1: \documentclass [11pt] {amsart} 

2: \usepackage {amsmath, amssymb, amscd, mathrsfs, graphicx, color, url}

3: 

4: \setlength{\oddsidemargin}{.15in} \setlength{\evensidemargin}{.15in} 

5: \setlength{\textwidth}{6in} \setlength{\textheight}{8.75in} 

6: \setlength{\topmargin}{0pt} \setlength{\headheight}{.2in} 

7: \setlength{\parskip}{0pt} \setlength{\labelsep}{10pt} 

8: \setlength{\parindent}{12pt} \setlength{\medskipamount}{2ex} 

9: \setlength{\smallskipamount}{1ex} \newenvironment{List}{%

10:     \begin{list}{}{\setlength{\labelwidth}{.15in}

11:     \setlength{\leftmargin}{.55in}

12:     \setlength{\rightmargin}{.25in}

13:     \setlength{\topsep}{0pt}

14:     \setlength{\partopsep}{0pt}

15:     }}{\end{list}}

16: 

17: \newtheorem {theorem}{Theorem}[section]

18: \newtheorem {definition}[theorem]{Definition}

19: \newtheorem {lemma}[theorem]{Lemma}

20: \newtheorem {proposition}[theorem]{Proposition}

21: \newtheorem {corollary}[theorem]{Corollary}

22: \newtheorem {conjecture}[theorem]{Conjecture}

23: \newtheorem {question}[theorem]{Question}

24: \newtheorem {remark}[theorem]{Remark}

25: \newtheorem {example}[theorem]{Example}

26: \newtheorem {fact}[theorem]{Fact}

27: \newcommand{\abs}[1]{\lvert#1\rvert}

28: 

29: \def\zz {{\mathbb{Z}}}

30: \def\rr {{\mathbb{R}}}

31: \def\cc {{\mathbb{C}}}

32: \def\qq {{\mathbb{Q}}}

33: \def\pp {{\mathbb{P}}}

34: 

35: \def\ze {{\mathcal{Z}}}

36: \def\bb {{\mathcal{B}}}

37: \def\gz {{\mathcal{G}}}

38: \def\oo {{\mathcal{O}}}

39: \def\ss {{\mathcal{S}}}

40: \def\ii {{\mathcal{I}}}

41: \def\lll {{\mathcal{L}}}

42: \def\rrr {{\mathcal{R}}}

43: \def\pc {{\mathcal{P}}}

44: \def\ss {{\mathcal{S}}}

45: \def\ccc {{\mathcal{C}}}

46: \def\yy {{\mathcal{Y}}}

47: \def\nnn {{\mathcal{N}}}

48: \def\ua {{\mathcal{U}}}

49: 

50: \def\hh {{\mathfrak{h}}}

51: \def\gg {{\mathfrak{g}}}

52: \def\sl {{\mathfrak{sl}}}

53: \def\so {{\mathfrak{so}}}

54: \def\spc {{\mathfrak{s}}}

55: \def\gl {{\mathfrak{gl}}}

56: \def\lfrak {{\mathfrak{L}}}

57: \def\tfrak {{\mathfrak{T}}}

58: \def\ggg {{\mathfrak{g}}}

59: \def\hhh {{\mathfrak{h}}}

60: \def\ppp {{\mathfrak{p}}}

61: 

62: \def\cpn {{\cc \pp^{n-1}}}

63: \def\ka {{L}}

64: \def\lk {{\kappa}}

65: \def\nn {{(n)}}

66: \def\ymnt {{\yy_{m, n, \tau}}}

67: \def\lambdav {{\vec{\lambda}}}

68: \def\muv {{\vec{\mu}}}

69: \def\inte {{\text{int}}}

70: 

71: \def\ff {{\mathscr{F}}}

72: \def\ee {{\mathscr{E}}}

73: \def\kr {{\mathscr{H}}}

74: \def\bb {{\mathscr{B}}}

75: \def\xx {{\mathscr{X}}}

76: \def\tt {{\mathscr{T}}}

77: 

78: \def\Reg {{\operatorname {Reg} \ }}

79: \def\symp {{\scriptscriptstyle{\mathrm{symp}}}}

80: \def\mapp {{q}}

81: \def\krs {{\kr_{\nn \symp}}}

82: \def\krss {{\kr_{\nn \symp}^*}}

83: \def\krsk {{\kr_{\nn \symp}^k}}

84: \def\ggp {{\hat \ggg^{\pi}}}

85: \def\gp {{\tilde \ggg^{\pi}}}

86: \def\gpi {{\ggg^{\pi}}}

87: \def\sln {{\sl(n)}}

88: \def\slan {{\sl(n, \cc)^{(1(n-1))}}}

89: \def\chin {{\chi^{(1(n-1))}}}

90: \def\vp {{\varphi}}   

91: \def\del {{\partial}}

92: \def\fin {{\hfill \square}}

93: \def\pr {{\text{pr}}}

94: \def\im {{\hspace{2pt} \text{Im }}}

95: \def\ker {{\hspace{2pt} \text{Ker }}}

96: \def\from {{\leftarrow}}

97: \def\gmn {{\gg^{\pi(m,n)}}}

98: \def\mat {{\mathfrak{m}}}

99: \def\lfrak {{\mathcal{L}}}

100: \def\pim {{\operatorname{proj}^{\scriptscriptstyle{\mathrm{im}}}}}

101: 

102: \def\jm {{\scriptscriptstyle{\mathrm{JM}}}}

103: \def\sjm {{\ss^{\jm}}}

104: \def\sing {{\operatorname{Sing}}}

105: \def\reg {{\scriptscriptstyle{\mathrm{reg}}}}

106: \def\red {{\scriptscriptstyle{\mathrm{red}}}}

107: \def\sub {{\scriptscriptstyle{\mathrm{sub}}}}

108: \def\resc {{\scriptscriptstyle{\mathrm{resc}}}}

109: \def\sym {{\operatorname{Sym}}}

110: \def\inv {{\operatorname{Inv}}}

111: \def\End {{\operatorname{End}}}

112: \def\hom {{\operatorname{Hom}}}

113: \def\id {{\operatorname{Id}}}  

114: \def\Id {{I}}

115: \def\iso {{\ \cong \ }}

116: 

117: % figures L0.eps, L+.eps, L_.eps are taken from the paper \cite{DGR}

118: 

119: \def\conf {{Conf}}

120: \def\confb {{BConf}}

121: 

122: \hbadness=100000

123: 

124: \begin{document}

125: 

126: \title [Link homology theories from symplectic geometry] 

127: {Link homology theories from symplectic geometry} \author 

128: [Ciprian Manolescu]{Ciprian Manolescu} 

129: \thanks {The author was supported by a Clay Research Fellowship.}

130: 

131: \begin {abstract} For each positive integer $n,$ Khovanov and Rozansky constructed an 

132: invariant of links in the form of a doubly-graded cohomology theory whose Euler 

133: characteristic is the $\sln$ link polynomial. We use Lagrangian Floer cohomology on 

134: some suitable affine varieties to build a similar series of link invariants, and we 

135: conjecture them to be equal to those of Khovanov and Rozansky after a 

136: collapse of 

137: the bigrading. Our work is a generalization of that of Seidel and Smith, who treated 

138: the case $n=2.$ \end {abstract}

139: 

140: \address {Department of Mathematics, Columbia University\\ 2990 Broadway, New York, 

141: NY 10027}

142: \email {cm@math.columbia.edu}

143: 

144: \maketitle

145: 

146: \section {Introduction}

147: 

148: For any $n > 0,$ the quantum $\sln$ polynomial invariant $P_{\nn}$ of an oriented link 

149: $\lk \subset S^3$ is uniquely determined by the skein relation: 

150: \begin{equation}

151: \label {skey}

152: q^n P_{\nn} \left( \, \includegraphics[scale=0.3]{L-.eps} \, \right)

153: - q^{-n} P_{\nn}  \left( \, \includegraphics[scale=0.3]{L+.eps} \,  \right)

154: = ( q - q^{-1}) P_{\nn} \left(\includegraphics[scale=0.3]{L0.eps} \right),

155: \end{equation}

156: together with the value on the unknot 

157: $P_{\nn}(\text{unknot}) = (q^n - q^{-n})/(q-q^{-1}).$ The invariant can also be defined 

158: in terms of the representation theory of the quantum group $U_q(\sln),$ hence the 

159: name.  When $n=2$ we obtain the Jones polynomial, up to a $q+q^{-1}$ factor. The same skein 

160: relation for $n=0$ with the normalization $P_{(0)}(\text{unknot})=1$ gives the 

161: Alexander polynomial, but the representation theoretic story is 

162: somewhat different in this case. The polynomials $P_{\nn}$ are all different 

163: specializations of a single link invariant, the two variable HOMFLY polynomial 

164: \cite{HOMFLY}, \cite{PT}.

165: 

166: Khovanov and Rozansky \cite{KR1} associated to every link $\lk$ a series of 

167: bigraded cohomology theories $\kr^{i, j}_{\nn}(\lk)$ for $n > 0$ and showed 

168: that they are link invariants. Their theories can be interpreted as 

169: categorifications of $P_{\nn},$ in the sense that $$ P_{\nn}(\lk) = \sum_{i, 

170: j\in \zz} (-1)^i q^j \dim_{\qq} \kr^{i, j}_{\nn} (\lk).$$ When $n=2,$ they 

171: recover the older categorification of the Jones polynomial due to Khovanov 

172: \cite{Kh1}.

173: 

174: Khovanov-Rozansky homology is particularly interesting because it is conjectured to 

175: be related to the knot Floer homology of Ozsv\'ath-Szab\'o and Rasmussen \cite{OS}, 

176: \cite{R}. Knot Floer homology is an invariant defined used Lagrangian Floer homology, 

177: and an important question is to find a way to compute it algorithmically. Its graded 

178: Euler characteristic is the Alexander polynomial corresponding to $n=0$ above. 

179: On the other hand, Khovanov-Rozansky homology is algorithmically computable by definition, 

180: and the hope is to be able to extract the case $n=0$ from the $n > 0$ theories. A 

181: precise conjecture in this direction was made by Dunfield, Gukov and Rasmussen in 

182: \cite{DGR}, and a potentially useful triply graded categorification of the HOMFLY 

183: polynomial was constructed by Khovanov and Rozansky in the sequel \cite{KR2}.

184: 

185: In this paper we construct a sequence of link invariants $\krss(\lk)$ using 

186: Lagrangian Floer theory. This has been done by Seidel and Smith in the case $n=2$ 

187: \cite{SS}, and our work is a generalization of theirs. We conjecture that our 

188: invariants are related to Khovanov-Rozansky homology of the mirror link $\lk^!$ in 

189: the following way:

190: 

191: \begin {conjecture}

192: \label {conj}

193: $$\krsk(\lk) \otimes \qq = \bigoplus_{i + j =k} \kr^{i, j}_{\nn}(\lk^!).$$

194: \end {conjecture}

195: 

196: Our construction is inspired from that of Seidel and Smith, with some differences coming from 

197: the fact that the standard (quantum) representation $V$ of $\sln$ is not self-dual for $n > 

198: 2.$ As in \cite{SS}, we start by presenting the link $\lk$ as the closure of an $m$-stranded 

199: braid $\beta \in Br_m.$ The rough idea is to find a symplectic manifold $(M, \omega)$ with an 

200: action of the braid group by symplectomorphisms $\phi: Br_m \to \pi_0(Symp (M, \omega)),$ to 

201: take a specific Lagrangian $L \subset M$ and to consider the Floer cohomology of $L$ and 

202: $\phi(\beta)L$ in $M.$ Following the ideas of Khovanov from \cite{Kh2}, we would like the 

203: Grothendieck group of the derived Fukaya category of $M$ to be related to the space $\inv 

204: (m,n)$ of invariants in the representation $V^{\otimes m} \otimes (V^{\otimes m})^*$ of 

205: $\sln.$ The reason is that $\inv (m,n)$ naturally occurs in the representation theoretic 

206: definition of the polynomial $P_{\nn}.$ On the other hand, the Grothendieck group of the 

207: derived Fukaya category is not a well-understood object in general, but it is related (and in 

208: some special cases equal) to the middle dimensional homology of $M.$ In the case $n=2$ for 

209: example, Seidel and Smith worked with a symplectic manifold whose middle dimensional Betti 

210: number is the $m$-th Catalan number, the same as the dimension of $\inv (m,2).$ Their 

211: construction uses the geometry of the adjoint quotient and nilpotent slices in the Lie algebra 

212: $\sl(2m),$ and our manifolds $M = \ymnt$ below are a natural extension of theirs, obtained by 

213: looking at the Lie algebra $\sl(mn)$ instead.

214: 

215: We define the bipartite configuration space $$ \confb^0_m = \Bigl\{ \bigl( \{ \lambda_1, 

216: \dots, \lambda_m \}, \{ \mu_1, \dots, \mu_m \} \bigr ) \ | \ \lambda_i, \mu_j \in \cc 

217: \text{ distinct}, \ \sum \lambda_i + (n-1)\sum \mu_j =0  \Bigr \}.$$

218: 

219: The elements in each of the two sets $\lambdav = \{ \lambda_1, \dots, \lambda_m \}$ and $\muv 

220: = \{ \mu_1, \dots, \mu_m \}$ are not ordered, but the pair $\tau = (\lambdav, \muv)$ is 

221: ordered. The fundamental group of $\confb_m^0$ is the colored braid group on two colors 

222: $Br_{m,m}.$ This has a (noncanonical) subgroup isomorphic to $Br_m,$ which corresponds to 

223: keeping $\muv$ fixed.

224: 

225: For each $m, n > 0$ and $\tau = (\lambdav, \muv) \in \confb^0_m,$ we construct a complex 

226: affine 

227: variety $\ymnt$ as follows. Let $N_{m, n}$ be a nilpotent element in 

228: $\sl(mn)$ with $n$ Jordan blocks of size $m.$ After a change of basis, we can write 

229: $$ N_{m,n} =  \left( \begin {array}{ccccc} 0 & I &  &  & \\

230:  & 0 & I & \ & \\

231:  & \ & \ & \ldots & \ \\

232:  & \ & \  & \  &  I \\

233:  & \ & \ & \ & 0

234: \end {array} \right ), $$

235: where $I$ and $0$ are the $n$-by-$n$ identity and zero matrix, respectively.

236: 

237: The following is a transverse slice to the adjoint orbit of $N_{m, n}$ in $\sl(mn):$

238: 

239: $$ \ss_{m, n} = \left \{ \left( \begin {array}{ccccc} Y_1 & I &  &  & \\

240: Y_2 & 0 & I & \ & \\

241: \ldots & \ & \ & \ldots & \ \\

242:  & \ & \  & \  &  I \\

243: Y_m & \ & \ & \ & 0

244: \end {array} \right ): Y_i \in \gl(n), \ \operatorname{trace}(Y_1) =0  \right \}.$$

245: 

246: Consider also the diagonal matrix $D_{\tau} \in \sl(mn)$ with eigenvalues $\lambda_1, \dots, 

247: \lambda_m$ with multiplicity $1$ (we call these {\it thin} eigenvalues), and $\mu_1, \dots, 

248: \mu_m,$ each with multiplicity $n-1$ (we call these {\it thick} eigenvalues). Let 

249: $\oo_{\tau}$ be the adjoint orbit of $D_{\tau}$ in $\sl(mn),$ and set

250: \begin {equation}

251: \label {ot}

252: \ymnt = \ss_{m,n} \cap \oo_{\tau}.

253: \end {equation}

254: 

255: We will show that, as $\tau$ varies over $\confb^0_m,$ the spaces $\yy_{m, n, \tau}$ form a 

256: symplectic fibration that admits good parallel transport maps. Moreover, for a specific 

257: $\tau$ we build a Lagrangian $L \subset \ymnt$ by iterating a relative vanishing $\cpn$ 

258: construction. The vanishing projective spaces which we introduce in this paper 

259: share many properties with the usual vanishing cycles in symplectic geometry, and may be of 

260: interest on their own. Indeed, Huybrechts and Thomas \cite{HT} pointed out that Lagrangian 

261: projective spaces can play a role in homological mirror symmetry, akin to the role of 

262: Lagrangian spheres.

263: 

264: The local model for the vanishing $\cpn$ construction is the space $Z$ of $n$-by-$n$ 

265: traceless matrices which have an eigenvalue of multiplicity at least $n-1.$ Let us assume 

266: $n > 2.$ (When $n=2,$ our vanishing $\cpn$ is a usual vanishing cycle.) Consider the map 

267: $\chi: Z \to \cc$ which takes a matrix to the corresponding high multiplicity eigenvalue, 

268: and denote by $Z_t$ the fiber over $t \in \cc.$ Note that $Z$ has a singularity at the 

269: origin. Equip (the smooth strata of) $Z$ and the fibers $Z_t$ with the restriction of the 

270: standard K\"ahler form on the space of all $n$-by-$n$ matrices, viewed as $\cc^{n^2}.$ We 

271: can then consider parallel transport in $Z$ with respect to $\chi,$ as long as we stay away 

272: from $Z_0.$ Taking a linear path from $t \in \cc^*$ to the origin, we let $L_t$ be the set 

273: of points in $Z_t$ which are taken to $0 \in Z_0$ in the limit of parallel transport along 

274: that path. It is not hard to describe $L_t$ explicitly.

275: 

276: \begin {lemma}

277: \label {rex}

278: Consider the diagonal matrix $E_t=diag(t,t,\dots,t, (1-n)t) \in Z_t.$ Then:

279: $$ L_t = \{ UE_tU^{-1} \ | \ U \in U(n) \}. $$

280: \end {lemma}

281: 

282: Thus $L_t$ is diffeomorphic to $U(n)/(U(n-1) \times U(1)) \cong \cpn.$ It is also easy to 

283: check that it is a Lagrangian subspace of $Z_t.$ This is the basic model for a 

284: \textit {vanishing projective space}. 

285: 

286: In Section~\ref{sec:vvo} we establish a more general version of Lemma~\ref{rex}, in which 

287: the smooth stratum of $Z$ is equipped with any K\"ahler form satisfying certain real 

288: analyticity and proportionality conditions. We define $L_t$ just as above, and show that it 

289: is a Lagrangian $\cpn$ in $Z_t.$ This is done by observing that $Z$ is the GIT quotient of 

290: $\cc^{2n}$ by the linear $\cc^*$-action with weights $1$ and $-1,$ each with multiplicity 

291: $n.$ We can then lift the spaces $L_t$ to the affine space $\cc^{2n},$ where we argue that 

292: they are vanishing cycles in a certain singular metric. Thus, the vanishing projective 

293: spaces are quotients of vanishing cycles by a circle action.

294: 

295: The spaces $Z_t$ are the same as the fibers $\yy_{1,n, \tau}$ from (\ref{ot}), for $m=1.$ 

296: Therefore, for $m=1$ Lemma~\ref{rex} gives a Lagrangian $\cpn$ in each $\yy_{1,n, \tau}.$ 

297: We then proceed to construct a Lagrangian $L\subset \ymnt$ inductively in $m,$ by using a 

298: relative version of the vanishing $\cpn$'s. The resulting $L$ is diffeomorphic to the 

299: product of $m$ copies of $\cpn.$ Given an element $b \in Br_m \subset Br_{m,m}$ whose 

300: closure is a link $\lk,$ we use parallel transport along the corresponding loop $\beta$ in 

301: $ \confb^0_m$ to construct a second Lagrangian $L'=h^{\resc}_{\beta}(L)$ in the same 

302: $\ymnt.$ Let $w$ be the writhe of the braid $b.$ Our main result is:

303: 

304: \begin {theorem}

305: \label {the}

306: Up to isomorphism of graded abelian groups, the shifted Floer cohomology groups

307: $$ \krss(\lk) = HF^{*+(n-1)(m+w)}(L, L')$$

308: depend only on the oriented link $\lk.$ 

309: \end {theorem}

310: 

311: The proof of the theorem involves checking invariance under the Markov moves $I$ and $II$ 

312: which relate braids with the same closure. 

313: 

314: We managed to compute the groups $\krs$ in a few examples. For the unknot we have 

315: $\krss(\text{unknot}) = H^{*+n-1}(\cpn),$ while for the unlink of $p$ components we get 

316: the tensor product of $p$ copies of the same group. The first nontrivial computation is for the 

317: trefoil, for which we have the following result, consistent with the formula in 

318: \cite[Proposition 6.6]{DGR}.

319: 

320: \begin {proposition}

321: \label {3f}

322: When $\lk$ is the left-handed trefoil, the groups $\krs$ are given by:

323: $$ \krss(\lk) = H^{*-n+1}(\cpn) \oplus H^{*-n-1}(UT\cpn),$$

324: where $UT\cpn$ is the unit tangent bundle to $\cpn.$

325: \end {proposition}

326: 

327: One advantage that our theory has over that of Khovanov and Rozansky is that it 

328: produces abelian groups rather than vector spaces over $\qq$. For example, there is a 

329: $\zz/n\zz$ torsion group appearing in the computation of $\krs$ of the trefoil, and 

330: that group is invisible in $\kr_{\nn}.$

331: 

332: Nevertheless, there is also an obvious shortcoming of our theory, the fact that it does 

333: not come with a bigrading. In the case $n=2,$ a bigrading for the Seidel-Smith cochain 

334: complex was constructed by the author in \cite{M}, but it is not yet clear whether it 

335: descends to cohomology. The bigrading was built using an open holomorphic embedding of the 

336: manifold $\yy_{m,2,\tau}$ into a Hilbert scheme. It would be interesting to study whether 

337: similar embeddings exist for $n > 2.$

338: 

339: It is worth noting here that there are several alternate descriptions of the spaces 

340: $\ymnt,$ and these could lead to further insights into our construction. (This 

341: observation was made by Seidel and Smith in their introduction to \cite{SS}, for $n=2.$) 

342: First, the spaces $\ymnt$ can be viewed as quiver varieties of type $A_{2m-1},$ cf. 

343: Nakajima's Conjecture 8.6 in \cite{N1}, proved by Maffei in \cite{Maf}. Via an ADHM 

344: transform, these can also be viewed as moduli spaces of rank $n$ instantons on an ALE 

345: space (cf. \cite{KN}), or as moduli spaces of solutions to Nahm's equations (cf. 

346: \cite{AB}, \cite{Kr2}, \cite{N1}). To some extent these alternate descriptions are 

347: conjectural, because most of the works cited in this paragraph only deal with nilpotent 

348: orbits. However, we expect the respective results to generalize to our situation.

349: 

350: This paper is organized as follows. In Section~\ref{sec:gaos} we study the general 

351: properties of intersections between transverse slices and adjoint orbits. In 

352: Section~\ref{sec:av} we apply these properties to study our objects of interest, the spaces 

353: $\ymnt,$ as well as their degenerations. In Section~\ref{sec:vvo} we present the 

354: construction of vanishing projective spaces. In Section~\ref{sec:fiba2} we study in detail 

355: a geometric situation that is key to the proof of Markov $II$ invariance. In 

356: Section~\ref{sec:floer} we review the definition of Floer cohomology and discuss some 

357: relevant properties. Section~\ref{sec:maind} contains the construction of the Lagrangians 

358: $L, L' \subset \ymnt.$ We prove Theorem~\ref{the} in Section~\ref{sec:markov} by showing 

359: invariance under the two Markov moves, and then do the trefoil computation in 

360: Section~\ref{sec:trefoil}. In the last section we speculate on the existence of other 

361: classes of link invariants: some that correspond to various other Lie algebras and 

362: representations, and some that we expect to arise by considering a particular involution on 

363: the spaces $\ymnt.$ The latter invariants should form a series parametrized by integers $n 

364: \geq 2,$ such that the $n=2$ case gives the Heegaard Floer homology of the double branched 

365: cover.

366: 

367: We should note that many of the arguments in this paper follow the ones given by Seidel and 

368: Smith in \cite{SS} closely, and we sometimes refer to their article for full details. 

369: However, there are also several aspects that are fundamentally new in our work as compared 

370: with the $n=2$ case, and these are the ones which we choose to emphasize in our exposition. 

371: The first is the distinction between thin and thick eignevalues, stemming from the fact 

372: that $V \not \cong V^*$ for the standard representation of $\sl(n)$ when $n > 2.$ A 

373: consequence of the appearance of thick eigenvalues is that the orbit $\oo_{\tau}$ from 

374: (\ref{ot}) is no longer maximal, and we need a careful study of the geometry of 

375: intersections between slices and non-maximal orbits. One new phenomenon is that the total 

376: space consisting of all $\ymnt$'s and their degenerations is singular for $n > 2.$ It is in 

377: this space where the vanishing $\cpn$ construction is made, and as far as we know this 

378: construction is new. We deal with vanishing projective spaces by viewing them as the 

379: quotients of ordinary vanishing cycles by an $S^1$-action. The corresponding vanishing 

380: cycles appear in a singular metric though, and in order to justify their existence we have 

381: to resort to a real analyticity condition on the metric. Also, the parallel transport 

382: estimates for vanishing $\cpn$'s in Section~\ref{sec:fiba2}, although similar in spirit to 

383: those in \cite{SS}, are somewhat more involved here. In particular, some care is needed in 

384: making sure that the K\"ahler metrics on our local models in singular spaces are equivalent 

385: to standard metrics. Finally, the discussion of orientations in Floer cohomology depends on 

386: the parity of $n,$ because for $n$ odd the complex projective space $\cpn$ is not spin.

387: 

388: \medskip \noindent \textbf{Acknowledgements.} I am grateful to Mikhail Khovanov, Peter 

389: Kronheimer, Duong Phong and Jacob Rasmussen for several valuable discussions, to Edward 

390: Bierstone, Krzysztof Kurdyka and Alec Mihailovs for helpful email correspondence, and to Paul 

391: Seidel and Ivan Smith for their interest in this work.

392: 

393: 

394: \section {The geometry of adjoint orbits and slices}

395: \label {sec:gaos}

396: 

397: This section parallels Section 2 in \cite{SS}. We collect some facts about partial 

398: Grothendieck resolutions and their intersections with transverse slices. Our main 

399: reference is the work of Borho and MacPherson \cite{BM}, where partial Grothendieck 

400: resolutions are studied in detail. We also drew inspiration from \cite{Kr2}, \cite{N1}, 

401: \cite{Ro}, \cite{Sl} and \cite{SS}. Some of the results there were formulated only for 

402: the full Grothendieck resolution, but admit straightforward generalizations to the 

403: partial case.

404: 

405: The discussion below can be made more general, but for the sake of concreteness we 

406: restrict our attention to the case of the group $G= SL(N, \cc)$ and its Lie algebra 

407: $\ggg = \sl(N, \cc).$ We also fix the standard basis for $\cc^N.$ We denote by $\hh 

408: \cong \cc^{N-1}$ the corresponding Cartan subalgebra of traceless diagonal matrices, and 

409: by $W = S_{N}$ its Weyl group.

410: 

411: \subsection {Partial Grothendieck resolutions}

412: Let $\pi=(\pi_1, \dots, \pi_s)$ be a partition of $N = \pi_1 + \dots + \pi_s,$ with $\pi_1 

413: \geq \pi_2 \geq \dots \geq \pi_s > 0.$ We denote by $m_{\pi}(k)\geq 0$ the number of 

414: times $k$ appears among the $\pi_j,$ for $k=1, \dots, N.$ Thus $\sum m_{\pi}(k) = s$ 

415: and $\sum k m_{\pi}(k) = N.$ Sometimes the partition $\pi$ is also written as $\pi = 

416: (1^{m_{\pi}(1)} 2^{m_{\pi}(2)} \dots N^{m_{\pi}(N)}).$ There is a dual partition $\pi^* 

417: = (\pi^*_1, \pi^*_2, \dots)$ with $\pi^*_j = m_{\pi}(j) + m_{\pi}(j+1) + \dots + 

418: m_{\pi}(N).$

419: 

420: Associated to $\pi$ is a partial flag variety 

421: $$ \ff^{\pi} = \{0=F_0 \subset F_1 \subset \dots \subset F_s=\cc^N \ | \ \dim (F_j) - 

422: \dim (F_{j-1}) = \pi_j \}.$$

423: 

424: We denote by $F^{st} \in \ff^{\pi}$ the standard flag $0 \subset \cc^{\pi_1} \subset 

425: \cc^{\pi_1 + \pi_2} \subset \dots \subset \cc^N.$ Note that for every flag $F=(F_0, 

426: \dots, F_s) \in \ff^{\pi}$ there is a corresponding parabolic subalgebra $\ppp(F) 

427: \subset \ggg$ consisting of those matrices $x\in \ggg$ preserving the flag, i.e. such 

428: that $x(F_j) \subset F_j$ for all $j.$ Thus we can identify $\ff^{\pi}$ with the space 

429: of parabolic Lie subalgebras of $\ggg$ conjugate to $\ppp(F^{st}).$

430: 

431: The adjoint quotient map $\chi: \ggg \to \ggg/G = \hh/W$ takes a matrix to the set of its 

432: generalized eigenvalues. Note that $\hh/W$ can be identified with $\cc^{N-1}$ via 

433: symmetric polynomials. Set

434: $$\ggp = \{(x, F) \ | \ F \in \ff^{\pi}, x \in \ppp(F) \}.$$

435: 

436: Consider also the subgroup $W_{\pi} \subset W$ given by $S_{\pi_1} \times \dots 

437: \times S_{\pi_s} \subset S_N.$ This is the Weyl group corresponding to the Levi subalgebra of 

438: $\ppp(F^{st}).$ The {\it partial simultaneous resolution} of $\chi$ associated to 

439: the partition $\pi$ consists of the commutative diagram: 

440: \begin {equation}

441: \label {psr}

442: \begin {CD}  \ggp @>>> \ggg 

443: \\ @V{\hat \chi^{\pi}}VV @V{\chi}VV \\ \hh/ W_{\pi} @>>> \hh/W 

444: \end {CD}

445: \end {equation}

446: Observe that for every $(x, F) \in \ggp$ there is an induced action of $x$ on each

447: quotient $F_j/F_{j-1},$ for $j=1, \dots, s.$ This gives elements $x_j^F \in

448: \gl(F_j/F_{j-1}).$  The map 

449: $\hat \chi^{\pi}$ is defined to take a pair $(x, F)$ to the sets of generalized 

450: eigenvalues of $x_j^F$ for all $j.$

451: 

452: The map $\tilde \chi = \hat \chi^{\pi}$ for $\pi=(1^N)$ was the one 

453: originally studied by Grothendieck. In that case $W_{\pi} = 1,\ \tilde \ggg = \ggp$ is a 

454: smooth manifold and $\tilde \chi$ is a {\it simultaneous resolution} in the following 

455: sense: $\tilde \chi$ is a submersion with the property that for each $\tilde t \in \hh, 

456: \tilde \chi^{-1}(\tilde t)$ is a resolution of singularities for $\chi^{-1}(t),$ where 

457: $t$ is the image of $\tilde t$ in $\hh/W.$ Diagramatically,

458: \begin {equation}

459: \label {grothen}

460: \begin {CD}  \tilde \ggg @>>> \ggg

461: \\ @V{\tilde \chi}VV @V{\chi}VV \\ \hh @>>> \hh/W.

462: \end {CD}

463: \end {equation}

464: 

465: For general $\pi,$ the variety $\ggp$ is not smooth. $\hat \chi^{\pi}$ is called a 

466: partial resolution 

467: because the map $\tilde \ggg \to \ggg$ factors through $\ggp.$ However, as explained 

468: below, if we restrict $\ggp \to \ggg$ to a certain subset of $\ggp$ we do get an honest 

469: simultaneous resolution.

470: 

471: 

472: \subsection {Restricted partial Grothendieck resolutions} \label{sec:rpr} Set 

473: $$\gp = \{(x, F) \in \ggp \ | \ x_j^F \in Z(\gl(F_j/F_{j-1})) \text{ for all } j \}. $$

474: 

475: Consider the subspace $\hh^{\pi} \subset \hh/W_{\pi}$ made of traceless diagonal 

476: matrices of the type $\alpha = diag(\alpha_1, \dots, \alpha_1, \alpha_2, \dots, 

477: \alpha_2, \dots, \alpha_s),$ where each $\alpha_j$ appears exactly $\pi_j$ times. 

478: For every $(x, F) \in 

479: \gp$ we have $x_j^F = \alpha_j \cdot I$ for some $\alpha_j \in \cc.$ Therefore, there 

480: is a induced map $\tilde \chi^{\pi}: \gp \to \hh^{\pi}.$ This fits into a commutative diagram 

481: \begin {equation}

482: \label {rpr}

483: \begin {CD}

484: \gp @>>> \gpi \\ @V{\tilde \chi^{\pi}}VV @V{\chi^{\pi}}VV \\ \hh^{\pi} @>>> 

485: \hh^{\pi}/W^{\pi} \end {CD}

486: \end {equation}

487: 

488: We call this diagram the {\it restricted partial simultaneous resolution} associated to 

489: $\pi.$ The subset $\gpi \subset \ggg$ is defined to be the image of $\gp$ under the map 

490: $\ggp \to \ggg$ from (\ref{psr}), and the {\it partial Weyl group} $W^{\pi}$ is defined as 

491: $W^{\pi} = S_{m_{\pi}(1)} \times \dots \times S_{m_{\pi}(N)} \subset S_s.$ The vertical map 

492: $\chi^{\pi}$ is naturally induced from the adjoint quotient map $\chi.$ Note that 

493: $\hh^{\pi} \cong \cc^{s-1},$ while $\hh^{\pi}/W^{\pi}$ is naturally the quotient by $\cc$ 

494: of a product of symmetric spaces $\sym^{m_{\pi}(1)}(\cc) \times \dots \times 

495: \sym^{m_{\pi}(N)}(\cc).$ (This quotient can also be identified with $\cc^{s-1},$ using 

496: symmetric polynomials.)

497: 

498: \begin {example}

499: \label {ex0}

500: For $\pi = (N-1,1),$ the space $\gpi$ consists of traceless $N$-by-$N$ matrices having

501: an eigenspace of dimension at least $N-1.$ (This is the space $Z$ mentioned in the 

502: introduction.) The map $\chi^{\pi}$ takes a 

503: matrix to the corresponding high multiplicity eigenvalue in $\hh^{\pi}/W^{\pi} \cong \cc.$ 

504: \end {example}

505: 

506: In general, it is easy to see that $\gpi$ is a closed subvariety of $\ggg.$ Also, the 

507: adjoint action $Ad$ of $G$ on $\ggg$ induces actions of $G$ on all the spaces in 

508: (\ref{rpr}), and all the maps there are $G$-equivariant. It follows that $\gpi$ is a union 

509: of adjoint orbits in $\ggg.$

510: 

511: The proof of the following proposition is analogous to that of Lemma 4 in \cite{SS}:

512: 

513: \begin {proposition}

514: \label {aboutrpr}

515: The diagram (\ref{rpr}) is a simultaneous resolution of $\chi^{\pi},$ and the map $\tilde 

516: \chi^{\pi}$ is naturally a differentiable fiber bundle.

517: \end {proposition}

518: 

519: In fact, we can exhibit an explicit trivialization of the fiber bundle $\tilde 

520: \chi^{\pi}.$ Every flag $F \in F^{\pi}$ is of the form $u(F)F^{st},$ where $u(F) \in 

521: U(N)$ is unique up to right translation by an element in $U^{\pi} = U(\pi_1) \times 

522: \dots \times U(\pi_s).$ The map

523: \begin {equation}

524: \label {trivia}

525: (\tilde \chi^{\pi})^{-1}(0) \times \hh^{\pi} \to \gp \ , \ \ (x, F, \alpha) 

526: \to (x+Ad(u(F))\alpha, F) 

527: \end {equation}

528: is a trivialization for $\tilde \chi^{\pi}.$ Note that this trivialization is not natural, 

529: because it depends on our chosen basis for $\cc^N.$

530: 

531: Now consider the open subset $\hh^{\pi, \reg} \subset \hh^{\pi}$ where $\alpha_1, \dots, 

532: \alpha_s$ are pairwise distinct. Its image $\conf^0_{\pi} = \hh^{\pi, \reg}/W^{\pi}$ is 

533: called the {\it $\pi$-colored configuration space}. The superscript $0$ stands for 

534: traceless. If we don't impose the condition $\sum \pi_j \alpha_j =0,$ we get a larger 

535: but homotopy equivalent configuration space $\conf_{\pi}.$

536: 

537: \begin {definition}

538: The {\it $\pi$-colored braid group} is the fundamental group $Br^{\pi} = 

539: \pi_1(\conf^0_{\pi}).$ 

540: \end {definition}

541: 

542: Observe that the $(1^N)$-colored braid group is the usual braid group 

543: on $N$ strands $Br_N,$ while at the other extreme the $(N)$-colored braid group is the 

544: pure braid group $PBr_N.$ (In the literature $PBr_N$ is sometimes called the colored braid 

545: group.) In general, $Br^{\pi}$ is intermediate between these two cases, in the sense 

546: that the natural map $PBr_s \to Br_s$ factors through $Br^{\pi},$ in correspondence to a 

547: factoring of covering spaces. 

548: 

549: Let $\ggg^{\pi, \reg}$ be the preimage of $\hh^{\pi, \reg}/W^{\pi}$ under the map 

550: $\chi^{\pi}.$ For $x \in \ggg^{\pi, \reg},$ the choice of a partial flag $F$ with $(x, 

551: F) \in \gp$ is unique once we fix an ordering of the $\alpha_j$'s with the same $\pi_j.$ 

552: A quick consequence of this is the following:

553: 

554: \begin {proposition}

555: \label {regular}

556: The points $x \in \ggg^{\pi, \reg}$ are regular for the maps $\chi^{\pi},$ 

557: and the restriction of $\chi^{\pi}$ to $\ggg^{\pi, \reg}$ is naturally a differentiable 

558: fiber bundle over $\conf^0_{\pi}.$

559: \end {proposition}

560: 

561: Since $\gg^{\pi}$ is typically singular, we should clarify some terminology. In 

562: Proposition~\ref{regular} and everywhere below, a point $x \in \gg^{\pi}$ is called 

563: regular for 

564: the map $\chi^{\pi}$ if $\gg^{\pi}$ is smooth at $x,$ and the differential $(d\chi^{\pi})_x$ 

565: is onto; $x$ is called critical otherwise. (The same goes for points on any singular 

566: variety mapped to a smooth one.)

567: 

568: 

569: \subsection {Reorderings of a partition}

570: By a \textit{reordering} $\pi_{\tau}$ of the partition $\pi = (\pi_1, \dots, \pi_s)$ we 

571: simply mean the ordered $s$-tuple $(\pi_{\tau(1)}, \dots, \pi_{\tau(s)})$ for some 

572: permutation $\tau$ of $\{1,2,\dots, s\}.$ Given such a reordering, we can define partial 

573: flags of type $\pi_{\tau}$ so that the differences in consecutive dimensions are 

574: $\pi_{\tau(1)}, \dots, \pi_{\tau(s)}$ in that order. We can then define $\tilde 

575: \gg^{\pi_{\tau}}, \ggg^{\pi_{\tau}}, \hh^{\pi_{\tau}}$ and $W^{\pi_{\tau}}$ as in the 

576: previous subsection, an adjoint quotient map $\chi^{\pi_{\tau}}$ and an analogue of 

577: the diagram (\ref{rpr}).  Note that there are isomorphisms $\hh^{\pi}  \cong 

578: \hh^{\pi_{\tau}}$ (from reordering the diagonal elements) and $W^{\pi} \cong 

579: W^{\pi_{\tau}}.$ We denote by $\iota_{\tau}$ the induced isomorphism 

580: $\hh^{\pi}/W^{\pi} \to \hh^{\pi_{\tau}}/W^{\pi_{\tau}}.$ The following result will 

581: prove useful: 

582: 

583: \begin {lemma}

584: \label {reorder}

585: The subvariety $\ggg^{\pi_{\tau}} \subset \gg$ is independent of $\tau,$ and in 

586: particular identical to $\gg^{\pi}.$ The map $\chi^{\pi_{\tau}} : \ggg^{\pi_{\tau}} 

587: \to \hh^{\pi_{\tau}}/W^{\pi_{\tau}}$ equals the composite $\iota_{\tau} \circ 

588: \chi^{\pi}.$   

589: \end {lemma}

590: 

591: \noindent \textbf{Proof.} We first deal with the case when $\pi = (a, b)$ with 

592: $a \geq b, \ a+b=N$ and $\tau$ is the transposition, so that $\pi_{\tau} = (b,a).$ An 

593: element $x$ is in $\gg^{\pi}$ if it fixes some subspace $V= F_1  \subset 

594: \cc^{N}$ of dimension $a,$ and $x$ acts on $V$ diagonally by $\alpha \cdot \Id$ and on 

595: $\cc^N/V$ diagonally by $\beta \cdot I,$ for some $\alpha, \beta \in \cc.$ We seek to 

596: show that $x$ is in $\gg^{\pi_{\tau}}$ by constructing a subspace $W \subset \cc^N$ of 

597: dimension $b$ such that $x$ acts by $\beta \cdot \Id$ on $W$ and by $\alpha \cdot I$ on 

598: $\cc^N/W.$ If $\alpha \neq \beta,$ we simply take $W$ to be the $\beta$-eigenspace of $x.$ If 

599: $\alpha = \beta,$ then $(x - \alpha\Id)$ acts trivially on both $V$ and $\cc^N/V$ and 

600: therefore is determined by an induced map $\cc^N/V \to V.$ In this case we can 

601: choose $W$ to be any $b$-dimensional subspace of $V$ containing the image of $(x - 

602: \alpha \Id).$

603: 

604: Conversely, if we know that $X \in \gg^{\pi_{\tau}},$ then there exists a 

605: $b$-dimensional subspace $W \subset \cc^N$ as in the previous paragraph, and we want to 

606: construct $V.$ 

607: If $\alpha \neq \beta,$ we take $V$ to be the $\alpha$-eigenspace of $x.$ If $\alpha = 

608: \beta,$ then $x$ is determined by a map $\cc^N/W \to W$ induced by $(x-\alpha \Id),$ and 

609: we can take $V$ to be any $a$-dimensional subspace of the kernel of $(x - \alpha \Id),$ 

610: containing $W.$ 

611: 

612: The case of a general reordering follows from this by induction, using the fact that 

613: every permutation $\pi$ is a product of transpositions. $\hfill \fin$

614: 

615: 

616: \subsection {Slices at semisimple elements}

617: \label {sec:semis}

618: 

619: Let $x$ be an element of $\ggg.$ We denote its stabilizer subgroup by $G_x$ and the 

620: corresponding Lie subalgebra by $\ggg_x = \{y\in \ggg \ | [x,y]=0\}$. A local {\it 

621: transverse slice} to the adjoint orbit $\oo(x) = Gx$ of $x$ is a local complex submanifold 

622: $\ss \subset \ggg$ such that $x\in \ss$ and the tangent spaces at $x$ of $\oo(x)$ and 

623: $\ss$ are complementary. Note that $$ T_x \oo(x) = ad(\ggg)x = \{[y,x] \ | \ y \in 

624: \ggg\}.$$

625: 

626: Let $x$ be semisimple. The dimensions of the eigenspaces of $x$ form a partition 

627: $\sigma=(\sigma_1, \dots, \sigma_l)$ of $N.$ The Lie algebra $\ggg_x$ splits into a direct 

628: sum of its center $Z(\ggg_x)$ and a reduced part $\ggg_x^{\red}=[\ggg_x, \ggg_x],$ which is 

629: a product of factors $\ggg_x^{\red} [ i ]$ isomorphic to $\sl(\sigma_i, \cc).$ There is 

630: an 

631: adjoint quotient map for $\ggg_x^{\red},$ denoted by 

632: \begin {equation}

633: \label {chix}

634: \chi_x^{\red}: \ggg^{\red}_x \to \ggg^{\red}_x/G_x = \hhh^{\red}_x/W^{\red}_x.

635: \end {equation}

636: Here $\hhh^{\red}_x \subset \hhh$ corresponds to block diagonal matrices such that each 

637: block has trace zero, and $W^{\red}_x= W_{\sigma}.$

638: 

639: Using transverse slices, it can be shown that a fibered version of $\chi_x^{\red}$ gives a 

640: a local model for the adjoint quotient map $\chi$ near the orbit $\oo(x).$ We sketch this 

641: here (following \cite[Section 2(B)]{SS}), and then explain how to get a local model 

642: for the restricted map $\chi^{\pi}$ near $\oo(x)$ when that orbit lies in $\gg^{\pi}.$

643: 

644: A canonical transverse slice for $x$ is $\ss^{ss} = x + \ggg_x.$ 

645: This has the property that, given any other slice $\ss,$ there is a canonical local 

646: isomorphism between $\ss$ and $\ss^{ss}$ obtained as follows. We take a 

647: neighborhood $V$ of zero inside $ad(\ggg)x.$ Then $exp(V) \subset G$ is tranverse to $G_x$ 

648: and we can use the composition

649: \begin {equation}

650: \label {compose}

651: \begin {CD}

652: \ss \hookrightarrow \ggg @>\operatorname{local \ \cong}>> exp(V) \times \ss^{ss} 

653: @>\operatorname{projection}>> \ss^{ss}.

654: \end {CD}

655: \end {equation}

656: 

657: The slice $\ss^{ss}$ is $G_x$-invariant, and therefore the map

658: \begin {equation}

659: \label {fibered}

660:  G \times_{G_x} \ss^{ss} \to \ggg, \ (g, y) \to Ad(g)y

661: \end {equation}

662: is a local isomorphism between neighborhoods of $G/G_x \times \{x\}$ and $\oo(x).$

663: Using the fact that $G_x$ acts trivially on the first summand in the splitting 

664: $\ggg_x = Z(\ggg_x) \oplus \ggg_x^{\red},$ we obtain a commutative diagram

665: \begin {equation}

666: \label {cairo}

667: \begin {CD}

668: Z(\ggg_x) \times (G \times_{G_x} \ggg_x^{\red}) @>>> \ggg \\

669: @VVV @V{\chi}VV \\

670: Z(\ggg_x) \times \hhh^{\red}_x/W^{\red}_x @>>> \hhh/W.

671: \end {CD}

672: \end {equation}

673: 

674: The top $\rightarrow$ is (\ref{compose}), and the 

675: bottom one is the induced local isomorphism on quotients, in both cases identifying 

676: $\ss^{ss}$ with $\ggg_x  = Z(\ggg_x) \times \ggg_x^{\red}$ by translation. The vertical map 

677: on the left is identity times the quotient map, which means that $\chi$ looks locally 

678: like a fibered version of (\ref{chix}).

679: 

680: Let us go back to the restricted partial resolution (\ref{rpr}) and assume that $x \in 

681: \ggg^{\pi}.$ Since $x$ is semisimple, it must be conjugate to a matrix $diag(\alpha_1, 

682: \dots, \alpha_1, \alpha_2, \dots, \alpha_2, \dots, \alpha_s) \in \hh^{\pi}.$ (Note that 

683: this diagonal matrix is not typically unique, because of the $W^{\pi}$ symmetry.) The 

684: set of indices $\{1,2,\dots, s\}$ has a decomposition into a disjoint union $A_1 

685: \amalg \dots \amalg A_l$ such that $\alpha_j = \alpha_k$ if and only if $j$ and $k$ are 

686: in the same $A_i.$ The sizes $\sigma_i$ of the sets $A_i$ form the partition $(\sigma_1, 

687: \dots, \sigma_l)$ of $N$ that was described for any semisimple element. Since $x \in 

688: \ggg^{\pi},$ we can be more specific and say that the partition $\pi$ of $N$ ``breaks'' in 

689: the following sense:

690: 

691: \begin {definition}

692: Let $\pi=(\pi_1, \dots, \pi_s)$ and $\sigma = (\sigma_1, \dots, \sigma_l)$ be two 

693: partitions of the same positive integer $N.$ A breaking $b = (b[1], b[2], \dots, b[l])$ 

694: of $\pi$ according to $\sigma$ consists of partitions $b[i]$ of each $\sigma_i$ such that 

695: their concatenation is a reordering of the partition $\pi.$ The set of breakings of $\pi$ 

696: according to $\sigma$ is denoted $\bb_{\pi \sigma}.$

697: \end {definition} 

698: 

699: In our situation, $\pi$ breaks into $b = (b[1], b[2], \dots, b[l]),$ where $b[i]$ is 

700: composed of all the $\pi_j$'s with $j \in A_i.$ By varying the diagonal matrix in 

701: $\hh^{\pi}$ conjugate to $x,$ all the breakings in $\bb_{\pi \sigma}$ can occur.

702: 

703: \begin {example}

704: \label {ex1}

705: If $\pi=(2,1,1)$ and $\sigma=(2,2),$ then there are two possible breakings of $\pi$ 

706: according to $\sigma: \ \bigl( b[1]=(2),\ b[2]=(1,1) \bigr)$ and $\bigl( b[1]=(1,1),\  

707: b[2]=(2) \bigr).$ More informally, we say that the latter breaking, for example, consists 

708: of the first $2$ in $\sigma=(2,2)$ being broken as $1+1$ and the second $2$ being broken 

709: trivially.

710: \end {example}

711: 

712: \begin {example}

713: \label {ex2}

714: For $m,n \geq 1,$ there is a unique breaking $b$ of $\pi =

715: (1^m(n-1)^m)$ according to $\sigma=(1^{m-1}(n-1)^{m-1}n),$ namely $n$ breaks as $(n-1)+1$ 

716: and all the $1$'s and $(n-1)$'s break trivially.

717: \end {example}

718: 

719: \begin {example}

720: \label {ex3}

721: For $m, n \geq 1,$ there are exactly $m$ breakings of $\pi=(1^{m+n-1}(n-1)^{m-1})$ 

722: according to $\sigma = (1^{m-1}(n-1)^{m-1}n):$ either the $n$ summand breaks as $1+ \dots 

723: +1$ and the rest break trivially, or the $n$ summand breaks as $(n-1)+1,$ one of the 

724: $(n-1)$'s breaks as a sum of $1$'s and the rest break trivially.

725: \end {example}

726: 

727: \begin {example}

728: \label {ex4}

729: For $m \geq 2$ and $n > 3,$ there is a unique breaking of $\pi =

730: (1^m(n-1)^m)$ according to $\sigma=(1^{m-2}2(n-1)^m),$ namely $2$ breaks as $1+1$ and the 

731: rest trivially.

732: \end {example}

733: 

734: \begin {example}

735: \label {ex5}

736: For $m \geq 2$ and $n \geq 1,$ there is again a unique breaking of $\pi =

737: (1^m(n-1)^m)$ according to $\sigma = (1^{m-2}(n-1)^{m-1}(n+1)),$ namely $(n+1) = 

738: (n-1) + 1 +1$ and the rest trivial. 

739: \end {example}

740: 

741: 

742: With respect to transverse slices at $x \in \gg^{\pi}$, the first observation which needs 

743: to be made is that the local isomorphism (\ref{compose}) moves points only inside their 

744: adjoint orbits, and therefore induces a canonical local isomorphism

745: \begin {equation}

746: \label {caniso}

747:  \ss \cap \ggg^{\pi} \cong \ss^{ss} \cap \ggg^{\pi}.

748: \end {equation}

749: 

750: Second, the same reasoning applies to the map (\ref{fibered}), giving a local isomorphism

751: \begin {equation}

752: \label {fibered2}

753: G \times_{G_x} (\ss^{ss} \cap \ggg^{\pi}) \to \ggg^{\pi}.

754: \end {equation}

755: 

756: The left hand side of (\ref{fibered}) can be made more explicit as follows. Recall that 

757: $\ss^{ss} = x + \ggg_x$ and $\ggg_x$ decomposes as $Z(\ggg_x) \times \prod_i 

758: \ggg_x^{\red}[i],$ with $\gg_x^{\red}[i] \cong \sl(\sigma_i, \cc).$ We have a restricted 

759: partial simultaneous resolution associated to each factor $\ggg_x^{\red}[i]$ and the 

760: corresponding partition $b[i]$ in a breaking of $\pi$ according to $\sigma.$ The role of 

761: $\chi^{\pi}$ in the left half of the diagram (\ref{rpr}) is played by maps 

762: $$\chi^{b[i]} : \gg_x^{\red}[i]^{b[i]} \to  \hh_x^{\red}[i]^{b[i]}/W^{b[i]}$$

763: for each $i.$

764: 

765: \begin {lemma}

766: \label {hello}

767: Translation $y \to (y-x)$ induces an identification of 

768: $\ss^{ss} \cap \ggg^{\pi}$ with

769: $$Z(\gg_x) \times \bigcup_{b \in \bb_{\pi \sigma}} \prod_i \gg_x^{\red}[i]^{b[i]}.$$  

770: \end {lemma}

771: 

772: \noindent \textbf{Proof.} First observe that an element in $\gg_x$ preserves each eigenspace 

773: $E_i$ of $x.$ The direct sum of all $E_i$ is $\cc^N,$ the sizes of $E_i$ form the partition 

774: $\sigma,$ and $ \gg_x^{\red}[i]$ is the Lie alegbra of traceless endomorphisms of $E_i.$

775: 

776: Now choose some $y$ such that $y -x $ is in $Z(\gg_x) \times \prod_i 

777: \gg_x^{\red}[i]^{b[i]}$ for some $b \in  \bb_{\pi \sigma}.$ We have that $y$ is in $\gg_x,$

778: and $y$ acts on each $E_i$ by a central element plus something in $\gg_x^{\red}[i]^{b[i]}$ 

779: This means that we can find partial flags of type $b[i]$ made of subspaces of $E_i$ that are 

780: preserved by $y$ and such that $y$ acts diagonally on consecutive quotients. Putting these flags 

781: together (by taking direct sums) we form a partial flag made of subsets of $\cc^N.$ Its type is 

782: a reordering of the partition $\pi.$ Lemma~\ref{reorder} implies that $y$ must be in 

783: $\gg^{\pi}.$

784: 

785: Conversely, take some $y \in (x + \gg_x) \cap \ggg^{\pi},$ sufficiently close to $x.$ Since 

786: $y$ is in $\gg^{\pi}$ it must preserve a partial flag $F$ of type $\pi$ and act diagonally 

787: on the consecutive quotients. We also know that $y$ preserves each $E_i,$ and therefore the 

788: intersections of $F$ with each $E_i$ are also preserved, with diagonal actions on 

789: consecutive quotients. The resulting partial flags are of type $b[i],$ for some breaking of 

790: $\pi$ according to $\sigma.$ We get that the traceless part of $y|_{E_i}$ is in  

791: $\gg_x^{\red}[i]^{b[i]}$ for all $i. \ \hfill \fin$

792: 

793: \medskip

794: 

795: Using the result of Lemma~\ref{hello}, we get a restricted version of (\ref{cairo}):

796: \begin {equation}

797: \label {cairo2}

798: \begin {CD}

799: Z(\ggg_x) \times \bigl (G \times_{G_x}  \bigcup_{b \in \bb_{\pi \sigma}}(\prod_i  

800: \gg_x^{\red}[i]^{b[i]}) \bigr ) @>>> \ggg^{\pi} \\

801: @VVV @V{\chi^{\pi}}VV \\

802: Z(\ggg_x) \times  \bigcup_{b \in \bb_{\pi \sigma}} (\prod_i  \hh_x^{\red}[i]^{b[i]}/W^{b[i]}) 

803: @>>> \hhh^{\pi}/W^{\pi}.

804: \end {CD}

805: \end {equation}

806: 

807: The map going vertically on the left is identity times a quotient map (coming from the 

808: product of all $\chi^{b[i]}$'s), and the horizontal arrows are local isomorphisms. It 

809: follows that locally near $\oo(x) \subset \gg^{\pi},$ the map $\chi^{\pi}$ looks like a 

810: fibered version of the union of products of the $\chi^{b[i]}$'s, taken over all $b \in \bb_{\pi 

811: \sigma}.$

812: 

813: 

814: \subsection {Invariant slices at nilpotent elements}

815: 

816: Now let $x \in \ggg$ be nilpotent. One way to construct transverse slices at $x$ with nice 

817: global properties is using the Jacobson-Morozov lemma, which claims the existence of a 

818: triple $(n^+=x, n^-, h)$ of elements of $\ggg$ such that

819: \begin {equation}

820: \label {nueva}

821: [h, n^+]=2n^+, \ [h, n^-]=-2n^-, \ [n^+, n^-]=h.

822: \end {equation}

823: 

824: This gives a representation of $\sl(2,\cc)$ into $\ggg,$ which in turn 

825: produces a splitting $\ggg = \im ad(n^+) \oplus \ker ad(n^-).$ It follows that:

826: 

827: \begin {lemma}

828: The affine subspace $\sjm = n^+ + \ker ad(n^-)$ is a local transverse slice at $x.$  

829: \end {lemma}

830: 

831: Furthermore, we have a linear $\cc^*$-action on $\ggg$

832: \begin {equation}

833: \label {act1}

834:  \lambda_r: \cc^* \to \operatorname{Aut}(\ggg), \ \ r\in \cc^* \ : \  y \to 

835: r^2Ad(r^{h})y.

836: \end {equation}

837: 

838: This restricts to an action on the Jacobson-Morozov slice $\sjm,$ and has the property 

839: that it contracts $\sjm$ to $0$ as $r\to 0.$ Using this action we can show that the 

840: behavior of $\sjm$ globally is determined by the behavior near $x.$ In particular, 

841: $\sjm$ intersects all adjoint orbits transversely.

842: 

843: In \cite{SS}, Seidel and Smith use a more general notion:

844: 

845: \begin {definition}

846: Given a triple $(n^+, n^-, h)$ satisying (\ref{nueva}), a $\lambda$-invariant slice is an 

847: affine subspace $\ss \subset \ggg$ invariant under $\ggg$ and such that it represents a local 

848: transverse slice at $n^+=x.$ 

849: \end {definition}

850: 

851: In fact, $\lambda$-invariant slices share many properties with Jacobson-Morozov slices:

852: 

853: \begin {proposition}[Lemma 11(i) and Lemma 14 in \cite{SS}] 

854: \label {invslice}

855: (i) A $\lambda$-invariant slice $\ss$ has transverse intersections with all adjoint 

856: orbits. (ii) For $\lambda$-invariant $\ss$ and Jacobson-Morozov slice $\sjm$ (possibly 

857: coming from a  

858: different choice of a triple with $n^+=x$), there exists a (noncanonical) 

859: $\cc^*$-equivariant isomorphism $\ss \cong \sjm,$ which moves points only in their 

860: adjoint orbits.

861: \end {proposition}

862: 

863: Now assume that $x \in \gpi.$ Because the diagram (\ref{rpr}) is $G$-equivariant, it 

864: behaves well under intersecting with a $\lambda$-invariant transverse slice $\ss.$ 

865: More precisely, we have:

866: 

867: \begin {proposition}

868: \label {big}

869: (i) The isomorphism in Proposition~\ref{invslice} (ii) maps $\ss \cap \gpi$ into $\sjm 

870: \cap \gpi.$ (ii) A point of $\ss \cap \gpi$ is a critical point of $\chi^{\pi}$ iff it 

871: is a critical point of $\chi^{\pi}|_{\ss}.$ (iii) Let $\tilde \ss$ be the 

872: preimage of $\ss$ in $\tilde \ggg.$ Then 

873: \begin {equation}

874: \label {slice}

875: \begin {CD}

876: \tilde \ss \cap \gp @>>> \ss \cap \gpi \\ @V{\tilde \chi^{\pi}}VV @V{\chi^{\pi}}VV \\ 

877: \hh^{\pi} 

878: @>>> \hh^{\pi}/W^{\pi} \end {CD}

879: \end {equation}

880: is a simultaneous resolution. (iv) $\tilde \chi^{\pi}: \tilde \ss \cap \gp \to 

881: \hh^{\pi}$ is naturally a differentiable fiber bundle. 

882: \end {proposition}

883: 

884: Point (i) is a direct implication of Proposition~\ref{invslice} and the fact that 

885: $\gpi$ is a union of adjoint orbits. The proofs of the other assertions use the 

886: properties of $\lambda,$ and are completely analogous to those in \cite[Lemma 11 and 

887: Section 3(D)]{SS}.

888: 

889: Putting propositions \ref{regular} and \ref{big}(ii) together we get that the points of 

890: $\ss \cap \ggg^{\pi, \reg}$ are regular for the restriction of $\chi^{\pi}$ to $\ss.$ 

891: In fact, we also have:

892: 

893: \begin {proposition}

894: \label {luna}

895: The restriction of $\chi^{\pi}$ to $\ss \cap \ggg^{\pi, \reg}$ is naturally a 

896: differentiable fiber bundle over $\conf^0_{\pi}.$

897: \end {proposition}

898: 

899: \subsection {An example} \label {sec:exa} Consider the following nilpotent in $\sl(n+1, 

900: \cc):$

901: $$ x=n^+ = \left( \begin {array}{ccccc} 0 & 1 &  &  & \\

902:  & 0 & \ & \ & \\

903:  & \ & 0 & \ & \ \\

904:  & \ & \  & \ldots  &  \ \\

905:  & \ & \ & \ & 0

906: \end {array} \right ) .$$

907: We complete it to a Jacobson-Morozov triple with:

908: $$n^- =  \left( \begin {array}{ccccc} 0 &  &  &  & \\

909: 1 & 0 & \ & \ & \\

910:  & \ & 0 & \ & \ \\

911:  & \ & \  & \ldots  &  \ \\

912:  & \ & \ & \ & 0

913: \end {array} \right ) \ , \  

914: h = \left( \begin {array}{ccccc} 1 &  &  &  & \\

915:  & -1 & \ & \ & \\

916:  & \ & \ 0 & \ \\

917:  & \ & \  & \ldots  &  \ \\

918:  & \ & \ & \ & 0

919: \end {array} \right ).$$

920: 

921: The associated slice $\sjm$ consists of all matrices of the form

922: \begin {equation}

923: \label {sjmm}

924:   \left( \begin {array}{ccccc} \alpha & 1 & 0 & \cdots & 0\\

925: a_{11} & \alpha & a_{12} & \cdots & a_{1n}\\

926: a_{21} & 0 & a_{22} & \cdots & a_{2n} \\

927:  & \cdots & \  & \cdots  &  \ \\

928: a_{n1} & 0 & a_{n2} & \cdots & a_{nn}

929: \end {array} \right ),

930: \end {equation}

931: with $\alpha, a_{ij} \in \cc$ such that $2\alpha + a_{22} + \cdots + a_{nn} = 0.$

932: 

933: Consider the partition $\pi = (n-1,1,1)$ of $n+1.$ Then $\gg^{\pi} \subset \gg=\sl(n+1, 

934: \cc)$ consists of the traceless matrices which admit an $(n-1)$-dimensional eigenspace. For 

935: future reference, we will denote by $\xx_n$ the intersection $\sjm \cap 

936: \gg^{\pi}$ in this case, 

937: and by $\mapp$ the restriction of $\chi^{\pi}$ to $\sjm.$  More precisely,

938: \begin {equation}

939: \label {mapp}

940: \mapp : \xx_n \to \hh^{(n,1)}/W^{(n,1)} \cong \sym^2(\cc),

941: \end {equation}

942: where the identification with $\sym^2(\cc)$ comes form taking the diagonal matrix 

943: $diag(z_1, z_2, 

944: z_3,$ $\cdots, z_3) \in \hh^{\pi}$ to the pair $(z_1, z_2).$ (Note that $z_3 = -(z_1 + 

945: z_2)/(n-1).$) Of course, we can also further identify $\sym^2(\cc)$ with $\cc^2$ by taking 

946: $(z_1, z_2)$ to $(z_1 + z_2, z_1z_2).$ 

947: 

948: \begin {remark}

949: \label {act}

950: More canonically, we define a linear quadruple $T=(F_1, T_1, T_2, f)$ to be the data 

951: consisting of complex vector spaces $F_1, T_1, T_2$ of dimensions $1, n-1, 1,$ 

952: respectively, and a linear isomorphism $f: T_2 \to F_1.$ To every linear quadruple we can 

953: associate an $(n+1)$-dimensional complex vector space $E = F_1 \oplus T_2 \oplus T_1$ and a 

954: flag $F$ of type $1+(n-1)+1$ given by

955: \begin {equation}

956: \label {drapel}

957:  0 \subset F_1 \subset F_2 \subset E, \ \ F_2 = F_1 \oplus T_1.

958: \end {equation}

959: 

960: The isomorphism $f$ produces a nilpotent $n^+=x \in \sl(E)$ with $F_1 = \im (n^+), \ F_2 = 

961: \ker (n^+),$ while its inverse $f^{-1}$ gives rise to a nilpotent $n^- \in \sl(E)$ with 

962: $T_2 = \im (n^-), \ T_2 \oplus T_1 = \ker(n^-).$ We can complete this to a Jacobson-Morozov 

963: triple by setting $h=[n^+, n^-].$ It follows that to $T$ we can naturally associate a space 

964: $\xx(T)$ isomorphic to $\xx_n.$ There is still a natural map $q: \xx(T) \to \sym^2(\cc).$ 

965: \end {remark}

966: 

967: 

968: \subsection {More on nilpotents} The adjoint orbits of nilpotents in $\ggg$ are 

969: classified by the partitions of $N.$ Given the partition $\pi,$ the parabolic subalgebra 

970: $\ppp(F^{st})$ decomposes into a Levi part and a nilpotent part. The nilpotent part 

971: $\mathfrak{n}_{\pi}$ has a unique dense orbit $\oo(x_{\pi}),$ where the nilpotent 

972: element $x_{\pi}$ can be taken to be the Jordan matrix with Jordan cells whose sizes 

973: give the dual partition $\pi^*.$ For example, when $\pi=(N),$ the corresponding 

974: parabolic subalgebra is all of $\ggg,$ the nilpotent $x_{(N)}$ has $N$ Jordan blocks of 

975: size one, and $(x_{(N)}) = \{0\}.$ At the other extreme, when $\pi = (1^N),$ the 

976: corresponding parabolic subalgebra consists of the upper triangular matrices, the 

977: nilpotent $x_{(1^N)}$ is just one Jordan block of size $N,$ and $\oo(x_{(1^N)})$ is the 

978: so-called regular nilpotent orbit. In general, a nilpotent in $\oo(x_{\pi})$ is called 

979: of type $\pi.$

980: 

981: There is a well-known partial ordering on partitions of $N.$ Consider two partitions 

982: $\pi= (\pi_1, \pi_2, \dots)$ and $\rho = (\rho_1,\rho_2, \dots),$ with $\pi_1 \geq 

983: \pi_2 \geq \dots$ and $\rho_1 \geq \rho_2 \geq \dots$ as before, both adding up to $N.$

984: We write $\pi \preccurlyeq \rho$ if $\pi_1 + \dots + \pi_j \leq \rho_1 + \dots + \rho_j$ for 

985: all $j \geq 1.$ The following lemma is elementary:

986: 

987: \begin {lemma}

988: \label {lemm}

989: The closure of $\oo(x_{\pi})$ in $\ggg$ consists of the union of all orbits 

990: $\oo(x_{\rho})$ with $\pi \preccurlyeq \rho.$ These are exactly the nilpotent orbits that 

991: appear in $\gpi.$

992: \end {lemma}

993: 

994: \subsection {Slices at general points} \label {sec:gen} A general element $x \in \gg$ has 

995: a unique decomposition $x=x_s + x_n$ into a semisimple and a nilpotent part. Just as in 

996: Section~\ref{sec:semis}, $\gg_{x_s}$ decomposes into the direct sum of $Z(\gg_{x_s})$ and 

997: $\gg_{x_s}^{\red} = \oplus_i \gg_{x_s}^{\red}[i],$ and each $\gg_{x_s}^{\red}[i]$ is 

998: composed of the traceless endomorphisms of an eigenspace $E_i$ of $x_s.$ The sizes of 

999: these eigenspaces form a partition $\sigma$ of $N.$ The nilpotent part $x_n$ always lies 

1000: in $\gg_{x_s}^{\red};$ denote by $x_n[i]$ its piece in $\gg_{x_s}^{\red}[i].$ We choose 

1001: Jacobson-Morozov slices $\ss^{\jm, \red}[i]$ at $x_n[i]$ in $\gg_{x_s}^{\red}[i],$ and let 

1002: $\ss^{\jm, \red}$ be their direct sum.

1003: 

1004: In \cite[Section 2(D)]{SS} it is proved that $x_s + (Z(\gg_{x_s}) \times \ss^{\jm, \red})$

1005: is a transverse slice at $x$ in $\ggg.$ Using this, it follows that locally near $x,$ the 

1006: adjoint map $\chi$ looks like a linear projection times the product of the 

1007: restriction of the adjoint map of  $\gg_{x_s}^{\red}$ to $\ss^{\jm, \red}.$ This local model 

1008: enables one to describe explicitly which points of $\gg$ are regular for the map $\chi:$ 

1009: namely, the ones for which $x_n$ is a regular nilpotent in $\gg_{x_s}^{\red}.$

1010: 

1011: When $x \in \gg^{\pi},$ by Lemma~\ref{hello} we have that the nilpotents $x_n[i]$ are in 

1012: $\gg_{x_s}^{\red}[i]^{b[i]}$ for some breaking $b \in \bb_{\pi \sigma}.$ We can restrict the 

1013: local 

1014: model to $\gg^{\pi}$ along the lines of Section~\ref{sec:semis}, with the following result:

1015: 

1016: \begin {lemma}

1017: \label {locmodel}

1018: A point $x = x_s + x_n \in \gg^{\pi}$ is regular for the map $\chi^{\pi}$ if and only if 

1019: there exists a breaking $b = (b[1], \dots, b[l])$ of $\pi$ according to $\sigma,$ such that 

1020: $x_n[i] \in \gg_{x_s}^{\red}[i]$ is a nilpotent of type $b[i].$  

1021: \end {lemma}

1022: 

1023: Furthermore, note that a fiber of the map $\chi^{\pi}:\gg^{\pi} \to \hh^{\pi}/W^{\pi}$ is 

1024: uniquely determined by the semisimple orbit $\oo(x_s)$ that it contains, corresponding to some 

1025: partition $\sigma.$ Using Lemma~\ref{lemm}, we get that:

1026: 

1027: \begin {lemma}

1028: \label {structure}

1029: The fiber of $\chi^{\pi}$ containing $\oo(x_s)$ is composed of the orbits of $x_s + x_n,$ 

1030: where $x_n$ is a nilpotent decomposing into $x_n[i] \in \gg_{x_s}^{\red}[i]$ for some 

1031: breaking $b = (b[1], \dots, b[l]) \in \bb_{\pi \sigma},$ and with the types $\rho[i]$ of 

1032: $x_n[i]$ satisfying $b[i] \preccurlyeq \rho[i].$ 

1033: \end {lemma}

1034: 

1035: \subsection {The topology of the fibers} 

1036: \label{sec:top}

1037: Our main objects of interest are the fibers of 

1038: the bundle map $\chi^{\pi}|_{\ss \cap \ggg^{\pi, \reg}}$ from Proposition~\ref{luna}. By 

1039: Proposition~\ref{big}, they are homeomorphic to an arbitrary fiber of the 

1040: restriction of $\tilde \chi^{\pi}$ to $\tilde \ss \cap \gp.$ We pick the nilpotent fiber 

1041: \begin {equation}

1042: \label {nnn}

1043: \nnn_{\rho \pi} = \bigl(\tilde \chi^{\pi} |_{\tilde \ss \cap \gp}\bigr)^{-1}(0).

1044: \end {equation}

1045: 

1046: Here we assumed that the nilpotent $x$ where we took the slice $\ss = \ss_{\rho}$ was 

1047: in the orbit of $x_{\rho},$ for some partition $\rho$ with $\pi \preccurlyeq \rho.$ Observe 

1048: that by Proposition~\ref{invslice}(ii), the topology is independent of which invariant 

1049: slice we choose.

1050: 

1051: Using Lemma~\ref{lemm} we see that there is a natural map

1052: $$ \Pi: \nnn_{\rho \pi} \to \ss_{\rho} \cap \overline{\oo(x_{\pi})}, \ \ (x, F) \to x. $$

1053: 

1054: Using the $\cc^*$-action $\lambda,$ it can be shown that $\nnn_{\rho \pi}$ deformation 

1055: retracts into its ``compact core''

1056: $$ \Pi^{-1}(x) = \{F \in \ff^{\pi} \ | \ x \text{ fixes } F \text { and acts 

1057: trivially on all quotients } F_j/F_{j-1} \}. $$

1058: 

1059: $\Pi^{-1}(x)$ is called the {\it Spaltenstein variety} and appeared first in \cite{Sp}. 

1060: It is a projective variety of half the dimension of $\nnn_{\rho \pi}.$

1061:  

1062: \begin {example}

1063: When $\pi=(1^N), \ \Pi^{-1}(x)$ is the Springer variety of complete flags fixed by a 

1064: given nilpotent element. 

1065: \end {example}

1066: 

1067: \begin {example}

1068: When $\rho = (N),$ we have $x = 0$ and $\Pi^{-1}(x)$ is the partial flag variety 

1069: $\ff^{\pi}.$ In this case $\nnn_{\rho \pi}$ can be identified with the cotangent bundle 

1070: $T^*\ff^{\pi}.$

1071: \end {example}

1072: 

1073: In general, the real dimension $d_{\rho \pi}$ of the Spaltenstein variety  

1074: $\Pi^{-1}(x)$ is given by the formula

1075: \begin {equation}

1076: \label {dimension}

1077: d_{\rho \pi} = \sum_{j\geq 1} \rho_j^2 - \sum_{j\geq 1} \pi_j^2.

1078: \end {equation}

1079: 

1080: In \cite[Sections 3.4 and 3.5]{BM}, Borho and MacPherson studied the rational cohomology of 

1081: a general Spaltenstein variety. For example, they found that the number of $d_{\rho 

1082: \pi}$-dimensional irreducible components of $\Pi^{-1}(x)$ (which is the middle dimensional 

1083: Betti number of $\nnn_{\rho \pi}$) is equal to the Kostka number $K_{\rho \pi}$ which 

1084: counts semistandard Young tableaux of shape $\rho$ and weight $\pi.$ For more information 

1085: on the combinatorics of Kostka numbers and their relevance to representation theory, we 

1086: refer to \cite{McD}. A different viewpoint on these cohomology computations is in 

1087: \cite{N2}.

1088: 

1089: 

1090: \section {The relevant affine variety and its degenerations}

1091: \label{sec:av}

1092: 

1093: From now on we specialize the discussion in the previous section to the case of interest to 

1094: us, by setting $\pi = (1^m(n-1)^m)$ and $\rho = (n^m),$ with $N = nm.$ Note that $W^{\pi} = 

1095: S_m \times S_m$ for $n > 2.$ The case $n=2$ is slightly different, because  $W^{\pi} = 

1096: S_{2m}.$ Since that case was treated by Seidel and Smith in \cite{SS}, we shall restrict our 

1097: attention to $n > 2.$ (The case $n=1$ is trivial, and the resulting link invariant is $\zz$ 

1098: for any link.)

1099: 

1100: \subsection {Motivation} \label {sec:motiv}

1101: Let us briefly explain why our choice of $\pi = (1^m(n-1)^m)$ and 

1102: $\rho = (n^m)$ is natural if we aim at constructing an analogue of Khovanov-Rozansky 

1103: homology. As mentioned in the introduction, the Euler characteristic of Khovanov-Rozansky 

1104: homology is the link polynomial $P_{\nn}.$ Let $V$ be the standard representation of 

1105: $\sl(n, \cc).$ The polynomial $P_{\nn}$ can be defined as follows: we present 

1106: the link as the closure of a braid $b$ on $m$ strands, associate to $b$ a map 

1107: $F_{b}(q): V^{\otimes m} \to V^{\otimes m}$ depending on a quantum parameter $q,$ and 

1108: then take its trace. This is equivalent to looking at the image of $1$ under a map 

1109: \begin {equation}

1110: \label {poly}

1111: \begin {CD}

1112: \cc \to  V^{\otimes m} \otimes (V^{\otimes m})^* @>{F_{b}(q) \times id}>> V^{\otimes m} 

1113: \otimes (V^{\otimes m})^* \to \cc.

1114: \end {CD} 

1115: \end {equation}

1116: 

1117: The composite factors through the space

1118: of invariants 

1119: $$\inv(m,n) = \hom_{\sl(n, \cc)} (\cc, V^{\otimes m} \otimes (V^{\otimes 

1120: m})^*).$$  

1121: 

1122: The dimension of $\inv(m,n)$ is equal to $d(m,n)=$ the number of permutations of $m$ 

1123: elements with longest increasing subsequence of length $\leq n.$ (A table of the values of 

1124: $d(m,n)$ can be found in \cite{O}.) According to Khovanov's principles for 

1125: categorification \cite{Kh2}, we should look for a triangulated category whose Grothendieck 

1126: group has dimension $d(m,n)$ or, more geometrically, for an exact symplectic manifold 

1127: whose middle dimensional Betti number is $d(m,n).$ As explained at the end of 

1128: Section~\ref{sec:top}, the middle dimensional Betti numbers of the smooth fibers of 

1129: $\chi^{\pi}$ are the Kostka numbers. By the Schensted correspondence \cite{Sc}, $d(m,n)$ 

1130: is the same as the Kostka number $K_{(n^m), (1^m(n-1)^m)},$ which explains our choice for 

1131: $\pi$ and $\rho.$

1132: 

1133: 

1134: \subsection {A few properties}

1135: \label {sec:props}

1136: 

1137: From the orbit of $x_{\rho}$ we choose the following nilpotent element, written as a $m 

1138: \times m$ matrix made of $n \times n$ blocks:

1139: $$ N_{m,n} =  \left( \begin {array}{ccccc} 0 & I &  &  & \\

1140:  & 0 & I & \ & \\

1141:  & \ & \ & \ldots & \ \\

1142:  & \ & \  & \  &  I \\

1143:  & \ & \ & \ & 0

1144: \end {array} \right ). $$

1145: Here $I$ and $0$ are the $n$-by-$n$ identity and zero matrix, respectively.

1146: 

1147: We complete this to a Jacobson-Morozov triple $(N^+=N_{m,n}, N^-, H)$ with

1148: $$ H =  \left( \begin {array}{ccccc} (m-1)I &  &  &  & \\

1149:  & (m-3)I & \ & \ & \\

1150:  & \ & (m-5)I & \ & \ \\

1151:  & \ & \  & \dots  &  \ \\

1152:  & \ & \ & \ & (-m+1)I

1153: \end {array} \right ), $$

1154: $$ N^- =  \left( \begin {array}{cccccc} 0 & & &  &  & \\

1155: (m-1)I & 0 & \ & & \\ & \\

1156:  & 2(m-2)I & 0 & \ & \ & \ \\

1157:  & \ & 3(m-3)I & \  & \  &  I \\

1158: & \ & \ & \dots & \ & \\

1159:  & \ & \ & \ & (m-1)I & 0

1160: \end {array} \right ), $$

1161: written in the same form. The induced $\cc^*$ action on $\ggg = \sl(mn)$ is:

1162: \begin {equation}

1163: \label {action}

1164:  \lambda_r \ : Y \to  \left( \begin {array}{ccccc} r^2Y_{11} & Y_{12} & \dots &  &  

1165: r^{4-2m} Y_{1m}\\

1166: r^4 Y_{21} & r^2Y_{22} &  & \ & \\

1167: \dots & \ & \dots & \ & \ \\

1168:  & \ & \  & \  &  Y_{m-1,m} \\

1169: r^{2m}Y_{m1} & \ & \dots & r^4Y_{m,m-1} & r^2Y_{mm}

1170: \end {array} \right ).

1171: \end {equation}

1172: 

1173: Consider the affine space

1174: \begin {equation}

1175: \label {smn}

1176:  \ss_{m, n} = \left \{ \left( \begin {array}{ccccc} Y_{11} & I 

1177: &  &  & \\

1178: Y_{21} & 0 & I & \ & \\

1179: \ldots & \ & \ & \ldots & \ \\

1180:  & \ & \  & \  &  I \\

1181: Y_{m1} & \ & \ & \ & 0

1182: \end {array} \right ): Y_{m1} \in \gl(n) \text{ for } m>1\ , \ Y_{11} \in \sl(n)  \right 

1183: \}.

1184: \end {equation}

1185: 

1186: \begin {lemma}

1187: $\ss_{m,n}$ is a $\lambda$-invariant slice to the adjoint orbit of $N_{m, n}$ in $\sl(mn).$

1188: \end {lemma}

1189: 

1190: \noindent \textbf{Proof. } The adjoint orbit of $N_{m,n}$ has (complex) dimension $(mn)^2 - 

1191: mn^2,$ and $\ss_{m,n}$ has complementary dimension $mn^2 -1.$  Also, $\lambda$-invariance is 

1192: clear from (\ref{action}). Therefore the only thing to show is that $\ss_{m,n}$ intersects 

1193: the tangent space to the adjoint orbit trivially. Let $N^+ + X$ be an element in their 

1194: intersection. Then all but the first $n$ columns of $X$ are nonzero, and $X$ is of the form 

1195: $[N^+, Z]$ for some $Z \in \ggg.$ A quick calculation shows that $Z$ must be upper 

1196: triangular in our block form, which implies that $X = [N^+, Z] =0. \hfill \fin$

1197: \medskip

1198: 

1199: We are interested the simultaneous resolution (\ref{slice}) with $\ss = 

1200: \ss_{m,n}$ and $\pi = \pi(m,n) = (1^m(n-1)^m),$ and in particular in understanding the 

1201: fibers of

1202: $$ \chi^{\pi}|_{\ss}: \ss \cap \ggg^{\pi} \longrightarrow \hh^{\pi}/W^{\pi}.$$

1203: 

1204: We denote a typical element of $\hh^{\pi}/W^{\pi}$ by $\tau = (\lambdav, \muv).$ Here 

1205: $\lambdav = ( \lambda_1, \dots, \lambda_m )$ and $\muv = (\mu_1, \dots, \mu_m )$ are 

1206: unordered $m$-tuples of elements of $\cc,$ with repetitions allowed and such that $\sum 

1207: \lambda_i + (n-1)\sum \mu_j =0.$ Strictly speaking, the element in $\hh^{\pi}/W^{\pi}$ 

1208: associated to $\tau$ is the class of the diagonal matrix $D_{\tau} \in \sl(mn)$ with 

1209: eigenvalues $\lambda_1, \dots, \lambda_m$ with multiplicity $1$ (called {\it thin} 

1210: eigenvalues), and $\mu_1, \dots, \mu_m,$ each with multiplicity $n-1$ (called {\it thick} 

1211: eigenvalues). The fiber $\ymnt=(\chi^{\pi}|_{\ss})^{-1}(\tau)$ is then the intersection of 

1212: the adjoint orbit $\oo_{\tau}$ of $D_{\tau}$ with our chosen slice:

1213: $$ \ymnt = (\chi^{\pi}|_{\ss_{m,n}})^{-1}(\tau) = \ss_{m,n} \cap \oo_{\tau}.$$

1214: 

1215: Inside of $\hh^{\pi}/W^{\pi}$ we have the bipartite configuration space $\conf^0_{\pi} = 

1216: \confb^0_m$ mentioned in the introduction, which corresponds to all the $\lambda_i$ and 

1217: $\mu_j$ being distinct. According to Proposition~\ref{luna}, the fibers $\ymnt$ are smooth 

1218: for $\tau \in \confb^0_m.$  

1219: 

1220: The following two lemmas are generalizations of lemmas 18 and 19 in \cite{SS}:

1221: 

1222: \begin {lemma} 

1223: \label {newl}

1224: For any $Y \in \ss_{m,n}$ and $\mu \in \cc,$ projection to the first $n$ coordinates 

1225: produces an injective map $\ker (\mu I - Y) \to \cc^n.$

1226: \end {lemma}

1227: 

1228: \noindent \textbf{Proof. } Suppose the contrary is true, so that $\ker(\mu I - Y)$ has nonzero 

1229: intersection with  $\{0\}^n \times \cc^{(m-1)n}$. Using

1230: the $\cc^*$-action, one sees that the same holds for $\ker(r^2\mu I

1231: - \lambda_r(Y)).$ In the limit $r \rightarrow 0$ we obtain a nonzero

1232: element in $\ker(N^+) \cap (\{0\}^n \times \cc^{(m-1)n})$, which is a contradiction.

1233: $\hfill \fin$

1234: \medskip

1235: 

1236: 

1237: \begin {lemma} 

1238: \label {identh}

1239: Let $\tau=(\lambdav, \muv) \in \hh^{\pi}/W^{\pi}$ be such that 

1240: $\lambdav=(\lambda_1=0, \lambda_2, \dots, \lambda_m), \muv=(\mu_1=0, \mu_2, \dots, \mu_m).$ 

1241: Denote by $\bar \tau=\bigl( (\lambda_2, \dots, \lambda_m), (\mu_2, \dots, \mu_m) \bigr).$ 

1242: Then the subspace of $Y \in \ymnt$ such that $\ker(Y)$ is $n$-dimensional can be canonically 

1243: identified with $\yy_{m-1, n, \bar \tau}.$ 

1244: \end {lemma}

1245: 

1246: \noindent \textbf{Proof. } From the previous lemma we know that a vector in $\ ker(Y)$ 

1247: is uniquely determined by its first $n$ entries. There are $n$ linearly independent such 

1248: vectors if and only if $Y_{m1} = 0$. The subspace of $\ss_{m,n}$ with $Y_{m1}=0$ can be 

1249: identified with $\ss_{m-1,n}$ in a straightforward way. If $\bar Y \in \yy_{m-1, n, \bar 

1250: \tau} \subset \ss_{m-1,n},$ then $Y$ must fix a partial flag $\bar F \in 

1251: \ff^{(1^{m-2}(n-1)^{m-2})}$ and act diagonally (by the components of $\bar \tau$) on the 

1252: successive quotients. The corresponding matrix $Y \in \ss_{m,n}$ with $Y_{m1} =0$ fixes 

1253: the partial flag $F$ obtained from $\bar F$ by taking direct sum with an arbitrary flag 

1254: of type $(1, n-1)$ in $\cc^n \times \{0\}^{(m-1)n} \subset \ker(Y).$ Using 

1255: Lemma~\ref{reorder}, we get that $Y$ is in $\ymnt.$ Conversely, starting with $F$ fixed 

1256: by $Y \in \ymnt$ we can construct a $\bar F \in \ff^{(1^{m-2}(n-1)^{m-2})}$ fixed by 

1257: $\bar Y$ by intersecting everything with the subspace $\{0\}^n \times \cc^{(m-1)n} 

1258: \subset \cc^n. \hfill \fin$ \medskip

1259: 

1260: 

1261: \subsection {The case $m=1$} \label {sec:mone} When $m=1,$ the slice $\ss_{1,n}$ is the whole 

1262: $\sl(n,\cc).$ The subvariety $\slan$ consists of traceless matrices having 

1263: an eigenspace of dimension at least $n-1,$ cf. Example~\ref{ex0}.  The spaces $\yy_{1,n, 

1264: \tau}$ are the fibers of the map

1265: \begin {equation}

1266: \label {m1}

1267: \chin : \slan \to \cc,

1268: \end {equation}

1269: which takes a matrix into its eigenvalue of multiplicity $\geq n-1.$ This map will play an 

1270: important role in our paper, being part of the local model for constructing vanishing 

1271: projective spaces in Section~\ref{sec:vvo}.

1272: 

1273: \subsection {A thick and a thin eigenvalue coincide} \label{sec:thickthin} As mentioned in 

1274: Section~\ref{sec:props}, the fiber $\ymnt$ is smooth when $\lambda_i$ and $\mu_j$ are all 

1275: distinct. In this subsection we study the structure and form of the singularities in a 

1276: fiber $\ymnt,$ where the $\lambda_i$ and $\mu_j$ that appear in $\tau$ satisfy $\lambda_1 

1277: = \mu_1 = \lambda,$ but are otherwise distinct.

1278: 

1279: Lemma~\ref{structure} tells us the structure of the fiber of $\chi^{\pi}: \gg^{\pi} \to 

1280: \hh^{\pi}/W^{\pi}$ over $\tau.$ The partition $\sigma$ associated to a semisimple element 

1281: in that fiber is $\sigma=(1^{m-1}(n-1)^{m-1}n).$ There is a unique breaking $b$ of $\pi = 

1282: (1^m(n-1)^m)$ according to $\sigma,$ cf. Example~\ref{ex2}. It follows that there are two 

1283: adjoint orbits in the fiber 

1284: $(\chi^{\pi})^{-1}(\tau):$ a regular orbit $\oo^{\reg}$ consisting of matrices with a 

1285: Jordan block of size two and $n-2$ blocks of size one for the eigenvalue $\lambda,$ and a 

1286: subregular orbit $\oo^{\sub}$ of matrices having $n$ independent $\lambda$-eigenvectors. 

1287: The orbit $\oo^{\reg}$ is open and dense in $(\chi^{\pi})^{-1}(\tau),$ while $\oo^{\sub}$ 

1288: is closed. By Lemma~\ref{locmodel}, the points of $\oo^{\reg}$ are regular for 

1289: $\chi^{\pi},$ and then from Proposition~\ref{big}(ii) the points of $\oo^{\reg} \cap 

1290: \ss_{m,n}$ are regular for the restriction of $\chi^{\pi}$ to the slice $\ss_{m,n}.$

1291: 

1292: \begin {remark}

1293: When $\tau$ is as above, we denote by $\ccc_{m,n, \tau} = \oo^{\sub} \cap \ss_{m,n}$ the 

1294: singular set of $\ymnt.$ This 

1295: is a regular fiber of the restriction of $\chi^{\sigma}$ to $\ss_{m,n},$ with $\sigma = 

1296: (1^{m-1}(n-1)^{m-1}n).$ The union of all $\ccc_{m,n, \tau}$ is $\ccc_{m,n} = \ss_{m,n} \cap 

1297: \gg^{\sigma, \reg}.$ By Proposition~\ref{luna}, we have a differentiable fiber bundle: 

1298: \begin {equation}

1299: \label {cmnt}

1300: \ccc_{m,n} \to \hh^{\sigma, reg}/W^{\sigma} = \conf^0_{\sigma}.

1301: \end {equation}

1302: \end {remark} 

1303: 

1304: The structure of the map $\chi^{\pi}|_{\ss_{m,n}}$ near the singular stratum $\oo^{\sub} 

1305: \cap \ss_{m,n}$ is described by the lemma below, which basically states that the local model 

1306: is a trivially fibered version of the $m=1$ case, in other words of the map (\ref{m1}).

1307: 

1308: \begin {lemma}

1309: \label {thickthin}

1310: Let $D \subset \hh^{\pi}/W^{\pi}$ be a disk corresponding to $\lambdav = ( \lambda - 

1311: (n-1)z, \lambda_2, \dots, \lambda_m)$ and $\muv =(\lambda + z, \mu_2, \dots, \mu_m)$ 

1312: with $z \in \cc$ small. Then there is a neighborhood of $\oo^{\sub} \cap \ss_{m,n}$ inside 

1313: $(\chi^{\pi})^{-1}(D) \cap \ss_{m,n},$ and an isomorphism of that with a neighborhood of 

1314: $(\oo^{\sub} \cap \ss_{m,n}) \times \{0\}$ inside  $(\oo^{\sub} \cap \ss_{m,n}) \times 

1315: \slan.$ (Here $0 \in \sl(n, \cc)$ is the zero matrix.) The isomorphism 

1316: fits into a commutative diagram:

1317: $$ \begin {CD}

1318: (\chi^{\pi})^{-1}(D) \cap \ss_{m,n} @>{\text{local } \cong \ \text{near} \ \oo^{\sub} 

1319: \cap \ss_{m,n}}>> 

1320: (\oo^{\sub} \cap \ss_{m,n}) \times \slan \\

1321: @V{\chi^{\pi}}VV   @V{\chin}VV \\

1322: D @>{\quad\quad\quad z \quad\quad\quad}>> \cc

1323: \end {CD} $$

1324: \end {lemma}

1325: 

1326: \noindent \textbf{Proof. } Let us first look at a neighborhood 

1327: $\oo^{\sub}$ inside $(\chi^{\pi})^{-1}(D),$ without restricting to $\ss_{m,n}.$ From 

1328: Section~\ref{sec:semis} we know that $\chi^{\pi}$ looks locally like a fibered version 

1329: of the product of $\chin,$ $n-1$ copies of $\chi^{(n-1)},$ and $n-1$ copies of 

1330: $\chi^{(1)},$ see (\ref{cairo2}). Since $\chi^{(n-1)}$ and $\chi^{(1)}$ correspond to 

1331: trivial partitions, they are all just the identity map for a point. Thus we know 

1332: that $\chi^{\pi}$ looks like a fibered version of $\chin$ near $\oo^{\sub}.$ 

1333: 

1334: We need to show that when we restrict to $\ss_{m,n},$ this fibered structure is preserved 

1335: and, moreover, that the normal data along $\oo^{\sub} \cap \ss_{m,n}$ is trivial. At every 

1336: point $Y \in \oo^{\sub} \cap \ss_{m,n},$ we choose a subspace $R_Y \subset T_Y\ss_{m,n}$ 

1337: which is complementary to $T_Y(\oo^{\sub} \cap \ss_{m,n})$ and depends holomorphically on 

1338: $Y.$ This is possible because $\oo^{\sub} \cap \ss_{m,n}$ is affine, and therefore the 

1339: relevant $Ext^1$ obstruction group is zero. The spaces $(Y + R_Y) \cap \gg^{\pi}$ form a 

1340: local tubular neighborhood of $\oo^{\sub} \cap \ss_{m,n}$ inside $\gg^{\pi} \cap \ss_{m,n}.$ 

1341: On the other hand, we also have the canonical slice $\ss^{ss} = \ss^{ss}_Y$ to $\oo^{\sub},$ 

1342: and a canonical local isomorphism (\ref{caniso}) between $(Y+R_Y) \cap \gg^{\pi}$ and 

1343: $\ss^{ss}_Y \cap \gg^{\pi},$ which only moves points inside their adjoint orbits. Hence the 

1344: restriction of $\chi^{\pi}$ to $\ss^{ss}_Y \cap \gg^{\pi}$ serves as a local model for 

1345: $\chi^{\pi}|_{\ss_{m,n}}$ near $\oo^{\sub} \cap \ss_{m,n}.$

1346: 

1347: It remains to check triviality of the normal data. By Lemma~\ref{hello}, the intersection 

1348: $\ss^{ss}_Y \cap \gg^{\pi}$ can be identified via translation with 

1349: $$Z(\gg_Y) \times \sl(E_Y(\lambda))^{(1(n-1))} \times \prod_{i=2}^m 

1350: \sl(E_Y(\lambda_i))^{(1)} \times \prod_{i=2}^m \sl(E_Y(\mu_i))^{(n-1)}.$$

1351: 

1352: Here by $E_Y(\alpha)$ we denoted the eigenspace of $Y$ with eigenvalue $\alpha.$ Note that 

1353: each of the $\sl(E_Y(\lambda_i))^{(1)}$ and $ \sl(E_Y(\mu_i))^{(n-1)}$ is just a point, 

1354: while $Z(\gg_Y)$ can be identified with $\cc^{2m-2}$ in a canonical way. Finally, by 

1355: Lemma~\ref{newl} we have a preferred isomorphism of $E_Y(\lambda)$ with $\cc^n,$ and thus 

1356: of $\sl(E_Y(\lambda))^{(1(n-1))}$ with $\slan.$ This completes the proof of 

1357: triviality for the normal data. It is easy to see that all the 

1358: isomorphisms which we used depended holomorphically on $Y.$ $\hfill \fin$

1359: \medskip

1360: 

1361: \begin {remark} \label{remcom} Consider the partition $\pi^+=(1^{m+n-1}(n-1)^{m-1}).$ Since 

1362: $\pi^+ \preccurlyeq \pi,$ we have that $$(\chi^{\pi})^{-1}(D) \cap \ss_{m,n}\ \subset \ 

1363: \gg^{\pi} \cap \ss_{m,n} \ \subset \ \gg^{\pi^+} \cap \ss_{m,n}.$$ We can proceed as in 

1364: the proof of Lemma~\ref{thickthin} and 

1365: study the structure of $\gg^{\pi^+} \cap \ss_{m,n}$ near $(\oo^{\sub} \cap \ss_{m,n}).$ There 

1366: are now several breakings of $\pi^+$ according to $\sigma = (1^{m-1}(n-1)^{m-1}n),$ cf. 

1367: Example~\ref{ex3}. One of them is $b^+$ given by $n=1+\cdots +1, \ (n-1), \cdots, (n-1), 1, 

1368: \cdots, 1.$ 

1369: Lemma~\ref{hello} says that a neighborhood of $(\oo^{\sub} \cap \ss_{m,n})$ in $\gg^{\pi^+} 

1370: \cap \ss_{m,n}$ has several components. We denote the one corresponding to $b^+$ by 

1371: $(\gg^{\pi^+} \cap \ss_{m,n})_{b^+}.$ It is isomorphic to $(\oo^{\sub} \cap \ss_{m,n}) \times 

1372: \sl(n, \cc).$ (The normal data is trivial just as in Lemma~\ref{thickthin}.) Furthermore, 

1373: this local isomorphism is compatible with the one in Lemma~\ref{thickthin}, in the sense that 

1374: there is a commutative diagram 

1375: $$ \begin {CD} 

1376: \gg^{\pi} \cap \ss_{m,n} @>{\text{local } \cong }>> (\oo^{\sub} \cap \ss_{m,n}) 

1377: \times \slan \\ @VVV @VVV \\ (\gg^{\pi^+} \cap \ss_{m,n})_{b^+} @>{\text{local 

1378: } \cong }>> (\oo^{\sub} \cap \ss_{m,n}) \times \sl(n, \cc), 

1379: \end {CD} $$ 

1380: where the vertical maps are inclusions. \end {remark}

1381: 

1382: We can generalize Lemma~\ref{thickthin} to the case when several pairs of one thick and one 

1383: thin eigenvalue come together. Let $\tau = (\lambdav, \muv)$ be a point in $\hh^{\pi}/W^{\pi}$ 

1384: such that $\lambda_1 = \mu_1, \lambda_2 = \mu_2, \dots,  \lambda_k = \mu_k,$ and these 

1385: are the only coincidences. There are $2^k$ orbits in the fiber of $\chi^{\pi}$ 

1386: corresponding to which of the $k$ relevant Jordan blocks are semisimple. The smallest 

1387: orbit $\oo^{\min}$ where all the $k$ blocks are semisimple is closed in 

1388: $(\chi^{\pi})^{-1}(\tau).$ An adaptation of the arguments in the proof of 

1389: Lemma~\ref{thickthin} gives:

1390: 

1391: \begin {lemma}

1392: \label {multithickthin}

1393: Let $P \subset \hh^{\pi}/W^{\pi}$ be a polydisk corresponding to $\lambdav = ( \lambda -

1394: (n-1)z_1, \dots, \lambda_k - (n-1)z_k, \lambda_{k+1}, \dots, \lambda_m)$ and $\muv =(\lambda + 

1395: z_1, \dots, \lambda_k+z_k, \mu_{k+1}, \dots, \mu_m)$

1396: with the $z_k$'s small. Then there is a neighborhood of $\oo^{\min} \cap \ss_{m,n}$ 

1397: inside $(\chi^{\pi})^{-1}(P) \cap \ss_{m,n},$ and an isomorphism of that with a 

1398: neighborhood of

1399: $(\oo^{\min} \cap \ss_{m,n}) \times \{0\}^k$ inside  $(\oo^{\min} \cap \ss_{m,n}) \times

1400: \bigl( \slan \bigr)^k.$ The isomorphism fits into

1401: a commutative diagram:

1402: $$ \begin {CD}

1403: (\chi^{\pi})^{-1}(P) \cap \ss_{m,n} @>{\text{local } \cong \ \text{near} \ \oo^{\min}

1404: \cap \ss_{m,n}}>>

1405: (\oo^{\min} \cap \ss_{m,n}) \times \bigl( \slan \bigr)^k \\

1406: @V{\chi^{\pi}}VV   @V{(\chin, \dots, \chin)}VV \\

1407: P @>{\quad\quad(z_1, \dots, z_k)\quad\quad}>> \cc^k

1408: \end {CD} $$

1409: \end {lemma}

1410: 

1411: 

1412: \subsection {Two thin eigenvalues coincide} Although not necessary for the rest of the 

1413: paper, it is instructive to consider the case when $\tau = (\lambdav, \muv)$ has $\lambda_1 

1414: = \lambda_2 = \lambda,$ and all the other $\lambda_i$'s and $\mu_j$'s are distinct from 

1415: each other and from $\lambda.$ Let us assume that $n > 3.$ The corresponding partition 

1416: $\sigma$ is now $(1^{m-2}2(n-1)^m),$ and there is again a unique breaking of $\pi$ 

1417: according to $\sigma,$ cf. Example~\ref{ex4}. We get two orbits $\oo^{\reg}$ and 

1418: $\oo^{\sub}$ in the fiber, just as in Section~\ref{sec:thickthin}. The orbit $\oo^{\sub}$ 

1419: is closed, and the local model around it is the map $\chi^{(1^2)} : \sl(2,\cc) \to \cc,$ 

1420: which is basically the $A_1$-singularity $\cc^3 \to \cc, \ (a,b,c) \to a^2 + b^2 + c^2.$ 

1421: The same arguments as in the proof of Lemma~\ref{thickthin} show that the local model for 

1422: $\chi^{\pi}|_{\ss_{m,n}}$ near $\oo^{\sub} \cap \ss_{m,n}$ is a fibered version of the 

1423: $A_1$-singularity. However, note that we do not have a canonical identification of the 

1424: $\lambda$-eigenspace with $\cc^2,$ and so we cannot claim the triviality of the normal 

1425: data.

1426: 

1427: 

1428: \subsection {One thick and two thin eigenvalues coincide} \label {sec:ttt} A slightly more 

1429: involved situation is when $\tau = (\lambdav, \muv)$ has $\lambda_1 = \lambda_2 = \mu_1 = 

1430: \lambda,$ and these are the only coincidences. The partition $\sigma$ is 

1431: $(1^{m-2}(n-1)^{m-1}(n+1)),$ and there is still a unique breaking of $\pi$ according to 

1432: $\sigma,$ namely $(n+1) = (n-1) + 1 +1$ and the rest trivial, cf. Example~\ref{ex5}. By 

1433: Lemma~\ref{structure} the 

1434: fiber $(\chi^{\pi})^{-1}(\tau)$ consists of three orbits: $\oo^{\reg}$ made of matrices 

1435: with a Jordan block of size $3$ and $n-2$ blocks of size $1$ for the eigenvalue $\lambda;$ 

1436: a subregular orbit $\oo^{\sub}$ with a Jordan block of size $2$ and $n-1$ blocks of size 

1437: $1$ for $\lambda$; and a minimal semisimple orbit with $n+1$ independent 

1438: $\lambda$-eigenvectors. However, according to Lemma~\ref{newl}, the minimal orbit does not 

1439: intersect $\ss_{m,n}.$ Thus $\ymnt$ is only made of two strata: an open and dense part 

1440: $\oo^{\reg} \cap \ss_{m,n},$ and a closed part $\oo^{\sub} \cap \ss_{m,n}.$

1441: 

1442: The discussion from Section~\ref{sec:gen} can be applied here to find a local model for the 

1443: map $\chi^{\pi}$ near $\oo^{\sub}.$ One can show that the same model is valid for the 

1444: restriction of $\chi^{\pi}$ to $\ss_{m,n},$ along the lines of the proof of 

1445: Lemma~\ref{thickthin}, and then study the normal data in the fibers (which turns out to be 

1446: nontrivial). To be precise, for every $Y \in \oo^{\sub} \cap \ss_{m,n},$ we denote by $E_Y$ 

1447: the kernel of $(\lambda I - Y)^2,$ by $F_{1,Y}$ the image of $E_Y$ under the map $\lambda I 

1448: - Y,$ and by $F_{2,Y}$ the kernel of $\lambda I - Y.$ We let $\ee, \ff_1, \ff_2$ be the 

1449: complex vector bundles over $\oo^{\sub} \cap \ss_{m,n}$ whose fibers over $Y$ are $E_Y, 

1450: F_{1,Y}, F_{2,Y},$ respectively. Since $Y$ is affine, we can choose complements $T_{1,Y}$ 

1451: of $F_{1,Y}$ in $F_{2, Y}$ and $T_{2,Y}$ of $F_{2,Y}$ in $E_Y$ that depend holomorphically 

1452: on $Y.$ The map $\lambda I - Y$ induces a linear isomorphism from $T_{2,Y}$ to $F_{1, Y}.$ 

1453: This gives a linear quadruple $T_Y$ in the sense of Remark~\ref{act}. Since the data $T_Y$ 

1454: depends holomorphically on $Y,$ we call the whole thing a {\it holomorphic quadruple 

1455: bundle} over $\oo^{\sub} \cap \ss_{m,n},$ and denote it by $\tt.$ Using Remark~\ref{act}, 

1456: we have an associated space $\xx(T_Y)$ for each $Y;$ the corresponding fiber bundle over 

1457: $\oo^{\sub} \cap \ss_{m,n}$ is denoted $\xx(\tt).$ The fibers of $\xx$ are isomorphic to 

1458: the space $\xx_n$ which appears in (\ref{mapp}). There is also a map $q: \xx(\tt) \to 

1459: \sym^2(\cc)$ which is given by (\ref{mapp}) in each fiber.

1460: 

1461: \begin {lemma}

1462: \label {ttt}

1463: Consider the bidisk $P \subset \hh^{\pi}/W^{\pi}$ corresponding to $\lambdav = ( 

1464: \lambda + z_1, \lambda+z_2, \dots, \lambda_m)$ and $\muv =(\lambda + z_3, \mu_2, \dots, 

1465: \mu_m)$ with $z_1, z_2, z_3 \in \cc$ small, $z_1 + z_2 + (n-1)z_3 = 0.$

1466: 

1467: Then there is a neighborhood of $\oo^{\sub} \cap \ss_{m,n}$ inside

1468: $(\chi^{\pi})^{-1}(P) \cap \ss_{m,n},$ and an isomorphism of that with a neighborhood of

1469: the zero-section inside  $\xx(\tt).$ The isomorphism fits 

1470: into a commutative diagram:

1471: $$ \begin {CD}

1472: (\chi^{\pi})^{-1}(P) \cap \ss_{m,n} @>{\text{local } \cong \ \text{near} \ \oo^{\sub}

1473: \cap \ss_{m,n}}>> \xx(\tt) \\

1474: @V{\chi^{\pi}}VV   @VV{q=(q_1, q_2)}V \\

1475: P @>{\quad\quad(z_1,z_2)\quad\quad}>> \sym^2(\cc).

1476: \end {CD} $$

1477: \end {lemma}

1478: 

1479: In short, this says that the map $\chi^{\pi}|_{\ss_{m,n}}$ looks locally near $\oo^{\sub} 

1480: \cap \ss_{m,n}$ like a fibered version of the map (\ref{mapp}). The possible nontriviality

1481: of the normal data is encoded in the structure of the fiber bundle $\xx(\tt).$ In 

1482: particular, the bundles $\ee$ and $\ff_1$ can be nontrivial. On the other hand, note that 

1483: by Lemma~\ref{newl}, the bundle $\ff_2$ is always trivial.

1484: 

1485: There is also an analogue of Remark~\ref{remcom}:

1486: \begin {remark}

1487: \label {remcom2}

1488: One breaking of $\pi^+=(1^{m+n-1}(n-1)^{m-1})$ according to $\sigma=(1^{m-2}(n-1)^{m-1}(n+1))$

1489: is $\beta^+$ given by $(n+1)=1+\cdots+1, \ (n-1), \cdots, (n-1), 1, \cdots, 1.$ Take a 

1490: neighborhood of $\oo^{\sub} \cap \ss_{m,n}$ in $\gg^{\pi^+} \cap \ss_{m,n}$ as in 

1491: Lemma~\ref{hello}. Denote by  $(\gg^{\pi^+} \cap \ss_{m,n})_{\beta^+}$ the part corresponding 

1492: to $\beta^+.$ Then there is a commutative diagram made of inclusions and local isomorphisms:

1493: $$ \begin {CD}

1494: \gg^{\pi} \cap \ss_{m,n} @>{\text{local } \cong}>>  \xx(\tt)  \\ 

1495: @VVV @VVV \\ (\gg^{\pi^+} \cap \ss_{m,n})_{\beta^+} 

1496: @>{\text{local } \cong }>>  \sl(\ee).

1497: \end {CD} $$

1498: \end {remark}

1499: 

1500: 

1501: \section {Parallel transport and various vanishing objects}

1502: \label {sec:vvo}

1503: 

1504: \subsection {Parallel transport} \label{sec:parallel}

1505: 

1506: In this section we review the definition of rescaled parallel transport in a Stein fibration, 

1507: following Seidel and Smith \cite[Section 4(A)]{SS}.

1508: 

1509: Let $p: Y \to T$ be a holomorphic map between complex manifolds, which is also a submersion 

1510: with fibers $Y_t.$ Let $\gamma: [0,1] \to T$ be a path on the base. The parallel transport 

1511: vector field $H_{\gamma}$ on the pullback $\gamma^*Y \to [0,1]$ consists of the sections of 

1512: $TY|Y_{\gamma(s)}$ which project to $\gamma'(s)$ and are orthogonal to the vertical tangent 

1513: bundle $T(Y_{\gamma(s)}).$ Thus,

1514: \begin {equation}

1515: \label {hgam}

1516: H_{\gamma} = \frac{\nabla p}{\| \nabla p \|^2 } \gamma'(s).

1517: \end {equation}

1518: 

1519: In some cases, for example if $p$ is proper or if we have 

1520: suitable 

1521: estimates on $H_{\gamma},$ integrating $H_{\gamma}$ yields a symplectic isomorphism $h_{\gamma} 

1522: : Y_{\gamma(0)} \to Y_{\gamma(1)},$ called \emph{(naive) parallel transport.} In general, 

1523: however, the lack of compactness means that integral lines may not exist everywhere, so 

1524: that $h_{\gamma}$ is only defined on compact subsets $P \subset Y_{\gamma(0)}$ for a short time 

1525: $\epsilon > 0$ which depends on $P.$

1526: 

1527: To fix this problem, we use \emph {rescaled parallel transport}. Assume that there exists a 

1528: function $\psi: Y \to \rr$ with the following properties:

1529: \begin {eqnarray}

1530: \label {cond1}

1531: &\bullet & \psi  \text{ is proper and bounded below. } \\

1532: \label {cond2}

1533: &\bullet & -dd^c \psi > 0, \text{ so that } \Omega=-dd^c\psi \text{ is a K\"ahler form on }Y. 

1534: \\

1535: \label{cond3}

1536: & \bullet & \text{ Outside a compact set of }Y, \ \text{we have} \ \|\nabla \psi\| \leq

1537: \rho \psi \text{ for some }\rho > 0; \\

1538: \label {cond4}

1539: & \bullet & \text{ The fiberwise critical set } \{y \in Y : d\psi|_{\ker(Dp)} = 0\}  \text{ 

1540: maps properly to }T.

1541: \end {eqnarray}

1542: 

1543: The Liouville vector field $Z_t = \nabla (\psi|_{Y_t})$ on $Y_t$ is well-defined for all times 

1544: because of condition (\ref{cond3}). Given the path $\gamma,$ condition (\ref{cond4}) implies 

1545: that there exists $c > 0$ such that the critical values of $\psi$ on the fibers $Y_{\gamma(s)}$ 

1546: all lie in $[0,c).$ Using this and (\ref{cond1}) we can find $\sigma > 0$ such that the 

1547: integral lines of $ \tilde H_{\gamma} = H_{\gamma} - \sigma Z_{\gamma(s)}$ stay inside 

1548: $\psi^{-1}([0,c]).$ Integrating the flow $\tilde H_{\gamma}$ and then composing with the time 

1549: $\sigma$ map of the Liouville flow on $Y_{\gamma(1)}$ yields a symplectic embedding

1550: $$ h_{\gamma}^{\resc}: Y_{\gamma(0)} \cap \psi^{-1}([0,c]) \longrightarrow Y_{\gamma(1)}.$$

1551: 

1552: This is called rescaled parallel transport. It depends on $\sigma$ only up to isotopy in the 

1553: class of symplectic embeddings. Replacing $c$ by a larger value yields a map defined on a 

1554: bigger set, whose restriction to the smaller set is isotopic to the original one. As a 

1555: consequence, the image $h_{\gamma}^{\resc}(L)$ is well-defined (up to Lagrangian isotopy) for 

1556: any compact Lagrangian submanifold $L \subset Y_{\gamma(0)}.$

1557: 

1558: 

1559: \subsection {Vanishing cycles for singular metrics.} \label{sec:vsm} Let $\Omega$ be an 

1560: arbitrary 

1561: K\"ahler form on $Y=\cc^n,$ and denote by $g$ the corresponding K\"ahler metric. Consider the 

1562: projection

1563: $$ p: \cc^n \to \cc, \ p(z_1, \dots, z_n) = z_1^2 + \dots + z_n^2,$$

1564: and equip the fibers with the induced metrics and forms.

1565: 

1566: The real part $re(p)$ is a Morse function with a critical point at the origin. The 

1567: \textit{stable manifold} $W$ is defined as the set of points $y \in \cc^n$ such that the flow 

1568: line of $-\nabla re(p)$ starting at $y$ exists for all times $s \geq 0,$ and converges to 

1569: zero as $s \to \infty.$ 

1570: 

1571: Since $\Omega$ is K\"ahler, the negative gradient flow of $re(p)$ is the same as the 

1572: Hamiltonian vector field of the imaginary part $im(p).$ This preserves $im(p),$ hence $W$ 

1573: lies in the preimage of $\rr_{\geq 0}.$ The intersection $C_t$ of $W$ with a fiber 

1574: $Y_t=p^{-1}(t)$ for $t >0$ is called a \textit{vanishing cycle}. It is well-known that for 

1575: $t$ small $C_t$ is a smooth manifold and, in fact, is diffeomorphic to $S^{n-1}.$ (See 

1576: \cite{J}, for example.) An alternate definition is as follows. Consider the path $\gamma: 

1577: [0,1] \to \cc, \ \gamma(s)=(1-s)t.$ Since $-\nabla re(p)$ is proportional to the naive 

1578: parallel transport $H_{\gamma},$ we have \begin {equation} \label {altvc} C_t = \{ y \in Y_t 

1579: \ | \ h_{\gamma|[0,s]} \text{ is defined near }y \text{ for } s <1, \text{ and } 

1580: h_{\gamma|[0,s]}(y) \to 0 \text{ as } s \to 1 \}. \end {equation}

1581: 

1582: Because the gradient flow preserves $\Omega,$ by taking the limit $s \to 1$ we find that 

1583: $\Omega$ vanishes on $TC_t.$ In other words, $C_t \subset Y_t$ is Lagrangian.

1584: 

1585: We need a generalization of this construction in the case when the K\"ahler form 

1586: $\Omega$ is degenerate at the origin. In that situation the gradient flow $-\nabla re(p)$ can 

1587: take a nonzero point into zero in finite time. However, we can still define a vanishing cycle 

1588: $C_t$ using parallel transport maps. 

1589: 

1590: \begin {proposition}

1591: \label {vansing}

1592: Let $\Omega$ be a smooth two-form on $\cc^n$ that is real analytic in a neighborhood 

1593: of zero and K\"ahler on $\cc^{n} - \{ 0 \}.$ For $t > 0,$ define $C_t$ by (\ref{altvc}).

1594: Then, for $t > 0$ sufficiently small, $C_t$ is a Lagrangian $(n-1)$-sphere in $Y_t.$

1595: \end {proposition}

1596: 

1597: (It is conceivable that the real analyticity consition is unnecessary. However, our proof 

1598: is fundamentally different from the one in the nondegenerate case, and makes essential use of 

1599: real analyticity.)

1600: 

1601: \medskip

1602: 

1603: \noindent \textbf {Proof of Proposition ~\ref{vansing}. } For $t \in \rr,$ we will denote by 

1604: $Y_{t+i\rr}$ the preimage of $t$ under the map $re(p): Y \to \rr.$ This is the union of 

1605: $Y_{t+i\tau}$ for all $\tau \in \rr.$ Note that $Y_{t+i\rr}$ is diffeomorphic to $S^{n-1} 

1606: \times \rr^n$ for all $t \neq 0.$

1607: 

1608: Consider the projection from $\cc^n$ to its imaginary part 

1609: $$\pim: \cc^n \to \rr^n, \ \pim(z_1, \dots, z_n) = (im(z_1), \dots, im(z_n)).$$ 

1610: 

1611: Denote by $\iota$ the inclusion of $Y_{i\rr}$ into $\cc^n.$ Take the preimage $V=(\pim 

1612: \circ \iota)^{-1}(B) \subset Y_{i\rr}$ of a ball $B^n=B^n(\epsilon) \subset \rr^n$ centered 

1613: at the 

1614: origin. Given a point $y \in V -\{0\}$ with $p(y)=i\tau,$ we can consider the translate 

1615: $\gamma_{\tau}$ of $\gamma$ by $i \tau,$ as a path in $\cc.$ Reverse parallel transport along 

1616: the corresponding $\gamma_{\tau}$ can be applied to all points in $V-\{0\}$ for at least some 

1617: small time $t > 0.$ (Up to a reparametrization of time, this is equivalent to going along the 

1618: forward gradient flow of $re(p).$) The image of this reverse transport, together with the 

1619: vanishing cycle $C=C_t$, forms an open set $U \subset Y_{t+i\rr}.$ We can define $h: U \to V$ 

1620: by setting $h(y)=0$ for $y \in C,$ and $h(y)=h_{\gamma_{\tau}}(y)$ for $y\in V-C,$ with 

1621: $\tau$ being the imaginary part of $p(y).$

1622: 

1623: If our parameters $\epsilon$ and $\tau$ were sufficiently small, the whole parallel transport 

1624: of $U$ along the $\gamma_{\tau}$'s happens in the neighborhood where $\Omega$ is real 

1625: analytic. The solution of an ODE with real analytic coefficients and initial data is a 

1626: real analytic function by the Cauchy-Kovalevsky theorem. Therefore the map $\iota \circ h : 

1627: U \to \cc^n$ is real analytic, and so must be 

1628: $$f=\pim \circ h: U \to B^n.$$

1629: 

1630: The restriction of $h$ to $U - C$ is a diffeomorphism onto $V-\{ 0 \}.$ It follows 

1631: that $f: U \to B^n$ is an $S^{n-1}$-fibration over $B^n - \{0\},$ outside of the vanishing 

1632: cycle $C.$ We aim to show that $C=f^{-1}(0)$ is also an $(n-1)$-sphere.

1633: 

1634: Note that $C$ is an analytic subvariety of the real analytic manifold $Y_{t+i\rr} = S^{n-1} 

1635: \times \rr^n.$ Let us recall a few facts about real analytic varieties, cf. \cite[Section 

1636: 6.3]{KP}, \cite{Lo}. Every analytic variety $A$ admits a Whitney stratification and, in 

1637: particular, it has a top-dimensional stratum of dimension $d=\dim A.$ Every point $y \in A$ 

1638: admits a Zariski tangent space $T_yA,$ and (assuming $A$ is connected) we have $\dim (T_yA) 

1639: \geq \dim A$ for all $y,$ with equality if and only if $y$ is a regular point, i.e. part of 

1640: the top-dimensional stratum.

1641: 

1642: In our case, since parallel transport preserves $\Omega,$ it follows that $\Omega$ must 

1643: vanish on $T_yC$ for all $y \in C,$ cf. the remark following (\ref{altvc}). Since $C 

1644: \subset Y_t$ and the 

1645: restriction of $\Omega$ to the complex submanifold $Y_t \subset Y$ is nondegenerate, 

1646: this implies that $\dim (T_yC) \leq n-1$ for all $y \in C.$ In particular, the dimension 

1647: of $C$ is at most $n-1.$

1648: 

1649: On the other hand, each fiber $f^{-1}(x)$ for $x \neq 0$ represents a generator of 

1650: $H_{n-1}(S^{n-1} \times \rr^n).$ Hence $f^{-1}(x)$ must intersect each $\{z\} \times \rr^n 

1651: \subset S^{n-1} \times \rr^n$ nontrivially. Taking the limit $x \to 0,$ we find that the 

1652: projection of $f^{-1}(0) = C \subset S^{n-1} \times \rr^n$ to the $S^{n-1}$ factor is 

1653: surjective as well. As a consequence, $C$ must have dimension exactly $n-1.$

1654: 

1655: For $n >1,$ we also claim that $C$ is connected. If it were disconnected, then a small 

1656: neighborhood $f^{-1}(B^n(\epsilon'))$ for $\epsilon' < \epsilon$ would also be 

1657: disconnected, and the same would hold for its boundary $f^{-1}(S^{n-1}(\epsilon')).$ 

1658: However, this boundary is diffeomorphic to $S^{n-1} \times S^{n-1}$, and we arrive at a 

1659: contradiction.

1660: 

1661: Since $C$ is a connected analytic variety of dimension $n-1$ and $\dim (T_yC) \leq n-1$ at any 

1662: $y \in C,$ we deduce that all of its points are regular. In other words, $C$ is an 

1663: $(n-1)$-dimensional real analytic manifold. As we already noted, $\Omega$ vanishes on its 

1664: tangent space.

1665: 

1666: To show that $C$ is a sphere, we interpolate between $\Omega^{(0)}=\Omega$ and a form 

1667: $\Omega^{(1)}$ that is nondegenerate over all of $\cc^n.$ For example, if $N$ is the null 

1668: space of $\Omega$ at the origin and $N^{\perp}$ is its orthogonal complement in the 

1669: standard Hermitian metric, we can choose $\Omega^{(1)}$ to be constant over all of $\cc^n$ 

1670: (with respect to the standard holomorphic coordinates), identical to $\Omega$ on 

1671: $N^{\perp}$ (at the origin), and nondegenerate on $N.$ Then, all the forms in the family 

1672: $\Omega^{(r)} = (1-r)\Omega^{(0)} + r\Omega^{(1)}, \ r\in (0,1],$ are real analytic and 

1673: K\"ahler in a neighborhood of zero. For $t$ sufficiently small, we can thus relate the 

1674: vanishing cycles $C=C_t^{(0)}$ to $C_t^{(1)}$ by a smooth family of manifolds. Note that 

1675: $\Omega^{(1)}$ is nondegenerate, hence we already know that $C_t^{(1)}$ is a sphere. Since 

1676: all the manifolds in a smooth family are diffeomorphic, we conclude that $C$ is also a 

1677: sphere.

1678: 

1679: For $n=1,$ a similar argument shows that $C$ is a disjoint union of exactly two points. 

1680: $\hfill \fin$

1681: 

1682: \begin {remark}

1683: \label {multsc}

1684: By multiplying $p$ with some scalar in $S^1,$ we can define vanishing cycles $C_t \subset 

1685: Y_t$ for all sufficiently small $t \in \cc^*.$

1686: \end {remark}

1687: 

1688: This discussion can be generalized to relative vanishing cycles. Take any complex manifold 

1689: $X$ and consider the projection

1690: $$ p: Y=X \times \cc^n \to \cc, \ p(x,z_1, \dots, z_n) = z_1^2 + \dots + z_n^2.$$

1691: 

1692: Equip $Y$ with any K\"ahler form $\Omega$, possibly degenerate on $X \times \{ 0 \}^n.$ The 

1693: function $re(p)$ is now Morse-Bott, and its critical point set can be identified with $X.$ Let 

1694: $K \subset X$ be a compact Lagrangian submanifold. For $t \in \cc^*,$ the \textit{relative 

1695: vanishing cycle} $C_t$ associated to $K$ is defined as the set of points $y \in Y_t=p^{-1}(t)$ 

1696: which are taken into $K$ by parallel transport along the path $\gamma: [0,1] \to \cc, \ 

1697: \gamma(s)=(1-s)t,$ as $s \to 1.$ The same arguments as in the proof of 

1698: Proposition~\ref{vansing} show:

1699: 

1700: \begin {proposition}

1701: \label {relvs}

1702: Assume that $\Omega$ is real analytic in a neighborhood of $X.$ Then for sufficiently small 

1703: $t \in \cc^*, \ C_t$ is a Lagrangian submanifold of $Y_t$ diffeomorphic to $K \times 

1704: S^{n-1}.$ 

1705: \end {proposition}

1706: 

1707: The case when $\Omega$ is nondegenerate appears in \cite[Lemma 26]{SS}.

1708: 

1709: 

1710: \subsection {Vanishing projective spaces} \label {sec:van}

1711: 

1712: We now describe a construction similar to that of vanishing cycles, but where instead of a 

1713: sphere we obtain a complex projective space $\cpn$. Our main interest lies in a fibered 

1714: version of this (treated in the next section). However, we decided to present the simpler 

1715: situation first in order to make clear all the ideas involved.

1716: 

1717: The construction takes place in the space $Z=\slan$ from 

1718: Section~\ref{sec:mone}, for $n > 2.$ Recall that

1719: $$Z = \{ A \in \sl(n, \cc) \ | \text{ there exists } t \in \cc \text{ with } \dim \ker (A- tI) 

1720: \geq n-1 \},$$

1721: and the map $\chin$ takes a matrix $A$ to the corresponding $t \in \cc.$ 

1722: 

1723: The following alternate description is also helpful:

1724: 

1725: \begin {lemma}

1726: \label {git}

1727: Consider the linear action of $\cc^*$ on $\cc^{2n}$ with weights $1$ and $-1,$ each with 

1728: multiplicty $n.$ Then the space $Z$ is the GIT quotient of $\cc^{2n}$ by this action.

1729: \end {lemma}

1730: 

1731: \noindent \textbf {Proof. } Write $\cc^{2n}$ as the space of pairs $(v, w),$ with $v, w \in 

1732: \cc^n,$ and the $\cc^*$ action given by 

1733: \begin {equation} 

1734: \label {cstar}

1735: (v, w) \to (\zeta v, \zeta^{-1}w)

1736: \end {equation}

1737:  for $\zeta \in 

1738: \cc^*.$ All orbits corresponding to $v, w \neq 0$ are stable. There is one semistable orbit 

1739: given by $v=w=0,$ and two unstable orbits corresponding to only one of $v$ and $w$ being 

1740: zero.

1741: 

1742: We define an algebraic map

1743: $$ f: \cc^{2n} \to Z, \ f(v,w) = v^Tw - \frac{1}{n}(v\cdot w)I.$$

1744: 

1745: This is $\cc^*$-invariant and therefore factors through the GIT quotient $\cc^{2n}/\cc^*.$ It 

1746: is easy to see that the induced map $\cc^{2n}/\cc^* \to Z$ is bijective. $\hfill \fin$

1747: 

1748: \medskip

1749: 

1750: Note that $Z$ has an isolated singularity at the origin, and the map $\chin$ is a 

1751: submersion everywhere except over zero. Given a K\"ahler metric $\Omega$ on $Z - \{0\}$, we 

1752: can study parallel transport with respect to $\chin.$ For $t \in \cc^*,$ we take 

1753: the path $\gamma: [0,1] \to \cc, \ \gamma(s) = (1-s)t$ as in the previous section, and define 

1754: \begin {equation}

1755: \label {vcps}

1756:  L_t = \{ y \in Z_t

1757: \ | \ h_{\gamma|[0,s]} \text{ is defined near }y \text{ for all } s <1, \ 

1758: h_{\gamma|[0,s]}(y) \to 0 \text{ as } s \to 1 \}. 

1759: \end {equation}

1760: 

1761: Before discussing the general case, let us work with a specific example. Take $\Omega = 

1762: \Omega_{st}$ to be the restriction of the standard form on $\cc^{n^2}$ under the inclusion $Z 

1763: \hookrightarrow \gl(n, \cc) \cong \cc^{n^2}.$ For each $t \in \cc^*,$ consider the diagonal 

1764: matrix $E_t=diag(t,t,\dots,t, (1-n)t) \in Z_t,$ and let

1765: \begin {equation}

1766: \label {ua}

1767: \ua_t = \{ UE_tU^{-1} \ | \ U \in U(n) \}.

1768: \end {equation}

1769: 

1770: Note that $\ua_t$ is diffeomorphic to $U(n)/(U(n-1) \times U(1)) \cong \cpn.$

1771: 

1772: \begin {lemma}

1773: \label {ost}

1774: The vanishing space $L_t \subset Z_t$ for $\Omega_{st}$ contains $\ua_t.$

1775: \end {lemma}

1776: 

1777: \noindent \textbf {Proof. } We claim that parallel transport along $\gamma|[0,s]$ takes 

1778: $UE_tU^{-1}$ into $UE_{t(1-s)}U^{-1}$ for $s<1$ and any $U \in U(n).$ From this it will follow 

1779: that in the limit $s \to 1,$ the space $\ua_t$ is indeed sent to the origin.

1780: 

1781: It suffices to verify our claim infinitesimally, i.e. to show that the parallel transport

1782: vector field at $UE_tU^{-1}$ is proportional to $UE_1U^{-1}.$ Since the K\"ahler metric is 

1783: $U(n)$-invariant, it also suffices to check this at $U=I.$ There the statement is that $E_1$ 

1784: is perpendicular to the tangent space to $Z_t$ at $E_t.$ The tangent space $T_{E_t}Z_t$ 

1785: consists of the commutators $[E_t, B]$ for $B \in \gl(n,\cc).$ A simple computation shows that 

1786: $[E_t, B]$ is always perpendicular to $E_1$ in $\cc^{n^2}. \hfill \fin$

1787: 

1788: \medskip

1789: 

1790: We now turn our attention to a more general class of K\"ahler metrics:

1791: 

1792: \begin {definition}

1793: \label {eff}

1794: Let $i$ be the standard inclusion of $Z=\slan$ in $\sl(n, \cc).$ A K\"ahler 

1795: form 

1796: $\Omega$ on $Z - \{0\}$ is called effective if it is of the form $i^*\omega,$ where $\omega$

1797: is a K\"ahler form on $\sl(n, \cc)$ that is real analytic in a neighborhood of zero.  

1798: \end {definition}

1799: 

1800: \begin {lemma}

1801: \label {commens}

1802: Let $\Omega$ be an effective K\"ahler form on $Z - \{0\},$ and $\Omega_{st}$ the standard 

1803: form from Lemma~\ref{ost}. Then the corresponding K\"ahler metrics $g$ and $g_{st}$ are 

1804: equivalent in 

1805: the following sense: given a relatively compact open set $V \subset Z,$ there exists a 

1806: constant $c > 0$ such that $c^{-1}\cdot g_{st} \leq g \leq c\cdot g_{st}$ on $V - \{0\}.$ 

1807: \end {lemma}

1808: 

1809: \noindent \textbf {Proof. } This is a direct consequence of the fact that both metrics are 

1810: obtained by pull-back from the nonsingular space $\sl(n, \cc). \hfill \fin$

1811: 

1812: \begin {lemma}

1813: \label {exist}

1814: Let $\Omega$ be an effective K\"ahler form on $Z - \{0\},$ and define $L_t$ as in 

1815: (\ref{vcps}). Then for any sufficiently small $t \in \cc^*, \ L_t$ is a smooth manifold 

1816: diffeomorphic to $\cpn.$

1817: \end {lemma}

1818: 

1819: \noindent \textbf {Proof. } Let $f: Y=\cc^{2n} \to Z$ be the GIT quotient map from 

1820: Lemma~\ref{git}. Observe that $f^*\Omega$ is nondegenerate on the union of all stable 

1821: $\cc^*$-orbits, except in the directions of those orbits. Consider the Hamiltonian $H: 

1822: \cc^{2n} \to \rr, \ H(v,w) = |v|^2-|w|^2$ associated to the action of $S^1 \subset \cc^*.$ The 

1823: restriction of the map $|H|^2: \cc^{2n} \to \cc$ to any nonzero (stable or unstable) 

1824: $\cc^*$-orbit is strictly plurisubharmonic. Let $g_H$ be the symmetric two-tensor 

1825: associated to $-dd^c|H|^2$ by the complex structure. An explicit calculation shows that 

1826: for any neighborhood $N$ of zero in $\cc^n,$ we can find some small $\epsilon > 0$ such that

1827: $f^*g_{st} + \epsilon g_H > 0$ on $N-\{0\}.$ Lemma~\ref{commens} implies that the same must be 

1828: true for $f^*g$ instead of $f^*g_{st}$ (with possibly a different $\epsilon.$) It follows that

1829: $$ \tilde \Omega = f^*\Omega - \epsilon dd^c |H|^2$$ 

1830: is K\"ahler on $N - \{0\}.$ 

1831: 

1832: We can now apply Proposition~\ref{vansing} to our situation; see also Remark~\ref{multsc}. 

1833: (The fact that $\tilde \Omega$ is not K\"ahler everywhere on $\cc^{2n}$ does not make a 

1834: difference.) In place of $p: \cc^{2n} \to \cc$ we take the map $(v,w) \to (v \cdot w)/n$ which 

1835: is the composition of $f$ with $\chin.$ This is the same as the map $p$ from 

1836: Section~\ref{sec:vsm}, up to a linear change of coordinates.

1837: 

1838: We obtain vanishing cycles $C_t \subset Y_t$ diffeomorphic to $S^{2n-1},$ for any $t \in 

1839: \cc^*$ small. Note that $\tilde \Omega$ is $S^1$-equivariant, hence the same is true 

1840: for parallel transport in $Y=\cc^{2n}.$ If we think of $Z$ as the quotient 

1841: $H^{-1}(0)/S^1,$ then parallel transport along $\gamma$ in $(Y, \tilde \Omega)$ corresponds to 

1842: parallel transport in $(Z, \Omega).$ The vanishing cycle $C_t$ is preserved by the $S^1$ 

1843: action, and its quotient is $L_t.$ Since the action is free outside zero, we find that $L_t$ 

1844: must be a smooth, connected $(2n-2)$-dimensional manifold. 

1845: 

1846: In particular, in the case when $\Omega$ is the standard $\Omega_{st}$ used in 

1847: Lemma~\ref{ost}, we find that $\ua_t \cong \cpn$ is the whole of $L_t.$ To see that 

1848: $L_t$ is diffeomorphic to $\cpn$ for a general $\Omega,$ interpolate between $\Omega$ and 

1849: $\Omega_{st}$ by a family of effective forms, and use the fact that all manifolds in a 

1850: smooth family are diffeomorphic.  $\hfill \fin$

1851: 

1852: \medskip

1853: 

1854: We call $L_t$ a {\it vanishing projective space} in $\slan.$ As seen in the 

1855: proof of Lemma~\ref{exist}, $L_t$ is the quotient of an ordinary vanishing cycle 

1856: $S^{2n-1}$ by an $S^1$ action. It is also a Lagrangian submanifold of $Z_t.$

1857: 

1858: \subsection {Relative vanishing projective spaces.} \label {sec:relvan}

1859: 

1860: Let $X$ be a smooth complex manifold. With an eye towards Lemma~\ref{thickthin}, we 

1861: look at the product $Z= X \times \slan.$ A similar discussion to that in the 

1862: previous section applies here, using Proposition~\ref{relvs}. Let us identify 

1863: $X$ with $X \times \{0\}.$ Consider the map $$ 

1864: \pi: Z \to \cc, \ (x, A) \to \chin(A),$$ and denote by $Z_t$ its fibers. Let $i$ be 

1865: the standard inclusion of $Z$ into $X \times \sl(n,\cc).$ By analogy with Definition~\ref{eff}, we 

1866: say that a K\"ahler form $\Omega$ on $Z - X = X \times (\slan - \{ 0\})$ is {\it 

1867: effective} if it is obtained by pull-back from a form on $X \times \sl(n, \cc)$ that is real 

1868: analytic around $X \times \{ 0 \}.$ 

1869: 

1870: For the remaining of this section we will assume that $Z-X$ is endowed with an effective 

1871: K\"ahler form $\Omega.$ Given a compact Lagrangian $K \subset X,$ we obtain a {\it relative 

1872: vanishing projective space} $L_t$ in $Z_t$ (for small $t$), diffeomorphic to $K \times \cpn.$ 

1873: 

1874: We denote by $g$ be the K\"ahler metric corresponding to $\Omega,$ and by $g_{st}$ the 

1875: product of an arbitrary metric on $X$ with the pull-back of standard metric on $\sl(n,\cc).$ 

1876: The analogue of Lemma~\ref{commens} holds true, so that $g$ and $g_{st}$ are equivalent on 

1877: the complement of $X$ in any relatively compact subset of $Z.$

1878: 

1879: Fix a relatively compact open set $W \subset X$ and a ball $B$ in $\slan$ around 

1880: the origin (in the standard pulled-back metric, say). We assume that the form $\Omega$ is real 

1881: analytic on $V=W \times B \subset Z.$ We are interested in estimating the parallel transport 

1882: maps in $V$ obtained from the map $\pi$ and the form $\Omega.$

1883: 

1884: \begin {lemma}

1885: \label {piano}

1886: The quantity $ \| \nabla \pi \| $ is bounded below on $V - X.$

1887: \end {lemma}

1888: 

1889: \noindent \textbf {Proof. } Since $g$ and $g_{st}$ are equivalent, it suffices to establish 

1890: the result when the gradient is taken with respect to $g_{st}.$ In that case it turns out 

1891: that $ \| \nabla \pi \| $ is bounded below globally. Indeed, recall from (\ref{trivia}) 

1892: that every matrix $A \in \slan$ is of the form $S + tUE_1U^{-1},$ where $S$ 

1893: is a matrix of rank at most one, $U$ is unitary, $E_1$ is the diagonal matrix $diag(1, 

1894: \dots, 1, 1-n)$, and $t \in \cc.$ At a point $(x,A) \in Z$ we have $\pi(x,A) = t.$ Fix a 

1895: representation of $A$ as $S+tUE_1U^{-1},$ and consider the one-dimensional complex subspace 

1896: $\{x\} \times \{ S+\tau UE_1U^{-1} \ | \ \tau \in \cc \} \subset Z.$ The gradient of the 

1897: restriction of $\pi$ to this subspace is $UE_1U^{-1}.$ This always has norm 

1898: $(n^2-n)^{1/2},$ because the metric is $U(n)$-invariant. It follows that the actual 

1899: gradient of $\pi$ on $Z$ (with respect to $g_{st}$) has norm at least $(n^2-n)^{1/2}. 

1900: \hfill \fin$

1901: 

1902: \medskip

1903: 

1904: Now let $K \subset U \subset X$ be a compact Lagrangian submanifold, and denote by $\delta > 

1905: 0$ its distance from $\partial V$ in the metric $g.$ Let also $\nu$ be the lower bound on $\| 

1906: \nabla \pi \|$ in $V$ which is given by the previous lemma.

1907: 

1908: \begin {lemma}

1909: \label {wdw}

1910: For any $t$ with $0 < |t| < \nu \delta/2,$ the relative vanishing cycle $L_t$ is well-defined 

1911: and lies in $V.$ 

1912: \end {lemma}

1913: 

1914: \noindent \textbf {Proof. } Without loss of generality we can assume that $t$ is real and 

1915: positive. Think in terms of parallel transport along $\gamma(s)=s.$ By the previous lemma, the 

1916: the horizontal vector field $H_{\gamma} = \nabla \pi/\| \nabla \pi \|^2$ satisfies

1917: \begin {equation}

1918: \label {hga}

1919: \| H_{\gamma} \| \leq \nu^{-1}.

1920: \end {equation}

1921: Say we are given a flow line of $H_{\gamma}$ defined for $s \in (0,t)$ and 

1922: converging to a point in $K$ as $s \to 0.$ Then by integrating (\ref{hga}) we obtain that the 

1923: whole flow line lies at distance at most $\nu^{-1}t < \delta/2$ from $K,$ hence it extends to 

1924: $s=t.$ The desired claim follows. $\hfill \fin$

1925: 

1926: \medskip

1927: 

1928: Consider now the circle $\gamma: [0,2\pi] \to \cc^*, \ \gamma_t(s) = t \ exp(is).$ A 

1929: similar argument to that in the proof of Lemma~\ref{wdw} shows that for $0 < |t| < \nu 

1930: \delta/2,$ parallel transport $h_{\gamma_t}(y)$ is well-defined and lies in $V$ for any $y 

1931: \in L_t.$ Furthermore, we have

1932: 

1933: \begin {lemma}

1934: The relative vanishing cycle $L_t$ is Lagrangian isotopic to $h_{\gamma_t}(L_t)$ inside $V 

1935: \cap Z_t.$

1936: \end {lemma}

1937: 

1938: \noindent \textbf {Proof. } By our estimates, the vanishing cycle $L_{\tau}$ is well-defined 

1939: for any $\tau = \gamma_t(s),$ and it is moved by parallel transport along $\gamma_t|_{[s, 

1940: 2\pi]}$ without going outside $V.$ The required isotopy is given by $h_{\gamma_t|{[s,

1941: 2\pi]}}(L_{\tau}). \hfill \fin$

1942: 

1943: \subsection {Iterating the relative $\cpn$ construction.}

1944: Now we deal with the situation relevant for Lemma~\ref{multithickthin}. Take $Z = X \times 

1945: \bigl ( \slan \bigr )^k,$ with the map

1946: $$ \pi: Z \to \cc^k, \ \pi(x, A_1, \dots, A_k) = \bigl (\chin(A_1), \dots, 

1947: \chin(A_k) \bigr ), $$

1948: and an effective K\"ahler form (defined just like in the previous section).

1949: 

1950: Starting with a compact Lagrangian submanifold $K$ of $X= X \times \{0\}^k \subset Z,$ one can 

1951: use the first component of $\pi$ to construct a relative vanishing projective space $L_{t_1}$ 

1952: in $ X \times \slan \times \{0\}^{k-1},$ for $t_1 \in \cc^*$ small. Using the 

1953: second component next, and then iterating the process produces a Lagrangian submanifold 

1954: $L_{t_1, \dots, t_k} \subset \pi^{-1}(t_1, \cdot, t_k)$ diffeomorphic to $X \times (\cpn)^k.$ 

1955: Parallel transport estimates show that this is well-defined whenever $0 \leq |t_j| \leq 

1956: \sigma$ for all $1\leq j \leq k,$ where $\sigma$ is a bound independent of $j.$

1957: 

1958: \begin {lemma}

1959: \label {changeorder}

1960: Changing the order in which the components of $\pi$ are used to construct $L_{t_1, \dots, 

1961: t_k}$ produces the same outcome up to Lagrangian isotopy, at least as long as all the $|t_j|$ 

1962: are sufficiently small.

1963: \end {lemma}

1964: 

1965: \noindent \textbf {Proof. } The statement is trivial when the K\"ahler form is the product of 

1966: a form on $X$ with the standard forms on each $\slan.$ The case of a general 

1967: effective K\"ahler form follows from this by a Moser Lemma argument similar to the one in 

1968: Lemma 30 in \cite{SS}. $\hfill \fin$

1969: 

1970: 

1971: \section {An analogue of the fibered $A_2$ singularity} \label {sec:fiba2}

1972: 

1973: Our goal for this section is to understand the behavior of vanishing projective spaces 

1974: under parallel transport in the situation of Section~\ref{sec:ttt}. The case $n=2$ is a 

1975: fibered $A_2$ singularity and was treated in \cite[Section 4(C)]{SS}. The general case 

1976: discussed here runs parallel to that, but the geometry is somewhat more complicated.

1977: 

1978: \subsection{The non-fibered case} \label{sec:non} We start with analyzing the situation 

1979: in Section~\ref{sec:exa}. There we had the space $\xx_n \subset \sl(n+1, \cc)$ 

1980: consisting of the matrices of the form (\ref{sjmm}) which admit an $(n-1)$-dimensional 

1981: eigenspace. A matrix $A \in \sl(n+1, \cc)$ has a ``thick'' eignevalue $d$ of multiplicity at 

1982: least $n-1,$ and two other ``thin'' eigenvalues which we write as $d+z_1$ and $d+z_2.$ (Of 

1983: course, there can be coincidences giving higher multiplicities.) The zero trace condition 

1984: implies that $z_1 + z_2 = -(n+1)d.$ We also have the map (\ref{mapp}) given by

1985: $$ \mapp: \xx_n \to \sym^2(\cc) \cong \cc^2.$$

1986: 

1987: As coordinates on $\cc^2$ we take $d$ and the product $z = z_1z_2,$ so that $\mapp(A) = 

1988: (d,z).$ By Proposition~\ref{luna} the map $\mapp$ is a smooth fibration except over the 

1989: critical value sets corresponding to $z=0$ (a thick-thin coincidence) or $z= ((n+1)d/2)^2$ (a 

1990: thin-thin coincidence). 

1991: 

1992: We view $d$ as an auxiliary parameter. (Our choice of $d$ is somewhat different from the one 

1993: in \cite{SS}.) We denote the restriction of $\mapp$ to some $\xx_{n,d} = \mapp^{-1}(\{d \} 

1994: \times \cc)$ by $\mapp_d : \xx_{n,d} \to \{d \} \times \cc = \cc,$ and we write $\xx_{n,d,z}$ 

1995: for a fiber of $\mapp_d.$ For $d \neq 0,$ the map $\mapp_d$ has two critical values $0$ and 

1996: $\zeta_d = ((n+1)d/2)^2.$ These coalesce for $d =0.$

1997: 

1998: Let us make this picture more explicit. A matrix $A \in \xx_{n,d}$ is of the form

1999: \begin {equation}

2000: \label {sjmm2}

2001: dI +   \left( \begin {array}{ccccc} \alpha & 1 & 0 & \cdots & 0\\

2002: a_{11} + \alpha^2 & \alpha & a_{12} & \cdots & a_{1n}\\

2003: a_{21} & 0 & a_{22} & \cdots & a_{2n} \\

2004:  & \cdots & \  & \cdots  &  \ \\

2005: a_{n1} & 0 & a_{n2} & \cdots & a_{nn}

2006: \end {array} \right ).

2007: \end {equation}

2008: 

2009: (As compared to (\ref{sjmm}), we changed the meaning of the notation $a_{ij}.$) The fact that 

2010: $A - dI$ has rank at most two translates into the condition that the associated {\it reduced 

2011: matrix}

2012: \begin {equation}

2013: \label {sjmm3}

2014: A_{\red}=   \left( \begin {array}{cccc} 

2015: a_{11} & a_{12} & \cdots & a_{1n}\\

2016: a_{21} &  a_{22} & \cdots & a_{2n} \\

2017:  \cdots & \  & \cdots  &  \ \\

2018: a_{n1}  & a_{n2} & \cdots & a_{nn}

2019: \end {array} \right )

2020: \end {equation}

2021: has rank at most one. Note that 

2022: \begin {equation}

2023: \label {alfa}

2024: 2\alpha + a_{22} + \cdots + a_{nn} = -(n+1)d. 

2025: \end {equation}

2026: 

2027: If we denote $s=a_{22} + \cdots + a_{nn},$ an evaluation of the sum of $2$-by-$2$ minors of $A$ 

2028: gives $z=-a_{11}+2\alpha s,$ or

2029: \begin {equation}

2030: \label {betha}

2031: z = -a_{11} -(n+1)d\cdot s - s^2.

2032: \end {equation}

2033: 

2034: Thus $\xx_{n,d}$ can be identified with the space of matrices $A_{\red}$ of rank at most one,

2035: and the map $\mapp_d: \xx_{n,d} \to \cc$ is given by 

2036: \begin {equation}

2037: \label {rede}

2038: \mapp_d(A_{\red}) = -a_{11} - (n+1)d\sum_{k \geq 2} a_{kk} - \bigl ( \sum_{k \geq 2}  

2039: a_{kk} \bigr )^2.

2040: \end {equation}

2041: 

2042: Over the critical value $z=0$ we have the element in $\xx_{n,d}$ corresponding to $A_{\red} 

2043: =0.$ 

2044: 

2045: \begin {lemma}

2046: \label {biholo}

2047: There exists a biholomorphism $f$ between a neighborhood of $A_{\red}=0$ in $\xx_{n,d}$ and a 

2048: neighborhood of zero in the space $\slan$ from Section~\ref{sec:mone}, having 

2049: the property that 

2050: \begin {equation}

2051: \label {propr}

2052: \chin \circ  f = \mapp_d.

2053: \end {equation}

2054: 

2055: This biholomorphism extends to one 

2056: between neighborhoods of zero in $\gl(n, \cc),$ in the sense that there is a commutative 

2057: diagram

2058: $$ \begin {CD}

2059: \xx_{n,d} @>{\text{local } \cong \ \text{near} \ 0}>> \slan \\

2060: @VVV   @VVV \\

2061: \gl(n, \cc) @>{\text{local } \cong \ \text{near} \ 0}>> \gl(n, \cc),

2062: \end {CD} $$

2063: where the vertical maps are the natural inclusions.

2064: \end {lemma}

2065: 

2066: \noindent \textbf {Proof. } The spaces $\xx_{n,d}$ and $\slan$ are clearly 

2067: biholomorphic: one can subtract a suitable multiple of the identity from a reduced matrix 

2068: in $\xx_{n,d}$ and make it traceless. However, this obvious biholomorphism does not satisfy 

2069: the required condition (\ref{propr}). What we need is to find a biholomorphic change of 

2070: coordinates near zero 

2071: in $\xx_{n,d}$ that takes the expression 

2072: \begin {equation}

2073: \label {eke}

2074:  -\frac{1}{n} (a_{11} + \cdots + a_{nn})

2075: \end {equation}

2076: into the right-hand side of (\ref{rede}). 

2077: 

2078: It is helpful to work with a different set of coordinates on $\xx_{n,d}.$ By Lemma~\ref{git}, 

2079: the variety $\slan$ and hence $\xx_{n,d}$ can be identified with a GIT 

2080: quotient $\cc^{2n}/\cc^*.$ We denote by $v_i, w_j \ (i, j = 1, \dots, n)$ the coordinates on 

2081: $\cc^{2n},$ so that the entries $a_{ij}$ of a reduced matrix $A_{\red}$ (for an element of  

2082: $\xx_{n,d}$) are $a_{ij} = v_i w_j.$ Consider the holomorphic map from $\cc^{2n}$ to itself 

2083: which keeps all $v_i$ fixed, takes $w_1$ into $nw_1,$ and

2084: $$ w_j \to n(n+1) d\cdot w_j + nw_j(v_2w_2 + \dots + v_nw_n) \ \text{for } j=2, \dots, n.$$

2085: 

2086: The differential of this map at zero is invertible, hence this is a biholomorphism between 

2087: neighborhoods of zero in $\cc^{2n}.$ Since it is $\cc^*$-invariant with respect to 

2088: (\ref{cstar}), it induces a local biholomorphism between the GIT quotients. Written in terms 

2089: of $a_{ij},$ this takes (\ref{eke}) into (\ref{rede}), as desired. 

2090: 

2091: The same formulae in terms of $a_{ij}$ also give the required extension to a local 

2092: biholomorphism in $\gl(n, \cc). \hfill \fin$ 

2093: 

2094: \begin {remark} \label {remc} More canonically, suppose that instead of $\xx_n$ we have a 

2095: space $\xx(T) \subset \sl(E)$ constructed from a linear quadruple $T$ as in 

2096: Remark~\ref{act}. A reduced matrix represents a linear map from $F_1 \oplus T_1$ to $T_2 

2097: \oplus T_1.$ Using the isomorphism $T_2 \to F_1$ from the definition of a linear quadruple, 

2098: we can think of reduced matrices as automorphisms of $F_2 = F_1 \oplus T_1.$ Hence,   

2099: Lemma~\ref{biholo} provides a local biholomorphism between neighborhoods of zero in 

2100: $\xx(T)_d$ and $\sl(F_2)^{(1(n-1))},$ respectively. This extends to a local 

2101: self-biholomorphsim of $\gl(E).$ \end {remark}

2102: 

2103: \subsection {Effective K\"ahler forms}

2104: \label {sec:ek}

2105: 

2106: By analogy with Definition~\ref{eff} we make the following

2107: 

2108: \begin {definition}

2109: \label {efftwo}

2110: Denote by $\Reg \xx_n$ the open dense set of smooth points of the variety $\xx_n,$ and by $i$  

2111: the standard inclusion of $\xx_n$ in $\sl(n+1, \cc).$ A K\"ahler form

2112: $\Omega$ on $\Reg \xx_n$ is called effective if it is of the form $i^*\omega,$ where 

2113: $\omega$ is a K\"ahler form on $\sl(n+1, \cc)$ that is real analytic in a neighborhood of 

2114: zero.

2115: \end {definition}

2116: 

2117: If we denote by $(b_{ij}), \ i,j=1, \dots, n+1$ the entries of a matrix $B \in \sl(n+1, 

2118: \cc),$ we call $$ -dd^c \Bigl ( |b_{11} - b_{22}|^2 + \sum_{(i,j) \neq (1,1), (2,2)} 

2119: |b_{ij}|^2 \Bigr ) $$ the {\it standard K\"ahler form} on $\sl(n+1, \cc).$

2120: Its pullback to $\xx_{n,d}$ is called the standard K\"ahler form on that space, and denoted 

2121: $\Omega_{st}.$ (Strictly speaking, because of the zero trace condition, there is no 

2122: ``standard'' choice of basis for $\sl(n+1, \cc).$ Our choice is the most natural here 

2123: because we deal with matrices of the form \ref{sjmm2}.)

2124: 

2125: Fix a small (relatively compact) neighborhood $V$ of zero in $\xx_n.$ The analogue of 

2126: Lemma~\ref{commens} still holds. If we restrict $\Omega_{st}$ and any other effective 

2127: K\"ahler form on $\Reg \xx_{n,d}$ to the regular set $\Reg V$, then the corresponding 

2128: metrics $g$ and $g_{st}$ are equivalent.

2129: 

2130: For small $d \in \cc, $ we can define two natural K\"ahler forms on $\Reg \xx_{n,d}.$ One is 

2131: the restriction of $\Omega_{st},$ which we call the {\it first standard K\"ahler form} and 

2132: denote by $\Omega_{st(1)}.$ The other, denoted $\Omega_{st(2)}$ and called the {\it second 

2133: standard K\"ahler form}, is the pullback of the usual K\"ahler form on $\cc^{n^2}$ under the 

2134: inclusion of $\xx_{n,d}$ (viewed as the set of reduced matrices) into $\gl(n, \cc) = 

2135: \cc^{n^2}.$

2136: 

2137: The only difference between $\Omega_{st(1)}$ and $\Omega_{st(2)}$ appears because of the 

2138: entry $a_{11} + \alpha^2$ in (\ref{sjmm2}). Using (\ref{alfa}) we see that this quadratic 

2139: discrepancy is negligible for $V$ and $d$ sufficiently small. More precisely, the first and 

2140: second standard metrics $g_{st(1)}$ and $g_{st(2)}$ induced by $\Omega_{st(1)}$ and 

2141: $\Omega_{st(2)}$ on $\xx_{n,d}$ are equivalent on $\Reg (V \cap \xx_{n,d}).$ Moreover, the 

2142: constants appearing in the definition of equivalence for metrics can be taken to be 

2143: independent of $d,$ for $d$ sufficinetly small. Note that from here it follows that the 

2144: same statements are also true for $g_{st(2)}$ and any metric induced by the restriction of 

2145: an effective K\"ahler form on $\Reg \xx_n.$

2146: 

2147: \subsection {Estimating parallel transport} \label {sec:seces}

2148: 

2149: We work on the same open set $V \subset \xx_n$ as in the previous subsection. Endow 

2150: $\Reg \xx_n$ with an effective K\"ahler form $\Omega$ in the sense of 

2151: Definition~\ref{efftwo}. It is easy to see that the restrictions of $\Omega$ to $\Reg 

2152: \xx_{n,d} \iso \slan - \{ 0 \}$ are effective in the sense of Definition~\ref{eff}. 

2153: 

2154: Using Lemmas \ref{biholo} and \ref{exist} we can find a natural Lagrangian vanishing 

2155: projective space (with respect to the map $\mapp_d$)

2156: $$ L_{d,z} \cong \cpn \subset \xx_{n,d,z},$$ for any $z$ with $ |z| \ll |d|.$ If $z$ and $d$ 

2157: are small enough this lies inside $V.$

2158: 

2159: For $d, z > 0$ real ($z \ll d$), consider the path $\gamma_{d,z}$ in $\cc - \{0, \zeta_d 

2160: \}$ going from $z$ to $\zeta_d/2$ on the real axis, then making a positive full circle 

2161: around $\zeta_d,$ and going back to $z$ along the real axis. (See Figure~\ref{fig:path}.)

2162: Parallel transport along this path basically corresponds to swapping the order of the two 

2163: thin eigenvalues $d+z_1$ and $d+z_2.$

2164: 

2165: \begin {figure}

2166: \begin {center}

2167: \input {path.pstex_t}

2168: \end {center}

2169: \caption {Swapping the thin eigenvalues.}

2170: \label {fig:path}

2171: \end {figure}

2172: 

2173: The image of $L_t$ under parallel transport along $\gamma_{d,z}$ is another Lagrangian 

2174: projective space

2175: \begin {equation}

2176: \label {imago}

2177:  h_{\gamma_{d,z}}(L_{d,z}) \subset \xx_{n,d,z}.

2178: \end {equation}

2179: 

2180: In order to show that this is well-defined, we need to estimate the size 

2181: of the parallel transport vector fields.

2182: 

2183: \begin {lemma}

2184: \label {maines}

2185: On each $\Reg (V \cap \xx_{n,d})$ we have

2186: \begin {equation}

2187: \label {deq}

2188:  \| \nabla \mapp_d \|^2 \geq \nu \cdot |\mapp_d - \zeta_d|,

2189: \end {equation}

2190: where $\nu > 0$ is a constant that depends only on $n, V,$ and the K\"ahler form $\Omega.$

2191: \end {lemma}

2192: 

2193: \noindent \textbf{Proof. } By the observation at the end of Section~\ref{sec:ek}, it suffices 

2194: to prove the required bound using the second standard K\"ahler form on each $\Reg \xx_{n,d}.$ 

2195: 

2196: If we think of $\xx_{n,d}$ as the space of reduced matrices, we have an action of $U(n-1)$ on 

2197: this space given by conjugation with the block matrix

2198: 

2199: $$ \left( \begin {tabular}{c|c}

2200: $1$  &  $\begin {array}{ccc}

2201: 0 & \cdots & 0 \\

2202: \end {array}$

2203: 

2204:  \\

2205: \hline

2206: 

2207: $\begin {array}{c}

2208: 0 \\

2209: \cdots \\

2210: 0 \\

2211: \end {array}$ 

2212:   &

2213: 

2214: $ \begin {array}{ccc} 

2215: & & \\

2216: & U & \\

2217: & & \\

2218: \end {array} $

2219: 

2220: \end {tabular} \right)

2221: $$

2222: for $U \in U(n-1).$ By using this action we can transform every matrix

2223: $A_{\red}= (a_{ij})$ of rank at most one into one satisfying $a_{ij}=0$ for all $i \geq 3, 

2224: j

2225: \geq 2.$ Thus it suffices to prove the inequality (\ref{deq}) for matrices of this form, 

2226: because both the K\"ahler metric $g_{st(2)}$ and the expression (\ref{rede}) are 

2227: $U(n-1)$-invariant.

2228: 

2229: Since all the $2$-by-$2$ minors of $A_{\red}$ are zero, we can in fact assume that $a_{ij} 

2230: = 0$ 

2231: for $i \geq 3$ and any $j.$ (Indeed, if $a_{i1} =0$ for some $i \geq 3,$ then all columns 

2232: except the first are zero, and by transposing we are back in the same case.)

2233: 

2234: Just as in the proof of Lemma~\ref{piano}, it is now helpful to restrict $\mapp_d$ to various 

2235: subspaces where the size of its gradient is easier to estimate. The first restriction is to 

2236: the space of reduced matrices with the property mentioned in the previous paragraph: 

2237: \begin {equation}

2238: \label {aij}

2239: a_{ij} = 0 \ \text{ for } \ i \geq 3 \ \text{ and any  }  j. 

2240: \end {equation}

2241: 

2242: Inside of this space we have the open dense subset $W$ consisting of matrices of the form

2243: $$ \left( \begin {array}{cccc}

2244: uw_1 & uw_2 & \cdots & uw_n\\

2245: w_1 & w_2 & \cdots & w_n \\

2246: 0 & 0 & \cdots & 0 \\

2247: & \cdots & \  & \cdots  \\

2248: 0 & 0 & \cdots & 0

2249: \end {array} \right ), $$

2250: with $u, w_j \in \cc.$ The only matrices that satisfy (\ref{aij}) and are not in $W$ have 

2251: only zeros in the second row, but not so in the first. Note that if we establish 

2252: (\ref{deq}) for 

2253: matrices in the dense set $W,$ by taking limits the same inequality must be true for the 

2254: remaining matrices too.

2255: 

2256: Therefore we just focus our attention on the set $W.$ In terms of the coordinates $u, w_j,$ 

2257: the restriction of $q_d$ to $W \cong \cc^{n+1}$ is given by

2258: $$ p_d = q_d|_W : W \to \cc, \ \ (u, w_j) \to  -uw_1 - (n+1)d\cdot w_2 - w_2^2.  $$

2259: 

2260: Since $\| \nabla p_d \| \leq  \| \nabla q_d \|,$ the inequality (\ref{deq}) would follow if we 

2261: were able to prove

2262: \begin {equation}

2263: \label {what}

2264: \| \nabla p_d \|^2 \geq \nu \cdot |p_d - \zeta_d|.

2265: \end {equation}

2266: 

2267: The restriction of the second standard K\"ahler form to $W$ is

2268: $$ -dd^c \Bigl ( (|u|^2 + 1) (|w_1|^2 + \dots + |w_n|^2) \Bigr ). $$

2269: 

2270: Let us consider the restriction of $p_d$ to the subspace $W_1 \cong \cc \subset W$ which 

2271: corresponds to keeping all coordinates except $w_1$ fixed. We get

2272: \begin {equation}

2273: \label {in1}

2274: \| \nabla p_d \|^2 \geq \| \nabla p_d|_{W_1} \|^2 = \frac{|u|^2}{|u|^2 + 1}.

2275: \end {equation}

2276: 

2277: Doing the same thing with for the $w_2$-coordinate we obtain

2278: \begin {equation}

2279: \label {in2}

2280: \| \nabla p_d \|^2 \geq  \frac{|(n+1)d + 2w_2|^2}{|u|^2 + 1}.

2281: \end {equation}

2282:  

2283: Finally, using the $u$-coordinate:

2284: \begin {equation}

2285: \label {in3}

2286: \| \nabla p_d \|^2 \geq \frac{|w_1|^2}{|w_1|^2 + \dots + |w_n|^2}.

2287: \end {equation}

2288: 

2289: The matrices for which we seek to prove (\ref{what}) all lie inside the relatively compact 

2290: set $V,$ which means that there is an a priori upper bound $R$ on the absolute values of all 

2291: $w_j$'s. There is no such bound on $u.$ Nevertheless, if $|u| > 1,$ then (\ref{in1}) would 

2292: automatically imply $\| \nabla p_d \| \geq 1/2,$ and then (\ref{what}) would follow from the a 

2293: priori bound on its right-hand side. Thus we can assume $|u| \geq 1.$ Putting (\ref{in1}) 

2294: and (\ref{in3}) together and using these bounds we obtain

2295: $$ \| \nabla p_d \|^2 \geq \max \bigl( |u|^2/2, |w_1|^2/(nR^2) \bigr ) \geq (2n)^{-1/2}R^{-1} 

2296: |uw_1|. $$

2297: 

2298: Combining this with (\ref{in2}) gives

2299: \begin {eqnarray*}

2300:  \| \nabla p_d \|^2 & \geq & \max \bigl((2n)^{-1/2}R^{-1} |uw_1|, 2|(n+1)d/2 + w_2|^2) \\

2301: & \geq & \nu\cdot |uw_1 + ((n+1)d/2+w_2)^2| = \nu \cdot | p_d - \zeta_d |, 

2302: \end {eqnarray*} 

2303: as desired. $\hfill \fin$

2304: 

2305: \begin {lemma}

2306: For $0 < z \ll d$ small, parallel transport along $\gamma_{d,z}$ is well-defined near the 

2307: vanishing projective space $L_{d,z} \subset V,$ and the image (\ref{imago}) still lies in $V.$

2308: \end {lemma}

2309: 

2310: \noindent \textbf{Proof. } Note that the length of $\gamma_{d,z}$ is $(1+\pi)\zeta_d,$ and all 

2311: of its points are at distance at least $\zeta_d/2$ from $\zeta_d.$ Using Lemma~\ref{maines} 

2312: and (\ref{hgam}), we deduce that any flow line of $H_{\gamma_{d,z}}$ contained in $W$ must 

2313: satisfy $$ \int 

2314: \|H_{\gamma_{d,z}} \| = \int \|\nabla q_d \|^{-1} \leq (1+\pi)\zeta_d \cdot \nu^{-1/2} 

2315: (\zeta_d/2)^{-1/2} = C \cdot d, $$

2316: where $C$ is a constant depending only on $n,$ $V,$ and $\Omega.$ The conclusion follows just 

2317: as in the proof of Lemma~\ref{wdw}.

2318: $\hfill \fin$

2319: 

2320: \subsection {Projection to the $\alpha$-coordinate} 

2321: \label {sec:ac} Let us assume that $\Reg \xx_{n}$ is 

2322: endowed with its standard K\"ahler form, so that on each $\xx_{n,d}$ (and its subspaces) we 

2323: have the first standard K\"ahler form. In this situation, a more concrete picture of the 

2324: vanishing cycles can be obtaines by projecting $\xx_{n,d,z}$ to the $\alpha$-coordinate, where 

2325: $\alpha$ is given by the expression (\ref{alfa}) in terms of the $a_{ij}$'s.

2326: 

2327: Denote by $\xx_{n,d,z,\alpha}$ the subspace of $\xx_{n,d,z}$ corresponding to fixed $\alpha.$ 

2328: Using (\ref{alfa}), (\ref{betha}), and the representation of $\xx_{n,d}$ in terms of the 

2329: variables $v_i, w_j$ as in the proof of Lemma~\ref{biholo}, we get that $\xx_{n,d,z,\alpha}$ 

2330: is isomorphic to the variety

2331: $$ \mathscr{V}_{c_1, c_2} = \{ v_i, w_j \in \cc \ (i,j=1, \dots,n)\ | \ v_1w_1 = c_1, \ 

2332: v_2w_2 + \dots + v_n w_n = c_2 \}/\cc^*.$$

2333: Here the $\cc^*$-action is given by (\ref{cstar}), and

2334: \begin {eqnarray*}

2335: c_1 &=& -z-2(n+1)d \cdot \alpha - 4\alpha^2; \\

2336: c_2 & = & -(n+1)d - 2\alpha.

2337: \end {eqnarray*}

2338: 

2339: \begin {lemma}

2340: The variety $ \mathscr{V}_{c_1, c_2} $ is nonsingular if and only if $c_1 \neq 0$ and $c_2 

2341: \neq 0.$ 

2342: \end {lemma}

2343: 

2344: \noindent \textbf{Proof. } If $c_1=0,$ the variety has two irreducible components 

2345: corresponding to $v_1 = 0$ and $w_1=0,$ respectively, and the two components have a nontrivial 

2346: intersection.

2347: 

2348: If $c_1 \neq 0,$ we can use the $\cc^*$-action to fix $v_1 =c_1, \ w_1 =1.$ We get that $ 

2349: \mathscr{V}_{c_1, c_2}$ is the quadric given by the equation $v_2w_2 + \dots + v_n w_n = c_2.$

2350: This is singular if and only if $c_2 =0. \hfill \fin$

2351: 

2352: \medskip

2353: 

2354: Therefore, for fixed $n,d,z,$ there are typically three values of $\alpha$ for which 

2355: $\xx_{n,d, z,\alpha}$ is singular: $\alpha_1 = -(n+1)d/2,$ and the two roots $\alpha_2$ and 

2356: $\alpha_3$ of the equation $4\alpha^2 + 2(n+1)d\alpha + z=0.$ Note that for $0 < z \ll d,$ 

2357: all these three values are real and negative. The leftmost one is $\alpha_1,$ then one of 

2358: $\alpha_2$ and $\alpha_3$ (say $\alpha_2$) is slightly to the right of $\alpha_1,$ and 

2359: the remaining value $\alpha_3$ is close to zero. (See Figure ~\ref{fig:lag}.)

2360: 

2361: Let us consider the $U(n-1)$-action on $\xx_{n,d}$ that appeared in the beginning of 

2362: the proof of Lemma~\ref{maines}. The fibers $\xx_{n,d,z, \alpha}$ are preserved by this 

2363: action, which in terms of the coordinates $\vec{v} =(v_2, \dots, v_n), \vec{w} =(w_2, 

2364: \dots, w_n)$ is simply 

2365: \begin {equation}

2366: \label {un}

2367: U \in U(n-1) \ : \ (\vec{v}, \vec{w}) \to (U\vec{v}, U^{-1}\vec{w}).

2368: \end {equation}

2369: 

2370: This action is Hamiltonian with respect to $\Omega_{st(1)}.$ If $\mu: \xx_{n,d,z, \alpha} 

2371: = \mathscr{V}_{c_1, c_2} \to \mathfrak{u}(n-1)$ denotes the moment map, the set 

2372: $\mu^{-1}(0)$ is determined by the system of equations

2373: $$ v_i \bar{v}_j = \bar{w}_i w_j \ \ (i,j =2, \dots, n). $$

2374: 

2375: If these equations are satisfied, then first of all $|v_i| = |w_i|$ for all $i,$ and then 

2376: we know that $\bar{w}_i/v_i$ is independent of $i.$ This means that $w_i = \lambda \bar 

2377: v_i$ for some $\lambda \in S^1.$ The equation $v_2w_2 + \dots + v_n w_n=c_2$ becomes 

2378: $|v_2|^2 + \dots + |v_n|^2 = \lambda^{-1}c_2,$ and (unless $c_2=0$) the value of $\lambda$ 

2379: is determined by $c_2$ in such a way as to make $\lambda^{-1}c_2$ a positive real number.

2380: 

2381: In light of this, the intersection

2382: \begin {equation} \label {cndz}

2383: \mathscr{C}_{n,d,z,\alpha} = \xx_{n,d,z,\alpha} \cap \mu^{-1}(0) 

2384: \end {equation}

2385: is easily seen to be a $(2n-3)$-dimensional sphere if $c_1, c_2 \neq 0,$ a point if $c_2 = 

2386: 0,$ and a copy of $\mathbb{CP}^{n-2}$ if $c_1=0, c_2 \neq 0.$ 

2387: 

2388: \begin {remark}

2389: \label {klm}

2390: If we are in the situation from Remark~\ref{remc}, with $\xx(T)$ instead of 

2391: $\xx_n,$ then for $c_1, c_2 \neq 0,$ the space (\ref{cndz}) can be naturally identified 

2392: with a zero-centered sphere in $\text{Hom}(F_1, T_1)=\text{Hom}(F_1, F_2/F_1).$ Indeed, 

2393: if we use the 

2394: $\cc^*$-action to set $w_1=1,$ then the $v_j$'s ($j \geq 2$) are elements on the 

2395: first column of the reduced matrix $A_{\red}.$ This matrix represents a linear operator 

2396: in $\gl(F_2),$ written in a basis in which the first component is in $F_1.$ 

2397: \end {remark}

2398: 

2399: Given a path $\delta: [0,1] 

2400: \to \cc$ such that $\delta(0)=\alpha_1, \ \delta(1) \in \{\alpha_2, \alpha_3\}$ and 

2401: $\delta(t) \not \in \{\alpha_1, \alpha_2, \alpha_3\}$ for $t\in (0,1)$, we can construct a 

2402: Lagrangian $\cpn \subset \xx_{n,d,z}:$ 

2403: \begin {equation}

2404: \label {bigla}

2405: \Lambda_{\delta} = \bigcup_{t\in [0,1]} \mathscr{C}_{n,d,z, \delta(t)} 

2406: \end {equation}

2407: as the union of a point sitting over $\alpha_1,$ a family of $S^{2n-3}$'s sitting over 

2408: $\delta(t), t \in (0,1),$ and a $\mathbb{CP}^{n-2}$ sitting over the other endpoint 

2409: ($\alpha_2$ or $\alpha_3$).

2410: 

2411: Suppose now that $0 < z \ll d.$

2412: 

2413: \begin {lemma}

2414: \label {earth}

2415: The vanishing projective space $L_{d,z} \subset \xx_{n,d,z}$ with respect to 

2416: $\Omega_{st(1)}$ and its 

2417: monodromy image (\ref{imago}) are Lagrangian isotopic to the projective spaces 

2418: (\ref{bigla}) associated to the paths $\delta_1$ and $t_{\delta_2}(\delta_1)$ shown in 

2419: Figure~\ref{fig:lag}. (Here $t_{\delta_2}$ stands for a half-twist around $\delta_2.$)

2420: \end {lemma}

2421: 

2422: \begin {figure}

2423: \begin {center}

2424: \input {lag.pstex_t}

2425: \end {center}

2426: \caption {Projections of the Lagrangians.}

2427: \label {fig:lag}

2428: \end {figure}

2429: 

2430: \noindent \textbf{Proof. } This is completely similar to the proof of Lemma 32 in 

2431: \cite{SS}. A short computation shows that the parallel transport vector 

2432: fields for $\mapp_d: \xx_{n,d} \to \cc$ are invariant with respect to the $U(n-1)$-action 

2433: (\ref{un}), and $d\mu$ vanishes on them. The critical point corresponding to $A_{\red}=0$ 

2434: is a fixed point of the $U(n-1)$-action, and lies in $\mu^{-1}(0).$ It follows that the 

2435: vanishing cycles are $U(n-1)$-invariant and lie in $\mu^{-1}(0)$ as well. Any such 

2436: Lagrangian is of the form $\Lambda_{\delta}$ for some path $\delta$ going from $\alpha_1$ 

2437: to either $\alpha_2$ or $\alpha_3.$ This information, and the fact that $L_{d,z}$ must lie 

2438: close to the critical point $A_{red}=0$ for $z$ small, determines the isotopy class of 

2439: $L_{d,z}$ uniquely. A similar argument works for the mondromy image, but in that case the 

2440: thin eigenvalues (and hence $\alpha_2$ and $\alpha_3$) are being swapped. $\hfill \fin$

2441: 

2442: \begin {remark}

2443: \label {moon}

2444: The Lagrangians $L_{d,z}$ and $h_{\gamma_{d,z}}(L_{d,z})$ (constructed using the first 

2445: standard K\"ahler form) can be isotoped to intersect exactly in one point.

2446: \end {remark}

2447: 

2448: \subsection {The fibered case.} \label {sec:fibcase}

2449: We now consider the situation appearing in 

2450: Lemma~\ref{ttt} and Remark~\ref{remcom2}. Our discussion parallels that at the end 

2451: of Section 4(C) in \cite{SS}.

2452: 

2453: Let $X$ be a complex manifold and $\tt$ a holomorphic quadruple bundle over $X$ as in 

2454: Section~\ref{sec:ttt}. Over $X$ there are associated holomorphic vector bundles $\ff_1 

2455: \subset \ff_2 

2456: \subset \ee$ of ranks $1, n, n+1,$ respectively. We also have a fiber bundle $\xx(\tt) 

2457: \subset \sl(\ee)$ over $X$ with fibers isomorphic 

2458: to $\xx_n,$ constructed using Remark~\ref{act}. We make the assumption that the 

2459: bundle $\ff_2 \to X$ is trivial. (This is satisfied in Lemma~\ref{ttt}, cf. the 

2460: observation preceding Remark~\ref{remcom2}.) Consider also the map $q: 

2461: \xx(\tt) \to \sym^2(\cc) \cong \cc^2$ which is equal to (\ref{mapp}) in every $\xx_n$ 

2462: fiber. 

2463: 

2464: We identify $X$ with the zero section in $\sl(\ee),$ and endow $\sl(\ee)$ with a K\"ahler 

2465: metric $\omega.$ Let $K, K'$ be two closed Lagrangian submanifolds of $X,$ and assume that 

2466: $\omega$ is real analytic in a neighborhood of $K, K' \subset \sl(\ee).$ Using parallel 

2467: transport in the union of the critical value sets $Crit(q_d) \cong X \subset 

2468: \xx(\tt)_{d,0}$ 

2469: corresponding to $A_{\red} = 0$, we obtain Lagrangian submanifolds $K_d, K'_d$ in 

2470: $Crit(q_d)$ for any small $d.$

2471: 

2472: Applying the results of Section~\ref{sec:relvan} and taking into account the triviality 

2473: of $\ff_2$ and Remark~\ref{remc}, we get associated relative vanishing 

2474: projective spaces for $\pi_d$ in the form of Lagrangian submanifolds

2475: \begin {equation}

2476: \label {eins}

2477:  L_{d, z}, \ L'_{d, z} \subset \xx(\tt)_{d,z}

2478: \end {equation}

2479: for $0 < z \ll d.$ These are diffeomorphic to $K \times \cpn$ and $K' \times \cpn,$ 

2480: respectively.

2481: 

2482: Fix a relatively compact open subset $W \subset X$ containing $K, K',$ and an open subset 

2483: $V \subset \xx(\tt)$ which is the unit ball bundle over $W$ with respect to the given 

2484: metric. 

2485: For $0 < z \ll d,$ the Lagrangians $L_{d,z}, L'_{d,z}$ will lie in $V.$ The estimates for 

2486: parallel transport from Section~\ref{sec:seces} apply in this situation as well, and show 

2487: that there is another well-defined Lagrangian submanifold

2488: \begin {equation}

2489: \label {zwei}

2490: L''_{d,z} = h_{\gamma_{d,z}}(L'_{d,z}) \subset \xx(\tt)_{d,z} \cap V.

2491: \end {equation}

2492: 

2493: Let us now look at a particular construction for the K\"ahler form $\omega$. Given an 

2494: arbitrary K\"ahler form on $X$ and a Hermitian metric on $\ee \to X$ (both with good local 

2495: real analyticity properties), we can combine them to obtain an associated form $\omega$ on 

2496: $\sl(\ee)$ which is K\"ahler at least in a neighborhood of the zero set of the moment map 

2497: $\mu$ for (\ref{un}). (For details of this construction, we refer to Remark 33 in 

2498: \cite{SS}.) Given $\alpha \in \cc$ and a path $\delta$ in $\cc$ as in 

2499: Section~\ref{sec:ac}, we get a Lagrangian submanifolds $\Lambda_{d,z,K, \delta}$ by 

2500: applying the construction (\ref{bigla}) fiberwise over a Lagrangian $K \subset X.$ The 

2501: arguments in the proof of Lemma~\ref{earth} carry over to give the following 

2502: generalization:

2503: 

2504: \begin {lemma}

2505: \label {earth2}

2506: Equip $\Reg \xx(\tt)$ and the smooth fibers $\xx(\tt)_{d,z}$ with the restriction of a 

2507: K\"ahler form $\omega$ 

2508: constructed as in the previous paragraph. Given closed Lagrangian submanifolds $K, K' 

2509: \subset X,$ consider the relative vanishing projective spaces (\ref{eins}) and the 

2510: mondromy image (\ref{zwei}). Then, up to Lagrangian isotopy,

2511: $$ L_{d,z} = \Lambda_{d, z, K, \delta_1}, \ \ \ L''_{d,z} = \Lambda_{d,z, K', 

2512: t_{\delta_2}\delta_1}.$$

2513: Here the paths $\delta_1$ and $t_{\delta_2}\delta_1$ are as in Figure~\ref{fig:lag}.

2514: \end {lemma}

2515: 

2516: 

2517: \section {Floer cohomology}

2518: \label{sec:floer}

2519: 

2520: Lagrangian Floer cohomology (\cite{F}, \cite{FOOO}) can be defined in various settings, and it 

2521: comes with various amounts of structure depending on how restrictive our assumptions are. We 

2522: work here in the setting of \cite[Section 4(D)]{SS}, except that we need to relax the spinness 

2523: condition on Lagrangians. 

2524: 

2525: \subsection {Definition} \label {sec:flo} Let $M$ be a Stein manifold endowed with an 

2526: exact K\"ahler form $\omega.$ We assume that $c_1(M) = 0$ and $H^1(M) =0.$ Let $L,L'$ be 

2527: two closed, connected, oriented Lagrangian submanifolds of $M$ satisfying $H_1(L) = 

2528: H_1(L') =0.$ We also assume that the pair of Lagrangians $(L, L')$ is relatively spin in 

2529: the sense of \cite{FOOO}, meaning that there exists a class $st \in H^2(M; \zz/2\zz)$ that 

2530: restricts to $w_2(L)$ on $L$ and to $w_2(L')$ on $L'.$ The choice of the class $st$ is 

2531: called a {\it relative spin structure}.

2532: 

2533: By doing a small Lagrangian perturbation we can arrange so that the intersection $L \cap L'$ is 

2534: transverse. The Floer cochain complex is then defined to be the abelian group

2535: $$ CF(L, L') = \bigoplus_{x \in L \cap L'} O_x,$$

2536: where $O_x$ is the orientation group of $x.$ This is the abelian group 

2537: (noncanonically isomorphic to $\zz$) which is generated by the two possible orientations of 

2538: $x,$ with relation that their sum is zero. Our assumptions allow one to define a relative 

2539: Maslov index $\Delta gr(x,y) \in \zz$ for every $x,y \in L \cap L'.$ The relative index 

2540: staisfies $\Delta gr(x,y) + \Delta gr(y,z) = \Delta gr(x,z)$ and induces a relative 

2541: $\zz$-grading on the cochain complex.

2542: 

2543: The Lagrangian Floer cohomology groups $HF^*(L,L')$ are the cohomology groups of $CF^*(L, L')$ 

2544: with respect to the differential $d_J$ defined on generators by

2545: $$ d_J(x) = \sum_y n_{xy} y.$$

2546: Here $n_{xy} \in \zz$ is the signed count of isolated solutions (modulo translation in $s$) to 

2547: Floer's equation

2548: \begin{equation} \label{floer}

2549: \left\{

2550: \begin{aligned}

2551:  & u: \rr \times [0,1] \rightarrow M, \\

2552:  & u(s,0) \in L, \quad u(s,1) \in L', \\

2553:  & \partial_s u + J_t(u) \partial_t u = 0, \\

2554:  & \lim_{s \rightarrow +\infty} u(s,\cdot) = x, \  \lim_{s \rightarrow -\infty} u(s,\cdot) = y,

2555: \end{aligned}

2556: \right.

2557: \end{equation}

2558: where $J = (J_t)_{0 \leq t \leq 1}$ is a generic smooth family of

2559: $\omega$-compatible almost complex structures, which all agree

2560: with the given complex structure outside a compact subset. The solutions to (\ref{floer}) are 

2561: called pseudo-holomorphic disks.

2562: 

2563: The fact that $M$ is Stein implies that the solutions of (\ref{floer}) remain inside a 

2564: fixed compact subset of $M.$ The exactness of $\omega$ and the fact that $H^1(L) = H^1(L') 

2565: = 0$ ensure that no bubbling occurs and thus the moduli spaces of solutions to 

2566: (\ref{floer}) have well-behaved compactifications. Finally, the condition that the 

2567: Lagrangians are relatively spin is necessary in order to define orientations on the moduli 

2568: spaces. The orientations depend on some additional data, cf. \cite[p.192]{FOOO}: the class 

2569: $st \in H^2(M; \zz/2\zz)$ which we called a relative spin structure, as well as 

2570: (in principle) choices of spin structures on certain bundles over the two-skeleta of $L$ 

2571: and $L'.$ However, in our case the latter piece of information is vacuous: since our 

2572: assumptions imply that $H^1(L; \zz/2\zz) = H^1(L';\zz/2\zz)=0,$ the spin structures are 

2573: unique. (This follows from \cite[p. 81, Corollary 1.5]{LM}, for example.) 

2574: 

2575: An important property of the Floer cohomology groups $HF^*(L, L')$ in our setting is that 

2576: they are invariant under Lagrangian isotopies of either $L$ and $L'.$ Furthermore, they 

2577: are invariant under any smooth deformation of the objects involved in their definition 

2578: (for example, the K\"ahler metric), as long as all the assumptions which we made are still 

2579: satisfied.

2580: 

2581: \subsection {Absolute gradings.} \label{sec:abs} As defined in the previous subsection, 

2582: the Floer cohomology $HF^*(L, L')$ groups are only relatively $\zz$-graded. However, in 

2583: the presence of some additional data, one can improve this to an absolute grading. This 

2584: improvement is due to Seidel \cite{Se}, who was inspired by the ideas of Kontsevich 

2585: \cite{Ko}.

2586: 

2587: Since $c_1(Y) = 0,$ we can pick a complex volume form $\Theta$ on $Y,$ i.e. a nowhere vanishing 

2588: section of the canonical bundle. This determines a square phase map

2589: \begin {equation}

2590: \label {sqp1}

2591:  \theta : \lfrak \to

2592: \cc^*/\rr_+, \ \theta(V) = \Theta(e_1 \wedge \dots \wedge e_n)^2

2593: \end {equation}

2594: for any orthonormal basis $e_1, \dots, e_n$ of $T_xL.$ 

2595: We can identify $\cc^*/\rr_+$ with $S^1$ by the contraction $z \to z/|z|.$ The condition 

2596: $H^1(\lll) = 0$ allows us to lift $\theta_{\lll}$ to a real-valued map. A {\it grading} on 

2597: $\lll$ 

2598: is a choice $\tilde \theta_{\lll} :\lll \to \rr$ of such a lift. If we choose a grading for 

2599: $\lll'$ 

2600: as well, then every

2601: point $x$ in the intersection $\lll \cap \lll'$ (which was assumed to be

2602: transverse) has a well-defined absolute Maslov index $gr(x) \in \zz$ \cite{Se}, such that 

2603: $\Delta gr(x,y) = gr(x) - gr(y).$

2604: In turn, this gives an absolute $\zz$ grading on the cochain 

2605: complex and on cohomology. The result does not depend on the choice of $\theta,$

2606: because the condition $H^1(Y) =0$ ensures that different choices are

2607: homotopic in the class of smooth trivializations of the canonical bundle.

2608: 

2609: We should note that if we denote by $L \to L[1]$ the process which subtracts the 

2610: constant $1$ from the grading, then we have

2611: \begin {equation}

2612: \label {grad}

2613:  HF^*(L, L'[1]) = HF^*(L[-1], L') = HF^{*+1} (L, L').

2614: \end {equation}

2615: 

2616: 

2617: \subsection {A K\"unneth formula} Let us consider Floer cohomology in the geometric 

2618: situation from Section~\ref{sec:relvan}. We work in the following context, which is 

2619: slightly more general than the one in Section~\ref{sec:relvan}:

2620:  

2621: \begin{itemize} 

2622: \item 

2623: $Z$ is a complex affine variety equipped with a holomorphic function $\pi: Z \rightarrow 

2624: \cc.$ There is a neighborhood $D$ of the origin in $\cc$ such that the open set 

2625: $\pi^{-1}(D - \{ 0 \}) \subset Z$ is smooth, and the restriction of $\pi$ to this open set 

2626: is a smooth submersion.

2627: 

2628: \item

2629: We have a smooth complex subvariety $X \subset Z$ and an isomorphism between a 

2630: neighbourhood of that subvariety and a neighbourhood of $X \times \{0\} \subset X \times 

2631: \slan$, such that the following diagram commutes: 

2632: \begin{equation} \label {eql}

2633: \begin{CD}

2634:  Z @>{\text{local $\iso$ defined near $X$}}>> X \times \slan \\

2635:  @V{\pi}VV @V{\chin}VV \\

2636:  \cc @>{\quad\quad\qquad\qquad\quad}>> \cc

2637: \end{CD}

2638: \end{equation}

2639: 

2640: \item The regular set $\Reg Z$ of $Z$ carries an exact K{\"a}hler form $\Omega$ that is 

2641: effective near $X$ in the sense of Section~\ref{sec:relvan}. In particular, since $\Omega$ 

2642: is locally obtained by pullback from $X \times \sl(n, \cc),$ this automatically equips $X$ 

2643: with a K\"ahler form as well.

2644: 

2645: \item

2646: We assume that the first Chern classes $c_1(Z_t)$ for 

2647: $t\neq 0,$ as well as $c_1(X),$ are zero. Also, $H^1(Z_t) = 0$ for small $t \neq 0$, and

2648: $H^1(X) = 0$.

2649: \end{itemize}

2650: 

2651: We endow all the smooth fibres $Z_t$ with the restrictions of $\Omega$. Note that 

2652: (\ref{eql}) implies that every $Z_t$ (for $t \neq 0$ small) has an open set $U_t \subset 

2653: Z_t$ diffeomorphic to $X \times T^*\cpn.$ Let $K,K'$ be closed Lagrangian submanifolds of 

2654: $X$ which have the properties necessary to define the Floer cohomology $HF(K,K').$ In 

2655: particular, the pair $(K, K')$ comes with a relative spin structure $st \in H^2(X; 

2656: \zz/2\zz).$ For sufficiently small $t \neq 0$ we have associated relative vanishing 

2657: projective spaces $L_t,L_t' \subset U_t \subset Z_t,$ as in Section~\ref{sec:relvan}. 

2658: These are products of $K,K'$ with $\cpn$, so their Floer cohomology $HF(L_t,L_t')$ is 

2659: again well-defined, provided we make sure that the pair $(L_t, L'_t)$ is relatively spin. 

2660: This last condition is not automatic, so we introduce an additional piece of data:

2661: 

2662: \begin {itemize}

2663: \item 

2664: For small $t \neq 0,$ we have a class $\widetilde{st} \in H^2(Z_t; \zz/2\zz)$ such that 

2665: its restriction to each $U_t \iso X \times T^*\cpn \subset Z_t$ is isomorphic to the product 

2666: of $st$ and $\xi_n.$ Here $\xi_n$ is zero for $n$ even, and equals the generator of 

2667: $H^2(T^*\cpn; \zz/2\zz)$ for $n$ odd. In other words, $\xi_n$ is the pull-back of the 

2668: second Stiefel-Whitney class of $\cpn$ under the projection $T^*\cpn \to \cpn.$  

2669: \end {itemize}

2670: 

2671: We use $\widetilde{st}$ to define $HF(L_t, L_t').$ Note that under these hypotheses, the 

2672: Floer cohomology $HF(L_t, L'_t)$ is independent of $t$ by the invariance principle 

2673: mentioned at the very end of Section~\ref{sec:flo}.

2674: 

2675: \begin{lemma} \label{kunneth}

2676: $HF(L_t,L_t') \iso HF(K,K') \otimes H^*(\cpn)$, where $H^*(\cpn)$ is given its standard 

2677: grading.

2678: \end{lemma}

2679: 

2680: \noindent {\bf Proof. } This is completely analogous to the proof of Lemma 36 in 

2681: \cite{SS}. One can find a holomorphically weakly convex 

2682: neighborhood $V$ of $X$ in $Z$ on which the local isomorphism (\ref{eql}) is well-defined. 

2683: For $t$ small, the Lagrangians $L_t$ and $L'_t$ are in $V.$ Pseudoholomorphic disks in 

2684: $Z_t$ with boundaries in $V \cap Z_t$ cannot go outside $V,$ so the Floer cohomology can 

2685: be computed in $V \cap Z_t.$ On that set we can deform the K\"ahler form to be the product 

2686: of the form on $X$ with the restriction of the standard form $\Omega_{st}$ from the 

2687: example following (\ref{vcps}). Floer cohomology is not changed by this deformation, and 

2688: in the new K\"ahler form $L_t$ and $L_t'$ become products $K \times \cpn$ and $K' \times 

2689: \cpn,$ respectively, where the $\cpn$ factor is the same. The result now follows from a 

2690: straightforward K\"unneth product formula in Floer cohomology, similar to the one in Morse 

2691: theory. The Floer cohomology orientations were chosen to exactly match. $\hfill \fin$

2692: 

2693: 

2694: \subsection {Floer cohomology in the setting of Section~\ref{sec:fibcase}} We now analyze

2695: the following situation:

2696: 

2697: \begin{itemize}

2698: \item

2699: $Z$ is a complex affine variety equipped with a holomorphic map $\pi: Z \to \cc^2.$

2700: The variety $Z$ is smooth over the set of pairs $(d,z) \in \cc^2$ with $0 < z \ll d$

2701: sufficiently small, and the map $\pi$ is a submersion there.

2702: 

2703: \item

2704: We have a smooth complex subvariety $X \subset Z$ and a holomorphic quadruple bundle 

2705: $\tt \to \xx$ in the sense of Section~\ref{sec:ttt}. There are associated holomorphic 

2706: bundles $\ff_1 \subset \ff_2 \subset \ee$ over $X$ of ranks $1, n, n+1,$ respectively, and 

2707: we assume that $\ff_2$ is trivial. From here we obtain a fiber bundle $\xx(\tt) 

2708: \to X$ as a 

2709: subset of $\sl(\ee),$ and a map $q:\xx(\tt) \to \cc^2.$ We assume that there is an 

2710: isomorphism 

2711: between a neighborhood of $X$ in $Z$ and a neighborhood of the zero section in $\xx(\tt) 

2712: \subset \sl(\ee),$ such that the following diagram commutes:

2713: \begin{equation} \label{eql2}

2714: \begin{CD}

2715:  Z

2716:  @>{\text{local $\iso$ defined near $X$}}>>

2717:  {\xx(\tt)} \\

2718:  @V{\pi}VV @V{q}VV \\

2719:  {\cc^2} @>{\quad\qquad\qquad\qquad\quad}>> \cc^2

2720: \end{CD}

2721: \end{equation}

2722: 

2723: \item 

2724: The regular set $\Reg Z$ carries an exact K{\"a}hler form $\Omega$ which in a neighborhood 

2725: of $X$ is obtained by pull-back from a real analytic K\"ahler form on an open set in 

2726: $\sl(\ee).$ In particular, $X$ also has an induced K\"ahler form. 

2727: 

2728: \item

2729: We require that $c_1(Z_{d,z}) = 0$ and $H^1(Z_{d,z})=0$ for $0 < z \ll d.$ Also, 

2730: $c_1(X) =0$ and $H^1(X) =0.$

2731: \end{itemize}

2732: 

2733: Take closed Lagrangian submanifolds $K,K' \subset X$, satisfying the conditions in 

2734: Section~\ref{sec:flo}, so that $HF(K,K')$ is well-defined. For sufficiently small $0 < z 

2735: \ll d,$ the construction in Section~\ref{sec:fibcase} produces some new 

2736: Lagrangians $L_{d,z},L_{d,z}', L''_{d,z} \subset Z_{d,z}.$ These can be assumed to 

2737: lie inside an open subset $U_{d,z} \subset Z_{d,z}$ diffeomorphic to 

2738: $X \times T^*\cpn,$ and where the isomorphism (\ref{eql2}) is well-defined. Given the 

2739: relative spin structure $st$ for the pair $(K, K'),$ we make the following 

2740: additional assumption.

2741: 

2742: \begin {itemize} \item For small $0 < z \ll d,$ we have a class $\widetilde{st} \in 

2743: H^2(Z_{d,z}; \zz/2\zz)$ such that its restriction to each $U_{d,z} \iso X \times T^*\cpn 

2744: \subset Z_{d,z}$ is isomorphic to the product of $st$ and $\xi_n.$ Here $\xi_n$ is zero 

2745: for $n$ even, and equals the generator of $H^2(T^*\cpn; \zz/2\zz)$ for $n$ odd. \end 

2746: {itemize}

2747: 

2748: Under these hypotheses, we can use $\widetilde{st}$ to define the Floer cohomology  

2749: $HF(L_{d,z}, L''_{d,z}),$ for $0 < z \ll d$ small.

2750: 

2751: \begin {lemma}

2752: \label {thom}

2753: $HF(L_{d,z}, L''_{d,z}) \iso HF(K, K').$

2754: \end {lemma}

2755: 

2756: \noindent {\bf Proof. } This goes just like the proof of Lemma 38 in \cite{SS}, so we only 

2757: sketch the argument. We can assume that $K$ and $K'$ intersect transversely. Floer 

2758: cohomology can be calculated inside a suitable holomorphically weakly convex subset of 

2759: $Z_{d,z}$ where (\ref{eql2}) is defined. On that set we can deform the K\"ahler form into 

2760: one that is induced by the form on $X$ and a Hermitian form on $\ee,$ as in the discussion 

2761: at the end of Section~\ref{sec:fibcase}. Lemma~\ref{earth2} and Remark~\ref{moon} show 

2762: that $L_{d,z}$ and $L''_{d,z}$ can be isotoped so that their intersection points exactly 

2763: correspond to intersection points of $K$ and $K'.$ With a little more care we can also 

2764: arrange that pseudoholomorphic disks in the two settings are in one-to-one correspondence 

2765: as well. This implies that the Floer cohomology groups are isomorphic. $\hfill \fin$

2766: 

2767: 

2768: 

2769: \section {Definition of the invariants}

2770: \label {sec:maind}

2771: 

2772: This section contains the construction of the Floer cohomology groups $\krss.$ We apply the 

2773: symplectic geometric techniques which we developed so far to the map

2774: \begin {equation}

2775: \label {our}

2776: \chi^{\pi}|_{\ss_{m,n}}: \ss_{m,n} \cap \gg^{\pi} \to \hh^{\pi}/W^{\pi} 

2777: \end {equation}

2778: from Section~\ref{sec:props}. Recall that $\gg = \sl(mn, \cc), \ \pi$ is the partition 

2779: $(1^m(n-1)^m),$ and the fibers of $\chi^{\pi}|_{\ss_{m,n}}$ are denoted $\ymnt.$ The 

2780: restriction of (\ref{our}) to the set $\ss_{m,n} \cap \gg^{\pi, \reg}$ lying over 

2781: the bipartite configuration space $BConf^0_m$ is a symplectic fibration.

2782: 

2783: We extend the notation $\ymnt$ to $\tau = (\lambdav, \muv)$ such that the trace $T=\sum 

2784: \lambda_j + (n-1) \sum \mu_j$ is not necessarily zero. (When the $\lambda$'s and $\mu$'s are 

2785: distinct, the values of $\tau$ form a configuration space $BConf_m = Conf^{\pi}$ as defined in 

2786: Section~\ref{sec:rpr}.) In this more general situation, by $\ymnt$ we mean the fiber of 

2787: (\ref{our}) over the normalized configuration made of $\lambda_j - T/(nm)$ and $\mu_j - 

2788: T/(nm).$

2789: 

2790: \subsection {Parallel transport in $\ss_{m,n} \cap \gg^{\pi, \reg}$} 

2791: \label {sec:kh}

2792: We seek a K\"ahler metric $\Omega = -dd^c \psi$ on the total space of our symplectic fibration

2793: \begin {equation}

2794: \label {ourf}

2795:  \ss_{m,n} \cap \gg^{\pi, \reg} \to BConf^0_m

2796: \end {equation}

2797: satisfying the conditions (\ref{cond1})-(\ref{cond4}), so that rescaled parallel 

2798: transport is well-defined.

2799: 

2800: The function $\psi$ will be the restriction of some $\hat \psi: \ss_{m,n} \to \rr$ whose 

2801: construction is completely analogous to that in \cite[Section 5(A)]{SS}. Pick some $\alpha > 

2802: m.$ For each $k=2,4, \dots, 2m,$ apply the function $\xi_k(z) = |z|^{2\alpha/k}$ to the 

2803: coordinates of $\ss_{m,n}$ on which the action (\ref{action}) is by weight $k$ or, in other 

2804: words, to the entries of $Y_{1k}$. Sum up these terms to get a function $\xi: \ss_{m,n} \to 

2805: \cc.$ This is not smooth at the origin, but we can perturb it using compactly supported 

2806: functions $\eta_k$ on $\cc$ such that $\psi_k = \eta_k + \xi_k$ is $C^{\infty}.$ We 

2807: choose the $\psi_k$ to be real analytic in a neighborhood of the origin, and strictly 

2808: plurisubharmonic everywhere. Adding up all 

2809: the $\psi_k$'s we obtain the function $\hat \psi$ on $\ss_{m,n},$ whose restriction 

2810: to $ \gg^{\pi, \reg}$ is $\psi.$ We set $\Omega = -dd^c \psi.$ 

2811: \begin {lemma}

2812: \label {rescpt}

2813: The function $\psi$ satisfies (\ref{cond1})-(\ref{cond4}).

2814: \end {lemma}

2815: 

2816: The proof of Lemma~\ref{rescpt} is identical to those of Lemmas 41 and 42 in \cite{SS}, so we 

2817: omit the details. The main ingredient in the proof of (\ref{cond2}) is that by construction 

2818: $\psi$ is asymptotically homogeneous for the action of $\rr_+ \subset \cc^*$ by (\ref{action}). 

2819: We can then control its critical points using this $\rr_+$-action, after pulling back to the 

2820: resolution $\tilde \ss_{m,n} \cap \tilde \gg^{\pi}$ from (\ref{big}).

2821: 

2822: According to the discussion in Section~\ref{sec:parallel}, the function $\psi$ defines rescaled 

2823: parallel transport maps 

2824: $$ h^{\resc}_{\beta}: \yy_{m,n,\beta(0)} \to \yy_{m,n, \beta(1)}$$

2825: for every path $\beta:[0,1] \to BConf^0_m.$ (Strictly speaking, $h_{\beta}$ is only 

2826: well-defined on compact subsets, but these can get arbitrarily large, cf. the note at the end 

2827: of Section~\ref{sec:parallel}). We also extend the notation $h_{\beta}$ to paths in $Conf^0_m,$ 

2828: with the understanding that each point of the path is translated by $T/(mn),$ where $T$ is 

2829: its trace.

2830: 

2831: Rescaled parallel transport also exists for the symplectic fibration (\ref{cmnt}). The total 

2832: space $\ccc_{m,n}$ lives inside $\ss_{m,n},$ so we can take the restriction of the function 

2833: $\hat \psi$ and the corresponding K\"ahler form. Given a path $\bar \beta: [0,1] \to 

2834: Conf^0_{\sigma}$ with $\sigma=(1^{m-1}(n-1)^{m-1}n)$ as in Section~\ref{sec:thickthin}, we get 

2835: a rescaled parallel transport map $h^{\resc}_{\bar \beta}.$ Denote by $BConf^{0*}_{m-1} \subset 

2836: Conf^0_{\sigma}$ the subset corresponding to the multiplicity $n$ eigenvalue being zero. We can 

2837: identify $BConf^{0*}_{m-1}$ to an open subset of $BConf^0_{m-1}$ in the obvious way. Note that 

2838: according to Lemma~\ref{identh}, a fiber $\ccc_{m,n,\tau}$ with $\tau=\{(0, \lambda_2, \dots, 

2839: \lambda_m), (0, \mu_2, \dots, \mu_m)\}$ can be identified with $\yy_{m-1,n, \bar \tau},$ with 

2840: $\bar \tau =\{(\lambda_2, \dots, \lambda_m), (\mu_2, \dots, \mu_m)\}.$ This identification is 

2841: compatible with our choice of symplectic forms, provided we take the same $\alpha > 0$ and 

2842: functions $\psi_k$ for both $m$ and $m-1.$ Thus parallel transport in $\ccc_{m,n}$ over paths 

2843: in $BConf^{0*}_{m-1}$ is the same as parallel transport in $\ss_{m-1,n} \cap 

2844: \sl((m-1)n,\cc)^{\pi'},$ where $\pi'=(1^{m-2}(n-1)^{m-2}).$

2845: 

2846: 

2847: \subsection {The Lagrangians} \label{sec:tl} Let $\tau =\{(\lambda_1, \lambda_2, \dots, 

2848: \lambda_m), (\mu_1, \mu_2, \dots, \mu_m)\}$ be a point in the bipartite configuration space 

2849: $BConf_m.$

2850: 

2851: \begin {definition}

2852: A crossingless matching $\mat$ with endpoints $\tau$ is a collection of $m$ disjoint embedded 

2853: arcs $(\delta_1, \dots, \delta_m)$ in $\cc$ such that each $\delta_k: [0,1] \to \cc$ satisfies 

2854: $\delta_k(0) = \lambda_k$ and $\delta_k(1) = \mu_{\nu(k)},$ where $\mu$ is a permutation of 

2855: $\{1,2, \dots, m\}.$ \end {definition}

2856: 

2857: To each crossingless matching $\mat$ we associate a Lagrangian submanifold $L_{\mat} \subset 

2858: \ymnt,$ diffeomorphic to the product of $m$ copies of $\cpn,$ and unique up to Lagrangian 

2859: isotopy. Since we had not fixed an ordering of $(\mu_1, \mu_2, \dots, \mu_m),$ we can assume 

2860: that $\nu$ is the identity permutation, so that $\delta_k$ joins $\lambda_k$ and $\mu_k.$ We 

2861: choose a path $[0,1) \to BConf_m$ that starts at $\tau$ and moves the endpoints of $\delta_1$ 

2862: towards each other along that arc, while keeping all the other components of $\tau$ constant. 

2863: The endpoints $\lambda_1$ and $\mu_{\nu(1)}$ should collide as $s \to 1.$ We assume that 

2864: $\delta_1$ is a straight line near its midpoint $\lambda$, and that the colliding points move 

2865: towards the midpoint with the same speed as $s \to 1.$ We can translate this whole picture into 

2866: $BConf^0_m \subset \hh^{\pi}/W^{\pi}$ as before. We obtain 

2867: a path $\gamma: [0,1] \to 

2868: \hh^{\pi}/W^{\pi}$ with $\gamma(1)= \{(\lambda', \lambda'_2, \dots, \lambda'_m), (\lambda', 

2869: \mu'_2, \dots, \mu'_m)\},$ such that $\lambda_k' = \lambda_k + \lambda' - \lambda$ and $\mu'_k 

2870: = \mu_k - \lambda' - \lambda,$ for all $k=2, \dots, m.$

2871: 

2872: The construction of $L_{\mat}$ proceeds by induction on $m,$ with $n$ being kept fixed. In 

2873: the case $m=1$ we have the vanishing $\cpn$ from Section~\ref{sec:van} in the fibers over 

2874: $\gamma(1-s)$ for small $s.$ We use reverse (rescaled) parallel transport along $\gamma$ 

2875: to move this back to the fiber $\yy_{1,n, \tau}$ over $\gamma(0).$ For the inductive step, 

2876: denote by $\bar \mat$ the crossingless matching of $(2m-2)$ points which is obtained from 

2877: $\mat$ after removing $\delta_1.$ By assumption, we have a Lagrangian $L_{\bar \mat} 

2878: \subset \yy_{m-1, n, \bar \tau},$ where $\bar \tau$ consists of the endpoints of $\bar 

2879: \mat.$ According to Lemma~\ref{identh}, the space $\yy_{m-1, n,\bar \tau}$ can be 

2880: identified with a fiber of the singular set fibration $\ccc_{m,n} \to Conf^0_{\sigma}.$ 

2881: Using parallel transport in $\ccc_{m,n},$ we can move the Lagrangian into the singular 

2882: locus of $\yy_{m,n, \gamma(1)}.$ The local models from Lemma~\ref{thickthin} and 

2883: Remark~\ref{remcom} tell us that 

2884: we can apply the relative $\cpn$ construction from Section~\ref{sec:relvan} and get a 

2885: Lagrangian submanifold in $\yy_{m,n, \gamma(1-s)}$ for small $s.$ Using reverse parallel 

2886: transport along $\gamma$ we move this to the fiber over $\gamma(0).$ The result is the 

2887: desired Lagrangian $L_{\mat} \subset \ymnt.$ Note that the necessary real analyticity 

2888: condition on the K\"ahler metrics (cf. Definition~\ref{eff} and Section~\ref{sec:relvan}) 

2889: is satisfied because of the way we chose the $\psi_k$'s in Section~\ref{sec:kh}.

2890: 

2891: Let us note a few properties of the Lagrangians that arise from this construction. Observe that 

2892: the construction can be done in families as well. If $\mat(s)\  (s \in [0,1])$ is a 

2893: smooth family of crossingless matchings with endpoints $\beta(s),$ one can construct 

2894: Lagrangians $L_{\mat(s)} \subset \yy_{m,n, \beta(s)}$ depending smoothly on $s.$ Parallel 

2895: transport along $\beta|_{[s,1]}$ can be used to carry them into a common fiber. We obtain a 

2896: Lagrangian isotopy

2897: \begin {equation}

2898: \label {lagi}

2899: L_{\mat(1)}\ \iso \ h_{\beta}^{\resc}(L_{\mat(0)}).

2900: \end {equation}

2901: 

2902: One corollary is that a smooth family of crossingless matchings with fixed endpoints always 

2903: yields a family of isotopic Lagrangians.

2904: 

2905: Note that in the first step of the construction of $L_{\mat}$ (corresponding to $m=1$), 

2906: the matching is simply a path from $\lambda_1$ to $\delta_1,$ and therefore its isotopy 

2907: class is unique. This implies that the procedure does not depend essentially on 

2908: $\delta_1.$ A quick consequence is the following:

2909: 

2910: \begin {lemma}

2911: \label {mm}

2912: Let $\mat = (\delta_1, \delta_2, \delta_3, \dots, \delta_m)$ and $\mat'=(\delta_1, \delta'_2, 

2913: \delta_3, \dots, \delta_m)$ be two crossingless matchings related to each other 

2914: as shown in Figure~\ref{fig:match}: $\delta_2'$ is obtained from $\delta_2$ by sliding it 

2915: over $\delta_1,$ using a dashed path (shown dashed in the figure) that does not intersect any 

2916: of the other arcs. Then $L_{\mat}$ and $L_{\mat'}$ are Lagrangian isotopic.

2917: \end {lemma}

2918: 

2919: \begin {figure}

2920: \begin {center}

2921: \input {match.pstex_t}

2922: \end {center}

2923: \caption {Matchings $\mat$ and $\mat'$.}

2924: \label {fig:match}

2925: \end {figure}

2926: 

2927: Recall that in the local model given by Lemma~\ref{multithickthin}, we can change the order in 

2928: an iteration of the vanishing $\cpn$ procedure, cf. Lemma~\ref{changeorder}. Applying this to 

2929: our construction, we obtain:

2930: 

2931: \begin {lemma}

2932: \label {indy}

2933: Up to Lagrangian isotopy, $L_{\mat}$ is independent of the ordering of the components of the 

2934: matching $\mat.$

2935: \end {lemma}

2936: 

2937: 

2938: \subsection {Floer cohomology} \label{sec:fc}

2939: 

2940: Given an oriented link $\lk \subset S^3$ we can represent it as the closure of an $m$-stranded 

2941: braid $b \in Br_m.$ (See Figure~\ref{fig:braid} for a presentation of the left-handed trefoil.)

2942: Note that the braid group $Br_m$ is the fundamental group of the configuration space 

2943: $Conf_{(1^m)}$ of $m$ unordered points in the plane.

2944: 

2945: 

2946: \begin {figure}

2947: \begin {center}

2948: \input {braid.pstex_t}

2949: \end {center}

2950: \caption {The trefoil as a braid closure.}

2951: \label {fig:braid}

2952: \end {figure}

2953: 

2954: Consider the standard crossingless matching $\mat_0$ shown in Figure~\ref{fig:std}, where all 

2955: the $\lambda$'s and the $\mu$'s are on the real line, with the ``thin points'' $\lambda$'s to 

2956: the left of the ``thick points'' $\mu$'s, and all the arcs lying in the upper half-plane. Let 

2957: $D \subset \cc$ be a disk that contains all the thin points and none of the thick ones. We 

2958: denote by $U \subset BConf_m$ the open subset corresponding to bipartite configurations $\tau = 

2959: \{(\lambda_1, \lambda_2, \dots, \lambda_m), (\mu_1,\mu_2, \dots, \mu_m)\},$ where the 

2960: $\mu_k$'s are fixed to be the ones chosen for $\mat_0,$ while the $\lambda_k$'s are free to 

2961: move inside the disk $D.$ Using a deformation retraction $\cc \to D,$ we can identify $Br_m$ 

2962: with the fundamental group of $U.$

2963: 

2964: \begin {figure}

2965: \begin {center}

2966: \input {std.pstex_t}

2967: \end {center}

2968: \caption {The standard crossingless matching $\mat_0$.}

2969: \label {fig:std}

2970: \end {figure}

2971: 

2972: In this fashion, the braid $b$ representing $\lk$ induces a path $\beta: [0,1] \to U$ with 

2973: $\beta(0) = \beta(1)$ being the configuration $\tau_0$ formed from the endpoints of $\mat_0.$ 

2974: We consider the pair of Lagrangians $(L, L')$ in $M= \yy_{m,n, \tau_0},$ where $L$ is 

2975: $L_{\mat_0}$ and $L'= h_{\beta}^{resc}(L)$ is obtained from $L$ by parallel transport. We 

2976: denote by $w$ writhe of the braid $b,$ and set

2977: \begin {equation}

2978: \label {defff}

2979: \krss(\lk) = HF^{*+(n-1)(m+w)}(L, L').

2980: \end {equation}

2981: 

2982: The fact that these groups are link invariants will be shown in the next section. For now, 

2983: let us check that we are in the setting of Section~\ref{sec:flo}, so that the Floer cohomology

2984: groups on the right-hand side of (\ref{defff}) are well-defined. 

2985: 

2986: The complex manifold $M=\yy_{m,n, \tau_0}$ is an affine variety, and therefore Stein. The 

2987: K\"ahler form $\omega = -dd^c \psi$ is exact by construction. The simultaneous resolution 

2988: (\ref{slice}) shows that $M$ is deformation equivalent to the nilpotent fiber $\nnn_{\rho 

2989: \pi}$ from (\ref{nnn}). It follows from the work of Maffei \cite{Maf} that $\nnn_{\rho \pi}$ 

2990: is a quiver variety in the sense of Nakajima \cite{N1}, and hence hyper-K\"ahler. This 

2991: implies that $c_1 = 0$ for $\nnn_{\rho \pi}$ and then the same must be true for $M.$ 

2992: 

2993: As noted in Section~\ref{sec:top}, the fiber $\nnn_{\rho \pi}$ is homotopy equivalent to the 

2994: Spaltenstein variety $\Pi^{-1}(N_{m,n})$, whose rational cohomology was studied in 

2995: \cite[Section 3.4]{BM}. Borho and MacPherson proved that, up to an even dimension shift, 

2996: $H^*(\Pi^{-1}(N_{m,n}); \qq)$ is equal to a space of anti-invariants in the rational 

2997: cohomology of the Springer variety of complete flags fixed by $N_{m,n}.$ Since the 

2998: odd-dimensional Betti numbers of Springer varieties are zero (cf. \cite{CP}), the same must 

2999: be true for those of Spaltenstein varieties. In particular, it follows that $H^1(M) =0.$

3000: 

3001: The Lagrangians $L$ and $L'$ are diffeomorphic to $(\cpn)^m$, so they satisfy $H_1 = 0$ as 

3002: required. The Stiefel-Whitney class $w_2((\cpn)^m)$ is zero for $n$ even, while for $n$ 

3003: odd it is the class $$(w,w,\dots, w) \in H^2((\cpn)^m; \zz/2\zz)\ \iso \ \bigoplus_{i=1}^m 

3004: H^2(\cpn; \zz/2\zz),$$ where $w$ is the generator of $H^2(\cpn; \zz/2\zz) \ \iso \ 

3005: \zz/2\zz.$

3006: 

3007: When $n$ is even, the Lagrangians $L$ and $L'$ are spin. The following lemma shows that the 

3008: pair $(L, L')$ is relatively spin when $n$ is odd:

3009: 

3010: \begin {lemma}

3011: \label {class}

3012: Every element $Y \in \ss_{m,n} \cap \gg^{\pi, \reg}$ has exactly $m$ one-dimensional 

3013: eigenspaces $E(\lambda_1), E(\lambda_2), \dots, E(\lambda_m),$ coresponding to eigenvalues 

3014: $\lambda_j$ for

3015: $j=1,\dots, m.$ Let $V$ be the complex line bundle over $\ss_{m,n} \cap \gg^{\pi, \reg}$ 

3016: whose fibers are $E(\lambda_1) \otimes E(\lambda_2) \otimes \dots \otimes E(\lambda_m).$

3017: Then the restriction of $w_2(V)$ to any Lagrangian $L_{\mat}$ coming from a matching $m$ is the 

3018: class $(w,w, \dots, w) \in H^2((\cpn)^m; \zz/2\zz).$ 

3019: \end {lemma}

3020: 

3021: \noindent \textbf{Proof.} Lemma~\ref{indy} says that the ordering of the factors is not 

3022: essential in the construction of $L_{\mat}.$ Therefore, it suffices to show that the first 

3023: component of $w_2(V)$ is $w.$ Since continuous deformations do not change the Stiefel-Whitney 

3024: class, this is equivalent to showing that $w_2(V|_C)=w,$ where $C$ is a vanishing $\cpn$ 

3025: appearing in the last step of the construction of $\cpn.$ More precisely, in the notation of 

3026: Section~\ref{sec:tl}, we pick a point $Y_0$ in the singular locus of on $\yy_{m, n, 

3027: \gamma(1)},$ and define $C$ to be the associated vanishing $\cpn$ in a nearby fiber $\yy_{m,n, 

3028: \gamma(1-s)},$ arising from the local model in Lemma~\ref{thickthin}.

3029: 

3030: Recall that $\gamma(1)= \{(\lambda', \lambda'_2, \dots, \lambda'_m), (\lambda', \mu'_2, \dots, 

3031: \mu'_m)\}.$ Let $N$ be a small neighborhood of $\gamma(1)$ in $\hh^{\pi}/W^{\pi}$ and $\tilde 

3032: N$ its preimage in $\ss_{m,n} \cap \gg^{\pi}$ under (\ref{our}). Note that for $k=2, \dots, m,$ 

3033: there are well-defined line bundles $E_k$ over $\tilde N$ whose fibers are the eigenspaces 

3034: corresponding to the eigenvalue close to $\lambda_k'.$ There is also a line bundle $E_1$ over 

3035: $\tilde N \cap \gg^{\pi, \reg}$ corresponding to the eigenspace with eigenvalue close to 

3036: $\lambda',$ but $E_1$ cannot be extended to all of $\tilde N.$

3037: 

3038: Inside $\tilde N,$ the vanishing space $C$ is the boundary of a cone with vertex $x.$ Since the 

3039: cone is contractible, it follows that all $E_k$ can be trivialized over $C,$ for $k=2, \dots, 

3040: m.$ Consequently, $V|_C$ is isomorphic to $E_1|_C.$ Since we only care about the topology, we 

3041: can move $C$ continuously into some $C' \ \iso \ \cpn$ that lies in the canonical semisimple 

3042: slice $\ss^{ss}$ at $Y_0.$ It follows from the discussion in the proof of Lemma~\ref{thickthin} 

3043: that $\ss^{ss} \cap \gg^{\pi}$ (after translation by $Y_0$) can be identified with $\cc^{2m-2} 

3044: \times \slan.$ We can transport $C'$ further into some $C'' \ \iso \ \cpn \subset 

3045: \{0\} \times \slan.$ The pullback of $E_1|_{C''}$ under this identification 

3046: is the tautological bundle over some $\cpn \subset \slan,$ which has $w_2 = w. 

3047: \hfill \fin$ 

3048: 

3049: \medskip

3050: 

3051: In order to define orientations on the moduli spaces of pseudoholomorphic curves, we need 

3052: to choose the relative spin structure $st \in H^2(M; \zz/2\zz).$ We let $st$ be zero when 

3053: $n$ is even, and $w_2(V)$ for $n$ odd, where $V$ is the bundle from Lemma~\ref{class} (or, 

3054: more precisely, its restriction to $\yy_{m,n, \tau_0} \subset \ss_{m,n} \cap \gg^{\pi, 

3055: \reg}).$

3056: 

3057: One other thing needed in the definition of Floer cohomology is the choice of orientations 

3058: on the two Lagrangians. We make this inductively: when $m=1$ we give $\cpn \iso 

3059: U(n)/(U(1) \times U(n-1))$ its natural complex structure and hence a complex orientation; 

3060: then, at each step in the recursive definition of $L_{\mat}$ the new $\cpn$ factor can be 

3061: identified with the one appearing in the $n=1$ case, and therefore we can also give it its 

3062: complex structure and orientation. This defines an orientation on $L_{\mat}$ for any 

3063: matching.

3064: 

3065: Finally, in order to have a well-defined absolute grading for the Floer groups, we need to 

3066: endow the Lagrangians $L, L'$ with gradings as in Section~\ref{sec:abs}. Looking at the 

3067: fibration (\ref{ourf}), we start by choosing arbitrary trivializations for the canonical 

3068: bundles of the total space and the base. This produces a family of trivializations of the 

3069: canonical bundles on the fibers $\ymnt.$ If we choose a grading for the Lagrangian 

3070: $L_{\mat_0} \subset \yy_{m,n, \tau_0},$ we can then transport it in a continuous way to 

3071: gradings on $h^{\resc}_{\beta|[0,s]}(L_{\mat_0}).$ In particular when $s=1$ we get a grading 

3072: on the Lagrangian $L'.$ Relation (\ref{grad}) implies that the resulting $HF(L, L')$ does not 

3073: depend on our choice of grading for $L.$

3074: 

3075: 

3076: \section {Invariance under Markov moves} 

3077: \label {sec:markov}

3078: 

3079: The Floer cohomology groups $\krss$ defined in

3080: (\ref{defff}) are invariant under continuous deformations of the objects involved. We make 

3081: the following observation:

3082: 

3083: \begin {lemma} All the noncanonical choices made in the definition of $\krss,$ with the 

3084: exception of the braid element $b \in Br_m,$ are parametrized by weakly contractible 

3085: spaces. \end {lemma} 

3086: 

3087: \noindent \textbf{Proof.} There were some choices made in the 

3088: construction of the K\"ahler form $\Omega = -dd^c \psi:$ the constant $\alpha \in (m, 

3089: \infty)$ and the functions $\psi_k: \cc \to \rr.$ The functions $\psi_k$ are required to 

3090: satisfy certain properties, but these are preserved under linear interpolation. Therefore, 

3091: the corresponding parameter space is a convex subset of $C^{\infty}(\cc, \rr),$ and hence 

3092: contractible.

3093: 

3094: Another choice was made in Section~\ref{sec:fc}, where a particular path $\beta: [0,1] \to 

3095: U$ was selected as a representative for a given homotopy class in $\pi_1(U)=Br_m.$ The 

3096: space of these representatives is weakly contractible because $U$ is homotopy equivalent to 

3097: the configuration space $\conf_m,$ and the latter was shown to be aspherical in \cite{FN}.

3098: 

3099: There are other choices involved in the definition of Floer cohomology, such as almost 

3100: complex structures and perturbations. It is well-known that the respective parameter spaces 

3101: are weakly contractible. $\hfill \fin$

3102: \medskip

3103: 

3104: Thus, in order to prove that $\krss$ are link invariants, it suffices to show that they are 

3105: independent of the presentation of the link $\lk$ as a braid closure. Two braids have the 

3106: same closure if they are related by a sequence of Markov moves. The proof of 

3107: Theorem~\ref{the} will be completed once we show invariance under the Markov moves.

3108: 

3109: \subsection {Markov $I$}

3110: The type $I$ Markov move consists in replacing a braid $b$ by $s_k^{-1}bs_k,$ where $s_k$ 

3111: is one of the standard generators $s_1, \dots, s_{m-1}$ of $Br_m$, the positive 

3112: half-twist between the $k$th and $(k+1)$th strand. Choose a representative $\sigma_k$ for $s_k$ 

3113: in the form of a loop in the subset $U \subset BConf_m$ from Section~\ref{sec:fc}. We also 

3114: choose a representative $\sigma_{2m-k}$ for the half-twist between the thick points

3115: $\mu_k$ and $\mu_{k+1},$ in the form of a loop in $BConf_m$ that fixes the thin points; in 

3116: fact, we can assume that $\sigma_{2m-k}$ is the identity on the disk $D$ that contains the thin 

3117: points. This implies that the parallel transport maps $h^{\resc}_{\sigma_{2m-k}}$ and 

3118: $h^{\resc}_{\beta}$ do not interfere with each other and therefore commute.

3119: 

3120: The proof of Markov $I$ invariance is exactly as in \cite{SS}. Using (\ref{lagi}) and 

3121: Lemma~\ref{mm}, we get that the the image of $L_{\mat_0}$ under parallel transport along 

3122: $\sigma_{2m-k}^{-1} \circ \sigma_k$ is isotopic to $L_{\mat_0}.$ This fact, together with the

3123: symplectomorphism invariance of Floer cohomology, leads to the string of identities

3124: \begin{align*}

3125:  & HF(L_{\mat_0},h^{\resc}_{\sigma_k^{-1}} h^{\resc}_\beta

3126:  h^{\resc}_{\sigma_k}(L_{\mat_0})) 

3127: \  \iso \ HF(h^{\resc}_{\sigma_k}(L_{\mat_0}),

3128:  h^{\resc}_\beta h^{\resc}_{\sigma_k}(L_{\mat_0})) \\

3129:  & \iso \ HF(h^{\resc}_{\sigma_{2m-k}}(L_{\mat_0}),

3130:  h^{\resc}_\beta h^{\resc}_{\sigma_k}(L_{\mat_0})) 

3131: \  \iso \ HF(L_{\mat_0},

3132:  h^{\resc}_{\sigma_{2m-k}^{-1}} h^{\resc}_\beta

3133:  h^{\resc}_{\sigma_k}(L_{\mat_0})) \\

3134:  & \iso \ HF(L_{\mat_0},h^{\resc}_\beta

3135:  h^{\resc}_{\sigma_{2m-k}^{-1} \circ \,\sigma_k}(L_{\mat_0})) 

3136:  \ \iso \ HF(L_{\mat_0},h^{\resc}_\beta(L_{\mat_0})).

3137: \end{align*} 

3138: 

3139: \subsection {Markov $II$} Given a braid $\bar b \in Br_{m-1},$ the type $II^+$ Markov 

3140: move consists in adding a strand plus a half-twist of that strand with its neighbor,

3141: so that the result is $b = s_{m-1}(\bar b \times 1) \in Br_m.$ There is also a type $II^-$ 

3142: Markov move where instead of $s_{m-1}$ we have the negative half-twist $s_{m-1}^{-1},$ but 

3143: since its treatment is completely similar to that for Markov $II^+$, we will just focus on 

3144: $II^+.$ 

3145: 

3146: Consider the standard configuration $\tau_0 = (\lambda_1, \dots, \lambda_m, \mu_m, \dots, 

3147: \mu_1)$ as in Figure~\ref{fig:std}. The eigenvalues $\lambda_{m-1}$ and $\lambda_m$ are 

3148: those exchanged by $s_{m-1}.$ If $m \geq 3,$ we can assume that $\lambda_{m-1}, 

3149: \lambda_m,$ and $\mu_m$ are small and 

3150: \begin {equation}

3151: \label {ser}

3152: \lambda_{m-1} + \lambda_m + (n-1) \mu_m =0.

3153: \end {equation}

3154: 

3155: In other words, $(\lambda_{m-1}, \lambda_m)$ is a point in a small bidisk $P$ around a 

3156: configuration where two thin and one thick eigenvalues coincide, as in Lemma~\ref{ttt}. 

3157: Let $\bar \tau_0$ be the configuration $(\lambda_1, \dots, \lambda_{m-2}, 0, \mu_{m-1}, 

3158: \mu_{m-2}, \dots, 0)$ and $\bar \beta$ a loop in $BConf^0_{m-1}$ based at $\bar \tau_0$ 

3159: and representing the braid $\bar b.$ Consider the Lagrangians

3160: $$ K = L_{\bar \mat}, \ \ K' = h_{\bar \beta}^{\resc}(L_{\bar \mat}) $$

3161: in $\yy_{m-1,n, \bar \tau_0}.$ The local model described in Lemma~\ref{ttt} and 

3162: Remark~\ref{remcom2} allows us to apply the discussion in Section~\ref{sec:fibcase} and 

3163: construct relative vanishing projective spaces $L_{d,z}, L'_{d,z}$ in the fiber $\yy_{m,n, 

3164: \tau_0}.$ Here $d = \mu_m$ and $z = (\mu_m - \lambda_{m-1})(\mu_m - \lambda_m)$ as in 

3165: Section~\ref{sec:non}.  

3166: 

3167: Note that $L=L_{\mat_0}$ is isotopic to $L_{d,z}$ by construction. The second Lagrangian 

3168: $L'= h_{\beta}^{\resc}(L) $ is obtained from this by parallel transport. A deformation 

3169: argument as in \cite[Section 5(D)]{SS} shows that $L'$ is isotopic to the Lagrangian 

3170: obtained by taking $K',$ doing the relative vanishing $\cpn$ construction to get 

3171: $L'_{d,z},$ and then applying the twist $s_{n-1}.$ This last step corresponds to going 

3172: around the loop $\gamma_{d,z}$ from Figure~\ref{fig:path}. This new Lagrangian is exactly 

3173: $L''_{d,z}$ from (\ref{zwei}).

3174: 

3175: At this point we can apply Lemma~\ref{thom} and find that $HF(L, L') \iso HF(K, K').$ In 

3176: other words, the Floer groups associated to the braid $\bar b$ and its Markov $II^+$ 

3177: transform $b$ are isomorphic. The shift by $(n-1)(m+w)$ in (\ref{defff}) was 

3178: chosen so that the absolute gradings match as well. (See \cite[Section 6(A)]{SS} for more 

3179: details in the $n=2$ case.) 

3180: 

3181: We should note that the discussion above was only applicable for $m \geq 3.$ When $m=2,$ 

3182: the relation (\ref{ser}) cannot be arranged. However, we can bring $\lambda_{m-1}, 

3183: \lambda_m$ and $\mu_m$ close to a nonzero value instead, and then use suitable parallel 

3184: transport in $\ccc_{m,n}$ to bring that value back to zero.

3185: 

3186: \begin {remark}

3187: Instead of doing the twist $s_{n-1}$ in Markov $II^+$ we could just compare the situations 

3188: for the braids $\bar b$ and $\bar b \times 1.$  The Floer cohomology 

3189: of $L= L_{d,z}$ and $L'_{d,z}$ can be computed using Lemma~\ref{kunneth}. Let $O$ 

3190: denote the unknot. Taking into 

3191: account the absolute gradings, we get that our link invariants satisfy: 

3192: $$ \krss ( \lk  \amalg O)= \krss (\lk) \otimes H^{*+n-1}(\cpn).$$

3193: 

3194: In particular, the invariant of the unlink of $p$ components is the tensor product of $p$ 

3195: copies of $H^{*+n-1}(\cpn).$

3196: \end {remark}

3197: 

3198: 

3199: \section {Calculation for the trefoil}

3200: \label {sec:trefoil}

3201: 

3202: In this section we prove Proposition~\ref{3f}. We work with left-handed trefoil knot 

3203: $\lk,$ which is the closure of the braid $b=s_1^3 \in Br_2.$ We need to consider Floer 

3204: cohomology inside the space $Y=\yy_{2,n,\tau_0},$ where $\tau_0$ is the standard 

3205: configuration as in Figure~\ref{fig:std}.

3206: 

3207: Just as in the proof of Markov $II$ invariance, we bring the first three eigenvalues 

3208: $\lambda_1, \lambda_2, \mu_2$ close together. This allows us to work inside the setting of 

3209: Section~\ref{sec:fibcase} once again. For the base space $X$ we have a fiber of the map 

3210: $\chin: \slan \to \cc,$ which contains a vanishing projective space 

3211: $K \iso \cpn.$ This corresponds to the first step in the construction of the Lagrangian 

3212: $L_{\mat},$ i.e. to parallel transport by bringing $\lambda_1$ and $\mu_1$ together. Over 

3213: $X$ we have three bundles $\ff_1 \subset \ff_2 \subset \ee.$ Here the fiber of $\ff_1$ 

3214: over $A \in X$ is the eigenspace for the thin eigenvalue of $A,$ while $\ff_2$ is the whole 

3215: space on which $A$ acts. (Hence, in fact $\ff_2$ is a trivial bundle.) We can choose a 

3216: compatible holomorphic quadruple bundle $\tt \to X$ as in Section~\ref{sec:ttt}, and then 

3217: get a fiber bundle $\xx(\tt) \subset \sl(\ee) \to X.$ This describes the local structure of 

3218: $Y$ near $X.$ Locally near $X$ the K\"ahler form can be deformed to be one obtained by 

3219: combining the standard form on $X$ (from Section~\ref{sec:van}) with a Hermitian form on 

3220: $\ee.$ Recall that if $X$ is equipped with the standard form then $K$ can be described 

3221: explicitly by (\ref{ua}).

3222: 

3223: Applying the same reasoning as in the proof of Markov $II,$ we find that, up to 

3224: isotopy, the Lagrangian $L = L_{\mat_0} \subset Y$ is obtained from $K$ by taking 

3225: fiberwise over it the space $\Lambda_{\delta_1}$ from (\ref{bigla}). Also, the second 

3226: Lagrangian $L' = h_{\beta}^{\resc}(L)$ is obtained from $K$ in the same way, but using 

3227: fibers $\Lambda_{\delta_3}$ instead. Here $\delta_3 = t^3_{\delta_2}(\delta_1)$ as in 

3228: Figure~\ref{fig:t3}.

3229: 

3230: \begin {figure}

3231: \begin {center}

3232: \input {t3.pstex_t}

3233: \end {center}

3234: \caption {Projections of the two Lagrangians.}

3235: \label {fig:t3}

3236: \end {figure}

3237: 

3238: The paths $\delta_1$ and $\delta_3$ intersect in one endpoint and one interior point. By 

3239: the discussion in Section~\ref{sec:ac}, the corresponding intersections of 

3240: $\Lambda_{\delta_1}$ and $\Lambda_{\delta_2}$ are a point and a $(2n-3)$-dimensional 

3241: sphere, respectively. This implies that $L$ and $L'$ intersect in the disjoint union of a 

3242: $\cpn$ and a $S^{2n-3}$-bundle over $\cpn.$ Using Remark~\ref{klm}, the latter turns out 

3243: to be exactly the unit tangent bundle of $\cpn.$ Indeed, the restriction of $\ff_1$ to $K 

3244: \iso \cpn$ is the tautological complex line bundle $\xi,$ and we have $T\cpn = 

3245: \text{Hom}(\xi, \cc^n/\xi).$

3246: 

3247: The intersection $L \cap L'$ is clean in the sense of Pozniak \cite{P}. This gives a 

3248: Morse-Bott long exact sequence

3249: $$ \cdots \to H^{*-2}(UT\cpn) \to HF^*(L, L') \to H^*(\cpn) \to H^{*-1}(UT\cpn) \to 

3250: \cdots $$

3251: 

3252: The absolute gradings can be understood from the geometry of the fibers. Observe that the 

3253: differential $H^*(\cpn) \to H^{*-1}(UT\cpn) $ is zero for simple algebraic reasons. After 

3254: taking into account the shift factor $(n-1)(m+w) = -n+1$ in (\ref{defff}), the proof of 

3255: Proposition~\ref{3f} is completed. Explicitly, we have 

3256: $$ \krsk(\lk)=

3257: \begin{cases}

3258: \zz & \text{ for $k=n-1$;}\\

3259: \zz^2 & \text{ for $k=n+1+2j, \ 0 \leq j \leq n-2$;}\\

3260: \zz/n\zz & \text{ for $k=3n-1$;}\\

3261: \zz & \text{ for $k=3n+2j, \ 0 \leq j \leq n-2$;}\\

3262: 0 & \text{ otherwise.}

3263: \end{cases}

3264: $$

3265: 

3266: \section {Other link invariants}

3267: \label {sec:other}

3268: 

3269: This section is more speculative in nature. We suggest some further avenues of research, 

3270: by explaining how variants of our construction could lead to other link invariants.

3271: 

3272: \subsection {Other Lie algebras and representations} In Section~\ref{sec:motiv} we noted 

3273: that the link polynomial $P_{(n)}$ from (\ref{skey}) is associated to the standard 

3274: $n$-dimensional representation $V$ of $\sl(n, \cc).$ The procedure described there can in 

3275: fact be applied to any complex simple Lie algebra $\ggg,$ the result being a polynomial 

3276: invariant of decorated links, where the link components are decorated by 

3277: finite-dimensional irreducible representations of $\ggg$ (\cite{RT}). In the case of 

3278: $P_{(n)},$ all components are decorated by $V.$

3279: 

3280: Let us consider the situation in which the Lie algebra is still $\sl(n, \cc),$ but the 

3281: components of the link are decorated by various exterior products of the standard 

3282: representation $V.$ Represent the link as the closure of a braid $b$, and suppose that 

3283: $b$ has $m_i$ strands colored in the representation $\Lambda^kV,$ for $0 < k < n.$ Let 

3284: the total number of strands be $m = m_1 + \dots + m_{n-1}.$ Form the representation $$ W = 

3285: \bigotimes_{k=1}^{n-1} \ (\Lambda^kV)^{\otimes m_k}.$$

3286: 

3287: To the braid $b$ one associates a map $F_b(q):W \to W,$ and then the  

3288: polynomial invariant is obtained as the image of $1$ under a map

3289: $$\begin {CD}

3290: \cc \to  W \otimes W^* @>{F_b(q) \times id}>> W \otimes W^* \to \cc. \end {CD} $$

3291: similar to (\ref{poly}). Note that the dual of $\Lambda^k V$ is $\Lambda^{n-k}V,$ so that

3292: $$ W \otimes W^* = \bigotimes_{k=1}^{n-1} \ (\Lambda^kV)^{\otimes (m_k + m_{n-k})}.$$

3293: 

3294: In \cite[Section 11]{KR1}, Khovanov and Rozansky sketched the construction of a 

3295: homology theory that categorifies this particular polynomial invariant. We conjecture that 

3296: a similar theory can be constructed from Floer homology, and that it is equivalent to

3297: that of Khovanov-Rozansky. One should still use intersections of transverse slices and 

3298: adjoint orbits in $\sl(nm, \cc),$ but the partition $\pi$ from Section~\ref{sec:av} should 

3299: be chosen as

3300: $$ \pi = \bigl ( 1^{m_1 + m_{n-1}} 2^{m_2 + m_{n-2}} \dots (n-1)^{m_{n-1} + m_1}  \bigr ). 

3301: $$ 

3302: 

3303: The second partition $\rho$ should still be $(n^m).$ To define the two Lagrangians one 

3304: would need a generalization of the vanishing $\cpn$ construction, in the form of vanishing 

3305: Grassmannians $Gr(n,k),$ for $0 < k < n.$ The resulting Lagrangians should be 

3306: diffeomorphic to products of $m_k$ copies of the $Gr(n, k),$ over all $k.$ We expect that 

3307: many of the arguments in this paper would go through in this more general setting as well. 

3308: However, since the local models near degenerations are more complicated, it is possible 

3309: that additional technical difficulties could appear, especially in the proof of Markov $II$ 

3310: invariance.

3311: 

3312: More categorifications of quantum polynomial invariants were constructed by Gukov and 

3313: Walcher in \cite{GW}. They mainly work with representations of the Lie algebras $\so(n, 

3314: \cc),$ but also discuss representations of some exceptional Lie algebras. We expect that 

3315: one can define similar link homology invariants using Floer theory. One should use 

3316: transverse slices and adjoint orbits in a suitable Lie algebra, instead of those in 

3317: $\sl(nm, \cc).$

3318: 

3319: 

3320: \subsection {An involution} In some unpublished work (\cite{SS2}), Seidel and Smith define 

3321: an involution on their manifold $\yy_{m,2, \tau},$ and do Floer theory in the 

3322: fixed point set of that involution. (See also \cite[Section 7]{M}.) They show that the 

3323: resulting Floer homology groups are isomorphic to $\widehat{HF} (\Sigma(\lk) \# S^1 

3324: \times S^2),$ where $\widehat{HF}$ stands for a variant of Heegaard Floer homology 

3325: (\cite{OS0}), and $\Sigma(\lk)$ for the double cover of $S^3$ branched over the link 

3326: $\lk.$

3327: 

3328: There is a similar involution $\iota$ in our theory, for any $n \geq 2.$ On the slice 

3329: (\ref{smn}), it takes each block $Y_{k1}$ into its transpose $Y_{k1}^T.$ A linear algebra 

3330: exercise shows that $\iota$ does not change the conjugation class of a matrix, and 

3331: therefore descends to an involution on each $\ymnt.$ By doing all constructions in this 

3332: paper $\zz/2\zz$-equivariantly, we can arrange so that the involution also acts on the 

3333: Lagrangians $L$ and $L'.$ We can then try to apply Floer's theory to the fixed point sets 

3334: of $\iota|_L, \iota|_{L'}$, viewed inside the fixed point set of $\iota|_{\ymnt}.$ We 

3335: expect the resulting Floer cohomology groups to form a series of link invariants $\kr_{\nn 

3336: \iota}^* (\lk),$ for $n \geq 2,$ with the property that $\kr_{\nn \iota}^* (\text{unknot}; 

3337: \zz/2\zz) = H^{*+(n-1)/2}(\mathbb{RP}^{n-1}; \zz/2\zz).$

3338: 

3339: 

3340: \begin {thebibliography}{99999}

3341: 

3342: \bibitem {AB}

3343: M. Atiyah and R. Bielawski, {\it Nahm's equations, configurations spaces and flag 

3344: manifolds,} {Bull. Braz. Math. Soc.} {\bf 33} {(2002), 157--176.}

3345: 

3346: \bibitem {BM}

3347: W. Borho and R. MacPherson, {\it Partial resolutions of nilpotent varieties,} 

3348: {Ast\'erisque} {\bf 101-102} {(1983), 23--74.}

3349: 

3350: \bibitem {CP}

3351: C. De Concini and C. Procesi, {\it Symmetric functions, conjugacy classes, and the flag 

3352: variety,} {Invent. Math.} {\bf 64} {(1981), 203--219.}

3353: 

3354: \bibitem {DGR}

3355: N. Dunfield, S. Gukov and J. Rasmussen, {\it The superpolynomial for knot 

3356: homologies,} {math.GT/0505662, Experiment. Math., to appear.}

3357: 

3358: \bibitem{FN}

3359: E. Fadell and L. Neuwirth, {\it Configuration spaces,}

3360: Math. Scand. {\bf 10} (1962), 111--118.

3361: 

3362: \bibitem {F}

3363: A. Floer, {\it Morse theory for Lagrangian intersections, } {J. Diff.

3364: Geom.} {\bf 28} {(1988), 513--547.}

3365: 

3366: \bibitem {GW}

3367: S. Gukov and J. Walcher, {\it Matrix factorizations and Kauffman homology,} {preprint 

3368: (2005), hep-th/0512298.}

3369: 

3370: \bibitem {HOMFLY}

3371: P. Freyd, D. Yetter, J. Hoste, W. B. R. Lickorish, K. Millett and A. Ocneanu, 

3372: {\it A new polynomial invariant of knots and links,} {Bull. Amer. Math. Soc. (N.S.)} 

3373: {\bf 12} {(1985), no. 2, 239--246.}

3374: 

3375: \bibitem {FOOO}

3376: K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, {\it Lagrangian intersection Floer theory - anomaly 

3377: and obstruction,} preprint (2000), available at \url{www.math.kyoto-u.ac.jp/~fukaya/fukaya.html}

3378: 

3379: \bibitem {HT}

3380: D. Huybrechts and R. Thomas, {\it $\mathbb{P}$-objects and autoequivalences of 

3381: derived categories,}  Math. Res. Lett. {\bf 13} {(2006), no. 1, 87--98.} 

3382: 

3383: \bibitem {J}

3384: J. Jost, {\it Riemannian geometry and geometric analysis,} Springer Verlag, Berlin (2002).

3385: 

3386: \bibitem {Kh1}

3387: M. Khovanov, {\it A categorification of the Jones polynomial,} 

3388: {Duke Math. J.} {\bf 101} {(2000), no. 3, 359--426.}

3389: 

3390: \bibitem {Kh2} 

3391: M. Khovanov, {\it A functor-valued invariant of tangles,}  Algebr. Geom. 

3392: Topol. {\bf 2} {(2002), 665--741.}

3393: 

3394: \bibitem {KR1}

3395: M. Khovanov and L. Rozansky, {\it Matrix factorizations and link 

3396: homology,} {preprint (2004), math.QA/0401268.}

3397: 

3398: \bibitem {KR2}

3399: M. Khovanov and L. Rozansky, {\it Matrix factorizations and link

3400: homology II,} {preprint (2005), math.QA/0505056.}

3401: 

3402: \bibitem {Ko}

3403: M. Kontsevich, {\it Homological algebra of mirror symmetry,} in

3404: Proceedings of the International Congress of Mathematicians (Z\"urich,

3405: 1994), p. 120-139, Birkh\"auser, 1995.

3406: 

3407: \bibitem {KP}

3408: S. G. Krantz and H. R. Parks, {\it A primer of real analytic functions,} 2nd ed., Birkh\"auser, 

3409: Boston (2003).

3410: 

3411: \bibitem {Kr2}

3412: P. B. Kronheimer, {\it Instantons and the geometry of the nilpotent

3413: variety,} { J. Diff. Geom.} {\bf 32} {(1990), 473--490.}

3414: 

3415: \bibitem {KN}

3416: P. B. Kronheimer and H. Nakajima, {\it Yang-Mills instantons on ALE gravitational 

3417: instantons,} { Math. Ann.} {\bf 288} {(1990), 263--307.}

3418: 

3419: \bibitem {LM}

3420: H. B. Lawson and M.-L. Michelsohn, {\it Spin geometry,} Princeton University Press, 

3421: Princeton (1989).

3422: 

3423: \bibitem {Lo}

3424: S. Lojasiewicz, {\it Introduction to complex analytic geometry,} Birkh\"auser, Boston (1991).

3425: 

3426: \bibitem {Maf}

3427: A. Maffei, {\it Quiver varieties of type A,} Comment. Math. Helv. {\bf 80} (2005),  no. 1, 

3428: 1--27.

3429: 

3430: \bibitem {M}

3431: C. Manolescu, {\it Nilpotent slices, Hilbert schemes, and the Jones polynomial,} 

3432: Duke Math. J. {\bf 132} {(2006), no. 2, 311--369.}

3433: 

3434: \bibitem {McD}

3435: I. G. MacDonald, {\it Symmetric functions and Hall polynomials,} {Oxford University 

3436: Press, 2nd edition (1995).}

3437: 

3438: \bibitem {N1}

3439: H. Nakajima, {\it Instantons on ALE spaces, quiver varieties, and 

3440: Kac-Moody algebras,} {Duke Math. J.} {\bf 76} {(1994) no. 2, 365--416.}

3441: 

3442: \bibitem {N2}

3443: H. Nakajima, {\it Homology of moduli spaces of instantons on ALE spaces I,} {J. Diff. Geom.}

3444: {\bf 40} {(1997), 105--127.}

3445: 

3446: \bibitem {O}

3447: {\it On-Line Encyclopedia of Integer Sequences, } maintained by N. J. A. Sloane, entry A047888, 

3448: \url {http://www.research.att.com/projects/OEIS?Anum=A047888}

3449: 

3450: \bibitem {OS0}

3451: P. Ozsv\'ath and Z. Szab\'o, {\it Holomorphic disks and topological

3452: invariants for closed three-manifolds,} {Annals of Math. } {\bf 159} {(2004), no. 3,

3453: 1027--1158.}

3454: 

3455: \bibitem {OS}

3456: P. Ozsv\'ath and Z. Szab\'o, {\it Holomorphic disks and knot invariants,} 

3457: Adv. Math. {\bf 186} {(2004), no. 1, 58--116.}

3458: 

3459: \bibitem {P}

3460: M. Pozniak, {\it Floer homology, Novikov rings and clean intersections,}

3461: in {Northern California Symplectic Geometry Seminar, p. 119-181, Amer.

3462: Math. Soc., 1999.}

3463: 

3464: \bibitem {PT}

3465: J. H. Przytycki and P. Traczyk, {\it Invariants of links of Conway type,} {Kobe J. 

3466: Math.} {\bf 4} {(1988), no. 2, 115--139.}

3467: 

3468: \bibitem {R}

3469: J. Rasmussen, {\it Floer homology and knot complements,} {Ph. D. Thesis, Harvard

3470: University (2003), math.GT/0306378.}

3471: 

3472: \bibitem {RT}

3473: N. Reshetikhin and V. Turaev, {\it Ribbon graphs and their invariants derived from quantum 

3474: groups,} {Comm. Math. Phys.} {\bf 127} {(1990), no.1, 1--26.}

3475: 

3476: \bibitem {Ro}

3477: W. Rossmann, {\it Picard-Lefschetz theory for the coadjoint quotient of a semisimple Lie 

3478: algebra,} {Invent. Math.} {\bf 121}  {(1995), no. 3, 531--578.}

3479: 

3480: \bibitem {Sc}

3481: C. Schensted, {\it Longest increasing and decreasing subsequences, }

3482: {Canad. J. Math.} {\bf 13} {(1961), 179--191.}

3483: 

3484: \bibitem {Se}

3485: P. Seidel, {\it Graded Lagrangian submanifolds,} {Bull. Soc. Math. 

3486: France} {\bf 128} {(2000), 103--146.}

3487: 

3488: \bibitem {SS}

3489: P. Seidel and I. Smith, {\it A link invariant from the symplectic 

3490: geometry of nilpotent slices,} {preprint, math.SG/0405089.}

3491: 

3492: \bibitem {SS2}

3493: P. Seidel and I. Smith, {in preparation.}

3494: 

3495: \bibitem {Sl}

3496: P. Slodowy, {\it Simple singularities and simple algebraic groups,} {Lecture Notes in 

3497: Math.} {\bf 815,} Springer, Berlin (1980).

3498: 

3499: \bibitem {Sp}

3500: N. Spaltenstein, {\it The fixed point set of a unipotent transformation on the flag 

3501: manifold,} {Nederl. Akad. Wetensch. Proc. Ser. A} {\bf 79} {(1976), no. 5, 

3502: 452--456.}

3503: 

3504: \end{thebibliography}

3505: \end{document}

3506: 

3507: