

30: \emph{and} exterior magnetic billiards display the universal

31: characteristics of random matrix theory (RMT) if the corresponding

32: skipping motion is hyperbolic.  Our quantity of choice to characterize

33: the spectrum statistically is the spectral form factor $K(\tau)$.  It

34: is sensitive to correlations of the eigenenergies beyond the mean

35: level spacing \cite{Bohigas91}.  The standard form factor was already

36: used in Chapter \ref{chap:numres} to study the two-point correlations

37: in the unweighted spectra of interior billiards.  For edge spectra

38: $K(\tau)$ is readily defined in terms of the 2-point autocorrelation

39: function of the edge density,

40: %

41: \begin{equation}

42:   R_{\nu_0}(\nu)=\int

43:   \dedge^{\rm osc}\Big(\nu'+\frac{\nu}{2}\Big)

44:   \,

45:   \dedge^{\rm osc}\Big(\nu'-\frac{\nu}{2}\Big)

46:   \,

47:   g_1(\nu'-\nu_0)

48:   \,

49:   \rmd\nu'

50:   \PO

51: \end{equation}

52: %

53: Here, we included a normalized Gaussian window function $g_1$

54: to pick up a spectral interval centered at $\nu_0$.

55: 

56: %

57: %

58: %

59: Before comparing to RMT it is advantageous to remove the trivially

60: system-dependent properties of the spectrum by ``unfolding'' it

61: \cite{Bohigas91}.  This is a transformation of the spectrum

62: which renders it dimensionless and of unit mean density.  Dealing with a

63: \emph{weighted} spectrum the unfolding procedure must transform both

64: the energies and the weights.  The natural choice involves the smooth

65: edge state counting function $\Nsm_{\rm edge}$ and the average weight

66: $\langle w^2\rangle / \langle w \rangle$ in the spectral interval

67: considered:

68: \begin{align}

69:   \label{eq:unfoldnuandw}

70:   \unfolded{\nu}_n 

71:   \defas 

72:   \frac{ \langle w \rangle}{ \langle w^2 \rangle}

73:   \,

74:   \Nsm_{\rm edge}(\nu_n)

75: \qq\text{and}\qq

76: %

77: %

78:   \unfolded{w}_n 

79:   \defas

80:   \frac{ \langle w \rangle}{ \langle w^2 \rangle}

81:   \,

82:   w_n

83: \PO

84: \end{align}

85: Here,  the first and second moments of the weights,

86: \begin{align}

87:   \langle w\rangle &= \sum_{n=1}^{\infty} w_n

88:   \;

89:   g(\Nsm_{\rm edge}(\nu_n)

90:   -\unfolded{\nu}_0)

91: \intertext{and}

92:   \langle w^2\rangle &= \sum_{n=1}^{\infty} w_n^2

93:   \;

94:   g(\Nsm_{\rm edge}(\nu_n)

95:   -\unfolded{\nu}_0)

96: \CO

97: \end{align}

98: are taken locally in the spectrum in terms of the window function $g$

99: (a normalized Gau{ss}ian of width $\sigma_g$.)  As a result of this

100: unfolding, both the weights and the weighted density have unit mean.

101: 

102: Since we are dealing with a discrete spectrum, the form factor must be

103: averaged to be well-defined. The standard procedure is to take the

104: spectral average over non-overlapping parts of the spectrum,

105: \begin{align}

106:   \label{eq:formfactordef}

107:   K(\tau)=

108:    \left\langle

109:      \int

110:      \e^{2\pi\rmi\unfolded{\nu}\tau}

111:      R_{\unfolded{\nu}_0}(\unfolded{\nu})\,

112:      g_2(\unfolded{\nu})\,

113:      \rmd\unfolded{\nu}

114:   \right\rangle_{\unfolded{\nu}_0}

115: \end{align}

116: as indicated by the triangular brackets.  According to the spectral

117: ergodicity hypothesis \cite{BFFMPW81} this should be equivalent to an

118: ensemble average for hyperbolic systems.

119: 

120: \begin{figure}[tp]%

121: \begin{center}

122:   \includegraphics[width=\linewidth]

123:   {figs/form2x2.eps}

124: \figurecaption{%

125:   Form factors \eref{eq:formfactorqm} of the interior (a,b) and

126:   exterior (c,d) edge state spectra for the asymmetric stadium (a,c)

127:   and skittle (b,d) billiard at $\rho=1.2$. The shapes are defined in

128:   Fig.~\ref{fig:shapes}.  The functions follow the RMT predictions of

129:   the GOE and GUE ensembles \cite{Bohigas91}, respectively (dashed

130:   lines). The heavy lines correspond to stronger spectral averaging

131:   than the thin lines ($\sigma_g=10$ and $\sigma_g=3$, respectively.)

132:   }

133: \label{fig:edgeformfactor}

134: \end{center}

135: \end{figure}

136: 

137: 

138: If we choose the widths of the Gaussians $g_1$ and $g_2$ as

139: $\sigma_g/\sqrt{2}$ and $\sigma_g\sqrt{2}$, respectively, the Fourier

140: transform of the autocorrelation function leads directly to the power

141: spectrum.  The form factor is then given by the \emph{weighted} sum

142: %

143: \begin{align}

144:   \label{eq:formfactorqm}

145:   K(\tau)=

146:   \left\langle

147:     \frac{2\sqrt{2\pi}\sigma}{ \langle w^2\rangle }

148:     \left| \sum_{n=1}^{\infty}

149:       \unfolded{w}_n\, 

150:       \rme^{2\pi\rmi ( \unfolded{\nu}_n- \unfolded{\nu}_0)\tau}

151:       g(\unfolded{\nu}_n-\unfolded{\nu}_0)

152:       -\hat{g}(\tau)

153:     \right|^2

154:   \right\rangle_{\unfolded{\nu}_0} \ ,

155: \end{align}

156: where the Fourier transform of  $g$ is denoted by $\hat{g}$.

157: 

158: The previous discussion holds for both definitions,

159: \eref{eq:defweight} and \eref{eq:wnmagdef}, of the spectral density of

160: edge states. However, it is necessary to keep the type of the

161: underlying classical motion unchanged during the spectral averaging.

162: This is conveniently done by taking the spectrum in the semiclassical

163: rather than the conventional direction, see Sects.~\ref{sec:scaling}

164: and \ref{sec:qspec}.  In this case the quantum weights $w_n$ are

165: simply obtained by taking the derivatives with respect to $\Lambda$ at

166: fixed $\rho$, see Appendix \ref{app:rho}.  In the present section

167: we use only this first definition of a spectral density of edge states

168: since it is not possible to define magnetization-based weights for the

169: semiclassical direction.

170: 

171: \begin{figure}[t]%

172: \begin{center}

173:   \psfrag{x}{$w$}

174:   \psfrag{y}{$p(w)$}

175:   \includegraphics[width=0.8\linewidth]

176:   {figs/histoweights.eps}

177:   \figurecaption{%

178:     Distribution of the quantum weights $w_n>0.1$ of the interior

179:     (shaded) and exterior (transparent) skittle spectrum at

180:     $\rho=1.2$. 

181:     %

182:     The histograms show peaks whose positions are well reproduced by the

183:     phase space estimates \eref{eq:wavint} and \eref{eq:wavext}

184:     (indicated by the arrows).  Unlike the interior case, the exterior

185:     distribution shows a tail due to the transitional states which ranges

186:     to the small weights. (For normalization (bulk) states with

187:     weights smaller than $0.1$ had to be disregarded.)

188:     }

189:   \label{fig:histoweights}

190: \end{center}

191: \end{figure}

192: 

193: 

194: Figure \ref{fig:edgeformfactor}  shows the form factors for the

195: interior (top) and exterior (bottom) edge state spectra for the

196: asymmetric stadium (left) and skittle (right) billiard, respectively.

197: The spectra were obtained in the semiclassical direction, at fixed

198: $\rho=1.2$, ie, for the same situation as in

199: Fig.~\ref{fig:formfactor}. The weights were obtained by numerical

200: differentiation with respect to $\Lambda$.

201: %

202: We observe that the interior form factors follow the RMT prediction of

203: the Gaussian Orthogonal and Gaussian Unitary Ensembles, respectively,

204: as expected from the specific symmetry properties of the Hamiltonians.

205: In the interior case this is not surprising.  To ensure essential

206: hyperbolicity of the classical motion the value of $\rho$ had to be

207: chosen large such that the interior phase space consists only of

208: skipping trajectories which cover it ergodicly.  As a consequence, one

209: expects that all \emph{interior} states are edge states to an equal

210: degree.  Indeed, the interior weights are distributed narrowly around

211: a mean value $\overline{w}$, given by the ratio of weighted and

212: unweighted mean densities,

213: %

214: \begin{align}

215:   \label{eq:wavint}

216:   \overline{w}=\frac{\dedgesm^{(\rho)}(\nu)}

217:   {\overline{d}_{\rm tot}^{(\rho)}(\nu)}

218:   = \frac{\Len\rho}{4\Area}

219: \CO

220: \end{align}

221: %

222: as can be observed from the shaded histogram in 

223: Fig.~\ref{fig:histoweights}.  The weights do not provide   additional

224: information in this case, which explains why $K(\tau)$ reproduces the

225: RMT prediction, like in the unweighted case.

226: 

227: In contrast, the standard form factor -- like any other standard

228: statistical function -- does not even exist for the exterior spectrum,

229: which is dominated by infinitely many bulk states.  Nonetheless, we

230: find that the exterior spectrum closely obeys the predictions of

231: random matrix theory (bottom row of Fig.~\ref{fig:formfactor}) if

232: viewed in an appropriate way, ie, by means of the edge state density.

233: This way, a crucial test for the consistency of the spectral measure

234: of edge states is passed.  The quantum weights succeed to filter out

235: \emph{selectively} the relevant edge states, which in turn exhibit the

236: universal characteristics expected for chaotic motion.

237: 

238: The distribution of the exterior weights is given by the transparent

239: histogram in Figure \ref{fig:histoweights}. 

240: %

241: One observes that the

242: distribution of large weights is peaked like in the interior case.

243: Again, the peak position is well described by the ratio of weighted

244: and unweighted densities,

245: %

246: \begin{align}

247:   \label{eq:wavext}

248:   \overline{w}=

249:   \frac{\dedgesm^{(\rho)}(\nu)}

250:   {\overline{d}_{\rm skip}^{(\rho)}(\nu)}

251:   \simeq 

252:   \frac{\Len\rho}{2\Area_{\rm skip}^{\rm ext}(\rho)}

253:   \CO

254: \end{align}

255: %

256: with the mean unweighted density now given by the phase space estimate

257: \eref{eq:dskip} of skipping states. Unlike the interior case, the

258: distribution has a tail of transitional states which ranges to the

259: infinitely many bulk states with small weights. 

260: 

261: 

262: 

263: \begin{figure}[t]%

264: \begin{center}

265:   \psfrag{t}{$t$}

266:   \psfrag{y1}{\hspace*{-5.1em}

267:     $-|\widehat{d}^{\rm (\rho, sc)}_{\rm edge}\!(t)|

268:     \q|\widehat{d}^{\rm (\rho)^{osc}}_{\rm edge}\!(t)|$}

269:   \psfrag{y2}{\hspace*{-5.1em}

270:     $-|\widehat{d}^{\rm (\rho, sc)}_{\rm edge}\!(t)| 

271:     \q|\widehat{d}^{\rm (\rho)^{osc}}_{\rm edge}\!(t)|$}

272:   \includegraphics[width=\linewidth]

273:   {figs/accompext.eps}

274:   \figurecaption{%

275:     Action spectrum of the \emph{exterior} disk at $\rho=2 R$.  The

276:     positive values give the Fourier transform \eref{eq:dftdef} of the

277:     exterior edge density (absolute values).  The positions of the

278:     peaks are well reproduced by the trace formula \eref{eq:dedgecirc}

279:     (negative values) -- except for the small peaks at integer $t$

280:     which are remnants of the bulk states. The peak heights match well

281:     in most cases; they are expected to fit better if a spectral

282:     interval larger than $\nu\in[0;48]$ is used.  }

283:   \label{fig:actionsext}

284: \end{center}

285: \end{figure}

286: 

287: \begin{figure}[pt]%

288: \begin{center}

289:   \psfrag{y1}{\hspace*{-4.1em}\small

290:     $-|\widehat{d}^{\rm (\rho, sc)}_{\rm edge}\!(t)|

291:     \q|\widehat{d}^{\rm (\rho)^{osc}}_{\rm edge}\!(t)|$}

292:   \psfrag{y2}{\hspace*{-4.1em}\small

293:     $-|\widehat{d}^{\rm (\rho, sc)}_{\rm edge}\!(t)| 

294:     \q|\widehat{d}^{\rm (\rho)^{osc}}_{\rm edge}\!(t)|$}

295:   \psfrag{y3}{\hspace*{-4.1em}\small

296:     $-|\widehat{d}^{\rm (\rho, sc)}_{\rm edge}\!(t)| 

297:     \q|\widehat{d}^{\rm (\rho)^{osc}}_{\rm edge}\!(t)|$}

298:   \psfrag{t}{$t$}

299:   \includegraphics[width=\linewidth]

300:   {figs/accompint.eps}

301:   \figurecaption{%

302:     Action spectrum of the \emph{interior} disk at a cyclotron radius

303:     $\rho=0.4\times R$ small enough to enable bulk states.  The

304:     Fourier transform \eref{eq:dftdef} of the interior edge density

305:     (positive values, $\nu\in[0;60]$) is well reproduced by the trace

306:     formula (negative values).  Note that in the top part, which shows

307:     the remnant peaks of bulk contributions, the y-axis has a different

308:     scale.}

309:   \label{fig:actionsint}

310: \end{center}

311: \end{figure}

312: 

313: 

314: 

315: \subsection{The action spectrum}

316: \label{sec:acspec}

317: 

318: We turn from the statistical analysis of edge spectra to their

319: semiclassical description.  Here, the main purpose is to show that the

320: trace formula for the edge state density -- which rates each periodic

321: orbit with a classical weight -- succeeds in approximating the exact

322: edge spectrum.

323: 

324: We choose the disk billiard for which an explicit periodic orbit

325: formula is readily obtained from equation \eref{eq:Noscdisk2}. For the

326: exterior case and $\Gam=R/\rho<1$ we find, see \eref{eq:dedgedef2},

327: %

328: \begin{align}

329:   \label{eq:dedgecirc}

330:     d^{\rm osc}_{\rm edge}(\nu) = &

331:   \left(\frac{2\nu}{\pi}\right)^\oh

332:   \sum_{\n=2}^\infty\,

333:   \frac{2}{\n^{1/2}}

334:   \sum_{\Delta\phi\in\PSet_{\rm ext}^\n} 

335:   \Big|\sin\Big(\alpha-\frac{\Delta\phi}{2}\Big)\Big|

336:   \\

337:  & \times

338:   \frac{\sin(\alpha)\cos(\alpha)-\oh\Gam^2\sin(\Delta\phi)}

339:     {(\sin(\alpha)\cos(\alpha))^\oh}

340:     \cos\left(2\pi\nu \n\,\ga_{\rm L}(\Delta\phi)+\n\frac{\pi}{2}-\piof\right)

341: \CO

342: \nn

343: \end{align}

344: %

345: with $\alpha$ defined by equation \eref{eq:sadisk}. (The term

346: $|\sin(\alpha-\frac{\Delta\phi}{2})|$ corresponds to the normal

347: component of the velocity, $|\nvec\,\vvech|$,

348: in \eref{eq:dedgeosc2}.)

349: %

350: Moreover, the exact quantum spectrum of the disk is calculated

351: %

352: relatively easily in terms of the roots of special functions, see

353: App.~\ref{app:diskexact}.  We calculated spectral intervals large

354: enough so that the Fourier transformation of the spectral densities,

355: %

356: \begin{align}

357:   \label{eq:dftdef}

358:   \widehat{d}^{{\rm osc}}_{\rm edge}\!(t) =

359:   \int \rme^{2\pi\rmi\nu t}  d^{{\rm osc}}_{\rm edge}\!(\nu) 

360: %

361: %

362:   \,h(\nu-\nu_0)\, \rmd\nu

363:   \CO

364: \end{align}

365: %

366: resolves the classical actions $t$ of the underlying periodic orbits.

367: Here, the function $h$ is a suitable window centered on the midpoint

368: $\nu_0$ of the spectral interval.

369: %

370: This \emph{action spectrum} may be readily compared to the

371: semiclassical  prediction based on

372: \eref{eq:dedgecirc}.

373: 

374: 

375: Like in the previous section, it is  convenient to take the 

376: spectrum in the semiclassical direction, at constant $\rho$.

377: %

378: %

379: %

380: In Figure \ref{fig:actionsext} we show the action spectrum for the

381: \emph{exterior} disk at a cyclotron radius $\rho=2 R$ (positive

382: values). The corresponding prediction of the trace formula

383: \eref{eq:dedgecirc} is given by the negative values.  One observes

384: that the peak positions match very well with the predictions of

385: semiclassical theory. The only exception are the small peaks at

386: integer actions which are not reproduced semiclassically.  They

387: %

388: are remnants of the infinite number of bulk states.

389: %

390: The peak heights are well reproduced most of the time, except if two

391: peaks overlap too strongly.  These deviations are expected to fade

392: as a larger spectral interval is used and the the widths of

393: the peaks decrease.

394: %

395: This is also seen in Figure \ref{fig:actionsint} where we present the exact

396: and semiclassical action spectra of the \emph{interior} magnetic disk -- based

397: on a large spectral interval ($\nu\in[0;60]$ at $\rho=0.4\times R$). Here,

398: the cyclotron radius was chosen small enough for bulk

399: states to exist in the interior.

400: %

401: %

402: One observes again that the latter are very efficiently

403: suppressed in the action spectrum giving rise only to the small peaks

404: at integer values (shown in the top part of Fig.~\ref{fig:actionsint}). 

405: %

406: %

407: In the Fourier transform of the \emph{unweighted} density, in

408: contrast, the bulk states  obliterate the edge contributions such

409: that not a single action is resolved (not shown).  Blaschke and Brack

410: \cite{BB97a} analyzed semiclassically the spectrum for the unweighted

411: interior problem. The contribution of the bulk states was estimated

412: and added by hand resulting in an unsatisfactory agreement between the

413: semiclassical and the quantum spectra.  

414: 

415: In conclusion, we find that the the semiclassical trace formula

416: succeeds in reproducing the quantum edge state density. It does so by

417: weighting each periodic orbit contribution with a classical weight

418: which vanishes for cyclotron orbits. This removes the bulk

419: contributions analogous to -- and consistent with -- the quantum

420: weights of the edge state density.

421: 

422: 

423: %

424: \begin{figure}[t]%

425:   \begin{center}%

426:     \psfrag{sqrt(5)/2}{$\ds\oh\sqrt{5}$}

427:     \psfrag{1.0}{$1.0$}

428:     \includegraphics[width=0.9\linewidth] 

429:     {figs/s14shape.eps}

430:     \figurecaption{%

431:       Definition of the Bunimovich stadium used in

432:       Sect.~\ref{sec:magstadium} and Chapter \ref{chap:cross}. 

433:       }

434:     \label{fig:s14shape}

435:   \end{center}%

436: \end{figure}

437: 

438: 

439: 

440: \subsection{Using the edge magnetization}

441: \label{sec:magstadium}

442: 

443: 

444: Finally, let us demonstrate that the edge state density may as well be

445: defined in terms of the magnetization as discussed in Section

446: \ref{sec:edgemag}.

447: 

448: 

449: Choosing the Bunimovich stadium billiard (defined in

450: Fig.~\ref{fig:s14shape}) we calculated the interior and exterior

451: magnetization spectrum in the high-energy direction, at $b=0.2$.  The

452: selected spectral interval $\nu\in[100;135]$ corresponds to large

453: cyclotron radii $\rho\in[2;2.32]$ giving rise to essentially

454: hyperbolic\footnote{The term ``essentially hyperbolic'' means that

455:   although there might be small integrable parts in phase space their

456:   combined area is much smaller than the uncertainty product

457:   $(b^2\pi)^2$.} classical motion.

458: %

459: Quantum mechanically, the problem exhibits one unitary and one

460: anti-unitary symmetry (rotation by $\pi$ and reflection at one axis,

461: respectively).  Hence, the spectrum decomposes into two symmetry

462: classes (a feature which will be used in the next chapter) while each

463: class should obey the characteristics of the Gaussian Orthogonal

464: Ensemble \cite{Bohigas91}.

465: %

466: %

467: 

468: 

469: \begin{figure}[tp]%

470:   \begin{center}%

471:     \psfrag{x}{$\nu_n$}

472:     \psfrag{yext}{\hspace*{-2.0em}exterior $w^\magn_n$}

473:     \psfrag{yint}{\hspace*{-2.0em}interior $w^\magn_n$}

474:     \includegraphics[width=0.9\linewidth] 

475:     {figs/mweights_stadium.eps}

476:     \figurecaption{%

477:       Weights as obtained from the edge magnetization for the stadium

478:       billiard at $b=0.2$ and high energy.  Note that the values of

479:       the weights differ in the (top) and exterior (bottom) by a

480:       factor of about ten.  Notwithstanding, the mean edge state

481:       densities are equal to leading order. The classical cyclotron

482:       radius which corresponds to this part of the spectrum is large,

483:       $\rho\in[2;2.32]$, giving rise to essentially hyperbolic

484:       classical motion.

485: %

486: %

487: %

488:  }

489:     \label{fig:s14magspec}

490:   \end{center}%

491: \end{figure}

492: 

493: 

494: 

495: \begin{figure}[tp]%

496:   \begin{center}%

497:     \includegraphics[width=0.66\linewidth] 

498:     {figs/forms14.eps}

499:     \figurecaption{%

500:       Form factor of the \emph{exterior} Bunimovich stadium

501:       (Fig.~\ref{fig:s14shape}) computed from the edge magnetization

502:       spectrum shown in Fig.~\ref{fig:s14magspec}, bottom part.  }

503:     \label{fig:mform}

504:   \end{center}%

505: \end{figure}

506: 

507: 

508: The weighted spectra are shown in Figure \ref{fig:s14magspec}.  Note

509: that the weights are very different in magnitude, although they lead

510: to the same average edge magnetization \eref{eq:Magsmboth}.  This

511: is explained by the different areas $\Area_{\rm skip}$ of the interior

512: and exterior skipping motion since the mean weight is asymptotically

513: determined by the ratio

514: %

515: \begin{align}

516:   \label{eq:meanu}

517:   \overline{w}^\magn = 

518:   \frac{\pm\overline{\mm}_{\rm edge}(\nu)}

519:   {\nu\,\overline{d}_{\rm skip}(\nu)}

520:   \simeq

521:  \frac{\Area}{\Area_{\rm skip}(\rho=\sqrt{\nu}b)}

522: \CO

523: \end{align}

524: %

525: see \eref{eq:dskip} and \eref{eq:mmedgesm}; (in the interior case

526: $\Area_{\rm skip}=\Area$). 

527: Similar to Fig.~\ref{fig:histoweights} 

528: the distributions

529: of the magnetization weights are localized

530: at $\overline{w}^\magn$ (not shown).

531: 

532: 

533: 

534: Figure \ref{fig:mform} presents the form factor

535: \eref{eq:formfactordef} of the exterior magnetization spectrum

536: restricting the energies to a single symmetry class. As one expects

537: the form factors follows the GOE prediction (dashed line).  This

538: indicates that the weights \eref{eq:wnmagdef} based on the

539: magnetization succeed to filter the bulk states consistently.  They

540: perform as well as the weights \eref{eq:defweight} based on the

541: boundary condition.

542: 

543: