1: 

2: \section[Quantization in the interior and the exterior]

3: {Quantization in the interior and the exterior: The boundary integral method}

4: 

5: \label{chap:bim}

6: 

7: 

8: In the present chapter, we show how to solve the quantization problem

9: for interior and exterior magnetic billiards by means of a boundary

10: integral method.  It provides the spectra and wave functions of

11: arbitrarily shaped billiard domains, and includes the general boundary

12: conditions discussed in Section \ref{sec:bcond}.  Moreover, the

13: boundary integral formalism constitutes the basis for the

14: semiclassical theory discussed in Chapter \ref{chap:trace}.

15: 

16: \subsection{Boundary methods}

17: 

18: As compared to the field free case, it is surprisingly difficult 

19: to obtain the quantum spectra of magnetic billiards.

20: %

21: %

22: %

23: %

24: So far, numerical studies were restricted to the interior problem and

25: performed almost exclusively by diagonalizing the

26: Hamiltonian\cite{NT88,BGOdAS95,YH95,JB95,deAguiar96}.  This requires

27: the choice and truncation of a basis, which is problematic for general

28: billiards, where no natural magnetic basis set exists.  Consequently,

29: results were limited to the first few hundred eigenvalues.

30: 

31: In the case of field free billiards

32: quantum spectra  are usually obtained by

33: transforming the eigenvalue problem into an integral equation of lower

34: dimension. The corresponding integral operator is defined in terms

35: of the free Green function, and depends only on the boundary

36: \cite{KR74,Riddell79,BW84,Boasman94,KS97,LRH98}.  

37: %

38: This method is known to be more efficient than diagonalization by an

39: order of magnitude \cite{MK88,CLH01}. 

40: %

41: %

42: %

43: %

44: We proceed to extend these ideas to magnetic billiards. A step in this

45:  direction was taken by Tiago \etal \cite{TCA97}, who

46:  essentially propose a null-field method\footnote{The authors of

47:  \cite{TCA97} inaccurately call their scheme a ``boundary integral

48:  method''. }  \cite{Martin82} for (interior) magnetic billiards.  It

49:  involves the irregular Green function

50: %

51: \eref{eq:Girrangular} in angular

52: momentum decomposition.  A drawback of the approach is that this function

53: must be known for large angular momenta, which turns out to be 

54: numerically impractical.

55: Moreover, the method does not apply for the exterior problem. 

56: 

57: In the following we derive the boundary integral method for magnetic

58: billiards.  Like in the field free case, it involves the regular

59: Green function in position space representation. We present the method

60: for the interior and the exterior problem, and general boundary

61: conditions.

62: 

63: 

64: \subsection{The boundary integral equations}

65: \label{sec:bie}

66: 

67:  

68: %

69: %

70: %

71: 

72: \subsubsection{Single and double layer equations}

73: 

74: The stationary eigenfunction 

75: %

76: of a magnetic billiard at

77: energy $\nu$ is defined by the differential equation

78: %

79: \begin{gather}

80: \label{eq:Schreq2}

81: \left({\tfrac{1}{2}}\big(-\rmi\grad_{\!r/b} -

82: \tilde{\mb{A}}(\rvec)\big)^2 

83: -2\nu\right)

84: \psi(\rvec)

85: =0\CO

86: \end{gather}

87: %

88: and a specification of the wave function on the billiard boundary

89: $\Boundary$.  

90: %

91: %

92: %

93: %

94: %

95: %

96: The free Green function, $\Gnu$, was shown

97: %

98: to satisfy the {inhomogeneous} Schr{\"o}dinger equation

99: %

100: \begin{align}

101: \label{eq:Gdef}

102: \left(\tfrac{1}{2}\big(-\rmi\grad_{\!r/b} 

103:   - \tilde{\mb{A}}(\rvec)\big)^2 -2\nu\right)

104: \Gnu(\rvec;\rvec_0)

105: = - \tfrac{1}{2}\,\delta\!\left(\frac{\rvec-\rvec_0}{b}\right)

106: \PO

107: \end{align}

108: %

109: %

110: %

111: %

112: %

113: %

114: %

115: %

116: %

117: %

118: %

119: %

120: %

121: %

122: %

123: %

124: 

125: Our goal is to cast the quantization problem into an integral

126: equation defined on the billiard boundary.

127: %

128: %

129: %

130: To that end,

131: we take the complex conjugate of \eref{eq:Schreq2} and

132: multiply it (from the left) with $\Gnu$.  Similarly, equation

133: \eref{eq:Gdef} is multiplied with $\psi^*$ and subtracted from the

134: former expression. One obtains an equation

135: %

136: \begin{gather}

137:   \label{eq:deleq}

138:   \psi^*\, \grad_{\!r/b}^2 \Gnu

139:   - \Gnu\,\grad_{\!r/b}^2 \psi^*

140:   -2 \rmi\, \grad_{\!r/b} \big( \tilde{\mb{A}}\psi^*\Gnu \big) 

141:   = \psi^*\,  \delta\!\left(\frac{\rvec-\rvec_0}{b}\right)\CO

142: \end{gather}

143: %

144: %

145: which has a form suitable for the Green and Gau{ss} integral theorems.

146: It holds everywhere in the plane, except for the boundary $\Boundary$,

147: where the boundary condition \eref{eq:bcond} introduces a

148: discontinuity in the derivative of $\psi$.

149: 

150: 

151: We start by considering the {interior} problem and sketch the treatment of

152: the {exterior} case afterwards.

153: Choosing the initial point of the Green function away from the

154: boundary, $\rvec_0\in \mathbb{R}^2\setminus\Boundary$, the integral of

155: \eref{eq:deleq} over the (interior) domain $\Domain$ may be transformed to a

156: line  integral,

157: %

158: %

159: %

160: %

161: %

162: %

163: %

164: %

165: %

166: %

167: %

168: %

169: %

170: %

171: %

172: %

173: %

174: \begin{multline}

175: \label{eq:split}

176: \int_{\Boundary}

177: \big[

178: \psi^*\, (\dnb\Gnu-\rmi\,\At_n\,\Gnu)

179: -\Gnu\,(\dnb\psi^*+\rmi\,\At_n\,\psi^*)

180: \big] \frac{{\rm d}\Boundary}{b}

181: \\

182: =

183: \begin{cases}

184: \psi^*(\rvec_0)&\text{if $\rvec_0\in \openDomain $ }

185: \\

186: 0&\text{if $\rvec_0\in \mathbb{R}^2 \setminus{\Domain}$. }

187: \end{cases}

188: \end{multline}

189: %

190: It is defined on the boundary $\Boundary$ (with the normal components

191: of the vector potential and the gradient denoted as

192: $\At_n=\nvec(\rvec)\tilde{\mb{A}}$ and $\dnb\defas

193: b\nvec(\rvec)\grad_{\!r}$, respectively).

194: %

195: Note that the vector potential part of the integrand was split

196: which is necessary for a gauge invariant formulation of the

197: integral equations. 

198: 

199: \myparagraph{The single layer equations}

200: 

201: 

202: We choose $\rvec_0 \in \Boundary$ and define

203: $\rvec_0^\pm\defas\rvec_0\pm\epsilon \nvec_0$, for small $\epsilon>0$.

204: By adding the two equations in \eref{eq:split}, one obtains

205: %

206: \begin{gather}

207: \label{eq:SLepsin}

208: \int_{\Boundary}

209: \big[

210: \psi^*\, ({\partial}_{n/b}\epscomb{\Gnu} -\rmi\, \At_n\, \epscomb{\Gnu} )

211: - \epscomb{\Gnu}\, (\dnb\psi^*+\rmi\,\At_n\,\psi^*) )

212: \big] 

213: \frac{{\rm d}\Boundary}{b}

214: =

215: {\tfrac{1}{2}}\psi^*(\rvec_0^-)\PO

216: \end{gather}

217: %

218: Here, we used the abbreviation

219: $\epscomb{\Gnu}={\tfrac{1}{2}}\Gnu(\rvec;\rvec_0^+)

220: +{\tfrac{1}{2}}\Gnu(\rvec;\rvec_0^-)$.

221: %

222: %

223: %

224: Equation \eref{eq:SLepsin} holds for all (sufficiently small)

225: $\epsilon>0$, hence the limit

226: $\epsilon\to0$ exists.  Moreover, observing the asymptotic

227: properties of the Green function (cf Sect. \ref{sec:Gprop}), it can be

228: shown, that the integration and the limit $\epscomb{\Gnu}\to\Gnu$,

229: ${\partial}_{n/b}\epscomb{\Gnu}\to\partial_{n/b}\Gnu$ may be

230: interchanged.  Inserting the boundary condition \eref{eq:bcond} we

231: obtain, after renaming the limiting function

232: $u=\dnb\psi^*+\rmi\At_n\psi^*$, $u_0\defas u(\rvec_0)$,

233: %

234: \begin{gather}

235:   \label{eq:SingleLayerin}

236: \int_{\Boundary}

237: \big[

238: \,\Gnu-\frac{\lambda}{b}

239: (\dnb\Gnu -\rmi\, \At_n\, \Gnu)

240: \big]\, u \,\frac{{\rm d}\Boundary}{b}

241: =

242: \frac{\lambda}{b}\, (- {\tfrac{1}{2}}u_0)

243: \CO

244: \end{gather}

245: %

246: %

247: an integral equation defined on the boundary $\Boundary$ \cite{HS00a}.

248: 

249: In order to derive the corresponding equation for the exterior

250: problem, consider a large disk $\mathcal{K}_p\supset\Domain$ of radius

251: $p$, and integrate \eref{eq:deleq} over

252: $\mathcal{K}_p\cap\openDomain$.  Once $\rvec_0$ lies in the vicinity

253: of $\Boundary$, the contribution of $\partial\mathcal{K}_p$ to the

254: boundary integral vanishes as $p\to\infty$, due to the exponential

255: decay of the regular Green function $\Gnu$.

256: %

257: %

258: Similar to eq \eref{eq:SLepsin} one  obtains an equation

259: %

260: %

261: %

262: %

263: %

264: %

265: %

266: %

267: %

268: %

269: %

270: %

271: %

272: %

273: %

274: which permits the limit $\epsilon\to 0$ to be taken before

275: performing the integration.  The resulting boundary integral equation

276: differs from \eref{eq:SingleLayerin} only by a sign.  In the

277: following, we shall treat both cases simultaneously, with the

278: convention that the upper sign stands for the interior problem, and

279: the lower sign for the exterior one,

280: %

281: \begin{gather}

282:   \label{eq:SingleLayer}

283: \int_{\Boundary}

284: \big[\,

285: \Gnu\mp\frac{\lambda}{b}

286: (\dnb\Gnu -\rmi\, \At_n\, \Gnu)

287: \big]\, u\, \frac{{\rm d}\Boundary}{b}

288: =

289: \frac{\lambda}{b}\, (- {\tfrac{1}{2}}u_0)

290: \PO

291: \end{gather}

292: %

293: %

294: In analogy to the Helmholtz problem  \cite{KR74}, 

295: we will refer to these

296: equations as the \emph{single layer equations} for the interior and

297: the exterior domain.

298: 

299: \myparagraph{The double layer equations}

300: 

301: A second kind of boundary integral equations can be derived by applying

302: the differential operator $(\dnnb +\rmi

303: \At_{n_0})\defas\nvec(\rvec_0)(\grad_{\!r_0/b}+\rmi\tilde{\mb{A}}(\rvec_0)$

304: on equation \eref{eq:SLepsin},

305: %

306: %

307: \setlength{\multlinegap}{0.1\textwidth}

308: \begin{multline}

309:   \label{eq:DLeps}

310: %

311:   \int_{\Boundary}

312:   \psi^*\,

313:   ({\partial}_{n_0/b}+\rmi\, \At_{n_0})

314:   ({\partial}_{n/b}\epscomb{\Gnu}-\rmi\, \At_n\,\epscomb{\Gnu} ) 

315:   \frac{{\rm d}\Boundary}{b}

316:   \\

317:   - 

318:   \int_{\Boundary}

319:   ({\partial}_{n_0/b}\epscomb{\Gnu} +\rmi\, \At_{n_0}\, \epscomb{\Gnu} )

320:   (\dnb\psi^*+\rmi\At_n\psi^*)

321:   \frac{{\rm d}\Boundary}{b}

322:   \\

323:   =

324:   \pm{\tfrac{1}{2}}

325:   (\dnnb+\rmi\,\At_{n_0})\psi^*(\mb{r^\mp_0})\PO

326: \end{multline}

327: %

328: This equation is true for all $\epsilon >0$, which means that the

329: limit $\epsilon\to 0$ exists. As for the second integral, we may again

330: permute the limit and the integration which yields a proper integral.

331: Consequently, the limit of the first integral is finite, too.

332: However, in the first integral we are not allowed to exchange the

333: integration with taking the limit because the limiting integrand

334: \eref{eq:QexpslN} has a $1/(\rvec-\mb{r_0})^2$-singularity which is

335: not integrable (see below).

336: 

337: 

338: Integral operators of this kind are named \emph{hypersingular}

339: \cite{Guiggiani98}. Similar to a Cauchy principal value integral, they

340: are defined by taking a special limit. However, compared to the

341: principal value the singularity is stronger by one order in the

342: present case. Below, in

343: Section~\ref{sec:bops}, 

344: we define which limit is to be taken.  It is

345: denoted by $\tfpint$ and should be read ``finite part of the

346: integral''. With this concept and equation \eref{eq:bcond}, we obtain the

347: \emph{double~ layer} equations,

348: %

349: \begin{multline}

350:   \label{eq:DoubleLayer}

351:   \int_{\Boundary}

352:    (\dnnb\Gnu +\rmi\, \At_{n_0}\,\Gnu  )

353:    \,u\,

354:    \frac{{\rm d}\Boundary}{b}

355:    \\

356:   \mp

357:   \frac{\lambda}{b}

358:    \fpint_{\Boundary}

359:    ({\partial}_{n_0/b} +\rmi\, \At_{n_0})

360:    ({\partial}_{n/b}\Gnu -\rmi\, \At_{n}\,\Gnu)

361:    \,u\, 

362:    \frac{{\rm d}\Boundary}{b}

363:   = 

364:   \mp \tfrac{1}{2}u_0

365: \CO

366: \end{multline}

367: %

368: which are again integral equations defined on the boundary

369: $\Boundary$.

370: 

371: \myparagraph{The spectral determinants}

372: 

373: It is useful to introduce a set of integral operators (whose labels D

374: and N indicate correspondence to pure Dirichlet or Neumann

375: conditions):

376: \begin{align}

377:   \label{eq:Opdef}

378:   \mathsf{Q}^{\rm D}_{\rm sl} [u] &= 

379:   \int_{\Boundary}\!

380:   {\rm d}\Boundary

381:   \,\Gnu\, u

382:   \\

383:   \label{eq:Opdef2}

384:   \mathsf{Q}^{\rm N}_{\rm sl} [u] &= 

385:   \int_{\Boundary}\!

386:   \frac{{\rm d}\Boundary}{b}

387:   (\dnb\Gnu -\rmi\, \At_n\, \Gnu)

388:   \,u

389:   \\

390:   \label{eq:Opdef3}

391:   \mathsf{Q}^{\rm D}_{\rm dl} [u] &= 

392:   \int_{\Boundary}\!

393:   \frac{{\rm d}\Boundary}{b}

394:    (\dnnb\Gnu   +\rmi\, \At_{n_0}\,\Gnu   )

395:    \,u

396:    \\

397:   \label{eq:Opdef4}

398:   \mathsf{Q}^{\rm N}_{\rm dl} [u] &=  \,\,

399:    \fpint_{\Boundary}\!

400:    \frac{{\rm d}\Boundary}{b^2}

401:    ({\partial}_{n_0/b} +\rmi \At_{n_0})

402:    ({\partial}_{n/b}\Gnu -\rmi\, \At_{n}\,\Gnu)

403:     \,u

404: \end{align}

405: %

406: They act in the space of square-integrable periodic functions,

407: $u\in\Hilbert(\Boundary)$, with the period given by the circumference

408: $\Len$.

409: 

410: Nontrivial solutions of the single  layer equations  \eref{eq:SingleLayer}

411: and double layer equations \eref{eq:DoubleLayer} exist

412: %

413: for energies where the corresponding Fredholm determinants vanish,

414: %

415: \begin{align}

416:   \label{eq:detsl}

417:   \det\left[

418:     \mathsf{Q}^{\rm D}_{\rm sl}

419:     \mp \lambda

420:     \mathsf{Q}^{\rm N}_{\rm sl}

421:     +  \frac{\lambda}{2} \mathsf{id}

422:   \right] 

423:   &= 0

424:   \q\q\mbox{(single layer)}

425: \\

426:   \label{eq:detdl}

427:   \det\left[

428:     \mathsf{Q}^{\rm D}_{\rm dl}

429:     \mp \lambda

430:     \mathsf{Q}^{\rm N}_{\rm dl}

431:     \pm \frac{1}{2} \mathsf{id}

432:   \right] 

433:   &= 0

434:   \q\q\mbox{(double layer).}

435: \end{align}

436: %

437: Hence,

438: these are secular equations although the explicit dependence on the

439: spectral variable is not shown in our abbreviated notation.  

440: However, each of the determinants \eref{eq:detsl}

441: and \eref{eq:detdl} may have roots, which do \emph{not} correspond to

442: solutions of the original eigenvalue problem given by

443: \eref{eq:Schreq2} and \eref{eq:bcond}.  For finite $\epsilon$, the

444: equations \eref{eq:SLepsin} and \eref{eq:DLeps}

445: are still equivalent to the latter. They acquire additional spurious

446: solutions only as they are transformed to boundary integral equations

447: by the limit $\epsilon\to0$.

448: 

449: \subsubsection{Spurious solutions and the combined operator}

450: \label{ssec:spurious}

451: 

452: The physical origin of the redundant zeros is apparent in our gauge

453: invariant formulation:  They are proper solutions for the domain

454: \emph{complementary} to the one considered.  This is obvious for the

455: single layer equation with Dirichlet boundary conditions ($\lambda=0$),

456: where the spectral determinant does not depend on the orientation of

457: the normals. The same is true for the double layer equation with

458: Neumann boundary conditions ($\lambda^{-1}=0$).

459: 

460: In general, the character of the spurious solutions may be summarized

461: as follows: Independently of the boundary conditions, the

462: \emph{single layer} equation includes the \emph{Dirichlet} solutions

463: of that domain which is complementary to the one considered.

464: Likewise, the \emph{double layer} equation is polluted by the

465: \emph{Neumann} solutions of the complementary domain, irrespective of

466: the boundary conditions employed.

467: %

468: This statement is easily proved by observing that the

469: single-layer-Neumann operator and the double-layer-Dirichlet operator

470: are \emph{adjoint} to each other, \mbox{$\mathsf{Q}^{\rm N}_{\rm

471: sl}=(\mathsf{Q}^{\rm D}_{\rm dl})^\dagger$}, while the operators

472: $\mathsf{Q}^{\rm D}_{\rm sl}$ and $\mathsf{Q}^{\rm N}_{\rm dl}$ are

473: self-adjoint (see below).

474: %

475: Now assume that $u$ is a complementary Dirichlet solution. In Dirac

476: notation,

477: %

478: \begin{align}

479:   \mathsf{Q}^{\rm D}_{\rm sl} \ket{u} =0&

480:   \q\wedge\q

481:   \mathsf{Q}^{\rm D}_{\rm dl} \ket{u} \mp \tfrac{1}{2} \ket{u} =0

482: \\

483:   {\Rightarrow}\qq

484:   \bra{u}\mathsf{Q}^{\rm D}_{\rm sl} =0&

485:   \q\wedge\q

486:   \bra{u}\mathsf{Q}^{\rm N}_{\rm sl} \mp \tfrac{1}{2} \bra{u} =0\PO

487:   \nn

488: \end{align}

489: %

490: Applying the dual of $u$ to the single layer operator yields

491: \begin{align}

492:   \bra{u} \mathsf{Q}^{\rm D}_{\rm sl}  

493:   \mp \lambda \Big\{\bra{u}\mathsf{Q}^{\rm N}_{\rm sl}  

494:                     \mp \tfrac{1}{2} \bra{u}\Big\}

495:   = 0\CO

496: \end{align}

497: which implies that the Fredholm determinant of the single layer

498: operator vanishes.  Similarly, if $u$ is a complementary Neumann

499: solution,

500: \begin{align}

501:   &\pm\mathsf{Q}^{\rm N}_{\rm sl} \ket{u} + \tfrac{1}{2} \ket{u} =0

502:   \q\wedge\q

503:   \mathsf{Q}^{\rm N}_{\rm dl} \ket{u} =0

504: \\

505:   {\Rightarrow}\qq

506:   &\pm\bra{u}\mathsf{Q}^{\rm D}_{\rm dl} + \tfrac{1}{2} \bra{u} =0

507:   \q\wedge\q

508:   \bra{u}\mathsf{Q}^{\rm N}_{\rm dl}  =0

509:   \nn

510: \end{align}

511: then its dual satisfies the double layer equation, again for any $\lambda$,

512: \begin{align}

513:   \pm\Big\{\pm\bra{u} \mathsf{Q}^{\rm D}_{\rm dl} + \tfrac{1}{2} \bra{u}\Big\}

514:   \mp \lambda \bra{u}\mathsf{Q}^{\rm N}_{\rm dl}

515:   = 0\PO

516: \end{align}

517: %

518: Since the spurious solutions are never of the same type, it is possible

519: to dispose of them by requiring that both, the single and the double

520: layer equations, should be satisfied by the \emph{same} solution $u$.

521: Therefore, one obtains a necessary and sufficient condition for the

522: definition of the spectrum by considering a \emph{combined} operator

523: %

524: \begin{gather}

525:   \label{eq:combdef}

526:   \mathsf{Q}_{\rm c}^\pm

527:   \defas

528:   \left(

529:     \mathsf{Q}^{\rm D}_{\rm dl}

530:     \mp \lambda

531:     \mathsf{Q}^{\rm N}_{\rm dl}

532:     \pm \frac{1}{2} \mathsf{id}

533:   \right)

534:   + \rmi \aconst    

535:   \left(

536:     \mathsf{Q}^{\rm D}_{\rm sl}

537:     \mp \lambda

538:     \mathsf{Q}^{\rm N}_{\rm sl}

539:     +  \frac{\lambda}{2} \mathsf{id}

540:   \right)\CO

541: \end{gather}

542: %

543: with an arbitrary constant  $\aconst$.

544: It has a zero eigenvalue only if both, single and double layer

545: operators do.  In practice, the spectrum is

546: obtained by finding the roots of the \emph{spectral function}

547: %

548: \begin{gather}

549:   \label{eq:xibim}

550:   \xi(\nu) = \det\left[\mathsf{Q}_{\rm c}^\pm\right]

551: \PO

552: \end{gather}

553: %

554: %

555: %

556: %

557: %

558: %

559: %

560: %

561:  

562: %

563: %

564: %

565: %

566: %

567: %

568: %

569: 

570: It is worthwhile noting that (for the interior problem) spurious

571: solutions will not appear if one uses the irregular Green function.

572: The reason is that the gauge-independent part of this function is

573: \emph{complex}, which destroys the mutual adjointness of the

574: operators.  This is why the irregular Green function had to be chosen

575: for the null-field method \cite{TCA97}.  For the boundary integral

576: method, the option to use this exponentially divergent solution of

577: \eref{eq:Gdef} is excluded, since the corresponding operator would get

578: arbitrarily ill-conditioned once the diameter of the domain $\Domain$

579: exceeds the cyclotron diameter.  The exterior problem cannot even

580: formally be solved using ${\rm G}^{\rm (irr)}_\nu$ (due to an

581: essential singularity at the origin).

582: 

583: 

584: A last remark is concerned with the important case of Dirichlet

585: boundary conditions.  Here, one could as well derive a pair of

586: boundary integral equations that are \emph{not} gauge-invariant.

587: (Just set $\psi^*=0$ in \eref{eq:split} and consider $u=\dnb\psi^*$.)

588: Of course, these equations would yield all the proper gauge-invariant

589: eigen-energies of the problem. However, the energies of the additional

590: spurious solutions would depend on the chosen gauge, and a

591: characterization of the latter in terms of solutions of a

592: complementary problem would not be possible.

593: 

594: The fact that the spurious solutions can be removed by

595: %

596: %

597: considering a \emph{combined} integral operator is of great practical

598: importance for numerical calculations \cite{HS00a,Hornberger01}.

599: %

600: %

601: An individual spurious

602: solution of the single or the double layer operator may 

603: %

604: be identified as well after  evaluating the corresponding wave

605: functions by observing in which domain it vanishes. 

606: 

607: \subsubsection{Wave functions}

608: 

609: The eigenfunctions at points off the boundary,

610: $\psi(\rvec_0\notin\Boundary)$, are determined by the null vectors $u$ 

611: corresponding to the roots of the spectral determinant.

612: %

613: %

614: %

615: %

616: %

617: From equation \eref{eq:split} we obtain immediately an integral

618: representation of the (un-normalized) wave function,

619: %

620: %

621: \begin{align}

622:   \label{eq:psiint}

623:   \psi(\rvec_0)

624:   =\,\pm

625:   \bigg[

626:   \int_{\Boundary}

627:   \!

628:   \frac{{\rm d}\Boundary}{b}

629:   \big[

630:   \pm\frac{\lambda}{b} (\dnb\Gnu-\rmi\At_n\Gnu)-\Gnu

631:   \big] u

632:   \bigg]^*

633: \CO

634: \end{align}

635: %

636: for $\rvec_0\notin\Boundary$.

637: According 

638: %

639: to eq. \eref{eq:split} the integral vanishes identically either

640: in the interior or in the exterior. This is indeed 

641: %

642: confirmed by our numerical calculations which are reported  in the

643: next chapter. 

644: %

645: %

646: 

647: In order to calculate the current density \eref{eq:defj} one needs the gauge

648: invariant gradient of the wave function. An integral formula

649: is obtained from equation \eref{eq:split}, after applying the

650: differential operator $\grad_{r_0/b}+\rmi\Avect_0$,

651: %

652: \begin{align}

653:   \label{eq:gradint}

654:   \grad_{r_0/b}\psi(\rvec_0)-\rmi\Avect(\rvec_0)\psi(\rvec_0)

655:   =\,\pm

656:   \bigg[

657:   \int_{\Boundary}

658:   \!

659:   \frac{{\rm d}\Boundary}{b}

660:   \begin{aligned}[t]

661:     \big[&

662:     \pm\frac{\lambda}{b} (\grad_{r_0/b}+\rmi\Avect_0)(\dnb\Gnu-\rmi\At_n\Gnu)

663:     \\

664:     &-(\grad_{r_0/b}\Gnu+\rmi\Avect_0\Gnu)

665:     \big] u 

666:     \bigg]^*  

667:     \PO

668:   \end{aligned}

669:   \nn

670:   \\

671: \end{align}

672: %

673: %

674: The densities of other observables can

675: be obtained by similar boundary integrals.

676: 

677: 

678: 

679: \subsection{The boundary operators}

680: \label{sec:bops}

681: 

682: In the following, we give explicit expressions for the boundary

683: integrals.  This allows to define the ``finite part integral'' appearing

684: in the double layer equation \eref{eq:DoubleLayer}.

685: 

686: %

687: %

688: 

689: The integral operators \eref{eq:Opdef} -- \eref{eq:Opdef4},

690: %

691: \begin{gather}

692:   \big(\mathsf{Q}[u]\big)(\rvec_0) 

693:   = \int_\Boundary\! \rmd\Boundary\, {\rm q}(\rvec;\rvec_0) u(\rvec)

694: \CO

695: \end{gather}

696: are defined by their  integral kernels ${\rm q}(\rvec;\rvec_0)$. 

697: The form of the Green function \eref{eq:GreenPhase} leads to the

698: expressions

699: %

700: \begin{align}

701: \label{eq:QexpslD}

702:   {\rm q}_{\rm sl}^{\rm D}(\rvec;\rvec_0)

703:   =&

704: \,{\rm E}(\rvec;\rvec_0)

705: %

706: %

707: %

708:   \,  \Gn_\nu(z)

709: \\

710: \label{eq:QexpslN}

711:   {\rm q}_{\rm sl}^{\rm N}(\rvec;\rvec_0)

712:   =&

713: \,{\rm E}(\rvec;\rvec_0)

714: %

715: %

716: %

717:   \,

718:   \bigg\{

719:     -\rmi \frac{(\rmrn)\times\nvec}{b^2} 

720:     \, \Gn_\nu(z)

721:     + 2 \frac{(\rmrn)\,\nvec}{(\rmrn)^2} 

722:     \,

723:     z\frac{\rmd}{\rmd z}\Gn_\nu(z)

724:   \bigg\}

725: \\

726: \label{eq:QexpdlD}

727:   {\rm q}_{\rm dl}^{\rm D}(\rvec;\rvec_0)

728:   =&

729: \,{\rm E}(\rvec;\rvec_0)

730: %

731: %

732: %

733:   \,

734:   \bigg\{

735:     -\rmi \frac{(\rmrn)\times\mb{n_0}}{b^2} 

736:     \, \Gn_\nu(z)

737:     - 2 \frac{(\rmrn)\,\nvec_0}{(\rmrn)^2} 

738:     \,

739:     z\frac{\rmd}{\rmd z}\Gn_\nu(z)

740:   \bigg\}

741: \\

742: \label{eq:QexpdlN}

743:   {\rm q}_{\rm dl}^{\rm N}(\rvec;\rvec_0)

744:   =&

745: \,{\rm E}(\rvec;\rvec_0)

746: %

747: %

748: %

749:   \,

750:   \bigg\{

751:     \left(-\frac{((\rmrn)\times\nvec_0)((\rmrn)\times\nvec)}{b^4}

752:       -\rmi \frac{\nvec\times\nvec_0}{b^2}

753:       \right)   \Gn_\nu(z)

754:   \nnn

755:     & \phantom{\,{\rm E}(\rvec;\rvec_0) \,\bigg\{}

756:         +\left(- 2\rmi \frac{\nvec\times\nvec_0}{b^2}

757:         - 2  \frac{\nvec\,\nvec_0}{(\rmrn)^2}       \right)    

758:        \,

759:        z\frac{\rmd}{\rmd z}\Gn_\nu(z)  

760:   \nnn & \phantom{\,{\rm E}(\rvec;\rvec_0) \,\bigg\{}

761:     - 4   \frac{((\rmrn)\nvec)((\rmrn)\nvec_0)}{(\rmrn)^4}

762:        \;       

763:        z^2\frac{\rmd^2}{\rmd z^2}\Gn_\nu(z)  

764:   \bigg\}

765: \CO

766: \intertext{with $\nvec=\nvec(\rvec)$, $\nvec_0=\nvec(\rvec_0)$, 

767: $z\defas(\rmrn)^2/b^2$, and  the abbreviation}

768: {\rm E}(\rvec;\rvec_0)\defas& \exp\left[{-\rmi\left(

769:   \frac{\rvec\times\rvec_0}{b^2}

770:   -\Chit(\rvec)+\Chit(\mb{r_0})\right)}\right]

771: \end{align}

772: %

773: for the gauge dependent part.  Note that the gauge freedom $\Chi$ has

774: cancelled in the pre\-factors and appears in the phase only. It can be

775: absorbed by the substitution $u(\rvec) \to \exp(+\rmi\Chi(\rvec))

776: u(\rvec)$, proving the manifest gauge invariance of the boundary

777: integral equations \eref{eq:SingleLayer}, \eref{eq:DoubleLayer}.  Note

778: also that expressions \eref{eq:QexpslN} and \eref{eq:QexpdlD} are related

779: by a permutation of $\rvec$ and $\rvec_0$ with subsequent complex

780: conjugation (since $\Gn_\nu$ is real), hence the operators are the

781: adjoints of each other. The self-adjoint nature of \eref{eq:QexpslD}

782: and \eref{eq:QexpdlN} follows likewise.

783: 

784: The derivatives appearing in \eref{eq:QexpslN} -- \eref{eq:QexpdlN}

785: may be stated in terms of the gauge independent part of the Green

786: function, $\Gn_\nu$, itself, at different energies $\nu$. 

787: They are given 

788: in Section \ref{sec:Gprop} together with their asymptotic

789: properties. $\Gn_\nu$ displays a logarithmic singularity as

790: $\rvec\to\rvec_0$, while the differential expressions are bounded.

791: %

792: In that limit, most of the quotients vanish for a smooth boundary,

793: others tend to the curvature \eref{eq:parametr3}

794: %

795: at the boundary point $\rvec_0$.

796: %

797: %

798: %

799: %

800: %

801: %

802: %

803: %

804: %

805: %

806: As a consequence, all the terms in \eref{eq:QexpslD} --

807: \eref{eq:QexpdlN} are integrable --- but for the one containing the

808: $(\nvec\,\nvec_0)/(\rmrn)^2$-singularity.  The latter gives rise to

809: the need for a finite part integral.

810: 

811: \subsubsection*{The hypersingular integral operator}

812: %

813: 

814: For finite $\lambda$ the double-layer equation contains a

815: hypersingular integral defined as

816: \begin{align}

817: \label{eq:fpi}

818:   \mathsf{Q}^{\rm N}_{\rm dl} [u] 

819:   =&\,\, 

820:   \fpint_{\Boundary}

821:    \frac{{\rm d}\Boundary}{b^2}

822:    ({\partial}_{n_0/b} +\rmi \At_{n_0})

823:    ({\partial}_{n/b}\Gnu -\rmi \At_{n}\Gnu)

824:    u 

825: \nnn

826:   \defas& 

827:   \lim_{\epsilon\to0}

828:   \int_{\Boundary}

829:    \frac{{\rm d}\Boundary}{b^2}

830:    ({\partial}_{n_0/b} +\rmi \At_{n_0})

831:    ({\partial}_{n/b}\epscomb{\Gnu} -\rmi \At_{n}\epscomb{\Gnu})

832:     u\PO

833: \end{align}

834: %

835: We want to replace the integrand by its limiting form.  To this end

836: the boundary is split into the part $\gamma_{c\epsilon}$, which lies

837: within a $(c\epsilon)$-vicinity around $\rvec_0$ (with arbitrary

838: constant $c$), and the remaining part $\Boundary_{\!c\epsilon}$,

839: \begin{align}

840:   \label{eq:fpieps}

841:  =  \lim_{\epsilon\to0} \!

842:   \bigg[&

843:   \int_{\Boundary_{\!\scriptstyle c\epsilon}}

844:     \!\! \frac{{\rm d}\Boundary}{b^2}

845:     (\partial_{n_0/b} +\rmi \At_{n_0})

846:     ({\partial}_{n/b}\epscomb{\Gnu} -\rmi \At_{n}\epscomb{\Gnu})

847:      u

848: \nnn

849:    +&  \int_{\gamma_{\scriptstyle c\epsilon}}

850:     \!\! \frac{{\rm d}\Boundary}{b^2}

851:     (\partial_{n_0/b} +\rmi \At_{n_0})

852:     ({\partial}_{n/b}\epscomb{\Gnu} -\rmi \At_{n}\epscomb{\Gnu})

853:      (u-u_0)

854: \\

855:     +&

856:     u_0

857:   \int_{\gamma_{\scriptstyle c\epsilon}}

858:     \!\! \frac{{\rm d}\Boundary}{b^2}

859:     (\partial_{n_0/b} +\rmi \At_{n_0})

860:     ({\partial}_{n/b}\epscomb{\Gnu} -\rmi \At_{n}\epscomb{\Gnu})

861:   \bigg]

862: \CO

863: \nn

864: \end{align}

865: with $u_0 \defas u(\rvec_0)$.  For sufficiently small $\epsilon$ the

866: boundary piece $\gamma_{c\epsilon}$ may be replaced by its

867: tangent\footnote{We emphasize  that we assume the boundary to be

868:   smooth throughout.} and the Green function by its asymptotic

869: expression, cf Sect. \ref{sec:Gprop}.  This way the third integral in

870: \eref{eq:fpieps} may be evaluated to its contributing order,

871: \begin{align}

872:   \label{eq:fpix3}

873:   &\int_{\gamma_{\scriptstyle c\epsilon}}

874:     \!\! \frac{{\rm d}\Boundary}{b^2}

875:     (\partial_{n_0/b} +\rmi \At_{n_0})

876:     ({\partial}_{n/b}\epscomb{\Gnu} -\rmi \At_{n}\epscomb{\Gnu})

877: \nnn

878:   =\,&

879:   \frac{1}{4\pi}

880:   \int_{-c\epsilon}^{c\epsilon}

881:     \!\!\!    \!\!\!

882:     \cos\! \Big(\frac{\rvec_0\nvec_0}{b^2}s\Big)

883:     \cos\!\left[\epsilon\left(\frac{\nvec_0\times\rvec_0}{b^2}-s\right)\right]

884:     \bigg(

885:       \frac{-2}{s^2+\epsilon^2}

886:       +4\frac{\epsilon^2}{(s^2+\epsilon^2)^2}

887:     \bigg)

888:     \,    \rmd s

889:     \nnn

890:   &+ \Or(\epsilon^2\log\epsilon)  

891: \nnn

892:  =\,&

893:   \frac{1}{2\pi}

894:   \int_{-c\epsilon}^{c\epsilon}

895:     \!\!\!    \!\!\!

896:     \rmd s

897:       \frac{\epsilon^2-s^2}{(s^2+\epsilon^2)^2}

898:     \,\,

899:   + \Or(\epsilon^2\log\epsilon)  

900:  =

901:   \frac{1}{\pi}   \frac{1}{c\epsilon}

902:   \frac{c^2}{c^2+1}

903:   + \Or(\epsilon^2\log\epsilon)

904: \nnn

905:  \approx&\,

906:   \frac{1}{\pi}   \frac{1}{c\epsilon}

907:   + \Or(\epsilon^2\log\epsilon)\PO  

908: \end{align}

909: %

910: Here, the explicit form of the integrand was obtained from

911: \eref{eq:QexpdlN} by the replacement $\rvec_0\to\rvec_0^\pm$.

912: The last approximation in \eref{eq:fpix3} holds because $c$ may be

913: chosen arbitrarily large.  In a similar fashion it can be shown that

914: the second integral in \eref{eq:fpieps} is of order

915: $\Or(\epsilon)$. In the first integral we may replace (again for large

916: $c$) the integrand by its limit, because $\epsilon$ is small compared

917: to $\min(|\rvec-\rvec_0|)=c\;\epsilon$.  Therefore, the limit in

918: \eref{eq:fpi} may be expressed as

919: \begin{multline}

920: \label{eq:fpilim}

921:   \fpint_{\Boundary}

922:    \frac{{\rm d}\Boundary}{b^2}

923:    ({\partial}_{n_0/b} +\rmi \At_{n_0})

924:    ({\partial}_{n/b} -\rmi \At_{n})

925:    \Gnu u 

926:    \\

927:   =  

928:  \lim_{\epsilon\to0}

929:  \bigg[

930:   \int_{\Boundary_{\!\scriptstyle \epsilon}}

931:     \!\! \frac{{\rm d}\Boundary}{b^2}

932:     (\partial_{n_0/b} +\rmi \At_{n_0})

933:     ({\partial}_{n/b}\Gnu -\rmi \At_{n}\Gnu)

934:      u

935:    \,\,

936:    + u_0 \frac{1}{\pi\epsilon}

937:  \bigg]\CO

938: \end{multline}

939: where we replaced $c\epsilon$ by $\epsilon$. This equation

940: defines the finite part integral.  It completes the derivation

941: of the boundary integral equations.

942: %

943: %

944: 

945: 

946: 

947: 

948: 

949: 

950: 

951: \subsection{Solving the integral equations}

952: %

953: 

954: As discussed above, the integral equations \eref{eq:SingleLayer} and

955: \eref{eq:DoubleLayer} 

956: determine the spectra and wave functions of arbitrary interior and

957: exterior magnetic billiards.  

958: In the stated form the equations are not yet suitable for numerical

959: evaluation, though, since the integral kernels display (integrable)

960: singularities.

961: 

962: %

963: Fortunately, 

964: it is possible to treat the singular behavior

965: analytically  which renders a highly accurate and efficient numerical

966: scheme.

967: %

968: %

969: %

970: %

971: %

972: %

973: %

974: %

975: %

976: %

977: %

978: %

979: %

980: %

981: %

982: In brief, the boundary integral equations are regularized using the

983: known asymptotic behavior of the Green function and its

984: derivatives, cf Sect.~\ref{sec:Gprop}.

985: %

986: %

987: Representing the periodic boundary functions in a Fourier basis then

988: leads to an exponential localization of the integral kernels.  This

989: permits a well controlled truncation of the corresponding matrix. The

990: roots of the (Fredholm) determinant are accurately obtained by

991: singular value decomposition.

992: %

993: We refer the reader to our recent publication \cite{HS00a} for the

994: technical details 

995: %

996: and a convergence analysis.\footnote{

997: Note that the equations in \cite{HS00a} are stated in complex

998: conjugated form since the focus is there on the wave functions rather

999: than the Green function.}

1000: 

1001: