1: \documentclass{amsart}

2: 

3: 

4: \parskip 0.2cm

5: \usepackage{amssymb}

6: \usepackage{amscd}

7: \usepackage{amsmath}

8: \usepackage{amsfonts}

9: \usepackage{a4}

10: \usepackage{amsthm}

11: \usepackage{color}

12: 

13: %\usepackage[hypertex]{hyperref}

14: %\usepackage[active]{srcltx}

15: 

16: 

17: %\usepackage[inner]{showlabels}

18: %\usepackage[notref, notcite]{showkeys} beaucoup mieux que le pr{\'e}c{\'e}dent

19: 

20: 

21: \theoremstyle{plain}

22: \newtheorem{Hyp}{H}

23: \newtheorem{theo}{Theorem}[section] 

24: \newtheorem{prop}[theo]{Proposition}

25: \newtheorem{lemme}[theo]{Lemme}

26: \newtheorem{cor}[theo]{Corollaire}

27: \newtheorem{defin}[theo]{Definition}

28: 

29: 

30: \theoremstyle{definition}

31: 

32: \newtheorem{rem}[theo]{Remarque}

33: \newtheorem{exe}[theo]{Exemple}

34: \newtheorem{exo}[theo]{Exercice}

35: \newtheorem{commentaire}[theo]{Commentaire} 

36: 

37: \newcommand{\eval}[1]{\lvert_{#1}}

38: 

39: \newcommand{\Hi}{{\mathcal{H}}}

40: \newcommand{\Ci}{C^{\infty}} 

41: 

42: 

43: \newcommand{\R}{{\mathbb{R}}}

44: \newcommand{\N}{{\mathbb{N}}}

45: \newcommand{\C}{{\mathbb{C}}}

46: \newcommand{\Z}{{\mathbb{Z}}}

47: 

48: \newcommand{\de}{{\delta}}

49: \newcommand{\be}{{\beta}}

50: \newcommand{\al}{{\alpha}}

51: \newcommand{\la}{{\lambda}}

52: \newcommand{\De}{{\Delta}}

53: \newcommand{\si}{{\sigma}}

54: \newcommand{\hb}{{\hbar}}

55: \newcommand{\ka}{{\kappa}}

56: \newcommand{\ga}{{\gamma}}

57: \newcommand{\te}{{\theta}}

58: \newcommand{\om}{{\omega}}

59: \newcommand{\Om}{\Omega}

60: \newcommand{\ph}{{\varphi}}

61: \newcommand{\Ph}{{\Phi}}

62: \newcommand{\Ga}{{\Gamma}}

63: \newcommand{\eps}{{\epsilon}}

64: \newcommand{\La}{{\Lambda}}

65: \newcommand{\Si}{{\Sigma}}

66: 

67: \newcommand{\operateur}{{\mathcal{Q}}}

68: \newcommand{\identite}{{\operatorname{Id}}}

69: \newcommand{\Trace}{{\operatorname{Tr}}}

70: \newcommand{\trace}{{\operatorname{tr}}}

71: \newcommand{\volume}{{\operatorname{Vol}}}

72: \newcommand{\squar}{{\operatorname{sq}}}

73: \newcommand{\argument}{{\operatorname{Arg }}}

74: \newcommand{\diagonale}{{\operatorname{Diag}}}

75: 

76: \newcommand{\Hilbert}{{\mathcal{H}}}

77: \newcommand{\Top}{\text{top}}

78: \newcommand{\Toeplitz}{{\mathcal{T}}}

79: \newcommand{\Lie}{{\mathcal{L}}}

80: \newcommand{\equivalence}{{\mathcal{E}}}

81: \newcommand{\lagrangien}{{\mathcal{L}}}

82: \newcommand{\Hessien}{{\operatorname{Hess}}}

83: 

84: \newcommand{\contravariant}{{\operatorname{cont}}}  

85: \newcommand{\normalised}{{\operatorname{norm}}} 

86: 

87: \newcommand{\End}{{\operatorname{End}}}

88: 

89: \newcommand{\Lag}{{\operatorname{Lag}}}

90: 

91: \newcommand{\Mas}{{\mathcal{M}}}

92: 

93: \newcommand{\ideal}{{\mathcal{I}}}

94: \newcommand{\Sym}{\operatorname{Sym}}

95: \newcommand{\no}{{\mathcal{N}}}

96: 

97: %\newcommand{\scratch}[1]{\textsl{[\textcolor{blue}{#1}]}}

98: \newcommand{\scratch}[1]{}

99: 

100: 

101: 

102: \author{L. Charles} 

103: 

104: \address{Universit{\'e} Pierre et Marie Curie-Paris6, UMR 7586 Institut de

105:   Math{\'e}matiques de Jussieu, Paris, F-75005 France.}

106: 

107: \email{charles@math.jussieu.fr}

108: 

109: \keywords{Geometric Quantization, Toeplitz operators, Bohr-Sommerfeld conditions, Half-form bundle}

110: 

111: \subjclass{53D12, 53D50, 53D55, 81S30, 47L80, 35P20}

112: 

113: \title{Symbolic calculus for Toeplitz operators with half-forms} 

114: 

115: \begin{document}

116: 

117: \begin{abstract}

118: This paper is devoted to the use of half-form bundles in the symbolic

119: calculus of Berezin-Toeplitz operators on K{\"a}hler manifolds. We state

120: the Bohr-Sommerfeld conditions and relate them to the

121: functional calculus of Toeplitz operators, a trace formula and the

122: characteristic classes in deformation quantization. We also develop

123: the symbolic calculus of Lagrangian sections, with the crucial

124: estimates of the subprincipal terms. 

125: \end{abstract}

126: 

127: \maketitle

128: 

129: 

130: \bibliographystyle{plain}

131: 

132: In semi-classical analysis we usually consider (pseudo) differential

133: operators depending on a small parameter and acting on a $L^2$ space,

134: the underlying classical limit being a cotangent space with its

135: canonical symplectic structure. In this paper we are interested in a

136: similar theory where the classical phase space is a compact K{\"a}hler

137: manifold endowed with a prequantum line bundle $L$. Here the quantum

138: Hilbert space consists of the holomorphic sections of $L^k$. The

139: small parameter is the inverse of the power $k$. The operators of

140: interest are the

141: Berezin-Toeplitz operators. This setting was mainly introduced by

142: Kostant \cite{Ko},

143: Souriau \cite{So} and Berezin \cite{Be} and the suitable microlocal techniques were

144: developed by Boutet de Monvel and Guillemin \cite{BoGu}. Since then many standard

145: results for pseudo-differential operators have been adapted to this

146: context, like for instance the Schnirelman theorem \cite{Ze}, the Gutzwiller trace formula

147: \cite{BP2} or 

148: the Bohr-Sommerfeld conditions \cite{oim2}. The statements of these results are

149: easily predictable as far as only the symplectic structure of the

150: phase space is concerned, because they are the same

151: for the cotangent and K{\"a}hler spaces. But the semi-classical results

152: for the pseudo-differential operators may involve also some invariants,

153: like the subprincipal symbol or the Maslov index, which

154: do not only depend on the symplectic structure and consequently 

155: are difficult to identify in the K{\"a}hler setting. Furthermore these

156: quantities generally appear as quantum corrections and are

157: difficult to compute. Nevertheless in the papers \cite{oim1} and \cite{oim2}, we

158: carried out successfully some techniques to handle this. To formulate our

159: result we used the Riemannian metric of the K{\"a}hler structure instead

160: of the vertical polarization of the cotangent bundle. Typically we

161: proved some Bohr-Sommerfeld conditions where the Maslov index is

162: replaced with a curvature integral. Actually we missed the right

163: formulation which uses the half-form bundles. 

164: The main purpose of this paper and the sequel \cite{new} is to develop this point

165: of view. In this part we focus on the Bohr-Sommerfeld conditions

166: whereas \cite{new} is devoted to the dependence of the quantization on the complex structure.

167: 

168: Concretely we alter the usual setting by defining the quantum space  as the

169: space of holomorphic sections of $L^k \otimes L_1 \otimes \delta

170: \rightarrow M$. Here

171: $L$ is the prequantum bundle as previously, $L_1$ is an auxiliary

172: line bundle and $\delta$ is a half-form bundle, {\em i.e.} a square

173: root of the canonical bundle of $M$. A priori artificial, this

174: decomposition enlightens the semi-classical results, even in the usual

175: case where $L_1 \otimes

176: \delta$ is the trivial bundle. Roughly speaking the contribution of

177: $L_1$ in the semi-classical limit is the same as the one

178: of $L^k$ (one can view $L^k \otimes L_1$ as a first order

179: deformation of $L^k$) whereas the half-form bundle contributes in a

180: specific way. This principle will be confirmed in all our

181: results.  Another important point is that there

182: is a topological obstruction to the existence of half-form bundles. To

183: avoid this problem we consider globally the bundle $L^k \otimes K$ and

184: write locally $K = L_1 \otimes \delta$. 

185: The situation here is analogous to that in Riemannian geometry where we

186: think any Clifford module, at least locally, as the spinor bundle

187: twisted with an auxiliary bundle.

188: 

189: The first section is devoted to basic properties of Toeplitz

190: operators and their symbolic calculus. In particular an important

191: subprincipal symbol is defined. We state the Bohr-Sommerfeld

192: conditions in section 2 and relate them to the symbolic calculus and

193: trace formula by adapting an argument of Colin de Verdi{\`e}re

194: \cite{Co}. Here the formulation with half-forms permits to check easily the

195: consistency of the results. The next sections contain the proof of the

196: Bohr-Sommerfeld conditions. In section 4, we introduce the Lagrangian

197: sections, which are similar to the Lagrangian distributions, and

198: develop their symbolic calculus. Bohr-Sommerfeld conditions follows immediately. 

199: A comparison with the usual setting

200: is included, where a $\Z_4$-bundle plays a role analogous to the

201: Maslov bundle. In section 5, the technical part of the paper, we

202: provide the proof for the symbolic calculus of  the Lagrangian

203: sections. We follow essentially the method of \cite{oim2} but avoid the

204: complicated computations involving the derivatives of the K{\"a}hler

205: metric. These simplifications rely on a version of the stationary

206: phase lemma stated in an appendix of this paper. In view of this

207: proof, we think that our result  should generalize mutatis mutandis to the case where the

208: symplectic manifold doesn't admit any integrable complex structure. 

209: 

210: 

211: 

212: {\bf Acknowledgments} We thank Y. Colin de Verdi{\`e}re who provided us

213: his preprint \cite{Co} and suggested us to adapt his argument to the Toeplitz

214: operators. This was actually one of our original motivations to develop

215: the half-form formalism. We also thank F. Faure for his kind interest. 

216: 

217: 

218: 

219:  

220: 

221: 

222: 

223: 

224: 

225: \section{The setting}

226: \subsection{Square root of line bundle}

227: 

228: Let $M$ be a manifold and $F \rightarrow M$ be a complex line bundle.  A {\em square root} $(\delta,

229: \ph)$ of $F$ is a line bundle $\delta \rightarrow M$ together with an

230: isomorphism of line bundle $\ph : \delta^{\otimes 2} \rightarrow

231: F$. If $M$ is a complex manifold, a square root of its canonical bundle

232: $\La ^{n,0} T^*M$ is called a {\em half-form bundle}. 

233: Let us state basic properties of square roots. 

234: 

235: If $F$ has a Hermitian structure and $(\delta,

236: \ph)$ is a square root of $F$, then $\delta$ has a unique Hermitian structure

237: such that $\ph$ is a isomorphism of Hermitian line bundle. In the same

238: way, if $F$ is holomorphic or flat, $\delta$ inherits the same

239: structure. If $D^F$ is a first order differential operator acting on

240: sections of $F \rightarrow M$, then there exists a unique first

241: order differential operator $D^\delta$ acting on section of $\delta$ such that 

242: $$ D^F \ph (s \otimes s ) = 2 \ph ( s \otimes D^\delta s), \quad \forall \; s \in \Ci

243: (M,\delta).$$

244: A line bundle admits a square root if and only if its Chern class is divisible by 2 in $H^2(M, \Z)$. 

245: Two square

246: roots $(\delta, \ph)$ and $(\delta' , \ph')$ of $F$ are {\em equivalent}

247: if there exists an isomorphism $\Psi : \delta \rightarrow

248: \delta'$ such that $\ph' \circ \Psi^2 = \ph$. 

249: 

250: \begin{prop} Assume that $F$ admits a square root. Then the set

251:   of

252:   equivalence classes of square roots of $F$ is a principal

253:   homogeneous space for the first group of cohomology of $M$ with coefficient

254:   in $\Z_2$. 

255: \end{prop}

256: 

257: \begin{proof} 

258: First, if $(\delta, \ph)$ is a square root of the trivial line bundle

259: $1_M = M \times \C$, then $\delta$ inherits a flat structure from $1_M$

260: with structure group $\Z_2$. Furthermore this flat structure

261: determines $\ph$. It is easily proved that this induces an

262: isomorphism between the set of equivalence classes of square roots of

263: $1_M$ and the set of equivalences of flat line bundles with structure

264: group $\Z_2$. The latter is isomorphic to $H^1(M, \Z_2)$. Now, observe

265: that the tensor product of a square root of $L$ with a square root of

266: $1_M$ is a square root of $L$. This defines an action of $H^1(M,

267: \Z_2)$ on  $\equivalence _L$, which is easily shown to be free and transitive.

268: \end{proof}

269: 

270: 

271: \subsection{Quantum spaces} 

272: Let $M$ be a connected compact K{\"a}hler manifold of complex dimension

273: $n$. Denote by $\om \in \Om ^2 (M, \R)$ the fundamental form of $M$. Assume $M$ is endowed with a prequantization bundle $$L

274: \rightarrow M,$$ that is a

275: Hermitian line bundle with a connection $\nabla^L$ of curvature $\frac{1}{i}

276: \om$. Since $\om$ is a $(1,1)$-form, $L$ has a natural holomorphic

277: structure defined in such a way that the (local) holomorphic sections

278: satisfy the Cauchy-Riemann equations: $\nabla _{\bar{Z}} s

279: = 0$ for every holomorphic vector field $Z$ of $M$.

280: 

281: Let $K \rightarrow M$ be a Hermitian holomorphic line bundle. For

282: every positive integer $k$ define the quantum space

283: $\Hilbert_k$ : 

284: $$ \Hilbert_k = \bigl\{ \text{holomorphic section of } L^k \otimes K

285: \bigr\}.$$  

286: Assume that $M$ carries a half-form bundle $(\delta, \varphi)$.

287: $\delta \rightarrow M$ inherits a Hermitian scalar product and a holomorphic structure from

288: $\La^{ n, 0 }T^* M$. Introduce the Hermitian holomorphic line bundle $L_1$

289: such that $$K = L_1 \otimes \delta$$

290: and let $\frac{1}{i} \om_1 $ be the curvature of the Chern connection of

291: $L_1$.

292: 

293: Since $M$ is

294: compact, $\Hilbert_k$ is finite dimensional and it follows

295: from  the Riemann-Roch-Hirzebruch theorem and

296: Kodaira vanishing theorem that  

297: \begin{gather} \label{Riemann-Roch} 

298:  \operatorname{dim} \Hilbert_k = \Bigl( \frac{k}{2 \pi} \Bigr)^{n}

299: \int_M ( \om + k^{-1} \om_1 )^{\wedge n} /n!  + O(k^{n-2}) \end{gather} 

300: To interpret this formula, we consider $L^k \otimes L_1$ and $\om +

301: \hbar \om_1$ as

302: deformations of $L^k$ and $\om$ which give the first quantum

303: corrections in the semi-classical limit. Indeed  the leading term 

304: $$\bigl( \tfrac{k}{2 \pi} \bigr)^{n} \int \om^{\wedge n} /n!$$  gives

305: the second-order correction when we replace $\om$ with $\om + k^{-1}

306: \om_1$. Furthermore in the case $M$ doesn't carry any half-form

307: bundle, equation \eqref{Riemann-Roch}  is still

308: valid if we define $\om_1$ by 

309: $$ \om_1 :=  \om_K - \om_c/2,$$ 

310: where $\frac{1}{i} \om_K$

311: and $\frac{1}{i} \om_c $ are the curvatures of the Chern connections of

312: $K$ and $\La^{n,0} T^*M$.  

313: 

314: 

315: 

316: \subsection{Toeplitz operators} 

317: Denote by $\Pi_k$ the orthogonal projector of $L^2(M,

318: L^k \otimes K)$ onto ${\mathcal{H}}_k$, where the scalar product of two sections of $L^k \otimes K$ is

319: defined from the Hermitian structures of $L$ and $K$ and the

320: Liouville form $\mu_M$.

321: 

322: A Toeplitz operator is any  sequence $(T_k: \Hilbert_k \rightarrow

323: \Hilbert_k )$ of operators of the form 

324: $$ T_k = \Pi_k f(.,k) + R_k , $$

325: where $f(.,k)$ is a sequence of $\Ci( M)$ with an asymptotic expansion

326: $f_0 + k^{-1} f_1 +...$ for the $\Ci$ topology and the norm of $R_k$

327: is $O(k^{-\infty})$. 

328: 

329: The set $\Toeplitz$  of Toeplitz operators is a semi-classical algebra associated

330: to $(M, \om)$ in the following sense. 

331: 

332: \begin{theo} \label{sc_algebra}

333: $\Toeplitz$ is closed under the

334: formation of product. So it is a star algebra, the identity is

335: $(\Pi_k)$. The symbol

336: map $$\si_{ \contravariant} : \Toeplitz \rightarrow \Ci(M)[[\hb]],$$ sending $T_k$ into the

337: formal series $f_0 + \hb f_1 + ...$ where the functions $f_i$ are the

338: coefficients of  the asymptotic expansion of the multiplicator

339: $f(.,k)$, is

340: well defined.  It is onto and its kernel is the ideal consisting of

341: $O(k^{-\infty})$ 

342: Toeplitz operators. More precisely for any integer $\ell$, 

343: $$ \| T_k \| = O(k^{-\ell} ) \text{ if and only if } \si_{\contravariant} (T_k)

344: = O( \hb^{\ell}). $$ 

345:  Furthermore, the induced product $*_{\contravariant}$ on $\Ci (M) [[\hbar]]$ is a

346:  star-product. 

347: \end{theo}

348: Following the terminology of Berezin in \cite{Be}, we call $\si_{

349:   \contravariant}$  the

350: contravariant symbol map. This result is essentially a consequence of the works of Boutet de Monvel and Guillemin

351: \cite{BoGu} (cf. also \cite{Gu}, \cite{BoMeSc} and \cite{oim1}). 

352: Let us recall that equivalence classes of star-products on $(M,\om)$

353: are parametrized by elements in

354: $$ \frac{1}{i \hb}  [\om] + H^2 (M, \C )[[\hb]] $$

355: called Fedosov characteristic classes. The following theorem was

356: proved by Karabegov and Schlichenmaier in \cite{ka} and \cite{KaSc},

357: in the case $K$ is the trivial line bundle. 

358: \begin{theo} \label{Fedosov}

359: The Fedosov class of the star-product $*_{\contravariant}$ is  $\frac{1}{i \hb} ( [\om] + \hb [\om_1])$.

360: \end{theo} 

361: Again it is interesting to note the appearance of $\om + \hb

362: \om_1$. We do not need this result but some related facts.

363: Let us define the {\em normalized symbol} of a Toeplitz operator by 

364: $$ \si_{\normalised} (T_k ) := (\identite + \tfrac{\hbar}{2} \Delta ) \si_{\contravariant} (T_k) $$

365: where $\Delta$ is the holomorphic Laplacian acting on $\Ci(M)$.

366: Actually we are only interested in the leading and second order terms

367: of $\si_{\normalised} (T_k)$ and modifying the definition of

368: $\si_{\normalised} (T_k)$ by a $O(\hb^2)$ term wouldn't change the statements

369: of our results.  To compare with our previous article \cite{oim2}, the Weyl symbol that

370: we introduced when $K$ is

371: the trivial line bundle is equal to the normalized symbol modulo $O(\hb^2)$.

372: 

373: The map $\si_{\normalised}: \Toeplitz \rightarrow \Ci(M)[[\hb]]$ satisfies

374: the same properties as $\si_{\contravariant}$ stated in theorem \ref{sc_algebra}.  Denote

375: by $*_{\normalised}$ the associated star product.

376: 

377: \begin{theo} \label{subprincipal_calculus} 

378: Let $f$ and $g$ belong to $\Ci(M)[[\hb]]$, then

379: $$ f *_{\normalised}  g = f.g + \tfrac{\hb}{2i} \langle \pi, df \wedge

380: dg \rangle + O(\hb^2) $$

381: and 

382: $$  i \hb^{-1} \bigl( f *_{\normalised}  g -  g *_{\normalised}  f

383: \bigr) =  \langle \pi + \hb \pi_1 , df \wedge

384: dg \rangle + O(\hb^2).$$

385: where $\pi$ is the Poisson bivector and $\pi_1$ is the bivector such that 

386: $\langle \pi_1, df \wedge dg

387: \rangle + \langle X_f \wedge X_g , \om_1 \rangle =0$

388: for every $f, g \in \Ci(M)$. 

389: \end{theo}

390: 

391: So $*_{\normalised}$ is a normalized star-product, in the sense that

392: the second order term in the first formula is antisymmetric, which explains

393: our terminology. 

394: Observe that $\pi + \hb \pi_1$ is the Poisson bivector associated to $\om

395: + \hb \om_1$ in the sense that 

396: $$ \langle \pi + \hb  \pi_1, df \wedge dg

397: \rangle = \langle (X_f + \hb X_f^1) \wedge (X_g + \hb X_g^1) , \om +

398: \hb \om_1

399: \rangle +O(\hb^2),$$ 

400: where $X_f + \hb X_f^1+O(\hb^2)$  is

401: the Hamiltonian vector field of $f$  with respect to $\om +

402: \hb \om_1$, that is 

403: $$df + \langle \om + \hb \om_1 , X_f + \hb X_f^1

404: \rangle = O(\hb^2),$$ and the same holds for $g$ and $X_g + \hb

405: X_g^1$. So it follows from theorem \ref{Fedosov} that there exists a star

406: product equivalent to $*$ satisfying the formulas of theorem

407: \ref{subprincipal_calculus}. This last result is more precise because the

408: equivalence is specified. 

409: We can prove it  using the methods of \cite{oim1} or 

410: \cite{KaSc}. But this leads to complicated computations. We will present in

411: \cite{new} a more conceptual proof. We stated this result because we

412: can deduce from it a part of the Bohr-Sommerfeld conditions 

413: (cf. sections \ref{trace}, \ref{func}, \ref{var}). 

414: 

415: \scratch{Last remark is that $\si_{\normalised}$ isn't the only symbol map which

416: defines a normalized star-product. Indeed consider a symbol

417: map $\si$ of the form

418: $$ \si (T_k) = (\identite + \hb P ) \si_{\normalised} (T_k)  +O(\hb^2)$$

419: where $P$ is a differential operator. 

420: Then the associated star-product $*$ is normalized if and only if $P$ acts

421: as a vector field $X$.}

422: 

423: 

424: 

425: \subsection{Relation with geometric Quantization}

426: 

427: Our definition of the normalized symbol agrees in some sense with the

428: usual procedure to quantize observables in geometric

429: quantization. Assume that the Hamiltonian flow of $f \in \Ci (M)$

430: preserves the complex structure of $M$. Assume also that $K = L_1

431: \otimes \delta$, where $(\delta, \varphi)$ is a half-form bundle. 

432: Then the following operator is well-defined  

433: \begin{gather}

434: \label{geometric_quantization}

435:  \operateur (f) := f + \frac{1}{ik} \bigl( \nabla_{X_f}^{L^k \otimes L_1}\otimes

436: \identite + \identite \otimes \Lie_{X_f} \bigr) : \Hilbert_k \rightarrow

437: \Hilbert_k. \end{gather}

438: Here $\Lie_{X}$ acts on sections of

439: $\delta$ by $ \varphi ( (\Lie _X s) \otimes s) = \frac{1}{2} \Lie _X 

440: \varphi ( s^{\otimes 2} )$, or

441: equivalently as a Lie derivative where the pull-back of sections of

442: $\delta$ by a

443: complex diffeomorphism $\zeta $ is defined in such a way that $\varphi

444: ((\zeta^* s)^{\otimes 2} )  =

445:   \zeta^* \varphi (s^{\otimes 2} )$.

446: 

447: The definition \eqref{geometric_quantization}

448: is natural for the following reason. Denote by $\Phi_t$ the

449: Hamiltonian flow of $X_f$. Let $\tilde{\Phi}_t$ be the lift

450: of $\Phi_t$ to $L^k \otimes L_1 \otimes \delta$

451: defined by the tensor product of the parallel transport along the trajectories of $X_f$ in $L^k

452: \otimes L_1$ and by the pull-back in $\delta$. Then the solution of the

453: Schr{\"o}dinger equation 

454: $$ \frac{1}{ik} \frac{d}{dt} \Psi(.,t) +  \operateur (f) \Psi (.,t) = 0$$

455: with initial condition $\Psi \in \Hilbert _k$ is given by 

456: $$\Psi (x,t) = e^{\frac{k}{i} \int_0^t f ( \Phi_{s-t} (x) ) \sl{ds}} \tilde{\Phi}_{t} \bigr( \Psi

457: (\Phi_{-t}(x)) \bigl) .$$

458: 

459: \scratch{La formule est v{\'e}rifi{\'e}e. Regarder dans le Woodhouse

460: s'il introduit qqch de ressemblant.} 

461: 

462: The important point for us is that $\operateur (f)$ is a Toeplitz

463: operators whose normalized symbol is $f$ modulo $O(\hb^2)$. We will prove a

464: more general result for every smooth function $f$, which simplifies

465: some further proofs. 

466: 

467: If $f$ is an arbitrary smooth function, formula \eqref{geometric_quantization} doesn't

468: make sense, because the Lie derivative with respect to $X_f$ doesn't

469: necessarily preserve $\Om^{n,0} (M)$. So we define for any vector field $X$ the

470: operator $D_X$, 

471: $$ D_X \al  = p (\Lie _X \al), \qquad  \al \in \Om^{n,0} (M) $$

472: where $p$ is the projection from $\La^{n} T^*M \otimes \C$ onto

473: $\La^{n,0} T^*M$ with kernel the sum

474: $$    \La^{n-1,1}T^* M \oplus \La^{n-2,2}T^* M \oplus ... \oplus

475: \La^{0,n} T^* M .$$ 

476: Next we let $D_X$ act on the sections of $\delta$, as the first-order

477: differential operator such that $ 2 \varphi( s \otimes D_X s) = D_X

478: \varphi( s^2)$ . 

479:  

480: \begin{theo} \label{normalised_symbol_quantisation}

481: For any $f\in \Ci(M)$, the operator 

482: $$  \operateur (f) := \Pi_k \Bigl( f + \frac{1}{ik} \bigl( \nabla_{X_f}^{L^k \otimes L_1}\otimes

483: \identite + \identite \otimes D_{X_f}  \bigr) \Bigr)  : \Hilbert_k \rightarrow

484: \Hilbert_k  $$

485: is a Toeplitz operator with principal symbol $f$ and vanishing subprincipal symbol. 

486: \end{theo}

487: 

488: This theorem is a consequence of the following lemma and an argument

489: of Tuynman \cite{Tu}. 

490: 

491: \begin{lemme} \label{l1}

492: Let $s$ be a half-form, then 

493: $$ D_{X_f} s = \nabla^{\delta}_{X_f} s + \tfrac{i}{2} (\Delta f) s $$

494: where $\nabla^\delta$ is the Chern connection of $\delta$ and $\Delta$

495: is the holomorphic Laplacian of $M$. 

496: \end{lemme}

497: 

498: It follows that 

499: \begin{gather} \label{defe}

500:   \nabla_{X_f}^{ L^k \otimes L_1}\otimes

501: \identite + \identite \otimes D_{X_f} = \nabla_{X_f}^{  L^k \otimes

502:   L_1 \otimes \delta } +  \tfrac{i}{2} (\Delta f) = \nabla_{X_f}^{  L^k \otimes

503:   K} +  \tfrac{i}{2} (\Delta f) \end{gather}

504:  Now we have for every $\Psi \in \Hilbert_k$, 

505: $$ \Pi_k ( \nabla_{X_f}^{  L^k \otimes

506:   K} \Psi)  = \tfrac{1}{i} \Pi_k ( \Delta f . \Psi ), $$

507: cf. \cite{Tu} or \cite{BoMeSc} for a proof. Hence

508: $$ \operateur (f) \Psi =  \Pi_k \bigl(

509: ( f - \tfrac{1}{2k} \Delta f ) \Psi \bigr)

510: $$

511: which proves theorem \ref{normalised_symbol_quantisation}.

512: 

513: The last expression in \eqref{defe} shows that the definition of

514: $\operateur (f)$ is

515: independent of the choice of the half-form bundle and generalizes in

516: the cases where no such bundle exists.

517: 

518: \begin{proof}[Proof of lemma \ref{l1}]

519: It suffices to prove that for every $\alpha \in \Om^{n,0} (M)$, we have 

520: $$  D_{X_f} \al = \nabla_{X_f} \al + i (\Delta f) \al $$

521: Introduce normal complex coordinates $z^1,...,z^n$ centered at

522: $x_0$. So if $\alpha = dz^1 \wedge ... \wedge dz^n$, then $\nabla \al =

523: 0$ at $x_0$. Let us write $\om = i G_{j,k} dz^j \wedge d

524: \bar{z}^k$. Then 

525: $$X_f = - i G^{j,k} (\partial_{z^j} f) \partial_{\bar{z}^k} + i

526: G^{j,k}(\partial_{\bar{z}^k} f) \partial_{z^j}

527: $$ 

528: Using that the first derivatives of $G^{j,k}$ vanish at $x_0$, we

529: obtain easily 

530: $$ D_{X_f} \al = i G^{j,k}  (\partial_{z^j}\partial_{\bar{z}^k} f) = i

531: \Delta f.$$

532: The result follows. 

533: \end{proof}

534: 

535: %\newpage

536: \section{Bohr-Sommerfeld Conditions} \label{BS_section}

537: 

538: \subsection{The result}

539: 

540: Assume that $M$ is $2$-dimensional. Let $(\delta,\varphi)$ be a

541: half-form bundle and let us write $K = L_1 \otimes \delta$ as

542: previously. Consider a self-adjoint Toeplitz

543: operator $(T_k)$. Its normalized symbol $$f_0 + \hb f_1 +...$$  is

544: real-valued. Bohr-Sommerfeld conditions give the spectrum of $T_k$ on

545: every open interval $I$ of regular values of $f_0$ in the

546: semi-classical limit. To simplify the statements, assume that $f_1$ vanishes. 

547: 

548: Let $\Ga^1$, ...,

549: $\Ga^m$ be the components of $f_0^{-1}(I)$. For every $i \in

550: \{ 1,...,m\}$, the map $$f_0 : \Ga^i \rightarrow I$$ is a trivial

551: fibration with fiber diffeomorphic to $

552: S^1$. For every $\la \in I$, fix an orientation on the fiber

553: $\Ga^{i}_\la = f_0^{-1} (\la) \cap \Ga^i$ depending continuously on

554: $\la$.

555: 

556: Let $a^i \in \Ci(I)$ be the {\em principal action}, defined in such a

557: way that the parallel transport in $L$ along $\Ga^i_\la$ is the

558: multiplication by $\exp \bigl( {i a^i(\la)}\bigr)$. Using $L_1$

559: instead of $L$, define in the same way the

560:   {\em subprincipal action} $a^i_1 \in \Ci(I)$.

561: 

562: Let us define an index $\epsilon^i$ from the half-form bundle

563: $(\delta, \varphi)$. Observe that the restriction $\delta_{i, \la}$ of

564: $\delta$ to $ \Ga^{i}_\la$ is a square root of $T^*

565: \Ga^{i}_\la \otimes \C$. Indeed, let us denote by $\iota$ the embedding $\Ga^{i}_\la

566: \rightarrow M$, then the map

567: $$ \varphi_{i,\la} :\delta_{i, \la}^2 \rightarrow T^*

568: \Ga^{i}_\la \otimes \C, \qquad u \rightarrow \iota^* \varphi (u)$$

569: is an isomorphism of line bundle. The set 

570: $$ \{ u \in \delta_{i,\la}; \; \varphi_{i,\la} (u ^{\otimes 2}) >0 \} $$

571: has one or two connected components. In the first case,

572: we set $\epsilon^{i}_\la =1$ and in the second case $\epsilon^{i}_\la

573: =0$. Observe that $\epsilon^{i}_\la$ doesn't depend on $\la$.

574: 

575: The Bohr-Sommerfeld conditions are   

576: \begin{gather} \label{BS_conditions}

577:   a^i(\la) + k^{-1}( a^i_1 (\la) + \epsilon^i \pi   ) \in \frac{2 \pi}{k} \Z \end{gather} 

578: Denote by $\Si^i(k)$ the set of $\la \in I$ satisfying

579: \eqref{BS_conditions}. When $k$ is sufficiently large, $\Si^i(k)$ is a

580: finite set containing 

581: $$\frac{k}{2\pi} \volume (\Ga^i ) + O( k^{-1})$$ points. 

582: Let $\Si(k)$ be the union of the $\Si^i(k)$. Define the

583: {\em multiplicity} of $\la \in \Si(k)$ as the number of $\Si^i(k)$ which contains

584: $\la$. The points of $\Si (k)$ approximate the eigenvalues of $T_k$

585: in the following sense.

586: 

587: \begin{theo} \label{BS_theo}

588:  Let $\la_-(k)$ and $\la_+(k)$ be two sequences of $I$ such that 

589: \begin{gather} \label{E1}

590: d ( \la_-(k) , \Si (k)) \geqslant  C k^{-1}, \quad d ( \la_+(k) ,

591: \Si (k))

592: \geqslant C k^{-1} \end{gather} 

593: for some positive $C$. Assume furthermore that there exists $\la_-,

594: \la_+ \in I$ such that 

595: \begin{gather} \label{E2}

596: \la_- \leqslant

597: \la_-(k) \leqslant \la_+(k) \leqslant \la_+ .\end{gather} 

598: Denote by $\la_1(k) \leqslant \la_2(k) \leqslant ....\leqslant \la_{N(k)}(k)$

599:  (resp. $\la'_1(k) \leqslant \la'_2(k) \leqslant ....\leqslant \la'_{N'(k)}(k)$)

600:  the eigenvalues of $(T_k)$ (resp. points of $\Si (k)$) contained in

601:  $(\la_-(k), \la_+(k))$ and counted with multiplicities. Then, when $k$ is

602:  sufficiently large, $N(k) = N'(k)$. Furthermore  

603: \begin{gather} \label{Approximation} 

604:  \la_j(k) =  \la_j' (k) + O(k^{-2}) \end{gather}

605: uniformly with respect to $j$.  

606: \end{theo} 

607: 

608: The interest of condition \eqref{E1} is to avoid any ambiguity in the counting of eigenvalues near the

609: endpoints of $(\la_- (k), \la_+ (k))$. It is not restrictive. Indeed  if

610: $\la_-(k)$, $\la_+ (k)$ are arbitrary sequences satisfying \eqref{E2},

611: then by modifying them by suitably chosen $O(k^{-1})$ sequences, we

612: obtain sequences satisfying both estimates \eqref{E1} and \eqref{E2}. 

613: 

614: Since the definition of the Toeplitz operators and of their normalized

615: symbol only depend on $K$ and not on the choice of the half form bundle, it is likely that the same holds for the Bohr-Sommerfeld

616: conditions. This is easily checked using that any other half-form bundle is of the form $\delta' =\delta \otimes

617: F$, where $F$ is a flat Hermitian line bundle with holonomy in $\Z_2$. So

618: $L_1' = L_1 \otimes F^{-1}$, and straightforward computations show

619: that the functions $a^i_1 +

620: \epsilon^i$ do not depend on the choice of $\delta$. 

621: 

622: To compare with our previous results in \cite{oim2}, when $K$ is

623: the trivial bundle, we defined the function $a^i_1 +

624: \epsilon^i$ as the integral of the geodesic curvature of

625: $\Ga_{\la}^i$. 

626: 

627: We can also approximate the eigenvalues up to a $O(k^{-\infty})$ error. More precisely,  there exist

628: sequences $(S^i(.,k))_k$ of $\Ci (I)$ such that the Bohr-Sommerfeld

629: conditions $$S^i(.,k) \in \tfrac{2 \pi}{ k} \Z$$ 

630: instead of \eqref{BS_conditions} lead to the same

631: result with a $O(k^{-\infty})$ error in \eqref{Approximation}. Furthermore, the sequences $S^i(.,k)$ admit asymptotic

632: expansions of the form 

633: $$S_0^i + k^{-1} S^i_1 + k^{-2} S^i_2 +...$$ 

634: The leading and second order terms are the same like in \eqref{BS_conditions}. Applying an argument of Colin de

635: Verdi{\`e}re \cite{Co}, we can prove that the derivatives of the $S^i_j$ only depend on the star-product $*_\normalised$ and the

636: normalized symbol of $T_k$.  In particular, assuming that $S_0^i$ is

637: increasing, we will deduce from

638: proposition \ref{subprincipal_calculus} that necessarily

639: \begin{gather} \label{Colin}

640: S^i_0 (\la) -   S^i_0 (\la') = \int_{D^i} \om \quad \text{and} \quad S^i_1 (\la ) -  S^i_1 (\la') = \int_{D^i} \om_1 \end{gather}

641: if $\la , \la' \in I$ are such that $\la' \leqslant \la$ and $D^i =

642: \Ga^i \cap f_0^{-1} ((\la , \la'))$.

643: This determines the Bohr-Sommerfeld conditions \eqref{BS_conditions} modulo a

644: constant term. 

645: Note that only

646: the derivatives of $S^i_0$ and $S^i_1$ are determined by the

647: star-product. Indeed, if we twist $L$ or $L_1$ by a flat Hermitian

648: line bundle, the actions of the non-contractible loops may change although the star-product remains the same. 

649: 

650: As a first step to deduce \eqref{Colin} from proposition

651: \ref{subprincipal_calculus}, let us state some results on traces

652: and functional calculus of Toeplitz operators.

653: 

654: \subsection{Traces} \label{trace}

655: 

656: In this section and the next one, we do not necessarily assume that $M$

657:  is $2$-dimensional. 

658:  It is a known result that the trace of a Toeplitz operator $T_k$ with normalized symbol $f_0 +

659: \hb f_1 +...$ admits an asymptotic

660: expansion of the following form:

661: $$ 

662:  \Trace ( T_k) = \Bigl( \frac{k}{2 \pi} \Bigr)^n \int_M (f_0 +

663: k^{-1} f_1 +...) (1+ k^{-1} d_1 + k^{-2} d_2 +...) \mu_M

664: +O(k^{-\infty}) $$

665: where $d_1$, $d_2$, $d_3$, ... are functions of $\Ci(M)$ which do not depend of

666: $T_k$ (cf. for instance \cite{oim1}). 

667: 

668: These functions may be computed in terms of the K{\"a}hler metric

669:  by using the methods of \cite{oim1}, but it is more convenient to relate them to the star-product $*_\normalised$. 

670: To do this, observe that the $\C[[\hb]]$-linear map

671: \begin{gather} \label{Trace}  \trace: \Ci(M) [[\hb ]] \rightarrow \C[\hb^{-1}, \hb]], \quad f(\hb)

672: \rightarrow (2 \pi \hb )^{-n} \int_M f(\hb) d(\hb)\; \mu_M \end{gather}

673: where $d( \hb) =1 + \hb d_1 + \hb^2 d_2 +...$ is a trace for the

674: star-product $*_{\normalised}$, in the sense that it satisfies 

675: $$\trace (f

676: *_{\normalised} g) = \trace( g *_{\normalised} f).$$ 

677: Following Fedosov

678: \cite{Fe} or Nest-Tsygan \cite{NeTs}, such a trace is

679: unique up to multiplication by an element of $\C [\hb^{-1}, \hb]]$ and

680: there exists a canonical one determined by the

681: following normalization condition: for every local equivalence $\Phi$

682: between  $*_{\normalised}$ and the

683: Weyl star-product, we have 

684: \begin{gather}  \label{trace_normalised}

685: \trace (f) =  (2 \pi \hb)^{-n} \int_M \Ph (f)\; \mu_M, \quad \text{

686:   where } \mu_M = \om^n/n!. \end{gather} 

687: We claim that the trace defined in

688: \eqref{Trace} is the canonical trace. This follows from the fact that

689: the quantization by Toeplitz operators is microlocally equivalent to

690: the usual Weyl quantization.

691: 

692: So the functions $d_i$ are determined by $*_\normalised$. In

693: particular, it follows from

694: proposition \ref{subprincipal_calculus} that 

695: \begin{gather} \label{sub_trace} 

696:  \trace (f) =  (2 \pi \hb)^{-n} \int_M f \; ( \om + \hb \om_1 )^{\wedge

697:   n}/ n! + O( \hb ^{-n +2}) \end{gather}  

698: 

699: \begin{proof}[Proof of formula \eqref{sub_trace}] Consider an equivalence of star-product of the form

700:   $$\Phi = \identite + \hb X + O(\hb^2)$$ where $X$ is a vector field. It is easily

701:   checked that the star-product 

702: $$ f *' g =  \Phi \bigl( \Phi^{-1} (f) *_{\normalised} \Phi^{-1} (g) \bigr)$$

703: is normalized and satisfies

704: $$  i \hb^{-1} \bigl( f *'  g -  g *'  f

705: \bigr) =  \langle \pi + \hb( \pi_1+ X.\pi) , df \wedge

706: dg \rangle + O(\hb^2).$$

707: Locally on can choose $X$ in such a way that $ \pi_1 + X.\pi = 0$. Indeed this

708: equation is equivalent to 

709: $\om_1 + d\al = 0,$ where $\al$ is the

710: $1$-form such that $ \om(X,.) = \al$. 

711:  

712: Next step is to introduce local Darboux coordinates, which define a Weyl star-product $*_{\operatorname{Weyl}}$. And  modifying $\Phi$

713: by a $O(\hb^2)$ term, one has $*_{\operatorname{Weyl}} = *'$

714: (cf. \cite{BeCaGu}).  Then it follows from

715: \eqref{trace_normalised} that 

716: $$ \trace (f) = (2 \pi \hb)^{-n} \int_M f\; \bigl(\mu_M - \hb (X\mu_M)

717: +O(\hb^2)\bigr) .$$

718: By definition of $\pi_1$ in theorem

719: \ref{subprincipal_calculus}, one has  $\langle \pi_1 , \om \rangle +

720: \langle \pi, \om_1 \rangle = 0 $. Since $X. \langle \pi, \om \rangle

721: =0$, we obtain 

722: $$\langle \pi , X \om \rangle = -  \langle X  \pi, \om

723: \rangle = \langle \pi_1, \om \rangle = - \langle \om_1, \pi \rangle.$$

724: Then a straightforward computation leads to 

725: $$\langle \pi^ {\wedge n}, \mu_M - \hb X

726: \mu_M \rangle  = \langle \pi^{\wedge n}, ( \om + \hb \om_1 )^{\wedge

727:   n}/ n! \rangle + O(\hb^2),$$ 

728: which proves the result.

729: \end{proof}

730: 

731: As a consequence of \eqref{sub_trace}, we obtain the estimate

732: \eqref{Riemann-Roch} of the dimension of $\Hilbert_k$, since this

733: dimension is the trace of the projector $\Pi_k$, whose normalized symbol is

734: $1$.

735:  Actually the index theorems of deformation quantization proved

736: in \cite{Fe} or \cite{NeTs} yield the asymptotic expansion of $\Trace (

737: \Pi_k)$ modulo $O(k^{-\infty})$ in terms of the Fedosov class

738: of $*_\normalised$. Since this trace is an integer, the

739: $O(k^{-\infty})$ error vanishes when $k$ is sufficiently

740: large. In this way we can deduce Riemann-Roch-Hirzebruch theorem from

741: theorem \ref{Fedosov} .  

742: 

743: 

744: \subsection{Functional calculus} \label{func}

745: 

746: Let $T_k$ be a self-adjoint Toeplitz operator with normalized symbol

747: $$f =f_0 + \hb f_1 +....$$ and let $g$ be a function of $\Ci

748: (\R, \C)$. Then it is known that $g(T_k)$ is a Toeplitz

749: operator (cf. for instance \cite{oim1}). Furthermore, the normalized symbol of $g(T_k)$ is given by

750: the following non-commutative Taylor formula: 

751: \begin{gather} \label{NC_Taylor}

752:  g^{*_{\normalised}}(f) (x)= \sum \frac{1}{\ell!}g^{(\ell)} (f_0 (x)) \; ( f(y) - f_0 (x)

753: )^{*_{\normalised} \ell} \eval{y = x}, \end{gather} 

754: where $g^{(\ell)}$ is the $\ell$th derivative of $g$ and $h^{*_{\normalised}

755:   \ell} = h*_{\normalised}...*_{\normalised} h$ repeated $\ell$ times. 

756: In particular, an easy computation

757: from proposition \ref{subprincipal_calculus} leads to

758: \begin{gather} \label{sub_functional_calculus} 

759: g^{*_{\normalised}}(f)=  g(f_0) + \hb g'(f_0)f_1 +

760: O(\hb^2). \end{gather} 

761: \scratch{

762: Notice that formula \eqref{NC_Taylor} makes sense for an arbitrary

763: star product $*$. This defines a formal functional calculus

764: satisfying the expected properties. For instance, one can check that $g_1^* (f) * g_2^*(f) = (g_1

765: g_2)^{*}(f)$. }

766: 

767: \subsection{On the variation of $S_0$ and $S_1$} \label{var}

768:  

769: Let us deduce formulas \eqref{Colin} on the variations of $S_0$ and

770: $S_1$ from the trace formula \eqref{sub_trace} and the functional symbolic calculus. 

771: Assume that

772: $f_0^{-1}(I)$ is connected. 

773: 

774: If  $g \in \Ci_o (I, \C)$, then we deduce from

775: \eqref{sub_functional_calculus}, \eqref{sub_trace} and the fact that

776: $f_1 =0$ that 

777: \begin{xalignat}{1} \notag 

778: \Trace (g(T_k)) = & \frac{k}{2 \pi} \int_M (g \circ f_0)  ( \om + k^{-1}

779: \om_1 ) \; + O(  k^{-1})\\\label{eq1}

780: =& \frac{k}{2 \pi} \int_I g \; \bigl( f_{0*}  [ \om + k^{-1}

781: \om_1 ] \bigr) \; + O( k^{-1})

782:  .\end{xalignat}

783: where  $f_{0*}$ is the push-forward $\Om^2 (M) \rightarrow \Om^1 (I)$

784: defined by 

785: $$  \int_M (f_0^* h ) \ \al  = \int_I g. f_{0*} \al, \qquad \forall h

786: \in \Ci_o(I) .$$

787: 

788: 

789: On the other hand, assume the Bohr-Sommerfeld condition is $$ S(\la,k) \in 2 \pi k^{-1} \Z $$ with

790: $S(\la,k) =S_0 + k^{-1}S_1 + O(k^{-2})$. If the derivative of $S_0$

791: doesn't vanish, then one can invert the functions $S(.,k)$ when $k$ is

792: sufficiently large and

793: $$ \Trace (g(T_k)) = \sum_{x \in 2\pi k^{-1} \Z} g ( S^{-1}(x,k))

794: + O(k^{-1}) $$

795: Interpreting this as a Riemann sum, it follows that

796: \begin{xalignat}{1} \notag

797:  \Trace (g(T_k)) =& \frac{k}{2 \pi} \int_\R g (S^{-1}(x,k )) dx

798:  +O(k^{-1}) \\ \label{eq2}

799:  = &  \frac{k}{2 \pi} \int_I g (\la) S' (\la, k ) d\la

800:  +O(k^{-1})

801: \end{xalignat}

802: with the same orientation of $I$ as before if $S'_0$ is positive.  

803: Since \eqref{eq1} and \eqref{eq2} hold for any function $g$ of $\Ci_o( I,

804: \C)$, we have 

805: $$ d S_0   + \hb d S_1  =  f_{0*}  ( \om + \hb \om_1 )$$ 

806: and \eqref{Colin} follows. 

807: 

808: 

809: 

810: 

811: 

812: 

813: 

814: 

815: %\newpage

816: \section{Lagrangian states}

817: 

818: 

819: \subsection{Definitions and symbolic calculus} \label{sec:resultat}

820: 

821: First we recall the definition of a local Lagrangian section

822: associated to a closed Lagrangian embedding $\iota: \Ga \rightarrow M$. 

823: 

824: Let $U$ be an

825: open set of $M$ such that $U_\Gamma := \iota^{-1} (U )$ is contractible. Since the

826: curvature of $\iota^* L$ vanishes, there exists a 

827: flat unitary section $t_\Gamma$ of $ \iota^* L \rightarrow U_\Gamma$. 

828: Introduce a formal

829: series 

830: $$\sum_{\ell =0 }^{\infty} \hb^\ell g_\ell \in \Ci ( U_\Gamma, \iota^*K)[[\hb]].$$ 

831: Let $V$  be an open set of $M$ such that $\overline{V} \subset U$. 

832: Then a

833:  sequence $\Psi_k \in {\mathcal{H}}_k$ is a {\em Lagrangian

834:  section}

835:  over $V$ associated to $(\Gamma, t_\Gamma)$ with symbol $\sum \hb^\ell

836:  g_\ell$ if 

837: \begin{gather*} % \label{Ansatz_Lag}

838:  \Psi_k(x) =  \Bigl( \frac{k}{2\pi} \Bigr)^{\frac{n}{4}} F^k (x)

839:  \tilde{g} (x,k) + O(k^{-\infty}) \text{ over $V$, } \end{gather*} 

840: where 

841: \begin{itemize} 

842:  \item  $F$ is a section of $L \rightarrow U$ such that 

843: $$\iota^* F =

844:  t_\Gamma \quad \text{ and } \quad \bar{\partial} F \equiv 0$$ modulo a section which

845:  vanishes to every order along $\iota (\Gamma)$. Furthermore $|F(x)|< 1$ if $x \notin \iota(\Gamma)$.  

846:  \item

847: $\tilde{g}(.,k )$ is a sequence of $\Ci ( U,K)$

848: with an asymptotic expansion $\textstyle{\sum} k^{-\ell} \tilde{g}_{\ell}$

849: in the $\Ci$ topology such that 

850: $$\iota^* \tilde{g}_\ell = g_\ell \text{

851:  and } \quad \bar{\partial} \tilde{g}_\ell \equiv 0$$ modulo a section which vanishes at

852: every order along $\iota(\Gamma)$.

853: \end{itemize}

854: We assume furthermore that $\Psi_k$ is admissible in the sense that $\Psi_k(x)$ is uniformly

855: $O(k^{N})$ for some $N$ and the same holds for its successive

856: covariant derivatives. 

857: 

858: 

859: It is not obvious that such a sequence exists.

860: \begin{theo} \label{existence}

861: For every series $\sum \hb^\ell g_\ell$ of $\Ci ( U_\Gamma,

862:  \iota^* K)[[\hb]]$, there exists a Lagrangian

863:   section  over $V$ associated to $(\Gamma, t_\Gamma)$ with symbol $\sum \hb^\ell

864:  g_\ell$. It is unique modulo a section which is $O(k^{-\infty})$

865:   over $V$. \end{theo}

866: 

867: In the statement of the following theorems, we consider that $K = L_1

868: \otimes \delta $ over $U$, where $(\delta, \varphi)$ is a half-form bundle. Recall that

869: $\iota^* \delta = \delta_{\Ga}$ is a square root of $\La ^n T^*\Ga \otimes

870: \C$ through the isomorphism 

871: $$ \varphi_{\Ga} : \delta^2_ \Ga

872: \rightarrow \La ^n T^*\Ga \otimes \C, \quad u \rightarrow \iota^*

873: \varphi( u).$$ 

874: Let us associate

875: to the {\em principal symbol} $g_0$ of a Lagrangian section a density

876: $m(g_0)$ where $m$ is the map:

877: $$ m : \iota^* L_1 \otimes  \delta_{\Ga} \rightarrow | \La | (\Ga) , \quad

878: u \otimes v \rightarrow \| u \|^2_{L_1} \ | \varphi_{\Ga} ( v^{\otimes

879:   2})|$$

880: We then have the following estimate of the norm of $\Psi_k$. 

881: \begin{theo} \label{norme} 

882: Let  $\xi \in \Ci_o ( V)$, then we have  

883: $$ \int_M \xi \  \| \Psi_k \|^2 _{L^k \otimes L_1 \otimes \delta } \; \mu_M

884:  = \int_{\Gamma} (\iota^* \xi) \  m(g_0) +

885:  O(k^{-1}).$$

886: \end{theo}

887: Next results describe how a Toeplitz operator acts on a Lagrangian

888: section.

889: 

890: \begin{theo}  \label{Toep_Lag}

891: Let $T_k$ be a Toeplitz operator with principal symbol

892: $f_0$. Then $T_k \Psi_k$ is a Lagrangian section over $V$ associated

893: to $(\Gamma, t_\Gamma)$ with symbol $(\iota^* f_0) g_0 +O(\hb)$. 

894: \end{theo}

895: 

896: 

897: To prove the Bohr-Sommerfeld conditions, we need to compute

898: the subsequent coefficient of the symbol of $T_k \Psi_k$, in the case

899: where $f_0$ is constant over $\Ga$. 

900:  

901: \begin{theo} \label{resultat_clef} 

902: Let $T_k$ be a Toeplitz operator with normalized symbol $f_0 + \hb

903:  f_1 + O(\hb^2)$. Assume that $f_0$ is constant along $\Gamma$. Then

904: the symbol of $ T_k \Psi_k$ is 

905: $$ (\iota^* ( f_0 + \hb f_1)).( g_0  + \hb g_1)  + \hb  \tfrac{1}{i} (

906: \nabla_{X}^{\iota^* L_1}

907: \otimes \identite + \identite \otimes \Lie^{\delta_{\Ga}}_X ).g_0  + O(\hb^2) $$

908: where 

909: \begin{itemize} 

910: 

911: \item $X$ is the Hamiltonian vector field of $f_0$,

912: \item  $\nabla^{\iota^* L_1}$ is the pull-back of the Chern connection of $L_1$, 

913: \item $ \Lie^{\delta_{\Ga}}_X$ is the first order differential operator

914:   acting on sections of $\delta_{\Ga}$ 

915:   such that $$ \varphi_{\Ga} ( \Lie^{\delta_{\Ga}}_X g  \otimes g)  = \tfrac{1}{2}

916: \Lie_X \varphi_{\Ga} ( g ^{\otimes 2}) $$

917: for every section $g$.

918: \end{itemize} 

919: \end{theo}

920: 

921: It is easily checked that the operator $\nabla_{X}^{\iota^* L_1}

922: \otimes \identite + \identite \otimes \Lie^{\delta_{\Ga}}_X$  doesn't

923: depend of the choice of the half-form bundle if we consider that it

924: acts on sections of $\iota^* K = \iota^* L_1 \otimes

925: \delta_\Ga$. The

926: same holds with the map $m$. 

927: 

928: \subsection{Proof of Bohr-Sommerfeld conditions}

929: Let us deduce from the previous theorems the Bohr-Sommerfeld conditions for $n$

930: self-adjoint commuting Toeplitz operators $T^1$, $T^2$,..., $T^n$, which is  a slight

931: generalization of \eqref{BS_conditions}. 

932: 

933: Denote by $f^i_0$ and $f^i_1$

934: the principal and subprincipal symbols of $T^i$. Let $E$ be a

935: regular value of $f=(f^1_0,...,f^n_0)$ and $\iota : \Ga \rightarrow M$ be

936: an embedding with image a connected component

937: of $f^{-1} (\{ E\})$. Since $M$ is compact, $\iota (\Ga)$ is a Lagrangian

938: torus. So there exists a half-form bundle $(\delta, \varphi)$ defined over

939: a neighborhood of $\iota (\Ga)$. It is not unique but as usual the

940: final result doesn't depend on the choice of $(\delta, \varphi)$. 

941: Introduce like in the previous section two open sets $U,V$ and a flat section

942: $t_\Ga$. Let us try to solve the eigenvalues equation 

943: \begin{gather} \label{vap_equation} 

944:  T^i \Psi = E^i \Psi + O(k^{-\infty}) \text{ over $V$} \end{gather}  

945: where $\Psi$ is a Lagrangian section associated to $(\Ga, t_{\Ga})$. By

946: theorem \ref{resultat_clef}, the symbol of $(T^i

947: -E^i) \Psi$ is $O(\hb)$ because $\iota ^* f^i = E^i$. Furthermore it

948: is $O(\hb^2)$ if and only if it satisfies the following transport equation

949: \begin{gather} \label{transport_equation}

950:  \bigr[  f^i_1 + ( \nabla_{X^i}^{\iota^*L_1  }

951: \otimes \identite + \identite \otimes \Lie^{\delta_{\Ga}}_{X^i} ) \bigl] g_0 = 0

952: \text{ over $V \cap \Ga$} \end{gather}

953: where $g_0$ is the principal symbol of $\Psi$. This equation can be

954: interpreted as $g_0$ being flat for a connection on $ \iota^* L_1 \otimes \delta_{\Ga}$ that we

955: describe  now. 

956: 

957: First a section $g$ of $\delta_{\Ga}$ is flat if $\mathcal{L}_{X^i}^{\delta_{\Ga}} g

958: = 0$ for every $i$. With this definition, $\varphi_{\Ga} : \delta_{\Ga}^2

959: \rightarrow \La ^n T^*\Ga \otimes \C $ is a morphism of flat bundles, if

960: we endow $\La ^n T^* \Ga \otimes \C $ with the Weinstein connection. Since

961: the form $\be$ such that 

962: $$ \langle  \be,   X^1 \wedge ... \wedge X^n \rangle =1 $$

963: is a global non vanishing flat section of $\La ^n T^* \Ga \otimes \C $,  $\delta_\Ga$ has  holonomy in $\Z_2 \subset U(1)$.

964: 

965: Let $\al \in \Om^1(\Gamma)$ be such that $ \frac{1}{i} \langle \al ,

966: X^j \rangle =  f^j_1$. Consider the connection  $\nabla ^{ \iota^* L_1} + \al$

967: on $\iota^* L_1$. Its flat sections satisfy 

968: $$(f_1^j + \tfrac{1}{i} \nabla^{\iota^* L_1}_{X^j}) s = 0.$$ 

969: Furthermore its curvature vanishes. Indeed, since $[X^i, X^j] = 0$, we have 

970: \begin{xalignat*}{2}

971:  \frac{1}{i}  \langle d \al , X^i \wedge X^j \rangle = &  ( X^i. f_1^j - X^j. f_1^i

972:   ) \\

973: = &  \om_1 ( X^i, X^j) \end{xalignat*}

974: which follows from proposition \ref{subprincipal_calculus} and the fact that

975: $[T^i, T^j] = 0$. 

976:  

977: This defines a structure of flat line bundle for $\iota^*L_1 \otimes

978: \delta_{\Ga} $, whose flat sections are the solutions of \eqref{transport_equation}.

979: Recall that 

980: $$ \Psi(x,k) =  t^k_\Ga (x)  (g_0(x) + O(k^{-1}))

981: \text{ over $V \cap \iota(\Gamma)$},$$

982: where $t_\Ga$ is a flat section of $\iota^* L $. The condition

983: to patch together these sections along $\Ga$ is the Bohr-Sommerfeld

984: condition:

985: $$ \text{$\iota^*( L^k \otimes   L_1) \otimes \delta_{\Ga} \rightarrow \Ga$ is trivial as a flat bundle.} $$

986: When $M$ is two-dimensional, this is equivalent to

987: \eqref{BS_conditions}. To prove theorem \ref{BS_theo} or a similar result in the

988: $2n$-dimensional case for the joint spectrum, we should consider

989: Lagrangian sections depending continuously on $\Ga$. Furthermore, we

990: can show by using a local normal form that the solutions of

991: \eqref{vap_equation} are necessarily Lagrangian sections associated to

992: $\Ga$. A complete proof is in \cite{oim2}. The only novelty here is

993: the formulation of theorems \ref{resultat_clef} and \ref{norme}, and

994: consequently of the Bohr-Sommerfeld conditions. 

995: 

996: \subsection{Comparison with the cotangent case} 

997: To compare theorems \ref{norme} and \ref{resultat_clef} with the similar statements in

998:   the case of pseudo-differential operators, we can introduce some

999:   kind of Maslov

1000:   bundle in the following way. Recall that we denote by $\varphi_{\Ga}$ the

1001:   isomorphism $\delta_{\Ga}^2 \rightarrow \La ^n T^*\Ga \otimes

1002:   \C$. Introduce

1003: $$ P := \bigl\{ u \in \delta_{\Ga} ; \; \varphi_{\Ga} ( u ^{\otimes 2} )

1004: \in \La^n T^* \Ga - \{0 \} \bigr\} .$$

1005: Let $\Z_4$ be the subgroup $\{ 1, -1, i, -i \}$ of $\C^*$. Then $P$ is

1006: a principal bundle with structure group $ \Z_4 \times \R_+$. Introduce

1007: the complex line bundles $| \delta_{\Ga} |$ and $\arg ( \delta_{\Ga}

1008: )$ associated to $P$ via the homomorphism $\Z_4 \times \R_+ \rightarrow

1009: \Z_4$ and $\Z_4 \times \R_+ \rightarrow \R_+$ respectively. Following Weinstein in \cite{We}, we call $\arg (\delta_{\Ga})$ the unitarization

1010: of $\delta_{\Ga}$. We have a

1011: canonical isomorphism

1012: $$ \delta_{\Ga} \rightarrow | \delta_{\Ga} | \otimes \arg

1013: (\delta_{\Ga} ).$$ 

1014: Furthermore, the map

1015: $$ |\delta_{\Ga} | = P

1016: \times_{\R_+} \C \ni  [u,z] \rightarrow z.| \varphi_{\Ga} (u^{\otimes

1017:   2})|^{\frac{1}{2}}\in  |\La |^{\frac{1}{2}} ( \Ga)$$

1018: is an isomorphism between $| \delta_{\Ga} |$ and

1019: the bundle of half-densities of $\Ga$. So we obtain an isomorphism

1020: $$ \zeta: \delta_{\Ga} \rightarrow |\La |^{\frac{1}{2}} ( \Ga) \otimes \arg

1021: (\delta_{\Ga} )  .$$

1022: The bundle $\arg (\delta_{\Ga})$ is a line bundle with structure group $\Z_4$ like the

1023: Maslov bundle. 

1024: The isomorphism $\zeta$ intertwines the operator

1025: $\Lie_X^{\delta_{\Ga}}$ of theorem \ref{resultat_clef} with the Lie

1026: derivative of half-densities. So in the case $L_1$ is trivial, the

1027: theorem \ref{resultat_clef} is similar to the formula 1.3.13 in

1028: \cite{Du3} p223, computing the symbol of an oscillatory integral acted

1029: on under a differential operator. 

1030: Furthermore the map $ \delta_{\Ga}

1031: \rightarrow |\La |(\Ga)$ used in  theorem \ref{norme} is the composition of

1032: $\zeta$ with the squaring map from half-densities into

1033: densities. Again theorem \ref{norme} is similar to formula 1.3.15 in

1034: \cite{Du3} p224, computing the norm of an oscillatory integral.

1035: 

1036: To end this comparison, we apply the previous construction to a

1037: symplectic vector space $E$ and prove that we obtain the usual Maslov

1038: bundle. Consider a one-dimensional vector space $\delta$ with an

1039: isomorphism $\varphi: \delta^2 \rightarrow \La^{n, 0} E^*$. 

1040: Let $\Lag (E)$ be the Lagrangian Grassmannian and $\eta

1041: \rightarrow \Lag (E) $ be the tautological vector bundle, that is the

1042: bundle whose fiber over $x$ is $x$ itself. In the same way we defined

1043: $\varphi_{\Ga}$, one has an isomorphism

1044: $$ \varphi_{\Lag} : \Lag (E) \times \delta^2 \rightarrow \La ^n \eta

1045: \otimes \C,  $$ 

1046: sending $(x,u)$ into $\iota_x^* \varphi(u)$ where $\iota_x$ is the

1047: embedding $ \eta_x \rightarrow E$. Introduce the

1048: $\Z_4 \times \R_+$ bundle 

1049: $$  \bigl\{ (x,u) \in \Lag (E)

1050: \times \delta; \;  \varphi_{\Lag} (x,u^2) \in \La_x^n \eta \text{ and }

1051: u \neq 0 \bigr\} $$

1052: Dividing by $\R_+$ we get a $\Z_4$-principal bundle $\Mas$. We claim

1053: that the holonomy in

1054: $\Mas$ of a loop of $\Lag (E)$ is the mod 4 reduction of its Maslov index.

1055: 

1056: \begin{proof}  Introduce linear

1057: Darboux coordinates $(p^i,q^i)$ and identify $E$ with $\R^{2n}$. Set $z^j= p^j + i q^j$ and let 

1058: $$ \delta := (dz^1\wedge...\wedge dz^n)^{\frac{1}{2}} \C$$ 

1059:  be the square root of $\La^{n,0} E = (dz^1\wedge...\wedge dz^n)\

1060: \C$. Recall that $\Lag (E)$ is isomorphic to $U(n)/O(n)$ (cf. lemma 2.31 of

1061: \cite{McSa}), through the map sending the unitary matrix $U =P +i Q$

1062:  into the range of  

1063: \begin{gather*}  

1064:  A_U = \left( 

1065: \begin{array}{c} P  \\ Q 

1066: \end{array} \right) \end{gather*}  

1067: Let us denote by $\al^1_U$,...,$\al^n_U$ the base of

1068: $\eta_x^*$ dual to the column vectors of $A_U$. A

1069: straightforward computation shows that $\varphi_{\Lag}$ sends the

1070: square of 

1071: $$([U], (dz^1\wedge...\wedge dz^n)^{\frac{1}{2}})

1072: \in \Lag (E) \times \delta $$ into

1073: $\det (U) \ \al^1_U \wedge... \wedge \al^n_U$. Consequently, 

1074: $$ \Mas \simeq \bigl\{ ( [U], v ) \in \Lag (E)

1075: \times \C^* ;

1076: \; U \in U(n) \text{ and } v^2 \det ( U) = \pm 1 \bigr\} $$

1077: Recall now that the

1078: Maslov index  of a loop  $x: \R/ \Z \rightarrow \Lag (E)$ is the degree of $\rho \circ x : S^1 \rightarrow

1079: S^1$ where $\rho$ is the map 

1080: $$ \Lag (E) \rightarrow S^1, \quad [U] \rightarrow \operatorname{det}^2 (U)$$ 

1081: (cf. \cite{McSa} page 53). Its mod 4 reduction is the holonomy of $x$

1082: in $ \Mas$.

1083: \end{proof}

1084: 

1085: 

1086: This result is related to the paper

1087: \cite{We} of Weinstein, where it is observed that the Maslov bundle of

1088: $\Lag (E)$ is a unitarization of a square root of $\La ^n \eta

1089: \otimes \C$.

1090: 

1091: Last remark is that in general the

1092: Maslov bundle of a Lagrangian submanifold of a cotangent space can be

1093: different of the bundle we construct. Indeed notice that the structure

1094: group of $\arg (\delta_{\Ga})$ reduces to $\Z_2$ if and only if $\Ga$

1095: is orientable. Consider a non-orientable manifold $Q$. Then as the

1096: null section of $T^*Q$, $Q$ is a

1097: Lagrangian submanifold and its Maslov bundle is the flat trivial

1098: bundle. So it can not be a unitarization of a square root of $\La^n T^*Q

1099: \otimes \C$.

1100: 

1101: 

1102: %\newpage

1103: \section{Proof}

1104: 

1105: We assume in the whole section 

1106: that there exists a globally defined half-form bundle $(\delta, \varphi)$ and

1107: $K = \delta$. There is no difficulty to generalize to the case where

1108: $K = \delta \otimes L_1$. 

1109: 

1110: \subsection{A preliminary result} 

1111: 

1112: Consider a sequence $\Psi_k \in \Ci(M,L^k \otimes \delta)$ of the

1113: form 

1114: $$\Psi_k(x) =  \Bigl( \frac{k}{2\pi} \Bigr)^{\frac{n}{4}} F^k (x)

1115:  \tilde{g} (x,k) + O(k^{-\infty}) \text{ over $V$, } $$

1116: where $F$ and $\tilde{g}(.,k)$ 

1117:   satisfies the same assumptions as in section \ref{sec:resultat} except that the coefficients

1118:   $\tilde{g}_k$ do not necessarily satisfy $\bar{\partial} \tilde{g}_k \equiv

1119:   0$.  Assume furthermore that $\Psi_k$ is admissible. 

1120: 

1121: \begin{theo} \label{theo:main}

1122: Let $T_k$ be a Toeplitz operator with principal symbol

1123:   $f_0$. Then $T_k \Psi_k$ is a Lagrangian section over $V$ with

1124:   symbol $\iota^* (f_0 \tilde{g}_0) + O(\hb)$. 

1125:  

1126: Furthermore, if $\tilde{g}_0$

1127:   and its first derivatives vanish along $\iota(\Ga)$, then the

1128:   symbol of $T_k \Psi_k$ is

1129: $$   \hb (\iota^* f_0) \bigl(  \square \tilde{g}_0 + \iota^* \tilde{g}_1) ) + O(\hb^2)

1130: $$

1131: where $\square \tilde{g}_0 \in \Ci(\Ga, \delta_{\Ga})$ and at every $x \in \Ga$ 

1132: $$  \square \tilde{g}_0  (x) = - \frac{1}{2}  \sum \bar{\partial}_i \bar{\partial}_i

1133: \tilde{g}_0 (\iota(x))$$

1134: if $\partial_1,...,\partial_n$ is a base of vectors of

1135: $T^{1,0}_{\iota(x)} M$ such that  

1136: $\frac{1}{i} \om (\partial_{i}, \bar{\partial}_j)

1137: = \delta_{ij}$

1138: and the vectors $\partial_{i} + \bar{\partial}_i$ are

1139: tangent to $\Ga$.

1140: \end{theo}

1141: 

1142: The proof starts from the following representation of the Schwartz

1143: kernel of the Toeplitz operator $T_k$:

1144: \begin{gather} \label{eq:noyau_Toeplitz}

1145:  T_k (x,y) = \Bigl( \frac{k}{2 \pi} \Bigr)^n E^{k} (x,y) \tilde{f}

1146: (x,y,k)  + O (k^{-\infty}) \end{gather}

1147: where, if we consider $M^2$ as a complex manifold with complex

1148: structure $(j , -j)$,

1149: \begin{itemize}

1150: \item 

1151: $E$ is a section of $L \boxtimes \bar{L}

1152: \rightarrow M^2$ satisfying  

1153: $$ E (x,x) = u \otimes \bar{u}, \quad \forall u \in L_x \text{ such

1154:   that }

1155:   \| u \| = 1,

1156: $$ $ \bar{\partial} E \equiv 0 $

1157: modulo a section vanishing to every order along the diagonal $\De$ of

1158: $M^2$ and  $\| E(x,y)

1159: \| <1 $ if $x \neq y$.

1160: \item

1161:   $\tilde{f}(.,k)$ is a sequence of sections of $ \delta \boxtimes

1162:   \bar{\delta} \rightarrow M^2$ with an asymptotic expansion of the form 

1163: $$ \tilde{f}(.,k) = \tilde{f}_0 + k^{-1} \tilde{f}_1 + ...$$

1164: whose coefficient satisfy $\bar{\partial} f_l \equiv 0  $

1165: modulo a section vanishing to every order along $\De$.  Furthermore, $$\tilde{f}_0 (x,x) = f_0 (x),$$

1166: where $f_0$ is the principal symbol of $T_k$. 

1167: \end{itemize}

1168: In other words, $T_k (.,.)$ is a

1169: Lagrangian section associated to the diagonal $\De$ of $M^2$. This

1170: result was proved in \cite{oim1}, without the additional bundle

1171: $\delta$. The generalization is straightforward. 

1172: 

1173: Since the norm of $E$ is $<1$ outside the diagonal and $\Psi_k$ is

1174: admissible, one has for every $x$ in $V$

1175: $$ (T_k \Psi _k ) (x)= \Bigl( \frac{k}{2 \pi} \Bigr)^{n+ \frac{n}{4}}  \int E^k

1176: (x,y).F^k(y) \ \tilde{f}(x,y,k). \tilde{g} (y,k)\ \mu_M(y) +

1177: O(k^{-\infty})$$ 

1178: where we integrate on a neighborhood of $x$. Introduce a unitary

1179: section $t$ of $L$ over $U$ such that $\iota^* t = t_\Gamma$ over

1180: $U_{\Gamma}$ and let us write  

1181: \begin{gather} \label{eq:def_phi}

1182:  E(x,y).F(y) = e^{i \phi(x,y) } t(x) .\end{gather} 

1183: Then the imaginary part of $\phi$ is non positive and vanishes only if

1184: $(x,y)$ belongs to 

1185: $$ C:= \{ (x,x) \in M^2; \; x \in \iota(\Ga) \} .$$

1186: To compute the derivatives of $\phi$ along $C$, recall the following

1187: lemma proved in \cite{oim2} (cf. proposition 2.2 p.1535). 

1188: \begin{lemme} \label{lem:derivee_F}

1189: If $\nabla^{L} F = \tfrac{1}{i} \al_F \otimes F$, then

1190:   $\al_F$ vanishes along $\iota(\Ga)$ and for every vector field $X,Y$ 

1191: $$ \Lie_X \langle \al_F , Y \rangle  = \om ( q X, Y)$$

1192: at $x \in \iota(\Ga)$, where $q$ is the projection onto $T_x^{0,1}M$ with kernel

1193: $T_x \iota(\Ga) \otimes \C$. 

1194: \end{lemme} 

1195: As a corollary, we have the following

1196: \begin{lemme} \label{lem:derivee_E}

1197: If $\nabla^{L \boxtimes \bar{L}} E =

1198: \tfrac{1}{i} \al_E \otimes E$, then $\al_E$ vanishes along the

1199: diagonal, and for every vector fields $X_1,Y_1,X_2,Y_2$ of

1200: $M$,

1201: $$ \Lie_{(X_1,Y_1)}. \langle \al_E, (X_2,Y_2) \rangle = \om (

1202: X_1^{0,1} - Y_1^{0,1}, X_2) + \om (

1203: X_1^{1,0} - Y_1^{1,0}, Y_2)$$

1204: along the diagonal, where we denoted by $X^{1,0}$ and $X^{0,1}$ the

1205: holomorphic and anti-holomorphic parts of $X$ respectively.

1206: \end{lemme}

1207: We deduce from both lemmas that $d_y \phi$ vanishes along $C$. Furthermore

1208: the kernel of the tangent map to $d_y \phi $ at $(x,x) \in C$ is 

1209: $$ \bigl( T_x ^{0,1}M \times (0) \bigr) \oplus \bigl( T_{(x,x)} C \otimes \C \bigr) $$

1210: Finally, we have along $C$, 

1211: $$ d^2_y \phi  (Y_1,Y_2) = \om ( Y_1^{1,0}, Y_2) - \om ( q Y_1, Y_2) $$ 

1212: and $d^2_y \phi $ is non-degenerate. So we can apply the stationary

1213: phase lemma (cf. \cite{Ho} section 7.7 or theorem \ref{theo:PS1}).

1214: One gets

1215: $$ (T_k \Psi_k ) (x)   = \Bigl( \frac{k}{2 \pi} \Bigr)^{\frac{n}{4}} e^{ik

1216:   \phi_r (x) } t^k(x) \ \tilde{h} (x,k) + O( k^{-\infty})$$

1217: where 

1218: \begin{gather} \label{eq:def_phir} 

1219: \phi_r (x) \equiv \phi (x,y) \end{gather} 

1220: modulo a linear combination with

1221:   $\Ci$ coefficients of the

1222:   $\partial_{y^i} \phi (x,y)$.

1223: \begin{lemme} $e^{ik

1224:   \phi_r (x) } t(x)$ satisfies the same assumption as the section

1225:   $F$. 

1226: \end{lemme}

1227: \begin{proof} 

1228: Since $\phi$ and $d_y \phi$ vanishes along $C$, $\phi_r$ vanishes

1229: along $\iota(\Ga)$

1230: and consequently $$ e^{i

1231:   \phi_r (x) } t(x) = t_{\Ga} (x)$$ for every $x$ in

1232: $\iota(\Ga)$. Introduce complex coordinates $x^1,...,x^n$ and write

1233: $$ \nabla t = \tfrac{1}{i}  t \otimes \sum a_j (x) dx^j + \bar{a}_j

1234: (x) d\bar{x}^j .$$

1235: Derivating \eqref{eq:def_phi}, it follows from $\bar{\partial} E

1236: \equiv 0$ that  

1237: $$ \partial_{\bar{x}^i} \phi (x,y) \equiv \bar{a}_j (x) \mod

1238: \ideal_{\De} (\infty),$$ i.e. modulo a function vanishing to infinite

1239: order along the diagonal. Derivating again, one has

1240: $$\partial_{\bar{x}^i} \partial_{y^j} \phi (x,y) \equiv 0 \mod

1241: \ideal_{\De} (\infty) $$

1242: Then we deduce from the two previous equations and

1243: \eqref{eq:def_phir} that for every multi-index $\al$,

1244: $$ \partial_{\bar{x}^1}^{\al(1)}...\partial_{\bar{x}^n}^{\al(n)} \bigl(

1245: \partial_{\bar{x}^i} \phi_r  - \bar{a}_i \bigr) (x) = 0 $$

1246: along $\iota(\Ga)$. And consequently 

1247: $$ \partial_{\bar{x}^i} \phi_r  \equiv \bar{a}_i \mod \ideal_{\iota(\Ga)}

1248: (\infty)$$

1249: which proves the result. 

1250: \end{proof}   

1251: 

1252: \begin{lemme} 

1253: The sequence $\tilde{h}(.,k)$ admits an asymptotic

1254:   expansion $\tilde{h}_0 + k^{-1} \tilde{h}_1 +...$  whose

1255:   coefficients satisfy $\bar{\partial} \tilde{h}_\ell \equiv 0$ modulo a

1256:   section vanishing to every order along $\Ga$. 

1257: \end{lemme} 

1258: 

1259: \begin{proof} First one deduces from lemma \ref{lem:derivee_F} that

1260:   the imaginary part of $\phi_r$ and its first derivatives vanishes

1261:   along $\iota(\Ga)$. Furthermore the Hessian of $\Im \phi_r$ along $\iota(

1262:   \Ga)$ is non-degenerate in the

1263:   transverse direction to $\iota(\Ga)$. As a consequence, if $\tilde{e}

1264:   (x,k)$ is a sequence with an asymptotic expansion $\tilde{e}_0 +

1265:   k^{-1} \tilde{e}_1 +...$ such that 

1266: $$ e^{ik \phi_r (x)}   \tilde{e} (x,k) = O ( k^{-\infty}) $$

1267:  then the coefficients $\tilde{h}_\ell$ vanish to every order along

1268:  $\iota(\Ga)$. This was proved in \cite{oim1} (cf. lemma 1, p.6). We apply this to the

1269:  sequence $\bar{\partial} T_k \Psi_k$ which vanishes, since $ \Pi_k T_k

1270:  = T_k$ implies that $T_k \Psi_k$ belongs to

1271: $\Hilbert_k$.  

1272: \end{proof} 

1273: 

1274: The two previous lemmas imply that $T_k \Psi_k$ is a Lagrangian section. 

1275: Then applying theorems \ref{theo:PS1} and \ref{theo:PS2}, we obtain

1276: the symbol of $T_k\Psi_k$ by computations of linear algebra, which are easily done using

1277: the tangent vectors $\partial_i$ introduced in the statement of

1278: theorem \ref{theo:main}.

1279: 

1280: \scratch{Quelques d{\'e}tails suppl{\'e}mentaires: Pour d{\'e}duire des

1281:   lemmes \ref{lem:derivee_E} et \ref{lem:derivee_F} l'application tangente {\`a} $d_y \phi$, on

1282:   d{\'e}rive \eqref{eq:def_phi}, ce qui nous donne

1283: $$  \al_E + \pi_y^* \al_F = - d \phi + \pi_x^* \al_t $$

1284: o{\`u} $\nabla t = \frac{1}{i} t \otimes \al_t$, et $\pi_x$ et $\pi_y$

1285: sont les projections de $M^2$ sur le premier et second facteur

1286: respectivement. On en d{\'e}duit que 

1287: $$ \Lie_{(X_1,Y_1)}. \langle d_y \phi , Y_2 \rangle = - \om (

1288: X_1^{0,1} - Y_1^{0,1}, X_2) - \om ( q Y_1,Y_2 ) $$

1289: On v{\'e}rifie imm{\'e}diatement que $ \bigl( T_x ^{0,1}M \times (0)

1290: \bigr) \oplus \bigl( T_{(x,x)} C \otimes \C \bigr) $ est inclus dans le

1291: noyau. Pour l'inclusion r{\'e}ciproque, il suffit de raisonner sur les dimension

1292: une fois que l'on a montr{\'e} que $d_y^2 \Phi$ est

1293: non-d{\'e}g{\'e}n{\'e}r{\'e}e. Pour montrer ceci, on calcule la matrice de

1294: $\frac{1}{i} d_y^2 \Phi$ dans la base $\frac{1}{\sqrt{2}} ( \partial_i

1295: + \bar{\partial}_i), \frac{i}{\sqrt{2}} ( \partial_i

1296: - \bar{\partial}_i)$, on obtient

1297: $$ \frac{1}{2} \left( 

1298: \begin{array}{cc} 1 & -i  \\ -i & 3 

1299: \end{array} \right)

1300: $$

1301: dont le d{\'e}terminant est $1$. Et ceci nous calcule de plus

1302: $\tilde{h}_0$ sur la diagonale (il ne faut pas oublier la mesure de

1303: Liouville qui ne contribue pas car la base pr{\'e}c{\'e}dente est orthonorm{\'e}e pour

1304: la m{\'e}trique $\om (X, JY)$). Ensuite pour calculer le terme suivant

1305: lorsque $\tilde{g}_0$ s'annule au second ordre le long de

1306: $\iota(\Ga)$, on applique le th{\'e}or{\`e}me \ref{theo:PS2}. Comme base

1307: de $ T_x ^{0,1}M \times (0)

1308: $ qui est dans le noyau de l'application tangente {\`a} $d_y

1309: \phi$ et en somme directe avec $ T_{(x,x)} C \otimes \C$, on choisit

1310: les $(\bar{\partial}_i,0)$. Come base de $(0) \times (T_x M \otimes

1311: \C)$, on choisit les $\partial_i, \bar{\partial_i}$. La matrice de

1312: $\frac{1}{i} d^2_y \phi$ dans cette base est 

1313: $$  \left( 

1314: \begin{array}{cc} -1 & 1  \\ 1 & 0 

1315: \end{array} \right)

1316: $$

1317: qui admet pour inverse 

1318: $$   \left( 

1319: \begin{array}{cc} 0 & 1  \\ 1 & 1 

1320: \end{array} \right)

1321: $$

1322: On obtient le r{\'e}sultat du th{\'e}or{\`e}me avec $$\square \tilde{g}_0 = \Delta

1323: \tilde{g}_0 = \sum \partial_i \bar{\partial}_i \tilde{g}_0 +

1324: \frac{1}{2}\bar{\partial}_i\bar{\partial}_i \tilde{g}_0$$ 

1325: On conclut en utilisant que $ \partial_i \bar{\partial}_i \tilde{g}_0

1326: = - \bar{\partial}_i\bar{\partial}_i \tilde{g}_0$ car $\partial_i +

1327: \bar{\partial}_i$ tangent {\`a} $\iota( \Ga)$. }

1328: 

1329: \subsection{Proofs of the theorems of part \ref{sec:resultat}}

1330:     

1331: A first corollary of theorem \ref{theo:main} is the existence of a

1332: Lagrangian section with an arbitrary symbol: applying theorem \ref{theo:main} with the Toeplitz operator $\Pi_k$,

1333: we construct a Lagrangian section with a prescribed principal symbol,

1334: then theorem \ref{existence} follows from Borel resummation. Theorem

1335: \ref{Toep_Lag} is a particular case of theorem \ref{theo:main}.

1336: 

1337: To prove theorem \ref{resultat_clef}, we can assume that $f_0$

1338: vanishes along $\Ga$. Since we compute the symbol of

1339: $T_k \Psi_k$  modulo $O(\hb^2)$, we can replace $T_k$ with every

1340: Toeplitz operators of symbol $f_0 + \hb f_1 + O(\hb^2)$. So by theorem

1341: \ref{normalised_symbol_quantisation}, we can choose

1342: $$ T_k = \Pi_k \Bigr( f_0 + k^{-1} f_1 + \frac{1}{ik} ( \nabla^{L^k}_X

1343: \otimes \identite + \identite \otimes D_X ) \Bigl)  $$

1344: where $X$ is the Hamiltonian vector field of $f_0$. So $T_k \Psi_k$ is

1345: equal to 

1346: \begin{gather} \label{La}

1347:  \Pi_k \Bigl[ \bigl( \tfrac{k}{2 \pi} \bigr)^{\frac{n}{4}} \bigl( [f_0 + \tfrac{1}{ik} \nabla_{X}^{L^k} ] F^k

1348: \bigr) \tilde{g}(.,k) \Bigr] + k^{-1} \Pi_k \Bigl[  \bigl( \tfrac{k}{2

1349:   \pi} \bigr)^{\frac{n}{4}} F^k \bigl( [f_1 +

1350: \tfrac{1}{i} D_X ] \tilde{g}(.,k) \bigr)  \Bigr] \end{gather}  

1351: By theorem \ref{theo:main}, each term of the sum is a Lagrangian

1352: section. Furthermore by the first part of this theorem, the symbol of the second one is 

1353: $$ \hb\ \bigl( (\iota^* f_1).g_0 + \iota^*(\tfrac{1}{i} D_X

1354: \tilde{g}_0) \bigr) +O(\hb^2).$$

1355: Since $X$ is tangent to $\iota(\Ga)$, $\iota ^* (D_X

1356: \tilde{g}_0)$ only depends on the restriction of $\tilde{g}_0$

1357: to $\iota(\Ga)$. So we can define the operator $\iota^* D_X$ acting on $\Ci

1358: (\Ga, \delta_\Ga)$ which sends $g_0$ to $ \iota ^* (D_X

1359: \tilde{g}_0)$. And the previous symbol is 

1360: \begin{gather} \label{2}

1361:  \hb\ \bigl( (\iota^* f_1) + \tfrac{1}{i} (\iota^* D_X) \bigr).g_0

1362:  +O(\hb^2). 

1363: \end{gather}

1364: 

1365: 

1366: To compute the symbol of the first term of \eqref{La}, let us write

1367: $$  [f_0 + \tfrac{1}{ik} \nabla_{X}^{L^k} ] F^k

1368:  = F^k a $$

1369: with $a$ defined on a neighborhood of $\iota(\Ga)$.  

1370: \begin{lemme} \label{li2}

1371: The function $a$ and its first derivatives vanish along $\iota(\Ga)$. If $Z$

1372: and $W$ are holomorphic vector fields of $M$, then

1373: $$ \Lie_{\bar{W}} \Lie_{\bar{Z}} a =  \om( \bar{W} , \Lie_X \bar{Z}  ) $$

1374: on $\iota( \Ga)$.

1375: \end{lemme}

1376: 

1377: \begin{proof} 

1378: Denote by $\alpha _F$ be the one-form such that $ \nabla^{L} F = 

1379: \frac{1}{i}\al_F \otimes F$. Then

1380: $$ a = f_0 - \langle \al_{F} , X \rangle .$$

1381: This vanishes along $\iota ( \Ga)$ because $\al_F$ vanishes along $\iota(\Ga)$

1382: (cf. lemma \ref{lem:derivee_F}).

1383: 

1384: Since $X$ is the Hamiltonian vector field of $f_0$, one has $\Lie_Y

1385: f_0 + \om (X,Y) =0$. Since the curvature of $L$ is $\frac{1}{i}\om$,

1386: one has $d \al_F = \om$ and consequently 

1387: $$\Lie_Y \langle \al_F, X

1388: \rangle = \Lie_X \langle \al_F, Y

1389: \rangle + \om(Y,X) + \langle \al_F, [Y,X] \rangle.$$ It follows that

1390: $$ \Lie_Y a = - \Lie_X \langle \al_F, Y \rangle - \langle \al_F, [Y,X]

1391: \rangle .$$

1392: This vanishes along $\iota(\Ga)$, because $\al_F$ vanishes along $\iota(\Ga)$

1393: and $X$ is tangent to $\iota(\Ga)$. 

1394: 

1395: Since $\bar{\partial} F \equiv 0$ modulo a flat section and $Z$ is holomorphic, $  \langle

1396: \al_F, \bar{Z} \rangle$ vanishes to every order along $\iota(\Ga)$. So choosing

1397: $Y =\bar{Z}$ in the previous equation, we obtain

1398: $$ \Lie_{\bar{W}}.\Lie_{\bar{Z}} a = - \Lie_{\bar{W}} \langle \al_F,

1399: [\bar{Z}, X] \rangle \qquad \text{ along } \iota(\Ga) .$$

1400: Using again that $\om = d\al_F$ and $\al_F $ vanishes along $\iota (\Ga)$, it

1401: follows that

1402: $$  \Lie_{\bar{W}}.\Lie_{\bar{Z}} a = - \om ( \bar{W}, [\bar{Z}, X ] )

1403: - \Lie_{[\bar{Z},X ]} \langle \al_F, \bar{W} \rangle $$

1404: along $\iota ( \Ga)$. The second term of the right side vanishes along

1405: $\iota ( \Ga)$ because $\bar{W}$ is an anti-holomorphic vector

1406: field. This gives the result.

1407: \end{proof} 

1408: 

1409: Since $\iota^* D_X$ and $\Lie_X^{\delta_{\Ga}}$ are first order

1410: differential operators which have the same symbol,

1411: $$ \iota^* D_X - \Lie_X^{\delta_{\Ga}} = b $$

1412: where $b$ is a function of $\Ga$. 

1413: 

1414: \begin{lemme} \label{li3} 

1415: The symbol of $ \Pi_k \bigl[ F^k a \tilde{g}(.,k) \bigr]$ is 

1416: $  i  \hb  b. g_0 + O(\hb^2)$

1417: \end{lemme} 

1418: 

1419: 

1420: So the symbol of $T_k \Psi_k$ is the sum of \eqref{2} and $ i 

1421: \hb b g_0 + O(\hb^2)$, which is equal to 

1422: $$ \hb(\iota^* f_1 +  \tfrac{1}{i} \Lie_X^{\delta_{\Gamma}}) . g_0 + O(\hb^2). $$

1423: Theorem \ref{resultat_clef} follows. 

1424: 

1425: \begin{proof} 

1426: Let us start with a local computation of the function $b$.  

1427: Let $u$ be a non-vanishing section of $\delta_{\Ga} \rightarrow

1428: \Ga$. Then one has 

1429: $$  b = \frac{ \bigl( \iota^* D_X - \Lie_X^{\delta_{\Ga}} \bigr) .u

1430: }{u}$$

1431: Let $\beta$ be a non-vanishing $(n,0)$-form of $M$ such that $\varphi

1432: (u^{\otimes 2}(x)) = \beta (\iota(x))$ if $x$ be\-longs to $\Ga$. Then

1433: $\iota^* D_X $ is defined in such a way that 

1434: $$ \frac{ ( \iota^* D_X).u}{u} = \frac{1}{2} \iota^* \Bigl(\frac{ p

1435:   \Lie_{X} \be}{\be}\Bigr) = \frac{1}{2} \frac{ \iota^* (p

1436:   \Lie_{X} \be)}{\iota^* \be} .$$

1437: where $p$ is the projection from $\La^{n} M \otimes \C$ onto

1438:   $\La^{n,0} M$ with kernel 

1439: $ \La^{0,n} M \oplus ... \oplus \La^{n-1,1} M.$ 

1440: On the other hand, since 

1441: $\varphi_{\Ga}(u^{\otimes 2} (x)) = \iota^* \varphi

1442:   (u^{\otimes 2} (x) ) = \iota^* \be (x) $, 

1443: one has

1444: $$ 

1445:  \frac{ \Lie_X^{\delta_{\Ga}}.u

1446: }{u} = \frac{1}{2} \frac{ \Lie_X \iota^* \be}{\iota^* \be} = \frac{1}{2} \frac{ \iota^* \Lie_X \be}{\iota^* \be} .$$

1447: Consequently

1448: $$ b = \frac{1}{2} \frac{ \iota^* \bigl(p \Lie_X \be-  \Lie_X

1449:   \be\bigr) }{\iota ^* \be}. $$

1450: 

1451: Now let us choose a frame $(\partial_1,...,\partial_n)$ of holomorphic

1452:   vector fields of $M$ such that the vectors $\partial_i + \bar{\partial_i}$ are tangent

1453:   to $\iota(\Ga)$. Denote by $(\te^{1},..., \te^n)$ the dual frame and set

1454: $$ \be = \te^1 \wedge... \wedge \te^n .$$ 

1455: Then using that $\Lie_X \te^i \equiv - \sum \langle \te^i, \Lie_X

1456: \bar{\partial}_j \rangle \bar{\te}^j $ modulo a linear combination of the

1457: $\te^i$, we obtain that $\iota^* (p \Lie_X \be-  \Lie_X

1458:   \be)$ is equal to 

1459: \begin{gather*} 

1460: \sum \langle \te^1, \Lie_X \bar{\partial}_j \rangle

1461:   \bar{\te}^j \wedge \te^2 \wedge ...\wedge \te^n +  \langle \te^2,

1462:   \Lie_X \bar{\partial}_j \rangle \te^1 \wedge \bar{\te}^j \wedge

1463:   \te^3 \wedge ...\wedge \te^n + ...\\

1464: +  \langle \te^n,

1465:   \Lie_X \bar{\partial}_j \rangle \te^1 \wedge ...\wedge \te^{n-1}

1466:   \wedge \bar{\te}^j \end{gather*} 

1467: It follows then from $\iota^* \te^i  = \iota^* \bar{\te}^i$ that 

1468: $$ b = \frac{1}{2} \sum \langle \te^i, \Lie_X

1469: \bar{\partial}_i \rangle .$$

1470: 

1471: To end the proof, assume furthermore that $\frac{1}{i} \om ( \partial_i

1472: ,\bar{\partial}_j) = \de_{ij}$. Then it follows from the second part of theorem \ref{theo:main}

1473: that the symbol of $ \Pi_k \bigl[ F^k a \tilde{g}(.,k) \bigr]$ is 

1474: $$ - \hb g_0  \frac{1}{2} \ \iota^* \sum \Lie_{\bar{\partial}_i}

1475: \Lie_{\bar{\partial}_i} a $$

1476: which by lemma \ref{li2} is equal to  

1477: $$ - \hb g_0 \frac{1}{2}  \ \iota^* \sum \om ( \bar{\partial}_i, \Lie_X

1478: \bar{\partial}_i)$$

1479: Using again that $\frac{1}{i} \om ( \partial_i

1480: ,\bar{\partial}_j) = \de_{ij}$, we obtain

1481:  $$   \hb g_0 \frac{i}{2} \ \iota^*  \sum \langle \te^i, \Lie_X

1482: \bar{\partial}_i \rangle .$$

1483: The final result follows. 

1484: \end{proof} 

1485: 

1486: Finally, let us prove theorem \ref{norme}.

1487: One has

1488: $$ \int_M \xi \  \bigl\| \Psi_k \bigr\|_{L^k \otimes \delta} ^2 \; \mu_M =

1489: \Bigl( \frac{k}{2\pi} \Bigr)^{\frac{n}{2}} \int_M  e^{-k c} \ \xi \

1490: \bigl\| \tilde{g}(.,k) \bigr\|_{\delta}^2 \; 

1491: \mu_M +O(k^{-\infty}) $$

1492: where $c(x) = -2 \ln \| F(x) \| _{L}$. The following is a

1493: consequence of lemma

1494: \ref{lem:derivee_F}.

1495: \begin{lemme} \label{lem:norme_F}

1496: The function $c$ and its first derivatives vanish along

1497: $\iota(\Ga)$. Furthermore, its Hessian at $x \in \iota(\Ga)$ is definite positive on

1498: $J T_x \iota(\Ga)$ and is given by 

1499: $$  X.Y.c  = 2 \om (X, JY), \quad X,Y \in J T_x \iota(\Ga).$$

1500: \end{lemme} 

1501: So integrating along transversal directions to $\Ga$, it follows from the

1502: stationary phase lemma that 

1503: $$  \int_M \; \xi \ \bigl\| \Psi_k \bigr\|_{L^k  \otimes \delta}

1504: ^2 \;  \mu_M = \int_{\Gamma} (\iota^* \xi)\  d +O(k^{-1})$$

1505: where $d$ is the density of $\Ga$ such that 

1506: $$ d\eval{x} ( X) =   \bigl\| g_0(x) \bigr\|^2_{ \delta}

1507: \  \mu_{M} \eval{x} (X \wedge Y) \ 2^{-\frac{n}{2}} \Bigl( \det \bigl[ \om (Y_i,J Y_j)

1508: \bigr]\Bigr)^{-\frac{1}{2}}.$$

1509: Here $(X_i)$ and $(Y_i)$ are bases of $T_x \Ga$ and $J T_x \Ga$

1510: respectively, and $X = X_1 \wedge...\wedge X_n$, $Y = Y_1

1511: \wedge ... \wedge Y_n$. To deduce

1512: theorem \ref{norme}, we have to check that  

1513: \begin{gather*}

1514: d = m (g_0)  

1515: \end{gather*}

1516: This is easily done, by introducing the same vector fields

1517: $\partial_i$ and forms $\te^i$ like in the

1518: proof of lemma \ref{li3}, setting $X_i = \frac{1}{\sqrt{2}}(

1519: \partial_i + \bar{\partial}_i)$, $Y_j = J X_j$ and choosing $g_0$ such that $ \varphi (g_0^{\otimes

1520:   2} ) = \theta_1 \wedge ...\wedge \theta^n$. 

1521: 

1522: \scratch{Quelques d{\'e}tails: Pour d{\'e}duire le lemme \ref{lem:norme_F}

1523:   du lemme \ref{lem:derivee_F}, on utilise que si $X \in J

1524:   T\iota(\Ga)$, alors $q(JX) = 0$ et $q(X+ i JX) = X+iJX$, donc $q(X)

1525:   = X + iJX$. Apr{\`e}s, on calcule

1526: $$  X.Y .c = - \frac{X .Y. \| F \|^2}{\| F \|^2} = - \frac{1}{i} X. \langle \al_F -

1527: \bar{\al}_F, Y \rangle .$$  

1528: Ce qui donne avec le lemme \ref{lem:derivee_F}

1529: $$ -\frac{1}{i} \om (X + iJX - (X - iJX), Y) 

1530: = -2 \om ( JX,Y) = 

1531: 2 \om (X, JY).$$

1532: Voil{\`a}. Ensuite, pour la v{\'e}rification de $d = m(g_0)$, avec les

1533: choix que j'indique $\om (Y_i, J Y_j) =\de_{ij}$, $ \mu_M ( X \wedge

1534: Y) = 1$, et $\te^1 \wedge ...\wedge \te^n$ {\'e}tant de norme 1, il

1535: vient $d(X) = 2^{-\frac{n}{2}}$. D'autre part $\langle \iota^*

1536: \te^i, X_j \rangle = \frac{1}{\sqrt{2}} \de_{ij} $, donc $\langle \iota^*

1537: \varphi (g_0^2) , X \rangle =  2^{-\frac{n}{2}}$, autrement dit $m(

1538: g_0) (X) = 2^{- \frac{n}{2}}$.  

1539: }

1540: 

1541: %\newpage

1542: \section{Appendix} 

1543: 

1544: Let $W$ be an open set of $\R^{n} \times \R^{k} \ni (x,y)$. Denote by

1545: $p$ the projection from $W$ onto $\R^n$. Let $\ph (x,y)$ be a $\Ci$

1546: function on $W$ whose imaginary part is $ \geqslant 0$. Let $a(x,y)$

1547: be a $\Ci$ function with compact support in $W$. Stationary phase

1548: lemma gives the asymptotic expansion  of  

1549: $$ I(a, \ph)(x, \tau ) = \int_{\R^k} e^{i \tau \ph (x,y) } a(x,y) |dy| $$

1550: when $\tau \rightarrow \infty$.

1551: First, if the support of $a$ doesn't meet the critical locus 

1552: $$ C := \{ (x,y)\in W  ; \; d_y \ph (x,y) = 0 \text{ and } \Im \ph (x,y) = 0 \}, $$ 

1553: then  $I(a, \ph)$ is $O(\tau^{-\infty})$. Introduce the functions

1554: $$\ph_{i,j}(x,y) = \partial_{y^i} \partial_{y^j} \ph

1555: (x,y), \quad i,j =1,...,k.$$ 

1556: The following theorem is proved in \cite{Ho} section 7.7. 

1557: 

1558: \begin{theo} \label{theo:PS1}

1559: Assume that at $(x_0,y_0) \in C$, the matrix $

1560: \bigl(\ph_{i,j}(x_0,y_0)\bigr)$ is invertible. Then there exists a

1561: neighborhood $U$ of $(x_0, y_0)$ such that if the support of $a$ is a

1562: subset of $ U$, one has 

1563: $$ I(\ph, a) (x, \tau) =( \tfrac{2 \pi }{ \tau} )^{\frac{k}{2}} \ d (x) \

1564: e^{i \tau \ph_r (x) }\ b(x,\tau) + O(\tau^{-\infty}) \qquad \text{ over

1565: } p(U)

1566: $$ 

1567: where  $d$, $\ph_r$ et $b(.,\tau)$ are $\Ci$ functions such that  

1568: \begin{itemize} 

1569: \item $d$ only depends on $\ph$. In particular, 

1570:  $$ d(x) = \operatorname{det}^{-\frac{1}{2}} [ \tfrac{1}{i}

1571:  \varphi_{j,k} (x,y)]_{j,k}, \quad  \text {if } (x,y) \in C \cap U.$$

1572: \item $\ph_r$ is such that $ \ph (x,y) \equiv \ph_r (x) $ on $U$

1573:   modulo a linear combination with $\Ci$ coefficient of the functions  $\partial_{y^j} \ph$. 

1574: \item  $b(.,\tau)$ has an asymptotic expansion for the $\Ci$ topology of the form 

1575:   $$ b_0 (x) + \tau b_1(x) + \tau^2 b_2 (x) +...$$ 

1576: Furthermore,   $b_0(x) = a(x,y)$

1577: if  $(x,y) \in C \cap U$. 

1578: \end{itemize} 

1579: \end{theo} 

1580: In \cite{Ho} the various terms of the asymptotic expansion are

1581: completely determined and not only their restriction at $p(C)$. But in the applications this leads to

1582: complicated computations that we prefer to avoid. Let us introduce an

1583: additional assumption.

1584: 

1585: Denote by $E_{(x,y)}$ the complexification of the tangent space to the

1586: fiber of $p$ at  $(x,y)$. 

1587: At $(x,y) \in C$ the tangent map to the section  $d_y

1588: \ph$ of $E^*$ 

1589: $$ T_{(x,y)} d_{y} \ph : T_{(x,y)} W \otimes \C \rightarrow E^*_{(x,y)} $$ 

1590:  is well-defined. Assume that  $(\varphi_{i,j})$ is invertible along

1591:  $C$, that is the kernel   $F_{(x,y)}$ of $ T_{(x,y)} d_{y} \ph$

1592: satisfies  

1593: \begin{gather} \label{H1}

1594:  \forall \ (x,y) \in C, \quad  F_{(x,y)} \oplus E_{(x,y)}  =

1595:   T_{(x,y)} W \otimes \C. \end{gather}

1596: Assume furthermore that 

1597: \begin{gather} \label{H2}

1598: \text{$C$ is a submanifold of $W$ and } 

1599: T C \otimes \C = F \cap

1600:   \bar{F} . \end{gather} 

1601: Finally, these two assumptions imply that the restriction $p : C \rightarrow

1602:   \R^n$ is an immersion. We assume it is an embedding. 

1603:    

1604: Observe that when the phase takes real values, the assumption

1605: \eqref{H2} is a consequence of 

1606: \eqref{H1}. We are interested in the opposite case, typically when the

1607: Hessian of the imaginary part of the phase is non-degenerate  in the

1608: transverse directions to $C$, for instance with $$\ph (x,y) = xy + \tfrac{i}{2}(x^2 +

1609: y^2).$$ 

1610: We can also consider intermediary cases, for example  $\ph (x,y) = xy + \frac{i}{2} y^2$. 

1611: 

1612: Under the previous assumptions, when the amplitude $a$ vanishes to

1613: order $m$ along $C$, i.e. when the partial derivatives of $a$ of order 

1614: $\leqslant m-1$ vanish along $C$, it follows from the result of

1615: \cite{Ho} that the functions $b_i$ vanish to order $m-2i$ along

1616: $C$. Furthermore one can easily compute 

1617: $b_i$ modulo a function vanishing to order   $m-2i+1$ along 

1618: $C$. 

1619: 

1620: To state the result, consider a free family $\partial_1,...,

1621: \partial_l$ of complex tangent vectors to $W$ at  $(x,y) \in

1622:    C$ such that 

1623: $$ \operatorname{Vect}_{\C} (\partial_1,..., \partial_l) \oplus  (TC

1624: \otimes \C) = F_{(x,y)}.$$

1625: If $a$ vanishes to order  $m$ along  $C$, we define the polynomial 

1626: $$ [a] (Z,Y) = \sum_{|\al| + |\be|=m} \frac{1}{\al ! \be !} \Bigl( 

1627: \partial_1^{\al(1)}...\partial_l^{\al(l)}\partial_{y^1}^{\be(1)}...\partial_{y^k}^{\be(l)}

1628: a(x,y) \Bigr) \ Z^{\al} Y^{\be}$$

1629: at $(x,y) \in C$. 

1630: Similarly, if $b(x)$ vanishes to order  $l$ along  $p(C)$, we set 

1631: $$ [b] (Z) = \sum_{|\al|=l} \frac{1}{\al ! \be !} \Bigl( 

1632: (p_*\partial_1)^{\al(1)}...(p_* \partial_l)^{\al(l)} b(x) \Bigr) \ Z^{\al}.$$

1633: at $x \in p(C)$. 

1634: \begin{theo} \label{theo:PS2}

1635: Under the assumptions \eqref{H1} and \eqref{H2}, if $a$ vanishes to

1636:   order  $m$ along  $C$, then for every  $i \leqslant \frac{m}{2}$ the

1637:   function  $b_i$ vanishes to order  $m-2i$ along  $p(C)$. Furthermore  

1638: $$ [b_i](Z) = \frac{1}{i!} \De^i A_{2i}(Z,Y) $$

1639: at $(x,y) \in C$, where 

1640: \begin{itemize} 

1641: \item $[a](Z,Y) = \sum_{l=0}^{m} A_l(Z,Y)$ and $A_l$ is homogeneous of 

1642:  degree $l$ in $Y$ and of degree  $m-l$ in $Z$. 

1643: \item  $\De = \tfrac{i}{2} \textstyle{\sum}_{j,k}

1644:   \ph^{j,k}(x,y) \partial_{Y^j} \partial_{Y^k}$, with  $(\ph^{j,k}(x,y))$

1645:   the inverse of $ (\ph_{j,k}(x,y))$.

1646: \end{itemize} 

1647: \end{theo} 

1648: 

1649: More intrinsically, denote by $\ideal ^m(C) \subset \Ci( W)$ the ideal

1650: of functions vanishing to order  $l$ along  $C$. Then $\ideal ^m (C)/

1651: \ideal ^{m-1} (C)$ is isomorphic to the space of sections of the $m$-th symmetric power of the

1652: complex conormal bundle $\no^* (C)$. By \eqref{H1}, we have an

1653: isomorphism of vector bundle over $C$, 

1654: $$  \no^* (C) \simeq \no ^* ( p (C))  \oplus E^* ,$$

1655: which associates  $\ga$ and $\al \oplus \be$ if 

1656: $$ \langle \ga, U +

1657: V \rangle = \langle \al, p_* U \rangle + \langle \beta, V \rangle,

1658: \quad \text{$U \in F$ and $V \in  E$} $$

1659: Consequently,  

1660: \begin{gather*} %\label{dec_sym} 

1661:  \Sym_m (  \no^* (C) ) = \bigoplus_{l=0}^{m} \Sym_{m-l} \bigl( \no^* (p(C) )  \bigr) \otimes \Sym_{l} ( E^* ) \end{gather*} 

1662: $ \Delta = \tfrac{i}{2} \textstyle{\sum}_{j,k} \ph^{j,k}

1663: \partial_{y^j} \partial_{y^k}$ defines a section of $\Sym_{2}(E)$, so

1664: $\Delta ^i$ acts as an operator 

1665: $$\Sym_{2i} (

1666: E^* ) \rightarrow \C  .$$ In theorem  \ref{theo:PS2}, we consider

1667: $[a]$ as a section of $\Sym_m (  \no^*

1668: (C) )$ and  $[b_i]$ as a section of  $\Sym_{m-2i} ( \no^* (p(C)))$ . Then we have  

1669: $$ [b_i] = \frac{1}{i!} (\identite \otimes \De^i) A_{2i} $$

1670: where 

1671: $$ [a] = \sum_{l=0}^{m} A_l, \quad A_l \in \Ci (C,\Sym_{m-l} \bigl( \no^*

1672: (p(C) )  \bigr) \otimes \Sym_{l} ( E^* )) .$$

1673: 

1674: 

1675: 

1676: \bibliography{biblio}

1677: 

1678: 

1679: \end{document}

1680: 

1681: 

1682: 

1683: 

1684: %%% Local Variables: 

1685: %%% mode: latex

1686: %%% TeX-master: t

1687: %%% End: 

1688: