1:  \documentclass[12pt]{article}

2:  \usepackage{amsmath,amsfonts,theorem}

3: \input epsf

4: \pagestyle{myheadings}

5:  \newcommand{\ntag}[1]{} %\tag{#1}}

6: \numberwithin{equation}{section}

7:  \newtheorem{proof}{Proof}

8:  \newtheorem{prop}{Proposition}

9:  \newtheorem{pro}{Proposition}

10:  \renewcommand{\thepro}{}

11:  \newtheorem{cor}{Corollary}

12:  \newtheorem{co}{Corollary}

13:  \renewcommand{\theco}{}

14:  \newtheorem{thm}{Theorem}

15:  \newtheorem{lem}{Lemma}

16: 

17:  \theorembodyfont{\rmfamily}

18:  \newtheorem{dfn}{Definition}

19:  \newtheorem{mcdn}{Main Condition}

20:  \newtheorem{warning}{Warning}

21:  \renewcommand{\thewarning}{}

22:  \newtheorem{rmk}{Remark}

23:  \renewcommand{\thermk}{}

24: 

25:  \newenvironment{pf}{\paragraph{Proof}}{\par\bigskip}

26:  \newcommand{\qed}{\ifhmode\unskip\nobreak\fi\quad\ensuremath\square}

27: 

28: \newenvironment{digr}[1]{\subsubsection*{#1}}{\quad$\diamondsuit$\par\bigskip}

29: 

30:  \newcommand{\Span}[1]{\left< #1 \right>}

31:  \newcommand{\half}{\textstyle\frac12}

32:  \newcommand{\dd}{\mathrm{d}}

33:  \newcommand{\id}{\operatorname{id}}

34:  \newcommand{\pr}{\operatorname{pr}}

35:  \newcommand{\ch}{\operatorname{ch}}

36:  \newcommand{\td}{\operatorname{td}}

37:  \newcommand{\hF}{\operatorname{hF}}

38:  \newcommand{\KH}{\mathrm{KH}}

39:  \newcommand{\LC}{\mathrm{LC}}

40:  \newcommand{\CLRep}{\mathrm{CLRep}}

41:  \newcommand{\II}{\mathrm{II}}

42:  \newcommand{\gen}{\mathrm{gen}}

43: % \newcommand{\can}{\mathrm{can}}

44: % \newcommand{\irr}{\mathrm{irr}}

45:  \newcommand{\res}{\mathrm{res}}

46: % \newcommand{\red}{_{\mathrm{red}}} % reduced subscheme

47: % \newcommand{\mi}{_{\mathrm{min}}} % minimal surface

48:  \newcommand{\rest}[1]{_{{\textstyle{|}}#1}} % restriction of map to

49: 

50: 

51:  \newcommand{\op}{\overline{\partial}}

52:  \newcommand{\ubar}{\overline{u}}

53:  \newcommand{\Ibar}{\overline{I}}

54:   \newcommand{\p}{\partial}

55:  \newcommand{\w}{\widetilde}

56:  \newcommand{\tensor}{\otimes}

57:  \newcommand{\ov}{\overline}

58: 

59: % Script letters

60:  \newcommand{\sA}{\mathcal A} % sheaf of algebras A

61:  \newcommand{\sB}{\mathcal B} % conductor ideal

62:  \newcommand{\sG}{\mathcal G} % sheaf G

63:  \newcommand{\sD}{\mathcal D}

64:  \newcommand{\sJ}{\mathcal J} % sheaf J

65:  \newcommand{\sI}{\mathcal I}

66:  \newcommand{\sC}{\mathcal C}

67: \newcommand{\sU}{\mathcal U}

68:  \newcommand{\sH}{\mathcal H}

69:  \newcommand{\sK}{\mathcal K}

70:  \newcommand{\sL}{\mathcal L}

71:  \newcommand{\sM}{\mathcal M}

72:  \newcommand{\Oh}{\mathcal O}

73:  \newcommand{\sR}{\mathcal R}

74:  \newcommand{\sS}{\mathcal S}

75:  \newcommand{\sF}{\mathcal F}

76:  \newcommand{\sE}{\mathcal E}

77:  \newcommand{\sP}{\mathcal P}

78: 

79: % short Greeks

80:  \newcommand{\al}{\alpha}

81:  \newcommand{\be}{\beta}

82:  \newcommand{\de}{\delta}

83:  \newcommand{\ep}{\varepsilon}

84:  \newcommand{\fie}{\varphi}

85:  \newcommand{\ga}{\gamma}

86:  \newcommand{\ka}{\kappa}

87:  \newcommand{\om}{\omega}

88:  \newcommand{\si}{\sigma}

89:  \newcommand{\De}{\Delta}

90:  \newcommand{\Ga}{\Gamma}

91:  \newcommand{\La}{\Lambda}

92:  \newcommand{\Om}{\Omega}

93:  \newcommand{\la}{\lambda}

94:  \newcommand{\Si}{\Sigma}

95:  \newcommand{\na}{\nabla}

96: 

97:  \newcommand{\PP}{\mathbb P}

98:  \newcommand{\C}{\mathbb C}

99:  \newcommand{\HH}{\mathbb H}

100:  \newcommand{\Q}{\mathbb Q}

101:  \newcommand{\R}{\mathbb R}

102:  \newcommand{\Z}{\mathbb Z}

103: 

104: % \mathops

105:  \newcommand{\ad}{\operatorname{Ad}} % adjoint bundle

106:  %\newcommand{\PP}{\operatorname{\mathcal P}}

107: % \newcommand{\alg}{\operatorname{alg}}

108: % \newcommand{\res}{\operatorname{res}}

109:  \newcommand{\rk}{\operatorname{rank}}

110:  \newcommand{\hcf}{\operatorname{hcf}}

111:  \newcommand{\mir}{\operatorname{mir}}

112:  \newcommand{\codim}{\operatorname{codim}} % codimension

113: % \newcommand{\diff}{\operatorname{diff}}

114:  \newcommand{\topo}{\operatorname{top}}

115: % \newcommand{\var}{\operatorname{var}}

116: % \newcommand{\vcodim}{\operatorname{v.codim}}

117:  \newcommand{\vdim}{\operatorname{v.dim}}

118: \newcommand{\Diff}{\operatorname{Diff}}

119: % \newcommand{\Ext}{\operatorname{Ext}}

120: % \newcommand{\Gr}{\operatorname{Gr}}

121: % \newcommand{\Grass}{\operatorname{Grass}} % Grassmann variety

122: % \newcommand{\Mod}{\operatorname{Mod}}

123: % \newcommand{\Mon}{\operatorname{Mon}}

124:  \newcommand{\Pic}{\operatorname{Pic}} % Picard scheme

125: \newcommand{\Tr}{\operatorname{Tr}}

126:  \newcommand{\End}{\operatorname{End}}

127:  \newcommand{\GFT}{\operatorname{GFT}}

128:  \newcommand{\sGFT}{\operatorname{sGFT}}

129:  \newcommand{\LaGr}{\operatorname{\La_{\uparrow}\!}}

130: % \newcommand{\orGr}{\operatorname{Gr_{\uparrow}\!}}

131:  \newcommand{\Hom}{\operatorname{Hom}}

132: % \newcommand{CLRep}{\operatorname{CLRep}}

133:  \renewcommand{\Im}{\operatorname{Im}}

134:  \renewcommand{\Re}{\operatorname{Re}}

135:  \newcommand{\Sing}{\operatorname{Sing}}

136: % \newcommand{\Spec}{\operatorname{Spec}}

137:  \newcommand{\Vol}{\operatorname{Vol}}

138: % \newcommand{\Stab}{\operatorname{Stab}} % Stabiliser group

139:  \newcommand{\Conj}{\operatorname{Conj}}

140: \newcommand{\CL}{\operatorname{CL}}

141: % Lie groups

142:  \newcommand{\SL}{\operatorname{SL}}

143:  \newcommand{\SO}{\operatorname{SO}}

144: %\newcommand{U(1)}{\operatorname U(1)} % unitary group

145:  \newcommand{\PU}{\operatorname{PU}}

146:  \newcommand{\SU}{\operatorname{SU}}

147:  \newcommand{\fsl}{\operatorname{\mathcal{\frak{sl}}}}

148:  \newcommand{\fsu}{\operatorname{{\frak{su}}}}

149: % \newcommand{\Spin}{\operatorname{Spin}}

150: % \newcommand{\SP}{\operatorname{Sp}}

151: % \newcommand{\odd}{\mathrm{odd}}

152:  \newcommand{\pt}{\mathrm{pt}}

153: 

154:  \renewcommand{\labelenumi}{(\arabic{enumi})}

155: 

156: \begin{document}

157: 

158:  \title{Quantization, Classical and Quantum Field Theory and Theta-Functions.}

159: 

160:   \author{\fbox{Andrei Tyurin}}

161:   \date{24.02.1940 -- 27.10.2002}

162:  

163: 

164: \maketitle

165: 

166: \begin{center}

167:  {\em To Igor Rostislavovich Shafarevich on his 80th birthday}

168:  \end{center}

169:  \bigskip

170: 

171: 

172: \section*{Introduction}

173: 

174:  Arnaud Beauville's  survey "Vector bundles on Curves and Generalized

175: Theta functions: Recent Results and Open Problems"

176: \cite{Be} appeared 10 years ago. This elegant survey is short (16 pages)  but

177: provides a complete introduction to a specific part of algebraic geometry.

178: To repeat his succes now we need more pages, even though we  assume that

179: the reader is already acquainted with the material presented

180: there. Moreover, in  Beauville's survey the  relation

181: between generalized theta functions and conformal field theories (classical and

182: quantum) was presented already.

183: 

184: Following  Beauville's  strategy we do not provide any proof or motivation.

185:  But we would like  to propose {\it all constructions} of

186: this large domain of mathematics in

187: such a way that the proofs    can be guessed

188:   from the geometric picture. Thus this text is not a mathematical monograph

189:    yet, but rather a digest of a field of mathematical investigations.

190: 

191: In the abelian case ( the

192:   subject of several beautiful classical books (\cite{B},

193:   \cite{C}, \cite{Wi}, \cite{F1} and  many others ) fixing some

194:   combinatorial structure

195:    (a so called

196:   theta structure of level $k$)   one obtains  special basis in the space

197:   of  sections of powers of the canonical polarization powers on jacobians.

198:   These sections can be presented  as holomorphic functions on the "abelian

199:   Schottky" space $(\C^*)^g$.

200:   This fact provides various applications of these  concrete analytic formulas

201:    to  integrable

202:   systems, classical mechanics and PDE's  (see the references in  \cite{DKN} ).

203: 

204:    Our practical goal

205:  is to do the same in the non-abelian case, that is, to give the answer to

206:  the final question of the Beauville's survey (Question 9 in

207:  \cite{Be}).

208: 

209:    It has been observed many time

210:    that the {\it construction} of theta functions with characteristics is

211:    intricately

212:    related to the paradigm of the quantization procedure (which is

213:    a quantum field theory in  dimension 1).

214: New features   came from Conformal Field Theory (which is a field theory

215: in dimension 2 = 1+1).

216:  This new stream brings the

217: standard physical paradigm of "symmetries, fields, equations etc ... and

218:  gluing properties corresponding to local

219: Lagrangians". New CFT methods provided powerful  computational tools while

220: "algebraic

221: geometers would have never dreamed of being able to perform such

222: computations" (A. Beauville).

223: 

224: In  future we hope to extend this digest to a mathematical monograph

225: with the title "VBAC".

226: 

227: {\it Acknowledgments.} These notes were written for a series of lectures given at the Centre de Recherches Mathematiques at the Universite de Montreal in September 2002. I am grateful to J. McKay, J. Hurtubise, J. Harnad, D. Korotkin

228: and A. Kokotov who made my stay in Montreal very agreeable and productive.  I would like to express my gratitude to

229: my collaborators C.~Florentino, J.~Mourao, J.P.~Nunes and

230: participants of Quantum Gravity seminar in IST (Lisboa, 2000). I would like to thank

231:  Jean Le Tourneux and Andre Montpetit for their assistance in the preparation of these notes. Special thanks  go to my daughter, Yulia Tiourina,

232:  for  catching  numerous mistakes and misprints in this huge text.

233: 

234: \tableofcontents

235: 

236: 

237: \section{Quantization procedure}

238: 

239: \subsection{The Framework}

240: 

241: The first question: what we want  to quantize? The

242: framework of geometric quantization is usually described as

243: follows ( see, for example \cite{GS1}). Let $(M, \om)$ be a

244: symplectic manifold which represents a classical mechanical

245: system with finite number of degrees of freedom or the bialgebra

246: $C^\infty(M)$ with the usual commutative multiplication  $f_1

247: \cdot f_2$ and Poisson structure given by well known construction:

248:  every $f \in C^\infty(M)$ defines a Hamiltonian vector field

249: \begin{equation}

250: H_f = \om^{-1}(df)

251: \end{equation}

252: and a pair $(f_1, f_2)$ defines the function

253: \begin{equation}

254: \{f_1, f_2\} = \om(H_{f_1}, H_{f_2})

255: \end{equation}

256: -Poisson brackets, such that

257: \begin{equation}

258: d \om = 0 \implies \text{ Jacobi equality}.

259: \end{equation}

260: 

261: If $M$ is compact, then even abelian multiplication is enough to

262: reconstruct $M$ and then the Lie algebra structure reconstructs

263: $\om$.

264: 

265:  So in this case the pair $(M, \om)$ and the (observables) algebra

266: $C^\infty(M)$ are equivalent objects.

267: 

268:  (Pre BRST)-quantization rules is a map $Q$ sending classical

269: objects to corresponding quantum objects:

270: \begin{enumerate}

271: \item $Q(M) = \sH$ is a Hilbert space (of wave functions);

272: \item $Q(f) \in Op(\sH)$ where $Op(\sH)$ is the space of

273:  self adjoint operators

274: \item which

275: should satisfy the {\it correspondence principle}

276: \begin{equation}

277: [Q(f_1), Q(f_2)] = i \cdot h \cdot Q(\{ f_1, f_2 \})

278: \end{equation}

279: (Dirac correspondence) and the representation of our Poisson

280: Lie-algebra  be irreducuble.

281: \end{enumerate}

282: 

283: Unfortunately, such construction couldn't exist at all due to van

284: Hove theorem. Nevertheless  one usually uses two basic examples:

285: Souriau - Kostant quantization doesn't satisfy the irreducibility

286: condition while  Berezin quantization doesn't satisfy the

287: correspondence principle.

288: 

289: But in both approaches we  extend classical mechanical

290:  date by

291:  a {\it prequantization date} $(L, \nabla)$, where $L$ is a

292: complex line bundle (with a Hermitian structure $h$) and $\nabla$

293: is a unitary connection on $L$ with the curvature form

294: \begin{equation}

295: F_\nabla = 2 \pi \cdot i \cdot \om.

296: \end{equation}

297: 

298:  This equality implies  very strong constraint to the

299:   symplectic structure:

300: \begin{equation}

301: [\om] \in H^2(M, \Z)

302: \end{equation}

303: that is the cohomology class of the symplectic form has to be

304: integer.

305:  Such quadruple

306: \begin{equation}

307: (M, \om, L, \nabla)

308: \end{equation}

309:  is ready to be quantized.

310: First of all the space of wave functions is the space of sections

311: \begin{equation}

312: \sH = \Ga^\infty (L)

313: \end{equation}

314: with the inner product given by the formula

315: \begin{equation}

316: <s, s'> = 1/n! \int_M (s, s')_h \om^{1/2 dim M},

317: \end{equation}

318: (usually one  extends this space to $L^2$ - sections). For every

319: function (a classical observable) we use the operator

320: \begin{equation}

321:   Q(f) = \nabla_{H_f} + 2 \pi i \cdot f,

322: \end{equation}

323: that is

324: $$

325: Q(f)(s) = \nabla_{H_f} s + 2 \pi i \cdot f \cdot s.

326: $$

327: Two simple exercises:

328: \begin{enumerate}

329: \item Dirac equality holds (2.4);

330: \item  the representation of

331: our Poisson Lie-algebra is very far  from being irreducible (even

332: for the simplest case $M = \R^2 = T^{*} \R)$

333: \end{enumerate}

334: 

335: %\begin{figure}[tbn]

336: %\centerline{\epsfxsize=2in\epsfbox{18.eps}}

337: %\caption{\sl complete integrable}

338: %\label{F1}

339: %\end{figure}

340: 

341: Well known remedy is a choice of a {\it polarization}. A

342: polarization is a distribution that is a sub bundle of the

343: complexified tangent space

344: \begin{equation}

345: F \subset T^\C M , \quad rk F = 1/2 dim M;

346: \end{equation}

347: 

348: 

349: Fixing such polarization  the wave function space can be obtained

350: as

351: \begin{equation}

352: Q(M)= \sH^F = \{ s \in \Ga^\infty (L) \vert \nabla_F s = 0 \}.

353: \end{equation}

354:  There are two natural choices:

355: \begin{enumerate}

356: \item $\overline{F} = F$  is a {\it real polarization};

357: \item $\overline{F} = F^\perp $  is a

358: {\it complex polarization}, that is, a choice of an almost complex

359: structure $I$ such that $(F = T^{0,1})$.

360: \end{enumerate}

361: 

362: Of course for the general choice of a polarization we don't know

363: so much but there is a couple of good choices:

364: 

365: in the real case: choose $F$  integrable and moreover completely

366: integrable that is defined by a Lagrangian fibration:

367: \begin{equation}

368: \pi \colon M \to B , \quad \pi^{-1}(b_{gen}) = T^n

369: \end{equation}

370: that is, the generic fiber here is  a Lagrangian torus of

371: dimension $n = 1/2 dim M$ (see Fig.1);

372: 

373:  and  in the complex case: choose  an

374: almost complex structure which is integrable and compatible with

375: $\om $ such that the pair $(\om, I)$ gives a Kahler metric with

376: condition (1.6) that is a Hodge metric.

377: 

378: 

379: \begin{figure}[tbn]

380: \centerline{\epsfxsize=3in\epsfbox{22c.eps}}

381: \caption{\sl Completely integrable system}

382: \label{F2}

383: \end{figure}

384: 

385: 

386: There is a couple of equivalent conditions for both of these

387: cases: for a real polarization, for any two vector fields

388: \begin{equation}

389: \p_1, \p_2 \in F \subset T^\C M

390: \end{equation}

391: the commutator

392: \begin{equation}

393: [\p_1, \p_2] \in F \subset T^\C M

394: \end{equation}

395: i.e.  vector fields from our real distribution form a Lie

396: subalgebra.

397:  (Otherwise their commutators form a quadratic form on $ F$ with

398:  coefficients in $T^\C M / F$ just like the second quadratic

399:  form in the standard differential geometry of a surface

400:  in $\R^3$ ).

401: 

402: For the complex case we have to add the equality of the

403:  same nature

404: \begin{equation}

405: [I \p_1, I \p_2] = [\p_1, \p_2] + I [I \p_1, \p_2]

406: + I [\p_1, I \p_2]

407: \end{equation}

408: where we consider $I$ as a section of $End TM$ with

409: \begin{equation}

410:  I^2 = -1

411: \end{equation}

412: $$

413: \om(\p_1, I \p_2) = - \om (I \p_1,  \p_2)

414: $$

415: $$

416: \om(\p_1, I \p_2) > 0.

417: $$

418: 

419: 

420: These cases are lying in different domains of mathematics: the

421: real polarizations  are contained in Symplectic Geometry and the

422: complex polarization belongs to Algebraic Geometry. A

423: quantization is {\it perfect} if it  admits both of these

424: polarizations. So the perfect quantization belongs to the union of

425: Algebraic and Symplectic Geometries.

426: 

427: 

428: \subsection{The real polarization.(Symplectic Geometry)}

429: 

430: 

431: 

432: 

433: Recall that a  subcycle $T \subset M$ is Lagrangian if it is a

434: middle dimensional subcycle such that

435: \begin{equation}

436: \om \vert_T = 0.

437: \end{equation}

438: Thus  the restriction of the pair $(L, \nabla)\vert_T$ is  the

439: trivial

440:  line bundle with a flat connection

441: \begin{equation}

442: F_\nabla \vert_T = 2 \pi i \om \vert_T = 0.

443: \end{equation}

444: Recall that a gauge class of a flat connection is given by

445:  a character

446: of the fundamental group of $T$

447: \begin{equation}

448: \chi \colon \pi_1(T)     \to U(1).

449: \end{equation}

450: 

451: 

452: \begin{dfn} A Lagrangian cycle $T$ is Bohr-Sommerfeld

453: if $\chi = 1$, that is, if there exists a covariant constant

454: section of $(L, \nabla)\vert_T$.

455: \end{dfn}

456:  To kill a character,  we need $rk H_1(T, \Z)$ conditions

457:  so we may expect that

458:  in a family

459: of Lagrangian cycles the subfamily of Bohr-Sommerfeld cycles has

460: this codimension. Moreover, it was shown

461:   that in any small

462:  Darboux-Weinstein

463:  neighborhood of a smooth

464:  Bohr-Sommerfeld cycle $T$ every other Bohr-Sommerfeld cycle $T'$

465:  intersects

466:  $T$.

467: 

468: For example,  our real polarization (1.13) defines a family

469:  of Lagrangian

470: cycles parameterized by a space $B$ of dimension $n = 1/2 \dim M$.

471: But  a general fiber of any real polarization fibration $\pi$

472: (1.13) is a n-dimensional torus, that is

473: $$

474:  rk H_1(T^n) = n = dim B.

475: $$

476:  Thus the set of Bohr-Sommerfeld fibers is discrete and if $B$ is

477:  compact it forms

478:   a finite set $BS \subset B$ of Bohr-Sommerfeld fibers of $\pi$.

479: 

480:  Then the Hilbert space of wave functions is given as

481:  the direct sum of lines

482: \begin{equation}

483:  Q_F(M) = \sH_\pi = \oplus_{b \in BS \subset B} \C \cdot

484: s_b

485: \end{equation}

486: where $s_b$ is a covariant constant section of $(L,

487: \nabla)\vert_{\pi^{-1}(b)}$).

488: 

489: %\begin{figure}[tbn]

490: %\centerline{\epsfxsize=2in\epsfbox{23.eps}}

491: %\caption{\sl twice integrable system}

492: %\label{F4}

493: %\end{figure}

494: 

495: Recall that every covariant constant section is defined up

496: to $U(1)$-scaling,

497: so we can define naturally an Hermitian scalar product on this space.

498: 

499: We can extend  the notion of

500:  Bohr-Sommerfeld Lagrangian cycles:

501: 

502: \begin{dfn} A Lagrangian cycle $T$ is Bohr-Sommerfeld of level $k$ (or $BS_k$ - cycle)

503:  if $\chi^k = 1$, that is, if there

504: exists a covariant constant section of $(L^k, \nabla_k)\vert_T$,

505: where $\nabla_k $ is the connection generated by $\nabla$ on the

506: tensor power $L^k$ of the line bundle $L$.

507: \end{dfn}

508: 

509: Reasoning  the same way as before,  we can prove that the subset

510: $BS_k \subset B$ is  finite and that the Hilbert space of wave

511: functions of level $k$ is given by the direct sum of lines

512: \begin{equation}

513:  Q_F^k(M) = \sH_\pi^k = \oplus_{b \in BS_k \subset B} \C \cdot

514: s_b

515: \end{equation}

516: where $s_b$ is a covariant constant section of $(L^k,

517: \nabla_k)\vert_{\pi^{-1}(b)}$).

518: 

519:  But we are quantizing a classical dynamical system $(M, \om)$!

520: So the result does not have to  depend on additional data  like

521: $(L, \nabla, \pi, I)$ and so on.

522: 

523: In our construction the wave space $\sH_\pi$ depends on a gauge

524: class of a connection $\nabla$ only and this class is given by

525: the curvature form if $M$ is simply connected (see \cite{AB}).

526: (Recall that in non-simple connected case

527:  we have to fix additionally  a point of the

528:  "jacobian" $H^1(M, \R) / H^1(M, \Z)$ of $M$ (see below).

529: 

530:   In any case

531:  our wave function spaces depend up to a shift of characters on the

532: curvature tensor which is $\om$ and on a choice of a real

533: polarization $\pi$.

534: 

535: \begin{figure}[tbn]

536: \centerline{\epsfxsize=3in\epsfbox{23c.eps}}

537: \caption{\sl twice integrable system}

538: \label{F4}

539: \end{figure}

540: 

541: 

542: But the most important thing is that the projectivization of the

543: wave functions

544:  space doesn't depend on a real polarization $\pi$ too.

545:   For any other Lagrangian fibration $\pi' \colon M \to B'$,

546: Kostant defined the canonical (up to constant )

547:  pairing between

548: $\sH_\pi$ and $\sH_{\pi'}$ (see for example \cite{JW2}). So we

549: get  the identification

550: \begin{equation}

551:  \PP \sH_\pi = \PP \sH_{\pi'}

552: \end{equation}

553: of projective spaces.

554: 

555:  Thus $\sH_\pi$  is the wave space of a quantization of $(M, \om)$

556:  and of nothing else.

557: 

558: \subsection{Kahler quantization. (Algebraic Geometry) }

559: 

560: In this case we  have a quadruple $(M, \om, \nabla, I)$ where a

561: complex structure $I$ is compatible with $\om$. Thus the pair

562: $(\om, I)$ defines a Kahler metric with the Kahler form $\om$.

563: Moreover, $\om$ is of Hodge type (1,1). Thus the connection

564: $\nabla $ on $L$ has the curvature form of type (1,1) and defines

565: a holomorphic structure on $L$ (see for, example, \cite{GH}).

566:  As

567: Kodaira (1958) proved, a Kahler metric  with  integer cohomology

568: class $[\om] \in H^2(M, \Z)$ of the Kahler form is a Hodge

569: metric, hence $M_I$ is a projective algebraic variety and $L =

570: \Oh_M (1)$ is the class of  hyperplane section of $M_I$.

571: 

572:  So, the complex polarizations bring us to Algebraic Geometry!

573: 

574: 

575: Naturally, the wave function space

576: \begin{equation}

577: Q_F(M) = \sH_I = H^0(M_I, L)

578: \end{equation}

579: is the space of holomorphic sections of the polarization $L$ and

580:  the projectivization $\PP H^0(M_I, L)$ is the classical

581:  complete linear system on $M_I$.

582: 

583: The Hermitian structure on $L$ gives an Hermitian structure on

584: $H^0(M_I, L)$ and provides the identification

585: \begin{equation}

586:  \sH_I = H^0(M_I, L) = H^0(M_I, L)^*

587: \end{equation}

588:    with changing of the complex structure to the complex

589: conjugate.

590: 

591: The standard construction of the Algebraic Geometry gives the

592: following map

593: \begin{equation}

594:  \phi \colon M_I \to \PP H^0(M_I, L)^*,

595: \end{equation}

596: that is, under the previous identification

597: \begin{equation}

598:  \ov{\phi} \colon M_I \to \PP H^0(M_I, L).

599: \end{equation}

600: 

601: Images of points of $M_I$ are called {\it coherent states}.  A

602: priori such image is defined up to a transformation

603:  from $PGL( \PP H^0(M_I, L))$. But its every

604:  {\it stable} orbit contains  unique

605: unitary orbit preserving our form  $\om$. If our form $\om$ is the

606: restriction of

607:  the Fubini-Study form

608: from $ \PP H^0(M_I, L)$, then we can switch on to  Geometric

609: Invariant Theory and reduce the theory of coherent states to it.

610: So in algebraic geometrical set up we can assume that our form is

611: a restriction of the Fubini-Study form. In the general case we

612: have to be more careful and to add so called J.Rawnsley

613: $\ep$-function of

614:  (see  \cite{R}).

615: 

616: Of course we lose the Dirac correspondence and have to use the

617: Berezin-Toeplitz quantization rules (see for example \cite{R}).

618: Thus we  introduce the level parameter $k \in \Z^+$. The space of

619: wave function of level $k$ is the space of holomorphic sections

620: \begin{equation}

621:  \sH_I^k = H^0(M_I, L^k).

622: \end{equation}

623: The Dirac correspondence can be restored in the

624: {\it quasi classical limit} $k \to \infty$ or $1/k \to 0$.

625:  So,  our complex polarization becomes a polarization in

626:  algebraic geometrical sence. Recall that if $k$ is large enough,

627:  then $\ov{\phi}$ (1.27) is an embedding  and the dimension of the

628:  corresponding projective space is given by the Riemann-Roch

629:  formula as a pure topological invariant.

630: 

631: So, every underlying symplectic manifold $(M, \om)$ of

632:  a polarized algebraic   variety $M_I, \Oh(1)$  with a

633:  natural complex polarization $\Oh(1)$ gives  the collection

634:   of wave functions

635:  spaces

636: $H^0(M_I, L^k) = H^0(M_{I}, \Oh(k))$

637: 

638:  But the quantization set up disturbs

639:   the Algebraic Geometry Peace.

640: The main question is:

641: 

642: do these spaces appear as a result of a successful quantization

643: procedure of a classical system given by the  underlying

644: symplectic manifold $(M, \om)$?

645: 

646:  That is,

647: 

648: is the projectivization of this space independent  on the complex

649: structure $I$ ?

650: 

651: (In physical symbols, $\frac{\p \PP H^0(M_I, L)}{\p I} = 0$ .)

652: 

653:  Mathematically correct question is the following:

654: let  $ \sM $ be the moduli space of such complex structures, that

655: is,  the moduli space of polarized  (in the algebraic geometryical

656: sense) varieties and

657: \begin{equation}

658: p \colon V_k \to \sM, \quad p^{-1}(M_I) = H^0(M_I, L^k)

659: \end{equation}

660: - a vector bundle of holomorphic sections of polarization line

661: bundles. Note that by the Kodaira vanishing theorem it is a vector

662: bundle indeed! Then

663:    there exists a holomorphic projective flat connection on $V$.

664: 

665: Existence of such connection constraints the topology type of the

666: vector bundle: if a vector bundle $V$ admits a projective flat

667: connection, then

668:  \begin{equation}

669:   c_t(V) = (1 - c \cdot t)^{rk V}.

670: \end{equation}

671: In particular,

672: \begin{equation}

673:   c_1(V) = rk V \cdot c, \quad c \in H^2(\sM, \Z).

674: \end{equation}

675: This is  very strong topological constraint!

676: 

677:  Thus   problems of new type are

678:  given inside of Algebraic Geometry:

679: 

680:  When a pair $M_I, \Oh(1)$ of an algebraic variety with

681:  a polarization is a result

682:   of a quantization procedure

683:  for the underlying symplectic manifold ?

684: 

685: Before  considering the simplest example when $\dim M = 2$ we give

686: some   general remarks:

687: \begin{enumerate}

688: \item of course our vector bundles $V_k$ (1.29) are

689: holomorphic but the  required projective flat connection could be

690: non holomorphic;

691: \item so, it is quite reasonable to construct may be not

692: holomorphic projective Hermitian connection;

693: \item in all classical examples

694: holomorphic projective flat connections are Hermitian by the

695: construction;

696: \item but in the last non-abelian case (which is of the main

697:   interest here) this fact isn't known

698:  up to now;

699: \item if such new connection exists automatically we get a Higgs field

700: on every vector bundle (1.29) on moduli spaces of complex

701: structures.

702: \item In the physical set up it may be  that we do not obtain  vector

703:  bundles (1.29) but have the collection of projective bundles

704:  which {\it aren't projectivizations of vector bundles};

705: \item even in the infinite dimensional case (when Hilbert  vector

706:  bundles are always trivial) the projective bundles can be non trivial

707:   and are given by a cohomology class from $H^3(\sM, \Z)$. This

708: Brauer group gives us non trivial analog of Atiyah's K-functor.

709: \end{enumerate}

710: 

711: In the conformal field theory context this connection always {\it

712: has to be holomorphic} but we do not have space and time to

713: discuss it here.

714: 

715: 

716: Because of the time-space problem this very interesting subject

717: is out of our considerations here.

718: 

719: 

720: \subsection{Extended Kodaira-Spencer theory}

721: 

722: The theory of deformations of complex structures was developed by

723: Kodaira and Spencer many years ago. Trying  "theoretically" to

724: solve  the

725:  problem of  successful

726: Kahler quantization we have to lift this theory to the problem

727: of

728:  holomorphic deformations of a triple $(I, L_I, (s)_0)$,

729:   where $I$ is an integrable complex structure, $L_I$ is

730:   our holomorphic line bundle and $s \in H^0(L_I)$ is

731:    zero divisor of a holomorphic section of this line bundle.

732:    The best reference for such lifting is   Welters paper

733:    \cite{W}. Recall that the holomorphic structure of $L_I$

734:     codes the connection $\nabla$ from the quadruple $(1.7)$ by the decomposition

735: \begin{equation}

736:  \nabla = \nabla^{1, 0} + \nabla^{0, 1}

737: \end{equation}

738: where

739: \begin{equation}

740:   \nabla^{1, 0} = 1/2 \cdot(1 - i I) \nabla, \quad \nabla^{0, 1} =

741:   1/2 \cdot (1 + i I) \nabla.

742: \end{equation}

743: Our history starts with an infinitesimal deformation of the

744: complex structure $\dot I$. Since it is an anti involution $I^2 =

745: -1$ we have the equality

746: \begin{equation}

747: \dot I I + I \dot I = 0

748: \end{equation}

749: and one can consider $\dot I$ as a projection of $-i$ eigenspace

750: of $I$ to the $+i$-eigenspace. That is

751: \begin{equation}

752: \dot I \in \Om^{0, 1} (M_I,T^{1, 0})

753: \end{equation}

754: (just as  usual Beltrami differential).

755: 

756: Linearization of the integrability condition (1.16) gives us

757: \begin{equation}

758: (\op \dot I)^{2, 0} = 0,

759: \end{equation}

760: that is, the (0, 2)-component of this tensor is trivial.

761: 

762: Suppose that we have the following list of properties.

763: \begin{enumerate} \item

764: Our polarization $L$ is a rational part of the canonical class $K$

765: of $M_I$. That is

766: \begin{equation}

767: - [K] = n [c_1(L)].

768: \end{equation}

769: \item Further on we suppose that the Levi-Civita connection of

770:  the Kahler

771:  metric $\om(\quad , I \quad )$ (1.17) on $K $ is proportional to

772:   the unitary  connection on $L$.

773: \item We consider a sufficiently large level $k$ such that

774: \begin{equation}

775: i >o \implies H^i(M_I, \Oh(k \cdot c_1(L) + 1/2 K)) = 0.

776: \end{equation}

777: (For example, $M_I$ is a Fano-variety of index $n$ and

778:  $L = \Oh(k)$).

779: \item Then we have the symplectic Kahler form

780: \begin{equation}

781: ( k + n/2) \om.

782: \end{equation}

783: \end{enumerate}

784: 

785: From this it is easy to see (for example, by  local

786: representation of our tensor) that

787:  there exists a  section

788: \begin{equation}

789: W_0 \in \Ga^\infty (S^2 T^{1, 0}M_I) \in (S^2)^{2, 0}

790: \end{equation}

791: where $S^2$ is the second symmetrical power of the

792: holomorphic tangent bundle such that the convolution of

793:  this tensor with our symplectic form gives us $\dot I$:

794: \begin{equation}

795: W_0 * (k + n/2) \om = \dot I.

796: \end{equation}

797: (Recall that $\om$ is of a type $(1.1)$ with respect to the

798: complex structure $I$.)

799: 

800:  To see this very important tensor we have to represent tensors

801:  as homomorphisms of tensors vector bundles!

802:   By this notation we

803:    emphasize

804:   that this construction is due to Welters \cite{W}.

805:    So our symplectic form

806: \begin{equation}

807: (k+n/2)\om \colon T^{1, 0} \oplus T^{0, 1} \to \Om^{1, 0} \oplus \Om^{0, 1}

808: \end{equation}

809: is of the Hodge type $(1,1)$ and it means that

810: \begin{equation}

811: \om(T^{1, 0}) \in \Om^{0,1} \text{ and } \om(T^{0, 1}) \in \Om^{1,0}.

812: \end{equation}

813: On the other hand,

814: \begin{equation}

815: \dot I \colon T^{0,1} \to T^{1,0}.

816: \end{equation}

817: Thus the composition of these homomorphisms

818: \begin{equation}

819: \dot I \circ \frac{1}{k+n/2} \om^{-1} = W \colon  \Om^{1,0} \to T^{1,0}

820: \end{equation}

821: is the very one symmetric tensor.

822: 

823: 

824: 

825: 

826: Now consider the cup product map

827: \begin{equation}

828: H^0(M_I, T^{1, 0}) \otimes H^1(M_I, \Om^{0, 1}) \to H^1(M_I, \Oh)

829: \end{equation}

830: and the restriction of it to

831:  $H^0(M_I, T^{1, 0}) \otimes [(k+n/2)\om]$:

832: \begin{equation}

833: [(k+n/2)\om] \colon H^0(M_I, T^{1, 0}) \to H^1(M_I, \Oh).

834: \end{equation}

835: Suppose that this in an {\it isomorphism}. For example

836: \begin{equation}

837: H^0(M_I, T^{1, 0}) = H^1(M_I, \Oh) = 0

838: \end{equation}

839: or $M_I$ is a polarized  abelian variety.

840: 

841: In this case our holomorphic line bundle $L_I$ can't be deformed

842: continuously

843:  under deformation of the complex structure.

844:   So we have to investigat  deformations of

845:  holomorphic sections only.

846: 

847: 

848: Having such symmetric form and following Nigel Hitchin let us

849: infinitesimally deform a section $s \in H^0(M_I, L_I)$ to a $\dot

850: I$-holomorphic section. Differentiating along a "time" $t$ the

851: condition $(1 + i I_t) \nabla s_t = 0$ we get the equality

852: \begin{equation}

853: i \dot I \nabla^{1,0} s = \nabla^{0,1} a

854: \end{equation}

855: where $-a$ is  a section of $L$. Such section depends linearly on

856: $s$ and $\dot I$ and defines a connection on the bundle $V$

857: (1.29).

858:  The main observation of

859: the extended Kodaira-Spencer theory is the fact that

860:  this equation admits

861: a cohomological interpretation. Namely LHS and RHS of (1.49) are

862: results of applying  first order

863:  differential

864: operators on $L$. So now we have to consider the vector

865: bundle $D^1(L_I)$ of

866: holomorphic differential operators on $L_I$ applied to

867: a holomorphic section

868:  in LHS and to the potential of a connection.

869: 

870: Let the exact sequence

871: \begin{equation}

872: 0 \to \Om^{1,0} \otimes L \to J^1(L) \to L \to 0

873: \end{equation}

874: be the extension defining 1-jet bundle of our line bundle $L$. By

875: the Atiyah (see \cite{A2}) theorem, the cocycle of this extension

876: is

877: \begin{equation}

878: [c_1(L)] = [\om] \in H^1(M_I, \Om^{1,0}).

879: \end{equation}

880: 

881: Now

882: \begin{equation}

883: D^1(L) = Hom (J^1(L), L )

884: \end{equation}

885: is the bundle of the sheaf of germs of differential operators

886: on $L$ of order

887: $\leq 1$. So this sheaf is the extension

888: \begin{equation}

889: 0 \to \Oh \to D^1(L) \to T^{1, 0} \to 0

890: \end{equation}

891: (given by the class $- [\om]$).

892: 

893: The long exact cohomology sequence of this exact triple

894:  decays in two parts

895: \begin{equation}

896: 0 \to H^0(M_I, \Oh) \to H^0(M_I,D^1(L)) \to 0

897: \end{equation}

898: and

899: \begin{equation}

900: 0 \to H^1(M_I, D^1(L)) \to H^1(M_I, T^{1, 0}) \to H^2(M_I, \Oh)

901: \dots

902: \end{equation}

903: because they are divided by the homomorphism (1.44).

904:  The first exact sequence states

905:  that an every globally defined first order holomorphic operator

906:  on $L$ is

907:  a {\it multiplication by a constant} and the second states

908:  that the {\it

909:  symbol map} $H^1(M_I, D^1(L)) \to H^1(M_I, T^{1,0})$ is a

910:  { \it monomorphism}.

911: 

912: Now for a holomorphic section  $s \in H^0(M_I, L_I)$ the

913:  valuation morphism

914: \begin{equation}

915: d_s \colon D^1(L) \to L , \quad d_s (D) = D s,

916: \end{equation}

917: gives the complex

918: \begin{equation}

919: 0 \to D^1(L) \to L  \to 0

920: \end{equation}

921: which hypercohomology $H_h^1(d_s)$  defines linear infinitesimal

922: deformations of a triple $(\dot I, L_I, s)$.

923: 

924: Each class of such hypercohomology can be done as a pair

925: \begin{equation}

926: (\{ \dot s_i\}, \{\al_{i,j} \}) \in C^0(\sU, L) \oplus C^1(\sU, D^1(L))

927: \end{equation}

928: where $\sU$ is a Cech covering. Indeed it is a 1-cocycle of the

929: total complex, associated with the double complex

930: \begin{equation}

931: \pm d_s \colon C^i(\sU, D^1(L)) \to C^i(\sU, L).

932: \end{equation}

933: The spectral sequence of hypercohomology (of a double complex)

934: yields  the exact sequence

935: \begin{equation}

936: H^0(M_I, D^1(L)) = \C \to H^0(L_I) \to H_h^1(d_s) \to H^1(D^1(L)) \to H^1(L_I)

937: \end{equation}

938: corresponding to the pairs (1.56).

939: 

940: On the other hand,    $d_s$-complex (1.57) defines the exact

941: quadruple

942: \begin{equation}

943: 0 \to ker_s \to D^1(L) \to L \to coker_s \to 0

944: \end{equation}

945: and the second spectral sequence gives

946: \begin{equation}

947: 0 \to H^1(ker) \to H_h^1(d_s) \to H^0(coker) \to H^2(ker).

948: \end{equation}

949: It is easy to see (this observation due to Welters \cite{W})

950:  that the support

951: of the sheaf $coker_s$ can be very small:

952: \begin{equation}

953: Supp F = Sing (s)_0

954: \end{equation}

955: -the singularity locus of the zero-divisor of a section $s$. In

956: particular, if a zero divisor

957:  is smooth (this is a general case)

958: then $F=0$  and the exact quadruple (1.60) is in fact a triple.

959: 

960: Now let us return to our main equation (1.49).

961: The RHS admits a solution if the

962: \begin{equation}

963: (\op LHS )^{0,2} = 0

964: \end{equation}

965: (by the  Dolbeault lemma). So we have to compute this RHS Hodge

966: type. But by ( 1.41 ) it is easy to see that

967: \begin{equation}

968: (\op (i \dot I \nabla^{1,0}))^{0,2} = 0.

969: \end{equation}

970: Here we are using

971: \begin{equation}

972: i \dot I \nabla^{1,0} \in \Om^{0,1} (M_I, D^1(L))

973: \end{equation}

974: and the infinitesimal integrability condition (1.34).

975: Now we have the interpretations of $i \dot I \nabla^{1,0}$ and

976:  $-a$ from (1.49)

977: as (0, 1) - forms with coefficients in $D^1(L)$ and we have

978: the complex

979: \begin{equation}

980: C^i = \Om^{0,i} (M_I, D^1(L)) \oplus \Om^{0, i-1}(M_I, L_I)

981: \end{equation}

982: with the differential

983: \begin{equation}

984: d(s) (D, a) = (\op D, \op a + (-1)^{i-1} D s)

985: \end{equation}

986: Now

987: \begin{equation}

988: \op s = 0 \implies d(s)^2 = 0

989: \end{equation}

990: and our pair

991: \begin{equation}

992: (i \dot I \nabla^{1,0}, -a) \in C^1.

993: \end{equation}

994: Moreover, equations (1.65) and (1.49) give

995: \begin{equation}

996: d(s) (i \dot I \nabla^{1,0}, -a) = 0.

997: \end{equation}

998: Thus, every solution to (1.49) defines a  hypercohomology class

999:  in $H_h^1(d_s)$!

1000: 

1001: Now suppose we can lift every pair

1002: \begin{equation}

1003: (\dot I, s) \in \Om^{0,1}(M_I, T^{1,0}) \oplus H^0(M_I, L_I)

1004: \end{equation}

1005: to a cocycle from $H_h^1(d_s)$ such that the natural "symbol" map

1006: \begin{equation}

1007: \si \colon H_h^1(d_s) \to H^1(M_I, T^{1,0})

1008: \end{equation}

1009: sends our pair to the cohomology class $[\dot I] \in H^1(M_I, T^{1,0})$.

1010: That is there exists a map

1011: \begin{equation}

1012: A \colon  \Om^{0,1}(M_I, T^{1,0}) \oplus H^0(M_I, L_I) \to H_h^1(d_s)

1013: \end{equation}

1014: such that the composition $\si A$ is the map to the cohomology class.

1015: Then such map defines a projective  connection on the vector bundle

1016: of holomorphic sections (1.29).

1017: 

1018: Indeed, we can presents a hypercohomology class as a pair (1.58)

1019: \begin{equation}

1020: (D, s') \in \Om^{0,1} (M_I, D^1(L)) \oplus \Om^{0, 0}(M_I, L_I).

1021: \end{equation}

1022: But now the symbols of operators $-i \si (D)$ and $i \dot I

1023: \nabla^{1,0}$ are cohomological  and, therefore the corresponding

1024: operators are cohomological in $\Om^{0,1}(M_I, D^1(L))$    because

1025: map to symbols is a monomorphism (1.55). Thus there exists a

1026: global operator

1027:  $\sD \Om^0(M_I, D^1(L))$ such that

1028: \begin{equation}

1029: D - i \dot I \nabla^{1,0} = \op \sD.

1030: \end{equation}

1031: Since $(D, s')$ is a cocycle ($d(s)(D, s') = 0 $, (see (1.71))

1032:  one gets

1033: \begin{equation}

1034: Ds + \op s' = i \dot I \nabla^{1,0} s + \op (\sD s + s').

1035: \end{equation}

1036: Thus

1037: \begin{equation}

1038: -a = \sD s + s'

1039: \end{equation}

1040: is a solution for (1.49).

1041: 

1042: Now for two such solutions hypercocycles

1043: $(i \dot I \nabla^{1,0}, a_1)$

1044:  and $(i \dot I \nabla^{1,0}, a_2)$ are cohomological and

1045:  the hypercocycle

1046: \begin{equation}

1047: (0, a_1 - a_2) = d(s) \sD'.

1048: \end{equation}

1049:  But from the definition of the complex (1.59) we have

1050:  \begin{equation}

1051: \op \sD' = 0 \implies a_1 - a_2 = \la s

1052: \end{equation}

1053: since every first order holomorphic operator is the multiplication

1054: by a constant. Geometrically this means that we get a holomorphic

1055: connection on

1056:  the projectivization of our vector bundle.

1057: 

1058: Thus our last task is to construct the lifting $A$ (1.74).

1059: But there is a canonical way to associate such lifting

1060:  with a holomorphic symmetrical tensor

1061: $W$ (1.40).

1062: 

1063: Consider the sheaf $D^2(L)$ of second order  holomorphic

1064: operators given by the

1065: standart extension

1066:  \begin{equation}

1067: 0 \to D^1(L) \to D^2(L) \to S^2(T^{1,0}) \to 0.

1068: \end{equation}

1069: Evaluating differential operators on a given section $s \in

1070: H^0(M_I, L_I)$ we get two homomorphisms

1071: \begin{equation}

1072: d_s \colon D^1(L) \to L

1073: \end{equation}

1074: $$

1075: d'_s \colon D^2(L) \to L

1076: $$

1077: giving two complexes: our complex (1.57) and its analog

1078: \begin{equation}

1079: 0 \to D^2(L) \to L \to 0.

1080: \end{equation}

1081: Considering the "trivial" third complex

1082: \begin{equation}

1083: 0 \to S^2(T^{1,0}) \to 0 \to 0

1084: \end{equation}

1085: we get the homomorphism of the vertical triples  to the triple

1086: $$

1087: 0 \to L \to L \to 0 \to 0

1088: $$

1089: and the exact triple of these complexes. The corresponding exact

1090: sequence of hypercohomology has the form

1091: \begin{equation}

1092: 0 \to H_h^0(d_s) \to H_h^0(d'_s) \to

1093: \end{equation}

1094: $$

1095: \to H^0(M_I, S^2(T^{1,0})) \to H_h^1(d_s) \to H_h^1(d'_s) \to \dots

1096: $$

1097: and every holomorphic quadratic form on the tangent bundle defines

1098: a  1-hypercohomology class (which may be zero). Of course our

1099: required map depends on a choice of a section $s$, so we denote

1100: this map by the symbol

1101: \begin{equation}

1102: \phi_s \colon H^0(M_I, S^2(T^{1,0})) \to H_h^1(d_s).

1103: \end{equation}

1104: The composition of this map with the forgetful homomorphism $\si$

1105: (1.73) gives the homomorphism

1106: \begin{equation}

1107: \mu_L \colon H^0(M_I, S^2(T^{1,0})) \to H^1(M_I, T^{1,0})

1108: \end{equation}

1109: which doesn't depend on the choice of $s$. More precisely

1110: \begin{equation}

1111: \mu_L (W_0) = -W * [c_1(L)] + \mu_{\Oh}

1112: \end{equation}

1113: It is quite natural to call it the "cohomological heat equation".

1114:  For example

1115: for polarized abelian varieties the map $\mu_\Oh = 0$. Moreover,

1116: all extensions $D^n(\Oh)$ are trivial hence all

1117:  coboundary homomorphisms are trivial.

1118: 

1119: 

1120: In particular, from this equality and (1.41) we can see that

1121: \begin{equation}

1122: \mu_\Oh (W_0) = - W_0 * 1/2 K

1123: \end{equation}

1124: because we can say more about sheaves of differential operators

1125: on the trivial line bundle. Namely the extension class for all

1126: jet-bundles is equal to

1127: \begin{equation}

1128: 1/2[K_{M_I}] \in \Om^{1,1}.

1129: \end{equation}

1130: From this it is easy to see that

1131: \begin{equation}

1132: \mu_L (W_0) = -W_0 * ([c_1(L)] + 1/2 [K_{M_I}])

1133: \end{equation}

1134: where $K_{M_I}$ is the canonical class of $M_I$ and the

1135:  cohomology class

1136: $[K_{M_I}]$ depends on $\om$ only (and not on the choice of an

1137: admissible complex

1138:  structure).

1139: 

1140:  Of course we can get more precise formula using representatives

1141:  of classes by

1142:  forms and using the Levi-Civita connections for tensor bundles

1143:   (see \cite{H1})

1144: 

1145: 

1146: 

1147: So the extended Kodaira-Spencer theory gives us (canonically) a

1148: holomorphic projective connection on the vector bundles (1.29) but

1149: of course to compute a curvature of one we have to present it as

1150:  a Dolbeault or Cech class.

1151: 

1152: An existence of Welters tensor $W_0$ (1.40) gives strong geometrical

1153: constraints. In particular $M_I$ can't be a variety of general

1154: type. More precisely the canonical class $K$ of $M_I$ can't be

1155: positive.

1156: 

1157: \subsection{Faithful functors}

1158: 

1159:       Suppose our $M_I$ which we consider as a point $I$ of

1160:        the moduli space $\sM$

1161: of all its deformation defines a new polarized algebraic variety

1162: $M'_{I'}$  which also we consider as a point $I'$ of the

1163: moduli space $\sM'$

1164: such a way, that

1165:  we can reconstruct $M_I$ (from the geometry of $M'_{I'}$).

1166:  Geometrically, mapping $M_I$ to $M'_{I'}$ defines a holomorphic

1167: embedding of moduli spaces

1168: \begin{equation}

1169: J \colon \sM \to \sM',

1170: \end{equation}

1171: and $J(I) = I'$. If a morphism of polarized varities from $\sM $

1172: induces a morphism of corresponding varieties from $\sM'$ then

1173: the map $M_I \to M'_{I'}$ is called a {\it faithful

1174:  functor}. From geometrical point of view these geometrical

1175:  images are  undistinguishable in spite of the fact

1176:  that these varieties can have different dimensions and so on.

1177:   An algebraic geometer is skillful enough if he can recognize

1178:   such geometrically undistinguishable varieties of different

1179:    dimensions and different shapes.

1180: 

1181: For example, a smooth intersection

1182: \begin{equation}

1183: X = Q_1 \cap Q_2 \subset \PP^5

1184: \end{equation}

1185:  of two quadrics in five dimensional projective space is

1186:   geometrically equivalent

1187:  to an algebraic curve $\Si_I$ of genus 2.

1188:  To reconstruct $\Si_I$ from $X$ it is enough to remark that

1189: \begin{enumerate} \item By the Lefshetz theorem $X$ is

1190: simply connected;

1191: \item $\Oh_X(-2) = K_X$ - the canonical class of $X$.

1192:  Hence the embedding of $X$ into $\PP^5$ is canonical.

1193: \item The space of all quadrics in $\PP^5$ through $X$ is

1194:  a pencil

1195: \begin{equation}

1196:  \vert 2 H - X \vert = \PP^1

1197: \end{equation}

1198: that is parameterized by the projective line.

1199: \item The subset of singular quadrics in this pencil is

1200:  six different points

1201: \begin{equation}

1202:  W =p_1 \cup ... \cup p_6 \subset \PP^1.

1203: \end{equation}

1204: \item The double cover

1205: \begin{equation}

1206:  \phi \colon \Si_I \to \PP^1

1207: \end{equation}

1208:  with  ramification in these six points $W$ is the required curve

1209: $\Si_I$ of genus 2.

1210: \end{enumerate}

1211: 

1212: Note that the previous double cover is given by the canonical

1213:  linear system of $\Si_I$.

1214: Now a simple exercise in  linear algebra shows that there exists

1215: just one pencil of quadrics in $\PP^5 $ up to a linear

1216: transformation with the subset of singular

1217:  quadrics which is equal to the  ramification locus of the

1218:   double cover.

1219: 

1220: If $M_I \to M'_{I'}$ is a faithful functor and geometries of both

1221: varietes are undistinguishable it is quite reasonable to consider

1222: a quantization of $\sM'$ as a quantization of $\sM$, that is,

1223:  to say

1224: \begin{equation}

1225: Q(M_I) = Q(M'_{I'})

1226: \end{equation}

1227: and so on ... We will say that such quantization is successful if

1228: the

1229:  quantization of $\sM$ is successful, that is, corresponding

1230:   vector bundles (1.29)

1231: on $\sM'$ admit projective flat holomorphic connections.

1232: 

1233: 

1234: \subsection{Perfect quantization}

1235: 

1236: Suppose our algebraic variety $(M_I, \Oh(1))$ with a Hodge form

1237: $\om$ admits as a symplectic manifold a real polarization that is

1238: a Lagrangian fibration:

1239: \begin{equation}

1240: \pi \colon M \to B , \quad \pi^{-1}(b_{gen}) = T^n

1241: \end{equation}

1242: that is the generic fiber is  a Lagrangian torus of dimension

1243: $n = 1/2 dim M$;

1244: 

1245: Again we may suppose that the subset $BS_k \subset B$ is

1246:  finite and the Hilbert space of wave functions of level $k$

1247:  is given by the direct sum of lines

1248: \begin{equation}

1249:  Q_F^k(M) = \sH_\pi^k = \oplus_{b \in BS_k \subset B} \C \cdot

1250: s_b

1251: \end{equation}

1252: where $s_b$ is a covariant constant section of $(L^k,

1253: \nabla_k)\vert_{\pi^{-1}(b)}$).

1254: 

1255: Suppose we can construct a natural isomorphism

1256: \begin{equation}

1257:   \PP Q_F^k(M) = \PP H^0(M_I, \Oh(k)).

1258: \end{equation}

1259: Then we have a perfect quantization providing a lot of

1260:  beautiful and very important

1261: properties.

1262: \begin{enumerate}

1263: \item First of all, both kinds of quantization are successful

1264:  : indeed, the LHS of the equality (1.100) doesn't depend on a complex

1265:   polarization at all. On the other hand, RHS doesn't depend on

1266:   a real polarization.

1267: \item LHS of the equality is decomposed into a sum of lines

1268:  so the RHS as a space of sections admits a special basis

1269:  (may be after fixing some additional structure like  so called

1270: theta - structure for abelian varieties \cite{Mum}).

1271: 

1272: \item A basis of such type is called the Bohr-Sommerfeld basis.

1273: \item Such situation can be called  {\it perfect quantization.}

1274: \end{enumerate}

1275: 

1276: Moreover, the first example of a perfect quantization is given by

1277: the classical theory of theta-functions.

1278: 

1279: 

1280:  Obviously it is not the simplest example.

1281: The simplest example is the following: consider the standard two

1282: sphere $S^2$ realized as the complex Riemann sphere with the

1283: standard $U(1) = S^1$-action (rotations around North and South

1284: Poles). Now  $U(1)$-invariant symplectic form $\om$ of volume 1

1285: gives the phase space of the classical mechanical system $S^2,

1286: \om$. To quantize this one we have to fix a complex structure, but

1287: it is unique and the real polarization

1288: \begin{equation}

1289:   \pi \colon S^2 \to [-1, 1]

1290: \end{equation}

1291: is given by the projection to the rotation axis.

1292: 

1293: As a prequantization date we get the line bundle $\Oh(1)$ of

1294: degree 1 with the standard $U(1)$-invariant unitary  connection

1295: $\nabla$. Then

1296:  the wave function spaces

1297: \begin{equation}

1298:   \sH^k_I = H^0(\Oh(k))

1299: \end{equation}

1300: and $U(1)$- action on this space defines the special eigenbasis of

1301:  Fourier polynomials. Fourier monomials

1302:   are in 1-1 correspondence with

1303:  Bohr-Sommerfeld fibers of the projection (1.101) in angle

1304:  coordinates.

1305:  Thus in this case we have the perfect quantization.

1306: 

1307:  It is easy to see that in this case all "good" theoretical

1308:  conditions

1309:  as (1.48) are

1310:  disturbed but the result of quantization is successful.

1311: 

1312:  If we put out the word "natural" from the statement (1.100)

1313:   we get the statement

1314:  about ranks of both wave functions spaces. In this case we say that

1315:  the quantization

1316:  is {\it numerically} perfect. The geometry behind this

1317:   numerical coincidence

1318:  will be described in (2.25) - (2.30) and \cite{T3}.

1319:  Such rank's equalities

1320:  were proved for K3 surfaces cases and many others. But historically first case

1321:  was the Gelfand-Cetlin system (see \cite{GS2}).

1322: 

1323: 

1324: 

1325:  \section{Algebraic curves = Riemann surfaces}

1326: 

1327: \subsection{Direct approach}

1328: 

1329: 

1330: 

1331:  Let $\Si$ be a compact smooth oriented Riemann surface and

1332: $\Si_I$ be a complex structure on it, thus, one has an algebraic

1333: curve of genus $g > 1$. Such curves admit the canonical

1334:  polarization by the cotangent bundle $T^* \Si_I = \Oh(K)$

1335:   where $K$ is the canonical class of $\Si_I$

1336:   (see for example \cite{DSS}). Moreover, if we fix a metric

1337:   on $\Si$ with the conformal

1338:  class given by the complex structure $I$ then

1339:  the Levi-Civita connection on the cotangent

1340:  bundle gives the prequantization data $(T^*\Si, \nabla_{LC})$.

1341:  This pair is equivalent to

1342: the holomorphic pair $(\Si_I, \Oh(K))$.

1343: 

1344: The union of all spaces $H^0(\Oh(k K))$ of holomorphic sections

1345: gives the collection of vector bundles

1346: \begin{equation}

1347: p_k \colon V_k \to \sM_g, \quad p^{-1}(M_I) = H^0(M_I, \Oh(k K))

1348: \end{equation}

1349: on the moduli space $\sM_g$ of curves of genus $g$.

1350: (Recall that $\dim \sM_g= 3g-3$).

1351: 

1352: %\begin{figure}[tbn]

1353: %\centerline{\epsfxsize=2in\epsfbox{24.eps}}

1354: %\caption{\sl Riemann surface = curve}

1355: %\label{F5}

1356: %\end{figure}

1357: 

1358: In particular, for $k=2$ the vector bundle $V_{2} = T^* \sM_g$ is

1359: the cotangent bundle of the moduli space, that is, $V_{2}$

1360:  is the bundle with spaces of quadratic differentials as fibers.

1361: In particular the canonical class $K_{\sM_{g}}$ is the first Chern

1362: class of the vector bundle $V_{2}$:

1363: \begin{equation}

1364:  c_1 (V_2) = K_{\sM_g}.

1365: \end{equation}

1366: Now according to Mumford  this canonical class has

1367: the decomposition

1368: \begin{equation}

1369: K_{\sM_g} = 13 \Theta + \text{boundary divisors $\dots$}

1370: \end{equation}

1371: where $\Theta$ is the diviser of zero theta constants.

1372: 

1373: If the vector bundle $V_{2}$ admits a holomorphic projective flat

1374: connection then by

1375:  (1.31) the number 13 has to be divided by $3g-3$.

1376:  This isn't the case thus $V_{2}$

1377:  doesn't admit

1378: any flat connection and an algebraic curve as a polarized

1379: algebraic variety isn't a

1380:  result of

1381: a successful quantization procedure of the classical dynamical

1382: system  $(\Si, \om)$ where $\om$ is the restriction of the

1383: Fubini-Study form.

1384: 

1385: But the history isn't finished! We can find a "faithful functor".

1386: 

1387: \subsection{Jacobians}

1388: 

1389: May be the first faithful functor for algebraic curve is mapping

1390: of

1391:  the curve to the

1392: polarized jacobian

1393: $(J(\Si_I), \Theta )$. There exists a lot of constructions but

1394:  for us it will be very convinient

1395: to consider $J(\Si_I)$ as the {\it moduli space of holomophic

1396: topological trivial line bundles}. Fixing a point $p_0 \in \Si_I$

1397: we consider for every other point $p\in \Si_I$ the topological

1398: trivial line bundle $\Oh(p - p_0)$, where we consider points as

1399: divisors of degree 1. So we get the map

1400: \begin{equation}

1401: a_{p_0} \colon \Si_I \to J(\Si_I).

1402: \end{equation}

1403: By the Riemann theorem this map is an embedding (if genus $g>0$)

1404: and induces the isomorphism

1405: \begin{equation}

1406: a \colon H_1(\Si, \Z) \to H_1(J(\Si_I), \Z).

1407: \end{equation}

1408: The moduli space of line bundles is a group (with respect to

1409: the tensor product of line bundles). So, we can consider the Pontrjagin map

1410: \begin{equation}

1411: a_{p_0}^{g-1} \colon \Si_I^{[g-1]} \to J(\Si_I)

1412: \end{equation}

1413: sending a finite set of points $\{p_1, ... , p_{g-1}\}$ to

1414:  the line bundle

1415: \begin{equation}

1416: \Oh(p_1 + ... + p_{g-1} - (g-1)p_0) \in J(\Si_I).

1417: \end{equation}

1418: The image of this map is a divisor $\Theta$ (so called

1419:  {\it theta diviser}) and the

1420: corresponding line bundle admits only one (up to $\C^*$)

1421: holomorphic section (with

1422: the zero set $\Theta$).

1423: 

1424: To reconstruct the curve $\Si_I$ from the pair

1425:  $(\Theta \subset J(\Si_I))$ we have to consider

1426:  the Gauss map

1427: \begin{equation}

1428: G \colon \Theta \to \PP T^* J(\Si_I)

1429: \end{equation}

1430: to the projectivization of the cotangent bundle sending

1431:  non singular point of $\Theta$ to  the tangent hyperplane

1432:   to $\Theta$ at this point.

1433:   But the cotangent bundle (of a group) is

1434:   trivial $\PP T^* J(\Si_I) = \PP^{g-1} \times  J(\Si_I)$, and

1435:   the composition of the projection to $\PP^{g-1}$ and

1436:    the Gauss map gives the rational cover

1437: \begin{equation}

1438: G \circ p \colon \Theta \to \PP^{g-1}.

1439: \end{equation}

1440: Now the  ramification diviser of this cover is the dual divisor

1441:  to

1442: the canonical embedding of $\Si_I$ into $(\PP^{g-1})^*$ (for

1443: hyperelliptic curves we use other arguments). By the classical

1444: projective duality theory we can reconstruct a curve by its dual

1445: divisor. Thus the sending of a curve $\Si_I \to (J(\Si_I),

1446: \Theta)$ is a faithful functor.

1447: 

1448: Now using our rule (1.97) we may quantize the polarized variety

1449:  $(J(\Si_I), \Theta)$.

1450: 

1451: The role of wave function spaces now are played by the spaces of

1452:  holomorphic sections:

1453: \begin{equation}

1454: \sH_{I}^k = H^0(J(\Si_I), \Oh (k \Theta)).

1455: \end{equation}

1456: These spaces are called spaces of theta function of

1457:  the Riemann surface $\Si_I$ of level $k$.

1458: 

1459: This quantization is successful! Corresponding vector bundles

1460:  on

1461: the moduli space admit projective flat connections (see

1462: \cite{W}). But the situation is much better than just a successful

1463: quantization.

1464: 

1465: 

1466: 

1467: \subsection{Algebro geometrical theory of theta functions}.

1468: 

1469: Consider a principal polarized abelian variety $(A, \Theta)$ of

1470:  dimension $g$. Then the line bundle $\Oh(k \Theta)$ admits

1471:  the finite group $K$ of translations preserving this line bundle.

1472: It is easy to see that

1473: \begin{equation}

1474:   K = A_k \subset A

1475: \end{equation}

1476: is the subgroup of points of order $k$ in the abelian group $A$.

1477:  Recall that we have a symplectic form on $K$ given by the wedge

1478:   product of 1-cycles coupled by the polarization $\Theta$.

1479: The decomposition

1480: \begin{equation}

1481:   K = K_+ \times K_-

1482: \end{equation}

1483: by two isotropic subgroups of this group is called

1484: theta-structure of level $k$.

1485: 

1486: The beautiful classical result reproduced by Mumford is the following

1487: \begin{enumerate}

1488: \item In the complete linear system $\vert  k \Theta \vert$ there

1489: exists only one divisor $D_+$ which is invariant with respect to

1490: the subgroup $K_+$ - action.

1491: \item Every translation $a \in K_-$ defines a divisor $D_+ + a$

1492: and the rational function $\theta_a$ with pole at the divisor

1493: $D_+$ and zeros at the divisor $D_+ + a$. This function is called

1494:  theta function with  characteristic $a \in K_-$.

1495: \item The space of holomorphic sections

1496: \begin{equation}

1497:  H^0(A, \Oh (k \Theta)) = H^0(A, \Oh(D_+)) = \oplus_{a \in K_-} \C \theta_a.

1498: \end{equation}

1499: \item This is the theta functions with characteristics decomposition.

1500: \end{enumerate}

1501: 

1502: Note, that the cardinalities of our finite abelian groups are

1503: given by formulas

1504: \begin{equation}

1505:  \vert K \vert = k^{2g} , \quad \vert K_{\pm} \vert = k^{g}.

1506: \end{equation}

1507: 

1508: Now it is quite useful to apply the extended Kodaira-Spenser

1509: theory to this situation. First of all for every abelian variety

1510: \begin{equation}

1511:   H^0(A, T) \to H^2(A, \Oh)

1512: \end{equation}

1513: is an isomorphism thus the condition (1.48) holds. Moreover, from

1514: cohomology sequence of (1.50) we have

1515: \begin{enumerate}

1516: \item an isomorphism

1517: \begin{equation}

1518:  H^0(A, S^2(T)) = H^1(A,D^1(k \Theta)));

1519: \end{equation}

1520: \item

1521: \begin{equation}

1522:   H^0(A, \Oh) = H^0(A, D^1(\Oh(k \Theta)));

1523: \end{equation}

1524: \item and from (1.60)

1525: we have

1526: \begin{equation}

1527:  0 \to H^0(A, \Oh(k \Theta) )/ \C \cdot s \to H_h^1(d_s) \to

1528:  H^1(A, D^1(\Oh(k \Theta))) \to 0;

1529: \end{equation}

1530: \item so under a deformation of the pair $(A, \Oh(k \Theta))$

1531: for every section $s$ there exits  unique deformation of this

1532: section which is {\it defined by the heat equation} (see the

1533: standard presentation (2.49) below).

1534: \end{enumerate}

1535: 

1536: \subsection{Symplectic-combinatorial theory of theta functions}

1537: 

1538: A principal polarization $\Theta$ on an abelian variety $A$ can

1539: be done by

1540:  some unimodular skew symmetrical integer form

1541: \begin{equation}

1542: \om  \in H^2(A, \Z)

1543: \end{equation}

1544: because  $A $ is a $2g$-dimensional real torus:

1545: \begin{equation}

1546: \om \in \wedge^2 H^1(A, \Z) = H^2(A,\Z).

1547: \end{equation}

1548: For a jacobian $J(\Si)$ this form is induced by the intersection

1549: form on 1-cycles on the Riemann surface $\Si$.

1550: 

1551: The theory of unimodular integer skew symmetrical forms predicts

1552:  that there are two isotropic $\Z$-sublattices in $H^1(A, \Z)$

1553:  such that our form $\om$ has the standard symplectic shape.

1554: 

1555: Recall that the underlying smooth manifold of our abelian variety

1556: is nothing else but a real torus $T^{2g}$ and the decomposition

1557:  of  $H^1(A, \Z)$ by two isotropic $\Z$-submodules  induces

1558:  the corresponding decomposition of this torus into two families of Lagrangian

1559:  subtori

1560: \begin{equation}

1561: T^{2g} = T^g_+ \times T^g_-.

1562: \end{equation}

1563: Considering the projection to the second $g$-torus

1564: \begin{equation}

1565: \pi \colon T^{2g} \to T^g_-

1566: \end{equation}

1567: we obtain an integrable system that is a real polarization

1568: of the torus $T^{2g} = A$.

1569: 

1570: %\begin{figure}[tbn]

1571: %\centerline{\epsfxsize=2in\epsfbox{35.eps}}

1572: %\caption{\sl tori fibration}

1573: %\label{F6}

1574: %\end{figure}

1575: 

1576: \begin{figure}[tbn]

1577: \centerline{\epsfxsize=3in\epsfbox{35c.eps}}

1578: \caption{\sl Bohr-Sommerfeld fibers}

1579: \label{Fig 2}

1580: \end{figure}

1581: 

1582: The direct interpretation of the prequantization line bundle

1583: shows that the subset of $k$-Bohr-Sommerfeld fibers

1584: \begin{equation}

1585: BS_k = (T^g_-)_k \subset T^g_-

1586: \end{equation}

1587: is the subgroup of points of order $k$ on our second torus

1588: $T^g_-$  (see Fig.5). So, the wave function space of the

1589: Bohr-Sommerfeld

1590:  quantization is the sum of lines

1591: \begin{equation}

1592: Q_\pi^k (T^{2g}, \om) = \oplus_{a \in (T^g_-)_k} \C \cdot s_a

1593: \end{equation}

1594: where $s_a$ are covariant constant sections of restrictions of

1595:  the prequantization line bundle.

1596: 

1597: 

1598: We can see that the results of the Kahler quantization and the

1599: Bohr-Sommerfeld quantization give wave function spaces of the

1600: same rank so the quantization is numericaly perfect (see the end

1601: of subsection 1.5).

1602: 

1603: This is the first amazing fact in this story.

1604: We would like to say couple of words about the geometry behind

1605: such coincidence.

1606: 

1607: Returning to the fibration (2.22) we can construct so called

1608: {\it dual fibration}

1609: \begin{equation}

1610: \pi' \colon (T')^{2g} \to T^g_-

1611: \end{equation}

1612: changing every torus-fiber $T^g_+$ by the dual torus $(T^g_+)^*$.

1613: This dual torus is nothing else but the space of gauge classes of

1614: all $U(1)$-flat connections on the trivial line bundle on

1615: $T^g_+$. We have got  a priori a new $2g$-torus with  new

1616: symplectic form $\om'$ of the same type as before.

1617: 

1618: Now consider any holomorphic line bundle  $L$ on our old

1619: complex torus $A$ with $U(1)$-connection which  curvature

1620: form has to be proportional to $\om$. (Just like our

1621: prequantization bundle $\Oh(k)$). Then for every point

1622: $a \in T^g_+$ the restriction of $L$ to the Lagrangian fiber

1623:  $\pi^{-1}(a)$ gives a flat line bundle, that is the point of the

1624:  dual fibers

1625: \begin{equation}

1626: s_L \in (T^g_+)^* = (\pi^{-1}(a))^*.

1627: \end{equation}

1628: Thus every such line bundle defines a section

1629: \begin{equation}

1630: s_L \colon  T^g_+ \to (T')^{2g},

1631: \end{equation}

1632: that is a middle dimensional submanifold $s_L$ (more precisely

1633: $s_L(T^g_+)$) in the manifold $(T')^{2g}$.

1634: 

1635: Of course  trivial connection defines  zero section of the

1636:  new fibration (2.25) that is we have a middle dimensional

1637:   submanifold

1638: \begin{equation}

1639: s_{\Oh(0)} \subset (T')^{2g}.

1640: \end{equation}

1641: 

1642: Obviously, the intersection set

1643: \begin{equation}

1644: s_{\Oh(0)} \cap s_{\Oh(k)} = BS_k

1645: \end{equation}

1646: is the set of Borh-Sommerfeld fibers of level $k$. Moreover, it

1647: is easy to see that both  our cycles are Lagrangian, oriented

1648:  and admit positive index intersections only. Thus

1649: the number of Bohr-Sommerfeld fibers is given by the

1650:  intersection index of cohomology classes

1651: \begin{equation}

1652: [s_{\Oh(0)}] \cdot [ s_{\Oh(k)}] = \# BS_k.

1653: \end{equation}

1654: Now using the first Chern class of the polarization we can easily

1655: compute these numbers.

1656: 

1657: The moral is the following: an abelian variety $A = T^{2g}$ has

1658:  a symplectic partner $(T')^{2g}$

1659: such that coherent sheaves geometry on $A$ can be encoded by the

1660: geometry of Lagrangian cycles (or super cycles) on the partner

1661:  $(T')^{2g}$. Usually this partner is called

1662:  {\it mirror partner}.

1663:   Mirror symmetry now is  far developed domain so we

1664:   can't  dive to this subject (see \cite{CK}).

1665:   In spite of the coincidence of our previous construction with

1666:   the Strominger-Yau-Zaslow mirror symmetry construction we would

1667:   like to stop here discussion of this subject but formally it can be extended

1668:   along this line (see \cite{T3}).

1669: 

1670:  Of course the dimensional coicidence can be also extended. Moreover there are

1671:   theta bases in both of wave function spaces.

1672:   To compare them we have to construct

1673:   a holomorphic object, namely, a section

1674:   of the theta bundle of level $k$ using only

1675:    Bohr-Sommerfeld fiber of the projection $\pi$ (2.22) (see Figure 1).

1676: 

1677:   We will do this using so called {\it coherent state

1678: transform} or the slightly generalized

1679: Segal-Bargmann isomorphism introduced in the

1680: context of the quantum theory as a transform from the Hilbert

1681: space of  square integrable functions on the configuration

1682: space to the space of holomorphic functions on the phase space.

1683: 

1684: Considering  zero fiber $T^g_+$ of the polarization map $\pi$

1685: (2.22) we can interpret our complex torus $A$ as a

1686: complexification of this real torus and input our geometrical

1687: situation to the situation of the classical coherent state

1688: transform.

1689: 

1690:  In the finite dimensional context a configuration space is

1691:  just $\R^g$ as the real part of the phase space $\C^g$.

1692:   The  coherent state transform

1693: \begin{equation}

1694: CST_t \colon L^2(\R^g, d^g x) \to L^2(\C^g, d\mu_t) \cap \sH(\C^g)

1695: \end{equation}

1696: is a unitary isomorphism.(Here $d\mu_t$ is the Gaussian measure

1697: and $\sH$ is the space of holomorphic functions.)

1698: 

1699:  For our abelian case we have to replace $\R^g$ by $T^g = U(1)^g$ and $\C^n$

1700:  by our complex

1701: torus $A$. Remark

1702:  that $T^g$ in our case is a special Lagrangian

1703: subtorus (see, for example \cite{T4}).

1704: 

1705: 

1706:  B. C. Hall proposed a generalization of the CST where $\R^n$ is replaced

1707: by $\SU(2)$ and $\C^n$ by $SL(2, \C)$ (see

1708:  \cite{Ha}) and we will use it later for non-abelian case. But here we use this

1709:  construction for  abelian group $U(1)^g \subset (\C^*)^g$.

1710: 

1711:  First of all, the decomposition (2.21) induces the decomposition

1712:  \begin{equation}

1713: H^1(A, \Z) = H^1(T^g_+, \Z) \oplus  H^1(T^g_-, \Z).

1714: \end{equation}

1715: Let us fix a basis $(a_1, \dots, a_g)$ in the lattice $H^1(T^g_+, \Z)$ and

1716: a basis $(b_1, \dots, b_g)$ in the lattice

1717:  $ H^1(T^g_-, \Z)$ such that the full system $(a_1, \dots, a_g ,b_1, \dots, b_g)$

1718: is a standard basis of 1-homology of $A$. For a complex Riemann

1719: surface $\Si$

1720:  consider the period matrix

1721: \begin{equation}

1722: \Om = \Vert \Om_{ij}\Vert = Re \Om + i \cdot Im \Om

1723: \end{equation}

1724:   as a point of the Siegel space $H$.

1725:   Let $\vec x =(x_1, \dots, x_g) \in [0,1]^g$

1726: be periodic coordinates on $U(1)^{g} = T^g_+$. Then we have

1727:  $U(1)$-invariant

1728: Laplacian

1729: \begin{equation}

1730: \De^{Im \Om} = \sum_{i,j = 1}^g \frac{Im \Om_{ij}}{2 \pi} \frac{\p^2}{\p x_i

1731: \p x_j}.

1732: \end{equation}

1733: The complexification of $T^g_+$ is $(\C^*)^g$ with coordinates $(e^{2\pi i z_1},

1734: \dots,

1735: e^{2\pi i z_g})$ for $\vec z = \vec x + i \vec y$. Then the Haar mesure on

1736:  $(\C^*)^g$ is $d \vec x d \vec y$ and we have  $U(1)^g$-invariant

1737:   complex

1738:  Laplacian

1739: \begin{equation}

1740: \Delta^{Im \Om}_\C = \sum_{i,j = 1}^g \frac{Im \Om_{ij}}{2 \pi} (\frac{\p^2}{\p x_i

1741: \p x_j} + \frac{\p^2}{\p y_i

1742: \p y_j}).

1743: \end{equation}

1744: 

1745: As the first step consider the fundamental solution $\mu_t$ at the

1746: identity of the

1747:  heat equation on $(\C^*)^g$

1748: \begin{equation}

1749: 4 \frac{\p u}{\p t} = \Delta^{Im \Om}_\C  u

1750: \end{equation}

1751: and average the measure $\mu_t d \vec x d \vec y$ with respect to the action of

1752: $U(1)^g$ :

1753: \begin{equation}

1754: \nu_t(\vec z) d \vec x d \vec y =

1755: (\int_{U(1)^g} \mu_t(\vec z + \vec x) d \vec x) \quad d \vec x d \vec y.

1756: \end{equation}

1757: The explicit formula is

1758: \begin{equation}

1759: \nu_t(\vec z) = (\frac{2}{t})^{g/2} (det (Im \Om))^{-1/2} e^{\frac{\pi}{2t}

1760: \sum_{i,j}

1761: (z_i - \ov{z_i}) (Im \Om)^{-1}_{ij}(z_j - \ov{z_j})}.

1762: \end{equation}

1763: Now consider  non-self adjoint Laplace operator

1764: \begin{equation}

1765: \Delta^{-i \Om} = \sum_{i,j = 1}^g \frac{i}{2 \pi} \Om_{ij}\frac{\p^2}{\p x_i

1766: \p x_j}.

1767: \end{equation}

1768: C. Florentino, J. Mourao and J. Nunes proved in \cite{FMN} the

1769: following

1770: \begin{prop} Let $\sC$ be the analytic continuation from $U(1)^g$ to

1771: $(\C^*)^g$. Then the transform

1772: \begin{equation}

1773: C_t^{(-i \Om)} = \sC \circ e^{t/2 \Delta^{(-i \Om)}} \colon L^2(U(1)^g, d \vec x)

1774: \to

1775: \end{equation}

1776: $$

1777: \to L^2((\C^*)^g, d\nu_t) \cap \sH((\C^*)^g)

1778: $$

1779: is unitary.

1780: \end{prop}

1781: So for any function $f \in L^2(U(1)^g, d \vec x)$ given by its Fourier

1782:  decomposition

1783: \begin{equation}

1784: f(\vec x) = \sum_{\vec n \in \Z^g} a_n e^{2\pi i \vec x \cdot \vec n}

1785: \end{equation}

1786: we have

1787: \begin{equation}

1788: (C_t^{(-i \Om)} f)(\vec z) =

1789: \sum_{\vec n \in \Z^g} a_n e^{ti\pi \vec n \Om \vec n}\cdot

1790:  e^{2\pi i \vec x \cdot \vec n}.

1791: \end{equation}

1792: This function is the analytic continuation to $(\C^*)^g$ of the solution of

1793: the complex heat equation

1794: \begin{equation}

1795: 2 \frac{\p u}{\p t} = \Delta^{(-i \Om)} u

1796: \end{equation}

1797: on $(\C^*)^g$ with the initial condition given by f.

1798: 

1799: But the coherent states transform (2.40) can be extended from

1800: $L^2(U(1)^g, d \vec x)$ to the space of distributions $(C^\infty(U(1)^g))' $

1801: given by Fourier series of the form (2.41) such that there exists an integer

1802: $N > 0$ such that

1803: \begin{equation}

1804: lim_{\vec n \cdot \vec n \to \infty} \frac{| a_n |}{(1 + \vec n \cdot \vec n)^N}

1805: = 0.

1806: \end{equation}

1807: The Laplace operator and its powers act as continuous linear operators on this space of distributions

1808: (by duality from the corresponding action on $C^\infty(U(1)^g)$)

1809: and define for ($t > 0$) the action of the operator $e^{t/2} \Delta^{(-i \Om)}$ on

1810: distributions $f$ of the form (2.41) as

1811: \begin{equation}

1812: e^{t/2} \Delta^{(-i \Om)} (\sum_{\vec n \in \Z^g} a_n e^{2\pi i \vec x \vec n}) =

1813: \end{equation}

1814: $$

1815: = \sum_{\vec n \in \Z^g} a_n e^{t i \pi \vec n \Om \vec n}

1816: e^{2\pi i \vec z \vec n}.

1817: $$

1818: 

1819: 

1820: In the same paper \cite{FMN} the authors proved the following

1821: \begin{prop} If a series $f \in (C^\infty(U(1)^g))' $ then the RHS of (2.45)

1822: defines a holomorphic function on

1823: $(\C^*)^g$.

1824: \end{prop}

1825: Now we are ready to map every Bohr-Sommerfeld fiber of the

1826: projection (2.22) to an analytic function on $(\C^*)^g$. It is

1827: enough to do it for  zero torus $T^g_+ = U(1)^g$ of this

1828: projection.

1829: 

1830: For this let us consider the distribution

1831: \begin{equation}

1832: \theta^\R_0 (x) = \sum_{\vec n \in \Z^g} e^{2\pi i k \vec x \vec n}.

1833: \end{equation}

1834: It is nothing else but the delta-function at the identity of level

1835: $k$. Let us apply  CST (2.40) to this distribution:

1836: \begin{equation}

1837: C_{1/k}^{(-i \Om)} (\theta_k^\R) =

1838: \end{equation}

1839: $$

1840: = \sum_{\vec n \in \Z^g} e^{\pi i k \vec n \cdot \Om / k  \vec n k } \cdot e^{

1841: 2 \pi i k\vec n \vec z}.

1842: $$

1843: For every $l \in \Z^g / k \Z^g$ we can start with the function

1844: \begin{equation}

1845: \theta^\R_l (x) = \sum_{\vec n \in \Z^g} e^{2\pi i  \vec x (l+ k\vec n)}

1846: \end{equation}

1847: to get the analytic function

1848: \begin{equation}

1849: C_{1/k}^{(-i \Om)} (\theta_l^\R) = \theta_l(z, \Om) =

1850: \end{equation}

1851: $$

1852: = \sum_{\vec n \in \Z^g} e^{\pi i (l+ k \vec n )\cdot \Om / k (l

1853: + k\vec n )} \cdot e^{ 2 \pi i (l + k\vec n ) \vec z}.

1854: $$

1855: 

1856:  Remark that we are substitute  our continious

1857: positive parameter $t$ coming from

1858:  heat kernels by its discrete conterpart $1/k$.

1859: 

1860: 

1861: We would like to emphasize that in spite of 150 years of

1862: development of the classical theory of theta functions  this

1863: "comparing quantization" approach was realized first quite

1864: recently by C. Florentino, J. Mourao and J. Nunes in \cite{FMN}.

1865: 

1866: What we have to do now is just to compare our holomorphic functions on

1867: $(\C^*)^g$ with holomorphic sections of the $\Theta$-line bundle on

1868: $A$.

1869: 

1870: \subsection{Abelian holomorphic flat connections}

1871: 

1872: The usual way to define a line bundle $L$ is

1873:   to consider some finite cover $\{U_i\}$ on the base and

1874:   a collection of {\it transition functions} $f_{ij}$ which are

1875:   regular

1876:    and regular inversed on $U_i \cap U_j$.

1877: On the intersection $U_i \cap U_j \cap U_k$ we have the cocycle

1878: equality

1879: \begin{equation}

1880: f_{ij} \cdot f_{jk} \cdot  f_{ki} = 1.

1881: \end{equation}

1882: A collection $f_{ij}$ is equivalent to a collection $f'_{ij}$

1883: iff there exists a collection $\{\beta_i\}$ of functions, where each

1884: $\beta_i$ is regular and regular inversed on $U_i $ such that

1885: \begin{equation}

1886: f'_{ij} = \beta_i f_{ij} \beta_j^{-1}.

1887: \end{equation}

1888: 

1889: 

1890: 

1891: The idea is to make these functions (2.51) as simple as possible using

1892: this equivalence: for example take all $f'_{ij}$

1893:  constant.

1894: 

1895: Starting with any collection of functions (2.50) let us

1896:  consider the collection of  differentials

1897: forms

1898: \begin{equation}

1899: \{  f_{ij}^{-1} d f_{ij} \} \in Z^{1}(\{U_{i} \} )

1900: \end{equation}

1901: Obviously, this cochain is a cocycle from $H^1(\Om) = H^{1,1}$

1902: whose cohomology class

1903: \begin{equation}

1904: c(L) \in H^{1,1}

1905: \end{equation}

1906: is the first Chern class of $L$.

1907: 

1908: If this class is zero  then the cocycle is trivial and there are

1909: matrices of differential forms $h_i$ such that

1910: \begin{equation}

1911: f_{ij}^{-1} d f_{ij} = h_i - h_j.

1912: \end{equation}

1913:  \begin{dfn} A collection of  differential forms $h_i$

1914:   on $U_i$ is called  flat holomorphic connection of the

1915:    vector bundle $L$ given by the collection of transition

1916:     functions

1917: $f_{ij}$.

1918: \end{dfn}

1919: 

1920: Having such collection we can solve  the system of

1921: linear equations on $U_i$

1922: \begin{equation}

1923: \beta_i^{-1} \cdot d \beta_i = h_i

1924: \end{equation}

1925: and obtain a new collection of transition functions (2.51).

1926: Now using all equations we can see that functions

1927: $f'_{ij} = \beta_i \al_{ij} \beta_j^{-1}$  are constant, that is

1928: \begin{equation}

1929:  d f'_ij = 0.

1930: \end{equation}

1931: If transition functions are constant then this vector bundle is

1932: called

1933:  {\it local system of coefficients} and we can trivialize it over

1934:  any simply connected open set. Thus  we have a  character of

1935:  the fundamental group of our base:

1936: \begin{equation}

1937: \rho \colon \pi_1 \to  \C^*.

1938: \end{equation}

1939: 

1940: For a curve we've proved the classical result of Poincare:

1941: 

1942: \begin{prop}

1943: Every $L \in J(\Si)$ admits  holomorphic flat

1944:  connection

1945:  given by a character $\chi \colon \pi_1(\Si) \to \C^*$.

1946: \end{prop}

1947: 

1948: How many characters  give the same vector bundle?

1949:  Just  as many as flat connections admitted by this bundle there are.

1950:  The difference of any two holomorphic connections can be

1951:  identified over

1952: $U_i \cap U_j$, thus

1953: \begin{equation}

1954: \{h_i\} - \{ h'_i\} \in H^0(\Oh_\Si (K_\Si)).

1955: \end{equation}

1956: 

1957: By Serre duality

1958: \begin{equation}

1959: H^1(\Oh)^* = H^0( \Oh( K_\Si))

1960: \end{equation}

1961:  is a fiber of

1962: the cotangent bundle $T^* J(\Si)$.

1963: 

1964: The full space $A$ of all

1965:  holomorphic flat connections is the space of all characters

1966: \begin{equation}

1967: A = (\C^*)^{2g}, \quad dim_\C A = 2g

1968: \end{equation}

1969: 

1970: 

1971: Every character defines a holomorphic vector bundle

1972:  by the standard construction

1973: \begin{equation}

1974: L = U \times \C / (\pi_1,\rho)

1975: \end{equation}

1976:  where $U$ is the universal cover of our base with the natural

1977:   action of the fundamental group of a base on $U$ and the $\rho$-action

1978:   on $\C$ .

1979: 

1980: Thus we have  the forgetful map

1981: \begin{equation}

1982: f \colon A \to J(\Si).

1983: \end{equation}

1984: Every fiber of this map is the set of holomorphic flat connections

1985: on a fixed line bundle. Thus the map $f$ sending each character

1986:  to the corresponding line bundle provides on $A$ the structure of an

1987:   {\it affine bundle over the cotangent bundle}.

1988: 

1989: So fibers of $f$ are affine spaces over $H^0(J(\Si),\Om)$ (but in

1990: this case the cotangent bundle is the trivial $g$-dimensional

1991: vector bundle). Over every small open set $U_i$ our affine bundle

1992: admits a section that is can be identified with the restriction

1993: of the cotangent bundle. Any descrepance  exists only  on the

1994: intersections $U_i \cap U_j$. The collection of differences of

1995: sections gives a 1-cocycle

1996: \begin{equation}

1997: \ep \in H^1(J(\Si), \Om).

1998: \end{equation}

1999: If this cocycle is trivial, our affine bundle coincides with its

2000: vector bundle while non trivial affine bundle is defined by this

2001: cocycle uniquely. By the Dolbeault theorem in our case the cocycle

2002: $\ep_A \in H^{1,1} = H^2(J(\Si), \C)$. Poincare proved that

2003: \begin{equation}

2004: \ep_A = [\Theta] = [\om].

2005: \end{equation}

2006: To prove this equality we have to use the functional equation

2007: for theta functions.

2008: 

2009: 

2010: The affine bundle (2.62) admits {\it non holomorphic sections}:

2011: let $U(1) \subset \C^*$ be the subgroup of unit norm numbers.

2012: There is a map

2013: \begin{equation}

2014: mod \colon \C^* \to U(1), \quad z \to \frac{z}{|z|}.

2015: \end{equation}

2016: Then we have the subspace

2017: \begin{equation}

2018: U(1) (\C^*)^{2g}

2019: \end{equation}

2020: of unitary characters of $\pi_1(\Si)$ which is a {\it section} of

2021:  the projection $f$ (2.62) of the  affine vector

2022:  bundle $(\C^*)^{2g}$. It  follows immediately from the maximum

2023: principle for a compact complex manifold.

2024: 

2025: 

2026: 

2027: So we have the smooth  identification

2028: \begin{equation}

2029:  J(\Si) = U(1)^{2g}.

2030: \end{equation}

2031: 

2032: 

2033:  Let $H_1(\Si, \Z) = <a_1, ... ,a_g, b_1,... ,b_g> $ be as before

2034:   (see (2.32)).Such

2035:   presentation defines

2036:  very important subspace of the character space, so

2037:  called  {\it abelian Schottky subspace}:

2038: \begin{equation}

2039: S^a_g = (\C^*)^g = \{ \chi \in (\C^*)^{2g} \vert \chi(a_i) = 1 \}.

2040: \end{equation}

2041: and

2042: \begin{equation}

2043: uS^a_g = (U(1))^g = \{ \chi \in U(1)^{2g} \vert \chi(a_i) = 1 \}

2044: \end{equation}

2045:  is

2046:  {\it abelian unitary Schottky space}.

2047: 

2048:  The restrictions of the forgetful map $f$ (2.62) onto this space has the

2049:  following properties

2050: \begin{enumerate}

2051:  \item $f \colon S^a_g = (\C^*)^g \to J(\Si)$ is infinite ($\Z^g$) cover,

2052: \item $ f^* (\Oh (\Theta))$ is trivial, so holomorphic sections =

2053: holomorphic functions = theta functions;

2054: \item this line bundle is given by the {\it automorphy factors}

2055: \begin{equation}

2056:  e_{a_i}(z) = 1

2057: \end{equation}

2058: $$

2059: e_{b_i}(z) = e^{-2\pi z_i - \pi i \Om_{ii}}

2060: $$

2061: where $\vec z$ are complex coordinates of the universal cover $U$ of $A$

2062: described after formula (2.34), $\Om$ is

2063:  the period matrix of $\Si$ and so on (see the set up around (2.32) - (2.35));

2064: let $\{ u_i\}$ be the basis in $U$ dual to $\{ a_i\}$ and

2065:  \begin{equation}

2066:  u_{g + i} = \sum_1^g \Om_{ij} u_i;

2067: \end{equation}

2068: \item thus using these automorphy factors we see that the space $H^0(J(\Si),

2069: \Oh(k \Theta))$ is naturaly identifyed with the space of holomorphic functions

2070: $\theta$

2071:  on $(\C^*)^g$ such that $\theta(z + u_i) = \theta$ and

2072: \begin{equation}

2073:  \theta(z + u_{g+i}) = e^{-2\pi k  z_i - \pi i k

2074:  \Om_{ii} };

2075: \end{equation}

2076: \item thus any $\theta(z)$ admits the following decomposition

2077: \begin{equation}

2078:  \theta (z) = \sum_{l \in \Z^g / k \Z^g} \theta_l(z, \Om)

2079: \end{equation}

2080: where theta functions with characteristics are functions (2.49).

2081: \end{enumerate}

2082: 

2083: \subsection{Perfect quantization}

2084: 

2085: Summarizing all results and constructions, we have

2086: 

2087: \begin{thm} Principal polarized abelian varieties

2088: (and hence algebraic curves) admit the perfect

2089: quantization by theta functions.

2090: \end{thm}

2091: 

2092: In particular, we proved  very important fact (which we want to

2093: generalize for the non-abelian case):

2094: 

2095: \begin{prop}

2096: The  Bohr-Sommefeld wave space (2.24) doesn't depend on the

2097: decomposition (2.21) determining  projection $\pi$ (2.22). This

2098: decomposition defines a basis in this space only and the result

2099: of Bohr-Sommerfeld quantization doesn't depend on the choice of a

2100: real polarization of an "arithmetical" type (2.22).

2101: \end{prop}

2102: Of course this observation is obvious in the set-up of the classical theory

2103: of theta functions, but in the set-up of non-abelian theta functions it is giving

2104: an important "independence condition" in CQFT and low dimensional topology.

2105: 

2106: 

2107: 

2108: 

2109: Let us list the sequence of tasks necessary to get a perfect quantization:

2110: \begin{enumerate}

2111: \item a $g$-torus $U(1)^g$ with the marked point $0 \in U(1)^g$ we identify with

2112: the zero fiber of the fibration $\pi : A \to T^g_-$ (2.22) containing zero point;

2113: \item for zero point $0 \in U(1)^g$ we construct $\de$-function $\theta_0^\R$

2114:  as the Fourier

2115: serie (2.46);

2116: \item using the period matrix $\Om$ of $A$ we construct  CST

2117: $C^{(-i \Om)}_{1/k}$ (2.40) , (2.41);

2118: \item applying this transform to the distribution $\theta_0^\R$

2119: we get the collection of holomorphic theta functions with characteristics $\{

2120: \theta_l(z, \Om)\}$  (2.49) on the complexification  $(\C^*)^g$ of $U(1)^g$;

2121: \item using the periods matrix $\Om$ we construct the cover $(\C^*)^g \to A$

2122:  and get sections of the line bundle $\Oh(k \Theta)$ as functions subjecting to

2123:  automorphity factors conditions (2.70);

2124:  \item to check that  functions $\{

2125: \theta_l(z, \Om)\}$  (2.49) correspond to sections of  $\Oh(k \Theta)$.

2126: \end{enumerate}

2127: 

2128: This is the end of story for the moduli spaces of topological trivial

2129:  vector bundles

2130:  of rank one on  algebraic curves=Riemann surfaces. But any curve $\Si$ also

2131:   has the other

2132:  "moduli space", namely the moduli space of topological trivial

2133:  vector bundles of rank 2. This is a faithful functor too. To implement

2134:  this theory we have to plug  Conformal Field Theory in dimension

2135:  $D = 2$. We will use the previous list of tasks as a pattern of

2136:  much more sophisticated  efforts in this "non-abelian" case.

2137: 

2138: 

2139:  Note that  $H^0(J(\Si), \Oh ( k \Theta))$ is the space of the irreducible

2140: representation of the Heisenberg

2141:  group of level $k$.

2142: 

2143: 

2144: This statement is in the very heart of the classical theory of theta

2145: functions. Actually the group $K$ (2.12) is the quotient of the

2146: Heisenberg group $\Ga_k$ with the kernal  $\Z_k$ (with respect to the

2147: multiplicative action). More precisely, the Heisenberg group acts

2148: on the line bundle $\Oh(k \Theta)$ and on the space of its

2149: holomorphic sections. This is an

2150: irreducible representation of $\Ga_k$ and such representation is

2151: unique up to projectiviazation. The existence of the special basis is provided

2152:   by this

2153: action. ( In particular, $rk H^0(J(\Si), \Oh ( k \Theta)) = k^g$).

2154: This property (to be the space of   unique irreducible

2155: representation of some group or algebra) is true for non-abelian

2156: case too. But in this case the space of theta functions is the

2157: space

2158:  of irreducible representation of the gauge algebra

2159: of WZW CQFT.

2160: 

2161: 

2162: 

2163: \section{Non-abelian theta functions}

2164: 

2165: 

2166: The direct generalization of  jacobians of  algebraic curves

2167:  as a faithful functor is the sending of an algebraic curve $\Si_I$

2168:  to the {\it moduli space} $M^{ss}$ of semi-stable

2169:   topologicaly trivial rk 2 vector bundles on this curve.

2170:   Of course, we can consider much more complicated situation,

2171:   but this case is the first {\it non-commutative} case which

2172:    is expressive enough to see all new features of the

2173:    geometrical situation.

2174: 

2175: Following  our pattern  we can quantize a Riemann surface $\Si$

2176: using the faithful functor

2177: \begin{equation}

2178:  \Si \to M^{ss}(\Si)

2179: \end{equation}

2180: sending a Riemann surface to the moduli space of  semi-stable

2181: vector bundles on it.

2182: 

2183: For this we have to repeat all steps of the abelian (classical) theory of

2184: theta functions.

2185: 

2186: 

2187: \subsection{Algebraic geometry of moduli spaces of vector bundles}

2188: 

2189: 

2190:  The algebro-geometric part of the

2191: non-abelian theory of theta functions is well developed by

2192: Narashimhan, Beauville,

2193:  Laszlo, Pauly, Oxbury, Ramanan, Sorger and many others.

2194: In non-abelian context important new notion of  semi-stability

2195: appears. Recall that for our case  semi-stability of $E$ means

2196: that $E$ doesn't contain  linear subbundles of positive degree

2197: (that is, with positive $c_1$). To work with rk 2 vector bundles

2198: we need some information about the structure of coherent sheaves

2199: on algebraic curves and more from the homological algebra. We

2200: have to know that every coherent sheaf on a smooth algebraic

2201: curve is a direct sum of a local free sheaf (= vector bundle) and

2202: a torsion sheaf with support in a finite set of points. From the

2203: homological algebra we need  Riemann-Roch theorem and first

2204: cohomology of sheaves. All  these facts can be found in any

2205: survey on this subject, for example in \cite{DSS}

2206: 

2207: Let $M^{ss} (\Si)$ be the moduli space of topologically trivial

2208:  semi-stable holomorphic bundles of rank 2 on $\Si$ where every

2209: non-stable (but semi stable) bundle is presented by the direct sum

2210: of two opposite

2211:  topological trivial line bundles.

2212: Then  $ dim M^{ss}(\Si) = 3g-3$.

2213: 

2214:  The tangent space to the moduli space at a point $E$

2215: has well known  shape

2216:  \begin{equation}

2217: TM^{ss}_E = H^1(\ad E)

2218: \end{equation}

2219: where $\ad E$ is the traceless part of the vector bundle of

2220: endomorphisms of $E$. For any  stable bundle $E$ one has

2221: $H^0(\Si_I, \ad E) = 0$ and thus by the Riemann-Roch theorem

2222: \begin{equation}

2223: dim M^{ss} = rk TM^{ss}_E = rk H^1(\ad E) = 3g-3.

2224: \end{equation}

2225: 

2226: Definitions of theta divisors are absolutely parallel to (2.6):

2227: let $\si \in Pic_{g-1}(\Si)$ be a line bundle on $\Si$ of degree

2228: $g-1$; then we have

2229: 

2230: \begin{prop}

2231: \begin{enumerate}

2232: \item $\Theta = \{ E \in M^{ss}(\Si) \vert H^0(\Si, E(\si))

2233: \neq 0\}$;

2234: \item as a variety $\Theta$ is birationally equivalent

2235:  to the projective space  $\PP^{3g-4}$;

2236: \end{enumerate}

2237: \end{prop}

2238: 

2239: Indeed, a general vector bundle from the divisor $\Theta$ twisted by $\si \in

2240: Pic_{g-1}(\Si)$

2241:  admits a section. (Recall that in VBAC-slang "twisted" means

2242:   tensor product with the line bundle corresponding to $\si$.)

2243:  We may expect that this section has no zeros, thus our general

2244:   vector bundle $E$ can be represented as an extension

2245: \begin{equation}

2246: 0 \to \Oh(-\si) \to E \to \Oh(\si) \to 0.

2247: \end{equation}

2248: Of course there exists  trivial extension $E = \Oh(-\si) \oplus

2249: \Oh(\si)$ but this bundle isn't stable at all.

2250: 

2251: Any other extension is given by non zero cocycle of the space

2252: \begin{equation}

2253: Ext^1(\Oh(\si), \Oh(-\si)) = H^1(\Si_I, \Oh(-2\si)) = \C^{3g-3}

2254: \end{equation}

2255: by the Riemann-Roch theorem. Automorphisms of a subbundle act

2256:  on these cocycles by  constant multiplication.

2257:   Thus if $E(\si)$ has no other section (what happens in   general case),

2258:   our vector bundle $E$ is defined uniquely by a point of

2259:   the projective space $\PP H^1(\Si_I, \Oh(-2\si)) = \PP^{3g-4}$.

2260: It is a simple exercise to estimate the dimension of spaces of

2261: bundles admitting a section with  non trivial zero set.

2262: 

2263: The moduli space $M^{ss}$ is close to be rational itself. Let us

2264: consider the space $Pic_g$ of classes of divisors of

2265:  degree $g$ (instead of $g-1$ as in Proposition 5) and twist

2266:   all bundles from $M^{ss}$ by some divisor $\si \in Pic_g(\Si)$.

2267:   Then for every vector bundle $E \in M^{ss}$

2268: we have

2269: \begin{equation}

2270: rk H^0(\Si_I, E (\si)) \geq 2.

2271: \end{equation}

2272: Now we have the canonical (evaluating) homomorphism

2273:  \begin{equation}

2274:  can \colon H^0(\Si_I, E(\si)) \otimes \Oh \to E(\si)

2275: \end{equation}

2276: Again by dimensional computations it is easy to see that this

2277: moduli space contains a Zariski open set $M_0$ of bundles

2278:  for which this homomorphism

2279: can be extended to the exact sequence

2280: \begin{equation}

2281: 0 \to  \C^2 \otimes \Oh \to E(\si) \to \oplus_{i=1}^{2g}

2282: \Oh_{p_i} \to 0

2283: \end{equation}

2284: where $2g$ points $\{ p_i \}$ form the sum which is an effective

2285: divisor equivalent to $2\si$:

2286: \begin{equation}

2287: p_1 + p_2 + .... + p_{2g} \in \vert 2\si \vert = \PP H^0(\Si_I,

2288: \Oh(2\si)) = \PP^g.

2289: \end{equation}

2290: Why does this sum of points give an effective divisor from $\vert

2291: 2\si\vert$? Well, the homomorphism

2292: \begin{equation}

2293:  \wedge^2 can = det \colon \Oh \to \Oh(2 \si) = c_1(E(\si))

2294: \end{equation}

2295: and the effective divisor (3.9) is just  zero set of this

2296: homomorphism.

2297: 

2298: 

2299: 

2300: Sending a vector bundle $E \in M_0$ to such effective divisor

2301:  we

2302: obtain  the map (of course algebraic)

2303: \begin{equation}

2304: det \colon M_0 \to \PP^g.

2305: \end{equation}

2306: The fiber can be obtained by the canonical homomorphism data:

2307: \begin{equation}

2308:  ker can_{p_i} \colon \C^2 \to E_{p_i}.

2309: \end{equation}

2310: So for every point $p_i$ we have a point on the projective line

2311: \begin{equation}

2312: \ga_i = \PP ker  can_{p_i} \in \PP \C^2 = \PP^1.

2313: \end{equation}

2314: Thus the fiber of the map (3.11) is given by a collection of

2315: points

2316: \begin{equation}

2317: (\ga_1, .... , \ga_{2g}) \subset \PP^1

2318: \end{equation}

2319: up to a projective linear transformation.

2320: 

2321: We can see that any fiber of the map $det$ is rational too

2322:  and $M^{ss}$ is an algebraic fibration with rational

2323:  $g$-dimensional base and rational $2g-3$-dimensional fiber.

2324: 

2325: The same trick we can use for vector bundles of higher ranks. But

2326: we would like to recall the old problem of the birational

2327: geometry:

2328: 

2329: {\it Is the moduli space $M^{ss}$ rational?}

2330: 

2331: In spite of many attempts to solve  this rationality problem,

2332: the answer isn't known up to now.

2333: 

2334: \begin{dfn}

2335: The space $H^0(M^{ss}(\Si), \Oh ( k \Theta))$ is called the space

2336: of non-abelian theta functions of level $k$.

2337: \end{dfn}

2338: Continuing the parallel description of the geometry of moduli

2339: spaces we get

2340: \begin{prop}

2341: \begin{enumerate}

2342: \item  Theta divisors are ample. Moreover

2343: \item  $H^2(M^{ss}, \Z) = \Z$.

2344: \item (Faltings 92) $H^0(M^{ss}(\Si), \Oh ( k \Theta))$ is the space

2345:   of irreducible representation of the gauge algebra

2346: of WZW CQFT.

2347: \end{enumerate}

2348: \end{prop}

2349: Such spaces have played a noted role in Conformal Field Theory.

2350: The Proposition  will be proved below.

2351: 

2352: 

2353: \subsection{Holomorphic flat connections}

2354: 

2355: Up to now we have considered any vector bundle as a local free

2356: sheaf.

2357:  As far as for line bundles   any vector bundle is defined by

2358:  some finite cover $\{U_i\}$ of the base and by

2359:   a collection of {\it transition matrices} $\al_{ij}$ regular

2360:    and regular inversed on $U_i \cap U_j$.

2361: On the intersection $U_i \cap U_j \cap U_k$ we have again the

2362: cocycle equality

2363: \begin{equation}

2364: \al_{ij} \cdot \al_{jk} \cdot  \al_{ki} = 1.

2365: \end{equation}

2366: A collection $\al_{ij}$ is equivalent to a collection $\al'_{ij}$

2367: iff there exists a collection $\beta_i$ matrices, each $\{\beta_i\}$ is

2368: regular and regular inversed on $U_i $ such that

2369: \begin{equation}

2370: \al'_{ij} = \beta_i \al_{ij} \beta_j^{-1}.

2371: \end{equation}

2372: 

2373: 

2374: For the matrix case "as simple as possible" representatives can be

2375: \begin{enumerate}

2376: \item triangle matrices (the theory of extensions, see (3.4));

2377: \item constant matrices (the theory of flat holomorphic

2378: connections).

2379: \end{enumerate}

2380: Again let us

2381:  consider the collection of matrices of differential

2382: forms

2383: \begin{equation}

2384: \{  \al_{ij}^{-1} d \al_{ij} \} \in Z^{1}(\{U_{i} \} )

2385: \end{equation}

2386: Obviously this cochain is a cocycle from $H^1(\Om (End E))$ which

2387: cohomology class

2388: \begin{equation}

2389: c(E) \in H^{1}(\Om(End E))

2390: \end{equation}

2391: is the full Chern class of $E$.

2392: 

2393: The automorphisms group $Aut E$ acts on the space $H^{1}(\Om(End E

2394: ))$ preserving this class.

2395: 

2396: For example, on a curve by the Serre duality

2397: \begin{equation}

2398:  H^{1}(\Om(End E))^{*} = H^{0}(End E)

2399: \end{equation}

2400: and it is easy to see that

2401: 

2402: \begin{prop}

2403: \begin{equation}

2404: <c(E), \si> \neq 0 \implies \quad \si \quad \text{is an idempotent

2405: in } \ H^0(End E)

2406: \end{equation}

2407: \end{prop}

2408: This simple observation belonging to Atiyah gives plenty  results

2409: about the structure of multidimensional bundles: the vector bundle

2410: $End E$ splits as

2411: \begin{equation}

2412: End E = \Oh \oplus ad E

2413: \end{equation}

2414: where the first component $\Oh$ is the trivial line bundle

2415: corresponding to homotheties  and $ad E$ corresponds to traceless

2416: endomorphisms. Thus we have the decomposition

2417: \begin{equation}

2418: \Om(End E) = \Om \oplus ad E \otimes \Om

2419: \end{equation}

2420: where the first component sends  endomorphisms to their  traces.

2421: Thus if $H^0(End E)$ contains one indempotent then it contains a

2422: second too and such vector bundle is the direct sum of line

2423: bundles.

2424: 

2425:  Vector bundle $E$ is called {\it indecomposable} if $E \neq E_{1}

2426: \oplus E_{2}$.

2427: 

2428: For such bundles

2429: \begin{equation}

2430: c(E) \in H^{1}(\Om)

2431: \end{equation}

2432: since $H^0(End E)$ doesn't contain any idempotents.

2433: 

2434: If $\{ \al_{ij}^{-1} \cdot d \al_{ij} \}$ is the cochain (3.15)

2435: defining $c(E)$ then

2436: \begin{equation}

2437:  tr (\al_{ij}^{-1} d \al_{ij}) \in H^{1} (\Om).

2438:  \end{equation}

2439:  By the Dolbeault isomorphism

2440: \begin{equation}

2441: H^1(\Om) = H^{1,1} \in H^2(\C).

2442: \end{equation}

2443: The formal matrix equality

2444: \begin{equation}

2445: tr (\al_{ij}^{-1} \cdot d \al_{ij}) = (det \Vert\al_{ij}

2446: \Vert)^{-1} d (det \Vert \al_{ij} \Vert)

2447: \end{equation}

2448: shows that

2449: \begin{equation}

2450: c(E) = c(det E)

2451: \end{equation}

2452: for any indecomposable vector bundle on  curve.

2453: 

2454: 

2455: It is easy to see (just by the direct interpretation) that this

2456: cohomology class is integer and moreover

2457: \begin{equation}

2458: \{\al_{ij}^{-1} d \al_{ij}\} = c_1(E)

2459: \end{equation}

2460: where $E$ is the indecomposable vector bundles given by

2461: $\{\al_{ij}\}$. If this class is zero as in our case of topologically

2462: trivial bundles then the cocycle is trivial and there are

2463: matrices of differential forms $h_i$ such that

2464: \begin{equation}

2465: \al_{ij}^{-1} d \al_{ij} = h_i - h_j.

2466: \end{equation}

2467:  \begin{dfn} Collection of matrix differential forms $h_i$

2468:   on $U_i$ is called  flat holomorphic connection on the

2469:    vector bundle $E$ given by the collection of transition

2470:     functions

2471: $\al_{ij}$.

2472: \end{dfn}

2473: 

2474: Having such collection we can solve  the system of linear

2475: equations on $U_i$

2476: \begin{equation}

2477: \beta_i^{-1} \cdot d \beta_i = h_i

2478: \end{equation}

2479: and  get a new collection of transition matrices. Now using all

2480: equations we can see that matrices $\al'_{ij} = \beta_i \al_{ij}

2481: \beta_j^{-1}$  are constant, that is

2482: \begin{equation}

2483:  d \al'_ij = 0.

2484: \end{equation}

2485: If matrices are constant,  this vector bundle is  called

2486:  {\it local system of coefficients} and we can trivialize it over

2487:  any simply connected open set and obtain the  representation of

2488:  the fundamental group of the base:

2489: \begin{equation}

2490: \rho \colon \pi_1 \to SL(2, \C).

2491: \end{equation}

2492: Actually, because of the equivalence relations  we obtain a class of

2493:  representations only. That is the representation $\rho$ up to

2494:  the adjoint action of $SL(2, \C)$.

2495: 

2496: 

2497: For  curve we have the classical result:

2498: 

2499: \begin{prop} (A. Weil) Every $E \in M^{ss}(\Si)$ admits a

2500: holomorphic flat connection

2501:  given by a class of representations $\rho \colon \pi_1(\Si) \to SL(2, \C)$.

2502: \end{prop}

2503: The author emphasizes   again that he've proved this statement

2504: already.

2505: 

2506: \begin{rmk}

2507: In the influential paper \cite{We} A. Weil proposed

2508: \begin{enumerate}

2509: \item the proof of the previous Proposition;

2510: \item the idea that high dimensional vector bundles should play

2511: the role of non-abelian analogue of Jacobians;

2512: \item the notion of  {\it class of matrix divisor} which is

2513: equivalent to the notion of  vector bundle.

2514: \end{enumerate}

2515: \end{rmk}

2516: 

2517: 

2518: 

2519: 

2520: Many classes of representations give the same vector

2521: bundle:

2522:  the difference of two holomorphic connections can be

2523:  identified over

2524: $U_i \cap U_j$, thus

2525: \begin{equation}

2526: \{h_i\} - \{ h'_i\} \in H^0(End E \otimes \Om).

2527: \end{equation}

2528:  An element of the last space is a homomorphism

2529: \begin{equation}

2530: \phi \colon E \to E \otimes \Om

2531: \end{equation}

2532: and for   case of curves (when $K_\Si = \Om_\Si$) it is called

2533: {Higgs field}.

2534: 

2535: Note that by Serre duality

2536: \begin{equation}

2537: H^1(\ad E)^* = H^0( \ad E \otimes K)

2538: \end{equation}

2539: that is the vector bundle over the moduli space $M^{ss}$ with

2540: fibers equal to spaces of Higgs fields is nothing else but the

2541: cotangent bundle $\Om$ of $M^{ss}$.

2542: 

2543: 

2544: For rank 2 vector bundles with fixed determinant we have to

2545: consider traceless Higgs fields only:

2546: \begin{equation}

2547: \phi \in H^0(\Si, \ad E \otimes K).

2548: \end{equation}

2549: 

2550: 

2551: The full space $A^{na}$ of all

2552:  holomorphic flat connections

2553: \begin{equation}

2554: A^{na} = CLRep(\pi_1(\Si), SL(2, \C))

2555: \end{equation}

2556: is the space of all classes of representations and

2557:  every representation defines a holomorphic vector bundle

2558: just by the standard construction

2559: \begin{equation}

2560: E  = U \times \C^2 / (\pi_1,\rho)

2561: \end{equation}

2562:  where $U$ is the universal cover of $\Si$ with the natural

2563:   action of the fundamental group of athe base.

2564: 

2565: Of course in the set of such bundles there are non-stable but

2566: indecomposable

2567:  vector bundles and even semi-stable but indecomposable. Let us remove

2568:  such representations and get the space $A_{na}^{ss}$ of classes

2569:  of representations which gives stable and decomposable  semi-stable bundles.

2570: 

2571: Consider  the forgetful  map

2572: \begin{equation}

2573:  f \colon A^{na}_{ss} = CLRep(\pi_1(\Si), \SU(2)), SL(2, \C)) \to M^{ss}.

2574: \end{equation}

2575: This is the affine bundle over the cotangent bundle (see (2.62

2576: )). Again such bundle is given by a cocycle

2577: \begin{equation}

2578: \ep_{na} \in H^1(M^{ss}, \Om) = H^{1,1}(M^{ss}, \C)

2579: \end{equation}

2580: 

2581: \begin{prop}

2582: \begin{equation}

2583: \ep_{na} = [\Theta_{na}] = [\om_{na}]

2584: \end{equation}

2585: \end{prop}

2586: That is, the answer is precisely the same as in the abelian case.

2587: We will prove this fact a little bit latter using special

2588: subspaces.

2589: 

2590: Both of affine bundles (abelian and non-abelian) (2.62) and (3.39)

2591: admit {\it non holomorphic sections}. Again constructions of these

2592: sections are quite parallel (see (2.67) for the abelian case) :

2593: the including $\SU(2) \subset SL(2, \C)$ of complexification

2594: defines

2595:  the subspace

2596: \begin{equation}

2597: CLRep (\pi_1(\Si), \SU(2)) \subset CLRep (\pi_1(\Si), SL(2, \C)

2598: \end{equation}

2599: which is the section of the projection $f$ (3.39). (Again it

2600: follows from the maximum principle for the compact curve.) But the

2601: new feature of non abelian

2602:  case is the following:

2603: 

2604:  we don't know if the restriction of forgetfull map to

2605:  $CLRep (\pi_1(\Si), \SU(2))$ is

2606:  onto. Does every semi-stable bundle admits Hermitian flat connection?

2607: 

2608:  We will prove this statement  later, and note that the restriction map defines

2609: the Narasimhan-Sesadri isomorphism

2610: \begin{equation}

2611:  NS  \colon  CLRep(\pi_1(\Si), \SU(2)) \to M^{ss}.

2612: \end{equation}

2613: 

2614: 

2615: So we have differential identification

2616: $$

2617:   CLRep(\pi_1(\Si), \SU(2)) = M^{ss}.

2618: $$

2619: 

2620: 

2621:  Let $\pi_1 = <a_1, ... ,a_g, b_1,... ,b_g \vert \prod_{i=1}^g [a_i, b_i] = 1>$

2622:  be a standard presentation

2623: of the fundamental group. Such presentation defines a couple of

2624:  very important subspaces of the spaces of representations, so

2625:  called Schottky subspaces:

2626: \begin{enumerate}

2627: \item $S_g = \{ \rho \in A^{na} \vert \rho (a_i) = 1 \}$ is

2628: non-abelian complex Schottky space.

2629: \item $S_g = SL(2, \C)^g / Ad_{diag}SL(2, \C)$

2630: \item $dim S_g = dim M^{ss}$

2631: \item $uS_g = S_g \cap CLRep(\pi_1(\Si), \SU(2))$ is  unitary Schottki space;

2632: \item $dim \quad uS_g = 1/2 dim M^{ss}$.

2633: \end{enumerate}

2634: 

2635: We saw in the abelian case that the restriction of the forgetful

2636: map to the Schottki space is a cover giving theta sections as

2637: functions.

2638: 

2639: 

2640: For the non-abelian case the situation is much more complicated:

2641: \begin{enumerate}

2642:  \item the forgetful map

2643:  \begin{equation}

2644:  f \colon S_g \to M^{ss}(\Si)

2645:  \end{equation}

2646:   is meromorphic only

2647:  (because of existence of non stable flat bundles).

2648:   For example, the Schottky uniformization of  curve  gives the

2649:   non stable bundles.

2650: \item Properties of this map are unknown yet.

2651: C. Florentino proved only that the differential of this map at the

2652: unitary Schottky space is non degenerated and in section 9 we show that for

2653:  general curve $\Si$ the image of $f$ (3.44) is Zariski dense in $M^{ss}(\Si)$.

2654: \end{enumerate}

2655:  To avoid this gap, we have to use Symplectic and Lagrangian

2656:  geometry and sophisticated analysis.

2657: 

2658: \subsection{Moduli of stable pairs

2659: and the desingularization of moduli spaces}

2660: 

2661: The construction of the moduli spaces was achived in the 1960's,

2662: mainly by D.Mumford and the mathematicians of  Tata institute.

2663: But after appearing of the gauge theory approach to vector bundles

2664: these moduli spaces were input in the collection of moduli spaces

2665: of {\it stable pairs} related by a chain of {\it flips} (see

2666: \cite{Re}). This chain of flips was used by Thaddeus, Bertram and

2667: Bredlow-Daskalopoulos for the computations of the cohomology ring

2668: of moduli spaces and by Seshadri \cite{Se} for the construction

2669: of a desingularisation of moduli spaces.

2670: 

2671: New object to study will be a pair $(E, s)$ consisting of a

2672:  vector bundle $E$ on a curve $\Si$ and a nonzero section

2673: $s \in H^{0}(E)$. S. Bredlow defined  new stability condition for

2674: such pairs and proved  theorem relating stable pairs to vortices

2675: on the Riemann surface. The vortex equation depends on  positive

2676: real parameter $t$ and so the stability condition also depends on

2677: $t$.

2678: \begin{rmk}

2679: The vortex equation requests a Kahler metric, not a conformal

2680: class only. We consider the algebro-geometrical interpretation

2681:  here. We fix the normalization of such metric such that the

2682:  volume $vol \Si = 4 \pi$.

2683: \end{rmk}

2684: 

2685: 

2686: 

2687: Our vector bundles admit nonzero sections so we have to consider

2688: some positive line bundle $D$ of degree $d$ such that $det E =

2689: D$. We obtain this case by twisting vector bundles from $M^{ss}$

2690: by an appropriate line bundle. The stability condition for  pair

2691: $(E, s)$ with rk 2

2692:  vector bundle is the

2693: following: a pair is semi-stable if for any line bundle $ L

2694: \subset E$

2695: \begin{equation}

2696: deg L \leq 1/2 d - t \quad \text{ if } s \in H^{0}(L)

2697: \end{equation}

2698: $$

2699: deg L \leq 1/2 d +t \quad \text{ otherwise. }

2700: $$

2701: The fondation of this theory can be found in \cite{BD},

2702: \cite{Ber}, \cite{Th2}.

2703: 

2704: Let $M(D, t)$ be the moduli space of stable pairs. It's easy to

2705: see that

2706: \begin{enumerate}

2707: \item $M(D, t)$ is nonempty if and only if $t  \leq d/2$;

2708: \item if $t$ is irrational then $M(D, t)$ is compact;

2709: \item if for $(E_{i}, s_{i}) \in M(D, t), \quad i=1,2$ and if there

2710: exists a homomorphism $\phi \colon E_{1} \to E_{2}$ such that

2711: $\phi(s_{1}) = s_{2}$ then $\phi$ is an isomorphism;

2712: \item for $(E, s) \in M(D, t)$ there are no endomorphisms of $E$

2713:  annihilating the section $s$ exept 0 and no endomorphisms preserving $s$

2714: exept  identity;

2715: \item if $(E, s) \in M(D, t)$ then $(E(\xi), s(\xi)) \in M(D(\xi), t)$

2716: for any effective divisor $\xi$.

2717: \end{enumerate}

2718: 

2719: The deformation theory of stable pairs is very simple: let $t$ be

2720: irrational, then  for $(E, s) \in M(D, t)$

2721: \begin{enumerate}

2722: \item there is a natural exact sequence

2723: \begin{equation}

2724: o \to H^{0}(ad E) \to H^{0}(E) \to TM(D,t)_{(E,s)} \to H^{1}(ad

2725: E) \to H^{1}(E);

2726: \end{equation}

2727: where homomorphisms are defined by the section $s$ .

2728: \item $rk

2729: TM(D,t)_{(E,s)} = d - g -2.$

2730: \end{enumerate}

2731: So for irrational $t$ the space $M(D, t)$ is smooth and we

2732: concentrate our attention on these smooth moduli spaces. More

2733: precisely we have the case if

2734: \begin{equation}

2735: t_{i} \in (max(0, d/2 - i - 1, d/2 - i)).

2736: \end{equation}

2737: For all $t_{i}$ from these intervals we have the same compact

2738: smooth moduli spaces $M(D, t_{i})$ related by the chain of flips

2739: (see \cite{Re}). We will use them later.

2740: 

2741: 

2742: In general case the stability condition for  pair $(E, s)$ is the

2743: following: a pair is semi-stable if for any proper subbundle $

2744: E_{1} \subset E$

2745: \begin{equation}

2746: \mu (E_{1}) \leq \mu (E) - t \quad \text{ if } s \in H^{0}(E_{1})

2747: \end{equation}

2748: $$

2749: \mu E_{1}  \leq \mu(E) +t \quad \text{ otherwise. }

2750: $$

2751: where $\mu$ as usual is  slope of the bundle. The fondations of

2752: this theory can be find in \cite{BD}.

2753: 

2754: 

2755: For  moduli spaces of stable pairs we have the same collections

2756: of statements as before.

2757: 

2758: To apply this method to desingularization of $M^{ss}$ we have to

2759: send a vector bundle $E \in M^{ss}$  to the pair $(End E, s)$

2760: where $s \in H^{0}(End E)$. For a stable $E$ the choice of

2761: section is unique: it has to be the line subbundle  of

2762: homotheties. For

2763:  semi-stable bundles there are two cases: $H^{0}(End E) = 2, 4$.

2764:  Considering the parameter $t$ which is very close to $0$ we get

2765:  the moduli space of stable pairs $\widetilde{M^{ss}}$ with the

2766:  birational projection to $M^{ss}$ which is a smooth variety.

2767:  Always below {\it we will consider this canonical desingularization

2768: will not mention  that specially}.

2769: 

2770: \begin{rmk}

2771: This desingularization was constructed by Seshadri \cite{Se}. He

2772: considered a {\it parabolic structure} on  vector bundle over

2773: fixed point of a curve and choose  special parameters of such

2774: structure to get the same non singular variety with a birational

2775: map to the $M^{ss}$.

2776: \end{rmk}

2777: 

2778: 

2779: 

2780: 

2781: 

2782: 

2783: 

2784: \subsection{Holomorpic symplectic geometry of Higgs fields.}

2785: 

2786: 

2787: Returning to Higgs fields (3.36),

2788:  we can send every pair $(E, \phi)$ where $\phi$ is a Higgs

2789:  field

2790: to the space of holomorphic quadratic differentials:

2791: \begin{equation}

2792: \wedge^2 \phi \colon \Oh \to \wedge^2 (E \otimes K_{\Si}).

2793: \end{equation}

2794: if $E$ is stable our pair is stable too and  $T^* M^{ss}$ is a

2795: moduli space of such pairs. Then

2796:  we get

2797: the holomorphic map

2798: \begin{equation}

2799: \pi \colon T^* M^{ss} \to H^0(\Si, K_{\Si_I}^2).

2800: \end{equation}

2801: 

2802: 

2803: As usual the cotangent bundle admits the holomorphic "action

2804: 1-form" $\al$ such that the   holomorphic differential of it $d

2805: \al = \Om$ is the holomorphic symplectic form defining the

2806: holomorphic skew symmetrical isomorphism

2807: \begin{equation}

2808: \Om \colon T T^* \to T^* T^*.

2809: \end{equation}

2810: One can see that we just reproduce notions and constructions of

2811: the classical symplectic geometry in the holomorphic set up. In

2812: particular, we can expect  existence of "completely integrable

2813: systems" that is a holomorphic map of  holomorphic symplectic

2814: variety $\sM$

2815: \begin{equation}

2816: \pi \colon \sM \to B

2817: \end{equation}

2818: such that every fiber is Lagrangian with respect to $\Om$ and the

2819: generic fiber is a middle dimensional complex torus or an abelian

2820: variety.We call such map as  holomorphic moment map.

2821: 

2822: The brilliant observation of Hitchin is that the map (3.50) gives

2823: such  integrable system. Moreover, the variety $T^* M^{ss}$ can be

2824: slightly extended as a holomphic symplectic variety such a way

2825: that fiber of $\pi$ become compact.

2826: 

2827: \begin{dfn}  Higgs pair $(E, \phi)$ is called stable if it doesn't

2828:  exist a line

2829: subbundle $L \in E$ of positive degree invariant with respect to

2830: $\phi$. That is, the restriction of $\phi$ to $L$ hasn't the form

2831: \begin{equation}

2832: \phi \vert_L \colon L \to L \otimes K_\Si \subset E \otimes K_\Si.

2833: \end{equation}

2834: \end{dfn}

2835: 

2836: It can be shown (see \cite{H2}) that the moduli space $MH_\Si$ of

2837: stable Higgs pairs

2838: \begin{enumerate}

2839: \item contains the cotangent bundle $T^* M^{ss}$ as a Zariski

2840:  dense subset;

2841: \item $T^* M^{ss}$ is different from $MH_\Si$ in codimension 2

2842: and we can use  Hartog's principle;

2843: \item the holomorphic symplectic form can be extended to such

2844: form (with the same notation) $\Om$;

2845: \item the map $\pi$ (3.50) can be extended on $MH_\Si$ to

2846:  a holomorphic

2847:  moment map

2848:  \begin{equation}

2849: \pi \colon MH_\Si \to H^0(\Si, \Oh(2K_\Si))

2850:  \end{equation}

2851:  with compact fibers.

2852:  \end{enumerate}

2853: 

2854: All  these statements can be proved directly and we just

2855:  describe  general

2856: fiber of the map $\pi$. General Higgs field we can consider as an

2857: endomorphism of the

2858:  projectivization

2859:  of our vector bundle $E$

2860:  $$

2861: \phi \colon \PP E \to \PP E = \PP E \otimes K_\Si.

2862: $$

2863: If such map is birational then over every point $p \in \Si$

2864: \begin{enumerate}

2865: \item either $\phi_p$ is a linear automorphism of $\PP E_p$ and

2866: has two fixed points $p_1, p_2 \in \PP E_p$,

2867: \item or the endomorphism $\phi_p$ is degenerated and the

2868: image gives us one point $p \in \PP E_p$.

2869: \end{enumerate}.

2870: Thus we have  double cover

2871: \begin{equation}

2872: \phi \colon \Si_\phi \to \Si

2873: \end{equation}

2874: with  ramification  divisor

2875: \begin{equation}

2876: \xi_\phi = (\wedge^2 \phi)_0 \in \vert 2K_\Si \vert

2877: \end{equation}

2878: (recall that all our endomorphisms are traceless).

2879: 

2880: The curve $\Si_\phi$ is called  spectral curve. It lies on the

2881: ruled surface $\PP E$ and we have the standard adjoint sequence of

2882: sheaves on this surface

2883: \begin{equation}

2884: 0 \to \Oh(- \Si_{\phi}) \to \Oh \to \Oh_{\Si_{\phi}} \to 0.

2885: \end{equation}

2886: 

2887: Multiplying this sequence by the Grothendieck line bundle $H$ on

2888: $\PP E$ (recall

2889:  that for the projection to base $R^0pr(H) = E, R^1pr (H) = 0 $) we get the

2890:   exact triple

2891:   \begin{equation}

2892: 0 \to \Oh(- \Si_\phi) \otimes H \to H  \to \Oh_{\Si_\phi}(H) \to

2893: 0.

2894: \end{equation}

2895: The exact sequence of this triple for direct images gives us the

2896:  isomorphism

2897: \begin{equation}

2898: 0 \to E \to R^0(\Oh_{\Si_{\phi}} (H)) \to 0

2899: \end{equation}

2900: because of the restriction

2901: $$

2902: \Oh(-\Si_phi) (H )\vert_{\PP E_p} = \Oh(-2) \otimes \Oh(1) =

2903: \Oh(-1)

2904: $$

2905: on the projective line.

2906: 

2907: 

2908: So  on the spectral curve $\Si_\phi$ our vector bundle $E$ defines

2909: the line bundle $\Oh_{\Si_{phi}} (H) \in Pic (\Si_\phi)$.

2910: 

2911: For  double cover $\phi$ (3.55) the norm map

2912: \begin{equation}

2913: Nm \colon J(\Si_\phi) \to J(\Si)

2914: \end{equation}

2915: has connected preimage of zero

2916: \begin{equation}

2917: Nm^{-1}(0) = Pr_\phi

2918: \end{equation}

2919: which is an abelian variety which is called  {\it Prym variety}.

2920: 

2921: The isomorphism (3.59) shows that

2922: \begin{equation}

2923: Nm (\Oh_{\Si_\phi} (H)) = \wedge^2 E = \Oh.

2924: \end{equation}

2925: Thus

2926: \begin{equation}

2927: \Oh_{\Si_\phi} (H) \in Pr_\phi.

2928: \end{equation}

2929: 

2930: 

2931: So,  general quadratic differential

2932:  $w \in H^0(\Si, \Oh(2 K_\Si)$ has

2933:  zero-set

2934: \begin{equation}

2935: (w)_0 = p_1, \dots , p_{4g-4}

2936: \end{equation}

2937: and defines the double cover

2938: \begin{equation}

2939: \phi \colon \Si_w \to \Si

2940: \end{equation}

2941: with the  ramification locus (3.64).

2942: 

2943: It is easy to see that every $L \in Pr_\phi$ defines rk2 vector

2944:  bundle on $\Si$:

2945: \begin{equation}

2946: E_L = R^0 L

2947: \end{equation}

2948: with $\wedge^2 E_L = \Oh.$ This bundle is stable as a Higgs

2949: bundle if the cover is irreducible.

2950: 

2951: Thus the fiber of the map $\pi$ (3.54) is

2952: \begin{equation}

2953: \pi^{-1} (w) = Pr_\phi.

2954: \end{equation}

2955: 

2956: 

2957: 

2958: 

2959: 

2960: \subsection{Gauge theory on  Riemann surface. Higgs fields.}

2961: 

2962: 

2963:  The new features of the gauge theory is a necessity to

2964:  consider objects  of different dimensions.

2965:  For example,

2966: Riemann surface bounds  threefold (handlebody) and we come to

2967: 3-dimensional theory to get a result in 2-dimensional case. In

2968: the same vein sometimes it is reasonable to consider graphs

2969: (1-dimensional complexes)

2970:  instead of  Riemann

2971: surfaces and so on.

2972: 

2973: A classical field theory on a manifold $M$ has three

2974: ingredients:

2975:  \begin{enumerate}

2976:  \item a collection $\sA$ of {\it fields} on $M$, which are

2977:  geometric

2978: objects, such as sections of vector bundles, connections on

2979:  vector bundles,

2980: maps from $M$ to some auxiliary manifold (the target space) and

2981: so on;

2982: \item an {\it action functional}

2983: $ S \colon \sA\to\C$ which is an integral of a function $L$ (the

2984: Lagrangian) of fields;

2985: \item a collection of observable functionals on the space of

2986:  fields,

2987: $ w \colon \sA\to\C.$

2988:  \end{enumerate}

2989: For example: {\it  Chern--Simons functional}. Here

2990: \begin{enumerate}

2991: \item

2992:   $M$ is a

2993: 3-manifold, \item the collection of fields

2994:  \[

2995:  \sA=\Om^1(M) \tensor \fsu (2)

2996:  \]

2997: is the space of 1-form with coefficients in the Lie algebra,

2998: \item the action functional

2999:  \begin{equation}

3000: S(a)=\frac{1}{8} \pi^2\int_M tr(a \cdot d a + \frac{2}{3} a^3),

3001:  \end{equation}

3002: \item as an observable  we can consider a {\it Wilson loop},

3003: given by some knot $K\subset M$:

3004:  \[

3005:  W_K(a)= tr(Hol_K(a))

3006:  \]

3007: - the trace of  holonomy of  connection $a$ around  knot $K$.

3008: \end{enumerate}

3009: \begin{rmk} These functions on the space of fields $\sA$ form

3010: an algebra because of the Mandelstam constraint:

3011: $$

3012: tr(A ) \cdot tr(B) = tr(A \cdot B) + tr( A^{-1} \cdot B).

3013: $$

3014: \end{rmk}

3015: Our space $CLRep(\pi_1(\Si), \SU(2))$ consists  of classes of

3016: $\SU(2)$-representations of the fundamental group of a Riemann

3017: surface of genus $g$. This space contains  the subspace of

3018: reducible representations

3019:  \begin{equation}

3020:  CLRep(\pi_1(\Si), \SU(2))^{red} \subset CLRep(\pi_1(\Si),\SU(2)).

3021:  \end{equation}

3022: To get this space as the phase space of some mechanical system,

3023:  consider a

3024: compact smooth Riemann surface $\Si$ of genus $g > 1$ and trivial

3025: Hermitian vector bundle $E_h$ of rank 2 on it.

3026:  As usual, let $\sA_h$ be the

3027: affine space (over the vector space $\Om^1(\End E_h)$)

3028:  of Hermitian

3029: connections and $\sG_h$ is the corresponding Hermitian gauge

3030: group. This space has the subspace:

3031:  \[

3032:  \sA_h^{red} \subset \sA_h

3033:  \]

3034: where the left-hand side is the subset of reducible connections.

3035:  As usual,

3036: let

3037:  \[

3038: \sA_h^*=\sA_h-\sA_h^{red}.

3039:  \]

3040: 

3041: The sending of  connection to its curvature tensor defines a

3042: $\sG_h$-equivariant map

3043:  \begin{equation}

3044: F \colon \sA(E_h)\to\Om^2(End E_h)=Lie(\sG_h)^*

3045:  \end{equation}

3046: to the coalgebra Lie of the gauge group.

3047: 

3048: We can consider this map as the moment map with respect to the

3049: action of $\sG_h$. The subset

3050:  \begin{equation}

3051: F^{-1}(0)=\sA_F

3052:  \end{equation}

3053: is the subset of flat connections.

3054: 

3055: For a connection $a\in \sA_F$  a tangent vector to $\sA_h$ at $a$ is

3056:  \begin{equation}

3057:  \om\in\Om^1(\End E_h)=T \sA_h,

3058:  \end{equation}

3059: and for it the following condition holds

3060:  \begin{equation}

3061:   \na_a (\om)=0.

3062:  \end{equation}

3063: The trivial vector bundle $E_h$ admits  zero connection

3064:  $\theta $,

3065: which is interesting and important from many points of view. In

3066: particular it provides possibility to identify $\sA_h$ with

3067: $\Om^1(End E_h)$  sending  connection $a$ to the

3068:  form $a - \theta$. We

3069: will identify forms and connections  this way.

3070: 

3071:  The space $\sA_h=\Om^1(End E_h)$ is the {\it collection of fields} of

3072: YM-QFT with the {\it Yang--Mills functional}

3073:  \[

3074:  S(a)=\int_{\Si} |F_a|^2.

3075:  \]

3076: Thus $\sA_F$ is a {\it classical phase space}, that is, the space

3077: of solutions of the {\it Euler--Lagrange equation} $\de S(a)=0$.

3078: 

3079: There exists a symplectic structure on the affine space $\sA_h$,

3080: induced by the canonical 2-form given on the tangent space

3081: $\Om^1(End E_h)$ at a connection $a$ by the formula

3082:  \begin{equation}

3083:  \om (\ga_1,\ga_2)=\int_{\Si} tr (\ga_1 \wedge\ga_2).

3084:  \end{equation}

3085: This form is $\sG_h$-invariant, and its restriction to

3086: $\sA_F^{irr}$ is degenerate along $\sG_h$-orbits: at a connection

3087: $a$, for a tangent vector $\ga \in\Om^1(End E_h)$, we have

3088:  \[

3089:  \om\in T\sG_h \text{if and only if}  \om=\na_a \phi \quad\text{for $

3090:  \phi\in\Om^0(\End E_h)=Lie(\sG_h)^*$,}

3091:  \]

3092: and

3093:  \[

3094:  \int_{\Si} tr (\na_a \fie \wedge\om)=\int_{\Si}tr(\fie\wedge\na_a\om)=0.

3095:  \]

3096: Hence

3097:  \begin{equation}

3098:  \om\in T \sA_F \text{if and only if} \Om_0 (\na_a \fie,\om)=0.

3099:  \end{equation}

3100: 

3101: Interpreting (3.70) as a moment map and using symplectic reduction

3102: arguments, we get a non degenerate closed symplectic form $\om$ on

3103:  the space

3104:  \begin{equation}

3105:  \sA_F /\sG_h= CLRep(\pi_1(\Si), \SU(2))

3106: \end{equation}

3107: of classes of $\SU(2)$-representations of the fundamental group

3108: of the Riemann surface, and a stratification of this space. The

3109: form $\om$ defines  symplectic structure on $CLRep(\pi_1(\Si),

3110: \SU(2))^{irr})$ and  symplectic orbifold structure on

3111: $CLRep(\pi_1(\Si), \SU(2))$.

3112: 

3113: \begin{rmk}

3114: The symplectic structure $\om$ is defined in  pure topological

3115: way (see \cite{G1}).

3116: \end{rmk}

3117: 

3118: 

3119: On the other hand, the form $\om_0$ on $\sA_h$ is the differential

3120:  of the

3121: 1-form $D$ given by the formula

3122:  \begin{equation}

3123: D(\om)=\int_{\Si} tr((a) \wedge\om).

3124:  \end{equation}

3125: We consider this form as a unitary connection $A_0$ on the trivial

3126: principal \hbox{$U(1)$-bundle} $L_0$ on $\sA_h$.

3127: 

3128: To descend this Hermitian bundle and its connection to the

3129:  orbit space,

3130: one defines  $\Theta$-cocycle (or $\Theta$-torsor) on the

3131:  trivial line

3132: bundle (see \cite{RSW}). This cocycle is  $U(1)$-valued function

3133: $\Theta$ on $\sA_h\times\sG_h$ defined as follows: for any triple

3134: $(\Si,a,g)$ where $(a,g)\in\sA_h\times\sG_h$, we can find a

3135: triple $(Y,A,G)$ where $Y$ is a smooth compact 3-manifold, $A$ a

3136: $\SU(2)$-connection on the trivial vector bundle $\sE$ on $Y$ and

3137: $G$ is a gauge transformation of it, such that

3138:  \begin{equation}

3139:  \p Y=\Si, \quad a=A \vert_{\Si} \quad \text{and} \quad g=G\vert_{\Si}.

3140:  \end{equation}

3141: Then

3142:  \begin{equation}

3143:  \Theta (a, g)=e^{i(CS(A^G)-CS(A))}.

3144:  \end{equation}

3145: 

3146: Recall that the Chern--Simons functional on the space $\sA

3147: (\sE_h)$ of unitary connections on the trivial vector bundle is

3148: given by the formula

3149:  (3.68).

3150: It can be checked that the function (3.79) does not depend

3151:  on the

3152: choice of the triple $(Y,A,G)$ (see \cite{RSW}, \S2).

3153: 

3154: The differential of $\Theta$ at $(a, g)$ is given by the formula

3155:  \begin{multline}

3156:  d\Theta(\om,\phi)= \\

3157:  \frac{\pi i}{4}\,\Theta\int_{\Si}\Bigl(tr(g^{-1} d g\wedge g^{-1}\om

3158:  g)-tr(a\wedge\na_{a^g}\phi)+2tr(F_{a^g} \wedge \phi) \Bigr),

3159:  \end{multline}

3160: where $\om\in\Om^1(End E_h)$ and $\phi\in\Om^0(\End

3161: E_h)=Lie(\sG_h)$.

3162: 

3163: But the restriction of this differential to the subspace of flat

3164: connections is much simpler:

3165:  \begin{equation}

3166:  d \Theta (\om, \phi)=\frac{\pi i}{4}\,\Theta\int_{\Si} tr(g^{-1}d g

3167:  \wedge g^{-1}\om g),

3168:  \end{equation}

3169: and is independent of the second coordinate.

3170: 

3171: That this function is  a cocycle ndeed results from the functional

3172: equation

3173:  \begin{equation}

3174: \Theta (a, g_1 g_2)=\Theta (a, g_1) \Theta (a^{g_1}, g_2).

3175:  \end{equation}

3176: Using this function as a torsor $\sA_h \times_{\Theta}U(1)$, we

3177: get a principal $U(1)$-bundle $S^1(L)$ on the orbit space

3178: $\sA_h/\sG_h$:

3179:  \begin{equation}

3180: S^1(L)=(\sA_h\times S^1)/\sG_h,

3181:  \end{equation}

3182: where the gauge group $\sG_h$ acts by

3183:  \begin{equation}

3184: g(a, z)=(a^g, \Theta(a,g) z),

3185:  \end{equation}

3186: on the line bundle $L$ with a Hermitian structure.

3187: 

3188: Following \cite{RSW}, let us restrict this bundle to the subspace

3189: of flat connections $\sA_F$. Then one can check that the

3190: restriction of the form $D$ (3.77) to $\sA_F$ defines a

3191: $U(1)$-connection $A_{CS}$ on the line bundle $L$.

3192: 

3193: By definition, the curvature form of this connection is

3194:  \begin{equation}

3195: F_{A_{CS}}=i\cdot\Om.

3196:  \end{equation}

3197: 

3198: Thus the quadruple

3199:  \begin{equation}

3200: (CLRep(\pi_1(\Si), \SU(2)),\om, L, A_{CS})

3201:  \end{equation}

3202: is a {\it prequantum system} and we are ready to perform the

3203: Geometric Quantization Procedure described in section 1.

3204: 

3205: 

3206: 

3207: 

3208: 

3209: \subsection{Complex polarization of $CLRep(\pi_1(\Si), \SU(2))$}

3210: 

3211: The standard way of getting a complex polarization is to give to

3212: Riemann surface $\Si$ of genus $g$ a conformal structure $I$.

3213:  We get a complex

3214: structure on the space of classes of representations

3215: $CLRep(\pi_1(\Si), \SU(2))$ as follows: let $E$ be our complex vector

3216: bundle and $\sA$ the space of all

3217:  connections on

3218: it. Every connection $a\in\sA$ defines the correspoding covariant

3219: derivative $\na_a\colon\Ga(E)\to\Ga(E\tensor T^*X)$, a first

3220: order differential operator with the ordinary derivative $d$ as

3221: the principal symbol. A complex structure gives the decomposition

3222: $\p+\op$, so any covariant derivative can be decomposed as

3223: $\na_a= \nabla^{1,0} + \na^{0,1}$,

3224:  where

3225: $\na^{1,0}\colon\Ga(E)\to\Ga(E\tensor\Om^{1,0})$ and $\na^{0,1}

3226: \colon\Ga(E)\to\Ga(E\tensor\Om^{0,1})$. Thus the space of

3227: connections admits a decomposition

3228:  \begin{equation}

3229:  \sA=\sA'\times \sA'',

3230:  \end{equation}

3231: where $\sA'$ is an affine space over $\Om^{1,0}(ad E)$ and $\sA''$

3232: is an affine space over $\Om^{0,1}(ad E)$.

3233: 

3234: The group $\sG$ of all automorphisms of $E$ acts as the group of

3235: gauge transformations, and the projection $pr \colon \sA\to\sA''$

3236: to

3237:  the space

3238: $\sA''$ of $\op$-operators on $E$ is equivariant with respect to

3239: the $\sG$-action.

3240: 

3241: Giving $E$ a Hermitian structure $h$, we get the subspace

3242: $\sA_h\subset \sA$ of Hermitian connections, and the restriction

3243: of the

3244:  projection $pr$

3245: to $\sA_h$ is one-to-one. Under this Hermitian metric $h$, every

3246: element $g\in\sG$ gives an element $\overline{g} = (g^*)^{-1}$

3247: such that

3248:  \[

3249:  \overline{g} = g \text{if and only if}  g\in\sG_h.

3250:  \]

3251: Now for $g\in\sG$, the action of $\sG$ on the component $\sA''$ is

3252: standard:

3253:  \[

3254:  \na^{0,1}_{g(a)}=g\cdot \na^{0,1}_a\cdot g^{-1}=\na^{0,1}_a-(\na^{o,1}_a g)

3255:  \cdot g^{-1};

3256:  \]

3257: and the action on the first component $\sA'$ of $\p$-operators is

3258:  \[

3259:  \na^{1,0}_{g(a)}=\overline{g}\cdot\na^{1,0}_a\cdot\overline{g}^{-1}=

3260:  \na^{1,0}_a-((\na^{0,1}_a g)\cdot g^{-1})^*.

3261:  \]

3262: It is easy to see directly that the action just described

3263: preserves unitary connections:

3264:  \begin{equation}

3265:  \sG(\sA_h)=\sA_h,

3266:  \end{equation}

3267: and that the identification $\sA_h=\sA''$ is equivariant with

3268: respect to this action.

3269: 

3270: It is easy to see that $(\na^{0,1}_a)^2 = 0\in\Om^{0,2}(ad E)$.

3271:  Thus the orbit

3272: space

3273:  \begin{equation}

3274: \sA'' /\sG=\bigcup \sM_i

3275:  \end{equation}

3276: is the union of all components of the moduli space of

3277: topologically trivial $I$-holomorphic bundles on $\Si_I$. (Although this

3278: union doesn't admit any good structure, as it contains all

3279: unstable vector bundles, however we can use the notion {\it stack}

3280: for such situation .)  Finally, the image of $\sA_F\in \sA_h$ is

3281: the component $M^{ss}$ of maximal dimension ($3g-3$) of s-classes

3282: of  semi-stable vector bundles. Thus by classical technique of GIT

3283: of Kempf--Ness type we get:

3284: 

3285:  \begin{prop} {\it (Narasimhan--Seshadri)}

3286:  \[

3287:  CLRep(\pi_1(\Si), \SU(2))=M^{ss}.

3288:  \]

3289:  \end{prop}

3290: 

3291:  \begin{prop} The form $F_{A_{CS}}$ (3.85) is a $(1,1)$-form and

3292: the line bundle $L$ admits  unique holomorphic structure

3293: $\Oh(\Theta)$ compatible with the Hermitian connection $A_{CS}$.

3294:  \end{prop}

3295: 

3296: On the other hand,  complex structure $I$ on $\Si$ defines a

3297: K\"ahler metric on $M^{ss}$ (the so-called Weil--Petersson metric)

3298: with K\"ahler form

3299:  \begin{equation}\om_{\mathrm{WP}}=i F_{A_{CS}}=i\cdot\om.

3300:  \end{equation}

3301: This metric defines the Levi-Civita connection on the complex

3302: tangent bundle $T M^{ss}$, and hence a Hermitian connection

3303: $A_{LC}$ on the line bundle

3304:  \begin{equation}

3305: \det T M^{ss}= \Oh(4 \Theta),

3306:  \end{equation}

3307: and a Hermitian connection $\frac{1}{4}A_{LC}$ on $L$ compatible

3308: with the holomorphic structure on $L = \Oh(\Theta)$. Thus we have

3309: 

3310:  \begin{prop}

3311:  \[

3312:  \frac{1}{4} A_{LC}=A_{CS}.

3313:  \]

3314:  \end{prop}

3315: 

3316: Finally, considering $M^{ss}$ as a family of $\op$-operators, we

3317: get the Quillen determinant line bundle $L$ with a Hermitian

3318: connection $A_Q$ with curvature form

3319:  \begin{equation}

3320: F_{A_Q}=i\cdot\om.

3321:  \end{equation}

3322: Hence we can extend the equality of Proposition 4:

3323:  \begin{prop}

3324:  \[

3325:  \frac{1}{4} A_{LC}=A_{CS}=A_Q.

3326:  \]

3327:  \end{prop}

3328: 

3329: Now we would like to describe Higgs fields in  gauge terms.

3330: 

3331: A holomorphic Higgs field $\phi$ is given by a section from

3332: $H^0(\Si_I, ad E \otimes K)$ and corresponding quadratic

3333: differential

3334: \begin{equation}

3335: \pi (E, \phi) = tr(\phi^2).

3336: \end{equation}

3337: Every quadratic differential $w$ is a holomorphic function on the

3338: dual space $\Om^{0,1}(\Si_I, K^{-1})$ of Beltrami differentials.

3339:  This function is given

3340: by the integral

3341: \begin{equation}

3342: H_w = \int_\Si tr(\phi^2) \cdot \dot I.

3343: \end{equation}

3344:  Obviously this function from the cotangent bundle

3345: $T^* M^{ss}$ can be extended to the full moduli space of Higgs pair

3346: $MH_\Si$.

3347: 

3348: \begin{prop} For every two holomorphic quadratic differentials

3349:  $w, w'$

3350: functions $H_w$ and $H_{w'}$ are Poisson commuting with respect

3351: to the holomorphic symplectic structure.

3352: \end{prop}

3353: Indeed this question can be lifted to the affine space of

3354: connections or Cauchy- Riemann operators on $E$ (3.87). The new

3355: affine space is

3356: \begin{equation}

3357: T^* \sA'' = \sA'' \times \Om^{0,1}(ad E).

3358: \end{equation}

3359: But every function $H_w$ as a function on this flat space

3360: doesn't depend on $\sA''$  - variables. But $\sA''$-variables and

3361: $\Om^{0,1}(ad E)$ - variables are conjugate with respect to the

3362: canonical symplectic form on the flat space. The zero-level of

3363: the moment map for the complex gauge group action (see (3.88)) is

3364: precisely the space of holomorphic Higgs fields.  Thus $H_w$ and $H_{w'}$

3365: Poisson commute.

3366: 

3367: \subsection{Computations of ranks}

3368: 

3369: Computations of ranks in non-abelian setup is in a sense

3370: absolutely new procedure using a new idea -  consideration

3371:  of full collection of moduli spaces

3372:  \begin{equation}

3373:  M^{ss}_{2}, M^{ss}_{3}, \dots , M^{ss}_{g}, \dots

3374:  \end{equation}

3375: for all genus. In the complex polarization case ideas are coming

3376: from the Donaldson theory for 4-folds (see, for example, \cite{T2}

3377: and \cite{T3}) where the second Chern class $c_{2}(E)$ is playing

3378: the role of the parameter $g$. In the real polarization case

3379:  constructions are coming from  Conformal Field Theory.

3380:  \begin{rmk}

3381:  It will be very useful  to relate the Donaldson theory and CFT directly

3382:   as it was done by Nakajima for Hilbert schemes $Hilb^{g} S$ of

3383:   an algebraic surface $S$.

3384:  \end{rmk}

3385: 

3386: So let $\Si$ be a smooth algebraic curve of genus $g$ and

3387:  \begin{equation}

3388: \sS (\Si)= Sym (H_{even}(\Si,\R)) \otimes \Lambda(H_{odd}(\Si, \R))

3389: \end{equation}

3390:  be the graded algebra where $\deg(\si)=4-i$ if $\si \in

3391: H_i(\Si)$. Let $M^{ss}_g$ be the moduli space of rank 2

3392: semi-stable vector bundles on $\Si$ with trivial determinant. On

3393: the direct product $ \Si \times M^{ss}_g$ there isn't the

3394: universal bundle $\sE$ but there exists the bundle $\ad \sE$ with

3395: its Chern class

3396:  \begin{equation}

3397: - p_1(\sE)= c_1(\sE)^2 - 4 c_2(\sE)= c_2(\ad \sE) \in H^4(\Si \times

3398: M^{ss}_g).

3399:  \end{equation}

3400: This class is defined by the slant product, that is, the

3401: homomorphism $\mu\colon H_*(\Si)\to H^*(M^{ss}_g)$ defined by

3402:  \begin{equation}

3403: \mu (\si)=-\frac{1}{4} p_1(\sE)/ \si.

3404: \end{equation}

3405: So, $ \deg(\al)=i \implies \mu(\al) \in H^i(M^{ss}_g)$.

3406: 

3407: More precisely,

3408:  \begin{equation}

3409: -p_1(\sE)=-2[\Si] \otimes [\Theta]+\pt \otimes x - 4

3410: \sum_{i=1}^g(a_i \otimes \mu(a_i)+b_i \otimes \mu(b_i)),

3411: \end{equation}

3412: where the cohomology class of the theta diviser

3413:  \begin{equation}

3414: [\Theta] \in H^2(M^{ss}_g, \Z);\quad  x \in H^4(M^{ss}_g, \Z);

3415:  \end{equation}

3416: $[\Si]$ is the fundamental class of $C$; $[\pt]$ is the class of a

3417: point of $\Si$ and $ a_i, b_i, i=1,\dots,g$ is a symplectic basis

3418: of $H^1(\Si, \Z)$.

3419: 

3420: \begin{dfn}

3421: The Newstead invariant is a linear function

3422:  $$

3423: N_g \colon \sS (\Si) \to \Q

3424:  $$

3425: sending a typical monomial $\al= \si_1 \cdot \cdot \cdot \si_r$

3426: where $\si_i \in H_{n_i}(\Si)$ of total degree

3427: $d(\al)=\sum_{i=1}^{r} (4-n_i)$

3428:  to 0 if $d \neq 6g-6= dim M^{ss}_g$ and to the index of

3429:  intersection

3430:  \begin{equation}

3431:  \mu(\si_1) \cdot \cdot \cdot \mu(\si_r)

3432:  \end{equation}

3433:  if $d=6g-6= dim M^{ss}_g$.

3434: Such homogeneous polynomial of degree $3(g-1)$ we will call {\it

3435: Newstead polynomial}.

3436: \end{dfn}

3437:  \begin{rmk}

3438:   Of course we can apply this constructions to any

3439:  family of vector bundles, that is, to any vector bundle $\sE$ over

3440:  $ \Si  \times B $ where $B$ is a base of this family. Of course in this

3441:  case we have one

3442:  polynomial  which we will denote by $ N(\sE)$.\end{rmk}

3443: 

3444: Every Newstead polynomial  is  $Diff \Si$-invariant. From this we

3445: can see its shape:

3446: \begin{enumerate}

3447: \item $$

3448:  \quad N_g( \mu([\Si])^m \cdot \mu([\pt])^n \cdot \mu(a_{i_{\min}

3449:  })\cdot \mu(b_{j_{\min} }) \cdot \cdot \cdot) \neq 0 \implies i_{\min}

3450: =j_{\min}.

3451:  $$

3452:  Indeed, we  use the commutation relations,

3453:  the diffeomorphism invariance of

3454:  the Newstead polynomials and the existence of an orientation

3455:  preserving

3456:  diffeomorphism $\phi \colon \Si \to \Si$ satisfying

3457:  $$

3458:  \phi^* ( a_i)= -a_i;

3459:  \phi^*(b_i)=-b_i; \phi^*(a_j)=a_j, \phi^*(b_j)=b_j, i\neq j

3460:  $$

3461:  (the

3462:  Dehn twist) and get the statement.

3463: \item The value

3464:  $ N_g( \mu([\Si])^m \cdot \mu([\pt])^n \cdot \mu(a_{i_1

3465: })\cdot \mu(b_{i_1}) \cdots )$ is independent on the choice of

3466: handles $i_j$ because of

3467:  the existence of an orientation preserving diffeomorphism

3468: interchanging any pair of handles;

3469:  \item the expression $ \ga=2

3470: \sum_{i=1}^{g} \mu(a_i) \cdot \mu(b_i)$ is independent on the

3471: choice of symplectic basis $\{a_i, b_i \}$.

3472: \end{enumerate}

3473: 

3474: From this it is easy to see that if

3475: \begin{equation}

3476: \ga =2 \sum_{i=1}^g \mu(a_i) \cdot \mu(b_i) \in H^6(M^{ss}_g )

3477: \end{equation}

3478: then

3479:  \begin{equation}

3480: N_g \in \C[[\pt],[\Si], \ga].

3481: \end{equation}

3482: Thus this polynomial is nonzero only on the monomials of the type

3483: $ [C]^m \pt^n \ga^p;\text{ for } m+2n+3p=3g-3 $.

3484: 

3485: Since the nondegenerate pairing (3.102) only involves the classes

3486: $[\Si], [\pt], \ga $  and does not depend on the conformal

3487: structure on $\Si$, we can reduce our  of the graded algebra (3.97)

3488: to the "universal" algebra

3489:  \begin{equation}

3490: \Bbb S_3=  Sym ( \al \cdot \Z \oplus \be \cdot \Z \oplus \ga \cdot

3491: \Z),

3492: \end{equation}

3493: where $\deg \al=1; \deg \be=2, \deg \ga=3 $. Then the collection

3494: of the polynomials $N_g$ becomes the linear map

3495:  \begin{equation}

3496: N \colon \Bbb S_3 \to \Q

3497: \end{equation}

3498: sending a monomial of degree $d$ not divisible by 3 to 0 and a

3499: monomial of degree $3n$ to $N_{n+1}$ (3.102) of this monomial. So

3500: we have non homogeneous polynomial $N$ with homogeneous components

3501: of degree $3g-3$.

3502: 

3503: Moreover, there is the following recurrence relation between homogeneous

3504: components of $N$ (3.106):

3505:  \begin{equation}

3506: N_g (\ga \cdot \al)=g \cdot N_{g-1}(\al)

3507: \end{equation}

3508: (see for example \cite{Th2}, Proposition 26). This is very

3509: important point: we get the inductive statement relating the

3510: Newstead polynomials for different genus.

3511: \begin{rmk}

3512: We will see later the fusion rule of the same type in the real

3513: polarization case.

3514: \end{rmk}

3515: 

3516: It is convenient to use the following natural normalization

3517: \begin{equation}

3518:  N^0_{3n}=\frac{1}{(n+1)!}

3519: N_{n+1}.

3520: \end{equation}

3521: We call such function the normalized Newstead polynomial. For

3522: this polynomial

3523:  \begin{equation}

3524:  N^0\colon  \Bbb S_3 \to \Q

3525: \end{equation}

3526: we have the equality

3527:  \begin{equation}

3528:  N^0(\ga \cdot z) = N^0(z)

3529:  \end{equation}

3530:  for the generator $\ga$ and arbitrary $z \in \Bbb S$.

3531: So we can exclude the generator $\ga$ and consider $N^0$ as a

3532: linear map

3533:  \begin{equation}

3534: N^0 : \Bbb S_2= Sym_*(\al \cdot \Z \oplus \be \cdot \Z) \to \Q.

3535: \end{equation}

3536: The  so-called Newstead Conjecture ( see [Th2]) predicts that

3537:  \begin{equation}

3538: N^0(\al^m \cdot \be^k) \neq 0 \implies k \le \frac{1}{3}(m+2k).

3539: \end{equation}

3540:  So it is

3541: convenient to let $\al\cdot\be=\om$ and such that the Newstead

3542: polynomial is

3543:  \begin{equation}

3544: N^0 : Sym_*(\al \cdot \Z \oplus \om \cdot \Z) \to \Q

3545: \end{equation}

3546: satisfying the relation

3547: \begin{equation}

3548: N^0(\al^{3k} \cdot \om^{n})=N_{3(n+k)+1}(\al^{n+3k } \cdot \be^n).

3549:  \end{equation}

3550: From this we have the main equality:

3551:  \begin{equation}

3552: N^0(\al^{3k} \cdot \om^{n})=\frac{(-4)^{n+k}\cdot

3553: (n+3k)!}{(n+k+1)!} \cdot (2k)! \cdot(4^k-2)\cdot B_{2k},

3554: \end{equation}

3555: where $B_i$ is the $i$th Bernoulli number (see \cite{Th2}, formula

3556: (29)).

3557: 

3558: For any linear form $L \in

3559:  Sym_*(\al \cdot \Z \oplus \be \cdot \Z) \oplus \ga \cdot \Z$, the

3560:  interior product by any homogeneous element $x$ gives the differential

3561:   operator

3562:  \begin{equation}

3563:  \frac{\partial}{\partial x} L(z)=(\deg z+\deg x) L(x \cdot z).

3564:  \end{equation}

3565:   Thus we can consider the relations of the type (3.110)

3566:  as a differential equation. From this we  get the Newstead

3567:   polynomials as  solutions to such equations. Moreover we

3568:   recognize  the

3569:  generating functions of the pairing (3.102) as the standard elementary

3570:  functions (see for example \cite{Th1}).

3571: 

3572: \begin{rmk} \begin{enumerate} \item

3573:  For the simplest graded algebra $ Sym_*(\Z \oplus \cdots

3574: \oplus \Z)$ the product $ z_1 \cdot z_2 \in  Sym_{d_1+d_2}$of

3575:  two homogeneous elements $ z_i \in Sym_{d_i}$ is defined as

3576:  $$

3577: z_1 \cdot z_2(\si_1, \dots, \si_{d_1+d_2})=

3578:  $$

3579:  \begin{equation}

3580: = \frac{1}{(d_1+d_2)!}\sum_{g \in \Sigma_{d_1+d_2}}

3581: z_1(\si_{g(1)}, \dots, \si_{g(d_1)}) \cdot z_2(\si_{g(d_1+1)},

3582: \dots, \si_{g(d_1+d_2)}),

3583: \end{equation}

3584: where $\Sigma_n$ is the symmetric group on $n$ letters

3585: \item Witten's approach deals with the volumes of moduli spaces

3586: \begin{equation}

3587: vol (M_{g}^{ss}) = N^{0}(\al^{3g-3})

3588: \end{equation}

3589: and the other values of the Newstead polynomials can be read off

3590: from these formulae.

3591: \item Historically the

3592:  procedure

3593:  was the following: Thaddeus reconstructed values of the Newstead

3594:  polynomial (3.115)

3595:  from the beautiful form of the generating functions for

3596:  ranks

3597:  of   conformal block spaces given by the Verlinde formula

3598:  (see below).

3599: \item To use these computations Thaddeus proposed the following

3600: conjecture: {\it The quantization of $M^{ss}_{g}$ is numerically

3601: perfect} (see subsection 1.5).

3602: \item To avoid Thaddeus's conjecture we  use the

3603: Donaldson's construction from \cite{Do}.

3604: \item Other regular way to get the pairing formuli and the computation of

3605: the collection of Newstead polynomials is the Geometric

3606: Approximation Procedure: there is a chain of birational

3607: transformations or flips (see \cite{Re})

3608: \begin{equation}

3609:  \text{MP}_{\max}\leftrightarrow\dots\leftrightarrow\text{MP}_1

3610: \leftrightarrow\text{MP}_0,

3611:  \end{equation}

3612: where every $\text{MP}_i$ is the moduli space of $t$-stable pairs

3613: (see \cite{Th2}, \cite{BD} or \cite{Ber}), and can be realised as

3614: a family of vector bundles over $\Si$. Thus using the same

3615: construction  we can define  polynomials $NP_{g,i}$ analogeous of

3616: the Newstead polynomials, and the linear forms $NP_i$,  the

3617: analogs of (3.114) and (3.115). Recall that the moduli space on

3618: the extreme left admits the $\PP^{1}$-bundle structure that is

3619: regular map

3620:  \begin{equation}

3621: \text{MP}_{\max}\to M_g^{ss}

3622: \end{equation}

3623: to the moduli space of stable bundle with trivial determinant.

3624: From this it is easy to see that the form

3625:  \begin{equation}

3626: NP^{0}_{\max}\cdot H=N^0

3627:  \end{equation}

3628: where $H$ is the Grothendieck class of the $\PP^{1}$-bundle,

3629: $N^{0}$ is the Newstead form.

3630: \item

3631: Now the moduli space $\text{MP}_0$ on the extreme right-hand end

3632: of (3.119) is the projective space of dimension $3g-2$.

3633:  \begin{equation}

3634: \text{MP}_0=\PP^{3g-2}.

3635: \end{equation}

3636: \item Using the description of every flip in the chain (3.119) we can

3637: reconstruct all polynomials step by step begining with

3638: $\PP^{3g-2}$.

3639: \end{enumerate}

3640: \end{rmk}

3641: 

3642: Using the Hirzebruch-Riemann-Roch theorem we have

3643: \begin{equation}

3644: \sum_{i} rk H^{i}(M_{g}^{ss}, \Oh(k \Theta)) = (ch(\Oh(k\Theta))

3645: \cdot td (M_{g}^{ss}))_{3g-3}.

3646: \end{equation}

3647: Since the divisor $\Theta$ is positive the LHS of previous

3648: equality is equal to

3649: $$

3650: rk \sH^{r}_{I} = rk H^{0}(M_{g}^{ss}, \Oh(k \Theta))

3651: $$

3652: by Kodaira vanishing theorem. On the other hand, Newstead showed

3653: (see \cite{Ne}) that

3654: \begin{equation}

3655: c_{1}(M^{ss}) = 4 \cdot [\Theta] \quad \text{ and } p(M_{g}^{ss})

3656: = (1 + \beta)^{2g-2}.

3657: \end{equation}

3658: 

3659: Hence  the Todd class

3660: \begin{equation}

3661: td(M^{ss}) = exp(2 [\Theta]) (\frac{1/2 \sqrt{\beta}}{sinh(1/2

3662: \sqrt{\beta})})^{2g-2}

3663: \end{equation}

3664: Using values of the Newstead polynomials we get the RHS of (3.123)

3665: :

3666: \begin{equation}

3667: rk H^{0}(M_{g}^{ss}, \Oh(k \Theta)) = (\frac{k+2}{2})^{g-1}

3668: \sum_{j}\frac{1}{sin (\frac{j \pi}{k+2})^{2g-2}}.

3669: \end{equation}

3670: 

3671: \begin{rmk}

3672: All of the computations are due to  Zagier and Macdonald (see

3673: \cite{Th1} for the corresponding references).

3674: \end{rmk}

3675: 

3676: \subsection{Hitchin's  connections}

3677: 

3678: Here we will see that

3679: \begin{enumerate}

3680: \item The sending $ \Si_I$ to $ M^{ss}(\Si_I)$ (3.1) is a

3681:  faithful functor and

3682: \item the sending $\Si_I \to H^0(M^{ss}(\Si_I), \Oh(k \Theta))$ is a

3683: successful quantization.

3684: \end{enumerate}

3685: That is, for $g > 2$ we can reconstruct $\Si_I$ from $M^{ss}$ and

3686: every vector bundle from the collection of vector bundles on the

3687: moduli space $\sM_{g}$ of curves of genus $g$

3688: \begin{equation}

3689:  p_k \colon V_k \to \sM_g ; \quad p_k^{-1}(\Si_I) =  H^0(M^{ss}(\Si_I),

3690:  \Oh(k \Theta))

3691:   \end{equation}

3692: admits a projective flat connection. So, locally the

3693: projective space $\PP H^0(M^{ss}(\Si_I), \Oh(k \Theta))$ doesn't

3694: depend on $I$.

3695: 

3696: The first statement is almost obvious since

3697: \begin{equation}

3698: Sing M^{ss} = K(\Si_I)

3699:  \end{equation}

3700: is the Kummer variety of our curve $\Si_I$. Blowing up the

3701:  finite set $Sing K(\Si_I)$, considering the double cover

3702:  with  ramification in the exceptional divisor and then blowing

3703:   down exceptional

3704:   divisor we get the Jacobian $J(\Si)$. It is easy to see that

3705:    the restriction of non-abelian theta divisor $\Theta$ to

3706:    the Kummer variety gives the divisor $2 \Theta$ on the

3707:    Jacobian.

3708: Now the classical Torelli theorem gives us our curve $\Si_I$.

3709: 

3710: Recall that in the abelian case  under deformations of

3711:  a complex structure on $\Si$ we get vector bundles

3712: 

3713: $\pi \colon V_k \to \sM_g, \quad \pi^{-1}(\Si) = H^0(J(\Si), \Oh

3714: (k \Theta))) $.

3715: 

3716: It is the classical result  ( we saw this already in

3717: terms of the extended Kodaira-Spencer theory  \cite{W})  that $V_k$

3718: (1.29) admits a projective flat connection.

3719: 

3720: In parallel to this fact  Hitchin proved that even for

3721:  non-abelian

3722: theta function case $V_k$ admits

3723:  a projective flat connection.

3724: 

3725: To construct the holomorphic projective connection we can use

3726:  the extended Kodaira-

3727: Spencer theory again. First of all our goodness conditions (1.48)

3728:  holds:

3729: \begin{equation}

3730: H^0(T M^{ss}) = H^1(M^{ss}, \Oh) = 0

3731:  \end{equation}

3732: by Narasimhan-Ramanan result \cite{NR}. Now we have to construct

3733: a holomorphic section $W_0 \in H^0(S^2 (T M^{ss}))$. To do this,

3734: consider a  semi-stable vector bundle $E \in M^{ss}$. Then the fiber

3735: \begin{equation}

3736: T M^{ss}_E = H^1(\Si, \ad E)

3737:  \end{equation}

3738: where $ad E \subset End E$ is the traceless part of the sheaf of

3739:  endomorphisms. The

3740: dual space (using the Serre duality)

3741: \begin{equation}

3742: T^* M^{ss}_E = H^0(\Si, \ad E \otimes K_\Si)

3743:  \end{equation}

3744: where $K_\Si$ is the canonical class of $\Si$.

3745: 

3746: Now a tangent vector to the moduli space $\sM_g$ of

3747:  Riemann surfaces at a point

3748: $\Si_I$ is given by a Beltrami differential

3749: \begin{equation}

3750: \dot I \in H^1(\Si_I, K_\Si^*).

3751:  \end{equation}

3752: We have the cup-product map

3753: \begin{equation}

3754: H^1(\Si_I, K_\Si^*) \otimes H^0(\Si, \ad E \otimes K_\Si) \to

3755: H^1(\Si, \ad E)

3756:  \end{equation}

3757: which is nothing else but a quadratic form on $T^* M^{ss}_E$ and

3758: being extended on full moduli space $M^{ss}$ it gives the required

3759: section of the symmetric power of the tangent bundle. Using the

3760: standard machinery of the end of section 5 we get

3761:  a holomorphic projective connection.

3762: 

3763: The next question is about the curvature. Here it is quite

3764: reasonable to return to the gauge theory terminology.

3765: 

3766: Recall that the affine space of all unitary connections $\sA_h$

3767: is canonically a symplectic manifold. Since the tangent space to

3768: $\sA$ at a connection $a \in \sA$ is given by the space of forms

3769: $\Om^1(\fsu(2))$, the symplectic form is given by the formula

3770: (3.74). This form doesn't depend on $a$, so it is closed. The

3771: gauge group $\sG$ acts on the space $\sA_h$ in the usual way  and

3772: preserves the form $\om$.

3773: 

3774:  For the moment map (3.70) the symplectic quotient is

3775: \begin{equation}

3776: m^{-1} (0) / \sG = CLRep(\pi_1(\Si), \SU(2)).

3777:  \end{equation}

3778: (see (3.70)). The symplectic reduction of the symplectic form

3779: defines a symplectic structure on this space. Thus

3780: \begin{equation}

3781: (CLRep(\pi_1(\Si), \SU(2)), \om)

3782:  \end{equation}

3783: is a phase space of the classical mechanical system. Moreover we

3784: have the prequantization line bundle (3.83) with the connection

3785: (3.85).

3786: 

3787: For every connection $a$ its covariant derivative $\nabla_a$

3788: defines the complex

3789: \begin{equation}

3790: \Om^0(\fsu(2)) \to \Om^1(\fsu(2)) \to \Om^2(\fsu(2))

3791:  \end{equation}

3792: and the tangent space to $CLRep(\pi_1(\Si),\SU(2))$ at a point $a$

3793: is given as the first cohomology group of this complex

3794: \begin{equation}

3795: T CLRep(\pi_1(\Si), \SU(2))_a = H^1_a.

3796:  \end{equation}

3797: 

3798:  Kahler polarization is just the choice of a conformal structure

3799:  which can be done using

3800: the Hodge star operator $*$ acting on the space of forms

3801: \begin{equation}

3802: * \colon \Om^1 \to \Om^1.

3803:  \end{equation}

3804: Our complex structure $I$ acts as a endomorphism

3805: \begin{equation}

3806: I = - *

3807:  \end{equation}

3808: and

3809: \begin{equation}

3810: \om( \ga, I \ga) = - \om( I \ga,  \ga) = - \int_\Si tr(\ga, *\ga).

3811:  \end{equation}

3812: 

3813: Such complex structure defines the decomposition

3814: \begin{equation}

3815: \nabla_a = \nabla^{1,0} + \nabla^{0,1}

3816:  \end{equation}

3817: for every $a \in \sA$ and identifies $\sA$ with  the complex

3818: affine space

3819:  of Cauchy-Riemann operators on $E$ taking (0,1)-part of $\nabla_a$. Thus the

3820:  the tangent space

3821:  \begin{equation}

3822: T^{1,0}_a CLRep(\pi_1(\Si), \SU(2))_I = \Om^{0,1} (\fsu(2)).

3823:  \end{equation}

3824: This complex structure is invariant with respect to the gauge

3825: group $\sG$ action thus we get an integrable complex structure as

3826: the result of the symplectic reduction (3.89).

3827: 

3828: This complex structure $*$ on $\Si$ induces the complex structure

3829: $I$ on the space of classes of representations $CLRep(\pi_1(\Si),

3830: \SU(2))$ and we have the main identification

3831: \begin{equation}

3832: CLRep(\pi_1(\Si))_I = M^{ss} (\Si_{-*}).

3833:  \end{equation}

3834: 

3835: The last remark which we make before the curvature computations is

3836: following:

3837: \begin{prop} The map

3838: \begin{equation}

3839: f \colon H^1(\Si_I, T^{1,0}) \to H^0(M^{ss}, S^2(T^{1,0} M^{ss}))

3840:  \end{equation}

3841: given by the formula

3842: \begin{equation}

3843: f(\dot I)(\phi) = \int_\Si tr(\phi^2) \wedge \dot I

3844:  \end{equation}

3845: is an isomorphism.

3846: \end{prop}

3847: 

3848: Indeed by the Hartog's principle we can extend every holomorphic

3849: quadratic form from $T^* M^{ss}$ to $MH_\Si$. But every

3850: holomorphic form is  constant on the fibers of the map $\pi$

3851: (3.54) (which are compact abelian varieties). Hence the space of

3852: holomorphic quadratic forms is the dual space to the space of

3853: holomorphic quadratic differentials on $\Si_I$ and we are done.

3854: 

3855: Now let $0 \in B \subset \C^m$ be an open set and

3856: \begin{equation}

3857: p \colon \tilde{M^{ss}} \to B , \quad p^{-1} (0) = M^{ss}_I

3858: \end{equation}

3859: be a body of holomorphic deformations of complex structure.

3860: Consider a section

3861: \begin{equation}

3862: t \in H^0(B, TB)

3863:  \end{equation}

3864: that is a holomorphic vector field on $B$. By the classical

3865:  Kodaira-Spencer

3866: theory it define a section

3867: \begin{equation}

3868: \xi \in H^0(B, R^1 T\tilde{M^{ss}}_{/p})

3869:  \end{equation}

3870: that is the family of Kodaira-Spencer classes for deformations

3871: along $t$. Moreover considering our vector field $t$ as a

3872: differential operator on elements of a small open cover of fibers

3873: we get the section

3874: \begin{equation}

3875: D_t \in H^0(B, R^1(D^1(L))_{}/p)

3876:  \end{equation}

3877: that is the family of cohomology classes $H^1(M^{ss}_b, D^1(L))$

3878: over points of $B$. This section describes deformations of a

3879: polarization line bundles over $t$.

3880: 

3881: Having the holomorphic quadratic form (3.133)

3882: \begin{equation}

3883: W \in H^0(B, R^0(S^2(T_{/p})

3884:  \end{equation}

3885: we can lift it to the family of Laplace-Beltrami operators

3886: $\Delta$ (using the Levi-Civita

3887:  connection for the metric (3.140)) over elements of open covers

3888:  of fibers of $p$,

3889: and their differences  define

3890: \begin{equation}

3891: D \in H^0(B, R^1(D^2(L)_{/p}))

3892:  \end{equation}

3893: because the operators of second order have the same principal

3894: symbols. Described in (1.86) general construction gives us the

3895: projective connection

3896:  on the vector bundles $V_k \vert_B$ (3.146) with the principal symbol

3897:  \begin{equation}

3898: - \frac{1}{2k - 4} \si \cdot \de (W) \in H^0(B, R^1

3899: T\tilde{M^{ss}}_{/p}).

3900:  \end{equation}

3901: Moreover, the combination of these operators

3902: \begin{equation}

3903: \p_t + 1/(2k -4) i \Delta +i D

3904:  \end{equation}

3905: gives  globally defined holomorphic {\it heat operator} on $L_I$

3906: \begin{equation}

3907: \p_t + P_t

3908:  \end{equation}

3909: and the principal symbol of $P_t$ is a quadratic form lifted from

3910: the target of the moment map $\pi$ (3.93). Thus by Proposition 14

3911: these symbols  commute as  functions on the cotangent

3912: bundle.

3913: 

3914: 

3915: What we have to do now is just to compute the commutator of these

3916: operators for two holomorphic vector fields $t$ and $t'$ from

3917: (3.147):

3918: \begin{equation}

3919: \p_t P_{t'} - \p_{t'} P_t - [P_t, P_{t'}].

3920:  \end{equation}

3921: The last term  commutator is of order 3  a priori. But its

3922: principal symbol  is the Poisson brackets of symbols of operators

3923: (3.154)

3924:  (see \cite{GS1}

3925: ). They are Poisson  commuting. Thus this commutator

3926: \begin{equation}

3927: \p_t P_{t'} - \p_{t'} P_t - [P_t, P_{t'}] \in H^0(M^{ss}, D^2(L))

3928:  \end{equation}

3929: for any complex structure from $B$.

3930: 

3931: \begin{prop} This operator has to be  constant.

3932: \end{prop}

3933: 

3934: Indeed, in the cohomology sequence of the triple (1.81) there is

3935: the part

3936: \begin{equation}

3937: \to H^0(M^{ss}, S^2 T) \to H^1(M^{ss}, D^1(L))

3938:  \end{equation}

3939: which is a monomorphism because the combination of this

3940: homomorphism with the symbol projection $H^1(M^{ss}, D^1(L)) \to

3941: H^1(T)$ is an isomorphism by Proposition 15 (3.144). Thus the

3942: beginning part of this exact sequence

3943: \begin{equation}

3944: H^0(M^{ss}, D^1(L)) \to H^0(M^{ss}, D^2(L)) \to 0.

3945:  \end{equation}

3946: In particular,

3947: \begin{equation}

3948: H^0(M^{ss}, D^1(L)) = H^1(M^{ss}, D^2(L))

3949: \end{equation}

3950: and in the same vein

3951: \begin{equation}

3952: H^0(M^{ss}, D^1(L)) = \C

3953: \end{equation}

3954: Thus the commutator in (1.154) is a constant and our projective

3955: connection is projectively flat.

3956: 

3957: Problem: is this connection unitary? Or is there another unitary

3958: projective flat

3959:  connection?

3960: 

3961: Summarizing that we have

3962: 

3963: \begin{thm} The Kahler quantization of moduli spaces of topological trivial

3964:  rk2 vector bundles is succesful.

3965: \end{thm}

3966: 

3967: 

3968: 

3969: 

3970: \section{Symplectic geometry of moduli spaces of vector bundles

3971:  }

3972: 

3973: 

3974: \subsection{Goldman $U(1)$-action}

3975: 

3976:  The symplectic-geometrical part of the

3977: non-abelian theory of theta functions is well developed by W.

3978: Goldman, J. Weitsman,

3979:  L. Jeffrey and  J. Hurtubise.

3980: 

3981:  The starting point is the Narasimhan-Seshadri identification

3982:  $ M^{ss}(\Si) = CLRep(\pi_1, \SU(2))$ as a result of the

3983:   Symplectic Reduction described in the previous subsection:

3984:  $\sA_h $ is the space of $\SU(2)$-connections, $\sG_h$ is

3985:  the $\SU(2)$-gauge group, and

3986:  $ \om (\ga_1, \ga_2) = \int_\Si tr (\ga_1 \wedge \ga_2)$ is

3987:  the $\sG_h$-invariant symplectic form on $\sA_h$ (see (3.74)).

3988: 

3989: The map $F \colon \sA \to \Om^2(ad)$ (3.66) sending

3990:   connection to its curvature form is

3991:  the moment map for $\sG_h$ - action. Hence

3992:  $F^{-1}(0) = \sA_F$ is the space of flat connections and

3993:  $( CLRep(\pi_1, \SU(2)), \om )$ is the result of

3994:  the symplectic reduction with respect to this action.

3995: 

3996:  Any simple closed loop $C$ on $\Si$ defines the Goldman

3997:   function on the space of classes of representations

3998:   $CLRep(\pi_1, \SU(2))$:

3999: \begin{equation}

4000: c_C(\rho) = 1/\pi \cdot cos^{-1}(1/2 tr(\rho([C]))) \in [0, 1].

4001: \end{equation}

4002: In \cite{G2} Goldman proved that $c_C$ is a Hamiltonian of

4003: $U(1)$-action on $CLRep(\pi_1, \SU(2))$. This action can be

4004: described as follows:

4005: \begin{enumerate}

4006: \item for every $g \in \SU(2) , \quad g \neq \pm id$  there are exist

4007:  unique element

4008: \begin{equation}

4009: log\quad g \in \fsu(2), \quad |log\quad g|^2 = 2,

4010: \end{equation}

4011: and a real number $t \in \R$  such that

4012: \begin{equation}

4013:  g = e^{t \cdot log\quad g};

4014: \end{equation}

4015: \item obviously  $[g, e^{t \cdot log\quad g}] = 0$ for every

4016: $t \in \R$.

4017: \item Suppose that the class $[C] = a_1 $ in a standard presentation of

4018: the fundamental group of $\Si$:

4019: \begin{equation}

4020:  \pi_1 = <a_1, \dots ,a_g, b_1, \dots , b_g \vert

4021:   \prod_{i=1}^g [a_i, b_i] = id>,

4022: \end{equation}

4023: \item for any $\rho \in CLRep(\pi_1, \SU(2))$

4024: such that $\rho (a_1) \neq \pm id$ the formula

4025: \begin{equation}

4026: \rho (b_1) \to \rho (b_1) \cdot e^{t \cdot log \rho(a_1)}

4027: \end{equation}

4028: defines Hamiltonian $U(1)$-action on $CLRep(\pi_1, \SU(2))$.

4029: \end{enumerate}

4030: It is easy to see that

4031: \begin{equation}

4032: C_1 \cap C_2 = \emptyset \implies \{c_{C_1}, c_{C_2} \}_\om = 0

4033: \end{equation}

4034: where the Poisson brackets are taken with respect to the canonical

4035: symplectic structure $\om$ (3.74). Thus we have to require a

4036: maximal set of disjoint homotopy inequivalent loops which we call

4037: circles . It easy to see (you will see this in a minute) that the

4038: maximal number of circles in such set is equal to 3g-3. Moreover,

4039: let a maximal set of disjoint inequivalent circles be

4040: \begin{equation}

4041: \{ C_1, ... , C_{3g-3} \}.

4042: \end{equation}

4043: 

4044: \begin{figure}[tbn]

4045: \centerline{\epsfxsize=3in\epsfbox{27c.eps}} \caption{\sl Maximal

4046: collection of simple loops} \label{Fig 2}

4047: \end{figure}

4048: 

4049: 

4050: Then removing these circles we get

4051: \begin{equation}

4052: \Si - \{ C_1, ... , C_{3g-3} \} = \cup_{i=1}^{2g-2} \tilde{v_i}

4053: \end{equation}

4054:  is a finite set of trinions (or "pairs of pants").

4055: 

4056: \begin{figure}[tbn]

4057: \centerline{\epsfxsize=3in\epsfbox{28c.eps}} \caption{\sl Pairs

4058: of pants} \label{Fig 2}

4059: \end{figure}

4060: 

4061:  A trinion (or "pair of pants")  plays the role of points

4062: in algebraic geometry under investigations of geometry of

4063: connections: just like we restrict functions to a point we can

4064: restrict connections to a trinion. Roughly speaking, a trinion is

4065: the minimal geometrical object admitting non trivial non-abelian

4066: gauge theory.

4067: 

4068:  Riemann surface with a trinion decomposition (4.8) is called

4069: {\it marked} Riemann surface.

4070:  Any marked surface defines  the {\it dual

4071:  trivalent graph} $\Ga$

4072: \begin{enumerate}

4073: \item  the set of vertices of which

4074: \begin{equation}

4075: V(\Ga) = \{ v_i \} = \{ \tilde{v_i} \}

4076: \end{equation}

4077: is the set of trinions;

4078: \item  and with the set of edges

4079: \begin{equation}

4080: E(\Ga) = \{ e_i \} = \{ C_i \};

4081: \end{equation}

4082: \item and two vertices $v_i$ and $v_j$ are joined by the

4083:  edge $e_l$ iff the corresponding circle

4084: \begin{equation}

4085: C_l = \p \tilde{v_i} \cap \p \tilde{v_j}.

4086: \end{equation}

4087: \end{enumerate}

4088: 

4089: 

4090: \subsection{Real polarization}

4091: 

4092: Every marked Riemanian surface $\Si_\Ga$ defines a real

4093: polarization of the prequantized phase space

4094: \begin{equation}

4095:  (CLRep(\pi_1, \SU(2)),\om, L, A_{CS}).

4096:  \end{equation}

4097: 

4098: It is given in a very geometric way in the set-up of perturbation

4099: theory of 3-dimensional  Chern--Simons theory. As we saw, a mark

4100: of  Riemann surfaces is given by the choice of  maximal

4101: collection of disjoint, noncontractable, pairwise nonisotopic

4102: smooth circles on $\Si$. Any such system contains $3g-3$ simple

4103: closed circles (4.7).

4104: 

4105: 

4106: The isotopy class of a trinion decomposition $\{C_i\}$ gives the

4107: map

4108:  \begin{equation}

4109: \pi_\Ga \colon CLRep(\pi_1, \SU(2)) \to \R^{3g-3}

4110:  \end{equation}

4111: with fixed coordinates $(c_1, \dots, c_{3g-3})$ such that

4112:  \begin{equation}

4113:  c_i (\pi_{\{C_i\}} (\rho))=\frac{1}{\pi} cos^{-1}(\frac{1}{2} tr

4114: \rho([C_i]))\in [0, 1].

4115:  \end{equation}

4116: which are just Goldman functions (4.1).

4117: 

4118: Thus

4119: \begin{enumerate}

4120:  \item

4121: this map is the moment map for the Hamiltonian

4122: $U(1)^{3g-3}$-action on $CLRep(\pi_1, \SU(2))$, thus gives a real

4123: polarization of the system (4.12).

4124:  \item The coordinates $c_i$ are {\it action coordinates} for this

4125: Hamiltonian system ( \cite{GS1}).

4126:  \end{enumerate}

4127: 

4128: 

4129: These functions $c_i$ are continuous on all $CLRep(\pi_1,

4130: \SU(2))$ and smooth inside of the unit cube  $[0,1]^{3g-3}$.

4131: 

4132: Thus we have

4133:  \begin{enumerate}

4134:  \item The image of $CLRep(\pi_1, \SU(2))$ under $\pi_\Ga$ is

4135:  a convex polyhedron

4136:  \begin{equation}

4137:  \Delta_\Ga\subset [0, 1]^{3g-3}.

4138:  \end{equation}

4139: \item The symplectic volume of $CLRep(\pi_1, \SU(2))$ is equal to

4140: the Euclidean volume of $\Delta_\Ga $:

4141:  \begin{equation}\int_{CLRep(\pi_1, \SU(2))}\om^{3g-3}=Vol

4142:  \Delta_\Ga.

4143:  \end{equation}

4144: \item The  set of Bohr--Sommerfeld orbits of the real

4145: polarization $\pi_\Ga$ of level $k$

4146:  \begin{equation}

4147:  BS_{k}(\pi_{\Ga},CLRep(\pi_1, \SU(2)),\om,L,A_{CS})

4148:  \end{equation}

4149:  is   the set of $1/2k$- integer points in

4150:   the polyhedron $\Delta_\Ga$,

4151: and

4152:  \begin{equation}

4153:  lim_{k\to\infty} k^{3-3g}\cdot N_{k-BS}=

4154:  \int_{CLRep(\pi_1, \SU(2))}\om^{3g-3}

4155:  =Vol \Delta_\Ga

4156:  \end{equation}

4157:  just as predicted by general quantization rules.

4158:  \end{enumerate}

4159: 

4160: \begin{figure}[tbn]

4161: \centerline{\epsfxsize=3in\epsfbox{31c.eps}} \caption{\sl Moment

4162: map} \label{Fig 2}

4163: \end{figure}

4164: 

4165: 

4166: To see the statements about Bohr-Sommerfeld fibers remark that

4167:  every vector $w\in \R^{3g-3}$ can be interpreted as a {\it

4168: differential $1$-form} on $\R^{3g-3}$, and by the usual

4169: construction using the symplectic form $\Om$, this defines a

4170: vector field $\xi_w$ tangent to the fibers of $\pi$. Integrating

4171: such vector fields, one defines the collection of transformations

4172:  \begin{equation}

4173:  \{t_w\}= \{e^{\xi_w} \} \subset Diff^+ (CLRep(\pi_1, \SU(2)).

4174:  \end{equation}

4175: 

4176: These transformations preserve the curvature form of the

4177: connection $A_{CS}$. Thus (because $CLRep(\pi_1, \SU(2))$ is

4178: simple connected), there exists a collection of gauge

4179: transformations $\al_w\in\sG_L$ of $L$ such that

4180:  \begin{equation}

4181:  (t_w)^*(A_{CS})=A_{CS}^{\al_w}

4182:  \end{equation}

4183:  (see \cite{AB} for the description of the gauge-orbit with fixed

4184:  curvature form)

4185: 

4186: We can view such gauge transformations as $U(1)$-{\it torsors},

4187: just as in the description of the formulas for classical theta functions

4188: for Abelian varieties (2.49). From this

4189:   the subset $BS_{k} \subset \Delta_{\Ga}$

4190: of Bohr-Sommerfeld fibers of level $k$ is the subset of

4191: $1/2k$-integer points of $\Delta$.

4192: 

4193: To describe the polyhedron $\Delta_{\Ga}$ we have to describe the

4194: space of flat $\SU(2)$-connections on a trinion $pp$ (pair of

4195: pants): let

4196: \begin{equation}

4197:   \p pp = C_1 \cup C_2 \cup C_3.

4198: \end{equation}

4199: Then we have the map

4200: \begin{equation}

4201:   \pi \colon CLRep(\pi_1(pp), \SU(2)) \to \R^3

4202: \end{equation}

4203: with fixed coordinates $(c_1, c_2, c_3)$ just like in (3.139). It

4204: is easy to check (see \cite{JW1}) the following fact

4205: 

4206: \begin{figure}[tbn]

4207: \centerline{\epsfxsize=3in\epsfbox{32c.eps}} \caption{\sl Moment

4208: tetrahedron} \label{Fig 2}

4209: \end{figure}

4210: 

4211: \begin{prop} \begin{enumerate}

4212: \item the image

4213: \begin{equation}

4214:  \pi (CLRep(\pi_1(pp), \SU(2)) = \De_\Theta

4215: \end{equation}

4216: is the tetrahedron given by the following inequalities

4217: \begin{equation}

4218: 0 \leq c_i \leq 1, \quad |c_1-c_2| \leq c_3 \leq min(c_1 + c_2, 2

4219: - c_1 -c_2);

4220: \end{equation}

4221: \item  $\pi$ is 1-1 map, smooth inside of the tetrahedron and continious

4222: everywhere.

4223: \end{enumerate}

4224: \end{prop}

4225: 

4226: Using the dual 3-valent graph $\Ga$ of the trinion decomposition

4227: (4.8) we can describe the polyhedron $\De_\Ga$ (4.15):

4228: 

4229: \begin{prop} The polyhedron $\De_\Ga$ is given by inequalities (4.24) for every

4230:  triple $(c_i, c_j, c_k)$ of the coordinates in $\R^{3g-3}$ corresponding to

4231:  a boundary of any trinion in the decomposition (4.8).

4232:  \end{prop}

4233: 

4234: \subsection{ Bohr-Sommerfeld fibers}

4235: 

4236: 

4237: In these terms we can describe Bohr-Sommerfeld fibers  (using

4238: action coordinates $c_i$ ) as  functions

4239:  \begin{equation}

4240:  w\colon E(\Ga) \to\frac{1}{2k}\{0,1,2,\dots,k\}

4241:  \end{equation}

4242: on the collection of edges of the 3-valent graph $\Ga $ to the

4243: collection of $\frac{1}{2k}$ - integers, such that, for any three

4244: edges $e_{1}, e_{2}, e_{3}$ meeting at a vertex $v$, the following

4245: 3 conditions hold:

4246:  \begin{enumerate}

4247:  \item $w(e_{1}) + w(e_{2}) + w(e_{3})\in \frac{1}{k}\cdot \Z$;

4248: 

4249:  \item $w(e_{1}) + w(e_{2}) + w(e_{3}) \leq 1$;

4250: 

4251:  \item and

4252:  \begin{equation}

4253:  |w(e_{1})-w(e_{2})|\leq w (e_{3}) \leq min( w(e_{1}) + w(e_{2}),

4254:   2 - w(e_{1} - e_{2})).

4255:  \end{equation}

4256:  \end{enumerate}

4257: Such a function $w$ is called an {\it admissible integer weight of

4258: level} $k$ on the graph $\Ga$.

4259: 

4260: \begin{figure}[tbn]

4261: \centerline{\epsfxsize=3in\epsfbox{34c.eps}} \caption{\sl

4262: Bohr-Sommerfeld fibers} \label{Fig 2}

4263: \end{figure}

4264: 

4265: Let $W^{k}(\Ga)$ be the set of such weights.

4266: 

4267: \begin{rmk}

4268: The geometrical meaning of the last inequality (4.26) is the

4269: following: it is the triangle inequality for triangles on

4270: 2-sphere. Thus if we do our graph flat that is fix a cyclic

4271: orientation around every vertex  and write down around every

4272: vertex a triangle with sides orthogonal to edges and of lengths

4273: $w(e_{1}), w(e_{2}), w(e_{3}))$, identifying equal sides we get

4274: triangulated Riemann surface

4275:  $\Si_{w}$  which is

4276: a finite combinatorial cover of $S^{2}$

4277: \begin{equation}

4278: \phi_{w} \colon \Si_{w} \to S^{2}.

4279: \end{equation}

4280: For small genus, for example for the graph $\Theta$, this cover

4281: has degree one and gives just a triangulation of the basic sphere

4282:  $S^{2}$. But from metric point of view this is a $1/2k$-integer

4283:   triangulation.

4284: \end{rmk}

4285: Here is the list of simple but very important properties of

4286: weights:

4287: \begin{prop}

4288: \begin{enumerate}

4289:  \item the number $|W^k(\Ga)|$ of admissible weights of level $k$

4290: is independent of the graph $\Ga$ but depends on genus $g$ only;

4291:  \item

4292:  \begin{equation}

4293:  |W^{k}(\Ga)| =|W_g^k| =\frac{(k+2)^{g-1}}{2^{g-1}}

4294: \sum_{n=1}^{k+1}

4295:  \frac{1}{(\sin(\frac{n\pi}{k+2}))^{2g-2}}

4296:  \end{equation}

4297:  is the Verlinde number (3.126).

4298:  \end{enumerate}

4299: \end{prop}

4300: 

4301: 

4302: To describe the geometry of Bohr-Sommerfeld fibers given by an

4303:  admissible integer weight $w$ (4.25) we have to consider the

4304:  levels of this function:

4305: for $\al\in \{0, \frac{1}{2k},\dots,1\}$ let

4306:  \begin{equation}

4307: w^{-1}(\al)=\Ga_1(\al)\cup\cdots\cup\Ga_n(\al)\subset\Ga

4308: \label{eq5.1}

4309:  \end{equation}

4310: be the decomposition into connected components. Then every

4311: component $\Ga_i(\al)$ is a 3-valent graph with $n_i$ univalent

4312: vertices $a_1,\dots, a_{n_i}$ (see section 6).

4313: 

4314: Every gauge class of connections contains a connection $a_0$

4315: adapted to a trinion decomposition (see \cite{JW1},

4316: Definition~2.2). Fix the filtration

4317:  \begin{equation}

4318: Z(\SU(2))=\Z_2\subset U(1)\subset \SU(2),

4319:  \end{equation}

4320: and view it as the triple $ \{\Z_2, U(1), \SU(2)\}$

4321: 

4322: For a connection $[a]\in \pi^{-1}(w)$, we have the function

4323:  \begin{equation}

4324: e_w\colon E(\Ga)\to \{\Z_2, U(1), \SU(2)\}

4325:  \end{equation}

4326:  sending every loop $C_j$ to the element of $G$ conjugate to the

4327: stabilizer of the monodromy of $[a]$ around this loop, and the

4328: function

4329:  \begin{equation}

4330: v_w\colon V(\Ga)\to \{\Z_2, U(1), \SU(2)\}

4331:  \end{equation}

4332: sending  trinion $P_i$ to the stabilizer of the flat connection

4333: $a\rest{P_n}$. Of course,

4334:  \begin{align}

4335:  & C_j\subset \partial P_n \implies v_w(P_n)\subset e_w(C_j); \notag \\

4336:  & C_1 \cup C_2 \cup C_3=\partial P_n \quad \text{and} \quad

4337: e_w(C_1)=e_w(C_2)=\SU(2) \implies

4338:  \\

4339:  & \qquad\qquad e_w(C_3)=\SU(2) \implies v_w(P_n)=\SU(2), \notag

4340:  \end{align}

4341: and so on.

4342: 

4343: Obviously

4344:  \begin{equation}

4345: e_w(C_j)= \text{$U(1)$ or $\SU(2)$.}

4346:  \end{equation}

4347: 

4348:  \begin{prop} The functions $e_w$ and $v_w$ depend on  $w$ and not on

4349: the choice of $[a]\in \pi^{-1}(w)$.

4350:  \end{prop}

4351: 

4352: More precisely, they depend on the combinatorics of the

4353: decomposition (4.29).

4354: 

4355: Thus $w$ defines the direct products

4356:  \[

4357:  \prod_{C\in E(\Ga)} e_w(C) \quad\text{and}\quad

4358:  \prod_{P\in V(\Ga)} v_w(P),

4359:  \]

4360: and $\prod_{P\in V(\Ga)} v_w(P)$ acts on $\prod_{C\in E(\Ga)}

4361: e_w(C)$ as follows: for

4362:  \begin{gather*}

4363:  g=(g_1,\dots, g_{2g-2})\in \prod_{P\in V(\Ga)} v_w(P)

4364:  \quad\text{with $g_i\in v_w(P_i)$, and}\\

4365: (t_1,\dots, t_{3g-3})\in \prod_{C\in E(\Ga)} e_w(C)

4366:  \quad\text{with $t_n\in e_w(C_n)$,}

4367:  \end{gather*}

4368:  if $C_n\subset \partial P_i\cap\partial

4369: P_j$ then

4370:  \begin{equation}

4371:  g(t_n)=g_i \circ t_n \circ g_j^{-1}.

4372:  \end{equation}

4373: 

4374: \begin{rmk}

4375: This  action described in detail in subsection 6.3 under the

4376: consideration of the gauge theory on graphs.

4377: \end{rmk}

4378: 

4379:  \begin{prop}[\cite{JW1}, Theorem~2.5] The fibre $\pi^{-1}(w)$ is given by

4380:  \begin{equation}

4381:  \pi^{-1}(w)=\prod_{C\in E(\Ga)} e_w(C) \Bigm/ \prod_{P\in V(\Ga)} v_w(P).

4382:  \end{equation}

4383:  \end{prop}

4384: 

4385: 

4386:  \begin{cor}

4387:  The fibre $\pi^{-1}(w)$ is isomorphic to

4388:  \begin{equation}

4389:  \pi^{-1}(w) = T^t \times [(S^3)^p \times (S^2)^s] / G_w,

4390:  \end{equation}

4391: where $t, p$ and $s$ are nonnegative integers and $G_w$ is the

4392: finite Abelian group defined by $w$, or more precisely by the

4393: combinatoric data (4.1); moreover,

4394:  \begin{equation}

4395:  H_1(\pi^{-1}(w))=\Z^t \oplus \Z_2^p.

4396:  \end{equation}

4397:  \end{cor}

4398: 

4399: Translations along the torus $T^t$ in (4.37) are induced by

4400: Hamiltonians lifted from the target space $\R^{3g-3}$ of $\pi$.

4401: 

4402: Jeffrey and Weitsman \cite{JW1} use the normalization of the

4403: action coordinates via branched covers to construct a covariant

4404: constant section $s_w$ of the restrictions of $(L^k,A_{CS})$ to

4405: $\pi^{-1}(w)$.

4406: 

4407: 

4408: 

4409: 

4410: 

4411: 

4412: 

4413: 

4414: Following the real polarization quantization rules from section 1

4415: (formula (1.21)) the wave function space is given by the direct

4416: sum

4417: \begin{equation}

4418: \sH^{k}_{\Ga} = \oplus_{b \in BS_{k}} \C \cdot s_{b}

4419: \end{equation}

4420: where $s_{b}$ is a covariant constant section of the restriction

4421: of our line bundle and connection to $\pi^{-1}(b)$.

4422: 

4423:  Geometrically  this combinatorial description of Bohr-Sommerfeld

4424:  fibers can be done

4425: as follows: consider the space $\R^{3g-3}$ with action coordinates

4426: $c_i$ (4.14). This space contains the integer sublattice

4427: $\Z^{3g-3}\subset \R^{3g-3}$, and we can consider the {\it action

4428: torus}:

4429:  \begin{equation}

4430:  T^A=\R^{3g-3} / \Z^{3g-3}.

4431:  \end{equation}

4432: 

4433: In particular, we get a map

4434:  \begin{equation}

4435:  \pi_A \colon CLRep(\pi_1, \SU(2)) \to T^A

4436:  \end{equation}

4437: which glues at most points of the boundary of $\Delta_{\Ga}$.

4438: 

4439: 

4440: Now every integer weight $w$ (4.25) satisfying (1) and (2), but

4441: {\it a priori} without the Clebsch--Gordan conditions  (3) (4.26),

4442: defines a point of order $2k$ on the action torus

4443:  \[

4444:  w\in T^A_{2k}.

4445:  \]

4446: In particular, the collection $W^k(\Ga)$  of admissible integer

4447: weights (subject to (3) (4.26)) can be considered as a subset of

4448: points of order $2k$ on the action torus:

4449:  \begin{equation}

4450:  W^k(\Ga)\subset T^A_{2k}.

4451:  \end{equation}

4452: Thus for different $\Ga$ of genus $g$ we have {\it a priori}

4453: different subsets $\{W^{k}(\Ga)\}$ of the set of points of order

4454: $2k$ on the action torus $T^{A}$ (4.40).

4455: 

4456: The linear decomposition of the wave function space (4.39)

4457: defines the space uniquely up to the coordinatewise

4458: $|W^k(\Ga)|$-torus action because every covariant constant

4459: section is defined up to $U(1)$-action. But using admissible

4460: integer weights we can define this space as a space

4461:  with a fixed basis

4462:  \begin{equation}

4463: \sH^{k}_{\Ga} = \oplus_{w \in W^{k}_\Ga} \C \cdot w

4464: \end{equation}

4465: and the integer sublattice

4466: \begin{equation}

4467: \Z^{|W^k_\Ga|} \subset    \sH^{k}_{\Ga}.

4468: \end{equation}

4469: All of these spaces are embedded in one large space

4470: \begin{equation}

4471: \sH^{k}_{\Ga} \subset \sH^k =  \oplus_{w \in T^A_{2k}} \C \cdot w

4472: \end{equation}

4473: (see (4.42)) with the integer lattice

4474: \begin{equation}

4475: \Z^{|T^A_{2k}|} \subset    \sH^{k}.

4476: \end{equation}

4477: 

4478: 

4479: \subsection{Delzant model}

4480: 

4481: 

4482: Returning to the moment map (3.148) we come into the set-up of the

4483: toric symplectic  geometry: let $(M, \om)$ be a compact

4484: symplectic smooth manifold of dimesion $2n$ with a {\it smooth}

4485: Hamiltonian action of $n$-dimensional torus $T^n$. Then

4486: action-angle coordinates define the moment map

4487: \begin{equation}

4488:  \pi \colon M \to \De \subset \R^n

4489: \end{equation}

4490: whose image is a convex polyhedron $\De$ in Euclidean $n$-space.

4491: This polyhedron satisfies so called Delzant conditions and

4492: contains complete information on the symplectic geometry of $(M,

4493: \om)$. That is, $\De$ determines the manifold, the symplectic form

4494: and the $T^n$-action (see \cite{D}). Moreover, if a polyhedron

4495: $\De \subset \R^n$ satisfies the Delzant conditions then we can

4496: construct a smooth symplectic manifold with Hamiltomian

4497: $T^n$-action such that the image of the moment map is precisely

4498: $\De$.

4499: 

4500: Moreover, if a phase space $(M, \om)$ is prequantized

4501: and $M$ has a Hodge structure  whose Kahler form is $\om$, then

4502: this Hodge structure can also be reconstructed from $\De$ (see

4503: \cite{G1}).

4504: 

4505: We saw that our space $CLRep(\pi_1(\Si), \SU(2))$ admits the

4506: Hamiltonian $T^{3g-3}$-torus action, but

4507: \begin{enumerate}

4508: \item our phase space $CLRep(\pi_1(\Si), \SU(2))$ is singular if $g>2$;

4509: \item the action isn't smooth over the boundary of $\De$.

4510: \end{enumerate}

4511: 

4512: Nevertheless the polyhedron satisfies  Delzant conditions hence

4513: determines the smooth compact Hodge manifold $DM_\Ga$ with the

4514: smooth $T^{3g-3}$-action which is called  {\it Delzant model of

4515: moduli spaces} (see \cite{T8}). For a complex Riemann surface $\Si_I$ the moduli

4516: space $M^{ss}(\Si_I)$ is an algebraic variety. Every such variety

4517: admits a birational morphism

4518: \begin{equation}

4519:  f \colon M^{ss}(\Si_I) \to DM_\Ga.

4520: \end{equation}

4521: 

4522: The list of properties is the following

4523: \begin{enumerate}

4524: \item The Delzant model $DM_\Ga$ is canonically polarized by the line bundle

4525: $H$ corresponding to the theta polarization of every $M^{ss}$.

4526: \item  The Delzant model is rigid as a complex manifold and hence

4527: the Kahler quantization of $(DM_\Ga, H)$ is succesful. The wave

4528: function spaces are given as spaces of holomorphic sections

4529: \begin{equation}

4530:  H^0(DM_\Ga), H^k)

4531:  \end{equation}

4532: and dimensions of these spaces can be computed by the

4533: Duistermaat-Heckman formula.

4534: \item Bohr-Sommerfeld fibers for $CLRep(\pi_1(\Si), \SU(2))$ and $DM_\Ga$

4535: as sets of points of $\De_\Ga$ coincide.

4536: \item The rank of wave spaces (4.49) is equal to the number

4537:  of Bohr-Sommerfeld fibers

4538: thus the quantization of the dynamical system $(DM_\Ga, \om)$ is

4539: numericaly perfect (see \cite{T8}).

4540: \end{enumerate}

4541: 

4542: 

4543: \section{Two versions of CQFT}

4544: 

4545: Result of any quantization procedure is  creation of a quantum

4546: mechanics which is  QFT in dimension d = 1 when there is only one

4547: variable - a time.

4548:  But our main geometrical object - Riemann surfaces - are also the basic object

4549: of  conformal field theories in dimension 2 (or, more precisely, (1,1)).

4550: 

4551:  Some years ago the constraint of conformal invariance was proposed as

4552: the fundamental principle of quantum field theory. All two-dimensional

4553:  conformal field theories are duing to the theory of

4554: critical phenomena and to the string theory. So, the new  powerful

4555: mathematical tool appeared since large data on the geometry of

4556: Riemann surfaces is encoded in this rich mathematical structure.

4557: Moreover, the results of a succesful Kahler quantization procedure

4558: = the wave function spaces $H^0(M^{ss}(\Si), \Oh ( k \Theta))$

4559: appeared first as spaces of conformal blocks of CQFT.

4560: 

4561: On the

4562: other hand,

4563:  the spaces

4564:  $\sH_\pi^k = \oplus_{b \in BS_k \subset B} \C \cdot

4565: s_b$ as  results of the Bohr-Sommerfeld quantization procedure

4566:  appeared as fibers of projective flat vector bundles over

4567: the moduli space $\sM_g$ of curves of genus $g$ in the

4568: combinatorial version of CQFT in

4569:  \cite{FS} and \cite{MS}. Moreover, a 2-dimensional conformal field theory

4570: was formulated  as algebraic geometry of the moduli spaces of

4571: Riemann surfaces.

4572: 

4573: A very important case of that a pinched Riemann sphere was

4574: investigated by Belavin-Knijnik-Zamolodchikov in \cite{KZ} which

4575: produce  the general ideology as well as thechnical fondations.

4576: 

4577: We can start with the simple "classical" version of CFT.

4578: 

4579: \subsection{WZW-version}

4580: 

4581: 

4582: Recall that the space of classical theta functions has very

4583: important  property to be a space of   unique irreducible

4584: representation of some group or algebra (namely the Heisenberg

4585: group). The space of holomorphic sections $H^0(M^{ss}(\Si), \Oh (

4586: k \Theta))$ is also the space

4587:   of the irreducible representation of the gauge algebra

4588: of WZW RCFT. Such spaces have played a noted role in conformal

4589: field theory.

4590: 

4591: It turn out that in  rational conformal field theories, one encounters the

4592: "conformal blocks" of Belavin, Polyakov and Zamolodchikov coincidencing with

4593:  our spaces  $H^0(M^{ss}(\Si), \Oh ( k \Theta))$ .

4594: 

4595:  First of all recall Wess-Zumino-Witten CQFT

4596:   (WZW for short) set up: let $t$ be

4597: a local parameter in a point $p \in \Si$. Then

4598:   the {\it loop algebra}

4599: \begin{equation}

4600:  sl_2^t = sl(2, \C)\otimes \C[t, t^{-1}]

4601: \end{equation}

4602:  is the {\it horisontal algebra} of WZW.

4603: 

4604: The central extention by the cocycle $k \cdot c$

4605: \begin{equation}

4606: \ov{sl_2^t} = sl(2, \C) \otimes \C[t, t^{-1}] \oplus \C \cdot e

4607: \end{equation}

4608: is the k-{\it gauge algebra of} WZNW and the bracket of two

4609: elements of the loop algebra is given by the formula

4610: \begin{equation}

4611: [X \otimes f, Y \otimes g] = [X, Y] \otimes fg + k \cdot tr(XY)

4612: Res_0(gdf)

4613: \end{equation}

4614: 

4615: The theory of Kac-Moody algebras tells us that this algebra admits

4616:  unique irreducible representation $\sH^k$ such that there

4617: exists a non-zero vector annihilated by

4618:  $sl(2, \C)\otimes \C[t]$. Remark that the central element

4619:   of $\ov{sl_2^t}$ acts by multiplication by $k$ (which is the

4620:   level number).

4621: 

4622: On the other hand, the field $\C (\Si)$ of meromorphic functions

4623:  on $\Si$  has the natural embedding:

4624: \begin{equation}

4625: i \colon sl(2, \C) \otimes \C (\Si) \to \ov{sl_2^{t}}.

4626: \end{equation}

4627: This field $\C (\Si)$ contains the subring $\C (\Si)_p$ of

4628: regular functions in $\Si - p$. The structure of algebra on

4629: \begin{equation}

4630: sl(2, \C) \otimes \C (\Si)_p

4631: \end{equation}

4632: is given by the formula

4633: \begin{equation}

4634: [X \otimes f, Y \otimes g] = [X, Y] \otimes fg

4635: \end{equation}

4636: The restriction of the embedding (4.4) to this subring realizes it as the

4637: subalgebra of $\ov{sl_2^t} $.

4638: 

4639: Thus the algebra $sl(2, \C) \otimes \C (\Si)_p$ acts on the

4640: representation $\sH^k$

4641:  described before.

4642: 

4643:  \begin{dfn} The space of conformal blocks $V_k$ of level $k$

4644:  is the subspace of the dual

4645: $(\sH^k)^*$ annihilated by the Lie algebra

4646:  $sl(2, \C) \otimes \C (\Si)_p$.

4647: \end{dfn}

4648: 

4649: Why such space is related to vector bundles on $\Si$? First of

4650: all, we note, that every vector bundle $E$ on $\Si - p$ is

4651: trivial and of course it is trivial on a

4652:  small disc $D_p$ containing $p$. Fixing

4653: these trivializations we send a vector bundle $E$ to the Laurent

4654: series

4655: \begin{equation}

4656: \ga_E \in SL(2, \C)\otimes \C[t, t^{-1}].

4657: \end{equation}

4658: To get rid of the trivializations, we have to mod out by the

4659: automorphisms groups of trivial bundles on $D_p$ and $\Si - p$.

4660: So the set of vector bundles is

4661: \begin{equation}

4662: \{ E \} =   SL(2, \C) \otimes \C (\Si)_p \backslash  \fsl(2,

4663: \C)\otimes \C[t, t^{-1}] / SL(2, \C)\otimes \C[t].

4664: \end{equation}

4665: Remark that we don't discuss the stability condition and our

4666:  double coset space is so called {\it algebraic stack},

4667:   not a moduli space.

4668: 

4669: In any case our, stack $\{ E \}$ is the quotient of the universal

4670: homogeneous space

4671: \begin{equation}

4672: Q = SL(2, \C) \otimes \C[t, t^{-1}] / SL(2, \C)\otimes \C[t]

4673: \end{equation}

4674: and the line bundle $\Oh( \Theta)$ can be lifted to the universal

4675: line bundle $L_\chi$ on this homogeneous space. The theorem of

4676: Kumar and Mathueu provides the isomorphism

4677: \begin{equation}

4678: H^0(Q, L_\chi^k) = \sH^k

4679: \end{equation}

4680: and $H^0(M^{ss}, Oh(k \Theta))$ coincides with the subspace of

4681:  $(\sH^k)^*$ invariant under Lie algebra

4682:   $sl(2, \C) \otimes \C (\Si)_p$ action.

4683: 

4684: To see that this space is finite dimensional over $\C$, we need

4685: only the standard experience with matrix divisors (for example,

4686: like in the survey \cite{T2}). But to compute the dimension we

4687: have to use more sophisticated versions of all objects.

4688: 

4689: First of all, on our Riemann surface $\Si$ we choose a finite set

4690: of different points

4691: \begin{equation}

4692:  p_1, p_2, ....., p_n \subset \Si

4693: \end{equation}

4694: different from our fixed point $p \in \Si$ and labeled by the set of

4695: irreducible representations of $sl(2, \C)$, that is, by positive

4696:  integers $n_1, n_2, ...., n_n$ (twice spin-numbers).

4697: 

4698: Now our algebra $sl(2, \C) \otimes \C (\Si)_p$ acts on the space

4699: $V_{n_i}$ of the irreducible representation $n_i$ by the formula

4700: \begin{equation}

4701: (X \otimes f) v  = f(p_i) X v

4702: \end{equation}

4703: where $p_i$ is the point corresponding to the representation

4704: $n_i$ and $v \in V_{n_i}$.

4705: 

4706: New vector space is given now by the formula

4707: \begin{equation}

4708: V_k (\{ (p_i, n_i) \})_\Si = Hom_{sl(2, \C) \otimes \C (\Si)_p}(

4709: \sH^k, \otimes_{i=1}^n V_{n_i}).

4710: \end{equation}

4711: 

4712: These spaces have a number of good properties.

4713:  For example, they don't depend on moduli of curves, the choice

4714:   of points and so on. But the most important property is the

4715:   behavior with respect to a simplest degeneration of

4716:    curve $\Si$ to  curve $\Si_0$ with one double point.

4717:    Geometrically this means that we have a smooth curve $\Si_0$

4718:    of genus $g-1$ with a fixed pair of points $(q_+, q_-)$.

4719:    Then

4720: \begin{equation}

4721: V_k (\{ (p_i, n_i) \})_\Si = \oplus_{n \in \Z^+} V_k (\{ (p_i,

4722: n_i), (q_+, n), (q_-, n) \})_{\Si_0}.

4723: \end{equation}

4724: These properties provide the induction by genus (just like formula (3.107)

4725: ).

4726: 

4727: Now we are interested only in ranks of previous spaces.

4728:  We may forget about curves, points and others geometrical

4729:   features (how we did after (3.109) in subsection 3.7) and consider the monoid $M$ of formal sums

4730: \begin{equation}

4731: N = n_1 + n_2 + ... + n_n \in M

4732: \end{equation}

4733: providing the  present labeling of points. Every rank of space $V_k (\{ (p_i,

4734: n_i) \})_\Si$ can be done as a symbol

4735: \begin{equation}

4736: rk_g(N) , \quad \text{where}\quad  N =  n_1 + n_2 + ... + n_n \in

4737: M.

4738: \end{equation}

4739: The list of properties of these numbers is the following:

4740: \begin{enumerate}

4741: \item $rk_0(0) = 1;$

4742: \item $rk_g(N) = \sum_{n \in \Z^+}rk_{g-1}(N +n +n)$;

4743: \item let $g = g' + g''$, then

4744: \begin{equation}

4745: rk_g(N' + N'') = \sum_{n \in \Z^+} rk_{g'} (N' + n)\cdot

4746: rk_{g''}(N'' + n).

4747: \end{equation}

4748: \end{enumerate}

4749: 

4750: Using these equalities, we can decent genus $g$ to 0 and  get so

4751: called {\it fusion rules}:

4752: \begin{enumerate}

4753: \item $rk_0(0) = 1$;

4754: \item $rk_0(N' + N'') = \sum_{ n \in \Z^+} rk_0(N' + n) rk_0(N'' + n)$.

4755: \end{enumerate}

4756: 

4757: These fusion rules give the structure of commutative ring on our

4758: monoid $M$:

4759: \begin{equation}

4760:  N \cdot N' = \sum_{ n \in \Z^+} rk_0(N + N' + n) \cdot n.

4761: \end{equation}

4762: 

4763: This commutative ring is called the  {\it fusion ring}.

4764: 

4765: \begin{rmk} We don't prove here the factorisation rules (5.17) in the algebraic

4766: geometrical set-up, that is, in the set-up of vector bundles. We

4767: will get them as bypass results at the end of our long story. But

4768: remark that there are exist approaches to prove these rules in

4769: purely geometrical way.

4770: \end{rmk}

4771: 

4772: We saw that our fusion ring is very close to be  the ring of

4773: irreducible representations $R(sl(2, \C))$ where the multiplication

4774: is  the tensor product of representations. But there is

4775:  a small difference coming from level $k$ condition.

4776: 

4777:  According to the Clebsh-Gordan rules, a tensor product of

4778:   irreducible representation has the decomposition:

4779: \begin{equation}

4780: n_1 \otimes n_2 = (n_1 + n_2) \oplus  (n_1 + n_2 - 2) \oplus ....

4781: \oplus (n_1 - n_2)

4782: \end{equation}

4783: (here $n_1 \geq n_2$). But the level condition is $n \leq k$ and

4784:  the multiplication is given by the formulas:

4785: \begin{enumerate}

4786: \item $ n_1 + n_2 \leq k \implies n_1 \circ n_2 = n_1 \otimes n_2$;

4787: \item $n_1 + n_2 \geq k \implies n_1 \circ n_2 = (2k - n_1 - n_2) \oplus ... \oplus (n_1-n_2)$.

4788: \end{enumerate}

4789: From this it is easy to see that the fusion ring $M_k$ of level

4790:  $k$ is the quotient of

4791:  $R(sl(2, \C))$ by the ideal generated by $(k+1)$.

4792: 

4793: The fusion ring $M_k$ has the element

4794: \begin{equation}

4795: c = \sum_{n=0}^k n \circ n.

4796: \end{equation}

4797: 

4798: Let $Spec M_k$ be the set of characters (homomorphisms to $\C$)

4799: of $M_k$. Then after elementary computations we get the equality

4800: \begin{equation}

4801: rk_g(\emptyset) = \sum_{\chi \in Spec M_k} \chi(c)^{g-1}

4802: \end{equation}

4803: and

4804: \begin{equation}

4805: \chi(c) = \sum_{n=0}^k \vert \chi(n) \vert^2.

4806: \end{equation}

4807: 

4808: Recall that every character of the representations ring $R(sl(2,

4809: \C))$ is given by $z \in \C$ and

4810: \begin{equation}

4811: \chi_z(n) = \frac{sin((n+1)z)}{sin z}.

4812: \end{equation}

4813: This character is from $Spec M_k$, that is, it vanishes on $(k+1)$

4814: iff

4815: \begin{equation}

4816: z = \frac{n \pi}{k+2}, \quad (1 \leq n \leq k+1).

4817: \end{equation}

4818: Now elementary computations and formula (8.35) give

4819: \begin{equation}

4820: rk_g(\emptyset) = rk V_k =\frac{(k+2)^{g-1}}{2^{g-1}}

4821: \sum_{n=1}^{k+1}

4822:  \frac{1}{(\sin(\frac{n\pi}{k+2}))^{2g-2}}

4823: \end{equation}

4824:  which is  the Verlinde number again.

4825: 

4826:  In the conformal blocks set up  very important observation is

4827:   the absence of Hermitian structure on all vector spaces of this

4828:   theory. Indeed we are working always with "sl" objects, not

4829:    with "su". This feature is distiguishing from the quantization

4830:    set up where Hermitian structure  appears manifestly.

4831: 

4832: \subsection{WZW-connection}

4833: 

4834: On the moduli space $\sM_{g,1}$ of  curves of genus $g$ with a fixed

4835: (pinched) point  we

4836: have vector bundles (1.29) interpreted as vector bundles of

4837: conformal blocks. This identification induces another interpretation of the

4838: Hitchin's  projective flat connection coming from CQFT (see

4839: \cite{TK} or  beautiful survey  \cite{S}).

4840: 

4841: The construction and investigation of the connection in these

4842: terms are almost parallel to the constructions described in

4843: subsection 3.8.

4844:  The difference is following: instead of the moduli space of

4845: Higgs bundles (which is nothing else as the space of $SL(2,

4846: \C)$-representations of the fundamental group of  Riemann

4847: surface) we consider the infinite dimensional space $Q$ (5.9) and

4848: lift

4849:  all of our problems to it. Actually before that we have to develop a

4850:  technique of working with

4851:  infinite dimensional objects of such type rigorously. This was done

4852: in \cite{BL}. Saveing the time-space here we describe this

4853: approach only briefly.

4854: 

4855: Recall that the ind-homogeneous ("ind" is a term from \cite{BL})

4856:  space $Q$ (5.9) describes

4857: $\SU(2)$-vector bundles trivialized on $\Si - p$ where $(\Si, p)

4858: \in \sM_{g, 1}$. Of course we consider the dense open subset

4859: \begin{equation}

4860: Q_0 \subset Q

4861: \end{equation}

4862: corresponding to  semi-stable bundles. For this space the

4863: following fact was proved in \cite{DS}: the natural map

4864: \begin{equation}

4865: f \colon    Q_0 \to M^{ss}(\Si)

4866: \end{equation}

4867: is a locally trivial torsor in the etale topology. (The proof of

4868: this fact is contained also in  unpublished preprint of Beilinson

4869: and Kazhdan.)

4870: 

4871: In this set up Yves Laszlo constructed a family of second order

4872: differential operators $\theta (\dot I)$ over the family of

4873: deformations of complex structures. This construction commutes

4874: with the construction given by the composite morphisms $\phi_s$

4875: (1.86) with the cohomology sequence of the exact triple  (1.81)

4876: which gives

4877: \begin{equation}

4878: H^1(\Si, T \Si) \to H^0(S^2(TM^{ss}(\Si))) \to H_h^1(d_s).

4879: \end{equation}

4880: This lifting is, in a  sense, a lifting of the Sugawara tensor

4881: $T(\dot I)$ (modulo the ind-scheme technique). Namely for every

4882: function $f$ on $\sM_{g,1}$ and a local section $\tilde{e}$ of

4883: the family of dual to (5.10) spaces, the WZW-connection is given

4884: by the formula

4885: \begin{equation}

4886: \nabla ( f \otimes \tilde{s}) = \dot I(f) \cdot \tilde{s} +

4887: \frac{1}{2k-4} \cdot T(\dot I) \tilde{s} \otimes f \quad mod

4888: (\tilde{s} \otimes f).

4889: \end{equation}

4890: From this point the equality of both  connections is obvious

4891: because the only non trivial term in the previous formula is the

4892: Sugawara term just as in the formula (3.152). Thus the Hitchin's

4893: connection is the WZW-connection indeed.

4894: 

4895: 

4896: \subsection{Monodromy representations}

4897: 

4898: 

4899: All of our projective and vector connections are on vector bundles

4900: over the Teichmuller space $\tau_g$. The modular

4901:  group

4902: \begin{equation}

4903: Mod_g = Diff^+ \Si / Diff_0 \Si

4904: \end{equation}

4905:  acts on the Teihmuller space $\tau_g$ such that

4906: \begin{equation}

4907:  \sM_g = \tau_g / Mod_g.

4908: \end{equation}

4909: Thus all of them define  projective or vector representation of

4910: the modular group $Mod_g$

4911: \begin{equation}

4912: \rho \colon Mod_g \to PGL(N_k, \C)

4913: \end{equation}

4914: where $N_k$ is rank of the vector bundle $V_k$.

4915: 

4916: For example, for the classical theory of theta functions we have

4917: two vector representations

4918: \begin{equation}

4919: Mon_w^k \colon Mod_g \to PGL(k^g, \C)

4920: \end{equation}

4921: given by the  Welters projective flat connection, that is, by

4922: applying the full procedure of extended Kodaira-Spencer theory

4923: (see \cite{W}). On the other hand, the symplectic theory

4924:  yields another combinatorial connection (see \cite{K1}

4925:  and another monodromy representation

4926: \begin{equation}

4927: Mon_{BS}^k \colon Mod_g \to PGL(k^g, \C).

4928: \end{equation}

4929: 

4930: Fortunately, the perfect quantization identifies both of these

4931: connections and representations to give unique abelian

4932: representation

4933: \begin{equation}

4934: Mon_w^k = Mon_{BS}^k = Mon^k_a.

4935: \end{equation}

4936: Moreover,  the monodromy group of this representation is finite

4937: \begin{equation}

4938: Mon_k^a = Sp(2g, \Z_{2k}).

4939: \end{equation}

4940: 

4941: For non-abelian case we have the representation

4942: \begin{equation}

4943: Mon_H^k \colon Mod_g \to PGL(rk_g(\emptyset),\C)

4944: \end{equation}

4945:  given by the Hitchin connection.

4946: 

4947: On the other hand, using identifications from \cite{TUY} and \cite{TK}

4948:  Kohno constructed this

4949: representation in the pure combinatorial way

4950: \begin{equation}

4951: Mon_c^k \colon Mod_g \to PGL(rk_g(\emptyset),\C).

4952: \end{equation}

4953:  (The subscript

4954: $"c"$ here stands for "combinatorial".) The lines decompositions (4.39)

4955: of fibers

4956:  is parallel to the decomposition of the highest

4957: weight representation of the affine Lie algebra of $\fsl(2,\C)$

4958: by eigenspaces of the operator $L_0$ from Sugawara construction

4959: of the representation of the Virasoro Lie algebra (see

4960: \cite{K2}). Manifestly this representation is projective

4961: Hermitian. It  was investigated by Macsbaum and Funar. The main

4962: result is the following: the monodromy group of this

4963: representation

4964:  is infinite if $k \neq 2,3,4,6$ and $k \neq  10$ for $g=2$.

4965: 

4966: Moreover, in paper \cite{An} J. Andersen proved {\it asymptotic 

4967: faithfulness} of the collection of representations that is the intersection

4968:  of kernels

4969: $$

4970: \bigcap_{k=1}^\infty ker Mon^k_H = 1 \text{ or } \Z_2 \text{ for } g = 2.

4971: $$

4972: (Of course this result is proved for any rank of vector bundles.)

4973: More precisely he proved that that Berezin-Toeplitz endomorphisms associated to smooth functions on $M^{ss}$ are asymmptotically flat.

4974: 

4975: The appearance of  the Sugawara construction in the procedure of

4976:  definition of this connection suggested the equivalence of

4977: these representations. It was proved using the

4978:  Borthwick-Paul-Uribe method in

4979:  \cite{T3}. This method is quite different from CQFT and it will be

4980:   productive to use generalized Sugawara construction \cite{Sc}

4981:  to get this coincidence directly

4982:  as the generalization of Kniznik-Zamolodchikov construction.

4983: 

4984: 

4985: 

4986: 

4987: \section{3-valent graphs}

4988: 

4989: \subsection{Spin networks}

4990: 

4991:  Friedan and Shenker \cite{FS}, Moor and Seiberg

4992: \cite{MS} proposed a new approach to CFT - the {\it axiomatic

4993: description of a rational

4994:  conformal field theory}. We will call it the MS-axiomatic. The main ingredient

4995:  of this theory is 3-valent graphs which  appeared in constructions of

4996:  section 4  already. We saw that every 3-valent graph is equivalent

4997:  to a trinion-decomposed Riemann surface (see (4.9)-(4.11)).

4998: 

4999: 

5000: In a similar way, we can extend these constructions to an

5001: orientable surface $\Si$ of genus $g$ with $n$ holes.

5002: Geometrically such surface with a trinion decomposition is

5003: equivalent to its dual 3-1-valent graph $\Ga$ or a history

5004: (3-valent graph of

5005:  genus $g$ with $n$  parabolic edges which we describe in the special

5006:  subsection).

5007: In this case admissible integer weights are defined in the same

5008: way as in (4.25)

5009:  and (4.26) but the set of weights

5010:  \begin{equation}

5011:  W^k(\Ga, v_1, \dots, v_n)

5012:  \end{equation}

5013: is the set of functions $w$ (4.25)-(4.26) satisfying the

5014: conditions

5015:  \begin{equation}

5016:  w(a_i) = v_i

5017:  \end{equation}

5018: for every parabolic edge $a_i, \quad 1 \leq i \leq n$, that is, the

5019: values of these functions on parabolic edges are  fixed constants.

5020: So in the same vein we have spaces

5021:  \begin{equation}

5022: \sH^{k}_{\Ga, v_1, \dots, v_n} = \oplus_{w \in W^{k}(\Ga, v_1,

5023: \dots, v_n)} \C \cdot w

5024: \end{equation}

5025:  The {\it

5026: Clebsch--Gordan conditions} (1), (2) and (3) from (4.26) provide a  very

5027:  surprising

5028: point of all these constructions: namely, let us multiply all our integer

5029:  weights $w \in W^{k}(\Ga)$ by level $k$. We get the maps

5030: \begin{equation}

5031:  k \cdot w = j \colon E(\Ga) \to 1/2 \Z_{\geq 0}.

5032:  \end{equation}

5033: But we can consider the target set  as the set of irreducible

5034: representations of $\SU(2)$:

5035: \begin{equation}

5036: \widehat{\SU(2)} = 1/2 \Z_{\geq 0}.

5037: \end{equation}

5038: So every integer weight  can be considered as a coloring of edges by

5039: irreducible representations of $\SU(2)$. The conditions (1), (2)

5040: and (3) (4.26) give that for every triple $e_{1}, e_{2}, e_{3}$

5041: via one vertex $v$ the tensor product of representations

5042: \begin{equation}

5043: 0 \in j(e_{1}) \otimes j(e_{2}) \otimes j(e_{3})

5044: \end{equation}

5045: that is this tensor product contains the trivial representation.

5046: Or the tensor product $j(e_{1}) \otimes j(e_{2})$ admits an

5047: intertwiner

5048: \begin{equation}

5049: i_{v} \colon j(e_{1}) \otimes j(e_{2}) \to j(e_{3})

5050: \end{equation}

5051: as  $\SU(2)$-equivariant homomorphism.

5052: 

5053: Such coloring of graph edges is called  $\SU(2)$-{\it spin

5054: network} by Penrose \cite{P}( see also \cite{T1}). Thus the set

5055: $W^{k}(\Ga)$ is the set of spin networks of level $k$ over a graph

5056: $\Ga$. We will see in the following section that for every

5057: integer weight $w \in

5058:  W^{k}(\Ga)$ the function

5059: \begin{equation}

5060:  2k \cdot w = t \colon E(\Ga) \to  \Z_{\geq 0}.

5061:  \end{equation}

5062: is the type of a loop  on our graph $\Ga$. It means, that   a

5063: colouring can be defined purely in terms of the geometry of graph

5064: as a {\it topological space}.

5065: 

5066: The main fact here is the identification of the set $SNW^k(\Ga)$

5067: of $\SU(2)$-spin networks  of level $k$

5068: $$

5069: SNW^k_\Ga = W^k(\Ga)

5070: $$

5071: and the set of integer weights $W^k(\Ga)$.

5072: 

5073: Our first task  is

5074: to compare spaces $\sH^{k}_{\Ga}$

5075:  for different graphs of genus $g$.

5076: As a result of this comparing we  will construct, following Kohno

5077: \cite{K1}, the holomorphic vector bundle with a holomorphic

5078: projective flat connection over the moduli space $\sM_{g}$ of

5079: Riemann surfaces of genus $g$. This construction is purely

5080: topological. So first of all we have to construct combinatorial

5081: models of the Teichmuller space and the trivalent graphs complex.

5082: These ideas and constructions came from low dimensional topology

5083: and classical and quantum Conformal Field Theories.

5084: 

5085: \begin{figure}[tbn]

5086: \centerline{\epsfxsize=3in\epsfbox{30c.eps}} \caption{\sl Pumping

5087: up trick} \label{Fig 2}

5088: \end{figure}

5089: 

5090: 

5091: \subsection{3-dimensional topology}

5092: 

5093: 

5094:  Recall again that any 3-valent graph is equivalent to  a trinion decomposition

5095:  of a Riemann surface. It is easy to see that such surface is a boundary

5096:   of uniquely determined  handlebody.

5097:    We explain this fact more carefully.

5098: 

5099: Returning to  a maximal set of disjoint inequivalent curves $ \{

5100: C_1, ... , C_{3g-3} \} $ (4.7) and removing these curves we get

5101: a finite set of trinions $ \Si - \{ C_1, ... , C_{3g-3} \} =

5102: \cup_{i=1}^{2g-2} \tilde{v_i} $ (4.8) (see Fig.2).

5103: 

5104: Every oriented trinion defines the handlebody inside  it and

5105: gluing (sewing) these handlebodies together we get the handlebody

5106: $H_\Ga$ such that

5107: \begin{equation}

5108: \p H_\Ga = \Si_\Ga

5109:  \end{equation}

5110:  Any diffeomorphism $h \in Mod_g$ (where $Mod_g$ is the modular group (5.30))

5111:   gives us a compact  3-manifold

5112:  \begin{equation}

5113: X_{\Ga} = H_\Ga \cup_h  H_\Ga

5114:  \end{equation}

5115: where we are gluing two handlebodies with the boundaries by the

5116: diffeomorphism

5117:  \begin{equation}

5118: h \colon \p H_\Ga = \Si_\Ga \to \Si_\Ga = \p H_\Ga.

5119:  \end{equation}

5120:  For the 3-manifold such presentation is called a {\it Heegaard splitting}.

5121:  The subgroup

5122:  \begin{equation}

5123: M_\Ga \subset M_g

5124:  \end{equation}

5125: of diffeomorphisms which can be extended to diffeomorphisms of

5126: the handlebody $H_\Ga$, acts on $Mod_g$ on the left and right and we

5127: have the set of 3-manifolds

5128: \begin{equation}

5129: \{ X_\Ga \} = M_\Ga \backslash Mod_g \slash M_\Ga

5130:  \end{equation}

5131: as the set of double cosets.

5132: 

5133:  In  Heegard splitting (6.10) let us add a new handle to the surface and extend

5134:  the diffeomorphism $g$ by the identity on this new handle. This procedure

5135:   doesn't

5136:  change our 3-manifold and is called  {\it stabilization}.

5137:  Singer proved in \cite{Si} the following statement

5138: 

5139:  \begin{prop} The equivalence relation between elements $h \in Mod_g$ when $g \to

5140:  \infty$ creating homeoomorphic 3-manifolds, is generated by isotopies, left

5141:  and right

5142:  actions of $M_\Ga$ and a stabilization.

5143:  \end{prop}

5144: 

5145:  To construct an invariant of 3-manifolds we have to send a marked

5146:  Riemann surface $\Si_\Ga$ to the wave function vector space $\sH^k_\Ga$

5147:   (4.43) with fixed vector $w_{H_\Ga}$ given by the function (4.25) such that

5148:   \begin{equation}

5149: w_{H_\Ga}(e) = 0 \quad \text{for every} \quad e \in E(\Ga).

5150: \end{equation}

5151: Here

5152: \begin{equation}

5153: w_{H_\Ga} \in \sH^k_\Ga

5154: \end{equation}

5155: is called the {\it vacuum vector} for the handlebody $H_\Ga$.

5156: 

5157:  The space $\sH^k_\Ga$ admits the standard Hermitian structure which identifies

5158:  this space with the dual.

5159: 

5160:  Now if we have a linear representation

5161:  \begin{equation}

5162: \rho \colon Mod_g \to GL (\sH^k_\Ga)

5163: \end{equation}

5164:  then

5165:  \begin{equation}

5166: I(X_{\Ga, g}) = \frac{< w_{H_\Ga}, \rho(g)(w_{H_\Ga})>}{\Vert

5167: w_{H_\Ga} \Vert^2}

5168: \end{equation}

5169: has to be an invariant of $X_{\Ga}$ as 3-manifold since it

5170: easy to see that

5171: \begin{enumerate}

5172: \item the vector $w_{H_\Ga}$ is invariant with respect to $M_\Ga$-action and

5173: \item under normalization (6.17) this vector is invariant under stabilization.

5174: \end{enumerate}

5175: 

5176: But the situation is a slightly more complicated then just

5177: described. Namely, Kohno constructed nonlinear representation

5178: $\rho$ but made it slightly projective,

5179:  that is, determined up to the usual phase amplitude

5180:  \begin{equation}

5181: e^{\frac{\pi i \cdot k}{4 (k+2)} }.

5182: \end{equation}

5183: Thus the invariant (6.17) is determined up to the same phase

5184: amplitude.

5185: 

5186: To construct such slightly projective representation (6.16) one has to

5187: treat combinatorial constructions in the next special section.

5188: 

5189: 

5190: 

5191: \subsection{Geometry of graphs}

5192: 

5193: 

5194: 

5195: 

5196: 

5197:  A  graph is  one

5198: dimensional complex containing edges -

5199:  cells of one dimensional skeleton and vertices -cells of

5200:  zero dimensional

5201:  skeleton.

5202: 

5203: The combinatorial invariant of a pinched Riemann surfaces with

5204:  a trinion decomposition

5205: is a (3-1)-valent graphs or a {\it history}. Such graphs have

5206: 3-valent and 1-valent vertices. The last vertices are half

5207: intervals with one vertex. Vertices of such edges are called

5208: {\it parabolic}.

5209: 

5210:   Graphs and histories are Feynman diagrams - objects of the standard

5211:   formalism

5212:  in fields (for 3-valent graph case, $\phi^3$ - theories).

5213: 

5214: \begin{figure}[tbn]

5215: \centerline{\epsfxsize=3in\epsfbox{2c.eps}} \caption{\sl Standard

5216: Feynman diagrams} \label{Fig 2}

5217: \end{figure}

5218: 

5219: 

5220: In classical and quantum field theory a 3-valent graph without

5221: parabolic vertices is called a vacuum-polarization graph.  So we will

5222: consider such

5223: graph as a geometrical  object. Note that

5224:  an closed 1-valent graph is a disjoint finite union of compact

5225:   intervals and

5226:  a closed 2-valent graph is a disjoint finite union of compact

5227:  circles.

5228: 

5229: Every graph $\Ga$ has the following ingredients:

5230:  \begin{enumerate}

5231: \item $E(\Ga)$ is the set of edges;

5232:  \item  $V( \Ga)$ is the set of vertices,

5233:  \item  $S(v)$ is the star of a vertex $v$ that is the set of local edges incident  to  $v$,

5234:  \item   $F(\Ga)=\{v\in e \}$ is the set of flags (edge + vertex).

5235:  It is the same as

5236: the set of oriented edges $\vec E(\Ga)$,

5237: \item   $L( \Ga)$ is the set of circles, that is, of edges with only one vertex.

5238: \end{enumerate}

5239: We will consider homogeneous graphs that is the graphs of the same valence $d$ at every vertex.

5240: 

5241: For example for 3-valent graphs two projections   $p_e \colon F(\Ga)\to E(\Ga)$ and

5242:  $v \colon F(\Ga)\to V(\Ga)$ are finite covers

5243:  of degrees

5244:   2 and 3 with the same ramification locus

5245:   $W_e = W_v \subset F(\Ga)$, containing pairs

5246:   $v\in e$, where   $e$ is a loop:

5247: \begin{equation}

5248: R_e = L( \Ga) \subset E( \Ga) ; \quad R_v = L( \Ga) \subset V(

5249: \Ga)

5250: \end{equation}

5251: Let $\vert\ \vert$ be the cardinality of a finite set, then

5252:  \begin{equation}

5253: 2 \cdot \vert E(\Ga) \vert-\vert L(\Ga) \vert=\vert F(\Ga) \vert=

5254: 3 \cdot \vert V(\Ga) \vert-\vert L(\Ga)\vert.

5255:  \end{equation}

5256: 

5257:  Thus $ \vert V(\Ga) \vert=2g-2$ and $ \vert E(\Ga)

5258: \vert=3g-3$, where $g>1$ is an integer number called  {\it genus}

5259: of

5260:  $\Ga$.

5261: 

5262: There exists only finite set $TG_{g}$ of 3-valent graphs of genus

5263: $g$. But up to now nobody knows how many graphs it contains

5264: \begin{equation}

5265: \vert TG_{g} \vert = ?

5266: \end{equation}

5267: (besides of small $g = 2, 3,\dots, 11 $).

5268: 

5269: 

5270: Consider any graph $\Ga$ with vertices of any valence $ > 2$. Let $e \in E(\Ga)$ be 

5271: any edge with distinct two vertices $ \p e = v_1 \bigcup v_2$ and $v_1 \neq v_2$. Remove the edge $e$ and identify vertices $v_1$ and $v_2$ to get a single new vertex

5272:  $v_{new}$. We get a new graph $\Ga'$ which is called  {\it contraction} of $\Ga$

5273:  along the edge. Such contraction decreas $\vert E(\Ga) \vert$ and   $\vert V(\Ga) \vert$ by one. The minimal under this procedure graph $\Ga_{min}$ has only one vertex.  Obviously every graph can be obtained from a 3-valent graph by applying a chain of contractions. The inverse operation is {\it expansion} which is the blow up of a vertex $v$ with  $\vert S(v) \vert > 3$ to a new edge $e_{new}$ with

5274:  two new vertices $v_{new}, v'_{new} = \p e_{new}$ and partitions

5275: $$

5276: (S(v_{new} - e_{new}) \bigcup  ( S(v'_{new}) - e_{new}) = S(v).

5277: $$

5278:  The maximal under this procedure graph is a trivalent graph $\Ga_{max}$.

5279: 

5280: Every 3-valent graph $\Ga$ has {\it closed relatives}: let $v, v'

5281: $ be vertices of one edge $e$ and $e_{1}, e_{2}, e$ are edges via

5282: $v$ and $e'_{1}, e'_{2}, e$ are edges via $v'$. Let us contract

5283: the edge $e$ to obtain 4-valent vertex $v_{new}$. For this vertex the star $S(v_{new}) = e_{1} \bigcup e_{2} \bigcup e'_{1}

5284: \bigcup e'_{2}$. Now divide this quadruple in two pairs, for

5285: example, $(e_{1}, e'_{1})$ and $(e_{2}, e'_{2})$. Then  extend

5286: the vertex $v_{new}$ to new edge $e_{new}$ such a way that we get

5287: \begin{enumerate}

5288: \item

5289: two new vertices $v_{1}$ and

5290:  $v_{2}$ such that $\p e_{new} = v_{1}, v_{2}$,

5291: \item

5292:  with stars  $S(v_{1}) = e_{1}, e'_{1}, e_{new}$ and

5293:  $S(v_{2}) = e_{2}, e'_{2}, e_{new}$.

5294:  \end{enumerate}

5295:  We  get a new

5296:    graph $\Ga'_{e}$. This is the composition of one contraction and one expansion

5297: preserving valence of the graph.

5298: 

5299:    \begin{dfn} The transformation of graphs

5300:    $$

5301:    \Ga \to \Ga'_{e}

5302:    $$

5303: is called  {\it elementary transformation}.

5304:    \end{dfn}

5305: (We will see later that this transformation lies under {\it elementary move for  trinion decompositions.})

5306: 

5307: Sending the edge $e$ of $\Ga$ to the new edge $e_{new}$ of $\Ga'_e$ we get the identification 

5308: \begin{equation}

5309:   \na \colon E( \Ga) \to E(\Ga'_{e}).

5310:    \end{equation}

5311: 

5312: 

5313: 

5314: 

5315: \begin{figure}[tbn]

5316: \centerline{\epsfxsize=3in\epsfbox{1c.eps}} \caption{\sl

5317: Elementary transformations} \label{Fig 11}

5318: \end{figure}

5319: 

5320: 

5321: So, around any 3-valent graph there exists the collection of

5322: 3-valent graphs which differ from $\Ga$ by  elementary transformation. (This triple corresponds to the triple of vertices of the triangle of Fig. 13.)

5323: 

5324: By the induction on $g$ it is easy to prove that any 3-valent

5325:  graph $\Ga$ of genus $g$ can be transformed to any other

5326:  3-valent graph $\Ga'$ by a finite chain of elementary transformations.

5327:  Thus the elementary transformations act on $TG_{g}$  transitively.

5328: 

5329: In the same vein we can consider the composition of two contractions and two

5330:  expansions preserving valence of verticies and so on. 

5331: 

5332: Now for any graph $\Ga$ let 

5333: $$

5334: \sE(\Ga) = \{ \Ga_{max} \}

5335: $$

5336: be the set of all expansions of $\Ga$ to a trivalent graph. This set of 3-valent graphs we call  {\it nest} of 3-valent graphs.

5337: 

5338: Of course we can consider compositions of expansions for every vertex $v \in V(\Ga)$ with $\vert S(v) \vert > 3$ independentely. So for every such vertex

5339: the set of expansions $\sE(v)$ is in 1-1 correspondence with the set  of 3-valent trees $T$ with parabolic edges identifyed with the set $S(v)$:

5340: $$

5341: \p T = S(v).

5342: $$

5343: Thus 

5344: $$

5345: \vert \sE(v) \vert = \sum_{T \vert \p T = S(v)} \frac{1}{\vert Aut T \vert}.

5346: $$

5347: 

5348:  So for a graph $\Ga$ with one 

5349: 4-valent vertex 

5350: $\vert \sE(\Ga) \vert = 3 $

5351:  and so on.

5352: 

5353: The combinatoric behind of these constructions can be described perfectly  by the standard pairing scheme in the {\it Feynman diagram technique}.

5354: This technique  was applyed  as well to so called

5355: {\it ribbon graphs}. Recall that a ribbon graph structure on a graph is  cyclic

5356:  orders choosen  on all stars of edges. Geometricaly it is equaivalent to an embedding of a graph into an oriented surface. Such cyclic orders are required by applications to  matrix models (see \cite{MP} and \cite{Mu}). 

5357: 

5358: For such graphs contractions and expansions preserve the ribbon structures. For a homogeneous d-valent graph $\Ga$ a ribbon graph structure is equivalent to the choice of a {\it connection} on this graph (see \cite{BGH} for the consideration of graphs as manifolds). 

5359: 

5360: Recall that a path of lenth 1 on  $\Ga$ is just an oriented edge

5361:  $\vec e$. Let the set

5362: $P_1(\Ga) = \vec E(\Ga)$ be the set  of  1-paths on  $\Ga$.

5363: Every such path  $\vec e $ has  vertices of two types

5364: \begin{equation}

5365: v_s(\vec e), \quad v_t(\vec e) \in V(\Ga)

5366: \end{equation}

5367:  {\it source} and  {\it target} which are equal for  loops.

5368: 

5369: A path of length  $d$ in  $\Ga$ is an ordered sequence $(\vec e_1,

5370: ..., \vec e_d)$ of oriented edges such that for every $i$

5371: \begin{equation}

5372: v_t(\vec e_i) = v_s(\vec e_{i+1}).

5373: \end{equation}

5374: Let $e_{d+1} = e_1$, that is, our order is cyclic. A path $(\vec

5375: e_1, ..., \vec e_d)$     is called irreducible if   $e_i \neq

5376: e_{i+1}  $

5377:  for every

5378: $i$ (including $i = d+1 $).

5379: 

5380:  A connection $\na$ on a homogeneous graph $\Ga$ is a collection of

5381: identifications of the stars 

5382: \begin{equation}

5383: \na_{\vec e} \colon S(v_{s}(\vec e)) \to  S(v_{t}(\vec e)) 

5384: \end{equation}

5385: such that 

5386: $$

5387: \na_{\vec e} (e) = e

5388: $$

5389: where $v_{s}, v_{t} = \p \vec e$.

5390: 

5391: 

5392: A path   $(\vec

5393: e_1, ..., \vec e_n)$     is called {\it geodesic} if   

5394: \begin{equation}

5395: e_{i+2} =  \na_{\vec e_{i+1}} (e_i) 

5396: \end{equation}

5397:  for every

5398: $i$ (including $i = n+1 $) when we have closed geodesic.

5399: 

5400: Obviously for every pair of edges from the star of a vertex there exists

5401:  unique  closed geodesic passing through the edges. Hence the set of closed geodesics defines our connection

5402:  uniquely. 

5403: 

5404: In the same vein we can say that a subgraph $\Ga' \subset \Ga$ is totally geodesic if every geodesic starting in $\Ga'$ stay within $\Ga$.

5405: 

5406: Now for every closed geodesic  $l = (\vec

5407: e_1, ..., \vec e_d = \vec e_1)$ we have the set of vertices $V_l$ of the edges in this consequence. For every $v \in V_l$ we have  a monodromy permutation

5408: $$

5409: m_{v} (l) \colon S(v) \to S(v)

5410: $$

5411: So every closed geodesic $l$ defines a conjugation class 

5412: $$

5413: m(l) \in Conj(S_d)

5414: $$

5415: where $S_d$ is the group of permutations with $d$ elements.

5416: 

5417: A closed geodesic is {\it flat} if $m(l) = id$.

5418: 

5419: Let $G_\Ga$ be the set of simple closed irreducible geodesics of $\Ga$. 

5420: Then the map

5421: $$

5422: c_\na \colon G_\Ga \to  Conj(S_d)

5423: $$

5424: is an invariant of  connection $\na$ and the preimage of the unit class is the subset $G_\Ga^F$ of flat geodesics.

5425: Moreover this map is an invariant for gauge classes of connections:

5426: for every $v \in V(\Ga)$ consider the group 

5427: $S_v = Aut S(v) = S_d $

5428: where the last equality is non canonical, of course, and the product

5429: $$

5430: \sG = \prod_{v \in V(\Ga)} S_v.

5431: $$

5432: This group acts on the set $\sA_\Ga$ of all connections on $\Ga$ by the formula:

5433: for every $\vec e$ with $\p \vec e = v_{s}, v_{t}$ 

5434: \begin{equation}

5435: g (\na)_{\vec e} = g_{s} \na_{\vec e} g_{t}^{-1}

5436: \end{equation}

5437: where $g_v$ are components of the decomposition of our discrete gauge group.

5438: 

5439: Then the set

5440: $$

5441: \sB = \sA / \sG

5442: $$

5443: is the set of orbits with respect to  the gauge group  and the function $c_\na$ 

5444: is an invariant of the orbits.

5445: 

5446: Now gluing  2-cell to every simple closed flat geodesic we get a 2-complex

5447: $C_\na(\Ga)$  which is a closed oriented surface corresponding to this ribbon graph.

5448: 

5449: A ribbon graph is called {\it plane} if the surface $C(\Ga) = S^2$ is 2-sphere.

5450: 

5451: 

5452: 

5453: For ribbon graphs the set of expansions is "reducible". In particular we can

5454: choose from any pair of graphs correlated by  an elemetary transformation one

5455:  of them. Recall that in the procedure of expansion we identify two expanded graphs if there is a ribbon graph isomorphism from one to the other preserving all local edges. In particular if $v \in V(\Ga)$ is d-valent vertex then

5456: $\vert \sS(v) \vert = \frac{1}{d-1} C^{2d-4}_{d-2}$ which is the Catalan number.

5457: 

5458: Now forget the ribbon structure:  

5459: let  $P_d(\Ga)$ be the set of irreducible paths of length  $d$ in

5460: $\Ga$. Every path

5461:  $(\vec e_1, ..., \vec e_d) \in P_d(\Ga)$ defines two vertices

5462: \begin{equation}

5463: v_s((\vec e_1, ..., \vec e_d)), \quad v_t((\vec e_1, ..., \vec

5464: e_d)) \in V(\Ga)

5465: \end{equation}

5466: - source and  target and two maps

5467: \begin{equation}

5468: v_s \colon P_d(\Ga) \to V(\Ga)

5469: \end{equation}

5470: $$

5471: v_t \colon P_d(\Ga) \to V(\Ga)

5472: $$

5473: $$

5474: (v_s \times v_t) \colon P_d(\Ga) \to V(\Ga) \times V(\Ga).

5475: $$

5476: The preimage of the intersection

5477: \begin{equation}

5478: L_d(\Ga) = (v_s \times v_t)^{-1} ( V(\Ga)_\Delta )

5479: \end{equation}

5480: is the set of oriented irreducible loops of length   $d$. In

5481: particular

5482: $$

5483: F(L_1)(\Ga)= L(\Ga),

5484: $$

5485: where  $F$ is the projection of a path to the graph.  For every

5486: vertex á

5487:  $v \in V(\Ga)$ we have the set of irreducible loops marked by $v$

5488: \begin{equation}

5489: L_d(\Ga)_v = (v_s \times v_t)^{-1} ( v).

5490: \end{equation}

5491: The union

5492: \begin{equation}

5493: L_\infty(\Ga)_v = \cup_{d=1}^\infty L_d(\Ga)_v

5494: \end{equation}

5495: admits a group structure

5496: \begin{equation}

5497: \pi_1^C(\Ga)_v = L_\infty(\Ga)_v.

5498: \end{equation}

5499: if we consider only irreducible fragments of compositions.

5500: 

5501: Obviously, this group depends on the marking point $v$.

5502: 

5503: Let $\pi_1(\Ga)$ be the standard fundamental group of $\Ga$ (as

5504: 1-complex). Then the natural epimorphism  $r \colon \pi_1^C(\Ga)

5505: \to \pi_1(\Ga)$ is an isomorphism. Obviously, if $\Ga$ is a 3-valent graph 

5506: of genus $g$ then $\pi_1(\Ga) = F_g $

5507: is  free group with $g$ generators.

5508: 

5509: For any 3-valent graph  $  \Ga  $  a choice of  two  edges  $e, e'$  (may be not

5510:    different) defines a new graph  $\Ga_{e, e'}$  by the

5511: following procedure: let us choose a pair of inner points

5512:  $v \in e$ and   $v' \in e'$ and joint them by the edge

5513: $e_{new}$. We get a new  3-valent graph  with  the pair of new

5514: vertices  and the triple of new edges

5515:  $e_{new}, e{1/2} , e_{ 1/2}$, where the last two edges are just halves of the edges

5516:   obtained by dividing  previous edgesá $e, e'$.

5517: 

5518: The inverse  operation is   much more simple  :  we just remove  an

5519: edge and a pair of its vertices.

5520: 

5521: \begin{figure}[tbn]

5522: \centerline{\epsfxsize=3in\epsfbox{3c.eps}} \caption{\sl Two

5523: paths} \label{Fig 3o}

5524: \end{figure}

5525: %\epsfbox{3o.fig}}

5526: If  our  edges  were  in two  paths  $

5527:   e \in p = (\vec e_{1}, ...,

5528: \vec e_{n})$ and  $e' \in p' = (\vec e'_{1}, ..., \vec e'_{m})$,

5529: then on the new

5530:  graph  $\Ga_{e, e'}$ we would have (formally) two paths

5531: $p_{new}, p'_{new}$.

5532: 

5533: It is easy to see  that

5534: \begin{enumerate}

5535: \item $ p \neq p' \implies p_{new} = p'_{new}$;

5536: \item $ p = p' \implies p_{new} \neq p'_{new}$.

5537: \end{enumerate}

5538: 

5539: 

5540: By   induction one obtains:

5541: \begin{enumerate}

5542: \item Any   3-valent  graph  can be created by  a finite  chain  of such

5543: elementary  operations starting with  the disconnected union of circles

5544:  $0 \cup 0 ... \cup 0$;

5545: \item  every path on $\Ga$  can be created by a finite chain  of

5546:   elementary

5547:  operations  starting from a path on a disjoint finite set of circles.

5548: \end{enumerate}

5549: These statements give the induction procedure which allows to prove  all simple

5550: facts about trivalent graphs. There are two types of such problems: {\it Eulerian paths} and {\it coloring problems}.

5551: 

5552:  Recall that a path $p = (\vec e_{1}, ...,

5553: \vec e_{n}) \in P_d(\Ga)$ is called {\it Eulerian} if it contains every oriented edge and exactly one time. A system of paths $P_E$ is called Eulerian if every oriented edge is contained and exactly one time in one and exactly one path from

5554: this system. Such system is called minimal if $\vert P_E  \vert = min$. This is

5555: an invariant of the graph $e(\Ga)$.

5556: 

5557: We know already that an elementary operation tranform an Eulerian system $P_E$

5558: on $\Ga$ to the Eulerian system $P'_E$ on  $\Ga_{e, e'}$. Moreover

5559: $$

5560: \vert P'_E  \vert = \vert P_E  \vert \pm 1.

5561: $$

5562: From this it easy to see that for every 3-valent graph $\Ga$ of genus $g$

5563: \begin{enumerate}

5564: \item $1 \leq e(\Ga) \leq g+1$;

5565: \item $e(\Ga) = g-1 mod  2$;

5566: \item there exists a chain of elemetary operations with result graph $\Ga$ such that $e(\Ga') = 1$.

5567: \end{enumerate}

5568: 

5569: We can consider every plane ribbon graph $\Ga$  as a graph on the plane $\R^2$.

5570: Such embedded graph defines the geographic map with connected componets of the complement as the countries. This is the {\it 4-colors setup}. This problem is stated for embedded

5571:  graphs.

5572: 

5573: Other "coloring" problem for graphs appears in the Feynman diagram technique

5574:  (for the problem of dependence on the form of Lagrangian and of limits 

5575: in matrix models). Let $C = (c_1, ... , c_N)$ 

5576: is a set of colors. Any coloring is just a map

5577: $$

5578: c \colon E(\Ga) \to C

5579: $$

5580: such that for every vertex $v \in V(\Ga)$ the map $c \colon S(v) \to C$ is an

5581: {\it embedding}. That is around every vertex all edges have different colors.

5582: If the number $N$ of colors is minimal then  it is called the {\it chromatic number } $N_{min} = \chi (\Ga)$. Easely we have to consider graphs without 

5583: 1-loops. Then for 3-valent graphs

5584: $$

5585: 3 \leq \chi(\Ga) \leq 4.

5586: $$  

5587: A little more efforts you need to prove the following fact

5588: 

5589: $\chi(\Ga) = 3$ if and only if there is a set $(l_1, ... , l_n)$ of disjoint

5590: simple loops on $\Ga$ such that

5591: \begin{enumerate} 

5592: \item every loop $l_i$ contains an even number of edges;

5593: \item for every vertex $v \in V(\Ga)$ there exists a loop which contains it:

5594: $$

5595: V(\Ga) \subset \bigcup_{i=1}^n l_i.

5596: $$

5597: \end{enumerate}

5598: 

5599: Both of these "coloring" problems are related by the following equivalence:

5600: two  statements 

5601: \begin{enumerate}

5602: \item The 4-colors problem admits  positive solution;

5603: \item For every plane 3-valent graph $\Ga$ without 1-loops $\chi(\Ga) = 3$

5604: \end{enumerate}

5605: are equivalent.

5606: These problems of "classical theory" of graphs. Considering every homogeneous 

5607: graph as a manifold (in style of \cite{BGH}) and following  Segal's ideas we 

5608: have a reason

5609:  to consider the "Loop space" of every graph.

5610: 

5611:  

5612: 

5613: 

5614: 

5615: Consider  standard interval $[0, 1]$ and divide it by a finite

5616: set of points to a complex $[0, p_1, .... , p_{d-1}, 1]$. We can

5617: consider this as a 2-valent history with two parabolic edges $[0,

5618: p_1]$ and $[p_{d-1}, 1]$. The linking of the parabolic edges

5619:  of such history gives  a cycle (or a loop) that is a closed 2-valent graph.

5620: 

5621: As we know every continuous combinatorial map

5622: \begin{equation}

5623: p \colon [0, p_1, .... , p_{d-1}, 1]\to \Ga

5624: \end{equation}

5625:  is called  path of length $d$ in $\Ga$.

5626: 

5627: If this path is irreducible, the preimage of every edge

5628: \begin{equation}

5629:       p^{-1}(e) = g_1 \cup ... \cup g_{col(e)}

5630: \end{equation}

5631: is a disjoint union of a finite set of compact 1-graphs and the

5632: number of components is equal to $col(e)$ (without any orientation).

5633: 

5634:  Now we consider the set $\sL$ of irreducible loops that is closed

5635:  paths (0=1). The type of  loop $l \in \sL $ is   the non negative

5636: function

5637: \begin{equation}

5638: col \colon E(\Ga) \to \Z_{\geq 0}.

5639: \end{equation}

5640: (recall that we  consider irreducible unmarked loops only).

5641: 

5642: For any type of loops for every vertex $v \in V(\Ga)$ with the set

5643: of edges

5644:   $(e_{1,v}, e_{2,v}, e_{3,v})$  the following constraints hold:

5645:  \begin{enumerate}

5646: \item

5647: \begin{equation}

5648:  col (e_{1,v}) + col( e_{2,v}) + col( e_{3,v}) = 0 \mod 2;

5649: \end{equation}

5650: \item

5651: \begin{equation}

5652: |col( e_{1,v})-col(e_{2,v})| \leq col ( e_{3,v}) \leq col(

5653: e_{1,v}) + col(e_{2,v}).

5654: \end{equation}

5655: \end{enumerate}

5656: Indeed, consider a vertex $v$ and the star $S(v) = e_1, e_2, e_3$.

5657: Let $n_{12}$ be the number equal to how many times  we come to $v$ along

5658:  the edge $e_1$ and turn to the edge $e_2$. In the same vein we have

5659: $n_{ij}$ for every pair of edges from the star. Then

5660: \begin{equation}

5661:  col(e_i) = \sum_{j \neq i} n_{ij}.

5662: \end{equation}

5663: From this we  get the statement immediately.

5664: 

5665: 

5666: Now we say that a loop $l \in \sL$ has the type of level $k$ if

5667: for

5668:  every edge $e$

5669: $$

5670:  col (e) \leq k.

5671: $$

5672: In the same vein we can prove that the type function (6.36) gives

5673: a type of

5674:  loop of level $k$  iff

5675:  for every vertex $v \in V(\Ga)$ with the set of local edges

5676:   $(e_{1,v}, e_{2,v}, e_{3,v})$   almost the same constraint

5677:    holds:

5678:  \begin{enumerate}

5679: \item

5680: $$

5681:  col (e_{1,v}) + col( e_{2,v}) + col( e_{3,v}) = 0 mod 2;

5682: $$

5683: \item  and a slightly  sophisticated  condition 2)

5684: \begin{equation}

5685: |col( e_{1,v})-col(e_{2,v})| \leq col ( e_{3,v}) \leq

5686: \end{equation}

5687: $$

5688:  min (col( e_{1,v}) + col(e_{2,v}), 2k -

5689:  col( e_{1,v}) - col(e_{2,v}).

5690: $$

5691: \end{enumerate}

5692: 

5693: Note that the inequality (6.38) is just the triangle inequality

5694: and the inequality

5695:  (6.40) is the {\it spherical} triangle inequality

5696:  (with $\sqrt{k}$ proportional to the radius of the sphere).

5697: 

5698: Note also that it is quite reasonable to call every

5699: loop  as {\it knot}.

5700: 

5701: So every knot defines the type function (6.36)  coloring edges of

5702: our graph in $\Z_{\geq 0}$ colors. {\it Dividing this function by 2 we

5703: get the map}

5704: $$

5705: 1/2 col \colon E(\Ga) \to 1/2 \Z_{\geq 0}  = \widehat{\SU(2)}

5706: $$

5707: {\it where the last set is the set of irreducible representations of} $\SU(2)$. Conditions (6.38) and (6.40) mean that any pair $(\Ga, cal)$ is

5708: a $\SU(2)$- {\it spin network}.

5709: 

5710: Recall that to get a marked Riemann surface from a trivalent graph $\Ga$ we

5711: need to pump up our graph,

5712:  that is, to replace every  vertex by a trinion and to glue

5713:   trinions by tubes along the edges.

5714: 

5715: 

5716: \subsection{ Circle complex and Graph complex}

5717: 

5718: 

5719:   Let $\Si_{g,n}$ be an oriented surface of genus $g$ with $n$

5720:   punctures and $C$ be a simple, closed, homotopy non-trivial,

5721:    non-bounded pinchs loop on $\Si_{g,n}$. We call such loop a

5722:      {\it circle} and  consider it up to the isotopy class.

5723:      

5724: {\it Circle

5725: complex} or {\it Curve complex} $C(\Si_{g,n})$ has the set  of

5726: circles as   0-skeleton $C_{0}(\Si_{g,n})$,

5727: \begin{enumerate}

5728: \item 1-skeleton $C_{1}(\Si_{g,n})$ is the set of

5729:  pairs of classes of disjoint circles;

5730: \item triangles correspond to triples of classes of disjoint cirles ;

5731: \item every $i$-simplex corresponds to a set of $i+1$ classes

5732:  mutually

5733: disjoint circles,

5734: \item so, simplexes of maximal dimension are collections of $3g-3$

5735:  disjoint

5736: circles just as (4.7).

5737: \end{enumerate}

5738: The main fact which we will use is the following: the group of

5739: combinatorial automorphisms

5740: \begin{equation}

5741: Aut C(\Si_{g}) = Mod_{g}

5742: \end{equation}

5743: is the modular group (5.30).

5744: \begin{rmk}

5745: By the Roiden's theorem

5746: \begin{equation}

5747: Isom \quad \tau_{g} = Mod_{g}

5748: \end{equation}

5749: where the Teichmuller space $\tau_{g}$ (5.31) is equipped with

5750: the Teichmuller metric. Moreover as a metric space $C(\Si_{g,n})$

5751: is very closed to $\tau_{g}$. In particular, $C(\Si_{g,n})$ is

5752: $\delta$-hyperbolic (\cite{MM}). The same facts are true for the

5753: Weil-Peterson metrics.

5754: \end{rmk}

5755: Recall that in  the case of a pinched curve $\Si_{g, n}$ the circle

5756: can't be isotopic to a boundary component.

5757: 

5758: 

5759: 

5760: 

5761: \begin{figure}[tbn]

5762: \centerline{\epsfxsize=3in\epsfbox{7c.eps}} \caption{\sl Triangles} \label{Fig 13}

5763: \end{figure}

5764: 

5765: 

5766: 

5767: 

5768: 

5769:  For two special but very important  cases $\Si_{0, 4}$ and

5770:   $\Si_{1,1}$ the

5771: definition is slightly different

5772:  (since there are no disjoint pairs of circles

5773: ): two vertices are connected by an edge if circles have 2

5774: intersections in $\Si_{0,4}$-case and 1 intersection in

5775: $\Si_{1,1}$-case. In both cases

5776: \begin{equation}

5777: C(\Si_{0,4}) = C(\Si_{1,1})

5778: \end{equation}

5779: is the ideal triangulation of  2-disc (so called Farey Graph,

5780: see Fig. 14).

5781: 

5782: \begin{figure}[tbn]

5783: \centerline{\epsfxsize=3in\epsfbox{4c.eps}} \caption{\sl Farey

5784: graph (the ideal triangulation of disk)} \label{Fig 13}

5785: \end{figure}

5786: 

5787: 

5788:   {\it Trinion decomposition

5789:  graph } $C_1{TD(\Si)}$ has the set of trinion decompositions

5790:   as 0-skeleton

5791: \begin{equation}

5792: V(C_{1}(TD(\Si))) = \{(C_{1}, ... , C_{3g-3})\}

5793: \end{equation}

5794: (see (4.7)) and a pair of trinion decompositions is connected by

5795: an edge if the decompositions differ by an {\it elementary move}. That

5796: is, in the system of circles $C_{1}, \dots, C_{3g-3}$ on a

5797: Riemann surface

5798:  $\Si$ we replace  only one circle, for example $C_{1}$, with

5799:  another circle $C'_{1}$ intersecting twice if $C_{1}$ and

5800:  $C'_{1}$ don't lie on the same trinion  and intersecting

5801:  once otherwise (minimal intersections):

5802: \begin{equation}

5803: \{C_{1}, \dots, C_{3g-3}\} \to \{C'_{1}, \dots, C_{3g-3}\}.

5804: \end{equation}

5805: Recall that every trinion decomposition $C_{1}, ... , C_{3g-3}$

5806: defines the 3-valent graph $\Ga$ (4.9)- (4.11). Thus we have the

5807: {\it finite graph} $C_{1}(TG_{g})$ with the set of vertices

5808: $$

5809: V(C_{1}(TG_{g})) = TG_{g}

5810: $$

5811: (see (6.21)) and  a pair of graphs $\Ga, \Ga'$  is connected by an

5812: edge if they differ by an elementary transformation 

5813: (see Definition 9) corresponding to an elementary move of trinion

5814:  decomposition (6.45). The fibers of the map

5815: $$

5816: \ga \colon V(C_{1}(TD(\Si)))  \to TG_{g}

5817: $$

5818: are described by the formula

5819: $$

5820: Mod_{g} / Aut H_{\Ga}

5821: $$

5822: where $H_{\Ga}$ is the handlebody (6.9) and $Aut$ is the group of

5823: isotopy classes of diffeomorphisms of $\Si = \Si_{\Ga}$

5824: extendable to diffeomorphisms of the handlebody.

5825: 

5826: The finite TG-graph $C_{1}(TG_g)$ is  1-skeleton  of the finite   $TG_g$-complex

5827: $C(TG_g)$ which can be constructed by the following procedure: first of all let us glue by 2-cells all triangles coming from elementary transformations.

5828: Over such triangles in TG-graph we have triangles from Fig. 13 of the infinite TD-graph. Up to such triangles we have the canonical connection $\na$ (6.22) on

5829:  TG-graph and we can apply the "geodesic technique" (6.25)-(6.28). 

5830: In particular,  we obtain  2-complex $C_\na(TG_g) = C(TG_g)$. The important

5831:  fact (see Remark at the end of this subsection) proved in \cite{HT} is the following

5832: \begin{enumerate}

5833: \item $C(TG_g)$ is simply connected that is

5834: $$

5835: C(TG_g) = S^2,

5836: $$

5837: \item hence our TG-graph is plane;

5838: \item hence we may ask about this graph all classical questions, for example,

5839: on the chromatic number and so on.

5840: \end{enumerate}

5841: 

5842: However  the standard way to check these properties is the following

5843: \begin{enumerate}

5844: \item to construct the infinite complex $C(TD(\Si))$ "by our hands";

5845: \item to check described properties;

5846: \item to desend all constructions to TG-complex using the map $\ga : C_1(TD) \to C_1(TG_g)$ and to get the complex $C(TG_g)$;

5847: \item to check the required properties for this finite complex.

5848: \end{enumerate} 

5849: 

5850: \begin{rmk}

5851: It is quite natural to apply this geodesic technique to the infinite graph $C_1(TD(\Si))$ directly but we have to go passing the standard way.

5852: \end{rmk}

5853: 

5854: 

5855:  For our special cases  of Riemann surfaces it  can be proved that

5856: \begin{equation}

5857: C_{1}(TD(\Si_{0,4})) = C_{1}(TG(\Si_{1,2})) = C_{1}(\Si_{0,4})=

5858: C_{1}(\Si_{1,2})

5859: \end{equation}

5860: is the Farey graph that is 1-skeleton of the Farey complex again.

5861: 

5862: 

5863: \begin{figure}[tbn]

5864: \centerline{\epsfxsize=3in\epsfbox{8c.eps.eps}} \caption{\sl Squares} \label{Fig 13}

5865: \end{figure}

5866: 

5867: 

5868: 

5869: The infinite complex $C(TD)$ is 2-dimensional. Its 2-cells

5870: represent relations between special moves, that is, loops in

5871: $TD$-graph

5872:   and they can be triangles, squares,

5873:  pentagones and hexagons. Every trinion decomposition of $\Si_{g,n}$

5874:   contains trinion decompositions of surfaces with smaller $g$ and $n$ as

5875:   fragments of it. We call such fragments subsurfaces.

5876:  The relations,  coming from such subsurfaces are the following:

5877: 

5878: 

5879: \begin{figure}[tbn]

5880: \centerline{\epsfxsize=3in\epsfbox{9c.eps.eps}} \caption{\sl Pentagons}

5881:  \label{Fig 13}

5882: \end{figure}

5883: 

5884: 

5885: 

5886: 

5887: \begin{enumerate}

5888: \item $\Si_{1,1}$ and $\Si_{0,4}$ for triangles;

5889: \item \begin{equation}

5890: \Si_{1,1} \cup \Si_{0,4} = \Si_{1,3}

5891: \end{equation}

5892: for squares;

5893: \item $\Si_{0,5}$ for pentagons and

5894: \item $\Si_{1,2}$ for hexagons

5895: \end{enumerate}

5896: (see Figures 13, 15-17).

5897: 

5898: 

5899: 

5900: \begin{figure}[tbn]

5901: \centerline{\epsfxsize=3in\epsfbox{10c.eps.eps}} \caption{\sl Hexagons} \label{Fig 13}

5902: \end{figure}

5903: 

5904: 

5905: 

5906: 

5907: Amazing fact is that we can define 2-cells of this complex in

5908: terms of  $TD$-graph. It means that $C(TD)$-complex carries no

5909: more information than its 1-skeleton:

5910: \begin{enumerate}

5911: \item triangles are contained in subcomplexes

5912:  $C(TD(\Si_{1,1}))$ or  $C(TD(\Si_{0,4}))$ which are

5913:  the Farey complexes.

5914: Every edge of this complex  is a side of two triangles

5915: corresponding

5916:  to two subsurfaces

5917:  $C(TD(\Si_{1,1}))$ or  $C(TD(\Si_{0,4}))$;

5918: \item moreover, if we fix one vertex in any Farey subgraph of $TGD$-graph,

5919:  we get

5920: a circle which is a vertex of $C(\Si_{g,n})$ (indeed, fixing any

5921: Farey subgraph of $TD$-graph we get the collection $\{ C_1, \dots

5922: , C_{3g-3} \}$ of circles  of

5923:  trinion decompositions with fixed set of circles

5924:   $( C_2, \dots , C_{3g-3} )$,

5925:  since our Farey subgraph corresponds to a choice of $C_1$ );

5926: \item if $C$ is a vertex of  $C(\Si)$ (as $C_1$  before ) then there

5927: exists a Farey subgraph with fixed vertex but not unique (indeed

5928: we can extend

5929:  our circle $C_1$ to a complete set  $\{ C_1, \dots , C_{3g-3} \}$

5930:  and consider all

5931: trinion decompositions with fixed  $( C_2, \dots , C_{3g-3} )$;

5932: we get a Farey subgraph $F_{C}$  (just a possible choice) with

5933: fixed point $p \in F_C$);

5934: \item now we can define a map

5935: \begin{equation}

5936: \phi \colon Aut C_1(TD) \to Aut C_{0}(\Si_{g,n})

5937: \end{equation}

5938: sending a circle $C \in  C_{0}(\Si_{g,n})$ to  unique vertex in

5939: $C_{0}(\Si_{g,n})$

5940:  corresponding to the marked Farey graph  $g(p) \in g (F_C)$;

5941: \item   square, pentagon or hexagon ($n$-gones for small $n$)

5942:  can be  union of  triangle and

5943: $n-1$-gon  gluied along an edge. To distinguish this trivial

5944: situation from "non trivial" or {\it

5945:  alternating}

5946:  we consider a loop with $n$ consecutive points  such

5947: that  every consecutive triple $p_{i},p_{i+1},p_{i+2} $ and the

5948: pair of edges

5949:  between  arn't contained in any Farey subcomplex;

5950: \item at last, a hexagon can be {\it almost} alternating when its three

5951: consecutive

5952: vertices can be extended to the set of vertices of a square loop.

5953: \end{enumerate}

5954: 

5955: Using these simple rules Margalit proved in \cite{M} that

5956: \begin{enumerate}

5957: \item the  map $\phi$ (6.47) can be extended to a

5958:  well defined map

5959:  \begin{equation}

5960:  Aut C_1(TD) \to Aut C(TD),

5961: \end{equation}

5962: \item and this map can be extended to an isomorphism

5963: \begin{equation}

5964: \phi \colon Aut C(TD) \to Aut C(\Si),

5965: \end{equation}

5966: \item such that

5967: \begin{equation}

5968:  Aut C_{1}(TD)= Aut C(TD)=  Aut C(\Si)= Mod_g.

5969: \end{equation}

5970: \end{enumerate}

5971:  TD-complex  $C(TD)$ was introduced by Hatcher and Thurston

5972:  in \cite{HT} where they proved that this complex is connected and

5973:   simple connected. But  the equality (6.50) was proved quite

5974:   recently in \cite{M}.

5975: 

5976:  \begin{rmk} Returning to the Rational Conformal Field Theory

5977:   WZW we can consider the vector space spanned by  different

5978:   conformal blocks (see Definition 8) interpreted as the

5979:   holomorphic sections of some power of theta-line bundle on the

5980:   moduli space of semi-stable vector bundles on a Riemann

5981:    surface $\Si$ (see section 3).

5982:     This surface can be

5983:   formed by gluing a number of trinions. The conformal

5984:   blocks are obtained by summing over the intermediate states

5985:   passing through the circles along which we glue our trinions.

5986:   Different gluing procedures lead to different bases of the same

5987:   vector space. Hence vectors of

5988:   this space obtained by one gluing procedure can be expressed as

5989:   linear combinations of vectors found by another way of gluing

5990:    the same surface. These linear transformations are given by

5991:    {\it dual matrices}, which  have to satisfy some

5992:    {\it consistency conditions}. To describe these conditions it is

5993:    quite convinient to use the finite complex $C(TG)$.

5994:     First of all, we

5995:    have some very simple duality matrices defined by the elementary moves. The complex $C(TG)$ is connected, hence every duality matrix can be represented as a

5996:     product of the simple ones (maybe not uniquely). To get

5997:    an  unambiguous description of all duality matrices, we have to

5998:      use consistency conditions on simple duality matrices which

5999:      are given by loops of simple matrices. So we have to find all

6000:      independent conditions. Every loop of simple matrices

6001:      corresponds to the statement that the product of the corresponding

6002:      duality matrices is equal to one. Fixing a collection of

6003:       {\it fundamental loops}, like triangles, squares , pentagons

6004:       and hexagons before, and filling faces of such loops

6005:      gives us the graphs complex $C(TG)$. The relations for the

6006:      loops are complete iff the complex is simply connected.

6007:    Thus we get a system of (polynomial) equations for duality

6008:     matrices. Solutions to this system of equations give the {\it

6009:     moduli space of rational conformal field theories}. Sending

6010:  the   level $k$ to infinity we get a fusion algebra which for our

6011:      $\SU(2)$-case is the representation algebra $R(sl(2,\C))$

6012:       from subsection 5.1.

6013: \end{rmk}

6014: 

6015: Returning to wave function spaces $\sH^k_\Ga$ (4.43)

6016:  we will see that the system

6017: of these spaces defines a vector bundle

6018: \begin{equation}

6019:  \sH^k_{TG} \to C(TG_g) = S^2

6020: \end{equation}

6021:   on the $TG$- complex  with a special structure-some local

6022: linear decomposition on line subbundles.

6023: 

6024: Remark that the bundles is a subbundle of the trivial bundle

6025: (see (4.45))

6026: \begin{equation}

6027:  \sH^k_{TG} \subset \C^{(2k)^{3g-3}} \times C_{TG} = (\oplus_{\al \in T^A_{2k}} \C_\al)

6028: \times C_{TG}.

6029: \end{equation}

6030: 

6031: 

6032:  All these constructions  suggest  existence of

6033: a flat connection $a_K$. Since the complex  $C_{TG}$ is simply

6034: connected the bundle $\sH^k_{TG}$ has to be trivial with the

6035: trivialization defined up to scaling. From this we get the

6036: projective  bundle

6037: \begin{equation}

6038: \PP \sH^k \to \sM_g

6039: \end{equation}

6040: on the moduli space of curves of genus $g$ with the projective

6041: flat connection. But this bundle is a slight projective bundle

6042: with the phase amplitude (6.18) only.

6043: 

6044: Returning to the graph $C^{1}_{TG_g}$ we will see that

6045: \begin{enumerate}

6046: \item the system of duality matrices defines a flat connection

6047: on $C_{1}(TG_g)$;

6048: \item this connection has  trivial projective monodromy, thus

6049: \item this connection defines a representation

6050: \begin{equation}

6051: \rho \colon Aut C_{1}(TG_g) \to PGL(rk_{g}(\emptyset)), \C).

6052: \end{equation}

6053: \item this representation is equal to the desired Kohno representation.

6054: \end{enumerate}

6055: 

6056: 

6057: 

6058: 

6059: 

6060: \subsection{Gauge theory on graphs.}

6061: 

6062: A connection $a$ on the trivial $\SU(2)$-bundle on $\Ga$ is a map

6063: \begin{equation}

6064: a\colon P_1(\Ga)\to \SU(2)

6065: \end{equation}

6066: such that for the  orientation reversing involution $i_e$ we have

6067: \begin{equation}

6068: a(i_e(\vec e))=a(\vec e)^{-1}.

6069: \end{equation}

6070: Then the ``path integral'' is

6071: \begin{equation}

6072: a( (\vec e_1,\dots, \vec e_d))=a(\vec e_1) \cdot \cdots \cdot

6073: a(\vec e_d) \in \SU(2)

6074: \end{equation}

6075: that is, it is the map

6076: \begin{equation}

6077: a\colon P_d(\Ga)\to \SU(2)

6078: \end{equation}

6079: such that for the orientation reversing involution $i_\De$ we have

6080: \begin{equation}

6081: a(i((\vec e_1,\dots, \vec e_d)))=a((\vec e_1,\dots, \vec

6082: e_d))^{-1}.

6083: \end{equation}

6084: In the same vein we have the monodromy map for loops

6085: \begin{equation}

6086: a\colon L_d(\Ga)_v \to \SU(2).

6087: \end{equation}

6088: Obviously every such connection is {\it flat }.

6089: 

6090: Let $\sA(\Ga)$ be the space of connections that is the space of

6091: functions (6.56) subjecting to the constraint (6.57):

6092: \begin{equation}

6093: \sA(\Ga)=\{ a\in \SU(2)^{P_1(\Ga)} \vert a(i_e(\vec e))=a(\vec

6094: e)^{-1} \}.

6095: \end{equation}

6096: 

6097: 

6098: 

6099: Every element $\w{g}$ of the gauge transformations group

6100: $\sG(\Ga)$ is a function

6101: \begin{equation}

6102: \w{g}\colon V(\Ga)\to \SU(2)

6103: \end{equation}

6104: that is

6105: \begin{equation}

6106: \sG(\Ga)=\SU(2)^{V(\Ga)}

6107: \end{equation}

6108: with the componentwise multiplication.

6109: 

6110: This group acts on the space of connections $\sA(\Ga)$ by the

6111: following rule

6112: \begin{equation}

6113: \w{g}(a(\vec e))=\w{g}(v_s(\vec e)) \cdot a(\vec e) \cdot

6114: \w{g}(v_t(i(\vec e))).

6115: \end{equation}

6116: Recall that $\w{g}(v_t(i(\vec e)))= \w{g}(v_t(\vec e))^{-1}.$

6117: 

6118: The space of gauge orbits

6119: \begin{equation}

6120: \sB(\Ga)=\sA(\Ga) /\sG(\Ga)

6121: \end{equation}

6122: is the space of classes of representations of the fundamental

6123: group of $\Ga$ as it is expected because all our connections are

6124: flat:

6125: \begin{equation}

6126:  \sA(\Ga) /\sG(\Ga)=CLRep(\pi_1(\Ga), \SU(2))

6127: \end{equation}

6128:  is the quotient of the space of representations  by the adjoint action

6129: \begin{equation}

6130:  CLRep(\pi_1(\Ga), \SU(2))=\Hom(\pi_1(\Ga) , \SU(2)) /Ad \SU(2)

6131: \end{equation}

6132: - the space of classes of representations of this group.

6133: 

6134: Obviously the fundamental group of 3-valent graph $\Ga$ of genus

6135: $g$ is the free group with $g$ generators. Thus  the quotient

6136: (6.66)

6137: \begin{equation}

6138: CLRep(\pi_1(\Ga), \SU(2)) = uS_g

6139: \end{equation}

6140: is the {\it unitary Schottki space}. (see the definition (4) in 3.2

6141: after the formula (3.44)).

6142: 

6143: 

6144: The gauge transformation group $\sG(\Ga)$ contains the diagonal

6145: subgroup

6146: \begin{equation}

6147: \SU(2)_\De \subset \sG(\Ga)

6148: \end{equation}

6149: of constant functions (6.63). The action of this subgroup defines

6150: the space of constant orbits

6151: \begin{equation}

6152:  CL \sA(\Ga)=\sA(\Ga) / \SU(2)_\De.

6153: \end{equation}

6154: 

6155: For every vertex $v\in V(\Ga)$ we have the subgroup

6156: \begin{equation}

6157: \sG(\Ga)_v=\{\w{g}\in \sG(\Ga) \vert \w{g}(v)=\id \}

6158: \end{equation}

6159: of gauge transformations preserving the framing at $v$. This is a

6160: normal subgroup, and

6161: \begin{equation}

6162: \sG(\Ga) /\sG(\Ga)_v=\SU(2)_\De

6163: \end{equation}

6164: Thus the full group of gauge transformations is a semidirect

6165: product of $\sG(\Ga)_v$ and $\SU(2)_\De$.

6166: 

6167: The quotient

6168: \begin{equation}

6169: \sA(\Ga) /\sG(\Ga)_v=Rep(\pi_1(\Ga, v), \SU(2))

6170: \end{equation}

6171: is the orbit space of framed (at $v$) connections. This space

6172: depends on

6173:  the choice of a point $v$.

6174: 

6175: Thus the quotient map (6.67)

6176: \begin{equation}

6177: P\colon \sA(\Ga)\to CLRep(\pi_1(\Ga), \SU(2))

6178: \end{equation}

6179: can be decomposed as follows

6180: \begin{equation}

6181: \sA(\Ga) \xrightarrow{\,P\,} Rep(\pi_1(\Ga), \SU(2))

6182: \xrightarrow{/\SU(2)\De} CLRep(\pi_1(\Ga), \SU(2))

6183: \end{equation}

6184: or

6185: \begin{equation}

6186: \sA(\Ga)\xrightarrow{/\SU(2)_\De} CL\sA(\Ga) \xrightarrow{P_{cl}}

6187: CLRep(\pi_1(\Ga), \SU(2)).

6188: \end{equation}

6189: The involution $i_\De$ acts on the space $\sA(\Ga)$ of connections

6190: \begin{equation}

6191: i^*_\De\colon \sA(\Ga)\to \sA(\Ga)

6192: \end{equation}

6193: and there exists an element

6194: \begin{equation}

6195: \w{g}_i=\begin{pmatrix}0&1\\-1&0\end{pmatrix} \in\SU(2)_\De

6196: \subset \sG(\Ga)

6197: \end{equation}

6198: such that

6199: \begin{equation}

6200: i^*_\De=\w{g}_i.

6201: \end{equation}

6202:  Thus the involution $i^*_\De$ (3.23) acts

6203: trivially on $CL\sA(\Ga)$. Recall that a gauge fixing is a

6204: section of the projection $P$ (6.75).

6205: 

6206: 

6207: Recall that the set $\Conj (\SU(2))$ of conjugacy classes of

6208: elements of $\SU(2)$ can be described as the interval $[0,1]$

6209: with respect to the map

6210: \begin{equation}

6211: conj\colon \SU(2)\to \Conj (\SU(2))=[0,1]

6212: \end{equation}

6213: sending a matrix $g\in \SU(2)$ to

6214: \begin{equation}

6215:  conj g=\frac{1}{\pi} \cdot \cos^{-1}\Bigl(\frac{1}{2}\Tr g \Bigr)\in

6216:  [0,1].

6217:  \end{equation}

6218: 

6219: Using this map coordinatewise one obtains the map

6220: \begin{equation}

6221: conj\colon \CL\sA(\Ga)\to [0,1]^{3g-3}=\prod_{e\in E(\Ga)}

6222: [0,1]_e,

6223: \end{equation}

6224: which is obviously surjective.

6225: 

6226: The map $conj$ is the composite

6227: \begin{equation}

6228:  \sA(\Ga) \xrightarrow{\SU(2)_\De} \CL\sA(\Ga)\to

6229: [0,1]^{3g-3}=\prod_{e\in E(\Ga)} [0,1]_e

6230: \end{equation}

6231: and the involution $i_\De$ preserves its fibers.

6232: 

6233: 

6234: 

6235: 

6236:  \subsection{Abelian gauge theory and $U(1)$-spin networks}

6237: 

6238: 

6239: 

6240: 

6241: 

6242:  Abelian gauge theory on $\Ga$ is a good model for non-Abelian gauge

6243: theory. Abelian spin networks can be considered as

6244: theta-characteristics under numeration of theta functions.

6245:  A $U(1)$-connection $a$ is a map

6246: \begin{equation}

6247: a\colon P_1(\Ga)\to U(1)

6248: \end{equation}

6249: such that

6250: \begin{equation}

6251: a(i_{e}(\vec e))=a(\vec e)^{-1}.

6252: \end{equation}

6253: The ``path integral'' is given by

6254: \begin{equation}

6255: a( (\vec e_1,\dots, \vec e_d))=\prod_{\vec e\in P_1(\Ga)} a(\vec

6256: e)\in U(1),

6257: \end{equation}

6258: and so on. Again let

6259: \begin{equation}

6260: \sA_{U(1)}(\Ga)=\bigl\{ a\in U(1)^{P_1(\Ga)} \bigm| a(i_e(\vec

6261: e))=a(\vec e)^{-1} \bigr\}.

6262: \end{equation}

6263: be the space of $U(1)$-connections. Then we have the same

6264: involution

6265: \begin{equation}

6266:  \sA_{U(1)}(\Ga) \xrightarrow{i^*_e} \sA_{U(1)}(\Ga).

6267: \end{equation}

6268: Every element $\w{u}$ of the gauge transformation group

6269: $\sG_{U(1)}(\Ga)$ is a function

6270: \begin{equation}

6271: \w{u}\colon V(\Ga)\to U(1).

6272: \end{equation}

6273: This group acts on the space of connections $\sA_{U(1)}(\Ga)$ by

6274: the same rule:

6275: \begin{equation}

6276: \w{u}(a(\vec e))=\w{u}(v_s(\vec e)) \cdot a(\vec e) \cdot

6277: \w{u}(v_t(i(\vec e))).

6278: \end{equation}

6279: 

6280: However, there is one important difference between Abelian and

6281: non-Abelian theories. Namely, the diagonal group

6282: \begin{equation}

6283: U(1)_\De \subset \sG_{U(1)}(\Ga)

6284: \end{equation}

6285: has  trivial action. Thus

6286: \begin{equation}

6287: \sB_{U(1)}(\Ga)=\sA_{U(1)}(\Ga) /\sG_{U(1)}(\Ga)=\Hom(\pi_1(\Ga) ,

6288: U(1))=U(1)^g.

6289: \end{equation}

6290: Remark that this is abelian unitary Schottki space, that is, the

6291: zero fiber of the projection $\pi$ (2.22) on which we are

6292: constructing our delta functions (2.46) and (2.48) for the

6293: application of the coherent state transform to get an analytical

6294: presentation of theta functions on the complex Schottki space

6295: $(\C^*)^g$. We would like to do this for non-abelian case too.

6296: 

6297: Let

6298: \begin{equation}

6299: P_a\colon \sA_{U(1)}(\Ga)\to U(1)^g

6300: \end{equation}

6301: be the projection map.

6302: 

6303: The abelian and non-abelian theories are related by the following

6304: map:

6305: \begin{equation}

6306:  d\colon \sA_{U(1)}(\Ga)\to \sA(\Ga)

6307:  \quad\hbox{given by}\quad

6308:  d(a(\vec e))=\begin{pmatrix} e^{i \phi}, 0\\ 0, e^{- i \phi}

6309: \end{pmatrix}.

6310: \end{equation}

6311: This map $d$ is equivariant with respect to every involution

6312: $i^*_e$. Obviously the image $d (\sA_{U(1)}(\Ga) ) \subset

6313: \sA(\Ga)$ is a 2-section of the projection $conj$, that is, the

6314: composite

6315: \begin{equation}

6316:  conj \circ d\colon \sA_{U(1)}(\Ga)\to \prod_{e\in E(\Ga)}

6317: [0,1]_e

6318: \end{equation}

6319:  is the factorization by the involution $i^*_\De$.

6320: 

6321: It is easy to check

6322: 

6323: \begin{prop} In the chain of maps

6324: \begin{equation}

6325: \sA_{U(1)}(\Ga) \xrightarrow{\,d\,} \sA(\Ga) \xrightarrow{\,P\,}

6326: U(1)^g\in \CLRep (\pi_1(\Ga))

6327: \end{equation}

6328: the composite $d \circ P$ is the projection map $P_a$ (5.75).

6329: \end{prop}

6330: 

6331: 

6332: The set of irreducible representations of $U(1)$ is

6333: \begin{equation}

6334: \widehat{U(1)} = \Z

6335: \end{equation}

6336: and a spin $U(1)$-network on a graph $\Ga$ is given by a function

6337: \begin{equation}

6338: w \colon E(\Ga) \to \widehat{U(1)} = \Z

6339: \end{equation}

6340: (like (4.46)) such that for every triple of edges meeting at a

6341: vertex $v \in V(\Ga)$ the following condition holds

6342: \begin{equation}

6343:  w(e_1) + w(e_2) + w(e_3) = 0.

6344: \end{equation}

6345: Thus the triangle inequality  (Clebsh-Gordan conditions)

6346: becomes the equality.

6347: 

6348: Again a spin network is of level $k$  when $w(e) \leq 2k$.

6349: 

6350: Considering the basis $(a_1 , \dots , a_g, b_1, \dots b_g)$

6351: (2.32) we have the decomposition of the jacobian of the curve

6352: $\Si_\Ga$

6353: \begin{equation}

6354: J_{\Si_\Ga} = Hom(\pi_1(\Si_\Ga), U(1)) = \prod_{i=1}^g

6355: U(1)_{a_i}\times \prod_{i=1}^g U(1)_{b_i}

6356: \end{equation}

6357: which coincides with  the decomposition (2.21),(2.32).

6358: 

6359: Now we can consider the coordinates (4.14) of the map (4.41) as

6360: elements of

6361:  the target group $U(1)$.  Hence the map $\pi_\Ga$ (4.41) has the

6362:   form

6363:  \begin{equation}

6364: \pi_\Ga \colon J_{\Si_\Ga} \to U(1)^{E(\Ga)} = T^A

6365: \end{equation}

6366: where the target is the action torus (4.40). The image of this

6367: map is a g-dimensional torus

6368: \begin{equation}

6369: \De_\Ga = T^g_- \subset T^A.

6370: \end{equation}

6371: Now we can send every spin network that is a function (6.98)

6372: satisfying (6.99) to a point of order $k$ on $T^A$  dividing by

6373: $k$

6374: \begin{equation}

6375: w(e) \to 1/k w(e).

6376: \end{equation}

6377:  We get the map of the set $SNW_a^k$

6378: of abelian spin works of level $k$

6379: \begin{equation}

6380: 1/k \colon SNW_a^k \to (T^A)_k.

6381: \end{equation}

6382: Amusing fact is that

6383: \begin{equation}

6384: 1/k (SNW_a^k) \subset \De_\Ga = T^g_-.

6385: \end{equation}

6386: Moreover

6387: \begin{equation}

6388: 1/k(SNW_a^k) = W^k(\Ga) = BS_k(\Ga).

6389: \end{equation}

6390: Thus we get the canonical identification of the set $\Z^g/ k\Z^g$ of

6391: theta characteristics of level $k$ (2.48) with the set of spin

6392: networks

6393: \begin{equation}

6394: (\Z^g / k\Z^g) = SNW_a^k.

6395: \end{equation}

6396: So in the classical case we can use $U(1)$-spine networks for the

6397: enumeration of theta characteristics. In the same vein we will use

6398: $\SU(2)$-spin networks

6399:  for the enumeration of non-abelian theta characteristics.

6400: 

6401: Also we saw  that the harmonic analysis on $T^g_+ = U(1)^g$ is

6402:  the standard Fourier analysis on a torus. We use the Fourier

6403: decomposition

6404:  to define our delta-functions (2.46) and (2.48) which CST $C_t^{-i \Om}$

6405:  (2.42) sends to classical theta functions. Now we have to extend these

6406:  constructions to the non-abelian case.

6407: 

6408:  \subsection{Harmonic analysis of $\SU(2)$-spin networks}

6409: 

6410: Recall that a spin network defines a map $ j \colon E(\Ga)\to

6411: \widehat{SU(2)}$ with constraints. For a triple of representations

6412: $j_{e_1}, j_{e_2}, j_{e_3}$  an

6413:  intertwiner is  trivial component of the tensor product

6414: $j_{1} \tensor j_{2} \tensor j_{3}$. Such a component exists iff

6415: under the traditional identification $\widehat{SU(2)}=\half

6416: \Z^{+}$ the Clebsch--Gordan conditions

6417:  \begin{equation}

6418:  j_{1}+j_{2}+j_{3}\in \Z \quad\text{and}\quad \vert j_{1}-j_{2} \vert \le

6419:  j_{3} \le j_{1}+j_{2}

6420:  \end{equation}

6421:  (the triangle inequality) hold. Function $ j \colon E(\Ga)\to

6422: \widehat{SU(2)}$ defines a  spin network $\Ga_j$ iff these

6423: conditions hold for every triple $j_{v,1}, j_{v,2}, j_{v,3}$ of

6424: representations around every vertex $v\in V(\Ga)$.

6425: 

6426: Any intertwiner $ i_v\in j_{v,1} \tensor j_{v,2} \tensor j_{v,3}$

6427: is defined uniquely so one  omits any labeling of the vertices

6428: and denotes  $\SU(2))$-spin network by the symbol $\Ga_{j}$.

6429: 

6430:  Spin network $\Ga_{j}$

6431: is of  level $k$ if $j_{i} \le k$ for every edge $e_i$ and in the

6432: last inequality we take $min(j_{1}+j_{2}, k - j_{1}-j_{2})$

6433: instead of $j_{1}+j_{2}$. Of course if $k' > k$ then every spin

6434: network of  level $k$ is of level $k'$ automatically. We would

6435: like to fix a level and  denote a spin network of level $k$ by the

6436: symbol $\Ga_{j^k}$ and the set of all spin net works by the symbol

6437:  $SNW^k(\Ga)$. Recall that we have the identification $SNW^k(\Ga) = W^k(\Ga)

6438:  $ (see the end of subsection 6.1).

6439: 

6440: 

6441: 

6442: Spin network $\Ga_{j, k}$ is a purely combinatorial object but we

6443: can identify it with

6444:  a function on the special space - the unitary Schottky space $uS_g$

6445: given by the formula

6446:  \begin{equation}

6447:  uS_{g}=\SU(2)^{g} / Ad_{\mathrm{diag}} \SU(2)

6448:  \end{equation}

6449: where $Ad_{\mathrm{diag}}\SU(2)$ is the diagonal adjoint action

6450: on the direct product. This space doesn't depend on a three

6451: valent graph modulo the choice of appropriate topological data.

6452: Only one invariant of a graph is used, namely its

6453:  genus $g$.However, it is quite natural to use a graph $\Ga$

6454: as a starting point of the construction.

6455: 

6456: Consider the product

6457:  \begin{equation}

6458:  \SU(2)^{E(\Ga)}=\prod_{e\in E(\Ga)} \SU(2)_e

6459:  \end{equation}

6460: with $\SU(2)$ components enumerated by edges of $\Ga$ which we

6461: can identify

6462:  with the space of flat connections $\sA$ (6.88) (under a choice of any

6463:   orientation of $\Ga$).

6464: 

6465: Let $dx$ be the Haar measure on $\SU(2)$ normalized by the

6466: condition $\int_{\SU(2)}dx=1$ and $\vec dx$ the product measure

6467: on $\SU(2)^{E(\Ga)}$ normalized by $\int \vec dx=1$. Then by the

6468: Peter--Weyl formula, any function $f\in L^{2}(\SU(2)^{E(\Ga)},

6469: \vec dx)$ has the decomposition

6470:  \begin{equation}

6471:  f(x)=\sum_{\vec{\rho}\in \widehat{\SU(2)^{E(\Ga)}}}

6472:  \Tr[B_{\vec{\rho},f} \vec{\rho}(x)],

6473:  \end{equation}

6474: where $\widehat{\SU(2)^{E(\Ga)}}$ is the space of irreducible

6475: representations of $\SU(2)^{E(\Ga)}$, and $B_{\vec{\rho},f}$ are

6476: endomorphisms of the space $V_{\vec{\rho}}$ of the representation

6477: $\vec{\rho}$, given by

6478:  \begin{equation}

6479:  B_{\vec{\rho},f}=\frac{1}{\dim V_{\vec{\rho}}}\int_{\SU(2)^{E(\Ga)}}

6480:  f(x) \vec{\rho}\,^{-1}(x) \vec dx.

6481:  \end{equation}

6482: Every irreducible representation of $\SU(2)^{E(\Ga)}$ is given by

6483: tensor product of irreducible representations of $\SU(2)$ $

6484: \vec\rho=\rho_{1}\tensor \cdots \tensor\rho_{3g-3}$. So every spin

6485: network $\Ga_{j}$ of genus $g$ defines a representation of

6486: $\SU(2)^{E(\Ga)}$ by the tensor product of all labels

6487:  $\vec j=\bigotimes_{e\in E(\Ga)} j_e$.

6488: 

6489: Of course this is an analog of the standard Fourier decomposition

6490: (2.41) or even (2.46). Here the representation $\vec \rho$ is a

6491: label of {\it frequency} and an endomorphism $B_{\vec \rho, f}$

6492: is the {\it Fourier coefficient } that is the analog of a {\it

6493: number}. The formula (6.113) is nothing else than the integral

6494: formula for a Fourier coefficient.

6495: 

6496: Every endomorphism of the space $V_{\vec j}$ is a vector in the

6497: tensor product

6498: $$

6499:  \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr)

6500:  \tensor \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr)^*=

6501:  \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr)

6502:  \tensor \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr),

6503:  $$

6504: since we are dealing with $\SU(2)$-representations.

6505: 

6506: But after symmetrizing components of the final product can be

6507: labeled by elements of the set $F(\Ga)$, and we can decompose it

6508: as

6509:  \begin{equation}

6510:  \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr)

6511:  \tensor \Bigl(\bigotimes_{e\in E(\Ga)} j_e\Bigr)^*=

6512:  \bigotimes_{v\in V(\Ga)}(j_{v,1} \tensor j_{v,2} \tensor j_{v,3}).

6513:  \end{equation}

6514: 

6515: For every triple of representations around a vertex $v\in V(\Ga)$ we

6516: have a vector $ i_v\in j_{v,1} \tensor j_{v,2} \tensor j_{v,3}$

6517:  and their tensor product gives us the vector

6518:  \begin{equation}

6519:  B(\Ga_j)=\bigotimes_{v\in V(\Ga)} i_v \in  End V_{\vec j}\,.

6520:  \end{equation}

6521: So one has the function

6522:  \begin{equation}

6523:  f_{\Ga_{j}}(x)= Tr[B(\Ga_j) \vec j(x)]\in L^{2}(\SU(2)^{E(\Ga)},

6524:  d \vec x).

6525:  \end{equation}

6526: 

6527: This function is invariant with respect to the action on

6528: $\SU(2)^{E(\Ga)} = \sA $ of the gauge group on $\Ga$  $

6529: \SU(2)^{V(\Ga)}=\prod_{v\in V(\Ga)} \SU(2)_v$ with components

6530: enumerated by the vertices and an orientation of $\Ga$.

6531: 

6532: It is easy to see that the endomorphism $B(\Ga_j)$ (5.96) is an

6533: intertwiner with respect to such action. Then by the standard

6534: result of harmonic analysis on groups our function $f_{\Ga_{j}}$

6535: is invariant with respect to

6536:  this action. That is, $f_{\Ga_{j}}$ is a function on the homogeneous

6537: space

6538:  \begin{equation}

6539:  uS_g = \sA  / \sG.

6540:  \end{equation}

6541: 

6542: But the set $\{ \Ga_j\} = SNW^k(\Ga) = W^k(\Ga)$ enumerates the

6543: collection

6544:  of functions $f_{\Ga_j}$ (6.115) each of which is an analog of the

6545:   function

6546:  (2.48) enumerated by the set $(\Z^g / k\Z^g)$ of theta characteristics.

6547: 

6548:  \begin{prop} Graph $\Ga$ and  function $f_{\Ga_j}$ determine

6549:   spin network

6550:  $\Ga_j$ uniquely.

6551:  \end{prop}

6552:  Indeed,

6553:  \begin{enumerate}

6554:  \item knowing  graph we can reconstruct the map $P \colon \sA \to uS_g$;

6555:  \item lifting  function $f_{}\Ga_j$ on $\sA = \SU(2)^{E(\Ga)}$ and using

6556:  the harmonic decomposition of this function on $\sA$

6557:   (6.61)

6558:  we determine  the representation $j_1 \otimes \dots \otimes j_{3g-3}$ and

6559:  the space $V$ of this representation;

6560:  \item  by  the Peter-Weyl formula we determine the endomorphism $B(\Ga_j)$

6561:   (6.114);

6562:   \item decomposing this endomorphism on blocks (6.113)

6563:   correspoding to

6564:    vertices we reconstruct the function $j$ (6.4).

6565:    \end{enumerate}

6566:  and we are done.

6567: 

6568:  Now we are ready to construct  CST for the pair $uS_g \subset S_g$

6569:  (see the end of 3.2) since the complex Schottki space $S_g$ is

6570:  the complexification of $uS_g$.

6571: 

6572: 

6573: 

6574: 

6575: 

6576: \subsection{RCFT  and Kohno representation}

6577: 

6578: 

6579: The usual data for RCFT include an algebra $A$ and

6580: a discrete set $R$ of representations

6581:  (a finite set for level $k$) $\{ \rho_i \}$

6582: (just like in subsection 5.1) such that

6583: \begin{equation}

6584: \rho \in R \implies \rho^* \in R,

6585: \end{equation}

6586: and a distinguished element $1$ such that $1^* = 1$. Moreover for every

6587: ordered triple $\rho_i, \rho_j, \rho_k$ we have a finite (rationality of RQFT)

6588: dimensional space $V^{\rho_i}_{\rho_j, \rho_k}$.

6589: \begin{equation}

6590: dim V^1_{\rho_i, \rho_j} = 1 \quad \text{ if } \quad \rho_i =

6591: \rho_j^*

6592: \end{equation}

6593: and 0 otherwise.

6594: 

6595: \begin{rmk} So  we can extract a {\it modular tensor category}

6596: from RCFT.

6597: \end{rmk}

6598: 

6599: Having such $RCFT$ and 3-1-valent graph $\Ga$ and a spin network

6600: $\Ga_j$ we get from every vertex $v \in V(\Ga)$ the spaces with

6601: edges $e_1, e_2, e_3$ through it thus we have the space of

6602: intertwiners

6603:  \begin{equation}

6604: V^{j_1}_{j_2, j_3} \in j_1 \otimes j_2 \otimes j_3

6605: \end{equation}

6606: (like in (6.6)). Taking tensor products of them together over all

6607: vertices

6608:  we get the space

6609:  \begin{equation}

6610: \sH^k_{\Ga_j} = \otimes_{v \in V(\Ga)} V^{j_1}_{j_2, j_3}

6611: \end{equation}

6612:  and taking a direct product over all possible networks (labelings) we get

6613:  the space

6614:  \begin{equation}

6615: \sH^k_{\Ga} = \oplus_{j \in SNW^k(\Ga)} \sH^k_{\Ga_j}.

6616: \end{equation}

6617: For $\SU(2)$-case we have

6618: \begin{enumerate}

6619: \item $ dim V^{j_1}_{j_2, j_3} = 1$;

6620: \item $ dim \sH^k_{\Ga_j} = 1$;

6621: \item the space (6.121) is precisely our space (4.39), (4.43)

6622:  of wave functions under

6623: Bohr-Sommerfeld quantization.

6624: \end{enumerate}

6625: It is useful to fix bases in all our spaces. (In $\SU(2)$-case

6626: where all spaces are 1-dimensional we can do it using the

6627: standard Wigner 3j-symbols (for recalling and quantum groups

6628: extention see \cite{KR}). We equipe

6629:  the wave function spaces with  bases.

6630: 

6631:  This construction hasn't

6632: to be invariant with respect to changes in trinion decompositions

6633:  (or changes of bases). But having the complex $TG_g$ we know

6634:  the obstackles.

6635: 

6636:  To get such invariance we have to produce matrices

6637:  of transformations to new bases and check that these matrices satisfy the

6638:  same algebraic identities as the changes of decompositions themselves.

6639:  We have to check that these matrices satisfy the same algbraic

6640:  equations as

6641:  our transformations itselves (see Fig.11, Definition 9 and 1)-4) around (6.47)).

6642: 

6643:  Fortunately there are 5 elementary moves only described in \cite{MS} with

6644:   traditional symbols

6645:  \begin{equation}

6646: F, \quad S, \quad, T, \quad, \Om, \quad , \Theta

6647: \end{equation}

6648: satisfying the required algebraic relations.

6649: 

6650: 

6651: Recall that elementary transformations of 3-valent graphs admit

6652: relations 1)-4) around (6.47). So our intertwiners vector spaces must

6653: satisfy the  system of identities described below:

6654: \begin{enumerate}

6655: \item  3 transformations are diagonal matrices of phases (6.18)

6656:  (in an appropriate basis)

6657: \begin{equation}

6658: \Om^{\pm}, \quad \Theta, \quad T \colon  V^{\rho_i}_{\rho_j, \rho_k}

6659: \to V^{\rho_i}_{\rho_j, \rho_k};

6660: \end{equation}

6661: \item "genus 1 equality "

6662: \begin{equation}

6663: S \colon V^{\rho_i}_{\rho_j, \rho_j^*} \to

6664: \sum_l V^{\rho_i}_{\rho_l, \rho_l^*};

6665: \end{equation}

6666: \item the fusing matrix

6667: \begin{equation}

6668: F \colon V^{\rho_i}_{\rho_j, \rho_k} \otimes V^{\rho_k}_{\rho_t, \rho_m} \to

6669: \sum_{l} V^{\rho_j}_{\rho_t, \rho_l} \otimes V^{\rho_l}_{\rho_j, \rho_m};

6670: \end{equation}

6671: \item the braiding matrix

6672: \begin{equation}

6673: B^\pm = F \Om^\pm F^{-1}

6674: \end{equation}

6675: subjecting the ordinary Reidemeister relations

6676: \begin{equation}

6677: B_{12} B_{23} B_{12} = B_{23} B_{12} B_{23}.

6678: \end{equation}

6679: \end{enumerate}

6680: We can see that the most general topological (geometrical) setting

6681: for applying of discussed TQFT is a 3-manifold containing a

6682: 3-valent graph (instead a loop = knot). So the most striking

6683: difference between such objects and spin networks is following:

6684: in spin networks the holonomy of one edge around another is

6685: trivial that is any "braiding" relations are absent. In spin

6686: networks it isn't necessary to distinguish over and under crossing

6687: when a 3-valent graph is projected on a plane. Otherwise all

6688: rules for evaluating spin networks closely parallel the skein

6689: rules for the Jones polynomials. Here we are translating RCFT

6690: setting to Penrose's interpretation of the quantum gravity.

6691: 

6692: Recall that all braid properties of all matrices in the game are

6693: arising from the holonomy of the Knizhnik-Zamolodchikov equation

6694: (see \cite{KZ}) and the polynimial equations for these matrices

6695: expected from the Conformal Field Theory were proposed by Moore

6696: and Seiberg in \cite{MS}.

6697: 

6698: Consider a 1-3-valent graph $\Ga$ which corresponds to a  trinion

6699: decomposition of a pinched Riemann sphere $ \Si = \PP^{1} - p_{1}

6700: - \dots - p_{n} $.

6701: 

6702: Let

6703: \begin{equation}

6704: \{I_{\mu}\}, \quad \mu = 1,2,3

6705: \end{equation}

6706: be any orthonormal basis of $sl(2,\C)$ with respect to the

6707: Cartan-Killing form and

6708: \begin{equation}

6709: \Om = \sum_{\mu} I_{\mu} \otimes I_{\mu} \in End (1) =

6710: End(sl(2,\C)).

6711: \end{equation}

6712: Let $\pi_{i}(I_{\mu})$ be the action on i-th component of the

6713: tensor product $j_{1} \otimes \dots \otimes j_{n}$ (of some

6714: labeling of the graph) and

6715: \begin{equation}

6716: \Om_{ij} = \sum_{\mu} \pi_{i}(I_{\mu}) \pi_{j}(I_{\mu}) \in

6717: End(j_{1} \otimes \dots \otimes j_{n}).

6718: \end{equation}

6719: Consider the configuration space

6720: \begin{equation}

6721: \C^{n} - \De = \{(z_{1}, \dots, z_{n}) \vert i \neq j \implies

6722: z_{i} \neq z_{j}\}

6723: \end{equation}

6724: and the trivial vector bundle on it with the fiber $ j_{1} \otimes

6725: \dots \otimes j_{n}$. It is easy to check that the KZ-equation

6726: \begin{equation}

6727: \p \Phi / \p z_{i} = \frac{1}{k+2} \sum_{i \neq j}

6728: \frac{\Om_{ij}}{z_{i}-z_{j}} \Phi, \quad 1 \leq i \leq n

6729: \end{equation}

6730: defines a slightly projective flat connection

6731: \begin{equation}

6732: \om = \frac{1}{k+2} \sum_{i < j} \Om_{ij} d log(z_{i}-z_{j})

6733: \end{equation}

6734: on this trivial bundle. Any solution of the KZ-equation gives a

6735: {\it covariant constant} (horizontal) section of this bundle.

6736: 

6737: Using the detailed information on the monodromy of this equation

6738: Tsuchiya and Kanie proposed the operator formalism  two

6739: dimensional conformal field theory in \cite{TK}. We can use this

6740: information too but we would like to present the matrices of

6741: MS-axiomatic description in more background independent way.

6742: 

6743: First of all consider the sphere with 4 holes $\Si_{0,4}$ and

6744:  two graphs $\Ga$ and $\Ga'_{e}$ corresponding to two types of

6745:   trinion decompositions of it (see Fig.11).

6746: 

6747: Symbolically we can note every labeling of $\Ga$ by the matrix

6748:  \begin{equation}

6749: \begin{pmatrix}

6750: j_{2}&&&&j_{3}\\

6751: &&j&&\\

6752: j_{1}&&&&j_{4}

6753: \end{pmatrix}

6754:  \end{equation}

6755: Then for the graph $\Ga'_{e}$ the matrix has the form

6756:  \begin{equation}

6757: \begin{pmatrix}

6758: j_{3}&&&&j_{4}\\

6759: &&j'&&\\

6760: j_{2}&&&&j_{1}

6761: \end{pmatrix}

6762:  \end{equation}

6763: Remark that the sets of parabolic edges are the same

6764: \begin{equation}

6765: P(\Ga) = P(\Ga'_{e}) = e_{1}, e_{2}, e_{3}, e_{4}

6766: \end{equation}

6767: but the subsets $E_{v}$ are different. Thus we have two spaces

6768: \begin{equation}

6769: \sH^{k}_{\Ga, j_{1}, j_{2}, j_{3}, j_{4}} = Hom_{\SU(2)}(j_{1}

6770: \otimes j_{2}, j_{3} \otimes j_{4})

6771: \end{equation}

6772: (6.3) with the basis given by the set of weights $ W^{k} (\Ga,

6773: j_{1}, j_{2}, j_{3}, j_{4}) $ (6.1) and

6774: \begin{equation}

6775: \sH^{k}_{\Ga'_{e}, j_{1}, j_{2}, j_{3}, j_{4}} =

6776: Hom_{\SU(2)}(j_{2} \otimes j_{3}, j_{4} \otimes j_{1})

6777: \end{equation}

6778: with the basis $ W^{k} (\Ga'_{e}, j_{1}, j_{2}, j_{3}, j_{4}) $.

6779: 

6780: Obviously these spaces of (conformal) blocks coincide but the

6781: bases are different.

6782: 

6783: \begin{dfn}

6784: The matrix

6785: \begin{equation}

6786: F_{e} \colon \sH^{k}_{\Ga, j_{1}, j_{2}, j_{3}, j_{4}} =

6787: Hom_{\SU(2)}(j_{1} \otimes j_{2}, j_{3} \otimes j_{4}) \to

6788: \sH^{k}_{\Ga'_{e}, j_{1}, j_{2}, j_{3}, j_{4}} =

6789: Hom_{\SU(2)}(j_{2} \otimes j_{3}, j_{4} \otimes j_{1})

6790: \end{equation}

6791: sending the first basis to the second is called the fusion matrix.

6792: \end{dfn}

6793: 

6794: Let the symbol of the first basis be

6795: \begin{equation}

6796:  \begin{pmatrix}

6797: j_{2}&j_{3}\\

6798:  j_{1}&j_{4}

6799: \end{pmatrix}

6800: \end{equation}

6801: and of the second be

6802: \begin{equation}

6803:  \begin{pmatrix}

6804: j_{3}&j_{4}\\

6805:  j_{2}&j_{1}

6806: \end{pmatrix}

6807: \end{equation}

6808: Then the fusion matrix

6809: \begin{equation}

6810: F \begin{pmatrix}

6811: j_{2}&j_{3}\\

6812:  j_{1}&j_{4}

6813: \end{pmatrix} = \begin{pmatrix}

6814: j_{3}&j_{4}\\

6815:  j_{2}&j_{1}

6816: \end{pmatrix}

6817: \end{equation}

6818: can be considered as a transformation of solutions of KZ-equation

6819: with coefficients

6820: \begin{equation}

6821: F_{ij} \begin{pmatrix}

6822: j_{2}&j_{3}\\

6823:  j_{1}&j_{4}

6824: \end{pmatrix}

6825: \end{equation}

6826: In these notations it is easy to check the following equalities

6827: \begin{enumerate}

6828: \item \begin{equation}

6829: \sum_{j} F_{ij}\begin{pmatrix}

6830: j_{2}&j_{3}\\

6831:  j_{1}&j_{4}

6832: \end{pmatrix} F_{jk}\begin{pmatrix}

6833: j_{3}&j_{4}\\

6834:  j_{2}&j_{1}

6835: \end{pmatrix} = \delta_{ik};

6836: \end{equation}

6837: \item \begin{equation}

6838: F_{ij}\begin{pmatrix}

6839: j_{2}&j_{3}\\

6840:  j_{1}&j_{4}

6841: \end{pmatrix} = F_{ij}\begin{pmatrix}

6842: j_{4}&j_{1}\\

6843:  j_{3}&j_{2}

6844: \end{pmatrix}

6845: \end{equation}

6846: \item for $\Si_{0,5}$ five fusion matrices are subjecting the

6847: pentagone equation (see 3) below (6.47)).

6848: \end{enumerate}

6849: 

6850: Now using fusion matrices and  simply connectedness   of $C(TG)$

6851: from 5.2 we can identify uniquely all spaces (4.39), (4.43) and

6852: (6.121).

6853: 

6854: Using the same symbols we can define the collection of {\it

6855: braiding matrices}. Again we can describe it as monodromies of

6856: covariant constant solutions to the KZ-equation. Symbolically we

6857: have the chain of transformations

6858: \begin{equation}

6859: \begin{pmatrix}

6860: j_{2}&j_{3}\\

6861:  j_{1}&j_{4}

6862: \end{pmatrix} \to \begin{pmatrix}

6863: j_{1}&j_{3}\\

6864:  j_{2}&j_{4} \end{pmatrix} \xrightarrow{ F } \begin{pmatrix}

6865: j_{3}&j_{4}\\

6866:  j_{1}&j_{2}

6867: \end{pmatrix} \to \begin{pmatrix}

6868: j_{3}&j_{2}\\

6869:  j_{1}&j_{4}

6870: \end{pmatrix}

6871: \end{equation}

6872: \begin{dfn} The composition of these transformations

6873:  is called a braid matrix.

6874: \end{dfn}

6875: Again we denote coefficients of this matrix by the symbol

6876: \begin{equation}

6877: B_{ij}\begin{pmatrix}

6878: j_{2}&j_{3}\\

6879:  j_{1}&j_{4}

6880: \end{pmatrix}

6881: \end{equation}

6882: This braiding matrix can be diagonalized by the fusion matrix:

6883: \begin{equation}

6884: B \begin{pmatrix}

6885: j_{2}&j_{3}\\

6886:  j_{1}&j_{4}

6887: \end{pmatrix} = F \begin{pmatrix}

6888: j_{2}&j_{3}\\

6889:  j_{1}&j_{4}

6890: \end{pmatrix}^{-1} D F \begin{pmatrix}

6891: j_{2}&j_{3}\\

6892:  j_{1}&j_{4}

6893: \end{pmatrix}

6894: \end{equation}

6895: where $D = \{d_{i}\}$ is the diagonal matrix with elements

6896: \begin{equation}

6897: d_{i} = (-1)^{j_{2} + j_{3} - i} e^{\pi i (\frac{i(i+1)}{k+2} -

6898: \frac{j_{2}(j_{2}+1)}{k+2} - \frac{j_{3}(J_{3}+1)}{k+2})}.

6899: \end{equation}

6900: Moreover

6901: \begin{equation}

6902: B_{ij} \begin{pmatrix}

6903: j_{2}&j_{3}\\

6904:  j_{1}&j_{4}

6905: \end{pmatrix} = (-1)^{j_{1} + j_{4} - i - j} e^{\pi i (-1)

6906: (\frac{i(i+1)}{k+2} + - \frac{j(j+1)}{k+2} -

6907: \frac{j_{1}(j_{1}+1)}{k+2} - \frac{j_{4}(J_{4}+1)}{k+2})} F_{ij}

6908: \begin{pmatrix}

6909: j_{1}&j_{3}\\

6910:  j_{2}&j_{4}

6911: \end{pmatrix}

6912: \end{equation}

6913: All of these equalities can be checked using monodromies of

6914: KZ-equation for $n = 4$ (or for $n = 3$ plus $\infty$ as in

6915: \cite{K1}).

6916: 

6917: Using braid matrix $B$ we can defined matrices $\Om^{\pm}$ (6.123)

6918: by the formula (6.127).

6919: 

6920: For every edge $e \in E(\Ga)$ we have the diagonal transformation

6921: of the space $\sH^{k}_{\Ga}$ (4.41) given in the basis $\{w\}$

6922: (4.39) by the formula

6923: \begin{equation}

6924: T_{e}(w) = e^{2\pi i (\frac{w(e)(w(e) + 1)}{k+2} -

6925: \frac{k}{8(k+2)})} w.

6926: \end{equation}

6927: 

6928: 

6929: Now consider the first "elliptic" elementary transformation in

6930: Fig.11. This transformation preserves the shape of the graph

6931: $\Ga_{1,1}$ but changes "a" and "b" standard generators of the

6932: fundamental group

6933:  of 1-torus. This graph has one loop edges $a$, one 3-valent vertex

6934: $v$ and one parabolic edge $p$ and corresponds to 1-holed torus.

6935: For 2-holed torus  the graph $\Ga_{1,2}$ has two edges $a$ and

6936: $a'$, two vertices $v_{1}$ and $v_{2}$ and two corresponding

6937: parabolic edges $p_{1}$ and $p_{2}$.

6938: 

6939: 

6940: So we have the spaces (6.120)

6941: \begin{equation}

6942: \sH^{k}_{\Ga_{1,1}, j} \quad \text{ and } \sH^{k}_{\Ga_{1,2},

6943: j_{1}, j_{2}}.

6944: \end{equation}

6945: (of course the number $j$ has to be integer in this case).

6946: 

6947: Using fusion matrices around edges "a" we construct the

6948: isomorphism

6949: \begin{equation}

6950: f \colon \sH^{k}_{\Ga_{1,2}, j_{1}, j_{2}} \to \oplus_{j}

6951: \sH^{k}_{\Ga_{1,1}, j}

6952: \end{equation}

6953: where $j$  satisfies the Clebsh-Gordan conditions with $j_{1}+

6954: j_{2} + j \leq k$. The source space admits the basis $w_{kl}$

6955: which corresponds  to the admissible weights $w(a) = k$, $w(a') =

6956: l$. In this basis using braid matrix we define the linear operator

6957: \begin{equation}

6958: \sB (w_{kl}) = \sum_{n} B_{nk} \begin{pmatrix}

6959: j_{1}&j_{2}\\

6960:  l & l

6961: \end{pmatrix} w_{nl}

6962: \end{equation}

6963: on the space $\sH^{k}_{\Ga_{1,2}, j_{1}, j_{2}}$.

6964: 

6965: Now we are ready to define {\it switching operators} representing

6966: the action of the modular group $Mod_{1,1}$ on the space

6967: $\sH^{k}_{\Ga_{1,1}, j}$. More precisely, it is the diffeomerphism

6968: of the elliptic elementary transformation in Fig. 11.

6969: 

6970: 

6971: 

6972: \begin{dfn}

6973: An endomorphism

6974: \begin{equation}

6975: S_{k}(j) \in End \sH^{k}_{\Ga_{1,1}, j}

6976: \end{equation}

6977: is called a  switching operator if the following 3 conditions are

6978: satisfied

6979: \begin{enumerate}

6980: \item \begin{equation}

6981: S_{k}(j)^{2}= (-1)^{j} e^{\pi i \frac{j(j+1)}{k+2}} \circ id;

6982: \end{equation}

6983: \item

6984: \begin{equation}

6985: (S_{k}(j) T_{e=a})^{3} = S_{k}(j)^{2};

6986: \end{equation}

6987: \item let \begin{equation}

6988: S_{k}(j_{1}, j_{2}) = f^{-1} \circ (\oplus_{j}S_{k}(j)) \circ f

6989: \end{equation}

6990: be the endomorphism of $\sH^{k}_{\Ga_{1,2}, j_{1}, j_{2}}$. Then

6991: \begin{equation}

6992: S_{k}(j_{1, j_{2}}) \circ T_{e=a}^{-1} \circ T_{e=a'}\circ

6993: S_{k}(j_{1}, j_{2})^{-1} = \sB.

6994: \end{equation}

6995: \end{enumerate}

6996: \end{dfn}

6997: (Geometrically the operator $S_{k}(j_{1}, j_{2})$ has to be an

6998: intertwiner of diffeomerphsims corresponding to sliding the hole

6999: $p_{2}$ along "a" and "b" paths on 2-torus).

7000: 

7001: The existence of solutions for small $k = 1, 2, 3$ can be varified

7002:  by elementary computations and for general $k$ can be obtained

7003:  from $\SU(2)$-WZW model over 2-holed torus (see for example

7004:  \cite{TUY}).

7005: 

7006:  \begin{rmk}

7007:  We can see that for a closed graph (without parabolic edges) we

7008:  are using braiding matrices in the equality (6.160) only. Moreover

7009:  in this equality we are using the combination $\sB$ of braiding

7010:  matrices. The geometrical meaning of this observation is the

7011:  following: the construction of switching operators almost

7012:  independent on over and under crossings when 3-valent graph is

7013:  projected on a plane. So we can hope that in high genus closed

7014:  case $\sM_{g}$ evaluation for spin networks and ones for RCFT

7015:  coincide.

7016:  \end{rmk}

7017: 

7018: Now following Kohno \cite{K1} and using the fusion matrices $F$

7019: (6.143), switching operators $S$ (6.156) and diagonal operators

7020: $T_{e}$ (6.152) we construct the slightly projective

7021: representation (6.54).

7022: 

7023: 

7024: \begin{figure}[tbn]

7025: \centerline{\epsfxsize=3in\epsfbox{6c.eps}} \caption{\sl Graphs $\Theta_g$ and

7026:  $\Ga_{g0}$} \label{Fig 14}

7027: \end{figure}

7028: 

7029: 

7030: 

7031: 

7032: Consider  multi theta graph $\Theta_{g-1}$ of genus $g$ as in

7033: \cite{T3} (see Figure 18). On the Riemann surface

7034: $\Si_{\Theta_{g-1}}$ there exists the collection of simple circles

7035: \begin{equation}

7036: a_{1}, \dots, a_{g}, b_{1}, \dots, b_{g}, \de_{2}, \dots,

7037: \de_{g}, \ep_{2}, \dots, \ep_{g-1}

7038: \end{equation}

7039: as in Fig.12b in \cite{K1}. The collection of circles

7040: \begin{equation}

7041: a_{1}, \dots, a_{g}, \de_{2}, \dots, \de_{g}, \ep_{2}, \dots,

7042: \ep_{g-1}

7043: \end{equation}

7044: is the collection (4.7) of disjoint inequivalent circles. These

7045: curves correspond to edges of the graph $\Theta_{g-1}$. The

7046: identification of these circles with corresponding edges give us

7047: the collection of diagonal operators

7048: \begin{equation}

7049: T_{a_{1}}, \dots, T_{a_{g}}, T_{\de_{2}}, \dots, T_{\de_{g}},

7050: T_{\ep_{2}}, \dots, T_{\ep_{g-1}}

7051: \end{equation}

7052: (6.152) on the space $\sH^{k}_{\Theta_{g-1}}$ (4.39), (4.43).

7053: 

7054: These operators represent the Dehn twists around the circles

7055: (6.162).

7056:  But it isn't enough to generate full $Mod_{g}$. We have to add

7057: the Dehn twists around $\{ b_{i} \}$. To represent these twists

7058:  we have to change the graph.

7059: 

7060: Consider the graph $\Ga_{g0}$ (see Fig. 18) which is a chain of circles $C_{1},

7061: \dots, C_{g}$ joint by chain of consequent edges $ B_{1}, \dots,

7062: B_{g-1}$ such that $B_{i}$ joints $C_{i}$ and $C_{i+1}$. Then

7063: circles $C_{2}, \dots, C_{g-1}$ contain two edges

7064: \begin{equation}

7065: C_{i} = C'_{i} \cup C''_{i}.

7066: \end{equation}

7067: such that the set of edges

7068: \begin{equation}

7069: E(\Ga_{g0}) = C_{1}, C_{g}, C'_{2}, \dots, C'_{g-1}, C''_{2},

7070: \dots, C''_{g-1}, B_{1}, \dots, B_{g-1}.

7071: \end{equation}

7072: 

7073:  On

7074: the Riemann surface $\Si_{\Ga_{g0}}$ there are circles which

7075: correspond to edges $\{ B_{i} \}$. We cut this surface along

7076: these circles to obtain two 1-holed tori $\Si_{1,1}^{1}$ and

7077: $\Si_{1,1}^{g}$ and $g-2$ 2-holed tori $\{\Si_{1,2}^{i}\}, \quad

7078: i= 2,\dots, g-1$. The 1-holed torus $\Si_{1,1}^{1}$ conains the

7079: curve $b_{1}$ and $\Si_{1,1}^{g}$ contains $b_{g}$ from the

7080: collection (6.44). On the

7081:  2-holed torus $\Si_{1,2}^{i}$ we have the curves $ b_{i}, \quad

7082:  i = 2,\dots, g-1$. The operator corresponding to the Dehn twist

7083:   about $b_{i}$ acts on the correspoding block of the 1-holed and

7084:    2-holed tori (6.153). So we have the collection of representatives

7085:     of

7086:    these Dehn twists

7087: \begin{equation}

7088: T_{A_{1}} \circ S_{k}(j(B_{1})) \circ T_{A_{1}}, T_{A''_{2}} \circ

7089: S_{k}(j(B_{1}), j(B_{2})) \circ T_{A''_{2}}, \dots,

7090: \end{equation}

7091: $$

7092: T_{A''_{g-1}} \circ S_{k}(j(B_{g-2}), J(B_{g-1})) \circ

7093: T_{A''_{g-1}}, T_{A_{g}} \circ S_{k}(j(B_{g-1})) \circ T_{A_{g}}.

7094: $$

7095: (We identify the space $\sH^{k}_{\Theta_{g-1}}$ to

7096: $\sH^{k}_{\Ga_{g0}}$ with the basis (4.39) enumerated by spin

7097: networks $\{\Ga_{g0}\}$ of level $k$).

7098: 

7099: Using the description of generators of $Mod_{g}$  and

7100: relations in \cite{Wa} we can check that all these operators give

7101: the slightly projective representation (5.38) (see \cite{K1}).

7102: 

7103: So now we have two slightly projective representations (5.37) and

7104: (5.38) corresponding to two versions of CFT WZW and MS-axiomatic

7105: vertion. Fibers of the corresponding flat vector bundles are wave

7106: function spaces for two different kind of polarizations: complex

7107:  and real. In paper \cite{K1} Kohno considered his version of

7108:  CFT as the combinatorial description of the space of conformal

7109:  blocks that is the space of holomorphic sections (3.127).

7110: But the direct identification can be constructed at the end of

7111: this text only.

7112: 

7113: 

7114: \section{Analytical aspect of the theory of

7115:  non abelian theta functions}

7116: 

7117: \subsection{The unitary Schottky space}

7118: 

7119: 

7120: 

7121: This section is absolutely parallel to the subsection 2.4. The

7122: foundation for the following constructions is  preprint

7123: \cite{FMNT}.

7124: 

7125: The basis of the space $\sH^{k}_{\Ga}$ (6.3) is enumerated by

7126:  the set of spin networks corresponding to the set of functions

7127:  $\{ f_{\Ga_{j}}\}$ on the unitary Schottki space $uS_{g}$

7128:  (6.110). On the other hand the unitary Schottky space is the fiber

7129:  of the real polarization (4.13):

7130:  \begin{equation}

7131:  \pi_{\Ga}^{-1}(0, \dots , 0) = uS_{g}.

7132:  \end{equation}

7133: 

7134: The unitary Schottky space $uS_g = \SU(2)^{g} /

7135:  Ad_{\mathrm{diag}} \SU(2)$ is a singular manifold. To avoid this

7136: problem we will consider geometrical objects on it as the

7137: corresponding objects on the direct product $\SU(2)^{g}$

7138: invariant with respect to

7139:  $ Ad_{\mathrm{diag}} \SU(2)$-action.

7140:  Our group $\SU(2)$ is embedded as a real part into its

7141:   complexification

7142: $\SL(2, \C)$. The principal fact is that the theory of finite

7143: dimensional (non unitary) representations of $\SL(2, \C)$

7144: completely the same answers to all questions. Obviously

7145: \begin{equation}

7146:  \widehat{SU(2)}= \widehat{\SL(2, \C)} =  \half \Z^{+}

7147:  \end{equation}

7148: and we can note both type of representations by the same symbols

7149: $j \in \half \Z^{+}$. Remark that all spaces $V_j$ of

7150: representations of both types of groups are precisely the same,

7151: $rk V_j = (2j+1)$. Moreover all intertwiners in tensor

7152:  products of representations are the same. But of course $\SL(2,

7153: \C)$-action doesn't preseves the inner product.

7154: 

7155: So here for convenience of the reader and also to establish

7156: notation we start by briefly recalling   from \cite{Ha} the

7157: coherent state transform for Lie groups.

7158: 

7159: \subsection{$g$-extended Hall's construction of  CST for $\SU(2)$}

7160: 

7161: Using the heat kernel $\rho_{t}$ for a Laplacian $\Delta$ on

7162: $\SU(2)$ associated with a given bi-invariant metric  B. C. Hall

7163: constructed in \cite{Ha} the map

7164: $$

7165: C_{t}  :  L^{2}(\SU(2),dx)\mapsto {\cal H}(\SL(2, \C))

7166: $$

7167: \begin{equation}

7168: C_{t}(f)(z)  =  \int_{\SU(2)} f(x) \rho_t (x^{-1} z)dx, \qquad

7169:  f\in L^{2}(\SU(2),dx), z\in \SL(2, \C)

7170: \end{equation}

7171: where $dx$ is the normalized Haar measure on $\SU(2)$ and  ${\cal

7172: H}(\SL(2, \C))$ is the space of holomorphic functions on $\SL(2,

7173: \C)$. Recall $\rho_t$ has  unique analytic continuation to

7174: $\SL(2, \C)$ also denoted by $\rho_t$. This map is called

7175: $\SU(2)$-averaged coherent state transform and for each $f\in

7176: L^{2}(\SU(2),dx)$, the function $C_{t}f(x)$ is just the solution

7177: of the heat equation,

7178: \begin{equation}

7179: \frac{\partial  u}{\partial t} = \frac{1}{2}\Delta u,

7180: \end{equation}

7181: with initial condition given by $u(0,x)=f(x)$. Therefore,

7182: $C_{t}(f)$ is given by

7183: \begin{equation}

7184: C_{t}f(z) = ({\cal C}\circ \rho_t \star f)(z) = \left({\cal

7185: C}\circ e^{t\frac{\Delta}{2}}f \right)(z),

7186: \end{equation}

7187: where $\star$ denotes the convolution in $\SU(2)$ and ${\cal C}$

7188: denotes

7189: analytic continuation from $\SU(2)$ to $\SL(2, \C)$.

7190: 

7191: Again for every $f\in L^{2}(\SU(2),dx)$ there exists  the

7192: expansion given by the Peter-Weyl's theorem:

7193: \begin{equation}

7194: f(x) = \sum_{j \in \widehat{\SU(2)}} \mbox{tr} (j (x)B_{j ,f} )

7195: \end{equation}

7196: where the sum is taken over the set of equivalence classes of

7197: irreducible representations of $\SU(2)$ and $B_{j ,f}\in End \

7198: V_j$ is given by

7199: \begin{equation}

7200: B_{j ,f} = \left( \dim V_j \right) \int_{\SU(2)} f(x) j

7201: (x^{-1})dx,

7202: \end{equation}

7203: $V_j$ being the representation space for $j$. Then we obtain:

7204: \begin{equation}

7205: C_{t}f(z) = \sum_{j \in \widehat{\SU(2)}}

7206: e^{-t\frac{\lambda_{j}}{2}}\mbox{tr} (j (z)B_{j ,f } )

7207: \end{equation}

7208: where $\lambda_{j}$ is the eigenvalue of $-\Delta$ on functions

7209: of the type

7210: $$\mbox{tr} (B_j(x) ), \quad B\in End(V_j).

7211: $$

7212: Remark that in this formula $j \in \widehat{\SL(2, \C)}$ but $B_j

7213: $ is the same  endomorphism of the same space as for $j \in

7214: \widehat{\SU(2)}$.

7215: 

7216: 

7217: It turns out that there exists a natural extension of $C_{t}$ to

7218: appropriately chosen distributions on $\SU(2)$ and we get

7219: holomorphic functions on $\SL(2, \C)$ corresponding to

7220: holomorphic sections of line bundles on it.

7221: 

7222: This program can be extended to $\SU(2)^g \subset \SL(2, \C)^g$

7223: wordwisely. Moreover all constructions can be decented to

7224: quotients

7225: \begin{equation}

7226: uS_g = \SU(2)^{g} /  Ad_{\mathrm{diag}} \SU(2), \quad S_g =\SL(2,

7227: \C)^g/  Ad_{\mathrm{diag}} \SL(2, \C).

7228: \end{equation}

7229: 

7230: A natural $g$-extension of Hall's construction of  CST can be done

7231: immediately but an extention of the previous construction to the

7232: complexification

7233: \begin{equation}

7234: uS_g = \SU(2)^{g} /  Ad_{\mathrm{diag}} \SU(2) \subset S_g

7235: =\SL(2, \C)^g/  Ad_{\mathrm{diag}} \SL(2, \C)

7236: \end{equation}

7237: admits new parameters. We are starting  by precise description of

7238: them.

7239: 

7240:  Let  $\Omega = (\Omega_{\al \be})_{\al , \be = 1}^g$ be a complex

7241: symmetric matrix with  positive imaginary part that is a point of

7242: Siegel upper half space $H$ and

7243: \begin{equation}

7244: \Omega = Re \Omega +  i\cdot Im \Omega

7245: \end{equation}

7246: be the decomposition on real and imagine  parts.

7247: 

7248: Consider on $uS_g$ the invariant Laplacian given by

7249: \begin{equation}

7250: \Delta^{(im \Omega)} = \sum_{\al, \be = 1}^g \frac{im \Omega_{\al

7251: \be}}{2\pi i}  \frac{\partial^2 }{\partial x_\al \partial x_\be}

7252: \end{equation}

7253: where $x_\al,x_\be $ are invariant vector fields on $\SU(2)$.

7254: 

7255: Let $d\nu_t^{(im \Omega)}$ be the averaged heat kernel mesure on

7256: $uS_g$ corresponding to the heat equation

7257: \begin{equation}

7258: \frac{\partial u}{\partial t} = \frac{1}{ 4}\Delta_\C^{(im

7259: \Omega)} u  = \sum_{\al , \be = 1}^n \frac{im \Omega_{\al \be}}{8

7260: \pi }\left( \frac{\partial^2}{\partial x_\al \partial x_\be } +

7261: \frac{\partial^2 }{

7262: \partial y_\al \partial y_\be }\right) u

7263: \end{equation}

7264: that is if $\mu_t^{(im \Omega)}$ denotes the fundamental solution

7265: of this heat equation then $d\nu_t^{(im \Omega)}$ is obtained by

7266: averaging $\mu_t^{(im \Omega)} dw$ with respect to the left

7267: action of $uS_g$ where $dw\equiv dxdy$ is the Haar measure on

7268: $S_g$. We get the transform

7269: $$

7270: C_t^{(im \Omega)} : L^2(uS_g,dx) \rightarrow  \sH (S_g)\cap

7271: L^2((S_g, d\nu_t^{(im \Omega)})

7272: $$

7273: \begin{equation}

7274: C_t^{(im \Omega)}(f)(w) = \left( {\cal C} \circ e^{\frac{t}{

7275: 2}\Delta^{(im \Omega)}}f \right) (w) \quad , \ t > 0.

7276: \end{equation}

7277: Now consider  non-self-adjoint Laplace operator on $uS_g$ of the

7278: form

7279: \begin{equation}

7280: \Delta^{(-i\Omega )} = - \sum_{\al, \be = 1}^g \frac{i }{ 2 \pi

7281: }\Omega_{{\al \be}} \frac{\partial^2}{ \partial x_\al \partial

7282: x_\be}

7283: \end{equation}

7284: It is easy to see that the imaginary part of this Laplacian does

7285: not affect the unitarity properties of the extended Hall's CST

7286: defined in this way. Therefore we have the following

7287: \begin{prop} For any $\Omega\in H_g$ and $t>0$ the transform

7288: $$

7289: C_t^{(-i\Omega)}  ={\cal C} \circ e^{\frac{t }{ 2}

7290: \Delta^{-i(\Omega)}} \colon L^2(uS_g,dx) \rightarrow \sH (S_g)

7291: \cap L^2((S_g,(d\nu_t^{(im \Omega)})

7292: $$

7293: is unitary.

7294: \end{prop}

7295: Indeed we can decompose this transform as

7296: \begin{equation}

7297: C_t^{(-i\Omega)}  = \left( {\cal C} \circ e^{\frac{t }{

7298: 2}\Delta^{(im \Omega)}} \right) \circ  e^{\frac{t }{ 2}

7299: \Delta^{(-i re \Omega)}}.

7300: \end{equation}

7301: Then the unitarity of ${\cal C} \circ e^{\frac{t }{ 2}\Delta^{(im

7302: \Omega)}}$ is well known and that of the operator $e^{\frac{t }{

7303: 2} \Delta^{(-i re \Omega)}} = e^{-i \frac{t }{ 2}\Delta^{(re

7304: \Omega)}}$ (for any $t\in \R$) follows from the fact that

7305: $\Delta^{(re \Omega)}$ is a self-adjoint operator.

7306: 

7307: 

7308: Again  every $f\in L^{2}(uS_g,d\vec x)$ we can consider as $

7309: Ad_{\mathrm{diag}} \SU(2)$-invariant  function on $\SU(2)^{g}$

7310: and  use  the expansion given by the Peter-Weyl theorem:

7311: \begin{equation}

7312: f(x) = \sum_{\vec j \in \widehat{\SU(2)^g}} \mbox{tr} (\vec j

7313: (\vec x)B_{\vec j ,f} )

7314: \end{equation}

7315: where the sum is taken over the set of equivalence classes of

7316: irreducible representations of $\SU(2)^g$ and $B_{\vec j ,f}\in

7317: End \ V_{\vec j}$ is given by the usual formula. It has to be an

7318: intertwiner with respect to $ Ad_{\mathrm{diag}} \SU(2)$. Recall

7319: that every class $\vec j \in \widehat{\SU(2)^g}$ is a tensor

7320: product $\vec j = j_1 \otimes ... \otimes j_g$ where $j_i \in

7321: \widehat{\SU(2)}$ and the same is true for $\SL(2, \C)^g$. Its

7322: image under $C_t^{(-i\Omega)}$ is given by

7323: \begin{equation}

7324: C_{t}^{(-i\Omega)}f(\vec z) = \sum_{\vec j \in \widehat{\SU(2)^g}}

7325: e^{-t\frac{\lambda_{\vec j}^{(-i\Omega)}}{2}}\mbox{tr} (\vec j

7326: (z)B_{\vec j ,f } )

7327: \end{equation}

7328: where $\lambda_{\vec j}^{(-i\Omega)}$ is the eigenvalue of

7329: $-\Delta^{(-i\Omega)}$ on functions of the type

7330: $$\mbox{tr} (B_{\vec j} (\vec z)), \quad B\in End(V_{\vec j}).

7331: $$

7332: Again in this formula $j \in \widehat{\SL(2, \C)^g}$ and $B_{\vec

7333: j} $ is the same  endomorphism of the same space as for $j \in

7334: \widehat{\SU(2)^g}$. These endomorphisms are intertwiners with

7335: respect to $ Ad_{\mathrm{diag}} \SL(2, \C)$-action thus

7336: $C_{t}^{(-i\Omega)}f(\vec z)$ is a function on $S_g$.

7337: 

7338: This function is the analytic continuation to $\SL(2, \C)^g$ of

7339: the solution of the complex heat equation on $\SU(2)^g$

7340: \begin{equation}

7341: \frac{\partial u }{ \partial t} = \frac{1 }{ 2}

7342: \Delta^{(-i\Omega)} u

7343: \end{equation}

7344: with initial condition given by $f$.

7345: 

7346: Now this coherent states transform can be extended from

7347: $L^{2}(uS_g,d\vec x)$ to the space of distributions

7348: $L^{2}(uS_g,d\vec x)'$ (recall that the unitary Schottky space

7349: $uS_g$ is compact). More precisely we have

7350: \begin{prop} For any $\Omega\in H_g$ and $t>0$ the transform

7351: \begin{equation}

7352:  C_t^{(-i\Omega)} \colon (C^\infty(uS_g)' \to \sH ((\SL(2,

7353: \C))^g)

7354: \end{equation}

7355: $$

7356: f = \sum_{\vec j \in \widehat{\SU(2)^g}} \mbox{tr} (\vec j (\vec

7357: x)B_{\vec j ,f} ) \mapsto \sum_{\vec j \in \widehat{\SU(2)^g}}

7358: e^{-t\frac{\lambda_{\vec j}^{(-i\Omega)}}{2}}\mbox{tr} (\vec j

7359: (z)B_{\vec j ,f} )

7360: $$

7361: is  well defined linear map.

7362: \end{prop}

7363: 

7364: Indeed, the action of the Laplace operator and of its powers on

7365: the space of distributions is defined by duality from the

7366: corresponding action on $C^\infty(uS_g)$. For $f \in

7367: (C^\infty(uS_g)'$ with $f = \sum_{\vec j \in \widehat{\SU(2)^g}}

7368: \mbox{tr} (\vec j (\vec x)B_{\vec j ,f} )$ there exists positive

7369: a constant $C_1$  such that for all $\vec j \in

7370: \widehat{\SU(2)^g}$

7371: \begin{equation}

7372: \Vert  B_{\vec j ,f}   \Vert < C_1 \cdot e^{\Vert \vec j   \Vert}

7373: \end{equation}

7374: where

7375: $$

7376: \Vert \vec (j_1, ... , j_g)   \Vert = \sqrt{\sum_{l=1}^g (2j_l +

7377: 1)^2}.

7378: $$

7379: To show that the image of any distribution $f$ is an holomorphic

7380: function it is enouph to observe that there exists a positive

7381: constant $C_2$  such that for all $\vec j \in \widehat{\SU(2)^g}$

7382: \begin{equation}

7383:  |e^{-t\frac{\lambda_{\vec j}^{(-i\Omega)}}{2}}|\cdot \Vert  B_{\vec j ,f}

7384: \Vert < e^{C_2 \cdot (\Vert \vec j   \Vert)^2}.

7385: \end{equation}

7386: for $\Vert \vec j   \Vert$ sufficiently large. The existence

7387: follows from positivity of  imaginary part of $\Om$.

7388: 

7389: Our non-abelian  theta functions on $S_g$ are images under the CST

7390: transform defined in previous Proposition  of linear combinations

7391: of delta functions type distributions.

7392: 

7393: 

7394: Again  let us consider the distribution

7395: \begin{equation}

7396: \theta^{uS_{g}}_{k} (x) = \sum_{\vec j \in \widehat{\SU(2)^g}}

7397: \mbox{tr} (\vec j (\vec x) )

7398: \end{equation}

7399: that is $B_{\vec j ,f} = \id$

7400: 

7401: 

7402: It is nothing else as  delta-function at identity of level $k$.

7403: Let us apply  CST (7.20) to this distribution:

7404: \begin{equation}

7405: C_{1/k}^{(-i \Om)} (\theta_k^{uS_{g}}) = \sum_{\vec j \in

7406: \widehat{\SU(2)^g}} e^{-t\frac{k \lambda_{\vec

7407: j}^{(-i\Omega)}}{2}}\mbox{tr} (\vec j (z) )

7408: \end{equation}

7409: 

7410: For every spin network $\Ga_{j} \in SNNW^{k}_{\Ga}$ we can start

7411: with the function

7412: \begin{equation}

7413: \theta^{\Ga_{j}} (x) = f_{\Ga_{j}} \sum_{\vec j \in

7414: \widehat{\SU(2)^g}} \mbox{tr} (\vec j (\vec x) )

7415: \end{equation}

7416: to get the analytic function

7417: \begin{equation}

7418: C_{1/k}^{(-i \Om)} (\theta^{\Ga_{j}})

7419: \end{equation}

7420: 

7421: 

7422: \begin{rmk} Again we are specializing our continious

7423: positive parameter $t$ coming from

7424:  heat kernels by its discret contrepart $1/k$.

7425:  \end{rmk}

7426: 

7427: This collection of holomorphic functions on the Schottki space

7428: \begin{equation}

7429: \{ C_{1/k}^{(-i \Om)} (\theta^{\Ga_{j}})   \}

7430: \end{equation}

7431: depending on a period matrix of {\it any}  abelian variety is the

7432: combinatorial version of non abelian theta functions - theta

7433: functions with characteristics $\Ga_{j} \in SNW^{k}_{\Ga}$.

7434: 

7435: \section{BPU-map}

7436: 

7437: 

7438: \subsection{Geometry of Lagrangian cycles}

7439: 

7440: 

7441: We can see that the ranks of wave spaces  of the Kahler

7442: quantization and the Bohr-Sommerfeld quantizations are precisely

7443: the same that is the quantization is numerically perfect.

7444:  We can identify projectivizations of both

7445:  wave function spaces. But, is this identification natural?

7446:   It will be the case if we can send every

7447: covariant constant section to a holomorphic section of the

7448:  line bundle by some canonical way.

7449: 

7450: The natural way was used firstly that is the  coherent state

7451: transform or the Segal-Bargmann isomorphism introduced in the

7452: context of the quantum theory as a transform from the Hilbert

7453: space of of square integrable functions on the configuration

7454: space to the space of holomorphic functions on the phase space.

7455: In the abelian case (see subsection 2.4) considering the zero

7456: fiber $T^g_+$ of the polarization (2.22) we can interpret our

7457: complex torus $A$ as a complexification of this real torus and

7458: input our geometrical situation to the situation of the classical

7459: coherent state transform.

7460: 

7461:  In the very classical  context a configurations space is

7462:  just $\R^g$ as the real part of the phase space $\C^g$ but the

7463:  construction was extended to the cases when

7464: \begin{enumerate}

7465: \item $\R^n$ is replaced by $T^n$ and $\C^n$ by any complex

7466: torus $A$ such that $T^n$ is a special Lagrangian subtorus (the

7467: case of subsection 2.3);

7468: \item $\R^n$ is replaced by $SU(2)$ and $\C^n$ by $SL(2, \C)$

7469: (Hall, \cite{Ha}) in the section 7;

7470: \item $\R^n$ is replaced by $SU(2)^{[g]} / diag Ad SU(2)$

7471:  (= unitar

7472: Schottky space) and $\C^n$ by $SL(2, \C)^{[g]}/diag Ad SL(2, \C)$

7473: (= complex Schottky space) in the subsection 7.1;

7474: \item $\R^n$ is replaced by any Bohr-Sommerfeld Lagrangian cycle

7475: (more precisely a Legendrian cycle in the boundary of pseudo

7476: convex domain) and $\C^n$ by any compact Kahlerian symplectic

7477: manifold (Borthwick-Paul-Uribe map \cite{BPU}).

7478: \end{enumerate}

7479: 

7480: The last case is the point of applications of the beautiful part

7481: of calculus (Fourier integral operators, Legendrian distributions

7482: and so on) and is a very strong tool to compare results of the

7483: Bohr-Sommerfeld and Kaehler quantizations.

7484: 

7485: Actually the BPU-map is the extention of the classic WKB-method

7486: "acting in the opposite direction": WKB sends a wave function to

7487: a Lagrangian cycle with semi density in a phase space and BPU

7488: sends any Bohr-Sommerfeld cycle with a half form to a wave

7489: function (state) up to $U(1)$-scaling.

7490: 

7491: To start comparing the complex and symplectic geometries we

7492: consider a polarized Kahler manifold $(M_{I}, \om, L, \nabla)$

7493:  such that the curvature $F_{\na} = 2 \pi i \om$.

7494: We consider every Lagrangian cycle as a class of a

7495:  differential map of a compact smooth $1/2 dim M_{I}$-manifold

7496: \[

7497: \phi \colon S \to M_{I} \quad \text{ such that } \quad \phi^* \om

7498: = 0

7499: \]

7500: non degenrated at some point (that is $ker d \phi_{p} = 0$ ) up to

7501: the natural action of the group $\Diff^+$ of orientation

7502: preserving diffeomorphisms of the manifold. Fondations and

7503: constructions of the theory of such cycles are contained in

7504: \cite{GT}.

7505:  The main property of any

7506:  such cycle is that  {\em it can't be contained by any proper

7507:  algebraic subvariety}  (in particular, in a divisor). Thus any

7508:  holomorphic object is uniquely determined by its restriction

7509:  ( that is by

7510: the operation $\phi^*$ ) to a Lagrangian cycle $\phi \colon S \to

7511: X$. So restrictions to $S$ can serve as boundary conditions for

7512: holomorphic sections of line bundles with curvature proportional

7513: to $\om$.

7514: 

7515: 

7516:  Geometrically, if we consider Lagrangian cycles as supports of boundary

7517: conditions for holomorphic objects, they have the minimal possible

7518: dimension. Usual boundary conditions deal with boundaries of

7519: complex domains of real codimension 1. Thus it is only for

7520: Riemann surfaces that Lagrangian boundary conditions coincide

7521: with the usual boundary conditions. In this case, in the modern

7522: theory of integrable systems the restriction of holomorphic

7523: objects to a small circle around a point reduces many analytical

7524: problems to algebraic geometry of curves (see for example the

7525: survey \cite{DKN}). It seems reasonable to expect that

7526: restrictions to Lagrangian submanifolds give a higher dimensional

7527: generalization of the modern version of the theory of integrable

7528: systems.

7529: 

7530: There are no invariants of an embedding of a Lagrangian

7531: submanifold $S$ in a symplectic manifold. There are two ways of

7532: getting invariants:

7533:  \begin{enumerate}

7534:  \item consider families of Lagrangian manifolds admitting

7535:  invariants

7536: (in particular limit singular subcycles); or

7537:  \item to endow the submanifolds an additional structure (such as a

7538:  section of

7539: some bundle or a Hermitian connection on the trivial line bundle).

7540:  \end{enumerate}

7541:  The restriction of the pair $(L, \nabla)$ to a Lagrangian cycle $S$

7542: gives this type of additional structure. It defines the space of

7543: covariant constant sections: $ H^0_\na ((L, \nabla)\rest{S}) $

7544: which can be nontrivial. Indeed, the restriction  to any

7545: Lagrangian submanifold $S$ gives  topologically trivial line

7546: bundle on $S$ with flat connection. A connection of this type is

7547: defined by its monodromy character $

7548:  \chi\colon \pi_1(S)\to U(1),

7549:  $

7550: and it admits a covariant constant section iff this character is

7551: trivial.

7552: 

7553:  We know that a Lagrangian cycle $S$ is a  level $k$

7554: Bohr--Sommerfeld (BS$_k$) cycle if the character is trivial.

7555: 

7556: Moreover, such a section defines a  trivialization of the

7557: restriction , which identifies $C^\infty$ sections with complex

7558: valued functions on $S$:

7559:  \[

7560:  \Ga (L^{k}\rest{S})=C^{\infty}_\C(S).

7561:  \]

7562: Thus the restriction to $S$ defines an embedding

7563:  \begin{equation}

7564:  \res\colon H^0(M_{I},L^{k}) \hookrightarrow C^{\infty}_\C(S).

7565:  \end{equation}

7566: up to constant.

7567:  \begin{dfn} The image

7568:  \begin{equation}

7569:  \res(H^0(M_{I},L^k))=\sH_S\subset C^\infty(S)

7570:  \end{equation}

7571: is called  {\em analog of the Hardy space} of level $k$.

7572:  \end{dfn}

7573: 

7574: Fixing a half-form $\hF$ on $S$; we call a pair $(S,\hF)$ a {\em

7575: half-weighted Lagrangian} cycle or a Lagrangian cycle marked with

7576: a half-form. Now we can identify the space of functions with the

7577: space of half-forms

7578:  \begin{equation}

7579:  \Ga(L^{k})\rest{S} \cdot\hF =\Ga(\De^{1/2}),

7580:  \end{equation}

7581: where $\De$ is the complex volume bundle on $S$. This space is

7582: self adjoint with respect to the natural Hermitian form given by

7583: the integral of product of half-forms.

7584: 

7585: 

7586: \subsection{Legendrian distributions}

7587: 

7588: 

7589: 

7590: Following Borthwick, Paul and Uribe \cite{BPU}, we can construct a

7591: distribution in some completion of $C^\infty(S)\cdot\hF$. Its

7592: restriction to the image of $\sH_I^k$ gives a covector or a

7593: state.  BPU method uses usual codimension 1 boundary conditions

7594: rather than Lagrangian boundary conditions, and the original

7595: Hardy spaces for strictly pseudoconvex domains rather than analog

7596: of Hardy space. We refer the reader to the beautiful paper

7597: \cite{BPU} for the details because  in the paper \cite{GT} using

7598: Rawnsley's results we related this construction closely to the

7599: standard constructions of Algebraic geometry. The original

7600: construction is the following:

7601:   \begin{enumerate}

7602:    \item Our

7603: Hermitian connection on $L^*$ defines a contact structure on the

7604: unit circle bundle $P$ of $L^*$.

7605: \item The disc bundle in $L^*$ is a strictly

7606:  pseudoconvex domain, and there is the Szeg\"o orthogonal projector

7607:  $\Pi\colon L^2(P)\to \sH$ to the Hardy space of boundary values of

7608:  holomorphic functions on the disc bundle.

7609:   \item The contact manifold $P$

7610:  is a principal $U(1)$-bundle, and the natural $U(1)$-action on $P$

7611: commutes with $\Pi$ and gives a decomposition $\sH=\bigoplus_k

7612:  H^0(M_{I},L^{k})$ of the Hardy space.

7613:    \item If we fix a metaplectic

7614: structure on $P$, we can lift every $BS_{k}$ submanifold to a

7615:  Legendrian submanifold $\La\subset P$ over it, marked with the lifted

7616: half-form $\hF$.

7617:  \item $\La$ has an associated space of Legendrian

7618: distributions of order $m$, which is the Szeg\"o projection of

7619: space of

7620:  conormal distributions to $\La$ of order $m + \half\dim M_{I}$ (see

7621: \cite{BPU}, 2.1).

7622:  \item A half-form on $\La$ is identified with

7623: the symbol of a Legendrian distribution of order $m$ (see

7624: \cite{BPU}, 2.2); thus at the level of symbols, all Legendrian

7625: distributions look like delta

7626:  functions or their derivatives.

7627:   \item For a Legendrian submanifold $\La$

7628: with a half-form we fix the Legendrian distribution of order

7629: $\frac{1}{2}$ with symbol $\hF$ which is the Szeg\"o projection

7630: of the delta function

7631:  $\de_\La$.

7632:  \end{enumerate}

7633: 

7634: So we have:

7635:    \begin{enumerate}

7636:     \item

7637: For every lifting $\La\subset P$ of a $BS_{k}$ submanifold $S$

7638: marked with a half-form $\hF$ we have a vector

7639: \begin{equation}

7640: BPU_{k}(\La,\hF)=\Pi^k_{\hF} (\de_\La)\in H^0(M_{I},L^{k}),

7641: \end{equation}

7642:  where $\Pi^k_{\hF}$ is the Szeg\"o projection to the

7643: $k$th component of the Hardy space of the distribution with

7644: symbol $\hF$.

7645:  \item Every such lifting is defined up to $U(1)$-action on $P$;

7646:  thus a pair $(S,\hF)$ defines a point of the projectivization

7647: \begin{equation} BPU_{k} (S,\hF)=\PP (\Pi^k_{\hF} (\de_\La))\in \PP

7648:  H^0(M_{I}, L^{k}).

7649:  \end{equation}

7650:  \end{enumerate}

7651: To apply this construction to our case $(M^{ss}_{\Si_{\Ga}}, \om,

7652: \Oh(k \Theta), A_{CS})$ (4.12) we have to observe that

7653:  \begin{enumerate}

7654:  \item This construction holds literally in the case

7655:  when $\phi \colon S

7656:  \to M_{I} $ is an immersion or $S$ has the structure of

7657:  a smooth orbifold.

7658:  \item By (3.92) and Propositions 12 and 13 there exists

7659:  the canonical {\em geodesic} lifting of  Bohr-Sommerfeld cycles

7660:  to Planckian cycles.

7661:   \end{enumerate}

7662: 

7663: The following analog of Serre's Theorems~A and~B proved in

7664: \cite{BPU}, Section~3:

7665: 

7666:  \begin{thm} If\/ $n$ is large enough then for any halfweighted

7667:  $BS_k$-cycle $(S,hF)$\/ $BPU_{nk}(\La,\hF)\ne0$.  \end{thm}

7668: 

7669: There are two or three canonical ways to endow any Lagrangian

7670: submanifold $S$ with a half-form:

7671:  \begin{enumerate}

7672:  \item If $X$ is a K\"ahler manifold with a metaplectic structure. Then

7673: this metaplectic structure defines a metalinear structure on $S$

7674: (see for example Guillemin \cite{Gu2}), and the K\"ahler metric

7675: $g$ defines a half-form $\hF_g$ on $S$. (This method is of course

7676: the most important for our applications.)

7677: \item The graph of a metasymplectomorphism with symplectic volume as

7678: square of the half-form.

7679:  \end{enumerate}

7680: 

7681: Every half weighted cycle $(S, hF)$ is {\it weighted } by the

7682: volume form $hF^2$ and defines the number

7683: \begin{equation}

7684:  v(S, hF) = \int_S hF^2

7685: \end{equation}

7686: -the {\it volume } of this weighted cycle.

7687: 

7688: 

7689:  We write

7690:  \begin{equation} BS_{k}^v

7691:  \end{equation}

7692:  for the family of half-weight $BS$ cycles

7693: of a volme $v$. Then this space is an infinite dimensional {\it

7694: complex Kahler manifold } (see \cite{T5} and \cite{GT}).

7695: 

7696:   The BPU construction gives a

7697: ``rational'' map

7698:  \begin{equation} \PP \fie_k\colon BS^v_{k} \to \PP

7699: H^0(M^{ss}_{I}, L^{k})^*

7700:  \end{equation}

7701:   the differential of which

7702: is dual to the restriction monomorphism (1.27)

7703:  (see \cite{T7} and \cite{GT}).

7704: 

7705: Recall that to send a half-weighted $BS_{k}$-cycle $(S, hF)$ to a

7706: section $BPU_{k}(S, hF)$ we have to lift it to a Planckian cycle.

7707: If the canonical class $K_{M_{I}}$ is proportional to

7708: $c_{1}(L^{k}) = k \cdot[\om]$, we can do this up to finite set of

7709: possibilities which we call a {\it fixing of theta-structure of

7710: level} $k$ (see Definition 16 below). More precisely for every level $k$ we have the tower

7711:  of covers:

7712:  \begin{equation}

7713:  M_{k-theta} \to M_{SNW(k)} \to M_{\Ga_{g}} \to \sM_{g}

7714:  \end{equation}

7715: where $\sM_{g}$ is the moduli space of curves of genus $g$ (5.31),

7716: $M_{\Ga_{g}}$ is the moduli space of marked Riemann surfaces (4.8),

7717:  $M_{SNW(k)}$ is moduli space of marked curves $\Si_{\Ga}$ with

7718: fixed a spin network (4.42) and $M_{k-theta}$ is a curve with fixed

7719: theta-structure for the fixed half-weighted  $k$-Bohr-Sommerfeld

7720: fiber of the projection $\pi_{\Ga}$ (4.13). Let

7721: \begin{equation}

7722: \phi_{k-theta} \colon M_{k-theta} \to \sM_{g}

7723: \end{equation}

7724: be the composition of all these covers. Then the preimage of our

7725: vector bundle $\sH^{k} \to \sM_{g}$ (1.29) on this cover

7726: \begin{equation}

7727: \PP \phi^{*}_{k-theta}  \sH^{k} = \PP(\oplus_{w \in SNW_{k}(\Ga)} \Oh

7728: \cdot \theta_{w})

7729: \end{equation}

7730: where

7731: \begin{equation}

7732: \theta_{w} = BPU_{k-theta} (\widetilde{\pi_{\Ga}^{-1}(w)},

7733: hF_{can}).

7734: \end{equation}

7735: Here $\widetilde{\pi_{\Ga}^{-1}(w)}$ is the corresponding to the

7736: theta-structure  Planckian cycle (8.35).

7737: 

7738: The coincidence of the Hitchin, WZW and Kohno connections implies

7739: 

7740: \begin{thm} The lifting of the Hitchin connection to the cover

7741:  $\phi_{k-theta}$ (8.10) is

7742: Hermitian.

7743: \end{thm}

7744: Obviously this Hermitian structure isn't invariant with respect to this cover.

7745: 

7746: To use  the BPU construction described in (8.4), (8.5) of this

7747:  section we must repeat some of the details.  We stay in

7748: the situation of a complex polarization of a prequantized phase

7749: space $(M_{I}, \om, L, \nabla)$ where $(L, \nabla)$ is

7750: $\Oh(\Theta)$.

7751: 

7752: Consider the principal $U(1)$-bundle $P$ of the dual line bundle

7753: $L^*$, the unit circle bundle in $L^*$. Let $D\subset L^*, \quad

7754: \partial D=P $

7755:  be the unit disc subfibration with boundary.

7756: 

7757:  Our complex polarization $I$, that is a compatible complex

7758:  structure $I$ on $M$, defines a complex structure $D_I$ on $D$ as

7759:  a strictly pseudoconvex domain.(Recall that $L$ is a positive line

7760:  bundle).

7761: 

7762:    The Hermitian

7763:  connection on $L^*$ iduced by $\nabla$ is given by 1-form $\al$

7764:   on $P$ which defines a

7765:  contact structure on $P$ with volume form \[ \frac{1}{2\pi} \al \wedge

7766:  d\al^n, \quad \text{where} \quad n=\dim_{\C} M_{I}.  \] The null space of

7767:  $\al$ at a point $p\in P$ is the maximal complex subspace of the tangent

7768:  space for any complex structures.

7769: 

7770:  For every our $I$ the Hardy subspace

7771:  \begin{equation}

7772:  \sH_I\subset L^2(P)

7773:  \end{equation}

7774: consists of boundary values of holomorphic functions on $D_I$. We

7775: have the Szeg\"o orthogonal projector

7776:  \begin{equation}

7777:  \Pi_I\colon L^2(P)\to \sH_I.

7778:  \end{equation}

7779: The natural action of $U(1)= Aut P$ on $P$ as a principal bundle

7780: commutes with $\Pi_I$ and decomposes the space $\sH_I$ as a

7781: Hilbert direct sum of isotypes:

7782:  \begin{equation}

7783: \sH_I=\bigoplus_{k=0}^{\infty} \sH_I^k ; \quad \sH_I^k

7784: =H^0(M_{I}, L^{k})),

7785:  \end{equation}

7786: with only positive characters. Thus

7787:  \begin{equation}

7788:  \Pi_I=\bigoplus_{k=0}^{\infty} \Pi_k.

7789:  \end{equation}

7790:  Remark that in our situation the vector bundle $\sH^k \to \sM_{g}$

7791:   (1.29) is embedded

7792:  to the trivial

7793:  bundle $L^2(P) \times \sM_{g}$ with the space (8.1) as a fibre

7794:  and there

7795:  exists the Szego projection

7796:  \[

7797:  \Pi \colon  L^2(P) \times \sM \to \sH = \oplus_{k=0}^{\infty} \sH^k,

7798:  \]

7799:  which is (8.2) fiberwise.

7800: 

7801:  Recall that the vector bundles  $\sH^k \to \sM_g$   admit

7802:  projective flat connections. So locally the space $\PP\sH^k_I$ doesn't

7803:  depend on  complex structure $I$.  Then there exists a

7804: central extension $\widetilde{Mod_{g}}$:

7805:   \begin{equation} 1\to U(1)\to\widetilde{Mod_{g}}\to Mod_{g} \to 1,

7806:  \end{equation}

7807: where the centre $U(1)= Aut P$ acts on $P$ as the proper group of

7808: a principal bundle. This action induces a natural representation

7809:  \begin{equation}

7810:  \rho\colon \widetilde{Mod_{g}} \to Op(L^2(P))

7811:  \end{equation}

7812: to the operator algebra of the space of functions.

7813: 

7814: For every $g \in \widetilde{Mod_{g}} $ the projectors $\Pi_I$

7815: (8.4) define the linear transformation

7816:  \begin{equation}

7817: p_{I, g(I)} \circ  \Pi_{g(I)} \circ \rho(g)  \circ \Pi_I \in

7818: Op(\sH_I)

7819: \end{equation}

7820:  which commutes

7821: with the action of $U(1)$. So this transformation is decomposed

7822:   as a Hilbert direct sum of isotypes

7823:    \begin{equation}

7824:   Op(\sH_I) = \oplus_k       End H^0(L^k).

7825:   \end{equation}

7826:   These transformations  can be projectivized to the representation

7827:  \[

7828:  \PP\rho_I^k\colon Mod_{g} \to Aut\PP H^0(L^k).

7829:  \]

7830: Recall that our $M_{I}$ and the principal bundle $P$ have given

7831: metaplectic structures.

7832: 

7833: Now if $(S, hF)$ is a half-weight $BS$-orbifold, the complex

7834: conjugate of a covariant constant section gives a lift of it to a

7835: half weighted Legendrian cycle  $(\La,\hF)$, and the BPU --

7836: construction (8.4) defines a section $ BPU(\La,\hF)\in

7837: H^0(M_{I}, L) $.

7838:  In the same vein, we get lifting $\La_{k}$ of $S$ from $BS_{k}$

7839:  to the Legendrian cycle on $P$ which is a cyclic $(k)$-cover of

7840:  it and

7841: the system of sections

7842:  \begin{equation}

7843:  BPU_{k}(\La_{k},\hF)\in

7844: H^0(X, L^{k}).

7845:  \end{equation}

7846: 

7847:  The Lagrangian cycle $S$ can be reconstructed from the set of

7848:  its BPU

7849: images as the quasi-classic limit as $1/k=\text{Planck's

7850: constant}\to0$: the wave fronts of distributions concentrate on

7851: $S \subset M_{I}$ (see \cite{BPU} and the references given there).

7852: 

7853: Now for two Lagrangian cycles $(S_1,\hF_1)$ and $(S_2,\hF_2)$, the

7854: asymptotic behaviour of the scalar product

7855: $\Span{BPU_k(S_1,\hF_1), BPU_k(S_2,\hF_2)}$ (see (1.9)) can be

7856: computed in terms of the intersection $S_1\cap S_2$ (see

7857: \cite{BPU}). In particular,

7858:  \begin{equation}

7859: S_1\cap S_2=\emptyset \implies BPU_k(S_1,\hF_1) \perp

7860: BPU_k(S_2,\hF_2)

7861:  \end{equation}

7862: asymptotically as $k\to\infty$. (For the orbifold case these

7863: asymptotics are somewhat weaker, but are still quite expressive

7864: for geometric corollaries). This asymptotic technique comes from

7865: the physical interpretation of this set-up as the ``classical''

7866: geometric quantization. More precisely the asymptotic analysis of

7867: quantum states gives

7868:  \begin{multline}

7869:  \Span{BPU_k(S,\hF), BPU_k(S,\hF)} \ \sim\

7870:  \left(\frac{k}{\pi}\right)^{\half dim M_{I}}\int_{S} \vert\hF \vert^2 \\

7871:  +O(k^{\half(dim M_{I}-1)}),

7872:  \end{multline}

7873:  and if $S_1\cap S_2=\emptyset$ then

7874:  \begin{multline}

7875:  \Span{BPU_k(S_1,\hF_1), BPU_k(S_2,\hF_2)} \ \sim\

7876:  \left(\frac{k}{\pi}\right)^{\half(dim M_{I}-1)}+ \\

7877:  O(k^{\half(dim M_{I}-2)}).

7878:  \end{multline}

7879:  From this  and the description of the differential of BPU-map we get

7880:   \begin{prop} Let $\pi\colon M_{I}\to B$ be any real

7881:  polarization  with  transversal set of

7882: BS$_{k}$ - fibers.  Then we can fix such half forms on these

7883: fibers then for $k\gg0$, the BPU vectors in $H^0(M_{I},L^{k})$ are

7884: linear independent:

7885:   \[ rk

7886: \Span{BPU_{k}(BS_{k} \cap B)}=\#(BS_k \cap B).  \]

7887: \end{prop}

7888: 

7889:  \begin{rmk} A more sophisticated analysis of the asymptotics of quantum

7890: states extends this observation as follows: let

7891:  \[

7892: S_1,\dots, S_{N_{\max}}\subset BS_{k}

7893:  \]

7894: be a maximal collection of {\em disjoint\/} $BS_{k}$ cycles. Then

7895:  \[

7896: N_{\max} \le rk H^0(L^{k}),

7897:  \]

7898: and the right-hand side is given by the Riemann--Roch theorem.

7899: \end{rmk}

7900: 

7901: 

7902: \subsection{Geodesic lifting}

7903: 

7904: 

7905: 

7906: 

7907: The lifting of a BS-Lagrangian cycle in $M_{I}$ to a Legendrian

7908: cycle on $P$ we have described is defined up to the natural

7909: $U(1)$-action on $P$ and the states $BPU_k(S,\hF)$ are defined

7910: up to a phase. But in our situation we can do this almost

7911: canonically (up to a finite ambiguity called a choice of a

7912: theta-structure) and get an actual basis of $H^0(M_{I}, L^{k})$.

7913: 

7914: To describe this almost canonical lifting we must consider the

7915: Lagrangian Grassmannization of the tangent bundle of $M_{I}$ as

7916: described in \cite{T3}. Pointwise, the tangent space $(TM_{I})_x$

7917: at a point $x\to M_{I}$ is $\C^n$ with  constant symplectic form

7918: $\Span{\ \,,\ }=\om_x$ and  constant Euclidean metric $g_x$,

7919: giving the Hermitian triple $(\om_x,I_x,g_x)$. Define the

7920: Lagrangian Grassmannian $(LaGr)_x =LaGr(TM_{I})_x$ to be the

7921: Grassmannian of oriented Lagrangian subspaces in $(TM_{I})_x$.

7922: Taking this space over every point of $M_{I}$ gives the oriented

7923: Lagrangian Grassmannization of $TM_{I}$

7924:  \begin{equation}

7925:  \pi\colon LaGr(TM_{I})\to M_{I}

7926: \quad\text{with} \quad \pi^{-1} (x) =(LaGr)_x.

7927:  \end{equation}

7928: A complex structure $I_x$ on $(TM_{I})_x$ gives the standard

7929: identification

7930:  \begin{equation}

7931: (LaGr)_x=U(n) / SO(n).

7932:  \end{equation}

7933: This space admits a canonical map

7934:  \begin{equation}

7935:  det\colon(LaGr)_x\to U(1)=S^1_x

7936:  \quad\text{sending $u \in U(n)$ to $det u \in   U(1)=S^1$.}

7937:  \end{equation}

7938:  Taking this map over every point

7939: of $M_{I}$ gives the map

7940:  \begin{equation}

7941:  det\colon LaGr(TM_{I})\to S^1(L_{-K}),

7942:  \end{equation}

7943: where $S^1(L_{-K})$ is the unit circle bundle of the line bundle

7944: $\bigwedge^{n} TM_{I} =det TM_{I}$, with the first Chern class

7945:  \[

7946:  c_1(det TM_{I})=-K_{M_{I}}

7947:  \]

7948:  the canonical class of $M_{I}$ (see for example \cite{T2} and

7949: \cite{T3}).

7950: 

7951: We have already noted that our Lagrangian cycles does not usually

7952: have an orientation defined a priori. Thus we must consider the

7953: Lagrangian Grassmannian $\La(TM_{I})$ forgetting orientations.

7954: Then we get a map

7955:  \[

7956: det\colon \La(TM_{I})\to S^1(L_{-K/2})

7957:  \]

7958: in place of (8.28).

7959: 

7960: Now for every oriented Lagrangian cycle $S\subset X $, we have the

7961: Gaussian lifting of the embedding $i\colon S\to X$ to a section

7962:  \begin{equation}

7963:  G(i)\colon S\to La(TM_{I})\vert_{S},

7964:  \end{equation}

7965:  sending $x\in S$ to the subspace $TS_x\subset(TM_{I})_x$. The composite

7966: of this Gauss map with the projection $det$ gives the map

7967:  \begin{equation}

7968:  {det}\circ G(i)\colon S\to S^1(L_{-K/2})\vert_{S}.

7969:  \end{equation}

7970: Thus every Lagrangian cycle $S$ defines a Legendrian subcycle

7971:  \begin{equation}

7972:  \La={det}\circ G(i) (S)\subset S^1(L_{-K/2}).

7973:  \end{equation}

7974: The Levi-Civita connection of the K\"ahler metric defines a

7975: Hermitian connection $A_{LC}$ on $L_{-K/2}$.

7976: 

7977:  \begin{dfn} A Lagrangian cycle $S$ is  almost geodesic if the

7978: Legendrian cycle $\La={det}\circ G(i)(S)$ is horizontal with

7979: respect to the Levi-Civita connection $A_{LC}$ on $L_{-K/2}$.

7980: \end{dfn}

7981: 

7982: We  use Proposition 13 now. The line bundle $L_{-K/2}$ is $\Oh(2

7983: \Theta )$ and the Levi-Civita connection is induced by the

7984: connection $\nabla$ on $L = \Oh(\Theta)$. Then we have

7985:  \begin{prop} A Lagrangian cycle $S$ is $BS$ if and only if it is

7986: almost geodesic.

7987:  \end{prop}

7988: 

7989:  The Hermitian structures of our line bundles define a map

7990:  \begin{equation}

7991:  \mu_2\colon S^1(L^*)\to S^1(L_{-K/2})

7992:  \end{equation}

7993: of the principal $U(1)$-bundles of these line bundles, which

7994: fibrewise is minus  isogeny of degree $2$. Thus {\em every}

7995: Lagrangian cycle $S$ defines an oriented Legendrian subcycle (see

7996: (8.31))

7997:  \begin{equation}

7998:  \La=\mu_2^{-1} ({det}\circ

7999: G(i) (S))\subset S^1(L^*)=P

8000:   \end{equation}

8001: (not almost geodesic {\it a priori}).

8002: Now consider the pair of isogenies

8003:  \begin{equation}

8004: \mu_{k}\colon S^1(L^2)\to S^1(L^{2 k})

8005:  \end{equation}

8006:  \[

8007: \mu_2\colon S^1(L^{k})\to S^1(L^{2 k})

8008:  \]

8009: and the lifting $ l\colon S\to S^1(L^{k}) $ given by a covariant

8010: constant section over a $BS_{k}$ cycle $S$.

8011:  \begin{dfn} The lift $l$ is  almost geodesic if

8012:  \[

8013:  \mu_{k} \circ {det}\circ G(i) (S)=\mu_2 \circ l (S).

8014:  \]

8015:  \end{dfn}

8016: 

8017: The number of geodesic lifts is obviously $\leq 2(k+d)$.

8018: 

8019: In summary, let $\La_k$ be the space of Legendrian subcycles of

8020: $P$ the images of whose projection to $M_{I}$ is the $k$th root of

8021: unity cover of a $BS_{k}$ Lagrangian cycle on $M_{I}$ (that is

8022: Planckian cycles). Then the natural projection

8023:  $ p\colon \La_{k}\to BS_{k} $

8024: which sends  Legendrian cycle to Lagrangian cycle is a principal

8025: $U(1)$-bundle.

8026: 

8027: The geodesic lifting

8028:  \begin{equation}

8029: l\colon \widetilde{BS_{k}} \to \La_{k}

8030:  \end{equation}

8031: we have described is a multisection of this principal bundle and

8032:  $ p\colon \widetilde{BS_{k}}\to BS_{k} $

8033: is a finite cyclic cover.

8034: 

8035: Consider a real polarization $\pi\colon M_{I}\to \De$ (1.13). Then

8036: we have a finite set of Bohr--Sommerfeld fibres

8037:  \[

8038:  \De \cap BS_{k}=\{S_i\}, \quad \text{for $i=1,\dots, \vert SNW_{k}$.}

8039:  \]

8040: 

8041:  \begin{dfn} A choice of geodesic lifts

8042:  \[

8043: \{\tilde{S_i}\}\subset \La_{k}

8044:  \]

8045:  is called a choice of  theta structure of the real polarization

8046: $\pi_{\Ga}$.

8047:  \end{dfn}

8048: 

8049: Marking these Lagrangian cycles with the half-forms  we get a

8050: finite set

8051:  \[

8052:  \{\tilde{S_i},\hF_i\}

8053:  \]

8054: of half-weight Legendrian cycles.

8055: 

8056: We know that

8057:  \[

8058: BPU_k (\{\tilde{S_i},\hF_i\})\subset \PP H^0(M_{I},L^{k})

8059:  \]

8060: is a linear independent system of vectors (states) if $k\gg0$ and

8061: an accurate choice of half forms.  In particular, if

8062: \begin{equation}

8063:  \#(BS_{k} \cap B)=rk H^0(M_{I},L^{k}),

8064:  \end{equation}

8065:   we

8066:  get a Bohr--Sommerfeld basis.

8067: 

8068:  \begin{rmk}  For other descriptions and

8069: applications of the geodesic lift from Lagrangian to Legendrian

8070: cycles see \cite{T1}, \cite{T2} and \cite{T3}.

8071:  \end{rmk}

8072: 

8073: Now on every $BS$-fiber we can fix the canonical half-form. For

8074: this return to the description of $BS$-fibers (4.36) - (4.38).

8075: Remark that  groups (4.30) admit bi-invariant half-forms

8076: $\hF_1$ on $U(1)$ and $\hF_2$ on $\SU(2)$. For every $w$ we can

8077: normalize these form $\hF_1(w)$ and $\hF_2(w)$ so that the

8078: half-form

8079: \begin{equation}

8080:  \hF_w=(\hF_1(w))^{t-s} \cdot (\hF_2(w))^{p+s}

8081:  \end{equation}

8082: is homogeneous of degree 1 on $\pi^{-1}(w)$ (see (4.37)) with

8083: respect to scaling $\hF_i(w)\to t\cdot\hF_i(w)$. We say that such

8084: half-form is homogeneously normalized.

8085: 

8086: It's easy to see (\cite{JW2}, 4.7) that a normalized half-form

8087: for a nonsingular $BS_{k}$ fibre is given by a Hamiltonian vector

8088: field with Hamiltonian in $\R^{3g-3}$ of volume 1.

8089: 

8090: Thus every $BS_{k}$ fibre is endowed with the covariant constant

8091: half-form (8.37), and we can proceed to construct the

8092: corresponding Legendrian distributions in $P$. Recall that

8093: $M_{I}$ is almost homogeneous with respect to the Goldman torus

8094: action (4.5). Thus the Schwartz kernel of coherent states does

8095: not depend on points and, outside singular points of fibres, they

8096: behave as in the homogeneous case (see \cite{BPU}, (10--13)). By

8097: lifting to $P$ every $BS_{k}$ fibre $\pi^{-1}(w)$ defines

8098: Legendrian subcycle $\La_w\subset P$ marked with the half-form

8099:  \begin{equation}

8100:  \bar{hF}_w=(\fie_4\circ {det}\circ G(i))^* hF_w,

8101:  \end{equation}

8102: and having monodromy a $k$th root of 1.

8103: 

8104: To apply the BPU construction the principal bundle $P=S^1(L^*)$

8105: must be given a metaplectic structure. This can be done at once

8106: using the equality $K_{M_{I}} = - 4 \Theta$.

8107: 

8108: 

8109: Our long trip is finishing at the final statement:

8110: 

8111: {\it the collection of functions}

8112: \begin{equation}

8113: \{ \theta^{\Ga_j} \}, \quad \Ga_j \in SNW^k_\Ga

8114: \end{equation}

8115: (7.25) {\it coincides with the collection of sections}

8116: \begin{equation}

8117: \{\theta_w = BPU_{k-theta} (\widetilde{\pi^{}-1}_\Ga(w), \bar{hF}_{w}) \}

8118: \end{equation}

8119: (see (8.12) and (8.38)) {\it under the meromorphic map} (3.44).

8120: 

8121: 

8122: Now one can prove this up to phases (6.18). May be it is productive to input

8123: these phase ambiguities into the non-abelian theta-structure notion?

8124: 

8125: \section{The main weapon}

8126: 

8127: \subsection{Large limit curves}

8128: 

8129: To see how all the constructions could be exploited  in the main game we have to show  behavior on the moduli space "pointwise" way. For this we are using the following strategy:

8130: \begin{enumerate}

8131: \item we choose special complex curves which are convinient for applications

8132: of  constructions as algebro-geometrical so analytical  constructions and

8133: \item extend obtained structures to general curves using the projective connections.

8134: \end{enumerate}

8135: Recall that the analytic geometry formulation of two-dimensional conformal field theory

8136: (see \cite{FS})  predicts the existence of an extension of all geometrical objects like vector

8137:  bundles (1.29) to the compactification $\ov{\sM_g}$ of the moduli space of curves. So our "special" curves may  be non smooth. Moreover our curves are

8138: "maximally" non smooth and reducible. They look like

8139:  maximally unipotent boundary points of moduli spaces  of 

8140: Calabi-Yau threefolds.

8141: Now  we define such type points for moduli spaces of algebraic

8142:  curves  and investigate the geometry of their deformations. This

8143:  geometry reflects

8144:  mathematical relations between (1,1)-conformal field theories

8145: and string theories observed  by physisists in a number of

8146: papers. Recall that maximally unipotent boundary points  of

8147: CY-moduli spaces correspond to large radius limit point according to 

8148: mirror symmetry. We have not time to discuss all parallel to mirror symmetry

8149: for algebraic curves here. However we  call our curves {\it  large limit

8150: curves} (ll-curves for short).

8151:  

8152: Let $\Ga$ be any 3-valent graph of genus $g$.

8153: Topologically it is equivalent to

8154: 3-dimensional handlebody $H_{\Ga}$ with boundary

8155: $\p H_{\Ga} = \Si_{\Ga}$ (6.9)

8156: where $\Si_{\Ga}$ is the Riemann surface given by the pumping up

8157: trick (see Fig. 9) : we pump every edge of $\Ga$ to a tube (a {\it channel} in

8158: terms of CQFT) and every vertex to a trinion (=  2-sphere with 3

8159: holes). By the construction our Riemann surface $\Si_{\Ga}$ has a

8160: trinions (or  "pair of pants") decomposition given by removing all

8161: tubes.

8162: 

8163: 

8164: Of course we  get a surface without any complex structure.

8165: Thus it is not an algebraic curve, but we can certainly define uniquely   a

8166: reducible algebraic curve $P_{\Ga}$. Namely, send every vertex $v \in

8167: V(\Ga)$ to the complex Riemann sphere $P_{v} = \C\PP^{1}$ with a

8168: triple points $p_{e_1}, p_{e_2}, p_{e_3}$ corresponding to edges of

8169:  the star $S(v)$. Now for every edge $e$ with vertices $\p e = v, v'$

8170: we identify points $ p_{e}$ of components $P_{v}$ and $P_{v'}$.

8171: As the result we obtain a reducible algebraic curve with

8172: properties:

8173: 

8174: \begin{enumerate}

8175: \item  arithmetical genus of the connected reducible curve

8176:  $P_{\Ga_{g}}$ is equal to $g$;

8177:  \item for any 3-valent graph $\Ga_{g}$ the curve $P_{\Ga_{g}}$ is

8178:   Deligne - Mumford stable and called  the {\it large limit

8179:   curve};

8180:   \item the curve $P_{\Ga_{g}}$ defines a point of the Deligne-Mumford

8181:    compactification $\ov{\sM_{g}}$ (with the same notation);

8182:   \item   this

8183: point is contained by the boundary divisor $D = \ov{\sM_{g}} -

8184: \sM_{g}$;

8185: \item moreover,

8186: \begin{equation}

8187: \{ P_{\Ga}  \} \in D_{0} \cap ( \cap_{i=1}^{g-1} D_{i})

8188: \end{equation}

8189: where  general point of divisor $D_{0}$ corresponds to an

8190: irreducible curve with one node and general point of $D_{i}$ is a

8191: bouquet of smooth curves of genus $i$ and $g-i$ and

8192: \begin{equation}

8193: D = D_{0} \cup ( \cup_{i=1}^{g-1} D_{i})

8194: \end{equation}

8195: \end{enumerate}

8196: 

8197: So in the Deligne-Mumford compactification $\ov{\sM_{g}}$ of the

8198: moduli space $\sM_{g}$ of smooth curves of genus $g$ we get the finite

8199: configuration of points

8200: \begin{equation}

8201: \sP \subset \ov{\sM_{g}}

8202: \end{equation}

8203: enumerated by the set of 3-valent

8204:  graphs $TG_{g}$ - the set of large limit curves now.

8205: 

8206:  \begin{rmk}

8207:  These curves make a problem of the necessary part of two-dimensional conformal

8208:  field theory formulation as analytic geometry (see \cite{FS}).

8209:   Recall that the correlation functions of all local quantum

8210:   fields can be recovered from the partition function when all

8211:   channels (tubes)  of the surface are constricted down to nodes.

8212:   For the surface $\Si_{\Ga}$ the constriction of all 3g-3 tubes

8213:   produces our reducuble curve $P_{\Ga}$ with uniquely defined

8214:   complex structure.   On the other hand from the point of view of  dynamic

8215:   triangulation theory  these curves are the

8216:   polymer

8217:   phases of asymptotic  triangulations.

8218:  \end{rmk}

8219: 

8220: \subsection{Canonical classes and canonical maps}

8221: 

8222: For every curve $P_{\Ga}$ the canonical class $K_{\Ga}$ is a line

8223: bundle: a restriction of it to every component $P_{v}, v \in

8224: V(\Ga)$ is the sheaf of meromorphic differentials $\om$ with

8225: simple poles at $p_{e_1}, p_{e_2}, p_{e_3}$ where $e_1 \bigcup e_2 \bigcup e_3 =

8226: S(v)$. Thus

8227: \begin{equation}

8228: K_{\Ga} \vert_{P_{v}} = K_{P_{v}}(p_{e_1} + p_{e_2} + p_{e_3} ) =

8229: \Oh_{P_{\Ga}}(1).

8230: \end{equation}

8231: and in the Neron-Severi lattice $NS(P_\Ga) = \prod_{v \in V(\Ga) Pic(P_v)}$

8232: one has

8233: \begin{equation}

8234: c (K_{\Ga}) = (1,1,...,1) \in NS(P_\Ga).

8235: \end{equation}

8236: Every holomorphic section $s$ of $K_{\Ga}$ is a collection of

8237: meromorphic differentials $\{ \om_{v}\}$ on the components $P_{v}$

8238: with poles at $\bigcup_{e \in E()\Ga} p_{e}$ with constraints: for every $e$ such that $\p e = v, v'$

8239: \begin{equation}

8240: res_{p_{e}}\om_{v} +  res_{p_{v'}} \om_{v'} = 0.

8241: \end{equation}

8242: On the other hand we have 2g-2 linear relations: for every $v \in

8243: V(\Ga) $ with $(e_1,e_2, e_3 )= S(v)$

8244: \begin{equation}

8245: res_{p_{e_1}} \om_{v} + res_{p_{e_2}} \om_{v} + res_{p_{e_3}}

8246: \om_{v} = 0

8247: \end{equation}

8248: The {\it thickness} $th(\Ga)$ of the graph is the minimal number of edges that may be removed to make the graph disconnected. In \cite{Ar} Artamkin proved the following

8249: 

8250: \begin{prop} The canonical linear system $\vert K_{P_\Ga} \vert$ is

8251: \begin{enumerate}

8252: \item base points free iff $th(\Ga) \geq 2$;

8253: \item very ample iff  $th(\Ga) \geq 3$.

8254: \end{enumerate}

8255: \end{prop}

8256: 

8257: For simplicity below we are working with 3-valent graphs of thickness 3 only.

8258: Then we have the canonical embedding

8259: \begin{equation}

8260: \phi_\Ga

8261:  \colon P_\Ga \to \PP^{g-1} 

8262: \end{equation}

8263: and it is easy to check

8264: 

8265: \begin{prop} In $\PP^{g-1}$  dimension of  complete linear system

8266: \begin{equation}

8267: \vert 2h - \phi_{K_{\Ga}}(P_{\Ga}) \vert

8268: \end{equation}

8269: of quadrics through the canonical curve is equal to

8270: \begin{equation}

8271: dim \vert 2h \vert - 3g + 3.

8272: \end{equation}

8273: \end{prop}

8274: The proof is quite classical: we just repeat here the classical argument of Castelnuovo.

8275: 

8276: \begin{cor}

8277: For  3-valent graph $\Ga_{g}$ the large limit curve $P_{\Ga}

8278: \in \ov{\sM_{g}}$ is a smooth point of $\ov{\sM_{g}}$ as orbifold

8279: and the fiber of the cotangent bundle at this point is given by the following 

8280: equality

8281: \begin{equation}

8282: T^{*}_{P_{\Ga}} \ov{\sM_{g}} = H^{0}(P_{\Ga}, \Oh_{P_\Ga} (2K_{P_\Ga}).

8283: \end{equation}

8284: \end{cor}

8285: Now consider  the double  canonical map of $P_\Ga$ given  by the

8286: complete linear system $\vert 2K_{\Ga}\vert$

8287: \begin{equation}

8288: \phi_{2K_{\Ga}} \colon P_{\Ga} \to \PP^{3g-4}.

8289: \end{equation}

8290: Remark that the target projective space has the interpretation

8291: \begin{equation}

8292: \PP^{3g-4} = \vert 2K_{\Ga}\vert^{*} = \PP

8293: T_{P_{\Ga}} \ov{\sM_{g}}

8294: \end{equation}

8295: is the projectivization of the tangent space to the $\ov{\sM_{g}}$

8296: at $P_{\Ga}$.

8297: 

8298: As any double canonical curve our configuration $\phi_{2K_{\Ga}} (P_\Ga)$ of conics and

8299: double points is determined up to a projective automorphism of

8300: $\PP^{3g-4} = \PP T_{P_{\Ga}} \ov{\sM_{g}} $ only. But in the next

8301: subsection we show that  {\it the double canonical

8302: model of a large limit curve is determined absolutely canonically 

8303: without the projective transformation ambiguity} (unlikely of

8304: canonical model of general algebraic curve).

8305: 

8306: Indeed, the images  of nodes 

8307: \begin{equation}

8308: \{ \phi_{2K_{\Ga}}(\bigcup_{e \in E(\Ga)} p_{e}) \}

8309: \end{equation}

8310: define the configuration of 

8311: $$3g-3 = rk H^0(P_\Ga, \Oh(2K_{P_\Ga}))$$ 

8312: linear independent  points ( since components are coming to irreducible conics

8313: and every conic is defined by any triple of points on it).  Thus this configuration of points gives the decomposition of the tangent space

8314: \begin{equation}

8315: T_{P_{\Ga}} \ov{\sM_{g}}= \bigoplus_{e \in E(\Ga)} \C_{e}

8316: \end{equation}

8317: where $\PP \C_{e} = \phi_{2K_{\Ga}}(p_{e})$.

8318: 

8319: In the next

8320: subsection we show that every tangent direction $\C_{e}$ corresponds

8321: to a rational curve in $\ov{\sM_{g}}$. This identification of images of nodes

8322: under the double canonical embedding and the directions of special deformations cancels  the projective transformation ambiguity.

8323: 

8324: \subsection{Special 1-parameter deformation of large limit curves}

8325: 

8326: Let $ e \subset \Ga$ be  a 3-valent graph $\Ga$ with

8327: fixed edge  $e \in E(\Ga)$ such that $v, v' = \p e$ are two

8328: different points with stars $S(v)= e, e_{1}, e_{2}$ and $S(v') =

8329: e, e'_{1}, e'_{2}$. That is $e$ is not a loop. We call such pair

8330: a simple {\it flag}.  Let us blow down this edge in our graph. We

8331: get  new graph with 4-valent vertex $v_{new} = v = v'$ with the

8332: star $S(v_{new}) = e_{1}, e_{2}, e'_{1}, e'_{2}$. There are 3

8333: partitions this edges to pairs: the old one $(e_{1}, e_{2}) \vert

8334: (e'_{1}, e'_{2})$ and couple of new ones:

8335: \begin{equation}

8336: (e_{1}, e'_{2} \vert e_{2}, e'_{1}) \text{ and } (e_{1}, e'_{1}

8337: \vert e_{2}, e'_{2}).

8338: \end{equation}

8339: Now we can blow up the vertex  $v_{new}$ to the edge $e_{new}$

8340: with vertices $\p e_{new} = v_{new}, v'_{new}$ and stars

8341: \begin{enumerate}

8342: \item $S(v_{new}) = e_{new}, e_{1}, e_{2}$ and $S(v'_{new}) = e_{new},

8343: e'_{1}, e'_{2}$. This is our starting graph $\Ga$.

8344: \item $S(v_{new}) = e_{new}, e_{1}, e'_{2}$ and $S(v'_{new}) = e_{new},

8345: e'_{1}, e_{2}$. This is the first new graph $\Ga'$.

8346: \item $S(v_{new}) = e_{new}, e_{1}, e'_{1}$ and $S(v'_{new}) = e_{new},

8347: e'_{2}, e'_{2}$. This is the second new graph $\Ga''$.

8348: \end{enumerate}

8349: Remark that as a result of this construction every obtained graph

8350:  has distinguished edge that is we have the triple of flags

8351: \begin{equation}

8352: (e \subset \Ga), ( e_{new} \subset \Ga'), ( e_{new} \subset

8353: \Ga'').

8354: \end{equation}

8355: Such triple we call a {\it nest} of flags. Every nest is defined

8356: uniquely by any flag $(e \subset) \Ga$ of the triple. Moreover,

8357: let $S(e)$ be the  star  of an edge $e$ that is  the union

8358: of stars of the boundary $v, v' = \p e$

8359: \begin{equation}

8360: S(e) = S(v)\bigcup S(v').

8361: \end{equation}

8362: Then we have the canonical identification of graphs

8363: \begin{equation}

8364: \Ga - S(e) = \Ga' - S(e_{new}) = \Ga'' - S(e_{new}).

8365: \end{equation}

8366: 

8367: \begin{rmk}

8368: It is easy to see that if $e$ is a loop this construction gives

8369: the same graph with the same loop $e$ again.

8370: \end{rmk}

8371: Consider two componets $P_{v}, P_{v'}$ of a curve $P_{\Ga}$ with

8372:  common point $p_{e}$ where $v, v' = \p e$. Remove this point and

8373:  glue $P_{v}$ and $P_{v'}$ by a tube that is, consider the

8374:  connected sum

8375:  \begin{equation}

8376:  P_{v}\#_{p(e)} P_{v'} = P_{v,v'} = S^{2}.

8377:  \end{equation}

8378: This is a 2-sphere with two pairs of points $(p_{e_{1}}, p_{e_{2}})$

8379: and $(p_{e'_{1}}, p_{e'_{2}})$ where $(e, e_{1}, e_{2}) = S(v)$

8380: and $(e, e'_{1}, e'_{2}) = S(v')$. If we fix a complex structure

8381: on $S^{2}$ and consider the double cover

8382: \begin{equation}

8383: \phi \colon E \to \PP^{1}

8384: \end{equation}

8385: with ramification points

8386: \begin{equation}

8387: W = p_{e_{1}} \bigcup p_{e_{2}} \bigcup p_{e'_{1}} \bigcup

8388: p_{e'_{1}}

8389: \end{equation}

8390: we obtain an elliptic curve $E$ with a point of second order

8391: \begin{equation}

8392: \si = p_{e_{1}} + p_{e_{2}} - p_{e'_{1}} - p_{e'_{2}}.

8393: \end{equation}

8394: So the moduli space of complex structures on $S^{2}$ is equal to

8395: $\sM_{1}^{2}$ - the moduli space of smooth elliptic curves with fixed

8396: point of order 2.

8397: 

8398: Every such complex structure $\tau \in \sM_{1}^{2}$ on $S^{2}$ and the

8399: standard complex structures on all others components define a

8400: stable algebraic reducuble curve $P_{e \subset\Ga, \tau}$. Thus

8401: we obtain an embedding

8402: \begin{equation}

8403: \psi_{e \subset \Ga} \colon \sM_{1}^{2} \to \ov{\sM_{g}}.

8404: \end{equation}

8405: The moduli space

8406: \begin{equation}

8407: \sM_{1}^{2} = \PP^{1} - (\si, \si', \si'')

8408: \end{equation}

8409: is the projective line without 3 points. These 3 points

8410: correspond to 3 possibility to divide 4 points $p_{e_{1}},

8411: p_{e_{2}}, p_{e'_{1}}, p_{e'_{2}}$ in 2 pairs that is a choice of

8412: a point of order 2 on an elliptic curve.

8413: 

8414: It is easy to see that

8415: 

8416: \begin{prop}

8417: \begin{enumerate}

8418: \item The map $\psi_{e \subset \Ga}$ (9.24) can be extended to the map

8419: \begin{equation}

8420: \psi_{e \subset \Ga} \colon \PP^{1} \to \ov{\sM_{g}};

8421: \end{equation}

8422: \item 

8423: \begin{equation}

8424: \psi_{e \subset \Ga} (\si) = p_{e} \in  P_{\Ga}

8425: \end{equation}

8426: where $\si$ is given by (9.23) and $P_{\Ga}$ is a large limit curve with

8427: fixed node corresponding to the edge $e$,

8428: \item

8429: \begin{equation}

8430: \psi_{e \subset \Ga} (\si') = P_{\Ga'};

8431: \end{equation}

8432: and

8433: \item

8434: \begin{equation}

8435: \psi_{e \subset \Ga} (\si'') = P_{\Ga''};

8436: \end{equation}

8437: where the  triple

8438: \begin{equation}

8439: (e \subset \Ga), (e_{new} \subset \Ga'), ( e_{new} \subset \Ga'')

8440: \end{equation}

8441: is a nest of flags (9.17).

8442: \item Now we can identify the tangent direction

8443: \begin{equation}

8444: T\psi_{e \subset \Ga} (\PP^{1})_{P_{\Ga}} = \C_{e}

8445: \end{equation}

8446: from the decomposition (9.15) with the corresponding node of the

8447: double canonical curve. We obtain the canonical pluricanonical  model.

8448: \end{enumerate}

8449: \end{prop}

8450: \begin{rmk} If  edge $e$ is a loop with a vertex $v$ such that

8451: the star $S(v) = e, e'$ then $P_{v}$ is a rational curve with one

8452: double point $p_{e}$ and the smooth point $p_{e'}$. The operation

8453: of connected summing  around double point $p_{e}$ gives a smooth

8454: 2-torus $T^{2}$ with fixed point $p_{e'}$ and the isotopy class

8455: $a \in H_{1}(T^{2}, \Z)$ which is the class of the neck of gluing

8456: tube. The class $a$ mod 2 gives a point of order 2 on $T^{2}$.

8457: The space $\sM_{1}^{2}$ of complex structures on $T^{2}$  with

8458: such additional data is $\C^{*} - 1$ that is $\PP^{1}$ without 3

8459: points again. In this case the rational curve $\psi_{e \subset

8460: \Ga} (\sM_{1}^{2})$ admits the compactification by the double

8461: point corresponding to $P_{\Ga}$.

8462: \end{rmk}

8463: 

8464: So, the Deligne-Mumford  compactification $\ov{\sM_{g}}$ contains

8465: the configuration $\sC$ of rational curves

8466: \begin{equation}

8467: \{ C \} = \sC \text{ every } C = \psi_{e \subset \Ga} (\PP^{1})

8468: \end{equation}

8469: for some flag $e \subset \Ga$.  We can consider this configuration of rational

8470: curves as a reducible curve

8471: \begin{equation}

8472: \sC = \bigcup C

8473: \end{equation}

8474: that is the union of all components. It is easy to see that

8475: 

8476: \begin{prop}

8477: \begin{enumerate}

8478: \item for every  component $C$ of $\sC$

8479: \begin{equation}

8480:   C \bigcap \sP  = P_{\Ga} + P_{\Ga'} + P_{\Ga''}

8481: \end{equation}

8482: where $(\Ga, \Ga', \Ga'')$ is a nest of graphs (9.17 );

8483: \item for a pair $C, C'$ of componets  either the intersection $C \bigcap C'$

8484:  is empty or it is transversal and

8485:  \begin{equation}

8486:  C \bigcap C' \in \sP

8487:  \end{equation}

8488:  \item the set $S(\Ga)$ of components through every point

8489:   $P_{\Ga} \in \sP$ is enumerated by the set $E(\Ga)$;

8490: \end{enumerate}

8491: \end{prop}

8492: 

8493: We saw that if an edge $e \in E(\Ga)$ is not a loop then the map

8494: $\psi_{e \subset \Ga}$ (9.26) is an embedding and the component $C$

8495: (9.24) of the reducible curve $\sS \subset \ov{\sM_{g}}$ is a

8496: smooth rational curve with fixed 3 points $ ( P_{\Ga}, P_{\Ga'},

8497: P_{\Ga''})$ such that the corresponding 3 graphs  form a nest

8498: (9.17). Over every of such points the tangent spaces admit

8499: decompositions (9.15). Geometry of the other curves from the family

8500: of curves parametrized by $C$ is very near to the geometry of

8501: large limit curves: let $e \in E(\Ga)$, $v, v' = \p e$ and $P_{E,

8502: \si}$ is the point of $C$ corresponding to the elliptic curve

8503: (9.21) with a point of order 2 (9.23). Then $P_{E, \si}$ has 2g-4

8504:  old components

8505:  \begin{equation}

8506:  \bigcup_{v'' \neq v, v'} P_{v''}

8507:  \end{equation}

8508: with triples of points and 3g-1 nodes $p_{e'}, e \neq e'$ and one

8509: new component $P_{v, v'}$  with the quadruple of points (see

8510: (9.16) and 1), 2) and 3) after this formula). Then

8511: 

8512: \begin{enumerate} \item

8513: the canonical class $K_{P_{E, \si}}$ is a line bundle: a

8514: restriction of it to every component $P_{v''}, v'' \neq v, v' $

8515: is the sheaf of meromorphic differentials $\om$ with simple poles

8516: at $p_{e}, p_{e'}, p_{e''}$  where $\{e,e', e''\} = S(v)$;

8517: \item the

8518: restriction of the canonical class to the component $P_{v, v'}$

8519: is the sheaf of meromorphic differentials $\om$ with simple poles

8520: at $p_{e_{1}}, p_{e_{2}}, p_{e'_{1}}, p_{e''_{2}}$ (see (9.16)).

8521: \item

8522: Thus 

8523: \begin{equation}

8524: c (K_{P_{E, \si}}) = (2,1,...,1) \in NS_{P_{E, \si}}

8525: \end{equation}

8526: where the first coordinate corresponds to $P_{v, v'}$.

8527: \item Again

8528: every holomorphic section $s$ of the canonical class  is a

8529: collection of meromorphic differentials $\{ \om_{v''}\}$ on the

8530: components $P_{v''}$ with poles at $p_{e}, p_{e'}, p_{e''}$ where

8531: $\{e,e', e''\} = S(v)$ and a meromorphic differential $\om_{v,

8532: v'}$ on the component $P_{v, v'}$ with poles at the quadruple

8533: with the same constraints (1.29) and (1.30).

8534: \item

8535: The canonical map defined by the complete linear system $\vert K_{P_{E,

8536: \si }}\vert$

8537: \begin{equation}

8538: \phi_{K} \colon P_{E, \si} \to \PP^{g-1}

8539: \end{equation}

8540: sends $P_{v''}$ to a configuration of lines and $P_{v, v'}$ to a

8541: conic  in $\PP^{g-1}$.

8542: \item Again the dimension of the space of quadratic differentials

8543: on $P_{E, \si}$ is equal to 3g-3 and this curve is an orbifold

8544: smooth point of $\ov{\sM_{g}}$.

8545: \item

8546: The double  canonical map of $P_{E, \si}$  given  by the complete

8547: linear system $\vert 2K_{\Ga}\vert$ is an embedding

8548: \begin{equation}

8549: \phi_{2K_{P_{E, \si}}} \colon P_{E, \si} \to \PP^{3g-4} = \PP

8550: T\ov{\sM_{g}}_{P_{E, \si}}.

8551: \end{equation}

8552: (see (9.11)).

8553: \item The images  of nodes

8554: \begin{equation}

8555: \{ \phi_{2K_{P_{E, \si}}}(p_{e'}) \}, \quad e \neq e' \in

8556: E(\Ga)

8557: \end{equation}

8558: define the decomposition of the restriction of the tangent bundle

8559: \begin{equation}

8560: T\ov{\sM_{g}} \vert_{C} = TC \bigoplus (\bigoplus_{e \neq e' \in

8561: E(\Ga)} L_{e'})

8562: \end{equation}

8563: where the fiber of the line bundle $L_{e'}$ over a point is the

8564: component of the decomposition (9.15).

8565: \item Thus every line bundle $L_{e'}$ from the previous decomposition

8566: is the tautological line bundle

8567: \begin{equation}

8568: L_{e'} = \Oh_{C} (-1),

8569: \end{equation}

8570: \item hence the previous decomposition is

8571: \begin{equation}

8572: T\ov{\sM_{g}} \vert_{C} = \Oh_{C}(2)\bigoplus (3g-4) \Oh_{C}(-1).

8573: \end{equation}

8574: \item The restriction to $C$ of the canonical class of $\ov{\sM_{g}}$

8575: is

8576: \begin{equation}

8577: K_{\ov{\sM_{g}}} \vert_C = \Oh_{C}(3(g-2)).

8578: \end{equation}

8579: \end{enumerate}

8580: 

8581: \subsection{ Special 2-parameter deformations of large limit curves}

8582: 

8583: Let $v \subset \Ga$ be a pair of a 3-valent graph $\Ga$ and fixed

8584: vertex $v \in V(\Ga)$ and in the star  $ S(v) = e \bigcup e_{1}

8585: \bigcup e_{2}$ let us choose a pair $e_{1}$ and $e_{2}$ of edges.

8586: Then we have the pair of vertices $v_{1}$ and $v_{2}$ defined by

8587: equalities $\p e_{1} = v, v_{1}$ and $\p e_{2} = v, v_{2}$.

8588: 

8589: Thus such choices are given by 3-flag on a graph $\Ga$

8590: \begin{equation}

8591:  v \subset e \subset \Ga

8592: \end{equation}

8593: where

8594: \begin{equation}

8595: e_{1} \bigcup e_{2} = S(v) - e

8596: \end{equation}

8597:  Now let us blow down

8598: edges $e_{1}$ and $e_{2}$. We obtain a new graph $\Ga_{new}$

8599: with a new 5-valent vertex $v_{new} \in V(\Ga_{new})$ and fixed

8600: additional choices on the star of $v_{new}$:

8601: \begin{enumerate}

8602: \item fixed edge $e \in S(v_{new}) $ through $v_{new}$ and

8603: \item a partition of other quadruple edges from the star $S(v_{new})$

8604:  in two pairs $S(v_{1}) - e_{1}$ and $S(v_{2} - e_{1})$.

8605: \end{enumerate}

8606: 

8607: 

8608: There are exist 15 such choices divided into 5 triples corresponding

8609: to the first choice of $e' \in S(v_{new}) $. For every such a choice

8610:  let us blow up from $v_{new}$ two new edges $e'_{1}$ and $e'_{2}$

8611:  such a way that

8612: \begin{enumerate}

8613: \item  $v_{new}$ becomes a new 3-valent vertex $v'$

8614:  and $S(v') = e', e'_{1}, e'_{2}$,

8615:  \item two vertices appear $v'_{1}$ and $v'_{2}$ such that

8616:  $\p e'_{1} = v', v'_{1}$ and $\p e'_{2} = v', v'_{2}$.

8617:  \end{enumerate}

8618: Thus we get a new 3-flag

8619: \begin{equation}

8620: v' \subset e' \subset \Ga'

8621: \end{equation}

8622: and full collection of choices gives a collection of 3-flags

8623: \begin{equation}

8624: (v \subset e \subset \Ga),   (v' \subset e' \subset \Ga'), (v''

8625: \subset e'' \subset \Ga''), ........

8626: \end{equation}

8627: We call this collection of 15 3-flags  a {\it nest of 3-flags}.

8628: 

8629: A partial case of the choice is the following:

8630: \begin{enumerate}

8631: \item let us blow down just one edge, for example, $e_{1}$,

8632: \item then we get 4-valent vertex $v_{new}$ with the star

8633: $S(v') = e, e_{2}, e_{3}, e_{4}$  where $e_{3} \bigcup e_{4} =

8634: S(v_{1}) - e_{1}$.

8635: \item Now let us change $e$ by any edge from the pair $e_{3},

8636: e_{4}$, for example, by $e_{3}$ and blow up the vertex $v_{new}$

8637: to a new edge $e'$ with vertices $\p e' = v', v''$ such that

8638: \begin{equation}

8639: S(v') = e' \bigcup e \bigcup e_{3} \text{ and } S(v'') = e'

8640: \bigcup e_{2} \bigcup e_{4}.

8641: \end{equation}

8642: Then we have a pair of 3-flags

8643: \begin{equation}

8644: ((v \subset e \subset \Ga)) \text{ and } (v' \subset e' \subset

8645: \Ga')

8646: \end{equation}

8647: from the same nest of 3-flags such that the flags

8648: \begin{equation}

8649: (e_{1} \subset \Ga) \text{ and } (e' \subset \Ga')

8650: \end{equation}

8651: belong to the same nest (9.17).

8652: \item if in this construction we replace $e$ by $e_{4}$ we get the

8653: nest triple

8654: \begin{equation}

8655: (e_{1} \subset \Ga), (e' \subset \Ga'), (e'' \subset \Ga'')

8656: \end{equation}

8657: (see (9.17)).

8658: \end{enumerate}

8659: 

8660: 

8661: Any flag $(v \subset e \subset \Ga)$ defines

8662: \begin{enumerate}

8663: \item the reducible curve $P_{\Ga}$,

8664: \item its irreducible component $P_{v}$,

8665: \item a pair of its points $p_{e_{1}}, p_{e_{2}} \subset P_{v}$ and

8666: \item

8667: two components $P_{v_{1}}$ and $P_{v_{2}}$.

8668: \end{enumerate}

8669: Remove nodes $p_{e_{i}}, i = 1, 2$ and glue components by tubes

8670: that is consider the connected sum

8671: \begin{equation}

8672:  P_{v_{1}} \#_{p_{e_{1}}} P_{v} \#_{p_{e_{2}}} P_{v_{2}} = S^{2}.

8673: \end{equation}

8674: This is 2-sphere with fixed 5 points with fixed the point $p_{e}$

8675: and two pairs of points corresponding to the following two pairs of edges

8676: \begin{equation}

8677: (S(v_{1}) - e_{1}), (S(v_{2}) - e_{2}).

8678: \end{equation}

8679: The moduli space of complex structures on $S^{2}$ with fixed 5

8680: points subjecting such partition conditions is

8681: \begin{equation}

8682: S((v \subset e \subset \Ga)) = \PP^{1} \times S^{2}(S^{2}

8683: \PP^{1}) / PGL(2, \C),

8684: \end{equation}

8685: so it is rational affine surface. Every complex line with 5 points can

8686: be uniquely extended to a reducible curve adding the corresponding

8687: complex lines with 3 points. Thus the surface $S((v \subset e

8688: \subset \Ga))$ is embedded

8689: \begin{equation}

8690: \psi_{(v \subset e \subset \Ga)} \colon S((v \subset e \subset

8691: \Ga)) \to \ov{\sM_{g}}

8692: \end{equation}

8693: and its compactification belongs to

8694: \begin{equation}

8695: \ov{\psi_{(v \subset e \subset \Ga)}S((v \subset e \subset \Ga))}

8696: \subset \ov{\sM_{g}}.

8697: \end{equation}

8698: The compactification divisor is

8699: \begin{equation}

8700: \ov{\psi_{(v \subset e \subset \Ga)}S((v \subset e \subset \Ga))}

8701: - \psi_{(v \subset e \subset \Ga)}S((v \subset e \subset \Ga)) =

8702: \bigcup C

8703: \end{equation}

8704: where 15 components of the curve $\sC$ corresponding to the nests

8705: (9.48) and (9.52).

8706: 

8707: Again the Deligne-Mumford  compactification $\ov{\sM_{g}}$

8708: contains the configuration $\sS$ of rational surfaces

8709: \begin{equation}

8710: \{ S \} = \sS \text{ every } S = \psi_{v \subset e \subset \Ga}

8711: (S((v \subset e \subset \Ga)))

8712: \end{equation}

8713: for some 3-flag $v \subset e \subset \Ga$.  We can  consider 

8714: this configuration of

8715: rational surfaces as a reducible surface

8716: \begin{equation}

8717: \sS = \bigcup S

8718: \end{equation}

8719: that is the union of all components. It is easy to see that

8720: 

8721: \begin{prop}

8722: \begin{enumerate}

8723: \item for every  component $S$ of $\sS$

8724: \begin{equation}

8725:   S \bigcap \sP  = \bigcup C

8726: \end{equation}

8727: where the curves $ C $ belongs the collection  (9.32);

8728: \item for a pair $S, S'$ of componets the intersection $C \bigcap C'$

8729:  is either empty or it is transversal and

8730:  \begin{equation}

8731:  C \bigcap C' \subset  \sC

8732:  \end{equation}

8733: \end{enumerate}

8734: \end{prop}

8735: 

8736: We saw that if an edge $e \in E(\Ga)$ from 3-flag $v \subset e

8737: \subset \Ga$ is not a loop then the map $\psi_{v \subset e \subset

8738: \Ga}$ (3.12) is an embedding and the component $S$ (3.13) of the

8739: reducible surface $\sS \subset \ov{\sM_{g}}$ is a  rational

8740: surface with fixed 15 components of the reducible curve $\sC$.

8741: Over any of such irreducuble curve the restriction of the

8742: tangent bundle of $\ov{\sM_{g}}$ admits decompositions (9.41) and

8743: (9.43). The structure of a general stable curve $P_{s}, s \subset S$

8744: from the 2-dimensional

8745:  family parametrized by $S$

8746:  is a slightly different from  the geometry of the large limit curves:

8747:   $P_{s}$ has 2g-5

8748:  old components

8749:  \begin{equation}

8750:  \bigcup_{v'' \neq v, v_{1}, v_{2}} P_{v''}

8751:  \end{equation}

8752:  and one

8753: new component $P_{v, v_{1}, v_{2}}$  with 5 points. Then

8754: 

8755: \begin{enumerate} \item

8756: the canonical class $K_{s}$ is a line bundle: the restriction of it

8757: to every component $P_{v''}, v'' \neq v, v_{1}, v_{2} $ is the

8758: sheaf of meromorphic differentials $\om$ with simple poles at

8759: $p_{e}, p_{e'}, p_{e''}$  where $e \bigcup e' \bigcup e'' = S(v)$;

8760: \item the

8761: restriction of the canonical class to the component $P_{v, v_{1},

8762: v_{2} }$ is the sheaf of meromorphic differentials $\om$ with

8763: simple poles at $p_{e}, p_{e_{1}}, p_{e_{2}}, p_{e'_{1}},

8764: p_{e'_{2}}$.

8765: \item

8766: Thus in terms of subsection 1.1

8767: \begin{equation}

8768: c (K_{P_{s}}) = (3,1,...,1) \in NS_{P_{s}}.

8769: \end{equation}

8770: where the first coordinate corresponds to $P_{v, v'_{1}, v_{2}}$.

8771: \item

8772: The canonical map by the complete linear system $\vert

8773: K_{P_{s}}\vert$

8774: \begin{equation}

8775: \phi_{K} \colon P_{s} \to \PP^{g-1}

8776: \end{equation}

8777: sends $P_{v''}$ to a configuration of lines and $P_{v, v_{1},

8778: v{2}}$ to a cubic  in $\PP^{g-1}$.

8779: \item Again the dimension of the space of quadratic differentials

8780: on $P_{s}$ is equal to 3g-3 and this curve is an orbifold smooth

8781: point of $\ov{\sM_{g}}$.

8782: \item

8783: The double  canonical map of $P_{E, \si}$  given  by the complete

8784: linear system $\vert 2K_{P_{s}}\vert$ is an embedding

8785: \begin{equation}

8786: \phi_{2K_{P_{s}}} \colon P_{s} \to \PP^{3g-4} = \PP

8787: T\ov{\sM_{g}}_{s}.

8788: \end{equation}

8789: \item The images  of nodes

8790: \begin{equation}

8791: \{ \phi_{2K_{P_{s}}}(p_{e'}) \}, \quad e \neq e_{1}, e_{2} \in

8792: E(\Ga)

8793: \end{equation}

8794: define the decomposition of the restriction of the tangent bundle

8795: \begin{equation}

8796: T\ov{\sM_{g}} \vert_{S} = TS \bigoplus (\bigoplus_{e \neq e_{1},

8797: e_{2} \in E(\Ga)} L_{e})

8798: \end{equation}

8799: where the fiber of the line bundle $L_{e}$ over a point is the

8800: component of the decomposition (9.43).

8801: \end{enumerate}

8802: 

8803: \subsection{Modular configuration}

8804: 

8805: We understands the moduli subspaces (9.33) and (9.60) as partial cases of the general construction.

8806: For  any graph $\Ga$ with vertices of any valence $ > 2$ consider the corresponding reducible curve

8807: \begin{equation}

8808: P_{\Ga} = \bigcup_{v \in V(\Ga)} P_v

8809: \end{equation}

8810: where each component equals to $\PP^1$  with $\vert S(v) \vert$

8811: fixed points of transversal intersections with other components along the incidence correspondence  of the graph. If our graph

8812:  isn't 3-valent such curve has moduli. These moduli are defined by  the positions of  $\vert S(v) \vert$

8813:  points on the component $P_v$ and by the combinatoric of the set of expansions $\sE(\Ga)$ (see subsection 6.3). For example, if

8814: $\Ga$ is the result of one contraction of 3-valent graph $\Ga_{max}$ the corresponding moduli space $\sM^2_1$ (9.24) is a 3-cover of the moduli space of smooth elliptic curves. 

8815: For graphs with one 5-valent vertex the corresponding moduli space is the rational surface $S$ from the configuration

8816:  (9.60) and so on. 

8817: 

8818: Every such moduli space $\sM_\Ga$ is embedded into the Deligne-

8819: Mumford compactification and it admits itself the Deligne-Mumford compactification 

8820: \begin{equation}

8821: \ov{\sM_\Ga} \subset \ov{\sM_g}

8822: \end{equation}

8823: such that

8824: \begin{equation}

8825: \ov{\sM_\Ga} = \sM_v \bigcup (\bigcup_{\Ga'> \Ga}) \ov{\sM_{\Ga'}} 

8826: \end{equation}

8827: where  graph $\Ga'$ is an expansion of $\Ga$.

8828: 

8829: Obviously the maximal dimension of such moduli space

8830: is $2g-3$. 

8831: 

8832: So in the compactification $\ov{\sM_g}$ of the moduli space of  genus $g$

8833: curves  we have the configuration of the moduli spaces

8834: \begin{equation}

8835: \sM = \bigcup_{\Ga} \sM_\Ga \subset \ov{\sM_g}

8836: \end{equation}

8837: such that 

8838: \begin{enumerate}

8839: \item

8840: \begin{equation}

8841: \ov{\sM_\Ga'} \bigcap \ov{\sM_{\Ga''}}

8842: = \bigcup_{\Ga > \Ga', \Ga'' } \sM_\Ga;

8843: \end{equation}

8844: \item the restriction of the tangent bundle of $\ov{\sM_g}$

8845: to every component $\sM_\Ga$ admits the decomposition like (9.68).

8846: \end{enumerate} 

8847: 

8848: 

8849: 

8850: 

8851: 

8852: \subsection{ $Pic_0(P_\Ga)$ and moduli spaces $M_{vb}^{ss}(P_\Ga)$ of ll-curves}

8853: 

8854: For a singular curve the moduli spaces of  vector bundles are not compact. These moduli spaces admit the compactifications by torsion free sheaves. These sheaves are not local free  only over

8855: nodes but the theory of compactification is very close to

8856: such theory for algebraic surfaces. All details of this theory

8857: can be found in Artamkin paper \cite{Ar}.

8858: 

8859: For every torsion free sheaf $F$ on $P_\Ga$ we have the standard exact surface

8860: \begin{equation}

8861: 0 \to F \to F^{**} \to C(F) \to 0

8862: \end{equation}

8863: where 

8864: \begin{enumerate}

8865: \item the quotient sheaf has the form

8866: $$

8867: Supp C(F) \subset \bigcup_{e \in E(\Ga)} p_e;

8868: $$

8869: \item the double dual sheaf $F^{**}$ is a vector bundle on the normalization of the curve.

8870: \end{enumerate}

8871: The cardinality of $Supp F$ is the measure of deviation from the local freedom of

8872:  $F$. Let this number be

8873: \begin{equation}

8874: \vert Supp C(F) \vert = c_2(F).

8875: \end{equation}

8876: So $c_2(F) = 0$ implies $F$ is local free thus it is a vector bundle.

8877: 

8878: A vector bundle $E$ on $P_\Ga$ is called topologicaly trivial if

8879: the restrictions $E \vert_{P_v}$ are trivial for all $v \in V(\Ga)$.

8880: Below we describe the moduli spaces of topologicaly trivial 

8881: \begin{enumerate}

8882: \item line bundles $Pic_0 (P_\Ga)$,

8883: \item rk 2 semi-stable bundles $M_{vb}^{ss}$;

8884: \item the compactification $J(P_\Ga) = \ov{Pic_0(P_\Ga)}$ of 

8885: $Pic_0(P_\Ga)$ by torsion free sheaves $F$ (for which obviously

8886: $F^{**} \in Pic_0(P_\Ga)$);

8887: \item  the compactification $M^{ss}(P_\Ga) = \ov{M^{ss}_{vb}(P_\Ga)}$ of 

8888: $Pic_0(P_\Ga)$ by torsion free sheaves $F$ (for which obviously

8889: $F^{**} \in M^{ss}_{vb}(P_\Ga)$).

8890: \end{enumerate}

8891: 

8892: 

8893: Since each  component $P_{v}$ is  projective line the restriction of any topologicaly trivial line bundle equals

8894: \begin{equation}

8895: L \vert_{P_{v}} = \Oh_{P_{v}}.

8896: \end{equation}

8897: 

8898: 

8899: 

8900: To describe a line bundle on $P_\Ga$ consider a collection of any trivializations of $L$ on all components $P_v$ of $P_\Ga$ and denote the line bundle with such addition structure by $L_0$.

8901: 

8902: Now to get a line bundle $L$ on $P_{\Ga}$ we have to concord every

8903:  pair of line bundles $\Oh_{P_{v}}, \Oh_{P_{v'}}$

8904: at common point $p_{e}$ if $v, v' = \p e$. Under our trivializations such concordance is

8905: given by a multiplicative constant $ a (\vec e) \in \C^{*}$ such that under the orientation reversing involution $i_e$

8906: $$

8907: a(i_e(\vec e)) = a(\vec e)^{-1}

8908: $$

8909: (compare with (6.86)). Thus $L_0$ defines a map

8910: $$

8911: a \colon \vec E(\Ga) \to \C^*

8912: $$

8913: (compare with (6.85)). 

8914: 

8915: The changing of trivialization is given by a function

8916: $$

8917: \tilde{\la} \colon V(\Ga) \to \C^*

8918: $$

8919: (compare with (6.90)) which acts on the functions $a$  by the formula

8920: $$

8921: \tilde{\la}(a(\vec e)) = \tilde{\la}(v_s) \cdot a(\vec e) \cdot \tilde{\la}(v_t)^{-1}

8922: $$

8923: where $\p \vec e = v_s \bigcup v_t$ (compare with (6.91)).

8924: 

8925: Let $\sA_{\C}$ be the space of trivialized topologicaly trivial line bundles that is the space of functions $a$ 

8926:  and $\sG_{\C}$ be the group of trivializations.

8927: Then in terms of subsection 6.6 the space  $\sA_{\C}$ is the space of flat $\C^*$-connections,  $\sG_{\C}$ is the complexification of the unitar 

8928: abelian gauge group and

8929: \begin{equation}

8930: Pic_0(P_\Ga) = \sA_{\C} / \sG_{\C}.

8931: \end{equation}

8932: Thus 

8933: \begin{equation}

8934: Pic_0(P_\Ga) = Hom(\pi_1(\Ga), \C^*) = S^a_g

8935: \end{equation}

8936: is the abelian Schottky space (2.68).

8937: 

8938: 

8939: To describe a rk 2 bundle $E$ on $P_\Ga$ consider a collection of its trivializations  on all components $P_v$ of $P_\Ga$ and denote the  bundle with such addition structure by $E_0$.

8940: 

8941: Now to get a rk 2 vector  bundle $E$ on $P_{\Ga}$ we have to glue every

8942:  pair of line bundles $\Oh_{P_{v}} \oplus \Oh_{P_{v}}, \Oh_{P_{v'}} \oplus \Oh_{P_{v'}}$

8943: at common point $p_{e}$ if $v, v' = \p e$. Under our trivializations such concordance is

8944: given by a matrix $ a (\vec e) \in SL(2, \C)$ with the standard

8945:  property (6.57).

8946: 

8947: 

8948: 

8949:  Thus $E_0$ defines a flat $SL(2, \C)$-connection

8950: $$

8951: a \colon \vec E(\Ga) \to SL(2, \C)

8952: $$

8953: (compare with (6.56)). 

8954: 

8955: The changing of the trivialization is given by a function

8956: $$

8957: \tilde{g} \colon V(\Ga) \to SL(2, \C)

8958: $$

8959: (compare with (6.63)) which acts on the functions $a$  by the formula

8960: $$

8961: \tilde{g}(a(\vec e)) = \tilde{u}(v_s) \cdot a(\vec e) \cdot \tilde{g}(v_t)^{-1}

8962: $$

8963: where $\p \vec e = v_s \bigcup v_t$ (compare with (6.65)).

8964: 

8965: Thus the space  $\sA_{ \C}$  of trivialized topologicaly trivial rk 2 bundles coincides with the space of  flat  $SL(2, \C)$-connections

8966:  and  the group of trivializations $\sG_{ \C}$ coincides with the complex gauge group.

8967: Then in terms of subsection 6.5

8968: \begin{equation}

8969: M^{ss}_{vb}(P_\Ga) = \sA_{\C} / \sG_{\C}.

8970: \end{equation}

8971: Thus 

8972: \begin{equation}

8973: M^{ss}_{vb}(P_\Ga) = \CLRep(\pi_1(\Ga), SL(2, \C)) = S_g

8974: \end{equation}

8975: is the  Schottky space (see subsection 3.2).

8976: 

8977: Let us return to the subsection 3.2 and the handlebodies (6.9). Every such

8978:  handlebody defines the exact sequence

8979: \begin{equation}

8980: 1 \to  F_g \to \pi_1 (\Si_\Ga) \to \pi_1 (\Ga) \to 1

8981: \end{equation}

8982: where the normal subgroup is a free group with $g$ generators. Using this sequence we can choose the standard generators of $\pi_1 (\Si_\Ga)$ 

8983: $$

8984: \pi_1 = <a_1, ... ,a_g, b_1, ... , b_g \vert \prod_{i=1}^g [a_i, b_i] = 1>

8985: $$

8986: such that

8987: \begin{equation}

8988: F_g = <a_1, ... , a_g>

8989: \end{equation}

8990: $$

8991: \pi_1 (\Ga) =  <b_1, ... , b_g>.

8992: $$

8993: Then we can consider the space of classes of $SL(2, \C)$-representations

8994: \begin{equation}

8995: \CLRep(\pi_1(\Ga), SL(2, \C)) = S_g

8996: \end{equation}

8997: as a family of flat vector bundles on $\Si_\Ga$. If we fix a complex structure

8998: $I$ on $\Si_\Ga$ we get the forgetful map

8999: \begin{equation}

9000: f_I \colon S_g \to M^{ss}(\Si_I)

9001: \end{equation}

9002: (3.44).

9003: The space of complex structures on $\Si_\Ga$ under fixing of the basis  is

9004: the  Teichmuller space $\tau_g$ (5.31) and we can consider the universal family 

9005: of the forgetful maps

9006: \begin{equation}

9007: F \colon S_g \times \tau_g \to \sM

9008: \end{equation}

9009: where $u \colon \sM \to \tau_g$ is the universal family of moduli spaces of vector bundles such that $u^{-1}(I) =  M^{ss}(\Si_I)$.

9010: 

9011: Now it is easy to see that 

9012: \begin{enumerate}

9013: \item the ll-curve $P_\Ga$ is contained by the Deligne-Mumford 

9014: compactification of $\tau_g$,

9015: \item the moduli space $ \ov{M^{ss}(P_\Ga)}$ is contained by the compactification of

9016: $\sM$ corresponding to  the Deligne-Mumford 

9017: compactification of $\tau_g$,

9018: \item the specialization of the universal forgetful map F ( . ) on $P_\Ga$ coincides with the standard embedding

9019: \begin{equation} 

9020: F_{P_\Ga} = i \colon S_g =  M_{vb}^{ss}(P_\Ga)  \to  \ov{M^{ss}(P_\Ga)}

9021: \end{equation}

9022: described before.

9023: \end{enumerate}

9024: 

9025: From this it easy to prove  the following partial solution of the Narasimhan problem

9026: 

9027: \begin{prop} For a general smooth algebraic curve $\Si_I$ of genus $g$ the image of the forgetful

9028: map $f_I : S_g \to M^{ss}(\Si)$ (3.44) is Zariski dense in this moduli space.

9029: \end{prop}

9030: Indeed, this fact is true for $P_\Ga \in \sM_g$ and this property is preserved

9031: by small deformations in $\sM_g$.  (I thank Gerd Detloff for the simplification of my arguments.)

9032: 

9033: This fact is quite productuve even in the classical abelian case (see subsection 2.4): the standard embedding

9034: \begin{equation} 

9035: i \colon (\C^*)^g = S^a_g \to \ov{Pic_0(P_\Ga)}

9036: \end{equation}

9037: is a limit of the standard forgetful covers $f_I \colon (\C^*)^g \to J(\Si_I)$

9038: (see (2.68)- (2.73)). We can see what happens with the classical theta functions (2.73) under this limit procedure. In particular we can see why 

9039: $\delta$-function (2.46) appears. 

9040: 

9041: Moreover, the theory of classical theta functions can be constructed by CST-method from subsection 2.3 on $Pic_0(P_\Ga)$ and can be extended to any smooth curve using

9042:  the classical holomorphic flat connection perfectly described by Welters in

9043: \cite{W}. We use the same strategy as well as in the non-abelian case:

9044: CST-constructions from section 6 can be interpreted as the theory of non-abelian theta functions on $ M_{vb}^{ss}(P_\Ga)$. After this  theory

9045: can be extended to any smooth curve using the Hitchin projective flat connection. In this situation to avoid the meromorphicy of the forgetful map

9046: we apply the geometry of Lagrangian and Legandrian cycles and Bohr-Sommerfeld

9047: quantization. 

9048: 

9049: Here we use  existence of an algebraic-geometrical compactification

9050: of the spaces of vector bundles on ll-curves. Now we describe these compactifications precisely.   

9051: 

9052: 

9053: \subsection{Compactifications and AG-theory of  theta functions

9054:  for ll-curves}

9055: 

9056: Both of our spaces $Pic_0(P_\Ga)$ and $M^{ss}_{vb}(P_\Ga)$ are

9057: affine varieties admitting canonical compactifications.

9058: The compactification in the abelian case  will be called {\it jacobian}:

9059:  \begin{equation}

9060: J(P_{\Ga}) = \ov{Pic_{0}(P_{\Ga})}.

9061: \end{equation}

9062: 

9063:  Every rk 1

9064: torsion free sheaf $F$ is given by a function

9065: $a$ as before  but  we have to

9066: compactify the target space  by two points $0$ and $\infty$. So the new target

9067: space is the projective line $\PP^{1}_{\C^{*}}$ with fixed points

9068: $0$ and $\infty$ or fixed standard $\C^{*}$-action

9069: (distinguishing fixed points $0$ and $\infty$).

9070: 

9071: So the compactification is

9072: \begin{equation}

9073: J(P_{\Ga}) = (\PP^{1}_{\C^{*}})^{3g-3} / \sG_{\C^{*}}.

9074: \end{equation}

9075: The boundary is the union of $6g-6$ divisors

9076: \begin{equation}

9077: D_{\Ga} = J(P_{\Ga}) - Pic_{0}(\Ga) = \bigcup_{e \in E(\Ga)}

9078: (D_{+}(e) \bigcup D_{-}(e))

9079: \end{equation}

9080: where

9081: \begin{equation}

9082: D_{+}(e) = \{ a \vert a (\vec e) = 0\},

9083: \end{equation}

9084: \begin{equation}

9085: D_{-}(e) = \{ a \vert a (\vec e) = \infty\}

9086: \end{equation}

9087: and

9088: \begin{equation}

9089: D_{+}(e) \bigcup D_{-} (e)= \{F \vert  p_{e} \subset Sing F \}.

9090: \end{equation}

9091: 

9092: Now consider an abelian spin network $\Ga_w$ (6.99)  on our graph $\Ga$

9093: given by a function $w$ (6.99) subjecting to  the conditions (6.100).

9094: Then we have  divisor

9095: \begin{equation}

9096: D_w = \sum w(e)\cdot D_{sign (w(e))}

9097: \end{equation}

9098: with properties

9099: \begin{enumerate}

9100: \item this divisor is principal that is it is divisor of rational

9101: function $\theta_w$;

9102: \item this function is regular on $Pic_0(P_\Ga)$ being the  limit of the ordinary theta

9103: functions for smooth curves;

9104: \item if $\Ga_w \in SNW^k_a$ that is $w $ is of level $ k$ (6.104)-(6.107) Then

9105: we can fix a pole to make all $w(e)$ non negative and get the canonical basis

9106: of theta functions with characteristics. 

9107: \end{enumerate}

9108: 

9109:  Every rk 2

9110: torsion free sheaf $F$ is given by the same type function

9111: $a$  but  we have to consider as the target space  the compactification

9112: \begin{equation}

9113: SL(2, \C) \subset \PP^3 = \PP End(\C^2).

9114: \end{equation}

9115: with fixed quadric $Q$.

9116: 

9117: So 

9118: \begin{equation}

9119: M^{ss}(P_{\Ga}) = \ov{M^{ss}_{vb}}(P_\Ga) = (\PP^{3})^{E(\Ga)} / \sG_{\C}

9120: \end{equation}

9121: where the gauge group is the same (9.79).

9122: 

9123:  The boundary $D = M^{ss}(P_{\Ga}) - M^{ss}_{vb}(P_{\Ga}) $ is the union of $3g-3$ divisors

9124: \begin{equation}

9125: D =   \bigcup_{e \in E(\Ga)}

9126: D(e). 

9127: \end{equation}

9128: Recall (see the text after formula (6.33) and Fig.12) that we can remove the edge $e$ from $\Ga$ and forget  the vertices $v_1, v_2 = \p e$. We get a new graph

9129: $\Ga_e$ of genus $g-1$. Then the corresponding divisor can be expressed

9130: \begin{equation}

9131: D(e) = Q_e \times M^{ss}(P_{\Ga_e}).

9132: \end{equation}

9133: 

9134: For every spin network $\Ga_w$ of level 1 the divisor

9135: \begin{equation}

9136: \Theta_w = \sum_{e \in E(\Ga)} w(e) \cdot D(e)

9137: \end{equation}

9138: is the limit of effective theta divisors on smooth curves and the collection of these divisors

9139: \begin{equation}

9140: \{ \Theta_w\} \quad , w \in SNW_\Ga^1 = W^1(\Ga)

9141: \end{equation}

9142: gives the Mumford basis (2.13) which coincides with the Bohr-Sommerfeld basis (4.39) in the wave function space (see Definition 4) for

9143: ll-curve $P_\Ga$. 

9144: 

9145: Let us fix a spin network of level 1 $w_0 \in W^1(\Ga)$. Then one has

9146: \begin{enumerate}

9147: \item  the divisor $k \Theta_{w_0}$;

9148: \item the collection of divisors

9149: \begin{equation}

9150: \{ \l.c.m.(\Theta_w, k \Theta_{w_0})\} \quad , w \in SNW_\Ga^k = W^k(\Ga)

9151: \end{equation}

9152: \item this collection of effective divisors from the complete linear system $| k \Theta_{w_0}|$ gives the  Bohr-Sommerfeld basis (4.39) in the wave function space  for

9153: ll-curve $P_\Ga$ and the polarization of level $k$.

9154: \end{enumerate}

9155: By the restriction to $uS_g$ we can check that the extension of this basis to smooth curves due to the Hitchin connection coincides with the Bohr-Sommerfeld basis given by BPU-construction. 

9156: 

9157: 

9158: 

9159: 

9160: 

9161: \begin{thebibliography}{99}

9162: 

9163: \bibitem[A1]{A1}  Atiyah M. ``Reflections on Geometry and

9164: Physics", p. 423

9165: 

9166: \bibitem[A2]{A2} Atiyah M. "Complex analytic connections in

9167: fibre bundles", Trans.Amer. Math. Soc., 85, 1957, 181-207.

9168: 

9169: \bibitem[A3]{A3} Atiyah M. ``The path integral formulation'', Oxford

9170: seminar on Jones--Witten theory, Michaelmas Term 1988. Seminar 6,

9171: p.~106.

9172: 

9173: \bibitem[AB]{AB}  Atiyah M. and  Bott R. "The Yang-Mills equations over Riemann

9174:  surfaces.", Phil. Trans. R. Soc. Lond. A 308, 1982, 523-615.

9175: 

9176: \bibitem[An]{An} J. E. Andersen "Asymptotic faithfulness of quantum $SU(n)$

9177: representations of the mapping class groups", QA/0204084.

9178: 

9179: \bibitem[Ar]{Ar} I. V. Artamkin "Algebraic geometry of ll-curves." appears in

9180:  Izv. RAN,ser.Math.

9181: 

9182: \bibitem[B]{B} Baker H. "Abel's theorem and the allied theory includinng the

9183:  theory of theta functions.",

9184: Cambridge Univ. Press, 1897.

9185: 

9186: \bibitem[Be]{Be} Beauville A."Vector Bundles on Curves and

9187:  Generalized Theta Functions: Recent Results and Open Problems"

9188: Complex Algebraic Geometry, MSRI Publications, vol.28, 1995,

9189: 17-33.

9190: 

9191: 

9192: \bibitem[BGH]{BGH} E. D. Bolker, V. Guillemin, T. S. Holm,

9193: "How is a graph like a manifold?", CO/0206103.

9194: 

9195: 

9196: 

9197: \bibitem[BL]{BL} Beauville A. and Laszlo Y. , "Conformal blocks and

9198: Generalized Theta Functions", Commun.Math. Phys. 164, 1994,

9199: 385-419.

9200: 

9201: \bibitem[BPU]{BPU} D. Borthwick, T. Paul and A. Uribe, Legendrian

9202: distributions with applications to the non-vanishing of

9203: Poincar\'e series of large weight, Invent. math, {\bf122} (1995),

9204: 359--402 or hep-th/9406036

9205: 

9206: 

9207: 

9208: \bibitem[Ber]{Ber} Bertram A.," Moduli of rank 2 vector bundles, theta

9209: divisors and the geometry of curves in projective space", J.

9210: Diff. Geom., 35,1992, 429--469.

9211: 

9212: 

9213: 

9214:  \bibitem[BD]{BD}

9215:    Bradlow S. and Daskalopoulos G.

9216:  " Moduli of stable pair for holomorphic bundles over Riemann

9217:  surfaces", Int. J. Math., 2, 1991, 477--513.

9218: 

9219: 

9220: \bibitem[C]{C} Coble A. "Algebraic geometry and theta functions",

9221: AMS Coll.Publ. 10, Providence, R. I., 1929 (3 ed., 1969).

9222: 

9223: 

9224: \bibitem[CK]{CK} Cox D. and Katz S. "Mirror symmetry and Algebraic

9225: Geometry",AMS, Math.Surveys and Monographs, v. 68, 1999.

9226: 

9227: \bibitem[DSS]{DSS} Danilov V., Shokurov V. and Shafarevich I.

9228: "Algebraic curves, algebraic manifolds and schemes", Encyclopedia

9229: of math. scinces. v. 23, Springer, 1998.

9230: 

9231: 

9232: 

9233: \bibitem[D]{D} Delzant T. , "Hamiltonians periodiques et images

9234:  convexed de l'application moment", Bull.Soc. Math. France, 116,

9235:  1988,315-339.

9236: 

9237:  \bibitem[Do]{Do} S. Donaldson ``Gluing techniques in the cohomology of

9238: moduli spaces'', in Topological methods in modern mathematics

9239: (Stony Brook, NY, 1991), Publish or Perish, Houston, TX, 1993,

9240: pp. 137--170.

9241: 

9242: 

9243: \bibitem[DS]{DS} Drinfeld V., and Simpson C., "B-structures on G-bundles

9244: and local triviality", Math. Res.-Letters, 2, 1995, 823-829.

9245: 

9246: 

9247: \bibitem[DKN]{DKN} Dubrovin B., Krichever I. and Novikov S.

9248: " Integrable systems I", Encycl. Math. Sci., v. 4, Dynamical

9249: systems IV, Springer-Berlin, 1990, 173-280.

9250: 

9251: 

9252: 

9253: 

9254: 

9255: \bibitem[F1]{F1} Fay J. "Theta functions and Riemann surfaces",

9256: LNM, v 352, Springer-verlag, 1973.

9257: 

9258: \bibitem[F2]{F2} Fay J. "Kernel Functions, Analytic Torsion and

9259:  Moduli Spaces", AMS, Memoirs, v.98, 464, 1992.

9260: 

9261:  \bibitem[F3]{F3} Fay J. "The Non-Abelian Szego Kernel and

9262:   Theta-Diviser", Contemporary Mathematics, v. 136, 1992, 171-183.

9263: 

9264: \bibitem[FS]{FS} Friedan D. and  Shenker S. "The analytic

9265:  geometry of two-dimensional conformal field thory", Nuccl. Phys.

9266:  B 281, 1987, 509.

9267: 

9268: 

9269: \bibitem[FMN]{FMN} C. Florentino, J. Mourao, J. Nunes,

9270: ``Analytical aspects of the theory abelian theta functions.''

9271: Preprint

9272: 3/2001, February 2001, IST, Lisboa.

9273: 

9274: 

9275: \bibitem[FMNT]{FMNT} C. Florentino, J. Mourao, J. Nunes and A. Tyurin

9276: ``Analytical aspects of the theory non-Abelian theta functions.''

9277: Preprint IST 2000.

9278: 

9279: \bibitem[GH]{GH} Griffits Ph, Harris J. "Principles of Algebraic Geometry",

9280: Wiley-Intersience, New York, 1978

9281: 

9282: 

9283: 

9284: 

9285: \bibitem[GT]{GT} A. Gorodentsev and A. Tyurin, "ALAG", Izv. Ross.

9286: Akad. Nauk Ser. Mat. {\bf65}, n. 3, (2001), 15--50; English

9287: transl. in Russian Acad. Sci. Izv. Math.{\bf65}:3 (2001), avalable

9288: as  Max-Planck-Institute fur Mathematik, Preprint Series 1999

9289: (130). 1-34.

9290: 

9291: 

9292: \bibitem[Gu1]{Gu1} Guillemin V.,"Moment maps and combinatorial invariants

9293:  of Hamiltonian $T^n$-spaces",Birkhauser (Progress in Mathematics 122), 1994.

9294: 

9295: \bibitem[Gu2]{Gu2} V. Guillemin, Symplectic spinors and partial

9296: differential equations, Coll. Inter. C.N.R.S., Aix-en-Provence,

9297: 1974) (1975), 217--252

9298: 

9299: 

9300: 

9301: 

9302: 

9303: \bibitem[GS1]{GS1} Guillemin V. and Sternberg S. " Geometric asymptotics",

9304: Mathematical surveys, Vol.14, Providence, RI: AMS, 1977.

9305: 

9306: 

9307: \bibitem[GS2]{GS2} Guillemin V. and Sternberg S. "The Gelfand-Cetlin System and

9308: Quantization of the Complex Flag Manifolds" Journal of func.

9309: analysis, 52, 1983, 106-128. Compos. Math.,49, 173-194.

9310: 

9311: \bibitem[G1]{G1} W. Goldman, The symplectic nature of fundamental groups of

9312: surfaces, Adv. in Math. {\bf54} (1984) 200--225

9313: 

9314: \bibitem[G2]{G2} Goldman W. "Invariant functions on

9315:  Lie groups and Hamiltonian flows of surface group

9316:   representations", Invent. Math., 85, 1986, 263-302.

9317: 

9318: 

9319: \bibitem[Ha]{Ha} B.C. Hall, ``The Segal-Bargmann coherent state

9320: transform for compact Lie groups'', J. Funct. Anal. {\bf 122}

9321: (1994), 103-151.

9322: 

9323: 

9324: \bibitem[HT]{HT} Hatcher A. and Thurston W. "A presentation for the

9325:  mapping class group of closed oriented surface", Topology,

9326:  19 (3),1980,221-237.

9327: 

9328: \bibitem[H1]{H1} Hitchin N. "Flat connections and Geometrical Quantization",

9329: Commun.Math.Phys. 131, 1990,347-380.

9330: 

9331: \bibitem[H2]{H2} Hitchin N. "Stable bundles and integral systems." Duke

9332: Math. Jour. ,v.54, n.1, !987, 91-114.

9333: 

9334: \bibitem[HS]{HS} N. Hitchin and J. Sawon, ``Curvature and characteristic

9335: numbers of hyper\-K\"ahler manifolds'', math.DG/9908114, 17 pp.

9336: 

9337: 

9338: 

9339: \bibitem[JW1]{JW1} L. C. Jeffrey and J. Weitsman, Bohr--Sommerfeld orbits

9340: in the moduli space of flat connections and the Verlinde

9341: dimension formula, Commun. Math. Phys. {\bf150} (1992), 593--630

9342: 

9343: 

9344:  \bibitem[JW2]{JW2} L. C. Jeffrey and J. Weitsman, Half density

9345: quantization of the moduli space of flat connections and Witten's

9346: semiclassical invariants, Topo\-logy {\bf32} (1993), 509--529

9347: 

9348: 

9349: \bibitem[K1]{K1} Kohno T., "Topological invariants for 3-manifolds using

9350: representation of mapping class group I", Topology 31, 1992,

9351: 203-230.

9352: 

9353: \bibitem[K2]{K2} Kohno T., "Topological invariants for 3-manifolds using

9354: representation of mapping class group II. Estimation tunnel

9355: number of knots",

9356:  Commentary mathematics 175, 1994, 193-217.

9357: 

9358: \bibitem[KZ]{KZ} Kninik V. and Zamolodchikov A. "Current algebra

9359: and Wess-Zumino model in two dimensions", Nucl. Phys. B,247,

9360: 1984, 83.

9361: 

9362: \bibitem[KR]{KR} Kirillov A. and Reshetikhin N. "Representation of the algebtra

9363:  $U_q(sl(2, \C))$, q-orthogonal polynomials  and invariants of links,

9364:  Infinite dimensional Lie algebras and groups.", World Scientific, 1988,

9365:  285-342.

9366: 

9367: \bibitem[L]{L} Laszlo Yves, "Hitchin's and WZW connections are the same",

9368: J. Differential Geometry, 49, 1998, 547-576.

9369: 

9370: \bibitem[MS]{MS} Moore G., and Seiberg N., "Classical and quantum conformal

9371: field theory", Commun. Math. Phys. 123, 1989,177-254.

9372: 

9373: 

9374: \bibitem[Mum]{Mum} D. Mumford, Tata lectures on theta. I: Progr. Math

9375: {\bf28}, Birkh\"auser (1983). II. Jacobian theta functions and

9376: differential equations: Progr. Math {\bf43}, Birkh\"auser (1984).

9377: III. Progr. Math {\bf97}, Birkh\"auser (1991)

9378: 

9379: 

9380: \bibitem[M]{M} Margalit D. "The Automorphisms Group of the

9381: Complex of Pant Decompositions", GT/0201319.

9382: 

9383: 

9384: 

9385: \bibitem[MM]{MM} Masur H. and Minsky Y.

9386: "Geometry of the complex of curves I. Hyperbolicity", Invent.

9387: Math., 138 (1), 1999,103-149.

9388: 

9389: \bibitem[MP]{MP} M. Mulase, M. Penkava, "Ribbon graphs, quadratic

9390:  differentials on Riemann surfaces and algebraic curves defined over $\ov{\Q}$",

9391: math-ph/9811024.

9392: 

9393: \bibitem[Mu]{Mu}  M. Mulase, "Lectures on the asymptotic expansion of a Hermitian matrix integral", math-ph/9811023.

9394: 

9395: 

9396: \bibitem[NN]{NN} Newlender, A

9397:  and Nirenberg, L. Ann. of Math. 65 (1957), 391-404.

9398: 

9399: 

9400: \bibitem[Ne]{Ne} Newstead P.E. "Characteristic

9401:  classes of stable bundles over an algebraic curve ", Trans. Am.

9402:  Math. Soc., 169,1972, 337-345.

9403: 

9404: 

9405: \bibitem[NR]{NR} Narasimhan M. and Ramanan R. "Deformations of

9406: the moduli space of vector bundles over n algebraic curve", Ann.

9407: of Math., v.101, 1975, 391-417.

9408: 

9409: 

9410: 

9411: 

9412: \bibitem[P]{P} Penrose R. "An approach to combinatorial space-time"

9413: in "Quantum Theory and Beyond", ed. T. Bastin, Cambridge Univ.

9414: press, 1971, 151-180.

9415: 

9416: \bibitem[RS]{RS} Rovelli C. and Smolin L. "Spin Networks and

9417:  Quantum Gravity", gr-qu/9505006.

9418: 

9419: \bibitem[R]{R}

9420: ~Rawnslay J. Coherent States And K\"ahler Manifolds, Qart. J. of

9421: Maths. 28, 1997, pp.~404--415.

9422: 

9423: 

9424: 

9425:  \bibitem[RSW]{RSW} T. R. Ramadas, L. M. Singer and J. Weitsman, Some comments

9426: on Chern--Simons gauge theory, Commun. Math. Phys. {\bf126}

9427: (1989) 409--420

9428: 

9429: \bibitem[Re]{Re} Reid M. "What is a flip?", Preprint Uta, 1993.

9430: 

9431: \bibitem[Se]{Se} Seshadri C.S. "Desingurisation des varietes

9432: de modules de fibres semi-stables.", Asterisque, 98, 1982, 110-130.

9433: 

9434: 

9435: 

9436: \bibitem[S]{S} Sorger C., "La vormula de Verlinde", Sem. Bourbaki, 794, 1994.

9437: 

9438: 

9439: \bibitem[Sc]{Sc} Schlichenmaier M."Sugawara construction for higher

9440: genus Riemann surfaces", Mannheimer Manuskripte, 231, 1998,1-19,

9441: QA/9806032.

9442: 

9443: \bibitem[Si]{Si} Singer J. "Three dimensional manifolds  and their Heegaard

9444: diagrams", Trans. Am. Math. Soc., 35, 1933, 88-111.

9445: 

9446: 

9447:  \bibitem[Th1]{Th1} Thaddeus M. "Conformal field theory and

9448:  the cohomology of the moduli space

9449:  of stable bundles", J. Diff. Geom., 35, 1992, 131--149

9450: 

9451: 

9452: \bibitem[Th2]{Th2}  Thaddeus M.

9453:  "Stable pairs,linear system and the Verlinde formula",

9454:  Invent. Math., 117, 1994, 317--353.

9455: 

9456: 

9457: 

9458: \bibitem[TUY]{TUY} Tsuchiya A., Ueno K. and Yamada Y., "Conformal field theory on iniversal family of stable curves with gauge symmetries", Adv. Stud. Pure Math. 19, 1989, 459-566.

9459: 

9460: 

9461: \bibitem[TK]{TK} Tsuchiya A. and Kanie, "Vertex operators in

9462: conformal field on $\PP^{1}$ and monodromy representation of

9463: braid group", Advanced Studies in Pure Math. 16, 1988,297-372,

9464: (Errata) ibid. 19, 1990,675-682.

9465: 

9466: 

9467: 

9468: \bibitem[T1]{T1} Tyurin A. "The geometry of moduli of vector bundles,

9469:     Uspekhi Mat.Nauk  29:6 (1974), 59-88. Russian Math. Surveys

9470:     29:6 (1974), 57-88."

9471: 

9472: \bibitem[T2]{T2} Tyurin A, "Algebraic geometric aspects of

9473: smooth structures. I, Donaldson polynomials", Russian Math.

9474: Surveys, 44:3, 1989, 117-178

9475: 

9476: 

9477: \bibitem[T3]{T3} Tyurin A. "Six lectures on four manifolds",

9478: Lect.Notes in Math.,1646, 186-246.

9479: 

9480: \bibitem[T4]{T4} A. N. Tyurin ``Special Langrangian geometry and slightly

9481: deformed algebraic geometry (geometric quantization and mirror

9482: symmetry).'', Izv. RAN,ser.Math. 64:2, 141-224,

9483: \newline math.AG/9806006, 1-45.

9484: 

9485: 

9486:  \bibitem[T5]{T5} A. Tyurin, ``Three mathematical facets of $\SU(2)$-spin

9487: networks'',\newline math.DG/0011035, 1-20.

9488: 

9489: 

9490:  \bibitem[T6]{T6} A. Tyurin, "On Bohr-Sommerfeld bases, Izv. Ross.

9491: Akad. Nauk Ser. Mat. {\bf64} (2000), 163--196; English transl. in

9492: Russian Acad. Sci. Izv. Math.{\bf64}:5 (2000), avalable as

9493: math.AG/9909084;

9494: 

9495: \bibitem[T7]{T7} Andrei Tyurin, Complexification of

9496:  Bohr-Sommerfeld conditions, Preprint series, Inst. of Math of

9497:  university of Oslo, {\bf 15}, 1999, 1-32.

9498: 

9499: \bibitem[T8]{T8} Andrei Tyurin, Delzant model of moduli spaces, 

9500: AG/0105216, appear in Russian Acad. Sci. Izv. Math.{\bf65}:3 (2002)

9501: 

9502: \bibitem[W]{W} Welters G. Polarized abelian varieties and the heat euqation.

9503: Compos. Math.,49, 173-194.

9504: 

9505: \bibitem[Wa]{Wa} Wajnryb B. "A simple presentation for

9506:  the mapping class group", Israel J. Math. 45. 1983, 157-174.

9507: 

9508: 

9509: 

9510: \bibitem[We]{We} Weil A. "Generalization des functions abeliennes"

9511: J. de Math. P. et Appl.,17,no.9,1938, 47-87.

9512: 

9513: \bibitem[Wi]{Wi} Wirtinger W. "Untersuchengen uber Thetafunctionen"

9514: Teubner, Leipzig, 1895

9515: 

9516:  \end{thebibliography}

9517:  \end{document}

9518: