1: \documentclass[letterpaper,11pt]{amsart}

2: 

3: \usepackage{amssymb}

4: \usepackage{eucal}

5: \usepackage{array}

6: \usepackage{epsfig}

7: \usepackage[all]{xy}

8: \CompileMatrices \numberwithin{equation}{section}

9: \SelectTips{cm}{}

10: 

11: \title{More about vanishing cycles and mutation}

12: \author{Paul Seidel}

13: \date{October 5, 2000.}

14: 

15: %

16: %------------------------ NON-MATH MACROS --------------------------

17: %

18: \renewcommand{\thefootnote}{\fnsymbol{footnote}}

19: 

20: \renewcommand{\subsection}{\vspace{1ex}

21: \hspace{-\parindent}%

22: \stepcounter{subsection}%

23: ({\sc \thesection\alph{subsection}})}

24: 

25: \setlength{\extrarowheight}{4pt}

26: 

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

28: \begin{figure}[#3]

29: \begin{center}

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

31: \end{center}

32: \end{figure}}

33: 

34: \newenvironment{myitemize}%

35: {\begin{itemize}\itemsep0.5em} {\end{itemize}}

36: 

37: %

38: %----------------------- GENERIC MATH MACROS -----------------------

39: %

40: \newcommand{\R}{\mathbb{R}}

41: \newcommand{\Z}{\mathbb{Z}}

42: \newcommand{\C}{\mathbb{C}}

43: \newcommand{\Q}{\mathbb{Q}}

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

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

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

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

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

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

50: \newcommand{\leftsc}{\langle}

51: \newcommand{\rightsc}{\rangle}

52: \newcommand{\Rgeq}{\R^{\scriptscriptstyle \geq 0}}

53: \newcommand{\Rleq}{\R^{\scriptscriptstyle \leq 0}}

54: \newcommand{\suchthat}{\; | \;}

55: 

56: \newcommand{\id}{\mathrm{id}}

57: \newcommand{\ind}{\mathrm{ind}}

58: \newcommand{\re}{\mathrm{re}}

59: \newcommand{\im}{\mathrm{im}}

60: \renewcommand{\ker}{\mathrm{ker}}

61: \newcommand{\coker}{\mathrm{coker}}

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

63: \newcommand{\Hom}{\mathrm{Hom}}

64: \newcommand{\End}{\mathrm{End}}

65: 

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

67: \renewcommand{\o}{\omega}

68: \renewcommand{\O}{\Omega}

69: \newcommand{\Aut}{\mathrm{Aut}}

70: \newcommand{\Ham}{\mathrm{Ham}}

71: \newcommand{\Symp}{\mathrm{Symp}}

72: \newcommand{\Diff}{\mathrm{Diff}}

73: 

74: \theoremstyle{plain}

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

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

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

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

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

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

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

82: \newtheorem{conjecture}[thm]{Conjecture}

83: \newtheorem{question}[thm]{Question}

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

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

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

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

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

89: \newtheorem{examples}[thm]{Examples}

90: 

91: \theoremstyle{remark}

92: \newtheorem*{additional}{Additional references}

93: \newtheorem*{acknowledgments}{Acknowledgments}

94: 

95: %

96: %-------------------- SPECIFIC MATH MACROS, PART I ----------------------

97: %

98: \newcommand{\Sympe}{\Symp^e}

99: \newcommand{\crit}{\mathrm{crit}}

100: \newcommand{\Phirel}{\Phi_{rel}}

101: \newcommand{\A}{\mathcal A}

102: \newcommand{\B}{\mathcal B}

103: \newcommand{\Cone}{\mathrm{Cone}}

104: \newcommand{\tL}{\widetilde{L}}

105: \newcommand{\tphi}{\tilde{\phi}}

106: \newcommand{\ttau}{\tilde{\tau}}

107: \newcommand{\J}{\mathbf{J}}

108: \newcommand{\Lag}{{\mathit{Lag}}^\rightarrow}

109: \renewcommand{\hom}{\mathit{hom}}

110: \newcommand{\Tw}{\mathrm{Tw}}

111: \newcommand{\Ob}{\mathrm{Ob}}

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

113: \newcommand{\RR}{\mathcal{R}}

114: \renewcommand{\SS}{\mathcal{S}}

115: %

116: %------------------ SPECIFIC MATH MACROS, PART II -----------------

117: %

118: \newcommand{\sing}{\mathrm{sing}}

119: \newcommand{\Coh}{\mathrm{Coh}}

120: \newcommand{\Morse}{\mathit{Morse}}

121: \newcommand{\fund}[1]{[\hspace{-0.15em}[ #1 ]\hspace{-0.15em}]}

122: 

123: %

124: %--------------------------- THE PAPER ----------------------------

125: %

126: \begin{document}

127: \maketitle

128: 

129: \section{Introduction}

130: %

131: Not surprisingly, this is a sequel to ``Vanishing cycles and mutation''

132: \cite{seidel00}. Notions from that paper will be used freely later on, but for

133: starters a short recapitulation seems appropriate. Let $D \subset \C$ be the

134: closed unit disc. An {exact Morse fibration} $\pi: E^{2n+2} \rightarrow D$ can

135: be described roughly as a family of $2n$-dimensional exact symplectic manifolds

136: parametrized by $D$, with singularities modelled on Morse-type critical points

137: of holomorphic functions. Such objects are suitable for developing

138: Picard-Lefschetz theory in a symplectic context. Suppose that $(E,\pi)$ comes

139: equipped with a relative Maslov map $\delta_{E/D}$, which is basically a

140: trivialization of $(\Lambda^{n+1}_{\C} TE)^{\otimes 2}$. Then one can associate

141: to it an algebraic invariant, the derived directed Fukaya category

142: $D^b\Lag(\widetilde{\Gamma})$, which encodes the symplectic geometry of the

143: vanishing cycles of $(E,\pi)$. It is constructed in three steps:

144: 

145: \begin{myitemize}

146: \item[(i)]

147: Fix a base point $z_0 \in \partial D$, for definiteness $z_0 = -i$; the fibre

148: $M = E_{z_0}$ is an exact symplectic manifold. Then make an admissible choice

149: of paths \cite[Figure 2]{seidel00} from $z_0$ to the critical values of $\pi$.

150: This determines a {distinguished basis of vanishing cycles} in $M$, which is an

151: ordered family $\Gamma = (L_1,\dots,L_m)$ of embedded exact Lagrangian spheres

152: $L_i \subset M$ (there is one more bit of information, so-called framings of

153: the $L_i$, which we ignore for the moment). We will sometimes call such

154: families Lagrangian configurations in $M$. Any other distinguished basis,

155: arising from a different choice of paths, can be obtained from $\Gamma$ through

156: a sequence of Hurwitz moves.

157: 

158: \item[(ii)]

159: $\delta_{E/D}$ induces a Maslov map $\delta_M$ on $M$, thus allowing one to

160: speak of graded Lagrangian submanifolds in $M$. By choosing gradings in an

161: arbitrary way, one lifts $\Gamma$ to a graded Lagrangian configuration

162: $\widetilde{\Gamma} = (\tL_1,\dots,\tL_m)$. Next one introduces the {directed

163: Fukaya category} $\A = \Lag(\widetilde{\Gamma})$, which is an

164: $A_\infty$-category with objects $\tL_1,\dots,\tL_m$. The spaces of morphisms

165: in $\A$ are essentially given by the natural cochain complexes underlying the

166: Floer cohomology groups $HF^*(\tL_i,\tL_k)$, but the ordering of the objects

167: also plays a role in the definition, which is what the word ``directed'' refers

168: to.

169: 

170: \item[(iii)]

171: The {derived category} $D^b(\A)$ of $\A$ is no longer an $A_\infty$-category

172: but rather a category in the ordinary sense; in fact it is a triangulated

173: category, linear over $\Z/2$ since we are using $\Z/2$-coefficients. Its

174: objects are twisted complexes in $\A$, which are $A_\infty$-analogues of chain

175: complexes, and morphisms are chain homotopy classes of maps in a suitable

176: sense. Unlike $\A$ itself, $D^b(\A)$ is independent of the choice of $\Gamma$

177: (and of $\widetilde{\Gamma}$), hence is an invariant of $(E,\pi)$ and

178: $\delta_{E/D}$. The reason for this is a relation between Hurwitz moves of

179: $\Gamma$ and the purely algebraic notion of mutations of $\A$.

180: \end{myitemize}

181: %

182: All this is explained in more detail, although still without proofs, in

183: \cite{seidel00}. The present paper adds computations and some more far-reaching

184: conjectures. It is divided into four more or less independent parts. The first

185: part, formed by Sections \ref{sec:dimension-zero}--\ref{sec:examples},

186: introduces some classes of examples. Section \ref{sec:dimension-zero} concerns

187: the toy model case when the fibre dimension is zero. Section \ref{sec:examples}

188: explains how to construct exact Morse fibrations from algebro-geometric

189: objects, namely Lefschetz pencils or isolated hypersurface singularities. Also

190: included is a discussion of the mirror manifold of $\CP{2}$.

191: 

192: The subject of the second part, spanning Sections

193: \ref{sec:tqft2}--\ref{sec:donaldson}, is Hochschild cohomology. The basic

194: observation is that like $D^b(\A)$, the Hochschild cohomology $HH^*(\A,\A)$ is

195: an invariant of $(E,\pi)$ and $\delta_{E/D}$. One would expect this to have a

196: more geometric meaning. Following a suggestion of Donaldson we propose a

197: conjecture in this direction, which relates Hochschild cohomology to the global

198: monodromy of the fibration. As a preliminary to this, Section \ref{sec:tqft2}

199: enlarges the topological quantum field theory framework from \cite[Section

200: 3]{seidel00} to include the Floer cohomology of symplectic automorphisms.

201: 

202: The third part, Sections \ref{sec:morse}--\ref{sec:real}, concerns exact Morse

203: fibrations with a real involution. In that situation, and under certain

204: additional assumptions, the category $D^b(\A)$ should have a description in

205: terms of Morse theory on the real part. We state a precise conjecture, which is

206: a theorem at least in the lowest nontrivial dimension $n = 1$.

207: 

208: Finally, Section \ref{sec:induction} is a reflection on the aspect of

209: dimensional reduction inherent in the theory. By this we mean that $\A$, which

210: is defined in terms of pseudo-holomorphic curves in the $2n$-dimensional

211: manifold $M$, produces invariants of the $(2n+2)$-dimensional object $(E,\pi)$.

212: In an attempt to push this further, we give a conjectural algorithm for

213: determining the Floer cohomology of certain Lagrangian spheres in $E$ through

214: computations in $D^b(\A)$. This, if correct, might be the first true

215: application of homological algebra methods in symplectic geometry.

216: 

217: \begin{acknowledgments}

218: This paper owes much to Donaldson; several of the ideas presented here arose

219: during conversations with him. Givental and Kontsevich provided helpful

220: explanations concerning mirror symmetry.

221: \end{acknowledgments}

222: 

223: \newpage

224: 

225: \section{Branched covers\label{sec:dimension-zero}}

226: %

227: \subsection{}

228: In \cite{seidel00} it was tacitly assumed that all symplectic manifolds were of

229: dimension $>0$ (and that assumption will be resumed as soon as this section is

230: finished). The zero-dimensional case is rather trivial, but it still requires

231: some minor adaptations, which we now explain. Let's start with a basic

232: dictionary of zero-dimensional symplectic geometry:

233: 

234: \begin{center}

235: \begin{tabular}{l||l}

236:  compact symplectic manifold & finite set $M$ \\ \hline

237:  Lagrangian sphere & subset $L \subset M$, $|L| = 2$

238:  \\ \hline

239:  Floer cohomology & $HF(L_1,L_2) = (\Z/2)^{L_1 \cap L_2}$

240:  \\ \hline

241:  Dehn twist & \begin{minipage}{16em} \vspace{2pt}

242:  \raggedright

243:  $\tau_L: M \rightarrow M$ swapping the two

244:  points of $L$, leaving the rest fixed \vspace{2pt}

245:  \end{minipage}

246:  \\ \hline

247:  exact Morse fibration & generic branched cover $\pi: E \rightarrow S$

248: \end{tabular}

249: \end{center}

250: Here a {\bf generic branched cover} is a proper map $\pi: E \rightarrow S$

251: between oriented surfaces with boundary, with $\pi^{-1}(\partial S) = \partial

252: E$, having finitely many critical points, which are locally modelled on $\pi(z)

253: = z^2$; and no two critical points may lie in the same fibre.

254: 

255: The TQFT framework deserves a brief mention, if only for its simplicity. Let

256: $S$ be a compact oriented surface with boundary, $\Sigma \subset \partial S$ a

257: finite set of marked points, and $(E,\pi)$ a generic branched cover over $S^* =

258: S \setminus \Sigma$. Suppose moreover that we have a submanifold $Q \subset

259: \pi^{-1}(\partial S^*)$ such that $\pi|Q: Q \rightarrow \partial S^*$ is a

260: double cover; this corresponds to what in \cite{seidel00} was called a

261: Lagrangian boundary condition for $(E,\pi)$. The behaviour of $Q$ near a point

262: $\zeta \in \Sigma$ can be described by a pair $(L_{\zeta,+},L_{\zeta,-})$ of

263: Lagrangian zero-spheres in some fibre of $E$ close to $\zeta$. A continuous

264: section $u: S^* \rightarrow E$ of $\pi$ which satisfies $u(\partial S^*)

265: \subset Q$ singles out a point of $L_{\zeta,+} \cap L_{\zeta,-}$ for each

266: $\zeta$, and hence determines an element of $\bigotimes_\zeta HF(L_{\zeta,+},

267: L_{\zeta,-})$. Taking the sum over all sections defines an invariant

268: \begin{equation} \label{eq:rel}

269: \Phirel(E,\pi,Q) \in \bigotimes_{\zeta \in \Sigma} HF(L_{\zeta,+},L_{\zeta,-}).

270: \end{equation}

271: For trivial reasons, these invariants satisfy the gluing law formulated in

272: \cite[Section 3]{seidel00}. One can use them to find the correct

273: zero-dimensional analogues of various familiar maps in Floer theory. For

274: instance, the pair-of-pants product $HF(L_2,L_3) \otimes HF(L_1,L_2)

275: \rightarrow HF(L_1,L_3)$ is derived from an invariant \eqref{eq:rel} with $S =

276: D$, $\Sigma = \{\text{\em three points}\}$ and $E = S^* \times M$, together

277: with a boundary condition given by $L_1,L_2,L_3$ over the three connected

278: components of $\partial S^*$. Working this out explicitly shows that the

279: product takes a basis element $\bar{x} \otimes \bar{y} \in HF(L_2,L_3) \otimes

280: HF(L_1,L_2)$, $x \in L_2 \cap L_3$ and $y \in L_1 \cap L_2$, to $\bar{x} \in

281: HF(L_1,L_3)$ if $x = y$, and to zero otherwise. There is the same long exact

282: sequence in Floer cohomology as in \cite[Theorem 3.3]{seidel00} and again, the

283: more fundamental fact underlying it is a quasi-isomorphism

284: \[

285: \Cone(a: HF(L,L_2) \otimes HF(L_1,L) \rightarrow HF(L_1,L_2))

286: \xrightarrow{(h,b)} HF(L_1,\tau_L(L_2)).

287: \]

288: However, while $a$ and $b$ are as in higher dimensions, $h$ is the product

289: $HF(L,L_2) \otimes HF(L_1,L) \iso HF(L,\tau_L(L_2)) \otimes HF(L_1,L)

290: \rightarrow HF(L_1,\tau_L(L_2))$. Of course, both the result and proof are now

291: elementary!

292: 

293: The next topic are gradings. We summarize their zero-dimensional theory in

294: another dictionary:

295: 

296: \begin{center}

297: \begin{tabular}{l||l}

298:  \begin{minipage}{9em} \vspace{2pt}

299:  Maslov map on a \newline symplectic manifold \vspace{2pt}

300:  \end{minipage}

301:  & $\delta_M: M \rightarrow S^1$

302:  \\ \hline

303:  \begin{minipage}{9em} \vspace{2pt} \raggedright

304:  graded Lagrangian sphere \vspace{2pt}

305:  \end{minipage}

306:  &

307:  \begin{minipage}{18em} \vspace{2pt} \raggedright

308:  $L \subset M$ with a map $\tL: L \rightarrow \R$ such

309:  that $\exp(2\pi i \tL) = \delta_M|L$ \vspace{2pt}

310:  \end{minipage}

311:  \\ \hline

312:  \begin{minipage}{9em} \vspace{2pt} \raggedright

313:  graded Floer cohomology \vspace{2pt}

314:  \end{minipage}

315:  &

316:  \begin{minipage}{18em} \vspace{2pt} \raggedright

317:  in $HF^*(\tL_1,\tL_2)$, the degree of $\bar{x}$ is

318:  $\tL_2(x) - \tL_1(x)$

319:  \vspace{2pt}

320:  \end{minipage}

321:  \\ \hline

322:  shifting the grading &

323:  $\tL[\sigma] = \tL - \sigma$, $\sigma \in \Z$

324:  \\ \hline

325:  \begin{minipage}{9em} \vspace{2pt} \raggedright

326:  graded symplectic automorphism \vspace{2pt}

327:  \end{minipage}

328:  &

329:  \begin{minipage}{18em} \vspace{2pt}

330:  $\phi: M \rightarrow M$ with a map $\tphi: M \rightarrow \R$ such that

331:  $\exp(2\pi i \phi(x)) = \delta_M(x)/\delta_M(\phi^{-1}(x))$

332:  \vspace{2pt}

333:  \end{minipage}

334:  \\ \hline

335:  relative Maslov map &

336:  \begin{minipage}{18em} \vspace{2pt} \raggedright

337:  on $\pi: E \rightarrow S$, a trivialization

338:  $\delta_{E/S}$ of $TE^{\otimes 2} \otimes (\pi^*TS)^{\otimes -2}$

339:  \end{minipage}

340: \end{tabular}

341: \end{center}

342: %

343: The action of a graded symplectic automorphism on graded Lagrangian

344: submanifolds is defined by $(\tphi \tL)(x) = \tL(\phi^{-1}(x)) + \tphi(x)$. A

345: grading $\tL$ determines a preferred grading $\ttau_{\tL}$ of the associated

346: Dehn twist, characterized by

347: \[

348: \ttau_{\tL}(\tL) = \tL[1] \quad \text{ and } \quad \ttau_{\tL}(x) = 0 \text{

349: for } x \notin L;

350: \]

351: compare \cite[Equation (4)]{seidel00} for the first property. Graded Lagrangian

352: configurations $\widetilde{\Gamma} = (\tL_1,\dots,\tL_m)$ are defined as one

353: would expect. Their Hurwitz moves are

354: \begin{itemize}

355: \item

356: $\widetilde{\Gamma} \rightsquigarrow (\tL_1[\sigma_1],\dots,\tL_m[\sigma_m])$

357: for $\sigma_1,\dots,\sigma_m \in \Z$;

358: \item

359: $\widetilde{\Gamma} \rightsquigarrow c\widetilde{\Gamma} =

360: (\ttau_{\tL_1}(\tL_2),\dots,\ttau_{\tL_1}(\tL_m),\tL_1)$;

361: \item

362: $\widetilde{\Gamma} \rightsquigarrow r\widetilde{\Gamma} =

363: (\tL_1,\dots,\tL_{m-2},\ttau_{\tL_{m-1}}(\tL_m),\tL_{m-1})$.

364: \end{itemize}

365: Let $(E,\pi)$ be a generic branched cover over $D$, with a relative Maslov map

366: $\delta_{E/D}$. A drastically simplified version of the usual Picard-Lefschetz

367: argument produces from it a graded (with respect to the induced Maslov map

368: $\delta_M$) Lagrangian configuration in $M = E_{z_0}$, unique up to Hurwitz

369: equivalence.

370: 

371: \begin{remark}

372: One can observe here two minor differences with respect to the

373: positive-dimensional situation. Firstly, $\ttau_{\tL}$ depends on $\tL$;

374: secondly, the Hurwitz equivalence class of a graded configuration contains

375: information that is lost if one forgets the gradings. Both phenomena go back to

376: the obvious fact that there is a $\Z^2$ ambiguity in the choice of grading for

377: a Lagrangian zero-sphere, as opposed to $\Z$ in higher dimensions.

378: \end{remark}

379: 

380: The directed Fukaya category $\A = \Lag(\widetilde{\Gamma})$ attached to a

381: zero-dimensi\-onal graded configuration has objects $\tL_i$. The morphisms

382: $hom_\A(\tL_i,\tL_k)$ are $HF^*(\tL_i,\tL_k)$ for $i<k$, $\Z/2 \cdot

383: \id_{\tL_i}$ for $i = k$, and zero for $i>k$. The nontrivial compositions

384: $HF^*(\tL_j,\tL_k) \otimes HF^*(\tL_i,\tL_j) \rightarrow HF^*(\tL_i,\tL_k)$,

385: $i<j<k$, are given by the pair-of-pants product. Even though $\A$ is an

386: ordinary $\Z/2$-linear and $\Z$-graded category, we prefer to regard it as a

387: directed $A_\infty$-category where all composition maps $\mu_\A^d$ of order $d

388: \neq 2$ vanish. The details of the proof are slightly different than in the

389: higher-dimensional case, but the main result of \cite{seidel00} remains true,

390: which is that Hurwitz moves of $\widetilde{\Gamma}$ give rise to mutations of

391: $\A$; and one arrives at the familiar conclusion that

392: $D^b\Lag(\widetilde{\Gamma})$ is an invariant of $(E,\pi)$, $\delta_{E/D}$.

393: 

394: Before going on to concrete examples, we need to recall some algebraic

395: terminology. Let $T$ be a quiver (an oriented graph). The path category

396: ${\mathcal P}T$ has one object for each vertex, and the space of morphisms

397: between two objects is the $\Z/2$-vector space freely generated by all paths in

398: $T$ going from one vertex to the other. One can further divide the morphism

399: spaces by some two-sided ideal, in order to kill certain designated morphisms.

400: This is usually referred to as describing a category by a ``quiver with

401: relations''. If the quiver is directed, which means that its vertices are

402: numbered $1,\dots,m$ such that there are no paths from the $i$-th one to the

403: $k$-th one unless $i<k$, ${\mathcal P}T$ and its quotients can again be seen as

404: special cases of directed $A_\infty$-categories, with morphisms only in degree

405: zero and vanishing composition maps of order $\neq 2$. Note that in this

406: situation our notion of derived category (defined through twisted complexes)

407: agrees with the classical one (defined by considering the categories as

408: algebras, and taking chain complexes of right modules over them).

409: 

410: \subsection{}

411: Take the following quiver with $m$ vertices, with the arrows oriented in an

412: arbitrary way:

413: \begin{equation} \label{eq:bgp}

414: \xymatrix{

415:  {\bullet} \ar[r] &

416:  {\bullet} & \ar[l]

417:  {\bullet} \ar[r] &

418:  {\dots} &

419:  {\bullet} \ar[l]

420: }

421: \end{equation}

422: and let $\A$ be its path category. It is a classical result that $D^b(\A)$ is

423: independent of the orientation of the arrows; what we will do is to explain

424: this geometrically. First of all, one can find a configuration of Lagrangian

425: zero-spheres $\Gamma = (L_1,\dots,L_m)$ in $M = \{1,\dots,m+1\}$ which, when

426: lifted to a graded configuration in the trivial way ($\delta_M = 0$ and $\tL_i

427: = 0$), has $\A$ as its directed Fukaya category. The way to do that is best

428: explained by an example:

429: \begin{equation} \label{eq:makebasis}

430: \begin{split}

431: & \xymatrix@R=-0.25em{

432:  \bullet \ar[r] &

433:  \bullet \ar[r] &

434:  \bullet &

435:  \bullet \ar[l] &

436:  \bullet \ar[l] &

437:  \bullet \ar[l] \ar[r] &

438:  \bullet

439:  \\

440:  {\hspace{-1em} \{1,2\} \hspace{-1em}} &

441:  {\hspace{-1em} \{1,3\} \hspace{-1em}} &

442:  {\hspace{-1em} \{1,4\} \hspace{-1em}} &

443:  {\hspace{-1em} \{5,4\} \hspace{-1em}} &

444:  {\hspace{-1em} \{6,4\} \hspace{-1em}} &

445:  {\hspace{-1em} \{7,4\} \hspace{-1em}} &

446:  {\hspace{-1em} \{7,8\} \hspace{-1em}}

447: }

448: \\[0.5em]

449: \Gamma & = (\{1,2\},\{1,3\},\{7,4\},\{6,4\},\{5,4\},\{1,4\},\{7,8\}).

450: \end{split}

451: \end{equation}

452: Next one constructs an $(m+1)$-fold generic branched cover $\pi: E \rightarrow

453: D$ such that $\Gamma$ is one of its distinguished bases of vanishing cycles.

454: Inspection of \eqref{eq:makebasis} and an Euler characteristic computation show

455: that the total space $E$ is connected and in fact a disc. It is a consequence

456: of the classification of branched covers that there is just one such $(E,\pi)$

457: up to isomorphism, and moreover it admits precisely one homotopy class of

458: relative Maslov maps. Therefore all categories $D^b(\A)$ arise from the same

459: geometric situation, hence must be equivalent on grounds of the general theory.

460: 

461: \subsection{}

462: Let $\pi: E \rightarrow D$ be a double cover branched along $2g+1 \geq 5$

463: points, so that the total space is a genus $g \geq 2$ surface with one boundary

464: component. Write $\iota$ for the nontrivial covering transformation, which is a

465: hyperelliptic involution of $E$. There is a unique homotopy class of relative

466: Maslov maps which are invariant under $\iota$; take $\delta_{E/D}$ in that

467: class. All vanishing cycles of $(E,\pi)$ are the same, and if we choose their

468: gradings in the most obvious way, the resulting directed Fukaya category $\A_g$

469: can be described by the quiver with relations

470: \[

471: \xymatrix{

472:  {\bullet} \ar@/^1pc/[r]^{a_1} \ar@/_1pc/[r]_{b_1} &

473:  {\bullet} \ar@/^1pc/[r]!<-1em,0em>^{a_2} \ar@/_1pc/[r]!<-1em,0em>_{b_2} &

474:  {\bullet\; \dots\; \bullet}

475:  \ar@/^1pc/[]!<1em,0em>;[r]^{a_{2g}}

476:  \ar@/_1pc/[]!<1em,0em>;[r]_{b_{2g}} &

477:  {\bullet}

478: } \qquad \qquad

479: \begin{cases}

480:  b_{i+1}a_i = 0, & \\

481:  a_{i+1}b_i = 0.

482: \end{cases}

483: \]

484: We will now make some further remarks concerning $\A_g$ and its derived

485: category. These may seem a bit unmotivated, but they will be at least partially

486: put into context later on; see Remark \ref{th:zero-spheres}. The following

487: definition is taken from \cite{seidel-thomas99}:

488: 

489: \begin{defn} \label{def:spherical}

490: Let ${\mathcal C}$ be a triangulated category, linear over $\Z/2$ and such that

491: the spaces $\Hom^*_{\mathcal C}(-,-)$ are finite-dimensional. $C \in

492: \Ob\,{\mathcal C}$ is called spherical of dimension $(n+1)$ if

493: $\Hom^*_{\mathcal C}(C,C) \iso H^*(S^{n+1};\Z/2)$ and the composition

494: \[

495: \Hom^*_{\mathcal C}(X,C) \otimes \Hom^{n+1-*}_{\mathcal C}(C,X) \rightarrow

496: \Hom^{n+1}_{\mathcal C}(C,C) \iso \Z/2

497: \]

498: is a nondegenerate pairing for any $X \in \Ob\,{\mathcal C}$.

499: \end{defn}

500: 

501: Under some additional assumptions on ${\mathcal C}$, which are satisfied for

502: derived categories of directed $A_\infty$-categories, one can associate to any

503: spherical object $C$ an exact self-equivalence of ${\mathcal C}$, the twist

504: functor $T_C$, which is well-defined up to isomorphism. If $C_1,C_2$ are

505: spherical with $\Hom^*_{\mathcal C}(C_1,C_2) = 0$ then their twist functors

506: commute; and if $\Hom^*_{\mathcal C}(C_1,C_2)$ is one-dimensional, one gets a

507: braid relation $T_{C_1}T_{C_2}T_{C_1} \iso T_{C_2}T_{C_1}T_{C_2}$.

508: 

509: Denote the objects of $\A_g$ by $(X^1,\dots,X^{2g+1})$. The twisted complexes

510: $C_i = (X^i[1] \oplus X^{i+1}, \left(\begin{smallmatrix} 0 & 0 \\ a_i+b_i & 0

511: \end{smallmatrix}\right) )$, $1 \leq i \leq 2g$, are spherical objects of

512: dimension one in $D^b(\A_g)$, and moreover

513: \[

514:  \dim\, \Hom^*_{D^b(\A_g)}(C_i,C_k) =

515:  \begin{cases}

516:  0 & |i-k| \geq 2, \\

517:  1 & |i-k| = 1.

518:  \end{cases}

519: \]

520: This implies that $T_{C_1},\dots,T_{C_{2g}}$ generate an action (in a weak

521: sense) of the braid group $B_{2g+1}$ on $D^b(\A_g)$. Now let $D^{per}(\A_g)$ be

522: the category defined like $D^b(\A_g)$ but using $\Z/2$-graded twisted

523: complexes. This is triangulated, and its two-fold shift functor $[2]$ is

524: isomorphic to the identity; we call it the periodic derived category of $\A_g$.

525: One can adapt the theory of spherical objects and twist functors to such

526: categories. In our case, this means that there are $C_i \in D^{per}(\A_g)$

527: defined as before, with the same properties. In addition there is now another

528: spherical object

529: \[

530:  C_0 = (X^1 \oplus X^2[1] \oplus X^3 \oplus X^4[1],

531:  \mbox{\smaller $\begin{pmatrix}

532:  0         & 0   & 0   & 0 \\

533:  a_1       & 0   & 0   & 0 \\

534:  0         & b_2 & 0   & 0 \\

535:  b_3b_2b_1 & 0   & a_3 & 0

536:  \end{pmatrix}$} )

537: \]

538: The dimension of $\Hom^*_{D^{per}(\A_g)}(C_0,C_i)$ is one if $i = 4$ and zero

539: for all other $i>0$. The diagram representing this situation,

540: \[

541: \xymatrix{

542:   & & & {C_0} \ar@{-}[d] \\

543:   {C_1} \ar@{-}[r] &

544:   {C_2} \ar@{-}[r] &

545:   {C_3} \ar@{-}[r] &

546:   {C_4} \ar@{-}[r] &

547:   {C_5} \ar@{-}[r] &

548:   \dots \ar@{-}[r] &

549:   {C_{2g}}

550: }

551: \]

552: has the same structure as the configuration of curves in Wajnryb's presentation

553: of $R_{g,1}$, the mapping class group of a genus $g$ surface with one marked

554: point \cite{wajnryb83}; and one can ask

555: 

556: \begin{question}

557: Let ${\mathcal C} \subset D^{per}(\A_g)$ be the full triangulated subcategory

558: generated by the $C_i$. Do $T_{C_0},\dots,T_{C_{2g}}$ generate an action of

559: $R_{g,1}$ on ${\mathcal C}$?

560: \end{question}

561: 

562: \begin{additional}

563: The classical paper about the quiver \eqref{eq:bgp} is by Bernstein, Gelfand,

564: and Ponomarev \cite{bernstein-gelfand-ponomarev72}. The category $D^b(\A_g)$ is

565: closly related to those studied in \cite{khovanov-seidel98}, even though the

566: quiver presentation looks different. Those papers do not consider periodic

567: derived categories and therefore miss out on the additional object $C_0$.

568: \end{additional}

569: 

570: \section{Examples from algebraic geometry\label{sec:examples}}

571: %

572: \subsection{}

573: Let $X$ be a smooth projective variety, $\xi \rightarrow X$ an ample line

574: bundle, and $\sigma_0,\sigma_1$ two holomorphic sections of $\xi$ which

575: generate a Lefschetz pencil of hypersurfaces $Y_z = \{ x \in X \suchthat

576: \sigma_0(x)/\sigma_1(x) = z\}$, $z \in \CP{1} = \C \cup \{\infty\}$. Suppose,

577: in addition to the Lefschetz condition, that $Y_\infty$ is smooth. What we want

578: to look at is, in principle, the holomorphic Morse function

579: \[

580: \sigma_0/\sigma_1 : X \setminus Y_\infty \longrightarrow \C.

581: \]

582: To obtain additional symplectic data, one chooses a metric on $\xi$, with

583: corresponding connection $A$, such that $\o_X = (i/2\pi)F_A$ is a K{\"a}hler

584: form. By a standard construction, $A$ and $\sigma_1$ give rise to a one-form

585: $\theta_X$ defined on $X \setminus Y_\infty$, such that $d\theta_X = \o_X$.

586: This still doesn't fit the definition of an exact Morse fibration, making some

587: further modifications necessary.

588: 

589: Let $Z = Y_0 \cap Y_\infty$ be the base locus of the pencil, $G = \{ (z,x) \in

590: \CP{1} \times X \suchthat \sigma_0(x)/\sigma_1(x) = z\}$ its graph, and $p: G

591: \rightarrow \CP{1}$ the projection. $(G,p)$ is a Lefschetz fibration with

592: fibres $Y_z$, and contains a trivial subfibration $\CP{1} \times Z \subset G$.

593: Let $\O_G \in \Omega^2(G)$ be the pullback of $\o_X$; it is closed,

594: nondegenerate on the vertical tangent spaces $\ker(Dp) \subset TG$, and hence

595: defines a symplectic connection away from the critical points of $p$. Moreover,

596: the restriction of this connection to $\CP{1} \times Z \subset G$ is trivial.

597: Similarly, pulling back $\theta_X$ gives a one-form $\Theta_G$ defined on the

598: complement of $(\{\infty\} \times Y_\infty) \cup (\CP{1} \times Z) \subset G$,

599: with $d\Theta_G = \Omega_G$.

600: 

601: After possibly rescaling $\sigma_0$, we may assume that $Y_z$ is smooth for

602: $|z| \geq 1$. Let $U \subset Y_0$ be a small open neighbourhood of $Z$. Using

603: the symplectic parallel transport in radial directions one constructs an

604: embedding $\Xi$,

605: \[

606: \xymatrix{

607:  {D \times U} \ar[rr]^-{\Xi} \ar[dr] &&

608:  {p^{-1}(D) \subset G} \ar[dl]^p \\

609:  & {D} &

610: }

611: \]

612: with $\Xi\, |\, D \times Z$ and $\Xi\, |\,\{0\} \times U$ the obvious

613: inclusions, which has the property that $\Xi^*\Omega_G - \o_X|U$ vanishes on

614: each diameter $[-z;z] \times U$, $z \in S^1$. Let $h \in \smooth_c(U,[0;1])$ be

615: a function which is $\equiv 1$ on some smaller neighbourhood $U' \subset U$ of

616: $Z$. Take the map $H: p^{-1}(D) \rightarrow p^{-1}(D)$ which is the identity

617: outside $\im(\Xi)$ and satisfies $H\,\Xi(z,x) = \Xi((1-h(x))z,x)$.

618: 

619: Set $E = p^{-1}(D) \setminus \Xi(D \times U'')$ for an even smaller open

620: neighbourhood $U'' \subset U'$ of $Z$, which we choose such that $\partial U''$

621: is smooth. Together with $\pi = p|E$, $\Omega = (H^*\Omega_G|E) \in

622: \Omega^2(E)$, $\Theta = (H^*\Theta_G|E) \in \Omega^1(E)$, and the given complex

623: structure near the critical points of $\pi$, this forms an exact Morse

624: fibration over $D$. Indeed, the required trivialization near $\partial_hE$ is

625: given by $\Xi$ itself; and the fact that $H$ contracts along the same radial

626: directions which are used to define $\Xi$ ensures that $\Omega$ equals

627: $\Omega_G$ on $\ker(D\pi) \subset TE$, hence is nondegenerate there.

628: 

629: Suppose now that the canonical bundle $K_X$ satisfies

630: \begin{equation} \label{eq:relative-maslov}

631: K_X^{-2} \iso \xi^a \, \text{ holomorphically, for some $a \in \Z$.}

632: \end{equation}

633: Then $\sigma_1^a$ gives a trivialization of $K_X^{-2}$ over $X \setminus

634: Y_\infty$, determined up to multiplication with a constant. Since the

635: projection $G \rightarrow X$ identifies $E$ with a compact subset of $X

636: \setminus Y_\infty$, and $\O$ agrees with the pullback of $\o_X$ except in a

637: small neighbourhood of $\partial_hE$, one gets, at least up to homotopy, a

638: preferred relative Maslov map $\delta_{E/D}$. The category

639: $D^b\Lag(\widetilde{\Gamma})$ associated to $(E,\pi)$ and $\delta_{E/D}$

640: actually depends only on $X$ and on the Lefschetz pencil, and not on the other

641: choices made during the construction of $E$. That is because different choices

642: lead to exact Morse fibrations which are deformation equivalent in a suitable

643: sense. For the same reason, if $\xi$ is such that any two generic sections

644: define a Lefschetz pencil, the category is an invariant of the pair $(X,\xi)$.

645: %

646: \includefigure{degree2}{degree2.eps}{hb}%

647: 

648: As a concrete example take $X = \CP{2}$, $\xi = {\mathcal O}(2)$. The fibre $M

649: = E_{z_0}$ of $(E,\pi)$ is, topologically, $\CP{1}$ with four small discs

650: removed. The relative Maslov map $\delta_{E/D}$ coming from

651: \eqref{eq:relative-maslov} induces a Maslov map $\delta_M$ on $M$, which is

652: characterized by having the same behaviour over each component of $\partial M$.

653: Figure \ref{fig:degree2} shows a distinguished basis $\Gamma = (L_1,L_2,L_3)$

654: of vanishing cycles. One can read off directly that the directed Fukaya

655: category $\A = \Lag(\widetilde{\Gamma})$, for a suitable choice of gradings, is

656: described by the quiver with relations

657: \begin{equation} \label{eq:cp2-quiver}

658: \xymatrix{

659:  {\bullet} \ar@/^1pc/[rr]^{a_1} \ar@/_1pc/[rr]_{b_1} &&

660:  {\bullet} \ar@/^1pc/[rr]^{a_2} \ar@/_1pc/[rr]_{b_2} &&

661:  {\bullet}

662: } \qquad

663: \begin{cases}

664:  a_2a_1 = b_2b_1, & \\

665:  b_2a_1 = a_2b_1. &

666: \end{cases}

667: \end{equation}

668: Namely, the intersections $L_i \cap L_k$, $i<k$, consist of two points which

669: are essential, in the sense that they cannot be removed by an isotopy in $M$.

670: This implies that $hom_\A(\tL_i,\tL_k)$ is two-dimensional and $\mu^1_\A = 0$.

671: And there are four triangles in $M$ whose sides (in positive order) map to

672: $(L_1,L_2,L_3)$, giving rise to the nonzero products $\mu^2_\A(a_2,a_1)$,

673: $\mu^2_\A(b_2,a_1)$, $\mu_\A^2(a_2,b_1)$, $\mu^2_\A(b_2,b_1)$.

674: 

675: \begin{remark}

676: Suppose that \eqref{eq:relative-maslov} holds but with $a \in \Q$, which is to

677: say $K_X^{-2c} \iso \xi^b$ for coprime $b,c$. Then $\sigma_1^b$ defines a

678: $c$-sheeted cyclic covering of $X \setminus Y_\infty$, hence also a covering of

679: $E$. The total space of the latter can be made into an exact Morse fibration

680: which then has a canonical relative Maslov map, so that our theory can be

681: applied to it. The $\Z/c$-action on $E$ by covering transformations induces a

682: self-equivalence $\sigma$ of $D^b\Lag(\widetilde{\Gamma})$ which satisfies

683: $\sigma^c \iso [1]$. One can view this as a category with ``fractional

684: gradings'', thinking of $\Hom(X,\sigma^iY)$ as ``the group $\Hom^{i/c}(X,Y)$ of

685: morphisms of degree $i/c$''.

686: \end{remark}

687: 

688: \subsection{}

689: We need to extend the previous discussion to a degenerate situation. Let $X

690: \subset \CP{N}$ be a projective variety with singular set $X^\sing$, and

691: $\sigma_0,\sigma_1 \in (\C^{N+1})^*$ two linear forms which generate a pencil

692: $(Y_z)$ of hyperplane sections of $X$; so the line bundle we are looking at is

693: $\xi = {\mathcal O}_X(1)$. Suppose that the following conditions are satisfied:

694: $X^\sing \subset Y_\infty \setminus Y_0$; the base locus $Z = Y_0 \cap

695: Y_\infty$ is smooth; and $(Y_z)_{z \neq \infty}$ satisfies the same

696: nondegeneracy conditions as a Lefschetz pencil. Then the associated graph $p: G

697: \rightarrow \CP{1}$ is well-behaved except over $\infty \in \CP{1}$. The

698: construction made above, in which the fibre at infinity plays no role, goes

699: through without any changes, yielding an exact Morse fibration $(E,\pi)$ over

700: $D$. If we suppose moreover that $X$ is locally a complete intersection, so

701: that there is a well-defined canonical bundle $K_X$, then condition

702: \eqref{eq:relative-maslov} again ensures the existence of preferred relative

703: Maslov maps $\delta_{E/D}$.

704: %

705: \includefigure{mirrorp2}{mirrorp2.eps}{hb}%

706: 

707: The example which we have in mind is $X = \{ x \in \CP{3} \suchthat x_0^3 =

708: x_1x_2x_3\}$, $\sigma_0(x) = x_1+x_2+x_3$, $\sigma_1(x) = x_0$. One can

709: identify $X \setminus Y_\infty$ with the affine hypersurface $x_1x_2x_3 = 1$,

710: and then $\sigma_0/\sigma_1$ becomes the function $x_1 + x_2 + x_3$, which

711: appears in Givental's work \cite[Theorem 5]{givental94} as the object mirror

712: dual to the projective plane. The next result exhibits the mirror phenomenon

713: within our framework.

714: 

715: \begin{prop} \label{th:mirror}

716: Let $(E,\pi)$ be the exact Morse fibration constructed from

717: $(X,\sigma_0,\sigma_1)$, with relative Maslov map $\delta_{E/D}$ obtained from

718: \eqref{eq:relative-maslov}. Then for any graded distinguished basis of

719: vanishing cycles ones has

720: \[

721: D^b\Lag(\widetilde{\Gamma}) \iso D^b Coh({\mathbb P}^2)

722: \]

723: where $D^b Coh({\mathbb P}^2)$ is the derived category, in the classical sense,

724: of coherent sheaves on the projective plane over the field $\Z/2$.

725: \end{prop}

726: 

727: The fibre $M = E_{z_0}$ is a torus with three small discs removed. Figure

728: \ref{fig:mirrorp2} shows a particular distinguished basis of vanishing cycles

729: $\Gamma = (L_1,L_2,L_3)$. In that picture, the Maslov map $\delta_M$ is that

730: given by the flat structure of $M$. One can therefore choose gradings such that

731: the groups $CF^*(\tL_i,\tL_k)$, $i<k$, are concentrated in degree zero. Then

732: $\A = \Lag(\widetilde{\Gamma})$ is given by the quiver with relations

733: \[

734: \xymatrix{

735:  {\bullet} \ar@/^1.5pc/[rr]^{a_1} \ar[rr]^{a_2} \ar@/_1.5pc/[rr]_{a_3} &&

736:  {\bullet} \ar@/^1.5pc/[rr]^{b_1} \ar[rr]^{b_2} \ar@/_1.5pc/[rr]_{b_3} &&

737:  {\bullet}

738: } \qquad

739: \begin{cases}

740:  b_i a_j = b_j a_i & i \neq j, \\

741:  b_i a_i = 0.

742: \end{cases}

743: \]

744: A theorem of Beilinson \cite{beilinson78} says that $D^b(\A) \iso D^b

745: Coh({\mathbb P}^2)$. Since $D^b(\A)$ remains the same for all other

746: distinguished bases, the Proposition is proved. The same approach can be used

747: for the mirror partner of $\CP{1} \times \CP{1}$, with analogous results; I

748: have not checked any further cases.

749: 

750: \begin{remarks}

751: (i) It is maybe helpful to mention one way of making drawings like Figure

752: \ref{fig:mirrorp2}. By projecting to one coordinate, each fibre

753: $(\sigma_0/\sigma_1)^{-1}(z)$ of $\sigma_0/\sigma_1: X \setminus Y_\infty

754: \rightarrow \C$ can be represented as a double covering of $\C^*$ branched over

755: three points. The vanishing cycles are preimages of paths in $\C^*$ joining two

756: branch points, and one finds them by looking at how the branch points come

757: together as one varies $z$.

758: 

759: (ii) One sees from Figure \ref{fig:mirrorp2} that the Dehn twists $\tau_{L_i}$

760: can be represented by affine maps, at least away from a small neighbourhood of

761: the missing discs. It follows that any Lagrangian configuration $\Gamma' =

762: (L_1',L_2',L_3')$ which is Hurwitz equivalent to $\Gamma$ consists of curves

763: $L_i'$ isotopic in $M$ to straight lines. An easy argument using the topology

764: of $E$ %specifically $\pi_2(E) = 0$

765: shows that no two $L_i'$ can ever be isotopic to each other. Therefore

766: $HF^*(\tL_i',\tL_k')$, $i<k$, is concentrated in a single degree, for any

767: choice of gradings. This is related to the fact that mutations of the standard

768: exceptional collection in $D^b Coh({\mathbb P}^2)$ remain strongly exceptional;

769: see \cite{bondal-polishchuk94} for much more about this subject.

770: \end{remarks}

771: 

772: \subsection{}

773: Let $f \in \C[x_1,\dots,x_{n+1}]$ be a polynomial with an isolated critical

774: point at the origin. A Morsification of $f$ is a smooth family of polynomials

775: $(f_t)_{0 \leq t < \rho}$ with $f_0 = f$ and such that the critical points of

776: $f_t$, $t>0$, which lie near the origin are nondegenerate. For sufficiently

777: small $0<\epsilon$, $0<\delta \ll \epsilon$, and $0 < t \ll \delta$, one finds

778: that

779: \[

780: X = \{ x \in \C^{n+1} \suchthat |x| \leq \epsilon, \; |f_t(x)| \leq \delta\}

781: \]

782: is a manifold with corners, and $f_t$ a holomorphic Morse function on it. One

783: can equip $X$ with the standard symplectic form on $\C^{n+1}$, and its

784: canonical bundle has a standard trivialization obtained from the constant

785: holomorphic volume form. After some modifications, this becomes an exact Morse

786: fibration $(E,\pi)$ over $D$ together with a canonical homotopy class of

787: relative Maslov maps $\delta_{E/D}$. We will not explain the details since the

788: procedure is similar to that for Lefschetz pencils. A deformation equivalence

789: argument shows that the resulting category $D^b\Lag(\widetilde{\Gamma})$ is

790: independent of all choices, and even of the Morsification, so that it is an

791: invariant of $f$.

792: 

793: \begin{prop} \label{th:ade}

794: Let $f \in \C[x_1,x_2]$ be an ADE singularity. Then the category

795: $D^b\Lag(\widetilde{\Gamma})$ associated to it is equivalent to the derived

796: path category of the Dynkin quiver of the same type, with an arbitrary

797: orientation of the arrows.

798: \end{prop}

799: 

800: A'Campo \cite[pp.\ 13--17]{acampo75b} has constructed particularly nice

801: Morsifications and distinguished bases $\Gamma = (L_1,\dots,L_m)$ for these

802: singularities. These have the properties that (i) there is a bijection between

803: the $L_i$ and the vertices of the corresponding Dynkin diagram, such that $L_i

804: \cap L_k$ consists of a single point if the $i$-th and $k$-th vertex are

805: connected by a line, and is empty otherwise; (ii) there are no $i<k<l$ with

806: $L_i \cap L_k \neq \emptyset$ and $L_k \cap L_l \neq \emptyset$. Figure

807: \ref{fig:e6} shows the $E_6$ case. One can choose gradings $\tL_i$ such that

808: $HF^*(\tL_i,\tL_k)$ is concentrated in degree zero for all $i<k$. Then

809: $\Lag(\widetilde{\Gamma})$ is the path category of the Dynkin quiver for a

810: particular orientation, which is such that each vertex is either a sink or a

811: source. Changing the orientation does not affect the derived category of the

812: path category, which completes the proof.

813: %

814: \includefigure{e6}{e6.eps}{ht}%

815: 

816: We mention two more expected properties of the categories

817: $D^b\Lag(\widetilde{\Gamma})$ as invariants of singularities. Strictly speaking

818: these are conjectures, but the proofs should not be difficult.

819: \begin{myitemize}

820: \item ($\mu$-invariance)

821: They should remain the same when one deforms $f$ inside its $\mu = \text{\em

822: const.}$ stratum. Moreover, if $f$ is adjacent to $g$ then the category

823: belonging to $f$ should contain that of $g$ as a full triangulated subcategory.

824: \item (Stabilization)

825: The categories associated to $f(x)$ and $\hat{f}(x,y) = f(x) + y^2_1 + \dots +

826: y_q^2$ should be equivalent, for any $q$.

827: \end{myitemize}

828: For instance, the reader may have noticed that the categories of

829: $A_m$-singula\-ri\-ties in Proposition \ref{th:ade} are the same as those

830: arising from certain branched covers, which were considered in Section

831: \ref{sec:dimension-zero}b. This is an extreme case of the stabilization

832: property, since the branched covers can be viewed as Morsifications of the

833: one-dimensional singularities $f(x) = x^{m+1}$.

834: 

835: \begin{additional}

836: Hori, Iqbal and Vafa \cite{hori-iqbal-vafa00} discuss mirror symmetry for Fano

837: varieties from a physics point of view. The construction of exact Morse

838: fibrations from singularities is similar to the definition of symplectic

839: monodromy in \cite[Section 6]{khovanov-seidel98}.

840: \end{additional}

841: 

842: \section{Floer cohomology for automorphisms\label{sec:tqft2}}

843: %

844: \subsection{}

845: Let $(M,\o,\theta)$ be an exact symplectic manifold (which is, as always,

846: assumed to have contact type boundary). Floer cohomology associates to each

847: $\phi \in \Sympe(M)$ a pair of vector spaces $HF(\phi,-)$ and $HF(\phi,+)$.

848: These correspond to two ways of treating the fixed points near $\partial M$:

849: let $H \in \smooth(M,\R)$ be a function with $H|\partial M \equiv 0$, whose

850: Hamiltonian flow $(\phi_t^H)$ is equal to the Reeb flow on $\partial M$. To

851: define $HF(\phi,-)$ and $HF(\phi,+)$ one perturbs $\phi$ to $\phi \circ

852: \phi_t^H$ for some small $t<0$, respectively $t>0$. The difference is measured

853: by a long exact sequence

854: \begin{equation} \label{eq:boundary-term}

855: \xymatrix{

856:  {HF(\phi,-)} \ar[r] & {HF(\phi,+)} \ar[r] & {H^*(\partial M;\Z/2).}

857:  \ar@/^1.5em/[ll]_{ }

858: }

859: \end{equation}

860: For $\phi = \id_M$ this reduces to the usual cohomology exact sequence, with

861: $HF(\id_M,-) \iso H^*(M,\partial M;\Z/2)$ and $HF(\id_M,+) \iso H^*(M;\Z/2)$.

862: Other properties of $HF(\phi,\pm)$ are its invariance under isotopies within

863: $\Sympe(M)$; conjugation invariance, $HF(\phi_2\phi_1,\pm) \iso

864: HF(\phi_1\phi_2,\pm)$; and Poincar{\'e} duality, $HF(\phi,\pm) \iso

865: HF(\phi^{-1},\mp)^\vee$. If $M$ is equipped with a Maslov map $\delta_M$, a

866: grading $\tphi$ determines a $\Z$-grading on Floer cohomology, denoted by

867: $HF^*(\tphi,\pm)$. This satisfies $HF^*(\tphi[\sigma],\pm) \iso

868: HF^{*-\sigma}(\tphi,\pm)$ for $\sigma \in \Z$.

869: 

870: Let $S$ be a closed oriented surface together with a finite set $\Sigma \subset

871: S$ of marked points, and $(E,\pi) = (E,\pi,\Omega,\Theta,J_0,j_0)$ an exact

872: Morse fibration over $S^* = S \setminus \Sigma$, with smooth fibres isomorphic

873: to $M$. Suppose that around each $\zeta \in \Sigma$ we have oriented local

874: coordinates $\psi_\zeta: D \rightarrow S$, such that there are commutative

875: diagrams

876: \begin{equation} \label{eq:local-structure}

877: \xymatrix{

878:  {\Rleq \times T(\phi_\zeta)} \ar[d] \ar[rr]^-{\Psi_{\zeta}} &&

879:  {E} \ar[d]^{\pi} \\

880:  {D^*} \ar[rr]^-{\psi_{\zeta}|D^*} &&

881:  {S^*.}

882: }

883: \end{equation}

884: Here $T(\phi_\zeta)$ is the mapping torus of $\phi_\zeta \in \Sympe(M)$,

885: meaning the quotient of $\R \times M$ by $(t,x) \sim (t-1,\phi_\zeta(x))$. The

886: map from $\Rleq \times T(\phi_\zeta)$ to $D^* = D \setminus \{0\}$ is

887: $(s,[t,x]) \mapsto \exp\,2\pi(s+it)$. The pullback of $\o$ to $\R \times M$ is

888: invariant under the $\Z$-action, hence descends to a two-form $\o_{\phi_\zeta}$

889: on $T(\phi_\zeta)$; and a choice of function $K \in \smooth_c(M \setminus

890: \partial M,\R)$ with $dK = \phi^*\theta - \theta$ yields a one-form

891: $\theta_{\phi_\zeta}$ on the same space. $\Psi_{\zeta}$ is an isomorphism of

892: differentiable fibre bundles between $\Rleq \times T(\phi_\zeta)$ and

893: $E|\psi_\zeta(D^*)$, satisfying $\Psi_\zeta^*\Omega = \o_{\phi_\zeta}$,

894: $\Psi_\zeta^*\Theta =\theta_{\phi_\zeta}$. By and large, this means that the

895: behaviour of $(E,\pi)$ around $\zeta$ is described by the exact symplectic

896: automorphism $\phi_\zeta$. One then has a relative invariant

897: \begin{equation} \label{eq:relative2}

898: \Phirel(E,\pi) \in \bigotimes_{\zeta \in \Sigma} HF(\phi_\zeta,\pm),

899: \end{equation}

900: where the signs $\pm$ can be chosen arbitrarily, subject to the restriction

901: that there must be at least one $\zeta$ labeled $+$ in each connected component

902: of $S$; in particular $\Sigma$ may not be empty. Changing an existing $-$ to a

903: $+$ gives a new relative invariant which is the image of the old one under the

904: map $HF(\phi_\zeta,-) \rightarrow HF(\phi_\zeta,+)$ from

905: \eqref{eq:boundary-term}. To define the relative invariant, one chooses a

906: suitable complex structure on $S^*$ and almost complex structure on $E$, and

907: considers the moduli spaces of pseudo-holomorphic sections.

908: 

909: \begin{remark}

910: The asymmetry between $+$ and $-$ can be explained as follows. By definition of

911: an exact Morse fibration, one has a trivialization near $\partial E$. However,

912: the almost complex structures $J$ on $E$ used here do not have the obvious

913: product structure with respect to this trivialization; this is made necessary

914: by the perturbation of $\phi_\zeta$ near $\partial M$ which defines

915: $HF(\phi_\zeta,\pm)$. The convexity of $\partial E$ with respect to

916: $J$-holomorphic sections now becomes a more delicate matter, and it is there

917: that the sign question appears.

918: \end{remark}

919: 

920: Suppose that $\zeta \neq \zeta'$ are points in $\Sigma$ with $\phi_{\zeta'} =

921: (\phi_{\zeta})^{-1}$, and look at a relative invariant $\Phirel(E,\pi)$ in

922: which $\zeta,\zeta'$ have different signs. The surface $\overline{S}$ obtained

923: by connect summing together $\zeta,\zeta'$ comes with marked points

924: $\overline{\Sigma}$ inherited from $\Sigma \setminus \{\zeta,\zeta'\}$, and

925: with an exact Morse fibration over $\overline{S} \setminus \overline{\Sigma}$.

926: The relative invariant of this new fibration (keeping the signs as before) can

927: be computed, if it is well-defined, by applying the Poincar{\'e} duality

928: pairing $HF(\phi_\zeta,\pm) \otimes HF(\phi_{\zeta'},\mp) \rightarrow \Z/2$ to

929: $\Phirel(E,\pi)$.

930: 

931: This setup can be generalized and unified with that in \cite[Section

932: 3]{seidel00}. In the generalization one considers oriented compact surfaces $S$

933: with boundary, with marked points $\Sigma \subset S$ which may lie on the

934: boundary or in the interior, together with an exact Morse fibration $(E,\pi)$

935: over $S^* = S \setminus \Sigma$ having a Lagrangian boundary condition $Q$. The

936: structure of $(E,\pi)$ near a marked point in the interior remains as before,

937: while near boundary marked points $(E,\pi,Q)$ is as in \cite{seidel00}. One

938: then gets relative invariants of a mixed kind,

939: \begin{equation} \label{eq:mixed}

940:  \Phirel(E,\pi,Q) \in \bigotimes_{\zeta \in \Sigma \cap \partial S}

941:  HF(L_{\zeta,+},L_{\zeta,-}) \otimes \bigotimes_{\zeta \in \Sigma \setminus

942:  \partial S} HF(\phi_\zeta,\pm).

943: \end{equation}

944: The restriction that there must be at least one $+$ now applies only to those

945: connected components of $S$ which have no boundary; and there are two different

946: gluing formulas, for boundary and interior marked points respectively.

947: 

948: \subsection{}

949: We can now explain several kinds of maps between Floer cohomology groups as

950: special cases of the relative invariants \eqref{eq:relative2} and

951: \eqref{eq:mixed}.

952: 

953: \begin{myitemize}

954: \item

955: Take an exact symplectic manifold $M$ and $\phi_1,\phi_2 \in \Sympe(M)$. Let $S

956: = S^2$ and $\Sigma = \{\text{\em 3 points}\}$. There is a unique exact Morse

957: fibration $(E,\pi)$ over $S^*$ which has no critical points and is even flat

958: (locally trivial), such that the symplectic monodromy around the three missing

959: points is respectively, $\phi_1$, $\phi_2$ and $(\phi_1\phi_2)^{-1}$. By

960: marking the points with suitable signs and using Poincar{\'e} duality, one

961: obtains from $\Phirel(E,\pi)$ a map $HF(\phi_1,\pm) \otimes HF(\phi_2,+)

962: \rightarrow HF(\phi_1\phi_2,\pm)$, which is the ``pair-of-pants'' product in

963: the Floer cohomology of symplectic automorphisms.

964: 

965: \item

966: Let $L \subset M$ be an exact framed Lagrangian sphere. Then there is an exact

967: Morse fibration $(E,\pi)$ over $D$ with a single critical point, such that the

968: monodromy around $\partial D$ is $\tau_L$. If one extends this to $\C = S^2

969: \setminus \{\infty\}$ in such a way that it is flat near $\infty$, one gets an

970: invariant $\Phirel(E,\pi) \in HF(\tau_L^{-1},+)$.

971: 

972: \item

973: Take the pair-of-pants product with $\phi_1 = \phi\tau_L$ and $\phi_2 =

974: \tau_L^{-1}$, and plug in the distinguished element of $HF(\tau_L^{-1},+)$

975: constructed above. This gives a canonical map $HF(\phi\tau_L,\pm) \rightarrow

976: HF(\phi,\pm)$, which we call $c$. Thanks to the gluing formula, $c$ can also be

977: described as the relative invariant of a certain exact Morse fibration over

978: $S^2 \setminus \{\text{\em 2 points}\}$ with a single critical point.

979: 

980: \item

981: For any $\phi \in \Sympe(M)$, $\Rleq \times T(\phi^{-1})$ is an exact Morse

982: fibration over $D^*$. Denote by $(E,\pi)$ its restriction to $D \setminus

983: \{0,i\}$. Parallel transport along $\partial D$ starting at the base point $z_0

984: = -i$ yields a trivialization $E\,|\,\partial D \setminus \{i\} \iso (\partial

985: D \setminus \{i\}) \times M$. Given an exact Lagrangian submanifold $L \subset

986: M$, one can introduce the Lagrangian boundary condition which in this

987: trivialization is $(\partial D \setminus \{i\}) \times L$. Because of the

988: monodromy around $0$, $\Phirel(E,\pi,Q)$ takes values in $HF(\phi^{-1},\pm)

989: \otimes HF(\phi(L),L)$. Equivalently, one can see it as a map $d: HF(\phi,\pm)

990: \rightarrow HF(\phi(L),L)$.

991: \end{myitemize}

992: 

993: \begin{theorem} \label{th:exact-sequence2}

994: Let $(M,\o,\theta)$ be an exact symplectic manifold, $L \subset M$ an exact

995: framed Lagrangian sphere, and $\phi \in \Sympe(M)$ an exact automorphism.

996: Suppose that $2c_1(M,L) \in H^2(M,L)$ is zero. Then there is a long exact

997: sequence, with maps $c$ and $d$ as defined above,

998: \begin{equation} \label{eq:sequence2}

999: \xymatrix{

1000:  {HF(\phi\tau_L,\pm)} \ar[r]^-{c} &

1001:  {HF(\phi,\pm)} \ar[d]^-{d} \\ &

1002:  {HF(\phi(L),L).} \ar@/^1pc/[ul]

1003: }

1004: \end{equation}

1005: \end{theorem}

1006: 

1007: If $M$ has a Maslov map $\delta_M$, and $\phi$, $L$ have gradings $\tphi$,

1008: $\tL$ then, with respect to the natural gradings of the groups in

1009: \eqref{eq:sequence2}, the maps $c,d$ have degree zero while the remaining one

1010: has degree one. The proof of the exact sequence is similar to that of

1011: \cite[Theorem 3.3]{seidel00}.

1012: 

1013: \begin{example}

1014: By taking $\phi = \id_M$ resp.\ $\phi = \tau_L^{-1}$ one gets exact sequences

1015: \begin{align*}

1016:  \xymatrix{

1017:  {HF(\tau_L,+)} \ar[r]

1018:  & {H^*(M;\Z/2)} \ar[d] \\ &

1019:  {H^*(L;\Z/2),} \ar@/^1pc/[ul]

1020:  }

1021:  \qquad

1022:  \xymatrix{

1023:  {H^*(M;\Z/2)} \ar[r]

1024:  & {HF(\tau_L^{-1},+)} \ar[d] \\ &

1025:  {H^*(L;\Z/2).} \ar@/^1pc/[ul]

1026: }

1027: \end{align*}

1028: In the case when $M$ is a surface, this fits in with the results of direct

1029: computation in \cite{seidel96b}, which are that $HF(\tau_L,+) \iso

1030: H^*(M,L;\Z/2)$ and $HF(\tau_L^{-1},+) \iso H^*(M \setminus L;\Z/2)$. In fact,

1031: the same holds in all dimensions, as one can show by taking a closer look at

1032: the map $d$ in the exact sequence.

1033: \end{example}

1034: %

1035: \includefigure{skein2}{skein2.eps}{hb}%

1036: 

1037: Now take an exact Morse fibration $(E,\pi)$ with arbitrary base $S$. To any

1038: loop $l: S^1 \rightarrow S \setminus S^\crit$ one can associate its symplectic

1039: monodromy $\rho_l$ and hence Floer cohomology groups, which are invariant under

1040: homotopies of $l$ that do not pass across $S^\crit$. We write $HF(l,\pm)$

1041: instead of $HF(\rho_l,\pm)$, in parallel with the notation $HF(c)$ adopted at

1042: the end of \cite[Section 3]{seidel00} for paths $c: [0;1] \rightarrow S$ with

1043: $c^{-1}(S^\crit) = \{0;1\}$. Theorem \ref{th:exact-sequence2} implies that if

1044: one has loops $l_-,l_+$ and a path $c$ as in Figure \ref{fig:skein2}, there is

1045: a long exact sequence

1046: \[

1047: \xymatrix{

1048:  {HF(l_-)} \ar[r] & {HF(l_+)} \ar[r] & {HF(c).} \ar@/^1.5pc/[ll]^{ }

1049: }

1050: \]

1051: The connection with the original formulation is made by the Picard-Lefschetz

1052: theorem: $l_+$ differs from $l_-$ by a clockwise turn around $z \in S^\crit$,

1053: so that $\rho_{l_+}^{-1}\rho_{l_-}$ is the Dehn twist along the vanishing cycle

1054: arising from $z$.

1055: 

1056: \section{Hochschild cohomology}

1057: %

1058: \subsection{}

1059: The Hochschild cohomology $HH^*(\A,\A)$ of a (small) $A_\infty$-category $\A$

1060: is a graded vector space, defined via the following cochain complex

1061: $CC(\A,\A)$. A cochain of degree $r$ is a sequence $h = (h^d)_{d \geq 0}$ of

1062: which each member $h^d$ is a family of graded linear maps

1063: \[

1064: \bigotimes_{i=1}^d hom_\A(X_i,X_{i+1}) \rightarrow hom_\A(X_1,X_{d+1})[r-d],

1065: \]

1066: one for each $(d+1)$-tuple $X_1,\dots,X_{d+1} \in \Ob\,\A$, satisfying

1067: $h^d(a_d,\dots,a_1) = 0$ whenever some $a_i$ is the identity map $\id_{X_i}$.

1068: In particular, $h^0$ consists of an element in $hom_\A^r(X,X)$ for each $X$.

1069: The differential is

1070: \[

1071: \begin{split}

1072:  & (\partial h)^d(a_d,a_{d-1},\dots,a_1) = \\

1073:  & \qquad

1074:  = \sum_{i+j \leq d+1}

1075:  \mu_\A^{d+1-j}(a_d,\dots,a_{i+j},h^j(a_{i+j-1},\dots,a_i),

1076:  a_{i-1},\dots,a_1)

1077:  \\

1078:  & \quad \qquad

1079:  + \sum_{i+j \leq d+1}

1080:  h^{d+1-j}(a_d,\dots,a_{i+j},\mu_\A^j(a_{i+j-1},\dots,a_i),

1081:  a_{i-1},\dots,a_1).

1082: \end{split}

1083: \]

1084: When $\A$ is directed, $CC(\A,\A)$ and hence $HH^*(\A,\A)$ are

1085: finite-dimensional.

1086: 

1087: Hochschild cohomology has two related but different interpretations. One is in

1088: terms of first order deformations of the maps $\mu^d_\A$, and $\partial h = 0$

1089: appears there as the linearization of the structure equations of an

1090: $A_\infty$-category, extended by an $h^0$ term. The other involves the

1091: $A_\infty$-category $funct(\A,\A)$ of $A_\infty$-functors from $\A$ to itself,

1092: and it says that

1093: \begin{equation} \label{eq:functors}

1094: HH^*(\A,\A) = H(hom_{funct(\A,\A)}(\mathrm{Id}_\A,\mathrm{Id}_\A))

1095: \end{equation}

1096: is the space of $A_\infty$-natural transformations from the identity functor to

1097: itself. In the case where $\mu^d_\A = 0$ for $d \geq 3$, which means $\A$ is a

1098: dg-category, there is yet another approach which is essentially equivalent to

1099: \eqref{eq:functors} but more in line with classical homological algebra. Let

1100: $\A\!-\!mod\!-\!\A$ be the dg-category of dg-functors from $\A^{opp} \times \A$

1101: to chain complexes of vector spaces, and $D(\A\!-\!mod\!-\!\A)$ its derived

1102: category, defined by inverting quasi-isomorphisms as in \cite[Chapter

1103: 10]{bernstein-lunts94}. Take the dg-functor $\Delta_{\A} \in

1104: \Ob\,\A\!-\!mod\!-\!\A$ which assigns to $(X,Y) \in \Ob(\A^{opp} \times \A) =

1105: \Ob\,\A \times \Ob\,\A$ the complex $hom_\A(X,Y)$ (to motivate the notation, we

1106: should say that if $\A$ had only a single object $Z$, objects of

1107: $\A\!-\!mod\!-\!\A$ would be just dg-bimodules over the dg-algebra

1108: $hom_\A(Z,Z)$, and $\Delta_\A$ would be $hom_\A(Z,Z)$ considered as a bimodule

1109: over itself). Then

1110: \begin{equation} \label{eq:bimodule}

1111: HH^*(\A,\A) = \Hom^*_{D(\A\!-\!mod\!-\!\A)}(\Delta_\A,\Delta_\A).

1112: \end{equation}

1113: The proof goes by observing that the Hochschild complex coincides with

1114: $hom_{\A\!-\!mod\!-\!\A}(B_{\A}, \Delta_{\A})$ where $B_{\A}$, the bar

1115: construction, is quasi-isomorphic to $\Delta_{\A}$ and $K$-projective in the

1116: sense of \cite[Definition 10.12.2.1]{bernstein-lunts94}.

1117: 

1118: \subsection{}

1119: An informal principle says that $HH^*(\A,\A)$ should be an invariant of

1120: $D^b(\A)$. It is unknown whether this is rigorously true, but there is a weaker

1121: result which is sufficient for our purpose.

1122: 

1123: \begin{prop} \label{th:hochschild}

1124: Mutation of a directed $A_\infty$-category does not change its Hochschild

1125: cohomology.

1126: \end{prop}

1127: 

1128: To prove this one goes through the list of moves \cite[Definition

1129: 5.2]{seidel00}. If $F: \A \rightarrow \B$ is a quasi-isomorphism, there are

1130: natural maps

1131: \[

1132: HH^*(\A,\A) \rightarrow H(hom_{funct(\A,\B)}(F,F)) \leftarrow HH^*(\B,\B),

1133: \]

1134: both of which can be shown to be isomorphisms by looking at the spectral

1135: sequence associated to the length filtration. Next, changing the gradings of

1136: the morphism groups $hom_\A(X^i,X^k)$ by some $\sigma_i-\sigma_k$ does not

1137: affect the Hochschild complex. Now consider $\A \rightsquigarrow c\A$. One can

1138: always find a directed dg-category ${\mathcal C}$ with a quasi-isomorphism

1139: ${\mathcal C} \rightarrow \A$, and then $c{\mathcal C}$ is quasi-isomorphic to

1140: $c\A$. Hence it is sufficient to prove $HH^*(c{\mathcal C},c{\mathcal C}) \iso

1141: HH^*({\mathcal C},{\mathcal C})$, which reduces the problem to the case of

1142: dg-categories. In that case there is a derived Morita equivalence between

1143: ${\mathcal C}$ and $c{\mathcal C}$ , which means that there are objects $P \in

1144: \Ob\,D(c{\mathcal C}\!-\!mod\!-\!{\mathcal C})$, $Q \in \Ob\,D({\mathcal

1145: C}\!-\!mod\!-\!c{\mathcal C})$ with

1146: \[

1147:  P \otimes^{{\bf L}}_{{\mathcal C}} Q \iso \Delta_{c{\mathcal C}}, \quad

1148:  Q \otimes^{{\bf L}}_{c{\mathcal C}} P \iso \Delta_{{\mathcal C}};

1149: \]

1150: for the definition of $\otimes^{{\bf L}}$ see again \cite[Chapter

1151: 10]{bernstein-lunts94}. The desired equality is a formal consequence of this

1152: and \eqref{eq:bimodule}. The argument for $\A \rightsquigarrow r\A$ is

1153: identical.

1154: 

1155: \begin{additional}

1156: This section is essentially a review of known facts. $HH^*(\A,\A)$ is defined

1157: in \cite{penkava-schwarz95}; the $A_\infty$-categories $funct(-,-)$ in

1158: \cite{fukaya97}; and an account of derived Morita equivalences can be found in

1159: \cite{koenig-zimmermann98}. These are just sample references, and by no means

1160: the first ones historically.

1161: \end{additional}

1162: 

1163: \section{Hochschild cohomology and global monodromy\label{sec:donaldson}}

1164: %

1165: \subsection{}

1166: What follows is my attempt to formulate an idea of Donaldson. Let $(E,\pi)$ be

1167: an exact Morse fibration over $D$ with a relative Maslov map $\delta_{E/D}$,

1168: $M$ the fibre over the base point with its induced Maslov map $\delta_M$, and

1169: $\mu \in \Sympe(M)$ the global monodromy, defined to be the symplectic parallel

1170: transport around $\partial D$ in positive direction. This comes with a

1171: canonical grading $\tilde{\mu}$, characterized by being zero near $\JJ_M \times

1172: \partial M$. After making an admissible choice of paths $(c_1,\dots,c_m)$, with

1173: corresponding distinguished basis $\Gamma = (L_1,\dots,L_m)$, one has that

1174: $\tilde{\mu}$ is isotopic to $\ttau_{L_1}\dots\ttau_{L_m}$ within the group of

1175: graded exact symplectic automorphisms. Choose gradings $\tL_i$ and let $\A =

1176: \Lag(\widetilde{\Gamma})$ be the directed Fukaya category. By Proposition

1177: \ref{th:hochschild}, $HH^*(\A,\A)$ is independent of the choice of $\Gamma$ and

1178: of the gradings $\widetilde{\Gamma}$.

1179: 

1180: \begin{conjecture} \label{th:donaldson}

1181: There is a long exact sequence

1182: \begin{equation} \label{eq:donaldson}

1183: \xymatrix{

1184:  {HF^*(\tilde{\mu},+)} \ar[r] &

1185:  {H^*(E;\Z/2)} \ar[d] \\

1186:  & {HH^*(\A,\A)} \ar[ul]

1187: }

1188: \end{equation}

1189: with the $\nwarrow$ map having degree one, and the others degree zero.

1190: \end{conjecture}

1191: 

1192: It is elementary to check, using the classical Picard-Lefschetz formula, that

1193: the Euler characteristics add up in the right way. There is also an informal

1194: argument showing that, in a somewhat loose sense, $HF^*(\tilde{\mu},+)$ is

1195: ``constructed'' from nearly the same pieces as $HH^*(\A,\A)[-1]$, with

1196: $H^*(E;\Z/2)$ compensating for the small difference. We will now reproduce this

1197: argument, both because it seems to have been Donaldson's original motivation

1198: and because it uses nicely the two exact sequences in Floer cohomology. As in

1199: the discussion at the end of Section \ref{sec:tqft2}, it is convenient to use

1200: the notation $HF(l,\pm)$ and $HF(c)$ for Floer cohomology groups.

1201: %

1202: \includefigure{skd1}{skd1.eps}{hb}

1203: 

1204: \begin{myitemize}

1205: \item[(i)]

1206: Let $l_0,\dots,l_m$ be the loops in $D \setminus D^{\crit}$ shown in Figure

1207: \ref{fig:skd1}. To each of them is associated a Floer group $HF(l_k,+)$. Our

1208: aim is to successively ``decompose'' $HF(l_m,+)$. To begin, by Theorem

1209: \ref{th:exact-sequence2} one has long exact sequences

1210: \[

1211: \xymatrix{

1212:  {HF(l_k,+)} \ar[r] & {HF(l_{k-1},+)} \ar[r] &

1213:  {HF(d_{1,k}),} \ar@/^1.5pc/[ll]^{ }

1214: }

1215: \]

1216: where the $d_{1,k}$ are as in Figure \ref{fig:skd2}. We interpret this as

1217: saying that $HF(l_m,+)$ is ``constructed'' by putting together the pieces

1218: $HF(l_0,+)$, $HF(d_{1,1})$, \dots, $HF(d_{1,m})$.

1219: %

1220: \includefigure{skd2}{skd2.eps}{hb}

1221: 

1222: \item[(ii)]

1223: We next analyze $HF(d_{1,k})$ further. Take the paths $e_{i,k}$ and $e'_{i,k}$,

1224: $i<k$, from Figure \ref{fig:skd2}. \cite[Theorem 3.3]{seidel00} yields long

1225: exact sequences

1226: \[

1227: \xymatrix{

1228:  {HF(d_{i,k})} \ar[r] & {HF(d_{i+1,k})} \ar[r] &

1229:  {HF(e_{i,k}) \otimes HF(e'_{i,k}),} \ar@/^1.5pc/[ll]^{ }

1230: }

1231: \]

1232: so that $HF(d_{1,k})$ is ``built'' from $HF(d_{k,k})$ and $HF(e_{i,k}) \otimes

1233: HF(e'_{i,k})$, $1 \leq i < k$.

1234: 

1235: \item[(iii)]

1236: A similar argument as in (ii) shows that $HF(e_{i,k}')$ can be ``decomposed''

1237: into $HF(e_{i,k})^\vee$ together with $HF(e_{i,j})^\vee \otimes HF(e_{j,k}')$

1238: for all $i < j < k$ (the dual vector space appears because the paths $e_{i,k}$,

1239: $e_{i,j}$ occur with reversed orientation). One can apply the same argument

1240: recursively to the second component $HF(e_{j,k}')$ of the tensor product, and

1241: this result in a different ``decomposition'' of $HF(e_{i,k}')$, of which the

1242: pieces are $(HF(e_{i_d,i_{d+1}}) \otimes \dots \otimes HF(e_{i_1,i_2}))^\vee$

1243: ranging over all $d \geq 1$ and $i = i_1 < i_2 < \dots < i_{d+1} = k$.

1244: \end{myitemize}

1245: %

1246: Putting all of this together yields a list of ``elementary pieces'' which make

1247: up $HF(l_m,+)$. At this point we revert to the conventional notation, so that

1248: $HF(l_m,+) = HF(\mu,+)$, $HF(l_0,+) = HF(\id_M,+) \iso H^*(M;\Z/2)$,

1249: $HF(d_{k,k}) = HF(L_k,L_k) \iso H^*(L_k;\Z/2)$, and $HF(e_{i,k}) =

1250: HF(L_i,L_k)$. We also restore the gradings, which were omitted up to now for

1251: the sake of simplicity. Then the ``elementary pieces'' are:

1252: \[

1253: \left\{

1254: \begin{split}

1255:  & H^*(M;\Z/2); \\

1256:  & (\Z/2)^m[-n\!-\!1]; \\

1257:  & (\Z/2)^m[-1]; \text{ and} \\

1258:  & HF^*(\tL_{i_1},\tL_{i_{d+1}})[-d\!-\!1] \otimes

1259:  (HF^*(\tL_{i_d},\tL_{i_{d+1}}) \otimes \dots \otimes

1260:  HF^*(\tL_{i_1},\tL_{i_2}))^\vee \\

1261:  & \hspace{10em} \text{ for all $d \geq 1$ and $1\leq i_1 < \dots < i_{d+1} \leq m$.}

1262: \end{split}

1263: \right.

1264: \]

1265: The second and third piece are the result of splitting each $H^*(L_k;\Z/2)[-1]

1266: = \Z/2[-1] \oplus \Z/2[-n\!-\!1]$. Now $E$ is obtained up to homotopy

1267: equivalence by attaching $m$ cells of dimension $(n+1)$ to $M$. Hence the first

1268: two pieces can be ``put together'' again, in the same informal sense as before,

1269: to form $H^*(E;\Z/2)$. On the other hand the spectral sequence induced by the

1270: length filtration, which converges to $HH^*(\A,\A)[-1]$, has $E_1$ term equal

1271: to the direct sum of the remaining pieces (that is, all except the first two).

1272: Of course this sort of bookkeeping falls far short of a proof of Conjecture

1273: \ref{th:donaldson}, since it ignores all the maps in the long exact sequences,

1274: and equally the differentials in the spectral sequence.

1275: 

1276: \subsection{}

1277: For a better understanding of Conjecture \ref{th:donaldson} one would need to

1278: know what the maps in it are. We outline here a possible description of one of

1279: them, namely $H^*(E;\Z/2) \rightarrow HH^*(\A,\A)$. This is pure speculation,

1280: supported mainly by the fact that the construction, apart from technicalities,

1281: seems natural.

1282: 

1283: One may suppose that the $L_i$ are in generic position. Choose Morse functions

1284: $f$ on $M$ and $f_i$ on $L_i$, such that $f|L_i = f_i$ and the restriction of

1285: $f$ to a tubular neighbourhood of each $L_i$ is given by the sum of $f_i$ and a

1286: positive definite quadratic form on the normal bundle to $L_i$. Moreover,

1287: $f|\partial M = const.$ and the gradient should point outwards there. After

1288: choosing suitable Riemannian metrics, one gets Morse cohomology complexes

1289: $C_{Morse}(M)$ and $C_{Morse}(L_i)$ with $\Z/2$ coefficients, together with

1290: natural restriction maps $r_i: C_{Morse}(M) \rightarrow C_{Morse}(L_i)$.

1291: Enlarge $C_{Morse}(L_i)$ to a complex $\tilde{C}_{Morse}(L_i)$ by adding one

1292: generator in degree $-1$ whose boundary is the only zero-dimensional cocycle

1293: (the sum of all local minima of $f_i$). Composing $r_i$ and the inclusions

1294: $C_{Morse}(L_i) \hookrightarrow \tilde{C}_{Morse}(L_i)$ gives a map $\tilde{r}:

1295: C_{Morse}(M) \rightarrow \bigoplus_i \tilde{C}_{Morse}(L_i)$, whose mapping

1296: cone satisfies

1297: \begin{equation} \label{eq:morse-homology}

1298: H^*(\Cone(\tilde{r})) \iso H^*(E;\Z/2)[1].

1299: \end{equation}

1300: The point of using this rather strange model for the cohomology of $E$ is that

1301: there are natural maps

1302: \begin{equation} \label{eq:first-order}

1303: \tilde{C}_{Morse}(L_i) \rightarrow CC(\A,\A)[1], \quad C_{Morse}(M) \rightarrow

1304: CC(\A,\A).

1305: \end{equation}

1306: The first of these is a chain homomorphism; the second isn't, but the direct

1307: sum of the two is a chain map $\Cone(\tilde{r}) \rightarrow CC(\A,\A)[1]$,

1308: which in combination with \eqref{eq:morse-homology} gives the desired map

1309: $H^*(E;\Z/2) \rightarrow HH^*(\A,\A)$.

1310: 

1311: Both maps in \eqref{eq:first-order} are constructed by the method of ``cutting

1312: down moduli spaces'' which is familiar from the definition of the cap product

1313: on Floer cohomology. We will describe the first map in detail, and the second

1314: one only briefly. Fix $i \in \{1,\dots,m\}$ and let $Z \subset L_i$ be the

1315: unstable manifold, under the gradient flow, of some critical point of $f_i$ of

1316: Morse index $p$. Suppose for the moment that $p>0$, so that $Z$ has positive

1317: codimension. Recall that the boundary operator $\mu_{\A}^1$ on $CF^*(\tL_{i_1},

1318: \tL_{i_2})$, $i_1<i_2$, is defined using pseudo-holomorphic maps $u: \R \times

1319: [0;1] \rightarrow M$ with boundary conditions $u(\R \times \{2-\nu\}) \subset

1320: L_{i_\nu}$. If $i_\nu = i$ for some $\nu \in \{1,2\}$, one can consider the

1321: subset of those maps $u$ which satisfy $u(0,2-\nu) \in Z$. If one assumes that

1322: the almost complex structures have been chosen generically (similar assumptions

1323: will be made tacitly later on) then counting isolated points in that subset

1324: yields a map

1325: \begin{equation} \label{eq:h1}

1326: h^1_Z: CF^*(\tL_{i_1},\tL_{i_2}) \rightarrow CF^{*+p}(\tL_{i_1},\tL_{i_2}).

1327: \end{equation}

1328: Extend this to the other Floer cochain groups in $\A$, that is to say to

1329: $CF^*(\tL_{i_1},\tL_{i_2})$ with $i_1,i_2 \neq i$, by setting $h_Z$ to be zero

1330: there. Now consider the composition maps of order $d \geq 2$, $\mu_\A^d:

1331: CF^*(\tL_{i_d},\tL_{i_{d+1}}) \otimes \dots \otimes CF^*(\tL_{i_1}, \tL_{i_2})

1332: \rightarrow CF^{*+2-d}(\tL_{i_1},\tL_{i_{d+1}})$ for $i_1 < \dots < i_{d+1}$.

1333: Their definition uses moduli spaces of pairs $(r,u)$, where $r \in \RR^{d+1}$

1334: is a point in the moduli space of discs with $d+1$ marked boundary point,

1335: $\SS^{d+1,*}_r$ is the corresponding disc with the marked points removed, and

1336: $u: \SS^{d+1,*}_r \rightarrow M$ is a map satisfying a pseudo-holomorphicity

1337: equation. $\partial \SS^{d+1,*}_r$ consists of intervals $I^{d+1}_{r,\nu}$, $1

1338: \leq \nu \leq d+1$, and the boundary conditions are

1339: \[

1340: u(I^{d+1}_{r,\nu}) \subset L_{i_\nu}.

1341: \]

1342: Assuming that $i_{\mu} = i$ for some $\mu$, one considers the moduli space of

1343: triples $(r,u,z)$, where $z$ is a point in $I^{d+1}_{r,\mu}$ such that $u(z)

1344: \in Z$. Counting isolated points in this moduli space defines a map

1345: \[

1346: h^d_Z: CF^*(\tL_{i_d},\tL_{i_{d+1}}) \otimes \dots \otimes CF^*(\tL_{i_1},

1347: \tL_{i_2}) \rightarrow CF^{*+p+1-d}(\tL_{i_1},\tL_{i_{d+1}}).

1348: \]

1349: Again, this is extended trivially to the cases when $i_1,\dots,i_{d+1} \neq i$.

1350: The sequence $(h_Z^0 = 0,h_Z^1,h_Z^2,\dots)$ is a Hochschild cochain $h_Z \in

1351: CC^{p+1}(\A,\A)$.

1352: 

1353: The case $p = 0$, which we had excluded above, is considerably easier: one

1354: takes \eqref{eq:h1} to be the projection to the subspace spanned by the points

1355: of $L_{i_1} \cap L_{i_2}$ which lie on $Z$, and all other $h^d_Z$ to be zero.

1356: Finally there is a canonical element $\tilde{h} \in CC^0(\A,\A)$, defined by

1357: setting $\tilde{h}^0$ to be the unit element in $hom_\A^0(\tL_i,\tL_i) \iso

1358: \Z/2$, with all other components equal to zero. The assignment $Z \mapsto h_Z$,

1359: extended to $\tilde{C}_{Morse}(L_i)$ by mapping the generator of degree $-1$ to

1360: $\tilde{h}$, gives the first map in \eqref{eq:first-order}.

1361: 

1362: The definition of the second map uses moduli spaces of triples $(r,u,z)$ with a

1363: point $z$ which can lie anywhere in $\SS^{d+1,*}_r$, such that $u(z)$ is in

1364: some stable manifold of $f$. The failure of this to be a chain map reflects the

1365: fact that $z$ can move to the boundary; in that case however $u(z)$ necessarily

1366: lies on some $L_i$, which provides the connection with the first map.

1367: 

1368: \begin{remarks}

1369: (i) This discussion may seem rather abstract, but it has concrete computational

1370: implications, most strikingly for $n = 1$ when $M$ is a surface. In that case

1371: the $A_\infty$-category $\A$ is constructed from $\Gamma$ by a purely

1372: combinatorial count of immersed polygons, and the same is true for the maps

1373: \eqref{eq:first-order}. Suppose that Conjecture \ref{th:donaldson} is true and

1374: that our guess for the map $H^*(E;\Z/2) \rightarrow HH^*(\A,\A)$ which occurs

1375: in it is correct. Then one gets a combinatorial algorithm for computing the

1376: Floer cohomology $HF(\tau_{L_1}\dots\tau_{L_m},+)$ of an arbitrary product of

1377: Dehn twists in $M$ (along exact curves $L_i$ which admit gradings, to be

1378: precise).

1379: 

1380: (ii) From a deformation theory point of view, what we have described is a first

1381: order infinitesimal deformation of $\A$ parame\-trized by the graded vector

1382: space $H^*(E;\Z/2)$. This was done by inserting one marked point into the

1383: Riemann surfaces which define $\mu^d_\A$. It seems natural to expect that using

1384: more marked points will allow one to extend this to higher order. This is

1385: interesting because in principle there are obstructions to such an extension,

1386: which are expressed by a natural graded Lie bracket on $HH^*(\A,\A)[1]$. If the

1387: extension to higher order is indeed possible, it would mean that the Lie

1388: bracket vanishes on the image of $H^*(E;\Z/2) \rightarrow HH^*(\A,\A)$.

1389: \end{remarks}

1390: 

1391: \subsection{}

1392: To round off the discussion of Conjecture \ref{th:donaldson} we look at its

1393: implications in some specific cases. Suppose first that $(E,\pi)$ and

1394: $\delta_{E/D}$ come from a Lefschetz pencil satisfying

1395: \eqref{eq:relative-maslov} for some $a \in \Z$. Then $\mu$ can be isotoped to

1396: the identity but only within a group of symplectic automorphisms which is

1397: larger than $\Sympe(M)$, the isotopy being nontrivial along $\partial M$. Floer

1398: cohomology is not invariant under such isotopies. However, a more careful

1399: analysis shows that $HF(\mu,+) \iso HF(\id_M,-) \iso H^*(M,\partial M;\Z/2)$.

1400: More precisely, taking the canonical grading into account, this formula reads

1401: \[

1402: HF^*(\tilde{\mu},+) \iso H^{*+a-4}(M,\partial M;\Z/2).

1403: \]

1404: For $X = \CP{2}$, $\xi = {\mathcal O}(2)$, $\A$ has been determined in Section

1405: \ref{sec:examples}a, and one finds by explicit computation that $HH^r(\A,\A)

1406: \iso H^r(M,\partial M;\Z/2) \oplus H^r(E;\Z/2)$ for all $r$. This is compatible

1407: with an exact sequence \eqref{eq:donaldson} in which the horizontal arrow would

1408: be the zero map. There are reasons to believe that this arrow will in fact

1409: vanish for all Lefschetz pencils except the trivial one (the degree one pencil

1410: on $X = \CP{n+1}$).

1411: %

1412: \includefigure{mirror-monodromy}{mirmon.eps}{hb}

1413: 

1414: Next consider the mirror dual of the projective plane, as in Section

1415: \ref{sec:examples}b. From the structure of the singular fibre $Y_\infty$ one

1416: sees that the global monodromy can be written as a product of Dehn twists and

1417: their inverses along disjoint curves, namely

1418: \[

1419: \mu = \prod_{i=1}^3

1420: (\tau_{B_i}\tau_{C_i}^{-1}\tau_{C_i'}^{-1}\tau_{C_i''}^{-1})

1421: \]

1422: where the curves are as shown in Figure \ref{fig:mirror-monodromy}. Using

1423: \cite{seidel96b} one computes that $HF^*(\tilde{\mu},+) = H^{*-2}(M \setminus

1424: \bigcup_{i=1}^3 C_i \cup C_i' \cup C_i'',\partial M;\Z/2)$. Comparing

1425: dimensions shows that $HH^r(\A,\A) \iso HF^{r+1}(\tilde{\mu},+) \oplus

1426: H^r(E;\Z/2)$, as in the previous example.

1427: 

1428: As a third and final case, suppose that $(E,\pi)$ and $\delta_{E/D}$ come from

1429: a Morsification of an isolated hypersurface singularity. We conjecture that for

1430: all nontrivial singularities

1431: \begin{equation} \label{eq:acyclicity}

1432: HF(\mu) = 0;

1433: \end{equation}

1434: this can be checked easily in many cases, e.g.\ for weighted homogeneous

1435: singularities. From \eqref{eq:acyclicity} and Conjecture \ref{th:donaldson} it

1436: would follow that $HH^*(\A,\A) \iso H^*(E;\Z/2) = \Z/2$. Because of the role of

1437: Hochschild cohomology in deformation theory, this means that $\A$ is rigid: it

1438: admits no nontrivial first order deformations of any degree. As of now, there

1439: is no deeper explanation for this phenomenon.

1440: 

1441: \begin{additional}

1442: For the cap product on Floer cohomology see \cite{le-ono97}, \cite{schwarz96}.

1443: Equation \eqref{eq:acyclicity}, if true, improves on old topological results of

1444: A'Campo \cite{acampo73b} and L{\^e} \cite{le74}.

1445: \end{additional}

1446: 

1447: \section{Morse categories\label{sec:morse}}

1448: %

1449: \subsection{}

1450: Let $N^{n+1}$ be a compact manifold with codimension two corners. More

1451: precisely, we assume that $\partial N$ consists of three codimension one faces

1452: $\partial_- N$, $\partial_0 N$, $\partial_+ N$ (any of which may be empty) with

1453: $\partial_-N \cap \partial_+N = \emptyset$, and where the corners are

1454: $(\partial_- N \cup \partial_+N) \cap \partial_0 N$. Let $p: N \rightarrow

1455: [-1;1]$ be a Morse function satisfying $\partial_-N = p^{-1}(-1)$, $\partial_+N

1456: = p^{-1}(1)$, and such that $p|\partial_0N$, as a function on $\partial_0N$,

1457: has no critical points. In addition we require that if $x \in N$ is a critical

1458: point of $p$ with Morse index $i(x)$, then

1459: \begin{equation} \label{eq:self-indexing}

1460:  \begin{cases}

1461:  -1 < p(x) < 0 & \text{if $i(x) = 0$,} \\

1462:  p(x) = 0 & \text{if $0<i(x)<n+1$,} \\

1463:  0 < p(x) < 1 & \text{if $i(x) = n+1$.} \end{cases}

1464: \end{equation}

1465: For $n = 1$ this is more or less the classical notion of self-indexing Morse

1466: function, but in higher dimensions it is far more restrictive. To any function

1467: $p$ with these properties one can associate a {\bf Morse category}

1468: $Morse(N,p)$. To do that, choose a Riemannian metric such that $\nabla p$ is

1469: parallel to $\partial_0N$. For two critical points $x,y$ of $p$, let ${\mathcal

1470: G}(x,y)$ be the space of unparametrized gradient trajectories going from $x$ to

1471: $y$. This is a smooth manifold; it is compact unless $i(x) = 0$ and $i(y) =

1472: n+1$, in which case there is a natural compactification to a manifold with

1473: boundary $\overline{\mathcal G}(x,y)$ such that

1474: \begin{equation} \label{eq:compactification}

1475: \partial\overline{\mathcal G}(x,y) = \bigsqcup_{0 < i(w) < n+1} {\mathcal

1476: G}(w,y) \times {\mathcal G}(x,w).

1477: \end{equation}

1478: These are all standard facts from Morse theory. What is important is that there

1479: is no need to choose the metric generically, since \eqref{eq:self-indexing}

1480: implies that $\nabla p$ is always Morse-Smale. A cobordism argument based on

1481: this fact shows that the topological type of the spaces ${\mathcal G}(x,y)$ and

1482: of their compactifications is independent of the metric.

1483: 

1484: The objects of $\A = Morse(N,p)$ are the critical points of $p$, and the

1485: morphisms are $\hom_\A(x,y) = H_{-*}({\mathcal G}(x,y);\Z/2)$; note the sign in

1486: front of the grading. The composition maps

1487: \begin{equation} \label{eq:multiplication}

1488: H_{-*}({\mathcal G}(w,y)) \otimes H_{-*}({\mathcal G}(x,w)) \rightarrow

1489: H_{-*}({\mathcal G}(x,y))

1490: \end{equation}

1491: are as follows. If $x = w$ then $[point] \in H_0(point;\Z/2) = hom^0_\A(x,w)$

1492: acts as identity morphism, and similarly for $w = y$. If $i(x) = 0$, $0 < i(w)

1493: < n+1$, and $i(y) = n+1$, the composition comes from the inclusion of the

1494: boundary stratum \eqref{eq:compactification}, together with the fact that

1495: ${\mathcal G}(x,y) \hookrightarrow \overline{\mathcal G}(x,y)$ is a homotopy

1496: equivalence. In all remaining cases \eqref{eq:multiplication} is automatically

1497: zero, because the tensor product on the left hand side vanishes. Thus $\A$ is a

1498: $\Z/2$-linear and $\Z$-graded category. Ordering the critical points according

1499: to $p(x)$ shows that it can also be viewed as a directed $A_\infty$-category

1500: with $\mu^d_\A = 0$ for $d \neq 2$.

1501: 

1502: Let $B(N,p) \subset N \setminus \partial N$ be the subspace of points whose

1503: gradient flow line exists for all time, which means that it converges to a

1504: critical point in both positive and negative direction. Define the fundamental

1505: object $\fund{B(N,p)} \in \Ob\,D^b(\A)$ to be the twisted complex

1506: \[

1507: (C,\delta_C) = \Big(\bigoplus_{dp(x) = 0} x[-i(x)], (\delta_{xy}) \Big),

1508: \]

1509: where $\delta_{xy} \in hom^1_\A(x[-i(x)],y[-i(y)])$ is the fundamental homology

1510: class $[{\mathcal G}(x,y)]$ if that space is compact and $x \neq y$, and zero

1511: otherwise. The generalized Maurer-Cartan equation reduces to

1512: $\mu^2_\A(\delta_C,\delta_C) = 0$, which is a consequence of

1513: \eqref{eq:compactification}. The notation $\fund{B(N,p)}$ is motivated by the

1514: fact that

1515: \begin{equation} \label{eq:sheafcohomology}

1516: \Hom^*_{D^b(\A)}(\fund{B(N,p)},\fund{B(N,p)}) \iso H^*(B(N,p);\Z/2).

1517: \end{equation}

1518: To see this one identifies $hom_{\Tw\,\A}(C,C)$ with the $E^1$ term of the

1519: spectral sequence, converging to $H^*(B(N,p);\Z/2)$, which arises from a

1520: decomposition of $B(N,p)$ determined by the gradient flow. There is only one

1521: nonzero differential in this spectral sequence, and that coincides with the

1522: differential on $hom_{\Tw,\A}(C,C)$, which completes the proof. If $N$ is

1523: closed, so that $B(N,p) = N$, the fundamental object $\fund{N}$ has another

1524: property. Namely, for any $X \in \Ob\,D^b(\A)$ the composition gives rise to a

1525: nondegenerate pairing

1526: \begin{equation} \label{eq:verdier}

1527:  \Hom^*_{D^b(\A)}(X,\fund{N}) \otimes \Hom^{n+1-*}_{D^b(\A)}(\fund{N},X)

1528:  \rightarrow

1529:  \Hom^{n+1}_{D^b(\A)}(\fund{N},\fund{N}) \iso \Z/2.

1530: \end{equation}

1531: 

1532: \begin{remarks}

1533: (i) Consider $B(N,p)$ as a space stratified by the stable manifolds of $\nabla

1534: p$. It was pointed out to me by Khovanov that the bounded derived category of

1535: sheaves of $\Z/2$-vector spaces, constructible with respect to this

1536: stratification, is equivalent to $D^b(\A)$; this appears to be a reformulation

1537: of a familiar result \cite{kapranov90}. The objects $x[-i(x)]$ correspond to

1538: the constant sheaves along the strata, and $\fund{B(N,p)}$ to the constant

1539: sheaf on all of $B(N,p)$. This explains the properties listed above:

1540: \eqref{eq:sheafcohomology} computes the cohomology of the constant sheaf, and

1541: \eqref{eq:verdier} is Verdier duality.

1542: 

1543: (ii) The Morse categories for $p$ and $-p$ are not equivalent. However, they

1544: are Koszul dual, so that their derived categories are equivalent.

1545: \end{remarks}

1546: 

1547: \subsection{}

1548: Assume now that $n = 1$, so that $p^{-1}(0)$ is a nodal curve on the surface

1549: $N$. Put a sign $\ominus,\oplus$ into each connected component of $N \setminus

1550: p^{-1}(0)$ that contains a local minimum respectively maximum of $p$. This

1551: diagram retains enough information about $p$ to reconstruct $\Morse(N,p)$.

1552: %

1553: \includefigure{triple-point}{triple.eps}{ht}%

1554: 

1555: Let's say that two functions are related by a triple point move if the

1556: corresponding diagrams differ as shown in Figure \ref{fig:triple-point}. By a

1557: sequence of mutations that mimics the base changes in \cite[Volume\ 2, I \S 4]

1558: {arnold-gusein-zade-varchenko}, one can prove

1559: 

1560: \begin{prop} \label{th:triple-point}

1561: Up to equivalence, a triple point move does not change the derived Morse

1562: category $D^bMorse(N,p)$.

1563: \end{prop}

1564: 

1565: It would be interesting to find a topological interpretation of $HH^*(\A,\A)$

1566: for $\A = Morse(N,p)$. I have not succeeded in doing that, except for one

1567: special situation. Call $(N,p)$ cellular if all spaces ${\mathcal G}(x,y)$ are

1568: contractible (we are still assuming that $n = 1$; anyway cellularity is

1569: impossible in higher dimensions, except for trivial cases).

1570: 

1571: \begin{prop} \label{th:cellular}

1572: If $(N,p)$ is cellular, $HH^*(\A,\A) \iso H^*(B(N,p);\Z/2)$.

1573: \end{prop}

1574: 

1575: The first step in the proof is to note that the only possible nonzero groups

1576: are $HH^r(\A,\A)$ for $r = 0,1,2$. $HH^0(\A,\A) \iso H^0(B(N,p);\Z/2)$ is

1577: always true. From the cellularity one derives that $HH^1(\A,\A) \iso

1578: H^1(B(N,p);\Z/2)$, and also that the Euler characteristic of $HH^*(\A,\A)$

1579: equals that of $B(N,p)$, which completes the argument.

1580: 

1581: \begin{additional}

1582: Objects similar to our Morse categories were introduced several years ago by

1583: Cohen, Jones and Segal \cite{cohen-jones-segal95} under the name flow

1584: categories. Proposition \ref{th:cellular} was inspired by A'Campo's paper

1585: \cite{acampo00}.

1586: \end{additional}

1587: 

1588: \section{Real structures\label{sec:real}}

1589: %

1590: \subsection{}

1591: For the duration of this section, we extend the definition of exact Morse

1592: fibration by allowing each fibre to contain several critical points. This does

1593: not affect the theory seriously. The notion of admissible choice of paths

1594: remains the same; each path may give rise to several disjoint vanishing cycles,

1595: which one can place in arbitrary order in the distinguished basis. The

1596: important thing is that such bases are still unique up to Hurwitz moves.

1597: 

1598: Let $(E,\pi) = (E,\pi,\Omega,\Theta,J_0,j_0)$ be an exact Morse fibration over

1599: $D$, with $j_0$ the standard complex structure on $D$, and $\delta_{E/D}$ a

1600: relative Maslov map. A real structure on it is an involution $\iota: E

1601: \rightarrow E$ such that

1602: \[

1603:  \pi(\iota(x)) = \overline{\pi(x)},\;\; \iota^*\Omega = -\Omega,\;\;

1604:  \iota^*\Theta = -\Theta,\;\; \iota^*J_0 = -J_0.

1605: \]

1606: There is always a $(j,J) \in \JJ_{E/D}$ with $j$ the standard complex structure

1607: on $D$ and such that $\iota^*J = -J$. Then the line bundle

1608: \begin{equation} \label{eq:delta}

1609: \Delta_{E/D} \,|\, \{(j,J)\} \times E = \Lambda^{n+1}(TE,J)^{\otimes 2} \otimes

1610: \pi^*(TD,j)^{\otimes -2}

1611: \end{equation}

1612: has a natural real involution $\iota_{(j,J)}$ which covers $\iota$. Call

1613: $\iota$ compatible with the relative Maslov map if $\delta_{E/D}\,|\,\{(j,J)\}

1614: \times E$ is homotopic to a trivialization of \eqref{eq:delta} which takes

1615: $\iota_{(j,J)}$ to complex conjugation in $\C$. This condition is independent

1616: of the choice of $(j,J)$.

1617: 

1618: Given a real structure, one can consider $N = E^\iota$ with the Morse function

1619: $p = \pi|N: N \rightarrow [-1;1]$. $N$ is a manifold with corners of the kind

1620: considered in the previous section, with $\partial_-N \cup \partial_+N = (N

1621: \cap \partial_vE) = p^{-1}(\{\pm 1\})$ and $\partial_0N = N \cap \partial_hE$;

1622: and $p|\partial_0N$ has no critical points because the same holds for

1623: $\pi|\partial_hE$. We say that $\iota$ a good real structure if the critical

1624: points of $p$ satisfy \eqref{eq:self-indexing}, in which case there is a

1625: well-defined Morse category $Morse(N,p)$. We say that $\iota$ is complete if

1626: all critical points of $\pi$ lie on $N$.

1627: 

1628: \begin{conjecture} \label{th:real}

1629: Let $(E,\pi)$ be an exact Morse fibration over $D$ with a good and complete

1630: real structure $\iota$, which is also compatible with a relative Maslov map

1631: $\delta_{E/D}$. Take an admissible choice of paths $(c_1,\dots,c_m)$, each of

1632: which lies in the lower half-disc $D \cap \{\im(z) \leq 0\}$, and let $\Gamma$

1633: be the corresponding distinguished basis. Then $\Lag(\widetilde{\Gamma})$, for

1634: a certain choice of grading $\widetilde{\Gamma}$, is quasi-isomorphic to

1635: $Morse(N,p)$.

1636: \end{conjecture}

1637: 

1638: For $n = 1$ there is an elementary way of drawing the vanishing cycles which

1639: constitute $\Gamma$, starting from $(N,p)$; see e.g.\ \cite{acampo98}. Using it

1640: I have proved Conjecture \ref{th:real} in that dimension. There is also a

1641: general argument in support of it which, while falling short of a proof, is

1642: quite suggestive. Take an interval $[a;b] \subset [-1;1] = D \cap \R$ such that

1643: $[a;b] \cap D^\crit = \{a;b\}$. The symplectic parallel transport along $(a;b)$

1644: is $\iota$-equivariant, and hence preserves the real part $N$; in fact its

1645: restriction to $N$ is equal, up to reparametrization of the orbits, to the flow

1646: of $\nabla p$ with respect to some metric. Using property

1647: \eqref{eq:self-indexing} one can show that intersection points of the vanishing

1648: cycles arising from the endpoints $\{a,b\}$ are in one-to-one correspondence

1649: with gradient flow lines joining the corresponding critical points of $p$.

1650: 

1651: \begin{remark}

1652: The combination of Conjecture \ref{th:real}, Proposition \ref{th:cellular} and

1653: Conjecture \ref{th:donaldson} leads to intriguing relations between the global

1654: monodromy of an exact Morse fibration on one hand, and its possible good and

1655: complete real structures on the other hand.

1656: \end{remark}

1657: 

1658: Real structures are not the most general context in which there is a relation

1659: between directed Fukaya categories and Morse categories. Take an exact Morse

1660: fibration $(E,\pi)$ over $D$ with relative Maslov map $\delta_{E/D}$. A

1661: pseudo-real submanifold is a pair of embeddings

1662: \begin{equation} \label{eq:pseudo-real}

1663: \xymatrix{

1664:  {N} \ar[d]_{p} \ar[r]^-{C} & E \ar[d]^{\pi} \\

1665:  {[-1;1]} \ar[r]^{c} & D.

1666: }

1667: \end{equation}

1668: Here $c$ is a curve in $D$ with $c^{-1}(\partial D) = \{\pm 1\}$, meeting

1669: $\partial D$ transversally. $N$ is an $(n+1)$-dimensional manifold with corners

1670: as before. $C$ is an embedding transverse to $\partial_hE$ and $\partial_vE$,

1671: such that $C^{-1}(\partial_vE) = \partial_-N \cup \partial_+N$ and

1672: $C^{-1}(\partial_hE) = \partial_0N$; we require that $C^*\Omega$ is zero and

1673: $C^*\Theta$ is an exact one-form. $p$ is a Morse function such that

1674: $\partial_{\pm}N = p^{-1}(\pm 1)$. Furthermore, using the fact that $\im(C)$ is

1675: totally real with respect to $J$ for any $(j,J) \in \JJ_{E/D}$, one finds that

1676: there is a canonical homotopy class of sections of $S(\Delta_{E/D})$ over

1677: $\{(j,J)\} \times \im(C)$. If a section in that homotopy class can be chosen

1678: such that its composition with $\delta_{E/D}$ is a constant map $N \rightarrow

1679: S^1$, we say that our pseudo-real submanifold is compatible with the relative

1680: Maslov map. The notion of good pseudo-real submanifold is defined as before.

1681: 

1682: \begin{conjecture}

1683: Suppose that $(N,p)$ is a good pseudo-real submanifold in $(E,\pi)$, compatible

1684: with $\delta_{E/D}$. Then $D^b\Lag(\widetilde{\Gamma})$ contains

1685: $D^bMorse(N,p)$ as a full triangulated subcategory. In particular there is an

1686: object $\fund{B(N,p)} \in \Ob\,D^b\Lag(\widetilde{\Gamma})$ whose endomorphism

1687: ring is $H^*(B(N,p);\Z/2)$.

1688: \end{conjecture}

1689: 

1690: Again, it is not difficult to verify this for $n = 1$.

1691: 

1692: \subsection{}

1693: As examples one can consider Lefschetz pencils as in Section

1694: \ref{sec:examples}a which are real, meaning that $X$ and $\xi$ have real

1695: involutions such that the composition of the two preserves $\sigma_0,\sigma_1$.

1696: Then the resulting exact Morse fibration has a natural real structure. To get

1697: the right kind of relative Maslov map, one must ask that the isomorphism

1698: \eqref{eq:relative-maslov} be compatible with the involutions on both sides. An

1699: example is $X = \CP{2}$, $\xi = {\mathcal O}(2)$, $\sigma_0(x) = x_0^2-x_2^2$,

1700: $\sigma_1(x) = x_0^2 + x_1^2 + x_2^2$, with complex conjugation. This gives

1701: rise to a complete and good real structure such that $(N,p)$ is $\RP{2}$ with

1702: its standard Morse function (having three critical points). One verifies easily

1703: that $Morse(N,p)$ is precisely given by the quiver with relations

1704: \eqref{eq:cp2-quiver}. Another possible application is to isolated hypersurface

1705: singularities which have real Morsifications. For instance, the Morsifications

1706: used in Section \ref{sec:examples}c for ADE plane curve singularities are real,

1707: and this provides another way of computing the associated categories.

1708: 

1709: \begin{additional}

1710: The basic idea of relating vanishing cycles and real structures comes from work

1711: of A'Campo \cite{acampo75b} and Gusein-Zade \cite{gusein-zade74}.

1712: \end{additional}

1713: 

1714: \section{Matching pairs and matching paths\label{sec:induction}}

1715: %

1716: \subsection{}

1717: Let ${\mathcal C}$ be a triangulated category, linear over $\Z/2$ and such that

1718: the spaces $\Hom^*_{\mathcal C}(X,Y)$ are finite-dimensional. Suppose that

1719: $(Y^1,\dots,Y^m)$ is a full exceptional collection in ${\mathcal C}$ such that

1720: for some $1 \leq i < m$, the following properties hold. Firstly

1721: \begin{equation} \label{eq:matching}

1722: \Hom^*_{\mathcal C}(Y^i,Y^{i+1}) = H^*(S^n;\Z/2)

1723: \end{equation}

1724: for some $n>0$. Secondly, if $a$ denotes the unique nontrivial morphism $Y^i

1725: \rightarrow Y^{i+1}$ of degree zero, the compositions

1726: \begin{align*}

1727:  & a \circ -: \Hom^*_{\mathcal C}(Y^k,Y^i) \longrightarrow

1728:  \Hom^*_{\mathcal C}(Y^k,Y^{i+1}) && \text{for $k<i$, and} \\

1729:  & - \circ\, a: \Hom^*_{\mathcal C}(Y^{i+1},Y^l) \longrightarrow

1730:  \Hom^*_{\mathcal C}(Y^i,Y^l) && \text{for $l>i+1$}

1731: \end{align*}

1732: are isomorphisms. We then call $(Y^i,Y^{i+1})$ a {\bf matching pair} of

1733: dimension $n$. Let $C \in \Ob\, {\mathcal C}$ be the cone of $a$, which is

1734: well-defined up to isomorphism.

1735: 

1736: \begin{lemma} \label{th:matching-pair}

1737: $C$ is a spherical object of dimension $(n+1)$ in the sense of Definition

1738: \ref{def:spherical}.

1739: \end{lemma}

1740: 

1741: For future use we need to mention one more fact about $C$. Assume that

1742: ${\mathcal C} = D^b(\A)$ is the derived category of some directed

1743: $A_\infty$-category $\A$, with $\Ob\,\A = \{X^1,\dots,X^m\}$, and moreover that

1744: $(Y^1,\dots,Y^m)$ is obtained from $(X^1,\dots,X^m)$ by mutation, that is to

1745: say by applying the transformations \cite[Equations (6) and (7)]{seidel00} and

1746: their inverses, plus possibly shifting each object by some amount. In this

1747: situation the following holds:

1748: 

1749: \begin{lemma} \label{th:invariant}

1750: If $F: \A \rightarrow \A$ is a quasi-isomorphism which preserves the numbering

1751: of the objects, then $(D^bF)(C) \iso C$.

1752: \end{lemma}

1753: 

1754: The proof is a general nonsense argument. Consider first a general triangulated

1755: category ${\mathcal C}$ as before with an exact self-equivalence $G$, and

1756: define $Inv(G) \subset \Ob\,{\mathcal C}$ to be the class of objects $X$ such

1757: that $G(X) \iso X$. This has the following properties: (i) if $X$ lies in

1758: $Inv(G)$ then so does $X[\sigma]$ for any $\sigma \in \Z$, as well as any

1759: object isomorphic to $X$; (ii) if $X,Y \in Inv(G)$ and there is a unique

1760: nonzero map $X \rightarrow Y$ of degree zero, then the cone over it lies in

1761: $Inv(G)$; (iii) if $X,Y \in Inv(G)$ then $T_X(Y)$, $T_X'(Y) \in Inv(G)$. Here

1762: $T_X(Y)$ is as in \cite[Section 5]{seidel00}, and $T_X'(Y)$ is the object which

1763: fits into an exact triangle

1764: \[

1765: T_X'(Y) \rightarrow Y \xrightarrow{ev^\vee} \Hom^*_{\mathcal C}(Y,X)^\vee

1766: \otimes X \rightarrow T_X'(Y)[1],

1767: \]

1768: where ${ev}^\vee$ is the transpose of the evaluation $ev: \Hom^*_{\mathcal C}

1769: (Y,X) \otimes Y \rightarrow X$. Now specialize to ${\mathcal C} = D^b(\A)$, $G

1770: = D^bF$. Then $X^i \in Inv(G)$, and by (iii) the same holds for the objects

1771: $Y^i$ of any mutated exceptional collection. Applying (ii) shows that $C \in

1772: Inv(G)$.

1773: 

1774: \subsection{}

1775: Let $(E,\pi) = (E,\pi,\Omega,\Theta,J_0,j_0)$ be an exact Morse fibration over

1776: a compact base $S$, with $\dim\,E = 2n+2$. As explained in \cite[Section

1777: 3]{seidel00}, to any smooth path $c: [0;1] \rightarrow S$ with $c^{-1}(S^\crit)

1778: = \{0;1\}$ and $c'(0),c'(1) \neq 0$ one can associate two vanishing cycles

1779: $V_{c,0},V_{c,1}$, which are exact framed Lagrangian spheres in, say,

1780: $E_{c(1/2)}$. Now suppose that $c$ is an embedding, does not intersect

1781: $\partial S$, and that its vanishing cycles are isotopic, which means that

1782: there is a smooth family $I = (I_t)_{0 \leq t \leq 1}$ inside the space of all

1783: exact Lagrangian submanifolds of $E_{c(1/2)}$ joining $I_0 = V_{c,0}$ and $I_1

1784: = V_{c,1}$. Pairs $(c,[I])$ of this kind, where $[I]$ denotes the homotopy

1785: class of $I$ rel endpoints, are called {\bf matching paths}. The interest of

1786: this notion lies in a construction which I learnt from Donaldson. Roughly

1787: speaking (see below for a precise formulation) this produces from a matching

1788: path a submanifold of the total space $E$, which is Lagrangian in a suitable

1789: sense. The idea is easy to see in the case where $V_{c,0} = V_{c,1}$ and $I$ is

1790: the constant isotopy. By the definition of vanishing cycle, there are canonical

1791: $(n+1)$-dimensional balls $D_{c,\nu} \subset E$, $\nu = 0,1$, such that

1792: $\pi(D_{c,0}) = c([0;1/2])$, $\pi(D_{c,1}) = c([1/2;1])$, $\partial D_{c,\nu} =

1793: V_{c,\nu}$ and $\Omega|D_{c,\nu} = 0$. Under the assumption that $V_{c,0} =

1794: V_{c,1}$, the two balls fit together smoothly to give a closed submanifold of

1795: $E$. Fix a positive two-form $\beta$ on $S$. Then the base $\Omega^{(r)} =

1796: \Omega + r\cdot\pi^*\beta$ is symplectic for large positive $r$, and $D_{c,0}

1797: \cup D_{c,1}$ is Lagrangian with respect to it.

1798: 

1799: In general, rather than choosing some $r$, it seems more natural to work with

1800: the entire family of forms $\Omega^{(r)}$. This requires a suitably adapted

1801: notion of Lagrangian submanifold. Let $\Lambda \subset \Rgeq \times (E

1802: \setminus \partial E)$ be a $(n+2)$-dimensional submanifold such that the

1803: projection $\Lambda \rightarrow \Rgeq$ is a proper submersion; this is the same

1804: as a smooth family of closed $(n+1)$-submanifolds $\Lambda^{(r)} \subset E$,

1805: $\Lambda^{(r)} = \Lambda \cap (\{r\} \times E)$. For simplicity assume that

1806: $H^1(\Lambda;\R) = 0$. Such a $\Lambda$ is called eventually Lagrangian if

1807: $\Omega^{(r)}|\Lambda^{(r)} = 0$ for $r \gg 0$. We can now state precisely the

1808: nature of the construction mentioned above: for any matching pair $(c,[I])$ it

1809: gives an eventually Lagrangian submanifold $\Lambda_{(c,[I])}$, unique up to

1810: isotopy.

1811: 

1812: The first step is to trivialize $E$ over $c$, away from its endpoints, using

1813: symplectic parallel transport. The trivialization is the unique embedding

1814: $\Psi: (0;1) \times E_{c(1/2)} \rightarrow E$ which satisfies $\Psi(\{t\}

1815: \times E_{c(1/2)}) = E_{c(t)}$, $\Psi\,|\,\{1/2\} \times E_{c(1/2)} = \id$, and

1816: $\Psi^*\Omega = \Omega|E_{c(1/2)}$. Because $D_{c,0}$ and $D_{c,1}$ are

1817: likewise defined by parallel transport, they satisfy $\Psi^{-1}(D_{c,0}) =

1818: (0;1/2] \times V_{c,0}$, $\Psi^{-1}(D_{c,1}) = [1/2;1) \times V_{c,1}$. One can

1819: assume that the isotopy $(I_t)$ is constant near the endpoints. Then

1820: \[

1821:  A = \{x_0\} \cup \{x_1\} \cup

1822:      \bigcup_{t \in [0;1]} \Psi(\{t\} \times I_t),

1823: \]

1824: where $x_\nu \in E_{c(\nu)}$ are the critical points of $\pi$, is a closed

1825: submanifold of $E \setminus \partial E$ and contained in $\pi^{-1}c([0;1])$.

1826: 

1827: By dualizing the normal vector field $\partial I_t/\partial t$ one gets a

1828: function $h_t$ (unique up to a constant) on each $I_t$. Extend the $h_t$ to

1829: functions $H_t$ on $E_{1/2}$ supported in a neighbourhood of $I_t$; these may

1830: be chosen such that $H_t = 0$ for $t$ close to $0$ or $1$. One can then find an

1831: $\alpha \in \Omega^1(E)$ vanishing near $\partial E$, which satisfies

1832: $\alpha\,|\,\ker(D\pi) = 0$ and $\Psi^*\alpha = H_t\,dt$. With respect to the

1833: modified symplectic forms $\Omega^{(r)} + d\alpha$, $r \gg 0$, $A$ is a

1834: Lagrangian submanifold. Moser's Lemma shows that $\Omega^{(r)} + d\alpha$ and

1835: $\Omega^{(r)}$ are diffeomorphic, and the diffeomorphism can be made to depend

1836: smoothly on $r$. Applying these diffeomorphisms to $A$ yields a smooth family

1837: of submanifolds of $E$, parametrized by $r \gg 0$, which are Lagrangian with

1838: respect to $\Omega^{(r)}$. One extends this arbitrarily to small values of $r$

1839: to complete the definition of $\Lambda_{(c,[I])}$.

1840: 

1841: \begin{remarks}

1842: (i) We will often write $c$, $\Lambda_c$ instead of $(c,[I])$,

1843: $\Lambda_{(c,[I])}$, for the sake of brevity. This is fully justified only when

1844: $n = 1$: in that case each component of the space of exact Lagrangian circles

1845: on the surface $E_{c(1/2)}$ is simply-connected, so that $[I]$ contains no

1846: information.

1847: 

1848: (ii) By definition each $\Lambda^{(r)}_c \subset E$, $r \in \Rgeq$, is

1849: diffeomorphic to $A$. Since $A$ is glued together from two balls, it is a

1850: homotopy $(n+1)$-sphere, but it may not be the standard one. To get standard

1851: spheres from the construction, one has to impose the additional condition that

1852: the composition

1853: \[

1854: \xymatrix{

1855:  {S^n} \ar[r]^-{f_0} & {V_{c,0}} \ar[r] & {V_{c,1}} & \ar[l]_-{f_1} {S^n}

1856: }

1857: \]

1858: can be isotoped within $\Diff(S^n)$ to an element of $O(n+1)$. Here $f_0,f_1$

1859: are the framings of the vanishing cycles $V_{c,0},V_{c,1}$; and the middle

1860: arrow is the diffeomorphism determined, up to isotopy, by $(I_t)$.

1861: \end{remarks}

1862: 

1863: Assume now that no connected component of $S$ is closed, so that $\beta \in

1864: \Omega^2(S)$ is exact. One then defines the Floer cohomology of two eventually

1865: Lagrangian submanifolds $\Lambda,\Lambda' \subset \R^{\geq 0} \times E$ by

1866: \begin{equation} \label{eq:eventually}

1867: HF(\Lambda,\Lambda') = HF(\Lambda^{(r)},(\Lambda')^{(r)})

1868: \end{equation}

1869: where the right hand side is with respect to $\Omega^{(r)}$, for some $r \gg

1870: 0$. This makes sense because $\Omega^{(r)}$ is exact, and because there is a

1871: natural class of compatible almost complex structures on $E$ such that

1872: pseudo-holomorphic curves do not reach out to $\partial E$. Moreover, one can

1873: show that the choice of $r$ does not affect \eqref{eq:eventually}.

1874: 

1875: \subsection{}

1876: We will now explain the relation between the algebra and geometry introduced

1877: above. Let $(E,\pi)$ and $\delta_{E/D}$ be an exact Morse fibration over $D$

1878: with a relative Maslov map; the notation $M$ and $\delta_M$ is as usual. Let

1879: $c$ be a matching path. One can find an admissible choice of paths

1880: $(c_1,\dots,c_m)$ such that $c_i$ and $c_{i+1}$, for some $1 \leq i < m$, are

1881: as in Figure \ref{fig:matching}, and such that no other $c_j$ intersects $c$.

1882: In the corresponding distinguished basis $\Gamma = (L_1,\dots,L_m)$, $L_i$ and

1883: $L_{i+1}$ are isotopic. Now choose gradings $\widetilde{\Gamma} =

1884: (\tL_1,\dots,\tL_m)$ such that $\tL_i$ is isotopic to $\tL_{i+1}$, and let $\A

1885: = \Lag(\widetilde{\Gamma})$ be the directed Fukaya category. Then

1886: $(\tL_i,\tL_{i+1})$ is a matching pair in $D^b(\A)$. In fact one has

1887: \begin{align*}

1888: \Hom_{D^b(\A)}(\tL_i,\tL_{i+1}) & = H(hom_\A(\tL_i,\tL_{i+1})) \\ & =

1889: HF^*(\tL_i,\tL_{i+1}) \iso H^*(L_i;\Z/2)

1890: \end{align*}

1891: which proves \eqref{eq:matching}, and the other property is similarly easy. The

1892: informal idea is that the spherical object $C \in \Ob \, D^b(\A)$ associated to

1893: this matching pair ``represents'', in some sense, the eventually Lagrangian

1894: submanifold $\Lambda_c$. We will now make a concrete conjecture based on this

1895: philosophy.

1896: %

1897: \includefigure{matching}{matching.eps}{hb}%

1898: 

1899: Suppose that $c,c'$ are two matching paths. For each of them one can find an

1900: admissible choice of paths of the kind considered above, but in general these

1901: choices cannot be the same for both paths. Therefore one gets two distinguished

1902: bases $\Gamma,\Gamma'$ of vanishing cycles, and two directed Fukaya categories

1903: $\A,\A'$ together with spherical objects $C \in \Ob \, D^b(\A)$ and $C' \in \Ob

1904: \, D^b(\A')$, constructed from matching pairs of objects in $\A,\A'$ in the way

1905: which we have explained. We know of course that $D^b(\A)$ and $D^b(\A')$ are

1906: equivalent, but for the present purpose we need to make this slightly more

1907: precise. There is a chain $\A = \A_0,\A_1,\dots,\A_r = \A'$ of directed

1908: $A_\infty$-categories, of which each is related to the preceding one in one of

1909: four possible ways: either $\A_{i+1}$ is quasi-isomorphic to $\A_i$; or it is

1910: obtained from $\A_i$ by shifting the grading of the morphism spaces; or else

1911: $\A_{i+1} = c\A$ or $r\A$. In the three last cases there are canonical

1912: equivalences $D^b(\A_i) \iso D^b(\A_{i+1})$; in the first case we choose an

1913: equivalence induced by an arbitrary quasi-isomorphism. Composing all of them

1914: gives an equivalence $G: D^b(\A) \rightarrow D^b(\A')$.

1915: 

1916: \begin{conjecture} \label{th:induction}

1917: Let $\Lambda_c,\Lambda_{c'}$ be the eventually Lagrangian submanifolds

1918: associated to $c,c'$. Then

1919: \begin{equation} \label{eq:induction}

1920: HF(\Lambda_c,\Lambda_{c'}) \iso \bigoplus_{d \in \Z}

1921: \Hom^d_{D^b(\A')}(G(C),C').

1922: \end{equation}

1923: \end{conjecture}

1924: 

1925: One can refine the statement by introducing suitable graded Floer cohomology

1926: groups for $(\Lambda_c,\Lambda_{c'})$ but we prefer to skip this, since it does

1927: not involve any really new ideas.

1928: 

1929: The two admissible choices of paths which lead to $\Gamma,\Gamma'$ are not

1930: unique, and neither is the sequence of Hurwitz moves joining them; moreover, in

1931: the definition of $G$ we admitted arbitrarily chosen quasi-isomorphisms. It is

1932: implicitly part of the conjecture that the right hand side of

1933: \eqref{eq:induction} is independent of all these choices. The most

1934: doubtful-looking point is the bit about quasi-isomorphisms. Imagine for

1935: instance that $\A_i$ and $\A_{i+1}$ are directed Fukaya categories arising from

1936: the same Lagrangian configuration, but with different choices of almost complex

1937: structures. As mentioned in \cite[Section 6]{seidel00}, there is an explicit

1938: construction of quasi-isomorphisms between $\A_i$ and $\A_{i+1}$, using

1939: pseudo-holomorphic curves for a one-parameter family of choices of almost

1940: complex structures. One might think that for \eqref{eq:induction} to be

1941: correct, it would be necessary to choose the step between $D^b(\A_i)$ and

1942: $D^b(\A_{i+1})$ in the construction of $G$ to be such an analytically defined

1943: quasi-isomorphism. The answer to this objection is Lemma \ref{th:invariant},

1944: which says that choices of quasi-isomorphisms can never affect how objects like

1945: $C$ are mapped.

1946: 

1947: To see what Conjecture \ref{th:induction} does, consider the case $n = 1$. Then

1948: $M$ is a surface, so that $\A$ and $\A'$ can be determined combinatorially, and

1949: so can $G: \A \rightarrow \A'$ (this is only true thanks to the freedom of

1950: choosing quasi-isomorphisms, which is why we were insisting on it). On the

1951: other hand, the left hand side of \eqref{eq:induction} is the Floer cohomology

1952: of Lagrangian two-spheres in the symplectic four-manifold $(E,\Omega^{(r)})$,

1953: which is difficult or even impossible to compute directly.

1954: 

1955: \begin{remark} \label{th:zero-spheres}

1956: With some modifications, the discussion can be adapted to the trivial dimension

1957: $n = 0$. Here we only want to point out one example, namely the double branched

1958: cover $\pi: E \rightarrow D$ from Section \ref{sec:dimension-zero}c. Because

1959: the smooth fibre consists of two points, any path is a matching path, so that

1960: there is a multitude of one-dimensional spherical objects. This explains the

1961: geometric meaning of the objects $C_1,\dots,C_{2g} \in \Ob \, D^b(\A_g)$ which

1962: we wrote down earlier. By pushing the argument further, one can show that the

1963: zero-dimensional version of Conjecture \ref{th:induction} implies the algebraic

1964: formulae for geometric intersection numbers on $E$ which were derived, by a

1965: quite different argument, in \cite{khovanov-seidel98}.

1966: \end{remark}

1967: 

1968: %

1969: %--------------------------- BIBLIOGRAPHY ---------------------------

1970: %

1971: \providecommand{\bysame}{\leavevmode\hbox to3em{\hrulefill}\thinspace}

1972: \begin{thebibliography}{VC}

1973: \bibitem[VC]{seidel00}

1974: P.~Seidel, \emph{Vanishing cycles and mutation}, Proceedings of the 3rd

1975: European Congress of Mathematics (Barcelona, 2000), Birkh{\"a}user, to appear.

1976: 

1977: \bibitem{acampo00}

1978: N.~A'Campo, \emph{A combinatorial property of generic immersions of curves},

1979:   Preprint.

1980: 

1981: \bibitem{acampo73b}

1982: \bysame, \emph{Le nombre de lefschetz d'une monodromie}, Indag. Math.

1983:   \textbf{35} (1973), 113--118.

1984: 

1985: \bibitem{acampo75b}

1986: \bysame, \emph{Le groupe de monodromie du d{\'e}ploiement des singularit{\'e}s

1987:   isol{\'e}es de courbes planes {I}}, Math. Ann. \textbf{213} (1975), 1--32.

1988: 

1989: \bibitem{acampo98}

1990: \bysame, \emph{Real deformations and complex topology of plane curve

1991:   singularities}, Annales Fac. Sci. Toulouse \textbf{8} (1999), 5--23.

1992: 

1993: \bibitem{arnold-gusein-zade-varchenko}

1994: V.~I. Arnold, S.~M. Gusein-Zade, and A.~N. Varchenko, \emph{Singularities of

1995:   differentiable maps}, Birkh{\"a}user, 1988.

1996: 

1997: \bibitem{beilinson78}

1998: A.~Beilinson, \emph{Coherent sheaves on {$\mathbb{P}^n$} and problems of

1999:   linear algebra}, Funct. Anal. Appl. \textbf{12} (1978), 214--216.

2000: 

2001: \bibitem{bernstein-gelfand-ponomarev72}

2002: J.~Bernstein, I.~Gelfand, and V.~Ponomarev, \emph{Coxeter functors and

2003:   {G}abriel's theorem}, Russian Math. Surveys \textbf{28} (1973), 17--32.

2004: 

2005: \bibitem{bernstein-lunts94}

2006: J.~Bernstein and V.~Lunts, \emph{Equivariant sheaves and functors}, Lecture

2007:   Notes in Mathematics, vol. 1578, Springer, 1994.

2008: 

2009: \bibitem{bondal-polishchuk94}

2010: A.~Bondal and A.~Polishchuk, \emph{Homological properties of associative

2011:   algebras: the method of helices}, Russian Math. Izvestiya \textbf{42} (1994),

2012:   219--260.

2013: 

2014: \bibitem{cohen-jones-segal95}

2015: R.~L. Cohen, J.~D.~S. Jones, and G.~B. Segal, \emph{{F}loer's

2016:   infinite-dimensional {M}orse theory and homotopy theory}, The {F}loer

2017:   memorial volume (H.~Hofer, C.~Taubes, A.~Weinstein, and E.~Zehnder, eds.),

2018:   Progress in Mathematics, vol. 133, Birkh{\"a}user, 1995, pp.~297--325.

2019: 

2020: \bibitem{fukaya97}

2021: K.~Fukaya, \emph{Floer homology for three-manifolds with boundary {I}},

2022:   Preprint, 1997.

2023: 

2024: \bibitem{givental94}

2025: A.~Givental, \emph{Homological geometry and mirror symmetry}, Proceedings of

2026:   the International Congress of Mathematics, Z{\"u}rich, vol.~1,

2027:   Birkh{\"a}user, 1994, pp.~472--480.

2028: 

2029: \bibitem{gusein-zade74}

2030: S.~M. Gusein-Zade, \emph{Dynkin diagrams for singularities of functions of two

2031:   variables}, Functional Anal. Appl. \textbf{8} (1974), 295--300.

2032: 

2033: \bibitem{hori-iqbal-vafa00}

2034: K.~Hori, A.~Iqbal, and C.~Vafa, \emph{{$D$}-branes and mirror symmetry},

2035:   Preprint hep-th/0005247.

2036: 

2037: \bibitem{kapranov90}

2038: M.~M. Kapranov, \emph{{M}utations and {S}erre functors on constructive

2039:   bundles}, Functional Anal. Appl. \textbf{24} (1990), 155--156.

2040: 

2041: \bibitem{khovanov-seidel98}

2042: M.~Khovanov and P.~Seidel, \emph{Quivers, {F}loer cohomology, and braid group

2043:   actions}, Preprint Math.QA/0006056.

2044: 

2045: \bibitem{koenig-zimmermann98}

2046: S.~Koenig and A.~Zimmermann (eds.), \emph{Derived equivalences for group

2047:   rings}, Lecture notes in Math., vol. 1685, Springer, 1998.

2048: 

2049: \bibitem{le74}

2050: D.~T. L{\^e}, \emph{La monodromie n'a pas de points fixes}, J. Fac. Sci. Univ.

2051:   Tokyo Sect. IA \textbf{22} (1975), 409--427.

2052: 

2053: \bibitem{le-ono97}

2054: H.~V. L{\^e} and K.~Ono, \emph{Cup-length estimates for symplectic fixed

2055: points},

2056:   Contact and symplectic geometry (Cambridge, 1994), Cambridge Univ. Press,

2057:   1996, pp.~268--295.

2058: 

2059: \bibitem{penkava-schwarz95}

2060: M.~Penkava and A.~Schwarz, \emph{{$A_\infty$}-algebras and the cohomology of

2061:   moduli spaces}, Amer. Math. Soc. Translations, vol. 169, pp.~91--108,

2062:   American Mathematical Society, 1995.

2063: 

2064: \bibitem{schwarz96}

2065: M.~Schwarz, \emph{A quantum cup-length estimate for symplectic fixed points},

2066:   Invent. Math. \textbf{133} (1998), 353--397.

2067: 

2068: \bibitem{seidel96b}

2069: P.~Seidel, \emph{The symplectic {F}loer homology of a {D}ehn twist}, Math.

2070:   Research Lett. \textbf{3} (1996), 829--834.

2071: 

2072: \bibitem{seidel-thomas99}

2073: P.~Seidel and R.~Thomas, \emph{Braid group actions on derived categories of

2074:   coherent sheaves}, Preprint math.AG/0001043. To appear in {\em Duke Math.\

2075:   J}.

2076: 

2077: \bibitem{wajnryb83}

2078: B.~Wajnryb, \emph{A simple presentation for the mapping class group of an

2079:   orientable surface}, Israel J. Math. \textbf{45} (1983), 157--174.

2080: \end{thebibliography}

2081: \end{document}

2082: