1: 

2: \documentclass[12pt]{article}

3: \usepackage{epsfig,amsfonts,amssymb}

4: \usepackage{cite}

5: \input epsf.sty

6: \topmargin -.5cm

7: \textheight 21cm

8: \oddsidemargin -.125cm

9: \textwidth 16cm

10: 

11: 

12: \def\ZZZ{{\hbox{ Z\kern-1.6mm Z}}}

13: \def\zzz{{\hbox{z\kern-1mm z}}}

14: 

15: \newcommand{\vt}{\vartheta}

16: 

17: \newcommand{\vtau} {\vec \tau}

18: \newcommand{\vj} {\vec J}

19: \newcommand{\vxi} {\vec \xi}

20: \newcommand{\vu} {\vec u}

21: \newcommand{\htau} {\vec \eta}

22: \newcommand{\vc}{\vec\chi}

23: \newcommand{\vpsi} {\vec \psi}

24: 

25: 

26: 

27: 

28: 

29: \newcommand{\rrho}{r}

30: \newcommand{\bA}{{\bf A}}

31: \newcommand{\tx}{\wt x}

32: \newcommand{\bG}{{\bf G}}

33: \newcommand{\bF}{{\bar F}}

34: \newcommand{\bbb}{{\bar b}}

35: \newcommand{\gam}{\tau}

36: \newcommand{\eps}{\epsilon}

37: \newcommand{\ra}{\rangle}

38: \newcommand{\la}{\langle}

39: \newcommand{\T}{\chi_{T}(k)}

40: \newcommand{\Tm}{\chi_{T}(k')}

41: \newcommand{\Cn}{{\cal C}_n}

42: \newcommand{\vp}{\varphi}

43: \newcommand{\ve}{\varepsilon}

44: \newcommand{\tl}{\lambda}

45: \newcommand{\dt}{(\vec \nabla T)^2}

46: \newcommand{\hp}{{\wh\Phi}}

47: \newcommand{\hq}{{\wh Q_B}}

48: \newcommand{\he}{{\wh\eta_0}}

49: \newcommand{\ha}{{\wh{A}}}

50: \newcommand{\lllb}{\Bigl\langle\Bigl\langle}

51: \newcommand{\rrrb}{\Bigr\rangle\Bigr\rangle}

52: \newcommand{\tf}{\wt f}

53: \newcommand{\sss}{{\cal L}_{av}}

54: \newcommand{\bx}{\bar x}

55: \newcommand{\bw}{\bar w}

56: \newcommand{\ws}{{\wt\sigma}}

57: \newcommand{\wrh}{{\wt\rho}}

58: \newcommand{\wv}{{\wt v}}

59: 

60: 

61: 

62: 

63: 

64: \newcommand{\vv} {\bar v}

65: \newcommand{\uu} {\bar u}

66: \newcommand{\K}{{\rm K_1}}

67: \newcommand{\Kt}{{\rm \widetilde K_1}}

68: 

69: \newcommand{\B}{b'}

70: \newcommand{\C}{c'}

71: \newcommand{\bB}{\bar b'}

72: \newcommand{\Bu}{B_{\vec u}}

73: \newcommand{\VV}{{\cal V}}

74: \newcommand{\BB}{{\cal B}}

75: \newcommand{\II}{{\cal I}}

76: \newcommand{\AAA}{{\cal A}}

77: \newcommand{\GG}{{\cal G}}

78: \newcommand{\KK}{{\cal K}}

79: \newcommand{\fff}{{\bf f}}

80: \newcommand{\ccc}{{\bf c}}

81: \newcommand{\FF}{{\cal F}}

82: \newcommand{\JJ}{{\cal J}}

83: \newcommand{\HH}{{\cal H}}

84: \newcommand{\MM}{{\cal M}}

85: \newcommand{\CC}{{\cal C}}

86: \newcommand{\bC}{{\bf C}}

87: \newcommand{\OO}{{\cal O}}

88: \newcommand{\QQ}{{\cal Q}}

89: \newcommand{\PP}{{\cal P}}

90: \newcommand{\EE}{{\cal E}}

91: \newcommand{\LL}{{\cal L}}

92:  \newcommand{\rrr}{\rangle\rangle}

93: \newcommand{\half}{{1\over 2}}

94: \newcommand{\wt}{\widetilde}

95: \newcommand{\wh}{\widehat}

96: \newcommand{\wc}{\wt}

97: \newcommand{\wb}{\bar}

98: \newcommand{\bd}{\bar{\rm D}}

99: \newcommand{\RR}{{\cal R}}

100: \newcommand{\NN}{{\cal N}}

101: \newcommand{\TT}{{\cal T}}

102: \newcommand{\bg}{\bar g}

103: \newcommand{\bb}{\bar b}

104: \newcommand{\bT}{\bar \Theta}

105: \newcommand{\SSS}{{\cal S}}

106: \newcommand{\tlx}{\left(\tilde \lambda ; X^0(0) \right)}

107: \newcommand{\al}{\alpha}

108: 

109: \newcommand{\omk}{\omega_n(\vec k)}

110: \newcommand{\onk}{\omega^{(N)}_{\vec k_\perp}}

111: \newcommand{\tI}{\wt\II}

112: \newcommand{\hI}{\wh\II}

113: \newcommand{\nI}{\II}

114: \newcommand{\cp}{\wh\Phi}

115: 

116: 

117: \newcommand{\be}{\begin{equation}}

118: \newcommand{\ee}{\end{equation}}

119: \newcommand{\ben}{\begin{eqnarray}\displaystyle}

120: \newcommand{\een}{\end{eqnarray}}

121: 

122: \newcommand{\bea}[1]{\begin{eqnarray}\label{#1} }

123: \newcommand{\eea}{\end{eqnarray}}

124: 

125: \newcommand{\refb}[1]{(\ref{#1})}

126: \newcommand{\p}{\partial}

127: \newcommand{\sectiono}[1]{\section{#1}\setcounter{equation}{0}}

128: \newcommand{\subsectiono}[1]{\subsection{#1}\setcounter{equation}{0}}

129: 

130: \def\one{{\hbox{ 1\kern-.8mm l}}}

131: \def\zero{{\hbox{ 0\kern-1.5mm 0}}}

132: 

133: \renewcommand{\theequation}{\thesection.\arabic{equation}}

134: 

135: \begin{document}

136: 

137: {}~

138: {}~

139: \hfill\vbox{\hbox{hep-th/yymmnnn}

140: }\break

141: 

142: \vskip .6cm

143: 

144: {\baselineskip20pt

145: \begin{center}

146: {\Large \bf Walls of Marginal Stability and 

147: Dyon Spectrum in $\NN=4$ Supersymmetric String Theories

148: } 

149: 

150: \end{center} }

151: 

152: \vskip .6cm

153: \medskip

154: 

155: \vspace*{4.0ex}

156: 

157: \centerline{\large \rm

158: Ashoke Sen}

159: 

160: \vspace*{4.0ex}

161: 

162: 

163: \centerline{\large \it Harish-Chandra Research Institute}

164: 

165: 

166: \centerline{\large \it  Chhatnag Road, Jhusi,

167: Allahabad 211019, INDIA}

168: 

169: \vspace*{1.0ex}

170: 

171: \centerline{\it and}

172: 

173: \vspace*{1.0ex}

174: 

175: \centerline{\large \it Department of Physics, California Institute

176: of Technology}

177: 

178: \centerline{\large \it Pasadena, CA91125, USA}

179: 

180: \vspace*{1.0ex}

181: 

182: \centerline{E-mail:

183: sen@mri.ernet.in, ashokesen1999@gmail.com}

184: 

185: \vspace*{5.0ex}

186: 

187: \centerline{\bf Abstract} \bigskip

188: 

189: The spectrum of quarter BPS dyons in 

190: $\NN=4$ supersymmetric string theories can change as the

191: asymptotic moduli cross

192: walls of marginal stability on which

193: the dyon can break apart into a pair of half BPS states. 

194: In this paper we

195: classify these marginal stability walls and

196: examine this phenomenon in the context of exact dyon spectrum

197: found in a class of $\NN=4$ supersymmetric string theories.

198: We argue that the dyon partition functions

199: in different domains separated by marginal stability walls are the

200: same, but the choice of integration contour needed

201: for extracting the degeneracies from the partition function differ

202: in these different regions. We also find that in the limit of large

203: charges the change in the degeneracy 

204: is exponentially suppressed compared to the

205: leading contribution. This is consistent with the fact that in the

206: computation of black hole entropy we do not encounter any change

207: as the asymptotic moduli fields move across the walls of marginal

208: stability. Finally we carry out some tests of S-duality invariance

209: in the theory.

210: 

211: \vfill \eject

212: 

213: \baselineskip=18pt

214: 

215: \tableofcontents

216: 

217: \sectiono{Introduction and Summary} \label{sintro}

218: 

219: In a series of papers 

220: we computed the exact degeneracy of quarter

221: BPS dyons in a

222: class of $\NN=4$ supersymmetric string compactifications in

223: an appropriate corner of the moduli space of these 

224: theories\cite{0605210,0607155,0609109}

225: verifying and generalizing

226: earlier conjectures\cite{9607026,0510147,0602254}.

227: Alternative approaches to this problem leading to similar results

228: have also been 

229: developed\cite{9607026,0505094,0506249,0612011,0603066}. 

230: The result for the degeneracy takes the form of integration over a

231: three real

232: dimensional subspace (a contour)

233: of the Siegel upper half plane parametrizing

234: genus two Riemann surfaces, 

235: and the integrand involves inverse of a certain meromorphic

236: modular form of a subgroup of the Siegel modular group. 

237: 

238: It is well known however that for a dyon with a given set of

239: charges the moduli space of $\NN=4$ supersymmetric string theory

240: contains subspaces on which the original dyon becomes 

241: marginally unstable

242: against decay into a pair of other dyons\cite{9712211,9804160}.

243: In particular the moduli space contains

244: walls of marginal stability  -- 

245: codimension one subspaces --

246: on which the  mass of the quarter BPS dyon becomes equal

247: to the sum of masses of a pair of half BPS dyons whose charges

248: add up to that of the original quarter BPS dyon. On this subspace

249: the original dyon becomes marginally unstable against decay into

250: this pair of half BPS dyons, and typically the spectrum of the original

251: dyon changes discontinuously as we move through these marginal

252: stability walls in the moduli space\cite{9712211,9804160}.

253: Thus an important question is: how does the dyon spectrum

254: computed in \cite{0605210,0607155,0609109} change as we

255: move away from the particular corner of the moduli space

256: in which the degeneracy was computed?

257: 

258: A glimpse of this issue was already seen in the analysis of

259: \cite{0605210,0607155,0609109} where it was found that even

260: in the corner of the moduli space where the result was computed,

261: -- in a weakly coupled type IIB string theory compactified on

262: a certain orbifold, -- the result for the degeneracy changes

263: discontinuously as the angle between certain pair of

264: circles of the compact manifold

265: passes through zero. 

266: This change could be attributed to the fact that precisely at this

267: point the system under consideration became marginally stable

268: against decay into a pair of half BPS states.

269: However the change was such that

270: the expression for the degeneracy continues to be given by a

271: similar integral with identical integrand, but the contour over which

272: the integral is to be performed gets changed. If we try to 

273: deform the new contour into the original contour we encounter

274: a pole of the integrand and hence the two contributions differ by

275: the residue at the pole. One can also interpret the  result by saying

276: that the dyon partition function formally remains the same as we

277: move through the marginal stability wall, but the point in the

278: Siegel moduli space around which we should series expand the 

279: partition function to extract the degeneracies changes as we move

280: through these walls.

281: 

282: In this paper we classify these marginal stability walls in the

283: moduli space  

284: of the theory and explore what happens when we move across these

285: walls.

286: Although the walls are complicated codimension one surfaces

287: in the moduli space, one gets a simpler picture by

288: regarding them as curves in the

289: axion-dilaton moduli space for fixed values of the other moduli.

290: We find that these curves are circles and straight lines, and 

291: could intersect on the real axis or at $i\infty$, 

292: but have no intersection in the

293: interior of the upper half plane. 

294: Furthermore although the slopes of the lines and the radii and the

295: centres of the circles depend on the charges and other moduli,

296: the points where they intersect the real axis and each other are

297: universal.

298: As a result a region bounded

299: by these curves has universal vertices but boundaries which depend

300: on the other moduli and charges.  One such region has been

301: displayed in Fig.\ref{f1} in \S\ref{smarginal}.

302: As we move from one of these regions to another, we cross marginal

303: stability walls and as a result the dyon spectrum could change.

304: 

305: Typically most of these walls lie outside the domain in which the

306: approximation made in our 

307: computation of the degeneracy can be trusted. However

308: using T- and S-duality invariance of the theory we can extract

309: useful information about these walls. A useful input in reaching this conclusion

310: is the observation made in \cite{appear} that under an S-duality

311: transformation the degeneracy formula does not remain invariant, but

312: there is a change in the integration contour. This can be attributed to

313: the fact that a duality transformation acts both on the charges as well

314: as the asymptotic values of the moduli fields, whereas the degeneracy

315: formula is computed for different charges but in the same region

316: of the moduli space, in a sense that will be made precise in

317: \S\ref{sdualtwo}. The apparent lack of duality symmetry of the 

318: degeneracy formula is due to the fact that the region in which the

319: degeneracy formula is calculated and the region obtained after a

320: duality transformation are typically separated by walls of marginal

321: stability. Thus the knowledge of how the contour changes under a

322: duality transformation can be used to extract information about

323: how it changes from one region of the moduli space to another as

324: we move through these marginal stability walls.

325: This way the change in the

326: spectrum across any

327: wall of marginal stability can be encoded as the result of changing the

328: integration contour leaving the integrand unchanged, or equivalently

329: as shifting the point around which we expand the partition function

330: to extract the degeneracy. 

331: 

332: One can try to find a physical interpretation of these changes in the

333: degeneracy by explicitly evaluating the residues at the poles picked

334: up by the contour as we move it from the position associated

335: with one side of a wall of marginal stability to that on the other side

336: of the wall. We can do this easily in the context of the marginal

337: stability wall found in the original analysis of \cite{0605210}.

338: We find that the change in the degeneracy is proportional to the

339: total degeneracy of the half-BPS

340: electric and magnetic states into which the

341: original dyon can decay on this particular marginal stability wall.

342: 

343: Given these results, it is natural to ask what 

344: happens in the large charge

345: limit where the statistical entropy given by the logarithm of

346: the degeneracy can be compared with the black hole entropy. We find

347: that in this limit the change in the degeneracy as we move across a

348: wall of marginal stability, -- encoded in the residues at the poles which

349: we encounter while deforming the new contour to the old one, --

350: is exponentially suppressed relative to the

351: leading contribution to the degeneracy. As a result the change in the

352: statistical entropy as we move across the wall of marginal stability

353: is exponentially suppressed. This is consistent with the fact that the

354: entropy of the corresponding black hole is controlled solely by the

355: charges and is independent of the asymptotic moduli due to

356: attractor mechanism. Thus as the asymptotic values of the

357: moduli fields move across the wall of marginal stability the

358: answer for the black hole entropy does not change. Our results

359: indicate however that if we are able to incorporate non-perturbative

360: (in inverse charges) 

361: corrections in the computation of the 

362: black hole entropy then we should see 

363: a dependence of the entropy on the asymptotic

364: values of the moduli fields, possibly along the line of

365: \cite{0010222,0101135}.

366: 

367: Since we use duality invariance to find how the contour should be

368: deformed as we move across various walls of marginal stability,

369: one could ask if there is any non-trivial test of duality that one

370: could perform. If we

371: could identify duality transformations which leave a region

372: invariant, -- in a sense that will be made precise in

373: \S\ref{sdualtwo}, --

374: then the expression for the degeneracy should not change

375: under such a duality transformation. This requires that

376:  under such a duality transformation either the

377: contour should remain unchanged or it should move to a new

378: position such that in deforming it from the new to the old

379: position we do not encounter a pole in the integrand. 

380: This can then be explicitly tested.

381: There is also

382: a possibility that a duality transformation 

383: maps one region to another such that the degeneracy

384: in each of these regions can be computed directly. In this case 

385: we have an independent result of how the contour should change

386: as we move from the first region to the second and this can then be

387: compared with the predictions coming from duality. 

388: We identify some duality transformations

389: of these types and carry out the required consistency checks.

390: 

391: The rest of the paper is organized as follows. In \S\ref{sreview}

392: we review the results of

393: \cite{0605210,0607155,0609109}

394: about the degeneracy of quarter BPS

395: dyons in a class of $\NN=4$ supersymmetric string theories.

396: In \S\ref{smarginal} we determine the locations of the

397: walls of marginal stability

398: in $\NN=4$ supersymmetric string theories. In \S\ref{sdualtwo}

399: we determine how the different domains of the moduli space,

400: bounded by the marginal stability walls, are mapped to each other

401: under T- and S-duality symmetries of the theory and

402: use this information to determine how the spectrum should change as

403: we pass through a particular marginal stability wall. In \S\ref{sblack}

404: we show that the change in the statistical entropy as we

405: move across the marginal stability walls is non-leading compared to

406: the full entropy. In \S\ref{stest} we perform some tests of

407: S-duality invariance of the theory.

408: 

409: Finally we would like to remind the reader 

410: that our analysis will focus on the

411: marginal stability walls associated with decay of the dyon into a

412: pair of half-BPS states. It will be worth exploring if there are

413: interesting phenomena associated with decay into a pair of

414: quarter BPS states.

415: 

416: Some related issues have been discussed in \cite{appear}.

417: 

418: \sectiono{Dyon Spectrum

419: in a Class of $\NN=4$ Supersymmetric Models} \label{sreview}

420: 

421: In this section we shall review the dyon spectrum in a

422: class of $\NN=4$ supersymmetric

423: models analyzed in \cite{0510147,0602254,0605210,

424: 0607155,0609109}. Somewhat different approaches leading to

425: similar results have been developed in 

426: \cite{9607026,0505094,0506249,0603066,0612011}.

427: 

428: These theories are constructed by taking a $\ZZZ_N$ orbifold of

429: type IIB string theory on $\MM\times S^1\times \wt S^1$ where

430: $\MM$ is either K3 or $T^4$. 

431: The generator $g$ of the $\ZZZ_N$ group involves

432: $1/N$ unit of shift along the circle $S^1$ 

433: together with an order

434: $N$ transformation

435: $\wt g$ in $\MM$. $\wt g$ is chosen so that it commutes with

436: an $\NN=4$ supersymmetry algebra of the parent theory.

437: Thus the final

438: theory has  $\NN=4$ supersymmetry. 

439: 

440: The description of the theory given above will be referred to as

441: the first

442: description of the theory. Another useful description is

443: obtained by a series of duality transformations. We first

444: make an S-duality transformation in the type IIB theory.  Next

445: we make 

446: an $R\to 1/R$ duality on the circle $\wt S^1$ that takes type IIB

447: string theory on $\MM\times S^1\times\wt S^1$ to type IIA  string

448: $\MM\times S^1\times\wh S^1$ where $\wh S^1$ is

449: the circle dual to $\wt S^1$. Finally  using 

450: six dimensional string-string

451: duality we relate this to a heterotic string theory on $T^4\times S^1

452: \times \wh S^1$ for 

453: $\MM=K3$

454: and type IIA string theory on

455: $T^4\times S^1

456: \times \wh S^1$ for $\MM=T^4$. 

457: Under this duality the transformation

458: $\wt g$ gets mapped to a transformation $\wh g$ that acts only as

459: a shift on the right-moving degrees of freedom on the world-sheet

460: and as a shift plus rotation on the left-moving degrees of freedom.

461: In the final theory, obtained by taking the orbifold of heterotic or

462: type IIA

463: string theory on $T^4\times S^1\times \wh S^1$ by a $1/N$

464: unit of shift along $S^1$ together with the transformation $\wh g$, all

465: the space-time supersymmetries come from the right-moving sector

466: of the world-sheet. We shall call this the second description of the

467: theory. 

468: 

469: These theories typically have several moduli fields.

470: In the second description one complex modulus scalar 

471: arises

472: from the axion ($a$) - dilaton ($\phi$) 

473: combination, where the axion by definition

474: is the scalar field obtained by dualizing the NSNS sector

475: 2-form field. We shall denote by $\tau$ the combination

476: $a + i S$ where $S=e^{-2\phi}$. Other

477: moduli fields arise from the $\wh g$ invariant

478: components of the metric, antisymmetric tensor field and gauge

479: fields (in case of heterotic string theory)

480: along the six dimensional internal torus and 

481: may be encoded in an $r\times r$ matrix valued field $M$ satisfying

482: \be \label{eag1}

483: MLM^T=M, \quad M^T=M \, ,

484: \ee

485: where $L$ is a matrix with 6 eigenvalues +1 and 

486: $(r-6)$ eigenvalues $-1$. Here $r$ is the rank of the gauge group

487: and depends on the specific model being considered.

488: These U(1)

489: gauge fields  arise from the $\wh g$ invariant ten dimensional gauge fields 

490: (in case of heterotic string theory) as well as $\wh g$ invariant

491: components of the

492: metric and the 2-form field with one leg along the

493: internal   torus and one leg along the non-compact directions.

494: 

495: Following the chain of dualities relating the first description to the

496: second description one can work out the origin of the various

497: fields in the first description. In particular the modulus $\tau=a+iS$

498: can be shown to correspond to the complex structure modulus of

499: the torus spanned by the $S^1$ and $\wt S^1$ directions.

500: 

501: We shall now give the precise relationship between 

502: some of the moduli fields and the

503: geometric quantities associated with the compactification in

504: the second description and also

505: give precise expression for some of the charges

506: in terms of physical quantum numbers

507: carried by a state. To do this we need to fix a normalization convention

508: for the coordinates along $S^1$ and $\wh S^1$. We choose 

509: $x^4$ and $x^5$ to be the coordinates along $\wh S^1$

510: and $S^1$ respectively and choose their periods before

511: orbifolding to be $2\pi\sqrt{\alpha'}$ 

512: and $2\pi N\sqrt{\alpha'}$ respectively. Thus after

513: orbifolding both can be regarded as having period 

514: $2\pi\sqrt{\alpha'}$; however

515: since periodicity under a $2\pi\sqrt{\alpha'}$ translation 

516: action $S^1$ also

517: involves an action of the generator $\wh g$ of the internal $\ZZZ_N$

518: symmetry group, the momentum along $S^1$, measured in units

519: of $1/\sqrt{\alpha'}$, could be 

520: fractional,

521: -- some multiple of $1/N$, -- if

522: $\wh g$ acts non-trivially on the state. 

523: A string will be said to carry one 

524: unit of winding

525: along $S^1$ (or $\wt S^1$) if, as we go once

526: around the string, the $x^4$ ($x^5$) coordinate 

527: shifts by $2\pi\sqrt{\alpha'}$.

528: Thus an untwisted sector state whose coordinate along $S^1$

529: changes by multiples of $2\pi N\sqrt{\alpha'}$ 

530: will carry winding charge along

531: $S^1$ in

532: multiples of $N$, but twisted sector states can carry generic integer

533: winding charges.

534: A single H-monopole associated with $S^1$

535: will correspond to an

536: array of NS 5-branes wrapped on $\wh S^1\times T^4$

537: and placed at

538: intervals of $2\pi\sqrt{\alpha'}$ along $S^1$. Finally the original 

539: Kaluza-Klein monopole associated  with $S^1$,

540: represented by a

541: Taub-NUT

542: space with an asymptotic circle of radius $2\pi N\sqrt{\alpha'}$ 

543: along $x^4$,

544: will develop a $\ZZZ_N$ singularity at its centre after the

545: orbifolding and has to be 

546: regarded as carrying $N$ units of  Kaluza-Klein

547: monopole charge associated with $S^1$. 

548: Thus the Kaluza-Klein monopole charge associated with $S^1$

549: will be quantized

550: in units of $N$. Similar definition can be given for the $H$ and

551: Kaluza-Klein monopole charges associated with $\wh S^1$, but

552: in this case the normalization is straightforward and both the

553: charges are allowed to take arbitrary integer values. 

554: 

555: 

556:  

557: Let  $x^\mu$ ($0\le

558: \mu\le 3$) denote the coordinates along

559: the non-compact coordinates.

560: For our analysis it will be 

561: useful to study in detail a subsector

562: of the theory in which we include only those gauge fields 

563: which are associated with the $4\mu$ and $5\mu$

564: components of the metric and the

565: anti-symmetric tensor field, 

566: only those components of $M$ which encode information

567: about the

568: components of the metric and the

569: anti-symmetric tensor field

570: along $S^1\times \wh S^1$, the axion-dilaton field,

571: and the four dimensional metric. In this subsector there are

572: four gauge fields $A^{(i)}_\mu$ ($1\le i\le 4$) and 

573: a $4\times 4$ matrix valued field $M$ satisfying

574: \be\label{enewm}

575: M^T = M, \quad M L M^T = L, \quad L\equiv\pmatrix{0 & I_2\cr

576: I_2 & 0}\, .

577: \ee

578: The fields $A^{(i)}_\mu$ and $M$

579: are related to the ten dimensional string metric $G_{MN}$ and

580: 2-form field $B_{MN}$ via the relations\cite{9207016,9402002}:

581: \ben \label{etenfour}

582: && \wh G_{mn} \equiv G^{(10)}_{mn}, \quad 

583: \wh B_{mn} \equiv B^{(10)}_{mn}\, ,  

584: \qquad 

585: m,n=4,5\, , 

586: \nonumber  \\ 

587: && M =  

588: \pmatrix{ \wh G^{-1} & \wh G^{-1} B \cr -\wh B \wh G^{-1} & \wh 

589: G - \wh B \wh G^{-1} \wh B} \nonumber \\

590: && A^{(m-3)}_\mu = {1\over 2} (\wh G^{-1})^{mn} 

591: G^{(10)}_{m\mu} , \quad

592: A^{(m-1)}_\mu = {1\over 2} B^{(10)}_{m\mu} - 

593: \wh B_{mn} A^{(m-3)}_\mu, \nonumber \\

594: && \qquad  4\le m,n\le 5, \quad 

595: 0\le \mu, \nu \le 3 \, . 

596: \een

597: 

598: A general

599: dyonic state in the theory is characterized by an $r$ dimensional

600: electric charge

601: vector $\vec Q$ and an $r$ dimensional 

602: magnetic charge vector $\vec P$. 

603: However if we consider a dyon that is charged only under the

604: gauge fields $A_\mu^{(i)}$ ($1\le i\le 4$) introduced in

605: \refb{etenfour}, the corresponding charge vectors can be taken to be

606: four dimensional. 

607: If  we consider a state with 

608: momentum $\wh n$ and

609: winding $-\wh w$ along $\wh S^1$, 

610: momentum $n'$ and

611: winding $-w'$ along $S^1$,

612: Kaluza-Klein monopole charge  $\wh N$ 

613: and 

614: H-monopole

615: charge  $-\wh W$ associated with 

616: $\wh S^1$ and 

617: Kaluza-Klein monopole charge $N'$ and

618: H-monopole

619: charge $-W'$ associated with 

620: $S^1$, then we define the four dimensional electric

621: charge vector $\vec Q$ and magnetic charge vector $\vec P$

622: characterizing the state to be

623: \be\label{e2dcharge}

624: Q=\pmatrix{\wh n\cr n'\cr \wh w\cr w'}, \qquad 

625: P = \pmatrix{\wh W\cr W'\cr

626: \wh N\cr N'}\, .

627: \ee

628: {}From our earlier discussion it follows that $n'$ is quantized in

629: units of $1/N$, $N'$ is quantized in units of $N$ and all other

630: quantum numbers appearing in \refb{e2dcharge} are quantized

631: in integer units. The precise relation between $Q$, $P$ and the

632: electric and magnetic charges associated with the gauge fields

633: $A_\mu^{(i)}$ has been derived 

634:  in \cite{0508042} by working

635: in the

636: $\alpha'=16$ unit.

637: 

638: 

639: Restricted to this subspace the 

640: T-duality transformation of the theory is parametrized by

641: a $4\times 4$ matrix $\Omega$ preserving the charge lattice

642: and satisfying

643: \be \label{eag9}

644: \Omega^T L \Omega=L\, .

645: \ee 

646: The action of

647: T-duality takes us from a charge vector $(\vec Q, \vec P)$ to another

648: charge vector $(\vec Q',\vec P')$ given by

649: \be \label{sadd1}

650: \vec P'= (\Omega^T)^{-1}  \, \vec P, \qquad \vec Q'=(\Omega^T)^{-1} 

651: \, \vec Q\, ,

652: \ee

653: and the moduli field $M$ to $M'$ given by

654: \be \label{emtrs}

655: M'=\Omega M \Omega^T \, ,

656: \ee

657: leaving $\tau$ unchanged.

658: {}From eqs.\refb{eag9}, \refb{sadd1} it follows that the following

659: inner products are invariant under T-duality transformation

660: \ben\label{echsq}

661: && Q^2 \equiv Q^T L Q

662: = 2(\wh n \wh w + n'w'), \qquad P^2 \equiv P^TLP

663: = 2(\wh N \wh W

664: +N'W'), \nonumber \\ &&

665: P\cdot Q \equiv P^TLQ

666: = \wh N \wh n + \wh W \wh w+ N' n' + W' w'\, .

667: \een

668: These combinations are independent of the moduli fields $M$.

669: Using the moduli field we can construct more general T-duality

670: invariant combinations like $Q^TMQ$, $P^TMP$ and $Q^TMP$.

671: We shall make use of these quantities in \S\ref{smarginal}.

672: 

673: In the full theory $\vec Q$ and $\vec P$ are $r$ dimensional vectors

674: and $\Omega$ and $M$ are $r\times r$ matrices, and $L$ is also an

675: $r\times r$ matrix with 6 eigenvalues 1 and $(r-6)$ eigenvalues $-1$.

676: However the form of eqs.\refb{eag9}-\refb{emtrs} 

677: remains

678: the same.

679: 

680:  

681: The theories under consideration also have  

682: S-duality symmetry which leaves the field $M$ unchanged,

683: changes the vector $(\vec Q, \vec P)$

684: to another vector $(\vec Q'', \vec P'')$ via the formula

685: \be \label{sadd3}

686: \pmatrix{\vec Q''\cr \vec P''} = \pmatrix{\alpha & \beta\cr \gamma

687:  & \delta}

688: \pmatrix{\vec Q\cr \vec P}\, , \qquad 

689: \pmatrix{\alpha & \beta\cr \gamma

690:  & \delta}\in \Gamma_1(N)\, ,

691: \ee

692: and transforms $\tau=a+iS$ to

693: \be \label{sadd4}

694: \tau''={\alpha\tau+\beta\over \gamma\tau+\delta}\, ,

695: \ee

696: where the $\Gamma_1(N)$ group of matrices 

697: is defined by the conditions

698: \be \label{egammacond}

699: \alpha\delta - \beta\gamma=1, \qquad \alpha, \delta = 

700: \hbox{1 mod $N$}, \qquad \gamma = \hbox{0 mod $N$},

701: \qquad \beta\in \ZZZ\, .

702: \ee

703: 

704: 

705: We now consider in the first description of this theory

706:  a configuration with a single D5-brane wrapped on 

707:  $\MM\times S^1$, $Q_1$ D1-branes wrapped on $S^1$, a single

708:  Kaluza-Klein monopole associated with the circle $\wt S^1$,

709:  momentum $-n/N$ along $S^1$ and momentum $J$

710:  along $\wt S^1$\cite{0505094}. Since a 

711:  D5-brane wrapped on $\MM$

712:  carries, besides the D5-brane charge, $-\zeta$ 

713:  units of D1-brane charge

714:  with $\zeta$ given by the Euler character of $\MM$ 

715:   divided by 24 \cite{9511222}, the net 

716:   D1-brane charge carried by the system

717:   is $Q_1-\zeta$.

718:  By following the chain of dualities described earlier and a

719:  suitable sign convention for the charges in the first description,

720:  we can map this to a configuration

721:  in the second description with momentum $-n/N$ 

722:  along $S^1$, a single Kaluza-Klein monopole associated with 

723:  $\wh S^1$,

724:   $(-Q_1+\zeta)$  NS 5-brane charge wrapped along $T^4\times

725:   S^1$, $J$  NS 5-brane charge wrapped along 

726:   $T^4\times \wh S^1$

727:   and  unit fundamental string winding charge along 

728:   $S^1$\cite{0605210}. In particular the Kaluza-Klein monopole

729:   charge associated with $\wt S^1$ in the first description gets mapped to

730:   the fundamental string winding number along $S^1$ in the second

731:   description and the D5-brane wrapped on $\MM\times S^1$ in the

732:   first description gets mapped to Kaluza-Klein monopole charge

733:   associated with $\wh S^1$ in the second description.

734:   Using \refb{e2dcharge} we see that 

735:   this corresponds to the charge vectors\footnote{Recall that an

736:   NS 5-brane wrapped on $T^4\times S^1$ represents an H-monopole

737:   associated with $\wh S^1$ and an

738:   NS 5-brane wrapped on $T^4\times \wh S^1$ 

739:   represents an H-monopole

740:   associated with $S^1$.}

741:   \be \label{echvec}

742:   Q=\pmatrix{0 \cr -n/N \cr 0 \cr -1}, \quad P = \pmatrix{Q_1-\zeta

743:  \cr -J\cr 1 \cr 0}\, .

744:   \ee

745:     This gives

746:  \be\label{eqdef}

747:  Q^2  

748:  = 2 n/N, \qquad P^2   

749:  = 2 (Q_1- \zeta   ), \qquad Q\cdot P = J\, .

750: \ee

751: 

752: The counting of states of this system in the weak coupling region

753: of the first description was carried out in 

754: \cite{0605210,0607155,0609109}. 

755: We shall now summarize the results of this analysis.

756: We denote by 

757: $d(\vec Q,\vec P)$ the number of bosonic minus fermionic

758: quarter BPS supermultiplets carrying a given set of charges

759: $(\vec Q, \vec P)$, a supermultiplet being considered bosonic

760: (fermionic) if it is obtained by tensoring the basic 64 dimensional

761: quarter BPS supermultiplet, with helicity ranging from $-{3\over 2}$

762: to ${3\over 2}$,  with a supersymmetry singlet  

763: bosonic (fermionic) state.

764: Our result for $d(\vec Q,\vec P)$  is

765: \be\label{egg1int}

766: d(\vec Q,\vec P) = {1\over N}\, \int _\CC d\wt\rho \, 

767: d\wt\sigma \,

768: d\wt v \, e^{-\pi i ( N\wt \rho Q^2

769: + \wt \sigma P^2/N +2\wt v Q\cdot P)}\, {1

770: \over \wt\Phi(\wt \rho,\wt \sigma, \wt v)}\, ,

771: \ee

772: where $\CC$ is a three real dimensional subspace of the

773: three complex dimensional space labelled by $(\wt\rho=\wrh_1+i

774: \wrh_2,\ws=\ws_1+i\ws_2,\wv=\wv_1+i\wv_2)$. $\CC$ corresponds

775: to the subspace:\footnote{The sign of $\wt v_2$ chosen here differs

776: from the ones used in \cite{0605210,0607155,0609109}. 

777: As will be discussed later, this choice of sign is valid for a

778: specific choice of sign of the axion field; for the other choice

779: the sign of $\wv_2$ needs to be reversed. The other difference is

780: that we have put the $M_3<<M_1, M_2$ condition to make

781: explicit the fact that while interpreting $d(\vec Q, \vec P)$ as the

782: coefficients of the Fourier expansion of $1/\wt\Phi(\wrh,\ws,\wv)$

783: as in eqs.\refb{efo1}, \refb{efo2} 

784: we need to first carry out the expansion in

785: powers of $e^{2\pi i\wrh}$ and $e^{2\pi i\ws/N}$ and then carry out

786: the expansion in powers of $e^{-2\pi i\wt v}$. This was implicit

787: in the results of \cite{0605210,0607155,0609109} but was not

788: stated explicitly.}

789: \bea{ep2kk}

790:  \wt \rho_2=M_1, \quad  \wt\sigma_2 = M_2, \quad

791:  \wt v_2 = -M_3, \nonumber \\

792:  0\le  \wt\rho_1\le 1, \quad

793: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, ,

794: \een

795: $M_1$, $M_2$ and $M_3$ being large but fixed positive

796: numbers with $M_3<< M_1, M_2$, and

797: \bea{edefwtphi}

798: && \wt \Phi(\wt \rho,\wt \sigma,\wt v ) =

799: e^{2\pi i (\wt \alpha\wt\rho + \wt \gamma\ws 

800: + \wt v)} \nonumber \\

801: && \qquad \times \prod_{b=0}^1\, 

802:  \prod_{r=0}^{N-1}

803: \prod_{k'\in \zzz+{r\over N},l\in\zzz,j\in 2\zzz+b

804: \atop k',l\ge 0, j<0 \, {\rm for}

805: \, k'=l=0}

806: \left( 1 - \exp\left(2\pi i ( k'\wt \sigma   +  l\wt \rho +  j\wt v)

807: \right)\right)^{

808: \sum_{s=0}^{N-1} e^{-2\pi i sl/N } c^{(r,s)}_b(4k'l - j^2)} \, .

809: \nonumber \\

810: \eea

811: The coefficients $c^{(r,s)}_b(u)$, $\wt\alpha$,

812: $\wt\gamma$ encode information about

813: the spectrum of two dimensional superconformal $\sigma$-model

814: with target space $\MM$ and are defined as follows.

815: First we define

816: \be\label{esi4aint}

817: F^{(r,s)}(\tau,z) \equiv {1\over N} Tr_{RR;\wt g^r} \left(\wt g^s

818: (-1)^{F_L+F_R}

819: e^{2\pi i \tau L_0} 

820: e^{-2\pi i \bar\tau \bar L_0}

821: e^{2\pi i F_L z}\right), \qquad 0\le r,s\le N-1\, ,

822:  \ee

823: where

824: $Tr_{RR;\wt g^r}$ denotes trace 

825: over all the Ramond-Ramond (RR) sector 

826: states twisted by $\wt g^r$ in the SCFT described above, 

827: $L_n$, $\bar L_n$ denote the left- and right-moving

828: Virasoro generators and $F_L$ and $F_R$ 

829: denote the world-sheet fermion 

830: numbers

831: associated with left and right-moving sectors in this 

832: SCFT. In defining $L_0$ and $\bar L_0$ of a state we subtract 

833: $c_L/24$ and $c_R/24$ from the conformal weights of the

834: corresponding operators so that the RR sector ground state has

835: $L_0=\bar L_0=0$.

836: Due to the insertion of $(-1)^{F_R}$ factor in the trace the

837: contribution to $F^{(r,s)}$ comes only from the $\bar L_0=0$

838: states. As a result $F^{(r,s)}$ does not depend on $\bar\tau$.

839: Furthermore, using the existence of an $SU(2)_L\times SU(2)_R$

840: R-symmetry current algebra in this theory

841: one can show that $F^{(r,s)}(\tau,z)$ have expansions of the form

842: \be\label{enewint}

843: F^{(r,s)}(\tau,z) =\sum_{b=0}^1\sum_{j\in2\zzz+b, n\in \zzz/N

844: \atop 4n-j^2\ge -b^2} 

845: c^{(r,s)}_b(4n -j^2)

846: e^{2\pi i n\tau + 2\pi i jz}\, ,

847: \ee

848: for some coefficients $c^{(r,s)}_b(u)$.  This defines the coefficients

849: $c^{(r,s)}_b(u)$.

850: We now define

851: \be\label{eqrsrev}

852: Q_{r,s} = N\, 

853: \left( c^{(r,s)}_0(0)+ 2 \, c^{(r,s)}_1(-1)\right)\, ,

854: \ee

855: \be\label{enn9d}

856: \wt \alpha={1\over 24N} \, Q_{0,0} - {1\over 2N}

857: \, \sum_{s=1}^{N-1} Q_{0,s}\, {e^{-2\pi i s/N}\over

858: (1-e^{-2\pi i s/N})^2 } \, 

859: , \qquad 

860: \wt \gamma= {1\over 24N} \, Q_{0,0}  \, .

861: \ee

862: This defines all the coefficients appearing in \refb{egg1int}.

863: 

864: As an alternative to \refb{egg1int}, \refb{ep2kk}

865: we can express $d(\vec Q,\vec P)$ as

866: \be\label{efo1}

867: d(\vec Q, \vec P) = g\left({N\over 2} Q^2 , {1\over 2\, N}\, P^2,

868: Q\cdot P\right)\, ,

869: \ee

870: where $g(m,n,p)$ are

871: the coefficients of Fourier expansion of the function

872: $1/ \wt\Phi(\wt \rho,\wt \sigma, \wt v)$:

873: \be\label{efo2}

874: {1

875: \over \wt\Phi(\wt \rho,\wt \sigma, \wt v)}

876: =\sum_{m,n,p} g(m,n,p) \, e^{2\pi i (m\, \wt \rho + n\,

877: \wt\sigma

878: + p\, \wt v)}\, .

879: \ee

880: 

881: Let us denote by $G$ the group of $4\times 4$ matrices generated by:

882: \ben \label{egroup}

883:    \pmatrix{A&B\cr C&D}  &=& \pmatrix{ a & 0 & b & 0 \cr

884:      0 & 1 & 0 & 0\cr c & 0 & d & 0\cr 0 & 0 & 0 & 1}\, ,

885:    \qquad ad-bc=1, \quad \hbox{$c=0$ mod $N$, \quad $a,d=1$

886:      mod $N$}

887:    \nonumber \\

888:    \pmatrix{A&B\cr C&D}  &=& 

889:    \pmatrix{0 & 1 & 0 & 0 \cr -1 & 0 & 0 & 0\cr

890:      0 & 0 & 0 & 1\cr 0 & 0 & -1 & 0}\, , \nonumber \\

891:    \pmatrix{A&B\cr C&D}  &=& 

892:    \pmatrix{ 1 & 0 & 0 & \mu \cr

893:      \lambda & 1 & \mu & 0\cr 0 & 0 & 1 & -\lambda\cr

894:      0 & 0 & 0 & 1}\, , \qquad \lambda, \mu \in \ZZZ,

895: \een

896: and let $\wt G$ denote the group of $4\times 4$ matrices

897: satisfying the requirement that:

898: \be\label{ebelong}

899: \wt g\in \wt G \quad \hbox{iff} \quad U^{-1} \, \wt g U\in G\, ,

900: \ee

901: where\footnote{Even though $U$ contains factors of $\sqrt{N}$

902: the elements of $\wt G$ are integers due to the

903: fact that for $\pmatrix{A & B\cr C & D}\in G$, the elements of

904: $C$ are multiples of $N$\cite{0510147}.}

905: \be\label{edefu}

906: U = \pmatrix{0 & 0 & 0 & -1/\sqrt N\cr -\sqrt N & 0 & 0 & 0\cr

907: 0 & \sqrt N & 0 & 0\cr 0 & 0 & -1/\sqrt N & 0}\, .

908: \ee

909: An element $\pmatrix{\wt A & \wt B\cr \wt C & \wt D}\in \wt G$

910: induces a transformation on $(\wrh,\ws,\wv)$ via the relations:

911: \be\label{enatural}

912: \pmatrix{\wrh' & \wv'\cr \wv' & \ws'}

913: = (\wt A \wt\Omega +\wt B) (\wt C\wt\Omega + \wt D)^{-1},

914: \qquad \wt\Omega\equiv \pmatrix{\wrh & \wv\cr \wv & \ws}\, ,

915: \ee

916: and one can show that

917: $\wt\Phi$ transforms as a Siegel modular form of weight $k$

918: under this transformation\cite{0609109}:

919: \be \label{ewtphitrs}

920: \wt\Phi(\wrh',\ws',\wv') 

921: = \det(\wt C\wt\Omega+\wt D)^k \, \wt\Phi(\wrh,\ws,\wv) \, ,

922: \ee

923: where

924: \be\label{ekvalue}

925: k={1\over 2}\, \sum_{s=0}^{N-1} \, c_0^{(0,s)}(0)\, .

926: \ee

927: 

928: This finishes our review of the main results. However

929: two points about the degeneracy formula given  above need 

930: special mention. Eqs.\refb{egg1int} and \refb{efo1} are equivalent

931: only if the sum over $m$, $n$, $p$ in \refb{efo2}

932: are convergent on the contour $\CC$. For the choice of $\CC$

933: given in \refb{ep2kk} this requires that $\wt\Phi(\wrh,\ws,\wv)^{-1}$

934: has a power series expansion in positive powers of $e^{2\pi i\wrh}$

935: and $e^{2\pi i\ws/N}$ except possibly for a finite number of negative

936: powers, and that the  coefficient  of any

937: given term in this double power series expansion has an expansion

938: in positive powers of $e^{-2\pi i \wv}$ 

939: except possibly for a finite number of negative

940: powers.

941: This in particular 

942: requires that the sum over $m$ and $n$ in \refb{efo2}

943: are bounded from below,

944: and that for fixed $m$ and $n$

945: the sum over $p$ is bounded from above.  By examining the

946: formula \refb{edefwtphi} for $\wt\Phi$ and the result that the

947: coefficients $c^{(r,s)}_b(u)$ are non-zero only for $u\ge -b^2$,

948: we can verify that with the exception of the contribution from the

949: $k'=l=0$ term in this product, the other terms, when expanded

950: in a power series expansion in $e^{2\pi i\wrh}$, $e^{2\pi i\ws}$

951: and $e^{2\pi i\wv}$, 

952: does have the form of \refb{efo2} with $p$ bounded from above

953: (and below)

954: for fixed $m$, $n$.

955: However for the $k'=l=0$ term, which  gives a contribution

956: $e^{-2\pi i \wv} / (1 - e^{-2\pi i \wv})^2$, 

957: there is an ambiguity in

958: carrying out the series expansion. We could either use the form given

959: above and expand the denominator in a series 

960: expansion in $e^{-2\pi i \wv}$ so that the criterion described

961: above is satisfied, or express it as

962: $e^{2\pi i \wv} / (1 - e^{2\pi i \wv})^2$ and expand it in 

963: a series 

964: expansion in $e^{2\pi i \wv}$ in which case the sum over $p$

965: will be bounded from below rather than from

966: above.  One finds that

967: depending on the angle between $S^1$ and $\wt S^1$,

968: only one of these expansions produces the degeneracy

969: correctly via \refb{efo2}\cite{0605210}. The physical 

970: spectrum actually changes as this angle

971: passes through $90^\circ$ since at this point the system is only

972: marginally stable.

973: On the other hand our degeneracy formula

974: \refb{egg1int}, \refb{ep2kk} 

975: implicitly assumes that we have

976: expanded this factor in powers

977: of $e^{-2\pi i \wv}$ since only in this case the sum over $p$ in

978: \refb{efo2} is bounded from above for fixed $m$, $n$.

979: Thus as it stands the expression for $d(\vec Q,\vec P)$

980: given in \refb{egg1int}, \refb{ep2kk} 

981: is valid for a

982: specific range of values of the angle between $S^1$ and $\wt S^1$,

983: which, in the second

984: description of the system, corresponds to the sign of the axion field.

985: For the other sign of the axion we need to carry out the integral

986: over a different contour $\wh \CC$

987: defined as

988: \ben \label{ecchat}

989:  \wt \rho_2=M_1, \quad  \wt\sigma_2 = M_2, \quad

990:  \wt v_2 = M_3, \nonumber \\

991:  0\le  \wt\rho_1\le 1, \quad

992: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, ,

993:  \een

994: where $M_1$, $M_2$ and $M_3$ are large positive numbers

995: with $M_3<<M_1,M_2$.

996: In the

997: convention that we shall be using in this paper, we need to use

998: the contour $\CC$ for positive sign of the axion and the

999: contour $\wh \CC$ for negative sign of the axion.

1000: 

1001: It turns out that walls of marginal stability, -- codimension

1002: one subspaces of the asymptotic moduli space on which

1003: the BPS mass of the 

1004: system becomes equal to the sum of masses of two  or more

1005: other

1006: states carrying the same total charge, -- are quite generic for

1007: quarter BPS states in $\NN=4$ supersymmetric string

1008: theories\cite{9712211}, and we expect the 

1009: spectrum to change discontinuously as

1010: the asymptotic moduli fields pass through any of the

1011: walls of marginal 

1012: stability.\footnote{If a state becomes

1013: marginally stable on a surface of codimension $\geq 2$, then we can

1014: always move around this subspace in going from one point to 

1015: another and hence the spectrum cannot change discontinuously.}

1016: Thus the expression for the degeneracy given in this section 

1017: holds only in

1018: a finite region of the moduli space, bounded by the walls of 

1019: marginal stability. This will be discussed in more detail in

1020: \S\ref{smarginal}.

1021: 

1022: Another point about the formula \refb{egg1int} is that although 

1023: it was

1024: derived for special

1025: charge vectors $\vec Q$, $\vec P$ described in \refb{echvec}, it has

1026: been expressed as a function of the T-duality invariant combinations

1027: $P^2$, $Q^2$ and $Q\cdot P$.

1028: Even though we expect T-duality to be  a symmetry of the theory,

1029: it is not guaranteed that the formula written in this fashion hold

1030: for all charge vectors. First of all a T-duality transformation acts not

1031: only on the charges but also on the asymptotic moduli. 

1032: Had the spectrum been

1033: independent of the asymptotic moduli, we could have demanded

1034: that the spectrum remains invariant under T-duality transformation

1035: of the charges. However if 

1036: a T-duality

1037: transformation takes the asymptotic moduli fields 

1038: across a wall of marginal stability, then

1039: all we can say is that the spectrum remains unchanged under a

1040: simultaneous T-duality transformation of the moduli fields

1041: and the charges, but if we are sitting at a fixed point in the moduli

1042: space then the spectrum is not invariant under T-duality

1043: transformation on the charges. Second point is that even if we

1044: ignore the issues related to the walls of marginal stability,

1045: two charge vectors carrying the same values of $P^2$, $Q^2$

1046: and $Q\cdot P$ may not necessarily

1047: be related by a T-duality

1048: transformation.\footnote{Generically they are related by a

1049: continuous T-duality transformation but only a discrete subgroup

1050: of this is a genuine symmetry of the theory.} In that case the

1051: degeneracy of states for these two charge vectors could be

1052: different. An example of this is that a state that carries

1053: fractional momentum along

1054: $S^1$ can never be related to a state carrying

1055: integer momentum along $S^1$, although they may carry 

1056: same values of $Q^2$, $P^2$ and $Q\cdot P$. Both these issues,

1057: together with the S-duality transformation properties of the

1058: degeneracy formula, will be discussed in \S\ref{sdualtwo}.

1059: 

1060: 

1061: 

1062: 

1063: \sectiono{Walls of Marginal Stability}

1064: \label{smarginal}

1065: 

1066: As has been briefly mentioned in \S\ref{sreview}, the 

1067: degeneracy formula given in \refb{egg1int},  \refb{ep2kk}

1068: is expected to be valid within a certain region of the

1069: moduli space bounded by codimension one subspaces on which the

1070: BPS state under consideration becomes marginally stable. As we

1071: cross this subspace of the moduli space, the spectrum can change

1072: discontinuously. In this section we 

1073: shall study in some detail the locations

1074: of these walls of marginal stability so that we can identify the region

1075: within which our degeneracy formula will remain valid.

1076: 

1077: Let us 

1078: consider a state carrying electric charge $\vec Q$ and magnetic charge

1079: $\vec P$ and examine under what condition it can decay into a

1080: pair of half-BPS states.

1081: This happens

1082: when its mass is equal to the sum of the masses of a pair of half BPS

1083: states whose electric and magnetic charges add up to $\vec Q$ and

1084: $\vec P$ respectively. Since for half BPS states the electric and magnetic

1085: charges must be parallel, these pair of states must

1086: have charge vectors of

1087: the form $(a\vec M, c \vec M)$ and $(b \vec N, d \vec N)$ for some

1088: constants $a$, $b$, $c$, $d$ and a pair of $r$-dimensional vectors

1089: $\vec M$, $\vec N$. We shall normalize $\vec M$, 

1090: $\vec N$ such that 

1091: \be \label{enorm}

1092: ad-bc=1\, .

1093: \ee

1094: Then the requirement that the charges add up to $(\vec Q, \vec P)$ gives

1095: \be \label{esm2}

1096: \vec M = d\vec Q - b\vec P, \qquad \vec N = -c\vec Q + a\vec P\, .

1097: \ee

1098: Thus the charges of the decay products are given by

1099: \be \label{escale0}

1100: (ad\vec Q-ab\vec P, cd\vec Q-cb\vec P) \quad \hbox{and} \quad

1101: (-bc\vec Q + ab\vec P, -cd\vec Q+ad\vec P)\, .

1102: \ee

1103: Note  that under the scale

1104: transformation

1105: \be \label{escale1}

1106: \pmatrix{a & b\cr c & d}\to 

1107: \pmatrix{a & b\cr c & d} \, \pmatrix{\lambda & 0\cr 0 

1108: & \lambda^{-1}}

1109: \ee

1110: eqs.\refb{enorm} and \refb{escale0} remain unchanged.

1111: There is another discrete transformation

1112: \be \label{esc12}

1113: \pmatrix{a & b\cr c & d}\to 

1114: \pmatrix{a & b\cr c & d} \, \pmatrix{0 & 1\cr -1 & 0}\, ,

1115: \ee

1116: which  leaves \refb{enorm} unchanged and exchanges the

1117: two decay products in \refb{escale0}.

1118: A pair of matrices $\pmatrix{a & b\cr c & d}$ related by

1119: \refb{escale1} or \refb{esc12} describe identical decay channels.

1120: 

1121: In order that the charge vectors of the decay products

1122: given in \refb{escale0} satisfy the

1123: charge quantization rules we must ensure that $a\vec M

1124: =ad\vec Q-ab\vec P$

1125: and $b\vec N=-bc\vec Q + ab\vec P$ belong to the lattice of electric

1126: charges and that $c\vec M=cd\vec Q-cb\vec P$ and 

1127: $d\vec N=-cd\vec Q+ad\vec P$ belong to

1128: the lattice of magnetic charges. 

1129: For the charge vectors $\vec Q$, $\vec P$ given in

1130: \refb{echvec} this would require 

1131: \be \label{egen}

1132: ad,ab,bc\in \ZZZ, \qquad 

1133: cd\in N\ZZZ\, .

1134: \ee

1135: The condition $cd\in N\ZZZ$ comes from the requirement

1136: that $cd\vec Q-cb\vec P$ is an allowed magnetic charge.

1137: In particular for a $\vec Q$ of the form \refb{echvec}, a magnetic

1138: charge $cd\vec Q$ represents a state with

1139: Kaluza-Klein monopole charge $-cd$

1140: associated with $S^1$. Since this charge is quantized in units

1141: of $N$, $cd$ must be a multiple of $N$.

1142: We shall denote by $\AAA$ the set of matrices $\pmatrix{a & b\cr c

1143: & d}$ subject to the equivalence relations \refb{escale1}, \refb{esc12}

1144: and satisfying \refb{enorm}, \refb{egen}.

1145: 

1146: It is instructive to determine the structure of the set $\AAA$. We shall

1147: first

1148: show that using the scale transformation \refb{escale1} we can

1149: always choose $a$, $b$, $c$ and $d$ to be integers and furthermore

1150: the solution is unique for given $ad$, $ab$, $bc$ and $cd$. 

1151: Since

1152: $ad$, $ab$, $bc$ and $cd$ are all integers, we can express them as

1153: products of prime factors:

1154: \be \label{eprime1}

1155: |ab| = \prod_i \, p_i^{r_i}, \qquad |cd| = \prod_i p_i^{s_i}, \qquad

1156: |ad| = \prod_i p_i^{u_i}, \qquad |bc| = \prod_i p_i^{v_i},

1157: \ee

1158: where the product over $i$ runs over the prime numbers $p_i$ and

1159: $r_i$, $s_i$, $u_i$ and $v_i$ are non-negative integers satisfying

1160: \be \label{eprime2}

1161: r_i + s_i = u_i + v_i \quad \forall\, i\, .

1162: \ee

1163: Furthermore since $ad$ and $bc$ differ by 1, they cannot have a

1164: common factor. This shows that either $u_i$ or $v_i$ must vanish:

1165: \be \label{eprime3}

1166: u_i \, v_i = 0 \quad \forall \, i\, .

1167: \ee

1168: Let us now look for integer $a$, $b$, $c$ and $d$ satisfying

1169: \refb{eprime1}. For this we use the ansatz:

1170: \be \label{eprime4}

1171: |a| = \prod_i \, p_i^{a_i}, \qquad |b| = \prod_i p_i^{b_i}, \qquad

1172: |c| = \prod_i p_i^{c_i}, \qquad |d| = \prod_i p_i^{d_i},

1173: \ee

1174: where $a_i$, $b_i$, $c_i$ and $d_i$ are non-negative integers.

1175: Eq.\refb{eprime1} now gives

1176: \be \label{eprime5}

1177: a_i + b_i = r_i, \qquad c_i+d_i = s_i, \qquad a_i+d_i=u_i, \qquad

1178: b_i + c_i = v_i\, .

1179: \ee

1180: Now \refb{eprime3} tells us 

1181: that for any given $i$ either $u_i$ or $v_i$ (or both) are

1182: zero. If $u_i=0$ then the only possible solution to

1183: \refb{eprime5} is

1184: \be \label{eprime6}

1185: a_i=0, \qquad d_i=0, \qquad b_i=r_i, \qquad c_i=s_i\, .

1186: \ee

1187: On the other hand if $v_i=0$ then we must have

1188: \be \label{eprime7}

1189: b_i=0, \qquad c_i=0, \qquad a_i=r_i, \qquad d_i=s_i\, .

1190: \ee

1191: This gives a unique expression for $|a|$, $|b|$, $|c|$ and $|d|$ using

1192: prime factorization. Up to an overall factor of $-1$ which can be

1193: removed with the help of the scale transformation 

1194: \refb{escale1}, we can

1195: determine the signs of $a$, $b$, $c$ and $d$ in terms of the

1196: signs of $ab$, $cd$, $ad$ and $bc$.

1197: 

1198: The requirement that $cd$ is a multiple of $N$ implies that the

1199: prime factors of $N$ are shared by $c$ and $d$. In case $N$ is

1200: prime either $c$ or $d$ must be a multiple of $N$. Using 

1201: the freedom \refb{esc12} we can ensure that $c$ is a multiple

1202: of $N$.  In this case the matrices $\pmatrix{a & b\cr c & d}$ 

1203: describe elements of $\Gamma_0(N)$ 

1204: modulo multiplication

1205: by $-1$.\footnote{$\Gamma_0(N)$ contains matrices

1206: $\pmatrix{a&b\cr c&d}$ satisfying $ad-bc=1$, $a,b,d\in\ZZZ$,

1207: $c\in N\ZZZ$.}

1208: Since $\Gamma_0(2)=\Gamma_1(2)$, for $N=2$ the set of matrices

1209: $\pmatrix{a&b\cr c&d}$ may be identified as the elements of

1210: $\Gamma_1(N)$ modulo multiplication by $-1$. On the other

1211: hand using the freedom of multiplication by $-1$ we can convert

1212: any $\Gamma_0(3)$ matrix to a $\Gamma_1(3)$ matrix.

1213: Thus for $N=3$ the matrices $\pmatrix{a &b\cr c & d}$ may be

1214: chosen to be $\Gamma_1(3)$ matrices.

1215: 

1216: The case $N=1$ with $\MM=K3$, corresponding to heterotic string

1217: theory on $T^6$ in the second description, is somewhat special. In

1218: this case the set $\AAA$ consists of $PSL(2,\ZZZ)$ matrices subject

1219: to the equivalence relation \refb{esc12}.

1220: 

1221: We shall now determine the wall of marginal stability corresponding

1222: to the decay channel given in \refb{escale0}.

1223: Our starting point will be the formula for the mass $m(\vec Q,

1224: \vec P)$

1225: of a BPS state carrying electric charge $\vec Q$ and magnetic charge

1226: $\vec P$\cite{9507090,9508094}

1227: \ben \label{esm1}

1228: m(\vec Q, \vec P)^2 &=& {1\over S_\infty} 

1229: (Q - \bar\tau_\infty P)^T

1230: (M_\infty + L) (Q -\tau_\infty P) \nonumber \\

1231: &&

1232: + 2 \left[ (Q^T (M_\infty + L) Q) (P^T (M_\infty + L) P)

1233: - (P^T (M_\infty + L) Q)^2\right]^{1/2}\, , \nonumber \\

1234: \een

1235: where $\tau=a+iS$ and the subscript $\infty$ denotes asymptotic

1236: values of various fields. This expression is manifestly invariant under

1237: the T- and S-duality transformations described in 

1238: eqs.\refb{eag9}-\refb{emtrs}

1239: and \refb{sadd3}-\refb{egammacond}.

1240: In order that the state $(\vec Q,\vec P)$ is marginally stable against decay

1241: into $(ad\vec Q-ab\vec P, cd\vec Q-cb\vec P)$ and $(-bc\vec Q

1242: +ab\vec P, -cd\vec Q+ad\vec P)$, we need

1243: \be \label{esm3}

1244: m(\vec Q,\vec P) = m(ad\vec Q-ab\vec P, cd\vec Q-cb\vec P)

1245: + m(-bc\vec Q

1246: +ab\vec P, -cd\vec Q+ad\vec P)\, .

1247: \ee

1248: Using \refb{esm1}, \refb{esm3} 

1249: and some tedious algebra, we arrive at the

1250: condition

1251: \be \label{esm4}

1252: \left(a_\infty - {ad+bc\over 2cd}\right)^2 

1253: + \left( S_\infty +{E\over 2 cd}

1254: \right)^2 = {1\over 4 c^2 d^2} (1 + E^2)\, ,

1255: \ee

1256: where

1257: \be \label{esm5}

1258: E \equiv { cd (Q^T (M_\infty + L) Q)

1259: + ab (P^T (M_\infty + L) P) -(ad + bc) (P^T (M_\infty + L) Q) \over 

1260: \left[ (Q^T (M_\infty + L) Q) (P^T (M_\infty + L) P)

1261: - (P^T (M_\infty + L) Q)^2\right]^{1/2}

1262: }\, .

1263: \ee

1264: Note that $E$ depends on $M_\infty$, the constants $a,b,c,d$ and

1265: the charges $\vec Q$, $\vec P$, but is 

1266: independent of $\tau_\infty$.

1267: Thus for fixed $P$, $Q$ and $M_\infty$, the wall of marginal

1268: stability describes a circle in the $(a_\infty, S_\infty)$

1269: plane with radius 

1270: \be \label{esm6}

1271: R = \sqrt{1 + E^2}/2|cd|\, ,

1272: \ee 

1273: and center at

1274: \be \label{esm7}

1275: C = \left( {ad+bc\over 2cd}, -{E\over 2 cd}\right)\, .

1276: \ee

1277: This circle 

1278: intersects the real $\tau_\infty$

1279: axis at

1280: \be \label{eireal}

1281: {a/c} \quad \hbox{and} \quad {b/d}\, .

1282: \ee

1283: 

1284: The cases where either $c$ or $d$ vanish require special attention.

1285: First consider the case $c=0$. In this case the condition

1286: $ad-bc=1$ implies that $a=d=1$. By taking the $c\to 0$

1287: limit of \refb{esm4}, \refb{esm5} 

1288: we see that  the wall of marginal

1289: stability becomes a straight line in the $(a_\infty, S_\infty)$ plane

1290: for a fixed $M_\infty$:

1291: \be\label{ecurve1}

1292: a_\infty - {b (P^T (M_\infty + L) P)-(P^T (M_\infty + L) Q)  \over 

1293: \left[ (Q^T (M_\infty + L) Q) (P^T (M_\infty + L) P)

1294: - (P^T (M_\infty + L) Q)^2\right]^{1/2}

1295: }\, S_\infty -b =0\, .

1296: \ee

1297: On the other hand for $d=0$ we have $bc=-1$ and

1298: we can choose $b=-1$, $c=1$. In the $d\to 0$ limit of \refb{esm4}, 

1299: \refb{esm5} we get

1300: another straight line

1301: \be\label{ecurve2}

1302: a_\infty - {  a (P^T (M_\infty + L) P)  -(P^T (M_\infty + L) Q) \over 

1303: \left[ (Q^T (M_\infty + L) Q) (P^T (M_\infty + L) P)

1304: - (P^T (M_\infty + L) Q)^2\right]^{1/2}

1305: }\, S_\infty -a =0\, .

1306: \ee

1307: We now notice that this has exactly the same form as \refb{ecurve1}

1308: with $b$ replaced by $a$. Thus these do not give rise to new walls

1309: of marginal stability. In fact the $c=0$ and $d=0$ cases are related

1310: by the equivalence relation \refb{esc12}.

1311: 

1312: In order to get some insight into the geometric structure of the domain

1313: bounded by these marginal stability walls it will be useful to

1314: study the possible intersection points of these walls in the upper

1315: half $\tau_\infty$ plane. Let us consider a pair of such walls

1316: characterized by the matrices $\pmatrix{a_1 & b_1\cr c_1 & d_1}$

1317: and $\pmatrix{a_2 & b_2\cr c_2 & d_2}$. 

1318: A convenient

1319: procedure for studying their intersection is to convert one

1320: of them (say the first one) to a straight line

1321: by an SL(2,$\ZZZ)$ transformation.

1322: We define

1323: \be \label{etrs1}

1324: \tau'_\infty \equiv a'_\infty + i S'_\infty = {d_1 \tau - b_1\over

1325: -c_1\tau + a_1}\, .

1326: \ee

1327: Then it is easy to see that in the $(a'_\infty, S'_{\infty})$ plane

1328: the two walls get mapped to the curves

1329: \be \label{etrs2}

1330: a'_\infty + {   \wt P^T (M_\infty + L) \wt Q  \over 

1331: \left[ (\wt Q^T (M_\infty + L) \wt Q) (\wt P^T (M_\infty + L) \wt

1332: P)

1333: - (\wt P^T (M_\infty + L) \wt Q)^2\right]^{1/2}

1334: }\, S'_\infty = 0\, ,

1335: \ee

1336: and

1337: \be \label{etrs4}

1338: \left(a'_\infty - {\wt a\wt d+\wt b\wt c\over 2\wt c\wt 

1339: d}\right)^2 

1340: + \left( S'_\infty +{\wt E\over 2 \wt c\wt d}

1341: \right)^2 = {1\over 4 \wt c^2 \wt d^2} (1 + \wt E^2)\, ,

1342: \ee

1343: where

1344: \be \label{etrs6}

1345: \vec{\wt Q} = d_1\vec Q - b_1\vec P, \qquad 

1346: \vec{\wt P} = -c_1\vec Q +a_1 \vec P

1347: \, ,

1348: \ee

1349: \be\label{etrs7}

1350: \pmatrix{\wt a & \wt b\cr \wt c & \wt d} =

1351: \pmatrix{d_1 & -b_1\cr -c_1 & a_1} \pmatrix{a_2 & b_2\cr

1352: c_2 & d_2}\, ,

1353: \ee

1354: \be \label{etrs5}

1355: \wt E \equiv { \wt c\wt d (\wt

1356: Q^T (M_\infty + L) \wt Q)

1357: + \wt a\wt b (\wt P^T (M_\infty + L) \wt P) 

1358: -(\wt a\wt d + \wt b\wt c) 

1359: (\wt P^T (M_\infty + L) \wt Q) \over 

1360: \left[ (\wt Q^T (M_\infty + L) \wt Q) (\wt P^T (M_\infty + L) \wt

1361: P)

1362: - (\wt P^T (M_\infty + L) \wt Q)^2\right]^{1/2}

1363: }\, .

1364: \ee

1365: If either $\wt c$ or $\wt d$ vanishes \i.e.\ if either 

1366: $a_1/c_1=a_2/c_2$

1367: or $a_1/c_1=b_2/d_2$ then 

1368: \refb{etrs4} reduces to a straight line of the form \refb{ecurve1}

1369: or \refb{ecurve2} with $P$, $Q$, $b$ (or $a$)

1370: replaced by $\wt P$, $\wt Q$, $\wt b$ (or $\wt a$)

1371: respectively, 

1372: and one finds that the only point of

1373: intersection of this line with \refb{etrs2}

1374: in the upper half $\tau'_\infty$ plane is at

1375: $i\infty$. In the $\tau_\infty$ plane this corresponds to the

1376: point $a_1/c_1$. If neither $\wt c$ nor $\wt d$ vanishes, then

1377: by

1378: eliminating $a'_\infty$ from \refb{etrs2} and \refb{etrs4}

1379: we get

1380: \ben \label{etrs8}

1381: && (S'_\infty)^2 \, { \wt

1382: Q^T (M_\infty + L) \wt Q\, \wt P^T (M_\infty + L) \wt

1383: P \over (\wt Q^T (M_\infty + L) \wt Q) (\wt P^T (M_\infty + L) \wt

1384: P)

1385: - (\wt P^T (M_\infty + L) \wt Q)^2} \nonumber \\

1386: && + S'_\infty \, { \wt Q^T (M_\infty + L) \wt Q + (\wt a\wt b/\wt c

1387: \wt d)

1388: \wt P^T (M_\infty + L) \wt

1389: P \over \left[ (\wt Q^T (M_\infty + L) \wt Q) 

1390: (\wt P^T (M_\infty + L) \wt

1391: P)

1392: - (\wt P^T (M_\infty + L) \wt Q)^2\right]^{1/2}

1393: } \nonumber \\

1394: && + {(\wt a\wt d+\wt b\wt c)^2 -1\over 4 \wt c^2 \wt d^2} = 0\, .

1395: \een

1396: Using the conditions $\wt a\wt d-\wt b\wt c=1$ and $\wt a, 

1397: \wt b, \wt c, \wt d \in \ZZZ$ we see that $\wt a \wt d$ and

1398: $\wt b \wt c$ have same signs and also that 

1399: $(\wt a\wt d+\wt b\wt c)^2 \ge 1$. As a result for $S'_\infty\ge 0$,

1400: each term in the left hand side of \refb{etrs8} is non-negative

1401: and the only possible solution is 

1402: \be \label{etrs9}

1403: \wt b=0\quad \hbox{or} \quad

1404: \wt a=0\, , \qquad S'_\infty = 0 \, . 

1405: \ee

1406: Going back to the original variables using 

1407: \refb{etrs1}, \refb{etrs7}  we 

1408: see that these correspond to the following two cases:

1409: \be \label{enewcases}

1410: {b_1\over d_1}={a_2\over c_2} \quad \hbox{or} \quad

1411:  {b_1\over d_1}={b_2\over d_2}, \qquad \tau_\infty={b_1\over d_1}

1412:  \, .

1413:  \ee

1414: Collecting all the cases together we see that the

1415: pair of circles characterized by the matrices $\pmatrix{a_1 & b_1

1416: \cr c_1 & d_1}$ and $\pmatrix{a_2 & b_2

1417: \cr c_2 & d_2}$ never intersect in the interior of the upper

1418: half plane and intersect on the real axis if and only if the sets

1419: \be \label{etrs11a}

1420: \left\{{a_1\over c_1}, {b_1\over d_1}\right\} \quad \hbox{and}

1421: \quad

1422: \left\{{a_2\over c_2}, {b_2\over d_2}\right\} \, 

1423: \ee

1424: have an overlap.

1425: 

1426: 

1427: {}From this analysis we see that

1428: the set of marginal stability walls divides the upper half

1429: $\tau_\infty$ plane into many domains, with each domain

1430: bounded by a set of walls described by a set of matrices

1431: $\left\{\pmatrix{a_i & b_i\cr c_i & d_i}\right\}$. Since the

1432: walls meet on the real axis or at $i\infty$, the regions have

1433: possible vertices on the real axis or at $i\infty$, 

1434: but never in the interior of

1435: the upper half $\tau_\infty$ plane.  We also note that

1436: while the shape of the walls in the $\tau_\infty$ plane

1437: depends on the charges and the

1438: moduli $M_\infty$, their intersetion points are determined

1439: only in terms of the associated matrices 

1440: $\pmatrix{a_i & b_i\cr c_i & d_i}$.

1441: 

1442: How do we determine the walls 

1443: bordering a given domain? We shall illustrate 

1444: this  for the domain that includes the

1445: large $S_\infty$ region and is bounded on

1446: the left (right) by the

1447: line \refb{ecurve1} for $b=0$ ($b=1$). Since these lines

1448: intersect the real $\tau_\infty$ axis at $0$ and 1 respectively,

1449: the domain is bounded from below by  set of circle segments

1450: in the upper half plane described by \refb{esm4},  

1451: of which the

1452: first one begins

1453: at 0 and ends at some point $P_1$

1454: on the positive real axis, the second

1455: one begins at $P_1$ and ends at some point $P_2$ to the right

1456: of $P_1$ etc. with the final segment ending at 1. The parameters

1457: $a,b,c,d$ for each segment

1458: must be chosen such that given the starting point, it travels

1459: maximum possible distance to the right; any other segment that

1460: travels less distance will lie beneath the maximal segment and will

1461: not be of relevance for this computation. 

1462: Thus in order to determine these

1463: segments we first need to study the following general question:

1464: given a rational number $p/q\ge0$, what is the maximum

1465: distance a wall can travel given that it starts at $p/q$?

1466: 

1467: 

1468: Let us first consider the case $p/q>0$.

1469: We choose $p$, $q$ to be relatively prime

1470: and $p>0$, $q>0$.

1471: Now eq.\refb{eireal}  shows that 

1472: the circle \refb{esm4} intersects the real axis at $a/c$ and $b/d$.

1473: Thus either $a/c$ or $b/d$ should be equal to $p/q$. Without

1474: any loss of generality we can choose $a/c=p/q$ and hence 

1475: $a=p$, $c=q$ up to an irrelevant overall sign.

1476: The constraint $ad-bc=1$ now gives

1477: \be \label{ebbyd}

1478: {b\over d} = {p\over q} - {1\over qd}\, .

1479: \ee

1480: We need to choose $d$ so as to maximize the right hand side.

1481: Thus $d$ must be chosen to be the maximum possible negative

1482: number subject to the requirement that $cd=qd$ is a multiple of

1483: $N$ and $b=(pd-1)/q$ is an integer. 

1484: 

1485: In the special case where $p/q=0$, we have $a/c=0$ and

1486: hence $a=0$. The condition

1487: $ad-bc=1$ tells us that we have (up to an overall sign) 

1488: $b=-1$, $c=1$. The second intersection point $b/d=-1/d$ is

1489: maximized for maximum negative

1490: value of $d$ subject to the

1491: requirement that $cd$ is a multiple of $N$. This gives

1492: $d=-N$ and hence

1493: \be \label{ebbyd2}

1494: {b\over d} = {1\over N}\, .

1495: \ee

1496: 

1497: \begin{figure}

1498: \leavevmode

1499: \begin{center}

1500: \hbox{

1501: \epsfysize=3cm \epsfbox{fig1.eps}

1502: \epsfysize=3cm \epsfbox{fig2.eps}

1503: \epsfysize=3cm \epsfbox{fig3.eps}

1504: }

1505: \end{center}

1506: \caption{A schematic diagram representing the domain $\RR$

1507: in the upper half $\tau_\infty$ plane, bounded by the walls of

1508: marginal stability,  for $\ZZZ_1$, $\ZZZ_2$ and $\ZZZ_3$ orbifolds. 

1509: The shapes of the circles and

1510: the slopes of the straight lines bordering the domain depend

1511: on the charges and other asymptotic moduli, but the vertices

1512: are universal.} \label{f1}

1513: \end{figure}

1514: 

1515: 

1516: 

1517: Following this rule we can now construct the circles 

1518: which border

1519: the domain from below. 

1520: For the purpose of illustration we shall carry out the first few steps. 

1521: Using \refb{ebbyd2} we see that the

1522: first circle segment beginning at 0 ends at $1/N$. 

1523: For $N=1$ this completes the story since we have already

1524: reached the

1525: point 1. For $N\ge 2$ we need to proceed further. Taking $p/q=1/N$

1526: we see from \refb{ebbyd} that $b=(d-1)/N$. Since $q=N$, the

1527: condition that $qd$ is a multiple of $N$ is trivially satisfied.

1528: Thus we need to choose $d$ to be the maximum negative number

1529: for which $b=(d-1)/N$ is integer. This gives $d=-(N-1)$ and $b=-1$.

1530: Thus the circle ends at $b/d=1/(N-1)$. For $N=2$ this completes

1531: the story but for $N\ge 3$ we need to proceed further. At the next

1532: stage we begin with $p/q=1/(N-1)$ and get $b=-(N-1)$, 

1533: $d=-N(N-2)$. Thus $b/d=(N-1)/((N(N-2))$. This does not complete

1534: the story for any $N\ge 3$; {\it e.g.}

1535: for $N=3$ this gives $b/d=2/3$. By 

1536: continuing

1537: this process one can show that for $N=3$ we reach the point

1538: 1 at the next step via the wall corresponding

1539: to $\pmatrix{a & b\cr c & d}=\pmatrix{2 & -1\cr 3 & -1}$ but

1540: for higher $N$ the story continues further. A schematic diagram

1541: representing these domains for $N=1$, 2 and 3 

1542: have been shown in Fig.~\ref{f1}. As will be discussed in

1543: \S\ref{stest}, for $N\ge 4$ the number of

1544: such walls is infinite. 

1545: 

1546: This finishes our general analysis of marginal stability walls and

1547: domains bounded by them. Now we focus on one particular domain,

1548: -- the one in which the degeneracy formula given in

1549: \refb{egg1int}, \refb{ep2kk} is valid. 

1550: Since the calculation leading to

1551: \refb{egg1int}, \refb{ep2kk} was performed in a specific corner of the

1552: moduli space, -- the weakly coupled type IIB string theory,

1553: -- all we need to know is how this region is situated 

1554: with respect to the

1555: various marginal stability

1556: walls described here.

1557: However to address this issue we need to 

1558: determine

1559: the relation beween the moduli parameters in the first description,

1560: -- as an orbifold of type IIB string theory, -- and the moduli 

1561: $\tau_\infty$, $M_\infty$ appearing in the BPS mass formula

1562: \refb{esm1}.

1563: We have already stated that

1564: $\tau_\infty=a_\infty+iS_\infty$ 

1565: denotes the asymptotic value of the complex structure

1566: of the torus $(S^1\times \wt S^1)/\ZZZ_N$. We have worked in

1567: a region of the moduli space where it is finite, \i.e.\

1568: $S_\infty$ is neither too large nor too small. The relation between

1569: the other moduli fields in the first description and the matrix valued

1570: moduli field $M$ can be found by following carefully the duality

1571: chain that takes the theory to its second description, 

1572: and using the

1573: identification of $M$ with the geometric quantities in this

1574: description as given in \refb{etenfour}. Let us denote, 

1575:  in the

1576: first description of the theory, 

1577: by $g$ the ten dimensional coupling constant, by 

1578: $(2\pi\sqrt{\alpha'})^4V$ the volume

1579: of $\MM$ measured in the string metric, and by 

1580: $(2\pi\sqrt{\alpha'})^2A$ the area of

1581: $S^1\times \wt S^1/\ZZZ_N$ measured in the string metric. 

1582: Let us also set the anti-symmetric tensor field and

1583: all the Ramond-Ramond fields to zero. Then using the standard

1584: duality transformation rules one can show that the

1585: relevant $4\times 4$ component part of the matrix $M$ that couples

1586: to $Q$ and $P$ are given by

1587: \be \label{ematrix}

1588: M_\infty= \pmatrix{{1/ V} &&& \cr & {g^2/(A^2

1589:  V)} &&\cr

1590: &&  V & \cr &&& {A^2  V/ g^2}}\, .

1591: \ee

1592: Using

1593: the expression for $Q$ and $P$

1594: given in \refb{echvec} one now gets

1595: \ben\label{egets}

1596: Q^T(M_\infty+L)Q &=& {g^2\over A^2  V} \left({n\over N}\right)^2

1597: + {A^2 V\over g^2} + 2{n\over N}, \nonumber \\

1598: P^T(M_\infty+L)P&=&

1599: {(Q_1-\zeta)^2\over V} +{g^2\over A^2  V} J^2

1600: +  V + 2(Q_1-\zeta), \nonumber \\

1601: P^T(M_\infty+L)Q &=& {g^2\over A^2  V} {nJ\over N} + J\, .

1602: \een

1603: The computation of the degeneracy was 

1604: done in a region where $g$ is small but

1605: $A$ and $V$ are of order 1. Using \refb{egets} 

1606: one can now see that in this region

1607: \be\label{escale3}

1608: Q^T(M_\infty+L)Q >> P^T (M_\infty+L) P, \, 

1609: |P^T (M_\infty+L) Q|\, .

1610: \ee

1611: Even when we deform $M_\infty$ away a little from its diagonal form

1612: \refb{ematrix}, eq.\refb{escale3} continues to hold.

1613: 

1614: We now study the implication of \refb{escale3} 

1615: on  the location of the 

1616: walls of marginal stability in the $\tau_\infty$ plane.

1617: First consider the case where $cd\ne 0$. In this case we see from

1618: \refb{esm5}, \refb{escale3} that 

1619: \be \label{escale4}

1620: {E\over cd}  \simeq \sqrt{Q^T(M_\infty+L)Q\over

1621: P^T (M_\infty+L) P}>> 1\, .

1622: \ee

1623: Hence for the circle in the $\tau_\infty$ plane described by

1624: \refb{esm4} we have

1625: \be \label{escale5}

1626: |S_\infty| <  {\sqrt{1 + E^2}\over |2cd|} - {E\over 2cd} << 1\, .

1627: \ee

1628: In other words in the $(a_\infty, S_\infty)$ plane the uppermost point

1629: on the circle \refb{esm4} lies little above the $S_\infty=0$ axis and

1630: its center lies deep down in the lower half plane. 

1631: Since $S_\infty\sim 1$ in

1632: the region of the moduli space in which we have worked, we see that

1633: this region lies {\it above} all the circles described by

1634: \refb{esm4} in the $\tau_\infty$ plane.

1635: 

1636: Next we consider the case $c=0$, -- as discussed earlier the $d=0$

1637: case is equivalent to this. In this case eq.\refb{egets} 

1638: shows that the coefficient of $S_\infty$

1639: in \refb{ecurve1} is small for

1640: finite $b$ and small $g$. Thus for fixed

1641: $M_\infty$ these lines

1642: are almost vertical in the $(a_\infty, S_\infty)$ plane and are

1643: given by

1644: \be \label{evert1}

1645: a_\infty \simeq b\, .

1646: \ee

1647: The $b=0$ line corresponds to the wall of marginal stability

1648: described at the end of \S\ref{sreview}, and as we described in

1649: \S\ref{sreview}, the degeneracy formula actually jumps across

1650: this wall.\footnote{In the analysis of \cite{0605210} the jump 

1651: occured exactly across the line $a_\infty=0$. This can be traced

1652: to the fact that there the analysis was carried out 

1653: using weak coupling approximation

1654: and did not take into account the backreaction

1655: due to switching on $J$. Indeed for $J=0$ we have $P^T(M_\infty+L)

1656: Q=0$, and \refb{ecurve1} for $b=0$ reduces to the vertical line

1657: $a_\infty=0$.} Since our degeneracy formula has been derived in

1658: the small $a_\infty$ region, we see that

1659: the region of validity of our formula is

1660: bounded by the $b=-1$ line on the left, $b=1$ line on the

1661: right and a set of circle segments below.

1662: 

1663: To summarize, the region of the moduli space  

1664: in which we have carried out our

1665: analysis consists of two domains. One of them, lying between the

1666: $\pmatrix{a & b\cr c & d}=\pmatrix{1 & 0\cr 0 & 1}$ line

1667: and $\pmatrix{a & b\cr c & d}=\pmatrix{1 & 1\cr 0 & 1}$ line

1668: extends to the large $S_\infty$ region, and 

1669: is bounded from below by a set of circles. 

1670: For later reference we 

1671: shall call this the right domain $\RR$ and

1672: denote by $\BB_R$ the set of matrices $\pmatrix{a & b

1673: \cr c & d}$ labelling the boundaries of this domain.

1674: The other domain,

1675: lying between the

1676: $\pmatrix{a & b\cr c & d}=\pmatrix{1 & -1\cr 0 & 1}$ line

1677: and $\pmatrix{a & b\cr c & d}=\pmatrix{1 & 0\cr 0 & 1}$ line

1678: also extends to the large $S_\infty$ region and

1679: is  bounded from below by a set of circles. We shall call this the

1680: left domain $\LL$, and denote by $\BB_L$ the set of matrices

1681: $\pmatrix{a & b

1682: \cr c & d}$ labelling the boundaries of this domain.  We shall

1683: argue in \S\ref{stest} that the

1684: set $\BB_L$ is obtained simply by multiplying the elements

1685: of $\BB_R$ by the matrix $\pmatrix{1 & -1\cr 0 & 1}$ from the

1686: left. 

1687: 

1688: One important issue that we would like to address is: how does the

1689: degeneracy formula change as we move across a wall of marginal

1690: stability? We already 

1691: know that as we move across the wall \refb{ecurve1} for $b=0$

1692: the degeneracy formula continues to be given by \refb{egg1int}

1693: except for a change in the location of the integration contour

1694: from $\CC$ to $\wh\CC$. We shall argue in

1695: \S\ref{sdualtwo} that this is a general phenomenon; as we move

1696: across any line of marginal stability the 

1697: degeneracy formula will be given

1698: by the same expression except for

1699: a change in the integration contour.  

1700: 

1701: \sectiono{Duality Transformation of the Degeneracy Formula} 

1702: \label{sdualtwo}

1703: 

1704: As noted in \S\ref{sreview}, 

1705: the degeneracy formula \refb{egg1int},

1706: \refb{ep2kk} has been 

1707: written in terms

1708: of T-duality invariant combinations $Q^2$, $P^2$ and $Q\cdot P$

1709: although we have derived the formula only for a special class of

1710: charge vectors. In this section we shall discuss what information about

1711: the degeneracy formula can be extracted using the 

1712: T- and S-duality symmetries of the theory.

1713: 

1714: We begin by studying the consequences of 

1715: the T-duality symmetries of the theory. It follows from

1716: \refb{eag9}, \refb{sadd1} that if a T-duality transformation takes

1717: a charge vector $(\vec Q, \vec P)$ to $(\vec Q',\vec P')$ then

1718: \be \label{sadd2}

1719: Q^{\prime 2}=Q^2, \quad P^{\prime 2}=P^2, \quad Q'\cdot P'

1720: = Q\cdot P\, .

1721: \ee

1722: However there may be

1723: pairs of charge vectors with the same $Q^2$, $P^2$ and

1724: $Q\cdot P$ which are not related by a T-duality transformation.

1725: Clearly T-duality invariance of the theory cannot give us

1726: any relation between the degeneracies associated with

1727: such a pair of charge vectors.

1728: In what follows we shall focus on charge vectors $(\vec Q',\vec P')$

1729: which are in the same T-duality orbit of a charge vector $(\vec Q,

1730: \vec P)$ for which we have derived \refb{egg1int}. 

1731: 

1732: We have denoted by $\RR$ the right domain of the

1733: region of the moduli space described in \S\ref{smarginal} in which

1734: the original formula for $d(\vec Q, \vec P)$ is valid. It is bounded

1735: by a set of marginal stability walls labelled by 

1736: $\pmatrix{a & b\cr c & d} \in\BB_R$.  Let

1737: $\RR'$ denote the image of

1738: $\RR$ under the T-duality map. 

1739: In this case we

1740: expect $d(\vec Q', \vec P')$ in the region

1741: $\RR'$ to be equal to $d(\vec Q, \vec P)$ given

1742: in \refb{egg1int}:

1743: \ben\label{egg1intrp}

1744: d(\vec Q',\vec P') &=& {1\over N}\, \int _\CC d\wt\rho \, 

1745: d\wt\sigma \,

1746: d\wt v \, e^{-\pi i ( N\wt \rho Q^2

1747: + \wt \sigma P^2/N +2\wt v Q\cdot P)}\, {1

1748: \over \wt\Phi(\wt \rho,\wt \sigma, \wt v)}\, , \nonumber \\

1749: &=& {1\over N}\, \int _\CC d\wt\rho \, 

1750: d\wt\sigma \,

1751: d\wt v \, e^{-\pi i ( N\wt \rho Q^{\prime 2}

1752: + \wt \sigma P^{\prime 2}2/N +2\wt v Q'\cdot P')}\, {1

1753: \over \wt\Phi(\wt \rho,\wt \sigma, \wt v)}\, ,

1754: \een

1755: where $\CC$ has been defined in \refb{ep2kk}.

1756: In going from the first

1757: to the second line of \refb{egg1intrp} we have used \refb{sadd2}.

1758: 

1759: Let us now determine the region $\RR'$.

1760: Since under a T-duality

1761: transformation $M\to \Omega M \Omega^T$, and since $\RR'$

1762: is the image of $\RR$ under this map, $\RR'$

1763: is bounded by walls of marginal stability described in \refb{esm4},

1764: \refb{esm5} with $\pmatrix{a & b\cr c & d}\in \BB_R$

1765: and $M_\infty$ in \refb{esm5}

1766: replaced by $\Omega^{-1} M_\infty (\Omega^T)^{-1}$.

1767: Using \refb{sadd1}

1768: we see that this effectively replaces $(\vec Q, \vec P)$ by $(\vec Q',

1769: \vec P')$ in \refb{esm5}. Thus

1770: $\RR'$ is the region of the upper half plane bounded by the

1771: circles:

1772: \be \label{esm4rp}

1773: \left(a_\infty - {ad+bc\over 2cd}\right)^2 

1774: + \left( S_\infty +{E'\over 2 cd}

1775: \right)^2 = {1\over 4 c^2 d^2} (1 + E^{\prime 2})\, , 

1776: \quad \pmatrix{a & b\cr c & d}\in \BB_R\, ,

1777: \ee

1778: where

1779: \be \label{esm5rp}

1780: E' \equiv {   cd (Q^{\prime T} (M_\infty + L) Q')

1781: + ab (P^{\prime T} (M_\infty + L) P') 

1782: -(ad + bc) (P^{\prime

1783: T} (M_\infty + L) Q') \over 

1784: \left[ (Q^{\prime T} (M_\infty + L) Q') (P^{\prime T} (M_\infty + L) 

1785: P')

1786: - (P^{\prime T} (M_\infty + L) Q')^2\right]^{1/2}

1787: }\, .

1788: \ee

1789: \refb{egg1intrp}-\refb{esm5rp} are valid for any charge vector

1790: $(\vec Q',\vec P')$ which can be related to the charge vectors

1791: given in \refb{echvec} via a T-duality transformation.

1792: 

1793: Next we shall analyze the consequences of S-duality symmetry.

1794: An S-duality transformation changes the vector $(\vec Q, \vec P)$

1795: to another vector $(\vec Q'', \vec P'')$ 

1796: and $\tau$ to $\tau''$ 

1797: via the formul\ae\ \refb{sadd3}, \refb{sadd4}.

1798: Thus if $\RR''$ denotes

1799: the image of the region $\RR$ under the  map \refb{sadd4}, then

1800: S-duality invariance implies that

1801: inside $\RR''$ the degeneracy $d(\vec Q'', \vec P'')$ is given by

1802: the same expression \refb{egg1int} for $d(\vec Q, \vec P)$:

1803: \be\label{sadd5}

1804: d(\vec Q'',\vec P'') = {1\over N}\, \int _\CC d\wt\rho \, 

1805: d\wt\sigma \,

1806: d\wt v \, e^{-\pi i ( N\wt \rho Q^2

1807: + \wt \sigma P^2/N +2\wt v Q\cdot P)}\, {1

1808: \over \wt\Phi(\wt \rho,\wt \sigma, \wt v)}\, .

1809: \ee

1810: We would like to express the right hand side of \refb{sadd5} in

1811: terms of the vectors $\vec Q''$ and $\vec P''$. For this we define

1812: \be\label{epm}  

1813: \pmatrix{\tilde \alpha &\tilde \beta\cr\tilde \gamma &\tilde \delta} = 

1814: \pmatrix{\delta  & \gamma /N \cr \beta  N & \alpha }\in

1815: \Gamma_1(N)\, .

1816: \ee

1817: and

1818: \be\label{e3.3}

1819:   \pmatrix{\wrh''\cr \ws''\cr \wv''}  \equiv 

1820:   \pmatrix{\wrh''_1+i\wrh''_2\cr \ws''_1+i\ws''_2\cr \wv''_1

1821:   +i\wv''_2}=

1822:   \pmatrix{\tilde \alpha^2& \tilde \beta^2&- 2\tilde 

1823:    \alpha

1824:    \tilde \beta\cr

1825:    \tilde \gamma^2&\tilde \delta^2 & - 

1826:    2\tilde

1827:    \gamma\tilde \delta\cr

1828:     -\tilde \alpha\tilde \gamma& - \tilde \beta\tilde 

1829:    \delta & (\tilde \alpha\tilde \delta + \tilde 

1830:    \beta\tilde \gamma)} \pmatrix{\wrh\cr \ws\cr \wv}\, .

1831:    \ee

1832: Using \refb{sadd3}, \refb{epm},

1833: \refb{e3.3} one can easily verify that

1834: \be\label{e3.4}

1835: e^{-\pi i ( N\wt \rho Q^2

1836: + \wt \sigma P^2/N +2\wt v Q\cdot P)} = 

1837: e^{-\pi i ( N\wt \rho'' Q^{\prime\prime 2}

1838: + \wt \sigma'' P^{\prime \prime 2}/N +2\wt v'' 

1839: Q''\cdot P'')}   \, ,

1840: \ee

1841: and

1842: \be\label{e3.5}

1843: d\wt\rho \, 

1844: d\wt\sigma \,

1845: d\wt v = d\wt\rho'' \, 

1846: d\wt\sigma'' \,

1847: d\wt v'' \, .

1848: \ee

1849: Furthermore, with the help of eq.\refb{ewtphitrs}

1850: one can show that\cite{0609109}

1851: \be\label{e3.6}

1852:    \wt\Phi(\wt \rho'',\wt\sigma'',\wt v'') = \wt\Phi(

1853:    \wt \rho,\wt\sigma,\wt v)\, .

1854: \ee

1855: If $\CC''$ denotes the image of $\CC$ under the map \refb{e3.3}

1856: then 

1857: eqs.\refb{e3.4}-\refb{e3.6} allow us to express

1858: \refb{sadd5} as

1859: \be\label{e3.7}

1860: d(\vec Q'',\vec P'') = {1\over N}\, \int _{\CC''} d\wt\rho'' \, 

1861: d\wt\sigma'' \,

1862: d\wt v'' \, e^{-\pi i ( N\wt \rho'' Q^{\prime\prime 2}

1863: + \wt \sigma'' P^{\prime \prime 2}/N +2\wt v Q''\cdot P'')}\, {1

1864: \over \wt\Phi(\wt \rho'',\wt \sigma'', \wt v'')}  \, .

1865: \ee

1866: To find the location of $\CC''$ we note that under the map

1867: \refb{e3.3} the real parts of $\wrh$, $\ws$ and $\wv$ mix

1868: among themselves and the imaginary parts of 

1869: $\wrh$, $\ws$ and $\wv$ mix among themselves.

1870: The initial contour $\CC$ corresponded to a unit cell of

1871: the cubic lattice in the

1872: ($\wrh_1$,$\ws_1$,$\wv_1$) space spanned by the basis vectors

1873: $(1,0,0)$, $(0,N,0)$ and $(0,0,1)$. The unimodular

1874: map \refb{e3.3} transforms this into a different unit cell

1875: of the same lattice. We can now use

1876: the 

1877: shift symmetries

1878: \be\label{eshift}

1879: \wt\Phi(\wt\rho,\wt\sigma,\wt v) =

1880: \wt\Phi(\wt\rho+1,\wt\sigma,\wt v) =

1881: \wt\Phi(\wt\rho,\wt\sigma+N,\wt v) =

1882: \wt\Phi(\wt\rho,\wt\sigma,\wt v+1)\, ,

1883: \ee

1884: which are manifest from \refb{edefwtphi}, to

1885: bring the integration region back to the original unit cell.

1886: Thus $\CC''$ and $\CC$ differ only in the values of the imaginary

1887: parts of $\wrh$, $\ws$ and $\wv$. Using \refb{ep2kk},

1888: \refb{e3.3} we see that for the contour $\CC''$,

1889: \ben \label{sadd7}

1890:     \wt \rho''_2 =\tilde \alpha^2\, M_1 +\tilde \beta^2\, M_2 + 2\tilde 

1891:    \alpha

1892:    \tilde \beta \, M_3 \, ,

1893:    \nonumber \\

1894:     \wt\sigma''_2 =\tilde \gamma^2\, M_1 +\tilde \delta^2 \, M_2 + 

1895:    2\tilde

1896:    \gamma\tilde \delta\, M_3 \, ,

1897:    \nonumber \\

1898:     v''_2 = -\tilde \alpha\tilde \gamma\, M_1 - \tilde \beta\tilde 

1899:    \delta \, M_2 - (\tilde \alpha\tilde \delta + \tilde 

1900:    \beta\tilde \gamma)\, M_3 \, .

1901: \een

1902: Thus $\CC''$ is not identical to $\CC$, -- a fact first noticed in

1903: \cite{appear}.

1904: We could try to deform $\CC''$ back to $\CC$, but in that process

1905: we might pick up contribution from the residues at the poles of

1906: $\wt\Phi(\wt \rho'',\wt \sigma'', \wt v'')$. Thus we see that the

1907: degeneracy formula \refb{e3.7} for $d(\vec Q'', \vec P'')$ is not

1908: obtained by simply replacing $(\vec Q, \vec P)$ by

1909: $(\vec Q'', \vec P'')$ in the

1910: expression for $d(\vec Q, \vec P)$. The integration contour $\CC$

1911: also gets deformed to a new contour $\CC''$.

1912: 

1913: Let us now analyze the region $\RR''$

1914: of the asymptotic moduli space in

1915: which \refb{e3.7} is valid. This is obtained by taking the image

1916: of the region $\RR$ under the transformation \refb{sadd4}.

1917: To determine this region we need to first study the images of the

1918: curves described in \refb{esm4} in the

1919: $a_\infty-S_\infty$ plane. A straightforward analysis shows that

1920: the image of \refb{esm4} is described by the curve

1921: \be \label{esm4rprp}

1922: \left(a_\infty - {a''d''+b''c''\over 2c''d''}\right)^2 

1923: + \left( S_\infty +{E''\over 2 c''d''}

1924: \right)^2 = {1\over 4 c^{\prime\prime 2} d^{\prime\prime 2}} (1 + 

1925: E^{\prime\prime 2})\, ,

1926: \ee

1927: where

1928: \be \label{sadd8}

1929: \pmatrix{a'' & b'' \cr c'' & d''} = \pmatrix{\alpha & \beta\cr \gamma

1930: & \delta} \pmatrix{a & b\cr c & d}\, ,

1931: \ee

1932: and

1933: \be \label{esm5rprp}

1934: E'' \equiv { c''d'' (Q^{\prime\prime 

1935: T} (M_\infty + L) Q'')

1936: + a''b'' (P^{\prime\prime  T} (M_\infty + L) P'') 

1937:  -(a''d'' + b''c'') (P^{\prime\prime

1938: T} (M_\infty + L) Q'') \over 

1939: \left[ (Q^{\prime \prime T} (M_\infty + L) Q'') 

1940: (P^{\prime \prime T} (M_\infty + L) 

1941: P'')

1942: - (P^{\prime \prime T} (M_\infty + L) Q'')^2\right]^{1/2}

1943: }\, .

1944: \ee

1945: It can be easily seen that $\pmatrix{a'' & b'' \cr c'' & d''}$ satisfy

1946: the relations \refb{enorm}, \refb{egen}, and the equivalence

1947: relations \refb{escale1}, \refb{esc12} translate to identical

1948: equivalence relations on $\pmatrix{a'' & b'' \cr c'' & d''}$.

1949: Thus the collection of all the

1950: matrices $\pmatrix{a'' & b'' \cr c'' & d''}$ describes

1951: same set $\AAA$ as the collection of all the

1952: matrices $\pmatrix{a & b\cr c & d}$.

1953: 

1954: Now recall that the original region $\RR$ was bounded by a

1955: set of marginal stability walls $\left\{\pmatrix{a & b\cr

1956: c & d}\right\}\in\BB_R$. 

1957: Thus the region $\RR''$ 

1958: is bounded by the collection of walls described by  

1959: \refb{esm4rprp}, \refb{esm5rprp} with

1960: $\pmatrix{a'' & b''\cr c'' & d''}\in \pmatrix{\alpha & \beta\cr

1961: \gamma & \delta}\BB_R$.

1962: 

1963:  At this stage it will be instructive to 

1964:  compare the expressions for the degeneracies 

1965: $d(\vec Q,\vec P)$ and $d(\vec Q'', \vec P'')$. There are two

1966: key differences. First of all although both are represented

1967: by similar looking integrals, the contours of integration are

1968: different in the two cases. Second the region of validity $\RR''$

1969: of

1970: the expression for $d(\vec Q, \vec P)$ is not obtained by simple

1971: replacement of $(\vec Q, \vec P)\to (\vec Q'',\vec P'')$ in the

1972: expression for $d(\vec Q, \vec P)$; 

1973: the original region of validity of our formula

1974: was bounded by a set of marginal stability

1975: walls corresponding to a set of matrices

1976: $\pmatrix{a & b\cr c & d}\in\BB_R $, whereas the new region

1977: of validity of the formula is bounded by 

1978: a set of walls corresponding

1979: to the matrices

1980: $ \pmatrix{a'' & b''\cr c'' & d''}\in \pmatrix{\alpha & \beta\cr

1981: \gamma & \delta}\BB_R$.

1982: 

1983: Typically the new charge vectors $(\vec Q'', \vec P'')$ are not of

1984: the type \refb{echvec}, but in some cases 

1985: it is possible to obtain the

1986: charge vectors $(\vec Q'', \vec P'')$ by T-duality transformation of

1987: another charge vector $(\vec {\wt Q}, \vec {\wt P})$ 

1988: of the form

1989: given in \refb{echvec} satisfying

1990:  \be\label{eqdefrp}

1991: \wt Q^2 = Q^{\prime\prime 2}, \qquad \wt 

1992: P^2 = P^{\prime\prime 2}, \qquad \wt Q\cdot\wt P

1993: = Q''\cdot P''\, .

1994: \ee

1995:  Then for $d(\vec {\wt Q}, \vec {\wt P})$ our

1996: original formula for degeneracy holds.

1997: Using the results given in \refb{egg1intrp}-\refb{esm5rp}

1998: we can now conclude that in the region $\wt\RR''$ 

1999: bounded by the walls \refb{esm5rprp}  with

2000: \be \label{eww1}

2001: \pmatrix{a'' & b''\cr c'' & d''} \in \BB_R

2002: \, ,

2003: \ee

2004: the

2005: degeneracy $d(\vec Q'', \vec P'')$ is given by 

2006:  \refb{e3.7}

2007: with the contour $\CC''$ replaced by $\CC$. 

2008: 

2009: Thus we now have expressions for $d(\vec Q'', \vec P'')$ in two

2010: different domains, -- $\wt \RR''$ and $\RR''$. In both regions the

2011: degeneracy is given by an integral. The integrand in both cases are

2012: same, but in one case the integration contour is $\CC$ while in

2013: the other case it is $\CC''$.

2014: This shows that

2015: as we cross the walls of marginal stability to move from the

2016: region $\wt\RR''$ to $\RR''$ the expression

2017: for $d(\vec Q'', \vec P'')$ changes by a modification in the location

2018: of the contour of integration.  

2019: 

2020: Even though the result

2021: was derived under the assumption that $(\vec Q'', \vec P'')$

2022: can be related to a charge vector of the type \refb{echvec} by

2023: T-duality, it is natural to assume that this phenomenon is more

2024: general. In particular, a natural postulate will be that

2025: {\it as we move from the domain 

2026: $\RR$ corresponding

2027: to the set $\BB_R$ to

2028: another domain corresponding to the set $\pmatrix{\alpha & \beta

2029: \cr \gamma & \delta}\BB_R$, with $\pmatrix{\alpha &\beta\cr

2030: \gamma & \delta}\in\Gamma_1(N)$, 

2031: the expression for $d(\vec Q, \vec P)$

2032: gets modified by a replacement of the integration contour

2033: $\CC$ to the contour $\CC''$ corresponding to

2034: \ben \label{sadd7rp}

2035:     \wt \rho_2 =\tilde \alpha^2\, M_1 +\tilde \beta^2\, M_2 + 2\tilde 

2036:    \alpha

2037:    \tilde \beta \, M_3 \, ,

2038:    \nonumber \\

2039:     \wt\sigma_2 =\tilde \gamma^2\, M_1 +\tilde \delta^2 \, M_2 + 

2040:    2\tilde

2041:    \gamma\tilde \delta\, M_3 \, ,

2042:    \nonumber \\

2043:     v_2 = -\tilde \alpha\tilde \gamma\, M_1 - \tilde \beta\tilde 

2044:    \delta \, M_2 - (\tilde \alpha\tilde \delta + \tilde 

2045:    \beta\tilde \gamma)\, M_3 \nonumber \\

2046:     0\le  \wt\rho_1\le 1, \quad

2047: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, ,

2048: \een

2049: $M_1$, $M_2$ and $M_3$ being large but fixed positive

2050: numbers with $M_3<< M_1, M_2$ and

2051: $ \wt\alpha=\delta$, $\wt\beta = \gamma/N$, $\wt\gamma

2052:    =N\beta$, $\wt\delta = \alpha$.}

2053: 

2054: 

2055: Can every domain be related to $\RR$ or $\LL$ via S-duality

2056: transformation? 

2057: If this is so then we can

2058: use duality invariance together with the information about

2059: the  contours $\CC$ and $\wh\CC$ appropriate for

2060: $\RR$ and $\LL$, -- as given in \refb{ep2kk} 

2061: and \refb{ecchat},

2062: --  to find out the location of the integration contour in

2063: every other domain. For $N=1,2,3$ we can answer this question

2064: in the affirmative as  follows.  In these cases the matrix

2065: $\pmatrix{a & b\cr c & d}$ labelling any wall can be

2066: represented by an element of $\Gamma_1(N)$ and hence

2067: can be 

2068: related to the matrix $\pmatrix{1 & 0\cr 0 & 1}$ via a

2069: duality transformation of the form \refb{sadd8}.  Since

2070: $\pmatrix{1 & 0\cr 0 & 1}$ describes the boundary between

2071: the domains $\RR$ and $\LL$, 

2072: the two domains bordering the wall $\pmatrix{a & b\cr c & d}$

2073: must be related to the domains $\RR$ and $\LL$ 

2074: by the same duality transformation. Since this can be done for

2075: every wall of marginal stability, we see that every domain

2076: must be related to either $\RR$ or $\LL$ via a duality

2077: transformation.

2078: 

2079: For $N\ge 4$ not all the walls can be related to the wall

2080: $\pmatrix{1 & 0\cr 0 & 1}$ via duality transformation.

2081: Nevertheless it may  still be possible to

2082: relate all the domains to $\RR$ and $\LL$ via duality 

2083: transfomation, with the walls not related to 

2084: $\pmatrix{1 & 0\cr 0 & 1}$ being related to some other wall

2085: of $\RR$ or $\LL$. We shall not attempt to settle this

2086: issue here.

2087: 

2088: 

2089: Given two contours on two sides of a marginal

2090: stability wall,

2091: if we try to deform one to the other then typically

2092: we shall encounter poles of the integrand, and the two results

2093: will differ by the sum of the residues at these poles. To get a physical

2094: insight into how much this change is as we move across a

2095: given wall, let us consider the wall

2096: corresponding to the matrix $\pmatrix{1 & 0\cr 0 & 1}$. 

2097: This is the wall separating the regions 

2098: $\RR$ and $\LL$, and 

2099: across this wall the original dyon 

2100: becomes unstable against

2101: decay into states with charges 

2102: $(\vec Q, 0)$ and $(0,\vec P)$.

2103: Since across this wall the contour changes from constant

2104: negative $\wt v_2$ to constant

2105: positive $\wt v_2$, we pick up the residue at the pole

2106: $\wv=0$. Now near $\wt v=0$ the function $\wt\Phi$ has the

2107: behaviour\cite{0510147,0602254,0607155,0609109}

2108: \be \label{enearpole}

2109: \wt\Phi(\wrh,\ws,\wv) \simeq -4\pi^2 \wt v^2 \, f(N\wt\rho) 

2110: g(\wt\sigma/N) + \OO(\wv^4)

2111: \, ,

2112: \ee

2113: where $f(\tau)$ and $g(\tau)$ are two functions which have the

2114: interpretation of inverse of partition functions associated with

2115: electric and magnetic half BPS states of the 

2116: theory.

2117: Performing the integral over $\wt v$ in \refb{egg1int}

2118: around the $\wt v=0$ point now gives

2119: \be \label{echange}

2120: -(Q\cdot P) \left\{\int_0^1 d\wrh e^{-i\pi  N \wrh Q^2} 

2121: \left( f(N\wrh)\right)^{-1}\right\}  \left\{{1\over N} \int_0^N

2122: d\ws e^{-i \pi  \ws P^2/N} 

2123: \left( g(\ws/N)\right)^{-1}\right\}\, .

2124: \ee

2125: This formula can be given a simple physical interpretation.

2126: The second and the third factors represent the

2127: degeneracies of the electric and magnetic half BPS states

2128: into which the original dyon decays on the marginal stability

2129: wall. The $Q\cdot P$ factor on the other hand is associated

2130: with the supersymmetric quantum mechanics describing the

2131: relative motion of the electric and the magnetic 

2132: system\cite{0605210}.

2133: In particular it represents the number of states whose binding

2134: energy vanishes as we reach the marginal

2135: stability wall\cite{pope,9912082}. Thus this is the number

2136: of states which disappear from the spectrum as we cross the

2137: wall.

2138: 

2139: 

2140: 

2141: \sectiono{The Large Charge Limit} \label{sblack}

2142: 

2143: In this section we shall argue that even though the complete

2144: spectrum changes discontinuously when the asymptotic value

2145: of the axion field changes sign, the large charge expansion

2146: is not affected by this change. 

2147: Our starting point is the result derived in

2148: \cite{0605210,0607155,0609109} 

2149: that the poles in the integrand in the expression for 

2150: $d(\vec Q,\vec P)$ come

2151: from the  second order zeroes of $\wt\Phi$  at\footnote{For $\MM=K3$

2152: this result can be found in appendix E of \cite{0605210}. For

2153: the general case the set of all the zeroes and poles of $\wt\Phi$ were

2154: listed in \cite{0609109}, but we did not attempt to separate out

2155: the zeroes from the poles. A careful analysis shows that the only

2156: zeroes come from the set \refb{eonly}.}

2157: \ben\label{eonly}

2158: &&   n_2 ( \ws \wrh  -\wv ^2) + j\wv  + 

2159: n_1 \ws  -m_1 \wrh  + m_2

2160:  =0\, , \nonumber \\

2161:  && 

2162: m_1 n_1 + m_2 n_2 +\frac{j^2}{4} = {1\over 4} \, ,\nonumber \\

2163: && m_1\in N\ZZZ, 

2164: \quad n_1, m_2, n_2 \in \ZZZ, \quad j\in 2\ZZZ+1

2165: \, . 

2166: \een

2167: For large charges

2168: the leading contribution to the degeneracy comes from

2169: the pole 

2170: at $n_2=1$\cite{9607026,0412287,0510147,0605210,

2171: 0607155,0609109}. Contribution from the poles with

2172: $n_2\ne 1$ are exponentially suppressed compared to the

2173: leading contribution.

2174: 

2175: Consider now the contours $\CC$ and $\CC''$ given in

2176: \refb{ep2kk} and \refb{sadd7rp} respectively. 

2177: Both contours have the same range of integration

2178: over the real parts of $\wrh$, $\ws$ and $\wv$

2179: \be \label{erho1}

2180: 0\le\wrh_1<1, \qquad 0\le\ws_1< N, \qquad

2181: 0\le \wv<1\, .

2182: \ee

2183: However they

2184: differ in the values of $\wrh_2$, $\ws_2$ and $\wv_2$.

2185: We have for

2186: \ben\label{etwocontour}

2187: \CC &:&  \wrh_2=M_1, \quad \ws_2=M_2, \quad \wv_2=-M_3\, ,

2188: \nonumber \\

2189: \CC'' &:&  \wrh_2=\tilde\alpha^2\, M_1 

2190: +\tilde \beta^2\, M_2 + 2\tilde 

2191:    \alpha

2192:    \tilde \beta \, M_3, \quad \ws_2=\tilde \gamma^2\, M_1 

2193:    +\tilde \delta^2 \, M_2 + 

2194:    2\tilde

2195:    \gamma\tilde \delta\, M_3, \nonumber \\

2196:    && \wv_2=

2197:    -\tilde \alpha\tilde \gamma\, M_1 - \tilde \beta\tilde 

2198:    \delta \, M_2 - (\tilde \alpha\tilde \delta + \tilde 

2199:    \beta\tilde \gamma)\, M_3 \, ,

2200: \een

2201: where $M_1$, $M_2$ and $M_3$ are fixed 

2202: numbers with $M_1,M_2>>1$ and

2203: $|M_3|<<M_1, M_2$.

2204: Our goal will be to show that we can deform the contour $\CC$

2205: to $\CC''$ without hitting any of the poles given in

2206: \refb{eonly} except those with $n_2=0$. Since for large charges

2207: the contribution

2208: to the degeneracy comes from the poles at $n_2=1$

2209: up to

2210: exponentially suppressed terms, this will show that

2211: the change in the degeneracy due to the change in the contour

2212: of integration is exponentially suppressed compared to the

2213: leading contribution.

2214: 

2215: To proceed we shall choose a specific path along

2216: which we deform the contour. We take this to be along the

2217: straight line

2218: joining the points $(M_1, M_2, M_3)$ and 

2219: $(\tilde\alpha^2\, M_1 

2220: +\tilde \beta^2\, M_2 + 2\tilde 

2221:    \alpha

2222:    \tilde \beta \, M_3, \tilde \gamma^2\, M_1 

2223:    +\tilde \delta^2 \, M_2 + 

2224:    2\tilde

2225:    \gamma\tilde \delta\, M_3,

2226: -\tilde \alpha\tilde \gamma\, M_1 - \tilde \beta\tilde 

2227:    \delta \, M_2 - (\tilde \alpha\tilde \delta + \tilde 

2228:    \beta\tilde \gamma)\, M_3)$ in the $(\wrh_2, \ws_2,\wv_2)$

2229:    space. Points on this line can be parametrized by a real

2230:    number $\lambda$ lying between 0 and 1, with

2231: \ben \label{estrai}

2232: \wrh_2 &=& M_1 + \lambda\, \{ (\tilde\alpha^2-1)\, M_1 

2233: +\tilde \beta^2\, M_2 + 2\tilde 

2234:    \alpha

2235:    \tilde \beta \, M_3\} \nonumber \\

2236: \ws_2 &=& M_2 + \lambda \, \{ \tilde \gamma^2\, M_1 

2237:    +(\tilde \delta^2-1) \, M_2 +

2238:    2\tilde

2239:    \gamma\tilde \delta\, M_3\} \nonumber \\

2240: \wv_2 &=& -M_3 +\lambda \, \{ -\tilde \alpha\tilde \gamma\, M_1 

2241: - \tilde \beta\tilde 

2242:    \delta \, M_2 - (\tilde \alpha\tilde \delta + \tilde 

2243:    \beta\tilde \gamma-1)\, M_3\}\, .

2244: \een

2245: For each $\lambda$ we take the real parts of $\wrh$, $\ws$ and $\wv$ 

2246: to be

2247: in the range \refb{erho1}. 

2248: 

2249: We want to show that \refb{eonly} and \refb{estrai} have no

2250: simultaneous solutions except for $n_2=0$. For this we write down

2251: separately the real and imaginary parts of eq.\refb{eonly}, as

2252: well as the constraint on $\vec m$, $\vec n$, $j$:

2253: \be \label{esep1}

2254: \ws_2 \wrh_2 - \wv_2^2 - \ws_1\wrh_1 + \wv_1^2

2255: -{j\over n_2} \wv_1 - {n_1\over n_2} \ws_1

2256: + {m_1\over n_2} \wrh_1 -{m_2\over n_2} = 0\, ,

2257: \ee

2258: \be \label{esep2} 

2259: n_2 (\ws_1 \wrh_2 + \ws_2 \wrh_1 -2 \wv_1 \wv_2)

2260: + j\wv_2 +n_1\ws_2-m_1 \wrh_2 = 0\, ,

2261: \ee

2262: \be \label{esep3}

2263: m_1 n_1 + m_2 n_2 +{j^2\over 4}={1\over 4}\, .

2264: \ee

2265: We now eliminate $m_1$ and $m_2$ using \refb{esep2}, 

2266: \refb{esep3}

2267: to write  \refb{esep1} as

2268: \be \label{esep4}

2269:  (\ws_2 \wrh_2 - \wv_2^2) + \left[\pmatrix{j & n_1}\, A

2270: \, \pmatrix{j\cr n_1} + aj+ bn_1 + c\right]=0\, ,

2271: \ee

2272: where

2273: \be \label{esep5}

2274: A = {1\over 4 n_2^2} 

2275:  \pmatrix{1 &  {2\wv_2/

2276: \wrh_2} \cr 

2277: 2\wv_2/

2278: \wrh_2 & {4\ws_2/\wrh_2}}\, ,

2279: \ee

2280: \be \label{esep6}

2281: a = {1\over  n_2}\, \left({\wv_2 \over \wrh_2}\wrh_1 

2282: -\wt v_1\right), \quad

2283: b = 2{\ws_2\over n_2\wrh_2} \left(\wrh_1 - {\wv_2\over \ws_2}\wv_1

2284: \right), \quad c = v_1^2   

2285: + \wrh_1 \left( {\ws_2\over \wrh_2} \wrh_1

2286: - 2{\wv_2\over \wrh_2} \wv_1 \right)-{1\over 4 n_2^2}\,.

2287: \ee

2288: Now for our contour $\wrh_1$, $\ws_1$ and $\wv_1$ are always finite.

2289: For $M_1$ and $M_2$  large and of the same order with 

2290: $|M_3|<<M_1,M_2$,  the quantities

2291: $\wrh_2$ and $\ws_2$  given in \refb{estrai} always remain large

2292: and positive, and the ratios $\ws_2/\wrh_2$, $\wv_2/\wrh_2$,

2293: $\wv_2/\ws_2$ etc. are bounded from above by finite numbers.

2294: Thus the quantities $a$, $b$ and $c$ remain finite.

2295: On the other hand in this limit we have 

2296: \be \label{esep7}

2297: \wrh_2\ws_2 - \wv_2^2 \simeq \lambda (1-\lambda) 

2298: \left( \tilde\gamma^2 M_1^2 + \tilde\beta^2 M_2^2 \right)

2299: + \left( \lambda^2 + (1-\lambda)^2 +  \lambda (1-\lambda)

2300: (\tilde\alpha^2 +\tilde\delta^2)\right) M_1 M_2\, .

2301: \ee

2302: In the range $0\le\lambda\le 1$ each term in this expression is

2303: non-negative, and  \refb{esep7}

2304: remains large and positive, -- of order $M_1M_2$, -- 

2305: in the limit

2306: of large $M_1$, $M_2$. This also shows that the matrix

2307: $A$ defined in \refb{esep5} is nondegenerate in this limit and in

2308: fact has finite positive eigenvalues. As a result the term in the square

2309: bracket in \refb{esep4} is bounded from below by a finite number

2310: \be \label{efinite}

2311: c- {1\over 4} \pmatrix{a & b} \, A^{-1}\, \pmatrix{a\cr b}\, ,

2312: \ee 

2313: and can never cancel the first term in \refb{esep4}

2314: for any value of $j$ and $n_1$

2315: for sufficiently large $M_1$ and $M_2$.

2316: Hence \refb{esep4} cannot be satisfied.

2317: This shows that it is impossible to find a similtaneous solution to the

2318: eqs.\refb{eonly} and \refb{estrai} for $n_2\ne 0$, 

2319: and hence the contour $\CC$ can

2320: be deformed to $\CC''$ along the path \refb{estrai} without

2321: encountering the poles of the integrand given in \refb{eonly}

2322: for $n_2\ne 0$.

2323: 

2324: 

2325: This establishes that in the limit of large charges the degeneracy

2326: remains  the same in different domains in the moduli space up

2327: to nonperturbative terms.

2328: This result is consistent with the fact that for a black hole of

2329: charge $(\vec Q, \vec P)$ in this theory  we have a

2330: stable supersymmetric attractor for $P^2>0$, $Q^2>0$,

2331: $P^2 Q^2 > (Q\cdot P)^2$. Thus the near horizon geometry of these

2332: black holes is always given by this attractor point and is independent

2333: of the asymptotic moduli even if this requires the attractor flow

2334: to cross one or more walls of marginal stability.

2335: 

2336: \sectiono{Test of S-duality Invariance} \label{stest}

2337: 

2338: In \S\ref{sdualtwo} we used S-duality invariance to determine the

2339: locations of the integration contour in the degeneracy formula in

2340: different domains in the moduli space. However this did not provide

2341: a test of S-duality. In this section we shall describe some tests of

2342: S-duality that one could perform.

2343: \begin{enumerate}

2344: \item If there is an S-duality transformation that leaves the set

2345: $\BB_R$ invariant, then under such a transformation the contour

2346: $\CC$ either should not transform, or should transform to another

2347: contour that is continuously deformable to $\CC$ without passing

2348: through any poles.

2349: \item Analysis of \cite{0605210} has shown that inside the 

2350: left domain

2351: $\LL$ corresponding to the set of matrices $\BB_L$, the degeneracy

2352: is given by performing integration over the

2353: contour $\wh\CC$ described in \refb{ecchat}.

2354: Thus if there is an S-duality transformation that maps the set

2355: $\BB_R$ to the set $\BB_L$ then such a transformation must

2356: map the contour $\CC$ to the contour $\wh\CC$ or another

2357: contour deformable to $\wh\CC$ without passing through

2358: any pole.

2359: \end{enumerate}

2360: In fact for all values of $N$ 

2361: there is an S-duality transformation that maps

2362: $\BB_R$ to $\BB_L$. It is given by

2363: \be \label{egiven}

2364: \pmatrix{\alpha & \beta \cr \gamma & \delta}

2365: = \pmatrix{1 & -1\cr 0 & 1}\, .

2366: \ee

2367: To see this we note that it maps the $\pmatrix{1 & 0\cr 0 & 1}

2368: \in\BB_R$ to $\pmatrix{1 & -1\cr 0 & 1}\in \BB_L$ and

2369: $\pmatrix{1 & 1\cr 0 & 1}

2370: \in\BB_R$ to $\pmatrix{1 & 0\cr 0 & 1}\in \BB_L$. Since

2371: two domains

2372: sharing two common boundaries must be identical, the action

2373: of \refb{egiven} must map $\BB_R$ to $\BB_L$.

2374: Using this transformation we can convert all tests of the first type into

2375: tests of the second type; all we need to do is to left multiply the duality

2376: transformation preserving $\BB_R$ by \refb{egiven} to construct

2377: a duality transformation that maps $\BB_R$ to $\BB_L$.

2378: 

2379: 

2380: We shall now try to verify that 

2381: the transformation \refb{egiven} maps $\CC$ to $\wh\CC$ or a 

2382: contour deformable to $\wh\CC$ without passing through any poles.

2383: Using \refb{epm}, \refb{sadd7rp} we see that this 

2384: transformation maps the

2385: contour $\CC$ given in \refb{ep2kk} to

2386: \ben \label{emaps}

2387: \wrh_2 = M_1, \qquad \ws_2 = N^2 M_1 + M_2 - 2 N M_3, \qquad

2388: \wv_2 = NM_1 -M_3\, , \nonumber \\

2389: 0\le  \wt\rho_1\le 1, \quad

2390: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, .

2391: \een

2392: This is different from $\wh\CC$ given in

2393: \refb{ecchat} for which $(\wrh_2,\ws_2,\wv_2)

2394: =(M_1,M_2,M_3)$. Thus we need to verify that

2395: we can deform the contour \refb{emaps} to

2396: $\wh\CC$  without encountering

2397: any pole. {}From the analysis in \S\ref{sblack} we already know that

2398: the poles at \refb{eonly} for $n_2\ne 0$ are not encountered; thus

2399: we need to look for poles with $n_2=0$. They occur at

2400: \be\label{eonlysp}

2401: \left( j\wv  + 

2402: n_1 \ws  -m_1 \wrh  + m_2

2403: \right)=0\, , 

2404: \ee

2405: with 

2406: \be \label{enra}

2407: m_1\in N\ZZZ,

2408: \quad n_1, m_2  \in \ZZZ, \quad j\in 2\ZZZ+1, \quad

2409: m_1 n_1   +\frac{j^2}{4} = {1\over 4}\, . \nonumber \\

2410: \ee

2411: Taking the imaginary part of eq.\refb{eonlysp} we get

2412: \be \label{eks1}

2413:  j\wv_2  + 

2414: n_1 \ws_2  -m_1 \wrh_2 = 0\, .

2415: \ee

2416: For fixed $j$, $n_1$, $m_1$

2417: this describes a plane in the $(\wrh_2,\ws_2,\wv_2)$ space.

2418: Our job is to show that points in the $(\wrh_2,\ws_2,\wv_2)$ space

2419: given in \refb{ecchat} and \refb{emaps} lie on the

2420: same side of this plane so that we can deform them to each other

2421: without going through this plane. For this we need to show that

2422: the left-hand side of \refb{eks1}, evaluated at \refb{ecchat} and

2423: \refb{emaps} have the same sign:

2424: \be \label{eks2}

2425: (jM_3 + n_1 M_2 -m_1 M_1) \left(

2426: j (NM_1 -M_3) + n_1 (N^2 M_1 + M_2 - 2 N M_3)

2427: - m_1 M_1

2428: \right) >0\, .

2429: \ee

2430: We can simplify the analysis by setting $M_1=M_2=M$. In this

2431: case the left hand side of \refb{eks2} has the form

2432: \ben \label{eks3}

2433: &&  (n_1 - m_1) \{Nj + (N^2 + 1) n_1 - m_1\} M^2 

2434: \nonumber \\

2435: && + \{(Nj + (N^2 + 1) n_1 - m_1) j

2436:  -(n_1 - m_1) ( j + 2 N n_1) \} \, M \, M_3 \nonumber \\

2437: && - j (j+2Nn_1) M_3^2 \, .

2438: \een

2439: Using \refb{enra} the coefficient of the $M^2$ term can be

2440: brought to the form

2441: \be \label{em2}

2442: \left( {Nj\over 2} + n_1 - m_1\right)^2 + N^2 n_1^2 - 

2443: {N^2\over 4}\, .

2444: \ee

2445: If $n_1\ne 0$ then this is strictly positive. If $n_1=0$ then

2446: from \refb{enra} we have $j=\pm 1$. We shall choose $j=1$ by

2447: using the freedom of changing the signs of $n_1$, $m_1$,

2448: $n_2$, $m_2$ and $j$

2449: simultaneously without changing the location of the pole.

2450: Since $m_1\in N\ZZZ$,

2451: in this case \refb{em2} is

2452: strictly positive if $m_1\ne 0,N$, and vanishes for $m_1=0,N$.

2453: 

2454: Now as long as \refb{em2} is non-vanishing and positive,

2455: we can make the first term of \refb{eks3} dominate over others

2456: by taking $M$ to be arbitrarily large, and hence \refb{eks3}

2457: is positive as required. Thus we only need to worry

2458: about is the case $n_1=0$, $j=1$, $m_1=0,N$ when the order

2459: $M^2$ term vanishes. In both cases \refb{eks3} takes the

2460: form:

2461: \be \label{em3}

2462: N\, M\, M_3 - M_3^2\, .

2463: \ee

2464: Since $M$ is large and positive, and $M>>M_3>0$, this is strictly

2465: positive. This shows that \refb{eks2} holds for all

2466: $j$, $m_1$, $n_1$ and hence we do not encounter any pole while

2467: deforming the contour \refb{emaps} to \refb{ecchat}.

2468: 

2469: For the special case of $N=1$ one can consider another map that

2470: takes us from the set $\BB_R$ to $\BB_L$. It is

2471: \be \label{em4}

2472: \pmatrix{\alpha & \beta \cr \gamma & \delta} =

2473: \pmatrix{0 & 1\cr -1 & 0}\, .

2474: \ee

2475: Under this map the vertices 0, 1 and $i\infty$ of the domain $\RR$

2476: get mapped to the points $i\infty$, $-1$ and $0$ respectively.

2477: The latter set is precisely the vertices of the domain $\LL$.

2478: 

2479: Using \refb{epm}, \refb{sadd7rp} 

2480: we see that this transformation maps the

2481: original contour to\cite{appear}

2482: \ben \label{em5}

2483: \wt \rho_2=M_2, \quad  \wt\sigma_2 = M_1, \quad

2484:  \wt v_2 = M_3, \nonumber \\

2485:  0\le  \wt\rho_1\le 1, \quad

2486: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, .

2487:  \een

2488:  This exactly coincides with \refb{ecchat} up to an exchange

2489:  of $M_1$ and $M_2$. Since exchange of $M_1$ and $M_2$

2490:  does not change the integral (we could take $M_1=M_2$)

2491:  we see that

2492:  this constraint of S-duality is satisfied trivially.

2493:  

2494:  Returning to the case of general $N$, one can identify the

2495:  following additional $\Gamma_1(N)$

2496:  transformation 

2497:  that maps the set $\BB_R$ to $\BB_L$:

2498:  \be \label{ede1}

2499:  \pmatrix{\alpha & \beta \cr \gamma & \delta}

2500:  = \pmatrix{1 & 0\cr -N & 1}\, .

2501:  \ee

2502:  This maps the contour $\CC$ to

2503:  \ben \label{ede2}

2504: \wrh_2 = M_1+M_2-2M_3, 

2505: \qquad \ws_2 = M_2, \qquad

2506: \wv_2 = M_2 -M_3\, , \nonumber \\

2507: 0\le  \wt\rho_1\le 1, \quad

2508: 0\le  \wt\sigma_1\le N, \quad 0\le  \wt v_1\le 1\, .

2509: \een

2510: Consistency with duality invariance requires that we must be

2511: able to deform this contour to $\wh\CC$ without encountering

2512: any pole.

2513: One can proceed to analyze this exactly in the same manner as

2514: we did for \refb{emaps} and arrive at the condition that 

2515: in order to be able to deform this contour to $\wh\CC$ 

2516: given in \refb{ecchat} we need

2517: the following quantity to be positive (analog of eq.\refb{eks3}):

2518: \ben \label{ede3}

2519: && (n_1-m_1) (n_1 - 2 m_1 +j)M^2 \nonumber \\

2520: && + \{ j(n_1 - 2m_1+j) + (n_1-m_1) (2 m_1-j)\} MM_3

2521: \nonumber \\

2522: && + j (2m_1 -j) M_3^2\, .

2523: \een

2524: Using \refb{enra} the coefficient of the $M^2$ term can be

2525: brought to the form

2526: \be \label{ede4}

2527: \left({j\over 2} + n_1 - m_1\right)^2 + m_1^2 -{1\over 4}\, .

2528: \ee

2529: If $m_1\ne 0$ it is strictly positive. For $m_1=0$ from

2530: \refb{enra} we have $j=1$. 

2531:  Since $n_1\in \ZZZ$,

2532: in this case \refb{ede4} is

2533: strictly positive if $n_1\ne 0,-1$, and vanishes for $n_1=0,-1$.

2534: As long as \refb{ede4} is non-vanishing and positive,

2535: we can make the first term of \refb{ede3} dominate over others

2536: by taking $M$ to be arbitrarily large, and hence \refb{ede3}

2537: is positive as required. Thus we only need to worry

2538: about is the case $m_1=0$, $j=1$, $n_1=0,-1$ when the order

2539: $M^2$ term vanishes. In both cases \refb{ede3} takes the

2540: form:

2541: \be \label{ede5}

2542: M\, M_3 - M_3^2\, .

2543: \ee

2544: Since $M$ is large and positive, and $M>>M_3>0$, this is strictly

2545: positive. This shows that \refb{ede3} is positive for all

2546: $j$, $m_1$, $n_1$ and hence we do not encounter any pole while

2547: deforming the contour \refb{ede2} to \refb{ecchat}.

2548: 

2549: If we consider the $\Gamma_1(N)$ element 

2550: \be \label{ede6}

2551: g_0= \pmatrix{1 & 1\cr 0 & 1} \pmatrix{1 & 0 \cr -N & 1}

2552: = \pmatrix{1-N & 1 \cr -N & 1}\, ,

2553: \ee

2554: then it clearly leaves the set $\BB_R$ unchanged since it is given

2555: by a map from $\BB_R$ to $\BB_L$ followed by the inverse of

2556: a map from $\BB_L$ to $\BB_R$.

2557: Our results establish that under this transformation 

2558: the contour $\CC$ is

2559: mapped to another contour that is continuously deformable to $\CC$.

2560: This is turn establishes that any power of $g_0$ will also have the same

2561: property. We can now follow this up with the transformation

2562: \refb{egiven} to construct a set of duality transformations that maps

2563: the set $\BB_R$ to $\BB_L$ and maps the contour $\CC$ to another

2564: contour deformable to $\wh \CC$, thereby providing a test of the

2565: corresponding duality transformtion. 

2566: For $N=1$ this includes in particular

2567: the element \refb{em4} considered earlier.

2568: 

2569: Can every element of $\Gamma_1(N)$ that preserves the set $\BB_R$

2570: be expressed as a positive or negative power of $g_0$? If so

2571: then our test of S-duality invariance of the degeneracy formula

2572: would be

2573: complete. To address this issue

2574: note that a $\Gamma_1(N)$ 

2575: transformation that maps $\BB_R$ to $\BB_R$ must take

2576: adjacent walls to adjacent walls. Furthermore the map must

2577: be orientation preserving. Thus the action of any such

2578: transformation on $\BB_R$ must preserve the cyclic ordering

2579: of the walls and vertices. The map $g_0$ 

2580: indeed has this property. It moves

2581: the walls and vertices by two steps clockwise, taking 0 to 1,

2582: $i\infty$ to $1 - {1\over N}$, 1 to $1 - {1\over N-1}$ etc.

2583: $g_0^{-1}$ causes a shift by two steps in the anti-clockwise

2584: direction. Powers of $g_0$ will move the walls clockwise or

2585: anti-clockwise by even number of steps. 

2586: Are there elements of $\Gamma_1(N)$ which

2587: shift the walls by odd number of steps? If so then by

2588: combining it with appropriate positive or negative powers of

2589: $g_0$ we can generate a transformation that

2590: shifts every wall of $\BB_R$ by

2591: one step in the clockwise direction. Such a move will map

2592: 0 to $i\infty$, $i\infty$ to 1 and 1 to $1-{1\over N}$. The unique

2593: SL(2,R) map that implements this is the matrix

2594: $\pmatrix{\sqrt N & -1/\sqrt N\cr \sqrt N & 0}$. This is 

2595: clearly not an element of $\Gamma_1(N)$ for any $N$ other

2596: than $N=1$.

2597: 

2598: This would seem to indicate that all elements of

2599: $\Gamma_1(N)$ preserving $\BB_R$ are generated by

2600: $g_0$. However there is an

2601: additional subtlety arising out of the fact that for $N> 4$

2602: the element $g_0$ has a pair of fixed points on the real line

2603: corresponding to\footnote{For $N=4$ there is a single fixed point

2604: and hence it does not divide the vertices into two sets. Thus every

2605: pair of vertices separated by even number of steps can still be

2606: related by a $g_0$ transformation.}

2607: \be \label{efi1}

2608: \tau_\infty={1\over 2} \left( 1 \pm \sqrt{1 -{4\over N}}\right)\, .

2609: \ee

2610: These provide accumulation points of the vertices of $\RR$; indeed

2611: if we begin with any vertex of $\RR$ 

2612: and apply $g_0$ or $g_0^{-1}$ transformation

2613: successively to generate other vertices they accumulate at the

2614: points given in \refb{efi1}. As a result there are infinite

2615: number of walls bordering $\RR$, and

2616: a wall or a vertex that is situated within the

2617: range

2618: \be \label{efi2}

2619: {1\over 2} \left( 1 - \sqrt{1 -{4\over N}}\right) <\tau_\infty

2620: < {1\over 2} \left( 1 + \sqrt{1 -{4\over N}}\right) \, ,

2621: \ee

2622: can never be related to a

2623: wall or a vertex lying outside this range by a $g_0$ transformation.

2624: This opens up the possibility that there may be additional

2625: elements of $\Gamma_1(N)$ which map the walls and vertices

2626: outside the range \refb{efi2} to walls and vertices

2627: inside this range and vice versa, preserving the cyclic ordering.

2628: 

2629: To examine this issue we need to first identify 

2630: some 

2631: vertices lying within the range \refb{efi2}. 

2632: For this we shall first prove that the point $1/2$ must be a vertex.

2633: If it is not a vertex then there must be a wall that goes over 1/2; by

2634: symmetry of the problem the vertices at the two ends of this wall must

2635: be

2636: situated symmetrically about 1/2. Let us take them to be

2637: \be \label{epq1}

2638: {1\over 2} \pm {p\over q} = {q\pm 2p\over 2q}\, ,

2639: \ee

2640: with $p,q$ relatively prime. This wall will correspond to a

2641: matrix $\pmatrix{a & b\cr c & d}$ with $a/c=(q-2p)/2q$ and

2642: $b/d = (q+2p)/2q$. Now we see that if $q$ is odd then 

2643: $(q\pm 2p)$ are odd and hence both $c$ and $d$ must be divisible

2644: by 2. This of course is incompatible with the relation $ad-bc=1$.

2645: If on the other hand $q$ is even (say $2m$) then $p$ must be

2646: odd and we can express \refb{epq1} as

2647: \be \label{epq2}

2648: {m\pm p\over 2m}\, .

2649: \ee

2650: Thus we have $a/c=(m-p)/2m$ and $b/d=(m+p)/2m$. $m$ must be

2651: odd so that $(m\pm p)$ are even; otherwise we again run into the

2652: problem of both $c$ and $d$ being even. Now since

2653: $c$ and $d$ cannot have a common factor, between $m+p$ and

2654: $m-p$ they should be able to cancel all the factors in $2m$. 

2655: This in particular will mean that their product

2656: $(m^2-p^2)$ must be divisible by $m$. Thus $p^2$ should be divisible

2657: by $m$. This is in contradiction with the assumption that $p$

2658: and $q$ (and hence $p$ and $m$) are relatively prime except for

2659: $m=1$. The latter corresponds to the wall connecting 0 and 1

2660: and is not relevant for us.

2661: 

2662: This shows that $1/2$ must be a vertex (or accumulation point

2663: of vertices) of $\RR$. We can now begin with

2664: ${1\over 2}$ and identify the vertices to the left and right

2665: of this using the analysis described below \refb{ebbyd}. For odd

2666: $N$ this gives three points

2667: \be\label{efi3}

2668: {1\over 2} - {1\over 2N}, \quad {1\over 2}, \quad {1\over 2}

2669: +{1\over 2N}\, ,

2670: \ee

2671: in ascending order. On the other hand for $N=2M$ with $M$ odd

2672: we get the three points to be

2673: \be\label{efi4}

2674: {1\over 2} - {1\over 2M}, \quad {1\over 2}, \quad {1\over 2}

2675: +{1\over 2M}\,.

2676: \ee

2677: If $N$ is a multiple of 4 then

2678: one can show that there is no wall ending

2679: at 1/2 and hence 1/2 must be an accumulation point. We shall not

2680: deal with this case here.

2681: We can now ask how $g_0$ acts on these vertices. For $N=5$

2682: and $N=6$ one can check that it moves the left-most vertex

2683: in the set \refb{efi3} or \refb{efi4} to the right-most vertex. By

2684: continuity we can conclude that in the range \refb{efi2},

2685: $g_0$ moves every vertex and wall by two steps in the

2686: {\it anti-clockwise} direction. There is no contradiction with

2687: the fact that outside the range \refb{efi2} $g_0$ moves the points

2688: in the clockwise direction since the boundaries of this region are

2689: fixed points of $g_0$. Thus every wall in the range \refb{efi2}

2690: can be related by $g_0$ action

2691: to one of the two walls connecting the vertices

2692: given in \refb{efi3} or \refb{efi4} for $N=5,6$.

2693: For $N\ge 7$ there are more than two

2694: vertices between a given vertex and its $g_0$ image and the situation

2695: is more complicated. 

2696: 

2697: Restricting ourselves to the cases $N=5,6$ we now ask the following

2698: question: is there a $\Gamma_1(N)$

2699: transformation that maps the vertices lying 

2700: inside the range \refb{efi2} to

2701: vertices outside

2702: the range \refb{efi2}? We focus on the possible action of this map

2703: on the three vertices 

2704: given in \refb{efi3} or

2705: \refb{efi4}. If such a transformation 

2706: exists then by left multiplying it with powers of

2707: $g_0$ we can always bring the three final points to

2708: 1, $i\infty$, 0 or $i\infty$, 0, $1/N$. Examining the two cases

2709: separately we find that there is no such $\Gamma_1(N)$ transformation

2710: mapping \refb{efi3} or

2711: \refb{efi4}  to 

2712: these points (although for odd $N$ there is a $\Gamma_0(N)$ 

2713: transformation

2714: $ \pmatrix{2 & -1\cr N & -(N-1)/ 2} $

2715: which takes the three points

2716: given in \refb{efi3} to 

2717: $i\infty$, 0 and $1/N$). 

2718: 

2719: This finally establishes that 

2720: for $N\le 6$, all the elements

2721: of $\Gamma_1(N)$ which preserve the set $\BB_R$ are obtained

2722: by taking positive or negative powers of $g_0$. Since we have

2723: checked that these transformations take the contour $\CC$ to

2724: another contour deformable to $\CC$ without encountering any

2725: poles, this completes our test of S-duality invariance for $N\le 6$.

2726: 

2727: 

2728: 

2729:  \bigskip

2730: 

2731: {\bf Acknowledgement:} I would like to thank Atish Dabholkar,

2732: Justin David, Tohru Eguchi, 

2733: Davide Gaiotto, Dileep Jatkar, Suresh Nampuri and

2734: Hirosi Ooguri for

2735: useful discussions. This work was supported  generously

2736: by the people of India, J.C. Bose fellowship of the

2737: Department of Science and Technology of Govt. 

2738: of India and Moore distinguished scholarship at

2739: California Institute of Technology.

2740: 

2741: 

2742: 

2743: \begin{thebibliography}{99}

2744: 

2745:  

2746: 

2747: \bibitem{0605210}

2748:   J.~R.~David and A.~Sen,

2749:   ``CHL dyons and statistical entropy function from D1-D5 system,''

2750:   arXiv:hep-th/0605210.

2751:   %%CITATION = HEP-TH 0605210;%%

2752: %\cite{Sahoo:2006rp}

2753: 

2754: \bibitem{0607155}

2755:   J.~R.~David, D.~P.~Jatkar and A.~Sen,

2756:   ``Dyon spectrum in N = 4 supersymmetric type II string theories,''

2757:   arXiv:hep-th/0607155.

2758:   %%CITATION = HEP-TH 0607155;%%

2759: %\cite{Astefanesei:2006dd}

2760: 

2761: \bibitem{0609109}

2762:   J.~R.~David, D.~P.~Jatkar and A.~Sen,

2763:   ``Dyon spectrum in generic N = 4 supersymmetric Z(N) orbifolds,''

2764:   arXiv:hep-th/0609109.

2765:   %%CITATION = HEP-TH 0609109;%

2766:   

2767: \bibitem{9607026}

2768: R.~Dijkgraaf, E.~P.~Verlinde and H.~L.~Verlinde,

2769: ``Counting dyons in N = 4 string theory,''

2770: Nucl.\ Phys.\ B {\bf 484}, 543 (1997)

2771: [arXiv:hep-th/9607026].

2772: %%CITATION = HEP-TH 9607026;%%

2773: 

2774: \bibitem{0510147}

2775:   D.~P.~Jatkar and A.~Sen,

2776:   ``Dyon spectrum in CHL models,''

2777:   JHEP {\bf 0604}, 018 (2006)

2778:   [arXiv:hep-th/0510147].

2779:   %%CITATION = HEP-TH 0510147;%%

2780: %\cite{Sen:2005iz}

2781: 

2782: \bibitem{0602254}

2783:   J.~R.~David, D.~P.~Jatkar and A.~Sen,

2784:   ``Product representation of dyon partition function in CHL models,''

2785:   JHEP {\bf 0606}, 064 (2006)

2786:   [arXiv:hep-th/0602254].

2787:   %%CITATION = HEP-TH 0602254;%%

2788: %\cite{Sahoo:2006vz}

2789: 

2790: \bibitem{0505094}

2791: D.~Shih, A.~Strominger and X.~Yin,

2792: ``Recounting dyons in N = 4 string theory,''

2793: arXiv:hep-th/0505094.

2794: %%CITATION = HEP-TH 0505094;%%

2795: 

2796: \bibitem{0506249}

2797: D.~Gaiotto,

2798: ``Re-recounting dyons in N = 4 string theory,''

2799: arXiv:hep-th/0506249.

2800: %%CITATION = HEP-TH 0506249;%%

2801: 

2802: \bibitem{0612011}

2803:   A.~Dabholkar and D.~Gaiotto,

2804:   ``Spectrum of CHL dyons from genus-two partition function,''

2805:   arXiv:hep-th/0612011.

2806:   %%CITATION = HEP-TH 0612011;%%

2807:   

2808: \bibitem{0603066}

2809:   A.~Dabholkar and S.~Nampuri,  

2810:   ``Spectrum of dyons and black holes in 

2811:   CHL orbifolds using Borcherds lift,''

2812:   arXiv:hep-th/0603066.

2813:   %%CITATION = HEP-TH 0603066;%%

2814: 

2815: \bibitem{9712211}

2816:   O.~Bergman,

2817:    ``Three-pronged strings and 1/4 BPS states in N=4 super-Yang-Mills

2818:   theory,''

2819:   Nucl.\ Phys.\ B {\bf 525}, 104 (1998)

2820:   [arXiv:hep-th/9712211].

2821:   %%CITATION = HEP-TH 9712211;%%

2822:   

2823: \bibitem{9804160}

2824:   O.~Bergman and B.~Kol,

2825:   ``String webs and 1/4 BPS monopoles,''

2826:   Nucl.\ Phys.\  B {\bf 536}, 149 (1998)

2827:   [arXiv:hep-th/9804160].

2828:   %%CITATION = NUPHA,B536,149;%%

2829: 

2830: \bibitem{appear}

2831: A.~Dabholkar, D.~Gaiotto and S.~Nampuri, private communications

2832: and to appear.

2833: 

2834: \bibitem{0010222}

2835: F.~Denef,

2836: ``On the correspondence between D-branes 

2837: and stationary supergravity solutions of type

2838: II Calabi-Yau compactifications'', 

2839: arXiv:hep-th/0010222.

2840: %%CITATION = HEP-TH 0010222;%%

2841: 

2842: \bibitem{0101135}

2843: F.~Denef,

2844: ``Split attractor flows and the spectrum of 

2845: BPS D-branes on the Quintic'', 

2846: arXiv:hep-th/0101135.

2847: %%CITATION = HEP-TH 0101135;%%

2848: 

2849: \bibitem{9207016}

2850:   J.~Maharana and J.~H.~Schwarz,

2851:   ``Noncompact symmetries in string theory,''

2852:   Nucl.\ Phys.\ B {\bf 390}, 3 (1993)

2853:   [arXiv:hep-th/9207016].

2854:   %%CITATION = HEP-TH 9207016;%%

2855: 

2856: \bibitem{9402002}

2857:   A.~Sen,

2858:   ``Strong - weak coupling duality in four-dimensional string theory,''

2859:   Int.\ J.\ Mod.\ Phys.\ A {\bf 9}, 3707 (1994)

2860:   [arXiv:hep-th/9402002].

2861:   %%CITATION = HEP-TH 9402002;%%

2862: 

2863: \bibitem{0508042}

2864:   A.~Sen,

2865:   ``Entropy function for heterotic black holes,''

2866:   JHEP {\bf 0603}, 008 (2006)

2867:   [arXiv:hep-th/0508042].

2868:   %%CITATION = JHEPA,0603,008;%%

2869: 

2870: \bibitem{9511222}

2871:   M.~Bershadsky, C.~Vafa and V.~Sadov,

2872:   ``D-Branes and Topological Field Theories,''

2873:   Nucl.\ Phys.\ B {\bf 463}, 420 (1996)

2874:   [arXiv:hep-th/9511222].

2875:   %%CITATION = HEP-TH 9511222;%%

2876: 

2877:  \bibitem{9507090}

2878:   M.~Cvetic and D.~Youm,

2879:   ``Dyonic BPS saturated black holes of heterotic string on a six torus,''

2880:   Phys.\ Rev.\ D {\bf 53}, 584 (1996)

2881:   [arXiv:hep-th/9507090].

2882:   %%CITATION = HEP-TH 9507090;%%

2883:  

2884:  \bibitem{9508094}

2885:   M.~J.~Duff, J.~T.~Liu and J.~Rahmfeld,

2886:   ``Four-Dimensional String-String-String Triality,''

2887:   Nucl.\ Phys.\ B {\bf 459}, 125 (1996)

2888:   [arXiv:hep-th/9508094].

2889:   %%CITATION = HEP-TH 9508094;%%

2890: 

2891: \bibitem{pope}

2892:  C.~N.~Pope,

2893:   ``Axial Vector Anomalies And The Index Theorem In Charged 

2894:  Schwarzschild And

2895:   Taub - Nut Spaces,''

2896:   Nucl.\ Phys.\ B {\bf 141}, 432 (1978).

2897:   %%CITATION = NUPHA,B141,432;%% 

2898: 

2899:  \bibitem{9912082}

2900: J.~P.~Gauntlett, N.~Kim, J.~Park and P.~Yi,

2901:  ``Monopole dynamics and BPS dyons in 

2902: N = 2 super-Yang-Mills theories,''

2903:   Phys.\ Rev.\ D {\bf 61}, 125012 (2000)

2904:   [arXiv:hep-th/9912082].

2905:   %%CITATION = HEP-TH 9912082;%%

2906: 

2907: \bibitem{0412287}

2908: G.~L.~Cardoso, B.~de Wit, J.~Kappeli and T.~Mohaupt,

2909: ``Asymptotic degeneracy of dyonic N = 4 string states

2910: and black hole

2911: entropy,''

2912: JHEP {\bf 0412}, 075 (2004) [arXiv:hep-th/0412287].

2913: %%CITATION = HEP-TH 0412287;%%

2914: 

2915: \end{thebibliography}

2916: 

2917: 

2918: 

2919: \end{document}

2920: 

2921: \bye

2922: 

2923: 

2924: