1: \documentclass[11pt]{amsart}

2: \textwidth5.22in \textheight8.2in \raggedbottom

3: 

4: \usepackage{array}

5: \usepackage{float}

6: \usepackage{longtable} \setlongtables

7: \usepackage{hhline}

8: \usepackage{graphics}

9: \usepackage{color}

10: \usepackage{amssymb}

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

12: \CompileMatrices

13: \usepackage{epsfig}

14: \usepackage{euscript}

15: \usepackage{xspace}

16: \usepackage{ulem} \normalem

17: 

18: \parindent0em

19: \parskip1em

20: \renewcommand{\thesubsection}{\arabic{section}\alph{subsection}}

21: \renewcommand{\subsection}{\hspace{-\parindent}\refstepcounter{subsection}{\bf

22: (\arabic{section}\alph{subsection})}}

23: \newcommand{\subsectionstar}[1]{\hspace{-\parindent}{\bf #1.}}

24: \renewcommand{\proof}{{\sc Proof.}\xspace}

25: \numberwithin{equation}{section}

26: %

27: %------------- Theorem environments ----------------------

28: %

29: \newtheoremstyle{my}{1.5em}{0.5em}{\em}{}{\sc}{.}{0.5em}{}

30: \newtheoremstyle{mydef}{1.5em}{0.5em}{}{}{\sc}{.}{0.5em}{}

31: % #1 = name

32: % #2 = preskip

33: % #3 = postskip

34: % #4 = bodyfont

35: % #5 = noindent?

36: % #6 = headfont

37: % #7 = headpunct, e.g. "."

38: % #8 = labelsep (between label and statement}

39: % #9 = apparently overrides the whole header

40: 

41: \theoremstyle{my}

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

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

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

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

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

47: \newtheorem{sublemma}[thm]{Sublemma}

48: \newtheorem{addendum}[thm]{Addendum}

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

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

51: \newtheorem{assumptions}[thm]{Assumptions}

52: \newtheorem{assumption}[thm]{Assumption}

53: 

54: % \theoremstyle{mydef}

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

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

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

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

59: \newtheorem{notation}[thm]{Notation}

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

61: \newtheorem{claim}[thm]{Claim}

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

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

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

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

66: 

67: \newcommand{\acknowledgments}{{\em Acknowledgments.} }

68: 

69: %-----------------------------------------------------------------

70: 

71: 

72: \newcommand{\GAP}{%

73: \typeout{There's a gap here.}%

74: \begin{marginpar}

75:  {\bf gap}

76: \end{marginpar}%

77: }

78: 

79: \newcommand{\FIGUREGAP}[1]{

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

81: \setlength{\unitlength}{1cm} \framebox[10cm]{\begin{picture}(5,3)

82: \end{picture}}

83: \caption{#1 (missing)%

84: \label{#1}}

85: \end{center} \end{figure}}

86: 

87: \newcommand{\FIGURE}[3]{%

88: \begin{figure}[#3]

89: \begin{center}

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

91: \end{center}

92: \end{figure}}

93: 

94: \newenvironment{alphalist}%

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

96: \renewcommand{\labelenumi}{\theenumi}

97: \begin{enumerate}

98: \itemsep1em \leftmargin2em

99: }{\end{enumerate}}

100: 

101: \newenvironment{theoremlist}%

102: {\begin{list}{{\rm(\alph{enumi}) }}{\usecounter{enumi}

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

104: \renewcommand{\labelenumi}{\theenumi}

105: \leftmargin0cm \labelsep0cm \rightmargin0cm \parsep1em

106: \listparindent0em \itemsep-1em \topsep1em \parskip1em

107: \setlength{\labelwidth}{\fill}}} {\parskip0em \end{list}}

108: 

109: \newenvironment{normallist}%

110: {\renewcommand{\theenumi}{(\arabic{enumi})}

111: \renewcommand{\labelenumi}{\theenumi}

112: \begin{enumerate}}{\end{enumerate}}

113: 

114: \newenvironment{romanlist}%

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

116: \renewcommand{\labelenumi}{\theenumi}

117: \begin{enumerate} \parsep0cm

118: \itemsep0cm \parskip0cm}{\end{enumerate}}

119: 

120: \newenvironment{primelist}%

121: {\renewcommand{\theenumi}{(\arabic{enumi}')}

122: \renewcommand{\labelenumi}{\theenumi}

123: \begin{enumerate}}{\end{enumerate}}

124: 

125: %---------------------------- Standard math macros ---------------

126: 

127: \newcommand{\R}{\mathbb{R}}

128: \newcommand{\Z}{\mathbb{Z}}

129: \newcommand{\Q}{\mathbb{Q}}

130: \newcommand{\C}{\mathbb{C}}

131: \newcommand{\N}{\mathbb{N}}

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

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

134: 

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

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

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

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

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

140: \newcommand{\leftsc}{\langle}

141: \newcommand{\rightsc}{\rangle}

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

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

144: \newcommand{\suchthat}{\; : \;}

145: 

146: \newcommand{\id}{\mathrm{id}}

147: \newcommand{\ind}{\mathrm{ind}}

148: \newcommand{\re}{\mathrm{re}}

149: \newcommand{\im}{\mathrm{im}}

150: \renewcommand{\ker}{\mathrm{ker}}

151: \newcommand{\coker}{\mathrm{coker}}

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

153: \newcommand{\Hom}{\mathrm{Hom}}

154: \newcommand{\End}{\mathrm{End}}

155: 

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

157: \renewcommand{\o}{\omega}

158: \renewcommand{\O}{\Omega}

159: \newcommand{\Aut}{Aut}

160: \newcommand{\Ham}{Ham}

161: \newcommand{\Symp}{Symp}

162: \newcommand{\Diff}{Diff}

163: \newcommand{\Lag}{Lag}

164: 

165: %---------------- Introduction -----------------------------------

166: 

167: \newcommand{\Fuk}{{\mathcal F}}

168: 

169: %---------------- Sections 1-3: Algebraic preliminaries ----------

170: 

171: \newcommand{\Cat}{{\mathcal C}}

172: \newcommand{\A}{\mathcal A}

173: \newcommand{\B}{\mathcal B}

174: \newcommand{\T}{\mathcal T}

175: \newcommand{\F}{\mathcal F}

176: \newcommand{\G}{\mathcal G}

177: \newcommand{\M}{\mathcal M}

178: \newcommand{\NN}{\mathcal N}

179: \newcommand{\OO}{\mathcal O}

180: \renewcommand{\AA}{\mathfrak A}

181: \newcommand{\GG}{\mathfrak G}

182: \newcommand{\bimodA}{\A\text{--}\A}

183: \newcommand{\bimodB}{\B\text{--}\B}

184: \newcommand{\g}{\mathfrak g}

185: \newcommand{\semidirect}{\rtimes}

186: \newcommand{\inner}{\llcorner\,}

187: \newcommand{\ampersand}{\,\&\,}

188: \newcommand{\E}{\mathcal E}

189: 

190: \newcommand{\UU}{\mathfrak U}

191: \newcommand{\SHom}{\underline{Hom}}

192: \newcommand{\cech}[1]{\check{#1}}

193: \newcommand{\I}{\mathcal I}

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

195: 

196: %--------------- Sections 5-8: geometric preliminaries -----------

197: 

198: \newcommand{\lcm}{l.c.m.}

199: \newcommand{\LL}{{\mathfrak L}}

200: \newcommand{\orient}{\mathrm{or}}

201: \newcommand{\Iso}{Iso}

202: \newcommand{\alphagr}{\tilde{\alpha}}

203: \newcommand{\Lgr}{\tilde{L}}

204: \newcommand{\phigr}{\tilde{\phi}}

205: \newcommand{\psigr}{\tilde{\psi}}

206: \newcommand{\hgr}{\tilde{h}}

207: \newcommand{\Autgr}{\widetilde{\Aut}}

208: \newcommand{\taugr}{\tilde{\tau}}

209: \renewcommand{\Diff}{\text{\it Diff}\,}

210: 

211: \renewcommand{\O}{\mathcal{O}}

212: \newcommand{\K}{\mathcal{K}}

213: 

214: \newcommand{\red}{\mathrm{red}}

215: \newcommand{\crit}{\mathrm{crit}}

216: 

217: \newcommand{\Spin}{\$}

218: \newcommand{\br}{{\scriptscriptstyle \flat}}

219: \newcommand{\rbr}{{\scriptscriptstyle \sharp}}

220: \newcommand{\mbr}{{\scriptscriptstyle \natural}}

221: 

222: \newcommand{\trans}{\pitchfork}

223: \newcommand{\CC}{\mathfrak{C}}

224: \newcommand{\MM}{\mathfrak{M}}

225: \newcommand{\Fol}{\mathcal{F}}

226: 

227: %--------------------- Actual computation --------------------

228: 

229: \newcommand{\QQ}{\mathcal Q}

230: \newcommand{\U}{\mathcal U}

231: \newcommand{\One}{I}

232: \newcommand{\twist}[1]{\langle #1 \rangle}

233: 

234: %-------------------------------------------------------------

235: 

236: \begin{document}

237: \title[Homological mirror symmetry]{Homological mirror symmetry\\ for the quartic surface}

238: \author{Paul Seidel}

239: \date{This copy {\LaTeX}ed: \today}

240: \maketitle

241: 

242: \begin{quote}

243: {\bf Warning.} This preprint is in finished form, and should be

244: readable by itself. However, the argument uses certain general

245: properties of Fukaya categories, the proofs of which will appear in

246: \cite{seidel04} (under preparation). I apologize for this

247: unsatisfactory state of affairs.

248: 

249: \parskip0em

250: \tableofcontents

251: \end{quote}

252: %\input{todo}

253: 

254: \section{Introduction}

255: 

256: This paper deals with a special case of Kontsevich's ``homological

257: mirror symmetry'' conjecture \cite{kontsevich94}. Before formulating

258: the precise statement, it is necessary to introduce the relevant

259: coefficient rings.

260: 

261: \begin{notation}

262: Let $\Lambda_\N = \C[[q]]$ be the ring of formal power series in one

263: variable. We denote by $\Lambda_\Z$ its quotient field, obtained by

264: formally inverting $q$. By adjoining roots $q^{1/d}$ of all orders to

265: $\Lambda_\Z$, one obtains its algebraic closure $\Lambda_\Q$, known

266: in symplectic geometry as the rational Novikov field. Elements of

267: this field are formal series

268: \[

269:  f(q) = \sum_m a_m q^m

270: \]

271: where $m$ runs over all numbers in $(1/d)\Z \subset \Q$ for some $d

272: \geq 1$ which depends on $f$, and $a_m \in \C$ vanishes for all

273: sufficiently negative $m$. Geometrically, one should think of

274: $\Lambda_\N$ as a small disc, of $\Lambda_\Z$ as the punctured disc,

275: and of $\Lambda_\Q$ as its universal cover. Consider the semigroup

276: $End(\Lambda_\N)$ of $q$-adically continuous $\C$-algebra

277: endomorphisms of $\Lambda_\N$. More concretely, these are

278: substitutions

279: \[

280:  \psi^* : q \longmapsto \psi(q)

281: \]

282: with $\psi \in q \C[[q]]$. Each nonzero $\psi$ extends to an

283: endomorphism of the quotient field, and the $q$-adically continuous

284: Galois group of $\Lambda_\Z/\C$ is the group of invertibles

285: $End(\Lambda_\N)^\times$, consisting of those $\psi$ with $\psi'(0)

286: \neq 0$. Moreover, the Galois group of $\Lambda_\Q/\Lambda_\Z$ is the

287: profinite group $\hat{\Z}$, whose topological generator $\hat{1}$

288: takes $q^{1/d}$ to $e^{2\pi i/d} q^{1/d}$.

289: \end{notation}

290: 

291: On the symplectic side of mirror symmetry, take any smooth quartic

292: surface $X_0 \subset \CP{3}$, with its standard symplectic structure.

293: To this we associate a triangulated category linear over

294: $\Lambda_\Q$, the split-closed derived Fukaya category

295: $D^\pi\Fuk(X_0)$, defined using Lagrangian submanifolds of $X_0$ and

296: pseudo-holomorphic curves with boundary on them. The use of formal

297: power series to take into account the area of pseudo-holomorphic

298: curves is a familiar device. $\Lambda_\Q$ appears because we allow

299: only a certain kind of Lagrangian submanifolds, namely the rational

300: ones (there is also another restriction, vanishing of Maslov classes,

301: which is responsible for making the Fukaya category $\Z$-graded).

302: 

303: On the complex side, we start with the quartic surface in ${\mathbb

304: P}^3_{\Lambda_\Q}$ defined by

305: \begin{equation} \label{eq:fermat}

306: y_0 y_1 y_2 y_3 + q(y_0^4 + y_1^4 + y_2^4 + y_3^4) = 0.

307: \end{equation}

308: Note that here $q$ is a ``constant'' (an element of the ground

309: field). The group $\Gamma_{16} = \{ [diag(a_0,a_1,a_2,a_3)] \suchthat

310: a_k^4 = 1, \; a_0a_1a_2a_3 = 1 \} \subset PSL(V)$, $\Gamma_{16} \iso

311: \Z/4 \times \Z/4$, acts on this surface in the obvious way. We denote

312: by $Z_q^*$ the unique minimal resolution of the quotient orbifold.

313: Non-algebraic-geometers may wish to remind themselves that

314: $\Lambda_\Q$ is an algebraically closed field of characteristic zero,

315: so that all the geometric constructions like minimal resolutions go

316: through as in the familiar case of $\C$. In fact, $Z_q^*$ could

317: already be defined over the smaller field $\Lambda_\Z$, since the

318: fixed points of the action that need to be resolved are

319: $\Lambda_\Z$-points. The associated category is the bounded derived

320: category of coherent sheaves, $D^bCoh(Z_q^*)$, which is also a

321: triangulated category linear over $\Lambda_\Q$.

322: 

323: \begin{theorem} \label{th:main}

324: There is a $\psi \in End(\Lambda_\N)^\times$ and an equivalence of

325: triangulated categories,

326: \[

327: D^\pi\Fuk(X_0) \iso \hat{\psi}^*D^bCoh(Z_q^*).

328: \]

329: \end{theorem}

330: 

331: Here $\hat\psi$ is a lift of $\psi$ to an automorphism of

332: $\Lambda_\Q$, which we use to change the $\Lambda_\Q$-module

333: structure of $D^bCoh(Z_q^*)$, by letting $\hat\psi^*f$ act instead of

334: $f$. The theorem says that the outcome of this ``base change'' is

335: equivalent to $D^\pi\Fuk(X_0)$. The choice of lift $\hat\psi$ is

336: irrelevant because $D^bCoh(Z_q^*)$ carries a $\hat{\Z}$-action (and

337: so does $D^\pi\Fuk(X_0)$ by \cite{fukaya03}).

338: 

339: The function $\psi$ which occurs in the theorem is unknown at present

340: (or rather, we cannot verify that it agrees with the standard

341: ``mirror map''). Nevertheless, certain consequences can already be

342: drawn from the result as given. For instance, let $\mathcal{M}^*$ be

343: the classifying space (or moduli stack, depending on the reader's

344: preference) for smooth $K3$ surfaces equipped with an ample

345: cohomology class $A$ whose square is $A \cdot A = 4$, and with a

346: choice of nonzero holomorphic two-form. There is a canonical

347: homomorphism $\pi_1(\mathcal{M}^*) \rightarrow \pi_0(Aut(X_0))$,

348: which is defined by symplectic monodromy of families of $K3$ surfaces

349: equipped with K{\"a}hler forms in the class $A$. On the other hand,

350: since $D^\pi\Fuk(X_0)$ is a symplectic invariant, we have a map

351: $\pi_0(Aut(X_0)) \rightarrow Aut(D^\pi\Fuk(X_0)/\Lambda_\Q)/[2]$,

352: where now $Aut$ on the right hand side is the group of isomorphism

353: classes of $\Lambda_\Q$-linear autoequivalences, and we divide by the

354: subgroup generated by even translations. Actually, the monodromy maps

355: obtained from loops in ${\mathcal M}^*$ are naturally graded, so the

356: shift ambiguity can be removed, and we get a homomorphism

357: \begin{equation} \label{eq:alg-monodromy}

358: \pi_1(\mathcal{M}^*) \longrightarrow Aut(D^bCoh(Z_q^*)/\Lambda_\Q).

359: \end{equation}

360: This had been postulated by several people, starting with Kontsevich

361: himself. Interestingly, even though both $\mathcal{M}^*$ and

362: $D^bCoh(Z_q^*)$ are objects from algebraic geometry, there seems to

363: be no definition of the map which avoids symplectic geometry and

364: mirror symmetry. Previous work on Dehn twists \cite{seidel-thomas99}

365: shows that \eqref{eq:alg-monodromy} is highly nontrivial, but the

366: exact kernel and cokernel remain unknown. Another direct implication

367: of Theorem \ref{th:main} is the equality of certain simpler

368: invariants, such as the Grothendieck groups

369: \[

370: K_0(D^\pi\Fuk(X_0)) \iso K_0(D^b Coh(Z_q^*)) = K_0(Z_q^*).

371: \]

372: Every object of the Fukaya category (in particular, every oriented

373: Lagrangian sphere in $X_0$) determines a class in

374: $K_0(D^\pi\Fuk(X_0))$, and $\pi_0(Aut(X_0))$ acts linearly on this

375: group. On the other hand $K_0(Z_q^*) \otimes \Q \iso CH_*(Z_q^*)

376: \otimes \Q$ by Grothendieck's theorem, so it contains $CH_0(Z_q^*)

377: \otimes \Q$ which is quite an interesting object.

378: 

379: Having stated our result, it is maybe time to review briefly some

380: other work on homological mirror symmetry. The first case to be

381: considered was that of elliptic curves, where Polishchuk and Zaslow

382: \cite{polishchuk-zaslow98} proved a result roughly similar to Theorem

383: \ref{th:main}, with an explicit mirror map $\psi$ but not taking into

384: account the triangulated structure (subsequent work on Massey

385: products \cite{polishchuk98} largely makes up for this deficiency).

386: There are many beautiful and deep results on abelian varieties

387: \cite{fukaya02b, kontsevich-soibelman00, fukaya02, nishinou03}, but

388: none of them comes close to describing the entire Fukaya category. A

389: common thread of all these papers is to establish an explicit

390: bijection between objects on both sides (Lagrangian submanifolds and

391: sheaves). The prevailing view is that this correspondence should be

392: set up as a Fourier-Mukai type transform associated to a

393: Strominger-Yau-Zaslow torus fibration. This is the approach of

394: \cite{kontsevich-soibelman00}, in the situation where the SYZ

395: fibration has no singularities, and where the Lagrangian submanifolds

396: are supposed to be transverse to the fibres. Both restrictions are

397: serious ones from the point of symplectic topology: work on removing

398: them has started, see in particular \cite{fukaya02b}, but it may be

399: some time before a complete thery emerges.

400: 

401: Actually, this entire circle of ideas is of little importance for our

402: proof (although the SYZ conjecture does serve as a source of

403: intuition in some places). We rely instead on techniques from

404: symplectic topology and homological algebra, which are maybe closer

405: to the spirit of Kontsevich's original paper. The computation on the

406: algebraic geometry side relies on a classical idea of Beilinson,

407: together with a Massey product computation which the author learned

408: from \cite{douglas-govindarajan-jayaraman-tomasiello01}. On the

409: symplectic side, where the overwhelming majority of the work is

410: invested, there are three steps, namely (slightly simplified, and in

411: reverse logical order):

412: \begin{itemize} \itemsep1em

413: 

414: \item

415: First, we prove that $D^\pi\Fuk(X_0)$ can be entirely reconstructed

416: from the full $A_\infty$-subcategory of $\Fuk(X_0)$ consisting of a

417: particular set of 64 Lagrangian two-spheres, which are vanishing

418: cycles for the standard Fermat pencil (the choice of these cycles is

419: best explained by physics considerations, compare for instance

420: \cite{hori-iqbal-vafa00}). The necessary ``split-generation''

421: argument resembles the ones in \cite{seidel04}, with some added

422: geometric input concerning the monodromy around the singular fibre at

423: infinity.

424: 

425: \item

426: By construction, all 64 vanishing cycles lie in an affine

427: Zariski-open subset $M_0 \subset X_0$. Following a proposal from

428: \cite{seidel02} the relation between the Fukaya categories

429: $\Fuk(M_0)$ and $\Fuk(X_0)$ can be formulated in terms of the

430: deformation theory of $A_\infty$-structures. This recourse to

431: abstract deformation theory is the reason why we cannot determine

432: $\psi$ explicitly.

433: 

434: \item

435: Having reduced our problem to a finite collection of Lagrangian

436: two-spheres in an affine $K3$ surface, we now apply a general

437: dimensional induction argument from \cite{seidel04} to compute the

438: relevant subcategory of $\Fuk(M_0)$. This starts with an application

439: of Picard-Lefschetz theory in one dimension lower than before,

440: fibering our $K3$ surface by a pencil of curves.

441: \end{itemize}

442: 

443: The last-mentioned argument involves a certain amount of computation,

444: of a combina\-torial (drawing loops on surfaces, and counting

445: polygons) and homological (determining morphisms between certain

446: twisted complexes in $A_\infty$-categories) nature. Although the

447: algorithm is simple, the computations themselves are not particularly

448: enlightening, and in fact a computer was used to do parts of them.

449: The reason why this step in the proof is less transparent is that,

450: unlike the previous two, it has no known meaning in terms of the

451: mirror. This is partially due to the fact that the fibres of our

452: pencil are curves of genus $3$, hence outside the scope of mirror

453: symmetry.

454: 

455: In principle, the approach taken here is fairly general. Suppose for

456: instance that one wanted to adapt it to the quintic threefold case.

457: The deformation theoretic part actually becomes easier since, unlike

458: the situation for $K3$s, there are no noncommutative deformations of

459: the mirror variety. On the other hand, the definition of the Fukaya

460: category is much more complicated \cite{fooo}, and one would have to

461: generalize the twist exact sequence \cite{seidel01} accordingly, but

462: these are on the whole technical issues. There are also some

463: elementary geometric complications, which arise where we have relied

464: on intuition provided by the SYZ conjecture. The final obstacle is in

465: the computational part, where the procedure remains precisely the

466: same but the complexity rises by a factor of at least 10. This raises

467: serious practical concerns, but it may still be feasible, or one can

468: try to find a more conceptual way of carrying it out (ideas of

469: Moishezon might prove helpful, see the comments and computations in

470: \cite{auroux-donaldson-katzarkov-yotov01}).

471: 

472: The paper is structured as follows. In the first two parts we

473: assemble the basic algebraic (Sections

474: \ref{sec:generators}--\ref{sec:sheaves}) and geometric (Sections

475: \ref{sec:geometry}--\ref{sec:induction}) techniques. Much of this is

476: not really new, but the application to mirror symmetry often results

477: a change of perspective, which means that it is hard to find the

478: theorems stated elsewhere in the form in which we need them. The

479: level of generality varies somewhat: we have tried to include

480: generalizations which were reasonably easy or natural, and to exclude

481: those which are either irrelevant to homological mirror symmetry, or

482: which would require a significant amount of additional techniques or

483: terminology. The last part of the paper (Sections

484: \ref{sec:4-64}--\ref{sec:computation}) contains the argument which is

485: really specific to the quartic surface, including figures and tables

486: which list the results of the computational step. The aim of that was

487: to give enough data so that the reader desirous of verifying the

488: computation independently could break up that process into manageable

489: steps.

490: 

491: \acknowledgments This paper could not have been written without the

492: assistance of Michael Douglas, who showed me the ``quiver

493: presentation'' of the derived category of coherent sheaves which

494: appears as the algebras $Q_4$ and $Q_{64}$ in this paper, and also

495: the crucial Massey product formula (Lemma \ref{th:massey}) from

496: \cite{douglas-govindarajan-jayaraman-tomasiello01}. Maxim Kontsevich

497: helped me in many ways, one of them being the Hochschild cohomology

498: computations for semidirect products (Lemma \ref{th:twisted-hkr}).

499: Simon Donaldson introduced me to the notion of matching cycle, on

500: which the main dimensional induction procedrure is based. To all of

501: them, thanks! ETH Z\"urich invited me to give a lecture course on

502: Fukaya categories, and provided a responsive audience on which parts

503: of this paper were tested. The following computer programs have been

504: used: {\sc singular} for elimination theory, {\sc mathematica} for

505: visualization, and {\sc maple} for the homological algebra part.

506: 

507: 

508: \section{$A_\infty$-categories\label{sec:generators}}

509: 

510: This section is designed to serve as quick reference for the basic

511: algebraic terminology, for the benefit of readers with a geometry

512: background. The main construction is that of the bounded derived

513: category of a $A_\infty$-category, following Kontsevich

514: \cite{kontsevich94}. We then consider the idempotent (Karoubi, or

515: split-closed) completion, and the corresponding notion of

516: split-generators. A brief look at twist functors and exceptional

517: collections completes the tour. Some references for more in-depth

518: treatments of various matters touched on here are \cite{kontsevich98,

519: keller99, lefevre, rudakov90, fooo, fukaya00, seidel04}.

520: 

521: \subsection{}

522: A non-unital $A_\infty$-category $\Cat$ consists of a set (all our

523: categories are small) of objects $Ob\,\Cat$, together with a graded

524: $\C$-vector space $hom_\Cat(X_0,X_1)$ for any two objects, and

525: multilinear maps

526: \begin{equation} \label{eq:mu}

527:  \mu^d_\Cat: hom_\Cat(X_{d-1},X_d) \otimes \dots \otimes

528:  hom_\Cat(X_0,X_1) \longrightarrow hom_\Cat(X_0,X_d)[2-d]

529: \end{equation}

530: for all $d \geq 1$ and all $(d+1)$-tuples of objects

531: $(X_0,\dots,X_d)$, satisfying the $A_\infty$-associativity equations

532: \begin{equation} \label{eq:ainfty}

533:  \sum_{e,i} (-1)^\ast \mu_\Cat^{d-e+1}(a_d,\dots,

534:  a_{i+e+1},\mu_\Cat^e(a_{i+e},\dots,a_{i+1}),a_i,\dots,a_1) = 0,

535: \end{equation}

536: where $\ast = |a_1|+\dots+|a_i|-i$ (the decreasing numbering of the

537: $a_k$ is the result of trying to bring together various standard

538: conventions). The underlying (non-unital) cohomological category

539: $H(\Cat)$ has the same objects as $\Cat$, with

540: \[

541: Hom^*_{H(\Cat)}(X_0,X_1) = H^*(hom_{\Cat}(X_0,X_1),\mu^1_\Cat)

542: \]

543: and composition induced by the cochain level map $a_2a_1 =

544: (-1)^{|a_1|}\mu^2_\Cat(a_2,a_1)$. There is also the full subcategory

545: $H^0(\Cat)$ which retains only the morphisms of degree zero. One says

546: that $\Cat$ is {\em c-unital} (cohomologically unital) if $H^0(\Cat)$

547: has identity morphisms. It is true, but by no means obvious, that

548: this is a good class of $A_\infty$-categories to work with: the

549: proofs of many properties rely on the fact that c-unitality turns out

550: to be equivalent to the less flexible but more visibly well-behaved

551: notion of strict unit \cite{lefevre,seidel04}. {\em From now on, all

552: $A_\infty$-categories are assumed to be c-unital}. Similarly, when

553: considering $A_\infty$-functors $\F: \Cat \rightarrow {\mathcal D}$,

554: we always require that $H(\F): H(\Cat) \rightarrow H({\mathcal D})$

555: be unital (carries identity morphisms to identity morphisms). $\F$ is

556: called a quasi-equivalence if $H(\F)$ is an equivalence. One can

557: prove that then, there is a $\G: {\mathcal D} \rightarrow \Cat$ such

558: that $H(\G)$ is an inverse equivalence to $H(\F)$.

559: 

560: \begin{remark}

561: A linear category with one object is the same as an associative

562: algebra, and similarly, an $A_\infty$-category with one object is an

563: $A_\infty$-algebra. Slightly more generally, given a category $C$

564: with only finitely many objects $X_1,\dots,X_m$, one can form the

565: {\em total morphism algebra}

566: \[

567:  A = \bigoplus_{j,k} Hom_C(X_j,X_k)

568: \]

569: which is linear over the semisimple ring $R_m = \C \oplus \dots

570: \oplus \C$ ($m$ summands). Again, the same holds for

571: $A_\infty$-categories versus $R_m$-linear $A_\infty$-algebras, and we

572: will use that frequently. Our notation will be that $u_1,\dots,u_m$

573: are the orthogonal idempotents in $R_m$, so $u_k A u_j =

574: Hom_C(X_j,X_k)$.

575: \end{remark}

576: 

577: \begin{remark}

578: For concreteness, we stick to the ground field $\C$ for our

579: exposition of the basic algebraic theory, but any other field would

580: do just as well (later on though, we will make use of deformation

581: theoretic ideas which work best in characteristic $0$). Defining

582: $A_\infty$-categories over a ring is a more difficult issue, and the

583: only simple solution is to assume that all the $hom$ spaces are

584: projective modules. That of course holds automatically for the

585: semisimple rings $R_m$. The only other ring of interest to us is

586: $\Lambda_\N$, which requires special treatment because of its

587: topology. Recall that for a $\Lambda_\N$-module $V_q$ which is

588: $q$-adically complete, the following conditions are equivalent: (1)

589: $V_q$ is torsion-free, which means that multiplication by $q$ is an

590: injective endomorphism; (2) $V_q$ is topologically free, which means

591: that projection $V_q \rightarrow V$ to the vector space $V = V_q

592: \otimes_{\Lambda_q} \C = V_q/qV_q$ can be lifted to a topological

593: isomorphism $V_q \rightarrow V \hat\otimes \Lambda_\N$. A {\em

594: topological $A_\infty$-algebra} over $\Lambda_\N$ consists of a

595: graded $\Lambda_\N$-module $\A_q$ such that each degree $k$ piece is

596: complete and torsion-free. If $H(\A_q)$ is torsion-free, it is

597: automatically also complete.

598: %

599: %, together with maps $\mu^d_{\A_q}$ which are $q$-adically

600: % continuous. This condition is automatic!!!

601: \end{remark}

602: 

603: \subsection{}

604: The twisted complex construction embeds $\Cat$ into a larger

605: $A_\infty$-category which is closed under (appropriately understood)

606: mapping cones. This can be broken into two steps:

607: \begin{itemize}

608: \item[(1)]

609: The additive enlargement $\Sigma\Cat$ has objects which are formal

610: sums

611: \[

612: X = \bigoplus_{f \in F} X_f[\sigma_f],

613: \]

614: where $F$ is some finite set, $X_f \in Ob\,\Cat$, and $\sigma_f \in

615: \Z$. The morphisms between any two objects are

616: \[

617:  hom_{\Sigma\Cat}\Big(\bigoplus_{f \in F} X_f[\sigma_f],

618:  \bigoplus_{g \in G} Y_g[\tau_g]\Big) =

619:  \bigoplus_{f,g} hom_{\Cat}(X_f,Y_g)[\tau_g-\sigma_f],

620: \]

621: and the compositions are defined using those of $\Cat$ and the

622: obvious ``matrix multiplication'' rule, with additional signs put in

623: as follows: for $a_1 \in hom(X_0[\sigma_0],X_1[\sigma_1]), \dots, a_d

624: \in hom(X_{d-1}[\sigma_{d-1}],X_d[\sigma_d])$,

625: \[

626:  \mu^d_{\Sigma\Cat}(a_d,\dots,a_1) = (-1)^{\sigma_0}

627:  \mu^d_\Cat(a_d,\dots,a_1).

628: \]

629: 

630: \item[(2)]

631: A twisted complex is a pair $(C,\delta)$ consisting of $C \in

632: Ob\,\Sigma\Cat$ and $\delta \in hom^1(C,C)$, with the ``upper

633: triangularity'' property that the indexing set $F$ for $C =

634: \bigoplus_{f \in F} C_f$ can be ordered in such a way that all

635: components $\delta_{fg}$ with $f \geq g$ are zero, and subject to the

636: generalized Maurer-Cartan equation

637: \[

638:  \mu^1_{\Sigma\Cat}(\delta) + \mu^2_{\Sigma\Cat}(\delta,\delta) + \dots = 0.

639: \]

640: Upper triangularity says that this is a finite sum, hence the

641: condition makes sense. We define an $A_\infty$-category $Tw\Cat$ of

642: which the $(C,\delta)$ are the objects. The spaces of morphisms are

643: the same as for $\Sigma\Cat$, but composition involves the $\delta$s:

644: for $a_k \in hom((C_{k-1},\delta_{k-1}),(C_k,\delta_k))$,

645: \[

646:  \mu^d_{Tw\Cat}(a_d,\dots,a_1) = \sum_{j_0,\dots,j_d \geq 0}

647:  \mu^d_{\Sigma\Cat}(\overbrace{\delta_d,\dots,\delta_d}^{j_d},a_d,

648:  \overbrace{\delta_{d-1},\dots,\delta_{d-1}}^{j_{d-1}},\dots).

649: \]

650: \end{itemize}

651: The c-unitality of $\Cat$ implies that of $Tw\Cat$. The underlying

652: cohomological category in degree zero, which means we retain only the

653: $H^0$ of the morphisms spaces, is (somewhat improperly) called the

654: {\em bounded derived category} $D^b(\Cat) = H^0(Tw\Cat)$. Given two

655: twisted complexes $(C_k,\delta_k)$, $k = 0,1$, and a morphism $a \in

656: hom^0(C_0,C_1)$ such that $\mu^1_{Tw\Cat}(a) = 0$, one can define the

657: mapping cone as usual:

658: \[

659:  Cone(a) = \Big\{C_0 \stackrel{a}{\rightarrow} C_1\Big\} \stackrel{\text{def}}{=}

660:  \Big(C_0[1] \oplus C_1, \begin{pmatrix} \delta_0 & 0 \\ a &

661:  \delta_1 \end{pmatrix}\Big)

662:  \in Ob\,Tw\,\Cat.

663: \]

664: The isomorphism class of this in the derived category depends only on

665: the cohomology class $\alpha = [a] \in Hom_{D^b(\Cat)}(C_0,C_1)$,

666: hence we will sometimes write sloppily $Cone(\alpha)$. Declare {\em

667: exact triangles} in the derived category to be those which are

668: isomorphic to the standard mapping cone triangles $C_0 \rightarrow

669: C_1 \rightarrow Cone(a) \rightarrow C_0[1]$. This makes $D^b(\Cat)$

670: into a triangulated category. It contains $H^0(\Cat)$ as a full

671: subcategory, and the objects of that subcategory {\em generate}

672: $D^b(\Cat)$ in the usual triangulated sense, that is, any object of

673: $D^b(\Cat)$ can be obtained from objects of $H^0(\Cat)$ by successive

674: mapping cones and isomorphism. However, this does not mean that

675: $H^0(\Cat)$ determines $D^b(\Cat)$. For instance, triangulated

676: categories have intrinsically defined Massey product operations

677: \cite{gelfand-manin}, and in our case these involve the higher order

678: $\mu^d_\Cat$.

679: 

680: The idempotent completion of an arbitrary linear category has objects

681: which are pairs $(X,p)$, where $X$ is an object of the original

682: category and $p \in Hom(X,X)$ an idempotent endomorphism. One thinks

683: of this as the ``direct summand $im(p)$'', and so

684: $Hom((X_0,p_0),(X_1,p_1)) = p_1 Hom(X_0,X_1) p_0$. It is proved in

685: \cite{balmer-schlichting01} that the idempotent completion of a

686: triangulated category is again triangulated, in a unique way which

687: makes the inclusion $X \mapsto (X,id_X)$ into an exact functor. The

688: result of applying this construction to the bounded derived category

689: of an $A_\infty$-category is called the {\em split-closed derived

690: category} $D^\pi(\Cat)$. Suppose that we have a set of objects in an

691: idempotent complete triangulated category; these are called {\em

692: split-generators} if one can obtain any object from them by

693: successive mapping cones, splitting off direct summands, and

694: isomorphism.

695: 

696: Here are some obvious but useful facts about derived categories:

697: 

698: \begin{lemma}

699: If two $A_\infty$-categories are quasi-equivalent, their derived

700: categories $D^b$ and $D^\pi$ are equivalent triangulated categories.

701: \end{lemma}

702: 

703: \proof Any $A_\infty$-functor $\F: \Cat \rightarrow {\mathcal D}$

704: admits a canonical extension $Tw\F: Tw\Cat \rightarrow Tw{\mathcal

705: D}$, and $D^b\F = H^0(Tw\F): D^b(\Cat) \rightarrow D^b({\mathcal D})$

706: is exact (it takes cones to cones). Suppose that $\F$ is a

707: quasi-equivalence, and consider the subcategory of $D^b(\Cat)$

708: consisting of those objects $X$ such that $D^b\F$ maps $Hom^*(X,Y)$

709: isomorphically to $Hom^*(D^b\F(X),\F(Y))$ for all $Y \in

710: Ob\,H^0(\Cat)$. By assumption this subcategory contains $H^0(\Cat)$,

711: and it is a standard exercise to show that it is triangulated, so it

712: is the whole of $D^b(\Cat)$. Repetition of the argument proves that

713: $D^b\F$ is full and faithful. Its image is a triangulated subcategory

714: of $D^b({\mathcal D})$ which, up to isomorphisms, contains all

715: objects of $H^0({\mathcal D})$. Since these are generators for

716: $D^b({\mathcal D})$, we conclude that $D^b(\F)$ is an equivalence.

717: The corresponding statement for $D^\pi$ follows immediately. \qed

718: 

719: \begin{lemma} \label{th:generating-subcategory}

720: Let $\Cat' \subset \Cat$ a full $A_\infty$-subcategory such that the

721: set of all objects of $\Cat'$ split-generates $D^\pi(\Cat)$. Then

722: $D^\pi(\Cat') \iso D^\pi(\Cat)$ as triangulated categories.

723: \end{lemma}

724: 

725: \proof Inclusion induces a full and faithful functor $D^\pi(\Cat')

726: \rightarrow D^\pi(\Cat)$, the image of which contains the objects of

727: $H^0(\Cat')$ and is closed under cones and idempotent splittings.

728: \qed

729: 

730: Combining the two previous Lemmas, we have

731: 

732: \begin{lemma} \label{th:derived-equivalence}

733: Let $\Cat,{\mathcal D}$ be two $A_\infty$-categories. Suppose that

734: they contain full $A_\infty$-subcategories $\Cat' \subset \Cat,

735: {\mathcal D}' \subset {\mathcal D}$ which are quasi-equivalent to

736: each other. If the objects of $\Cat',{\mathcal D}'$ split-generate

737: $D^\pi(\Cat)$ respectively $D^\pi({\mathcal D})$, then these two

738: derived categories are equivalent. \qed

739: \end{lemma}

740: 

741: \subsection{}

742: Let $C$ be a $\C$-linear triangulated category, and $X,Y$ two objects

743: such that $Hom_C^*(X,Y) = \bigoplus_{k \in \Z} Hom_C(X,Y[k])$ is

744: finite-dimensional. Then one can form two new objects $T_X(Y)$,

745: $T_Y^\vee(X)$ which fit into exact triangles

746: \[

747: \begin{aligned}

748:  & T_X(Y)[-1] \rightarrow Hom_C^*(X,Y) \otimes X \xrightarrow{ev} Y \rightarrow T_X(Y), \\

749:  & T_Y^\vee(X) \rightarrow X \xrightarrow{ev^\vee} Hom^*_C(X,Y)^\vee \otimes Y \rightarrow

750:  T_Y^\vee(X)[1].

751: \end{aligned}

752: \]

753: The notation means that we choose a homogeneous basis $\{f_k\}$ of

754: $Hom^*(X,Y)$, set $Hom^*(X,Y) \otimes X = \bigoplus_k X[-deg(f_k)]$,

755: and take the evaluation map $ev = \bigoplus_k f_k[-k]$. The dual map

756: $ev^\vee$ is defined in the same way. These algebraic {\em twist} and

757: {\em untwist} operations are well-defined up to non-unique

758: isomorphism, and the isomorphism type of the new objects depends only

759: on those of $X$ and $Y$.

760: 

761: The construction becomes somewhat more functorial in the case where

762: $C = D^b(\Cat)$ is the bounded derived category of an

763: $A_\infty$-category $\Cat$ such that (1) the cochain level morphism

764: space $hom_{Tw\Cat}(X,Y)$ is itself finite-dimensional, and (2) $X$

765: has a strict cochain level identity morphism $e_X$. One can then form

766: the tensor product $hom(X,Y) \otimes X \in Ob\,Tw(\Cat)$, which is a

767: finite direct sum of shifted copies of $X$ with a differential

768: $\mu^1_{hom(X,Y)} \otimes e_X$, and $T_X(Y)$ can be defined as a

769: twisted complex:

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

771:  T_X(Y) = \{hom^*(X,Y) \otimes X \longrightarrow Y\}.

772: \end{equation}

773: In fact, if the finite-dimensionality condition holds for all $Y$,

774: one can realize $T_X$ by an $A_\infty$-endofunctor of $Tw\Cat$,

775: inducing an exact functor on $D^b(\Cat)$. Similarly, if (1) holds and

776: (2') $Y$ has a strict cochain level identity morphism $e_Y$, then

777: $T_Y^\vee(X) = \{X \rightarrow hom^*(X,Y)^\vee \otimes Y\}$ will do.

778: Fukaya categories are not in general strictly unital, so the previous

779: observation does not apply directly to them. This can be remedied,

780: but instead of entering into a full discussion, we just record the

781: technical statement that will be used later on:

782: 

783: \begin{lemma} \label{th:recognize-twist}

784: Let $\Cat$ be a c-unital $A_\infty$-category whose $hom$ spaces are

785: finite-dimensional, and $X,Y,Z$ objects of it. Any pair

786: \[

787:  k \in hom^0_\Cat(Y,Z), \quad

788:  h : hom_\Cat(X,Y) \longrightarrow hom_\Cat(X,Z)[-1]

789: \]

790: satisfying

791: \begin{equation} \label{eq:coupled-morphism}

792:  \mu^1_\Cat(k) = 0, \quad

793:  \mu^1_\Cat h(a) + h(\mu^1_\Cat a) + \mu^2_\Cat(k,a) = 0

794: \end{equation}

795: for all $a \in hom_\Cat(X,Y)$, determines a morphism $T_X(Y)

796: \rightarrow Z$ in $D^b(\Cat)$. This is an isomorphism if for all

797: objects $W$ of $\Cat$, the following complex of vector spaces is

798: acyclic:

799: \begin{equation} \label{eq:total-cone}

800: \begin{split}

801:  & (hom(X,Y) \otimes hom(W,X))[2] \oplus hom(W,Y)[1] \oplus hom(W,Z), \\

802:  & \partial(a_2 \otimes a_1,b,c) =

803:  \big(\mu^1_\Cat(a_2) \otimes a_1 +

804:  (-1)^{|a_2|-1} a_2 \otimes \mu^1_\Cat(a_1),

805:  \mu^1_\Cat(b) \\ & \qquad + \mu^2_\Cat(a_2,a_1),

806:  -\mu^1_\Cat(c) + \mu^2_\Cat(k,b) + \mu^3_\Cat(k,a_2,a_1) +

807:  \mu^2(h(a_2),a_1)\big).

808: \end{split}

809: \end{equation}

810: \end{lemma}

811: 

812: \proof Suppose first that $\Cat$ is strictly unital. Then $(k,h)$ do

813: indeed define a $\mu^1$-closed cochain level morphism from the object

814: $T_X(Y)$ as defined in \eqref{eq:explicit-twist} to $Z$. Composition

815: with this morphism is a map of complexes of vector spaces

816: \[

817:  hom_{Tw\Cat}(W,T_X(Y)) \longrightarrow hom_{\Cat}(W,Z)

818: \]

819: and \eqref{eq:total-cone} is the mapping cone of it, so the

820: assumption that this is acyclic means that our morphism induces

821: isomorphisms

822: \[

823: Hom^*_{D^b(\Cat)}(W,T_X(Y)) \iso Hom^*_{D^b(\Cat)}(W,Z)

824: \]

825: for all $W \in Ob\,\Cat$. A straightforward long exact sequence

826: argument extends this to objects $W$ in the derived category, and

827: then the Yoneda Lemma implies that $T_X(Y) \iso Z$.

828: 

829: In the c-unital case, one argues as follows. There is a differential

830: graded category $\Cat^\vee = {\mathcal Fun}(\Cat,Ch)$ of c-unital

831: $A_\infty$-functors from $\Cat$ to the dg category $Ch$ of

832: finite-dimensional chain complexes over $\C$, and a natural

833: $A_\infty$-functor $\Cat \rightarrow \Cat^\vee$, which on cohomology

834: is full and faithful. This extends to an $A_\infty$-functor $Tw\Cat

835: \rightarrow \Cat^\vee$, which induces a full embedding of

836: triangulated categories $D^b(\Cat) \hookrightarrow H^0(\Cat^\vee)$.

837: Since $\Cat^\vee$ is strictly unital, one can define the object

838: $T_X(Y)$ in $H^0(\Cat^\vee)$ as in \eqref{eq:explicit-twist}; of

839: course, this will be non-canonically isomorphic to the object of the

840: same name in the subcategory $D^b(\Cat)$. By unravelling the

841: definitions one sees that the pair $(h,k)$ gives a morphism $T_X(Y)

842: \rightarrow Z$ in $H^0(\Cat^\vee)$; one proves as before that this is

843: an isomorphism. \qed

844: 

845: \begin{lemma} \label{th:twist-generators}

846: Let $C$ be a $\C$-linear idempotent closed triangulated category,

847: such that $Hom^*_C(X,Y)$ is finite-dimensional for all $X,Y$. Take a

848: collection of objects $X_1,\dots,X_m$ and another object $Y$, and

849: consider the sequence of objects $Y_0 = Y$, $Y_{d+1} = T_{X_1} \dots

850: T_{X_m}(Y_d)$. Suppose there is a $d>0$ such that $Hom_C(Y,Y_d) = 0$

851: (that is, there are no nontrivial morphisms of degree zero). Then $Y$

852: lies in the idempotent closed triangulated subcategory of $C$

853: split-generated by $X_1,\dots,X_m$.

854: \end{lemma}

855: 

856: \proof By definition, we have exact triangles

857: \begin{equation} \label{eq:taut-triangle}

858: \xymatrix{

859:  {T_{X_{i+1}} \dots T_{X_m}(Y_j)} \ar[rr] &&

860:  {T_{X_i}\dots T_{X_m}(Y_j)} \ar_{[1]}[dl] \\

861:  & {\hspace{-5em}

862:  Hom^*_C(X_i,T_{X_{i+1}}\dots T_{X_m}(Y_j)) \otimes X_i

863:  \hspace{-5em} } \ar_-{ev}[ul] &

864: }

865: \end{equation}

866: The octahedral axiom allows us to conflate these, for all $i$ and all

867: $j < k$, into a single exact triangle

868: \begin{equation} \label{eq:new-triangle}

869: \xymatrix{

870:  Y \ar[rr] && Y_d \ar_-{[1]}[dl] \\

871:  & R_d \ar[ul] &}

872: \end{equation}

873: where $R_d$ is an object built by taking repeated mapping cones from

874: the objects in the same positions in \eqref{eq:taut-triangle}, hence

875: which lies in the triangulated subcategory generated by

876: $X_1,\dots,X_m$. The assumption says that the $\rightarrow$ in

877: \eqref{eq:new-triangle} must be zero, hence $R_d \iso Y \oplus

878: Y_d[-1]$. \qed

879: 

880: Our applications of this Lemma will be along the following line: if

881: $C = D^\pi(\Cat)$ and there are fixed $X_1,\dots,X_m \in

882: Ob\,H^0(\Cat)$ such that any $Y \in Ob\,H^0(\Cat)$ has the property

883: required above, then the $X_k$ are split-generators for

884: $D^\pi(\Cat)$.

885: 

886: \subsection{}

887: Let $C^\rightarrow$ be a linear graded category with only finitely

888: many objects $(X_1,\dots,X_m)$, ordered in a fixed way. We say that

889: $C^\rightarrow$ is {\em directed} if

890: \begin{equation} \label{eq:directedness}

891:  Hom^*_{C^\rightarrow}(X_i,X_j) = \begin{cases}

892:  0 & i>j, \\ \C \cdot id_{X_i} & i = j, \\

893:  \text{finite-dimensional} & i < j.

894:  \end{cases}

895: \end{equation}

896: If $C$ is a linear $\C$-graded category with finite-dimensional

897: $Hom^*$'s, and $\Gamma = (X_1,\dots,X_m)$ an ordered finite family of

898: nonzero objects in it, we can define the associated directed

899: subcategory to be the subcategory $C^\rightarrow \subset C$ with the

900: same objects, which is directed and satisfies

901: $Hom^*_{C^\rightarrow}(X_i,X_j) = Hom^*_C(X_i,X_j)$ for $i<j$. This

902: is a full subcategory iff the $X_i$ form an {\em exceptional

903: collection}, which means that their morphisms in $C$ already satisfy

904: the conditions of \eqref{eq:directedness}.

905: 

906: Similarly, let $\Cat^\rightarrow$ be an $A_\infty$-category with

907: objects $(X_1,\dots,X_m)$. We call it directed if the $hom$-spaces

908: satisfy the same conditions as in \eqref{eq:directedness}, with the

909: additional condition that there are generators $e_i$ of

910: $hom(X_i,X_i)$ which are units in the strict sense. Given a finite

911: collection of objects $(X_1,\dots,X_m)$ in an arbitrary

912: $A_\infty$-category $\Cat$ with finite-dimensional $hom$'s, one can

913: define an associated directed $A_\infty$-category $\Cat^\rightarrow$

914: by requiring that $hom_{\Cat^\rightarrow}(X_i,X_j) =

915: hom_\Cat(X_i,X_j)$ for $i<j$, and

916: $\mu^d_{\Cat^\rightarrow}(a_d,\dots,a_1) =

917: \mu^d_{\Cat}(a_d,\dots,a_1)$ for all $a_k \in

918: hom(X_{i_{k-1}},X_{i_k})$ with $i_0 < \dots < i_d$. This is not

919: really an $A_\infty$-subcategory of $\Cat$ (unless that happens to

920: have strict units), but after some more work one can still construct

921: an $A_\infty$-functor $\Cat^\rightarrow \rightarrow \Cat$ which maps

922: the $hom(X_i,X_j)$, $i<j$, trivially. We will therefore tolerate a

923: slight abuse of terminology, and call $\Cat^\rightarrow$ the {\em

924: directed $A_\infty$-subcategory} associated to $(X_1,\dots,X_m)$.

925: 

926: The main interest in exceptional collections and directed categories

927: comes from the notion of mutations. Occasional glimpses of this

928: theory will be visible later in the paper, in particular in Section

929: \ref{sec:induction}, but it is not strictly necessary for our

930: purpose, so we will just refer the interested reader to

931: \cite{rudakov90,kontsevich98,seidel00}.

932: 

933: \section{Deformation theory\label{sec:deformations}}

934: 

935: We need to put together some material (well-known in the right

936: circles) from the classification theory of $A_\infty$-structures.

937: Using the Homological Perturbation Lemma

938: \cite{gugenheim-lambe-stasheff90}, the ``moduli space'' of

939: quasi-isomorphism types of $A_\infty$-algebras with fixed cohomology

940: algebra can be written as the quotient of an infinite-dimensional

941: space of solutions to the $A_\infty$-equations, by the action of an

942: infinite-dimensional ``gauge'' group. There is a canonical base

943: point, represented by the trivial $A_\infty$-structure (the one where

944: the higher order composition maps are all zero). The formal ``Zariski

945: tangent space'' at this point is a product of suitable Hochschild

946: cohomology groups of the cohomology algebra. Moreover, the ``moduli

947: space'' carries a $\C$-action which contracts it to the base point.

948: As noted by Kadeishvili \cite{kadeishvili88}, this implies that the

949: ``tangent space'' already carries a lot of information about the

950: ``moduli space''. A more straightforward argument along the same

951: lines applies to the classification of infinitesimal deformations of

952: a fixed possibly nontrivial $A_\infty$-structure, using a suitable

953: generalization of Hochschild cohomology \cite{getzler-jones90}. We

954: will recall these and some similar ideas, aiming in each case for

955: uniqueness results with easily verifiable criteria, tailored to the

956: specific computations later on. The proofs are exercises in abstract

957: deformation theory \`a la Deligne \cite{goldman-millson88}, and we

958: postpone them to the end.

959: 

960: \subsection \label{subsec:classification}

961: Let $A$ be a graded algebra. Consider the set $\AA(A)$ of those

962: $A_\infty$-algebras $\A$ whose underlying graded vector space is $A$,

963: and where the first two composition maps are

964: \begin{equation} \label{eq:mu1zero}

965:  \mu^1_\A = 0, \quad

966:  \mu^2_\A(a_2,a_1) = (-1)^{|a_1|} a_2a_1.

967: \end{equation}

968: This means that the cohomology algebra is $H(\A) = A$ itself. $\A,\A'

969: \in \AA(A)$ are considered equivalent if there is an

970: $A_\infty$-homomorphism $\G: \A \rightarrow \A'$ whose underlying

971: linear map is $\G^1 = id_A$. After writing out the relevant

972: equations, of which the first one is

973: \begin{equation} \label{eq:pullback}

974: \begin{split}

975:  \mu^3_{\A'}(a_3,a_2,a_1)

976:  & = \mu^3_\A(a_3,a_2,a_1) \\ &

977:  + \G^2(a_3,\mu^2_\A(a_2,a_1)) + (-1)^{|a_1|-1} \G^2(\mu^2_\A(a_3,a_2),a_1) \\

978:  & - \mu^2_\A(a_3,\G^2(a_2,a_1))

979:  - \mu^2_\A(\G^2(a_3,a_2),a_1),

980: \end{split}

981: \end{equation}

982: one sees that $\A$ and $\G$ determine $\A'$ completely, and that

983: there are no constraints on $\G$. In other words, one can define a

984: group $\GG(A)$ of {\em gauge transformations} which are arbitrary

985: sequences of multilinear maps $\{\G^1 = id_A, \G^2: A \otimes A

986: \rightarrow A[-1], \dots\}$, and then \eqref{eq:pullback} and its

987: successors define an action of $\GG(A)$ on $\AA(A)$, whose quotient

988: is the set of equivalence classes (this becomes entirely transparent

989: in a geometric formulation, where it is the action of formal

990: noncommutative diffeomorphisms on integrable odd vector fields).

991: Additionally, there is a canonical action of the multiplicative

992: semigroup $\C$ on $\AA(A)$, by multiplying $\mu^d$ with

993: $\epsilon^{d-2}$, which we denote by $\A \mapsto \epsilon^*\A$. For

994: $\epsilon \neq 0$, it is still true that $\A$ is isomorphic to

995: $\epsilon^*\A$, even though not by a gauge transformation: the

996: isomorphism is multiplication by $\epsilon^k$ on the degree $k$

997: component. The $\C$-action obviously descends to $\AA(A)/\GG(A)$.

998: 

999: The main justification for imposing conditions \eqref{eq:mu1zero} is

1000: the Homological Perturbation Lemma, which says the following. Let

1001: $\B$ be an $A_\infty$-algebra with an isomorphism $F: A \rightarrow

1002: H(\B)$. Then one can find an $A_\infty$-structure $\A$ on $A$

1003: satisfying \eqref{eq:mu1zero} and an $A_\infty$-homomorphism $\F: \A

1004: \rightarrow \B$ such that $H(\F) = F$. This establishes a bijection

1005: \[

1006:  \frac{\AA(A)}{\GG(A)} \iso \frac{

1007:  \{\text{pairs $(\B,F: A \rightarrow H(\B))$}\}

1008:  }

1009:  {

1010:  \{\text{$A_\infty$-quasi-isomorphisms compatible with $F$}\}

1011:  }.

1012: \]

1013: 

1014: \begin{remark} \label{th:explicit-hpt}

1015: The construction of $\A$ is explicit: take maps

1016: \begin{equation} \label{eq:hpt-data}

1017:  \pi: \B \rightarrow A, \quad

1018:  \lambda: A \rightarrow \B, \quad

1019:  h: \B \rightarrow \B[-1]

1020: \end{equation}

1021: such that $\pi \mu^1_\B = 0$, $\mu^1_\B \lambda = 0$ (with the map

1022: induced by $\lambda$ on cohomology being $F$), $\pi \lambda = id_A$,

1023: $\lambda \pi = id_\B + \mu^1_\B h + h \mu^1_\B$. Then $\mu_\A^d$ is a

1024: sum over planar trees with $d$ leaves and one root. The contribution

1025: from each tree involves only the $\mu^k_\B$ for $k$ at most $d$, and

1026: the auxiliary data \eqref{eq:hpt-data} (this particular formulation

1027: is taken from \cite{kontsevich-soibelman00}). For instance, writing

1028: $b_k = \lambda(a_k)$ one has

1029: \begin{align} \label{eq:mu4-hpt}

1030:  & \mu^4_\A(a_4,a_3,a_2,a_1) =

1031:  \pi \mu^4_\B(b_4,b_3,b_2,b_1) + {} \\ \notag

1032:  & \qquad \pi \mu^2_\B(h\mu^3_\B(b_4,b_3,b_2),b_1)

1033:  + \pi \mu^2_\B(b_4,h\mu^3_\B(b_3,b_2,b_1)) + {} \\ \notag

1034:  & \qquad \pi \mu^3_\B(h\mu^2_\B(b_4,b_3),b_2,b_1)

1035:  + \pi \mu^3_\B(b_4,h\mu^2_\B(b_3,b_2),b_1) + {} \displaybreak[0] \\ \notag

1036:  & \qquad \pi \mu^3_\B(b_4,b_3,h\mu^2_\B(b_2,b_1))

1037:  + \pi \mu^2_\B(h\mu^2_\B(b_4,b_3),h\mu^2_\B(b_2,b_1)) + {} \\ \notag

1038:  & \qquad \pi \mu^2_\B(h\mu^2_\B(h\mu^2_\B(b_4,b_3),b_2),b_1)

1039:  + \pi \mu^2_\B(h\mu^2_\B(b_4,h\mu^2_\B(b_3,b_2)),b_1) + {} \\ \notag

1040:  & \qquad \pi \mu^2_\B(b_4,h\mu^2_\B(h\mu^2_\B(b_3,b_2),b_1))

1041:  + \pi \mu^2_\B(b_4,h\mu^2_\B(b_3,h\mu^2_\B(b_2,b_1))).

1042: \end{align}

1043: \end{remark}

1044: 

1045: The Hochschild cohomology of $A$ is a bigraded vector space. In our

1046: slightly unusual notation, elements of $HH^{s+t}(A,A)^t$ are

1047: represented by cocycles $\tau \in CC^{s+t}(A,A)^t$, which are

1048: multilinear maps $\tau: A^{\otimes s} \rightarrow A$ of degree $t$.

1049: Suppose that for $\A \in \AA(A)$, the compositions $\mu_\A^s$ vanish

1050: for $2 < s < d$. Then $\mu^d_\A$ is a Hochschild cocycle; its class

1051: \[

1052: m_\A^d = [\mu^d_\A] \in HH^2(A,A)^{2-d},

1053: \]

1054: the order $d$ {\em deformation class} of $\A$, forms the obstruction

1055: to making $\mu^d_\A$ trivial by a gauge transformation which is

1056: itself equal to the identity to all orders $<d-1$. In particular, if

1057: all the obstruction groups $HH^2(A,A)^{2-s}$, $s > 2$, are zero,

1058: $\AA(A)/\GG(A)$ reduces to a point, which means that each

1059: $A_\infty$-algebra structure on $A$ can be trivialized by a gauge

1060: transformation (this is the {\em intrinsically formal} situation from

1061: \cite{halperin-stasheff79, kadeishvili88}). A slightly more general

1062: result allows the total obstruction group to be one-dimensional:

1063: 

1064: \begin{lemma} \label{th:versal-1}

1065: Suppose that there is some $d>2$ such that

1066: \[

1067: \begin{cases}

1068: HH^2(A,A)^{2-d} = \C, \\ HH^2(A,A)^{2-s} = 0 & \text{for all $s > 2$,

1069: $s \neq d$}.

1070: \end{cases}

1071: \]

1072: Suppose also that there exists an $\A \in \AA(A)$ such that $\mu_\A^s

1073: = 0$ for $2 < s < d$, and whose deformation class $m_\A^d$ is

1074: nonzero. Then any $\A' \in \AA(A)$ is equivalent to $\epsilon^*\A$

1075: for some $\epsilon \in \C$, which means the $\C$-orbit of $[\A]$ is

1076: the whole of $\AA(A)/\GG(A)$.

1077: \end{lemma}

1078: 

1079: We say that $\A$ is {\em versal}. All other problems that we will

1080: consider have the same structure: a set of equivalence classes

1081: described as a quotient, often involving the appeal to some kind of

1082: Perturbation Lemma; an additional group of reparametrizations acting

1083: on the quotient; a suitable version of Hochschild cohomology, and

1084: notion of deformation class; and ultimately, results expressing

1085: intrinsic formality or versality.

1086: 

1087: \begin{remark} \label{th:versal-2}

1088: Lemma \ref{th:versal-1} can be somewhat generalized, even though that

1089: does not fit in as well with our general philosophy as the original

1090: statement. The generalization goes as follows: suppose that

1091: $HH^2(A,A)^{2-s} = 0$ for all $s>2$ except $s = d$, with

1092: $HH^2(A,A)^{2-d}$ arbitrary. Then every point in $\AA(A)/\GG(A)$ has

1093: a representative $\A$ with $\mu^s_\A = 0$ for $2<s<d$; and if two

1094: such representatives $\A,\A'$ have the same order $d$ deformation

1095: class, they are equivalent.

1096: \end{remark}

1097: 

1098: \subsection{}

1099: Let $\A$ be an $A_\infty$-algebra. A {\em one-parameter deformation}

1100: of $\A$ is a topological $A_\infty$-algebra $\A_q$ over $\Lambda_\N$,

1101: whose underlying graded vector space is $\A \hat\otimes \Lambda_\N$,

1102: and whose composition maps reduce to those of $\A$ if one sets $q =

1103: 0$. More concretely,

1104: \[

1105: \mu_{\A_q}^d = \mu_{\A}^d + q \mu_{\A_q,1}^d + q^2 \mu_{\A_q,2}^d +

1106: \dots =\mu_{\A}^d + O(q),

1107: \]

1108: where each coefficient is a $\C$-linear map $\A^{\otimes d}

1109: \rightarrow \A[2-d]$. Two one-parameter deformations are equivalent

1110: if there is a $\Lambda_\N$-linear $A_\infty$-homomorphism $\G_q: \A_q

1111: \rightarrow \A'_q$ between them of the form

1112: \begin{equation} \label{eq:zero-phi}

1113: \G_q^d = \begin{cases} id + O(q) & d = 1, \\ O(q) & d>1. \end{cases}

1114: \end{equation}

1115: Let $\AA_q(\A)$ be the set of all one-parameter deformations, and

1116: $\GG_q(\A)$ the group of sequences of multilinear maps $\{\G^1_q,

1117: \G^2_q,\dots\}$ on $\A \hat{\otimes} \Lambda_\N$ which satisfy

1118: \eqref{eq:zero-phi}. Then $\AA_q(\A)/\GG_q(\A)$ is the set of

1119: equivalence classes. The semigroup $End(\Lambda_\N)$ acts on

1120: $\AA_q(\A)$ by reparametrizations $q \mapsto \psi(q)$; we denote the

1121: action by $\A_q \mapsto \psi^*\A_q$. In this context, versality

1122: therefore means that the $End(\Lambda_\N)$-orbit of some equivalence

1123: class $[\A_q]$ is the whole of $\AA_q(\A)/\GG_q(\A)$.

1124: 

1125: \begin{remark}

1126: We will also occasionally use terminology from finite order

1127: deformation theory. An order $d$ infinitesimal deformation of $\A$

1128: would be an $A_\infty$-algebra $\A_q$ as before, except that the

1129: ground ring is now $\C[[q]]/q^{d+1}$. The notion of equivalence is

1130: obvious. Of course, a one-parameter deformation can be truncated to

1131: any desired finite order in $q$.

1132: \end{remark}

1133: 

1134: Hochschild cohomology generalizes to $A_\infty$-algebras

1135: \cite{getzler-jones90} as a graded (no longer bigraded) space

1136: $HH^*(\A,\A)$. Consider $CC^d(\A,\A) = \prod_{s+t = d}

1137: Hom_\C(\A^{\otimes s},\A[t])$ with the Gerstenhaber product

1138: \begin{equation} \label{eq:gerstenhaber}

1139:  (\sigma \circ \tau)^d(a_d,\dots,a_1) =

1140:  \sum_{i,j} (-1)^\sharp \sigma^{d-j+1}(a_d,\dots,

1141:  \tau^j(a_{i+j},\dots,a_{i+1}),\dots,a_1),

1142: \end{equation}

1143: where $\sharp = (|\tau|-1)(|a_1|+\dots+|a_i|-i)$, and the

1144: corresponding Lie bracket $[\sigma,\tau] = \sigma \circ \tau -

1145: (-1)^{(|\sigma|-1)(|\tau|-1)} \tau \circ \sigma$. As $\mu_\A^*$

1146: itself lies in $CC^2(\A,\A)$ and satisfies $\mu_A^* \circ \mu_A^* =

1147: 0$ by definition, one can define a differential $\delta\tau =

1148: [\mu_\A^*,\tau]$, and this is the $A_\infty$ version of the

1149: Hochschild differential. Filtration by length yields a spectral

1150: sequence with

1151: \[

1152: E_2^{s,t} = HH^{s+t}(A,A)^t

1153: \]

1154: for $A = H(\A)$, and this converges to $HH^*(\A,\A)$ under suitable

1155: technical assumptions (for instance, if the nonzero terms in the

1156: $E_2$ plane are concentrated in a region cut out by $t \leq c_1$, $t

1157: \geq c_2 - c_3 s$ for $c_1,c_2$ arbitrary and $c_3 < 1$). If

1158: $\mu_{\A}^d$, $d \geq 3$, is the first nonzero higher order

1159: composition, then the first potentially nontrivial differential in

1160: the spectral sequence is

1161: \begin{equation} \label{eq:bracket-differential}

1162:  \delta_{d-1} = [m_{\A}^d,\cdot]:

1163:  E_2^{s,t} \longrightarrow E_2^{s+d-1,t+2-d}

1164: \end{equation}

1165: where $[\cdot,\cdot]$ is the Gerstenhaber bracket on $HH^{*+1}(A,A)$,

1166: and $m_{\A}^d$ is the order $d$ deformation class. Actually, the

1167: spectral sequence is a quasi-isomorphism invariant, so that one can

1168: study it under the assumption \eqref{eq:mu1zero}. Then, either $\A$

1169: is equivalent to the trivial $A_\infty$-structure on $A$ and the

1170: spectral sequence degenerates; or else, after using gauge

1171: transformations to make as many higher order terms zero as possible,

1172: one encounters a nontrivial deformation class $m_{\A}^d$ for some

1173: $d>2$. In that case, taking the Euler element $\tau \in HH^1(A,A)^0$

1174: given by the derivation which is $k$ times the identity on $A^k$, one

1175: finds that

1176: \begin{equation} \label{eq:euler-element}

1177:  [m_\A^d,\tau] = (d-2) m_\A^d.

1178: \end{equation}

1179: By \eqref{eq:bracket-differential}, $\delta_{d-1}$ is nonzero and

1180: kills $m_\A^d \in E_{d-1}^{d,2-d} = HH^2(A,A)^{2-d}$.

1181: 

1182: Having come this far by straightforward adaptation of the previously

1183: used notions, we encounter a slight snag, which is that the

1184: deformation problem governed by $HH^*(\A,\A)$ reaches beyond the

1185: class of $A_\infty$-algebras into that of ``curved'' or

1186: ``obstructed'' ones. To exclude these additional deformations, one

1187: can proceed as follows. Let $\A$ be an $A_\infty$-algebra with

1188: $\mu^1_{\A} = 0$. The {\em truncated Hochschild cohomology} is a

1189: graded vector space $HH^*(\A,\A)^{\scriptscriptstyle \leq 0}$, built

1190: from the subcomplex $CC^*(\A,\A)^{\scriptscriptstyle \leq 0}$ of

1191: Hochschild cochains $\tau: \A^{\otimes s} \rightarrow \A[t]$ with $t

1192: \leq 0$. There is a spectral sequence leading to it as before, where

1193: now

1194: \begin{equation} \label{eq:truncated-ss}

1195:  E_2^{s,t} = \begin{cases}

1196:  HH^{s+t}(A,A)^t & \text{$t \leq 0$}, \\

1197:  0 & \text{otherwise.}

1198:  \end{cases}

1199: \end{equation}

1200: All remarks made above carry over to this modification, including

1201: \eqref{eq:euler-element} because the Euler element lies in

1202: \eqref{eq:truncated-ss}. For the application to deformation theory,

1203: consider the subset $\AA_q(\A)^{\scriptscriptstyle \leq 0} \subset

1204: \AA_q(\A)$ of those one-parameter deformations $\A_q$ during which

1205: the differential remains trivial, $\mu^1_{\A_q} = 0$. This still

1206: carries actions of $\GG_q(\A)$ and of $End(\Lambda_\N)$. The order

1207: $q$ term of a deformation of this kind defines a primary deformation

1208: class

1209: \[

1210: m_{\A_q}^{\scriptscriptstyle \leq 0} = [\mu_{\A_q,1}^*] \in

1211: HH^2(\A,\A)^{\scriptscriptstyle \leq 0},

1212: \]

1213: which is the obstruction to transforming $\A_q$ into a deformation of

1214: order $q^2$ (or in other words, making the associated first order

1215: deformation trivial). The desired versality result is:

1216: 

1217: \begin{lemma} \label{th:versal-3}

1218: Suppose that $HH^2(\A,\A)^{\scriptscriptstyle \leq 0} \iso \C$. Let

1219: $\A_q$ be a one-parameter deformation with $\mu^1_{\A_q} = 0$, whose

1220: primary deformation class is nonzero. Then $\A_q$ is versal in

1221: $\AA_q(\A)^{\scriptscriptstyle \leq 0}/\GG_q(\A)$.

1222: \end{lemma}

1223: 

1224: Take an arbitrary $A_\infty$-algebra $\B$, and use the Perturbation

1225: Lemma to find an $\A$ with $\mu^1_\A = 0$ and a quasi-isomorphism

1226: $\F: \A \rightarrow \B$. Suppose that $\B_q$ is a one-parameter

1227: deformation of $\B$ such that $H(\B_q)$ is torsion-free. Then

1228: $H(\B_q) \iso H(\B) \hat\otimes \Lambda_\N$ non-canonically, and a

1229: straightforward extension of the Perturbation Lemma allows one to

1230: construct a one-parameter deformation $\A_q$ of $\A$ with vanishing

1231: differential, and an $\F_q: \A_q \rightarrow \B_q$ which deforms

1232: $\F$. This in fact yields a bijection

1233: \begin{equation} \label{eq:deformation-comparison}

1234:  \frac{\AA_q(\A)^{\scriptscriptstyle \leq 0}}{\GG_q(\A)} \iso

1235:  \frac{\{\B_q \in \AA_q(\B), \text{$H(\B_q)$ torsion-free}\}}{\GG_q(\B)}.

1236: \end{equation}

1237: 

1238: \subsection{}

1239: Lemma \ref{th:versal-1} is essentially sufficient for our needs, but

1240: Lemma \ref{th:versal-3} is not, since

1241: $HH^2(\A,\A)^{\scriptscriptstyle \leq 0}$ turns out to be too large

1242: in our example. This means that we need to restrict the class of

1243: allowed deformations further, so as to whittle down the dimension of

1244: the relevant ``moduli space'' to one. The solution adopted here

1245: (largely motivated by the desire to minimize work on the symplectic

1246: geometry side later on) is to consider $A_\infty$-algebras equipped

1247: with an automorphism, and deformations of those. In fact, to turn

1248: this into a standard deformation problem, we replace the automorphism

1249: by its graph, which is an $A_\infty$-bimodule.

1250: 

1251: Some background (see \cite{tradler01} for a more detailed

1252: exposition): to any $A_\infty$-algebra $\A$ one can associate the dg

1253: category $\bimodA$ of $A_\infty$-bimodules. An object is a graded

1254: vector space $\M$ with operations $\mu^{f,e}_\M: \A^{\otimes f}

1255: \otimes \M \otimes \A^{\otimes e} \rightarrow \M[1-e-f]$, for all

1256: $e,f \geq 0$, satisfying certain structural equation, and also

1257: subject to a certain c-unitality condition (see below). The cochain

1258: space of morphisms $\M \rightarrow \NN$ of degree $k$ is the space of

1259: sequences of multilinear maps $\beta^{f,e}: \A^{\otimes f} \otimes \M

1260: \otimes \A^{\otimes e} \rightarrow \NN[k-e-f]$. Note that

1261: $\mu_\M^{*,*}$ itself is an element in $hom^1(\M,\M)$. To give

1262: explicit formulae for the dg structure, it is best to start with

1263: composition of morphisms, which is somewhat similar to the

1264: Gerstenhaber product:

1265: \begin{multline*}

1266:  (\gamma \circ \beta)^{f,e}

1267:  (b_f,\dots,b_1,m,a_e,\dots,b_1) = \\ =

1268:  \sum_{i,j} (-1)^\dag \gamma^{f-j,i}(b_f,\dots,b_{j+1},

1269:  \beta^{j,e-i}(b_j,\dots,m,\dots,a_{i+1}),a_i,\dots,a_1),

1270: \end{multline*}

1271: where $\dag = |\beta|(|a_i|+\dots+|a_1|-i)$. We also have a

1272: preliminary version of the differential, involving only the

1273: $A_\infty$-structure of $\A$:

1274: \begin{multline*}

1275:  (-1)^{|\beta|+1}(\delta_{\text{pre}}\beta)^{f,e}(b_f,\dots,b_1,m,a_e,\dots,a_1) = \\ =

1276:  \sum_{i,j} (-1)^{\ast}

1277:  \beta^{f,e-j+1}(b_f,\dots,m,\dots,a_{i+j+1},\mu^j_\A(a_{i+j},\dots,a_{i+1}),a_i,\dots,a_1) \\ +

1278:  \sum_{i,j} (-1)^{\natural}

1279:  \beta^{f-j+1,e}(b_f,\dots,b_{i+j+1},\mu^j_\A(b_{i+j},\dots,b_{i+1}),b_i,\dots,m,\dots,a_1),

1280: \end{multline*}

1281: with $\ast$ as in \eqref{eq:ainfty} and $\natural = |a_1| + \dots +

1282: |a_e| + |m| + |b_1| + \dots + |b_i| - e - 1 - i$. This satisfies

1283: $\delta_{\text{pre}}^2 = 0$ and $\delta_{\text{pre}}(\gamma \circ

1284: \beta) = \delta_{\text{pre}}(\gamma) \circ \beta + (-1)^{|\gamma|}

1285: \gamma \circ \delta_{\text{pre}}(\beta)$. Having that, one can write

1286: the structural equations of an $A_\infty$-bimodule as

1287: \begin{equation} \label{eq:module-equation}

1288: \delta_{\text{pre}}\mu^{*,*} + \mu^{*,*} \circ \mu^{*,*} = 0,

1289: \end{equation}

1290: and the actual differential on $hom(\M,\NN)$ as $\delta \beta =

1291: \delta_{\text{pre}}\beta + \mu_\NN^{*,*} \circ \beta - (-1)^{|\beta|}

1292: \beta \circ \mu_\M^{*,*}$. The cohomology $M = H(\M,\mu_\M^{0,0})$ of

1293: any $A_\infty$-bimodule is naturally a graded bimodule over $A =

1294: H(\A)$, and we require that the unit of $A$ acts as the identity from

1295: both sides. By using the length filtration of

1296: $hom_{\bimodA}(\M,\NN)$, one gets a spectral sequence whose $E_1$

1297: term can be identified with $E_1^{s,t} =

1298: Hom_{A-A}(B^{-s}_{A-A}M,N[t])$, where $B^*_{A-A}M$ is the classical

1299: bimodule bar resolution. This is compatible with the differential, so

1300: that

1301: \begin{equation} \label{eq:bimodule-ss}

1302:  E_2^{s,t} = Ext_{A-A}^s(M,N[t]).

1303: \end{equation}

1304: In particular, if $\M$ or $\NN$ are acyclic, then so is

1305: $hom_{\bimodA}(\M,\NN)$. If we allow ourselves to loosely borrow

1306: language from from \cite{spaltenstein88}, the fact that the bar

1307: resolution is built into the definition of our category ensures that

1308: ``all $A_\infty$-bimodules are $K$-projective (and $K$-injective)''.

1309: It follows that any quasi-isomorphism $\M \rightarrow \NN$ has a

1310: quasi-inverse, hence is an isomorphism in the category

1311: $H^0(\bimodA)$.

1312: 

1313: There is a ``coupled Hochschild cohomology'' theory associated to a

1314: pair $(\A,\M)$ consisting of an $A_\infty$-algebra and an

1315: $A_\infty$-bimodule over it. It sits in a long exact sequence

1316: \begin{multline} \label{eq:coupled-sequence}

1317:  \quad \cdots \longrightarrow HH^{*-1}(\A,\A) \stackrel{d}{\longrightarrow}

1318:  H^{*-1}(hom_{\bimodA}(\M,\M)) \stackrel{a}{\longrightarrow} \\ \longrightarrow

1319:  HH^*(\A \ampersand \M, \A \ampersand \M) \stackrel{b}{\longrightarrow}

1320:  HH^*(\A,\A) \longrightarrow \cdots \quad

1321: \end{multline}

1322: Later on, when we consider it from the point of view of deformation

1323: theory, the meaning of the sequence will be clear: $b$ forgets $\M$

1324: and concentrates on the deformation of $\A$, while the image of $a$

1325: consists of deformations of $\M$ with respect to a fixed $\A$. To

1326: define $HH^*(\A \ampersand \M,\A \ampersand \M)$ one can use the

1327: triangular matrix trick, see for instance \cite{guccione-guccione01}.

1328: Define an $A_\infty$-algebra

1329: \begin{equation} \label{eq:t}

1330: \T = \begin{pmatrix} \T_{11} & \T_{12} \\ \T_{21} & \T_{22}

1331: \end{pmatrix} = \begin{pmatrix} \A & 0 \\ \M & \A \end{pmatrix}

1332: \end{equation}

1333: whose underlying graded vector space is the direct sum of two copies

1334: of our $A_\infty$-algebra, $\T_{11} = \T_{22} = \A$, with one copy of

1335: the bimodule, $\T_{12} = \M$. The only nonzero compositions are

1336: \begin{align*}

1337:  & \mu^d_\T = \mu^d_\A: \T_{kk}^{\otimes d} \longrightarrow \T_{kk},

1338:  \;\;\quad k = 1,2, \\

1339:  & \mu_\T^{e+f+1} = \mu^{f,e}_\M: \T_{22}^{\otimes f} \otimes \T_{21} \otimes \

1340:  \T_{11}^{\otimes e} \longrightarrow \T_{21}.

1341: \end{align*}

1342: Consider the subcomplex $CC^*(\A \ampersand \M, \A \ampersand \M)

1343: \subset CC^*(\T,\T)$ of cochains of the same form, meaning that they

1344: are made up of maps $\T_{kk}^{\otimes d} \rightarrow \T_{kk}$, which

1345: should be the same for $k = 1,2$, and maps $\T_{22}^{\otimes f}

1346: \otimes \T_{21} \otimes \T_{11}^{\otimes e} \rightarrow \T_{21}$. The

1347: cochains of the second kind form a subcomplex of $CC^*(\A \ampersand

1348: \M, \A \ampersand \M)$ which is isomorphic to $hom(\M,\M)[-1]$, and

1349: the quotient complex is just $CC^*(\A,\A)$, which makes

1350: \eqref{eq:coupled-sequence} clear.

1351: 

1352: The bimodules that we will be interested in are graphs $\M =

1353: Graph(\F)$ of $A_\infty$-homomorphisms $\F: \A \rightarrow \A$. The

1354: underlying graded vector space of $\M$ is the same as for $\A$, and

1355: the operations are

1356: \begin{multline*}

1357:  \mu^{f,e}_\M(b_f,\dots,b_1,m,a_e,\dots,a_1) = \\ = \!\!

1358:  \sum_{i_1 + \dots + i_r = f} \!\!

1359:  \mu_\A^{f+r+1}(b_f,\dots,b_1,m,\F^{i_r}(a_e,\dots,a_{e-i_r+1}),

1360:  \dots,\F^{i_1}(a_{i_1},\dots,a_1)).

1361: \end{multline*}

1362: 

1363: \begin{lemma} \label{th:graph}

1364: For a quasi-isomorphism $\F$, one can map $H^*(hom_{\bimodA}(\M,\M))

1365: \iso HH^*(\A,\A)$ in such a way that the differential $d$ in

1366: \eqref{eq:coupled-sequence} becomes $id-\F^*$, where $\F^*$ is the

1367: automorphism of Hochschild cohomology induced by $\F$.

1368: \end{lemma}

1369: 

1370: Our proof of this will be rather pedestrian, but a more satisfactory

1371: treatment requires all the machinery of $A_\infty$-categories of

1372: $A_\infty$-functors, for which we have no real use elsewhere. To get

1373: started, we need to recall how quasi-isomorphism invariance of

1374: Hochschild cohomology for $A_\infty$-algebras is proved. The most

1375: obvious scheme does not apply because $HH^*$ is neither co- nor

1376: contravariantly functorial, and instead one passes through a

1377: generalization, which is the {\em functor Hochschild cohomology}

1378: $HH^*(\F,\F)$ of an $A_\infty$-homomorphism $\F: \A \rightarrow \B$.

1379: We will not give the full definition of this, but the underlying

1380: cochain complex is $CC^d(\F,\F) = \prod_{s+t = d} Hom(\A^{\otimes

1381: s},\B[t])$, and the $E_2$ term of the filtration-by-length spectral

1382: sequence is $Ext^s_{A-A}(A,B[t])$, where $B$ is considered as an

1383: $A$-bimodule through pullback by $F = H(\F)$. There are natural maps

1384:  $\F^{*,\text{left}}: HH^*(\A,\A) \rightarrow HH^*(\F,\F)$,

1385:  $\F^{*,\text{right}}: HH^*(\B,\B) \rightarrow HH^*(\F,\F)$

1386: defined on the cochain level by

1387: \begin{align*}

1388:  & (\F^{*,\text{left}}\tau)(a_d,\dots,a_1) =

1389:  \sum_{i,j} (-1)^\sharp

1390:  \F^{d-j+1}(a_d,\dots,\tau^j(a_{i+j},\dots,a_{i+1}),\dots,a_1), \\

1391:  &

1392:  (\F^{*,\text{right}}\tau)(a_d,\dots,a_1) = \\ & \qquad \qquad \qquad =

1393:  \sum_{i_1 + \dots + i_r = d}

1394:  \tau^r(\F^{i_r}(a_d,\dots,a_{d-i_r+1}),\dots,\F^{i_1}(a_{i_1},\dots,a_1)).

1395: \end{align*}

1396: with $\sharp$ as in \eqref{eq:gerstenhaber}. If $\F$ is a

1397: quasi-isomorphism, a spectral sequence comparison argument shows that

1398: both maps are isomorphisms. One then defines $\F^*$ to be the

1399: composition $(\F^{*,\text{left}})^{-1} \circ \F^{*,\text{right}}:

1400: HH^*(\B,\B) \rightarrow HH^*(\A,\A)$.

1401: 

1402: Suppose from now on that $\B = \A$, so we are in the situation of

1403: Lemma \ref{th:graph}, and $\F^*$ is the automorphism of Hochschild

1404: cohomology induced by $\F$. There is a natural isomorphism $c:

1405: HH^*(\F,\F) \rightarrow H^*(hom_{\bimodA}(\M,\M))$ for $\M =

1406: Graph(\F)$, given on cochains by

1407: \begin{multline*}

1408:  (c\tau)^{f,e}(b_f,\dots,b_1,m,a_e,\dots,a_1) =

1409:  \\ = \sum_{\substack{k \leq r \\ i_1 + \dots + i_r = e}}

1410:  (-1)^\flat \mu_\A^{f+1+r}(b_f,\dots,b_1,m,\F^{i_r}(a_e,\dots,a_{e-i_r+1}),

1411:  \dots, \qquad \\[-1.5em] \tau^{i_k}(a_{i_1+\dots+i_k},\dots,a_{i_1+\dots+i_{k-1}+1}),

1412:  ,\dots,\F^{i_1}(a_{i_1},\dots,a_1)).

1413: \end{multline*}

1414: $\flat = (|\tau|-1)(|a_1|+\dots+|a_{i_{k-1}}|-i_1-\dots-i_{k-1})$. To

1415: prove that this is an isomorphism, one considers the $E_2$ term of

1416: the usual spectral sequences, and checks that the action of $c$ on it

1417: can be traced back to a comparison map between two different

1418: resolutions of the bimodule $M = H(\M)$. Finally, the following

1419: diagram, where $d$ is from \eqref{eq:coupled-sequence}, commutes:

1420: \begin{equation} \label{eq:f-f}

1421: \xymatrix{

1422:  {HH^*(\A,\A)}

1423:  \ar[rr]^-{\F^{*,\text{left}}-\F^{*,\text{right}}}

1424:  \ar@/_1pc/[drr]_-{d}

1425:  && {HH^*(\F,\F)} \ar[d]_{\iso}^{c} \\

1426:  && {H^*(hom_{\bimodA}(\M,\M))}

1427: }

1428: \end{equation}

1429: In view of the definition of $\F^*$, this proves Lemma

1430: \ref{th:graph}. Actually \eqref{eq:f-f} does not strictly commute on

1431: cochains, but the necessary homotopy can be easily provided,

1432: \begin{multline*}

1433: (h\tau)(b_f,\dots,m,\dots,a_1) = \\ = \!\! \sum_{i_1+\dots+i_r = e}

1434:  \!\!\!

1435:  \tau(b_f,\dots,m,\F^{i_r}(a_e,\dots,a_{e-i_r+1}),

1436:  \dots,\F^{i_1}(a_{i_1},\dots,a_1)).

1437: \end{multline*}

1438: 

1439: \subsection{}

1440: Take a graded algebra $A$, and let $\A$ be an $A_\infty$-structure on

1441: it as in \eqref{eq:mu1zero}. Given a graded $A$-bimodule $M$, we want

1442: to look at $A_\infty$-bimodule structures $\M$ on it such that the

1443: analogue of \eqref{eq:mu1zero} holds,

1444: \begin{equation} \label{eq:m-mu2}

1445:  \mu^{0,0}_\M = 0, \quad

1446:  \mu^{1,0}_\M(a,m) = (-1)^{|a|} ma, \quad

1447:  \mu^{0,1}_\M(m,a) = (-1)^{|m|} am.

1448: \end{equation}

1449: 

1450: \begin{lemma} \label{th:formal-m}

1451: Suppose that $Ext_{A \otimes A^{op}}^k(M,M[1-k]) = 0$ for all $k \geq

1452: 2$. Then, up to isomorphism, there is at most one $\M$ satisfying

1453: \eqref{eq:m-mu2}.

1454: \end{lemma}

1455: 

1456: It might seem that there is a straightforward ``linear'' approach to

1457: this: take two $A_\infty$-bimodules $\M,\M'$ satisfying

1458: \eqref{eq:m-mu2}, and look at the spectral sequence

1459: \eqref{eq:bimodule-ss} associated to $hom(\M,\M')$. The assumption in

1460: the Lemma ensures that $id_M \in E_2^{0,0}$ survives to $E_\infty$,

1461: and if the spectral sequence converges, one could use a standard

1462: comparison argument to show that the resulting element of

1463: $H^0(hom(\M,\M'))$ is an isomorphism. However, there seems to be no

1464: reason why convergence should hold in general, and our proof will

1465: rely instead on the ``nonlinear'' methods of deformation theory. To

1466: complete our discussion, it is maybe useful to say that there is a

1467: version of the Perturbation Lemma for $A_\infty$-bimodules, which

1468: shows that any such bimodule over an $A_\infty$-algebra $\A$ with

1469: $\mu^1_\A = 0$ is quasi-isomorphic to one with $\mu^{0,0}_\M = 0$,

1470: hence which is of the form \eqref{eq:m-mu2}. To put this on a more

1471: formal footing, consider the dg subcategory

1472: $\bimodA^{\scriptscriptstyle \leq 0} \subset \bimodA$ whose objects

1473: are $A_\infty$-bimodules $\M$ with $\mu^{0,0}_\M = 0$, and where the

1474: morphisms $\beta$ of degree $d$ must satisfy $\beta^{p,q} = 0$ for

1475: $p+q < d$. By definition, the functor induced by inclusion is a full

1476: embedding $H^0(\bimodA^{\scriptscriptstyle \leq 0}) \rightarrow

1477: H^0(\bimodA)$. The bimodule Perturbation Lemma shows that it is an

1478: equivalence.

1479: 

1480: We now turn to the ``coupled'' deformation problem which is our main

1481: interest in this matter. Let $A$, $\A$ and $M$ be as before, and

1482: suppose that we have some fixed $A_\infty$-bimodule structure $\M$

1483: satisfying \eqref{eq:m-mu2}. Let $\AA_q(\A \ampersand

1484: \M)^{\scriptscriptstyle \leq 0}$ be the set of pairs $(\A_q,\M_q)$

1485: where $\A_q$ is a one-parameter deformation of $\A$, and $\M_q$ a

1486: corresponding one-parameter deformation of $\M$ (that is, an

1487: $\A_q$-bimodule structure on $\M \hat{\otimes}_\C \Lambda_\N$ having

1488: the obvious kind of $q$-expansion), such that

1489: \begin{equation} \label{eq:truncated-pair}

1490:  \mu^1_{\A_q} = 0, \quad \mu^{0,0}_{\M_q} = 0.

1491: \end{equation}

1492: The formal gauge group $\GG_q(\A \ampersand \M)$ for this problem can

1493: be defined as a suitable subgroup of the corresponding group for

1494: deformations of the triangular algebra $\T$. For us, what matters is

1495: that gauge equivalence of pairs implies equivalence of the

1496: $A_\infty$-deformations $\A_q$, and we will later state the versality

1497: result in a correspondingly weak form. Let $HH^*(\A \ampersand \M,\A

1498: \ampersand \M)^{\scriptscriptstyle \leq 0}$ be the truncated version

1499: of coupled Hochschild cohomology, defined by looking only at cochains

1500: $\tau_{11} = \tau_{22}: \A^{\otimes s} \rightarrow \A[t]$,

1501: $\tau_{12}: \A^{\otimes f} \otimes \M \otimes \A^{\otimes e}

1502: \rightarrow \M[t]$ with $t \leq 0$. This sits in a suitable version

1503: of \eqref{eq:coupled-sequence},

1504: \begin{multline} \label{eq:coupled-sequence-2}

1505:  \dots \longrightarrow HH^{*-1}(\A,\A)^{\leq 0} \xrightarrow{d}

1506:  H^{*-1}(hom_{\bimodA^{\scriptscriptstyle\leq 0}}(\M,\M)) \longrightarrow

1507:  \\ \rightarrow

1508:  HH^*(\A \ampersand \M, \A \ampersand \M)^{\scriptscriptstyle \leq 0} \longrightarrow

1509:  HH^*(\A,\A)^{\scriptscriptstyle \leq 0} \xrightarrow{d} \dots

1510: \end{multline}

1511: In the case where $\M = Graph(\F)$, the obvious variation of Lemma

1512: \ref{th:graph} applies. Any pair $(\A_q,\M_q) \in \AA_q(\A \ampersand

1513: \M)^{\scriptscriptstyle \leq 0}$ has a primary deformation class in

1514: truncated coupled Hochschild cohomology. Again, the important thing

1515: for us is that this maps to the deformation class of $\A_q$ under

1516: \eqref{eq:coupled-sequence-2}.

1517: 

1518: \begin{lemma} \label{th:versal-4}

1519: Suppose that $HH^2(\A \ampersand \M,\A \ampersand

1520: \M)^{\scriptscriptstyle \leq 0} \iso \C$, and that there is a

1521: deformation $(\A_q,\M_q)$ satisfying \eqref{eq:truncated-pair} with

1522: nonzero $m_{\A_q}^{\scriptscriptstyle \leq 0} \in

1523: HH^2(\A,\A)^{\scriptscriptstyle \leq 0}$. Then, for any other pair

1524: $(\A_q',\M_q') \in \AA_q(\A \ampersand \M)^{\scriptscriptstyle \leq

1525: 0}$, there is a $\psi \in End(\Lambda_\N)$ such that $\A_q' \iso

1526: \psi^*\A_q$ as a one-parameter deformation of $A_\infty$-algebras.

1527: \end{lemma}

1528: 

1529: \subsection{}

1530: The framework of abstract one-parameter deformation theory is as

1531: follows. Let $\g$ be a dg Lie algebra. We consider the space

1532: $\AA_q(\g)$ of elements $\alpha_q \in \g^1 \hat{\otimes}\,

1533: q\Lambda_\N$ which satisfy the Maurer-Cartan equation

1534: \begin{equation} \label{eq:classical-mc}

1535:  \delta\alpha_q + \half [\alpha_q,\alpha_q] = 0.

1536: \end{equation}

1537: In particular, if we expand $\alpha_q = \alpha_{q,1}q + \alpha_{q,2}

1538: q^2 + \cdots$, the leading order term defines a primary deformation

1539: class $[\alpha_{q,1}] \in H^1(\g)$. An infinitesimal abstract gauge

1540: transformation $\gamma_q \in \g^0 \hat{\otimes} q\Lambda_\N$ defines

1541: a formal vector field $X_q$ on $\AA_q(\g)$, $X_q(\alpha_q) =

1542: \delta\gamma_q + [\alpha_q,\gamma_q]$. These vector fields

1543: exponentiate to an action of a group $\GG_q(\g)$ on $\AA_q(\g)$,

1544: which we denote by $\exp(\gamma_q)(\alpha_q)$. The primary

1545: deformation class is an invariant of this action. As usual, there is

1546: also an action of $End(\Lambda_\N)$ on $\AA_q(\g)$ by

1547: reparametrizations, hence a notion of versality. The abstract

1548: formality and versality theorems are

1549: 

1550: \begin{lemma} \label{th:formal-abstract}

1551: If $H^1(\g) = 0$, $\AA_q(\g)/\GG_q(\g)$ is a point.

1552: \end{lemma}

1553: 

1554: \begin{lemma} \label{th:versal-abstract}

1555: Suppose that $H^1(\g) \iso \C$. Then, every $\alpha_q \in \AA_q(\g)$

1556: whose primary deformation class is nonzero is versal.

1557: \end{lemma}

1558: 

1559: These are both standard results, but for the convenience of the

1560: reader we reproduce the conventional induction-by-order proof of the

1561: second one. Take an arbitrary $\beta_q \in \AA_q(\g)$. Suppose that

1562: for some $d \geq 1$, we have $\psi^d \in End(\Lambda_\N)$ and

1563: $\beta^d_q \in \AA_q(\g)$ gauge equivalent to $\beta_q$, such that

1564: \begin{equation} \label{eq:discrepancy}

1565:  \beta^d_q - (\psi^d)^*\alpha_q \stackrel{\text{def}}{=} \epsilon_q \in O(q^d).

1566: \end{equation}

1567: Because $\delta\epsilon_q = \half

1568: [(\psi^d)^*\alpha_q+\beta^d_q,\epsilon_q] \in O(q^{d+1})$, the

1569: leading order term $\epsilon_{q,d} \in \g^1$ is closed. By

1570: assumption, one can therefore write $\beta^d_q = (\psi^d)^*\alpha_q +

1571: \epsilon_q = (\psi^d)^* \alpha_q + c q^d \alpha^{d,1} + q^d \delta b

1572: + O(q^{d+1})$ for some $c \in \C$, $b \in \g^0$. With

1573: \begin{equation} \label{eq:new-approximation}

1574:  \psi^{d+1}(q) = \psi^d(q) + c q^d, \quad

1575:  \beta^{d+1}_q = \exp(-q^d b)(\beta^d_q)

1576: \end{equation}

1577: one finds that \eqref{eq:discrepancy} holds for $d+1$ as well. Since

1578: the parameter change and gauge transformation in

1579: \eqref{eq:new-approximation} are equal to the identity up to

1580: $O(q^d)$, the process converges to $(\psi^\infty)^*\alpha_q =

1581: \beta^\infty_q$, with $\beta^\infty_q$ gauge equivalent to $\beta_q$.

1582: 

1583: When applied to $\g_\A = CC^{*+1}(\A,\A)^{\scriptscriptstyle \leq

1584: 0}$, this yields Lemma \ref{th:versal-2}. Similarly, for the coupled

1585: case one sets $\g_{\A \ampersand \M} = CC^{*+1}(\A \ampersand \M,\A

1586: \ampersand \M)^{\scriptscriptstyle \leq 0}$ and obtains Lemma

1587: \ref{th:versal-4}. The other results stated above, which lack the

1588: parameter $q$, require modifications of the basic framework. Let $A$

1589: be a graded algebra, and $\g_A = CC^{*+1}(A,A)$ its shifted

1590: Hochschild complex with the Gerstenhaber bracket. One can reformulate

1591: the bigrading of Hochschild cohomology as a $\C^*$-action on $\g_A$,

1592: which has weight $t$ on $CC^{s+t}(A,A)^t$. We combine this with the

1593: $\C^*$-action on $\Lambda_\N$ which has weight $d$ on $\C q^d$, and

1594: obtain an action on $\g_A \otimes q\Lambda_\N$. Given an

1595: $A_\infty$-structure $\A$ on $A$ satisfying \eqref{eq:mu1zero}, one

1596: can define a one-parameter deformation $\A_q$ of the trivial

1597: $A_\infty$-structure by $\mu^d_{\A_q} = q^{d-2}\mu^d_\A$, and this is

1598: given by a $\C^*$-invariant element of $\g_A \otimes q\Lambda_\N$.

1599: The only possible changes of parameter in this context are the

1600: homogeneous ones $q \mapsto \epsilon q$, $\epsilon \in \C^*$, and

1601: with that in mind the proof of Lemma \ref{th:versal-1} is by a

1602: straightforward adaptation of Lemma \ref{th:versal-abstract}. In

1603: contrast, Lemma \ref{th:versal-3} requires a more substantial

1604: tweaking of the theory, where the basic object one considers is a dg

1605: Lie algebra $\g$ with a complete decreasing filtration $F_0\g = \g

1606: \supset F_1\g \supset \dots$ such that $\delta(F_i\g) \subset

1607: F_{i+1}\g$ and $[F_j\g,F_k\g] \subset F_{j+k}\g$. Take a solution

1608: $\alpha \in (F_1\g)^1$ of \eqref{eq:classical-mc}, and consider the

1609: corresponding deformed differential

1610: \[

1611: \delta_\alpha = \delta + [\alpha,\cdot].

1612: \]

1613: There is a series of obstruction groups, which are the terms

1614: $E_2^{d,1-d}$, $d \geq 2$, in the spectral sequence leading to the

1615: cohomology of $(\g,\delta_\alpha)$. A slightly refined version of the

1616: computation in the proof of Lemma \ref{th:versal-abstract} shows that

1617: if all of them vanish, then any other solution of

1618: \eqref{eq:classical-mc} with $\beta-\alpha \in F_2\g$ is equivalent

1619: to $\alpha$ by a gauge transformation, obtained by exponentiating the

1620: action of $(F_1\g)^0$. For our purpose, one applies this to $\g =

1621: \bigoplus_{e,f} Hom^*(A^{\otimes e} \otimes M \otimes A^{\otimes

1622: f},A)$, with $\delta$ and the bracket as suggested by

1623: \eqref{eq:module-equation}, and $F_i\g$ the filtration by length $i =

1624: e+f$. The obstruction groups are precisely those which occur in Lemma

1625: \ref{th:versal-3}.

1626: 

1627: \subsection{}

1628: The previous discussion generalizes effortlessly to

1629: $A_\infty$-categories. For didactic reasons, start with an ordinary

1630: category $C$ having finitely many objects $X_1,\dots,X_m$, and let

1631: $A$ be its total morphism algebra, which is linear over $R_m$.

1632: Consider Hochschild cochains that preserve this structure, which

1633: means that they are $R_m$-bimodule maps $A \otimes_{R_m} \dots

1634: \otimes_{R_m} A \rightarrow A$. By writing this out, one sees that

1635: the relevant complex can be generalized to arbitrary categories,

1636: leading to a Hochschild cohomology theory $HH^*(C,C)$. The same

1637: applies to Hochschild cohomology of $A_\infty$-categories, and all

1638: the other notions of deformation theory.

1639: 

1640: \begin{remark}

1641: For any category $C$ with finitely many objects, the Hochschild

1642: cohomology $HH^*(C,C)$ is the same as $HH^*(A,A)$ of the total

1643: morphism algebra. In other words, the inclusion of subcomplex of

1644: $R_m$-bilinear Hochschild cochains into the whole cochain complex is

1645: a quasi-isomorphism. The reason is that the relative bar resolution

1646: \[

1647: \cdots \longrightarrow A \otimes_{R_m} A \otimes_{R_m} \otimes A

1648: \longrightarrow A \otimes_{R_m} A \longrightarrow A

1649: \]

1650: is also a resolution by projective $A$-bimodules in the non-relative

1651: sense. By quasi-isomorphism invariance of abstract deformation

1652: theories, this shows that any $A_\infty$-structure $\A$ on $A$

1653: satisfying \eqref{eq:mu1zero} comes, up to gauge equivalence, from an

1654: $A_\infty$-category structure $\Cat$ on $C$. The previous observation

1655: combined with the standard spectral sequence argument also proves

1656: that for any finite $A_\infty$-category $\Cat$ with total morphism

1657: $A_\infty$-algebra $\A$, $HH^*(\Cat,\Cat) \iso HH^*(\A,\A)$, and

1658: similarly for the truncated version. This implies that the

1659: one-parameter deformation theory of $\A$ as an $A_\infty$-algebra is

1660: the same as that of $\Cat$ as an $A_\infty$-category. On occasion,

1661: these facts will allow us to be somewhat sloppy about the distinction

1662: between finite categories and their total morphism algebras.

1663: \end{remark}

1664: 

1665: To make the connection with Section \ref{sec:generators}, we should

1666: relate the deformation theory of $\Cat$ with that of its derived

1667: categories. However, this cannot be done in a truly satisfactory way

1668: within the restrictions of the present paper, because we have not

1669: introduced an $A_\infty$-category underlying $D^\pi(\Cat)$, and so we

1670: will limit ourselves to a single rudimentary statement, Let $\Cat_q$

1671: be a one-parameter deformation of $\Cat$. Suppose that we have a

1672: finite set $(X_1,\dots,X_m)$ of objects of $\Cat$ which are

1673: split-generators for $D^\pi(\Cat)$. Let $\A$ be their total morphism

1674: $A_\infty$-algebra, and $\A_q$ its one-parameter deformation induced

1675: from $\Cat_q$.

1676: 

1677: \begin{lemma} \label{th:deformation-generators}

1678: Assume that there are $Y_0,Y_1 \in Ob\, \Cat$ such that

1679: $H(hom(Y_0,Y_1))$ changes at first order in the deformation. More

1680: precisely, this means that the natural map

1681: \begin{equation} \label{eq:q2}

1682:  H(hom_{\Cat_q}(Y_0,Y_1) \otimes_{\Lambda_\N} \C[q]/q^2, \mu^1_{\Cat_q}) \otimes_{\C[q]/q^2} \C

1683:  \longrightarrow H(hom_\Cat(Y_0,Y_1),\mu^1_\Cat)

1684: \end{equation}

1685: is not an isomorphism. Then $\A_q$ is nontrivial at first order in

1686: $q$.

1687: \end{lemma}

1688: 

1689: \proof Suppose that the statement is false. After an $R_m$-linear

1690: gauge transformation, we can assume that $\A_q$ is a deformation of

1691: order $q^2$. Then the same holds for the induced one-parameter

1692: deformation $Tw\A_q$ of $Tw\A$. But the natural $A_\infty$-functor

1693: $\A_q \rightarrow \Cat_q$ extends to a $\Lambda_\N$-linear

1694: $A_\infty$-functor $Tw\F_q: Tw\A_q \rightarrow Tw\Cat_q$, which is

1695: full and faithful on cohomology. This means that \eqref{eq:q2} must

1696: be an isomorphism for all objects $Y_0,Y_1$ which are twisted

1697: complexes constructed from the $X_k$. Since that property carries

1698: over to direct summands, we have a contradiction. \qed

1699: 

1700: \section{Group actions\label{sec:group-actions}}

1701: 

1702: Semidirect products $\Lambda V \semidirect \Gamma$ of an exterior

1703: algebra with a finite abelian group appear naturally in the context

1704: of homological mirror symmetry for hypersurfaces in projective

1705: spaces. This is a consequence of Serre duality and Beilinson's

1706: description of $D^bCoh(\CP{n})$ \cite{beilinson78}. The Hochschild

1707: cohomology of such algebras can be computed by a version of the

1708: classical Hochschild-Kostant-Rosenberg (HKR) theorem

1709: \cite{hochschild-kostant-rosenberg62}, and this will later form the

1710: basis for applying the deformation theory from the previous section.

1711: Besides proving that theorem, we also use the opportunity to assemble

1712: some elementary facts about semidirect products, both of algebras and

1713: $A_\infty$-algebras. The exposition is by no means as general as

1714: possible, in particular, we shamelessly restrict ourselves to abelian

1715: groups whenever this makes things simpler. Dissatisfied readers may

1716: want to turn to \cite{montgomery92} and references therein.

1717: 

1718: \subsection{\label{subsec:actions-generalities}}

1719: Let $A$ be a graded algebra carrying an action $\rho$ of a finite

1720: group $\Gamma$. The semidirect product $A \semidirect \Gamma$ is the

1721: tensor product $A \otimes \C \Gamma$ with the twisted multiplication

1722: \[

1723:  (a_2 \otimes \gamma_2) (a_1 \otimes \gamma_1) =

1724:  a_2 \rho(\gamma_2)(a_1) \otimes \gamma_2\gamma_1.

1725: \]

1726: 

1727: The purpose of this construction is to turn the automorphisms

1728: $\rho(\Gamma)$ into inner ones: in fact, $(1 \otimes \gamma) (a

1729: \otimes e) (1 \otimes \gamma)^{-1} = \rho(\gamma)(a) \otimes e$.

1730: Therefore, it is not surprising that $A \semidirect \Gamma$ depends

1731: only on the ``outer part'' of the group action, in the following

1732: sense: suppose that $\tilde{\rho}$ is another action related to the

1733: previous one by

1734: \begin{equation} \label{eq:conjugate-action}

1735:  \tilde{\rho}(\gamma) = \iota(\gamma) \rho(\gamma) \iota(\gamma)^{-1},

1736: \end{equation}

1737: where $\iota: \Gamma \rightarrow (A^0)^\times$ is a map satisfying

1738: $\iota(\gamma_2\gamma_1) = \iota(\gamma_2)

1739: \rho(\gamma_2)(\iota(\gamma_1))$. Then $A \semidirect_\rho \Gamma$

1740: and $A \semidirect_{\tilde\rho} \Gamma$ are isomorphic by $a \otimes

1741: \gamma \mapsto a \iota(\gamma)^{-1} \otimes \gamma$. This kind of

1742: consideration appears naturally in connection with the following

1743: construction: suppose that there is a normal subgroup $\Gamma_0

1744: \subset \Gamma$ such that the projection $\Gamma \rightarrow

1745: \Gamma/\Gamma_0$ admits a right inverse $\Sigma$. Then $\Gamma \iso

1746: \Gamma_0 \semidirect (\Gamma/\Gamma_0)$, and there is a corresponding

1747: isomorphism

1748: \begin{equation} \label{eq:in-stages}

1749:  A \semidirect \Gamma

1750:  \iso (A \semidirect \Gamma_0) \semidirect (\Gamma/\Gamma_0).

1751: \end{equation}

1752: More precisely, let $\rho_\Gamma$ be the given $\Gamma$-action, and

1753: $\rho_{\Gamma_0}$ its restriction. Through the splitting $\Sigma$ we

1754: have an action $\rho_{\Gamma/\Gamma_0}$ of the quotient on $A

1755: \semidirect \Gamma_0$:

1756: \begin{equation} \label{eq:quotient-action}

1757:  \rho_{\Gamma/\Gamma_0}(\bar\gamma)(a \otimes \gamma_0) =

1758:  \rho_{\Gamma}(\Sigma(\bar\gamma))(a) \otimes

1759:  \Sigma(\bar\gamma) \gamma_0 \Sigma(\bar\gamma)^{-1},

1760: \end{equation}

1761: and these are the three actions involved in \eqref{eq:in-stages}.

1762: Different choices of $\Sigma$ lead to actions

1763: \eqref{eq:quotient-action} which differ by inner automorphisms as in

1764: \eqref{eq:conjugate-action}, and the discussion there shows that the

1765: right hand side of \eqref{eq:in-stages} is independent of the choice,

1766: as indeed it ought to be.

1767: 

1768: Any semidirect product $A \semidirect \Gamma$ carries a natural

1769: action $\rho_{\Gamma^*}$ of the group of scalar characters $\Gamma^*

1770: = Hom(\Gamma,\C^*)$; $\rho_{\Gamma^*}(\chi)(a \otimes g) = a\chi(g)

1771: \otimes g$. Additional automorphisms can be produced from those of

1772: $A$. We have already seen an instance of this in

1773: \eqref{eq:quotient-action}, and the general construction goes as

1774: follows: let $U$ be an automorphism of $A$, and suppose that there is

1775: an automorphism $\xi$ of $\Gamma$ as well as a map $\iota: \Gamma

1776: \rightarrow (A^0)^\times$ satisfying $\iota(\gamma_2\gamma_1) =

1777: \rho(\xi(\gamma_2))(\iota(\gamma_2)) \cdot \iota(\gamma_1)$, such

1778: that

1779: \begin{equation} \label{eq:inner-twist}

1780:  \rho(\xi(\gamma))(U(a))

1781:  = \iota(\gamma) U(\rho(\gamma)(a)) \iota(\gamma)^{-1}

1782: \end{equation}

1783: holds. Informally, this means that $U$ commutes with the group

1784: actions in a twisted sense given by $\xi$, and up to inner

1785: automorphism which are conjugations with the $\iota$'s. Then one can

1786: define an induced automorphism of $A \semidirect \Gamma$ by

1787: \begin{equation} \label{eq:twisted-u}

1788:  (U \semidirect \xi)(a \otimes \gamma) = U(a)\iota(\gamma)^{-1}

1789:  \otimes \xi(\gamma).

1790: \end{equation}

1791: Of course, $\iota $ is not unique: for instance, passing from

1792: $\iota(\gamma)$ to $\iota(\gamma)\chi(\gamma)$ for some $\chi \in

1793: \Gamma^*$ amounts to composing $U \semidirect \xi$ with

1794: $\rho_{\Gamma^*}(\chi)$.

1795: 

1796: In a slightly different direction, note that $A \semidirect \Gamma$

1797: is linear over the semisimple ring $\C \Gamma$. For simplicity,

1798: assume that $\Gamma$ is abelian, so that $\C \Gamma \iso

1799: R_{|\Gamma|}$ has a basis of orthogonal idempotents, which are the

1800: normalized characters $u_\chi = |G|^{-1}\chi$. This means that we can

1801: consider $A \semidirect \Gamma$ as a category with objects $\chi$ and

1802: morphism spaces

1803: \[

1804:  Hom(\chi_1,\chi_2) = u_{\chi_2} (A \semidirect \Gamma)

1805:  u_{\chi_1} = A^{\chi_1/\chi_2} \otimes \C u_{\chi_1}.

1806: \]

1807: where $\chi_2/\chi_1$ is the (pointwise) division in $\Gamma^*$, and

1808: $A^{\chi}$ is the subspace on which $G$ acts as on the representation

1809: corresponding to $\chi$. This suggests an obvious generalization,

1810: which is to consider an action of an abelian group $\Gamma$ on a

1811: linear graded category $C$; here and in related discussions below,

1812: the standing assumption is that the action on the set of objects is

1813: trivial. Then, the semidirect product category $C \semidirect \Gamma$

1814: has objects $(X,\chi) \in Ob\,C \times \Gamma^*$, and morphisms

1815: $Hom((X_1,\chi_1),(X_2,\chi_2)) = Hom(X_1,X_2)^{\chi_1/\chi_2}$. By

1816: construction, if $C$ has only finitely many objects and $A$ it its

1817: total morphism algebra, then $A \semidirect \Gamma$ is the total

1818: morphism algebra of $C \semidirect \Gamma$.

1819: 

1820: \subsection{\label{subsec:ainfty-actions}}

1821: A $\Gamma$-action on an $A_\infty$-algebra $\A$ is a linear action on

1822: $A$, such that all the $\mu_\A^d$ are equivariant. The semidirect

1823: product $A_\infty$-algebra $\A \semidirect \Gamma$ is then defined as

1824: $\A \otimes \C\Gamma$ with the compositions

1825: \begin{multline*}

1826:  \qquad \mu_{\A \semidirect \Gamma}^d(a_d \otimes \gamma_d, \dots, a_1 \otimes \gamma_1) = \\ =

1827:  \mu_\A^d(a_d, \rho(\gamma_d)(a_{d-1}), \rho(\gamma_d\gamma_{d_1})(a_{d-2}), \dots)

1828:  \otimes \gamma_d \cdots \gamma_1. \qquad

1829: \end{multline*}

1830: This generalizes to $A_\infty$-categories with $\Gamma$-actions; if

1831: there are only finitely many objects, these are the same as

1832: $R_m$-linear $\Gamma$-actions on $R_m$-linear $A_\infty$-algebras.

1833: Most of the discussion above applies to the $A_\infty$-case as well,

1834: the only problematic parts being those involving invertible elements

1835: and the associated inner automorphisms, since that notion does not

1836: carry over in a simple way. In the $R_m$-linear situation there is at

1837: least a rudimentary substitute. Namely, for $r = (r_1,\dots,r_m) \in

1838: I = (\C^*)^m$ and $a \in u_j A u_k$, define $r \cdot a = r_j a$ and

1839: $a \cdot r = a r_k$. In a slight abuse of terminology, we call $a

1840: \mapsto r \cdot a \cdot r^{-1}$ an {\em elementary inner

1841: automorphism}. One can use these instead of actual $xax^{-1}$'s in

1842: formulae like \eqref{eq:conjugate-action}, \eqref{eq:inner-twist},

1843: and this gives a valid $A_\infty$-version of the relevant

1844: observations for algebras, even though it is so reduced in scope as

1845: to be basically trivial.

1846: 

1847: Given an $A_\infty$-category $\Cat$ with a $\Gamma$-action, one can

1848: define the $A_\infty$-category $Tw_{\Gamma} \Cat$ of equivariant

1849: twisted complexes. This also carries a $\Gamma$-action, and if we

1850: forget that action, it becomes equivalent to a full

1851: $A_\infty$-subcategory of the ordinary $Tw\,\Cat$. The construction

1852: requires the following changes in the two basic steps: (1) the

1853: equivariant version $\Sigma_\Gamma\Cat$ of the additive completion

1854: has objects which are formal direct sums

1855: \begin{equation} \label{eq:bigoplus-f}

1856:  \bigoplus_{f \in F} (X_f \otimes V_f)[\sigma_f]

1857: \end{equation}

1858: where $V_f$ is some finite-dimensional representation of $\Gamma$.

1859: The morphism spaces are correspondingly

1860: \begin{multline*}

1861:  \qquad hom_{\Sigma_\Gamma\Cat}\Big(\bigoplus_f (X_f \otimes V_f)[\sigma_f],

1862:  \bigoplus_g (Y_g \otimes W_g)[\tau_g]\Big) = \\ =

1863:  \bigoplus_{f,g} hom_{\Cat}(X_f,Y_g)[\tau_g-\sigma_f] \otimes Hom_\C(V_f,W_g) \qquad

1864: \end{multline*}

1865: and they carry the obvious tensor product $\Gamma$-actions. (2) In

1866: the definition of an equivariant twisted complex, the differential

1867: $\delta$ has to belong to the $\Gamma$-invariant part of $hom^1$. The

1868: ``equivariant derived category'' $D^b_\Gamma(\Cat) =

1869: H^0(Tw_\Gamma\,\Cat)$ carries, in addition to the usual shift functor

1870: $[1]$, automorphisms $- \otimes \chi$ for any character $\chi \in

1871: \Gamma^*$, which change all $V_f$ in \eqref{eq:bigoplus-f} by

1872: tensoring with the corresponding one-dimensional representation.

1873: 

1874: \begin{remark} \label{th:twist-is-equivariant}

1875: $D^b_\Gamma(\Cat)$ is {\em not} triangulated, since mapping cones

1876: exist only for the $\Gamma$-invariant morphisms. Fortunately, the

1877: evaluation maps $ev$, $ev^\vee$ used in the definition of the twist

1878: and untwist operations are invariant, so $T_X(Y)$, $T_Y^\vee(X)$ are

1879: well-defined in $D^b_\Gamma(\Cat)$ (at least assuming that the

1880: necessary finite-dimensionality conditions hold). In particular, the

1881: theory of exceptional collections and mutation works well in the

1882: equivariant context.

1883: \end{remark}

1884: 

1885: \subsection{}

1886: Because we are working over a characteristic zero field, taking

1887: $\Gamma$-invariants is an exact functor. This means that the

1888: classification theory from the previous section extends in a

1889: straightforward way to the equivariant situation: one simply replaces

1890: all ``obstruction groups'' by their $\Gamma$-fixed parts. More

1891: concretely, suppose that we have a finite $\Gamma$ acting on a graded

1892: algebra $A$. Then $HH^*(A,A)^\Gamma$ is the cohomology of the

1893: subcomplex of $CC^*(A,A)^\Gamma$ of equivariant Hochschild cochains.

1894: There is a canonical map $J^\Gamma: HH^*(A,A)^\Gamma \rightarrow

1895: HH^*(A \semidirect \Gamma, A \semidirect \Gamma)$, induced by the

1896: cochain homomorphism

1897: \begin{equation} \label{eq:invariant-include}

1898: \begin{aligned}

1899:  & (J^\Gamma\tau)(a_s \otimes \gamma_s,\dots,a_1 \otimes \gamma_1) = \\

1900:  & \qquad \qquad = \tau(a_s,\rho(\gamma_s)(a_{s-1}),\rho(\gamma_s

1901:  \gamma_{s-1})(a_{s-2}),\dots) \otimes \gamma_s \cdots \gamma_1.

1902: \end{aligned}

1903: \end{equation}

1904: This is compatible with the Lie bracket on $HH^{*+1}$, and has an

1905: obvious deformation theoretic meaning: $\Gamma$-equivariant

1906: $A_\infty$-deformations of $A$ induce deformations of the semidirect

1907: product. In fact, \eqref{eq:invariant-include} is always injective.

1908: One can see this in several different ways. For a classical

1909: homological algebra argument, note that if $P$ is an $A$-bimodule

1910: carrying a $\Gamma$-action which is compatible with that on $A$, then

1911: $P \semidirect \Gamma$ is an $A \semidirect \Gamma$-bimodule, which

1912: for any $A \semidirect \Gamma$-bimodule $M$ satisfies

1913: \[

1914:  Hom_{A \semidirect \Gamma-A \semidirect \Gamma}(P \semidirect \Gamma, M)

1915:  \iso Hom_{A-A}(P,M)^\Gamma;

1916: \]

1917: here the $\Gamma$-action on the right hand side is obtained from that

1918: on $P$ and the diagonal action on $M$, which sends $x$ to $(1 \otimes

1919: \gamma)x(1 \otimes \gamma^{-1})$. In particular, if $P$ is projective

1920: then so is $P \semidirect \Gamma$. Applying this observation to a

1921: projective $A$-bimodule resolution $P^* \rightarrow A$ which carries

1922: a compatible $\Gamma$-action, one obtains

1923: \[

1924: \begin{aligned}

1925:  HH^*(A \semidirect \Gamma, A \semidirect \Gamma)

1926:  & = Ext_{A \semidirect \Gamma - A \semidirect \Gamma}^*(A \semidirect \Gamma,

1927:  A \semidirect \Gamma) \\

1928:  & = Hom_{A \semidirect \Gamma - A \semidirect \Gamma}^*(P^* \semidirect \Gamma,A

1929:  \semidirect \Gamma) \\

1930:  & = Hom^*_{A-A}(P^*, A \semidirect \Gamma)^\Gamma.

1931: \end{aligned}

1932: \]

1933: Under the diagonal action, $A \semidirect \Gamma$ splits into

1934: summands corresponding to conjugacy classes $K \subset \Gamma$, and

1935: this induces a splitting of $HH^*(A \semidirect \Gamma,A \semidirect

1936: \Gamma)$. In particular, the trivial class contributes

1937: $Hom^*_{A-A}(P^*,A)^\Gamma = HH^*(A,A)^\Gamma$, which is precisely

1938: the image of \eqref{eq:invariant-include}. More generally, the

1939: summand corresponding to $K$ is the direct sum of the graph bimodules

1940: $Graph(\rho(\gamma))$ for $\gamma \in K$, so that

1941: \begin{equation} \label{eq:summands}

1942:  HH^*(A \semidirect \Gamma,A \semidirect \Gamma) =

1943:  \bigoplus_K Ext^*_{A-A}\Big(A,\bigoplus_{\gamma \in K} Graph(\rho(\gamma))\Big)^\Gamma.

1944: \end{equation}

1945: Altenatively, one can approach the injectivity of $J^\Gamma$ directly

1946: in terms of the Hochschild complex. This involves looking at $A

1947: \semidirect \Gamma$ as an algebra linear over the semisimple ring $\C

1948: \Gamma$. In the abelian case, this can be put more forcefully by

1949: saying that we consider the $\Gamma^*$-action on $HH^*(A \semidirect

1950: \Gamma, A \semidirect \Gamma)$ induced from that on $A \semidirect

1951: \Gamma$ itself. The decomposition into different types of

1952: $\Gamma^*$-modules is precisely \eqref{eq:summands}, with the

1953: invariant part being the image of $J^\Gamma$. This ultimately finds a

1954: natural explanation in terms of the duality theory for group actions,

1955: which says in particular that $(A \semidirect \Gamma) \semidirect

1956: \Gamma^*$ is Morita equivalent to $A$ \cite{blattner-montgomery85}.

1957: 

1958: The HKR theorem describes the Hochschild cohomology of polynomial

1959: algebras, and more general smooth commutative algebras. There is a

1960: generalization to graded algebras \cite[Proposition 5.4.6]{loday},

1961: including exterior algebras as a special case, for which the

1962: statement is as follows. Take $A = \Lambda V$ for some

1963: finite-dimensional complex vector space $V$, with the usual grading.

1964: Let $SV^\vee$ be the symmetric algebra of the dual space. Then there

1965: is an explicit isomorphism

1966: \begin{equation} \label{eq:classical-hkr}

1967:  \phi_{HKR}: S^sV^\vee \otimes \Lambda^{s+t}V

1968:  \longrightarrow HH^{s+t}(A,A)^t.

1969: \end{equation}

1970: For instance, the cocycle $[\tau] = \phi_{HKR}(p \otimes 1)$

1971: corresponding to a homogeneous polynomial $p$ of degree $s$ is (in

1972: terms of a basis $v_j$ and dual basis $v_j^\vee$)

1973: \[

1974:  \tau(a_s, \dots, a_1) = \frac{1}{s!} \sum_{j_s,\dots,j_1}

1975:  \frac{\partial p}{\partial v_{j_1} \dots \partial v_{j_s}}

1976:  (v_{j_s}^\vee \inner a_s) \wedge \dots \wedge (v_{j_1}^\vee \inner a_1);

1977: \]

1978: and conversely, if $\tau \in CC^0(A,A)^{-s}$ is any cocycle, we

1979: recover a homogeneous polynomial $p \otimes 1 =

1980: \phi_{HKR}^{-1}([\tau])$ by setting

1981: \begin{equation} \label{eq:reverse-hkr}

1982:  p(v) = \tau(v,\dots,v).

1983: \end{equation}

1984: $\phi_{HKR}$ is compatible with the actions of $GL(V)$ on both sides

1985: (where the action on $V^\vee$ is contragredient to that on $V$).

1986: Moreover, it carries the canonical bracket on $HH^{*+1}(A,A)$ to the

1987: Schouten bracket on $SV^\vee \otimes \Lambda V$, which in particular

1988: satisfies

1989: \begin{equation} \label{eq:schouten}

1990:  [v_{i_1}^\vee \dots v_{i_r}^\vee \otimes 1,

1991:  v_{i_{r+1}}^\vee \dots v_{i_s}^\vee \otimes a] =

1992:  \sum_{k \leq r}

1993:  v_{i_1}^\vee \dots \widehat{v_{i_k}^\vee} \dots v_{i_s}

1994:  \otimes v_{i_k}^\vee \inner a

1995: \end{equation}

1996: for $a \in \Lambda V$. The standard proof of the HKR formula is via

1997: the Koszul bimodule resolution $P^* \rightarrow A$, which in our case

1998: consists of $P^s = A \otimes S^s V \otimes A[-s]$ with the

1999: differential

2000: \begin{multline*}

2001:  \qquad \partial(1 \otimes v_{i_1} \dots v_{i_s} \otimes 1) =

2002:  \sum_k v_{i_k} \otimes v_{i_1} \dots \widehat{v_{i_k}} \dots v_{i_s}

2003:  \otimes 1 + \\

2004:  +(-1)^s 1 \otimes

2005:  v_{i_1} \dots \widehat{v_{i_k}} \dots v_{i_s} \otimes v_{i_k}. \qquad

2006: \end{multline*}

2007: Because $A$ is supercommutative, the induced differential on

2008: $Hom(P^*,A)$ vanishes, which shows that an abstract isomorphism

2009: \eqref{eq:classical-hkr} exists; of course, the explicit formula and

2010: other properties have to be verified separately, by exhibiting a map

2011: between the Koszul and bar resolutions.

2012: 

2013: We now look at \eqref{eq:summands} in the special case where $A$ is

2014: an exterior algebra, and $\rho$ the action of a finite abelian group

2015: $\Gamma \subset GL(V)$. Taking the Koszul resolution as $P^*$, one

2016: sees that the $\gamma$-summand of $HH^*(A \semidirect \Gamma,A

2017: \semidirect \Gamma)$ can be obtained by computing the cohomology of

2018: the complex $Hom_{\C}(SV,\Lambda V) \iso SV^\vee \otimes \Lambda V$

2019: with the differential

2020: \[

2021:  (\delta_\gamma\eta)(v_{i_1} \dots v_{i_s}) =

2022:  (-1)^s \sum_k \eta(v_{i_1} \dots \widehat{v_{i_k}} \dots

2023:  v_{i_s}) \wedge (\gamma(v_{i_k}) - v_{i_k}),

2024: \]

2025: and then restricting to the $\Gamma$-invariant part of that. To carry

2026: out the computation, we decompose $V$ into eigenspaces of $\gamma$,

2027: and consider the corresponding tensor product decomposition of $S

2028: V^\vee \otimes \Lambda V$. On the fixed part $V^\gamma$, the

2029: differential vanishes as before; the remaining components can be

2030: identified with classical Koszul complexes, which have

2031: one-dimensional cohomology. After chasing the bidegrees and

2032: $\Gamma$-action through the argument, this amounts to a proof of the

2033: following:

2034: 

2035: \begin{prop} \label{th:twisted-hkr}

2036: The Hochschild cohomology of $A = \Lambda V \semidirect \Gamma$, for

2037: $\Gamma$ a finite abelian subgroup of $GL(V)$, is

2038: \begin{multline*}

2039:  HH^{s+t}(A, A)^t \iso \\ \iso

2040:  \bigoplus_{\gamma \in \Gamma} \Big( S^s(V^\gamma)^\vee \otimes

2041:  \Lambda^{s+t-codim(V^\gamma)}(V^\gamma) \otimes

2042:  \Lambda^{codim(V^\gamma)}(V/V^\gamma)

2043:  \Big)^\Gamma. \qed

2044: \end{multline*}

2045: \end{prop}

2046: 

2047: \section{Coherent sheaves\label{sec:sheaves}}

2048: 

2049: To apply the theory from Section \ref{sec:deformations} to the

2050: derived category of coherent sheaves, one needs to introduce an

2051: underlying differential graded category. Concretely, this means

2052: choosing a class of cochain complexes which compute $Ext$ groups,

2053: with cochain maps realizing the Yoneda product. Out of the many

2054: possibilities offered by the literature, we take the lowest-tech path

2055: via {\v C}ech cohomology, which is easy to work with but slightly

2056: more difficult to relate to the usual derived category (a few readers

2057: may prefer to skip that argument, and to take the {\v C}ech version

2058: as definition of the derived category; in which case some of the

2059: properties discussed later on have to be taken for granted).

2060: 

2061: \subsection{}

2062: Let $Y$ be a noetherian scheme over an algebraically closed field

2063: ($\C$ or $\Lambda_\Q$, in the cases of interest here). Fix a finite

2064: affine open cover $\UU$. We introduce a differential graded category

2065: $\cech{\SS}(Y)$ whose objects are locally free coherent sheaves

2066: (algebraic vector bundles), and where the morphism spaces are {\v

2067: C}ech cochain complexes with values in $\SHom$ sheaves,

2068: \[

2069: hom_{\cech{\SS}(Y)}(E,F) = \cech{C}^*(\UU,\SHom(E,F)).

2070: \]

2071: Composition is the shuffle product together with $\SHom(F,G)

2072: \otimes_{\OO_Y} \SHom(E,F) \rightarrow \SHom(E,G)$. One can think of

2073: any dg category as an $A_\infty$-category with vanishing compositions

2074: of order $\geq 3$, and therefore define a triangulated category

2075: $D^b\cech{\SS}(Y)$ in the way explained in Section

2076: \ref{sec:generators}, as the cohomological category of twisted

2077: complexes (we emphasize that there is no ``inverting of

2078: quasi-isomorphisms'' involved in this process). By refining covers

2079: and using standard results about {\v C}ech cohomology, one sees that

2080: $\cech{\SS}(Y)$ is independent of $\UU$ up to quasi-isomorphism,

2081: hence that $D^b\cech{\SS}(Y)$ is well-defined up to equivalence of

2082: triangulated categories.

2083: 

2084: \begin{lemma} \label{th:derived-is-derived}

2085: $D^b\cech{\SS}(Y)$ is equivalent to the full triangulated subcategory

2086: of the usual bounded derived category of coherent sheaves,

2087: $D^bCoh(Y)$, which is generated by locally free sheaves. If $Y$ is a

2088: regular projective variety, this is the whole of $D^bCoh(Y)$.

2089: \end{lemma}

2090: 

2091: \proof The usual approach to dg structures on derived categories of

2092: coherent sheaves is through injective resolutions, see for instance

2093: \cite[Section 2]{seidel-thomas99}. In our current language, this can

2094: be formulated as follows. For each locally free coherent sheaf $E$,

2095: fix a resolution $I^*_E$ by quasi-coherent injective sheaves.

2096: Consider a differential graded category $\I(Y)$ with the same objects

2097: as $\cech{\SS}(Y)$, but where $hom_{\I(Y)}(E,F) =

2098: hom_{\O_Y}(I^*_E,I^*_F)$ is the space of sheaf homomorphisms $I^*_E

2099: \rightarrow I^*_F$ of all degrees, with the obvious differential. If

2100: we then define $D^b\I(Y) = H^0(Tw\,\I(Y))$ through twisted complexes,

2101: it is equivalent to the subcategory of $D^bCoh(Y)$ appearing in the

2102: Lemma. To see that, one associates to each object of $Tw\,\I(Y)$ its

2103: total complex; this defines an exact, fully faithful functor from

2104: $D^b\I(Y)$ to the chain homotopy category ${\mathcal K}^+(Y)$ of

2105: bounded below complexes of injective quasi-coherent sheaves. The

2106: image of this functor is the triangulated subcategory generated by

2107: resolutions of locally free sheaves, and by standard properties of

2108: derived categories, this is equivalent to the triangulated

2109: subcategory of $D^bCoh(Y)$ generated by such sheaves. If $Y$ is

2110: smooth and projective, all coherent sheaves have finite locally free

2111: resolutions, so we get the whole of $D^bCoh(Y)$.

2112: 

2113: To mediate between {\v C}ech and injective resolutions, we use a

2114: triangular matrix construction. Given two locally free sheaves $E$

2115: and $F$, one can consider the {\v C}ech complex

2116: $\cech{C}^*(\SHom(E,I_F^*))$ with coefficients in the complex of

2117: sheaves $\SHom(E,I_F^*) \iso E^\vee \otimes_{\O_Y} I_F^*$. There are

2118: canonical cochain maps

2119: \begin{equation} \label{eq:t-maps}

2120:  hom_{\O_Y}(I_E^*,I_F^*) \longrightarrow

2121:  \cech{C}^*(\SHom(E,I_F^*)) \longleftarrow \cech{C}^*(\SHom(E,F)).

2122: \end{equation}

2123: The first one is defined by composing a sheaf homomorphism $I_E^*

2124: \rightarrow I_F^*$ with $E \rightarrow I_E^0$, and then restricting

2125: the associated section of $\SHom(E,I_F^*)$ to open subsets in $\UU$.

2126: The second one is just the map induced from $F \rightarrow I_F^0$.

2127: Both maps are quasi-isomorphisms: for the first map, note that the

2128: {\v C}ech complex computes the hypercohomology of $E^\vee \otimes

2129: I_F^*$, which on the other hand that is the ordinary cohomology of

2130: $hom_{\O_Y}(E^*,I_F^*)$ because each $E^\vee \otimes I_F^k$ is

2131: injective. For the second map one uses a standard spectral sequence

2132: comparison argument, together with the fact that the cohomology of

2133: $hom_{\O_U}(E|U, I_F^*|U)$ for each open subset $U$ is

2134: $Hom_{\O_U}(E|U,F|U)$. Along the same lines, there are canonical

2135: homomorphisms

2136: \begin{equation} \label{eq:t-bistructure}

2137: \begin{split}

2138:  & hom_{\O_Y}(I_F^*,I_G^*) \otimes \cech{C}^*(\SHom(E,I_F^*))

2139:  \longrightarrow \cech{C}^*(\SHom(E,I_G^*)), \\

2140:  & \cech{C}^*(\SHom(F,I_G^*)) \otimes \cech{C}^*(\SHom(E,F))

2141:  \longrightarrow \cech{C}^*(\SHom(E,I_G^*)).

2142: \end{split}

2143: \end{equation}

2144: We introduce another dg category $\T$, still keeping the same objects

2145: as before, and where the $hom$s are the direct sums of three

2146: components

2147: \[

2148:  hom_\T(E,F) =

2149:  \begin{pmatrix}

2150:  hom_{\O_Y}(I_E^*,I_F^*) & \cech{C}^*(\SHom(E,I_F^*))[1] \\

2151:  0 & \cech{C}^*(\SHom(E,F)).

2152:  \end{pmatrix}

2153: \]

2154: The differential consists of the given differentials on each summand

2155: together with the chain homomorphisms \eqref{eq:t-maps}. Similarly,

2156: composition in $\T$ combines the usual compositions of morphisms in

2157: $\I(Y)$ and $\cech{\SS}(Y)$ with the maps \eqref{eq:t-bistructure}.

2158: By definition, this comes with dg functors $\T \rightarrow \I(Y)$,

2159: $\T \rightarrow \cech{\SS}(Y)$, both of which are quasi-isomorphisms,

2160: and therefore one has induced exact equivalences $D^b\cech{\SS}(Y)

2161: \iso D^b\T \iso D^b\I(Y)$. \qed

2162: 

2163: \begin{remark} \label{th:equivariant-sheaves}

2164: At least if one supposes that the ground field has characteristic

2165: zero, the entire argument carries over to the derived category of

2166: equivariant coherent sheaves with respect to the action of a finite

2167: group $\Gamma$ on $Y$. First, by taking intersections over

2168: $\Gamma$-orbits one can find an invariant affine open cover. One then

2169: introduces the dg category $\cech{S}_\Gamma(Y)$ whose objects are

2170: locally free sheaves with $\Gamma$-actions, and whose morphism spaces

2171: are the $\Gamma$-invariant summands of the {\v C}ech complexes. In

2172: the regular projective case,

2173: \[

2174:  D^b\cech{\SS}_\Gamma(Y) \iso D^bCoh_\Gamma(Y),

2175: \]

2176: where $Coh_\Gamma(Y)$ is the abelian category of $\Gamma$-equivariant

2177: coherent sheaves. We will now make the connection between this

2178: observation and the material from Section \ref{sec:group-actions}.

2179: Suppose for simplicity that $\Gamma$ is abelian. If $E$ is a coherent

2180: sheaf with a $\Gamma$-action, then so is $E \otimes W$ for any

2181: finite-dimensional $\Gamma$-module $W$. Suppose for simplicity that

2182: $\Gamma$ is abelian, and take equivariant locally free sheaves

2183: $E_1,\dots,E_m$. Then the full dg subcategory $\Cat \subset

2184: \cech{\SS}(Y)$ with objects $E_k$ carries a $\Gamma$-action, and the

2185: semidirect product $\Cat \semidirect \Gamma$ can be identified with

2186: the full dg subcategory of $\cech{\SS}_\Gamma(Y)$ whose objects are

2187: $E_k \otimes W_\chi$ for all $k$ and all one-dimensional

2188: representations $W_\chi$.

2189: \end{remark}

2190: 

2191: \subsection{\label{subsec:formal-schemes}}

2192: Let $Y_q$ be a noetherian formal scheme, flat over $\Lambda_\N$. By

2193: definition \cite[ch.\ II \S 9]{hartshorne} there is a finite cover

2194: $\UU$ by open subsets, on each of which the scheme is ``given by''

2195: (actually, the formal spectrum of) a commutative algebra $A_{q,U}$

2196: over $\Lambda_\N$ which is complete and torsion-free as a

2197: $\Lambda_\N$-module. We consider locally free coherent sheaves over

2198: $Y_q$, which are $\O_{Y_q}$-module sheaves given over each $U$ by

2199: sections of a finitely generated free $A_{q,i}$-module. By the same

2200: procedure as before, one can make these into objects of a dg category

2201: $\cech{\SS}(Y_q)$ whose morphism spaces are again complete

2202: torsion-free $\Lambda_\N$-modules.

2203: 

2204: Geometrically, $Y_q$ is an infinitesimal one-parameter family of

2205: schemes. One can specialize the parameter $q$ to two different

2206: ``values''. Setting $q = 0$ yields the {\em special fibre} $Y = Y_q

2207: \times_{\Lambda_\N} \C$, and on the other hand, inverting $q$ and

2208: introducing formal roots of it gives the {\em general fibre} $Y_q^* =

2209: Y_q \times_{\Lambda_\N} \Lambda_\Q$. Both of them are noetherian

2210: schemes over algebraically closed fields. There are parallel

2211: constructions in the world of our differential graded categories,

2212: namely full and faithful dg functors

2213: \begin{equation} \label{eq:specialization}

2214:  \cech{\SS}(Y_q) \otimes_{\Lambda_\N} \C \longrightarrow \cech{\SS}(Y_0), \qquad

2215:  \cech{\SS}(Y_q) \otimes_{\Lambda_\N} \Lambda_\Q \longrightarrow \cech{\SS}(Y_q^*).

2216: \end{equation}

2217: To see what this means concretely, pick a finite collection of

2218: locally free sheaves $E_{0,1},\dots,E_{0,m}$ on $Y_0$, and suppose

2219: that each $E_{0,k}$ admits an extension $E_{q,k}$ to a locally free

2220: sheaf on $Y_q$, whose restriction to $Y_q^*$ we denote by

2221: $E_{q,k}^*$. The fact that the first map in \eqref{eq:specialization}

2222: is full and faithful on the cochain level means that the full dg

2223: subcategory of $\cech{\SS}(Y_q)$ with objects $E_{q,k}$ is a

2224: one-parameter deformation of the full dg subcategory of

2225: $\cech{\SS}(Y_0)$ with objects $E_{0,k}$. Moreover, if we take the

2226: first of these subcategories and tensor with $\Lambda_\Q$, we obtain

2227: the full dg subcategory of $\cech{\SS}(Y_q^*)$ with objects

2228: $E_{q,k}^*$. In this way, deformations of varieties and their effects

2229: on coherent sheaves have been brought into the fold of the abstract

2230: theory from Section \ref{sec:deformations}. This is still quite

2231: crude, for instance it excludes totally those sheaves on $Y_0$ that

2232: might not admit extensions to $Y_q$. One can give a much more natural

2233: and satisfactory treatment using obstructed dg or $A_\infty$

2234: structures, see for instance \cite{fukaya00}, but for our application

2235: the rudimentary version is sufficient.

2236: 

2237: \subsection{}

2238: We end our discussion by listing two properties of $D^bCoh(Y)$ for a

2239: regular projective variety over a field, which are well-known to

2240: specialists (I am indebted to Drinfeld for suggesting simplified

2241: proofs and giving references). The first result is classical, and

2242: explains why passage to $D^\pi$ is unnecessary.

2243: 

2244: \begin{lemma}

2245: $D^bCoh(Y)$ is idempotent closed.

2246: \end{lemma}

2247: 

2248: One way to prove this is to consider the chain homotopy category

2249: $\K^+(Y)$, which is idempotent closed by an infinite direct sum trick

2250: \cite[Proposition 1.6.8]{neeman} or \cite[Lemma 2.4.8.1]{levine}.

2251: Having done that, one recognizes that the direct summand of any

2252: object of $D^b(Y) \subset \K^+(Y)$ again lies in $D^b(Y)$, because

2253: its cohomology sheaves are coherent and almost all zero

2254: \cite[Proposition 2.4]{seidel-thomas99}. Alternatively, one can use

2255: the stronger result from \cite{bondal-vandenbergh02} which says that

2256: the bounded derived category is {\em saturated}.

2257: 

2258: \begin{lemma} \label{th:kontsevich}

2259: If $\iota: Y \hookrightarrow {\mathbb P}^N$ is a projective

2260: embedding, and $F_1,\dots,F_m$ are split-generators for

2261: $D^bCoh({\mathbb P}^N)$, their derived restrictions $E_k =

2262: L\iota^*(F_k)$ are split-generators for $D^bCoh(Y)$.

2263: \end{lemma}

2264: 

2265: This is due to Kontsevich, and the proof follows the same pattern as

2266: Lemma \ref{th:twist-generators}. By assumption, on ${\mathbb P}^N$

2267: one can express $\O(m)$ for any $m$ as direct summand of a complex

2268: built from the $F_k$. Hence, on $Y$ one can express $\O(m)$ in the

2269: same way using $E_k$. It therefore suffices to show that all the

2270: $\O(m)$ together are split-generators for $D^bCoh(Y)$. Any locally

2271: free coherent sheaf $E$ on $Y$ has a finite left resolution of the

2272: form

2273: \[

2274:  0 \rightarrow E' \rightarrow \O(m_l)^{\oplus r_l} \rightarrow

2275:  \cdots \rightarrow \O(m_1)^{\oplus r_1} \rightarrow E \rightarrow 0

2276: \]

2277: for some $E'$, and one can make $l$ arbitrarily large. For $l >

2278: dim\,Y$ we have $Ext^l(E,E') = H^l(E^\vee \otimes E') = 0$, hence the

2279: resulting exact triangle

2280: \[

2281:  \xymatrix{

2282:  \!\!\!\! {\big\{ \O(m_l)^{\oplus r_l} \rightarrow \dots \rightarrow

2283:  \O(m_1)^{\oplus r_1} \big\}} \ar[r] & {E} \ar[d] \\

2284:  & {E'[l]} \ar@/^1pc/[ul]^-{[1]}

2285: }

2286: \]

2287: splits in such a way that $E \oplus E'[l-1]$ is isomorphic to the top

2288: left complex. Once one has split-generated all locally free sheaves

2289: from the $\O(m)$, the rest follows as in Lemma

2290: \ref{th:derived-is-derived}.

2291: 

2292: 

2293: \section{Symplectic terminology\label{sec:geometry}}

2294: 

2295: The purpose of this section is to assemble the necessary nuts and

2296: bolts from elementary symplectic geometry. The origin of the main

2297: notions, to the author's best knowledge, is as follows. Rational

2298: Lagrangian submanifolds have a long history in quantization (going

2299: back to Einstein); the first place where they occur in connection

2300: with Floer homology is probably Fukaya's paper \cite{fukaya93}.

2301: Lagrangian submanifolds with zero Maslov class also originated in

2302: (Maslov's) quantization; the essentially equivalent notion of grading

2303: was introduced by Kontsevich \cite{kontsevich94}. The symplectic

2304: viewpoint on Lefschetz pencils and higher-dimensional linear systems

2305: emerged gradually from the work of Arnol'd \cite{arnold95}, Donaldson

2306: \cite{donaldson98,donaldson02}, Gompf \cite{gompf-stipsicz}, and

2307: Auroux \cite{auroux99,auroux00}. In particular, matching cycles were

2308: invented by Donaldson (unpublished).

2309: 

2310: \subsection{}

2311: By a {\em projective K{\"a}hler manifold} we mean a compact complex

2312: manifold $X$ carrying a unitary holomorphic line bundle $o_X$, which

2313: is positive in the sense that the curvature of the corresponding

2314: connection $A_X$ defines a K{\"a}hler form

2315: \[

2316: \o_X = \frac{i}{2\pi} F_{A_X}.

2317: \]

2318: From a symplectic viewpoint, this integrality condition on the

2319: symplectic class leads one to make corresponding restrictions on the

2320: other objects which live on $X$. A Lagrangian submanifold $L \subset

2321: X$ is called {\em rational} if the monodromy of the flat connection

2322: $A_X|L$ consists of roots of unity. This means that one can choose a

2323: covariantly constant multisection $\lambda_L$ of $o_X|L$ which is of

2324: unit length everywhere (as a multisection, this is the multivalued

2325: $d_L$-th root of a section $\lambda_L^{d_L}$ of $o_X^{\otimes

2326: d_L}|L$, for some $d_L \geq 1$). Rationality is invariant under exact

2327: Lagrangian isotopy. More precisely, take such an isotopy $L_t$, and

2328: embed it into a Hamiltonian isotopy $\phi^t$, with time-dependent

2329: Hamiltonian $H$ and vector field $Y$. We obtain covariantly constant

2330: multisections $\lambda_{L_t}$ by starting with $\lambda_{L_0}$ and

2331: solving the ODE

2332: \begin{equation} \label{eq:deform-lambda}

2333: \nabla_{Y_t} \lambda_{L_t} = - 2\pi i (H_t|L_t) \, \lambda_{L_t}.

2334: \end{equation}

2335: 

2336: Rationality takes its name from the effect on the periods of action

2337: functionals. Take two rational Lagrangian submanifolds $L_0,L_1$. For

2338: each of them, choose a covariantly constant unit length

2339: $d_{L_k}$-fold multisection $\lambda_{L_k}$. The {\em mod $\Q$

2340: action} $\bar{A}(x)$ of a point $x \in L_0 \cap L_1$ is defined by

2341: \begin{equation} \label{eq:modq-action}

2342: e^{2\pi i \bar{A}(x)} = \lambda_{L_1}(x)/\lambda_{L_0}(x).

2343: \end{equation}

2344: It is unique up to adding a number in $\lcm(d_{L_0},d_{L_1})^{-1} \Z

2345: \subset \Q$, whence the name. Now suppose that we have two

2346: intersection points $x_0,x_1$ and a connecting disc between them:

2347: this is a smooth map $u: \R \times [0;1] \rightarrow X$ with $u(\R

2348: \times \{k\}) \subset L_k$, such that $u(s,\cdot)$ converges to the

2349: constant paths at $x_0,x_1$ as $s \rightarrow -\infty,+\infty$. By

2350: definition of curvature,

2351: \begin{equation} \label{eq:modq-energy}

2352: \int u^*\o_X \in \bar{A}(x_0) - \bar{A}(x_1) +

2353: \lcm(d_{L_0},d_{L_1})^{-1}\Z.

2354: \end{equation}

2355: 

2356: Similarly, a symplectic automorphism $\phi$ of $X$ is rational if the

2357: monodromy of the induced flat connection on $Hom(o_X,\phi^*o_X)$

2358: consists of roots of unity. This is invariant under Hamiltonian

2359: isotopy. We write $Aut(X)$ for the group of such automorphisms, which

2360: acts in the obvious way on the set of rational Lagrangian

2361: submanifolds.

2362: 

2363: \subsection{}

2364: Equally important for us will be the relative situation, where in

2365: addition to the previous data we have a holomorphic section

2366: $\sigma_{X,\infty}$ of $o_X$ such that $X_\infty =

2367: \sigma_{X,\infty}^{-1}(0)$ is a divisor with normal crossings, along

2368: which $\sigma_{X,\infty}$ vanishes with multiplicity one, and such

2369: that each irreducible component $C \subset X_\infty$ is smooth (no

2370: self-intersections). We call $M = X \setminus X_\infty$ an {\em

2371: affine K{\"a}hler manifold}. One can use

2372: $\sigma_{X,\infty}/||\sigma_{X,\infty}||$ to trivialize $o_X|M$, and

2373: by writing the connection with respect to this trivialization as

2374: $A_X|M = d - 2\pi i \theta_M$, one gets a one-form $\theta_M$ which

2375: satifies $d\theta_M = \o_M = \o_X|M$, so that $M$ is an exact

2376: symplectic manifold in the usual sense of the word.

2377: 

2378: A compact Lagrangian submanifold $L \subset M$ is called {\em exact}

2379: if $\theta_M|L = dK_L$ for some function $K_L$. This condition is

2380: invariant under exact Lagrangian isotopy. Note that if $L$ is exact,

2381: \begin{equation} \label{eq:k-lambda}

2382: \lambda_L = \exp(2 \pi i K_L )

2383: \frac{\sigma_{X,\infty}}{||\sigma_{X,\infty}||}

2384: \end{equation}

2385: is a covariantly constant section of $o_X|L$, so that must be the

2386: trivial flat line bundle. Conversely, suppose that $L$ is some

2387: Lagrangian submanifold of $M$, such that $o_X|L$ is trivial and

2388: admits a covariantly constant section which can be written in the

2389: form \eqref{eq:k-lambda} for some $K_L$. Then $dK_L = \theta_M|L$,

2390: hence $L$ is exact. We have proved:

2391: 

2392: \begin{lemma} \label{th:check-exactness}

2393: $L$ is exact iff $o_X|L$ is trivial, and its nonzero covariantly

2394: constant sections lie in the same homotopy class of nowhere zero

2395: sections as $\sigma_{X,\infty}|L$. \qed

2396: \end{lemma}

2397: 

2398: Let $L_0$, $L_1$ be exact Lagrangian submanifolds of $M$, and choose

2399: functions $K_{L_0}$, $K_{L_1}$. One can then define the action of an

2400: intersection point $x \in L_0 \cap L_1$ to be

2401: \begin{equation} \label{eq:action}

2402: A(x) = K_{L_1}(x) - K_{L_0}(x) \in \R.

2403: \end{equation}

2404: For any two such points $x_0,x_1$ and any connecting disc $u$ in $X$,

2405: the intersection number $u \cdot X_\infty$ is well-defined, and

2406: \begin{equation} \label{eq:energy}

2407: \int u^*\o_X = A(x_0) - A(x_1) + (u \cdot X_\infty).

2408: \end{equation}

2409: In particular, the action functional is indeed exact if we allow only

2410: connecting discs which lie in $M$.

2411: 

2412: Let $\phi$ be a symplectic automorphism of $M$, which extends

2413: smoothly to $X$ in such a way that $X_\infty$ remains pointwise fixed

2414: (we denote the extension equally by $\phi$). We say that $\phi$ exact

2415: if $\phi^*\theta_M - \theta_M = dK_\phi$ for some function $K_\phi$

2416: on $M$. This condition is invariant under Hamiltonian isotopies

2417: keeping $X_\infty$ fixed. Exact symplectic automorphisms form a group

2418: $Aut(M)$, which acts naturally on the set of exact Lagrangian

2419: submanifolds. The analogue of Lemma \ref{th:check-exactness} is

2420: 

2421: \begin{lemma} \label{th:check-exactness-2}

2422: $\phi$ is exact iff $Hom(o_X,\phi^*o_X)$ is the trivial flat bundle,

2423: and every covariantly constant nonzero section has the property that

2424: its restriction to $M$ lies in the same homotopy class of nowhere

2425: zero sections as $\phi^*\sigma_{X,\infty}/\sigma_{X,\infty}$. \qed

2426: \end{lemma}

2427: 

2428: \begin{remark} \label{th:h1}

2429: The conditions for exactness which appear in the Lemma are

2430: automatically satisfied if $H_1(X) = 0$. This is obvious for the

2431: first one, and for the second one argues as follows: take the

2432: meridian $\gamma$ in $M$ around a component of $X_\infty$, and bound

2433: it by a small disc going through $X_\infty$ once. The homotopy class

2434: of covariantly constant sections of $Hom(o_X,\phi^*o_X)|\gamma$ is

2435: characterized by the fact that they extend to nowhere zero sections

2436: over the disc. On the other hand, $\sigma_{X,\infty}\gamma$ extends

2437: to a section with precisely one zero, and so does

2438: $\phi^*\sigma_{X,\infty}|\gamma$. Hence,

2439: $(\phi^*\sigma_{X,\infty}/\sigma_{X,\infty})|\gamma$ lies in the same

2440: homotopy class of nowhere zero sections as the covariantly constant

2441: ones.

2442: \end{remark}

2443: 

2444: %The way one goes from $K_\phi$ to $\gamma_\phi$ and back is by

2445: %writing

2446: %\[

2447: %\gamma_\phi|M = \exp(2 \pi i K_\phi) \cdot

2448: %\frac{\phi^*\sigma_{X,\infty}}{||\phi^*\sigma_{X,\infty}||} \left(

2449: %\frac{\sigma_{X,\infty}}{||\sigma_{X,\infty}||} \right)^{-1}.

2450: %\]

2451: %

2452: The relation between the affine and projective notions is clear:

2453: Lemma \ref{th:check-exactness} shows that an exact $L \subset M$ is a

2454: rational Lagrangian submanifold of $X$, and Lemma

2455: \ref{th:check-exactness-2} does the same for automorphisms. Given a

2456: pair of exact Lagrangian submanifolds $L_0,L_1$, the multivalued

2457: action \eqref{eq:modq-action} is a reduction of the single-valued one

2458: \eqref{eq:action}, and \eqref{eq:energy} is a refinement of

2459: \eqref{eq:modq-energy}.

2460: 

2461: \subsection{}

2462: Let $M$ be an affine K{\"a}hler manifold, coming from $X$ and

2463: $\sigma_{X,\infty}$ as always. Suppose that we have a second

2464: holomorphic section $\sigma_{X,0}$ of $o_X$, linearly independent of

2465: $\sigma_{X,\infty}$. Let $\{X_z\}_{z \in \CP{1}}$ be the pencil of

2466: hypersurfaces generated by these two sections. This means that we set

2467: \[

2468: \sigma_{X,z} = \sigma_{X,0} - z\sigma_{X,\infty} \quad \text{ and }

2469: \quad X_z = \sigma_{X,z}^{-1}(0)

2470: \]

2471: for $z \in \C$. Assume that $X_0 = \sigma_{X,0}^{-1}(0)$ is smooth in

2472: a neighbourhood of the base locus $X_{0,\infty} = X_0 \cap X_\infty$,

2473: and that it intersects each stratum of $X_\infty$ transversally; then

2474: the same will hold for each $X_z$, $z \in \C$. Every nonsingular

2475: $X_z$, $z \in \C$, is a projective K{\"a}hler manifold with ample

2476: line bundle $o_{X_z} = o_X|X_z$, and it carries a preferred

2477: holomorphic section $\sigma_{X_z} = \sigma_{X,\infty}|X_z$ whose zero

2478: set is $X_{0,\infty}$. By assumption, this is a divisor with normal

2479: crossings, so that $M_z = X_z \setminus X_{0,\infty}$ is an affine

2480: K{\"a}hler manifold. One can also see the $M_z$ as the regular fibres

2481: of the holomorphic function

2482: \[

2483: \pi_M = \sigma_{X,0}/\sigma_{X,\infty} :  M \longrightarrow \C.

2484: \]

2485: We denote by $Critv(\pi_M)$ its critical value set, which is finite

2486: by Bertini. Given $z,z' \in \C \setminus Critv(\pi_M)$, we denote by

2487: $Iso(M_z,M_{z'})$ the space of symplectic isomorphisms $\phi: M_z

2488: \rightarrow M_{z'}$ which extend to $X_z \rightarrow X_{z'}$ in such

2489: a way that $X_0 \cap X_\infty$ remains pointwise fixed, and such that

2490: $\phi^*\theta_{M_{z'}} - \theta_{M_z}$ is exact. These spaces form a

2491: topological groupoid under composition, and for $z = z'$ they reduce

2492: to the groups $Aut(M_z)$ defined before. Take a smooth path $c: [0;l]

2493: \rightarrow \C \setminus Critv(\pi_M)$. The $c'(t)$ can be lifted in

2494: a unique way to vector fields $Z_t \in \smooth(TM|M_{c(t)})$ which

2495: are horizontal (orthogonal to $ker(D\pi_M)$ with respect to the

2496: symplectic form). By integrating these one gets an exact {\em

2497: symplectic parallel transport} map

2498: \[

2499: h_c \in Iso(M_{c(0)},M_{c(l)}).

2500: \]

2501: To show that $h_c$ is well-defined, one embeds $M$ into the graph

2502: $\Gamma = \{(z,x) \in \C \times X_z \suchthat x \in X_z\}$. The

2503: projection $\pi_\Gamma: \Gamma \rightarrow \C$ is a properification

2504: of $\pi_M$. It is a standard fact that parallel transport with

2505: respect to the product symplectic form $\o_\Gamma = (\o_X +

2506: \o_{\C})|\Gamma$ gives a symplectic isomorphism $X_{c(0)} \rightarrow

2507: X_{c(l)}$. This extends the original $h_c$, because adding the

2508: pullback of $\o_{\C}$ to $\o_M$ does not change the vector fields

2509: $Z_t$. Moreover, at a point $(z,x) \in \C \times X_{0,\infty}$ we

2510: have

2511: \[

2512:  T\Gamma_{(z,x)} = \C \times \{ \xi \in TX_x \suchthat

2513:  D\sigma_0(x)\xi - z\, D\sigma_\infty(x)\xi = 0 \},

2514: \]

2515: and since this splitting is $\o_\Gamma$-orthogonal, the parallel

2516: transport vector field will lie in $\C \times \{0\}$. This implies

2517: that the extension of $h_c$ to $X_{c(0)}$ leaves $X_{0,\infty}$

2518: pointwise fixed. Finally, the exactness of $h_c^*\theta_{M_{c(l)}} -

2519: \theta_{M_{c(0)}}$ follows from the Cartan formula

2520: $

2521:  (L_{Z_t} \theta_M) \,|\, M_{c(t)} =

2522:  d(i_{Z_t}\theta_M) \,|\, M_{c(t)} +

2523:  i_{Z_t}\o_M \,|\, M_{c(t)}

2524: $, where the last term vanishes because of the horizontality of

2525: $Z_t$.

2526: 

2527: \subsection{}

2528: We say that $\{X_z\}$ is an {\em quasi-Lefschetz pencil} if $\pi_M$

2529: is nondegenerate in the sense of Picard-Lefschetz theory, meaning

2530: that it has only nondegenerate critical points, no two of which lie

2531: in the same fibre. This is the same as saying that the set of

2532: critical points $Crit(\pi_M)$ is regular ($D\pi_M$ is transverse to

2533: the zero-section) and $\pi_M: Crit(\pi_M) \rightarrow Critv(\pi_M)$

2534: is bijective. Lefschetz pencils are the special case where $X_\infty$

2535: is smooth.

2536: 

2537: An {\em embedded vanishing path} for a quasi-Lefschetz pencil is an

2538: embedded path $c: [0;1] \rightarrow \C$ with $c^{-1}(Critv(\pi_M)) =

2539: \{1\}$. By considering the limiting behaviour of parallel transport

2540: along $c|[0;t]$ as $t \rightarrow 1$, one defines an embedded

2541: Lagrangian disc $\Delta_c \subset M$, the {\em Lefschetz thimble},

2542: whose boundary is a Lagrangian sphere $V_c \subset M_{c(0)}$, the

2543: associated {\em vanishing cycle}. The Picard-Lefschetz formula

2544: \cite[Proposition 1.15]{seidel01} says that the monodromy around a

2545: loop $\gamma$ which doubles around $c$ in positive sense is the Dehn

2546: twist associated to the vanishing cycle, up to isotopy inside the

2547: group of exact symplectic automorphisms:

2548: \[

2549: h_\gamma \htp \tau_{V_c} \in Aut(M_{c(0)}).

2550: \]

2551: To be precise, $V_c$ carries a small additional piece of structure,

2552: called a framing in \cite{seidel00,seidel01}, and the definition of

2553: Dehn twist uses that too. However, framings contain no information if

2554: the vanishing cycles are of dimension $\leq 3$, so omitting them will

2555: not matter for our applications.

2556: 

2557: Now choose a base point $z_* \in \C \setminus Critv(\pi_M)$ and base

2558: path $c_*: [0;\infty) \rightarrow \C \setminus Critv(\pi_M)$, meaning

2559: an embedded path with $c_*(0) = z_*$ which goes off to infinity, in

2560: the sense that there is a $\zeta \in \C^*$ such that $c_*(t) = \zeta

2561: t$ for all $t \gg 0$. By a {\em distinguished basis} of embedded

2562: vanishing paths, we mean an ordered collection of such paths

2563: $(c_1,\dots,c_r)$, $r = |Critv(\pi_M)|$, starting at $z_*$ and with

2564: the following properties: the tangent directions $\Rgeq c_k'(0)$ are

2565: pairwise distinct and clockwise ordered, and different $c_k$ do not

2566: intersect except at $z_*$. Moreover, $\Rgeq c_*'(0)$ should lie

2567: strictly between $\Rgeq c_r'(0)$ and $\Rgeq c_1'(0)$, and no $c_k$

2568: may intersect $c_*$ except at $z_*$. The composition of the

2569: corresponding loops $\gamma_k$ is freely homotopic in $\C \setminus

2570: Critv(\pi_M)$ to a circle $\gamma_\infty(t) = R\zeta e^{it}$ of some

2571: large radius $R \gg 0$, and hence the product of all the associated

2572: Dehn twists is the monodromy around that circle. More precisely,

2573: taking into account the change of base point, one has

2574: \begin{equation} \label{eq:large-small-loop}

2575:  \tau_{V_{c_1}} \dots \tau_{V_{c_r}} \htp

2576:  (h_{c_*|[0;R]})^{-1} h_{\gamma_\infty} h_{c_*|[0;R]} \in Aut(M_{z_*}).

2577: \end{equation}

2578: This becomes more familiar for actual Lefschetz pencils, where one

2579: can extend the graph to $\CP{1}$,

2580: \begin{equation} \label{eq:bar-gamma}

2581: \bar{\Gamma} = \{(z,x) \in \CP{1} \times X \suchthat x \in X_z\}

2582: \xrightarrow{\pi_{\bar{\Gamma}}} \CP{1},

2583: \end{equation}

2584: so that $\infty$ is a regular value of $\pi_{\bar{\Gamma}}$.

2585: Proceeding as before, one can define symplectic parallel transport

2586: for paths in $\CP{1} \setminus Critv(\pi_M)$. In particular, since

2587: $\gamma_\infty$ can be contracted by passing over $\infty$, its

2588: monodromy is isotopic to the identity, so that

2589: \eqref{eq:large-small-loop} simplifies to the following relation

2590: between Dehn twists in the symplectic mapping class group

2591: $\pi_0(Aut(M_{z_*}))$:

2592: \begin{equation} \label{eq:relation-in-mcg}

2593: \tau_{V_{c_1}} \circ \dots \circ \tau_{V_{c_r}} \htp id.

2594: \end{equation}

2595: 

2596: \begin{remark} \label{th:multiple-fibres}

2597: It is no problem (except terminologically) to allow several

2598: nondegenerate critical points to lie in the same fibre. One should

2599: then choose a vanishing path for every critical point. Two paths

2600: $c_k,c_l$ going to the same critical value should either coincide or

2601: else satisfy $c_k(t) \neq c_l(t)$ for $t \in (0;1)$, and $\Rgeq

2602: c_k'(0) \neq \Rgeq c_l'(0)$. In the first case, the associated

2603: vanishing cycles in $M_{z_*}$ will be disjoint.

2604: \end{remark}

2605: 

2606: \subsection{}

2607: Let $\{X_z\}$ be a quasi-Lefschetz pencil. Consider a smooth embedded

2608: path $d: [-1;1] \rightarrow \C$ such that $d^{-1}(Critv(\pi_M)) =

2609: \{-1;1\}$. Split it into a pair of vanishing paths $c_-(t) = d(-t)$,

2610: $c_+(t) = d(t)$, $t \in [0;1]$, and consider the associated vanishing

2611: cycles

2612: \[

2613: V_{c_-},V_{c_+} \subset M_{d(0)}.

2614: \]

2615: We say that $d$ is a {\em matching path} if $V_{c_-}$, $V_{c_+}$ are

2616: isotopic as exact Lagrangian spheres. Actually, there is an

2617: additional condition concerning the framings, which we do not want to

2618: spell out here, since it is vacuous in the dimensions that we will be

2619: interested in. If $d$ is a matching path, one can glue together the

2620: Lefschetz thimbles $\Delta_{c_-}$, $\Delta_{c_+}$ to obtain a

2621: Lagrangian sphere in the total space, the {\em matching cycle}

2622: $\Sigma_d \subset M$. In the naive situation where the two vanishing

2623: cycles coincide, this would simply be the set-theoretic union of the

2624: Lefschetz thimbles. In general, the construction involves a choice of

2625: isotopy between the cycles, and $\Sigma_d$ might theoretically depend

2626: on that choice. Note however that this can never happen if the fibres

2627: are Riemann surfaces with boundary, since there, the space of exact

2628: Lagrangian submanifolds (circles) having a fixed isotopy class is

2629: contractible.

2630: 

2631: \subsection{\label{subsec:braid-monodromy}}

2632: Let $\sigma_{X,0},\sigma_{X,\infty}$ be two sections of $o_X$ which

2633: generate a quasi-Lefschetz pencil $\{X_z\}$. Suppose that we have yet

2634: another section $\sigma_{X,0}'$, linearly independent from the

2635: previous ones. We require that its zero set should be smooth in a

2636: neighbourhood of $X_0 \cap X_\infty$, and should intersect each

2637: stratum of that base locus transversally. Consider the associated map

2638: \begin{equation} \label{eq:bm-map}

2639: M \xrightarrow{b_M = \left(\frac{\sigma_{X,0}}{\sigma_{X,\infty}},

2640: \frac{\sigma_{X,0}'}{\sigma_{X,\infty}}\right)} \C^2,

2641: \end{equation}

2642: its critical point set $Crit(b_M)$ and critical value set $C =

2643: Critv(b_M)$. Concerning this, we make the following additional

2644: assumptions:

2645: \begin{itemize}

2646: \item

2647: $Crit(b_M)$ is regular, and $b_M: Crit(b_M) \rightarrow C$ is an

2648: embedding away from finitely many points.

2649: \end{itemize}

2650: Regularity means that $Db_M \in \Gamma(Hom(TM,\C^2))$ is never zero,

2651: and is transverse to the subset of rank one linear maps. The second

2652: statement implies that for generic $z \in \C \setminus Critv(\pi_M)$,

2653: the curve $Crit(b_M)$ is transverse to $b_M^{-1}(\{z\} \times \C)$,

2654: and the projection $b_M|Crit(b_M) \times b_M^{-1}(\{z\} \times \C)

2655: \rightarrow C \times (\{z\} \times \C)$ is a bijection. In this case,

2656: the sections $(\sigma_{X,0}'|X_z,\sigma_{X,\infty}|X_z)$ generate a

2657: quasi-Lefschetz pencil on $X_z$, and the critical value set of the

2658: associated holomorphic function $q_{M_z}: M_z \rightarrow \C$ is

2659: precisely

2660: \begin{equation} \label{eq:z-critical}

2661: Critv(q_{M_z}) = C \cap (\{z\} \times \C) \subset \C.

2662: \end{equation}

2663: There is a finite subset of $\C \setminus Critv(\pi_M)$ where this

2664: fails, and we denote it by $Fakev(\pi_M)$. These ``fake critical

2665: values'' will play a role similar to the real ones, see

2666: \eqref{eq:braid-monodromy} below.

2667: \begin{itemize}

2668: \item

2669: Let $x$ be a critical point of $\pi_M$, $\pi_M(x) = z$. Then the

2670: restriction of $D^2\pi_M$ to the complex codimension one subspace

2671: $ker(D_xb_M)$ is nondegenerate. Moreover, all critical points of

2672: $q_{M_z}|(M_z \setminus \{x\})$ are nondegenerate, and their

2673: $q_{M_z}$-values are pairwise distinct and different from

2674: $q_{M_z}(x)$.

2675: \end{itemize}

2676: As a consequence, $b_M: Crit(b_M) \rightarrow C$ is an embedding on

2677: $Crit(b_M) \cap M_z$, and moreover the projection $C \subset \C^2

2678: \rightarrow \C$ to the first variable has an ordinary (double) branch

2679: point at $b_M(x)$, and is a local isomorphism at all other points of

2680: $C \cap (\{z\} \times \C)$.

2681: 

2682: If the two conditions stated above are satisfied, we say that

2683: $\sigma_{X,0}'$ is a {\em generic auxiliary section} of $o_X$.

2684: Practically, the situation is that we have a quasi-Lefschetz pencil

2685: on $X$, almost every regular fibre $X_z$ of which again admits a

2686: quasi-Lefschetz pencil. Moreover, we have a good understanding of how

2687: these pencils degenerate as $X_z$ becomes singular. We will now

2688: explain the consequences of this for the vanishing cycles of $\pi_M$.

2689: For that, suppose that our base point $z_* \notin Fakev(\pi_M)$.

2690: Alongside the ordinary monodromy of the pencil,

2691: \begin{align}

2692:  \notag \pi_1(\C \setminus Critv(\pi_M),z_*) & \longrightarrow

2693:  \pi_0(Aut(M_{z_*})), \\

2694: \intertext{there is a ``relative'' or ``braid'' monodromy

2695: homomorphism}

2696:  \label{eq:braid-monodromy}

2697:  \pi_1(\C \setminus (Critv(\pi_M) \cup Fakev(\pi_M)),z_*)

2698:  & \longrightarrow

2699:  \pi_0(\Diff^c(\C,(C \cap \{z_*\} \times \C))),

2700: \end{align}

2701: which describes how the finite set \eqref{eq:z-critical} moves in

2702: $\C$ as $z$ changes. Take an embedded vanishing path for $\pi_M$, $c:

2703: [0;1] \rightarrow \C$, with the additional property that $c([0;1])

2704: \cap Fakev(\pi_M) = \emptyset$. Let $\gamma$ be a loop doubling

2705: around $c$ in positive sense. As we know, the monodromy around

2706: $\gamma$ is the Dehn twist along $V_c$. The corresponding statement

2707: for \eqref{eq:braid-monodromy} is that there is an embedded path $d:

2708: [-1;1] \rightarrow \C$, with $d^{-1}(Critv(q_{M_{z_*}})) = \{\pm

2709: 1\}$, such that the image of $\gamma$ under the relative monodromy is

2710: the half-twist along $d$. The relation between the two is as follows,

2711: see \cite{seidel04} for the proof:

2712: 

2713: \begin{prop} \label{th:vanishing-matching}

2714: $d$ is a matching path for $q_{M_{z_*}}$, and the vanishing cycle

2715: $V_c \subset M_{z_*}$ is, up to Lagrangian isotopy, a matching cycle

2716: $\Sigma_d$ for that path. \qed

2717: \end{prop}

2718: 

2719: \subsection{}

2720: We temporarily digress from K{\"a}hler to general symplectic

2721: geometry. Let $(M^{2n},\o_M)$ be any connected symplectic manifold

2722: with $c_1(M) = 0$. Take an almost complex structure $J_M$ which tames

2723: $\o_M$, and a complex volume form (a nowhere vanishing $(n,0)$-form

2724: with respect to $J_M$) $\eta_M$. Let $\LL_M \rightarrow M$ be the

2725: bundle of (unoriented) Lagrangian Grassmannians. The phase function

2726: associated to $\eta_M$,

2727: \[

2728: \alpha_M: \LL_M \longrightarrow S^1,

2729: \]

2730: is defined as follows. Given $x \in M$ and $\Lambda \in \LL_{M,x}$,

2731: take any (not necessarily orthonormal) basis $e_1,\dots,e_n$ of

2732: $\Lambda$, and set

2733: \begin{equation} \label{eq:squared-phase}

2734: \alpha_M(\Lambda) = \frac{\eta_M(e_1 \wedge \dots \wedge

2735: e_n)^2}{|\eta_M(e_1 \wedge \dots \wedge e_n)|^2}.

2736: \end{equation}

2737: We distinguish the special case where $J_M$ is $\o_M$-compatible by

2738: calling $\alpha_M$ a {\em classical phase function}. These are the

2739: most commonly used ones, and they are much better behaved. General

2740: phase functions will be needed later for technical reasons.

2741: 

2742: Given $\alpha_M$, one can associate to a Lagrangian submanifold $L

2743: \subset M$ or a symplectic automorphism $\phi: M \rightarrow M$ a

2744: phase function

2745: \[

2746: \begin{cases}

2747:  & \!\!\! \alpha_L: L \longrightarrow S^1, \quad

2748:  \alpha_L(x) = \alpha_M(TL_x), \\

2749:  & \!\!\! \alpha_\phi: \LL_M \longrightarrow S^1, \quad

2750:  \alpha_\phi(\Lambda) = \alpha_M(D\phi(\Lambda))/\alpha_M(\Lambda).

2751: \end{cases}

2752: \]

2753: A grading of $L$ or $\phi$ is a real-valued lift $\alphagr_L$ resp.\

2754: $\alphagr_\phi$, where the convention is that $\R$ covers $S^1$ by $a

2755: \mapsto \exp(2\pi i a)$. Pairs $\Lgr = (L,\alphagr_L)$, $\phigr =

2756: (\phi,\alphagr_\phi)$ are called graded Lagrangian submanifolds

2757: resp.\ graded symplectic automorphisms. The latter form a group, with

2758: a natural action on the set of graded Lagrangian submanifolds. The

2759: composition law and action are defined by

2760: \begin{equation}

2761: \label{eq:compose-gradings}

2762: \begin{cases}

2763:  & \!\!\! \alphagr_{\psi \circ \phi} = \alphagr_\psi \circ D\phi +

2764:  \alphagr_\phi, \\

2765:  & \!\!\! \alphagr_{\phi(L)} = \alphagr_L \circ \phi^{-1} + \alphagr_\phi

2766:  \circ D\phi^{-1}.

2767: \end{cases}

2768: \end{equation}

2769: In particular, for $k \in \Z$ we have the $k$-fold shift $[k] = (\phi

2770: = \id_M,\alphagr_{\phi} = -k)$, which reduces the grading of each

2771: Lagrangian submanifold by $k$. All the ``graded'' notions actually

2772: depend only on the homotopy class of $\eta_M^2$ as a trivialization

2773: of $K_M^2$, or equivalently on the cohomology class $\mu_M =

2774: [\alpha_M] \in H^1(\LL_M)$, called a global Maslov class in

2775: \cite{seidel99}. Namely, suppose that we have two cohomologous phase

2776: functions, which can be written as $\alpha_M' = \alpha_M e^{2\pi i

2777: \chi}$ for some $\chi: \LL_M \rightarrow \R$. Gradings with respect

2778: to the two are related in an obvious way,

2779: \begin{equation} \label{eq:new-phase}

2780: \begin{cases}

2781:  \!\!\! & \alphagr_L'(x) = \alphagr_L(x) + \chi(TL_x), \\

2782:  \!\!\! & \alphagr_\phi'(\Lambda) = \alphagr_\phi(\Lambda) +

2783:  \chi(D\phi(\Lambda)) - \chi(\Lambda).

2784: \end{cases}

2785: \end{equation}

2786: Note that $\chi$ is unique only up to an integer constant. This is

2787: irrelevant for $\alphagr_\phi'$; it affects $\alphagr_L'$ but not in

2788: a really important way, since changing $\chi$ shifts the grading of

2789: all Lagrangians simultaneously.

2790: 

2791: \begin{remark} \label{th:or}

2792: There is a canonical square root $\alpha_M^{1/2}: \LL^{\orient}

2793: \rightarrow \R$ of $\alpha_M$ on the bundle $\LL^{\orient}$ of

2794: oriented Lagrangian Grassmannians, defined by removing the squares

2795: from \eqref{eq:squared-phase} and using positively oriented bases

2796: $e_1,\dots,e_n$. Any graded Lagrangian submanifold carries a

2797: preferred orientation, characterized by

2798: \[

2799: \exp(\pi i \alphagr_L(x)) = \alpha_M^{1/2}(TL_x).

2800: \]

2801: \end{remark}

2802: 

2803: \subsection{}

2804: There is an obvious notion of graded symplectic fibre bundle, which

2805: is a locally trivial symplectic fibration $p: E \rightarrow B$ with

2806: an almost complex structure $J_{E_b}$ and complex volume form

2807: $\eta_{E_b}$ on each fibre $E_b$. The monodromy maps of such a

2808: fibration are naturally graded: given a loop $\gamma: [0;l]

2809: \rightarrow B$ and the corresponding parallel transport maps $\phi_t

2810: = h_{\gamma|[0;t]}: E_{\gamma(0)} \rightarrow E_{\gamma(t)}$, one

2811: takes the unique smooth family of functions $\tilde{a}_t$ on the

2812: Lagrangian Grassmannian bundle of $E_{\gamma(0)}$ with $\tilde{a}_0 =

2813: 0$ and

2814: \[

2815: \exp(2\pi i \tilde{a}_t) = (\alpha_{E_{\gamma(t)}} \circ

2816: D\phi_t)/\alpha_{E_{\gamma(0)}}.

2817: \]

2818: The graded symplectic monodromy is $\hgr_{\gamma} =

2819: (\phi_l,\alphagr_{\phi_l} = \tilde{a}_l)$. An example which is

2820: relevant for our purpose is where we have a trivial fibre bundle $E =

2821: \C^* \times M$ over $B = \C^*$ with a varying complex volume form

2822: $\eta_{E_z} = z^\mu \eta_M$. The ``invisible singularity'' of this

2823: family at $z = 0$ leads to a monodromy which, for a circle $\gamma$

2824: turning around the origin in clockwise direction, is the shift

2825: \begin{equation} \label{eq:shift-monodromy}

2826: \hgr_\gamma = [2\mu].

2827: \end{equation}

2828: 

2829: \subsection{}

2830: Gradings make it possible to assign absolute Maslov indices to

2831: Lagrangian intersection points. This depends on the following notion

2832: from symplectic linear algebra: a smooth path of Lagrangian subspaces

2833: $\Lambda_t$ ($0 \leq t \leq 1$) in a fixed symplectic vector space is

2834: {\em crossingless} if it satisfies $\Lambda_0 \cap \Lambda_t =

2835: \Lambda_0 \cap \Lambda_1$ for all $t>0$, and the crossing form

2836: \cite{robbin-salamon93} on $\Lambda_0/\Lambda_0 \cap \Lambda_1$

2837: associated to $(d\Lambda_t/dt)_{t = 0}$ is negative definite.

2838: Nonintersecting paths with given endpoints exist and are unique up to

2839: homotopy, see again \cite{robbin-salamon93}.

2840: 

2841: \begin{lemma} \label{th:angle-bound}

2842: Suppose that $\alpha_M$ is classical. Let $\Lambda_t$ be a

2843: crossingless path in $TM_x$ for some $x$. Choose $\tilde{a}: [0;1]

2844: \rightarrow \R$ such that $\exp(2\pi i \tilde{a}_t) =

2845: \alpha_M(\Lambda_t)$. Then

2846: \[

2847: \tilde{a}_1 - \tilde{a}_0 \in (-n;0].

2848: \]

2849: \end{lemma}

2850: 

2851: \proof We can identify $\Phi: TM_x \rightarrow \C^n$ in such a way

2852: that the complex structure and symplectic form become standard,

2853: $\Phi(\Lambda_0) = \R^n$ and $\Phi(\Lambda_1) = e^{i\pi c_1} \R

2854: \times \dots \times e^{i\pi c_n}\R$ with $c_k \in (-1;0]$. For the

2855: obvious crossingless path $\Lambda_t = e^{i\pi t c_1}\R \times \dots

2856: \times e^{i\pi t c_n}\R$ we have $\tilde{a}_1 - \tilde{a}_0 =

2857: c_1+\dots+c_n \in (-n;0]$. Because of homotopy uniqueness, the same

2858: holds for any other crossingless path. \qed

2859: 

2860: Take two graded Lagrangian submanifolds $\Lgr_0,\Lgr_1$, and a point

2861: $x \in L_0 \cap L_1$. Choose a crossingless path from $\Lambda_0 =

2862: TL_0$ to $\Lambda_1 = TL_1$, and take $\tilde{a}_t \in \R$ such that

2863: $\exp(2\pi i \tilde{a}_t) = \alpha_M(\Lambda_t)$. The absolute Maslov

2864: index is defined as

2865: \begin{equation} \label{eq:absolute-index}

2866: I(x) = (\alphagr_{L_1}(x) - \tilde{a}_1) - (\alphagr_{L_0}(x) -

2867: \tilde{a}_0) \in \Z.

2868: \end{equation}

2869: Lemma \ref{th:angle-bound} implies that this is approximately the

2870: difference of the phases:

2871: 

2872: \begin{lemma} \label{th:approximate-index}

2873: If $\alpha_M$ is classical, $I(x) - \alphagr_{L_1}(x) +

2874: \alphagr_{L_0}(x) \in [0;n)$. \qed

2875: \end{lemma}

2876: 

2877: This was pointed out to me by Joyce, but it is known to many other

2878: people in mathematics and physics, see for instance the proof of

2879: \cite[Theorem 4.3]{thomas-yau01}. In fact, ideas of this kind can be

2880: traced back at least to \cite{salamon-zehnder92}. Another easy

2881: consequence of Lemma \ref{th:angle-bound}, this time for symplectic

2882: automorphisms, will be useful later:

2883: 

2884: \begin{lemma} \label{th:graded-variation}

2885: Suppose that $\alpha_M$ is classical. Let $\phigr =

2886: (\phi,\alphagr_\phi)$ a graded symplectic automorphism of $M$. For

2887: any two Lagrangian subspaces $\Lambda_0,\Lambda_1 \in \LL_{M,x}$ at

2888: the same point,

2889: \[

2890: |\alphagr_\phi(\Lambda_1)-\alphagr_\phi(\Lambda_0)| < n.

2891: \]

2892: \end{lemma}

2893: 

2894: \proof Choose a crossingless path $\Lambda_t$. Lemma

2895: \ref{th:angle-bound} shows that as we go along this path,

2896: $\alpha_M(\Lambda_t)$ changes by an angle in $(-2\pi n;0]$, and so

2897: does $\alpha_M(D\phi(\Lambda_t))$. The angular change in

2898: $\alpha_M(D\phi(\Lambda_t))/\alpha_M(\Lambda_t)$ is the difference

2899: between the two, hence lies in $(-2\pi n;2\pi n)$. \qed

2900: 

2901: \subsection{}

2902: We now return to the case of an affine K{\"a}hler manifold $M = X

2903: \setminus X_\infty$. We say that $M$ is {\em affine Calabi-Yau} if it

2904: comes with a preferred isomorphism of holomorphic line bundles,

2905: \begin{equation} \label{eq:beta}

2906: \beta_X : \K_X \stackrel{\iso}{\longrightarrow} o_X^{\otimes -m_X}

2907: \end{equation}

2908: for some $m_X \in \Z$ (called the monotonicity index). On $M$ we have

2909: a preferred trivialization of $o_X$ given by $\sigma_\infty$, hence

2910: through $\beta_X$ a holomorphic volume form $\eta_M$. In the special

2911: case where $m_X = 0$, $X$ itself is Calabi-Yau, and $\eta_M$ extends

2912: to a holomorphic volume form $\eta_X$ on it. We denote by $\Autgr(M)$

2913: the group of exact symplectic automorphisms of $M$ equipped with a

2914: grading. Similarly, if $m_X = 0$, we write $\Autgr(X)$ for the graded

2915: rational automorphisms of $X$.

2916: 

2917: \begin{remark} \label{th:extend-grading}

2918: In the $m_X = 0$ case, there is a canonical embedding

2919: \[

2920: \Autgr(M) \longrightarrow \Autgr(X).

2921: \]

2922: We already know that every $\phi \in Aut(M)$ extends to a rational

2923: symplectic automorphism of $X$. There are no obstructions to

2924: extending gradings over subsets of codimension $\geq 2$, so the

2925: grading of $\phi$ extends uniquely from $M$ to $X$.

2926: \end{remark}

2927: 

2928: If $\{X_z \subset X\}$ is a quasi-Lefschetz pencil, we have for $z

2929: \in \C \setminus Critv(\pi_M)$ a canonical isomorphism $\beta_{X_z}:

2930: \K_{X_z} \iso (\K_X \otimes o_X)|X_z \iso o_X^{\otimes -m_X+1}$, so

2931: the $M_z$ are again affine Calabi-Yaus, with monotonicity index

2932: $m_{X_z} = m_X-1$. More explicitly, the complex volume form is

2933: obtained as a quotient $\eta_M/dz$, which means that it is the unique

2934: solution of

2935: \begin{equation} \label{eq:division}

2936: dz \wedge \eta_{M_z} = \eta_M.

2937: \end{equation}

2938: The first consequence of this is that the monodromy maps along loops

2939: $\gamma$ in $\C \setminus Critv(\pi_M)$ have canonical gradings,

2940: $\hgr_\gamma \in \Autgr(M_{\gamma(0)})$. If $c$ is a vanishing path,

2941: the associated vanishing cycle $V_c$ admits a grading; this is

2942: trivial if its dimension is $\geq 2$, and otherwise follows from the

2943: fact that it is bounded by a Lagrangian disc in $M$. Hence, the

2944: associated Dehn twist $\tau_{V_c}$ then has a canonical grading

2945: $\taugr_{V_c}$, which is zero outside a neighbourhood of $V_c$

2946: itself. Taking the loop $\gamma$ which doubles $c$, we have the

2947: following graded version of the Picard-Lefschetz theorem:

2948: \[

2949: \hgr_\gamma \htp \taugr_{V_c} \in \Autgr(M_{c(0)}).

2950: \]

2951: The analogue of \eqref{eq:relation-in-mcg} is

2952: 

2953: \begin{lemma} \label{th:shift-factor}

2954: Suppose that $\{X_z\}$ is a Lefschetz pencil, and $c_1,\dots,c_m$ a

2955: distinguished basis of vanishing cycles. Then

2956: \[

2957: \taugr_{V_{c_1}} \circ \dots \circ \taugr_{V_{c_r}} \htp [4-2 m_X]

2958: \in \Autgr(M_{z_*}).

2959: \]

2960: \end{lemma}

2961: 

2962: \proof What we need to show is that the graded monodromy around a

2963: large circle $\gamma_\infty$ is $[4-2 m_X]$. For simplicity, change

2964: variables to $\zeta = z^{-1}$, so that $\gamma_\infty$ becomes a

2965: small circle going clockwise around $\zeta = 0$. Recall that

2966: $\eta_{M_z}$ is obtained by dividing $\eta_M$, which has a zero or

2967: pole of order $-m_X$ along $X_\infty$, by $dz = d\zeta/\zeta^2$,

2968: which has order $-2$. Hence, after introducing the compactified graph

2969: $\bar{\Gamma}$ as in \eqref{eq:bar-gamma}, one has a locally trivial

2970: symplectic fibration near $\zeta = 0$, with a family of complex

2971: volume forms on the fibres that grows like $\zeta^{2-m_X}$. Now we

2972: are in the same situation as in \eqref{eq:shift-monodromy}. \qed

2973: 

2974: \section{Monodromy and negativity\label{sec:negativity}}

2975: 

2976: Continuing the discussion above, suppose that $m_X = 1$, so that the

2977: fibres $X_z$ are Calabi-Yau manifolds. In the Lefschetz pencil case,

2978: Lemma \ref{th:shift-factor} tells us that the ``large complex

2979: structure limit monodromy'' is a downward shift. We will need a

2980: generalization of this statement to the case when $X_\infty$ has

2981: normal crossings, where the geometry of the monodromy map becomes

2982: more complicated. Intuition is provided by two aspects of mirror

2983: symmetry: ideas related to the SYZ conjecture suggest that large

2984: complex structure limit monodromy maps should be fibrewise

2985: translations in a Lagrangian torus fibration; and in the context of

2986: homological mirror symmetry we have a formula for the mirror

2987: automorphism of the derived category of sheaves which goes back to

2988: Kontsevich, see \cite{aspinwall01}. Because of technical

2989: difficulties, our result will finally be limited to $K3$ surfaces;

2990: but large parts of the argument are more general, and it seems likely

2991: that with some more effort, the dimensional restriction could be

2992: lifted altogether.

2993: 

2994: \subsection{}

2995: We start with some generalities. Let $M^{2n}$ be a compact connected

2996: symplectic manifold (possibly with boundary), with a tame almost

2997: complex structure $J_M$, complex volume form $\eta_M$, and associated

2998: phase function $\alpha_M$.

2999: 

3000: \begin{defn} \label{th:negativity-definition}

3001: A graded symplectic automorphism $\phigr$ of $M$ is {\em negative} if

3002: there is a $d_0>0$ such that

3003: \[

3004: \alphagr_{\phi^{d_0}}(\Lambda) < 0 \quad \text{ for all $\Lambda \in

3005: \LL_M$.}

3006: \]

3007: \end{defn}

3008: 

3009: Here, the grading of $\phi^{d_0}$ is defined by the composition rule

3010: \eqref{eq:compose-gradings}. The same rule shows that if negativity

3011: holds, the grading of the iterates goes to $-\infty$. Namely, the

3012: grading of $\phi^{d_0}$ is $\leq -\epsilon$ for some $\epsilon>0$,

3013: and the grading of $\phi$ is $\leq C$ for some $C>0$. By writing $d =

3014: k d_0 + l$ with $0 \leq l \leq d_0-1$ one sees that

3015: \begin{equation} \label{eq:growth}

3016:  \alphagr_{\phi^d}(\Lambda) \leq C l - k\epsilon <

3017:  Cd_0 - \epsilon [d/d_0],

3018: \end{equation}

3019: as claimed. A closely related property is that the negativity of a

3020: given graded symplectic automorphism depends only on the global

3021: Maslov class $\mu_M$, and not on the choice of $\alpha_M$ within that

3022: class. For this take two cohomologous phase functions $\alpha_M' =

3023: e^{2 \pi i \chi} \alpha_M$. Since $\chi$ is bounded, it follows from

3024: \eqref{eq:new-phase} that the gradings of $\phi^d$ with respect to

3025: $\alpha_M,\alpha_M'$ differ by a uniformly bounded amount, which in

3026: view of \eqref{eq:growth} gives the desired property.

3027: 

3028: \begin{lemma} \label{th:maslov-growth}

3029: Let $\phigr$ be a negative graded symplectic automorphism, and

3030: $\Lgr_0,\Lgr_1$ two graded Lagrangian submanifolds. For any $k \in

3031: \Z$ one can find a $d>0$ and a small exact perturbation $L_1'$ of

3032: $L_1$ such that for each $x \in L_0 \cap \phi^d(L_1')$, the absolute

3033: Maslov index is

3034: \[

3035: I(x) < k.

3036: \]

3037: \end{lemma}

3038: 

3039: \proof Since negativity is independent of the choice of phase

3040: function, we may assume that that function is classical. Suppose that

3041: the grading of $L_1$ is bounded above by $C_1$, and that of $L_0$

3042: bounded below by $C_0$. Choose some $d$ such that $C_1 - C_0 + C d_0

3043: - \epsilon[d/d_0] \leq k-n$. From \eqref{eq:growth} and

3044: \eqref{eq:compose-gradings} it follows that

3045: \begin{equation} \label{eq:image-d}

3046:  \alphagr_{\phi^d(L_1)}(x) - \alphagr_{L_0}(x)

3047:  < k-n

3048: \end{equation}

3049: at every intersection point $x$. This estimate continues to hold if

3050: one perturbs $L_1$ slightly, to make the intersections transverse.

3051: Now apply Lemma \ref{th:approximate-index}. \qed

3052: 

3053: Negativity is well-behaved with respect to abelian symplectic

3054: reduction (this fact is related to the quotient construction of

3055: special Lagrangian submanifolds, see for instance \cite{gross01b}).

3056: The situation is as follows: $M = M^{2n}$ carries a free Hamiltonian

3057: $T^k$-action, with moment map $\mu: M \rightarrow \R^k$. We have an

3058: almost complex structure $J_M$ which is $\o_M$-compatible and

3059: $T^k$-invariant, and a $T^k$-invariant complex volume form $\eta_M$,

3060: with associated (classical and $T^k$-invariant) phase function

3061: $\alpha_M$. Denote by $K_1,\dots,K_k$ the Killing fields of the

3062: action. Let $\phi$ be a symplectic automorphism which preserves

3063: $\mu$, hence is $T^k$-equivariant; and $\alphagr_\phi$ a grading of

3064: it (necessarily $T^k$-invariant). For every moment value $r$ we have

3065: the symplectic quotient $M^{\red,r} = \mu^{-1}(r)/T^k$. This carries

3066: the reduced symplectic form $\o_{M^{\red,r}}$ and a compatible almost

3067: complex structure $J_{M^{\red,r}}$, the latter of which is obtained

3068: by taking the complexified tangent spaces along the orbits, and

3069: identifying their orthogonal complement with the tangent spaces of

3070: $M^{\red,r}$. Moreover, $\phi$ induces symplectic automorphisms

3071: $\phi^{\red,r}$. In our situation, we also have a reduced complex

3072: volume form $\eta_{M^{\red,r}}$, defined by restricting

3073: $

3074: i_{K_1} \dots i_{K_r} \eta_M

3075: $

3076: to the same orthogonal complement as before. Projection identifies

3077: Lagrangian subspaces $\Lambda \subset TM$ containing the tangent

3078: space to the orbit with Lagrangian subspaces $\Lambda^{\red} \subset

3079: TM^{\red,r}$. The phase functions defined by $\eta_M$,

3080: $\eta_{M^{\red,r}}$ are related by

3081: $\alpha_{M^{\red,r}}(\Lambda^{\red}) = \alpha_M(\Lambda)$. In

3082: particular, we can define a grading of the reduced symplectic

3083: automorphisms by

3084: \begin{equation} \label{eq:lift-lambda}

3085: \alphagr_{\phi^{\red,r}}(\Lambda^{\red}) = \alphagr_\phi(\Lambda).

3086: \end{equation}

3087: 

3088: \begin{lemma} \label{th:negative-reduction}

3089: Suppose that the $\phigr^{\red,r}$ are all negative. Then $\phigr$ is

3090: negative.

3091: \end{lemma}

3092: 

3093: \proof Choose some $r$. By assumption there is a ${d_0}$ such that

3094: the grading of $(\phigr^{\red,r})^{d_0}$ is negative, and then the

3095: same thing holds for neighbouring $r$. By a covering argument, one

3096: sees that there is a $d_0$ such that the gradings of the $d_0$-th

3097: iterates of all reduced maps are negative. Applying \eqref{eq:growth}

3098: shows that for some large $d$, the grading of each

3099: $(\phigr^{\red,r})^d$ will be $<-n$ everywhere. By

3100: \eqref{eq:lift-lambda}, this means that at each point of $M$ there

3101: are some Lagrangian tangent subspaces whose $\phigr^d$-grading is

3102: $<-n$. From this and Lemma \ref{th:graded-variation} one gets the

3103: desired conclusion. \qed

3104: 

3105: \subsection{}

3106: We will now look at the local models for the ``large complex

3107: structure limit'' monodromy, first under the rather unrealistic

3108: assumption that both the symplectic and complex structures are

3109: standard, which allows one to integrate the monodromy vector field

3110: explicitly.

3111: 

3112: \begin{assumptions} \label{as:1}

3113: Fix $n \geq 2$ and $2 \leq k \leq n$. We take $Y = \C^n$ with the

3114: standard complex structure and symplectic form $\o_Y$, and the map

3115: \[

3116: p: \C^n \longrightarrow \C, \quad y \longmapsto y_1 y_2 \dots y_k.

3117: \]

3118: $Y$ should carry the complex volume form (away from the zero fibre)

3119: \[

3120: \frac{dy_1}{y_1} \wedge \dots \wedge \frac{dy_k}{y_k} \wedge dy_{k+1}

3121: \wedge \dots \wedge dy_n.

3122: \]

3123: \end{assumptions}

3124: 

3125: The monodromy vector field $Z$ is the unique horizontal lift of the

3126: clockwise rotational field $-i\zeta \partial_\zeta$. Explicitly,

3127: \[

3128:  Z = -i

3129:  \frac{\left(\frac{y_1}{|y_1|^2}, \dots, \frac{y_k}{|y_k|^2},0,\dots,0\right)}

3130:  {\frac{1}{|y_1|^2} + \dots + \frac{1}{|y_k|^2}},

3131: \]

3132: In particular, since each $|y_j|^2$ is invariant, the flow $\psi_t$

3133: of $Z$ is well-defined for all times, in spite of the noncompactness

3134: of the fibres. As in our discussion of Lefschetz pencils, we define

3135: complex volume forms $\eta_{Y_z}$ on the fibres by (formally)

3136: dividing $\eta_Y$ by some complex one-form on the base, which in this

3137: case will be $d\zeta/\zeta^2$. As a consequence, if at some point $y

3138: \in Y_\zeta = p^{-1}(\zeta)$ we have a Lagrangian subspace $\Lambda

3139: \subset TY_y$ with $Dp(\Lambda) = a\R$, $a \in S^1$, then the phase

3140: of the vertical part $\Lambda^v = \Lambda \cap \ker(Dp)$ inside

3141: $Y_\zeta$ is related to the phase of $\Lambda$ by

3142: \begin{equation} \label{eq:divide-degree}

3143: \alpha_{Y_\zeta}(\Lambda^v) = \frac{\zeta^4}{a^2 |\zeta|^4}

3144: \alpha_Y(\Lambda).

3145: \end{equation}

3146: 

3147: For each point $y$ with $p(y) = \zeta \neq 0$, consider the

3148: Lagrangian tangent subspace $\Lambda_y = \R i y_1 \oplus \dots \oplus

3149: \R i y_k \oplus \R^{n-k} \subset TY_y$. This satisfies

3150: $\alpha_Y(\Lambda_y) = (-1)^k$ and $Dp(\Lambda_y) = i\zeta \R$, hence

3151: \eqref{eq:divide-degree} says that

3152: \begin{equation} \label{eq:lambda-shift}

3153: \alpha_{Y_z}(\Lambda_y^v) = \frac{\zeta^2}{|\zeta|^2} (-1)^{k-1}.

3154: \end{equation}

3155: $\psi_t$ takes the $\Lambda_y$ to each other, and rotates the base

3156: coordinate $\zeta$, hence it also maps the $\Lambda_y^v$ to each

3157: other. Let $h_\zeta = \psi_{2\pi}|Y_z: Y_z \rightarrow Y_z$ be the

3158: monodromy on the fibre over $\zeta$. By definition and

3159: \eqref{eq:lambda-shift}, the canonical grading of this map satisfies

3160: \begin{equation} \label{eq:integrate-shift}

3161: \begin{aligned}

3162:  \alphagr_{h_{\zeta}}(\Lambda_y^v)

3163:  & = \frac{1}{2\pi} \int_0^{2 \pi}

3164:  \frac{d}{dt} \arg(\alpha_{Y_z}(D\psi_t(\Lambda_y^v))) \, dt

3165:  \\ & = \frac{1}{2\pi} \int_0^{2 \pi}

3166:  \frac{d}{dt} \arg( (-1)^{k-1} e^{-2it}) \, dt = -2.

3167: \end{aligned}

3168: \end{equation}

3169: If one passes to a sufficiently large iterate such as $h_\zeta^n$,

3170: the grading on $\Lambda_y^v$ is $-2n$, hence by Lemma

3171: \ref{th:graded-variation} the grading of any $\Lambda$ is $<-n-1$. In

3172: a slight abuse of terminology (since $Y_\zeta$ is not compact), we

3173: conclude that the monodromy is a negative graded symplectic

3174: automorphism, in a way which is uniform for all $\zeta$.

3175: 

3176: \subsection{}

3177: A different local model appears at the singular points of the base

3178: locus of our pencil. In Ruan's work on the SYZ conjecture, see in

3179: particular \cite{ruan02b}, these are responsible for the

3180: singularities of the Lagrangian torus fibration. For closely related

3181: reasons, the negativity of the monodromy breaks down at such points.

3182: 

3183: \begin{assumption} \label{as:2}

3184: Fix $n \geq 3$ and $3 \leq k \leq n$. We again take $Y = \C^n$ with

3185: the standard complex and symplectic forms, but now with the rational

3186: map

3187: \[

3188: p: Y \dashrightarrow \C, \qquad p(y) = \frac{y_2 \dots y_k}{y_1},

3189: \]

3190: in the presence of the complex volume form (away from the closure of

3191: the zero fibre)

3192: \[

3193: \eta_Y = dy_1 \wedge \frac{dy_2}{y_2} \wedge \dots \wedge

3194: \frac{dy_k}{y_k} \wedge dy_{k+1} \wedge \dots \wedge dy_n.

3195: \]

3196: \end{assumption}

3197: 

3198: Denote by $Y_\zeta$ the closure of $p^{-1}(y)$. For $\zeta \neq 0$,

3199: each such fibre can be identified with $\C^{n-1}$ by projecting to

3200: $(y_2,\dots,y_n)$. We will use this identification freely, hoping

3201: that this does not cause too much confusion. For instance, the

3202: symplectic form on $Y_\zeta \iso \C^{n-1}$ is given by

3203: \begin{equation} \label{eq:zeta-symplectic-form}

3204:  \o_{Y_\zeta} = \frac{i}{2} \partial\bar\partial

3205:  \Big(\frac{|y_2\dots y_k|^2}{|\zeta|^2}

3206:  + |y_2|^2 + \dots + |y_n|^2 \Big)

3207: \end{equation}

3208: As in our previous model, we write $Z$ for the clockwise monodromy

3209: vector field (defined by passing to the graph of $p$). In view of our

3210: identification, this becomes a $\zeta$-dependent vector field on

3211: $\C^{n-1}$, namely:

3212: \[

3213: Z = -i

3214: \frac{\left(\frac{y_2}{|y_2|^2},\dots,\frac{y_k}{|y_k|^2},0,\dots,0\right)}

3215: {\frac{1}{|y_1|^2} + \frac{1}{|y_2|^2} + \dots + \frac{1}{|y_k|^2}}

3216: \]

3217: where $y_1 = y_2 \dots y_k/\zeta$. Again, this preserves $|y_2|^2,

3218: \dots, |y_n|^2$, so that the flow $\psi_t$ is well-defined for all

3219: times.

3220: 

3221: Taking $\eta_Y$ on the total space and $d\zeta/\zeta^2$ on the base,

3222: one finds that the induced volume form $\eta_{Y_\zeta}$ is the

3223: standard form on $\C^{n-1}$. At each point $y \in Y_\zeta$ with $y_2

3224: \dots y_k \neq 0$, consider the Lagrangian subspace $\Lambda_y^v = \R

3225: i y_2 \oplus \dots \oplus \R i y_k \oplus \R^{n-k}$. These subspaces

3226: are preserved by the flow $\psi_t$, and a computation similar to

3227: \eqref{eq:integrate-shift} shows that the canonical grading of the

3228: monodromy map $h_\zeta = \psi_{2\pi}|Y_\zeta$ satisfies

3229: \[

3230:  \alphagr_{h_\zeta}(\Lambda_y^v) = - 2

3231:  \frac{1}{1 + \frac{1}{\frac{|y_1|^2}{|y_2|^2} +

3232:  \cdots + \frac{|y_1|^2}{|y_k|^2}}}.

3233: \]

3234: This is obviously negative, but of course we also need to consider

3235: the grading at the points where $\Lambda_y^v$ is not defined. Suppose

3236: for simplicity that the sizes of the first $k-1$ coordinates of $y

3237: \in Y_\zeta$ are ordered, $|y_2| \leq |y_3| \leq \dots |y_{k-1}|$.

3238: Then

3239: \[

3240:  \alphagr_{h_\zeta}(\Lambda_y) \leq -2

3241:  \frac{1}{1 + \frac{|\zeta|^2}{|y_3|^{2(k-1)}}},

3242: \]

3243: which means that the grading is bounded above by something which

3244: depends on the size of the second smallest coordinate. As a

3245: consequence, if one has a sequence

3246: \[

3247:  y^{(m)} \rightarrow y^{(\infty)}, \quad

3248:  y_2^{(m)} \dots y_k^{(m)} \neq 0 \text{ for all $m$,}

3249: \]

3250: and such that only one of the $(y_2^{(\infty)}, \dots,

3251: y_k^{(\infty)})$ vanishes, the

3252: $\alphagr_{h_\zeta}(\Lambda^v_{y^{(m)}})$ are bounded above by a

3253: negative constant, hence cannot go to zero. Roughly speaking, this

3254: means that $h_\zeta$ is negative away from the subset where at least

3255: two of the $y_2,\dots,y_k$ vanish. In the special case $k = n = 3$,

3256: this subset is a point $0 \in Y_{\zeta}$; one can then apply

3257: \eqref{eq:compose-gradings} and Lemma \ref{th:graded-variation} much

3258: as in our previous local model, to get the following statement:

3259: 

3260: \begin{lemma} \label{th:bad-points}

3261: Suppose that $k = n = 3$. Let $W \subset Y_\zeta$ be some open ball

3262: around the origin. Then $h_\zeta(W) = W$, and there is a $d>0$ such

3263: that the grading of $h_\zeta^d|(Y_\zeta \setminus W)$ is negative.

3264: \qed

3265: \end{lemma}

3266: 

3267: \begin{remark}

3268: As mentioned before, ideas associated to the SYZ conjecture suggest

3269: that the monodromy should be a fibrewise translation in a SLAG

3270: fibration. From this point of view, the negativity of the monodromy

3271: is based on the idea that the phases of the tangent spaces to the

3272: fibration should be shifted by $-2$. In the situation of Assumptions

3273: \ref{as:1}, this is indeed the case: $L_r \iso T^{k-1} \times

3274: \R^{n-k} \subset Y_\zeta$, given by

3275: \[

3276:  L_r = \{|y_1|^2 = r_1, \dots, |y_k|^2 = r_k, \im(y_{k+1}) = r_{k+1},

3277:  \dots, \im(y_n) = r_n\}

3278: \]

3279: with $r_1 \dots r_k = |\zeta|^2$, are special Lagrangian with respect

3280: to $\eta_{Y_\zeta}$. The monodromy translates each of them, and the

3281: shift is computed in \eqref{eq:integrate-shift}. For Assumptions

3282: \ref{as:2}, there is a standard SLAG fibration with singularities on

3283: $Y_\zeta \iso \C^{n-1}$ (with respect to

3284: \eqref{eq:zeta-symplectic-form} and the standard complex volume

3285: form), see e.g.\ \cite{gross01b}:

3286: \begin{multline} \label{eq:syz}

3287:  \qquad L_r = \{

3288:  |y_3|^2 - |y_2|^2 =

3289:  r_1, \, \dots \, |y_k|^2 - |y_2|^2 = r_{k-2}, \\

3290:  \im(y_2\dots y_k) = r_{k-1}, \,

3291:  \im(y_{k+1}) = r_k, \dots, \im(y_n) = r_{n-1}\}. \qquad

3292: \end{multline}

3293: However, the monodromy preserves this only asymptotically far away

3294: from the singular locus, and hence the grading is only asymptotically

3295: equal to $-2$. To see more precisely what happens, we apply a

3296: symplectic reduction procedure, which means mapping $Y_\zeta

3297: \rightarrow \C$ by sending $y$ to $w = y_2y_3 \dots y_k$. The fibres

3298: of the SLAG fibration lie over horizontal lines in the base $\C$. The

3299: clockwise monodromy also fibres over a diffeomorphism of $\C$, which

3300: rotates each point $w$ by some $|w|$-dependent angle. For $|w| \gg

3301: 0$, this is almost a full clockwise rotation, hence it approximately

3302: preserves the horizontal lines, but near $w = 0$ the angle of

3303: rotation undergoes a full change $0$ to $-2\pi$.

3304: 

3305: On the other hand, as pointed out to me by Mark Gross, in the $K3$

3306: case one can construct a symplectic automorphism which preserves all

3307: the fibres of a SLAG fibration. This is obtained by hyperk{\"a}hler

3308: rotation from a complex automorphism, and it is presumably isotopic

3309: to the monodromy map, even though that has not strictly speaking been

3310: proved. So, it may well be that the failure of our local models to

3311: preserve the fibrations \eqref{eq:syz}, and the corresponding

3312: breakdown in negativity, could be repaired by an isotopy which is

3313: more global in nature.

3314: \end{remark}

3315: 

3316: \subsection{}

3317: We now relax our previous assumptions on the complex and symplectic

3318: structures, for the first local model. We choose to work in Darboux

3319: coordinates, and with a complex structure and holomorphic function

3320: which are not standard, but which still preserve a $T^k$ symmetry.

3321: 

3322: \begin{assumptions} \label{as:1prime}

3323: Fix $n \geq 2$ and $2 \leq k \leq n$.

3324: \begin{itemize} \itemsep1em

3325: \item

3326: $Y \subset \C^n$ is an open ball around the origin, carrying the

3327: standard symplectic form $\o_Y$ and the standard diagonal

3328: $T^k$-action

3329: \[

3330: \rho_s(y) = (e^{is_1}y_1,\dots,e^{is_k}y_k,y_{k+1},\dots,y_n),

3331: \]

3332: with moment map $\mu: Y \rightarrow (\Rgeq)^k$. For any regular

3333: moment value $r$, the quotient $Y^{\red,r}$ can be identified with an

3334: open subset of $\C^{n-r}$, with the standard symplectic form

3335: $\o_{Y^{\red,r}}$. We denote points in the reduced spaces by

3336: $y^{\red} = (y_{k+1},\dots,y_n)$.

3337: 

3338: \item

3339: $J_Y$ is a complex structure which is tamed by $\o_Y$. At the point

3340: $y = 0$ (but not necessarily elsewhere), it is $\o_Y$-compatible and

3341: $T^k$-invariant.

3342: 

3343: \item

3344: $p: Y \rightarrow \C$ is a $J_Y$-holomorphic function with the

3345: following properties:

3346: \begin{romanlist}

3347: \item \label{item:t-sym}

3348: $p(\rho_s(y)) = e^{i(s_1+\dots+s_k)}p(y)$.

3349: \item \label{item:nondeg}

3350: $\partial_{y_1}\dots \partial_{y_k} p$ is nonzero at the point $y =

3351: 0$.

3352: \end{romanlist}

3353: 

3354: \item

3355: $\eta_Y$ is a $J_Y$-complex volume form on $Y \setminus p^{-1}(0)$,

3356: such that $p(y)\eta_Y$ extends to a smooth form on $Y$, which is

3357: nonzero at $y = 0$. Let $\alpha_Y$ be the corresponding phase

3358: function.

3359: \end{itemize}

3360: \end{assumptions}

3361: 

3362: \begin{lemma} \label{th:product-function}

3363: One can write

3364: \[

3365: p(y) = 2^{-k/2} y_1 y_2\dots y_k\, q(\half |y_1|^2,\dots,\half

3366: |y_k|^2,y_{k+1},\dots,y_n)

3367: \]

3368: for some smooth function $q$ satisfying $q(0) \neq 0$.

3369: \end{lemma}

3370: 

3371: \proof Consider the Taylor expansion of $p$ around $y = 0$. From

3372: property \ref{item:t-sym} of $p$ one sees that each monomial which

3373: occurs with nonzero coefficient in that expansion contains one of the

3374: factors $y_l$ for each $1 \leq l \leq k$. As a consequence,

3375: $p(y)/(y_1\dots y_k)$ extends smoothly over $y = 0$. The same

3376: argument can be applied to points where only certain of the

3377: coordinates $y_1 \dots y_k$ vanish. Again applying \ref{item:t-sym},

3378: one sees that $p(y)/(y_1\dots y_k)$ is $T^k$-invariant, which means

3379: that it can be written in the form stated above. Finally, $q(0) \neq

3380: 0$ is property \ref{item:nondeg}. \qed

3381: 

3382: \begin{lemma} \label{th:shrink}

3383: After making $Y$ smaller if necessary, one has

3384: \begin{align*}

3385:  p(y) = 0 & \Longleftrightarrow  y_l = 0

3386:  \text{ for some } l = 1,\dots,k; \\

3387:  Dp(y) = 0 & \Longleftrightarrow y_l = 0

3388:  \text{ for at least two } l = 1,\dots,k.

3389: \end{align*}

3390: \end{lemma}

3391: 

3392: \proof Suppose that $q(y) \neq 0$ on the whole of $Y$. Then the first

3393: statement is obvious, and so is the $\Leftarrow$ implication in the

3394: second one. Conversely, suppose that $y$ is a critical point.

3395: Property \ref{item:t-sym} implies that $p(y) = 0$, so that $y_l = 0$

3396: for some $l = 1,\dots,k$. From 0 = $Dp_y = y_1 \dots \hat{y}_l \dots

3397: y_k \, q(y)\, dy_l$ one sees that another of the $y_1,\dots, y_k$ has

3398: to vanish too. \qed

3399: 

3400: We will assume from now on that the conclusions of Lemma

3401: \ref{th:shrink} hold. Consider the function $H(y) = -\half |p(y)|^2$,

3402: its Hamiltonian vector field $X$ and flow $\phi_t$. Since $H$ is

3403: $T^k$-invariant, the flow (where defined) is equivariant and

3404: preserves the level sets of the moment map. For every regular moment

3405: value $r$, we can consider the induced function

3406: \begin{equation} \label{eq:h-red}

3407: H^{\red,r}(y^\red) = -\half r_1 \dots r_k \,

3408: |q(r_1,\dots,r_k,y_{k+1},\dots,y_n)|^2,

3409: \end{equation}

3410: on $Y^{\red,r}$, its vector field $X^{\red,r}$ and flow

3411: $\phi_t^{\red,r}$. If the moment value is small, so are

3412: \eqref{eq:h-red} and all its $y^\red$-derivatives, which means that

3413: the reduced flow moves very slowly. We need a quantitative version of

3414: this:

3415: 

3416: \begin{lemma} \label{th:small-flow}

3417: Given $\epsilon_1,\epsilon_2>0$, there are

3418: $\delta_1,\delta_2,\delta_3>0$ such that for all

3419: \[

3420: \begin{cases}

3421:  r \in (\R^{\scriptscriptstyle>0})^k &

3422:  \text{with $||r|| < \delta_1$}, \\

3423:  y^\red \in Y^{\red,r} &

3424:  \text{with $||y^\red|| < \delta_2$}, \\

3425:  t \in \R &

3426:  \text{with $|t| < \delta_3 r_1^{-1} \dots r_k^{-1}$},

3427: \end{cases}

3428: \]

3429: the following holds: $\phi_t^{\red,r}$ is well-defined near $y^\red$,

3430: and satisfies

3431: \begin{align}

3432:  \label{eq:c0} &

3433:  ||\phi_t^{\red,r}(y^{\red}) || < \epsilon_1, \\

3434:  \label{eq:c1} &

3435:  ||(D\phi_t^{\red,r})_{y^{\red}} - \mathrm{Id} || < \epsilon_2.

3436: \end{align}

3437: \end{lemma}

3438: 

3439: \proof By Lemma \ref{th:product-function} there are $\delta_1>0$,

3440: $\delta_2 \in (0;\min(\epsilon_1/2,\epsilon_2/2))$ and a $C>0$, such

3441: that

3442: \[

3443: ||X^{\red,r}_{y^\red}||, \; ||DX^{\red,r}_{y^{\red}}|| \leq C

3444: r_1\dots r_k

3445: \]

3446: for each $||r|| < \delta_1$, $||y^{\red}|| < 2\delta_2$ (we use the

3447: Hilbert-Schmidt norm for matrices $DX^{\red,r}$, normalized to

3448: $||\mathrm{Id}|| = 1$). We claim that one can take $\delta_3 =

3449: \delta_2/C $. The first estimate \eqref{eq:c0} is clear: if one

3450: starts at a point $y^\red$ of norm less than $\delta_2$, and moves

3451: with speed at most $C r_1 \dots r_k$ for a time less than $\delta_2

3452: C^{-1} r_1^{-1} \dots r_k^{-1}$, the endpoint will be of norm less

3453: than $2\delta_2 < \epsilon_1$. Next, use the linearized flow to get

3454: an exponential growth bound for the time interval $t$ relevant to us:

3455: $||D\phi_t^{\red,r}|| \leq \exp(C r_1 \dots r_k t) \leq

3456: \exp(\delta_2)$. One may assume that $\exp(\delta_2) \leq 2$. By

3457: plugging this back into the linearized equation, one gets the linear

3458: estimate $||D\phi_t^{\red,r}-\mathrm{Id}|| \leq 2 C r_1 \dots r_k t

3459: \leq 2 \delta_2 < \epsilon_2$, from which \eqref{eq:c1} follows. \qed

3460: 

3461: We now need to deal with the fact that $J_Y$ and $\eta_Y$ are only

3462: approximately $T^k$-invariant. Let $J_Y'$ be the constant complex

3463: structure on $Y$ which agrees with $J_Y$ at $y = 0$. This is

3464: $\o_Y$-compatible and $T^k$-invariant, so

3465: \[

3466: J_Y' = \begin{pmatrix} i \cdot 1_k & 0 \\ 0 & J_{Y^{\red}}'

3467: \end{pmatrix}.

3468: \]

3469: $J_{Y^{\red}}'$ is the induced complex structure on $Y^{\red,r}$ for

3470: all $r$. Similarly, we consider the $J_Y'$-complex volume form

3471: $\eta'_Y$ which is obtained by taking the value of $p(y)\eta_Y$ at

3472: the point $y = 0$, and dividing that by $y_1\dots y_k$. This can be

3473: written as

3474: \[

3475: \eta_Y' = i^k \frac{dy_1}{y_1} \wedge \dots \wedge \frac{dy_k}{y_k}

3476: \wedge \eta_{Y^\red}',

3477: \]

3478: hence is $T^k$-invariant. $\eta_{Y^\red}'$, which is the induced

3479: complex volume form on the reduced spaces, is again constant and

3480: $r$-independent. Let $\alpha_Y'$ be the phase function associated to

3481: $(J_Y',\eta_Y')$ (unlike $\alpha_Y$, this is a classical phase

3482: function), and $\alpha_{Y^{\red,r}}'$ the induced phase functions on

3483: the quotients.

3484: 

3485: \begin{lemma} \label{th:phase-difference}

3486: Given $\epsilon_3>0$, there is $\delta_4>0$ such that for all $||y||

3487: < \delta_4$ with $p(y) \neq 0$, and all $\Lambda \in \LL_{Y,y}$,

3488: \[

3489: \Big|\frac{1}{2\pi}\arg(\alpha_Y'(\Lambda)/\alpha_Y(\Lambda))\Big| <

3490: \epsilon_3

3491: \]

3492: (by this we mean that there is a branch of the argument with that

3493: property).

3494: \end{lemma}

3495: 

3496: \proof The quotient $\alpha_Y'(\Lambda)/\alpha_Y(\Lambda)$ does not

3497: change if we multiply both $\eta_Y$, $\eta_Y'$ with $y_1\dots y_k$.

3498: But by construction,

3499: \begin{align*}

3500:  & y_1 \dots y_k \eta_Y = 2^{-k/2} q(\half |y_1|^2,\dots, \half |y_k|^2,

3501:  y_{k+1},\dots,y_n)^{-1} \sigma_y, \\

3502:  & y_1 \dots y_k \eta_Y' = 2^{-k/2} q(0)^{-1} \sigma_0

3503: \end{align*}

3504: where $\sigma$ is some smooth $n$-form with $\sigma_0 \neq 0$, so the

3505: associated phases also become close as $y \rightarrow 0$. \qed

3506: 

3507: $\phi_t$ has a canonical grading $\alphagr_{\phi_t}$ with respect to

3508: $\alpha_Y$, obtained by starting with the trivial grading for

3509: $\phi_0$ and extending continuously. Denote by $\alphagr_{\phi_t}'$

3510: the grading obtained in the same way, but with respect to

3511: $\alpha_Y'$. There are corresponding gradings

3512: $\alphagr_{\phi_t^{\red,r}}'$ on the reduced spaces, with respect to

3513: $\alpha_{Y^{\red,r}}'$.

3514: 

3515: \begin{lemma} \label{th:small-phase}

3516: Given $\epsilon_4>0$, one can find $\delta_5,\delta_6,\delta_7>0$

3517: such that for all

3518: \[

3519: \begin{cases}

3520:  r \in (\R^{\scriptscriptstyle>0})^k &

3521:  \text{with $||r|| < \delta_5$}, \\

3522:  y^\red \in Y^{\red,r} &

3523:  \text{with $||y^\red|| < \delta_6$}, \\

3524:  \Lambda^\red \in \LL_{Y^{\red,r},y^\red} & \text{and} \\

3525:  t \in \R &

3526:  \text{with $|t| < \delta_7 r_1^{-1} \dots r_k^{-1}$},

3527: \end{cases}

3528: \]

3529: $\phi_t^{\red,r}$ is well-defined near $y^\red$, and its grading is

3530: $|\alphagr_{\phi_t^{\red,r}}'(\Lambda^\red)| < \epsilon_4$.

3531: \end{lemma}

3532: 

3533: \proof This is an easy consequence of Lemma \ref{th:small-flow}: one

3534: can achieve that the derivative $(D\phi_t^{\red,r})_{y^\red}$ remains

3535: very close to the identity in the whole range of $t$'s that we are

3536: considering. Since the volume form $\eta_{Y^{\red,r}}$ is constant

3537: and independent of $r$, it follows that the grading remains small.

3538: \qed

3539: 

3540: \begin{lemma} \label{th:phi-grading-estimate}

3541: Given $\epsilon_5>0$, one can find $\delta_8,\delta_9,\delta_{10} >

3542: 0$ such that for all

3543: \[

3544: \begin{cases}

3545:  y \in Y & \text{with $0 < |p(y)| < \delta_8$ and $||y|| < \delta_9$,} \\

3546:  \Lambda \in \LL_{Y,y}, & \text{and} \\

3547:  t \in \R & \text{with $|t| < \delta_{10} |p(y)|^{-2}$,}

3548: \end{cases}

3549: \]

3550: $\phi_t$ is well-defined near $y$, and its grading satisfies

3551: $

3552: |\alphagr_{\phi_t}(\Lambda)| < n + \epsilon_5.

3553: $

3554: \end{lemma}

3555: 

3556: \proof Apply Lemma \ref{th:small-flow} to the corresponding point

3557: $y^\red$ in the quotient. In particular, one can choose

3558: $\delta_8,\delta_9,\delta_{10}$ in such a way that the reduced flow

3559: exists near $y^\red$ for $|t| < \delta_{10} |p(y)|^{-2}$. Since the

3560: fibres of the quotient map are compact (tori), it follows that

3561: $\phi_t$ is well-defined near $y$ in the same range of $t$. We now

3562: apply the same basic reasoning as in Lemma

3563: \ref{th:negative-reduction}. After possibly making

3564: $\delta_8,\delta_9$ smaller, Lemma \ref{th:small-phase} ensures that

3565: there is a $\Lambda \in \LL_{Y,y}$ such that

3566: $|\alphagr_{\phi^t}'(\Lambda)| < \epsilon_5/2$. By Lemma

3567: \ref{th:graded-variation}, the grading is $< n + \epsilon_5/2$ for

3568: all other Lagrangian subspaces at the same point. Finally, one uses

3569: Lemma \ref{th:phase-difference} to pass from the grading associated

3570: to $\eta_Y'$ to that for $\eta_Y$, again making the $\delta$s smaller

3571: so that the phase difference becomes $< \epsilon_5/2$. \qed

3572: 

3573: The next step is to relate $\phi_t$ to the monodromy. As usual, we

3574: take the vector field $-i\zeta \partial_\zeta$ on $\C^*$, and lift it

3575: in the unique way to a horizontal vector field $Z$ on $Y \setminus

3576: p^{-1}(0)$. This is well-defined because $\o_Y$ tames $J_Y$, hence

3577: the regular fibres $Y_\zeta$, $\zeta \neq 0$, are symplectic

3578: submanifolds. We claim that

3579: \begin{equation} \label{eq:f}

3580: Z = f X

3581: \end{equation}

3582: for some positive function $f$ on $Y \setminus p^{-1}(0)$. First,

3583: because $H$ is constant on each fibre, the associated vector field

3584: $X$ is horizontal. Secondly, because $\o_Y(X,Z) = 0$ and the space of

3585: horizontal vectors $TY^h$ is two-dimensional, it is true that $X =

3586: f^{-1}Z$ for some $f$. A little more thought shows that

3587: \begin{equation} \label{eq:explicit-psi}

3588: f = \frac{\o_Y|TY^h}{p^*\o_\C|TY^h} > 0,

3589: \end{equation}

3590: where $TY^h$ is the $\o_Y$-horizontal subspace (which is

3591: two-dimensional, so any two two-forms on it differ by a scalar). Let

3592: $(\psi_s)$ be the flow of $Z$, so that $h_\zeta =

3593: \psi_{2\pi}|Y_\zeta$ is the clockwise monodromy (of course, like

3594: $\phi_s$ this flow is only partially defined). As a consequence of

3595: \eqref{eq:f}, the unparametrized flow lines of $\phi_t$ and $\psi_t$

3596: agree, so

3597: \begin{equation} \label{eq:gt}

3598: \psi_t(y) = \phi_{g_t(y)}(y), \quad \text{where }\;\; g_t(y) =

3599: \int_0^t f(\psi_\tau(y)) d\tau.

3600: \end{equation}

3601: 

3602: \begin{lemma} \label{th:minimum-speed}

3603: For every $d>0$, $\epsilon_6>0$ there are $\delta_{11},\delta_{12}

3604: > 0$ such that the following holds. For all $0 < |\zeta| < \delta_{11}$ and

3605: all $y \in Y_\zeta$ with $||y|| < \delta_{12}$, the $d$-fold

3606: monodromy $h^d_\zeta$ is well-defined, and \eqref{eq:gt} holds for $t

3607: = 2\pi d$ with

3608: \[

3609: g_{2\pi d}(y) \leq \frac{\epsilon_6}{|\zeta|^2}.

3610: \]

3611: \end{lemma}

3612: 

3613: \proof Using property \ref{item:t-sym} of $p$ and its

3614: $J_Y$-holomorphicity, one obtains that $Dp_y(J_Y(-iy_1,0,\dots,0)) =

3615: p(y)$ and hence $dH_y(J_Y(-iy_1,0,\dots,0)) = |p(y)|^2 = |\zeta|^2$.

3616: It follows that for small $y$, $||X_y|| \geq c_1 |\zeta|^2/||y||$ for

3617: some $c_1>0$. Hence there is a $C_2>0$ such that

3618: \[

3619:  C_2^{-1} |\zeta|^4/||y||^2 \leq \o_Y(X,J_YX) = f^{-2} \o_Y(Z,J_YZ).

3620: \]

3621: On the other hand, $\o_Y(Z,J_YZ) = |\zeta|^2 f$ by definition of $Z$

3622: and \eqref{eq:explicit-psi}, so finally,

3623: \[

3624:  f(y) \leq C_2 \frac{||y||^2}{|\zeta|^2}.

3625: \]

3626: Integrating gives a similar inequality $g_{2\pi d}(y) \leq C_3

3627: ||y||^2/|\zeta|^2$. Clearly, by restricting to smaller $y$ one can

3628: make the right hand side less than $\epsilon_6/|\zeta|^2$ for any

3629: desired $\epsilon_6$ (this, by the way, is the crucial estimate in

3630: our whole computation). \qed

3631: 

3632: We use $J_Y$-complex volume forms $\eta_{Y_\zeta}$ on the fibres

3633: obtained in the usual way from $\eta_Y$, so that

3634: \eqref{eq:divide-degree} holds. Take $d>0$, $y \in Y_\zeta$, and a

3635: Lagrangian subspace $\Lambda \subset TY_y$ with $Dp(\Lambda) =

3636: i\zeta\R$. This is necessarily of the form $\Lambda = \Lambda^v

3637: \oplus \R Z_y$ for some Lagrangian subspace $\Lambda^v \subset

3638: T_y(Y_\zeta)$. By \eqref{eq:f} and \eqref{eq:gt} one has

3639: \begin{equation} \label{eq:relation-between-diff}

3640:  D(\psi_t)_y = D(\phi_{g_t})_y +

3641:  X_{\psi_t(y)} \otimes dg_t =

3642:  D(\phi_{g_t})_y + Z_{\psi_t(y)} \otimes \frac{dg_t}{f(\psi_t(y))}

3643: \end{equation}

3644: Since $\Lambda$ contains $Z_y$, and $\psi_t$ is the clockwise

3645: parallel transport flow, $D(\psi_t)_y(\Lambda)$ contains

3646: $Z_{\psi_t(y)}$. From this and \eqref{eq:relation-between-diff} it

3647: follows that

3648: \[

3649:  D\psi_t(\Lambda) = D(\phi_{g_t})(\Lambda), \qquad

3650:  D(\psi_t|Y_\zeta)(\Lambda^v) = D(\phi_{g_t})(\Lambda) \cap \ker(Dp),

3651: \]

3652: and hence by \eqref{eq:divide-degree} that

3653: $\alpha_{\psi_t|Y_\zeta}(\Lambda^v) = e^{-4\pi i t}

3654: \alpha_{\phi_{g_t}}(\Lambda)$. By considering the behaviour of this

3655: over the time $t \in [0;2\pi d]$ one finds that the canonical grading

3656: of the monodromy satisfies

3657: \begin{equation} \label{eq:relative-shift}

3658:  \alphagr_{h_\zeta^d}(\Lambda^v) =

3659:  \alphagr_{\phi_{g_{2\pi d}(y)}}(\Lambda)-2d.

3660: \end{equation}

3661: 

3662: \begin{lemma} \label{th:local-model-1}

3663: For every $d>0$ and $\epsilon_7>0$ there are $\delta_{13},

3664: \delta_{14} >0$ such that the following holds. For every $y \in

3665: Y_\zeta$ with $0 < |\zeta| < \delta_{13}$ and $||y|| < \delta_{14}$,

3666: and every Lagrangian subspace $\Lambda^v \subset T(Y_\zeta)_y$, we

3667: have that $h_\zeta^d$ is well-defined near $y$, and its canonical

3668: grading satisfies

3669: \[

3670: \alphagr_{h_\zeta^d}(\Lambda^v) < n-2d + \epsilon_7.

3671: \]

3672: \end{lemma}

3673: 

3674: \proof Lemma \ref{th:phi-grading-estimate} says that one can choose

3675: $\delta_{13},\delta_{14}$ so small that there is a $\delta_{15}>0$

3676: with

3677: \begin{equation} \label{eq:epsilon7}

3678: |\alphagr_{\phi_t}(\Lambda)| < n + \epsilon_7

3679: \end{equation}

3680: for all $|t| < \delta_{15}/|p(y)|^2$. On the other hand, after

3681: possibly making $\delta_{13}, \delta_{14}$ smaller, one knows from

3682: Lemma \ref{th:minimum-speed} that $|g_{2\pi d}| <

3683: \delta_{15}/|p(y)|^2$, so \eqref{eq:epsilon7} applies to the values

3684: of $t$ that are relevant for our monodromy map.

3685: \eqref{eq:relative-shift} completes the proof. \qed

3686: 

3687: This shows that the basic feature of our first local model, the

3688: uniform negativity of the monodromy, survives even if one generalizes

3689: from the idealized situation which we considered first (Assumptions

3690: \ref{as:1}) to a more realistic one (Assumptions \ref{as:1prime}). In

3691: principle, one should carry out a parallel discussion for the second

3692: local model. This would be rather more delicate, since the grading

3693: becomes zero along a submanifold; an arbitrary small perturbation can

3694: potentially make it positive, thereby spoiling the entire argument.

3695: Fortunately, in the $K3$ case the second model occurs only at

3696: isolated points, and one can achieve that the local picture near

3697: those points is standard (satisfies Assumptions \ref{as:2}), so that

3698: the previous discussion suffices after all.

3699: 

3700: \subsection{}

3701: We will now explain how to patch together the local models to

3702: understand of the ``large complex structure limit'' monodromy. This

3703: is somewhat similar to the strategy used by Ruan to construct

3704: (singular) Lagrangian torus fibrations \cite{ruan02b}. We consider a

3705: Fano threefold $X$ with $o_X = \K_X^{-1}$, and a pair of sections

3706: $\sigma_{X,0},\sigma_{X,\infty}$ which generate a quasi-Lefschetz

3707: pencil of hypersurfaces $\{X_z\}$. From $\sigma_{X,\infty}$ we get a

3708: rational complex volume form $\eta_X$ with a pole along $X_\infty$.

3709: There are canonical induced complex volume forms $\eta_{X_z}$ on the

3710: fibres $X_z$ (if that is smooth and $z \neq \infty$), which are

3711: therefore Calabi-Yau surfaces. In fact, they must be $K3$s, since

3712: $H^1(X) = 0$ for any Fano variety, hence $H^1(X_z) = 0$ by the

3713: Lefschetz hyperplane theorem.

3714: 

3715: \begin{lemma} \label{th:kaehler-1}

3716: One can find a K{\"a}hler form $\o_X'$ in the same cohomology class

3717: as $\o_X$, with the following properties. (i) Near every point where

3718: three components of $X_\infty$ meet, $(X,\o_X',1/\pi)$ is modelled on

3719: $(Y,\o_Y,p)$ from Assumptions \ref{as:1}, with $n = k = 3$. (ii) Near

3720: every point where $X_0$ meets two components of $X_\infty$,

3721: $(X,\o_X',1/\pi)$ is modelled on $(Y,\o_Y,p)$ from Assumptions

3722: \ref{as:2}, with $n = k = 3$.

3723: \end{lemma}

3724: 

3725: This is easy because the set of such points (which we will call

3726: ``lowest stratum points'' in future) is finite. Near each of them,

3727: there are local holomorphic coordinates in which $(X,1/\pi)$ becomes

3728: $(Y,p)$. One can then locally modify the K{\"a}hler form to make it

3729: standard, see e.g.\ \cite[Lemma 7.3]{ruan02b} or \cite[Lemma

3730: 1.7]{seidel01}.

3731: 

3732: \begin{lemma} \label{th:kaehler-2}

3733: One can find a K{\"a}hler form $\o_X''$ in the same cohomology class

3734: as $\o_X$, which agrees with $\o_X'$ in a neighbourhood of the lowest

3735: stratum points, and such that in addition, any two components of

3736: $X_\infty$ meet orthogonally. \qed

3737: \end{lemma}

3738: 

3739: Ruan proves a much more general result in \cite[Theorem

3740: 7.1]{ruan02b}. The main idea is to apply the techniques from the

3741: previous Lemma to each fibre of the normal bundle $\nu$ to the

3742: intersection of two components of $X_\infty$.

3743: %For instance, if one follows the approach from \cite{ruan02b} then one

3744: %first supposes that the kaehler metric on each fibre is quadratic.

3745: %then one multiplies with \psi(|y|^2) to make it vanish near the

3746: %zero-section, but still \geq 0 everywhere in fibre direction.

3747: %Now one adds some really small multiple of a compactly supported

3748: %form, which is negative only where the other summand is strictly

3749: %positive, and which has the desired behaviour near the zero section.

3750: %A rescaling argument shows that the whole is still Kaehler

3751: %on the total space.

3752: %

3753: %Ruan uses a diagonalization argument, which works in this case only

3754: %because coincidence of eigenvalues of hermitian matrices happens

3755: %generically along a codimension 3 subset, hence never over a Riemann

3756: %surface.

3757: 

3758: Before proceeding, we need some more notation. Denote by $C$ the

3759: irreducible components of $X_\infty$, and by $L_C = \O(-C)$ the

3760: corresponding line bundles. Let $\gamma_C$ be their canonical

3761: sections, with a zero along $C$. Take two components $C \neq C'$, and

3762: choose a small neighbourhood $U_{C,C'}$ of $C \cap C'$. On this

3763: consider the vector bundle

3764: \[

3765: N_{C,C'} = L_C \oplus L_C^{-1},

3766: \]

3767: We want to equip $L_C$ and hence $N_{C,C'}$ with hermitian metrics

3768: and the corresponding connections, with the following prescribed

3769: behaviour near the lowest stratum pointsin $U_{C,C'}$:

3770: \begin{itemize} \itemsep1em

3771: \item

3772: If $x$ is a point where $C \cap C'$ meets another component $C''$, we

3773: choose local coordinates for $X$ as given by Lemma

3774: \ref{th:kaehler-1}, so that Assumptions \ref{as:1} hold, and a local

3775: trivialization of $L_C$ such that $\gamma_C(y) = y_1$. Take the

3776: trivial metric on $L_C$ with respect to this trivialization.

3777: 

3778: \item

3779: If $x$ is a point where $C \cap C'$ meets $X_0$, we choose local

3780: coordinates for $X$ as given by Lemma \ref{th:kaehler-1}, so that

3781: Assumptions \ref{as:2} hold, and a local trivialization of $L_C$ such

3782: that $\gamma_C(y) = y_2$. Again, we take the trivial metric.

3783: \end{itemize}

3784: 

3785: Take the K{\"a}hler metric associated to $\o_X''$, and use its

3786: exponential map to define a retraction $r_{C,C'}: U_{C,C'}

3787: \rightarrow C \cap C'$ of an open neighbourhood onto $C \cap C'$. A

3788: corresponding use of parallel transport on $N_{C,C'}$, with respect

3789: to our chosen connection, yields an isomorphism $R_{C,C'}:

3790: N_{C,C'}|U_{C,C'} \rightarrow r_{C,C'}^*(N_{C,C'}|C \cap C')$. Taking

3791: the two objects together, we have a map

3792: \[

3793:  \nu_{C,C'}: U_{C,C'} \setminus X_0

3794:   \xrightarrow{\xi_{C,C'} = (\gamma_C,\pi_M^{-1}\gamma_C^{-1})}

3795:  N_{C,C'}|U_{C,C'} \xrightarrow{R_{C,C'}} N_{C,C'}|C \cap C'.

3796: \]

3797: 

3798: \begin{lemma} \label{th:kaehler-3}

3799: After making $U_{C,C'}$ smaller if necessary, it carries a unique

3800: $T^2$-action $\sigma_{C,C'}$ which, under $\nu_{C,C'}$, corresponds

3801: to the obvious fibrewise $T^2$-action on $N_{C,C'}|C \cap C'$. In

3802: particular,

3803: \begin{equation} \label{eq:rotate-base}

3804: \pi_M(\sigma_{C,C',s}(x)) = e^{-i(s_1+s_2)}\pi_M(x).

3805: \end{equation}

3806: Moreover, at each point of $C \cap C'$, the action preserves $\o_X''$

3807: and the complex structure; the same holds in a neighbourhood of the

3808: lowest stratum points.

3809: \end{lemma}

3810: 

3811: \proof It is sufficient to verify the existence and uniqueness

3812: statement locally. First, take a point $x \in C \cap C'$ which is not

3813: a lowest stratum point. $\gamma_C^{-1}$ has a simple pole along $C$,

3814: and $\pi_M^{-1}$ has a simple zero along $C \cup C'$, hence

3815: $\pi_M^{-1}\gamma_C^{-1}$ has a simple zero along $C'$. It follows

3816: that $\xi_{C,C'}$ is transverse to the zero-section at $x$, so

3817: $\nu_{C,C'}$ is a local diffeomorphism; we simply pull back the

3818: $T^2$-action by it. By definition of the connection on $N_{C,C'}$,

3819: there is a commutative diagram

3820: \[

3821: \xymatrix{

3822:  {U_{C,C'} \setminus X_0} \ar[r]^-{\xi_{C,C'}} \ar[dr]_-{\pi_M^{-1}} &

3823:  {N_{C,C'}|U_{C,C'}} \ar[r]^-{R_{C,C'}} \ar[d] &

3824:  {N_{C,C'}|C \cap C'} \ar[dl] \\

3825:  & {\C} &

3826: }

3827: \]

3828: where the unlabeled arrows are the canonical pairings $L_C \oplus

3829: L_C^{-1} \rightarrow \C$. This shows that our pullback $T^2$-action

3830: satisfies \eqref{eq:rotate-base}. At the point $x$ itself,

3831: \begin{equation} \label{eq:dnu}

3832: D\nu_{C,C'}: TX_x \longrightarrow T(C \cap C')_x \oplus N_x

3833: \end{equation}

3834: is given by orthogonal projection to $C \cap C'$ with respect to the

3835: $o_{X,2}$-metric, together with $D\xi_x: TX_x \rightarrow N_x$. This

3836: is $\C$-linear, which implies that $\sigma_{C,C',x}$ preserves the

3837: complex structure. As for the symplectic structure, observe that by

3838: Lemma \ref{th:kaehler-2} $TX_x$ splits into three complex

3839: one-dimensional pieces, namely $T(C \cap C')_x$ and the normal

3840: directions to that inside $C,C'$. On the right hand side of

3841: \eqref{eq:dnu}, this corresponds to the splitting of $N_x$ into line

3842: bundles. The symplectic structures on the each summand do not

3843: correspond, but they differ only by a scalar, which is enough to

3844: conclude that $\sigma_{C,C'}$ is symplectic at $x$.

3845: 

3846: It remains to deal with the situation near the lowest stratum points.

3847: For this, we choose local coordinates and the trivialization of

3848: $N_{C,C'}$ as arranged before. Near a point of $C \cap C' \cap C''$,

3849: one then finds that $\nu_{C,C'}(y) = (y_3,y_1,y_2y_3)$, where the

3850: first component is the coordinate on $C \cap C'$. Hence, our

3851: $\sigma_{C,C'}$ near that point is the standard $T^2$-action in the

3852: variables $y_1,y_2$. Similarly, near a point of $C \cap C' \cap X_0$,

3853: $\nu_{C,C'}(y) = (y_1,y_2,y_3/y_1)$, so that one ends up with the

3854: $T^2$-action in the variables $y_2,y_3$. These obviously have the

3855: desired properties. \qed

3856: 

3857: We now pick an ordering of every pair $\{C,C'\}$ of components, and

3858: will use only the $T^2$-actions $\sigma_{C,C'}$ in that order.

3859: 

3860: \begin{lemma} \label{th:kaehler-4}

3861: There is a symplectic form $\o_X'''$ which tames the complex

3862: structure, such that for each intersection $C \cap C'$ of two

3863: components of $X_\infty$, there is a neighbourhood $U_{C,C'}$ such

3864: that the local $T^2$-action $\rho_{C,C'}$ leaves $\o_X'''$ invariant.

3865: Moreover, $\o_X'''$ lies in the same cohomology class as $\o_X''$,

3866: and agrees with it at each point $x \in C \cap C'$, as well as in a

3867: neighbourhood of the lowest stratum points.

3868: \end{lemma}

3869: 

3870: \proof Let $\bar{\o}_{C,C'} \in \Omega^2(U_{C,C'})$ be the result of

3871: averaging $\o_X''$ with respect to $\rho_{C,C'}$. This is a closed

3872: two-form; it agrees with $\o_X''$ along $C cap C'$, and also in a

3873: neighbourhood of the lowest stratum points. Using the Poincar{\'e}

3874: lemma one can write $\bar{\o}_{C,C'} - \o_X'' = d\beta_{C,C'}$, where

3875: $\beta_{C,C'}$ is a one-form that vanishes to second order along $C

3876: \cap C'$, and which is again zero in a neighbourhood of the lowest

3877: stratum point. Take a compactly supported cutoff function

3878: $\psi_{C,C'}$ on $U_{C,C'}$, which is equal to one near $C \cap C'$.

3879: One sees easily that as one rescales the normal directions to make

3880: the support of $\psi_{C,C'}$ smaller, $d(\psi_{C,C'} \beta_{C,C'})$

3881: becomes arbitrarily small in $C^0$-sense. It follows that for

3882: sufficiently small support, $\o_X'' + d(\psi_{C,C'} \beta_{C,C'})$ is

3883: a symplectic form which tames the complex structure. This agrees with

3884: $\o_X''$ near the lowest stratum points, and for that reason, one can

3885: carry out the construction simultaneously for all intersections $C

3886: \cap C'$. \qed

3887: 

3888: \begin{lemma}

3889: Let $x \in C \cap C'$ be a point where two components of $X_\infty$

3890: meet, which is not a lowest stratum point. Then there are local

3891: $\rho_{C,C'}$-equivariant Darboux coordinates near $x$, in which $X$

3892: with the map $1/\pi$, the symplectic structure $\o_X'''$, the given

3893: complex structure $J_X$, and the meromorphic complex volume form

3894: $\eta_X$, is isomorphic to $(Y,p,\o_Y,J_Y,\eta_Y)$ as in Assumptions

3895: \ref{as:1prime}. \qed

3896: \end{lemma}

3897: 

3898: This is straightforward checking, which we leave to the reader. The

3899: outcome of the whole construction is this:

3900: 

3901: \begin{prop} \label{th:negativity}

3902: Let $\gamma_\infty: [0;2\pi] \rightarrow \C$ be a circle of large

3903: radius $R \gg 0$. Then the graded monodromy

3904: \[

3905: \hgr_{\gamma_\infty} \in \Autgr(M_R)

3906: \]

3907: is isotopic (within that group) to a graded symplectic automorphism

3908: $\phigr$ with the following property. There is a finite set $\Sigma

3909: \subset X_z \cap X_\infty$ which admits arbitrarily small open

3910: neighbourhoods $W \subset X_z$, such that $\phi(W) = W$ and

3911: $\phigr|(X_z \setminus W)$.

3912: \end{prop}

3913: 

3914: \proof The main point is to replace the given K{\"a}hler form $\o_X$

3915: by $\o_X'''$, and we need to convince ourselves that this is

3916: permitted. For each regular $X_z$ there is a symplectic isomorphism

3917: \[

3918: \phi_z: (X_z,\o_{X_z}) \longrightarrow (X_z,\o_X'''|X_z)

3919: \]

3920: which carries $X_{0,\infty}$ to itself. This follows from Moser's

3921: Lemma together with the isotopy theorem for symplectic surfaces with

3922: orthogonal normal crossings, see \cite[Theorem 6.5]{ruan02b} for a

3923: more general statement valid in all dimensions. Next, an inspection

3924: of the construction of monodromy maps for $\pi_M$ shows that they can

3925: be defined using any symplectic form which tames the complex

3926: structure, and in particular $\o_X'''$. The resulting diagram of

3927: symplectic maps

3928: \[

3929: \xymatrix{

3930:  (X_R,\o_{X_R}) \ar[d]^-{h_{\gamma_\infty}}

3931:  \ar[r]^-{\phi_R} &

3932:  (X_R,\o_X'''|X_R) \ar[d]^-{h_{\gamma_\infty}'''} \\

3933:  (X_R,\o_{X_R}) \ar[r]^-{\phi_R} & (X_R,\o_X'''|X_R)

3934: }

3935: \]

3936: commutes up to symplectic isotopy rel $X_{R,\infty}$. Since $X_R$ is

3937: a $K3$ surface, Remark \ref{th:h1} shows that $h_{\gamma_\infty}$ is

3938: isotopic to $\phi_R^{-1} \circ h_{\gamma_\infty}''' \circ \phi_R$

3939: within $Aut(M_R)$. It is no problem to take the gradings into

3940: consideration. Hence, if $h_{\gamma_\infty}'''$ has the desired

3941: negativity property for large $R$, so will $h_{\gamma_\infty}$.

3942: 

3943: Let $\Sigma$ be the set of those points where two components of

3944: $X_\infty$ intersect each other and $X_0$. This lies in the base

3945: locus of our pencil, hence in every fibre $X_z$. Around each point $x

3946: \in \Sigma$, choose a neighbourhood $U_x \subset X$ which is an open

3947: ball around the origin in local coordinates as in Assumptions

3948: \ref{as:2}, and set $U = \bigcup_x U_x$. We know from our discussion

3949: of the local model that $U$ is invariant under parallel transport

3950: along large circles. The claim is that for sufficiently large $R$,

3951: the grading of

3952: \[

3953: \psigr_R = (\hgr_{\gamma_\infty}''')^2

3954: \]

3955: (with respect to $\o_X'''$, the given complex structure and complex

3956: volume form $\eta_{X_R}$) is negative on $X_R \setminus U$. This is

3957: proved by contradiction: suppose that we have a sequence $R_k

3958: \rightarrow \infty$, and points $x_k \in X_{R_k} \setminus U$ such

3959: that the grading of $\psigr_R$ is not negative on some Lagrangian

3960: subspace in the tangent space at $x_k$. Consider the limit point of a

3961: subsequence, $x_\infty \in X_\infty \setminus U$. If $x_\infty$ is a

3962: smooth point of $X_\infty$, the grading of $\psigr_R$ at $x_{z_k}$

3963: converges to $-2$ for the same reason as in the case of Lefschetz

3964: pencils, see Lemma \ref{th:shift-factor}. If $x_\infty$ is a point

3965: where several components of $X_\infty$ meet, Lemma

3966: \ref{th:local-model-1} says that the grading of $\psigr_R$ at

3967: $x_{z_k}$ will be bounded above by $-1/2$ for large $k$ (when

3968: applying this local result, the reader should keep in mind the change

3969: of coordinates on the base, $\zeta = 1/z$ and $dz = d\zeta/\zeta^2$).

3970: 

3971: Fix some $R$ such that the claim made above applies. Take a

3972: neighbourhood $W \subset U \cap X_R$ of $\Sigma$. After making $W$

3973: smaller, we may again assume that in the local coordinates of

3974: Assumption \ref{as:2}, this is the intersection of the fibre

3975: $Y_{1/R}$ with an open ball around the origin. Lemma

3976: \ref{th:bad-points} tells us that $h_{\gamma_\infty}'''(W) = W$, and

3977: that $h_{\gamma_\infty}'''$ is negative on $U \setminus W$. \qed

3978: 

3979: \section{Fukaya categories\label{sec:fukaya}}

3980: 

3981: There are Fukaya categories for affine and projective Calabi-Yaus, as

3982: well as a relative version which interpolates between the two. All of

3983: them are c-unital $A_\infty$-categories, but with different

3984: coefficient rings:

3985: \begin{center}

3986: \begin{tabular}{l| l | l |l}

3987:  & notation for & & \\

3988:  & category & notation for objects & coefficient ring \\

3989:  \hline

3990:  affine & $\Fuk(M)$ &

3991:  $L^\br = (L,\alphagr_L,\Spin_L) $ & $\C$ \\

3992:  relative & $\Fuk(X,X_\infty)$ &

3993:  $L^\mbr = (L,\alphagr_L,\Spin_L,J_L) $ & $\Lambda_\N$ \\

3994:  projective & $\Fuk(X)$ &

3995:  $L^\rbr= (L,\alphagr_L,\Spin_L,\lambda_L,J_L) $ & $\Lambda_\Q$ \\

3996: \end{tabular}

3997: \end{center}

3998: While our definition of affine Fukaya categories is fairly general,

3999: in the projective and relative setups only Calabi-Yau surfaces ($K3$

4000: and tori) will be considered. This restriction allows us to avoid

4001: using virtual fundamental chains, and it also simplifies the relation

4002: between the different categories: up to quasi-equivalence,

4003: $\Fuk(X,X_\infty)$ reduces to $\Fuk(M)$ if the deformation parameter

4004: $q$ is set to zero, while tensoring with $\Lambda_\Q$ gives a full

4005: subcategory of $\Fuk(X)$ (for the definition of $\Fuk(X)$ in higher

4006: dimensions, see \cite{fooo}; and for its presumed relation with

4007: $\Fuk(M)$, \cite{seidel02}). We emphasize that all properties of

4008: Fukaya categories which we use are quite simple ones, and largely

4009: independent of the finer details of the definition.

4010: 

4011: \subsection{}

4012: The following is a summary of \cite[Chapter 2]{seidel04}, with minor

4013: changes in notation and degree of generality. Let $M = X \setminus

4014: X_\infty$ be an affine Calabi-Yau. Objects of $\Fuk(M)$ are {\em

4015: exact Lagrangian branes}

4016: \begin{equation} \label{eq:exact-brane}

4017: L^\br = (L,\alphagr_L,\Spin_L),

4018: \end{equation}

4019: where $L \subset M$ is an oriented exact Lagrangian submanifold,

4020: $\alphagr_L$ a grading which is compatible with the orientation (see

4021: Remark \ref{th:or}), and $\Spin_L$ a $Spin$ structure. To define the

4022: morphism spaces, choose for each pair of objects a {\em Floer datum}

4023: \[

4024: (H,J) = (H_{L_0^\br,L_1^\br},J_{L_0^\br,L_1^\br}).

4025: \]

4026: This consists of a function $H \in \smooth_c([0;1] \times M,\R)$, as

4027: well as a family $J = J_t$ of $\o_X$-compatible almost complex

4028: structures parametrized by $t \in [0;1]$, which for all $t$ agree

4029: with the given complex structure in a neighbourhood of $X_\infty$.

4030: The condition on $H$ is that the associated Hamiltonian isotopy

4031: $(\phi^t_H)$ should satisfy $\phi^1_H(L_0) \trans L_1$. We then

4032: define $\CC(L_0^\br,L_1^\br)$ to be the set of trajectories $x: [0;1]

4033: \rightarrow M$, $x(t) = \phi^t_H(x(0))$, with boundary conditions

4034: $x(k) \in L_k$. This is obviously bijective to $\phi^1_H(L_0) \cap

4035: L_1$, hence finite. To each such $x$ one can associate a Maslov index

4036: $I(x)$ and orientation set $or(x)$. For $I(x)$, one observes that the

4037: grading of $L_0$ induces one of $\phi^1_H(L_0)$, and then uses the

4038: index of an intersection point as defined in

4039: \eqref{eq:absolute-index}. $or(x)$ is a set with two elements, the

4040: two possible ``coherent orientations'' of $x$; we refer to

4041: \cite{seidel04,fooo} for its definition, which uses the $Spin$

4042: structures on $L_0,L_1$. The Floer cochain space is the graded

4043: $\C$-vector space

4044: \begin{equation} \label{eq:basic-floer-complex}

4045: CF^j_M(L_0^\br,L_1^\br) = \bigoplus_{\substack{x \in

4046: \CC(L_0^\br,L_1^\br)

4047: \\ I(x) = j}} \C_x,

4048: \end{equation}

4049: where $\C_x$ is defined as the quotient of the two-dimensional space

4050: $\C[or(x)]$ by the relation that the two elements of $or(x)$ must sum

4051: to zero (it is of course isomorphic to $\C$, but not canonically). In

4052: the Fukaya category, $CF^*_M$ is the space of $hom$s from $L_0^\br$

4053: to $L_1^\br$.

4054: 

4055: Given $x_0,x_1 \in \CC(L_0^\br,L_1^\br)$, we now consider the moduli

4056: spaces $\MM^1_M(x_0,x_1)$ of solutions $u: \R \times [0;1]

4057: \rightarrow M$ to Floer's ``gradient flow'' equation with limits

4058: $x_0,x_1$, divided by the $\R$-action by translation. As part of the

4059: definition of Floer datum, we require that these spaces should all be

4060: regular. Assume from now on that $I(x_1) = I(x_0)-1$. By an index

4061: formula $\MM^1_M(x_0,x_1)$ is zero-dimensional, and a compactness

4062: theorem says that it is a finite set. A few words about this: since

4063: the action functional is exact \eqref{eq:energy} and there is no

4064: bubbling, we only have to worry about sequences of solutions which

4065: become increasingly close to $X_\infty$. To see that this does not

4066: happen, one uses a maximum principle argument based on the function

4067: $-\log\,||\sigma_{X,\infty}||^2$, which goes to $+\infty$ at

4068: $X_\infty$. It is plurisubharmonic since

4069: \begin{equation} \label{eq:log-sigma}

4070: \partial \bar\partial \log\,||\sigma_{X,\infty}||^2 = -F_{\nabla_X} =

4071: 2\pi i \o_M.

4072: \end{equation}

4073: Each $u \in \MM^1_M(x_0,x_1)$ comes with a preferred identification

4074: $or(x_1) \iso or(x_0)$, which in turn defines a map $\C_{x_1}

4075: \rightarrow \C_{x_0}$. The sum of these over all $u$ is the Floer

4076: differential $\mu^1_M: CF^j_M(L_0^\br,L_1^\br) \rightarrow

4077: CF^{j+1}_M(L_0^\br,L_1^\br)$, which is the first composition map in

4078: the Fukaya category. Its cohomology is the Floer cohomology

4079: $HF^*_M(L_0^\br,L_1^\br)$, which is therefore the space of morphisms

4080: in the cohomological category $H\Fuk(M)$.

4081: 

4082: To define the rest of the $A_\infty$-structure, one proceeds as

4083: follows. Take some $d \geq 2$. A {\em $(d+1)$-pointed disc with

4084: Lagrangian labels}

4085: \begin{equation} \label{eq:s}

4086:  (S,\zeta_0,\dots,\zeta_d,L_0^\br,\dots,L_d^\br)

4087: \end{equation}

4088: is a Riemann surface $S$ isomorphic to the closed disc with $d+1$

4089: boundary points removed. Moreover, these points at infinity should

4090: have a preferred numbering $\zeta_0,\dots,\zeta_d$, compatible with

4091: their natural cyclic order. Denote the connected components of

4092: $\partial S$ by $I_0,\dots,I_d$, again in cyclic order and starting

4093: with the component $I_0$ which lies between $\zeta_0$ and $\zeta_1$.

4094: Finally, we want to have exact Lagrangian branes

4095: $L_0^\br,\dots,L_d^\br$ attached to the boundary components. A set of

4096: {\em strip-like ends} for $S$ consists of proper holomorphic

4097: embeddings

4098: \begin{align*}

4099:  \epsilon_{S,0} & : \Rleq \times [0;1] \longrightarrow S, &&

4100:  \textstyle\lim_{s \rightarrow -\infty} \epsilon_{S,0}(s,\cdot) = \zeta_0, \\

4101:  \epsilon_{S,1},\dots,\epsilon_{S,d} & : \Rgeq \times [0;1] \longrightarrow S, &&

4102:  \textstyle\lim_{s \rightarrow +\infty} \epsilon_{S,k}(s,\cdot) = \zeta_k

4103: \end{align*}

4104: taking $\Rleq \times \{0;1\}$, $\Rgeq \times \{0;1\}$ to $\partial S$

4105: and which have disjoint images. A {\em perturbation datum} on $S$ is

4106: a pair $(K_S,J_S)$ consisting of a family $\{J_{S,z}\}_{z \in S}$ of

4107: compatible almost complex structures on $X$, which agree with the

4108: given complex structure in a neighbourhood of $X_\infty$, and a

4109: one-form $K_S \in \Omega^1(S,\smooth_c(M,\R))$ with values in smooth

4110: functions, subject to the condition that $K_S(\xi)|L_k = 0$ for $\xi

4111: \in T(I_k) \subset T(\partial S)$. The behaviour of $(K_S,J_S)$ over

4112: the strip-like ends is fixed by the previously chosen Floer data:

4113: \[

4114:  J_{S,\epsilon_{S,k}(s,t)} = \begin{cases}

4115:  J_{L_0^\br,L_d^\br,t} & k = 0, \\

4116:  J_{L_{k-1}^\br,L_k^\br,t} & k > 0

4117:  \end{cases}

4118:  \quad \text{and} \quad

4119:  \epsilon_{S,k}^*K_S = \begin{cases}

4120:  H_{L_0^\br,L_d^\br,t}\, dt & k = 0, \\

4121:  H_{L_{k-1}^\br,L_k^\br,t} \, dt & k > 0.

4122:  \end{cases}

4123: \]

4124: Suppose that we have made a global choice of strip-like ends and

4125: perturbation data. This means that for every $(d+1)$-pointed disc

4126: with Lagrangian labels we have chosen a set of strip-like ends

4127: $(K_S,J_S)$ (the dependence on the other data in \eqref{eq:s} is

4128: suppressed from the notation for the sake of brevity). This needs to

4129: satisfy two additional kinds of conditions: first of all, it needs to

4130: vary smoothly if we change the complex structure on $S$ in a smooth

4131: way, which means that it is defined on the universal family of such

4132: discs. Secondly, it needs to be well-behaved with respect to the

4133: compactification of that family, which includes degenerations to

4134: discs with nodes. This in fact establishes recursive relations

4135: between the choices of perturbation data for different $d$, see

4136: \cite{seidel04} for details.

4137: 

4138: Supposing that all these conditions have been met, take $x_0 \in

4139: \CC(L_0^\br,L_d^\br)$ as well as $x_k \in \CC(L_{k-1}^\br,L_k^\br)$

4140: for $k = 1,\dots,d$. Define $\MM^d_M(x_0,\dots,x_d)$ to be the space

4141: of equivalence classes of pairs consisting of a $(d+1)$-marked disc

4142: with labels $L_0^\br,\dots,L_d^\br$, and a solution of the following

4143: generalization of Floer's equation for a map $u: S \rightarrow M$:

4144: \begin{equation} \label{eq:generalized-floer}

4145: \begin{cases}

4146:  \!\!\! & u(I_k) \subset L_k, \\

4147:  \!\!\! & (du(z)-Y_{S,z,u(z)})

4148:  + J_{S,z,u(z)} \circ (du-Y_{S,z,u(z)}) \circ i = 0, \\

4149:  \!\!\! & \lim_{s \rightarrow \pm \infty} u(\epsilon_{S,k}(s,\cdot)) = x_k.

4150: \end{cases}

4151: \end{equation}

4152: Here $Y_S \in \Omega^1(S,\smooth_c(TM))$ is the

4153: Hamiltonian-vector-field valued one-form determined by $K_S$. The

4154: equivalence relation on pairs is isomorphism of the underlying

4155: Riemann surfaces, compatible with all the additional structure, and

4156: composition of maps $u$ with that isomorphism. The resulting quotient

4157: space is (locally) the zero-set of a Fredholm section of a suitably

4158: defined Banach vector bundle. Generically, this will be transverse,

4159: and then $\MM^d_M(x_0,\dots,x_d)$ is smooth of dimension $I(x_0) -

4160: I(x_1) - \dots - I(x_d) + d - 2$. Moreover, the zero-dimensional

4161: spaces are finite. It is maybe worth while stating the basic energy

4162: equality which underlies the compactness argument. Choose primitives

4163: $K_{L_k}$ for $\theta|L_k$. Define the action of the limits $x_k$ by

4164: a generalization of \eqref{eq:action},

4165: \begin{align*}

4166:  A(x_k) & = - \int x_k^*\theta + \\

4167:  + & \begin{cases}

4168:  \int H_{L_0^\br,L_d^\br}(x_0(t))\, dt +

4169:  K_{L_d}(x_0(1)) - K_{L_0}(x_0(0)) & k = 0, \\

4170:  \int H_{L_{k-1}^\br,L_k^\br}(x_k(t))\,dt +

4171:  K_{L_k}(x_k(1)) - K_{L_{k-1}}(x_k(0)) & k = 1,\dots,d.

4172:  \end{cases}

4173: \end{align*}

4174: Then for $u \in \MM^d_M(x_0,\dots,x_d)$,

4175: \begin{equation} \label{eq:polygon-energy}

4176: \begin{aligned}

4177:  E(u) & \stackrel{\mathrm{def}}{=} \half \int_S | du - Y |^2 \\ & =

4178:  A(x_0) - \sum_{k=1}^d A(x_d) - \int_S u^*(-dK_S + \half \{K_S,K_S\}).

4179: \end{aligned}

4180: \end{equation}

4181: The integral on the right hand side can be bounded by a constant

4182: which is independent of $u$, because the curvature term $-dK_S +

4183: \half \{K_S,K_S\}$ vanishes over the strip-like ends of $S$. Of

4184: course \eqref{eq:log-sigma} also plays a role in the proof of

4185: compactness, as was the case for the Floer differential. To continue

4186: with the main story: any element of a zero-dimensional space

4187: $\MM^d_M(x_0,\dots,x_d)$ defines a canonical bijection

4188: \begin{equation} \label{eq:orientation-mechanism}

4189: or(x_0) \iso or(x_1) \times_{\Z/2} \dots \times_{\Z/2} or(x_d).

4190: \end{equation}

4191: The order $d$ composition map $\mu^d_M$ in $\Fuk(M)$ is the sum of

4192: the corresponding maps $\C_{x_d} \otimes \dots \otimes \C_{x_1}

4193: \rightarrow \C_{x_0}$ (the order is reversed according to our

4194: notational conventions for $A_\infty$-categories). There are

4195: additional signs which involve the Maslov indices $I(x_k)$, but since

4196: they depend on the conventions about determinant lines used in

4197: setting up the orientation spaces, we prefer not to spell them out

4198: here. $\Fuk(M)$ is independent of all the choices (Floer data,

4199: strip-like ends, perturbation data) made in its construction up to

4200: quasi-equivalence, and indeed quasi-isomorphism acting trivially on

4201: the objects.

4202: 

4203: \begin{remark} \label{th:class-of-j}

4204: Suppose that no irreducible component of $X_\infty$ is rational. Then

4205: one can use a wider class of almost complex structures, namely all

4206: those $\o_X$-compatible ones such that each irreducible component $C

4207: \subset X_\infty$ is an almost complex submanifold. Instead of using

4208: \eqref{eq:log-sigma} to prevent solutions of Floer's equation and its

4209: generalization from escaping into $X_\infty$, one now argues as

4210: follows: if the solution stays inside $M$, its intersection number

4211: with $C$ is zero. The limit of a sequence of solutions has the same

4212: property, and nonnegativity of intersection multiplicities shows that

4213: the limit is disjoint from $X_\infty$ (more precisely, one thinks of

4214: a solution of \eqref{eq:generalized-floer} as a map $u: S \rightarrow

4215: S \times X$ which is pseudo-holomorphic for a suitable almost complex

4216: structure, and considers its intersection with $S \times X_\infty$).

4217: Similarly, in the inhomogeneous terms one can use Hamiltonian

4218: functions on $X$ whose values and first derivatives vanish on

4219: $X_\infty$.

4220: \end{remark}

4221: 

4222: \subsection{\label{subsec:coverings}}

4223: Let $M$ be an affine Calabi-Yau, equipped with a homomorphism

4224: \[

4225: \rho: \pi_1(M) \longrightarrow \Gamma

4226: \]

4227: into some finitely generated abelian group $\Gamma$. Consider the

4228: associated abelian covering $M_\Gamma \rightarrow M$. Denote by

4229: $\Fuk(M)_\rho$ the full subcategory of $\Fuk(M)$ consisting of those

4230: branes $L^\br$ such that the composition $\pi_1(L) \rightarrow

4231: \pi_1(M) \rightarrow \Gamma$ is trivial. For each of these choose a

4232: lift $L_\Gamma \subset M_\Gamma$. Once one has done that, the

4233: $hom$-spaces split as

4234: \[

4235:  CF^*_M(L_0^\br,L_1^\br) =

4236:  \bigoplus_{\gamma \in \Gamma} CF^*_M(L_0^\br,L_1^\br)_\gamma,

4237: \]

4238: where the $\gamma$-summand consists of those $x \in

4239: \CC(L_0^\br,L_1^\br)$ whose unique lift $x_\Gamma: [0;1] \rightarrow

4240: M_\Gamma$ with $x_\Gamma(0) \in L_{0,\Gamma}$ satisfies

4241: $\gamma^{-1}(x_\Gamma(1)) \in L_{1,\Gamma}$. Because each

4242: pseudo-holomorphic polygon can be lifted to $M_\Gamma$, these

4243: splittings are compatible with the compositions, in the sense that

4244: \[

4245:  CF^*_M(L_{d-1}^\br,L_d^\br)_{\gamma_d} \otimes \dots \otimes

4246:  CF^*_M(L_0^\br,L_1^\br)_{\gamma_1} \xrightarrow{\mu^d_M}

4247:  CF^*_M(L_0^\br,L_d^\br)_{\gamma_1\dots \gamma_d}.

4248: \]

4249: This constitutes a $\Gamma$-{\em coaction} on our

4250: $A_\infty$-category. For general nonsense reasons, it defines an {\em

4251: action} of the character group $\Gamma^* = Hom(\Gamma,\C^*)$:

4252: elements $\chi \in \Gamma^*$ act on each

4253: $CF^*_M(L_0^\br,L_1^\br)_\gamma$ by scalar multiplication with

4254: $\chi(\gamma) \in \C^*$.

4255: 

4256: What we actually need is a parallel but slightly different geometric

4257: setup. Let $M = X \setminus X_\infty$ be an affine Calabi-Yau, and

4258: $\Gamma$ a finite abelian group acting on $X$ by K{\"a}hler

4259: isometries. This should come with a lift of the action to $o_X$,

4260: which preserves $A_X$ and is compatible with the induced action on

4261: $\K_X$ via the isomorphism $\beta_X$ \eqref{eq:beta}. Finally, we

4262: require that the action preserves $\sigma_{X,\infty}$ up to

4263: multiplication with $|m_X|$-th roots of unity (for $m_X = 0$ these

4264: may be arbitrary constants in $S^1$), and that it is free on $M$. As

4265: a consequence of these assumptions, the one-form $\theta_M$ and the

4266: holomorphic volume form $\eta_M$ are $\Gamma$-invariant. Let $\bar{M}

4267: = M/\Gamma$ be the quotient, with its inherited symplectic form

4268: $\o_{\bar{M}}$, one-form $\theta_{\bar{M}}$ and holomorphic volume

4269: form $\eta_{\bar{M}}$. Note that $-\log ||\sigma_{X,\infty}||^2$ also

4270: descends to a proper subharmonic function on $\bar{M}$. One can

4271: therefore define the Fukaya category $\Fuk(\bar{M})$ in the same way

4272: as for $M$ itself, using almost complex structures which agree with

4273: the given complex structure outside a compact subset (this is in

4274: spite of the fact that the obvious compactification $\bar{X} =

4275: X/\Gamma$ may be singular). There is a tautological homomorphism

4276: $\rho: \pi_1(\bar{M}) \rightarrow \Gamma$, whose associated covering

4277: is $\bar{M}_\Gamma = M$. The observation made above applies to this

4278: case as well, so one has a $\Gamma^*$-action on $\Fuk(\bar{M})_\rho$.

4279: Take the semidirect product $A_\infty$-category, as defined in

4280: Section \ref{subsec:ainfty-actions}.

4281: 

4282: \begin{lemma} \label{th:semidirect-lift}

4283: $\Fuk(\bar{M})_\rho \semidirect \Gamma^*$ is quasi-isomorphic to the

4284: full subcategory of $\Fuk(M)$ consisting of those branes in $M$ which

4285: satisfy $\gamma(L) \cap L = \emptyset$ for all nontrivial $\gamma \in

4286: \Gamma$. \qed

4287: \end{lemma}

4288: 

4289: The proof is straightforward, and we leave it to the reader. Note

4290: that the $\Gamma^*$-action is not quite unique, since it depends on

4291: choosing lifts for the branes; but different choices lead to group

4292: actions related by (elementary) inner automorphisms as in

4293: \eqref{eq:conjugate-action}, which explains why the semidirect

4294: product is the same for all choices.

4295: 

4296: \subsection{}

4297: Let $X$ be a projective Calabi-Yau surface, and $L \subset X$ a

4298: Lagrangian submanifold which admits a grading. An $\o_X$-compatible

4299: almost complex structure $J$ is called {\em regular with respect to

4300: $L$} if there are no non-constant $J$-holomorphic spheres, and no

4301: non-constant $J$-holomorphic discs with boundary on $L$.

4302: 

4303: \begin{lemma}

4304: Almost complex structures which are regular with respect to a given

4305: $L$ are dense within the set of all $\o_X$-compatible almost complex

4306: structures.

4307: \end{lemma}

4308: 

4309: \proof Since $c_1(X) = 0$, the virtual dimension of the space of

4310: unparametrized $J$-holomorphic spheres at a point $v$ is $4 + 2

4311: \leftsc c_1(X), [v] \rightsc - 6 = -2$. Standard transversality

4312: results tell us that for generic $J$, there are no $J$-holomorphic

4313: spheres that are somewhere injective; and since any non-constant

4314: sphere is a multiple cover of a somewhere injective one, there are no

4315: non-constant spheres at all. The argument for discs goes along the

4316: same lines: because of the grading, the Maslov number $\mu(w)$ of any

4317: disc in $(X,L)$ vanishes, and the virtual dimension is

4318: \begin{equation} \label{eq:minus-one}

4319: 2 + \mu(w) - 3 = -1.

4320: \end{equation}

4321: Again, this shows that for generic $J$, there are no somewhere

4322: injective discs. The decomposition theorem of Kwon-Oh

4323: \cite{kwon-oh96} and Lazzarini \cite{lazzarini98} says that if there

4324: is a non-constant $J$-holomorphic disc, there is also a somewhere

4325: injective one, which implies the desired result. \qed

4326: 

4327: Objects of $\Fuk(X)$ are {\em rational Lagrangian branes}

4328: \begin{equation} \label{eq:rational-brane}

4329: L^\rbr = (L,\alphagr_L,\Spin_L,\lambda_L,J_L)

4330: \end{equation}

4331: where $L$ is an oriented Lagrangian submanifold of $X$ with a grading

4332: $\alphagr_L$ and $Spin$ structure $\Spin_L$; $\lambda_L$ is a

4333: covariantly constant multisection of the circle bundle of $o_X|L$ of

4334: some degree $d_L \geq 1$; and $J_L$ is an $\o_X$-compatible almost

4335: complex structure which is regular with respect to $L$. For any pair

4336: of objects, we take a Floer datum

4337: \[

4338: (H,J) = (H_{L_0^\rbr,L_1^\rbr},J_{L_0^\rbr,L_1^\rbr})

4339: \]

4340: as before, except that the conditions on the behaviour of $H$ and $J$

4341: near $X_\infty$ are omitted, and the following new requirements

4342: imposed instead: $J_t$ must be equal to $J_{L_0}$, $J_{L_1}$ for $t =

4343: 0,1$; and there should be no non-constant $J_t$-holomorphic spheres

4344: for any $t$ (this is still true generically, because the virtual

4345: dimension of the parametrized moduli space is $-1$). Define the set

4346: $\CC(L_0^\rbr,L_1^\rbr)$ as before; and for any $x:[0;1] \rightarrow

4347: X$ in that set, write $\bar{A}(x) = \bar{A}(x(1))$ for the mod $\Q$

4348: action \eqref{eq:modq-action} of $x(1)$ as an intersection point of

4349: $\phi^1_H(L_0) \cap L_1$ (where we have used \eqref{eq:deform-lambda}

4350: to get a multisection of $o_X|\phi^1_H(L_0)$ from $\lambda_{L_0}$).

4351: To spell this out, we have

4352: \[

4353: \exp(2\pi i \bar{A}(x)) =

4354: \frac{\lambda_{L_1}(x(1))}{\kappa(\lambda_{L_0}(x(0)))} \exp\Big( 2

4355: \pi i \int_0^1 H_t(x(t))\, dt\Big),

4356: \]

4357: where $\kappa$ is parallel transport on $o_X$ along $x$. For any

4358: number $r \in \R/\Q$, we define a one-dimensional $\Lambda_\Q$-vector

4359: space $q^r\Lambda_{\Q}$ whose elements are formal series $f(q) =

4360: \sum_m a_m q^m$, with $m$ running over all real numbers in $r +

4361: \frac{1}{d}\Z$ for some natural number $d$ which depends on $f$, and

4362: $a_m \in \C$ vanishing for sufficiently negative $m$. Write

4363: $\Lambda_{\Q,x}$ for the quotient of

4364: $q^{\bar{A}(x)}\Lambda_{\Q}[or(x)]$ by the relation that the two

4365: elements of $or(x)$ sum to zero. Then the Floer cochain space is

4366: \[

4367: CF^j_X(L_0^\rbr,L_1^\rbr) = \bigoplus_{\substack{x \in

4368: \CC(L_0^\rbr,L_1^\rbr)

4369: \\ I(x) = j}} \Lambda_{\Q,x}.

4370: \]

4371: Denote by $\MM^1_X(x_0,x_1)$ the space of solutions of Floer's

4372: equation in $X$ with limits $x_0,x_1$ (this means we drop the

4373: previous assumption that the solutions stay in $M$). As before, we

4374: require that this is regular, and in particular zero-dimensional in

4375: the case $I(x_1) = I(x_0)-1$. Take a solution $u$ in such a

4376: zero-dimensional moduli space, and let $E(u) = \int |\partial

4377: u/\partial s|^2 \, ds\, dt

4378: > 0$ be its energy. A version of \eqref{eq:modq-energy} shows that

4379: $E(u) \in \bar{A}(x_0) - \bar{A}(x_1) + \lcm(d_{L_0},d_{L_1})^{-!}

4380: \Z$, so that we get a well-defined $\Lambda_\Q$-module map

4381: \[

4382: \pm q^{E(u)} : \Lambda_{\Q,x_1} \longrightarrow \Lambda_{\Q,x_0}.

4383: \]

4384: The sign of this is determined by the induced bijection of

4385: orientation sets, as before, and summing over all $u$ yields the

4386: Floer differential $\mu^1_X$ in $\Fuk(X)$. The sum is no longer

4387: finite, but it gives a well-defined $\Lambda_\Q$-module map by Gromov

4388: compactness; the basic point being that there is no bubbling off of

4389: holomorphic discs or spheres, by assumption on the Floer datum.

4390: Denote the resulting Floer cohomology by $HF^*_X(L_0^\rbr,L_1^\rbr)$.

4391: 

4392: The definition of the higher order compositions is largely similar.

4393: One may use only perturbation data $(K_S,J_S)$ where $J_S = J_{L_k}$

4394: over the boundary component $I_k \subset \partial S$. It is no longer

4395: possible to avoid $J_{S,z}$-holomorphic spheres for some values of $z

4396: \in int(S)$, but the argument from \cite{hofer-salamon95} shows that

4397: for a generic choice of $J_S$, these spheres will never occur as

4398: bubbles in the Gromov compactification of the spaces

4399: $\MM^d_X(x_0,\dots,x_d)$ as long as these are of dimension $\leq 1$,

4400: which is all that we will ever use. Equation

4401: \eqref{eq:polygon-energy} still holds if we replace all the actions

4402: by their mod $\Q$ counterparts. We define the contribution of a point

4403: $(S,u)$ in a zero-dimensional moduli space $\MM^d_X(x_0,\dots,x_d)$

4404: to $\mu^d_X$ to be

4405: \begin{equation} \label{eq:qe}

4406:  \pm q^{E(u) + \int_S u^*(-dK_S + \half\{K_S,K_S\})} :

4407:  \Lambda_{\Q,x_d} \otimes \dots \otimes \Lambda_{\Q,x_1}

4408:  \longrightarrow \Lambda_{\Q,x_0},

4409: \end{equation}

4410: with the sign again depending on \eqref{eq:orientation-mechanism}.

4411: There is a uniform lower bound for the $\int_S$ term. This, together

4412: with Gromov compactness, ensures that the sum is a well-defined

4413: $\Lambda_\Q$-multilinear map.

4414: 

4415: \begin{remark}

4416: The choice of $\lambda_L$ serves only to fix the action: the Floer

4417: cohomology groups for different choices are isomorphic in an obvious

4418: way, by multiplication with $q^m$ for some $m \in \Q$. In particular,

4419: changing $\lambda_L$ does not affect the quasi-isomorphism class of

4420: the resulting object in $\Fuk(X)$.

4421: \end{remark}

4422: 

4423: \subsection{}

4424: Take $X$ as before, but now equipped with a section

4425: $\sigma_{X,\infty}$ of $o_X$ whose zero set $X_\infty$ is a reduced

4426: divisor with normal crossings, so that $M = X \setminus X_\infty$ is

4427: an affine Calabi-Yau surface. Following Remark \ref{th:class-of-j},

4428: we also assume that $X_\infty$ has no rational components. The

4429: objects of the relative Fukaya category $\Fuk(X,X_\infty)$ are a

4430: mixture of the two previously introduced classes,

4431: \eqref{eq:exact-brane} and \eqref{eq:rational-brane}. One considers

4432: {\em relative Lagrangian branes}

4433: \begin{equation} \label{eq:relative-brane}

4434: L^\mbr = (L,\alphagr_L,\Spin_L,J_L)

4435: \end{equation}

4436: where $(L,\alphagr_L,\Spin_L)$ is an exact Lagrangian brane in $M$;

4437: and $J_L$ is an $\o_X$-compa\-tible almost complex structure on $X$

4438: such that each component of $X_\infty$ is an almost complex

4439: submanifold, and which is regular with respect $L$. The restriction

4440: on $J_L|X_\infty$ does not interfere with the regularity theory for

4441: somewhere injective pseudo-holo\-morphic spheres $v: S^2 \rightarrow

4442: X$, since $v^{-1}(X_\infty)$ is necessarily a finite set, while the

4443: set of somewhere injective points is open; and the same applies to

4444: pseudo-holomorphic discs. For any two objects, choose a Floer datum

4445: \[

4446: (H,J) = (H_{L_0^\mbr,L_1^\mbr},J_{L_0^\mbr,L_1^\mbr})

4447: \]

4448: such that $H \in \smooth([0;1] \times X,\R)$ vanishes along $[0;1]

4449: \times X_\infty$, along with its first derivative; and $J_t|X_\infty$

4450: makes the components of $X_\infty$ almost complex. In addition, we

4451: assume that $J_k = J_{L_k}$ for $k = 0,1$, and that there are no

4452: non-constant $J_t$-holomorphic spheres for any $t$.

4453: 

4454: We have the usual set $\CC(L_0^\mbr,L_1^\mbr)$ of $H$-trajectories

4455: joining $L_0$ to $L_1$, and for $x$ in that set we define

4456: $\Lambda_{\N,x}$ in the same way as $\C_x$, just using $\Lambda_\N$

4457: instead of $\C$ as the ground ring. The Floer cochain space in

4458: $\Fuk(X,X_\infty)$ is

4459: \[

4460:  CF_{X,X_\infty}^j(L_0^\mbr,L_1^\mbr) =

4461:  \bigoplus_{\substack{x \in \CC(L_0^\mbr,L_1^\mbr)

4462:  \\ I(x) = j}} \Lambda_{\N,x}.

4463: \]

4464: For the Floer differential, each $u \in \MM^1_X(x_0,x_1)$ contributes

4465: with $\pm q^{u \cdot X_\infty}: \Lambda_{\N,x_1} \rightarrow

4466: \Lambda_{\N,x_0}$. This makes sense for a single $u$ since the number

4467: $u \cdot X_\infty$ is nonnegative; and Gromov compactness, together

4468: with a version of \eqref{eq:energy}, says that the possibly infinite

4469: sum over all $u$ gives rise to a well-defined map

4470: $\mu^1_{X,X_\infty}$. Denote the resulting Floer cohomology by

4471: $HF^*_{X,X_\infty}(L_0^\mbr,L_1^\mbr)$. The definition of the higher

4472: order compositions works in the same way, and we will therefore omit

4473: the details.

4474: 

4475: \begin{remark} \label{th:class-of-j-2}

4476: Let $X$, $X'$ be two Calabi-Yau surfaces, with sections

4477: $\sigma_{X,\infty}$, $\sigma_{X',\infty}$ of their ample line bundles

4478: $o_X$, $o_{X'}$ which define reduced divisors with normal crossings,

4479: having no rational components. Suppose that there is a symplectic

4480: isomorphism

4481: \[

4482: \phi: (X,X_\infty) \longrightarrow (X',X'_\infty)

4483: \]

4484: with the following additional properties: $Hom(o_X,\phi^*o_{X'})$ is

4485: the trivial flat line bundle, and over $M = X \setminus X_\infty$,

4486: its covariantly constant sections lie in the same homotopy class of

4487: nowhere zero sections as

4488: $\phi^*\sigma_{X',\infty}/\sigma_{X,\infty}$. Moreover, under the

4489: isomorphism (unique up to homotopy) $\K_X \iso \phi^*\K_{X'}$ induced

4490: by deforming the almost complex structure on $X$ to the pullback of

4491: the one from $X'$, the nowhere zero sections $\eta_X$ and

4492: $\phi^*\eta_{X'}$ are homotopic. Then, by suitably choosing the Floer

4493: and perturbation data, one can define an induced quasi-isomorphism of

4494: $A_\infty$-categories over $\Lambda_\N$,

4495: \[

4496: \phi_*: \Fuk(X,X_\infty) \rightarrow \Fuk(X',X'_\infty).

4497: \]

4498: In the case where $H_1(X) = 0$, all the additional conditions imposed

4499: on $\phi$ are automatically satisfied, for the same reasons as in

4500: Remark \ref{th:h1}.

4501: \end{remark}

4502: 

4503: \subsection{}

4504: Finally, we need to establish the relation between the different

4505: kinds of Fukaya categories. Let $X$ be a Calabi-Yau surface with a

4506: suitable section $\sigma_{X,\infty}$, so that $\Fuk(M)$, $\Fuk(X)$

4507: and $\Fuk(X,X_\infty)$ are all defined.

4508: 

4509: \begin{prop} \label{th:q-zero}

4510: Consider $\C$ as a $\Lambda_\N$-module in the obvious way, with $q$

4511: acting trivially. Then the $\C$-linear $A_\infty$-category obtained

4512: from $\Fuk(X,X_\infty)$ by reduction of constants, $\Fuk(X,X_\infty)

4513: \otimes_{\Lambda_\N} \C$, is quasi-equivalent to $\Fuk(M)$.

4514: \end{prop}

4515: 

4516: \begin{prop} \label{th:invert-q}

4517: Consider $\Lambda_\Q \supset \Lambda_\N$ as a $\Lambda_\N$-module in

4518: the obvious way. Then the $\Lambda_\Q$-linear category obtained from

4519: $\Fuk(X,X_\infty)$ by extension of constants, $\Fuk(X,X_\infty)

4520: \otimes_{\Lambda_\N} \Lambda_\Q$, is quasi-isomorphic to a full

4521: $A_\infty$-subcategory of $\Fuk(X)$. This subcategory consists of all

4522: those $L^\rbr$ such that $L \cap X_\infty = \emptyset$, $o_X|L$ is

4523: the trivial flat bundle, and its covariantly constant sections lie in

4524: the same homotopy class as $\sigma_{X,\infty}|L$.

4525: \end{prop}

4526: 

4527: Both statements are essentially obvious. For Proposition

4528: \ref{th:q-zero}, one notices that setting $q = 0$ means that we count

4529: only those solutions of Floer's equation or its generalization which

4530: have zero intersection number with, hence are disjoint from,

4531: $X_\infty$. Then, the only difference between $\Fuk(M)$ and

4532: $\Fuk(X,X_\infty) \otimes_{\Lambda_\N} \C$ is that the latter

4533: category has more objects, consisting of the same exact Lagrangian

4534: brane with different choices of $J_L$. However, all these objects are

4535: isomorphic, whence the quasi-equivalence. For Proposition

4536: \ref{th:invert-q} one has to choose a function $K_L$, $dK_L =

4537: \theta_M|L$, for each exact Lagrangian brane in $M$. The associated

4538: object in $\Fuk(X)$ then carries the multisection $\lambda_L$ from

4539: \eqref{eq:k-lambda}, and one identifies

4540: \begin{equation} \label{eq:mq}

4541:  CF^*_{(X,X_\infty)}(L_0^\mbr,L_1^\mbr) \otimes_{\Lambda_\N}

4542:  \Lambda_\Q \stackrel{\iso}{\longrightarrow}

4543:  CF^*_X(L_0^\rbr,L_1^\rbr)

4544: \end{equation}

4545: by using the map $q^{A(x)}: \Lambda_{\N,x} \otimes_{\Lambda_\N}

4546: \Lambda_{\Q} \rightarrow \Lambda_{\Q,x}$ on each one-dimensional

4547: subspace. The compositions in the two categories are based on the

4548: same moduli spaces, and the fact that the weights used to count

4549: solutions correspond under \eqref{eq:mq} is a consequence of

4550: \eqref{eq:energy} and a suitable generalization. Clearly,

4551: $\Fuk(X,X_\infty) \otimes_{\Lambda_\N} \Lambda_\Q$ is not all of

4552: $\Fuk(X)$, since only the objects which are exact Lagrangian

4553: submanifolds in $M$ occur.

4554: 

4555: \subsection{}

4556: While the basic properties of affine Fukaya categories are familiar

4557: and quite simple, see for instance \cite{seidel04}, the relative and

4558: projective versions reserve some small surprises. These are all

4559: covered by the powerful general theory from \cite{fooo}, but we

4560: prefer a more elementary approach. The relative situation is the more

4561: important one for the arguments later on, and we will consider it

4562: primarily.

4563: 

4564: Fix an affine Calabi-Yau surface $M = X \setminus X_\infty$, without

4565: rational components in $X_\infty$. Take an exact Lagrangian brane

4566: $(L,\alphagr_L,\Spin_L)$ in $M$, and a one-parameter family $J_t$ ($0

4567: \leq t \leq 1$) of compatible almost complex structures, all of which

4568: make the components of $X_\infty$ almost complex, and are regular

4569: with respect to $L$. These give rise to a family of objects of

4570: $\Fuk(X,X_\infty)$, which we denote by $L_t^\mbr$. By suitably

4571: adapting the construction of Piunikhin-Salamon-Schwarz

4572: \cite{piunikhin-salamon-schwarz94}, one defines a canonical

4573: isomorphism

4574: \begin{equation} \label{eq:pss}

4575: H\Psi_{X,X_\infty} : H^*(L;\C) \otimes_\C \Lambda_\N

4576: \stackrel{\iso}{\longrightarrow}

4577: HF^*_{X,X_\infty}(L_0^\mbr,L_1^\mbr).

4578: \end{equation}

4579: In the special case where $J_t$ is constant, one can use this to

4580: obtain a distinguished element $e = H\Psi_{X,X_\infty}(1) \in

4581: HF^0_{X,X_\infty}(L_0^\mbr,L_0^\mbr)$, which is in fact the

4582: cohomological unit for $L_0^\mbr$ in $H\Fuk(X,X_\infty)$. The more

4583: general isomorphisms \eqref{eq:pss}, for different choices of almost

4584: complex structures, carry the ordinary cup product to the pair of

4585: pants product. This eventually shows that $L_0^\mbr$, $L_1^\mbr$ are

4586: quasi-isomorphic objects. A slight generalization, which in fact is

4587: already implicit in our use of Floer data, then proves the following

4588: 

4589: \begin{lemma} \label{th:move-object}

4590: Let $L_t$ be a family of Lagrangian submanifolds of $M$, and

4591: $L_t^\mbr$ a family of relative brane structures on them, varying

4592: smoothly with $t$. Then all $L_t^\mbr$ are quasi-isomorphic objects

4593: of $\Fuk(X,X_\infty)$. \qed

4594: \end{lemma}

4595: 

4596: Note that because all the $L_t$ are supposed to have relative brane

4597: structures, they are in particular exact in $M$, so that the isotopy

4598: is Hamiltonian.

4599: 

4600: \subsection{}

4601: We will now summarize the definition of $H\Psi_{X,X_\infty}$. Pick a

4602: cohomology class $c \in H^j(L;\Z)$, and represent its Poincar{\'e}

4603: dual by a map $\gamma: C \rightarrow L$ from a closed oriented

4604: manifold $C$ (recall that $L$ is a surface). Take the Riemann surface

4605: $S = \mathbb{H}$, the closed upper half plane, and consider a

4606: generalized Floer equation for pairs $(u,y)$ consisting of a map $u:

4607: S \rightarrow M$ and a point $y \in C$,

4608: \begin{equation} \label{eq:pss-equation}

4609: \begin{cases}

4610:  \!\!\! & u(\partial S) \subset L, \\

4611:  \!\!\! & u(0) = \gamma(c), \\

4612:  \!\!\! & (du - Y_{S,u(z),z}) + J_{S,z,u(z)} \circ (du - Y_{S,u(z),z} \circ i) = 0, \\

4613:  \!\!\! & \lim_{s \rightarrow +\infty} u(e^{\pi(s+it)}) = x_0(1-t),

4614:  \quad x_0 \in \CC(L_0^\mbr,L_1^\mbr).

4615: \end{cases}

4616: \end{equation}

4617: The choice of almost complex structures $J_S$ and inhomogeneous term

4618: $Y_S$ has to obey a whole series of conditions, of which we mention

4619: only the most relevant one: along the boundary $\partial S = \R$,

4620: $J_{S,z} = J_{\psi(z)}$ for some function $\psi: \R \rightarrow

4621: [0;1]$ with $\psi(s) = 0$ for very negative $s$; $\psi(s) = 1$ for

4622: very positive $s$; and $\psi'(s) > 0$ for all $s$ such that $\psi(s)

4623: \in (0;1)$.

4624: 

4625: Assume that the moduli spaces $\MM^{\Psi}_{X,X_\infty}(x_0;\gamma)$

4626: of solutions of \eqref{eq:pss-equation} are all regular. If $I(x_0) =

4627: j$, the relevant moduli space is zero-dimensional, and every $(u,y)$

4628: in it comes with a canonical choice of point in $or(x_0)$. This will

4629: define an element $\pm q^{u \cdot X_\infty} \in \Lambda_{\N,x_0}$,

4630: and the sum of all these is denoted by

4631: \[

4632: \Psi_{X,X_\infty}(\gamma) \in CF^j_{X,X_\infty}(L_0^\mbr,L_1^\mbr).

4633: \]

4634: The proof that this is a Floer cocycle relies on a standard gluing

4635: argument: one looks at the (one-dimensional) moduli spaces

4636: $\MM^{\Psi}_{X,X_\infty}(x_0;\gamma)$ with $I(x_0) = j+1$, and proves

4637: that points in the boundary of their Gromov compactification are

4638: pairs consisting of an isolated solution which contributes to

4639: $\Psi_{X,X_\infty}(\gamma)$ and a Floer gradient trajectory. The main

4640: point is that due to the choice of $J_S$, there is no bubbling off of

4641: pseudo-holomorphic discs. By definition, the Floer cohomology class

4642: of $\Psi_{X,X_\infty}(\gamma)$ is $H\Psi_{X,X_\infty}(c \otimes 1)$.

4643: 

4644: \subsection{}

4645: The assumption made up to now, that all the $J_t$ are regular, is not

4646: generic: while a single almost complex structure will typically admit

4647: no nonconstant pseudo-holomorphic discs with boundary on $L$, these

4648: can no longer be avoided in one-parameter families, as one can see

4649: from \eqref{eq:minus-one}. In other words, the space of all almost

4650: complex structures is divided into chambers by codimension one walls

4651: consisting of non-regular ones. Because there is an infinite number

4652: of possible homotopy classes of discs, there is potentially an

4653: infinite number of walls, which might well be everywhere dense. We

4654: will look at the simplest instance of the behaviour of Floer

4655: cohomology under wall-crossing (a complete treatment is beyond our

4656: scope here, since it requires techniques that can deal with the lack

4657: of transversality for multiply-covered discs, see \cite{fooo}).

4658: 

4659: \begin{assumptions} \label{as:wall-crossing}

4660: Fix an exact Lagrangian brane $(L,\alphagr_L,\Spin_L)$ in $M$, and a

4661: family $(J_t)$ of compatible almost complex structures, each of which

4662: makes the components of $X_\infty$ almost complex. Suppose that this

4663: family has the following properties:

4664: \begin{itemize}

4665: \item

4666: $J_0,J_1$ are regular with respect to $L$;

4667: \item

4668: for all $t$, there are no $J_t$-holomorphic spheres $v: S^2

4669: \rightarrow X$ such that $v \cdot X_\infty = 1$;

4670: \item

4671: there is a $t_0 \in (0;1)$ such that for all $t \neq t_0$, no

4672: $J_t$-holomorphic discs $w: (D,\partial D) \rightarrow (X,L)$ with $w

4673: \cdot X_\infty = 1$ exist. For $J_{t_0}$, there is exactly one such

4674: disc, which we denote by $w_0$. This is parametrized-regular, and

4675: $[w_0|\partial D] \in H_1(L)$ is nonzero.

4676: \end{itemize}

4677: \end{assumptions}

4678: 

4679: The terminology in the last sentence may require some explanation.

4680: Denote by $D\bar\partial_{w,X}: \Gamma(w^*TX,w^*TL) \rightarrow

4681: \Omega^{0,1}(w^*TX)$ the standard linearized operator associated to a

4682: $J_t$-holomorphic disc $w: (D,\partial D) \rightarrow (X,L)$. Here

4683: $\Gamma(E,F)$ means sections of the vector bundle $E \rightarrow D$

4684: with boundary values in the totally real subbundle $F \subset

4685: E|\partial D$. The parametrized version,

4686: \[

4687: D\bar\partial_{w,X}^{par}: \R \times \Gamma(w^*TX,w^*TL)

4688: \longrightarrow \Omega^{0,1}(w^*TX),

4689: \]

4690: adds a first component $1 \mapsto \half (\partial J_t/\partial t)

4691: \circ dw \circ i$. If this is onto in suitable Sobolev completions,

4692: we say that $w$ is parametrized-regular.

4693: 

4694: Consider the equation \eqref{eq:pss-equation} as before, choosing

4695: $\psi$ in such a way that $\psi(0) = t_0$. For solutions $(u,y) \in

4696: \MM^{\Psi}_{X,X_\infty}(x_0;\gamma)$ with $I(x_0) = j$, which occur

4697: in zero-dimensional moduli spaces, one can achieve by a suitable

4698: generic choice of $(K_S,J_S)$ that $u(z)$ does not meet the boundary

4699: of $J_{\psi(z)}$-holomorphic discs in $(X,L)$, for any $z \in \R$.

4700: Hence one can define Floer cochains $\Psi_{X,X_\infty}(\gamma)$ as

4701: before. Expand these and the Floer differential on

4702: $CF^*_{X,X_\infty}(L_0^\mbr,L_1^\mbr)$ into powers of $q$,

4703: \[

4704: \begin{aligned}

4705:  & \Psi_{X,X_\infty}(\gamma)

4706:  = \Psi_0(\gamma) + \Psi_1(\gamma) q + \Psi_2(\gamma) q^2 + \dots, \\

4707:  & \mu^1_{X,X_\infty} = \mu_0 + \mu_1 q + \mu_2 q^2 + \dots

4708: \end{aligned}

4709: \]

4710: 

4711: \begin{claim}

4712: $\mu_0 \Psi_0(\gamma) = 0$; indeed, $\Psi_0$ is an isomorphism

4713: between $H^*(L;\C) \otimes \Lambda_{\N}$ and the cohomology of

4714: $CF^*_{X,X_\infty}(L_0^\mbr,L_1^\mbr)$ with only $\mu_0$ as

4715: differential.

4716: \end{claim}

4717: 

4718: This is simply the statement of the Piunikhin-Salamon-Schwarz

4719: isomorphism for $L$ as an exact Lagrangian submanifold in the affine

4720: Calabi-Yau $M$: by Proposition \ref{th:q-zero}, $\mu_0 = \mu^1_M$ is

4721: the differential in $\Fuk(M)$, and similarly $\Psi_0$ counts

4722: solutions of \eqref{eq:pss-equation} which stay inside $M$.

4723: 

4724: \begin{claim}

4725: Suppose that $(C,\gamma) = (L,\id)$ is the whole surface. Then

4726: \[

4727: \mu_1 \Psi_0(\gamma) + \mu_0 \Psi_1(\gamma) = \pm \Psi_0(w_0|\partial

4728: D).

4729: \]

4730: \end{claim}

4731: 

4732: The $\Psi_0(w_0|\partial D)$ term comes from a new stratum in the

4733: compactification of the one-dimensional moduli spaces, which consists

4734: of solutions of \eqref{eq:pss-equation} with the disc bubble $w_0$

4735: attached to it at the point $z = 0$. Under the regularity condition

4736: on $w_0$ that we have imposed, these are smooth boundary points of

4737: the compactification.

4738: 

4739: A straightforward computation starting from the two claims above

4740: shows that

4741: \[

4742: HF^0_{X,X_\infty}(L_0^\mbr,L_1^\mbr) = 0.

4743: \]

4744: In particular $L_0^\mbr$, $L_1^\mbr$ are not quasi-isomorphic as

4745: objects of $\Fuk(X,X_\infty)$. Moreover, if we cut off the $q$-series

4746: after the first order, then by the same kind of computation as before

4747: the natural map

4748: \begin{equation} \label{eq:orders}

4749:  H^0(CF^*_{X,X_\infty}(L_0^\mbr,L_1^\mbr)

4750:  \otimes_{\Lambda_\N} \C[q]/q^2) \otimes_{\C[q]/q^2} \C

4751:  \longrightarrow H^0(CF^*_M(L_0^\br,L_1^\br))

4752: \end{equation}

4753: vanishes. More precisely, both sides are isomorphic to $\C$, but the

4754: left hand side is represented by the cocycle $q\Psi_0([L])$, which

4755: maps to zero. As a consequence,

4756: 

4757: \begin{cor} \label{th:wall-crossing}

4758: Suppose that there is some $(L,\alphagr_L,\Spin_L)$ such that

4759: Assumptions \ref{as:wall-crossing} hold. Then $\Fuk(X,X_\infty)$,

4760: seen as a one-parameter deformation of $A_\infty$-categories, is

4761: nontrivial at first order in $q$. \qed

4762: \end{cor}

4763: 

4764: \subsection{}

4765: Our assumption that $[w_0|\partial D] \in H_1(L)$ is nontrivial

4766: obviously excludes Lagrangian spheres, and indeed wall-crossing in

4767: trivial in that case:

4768: 

4769: \begin{lemma} \label{th:spheres}

4770: Up to quasi-isomorphism, a graded Lagrangian sphere $\Lgr$ in $M$

4771: gives rise to a unique object of $\Fuk(X,X_\infty)$.

4772: \end{lemma}

4773: 

4774: \proof Take two objects $L_0^\mbr,L_1^\mbr$ obtained by various

4775: choices of relative brane structures on $L$ (keeping the same

4776: grading). Clearly, these two objects are quasi-isomorphic in

4777: $\Fuk(X,X_\infty) \otimes_{\Lambda_\N} \C \iso \Fuk(M)$. Because

4778: there is no odd-dimensional cohomology, the quasi-isomorphism between

4779: them survives to an element of

4780: $HF^0_{X,X_\infty}(L_0^\mbr,L_1^\mbr)$, and an obvious spectral

4781: sequence argument shows that this is a quasi-isomorphism in

4782: $\Fuk(X,X_\infty)$. \qed

4783: 

4784: \begin{remark}

4785: Essentially all of our observations concerning relative Fukaya

4786: categories carry over to the projective case. First of all, we have

4787: an analogue of Lemma \ref{th:move-object}, saying that {\em the

4788: isomorphism class of an object of $\Fuk(X)$ remains constant under

4789: smooth deformation}. Secondly, {\em crossing a wall on which a single

4790: lowest area holo\-morphic disc $w_0$ sits does change the isomorphism

4791: class of an object}, assuming as before that $w_0$ is

4792: parameterized-regular and $[w_0|\partial D]$ is nontrivial. Finally,

4793: for graded Lagrangian spheres in $X$ we have the same statement as in

4794: Lemma \ref{th:spheres}. The proofs are quite similar, except that the

4795: filtration by intersection numbers with $X_\infty$ has to be replaced

4796: by a filtration by values of the action functional. This follows

4797: methods that were pioneered by Oh \cite{oh96} and vastly developed in

4798: \cite{fooo}. In our case, the rationality assumption simplifies the

4799: technical details somewhat, since it makes the filtration discrete.

4800: \end{remark}

4801: 

4802: \subsection{}

4803: As shown by Corollary \ref{th:wall-crossing}, one can use

4804: wall-crossing phenomena in the relative Fukaya category to prove its

4805: nontriviality in the sense of deformation theory; provided of course

4806: that one can come up with a Lagrangian submanifold for which

4807: wall-crossing actually occurs. To make that easier, we now suppose

4808: that the Calabi-Yau surface under discussion is the fibre of a

4809: quasi-Lefschetz pencil, so that the desired Lagrangian submanifold

4810: can be first constructed in the degenerate fibre at $\infty$.

4811: 

4812: For this discussion, we change notation as follows: $X$ is a Fano

4813: threefold with $o_X = \K_X^{-1}$, carrying a pair of sections

4814: $\sigma_{X,0},\sigma_{X,\infty}$ which generate a quasi-Lefschetz

4815: pencil of Calabi-Yau surfaces $\{X_z\}$. To be able to define the

4816: relative Fukaya categories, we assume that $X_{0,\infty}$ has no

4817: rational components.

4818: 

4819: Fix some irreducible component $C$ of $X_\infty$, and denote by $C'$

4820: the union of all the other components. There is an obvious short

4821: exact sequence

4822: \begin{equation} \label{eq:tangent-sequence}

4823:  0 \longrightarrow TC \longrightarrow

4824:  TX|C \xrightarrow{D(\sigma_{X,\infty}/\sigma_{X,0})}

4825:  \O_C(C \cap X_0 - C \cap C') \longrightarrow 0.

4826: \end{equation}

4827: From the resulting isomorphism $\K_C \iso \K_X|C \otimes

4828: \K_X^{-1}(-C')|C \iso \O_C(-C' \cap C)$ one obtains a meromorphic

4829: two-form $\eta_C$ on $C$ with simple poles along $C \cap C'$.

4830: Similarly, from $A_X|C$ and $\sigma_{X,0}|C$ one obtains a one-form

4831: $\theta_C$ on $C \setminus X_0$ with $d\theta_C = \o_C = \o_X|C$.

4832: 

4833: \begin{assumptions} \label{as:l-zero}

4834: Suppose that we have the following data $L_0$, $w_0$:

4835: \begin{itemize}

4836: \item

4837: $L_0 \subset C$ is a Lagrangian submanifold, disjoint from $C' \cup

4838: X_0$, exact with respect to $\theta_C$, and graded with respect to

4839: $\eta_C$. This implies that the Maslov number and area of any

4840: holo\-morphic disc $w: (D,\partial D) \rightarrow (C,L_0)$ can be

4841: computed as intersection numbers:

4842: \begin{equation} \label{eq:c-index-energy}

4843: \begin{aligned}

4844:  \mu(w) & = 2 (w \cdot C'), \\

4845:  \textstyle \int w^*\o_C & = w \cdot X_0.

4846: \end{aligned}

4847: \end{equation}

4848: 

4849: \item

4850: $w_0: (D,\partial D) \rightarrow (C,L_0)$ is a specific holo\-morphic

4851: disc with $w_0 \cdot X_0 = w_0 \cdot C' = 1$. The point $w_0(0)$

4852: should lie in $C' \cap X_0$, and $[w_0|\partial D] \in H_1(L_0)$

4853: should be nontrivial. Take the linearized operator of $w_0$, and

4854: restrict its domain to those sections $\xi$ with $\xi(0) = 0$. We ask

4855: that this restriction, denoted by

4856: \begin{equation} \label{eq:restricted-dbar}

4857:  D\bar\partial_{w_0,C}^{res}: \Gamma(w_0^*TC,w_0^*TL_0)^{res}

4858:  \longrightarrow \Omega^{0,1}(w_0^*TC),

4859: \end{equation}

4860: should have one-dimensional cokernel (in suitable Sobolev spaces).

4861: 

4862: \item

4863: Moreover,

4864: \begin{romanlist}

4865: \item \label{item:no-disc-1}

4866: every holo\-morphic disc $w$ in $(C,L_0)$ with $w \cdot C' = w \cdot

4867: X_0 = 1$ and $w(0) \in C' \cap X_0$ is equal to $w_0$, up to a

4868: possible action of $S^1 \subset Aut(D)$;

4869: \item \label{item:no-disc-2}

4870: there are no holomorphic discs $w$ in $(C,L_0)$ with $w \cdot C' =

4871: 0$, $w \cdot X_0 = 1$;

4872: \item \label{item:no-sphere}

4873: there are no holo\-morphic spheres $v$ in $C$ with $\int v^*\o_C =

4874: 1$.

4875: \end{romanlist}

4876: \end{itemize}

4877: \end{assumptions}

4878: 

4879: One can put $L_0$ into a smooth family $\{L_s\}$ (for $s \in \R$,

4880: $|s|$ small) of Lagrangian submanifolds, $L_s \subset X_{1/s}$, such

4881: that the union

4882: $

4883: L = \bigcup_s L_s

4884: $

4885: is Lagrangian (but of course not closed) in $X$. In fact, the

4886: condition on $L$ determines the $L_s$ uniquely, since it means that

4887: they get mapped into each other by symplectic parallel transport in

4888: $\R$-direction.

4889: 

4890: \begin{lemma}

4891: For each $s \neq 0$, $L_s$ is an exact Lagrangian submanifold in the

4892: affine K{\"a}hler surface $M_{1/s}$, and graded with respect to

4893: $\eta_{X_{1/s}}$.

4894: \end{lemma}

4895: 

4896: \proof Supposing that $s \neq 0$ remains sufficiently small, one has

4897: $L_s \cap X_0 = \emptyset$, and since $X_{1/s} \cap X_0 = X_{1/s}

4898: \cap X_\infty$, also $L_s \cap X_\infty = \emptyset$. Because

4899: $o_X|L_\infty$ is trivial and $L$ is Lagrangian, $o_X|L_s$ is trivial

4900: for any $s$. Moreover, its covariantly constant sections lie in the

4901: same homotopy class of nowhere zero sections as $\sigma_{X,0}|L_s$,

4902: because that is true for $s = 0$. On the other hand,

4903: $\sigma_{X,\infty}/\sigma_{X,0}$ is constant on $L_s$, so the

4904: covariantly constant sections are also homotopic to

4905: $\sigma_{X,\infty}|L_s$, which is equivalent to exactness. The

4906: argument with the grading is essentially the same. \qed

4907: 

4908: Exactness of $L_0$ with respect to $\theta_C$ means that the area of

4909: holo\-morphic discs in $(C,L_0)$ is integral. Our disc $w_0$ has

4910: minimal area, since by assumption $\int w_0^*\o_C = w_0 \cdot X_0 =

4911: 1$. By \cite{kwon-oh96,lazzarini98}, $w_0$ has a somewhere injective

4912: point. Standard transversality arguments tell us that there is a

4913: family $(J_{C,r})$, for $r \in \R$ with $|r|$ small, of compatible

4914: almost complex structures on $C$ such that $J_{C,0}$ is the given

4915: complex structure, and which makes $w_0$ parametrized-regular. Since

4916: the set of somewhere injective points is open, and $w_0$ intersects

4917: $C' \cup X_0$ only in $w_0(0)$, one can assume that all $J_{C,r}$ are

4918: equal to $J_{C,0}$ near $C \cap (C' \cup X_0)$. Having that, one can

4919: extend them to a family $(J_{X,r})$ of compatible almost complex

4920: structures on $X$, such that again $J_{X,0}$ is the given complex

4921: structure, and with the following two properties for all $r$:

4922: $J_{X,r} = J_{X,0}$ near $X_0$; and the projection $\pi^{-1} =

4923: \sigma_{X,\infty}/\sigma_{X,0}: X \setminus X_0 \rightarrow \C$ is

4924: $J_{X,r}$-holo\-morphic.

4925: 

4926: Consider the moduli space $\MM^{para}$ of pairs $(r,w)$, where $w:

4927: (D,\partial D) \rightarrow (X,L)$ is a $J_{X,r}$-holo\-morphic disc

4928: such that $w \cdot X_0 = 1$ and $w(0) \in X_0 \cap C'$, divided by

4929: the action of $S^1$ by rotation on $D$.

4930: 

4931: \begin{lemma} \label{th:is-in-fibre}

4932: For every $(r,w)$ in this moduli space, the image of $w$ is contained

4933: in some fibre $X_{1/s}$ of the Lefschetz pencil, with $s \in \R$.

4934: \end{lemma}

4935: 

4936: \proof Consider the meromorphic map $\pi^{-1} \circ w: D

4937: \dashrightarrow \C$. $w$ intersects $X_0$ only at $w(0)$ and with

4938: multiplicity one. At the same point it intersects $X_\infty$, and

4939: this implies that $\pi^{-1} \circ w$ is in fact holomorphic. Since

4940: its boundary values are real, it must be constant. \qed

4941: 

4942: \begin{lemma} \label{th:compactness}

4943: Let $(r_k,w_k)$ be a sequence of pairs, where $w_k: (D,\partial D)

4944: \rightarrow (X,L)$ is a $J_{X,r_k}$-holomorphic disc with $\int

4945: w_k^*\o_X = 1$, whose image is contained in $X_{1/s_k}$ for some $s_k

4946: \neq 0$. Suppose that $r_k,s_k \rightarrow 0$; then the $w_k$, after

4947: a possible reparametrization by automorphisms of $D$ and passing to a

4948: subsequence, converge to $w_0$.

4949: \end{lemma}

4950: 

4951: \proof Gromov compactness tells us that the limit of a subsequence is

4952: a stable disc in $X_\infty$, but in our case each non-constant

4953: component contributes an integral amount of energy, which means that

4954: there can be only one of them. Furthermore, the case where the

4955: non-constant component is a sphere is excluded by part

4956: \ref{item:no-sphere} of our assumptions. We conclude that after

4957: passing to a subsequence, the convergence is in $\smooth$-sense

4958: towards a limit which is a holomorphic disc $w: (D,\partial D)

4959: \rightarrow (C,L_0)$.

4960: 

4961: We know that $w \cdot X_0 = 1$, and \ref{item:no-disc-2} says that

4962: $w$ must intersect $C'$. Every $z \in D$ such that $w(z) \in C'$ must

4963: also satisfy $w(z) \in X_0$, otherwise the disc $w$ could not be the

4964: smooth limit of discs in the fibres $X_{1/s_k}$. In fact, a closer

4965: look at the local geometric situation shows that $w \cdot C' \leq w

4966: \cdot X_0$. Hence $w \cdot X_0 = w \cdot C' = 1$, and after

4967: reparametrization we may assume that $w(0) \in X_0 \cap C'$, so an

4968: appeal to \ref{item:no-disc-1} finishes the argument. \qed

4969: 

4970: For the next step, we need to compare the linearized operators

4971: associated to $w_0$ as a holomorphic disc in $C$ and $X$. From

4972: \eqref{eq:tangent-sequence} one gets a diagram with exact columns,

4973: \begin{equation} \label{eq:e-diagram}

4974: \xymatrix{

4975:  0 \ar[d] & 0 \ar[d] \\

4976:  \Gamma(w_0^*TC,w_0^*TL_0) \ar[r]^-{D\bar\partial_{w_0,C}} \ar[d] &

4977:  \Omega^{0,1}(w_0^*TC) \ar[d] \\

4978:  \Gamma(w_0^*TX,w_0^*TL) \ar[r]^-{D\bar\partial_{w_0,X}} \ar[d] &

4979:  \Omega^{0,1}(w_0^*TX) \ar[d] \\

4980:  \!\!\!\!\! \Gamma(w_0^*\O_C(C \cap X_0 - C \cap C'),w_0^*\Lambda) \ar[d]

4981:  \ar[r]^-{\bar\partial} & \Omega^{0,1}(w_0^*\O_C(C \cap X_0 - C \cap C')) \!\!\!\!

4982:  \ar[d] \\

4983:  0 & 0

4984: }

4985: \end{equation}

4986: where $\Lambda$ is the canonical real subbundle of $\O_C(C \cap X_0 -

4987: C \cap C')|L_0$, whose existence is due to the fact that $L_0$ is

4988: disjoint from $X_0 \cup C'$. As usual, we use suitable Sobolev spaces

4989: to make each row elliptic. The bottom row is a

4990: $\bar\partial$-operator on a line bundle over a disc: in fact, since

4991: $w_0$ intersects both $X_0$ and $C'$ only at $z = 0$ at with

4992: multiplicity one, this is just the standard $\bar\partial$-operator

4993: for functions with real boundary conditions. Hence, its Sobolev

4994: completion is onto with one-dimensional kernel, consisting of nowhere

4995: zero functions. Without losing exactness of the columns, we may

4996: restrict the domains of the top two $D\bar\partial$-operators to

4997: sections with $\xi(0) \in T(X_0 \cap C')$ (this tangent space is zero

4998: for the first row, and two-dimensional for the second one). Kernels

4999: and cokernels of the restrictions will fit into a long exact sequence

5000: \[

5001:  0 \rightarrow \ker\, D\bar\partial_{w_0,C}^{res} \rightarrow

5002:  \ker\, D\bar\partial_{w_0,X}^{res} \rightarrow \R \rightarrow

5003:  \coker\,D\bar\partial_{w_0,C}^{res} \rightarrow \coker\,D\bar\partial_{w_0,X}^{res}

5004:  \rightarrow 0.

5005: \]

5006: From this and \eqref{eq:c-index-energy} one sees that the expected or

5007: virtual dimension of the moduli space $\MM^{para}$ at $(0,w_0)$ is

5008: \[

5009: \text{v.dim}\, \MM^{para} = \ind\,D\bar\partial_{w_0,X}^{res} + 1 - 1

5010: = \ind\,D\bar\partial_{w_0,C} - 3 = 2 + \mu(w_0) - 3 = 1.

5011: \]

5012: (in the second expression, the $+1$ comes from the added parameter

5013: $r$, and the $-1$ from the action of $S^1$ by rotations). By

5014: assumption $\coker\,D\bar\partial_{w_0,C}^{res} \iso \R$, and by

5015: construction of $\{J_{C,r}\}$, this cokernel is spanned by $(\partial

5016: J_{C,r}/\partial r)_{r = 0} \circ dw_0 \circ i$. Hence, the cokernel

5017: of $D\bar\partial_{w_0,X}^{res}$ is spanned by $(\partial

5018: J_{X,r}/\partial r)_{r = 0} \circ dw_0 \circ i$. This means that the

5019: corresponding parametrized linearized operator is onto, so that

5020: $(0,w_0)$ is a regular point of $\MM^{para}$.

5021: 

5022: \begin{lemma}

5023: There is a family $(r_t,w_t)$ of regular points of $\MM^{para}$, for

5024: $t \in \R$ with $|t|$ small, with $r_0 = 0$ and $w_0$ our given

5025: holomorphic disc. This satisfies $w_t(D) \subset X_{1/s_t}$ for some

5026: $s_t \in \R$ smoothly depending on $t$, and with $s_0 = 0$,

5027: $(ds_t/dt)_{t = 0} \neq 0$.

5028: \end{lemma}

5029: 

5030: \proof We know that $\MM^{para}$ is locally a smooth one-manifold, so

5031: take a family $(r_t,w_t)$ which parametrizes it locally. The fact

5032: that $w_t \in X_{1/s_t}$ for some $s_t$ comes from Lemma

5033: \ref{th:is-in-fibre}, and the smoothness of $t \mapsto s_t$ is

5034: obvious. Now suppose that $(ds_t/dt)_{t = 0}$ vanishes. Then

5035: $(\partial r_t/\partial t, \partial w_t/\partial t)_{t = 0}$ lies in

5036: the kernel of the parametrized version of

5037: $D\bar\partial_{w_0,C}^{res}$. By construction this is onto, and for

5038: index reasons the only nontrivial kernel elements are those obtained

5039: from infinitesimal rotations of $D$. This contradicts the fact that

5040: we started with a local chart for $\MM^{para}$. \qed

5041: 

5042: \begin{lemma} \label{th:fibre-regular}

5043: For all small $t \neq 0$, $w_t$ is parametrized-regular as a

5044: $J_{r_t}$-holomor\-phic disc in $(X_{1/s(t)},L_{s(t)})$.

5045: \end{lemma}

5046: 

5047: The proof is by a comparison argument using a diagram of

5048: $\bar\partial$-operators similar to \eqref{eq:e-diagram}; we leave it

5049: to the reader. Now fix some small $\epsilon>0$ and $0 \neq t \ll

5050: \epsilon$. Then in the family $(J_{X,r}|X_{1/s(t)})$, $r \in

5051: [-\epsilon;\epsilon]$, of almost complex structures there is no

5052: pseudo-holomorphic disc in $(X_{1/s(t)},L_{s(t)})$ with unit area

5053: other than $w_t$. In fact, the contrary assumption for very small

5054: $\epsilon$ and $t$ leads to a contradiction with Lemma

5055: \ref{th:compactness} and the local structure of the moduli space

5056: $\MM^{para}$. By Lemma \ref{th:fibre-regular}, this situation

5057: persists under a small perturbation of our family of almost complex

5058: structures; by such a perturbation, we can make it regular with

5059: respect to $L_{s(t)}$ at the endpoints $r = \pm\epsilon$, and

5060: eliminate pseudo-holomorphic spheres in $X_{1/s(t)}$ for all values

5061: of $r$. We are now in the situation of Corollary

5062: \ref{th:wall-crossing}, which shows that

5063: $\Fuk(X_{1/s(t)},X_{1/s(t),\infty})$ is nontrivial at first order.

5064: Using Remark \ref{th:class-of-j-2} and symplectic parallel transport

5065: one can extend this to other fibres of our pencil, so the conclusion

5066: is:

5067: 

5068: \begin{cor} \label{th:wall-crossing-2}

5069: Suppose that Assumptions \ref{as:l-zero} hold. Take a smooth fibre

5070: $X_z$ of our pencil of Calabi-Yau surfaces. Then its relative Fukaya

5071: category $\Fuk(X_z,X_{0,\infty})$, seen as a one-parameter

5072: deformation of $A_\infty$-categories, is nontrivial at first order.

5073: \qed

5074: \end{cor}

5075: 

5076: \section{Computations in Fukaya categories\label{sec:induction}}

5077: 

5078: To effect the link between symplectic geometry and homological

5079: algebra, we basically rely on two tools. The first is the

5080: correspondence between Dehn twists and algebraic twists, which is a

5081: reformulation of the long exact sequence in Floer cohomology. The

5082: version in the literature \cite{seidel01,seidel04} is for the exact

5083: or affine situation, but as will be outlined below, the proof can be

5084: easily adapted to the projective and relative cases. Our second tool

5085: is a dimensional induction machine based on the notion of matching

5086: cycle, developed in \cite{seidel04}; this is strictly limited to the

5087: affine context, which is the reason why affine varieties appear in

5088: this paper at all. We will treat dimensional induction as a black

5089: box, stating its properties in an axiomatic way only; ultimately it

5090: is again an application of the long exact sequence, but the way in

5091: which this works is too complicated to be summarized briefly.

5092: 

5093: \subsection{}

5094: Let $M$ be an affine Calabi-Yau of dimension $n>1$, and $L \subset M$

5095: a Lagrangian sphere. Equip it with a grading and the unique $Spin$

5096: structure, making it into an object $L^\br$ of $\Fuk(M)$. Let

5097: $\tau_L$ be the Dehn twist along $L$, and $\taugr_L$ its canonical

5098: grading (again, there is the small matter of framing, which we

5099: ignore). Consider also the algebraic twist $T_{L^\br}$ along $L^\br$,

5100: which is a map of objects of $D^b\Fuk(M)$. Let $L_1^\br$ be any exact

5101: Lagrangian brane, with underlying Lagrangian submanifold $L_1$.

5102: Clearly $\tau_L^{-1}(L_1)$ carries an induced brane structure, making

5103: it into an object $\taugr_L^{-1}(L_1^\br)$ of the Fukaya category.

5104: What \cite{seidel04} says is that

5105: 

5106: \begin{prop} \label{th:exact-sequence-1}

5107: $T_{L^\br}(\taugr_L^{-1}(L_1^\br))$ is isomorphic to $L_1^\br$ in

5108: $D^b\Fuk(M)$.

5109: \end{prop}

5110: 

5111: On the algebraic side, this is an application of Lemma

5112: \ref{th:recognize-twist}, where $W$ is some other exact Lagrangian

5113: brane $L_0^\br$; we need to explain how a suitable moduli space setup

5114: provides the necessary pair $(k,h)$. This is a variation of the

5115: construction in \cite{seidel01}, based on pseudo-holomorphic sections

5116: of exact symplectic Lefschetz fibrations with Lagrangian boundary

5117: conditions. Specifically, taking $S = \mathbb{H}$ the upper half

5118: plane, there is such a fibration $E_L \rightarrow S$, which has a

5119: single critical point somewhere over $int(S)$. The regular fibres of

5120: $E_L$ are isomorphic to $M$; the monodromy around the singularity is

5121: of course $\tau_L$; and there is a canonical trivialization away from

5122: a compact subset of $S$. Given any exact Lagrangian brane in $M$ such

5123: as $L_1^\br$, one can define a subbundle $F_{L_1} \subset

5124: E_L|\partial S$ which, in the canonical trivialization, satisfies

5125: \[

5126:  F_{L_1} \cap E_s = \begin{cases}

5127:  \tau_L^{-1}(L_1) & s \ll 0, \\

5128:  L_1 & s \gg 0.

5129:  \end{cases}

5130: \]

5131: By considering an equation similar to \eqref{eq:generalized-floer}

5132: for sections of $E_L$ with boundary values on $F_{L_1}$, one then

5133: obtains a ``relative invariant'' in the style of

5134: \cite{piunikhin-salamon-schwarz94} taking values in

5135: $HF^0(\taugr_L^{-1}(L_1^\br),L_1^\br)$. Take $k$ to be the cocycle

5136: representing it, which of course depends on the specific choices of

5137: almost complex structures etc.\ used in the definition.

5138: %

5139: \FIGURE{fig:homotopy}{h.eps}{hb}%

5140: 

5141: For $h$ one uses a parametrized version of the same construction,

5142: applied to a family of Lefschetz fibrations $E^t$ with boundary

5143: conditions $F^t$, $t \in \R$. Figure \ref{fig:homotopy} shows the

5144: asymptotic behaviour of this family: for $t \ll 0$, $(E^t,F^t)$ is

5145: the fibre connected sum of $(E_L,F_{L_1})$ and a trivial fibration

5146: over the three-punctured disc, with boundary conditions over the

5147: different boundary components given by the Lagrangian submanifolds

5148: $L$, $\tau_L^{-1}(L_1)$ and $L_1$. The invariant associated to the

5149: degenerate limit $t = -\infty$ is therefore the composition

5150: \begin{equation} \label{eq:t-plus}

5151: \begin{aligned}

5152:  CF_M^*(L^\br,\taugr_L^{-1}(L_1^\br)) & \longrightarrow CF_M^*(L^\br,L_1^\br), \\

5153:  a & \longmapsto \mu^2_M(a,k).

5154: \end{aligned}

5155: \end{equation}

5156: For $t \gg 0$, $(E^t,F^t)$ again decomposes as a connected sum. Its

5157: first component is $E_L$ as before, but now with boundary condition

5158: $F_L$; and the second component is trivial, with boundary conditions

5159: given by $L$, $\tau_L^{-1}(L_1)$ and again $L$. In principle, the

5160: associated invariant is

5161: \begin{equation} \label{eq:t-minus}

5162: \begin{aligned}

5163:  CF_M^*(L^\br,\taugr_L^{-1}(L_1^\br)) \iso

5164:  CF_M^*(\taugr_L(L^\br),L_1^\br) &

5165:  \longrightarrow CF_M^*(L^\br,L_1^\br), \\

5166:  a & \longmapsto \mu^2_M(l,a),

5167: \end{aligned}

5168: \end{equation}

5169: where $l \in CF^0(L^\br,\taugr_L(L^\br)) \iso

5170: CF^0(\taugr^{-1}(L^\br),L^\br)$ is obtained from $(E_L,F_L)$.

5171: However, since $\taugr_L(\Lgr) = \Lgr[1-n]$ with $n>1$, we have

5172: $HF^0(L^\br,\taugr_L(L^\br)) = HF^{1-n}(L^\br,L^\br) \iso

5173: H^{1-n}(S^n;\C) = 0$, and by suitably choosing the Floer data, one

5174: can achieve that $CF^0(L^\br,\taugr_L(L^\br))$ vanishes too. Hence

5175: $l$ and the expression \eqref{eq:t-minus} are both zero, and from

5176: this one deduces that the parametrized moduli space of sections

5177: associated to the deformation from $t = -\infty$ to $t = +\infty$

5178: yields a chain homotopy $h$ between \eqref{eq:t-plus} and zero, as

5179: desired (this vanishing argument, which relies on the grading, is

5180: different from the one in \cite[Proposition 2.13]{seidel01}; we use

5181: it because it extends easily to the projective and relative cases).

5182: 

5183: Since all Lagrangian submanifolds involved are exact, there are

5184: real-valued action functionals for all the Floer cochain groups

5185: involved, so that in the relevant complex \eqref{eq:total-cone} one

5186: can introduce an $\R$-filtration by the values of the action. Its

5187: acyclicity is then proved by a spectral sequence argument. Briefly,

5188: one replaces the $\R$-filtration by a $\Z$-subfiltration, which

5189: divides the action values into intervals of some suitably chosen

5190: small size $\epsilon>0$. After passing to the associated graded space

5191: of the filtration, one finds that the Floer differentials $\mu^1$ and

5192: the terms $\mu^2(h(a_2),a_1)$, $\mu^3(k,a_2,a_1)$ disappear, while

5193: the remaining terms $\mu^2(a_2,a_1)$ and $\mu^2(k,b)$ make the

5194: complex exact, hence acyclic. For this to work, some care needs to be

5195: observed when choosing the map $\tau_L$ and the cochain level

5196: representatives for $k$ and $h$, see \cite[Chapter 3]{seidel01} for

5197: details.

5198: 

5199: \subsection{}

5200: We now turn to the consequences of Proposition

5201: \ref{th:exact-sequence-1}, combining arguments from \cite{seidel04}

5202: with the notion of negative graded symplectic automorphisms

5203: introduced in Section \ref{sec:negativity}.

5204: 

5205: \begin{lemma} \label{th:negative-exact}

5206: Suppose that $m_X = 0$, so $M$ is the affine part of a projective

5207: Calabi-Yau $X$. Let $L_1,\dots,L_m \subset M$ be Lagrangian spheres,

5208: and suppose that

5209: \[

5210: \taugr_{L_1}\dots \taugr_{L_m} \in \Autgr(M)

5211: \]

5212: is isotopic, within that group, to a graded symplectic automorphism

5213: $\phigr$ whose extension to $X$ is negative (see Remark

5214: \ref{th:extend-grading}). Then $L_1^\br,\dots,L_m^\br$ are

5215: split-generators for $D^\pi\Fuk(M)$.

5216: \end{lemma}

5217: 

5218: \proof Let $L^\br$ be any exact Lagrangian brane. From Lemma

5219: \ref{th:maslov-growth} we know that if $d$ is sufficiently large, a

5220: small exact perturbation $L'$ of $L$ will cause the Maslov index of

5221: any intersection point $x \in L' \cap \phi^d(L)$ to be strictly

5222: negative. By definition of the Floer cochain space

5223: \eqref{eq:basic-floer-complex}, it follows that

5224: \[

5225: Hom_{H^0\Fuk(M)}(L^\br,\phigr^d(L^\br)) =

5226: HF^0_M(L^\br,\phigr^d(L^\br)) = 0.

5227: \]

5228: On the other hand, $\phigr^d(L^\br)$ is isomorphic in $D^b\Fuk(M)$ to

5229: $(T_{L_1^\br} \dots T_{L_m^\br})^d(L^\br)$. Lemma

5230: \ref{th:twist-generators} says that $L^\br$ lies in the subcategory

5231: of $D^\pi\Fuk(M)$ split-generated by the $L_k^\br$; by construction

5232: of that category, this implies that these objects are

5233: split-generators. \qed

5234: 

5235: \begin{remark} \label{th:codimension-n}

5236: In Lemma \ref{th:negative-exact}, one can replace the negativity of

5237: $\phigr$ by the following slightly weaker assumption. There is a

5238: closed subset $\Sigma \subset X$ which is a finite union of real

5239: submanifolds of dimension $<n$, which admits arbitrarily small open

5240: neighbourhoods $W$ satisfying $\phi(W) = W$, and such that $\phigr$

5241: is negative on $X \setminus W$. The reason is that by a generic

5242: perturbation, one can make $\Lambda$ disjoint from $\Sigma$. Then it

5243: is disjoint from a sufficiently small $W$, and the same will hold for

5244: all iterates $\phi^d(\Lambda)$, so that the breakdown of negativity

5245: on $\Sigma$ can be disregarded.

5246: \end{remark}

5247: 

5248: \subsection{}

5249: To adapt Proposition \ref{th:exact-sequence-1} to the case of a

5250: projective Calabi-Yau surface $X$, with $L$ a Lagrangian sphere and

5251: $L_1^\rbr$ a rational brane, two points have to be taken into

5252: account. Firstly, more care needs to be exercised when defining the

5253: invariants derived from pseudo-holomorphic sections, to avoid

5254: bubbling off of holomorphic discs. This is not really new, since we

5255: have dealt with the same problem in the construction of $\Fuk(X)$.

5256: Note that up to shifts, $L$ gives rise to a unique object $L^\rbr$ of

5257: the Fukaya category, by Lemma \ref{th:spheres}. For

5258: $\taugr_L^{-1}(L_1^\rbr)$, one needs to choose the regular almost

5259: complex structure which is the $\tau_L$-pullback of that associated

5260: to the $L_1$. The second modification concerns the filtration

5261: argument. Our ground ring is now $\Lambda_\Q$, and each

5262: pseudo-holomorphic section contributes $\pm q^e$ as in \eqref{eq:qe}.

5263: In the $\R$-filtration of the Floer cochain groups, one now uses

5264: powers of $q$ as a replacement for values of the action functional.

5265: Since only a finite number of rational branes are involved ($L^\rbr$,

5266: $L_0^\rbr$, $L_1^\rbr$), one can actually work over a subfield

5267: $\Lambda_{(1/d)\Z} \subset \Lambda_\Q$ containing only roots of $q$

5268: of some fixed order $d$. An essentially equivalent and more familiar

5269: formulation is to say that all action functionals together have only

5270: a discrete set of periods. This allows one to carry over the previous

5271: argument, with its use of a $\Z$-subfiltration: it suffices to make

5272: sure that $\epsilon \ll 1/d$ (actually, this is the only point in the

5273: entire paper that would have to be modified substantially if one

5274: wished to include non-rational branes in $\Fuk(X)$). The outcome is

5275: 

5276: \begin{prop} \label{th:exact-sequence-2}

5277: $T_{L^\br}(\taugr_L^{-1}(L_1^\br))$ is isomorphic to $L_1^\br$ in

5278: $D^b\Fuk(X)$. \qed

5279: \end{prop}

5280: 

5281: There is also a version for the relative Fukaya category, where $L$

5282: is again a Lagrangian sphere in $M$ and $L_1^\mbr$ an object of that

5283: category. The definition of the relevant version of $(k,h)$ is as in

5284: the projective case, except sections $u$ now contribute with integer

5285: powers of $q$ according to their intersection with $X_\infty$. As for

5286: the acyclicity of \eqref{eq:total-cone}, one filters it by powers of

5287: $q$, and then taking the associated graded space brings one back to

5288: the affine situation, where we already know that acyclicity holds.

5289: 

5290: \begin{prop} \label{th:exact-equence-3}

5291: $T_{L^\mbr}(\taugr_L^{-1}(L_1^\mbr))$ is isomorphic to $L_1^\mbr$ in

5292: $D^b\Fuk(X,X_\infty)$. \qed

5293: \end{prop}

5294: 

5295: 

5296: The argument from Lemma \ref{th:negative-exact} carries over to the

5297: projective and relative cases without any modifications, and so does

5298: Remark \ref{th:codimension-n}. By combining this with Proposition

5299: \ref{th:negativity}, one arrives at the following conclusion:

5300: 

5301: \begin{cor} \label{th:generates-everything}

5302: Let $X$ be a projective K{\"a}hler threefold which is Fano, with $o_X

5303: = \K_X^{-1}$, equipped with sections $\sigma_{X,0}$,

5304: $\sigma_{X,\infty}$ which generate an almost Lefschetz pencil of

5305: Calabi-Yau surfaces $\{X_z\}$. Assume that $X_0 \cap X_\infty$ has no

5306: rational components. Pick some base point $z_* \in \C \setminus

5307: Critv(\pi_M)$, base path $c_*$, and a distinguished basis of

5308: vanishing cycles $V_{c_1},\dots,V_{c_r} \subset M_{z_*}$. Then the

5309: $V_{c_i}$, equipped with some gradings, are split-generators for all

5310: versions of the split-closed derived Fukaya category:

5311: $D^\pi\Fuk(M_{z_*})$, $D^\pi\Fuk(X_{z_*})$ and $D^\pi\Fuk(X_{z_*},

5312: X_{z_*,\infty})$. \qed

5313: \end{cor}

5314: 

5315: \subsection{\label{subsec:induction}}

5316: Let $M = X \setminus X_\infty$ be an affine Calabi-Yau of dimension

5317: $n > 1$, equipped with a quasi-Lefschetz pencil $\{X_z\}$ and the

5318: associated holomorphic function $\pi_M: M \rightarrow \C$. Choose a

5319: base point $z_*$ and base path $c_*$. Let $c_1,\dots,c_r$ be a basis

5320: of vanishing paths, and $V_{c_1},\dots,V_{c_r} \subset M_{z_*}$ the

5321: associated vanishing cycles. Make them into objects $V_{c_i}^\br$ of

5322: $\Fuk(M_{z_*})$; any choice of grading is allowed, and the $Spin$

5323: structure is unique except for $n = 2$, where one needs to take the

5324: {\em nontrivial} one. Let

5325: \begin{equation} \label{eq:directed}

5326: \F^\rightarrow = \Fuk^\rightarrow(V_{c_1}^\br,\dots,V_{c_r}^\br)

5327: \end{equation}

5328: be the directed $A_\infty$-subcategory of $\Fuk(M_{z_*})$ consisting

5329: of these objects. We associate to each embedded vanishing path $c$,

5330: subject to the conditions $im(c) \cap im(c_*) = \{z_*\}$ and $\Rgeq

5331: c'(0) \neq \Rgeq c_*'(0)$, an object $D_c$ of $D^b(\F^\rightarrow)$,

5332: unique up to isomorphism and shifts, according to the following

5333: rules:

5334: \begin{itemize}

5335: \item

5336: to $c_i$ belongs the object $D_{c_i} = V_{c_i}^\br$ of

5337: $\F^\rightarrow$ itself;

5338: \item

5339: an isotopy of $c$ within the class of such paths does not affect

5340: $D_c$;

5341: \item if we have vanishing paths $c,c',c''$ as in Figure

5342: \ref{fig:hurwitz}, then $D_{c'} \iso T_{D_c}(D_{c''})$ and $D_{c''}

5343: \iso T^\vee_{D_c}(D_{c'})$.

5344: \end{itemize}

5345: From this and the elements of the theory of mutations

5346: \cite{rudakov90}, one sees that all the $D_c$ are exceptional

5347: objects, which means that $Hom^*(D_c,D_c)$ is one-dimensional in

5348: degree zero, and zero in all other degrees.

5349: 

5350: Next, we associate objects $S_d$ of $D^b(\Fuk^\rightarrow)$ to

5351: matching paths which are disjoint from $c_*$. For this, take paths

5352: $c',c'',d$ as in Figure \ref{fig:triangle}. Suppose that the

5353: vanishing cycles associated to $c',c''$ are isotopic, so that $d$ is

5354: a matching paths. The objects associated to $c',c''$ will satisfy

5355: \begin{equation} \label{eq:matching-objects}

5356:  Hom^k_{D^b(\Fuk^\rightarrow)}(D_{c'},D_{c''}) =

5357:  \begin{cases} \C & k = l,l+n-1, \\

5358:  0 & \text{otherwise} \end{cases}

5359: \end{equation}

5360: for some $l \in \Z$. After a shift, we may assume that $l = 0$, and

5361: then the rule is:

5362: \begin{itemize}

5363: \item

5364: $S_d$ is the mapping cone $\{D_{c'} \rightarrow D_{c''}\}$ over some

5365: nontrivial morphism.

5366: \end{itemize}

5367: An easy computation shows that $Hom^*(S_d,S_d)$ is one-dimensional in

5368: degrees $0$ and $n$, and zero otherwise. This resemblance to the

5369: cohomology of an $n$-sphere is not accidental:

5370: %

5371: \FIGURE{fig:hurwitz}{hurwitz.eps}{ht}

5372: \FIGURE{fig:triangle}{triangle.eps}{ht} %

5373: 

5374: \begin{theorem} \label{th:induction}

5375: Let $d_1,\dots,d_m$ be matching paths which are disjoint from our

5376: base path $c_*$. Assign objects $S_{d_1},\dots,S_{d_m}$ in

5377: $D^b(\Fuk^\rightarrow)$ as explained above. Then the full

5378: $A_\infty$-subcategory of $Tw\Fuk^\rightarrow$ consisting of these

5379: objects is quasi-isomorphic to the full subcategory of the Fukay

5380: category of the total space, $\Fuk(M)$, whose objects are the

5381: matching cycles $\Sigma_{d_1},\dots,\Sigma_{d_m}$, with some choice

5382: of gradings. \qed

5383: \end{theorem}

5384: 

5385: This way of stating the result leaves several things implicit. Since

5386: one can get any vanishing path by starting with $c_1,\dots,c_r$ and

5387: applying the moves shown in Figure \ref{fig:triangle}, our rules

5388: determine the objects $D_c$. However, it is not a priori clear that

5389: the assignment $c \mapsto D_c$ is unambiguously defined, since one

5390: can get the same path by different sequences of moves. Similarly,

5391: \eqref{eq:matching-objects} and the well-definedness of $S_d$ are not

5392: obvious. The fact that all of this works is a nontrivial part of the

5393: statement. For proofs see \cite{seidel04}.

5394: 

5395: \subsection{\label{subsec:combinatorial}}

5396: The usefulness of Theorem \ref{th:induction} is particularly obvious

5397: for $n = 2$, when the vanishing cycles are curves on a surface, since

5398: then $\Fuk^\rightarrow$ can be computed in a purely combinatorial

5399: way, up to quasi-isomorphism (versions of this fact have been

5400: discovered independently in various contexts, see e.g.\

5401: \cite{desilva98,polishchuk-zaslow98,chekanov99}). We will now outline

5402: how to do this computation, referring to the papers just quoted and

5403: to \cite{seidel04} for further details and justification.

5404: 

5405: {\em Exactness:} For simplicity, we use a compact piece $N \subset

5406: M_{z_*}$ which is a surface with boundary, capturing all of the

5407: topology of $M_{z_*}$ (the inclusion is a homotopy equivalence). Draw

5408: simple closed curves $\nu_k$ on $N$ which are isotopic to $V_{c_k}$

5409: and in general position (any two intersect transversally, and there

5410: are no triple intersections). One should then verify that these

5411: curves are exact Lagrangian submanifolds for some choice of one-form

5412: $\theta_N$ with $d\theta_N > 0$. For this, consider the decomposition

5413: of $N$ given by the graph $\Gamma = \bigcup_i \nu_i$. Suppose that

5414: one can associate to each component $Z \subset N \setminus \Gamma$

5415: which is disjoint from $\partial N$ a positive weight $w(Z) > 0$,

5416: with the following property: if $\sum_k m_k \bar{Z}_k$, $m_k \in \Z$,

5417: is a two-chain whose boundary is a linear combination of the $\nu_k$,

5418: then

5419: \begin{equation} \label{eq:zero-energy}

5420: \sum_k m_k w(Z_k) = 0.

5421: \end{equation}

5422: One needs to check this only for a finite number of chains, which

5423: form a basis for the homology relations between the $\nu_k$. If this

5424: works out, one can find a symplectic form $\o_N$ with $\int_Z \o_N =

5425: w(Z)$, and that can be written as $\o_N = d\theta_N$ with

5426: $\int_{\nu_k} \theta_N = 0$ for all $k$.

5427: 

5428: \begin{remark} \label{th:painting-by-numbers}

5429: This ``painting by numbers'' step is rather tedious, so we will

5430: mention a way of reducing the amount of work. Relations in homology

5431: between the vanishing cycles correspond to elements of $H_2(M)$

5432: (because $H_2(M_{z_*}) = 0$, and $M$ is obtained topologically by

5433: attaching two-cells to the vanishing cycles). It is sufficient to

5434: check \eqref{eq:zero-energy} for relations corresponding to elements

5435: in $im(\pi_2(M) \rightarrow H_2(M))$. Equivalently, these are

5436: relations obtained from maps $\Sigma \rightarrow M_{z_*}$ where

5437: $\Sigma$ is a genus zero surface with boundary, each boundary

5438: component getting mapped to some $\nu_i$. This weaker condition does

5439: not guarantee that the vanishing cycles can be made exact, but it

5440: allows one to define the directed Fukaya category, and moving the

5441: $\nu_i$ within that class does not affect the quasi-isomorphism type

5442: of the category. Hence, after a (possibly non-Hamiltonian) isotopy,

5443: one sees that the directed Fukaya category is the same as for the

5444: exact choice of vanishing cycles. The appearance of $\pi_2(M)$ should

5445: remind the reader of the assumption $\o|\pi_2(M) = 0$ used frequently

5446: in symplectic topology, and indeed it addresses the same issue

5447: (getting a bound on the energy of pseudo-holomorphic polygons).

5448: \end{remark}

5449: 

5450: {\em Gradings:} Recall that the holomorphic one-form $\eta_{M_{z_*}}$

5451: defines a phase function $\alpha_{M_{z_*}}$. One can associate to

5452: this an unoriented foliation of $M_{z_*}$, consisting of those

5453: tangent lines $\Lambda$ for which $\alpha_{M_{z_*}}(\Lambda) = 1$. We

5454: draw some unoriented foliation $\Fol_N$ on $N$ which is homotopic to

5455: the one determined by $\alpha_{M_{z_*}}$. To equip the $\nu_k$ with

5456: gradings, one has to choose a point $b_k \in \nu_k$ which should not

5457: be an intersection point with any other $\nu_l$, and a homotopy

5458: between $\Fol_N$ and $T\nu_k$ at that point (within the $\RP{1}$ of

5459: lines in $TN_{b_k}$). This determines the Maslov indices $I(x)$ of

5460: all intersection points $x \in \nu_i \cap \nu_j$, $i < j$, and also

5461: orientations of the $\nu_i$. Note that $(-1)^{I(x)+1}$ is the local

5462: intersection number $(\nu_i \cdot \nu_j)_x$.

5463: 

5464: {\em $Spin$ structures:} For every $x$ as before with $(\nu_i \cdot

5465: \nu_j)_x = +1$, we choose a nonzero tangent vector $\xi_x$ to $\nu_j$

5466: at $x$. Of course, one can take these to be the vectors pointing

5467: along the orientation, but that is not strictly speaking necessary.

5468: Next, recall that we have to equip the $\nu_i$ with nontrivial $Spin$

5469: structures, or what is the same, with nontrivial double covers.

5470: Having already chosen points $b_k$, the simplest way to do this is to

5471: take the double cover which is trivial on $\nu_k \setminus b_k$, and

5472: whose two sheets get exchanged when passing over that point.

5473: 

5474: {\em Counting polygons:} We now introduce a directed

5475: $A_\infty$-category $\Cat^\rightarrow$ with objects

5476: $\nu_1,\dots,\nu_r$. Because of directedness, it is sufficient to

5477: define the $hom_\B(\nu_i,\nu_j)$ for $i<j$, and the composition maps

5478: between those spaces. Set as usual

5479: \[

5480:  hom_{\Cat^\rightarrow}^k(\nu_i,\nu_j) = \bigoplus_{I(x) = k} \C_x,

5481: \]

5482: but where now $\C_x$ is identified with $\C$ in a fixed way. Given

5483: $i_0 < \dots < i_d$ in $\{1,\dots,r\}$ and intersection points $x_0

5484: \in \nu_{i_0} \cap \nu_{i_d}$, $x_1 \in \nu_{i_0} \cap \nu_{i_1}$,

5485: \dots, $x_d \in \nu_{i_{d-1}} \cap \nu_{i_d}$, one considers oriented

5486: immersions $u: \Delta \rightarrow N$, where $\Delta$ is some model

5487: $(d+1)$-gon in the plane with numbered sides and corners. Under $u$,

5488: the sides should go to the $\nu_{i_k}$, and the corners to the $x_k$;

5489: moreover, the immersion should be a local diffeomorphism up to the

5490: corners, so that the image has locally an angle $<\pi$. One considers

5491: two such $u$ to be equivalent if they differ by a diffeomorphism of

5492: $\Delta$. Each equivalence class contributes to

5493: $\mu^k_{\Cat^\rightarrow}$ by

5494: \[

5495:  \pm 1: \C_{x_d} \otimes \dots \otimes \C_{x_1} \rightarrow \C_{x_0},

5496: \]

5497: and the sign is the product of the following factors: (i) for each $k

5498: \geq 1$ such that $(\nu_{i_{k-1}} \cdot \nu_{i_k})_{x_k} = 1$, we get

5499: a $-1$ iff the tangent vector $\xi_{x_k}$ points away from

5500: $u(\Delta)$. (ii) We get another $-1$ if $(\nu_{i_0} \cap

5501: \nu_{i_d})_{x_0} = 1$ and the tangent vector $\xi_{x_0}$ points

5502: towards $u(\Delta)$. (iii) Every time that $u|\partial\Delta$ passes

5503: over a point $b_k$, this adds a further $-1$.

5504: 

5505: \begin{lemma}

5506: $\Cat^{\rightarrow}$ is quasi-isomorphic to the directed subcategory

5507: $\Fuk^{\rightarrow}$ from \eqref{eq:directed}, at least up to a shift

5508: in the gradings of the $V_{c_i}^\br$. \qed

5509: \end{lemma}

5510: 

5511: Starting from this, Theorem \ref{th:induction} allows one to

5512: determine the full $A_\infty$-subcategory of $\Fuk(M)$ consisting of

5513: any finite number of matching cycles, and in particular the Floer

5514: cohomology of any two such cycles. Note that matching cycles are

5515: Lagrangian two-spheres in a four-manifold, so their Floer cohomology

5516: is not amenable to computation in any direct way.

5517: 

5518: \subsection{\label{subsec:addition}}

5519: We will need a minor addition to the previously explained

5520: computational method, which takes into account the presence of a

5521: homomorphism $\rho: \pi_1(M) \rightarrow \Gamma$ into a finitely

5522: generated abelian group $\Gamma$. Restriction yields a homomorphism

5523: from the fundamental group of the fibre $M_{z_*}$, and all vanishing

5524: cycles $V_{c_i}$ bound balls in $M$, hence the composition

5525: $\pi_1(V_{c_i}) \rightarrow \pi_1(M_{z_*}) \rightarrow \Gamma$ is

5526: zero. As discussed in Section \ref{subsec:coverings}, after choosing

5527: lifts $V_{c_i,\Gamma}$ to the associated $\Gamma$-cover

5528: $M_{z_*,\Gamma}$, one has a $\Gamma^*$-action on \eqref{eq:directed}.

5529: We can use the same rules as before to associate to each vanishing

5530: path $c$ an object $D_c$ of the equivariant derived category

5531: $D^b_{\Gamma^*}(\A)$, see Remark \ref{th:twist-is-equivariant}. For

5532: $c',c'',d$ are as in Figure \ref{fig:triangle}, the objects $D_{c'}$,

5533: the objects $D_{c''}$ will now satisfy

5534: \[

5535:  Hom^k_{D^b(\A)}(D_{c'},D_{c''}) \iso \begin{cases} W & k = l,l+n-1, \\

5536:  0 & \text{otherwise,} \end{cases}

5537: \]

5538: where $W$ is some one-dimensional $\Gamma^*$-representation. After

5539: shifting $D_{c'}$ and tensoring it with $W^*$, one may assume that $l

5540: = 0$ and $W$ is trivial. Then the cone $S_d = \{D_{c'} \rightarrow

5541: D_{c''}\}$ over a nontrivial degree zero morphism is again an object

5542: of the equivariant derived category. The equivariant version of

5543: Theorem \ref{th:induction} says that the full $A_\infty$-subcategory

5544: of $Tw_{\Gamma^*}\Fuk^\rightarrow$ with objects

5545: $S_{d_1},\dots,S_{d_m}$ is $\Gamma^*$-equivariantly quasi-isomorphic

5546: to the full subcategory of $\Fuk(M)$ with objects

5547: $\Sigma_{d_1},\dots,\Sigma_{d_m}$, where we assume that gradings and

5548: lifts to $M_\Gamma$ have been chosen in some appropriate way for

5549: these matching cycles. For the proof, which is not particularly

5550: difficult, there are two essentially equivalent viewpoints: one

5551: either goes through the relevant material in \cite{seidel04}, making

5552: sure that everything can be made compatible with the

5553: $\Gamma^*$-actions; or one works directly with the covering

5554: $M_\Gamma$ and its holomorphic function $M_\Gamma \rightarrow M

5555: \rightarrow \C$.

5556: 

5557: \section{The algebras $Q_4$ and $Q_{64}$\label{sec:4-64}}

5558: 

5559: Using split-generators, the categories which occur in homological

5560: mirror symmetry can be described in terms of $A_\infty$-structures on

5561: finite-dimensional algebras. This section introduces the algebra

5562: relevant for our computation, as well as a simpler version which

5563: serves as a stepping-stone. The main point is to use the general

5564: theory from Section \ref{sec:deformations} to partially classify

5565: $A_\infty$-structures on those algebras. We end by applying the

5566: results to coherent sheaves on the mirror of the quartic surface:

5567: this finishes work on the algebro-geometric side of mirror symmetry,

5568: and it also helps to clarify the amount of things still to be proved

5569: on the symplectic side.

5570: 

5571: \subsection{\label{subsec:q-preliminary}}

5572: We begin by recalling a general ``doubling'' construction. Let

5573: $C^\rightarrow$ be a directed ($\C$-linear graded) category, with

5574: objects $(X_1,\dots,X_m)$. The {\em trivial extension category} of

5575: degree $d > 0$ is a category $C$ with the same objects, and it

5576: satisfies:

5577: \begin{itemize}

5578: \item

5579: The directed subcategory of $C$ is the given $C^\rightarrow$.

5580: \item

5581: $C$ is a Frobenius category of degree $d$, in the sense that there

5582: are linear maps $\int_{X_j}: Hom_C^d(X_j,X_j) \rightarrow \C$, such

5583: that

5584: \begin{equation} \label{eq:int-pairing}

5585: \leftsc a,b \rightsc = \int_{X_j} ab

5586: \end{equation}

5587: for $a \in Hom_C^*(X_k,X_j)$, $b \in Hom_C^*(X_j,X_k)$ is a

5588: nondegenerate pairing. These pairings are graded symmetric, $\leftsc

5589: a,b \rightsc = (-1)^{deg(a)deg(b)} \leftsc b,a \rightsc$.

5590: \item

5591: One can split $Hom_{C}^*(X_j,X_k) = Hom_{C^\rightarrow}^*(X_j,X_k)

5592: \oplus J_{jk}$ in such a way that the product of two morphisms

5593: belonging to the $J$ spaces is zero (in view of the previous

5594: requirements, we have $J_{jk} = 0$ for $j<k$, $J_{jk} =

5595: Hom_C^*(X_j,X_k)$ for $j>k$; and since it is an isotropic subspace,

5596: $J_{jj}$ must be one-dimensional and concentrated in degree $d$).

5597: \end{itemize}

5598: These properties characterize $C$ up to isomorphism. Alternatively,

5599: one can define it explicitly by

5600: \begin{equation} \label{eq:double}

5601:  Hom_C(X_j,X_k) = Hom_{C^\rightarrow}(X_j,X_k)

5602:  \oplus Hom_{C^\rightarrow}(X_k,X_j)^\vee[-d]

5603: \end{equation}

5604: and $(a,a^\vee)(b,b^\vee) = (ab,a^\vee(b\,\cdot) +

5605: (-1)^{deg(a)(deg(b)+deg(b^\vee))} b^\vee(\cdot\, a))$.

5606: 

5607: The example relevant to us as is follows: Let $V$ be a

5608: four-dimensional complex vector space. Consider the (linear graded)

5609: directed category $C^{\rightarrow}_4$ having four objects

5610: $X_1,\dots,X_4$ and morphisms

5611: \[

5612:  Hom_{C^\rightarrow_4}(X_j,X_k) = \begin{cases}

5613:  \Lambda^{k-j}V & j \leq k, \\

5614:  0 & \text{otherwise,} \end{cases}

5615: \]

5616: with the obvious grading and (wedge product) composition. We define

5617: $C_4$ to be the trivial extension category of $C_4^\rightarrow$ in

5618: degree $d = 2$. Its total morphism algebra, linear over $R_4 \iso

5619: \C^4$, will be denoted by $Q_4$. Once one has fixed a choice of

5620: volume element in $\Lambda^4 V^\vee$, the duals in the general

5621: formula \eqref{eq:double} can be removed, so that then

5622: \begin{equation} \label{eq:c4-morphism-revised}

5623:  Hom_{C_4}(X_j,X_k) = \begin{cases}

5624:  \Lambda^{k-j}V & j < k, \\

5625:  \Lambda^0V \oplus (\Lambda^4V)[2], & j = k, \\

5626:  (\Lambda^{k-j+4}V)[2] & j > k.

5627:  \end{cases}

5628: \end{equation}

5629: 

5630: From a different viewpoint, take the exterior algebra $\Lambda V$,

5631: and consider the action of the center of $SL(V)$, $\Gamma_4 = \{\One,

5632: i\One, -\One, -i\One\}$, on it. For explicitness, pick the generator

5633: $\gamma = i\One$. From our general discussion of semidirect products,

5634: we know that $\Lambda V \semidirect \Gamma_4$ is an algebra over $\C

5635: \Gamma_4 \iso R_4$, with a basis of idempotents given by $u_j =

5636: \quarter\chi_j = \quarter(e + i^{-j} \gamma + i^{-2j} \gamma^2 +

5637: i^{-3j} \gamma^3)$ for $j = 1,\dots,4$. Concretely,

5638: \begin{equation} \label{eq:c4-chi}

5639:  u_k (\Lambda V \semidirect \Gamma_4) u_j =

5640:  \begin{cases}

5641:  \Lambda^{k-j} V \otimes \C u_j & j<k, \\

5642:  (\Lambda^0 V \oplus \Lambda^4 V) \otimes \C u_j & j = k, \\

5643:  \Lambda^{k-j+4} V \otimes \C u_j & j>k. \\

5644:  \end{cases}

5645: \end{equation}

5646: So $\Lambda V \semidirect \Gamma_4$ is the total morphism algebra of

5647: a category with four objects. If we view it in this way, the directed

5648: subcategory is isomorphic to $C_4^\rightarrow$. Moreover, after

5649: choosing a volume element on $V$, the resulting integration maps make

5650: $\Lambda V \semidirect \Gamma_4$ into a Frobenius category of degree

5651: 4. One can split the morphism spaces into those of the directed

5652: subcategory and a complementary summand $J$, consisting of those

5653: $\Lambda^jV \otimes \C u_k$ with $j+k > 4$. This satisfies the

5654: properties stated above, so we conclude that $\Lambda V \semidirect

5655: \Gamma_4$ is the trivial extension category of $C_4^\rightarrow$ in

5656: degree 4. Hence there is an isomorphism $\Lambda V \semidirect

5657: \Gamma_4 \iso Q_4$ of algebras, which is $R_4$-linear but only

5658: $\Z/2$-graded.

5659: 

5660: Later on, we will also need some understanding of the automorphisms

5661: of $Q_4$, as a graded algebra over $R_4$. To get the trivial thing

5662: out of the way, the group $I = (\C^*)^4$ acts by (elementary) inner

5663: automorphisms $\rho_I$, with $\rho_I(r_1,\dots,r_4)$ being

5664: multiplication with $r_k/r_j$ on $u_k Q_4 u_j$. Next, the canonical

5665: action of $G = GL(V)$ on $C_4^\rightarrow$ (leaving the objects

5666: fixed, and being the obvious representation on $\Lambda^{k-j}V$)

5667: extends to an action $\rho_G$ on $C_4$, determined by asking that

5668: \begin{equation} \label{eq:det}

5669: \int_{X_j} \rho_G(A)(a) = det(A) \int_{X_j} a.

5670: \end{equation}

5671: With respect to the isomorphism introduced above, this is the same as

5672: the $G$-action on $\Lambda V \semidirect \Gamma_4$ induced by the

5673: natural action on $V$. It is a general feature of semidirect products

5674: that $\Gamma_4 \subset G$ acts by inner automorphisms, more precisely

5675: \[

5676:  \rho_G(i\One) = \rho_I(1,i,i^2,i^3).

5677: \]

5678: This is the only overlap between $\rho_I$ and $\rho_G$, which

5679: together make up the whole automorphism group. To see this, note that

5680: any automorphism of $Q_4$ determines linear automorphisms of the

5681: spaces $u_{j+1}Q_4 u_j \iso V$ for all $j$ (in this case, the index

5682: is cyclic, so we are thinking of $u_1 Q_4 u_4$ as well). Moreover,

5683: the product of two elements for successive $j$ is zero iff they are

5684: linearly dependent in $V$, and this shows that the actions for

5685: various $j$ are by linear maps that differ only by multiplicative

5686: constants. By composing the given automorphism with an inner one,

5687: these constants can be made equal to 1, and then the remaining

5688: automorphism is easily seen to lie in the image of $\rho_G$. To

5689: summarize, we have

5690: \[

5691: \begin{aligned}

5692:  & Inn(Q_4/R_4) \iso (\C^*)^4/\C^*, \\

5693:  & Out(Q_4/R_4) \iso GL(V)/\Gamma_4, \\

5694:  & Aut(Q_4/R_4) \iso (\C^*)^4/\C^* \times_{\Z/4} GL(V).

5695: \end{aligned}

5696: \]

5697: 

5698: At this point, fix a maximal torus $H \subset SL(V)$, and let $T =

5699: H/\Gamma_4$ be its image in $PSL(V)$. Choose a splitting $\Sigma: T

5700: \rightarrow H$ of the projection, and denote by $\rho_T$ the

5701: $T$-action obtained by $\Sigma$ from $\rho_G$ (this should not be

5702: confused with $\rho_I$). Let $\Gamma_{64}$ be the subgroup of those

5703: $A \in H$ such that $A^4 = \One$, and $\Gamma_{16} =

5704: \Gamma_{64}/\Gamma_4$ its image in $T$. We use the restriction of

5705: $\rho_T$ to that finite group to form the semidirect product

5706: \[

5707:  Q_{64} \stackrel{\text{def}}{=} Q_4 \semidirect \Gamma_{16}.

5708: \]

5709: This is an algebra linear over $\C\Gamma_4 \otimes \C\Gamma_{16} =

5710: \C\Gamma_{64}$, so one can also view it as a category $C_{64}$ with

5711: 64 objects. The actions $\rho_T|\Gamma_{16}$ obtained from any two

5712: choices of $s$ differ by homomorphisms $\Gamma_{16} \rightarrow

5713: Inn(Q_4/R_4)$, which can always be lifted to $I$. This implies that

5714: $Q_{64}$ is independent of $\Sigma$ up to ($\C\Gamma_{64}$-linear)

5715: isomorphism, see \eqref{eq:conjugate-action}. Moreover, by combining

5716: our previous discussion with \eqref{eq:in-stages} one finds that

5717: there is a $\Z/2$-graded isomorphism

5718: \[

5719:  Q_{64} \iso (\Lambda V \semidirect \Gamma_4) \semidirect \Gamma_{16}

5720:  \iso \Lambda V \semidirect \Gamma_{64}.

5721: \]

5722: 

5723: \subsection{}

5724: We want to use the interpretation of $Q_4$ as a twisted exterior

5725: algebra to compute its Hochschild cohomology, and that requires

5726: paying careful attention to the different gradings. It is convenient

5727: to temporarily introduce a bigrading indexed by $\half \Z \times

5728: \Z/2$ on $A = \Lambda V \semidirect \Gamma_4$, with elements of

5729: $\Lambda^k V \otimes \C \Gamma_4$ given bidegree $(k/2, [k])$. One

5730: defines the Hochschild cohomology of such bigraded algebras by using

5731: the $\Z/2$-grading for parities and hence the signs in the Hochschild

5732: differential, and the $\half\Z$-grading to determine the degree of

5733: Hochschild cochains, hence the $t$-part of $HH^{s+t}(A,A)^t$. It is

5734: straightforward to adapt the proof of Proposition

5735: \ref{th:twisted-hkr} to the new gradings, so we have

5736: \begin{equation} \label{eq:fractional-hkr}

5737:  HH^{s+t}(A,A)^t \iso

5738:  \bigoplus_{\gamma \in \Gamma_4} \Big( S^s(V^{\gamma,\vee}) \otimes

5739:  \Lambda^{s+2t-codim(V^\gamma)}(V^\gamma) \Big)^{\Gamma_4}.

5740: \end{equation}

5741: Having done that, we modify the bigrading by adding $((k-j)/2,0)$ to

5742: the bidegree of each piece $(1 \otimes u_k)A(1 \otimes u_j)$, and

5743: denote the outcome by $B$. If we view our algebras as categories,

5744: this corresponds to a fractional shift or translation of the $j$-th

5745: object downwards by $j/2$. As a consequence, restriction to the

5746: quasi-isomorphic subcomplex of $R_4$-linear cochains shows that the

5747: Hochschild cohomologies of $A$ and $B$ are the same. On the other

5748: hand, a direct check using \eqref{eq:c4-chi} shows that $B$ is

5749: nontrivial only in bidegrees $(k,[k])$ with $k \in \Z$, so that we

5750: can consider it as an ordinary graded algebra. As such, it becomes

5751: isomorphic to $Q_4$, see \eqref{eq:c4-morphism-revised}; which shows

5752: that \eqref{eq:fractional-hkr} also computes the Hochschild

5753: cohomology of that algebra. In particular, the $\gamma \neq e$

5754: summands only contribute to $(s,t) = (0,2)$.

5755: 

5756: Let $\QQ_4$ be an $R_4$-linear and $\rho_T$-equivariant

5757: $A_\infty$-algebra whose underlying vector space is $Q_4$, which

5758: satisfies \eqref{eq:mu1zero}, and which is nontrivial in the sense of

5759: not being gauge equivalent to the trivial $A_\infty$-structure. As an

5760: aside, note that because the elementary inner automorphisms of $Q_4$

5761: are also automorphisms of $\QQ_4$, $\rho_T$-equivariance is actually

5762: the same as $\rho_G|H$-equivariance, hence independent of $\Sigma$.

5763: We have no need of proving the existence of such an

5764: $A_\infty$-algebra abstractly, since it will be constructed

5765: geometrically later on from coherent sheaves or as a Fukaya category.

5766: The uniqueness statement is:

5767: 

5768: \begin{lemma} \label{th:recognize-q4}

5769: $\QQ_4$ is unique up to ($R_4$-linear, $T$-equivariant)

5770: $A_\infty$-isomorphism.

5771: \end{lemma}

5772: 

5773: \proof We will use the equivariant version of the deformation theory

5774: from Section \ref{sec:deformations}, which was discussed extensively

5775: in Section \ref{sec:group-actions} in the case of finite groups. We

5776: need it for $T$, but that is not a problem since the $T$-action on

5777: all relevant Hochschild cochain spaces splits them into direct

5778: products of subspaces with given weights, which is all we needed.

5779: Recall that the HKR isomorphism is $G$-equivariant. From this and

5780: \eqref{eq:fractional-hkr}, one gets

5781: \[

5782:  \big(HH^2(Q_4,Q_4)^{2-d}\big)^T =

5783:  \big(S^dV^\vee \otimes \Lambda^{4-d}V \big)^H =

5784:  \begin{cases}

5785:  \C & d = 4, \\

5786:  0 & \text{for all other $d>2$}.

5787:  \end{cases}

5788: \]

5789: More explicitly, choose an identification $V = \C^4$ making the

5790: maximal torus $H$ standard. Denote the standard basis of $V$ by

5791: $v_0,\dots,v_3$, and the dual basis by $y_0,\dots,y_3$. Then the only

5792: nontrivial summand consists of multiples of the $H$-invariant

5793: polynomial $y_0y_1y_2y_3$. Since $u_k Q_4 u_j$ is nonzero only in

5794: degrees $\equiv k-j$ mod 2, any odd degree $R_4$-bimodule map

5795: $

5796: Q_4 \otimes_{R_4} \cdots \otimes_{R_4} Q_4 \longrightarrow Q_4

5797: $

5798: is necessarily zero. That applies to all the odd order

5799: $\mu^d_{\QQ_4}$ and in particular to $d = 3$, which shows that

5800: $\QQ_4$ has a well-defined order 4 deformation class \[m^4_{\QQ_4}

5801: \in \big(HH^2(Q_4,Q_4)^{-2}\big)^T.\] This is nonzero because

5802: otherwise, since all the higher obstruction groups vanish, $\QQ_4$

5803: would be gauge equivalent to the trivial $A_\infty$-structure, which

5804: is contrary to assumption. Using the equivariant version of Lemma

5805: \ref{th:versal-1}, we find that versality applies: any other

5806: $A_\infty$-structure with the same properties as $\QQ_4$ is gauge

5807: equivalent to $\epsilon^*\QQ_4$ for some $\epsilon \in \C^*$, and in

5808: particular isomorphic to $\QQ_4$. \qed

5809: 

5810: \begin{remark} \label{th:more-recognize-q4}

5811: One can formulate the previous Lemma as a more general recognition

5812: principle. Suppose that we have an $A_\infty$-algebra $\A$ linear

5813: over $R_4$ and carrying a $T$-action, whose cohomology algebra is

5814: equivariantly isomorphic to $Q_4$. Choose $\lambda$, $\pi$ and $h$ as

5815: in Remark \ref{th:explicit-hpt} (and $R_4$-linearly), so that we can

5816: run the explicit form of the Homological Perturbation Lemma to

5817: produce an $A_\infty$-structure $\B$ on $H(\A)$. This will satisfy

5818: $\mu^3_\B = 0$ for the same reasons as in the proof above. Take

5819: classes $a_4 \in u_1 H(\A) u_4$, $a_3 \in u_4 H(\A) u_3$, $a_2 \in

5820: u_3 H(\A) u_2$, $a_1 \in u_2 H(\A) u_1$, such that the product of any

5821: two successive ones is zero. Plug the $a_k$ into the expression

5822: \eqref{eq:mu4-hpt} for $\mu^4_\B$; if the outcome is nonzero, then

5823: $\A$ is quasi-isomorphic to $\QQ_4$. To see this, note that by the

5824: explicit formula \eqref{eq:reverse-hkr} for the HKR isomorphism, or

5825: rather a slight variation of it for semidirect products,

5826: $\mu^4_\B(a_4,a_3,a_2,a_1) = p(v,v,v,v) \otimes u_1$, where $p \in

5827: S^4V^\vee$ is the polynomial corresponding to the obstruction class

5828: $m^4_\B \in HH^2(Q_4,Q_4)^{-2}$, and $v \in V$ reflects the choice of

5829: the $a_k$. As soon as one knows that $m^4_\B$ is nonzero, Lemma

5830: \ref{th:recognize-q4} does the rest.

5831: \end{remark}

5832: 

5833: \subsection{\label{subsec:u4}}

5834: Since $\QQ_4$ is $\rho_T$-invariant, we can form the semidirect

5835: product $A_\infty$-algebra $\QQ_{64} = \QQ_4 \semidirect

5836: \Gamma_{16}$, whose underlying algebra we denote by $Q_{64}$.

5837: 

5838: \begin{lemma}

5839: The truncated Hochschild cohomology of $\QQ_{64}$ satisfies

5840: \begin{equation} \label{eq:hh12}

5841:  HH^1(\QQ_{64},\QQ_{64})^{\leq 0} \iso \C^3, \quad

5842:  HH^2(\QQ_{64},\QQ_{64})^{\leq 0} \iso \C^7.

5843: \end{equation}

5844: \end{lemma}

5845: 

5846: \proof We use the spectral sequence \eqref{eq:truncated-ss}. The same

5847: trick of passing through an appropriate bigrading which we used to

5848: compute the Hochschild cohomology of $Q_4$ also works for $Q_{64}$,

5849: yielding

5850: \begin{multline} \label{eq:hh-q64}

5851:  HH^{s+t}(Q_{64},Q_{64})^t \iso \\ \iso

5852:  \bigoplus_{\gamma \in \Gamma_{64}} \Big( S^s(V^{\gamma,\vee}) \otimes

5853:  \Lambda^{s+2t-codim(V^\gamma)}(V^\gamma) \otimes

5854:  \Lambda^{codim(V^\gamma)}(V/V^\gamma) \Big)^{\Gamma_{64}}.

5855: \end{multline}

5856: In particular, our spectral sequence satisfies $E_2^{s,t} = 0$ for $s

5857: \leq 0$ or $s + 2t < 0$, which ensures its convergence. Case-by-case

5858: study of the conjugacy classes in $\Gamma_{64}$ shows that the

5859: summands with nontrivial $\gamma$ contribute only to $E^{s,t}_2$ with

5860: $s+t \geq 3$. Since the differentials in the spectral sequence

5861: preserve each summand, we can concentrate on $\gamma = e$. Here is

5862: the relevant part of the $E_2$ term ($S^j\Lambda^k$ is shorthand for

5863: $(S^j V^\vee \otimes \Lambda^k V )^{\Gamma_{64}}$, and the crossed

5864: out summands vanish because the $\Gamma_4$-fixed point sets are

5865: already trivial):

5866: \[

5867: \begin{array}{r|ccccccc}

5868:  & s = 0 & s = 1 & s = 2 & s = 3 & s = 4 & s = 5 & s = 6

5869:  \\[0.25em]

5870:  \hline

5871:  t = 1 &

5872:  0

5873:  & \xout{S^1\Lambda^3}

5874:  & \xout{S^2\Lambda^4}

5875:  & 0

5876:  & 0

5877:  & 0

5878:  & 0

5879:  \\[0.25em]

5880:  t = 0 &

5881:  0

5882:  & S^1\Lambda^1

5883:  & S^2\Lambda^2

5884:  & S^3\Lambda^3

5885:  & S^4\Lambda^4

5886:  & 0

5887:  & 0

5888:  \\[0.25em]

5889:  t = -1 &

5890:  0

5891:  & 0

5892:  & \xout{S^2\Lambda^0}

5893:  & \xout{S^3\Lambda^1}

5894:  & \xout{S^4\Lambda^2}

5895:  & \xout{S^5\Lambda^3}

5896:  & \xout{S^6\Lambda^4}

5897:  \\[0.25em]

5898:  t = -2 &

5899:  0

5900:  & 0

5901:  & 0

5902:  & 0

5903:  & S^4\Lambda^0

5904:  & S^5\Lambda^1

5905:  & S^6\Lambda^2

5906:  \\[0.25em]

5907:  t = -3 &

5908:  0

5909:  & 0

5910:  & 0

5911:  & 0

5912:  & 0

5913:  & 0

5914:  & \xout{S^6\Lambda^0}

5915: \end{array}

5916: \]

5917: The first nonzero differential is $\delta_3$, which is the Schouten

5918: bracket with the order four deformation class. We identify $V = \C^4$

5919: as before, so that the deformation class is $y_0 y_1 y_2 y_3$ (up to

5920: a multiple, which we may safely ignore since it can be absorbed into

5921: a rescaling of $V$). In total degree $s+t = 1$, we have the

5922: $\Gamma_{64}$-invariant part of $V^\vee \otimes V$ which is spanned

5923: by elements $y_k \otimes v_k$, and these satisfy

5924: \begin{equation} \label{eq:d311}

5925:  \delta_3^{1,0}(y_k \otimes v_k) = y_0 y_1 y_2 y_3

5926: \end{equation}

5927: for all $k$. Note that $\half \sum_k y_k \otimes v_k$ is what we

5928: called the Euler element in Section \ref{sec:deformations}, and so

5929: the fact that its image under $\delta_3$ is twice the deformation

5930: class is an instance of the general property

5931: \eqref{eq:euler-element}. For $s+t = 2$ we have firstly the

5932: $\Gamma_{64}$-invariant part of $S^2V^\vee \otimes \Lambda^2V$,

5933: generated by the six elements $y_j y_k \otimes v_j \wedge v_k$ which

5934: satisfy

5935: \[

5936:  \delta_3^{2,0}(y_j y_k \otimes v_j \wedge v_k) =

5937:  (y_0 y_1 y_2 y_3) y_k \otimes v_k -

5938:  (y_0 y_1 y_2 y_3) y_j \otimes v_j;

5939: \]

5940: and secondly the invariant part of $S^4V^*$, whose basis consists of

5941: the $y_k^4$ together with $y_0y_1y_2y_3$, the latter being

5942: annihilated by being in the image of \eqref{eq:d311}. The structure

5943: of the $E_2$ term shows that the differentials of order $d>3$ all

5944: vanish, so our computation is complete. To summarize, explicit

5945: generators of $E_2$ which represent the Hochschild cohomology groups

5946: in \eqref{eq:hh12} are (with some slight redundancy): for $HH^1$,

5947: $y_0 \otimes v_0 - y_1 \otimes v_1$ and its three cyclic permutations

5948: (the sum of all four being zero). And for $HH^2$, $y_0y_1 \otimes v_0

5949: \wedge v_1 + y_12y_2 \otimes v_1 \wedge v_2 + y_2y_0 \otimes v_2

5950: \wedge v_0$ and its cyclic permutations (the sum being again zero),

5951: together with the monomials $y_k^4$, $k = 0,\dots,3$. \qed

5952: 

5953: Consider $U_4 \in Aut(Q_4/R_4)$ of the form $U_4 = \rho_I(r)

5954: \rho_G(A)$, where $Av_k = \lambda_kv_{k+1}$, $Av_3 = \lambda_4 v_0$

5955: with $\lambda_0\lambda_1\lambda_2\lambda_3 = 1$. Slightly more

5956: intrinsically, this class of automorphisms can be characterized by

5957: the following properties:

5958: \begin{itemize}

5959: \item

5960: $U_4$ acts on the second summand in $u_j Q_4 u_j \iso \C \oplus

5961: (\Lambda^4 V)[2]$ by $-1$;

5962: \item

5963: Let $\xi_H: H \rightarrow H$ be the shift $diag(t_0,\dots,t_3)

5964: \mapsto diag(t_1,t_2,t_3,t_0)$, and $\xi_T$ the induced automorphism

5965: of $T$. Then there is a homomorphism $\iota: \Gamma_{16} \rightarrow

5966: I$, such that

5967: \begin{equation} \label{eq:skew-commute}

5968: \rho_T(\xi_T(\gamma)) \circ U_4 = \rho_I(\iota(\gamma)) \circ U_4

5969: \circ \rho_T(\gamma).

5970: \end{equation}

5971: $\iota$ is necessary here because the splitting $\Sigma$ breaks some

5972: natural symmetries: essentially, it measures the difference between

5973: $\xi_H(\Sigma(\gamma))$ and $\Sigma(\xi_T(\gamma))$.

5974: \end{itemize}

5975: This is the situation from \eqref{eq:inner-twist}, so one can use

5976: $\iota$ to lift $U_4$ to an automorphism $U_{64} = U_4 \semidirect

5977: (\xi|\Gamma_{16})$ as in \eqref{eq:twisted-u}. Suppose that $\U_{64}$

5978: is an $A_\infty$-isomorphism from $\QQ_{64}$ to itself whose

5979: underlying algebra automorphism is $U_{64}$, and let $\M_{64}$ be its

5980: graph, which is an $A_\infty$-bimodule over $\QQ_{64}$. Consider a

5981: one-parameter deformation $\QQ_{64,q}$ of $\QQ_{64}$ which is

5982: nontrivial at first order in $q$, and has the property that $\U_{64}$

5983: can be simultaneously deformed to an $A_\infty$-homomorphism

5984: $\U_{64,q}$ from $\QQ_{64,q}$ to itself. The existence of these

5985: structures is again not an issue, but uniqueness is:

5986: 

5987: \begin{lemma} \label{th:recognize-q64}

5988: Up to the pullback action of $End(\Lambda_\N)^\times$ combined with

5989: equivalence of one-parameter deformations, $\QQ_{64,q}$ is unique.

5990: \end{lemma}

5991: 

5992: \proof We use coupled deformation theory for the pair

5993: $(\QQ_{64},\M_{64})$. The same parity issues as in Lemma

5994: \ref{th:recognize-q4} show that $\mu^1$ remains zero during the

5995: deformation to $\QQ_{64,q}$, and that also holds for the differential

5996: $\mu^{0,0}$ of the $A_\infty$-bimodule. Here is the relevant piece of

5997: the long exact sequence \eqref{eq:coupled-sequence-2}, with the

5998: information from Lemma \ref{th:graph} (or rather its analogue for

5999: truncated Hochschild cohomology) already built in:

6000: \begin{multline} \label{eq:dull-sequence}

6001:  \qquad HH^1(\QQ_{64},\QQ_{64})^{\scriptscriptstyle\leq 0} \xrightarrow{1-\U_{64}^*}

6002:  HH^1(\QQ_{64},\QQ_{64})^{\scriptscriptstyle\leq 0} \longrightarrow \\ \longrightarrow

6003:  HH^2(\QQ_{64} \& \M_{64},\QQ_{64} \& \M_{64})^{\scriptscriptstyle\leq 0} \longrightarrow \qquad \qquad \\

6004:  \longrightarrow HH^2(\QQ_{64},\QQ_{64})^{\leq 0} \xrightarrow{1-\U_{64}^*}

6005:  HH^2(\QQ_{64},\QQ_{64})^{\leq 0}.

6006: \end{multline}

6007: At the end of the proof of the previous Lemma, we wrote down four

6008: generators which give an $H$-equivariant isomorphism

6009: $HH^1(\QQ_{64},\QQ_{64})^{\scriptscriptstyle\leq 0} \iso

6010: \C^4/\C(1,1,1,1)$; since the coordinates are permuted cyclically by

6011: $U_{64}$, the first arrow in \eqref{eq:dull-sequence} is an

6012: isomorphism. $HH^2$ requires slightly more care, since two different

6013: pieces of the $E_\infty$ term contribute to it. What the spectral

6014: sequence shows is that there is a short exact sequence with an

6015: endomorphism,

6016: \[

6017: \xymatrix{

6018:  0 \ar[r]

6019:  & {\C^4} \ar[r] \ar[d]

6020:  & HH^2(\QQ_{64},\QQ_{64})^{\scriptscriptstyle\leq 0} \ar[r] \ar[d]^{\U_{64}^*}

6021:  & {\C^4/\C(1,1,1,1)} \ar[r] \ar[d]

6022:  & 0

6023:  \\

6024:  0 \ar[r]

6025:  & {\C^4} \ar[r]

6026:  & HH^2(\QQ_{64},\QQ_{64})^{\scriptscriptstyle\leq 0} \ar[r]

6027:  & {\C^4/\C(1,1,1,1)} \ar[r]

6028:  & 0

6029: }

6030: \]

6031: where both the left and right $\downarrow$ are cyclic permutations of

6032: the coordinates. This implies that the last arrow in

6033: \eqref{eq:dull-sequence} has one-dimensional kernel, hence the

6034: coupled truncated Hochschild cohomology group is one-dimen\-si\-onal.

6035: On the other hand, since $\QQ_{64,q}$ is nontrivial at first order,

6036: the primary deformation class of any pair $(\QQ_{64,q},\M_{64,q})$ in

6037: that group must be nonzero; at which point Lemma \ref{th:versal-4}

6038: takes over. \qed

6039: 

6040: It is maybe not entirely clear from the statement, but the proof of

6041: the Lemma above shows that we were talking about uniqueness of

6042: $\QQ_{64,q}$ with respect to a fixed choice of $U_4$, $\iota$ and

6043: $\U_{64}$. The next results removes the apparent dependence of those

6044: data:

6045: 

6046: \begin{lemma}

6047: Up to isomorphism of $A_\infty$-algebras linear over $\Lambda_\N$,

6048: combined with the action of $End(\Lambda_\N)^\times$, $\QQ_{64,q}$ is

6049: independent of all choices made in its construction.

6050: \end{lemma}

6051: 

6052: \proof It is convenient to take on the choices in reverse order. (1)

6053: Concerning the $A_\infty$-bimodule structure $\M_{64}$ on the fixed

6054: bimodule $M_{64} = Graph(U_{64})$, we know from \eqref{eq:hh-q64} and

6055: classical homological algebra that

6056: \[

6057:  Ext^k(M_{64},M_{64}[1-k]) \iso HH^1(Q_{64},Q_{64})^{1-k} = 0

6058: \]

6059: for all $k \geq 2$, and then Lemma \ref{th:formal-m} implies that

6060: $\M_{64}$ is unique up to isomorphism. (2) Any two choices of $\iota$

6061: differ by a $\chi \in \Gamma_{16}^*$, and that defines an

6062: automorphism of $\QQ_{64}$. Since all the Hochschild cohomology in

6063: \eqref{eq:hh12} lies in the $\Gamma_{16}^*$-invariant subspace, one

6064: can use the standard quasi-isomorphism theorem of abstract

6065: deformation theory \cite[Theorem 2.4]{goldman-millson88} to transform

6066: a given $\QQ_{64,q}$ into an equivalent deformation which is

6067: $\Gamma_{16}^*$-equivariant. Having done that, it is easy to see that

6068: if $\U_{64}$ deforms compatibly to an $\U_{64,q}$ for one choice of

6069: $\iota$, then it does for all the others. (3) Changing $U_4$ by an

6070: inner automorphism does not affect its graph (up to bimodule

6071: isomorphism), so we may assume that $U_4 = \rho_G(A)$. The remaining

6072: freedom is the choice of the $\lambda_k$, but one can relate

6073: different choices by an automorphism of $Q_4$ which rescales the

6074: different coordinates of $V$ in a suitable way. \qed

6075: 

6076: \subsection{}

6077: Take the Beilinson basis of the derived category of coherent sheaves

6078: on $P = {\mathbb P}(V)$, which consists of $F_k =

6079: \Omega^{4-k}(4-k)[4-k]$ for $k = 1,\dots,4$. The $\Omega^k$ are

6080: naturally $PGL(V)$-equivariant sheaves, and if we give $\O(-1)$ the

6081: $GL(V)$-action obtained by embedding it into $\O \otimes V$, the

6082: $\Omega^{4-k}(4-k) = \Omega^{4-k} \otimes \O(-1)^{\otimes k-4}$

6083: become $GL(V)$-equivariant. A straightforward computation, which is

6084: the basis of the argument in \cite{beilinson78}, shows that

6085: 

6086: \begin{lemma} \label{th:beilinson}

6087: The linear graded category with objects $F_1,\dots,F_4$, and where

6088: the morphisms are homomorphisms of all degrees in the derived

6089: category, is $GL(V)$-equiva\-ri\-antly isomorphic to

6090: $C_4^\rightarrow$. \qed

6091: \end{lemma}

6092: 

6093: Let $\iota_0: Y_0 \hookrightarrow P$ be the unique $H$-invariant

6094: quartic hypersurface, in appropriate coordinates $Y_0 =

6095: \{y_0y_1y_2y_3 = 0\}$. Write $E_{0,k} = \iota_0^*F_k$. Repeating an

6096: argument from \cite{seidel-thomas99}, we observe that

6097: \begin{align*}

6098:  Hom^*_{Y_0}(E_{0,j},E_{0,k})

6099:  & = Hom^*_P(F_j,(\iota_0)_*\iota_0^*F_k)

6100:  = Hom^*_P(F_j,\O_{Y_0} \stackrel{{\bf L}}{\otimes} F_k) \\

6101:  & = Hom^*_P(F_j,\{\K_P \otimes F_k \rightarrow \O_P \otimes F_k\})

6102: \end{align*}

6103: sits in a long exact sequence which one can write, using Serre

6104: duality on $P$, as

6105: \[

6106:  \cdots \rightarrow Hom^d_P(F_j,F_k) \stackrel{\iota_0^*}{\longrightarrow}

6107:  Hom^d_{Y_0}(E_{0,j},E_{0,k}) \stackrel{(\iota_0^*)^\vee}\longrightarrow

6108:  Hom^{2-d}_P(F_k,F_j)^\vee \rightarrow \cdots

6109: \]

6110: Since the $F_k$ form an exceptional collection, we have unique

6111: splittings

6112: \begin{equation} \label{eq:doubling}

6113:  Hom^*_{Y_0}(E_{0,j},E_{0,k}) = Hom^*_P(F_j,F_k) \oplus Hom^*_P(F_k,F_j)^\vee[-2].

6114: \end{equation}

6115: $Y_0$ has trivial dualizing sheaf, so Serre duality makes

6116: $D^bCoh(Y_0)$ into a Frobenius category of degree $d = 2$. If we

6117: denote by $J_{jk}$ the second summand in \eqref{eq:doubling}, then

6118: the composition of two morphisms in $J$ spaces is zero for dimension

6119: reasons (these morphisms are $Ext^2$'s of locally free sheaves on

6120: $Y_0$). We have now checked off the properties of a trivial extension

6121: category. Moreover, the induced $GL(V)$-action on

6122: $Hom^*_{Y_0}(E_{0,j},E_{0,k})$ satisfies \eqref{eq:det}, so

6123: 

6124: \begin{lemma}

6125: The linear graded category with objects $E_{0,k}$, and where the

6126: morphisms are maps of arbitrary degree in $D^bCoh(Y_0)$, is

6127: $GL(V)$-equiva\-riantly isomorphic to $C_4$. \qed

6128: \end{lemma}

6129: 

6130: Denote by $S_4$ the total morphism algebra of that subcategory, so

6131: $S_4 \iso Q_4$. This is the cohomology algebra of a dg algebra

6132: $\SS_4$, canonical up to quasi-isomorphism, which can be defined as

6133: total dg algebra of the same objects in the {\v C}ech dg category

6134: $\cech{\SS}(Y_0)$. The next step comes from

6135: \cite{douglas-govindarajan-jayaraman-tomasiello01}:

6136: 

6137: \begin{lemma} \label{th:massey}

6138: $\SS_4$ is not formal, meaning not quasi-isomorphic to its

6139: cohomology.

6140: \end{lemma}

6141: 

6142: \proof The skyscraper sheaf at any point $x = \C v \in P$ admits a

6143: locally free Koszul resolution

6144: \[

6145:  \big\{\Omega^3(3) \longrightarrow \Omega^2(2)

6146:  \longrightarrow \Omega^1(1) \longrightarrow \O\big\} \iso \O_x

6147: \]

6148: where each differential corresponds to $v \in V$ under the

6149: isomorphism from Lemma \ref{th:beilinson}. After restricting, it

6150: follows that there is a spectral sequence converging to

6151: $Hom^*_{Y_0}(\Omega^3(3)[3], L\iota_0^*\O_x)$ whose $E_1$ term is

6152: \[

6153: \begin{array}{r|ccccccc}

6154:  & s = 0 & s = 1 & s = 2 & s = 3 & s = 4

6155:  \\[0.25em]

6156:  \hline

6157:  t = 2 & \C & 0 & 0 & 0 & 0 \\

6158:  t = 1 & 0  & 0 & 0 & 0 & 0 \\

6159:  t = 0 & \C & V & \Lambda^2V & \Lambda^3V & 0

6160: \end{array}

6161: \]

6162: The differential $\delta_1: E_1^{s,t} \rightarrow E_1^{s+1,t}$ is

6163: wedge product with $v$ in the bottom row, which leaves just two

6164: nonzero terms $E_2^{0,2} \iso \C$, $E_2^{3,0} \iso \C$. If $x \notin

6165: Y_0$ then $L\iota_0^*\O_x = 0$, so these two terms must kill each

6166: other through the only remaining differential $\delta_3^{0,2}$. On

6167: the other hand, if $x \in Y_0$ then $L\iota_0^*\O_x = \O_x \oplus

6168: \O_x[1]$ has nonzero sections, so the spectral sequence degenerates

6169: at $E_2$.

6170: 

6171: The connection with the question we are interested in comes about as

6172: follows. If $\SS_4$ was formal, the triangulated subcategory of

6173: $D^bCoh(Y_0)$ generated by the $E_{0,k}$ could be modelled by twisted

6174: complexes defined on the category $C_4$ with its trivial

6175: $A_\infty$-structure. We know that $L\iota_0^*\O_x$ is a twisted

6176: complex of the form $(E_{0,1} \oplus E_{0,2} \oplus E_{0,3} \oplus

6177: E_{0,4},\delta)$, and the behaviour of first differential in the

6178: spectral sequence means that $\delta$ must be (up to irrelevant

6179: scalar multiples)

6180: \[

6181: \delta = \begin{pmatrix}

6182:  0 & & & \\

6183:  v & 0 & & \\

6184:  \ast & v & 0 & \\

6185:  \ast & \ast & v & 0

6186: \end{pmatrix},

6187: \]

6188: But for degree reasons the terms $\ast$ are necessarily equal to

6189: zero. Having identified the object, we can check that in $D^b(C_4)$

6190: the analogue of the spectral sequence above degenerates at $E_2$ for

6191: any nonzero $v$, which is a contradiction. \qed

6192: 

6193: By choosing the {\v C}ech covering to be $H$-invariant, one can

6194: arrange that $\SS_4$ has a natural $H$-action. Take the same

6195: splitting $\Sigma: T \rightarrow H$ as before, and consider the

6196: induced $T$-action. The $T$-equivariant version of the Homological

6197: Perturbation Theorem (which applies because the action on

6198: $hom_{\SS_4}$ is well-behaved), together with Lemma

6199: \ref{th:recognize-q4}, show that $\SS_4$ is

6200: $A_\infty$-quasi-isomorphic to $\QQ_4$, and moreover that

6201: quasi-isomorphism is $R_4$-linear and $T$-equivariant. As a

6202: consequence, the semidirect product $\SS_{64} = \SS_4 \semidirect

6203: \Gamma_{16}$ defined using the restriction of the $T$-action to

6204: $\Gamma_{16}$ is quasi-isomorphic to $\QQ_{64}$. On the geometric

6205: side, consider each $E_{0,k}$ as a $\Gamma_{16}$-equivariant sheaf in

6206: all 16 possible ways (mutually twisted by characters). This gives a

6207: total of 64 objects in the derived category

6208: $D^bCoh_{\Gamma_{16}}(Y_0)$, and Remark \ref{th:equivariant-sheaves}

6209: tells us that $\SS_{64}$ is the dg algebra underlying their total

6210: morphism algebra. The flat formal deformation $Y_q = \{p_q(v) = 0\}$

6211: of $Y_0$ given by \eqref{eq:fermat} still admits a

6212: $\Lambda_\N$-linear $\Gamma_{16}$-action. Since the $E_{0,k}$ are

6213: restricted from projective space, they have obvious equivariant

6214: extensions $E_{q,k}$ to locally free coherent sheaves over $Y_q$. By

6215: the discussion in Section \ref{subsec:formal-schemes}, or rather its

6216: equivariant analogue, this gives rise to a $\Gamma_{16}$-equivariant

6217: one-parameter deformation $\SS_{64,q}$ of $\SS_{64}$, which satisfies

6218: \begin{equation} \label{eq:stable-h}

6219:  H^*(\SS_{64,q}) \iso H^*(\SS_{64}) \otimes_\C \Lambda_\N.

6220: \end{equation}

6221: One can check \eqref{eq:stable-h} either by repeating the duality

6222: argument above for $Y_q$, or more simply by observing that the

6223: morphisms between any two $E_{0,k}$ in $D^bCoh(Y_0)$ are concentrated

6224: in a single degree, which precludes the possibility of any

6225: deformation of the differential affecting the cohomology.

6226: 

6227: \begin{lemma} \label{th:64-nontrivial}

6228: ${\SS}_{64,q}$ is nontrivial at first order in $q$.

6229: \end{lemma}

6230: 

6231: \proof This is quite similar to the previous Lemma. Let $x_q$ be a

6232: $\Lambda_\N$-point of $P \times_\C \Lambda_{\N}$, which is now given

6233: by a family $v_q$ of nonzero vectors in $V$ varying with the formal

6234: parameter $q$. We have the same spectral sequence as before for the

6235: derived restriction of its structure sheaf $\O_{x_q}$ to $\iota_q:

6236: Y_q \hookrightarrow P \times_\C \Lambda_\N$, and therefore

6237: $Hom^*_{Y_q}(\Omega^3(3)[3],L\iota_q^*\O_{x_q})$ is the cohomology of

6238: \[

6239:  \Lambda_\N \xrightarrow{q \mapsto p_q(v_q)} \Lambda_\N.

6240: \]

6241: $L\iota_q^*\O_{x_q}$ is not an equivariant sheaf, but one can remedy

6242: this by summing over the $\Gamma_{16}$-orbit of $x_q$. By taking

6243: $x_q$ in such a way that $p_q(v_q)$ has precisely order $O(q)$, one

6244: obtains examples of twisted complexes over $\SS_{64,q}$ (considered

6245: here as an $A_\infty$-category with 64 objects) such that the map

6246: \eqref{eq:q2} is not an isomorphism. Now apply Lemma

6247: \ref{th:deformation-generators} to $Tw\,\SS_{64,q}$. \qed

6248: 

6249: The automorphism $A \in GL(V)$, $A v_k = v_{k+1}$, $A v_3 = Av_0$

6250: preserves $Y_0$ and the objects $E_{0,k}$, hence defines an

6251: automorphism of $\SS_4$, unique up to elementary inner automorphisms.

6252: Since $det(A) = -1$, this automorphism acts by $-1$ on the space of

6253: sections of the canonical sheaf of $Y_0$, hence by Serre duality on

6254: $Hom^2(E_{0,k},E_{0,k})$ for any $k$. Moreover, it skew-commutes with

6255: the $T$-action on $Y_0$ in the obvious sense, so it can be identified

6256: on the level of cohomology with one of the automorphisms $U_4$ of

6257: $Q_4$ introduced before. Because of the twisted equivariance, one can

6258: define an induced automorphism of $\SS_{64}$, and since $A$ survives

6259: the deformation to $Y_q$, this can be deformed to an automorphism of

6260: $\SS_{64,q}$. Together with the previous Lemma, this allows us to

6261: apply Lemma \ref{th:recognize-q64} which shows that $\SS_{64,q}$ is

6262: quasi-isomorphic to $\QQ_{64,q}$, at least in the sense in which the

6263: latter object is unique, which means up to reparametrization by $\psi

6264: \in End(\Lambda_\N)^\times$.

6265: 

6266: Let $Y_q^* = Y_q \times_{\Lambda_\N} \Lambda_\Q$ be the general fibre

6267: of our formal scheme, and $E_{q,k}^*$ the restrictions of the

6268: $E_{q,k}$ to it. Let $Z_q^*$ be the minimal resolution of the

6269: quotient $Y_q^*/\Gamma_{16}$. We need two more basic facts:

6270: 

6271: \begin{lemma}

6272: The 64 equivariant versions of the $E_{q,k}^*$ are split-generators

6273: for $D^bCoh_{\Gamma_{16}}(Y_q^*)$.

6274: \end{lemma}

6275: 

6276: \proof Beilinson's resolution of the diagonal \cite{beilinson78}

6277: shows that the $F_{q,k}^*$ are generators for the derived category of

6278: $P \times_\C \Lambda_\Q$. His argument carries over to the

6279: equivariant case, which means that the 64 objects obtained by making

6280: the $F_{q,k}^*$ equivariant in all possible ways generate the derived

6281: category of $\Gamma_{16}$-equivariant sheaves. The equivariant

6282: analogue of Lemma \ref{th:kontsevich} shows that the restrictions of

6283: these objects to $Y_q^*$ split-generate its derived category of

6284: equivariant sheaves. \qed

6285: 

6286: \begin{lemma} \label{th:kapranov-vasserot00}

6287: $D^bCoh(Z_q^*) \iso D^bCoh_{\Gamma_{16}}(Y_q^*)$, as triangulated

6288: categories linear over $\Lambda_\Q$.

6289: \end{lemma}

6290: 

6291: This is a theorem of Kapranov and Vasserot

6292: \cite{kapranov-vasserot00}, valid for any finite group action on a

6293: smooth $K3$ surface which preserves the holomorphic two-form. We now

6294: have the situation where we know that

6295: \[

6296: D^bCoh_{\Gamma_{16}}(Y_q^*) \iso D^\pi(\SS(Y_q^*))

6297: \]

6298: is split-generated by certain objects, and where we know that the dg

6299: subcategory of $\SS_{\Gamma_{16}}(Y_q^*)$ consisting of those

6300: objects, denoted by $\SS_{64,q}$ above, is quasi-isomorphic to

6301: $\QQ_{64,q}$. Lemma \ref{th:derived-equivalence} then says that

6302: 

6303: \begin{cor} \label{th:mirror-1}

6304: There is some $\psi \in End(\Lambda_\N)^\times$, such that

6305: $D^bCoh(Z_q^*) \iso D^\pi(\psi^*\QQ_{64,q} \otimes_{\Lambda_\N}

6306: \Lambda_\Q)$. \qed

6307: \end{cor}

6308: 

6309: 

6310: \section{Counting polygons\label{sec:computation}}

6311: 

6312: We begin by showing that for a quartic surface in $\CP{3}$ whose part

6313: at infinity is the intersection with the ``simplex'' $x_0x_1x_2x_3 =

6314: 0$, the deformation from affine to relative Fukaya category is

6315: nontrivial. This is based on Corollary \ref{th:wall-crossing}, but

6316: the construction of the relevant Lagrangian torus involves geometric

6317: ideas related to the SYZ conjecture \cite{ruan02a}, in particular the

6318: appearance of amoebas in that context \cite{ruan02c,mikhalkin01}.

6319: Having done that, the next step is to apply Corollary

6320: \ref{th:generates-everything} to prove that the 64 vanishing cycles

6321: of the Fermat pencil of quartics are split-generators for the derived

6322: Fukaya category. This reduces us to studying the full

6323: $A_\infty$-subcategory formed by these cycles, and finally, the

6324: dimensional induction machinery from Section

6325: \ref{subsec:combinatorial} allows us to compute that subcategory

6326: combinatorially.

6327: 

6328: \subsection{}

6329: Let $X = \CP{3}$ with $o_X = \K_X^{-1} = \O(4)$, and

6330: $\sigma_{X,\infty} = x_0x_1x_2x_3$. Choose some section

6331: $\sigma_{X,0}$ which with $\sigma_{X,\infty}$ generates a

6332: quasi-Lefschetz pencil, and such that $X_0$ is regular.

6333: 

6334: \begin{lemma} \label{th:quartic-deformation}

6335: $\Fuk(X_0,X_{0,\infty})$, as a one-parameter deformation of

6336: $A_\infty$-cate\-go\-ries, is nontrivial at first order in $q$.

6337: \end{lemma}

6338: 

6339: The validity of the statement is independent of the particular choice

6340: of $\sigma_{X,0}$. Any two different choices $\sigma_{X,0}^0$,

6341: $\sigma_{X,0}^1$ can be joined by a smooth family $\sigma^t_{X,0}$,

6342: of sections of $o_X$, such that for all $t$,

6343: \[

6344: X^t_0 = (\sigma^t_{X,0})^{-1}(0)

6345: \]

6346: is a smooth quartic and intersects each stratum of $X_{\infty}$

6347: transversally. For the same reasons as in the proof of Proposition

6348: \ref{th:negativity}, there is a family of symplectic isomorphisms

6349: $\phi^t: X^0_0 \rightarrow X^t_0$ which maps $X^0_0 \cap X_\infty$ to

6350: $X^t_0 \cap X_\infty$. Since quartic surfaces are simply-connected,

6351: one sees from Remark \ref{th:class-of-j-2} that $\phi^1$ defines an

6352: equivalence of relative Fukaya categories.

6353: 

6354: It remains to prove Lemma \ref{th:quartic-deformation} in a single

6355: case. Let $C = \{x_3 = 0\} \iso \CP{2}$ be one of the irreducible

6356: components of $X_\infty$. The intersection with the other components

6357: is $\Delta = C \cap C' \iso \{x_0x_1x_2 = 0\} \subset \CP{2}$. By

6358: construction, $\CP{2}$ carries the Fubini-Study symplectic form

6359: $\o_C$, rescaled in such a way that a line has area $4$. We also have

6360: the meromorphic complex volume form $\eta_C$ with poles along

6361: $\Delta$. Take (this will not quite be our final word) a section of

6362: the form

6363: \begin{equation} \label{eq:prelim-choice}

6364: \sigma_{X,0} = x_0^4 + x_1^4 + \quarter x_2^4 + 2 x_0 x_1 x_2^2 +

6365: \text{(monomials containing $x_3$)}.

6366: \end{equation}

6367: $X_0 \cap C$ is the smooth quartic curve $Q = \{x_0^4 + x_1^4 +

6368: \quarter x_2^4 + 2 x_0 x_1 x_2^2 = 0\} \subset \CP{2}$, which

6369: intersects all the strata of $\Delta$ transversally. Hence, by

6370: suitably choosing the terms not spelled out in

6371: \eqref{eq:prelim-choice}, one can achieve that $\sigma_{X,0}$

6372: generates a quasi-Lefschetz pencil. Consider the family of Lagrangian

6373: torus orbits

6374: \[

6375:  T_\lambda = \{

6376:  \lambda_2 |x_0|^2 = \lambda_0 |x_2|^2, \;

6377:  \lambda_2 |x_1|^2 = \lambda_1 |x_2|^2

6378:  \} \subset \CP{2} \setminus \Delta,

6379: \]

6380: for $\lambda = (\lambda_0,\lambda_1,\lambda_2)$ with $\lambda_k >0$

6381: and $\sum_k \lambda_k = 1$. We take

6382: \[

6383: \textstyle L_0 = T_{\frac{1}{4},\frac{1}{4},\frac{1}{2}},

6384: \]

6385: which by an elementary estimate is disjoint from $Q$. The next

6386: question is how far this satisfies Assumptions \ref{as:l-zero}. Each

6387: $T_\lambda$ is graded with respect to $\eta_C$, and $O(4)|T_\lambda$

6388: is the trivial flat line bundle iff $\lambda_k \in \quarter\Z$ for

6389: all $k$, which is the case for $L_0$. Look at the holomorphic discs

6390: $w: (D,\partial D) \rightarrow (C,L_0)$ given by

6391: \begin{equation} \label{eq:8-discs}

6392:  w(z) = \begin{cases} [\zeta: z: \sqrt{2}], \\

6393:  [z: \zeta: \sqrt{2}] \end{cases}

6394:  \text{for some $\zeta \in \C$ with $\zeta^4 = -1$}.

6395: \end{equation}

6396: They satisfy $\int w^*\o_C = w \cdot Q = 1$, and their boundary

6397: circles generate $H_1(L_0)$. By Lemma \ref{th:check-exactness}, this

6398: implies that $L_0$ is exact with respect to $\theta_C$.

6399: 

6400: \begin{lemma} \label{th:regular-p2}

6401: Every holomorphic disc $w: (D,\partial D) \longrightarrow

6402: (\CP{2},T_\lambda)$ is regular.

6403: \end{lemma}

6404: 

6405: \proof Take the Euler sequence on $\CP{2}$,

6406: \begin{equation} \label{eq:euler}

6407:  0 \longrightarrow \O \longrightarrow \O(1) \otimes H^0(\O(1))^\vee

6408:  \longrightarrow T\CP{2} \longrightarrow 0.

6409: \end{equation}

6410: Choose a basis for $H^0(\O(1))^\vee$ dual to the three coordinate

6411: functions $x_0,x_1,x_2$. Then the first map is just given by these

6412: functions, and the second map takes $i(x_k \otimes x_k^\vee)$ to the

6413: vector field which infinitesimally rotates the $k$-th coordinate. It

6414: follows that over $T_\lambda$, we have a sequence of real subbundles

6415: of \eqref{eq:euler},

6416: \[

6417:  0 \longrightarrow i\R \longrightarrow

6418:  i\R x_0 \oplus i\R x_1 \oplus i\R x_2

6419:  \longrightarrow T(T_\lambda) \longrightarrow 0.

6420: \]

6421: After pulling this back to the disc via $w$ and introducing the

6422: associated Cauchy-Riemann operators, one finds that there is a

6423: surjection

6424: \[

6425:  \coker(D_0 \oplus D_1 \oplus D_2) \twoheadrightarrow \coker\, D_{w,\CP{2}},

6426: \]

6427: where $D_k$ is the $\bar\partial$-operator on $w^*\O(1)$ with totally

6428: real boundary conditions $w^*(i\R x_k)$. The Maslov number of

6429: $(w^*\O(1),w^*(i\R x_k))$ is twice the intersection number of $w$

6430: with the line $\{x_k = 0\}$, hence nonnegative, which means that

6431: $D_k$ has trivial cokernel. \qed

6432: 

6433: \begin{lemma} \label{th:maslov-2}

6434: Up to the $PSL(2,\R)$-action on $D$, any holomorphic disc $w:

6435: (D,\partial D) \rightarrow (\CP{2},T_\lambda)$ with Maslov number 2

6436: is either of the form

6437: \[

6438:  w(z) = [\lambda_0 : \zeta\lambda_1 : \lambda_2 z] \;\; \text{ or }

6439:  \;\; w(z) = [\lambda_0 : \zeta\lambda_1 : \lambda_2 z^{-1}]

6440: \]

6441: for some $|\zeta| \in S^1$, or else the same with the roles of the

6442: coordinates exchanged.

6443: \end{lemma}

6444: 

6445: \proof The virtual dimension of the moduli space of unparametrized

6446: discs is equal to one. By Lemma \ref{th:regular-p2}, this moduli

6447: space is indeed a smooth one-dimensional manifold. Since on the other

6448: hand, $T^2 \subset PU(3)$ acts on it, we find that there must be a

6449: sub-circle of $T^2$ which leaves $w$ invariant as an unparametrized

6450: disc. In other words, after applying an initial $PSL(2,\R)$-action

6451: one can assume that there are weights $d_0,d_1,d_2 \in \Z^3$, not all

6452: of them equal, such that

6453: \[

6454: w(e^{it}z) = \mathrm{diag}(e^{it d_0},e^{it d_1}, e^{it d_2})\, w(z).

6455: \]

6456: By holomorphicity $w(z) =

6457: \mathrm{diag}(z^{d_0},z^{d_1},z^{d_2})\,w(1)$. However, at $z = 0$

6458: our disc may intersect $\Delta$ only with multiplicity one, and that

6459: implies $d = (0,0,\pm 1)$ or else the same with the coordinates

6460: permuted. \qed

6461: 

6462: \begin{lemma} \label{th:8-discs}

6463: Up to the action of $S^1$ on $D$ by rotations, there are precisely

6464: eight holomorphic discs $w: (D,\partial D) \rightarrow (\CP{2},L_0)$

6465: with $w \cdot \Delta = w \cdot Q = 1$ and $w(0) \in \Delta \cap Q$.

6466: Moreover, for each of them $[w|\partial D] \in H_1(L_0)$ is

6467: nontrivial, and the restricted linearized operator

6468: $D\bar\partial_{w,\CP{2}}^{res}$ as in \eqref{eq:restricted-dbar} has

6469: one-dimensional cokernel.

6470: \end{lemma}

6471: 

6472: \proof Because $L_0$ is graded with respect to $\eta_C$, $w \cdot

6473: \Delta = 1$ implies that the Maslov number is 2, so we can apply

6474: Lemma \ref{th:maslov-2}. Out of the possible six choices of weights,

6475: only $d = (1,0,0)$ or $(0,1,0)$ lead to discs with $\int w^*\o_C = w

6476: \cdot Q = 1$. The condition that $w(0) \in \Delta \cap Q$ further

6477: restricts us to the eight discs from \eqref{eq:8-discs}. The

6478: statement about $[w|\partial D]$ is obvious, and that about the

6479: cokernel is a straightforward computation.

6480: % in fact, since we are in a clifford torus like situation,

6481: % the dbar operator is O(0) \oplus O(1), so restricting to a point = 0

6482: % yields precisely the desired cokernel \R on the O(0) factor.

6483: \qed

6484: 

6485: Now perturb the equation defining $Q$ slightly, in such a way that

6486: out of the eight points of $Q \cap \{x_0x_1 = 0\}$, one remains fixed

6487: while the others move in a generic way. Then if we replace $Q$ by its

6488: perturbed version $Q'$ in Lemma \ref{th:8-discs}, only one of the 8

6489: discs persists, which we call $w_0$. Take a corresponding slight

6490: perturbation $\sigma_{X,0}'$ of $\sigma_{X,0}$ which still generates

6491: a quasi-Lefschetz pencil. Then Assumptions \ref{as:l-zero} are

6492: satisfied: we already verified those concerning $L_0$ and $w_0$, and

6493: as for those excluding other pseudo-holomorphic maps, part

6494: \ref{item:no-disc-1} comes out of Lemma \ref{th:8-discs};

6495: \ref{item:no-disc-2} follows from Lemma \ref{th:regular-p2} because a

6496: disc $w$ which does not intersect $\Delta$ would have zero Maslov

6497: number, hence negative virtual dimension; and \ref{item:no-sphere} is

6498: obvious, since the $\o_C$-area of any $v: S^2 \rightarrow \CP{2}$ is

6499: a multiple of 4. An appeal to Corollary \ref{th:wall-crossing-2}

6500: completes the proof of Lemma \ref{th:quartic-deformation}.

6501: 

6502: \subsection{}

6503: As before take $X = \CP{3}$ with $\sigma_{X,\infty}(x) =

6504: x_0x_1x_2x_3$, so that the complement of $X_\infty$ can be written as

6505: a quotient by the action of $\Gamma_4^* = \{\pm \One, \pm i \One\}

6506: \subset SL_4(\C)$,

6507: \begin{equation} \label{eq:m-coordinates}

6508:  M = \{x \in \C^4 \suchthat x_0 x_1 x_2 x_3 = 1\}/\Gamma_4^*.

6509: \end{equation}

6510: We now specify $X_0$ to be the Fermat quartic $\sigma_{X,0} =

6511: \quarter( x_0^4+x_1^4+x_2^4+x_3^4)$. $X_0$ intersects the strata of

6512: $X_\infty$ transversally, and $\pi_M =

6513: \sigma_{X,0}/\sigma_{X,\infty}$ has 64 nondegenerate critical points,

6514: coming in groups of 16 which lie in the same fibre (this is not a

6515: problem, see Remark \ref{th:multiple-fibres}). We take $z_* = 0$ as

6516: our base point, and use the same embedded vanishing path for all

6517: critical points which lie in a given fibre. This yields a collection

6518: of 64 vanishing cycles in $M_0 = X_0 \setminus X_{0,\infty}$, divided

6519: into four groups of 16 mutually disjoint ones. We make them into

6520: branes by choosing gradings and, for the projective and relative

6521: versions of that notion, the other additional data. Notation:

6522: $\Fuk_{64}$ will be the full $A_\infty$-subcategory of $\Fuk(M_0)$

6523: whose objects are these vanishing cycles; and $\Fuk_{64,q}$,

6524: $\Fuk_{64,q}^*$ the corresponding $A_\infty$-subcategories of

6525: $\Fuk(X_0)$, $\Fuk(X_0,X_{0,\infty})$. From Propositions

6526: \ref{th:q-zero} and \ref{th:invert-q} we know that $\Fuk_{64,q}$ is a

6527: one-parameter deformation of $\Fuk_{64}$, and $\Fuk_{64,q}^* \iso

6528: \Fuk_{64,q} \otimes_{\Lambda_\N} \Lambda_\Q$.

6529: 

6530: \begin{lemma} \label{th:f64-generates}

6531: $D^\pi\Fuk(X_0) \iso D^\pi(\Fuk_{64,q}^*)$.

6532: \end{lemma}

6533: 

6534: \proof Corollary \ref{th:generates-everything} shows that the

6535: vanishing cycles are split-generators for $D^\pi\Fuk(X_0)$, and Lemma

6536: \ref{th:generating-subcategory} yields the desired conclusion. \qed

6537: 

6538: \begin{lemma} \label{th:f64-nontrivial}

6539: $\Fuk_{64,q}$, seen as a one-parameter deformation of $\Fuk_{64}$, is

6540: nontrivial at first order.

6541: \end{lemma}

6542: 

6543: \proof We know from Lemma \ref{th:quartic-deformation} that

6544: nontriviality is true for the whole of $\Fuk(X_0,X_{0,\infty})$. A

6545: closer look at the argument, and in particular at the proof of

6546: Corollary \ref{th:wall-crossing}, reveals that we have in fact

6547: established the assumptions of Lemma \ref{th:deformation-generators},

6548: which shows that the deformation remains nontrivial when restricted

6549: to any subcategory which has enough objects to split-generate the

6550: whole. \qed

6551: 

6552: To simplify the geometry of our pencil, consider the action of the

6553: group $\Gamma_{16}^* \subset PSL_4(\C)$ of diagonal matrices $A$

6554: satisfying $A^4 = \One$ on $X$. This action is free on $M$; it can be

6555: lifted to $o_X$ and preserves the Fubini-Study connection, as well as

6556: $\sigma_{X,0},\sigma_{X,\infty}$; and it is compatible with the

6557: isomorphism $o_X \iso \K_X^{-1}$. Hence, on the quotient $\bar{M} =

6558: M/\Gamma_{16}^*$ we have an induced symplectic form, a one-form

6559: primitive for it, a holomorphic volume form, and a function

6560: $\pi_{\bar{M}}: \bar{M} \rightarrow \C$. Explicitly,

6561: \[

6562:  \bar{M} = \{ u_0 u_1 u_2 u_3 = 1\}, \quad

6563:  \pi_{\bar{M}}(u) = \quarter(u_0 + u_1 + u_2 + u_3)

6564: \]

6565: where the coordinates are related to the previous ones by $u_k =

6566: x_k^4$. Since $\Gamma_{16}^*$ permutes the critical points in any

6567: given fibre transitively, $\pi_{\bar{M}}$ has one nondegenerate

6568: critical point in each each of the four singular fibres. In other

6569: words, the 64 vanishing cycles in $M_0$ are all possible lifts of the

6570: corresponding 4 vanishing cycles in the quotient $\bar{M}_0$. We need

6571: some basic information about the topology of this space:

6572: \begin{equation} \label{eq:pi12}

6573: \begin{split}

6574:  & \pi_1(\bar{M}_0) = \{ h \in \Z^4 \suchthat h_0 + \dots + h_3 = 0\}, \\

6575:  & im(\pi_2(\bar{M}_0) \otimes \Q \rightarrow H_2(\bar{M}_0;\Q)) \iso \Q^3

6576: \end{split}

6577: \end{equation}

6578: where the first group maps isomorphically to $\pi_1(\bar{M})$, with

6579: the generators $h_k$ becoming loops around the coordinates axes $u_k

6580: = 0$ at infinity; and the second is generated by the 4 vanishing

6581: cycles. Both things follow from elementary Morse and Lefschetz

6582: theory, since $\bar{M} \htp T^3$ is obtained by attaching four

6583: 3-cells to $\bar{M}_0$, which is itself homotopy equivalent to a

6584: 2-complex. As explained in Section \ref{subsec:coverings}, this means

6585: that we have an action of $\pi_1(\bar{M}_0)^*$ on a subcategory of

6586: $\Fuk(\bar{M}_0)$ which consists of branes with a certain condition

6587: on their fundamental groups, and in particular contains our vanishing

6588: cycles since they are spheres. Recall that in Section

6589: \ref{subsec:q-preliminary} we considered the group $T = H/\Gamma_4

6590: \subset PSL(V)$. If one chooses a basis of $V$ making the maximal

6591: torus $H \subset SL(V)$ standard, $T$ can be written as

6592: $(\C^*)^4/\C^*$ (the quotient of the group of all diagonal matrices

6593: by the multiples of the identity) and hence can be identified with

6594: $\pi_1(\bar{M}_0)^*$. The following is our main computational result,

6595: and will be proved later:

6596: 

6597: \begin{prop} \label{th:computation}

6598: For a suitable choice of the lifts to $M_0$ which define the

6599: $T$-action, full $A_\infty$-subcategory $\Fuk_4 \subset

6600: \Fuk(\bar{M}_0)$ whose objects are the 4 vanishing cycles of

6601: $\pi_{\bar{M}}$ is $T$-equivariantly quasi-isomorphic to $\QQ_4$.

6602: \end{prop}

6603: 

6604: The covering $M_0 \rightarrow \bar{M}_0$ provides an epimorphism

6605: $\pi_1(\bar{M}_0) \rightarrow \Gamma_{16}^*$, and the dual of that is

6606: the inclusion of the subgroup $\Gamma_{16} \hookrightarrow T$

6607: considered in Section \ref{subsec:q-preliminary}. Combining the

6608: statement above with Lemma \ref{th:semidirect-lift}, one finds that

6609: \[

6610:  \Fuk_{64} \iso \Fuk_4 \semidirect \Gamma_{16} \iso

6611:  \QQ_4 \semidirect \Gamma_{16} \iso \QQ_{64}.

6612: \]

6613: Since each morphism space $u_i Q_{64} u_j$ is concentrated in a

6614: single degree, the one-parameter deformation $\Fuk_{64,q}$

6615: necessarily satisfies $H^*(\Fuk_{64,q}) \iso H^*(\Fuk_{64})

6616: \otimes_\C \Lambda_\N$, so we can use the Perturbation Lemma to find

6617: a quasi-isomorphic deformation of $\QQ_{64}$. To complete the

6618: argument, consider the symplectic automorphism $\bar\phi$ of

6619: $\bar{M}$ given by $\bar\phi(u_0,u_1,u_2,u_3) = (u_1,u_2,u_3,u_0)$.

6620: This acts fibrewise with respect to $\pi_{\bar{M}}$, hence takes each

6621: vanishing cycle to itself, but it reverses their orientation since it

6622: acts with determinant $-1$ on the tangent space to each critical

6623: point. If we equip $\bar\phi_0 = \bar\phi|\bar{M}_0$ with a suitable

6624: ``odd'' grading $\alphagr_{\bar\phi_0}$, it will map each vanishing

6625: cycle to itself as a graded Lagrangian submanifold. Since it

6626: preserves the complex structure, it is no problem to make

6627: $\bar\phi_0$ act on the Fukaya category, yielding an

6628: $A_\infty$-homomorphism $W_4: \Fuk_4 \rightarrow \Fuk_4$ which acts

6629: as $-1$ on $HF^2(V,V)$ for each of the four vanishing cycles. This

6630: also skew-commutes with the $T$-action up to an automorphism $\xi$ of

6631: $T$, which is simply the dual of the action of $\bar\phi$ on

6632: $\pi_1(\bar{M})$, hence which permutes the coordinates cyclically.

6633: Under the isomorphism $H(\Fuk_4) \iso Q_4$, we can therefore identify

6634: $H(W_4)$ with one of the automorphisms $U_4$ considered in Section

6635: \ref{subsec:u4}. $\bar\phi_0$ lifts to an automorphism of $M_0$,

6636: $\phi_0(x_0,\dots,x_3) = (x_1,x_2,x_3,x_0)$, and this extends to the

6637: compactification $X_0$. This means that $W_{64} = W_4 \semidirect

6638: (\xi|\Gamma_{16})$ which can be identified with the action of

6639: $\phi_0$ on $\Fuk_{64}$, admits a deformation $W_{64,q}$ to an

6640: $A_\infty$-homomorphism from $\Fuk_{64,q}$ to itself. Together with

6641: Lemma \ref{th:f64-nontrivial}, this means that we have verified all

6642: the conditions of Lemma \ref{th:recognize-q64}, and therefore can

6643: conclude that $\Fuk_{64,q}$ is quasi-isomorphic to $\QQ_{64,q}$ up to

6644: some change of parameter in $End(\Lambda_\N)^\times$. Using Lemma

6645: \ref{th:f64-generates}, we obtain:

6646: 

6647: \begin{cor} \label{th:mirror-2}

6648: There is a $\psi \in End(\Lambda_\N)^\times$ such that

6649: \[ D^\pi\Fuk(X_0) \iso D^\pi(\Fuk_{64,q}^*) \iso

6650: \psi^*D^\pi(\QQ_{64,q} \otimes_{\Lambda_\N} \Lambda_\Q).\qed \]

6651: \end{cor}

6652: 

6653: This is the mirror statement to Corollary \ref{th:mirror-1}, and by

6654: comparing the two, Theorem \ref{th:main} follows.

6655: 

6656: 

6657: \subsection{}

6658: It remains to explain Proposition \ref{th:computation}. Following the

6659: strategy from Section \ref{subsec:braid-monodromy}, the first step in

6660: understanding the geometry of the vanishing cycles is to represent

6661: them as matching cycles, which involves the choice of a generic

6662: auxiliary section. Let us momentarily work ``upstairs'' in $M$. An

6663: ansatz for the section is

6664: \begin{equation} \label{eq:ansatz}

6665:  \sigma_{X,0}' = \quarter(\xi_0 x_0^4 + \xi_1 x_1^4 + \xi_2 x_2^4 + \xi_3 x_3^4)

6666: \end{equation}

6667: with pairwise distinct constants $\xi_k \in \C^*$. This always

6668: satisfies the first of the genericity assumptions, namely

6669: $(\sigma_{X,0}')^{-1}(0)$ intersects each stratum of $X_{0,\infty}$

6670: transversally. The critical point set of the associated map

6671: \eqref{eq:bm-map} is always smooth,

6672: \[

6673:  Crit(b_M) = \Big\{ x_k^4 = (s+\xi_k t)^{-1},

6674:  \text{ where } \prod_k (s+t \xi_k) = 1 \Big\}.

6675: \]

6676: The parameter value $t = 0$ corresponds precisely to the 64 critical

6677: points of $\pi_M$. At those points, $b_M: Crit(b_M) \rightarrow \C^2$

6678: is an immersion, and its second component

6679: $\sigma'_{X,0}/\sigma_{X,\infty} : Crit(b_M) \rightarrow \C$ a simple

6680: branched cover, iff the coefficients $\xi_k$ satisfy

6681: \begin{equation} \label{eq:nondegeneracy}

6682:  \sum_k \xi_k^2 \neq \quarter \Big(\sum_k \xi_k\Big)^2.

6683: \end{equation}

6684: Next, $b_M|Crit(b_M)$ cannot be a generic embedding because it is

6685: $\Gamma_{16}^*$-invariant, but after passing to the quotient we do

6686: get a generic embedding $b_{\bar{M}}: Crit(b_{\bar{M}}) \rightarrow

6687: \C^2$, for elementary reasons: the image $\bar{C}$ of this map is an

6688: irreducible algebraic curve in $\C^2$ having four distinct points at

6689: infinity, and since $Crit(b_{\bar{M}})$ has Euler characteristic -2,

6690: the map cannot be a multiple branch cover. The upshot is that the

6691: dimensional induction technique from Section \ref{subsec:induction}

6692: can be applied ``downstairs'', meaning to the vanishing cycles of

6693: $\pi_{\bar{M}}$.

6694: 

6695: Let's get down to concrete numbers. Our choice of weights is $\xi_0 =

6696: 5/4$, $\xi_1 = i$, $\xi_2 = -1$, $\xi_3 = -i$, which satisfies

6697: \eqref{eq:nondegeneracy}. The branch curve (determined by computer

6698: using elimination theory) is

6699: \[

6700: \begin{split}

6701:  \bar{C} = \{ & 128000 z^{12} - 76800 z^{11}w

6702:  + 92160 z^{10}w^2 - 108544 z^9w^3 - 258048 z^8w^4 \\ & + 86016z^7w^5

6703:  - 112640 z^6w^6 + 141312 z^5w^7 + 211968 z^4w^8 - 25600z^3w^9 \\

6704:  & + 36864 z^2 w^{10} - 49152 zw^{11} - 65536 w^{12}

6705:  - 391307 z^8 + 180632 z^7w \\ & + 1940 z^6w^2 - 503112 z^5w^3

6706:  - 2102866 z^4w^4 + 457512 z^3w^5 - 122764 z^2w^6 \\ & - 102904 zw^7

6707:  - 251483 w^8 + 399430 z^4 - 106312 z^3w - 155028z^2w^2 \\ & - 53752 zw^3

6708:  - 320858 w^4 - 136161 = 0 \}.

6709: \end{split}

6710: \]

6711: Figure \ref{fig:paths1} shows the $z$-plane $\C$ with the 4 points of

6712: $Critv(\pi_{\bar{M}}) = \{\pm 1, \pm i\}$, as well the 40 points of

6713: $Fakev(\pi_{\bar{M}})$. Luckily, none of the latter lie on $[-1;1]$

6714: or $i[-1;1]$, so we can take the basis of embedded critical paths

6715: which are straight lines $b_k(t) = i^{2-k}t$, $k = 1,\dots,4$. Denote

6716: by $V_{b_k} \subset \bar{M}_0$ the resulting vanishing cycles. We now

6717: turn to the function

6718: \begin{equation} \label{eq:q0}

6719: q_{\bar{M}_0} = \quarter \sum_k \xi_k u_k: \bar{M}_0 =

6720: \pi_{\bar{M}}^{-1}(0) \longrightarrow \C,

6721: \end{equation}

6722: which is also a quasi-Lefschetz fibration (or more accurately, the

6723: $\Gamma_{16}^*$-quotient of such a fibration on $M_0$). Figure

6724: \ref{fig:paths2} shows its basis $\C$ with the 12 critical values,

6725: which are also the points of $\bar{C} \cap \{z = 0\}$, and the

6726: matching paths $d_k$ obtained from the braid monodromy of $\bar{C}$

6727: over $b_k$. The associated matching cycles are our vanishing cycles:

6728: $V_{b_k} \htp \Sigma_{d_k}$.

6729: 

6730: Taking $0$ as a base point, choose a base path $c_*$ and a basis of

6731: vanishing paths $c_1,\dots,c_{12}$ for \eqref{eq:q0} as indicated in

6732: Figure \ref{fig:paths3}. This gives rise to vanishing cycles $V_{c_k}

6733: \subset \bar{M}_{0,0} = q_{\bar{M}_0}^{-1}(0)$, which are simple

6734: closed loops on a four-pointed genus three Riemann surface.

6735: 

6736: \begin{remark}

6737: If one took $\xi_k = i^k$ instead, condition \eqref{eq:nondegeneracy}

6738: would be violated but the structure of $q_{\bar{M}_0}$ would be much

6739: simpler: it would have 4 degenerate critical points of type $A_3$.

6740: The actually chosen $\xi_k$ are a small perturbation of these values,

6741: in the sense that one can still recognize the groups of 3 critical

6742: points obtained by ``morsifying'' the degenerate singularities. This

6743: allows one to choose the vanishing paths $c_k$ in such a way as to

6744: minimize the intersections between the vanishing cycles in each

6745: group.

6746: \end{remark}

6747: 

6748: Let $\Fuk^\rightarrow_{12}$ be the directed $A_\infty$-subcategory of

6749: the Fukaya category of $\bar{M}_{0,0}$ whose objects are the

6750: $V_{c_k}$. From the restriction of the universal cover of

6751: $\bar{M}_0$, we get an action of $\pi_1(\bar{M}_0)^* \iso T$ on this

6752: $A_\infty$-category. The equivariant version of Theorem

6753: \ref{th:induction} shows that $\Fuk_4$ is equivariantly

6754: quasi-isomorphic to the full $A_\infty$-subcategory of

6755: $Tw_T(\Fuk_{12}^\rightarrow)$ whose objects are the following four

6756: equivariant twisted complexes, obtained by applying the rules from

6757: Section \ref{subsec:induction} to the paths from Figures

6758: \ref{fig:paths2}, \ref{fig:paths3}:

6759: \begin{equation} \label{eq:s-objects}

6760: \begin{split}

6761:  &

6762:  S_1 = \big\{ T'_{V_{c_7}}T'_{V_{c_5}}(V_{c_3})

6763:  \longrightarrow T_{V_{c_{10}}}T_{V_{c_{11}}}(V_{c_{12}}) \big\}, \\

6764:  &

6765:  S_2 = \big\{ T_{V_{c_1}}T_{V_{c_2}}(V_{c_3})

6766:  \longrightarrow T'_{V_{c_{10}}}T'_{V_{c_8}}(V_{c_6}) \big\}, \\

6767:  &

6768:  S_3 = \big\{ T_{V_{c_1}}T_{V_{c_4}}T_{V_{c_5}}(V_{c_6})

6769:  \longrightarrow T_{V_{c_{11}}}'(V_{c_9}) \big\}, \\

6770:  &

6771:  S_4 = \big\{ T_{V_{c_2}}T_{V_{c_4}}T_{V_{c_7}}T_{V_{c_8}}(V_{c_9})

6772:  \longrightarrow V_{c_{12}} \big\}

6773: \end{split}

6774: \end{equation}

6775: where $\{\cdots \rightarrow \cdots\}$ is the cone over the lowest

6776: degree nonzero morphism, which is unique up to a $\C^*$ multiple.

6777: Actually, that morphism may not have degree zero, or it may not be

6778: $T$-invariant, so one needs to change one of the two objects forming

6779: the cone by a translation and tensoring with a suitable

6780: $T$-character. In that sense, the formulae given \eqref{eq:s-objects}

6781: are not entirely precise, but the appropriate corrections can not yet

6782: be determined at this stage in the computation.

6783: 

6784: The next step is to determine $\Fuk^\rightarrow_{12}$, or rather a

6785: directed $A_\infty$-category $\Cat^\rightarrow_{12}$ quasi-isomorphic

6786: to it, using the combinatorial description of directed Fukaya

6787: categories on Riemann surfaces. Following the procedure described in

6788: Section \ref{subsec:combinatorial}, we take a compact piece $N

6789: \subset \bar{M}_{0,0}$ which is a surface with boundary and a

6790: deformation retract. Figure \ref{fig:curve1} shows a decomposition of

6791: $N$ into eight polygons (I)--(VIII), with the solid lines the actual

6792: $\partial N$, and the dashed lines to be glued to the boundaries of

6793: other polygons as indicated. $\bar{M}_{0,0}$ inherits the structure

6794: of an affine Calabi-Yau from $\bar{M}$, and Figure \ref{fig:curve2}

6795: shows an unoriented foliation on $N$ lying in the same homotopy class

6796: as the distinguished trivialization of the squared canonical bundle.

6797: Moreover, restriction

6798: \[

6799: \Z^4/\Z(1,1,1,1) \iso H^1(\bar{M}) \longrightarrow H^1(\bar{M}_{0,0})

6800: \]

6801: gives four canonical elements of $H^1(N)$ whose sum is zero; the

6802: Poincar{\'e} duals of the first three elements are represented by the

6803: one-cycles $x,y,z$ in $(N,\partial N)$ shown in Figure

6804: \ref{fig:curve3}. Figures \ref{fig:vanishing1}--\ref{fig:vanishing8}

6805: show simple closed curves $\nu_k$ isotopic to the vanishing cycles

6806: $V_{c_k}$, drawn on each of the eight polygons which make up $N$.

6807: 

6808: \begin{remark}

6809: These pictures were arrived at by an application of braid monodromy

6810: (in one dimension lower than before), which means that we introduce

6811: another auxiliary holomorphic function on $\bar{M}_0$, whose

6812: restriction to any generic fibre of $q_{\bar{M}_0}$ represents that

6813: surface as a branched cover with only simple branch points. The

6814: function we use is simply a projection

6815: \[

6816:  r_{\bar{M}_0}(u) = u_2 \in \C^*,

6817: \]

6818: making $\bar{M}_{0,0}$ into a fourfold branched cover of $\C^*$ with

6819: eight simple branch points. Cutting the base of that into suitable

6820: pieces gives rise to our decomposition of $N$ into polygons.

6821: Elimination theory is used to compute the branch curve of

6822: \[

6823: (q_{\bar{M}_0},r_{\bar{M}_0}): \bar{M}_{0,0} \longrightarrow \C

6824: \times \C^*,

6825: \]

6826: and from that one obtains the vanishing cycles

6827: $V_{c_1},\dots,V_{c_{12}}$. We have omitted the details of this step,

6828: since it is essentially topological, and completely parallel to the

6829: procedure which led to the paths $b_1,\dots,b_4$ in Figure

6830: \ref{fig:paths2}.

6831: \end{remark}

6832: 

6833: When drawing the $\nu_k$, there is some freedom in their relative

6834: position with respect to each other. The only constraint, which comes

6835: from exactness, says that we should be able to assign positive areas

6836: to the connected components of $N \setminus \bigcup_k \nu_k$ such

6837: that the two-chains which represent relations between the $\nu_k$

6838: have signed area zero. In fact, as explained in Remark

6839: \ref{th:painting-by-numbers}, not all relations need to be

6840: considered; in our case \eqref{eq:pi12} shows that there are three

6841: essential ones. These can be read off from \eqref{eq:s-objects} and

6842: the intersection numbers of the $\nu_k$, and are

6843: \begin{equation}

6844: \begin{split}

6845:  & \nu_5 + \nu_8 + \nu_{10} - \nu_1 - \nu_2 - \nu_3 \sim 0, \\

6846:  & \nu_3 + \nu_5 + \nu_7 - \nu_{10} - \nu_{11} - \nu_{12} \sim 0, \\

6847:  & \nu_8 + \nu_2 + \nu_4 - \nu_7 - \nu_8 - \nu_0 \sim 0.

6848: \end{split}

6849: \end{equation}

6850: After having drawn the relevant two-chains explicitly in Figures

6851: \ref{fig:vanishing1}--\ref{fig:vanishing8}, one realizes that for

6852: each of them, there are pieces of $N \setminus \bigcup_k \nu_k$ which

6853: appear in it with either sign, and which do not appear in the other

6854: two-chains. This means that the configuration of curves $\nu_k$ can

6855: indeed be used as basis for the computation.

6856: 

6857: Choose gradings and $Spin$ structures for the $\nu_k$, lifts to the

6858: abelian covering of $N$ induced from $\pi_1(N) \iso

6859: \pi_1(\bar{M}_{0,0}) \rightarrow \pi_1(\bar{M}_0)$, and

6860: identifications $\C_x \iso \C$ for each intersection point of $\nu_k

6861: \cap \nu_l$ with $k<l$. One can then list all intersection points

6862: with their (Maslov index) degrees as well as their ($T$-action)

6863: weights. This is done in Tables \ref{table:intersections1} and

6864: \ref{table:intersections2}, where the notation is that

6865: \begin{equation} \label{eq:abcd}

6866:  m^d x^a y^b z^c

6867: \end{equation}

6868: is an intersection point with Maslov index $d$, and on which

6869: $[x,y,z,1] \in T$ acts as $x^ay^bz^c$, respectively. By inspecting

6870: the table, one sees that different intersection points of the same

6871: $\nu_k \cap \nu_l$ are always distinguished by $(a,b,c,d)$. We may

6872: therefore denote the generator of

6873: $hom_{\Cat^\rightarrow_{12}}(\nu_k,\nu_l)$ given by an intersection

6874: point \eqref{eq:abcd} by

6875: \[

6876:  \nu_k \cap \nu_l: m^d x^a y^b z^c.

6877: \]

6878: The next step is to count all immersed holomorphic polygons which

6879: have sides on the $\nu_k$ (in the correct order), with the

6880: appropriate signs. The outcome of that are the composition maps

6881: $\mu^d$ of $\Cat^{\rightarrow}_{12}$. It turns out that only

6882: $\mu^2,\mu^3$ are nonzero; their coefficients are listed in Tables

6883: \ref{table:mu2} and \ref{table:mu3}.

6884: 

6885: \begin{remark}

6886: The $T$-action is actually a big bonus in this computation, since the

6887: fact that the $\mu^d$ have to be homogeneous means that a lot of

6888: their coefficients have to vanish. For instance, the fact that $\mu^d

6889: = 0$ for $d > 3$ follows by inspection of the list of $T$-weights of

6890: the intersection points, so one does not need to verify geometrically

6891: that there are no immersed pentagons, hexagons, and so forth.

6892: \end{remark}

6893: 

6894: At this point, we can write down the objects $S_k$ from

6895: \eqref{eq:s-objects}, or rather the corresponding objects $C_k$ in

6896: $Tw_T(\Cat^{\rightarrow}_{12})$, more explicitly. They are

6897: \begin{align*}

6898:  &

6899:  C_1 = \nu_3[1]\twist{x} \oplus \nu_5\twist{y} \oplus \nu_7\twist{yz}

6900:  \oplus \nu_{10}\twist{x^{-1}y} \oplus \nu_{11}[-1]\twist{y} \oplus

6901:  \nu_{12}[-1], \\

6902:  & \delta_1 \!=\!

6903:  {\arraycolsep2.5pt \begin{pmatrix}

6904:  0 & 0 & 0 & 0 & 0 & 0 \\

6905:  \nu_3 \cap \nu_5\!:\!x^{-1}y \!\! & 0 & 0 & 0 & 0 & 0 \\

6906:  0 & \!\! \nu_5 \cap \nu_7\!:\!mz & 0 & 0 & 0 & 0 \\

6907:  0 & \!\! \nu_5 \cap \nu_{10}\!:\!mx^{-1} & 0 & 0 & 0 & 0 \\

6908:  0 & 0 & \nu_7 \cap \nu_{11}\!:\!z^{-1} & \nu_{10} \cap \nu_{11}\!:\!x & 0 & 0 \\

6909:  0 & 0 & 0 & 0 & \!\!\!\! \nu_{11} \cap \nu_{12}\!:\!my^{-1} & 0

6910:  \end{pmatrix}},

6911:  \\

6912:  &

6913:  C_2 = \nu_1[1]\twist{x^{-1}y} \oplus \nu_2\twist{y} \oplus \nu_3

6914:  \oplus \nu_6\twist{z^{-1}} \oplus \nu_8[-1]\twist{z^{-1}}

6915:  \oplus \nu_{10}[-1]\twist{x^{-1}}, \\

6916:  & \delta_2 =

6917:  {\arraycolsep2.5pt \begin{pmatrix}

6918:  0 & 0 & 0 & 0 & 0 & 0 \\

6919:  \nu_1 \cap \nu_2\!:\!x\!\! & 0 & 0 & 0 & 0 & 0 \\

6920:  0 & \nu_2 \cap \nu_3\!:\!my^{-1} & 0 & 0 & 0 & 0 \\

6921:  0 & \nu_2 \cap \nu_6\!:\!my^{-1}z^{-1} & 0 & 0 & 0 & 0 \\

6922:  0 & 0 & \!\!\!\!\!\! -(\nu_3 \cap \nu_8\!:\!z^{-1}) & \nu_6 \cap \nu_8\!:\!1 & 0 & 0 \\

6923:  0 & 0 & 0 & 0 & \!\! \nu_{8} \cap \nu_{10}\!:\!mx^{-1}z & 0

6924:  \end{pmatrix}},

6925:  \\

6926:  &

6927:  C_3 = \nu_1\twist{x^{-1}} \oplus \nu_4\twist{x^{-1}y} \oplus

6928:  \nu_5[-1]\twist{x^{-1}y} \oplus \nu_6[-1]\twist{x^{-1}z^{-1}} \\

6929:  & \qquad \qquad

6930:  \oplus \nu_9[-1]\twist{y^{-1}z^{-1}} \oplus \nu_{11}[-2]\twist{z^{-1}}, \\

6931:  & \delta_3 =

6932:  {\arraycolsep2.5pt \begin{pmatrix}

6933:  0 & 0 & 0 & 0 & 0 & 0 \\

6934:  0 & 0 & 0 & 0 & 0 & 0 \\

6935:  \nu_1 \cap \nu_5\!:\!y & \nu_4 \cap \nu_5\!:\!1 & 0 & 0 & 0 & 0 \\

6936:  0 & 0 & \nu_5 \cap \nu_6\!:\!my^{-1}z^{-1} & 0 & 0 & 0 \\

6937:  0 & 0 & \!\!\nu_5 \cap \nu_9\!:\!mxy^{-2}z^{-1}\!\! & 0 & 0 & 0 \\

6938:  0 & 0 & 0 & \nu_6 \cap \nu_{11}\!:\!x & \nu_9 \cap \nu_{11}\!:\!y & 0

6939:  \end{pmatrix}}, \displaybreak[0]

6940:  \\

6941:  &

6942:  C_4 = \nu_2[1]\twist{z^{-1}} \oplus \nu_4[1]\twist{x^{-1}} \oplus \nu_7[1]\twist{x^{-1}y}

6943:  \oplus \nu_8\twist{x^{-1}z^{-1}} \\ & \qquad \qquad \oplus \nu_9\twist{x^{-1}y^{-1}z^{-1}}

6944:  \oplus \nu_{12}\twist{y^{-1}z^{-1}}, \displaybreak[0] \\

6945:  & \delta_4 =

6946:  {\arraycolsep2.5pt \begin{pmatrix}

6947:  0 & 0 & 0 & 0 & 0 & 0 \\

6948:  \nu_2 \cap \nu_4 \!:\! mx^{-1}z & 0 & 0 & 0 & 0 & 0 \\

6949:  \nu_2 \cap \nu_7 \!:\! mx^{-1}yz & 0 & 0 & 0 & 0 & 0 \\

6950:  0 & \nu_4 \cap \nu_8\!:\!z^{-1} & \nu_7 \cap \nu_8\!:\!y^{-1}z^{-1} & 0 & 0 & 0 \\

6951:  0 & 0 & 0 & \nu_8 \cap \nu_9\!:\!my^{-1} & 0 & 0 \\

6952:  0 & 0 & 0 & \nu_8 \cap \nu_{12}\!:\!mxy^{-1} & 0 & 0

6953:  \end{pmatrix}}.

6954: \end{align*}

6955: where $\nu_k[d]\twist{x^ay^bz^c}$ means the object $\nu_k$ shifted

6956: down by $d$, and tensored with the one-dimensional representation of

6957: $T$ on which $[x,y,z,1]$ acts as $x^{-a}y^{-b}z^{-c}$.

6958: 

6959: It is now easy to compute explicitly the full $A_\infty$-subcategory

6960: of $Tw_T\Cat^{\rightarrow}_{12}$ consisting of these four equivariant

6961: twisted complexes $C_k$. More precisely, one verifies (by computer)

6962: first that the cohomology of this subcategory is $T$-equivariantly

6963: isomorphic to $Q_4$ and then, following the indications in Remark

6964: \ref{th:more-recognize-q4}, that a suitable fourth order Massey

6965: product is nonzero, which implies that our full subcategory is

6966: $T$-equivariantly quasi-isomorphic to $\QQ_4$. By Theorem

6967: \ref{th:induction}, or rather the equivariant version of it explained

6968: in Section \ref{subsec:addition}, we have a $T$-equivariant

6969: quasi-isomorphism of our subcategory with $\Fuk_4$, and this

6970: completes the proof of Proposition \ref{th:computation}.

6971: 

6972: \begin{figure}[H]

6973: \begin{centering}

6974: \begin{picture}(0,0)%

6975: \includegraphics{paths1.pstex}%

6976: \end{picture}%

6977: \setlength{\unitlength}{3947sp}%

6978: %

6979: \begingroup\makeatletter\ifx\SetFigFont\undefined%

6980: \gdef\SetFigFont#1#2#3#4#5{%

6981:   \reset@font\fontsize{#1}{#2pt}%

6982:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

6983:   \selectfont}%

6984: \fi\endgroup%

6985: \begin{picture}(4641,4716)(268,-3694)

6986: \put(2701,-736){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$b_1$}%

6987: }}}

6988: \put(3676,-1261){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$b_2$}%

6989: }}}

6990: \put(2701,-2011){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$b_3$}%

6991: }}}

6992: \put(1126,-1261){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$b_4$}%

6993: }}}

6994: \end{picture}

6995: \end{centering}

6996: \caption{\label{fig:paths1} The actual ($\bullet$) and fake ($\circ$)

6997: critical points of $\pi_{\bar{M}}$, together with our choice of

6998: vanishing paths $b_1,\dots,b_4$. The image has been distorted for

6999: better legibility.}

7000: \end{figure}

7001: 

7002: \begin{figure}[H]

7003: \begin{center}

7004: \begin{picture}(0,0)%

7005: \includegraphics{paths2.pstex}%

7006: \end{picture}%

7007: \setlength{\unitlength}{3947sp}%

7008: %

7009: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7010: \gdef\SetFigFont#1#2#3#4#5{%

7011:   \reset@font\fontsize{#1}{#2pt}%

7012:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7013:   \selectfont}%

7014: \fi\endgroup%

7015: \begin{picture}(3750,3562)(301,-3211)

7016: \put(1726,-2536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$d_2$}%

7017: }}}

7018: \put(530,-2850){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$d_3$}%

7019: }}}

7020: \put(3250,-2050){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$d_1$}%

7021: }}}

7022: \put(2701,-2536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$d_4$}%

7023: }}}

7024: \end{picture}

7025: \end{center}

7026: 

7027: \caption{\label{fig:paths2}The four matching paths obtained from the

7028: vanishing paths in Figure \ref{fig:paths1} by braid monodromy. These

7029: were drawn using a computer to watch how the points of $\bar{C} \cap

7030: \{z = b_k(t)\}$ move in $\C$ as one goes along the vanishing paths.

7031: The picture would be $\Z/4$-symmetric if we hadn't moved the paths by

7032: an isotopy, so as to avoid the future base path (the dashed line).}

7033: \end{figure}

7034: 

7035: \begin{figure}[H]

7036: \begin{center}

7037:  \epsfig{file=paths3.eps}

7038: 

7039:  \caption{\label{fig:paths3}

7040:  The base path and the 12 vanishing paths $c_1,\dots,c_{12}$

7041:  (ordered as they must be, clockwise with respect to

7042:  the derivatives at the base point).

7043:  }

7044: \end{center}

7045: \end{figure}

7046: 

7047: \begin{figure}[H]

7048: \begin{center}

7049:  \epsfig{file=curve1.eps}

7050: 

7051:  \caption{\label{fig:curve1}}

7052: \end{center}

7053: \end{figure}

7054: 

7055: \begin{figure}[H]

7056: \begin{center}

7057:  \epsfig{file=curve2.eps}

7058: 

7059:  \caption{\label{fig:curve2}}

7060: \end{center}

7061: \end{figure}

7062: 

7063: \begin{figure}[H]

7064: \begin{center}

7065:  \epsfig{file=curve3.eps}

7066:  \caption{\label{fig:curve3}}

7067: \end{center}

7068: \end{figure}

7069: 

7070: \begin{figure}[H]

7071: \begin{picture}(0,0)%

7072: \includegraphics{vanishing1.pstex}%

7073: \end{picture}%

7074: \setlength{\unitlength}{3947sp}%

7075: %

7076: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7077: \gdef\SetFigFont#1#2#3#4#5{%

7078:   \reset@font\fontsize{#1}{#2pt}%

7079:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7080:   \selectfont}%

7081: \fi\endgroup%

7082: \begin{picture}(4049,3690)(151,-3206)

7083: \put(475,-330){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7084: }}}

7085: \put(850, 14){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7086: }}}

7087: \put(3526,-586){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7088: }}}

7089: \put(4100,-2500){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_8$}%

7090: }}}

7091: \put(1375,-2500){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7092: }}}

7093: \put(2050,-3100){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7094: }}}

7095: \put(3475,-3100){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7096: }}}

7097: \put(3901,-2700){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7098: }}}

7099: \put(3650,-750){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7100: }}}

7101: \put(3800,-886){\makebox(0,0)[lb]{\smash{\SetFigFont{10}{12.0}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7102: }}}

7103: \end{picture}

7104: 

7105: \caption{\label{fig:vanishing1}}

7106: \end{figure}

7107: 

7108: \begin{figure}[H]

7109: \begin{picture}(0,0)%

7110: \includegraphics{vanishing2.pstex}%

7111: \end{picture}%

7112: \setlength{\unitlength}{3947sp}%

7113: %

7114: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7115: \gdef\SetFigFont#1#2#3#4#5{%

7116:   \reset@font\fontsize{#1}{#2pt}%

7117:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7118:   \selectfont}%

7119: \fi\endgroup%

7120: \begin{picture}(3974,3753)(226,-3269)

7121: \put(4100,-2490){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7122: }}}

7123: \put(3876,-2711){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7124: }}}

7125: \put(1800,-2921){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7126: }}}

7127: \put(2070,-3120){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7128: }}}

7129: \put(3440,-3150){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7130: }}}

7131: \put(550,-180){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7132: }}}

7133: \put(1050,280){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7134: }}}

7135: \put(3350,-466){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7136: }}}

7137: \put(3676,-800){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7138: }}}

7139: \put(3526,-630){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7140: }}}

7141: \end{picture}

7142: 

7143: \caption{\label{fig:vanishing2}}

7144: \end{figure}

7145: 

7146: \begin{figure}[H]

7147: \begin{picture}(0,0)%

7148: \includegraphics{vanishing3.pstex}%

7149: \end{picture}%

7150: \setlength{\unitlength}{3947sp}%

7151: %

7152: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7153: \gdef\SetFigFont#1#2#3#4#5{%

7154:   \reset@font\fontsize{#1}{#2pt}%

7155:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7156:   \selectfont}%

7157: \fi\endgroup%

7158: \begin{picture}(3674,3690)(526,-3206)

7159: \put(3601,-700){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7160: }}}

7161: \put(3376,-480){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7162: }}}

7163: \put(3770,-890){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7164: }}}

7165: \put(3901,-2686){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7166: }}}

7167: \put(4100,-2461){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7168: }}}

7169: \put(3526,-3061){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7170: }}}

7171: \put(2000,-3100){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7172: }}}

7173: \put(1410,-2500){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7174: }}}

7175: \put(576,-246){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7176: }}}

7177: \put(826,0){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7178: }}}

7179: \end{picture}

7180: 

7181: \caption{\label{fig:vanishing3}}

7182: \end{figure}

7183: 

7184: \begin{figure}[H]

7185: \begin{picture}(0,0)%

7186: \includegraphics{vanishing4.pstex}%

7187: \end{picture}%

7188: \setlength{\unitlength}{3947sp}%

7189: %

7190: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7191: \gdef\SetFigFont#1#2#3#4#5{%

7192:   \reset@font\fontsize{#1}{#2pt}%

7193:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7194:   \selectfont}%

7195: \fi\endgroup%

7196: \begin{picture}(3674,3753)(526,-3269)

7197: \put(1100,375){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7198: }}}

7199: \put(3440,-520){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7200: }}}

7201: \put(3725,-820){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7202: }}}

7203: \put(3575,-670){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7204: }}}

7205: \put(4051,-2536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7206: }}}

7207: \put(3901,-2686){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7208: }}}

7209: \put(3500,-3100){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7210: }}}

7211: \put(1821,-2936){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7212: }}}

7213: \put(570,-200){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7214: }}}

7215: \put(2101,-3251){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7216: }}}

7217: \end{picture}

7218: 

7219: \caption{\label{fig:vanishing4}}

7220: \end{figure}

7221: 

7222: \begin{figure}[H]

7223: \begin{picture}(0,0)%

7224: \includegraphics{vanishing5.pstex}%

7225: \end{picture}%

7226: \setlength{\unitlength}{3947sp}%

7227: %

7228: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7229: \gdef\SetFigFont#1#2#3#4#5{%

7230:   \reset@font\fontsize{#1}{#2pt}%

7231:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7232:   \selectfont}%

7233: \fi\endgroup%

7234: \begin{picture}(3622,3753)(578,-3269)

7235: \put(3451,-536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7236: }}}

7237: \put(3601,-686){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_8$}%

7238: }}}

7239: \put(3701,-836){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7240: }}}

7241: \put(606,-136){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7242: }}}

7243: \put(531,-286){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7244: }}}

7245: \put(1600,-2636){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7246: }}}

7247: \put(2000,-3136){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7248: }}}

7249: \put(3376,-3211){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7250: }}}

7251: \put(3826,-2761){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7252: }}}

7253: \put(4050,-2536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7254: }}}

7255: \end{picture}

7256: 

7257: \caption{\label{fig:vanishing5}}

7258: \end{figure}

7259: 

7260: \begin{figure}[H]

7261: \begin{picture}(0,0)%

7262: \includegraphics{vanishing6.pstex}%

7263: \end{picture}%

7264: \setlength{\unitlength}{3947sp}%

7265: %

7266: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7267: \gdef\SetFigFont#1#2#3#4#5{%

7268:   \reset@font\fontsize{#1}{#2pt}%

7269:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7270:   \selectfont}%

7271: \fi\endgroup%

7272: \begin{picture}(3622,3753)(578,-3269)

7273: \put(1626,-2711){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7274: }}}

7275: \put(3485,-3080){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7276: }}}

7277: \put(2121,-3211){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7278: }}}

7279: \put(3901,-2686){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_7$}%

7280: }}}

7281: \put(4100,-2461){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7282: }}}

7283: \put(3700,-811){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7284: }}}

7285: \put(3566,-661){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7286: }}}

7287: \put(3361,-466){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7288: }}}

7289: \put(1100,300){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7290: }}}

7291: \put(701,-106){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7292: }}}

7293: \end{picture}

7294: 

7295: \caption{\label{fig:vanishing6}}

7296: \end{figure}

7297: 

7298: \begin{figure}[H]

7299: \begin{picture}(0,0)%

7300: \includegraphics{vanishing7.pstex}%

7301: \end{picture}%

7302: \setlength{\unitlength}{3947sp}%

7303: %

7304: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7305: \gdef\SetFigFont#1#2#3#4#5{%

7306:   \reset@font\fontsize{#1}{#2pt}%

7307:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7308:   \selectfont}%

7309: \fi\endgroup%

7310: \begin{picture}(3974,3690)(226,-3206)

7311: \put(700,-75){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_8$}%

7312: }}}

7313: \put(460,-250){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{12}$}%

7314: }}}

7315: \put(1576,-2636){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7316: }}}

7317: \put(2056,-3136){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_8$}%

7318: }}}

7319: \put(3450,-536){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7320: }}}

7321: \put(3600,-686){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_5$}%

7322: }}}

7323: \put(3740,-836){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7324: }}}

7325: \put(4050,-2550){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{11}$}%

7326: }}}

7327: \put(3866,-2761){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7328: }}}

7329: \put(3500,-3100){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7330: }}}

7331: \end{picture}

7332: 

7333: \caption{\label{fig:vanishing7}}

7334: \end{figure}

7335: 

7336: \begin{figure}[H]

7337: \begin{picture}(0,0)%

7338: \includegraphics{vanishing8.pstex}%

7339: \end{picture}%

7340: \setlength{\unitlength}{3947sp}%

7341: %

7342: \begingroup\makeatletter\ifx\SetFigFont\undefined%

7343: \gdef\SetFigFont#1#2#3#4#5{%

7344:   \reset@font\fontsize{#1}{#2pt}%

7345:   \fontfamily{#3}\fontseries{#4}\fontshape{#5}%

7346:   \selectfont}%

7347: \fi\endgroup%

7348: \begin{picture}(3622,3753)(578,-3269)

7349: \put(1640,-2721){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7350: }}}

7351: \put(2101,-3181){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7352: }}}

7353: \put(4101,-2461){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_8$}%

7354: }}}

7355: \put(3556,-661){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_2$}%

7356: }}}

7357: \put(3361,-466){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_9$}%

7358: }}}

7359: \put(3720,-811){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_{10}$}%

7360: }}}

7361: \put(1080,280){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_6$}%

7362: }}}

7363: \put(721,-80){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_1$}%

7364: }}}

7365: \put(3500,-3066){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_3$}%

7366: }}}

7367: \put(3921,-2656){\makebox(0,0)[lb]{\smash{\SetFigFont{12}{14.4}{\familydefault}{\mddefault}{\updefault}{\color[rgb]{0,0,0}$\nu_4$}%

7368: }}}

7369: \end{picture}

7370: 

7371: \caption{\label{fig:vanishing8}}

7372: \end{figure}

7373: 

7374: \begin{table}[H] \extrarowheight0.2em

7375: \begin{equation*}

7376: \begin{array}{l||l|l|l|l|l|l|}

7377:  & 1 & 2 & 3 & 4 & 5 & 6 \\

7378:  \hhline{=#======} 1  \\

7379:  \hhline{--} 2 & x  \\

7380:  \hhline{---} 3 & & my^{-1}  \\

7381:  \hhline{----} 4 & 1 & mx^{-1}z & \\

7382:  & x && \\

7383:  \hhline{-----} 5 & x & x^{-1}y & x^{-1}y & 1  \\

7384:  & 1      & 1 && \\

7385:  & y &&& \\

7386:  \hhline{------} 6 & & my^{-1}z^{-1} & x^{-1}yz^{-1} && my^{-1}z^{-1}  \\

7387:  & & mx^{-1}z^{-1} & x^{-1} && \\

7388:  & & mx^{-1}y^{-1} &&& \\

7389:  \hline 7 & myz & mx^{-1}z && 1 & mz & \\

7390:  & x & mx^{-1}yz && x && \\

7391:  && mz &&&& \\

7392:  \hline 8 & xz^{-1} & z^{-1} & x^{-1} & xy^{-1}z^{-1} & z^{-1} & 1 \\

7393:  & xy^{-1}z^{-1} & mx^{-1}y^{-1} & z^{-1} & y^{-1}z^{-1} & xy^{-1}z^{-1} & \\

7394:  & z^{-1} && x^{-1}yz^{-1} & z^{-1} && \\

7395:  \hline 9 && my^{-1}z^{-1} & x^{-1} && mxy^{-2}z^{-1} & 1 \\

7396:  && mx^{-1}y^{-1} & my^{-1}z^{-1} && my^{-1}z^{-1} & y^{-1}z \\

7397:  && my^{-2} &&& my^{-2} & \\

7398:  \hline 10 & m & mx^{-2} && mx^{-1} & mx^{-1}y^{-1} & \\

7399:  & mx^{-1} & mx^{-1} && y^{-1}z^{-1} & mx^{-1} & \\

7400:  & & mx^{-1}y^{-1} & & & my^{-1} & \\

7401:  \hline 11 & xz^{-1} & my^{-1} & x^{-1}y & xz^{-1} & xz^{-1} & z \\

7402:  && mx^{-1} & 1 & xy^{-1}z^{-1} & my^{-1} & x \\

7403:  && & yz^{-1} & z^{-1} && y \\

7404:  \hline 12 && my^{-1} & my^{-1} && mxy^{-1}z^{-1} & z \\

7405:  &&& mz^{-1} && my^{-1} & mxy^{-1} \\

7406:  &&&&& mxy^{-2} & \\ \hline

7407: \end{array}

7408: \]

7409: 

7410: \caption{\label{table:intersections1} The points of $\nu_k \cap

7411: \nu_l$ for $1 \leq k \leq 6$ and $l > k$, with their Maslov indices

7412: and $T$-action weights.}

7413: \end{table}

7414: 

7415: \begin{table}[H] \extrarowheight0.2em

7416: \begin{equation*}

7417: \begin{array}{l||l|l|l|l|l|l|}

7418:  & 7 & 8 & 9 & 10 & 11 & 12 \\

7419:  \hhline{=#======} 7  \\

7420:  \hhline{--} 8 & y^{-1}z^{-1}  \\

7421:  \hhline{---} 9 & & my^{-1}  \\

7422:  \hhline{----} 10 & x^{-1}y^{-1}z^{-1} & mx^{-1}z & \\

7423:  & y^{-1}z^{-1} && \\

7424:  \hhline{-----} 11 & xy^{-1}z^{-1} & y & y & x \\

7425:  & y^{-1}z^{-1}      & x && \\

7426:  & z^{-1} &&& \\

7427:  \hhline{------} 12 & & mxy^{-1} & y && my^{-1}  \\

7428:  & & m & z && \\

7429:  & & my^{-1}z &&& \\

7430:  \hline

7431: \end{array}

7432: \]

7433: 

7434: \caption{\label{table:intersections2} The points of $\nu_k \cap

7435: \nu_l$ for $7 \leq k \leq 12$ and $l > k$, with their Maslov indices

7436: and $T$-action weights.}

7437: \end{table}

7438: 

7439: { \smaller\smaller

7440: \begin{longtable}[c]{ll|l|l}

7441:  $a_1$ & $a_2$ & $b$ & $\pm$ \\

7442:  \hline

7443:  $\nu_1 \cap \nu_2:x$

7444:  & $\nu_2 \cap \nu_5:1$

7445:  & $\nu_1 \cap \nu_5:x$

7446:  & $+$ \\

7447: $\nu_1 \cap \nu_2:x$

7448:  & $\nu_2 \cap \nu_5:x^{-1}y$

7449:  & $\nu_1 \cap \nu_5:y$

7450:  & $-$ \\

7451: $\nu_1 \cap \nu_4:x$

7452:  & $\nu_4 \cap \nu_5:1$

7453:  & $\nu_1 \cap \nu_5:x$

7454:  & $+$ \\

7455: $\nu_1 \cap \nu_4:1$

7456:  & $\nu_4 \cap \nu_5:1$

7457:  & $\nu_1 \cap \nu_5:1$

7458:  & $-$ \\

7459: $\nu_2 \cap \nu_3:my^{-1}$

7460:  & $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

7461:  & $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

7462:  & $-$ \\

7463: $\nu_2 \cap \nu_3:my^{-1}$

7464:  & $\nu_3 \cap \nu_6:x^{-1}$

7465:  & $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

7466:  & $-$ \\

7467: $\nu_2 \cap \nu_5:x^{-1}y$

7468:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7469:  & $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

7470:  & $-$ \\

7471: $\nu_2 \cap \nu_5:1$

7472:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7473:  & $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

7474:  & $+$ \\

7475: $\nu_4 \cap \nu_5:1$

7476:  & $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7477:  & $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

7478:  & $+$ \\

7479: $\nu_4 \cap \nu_5:1$

7480:  & $\nu_5 \cap \nu_8:z^{-1}$

7481:  & $\nu_4 \cap \nu_8:z^{-1}$

7482:  & $-$ \\

7483: $\nu_4 \cap \nu_7:x$

7484:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7485:  & $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

7486:  & $+$ \\

7487: $\nu_4 \cap \nu_7:1$

7488:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7489:  & $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

7490:  & $-$ \\

7491: $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7492:  & $\nu_6 \cap \nu_9:1$

7493:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

7494:  & $-$ \\

7495: $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7496:  & $\nu_6 \cap \nu_9:y^{-1}z$

7497:  & $\nu_5 \cap \nu_9:my^{-2}$

7498:  & $-$ \\

7499: $\nu_5 \cap \nu_8:z^{-1}$

7500:  & $\nu_8 \cap \nu_9:my^{-1}$

7501:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

7502:  & $-$ \\

7503: $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7504:  & $\nu_8 \cap \nu_9:my^{-1}$

7505:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

7506:  & $+$ \\

7507: $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7508:  & $\nu_8 \cap \nu_{11}:x$

7509:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

7510:  & $+$ \\

7511: $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7512:  & $\nu_8 \cap \nu_{11}:y$

7513:  & $\nu_7 \cap \nu_{11}:z^{-1}$

7514:  & $-$ \\

7515: $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7516:  & $\nu_{10} \cap \nu_{11}:x$

7517:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

7518:  & $+$ \\

7519: $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7520:  & $\nu_{10} \cap \nu_{11}:x$

7521:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

7522:  & $-$ \\

7523: $\nu_8 \cap \nu_9:my^{-1}$

7524:  & $\nu_9 \cap \nu_{12}:y$

7525:  & $\nu_8 \cap \nu_{12}:m$

7526:  & $-$ \\

7527: $\nu_8 \cap \nu_9:my^{-1}$

7528:  & $\nu_9 \cap \nu_{12}:z$

7529:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

7530:  & $-$ \\

7531: $\nu_8 \cap \nu_{11}:y$

7532:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

7533:  & $\nu_8 \cap \nu_{12}:m$

7534:  & $-$ \\

7535: $\nu_8 \cap \nu_{11}:x$

7536:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

7537:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

7538:  & $+$ \\

7539: $\nu_2 \cap \nu_4:mx^{-1}z$

7540:  & $\nu_4 \cap \nu_7:1$

7541:  & $\nu_2 \cap \nu_7:mx^{-1}z$

7542:  & $-$ \\

7543: $\nu_2 \cap \nu_4:mx^{-1}z$

7544:  & $\nu_4 \cap \nu_7:x$

7545:  & $\nu_2 \cap \nu_7:mz$

7546:  & $+$ \\

7547: $\nu_2 \cap \nu_5:1$

7548:  & $\nu_5 \cap \nu_7:mz$

7549:  & $\nu_2 \cap \nu_7:mz$

7550:  & $+$ \\

7551: $\nu_2 \cap \nu_5:x^{-1}y$

7552:  & $\nu_5 \cap \nu_7:mz$

7553:  & $\nu_2 \cap \nu_7:mx^{-1}yz$

7554:  & $-$ \\

7555: $\nu_3 \cap \nu_5:x^{-1}y$

7556:  & $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7557:  & $\nu_3 \cap \nu_8:z^{-1}$

7558:  & $-$ \\

7559: $\nu_3 \cap \nu_5:x^{-1}y$

7560:  & $\nu_5 \cap \nu_8:z^{-1}$

7561:  & $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

7562:  & $-$ \\

7563: $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

7564:  & $\nu_6 \cap \nu_8:1$

7565:  & $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

7566:  & $-$ \\

7567: $\nu_3 \cap \nu_6:x^{-1}$

7568:  & $\nu_6 \cap \nu_8:1$

7569:  & $\nu_3 \cap \nu_8:x^{-1}$

7570:  & $+$ \\

7571: $\nu_1 \cap \nu_2:x$

7572:  & $\nu_2 \cap \nu_7:mx^{-1}yz$

7573:  & $\nu_1 \cap \nu_7:myz$

7574:  & $+$ \\

7575: $\nu_1 \cap \nu_4:x$

7576:  & $\nu_4 \cap \nu_7:1$

7577:  & $\nu_1 \cap \nu_7:x$

7578:  & $-$ \\

7579: $\nu_1 \cap \nu_4:1$

7580:  & $\nu_4 \cap \nu_7:x$

7581:  & $\nu_1 \cap \nu_7:x$

7582:  & $+$ \\

7583: $\nu_1 \cap \nu_5:y$

7584:  & $\nu_5 \cap \nu_7:mz$

7585:  & $\nu_1 \cap \nu_7:myz$

7586:  & $+$ \\

7587: $\nu_2 \cap \nu_3:my^{-1}$

7588:  & $\nu_3 \cap \nu_8:x^{-1}$

7589:  & $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

7590:  & $+$ \\

7591: $\nu_2 \cap \nu_4:mx^{-1}z$

7592:  & $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

7593:  & $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

7594:  & $+$ \\

7595: $\nu_2 \cap \nu_5:x^{-1}y$

7596:  & $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7597:  & $\nu_2 \cap \nu_8:z^{-1}$

7598:  & $+$ \\

7599: $\nu_2 \cap \nu_5:1$

7600:  & $\nu_5 \cap \nu_8:z^{-1}$

7601:  & $\nu_2 \cap \nu_8:z^{-1}$

7602:  & $-$ \\

7603: $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

7604:  & $\nu_6 \cap \nu_8:1$

7605:  & $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

7606:  & $+$ \\

7607: $\nu_2 \cap \nu_7:mx^{-1}z$

7608:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7609:  & $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

7610:  & $-$ \\

7611: $\nu_3 \cap \nu_5:x^{-1}y$

7612:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

7613:  & $\nu_3 \cap \nu_9:my^{-1}z^{-1}$

7614:  & $+$ \\

7615: $\nu_3 \cap \nu_6:x^{-1}$

7616:  & $\nu_6 \cap \nu_9:1$

7617:  & $\nu_3 \cap \nu_9:x^{-1}$

7618:  & $-$ \\

7619: $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

7620:  & $\nu_6 \cap \nu_9:y^{-1}z$

7621:  & $\nu_3 \cap \nu_9:x^{-1}$

7622:  & $+$ \\

7623: $\nu_3 \cap \nu_8:z^{-1}$

7624:  & $\nu_8 \cap \nu_9:my^{-1}$

7625:  & $\nu_3 \cap \nu_9:my^{-1}z^{-1}$

7626:  & $-$ \\

7627: $\nu_1 \cap \nu_2:x$

7628:  & $\nu_2 \cap \nu_8:z^{-1}$

7629:  & $\nu_1 \cap \nu_8:xz^{-1}$

7630:  & $-$ \\

7631: $\nu_1 \cap \nu_4:x$

7632:  & $\nu_4 \cap \nu_8:z^{-1}$

7633:  & $\nu_1 \cap \nu_8:xz^{-1}$

7634:  & $-$ \\

7635: $\nu_1 \cap \nu_4:x$

7636:  & $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

7637:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

7638:  & $+$ \\

7639: $\nu_1 \cap \nu_4:1$

7640:  & $\nu_4 \cap \nu_8:z^{-1}$

7641:  & $\nu_1 \cap \nu_8:z^{-1}$

7642:  & $+$ \\

7643: $\nu_1 \cap \nu_4:1$

7644:  & $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

7645:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

7646:  & $+$ \\

7647: $\nu_1 \cap \nu_5:x$

7648:  & $\nu_5 \cap \nu_8:z^{-1}$

7649:  & $\nu_1 \cap \nu_8:xz^{-1}$

7650:  & $+$ \\

7651: $\nu_1 \cap \nu_5:y$

7652:  & $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7653:  & $\nu_1 \cap \nu_8:xz^{-1}$

7654:  & $+$ \\

7655: $\nu_1 \cap \nu_5:1$

7656:  & $\nu_5 \cap \nu_8:z^{-1}$

7657:  & $\nu_1 \cap \nu_8:z^{-1}$

7658:  & $+$ \\

7659: $\nu_1 \cap \nu_5:1$

7660:  & $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7661:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

7662:  & $-$ \\

7663: $\nu_1 \cap \nu_7:x$

7664:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

7665:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

7666:  & $+$ \\

7667: $\nu_2 \cap \nu_3:my^{-1}$

7668:  & $\nu_3 \cap \nu_9:x^{-1}$

7669:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

7670:  & $+$ \\

7671: $\nu_2 \cap \nu_5:1$

7672:  & $\nu_5 \cap \nu_9:my^{-2}$

7673:  & $\nu_2 \cap \nu_9:my^{-2}$

7674:  & $+$ \\

7675: $\nu_2 \cap \nu_5:1$

7676:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

7677:  & $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

7678:  & $+$ \\

7679: $\nu_2 \cap \nu_5:x^{-1}y$

7680:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

7681:  & $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

7682:  & $+$ \\

7683: $\nu_2 \cap \nu_5:x^{-1}y$

7684:  & $\nu_5 \cap \nu_9:my^{-2}$

7685:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

7686:  & $+$ \\

7687: $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

7688:  & $\nu_6 \cap \nu_9:1$

7689:  & $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

7690:  & $-$ \\

7691: $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

7692:  & $\nu_6 \cap \nu_9:y^{-1}z$

7693:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

7694:  & $+$ \\

7695: $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

7696:  & $\nu_6 \cap \nu_9:1$

7697:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

7698:  & $-$ \\

7699: $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

7700:  & $\nu_6 \cap \nu_9:y^{-1}z$

7701:  & $\nu_2 \cap \nu_9:my^{-2}$

7702:  & $-$ \\

7703: $\nu_2 \cap \nu_8:z^{-1}$

7704:  & $\nu_8 \cap \nu_9:my^{-1}$

7705:  & $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

7706:  & $+$ \\

7707: $\nu_5 \cap \nu_7:mz$

7708:  & $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7709:  & $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

7710:  & $-$ \\

7711: $\nu_5 \cap \nu_7:mz$

7712:  & $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7713:  & $\nu_5 \cap \nu_{10}:my^{-1}$

7714:  & $+$ \\

7715: $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7716:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

7717:  & $\nu_5 \cap \nu_{10}:my^{-1}$

7718:  & $+$ \\

7719: $\nu_5 \cap \nu_8:z^{-1}$

7720:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

7721:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

7722:  & $-$ \\

7723: $\nu_6 \cap \nu_8:1$

7724:  & $\nu_8 \cap \nu_{11}:x$

7725:  & $\nu_6 \cap \nu_{11}:x$

7726:  & $-$ \\

7727: $\nu_6 \cap \nu_8:1$

7728:  & $\nu_8 \cap \nu_{11}:y$

7729:  & $\nu_6 \cap \nu_{11}:y$

7730:  & $-$ \\

7731: $\nu_6 \cap \nu_9:1$

7732:  & $\nu_9 \cap \nu_{11}:y$

7733:  & $\nu_6 \cap \nu_{11}:y$

7734:  & $-$ \\

7735: $\nu_6 \cap \nu_9:y^{-1}z$

7736:  & $\nu_9 \cap \nu_{11}:y$

7737:  & $\nu_6 \cap \nu_{11}:z$

7738:  & $+$ \\

7739: $\nu_4 \cap \nu_5:1$

7740:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

7741:  & $\nu_4 \cap \nu_{10}:mx^{-1}$

7742:  & $+$ \\

7743: $\nu_4 \cap \nu_7:x$

7744:  & $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7745:  & $\nu_4 \cap \nu_{10}:y^{-1}z^{-1}$

7746:  & $-$ \\

7747: $\nu_4 \cap \nu_7:1$

7748:  & $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7749:  & $\nu_4 \cap \nu_{10}:y^{-1}z^{-1}$

7750:  & $+$ \\

7751: $\nu_4 \cap \nu_8:z^{-1}$

7752:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

7753:  & $\nu_4 \cap \nu_{10}:mx^{-1}$

7754:  & $+$ \\

7755: $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7756:  & $\nu_6 \cap \nu_{11}:z$

7757:  & $\nu_5 \cap \nu_{11}:my^{-1}$

7758:  & $+$ \\

7759: $\nu_5 \cap \nu_7:mz$

7760:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

7761:  & $\nu_5 \cap \nu_{11}:my^{-1}$

7762:  & $+$ \\

7763: $\nu_5 \cap \nu_8:z^{-1}$

7764:  & $\nu_8 \cap \nu_{11}:x$

7765:  & $\nu_5 \cap \nu_{11}:xz^{-1}$

7766:  & $+$ \\

7767: $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7768:  & $\nu_8 \cap \nu_{11}:y$

7769:  & $\nu_5 \cap \nu_{11}:xz^{-1}$

7770:  & $-$ \\

7771: $\nu_5 \cap \nu_9:my^{-2}$

7772:  & $\nu_9 \cap \nu_{11}:y$

7773:  & $\nu_5 \cap \nu_{11}:my^{-1}$

7774:  & $+$ \\

7775: $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

7776:  & $\nu_{10} \cap \nu_{11}:x$

7777:  & $\nu_5 \cap \nu_{11}:my^{-1}$

7778:  & $-$ \\

7779: $\nu_6 \cap \nu_8:1$

7780:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

7781:  & $\nu_6 \cap \nu_{12}:mxy^{-1}$

7782:  & $+$ \\

7783: $\nu_6 \cap \nu_9:y^{-1}z$

7784:  & $\nu_9 \cap \nu_{12}:y$

7785:  & $\nu_6 \cap \nu_{12}:z$

7786:  & $-$ \\

7787: $\nu_6 \cap \nu_9:1$

7788:  & $\nu_9 \cap \nu_{12}:z$

7789:  & $\nu_6 \cap \nu_{12}:z$

7790:  & $+$ \\

7791: $\nu_6 \cap \nu_{11}:x$

7792:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

7793:  & $\nu_6 \cap \nu_{12}:mxy^{-1}$

7794:  & $-$ \\

7795: $\nu_4 \cap \nu_5:1$

7796:  & $\nu_5 \cap \nu_{11}:xz^{-1}$

7797:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

7798:  & $-$ \\

7799: $\nu_4 \cap \nu_7:x$

7800:  & $\nu_7 \cap \nu_{11}:z^{-1}$

7801:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

7802:  & $-$ \\

7803: $\nu_4 \cap \nu_7:x$

7804:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

7805:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

7806:  & $+$ \\

7807: $\nu_4 \cap \nu_7:1$

7808:  & $\nu_7 \cap \nu_{11}:z^{-1}$

7809:  & $\nu_4 \cap \nu_{11}:z^{-1}$

7810:  & $+$ \\

7811: $\nu_4 \cap \nu_7:1$

7812:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

7813:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

7814:  & $+$ \\

7815: $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

7816:  & $\nu_8 \cap \nu_{11}:y$

7817:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

7818:  & $+$ \\

7819: $\nu_4 \cap \nu_8:z^{-1}$

7820:  & $\nu_8 \cap \nu_{11}:x$

7821:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

7822:  & $+$ \\

7823: $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

7824:  & $\nu_8 \cap \nu_{11}:y$

7825:  & $\nu_4 \cap \nu_{11}:z^{-1}$

7826:  & $+$ \\

7827: $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

7828:  & $\nu_8 \cap \nu_{11}:x$

7829:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

7830:  & $-$ \\

7831: $\nu_4 \cap \nu_{10}:y^{-1}z^{-1}$

7832:  & $\nu_{10} \cap \nu_{11}:x$

7833:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

7834:  & $+$ \\

7835: $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

7836:  & $\nu_6 \cap \nu_{12}:z$

7837:  & $\nu_5 \cap \nu_{12}:my^{-1}$

7838:  & $+$ \\

7839: $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7840:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

7841:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

7842:  & $+$ \\

7843: $\nu_5 \cap \nu_8:xy^{-1}z^{-1}$

7844:  & $\nu_8 \cap \nu_{12}:m$

7845:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

7846:  & $+$ \\

7847: $\nu_5 \cap \nu_8:z^{-1}$

7848:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

7849:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

7850:  & $+$ \\

7851: $\nu_5 \cap \nu_8:z^{-1}$

7852:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

7853:  & $\nu_5 \cap \nu_{12}:my^{-1}$

7854:  & $+$ \\

7855: $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

7856:  & $\nu_9 \cap \nu_{12}:y$

7857:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

7858:  & $-$ \\

7859: $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

7860:  & $\nu_9 \cap \nu_{12}:z$

7861:  & $\nu_5 \cap \nu_{12}:my^{-1}$

7862:  & $+$ \\

7863: $\nu_5 \cap \nu_9:my^{-2}$

7864:  & $\nu_9 \cap \nu_{12}:y$

7865:  & $\nu_5 \cap \nu_{12}:my^{-1}$

7866:  & $-$ \\

7867: $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

7868:  & $\nu_9 \cap \nu_{12}:z$

7869:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

7870:  & $-$ \\

7871: $\nu_5 \cap \nu_{11}:xz^{-1}$

7872:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

7873:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

7874:  & $+$ \\

7875: $\nu_2 \cap \nu_4:mx^{-1}z$

7876:  & $\nu_4 \cap \nu_{10}:y^{-1}z^{-1}$

7877:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

7878:  & $-$ \\

7879: $\nu_2 \cap \nu_5:x^{-1}y$

7880:  & $\nu_5 \cap \nu_{10}:my^{-1}$

7881:  & $\nu_2 \cap \nu_{10}:mx^{-1}$

7882:  & $+$ \\

7883: $\nu_2 \cap \nu_5:x^{-1}y$

7884:  & $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

7885:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

7886:  & $-$ \\

7887: $\nu_2 \cap \nu_5:1$

7888:  & $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

7889:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

7890:  & $+$ \\

7891: $\nu_2 \cap \nu_5:1$

7892:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

7893:  & $\nu_2 \cap \nu_{10}:mx^{-1}$

7894:  & $+$ \\

7895: $\nu_2 \cap \nu_7:mx^{-1}yz$

7896:  & $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7897:  & $\nu_2 \cap \nu_{10}:mx^{-1}$

7898:  & $-$ \\

7899: $\nu_2 \cap \nu_7:mx^{-1}yz$

7900:  & $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7901:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

7902:  & $-$ \\

7903: $\nu_2 \cap \nu_7:mz$

7904:  & $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7905:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

7906:  & $-$ \\

7907: $\nu_2 \cap \nu_7:mx^{-1}z$

7908:  & $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7909:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

7910:  & $-$ \\

7911: $\nu_2 \cap \nu_8:z^{-1}$

7912:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

7913:  & $\nu_2 \cap \nu_{10}:mx^{-1}$

7914:  & $+$ \\

7915: $\nu_3 \cap \nu_5:x^{-1}y$

7916:  & $\nu_5 \cap \nu_{11}:xz^{-1}$

7917:  & $\nu_3 \cap \nu_{11}:yz^{-1}$

7918:  & $-$ \\

7919: $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

7920:  & $\nu_6 \cap \nu_{11}:z$

7921:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

7922:  & $+$ \\

7923: $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

7924:  & $\nu_6 \cap \nu_{11}:x$

7925:  & $\nu_3 \cap \nu_{11}:yz^{-1}$

7926:  & $+$ \\

7927: $\nu_3 \cap \nu_6:x^{-1}$

7928:  & $\nu_6 \cap \nu_{11}:y$

7929:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

7930:  & $+$ \\

7931: $\nu_3 \cap \nu_6:x^{-1}$

7932:  & $\nu_6 \cap \nu_{11}:x$

7933:  & $\nu_3 \cap \nu_{11}:1$

7934:  & $-$ \\

7935: $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

7936:  & $\nu_8 \cap \nu_{11}:x$

7937:  & $\nu_3 \cap \nu_{11}:yz^{-1}$

7938:  & $+$ \\

7939: $\nu_3 \cap \nu_8:x^{-1}$

7940:  & $\nu_8 \cap \nu_{11}:x$

7941:  & $\nu_3 \cap \nu_{11}:1$

7942:  & $+$ \\

7943: $\nu_3 \cap \nu_8:z^{-1}$

7944:  & $\nu_8 \cap \nu_{11}:y$

7945:  & $\nu_3 \cap \nu_{11}:yz^{-1}$

7946:  & $-$ \\

7947: $\nu_3 \cap \nu_8:x^{-1}$

7948:  & $\nu_8 \cap \nu_{11}:y$

7949:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

7950:  & $-$ \\

7951: $\nu_3 \cap \nu_9:x^{-1}$

7952:  & $\nu_9 \cap \nu_{11}:y$

7953:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

7954:  & $+$ \\

7955: $\nu_1 \cap \nu_2:x$

7956:  & $\nu_2 \cap \nu_{10}:mx^{-1}$

7957:  & $\nu_1 \cap \nu_{10}:m$

7958:  & $+$ \\

7959: $\nu_1 \cap \nu_2:x$

7960:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

7961:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

7962:  & $+$ \\

7963: $\nu_1 \cap \nu_4:x$

7964:  & $\nu_4 \cap \nu_{10}:mx^{-1}$

7965:  & $\nu_1 \cap \nu_{10}:m$

7966:  & $+$ \\

7967: $\nu_1 \cap \nu_4:1$

7968:  & $\nu_4 \cap \nu_{10}:mx^{-1}$

7969:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

7970:  & $+$ \\

7971: $\nu_1 \cap \nu_5:y$

7972:  & $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

7973:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

7974:  & $+$ \\

7975: $\nu_1 \cap \nu_5:1$

7976:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

7977:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

7978:  & $-$ \\

7979: $\nu_1 \cap \nu_5:y$

7980:  & $\nu_5 \cap \nu_{10}:my^{-1}$

7981:  & $\nu_1 \cap \nu_{10}:m$

7982:  & $-$ \\

7983: $\nu_1 \cap \nu_5:x$

7984:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

7985:  & $\nu_1 \cap \nu_{10}:m$

7986:  & $+$ \\

7987: $\nu_1 \cap \nu_7:myz$

7988:  & $\nu_7 \cap \nu_{10}:x^{-1}y^{-1}z^{-1}$

7989:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

7990:  & $-$ \\

7991: $\nu_1 \cap \nu_7:myz$

7992:  & $\nu_7 \cap \nu_{10}:y^{-1}z^{-1}$

7993:  & $\nu_1 \cap \nu_{10}:m$

7994:  & $-$ \\

7995: $\nu_1 \cap \nu_8:xz^{-1}$

7996:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

7997:  & $\nu_1 \cap \nu_{10}:m$

7998:  & $-$ \\

7999: $\nu_1 \cap \nu_8:z^{-1}$

8000:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8001:  & $\nu_1 \cap \nu_{10}:mx^{-1}$

8002:  & $+$ \\

8003: $\nu_2 \cap \nu_3:my^{-1}$

8004:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

8005:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8006:  & $-$ \\

8007: $\nu_2 \cap \nu_3:my^{-1}$

8008:  & $\nu_3 \cap \nu_{11}:1$

8009:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8010:  & $+$ \\

8011: $\nu_2 \cap \nu_4:mx^{-1}z$

8012:  & $\nu_4 \cap \nu_{11}:z^{-1}$

8013:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8014:  & $+$ \\

8015: $\nu_2 \cap \nu_4:mx^{-1}z$

8016:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

8017:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8018:  & $-$ \\

8019: $\nu_2 \cap \nu_5:1$

8020:  & $\nu_5 \cap \nu_{11}:my^{-1}$

8021:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8022:  & $+$ \\

8023: $\nu_2 \cap \nu_5:x^{-1}y$

8024:  & $\nu_5 \cap \nu_{11}:my^{-1}$

8025:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8026:  & $+$ \\

8027: $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

8028:  & $\nu_6 \cap \nu_{11}:z$

8029:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8030:  & $+$ \\

8031: $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

8032:  & $\nu_6 \cap \nu_{11}:x$

8033:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8034:  & $-$ \\

8035: $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

8036:  & $\nu_6 \cap \nu_{11}:z$

8037:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8038:  & $-$ \\

8039: $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

8040:  & $\nu_6 \cap \nu_{11}:y$

8041:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8042:  & $-$ \\

8043: $\nu_2 \cap \nu_7:mx^{-1}z$

8044:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

8045:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8046:  & $-$ \\

8047: $\nu_2 \cap \nu_7:mz$

8048:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

8049:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8050:  & $+$ \\

8051: $\nu_2 \cap \nu_7:mx^{-1}yz$

8052:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

8053:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8054:  & $-$ \\

8055: $\nu_2 \cap \nu_7:mx^{-1}z$

8056:  & $\nu_7 \cap \nu_{11}:z^{-1}$

8057:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8058:  & $+$ \\

8059: $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

8060:  & $\nu_8 \cap \nu_{11}:y$

8061:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8062:  & $-$ \\

8063: $\nu_2 \cap \nu_8:mx^{-1}y^{-1}$

8064:  & $\nu_8 \cap \nu_{11}:x$

8065:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8066:  & $-$ \\

8067: $\nu_2 \cap \nu_9:my^{-2}$

8068:  & $\nu_9 \cap \nu_{11}:y$

8069:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8070:  & $+$ \\

8071: $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

8072:  & $\nu_9 \cap \nu_{11}:y$

8073:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8074:  & $+$ \\

8075: $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

8076:  & $\nu_{10} \cap \nu_{11}:x$

8077:  & $\nu_2 \cap \nu_{11}:my^{-1}$

8078:  & $-$ \\

8079: $\nu_2 \cap \nu_{10}:mx^{-2}$

8080:  & $\nu_{10} \cap \nu_{11}:x$

8081:  & $\nu_2 \cap \nu_{11}:mx^{-1}$

8082:  & $+$ \\

8083: $\nu_3 \cap \nu_5:x^{-1}y$

8084:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

8085:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8086:  & $+$ \\

8087: $\nu_3 \cap \nu_5:x^{-1}y$

8088:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

8089:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8090:  & $-$ \\

8091: $\nu_3 \cap \nu_6:x^{-1}$

8092:  & $\nu_6 \cap \nu_{12}:mxy^{-1}$

8093:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8094:  & $-$ \\

8095: $\nu_3 \cap \nu_6:x^{-1}yz^{-1}$

8096:  & $\nu_6 \cap \nu_{12}:mxy^{-1}$

8097:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8098:  & $+$ \\

8099: $\nu_3 \cap \nu_8:z^{-1}$

8100:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

8101:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8102:  & $+$ \\

8103: $\nu_3 \cap \nu_8:x^{-1}$

8104:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

8105:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8106:  & $-$ \\

8107: $\nu_3 \cap \nu_8:z^{-1}$

8108:  & $\nu_8 \cap \nu_{12}:m$

8109:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8110:  & $-$ \\

8111: $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

8112:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

8113:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8114:  & $-$ \\

8115: $\nu_3 \cap \nu_9:my^{-1}z^{-1}$

8116:  & $\nu_9 \cap \nu_{12}:y$

8117:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8118:  & $-$ \\

8119: $\nu_3 \cap \nu_9:my^{-1}z^{-1}$

8120:  & $\nu_9 \cap \nu_{12}:z$

8121:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8122:  & $+$ \\

8123: $\nu_3 \cap \nu_{11}:yz^{-1}$

8124:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8125:  & $\nu_3 \cap \nu_{12}:mz^{-1}$

8126:  & $-$ \\

8127: $\nu_3 \cap \nu_{11}:1$

8128:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8129:  & $\nu_3 \cap \nu_{12}:my^{-1}$

8130:  & $-$ \\

8131: $\nu_1 \cap \nu_4:1$

8132:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

8133:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8134:  & $+$ \\

8135: $\nu_1 \cap \nu_4:x$

8136:  & $\nu_4 \cap \nu_{11}:z^{-1}$

8137:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8138:  & $+$ \\

8139: $\nu_1 \cap \nu_5:1$

8140:  & $\nu_5 \cap \nu_{11}:xz^{-1}$

8141:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8142:  & $+$ \\

8143: $\nu_1 \cap \nu_7:x$

8144:  & $\nu_7 \cap \nu_{11}:z^{-1}$

8145:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8146:  & $-$ \\

8147: $\nu_1 \cap \nu_8:z^{-1}$

8148:  & $\nu_8 \cap \nu_{11}:x$

8149:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8150:  & $+$ \\

8151: $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

8152:  & $\nu_8 \cap \nu_{11}:y$

8153:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8154:  & $+$ \\

8155: $\nu_2 \cap \nu_5:x^{-1}y$

8156:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

8157:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8158:  & $+$ \\

8159: $\nu_2 \cap \nu_5:1$

8160:  & $\nu_5 \cap \nu_{12}:my^{-1}$

8161:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8162:  & $-$ \\

8163: $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

8164:  & $\nu_6 \cap \nu_{12}:z$

8165:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8166:  & $-$ \\

8167: $\nu_2 \cap \nu_8:z^{-1}$

8168:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

8169:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8170:  & $+$ \\

8171: $\nu_2 \cap \nu_9:my^{-2}$

8172:  & $\nu_9 \cap \nu_{12}:y$

8173:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8174:  & $+$ \\

8175: $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

8176:  & $\nu_9 \cap \nu_{12}:z$

8177:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8178:  & $-$ \\[1em]

8179: 

8180: \caption*{{\normalsize {\sc Table} \thetable. Products $\mu^2$ in the

8181: directed $A_\infty$-category $\Cat_{12}^\rightarrow$. The notation

8182: means that $\mu^2(a_2,a_1) = \pm b$. \label{table:mu2}}}

8183: \\

8184: \end{longtable}

8185: }

8186: 

8187: { \smaller\smaller

8188: \begin{longtable}[l]{lll|l|l}

8189:  $a_1$ & $a_2$ & $a_3$ & $b$ & $\pm$ \\

8190:  \hline

8191:  $\nu_1 \cap \nu_2:x$

8192:  & $\nu_2 \cap \nu_3:my^{-1}$

8193:  & $\nu_3 \cap \nu_5:x^{-1}y$

8194:  & $\nu_1 \cap \nu_5:1$

8195:  & $+$ \\

8196: $\nu_2 \cap \nu_4:mx^{-1}z$

8197:  & $\nu_4 \cap \nu_5:1$

8198:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8199:  & $\nu_2 \cap \nu_6:mx^{-1}y^{-1}$

8200:  & $-$ \\

8201: $\nu_4 \cap \nu_5:1$

8202:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8203:  & $\nu_6 \cap \nu_8:1$

8204:  & $\nu_4 \cap \nu_8:y^{-1}z^{-1}$

8205:  & $+$ \\

8206: $\nu_5 \cap \nu_7:mz$

8207:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8208:  & $\nu_8 \cap \nu_9:my^{-1}$

8209:  & $\nu_5 \cap \nu_9:my^{-2}$

8210:  & $-$ \\

8211: $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8212:  & $\nu_8 \cap \nu_9:my^{-1}$

8213:  & $\nu_9 \cap \nu_{11}:y$

8214:  & $\nu_7 \cap \nu_{11}:y^{-1}z^{-1}$

8215:  & $+$ \\

8216: $\nu_8 \cap \nu_{10}:mx^{-1}z$

8217:  & $\nu_{10} \cap \nu_{11}:x$

8218:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8219:  & $\nu_8 \cap \nu_{12}:my^{-1}z$

8220:  & $-$ \\

8221: $\nu_1 \cap \nu_2:x$

8222:  & $\nu_2 \cap \nu_3:my^{-1}$

8223:  & $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

8224:  & $\nu_1 \cap \nu_8:z^{-1}$

8225:  & $-$ \\

8226: $\nu_1 \cap \nu_2:x$

8227:  & $\nu_2 \cap \nu_3:my^{-1}$

8228:  & $\nu_3 \cap \nu_8:z^{-1}$

8229:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

8230:  & $+$ \\

8231: $\nu_2 \cap \nu_3:my^{-1}$

8232:  & $\nu_3 \cap \nu_5:x^{-1}y$

8233:  & $\nu_5 \cap \nu_7:mz$

8234:  & $\nu_2 \cap \nu_7:mx^{-1}z$

8235:  & $+$ \\

8236: $\nu_2 \cap \nu_7:mz$

8237:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8238:  & $\nu_8 \cap \nu_9:my^{-1}$

8239:  & $\nu_2 \cap \nu_9:my^{-2}$

8240:  & $-$ \\

8241: $\nu_2 \cap \nu_7:mx^{-1}yz$

8242:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8243:  & $\nu_8 \cap \nu_9:my^{-1}$

8244:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

8245:  & $+$ \\

8246: $\nu_1 \cap \nu_2:x$

8247:  & $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

8248:  & $\nu_6 \cap \nu_8:1$

8249:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

8250:  & $+$ \\

8251: $\nu_1 \cap \nu_2:x$

8252:  & $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

8253:  & $\nu_6 \cap \nu_8:1$

8254:  & $\nu_1 \cap \nu_8:z^{-1}$

8255:  & $+$ \\

8256: $\nu_1 \cap \nu_5:x$

8257:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8258:  & $\nu_6 \cap \nu_8:1$

8259:  & $\nu_1 \cap \nu_8:xy^{-1}z^{-1}$

8260:  & $+$ \\

8261: $\nu_1 \cap \nu_5:y$

8262:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8263:  & $\nu_6 \cap \nu_8:1$

8264:  & $\nu_1 \cap \nu_8:z^{-1}$

8265:  & $+$ \\

8266: $\nu_2 \cap \nu_4:mx^{-1}z$

8267:  & $\nu_4 \cap \nu_5:1$

8268:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

8269:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

8270:  & $-$ \\

8271: $\nu_2 \cap \nu_4:mx^{-1}z$

8272:  & $\nu_4 \cap \nu_5:1$

8273:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

8274:  & $\nu_2 \cap \nu_9:my^{-2}$

8275:  & $+$ \\

8276: $\nu_2 \cap \nu_4:mx^{-1}z$

8277:  & $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

8278:  & $\nu_8 \cap \nu_9:my^{-1}$

8279:  & $\nu_2 \cap \nu_9:my^{-2}$

8280:  & $+$ \\

8281: $\nu_2 \cap \nu_4:mx^{-1}z$

8282:  & $\nu_4 \cap \nu_8:z^{-1}$

8283:  & $\nu_8 \cap \nu_9:my^{-1}$

8284:  & $\nu_2 \cap \nu_9:mx^{-1}y^{-1}$

8285:  & $-$ \\

8286: $\nu_3 \cap \nu_5:x^{-1}y$

8287:  & $\nu_5 \cap \nu_7:mz$

8288:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8289:  & $\nu_3 \cap \nu_8:x^{-1}$

8290:  & $+$ \\

8291: $\nu_4 \cap \nu_5:1$

8292:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8293:  & $\nu_6 \cap \nu_{11}:y$

8294:  & $\nu_4 \cap \nu_{11}:z^{-1}$

8295:  & $-$ \\

8296: $\nu_4 \cap \nu_5:1$

8297:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8298:  & $\nu_6 \cap \nu_{11}:x$

8299:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

8300:  & $+$ \\

8301: $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8302:  & $\nu_6 \cap \nu_8:1$

8303:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8304:  & $\nu_5 \cap \nu_{10}:mx^{-1}y^{-1}$

8305:  & $+$ \\

8306: $\nu_5 \cap \nu_{10}:my^{-1}$

8307:  & $\nu_{10} \cap \nu_{11}:x$

8308:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8309:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

8310:  & $-$ \\

8311: $\nu_5 \cap \nu_{10}:mx^{-1}$

8312:  & $\nu_{10} \cap \nu_{11}:x$

8313:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8314:  & $\nu_5 \cap \nu_{12}:my^{-1}$

8315:  & $+$ \\

8316: $\nu_4 \cap \nu_5:1$

8317:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

8318:  & $\nu_9 \cap \nu_{11}:y$

8319:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

8320:  & $+$ \\

8321: $\nu_4 \cap \nu_5:1$

8322:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

8323:  & $\nu_9 \cap \nu_{11}:y$

8324:  & $\nu_4 \cap \nu_{11}:z^{-1}$

8325:  & $+$ \\

8326: $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

8327:  & $\nu_8 \cap \nu_9:my^{-1}$

8328:  & $\nu_9 \cap \nu_{11}:y$

8329:  & $\nu_4 \cap \nu_{11}:xy^{-1}z^{-1}$

8330:  & $+$ \\

8331: $\nu_4 \cap \nu_8:z^{-1}$

8332:  & $\nu_8 \cap \nu_9:my^{-1}$

8333:  & $\nu_9 \cap \nu_{11}:y$

8334:  & $\nu_4 \cap \nu_{11}:z^{-1}$

8335:  & $+$ \\

8336: $\nu_5 \cap \nu_7:mz$

8337:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8338:  & $\nu_8 \cap \nu_{12}:m$

8339:  & $\nu_5 \cap \nu_{12}:my^{-1}$

8340:  & $-$ \\

8341: $\nu_5 \cap \nu_7:mz$

8342:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8343:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

8344:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

8345:  & $+$ \\

8346: $\nu_5 \cap \nu_7:mz$

8347:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

8348:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8349:  & $\nu_5 \cap \nu_{12}:mxy^{-2}$

8350:  & $+$ \\

8351: $\nu_5 \cap \nu_7:mz$

8352:  & $\nu_7 \cap \nu_{11}:z^{-1}$

8353:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8354:  & $\nu_5 \cap \nu_{12}:my^{-1}$

8355:  & $-$ \\

8356: $\nu_6 \cap \nu_8:1$

8357:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8358:  & $\nu_{10} \cap \nu_{11}:x$

8359:  & $\nu_6 \cap \nu_{11}:z$

8360:  & $+$ \\

8361: $\nu_1 \cap \nu_2:x$

8362:  & $\nu_2 \cap \nu_3:my^{-1}$

8363:  & $\nu_3 \cap \nu_{11}:yz^{-1}$

8364:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8365:  & $-$ \\

8366: $\nu_2 \cap \nu_{10}:mx^{-1}$

8367:  & $\nu_{10} \cap \nu_{11}:x$

8368:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8369:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8370:  & $-$ \\

8371: $\nu_1 \cap \nu_2:x$

8372:  & $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

8373:  & $\nu_6 \cap \nu_{11}:x$

8374:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8375:  & $-$ \\

8376: $\nu_1 \cap \nu_2:x$

8377:  & $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

8378:  & $\nu_6 \cap \nu_{11}:y$

8379:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8380:  & $-$ \\

8381: $\nu_1 \cap \nu_2:x$

8382:  & $\nu_2 \cap \nu_9:my^{-1}z^{-1}$

8383:  & $\nu_9 \cap \nu_{11}:y$

8384:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8385:  & $+$ \\

8386: $\nu_1 \cap \nu_5:y$

8387:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8388:  & $\nu_6 \cap \nu_{11}:x$

8389:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8390:  & $-$ \\

8391: $\nu_1 \cap \nu_5:x$

8392:  & $\nu_5 \cap \nu_6:my^{-1}z^{-1}$

8393:  & $\nu_6 \cap \nu_{11}:y$

8394:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8395:  & $-$ \\

8396: $\nu_2 \cap \nu_3:my^{-1}$

8397:  & $\nu_3 \cap \nu_5:x^{-1}y$

8398:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

8399:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

8400:  & $-$ \\

8401: $\nu_2 \cap \nu_3:my^{-1}$

8402:  & $\nu_3 \cap \nu_5:x^{-1}y$

8403:  & $\nu_5 \cap \nu_{10}:my^{-1}$

8404:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

8405:  & $-$ \\

8406: $\nu_2 \cap \nu_3:my^{-1}$

8407:  & $\nu_3 \cap \nu_8:z^{-1}$

8408:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8409:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

8410:  & $+$ \\

8411: $\nu_2 \cap \nu_3:my^{-1}$

8412:  & $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

8413:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8414:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

8415:  & $-$ \\

8416: $\nu_2 \cap \nu_6:my^{-1}z^{-1}$

8417:  & $\nu_6 \cap \nu_8:1$

8418:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8419:  & $\nu_2 \cap \nu_{10}:mx^{-1}y^{-1}$

8420:  & $+$ \\

8421: $\nu_2 \cap \nu_6:mx^{-1}z^{-1}$

8422:  & $\nu_6 \cap \nu_8:1$

8423:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8424:  & $\nu_2 \cap \nu_{10}:mx^{-2}$

8425:  & $+$ \\

8426: $\nu_2 \cap \nu_4:mx^{-1}z$

8427:  & $\nu_4 \cap \nu_8:xy^{-1}z^{-1}$

8428:  & $\nu_8 \cap \nu_{12}:m$

8429:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8430:  & $-$ \\

8431: $\nu_2 \cap \nu_4:mx^{-1}z$

8432:  & $\nu_4 \cap \nu_8:z^{-1}$

8433:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

8434:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8435:  & $+$ \\

8436: $\nu_3 \cap \nu_5:x^{-1}y$

8437:  & $\nu_5 \cap \nu_7:mz$

8438:  & $\nu_7 \cap \nu_{11}:z^{-1}$

8439:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

8440:  & $+$ \\

8441: $\nu_3 \cap \nu_5:x^{-1}y$

8442:  & $\nu_5 \cap \nu_7:mz$

8443:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

8444:  & $\nu_3 \cap \nu_{11}:1$

8445:  & $+$ \\

8446: $\nu_1 \cap \nu_8:xz^{-1}$

8447:  & $\nu_8 \cap \nu_9:my^{-1}$

8448:  & $\nu_9 \cap \nu_{11}:y$

8449:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8450:  & $-$ \\

8451: $\nu_3 \cap \nu_5:x^{-1}y$

8452:  & $\nu_5 \cap \nu_{10}:mx^{-1}$

8453:  & $\nu_{10} \cap \nu_{11}:x$

8454:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

8455:  & $-$ \\

8456: $\nu_3 \cap \nu_5:x^{-1}y$

8457:  & $\nu_5 \cap \nu_{10}:my^{-1}$

8458:  & $\nu_{10} \cap \nu_{11}:x$

8459:  & $\nu_3 \cap \nu_{11}:1$

8460:  & $-$ \\

8461: $\nu_3 \cap \nu_8:x^{-1}yz^{-1}$

8462:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8463:  & $\nu_{10} \cap \nu_{11}:x$

8464:  & $\nu_3 \cap \nu_{11}:x^{-1}y$

8465:  & $-$ \\

8466: $\nu_3 \cap \nu_8:z^{-1}$

8467:  & $\nu_8 \cap \nu_{10}:mx^{-1}z$

8468:  & $\nu_{10} \cap \nu_{11}:x$

8469:  & $\nu_3 \cap \nu_{11}:1$

8470:  & $+$ \\

8471: $\nu_1 \cap \nu_5:x$

8472:  & $\nu_5 \cap \nu_9:my^{-1}z^{-1}$

8473:  & $\nu_9 \cap \nu_{11}:y$

8474:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8475:  & $+$ \\

8476: $\nu_1 \cap \nu_5:y$

8477:  & $\nu_5 \cap \nu_9:mxy^{-2}z^{-1}$

8478:  & $\nu_9 \cap \nu_{11}:y$

8479:  & $\nu_1 \cap \nu_{11}:xz^{-1}$

8480:  & $-$ \\

8481: $\nu_2 \cap \nu_4:mx^{-1}z$

8482:  & $\nu_4 \cap \nu_5:1$

8483:  & $\nu_5 \cap \nu_{12}:mxy^{-1}z^{-1}$

8484:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8485:  & $-$ \\

8486: $\nu_2 \cap \nu_4:mx^{-1}z$

8487:  & $\nu_4 \cap \nu_{11}:xz^{-1}$

8488:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8489:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8490:  & $+$ \\

8491: $\nu_2 \cap \nu_7:mx^{-1}yz$

8492:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8493:  & $\nu_8 \cap \nu_{12}:mxy^{-1}$

8494:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8495:  & $-$ \\

8496: $\nu_2 \cap \nu_7:mz$

8497:  & $\nu_7 \cap \nu_8:y^{-1}z^{-1}$

8498:  & $\nu_8 \cap \nu_{12}:m$

8499:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8500:  & $+$ \\

8501: $\nu_2 \cap \nu_7:mz$

8502:  & $\nu_7 \cap \nu_{11}:z^{-1}$

8503:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8504:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8505:  & $+$ \\

8506: $\nu_2 \cap \nu_7:mx^{-1}yz$

8507:  & $\nu_7 \cap \nu_{11}:xy^{-1}z^{-1}$

8508:  & $\nu_{11} \cap \nu_{12}:my^{-1}$

8509:  & $\nu_2 \cap \nu_{12}:my^{-1}$

8510:  & $-$ \\[1em]

8511: 

8512: \caption*{{\normalsize {\sc Table} \thetable. Products $\mu^3$ in

8513: $\Cat_{12}^\rightarrow$, written in the same way as in the previous

8514: table. \label{table:mu3}}}

8515: \\

8516: \end{longtable}

8517: }

8518: 

8519: %\bibliographystyle{amsplain}

8520: %\bibliography{../../../bib/all,../../../bib/new}

8521: \providecommand{\bysame}{\leavevmode\hbox

8522: to3em{\hrulefill}\thinspace}

8523: \providecommand{\MR}{\relax\ifhmode\unskip\space\fi MR }

8524: % \MRhref is called by the amsart/book/proc definition of \MR.

8525: \providecommand{\MRhref}[2]{%

8526:   \href{http://www.ams.org/mathscinet-getitem?mr=#1}{#2}

8527: } \providecommand{\href}[2]{#2}

8528: \begin{thebibliography}{10}

8529: 

8530: \bibitem{arnold95}

8531: V.~I. Arnol'd, \emph{Some remarks on symplectic monodromy of {M}ilnor

8532:   fibrations}, The {F}loer Memorial Volume (H.~Hofer, C.~Taubes, A.~Weinstein,

8533:   and E.~Zehnder, eds.), Progress in Mathematics, vol. 133, Birkh{\"a}user,

8534:   1995, pp.~99--104.

8535: 

8536: \bibitem{aspinwall01}

8537: P.~S. Aspinwall, \emph{Some navigation rules for {$D$}-brane

8538: monodromy}, J.

8539:   Math. Phys. \textbf{42} (2001), 5534--5552.

8540: 

8541: \bibitem{auroux99}

8542: D.~Auroux, \emph{{S}ymplectic 4-manifolds as branched coverings of

8543:   {$\CP{2}$}.}, Invent. Math. \textbf{139} (2000), 551--602.

8544: 

8545: \bibitem{auroux00}

8546: \bysame, \emph{Symplectic maps to projective spaces and symplectic

8547: invariants},

8548:   Turkish J. Math. \textbf{25} (2001), 1--42.

8549: 

8550: \bibitem{auroux-donaldson-katzarkov-yotov01}

8551: D.~Auroux, S.~Donaldson, L.~Katzarkov, and M.~Yotov,

8552: \emph{Fundamental groups

8553:   of complements of plane curves and symplectic invariants}, Preprint

8554:   math.GT/0203183.

8555: 

8556: \bibitem{balmer-schlichting01}

8557: P.~Balmer and M.~Schlichting, \emph{Idempotent completion of

8558: triangulated

8559:   categories}, J. Algebra \textbf{236} (2001), 819--834.

8560: 

8561: \bibitem{beilinson78}

8562: A.~A. Beilinson, \emph{Coherent sheaves on {$\mathbb{P}^n$} and

8563: problems of

8564:   linear algebra}, Funct. Anal. Appl. \textbf{12} (1978), 214--216.

8565: 

8566: \bibitem{blattner-montgomery85}

8567: R.~Blattner and S.~Montgomery, \emph{A duality theorem for {H}opf

8568: module

8569:   algebras}, J. Algebra \textbf{95} (1985), 153--172.

8570: 

8571: \bibitem{bondal-vandenbergh02}

8572: A.~Bondal and M.~Van den Bergh, \emph{Generators and representability

8573: of

8574:   functors in commutative and noncommutative geometry}, Preprint

8575:   math.AG/\-0204218.

8576: 

8577: \bibitem{chekanov99}

8578: Yu. Chekanov, \emph{Differential algebras of {L}egendrian links},

8579: Invent. Math.

8580:   \textbf{150} (2002), 441--483.

8581: 

8582: \bibitem{donaldson02}

8583: S.~K. Donaldson, \emph{Lefschetz pencils on symplectic manifolds}, J.

8584:   Differential Geom. \textbf{53} (1999), 205--236.

8585: 

8586: \bibitem{donaldson98}

8587: \bysame, \emph{Polynomials, vanishing cycles, and {F}loer homology},

8588:   Mathematics: frontiers and perspectives, Amer. Math. Soc., 2000, pp.~55--64.

8589: 

8590: \bibitem{douglas-govindarajan-jayaraman-tomasiello01}

8591: M.~Douglas, S.~Govindarajan, T.~Jayaraman, and A.~Tomasiello,

8592: \emph{{D}-branes

8593:   on {C}alabi-{Y}au manifolds and superpotentials}, Preprint hep-th/0203173.

8594: 

8595: \bibitem{fukaya02}

8596: K.~Fukaya, \emph{Asymptotic analysis, multivalued {M}orse theory, and

8597: mirror

8598:   symmetry}, Preprint 2002.

8599: 

8600: \bibitem{fukaya93}

8601: \bysame, \emph{Morse homotopy, {$A_\infty$}-categories, and {F}loer

8602:   homologies}, Proceedings of {GARC} workshop on Geometry and Topology (H.~J.

8603:   Kim, ed.), Seoul National University, 1993.

8604: 

8605: \bibitem{fukaya00}

8606: \bysame, \emph{Deformation theory, homological algebra, and mirror

8607: symmetry},

8608:   Geometry and physics of branes (Como, 2001), IOP, 2003, pp.~121--209.

8609: 

8610: \bibitem{fukaya03}

8611: \bysame, \emph{Galois symmetry on {F}loer cohomology}, Turkish J.

8612: Math.

8613:   \textbf{27} (2003), 11--32.

8614: 

8615: \bibitem{fooo}

8616: K.~Fukaya, Y.-G. Oh, H.~Ohta, and K.~Ono, \emph{Lagrangian

8617: intersection {F}loer

8618:   theory - anomaly and obstruction}, Preprint, 2000.

8619: 

8620: \bibitem{fukaya02b}

8621: Kenji Fukaya, \emph{Mirror symmetry of abelian varieties and

8622: multi-theta

8623:   functions}, J. Algebraic Geom. \textbf{11} (2002), 393--512.

8624: 

8625: \bibitem{gelfand-manin}

8626: S.~Gelfand and Yu. Manin, \emph{Methods of homological algebra},

8627: Springer,

8628:   1996.

8629: 

8630: \bibitem{getzler-jones90}

8631: E.~Getzler and J.~D.~S. Jones, \emph{{$A_\infty$}-algebras and the

8632: cyclic bar

8633:   complex}, Illinois J. Math. \textbf{34} (1990), 256--283.

8634: 

8635: \bibitem{goldman-millson88}

8636: W.~Goldman and J.~Millson, \emph{The deformation theory of the

8637: fundamental

8638:   group of compact {K}{\"a}hler manifolds}, IHES Publ. Math. \textbf{67}

8639:   (1988), 43--96.

8640: 

8641: \bibitem{gompf-stipsicz}

8642: R.~Gompf and A.~Stipsicz, \emph{$4$-manifolds and {K}irby calculus},

8643: Amer.

8644:   Math. Soc., 1999.

8645: 

8646: \bibitem{gross01b}

8647: M.~Gross, \emph{Examples of special {L}agrangian fibrations},

8648: Symplectic

8649:   geometry and mirror symmetry (Seoul), World Sci. Publishing, 2001,

8650:   pp.~81--109.

8651: 

8652: \bibitem{guccione-guccione01}

8653: J.~Guccione and J.~Guccione, \emph{Hochschild cohomology of

8654: triangular matrix

8655:   algebras}, Preprint math.KT/0104068.

8656: 

8657: \bibitem{gugenheim-lambe-stasheff90}

8658: V.~K. A.~M. Gugenheim, L.~A. Lambe, and J.~D. Stasheff,

8659: \emph{Algebraic aspects

8660:   of {C}hen's twisting cochain}, Illinois J. Math. \textbf{34} (1990),

8661:   485--502.

8662: 

8663: \bibitem{halperin-stasheff79}

8664: S.~Halperin and J.~Stasheff, \emph{Obstructions to homotopy

8665: equivalences},

8666:   Advances in Math. \textbf{32} (1979), 233--279.

8667: 

8668: \bibitem{hartshorne}

8669: R.~Hartshorne, \emph{Algebraic geometry}, Springer, 1977.

8670: 

8671: \bibitem{hochschild-kostant-rosenberg62}

8672: G.~Hochschild, B.~Kostant, and A.~Rosenberg, \emph{Differential forms

8673: on

8674:   regular affine algebras}, Trans. Amer. Math. Soc. \textbf{102} (1962),

8675:   383--408.

8676: 

8677: \bibitem{hofer-salamon95}

8678: H.~Hofer and D.~Salamon, \emph{Floer homology and {N}ovikov rings},

8679: The {F}loer

8680:   memorial volume (H.~Hofer, C.~Taubes, A.~Weinstein, and E.~Zehnder, eds.),

8681:   Progress in Mathematics, vol. 133, Birkh{\"a}user, 1995, pp.~483--524.

8682: 

8683: \bibitem{hori-iqbal-vafa00}

8684: K.~Hori, A.~Iqbal, and C.~Vafa, \emph{{$D$}-branes and mirror

8685: symmetry},

8686:   Preprint hep-th/0005247.

8687: 

8688: \bibitem{kadeishvili88}

8689: T.~V. Kadeishvili, \emph{The structure of the {$A_\infty$}-algebra,

8690: and the

8691:   {H}ochschild and {H}arrison cohomologies}, Trudy Tbiliss. Mat. Inst. Razmadze

8692:   Akad. Nauk Gruzin. SSR \textbf{91} (1988), 19--27 (Russian).

8693: 

8694: \bibitem{kapranov-vasserot00}

8695: M.~Kapranov and E.~Vasserot, \emph{Kleinian singularities, derived

8696: categories

8697:   and {H}all algebras}, Math. Ann. \textbf{316} (2000), 565--576.

8698: 

8699: \bibitem{keller99}

8700: B.~Keller, \emph{Introduction to {$A$}-infinity algebras and

8701: modules}, Homology

8702:   Homotopy Appl. (electronic) \textbf{3} (2001), 1--35.

8703: 

8704: \bibitem{kontsevich98}

8705: M.~Kontsevich, \emph{Lectures at {ENS} {P}aris, {S}pring 1998}, set

8706: of notes

8707:   taken by J.\ Bellaiche, J.-F. Dat, I. Marin, G. Racinet and H.

8708:   Randriambololona.

8709: 

8710: \bibitem{kontsevich94}

8711: \bysame, \emph{Homological algebra of mirror symmetry}, Proceedings

8712: of the

8713:   International Congress of Mathematicians (Z{\"u}rich, 1994), Birkh{\"a}user,

8714:   1995, pp.~120--139.

8715: 

8716: \bibitem{kontsevich-soibelman00}

8717: M.~Kontsevich and Y.~Soibelman, \emph{Homological mirror symmetry and

8718: torus

8719:   fibrations}, Symplectic geometry and mirror symmetry, World Scientific, 2001,

8720:   pp.~203--263.

8721: 

8722: \bibitem{kwon-oh96}

8723: D.~Kwon and Y.-G. Oh, \emph{Structure of the image of

8724: (pseudo)-holomorphic

8725:   discs with totally real boundary conditions}, Comm. Anal. Geom. \textbf{8}

8726:   (2000), 31--82.

8727: 

8728: \bibitem{lazzarini98}

8729: L.~Lazzarini, \emph{Existence of a somewhere injective

8730: pseudo-holomorphic

8731:   disc}, Geom. Funct. Anal. \textbf{10} (2000), 829--862.

8732: 

8733: \bibitem{lefevre}

8734: K.~Lefevre, \emph{Sur les {$A_\infty$}-cat\'egories}, Ph.D. thesis,

8735:   Universit\'e Paris 7, 2002.

8736: 

8737: \bibitem{levine}

8738: M.~Levine, \emph{Mixed motives}, Amer. Math. Soc., 1998.

8739: 

8740: \bibitem{loday}

8741: J.-L. Loday, \emph{Cyclic homology}, Springer, 2nd edition, 1997.

8742: 

8743: \bibitem{mikhalkin01}

8744: G.~Mikhalkin, \emph{Amoebas of algebraic varieties}, Preprint

8745: math.AG/0108225.

8746: 

8747: \bibitem{montgomery92}

8748: S.~Montgomery, \emph{Hopf algebras and their actions on rings},

8749: {CBMS} Regional

8750:   Conference Series, vol.~82, Amer. Math. Soc., 1992.

8751: 

8752: \bibitem{neeman}

8753: A.~Neeman, \emph{Triangulated categories}, Annals of Math. Studies,

8754: vol. 148,

8755:   Princeton Univ. Press, 2001.

8756: 

8757: \bibitem{nishinou03}

8758: T.~Nishinou, \emph{Lagrangian submanifold associated with a

8759: degenerating family

8760:   of stable bundles on $t^4$}, Preprint math.SG/0301324.

8761: 

8762: \bibitem{oh96}

8763: Y.-G. Oh, \emph{Floer cohomology, spectral sequences, and the

8764: {M}aslov class of

8765:   {L}agrangian embeddings}, Int. Math. Res. Notices (1996), 305--346.

8766: 

8767: \bibitem{piunikhin-salamon-schwarz94}

8768: S.~Piunikhin, D.~Salamon, and M.~Schwarz, \emph{Symplectic

8769: {F}loer-{D}onaldson

8770:   theory and quantum cohomology}, Contact and symplectic geometry (C.~B.

8771:   Thomas, ed.), Cambridge Univ. Press, 1996, pp.~171--200.

8772: 

8773: \bibitem{polishchuk98}

8774: A.~Polishchuk, \emph{{M}assey and {F}ukaya products on elliptic

8775: curves}, Adv.

8776:   Theor. Math. Phys. \textbf{4} (2000), 1187--1207.

8777: 

8778: \bibitem{polishchuk-zaslow98}

8779: A.~Polishchuk and E.~Zaslow, \emph{Categorical mirror symmetry: the

8780: elliptic

8781:   curve}, Adv. Theor. Math. Phys. \textbf{2} (1998), 443--470.

8782: 

8783: \bibitem{robbin-salamon93}

8784: J.~Robbin and D.~Salamon, \emph{The {M}aslov index for paths},

8785: Topology

8786:   \textbf{32} (1993), 827--844.

8787: 

8788: \bibitem{ruan02c}

8789: W.-D. Ruan, \emph{Newton polygon and string diagram}, Preprint

8790:   math.DG/\-0011012.

8791: 

8792: \bibitem{ruan02a}

8793: \bysame, \emph{Lagrangian torus fibration of quintic hypersurfaces.

8794: {I}.

8795:   {F}ermat quintic case}, Winter School on Mirror Symmetry, Vector Bundles and

8796:   Lagrangian Submanifolds (Cambridge, MA, 1999), AMS/IP Stud. Adv. Math.,

8797:   vol.~23, Amer. Math. Soc., 2001, pp.~297--332.

8798: 

8799: \bibitem{ruan02b}

8800: \bysame, \emph{Lagrangian torus fibration of quintic {C}alabi-{Y}au

8801:   hypersurfaces. {II}. {T}echnical results on gradient flow construction}, J.

8802:   Symplectic Geom. \textbf{1} (2002), 435--521.

8803: 

8804: \bibitem{rudakov90}

8805: A.~Rudakov et~al., \emph{Helices and vector bundles: {S}eminaire

8806: {R}udakov},

8807:   LMS Lecture Note Series, vol. 148, Cambridge University Press, 1990.

8808: 

8809: \bibitem{salamon-zehnder92}

8810: D.~Salamon and E.~Zehnder, \emph{Morse theory for periodic solutions

8811: of

8812:   {H}amiltonian systems and the {M}aslov index}, Comm. Pure Appl. Math.

8813:   \textbf{45} (1992), 1303--1360.

8814: 

8815: \bibitem{seidel99}

8816: P.~Seidel, \emph{Graded {L}agrangian submanifolds}, Bull. Soc. Math.

8817: France

8818:   \textbf{128} (2000), 103--146.

8819: 

8820: \bibitem{seidel00}

8821: \bysame, \emph{Vanishing cycles and mutation}, Proceedings of the 3rd

8822: European

8823:   Congress of Mathematics (Barcelona), vol.~2, Birkh{\"a}user, 2001,

8824:   pp.~65--85.

8825: 

8826: \bibitem{seidel02}

8827: \bysame, \emph{Fukaya categories and deformations}, Proceedings of

8828: the

8829:   International Congress of Mathematicians (Beijing), vol.~2, Higher Ed. Press,

8830:   2002, pp.~351--360.

8831: 

8832: \bibitem{seidel01}

8833: \bysame, \emph{A long exact sequence for symplectic {F}loer

8834: cohomology},

8835:   Topology \textbf{42} (2003), 1003--1063.

8836: 

8837: \bibitem{seidel04}

8838: \bysame, \emph{{F}ukaya categories and {P}icard-{L}efschetz theory},

8839:   {M}anuscript in preparation, 2004?

8840: 

8841: \bibitem{seidel-thomas99}

8842: P.~Seidel and R.~Thomas, \emph{Braid group actions on derived

8843: categories of

8844:   coherent sheaves}, Duke Math. J. \textbf{108} (2001), 37--108.

8845: 

8846: \bibitem{desilva98}

8847: V.~De Silva, \emph{Products in the symplectic {F}loer homology of

8848: {L}agrangian

8849:   intersections}, Ph.D. thesis, Oxford University, 1998.

8850: 

8851: \bibitem{spaltenstein88}

8852: N.~Spaltenstein, \emph{Resolutions of unbounded complexes},

8853: Compositio Math.

8854:   \textbf{65} (1988), 121--154.

8855: 

8856: \bibitem{thomas-yau01}

8857: R.~Thomas and S.-T. Yau, \emph{Special {L}agrangians, stable bundles

8858: and mean

8859:   curvature flow.}, Comm. Anal. Geom. \textbf{10} (2002), 1075--1113.

8860: 

8861: \bibitem{tradler01}

8862: T.~Tradler, \emph{Infinity-inner-products on {A}-infinity-algebras},

8863: Preprint

8864:   math.\-AT/0108027.

8865: 

8866: \end{thebibliography}

8867: 

8868: \end{document}

8869: