1: \documentclass[10pt,letterpaper]{article}

2: 

3: \usepackage[centertags]{amsmath}

4: \usepackage{amsthm}

5: \usepackage{amssymb}

6: \usepackage{amscd}

7: \usepackage{eucal}

8: \usepackage{epsfig}

9: \usepackage[matrix,arrow,curve]{xy}

10: \CompileMatrices

11: 

12: \renewcommand{\thesection}{\arabic{section}}

13: \renewcommand{\thesubsection}{\thesection.\arabic{subsection}}

14: 

15: \parskip0.75em

16: \parindent1.25em

17: 

18: \title{A long exact sequence for\\ symplectic Floer cohomology}

19: \author{Paul Seidel}

20: \date{Revised version, May 2002}

21: 

22: \begin{document}

23: \maketitle

24: 

25: %\input{commands}

26: 

27: 

28: \newcommand{\hatx}{\hat{x}}

29: \newcommand{\Sympe}{\mathrm{Symp}^e}

30: \newcommand{\crit}{{crit}}

31: \newcommand{\hatE}{\widehat{E}}

32: \newcommand{\hatO}{\widehat{\Omega}}

33: \newcommand{\hatQ}{\widehat{Q}}

34: \newcommand{\hatJ}{\widehat{J}}

35: \newcommand{\JJ}{\mathcal{J}}

36: \newcommand{\MM}{\mathcal{M}}

37: \newcommand{\WW}{\mathcal{W}}

38: \newcommand{\JJreg}{\JJ^{reg}}

39: \newcommand{\TT}{\mathcal{T}}

40: \newcommand{\BB}{\mathcal{B}}

41: \newcommand{\EE}{\mathcal{E}}

42: \newcommand{\FF}{\mathcal{F}}

43: \newcommand{\GG}{\mathcal{G}}

44: \newcommand{\loc}{{loc}}

45: \newcommand{\PP}{\mathcal{P}}

46: \newcommand{\gen}[1]{\leftsc #1 \rightsc}

47: \newcommand{\dist}{{dist}}

48: 

49: %---------------- GENERAL MATHS --------------------

50: 

51: \newcommand{\R}{\mathbb{R}}

52: \newcommand{\Z}{\mathbb{Z}}

53: \newcommand{\Q}{\mathbb{Q}}

54: \newcommand{\C}{\mathbb{C}}

55: \newcommand{\N}{\mathbb{N}}

56: \newcommand{\half}{{\textstyle\frac{1}{2}}}

57: \newcommand{\quarter}{{\textstyle\frac{1}{4}}}

58: 

59: \newcommand{\iso}{\cong}           %isomorphism sign

60: \newcommand{\htp}{\simeq}          %homotopy sign

61: \newcommand{\smooth}{C^\infty}

62: \newcommand{\CP}[1]{\C {\mathrm P}^{#1}}

63: \newcommand{\RP}[1]{\R {\mathrm P}^{#1}}

64: \newcommand{\leftsc}{\langle}

65: \newcommand{\rightsc}{\rangle}

66: %\newcommand{\Rgeq}{\R^{\scriptscriptstyle \geq 0}}

67: %\newcommand{\Rleq}{\R^{\scriptscriptstyle \leq 0}}

68: \newcommand{\suchthat}{\; : \;}

69: 

70: \newcommand{\id}{\mathrm{id}}

71: \newcommand{\ind}{\mathrm{ind}}

72: \newcommand{\re}{\mathrm{re}}

73: \newcommand{\im}{\mathrm{im}}

74: \renewcommand{\ker}{\mathrm{ker}}

75: \newcommand{\coker}{\mathrm{coker}}

76: \newcommand{\mymod}{\quad\text{mod }}

77: \newcommand{\Hom}{\mathrm{Hom}}

78: \newcommand{\End}{\mathrm{End}}

79: 

80: \newcommand{\mo}{(M,\omega)}

81: \renewcommand{\o}{\omega}

82: \renewcommand{\O}{\Omega}

83: \newcommand{\Diff}{\mathrm{Diff}}

84: 

85: %----------------- THEOREMS ------------------

86: 

87: \theoremstyle{plain}

88: \newtheorem{itheorem}{Theorem}

89: \newtheorem{thm}{Theorem}[section]

90: \newtheorem{theorem}[thm]{Theorem}

91: \newtheorem{cor}[thm]{Corollary}

92: \newtheorem{corollary}[thm]{Corollary}

93: \newtheorem{lemma}[thm]{Lemma}

94: \newtheorem{prop}[thm]{Proposition}

95: \newtheorem{proposition}[thm]{Proposition}

96: \newtheorem{defn}[thm]{Definition}

97: \newtheorem{definition}[thm]{Definition}

98: \newtheorem{definitions}[thm]{Definitions}

99: \newtheorem{rem}[thm]{Remark}

100: \newtheorem{remarks}[thm]{Remarks}

101: \newtheorem{remark}[thm]{Remark}

102: \newtheorem{example}[thm]{Example}

103: \newtheorem*{acknow}{Acknowledgments}

104: 

105: %----------------- LIST ENVIRONMENTS -----------

106: 

107: \newenvironment{condensedlist}%

108: {\renewcommand{\theenumi}{(\roman{enumi})}

109: \renewcommand{\labelenumi}{\theenumi}

110: \parskip0em

111: \begin{enumerate} \parsep0em

112: \parskip0em

113: \itemsep0em}{\vspace{-0.5em}\end{enumerate}}

114: 

115: \newenvironment{condensedprimelist}%

116: {\renewcommand{\theenumi}{(\roman{enumi}')}

117: \renewcommand{\labelenumi}{\theenumi}

118: \parskip0em

119: \begin{enumerate} \parsep0em

120: \parskip0em

121: \itemsep0em}{\vspace{-0.5em}\end{enumerate}}

122: 

123: \newenvironment{romanlist}%

124: {\renewcommand{\theenumi}{(\roman{enumi})}

125: \renewcommand{\labelenumi}{\theenumi}

126: \begin{enumerate} \parsep0em

127: \itemsep0em \parskip0.5em}{\end{enumerate}}

128: 

129: \newenvironment{Romanlist}%

130: {\renewcommand{\theenumi}{(\Roman{enumi})}

131: \renewcommand{\labelenumi}{\theenumi}

132: \begin{enumerate} \parsep0em

133: \itemsep0em \parskip0.5em}{\end{enumerate}}

134: 

135: \newenvironment{theoremlist}%

136: {\begin{list}{{\rm(\roman{enumi}) }}{\usecounter{enumi}

137: \renewcommand{\theenumi}{(\alph{enumi})}

138: \renewcommand{\labelenumi}{\theenumi}

139: \leftmargin0cm \labelsep0cm \rightmargin0cm \parsep1em \listparindent0em

140: \itemsep0em \topsep0em \parskip0.5em \setlength{\labelwidth}{\fill}}}

141: {\parskip0em \end{list}}

142: 

143: %------------ FIGURE MACROS -------------------

144: 

145: \newcommand{\printwarning}[1]{%

146: \typeout{ #1 }%

147: }

148: 

149: \newcommand{\FIGUREGAP}[1]{%

150: \printwarning{Figure #1 missing}%

151: \begin{figure}[htb] \begin{center}

152: \setlength{\unitlength}{1cm} \framebox[10cm]{\begin{picture}(6,4)

153: \end{picture}}

154: \caption{#1%

155: \label{fig:#1}}

156: \end{center} \end{figure}}

157: 

158: \newcommand{\includefigure}[3]{%

159: \begin{figure}[#3]

160: \begin{center}

161: \epsfig{file=#2} \\ \caption{\label{fig:#1}}

162: \end{center}

163: \end{figure}}

164: 

165: %\input{0}

166: 

167: \section*{Introduction}

168: 

169: Let $(M^{2n},\o,\alpha)$ be a compact symplectic manifold with contact type

170: boundary: $\alpha$ is a contact one-form on $\partial M$ which satisfies

171: $d\alpha = \o|\partial M$ and makes $\partial M$ convex. Assume in addition

172: that $[\o,\alpha] \in H^2(M,\partial M;\R)$ is zero, so that $\alpha$ can be

173: extended to a one-form $\theta$ on $M$ satisfying $d\theta = \o$. After fixing

174: such a $\theta$ once and for all, one can talk about exact Lagrangian

175: submanifolds in $M$. The Floer cohomology of two such submanifolds is

176: comparatively easy to define, since the corresponding action functional has no

177: periods, so that bubbling is impossible. The aim of this paper is to prove the

178: following result, which was announced in \cite{seidel00} (with an additional

179: assumption on $c_1(M)$ that has been removed in the meantime).

180: 

181: \begin{itheorem} \label{th:main}

182: Let $L$ be an exact Lagrangian sphere in $M$ together with a preferred

183: diffeomorphism $f: S^n \rightarrow L$. One can associate to it an exact

184: symplectic automorphism of $M$, the Dehn twist $\tau_L = \tau_{(L,[f])}$. For

185: any two exact Lagrangian submanifolds $L_0,L_1 \subset M$, there is a long

186: exact sequence of Floer cohomology groups

187: \begin{equation} \label{eq:exact} \xymatrix@C=-4em{

188:  {HF(\tau_L(L_0),L_1)} \ar[rr] && {HF(L_0,L_1)} \ar[dl] \\

189:  & {HF(L,L_1) \otimes HF(L_0,L).} \ar[ul]

190: }

191: \end{equation}

192: \end{itheorem}

193: 

194: The original inspiration for this came from the exact sequence in

195: Donaldson-Floer theory \cite{braam-donaldson94}, which can be translated into

196: symplectic geometry using various versions, proved \cite{dostoglou-salamon94}

197: and unproved, of the Atiyah-Floer conjecture. That line of thought should have

198: a Seiberg-Witten sibling, starting from \cite{carey-marcolli-wang98}, but the

199: corresponding Atiyah-Floer type relationships are only just beginning to be

200: understood \cite{salamon00}, \cite{ozsvath-szabo01}. In any case, the exact

201: sequences obtained from such speculations differ somewhat in generality from

202: that stated above. This reflects the fact that the first motivation has been

203: largely superseded by different ones, coming from mirror symmetry. Kontsevich's

204: homological mirror conjecture \cite{kontsevich94} for Calabi-Yau varieties

205: implies a relation between symplectic automorphisms and self-equivalences of

206: derived categories of coherent sheaves; see the survey \cite{aspinwall01} or

207: the papers \cite{seidel-thomas99}, \cite{horja01}. A particular class of

208: self-equivalences, ``twist functors along spherical objects'', is expected to

209: correspond to Dehn twists. By definition, twist functors give rise to an exact

210: sequence of the same form as \eqref{eq:exact}, with Floer cohomology replaced

211: by Ext-groups, so that the expected correspondence fits in well with our

212: result. With respect to the whole of Kontsevich's conjecture, this is a rather

213: peripheral issue. To see the exact sequence take on a more important role, one

214: has to pass to a related context, namely mirror symmetry for Fano varieties.

215: The derived categories of coherent sheaves on such varieties are often

216: generated by exceptional collections, which are subject to transformations

217: called mutations \cite{rudakov90}. The mirror dual notion is that of

218: distinguished basis of vanishing cycles, which is well-known in

219: Picard-Lefschetz theory. A rigorous connection between the two concepts is

220: established by \cite[Theorem 3.3]{seidel00} whose proof relies strongly on

221: Theorem \ref{th:main}; at present, this would seem to be its main application.

222: We conclude our discussion with some more concrete remarks about the statement

223: of the theorem:

224: 

225: (i) In this paper, Floer cohomology groups are treated as ungraded groups. One

226: can of course assume that $L,L_0,L_1$ are oriented, and then the groups become

227: $\Z/2$-graded. Different conventions are in use, but if one adopts that in

228: which the Euler characteristic of Floer cohomology is $(-1)^{n(n+1)/2}$ times

229: the intersection number, the degrees mod two of the maps in \eqref{eq:exact}

230: are

231: \begin{equation} \label{eq:grading}

232: \xymatrix{

233:  {\bullet} \ar[rr]^{0} && {\bullet} \ar[dl]^{n} \\

234:  & {\bullet} \ar[ul]^{1-n}

235: }

236: \end{equation}

237: Under more restrictive assumptions, one can introduce $\Z$-gradings. There are

238: several roughly equivalent ways of doing this; we adopt the approach of

239: \cite{kontsevich94} and \cite{seidel99}, in which one fixes a trivialization of

240: the bicanonical bundle $K^2_M$, thus establishing a notion of ``graded

241: Lagrangian submanifold''. Suppose that preferred gradings have been chosen for

242: $L,L_0,L_1$. Through the grading of $\tau_L$ itself, this induces a grading of

243: $\tau_L(L_0)$. All Floer cohomology groups in the exact sequence are then

244: canonically $\Z$-graded, and the degrees of the maps are as in

245: \eqref{eq:grading}. This is not difficult to show, it just requires a few

246: Maslov index computations.

247: 

248: (ii) We use Floer cohomology with $\Z/2$-coefficients. Inspection of the

249: discussion of coherent orientation in \cite{fukaya-oh-ohta-ono} suggests that

250: at least when $L_0,L_1$ are spin, the exact sequence should exist with

251: $\Z$-coefficients (with the $\otimes$ replaced by the cohomology of the

252: underlying tensor product of cochain groups, to avoid K{\"u}nneth terms).

253: However, I have not checked all the details.

254: 

255: (iii) The assumption $[\o,\alpha] = 0$ is the main limitation of the theorem as

256: it stands. We have adopted this ``exact'' framework with a view to the

257: application in \cite{seidel00}, and also because it simplifies a number of

258: technical issues, thereby hopefully allowing the basic ideas to stand out.

259: Almost the same proof goes through in a few other cases, such as when suitable

260: ``monotonicity'' conditions hold. On the other hand, a considerable amount of

261: work remains to be done to extend the exact sequence to the most general

262: situation where one would want to have it; a version for closed manifolds with

263: $c_1(M) = 0$ seems particularly desirable.

264: 

265: (iv) The map $\nwarrow$ in \eqref{eq:exact} is obtained by composing the

266: canonical isomorphism $HF(L_0,L) \iso HF(\tau_L(L_0),L)$ which exists because

267: $\tau_L(L) = L$, with a pair-of-pants product (a.k.a. Donaldson product). In

268: the spirit of \cite{piunikhin-salamon-schwarz94}, this can be seen as a kind of

269: relative Gromov invariant. A more general version of the same formalism,

270: involving pseudo-holomorphic sections of fibrations with singularities, yields

271: the second map $\rightarrow$. In contrast, $\swarrow$ appears as connecting map

272: in our construction, and is therefore defined only indirectly. A symmetry

273: consideration using the duality $HF(L_0,L_1) \iso HF(L_1,L_0)^\vee$ suggests

274: that this map should actually be a pair-of-pants coproduct. That is in fact

275: true, but we will not prove it here.

276: 

277: The exposition in the body of the paper follows a slightly indirect course, in

278: that we try to familiarize the reader with each ingredient separately, before

279: they all get mixed up into the main argument. There is even a small amount of

280: material which is not necessary for our immediate purpose, but which is closely

281: related and useful for further development. Thus, the whole first chapter is

282: elementary symplectic geometry, concentrating on topics related to

283: Picard-Lefschetz theory; the second one deals with pseudo-holomorphic curves,

284: which means setting up the relative invariants mentioned above, and introducing

285: certain techniques for partially computing them based on symplectic curvature;

286: and the only the third chapter addresses the actual proof.

287: 

288: \begin{acknow}

289: Much of this work was originally done in 1996. It goes without saying that I am

290: heavily indebted to Simon Donaldson, who was my advisor at that time.

291: Conversations with Michael Callahan, Mikhail Khovanov, Dietmar Salamon, and

292: Richard Thomas have been helpful. A preliminary version of this paper was

293: presented in a series of talks at Ecole Polytechnique in 1999; I thank my

294: colleagues there for their patience and comments. The final touches were put on

295: during a visit to the University of Michigan, which provided a hospitable

296: environment.

297: \end{acknow}

298: 

299: \newpage

300: \numberwithin{equation}{section}

301: 

302: %\input{1a}

303: 

304: \section[Dehn twists]{Dehn twists, and all that\label{ch:one}}

305: 

306: This chapter takes a look at basic Picard-Lefschetz theory from the symplectic

307: viewpoint. It has been known since Arnold's note \cite{arnold95} that such a

308: viewpoint makes sense, and it has been used for various purposes, see e.g.\

309: \cite{seidel98b}, \cite{khovanov-seidel98}. Still, the present paper seems to

310: be the most systematic attempt at an exposition so far. The reader may be

311: surprised by the exactness assumptions built into our framework. As far as the

312: elementary theory is concerned, there is no need to make such assumptions.

313: However, they greatly simplify the pseudo-holomorphic theory to be introduced

314: later on, and in order to keep the setup coherent, we have chosen to impose

315: them from the start.

316: 

317: \subsection{Exact symplectic geometry and fibrations\label{sec:basic}}

318: 

319: By an exact symplectic manifold we mean a compact manifold $M$ with boundary,

320: together with a symplectic form $\o$ and a one-form $\theta$ satisfying

321: $d\theta = \o$, such that $\theta|\partial M$ is a contact one-form and makes

322: $\partial M$ convex. An isomorphism of exact symplectic manifolds is a

323: diffeomorphism $\phi: M \rightarrow M'$ which is symplectic, satisfies

324: $\phi^*\theta' = \theta$ in some neighbourhood of $\partial M$, and such that

325: $[\phi^*\theta' - \theta] \in H^1(M,\partial M;\R)$ is zero. This means that

326: there is a unique function $K_\phi \in \smooth_c(M \setminus \partial M,\R)$

327: such that $\phi^*\theta' = \theta + dK_\phi$. We denote by $\Sympe(M)$ the

328: group of those exact symplectic automorphisms of $M$ which are the identity

329: near $\partial M$. Its Lie algebra consists of vector fields $X$ such that

330: $\o(\cdot,X) = dH$ for some $H$ which vanishes near $\partial M$, and is thus

331: identified with $\smooth_c(M \setminus \partial M,\R)$.

332: 

333: An exact Lagrangian submanifold in $M$ is a pair consisting of a Lagrangian

334: submanifold $L \subset M$ (always assumed to be disjoint from $\partial M$) and

335: a function $K_L$ on it such that $dK_L = \theta|L$. The image of $L$ under an

336: isomorphism $\phi: M \rightarrow M'$ of exact symplectic manifolds is again an

337: exact Lagrangian submanifold, in a canonical way; the associated function is

338: \begin{equation} \label{eq:induced-k}

339: K_{\phi(L)} = (K_L + K_\phi|L) \circ \phi^{-1}.

340: \end{equation}

341: In particular, $\Sympe(M)$ acts on the set of exact Lagrangian submanifolds. A

342: special situation which will occur later on is that $\phi \in \Sympe(M)$

343: satisfies $\phi(L) = L$ in the ordinary sense. Then (supposing $L$ to be

344: connected) $K_{\phi(L)} = K_L + c$ for some constant $c = K_\phi|L$, which

345: needs not be zero. This means that $\phi$ may not map $L$ to itself as an exact

346: Lagrangian submanifold, instead ``shifting'' it by some amount.

347: 

348: A notion of fibre bundle suitable for exact symplectic geometry is as follows.

349: Let $S$ be a smooth connected manifold, possibly with boundary (one could also

350: allow corners), and $\pi: E \rightarrow S$ a differentiable fibre bundle whose

351: fibres are compact manifolds with boundary. Write $\partial_hE \subset E$ for

352: the union of the boundaries of all the fibres. If the boundary of $S$ is empty,

353: $\partial_hE = \partial E$; otherwise $\partial E$ has another face

354: $\partial_vE = \pi^{-1}(\partial S)$, and the two faces meet at a codimension

355: two corner. An exact symplectic fibration is such an $(E,\pi)$ equipped with

356: $\Omega \in \Omega^2(E)$ and $\Theta \in \Omega^1(E)$, satisfying $d\Theta =

357: \Omega$, such that each fibre $E_z$ with $\o_z = \Omega|E_z$ and $\theta_z =

358: \Theta|E_z$ is an exact symplectic manifold. There is an additional condition

359: of triviality near $\partial_hE$, by which we mean the following: choose some

360: $z \in S$ and consider the trivial fibration $\tilde{\pi} : \tilde{E} = S

361: \times E_z \rightarrow S$, with the forms $\tilde{\Omega},\tilde{\Theta}$ which

362: are pullbacks of $\o_z,\theta_z$. There should be a fibrewise diffeomorphism

363: \begin{equation} \label{eq:boundary-trivialization}

364: \xymatrix{

365:  {N} \ar[rr] \ar[dr]_{\pi} && {\tilde{N}} \ar[dl]^{\tilde{\pi}} \\

366:  & {S} &

367: }

368: \end{equation}

369: betweeen neighbourhoods $N \subset E$, $\tilde{N} \subset \tilde{E}$ of

370: $\partial_hE$ resp.\ $\partial_h\tilde{E}$, which maps $\partial_hE$ to

371: $\partial_h\tilde{E}$, equals the identity on the fibre over $z$, and sends

372: $\Omega,\Theta$ to $\tilde{\Omega},\tilde{\Theta}$. Note that the choice of $z$

373: and the diffeomorphism are not considered to be part of the data defining an

374: exact symplectic fibration; only their existence is assumed.

375: 

376: \begin{lemma} \label{th:fibre-bundles}

377: Take a point $z \in S$ and a chart $\psi: U \rightarrow S$, with $U \subset

378: \R^k$ a contractible neighbourhood of $0$, such that $\psi(0) = z$. Then there

379: is a trivialization $\Psi: U \times E_z \rightarrow \psi^*E$, such that

380: $\Psi\,|\,\{0\} \times E_z = \id$ and

381: \[

382: \Psi^*\Theta = \theta_z + \sum_{i=1}^k H_i dt_i + dR.

383: \]

384: Here $H_1,\dots,H_k,R \in \smooth(U \times E_z,\R)$ are functions which vanish

385: near $U \times \partial E_z$, and $t_i$ the coordinates on $\R^k$. Moreover,

386: the difference between any two such trivializations $\Psi_1,\Psi_2$ is a map

387: $\Psi_2^{-1} \circ \Psi_1: (U,0) \rightarrow (\Sympe(E_z),\id)$. \qed

388: \end{lemma}

389: 

390: The proof is by a standard argument involving Moser's Lemma. The result means

391: first of all that any exact symplectic fibration $(E,\pi)$ has a structure of

392: $\Sympe(E_z)$-fibre bundle, where $E_z$ is any fibre. In addition to that, $E$

393: carries a preferred connection, which gives rise to canonical parallel

394: transport maps $\rho_c: E_{c(a)} \rightarrow E_{c(b)}$ over smooth paths $c:

395: [a;b] \rightarrow S$; these are exact symplectic isomorphisms, and lie in

396: $\Sympe(E_{c(a)})$ if $c$ is closed. To define the preferred connection, one

397: can use a local trivialization as before and set $A = \sum_i H_i\,dt_i$, which

398: is a one-form on $U$ with values in $\smooth_c(E_z \setminus \partial E_z,\R)$.

399: It is easy to check that this transforms in the proper way. A more intrinsic

400: approach is to observe that $TE_x = TE^h_x \oplus TE^v_x$ splits into a

401: horizontal and a vertical piece, given by $TE^v_x = \ker(D\pi_x)$ and

402: \begin{equation} \label{eq:connection}

403: TE^h_x = \{X \in TE_x \suchthat \Omega(X,\cdot)|TE^v_x = 0\}.

404: \end{equation}

405: Each $Z \in TS_z$ has a unique lift $Z^h \in \smooth(TE^h|E_z)$, and these

406: vectors define the same connection as before (conversely, one can show that

407: given a $\Sympe(M)$-fibre bundle and a compatible connection, one can equip its

408: total space with the structure of an exact symplectic fibration). In a local

409: trivialization as in Lemma \ref{th:fibre-bundles}, the curvature of the

410: connection is

411: \begin{equation} \label{eq:curvature}

412: F_A = \sum_{i<j} \left( -\frac{\partial H_i}{\partial t_j} + \frac{\partial

413: H_j}{\partial t_i} - \o_z(X_i,X_j) \right) dt_i \wedge dt_j,

414: \end{equation}

415: where $X_i$ is the family of Hamiltonian vector fields on $E_z$ corresponding

416: to $H_i(t,\cdot)$. If one takes the more intrinsic view, the curvature is a

417: two-form on $S$ with values in functions on the fibres, and is given by

418: $(Z_1,Z_2) \mapsto \Omega(Z_1^h,Z_2^h)$. In the case where the base $S$ is an

419: oriented surface, we say that $(E,\pi)$ is nonnegatively curved if for any

420: oriented chart $\psi$ and trivialization $\Psi$, the function in front of $dt_1

421: \wedge dt_2$ in \eqref{eq:curvature} is nonnegative; or equivalently, if

422: $\Omega|TE^h$ is nonnegative for the induced orientation of $TE^h$. To see what

423: this means, consider an exact symplectic fibration $(E,\pi)$ over the closed

424: unit disc $\bar{D}(1) \subset \C$, and the monodromy $\rho = \rho_{\partial

425: \bar{D}(1)} : E_1 \rightarrow E_1$ around the boundary in positive sense. If

426: the curvature is nonnegative, one can write $\rho$ as time-one map of some

427: (time-dependent) Hamiltonian on $E_1$ which vanishes near $\partial E_1$ and is

428: $\leq 0$ everywhere.

429: 

430: \begin{example} \label{ex:mapping-torus}

431: Let $E^\rho = \R \times M / (t,x) \sim (t-1,\rho(x))$ be the mapping

432: torus of $\rho \in \Sympe(M)$. To make this into an exact symplectic

433: fibration over $S^1 = \R/\Z$, one chooses a function $R_\rho \in

434: \smooth(\R \times M,\R)$ such that $R_\rho(t-1,\rho(x)) = R_\rho(t,x)

435: - K_\rho(x)$, and then sets $\Omega_{E^\rho} = \o$, $\Theta_{E^\rho}

436: = \theta + dR_\rho$. The preferred connection is the one induced from

437: the trivial connection on $\R \times M$, and its monodromy is $\rho$

438: itself. It is easy to prove that all exact symplectic fibrations over

439: a circle are of this form.

440: \end{example}

441: 

442: For future use, we make an observation on the compatibility of symplectic

443: parallel transport with Lagrangian submanifolds. Let $c: [a;b] \rightarrow S$

444: be a smooth embedded path. Suppose that we have an exact Lagrangian submanifold

445: in each fibre $E_{c(t)}$, depending smoothly on $t$; by this we mean a

446: subbundle $Q \subset E|\im(c)$ such that each fibre $Q_{c(t)} \subset E_{c(t)}$

447: is Lagrangian, together with a $K_Q \in \smooth(Q,\R)$ whose restrictions

448: $K_{Q_{c(t)}} = K_Q|Q_{c(t)}$ make the $Q_{c(t)}$ exact.

449: 

450: \begin{lemma} \label{th:submanifold}

451: Assume that all the $Q_{c(t)}$ are connected. Then the following conditions are

452: equivalent:

453: \begin{condensedlist}

454: \item

455: $\Omega|Q = 0$;

456: \item

457: $\Theta|Q = dK_Q + \pi^*\kappa_Q$ for some $\kappa_Q \in \Omega^1(\im(c))$;

458: \item \label{item:carry-along}

459: the maps $\rho_{c|[s;s']}: E_{c(s)} \rightarrow E_{c(s')}$ satisfy

460: $\rho_{c|[s;s']}(Q_{c(s)}) = Q_{c(s')}$ for all $s,s'$. \qed

461: \end{condensedlist}

462: \end{lemma}

463: 

464: The proof is straightforward. To be precise, \ref{item:carry-along} concerns

465: $Q_{c(s)}$ as Lagrangian submanifolds only. Taking the functions into account

466: and using \eqref{eq:induced-k} yields

467: \[

468: K_{\rho_{c|[s;s']}(Q_{c(s)})} = K_{Q_{c(s')}} + \int_{c|[s;s']} \kappa_Q.

469: \]

470: Hence $\rho_{c|[s;s']}(Q_{c(s)}) = Q_{c(s')}$ holds in the sense of exact

471: Lagrangian submanifolds iff $\kappa_Q = 0$, or what is the same, $\Theta|Q =

472: dK_Q$.

473: 

474: Just like any kind of fibre bundle with connection, exact symplectic fibrations

475: can be manipulated by cut-and-paste methods. As an example, take two oriented

476: surfaces $S^k$, $k = 1,2$, and boundary circles $C^k \subset \partial S^k$; and

477: let $S$ be the surface obtained by identifying $C^1,C^2$

478: orientation-reversingly. Suppose that we have exact symplectic fibrations

479: $(E^k,\pi^k,\Omega^k,\Theta^k)$ over $S^k$ whose monodromies around $C^k$,

480: taken in opposite senses, coincide. Then, assuming additionally that

481: $(E^k,\pi^k)$ is flat (has zero curvature) near $C^k$, one can construct from

482: them an exact symplectic fibration $(E,\pi)$ over $S$. It is maybe useful to

483: give some details of this. To start, take oriented collars $\psi^1: (-1;0]

484: \times \R/\Z \rightarrow S^1$, $\psi^2: [0;1) \times \R/\Z \rightarrow S^2$

485: around $C^1,C^2$ respectively, so that $S$ can be defined by using $\psi^2

486: \circ (\psi^1)^{-1}$ to identify the two circles. Our main assumption is that

487: there should be an exact symplectic isomorphism between the fibres of $E^k$

488: over $\psi^k(0,0)$, say, which relates the monodromies around $\psi^k(\{0\}

489: \times \R/\Z)$. Because of flatness, an equivalent formulation is that there is

490: some mapping torus $E^\rho$ as in Example \ref{ex:mapping-torus} and

491: diffeomorphisms

492: \[

493: \xymatrix{

494:  {(-1;0] \times E^\rho} \ar[d] \ar[r]^-{\Psi^1} & {E^1} \ar[d]^{\pi^1} \\

495:  {(-1;0] \times \R/\Z} \ar[r]^-{\psi^1} & {S^1}

496: } \qquad \xymatrix{

497:  {[0;1) \times E^\rho} \ar[d] \ar[r]^-{\Psi^2} & {E^2} \ar[d]^{\pi^2} \\

498:  {[0;1) \times \R/\Z} \ar[r]^-{\psi^2} & {S^2}

499: }

500: \]

501: such that $(\Psi^k)^*\Omega^k = \Omega_{E^\rho}$ and $(\Psi^k)^*\Theta^k =

502: \Theta_{E^\rho} + dR^k$ for some functions $R^k$. Clearly, one can introduce

503: modified forms $\tilde{\Theta}^k = \Theta^k - d\tilde{R}^k$ with suitable

504: functions $\tilde{R}^k$, in such a way that $(\Psi^k)^*\tilde{\Theta}^k =

505: \Theta_{E^\rho}$ near $\{0\} \times E^\rho$. Gluing together the $(E^k,\pi^k)$

506: along $C^k$ via $\Psi^2 \circ (\Psi^1)^{-1}$ yields a smooth fibration

507: $(E,\pi)$ over $S$, and the $\Omega^k, \tilde{\Theta}^k$ match up to forms

508: $\Omega,\Theta$ on it, making it an exact symplectic fibration. Since we have

509: not changed the symplectic connection, nonnegativity of the curvature of

510: $(E^k,\pi^k)$ implies the same for $(E,\pi)$.

511: 

512: \begin{remark} \label{re:squeeze-boundary}

513: The flatness condition on $(E^k,\pi^k)$ near $C^k$ can be removed. Namely,

514: suppose that it is not satisfied for $k = 1$. What one does then is to choose a

515: function $g \in \smooth([-1;0],\R)$ such that $g(s) = s$ for $s$ close to $-1$,

516: $g(s) = 0$ for $s$ close to $0$, and $g'(s) \geq 0$ everywhere; and define a

517: self-map $p$ of the surface $S^1$ by $p(z) = z$ for $z \notin \im(\psi^1)$,

518: $p(\psi^1(s,t)) = \psi^1(g(s),t)$. This collapses a small neighbourhood of

519: $C^1$ onto that boundary circle, so if we replace $(E^1,\pi^1)$ by its pullback

520: under $p$, it becomes flat near $C^1$, and the previous construction goes

521: through. It is noteworthy that this still preserves nonnegative curvature,

522: because $Dp$ has determinant $\geq 0$ everywhere.

523: \end{remark}

524: 

525: The basic objects of Picard-Lefschetz theory are fibrations over surfaces,

526: where the fibres are allowed to have certain particularly simple singularities.

527: Let $S$ be a connected oriented surface, possibly with boundary. An {\em exact

528: Lefschetz fibration}\footnote{Called ``exact Morse fibration'' in

529: \cite{seidel00}. The present terminology is more in line with general usage,

530: since the notion is closely related to, even though not quite the same as, the

531: symplectic Lefschetz fibrations considered by Donaldson, Gompf, and others.}

532: over $S$ consists of data $(E,\pi,\Omega,\Theta,J_0,j_0)$ as follows. $E$ is a

533: $(2n+2)$-dimensional manifold whose boundary is the union of two faces

534: $\partial_hE$ and $\partial_vE$, meeting at a codimension two corner. $\pi: E

535: \rightarrow S$ is a proper map with $\partial_vE = \pi^{-1}(\partial S)$ (so

536: this may be empty), and such that both $\pi|\partial_hE: \partial_hE

537: \rightarrow S$ and $\pi|\partial_vE: \partial_vE \rightarrow \partial S$ are

538: smooth fibre bundles. $\pi$ can have at most finitely many critical points, and

539: no two may lie on the same fibre (moreover, because of the previous

540: assumptions, they must lie in the interior of $E$). Denote by $E^\crit \subset

541: E$, $S^\crit \subset S$ the set of critical points resp.\ of critical values.

542: $J_0$ is an almost complex structure on a neighbourhood of $E^\crit$, and $j_0$

543: a positively oriented complex structure on a neighbourhood of $S^\crit$. These

544: are such that $\pi$ is $(J_0,j_0)$-holomorphic near $E^\crit$, and the Hessian

545: $D^2\pi$ at any critical point is nondegenerate as a complex quadratic form.

546: The closed two-form $\Omega \in \Omega^2(E)$ must be nondegenerate on $TE_x^v =

547: \ker(D\pi_x)$ for each $x \in E$, and a K{\"a}hler form for $J_0$ in some

548: neighbourhood of $E^\crit$. $\Theta \in \Omega^1(E)$ must satisfy $d\Theta =

549: \Omega$. We also require triviality near $\partial_hE$, which means the

550: existence of a map \eqref{eq:boundary-trivialization} with the same properties

551: as for exact symplectic fibrations. For brevity, exact Lefschetz fibrations

552: will usually be denoted by $(E,\pi)$ alone, as we have already done for exact

553: symplectic fibrations.

554: 

555: For any $x \in E$ there is a decomposition $TE_x = TE^h_x \oplus TE^v_x$ with

556: $TE^h_x$ defined as in \eqref{eq:connection}; the horizontal part is zero at

557: critical points, and projects isomorphically to $TS_z$, $z = \pi(x)$, at any

558: other point. We say that an exact Lefschetz fibration has nonnegative curvature

559: if $\Omega|TE^h_x \geq 0$ for each $x$. Note that if $x \notin E^\crit$ is

560: close to a critical point, $\Omega|TE^h_x$ is strictly positive anyway, because

561: of the K{\"a}hlerness assumption on $\Omega$. A standard argument based on this

562: shows

563: 

564: \begin{lemma} \label{th:add-base}

565: If $\beta \in \Omega^2(S)$ is a sufficiently positive two-form, $\Omega +

566: \pi^*\beta$ is a symplectic form on $S$. \qed

567: \end{lemma}

568: 

569: Symplectic parallel transport for an exact Lefschetz fibration is well-defined

570: as long as one avoids the critical fibres; indeed, if one removes those fibres,

571: the remainder is an exact symplectic fibration over $S \setminus S^\crit$. We

572: now take a look at the structure of the critical points. Take $z_0 \in S^\crit$

573: and local $j_0$-holomorphic coordinates $\xi: U \rightarrow S$, where $U

574: \subset \C$ is a neighbourhood of the origin, such that $\xi(0) = z_0$. By

575: assumption there is a unique critical point $x_0 \in E_{z_0}$. The holomorphic

576: Morse Lemma says that one can find a neighbourhood of the origin $W \subset

577: \C^{n+1}$ and a $J_0$-holomorphic chart $\Xi: W \rightarrow E$ with $\Xi(0) =

578: x_0$, such that

579: \[

580: (\xi^{-1} \circ \pi \circ \Xi)(x) = x_1^2 + \dots + x_{n+1}^2

581: \]

582: is the standard nondegenerate quadratic form on $\C^{n+1}$. We call $(\xi,\Xi)$

583: a holomorphic Morse chart. In general, it is not possible to choose $\Xi$ in

584: such a way that $\Xi^*\Omega$ is the standard K{\"a}hler form on $W \subset

585: \C^{n+1}$; however, one can remedy this by a suitable local deformation.

586: 

587: \begin{lemma} \label{th:local-deformation}

588: Let $(E,\pi,\Omega,\Theta,J_0,j_0)$ be an exact Lefschetz fibration, and $x_0$

589: a critical point of $\pi$. Then there are smooth families $\Omega^\mu \in

590: \Omega^2(E)$, $\Theta^\mu \in \Omega^1(E)$, $0 \leq \mu \leq 1$, such that

591: \begin{condensedlist}

592: \item

593: $\Omega^0 = \Omega$, $\Theta^0 = \Theta$;

594: \item

595: for all $\mu$, $\Omega^\mu = \Omega^0$ and $\Theta^\mu = \Theta^0$ outside a

596: small neighbourhood of $x_0$;

597: \item

598: each $(E,\pi,\Omega^\mu,\Theta^\mu,J_0,j_0)$ is an exact Lefschetz fibration;

599: \item

600: there is a holomorphic Morse chart $(\xi,\Xi)$ around $x_0$ such that

601: $\Xi^*\Omega^1$, $\Xi^*\Theta^1$ agree near the origin with the standard forms

602: $\o_{\C^{n+1}} = \frac{i}{2} \sum dx_k \wedge d\bar{x}_k$, $\theta_{\C^{n+1}} =

603: \frac{i}{4} \sum x_k d\bar{x}_k - \bar{x}_k dx_k$.

604: \end{condensedlist}

605: \end{lemma}

606: 

607: The proof is based on an elementary local statement about K{\"a}hler forms.

608: 

609: \begin{lemma} \label{th:kaehler}

610: Let $\o$ be a K{\"a}hler form on the ball $B = B^{2n+2}(r)$ of radius $r>0$ in

611: $\C^{n+1}$. Then there is another K{\"a}hler form $\o'$ which agrees with $\o$

612: near $\partial B$, and which close to the origin is some small multiple of

613: $\o_{\C^{n+1}}$.

614: \end{lemma}

615: 

616: \proof The first step is to find a K{\"a}hler form $\o''$ which is equal to

617: $\o$ near $\partial B$, and which has constant coefficients near the origin.

618: For this write\footnote{The definition of $d^c$ in this paper is such that

619: $\o_{\C} = -dd^c(\quarter|z|^2)$. This differs from the majority convention by

620: a negative constant.} $\o = \beta + dd^c f$, with $\beta$ constant and $f$

621: vanishing to second order at $x = 0$. Take a cutoff function $g \in

622: \smooth(\R^+,\R)$ such that $g(t) = 1$ for $t \leq 1$ and $g(t) = 0$ for $t

623: \geq 2$, and set $f_\epsilon(x) = g(||x||/\epsilon)f(x)$. A straightforward

624: computation shows that as $\epsilon \rightarrow 0$, the functions $f_\epsilon$

625: not only have increasingly small support, but also tend to $0$ in the $C^2$

626: topology. The desired form is, for small $\epsilon$,

627: \begin{equation} \label{eq:ohalf}

628: \o'' = \o - dd^c f_\epsilon = \beta + dd^c (f - f_\epsilon).

629: \end{equation}

630: 

631: In a second step, choose some small $t>0$ such that the constant form $\beta' =

632: \beta - t\o_{\C^{n+1}}$ is still K{\"a}hler. Take a two-form $\gamma$ on

633: $\C^{n+1}$ which is of type (1,1) and nonnegative everywhere, which vanishes

634: near the origin, and which is equal to $\o_{\C^{n+1}}$ outside a compact

635: subset; this can be obtained as $\gamma = -dd^c h(||x||^2)$ for a suitable

636: convex function $h$. Pulling $\gamma$ back by a linear map transforms it into

637: another nonnegative (1,1)-form $\gamma'$, zero near the origin and equal to

638: $\beta'$ outside a compact subset. Then $\gamma'' = \gamma' + t\o_{\C^{n+1}}$

639: is K{\"a}hler, equals $t\o_{\C^{n+1}}$ near the origin, and $\beta$ outside a

640: compact subset. That compact subset can be made arbitrarily small by retracting

641: linearly and rescaling; the two-form obtained in that way can be plugged into

642: $\o''$ locally near zero, yielding $\o'$. \qed

643: 

644: \proof[Proof of Lemma \ref{th:local-deformation}] Take some holomorphic Morse

645: chart $(\xi,\Xi)$ for $x_0$. Lemma \ref{th:kaehler} says that one can find a

646: two-form $\Omega^1$ on $E$ which agrees with $\Omega = \Omega^0$ outside

647: $\im(\Xi)$, such that $\Xi^*\Omega^1$ is K{\"a}hler and, near the origin,

648: equals $c\, \o_{\C^{n+1}}$ for some small constant $c>0$. The obstruction to

649: finding a $\Theta^1 \in \Omega^1(E)$ which agrees with $\Theta = \Theta^0$

650: outside $\im(\Xi)$ and satisfies $d\Theta^1 = \Omega^1$ lies in $H^2(B,\partial

651: B;\R)$, with $B$ a $(2n+2)$-dimensional ball; which is zero. An arbitrarily

652: chosen $\Theta^1$ needs to be modified to make $\Xi^*\Theta^1$ equal to

653: $c\,\theta_{\C^{n+1}}$ near $x = 0$, but that can be done by adding the

654: differential of some function to it. By restricting $\xi,\Xi$ to smaller

655: neighbourhoods, and rescaling them by $c^{-1/2}$ and $c^{-1}$ respectively, one

656: achieves that $\Xi^*\Omega^1 = \o_{\C^{n+1}}$ and $\Xi^*\Theta^1 =

657: \theta_{\C^{n+1}}$ near the origin. Finally, $\Omega^\mu$ and $\Theta^\mu$ are

658: defined by interpolating linearly between $\mu = 0$ and $1$; the required

659: properties are obvious. \qed

660: 

661: %\input{1b}

662: 

663: \subsection{The local model\label{sec:model}}

664: 

665: Consider $T = T^*\!S^n$ with its standard forms $\o_T \in \Omega^2(T)$,

666: $\theta_T \in \Omega^1(T)$. For concrete computations we use the coordinates $T

667: = \{ (u,v) \in \R^{n+1} \times \R^{n+1} \suchthat \leftsc u, v \rightsc = 0, \;

668: ||v|| = 1 \}$. For each $\lambda>0$, the subspace $T(\lambda)$ of cotangent

669: vectors of length $\leq \lambda$ is an exact symplectic manifold. We write

670: similarly $T(0) \subset T$ for the zero-section. The length function $\mu: T

671: \rightarrow \R$, $\mu(u,v) = ||u||$, generates a Hamiltonian circle action

672: $\sigma$ on $T \setminus T(0)$ which, after identifying $T \iso TS^n$ via the

673: standard metric, can be described as the normalized geodesic flow on $S^n$. In

674: coordinates

675: \[

676: \sigma_t(u,v) = (\cos(t)u - \sin(t) ||u|| v,\cos(t)v + \sin(t)\frac{u}{||u||}).

677: \]

678: $\sigma_\pi$ is the antipodal involution $A(u,v) = (-u,-v)$, hence extends

679: continuously over $T(0)$ (unlike any $\sigma_t$, $0 < t < \pi$). This can be

680: used to define certain symplectic automorphisms of $T$. The construction is by

681: now well-known, but we repeat it here since precise control over the parameters

682: will be important later on.

683: 

684: \begin{lemma} \label{th:on-invariant}

685: Let $R \in \smooth(\R,\R)$ be a function which vanishes for $t \gg 0$ and which

686: satisfies $R(-t) = R(t) - kt$ for small $|t|$, with some $k \in \Z$. Let

687: $(\phi_t^H)$ be the Hamiltonian flow of $H = R(\mu)$, defined on $T \setminus

688: T(0)$. Then $\phi_{2\pi}^H$ extends smoothly over $T(0)$ to a compactly

689: supported symplectic automorphism $\phi$ of $T$. The function $K = 2\pi

690: \,(R'(\mu)\mu - R(\mu))$ also extends smoothly over $T(0)$, and

691: \begin{equation} \label{eq:k-function}

692: \phi^*\theta_T - \theta_T = dK.

693: \end{equation}

694: \end{lemma}

695: 

696: \proof Two Hamiltonians $H_1,H_2$ which are both functions of $\mu$ always

697: Poisson-commute, so that $\phi_t^{H_1} \phi_t^{H_2} = \phi_t^{H_1 + H_2}$.

698: Decomposing $H$ into $H_1 = R(\mu) - (k/2)\mu$ and $H_2 = (k/2)\mu$, one gets

699: \[

700: \phi_{2\pi}^H = \phi_{2\pi}^{H_1} \circ \sigma_{k\pi}.

701: \]

702: We may assume that $R(-t) = R(t) - kt$ holds everywhere, since that can be

703: achieved by modifying $R$ for negative values only, which does not affect

704: $\phi$. Then $R(t) - (k/2)t$ is an even function, so it can be written as a

705: smooth function of $t^2$. This proves that $H_1$ and its flow extend smoothly

706: over $T(0)$. We know that $\sigma_{k\pi} = A^k$ extends smoothly, so the same

707: holds for $\phi_{2\pi}^H$. Since $H(y)$ vanishes for points with $\mu(y) \gg

708: 0$, $\phi$ is compactly supported (one can show that the compactly supported

709: symplectic automorphisms which are obtained in this way are precisely those

710: which are equivariant for the obvious $O(n+1)$-action). The function $R'(t)t -

711: R(t)$ is even, so $K$ extends smoothly over $T(0)$ for the same reason as

712: before. A computation shows that the Hamiltonian vector field $X$ of $H$

713: satisfies $L_{2\pi X}\theta_T = dK$. Since $H$ and $K$ are both functions of

714: $\mu$, $\phi_t^H$ preserves $K$, which implies \eqref{eq:k-function}. \qed

715: 

716: Clearly, if $supp(R) \subset (-\infty;\lambda)$ then $\phi$ restricts to a

717: symplectic automorphism of $T(\lambda)$ which is the identity near the

718: boundary. \eqref{eq:k-function} shows that this is an exact symplectic

719: automorphism. In the case $k = 1$ we call these automorphisms {\em model Dehn

720: twists}, and generally denote them by $\tau$; any two of them are isotopic in

721: $\Sympe(T(\lambda))$. An explicit formula is

722: \begin{equation} \label{eq:explicit-twist}

723: \tau(y) =

724:  \begin{cases}

725:  \sigma_{2\pi R'(\mu(y))}(y) & y \in T(\lambda) \setminus T(0),\\

726:  A(y) & y \in T(0),

727:  \end{cases}

728: \end{equation}

729: where the angle of rotation goes from $2\pi R'(0) = \pi$ to $2\pi R'(\lambda) =

730: 0$. Note that $\tau$ maps $T(0)$ to itself, and is the antipodal map on it. If

731: one considers $T(0)$ as an exact Lagrangian submanifold, with a function

732: $K_{T(0)} = const.$ associated to it, then by \eqref{eq:induced-k} and

733: \eqref{eq:k-function}

734: \begin{equation} \label{eq:self-shift}

735: K_{\tau(T(0))} = (K_{T(0)} + K_{\tau_L}|T(0)) \circ \tau_L^{-1} = K_{T(0)} -

736: 2\pi R(0).

737: \end{equation}

738: On occasion, it is useful to demand that the angle $R'(t)$ does not oscillate

739: too much. We say that $\tau$ is {\em $\delta$-wobbly} for some $0 < \delta <

740: 1/2$ if $R'(t) \geq 0$ for all $t \geq 0$, and $R''(t) < 0$ for all $t \geq 0$

741: such that $R'(t) \geq \delta$.

742: 

743: \begin{lemma} \label{th:local-intersections}

744: Suppose that $\tau$ is $\delta$-wobbly. Let $F_0 = T(\lambda)_{y_0}$, $F_1 =

745: T(\lambda)_{y_1}$ be the fibres of $T(\lambda) \rightarrow S^n$ at points

746: $y_0,y_1$, whose distance in the standard metric is $\dist(y_0,y_1) \geq 2\pi

747: \delta$. Then $\tau(F_0)$ intersects $F_1$ transversally and at a single point

748: $y$, which satisfies

749: \[

750: 2\pi R'(||y||) = \dist(y_0,y_1).

751: \]

752: In the special case where $y_1 = A(y_0)$ one has $y = y_1$; and then the

753: tangent space of $T(\lambda)$ at $y$ can be identified symplectically with

754: $\C^n$ in such a way that the subspaces tangent to $\tau(F_0)$, $T(0)$, $F_1$

755: become respectively, $\R^n$, $e^{2\pi i/3}\R^n$, and $e^{\pi i/3}\R^n$.

756: \end{lemma}

757: 

758: \proof Consider first the case when $y_1 \neq A(y_0)$. Suppose that $y \in F_1$

759: is a point with $\tau^{-1}(y) \in F_0$. By identifying $T \iso TS^n$ and using

760: the interpretation of $\sigma$ as normalized geodesic flow, one sees that $y$

761: must be a positive multiple of $c'(1)$, where $c: [0;1] \rightarrow S^n$ is the

762: minimal geodesic from $c(0) = y_0$ to $c(1) = y_1$. Moreover, the angle of

763: rotation must be $2\pi R'(||y||) = ||c'(1)|| = dist(y_0,y_1)$. These two

764: conditions are also sufficient. $\delta$-wobblyness implies that $2\pi R'(t) =

765: dist(y_0,y_1)$ has exactly one solution $t > 0$, which proves that there is

766: exactly one $y$. Combine the two conditions above into one, $c'(1) = 2\pi

767: R'(||y||)(y/||y||)$. Taking the derivative, one sees that a vector $Y \in

768: T(F_1)_y \iso T(S^n)_{y_1}$ satisfies $(T\tau)^{-1}(Y) \in T(F_0)$ iff

769: \[

770:  \textstyle R''(||y||) \leftsc \frac{y}{||y||}, Y \rightsc \frac{y}{||y||} +

771:  R'(||y||) \big( Y - \leftsc \frac{y}{||y||}, Y \rightsc \frac{y}{||y||} \big) = 0.

772: \]

773: We know that $R'(||y||) \geq \delta$; by $\delta$-wobblyness this implies

774: $R''(||y||) < 0$, which shows that $Y = 0$. Therefore $y \in \tau(F_0) \cap

775: F_1$ is a transverse intersection point.

776: 

777: Now consider the case when $y_1 = A(y_0)$. Then $y = y_1$ clearly lies in

778: $\tau(F_0) \cap F_1$, and because $R'(t) < 1/2$ for all $t>0$, there is no

779: other intersection point. In the notation from the proof of Lemma

780: \ref{th:on-invariant}, $\tau(F_0) = \phi^{H_1}_{2\pi}(A(F_0)) =

781: \phi^{H_1}_{2\pi}(F_1)$, and $y$ is a stationary point of $(\phi^{H_1}_t)$. It

782: follows that $T(\tau(F_0))_y$ is the image of $T(F_1)_y$ under the time $2\pi$

783: map of the linear Hamiltonian flow generated by the quadratic form $\half

784: Hess(H_1)_y$. If one identifies the tangent space to $T$ at $y$ with $T(S^n)_y

785: \oplus T(S^n)_y$ in such a way that the first summand is $T(F_1)_y$, then

786: \[

787: Hess(H_1)_y = \begin{pmatrix} R''(0) \cdot I & 0 \\ 0 & 0 \end{pmatrix}.

788: \]

789: Taking some isomorphism $T(S^n)_y \iso \R^n$ and its complexification $T(S^n)_y

790: \oplus T(S^n)_y \iso \C^n$, one finds that the tangent spaces of $\tau(F_0)$,

791: $T(0)$, and $F_1$ at $y$ correspond respectively to

792: \begin{equation} \label{eq:i-angle}

793: (1 + 2\pi i R''(0)) \R^n,\; i\R^n, \; \R^n \subset \C^n.

794: \end{equation}

795: In particular, since $R''(0) < 0$ by $\delta$-wobblyness, the intersection

796: $\tau(F_0) \cap F_1$ is transverse. At this point we need to recall a fact from

797: symplectic linear algebra, see e.g.\ \cite[p.\ 40]{lion-vergne}: the

798: classification of triples of mutually transverse linear Lagrangian subspaces,

799: up to the action of $Sp(2n)$, is equivalent to the classification of

800: nondegenerate quadratic forms on $\R^n$, up to $GL(n,\R)$. In particular there

801: is a finite number of equivalence classes, and deforming a triple continuously

802: while keeping transversality will not change its equivalence class. One can

803: clearly deform the three subspaces \eqref{eq:i-angle} in this way to $\R^n$,

804: $e^{2\pi i/3}\R^n$, $e^{\pi i/3}\R^n$, which proves the last part of the

805: statement. \qed

806: 

807: The next result links model Dehn twists to exact Lefschetz fibrations. The

808: connection has been known to algebraic geometers for a very long time, as

809: attested by the terminology ``Picard-Lefschetz transformations'' used for model

810: Dehn twists. But while the traditional approach ignores symplectic forms, they

811: are of course crucial for our purpose.

812: 

813: \begin{lemma} \label{th:model-fibrations}

814: Fix $\lambda>0$ and $r>0$, and let $\bar{D}(r) \subset \C$ be the closed disc

815: of radius $r$ around the origin. There is an exact Lefschetz fibration

816: $(E,\pi)$ over $\bar{D}(r)$, together with a diffeomorphism $\phi : E_r

817: \rightarrow T(\lambda)$ which respects both the symplectic forms and the exact

818: one-forms, such that the following holds. Denote by $\rho \in \Sympe(E_r)$ the

819: symplectic monodromy around $\partial \bar{D}(r)$, in positive sense. Then

820: $\tau = \phi \circ \rho \circ \phi^{-1} \in \Sympe(T(\lambda))$ is a model Dehn

821: twist.

822: \end{lemma}

823: 

824: \proof Take $\C^{n+1}$ with its standard forms $\o_{\C^{n+1}}$,

825: $\theta_{\C^{n+1}}$ and the function $q: \C^{n+1} \rightarrow \C$, $q(x) =

826: x_1^2 + \dots + x_{n+1}^2$. Even though $(\C^{n+1},q)$ is clearly not an exact

827: Lefschetz fibration (its fibres are not even compact), much of what was said in

828: the previous section carries over to it. The horizontal subspaces

829: \begin{equation} \label{eq:horizontal-model}

830:  T(\C^{n+1})_x^h =

831:  \{X \in \C^{n+1} \suchthat \o_{\C^{n+1}}(X,\ker\, Dq_x) = 0\} = \C \bar{x}

832: \end{equation}

833: define a symplectic connection away from the critical point $x = 0$, so that

834: one has parallel transport maps $q^{-1}(c(a)) \rightarrow q^{-1}(c(b))$ along

835: paths $c: [a;b] \rightarrow \C^*$. Consider the family of Lagrangian spheres

836: \begin{equation} \label{eq:model-spheres}

837: \Sigma_z = \sqrt{z}S^n = \{ (\sqrt{z}y_1, \dots, \sqrt{z}y_{n+1}) \suchthat y

838: \in S^n \subset \R^{n+1} \} \subset q^{-1}(z),

839: \end{equation}

840: $z \neq 0$, which for $z \rightarrow 0$ degenerate to $\Sigma_0 = \{0\} \subset

841: q^{-1}(0)$. Write $\Sigma^*$ for the union of all $\Sigma_z$, $z \neq 0$, and

842: $\Sigma = \Sigma^* \cup \Sigma_0$. One computes that

843: \begin{equation} \label{eq:theta-sigma}

844:  \theta_{\C^{n+1}} \, | \, \Sigma^* = q^* d^c({-\textstyle\quarter}|z|).

845: \end{equation}

846: Actually, the precise formula does not matter much for the moment. All we need

847: is that $\o_{\C^{n+1}} \,|\, \Sigma^*$ is the pullback by $q$ of some two-form

848: on $\C^*$, since that implies that parallel transport in any direction in

849: $\C^*$ takes the $\Sigma_z$ into each other; compare Lemma

850: \ref{th:submanifold}. The next observation is that if one removes $\Sigma$ then

851: parallel transport can be extended even to the singular fibre, so that for any

852: path $c$ in $\C$ one has a canonical symplectic isomorphism

853: \[

854: q^{-1}(c(a)) \setminus \Sigma_{c(a)} \rightarrow q^{-1}(c(b)) \setminus

855: \Sigma_{c(b)}.

856: \]

857: Since taking out $\Sigma_0$ removes the critical point, the only possible

858: problem is that a point in $\C^{n+1} \setminus \Sigma$ might move in horizontal

859: direction and converge to some point of $\Sigma$, which means that the flow of

860: some horizontal vector field would not be defined for all time. However, that

861: cannot happen since there is a function, $h(x) = ||x||^4 - |q(x)|^2$, which

862: satisfies $dh_x(\C\bar{x}) = 0$, hence is constant horizontally, and with

863: $h^{-1}(0) = \Sigma$. Therefore one can use parallel transport in radial

864: direction to trivialize $q: \C^{n+1} \setminus \Sigma \longrightarrow \C$

865: symplectically. In particular, if $\tilde{\rho}_s: q^{-1}(s) \rightarrow

866: q^{-1}(s)$ is the monodromy along the circle of radius $s>0$ around the origin,

867: $\tilde{\rho}_s\,|\,(q^{-1}(s) \setminus \Sigma_s)$ will be isotopic to the

868: identity in the group of all symplectic automorphisms of $q^{-1}(s) \setminus

869: \Sigma_s$. We have mentioned all this mainly to motivate the subsequent proof,

870: which is more computational.

871: 

872: Consider the map

873: \begin{equation} \label{eq:phi}

874: \begin{split}

875:  & \Phi: \C^{n+1} \setminus \Sigma \longrightarrow \C \times (T \setminus T(0)), \\

876:  & \Phi(x) = (q(x),\sigma_{\alpha/2}(-\im(\hatx)\,||\re(\hatx)||,

877:  \re(\hatx)\,||\re(\hatx)||^{-1})),

878: \end{split}

879: \end{equation}

880: where $s e^{i\alpha} = q(x)$ are polar coordinates on the base, and $\hatx =

881: e^{-i\alpha/2}x$. We claim that this is a diffeomorphism fibered over $\C$.

882: First of all, because of the use of polar coordinates, it is not obvious that

883: $\Phi$ is well-defined and smooth at $q^{-1}(0) \setminus \Sigma_0$. To dispel

884: any doubts about that one writes, after some manipulations,

885: \begin{gather*}

886: \Phi(x) = \big(q(x),

887:  - \half \im(x) \beta(x)

888:  - \half \im(\overline{q(x)}x) \beta(x)^{-1}, \\

889:  \qquad\qquad  h(x)^{-1/2} \re(x) \beta(x)

890:  - h(x)^{-1/2} \re(\overline{q(x)}x) \beta(x)^{-1}

891:  \big),

892: \end{gather*}

893: where $\beta(x) = (||x||^2 + h(x)^{1/2})^{1/2}$. The fact that $\Phi$ is a

894: diffeomorphism on each fibre is easy to see, either directly or by using the

895: symplectic forms and \eqref{eq:theta-pullback} below. Moreover, if one

896: restricts $\Phi$ to a fibre over $s > 0$, it extends to a diffeomorphism

897: $\phi_s: q^{-1}(s) \rightarrow T$. Such an extension does not exist for other

898: fibres, due to the non-continuity of $\sigma$ at $T(0)$. A computation yields

899: \begin{equation} \label{eq:theta-pullback}

900:  (\Phi^{-1})^*\theta_{\C^{n+1}} = \theta_T - \tilde{R}_s(\mu)\, d\alpha, \quad

901:  \tilde{R}_s(t) = \half t - \half\big(t^2 + s^2/4\big)^{1/2}.

902: \end{equation}

903: This implies that the restriction of $\Phi$ to any fibre is symplectic, and

904: actually maps the respective one-forms into each other. Of course, the same

905: will then be true for the continuous extensions $\phi_s$, $s>0$. In fact

906: \eqref{eq:theta-pullback} shows even more: if one restricts to any ray $\alpha

907: = const.$ in the basis then $\Phi^*\o_T = \o_{\C^{n+1}}$, which means that

908: $\Phi$ trivializes the symplectic parallel transport on $(\C^{n+1} \setminus

909: \Sigma,q)$ in radial directions, in accordance with the strategy which we set

910: out before. Consider $\tilde{\tau}_s = \phi_s \circ \tilde{\rho}_s \circ

911: \phi_s^{-1} : T \rightarrow T$, where $\tilde{\rho}_s$ is the monodromy map

912: introduced above. From \eqref{eq:theta-pullback} it follows that

913: $\tilde{\tau}_s$ restricted to $T \setminus T(0)$ is the time $2\pi$ map of the

914: Hamiltonian $\tilde{H}_s = \tilde{R}_s(\mu)$. Since $\tilde{R}_s(-t) =

915: \tilde{R}_s(t) - t$, this is quite close to the case $k = 1$ of Lemma

916: \ref{th:on-invariant}. The difference is that $\tilde{R}_s(t)$ does not vanish

917: for $t \gg 0$. Instead, it decays as follows:

918: \begin{equation} \label{eq:decay}

919:  0 > \tilde{R}_s(t) \geq -\textstyle{\frac{1}{16}} s^2t^{-1}, \quad

920:  0 < \textstyle{\frac{d}{dt}}

921:  \tilde{R}_s(t) \leq \textstyle{\frac{1}{16}} s^2t^{-2}.

922: \end{equation}

923: This is good enough to imply that $\tilde{\tau}_s$ is asymptotic to the

924: identity at infinity, for each $s>0$. It remains to tweak the given data

925: slightly, so as to produce a honest exact Lefschetz fibration, whose monodromy

926: is an actual model Dehn twist.

927: 

928: Fix $\lambda>0$, $r>0$. Choose a cutoff function $g \in \smooth(\R^+,\R)$ such

929: that $g'(t) \geq 0$ everywhere, $g(t) = 0$ for small $t$, and $g(t) = 1$ if $t$

930: is close to $\lambda$. We claim that there is a unique $\gamma \in

931: \Omega^1(\C^{n+1})$ with, again in polar coordinates on the base,

932: \begin{equation} \label{eq:cutoff-form}

933: (\Phi^{-1})^*\gamma = g(\mu)\tilde{R}_s(\mu) d\alpha.

934: \end{equation}

935: Since $\tilde{R}_0(t) = 0$, the function $\tilde{R}_s(t)/s$ extends smoothly to

936: $s = 0$, $t \neq 0$. Therefore the right hand side of \eqref{eq:cutoff-form}

937: can be written as $g(\mu)(\tilde{R}_s(\mu)/s)\, s d\alpha$, which means that

938: $\gamma$ is smooth at least on $\C^{n+1} \setminus \Sigma$. Now $\mu(\Phi(x)) =

939: (1/2) h(x)^{1/2}$; since $\Sigma = h^{-1}(0)$ and $g(t) = 0$ for small $t$, one

940: sees that $\Phi^*(g(\mu)\tilde{R}_s(\mu)d\alpha)$ vanishes near $\Sigma$, so

941: that $\gamma$ extends by zero over $\Sigma$. Set

942: \begin{equation} \label{eq:define-model}

943: \begin{cases}

944:  E = \Phi^{-1}(\bar{D}(r) \times (T(\lambda) \setminus T(0)))

945:  \cup (\Sigma \cap q^{-1}(\bar{D}(r))), \\

946:  \pi = q|E: E \longrightarrow \bar{D}(r), \\

947:  \Theta = (\theta_{\C^{n+1}} + \gamma) \,|\,E, \quad

948:  \Omega = d\Theta, \\

949:  \phi = \phi_r|E_r: E_r \longrightarrow T(\lambda).

950: \end{cases}

951: \end{equation}

952: $E \subset \C^{n+1}$ is cut out by the inequalities $h(x) \leq 4\lambda^2$,

953: $|q(x)| \leq r$. This makes it easy to show that it is a compact manifold with

954: corners, whose boundary faces are $\partial_vE = \pi^{-1}(\partial \bar{D}(r))$

955: and $\partial_hE = \{h(x) = 4\lambda^2\} = \{\mu(\Phi(x)) = \lambda\} =

956: \Phi^{-1}(\bar{D}(r) \times \partial T(\lambda))$. Because $\gamma$ vanishes

957: when restricted to any fibre, $\Omega|TE^v_x = \o_{\C^{n+1}} | \ker\, Dq_x$ is

958: nondegenerate for all $x$. In

959: \begin{equation} \label{eq:tilde-pullback}

960: \Theta\,|\,(E \setminus \Sigma) = \Phi^*(\theta_T + (g(\mu)-1)

961: \tilde{R}_s(\mu)\, d\alpha)

962: \end{equation}

963: the second term on the right hand side vanishes close to $\bar{D}(s) \times

964: \partial T(\lambda)$, so that $\Phi$ provides a trivialization near

965: $\partial_hE$ in the sense introduced in the previous section. Moreover, since

966: $\gamma$ vanishes near the critical point $x = 0$, equipping $E$ with the

967: standard complex structure $J_0$ near that point, and $\bar{D}(r)$ with the

968: standard complex structure $j_0$, turns $(E,\pi)$ into an exact Lefschetz

969: fibration. As for the statement about the monodromy $\rho$, one can repeat the

970: argument above, using \eqref{eq:tilde-pullback} instead of

971: \eqref{eq:theta-pullback}. This shows that $\tau = \phi \circ \rho \circ

972: \phi^{-1}$, when restricted to $T(\lambda) \setminus T(0)$, is the time $2\pi$

973: map of $R_r(\mu)$, where

974: \begin{equation} \label{eq:r-function}

975: R_r(t) = (1-g(t))\tilde{R}_r(t);

976: \end{equation}

977: by definition, this is a model Dehn twist. \qed

978: 

979: Let $(M,\o,\theta)$ be an exact symplectic manifold. A {\em

980: framed}\footnote{This has little or nothing to do with the usual topological

981: notion of framed manifold.} {\em exact Lagrangian sphere} is an exact

982: Lagrangian submanifold $L \subset M$ together with an equivalence class $[f]$

983: of diffeomorphisms $f: S^n \rightarrow L$. Here $f_1,f_2$ are equivalent iff

984: $f_2^{-1}f_1$ can be deformed inside $\Diff(S^n)$ to an element of $O(n+1)$. To

985: any such $(L,[f])$ one associates a Dehn twist $\tau_{(L,[f])} \in \Sympe(M)$

986: as follows. Choose a representative $f$ and extend it to a symplectic embedding

987: $\iota: T(\lambda) \rightarrow M$ for some $\lambda>0$. Take a model Dehn twist

988: $\tau$ which is supported in the interior of $T(\lambda)$, and define

989: \[

990: \tau_{(L,[f])} = \begin{cases}

991:  \iota \circ \tau \circ \iota^{-1} & \text{on $\im(\iota)$,} \\

992:  \id & \text{elsewhere.}

993: \end{cases}

994: \]

995: The exactness of $\tau_{(L,[f])}$ follows from that of $L$; moreover, the

996: analogue of \eqref{eq:self-shift} holds. It is not difficult to show that the

997: isotopy class $[\tau_{(L,[f])}] \in \pi_0(\Sympe(M))$ is independent of the

998: choices made in the definition. In contrast, it is unknown whether a change of

999: the framing $[f]$ can affect $[\tau_{(L,[f])}]$; if the answer is negative the

1000: notion of framing could be dropped altogether, but while the question is open

1001: one cannot do without it. Still, for the sake of brevity we will often omit

1002: framings from the notation and write $\tau_L$ instead of $\tau_{(L,[f])}$. We

1003: will say that $\tau_L$ is $\delta$-wobbly if the local model it is constructed

1004: out of has this property.

1005: 

1006: \begin{prop} \label{th:standard-fibrations}

1007: Let $(L,[f])$ be a framed exact Lagrangian sphere in $M$. Fix some $r>0$. There

1008: is an exact Lefschetz fibration $(E^L,\pi^L)$ over $\bar{D}(r)$ together with

1009: an isomorphism $\phi^L: E^L_r \rightarrow M$ of exact symplectic manifolds,

1010: such that if $\rho^L$ is the symplectic monodromy around $\partial \bar{D}(r)$,

1011: then $\tau_L = \phi^L \circ \rho^L \circ (\phi^L)^{-1}$ is a Dehn twist along

1012: $(L,[f])$.

1013: \end{prop}

1014: 

1015: We will prove this under the assumption that there is an embedding $\iota:

1016: T(\lambda) \rightarrow M$ as before, with $\iota^*\theta = \theta_T$. This is

1017: not really a restriction, since one can always satisfy it by adding the

1018: derivative of a function to $\theta$, which does not change $M$ up to exact

1019: symplectic isomorphism. On the other hand, it allows us to make the statement

1020: slightly sharper: $\phi^L$ will map the one-forms on $E^L_r$ and $M$ into each

1021: other, and the Dehn twist obtained from the monodromy of $(E^L,\pi^L)$ will be

1022: one constructed using the given embedding $\iota$.

1023: 

1024: \proof Take $(E,\pi)$ and $\phi$ from Lemma \ref{th:model-fibrations}, with the

1025: given $r$ and $\lambda$. We will construct $(E^L,\pi^L)$ by attaching a trivial

1026: piece to $(E,\pi)$. By construction, there is a neighbourhood $N \subset E$ of

1027: $\partial_hE$, a neighbourhood $V$ of $\partial T(\lambda)$ in $T(\lambda)$,

1028: and a diffeomorphism $\Phi: N \rightarrow \bar{D}(r) \times V$ fibered over

1029: $\bar{D}(r)$, such that $\Theta|N = \Phi^*\theta_T$. Moreover, $\Phi$ agrees

1030: with $\phi$ on $N \cap E_r$. Set

1031: \[

1032: E^L = E \cup_{\sim} \bar{D}(r) \times (M \setminus \iota(T(\lambda) \setminus

1033: V)),

1034: \]

1035: where $\sim$ identifies $N$ with $\bar{D}(r) \times \iota(V)$ through $(\id

1036: \times \iota) \circ \Phi$. One similarly defines $\pi^L$ from $\pi$ and the

1037: projection $\bar{D}(r) \times M \rightarrow \bar{D}(r)$. The forms

1038: $\Omega^L,\Theta^L$ on $E^L$ come from the corresponding ones on $E$ and the

1039: pullbacks of $\o,\theta$ on the trivial part. $\phi^L$ is constructed from

1040: $\iota \circ \phi$ and the identity map. The complex structure near the

1041: critical point and critical value are inherited from $E$. All properties stated

1042: above are obvious from the construction and the definition of Dehn twists. \qed

1043: 

1044: We call the $(E^L,\pi^L)$ {\em standard fibrations}. For future reference, we

1045: will now state certain properties which these fibrations inherit from the local

1046: model $(E,\pi)$, and which depend on the details of its construction.

1047: 

1048: \begin{lemma} \label{th:standard-properties}

1049: Any standard fibration $(E^L,\pi^L)$ has the following properties.

1050: \begin{romanlist}

1051: \item \label{item:lagrangian-subbundle}

1052: There is a closed subset $\Sigma^L \subset E^L$ such that

1053: \[

1054: \Sigma^L_z = \Sigma^L \cap E^L_z =

1055:  \begin{cases}

1056:  \text{is an embedded $n$-sphere} & \text{if $z \neq 0$,} \\

1057:  \text{is the unique critical point $x_0 \in E^L_0$} &

1058:  \text{if $z = 0$.}

1059:  \end{cases}

1060: \]

1061: In fact $(\Sigma^L)^* = \Sigma^L \setminus \Sigma^L_0$ is a smooth $n$-sphere

1062: bundle over $\bar{D}(r) \setminus \{0\}$, and satisfies

1063: \begin{equation} \label{eq:theta-sigma-l}

1064:  \Theta^L \, | \, (\Sigma^L)^* =

1065:  (\pi^L)^* d^c(-\textstyle\quarter|z|).

1066: \end{equation}

1067: As in the discussion following \eqref{eq:theta-sigma}, this implies that each

1068: $\Sigma^L_z \subset E^L_z$, $z \neq 0$, is an exact Lagrangian submanifold, and

1069: that symplectic parallel transport within $\bar{D}(r) \setminus \{0\}$ carries

1070: these spheres into each other. Moreover, $\phi^L(\Sigma^L_r) = L$.

1071: \item \label{item:preferred-charts}

1072: There are holomorphic Morse charts $(\xi,\Xi)$ around the unique critical point

1073: $x_0 \in E^L_0$, such that $\xi$ is the inclusion $U \hookrightarrow

1074: \bar{D}(r)$ of some neighbourhood $U \subset \C$ of the origin; $\Xi^*\Theta^L

1075: = \theta_{\C^{n+1}}$ and $\Xi^*\Omega^L = \omega_{\C^{n+1}}$; and

1076: $\Xi^{-1}(\Sigma^L_z) = \Sigma_z$ for all sufficiently small $z \in \C$.

1077: \item \label{item:nonnegatively-curved}

1078: $(E^L,\pi^L)$ has nonnegative curvature.

1079: \item \label{item:precise-twist}

1080: $\tau_L = \phi^L \circ \rho^L \circ (\phi^L)^{-1}$ is the Dehn twist defined

1081: using $\iota$ and the function $R_r$ from \eqref{eq:r-function}; in particular

1082: $R_r(0) = -r/4$. By making $r$ smaller while keeping all other choices in the

1083: construction fixed, one can achieve that $\tau_L$ is $\delta$-wobbly for an

1084: arbitrary $\delta$.

1085: \end{romanlist}

1086: \end{lemma}

1087: 

1088: \proof \ref{item:lagrangian-subbundle} From \eqref{eq:define-model} one sees

1089: that $\Sigma \cap q^{-1}(\bar{D}(r)) \subset E$, and that $\Theta =

1090: \theta_{\C^{n+1}}$ in a neighbourhood of that subset. $\Sigma^L$ is defined to

1091: be the image of this in $E^L$, so that \eqref{eq:theta-sigma-l} is a

1092: consequence of \eqref{eq:theta-sigma}. \ref{item:preferred-charts} The

1093: definition of $\xi,\Xi$ is obvious, and the claim about $\Xi^*\Theta^L$ follows

1094: from the fact that in \eqref{eq:define-model} $\Theta = \theta_{\C^{n+1}}$ near

1095: the critical point. \ref{item:nonnegatively-curved} One can compute the

1096: curvature of $(E,\pi)$ from \eqref{eq:theta-pullback} and

1097: \eqref{eq:cutoff-form}; it turns out to be

1098: \[

1099: \Phi^*\big( (g(\mu)-1) \textstyle\frac{\partial}{\partial s} \tilde{R}_s(\mu)

1100: \big) ds \wedge d\alpha,

1101: \]

1102: which is $\geq 0$ everywhere. This implies the same property for $(E^L,\pi^L)$.

1103: \ref{item:precise-twist} The nontrivial statement is $\delta$-wobblyness, which

1104: requires that we take another look at the function $R_r$. There is a $t_0>0$

1105: such that $g(t) = 0$ for $t \in [0;t_0]$, and in that interval $R_r(t) =

1106: \tilde{R}_r(t)$, so that $(\partial/\partial t) R_r(t) > 0$,

1107: $(\partial^2/\partial t^2) R_r(t) < 0$. For $t \in [t_0;\lambda]$ one estimates

1108: using \eqref{eq:decay} that

1109: \[

1110: 0 \leq {\textstyle\frac{\partial}{\partial t}}R_r(t) = -g'(t) \tilde{R}_r(t) +

1111: (1-g(t)) {\textstyle\frac{\partial}{\partial t}} \tilde{R}_r(t) \leq

1112: \textstyle{\frac{r^2}{16}} (||g'|| t_0^{-1} + t_0^{-2}).

1113: \]

1114: By choosing $r$ small while keeping $g$ and hence $t_0$ fixed, one can make

1115: this $< \delta$ for an arbitrary $\delta$. \qed

1116: 

1117: %\input{1c}

1118: 

1119: \subsection{Vanishing cycles\label{sec:picard-lefschetz}}

1120: 

1121: We have seen that all Dehn twists can be realized as monodromy maps (of

1122: standard fibrations). Conversely, the geometry of any exact Lefschetz fibration

1123: can be understood in terms in Dehn twists. This is again well-known on a

1124: topological level, and our exposition repeats the classical arguments while

1125: paying more attention to symplectic forms. The results will not be used again

1126: in this paper, but they are important in applications.

1127: 

1128: Let $(E,\pi)$ be an exact Lefschetz fibration over $S$. Take $z_0 \in S^\crit$,

1129: and the unique critical point $x_0 \in E_{z_0}$. Let $c: [a;b] \rightarrow S$

1130: be a smooth embedded path with $c(b) = z_0$, $c^{-1}(S^\crit) = \{b\}$; and let

1131: $\rho_{c|[s,s']}: E_{c(s)} \rightarrow E_{c(s')}$ be the parallel transport

1132: maps along it, which is defined for all $s \leq s' < b$. Following a suggestion

1133: of Donaldson, we define

1134: \begin{equation} \label{eq:vanishing-ball}

1135: B_c = \{x \in E_{c(s)}, \; a \leq s < b \suchthat \lim_{s' \rightarrow b}

1136: \rho_{c|[s;s']}(x) = x_0 \} \cup \{ x_0 \} \subset E.

1137: \end{equation}

1138: 

1139: \begin{lemma} \label{th:vanishing-ball}

1140: $B_c$ is an embedded closed $(n+1)$-ball, with $\partial B_c = B_c \cap

1141: E_{c(a)}$. The function $p = c^{-1} \circ \pi: B_c \rightarrow [a;b]$ has $x_0$

1142: as its unique critical point, which is a nondegenerate local maximum. Moreover,

1143: $\Omega|B_c = 0$.

1144: \end{lemma}

1145: 

1146: \proof Put a symplectic form $\Omega + \pi^*\beta$ on $E$ as in Lemma

1147: \ref{th:add-base}. Choose an oriented embedding $\tilde{c}:

1148: (a-\epsilon;b+\epsilon) \times (-\epsilon;\epsilon) \rightarrow S$ such that

1149: $\tilde{c}(0,t) = c(t)$. Let $h$ be the function defined on $\im(\tilde{c})$

1150: with $h(\tilde{c}(s,t)) = -t$. The Hamiltonian vector field $X$ of $H = h \circ

1151: \pi$ has the following properties:

1152: \begin{condensedlist}

1153: \item \label{item:h-one}

1154: it is horizontal everywhere, $X_x \in TE^h_x$;

1155: \item \label{item:h-two}

1156: for each $x \neq x_0$, $D\pi(X)_x$ is a positive multiple of

1157: $\partial\tilde{c}/\partial s$;

1158: \item \label{item:h-three}

1159: $x_0$ is a hyperbolic stationary point of $X$, with $n+1$ positive and negative

1160: eigenvalues.

1161: \end{condensedlist}

1162: \ref{item:h-one} and \ref{item:h-two} are straightforward; \ref{item:h-three}

1163: can be seen by looking at $X$ in a holomorphic Morse chart, where it is

1164: $J_0\nabla H$. Let $\tilde{B} \subset E$ be the stable manifold of $x_0$. It is

1165: an open $(n+1)$-ball, lies in $\pi^{-1}\tilde{c}((a-\epsilon;b] \times \{0\})$,

1166: and the projection

1167: \[

1168:  \tilde{p} = \tilde{c}^{-1}\pi : \tilde{B} \longrightarrow

1169:  (a-\epsilon;b] \times \{0\}

1170: \]

1171: is a proper map. The tangent space of $\tilde{B}$ at $x_0$ is the negative

1172: eigenspace of $DX_{x_0}$; one shows easily that $D^2\tilde{p}|T\tilde{B}_{x_0}$

1173: differs from $DX|T\tilde{B}_{x_0}$ only by a positive constant, which means

1174: that $x_0$ is a nondegenerate maximum of $\tilde{p}$. There are no other

1175: critical points, because elsewhere $d\tilde{p}(X)_x>0$ by \ref{item:h-two}.

1176: Finally, since the flow $X$ is symplectic and contracts the tangent spaces of

1177: $\tilde{B}$, these must be Lagrangian subspaces, so $(\Omega +

1178: \pi^*\beta)|\tilde{B} = \Omega|\tilde{B} = 0$.

1179: 

1180: We claim that $B_c = \tilde{B} \cap \pi^{-1}(\im(c))$, which implies all

1181: desired properties of $B_c$. Let $Y$ be the vector field on

1182: $\pi^{-1}(\im(\tilde{c})) \setminus \{x_0\}$ which is horizontal and satisfies

1183: $D\pi(Y) = \partial\tilde{c}/\partial s$. From \ref{item:h-one} and

1184: \ref{item:h-two} above one sees that $Y = gX$ for some function $g$, bounded

1185: from below by a positive constant, and which goes to $\infty$ as one approaches

1186: $x_0$. Therefore the orbits of $X$ and $Y$ coincide, except of course for

1187: $\{x_0\}$. Since $Y|\pi^{-1}(\im(c))$ defines parallel transport along $c$, the

1188: claim follows. \qed

1189: 

1190: \begin{lemma} \label{th:vanishing-cycle}

1191: $V_c = \partial B_c$ is an exact Lagrangian sphere in $E_{c(a)}$, and comes

1192: with a canonical framing.

1193: \end{lemma}

1194: 

1195: \proof Since $B_c$ is a ball and $\Theta|B_c$ is closed, there is a unique $K

1196: \in \smooth(B_c,\R)$ with $K(x_0) = 0$ and $dK = \Theta|B_c$. The restriction

1197: $K_{V_c} = K|V_c$ makes $V_c$ into an exact Lagrangian submanifold. It remains

1198: to explain the framing. In any chart on $B_c$ around $x_0$, the level sets

1199: $p^{-1}(s')$ of the function from Lemma \ref{th:vanishing-ball}, for $s'$ close

1200: to $b$, will be strictly convex hypersurfaces, so that one can map them to

1201: $S^n$ by radial projection. On the other hand, $V_c = p^{-1}(a)$ can be

1202: identified with $p^{-1}(s')$ by the gradient flow with respect to some metric.

1203: Combining these two maps gives a diffeomorphism $V_c \rightarrow S^n$, which is

1204: unique up to isotopy and action of $O(n+1)$; its inverse is our framing. \qed

1205: 

1206: The framed exact Lagrangian sphere $V_c \subset E_{c(a)}$ is called the {\em

1207: vanishing cycle} associated to $c$. It exists more generally for any path $c:

1208: [a;b] \rightarrow S$ which is smooth, not necessarily embedded, but still

1209: satisfies

1210: \begin{equation} \label{eq:path}

1211: c^{-1}(S^\crit) = \{b\}, \quad c'(b) \neq 0.

1212: \end{equation}

1213: To construct it in this situation, one first extends $c$ to a map

1214: $(a-\epsilon;b+\epsilon) \times (-\epsilon;\epsilon) \rightarrow S$ which is a

1215: local oriented diffeomorphism at $(b,0)$, pulls back $(E,\pi)$ by that map, and

1216: then applies Lemma \ref{th:vanishing-ball} to the pullback exact Lefschetz

1217: fibration. Note also that deforming $c$ smoothly, rel endpoints, within the

1218: class \eqref{eq:path} yields an exact Lagrangian isotopy of the corresponding

1219: vanishing cycles, which is compatible with their framings.

1220: 

1221: \includefigure{doubling}{doubling.eps}{hb}%

1222: %

1223: \begin{prop} \label{th:picard-lefschetz}

1224: Let $c$ be a path satisfying \eqref{eq:path}, and $l$ the loop in $S \setminus

1225: S^\crit$ obtained by ``doubling'' $c$, as in Figure \ref{fig:doubling}. Then

1226: the monodromy around $l$ is isotopic to the Dehn twist along the vanishing

1227: cycle $V_c$:

1228: \begin{equation} \label{eq:picard-lefschetz}

1229: [\rho_l] = [\tau_{V_c}] \in \pi_0(\Sympe(E_{c(a)})).

1230: \end{equation}

1231: \end{prop}

1232: 

1233: \proof We start with a rather special case. Let $(E,\pi)$ be an exact Lefschetz

1234: fibration with base $\bar{D}(r)$ for some $r>0$, and which has exactly one

1235: critical point $x_0 \in E_0$. Write $M = E_r$. In addition, $x_0$ should admit

1236: a holomorphic Morse chart $(\xi,\Xi)$ where $\Xi$ is defined on $W = \{x \in

1237: \C^{n+1} \suchthat ||x|| \leq 2\sqrt{r}, \; |q(x)| \leq r\}$, where $\xi =

1238: \id_{\bar{D}(r)}$, and such that $\Xi^*\Omega$, $\Xi^*\Theta$ are standard.

1239: This means that the symplectic geometry of $(E,\pi)$ equals that of the local

1240: model $(\C^{n+1},q)$ discussed in Lemma \ref{th:model-fibrations}, at least in

1241: a suitable neighbourhood of $x_0$. In particular, each fibre $E_z$, $z \neq 0$,

1242: contains an exact Lagrangian sphere $\Sigma_z^E = \Xi(\Sigma_z)$, degenerating

1243: to $\Sigma_0^E = \{x_0\}$. Write $L = \Sigma_r^E \subset M$; this inherits an

1244: obvious framing from $\Sigma_r = \sqrt{r}S^n$. We claim that the monodromy

1245: around $\partial \bar{D}(r)$, denoted by $\rho \in \Sympe(M)$, is isotopic to

1246: $\tau_L$.

1247: 

1248: From \eqref{eq:theta-sigma} one deduces that parallel transport in $\bar{D}(r)

1249: \setminus \{0\}$ carries the $\Sigma_z^E$ into each other. Moreover, by the

1250: same argument as in Lemma \ref{th:model-fibrations}, if one removes $\Sigma^E =

1251: \bigcup_z \Sigma^E_z$ from $E$, parallel transport can be extended over the

1252: singular fibre. Using parallel transport in radial directions one constructs a

1253: trivialization $\Phi^E: E \setminus \Sigma^E \longrightarrow \bar{D}(r) \times

1254: (M \setminus L)$ which is the identity on the fibre over $r$; this is such

1255: that, in radial coordinates $z = se^{i\alpha}$ on the basis,

1256: $((\Phi^E)^{-1})^*\Theta = \theta - R^E \wedge d\alpha + dS^E$ for some

1257: functions $R^E = R^E(s,\alpha,x)$ and $S^E = S^E(s,\alpha,x)$ which vanish for

1258: $x$ close to $\partial M$. This implies that $\rho$ restricted to $M \setminus

1259: L$ is the time-$2\pi$ map of the flow generated by the time-dependent

1260: Hamiltonian $(\alpha,x) \mapsto R^E(r,\alpha,x)$. Clearly, $[\rho] \in

1261: \pi_0(\Sympe(M))$ depends only on the behaviour of this function in an

1262: arbitrarily small neighbourhood of $L$, or equivalently on $(E,\pi)$ and

1263: $\Omega$ close to $\Sigma^E$; since $\Sigma^E \subset \Xi(W)$ by definition,

1264: this can be determined from the local model $(\C^{n+1},q)$. It remains to spell

1265: out the computation.

1266: 

1267: Define an embedding $\iota: T(\lambda) \rightarrow M$, for some $\lambda>0$, by

1268: combining $\Xi_r = \Xi|(W \cap q^{-1}(r)) \rightarrow M$ with the inverse of

1269: the isomorphism $\phi_r: q^{-1}(r) \rightarrow T$ from the proof of Lemma

1270: \ref{th:model-fibrations}. This satisfies $\iota^*\theta = \theta_T$,

1271: $\iota^*\o = \o_T$, and $\iota(T(0)) = L$. Because both $\Phi^E$ and the

1272: trivialization $\Phi$ from \eqref{eq:phi} are defined by radial parallel

1273: transport, there is a commutative diagram

1274: \[

1275: \xymatrix{

1276:  {\C^{n+1} \setminus \Sigma} \ar[r]^-{\Phi} &

1277:  {\C \times (T \setminus T(0))} \\

1278:  {W \setminus \Sigma} \ar@{^{(}->}[u] \ar[d]_{\Xi\,|\,W \setminus \Sigma} &

1279:  {\bar{D}(r) \times (T(\lambda) \setminus T(0))} \ar@{^{(}->}[u]

1280:  \ar[d]^{\id \times (\iota \,|\, T(\lambda) \setminus T(0))} \\

1281:  {E \setminus \Sigma^E} \ar[r]^-{\Phi^E} & {\bar{D}(r) \times (M \setminus L).}

1282: }

1283: \]

1284: From this, $\Xi^*\Omega = \o_{\C^{n+1}}$ and \eqref{eq:theta-pullback} it

1285: follows that

1286: \begin{align*}

1287:  & (\id \times \iota)^*(\theta_T - R^E d\alpha + dS^E)

1288:  = (\id \times \iota)^*((\Phi^E)^{-1})^*\Theta \\

1289:  & = (\Phi^{-1})^*\theta_{\C^{n+1}} = \theta_T - \tilde{R}_s(\mu) \wedge d\alpha,

1290: \end{align*}

1291: and in particular that $R^E(r,\alpha,\iota(y)) = \tilde{R}_r(\mu(y))$ for all

1292: $\alpha$. In view of the discussion above, and the fact that $\tilde{R}_r(-t) =

1293: \tilde{R}_r(t) - t$, this implies the desired equality $[\rho] = [\tau_L]$. Now

1294: consider the path $c: [0;r] \rightarrow \bar{D}(r)$, $c(s) = r-s$. Because

1295: symplectic parallel transport takes the $\Sigma^E_z$ into each other, the ball

1296: $B_c$ from Lemma \ref{th:vanishing-ball} must be the union of $\Sigma^E_z$ for

1297: all $z \in [0;r]$, and the vanishing cycle is $V_c = \Sigma^E_r = L \subset M$,

1298: with the same framing as before. Therefore, what we have done up to now proves

1299: \eqref{eq:picard-lefschetz} for this special class of exact Lefschetz

1300: fibrations $(E,\pi)$ and the particular path $c$.

1301: 

1302: More generally, let $(E,\pi)$ be a exact Lefschetz fibration with arbitrary

1303: base $S$, and $x_0 \in E_{z_0}$ a critical point which admits a holomorphic

1304: Morse chart $(\xi,\Xi)$ such that $\Xi^*\Omega$, $\Xi^*\Theta$ are standard. By

1305: restricting to a suitably small disc around $z_0$ in the base, and making the

1306: domains of $\xi,\Xi$ smaller, one can arrive at the situation considered

1307: before. This means that \eqref{eq:picard-lefschetz} is true at least for one

1308: (short) path $c$ with endpoint $z_0$. But from that it follows easily for all

1309: other paths with the same endpoint.

1310: 

1311: It remains to remove the assumption concerning $\Xi$. Let $(E,\pi)$ be an

1312: arbitrary exact Lefschetz fibration, $c$ a path as in \eqref{eq:path}, and $l$

1313: a corresponding loop. With respect to the critical point in $E_{c(b)}$, take

1314: smooth families $\Omega^\mu$, $\Theta^\mu$ as in Lemma

1315: \ref{th:local-deformation}. These can be chosen such that in a neighbourhood of

1316: $E_{c(a)}$, $\Omega^\mu = \Omega$ and $\Theta^\mu = \Theta$ for all $\mu$. The

1317: corresponding vanishing cycles $V^\mu_c \subset E_{c(a)}$ form a smooth isotopy

1318: of framed exact Lagrangian spheres, and similarly there is a smooth family of

1319: monodromies $\rho_l^{\mu}$. For $\mu = 1$ our previous assumption about

1320: holomorphic Morse charts is satisfied, and hence $[\rho_l] = [\rho_l^0] =

1321: [\rho_l^1] = [\tau_{V^1_c}] = [\tau_{V^0_c}] = [\tau_{V_c}] \in

1322: \pi_0(\Sympe(E_{c(a)}))$. \qed

1323: 

1324: \newpage

1325: %\input{2a}

1326: \section[Sections]{Pseudo-holomorphic sections\label{ch:two}}

1327: 

1328: Before, we have considered exact Lefschetz fibrations as geometric

1329: objects in the sense of elementary symplectic geometry; now we will

1330: apply the theory of pseudo-holomorphic curves to them. By today's

1331: standards, the necessary analysis is rather unsophisticated. The

1332: basic Gromov-type invariant which will be introduced first uses only

1333: the most familiar techniques, on the level of the book

1334: \cite{mcduff-salamon}. Of course ours is a Lagrangian boundary value

1335: problem, but the only part of the analysis which specifically

1336: concerns the boundary is bubbling off of holomorphic discs, which was

1337: addressed by Floer \cite{floer88c} and Oh \cite{oh92} (some more

1338: recent expositions are \cite[Section A.4.3]{ivashkovich-shevchishin},

1339: \cite{frauenfelder01}). Later, when considering relative invariants,

1340: we will use essentially the same analysis as in \cite{schwarz95},

1341: \cite{piunikhin-salamon-schwarz94}, \cite{desilva98}, even though the

1342: geometric setting is rather different. The remaining material

1343: (Sections \ref{sec:horizontal}, \ref{sec:vanishing}, and

1344: \ref{sec:horizontal-two}) is more specifically designed for

1345: application to the exact sequence, and has a greater claim to

1346: originality.

1347: 

1348: \subsection{A simple invariant\label{sec:simple-invariant}}

1349: 

1350: Let $(E^{2n+2},\pi)$ be an exact Lefschetz fibration over $S$. A {\em

1351: Lagrangian boundary condition} is an $(n+1)$-dimensional submanifold $Q \subset

1352: E|\partial S$ which is disjoint from $\partial_hE$ and such that $\pi|Q: Q

1353: \rightarrow \partial S$ is a submersion, together with $\kappa_Q \in

1354: \Omega^1(\partial S)$ and $K_Q \in \smooth(Q,\R)$, satisfying $\Theta|Q =

1355: \pi^*\kappa_Q + dK_Q$. This implies that each $Q_z = Q \cap E_z$, $z \in

1356: \partial S$, is an exact Lagrangian submanifold of $E_z$, with function

1357: $K_Q|Q_z$ (one could consider a more general situation in which the condition

1358: on $\pi|Q$ is dropped, so that the $Q_z$ could have singularities; but we will

1359: not do that). From Lemma \ref{th:submanifold} one sees that parallel transport

1360: along $\partial S$ carries the $Q_z$ into each other, in the ordinary sense of

1361: the word; this holds in the ``exact'' sense, that is to say as exact Lagrangian

1362: submanifolds, iff $\kappa_Q = 0$. Note also that if one equips $E$ with a

1363: symplectic form $\Omega + \pi^*\beta$ as in Lemma \ref{th:add-base}, $Q$ itself

1364: becomes a Lagrangian submanifold. The aim of this section is to introduce, in

1365: the case where $S$ is compact with $\partial S \neq \emptyset$, a Gromov-type

1366: invariant

1367: \[

1368: \Phi_1(E,\pi,Q) \in H_*(Q_\zeta;\Z/2),

1369: \]

1370: where $\zeta$ is some point of $\partial S$. In a nutshell, this is the cycle

1371: represented by the values at $\zeta$ of pseudo-holomorphic sections of $E$ with

1372: boundary in $Q$.

1373: 

1374: To begin, we remind the reader of a class of almost complex structures suitable

1375: for exact symplectic manifolds. The Liouville vector field $N$ on such a

1376: manifold, $i_N\o = \theta$, defines a collar $\R^- \times \partial M

1377: \hookrightarrow M$. Let $\sigma$ be the function on a neighbourhood of

1378: $\partial M$ whose composition with $\R^- \times \partial M \hookrightarrow M$

1379: is projection to the $\R^-$ factor. An $\o$-compatible almost complex $J$ is

1380: called convex near the boundary if $\theta \circ J = d(e^\sigma)$ near

1381: $\partial M$. This implies $d(d(e^\sigma) \circ J) = -\o$, which serves to

1382: control the behaviour of $J$-holomorphic curves. It is well-known that the

1383: space of these $J$ is contractible (in particular, nonempty).

1384: 

1385: Given an exact Lefschetz fibration $(E,\pi)$, one can consider the Liouville

1386: vector field on each fibre, and this gives rise to a function $\sigma$ on a

1387: neighbourhood of $\partial_hE$ in $E$. Choose a complex structure $j$ on the

1388: base $S$ (whenever we do that, now or later, $j$ is assumed to be positively

1389: oriented and equal to $j_0$ in some neighbourhood of $S^\crit$). An almost

1390: complex structure $J$ on $E$ is called {\em compatible relative to $j$} if

1391: \begin{condensedlist}

1392: \item

1393: $J = J_0$ in a neighbourhood of $E^\crit$;

1394: \item \label{item:projection}

1395: $D\pi \circ J = j \circ D\pi$;

1396: \item \label{item:partially-compatible}

1397: $\Omega(\cdot,J\cdot) | TE_x^v$ is symmetric and positive definite for any $x

1398: \in E$;

1399: \item \label{item:convexity}

1400: in a neighbourhood of $\partial_hE$, $J(TE^h) = TE^h$ and $\Theta \circ J =

1401: d(e^\sigma)$.

1402: \end{condensedlist}

1403: To see more concretely the meaning of these conditions, take $x \notin E^\crit$

1404: and split $TE_x = TE_x^h \oplus TE_x^v \iso TS_z \oplus TE_x^v$, $z = \pi(x)$.

1405: One can then write

1406: \begin{equation} \label{eq:compatible-j}

1407:  J_x = \begin{pmatrix} j_z & 0 \\ J_x^{vh} & J_x^{vv} \end{pmatrix}

1408: \end{equation}

1409: where $J_x^{vv} \in \End(TE_x^v)$ is a complex structure compatible with

1410: $\Omega|TE_x^v$, and $J_x^{vh} \in \Hom(TS_z,TE_x^v)$ is $\C$-antilinear with

1411: respect to $j$ and $J^{vv}$. This is a reformulation of \ref{item:projection},

1412: \ref{item:partially-compatible}. An immediate consequence is the following

1413: result, which sharpens Lemma \ref{th:add-base}:

1414: 

1415: \begin{lemma} \label{th:tame}

1416: Let $J$ be compatible relative to $j$. Then for any sufficiently positive

1417: $\beta \in \Omega^2(S)$, $\Omega + \pi^*\beta$ tames $J$. \qed

1418: \end{lemma}

1419: 

1420: Because of the lack of antisymmetry in \eqref{eq:compatible-j}, $J$ will not be

1421: compatible with $\Omega + \pi^*\beta$ in the ordinary sense of the word, unless

1422: $J^{vh} = 0$; we will return to this more restricted class of almost complex

1423: structures in the next section. Continuing with the analysis of the conditions

1424: above, suppose now that $x \in E$ is sufficiently close to $\partial_hE$. More

1425: precisely, we require that $\sigma(x)$ is defined and that \ref{item:convexity}

1426: applies. There is then a further splitting

1427: \begin{equation} \label{eq:fine-splitting}

1428: TE^v_x \iso \R N \oplus \R R \oplus (\ker\, \Theta \cap \ker\, d\sigma \cap

1429: TE^v_x),

1430: \end{equation}

1431: where $N$ is the Liouville vector field and $R$ is the Hamiltonian vector field

1432: of $e^\sigma|E_z$. The two parts of \ref{item:convexity} say that $J_x^{vh} =

1433: 0$ and that with respect to \eqref{eq:fine-splitting},

1434: \begin{equation} \label{eq:boundary-j}

1435:  J_x^{vv} = \begin{pmatrix} 0 & -1 & 0 \\ 1 & 0 & 0 \\ 0 & 0 & * \end{pmatrix}.

1436: \end{equation}

1437: The pointwise analysis which we have just carried out can be recast in terms of

1438: sections of fibre bundles, and one sees then that the space $\JJ(E,\pi,j)$ of

1439: almost complex structures which are compatible relative to $j$ is contractible.

1440: 

1441: Suppose now that we have a Lagrangian boundary condition $Q$. The theory of

1442: pseudo-holomorphic sections with boundary in $Q$ fits into a familiar framework

1443: of infinite-dimensional manifolds and maps. We will now review this, on a

1444: formal level, that is to say using $\smooth$ spaces and without assuming that

1445: $S$ is compact. In that sense, $\JJ(E,\pi,j)$ is an infinite-dimensional

1446: manifold; its tangent space at $J$ consists of sections $Y \in

1447: \smooth(\End(TE))$ which are zero near $E^\crit$ and can be written, in

1448: parallel with \eqref{eq:compatible-j}, as

1449: \begin{equation} \label{eq:tangent-compatible}

1450: Y_x = \begin{pmatrix} 0 & 0 \\ Y^{vh}_x & Y^{vv}_x \end{pmatrix}

1451: \end{equation}

1452: with $Y^{vv}_x$ an infinitesimal deformation of the compatible complex

1453: structure $J^{vv}_x$, and $Y^{vv}_xJ^{vh}_x + J^{vv}_x Y^{vh}_x = - Y^{vh}_x

1454: j_z$. There is a further requirement about $Y$ near $\partial_hE$, the

1455: linearization of \ref{item:convexity}, which we leave to the reader to write

1456: down. The space $\BB$ of sections $u: S \rightarrow E$ satisfying $u(\partial

1457: S) \subset Q$ is also an infinite-dimensional manifold, with $T\BB_u = \{ X \in

1458: \smooth(u^*TE^v) \suchthat X_z \in T(Q_z) \text{ for all } z \in \partial S\}$;

1459: note that $u^*TE^v \rightarrow S$ is really a vector bundle, since $u$ as a

1460: smooth section of $\pi$ avoids $E^\crit$. Consider the infinite-dimensional

1461: vector bundle $\EE \rightarrow \BB \times \JJ(E,\pi,j)$ whose fibre at $(u,J)$

1462: is $\Omega^{0,1}(u^*TE^v)$, the space of $(0,1)$-forms on $(S,j)$ with values

1463: in $u^*(TE^v,J|TE^v)$. It has a canonical section $\bar\partial^{univ}(u,J) =

1464: \half (Du + J \circ Du \circ j)$, and the zero set $\MM^{univ} =

1465: (\bar\partial^{univ})^{-1}(0)$ consists of pairs $(u,J)$ such that $u$ is

1466: $(j,J)$-holomorphic. We denote by $\bar\partial_J$, $\MM_J$ the restrictions of

1467: $\bar\partial^{univ}$, $\MM^{univ}$ to a fixed $J \in \JJ(E,\pi,j)$. The

1468: derivative of $\bar\partial_J$ at $u \in \MM_J$ is a map $D_{u,J}: T\BB_u

1469: \rightarrow \EE_{u,J}$. An explicit formula is

1470: $

1471:  D_{u,J}(X) = \half (L_{\tilde{X}}J) \circ Du \circ j

1472: $

1473: where $\tilde{X}$ is any section of $TE^v$, defined on a neighbourhood of

1474: $\im(u)$ in $E$, such that $u^*\tilde{X} = X$. After choosing a torsion-free

1475: connection $\nabla$ on $TE$ (away from $E^\crit$) which preserves the

1476: integrable subbundle $TE^v$, one transforms this into the more familiar

1477: expression

1478: \begin{equation} \label{eq:connection-formula}

1479:  D_{u,J}(X) = \bar{\partial}_{J,u^*\nabla} X +

1480:  \half (\nabla_X J) \circ Du \circ j,

1481: \end{equation}

1482: in which $\bar{\partial}_{J,u^*\nabla} = (u^*\nabla)^{0,1}$ is the

1483: $\bar\partial$-operator associated to the pullback connection $u^*\nabla$ on

1484: $u^*(TE^v,J|TE^v)$. The derivative of $\bar\partial^{univ}$, $D^{univ}_{u,J} :

1485: T\BB_u \times T\JJ(E,\pi,j)_J \longrightarrow \EE_{u,J}$, is obviously

1486: \begin{equation} \label{eq:universal-operator}

1487: \begin{split}

1488:  & D^{univ}_{u,J}(X,Y) = D_{u,J}(X) + \half Y \circ Du \circ j.

1489: \end{split}

1490: \end{equation}

1491: 

1492: From this point onwards, we pass to a more realistic situation, and assume that

1493: $S$ is compact with $\partial S \neq \emptyset$. Our first observation is a

1494: consequence of \ref{item:convexity}. Informally speaking, it says that for the

1495: purposes of pseudo-holomorphic sections, the boundary $\partial_hE$ of the

1496: fibres can be ignored.

1497: 

1498: \begin{lemma} \label{th:convexity}

1499: For every $J \in \JJ(E,\pi,j)$ there is a compact subset $K \subset E \setminus

1500: \partial_hE$ such that all $u \in \MM_J$ satisfy $u(S) \subset K$.

1501: \end{lemma}

1502: 

1503: \proof Let $W \subset E$ be a closed neighbourhood of $\partial_hE$ which,

1504: under the collar embedding provided by the Liouville vector fields on the

1505: fibres, corresponds to $[-\epsilon;0] \times \partial_hE$ for some

1506: $\epsilon>0$. By definition $\sigma(W) = [-\epsilon;0]$. After possibly making

1507: $W$ and $\epsilon$ smaller, we may assume that $W \cap Q = \emptyset$, that

1508: $\Omega|TE^h$ vanishes on $W$, and that \ref{item:convexity} holds there. From

1509: $\Omega|TE^h_x = 0$ and $J(TE^h_x) = TE^h_x$ it follows that $\Omega(X,JX) \geq

1510: 0$ for all $X \in TE_x$, $x \in W$. Take $u \in \MM_J$ and consider the

1511: function $h = e^\sigma \circ u$ on $U = u^{-1}(W)$. This satisfies $h|\partial

1512: U \equiv e^{-\epsilon}$ and is subharmonic, because $d(dh \circ j) = d(\Theta

1513: \circ J \circ Du \circ j) = -u^*\Omega \leq 0$. It follows that $h \leq

1514: e^{-\epsilon}$ everywhere, which shows that $K = E \setminus int(W)$ has the

1515: required property. \qed

1516: 

1517: The action of $u \in \BB$ is defined to be $A(u) = \int_S u^*\Omega$. This is

1518: actually the same for all $u$, since

1519: \begin{equation} \label{eq:action}

1520: \int_S u^*\O = \int_{\partial S} u^*\Theta = \int_{\partial S} \kappa_Q.

1521: \end{equation}

1522: Therefore, if one equips $E$ with a symplectic form $\Omega + \pi^*\beta$

1523: taming $J$, all the $u \in \MM_J$ become pseudo-holomorphic curves with the

1524: same energy $\half \int_S ||Du||^2 = A(u) + \int_S \beta$.

1525: 

1526: \begin{lemma} \label{th:compactness}

1527: $\MM_J$ is compact in any $C^r$-topology.

1528: \end{lemma}

1529: 

1530: \proof We apply the Gromov compactness theorem for $(j,J)$-holomorphic maps $S

1531: \rightarrow E$ with boundary in $Q$. The bubble components in the Gromov limit

1532: appear through a reparametrization which ``magnifies'' successively smaller

1533: parts of the domain. Since in our case all maps are sections, the bubbles are

1534: either nonconstant $J$-holomorphic spheres in some fibre $E_z$, or nonconstant

1535: $J$-holomorphic discs in $E_z$, $z \in \partial S$, with boundary on $Q_z$. But

1536: both are excluded by our assumptions, since $\Omega$ is exact and $Q_z \subset

1537: E_z$ an exact Lagrangian submanifold. \qed

1538: 

1539: A less formal version of the infinite-dimen\-sional framework introduced above

1540: involves spaces of $W^{1,p}$-sections, $p>2$. We omit the construction itself,

1541: and only mention its main consequence. For $(u,J) \in \MM^{univ}$, the

1542: differential operator $D_{u,J}$ extends to a Fredholm operator $\WW^1_u

1543: \rightarrow \WW^0_{u,J}$ from the $W^{1,p}$-completion $\WW^1_u$ of $T\BB_u$ to

1544: the $L^p$-completion $\WW^0_{u,J}$ of $\EE_{u,J}$. We denote this extension

1545: equally by $D_{u,J}$. If it is onto (in which case one says that $u$ is

1546: regular), $\MM_J$ is a smooth finite-dimensional manifold near $u$. If that

1547: holds for all $u \in \MM_J$, $J$ itself is called regular, and the space of

1548: such $J$ is denoted by $\JJreg(E,\pi,Q,j) \subset \JJ(E,\pi,j)$.

1549: 

1550: \begin{lemma} \label{th:transversality}

1551: $\JJ^{reg}(E,\pi,Q,j)$ is $\smooth$-dense in $\JJ(E,\pi,j)$. In fact the

1552: following stronger statement holds: take some nonempty open subset $U \subset

1553: S$ and a $J \in \JJ(E,\pi,j)$. Then there are $J' \in \JJ^{reg}(E,\pi,Q,j)$

1554: arbitrarily close to $J$, such that $J = J'$ outside $\pi^{-1}(U)$.

1555: \end{lemma}

1556: 

1557: \proof Even though this is a well-known argument, we recall part of it as a

1558: preparation for subsequent more refined versions. Fix $U$ and $J$. By Lemmas

1559: \ref{th:convexity} and \ref{th:compactness} there is an open subset $V \subset

1560: E$ with $\overline{V} \cap (\partial_hE \cup E^\crit) = \emptyset$, such that

1561: $\im(u) \subset V$ for each $u \in \MM_J$. A modified version of the

1562: compactness argument shows that this property, with the same $V$, remains true

1563: for all almost complex structures in $\JJ(E,\pi,j)$ which are sufficiently

1564: close to $J$. We want to make $J$ regular by perturbing it on $V \cap

1565: \pi^{-1}(U)$. Take $\TT \subset T\JJ(E,\pi,j)_J$ to be the subset of those $Y$

1566: which vanish outside $V \cap \pi^{-1}(U)$, and consider the operator

1567: \begin{equation} \label{eq:transversality-operator}

1568:  D_{u,J}^{univ}: \WW^1_u \times \TT \longrightarrow \WW^0_{u,J}

1569: \end{equation}

1570: given by the same formula as in \eqref{eq:universal-operator}. By a standard

1571: arguments surjectivity of this operator implies the desired result (strictly

1572: speaking, what comes up in this argument is a dense subspace of those $Y$ which

1573: have ``finite $\smooth_\epsilon$-norm'', but since the first component

1574: $D_{u,J}$ is Fredholm, this makes no difference as far as surjectivity is

1575: concerned). Let $F \rightarrow S$ be the bundle dual to

1576: $\Lambda^{0,1}(u^*TE^v)$, so that $(\WW^0_{u,J})^* \iso L^q(F)$, $p^{-1} +

1577: q^{-1} = 1$. Suppose that $\eta \in L^q(F)$ is orthogonal to the image of

1578: \eqref{eq:transversality-operator}. It then satisfies

1579: \begin{equation} \label{eq:adjoint}

1580:  D_{u,J}^*\eta = 0 \; \text{ on $S \setminus \partial S$,} \quad

1581:  \text{and} \quad \int_S \leftsc \eta, Y \circ Du \circ j \rightsc = 0

1582:  \; \text{ for $Y \in \TT$.}

1583: \end{equation}

1584: The first equation implies that $\eta$ is smooth away from the boundary.

1585: Suppose that $z \in U \setminus \partial S$ is a point where $\eta_z \neq 0$,

1586: and set $x = u(z) \in V \cap \pi^{-1}(U)$. Take a $(j,J)$-antilinear map $Z:

1587: TS_z \rightarrow TE^v_x$ such that $\leftsc \eta_z, Z \circ j \rightsc \neq 0$.

1588: One can see from \eqref{eq:tangent-compatible} that there is a $Y \in

1589: T\JJ(E,\pi,j)_J$ with $Y^{vh}_x = Z$, $Y^{vv}_x = 0$, and this will satisfy

1590: $\leftsc \eta_z, (Y \circ Du \circ j)_z \rightsc \neq 0$. By multiplying $Y$

1591: with a bump function supported near $x$, one can achieve that it lies in $\TT$

1592: and that $\int_S \leftsc \eta, Y \circ Du \circ j \rightsc \neq 0$, a

1593: contradiction. This means that $\eta\,|\, (U \setminus \partial S) = 0$. By

1594: unique continuation $\eta\,|\,(S \setminus \partial S) = 0$, which proves that

1595: $\eta = 0$. \qed

1596: 

1597: Take some $\zeta \in \partial S$ and consider the map $ev_\zeta: \BB

1598: \rightarrow Q_\zeta$, $ev_\zeta(u) = u(\zeta)$. The next result belongs to a

1599: type called ``transversality of evaluation''.

1600: 

1601: \begin{lemma} \label{th:ev-transversality}

1602: Let $g$ be a smooth map from some arbitrary manifold $G$ to $Q_\zeta$. Then,

1603: for any $J$ and $U$ as in the previous lemma, there are $J' \in

1604: \JJreg(E,\pi,Q,j)$ arbitrarily close to $J$, with $J' = J$ outside

1605: $\pi^{-1}(U)$, such that $ev_\zeta|\MM_{J'}$ is transverse to $g$.

1606: \end{lemma}

1607: 

1608: \proof In the same setup as before, one now has to prove that for $u \in \MM_J$

1609: and $x = u(\zeta) \in Q_\zeta$, the operator

1610: \[

1611: \WW^1_u \times \TT \longrightarrow \WW^0_{u,J} \times T(Q_\zeta)_x, \quad (X,Y)

1612: \longmapsto (D^{univ}_{u,J}(X,Y),X_\zeta)

1613: \]

1614: is onto. Take $(\eta,\xi)$ orthogonal to the image, with $\xi \in

1615: T(Q_\zeta)_x^\vee$. One still has \eqref{eq:adjoint} and as before it follows

1616: that $\eta = 0$. Then $\leftsc \xi, X_{\zeta} \rightsc = 0$ for all $X \in

1617: \WW^1_u$, so that $\xi = 0$ as well. \qed

1618: 

1619: For a given $(E,\pi,Q)$ and $\zeta$, one now proceeds as follows. After

1620: choosing some $j$ and a $J \in \JJreg(E,\pi,Q,j)$, one obtains a smooth compact

1621: moduli space $\MM_J$; and then one sets

1622: \[

1623: \Phi_1(E,\pi,Q) = (ev_\zeta)_*[\MM_J] \in H_*(Q_\zeta;\Z/2).

1624: \]

1625: This is independent of the choice of $j$, $J$ by a standard argument using

1626: parame\-trized moduli spaces. The same reasoning shows that it remains

1627: invariant under any ``smooth deformation'' of the geometric objects involved,

1628: that is to say of $Q,\Omega,\Theta$ or of the fibration $\pi: E \rightarrow S$

1629: itself, as long as one remains within the class of exact Lefschetz fibrations

1630: with Lagrangian boundary conditions. It seems pointless to formalize this

1631: notion of deformation, because the necessary conditions will be quite obviously

1632: satisfied in all our applications.

1633: 

1634: \begin{remark} \label{re:bordism}

1635: Since the regular spaces $\MM_J$ are actual manifolds, one can refine

1636: $\Phi_1(E,\pi,Q)$ by regarding it as an element of the unoriented bordism group

1637: $MO_*(Q_\zeta)$. But even with this refinement, it is far from capturing all

1638: the information contained in $\MM_J$. To get more sophisticated invariants one

1639: can use evaluation at several points, allowing those to move; we will now

1640: explain the simplest version of this. Given $\zeta \in \partial S$, choose a

1641: positively oriented path $c: [0;1] \rightarrow \partial S$, $c(0) = c(1) =

1642: \zeta$, parametrizing the boundary component on which $\zeta$ lies. Let

1643: $\phi_t: Q_{c(t)} \rightarrow Q_\zeta$ be the diffeomorphisms obtained from

1644: symplectic parallel transport along $c|[t;1]$. Denote by $\Delta, \Gamma

1645: \subset Q_\zeta^2$ the diagonal and the graph of $\phi_0$, respectively. For

1646: regular $J$, the parametrized evaluation map

1647: \[

1648: \widetilde{ev}_\zeta: [0;1] \times \MM_J \rightarrow [0;1] \times Q_\zeta^2,

1649: \quad \widetilde{ev}_\zeta(t,u) = (t,u(\zeta),\phi_t(u(c(t))))

1650: \]

1651: represents a class $\tilde{\Phi}_2(E,\pi,Q) \in H_*([0;1] \times Q_\zeta^2,

1652: \{0\} \times \Gamma \cup \{1\} \times \Delta;\Z/2)$, and this (as well as its

1653: cobordism version) is an invariant of $(E,\pi,Q)$. An example of this invariant

1654: will be computed in Remark \ref{re:vanishing}(iii).

1655: \end{remark}

1656: 

1657: Let $S^1,S^2$ be two compact surfaces with marked points $\zeta^k \in \partial

1658: S^k$, and denote by $S = S^1 \#_{\zeta^1 \sim \zeta^2} S^2$ their boundary

1659: connected sum (Figure \ref{fig:sum}). To be precise, one should choose oriented

1660: embeddings $\psi^k: \bar{D}^+(1) \rightarrow S^k$ of the closed half-disc

1661: $\bar{D}^+(1) = \{z \in \C \suchthat |z| \leq 1, \;\im\,z \geq 0\}$ into $S^k$,

1662: such that $\psi^k(0) = \zeta^k$ and $(\psi^k)^{-1}(\partial S^k) = [-1;1]$.

1663: Then, writing $D^+(\rho)$ for the open half-disc of some radius $0<\rho<1$, one

1664: forms $S$ by taking the two $S^k \setminus \psi^k(D^+(\rho))$ and identifying

1665: $\psi^1(z)$ with $\psi^2(-\rho^{-1}z)$. Suppose that we have exact Lefschetz

1666: fibrations $(E^k,\pi^k)$ over $S^k$, such that their symplectic connections are

1667: trivial on $\im(\psi^k)$; together with Lagrangian boundary conditions $Q^k

1668: \subset E^k$, such that the one-forms $\kappa_{Q^k}$ vanish on

1669: $\psi^k([-1;1])$; and finally, an exact syplectic manifold $M$ with an exact

1670: Lagrangian submanifold $L$, and isomorphisms $\phi^k: M \rightarrow

1671: (E^k)_{\zeta^k}$ satisfying $\phi^k(L) = (Q^k)_{\zeta^k}$. One can then glue

1672: the $(E^k,\pi^k)$ and $Q^k$ to an exact Lefschetz fibration over $S$. To do

1673: that, one first uses symplectic parallel transport to construct embeddings

1674: \[

1675: \xymatrix{

1676:  {\bar{D}^+(1) \times M} \ar[r]^-{\Psi^k} \ar[d] &

1677:  {E^k} \ar[d]^{\pi^k} \\

1678:  {\bar{D}^+(1)} \ar[r]^-{\psi^k} & {S^k}

1679: }

1680: \]

1681: such that $\Psi^k|\{0\} \times M = \phi^k$. These will satisfy

1682: \[

1683: (\Psi^k)^*\Omega^k = \o, \quad (\Psi^k)^*\Theta^k = \theta + dR^k, \quad

1684: (\Psi^k)^{-1}(Q^k) = [-1;1] \times L

1685: \]

1686: for some functions $R^k$. As in the pasting construction described in Section

1687: \ref{sec:basic}, one needs to introduce modified forms $\tilde{\Theta}^k$ such

1688: that $(\Psi^k)^*\tilde{\Theta}^k = \theta$ near $\{0\} \times M$. The functions

1689: $K_{Q^k}$ need to be modified accordingly, but that is rather straightforward,

1690: so we will not write it down explicitly. Now take the $E^k \setminus

1691: \Psi^k(D^+(\rho) \times M)$ and identify $\Psi^1(z,x)$ with $\Psi^2(-\rho

1692: z^{-1},x)$. This yields a manifold $E$ with a map $\pi: E \rightarrow S$, and

1693: the remaining data matches up, producing the structure of an exact Lefschetz

1694: fibration together with a Lagrangian boundary condition $Q$.

1695: %

1696: \includefigure{sum}{sum.eps}{hb}%

1697: 

1698: Assume that one has chosen complex structures $j^k$ on $S^k$ such that the

1699: $\psi^k$ are holomorphic; these determine a complex structure $j$ on $S$. Take

1700: $J^k \in \JJ(E^k,\pi^k,j^k)$ such that $(\Psi^k)^*(J^k)$ is the product of the

1701: standard complex structure on $\bar{D}^+(1)$ and of some fixed $\o$-compatible

1702: almost complex structure on $M$ (the same for both $k$). Then there is a

1703: canonical induced $J \in \JJ(E,\pi,j)$. Note that even though we have

1704: restricted the behaviour of $J^k$ over $\im(\Psi^k)$, it is still possible to

1705: choose them regular, by using the more precise statement in Lemma

1706: \ref{th:transversality}. Moreover, Lemma \ref{th:ev-transversality} says that

1707: for suitably chosen $J^k$, the evaluation maps $ev_{\zeta^k} | \MM_{J^k}:

1708: \MM_{J^k} \rightarrow Q^k_{\zeta_k} \iso L$ will be transverse to each other.

1709: 

1710: \begin{proposition} \label{th:gluing}

1711: Assume that $J^k \in \JJreg(E^k,\pi^k,Q^k,j^k)$ for $k = 1,2$, and that the

1712: $ev_{\zeta_k} | \MM_{J^k}$ are mutually transverse. Choose a sufficiently small

1713: parameter $\rho$ for the gluing. Then $J \in \JJreg(E,\pi,Q,j)$, and there is a

1714: diffeomorphism

1715: \[

1716: \MM_J \iso \MM_{J^1} \times_L \MM_{J^2},

1717: \]

1718: where the right hand side is the fibre product of

1719: $(ev_{\zeta^1},ev_{\zeta^2})$. \qed

1720: \end{proposition}

1721: 

1722: This is an average specimen of the ``gluing theorem'' type. The closest related

1723: argument in the literature would seem to be the gluing theory for

1724: pseudo-holomorphic discs from \cite[Section 18]{fukaya-oh-ohta-ono}, which is

1725: far more sophisticated than what we need here; as an alternative, one can

1726: probably adapt the proof of the more familiar gluing theorem for closed

1727: pseudo-holomorphic curves \cite[Section 6]{ruan-tian94}, \cite[Appendix

1728: A]{mcduff-salamon}, \cite{liu96}. The obvious next step would be to write down

1729: the outcome as a ``gluing formula''. However, while gluing will be important

1730: later, the situation then will be slightly different from that covered by

1731: Proposition \ref{th:gluing}. For this reason, further discussion is postponed

1732: to Section \ref{sec:relative-invariants}.

1733: 

1734: %\input{2b}

1735: 

1736: \subsection{Horizontality\label{sec:horizontal}}

1737: 

1738: Let $(E,\pi)$ be an exact Lefschetz fibration. Choose a complex structure $j$

1739: on its base. $J \in \JJ(E,\pi,j)$ is called horizontal if $J_x(TE^h_x) =

1740: TE_x^h$ for all $x \notin E^\crit$, or what is equivalent, if

1741: $\Omega(\cdot,J\cdot)$ is symmetric. In terms of \eqref{eq:compatible-j} these

1742: are just the $J$ with $J^{vh} = 0$, which shows that they form a contractible

1743: subspace $\JJ^h(E,\pi,j) \subset \JJ(E,\pi,j)$. The importance of horizontal

1744: almost complex structures is that they are sensitive to the geometry of the

1745: symplectic connection; the following is a particularly simple instance of this.

1746: 

1747: \begin{lemma} \label{th:first-illu}

1748: Let $(E,\pi)$ be an exact Lefschetz fibration over a compact surface $S$,

1749: $\partial S \neq \emptyset$, with a Lagrangian boundary condition $Q$. Assume

1750: that $(E,\pi)$ has nonnegative curvature, and that the boundary condition

1751: satisfies $\int_{\partial S} \kappa_Q < 0$. Then $\Phi_1(E,\pi,Q) = 0$.

1752: \end{lemma}

1753: 

1754: \proof For $J \in \JJ^h(E,\pi,j)$, take a symplectic form $\Omega + \pi^*\beta$

1755: as in Lemma \ref{th:tame}. Then $J$ is compatible with it in the ordinary sense

1756: of the word; we denote the associated metric by $||\cdot||$. Write $\Omega|TE^h

1757: = f(\pi^*\beta|TE^h)$ with $f \in \smooth(E \setminus E^\crit,\R)$. For any map

1758: $u$ from a compact Riemann surface, possibly with boundary, to $E$ there is the

1759: familiar equality $\half \int ||Du||^2 = \int u^*(\Omega + \pi^*\beta) + \int

1760: ||\bar\partial_J u||^2$. Specialize to sections $u$ and split $Du = (Du)^h +

1761: (Du)^v$ into horizontal and vertical parts. Since $||(Du)^h||^2 =

1762: 2(f(u)+1)\beta$, one obtains

1763: \begin{equation} \label{eq:deficiency}

1764: {\textstyle \half} \int_S ||(Du)^v||^2 + \int_S f(u)\beta = \int_S u^*\Omega +

1765: \int_S ||\bar\partial_J u||^2.

1766: \end{equation}

1767: This implies that $\MM_J = \emptyset$. In fact, the curvature assumption is

1768: just that $f \geq 0$, while for $u \in \MM_J$ one would have $\int_S u^*\Omega

1769: < 0$ by \eqref{eq:action} and the second assumption. \qed

1770: 

1771: A section $u: S \rightarrow E$ is called horizontal if $Du_z(TS_z) =

1772: (TE^h)_{u(z)}$ for all $z \in S$. To see the geometric meaning of this, it is

1773: convenient to exclude temporarily the presence of critical points, so that

1774: $(E,\pi)$ is an exact symplectic fibration.  If $u$ is horizontal, parallel

1775: transport along any path $c: [a;b] \rightarrow S$ carries $u(c(a)) \in

1776: E_{c(a)}$ to $u(c(b)) \in E_{c(b)}$. In other words, if $M$ is some fibre of

1777: $E$ and $x \in M$ the unique point through which $u$ passes, the structure

1778: group of the symplectic connection on $E$ is reduced from $\Sympe(M)$ to the

1779: subgroup $\Sympe(M,x)$ of maps preserving $x$. This entails a restriction on

1780: the curvature, namely, writing $\Omega|TE^h = f(\pi^*\beta|TE^h)$ for some

1781: positive $\beta \in \Omega^2(S)$, one has

1782: \begin{equation} \label{eq:critical-curvature}

1783: d(f|E_z)_{u(z)} = 0 \quad \text{for all $z \in S$}.

1784: \end{equation}

1785: If $u$ is horizontal, the symplectic vector bundle $u^*TE^v \rightarrow S$ has

1786: a preferred connection $\nabla^u$, obtained by linearizing parallel transport

1787: around $u$. Equivalently, this is induced from the connection on $E$ by the

1788: derivative map $\Sympe(M,x) \rightarrow Sp(TM_x)$. Explicitly $\nabla^u_ZX =

1789: u^*([Z^h,\tilde{X}])$, where $\tilde{X}$ is any section of $TE^v$ with

1790: $u^*\tilde{X} = X$. Using the canonical isomorphism $sp(V) \iso sym^2(V^*)$ for

1791: any symplectic vector space $V$, one can write the curvature of $\nabla^u$ as a

1792: two-form on $S$ with values in quadratic forms on the fibres of $u^*TE^v$. In

1793: those terms it is given by $F_{\nabla^u} = Hess(f|E_z)_{u(z)} \beta$, which is

1794: well-defined by \eqref{eq:critical-curvature}. We say that $\nabla^u$ is

1795: nonnegatively curved if all these Hessians are $\geq 0$. The ``infinitesimal

1796: deformations'' of a horizontal section $u$ are the covariantly constant

1797: sections, $\nabla^u X = 0$. If such a section exists, it further reduces the

1798: structure group of $(E,\pi)$ to the subgroup of maps in $\Sympe(M,x)$ which

1799: preserve a certain tangent vector at $x$. The resulting curvature restriction

1800: is

1801: \begin{equation} \label{eq:hessians}

1802: Hess(f)_{u(z)}(X(z),X(z)) = 0 \quad \text{for all $z \in S$.}

1803: \end{equation}

1804: Finally, if one readmits critical points, all the formulae derived above remain

1805: valid, with essentially the same proofs, except that when talking about the

1806: symplectic connection on $E$ it is necessary to restrict to a neighbourhood of

1807: a fixed horizontal section.

1808: 

1809: The connection between the two notions which we have introduced so far is that

1810: {\em a horizontal section is $(j,J)$-holomorphic for any horizontal $J$}. This

1811: provides a useful class of pseudo-holomorphic sections with a geometric origin,

1812: but it also raises a problem: if the space of horizontal sections has ``too

1813: large dimension'', one cannot find an almost complex structure which is both

1814: regular and horizontal. The rest of this section discusses this issue in more

1815: detail. Assume from now on that $S$ is compact with $\partial S \neq

1816: \emptyset$, and that we have a Lagrangian boundary condition $Q$. Write $\MM^h$

1817: for the space of horizontal sections $u$ which lie in $\BB$, meaning that

1818: $u(\partial S) \subset Q$. Since a horizontal section is determined by its

1819: value at any point, evaluation at $\zeta \in \partial S$ identifies $\MM^h$

1820: with a subset of $Q_\zeta$. As observed above, $\MM^h \subset \MM_J$ for all $J

1821: \in \JJ^h(E,\pi,j)$.

1822: 

1823: \begin{lemma} \label{th:horizontal-transversality}

1824: Let $U \subset S$ be a nonempty open subset, such that any partial section $u:

1825: U \rightarrow E|U$ which is horizontal and satisfies $u(\partial S \cap U)

1826: \subset Q$ is the restriction of some $u' \in \MM^h$. Then, given some $J \in

1827: \JJ^h(E,\pi,j)$, there are $J' \in \JJ^h(E,\pi,j)$ arbitrarily close to it and

1828: which agree with it outside $\pi^{-1}(U)$, with the property that any $u \in

1829: \MM_{J'} \setminus \MM^h$ is regular.

1830: \end{lemma}

1831: 

1832: \proof Take $V \subset E$ as in the proof of Lemma \ref{th:transversality}. In

1833: parallel with the argument there, we have to show that for any $u \in \MM_J

1834: \setminus \MM^h$,

1835: \begin{equation} \label{eq:h-universal}

1836: D_{u,J}^{univ}: \WW^1_u \times \TT^h \longrightarrow \WW^0_{u,J}

1837: \end{equation}

1838: is onto. Here $\TT^h = \TT \cap T\JJ^h(E,\pi,j)_J$ is the space of

1839: infinitesimal deformations $Y$ of $J$ within the class of horizontal almost

1840: complex structures, that is to say with $Y^{vh} = 0$ in

1841: \eqref{eq:tangent-compatible}, and such that $Y = 0$ outside $V \cap

1842: \pi^{-1}(U)$. As before, an $\eta$ which is orthogonal to the image of

1843: \eqref{eq:h-universal} is smooth away from the boundary and satisfies

1844: \begin{equation} \label{eq:h-orthogonal}

1845:  \int_S \leftsc \eta, Y \circ Du \circ j \rightsc = 0

1846:  \quad \text{for $Y \in \TT^h.$}

1847: \end{equation}

1848: Suppose that $u|U$ is horizontal. By assumption one could then find a $u' \in

1849: \MM^h$ with $u'|U = u|U$; unique continuation for pseudo-holomorphic curves

1850: would imply that $u' = u$, a contradiction. Hence there is a $z \in U$ such

1851: that $Du_z(TS_z) \neq (TE^h)_{u(z)}$. This is an open condition, so we can

1852: assume that $z \in U \setminus \partial S$. By choosing $Y$ suitably in

1853: $\TT^h$, one can make $(Y \circ Du \circ j)_z$ equal to any arbitrary

1854: $(j,J)$-antilinear homomorphism $TS_z \rightarrow (TE^v)_{u(z)}$. In

1855: particular, if $\eta_z \neq 0$ one can achieve that $\leftsc \eta_z, (Y \circ

1856: Du \circ j)_z \rightsc \neq 0$, which after multiplying with a cutoff function

1857: leads to a contradiction with \eqref{eq:h-orthogonal}. The same argument

1858: applies to all points close to $z$, proving that $\eta$ vanishes on a nonempty

1859: open subset, from which it follows that $\eta = 0$. \qed

1860: 

1861: In particular, taking $U = S$ shows that if $\MM^h = \emptyset$ then

1862: $\JJ^{reg,h}(E,\pi,Q,j) = \JJ^{reg}(E,\pi,Q,j) \cap \JJ^h(E,\pi,j) \subset

1863: \JJ^h(E,\pi,j)$ is a dense subset. In the same way one proves the following

1864: analogue of Lemma \ref{th:ev-transversality}:

1865: 

1866: \begin{lemma} \label{th:horizontal-ev-transversality}

1867: Let $g$ be a smooth map from an arbitrary manifold $G$ to $Q_\zeta$, for some

1868: $\zeta \in \partial S$. Let $U \subset S$ be a nonempty open subset, such that

1869: there are no horizontal partial sections $u: U \rightarrow E|U$ with

1870: $u(\partial S \cap U) \subset Q$. Given $J \in \JJ^h(E,\pi,j)$, there are $J'

1871: \in \JJ^{reg,h}(E,\pi,Q,j)$ arbitrarily close to it and which agree with it

1872: outside $\pi^{-1}(U)$, such that $ev_\zeta | \MM_{J'}$ is transverse to $g$.

1873: \qed

1874: \end{lemma}

1875: 

1876: Let $T_{Zar}(\MM^h)_u$ be the Zariski tangent space of $\MM^h$ at some section

1877: $u$. It consists of those $X \in \smooth(u^*TE^v)$ that satisfy $\nabla^uX = 0$

1878: and lie in $T\BB_u$, meaning that $X|\partial S \subset u^*(TQ \cap TE^v)$. The

1879: next result helps to determine when $u$, considered as a pseudo-holomorphic

1880: section for some horizontal almost complex structure, is regular.

1881: 

1882: \begin{lemma} \label{th:flat-deformations}

1883: Take $J \in \JJ^h(E,\pi,j)$ and $u \in \MM^h$. Then $T_{Zar}(\MM^h)_u \subset

1884: \ker\, D_{u,J}$, and if moreover $\nabla^u$ is nonnegatively curved, the two

1885: spaces are equal.

1886: \end{lemma}

1887: 

1888: \proof We take the second derivative of \eqref{eq:deficiency} at $u$. This is

1889: well-defined because the action $A$ is constant on $\BB$ and the other terms

1890: vanish to first order at $u$. The outcome is that for $X \in T\BB_u$,

1891: \begin{equation} \label{eq:weitzenboeck}

1892: \int_S ||\nabla^u X||^2 + \int_S Hess(f \circ u)(X,X) \,\beta = 2 \int_S

1893: ||D_{u,J}X||^2.

1894: \end{equation}

1895: If $\nabla^uX = 0$ then $Hess(f \circ u)(X,X) = 0$ by \eqref{eq:hessians},

1896: which implies that $D_{u,J}X = 0$. If $\nabla^u$ has nonnegative curvature, the

1897: converse also holds, since the second term in \eqref{eq:weitzenboeck} is $\geq

1898: 0$. \qed

1899: 

1900: Call $\MM^h$ clean if it is a smooth manifold and its tangent space is

1901: everywhere equal to $T_{Zar}\MM^h$. The next result can be considered as a

1902: limiting case of Lemma \ref{th:first-illu} (and is again just a sample

1903: application, in itself without any great importance, but hopefully

1904: instructive).

1905: 

1906: \begin{lemma}

1907: Assume that $(E,\pi)$ has nonnegative curvature, and that the Lagrangian

1908: boundary condition $Q$ satisfies $\int_{\partial S} \kappa_Q = 0$. In addition,

1909: assume that $\MM^h$ is clean and that its dimension agrees at every point with

1910: the index of $D_{u,J}$. Then $\Phi_1(E,\pi,Q) = [ev_{\zeta}(\MM^h)]$.

1911: \end{lemma}

1912: 

1913: \proof Take $J \in \JJ^h(E,\pi,j)$, and consider \eqref{eq:deficiency}. Since

1914: $f \geq 0$ and $\int_S u^*\Omega = 0$, it follows that $\MM_J = \MM^h$, and

1915: moreover that every $u \in \MM^h$ satisfies $f(u) \equiv 0$, which in turn

1916: implies $Hess(f|E_z)_{u(z)} \geq 0$. Applying Lemma \ref{th:flat-deformations}

1917: shows that $\dim\, \ker\, D_{u,J} = \dim\,T_{Zar}\MM^h = \dim\,\MM^h =

1918: \ind\,D_{u,J}$, from which one sees that $\coker\, D_{u,J} = 0$; hence $J$ is

1919: regular. \qed

1920: 

1921: %\input{2c}

1922: 

1923: \subsection{A vanishing theorem\label{sec:vanishing}}

1924: 

1925: Let $L$ be a framed exact Lagrangian sphere in an exact symplectic manifold

1926: $M$. According to Proposition \ref{th:standard-fibrations} one can associate to

1927: it a standard fibration $(E^L,\pi^L)$ over $\bar{D}(r)$ for some $r>0$. Each

1928: fibre $E^L_z$, $z \neq 0$, contains a distinguished Lagrangian sphere

1929: $\Sigma_z^L$, described by Lemma

1930: \ref{th:standard-properties}\ref{item:lagrangian-subbundle}; for $z = r$, the

1931: isomorphism $\phi^L: E^L_r \rightarrow M$ takes $\Sigma_r^L$ to $L$. Define the

1932: standard Lagrangian boundary condition for $(E^L,\pi^L)$ to be

1933: \begin{equation} \label{eq:standard-boundary}

1934: \begin{split}

1935: & \textstyle Q^L = \textstyle\bigcup_{z \in \partial\bar{D}(r)} \Sigma_z^L, \\

1936: & \kappa_{Q^L} =d^c(-\quarter|z|)\,|\,\partial\bar{D}(r), \qquad K_{Q^L} = 0.

1937: \end{split}

1938: \end{equation}

1939: This is indeed a Lagrangian boundary condition, as one can see from

1940: \eqref{eq:theta-sigma-l}. Our aim here is to prove the following result about

1941: the invariant $\Phi_1(E^L,\pi^L,Q^L) \in H_*(\Sigma_r^L;\Z/2) \iso

1942: H_*(L;\Z/2)$:

1943: 

1944: \begin{prop} \label{th:vanishing}

1945: $\Phi_1(E^L,\pi^L,Q^L) = 0$ for all $M$ and $L$.

1946: \end{prop}

1947: 

1948: The first part of the proof is a degeneration argument, in which one restricts

1949: the base to successively smaller discs. For $0< s \leq r$ set $Q^{L,s} =

1950: \textstyle \bigcup_{z \in \partial\bar{D}(s)} \Sigma^L_z$; together with

1951: $\kappa_{Q^{L,s}}$ and $K_{Q^{L,s}}$ as before, it is a Lagrangian boundary

1952: condition for $(E^{L,s},\pi^{L,s}) = (E^L,\pi^L)\,|\,\bar{D}(s)$. This

1953: constitutes a deformation of $(E^L,\pi^L,Q^L)$ in a suitable sense; which means

1954: that if one identifies $\Sigma_r^L$ with $\Sigma_s^L$ using parallel transport

1955: along $[s;r]$, then

1956: \begin{equation} \label{eq:deformation-equation}

1957: \Phi_1(E^L,\pi^L,Q^L) = \Phi_1(E^{L,s},\pi^{L,s},Q^{L,s})

1958: \end{equation}

1959: for all $s$. Now fix some $J \in \JJ^h(E^L,\pi^L,j)$, where $j$ is the standard

1960: complex structure on $\bar{D}(r)$. Let $\MM^{L,s}$ be the space of sections

1961: $\bar{D}(s) \rightarrow E^{L,s}$ which are holomorphic with respect to

1962: $j|\bar{D}(s)$ and $J|E^{L,s}$, and have boundary in $Q^{L,s}$. As $s

1963: \rightarrow 0$, $Q^{L,s}$ shrinks to the single critical point $\Sigma^L_0 =

1964: \{x_0\}$ of $\pi^L$, and we would like to apply a compactness argument to

1965: elements of $\MM^{L,s}$ in the limit. This looks a bit unpleasant as it stands,

1966: but one can modify the situation to make the degeneration of the $Q^{L,s}$ less

1967: singular, and then standard Gromov compactness is sufficient.

1968: 

1969: \begin{lemma} \label{th:resolution}

1970: For some $0<r'\leq r$, there is a compact almost complex manifold with corners

1971: $(\hatE^L,\hatJ)$, together with pseudo-holomorphic maps

1972: \[

1973: \xymatrix{

1974:  {\hatE^L} \ar[r]^{\eta^L} \ar[d]_{\hat\pi^L} & {E^L} \ar[d]^{\pi^L} \\

1975:  {\bar{D}(r') \times \bar{D}(1)} \ar[r]^-{m} & {\bar{D}(r)}

1976: }

1977: \]

1978: where $m(w,z) = wz$ is multiplication, such that the following properties are

1979: satisfied.

1980: \begin{romanlist}

1981: \item \label{item:resolution-one}

1982: Away from $Z^L = (\eta^L)^{-1}(x_0) \cap (\hat\pi^L)^{-1}(0,0)$, the map

1983: $\eta^L$ identifies $(\hatE^L,\hat\pi^L)$ with the pullback of $(E^L,\pi^L)$ by

1984: $m$. In particular, restricting to any $w = s>0$ gives a pseudo-holomorphic

1985: diffeomorphism

1986: \[

1987:  \hatE^{L,s} = (\hat\pi^L)^{-1}(\{s\} \times \bar{D}(1))

1988:  \longrightarrow E^{L,s}.

1989: \]

1990: \item \label{item:resolution-two}

1991: $\hatE^L$ carries a symplectic form $\hatO^L$ which tames its almost complex

1992: structure, and which is of the form

1993: \[

1994:  \hatO^L = (\eta^L)^*\Omega^L + (\hat\pi^L)^*(

1995:  \textstyle\frac{i}{2} dw \wedge d\bar{w} +

1996:  \frac{i}{2} dz \wedge d\bar{z} + d\gamma) + \delta,

1997: \]

1998: with $\gamma = \frac{1}{4} \im((r'-w)\bar{z}dz)$, and where $\delta$ is

1999: supported in a small neighbourhood of $Z^L$.

2000: 

2001: \item \label{item:resolution-three}

2002: $(\hatE^L,\hatO^L)$ contains a Lagrangian submanifold with boundary $\hatQ^L$

2003: which projects to $[0;r'] \times S^1 \subset \bar{D}(r') \times \bar{D}(1)$ and

2004: satisfies

2005: \[

2006: \eta^L(\hatQ^L \cap (\hat\pi^L)^{-1}(\{s\} \times S^1)) = \begin{cases} Q^{L,s}

2007: & s>0,

2008: \\ \{x_0\} & s = 0.

2009: \end{cases}

2010: \]

2011: \end{romanlist}

2012: \end{lemma}

2013: 

2014: It is convenient to begin with the local model for the construction. Choose

2015: some $r'>0$ and set

2016: \begin{equation} \label{eq:cylinder}

2017: \begin{split}

2018:  & E = \{x \in \C^{n+1} \suchthat ||x|| \leq 2\sqrt{r'},\; |q(x)| \leq r'\},

2019:  \qquad \Omega = \o_{\C^{n+1}}|E, \\

2020:  & \pi = q: E \longrightarrow \bar{D}(r'),

2021:  \qquad Q = \textstyle \bigcup_{z \in \partial \bar{D}(r')} \Sigma_z

2022: \end{split}

2023: \end{equation}

2024: where $q$ is our standard quadratic function, and $\Sigma_z$ is as in

2025: \eqref{eq:model-spheres}. In parallel with the reasoning above, one can also

2026: introduce $(E^s,\pi^s) = (E,\pi)|\bar{D}(s)$ and $Q^s = \bigcup_{z \in \partial

2027: \bar{D}(s)} \Sigma_z$ for $s \in (0;r']$. Take the pullback of $(E,\pi)$ by the

2028: multiplication map, $m^*E = \{(w,z,x) \in \bar{D}(r') \times \bar{D}(1) \times

2029: E \suchthat q(x) = wz\} \subset \C^{n+3}$. This has one singular point

2030: $(0,0,0)$, which can be resolved by blowing it up inside $\C^{n+3}$ and taking

2031: the proper transform of $m^*E$. The outcome, which we denote by $\hatE$, comes

2032: with maps

2033: \[

2034: \xymatrix{

2035:  {\hatE} \ar[r]^{\eta} \ar[d]_{\hat\pi} & {E} \ar[d]^{\pi} \\

2036:  {\bar{D}(r') \times \bar{D}(1)} \ar[r]^-{m} & {\bar{D}(r')}.

2037: }

2038: \]

2039: \begin{condensedprimelist}

2040: \item \label{item:model-resolution-one} \em

2041: $\eta$ is a pullback map away from $Z = \eta^{-1}(0) \cap \hat\pi^{-1}(0,0)$.

2042: \end{condensedprimelist}

2043: That is obvious from the definition. Consider the two-form

2044: \begin{equation} \label{eq:pre-form} \Omega + \textstyle

2045: \frac{i}{2} dw \wedge d\bar{w} + \frac{i}{2} dz \wedge d\bar{z} + d\gamma =

2046: \o_{\C^{n+3}} + d\gamma

2047: \end{equation}

2048: on $\bar{D}(r') \times \bar{D}(1) \times E$, where $\gamma$ is as in Lemma

2049: \ref{th:resolution}\ref{item:resolution-two}. We claim that this is symplectic

2050: and tames the obvious complex structure. All one needs to verify is that

2051: $(d\gamma)(\cdot,i\cdot)$ is nonnegative, which one can do by decomposing

2052: $d\gamma = (d\gamma)^{1,1} + (d\gamma)^{0,2} + (d\gamma)^{2,0}$; then

2053: $(d\gamma)^{1,1} = \re(r'-w)\frac{i}{4} dz \wedge d\bar{z}$ is nonnegative

2054: since $\re(r'-w) \geq 0$, while $(d\gamma)^{0,2}(\cdot,i\cdot)$ vanishes

2055: because it is a two-form of type $(0,2)$, and similarly for $(d\gamma)^{2,0}$.

2056: Now pull back \eqref{eq:pre-form} to the blowup, make it symplectic by adding a

2057: two-form $\delta$ supported near the exceptional divisor, and restrict that to

2058: $\hatE$. The outcome is:

2059: 

2060: \begin{condensedprimelist}

2061: \setcounter{enumi}{1}

2062: \item \label{item:model-resolution-two} \em

2063: There is a symplectic form $\hatO$ on $\hatE$ which tames the complex

2064: structure, and which is of the form $\eta^*\Omega + (\hat\pi)^*(\frac{i}{2} dw

2065: \wedge d\bar{w} + \frac{i}{2} dz \wedge d\bar{z} + d\gamma) + \delta$, with

2066: $\delta$ supported in a small neighbourhood of $Z$.

2067: \end{condensedprimelist}

2068: 

2069: The subset

2070: \begin{equation} \label{eq:hatq}

2071:  \{ (w,z,x) \suchthat w \in [0;r'], \; |z| = 1, \; x \in

2072:  \Sigma_{wz}\} \subset m^*E

2073: \end{equation}

2074: is a submanifold, since it is the image of the embedding $S^1 \times_{\Z/2}

2075: \bar{B}^{n+1}(1) \rightarrow \C^{n+3}$, $(z,y) \mapsto (|y|^2,z^2,zy)$. It is

2076: also Lagrangian with respect to \eqref{eq:pre-form}; which is in fact the

2077: reason for our choice of symplectic form. Define $\hatQ$ to be the preimage of

2078: \eqref{eq:hatq} in $\hatE$; provided that the support of $\delta$ has been

2079: chosen sufficiently small, this is again a Lagrangian submanifold. By

2080: construction one has

2081: 

2082: \begin{condensedprimelist}

2083: \setcounter{enumi}{2}

2084: \item \label{item:model-resolution-three} \em

2085: $\hat\pi(\hatQ) = [0;r] \times S^1$, and

2086: \[

2087: \eta(\hatQ \cap \hat\pi^{-1}(\{s\} \times S^1)) = \begin{cases} Q^s & s>0,

2088: \\ \{0\} & s = 0.

2089: \end{cases}

2090: \]

2091: \end{condensedprimelist}

2092: 

2093: \proof[Proof of Lemma \ref{th:resolution}] Take a holomorphic Morse chart

2094: $(\xi,\Xi)$ around $x_0 \in E_0^L$ as provided by Lemma

2095: \ref{th:standard-properties}\ref{item:preferred-charts}. After restricting the

2096: domains, we may assume that $\xi$ is the inclusion $\bar{D}(r') \hookrightarrow

2097: \bar{D}(r)$ for some $0<r' \leq r$, that $\Xi$ is defined on the set $E \subset

2098: \C^{n+1}$ from \eqref{eq:cylinder}, and that $Q^{L,s} = \Xi(Q^s)$ for all $0 <

2099: s \leq r'$. Construct a diagram

2100: \[

2101: \xymatrix{

2102:  {\hatE} \ar[d] \ar[r] \ar@/^1pc/[rr]^{\eta^L} &

2103:  {m^*E} \ar[d]^{m^*\Xi} \ar[r] & {E} \ar[d]^{\Xi} \\

2104:  {\hatE^L} \ar[dr]_{\hat\pi^L} \ar[r] &

2105:  {m^*E^L} \ar[r] \ar[d] &

2106:  {E^{L,r'}} \ar[d]^{\pi^{L,r'}} \\

2107:  & {\bar{D}(r') \times \bar{D}(1)} \ar[r]^{m} &

2108:  {\bar{D}(r')}

2109:  }

2110: \]

2111: as follows: pull back $E^L$ by multiplication $m$; the map $m^*\Xi$ induced by

2112: $\Xi$ identifies neighbourhoods of the singular points $(0,0,0) \in m^*E$ and

2113: $(0,0,x_0) \in m^*E^L$; which means that there is a resolution \[\hatE^L

2114: \longrightarrow m^*E^L\] modelled locally on $\hatE \rightarrow m^*E$. The

2115: given almost complex structure $J$ induces one on $m^*E$, for which $m^*\Xi$ is

2116: holomorphic; this and the complex structure on $\hatE$ define the almost

2117: complex structure $\hatJ$. Part \ref{item:resolution-one} of the lemma now

2118: follows from the corresponding statement \ref{item:model-resolution-one} in the

2119: local model. Next, take the two-form $\Omega^L + \frac{i}{2} dw \wedge d\bar{w}

2120: + \frac{i}{2} dz \wedge d\bar{z} + d\gamma$ on $m^*E^L$; due to the nonnegative

2121: curvature of standard fibrations, see Lemma

2122: \ref{th:standard-properties}\ref{item:nonnegatively-curved}, and to the fact

2123: that $J$ is horizontal, this will tame the almost complex structure away from

2124: the singular point. One gets $\hatO^L$ by gluing this together with the

2125: symplectic form $\hatO$ from the local model, and that proves

2126: \ref{item:resolution-two}. Similarly, the definition of $\hatQ^L$ follows that

2127: of $\hatQ$, and \ref{item:model-resolution-three} implies

2128: \ref{item:resolution-three}. \qed

2129: 

2130: Returning to the moduli spaces $\MM^{L,s}$ of pseudo-holomorphic sections, we

2131: can now carry out the compactness argument mentioned above:

2132: 

2133: \begin{lemma} \label{th:degeneration}

2134: Choose some neighbourhood of the critical point $x_0 \in E^L$. For sufficiently

2135: small $s$, the image of all $u \in \MM^{L,s}$ will lie in that neighbourhood.

2136: \end{lemma}

2137: 

2138: \proof Let $(s_k)$ be a sequence in $(0;r']$ converging to zero, and $u_k \in

2139: \MM^{L,s_k}$ a corresponding sequence of sections. By Lemma

2140: \ref{th:resolution}\ref{item:resolution-one}, there is for each $k$ a unique

2141: pseudo-holomorphic map \[\hat{u}_k: \bar{D}(1) \longrightarrow \hatE^L\] with

2142: $\eta^L(\hat{u}_k(z)) = u_k(s_kz)$ and $\hat{\pi}^L(\hat{u}_k(z)) = (s_k,z)$.

2143: By part \ref{item:resolution-three} of the same lemma, this maps $S^1$ to

2144: $\hatQ^L$. Moreover, its energy is independent of $k$, since

2145: \begin{align*}

2146:  & \int_{\bar{D}(1)} \hat{u}_k^*\hatO^L

2147:  = \int_{\bar{D}(s_k)} u_k^*\O^L + \int_{\bar{D}(1)}

2148:  {\textstyle \frac{i}{2} dz \wedge d\bar{z}} + \int_{\bar{D}(1)}

2149:  {\textstyle \frac{i}{4} (r'-s_k) \im(d\bar{z} \wedge dz)} +

2150:  \\ & \qquad + \int_{\bar{D}(1)}

2151:  \hat{u}^*_k\delta

2152:  = {\textstyle \frac{\pi}{2}} s_k + \pi +

2153:  {\textstyle \frac{\pi}{2}}(r'-s_k) + 0 = \pi(1 + \textstyle\frac{r'}{2}).

2154: \end{align*}

2155: Here we have used \eqref{eq:theta-sigma-l} for the first term; and the last

2156: term is zero because, without changing its cohomology class, one can modify

2157: $\delta$ to make its support arbitrarily close to $Z$, in which case, since

2158: $u_k$ avoids the critical point $x_0$, $\hat{u}_k$ would not meet the support

2159: of $\delta$. One can apply Gromov compactness to the sequence $\hat{u}_k:

2160: (\bar{D}(1), S^1) \rightarrow (\hatE^L,\hatQ^L)$; the limit of some subsequence

2161: will be a pseudo-holomorphic ``cusp disc'' or ``stable disc'' $\hat{u}_\infty$.

2162: Since $\im(\hat{u}_k) \subset (\hat\pi^L)^{-1}(\{s_k\} \times \bar{D}(1))$, we

2163: have $\im(\hat{u}_\infty) \subset (\hat\pi^L)^{-1}(\{0\} \times \bar{D}(1))$.

2164: Composing with $\eta^L$ yields a ``cusp disc'' $u_\infty$ in $E$ whose boundary

2165: lies in $\eta^L(\hatQ^L \cap (\hat{\pi}^L)^{-1}(\{0\} \times S^1)) = \{x_0\}$.

2166: Because $\Omega^L$ is exact, $u_\infty$ is necessarily constant equal to $x_0$.

2167: On the other hand, the image of $u_k$, for $k$ large and in our subsequence,

2168: lies in an arbitrarily small neighbourhood of the image of $u_\infty$. The rest

2169: is straightforward. \qed

2170: 

2171: The second part of the proof of Proposition \ref{th:vanishing} is an explicit

2172: computation in the local model $(E,\pi)$ from \eqref{eq:cylinder}. Note that

2173: while this is not an exact Lefschetz fibration, it still makes sense to

2174: consider the spaces $\MM^s$ of holomorphic sections $w: \bar{D}(s) \rightarrow

2175: E^s$ with boundary in $Q^s$. In fact $Q^s$ is a Lagrangian submanifold of

2176: $\C^{n+1}$, as one can see from \eqref{eq:theta-sigma}, and the maximum

2177: principle ensures that any holomorphic disc in $\C^{n+1}$ with boundary in

2178: $Q^s$ is actually contained in $E^s$.

2179: 

2180: \begin{lemma} \label{th:explicit-sections}

2181: $\MM^s$ consists of maps $w(z) = s^{-1/2} az + s^{1/2} \bar{a}$, where $a \in

2182: \C^{n+1}$ satisfies $q(a) = 0$ and $||a||^2 = 1/2$. All these maps are regular,

2183: in the sense that the associated Fredholm operators are surjective.

2184: \end{lemma}

2185: 

2186: \proof The holomorphic functions $v: \bar{D}(s) \rightarrow \C$ which satisfy

2187: $v(z) \in z^{1/2}\R$ for all $z \in \partial \bar{D}(s)$ are $v(z) = s^{-1/2} c

2188: z + s^{1/2} \bar{c}$, for $c \in \C$. All components of $w \in \MM^s$ must be

2189: of this form, and the conditions on $a$ come from $q(w(z)) = z$. As for

2190: regularity, since we are dealing with the standard complex structure $J_0$ on

2191: $\C^{n+1}$, $D_{w,J_0}$ is an actual $\bar\partial$-operator, see

2192: \eqref{eq:connection-formula}. Its kernel consists of holomorphic maps $X:

2193: \bar{D}(s) \rightarrow \C^{n+1}$ such that $X(z) \in \sqrt{z}\R^n$ for $z \in

2194: \partial \bar{D}(s)$, and $Dq(w(z))X(z) = 2 \sum_k w_k(z) X_k(z) = 0$ for all

2195: $z$. The same argument as before determines all such $X$ explicitly, the

2196: outcome being that $\ker\,D_{w,J_0} \iso \R^{2n-1}$. Using the Riemann-Roch

2197: formula for surfaces with boundary one computes that $\ind\,D_{w,J_0} = 2n-1$,

2198: which shows that the cokernel is zero. \qed

2199: 

2200: The condition on $a$ can be written as

2201: \[

2202: ||\re\,a||^2 = ||\im\,a||^2 = 1/4, \quad \leftsc \re\,a,  \im\,a \rightsc = 0

2203: \]

2204: so that $\MM^s$ can be identified with the sphere bundle $S(T^*S^n)$ by mapping

2205: $a$ to $(u,v) = (-2\, \im\,a, 2\, \re\,a)$. With this and the diffeomorphism

2206: $S^n \rightarrow \Sigma_s$, $x \mapsto \sqrt{s}x$, one can identify the

2207: evaluation $ev_s: \MM^s \rightarrow \Sigma_s$ with the projection to the base

2208: $S(T^*S^n) \rightarrow S^n$. This clearly represents the zero cycle in

2209: $H_*(S^n;\Z/2)$. Hence, if one defined an invariant $\Phi_1(E^s,\pi^s,Q^s)$ in

2210: this local model, it would vanish.

2211: 

2212: \proof[Proof of Proposition \ref{th:vanishing}] As in the proof of Lemma

2213: \ref{th:degeneration}, we consider a holomorphic Morse chart $(\xi,\Xi)$ with

2214: $\Xi$ defined on $E \subset \C^{n+1}$ for some $r'>0$. We know that for some

2215: sufficiently small $s>0$, all $u \in \MM^{L,s}$ lie in $\im(\Xi)$. Recall that

2216: $\Xi$ is holomorphic with respect to the standard complex structure $J_0$ on

2217: $E$ and to $J$ on $E^L$; and that it takes $Q^s$ to $Q^{L,s}$. This implies

2218: that composition with $\Xi$ yields a diffeomorphism

2219: \[

2220: \xymatrix{

2221:  {\MM^s} \ar[r]^{\iso} \ar[d]_{ev_s} & {\MM^{L,s}} \ar[d]_{ev_s} \\

2222:  {\Sigma_s} \ar[r]^{\Xi} & {\Sigma^L_s.}

2223: }

2224: \]

2225: It is easy to see that, for $w \in \MM^s$ and $u = \Xi \circ w \in \MM^{L,s}$,

2226: the kernels and cokernels of the associated operators $D_{w,J_0}$, $D_{u,J}$

2227: coincide. This shows that $J|E^{L,s}$ is regular, and that the evaluation cycle

2228: $\MM^{L,s} \rightarrow \Sigma^L_s$ can be identified with projection $S(T^*S^n)

2229: \rightarrow S^n$, which as observed before is zero in homology; together with

2230: \eqref{eq:deformation-equation} this completes the argument. \qed

2231: 

2232: \begin{remarks} \label{re:vanishing}

2233: (i) As pointed out in Remark \ref{re:bordism}, $\Phi_1$ can be refined to an

2234: invariant taking values in unoriented bordism. Proposition \ref{th:vanishing}

2235: also applies to this refinement, since \eqref{eq:deformation-equation} comes

2236: from a cobordism of moduli spaces and the projection $S(T^*S^n) \rightarrow

2237: S^n$ is obviously cobordant to zero.

2238: 

2239: (ii) Suppose that $(E^1,\pi^1,Q^1)$ is a standard fibration with its standard

2240: boundary condition, and that we have another exact Lefschetz fibration

2241: $(E^2,\pi^2,Q^2)$, such that the two can be glued together to $(E,\pi,Q)$ as in

2242: Section \ref{sec:simple-invariant}. Proposition \ref{th:vanishing} together

2243: with Proposition \ref{th:gluing} implies that $\Phi_1(E,\pi,Q)$ will then be

2244: zero. By pushing this reasoning a little further, one arrives at the following

2245: result: {\em let $(E,\pi)$ be an arbitrary exact Lefschetz fibration over a

2246: compact base $S$, with Lagrangian boundary condition $Q$. Assume that there is

2247: a path $c: [a;b] \rightarrow S$ with $c(a) \in \partial S$, $c^{-1}(S^\crit) =

2248: \{b\}$, and $c'(b) \neq 0$, whose vanishing cycle $V_{c(a)}$ is isotopic to

2249: $Q_{c(a)}$ as an exact Lagrangian submanifold in $E_{c(a)}$; then

2250: $\Phi_1(E,\pi,Q) = 0$}.

2251: 

2252: (iii) Despite the vanishing of their $\Phi_1$-invariant, the moduli spaces of

2253: pseudo-holomorphic sections of a standard fibration are never empty. To see

2254: this, one has to use the invariant $\tilde{\Phi}_2$ mentioned in Remark

2255: \ref{re:bordism}. The computation of this can be reduced to the local model in

2256: the same way as before, using Lemma \ref{th:degeneration}. The relevant

2257: parametrized evaluation map then becomes

2258: \begin{equation} \label{eq:para-e}

2259: \begin{split}

2260:  & [0;1] \times \MM^s \longrightarrow [0;1] \times

2261: \Sigma_s \times \Sigma_s, \\

2262:  & (t,w) \longmapsto (t,w(s),e^{\pi i(1-t)} w(s e^{2\pi i t})).

2263: \end{split}

2264: \end{equation}

2265: Here $\Sigma_{s e^{2\pi i t}} \rightarrow \Sigma_s$, $x \mapsto e^{\pi

2266: i(1-t)}x$ arises as parallel transport along $\partial\bar{D}(s)$ in positive

2267: direction. Using Lemma \ref{th:explicit-sections} one can make

2268: \eqref{eq:para-e} even more concrete, identifying it with

2269: \begin{align*}

2270:  & [0;1] \times S(T^*S^n) \longrightarrow [0;1] \times S^n \times S^n, \\

2271:  & (t,u,v) \longmapsto (t,v,-cos(\pi t)v - \sin(\pi t)u).

2272: \end{align*}

2273: It is now easy to see that $\tilde{\Phi}_2(E^L,\pi^L,Q^L)$ is nonzero: it is

2274: the image of the fundamental class $[S^n \times S^n]$ under the map

2275: \begin{multline*}

2276: H_*(S^n \times S^n;\Z/2) \rightarrow H_*(S^n \times S^n, \Delta \cup

2277: \overline{\Delta};\Z/2) \iso \\ \iso H_*([0;1] \times S^n \times S^n, \{0\}

2278: \times \overline{\Delta} \cup \{1\} \times \Delta;\Z/2)

2279: \end{multline*}

2280: where $\Delta$, $\overline{\Delta}$ are the diagonal and antidiagonal,

2281: respectively.

2282: \end{remarks}

2283: 

2284: %\input{2d}

2285: 

2286: \subsection{Relative invariants\label{sec:relative-invariants}}

2287: 

2288: A surface with strip-like ends is an oriented connected surface $S$, together

2289: with finite sets $I^-,I^+$ and oriented proper embeddings $\{\gamma_e: \R^-

2290: \times [0;1] \rightarrow S\}_{e \in I^-}$, $\{\gamma_e: \R^+ \times [0;1]

2291: \rightarrow S\}_{e \in I^+}$, such that $\gamma_e^{-1}(\partial S) = \R^{\pm}

2292: \times \{0;1\}$. The images of the $\gamma_e$ (the ends of $S$) should be

2293: mutually disjoint, and the complement of the union of all ends should be a

2294: relatively compact subset of $S$. We will always assume that there is at least

2295: one end.

2296: 

2297: Define an {\em exact Lefschetz fibration trivial over the ends of $S$} to be an

2298: exact Lefschetz fibration $(E,\pi)$, whose regular fibres are isomorphic to

2299: some exact symplectic manifold $M$, together with smooth trivializations

2300: $\{\Gamma_e: \R^- \times [0;1] \times M \rightarrow \gamma_e^*E\}_{e \in I^-}$,

2301: $\{\Gamma_e: \R^+ \times [0;1] \times M \rightarrow \gamma_e^*E\}_{e \in I^+}$,

2302: such that $\Gamma_e^*\Omega$, $\Gamma_e^*\Theta$ are equal to the pullbacks of

2303: $\o$,$\theta$ by the projection $\R^{\pm} \times [0;1] \times M \rightarrow M$.

2304: When considering a Lagrangian boundary condition $Q$ for such an exact

2305: Lefschetz fibration, we will always impose the additional condition that

2306: $\kappa_Q$ vanishes on $\im(\gamma_e) \cap \partial S$. To see the significance

2307: of this, recall that the family $Q_z \subset E_z$ of Lagrangian submanifolds is

2308: preserved by symplectic parallel transport along $\partial S$. Since the

2309: symplectic connection is trivial on the ends, it follows that

2310: $

2311:  \Gamma_e^{-1}(Q) =

2312:  (\R^\pm \times \{0\} \times L_{e,0}) \cup

2313:  (\R^\pm \times \{1\} \times L_{e,1})

2314: $

2315: for some pair of Lagrangian submanifolds $L_{e,0},L_{e,1} \subset M$; we say

2316: that $Q$ is modelled on $(L_{e,0},L_{e,1})$ over the end $e$. Using the

2317: definition of a Lagrangian boundary condition, one finds that

2318: \[

2319:  d(K_Q \circ \Gamma_e) \,|\, \R^{\pm} \times \{k\} \times L_{e,k} =

2320:  (\theta | L_{e,k}) - (\gamma_e^*\kappa_Q | \R^{\pm} \times \{k\}).

2321: \]

2322: Our assumption is that the second term on the right vanishes, which implies

2323: that $K_Q(\Gamma_e(s,k,y))$ is independent of $s$. It follows that $L_{e,k}$ is

2324: an exact Lagrangian submanifold in a canonical way, with associated function

2325: $K_{L_{e,k}}(y) = K_Q(\Gamma_e(s,k,y))$.

2326: 

2327: In this situation, and under the additional assumption that the intersections

2328: $L_{e,0} \cap L_{e,1}$ are transverse for all $e$, we will associate to

2329: $(E,\pi)$ and $Q$ a relative invariant, which is a map between Floer cohomology

2330: groups

2331: \[

2332:  \Phi_0^{rel}(E,\pi,Q) : \bigotimes_{e \in I^+} HF(L_{e,0},L_{e,1})

2333:  \longrightarrow \bigotimes_{e \in I^-} HF(L_{e,0},L_{e,1}).

2334: \]

2335: This is a modified version of the frameworks described in \cite{schwarz95},

2336: \cite{piunikhin-salamon-schwarz94} and \cite{seidel97}. The fact that

2337: $\Phi^{rel}_0$ goes from positive to negative ends has to do with the use of

2338: Floer {\em co}homology, which we think of as behaving contravariantly. This is

2339: of course largely a matter of convention. In any case, there is no real

2340: difference between positive and negative ends, as one can be turned into the

2341: other by switching from $\gamma_e(s,t)$ to $\gamma_e(-s,1-t)$. Doing that does

2342: not change the relative invariant, up to the ``Poincar{\'e} duality''

2343: isomorphism $HF(L_{e,0},L_{e,1}) \iso HF(L_{e,1},L_{e,0})^\vee$. One could

2344: therefore formulate the theory using only one kind of ends, bringing it closer

2345: to a (1+1)-dimensional TQFT. Finally, we should mention that the transverse

2346: intersection condition can be lifted. This goes by a standard argument, using

2347: the same ``continuation map'' technique as the proof of isotopy invariance of

2348: Floer cohomology. We will not discuss that further, since it is not necessary

2349: for our immediate purpose.

2350: 

2351: To begin with, a brief review of Floer cohomology, just to remind the

2352: reader of the special features of the ``exact'' situation. Let $M$ be

2353: an exact symplectic manifold and $(L_0,L_1)$ a pair of transversally

2354: intersecting exact Lagrangian submanifolds. The action functional on

2355: the path space $\PP(L_0,L_1) = \{c \in \smooth([0;1], M) \suchthat

2356: c(0) \in L_0, \; c(1) \in L_1 \}$ is

2357: \[

2358: a_{L_0,L_1}(c) = -\!\int c^*\theta \;+ K_{L_1}(c(1)) - K_{L_0}(c(0))

2359: \]

2360: (as pointed out to me by Oh, this form of the action functional

2361: appears in \cite{ohnew}, and is then used crucially in his papers

2362: with Milinkovic). Its critical points are the constant paths $c_x$ at

2363: points $x \in L_0 \cap L_1$. We write $a_{L_0,L_1}(x) =

2364: a_{L_0,L_1}(c_x) = K_{L_1}(x) - K_{L_0}(x)$ for their action. The

2365: Floer cochain space $CF(L_0,L_1)$ is the vector space over $\Z/2$

2366: with a basis given by these points; we denote the canonical basis

2367: vectors by $\gen{x}$. Let $\JJ(M)$ be the space of smooth families $J

2368: = (J_t)_{0 \leq t \leq 1}$ of almost complex structures on $M$, each

2369: of which is $\o$-compatible and convex near the boundary. For $J \in

2370: \JJ(M)$ and $x_{\pm} \in L_0 \cap L_1$, one considers the space

2371: $\FF_J(x_-,x_+)$ of maps

2372: \begin{equation} \label{eq:floer}

2373: \begin{cases}

2374:  & \!\! \sigma = \sigma(s,t): \R \times [0;1] \longrightarrow M, \\

2375:  & \!\! \sigma(\R \times \{0\}) \subset L_0, \;

2376:  \sigma(\R \times \{1\}) \subset L_1, \\

2377:  & \!\! \partial \sigma/\partial t = J_t \,\partial \sigma/\partial s, \\

2378:  & \!\! \lim_{s \rightarrow \pm \infty} \sigma(s,\cdot) = c_{x_{\pm}}

2379: \end{cases}

2380: \end{equation}

2381: where the limit is understood to be in the $C^1$-topology on $\PP(L_0,L_1)$.

2382: There is a natural $\R$-action on $\FF_J(x_-,x_+)$ by translation in

2383: $s$-direction. We denote by $\FF_J^*(x_-,x_+)$ the subspace of maps which are

2384: not $\R$-invariant. Solutions $\sigma$ of \eqref{eq:floer} can be interpreted

2385: as negative gradient flow lines for $a_{L_0,L_1}$ in an $L^2$-metric on

2386: $\PP(L_0,L_1)$; of course, this implies that $\FF_J^*(x_-,x_+) = \emptyset$

2387: whenever $a_{L_0,L_1}(x_-) \leq a_{L_0,L_1}(x_+)$.

2388: 

2389: To each $\sigma \in \FF_J(x_-,x_+)$ one can associate an operator

2390: $D_{\sigma,J}$ which linearizes \eqref{eq:floer}, and which is Fredholm in

2391: suitable Sobolev spaces. There is a dense subspace $\JJ^{reg}(M,L_0,L_1)

2392: \subset \JJ(M)$ of almost complex structures $J$ for which all $D_{\sigma,J}$

2393: are onto, and then the spaces $\FF_J(x_-,x_+)$, as well as the quotients

2394: $\FF_J^*(x_-,x_+)/\R$, are smooth manifolds. In that case one defines

2395: $n_J(x_-,x_+) \in \Z/2$ to be the number mod 2 of isolated points in

2396: $\FF_J^*(x_-,x_+)/\R$. The map

2397: \[

2398: d_J\gen{x_+} = \textstyle \sum_{x_-} n_J(x_-,x_+) \gen{x_-}

2399: \]

2400: has square zero, making $CF(L_0,L_1)$ into a differential vector space.

2401: $HF(L_0,L_1)$ is defined to be its cohomology $\ker\, d_J/\im\, d_J$. Unlike

2402: $d_J$ itself, Floer cohomology can be shown to be independent of $J$ up to

2403: canonical isomorphism. It is also invariant under deformations of $L_0$ or

2404: $L_1$ as exact Lagrangian submanifolds.

2405: 

2406: We now start constructing the relative invariant associated to $(E,\pi)$ and

2407: $Q$. Choose a point $x_e \in L_{e,0} \cap L_{e,1}$ for each end $e$, and write

2408: $\BB(\{x_e\})$ for the space of smooth sections $u: S \rightarrow E$ such that

2409: $u(\partial S) \subset Q$, and which over the ends have the form $\Gamma_e^{-1}

2410: \circ u \circ \gamma_e(s,t) = (s,t,\sigma_e(s,t))$ with maps $\sigma_e:

2411: \R^{\pm} \times [0;1] \rightarrow M$ satisfying $\lim_{s \rightarrow \pm

2412: \infty} \sigma_e(s,\cdot) = c_{x_e}$, in the same sense as in \eqref{eq:floer}.

2413: The action integral $A(u) = \int_S u^*\o$ is convergent for all $u \in

2414: \BB(\{x_e\})$, and one gets a formula analogous to \eqref{eq:action}:

2415: \begin{equation} \label{eq:relative-action}

2416:  A(u) = \sum_{e \in I^-} a_{L_{e,0},L_{e,1}}(x_e)

2417:  - \sum_{e \in I^+} a_{L_{e,0},L_{e,1}}(x_e)

2418:  + \int_{\partial S} \kappa_Q.

2419: \end{equation}

2420: Take a complex structure $j$ on $S$ (equal to $j_0$ near $S^\crit$ as always)

2421: which, on the ends, is induced from the standard complex structure $j_{\pm}$ on

2422: $\R^{\pm} \times [0;1]$. Choose also a $J_e = (J_{e,t})_{0 \leq t \leq 1} \in

2423: \JJ(M)$ for each $e$. Then $\JJ(E,\pi,j,\{J_e\})$ denotes the contractible

2424: space of almost complex structures $J$ on $E$ which are compatible relative to

2425: $j$, and such that $\Gamma_e^*J$, for each $e$, is the almost complex structure

2426: on $\R^{\pm} \times [0;1] \times M$ given by $j_{\pm} \times J_{e,t}$ at a

2427: point $(s,t,y)$. For any such $J$, we denote by $\MM_J(\{x_e\}) \subset

2428: \BB(\{x_e\})$ the subspace of sections $u$ which are $(j,J)$-holomorphic. The

2429: formal picture from Section \ref{sec:simple-invariant} still applies: one has

2430: the vector bundle $\EE_J \rightarrow \BB(\{x_e\})$, its canonical section

2431: $\bar\partial_J$ whose zero-set is $\MM_J(\{x_e\})$, as well as their

2432: ``universal'' versions $\EE^{univ}$, $\bar\partial^{univ}$. Turning this

2433: picture into an analytically realistic one is more complicated than in the

2434: compact case. Still, the procedure is by now standard, and so we will not say

2435: more about it, except to mention that the derivative $D_{u,J}$ of

2436: $\bar\partial_J$ again extends to a Fredholm operator from a suitably defined

2437: $W^{1,p}$-version $\WW^1_u$ of $T\BB(\{x_e\})_u$ to an $L^p$-version

2438: $\WW^0_{u,J}$ of $\EE_{u,J}$. This allows one to define the notion of

2439: regularity of $u \in \MM_J(\{x_e\})$, and of $J \in \JJ(E,\pi,j,\{J_e\})$, in

2440: the same way as before. The subspace of regular $J$ is denoted by

2441: $\JJreg(E,\pi,Q,j,\{J_e\})$.

2442: 

2443: \begin{remark}

2444: It is maybe helpful to mention that Floer's equation \eqref{eq:floer} itself

2445: can be made to fit into this framework. Namely, take the surface $S = \R \times

2446: [0;1]$ and the trivial exact symplectic fibration $\pi: E = S \times M

2447: \rightarrow S$, with the Lagrangian boundary condition $Q = (\R \times \{0\}

2448: \times L_0) \cup (\R \times \{1\} \times L_1)$; it is a tautology that this is

2449: modelled over the two ends of $S$ on $(L_0,L_1)$. Take the standard complex

2450: structure $j$ on $S$, the same $J_e \in \JJ(M)$ for both ends $e$, and define

2451: $J \in \JJ(E,\pi,j,\{J_e\})$ to be the product $j \times J_{e,t}$ at a point

2452: $(s,t,y)$. Sections $u: S \rightarrow E$ with $u(\partial S) \subset Q$ are of

2453: the form $u(s,t) = (s,t,\sigma(s,t))$, where $\sigma: \R \times [0;1]

2454: \rightarrow M$ is a map satisfying the boundary condition in \eqref{eq:floer}.

2455: Moreover, $u$ is $(j,J)$-holomorphic iff $\partial\sigma/\partial t = J_{e,t}

2456: \partial\sigma/\partial s$, so that one can identify $\MM_J(x_-,x_+) =

2457: \FF_{J_e}(x_-,x_+)$ for all $x_{\pm}$.

2458: \end{remark}

2459: 

2460: In parallel with the exposition in Section \ref{sec:simple-invariant}, we will

2461: now discuss the basic properties of the spaces $\MM_J(\{x_e\})$. The next two

2462: results are analogues of Lemmas \ref{th:convexity} and \ref{th:transversality},

2463: and their proofs are the same.

2464: 

2465: \begin{lemma} \label{th:relative-convexity}

2466: For every $J \in \JJ(E,\pi,j,\{J_e\})$ there is a closed subset $K \subset E

2467: \setminus \partial_hE$ such that $\pi|K : K \rightarrow S$ is proper, and which

2468: over the ends has the form $\Gamma_e^{-1}(K) = \R^{\pm} \times [0;1] \times

2469: K_e$ for some compact $K_e \subset M \setminus \partial M$, such that $u(S)

2470: \subset K$ for all $u \in \MM_J(\{x_e\})$ and all points $\{x_e\}$. \qed

2471: \end{lemma}

2472: 

2473: \begin{lemma} \label{th:relative-transversality}

2474: $\JJreg(E,\pi,Q,j,\{J_e\})$ is $\smooth$-dense in $\JJ(E,\pi,j,\{J_e\})$. More

2475: precisely, given some nonempty open subset $U \subset S$ which is disjoint from

2476: the ends, and a $J \in \JJ(E,\pi,j,\{J_e\})$, there are $J' \in

2477: \JJreg(E,\pi,Q,j,\{J_e\})$ arbitrarily close to $J$, such that $J = J'$ outside

2478: $\pi^{-1}(U)$. \qed

2479: \end{lemma}

2480: 

2481: The remaining issue is compactness. After taking a symplectic form $\Omega +

2482: \pi^*\beta$ on $E$ as in Lemma \ref{th:tame}, and the associated metric, one

2483: finds that $\half \int_W || Du ||^2 \leq A(u) + \int_W \beta$ for any compact

2484: subset $W \subset S$ and $u \in \MM_J(\{x_e\})$. Repeating the argument in

2485: Lemma \ref{th:compactness}, one derives from this that any sequence $(u_i)$ in

2486: $\MM_J(\{x_e\})$ has a subsequence which is $C^r$-convergent on compact

2487: subsets. It is necessary to go beyond this somewhat coarse result, and to study

2488: sequences $(u_i)$ in the more appropriate {\em Gromov-Floer topology}. A

2489: compactification $\overline{\MM}_J(\{x_e\})$ of $\MM_J(\{x_e\})$ in this

2490: topology can be constructed by adding ``broken sections''. Since a very similar

2491: notion is part of the standard analytical package underlying Floer theory, we

2492: will not define the topology, and only describe the compactification as a set.

2493: Each point of it consists of

2494: \begin{condensedlist}

2495: \item

2496: the ``principal component'' $u \in \MM_J(\{\hat{x}_e\})$ for some

2497: $\{\hat{x}_e\}$;

2498: \item

2499: for each end $e \in I^{\pm}$, a finite sequence of points $\hat{x}_{e,0},

2500: \dots, \hat{x}_{e,l_e} \in L_{e,0} \cap L_{e,1}$, $l_e \geq 0$. If the end is

2501: negative (positive), this should satisfy $\hat{x}_{e,0} = x_e$ and

2502: $\hat{x}_{e,l_e} = \hat{x}_e$ ($\hat{x}_{e,0} = \hat{x}_e$ and $\hat{x}_{e,l_e}

2503: = x_e$);

2504: \item

2505: Floer flow lines $\sigma_{e,m} \in

2506: \FF^*_{J_e}(\hat{x}_{e,m-1},\hat{x}_{e,m})/\R$, for each $e$ and $1 \leq m \leq

2507: l_e$.

2508: \end{condensedlist}

2509: Suppose that a sequence $(u_i)$ converges to such a limit. One sees from

2510: \eqref{eq:relative-action} that for all $i$,

2511: \begin{equation} \label{eq:energy-additivity}

2512:  A(u_i) = A(u) + \sum_e \sum_{1 \leq m \leq l_e}

2513:  (a_{L_{e,0},L_{e,1}}(\hat{x}_{e,m-1}) - a_{L_{e,0},L_{e,1}}(\hat{x}_{e,m}))

2514: \end{equation}

2515: Note that all terms of the $\sum$ are $>0$; therefore $A(u) \leq A(u_i)$, with

2516: equality iff $l_e = 0$ for all ends $e$. We will need another piece of

2517: information about the limit, which can be derived from the definition of the

2518: Gromov-Floer topology and a ``gluing theorem'' for linear elliptic operators.

2519: Namely, for $i \gg 0$,

2520: \begin{equation} \label{eq:index-additivity}

2521:  \ind\,D_{u_i,J} = \ind \, D_{u,J} + \sum_e \sum_{1 \leq m \leq l_e}

2522:  \ind \, D_{\sigma_{e,m},J_e}.

2523: \end{equation}

2524: 

2525: From now on assume that $J_e \in \JJreg(M,L_{e,0},L_{e,1})$ for all $e$, and

2526: that $J \in \JJreg(E,\pi,Q,j,\{J_e\})$. Then $\ind\,D_{u,J} \geq 0$ and

2527: $\ind\,D_{\sigma_{e,m},J_e} > 0$ on the right hand side of

2528: \eqref{eq:index-additivity}, so that the left hand side can be zero only if

2529: $\ind\,D_{u,J} = 0$ and $l_e = 0$ for all $e$. It follows that the

2530: zero-dimensional part of $\MM_J(\{x_e\})$, for any $\{x_e\}$, is compact, hence

2531: a finite set. Write $\nu_J(\{x_e\}) \in \Z/2$ for the number of points modulo

2532: two in this set, and consider the map

2533: \begin{align} \label{eq:relative-chain-map}

2534:  & C\Phi_0^{rel}(E,\pi,Q,J) : \bigotimes_{e \in I^+} CF(L_{e,0},L_{e,1})

2535:  \longrightarrow \bigotimes_{e \in I^-} CF(L_{e,0},L_{e,1}) \\

2536:  \intertext{given by the matrix with entries $\nu_J(\{x_e\})$, that is to

2537:  say}

2538: \notag

2539:  & C\Phi_0^{rel}(E,\pi,Q,J)(\underset{e \in I^+\!\!}{\otimes} \gen{x_e}) =

2540:  \sum_{\;\;\;\;\{x_e\}_{e \in I^-}\!\!\!\!} \nu_J(\{x_e\}_{e \in I^- \cup I^+})

2541:  (\underset{e \in I^-\!\!}{\otimes} \gen{x_e}).

2542: \end{align}

2543: A standard argument, involving the structure at infinity of the one-dimensional

2544: part of $\MM_J(\{x_e\})$, shows is that there is an even number of points in

2545: $\overline{\MM}_J(\{x_e\})$ of the following form: $l_e$ is $1$ for a single

2546: end $e=f$, and zero for all other $e$, so that the point is a pair

2547: $(u,\sigma_{f,1})$; and moreover $\ind\,D_{u,J} = 0$, $\ind \,

2548: D_{\sigma_{f,1},J_f} = 1$. Algebraically, what this says is that

2549: $C\Phi_0^{rel}(E,\pi,Q,J)$ is a chain map. The relative invariant

2550: $\Phi_0^{rel}(E,\pi,Q)$ is defined to be the induced map on cohomology.

2551: 

2552: The next step is to show that this is independent of the choice of $j$ and $J$,

2553: keeping the $J_e$ fixed for the moment. Any two $j^0,j^1$ can be connected by a

2554: family $j^\mu$, $0 \leq \mu \leq 1$, which remains constant on the ends;

2555: correspondingly, for $J^0 \in \JJreg(E,\pi,Q,j^0,\{J^e\})$, $J^1 \in

2556: \JJreg(E,\pi,Q,j^1,\{J^e\})$ one can find a family $J^\mu$ joining them, which

2557: is regular in the parametrized sense (this is not the same as saying that each

2558: $J^\mu$ should itself be regular, which would be impossible to achieve in

2559: general). The parametrized moduli spaces

2560: \[

2561: \MM^{para}_{(J^\mu)}(\{x_e\}) = \bigcup_{\!\!\! 0 \leq \mu \leq 1 \!\!\!} \,

2562: \{\mu\} \times \MM_{J^\mu}(\{x_e\})

2563: \]

2564: are then smooth manifolds, and they have a parametrized version of the

2565: Gromov-Floer compactification. In particular, the zero-dimensional parts are

2566: again finite sets, so that one can use the number of points $\lambda(\{x_e\})

2567: \in \Z/2$ in them to define a map $h(E,\pi,Q,(J^\mu))$ between the same groups

2568: as in \eqref{eq:relative-chain-map}. Arguing along the same lines as when

2569: proving that $C\Phi_0^{rel}$ is a chain map, one can show that $h$ is a

2570: homotopy between $C\Phi_0^{rel}(E,\pi,Q,J^0)$ and $C\Phi_0^{rel}(E,\pi,Q,J^1)$.

2571: The same argument can be used to show that $\Phi_0^{rel}(E,\pi,Q)$ remains

2572: invariant under deformations of the geometric data, that is to say of $Q$ or

2573: $(E,\pi)$ itself, as long as the structure of the ends remains unchanged.

2574: Finally, we should prove that the relative invariant is independent of the

2575: $J_e$, which more accurately means that it commutes with the canonical

2576: isomorphisms between Floer cohomology groups for different $J_e$. We omit this

2577: entirely, both because it is not important for our purpose, and because it

2578: would require a digression concerning ``continuation maps''.

2579: 

2580: Two ways of gluing together surfaces with strip-like ends will play a role

2581: later on. One of them is a close cousin of that considered as in Section

2582: \ref{sec:simple-invariant}. It can be formulated in various degrees of

2583: generality, but we will need only one special case. Suppose then that $S^1$ is

2584: a surface with strip-like ends, and $S^2$ a compact surface, together with

2585: points $\zeta^k \in \partial S^k$ ($\zeta^1$ should not lie on any end; that

2586: can of course always be achieved by making the ends smaller). Let $(E^k,\pi^k)$

2587: be exact Lefschetz fibrations over $S^k$ with Lagrangian boundary conditions

2588: $Q^k$; $(E^1,\pi^1)$ should be trivial over the ends. We also want to have $M$,

2589: $L$, maps $\phi^k: M \rightarrow (E^k)_{\zeta^k}$, and trivializations

2590: $(\psi^k,\Psi^k)$, with the same properties as in Section

2591: \ref{sec:simple-invariant}. The boundary connected sum $S = S^1 \#_{\zeta^1

2592: \sim \zeta^2} S^2$ is a surface with the same kind of strip-like ends as $S^1$.

2593: As before one constructs an exact Lefschetz fibration $(E,\pi)$ on it with a

2594: Lagrangian boundary condition $Q$, modelled over the ends on the same

2595: Lagrangian submanifolds as $Q^1$.

2596: 

2597: Choose complex structures $j^k$ on $S^k$ such that $(\psi^k)^*j^k$ is standard;

2598: $j^1$ should also be standard on the ends. Take $\{J_e\}$ and $J^1 \in

2599: \JJreg(E^1,\pi^1,Q^1,j^1,\{J_e\})$ as when defining the relative invariant of

2600: $(E^1,\pi^1,Q^1)$; as an additional condition, we want $(\Psi^1)^*J^1$ to be

2601: the product of the standard complex structure on $\bar{D}^+(1)$ and some fixed

2602: almost complex structure on $M$. By Lemma \ref{th:relative-transversality} this

2603: can be done while still achieving regularity. Using Lemmas

2604: \ref{th:transversality} and \ref{th:ev-transversality} one finds a $J^2 \in

2605: \JJreg(E^2,\pi^2,Q^2,j^2)$ with the same restriction on $(\Psi^2)^*J^2$, and

2606: such that the evaluations

2607: \begin{equation} \label{eq:two-evaluations}

2608: \begin{split}

2609:  & ev_{\zeta^1}\,|\,\MM_{J^1}(\{x_e\}) :

2610:  \MM_{J^1}(\{x_e\}) \longrightarrow Q^1_{\zeta^1} \iso L, \\

2611:  & ev_{\zeta^2}\,|\,\MM_{J^2}: \MM_{J^2} \longrightarrow Q^2_{\zeta^2} \iso L

2612: \end{split}

2613: \end{equation}

2614: are transverse to each other for every $\{x_e\}$. Let $J \in

2615: \JJ(E,\pi,j,\{J_e\})$ be the almost complex structure constructed from $J^1,

2616: J^2$. As in Proposition \ref{th:gluing}, this will be regular for small values

2617: of the parameter $\rho$, and

2618: \begin{equation} \label{eq:zero-gluing}

2619: \MM_J(\{x_e\})_{[0]} \iso (\MM_{J^1}(\{x_e\}) \times_L \MM_{J^2})_{[0]},

2620: \end{equation}

2621: where the $[0]$ denotes on both sides the zero-dimensional component of these

2622: manifolds. We will not need the full ``gluing formula'' which one can obtain

2623: from this, but only a special case:

2624: 

2625: \begin{lemma} \label{th:gluing-vanishing}

2626: If $\Phi_1(E^2,\pi^2,Q^2) = 0$ then $\Phi_0^{rel}(E,\pi,Q) = 0$.

2627: \end{lemma}

2628: 

2629: \proof For simplicity, suppose that $\Phi_1(E^2,\pi^2,Q^2)$ vanishes even when

2630: taken in the cobordism ring $MO_*(L)$. This means that there is a compact

2631: manifold with boundary $G$ and a smooth map $g: G \rightarrow L$, such that

2632: $\partial G = \MM_{J^2}$ and $g|\partial G = ev_{\zeta^2}$. After perturbing it

2633: slightly, one can assume that $g$ is transverse to all maps $ev_{\zeta^1}$ in

2634: \eqref{eq:two-evaluations}. Then the fibre products

2635: \begin{equation} \label{eq:fibre-product}

2636: \GG(\{x_e\}) = \MM_{J^1}(\{x_e\}) \times_L G

2637: \end{equation}

2638: are smooth manifolds. The evaluation map extends continuously to the

2639: Gromov-Floer compactification, so that one can define compactifications

2640: $\overline{\GG}(\{x_e\})$ in the obvious way. It is not difficult to see, using

2641: \eqref{eq:index-additivity}, that the zero-dimensional part

2642: $\GG(\{x_e\})_{[0]}$ is a finite set; one counts the points

2643: $\xi_{J,G,g}(\{x_e\}) \in \Z/2$ in it, and uses that to define a map $k$

2644: between the Floer cochain groups associated to $(E^1,\pi^1,Q^1)$, as in

2645: \eqref{eq:relative-chain-map}. We claim that this is a homotopy between

2646: $C\Phi^0_{rel}(E,\pi,Q,J)$ and the zero map. As usual, the proof is based on

2647: analyzing the ends of the one-dimensional moduli spaces. The closure in

2648: $\overline{\GG}(\{x_e\})$ of the one-dimensional part $\GG(\{x_e\})_{[1]}$ is a

2649: compact one-manifold with boundary, and its boundary points are of two kinds:

2650: first, boundary points of $\GG(\{x_e\})_{[1]}$ itself,

2651: $\partial\GG(\{x_e\})_{[1]} = (\MM_{J^1}(\{x_e\}) \times_L \partial G)_{[0]}

2652: \iso \MM_J(\{x_e\})_{[0]}$; their number modulo two is $\nu_J(\{x_e\})$ by

2653: definition. The second kind of boundary points are of the form

2654: $(u,\sigma_{f,1}) \times q \in \overline{\MM}_J(\{x_e\}) \times_L G$. This

2655: means that for some end $f$, and points $\{\hat{x}_e\}$ with $\hat{x}_e = x_e$

2656: for all $e \neq f$, one has

2657: \[

2658:  u \times q \in \GG(\{\hat{x}_e\})_{[0]}, \quad

2659:  \sigma_{f,1} \in \begin{cases}

2660:  (\FF^*(x_f,\hat{x}_f)/\R)_{[0]} & \text{if $e$ is a negative end,} \\

2661:  (\FF^*(\hat{x}_f,x_f)/\R)_{[0]} & \text{if $e$ is a positive end.}

2662:  \end{cases}

2663: \]

2664: The number of such boundary points is

2665: \[

2666: \sum_{f \in I^-} \sum_{\hat{x}_f}

2667:   n_{J_f}(x_f,\hat{x}_f) \,

2668:  \xi_{J,G,g}(\{\hat{x}_e\})

2669:  +

2670: \sum_{f \in I^+} \sum_{\hat{x}_f}

2671:  \xi_{J,G,g}(\{\hat{x}_e\}) \,

2672:  n_{J_f}(\hat{x}_f,x_f)

2673: \in \Z/2.

2674: \]

2675: The fact that this is equal to $\nu_J(\{x_e\})$ gives precisely the desired

2676: equality $dk + kd = C\Phi_0^{rel}(E,\pi,Q,J)$. \qed

2677: 

2678: The other and maybe more obvious gluing process is to join together two ends.

2679: Assume that $S^1,S^2$ are surfaces with strip-like ends, such that $S^1$ has a

2680: single positive end $e^1$, and $S^2$ a single negative end $e^2$ (this is not

2681: really a restriction since, as has been mentioned before, positive ends can be

2682: turned into negative ones and vice versa). Choose some $\sigma>0$, and define

2683: $S = S^1 \#_{e^1 \sim e^2} S^2$ by taking $S^1 \setminus

2684: \gamma_{e^1}((\sigma;\infty) \times [0;1])$ and $S^2 \setminus

2685: \gamma_{e^2}((-\infty;-\sigma) \times [0;1])$, and identifying

2686: $\gamma_{e^1}(s,t)$ with $\gamma_{e^2}(s-\sigma,t)$ for $(s,t) \in [0;\sigma]

2687: \times [0;1]$. The ends of $S$ are the negative ends of $S^1$ together with the

2688: positive ends of $S^2$. After choosing complex structures $j^k$ on $S^k$ which

2689: are standard over the ends, there is an obvious induced complex structure $j$

2690: on $S$.

2691: 

2692: Let $(L_0,L_1)$ be a pair of exact Lagrangian submanifolds in $M$. Suppose that

2693: we have exact Lefschetz fibrations $(E^k,\pi^k)$ over $S^k$, trivial over the

2694: ends, and Lagrangian boundary conditions $Q^1,Q^2$ for them modelled on

2695: $(L_0,L_1)$ over $e^1$ and over $e^2$, respectively. Then one can form an exact

2696: Lefschetz fibration $(E,\pi)$ over $S$ by identifying $\Gamma_{e^1}(s,t,y) \in

2697: E^1$ with $\Gamma_{e^2}(s-\sigma,t,y) \in E^2$, in parallel with the

2698: construction on the base. This comes with an obvious Lagrangian boundary

2699: condition $Q$. Choose almost complex structures $J^k$ on $E^k$ so as to define

2700: relative invariants $C\Phi_0^{rel}(E^k,\pi^k,Q^k,J^k)$. We require that the

2701: $J_{e^k} \in \JJreg(M,L_0,L_1)$ on which $J^k$ is modelled over the end $e^k$

2702: should be the same for $k = 1,2$. Then $J^1$ and $J^2$ match up to an almost

2703: complex structure $J$ on $E$, which is compatible relative to $j$.

2704: 

2705: \begin{prop} \label{th:gluing-ends}

2706: For fixed $J^1,J^2$, if one chooses $\sigma$ to be sufficiently large, then $J$

2707: is regular. Moreover, denoting again by $[0]$ the zero-dimensional components,

2708: and by $I^-,I^+$ the negative and positive ends of $S$, one has

2709: \[

2710: \MM_J(\{x_e\})_{[0]} \iso \bigcup_{\!\!\! x \in L_0 \cap L_1 \!\!\!}

2711: \MM_{J^1}(\{x_e\}_{e \in I^-},x)_{[0]} \times \MM_{J^2}(x,\{x_e\}_{e \in

2712: I^+})_{[0]}

2713: \]

2714: for any $\{x_e\}_{e \in I^- \cup I^+}$. \qed

2715: \end{prop}

2716: 

2717: The method used in the proof of this is the same as when setting up Floer

2718: cohomology. A thorough exposition of a closely related result can be found in

2719: \cite[Section 4.4]{schwarz95}. The implication for the relative invariants is

2720: clear:

2721: \begin{equation} \label{eq:chain-composition}

2722:  C\Phi_0^{rel}(E,\pi,Q,J) = C\Phi_0^{rel}(E^1,\pi^1,Q^1,J^1) \circ

2723:  C\Phi_0^{rel}(E^2,\pi^2,Q^2,J^2),

2724: \end{equation}

2725: which proves the main TQFT-style property of relative invariants, namely, that

2726: they are functorial if one regards the gluing $S = S^1 \#_{e^1 \sim e^2} S^2$

2727: as composition of the surfaces $S^1,S^2$.

2728: 

2729: %\input{2e}

2730: 

2731: \subsection{Horizontality and relative invariants\label{sec:horizontal-two}}

2732: 

2733: The aim of this section is to extend the methods of Section

2734: \ref{sec:horizontal} to surfaces with strip-like ends. Throughout, $(E,\pi)$

2735: will be an exact Lefschetz fibration over such a surface $S$, trivial on the

2736: ends, together with a Lagrangian boundary condition $Q$ modelled over the ends

2737: on pairs $L_{0,e},L_{1,e} \subset M$ of transversally intersecting exact

2738: Lagrangian submanifolds. For $\{x_e\}_{e \in I^- \cup I^+}$ with $x_e \in

2739: L_{0,e} \cap L_{1,e}$ write

2740: \[

2741: \chi(\{x_e\}) = \sum_{e \in I^-} a_{L_{0,e},L_{1,e}}(x_e) - \sum_{e \in I^+}

2742: a_{L_{0,e},L_{1,e}}(x_e).

2743: \]

2744: By \eqref{eq:relative-action} $A(u) = \chi(\{x_e\}) + \int_{\partial S}

2745: \kappa_Q$ for all $u \in \BB(\{x_e\})$. After fixing a complex structure $j$ on

2746: $S$ which is standard on the ends, and a $J_e \in \JJ(M)$ for each $e$, we

2747: consider the space

2748: \[

2749: \JJ^h(E,\pi,j,\{J_e\}) = \JJ^h(E,\pi,j) \cap \JJ(E,\pi,j,\{J_e\})

2750: \]

2751: of almost complex structures $J$, compatible relative to $j$, which are

2752: horizontal and have prescribed behaviour on the ends. The two conditions do not

2753: contradict each other, since the model on each end, the almost complex

2754: structure $j_{\pm} \times J_e$ on the trivial fibration $\R^{\pm} \times [0;1]

2755: \times M \rightarrow \R^{\pm} \times [0;1]$, is obviously horizontal. A little

2756: more thought shows that $\JJ^h(E,\pi,j,\{J_e\})$ is contractible. Choosing $J$

2757: in that space, one finds that for all $u \in \BB(\{x_e\})$,

2758: \begin{equation} \label{eq:rel-deficiency}

2759: {\textstyle \half} \int_S ||(Du)^v||^2 + \int_S f(u)\beta = A(u) + \int_S

2760: ||\bar\partial_J u||^2.

2761: \end{equation}

2762: Here $||\cdot||$ is the metric on $TE^v$ associated to $\Omega$ and $J$; $\beta

2763: \in \Omega^2(S)$ is a positive two-form; and $f$ is the function defined by

2764: $\Omega|TE^h = f(\pi^*\beta|TE^h)$. Because the symplectic connection is

2765: trivial on the ends, $f(u)$ is compactly supported. \eqref{eq:rel-deficiency}

2766: is obviously the analogue of \eqref{eq:deficiency}, and can be proved in the

2767: same way. Recall that $(E,\pi)$ has nonnegative curvature iff $f \geq 0$. From

2768: this we draw two conclusions:

2769: 

2770: \begin{lemma} \label{th:empty}

2771: Assume that $(E,\pi)$ has nonnegative curvature, and choose $J \in

2772: \JJ^h(E,\pi,j,\{J_e\})$. Then $\MM_J(\{x_e\}) = \emptyset$ for all $\{x_e\}$

2773: with $\chi(\{x_e\}) < -\int_{\partial S} \kappa_Q$. \qed

2774: \end{lemma} \vspace{\parskip}

2775: 

2776: \begin{lemma} \label{th:energy-compactness}

2777: In the same situation, let $\alpha$ be the minimum of $a_{L_{0,e},L_{1,e}}(x_-)

2778: - a_{L_{0,e},L_{1,e}}(x_+)$, taken over all $e$ and all $x_{\pm} \in L_{0,e}

2779: \cap L_{1,e}$ for which this is $>0$. Then the spaces $\MM_J(\{x_e\})$ is

2780: compact, in the Gromov-Floer topology, for all $\{x_e\}$ such that

2781: $\chi(\{x_e\}) < -\int_{\partial S} \kappa_Q + \alpha$.

2782: \end{lemma}

2783: 

2784: \proof Consider a sequence $(u_i)$ in $\MM_J(\{x_e\})$ which converges to

2785: $(u,\{\sigma_{e,m}\}) \in \overline{\MM}_J(\{x_e\})$. The definition of

2786: $\gamma$ implies that in \eqref{eq:energy-additivity}, each

2787: $a_{L_{e,0},L_{e,1}}(\hat{x}_{e,m-1}) - a_{L_{e,0},L_{e,1}}(\hat{x}_{e,m}) \geq

2788: \alpha$; therefore

2789: \[

2790: A(u) \leq A(u_i) - \big(\textstyle\sum_e l_e\big) \alpha.

2791: \]

2792: By assumption $A(u_i) = \chi(\{x_e\}) + \int_{\partial S} \kappa_Q < \alpha$,

2793: while on the other hand, $A(u) \geq 0$ by nonnegative curvature and

2794: \eqref{eq:rel-deficiency}. This is only possible if $l_e = 0$ for all $e$, so

2795: that the limit actually lies in $\MM_J(\{x_e\})$. \qed

2796: 

2797: As before, denote by $\MM^h$ the space of horizontal sections with boundary

2798: values in $Q$. The presence of strip-like ends makes this space somewhat

2799: simpler than in the case which we have encountered before. Take an arbitrary $u

2800: \in \BB$ and consider $u_e(s,t) = \Gamma_e^{-1} \circ u \circ \gamma_e(s,t) =

2801: (s,t,\sigma_e(s,t))$; $u$ is horizontal over the end $e$ iff $\sigma_e \equiv x

2802: \in M$ is constant, in which case the boundary conditions $\sigma_e(s,0) \in

2803: L_{e,0}$, $\sigma_e(s,1) \in L_{e,1}$ imply that $x \in L_{e,0} \cap L_{e,1}$.

2804: Since a horizontal section is determined by its value at any one point, for any

2805: $e$ and any $x \in L_{e,0} \cap L_{e,1}$ there can be at most one $u \in \MM^h$

2806: such that $\sigma_e \equiv x$. Thus $\MM^h$ is a finite disjoint union of the

2807: subsets $\MM^h(\{x_e\}) = \MM^h \cap \BB(\{x_e\})$, each of which consists of

2808: at most one element. One easily proves the following limiting case of Lemma

2809: \ref{th:empty}:

2810: 

2811: \begin{lemma} \label{th:almost-empty}

2812: For $(E,\pi)$ and $J$ as in Lemma \ref{th:empty}, suppose that $\{x_e\}$

2813: satisfy $\chi(\{x_e\}) = -\int_{\partial S} \kappa_Q$; then $\MM_J(\{x_e\}) =

2814: \MM^h(\{x_e\})$. \qed

2815: \end{lemma}

2816: 

2817: To translate these elementary observations into results about relative

2818: invariants, one needs to address again the question of regularity of horizontal

2819: almost complex structures; in other words, what is required are analogues of

2820: Lemmas \ref{th:horizontal-transversality} and \ref{th:flat-deformations}. For

2821: the first of these, both the statement and proof as essentially the same as in

2822: the original situation; the second needs to be adapted a little.

2823: 

2824: \begin{lemma} \label{th:relative-h-transversality}

2825: Let $U \subset S$ be a nonempty open subset disjoint from the ends, such that

2826: any partial section $w: U \rightarrow E|U$ which is horizontal and satisfies

2827: $w(\partial S \cap U) \subset Q$ is the restriction of a $u \in \MM^h$. Then,

2828: given some $J \in \JJ^h(E,\pi,j,\{J_e\})$, there are $J' \in

2829: \JJ^h(E,\pi,j,\{J_e\})$ arbitrarily close to it and which agree with it outside

2830: $\pi^{-1}(U)$, such that for all $\{x_e\}$, any $u \in \MM_J(\{x_e\}) \setminus

2831: \MM^h$ is regular. \qed

2832: \end{lemma} \vspace{\parskip}

2833: 

2834: \begin{lemma} \label{th:negative-index}

2835: Let $(E,\pi)$ and $J$ be as in Lemma \ref{th:empty}. If $u \in \MM^h$ is a

2836: horizontal section with $A(u) = 0$, then $ker \, D_{u,J} = 0$.

2837: \end{lemma}

2838: 

2839: \proof Formally, taking the second derivative of \eqref{eq:rel-deficiency} at

2840: $u$ yields the same formula \eqref{eq:weitzenboeck} as in the case of a compact

2841: $S$, but a little care needs to be exercised about its validity. It certainly

2842: holds for those elements of $T\BB_u = \{ X \in \smooth(u^*TE^v) \suchthat X_z

2843: \in T(Q_z) \text{ for } z \in \partial S\}$ which are compactly supported, and

2844: by continuity, for all $X$ in the $W^{1,2}$-completion; to be precise, this

2845: Sobolev space is with respect to the metric $||\cdot||$ and the connection

2846: $\nabla^u$ on $u^*TE^v$. We actually want to use the formula with $X \in

2847: \WW^1_u$ and this is a space of $W^{1,p}$-sections with $p>2$, hence not

2848: contained in $W^{1,2}$. However, if one assumes additionally that $D_{u,J}X =

2849: 0$ there is no problem, because any such $X$ is smooth and decays exponentially

2850: on the ends, as do its derivatives.

2851: 

2852: From $A(u) = 0$ and \eqref{eq:rel-deficiency} it follows that $f(u) \equiv 0$,

2853: so that the Hessians $Hess(f|E_z)_{u(z)}$ are nonnegative. One then sees from

2854: \eqref{eq:weitzenboeck} that any $X \in \WW^1_u$ with $D_{u,J}X = 0$ satisfies

2855: $\nabla^uX = 0$. Choose some end $e$ and write $x = x_e$. The trivialization

2856: $\Gamma_e$ induces a trivialization of the vector bundle $\gamma_e^*(u^*TE^v)

2857: \rightarrow \R^{\pm} \times [0;1]$, which identifies it with the trivial bundle

2858: with fibre $TM_x$. In this way, $Y = \gamma_e^*X$ becomes a map $\R^{\pm}

2859: \times [0;1] \rightarrow TM_x$. Since the connection $\nabla^u$ is compatible

2860: with the trivialization, $\nabla^uX = 0$ implies that $Y$ must be constant. On

2861: the other hand, the boundary conditions which are part of the definition of

2862: $T\BB_u$ and of $\WW^1_u$ tell us that $Y_{s,k} \in T(L_{e,k})_x$ for $k =

2863: 0,1$. Because the $L_{e,k}$ intersect transversally, it follows that $Y = 0$,

2864: hence that $X | \im(\gamma^e) = 0$. Since $X$ is covariantly constant, it must

2865: be zero everywhere. \qed

2866: 

2867: The next result summarizes what progress we have made so far, as well as the

2868: implications for the coefficients $\nu_J(\{x_e\})$ of the relative invariant.

2869: 

2870: \begin{prop} \parindent0em \label{th:nonnegative-one}

2871: Assume that $(E,\pi)$ has nonnegative curvature, and that any $u \in \MM^h$

2872: satisfies $A(u) = 0$ and $\ind\,D_{u,J} = 0$. Set $\kappa = \int_{\partial S}

2873: \kappa_Q$.

2874: 

2875: (i) Let $U \subset S$ be a nonempty open subset, disjoint from the ends, such

2876: that any partial horizontal section $w: U \rightarrow E|U$ with $w(\partial S

2877: \cap U) \subset Q$ extends to a $u \in \MM^h$. Then for any $J \in

2878: \JJ^h(E,\pi,j,\{J_e\})$ there are $J' \in \JJ^{reg,h}(E,\pi,Q,j,\{J_e\}) =

2879: \JJ^h(E,\pi,j,\{J_e\}) \cap \JJ^{reg}(E,\pi,Q,j,\{J_e\})$ arbitrarily close to

2880: $J$, and which agree with it outside $\pi^{-1}(U)$.

2881: 

2882: (ii) If $J_e \in \JJreg(M,L_{0,e},L_{1,e})$ for all $e$, and $J \in

2883: \JJ^{reg,h}(E,\pi,Q,j,\{J_e\})$, then

2884: \[

2885:  \nu_J(\{x_e\}) =

2886:  \begin{cases}

2887:  0 & \text{if $\chi(\{x_e\}) < -\kappa$,} \\

2888:  \#\MM^h(\{x_e\}) & \text{if $\chi(\{x_e\}) = -\kappa$.}

2889:  \end{cases}

2890: \]

2891: \end{prop}

2892: 

2893: \proof (i) Take a $J'$ as given by Lemma \ref{th:relative-h-transversality}.

2894: Then all $u \in \MM_{J'}(\{x_e\})$ are regular except possibly for the

2895: horizontal ones. These satisfy $A(u) = 0$, so we can apply Lemma

2896: \ref{th:negative-index} to them, showing that $\ker\,D_{u,J'} = 0$, and by

2897: assumption on the index, that $\coker\,D_{u,J'} = 0$; which means that they are

2898: regular as well. (ii) This follows immediately from Lemmas \ref{th:empty} and

2899: \ref{th:almost-empty}, in view of the definition of $\nu_J(\{x_e\})$. \qed

2900: 

2901: An algebraic language suitable for encoding results of this kind is that of

2902: $\R$-graded vector spaces, that is to say vector spaces $C$ equipped with a

2903: splitting $C = \bigoplus_{r \in \R} C_r$. All vector spaces occurring here will

2904: be over $\Z/2$ and finite-dimensional; in particular, their support $supp(C) =

2905: \{r \in \R \suchthat C_r \neq 0\}$ is always a finite set. Let $I \subset \R$

2906: be an interval. We say that $C$ has {\em gap $I$} if there are no $r,s \in

2907: supp(C)$ with $r-s \in I$. A map $f: C \rightarrow D$ between graded

2908: $\R$-vector spaces is said to be {\em of order $I$} if $f(C_r) \subset

2909: \bigoplus_{s \in r + I} D_s$ for all $r$.

2910: 

2911: Floer cochain groups are obvious examples: $C = CF(L_0,L_1)$ is canonically

2912: $\R$-graded, with $C_r$ the subspace spanned by those $\gen{x}$ with

2913: $a_{L_0,L_1}(x) = r$. Because of its gradient flow interpretation, the Floer

2914: differential $d_J$ is of order $(0;\infty)$. In a rather trivial way, this can

2915: always be strengthened slightly; namely, there is an $\alpha>0$ such that $C$

2916: has gap $(0;\alpha)$, and then $d_J$ is of order $[\alpha;\infty)$. One can

2917: reformulate Proposition \ref{th:nonnegative-one}(ii) in this language as

2918: follows:

2919: 

2920: \begin{lemma} \label{th:nonnegative-two}

2921: Take $(E,\pi)$, $Q$ and $\kappa$ as in Proposition \ref{th:nonnegative-one},

2922: with $J_e,J$ as in part (ii). Then the map $C\Phi^0_{rel}(E,\pi,Q,J)$ is of

2923: order $[-\kappa;\infty)$. For a more precise statement, take $\alpha>0$ to be

2924: the minimum of $\chi(\{x_e\})+\kappa$, ranging over all $\{x_e\}$ where this is

2925: positive. Then $C\Phi_0^{rel}(E,\pi,Q,J) = \phi + (C\Phi_0^{rel}(E,\pi,Q,J) -

2926: \phi)$, where the first summand is of order $\{-\kappa\}$ and the second of

2927: order $[-\kappa+\alpha;\infty)$. Moreover, $\phi$ is determined by the

2928: horizontal sections:

2929: \[

2930: \phi(\underset{e \in I^+\!\!}{\otimes} \gen{x_e}) =

2931:  \sum_{\;\;\;\;\{x_e\}_{e \in I^-}\!\!\!\!} \#\MM^h(\{x_e\}_{e \in I^- \cup I^+})

2932:  (\underset{e \in I^-\!\!}{\otimes} \gen{x_e}). \qed

2933: \]

2934: \end{lemma}

2935: 

2936: There is a version of this for the homotopies between the $C\Phi^0_{rel}$ for

2937: different choices of $(j,J)$. The proof applies the same ideas as before to

2938: parametrized moduli spaces, and is left to the reader.

2939: 

2940: \begin{lemma} \label{th:nonnegative-homotopy}

2941: Let $(E,\pi)$, $Q$ and $\kappa$ be as in Proposition \ref{th:nonnegative-one}.

2942: Consider two complex structures $j^k$ on $S$, $k = 0,1$, and correspondingly

2943: two almost complex structures $J^k \in \JJ^{reg,h}(E,\pi,Q,j,\{J_e\})$; note

2944: that the $J_e$ are supposed to be the same for both $k$. Then the maps

2945: $C\Phi^0_{rel}(E,\pi,Q,J^k)$ are homotopic by a chain homotopy which is of

2946: order $(-\kappa;\infty)$. \qed

2947: \end{lemma}

2948: 

2949: It is a familiar idea that when dealing with maps between $\R$-graded vector

2950: spaces, the ``lowest order term'' is usually the most important, and knowing it

2951: is often sufficient to resolve a question. A particular instance of this is

2952: relevant for our purpose.

2953: 

2954: \begin{lemma} \label{th:spectral-sequence}

2955: Let $D$ be an $\R$-graded vector space with a differential $d_D$ of order

2956: $[0;\infty)$. Suppose that $D$ has gap $[\epsilon;2\epsilon)$ for some

2957: $\epsilon>0$. One can then write $d_D = \delta + (d_D-\delta)$ with $\delta$ of

2958: order $[0;\epsilon)$, satisfying $\delta^2 = 0$, and $(d_D-\delta)$ of order

2959: $[2\epsilon;\infty)$. Suppose that in addition, $H(D,\delta) = 0$; then

2960: $H(D,d_D) = 0$.

2961: \end{lemma}

2962: 

2963: \proof Thanks to the gap assumption, $supp(D)$ can be decomposed into disjoint

2964: subsets $R_1,\dots,R_m$ such that for $r \in R_i$, $s \in R_j$,

2965: \[

2966: r-s \; \begin{cases}

2967:  \leq -2\epsilon & i < j, \\

2968:  \in (-\epsilon;\epsilon) & i = j, \\

2969:  \geq 2\epsilon & i > j.

2970: \end{cases}

2971: \]

2972: Define a descending filtration of $(D,d_D)$,

2973: \[

2974: F^k = \textstyle\bigoplus_{r \in R_k \cup R_{k+1} \cup \dots \cup R_m} D_r.

2975: \]

2976: There is a ``spectral sequence'' which takes the form of a sequence

2977: $(E^k,\partial^k)$ of differential vector spaces, such that $E^{k+1} =

2978: H(E^k,\partial^k)$. It starts with $E^0 = \bigoplus_i F^i/F^{i+1}$, which has a

2979: differential $\partial^0$ induced by $d_D$. In our case this can be identified

2980: with $(D,\delta)$, so the assumption says that $E^1 = 0$. On the other hand

2981: $E^k \iso H(D,d_D)$ for $k \gg 0$. \qed

2982: 

2983: \begin{lemma} \label{th:low-energy}

2984: Take three $\R$-graded vector spaces $C',C,C''$, each of them with a

2985: differential of order $(0;\infty)$. Suppose that we have differential maps $b:

2986: C' \rightarrow C$, $c: C \rightarrow C''$ and a homotopy $h: C' \rightarrow

2987: C''$ between $c \circ b$ and the zero map, such that that the following

2988: conditions are satisfied for some $\epsilon>0$:

2989: \begin{condensedlist}

2990: \item \label{item:gap-one}

2991: $C',C''$ have gap $(0;3\epsilon)$, and $C$ has gap $(0;2\epsilon)$.

2992: \item \label{item:gap-two}

2993: For all $r \in supp(C')$ and $s \in supp(C'')$, $|r-s| \geq 4\epsilon$.

2994: \item \label{item:gap-three}

2995: One can write $b = \beta + (b - \beta)$ with $\beta$ of order $[0;\epsilon)$

2996: and $(b - \beta)$ of order $[2\epsilon;\infty)$; and $c = \gamma + (c-\gamma)$

2997: with the same properties. The low order parts (which do not need be

2998: differential maps) fit into a short exact sequence of vector spaces

2999: \[

3000: 0 \rightarrow C' \xrightarrow{\beta} C \xrightarrow{\gamma} C'' \rightarrow 0.

3001: \]

3002: \item \label{item:gap-four}

3003: $h$ is of order $[0;\infty)$.

3004: \end{condensedlist}

3005: Then the maps on cohomology induced by $b,c$ fit into a long exact sequence

3006: \begin{equation} \label{eq:induced-sequence}

3007: \xymatrix{

3008:  {H(C',d_{C'})} \ar[r]^{b_*} & {H(C,d_C)} \ar[r]^{c_*} & {H(C'',d_{C''}).}

3009:  \ar@/^1.5pc/[ll]^{ }

3010: }

3011: \end{equation}

3012: \end{lemma}

3013: 

3014: \proof Consider intervals $I_r = [r;r+\epsilon)$ for $r \in supp(C')$, and $I_r

3015: = (r-\epsilon;r]$ for $r \in supp(C'')$. By \ref{item:gap-one},

3016: \ref{item:gap-two} these are pairwise disjoint, and the distance between any

3017: two of them is $\geq 2\epsilon$. From \ref{item:gap-three} one sees that

3018: $supp(C)$ is contained in the union of these intervals, which shows that $D =

3019: C' \oplus C \oplus C''$ has gap $[\epsilon;2\epsilon)$. Consider the

3020: differential $d_D = \delta + (d_D-\delta)$,

3021: \[

3022: d_D =

3023: \begin{pmatrix}

3024: d_{C'} & 0 & 0 \\ b & d_C & 0 \\ h & c & d_{C''}

3025: \end{pmatrix}, \qquad

3026: \delta =

3027: \begin{pmatrix}

3028: 0 & 0 & 0 \\ \beta & 0 & 0 \\ 0 & \gamma & 0

3029: \end{pmatrix}.

3030: \]

3031: We know that $d_{C'}$, $d_C$, $d_{C''}$, $(b-\beta)$, $(c-\gamma)$ are of order

3032: $[2\epsilon;\infty)$. Combining \ref{item:gap-two} with \ref{item:gap-four}

3033: shows that $h$ is of order $[4\epsilon;\infty)$, so that $(d_D-\delta)$ is of

3034: order $[2\epsilon;\infty)$. On the other hand $\delta$ is of order

3035: $[0;\epsilon)$, and \ref{item:gap-three} says that $H(D,\delta) = 0$. Lemma

3036: \ref{th:spectral-sequence} shows that $H(D,d_D) = 0$, which by a general fact

3037: implies the existence of a long exact sequence \eqref{eq:induced-sequence}.

3038: \qed

3039: 

3040: \newpage

3041: %\input{3a}

3042: 

3043: \section{Wrapping it up\label{ch:three}}

3044: 

3045: Technically, the proof of the exact sequence is an application of Lemma

3046: \ref{th:low-energy}. The spaces $C'$, $C$, $C''$ which occur in the lemma will

3047: be Floer cochain spaces, and the maps $b,c$ relative invariants. While the

3048: definition of $b$ is rather straightforward, that of $c$ uses some of the

3049: geometry from Chapter \ref{ch:one}, specifically the standard fibrations

3050: constructed in Section \ref{sec:model}. In both cases, the desired properties

3051: follow from nonnegative curvature, that is to say Proposition

3052: \ref{th:nonnegative-one} and related results. The homotopy $h$ is obtained by

3053: comparing two different constructions of a particular exact Lefschetz

3054: fibration. Nonnegative curvature again plays a role in analyzing it, but the

3055: vanishing theorem of Section \ref{sec:vanishing} is also important.

3056: 

3057: \subsection{Preliminaries\label{sec:setup}}

3058: 

3059: This section sets up the framework for the whole chapter. The data are: $M$ is

3060: an exact symplectic manifold. $L_0,L_1 \subset M$ are exact Lagrangian

3061: submanifolds. $L \subset M$ is an exact Lagrangian sphere, which comes with a

3062: diffeomorphism $f: S^n \rightarrow L$ and a symplectic embedding $\iota:

3063: T(\lambda) \rightarrow M$ for some $\lambda>0$, such that $\iota|T(0) = f$. We

3064: use the Dehn twist $\tau_L$ defined using $\iota$ and a function $R$. Also

3065: given are small constants $\epsilon,\delta>0$. The following conditions are

3066: required to hold:

3067: \begin{Romanlist}

3068: \item \label{cond:transverse}

3069: $L \cap L_0$, $L \cap L_1$, $L_0 \cap L_1$ are transverse intersections, and $L

3070: \cap L_0 \cap L_1 = \emptyset$.

3071: \item \label{cond:action}

3072: The actions $a_{L_0,L_1}(x)$ of distinct points $x \in L_0 \cap L_1$ differ by

3073: at least $3\epsilon$. Secondly, as $({x}_0,x_1)$ runs over $(L_0 \cap L) \times

3074: (L \cap L_1)$, the numbers $a_{L_0,L}({x}_0) + a_{L,L_1}(x_1)$ differ pairwise

3075: by at least $3\epsilon$. Thirdly, for any $x \in L_0 \cap L_1$, $({x}_0,x_1)

3076: \in (L_0 \cap L) \times (L \cap L_1)$ one has

3077: \[

3078: |a_{L_0,L_1}(x) - a_{L_0,L}({x}_0) - a_{L,L_1}(x_1)| \geq 5 \epsilon.

3079: \]

3080: \item \label{cond:distance}

3081: For all $y_k \in f^{-1}(L \cap L_k)$, $k = 0,1$, the distance $dist({y}_0,y_1)$

3082: in the standard metric on $S^n$ is $\geq 2\pi\delta$.

3083: \item \label{cond:local}

3084: $\iota^*\theta = \theta_T|T(\lambda)$ is the standard one-form, and the

3085: function $K_L$ associated to $L$ is zero. Moreover, each $\iota^{-1}(L_k)

3086: \subset T(\lambda)$ is a union of fibres; one can write this as

3087: \begin{equation} \label{eq:union-of-fibres}

3088: \iota^{-1}(L_k) = \bigcup_{y \in \iota^{-1}(L \cap L_k)} T(\lambda)_y.

3089: \end{equation}

3090: \item \label{cond:wobbly}

3091: $R$ satisfies $0 \geq 2\pi R(0) > -\epsilon$, and is such that $\tau_L$ is

3092: $\delta$-wobbly.

3093: \end{Romanlist}

3094: 

3095: \begin{remark} \label{re:scope}

3096: Since we will establish the exact sequence under these conditions, it is

3097: necessary to convince ourselves that they do not restrict its validity in any

3098: way. Suppose then that we are given arbitrary exact Lagrangian submanifolds

3099: $L_0,L_1$ and a framed exact Lagrangian sphere $(L,[f])$ in $M$. After

3100: perturbing the submanifolds slightly, one can assume that \ref{cond:transverse}

3101: holds. Another such perturbation achieves \ref{cond:action} for some

3102: $\epsilon>0$. This is an instance of a general fact: by moving one of two

3103: transverse exact Lagrangian submanifolds slightly, the action of the

3104: intersection points can be changed independently of each other, by arbitrary

3105: sufficiently small amounts. Choose some representative $f$ of the framing.

3106: Since $L_0 \cap L_1 \cap L = \emptyset$, \ref{cond:distance} is automatically

3107: true for some $\delta>0$. Because the intersections $L \cap L_k$ are

3108: transverse, one can find a symplectic embedding $\iota: T(\lambda) \rightarrow

3109: M$, for some $\lambda>0$, which extends $f$ and such that $\iota^{-1}(L \cap

3110: L_k)$ is a union of fibres. By replacing $\theta$ with $\theta + dH$ for a

3111: suitable $H$, and making $\lambda$ smaller, one can ensure that $\iota^*\theta

3112: = \theta_T$ is satisfied. Note that when one modifies $\theta$ in this way, the

3113: functions associated to exact Lagrangian submanifolds change accordingly. One

3114: can use this and the freedom in the choice of $H$ to arrange that $K_L$ becomes

3115: equal to zero. In any case, the values of the action functional at intersection

3116: points remain the same, so that this does not interfere with \ref{cond:action}.

3117: We have now satisfied \ref{cond:local}. It is no problem to choose $R$ such

3118: that \ref{cond:wobbly} holds for the previously obtained $\epsilon, \delta$.

3119: None of the changes which we have made affects Floer cohomology. Therefore,

3120: once the exact sequence is established for the modified data, it also holds for

3121: the original ones.

3122: \end{remark}

3123: 

3124: Next, we need to draw some elementary inferences. \ref{cond:local} implies that

3125: $dK_{L_k} = \theta|L_k$ vanishes on $L_k \cap \im(\iota)$; in other words

3126: $K_{L_k} \circ \iota$ is constant on each fibre in \eqref{eq:union-of-fibres}.

3127: Let $\tau$ be the model Dehn twist from which $\tau_L$ is constructed. The

3128: function $K_\tau$ associated to it was determined in Lemma

3129: \ref{th:on-invariant}. Now $K_{\tau_L}$ vanishes outside $\im(\iota)$ and,

3130: again by \ref{cond:local}, satisfies $K_{\tau_L} \circ \iota = K_\tau$.

3131: Concretely

3132: \begin{equation} \label{eq:kk-function}

3133: \begin{aligned}

3134:  K_{\tau_L}(\iota(y))

3135:  &= 2\pi( R'(||y||)||y|| - R(||y||)) \\

3136:  &= -2 \pi R(0) + 2\pi \int_0^{||y||} (R'(||y||) - R'(t))\, dt.

3137: \end{aligned}

3138: \end{equation}

3139: In particular $K_{\tau_L}|L = -2\pi R(0)$, which by \ref{cond:wobbly} lies in

3140: $[0;\epsilon)$. The same condition says that $R'$ decreases monotonically from

3141: $R'(0) = 1/2$ until it reaches the value $\delta$, and thereafter takes values

3142: in $[0;\delta)$. By combining this with \eqref{eq:kk-function} one obtains the

3143: estimate, valid for all $y \in T(\lambda)$ with $R'(||y||) \geq \delta$,

3144: \begin{equation} \label{eq:action-estimate}

3145: -2\pi R(0) \geq K_{\tau_L}(\iota(y)) \geq -2\pi R(0) - 2\pi \int_0^{\infty}

3146: R'(t) dt = 0.

3147: \end{equation}

3148: Now consider the $\R$-graded vector spaces

3149: \begin{align*}

3150:  & C' = CF(L,L_1) \otimes CF(\tau_L(L_0),L), \\

3151:  & C'' = CF(L_0,L_1).

3152: \end{align*}

3153: The first part of \ref{cond:action} implies that $C''$ has gap $(0;3\epsilon)$.

3154: Clearly, a point $\tilde{x}_0$ lies in $\tau_L(L_0) \cap L$ iff $x_0 =

3155: \tau_L^{-1}(\tilde{x}_0)$ lies in $L_0 \cap L_1$. By definition of

3156: $K_{\tau_L(L_0)}$ and the computation above,

3157: \begin{equation} \label{eq:tau-acts}

3158:  a_{\tau_L(L_0),L}(\tilde{x}_0) = a_{L_0,L}(x_0) - K_{\tau_L}(x_0) =

3159:  a_{L_0,L}(x_0) + 2\pi R(0).

3160: \end{equation}

3161: Hence $C'$ can be identified with $CF(L,L_1) \otimes CF(L_0,L)$ up to a shift

3162: in the grading, which is by a constant of size $<\epsilon$. It therefore

3163: follows from \ref{cond:action} that $C'$ has gap $(0;3\epsilon)$, and that the

3164: distance between the supports of $C',C''$ is at least $4\epsilon$. To

3165: summarize, what we have shown is that $C',C''$ satisfy the assumptions

3166: \ref{item:gap-one}, \ref{item:gap-two} of Lemma \ref{th:low-energy}.

3167: 

3168: \begin{lemma} \label{th:new-intersection-points}

3169: $\tau_L(L_0)$, $L_1$ intersect transversally, and there are injective maps

3170: \begin{align*}

3171:  & p: (\tau_L(L_0) \cap L) \times (L \cap L_1)

3172:  \longrightarrow \tau_L(L_0) \cap L_1, \\

3173:  & q: L_0 \cap L_1 \longrightarrow \tau_L(L_0) \cap L_1

3174: \end{align*}

3175: such that $\tau_L(L_0) \cap L_1$ is the disjoint union of their images. These

3176: maps have the following properties:

3177: \begin{romanlist}

3178: \item \label{item:q-map}

3179: $q$ is the inclusion $q(x) = x$. It preserves the values of the action

3180: functional, $a_{\tau_L(L_0),L_1}(x) = a_{L_0,L_1}(x)$. Moreover, for any $w \in

3181: \tau_L(L_0) \cap L_1$ and $x \in L_0 \cap L_1$ with $w \neq q(x)$ one has

3182: $a_{L_0,L_1}(x) - a_{\tau_L(L_0),L_1}(w) \notin [0;3\epsilon)$.

3183: 

3184: \item \label{item:p-map}

3185: Set $\tilde{x} = p(\tilde{x}_0,x_1)$. Then

3186: \begin{equation} \label{eq:epsilon-change}

3187:  0 \leq a_{\tau_L(L_0),L_1}(\tilde{x}) -

3188:  a_{\tau_L(L_0),L}(\tilde{x}_0) - a_{L,L_1}(x_1)

3189:  < \epsilon.

3190: \end{equation}

3191: Moreover, for any $w \in \tau_L(L_0) \cap L_1$ and $(\tilde{x}_0,x_1) \in

3192: (\tau_L(L_0) \cap L) \times (L \cap L_1)$ with $w \neq p(\tilde{x}_0,x_1)$ one

3193: has $a_{\tau_L(L_0),L_1}(w) - a_{\tau_L(L_0),L}(\tilde{x}_0) - a_{L,L_1}(x_1)

3194: \notin [0;3\epsilon)$.

3195: 

3196: \item \label{item:p-map-two}

3197: Suppose that there are $x_k \in L \cap L_k$, $k = 0,1$, whose preimages $y_k =

3198: \iota^{-1}(x_k)$ are antipodes on $S^n$. Since $\tau|S^n$ is the antipodal map,

3199: $\tilde{x}_0 = \tau_L(x_0)$ is equal to $x_1$ (hence $x_1 \in \tau_L(L_0) \cap

3200: L \cap L_1$, and these are all such triple intersection points). In that case

3201: $p(\tilde{x}_0,x_1) = \tilde{x}_0 = x_1$, and

3202: $a_{\tau_L(L_0),L_1}(p(\tilde{x}_0,x_1)) = a_{\tau_L(L_0),L}(\tilde{x}_0) +

3203: a_{L,L_1}(x_1)$.

3204: \end{romanlist}

3205: \end{lemma}

3206: 

3207: \proof \ref{cond:transverse} and \ref{cond:local} imply that $L_0 \cap L_1 \cap

3208: \im(\iota) = \emptyset$. Since $\tau_L$ is the identity outside $\im(\iota)$,

3209: one has $L_0 \cap L_1 = (\tau_L(L_0) \cap L_1) \setminus \im(\iota)$, so that

3210: $q$ can indeed be defined to be the inclusion. The equality

3211: $a_{\tau_L(L_0),L_1}(x) = a_{L_0,L_1}(x)$ follows from the fact that

3212: $K_{\tau_L}$ vanishes outside $\im(\iota)$.

3213: 

3214: There is a bijective correspondence between pairs $(\tilde{x}_0,x_1) \in

3215: (\tau_L(L_0) \cap L) \times (L \cap L_1)$ and $(y_0,y_1) \in \iota^{-1}(L_0

3216: \cap L) \times \iota^{-1}(L \cap L_1)$, given by setting $y_0 =

3217: \iota^{-1}(\tau_L^{-1}(\tilde{x}_0))$, $y_1 = \iota^{-1}(x_1)$. As a

3218: consequence of \ref{cond:local},

3219: \begin{equation} \label{eq:decompose-intersection}

3220: \iota^{-1}(\tau_L(L_0) \cap L_1) = \bigcup_{y_0,y_1} \tau(T(\lambda)_{y_0})

3221: \cap T(\lambda)_{y_1}.

3222: \end{equation}

3223: Since $\tau$ is $\delta$-wobbly \ref{cond:wobbly} and $dist(y_0,y_1) \geq

3224: 2\pi\delta$ \ref{cond:distance}, one can apply Lemma

3225: \ref{th:local-intersections} which tells us that each subset on the right hand

3226: side of \eqref{eq:decompose-intersection} consists of exactly one point. Fix

3227: temporarily some $(y_0,y_1)$ and write $\tau(T(\lambda)_{y_0}) \cap

3228: T(\lambda)_{y_1} = \{\tilde{y}\}$, $\tilde{x} = \iota(\tilde{y})$. One defines

3229: $p(\tilde{x}_0,x_1) = \tilde{x}$. Then

3230: \begin{align*}

3231:  a_{\tau_L(L_0),L_1}(\tilde{x})

3232:  & = K_{L_1}(\tilde{x}) - K_{\tau_L(L_0)}(\tilde{x}) \\

3233:  & = K_{L_1}(\tilde{x}) - K_{L_0}(\tau_L^{-1}(\tilde{x}))

3234:  - K_{\tau_L}(\tau_L^{-1}(\tilde{x})).

3235: \end{align*}

3236: By construction $\tilde{y}$ lies in the fibre $T(\lambda)_{y_1}$, and since

3237: $K_{L_1} \circ \iota$ is constant on fibres, $K_{L_1}(\tilde{x}) =

3238: K_{L_1}(x_1)$. The same reasoning shows that $K_{L_0}(\tau_L^{-1}(\tilde{x})) =

3239: K_{L_0}(x_0)$, where $x_0 = \tau_L^{-1}(\tilde{x}_0)$. Moreover, one sees from

3240: \eqref{eq:kk-function} that $K_{\tau_L}$ is invariant under $\tau_L$, so that

3241: $K_{\tau_L}(\tau_L^{-1}(\tilde{x})) = K_{\tau_L}(\tilde{x})$. With this and

3242: \eqref{eq:tau-acts} in mind, one continues the computation

3243: \begin{equation} \label{eq:finish-action}

3244: \begin{aligned}

3245:  a_{\tau_L(L_0),L_1}(\tilde{x})

3246:  & = K_{L_1}(x_1) - K_{L_0}(x_0) - K_{\tau_L}(\tilde{x}) \\

3247:  & = a_{L,L_1}(x_1) + a_{L_0,L}(x_0) - K_{\tau_L}(\tilde{x}) \\

3248:  & = a_{L,L_1}(x_1) + a_{\tau_L(L_0),L}(\tilde{x}_0) - K_{\tau_L}(\iota(\tilde{y}))

3249:  - 2\pi R(0).

3250: \end{aligned}

3251: \end{equation}

3252: Lemma \ref{th:local-intersections} also says that $R'(||\tilde{y}||) \geq

3253: \delta$. Combining this with \eqref{eq:action-estimate} and \ref{cond:wobbly}

3254: shows that $- K_{\tau_L}(\iota(\tilde{y})) - 2\pi R(0)$ lies in $[0;\epsilon)$,

3255: which completes our proof of \eqref{eq:epsilon-change}.

3256: 

3257: It is clear from their definitions that $p,q$ are injective. A point of

3258: $\tau_L(L_0) \cap L_1$ falls into $\im(q)$ or $\im(p)$ depending on whether it

3259: lies inside or outside $\im(\iota)$, hence the two images are disjoint and

3260: cover $\tau_L(L_0) \cap L_1$. The transversality follows from Lemma

3261: \ref{th:local-intersections} for $\im(p)$ and from that of $L_0 \cap L_1$ for

3262: $\im(q)$. We now turn to the claim made in the last sentence of

3263: \ref{item:q-map}. Supposing that $w$ is a point of $L_0 \cap L_1$ different

3264: from $x$, one has $|a_{\tau_L(L_0),L_1}(w) - a_{L_0,L_1}(x)| = |a_{L_0,L_1}(w)

3265: - a_{L_0,L_1}(x)| \geq 3\epsilon$ by \ref{cond:action}. In the remaining case,

3266: which is when $w = p(\tilde{x}_0,x_1)$, \eqref{eq:epsilon-change} shows that

3267: $|a_{\tau_L(L_0),L_1}(w) - a_{L_0,L_1}(x)| > |a_{\tau_L(L_0),L}(\tilde{x}_0) +

3268: a_{L,L_1}(x_1) - a_{L_0,L_1}(x)| - \epsilon$. We already know that the supports

3269: of $C',C''$ are at least $4\epsilon$ apart, and one concludes that

3270: $|a_{\tau_L(L_0),L_1}(w) - a_{L_0,L_1}(x)| > 3\epsilon$. A similar argument,

3271: paying a little more attention to signs, proves the parallel statement in

3272: \ref{item:p-map}. Finally, the only non-obvious things in \ref{item:p-map-two}

3273: are the fact that $p(\tilde{x}_0,x_1) = x_1$ and the statement about the

3274: action. But these follow from Lemma \ref{th:local-intersections} and the

3275: definition of $p$, respectively from \eqref{eq:finish-action} and $K_{\tau_L}|L

3276: = -2\pi R(0)$. \qed

3277: 

3278: Lemma \ref{th:new-intersection-points} and \ref{cond:action} imply that the

3279: actions $a_{\tau_L(L_0),L_1}(x)$ of different points $x \in \tau_L(L_0) \cap

3280: L_1$ differ by at least $2\epsilon$. In fact, for two such points which lie in

3281: $\im(q)$ the statement follows from \ref{item:q-map} in the lemma; for two

3282: points which lie in $\im(p)$, from \ref{item:p-map}; and combining the two

3283: parts shows it when one point lies in $\im(p)$ and the other in $\im(q)$. Set

3284: \[

3285: C = CF(\tau_L(L_0),L_1).

3286: \]

3287: We have just seen that this has gap $(0;2\epsilon)$. Define maps $\beta: C'

3288: \rightarrow C$, $\gamma: C \rightarrow C''$ by $\beta(\gen{x_1} \otimes

3289: \gen{\tilde{x}_0}) = \gen{p(\tilde{x}_0,x_1)}$ and

3290: $\gamma(\gen{p(\tilde{x}_0,x_1)}) = 0$, $\gamma(\gen{q(x)}) = \gen{x}$. The

3291: result above shows that these are of order $[0;\epsilon)$ and fit into a short

3292: exact sequence as in Lemma \ref{th:low-energy}\ref{item:gap-three}. What

3293: remains to be done is to realize them as ``low order parts'' of chain maps

3294: $b,c$, and to construct the homotopy $h$.

3295: 

3296: %\input{3b}

3297: 

3298: \subsection{The first map\label{sec:b}}

3299: 

3300: Let $S$ be the surface in Figure \ref{fig:y-piece}, which has three boundary

3301: components $\partial_kS$ and three strip-like ends (two positive and a negative

3302: one). Take the trivial exact symplectic fibration $\pi : E = S \times M

3303: \rightarrow S$, with $\Omega \in \Omega^2(E)$, $\Theta \in \Omega^1(E)$ pulled

3304: back from $\o$, $\theta$. Equip this with the Lagrangian boundary condition $Q

3305: = (\partial_1 S \times \tau_L(L_0)) \cup (\partial_2 S \times L) \cup

3306: (\partial_3 S \times L_1)$; $\kappa_{Q}$ is zero, and $K_{Q}(z,x)$ is equal to

3307: $K_{\tau_L(L_0)}(x)$, $K_L(x)$ or $K_{L_1}(x)$, for $z$ in the respective

3308: component $\partial_kS$. This gives rise to a relative invariant

3309: \[

3310:  \Phi^{rel}_0(E,\pi,Q) : HF(L,L_1) \otimes HF(\tau_L(L_0),L) \rightarrow

3311:  HF(\tau_L(L_0),L_1).

3312: \]

3313: The purpose of this section is to analyse this more closely, on the cochain

3314: level. Fix a complex structure $j$ on $S$, trivial over the ends, and let $U

3315: \subset S$ be the open set shaded in Figure \ref{fig:y-piece}.

3316: 

3317: \includefigure{y-piece}{y-piece.eps}{hb}

3318: 

3319: \begin{lemma} \label{th:transversality-b}

3320: $(E,\pi,Q)$ and $U$ satisfy the conditions of Proposition

3321: \ref{th:nonnegative-one}(i).

3322: \end{lemma}

3323: 

3324: \proof The curvature of $(E,\pi)$ is zero. A section $u(z) = (z,\sigma(z))$ is

3325: horizontal iff $\sigma(z) \equiv x \in M$ is constant, and it further satisfies

3326: $u(\partial S) \subset Q$ iff $x \in \tau_L(L_0) \cap L \cap L_1$. If $W

3327: \subset S$ is a connected open subset which intersects all three boundary

3328: components, the same description applies to partial horizontal sections $w: W

3329: \rightarrow E|W$ with $w(W \cap \partial S) \subset Q$. As a consequence, any

3330: such section can be extended to $u \in \MM^h$. This is in particular true for

3331: $W = U$.

3332: 

3333: Fix some $x \in \tau_L(L_0) \cap L \cap L_1$ and the corresponding constant

3334: section $u \in \MM^h$. By definition $A(u) = 0$, and it remains to prove that

3335: $\ind\,D_{u,J} = 0$. From the description of the points $x$ in Lemma

3336: \ref{th:new-intersection-points}\ref{item:p-map-two}, together with the

3337: corresponding local statement in Lemma \ref{th:local-intersections}, it follows

3338: that there is a symplectic isomorphism $TM_x \iso \C^n$ which takes the tangent

3339: spaces to $\tau_L(L_0)$, $L$, and $L_1$ to $\R^n$, $e^{2 \pi i/3}\R^n$, and

3340: $e^{\pi i/3}\R^n$ respectively. Since the index is independent of the almost

3341: complex structure, we may choose $J = j \times J^M$ to be the product of $j$

3342: and some $\o$-compatible $J^M$ on $M$. We may also assume that the isomorphism

3343: $TM_x \iso \C^n$ takes $J^M_x$ to the standard complex structure. By

3344: definition, the domain of $D_{u,J}$ are sections of the vector bundle

3345: $u^*(TE^v,J|TE^v) \iso S \times TM_x$, with boundary conditions given by the

3346: tangent spaces to $\tau_L(L_0)$, $L$, $L_1$. Its range are $(0,1)$-forms with

3347: values in the same vector bundle. The preceding discussion allows us to

3348: identify

3349: \begin{equation} \label{eq:triangle-conditions}

3350: \begin{split}

3351:  & \WW^1_u = \{ X \in W^{1,p}(S,\C^n) \suchthat

3352:  X_z \in e^{i(1-k)\pi/3}\R^n \text{ for $z \in \partial_kS$} \}, \\

3353:  & \WW^0_{u,J} = L^p(\Lambda^{0,1}S \otimes \C^n).

3354: \end{split}

3355: \end{equation}

3356: In \eqref{eq:connection-formula} take $\nabla = \nabla^{S} \times \nabla^M$ to

3357: be the product of torsion-free connections on $S$ and on $M$. Then the second

3358: term $(\nabla_X J) \circ Du \circ j$ in the formula vanishes, because $Du \circ

3359: j$ takes values in $TE^h$ whereas $\nabla_X J$ is nontrivial only on $TE^v$;

3360: and moreover, the pullback connection $u^*\nabla$ on $u^*TE^v$ is trivial. This

3361: shows that $D_{u,J}$ is the standard $\bar\partial$-operator for functions $S

3362: \rightarrow \C^n$, with boundary conditions \eqref{eq:triangle-conditions}.

3363: %

3364: \includefigure{sausage}{sausage.eps}{hb}%

3365: 

3366: There is a general index formula for such operators, but we prefer to use an ad

3367: hoc gluing trick instead. Consider the compact surface $\bar{S}$ in Figure

3368: \ref{fig:sausage}, which is of genus one with one boundary component.

3369: Parametrize the boundary by a closed path $l : [0;6] \rightarrow

3370: \partial\bar{S}$, such that $l(t) = z_t$ for $t \in \{0,1,2,3,4,5\}$ are the

3371: marked points in Figure \ref{fig:sausage}. Take a smooth nondecreasing function

3372: $\lambda: [0;6] \rightarrow \R$ such that

3373: \[

3374:  \lambda(t) = \begin{cases}

3375:  0 & 0 \leq t \leq 1, \\

3376:  1/3 & 2 \leq t \leq 3, \\

3377:  2/3 & 4 \leq t \leq 5. \\

3378:  1 & t = 6.

3379:  \end{cases}

3380: \]

3381: Let $\bar{D}$ be the $\bar\partial$-operator on the trivial bundle $\bar{S}

3382: \times \C^n \rightarrow \bar{S}$, with boundary condition given by the family

3383: of Lagrangian subspaces $\Lambda_{l(t)} = e^{\pi i \lambda(t)} \R^n \subset

3384: \C^n$. As a loop in the Lagrangian Grassmannian, this represents $n$ times the

3385: standard generator of the fundamental group. Riemann-Roch for compact surfaces

3386: with boundary therefore tells us that $\ind\,\bar{D} = n\chi(\bar{S}) + n = 0$.

3387: On the other hand, one can divide $\bar{S}$ into five pieces (two shaded and

3388: three unshaded ones) as indicated in Figure \ref{fig:sausage}, and add

3389: strip-like ends to each piece; the standard gluing theory for elliptic

3390: operators says that $\ind\,\bar{D}$ is the sum of the indices of the obvious

3391: corresponding operators on those pieces. For each shaded piece, this yields a

3392: copy of $D_{u,J}$ (in one of the two cases, the vector space $\C^n$ has been

3393: rotated by $e^{\pi i/3}$). The unshaded pieces give rise to operators of index

3394: zero. This can be derived from the index theorem of \cite{robbin-salamon93}, or

3395: else by directly deforming the operator to an invertible one. Consider for

3396: instance the two intervals of $\partial\bar{S}$ which are contained in the

3397: leftmost unshaded piece. The Lagrangian subspaces $\Lambda_z$ parametrized by

3398: the points $z$ in one of these intervals are all equal to $e^{\pi i/3}\R^n$;

3399: for the other interval, they are of the form $e^{\pi i s}\R^n$ for $2/3 \leq s

3400: \leq 1$. Since $e^{\pi i/3} \R^n \cap e^{\pi i s}\R^n = 0$ for all such $s$,

3401: the Maslov index for paths \cite{robbin-salamon93} is zero. The same holds for

3402: the other unshaded pieces, and one concludes that $0 = \ind\, \bar{D} = 2\,

3403: \ind \, D_{u,J}$. \qed

3404: 

3405: At this point, fix

3406: \begin{align*}

3407:  & J^{(1)} \in \JJ^{reg}(M,\tau_L(L_0),L), \\

3408:  & J^{(2)} \in \JJ^{reg}(M,L,L_1), \\

3409:  & J^{(3)} \in \JJ^{reg}(M,\tau_L(L_0),L_1).

3410: \end{align*}

3411: Proposition \ref{th:nonnegative-one}(i) tells us that there is a $J \in

3412: \JJ(E,\pi,j,J^{(1)},J^{(2)},J^{(3)})$ which is both horizontal and regular. By

3413: part (ii) of the same result, the coefficients $\nu_J(\tilde{x}_0,x_1,x)$ for

3414: $\tilde{x}_0 \in \tau_L(L_0) \cap L$, $x_1 \in L \cap L_1$, $x \in \tau_L(L_0)

3415: \cap L_1$, are zero whenever

3416: \begin{equation} \label{eq:action-difference}

3417: \chi(\tilde{x}_0,x_1,x) = a_{\tau_L(L_0),L_1}(x) -

3418: a_{\tau_L(L_0),L}(\tilde{x}_0) - a_{L,L_1}(x_1)

3419: \end{equation}

3420: is $<0$. Lemma \ref{th:new-intersection-points}\ref{item:p-map} shows that

3421: $\chi(\tilde{x}_0,x_1,x)$, if $\geq 0$, must be either in $[0;\epsilon)$ or

3422: $[3\epsilon;\infty)$, and that the first case only happens for $x =

3423: p(\tilde{x}_0,x_1)$. Hence, if one writes the relative invariant as

3424: \begin{equation} \label{eq:split-one}

3425: \begin{split}

3426:  & C\Phi_0^{rel}(E,\pi,Q,J) = \phi + (C\Phi_0^{rel}(E,\pi,Q,J)-\phi), \\

3427:  & \phi(\gen{\tilde{x}_0} \otimes \gen{x_1}) =

3428:  \nu_J(\tilde{x}_0,x_1,p(\tilde{x}_0,x_1)) \gen{p(\tilde{x}_0,x_1)}

3429: \end{split}

3430: \end{equation}

3431: then $\phi$ is of order $[0;\epsilon)$, while the second term is of order

3432: $[3\epsilon;\infty)$. The rest of this section contains the proof of the

3433: following result, which determines the low order part:

3434: 

3435: \begin{prop} \label{th:low-nu}

3436: $\nu_J(\tilde{x}_0,x_1,p(\tilde{x}_0,x_1)) = 1$ for all $(\tilde{x}_0,x_1)$.

3437: \end{prop}

3438: 

3439: There is one particular case of this which follows from the previous

3440: considerations. Namely, suppose that there is a pair $(\tilde{x}_0,x_1)$ with

3441: $\tilde{x}_0 = x_1 = x \in \tau_L(L_0) \cap L \cap L_1$. In that case $p(x,x) =

3442: x$ and $\chi(x,x,x) = 0$ by Lemma

3443: \ref{th:new-intersection-points}\ref{item:p-map-two}, and therefore

3444: $\nu_J(x,x,x) = \# \MM^h(x,x,x)$ by Proposition \ref{th:nonnegative-one}(ii).

3445: Now $\# \MM^h(x,x,x) = 1$ because, as we saw already when proving Lemma

3446: \ref{th:transversality-b}, the unique element of that space is the constant

3447: section $u(z) = (z,x)$. Our strategy will be to reduce the computation of all

3448: the $\nu_J(\tilde{x}_0,x_1,p(\tilde{x}_0,x_1))$ to this case, by using suitable

3449: deformations.

3450: 

3451: What ``deformation'' means here is keeping the submanifolds $L,L_0,L_1$ and the

3452: constant $\epsilon$ fixed, while changing the remaining data. More precisely,

3453: for $0 \leq \mu \leq 1$ we consider the following: a smooth family $f^\mu$ of

3454: diffeomorphisms $S^n \rightarrow L$; smoothly varying positive numbers

3455: $\lambda^\mu$, and symplectic embeddings $\iota^\mu: T(\lambda^\mu) \rightarrow

3456: M$ with $\iota^\mu|T(0) = f^\mu$; the Dehn twists $\tau_L^\mu$ defined using

3457: $\iota^\mu$ and functions $R^\mu$ supported in $(-\infty;\lambda^\mu)$; and

3458: constants $\delta^\mu$. These should agree with the given data

3459: $f,\lambda,\iota,R,\tau_L,\delta$ for $\mu = 0$. We also require that the

3460: analogues of \ref{cond:distance}--\ref{cond:wobbly} continue to hold for any

3461: $\mu$; where in \ref{cond:wobbly} we take the original $\epsilon$ throughout.

3462: 

3463: Choose $\tilde{x}_0 \in \tau_L(L_0) \cap L$, $x_1 \in L \cap L_1$. Given a

3464: ``deformation'' in the sense which we have just explained, one can set

3465: $\tilde{x}_0^\mu = \tau_L^\mu(\tau_L)^{-1}(\tilde{x}_0) \in \tau_L^\mu(L_0)

3466: \cap L$ and $x_1^\mu = x_1 \in L \cap L_1$, which ``continues'' the points

3467: smoothly into the deformed situation. We claim that $x = p(\tilde{x}_0,x_1)$

3468: fits similarly into a smooth family $x^\mu \in \tau_L^\mu(L_0) \cap L_1$. The

3469: point is that since \ref{cond:transverse}--\ref{cond:wobbly} continue to hold,

3470: Lemma \ref{th:new-intersection-points} can be applied to the situation for any

3471: $\mu$. This ensures that the intersections $\tau_L^\mu(L_0) \cap L_1$ remain

3472: transverse, which implies that a unique family $x^\mu$ exists. In fact, it even

3473: provides a smooth family of injective maps $p^\mu: (\tau_L^\mu(L_0) \cap L)

3474: \times (L \cap L_1) \rightarrow \tau_L^\mu(L_0) \cap L_1$, such that $x^\mu =

3475: p^\mu(\tilde{x}_0^\mu,x_1^\mu)$.

3476: 

3477: \begin{lemma} \label{th:antipodes}

3478: For any $(\tilde{x}_0,x_1)$ there is a ``deformation'' such that $\tilde{x}_0^1

3479: = x_1^1$.

3480: \end{lemma}

3481: 

3482: \proof From \ref{cond:distance} we know that $y_0 =

3483: f^{-1}(\tau_L^{-1}(\tilde{x}_0))$, $y_1 = f^{-1}(x_1)$ are points on $S^n$

3484: whose distance is $\geq 2\pi \delta$. Let $g^\mu \in \Diff(S^n)$ be an isotopy,

3485: $g^0 = \id$, such that $g^1(y_0)$, $g^1(y_1)$ are antipodes. There are $C^\mu

3486: \geq 1$, smoothly depending on $\mu$ and with $C^0 = 1$, with the property that

3487: \[

3488: C^\mu \geq ||D(g^\mu)_y||, \; ||D(g^\mu)^{-1}_y|| \quad \text{for all $y \in

3489: S^n$.}

3490: \]

3491: Consider the ``deformation'' $f^\mu = f \circ (g^\mu)^{-1}$, $\lambda^\mu =

3492: \lambda/C^\mu$, $\iota^\mu = \iota \circ G^\mu | T(\lambda^\mu)$, $\delta^\mu =

3493: \delta/C^\mu$; here $G^\mu \in \Sympe(T)$ is induced by $g^\mu$, in the sense

3494: that $G^\mu|T(0) = (g^\mu)^{-1}$. The bound on $D(g^\mu)$ implies that $G^\mu$

3495: maps $T(\lambda^\mu)$ to $T(\lambda)$, so that $\iota^\mu$ is well-defined. On

3496: the other hand, because of the bound on $D(g^\mu)^{-1}$, the distance between

3497: any point of $(f^\mu)^{-1}(L \cap L_0) = g^\mu f^{-1}(L \cap L_0)$ and any

3498: point of $(f^\mu)^{-1}(L \cap L_1) = g^\mu f^{-1}(L \cap L_1)$ is $\geq

3499: 2\pi\delta^\mu$, which shows that \ref{cond:distance} holds during the

3500: deformation. For \ref{cond:local} it is sufficient to note that $G^\mu$ takes

3501: the canonical one-form $\theta_T$ to itself and maps fibres to fibres. And it

3502: is no problem to find functions $R^\mu$ which satisfy \ref{cond:wobbly} with

3503: the $\delta^\mu$ defined above and the given $\epsilon$. By definition

3504: \[

3505:  \tau_L|L = f \circ A \circ f^{-1}, \quad

3506:  \tau_L^\mu|L = f^\mu \circ A \circ (f^\mu)^{-1}

3507:  = f \circ (g^\mu)^{-1} \circ A \circ g^\mu \circ f^{-1}.

3508: \]

3509: Since $g^1(y_1) = A(g^1(y_0))$ by construction, one sees that

3510: \[

3511: \tilde{x}_0^1 = \tau_L^1 (\tau_L^{})^{-1}(\tilde{x}_0)

3512:  = f \circ (g^1)^{-1} \circ A \circ g^1 (y_0) = f(y^1)

3513:  = x_1 = x_1^1. \qed

3514: \]

3515: 

3516: Given a ``deformation'', one can repeat the construction at the beginning of

3517: this section in a parametrized sense, which means equipping the trivial exact

3518: symplectic symplectic fibration $(E,\pi)$ with a smooth family $Q^\mu$ of

3519: Lagrangian boundary conditions, modelled over the ends on the pairs of exact

3520: Lagrangian submanifolds $(\tau_L^\mu(L_0),L)$, $(L,L_1)$, and

3521: $(\tau_L^\mu(L_0),L_1)$. Since their intersections are transverse for all

3522: $\mu$, it makes sense to consider parametrized moduli spaces of

3523: pseudo-holomorphic sections. To do that, take smooth families of almost complex

3524: structures $J^{(1),\mu}, J^{(2),\mu}, J^{(3),\mu} \in \JJ(M)$ which reduce to

3525: the previously chosen ones for $\mu = 0$, and similarly a family $J^\mu \in

3526: \JJ^h(E,\pi,j,J^{(1),\mu},J^{(2),\mu},J^{(3),\mu})$. The parametrized moduli

3527: space we are interested in is

3528: \[

3529: \MM^{para} = \bigcup_{\mu}\, \{\mu\} \times

3530: \MM_{J^\mu}(\tilde{x}_0^\mu,x_1^\mu,x^\mu)

3531: \]

3532: for points $\tilde{x}_0^\mu,x_1^\mu,x^\mu$ as introduced above.

3533: 

3534: \begin{lemma} \label{th:no-bubbles}

3535: $\MM^{para}$ is compact.

3536: \end{lemma}

3537: 

3538: \proof Consider $CF(L,L_1)$, $CF(\tau_L^\mu(L_0),L)$,

3539: $CF(\tau_L^\mu(L_0),L_1)$. We already know that the first of these $\R$-graded

3540: vector spaces has gap $(0;3\epsilon)$; for the second one, the same is true

3541: because it agrees with $CF(L_0,L)$ up to a constant shift in the grading; and

3542: by repeating the considerations after Lemma \ref{th:new-intersection-points},

3543: one can show that $CF(\tau_L^\mu(L_0),L_1)$ has gap $(0;2\epsilon)$. As another

3544: consequence of Lemma \ref{th:new-intersection-points}\ref{item:p-map} in the

3545: deformed situation, any $(\mu,u) \in \MM^{para}$ satisfies

3546: \[

3547:  A(u) = a_{\tau_L^\mu(L_0),L_1}(x^\mu) -

3548:  a_{\tau_L^r(L_0),L}(\tilde{x}_0^\mu) - a_{L,L_1}(x_1^\mu) \in [0;\epsilon).

3549: \]

3550: One can now argue exactly as in Lemma \ref{th:energy-compactness}; there are no

3551: points at infinity in the parametrized Gromov-Floer compactification, since the

3552: principal component of any such point would have negative action, which would

3553: contradict the fact that $(E,\pi)$ has nonnegative curvature. \qed

3554: 

3555: Suppose now that the $J^{(k),\mu}$ and $J^\mu$ for $\mu = 1$ have been chosen

3556: regular; for $J^1$ this can be achieved without leaving the class of horizontal

3557: almost complex structure, because Lemma \ref{th:transversality-b} equally

3558: applies to the deformed situation for $\mu = 1$. By using a parametrized

3559: version of the same lemma and of Proposition \ref{th:nonnegative-one}(i), one

3560: sees that in addition, the family $(J^\mu)$ can be chosen to be regular in the

3561: parametrized sense. Then the one-dimensional part of $\MM^{para}$ is a compact

3562: one-manifold, and its boundary points are precisely those with $\mu = 0$ or

3563: $1$. One concludes that

3564: \[

3565: \nu_J(\tilde{x}_0,x_1,x) = \nu_{J^1}(\tilde{x}_0^0,x_1^1,x^1).

3566: \]

3567: If the ``deformation'' is as in Lemma \ref{th:antipodes}, the situation for $r

3568: = 1$ is exactly the special case which we have already discussed, so that

3569: $\nu_{J^1}(\tilde{x}_0^0,x_1^1,x^1) = 1$. Since such deformations exist for all

3570: $(\tilde{x}_0,x_1)$, Proposition \ref{th:low-nu} is proved.

3571: 

3572: From now on, we fix some $J = J^{(4)} \in

3573: \JJ^{reg,h}(E,\pi,Q,j,J^{(1)},J^{(2)},J^{(3)})$ and write $b =

3574: C\Phi_0^{rel}(E,\pi,Q,J^{(4)}): C' \rightarrow C$ for the relative invariant on

3575: the cochain level. Let $\beta$ be the map defined at the end of Section

3576: \ref{sec:setup}. What \eqref{eq:split-one} and Proposition \ref{th:low-nu} say

3577: is that $b = \beta + (b - \beta)$, with $b-\beta$ of order

3578: $[3\epsilon;\infty)$, which is even slightly more than required by Lemma

3579: \ref{th:low-energy}.

3580: 

3581: \includefigure{triangle}{triangle.eps}{ht}%

3582: %

3583: \begin{remark}

3584: Our relative invariant is the well-known pair-of-pants product, or Donaldson

3585: product; in fact, if $u(z) = (z,\sigma(z))$ lies in $\MM_J(\tilde{x}_0,x_1,x)$,

3586: then $\sigma: S \rightarrow M$ is a ``pseudo-holomorphic triangle'' whose sides

3587: lie on $\tau_L(L_0), L, L_1$ and whose vertices are $\tilde{x}_0,x_1,x$.

3588: Proposition \ref{th:low-nu} asserts that there is an odd number of low-area

3589: triangles with certain specified vertices. In the lowest dimension, $n = 1$,

3590: one can see directly that there is precisely one such triangle (Figure

3591: \ref{fig:triangle}). In higher dimensions it is still easy to construct

3592: explicitly the analogue of this particular triangle, but proving that there are

3593: no others seems more difficult. For that reason, we have preferred to take the

3594: indirect approach via ``deformations'', which effectively meant moving

3595: $\tau_L(L_0)$ in such a way that the area of the triangle shrinks to zero.

3596: \end{remark}

3597: 

3598: %\input{3c}

3599: 

3600: \subsection{The second map\label{sec:c}}

3601: 

3602: From this point onwards, we add one more assumption to those in Section

3603: \ref{sec:setup}:

3604: \begin{Romanlist}

3605: \setcounter{enumi}{5}

3606: \item \label{cond:r}

3607: $R$ is the function $R_r$ which appears in the construction of exact Lefschetz

3608: fibrations in Lemma \ref{th:model-fibrations}, with the given $\lambda$ and

3609: some $0<r<1/2$ which we are free to choose; see more specifically

3610: \eqref{eq:r-function} for this function.

3611: \end{Romanlist}

3612: 

3613: Since that severely restricts the choice of $R$, one needs to worry about

3614: possible conflicts with the other conditions, or that it might restrict the

3615: ultimate scope of the exact sequence. An inspection of Remark \ref{re:scope}

3616: shows that the only issue is whether, using $R = R_r$, one can satisfy

3617: \ref{cond:wobbly} with arbitrarily small $\delta$ and $\epsilon$. That is taken

3618: care of by Lemma \ref{th:standard-properties}\ref{item:precise-twist}, which

3619: shows that it suffices to choose $r$ small.

3620: 

3621: \includefigure{c-pasting-ii}{c-ii.eps}{hb} %

3622: %

3623: The obvious reason for introducing \ref{cond:r} is that there is a standard

3624: fibration $(E^L,\pi^L)$ over a disc $\bar{D}(r)$, with $r$ small, whose

3625: monodromy around $\partial \bar{D}(r)$ is $\tau_L$. Following Remark

3626: \ref{re:squeeze-boundary}, we want to modify this by a pullback. Take $S^p =

3627: \bar{D}(1/2)$ and a map $p: S^p \rightarrow \bar{D}(r)$ of the form $p(z) =

3628: g(|z|)(z/|z|)$, where $g$ is a function with $g(t) = t$ for small $t$, $g(t) =

3629: r$ for $t \geq r$, and $g'(t) \geq 0$ everywhere. Then

3630: \[

3631: (E^p,\pi^p) = p^*(E^L,\pi^L)

3632: \]

3633: is again an exact Lefschetz fibration. It has nonnegative curvature; this

3634: follows from Lemma \ref{th:standard-properties}\ref{item:nonnegatively-curved}

3635: and the fact that $det(Dp) \geq 0$. Moreover, it is flat on the annulus $S^p

3636: \setminus D(r)$; and using the isomorphism $\phi^p: (E^p)_{1/2} \rightarrow M$

3637: inherited from $\phi^L: (E^L)_r \rightarrow M$, one can identify its monodromy

3638: around $\partial S^p$ with $\tau_L$. To represent this property, we draw

3639: $(E^p,\pi^p)$ as in Figure \ref{fig:c-pasting-ii}.

3640: 

3641: \includefigure{c-pasting-i}{c-i.eps}{hb} %

3642: %

3643: Now take the surface $S^f = (\R \times [-1;1]) \setminus D(1/2) \subset \R^2$,

3644: with coordinates $(s,t)$, and divide it into two parts $S^{f,\pm} = S^f \cap \{

3645: t \in \R^{\pm} \}$, so that $S^{f,+} \cap S^{f,-} = ((-\infty;-1/2] \cup

3646: [1/2;\infty)) \times \{0\}$. Consider trivial fibrations $\pi^{f,\pm}:

3647: E^{f,\pm} = S^{f,\pm} \times M \rightarrow S^{f,\pm}$ over the two parts, and

3648: equip them with differential forms $\Omega^{f,\pm}$, $\Theta^{f,\pm}$ as

3649: follows. $\Omega^{f,\pm}$ is the pullback of $\o \in \Omega^2(M)$, and

3650: similarly $\Theta^{f,+}$ is the pullback of $\theta$; finally $\Theta^{f,-} =

3651: \theta - d(\beta(s)K_{\tau_L})$, where $\beta$ is a function with $\beta(s) =

3652: 0$ for $s \geq 1/4$, $\beta(s) = 1$ for $s \leq -1/4$. Define a fibration

3653: $(E^f,\pi^f)$ over $S^f$ by identifying the fibres

3654: \[

3655: E^{f,+}_{(s,0)} \longrightarrow E^{f,-}_{(s,0)}

3656: \]

3657: via $\id_M$ for $s \geq 1/2$, respectively via $\tau_L$ for $s \leq -1/2$.

3658: $\Omega^{f,\pm}$ and $\Theta^{f,\pm}$ match up to forms $\Omega^f$, $\Theta^f$;

3659: the first because $\tau_L$ is symplectic, and the second as a consequence of

3660: our choice of $\Theta^{f,-}$. This makes $(E^f,\pi^f)$ into a (flat) exact

3661: symplectic fibration; it is represented in Figure \ref{fig:c-pasting-i}.

3662: 

3663: \includefigure{c-pasting-iii}{c-iii.eps}{hb} %

3664: %

3665: We now carry out a pasting construction of the kind discussed in Section

3666: \ref{sec:basic}. If one identifies the fibres of $E^p$ and $E^f$ at the point

3667: $1/2$ by using $\phi^p: (E^p)_{1/2} \rightarrow M = (E^f)_{1/2}$, then the

3668: monodromies around the circle $|z| = 1/2$ coincide, being both equal to

3669: $\tau_L$. Since the two fibrations are flat close to this circle, one can paste

3670: them together to an exact Lefschetz fibration, denoted by $(E,\pi)$, over $S =

3671: S^p \cup S^f = \R \times [-1;1]$; this is drawn in Figure

3672: \ref{fig:c-pasting-iii}. Equip $(E,\pi)$ with the Lagrangian boundary condition

3673: $Q$ which is the union of $\R \times \{1\} \times L_1 \subset E^{f,+}$ and $\R

3674: \times \{-1\} \times \tau_L(L_0) \subset E^{f,-}$, with $\kappa_Q = 0$, and

3675: with a function $K_Q$ which is $K_{L_1}$ on $\R \times \{1\} \times L_1$ and

3676: $K_{\tau_L(L_0)} - \beta(s)K_{\tau_L}|\tau_L(L_0)$ on $\R \times \{-1\} \times

3677: \tau_L(L_0)$. This is clearly modelled on $(\tau_L(L_0),L_1)$ over the positive

3678: end of $S$; over the negative end it is modelled on $(L_0,L_1)$, as one sees

3679: using the trivialization

3680: \begin{align*}

3681:  & \qquad \xymatrix{

3682:  {(-\infty;-1] \times [-1;1] \times M} \ar[r]^-{\Gamma} \ar[d] & {E} \ar[d]^\pi \\

3683:  {(-\infty;-1] \times [-1;1]} \ar[r]^-{\gamma} & {S}

3684:  } \\

3685:  &

3686:  \gamma = \text{inclusion}, \qquad

3687:  \Gamma(s,t,x) = \begin{cases}

3688:  (s,t,x) \in E^{f,+} & t \geq 0, \\

3689:  (s,t,\tau_L(x)) \in E^{f,-} & t \leq 0.

3690:  \end{cases}

3691: \end{align*}

3692: We now get a relative invariant $\Phi^{rel}_0(E,\pi,Q): HF(\tau_L(L_0),L_1)

3693: \rightarrow HF(L_0,L_1)$. Following the same pattern as in the previous section

3694: (but with considerably less technical difficulties), we need to determine

3695: partially the underlying cochain map.

3696: 

3697: \begin{lemma} \label{th:c}

3698: $(E,\pi,Q)$ together with $U = (-1;1) \times [-1;1] \subset S$ satisfies the

3699: conditions of Proposition \ref{th:nonnegative-one}(i). Moreover, given $w \in

3700: \tau_L(L_0) \cap L_1$ and $x \in L_0 \cap L_1$, the space $\MM^h(x,w)$ contains

3701: precisely one horizontal section if $w = q(x)$, and is empty otherwise.

3702: \end{lemma}

3703: 

3704: \proof Because the two parts from which it is assembled have nonnegative

3705: curvature, so does $(E,\pi)$. To any point $x \in L_0 \cap L_1$ one can

3706: associate a horizontal section $u^f: S^f \rightarrow E^f$ satisfying

3707: $u^f(\partial S) \subset Q$, which is defined by

3708: \[

3709:  u^f(s,t) = \begin{cases}

3710:  (s,t,x) \in E^{f,+} & t \geq 0, \\

3711:  (s,t,\tau_L(x)) \in E^{f,-} & t \leq 0.

3712:  \end{cases}

3713: \]

3714: Condition \ref{cond:transverse} and \ref{cond:local} imply that $x \notin

3715: \im(\iota)$. By construction $E^L$ contains a trivial part $\bar{D}(r) \times

3716: (M \setminus \im(\iota))$, see Proposition \ref{th:standard-fibrations}. The

3717: pullback $E^p$ has a corresponding property, which means that there is a

3718: horizontal section $u^p$ of $E^p$, matching up with $u^f$ to form a $u \in

3719: \MM^h(x,x) \subset \MM^h$.

3720: 

3721: $\int_{S^f} (u^f)^*\Omega^f = 0$ by definition of $\Omega^f$, and similarly

3722: $\int_{S^p} (u^p)^*\Omega^p = 0$ because the image of $u^p$ lies in the trivial

3723: part of $E^p$. It follows that the section we have constructed satisfies $A(u)

3724: = 0$. The connection $\nabla^u$ on $u^*TE^v$ is trivial; in fact, by inspecting

3725: the details of the construction, one can see that $(E,\pi)$ is symplectically

3726: trivial in a neighbourhood of $\im(u)$. By linearization of basic fact that

3727: symplectic parallel transport preserves any Lagrangian boundary condition, one

3728: finds that the subbundle $u^*(TQ \cap TE^v) \subset u^*TE^v|\partial S$ is

3729: preserved under $\nabla^u$; hence it is trivial. To summarize, we have found

3730: that one can identify $u^*(TE^v) \iso S \times \C^n$ symplectically, in such a

3731: way that $\nabla^u$ becomes trivial, and that $u^*(TQ \cap TE^v)$ is mapped to

3732: the subbundle $(\R \times \{-1\} \times \Lambda_{-1}) \cup (\R \times \{1\}

3733: \times \Lambda_1)$ for some Lagrangian subspaces $\Lambda_{\pm 1} \subset

3734: \C^n$; by looking at the positive end, one sees that $\Lambda_{-1}$ and

3735: $\Lambda_1$ are transverse. Then the index formula in terms of the Maslov index

3736: for paths \cite{robbin-salamon93} shows that $\ind\,D_{u,J} = 0$.

3737: 

3738: Next, suppose that $u \in \MM^h$ is an arbitrary horizontal section satisfying

3739: $u(\partial S) \subset Q$. When restricted to $S^{f,\pm}$, this is of the form

3740: $u^{f,\pm}(z) = (z,x^{\pm})$ for points $x^+ \in L_1$, $x^- \in \tau_L(L_0)$.

3741: The condition for the two parts to match along $S^{f,+} \cap S^{f,-}$ is that

3742: $x^- = x^+ = \tau_L(x^+)$. In particular $x^{\pm} \in L_0 \cap L_1$. Because of

3743: the strong unique continuation property of horizontal sections, it follows that

3744: the construction above yields all of $\MM^h$. By the same argument, any partial

3745: horizontal section $U \rightarrow E|U$ with boundary in $Q$ can be extended to

3746: some element of $\MM^h$. \qed

3747: 

3748: Let $j$ be some complex structure on $S$, standard over the ends. Take the same

3749: $J^{(3)} \in \JJreg(M,\tau_L(L_0),L_1)$ as in the previous section, and choose

3750: an additional $J^{(5)} \in \JJreg(L_0,L_1)$. By Lemma \ref{th:c} and

3751: Proposition \ref{th:nonnegative-one}, one can find a $J^{(6)} \in

3752: \JJ(E,\pi,Q,j,J^{(3)},J^{(5)})$ which is both horizontal and regular. Write

3753: \[

3754: c = C\Phi^{rel}_0(E,\pi,Q,J^{(6)}): C \longrightarrow C''

3755: \]

3756: for the chain map defined by this. In view of Lemmas \ref{th:nonnegative-two}

3757: and \ref{th:new-intersection-points}\ref{item:q-map} one can write $c = \phi +

3758: (c-\phi)$, where $\phi$ depends only on the sections in $\MM^h$ and is of order

3759: $\{0\}$, and the remaining term is of order $[3\epsilon;\infty)$. Again

3760: applying Lemma \ref{th:c}, one finds that $\phi$ is precisely the map $\gamma$

3761: defined at the end of Section \ref{sec:setup}. Again, this is marginally better

3762: than what is needed to apply Lemma \ref{th:low-energy}.

3763: 

3764: %\input{3d}

3765: 

3766: \subsection{The homotopy\label{sec:h}}

3767: 

3768: At this point it becomes necessary to change the notation slightly, in order to

3769: avoid conflicts. Thus, the fibration used in Section \ref{sec:b} to define the

3770: map $b$ will be denoted by $(E^b,\pi^b)$, its base by $S^b$, and its Lagrangian

3771: boundary condition by $Q^b$; and correspondingly we write $(E^c,\pi^c)$, $S^c$,

3772: $Q^c$ for the objects constructed in Section \ref{sec:c} to define $c$. Over

3773: the unique negative end of $S^b$, $Q^b$ is modelled on $(\tau_L(L_0),L_1)$, and

3774: the same holds for $Q^c$ over the positive end of $S^c$. As described in

3775: Section \ref{sec:relative-invariants}, one can glue these ends together to

3776: obtain a new exact Lefschetz fibration $(E^{bc},\pi^{bc})$ over a surface

3777: $S^{bc}$, with a Lagrangian boundary condition $Q^{bc}$. The outcome is

3778: represented schematically in Figure \ref{fig:h-map-one}.

3779: 

3780: \includefigure{h-map-one}{h-i.eps}{ht}%

3781: %

3782: $(E^{bc},\pi^{bc})$ has nonnegative curvature because $(E^b,\pi^b)$ and

3783: $(E^c,\pi^c)$ have that property. Moreover $\MM^h = \emptyset$, which means

3784: that there are no horizontal sections $u: S^{bc} \rightarrow E^{bc}$ with

3785: $u(\partial S^{bc}) \subset Q^{bc}$. In fact, assuming that such a $u$ exists,

3786: one could reverse the gluing construction and obtain a horizontal section $u^b$

3787: of $E^b$ with boundary in $Q^b$, as well as a similar section $u^c$ of $E^c$.

3788: We have seen previously that such $u^b$ correspond to points in $\tau_L(L_0)

3789: \cap L \cap L_1$, and $u^c$ to points in $L_0 \cap L_1$. In our case, the two

3790: sections would have to match over the ends used in the gluing process, which

3791: means that they would correspond to a point of $L_0 \cap L \cap L_1$; but that

3792: is impossible by \ref{cond:transverse}. The same argument shows that for a

3793: sufficiently large relatively compact subset $U \subset S^{bc}$, there are no

3794: horizontal $w: U \rightarrow E^{bc}$ with $w(\partial S^{bc} \cap U) \subset

3795: Q^{bc}$. Over the ends, $Q^{bc}$ is modelled on $(\tau_L(L_0),L)$, $(L,L_1)$

3796: and $(L_0,L_1)$, respectively. For these pairs of submanifolds we have already

3797: chosen almost complex structures $J^{(1)}$, $J^{(2)}$ and $J^{(5)}$,

3798: respectively. Let $j$ be some complex structure on $S^{bc}$ which is standard

3799: over the ends. Lemma \ref{th:nonnegative-one} ensures that one can choose a $J

3800: \in \JJ(E^{bc},\pi^{bc},Q^{bc},j,J^{(1)},J^{(2)},J^{(5)})$ which is both

3801: horizontal and regular.

3802: 

3803: \begin{lemma} \label{th:bc}

3804: For any such $J$,

3805: \[

3806: C\Phi_0^{rel}(E^{bc},\pi^{bc},Q^{bc},J): C' \longrightarrow C''

3807: \]

3808: is homotopic to $c \circ b$ by a chain homotopy which is of order $(0;\infty)$.

3809: \end{lemma}

3810: 

3811: \proof Suppose first that $j$ is induced from the complex structures on

3812: $S^b,S^c$, and that $J$ is similarly constructed from $J^{(4)}$ and $J^{(6)}$;

3813: this is automatically horizontal. Proposition \ref{th:gluing-ends} says that

3814: for large values of the gluing parameter $\sigma$, $J$ is regular; and then we

3815: have moreover $C\Phi_0(E^{bc},\pi^{bc},Q^{bc},J) = c \circ b$ by

3816: \eqref{eq:chain-composition}. On the other hand, the observations made above

3817: allow us to apply Lemma \ref{th:nonnegative-homotopy}, which shows that the

3818: maps $C\Phi_0$ for any two choices of $j$ and $J$ are homotopic by a chain

3819: homotopy of order $(0;\infty)$. \qed

3820: 

3821: \includefigure{h-map-two}{h-ii.eps}{hb} %

3822: %

3823: The proof that $c \circ b$ is chain homotopic to zero relies on an alternative

3824: construction of the same exact Lefschetz fibration. Consider the surface $S^o$

3825: from Figure \ref{fig:h-map-two}, embedded into $\R^2$ with coordinates $(s,t)$.

3826: As in the previous section, we divide it into $S^{o,\pm} = S^o \cap \{t \in

3827: \R^{\pm}\}$ and take the trivial fibrations $\pi^{o,\pm}: E^{o,\pm} = S^{o,\pm}

3828: \times M \rightarrow S^{o,\pm}$. We equip $E^{o,\pm}$ with the forms

3829: $\Omega^{o,\pm}$ pulled back from $\o$, and $E^{o,+}$ with the one-form

3830: $\Theta^{o,+}$ pulled back from $\theta$; while on $E^{o,-}$ we take

3831: $\Theta^{o,-} = \theta - d(\eta(t)K_{\tau_L})$, where $\eta$ is some function

3832: with $\eta(t) = 1$ for $t \geq -1$, $\eta(t) = 0$ for $t \leq -2$. One now

3833: identifies the fibres over $z \in \R^- \times \{0\} = S^{o,+} \cap S^{o,-}$ by

3834: using $\tau_L: E^{o,+}_z \rightarrow E^{o,-}_z$, which yields an exact

3835: symplectic fibration $(E^o,\pi^o)$ over $S^o$ (it is in fact trivial, but for

3836: us it is convenient to think of it as being built up in this particular way).

3837: Take the pieces of $\partial S^o$ labeled in Figure \ref{fig:h-map-three}, and

3838: construct a Lagrangian boundary condition $Q^o$ for $(E^o,\pi^o)$ as the union

3839: of $\partial_1S^{o,-} \times \tau_L(L_0), \,\, \partial_2S^{o,-} \times L

3840: \subset E^{o,-}$ and $\partial_2S^{o,+} \times L, \,\, \partial_3S^{o,+} \times

3841: L_1 \subset E^{o,+}$. The associated function $K_{Q^o}$ is equal to

3842: $K_{\tau_L(L_0)} - \eta(t)K_{\tau_L}|\tau_L(L_0)$ over $\partial_1S^{o,-}$, to

3843: $K_{L_1}$ over $\partial_3S^{o,+}$, and zero on the rest. $\kappa_{Q^o}$ is

3844: equal to $-(K_{\tau_L}|L) \eta'(t) dt$ on $\partial_2S^{o,-}$ and vanishes

3845: elsewhere; this makes sense because $K_{\tau_L}|L$ is constant, equal to $-2\pi

3846: R(0)$ by \eqref{eq:kk-function}. Combining this with \ref{cond:wobbly} shows

3847: that

3848: \begin{equation} \label{eq:kappao}

3849: \int_{\partial S^o} \kappa_{Q^o} = 2\pi R(0) \in (-\epsilon;0].

3850: \end{equation}

3851: $Q^o$ is modelled on $(\tau_L(L_0),L)$ and $(L,L_1)$ over the positive ends,

3852: and on $(L_0,L_1)$ over the negative end; a suitable trivialization of

3853: $(E^o,\pi^o)$ over that end can be defined in the same way as in Section

3854: \ref{sec:c}. The next statement is a straightforward consequence of

3855: \ref{cond:transverse}:

3856: 

3857: \begin{lemma} \label{th:o-sections}

3858: Take the subset $U^o \subset S^o$ shaded in Figure \ref{fig:h-map-three}. Then

3859: there are no horizontal sections $u: U^o \rightarrow E^o$ with $u(\partial S^o

3860: \cap U^o) \subset Q^o$. \qed

3861: \end{lemma}

3862: 

3863: \includefigure{h-map-three}{h-iii.eps}{ht} %

3864: %

3865: For the following step, we need the fibration $(E^p,\pi^p)$ over $S^p =

3866: \bar{D}(1/2)$ from Section \ref{sec:c}, which was defined as pullback of a

3867: standard fibration. The standard boundary condition from Section

3868: \ref{sec:vanishing} pulls back to a Lagrangian boundary condition $Q^p$ for it.

3869: As it stands $\kappa_{Q^p} = d^c(-\quarter|z|)$ is nonzero everywhere, but for

3870: us it is better to modify it by some exact one-form, in such a way that it

3871: becomes zero near $-1/2 \in \partial S^p$. This can be compensated by a change

3872: of $K_{Q^p}$, so that the whole remains a Lagrangian boundary condition. It is

3873: a consequence of Proposition \ref{th:vanishing} that the $\Phi_1$-invariant of

3874: $(E^p,\pi^p,Q^p)$ vanishes, since this can be joined to $(E^L,\pi^L,Q^L)$ by a

3875: smooth deformation of exact Lefschetz fibrations with Lagrangian boundary

3876: conditions.

3877: 

3878: \begin{lemma} \label{th:p-sections}

3879: Let $U^p \subset S^p$ be the complement of a sufficiently small neighbourhood

3880: of $-1/2 \in \partial S^p$. Then there are no partial horizontal sections $w:

3881: U^p \rightarrow E^p$ satisfying $w(\partial S^p \cap U^p) \subset Q^p$.

3882: \end{lemma}

3883: 

3884: \proof From the standard fibration, $E^p$ inherits a smooth family of

3885: Lagrangian spheres $\Sigma^p_z \subset E^p_z$, $z \neq 0$. These are carried

3886: into each other by parallel transport along any path, and they degenerate to

3887: the critical point $x_0 \in E^p_0$ as $z \rightarrow 0$. Now our boundary

3888: condition is made up of the $\Sigma^p_z$ for $z \in \partial S^p$; therefore a

3889: section $w$ with the properties stated above would satisfy $w(z) \in

3890: \Sigma^p_z$ for all $z \in \partial S^p \cap U^p$, and by parallel transport

3891: for all $z \in U^p \neq \{0\}$. In the limit this yields $w(0) = x_0$, but that

3892: is impossible since $x_0$ is a critical point. \qed

3893: 

3894: One can identify the fibre of $E^o$ over $\zeta^o = (0,0) \in \partial S^o$

3895: with the fibre of $E^p$ over $\zeta^p = -1/2 \in \partial S^p$ via

3896: \begin{equation} \label{eq:identify}

3897: (E^o)_{\zeta^o} \iso (E^{o,+})_{(0,0)} = M \xleftarrow{\phi^p} E^p_{1/2} \iso

3898: (E^p)_{\zeta^p}

3899: \end{equation}

3900: where the last isomorphism is parallel transport along the upper semi-circle

3901: $(1/2)e^{it}$, $0 \leq t \leq \pi$. By putting together the various

3902: definitions, one sees that \eqref{eq:identify} takes $(Q^o)_{(0,0)}$ to

3903: $(Q^p)_{-1/2}$. The fibrations $(E^o,\pi^o)$ and $(E^p,\pi^p)$ are flat near

3904: the points $\zeta^o,\zeta^p$, and the one-forms $\kappa_{Q^o},\kappa_{Q^p}$

3905: vanish near those points. This allows one to use the gluing construction

3906: discussed in Section \ref{sec:simple-invariant} and again in Section

3907: \ref{sec:relative-invariants} to produce an exact Lefschetz fibration $(E,\pi)$

3908: over $S = S^o \#_{\zeta^o \sim \zeta^p} S^p$, together with a Lagrangian

3909: boundary condition $Q$; see Figure \ref{fig:h-map-four}. Of course, this is

3910: still modelled over the ends on the same Lagrangian submanifolds as $S^o$. In

3911: what follows, we assume that the parameter $\rho$ in the gluing process has

3912: been chosen sufficiently small.

3913: 

3914: \includefigure{h-map-four}{h-iv.eps}{hb} %

3915: %

3916: \begin{lemma} \label{th:order-homotopy}

3917: There is a complex structure $j$ on $S$, standard over the ends, and a $J \in

3918: \JJ^{reg,h}(E,\pi,Q,j,J^{(1)},J^{(2)},J^{(5)})$, such that

3919: \[

3920: C\Phi_0^{rel}(E,\pi,Q,J): C' \longrightarrow C''

3921: \]

3922: is homotopic to zero by a chain homotopy of order $[-2\pi R(0);\infty)$.

3923: \end{lemma}

3924: 

3925: \proof Let $j^o$ be some complex structure on $S^o$, standard over the ends.

3926: Using Lemmas \ref{th:o-sections} and \ref{th:nonnegative-one} one can find a

3927: $J^o \in \JJ(E^o,\pi^o,j^o,J^{(1)},J^{(2)},J^{(5)})$ which is horizontal and

3928: regular. In fact, regularity can be achieved even while prescribing what $J^o$

3929: is outside $(\pi^o)^{-1}(U^o) \subset E^o$. This is useful because, for our

3930: intended gluing argument, $J^o$ needs to be of a particular form close to the

3931: fibre over $\zeta^o \notin U^o$: namely, in a local trivialization near that

3932: point, it needs to be the product of $j^o$ and some previously fixed almost

3933: complex structure on $M$. We now have evaluation maps, for $\tilde{x}_0 \in

3934: \tau_L(L_0) \cap L$, $x_1 \in L \cap L_1$, $x \in \tau_L(L_0) \cap L_1$,

3935: \begin{equation} \label{eq:ev-zetap}

3936: ev_{\zeta^o}: \MM_{J^o}(\tilde{x}_0,x_1,x) \longrightarrow Q^o_{\zeta^o}.

3937: \end{equation}

3938: Take some complex structure $j^p$ on $S^p$. One can use Lemmas

3939: \ref{th:horizontal-transversality} and \ref{th:p-sections} to find a $J^p \in

3940: \JJ^{reg,h}(E^p,\pi^p,Q^p,j^p)$ with fixed behaviour outside

3941: $(\pi^p)^{-1}(U^p)$; as before, since $\zeta^p \notin U^p$, one can use this to

3942: make $J^p$ suitable for gluing. At the same time, Lemma

3943: \ref{th:horizontal-ev-transversality} allows us to make the evaluation map

3944: $ev_{\zeta^p}: \MM_{J^p} \rightarrow Q^p_{\zeta^p}$ transverse to any given

3945: cycle. We take this cycle to be the disjoint union of \eqref{eq:ev-zetap} for

3946: all $\tilde{x}_0,x_1,x$, identifying $Q^o_{\zeta^o}$ and $Q^p_{\zeta^p}$ via

3947: \eqref{eq:identify}.

3948: 

3949: Let $j$ be the complex structure on $S$ glued together from $j^o,j^p$, and

3950: similarly $J \in \JJ^h(E,\pi,Q,j,J^{(1)},J^{(2)},J^{(5)})$ the almost complex

3951: structure obtained from $J^o$ and $J^p$. As discussed in Section

3952: \ref{sec:relative-invariants}, $J$ will be regular if the gluing parameter has

3953: been chosen sufficiently small. Moreover, the zero-dimensional spaces of

3954: $(j,J)$-holomorphic sections can be described as fibre products of those on

3955: both parts of the gluing, as in \eqref{eq:zero-gluing}. We know that

3956: $\Phi_1(E^p,\pi^p,Q^p)$ is zero (as pointed out in Remark

3957: \ref{th:vanishing}(i), this remains true even if we consider it as a cobordism

3958: class), and using Lemma \ref{th:gluing-vanishing} one concludes that

3959: $C\Phi_0^{rel}(E,\pi,Q,J)$ is homotopic to zero. Inspection of the proof of

3960: that lemma will show that the homotopy constructed there is of order $[-2\pi

3961: R(0);\infty)$. In fact, its coefficients are given by the number of points in

3962: fibre products

3963: \begin{equation} \label{eq:zero-product}

3964: \MM_{J^o}(\tilde{x}_0,x_1,x) \times_{Q^o_{\zeta^o}} G

3965: \end{equation}

3966: where $(G,g)$ is some cycle bounding $(\MM_{J^p},ev_{\zeta^p})$. Using

3967: \eqref{eq:kappao} and Lemma \ref{th:empty} one sees that

3968: \eqref{eq:zero-product} is empty unless $a_{L_0,L_1}(x) \geq

3969: a_{\tau_L(L_0),L}(\tilde{x}_0) + a_{L,L_1}(x_1) - 2\pi R(0)$, which provides

3970: the desired estimate. \qed

3971: 

3972: To put together Lemmas \ref{th:bc} and \ref{th:order-homotopy}, one observes

3973: that there are oriented diffeomorphisms

3974: \[

3975: \xymatrix{

3976:  {E^{bc}} \ar[r]^{\Psi} \ar[d]_{\pi^{bc}} & {E} \ar[d]^{\pi} \\

3977:  {S^{bc}} \ar[r]^{\psi} & {S}

3978: }

3979: \]

3980: with the following properties: on the ends, $\psi$ and $\Psi$ relate the local

3981: trivializations of the two fibrations. Next, $\Psi^*\Omega = \Omega^{bc}$, and

3982: $\Psi(Q^{bc}) = Q$. Finally, $\psi$ is holomorphic near the unique critical

3983: value, with respect to the complex structures defined there which are part of

3984: the structure of exact Lefschetz fibrations of $(E^{bc},\pi^{bc})$ and

3985: $(E,\pi)$; and the same holds for $\Psi$ near the unique critical point. This

3986: is not difficult, since both fibrations contain a copy of $(E^p,\pi^p)$ and are

3987: otherwise flat; comparing Figures \ref{fig:h-map-one} and \ref{fig:h-map-four}

3988: shows how the bases should be identified in order for the monodromies to match.

3989: It follows that one can take $j$ and $J$ in Lemma \ref{th:bc} to be the

3990: pullback of almost complex structures from Lemma \ref{th:order-homotopy}, and

3991: then $C\Phi_0^{rel}(E^{bc},\pi^{bc},Q^{bc},J)$ will be homotopic to zero by a

3992: chain homotopy of order $[-2\pi R(0);\infty)$, with $-2\pi R(0) > 0$. On the

3993: other hand, it is homotopic to $c \circ b$ by a homotopy of order $(0;\infty)$.

3994: Taking the two together, one has a homotopy $h: c \circ b \htp 0$ of order

3995: $(0;\infty)$, thus fulfilling the final requirement of Lemma

3996: \ref{th:low-energy}.

3997: 

3998: %\bibliographystyle{amsplain}

3999: \small

4000: %\bibliographystyle{plain}

4001: %\bibliography{../macros/all,../macros/new}

4002: 

4003: \begin{thebibliography}{10}

4004: 

4005: \bibitem{arnold95}

4006: V.~I. Arnol'd.

4007: \newblock Some remarks on symplectic monodromy of {M}ilnor fibrations.

4008: \newblock In H.~Hofer, C.~Taubes, A.~Weinstein, and E.~Zehnder, editors, {\em

4009:   The {F}loer Memorial Volume}, volume 133 of {\em Progress in Mathematics},

4010:   pages 99--104. Birkh{\"a}user, 1995.

4011: 

4012: \bibitem{aspinwall01}

4013: P.~S. Aspinwall.

4014: \newblock Some navigation rules for {$D$}-brane monodromy.

4015: \newblock Preprint hep-th/0102198.

4016: 

4017: \bibitem{braam-donaldson94}

4018: P.~Braam and S.~K. Donaldson.

4019: \newblock {F}loer's work on instanton homology, knots and surgery.

4020: \newblock In H.~Hofer, C.~Taubes, A.~Weinstein, and E.~Zehnder, editors, {\em

4021:   The {F}loer Memorial Volume}, volume 133 of {\em Progress in Mathematics},

4022:   pages 195--256. Birkh{\"a}user, 1995.

4023: 

4024: \bibitem{carey-marcolli-wang98}

4025: A.~Carey, M.~Marcolli, and B.~Wang.

4026: \newblock Exact triangles in {S}eiberg-{W}itten {F}loer theory. {P}art {I}: the

4027:   geometric triangle.

4028: \newblock Preprint math.DG/9907065.

4029: 

4030: \bibitem{dostoglou-salamon94}

4031: S.~Dostoglou and D.~Salamon.

4032: \newblock Self dual instantons and holomorphic curves.

4033: \newblock {\em Annals of Math.}, 139:581--640, 1994.

4034: 

4035: \bibitem{floer88c}

4036: A.~Floer.

4037: \newblock Morse theory for {L}agrangian intersections.

4038: \newblock {\em J. Differential Geom.} 28:513--547, 1988.

4039: 

4040: \bibitem{frauenfelder01}

4041: U.~Frauenfelder.

4042: \newblock {G}romov convergence of holomorphic discs.

4043: \newblock Diplomarbeit, ETH Z{\"u}rich, 2001.

4044: 

4045: \bibitem{fukaya-oh-ohta-ono}

4046: K.~Fukaya, Y.-G. Oh, H.~Ohta, and K.~Ono.

4047: \newblock Lagrangian intersection {F}loer theory - anomaly and obstruction.

4048: \newblock Preprint, 2000.

4049: 

4050: \bibitem{horja01}

4051: P.~Horja.

4052: \newblock Derived category automorphisms from mirror symmetry.

4053: \newblock Preprint Math.AG/0103231.

4054: 

4055: \bibitem{ivashkovich-shevchishin}

4056: S.~Ivashkovich and V.~Shevchishin.

4057: \newblock Complex curves in almost-complex manifolds and meromorphic hulls.

4058: \newblock {S}chriftenreihe des {G}raduiertenkollegs {G}eometrie und

4059:   mathematische {P}hysik der {U}niversit{\"a}t {B}ochum, {H}eft 36, 1999.

4060: 

4061: \bibitem{khovanov-seidel98}

4062: M.~Khovanov and P.~Seidel.

4063: \newblock Quivers, {F}loer cohomology, and braid group actions.

4064: \newblock Preprint math.QA/0006056.

4065: 

4066: \bibitem{kontsevich94}

4067: M.~Kontsevich.

4068: \newblock Homological algebra of mirror symmetry.

4069: \newblock In {\em Proceedings of the International Congress of Mathematicians

4070:   (Z{\"u}rich, 1994)}, pages 120--139. Birkh{\"a}user, 1995.

4071: 

4072: \bibitem{lion-vergne}

4073: G.~Lion and M.~Vergne.

4074: \newblock {\em The {W}eil representation, {M}aslov index and {T}heta series},

4075:   volume~6 of {\em Progress in Math.}

4076: \newblock Birkh{\"a}user, 1980.

4077: 

4078: \bibitem{liu96}

4079: G.~Liu.

4080: \newblock Associativity of quantum multiplication.

4081: \newblock {\em Commun. Math. Phys.}, 191:265--282, 1998.

4082: 

4083: \bibitem{mcduff-salamon}

4084: D.~McDuff and D.~Salamon.

4085: \newblock {\em {$J$}-holomorphic curves and quantum cohomology}, volume~6 of

4086:   {\em University Lecture Notes Series}.

4087: \newblock Amer. Math. Soc., 1994.

4088: 

4089: \bibitem{oh92}

4090: Y.-G.~Oh.

4091: \newblock Removal of boundary singularities of pseudo-holomorphic

4092: curves with {L}agrangian boundary conditions.

4093: \newblock {\em Commun. Pure Appl. Math.}, 45:121--139, 1992.

4094: 

4095: \bibitem{ohnew}

4096: Y.-G.~Oh.

4097: \newblock Symplectic topology as the geometry of the action functional {I}.

4098: \newblock {\em J. Differential Geom.}, 46:1--55, 1997.

4099: 

4100: \bibitem{ozsvath-szabo01}

4101: P.~Ozsvath and Z.~Szabo.

4102: \newblock Holomorphic disks and topological invariants for rational homology

4103:   three-spheres.

4104: \newblock Preprint math.SG/0101206.

4105: 

4106: \bibitem{piunikhin-salamon-schwarz94}

4107: S.~Piunikhin, D.~Salamon, and M.~Schwarz.

4108: \newblock Symplectic {F}loer-{D}onaldson theory and quantum cohomology.

4109: \newblock In C.~B. Thomas, editor, {\em Contact and symplectic geometry}, pages

4110:   171--200. Cambridge Univ. Press, 1996.

4111: 

4112: \bibitem{robbin-salamon93}

4113: J.~Robbin and D.~Salamon.

4114: \newblock The {M}aslov index for paths.

4115: \newblock {\em Topology}, 32:827--844, 1993.

4116: 

4117: \bibitem{ruan-tian94}

4118: Y.~Ruan and G.~Tian.

4119: \newblock A mathematical theory of {Q}uantum {C}ohomology.

4120: \newblock {\em J. Differential Geom.}, 42:259--367, 1995.

4121: 

4122: \bibitem{rudakov90}

4123: A.~Rudakov et~al.

4124: \newblock {\em Helices and vector bundles: {S}eminaire {R}udakov}, volume 148

4125:   of {\em LMS Lecture Note Series}.

4126: \newblock Cambridge University Press, 1990.

4127: 

4128: \bibitem{salamon00}

4129: D.~Salamon.

4130: \newblock {S}eiberg-{W}itten invariants of mapping tori, symplectic fixed

4131:   points, and {L}efschetz numbers.

4132: \newblock {\em Turkish J. of Maths.}, 23:117--143, 1999.

4133: 

4134: \bibitem{schwarz95}

4135: M.~Schwarz.

4136: \newblock {\em Cohomology operations from {$S^1$}-cobordisms in {F}loer

4137:   homology}.

4138: \newblock PhD thesis, {ETH} {Z}{\"u}rich, 1995.

4139: 

4140: \bibitem{seidel97}

4141: P.~Seidel.

4142: \newblock {\em Floer homology and the symplectic isotopy problem}.

4143: \newblock PhD thesis, Oxford University, 1997.

4144: 

4145: \bibitem{seidel98b}

4146: P.~Seidel.

4147: \newblock Lagrangian two-spheres can be symplectically knotted.

4148: \newblock {\em J. Differential Geom.}, 52:145--171, 1999.

4149: 

4150: \bibitem{seidel99}

4151: P.~Seidel.

4152: \newblock Graded {L}agrangian submanifolds.

4153: \newblock {\em Bull. Soc. Math. France}, 128:103--146, 2000.

4154: 

4155: \bibitem{seidel00}

4156: P.~Seidel.

4157: \newblock Vanishing cycles and mutation.

4158: \newblock In {\em Proceedings of the 3rd European Congress of Mathematics

4159:   (Barcelona, 2000)}. Birkh{\"a}user, to appear.

4160: 

4161: \bibitem{seidel-thomas99}

4162: P.~Seidel and R.~Thomas.

4163: \newblock Braid group actions on derived categories of coherent sheaves.

4164: \newblock Preprint math.AG/0001043. To appear in {\em Duke Math.\ J}.

4165: 

4166: \bibitem{desilva98}

4167: V.~De Silva.

4168: \newblock {\em Products in the symplectic {F}loer homology of {L}agrangian

4169:   intersections}.

4170: \newblock PhD thesis, Oxford University, 1998.

4171: 

4172: \end{thebibliography}

4173: \end{document}

4174: