1: \documentclass[11pt]{article}

2: \setlength{\textwidth}{14cm}

3: \setlength{\oddsidemargin}{1cm}

4: 

5: \usepackage[dvips]{graphicx,color}

6: \usepackage{amssymb,latexsym,epsfig,epsf,xypic,amsmath}

7: \usepackage[mathscr]{euscript}

8: 

9: \makeatletter \@addtoreset{equation}{section} \makeatother

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

11: 

12: \newtheorem{Lemma}[equation]{Lemma}

13: \newtheorem{Theorem}[equation]{Theorem}

14: \newtheorem{Prop}[equation]{Proposition}

15: \newtheorem{Definition}[equation]{Definition}

16: \newtheorem{Corollary}[equation]{Corollary}

17: \newtheorem{Conj}[equation]{Conjecture}

18: \newenvironment{Proof}{\noindent\emph{Proof\ }}{\hfill$\square$\\}

19: \newenvironment{Remarks}{\noindent\textbf{Remarks}\ }{\\}

20: \newenvironment{Remark}{\noindent\textbf{Remark}\ }{\\}

21: 

22: \newcommand{\rt}[1]{\stackrel{#1\,}{\rightarrow}}

23: \newcommand{\Rt}[1]{\stackrel{#1\,}{\longrightarrow}}

24: \newcommand\A{\mathcal A}

25: \newcommand\db{\bar\partial}

26: \newcommand\R{\mathbb R}

27: \newcommand\C{\mathbb C\,}

28: \newcommand\Z{\mathbb Z}

29: \newcommand\Pee{\mathbb P}

30: \newcommand\PD{\mathrm{PD}\,}

31: \newcommand\OO{\mathscr O}

32: \newcommand\I{\mathcal I}

33: \newcommand\To{\longrightarrow}

34: \newcommand\ot{\leftarrow}

35: \newcommand\Ot{\longleftarrow}

36: \newcommand\into{\hookrightarrow}

37: \newcommand\res{\arrowvert_}

38: \newcommand\End{\mathrm{End}\,}

39: \newcommand\ip{{\mbox\,\_\hspace{-1.5mm}\shortmid\hspace{1.5pt}}}

40: \newcommand\ImO{\,\mathrm{Im}\,\Omega}

41: \newcommand\Jd{J\,\widetilde{\!d\theta\,}}

42: \newcommand\Jdh{J\,\widetilde{\!dh\,}}

43: \DeclareMathOperator\MCV{MCV}

44: \DeclareMathOperator\id{id}

45: \DeclareMathOperator\tr{tr}

46: \DeclareMathOperator\vol{vol}

47: \DeclareMathOperator\ind{ind}

48: 

49: \title{\textbf{Special Lagrangians, stable bundles and mean

50: curvature flow}}

51: \author{R. P. Thomas and S.-T. Yau}

52: \date{}

53: 

54: \begin{document}

55: 

56: \maketitle

57: \begin{abstract} \noindent

58: We make a conjecture about mean curvature flow of Lagrangian

59: submanifolds of Calabi-Yau manifolds, expanding on that of \cite{Th}.

60: We give new results about the stability condition in \cite{Th}, and

61: propose a Jordan-H\"older-type decomposition of (special) Lagrangians.

62: The main results are the uniqueness of

63: special Lagrangians in hamiltonian deformation classes of

64: Lagrangians, under mild conditions, and a proof of the conjecture

65: in some cases with symmetry: mean curvature flow converging to

66: Shapere-Vafa's examples of SLags.

67: \end{abstract}

68: 

69: 

70: \section{Introduction}

71: 

72: Fix a Calabi-Yau manifold $X$ with a holomorphic $(n,0)$-form

73: $\Omega$. In \cite{Th} a stability condition for Lagrangians in $X$

74: was described,

75: conjectured to be equivalent to the existence of a special Lagrangian

76: (SLag) in the hamiltonian deformation class of a fixed Lagrangian.

77: This was motivated by an infinite dimensional set-up

78: in which $U(1)$ gauge transformations act on the (infinite

79: dimensional) space of Lagrangians

80: with flat $U(1)$ connections on them. There is a natural complex

81: structure and symplectic form on this space and, ignoring issues

82: of integrability of these structures (see \cite{Th}),

83: the formal complexification of the $U(1)$ gauge transformations

84: gives hamiltonian deformations of the Lagrangian, with moment map

85: the $n$-form $\ImO\res L$. The stability condition was

86: motivated by an example of Joyce and the `angle criterion', in terms

87: of splittings of the Lagrangian into Seidel's graded Lagrangian

88: connect sums (as defined in Section \ref{index} below) and family

89: versions thereof, with a

90: certain phase inequality. This led to a conjecture, a sort of

91: globalised version of the angle criterion \cite{L}, \cite{N},

92: that the hamiltonian

93: deformation class of a Lagrangian should contain a SLag if and only

94: if the Lagrangian is stable; this SLag representative should then be

95: unique. Here

96: we expand on the conjecture and relate it to mean curvature flow. It

97: was verified for the simplest case $T^2$ in \cite{Th}; here we prove

98: it in a series of $n$-dimensional examples with symmetry (Theorem

99: \ref{faq}), and prove uniqueness of smooth SLags in hamiltonian deformation

100: classes whose Floer cohomology \cite{FO3} is defined (Theorem \ref{!}).

101: 

102: We write $\approx$ for ``in the same hamiltonian

103: deformation class as'', and use $\,\widetilde{\ }\,$ for the

104: isomorphism $T^*L\to TL$ induced by the metric on a Riemannian

105: manifold $L$. Restricting the Ricci-flat metric on a Calabi-Yau

106: manifold $(X,\Omega)$ to a Lagrangian submanifold $L$ we get an induced

107: volume form $\vol$ on $L$, and by a short calculation

108: \begin{equation} \label{vol}

109: \Omega\res L=e^{i\theta}\vol

110: \end{equation}

111: defines an $S^1$-valued function $\theta$ on $L$, the

112: \emph{phase function of} $L$. A \emph{grading} of $L$ is a lift of

113: $\theta$ to a real valued function. By Lagrangian we will always mean

114: graded Lagrangian (thus the Maslov class of the Lagrangian, which

115: is the class of $d\theta$ in $H^1(L;2\pi\mathbb{Z})$, is assumed to vanish,

116: and we have chosen a lift of $\theta$). $L$ is special Lagrangian

117: (SLag) if $\theta$ is a constant; equivalently, replacing $\Omega$

118: by $e^{-i\theta}\Omega$, $\ImO\res L\equiv0$.

119: An average, cohomological, measure of the phase of a homology class

120: $[L]$ is given by taking the phase of the complex number $\int_L\Omega$;

121: for $L$ graded with the variation of $\theta$ less than $2\pi$, this lifts

122: naturally to give a real number

123: $\phi(L)$, which is the phase of any SLag in the same homology class.

124: 

125: We should point out that as in \cite{Th}, we do not fully understand the

126: role of holomorphic discs in the theory. These are of course crucial in

127: the definition and hamiltonian deformation invariance of Floer

128: cohomology; until this is fully set up \cite{FO3} and all of its

129: expected properties (such as the spectral sequences of \cite{Oh2} and

130: \cite{P}) are proved and extended to the Calabi-Yau case,

131: some of the arguments below are necessarily conjectural; it will be

132: clear which ones. We also

133: deal exclusively with smooth (S)Lags; how to modify our constructions

134: to include singularities is an important question. Using only

135: (family) Lagrangian connect sums as the degenerations necessary

136: to describe stability of Lagrangians is also probably too restrictive,

137: studying other singularities and splittings may also be necessary; the

138: conjecture in this paper is probably just the first step in understanding

139: SLags in hamiltonian isotopy classes.

140: 

141: \noindent \textbf{Acknowledgements.} The symplectic ideas and

142: suggestions of Paul Seidel have been absolutely invaluable throughout

143: this work. We have also benefitted from comments from Kenji Fukaya,

144: Edward Goldstein, Spiro Karigiannis, Conan Leung, Jun Li, Elizabeth

145: Mann, Yong-Geun Oh and Xiao Wei Wang, and would like to thank Mike

146: Gage for the reference [An]. The first author is supported by a

147: Royal Society university research fellowship and by Imperial College,

148: London; the second author is supported by DOE grant DE-FG02-88ER35065

149: and NSF grant DMS-9803347.

150: 

151: 

152: \section{Mean curvature flow}

153: 

154: We first give a well-known geometric calculation which

155: we learnt from unpublished lectures of Rick Schoen on his work with

156: Jon Wolfson, but which dates back at least as far as \cite {HaL},

157: \cite{Oh1} and others.

158: 

159: \begin{Lemma} \label{mcf}

160: In the above notation, the mean curvature vector of the Lagrangian

161: $L\subset X$ is $MCV=\Jd$.

162: \end{Lemma}

163: 

164: \begin{Proof}

165: We want to show that for any vector $X$ tangent to $L$,

166: $X\theta=-\langle\MCV,JX\rangle$.

167: 

168: Picking an orthonormal basis of $T_pL$ and parallel transporting it

169: along rays in $L$ to a frame field $(e_i)$, $(e_i,Je_i)$ forms a local

170: basis for $TX$ around $p$. Letting $(f_j,g_j=-f_j\circ J)$ be the dual

171: basis of 1-forms, it is clear that at $p$,

172: $$

173: \Omega=e^{-i\theta}\bigwedge_j(f_j+ig_j),

174: $$

175: with $\theta$ the phase function of $L$. Since $\Omega$ is parallel,

176: $\nabla_X\Omega=0$ yields

177: \begin{eqnarray}

178: iX(\theta)\bigwedge_j(f_j+ig_j)&=&\sum_k(f_1+ig_1)\wedge\ldots\wedge

179: \nabla_X(f_k+ig_k)\wedge\ldots\wedge(f_n+ig_n) \nonumber \\

180: &=&\sum_k\left[\nabla_X(f_k+ig_k)\left({1\over2}(e_k-iJe_k)\right)\right]

181: \bigwedge_j(f_j+ig_j). \label{formula}

182: \end{eqnarray}

183: 

184: Taking covariant derivatives on the Calabi-Yau (i.e. not on $L$), we have

185: $$

186: -\langle\MCV,JX\rangle=-\langle\sum_i\nabla_{e_i}e_i,JX\rangle=

187: \sum_i\langle\nabla_{e_i}Je_i,X\rangle,

188: $$

189: since $J$ is both skew adjoint and parallel. But as $Je_i$ and

190: $X$ are orthogonal, this is

191: $$

192: -\sum_i\langle Je_i,\nabla_{e_i}X\rangle=\sum_i\langle e_i,J\nabla_Xe_i

193: \rangle,

194: $$

195: as we may choose $X$ to have zero bracket with the $e_i$s.

196: 

197: So comparing with (\ref{formula}) we are left with showing that

198: $$

199: \sum_k\left[\nabla_X(f_k+ig_k)\left({1\over2}(e_k-iJe_k)\right)\right]

200: =i\sum_i\langle e_i,J\nabla_Xe_i\rangle,

201: $$

202: i.e. that $(\nabla_X(f_k+ig_k))(e_k-iJe_k)=2i\langle e_k,J\nabla_Xe_k

203: \rangle$.

204: 

205: But $(f_k+ig_k)(e_k-iJe_k)=2$ is a constant, so the left hand side

206: is $-(f_k+ig_k)(\nabla_X(e_k-iJe_k))=-f_k(\nabla_X(-iJe_k))-

207: ig_k(\nabla_Xe_k)$; the other terms vanish as $\nabla_Xe_i$ was

208: chosen to be perpendicular to $L$. Using $\nabla_XJ=0$ and

209: recalling that $g_k=-f_k\circ J$, we obtain $2i\langle e_k,

210: J\nabla_Xe_k\rangle$.

211: \end{Proof}

212: 

213: Another simple but important result is how the phase $\theta$ and

214: volume form $\vol_L$ vary under a hamiltonian deformation

215: $J\,\widetilde{\!dh\,}$ of $L$. Such calculations appear in various forms

216: in \cite{Oh1}, \cite{Sm}, for instance; we give short geometric proofs for

217: completeness.

218: 

219: \begin{Lemma}

220: Under a hamiltonian deformation $\Jdh$ of a

221: Lagrangian $L$, we have

222: \begin{eqnarray}

223: {d\over dt}\theta\!&=&\!-\Delta_L(h), \label{h} \\ \nonumber

224: {d\over dt}\vol_L\!&=&\!-\langle d\theta,dh\rangle\vol_L.

225: \end{eqnarray}

226: \end{Lemma}

227: 

228: \begin{Proof}

229: Take real and imaginary parts of $e^{-i\theta}$

230: times the following:

231: \begin{eqnarray*}

232: i\dot\theta e^{i\theta}\vol_L\!\!\!&+&\!\!\!e^{i\theta}{d\over dt}\vol_L

233: ={d\over dt}(e^{i\theta}\vol_L)=(\mathcal L_{\Jdh}\Omega)\res L=

234: d(\Jdh\ip\Omega)\res L

235: \\ &=&\!\! id(e^{i\theta}\,\widetilde{\!dh\,}\ip\vol_L)=-e^{i\theta}

236: d\theta\wedge(\,\widetilde{\!dh\,}\ip\vol_L)-ie^{i\theta}d^*dh\vol_L.

237: \end{eqnarray*}

238: Setting $\Delta_L=d^*d$ $(=-\sum_i\partial^2_{x_i}$ in geodesic

239: coordinates) gives the result.

240: \end{Proof}

241: 

242: We next show that, given a suitable metric on the Lie

243: algebra $C^\infty(L,\R)/\R$, the gradient flow of the norm square

244: $-|m|^2$ of the moment map $m=\ImO\res L$ of \cite{Th} is mean

245: curvature flow. The following standard calculation, applicable in any

246: K\"ahler reduction picture, shows that the gradient flow of

247: $-|m|^2$ is given by $JX_{m^*}$, where $J$ is the complex structure,

248: $m^*$ is the element of the Lie algebra $C^\infty(L)$ corresponding to

249: the moment map $m=\ImO\res L$ in the dual of the Lie algebra under the

250: metric on $C^\infty(L)$, and $X_{m^*}$ is its induced action on the

251: space \{Lagrangians with flat $U(1)$ connections on them\}.

252: $$

253: X(-|m|^2)=-X(m(m^*))=-\omega(X,X_{m^*})=\langle X,JX_{m^*}\rangle.

254: $$

255: By the definition of the group action in \cite{Th}, this

256: deformation $JX_{m^*}$ is just the hamiltonian deformation of the

257: Lagrangian $L$ with hamiltonian function $m^*$ on $L$.

258: 

259: Choosing the volume form Re\,$\Omega\res L$ on $L$ to define an $L^2$

260: metric on $C^\infty(L)$ gives $m^*=\tan\theta$, since

261: $m=\ImO\res L=\sin\theta\,$vol$\,=\tan\theta\,$Re\,$\Omega\res L$.

262: Similarly using the induced Riemannian volume form vol gives

263: $m^*=\sin\theta$, while using

264: $$

265: {\sin\theta\over\theta}\,\mathrm{vol}

266: $$

267: as volume form on $L$ yields $m^*=\theta$.

268: Any of these are suitable for small phase

269: $\theta:\,L\to\R$, and give similar flows down which the moment map

270: decreases. The last one, however, is precisely mean curvature flow, by

271: Lemma \ref{mcf}.

272: 

273: This and the previous lemma show that under mean curvature flow,

274: the phase $\theta$ satisfies a (time dependent) heat equation while

275: the Riemannian volume form decreases (as usual):

276: \begin{eqnarray}

277: \dot\theta\!&=&\!-\Delta\,\theta, \label{thetadot} \\

278: {d\over dt}\vol_L\!&=&\!-|d\theta|^2\vol_L. \label{voldot}

279: \end{eqnarray}

280: We therefore obtain a maximum principle for $\theta$, whose range must

281: always decrease, but it is important to note that the Laplacian $\Delta$

282: is time dependent as the metric on $L$ used to define it varies.

283: 

284: From these follow a series of identities and estimates, many of which

285: we use later, but none are strong enough to

286: give long term existence of the mean curvature flow, and with good

287: reason. Mean curvature flow is a complicated and much-studied

288: subject (understood only in codimension 1, dimension 1 \cite{Gr}, and,

289: in special cases, in

290: two dimensions \cite{Wa}), with known examples of finite

291: time blow-up. While we might expect it to behave better for Lagrangians

292: (locally functions of one variable instead of $n$), examples in

293: Section \ref{eg} show similar phenomena. But in our examples

294: there will be a way round these problems, and we will be able to make

295: a conjecture about the flow which may help in its study.

296: 

297: 

298: \section{Connect sums and Floer gradings} \label{index}

299: 

300: The stability definition in \cite{Th} made extensive use of

301: graded Lagrangian connect sums \cite{S2}; a description

302: of these and their relationship to Floer cohomology will be

303: important again here, as will knowledge of the Floer index of

304: Lagrangian intersections. We fix our conventions and definitions

305: now; in some places these differ in orientation from some of

306: the mirror symmetry literature and \cite{S2};

307: the problem seems to be deciding on whether to use the standard

308: symplectic form $dxdy$ on $T^2$, or the equally standard

309: $dpdq=-dxdy$ considering it as the cotangent bundle of its

310: SYZ base $S^1$ (divided by a lattice) \cite{SYZ}.

311: 

312: \subsection{The connect sum}

313: 

314: Suppose we have two Lagrangians $L_1,\,L_2$ hamiltonian isotoped

315: to intersect transversally in a finite number of points. We will

316: work at one of these points $p$. There we can pick a local

317: Darboux chart with coordinates $(x_i,y_i)$ and symplectic

318: form $\sum_idx_i\wedge dy_i$ such that $L_2=\{y_i=0\}$ is the

319: $x$-axes, and

320: \begin{equation} \label{def}

321: L_1=\{y_i=\tan(\alpha)x_i\}

322: \end{equation}

323: for some $\alpha\in(0,\pi)$. (It would be more usual to use

324: $\alpha=\pi/2$, of course, but that situation can be moved to

325: this one by an obvious symplectic (shear) transformation).

326: 

327: Using $z_i=x_i+iy_i$ coordinates to set up the obvious isomorphism

328: to $\C^n$ (notice that this complex structure and that inherited from

329: $X$ may be different), the $L_i$ are

330: $$

331: L=e^{i\alpha}[0,\infty).S^{n-1}:=\{z_j=re^{i\alpha}a_j\,:\,r\in

332: [0,\infty),\,\mathbf a=(a_j)_{j=1}^n\in S^{n-1}\subset\R^n\subset\C^n\},

333: $$

334: where $\alpha$ is set to zero to give $L_2$.

335: 

336: So given a curve $\gamma$ in $\C$, we define a Lagrangian

337: $$

338: L_\gamma=\gamma.S^{n-1}=\{z_j=\gamma a_j\,:\,\mathbf a=(a_j)_{j=1}^n

339: \in S^{n-1}\subset\R^n\subset\C^n\}.

340: $$

341: Then $L_2$ is represented by $\gamma_2=[0,\infty)\subset\C$,

342: $L_1$ by $\gamma_1=e^{i\alpha}[0,\infty)\subset\C$, and $L_1\cup L_2$

343: by the V-shaped union of these curves.

344: 

345: In this notation the \emph{Lagrangian connect sum} $L_1\#L_2$ is

346: represented by any smoothing $\gamma=:\gamma_1\#\gamma_2$

347: of $\gamma_1\cup\gamma_2$ staying inside the cone

348: $\{re^{i\beta}\,:\,r>0,\,\beta\in[0,\alpha]\}$

349: which is $\gamma_1\cup\gamma_2$ outside a compact set, and a

350: smooth curve cutting off the cone at the origin. (So here

351: $\gamma$ is \emph{not} a connect sum of the curves $\gamma_i$

352: in the topological sense; we only use the notation because the

353: resulting Lagrangians are topological connect sums.)

354: 

355: We want to analyse the phase of such a connect sum; initially

356: in the complex structure we picked on $T_pX\cong\C^n$ using the

357: $x_i+iy_i$ coordinates and so, up to scale, $\Omega\res p=

358: dz_1\ldots dz_n$. Then the phase function of the Lagrangian

359: $L_\gamma$ associated to a curve $\gamma$ is easily calculated to

360: be $\theta(\gamma')+(n-1)\theta(\gamma)+N\pi$ for any $N\in\Z$

361: (where $\theta(z)\in(-\pi,\pi]$ is the phase of a

362: complex number $z=re^{i\theta(z)}$). Orienting $\gamma_2$

363: such that $\gamma_2'$ is a positive real number, and choosing

364: $L_2$ to have phase 0, corresponds to choosing $N=0$ and so grading

365: $L_1$ by

366: \begin{equation} \label{pa}

367: (-\pi+\alpha)+(n-1)\alpha=n\alpha-\pi.

368: \end{equation}

369: In particular, choosing $\alpha=\pi/n$, and setting, for any $c>0$,

370: $$

371: \gamma^{\ }_c=

372: \{re^{i\theta}\,:\,r^n=c\sin(n\theta),\ \theta\in(0,\pi/n)\}

373: $$

374: gives a SLag $L_\gamma$ which has a grading of phase identically

375: zero, asymptotic

376: to $L_1$ and $L_2$ at infinity, and this is precisely the local

377: model of the example of Joyce, Harvey and Lawlor used so

378: extensively in \cite{Th}. 

379: 

380: While this is not strictly of the form $\gamma_1\#\gamma_2$ as defined

381: above (it is only asymptotic to the $\gamma_i$, not equal outside

382: a compact set), by taking $c$ small we can make it as close as we like

383: to such a connect sum, all in the same hamiltonian deformation class,

384: and the construction of Joyce is indeed a hamiltonian deformation of a

385: connect sum as claimed in \cite{Th}.

386: 

387: We plot these SLag curves $\gamma_c\subset\C$ in Figure 1

388: as the light lines, converging as $c\to0$ to the V-shaped

389: $\gamma_1\cup\gamma_2$ (with $\alpha=\pi/n$). Then the dark lines

390: depict connect sums $L_1\#L_2$ for $\phi(L_2)=0$ and $\phi(L_1)=

391: \pm\epsilon$. If $\phi(L_1)<0$, the \emph{stable case} as described

392: in \cite{Th}, then we can choose the connect sum

393: such that the phase of $L_1\#L_2$ varies monotonically between its

394: values on $L_1$ and $L_2$, i.e. between $-\epsilon$ and 0.

395: 

396: \begin{figure}[h]

397: \center{

398: \input{hash.pstex_t}

399: \caption{$\gamma_1\#\gamma_2$, and the resulting phase function

400: $\theta_{L_1\#L_2}$, for $\phi(L_1)=\pm\epsilon$}}

401: \end{figure}

402: 

403: If, however, $\phi(L_1)>0$, the \emph{unstable case} in \cite{Th},

404: then the phase of $L_1\#L_2$ must initially decrease to move away

405: from $L_2$ before increasing to reach $L_1$ (i.e. $\gamma$ must cross

406: the light lines one way then the other), giving a phase

407: function which necessarily goes outside the range $(0,\epsilon)$

408: (see Figure 1).

409: This will be important to us later -- under mean curvature flow

410: we expect the phase function $\theta$ to evolve to a constant

411: in the stable case (under the heat equation (\ref{thetadot})) and

412: to a Heaviside step function (with values $0$ and $\epsilon$)

413: in the unstable case. This does not then contradict the maximum

414: principle as the unstable case has the described non-monotonic phase.

415: 

416: While this defines the symplectic connect sum in general by means

417: of our Darboux chart, the analysis of phases depended on the choice

418: of complex structure on $T_pX$, which may not have been the one

419: restricted from the Calabi-Yau $X$. In the general case we can fix

420: $\theta_p(L_2)=0$, without loss of generality, by rotating $\Omega$,

421: and pick local complex

422: coordinates $z_i=x_i+iy_i$ such that $\Omega_p=dz_1\ldots dz_n$ and,

423: at the level of tangent spaces at $p$, (the tangent space to) $L_2$

424: is at $y_i=0\ \forall i$. (The tangent space to) $L_1$ will be of the

425: form

426: \begin{equation} \label{l1}

427: L_1=\{z_i=re^{i\alpha_i}\},

428: \end{equation}

429: for some $\alpha_i$s that are \emph{no longer necessarily all the same}

430: in these coordinates. We are now connect summing Lagrangians of pointwise

431: phase $0$ and $\sum_{i=1}^n(\alpha_i)-\pi$ (compare (\ref{pa})), but

432: the resulting phase function will not be as simple as before -- it

433: is not pulled back from $\gamma$ but will vary over the $S^{n-1}$

434: fibres. Its average phase over the $S^{n-1}$s will have a similar

435: form to that in Figure 1, however, and in the case of all the

436: $\alpha_i$s being the same we get the earlier simpler picture.

437: 

438: The dependence of the hamiltonian deformation class of

439: $L_1\#L_2$ on the choice of scale of the neck at each intersection point

440: was described in (\cite{Th} Section 4) (in particular if there is

441: only one intersection point the class is uniquely defined). We should

442: also point out that the graded connect sum (when it exists)

443: is also independent of hamiltonian deformations of $L_1$ and $L_2$. While

444: the $L_i$s intersect transversely this is clear; we need only understand

445: what happens in crossing the codimension one wall of Lagrangians

446: intersecting in a double point (i.e. creating or cancelling two

447: intersection points). But it will be clear from the definition of grading

448: below that two such points must have grading differing by one, and

449: so the connect sum along both of them cannot be graded (\ref{grhash}).

450: 

451: 

452: \subsection{The grading on Floer cohomology}

453: 

454: The Floer cohomology group $HF^*(L_2,L_1;\C)$ \cite{FO3}

455: is the cohomology of a cochain complex made from a copy of $\C$

456: for each intersection point of two \emph{graded} Lagrangians

457: hamiltonian isotoped to intersect transversally.

458: The differential is defined by counting holomorphic

459: strips, with boundary in the Lagrangians, running from one intersection

460: point to another. It is a symplectic refinement of the

461: topological intersection theory of $L_1,\,L_2$, and as such is invariant

462: only under hamiltonian deformations of the $L_i$.

463: What is important to us is the grading of a particular transverse

464: intersection point, as defined in \cite{S2}, \cite{FO3}.

465: 

466: While this can be defined completely topologically, it is most easily

467: (and equivalently) defined via a complex structure. Again we

468: work at the level of tangent spaces, pick local coordinates and,

469: without loss of generality, take $L_2$ to have phase 0 and to be

470: the $x$-axes: $L_2=\{y_i=0\}$. Write $L_1$ as

471: $$

472: L_1=\{z_i=re^{i\alpha_i}\},

473: $$

474: \emph{where the $\alpha_i$s are all in} $(0,\pi)$. Then

475: $\sum\alpha_i=\theta_p(L_1)$ mod $\pi$ is the phase of $L_1$ up to

476: multiples of $\pi$, and the following integer

477: is the \emph{definition} of the Floer index of the point $p$:

478: \begin{equation} \label{maslov}

479: \ind_p(L_2,L_1):={1\over\pi}\left(\sum\alpha_i-\theta_p(L_1)\right).

480: \end{equation}

481: Notice therefore that $\ind_p(L_2,L_1)+\ind_p(L_1,L_2)=n$.

482: Applying the definition (\ref{maslov}) to the connect sums defined

483: in the last section (for which $\theta_p(L_1)=\sum\alpha_i-\pi$),

484: we recover a result of Seidel \cite{S2}:

485: \begin{equation} \label{grhash}

486: L_1\#L_2 \text{\emph{ exists as a graded connect sum if and only if }}

487: \ind_p(L_2,L_1)=1.

488: \end{equation}

489: (The only if

490: part follows from the independence of gradings and the Floer index

491: from the complex structure; we may therefore pick the complex

492: structure locally to have the form of the local model above.)

493: Given $L_1$ there is at most one choice of the grading on $L_2$

494: such that $\ind_p(L_2,L_1)=1$ at all intersection points $p$, so that

495: $L_1\#L_2$ can be graded.

496: 

497: In fact connect sums $L_1\#L_2$ whose own Floer cohomology is well

498: defined \cite{FO3} should correspond to Floer coclosed cochains, i.e.

499: elements of $HF^1(L_2,L_1)$,

500: mirror to the fact that extensions of sheaves $0\to E_1\to E\to E_2\to0$

501: correspond to elements of Ext$^1(E_2,E_1)$, as discussed in \cite{Th}.

502: 

503: We can also deal with the connect sums mentioned in \cite{Th}

504: which are \emph{relative} versions of the above construction;

505: $(n-r)$-dimensional connect sums carried out in a smooth family

506: over an $r$-dimensional base. Then the same Floer

507: index can be defined; there are now $r$ angles between the Lagrangians

508: that are zero, and $(n-r)$ whose signs can be computed to get the Floer

509: index. (The signs are constant over the family since the intersection

510: of the Lagrangians $L_1\cap L_2$ fibres over the base of the family

511: with fibres of constant dimension; an angle going to zero would cause a

512: fibre dimension to increase.)

513: 

514: 

515: \section{Uniqueness}

516: 

517: In finite dimensional symplectic quotient problems, convexity

518: properties of the moment map prove uniqueness of its zeros (modulo

519: the action of the real group) in a complexified group orbit.

520: Translating this into our terms is not quite possible, because there

521: are hamiltonian deformations of $L$ which are not given by the flow

522: of a \emph{fixed} hamiltonian on $L$. By this we mean $L_0,\ L_1$ are

523: deformations given by a constant hamiltonian $h\in C^\infty(L_0;\R)$

524: if the flow

525: \begin{equation} \label{flow}

526: f_t:\,L\to W, \quad {df\over dt}=\Jdh, \qquad t\in[0,1],

527: \end{equation}

528: takes $L_0=f_0(L)$ to $L_1=f_1(L)$.

529: All small deformations of a Lagrangian are of this form; for more

530: general deformations we have to use a

531: different proof of uniqueness of a SLag representative of a

532: hamiltonian deformation class (Proposition \ref{!} below), but for

533: these constant hamiltonian deformations

534: we describe the moment map proof to show how the formalism works.

535: 

536: \begin{Lemma} \label{?}

537: If two SLags $L_0,\ L_1$ are time-independent hamiltonian deformations

538: of each other, in the sense above, then $L_0=L_1$.

539: \end{Lemma}

540: 

541: \begin{Proof}

542: Without loss of generality we may take $\phi(L_0)=\phi(L_1)=0$. Then

543: we compute, down the flow (\ref{flow}),

544: $$

545: {d\over dt}\int_L h\ImO=\int h\,\mathcal L_{\Jdh}

546: \ImO=\int h\,d(\Jdh\ip\ImO)=\int\cos\theta\,dh\wedge*dh,

547: $$

548: where the last identity (equation (3.2) of \cite{Th}) is an easy

549: computation in local coordinates.

550: (We have abused notation and written $\ImO$ for $f_t^*\ImO$.)

551: 

552: So for $\theta$ lying in $(-\pi/2,\pi/2)$ this is always strictly

553: positive, but $\int_L h\ImO$ is zero at $t=0,\,1$. Thus

554: the two SLags must in fact coincide.

555: 

556: However, we must show that $\theta$ stays in this range if it starts

557: in it, and deal with the case when it is not so bounded. The way

558: to do this in fact proves the whole Lemma in one go anyway: pick a

559: maximum $x\in L$ of $h$. Then by (\ref{h}) $\dot\theta(x)\le0

560: \ \ \forall t$, but under the flow $\theta$ starts and ends at the

561: same value (the \emph{cohomologically} determined phase $\phi$ of the

562: SLags). So $\Delta h=0$ and all the second derivatives $h_{xx}\le0$

563: in any direction $\partial_x$ vanish. Similarly then all third

564: derivatives of $h$ must vanish (since we are at a maximum). Apply

565: the same procedure to the second derivatives $h_{xx}$ of $h$:

566: $\dot\theta_{xx}=-\Delta_Lh_{xx}\le0$ at $x$, since the other terms

567: $[\partial^2_x,\Delta_L]\,h$ involve derivatives of the metric times

568: third and lower order derivatives of $h$. Thus $h$'s 4th order

569: derivatives vanish, and so on.

570: 

571: To get an integral form of this, to show that

572: $h$ is in fact constant, it is enough to show that $h$ is constant in

573: a small ball around any global maximum $x$ (with $h(x)=0$, without loss

574: of generality). Consider geodesic balls $B_r$ of radius $r$ about $x$,

575: and their boundary spheres $S_r$. Fix a standard

576: unit-volume $(n-1)$-form $d\mu$ on the spheres $S_r$, so that the

577: volume form induced by the metric is $\partial_r\ip\vol=A(r)d\mu$,

578: where $r^{1-n}A(r)=c_1+O(r^2)$. For $r$ sufficiently small,

579: $d/dr\{r^{1-n}A(r)\}$ is

580: bounded by $c_2r$ (for some $c_2$ dependent only on the maximum of the

581: curvatures of $L$ in a neighbourhood of $x$ over time $t\in[0,T]$

582: of the flow). Therefore

583: $$

584: {d\over dr}\left\{{\int_{S_r}h\,A(r)d\mu\over r^{n-1}}\right\}=

585: r^{1-n}\int_{S_r}h_r\,A(r)d\mu+e,

586: $$

587: where $|e|\le c_3r^{2-n}\left|\int_{S_r}h\,A(r)d\mu\right|$.

588: Integrating over $r\in(0,R)$

589: for $R$ sufficiently small, and using the divergence theorem,

590: $$

591: f(R):=R^{1-n}\int_{S_R}h\,A(R)d\mu=

592: \int_0^Rr^{1-n}\left(\int_{B_r}\Delta h\vol\right)dr+E,

593: $$

594: where $|E|\le c_3\int_0^R|rf(r)|dr$.

595: Since $\Delta h=\dot\theta$, and $\int_0^T\dot\theta dt=0$

596: (where the hamiltonian deformation is over time $t\in[0,T]$,

597: and $\theta$ starts and ends at the same value), we see that

598: $$

599: \int_0^T|f(R)|dt\le c_3\int_0^T\int_0^R|rf(r)|dtdr

600: $$

601: for all small $R$. From this it follows that $f(r)\equiv0$.

602: That is, the average value of $h\le0$

603: over all small spheres surrounding $x$ (averaged over time

604: as the metric on $L$ varies) is zero, and $h$ must be identically

605: zero in a neighbourhhod of $x$.

606: \end{Proof}

607: 

608: However, we can do better by mirroring the algebro-geometric argument

609: that a non-zero map between stable bundles of the same slope is an

610: isomorphism, using the grading on Floer cohomology (\ref{maslov}).

611: This will appear to be slightly magical; the crux of the argument is

612: the hamiltonian isotopy invariance of Floer cohomology, provided

613: by precisely the holomorphic discs in the theory about which we

614: have had so little to say. Again we work in $n$ dimensions.

615: 

616: \begin{Theorem} \label{!}

617: Pick a connected graded Lagrangian $L$ whose obstructions

618: \emph{\cite{FO3}}

619: to the existence of its Floer cohomology vanish, and whose second

620: Stieffel-Whitney class $w_2$ is the restriction of a class

621: $\in H^2(X;\Z/2)$ on the whole manifold (for instance if $L$ is spin).

622: 

623: Then there can be at most one smooth special Lagrangian in the

624: hamiltonian deformation class of $L$.

625: 

626: In particular, SLag homology spheres are unique in their

627: hamiltonian deformation class in dimension 3 and above.

628: \end{Theorem}

629: 

630: \begin{Proof}

631: Since Floer cohomology is independent of hamiltonian deformations

632: \cite{FO3}, any two SLags $L_1,\,L_2$ in this same hamiltonian

633: deformation class satisfy

634: $$

635: HF^0(L_1,L_2)=H^0(L_1;\C)=\C,

636: $$

637: given that the zeroth order piece of $H^*(L)$ survives in $HF^*(L,L)$

638: for $L$ with Maslov class zero (\cite{FO3} Theorem E 1.7.4).

639: Thus there must be at least one intersection point $p$ of $L_1$ and $L_2$.

640: 

641: We first want to show that the (constant) phases of the $L_i$ are the same;

642: all we know a priori is that they differ by $r\pi$ for some $r\in\Z$.

643: Using a hamiltonian perturbation we may assume then that there is at least

644: one transverse intersection between the $L_i$s of Floer index 0, with the

645: phases of the $L_i$ at this point differing by $r\pi+\epsilon$. Thus writing,

646: locally, $L_2$ as the graph in $T^*L_1$ of $df$, $f$ has Morse index $r$

647: at the intersection point, so $r\ge0$. Similarly there is

648: a point of Floer index $n$ (i.e. a point of Floer index 0 when the roles

649: of $L_1$ and $L_2$ are reversed) which must correspond locally to $d$ of

650: a function of Morse index $n+r$; so $r\le0$ also. Therefore $r=0$ as

651: required.

652: 

653: So we have SLags $L_1,\,L_2$ of the same pointwise phase with at

654: least one intersection point which, if isolated, must have Floer index

655: (\ref{maslov}) zero. Thus $\theta_p=0$ in (\ref{maslov}), and the definition

656: (\ref{maslov}) of the index is therefore always positive (in fact between $0$ and

657: $n$), and zero only if the relative angles $\alpha_i=0\ \ \forall i$.

658: So the $L_i$ are tangent at $p$.

659: 

660: So there is no isolated transverse intersection point of Maslov class zero.

661: In fact, working in a small neighbourhood of the intersection, we may choose

662: coordinates such that $L_1$ is the graph in $T^*L_2$ of a closed one-form

663: $\sigma$ on $L_2$ which is also coclosed in a certain metric on $T^*L_2$

664: in a first order infinitesimal neighbourhood of $L_2$ (coclosedness is

665: the \emph{special} Lagrangian condition). In this small open set, write

666: $\sigma=df$, so that $d^*df=0$ and $f$ is harmonic; thus by the maximum

667: principle, it has no local maxima or minima. We want to show that $f$

668: is in fact constant. The Floer index (\ref{maslov})

669: of intersection points $df=0$ now reduces to the Morse index of $f$

670: at isolated critical points, but we also have to deal with degenerate

671: critical points of $f$. Assuming for a contradiction

672: that the critical set of $f$ is not all of $L_2$, we may perturb

673: $f$ inside any connected component of a small

674: neighbourhood of its critical set such that its value is unchanged on

675: the boundary, \emph{where it attains its global maximum and minimum},

676: and is Morse in the interior. (That we may take the extrema to be

677: on the boundary is the key point and a consequence of the maximum

678: principle.) We can then perturb $f$ further to

679: arrange its index 1 critical points to be lower (with respect to $f$)

680: than all higher index points (by general position arguments

681: \cite{Mi} Theorem 4.8) and then cancel any local minima with them

682: (\cite{Mi} Theorem 8.1). (There must be index 1 critical points if

683: there are any interior minima, by connectivity of our neighbourhood.)

684: 

685: The upshot is a hamiltonian perturbation of $L_1$, using this new

686: function, with no Floer index zero intersection points with $L_2$.

687: Thus $HF^0(L_1,L_2)=0$, a contradiction, so in fact

688: $f$ was locally constant and $L_1=L_2$.

689: 

690: The final statement follows from the fact that the obstructions of

691: \cite{FO3} live in $H^2(L)$, and homology spheres are spin.

692: \end{Proof}

693: 

694: As Donaldson pointed out, this proof is similar in flavour to proofs

695: of the Arnold conjecture. If the local situation (of all hamiltonian

696: deformations coming from a fixed function) held globally, the proof

697: would be `trivial', i.e. that of Lemma \ref{?} above. Even more

698: simply, if one SLag is a graph in the cotangent bundle of another,

699: we reduce the problem to the uniqueness of harmonic functions of

700: integral zero on $L$, i.e. to $H^0(L;\C)=\C$. To extend this

701: argument globally we need to replace de Rham cohomology $H^0(L;\C)$

702: by Floer cohomology $HF^0(L,L;\C)$.

703: 

704: 

705: \section{Analogues of some properties of sheaves}

706: 

707: In this section we discuss more

708: properties of (S)Lags that mirror those of holomorphic vector bundles

709: on Calabi-Yau manifolds.  As they rely heavily on Floer cohomology

710: arguments, many of the topics are necessarily treated informally and

711: unrigorously for now.

712: 

713: \subsection{Twisting by line bundles}

714: 

715: Any coherent sheaf can be twisted by a sufficiently positive

716: line bundle $\OO(N)$ so that it has sections; equivalently there are

717: homomorphisms to the bundle from any sufficiently negative line bundle.

718: If the sheaf has global support, this homomorphism is injective,

719: exhibiting $E$ as an extension

720: $$

721: 0\to\OO(-N)\to E\to Q\to0.

722: $$

723: One test of our notion of subobject of Lagrangians (in terms of connect

724: sums), then, is that there should be appropriate connect sums mirroring

725: this extension.

726: 

727: A line bundle $\mathcal L$ defines a \emph{spherical object} \cite{ST}

728: of the derived category of sheaves on a Calabi-Yau manifold $X$;

729: that is Ext$^i(\mathcal L,\mathcal L)=H^{0,i}(X)\cong H^*(S^n;\C)$

730: is $\C$ in dimensions $0$

731: and $n$, and zero otherwise. These should be mirror to Lagrangian

732: homology spheres; we will consider only spheres here so that we can

733: use the graded Dehn twists \cite{S2} around them. Negativity

734: compared to some other Lagrangian may not make sense in general

735: (intuitively, the Lagrangian

736: might be mirror not to a sheaf but to an object of the

737: derived category with Homs in negative degrees, etc.) but instead

738: we can consider only those spheres $L$ with only degree zero

739: intersection points (\ref{maslov}) with a fixed Lagrangian $L'$.

740: 

741: Then it is indeed true that we can exhibit $L$ as a subobject of

742: $L'$: denoting by $T_L$ the (graded) symplectic Dehn twist about

743: $L$, simply note that

744: $$

745: L'\approx T_L^{-1}T_LL'\approx L\#[L'\#(L[\,1\,])]

746: $$

747: expresses $L'$ as a connect sum of $L$ and something else. These relations

748: can be shown by grading similar results in \cite{S1}. In general this will

749: not destabilise $L'$ due to the phase of $L$ being so negative.

750: 

751: 

752: \subsection{Stability of (S)Lags} \label{ssl}

753: 

754: It is usual in correspondences between stable objects in algebraic

755: geometry and solutions of the corresponding moment map PDE for one

756: direction of the correspondence to be reasonably straightforward to

757: prove, namely that objects which satisfy the PDE are stable.

758: 

759: While we cannot prove this for SLags, we can show, for SLags

760: satisfying Floer cohomology restrictions as in Theorem \ref{!}

761: (in particular for spheres), that they cannot be destabilised

762: by other SLags. (To test for stability of sheaves it is sufficient

763: to test only with \emph{stable} subsheaves; if the conjecture of

764: \cite{Th} is true then similarly we could test for stability of

765: Lagrangians by connect summing only SLags; this would then

766: be enough to prove the general stability of SLags.)

767: 

768: The idea is that if $\phi(L_1)>\phi(L)$, with both $L_1$ and $L$

769: SLags, then the Floer index of any intersection point of $L_1$

770: and $L$ is strictly positive (\ref{maslov}), almost by definition.

771: But if $L_1$ were to destabilise $L$, i.e. $L=L_1\#L_2$ for some

772: $L_2$, then there should be canonical morphisms $HF^0(L_1,L)\ne0$

773: and $HF^0(L,L_2)\ne0$, a contradiction.

774: 

775: The morphism from $L_1\#L_2$ to $L_2$, by which we mean an element of

776: \begin{equation} \label{hom}

777: HF^0(L_1\#L_2,L_2),

778: \end{equation}

779: can be described as follows (the element of $HF^0(L_1,L_1\#L_2)$

780: is similar). We use the description of the connect sum in Section

781: \ref{index}. Choose a Morse function $f$ on $L_2$ which has

782: local maxima at intersection points $p$ with $L_1$,

783: and in local Darboux charts as in Section \ref{index}, is

784: pulled up from a linear function on $\gamma_2$. Let the function

785: have a unique local minimum elsewhere on $L_2$, and now use

786: this to hamiltonian deform $L_2$ off $L_1\#L_2$. By construction

787: $L_1$ and $L_1\#L_2$ now intersect at the critical points of $f$

788: only, with Floer index the Morse index of $f$. In terms of

789: Figure 1, as $f$ has a maximum on $\gamma_2$ at the vertex of

790: $\gamma_2$, it defines a hamiltonian deformation of $\gamma_2$

791: \emph{downwards}, away from the connect-sum neck. As $L_2$ only

792: intersects $L_1$ near these connect-sum necks, we can make

793: our charts small enough that $L_2$ now only intersects $L_1\#L_2$

794: where its hamiltonian deformation intersects the old $L_2$, i.e.

795: at the critical points of $f$.

796: 

797: We now have a unique index zero point of $(L_1\#L_2)\cap L_2$ at

798: the unique local minimum of $f$. What we require is that

799: this survives in the passage to cohomology of the cochain complex

800: to give $HF^0(L_2,L_1\#L_2)\cong\C$. For instance, if there are no

801: index one points (i.e. $f$ is a Morse function with only minima and

802: index $\ge2$ critical points) then this will clearly be the case.

803: More generally there is a spectral sequence analogous to Po\'zniak's

804: \cite{P} with

805: $$

806: \mathrm{coker}\,\{\bigoplus_i\C_{p_i}[-n]\to H^*(L_2)\}\ 

807: \Longrightarrow\ HF^*(L_1\#L_2,L_2)

808: $$

809: (with a certain bigrading)

810: converging to $HF^*(L_1\#L_2,L_2)$. Here the notation means that a

811: copy of $\C$ is mapped to $H^n(L_2)$ (i.e. it is in degree $n$)

812: for every intersection point $p_i$ via the Morse theory for $f$ (whose

813: maxima are at the $p_i$). Therefore the degree zero part also survives

814: if, for instance, $H^1(L_2)=0$. If $L_2$ is a sphere we can proceed

815: more directly by applying Seidel's exact sequence \cite{S3}.

816: 

817: Using similar methods on Lagrangians rather than SLags, we can cut

818: down on the number of possible destabilising Lagrangians $L_1$

819: we must check to conclude that a given $L$ is stable, rather

820: analogously to only checking for subsheaves of vector bundles

821: amongst those of lower rank. There are

822: no morphisms (non-zero elements of $HF^0(L_1,L)$) if the phase

823: of $L_1$, at an intersection point $p$, is greater than that of

824: $L$; the Floer index at $p$ is strictly positive.

825: So for $L_1$ to destabilise $L$ (and so $HF^0(L_1,L)\ne0$ for

826: Lagrangians satisfying the same conditions as above and in (\ref{!}),

827: e.g. homology spheres) we must have

828: $$

829: \inf_{x\in L_1}\theta_{L_1}(x)<\sup_{x\in L}\,\theta_L(x),

830: $$

831: and in fact the corresponding phase inequality

832: at each point of intersection.

833: 

834: Thus we do not have to check all Lagrangians $L_1,\,L_2$ for the stability

835: of $L'$ in \cite{Th}, just those whose phase function satisfies

836: $$

837: \inf_{x\in L_1}\theta_{L_1}(x)\le\sup_{x\in L}\,\theta_L(x)

838: \quad\text{and}\quad

839: \sup_{x\in L_2}\theta_{L_2}(x)\ge\inf_{x\in L}\,\theta_L(x),

840: $$

841: where we can in fact replace the left hand sides of these inequalities

842: by the inf (respectively sup) over all Lagrangians in the same hamiltonian

843: deformation class.

844: 

845: Assuming the conjecture in \cite{Th}, so that we need only check

846: SLag destabilisers to show that stability of $L_0$, we are reduced to

847: checking for destabilising subobjects amongst those $L_1,\ L_2$ whose

848: homology classes sum to $[L_0]$ and satisfy

849: \begin{equation} \label{ineq}

850: \sup_{L\approx L_0}\left(\inf_{x\in L}\,\theta_L(x)\right)\le\phi(L_1)

851: \le\phi(L_2)\le\inf_{L\approx L_0}\left(\sup_{x\in L}\,\theta_L(x)\right).

852: \end{equation}

853: 

854: 

855: \subsection{A Jordan-H\"older decomposition for Lagrangians} \label{JH}

856: 

857: In order to understand limits of mean curvature flow it will be useful

858: to have the following concept;

859: an analogue for Lagrangians of the Jordan-H\"older filtration of

860: sheaves (see \cite{HuL} 1.5, for instance).

861: 

862: \begin{Definition}

863: Given two graded Lagrangians $L_1,\,L$, write $L_1\le L$ if there

864: exists a graded Lagrangian $L_1'$ such that $L\approx L_1\#L_1'$.

865: We then also write $L/L_1$ for $L'_1$, and say that $L_1$ is

866: a subobject of $L$.

867: 

868: A Jordan-H\"older filtration of $L$ is a sequence of graded

869: Lagrangians $L_i$ such that

870: $$

871: L_1\le L_2\le\ldots\le L_k=L,

872: $$

873: and $L_i':=L_{i+1}/L_i$ is stable. The Jordan-H\"older decomposition

874: of $L$ is the the singular union

875: \begin{equation} \label{decomp}

876: L_1\cup(L_2/L_1)\cup\ldots\cup(L/L_{k-1}).

877: \end{equation}

878: \end{Definition}

879: 

880: In sheaf theory the Jordan-H\"older filtration need not be unique,

881: but the decomposition is. For smooth \emph{connected} Lagrangians,

882: with connected $L_i$ for all $i$, however, we expect

883: the filtration to be unique too; the difference is essentially that

884: while direct sum is an operation on bundles, we are proposing that

885: its mirror is the (singular) union of Lagrangians, and this

886: cannot give a smooth Lagrangian if there is non-zero Floer cohomology

887: between the two Lagrangians.

888: 

889: If we assume the conjectures of \cite{Th}

890: and Section \ref{conj} below, and the properties of Floer

891: cohomology \cite{FO3} for all of the above Lagrangians (e.g. if they

892: are homology spheres), we can demonstrate the existence and uniqueness

893: of the Jordan-H\"older filtration for a Lagrangian $L$ whose phase

894: function of $L$ satisfies $\sup\theta_L-\inf\theta_L<\pi$.

895: 

896: Without loss of generality we may assume (by rotating $\Omega$) that 

897: $\theta$ lies between $\pi/2-\epsilon$ and $-\pi/2+\epsilon$, for

898: some $\epsilon>0$. By the inequality (\ref{ineq}) above, then, any

899: $L_1\#L'$ destabilising it will satisfy $\phi(L_1),\phi(L')\in

900: (-\pi/2+\epsilon,\pi/2-\epsilon)$.

901: 

902: We choose such an $L_1$ of maximal phase, and, amongst other such $L_1$s

903: of the same phase, minimal $\int_{L_1}\mathrm{Re}\,\Omega$ (for the

904: purposes of this proof we will call this quantity cohomological volume).

905: This still need not specify $L_1$ uniquely though.

906: 

907: We claim that such an $L_1$ must

908: be stable by construction. Any subobject of $l\le L_1$ would also be

909: a subobject of $L$ and so by the construction of $L_1$ must either

910: have smaller phase, which is not possible since it destabilises $L_1$,

911: or equal phase and greater or equal cohomological volume.

912: But $L_1=l\#l'$, where $l'=L_1/l$ has phase

913: $\phi(l')=\phi(L_1)\in(-\pi/2,\pi/2)$ and so positive cohomological

914: volume $\int_{l'}\mathrm{Re}\,\Omega$. So the complex numbers

915: $$

916: \int_{L_1}\Omega=\int_l\Omega+\int_{l'}\Omega

917: $$

918: all have positive real part, implying that the cohomological volume

919: of $l$ is strictly less that that of $L_1$, a contradiction.

920: 

921: We then apply the same

922: procedure to $L'$, producing an $L_2\into L'$, and so on.

923: By construction $\phi(L'/L_2)\le\phi(L')\le\phi(L)<\pi/2-\epsilon$,

924: and there is a canonical morphism (\ref{hom}) in

925: $HF^0(L,L'/L_2)\ne0$, making $\phi(L'/L_2)\ge\inf_L\theta_L>-

926: \pi/2+\epsilon$ by (\ref{ineq}).

927: 

928: Thus, inductively, we get the same inequalities at each stage, and

929: the cohomological volume of $L'$ decreases strictly with each

930: decomposition $L\approx L_1\#\ldots\#L_n\#L'$. The cohomological

931: volume of any $l$ with phase $\phi(l)\in(-\pi/2+\epsilon,\pi/2-\epsilon)$

932: is greater than (or equal to in the SLag case) $\cos(\pi/2-\epsilon)

933: \int_l\vol$, by (\ref{vol}), where $\vol$ is its

934: Riemannian volume form. This is bounded below above zero,

935: so the process can have at most a finite number of steps.

936: 

937: This gives us the Jordan-H\"older filtration; next we consider

938: uniqueness when the $L_i$s are connected (assuming the conjectures

939: of \cite{Th} and Section \ref{conj} and some Floer cohomology).

940: Suppose that $L_1'\le L_2'\le\ldots\le L$ is another such connected

941: decomposition where we take the $L_i'$s to be SLag assuming the

942: conjecture of \cite{Th}. If $HF^0(L_1',L_1)\ne0$ then

943: by the proof of Theorem \ref{!} (which applies as $L_1'$ and $L_1$

944: have the same phase), $L_1$ and $L_1'$ are equal, and we pass to $L_2$.

945: 

946: If, however, $HF^0(L_1',L_1)=0$, then we claim that

947: $HF^0(L_1',L/L_1)\ne0$. Again this should follow from standard

948: facts about Floer cohomology, in particular a long exact sequence

949: $HF^*(L_1',L_1)\to HF^*(L_1',L)\to HF^*(L_1',L/L_1)\to HF^{*+1}

950: (L_1',L_1)$. For $L/L_1$ a sphere this is Seidel's exact sequence

951: (\cite{S3} Theorem 3.3), and in general one can establish it at

952: the level of chains by good choices of hamiltonian perturbations

953: as in Section \ref{ssl}; as usual the problem is in controlling

954: the differential, i.e. holomorphic discs.

955: 

956: Assuming this we may pass to $L_2$; inductively we eventually obtain that

957: $L_1'$ is isomorphic to one of the graded pieces $L_{i+1}/L_i$ of

958: the original filtration, and is a subobject of $L_{i+1}$ but not of

959: $L_i$. But this gives us a contradiction (in contrast to the sheaf

960: analogue), since we have that both $L_{i+1}\approx L_i\#L_1'$ and

961: $L_1'$ is a subobject of $L_{i+1}$. The first condition ensures that

962: there are representatives of the hamiltonian deformation classes

963: such that $L_{i+1}$ and $L_1'$ have no index $n$ intersection points

964: by the construction of (\ref{hom}), so that $HF^n(L_{i+1},L_1')=0$.

965: But this is $HF^0(L_1',L_{i+1})^*$, which cannot vanish by the second

966: condition. (It is here we use the connectivity condition, i.e. that

967: the connect sum $L_{i+1}=L_1'\#(L_{i+1}/L_1')$ is not a trivial

968: disjoint union. Without the connectivity condition the usual

969: proof (e.g. \cite{HuL} 1.5) that the Jordan-H\"older

970: \emph{decomposition} (rather than filtration)

971: of sheaves is unique applies to Lagrangians, now that we have proved

972: or assumed all (the mirror analogues of) the algebraic facts used for

973: sheaves in terms of Floer cohomology instead of Exts. \\

974: 

975: So in the simplest case of instability, such as the example of Joyce

976: considered in \cite{Th}, where $L=L_1\#L_2$ is the only relevant

977: decomposition of $L$ with $\phi(L_1)\ge\phi(L_2)$, the Jordan-H\"older

978: decomposition (\ref{decomp}) would be simply $L_1\cup L_2$

979: (where the $L_i$ are SLag

980: representatives of their classes). This, like all such

981: decompositions, is in the closure of the hamiltonian deformation

982: orbit of $L$ while not being in the orbit itself.

983: 

984: This should have relevance to the Schoen-Wolfson programme \cite{SW} to

985: find canonical representatives (in a fixed hamiltonian

986: deformation class) of Lagrangian homology classes using volume

987: minimisers and so SLags (they do not use a flow, but regularity results

988: to study minimising currents). Our conjecture (as in \cite{Th} and later

989: in Section \ref{conj}) should either provide a unique SLag in a

990: hamiltonian deformation class, or a number of SLags in a Jordan-H\"older

991: decomposition.

992: 

993: For instance in the example above of $L_1\#L_2$ in 2 dimensions

994: we would produce

995: SLags in the classes of $L_1$ and $L_2$, but we could also form

996: $L_2\#L_1$; this could then be stable (it is no longer destabilised by

997: either of the $L_i$; if the phases of the $L_i$ are sufficiently close

998: one can show that in fact nothing else destabilises it either)

999: and we should recover a SLag in this class (and so in the same

1000: homology class in two dimensions).

1001: 

1002: Since in two dimensions SLags are just holomorphic curves with respect

1003: to a different complex structure, this places heavy restrictions

1004: on stability. Take the $L_i$ above to be spheres in $K3$ surfaces.

1005: Then any holomorphic sphere is unique in its homology class (it has

1006: negative self intersection $-2$, so does not lie in a pencil). Any

1007: other homologous hamiltonian deformation class

1008: must therefore be unstable. Good examples are provided by taking a

1009: stable (SLag/holomorphic) sphere, and applying the square of a Dehn

1010: twist $T^2_{L_1}$ to it; this preserves homology classes but can change

1011: hamiltonian deformation classes. If it does it should produce an

1012: unstable Lagrangian with copies of $L_1$ in its Jordan-H\"older

1013: decomposition; this happens in all simple cases.

1014: $L_1\#L_2$ is taken to $L_2\#L_1$,

1015: for instance; only one of these can be stable, the other having

1016: a Jordan-H\"older decomposition $L_1\cup L_2$ in the simplest case.

1017: 

1018: More generally, instead of studying the action on individual (S)Lags

1019: of symplectomorphisms like $T^2_{L_1}$ above, we could try to study

1020: them all at once by studying the Lagrangian graph of the

1021: symplectomorphism in $X\times X$, and its mean curvature flow. This

1022: looks for minimal energy representatives of the hamiltonian isotopy

1023: class of a symplectomorphism, and breaks graphs up into correspondences

1024: representing singular maps (birational maps in the hyperk\"ahler case)

1025: with singularities concentrated in loci whose stability is affected by

1026: the symplectomorphism. For a Dehn twist $T_L$, for instance, we would

1027: expect to get the graph $\Delta$ of the identity, union $L\times L$.

1028: This also shows what the analogue of a Dehn twist $T_L$ should be when

1029: $L$ is not a sphere but a rational homology sphere (so that it is still

1030: spherical to complex coefficients, and so mirror to a twist on the

1031: derived category of sheaves on the mirror Calabi-Yau \cite{ST}). Namely

1032: $\Delta\cup L\times L$ is a Lagrangian correspondence in $X\times X$

1033: which should give an automorphism of the derived Fukaya category of $X$

1034: (by the usual Fourier-Mukai-type construction) \emph{not} induced by a

1035: symplectomorphism of $X$.

1036: 

1037: 

1038: \section{An example: families of affine quadrics} \label{eg}

1039: 

1040: Here we consider an example suggested to us by both Paul Seidel and

1041: Cumrun Vafa, used in \cite{SV} and \cite{KS}. Consider the

1042: affine algebraic variety $X^n$ given by

1043: $$

1044: \sum_{i=1}^nx_i^2=p(t)

1045: $$

1046: in $\C^n\times\C$, where $p$ is some polynomial in $t\in\C$ with only

1047: simple zeros. Denote by $\pi:\,X^n\to\C$ the projection to the $t$

1048: coordinate. Here we use the K\"ahler structure restricted from

1049: $\C^{n+1}$, and the nowhere-zero holomorphic volume form given by

1050: taking the Poincar\'e residue (\cite{GH} p 147) of the standard form

1051: $dx_{1\ldots n}dt:=dx_1\ldots dx_{n\,}dt$ on $\C^{n+1}$;

1052: this can be written as

1053: \begin{equation} \label{form}

1054: (-1)^{n+i+1}{dx_{1\ldots\hat{\imath}\ldots n\,}dt\res{X^n}\over2x_i}=

1055: {dx_1\ldots dx_n\res{X^n}\over\dot p(t)}

1056: \end{equation}

1057: for any $i$ (so where $x_i=0\ \forall i$ we can use the second expression).

1058: Here $\hat{\imath}$ means that we omit the $dx_i$ term from the wedge product.

1059: This is then \emph{not} parallel, and the metric

1060: we have chosen is \emph{not the Ricci-flat one}. Nonetheless it is

1061: a good explicit testing ground for the conjecture; we can still define

1062: $\theta$ as the phase of $\Omega\res L$ and SLags as having constant phase,

1063: of course we then use flow

1064: by the $\Jd$ vector, rather than mean curvature flow in

1065: this metric. While the two flows are similar and would be the same in the

1066: Ricci-flat metric, only the former has SLags as its stationary points

1067: (for the latter we get minimal submanifolds, which in this metric are

1068: not quite SLag). As Edward Goldstein pointed out to us, the $\Jd$ flow

1069: is the gradient flow of the weighted volume functional $\int_L|\Omega|

1070: \vol$ instead of $\int_L\vol$; everything proceeds analogously

1071: to before on weighting all vols by $|\Omega|$, as we shall see.

1072: 

1073: Each smooth fibre over $t\in\C$ is an affine quadric with a natural

1074: Lagrangian $S^{n-1}$

1075: `real' slice, namely the intersection of the fibre with the slice

1076: $$

1077: x_i\in\sqrt{p(t)}\,\R\quad\forall i.

1078: $$

1079: It is invariant under the obvious $O(n)$ action on $X^n$, and is the

1080: vanishing cycle of every singular fibre (i.e. the fibres over the

1081: roots of $p$). Therefore any path

1082: $\gamma:\,I\to\C$ ($I\ni u$ being some interval in $\R$) from one

1083: zero of $p$ to another lifts to give a canonical

1084: $O(n)$-invariant Lagrangian $n$-sphere $\gamma^n$, $S^{n-1}$-fibred

1085: over $\gamma$ except at the endpoints where it closes up. Also,

1086: any vector $\gamma'\partial_t$ in the base $\C\ni t$ lifts canonically

1087: to a vector

1088: \begin{equation} \label{lift}

1089: \gamma'\left(\partial_t+{\dot p\over2p}\sum x_i\partial_{x_i}\right)

1090: \end{equation}

1091: tangent to the infinitesimal Lagrangian $\gamma^n$ lying above $\gamma'$.

1092: Here $'$ denotes $d/du$. Note that $\gamma^1$ is a closed curve double

1093: covering $\gamma$, branched over $\gamma$'s endpoints. We will use this

1094: curve $\gamma^1$ later to study $\gamma^n$.

1095: 

1096: The phase function $\theta$ on $\gamma^n$ is also $O(n)$-invariant and

1097: so a function of $t\in\C$ which we may calculate at $x_1=\sqrt{p(t)},\,

1098: x_i=0\ \forall i\ge2$. Choosing a basis of tangent vectors to $\gamma^n$

1099: at this point,

1100: \begin{equation} \label{basis}

1101: \gamma'(\partial_t+{\dot p\over2p^{1/2}}\partial_{x_1}),\,

1102: \sqrt p\,\partial_{x_2},\ldots,\sqrt p\,\partial_{x_n},

1103: \end{equation}

1104: wedging them together and evaluating against the $(n,0)$-form

1105: (\ref{form}) gives

1106: $$

1107: \gamma'{\dot p\over2p^{1/2}}{(\sqrt p)^{n-1}\over\dot p}=

1108: {1\over2}\gamma'p^{n/2-1}.

1109: $$

1110: Therefore the phase function on $\gamma^n$ is given by

1111: \begin{equation} \label{pf}

1112: \theta:=\theta(\gamma^n)=\theta(\gamma')+\big({n\over2}-1\big)

1113: \theta(p(\gamma)),

1114: \end{equation}

1115: where $\theta(\gamma')$ is the usual angle of the path $\gamma$, and

1116: $\theta(p)$ is the phase of the complex number $p$ evaluated at

1117: $t=\gamma$. So

1118: $$

1119: d\theta=\left({d\theta(\gamma')\over du}+\big({n\over2}-1\big)

1120: {d\theta(p)\over du}\right)du,

1121: $$

1122: where $du$ is the pullback to $\gamma^n$ under the projection $\pi$

1123: of the corresponding 1-form on $\gamma(I)\subset\C$. 

1124: 

1125: Using the metric and the orthogonal basis (\ref{basis}) we see

1126: that

1127: $$

1128: \,\widetilde{\!du\,}={\gamma'(\partial_t+{\dot p\over2p^{1/2}}

1129: \partial_{x_1})\over|\gamma'(\partial_t+{\dot p\over2p^{1/2}}

1130: \partial_{x_1})|^2}

1131: $$

1132: at the point $x_1=\sqrt{p(t)},\,x_i=0\ \forall i\ge2$.

1133: 

1134: Therefore, by $O(n)$-invariance and the holomorphicity of the

1135: projection $\pi$, $\Jd$ is the canonical lift (\ref{lift}) of

1136: $$

1137: {d(\theta(\gamma')+\big({n\over2}-1\big)\theta(p))\over du}\,

1138: {\pi_*\!\left[J\gamma'(\partial_t+{\dot p\over2p^{1/2}}

1139: \partial_{x_1})\right]\over|\gamma'|^2(1+|\dot p|^2/4|p|)}=

1140: {{1\over|\gamma'|}{d\over du}(\theta(\gamma')+\big({n\over2}-1\big)\theta(p))

1141: \over1+|\dot p|^2/4|p|}\,

1142: i{\gamma'\over|\gamma'|}\partial_t.

1143: $$

1144: Denoting by $\mathbf t=\partial_u/|\gamma'|=\gamma'\partial_t/|\gamma'|$

1145: and $\mathbf n=i\mathbf t$ the unit tangent and normal

1146: vectors to $\gamma$ at a point $\gamma(u)$, the above is

1147: $$

1148: {\mathbf t[\theta(\gamma')+\big({n\over2}-1\big)\theta(p)]\over

1149: 1+|\dot p|^2/4|p|}\,\mathbf n.

1150: $$

1151: By the Cauchy-Riemann equations for the holomorphic function

1152: $\log p=\log|p|+i\theta(p)$, $\mathbf t\theta(p)=-\mathbf n\log|p|$,

1153: so that our flow is the lift to $X^n$ of the flow of $\gamma$ with vector

1154: \begin{equation} \label{result1}

1155: V^n={1\over1+|\dot p|^2/4|p|}(\MCV+\,

1156: (1-n/2)\mathbf n(\log|p|)\,\mathbf n),

1157: \end{equation}

1158: where $\MCV$ is the usual mean curvature vector of $\gamma$ in

1159: the flat metric on $\C$.

1160: 

1161: So we can reduce studying our flow to studying the flow

1162: of a curve $\gamma$ with fixed endpoints (at zeros of $p$), under

1163: the above vector field. We would like to relate this to mean curvature

1164: flow of $\gamma\subset\C$ in a different metric, and also to

1165: both our flow and the mean curvature flow for the double

1166: $\gamma^1$ of $\gamma$ in the

1167: double cover $X^1$ of $\C$ branched over the zeros of $p$.

1168: The advantage of this is that we now have a flow for a closed

1169: curve instead of a boundary value problem (but since the flow

1170: has $O(1)=\Z/2$ symmetry it is equivalent to a flow of the

1171: original curve $\gamma$ with fixed endpoints). We need the following

1172: lemma.

1173: 

1174: \begin{Lemma}

1175: Let $\langle\,.\,,\,.\,\rangle$ be the standard metric on $\C$, and

1176: $g$ a positive real-valued function on $\C$. Then with respect

1177: to the metric $g\langle\,.\,,\,.\,\rangle$, the mean curvature vector

1178: of a curve $\gamma\subset\C$ is, in terms of the standard mean curvature

1179: vector $\MCV$ (and

1180: calculating the unit normal $\mathbf n$ in the standard metric),

1181: $$

1182: {1\over g}\left(\MCV-{1\over2}\mathbf n(\log g)\,\mathbf n\right).

1183: $$

1184: \end{Lemma}

1185: 

1186: \begin{Proof}

1187: The endomorphism-valued 1-form $\Gamma$ defined by

1188: $$

1189: \Gamma_XY={1\over2g}((Xg)Y+(Yg)X-\langle X,Y\rangle\,\widetilde{\!dg\,})

1190: $$

1191: is symmetric and so defines a torsion-free connection on $\C$. It

1192: is easily checked to be orthogonal with respect to the metric

1193: $g\langle\,.\,,\,.\,\rangle$, and so gives its Levi-Civita connection

1194: $\nabla+\Gamma$ (where $\nabla$ is the usual connection on $\C$).

1195: Then $\langle\Gamma_{\gamma'}\gamma',\mathbf n\rangle$

1196: (where $\mathbf n$ is calculated in the original metric) is

1197: $-{1\over2g}|\gamma'|^2\mathbf n(g)=-{1\over2}|\gamma'|^2\mathbf n(\log g)$.

1198: 

1199: Since the unit normal to $\gamma$ in the new metric is

1200: $g^{-1/2}\mathbf n$, the new mean curvature vector is

1201: $$

1202: {g\langle\gamma''+\Gamma_{\gamma'}\gamma',g^{-1/2}\mathbf n\rangle\over

1203: g|\gamma'|^2}\,g^{-1/2}\mathbf n={1\over g}\left(\MCV-{1\over2}

1204: \mathbf n(\log g)\,\mathbf n\right),

1205: $$

1206: as claimed.

1207: \end{Proof}

1208: 

1209: Using this we can get a number of geometrically interesting flows which

1210: are equivalent to our original flow in $X^n$. Namely, using the result

1211: (\ref{result1}), the above Lemma, and the fact

1212: that locally (away from branch points) $X^1$ is conformally equivalent

1213: to $\C$ with its metric scaled by $g=1+|\dot p|^2/4|p|$ (by

1214: holomorphicity and (\ref{lift})), we can deduce the following.

1215: 

1216: Denote by $V^n$ the flow vector $\pi_*\Jd$ of the

1217: curve $\gamma\subset\C$ under our flow in $X^n$. Denote by

1218: $\MCV^1_g$ the flow vector of $\gamma\subset\C$ under mean curvature

1219: flow of $\gamma^1$ in $X^1$, with $X^1$'s natural metric scaled by

1220: a $\Z/2$-invariant function $g$ (and omit the

1221: $g$ in the notation if $g\equiv1$). And denote by $\MCV_g$

1222: the mean curvature vector of $\gamma$ in $\C$ with metric $g

1223: \langle\,.\,,\,.\,\rangle$.

1224: 

1225: Letting $\mathbf n$ be the unit normal to $\gamma$ calculated in the

1226: standard metric on $\C$, and letting $f={|p|^{n-1}\over|p|+

1227: |\dot p|^2/4}$\,, we have the following relations between the

1228: various flows:

1229: \begin{equation} \label{result2}

1230: V^n=f.\MCV^1_f=f.\MCV_{|p|^{n-2}},

1231: \end{equation}

1232: and

1233: \begin{equation} \label{result4}

1234: V^n=\MCV^1-\left[{1\over2}|p|^{2-n}\mathbf n(f)\right]\mathbf n

1235: =V^1-{1\over2}(n-1){\mathbf n(|p|)\over|p|+|\dot p|^2/4}\,\mathbf n.

1236: \end{equation}

1237: 

1238: The problem with the first two is that on $\C\supset\gamma$ the flow

1239: is \emph{not} parabolic, it has degeneracies at the end points. As

1240: $X^1$ is so closely modelled on $X^n$ (and is in fact canonically

1241: embedded in it), however, we might expect better on $X^1$. This is

1242: more or less true; the result is that writing (\ref{result4})

1243: in terms of the unit normal $\mathbf n^1$ \emph{on} $X^1$, we get

1244: 

1245: \begin{Theorem} \label{result!}

1246: $$

1247: V^n=\MCV^1-{1\over2}(n-1)\mathbf n^1(\log|p|)\,\mathbf n^1+

1248: {1\over2}\mathbf n^1(\log(|p|+|\dot p|^2/4))\,\mathbf n^1.

1249: $$

1250: \end{Theorem}

1251: 

1252: The last term is bounded (as near a zero of $p$, $\dot p\ne0$

1253: by nondegeneracy of $p$'s zeros) and so unimportant, we shall

1254: see, and the flow

1255: resulting from the first term is well understood. The second term

1256: is more curious; it is of the order of $\mathbf t^1\theta(p)\approx

1257: \mathbf t^1\theta((\gamma^1)')/2$ (where $'=\mathbf t^1$ denotes

1258: differentiation with respect to arclength on $X^1$) whenever we are

1259: close to a point where $\gamma$ emanates from a zero of $p$

1260: (so that $p\approx Ct$ and $\theta(p)\approx\theta(\gamma)

1261: \approx\theta(\gamma')\approx\theta((\gamma^1)')/2$).

1262: (The last approximation is of course

1263: not valid if $\gamma$ simply passes close to a zero of $p$; then

1264: the equation blows up quickly as a glance at (\ref{result1}) shows,

1265: $\gamma$ flowing to

1266: this zero and breaking across it as discussed below; in the stable

1267: case we will be able to rule out this behaviour and need only consider

1268: $\gamma$ ending at the zero.)

1269: 

1270: But this is half the curvature of $\gamma^1$, so we get an

1271: approximation to the first term again, and something like mean

1272: curvature flow for $\gamma^1\subset X^1$. In fact in a small

1273: neighbourhood of (the double cover of) a zero of $p$, in coordinates

1274: $(x,y)$ in which $\gamma^1$ is a graph $y(x)$, the evolution

1275: PDE is of the general shape

1276: $$

1277: y_t=y_{xx}+{y_x\over x},

1278: $$

1279: where by the $\mathbb Z/2$-symmetry $y_x\res{x=0}=0$, so the second

1280: term is approximately $y_{xx}$.

1281: So for some analysis we use this flow for $\gamma^1$, while for the rest

1282: we pass back to $n$-dimensions, and work with the phase function

1283: $\theta$ instead, giving a more standard (but $n$-dimensional)

1284: parabolic equation.

1285: 

1286: We first assert how the flow behaves, before proving it in the

1287: stable case in the next section (Theorem \ref{faq}).

1288: Note that any deformation of

1289: $\gamma$ is a hamiltonian deformation of $\gamma^n$ (and SLag

1290: $\gamma^n$s have no moduli) since the $\gamma^n$s are spheres. We

1291: picture what happens in Figures 2 and 4 in the 2 and 3

1292: dimensional cases respectively. The dots represents zeros

1293: of $p$ in both cases, and the epsilons and zeros are phases.

1294: 

1295: \begin{figure}[h]

1296: \center{

1297: \input{stab2.pstex_t}

1298: \caption{The two connect sums $L_1\#L_2$ (1,\,2) and

1299: $L_2\#L_1$ (3,\,4) in 2-dimensions}}

1300: \end{figure}

1301: 

1302: In two dimensions the curves $\gamma$ whose Lagrangians $\gamma^2$ have

1303: constant phase are the straight lines, as can be seen from (\ref{pf}).

1304: Curves such as those marked 1 and 4 in Figure 2 flow towards

1305: a straight line (of some \emph{non-zero} angle) corresponding to a

1306: SLag, whereas curve 2 flows up until it `hangs' on a zero of $p$ (in

1307: finite time), where, on restarting the flow for 2 different curves,

1308: the separate flows form a kink and in the limit converge to

1309: destabilising SLags $L_1,\ L_2$ of different phases.

1310: These unions of SLags of different phases are still

1311: stationary for the volume functional (satisfying the second

1312: order variational equations, just not the first order SLag equations),

1313: and in fact are minimising in odd dimensions (the angle criterion

1314: \cite{N}, \cite{L} makes minimality of the singular union locally

1315: equivalent to the above destabilising phase condition; in even

1316: dimensions reversing the order of the Lagrangians reverses the

1317: inequality and the configuration is not minimal, just stationary).

1318: 

1319: Again we see how the phase or angle criterion comes to bear; curves

1320: 1 and 2 are in the same homology class, but the two different phase

1321: signs give very different results. As noted before in Figure 1,

1322: this is related to the necessity of the phase to vary non-monotonically

1323: to form the unstable connect sum; in Figure 3 we plot

1324: the phases of the two connect sums, and with dotted lines

1325: their limits under the heat flow (\ref{newdot}) (this is the correct

1326: modification of (\ref{thetadot}) in the non Ricci-flat case).

1327: 

1328: \begin{figure}[h]

1329: \center{

1330: \input{phase.pstex_t}

1331: \caption{The phase $\theta_{L=L_1\#L_2}$ of (1) $(\theta_{L_1}\equiv

1332: -\epsilon)$ and (2) $(\theta_{L_1}\equiv\epsilon)$ respectively.}}

1333: \end{figure}

1334: 

1335: Drawing $\gamma$ the other way round the zero of $p$ gives something in

1336: the same homology class (the Dehn twist around the root of $p$ does not

1337: alter the homology class of the $S^1$

1338: fibre over $\gamma$), which is the opposite connect sum discussed in

1339: \cite{Th} -- once the phase inequality becomes unstable for one

1340: connect sum it becomes stable for the other.

1341: 

1342: The two connect sums are related by monodromy, as in \cite{Th}.

1343: Take a one parameter family of polynomials $p$ which rotates

1344: two zeros $z_1,\,z_2$ of $p$ around each other. Then under the

1345: resulting monodromy a curve joining $z_1$ to a

1346: third zero $z_3$ is taken from being `above' $z_2$ to

1347: being below it, thus turning one connect sum into the other. \\

1348: 

1349: The three dimensional picture is similar. In Figure 4 we plot the

1350: lines corresponding to SLags of phase zero, and connect sums $L_1\#L_2$

1351: for $\phi(L_2)=0$ and $\phi(L_1)=\mp\epsilon$ (curves 1 and 2).

1352: Again we see

1353: the same behaviour with the phases behaving as in the graphs

1354: in Figure 3 and the flow getting hung on a zero of $p$ in

1355: the unstable case, splitting the Lagrangian.

1356: 

1357: Reversing the order of the connect sum in this case involves taking the

1358: $S^2$ fibre once around the zero of $p$; this effects a Dehn twist,

1359: reversing its orientation. Thus although curve 3 appears to give a

1360: Lagrangian in the same homology class, it is not; the phase of $L_1$

1361: once we have been round the root of $p$ has shifted by $\pi$ and we get

1362: the connect sum $L_2\#(L_1[-1])$ discussed in \cite{Th}. As is also

1363: discussed there, this can be seen to be unstable.

1364: 

1365: \begin{figure}[h]

1366: \center{

1367: \input{stab3.pstex_t}

1368: \caption{The two connect sums $L_1\#L_2$ and

1369: $L_2\#(L_1[-1])$ in 3-dimensions}}

1370: \end{figure}

1371: 

1372: Things are not quite as simple as we have portrayed them if the

1373: initial curve has very large phase variation. It is quite possible

1374: for a curve corresponding to a \emph{stable} Lagrangian, which is

1375: nonetheless very far from being a SLag, to pass close to a zero of

1376: $p$ without the large negative curvature away from the zero that the

1377: SLags exhibit. It can then flow into the zero, the Lagrangian being

1378: split into unions of Lagrangians of which it was a connect sum,

1379: despite their phases being such that they do not destabilise it.

1380: This limit is in the closure of the hamiltonian deformation orbit

1381: of the original Lagrangian, but does not contradict stability.

1382: 

1383: (A similar often-ignored subtlety occurs with stable bundles:

1384: when moduli of semistable sheaves are created by using GIT on part of

1385: a Quot scheme, the orbits of stable sheaves are \emph{not} closed in

1386: Quot -- the closures contain gradeds coming from any extension of

1387: sheaves forming the sheaf, stable or not. It is only in the

1388: locus of points of Quot representing

1389: semistable sheaves that the stable orbits are closed.)

1390: 

1391: The point about stable

1392: Lagrangians is that \emph{this can be avoided} by choosing a hamiltonian

1393: deformation of the

1394: Lagrangian to have sufficiently small variation in phase $\theta$

1395: (or sufficiently small volume) that it cannot be split into

1396: destabilising Lagrangians of different phase (or higher total volume);

1397: this we discuss now.

1398: 

1399: 

1400: \section{The conjecture} \label{conj}

1401: 

1402: It is now clear what our conjecture should be. Fix a (graded) Lagrangian

1403: submanifold $L$ of a Calabi-Yau $n$-fold $X$, and choose

1404: the phase of $\Omega$ such that the cohomological phase $\phi(L)=0$.

1405: Suppose first that the variation in

1406: $L$'s phase function $\theta$ is sufficiently small in the sense that

1407: \begin{equation} \label{close}

1408: [\phi(L_1),\phi(L_2)]\not\subseteq(\inf_L\theta,\sup_L\theta),

1409: \end{equation}

1410: for all graded connect sums $L_1\#L_2\approx L$ (by this we mean

1411: either the pointwise connect sums of Section \ref{index} or one of

1412: the relative connect sums discussed in \cite{Th}). This condition

1413: (\ref{close})

1414: is preserved by the flow, by the maximum principle and equation

1415: (\ref{thetadot}), and so prohibits $L$ splitting up as a

1416: connect sum under the flow, as in the limit of flowing to such a

1417: splitting (\ref{close}) would be violated.

1418: 

1419: We can also usefully consider volume instead of phase. If the

1420: Riemannian volume of our Lagrangian $L$ is less than the

1421: cohomological volume of any decomposition:

1422: \begin{equation} \label{vclose}

1423: \vol(L)\le\int_{L_1}e^{-i\phi(L_1)}\Omega+\int_{L_2}e^{-i\phi(L_2)}

1424: \Omega,

1425: \end{equation}

1426: for all $L_1,\,L_2$ such that $L\approx L_1\#L_2$, then we again expect

1427: convergence of mean curvature flow to a SLag representative for $L$.

1428: This is also preserved under the flow, by (\ref{voldot}), and so

1429: precludes the flow splitting $L$ into $L_1\cup L_2$, as this splitting

1430: would necessarily have higher volume. (As the referee pointed out,

1431: this is a global condition on $L$, whereas (\ref{close}) appears

1432: to be a pointwise one. The global nature of (\ref{close}) is that the

1433: points concerned are the extrema of $\theta$ over all of $L$.)

1434: 

1435: \begin{Conj}

1436: If $L$ satisfies either of the conditions (\ref{close}) or

1437: (\ref{vclose}) then mean curvature flow for $L$ exists for all time

1438: and converges to a special Lagrangian in its hamiltonian deformation

1439: class; the unique SLag conjectured in \emph{\cite{Th}}.

1440: \end{Conj}

1441: 

1442: It is of course a consequence of this and the conjecture in \cite{Th}

1443: that some hamiltonian deformation of $L$ satisfies (\ref{close}) if

1444: and only if it is stable. The SLag should also be unique in its

1445: hamiltonian deformation class as in Theorem \ref{!}.

1446: If $L$ is stable but not close enough to

1447: being SLag that (\ref{close}) fails, then mean curvature flow

1448: can become singular in finite time, (locally) splitting the Lagrangian

1449: in the reverse of a connect sum operation (i.e. with a vanishing cycle

1450: which is an $S^{n-1}$, or an $S^{n-r}$-bundle over an

1451: $(r-1)$-dimensional base in the relative connect sum case). We might

1452: then conjecture that the resulting pieces are smooth so we can

1453: begin the process again until we get a

1454: decomposition into different phase SLags. Typically, in the simplest

1455: case, we would get $L_1\cup L_2$ (with $\phi(L_1)<\phi(L_2)$ by

1456: stability) which is not a hamiltonian deformation of $L$ (though it

1457: is in the closure of such deformations).

1458: 

1459: If $L$ is unstable, we would again expect such finite time

1460: singularities and SLag splittings. But if $L$'s phase variation,

1461: or volume, is sufficiently small, we can hope for convergence

1462: to the Jordan-H\"older decomposition of Section \ref{JH}. That is,

1463: while the volume of $L$ must be larger than the cohomological volume

1464: of its Jordan-H\"older decomposition, if it is less than any other

1465: decomposition then it can only flow to the former. Again we expect

1466: the flow to become singular in finite time, the limit

1467: (locally) splitting $L$ into pieces for which we restart the flow.

1468: This splitting of the Lagrangian is a manifestation of the well

1469: known finite-time dumb-bell singularities in mean curvature flow.

1470: 

1471: 

1472: \subsection*{Proof for our example}

1473: 

1474: Under our flow (\ref{result!}) in the Shapere-Vafa example

1475: the proofs of the evolution equations for the phase

1476: function $\theta$ (\ref{thetadot}) and volume (\ref{voldot})

1477: show that the equations are modified (as the metric is not quite

1478: Ricci-flat) to

1479: \begin{eqnarray}

1480: {d\over dt}\theta\!&=&\!-\Delta\,\theta+{\langle d\theta,d|\Omega|

1481: \rangle\over|\Omega|}, \label{newdot} \\

1482: {d\over dt}(|\Omega|\vol_L)\!&=&\!-|d\theta|^2(|\Omega|\vol_L).

1483: \label{nvdot}

1484: \end{eqnarray}

1485: when $|\Omega|=|\Omega/\!\vol\!|$ is not $\equiv1$.

1486: Therefore the maximum principle still holds for $\theta$, and

1487: the condition (\ref{close}) is again preserved by the flow. Similarly

1488: if we measure volume with respect to $|\Omega|\vol_L$ then this

1489: is decreasing and (\ref{vclose}) is preserved by the flow. We can

1490: now prove the appropriate version of our conjecture in this example.

1491: From the proof it will also be clear that the original conjecture

1492: could be proved in this case in the $O(n)$-invariant Ricci flat

1493: metric if we knew it,

1494: we would just not be able to be as explicit about the flow equations.

1495: 

1496: \begin{Theorem} \label{faq}

1497: Suppose that $\gamma$ is a curve in $\C$, with endpoints at zeros of

1498: $p$, and otherwise missing the zeros of $p$, such that its

1499: pointwise phase $\theta$ (\ref{pf}) satisfies (\ref{close})

1500: for all Lagrangians $L_i=\gamma^n_i,\ i=1,2,$ fibred over curves

1501: $\gamma_i$ in the base, and also $S-I:=\sup_\gamma\theta-

1502: \inf_\gamma\theta<2\pi/3$. Then the flow (\ref{result!}) exists for all

1503: time and converges in $C^\infty$ to a smooth curve whose phase

1504: function (\ref{pf}) is constant.

1505: \end{Theorem}

1506: 

1507: We break the proof up into existence of the flow (best dealt with

1508: at the level of $\gamma^n\subset X^n$), controlling the angle

1509: variation (using $\gamma\subset\C$) to ensure no $180^o$ kinks

1510: appear in $\gamma$, and using this to show the

1511: flow exists for all time (for which we use the double cover

1512: $\gamma^1\subset X^1$

1513: and $\theta$ on $\gamma^n$). We follow \cite{An}, in parts heavily

1514: modified to take care of the endpoints of $\gamma$. Finally

1515: we will show that the flow converges to a SLag.

1516: 

1517: \begin{Lemma} \label{exist}

1518: The flow (\ref{result!}) exists while the curvature of $\gamma^1$

1519: is bounded.

1520: \end{Lemma}

1521: 

1522: \begin{Proof}

1523: Firstly, short term existence of the flow, given any initial curve

1524: $\gamma\subset\C$ missing the zeros of $p$ except at its endpoints

1525: and \emph{such that $\gamma^1$ is $H^{2+\alpha}$} for some

1526: $\alpha>0$ (i.e. $\gamma^1$ has H\"older continuous curvature),

1527: is in fact most easily proved at the level of the $H^{2+\alpha}$

1528: Lagrangian $\gamma^n$; see \cite{An} for the method in 1 dimension

1529: (which easily generalises to $n$ dimensions), and \cite{Ch} for

1530: a similar $n$-dimensional result. This is also done

1531: in (\cite{Sm} Proposition 1.6) using results of Hamilton \cite{H},

1532: for instance.

1533: 

1534: While the curvature of the curve $\gamma^1\subset X^1$ is bounded,

1535: so is the norm of the flow vector (the last term in (\ref{result!})

1536: is always bounded, and the second term can be bounded by the

1537: curvature at an intermediate point by Taylor's theorem). So at

1538: any finite time $T$ the flow converges to a limit curve $\gamma_T^1$

1539: pointwise. Parametrising the curves by their arclength on $X^1$,

1540: their first and second derivatives as maps to $X^1$ are therefore

1541: bounded, which

1542: by Arzel\`a-Ascoli implies that for a subsequence of $t$ we have

1543: convergence in $C^1$ to a $C^1$ curve with bounded (weak)

1544: curvature. By the uniqueness of the limit, then, $\gamma_T^1\subset

1545: X^1$ has bounded curvature.

1546: 

1547: Bounds on (the derivative of) the phase of $\gamma^1$ give

1548: corresponding bounds on (the derivative of) the phase of

1549: $\gamma^n$ (via (\ref{result!}) for $n$ and $n=1$). So the phase

1550: function $\theta$ of $\gamma^n$ is also $C^0$ convergent to the

1551: phase of $\gamma_T^n$, and satisfies the parabolic equation

1552: (\ref{newdot}). Putting this into local coordinates and

1553: differentiating with respect to arclength $s$, we get a

1554: uniformly parabolic equation with bounded coefficients and a

1555: bounded solution $\theta_s$ on $t\in[0,T]$. By (\cite{LSU}

1556: Section III Theorem 10.1), then, $\theta_s$ is

1557: in fact $\alpha$-H\"older continuous for some $\alpha>0$,

1558: and $\gamma^1_T$ is $H^{2+\alpha}$. By the existence of the flow

1559: for $H^{2+\alpha}$ initial conditions, then, the flow

1560: exists for some time $t>T$.

1561: \end{Proof}

1562: 

1563: \begin{Lemma} \label{cone}

1564: $\limsup_{|s-s'|\to0}|\theta(\gamma'(s,t))-\theta(\gamma'(s',t))|<\pi$

1565: for all time $t$ for which the flow exists, where $s$ is arclength

1566: along $\gamma(\,.\,,t)$, and $\theta(\gamma')$ is the argument

1567: of $\gamma'\in\C$.

1568: \end{Lemma}

1569: 

1570: \begin{Proof}

1571: Working outside a fixed neighbourhood of the zeros of $p$ at the

1572: endpoints of $\gamma$, this follows from the bounds

1573: on $\theta=\theta(\gamma')+(n/2-1)\theta(p(\gamma))$ (\ref{pf})

1574: coming from the $\sup\theta-\inf\theta<2\pi/3$ assumption

1575: (preserved under the flow by the maximum principle (\ref{newdot})),

1576: as the variation of $\theta(p(\gamma))$ can be made arbitrarily small

1577: with $|s-s'|$. Since $\gamma$ must stay at a bounded distance from

1578: other zeros of $p$ by the condition (\ref{close}) and the

1579: maximum principle for $\theta$ (\ref{newdot}), we are left with

1580: proving the lemma in an arbitrarily small neighbourhood in $\C$

1581: of the endpoints of $\gamma$.

1582: 

1583: Unfortunately, in this region, the bounds we want for

1584: $\theta(\gamma')$ do not follow from the

1585: bounds we have for $\theta$, and in fact come only from

1586: comparison with known solutions. Draw the SLags of phase $S,\,I$

1587: emanating from a zero of $p$, i.e. the curves in $\C$ solving

1588: $\theta(\gamma')+(n/2-1)\theta(p(\gamma))=S$ or $I$.

1589: (Since this is an ODE, there is no problem in finding solutions

1590: and extending them to either infinity or another zero of $p$; see

1591: \cite{SV}.) In the tangent space to the zero of $p$ this gives a

1592: cone of angle $(S-I)/n<2\pi/3n$ which $\gamma$ lies inside and

1593: cannot cross either at $t=0$ or any later time in the flow. So in

1594: a sufficiently small neighbourhood of an endpoint of $\gamma$,

1595: we may bound $\theta(\gamma)$ inside a cone of angle less than

1596: $2\pi/3n$, and also take $\theta(p(\gamma))$ to be within any given

1597: $\epsilon$ of $\theta(\gamma)+C$ (since the zero

1598: of $p$ is nondegenerate; here $C$ is the phase of $\dot p$ at the

1599: zero of $p$). Thus $\theta(p(\gamma))$ can be

1600: bounded inside a similar cone, so that (\ref{pf}) bounds the

1601: variation of $\theta(\gamma')$ by $2\pi/3+(n/2-1)2\pi/3n<\pi$.

1602: 

1603: The bounds on $\theta(\gamma')$ imply that the curve does not

1604: spiral round its endpoints but moves away from them with nonzero

1605: derivative inside the above cone until it is outside the small

1606: neighbourhood employed above. So the remaining case to consider

1607: is if the curve can pass arbitrarily close to one of its own

1608: endpoints at some bounded-below arclength from its endpoints,

1609: i.e. if the cone of SLags above starting from a zero of $p$

1610: passes either side of that same zero at some nonzero arclength.

1611: Then there would be a Slag fibred over a curve starting and

1612: ending at the same root of $p$, with our Lagrangian $\gamma^n$

1613: the connect sum of this SLag and some other $\gamma_1^n$.

1614: But $\gamma^n$ cannot flow arbitrarily close to such a connect sum

1615: as its phase variation would approach at least

1616: the difference between the phase of the SLag and $\phi(\gamma^n_1)$,

1617: contradicting (\ref{close}).

1618: \end{Proof}

1619: 

1620: By Lemma \ref{exist} the flow exists for all time unless, as we

1621: suppose now, the curvature of $\gamma^1$ becomes unbounded in finite

1622: time. Then to get a contradiction we start by scaling as in \cite{An}.

1623: Pick $s_i,\,t_i,\ i=1,2,\ldots$ such

1624: that $t_i$ tends to the blow up time and the curvature $\kappa_i$ of

1625: $\gamma^1(s_i,t_i)=y_i$ is maximal

1626: over the curvatures of $\gamma^1(s,t)$ for all $s$, and all

1627: $t\le t_i$ (here we parameterise by arclength $s$ on $X^1$, centred

1628: at a zero $z$ of $p$, i.e. $\gamma^1\res{s=0}=z$ lies over an

1629: endpoint of $\gamma$).

1630: 

1631: How we handle the blow up depends on

1632: whether it happens at the branch points of $\gamma^1$ (i.e.

1633: the endpoints of $\gamma$), by

1634: which we mean $|s_i|=O(|\kappa_i^{-1}|)$, or in the interior

1635: $|s_i|\gg|\kappa_i^{-1}|$, due to the different nature of

1636: (\ref{result!}) at the branch points.

1637: We first deal with the interior where the flow is a perturbation

1638: of mean curvature flow and so can be handled by \cite{An}:

1639: 

1640: \begin{Lemma}

1641: Supposing that the curvature blows up as above, then

1642: $|s_i\kappa_i|$ is bounded.

1643: \end{Lemma}

1644: 

1645: \begin{Proof}

1646: Firstly, if after passing to a subsequence of $i\in\mathbb N$,

1647: and centring $s$ about the other branch point of $\gamma^1$

1648: if necessary, the blow up

1649: occurs at a finite distance $|s_i|>\epsilon>0$ from either branch

1650: point of $\gamma^1$ 

1651: then in this interior the flow (\ref{result!}) is a finite

1652: perturbation of mean curvature

1653: flow satisfying the conditions of \cite{An}, so a $180^o$ kink

1654: must appear in $\gamma^1$, contradicting Lemma \ref{cone}. So

1655: we need only deal with the case of $s_i\to0$ (by passing to a

1656: subsequence to concentrate around one of the two branch

1657: points, if necessary) while $r_i:=|s_i\kappa_i|\to\infty$.

1658: 

1659: Then we rescale as in \cite{An};

1660: \begin{equation} \label{scale}

1661: s\mapsto \kappa_is, \quad g\mapsto \kappa_ig, \quad

1662: t\mapsto \kappa_i^2(t-t_i),

1663: \end{equation}

1664: where $g$ is the metric on $X^1$. This rescaled flow for

1665: $\gamma^1_i$ has the same form as (\ref{result!}),

1666: $$

1667: \dot\gamma_i^1=\MCV-(n-1)\mathbf n(\log|p|_i^{1/2})\,\mathbf

1668: n+{1\over2}\mathbf n(\log(|p|_i+|\dot p|_i^2/4))\,\mathbf n,

1669: $$

1670: but with $|p|$ and $|\dot p|$ replaced by their pullbacks

1671: $|p|_i,\ |\dot p|_i$ to

1672: the new Riemannian surface (here it is important that $\dot p$

1673: is still computed in the old coordinates, \emph{then} pulled

1674: back). Therefore their gradients are scaled by $\kappa_i^{-1}$.

1675: $\mathbf n$ denotes the unit normal to $\gamma^1$ in the

1676: new metric on $X^1$. The curvature gets scaled by

1677: $\kappa_i^{-1}$ and so has a maximum, over $t\le0$, of 1 at

1678: $y_i$ (at time $t=0$). We want to show that the two perturbation

1679: terms on the right hand side of the above flow tend to zero as

1680: $i\to\infty$.

1681: 

1682: In the rescaled variables, work in a geodesic disc in $X^1$ of

1683: radius $r_i/2$ ($r_i=|\kappa_is_i|\to\infty$ as $i\to

1684: \infty$) about $y_i$. As $s_i\to0$, for

1685: $i$ sufficiently large this is within an arbitrarily small

1686: neighbourhood of $z$ in the original metric,

1687: in which $\gamma'$ varies within an angle $<\pi$ cone

1688: as in the proof of Lemma \ref{cone}, i.e. $(\gamma^1)'$

1689: varies within an angle $<\pi/2$ cone on the double cover

1690: $X^1$. So arclength $s$ on $\gamma^1$ and radial distance $r$

1691: in $X^1$ are equivalent metrics on $\gamma^1$ in this disc;

1692: $r_s:=\partial r/\partial s$ and $s_r=\partial s/\partial r$

1693: are both bounded.

1694: 

1695: As $y_i\in\gamma_i^1$ is of arclength $s_i\ge r_i$ from $z$ (the

1696: zero of $p$) at $s=0$, we deduce that all points of our disc

1697: are of distance $\ge cr_i/2$ from the zero of $p$ (for some

1698: constant $c>0$ fixed for all $i\gg1$) in the new metric.

1699: Thus, for $i$ large enough, we have

1700: $$

1701: |p|_i^{1/2}\ge C\kappa_i^{-1}(cr_i/2),

1702: $$

1703: where $C$ is a constant just less than

1704: the norm of the derivative of $p^{1/2}$ at the zero $z$

1705: in the original metric on $X^1$ ($p^{1/2}$ pulls back

1706: to a well defined function on $X^1$ with a simple zero

1707: at $z$). We can therefore bound

1708: $$

1709: |\dot\gamma_i^1-\MCV|\le(n-1){\kappa_i^{-1}\sup|d(p^{1/2})|

1710: \over C\kappa_i^{-1}(cr_i/2)}+{1\over2}\kappa_i^{-1}

1711: \sup|d\log(|p|+|\dot p|^2/4)|,

1712: $$

1713: where both sups are taken over small neighbourhoods of

1714: $z$ \emph{in the original metric on} $X^1$. As $i\to\infty$,

1715: $\kappa_i,\,r_i\to\infty$, so the above bound tends to zero,

1716: while the radius of the disc we are working on

1717: $r_i/2\to\infty$. It follows that in the limit we get

1718: mean curvature flow of a curve inside an infinite flat

1719: disc $\R^2$; see (\cite{An} Section 9) for how to pass to the

1720: limit to conclude that for this

1721: blow up to occur a $180^o$ kink must appear in the curve

1722: $\gamma^1$ (by which we mean the limsup in Lemma \ref{cone}

1723: is $\ge\pi$). But this contradicts Lemma \ref{cone}.

1724: 

1725: (We do not repeat Angenent's argument here as we will

1726: give a slightly harder, $n$-dimensional, version of

1727: it around the endpoints of $\gamma$ in Lemma \ref{blowup}

1728: below. The point is just that in the rescaling we can

1729: get rid of the last two terms of our flow to reduce to the

1730: results of \cite{An}.)

1731: \end{Proof}

1732: 

1733: The remaining case

1734: we must dismiss is that of $\kappa_i$ blowing up at points

1735: $y_i=\gamma^1(s_i,t_i)$ with $|s_i|<A/|\kappa_i|$ for some

1736: fixed $A$. Here we must work harder than \cite{An}.

1737: 

1738: \begin{Lemma} \label{blowup}

1739: The curvature of $\gamma^1$ does not blow up in finite time.

1740: \end{Lemma}

1741: 

1742: \begin{Proof}

1743: By Lemma \ref{cone} we know that there is an $A$ such that

1744: $|s_i|<A/|\kappa_i|$.

1745: We rescale variables as in (\ref{scale}), and work on

1746: a length $\kappa_i^{1/2}\to\infty$ interval (in the new metric)

1747: on $\gamma^1$ centred ($s=0$) at the zero $z$ of $p$. This is

1748: contained inside the ball of radius $\kappa_i^{-1/2}\to0$

1749: about $z$ in $X^1$ in the original metric, so for $i$ sufficiently

1750: large we can assume that $p(t)-Ct$ is arbitrarily small

1751: \emph{in $C^2$ norm} (here $t\in\C$ is the base parameter,

1752: \emph{not} time, $C=\dot p(z)$, and the same is true of any $C^r$

1753: norm; $r=2$ is the case of interest for us). We start by obtaining

1754: bounds on the polar angle of the curve and its tangent vector.

1755: We shall confuse functions on $\C$ with their pullbacks to $X^1$

1756: (so writing things like $p(\gamma^1)$ etc.).

1757: 

1758: Taking $i$ sufficiently large that the metric on the

1759: radius $\kappa_i^{1/2}$ disc about $z$ in $X^1$ is sufficiently

1760: close to being flat, define geodesic polar coordinates on $X^1$,

1761: $r^1:=|\gamma^1|,\ \theta^1:=\theta(\gamma^1)$ (which is

1762: $\theta(\gamma)/2$ to within a constant). Then we can

1763: assume that $\theta^1_s$ is arbitrarily $C^1$ close to

1764: \begin{equation} \label{sin}

1765: {1\over r}\sin(\theta(\gamma^1_s)-\theta^1),

1766: \end{equation}

1767: which is the exact formula for a flat metric and polar

1768: coordinates. This bounds $|r\theta^1_s|$. Since by

1769: construction the curvature of $\gamma^1$ is not more than

1770: one, i.e. $|(\theta(\gamma^1_s))_s|\le1$, we can bound

1771: $|\theta(\gamma^1_s)|\le s$.

1772: 

1773: Note that (\ref{sin}) in flat space gives us the differential equation

1774: $$

1775: f_s=\kappa_i-{\sin f\over r}

1776: $$

1777: for $f=\theta(\gamma^1_s)-\theta^1$, with $f(0)=0$ and $|\kappa_i|\le1$.

1778: This implies that $|f(s)|\le|s|$ (consider a point where the graph

1779: of $f$ crosses that of $\pm s$, where $|f_s|\ge1$, for a contradiction),

1780: so for $i$ sufficiently large that our polar coordinates are sufficiently

1781: close to flat coordinates we can deduce a bound on $|\theta(\gamma^1_s)-

1782: \theta^1|/s$. Thus $\theta^1/s$ is also bounded, and $\theta^1/r$ by

1783: the uniform comparison bounds of $r$ and $s$ given by the cone argument

1784: in Lemma \ref{cone}.

1785: 

1786: Instead of considering the equation (\ref{result!}) for $\gamma^1$,

1787: we analyse the equation (\ref{newdot}) for $\theta$ on $\gamma^n$.

1788: After rescaling it becomes

1789: \begin{equation} \label{eqn}

1790: \dot\theta=\Delta\theta+{\langle d\theta,d(|\Omega|_i)\rangle\over

1791: |\Omega|_i},

1792: \end{equation}

1793: where $|\Omega|_i$ is the pullback of $|\Omega|:=|\Omega/\!\vol\!|$ to

1794: $\gamma^1$ with

1795: its new metric. Note also that pulling functions up from $\gamma$ and

1796: taking their exterior derivative $d$ on either $\gamma^1$ or on

1797: $\gamma^n$ gives the same result via the obvious inclusion

1798: $\gamma^1\subset\gamma^n$ commuting with the projections to $\gamma$.

1799: Again we want to control this evolution equation as $i\to\infty$.

1800: 

1801: $|d\theta|=|\theta_s|=|\partial_s[\theta(\gamma^1_s)/2+(n/2-1)

1802: \theta(p(\gamma^1))]|$, so this is bounded by the estimates above,

1803: for $i$ sufficiently large that $\theta(p(\gamma^1))$ is $C^1$

1804: close to $\theta^1/2+$\,const. in the disc in which we are working.

1805: So we can bound the last term in (\ref{eqn}) by a constant times

1806: $$

1807: \kappa_i^{-1}{\sup|d\Omega|\over\inf|\Omega|},

1808: $$

1809: where the sup and inf are taken in the original metric over a small

1810: neighbourhood of $(0,\ldots,0,z)\in X^n$. This tends to zero as

1811: $i\to\infty$.

1812: 

1813: Computing the Laplacian on the space $\gamma^n$, with a radial

1814: coordinate $s$ and rotational symmetry about the origin $s=0$,

1815: makes (\ref{eqn})

1816: \begin{equation} \label{tat}

1817: \theta_t=\theta_{ss}+(n-1){R^i_s\over R^i}\theta_s+O(\kappa_i^{-1}),

1818: \end{equation}

1819: where $R^i=R^i(s)$ is the radius of the sphere $S^{n-1}$ at $s$ in the

1820: new metric.

1821: 

1822: To compute $R^i$, we use our $\gamma^1\subset X^1$ arclength coordinate

1823: $s$, the radial coordinate $r$ on $X^1$, and a radial coordinate

1824: $\rho$ on $\C$. For $i$ sufficiently large, for $s\le\kappa_i^{1/2}$

1825: in the new metric, we can approximate $p$ linearly about $z$ and

1826: so assume that $R^i$ is as close as we like

1827: to $\kappa_i\sqrt{|\dot p(z)|\rho}$ in $C^2$. Therefore $R^i_r$ is

1828: approximated by

1829: \begin{equation} \label{grig}

1830: R^i_r=\frac{R^i_\rho}{r_\rho}=\frac{R^i_\rho}{\sqrt{\kappa_i^2+(R^i_\rho)^2}}

1831: =\frac{1}{\sqrt{1+\frac{\kappa_i^2}{(R^i_\rho)^2}}}\approx

1832: \frac{1}{\sqrt{1+\frac{4\rho}{|\dot{p}(z)|}}}\ ,

1833: \end{equation}

1834: which is bounded and tends to $1$ in the interval

1835: $s\in[0,\kappa_i^{1/2})$. Similarly $R^i_{rr}$ can be taken to be

1836: arbitrarily small on the same interval, for $i$ sufficiently large.

1837: 

1838: Since $r_s$ is bounded, this gives bounds on $R^i_s$, implying that,

1839: on passing to a subsequence if necessary, the functions $R^i(s)$

1840: are convergent as $i\to\infty$ by the Arzel\`a-Ascoli theorem.

1841: 

1842: Note also that for all $i$, $R^i_s(0)=1$. But to preserve this

1843: in the limit, we must similarly bound $R^i_{ss}$. Differentiating

1844: $r^2_s+r^2(\theta^1_s)^2=1$ and $R^i_s=R^i_rr_s$ gives

1845: \begin{equation} \label{final}

1846: R^i_{ss}=R^i_{rr}r_s^2-R^i_r(r(\theta^1_s)^2+

1847: r^2\theta^1_s\theta^1_{ss}/r_s).

1848: \end{equation}

1849: We have bounded $R^i_{rr},\ r_s,\ r_s^{-1},\ r\theta^1_s$ and

1850: $\theta^1_s$; all this leaves is the last term in (\ref{final}).

1851: 

1852: We have approximated $\theta^1_s$ in $C^1$ by

1853: ${1\over r}\sin(\theta(\gamma^1_s)-\theta^1)$; differentiating

1854: approximates $r^2\theta^1_s\theta^1_{ss}/r_s$ as closely as

1855: we like (as $i\to\infty$) to

1856: $$

1857: {r\theta^1_s\over r_s}(\kappa_i-\theta^1_s)\cos(\theta(\gamma^1_s)

1858: -\theta^1)-\theta^1_s\sin(\theta(\gamma^1_s)-\theta^1),

1859: $$

1860: which we have bounded already.

1861: In conclusion, after passing to a subsequence if necessary, $R^i$ is

1862: $C^1$ convergent to some $R$ with $R_s(0)=1$, and the phase function

1863: $\theta^\infty$ of the limit curve $\gamma^1_\infty$

1864: (which exists by Arzel\`a-Ascoli

1865: since $|\gamma^1_s|=1$ and $|\gamma^1_t|$ is bounded by the bound on

1866: its curvature $\kappa$) satisfies the limit of (\ref{tat}):

1867: \begin{equation} \label{lap}

1868: \theta^\infty_t=\theta^\infty_{ss}+(n-1){R_s\over R}\theta^\infty_s.

1869: \end{equation}

1870: 

1871: But this is just the heat equation for $\theta$ on $\R^n$ with radial

1872: coordinate $s$ and the $O(n)$-invariant metric in which the radius of

1873: the $S^{n-1}$ fibre over $s$ is $R(s)$. By construction of the time

1874: rescaling (\ref{scale}) it exists for all time $t\le0$, and the

1875: solution $\theta^\infty$ is bounded. Also the metric is uniformly

1876: elliptic compared to the flat metric, as $R_r=1$ (\ref{grig}) and

1877: $r_s$ is bounded above and below away from zero. Therefore, by

1878: Moser's Harnack inequality \cite{Mo}, $\theta^\infty$ is in fact

1879: constant.

1880: 

1881: But for all $i$, max\,$\theta_s=1$ by construction, and passing to

1882: a subsequence if necessary the point

1883: where the maximum is obtained is convergent. To show then that

1884: max\,$\theta^\infty_s=1$, to get our contradiction, we

1885: need only know that $\theta_s$ is, say, \emph{uniformly (in $i$)}

1886: H\"older continuous. This is again a consequence of (\cite{LSU}

1887: Section III Theorem 10.1) as follows.

1888: By the boundedness of $\theta_s$ and the parabolic

1889: equation it satisfies (different for each $i$), $\theta_s$ is

1890: in fact $\alpha$-H\"older continuous for some $\alpha>0$, with

1891: $H^\alpha$ norm bounded by the bounds on the coefficients of

1892: the parabolic equations. But these are bounded uniformly in $i$,

1893: as a glance at (\ref{eqn}) confirms: the correction term tends to

1894: zero, and the Laplacian term (and its derivative with respect to

1895: $s$) is controlled by $C^2$ bounds on the metric which we provided

1896: above by bounding $R^i(s),\,R^i_s$ and $R^i_{ss}$.

1897: \end{Proof}

1898: 

1899: Finally we show that this infinite time flow converges using

1900: standard techniques (see for instance \cite{C} for a harder result).

1901: Notice that the same scaling proof (\ref{blowup}) that the

1902: curvature of $\gamma^1$ does not blow up in finite time shows the

1903: same for our now infinite time flow. So, using the $O(n)$ symmetry,

1904: the curvature of the metric on $\gamma^n$ stays uniformly bounded,

1905: and we have a $C^1$ bound on $\theta$. Therefore in the equation

1906: (\ref{newdot}) for $\theta$ on $\gamma^n$, which we rewrite as

1907: \begin{equation} \label{ntd}

1908: \dot\theta=-\Delta^\Omega\theta:=|\Omega|^{-1}*d(|\Omega|*d\theta),

1909: \end{equation}

1910: the coefficients have at least uniform $C^1$ bounds; $\theta$ then

1911: acquires a uniform $C^3$ bound by parabolic Schauder estimates (see

1912: \cite{LSU} III Theorem 12.1, for instance). Again by $O(n)$ symmetry

1913: we now get uniform $C^2$ bounds on $\gamma^n$'s curvature. And so it

1914: goes on, giving $C^\infty$ bounds and allowing us to extract a

1915: subsequence of times for which the flow converges in $C^\infty$ to

1916: a Slag.

1917: 

1918: To see that the flow converges without having to pass to

1919: a subsequence we need only show convergence of $\theta$ in $L^2$;

1920: this way no other subsequence of the flow can converge to

1921: a different limit in $C^\infty$. In fact we use an $L^2$-norm

1922: weighted by $|\Omega|$, and compute using (\ref{ntd})

1923: and (\ref{nvdot}):

1924: $$

1925: {d\over dt}\int_{\gamma^n}(\theta-\bar\theta)^2|\Omega|\vol=

1926: \int_{\gamma^n}\left\{2(\theta-\bar\theta)(-\Delta^\Omega\theta-

1927: {d\over dt}\bar\theta)-(\theta-\bar\theta)^2|d\theta|^2\right\}

1928: |\Omega|\vol,

1929: $$

1930: where $\bar\theta=\int\theta|\Omega|\vol/\int|\Omega|\vol$ is

1931: constant on $\gamma^n$ (but not in time). This is then bounded

1932: above by

1933: $$

1934: -2\int(\theta-\bar\theta)\Delta^\Omega(\theta-\bar\theta)

1935: |\Omega|\vol.

1936: $$

1937: It is easily checked that $\Delta^\Omega=d_{\,}^{*^\Omega}

1938: d$, where $d_{\,}^{*^\Omega}$ is the adjoint of $d$ with

1939: respect to the $L^2$-metric $\int_{\gamma^n}\langle\ \cdot\ ,

1940: \ \cdot\ \rangle|\Omega|\vol$ we are using. So its kernel is just

1941: the constants, and by the uniform $C^\infty$ bounds on the metric

1942: of $\gamma^n$ and $|\Omega|$ we can get a uniform lower bound

1943: $\lambda>0$ for its first nonzero eigenvalue. This then gives

1944: a bound $\int(\mu\Delta\mu)|\Omega|\vol\ge\lambda\int\mu^2|\Omega|\vol$

1945: for functions $\mu$ of integral zero. Setting $\mu=\theta-\bar\theta$

1946: gives

1947: $$

1948: {d\over dt}\int(\theta-\bar\theta)^2|\Omega|\vol\le-2\lambda

1949: \int(\theta-\bar\theta)^2|\Omega|\vol,

1950: $$

1951: which therefore tends to zero as required. Theorem \ref{faq} is proved. \\

1952: 

1953: We end by noting that one can get cone-type bounds similar to

1954: those of Lemma \ref{cone} on the tangent

1955: direction $\theta(\gamma^1_s)$ even as a curve $\gamma$

1956: representing an \emph{unstable} Lagrangian $\gamma^n$ approaches

1957: and breaks across a zero of $p$. Suppose that the initial phase

1958: variation of $\gamma^n$ is, without loss of generality,

1959: in some $(-\delta,\delta)$. Draw the cone with boundary the

1960: SLags emanating from the zero of $p$ with phase $-\delta,\ \pi+

1961: \delta$ (the straight lines emanating from the zeros of $p$ drawn

1962: in Figures 2 and 4 display the $\delta=0$ cone for dimensions 2

1963: and 3 respectively). Then in a

1964: sufficiently small neighbourhood of the zero, the variation in

1965: $\theta(p)$ can be taken to be less than $2\pi/n+2\delta+\epsilon$

1966: for any $\epsilon>0$. Since we are free to make $\delta$ slightly

1967: smaller without violating the initial bounds on phase, we can

1968: ensure that there is a neighbourhood of the zero of $p$, and a cone

1969: with vertex at $p$ whose walls $\gamma$ cannot cross, such that for

1970: the part of $\gamma$ lying in this neighbourhood, the variation

1971: $$

1972: \sup\theta(p(\gamma))-\inf\theta(p(\gamma))<2\pi/n+2\delta. 

1973: $$

1974: Comparing with (\ref{pf}) gives

1975: $$

1976: \sup\theta(\gamma')-\inf\theta(\gamma')<(n/2-1)(2\pi/n+2\delta)+

1977: 2\delta=(1-2/n)\pi+n\delta

1978: $$

1979: so that again no $180^o$ kinks can occur while this is less than

1980: $\pi$, i.e. for $\delta\le2\pi/n^2$. So again the analysis should

1981: be tractable in this case (more general spiralling around a zero

1982: would make matters worse). However, we have not carried out the

1983: analysis necessary to show that at the moment $\gamma$ reaches

1984: the zero of $p$ it is sufficiently smooth that the two resulting

1985: curves it splits into give $C^2$ Lagrangians (whose flow we could

1986: restart). \\

1987: 

1988: Of course by just studying the simple examples above we cannot

1989: hope to know how bad the singularities are that arise in finite

1990: time in the general case.

1991: Also, as mentioned in \cite{Th}, we should perhaps restrict to

1992: those Lagrangians whose Floer cohomology is well defined \cite{FO3}.

1993: This includes all homology spheres, however.

1994: 

1995: We should also point out the obvious fact that most of the

1996: evidence for our conjecture, other than perhaps the mirror

1997: symmetry and study of Joyce's examples in \cite{Th}, has been

1998: essentially one-dimensional (either

1999: for $T^2$ in \cite{Th}, or by symmetry reduction in the examples

2000: above). This is unrepresentative, essentially because the angles at

2001: which Lagrangians intersect (the $\alpha_i$s of (\ref{l1}))

2002: are all the same in this situation, and so are determined by the

2003: phase (their sum). So interesting phenomena, where degrees in

2004: Floer cohomology change (e.g. a Hom becomes an Ext$^i$ on the mirror

2005: while the phase remains fixed) are largely lost due to them being

2006: controlled entirely by the phase.

2007: 

2008: 

2009: \begin{thebibliography}{FO3}

2010: 

2011: \bibitem[An]{An} S. Angenent,

2012: \textit{Parabolic equations for curves on surfaces. I. Curves

2013: with $p$-integrable curvature.} Ann. of Math. (2) \textbf{132}

2014: (1990), 451--483.

2015: 

2016: \bibitem[C]{C} H. D. Cao,

2017: \textit{Deformation of K\"ahler metrics to K\"ahler-Einstein metrics on

2018: compact K\"ahler manifolds.} Invent. Math. \textbf{81} (1985),

2019: 359--372.

2020: 

2021: \bibitem[Ch]{Ch} K.-C. Chang,

2022: \textit{Heat flow and boundary value problem for harmonic maps.}

2023: Ann. Inst. H. Poincar\'e Anal. Non Lin\'eaire \textbf{6} (1989),

2024: 363--395.

2025: 

2026: \bibitem[FO3]{FO3} K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono,

2027: \textit{Lagrangian intersection Floer theory -- anomoly and obstruction},

2028: preprint.

2029: 

2030: \bibitem[GH]{GH}

2031: P. Griffiths and J. Harris,

2032: \textit{Principles of algebraic geometry.} Wiley, New York. 1978.

2033: 

2034: \bibitem[Gr]{Gr} M. Grayson,

2035: \textit{Shortening embedded curves},

2036: Ann. of Math. \textbf{129} (1989), 71--111.

2037: 

2038: \bibitem[H]{H}

2039: R. S. Hamilton,

2040: \textit{Three manifolds with positive Ricci curvature},

2041: Jour. Diff. Geom. \textbf{23} (1986), 69--96.

2042: 

2043: \bibitem[HaL]{HaL} F. R. Harvey and H. B. Lawson,

2044: \textit{Calibrated Geometries}, Acta Math. \textbf{148} (1982), 47--157.

2045: 

2046: \bibitem[HuL]{HuL}

2047: D. Huybrechts and M. Lehn,

2048: \textit{Geometry of moduli spaces of shaves.}

2049: Aspects in Mathematics Vol. E31, Vieweg. 2000.

2050: 

2051: \bibitem[KS]{KS}

2052: M. Khovanov and P. Seidel,

2053: \textit{Quivers, Floer cohomology, and braid group actions},

2054: J. Amer. Math. Soc. \textbf{15} (2002), 203--271.

2055: 

2056: \bibitem[L]{L} G. Lawlor,

2057: \textit{The angle criterion}, Inv. Math. \textbf{95} (1989), 437--446.

2058: 

2059: \bibitem[LSU]{LSU} O. A. Ladyzhenskaya, V. A. Solonikov and

2060: N. N. Ural'ceva,

2061: \textit{Linear and quasilinear equations of parabolic type},

2062: Transl. Math. Monographs \textbf{23}, AMS. 1968.

2063: 

2064: \bibitem[Mi]{Mi} J. Milnor,

2065: \textit{Lectures on the h-cobordism theorem}. Princeton University

2066: Press. 1965.

2067: 

2068: \bibitem[Mo]{Mo} J. Moser,

2069: \textit{A Harnack inequality for parabolic differential equations},

2070: Comm. Pure Appl. Math. \textbf{17} (1964) 101--134.

2071: 

2072: \bibitem[N]{N} D. Nance,

2073: \textit{Sufficient conditions for a pair of planes to be area minizing},

2074: Math. Ann. \textbf{279} (1989) 161--164.

2075: 

2076: \bibitem[Oh1]{Oh1} Y.-G. Oh,

2077: \textit{Mean curvature vector and symplectic topology of Lagrangian

2078: submanifolds in Einstein-K{\"a}hler manifolds}, Math. Z. \textbf{216}

2079: (1994), 471--482. 

2080: 

2081: \bibitem[Oh2]{Oh2} Y.-G. Oh,

2082: \textit{Floer cohomology, spectral sequence and the Maslov class of

2083: Lagrangian embeddings}, Internat. Math. Res. Notices \textbf{7} (1996),

2084: 305--346.

2085: 

2086: \bibitem[P]{P} M. Po\'zniak,

2087: \textit{Floer homology, Novikov rings and clean intersections}, Ph.\,D.

2088: thesis, University of Warwick, 1994.

2089: 

2090: \bibitem[SW]{SW} R. Schoen and J. Wolfson,

2091: \textit{Minimizing area among Lagrangian surfaces: the mapping problem.}

2092: J. Differential Geom. \textbf{58} (2001), 1--86. 

2093: 

2094: \bibitem[S1]{S1} P. Seidel,

2095: \textit{Lagrangian two-spheres can be symplectically knotted}, Jour.

2096: Diff. Geom. \textbf{52} (1999), 145--171.

2097: 

2098: \bibitem[S2]{S2} P. Seidel,

2099: \textit{Graded Lagrangian submanifolds}, Bull. Soc. Math. France.

2100: \textbf{128} (2000), 103--146.

2101: 

2102: \bibitem[S3]{S3} P. Seidel,

2103: \textit{Vanishing cycles and mutation}, European Congress of Mathematics,

2104: Vol. II (Barcelona, 2000), 65--85, Progr. Math., 202, Birkh{\"a}user, Basel, 2001.

2105: 

2106: \bibitem[ST]{ST} P. Seidel and R. P. Thomas,

2107: \textit{Braid group actions on derived categories of sheaves},

2108: Duke Math. J. \textbf{108} (2001), 37--108.

2109: 

2110: \bibitem[SV]{SV} A. Shapere and C. Vafa,

2111: \textit{BPS structure of Argyres-Douglas superconformal theories},

2112: preprint hep-th/9910182.

2113: 

2114: \bibitem[Sm]{Sm} K. Smoczyk,

2115: \textit{A canonical way to deform a Lagrangian submanifold},

2116: preprint dg-ga/9605005.

2117: 

2118: \bibitem[SYZ]{SYZ} A. Strominger, S.-T. Yau and E. Zaslow,

2119: \textit{Mirror Symmetry is T-Duality}, Nucl. Phys. \textbf{B479}

2120: (1996), 243--259.

2121: 

2122: \bibitem[Th]{Th} R. P. Thomas, 

2123: \textit{Moment maps, monodromy and mirror manifolds}, Symplectic geometry

2124: and mirror symmetry (Seoul, 2000), 467--498, World Sci. Publishing, 2001.

2125: 

2126: \bibitem[Wa]{Wa} M. T. Wang,

2127: \textit{Deforming area-preserving diffeomorphisms of surfaces by

2128: mean curvature flow}, Math. Res. Lett. \textbf{8} (2001), 651--661.

2129: 

2130: \end{thebibliography}

2131: 

2132: \small \noindent {\tt richard.thomas@ic.ac.uk} \newline

2133: \noindent Department of Mathematics, Imperial College, Huxley

2134: Building, 180 Queen's Gate, London, SW7 2BZ. UK. \newline\newline

2135: \noindent {\tt yau@math.harvard.edu} \newline

2136: \noindent Department of Mathematics, Harvard University, One

2137: Oxford Street, Cambridge MA 02138. USA.

2138: 

2139: \end{document}

2140: 

2141: 

2142: % Local Variables: 

2143: % TeX-parse-self: t

2144: % TeX-auto-save: t

2145: % mode: latex

2146: % TeX-master: "thesis"

2147: % End:

2148: