1: \documentclass[10pt,a4paper]{article}

2: 

3: \usepackage[latin1]{inputenc}

4: \usepackage[T1]{fontenc}

5: \usepackage[english]{babel}

6: \usepackage{amsfonts}

7: \usepackage{euscript,amsthm,pstricks}

8: \usepackage[vcentermath]{youngtab}

9: 

10: \setlength{\textheight}{21.5cm}

11: 

12: \newtheorem{conj}{Conjecture}[section]

13: \newtheorem{theo}[conj]{Theorem}

14: \newtheorem{hypo}[conj]{Hypothesis}

15: \newtheorem{prop}{Proposition}[section]

16: \newtheorem{coro}[prop]{Corollary}

17: \newtheorem{lemm}{Lemma}[section]

18: \newtheorem{affi}[lemm]{Affirmation}

19: \newtheorem{clai}[lemm]{Claim}

20: \newtheorem{fait}[lemm]{Fact}

21: \newtheorem{defi}{Definition}[section]

22: \newtheorem{nota}[defi]{Notation}

23: \newtheorem{exem}[defi]{Example}

24: \newtheorem{naiv}[defi]{Naive guess}

25: \newenvironment{proo}{{\flushleft \bf Proof :}}{\hfill $\square$ \vspace{2mm}}

26: \newenvironment{rema}{{\flushleft \bf Remark :}}{\hfill}

27: \begin{document}

28: 

29: %\input /home/pe/latex/commandes_pe.tex

30: %\newcommand{\color}[1]{}

31: \input commandes_pe.tex

32: 

33: \title{Dual varieties of subvarieties of homogeneous spaces}

34: \author{Pierre-Emmanuel Chaput}

35: \maketitle

36: 

37: {\def\thefootnote{\relax}

38: \footnote{\hskip-0.6cm

39: {\it AMS mathematical classification \/}: 14N99, 14L35, 14L40. \\

40: {\it Key-words\/}: dual variety, homogeneous space, projective geometry.  }}

41: 

42: \begin{center}

43: {\bf Abstract}

44: \end{center}

45: 

46: If $X \subset \p^n_\C$ is an algebraic complex projective variety, one

47: defines the dual variety $X^* \subset {(\p^n)}^*$ as the set of

48: tangent hyperplanes to $X$.

49: The purpose of this paper is to generalise this notion when $\p^n$ is

50: replaced by a quite general partial flag variety. A similar

51: biduality theorem is proved, and the dual varieties of Schubert varieties

52: are described.

53: \lpara

54: 

55: \begin{center}

56: {\bf Introduction}

57: \end{center}

58: 

59: Let $X \subset \p V$ be a complex projective algebraic variety, with

60: $V$ a $\C$-vector space. If $h \in \p V^*$ is a hyperplane and $x \in X$

61: is a smooth point, we say that $h$ is tangent to $X$ at $x$ if $h$

62: contains the embedded tangent space of $X$ at $x$. Equivalently, the

63: intersection $X \cap h$ is singular at $x$. The closure of

64: the set of all $h \in \p V^*$

65: which are tangent at some smooth point of $X$

66: is denoted $X^*$ and called the

67: dual variety of $X$; for given $h \in X^*$, the closure

68: of the set of smooth points $x \in X$ such that $h$ is tangent at $x$

69: is called the tangency locus of $h$.

70: 

71: This notion of dual varieties is a very classical one, and it is used

72: plentifully in both classical and modern articles. The very powerfull

73: biduality theorem, to the effect that ${(X^*)}^* = X$,

74: and its corollary, which states that the tangency locus

75: at a smooth point $h \in X^*$ is a linear space, are ubiquitous.

76: To state only one example, this result is crucial in Zak's classification

77: of Severi varieties, since it allows proving that the entry locus of

78: a Severi variety is a smooth quadric \cite[proposition IV.2.1]{zak}.

79: 

80: This biduality

81: theorem deals with subvarieties of projective space, which have

82: been studied by so many classical algebraic geometers. More recently, work has

83: been done in a new direction which consists in

84: considering subvarieties of other homogeneous

85: spaces. For example, G. Faltings \cite{faltings}

86: and O. Debarre \cite{debarre1,debarre2} have shown

87: some connectivity theorems that hold

88: in an arbitrary homogeneous space, E. Arrondo has proved a classification

89: of some subvarieties of Grassmannians similar to Zak's result

90: \cite{arrondo_severi}, and some

91: topological results on low-codimensional subvarieties of some homogeneous

92: spaces emerge in works of E. Arrondo - J. Caravantes 

93: \cite{arrondo_low} and N. Perrin \cite{perrin_low}.

94: 

95: \lpara

96: 

97: Obviously, to study subvarieties of homogeneous spaces, a similar

98: notion of dual variety and a biduality theorem are lacking. The aim of

99: this article is to fill this gap as much as it is possible.

100: 

101: Since homogeneous spaces $G/P$ are by definition projective algebraic

102: varieties, it is certainly possible to embed them in a projective space, and

103: therefore a subvariety $X \subset G/P$ is a fortiori a

104: subvariety of a projective space, so that one can consider

105: the usual dual variety of $X$.

106: 

107: However I claim that in many cases this is not the best thing to

108: do. Let us consider an example. Let $V$ be a $\C$-vector space

109: equipped with a non-degenerate quadratic form. If $Q \subset \p V$ is 

110: the smooth

111: quadric it defines, then it is well-known that $Q^* \subset \p V^*$ is also a

112: smooth quadric, canonically isomorphic with $Q$. 

113: Now, let $r$ be an integer and let us consider the variety $\G_Q(r,V)$

114: parametrising $r$-dimensional isotropic subspaces as a subvariety of

115: a suitable projective space using Plücker embedding. Then clearly we no

116: longer have $\G_Q(r,V)^* \simeq \G_Q(r,V)$. On the contrary,

117: let us consider $\G_Q(r,V)$ as a subvariety of the Grassmannian

118: $\G(r,V)$;

119: proposition \ref{prop_symmetric}

120: shows that for my definition of dual varieties, $\G_Q(r,V) \subset \G(r,V)$ 

121: has a well-defined

122: dual variety in $\G(r,V^*)$ which is canonically isomorphic with

123: $\G_Q(r,V)$.

124: 

125: In fact, homogeneous spaces are often minimally

126: embedded in projective spaces of very big dimension, so that the usual

127: dual variety of a subvariety of a homogeneous space will happen to be

128: very large and often untractable. A notion of dual varieties within

129: homogeneous spaces is probably more suitable if one wants to deal with

130: low-dimensional or low-codimension subvarieties (of course, the price to

131: pay is that the ambient space is a bit more complicated than a projective

132: space).

133: 

134: \lpara

135: 

136: My definition of dual varieties uses a class of birational

137: transformations called stratified Mukai flops by Namikawa

138: \cite{namikawa}. These are birational maps

139: $\mu:T^*G/P \dasharrow T^*G/Q$ defined in terms of nilpotent orbits

140: for some semi-simple group $G$ and some parabolic subgroups

141: $P,Q$. For

142: given $G,P,Q$, if there exists such a map, then we say that

143: $G/P$ and $G/Q$ allow

144: duality. For $X \subset G/P$, we consider its conormal bundle 

145: $N^*X \subset T^*G/P$ and define the dual variety 

146: $X^Q = \pi_Q \circ \mu (N^*X) \subset G/Q$ ($\pi_Q : T^* G/Q

147: \rightarrow G/Q$ denotes the projection) if $N^*X$ meets the locus

148: where $\mu$ is defined (in which case we say that $X$ is suitable). 

149: For example, if $G = SL(V)$ and

150: $G/P = \p V$, the only possibility for $Q$

151: leads to $G/Q = \p V^*$; any

152: proper subvariety $X \subset \p V$ will be suitable and $X^Q = X^*$.

153: Another example is the fact that a Grassmannian $\G(r,V)$ and its dual

154: Grassmannian $\G(r,V^*)$ allow duality, as one could naturally expect.

155: 

156: One advantage of this definition is that it uses the so-called

157: Springer resolutions of the corresponding nilpotent orbit, which are

158: symplectic resolutions, and this

159: article uses heavily informations which come from the study of such

160: resolutions \cite{namikawa,flop_scorza}.

161: Another advantage is that it exhibits

162: the symplectic nature of dual varieties. In fact, 

163: $T^*G/P$ and $T^*G/Q$

164: are symplectic varieties and $N^*X$, as a subvariety of

165: $T^* G/P$, is a Lagrangian subvariety. These properties suffice to

166: show very easily the biduality theorem \ref{theo_bidualite} in our

167: setting.

168: 

169: \lpara

170: 

171: However, this definition also has its drawbacks. The most important is

172: probably that it is not so much intuitive, so that given $x \in X$ and

173: $h \in X^Q$, it is not obvious at all what the sentence ``$h$ is

174: tangent to $X$ at $x$'' should mean. However, 

175: in the case of a Grassmannian,

176: using the natural rational 

177: map $Hom(\C^r,V) \dasharrow \G(r,V)$, I show that the

178: dual variety of $X \subset \G(r,V)$ can be computed in terms of the usual

179: dual variety of an adequate subvariety of $\p Hom(\C^r,V)$ (see subsection

180: \ref{subsection_grassmannienne}).

181: Therefore, this is a way of understanding more easily

182: dual varieties in the case of Grassmannians.

183: In general however, there are two fundamental

184: differences between our setting and usual duality.

185: 

186: First of all, given $G,P$, there may be many different $Q$'s, or none,

187: such that $G/P$ and $G/Q$ allow duality. Therefore, given suitable 

188: $X \subset G/P$, we will get a dual variety $X^Q$ for each such $Q$. If one

189: restricts to maximal parabolic subgroups, thanks to

190: \cite{namikawa}, this difficulty disappears

191: because for given $G/P$ there will be at most one parabolic subgroup 

192: $Q$ such that

193: $G/P$ and $G/Q$ allow duality. Moreover, section

194: \ref{section_reduction} shows that one can understand all dual

195: varieties if they are understood when $P$ and $Q$ are maximal parabolic

196: subgroups. These cases are therefore called fundamental cases. They

197: include the duality between the Grassmannian $\G(r,V)$ and its dual

198: Grassmannian $\G(r,V^*)$, but also a duality

199: between the two spinor varieties of a quadratic space of dimension

200: $4p+2$, and between the exceptional homogeneous spaces 

201: $E_6/P_1 \leftrightarrow E_6/P_6$ and $E_6/P_3 \leftrightarrow E_6/P_5$.

202: 

203: The second difference is that not all proper subvarieties 

204: $X \subset G/P$ will have a dual variety. Note that $X = \p V$ has no

205: dual variety in $\p V^*$, because for any $x \in X$,

206: no non-zero cotangent form can vanish

207: on $T_xX$. From this point of view, the situation is

208: quite similar in our setting~: too big subvarieties $X$ of $G/P$ don't have

209: dual varieties because for any $x \in X$

210: there is no generic cotangent form in $T^*_x G/P$ which

211: vanishes on $T_xX$.

212: 

213: \lpara

214: 

215: In the classical setting, a hyperplane $h$ is tangent to $X$ at $x$

216: \iff the intersection $h \cap X$ is singular.

217: As I already alluded to, I have not been able to give a similar

218: geometric notion of ``tangent element''. The only sensible definition

219: seemed to state that $h \in X^Q$ is tangent to $X$ at $x \in X$ if $h$

220: belongs to the image of $N^*_xX$ under $\pi_Q \circ \mu$.

221: Since there is an incidence

222: variety in $G/P \times G/Q$ (the closed $G$-orbit), any $h \in G/Q$

223: still defines, exactly as in the classical situation, a subvariety $I_h

224: \subset G/P$. Lemma \ref{lemm_q} implies that if $h$ is tangent to $x$

225: at $X$, then the intersection $I_h \cap X$ is not transverse, but the

226: reciprocal of this fact is false.

227: 

228: Section \ref{section_tangency} deals with this matter. Corollary

229: \ref{coro_incident} states that if $h$ is tangent to $X$ at $x$, then

230: $x \in I_h$. For $x \in X$ with $X$

231: suitable, the tangent cone $\overline{T_xX} \subset G/P$ of $X$ at $x$

232: is defined in a roundabout manner as the dual variety 

233: of the variety of $h$'s in $X^Q$ which are

234: tangent to $X$ at $x$. Theorem \ref{theo_cone} implies that

235: $\overline{T_xX}$ is a ``cone'' with vertex $x$, where

236: definition \ref{def_cone}

237: generalises the classical notion of

238: cones from subvarieties of projective space to subvarieties of

239: fundamental homogeneous spaces.

240: 

241: \lpara

242: 

243: Finally, section \ref{section_exemple} studies dual

244: varieties of Schubert varieties. In the classical setting, the dual

245: variety of a linear subspace is again a linear subspace. In our

246: setting, it is a formal consequence of the definitions that the dual

247: variety of a Schubert variety is again a Schubert

248: variety (see proposition

249: \ref{prop_schubert} which relies on the functorial property of dual

250: varieties given in proposition \ref{foncteur}).

251: 

252: Let $B \subset G$ be a Borel subgroup. It turns out that the combinatorial

253: involution $X \mapsto X^Q$ between $B$-stable suitable

254: Schubert subvarieties of $G/P$ and $G/Q$

255: is no longer decreasing, as it was the case for $G/P = \p V$.

256: For this reason, the description of this map is quite intricate. In

257: the case of Grassmannians and spinor varieties, we give explicitly

258: in terms of

259: partitions the map $X \mapsto X^Q$,

260: see propositions \ref{grass_adapte} and \ref{prop_spinoriel}.

261: This relies on a general

262: recipy for finding $X^Q$ when $X$ is a Schubert variety

263: which is given in subsection

264: \ref{subsection_combinatoire}. For the exceptional cases, this recipy

265: theoretically defines the involution (there is only a finite number of

266: calculations to do to compute the dual variety of a Schubert

267: subvariety), but I will not give a more

268: explicit description of it. As a first step, I describe a criterion

269: for a Schubert subvariety to be suitable. Remarkably enough, this

270: criterion can be stated in a uniform way for all the fundamental

271: cases, using the combinatorics of some quivers studied in

272: \cite{carquois}~: see theorem \ref{theo_adapte}.

273: 

274: \lpara

275: {\bf Further questions :}

276: Of course this study only gives basic properties of our generalised

277: dual varieties~: if one compares with usual dual varieties, what

278: essentially has been proved is the biduality theorem and the

279: computation of the dual variety of a quadric and a linear subspace. The

280: power of the classical notion of dual varieties gives hope to me that much more

281: can be said on this topic, including~:

282: \begin{itemize}

283: \item

284: Is it true that for a smooth subvariety $X \subset G/P$

285: one has $\dim\ X^Q \geq \dim\ X$ ? This question has been raised by

286: Laurent Manivel.

287: \item

288: Many Fano 3-folds are defined as subvarieties of some homogeneous

289: spaces. What are the dual varieties of these Fano 3-folds ?

290: \item

291: What is the dual variety of a divisor in $G/P$ ? If this is a divisor,

292: what is the degree of this divisor ? The

293: answer to this question for $G/P = \G(2,V)$ or $G/P = E_6/P_1$ and a

294: divisor of degree 1 has been given in \cite{hermitian} (the dual

295: variety is again a divisor of degree 1).

296: \item

297: Classification problems~: for example find all smooth varieties

298: with dual variety a divisor of low degree.

299: \end{itemize}

300: 

301: \lpara

302: \noindent

303: {\bf Acknowledgements : }I am very grateful to M. Brion for suggesting

304: that maybe nilpotent orbits could help defining an

305: interesting equivariant rational map

306: $T^*G/P \dasharrow G/Q$, as it was finally exactly the case. Thanks

307: are also due to B. Fu who pointed to me the reference \cite{namikawa}.

308: 

309: \tableofcontents

310: 

311: \sectionplus{Definition of the dual variety}

312: 

313: \label{section_definition}

314: 

315: \subsectionplus{Notations and definition}

316: 

317: In this subsection, I introduce the (abstract) definition of dual 

318: varieties, which allows easy proofs of general results; in subsection

319: \ref{subsection_grassmannienne}, an equivalent but more ``down-to-earth'' 

320: definition will be given in the case of Grassmannians.

321: 

322: \lpara

323: 

324: Before giving this definition, which is not so intuitive, I give some

325: ``naive guesses'' and explain why the corresponding notion of dual

326: varieties would not be interesting. In this way, I hope to convince

327: the reader that it is not possible to avoid some technicalities.

328: Let us try our unsuccessful experiments in the case of Grassmannians.

329: 

330: So assume $G/P = \G(r,V)$ and $G/Q = \G(r,V^*)$ and assume

331: $2r<\dim V$. Any element

332: $h \in \G(r,V^*)$, representing a codimension $r$ 

333: subspace of $V$ denoted $L_h$, defines (at

334: least) two subvarieties in $\G(r,V)$. The first (resp. the second)

335: is the subvariety of $x \in \G(r,V)$ such that

336: $L_x \subset L_h$ (resp. $\dim(L_x \cap L_h)>0$). It will be denoted

337: $I_h$ (resp. $h^\bot$).

338: Assume $X \subset \G(r,V)$ is a subvariety and $x \in X$.

339: In the following, we give some naive definitions of the fact that $h$

340: is tangent to $X$ at $x$ in terms of the intersection of $X$, $h^\bot$

341: and $I_h$.

342: 

343: \begin{naiv}

344: ``$h$ is tangent to $X$ at $x$ if $x \in I_h$

345: and the intersection $h^\bot \cap X$ is

346: singular at $x$.''

347: \end{naiv}

348: \noindent

349: This is really stupid, because if $x \in I_h$, then

350: $h^\bot$ is singular at $x$, and so is the intersection

351: $h^\bot \cap X$. So any $h$

352: such that $L_x \subset L_h$ will satisfy this condition, regardless to

353: the tangent space $T_xX$.

354: 

355: \begin{naiv}

356: ``$h$ is tangent to $X$ at $x$ if $x \in h^\bot$

357: and the intersection $h^\bot \cap X$ is

358: singular at $x$.''

359: \end{naiv}

360: \noindent

361: For the same reason as above, it suffices that $L_h$ contains $L_x$ in

362: order that this condition holds. So if we define $X^*$ as the set of

363: $h$'s satisfying the above condition, we will not have a biduality

364: theorem. In fact, if for example $X=\{x\}$ is a point, then $X^*$ will

365: contain $\{h:L_h \supset L_x\}$ and ${(X^*)}^*$ will certainly not be

366: reduced to $\{x\}$.

367: 

368: \lpara

369: 

370: Therefore, it seems necessary to use the smooth subvariety $I_h$. In

371: this case, assuming that $I_h \cap X$ is singular is not quite

372: acurate, because $I_h$ has codimension larger than 1, so this

373: condition should be replaced by the fact that the intersection is not

374: transverse~:

375: \begin{naiv}

376: ``$h$ is tangent to $X$ at $x$ if $x \in I_h$ and the

377: intersection $I_h \cap X$ is not transverse at $x$.''

378: \end{naiv}

379: \noindent

380: Again, if we take $X = \{x\}$, then 

381: $X^* = \{h : L_h \supset L_x \}$, and

382: ${(X^*)}^* = \{y : \dim(L_x \cap L_y) > 0 \}$. So we don't have a

383: biduality theorem.

384: 

385: \lpara

386: 

387: Of course we could multiply such definitions; let us just consider one

388: more possibility~:

389: \begin{naiv}

390: ``$X^*$ is the intersection of the usual dual variety of $X$ (in the

391: Plücker embedding) with $\G(r,V^*)$.''

392: \end{naiv}

393: \noindent

394: Already in case $r=2$ and $\dim V$ even, it is easy to see that

395: biduality will not hold. Let again $X = \{x\}$. The usual dual variety

396: of $X$ in the Plücker embedding is a hyperplane; therefore $X^*$ will

397: be a hyperplane section of $\G(2,V^*)$. As it is well-known,

398: the dual variety of

399: $\G(2,V^*) \subset \p \wedge^2 V^*$ is a hypersurface in 

400: $\p \wedge^2 V$. Therefore it follows that the usual dual variety of

401: $X^* \subset \p \wedge^2 V^*$ 

402: will have codimension at most 2 in $\p \wedge^2 V$. Thus

403: its intersection ${(X^*)}^*$ with $\G(2,V)$ cannot be a point.

404: 

405: \para

406: 

407: I hope that the previous unsuccessfull experiments will convince the

408: reader to accept a more conceptual definition of generalised dual

409: varieties.

410: Let $G$ be a semi-simple simply-connected complex algebraic group with

411: Lie algebra $\g$, and let $\g^*$ be the dual vector space of $\g$.

412: We fix $T\subset B \subset G$ a

413: maximal torus and a Borel subgroup of $G$.

414: If $P \subset G$ is a

415: parabolic subgroup, let $G/P$ denote the corresponding flag variety.

416: If $X$ is a variety, let $T^*X$ denote its cotangent bundle;

417: we denote $t_P:T^*G/P \rightarrow \g^*$ the natual map.

418: 

419: \begin{defi}

420: Let $P,Q$ be parabolic subgroups of $G$.

421: We say that $G/P$ and $G/Q$ allow duality if there is a nilpotent

422: orbit ${\cal O} \subset \g^*$ such that 

423: $t_P:T^*G/P \rightarrow \g^*$ and $t_Q:T^*G/Q \rightarrow \g^*$ are birational

424: isomorphisms between the cotangent bundles and $\cal O$.

425: \label{defi_dualite}

426: \end{defi}

427: 

428: Assume that $G/P$ and $G/Q$ allow duality. The birational map

429: $t_Q^{-1} \circ t_P : T^*G/P \dasharrow T^* G/Q$ will be denoted

430: $\mu$. Let $\co$ be such that $t_P(T^*G/P) = t_Q(T^*G/Q) = \overline \co$.

431: Let $X \subset G/P$ be any subvariety. Let $X^{sm}$ denote its smooth

432: locus and let 

433: $N^*X \subset T^* G/P$ denote the conormal bundle to $X^{sm}$~: we have

434: $(x,f) \in N^*X$ \iff $x \in X^{sm}$,

435: $f \in T^*_xG/P$, and $f_{|T_xX} = 0$.

436: 

437: \begin{defi}\ 

438: \label{defi_adapte}

439: \begin{itemize}

440: \item

441: A form $f \in T^*G/P$ (resp. $f\in T^*G/Q$) is called generic if it

442: belongs to $t_P^{-1}(\co)$ (resp. $t_Q^{-1}(\co)$).

443: \item

444: A subvariety $X \subset G/P$ 

445: is suitable if it is irreducible and there are generic forms

446: in $N^*X$.

447: \item

448: A point $x$ of a suitable variety $X$ is itself suitable if there

449: are generic forms in $N^*_xX$.

450: Let $X^s$ denotable the locus of suitable

451: points of $X$.

452: \end{itemize}

453: \end{defi}

454: \rek

455: One could also consider reducible suitable subvarieties : they

456: would be subvarieties such that every irreducible component is

457: suitable; we could then define the dual variety of a reducible suitable

458: variety as the union of dual varieties of its irreducible

459: components.

460: 

461: \begin{nota}

462: Let $\pi_P$ denote the projection $T^* G/P \rightarrow G/P$.

463: \end{nota}

464: 

465: \begin{defi}

466: If $X \subset G/P$ is suitable then we define $X^Q \subset G/Q$ as

467: the image of $N^*X$ by the rational map

468: $\pi_Q \circ \mu$.

469: \end{defi}

470: 

471: In the rest of the article, $P$ and $Q$ will denote parabolic

472: subgroups of a reductive simply-connected group $G$ allowing duality.

473: Moreover, we denote

474: $p := \pi_P \circ \mu\ : T^* G/Q \dasharrow G/P$,\ 

475: $q := \pi_Q \circ \mu^{-1}\ : T^* G/P \dasharrow G/Q$ 

476: the relevant rational maps. Finally, let

477: $\co \subset \g^*$ be the $G$-orbit which is dense in 

478: $t_P(T^* G/P) = t_Q(T^* G/Q)$.

479: 

480: \begin{defi}\ 

481: \label{i_h}

482: \begin{itemize}

483: \item

484: Let $x \in X \subset G/P$. We say that $h \in G/Q$ is tangent to $X$

485: at $x$ if $h \in q(N^*_x X)$.

486: \item

487: If $h \in G/Q$, let $I_h$ denote the Schubert variety of elements in

488: $G/P$ which are incident to $h$, in the sense that $x$ is incident to

489: $h$ if the intersection of the stabilisors of $x$ and $h$ (in $G$)

490: contain a Borel subgroup.

491: \item

492: As a corollary of Borel's conjugacy theorem, $I_h$ is homogeneous

493: under the stabilisor of $h$.

494: \end{itemize}

495: \end{defi}

496: 

497: \noindent

498: The notion of tangency will be studied in more details in subsection

499: \ref{section_tangency}. Here we only remark the following~:

500: 

501: \begin{fait}

502: \label{fait_tangent}

503: If $h$ is tangent to $X$ at $x$, then the intersection

504: $I_h \cap X$ is not transverse at $x$.

505: \end{fait}

506: \noindent

507: The proof of this fact is postponed to section

508: \ref{section_tangency}~: see lemma \ref{lemm_q}.

509: Note that the converse does not hold in general, contrary

510: to the case when $G/P = \p V$.

511: 

512: 

513: 

514: 

515: \subsectionplus{Fundamental cases}

516: 

517: \label{subsection_fondamental}

518: 

519: \begin{defi}

520: Let $P,Q \subset G$ allow duality. We say that $P,Q,G$ is a

521: fundamental case if one of the following hold :

522: \begin{itemize}

523: \item

524: $G=SL_n$, $P$ and $Q$ are the stabilisors of supplementary subspaces

525: of $\C^n$.

526: \item

527: $G=Spin_{4p+2}$, $P$ and $Q$ are the stabilisors of supplementary

528: (and so of different families) isotropic subspaces of $\C^{4p+2}$.

529: \item

530: $G$ is of type $E_6$, and, with Bourbaki's notations 

531: \cite[p.261]{bourbaki} $(P,Q)$ correspond either to the roots 

532: $(\alpha_1,\alpha_6)$ or $(\alpha_3,\alpha_5)$.

533: \end{itemize}

534: If this holds, $G/P$ and $G/Q$ are called fundamental homogeneous spaces.

535: \label{fondamental}

536: \end{defi}

537: 

538: By \cite[theorem 6.1]{namikawa}, these examples are all the examples of 

539: maximal parabolic subgroups

540: allowing duality. Recall that the corresponding rational map

541: $\mu : T^* G/P \dasharrow T^* G/Q$ is called a stratified Mukai flop.

542: 

543: Moreover, in all the other cases, the rational map

544: $\mu : T^* G/P \dasharrow T^* G/Q$ (and, as we will see in subsection 

545: \ref{subsection_sequence}, the dual varieties) may be described using 

546: only fundamental stratified Mukai flops~: let us recall

547: this construction \cite[theorem 6.1]{namikawa}. 

548: Assume $P,Q \subset G$ are parabolic subgroups

549: included in a common parabolic subgroup

550: $R$. Then we have fibrations

551: $$

552: \begin{array}{ccccc}

553: G/P & & & & G/Q\\

554: & f_P\searrow & & \swarrow f_Q\\

555: && G/R

556: \end{array}

557: $$

558: with fibers $R/P$ and $R/Q$. Let $U(R)$ denote the unipotent radical of

559: $R$ and $Z(R)$ its connected center; let $L=R/Z(R)U(R)$; $R/U(R)$

560: is isomorphic with

561: a levi factor of $R$ and $L$ is semi-simple.

562: Moreover, $R/P$ and $R/Q$ are $L$-homogeneous varieties~:

563: denote $\pi : R \rightarrow L$ the projection, and denote

564: $P_L:=\pi(P)$ (resp. $Q_L:=\pi(Q)$) we have

565: $R/P \simeq L/P_L$ and $R/Q \simeq L/Q_L$. Assume now that $P_L,Q_L$

566: allow duality. Therefore there is a rational map

567: $\mu_L : T^* L/P_L \dasharrow T^* L/Q_L$ 

568: which can be used to define the stratified Mukai flop.

569: 

570: In fact, let $z \in G/R$, and denote $\cf_z:=f_P^{-1}(z)$

571: (resp. $\cg_z:=f_Q^{-1}(z)$), and let $i_z:\cf_z \rightarrow G/P$

572: (resp. $j_z:\cg_z \rightarrow G/Q$) be the natural inclusions. We have

573: $\cf_z \simeq L/P_L$ and $\cg_z \simeq L/Q_L$.

574: Let $L_z = R_z/Z(R_z)U(R_z)$ denote the group

575: isomorphic with $L$ acting on $\cf_z$ and $\cg_z$.

576: Because $\mu_L$ is canonical, it defines an algebraic family

577: of rational maps

578: $\mu_z : T^* \cf_z \dasharrow T^*\cg_z$ parametrised by $G/R$.

579: Now, if $\alpha$ is an element of $T^* G/P$,

580: say $\alpha \in T^*_x G/P$ with $x \in G/P$, then we can restrict this

581: linear form

582: to $T_x \cf_{f_P(x)}$; this gives an element in the bundle

583: $T^*\cf_{f_P(x)}$ which we denote $f_x$.

584: Finally, recall that $\pi_P:T^* G/P \rightarrow G/P$ and

585: $\pi_Q:T^* G/Q \rightarrow G/Q$ denote the bundle projections.

586: With these notations we have the following proposition~:

587: 

588: \begin{prop}

589: \label{prop_fibration}

590: If $f \in T_x^* G/P$ belongs to the open $G$-orbit, then

591: $f_x \in T^*\cf_{f_P(x)}$ belongs to the open $L_{f_P(x)}$-orbit, and

592: $\pi_Q(\mu(f)) = j_{f_P(x)} \circ \pi_{Q_L} \circ \mu_x(f_x)$.

593: \end{prop}

594: \noindent

595: Then, using \cite[theorem 4.1]{flop_scorza},

596: one can deduce a description of the flop

597: $T^* G/P \dasharrow T^* G/Q$.

598: \begin{proo}

599: Since both maps of the proposition are equivariant,

600: we can assume that $x$ corresponds to the base point in $G/P$.

601: If the restriction of $f$ to $T_x \cf_{f_P(x)}$ would belong to a

602: closed $L$-stable strict subvariety of $T^* \cf_{f_P(x)}$, then forms

603: in the

604: $G$-orbit of $f$ would restrict to non generic forms; therefore this

605: $G$-orbit could not be dense in $T^* G/P$.

606: 

607: Let $\u({\mathfrak r})$ and $\z({\mathfrak r})$ denote the

608: nilpotent part and the centraliser of the Lie algebra

609: ${\mathfrak r}$ of $R$.

610: Let $\plie$ be the Lie algebra of $P$. Under $t_P$,

611: $f$ is mapped to an element in $\g^*$ which is orthogonal to $\plie$

612: and therefore to

613: $\u({\mathfrak r}) \oplus \z({\mathfrak r})$. It thus defines an

614: element $\overline f$ in ${\mathfrak l}^*$, if ${\mathfrak l}$ denotes

615: the Lie algebra of $L$. If $y \in L/Q_L$ denotes the element

616: $\mu_L(\overline f)$, then, by definition of the Mukai flop,

617: $\overline f$ is orthogonal to

618: $\overline \q_y$ 

619: ($\overline \q_y$ denotes the Lie algebra of the stabilisor of $y$ in

620: $L$). Thus it follows that $f$ vanishes on $\q_{j(y)}$, the Lie

621: algebra of the stabiliser of $j(y)$ in $G/Q$. Therefore, $y$ equals

622: $\pi_Q \circ \mu(f)$.

623: \end{proo}

624: 

625: 

626: Now, \cite[theorem 6.1]{namikawa} states that for any pair $(P,Q)$

627: of parabolic

628: subgroups allowing duality, we can find a chain 

629: $(P_0=P,P_1,\ldots,P_n=Q)$ of parabolic subgroups such

630: that all the pairs $(P_i,P_{i+1})$ are as above and the corresponding

631: pair $P_L,Q_L \subset L$ is a fundamental case. Therefore, the description of

632: stratified Mukai flops in the fundamental cases is enough to understand

633: all stratified Mukai flops. As we will see in section \ref{section_reduction},

634: the same is true as far as dual varieties are concerned.

635: 

636: 

637: \subsectionplus{Recallections about fundamental homogeneous spaces}

638: 

639: \label{subsection_notation}

640: 

641: We now introduce some notations and recall some results

642: for fundamental homogeneous spaces

643: which will be used throughout the article.

644: In particular, we give in each case

645: a simple way of understanding the rational map

646: $q : T^* G/P \dasharrow G/Q$.

647: 

648: Let $r$ and $n$ be integers with $2r<n$.

649: The Grassmannian parametrising $r$-linear subspace of a fixed

650: $n$-dimensional vector space $V$ will be denoted

651: $\G(r,V)$. The dual Grassmannian, parametrising

652: codimension $r$ subspaces of $V$, will be denoted $\G(r,V^*)$.

653: Let $x \in \G(r,V)$; it represents a linear subspace of $V$ which will

654: be denoted $L_x$. Moreover, we have a natural identification

655: $T^*_x \G(r,V) \simeq Hom(V/L_x,L_x)$. If 

656: $\varphi \in Hom(V/L_x,L_x)$ is generic (that is, of rank $r$), then

657: its kernel is a codimension $r$ subspace of $V$ containing $L_x$. In

658: fact, we have $q(\varphi) = \ker \varphi$.

659: 

660: \lpara

661: 

662: Let $p$ be an integer. Let $V$ be a vector space of dimension $4p+2$,

663: equipped with a quadratic form. In case we need a basis for $V$, we

664: will take a hyperbolic one, of the form 

665: $(e_1^+,\ldots,e_{2p+1}^+,e_1^-,\ldots,e_{2p+1}^-)$, such that the

666: quadratic form is given by

667: $Q(\sum x_i^+e_i^+ + \sum x_i^-e_i^-) = \sum x_i^+x_i^-$.

668: Recall that the variety parametrising isotropic subspaces of $V$ of

669: dimension $2p+1$ has two connected components, which will be denoted

670: $G/P = \G_Q^+(2p+1,4p+2)$ and $G/Q = \G_Q^-(2p+1,4p+2)$. As in the

671: case of Grassmannians, for $x \in \G_Q^+(2p+1,4p+2)$ and

672: $h \in \G_Q^-(2p+1,4p+2)$, we denote $L_x,L_h$ the corresponding

673: isotropic subspaces. The relation $x \in I_h$ amounts to

674: $\dim(L_x \cap L_h) = 2p$. Given $x \in \G_Q^+(2p+1,4p+2)$ and

675: $L \subset L_x$ of dimension $2p$,

676: there is exactly one $h \in \G_Q^-(2p+1,4p+2)$ such that

677: $L_x \cap L_h = L$~: this yields a natural isomorphism between

678: $I_x$ and $\p L_x^*$.

679: 

680: The map $q$ may be defined as follows. Let $x \in \G_Q^+(2p+1,4p+2)$;

681: the cotangent space $T_x^* \G_Q^+(2p+1,4p+2)$ identifies with

682: $\wedge^2 L_x$. If $\omega \in \wedge^2 L_x$ is a skew form of rank

683: $2p$, let $L_\omega$ be its image. It is a hyperplane in $L_x$;

684: therefore it defines a unique element $h \in \G_Q^-(2p+1,4p+2)$

685: such that $L_x \cap L_h = L_\omega$. We have

686: $q(\omega) = h$.

687: 

688: \lpara

689: 

690: As far as the exceptional group $E_6$ is concerned, we denote $V_i$

691: the $i$-th fundamental representation of $E_6$, so that

692: $E_6/P_i \subset \p V_i$. We have $V_6 = V_1^*$ and $V_5 = V_3^*$. In

693: terms of this embedding, an element $h \in \p V_1^*$ belongs to

694: $E_6/P_6$ \iff it contains the linear span of two tangent spaces

695: $T_xE_6/P_1,T_yE_6/P_1$, for some distinct $x,y \in E_6/P_1$.

696: 

697: We refer to \cite{flop_scorza} for the

698: proofs of the following results. Let $x \in E_6/P_1$. The cotangent

699: space $T_x^* E_6/P_1$ identifies with $\oc \oplus \oc$, if $\oc$

700: denotes the algebra of complexified octonions, an 8-dimensional 

701: non-associative and non-commutative algebra over $\C$. This algebra is a

702: normed algebra~: there is a quadratic form $N:\oc \rightarrow \C$ such

703: that $N(z_1z_2) = N(z_1)N(z_2)$ for all $z_1,z_2 \in \oc$. The variety

704: $I_x$ is an 8-dimensional smooth quadric. It is convenient to denote

705: $Z = \C \oplus \oc \oplus \C$ a 10-dimensional space, equipped with

706: the quadratic form $Q(t,z,u) = tu - N(z)$. Then $I_x$ is the smooth

707: quadric defined by $Q$ and $q$ is defined by

708: $q((z_1,z_2)) = [N(z_1):z_1\overline z_2:N(z_2)] \in \p Z$

709: \cite[theorem 3.3 and corollary 3.2]{flop_scorza}.

710: 

711: \lpara

712: 

713: To visualise the homogneous space $E_6/P_3$ (resp. $E_6/P_5$), 

714: we use the fact

715: that its points parametrise

716: projective lines included in $E_6/P_1$ (resp. $E_6/P_6$)

717: \cite[theorem 4.3 p.82]{landsberg}.

718: To avoid confusions between points in $E_6/P_1$ and

719: $E_6/P_3$, we will denote the latters with greek letters.

720: Since the marked Dynkin diagrams of $E_6/P_3$ and $E_6/P_5$ are

721: respectively

722: $\dynkinep{9}{4}{2}$ and $\dynkinep{9}{4}{6}$, we see that for

723: $\kappa \in E_6/P_5$, $I_\kappa \simeq \G(2,5)$. Let us describe this

724: isomorphism $I_\kappa \simeq \G(2,5)$ more explicitly, since this will be

725: needed to describe the rational map $q$.

726: If $\alpha \in E_6/P_3$, we will denote $l_\alpha \subset E_6/P_1$ the

727: corresponding

728: line and $L_\alpha$ the linear subspace it represents. 

729: By \cite[proposition 3.6]{flop_scorza}, the span of 

730: the affine tangent spaces

731: $\widehat{T_x E_6/P_1}$ in $V_1$ for $x$ in

732: $l_\alpha$ is a 22-dimensional linear subspace in $V_1$ denoted 

733: $S_\alpha$.

734: Therefore, any 25-dimensional space which contains $S_\alpha$ defines a

735: pencil of hyperplanes which belong to $E_6/P_6 \subset \p V_1^*$, that

736: is, a point in $E_6/P_5$. Denoting $Q_\alpha = V_1/S_\alpha$, 

737: this shows that

738: $I_\alpha = \G(3,Q_\alpha) \simeq \G(2,5)$. Dually, for 

739: $\beta \in E_6/P_5$, $I_\beta \simeq \G(2,W_\beta)$, 

740: where $W_\beta \subset V_1$ is a

741: 5-dimensional linear subspace such that 

742: $\p W_\beta \subset E_6/P_1$.

743: 

744: A Levi factor of $P$

745: contains $L' \simeq SL_2 \times SL_5$, and $L_\alpha$ (resp. $Q_\alpha$)

746: is the natural representation of $SL_2$ (resp. $SL_5$).

747: These representations are usefull describing $T^* E_6/P_3$~:

748: let $[e] \in E_6/P_3$ denote the base point; according to

749: \cite[propositions 3.6 and 3.7]{flop_scorza},

750: $T^*_{[e]} E_6/P_3$ is no longer an irreducible

751: $L'$-module, but there are exact sequences of $L'$-representations

752: \begin{equation}

753: \label{suite_exacte}

754: 0 \rightarrow L_\alpha^* \otimes \wedge^2 Q_\alpha \rightarrow

755: T_{[e]} E_6/P_3 \rightarrow Q_\alpha^* \rightarrow 0

756: \end{equation}

757: $$

758: \label{suite_exacte_cotangent}

759: 0 \rightarrow Q_\alpha \rightarrow

760: T^*_{[e]} E_6/P_3 \stackrel{\pi}{\rightarrow}

761: L_\alpha \otimes \wedge^2 Q_\alpha^*

762: \rightarrow 0.

763: $$

764: 

765: We now describe the rational map $q$.

766: Choose a base

767: $e_1^*,e_2^*$ (resp. $f_1,\ldots,f_5$) of $L_\alpha^*$ (resp. 

768: $Q_\alpha$). The

769: rational map $q:T_{[e]}^* E_6/P_3 \dasharrow I_{[e]}$ factors through

770: $L_\alpha \otimes \wedge^2 Q_\alpha^*$, and the induced rational map

771: $\overline q:L_\alpha \otimes \wedge^2 Q_\alpha^* 

772: \dasharrow I_{[e]} = \G(2,Q_\alpha^*)$ is described as

773: follows : let $\varphi \in L_\alpha \otimes \wedge^2 Q_\alpha^* \simeq

774: Hom(L_\alpha \otimes \wedge^2 Q_\alpha^*)$ be generic. Its image in

775: $\G(2,W_5^*)$ under $\overline q$

776: represents the linear subspace generated by

777: \begin{itemize}

778: \item

779: the

780: orthogonal for the alternate form $\varphi(e_2^*)$ of the kernel of

781: $\varphi(e_1^*)$,\hspace{.4cm} and

782: \item the orthogonal for $\varphi(e_1^*)$ of the kernel

783: of $\varphi(e_2^*)$.

784: \end{itemize}

785: This is well-defined \iff $\varphi(e_1^*)$ and

786: $\varphi(e_2^*)$ have maximal rank 4 and the corresponding orthogonals

787: are different lines in $V_2^*$. This is proved in

788: \cite[theorem 4.3]{flop_scorza}.

789: 

790: 

791: 

792: 

793: \subsectionplus{Dual schemes}

794: 

795: For some purposes (for example \cite{hermitian}),

796: it may be usefull to extend the above definition of

797: dual varieties to more general schemes. The goal of this subsection is 

798: to explain how this is possible.

799: 

800: Let us first define the cotangent scheme of a subscheme. So let $S$ be

801: an arbitrary scheme and $f:X \rightarrow Y$ a morphism above $S$.

802: The cotangent scheme $T^*X$ of $X$ is

803: ${\bf Spec}\ S\point {\cal H}om(\Omega_{X/S},\co_X)$; it is a scheme

804: over $X$, equipped with a natural section, the zero section.

805: Now $f$ induces a natural morphism of sheaves

806: $f^*\Omega_{Y/S} \rightarrow \Omega_{X/S}$, and so a morphism

807: $f^* T^*Y \rightarrow T^*X$. We finally define the cotangent

808: scheme $N^*_{X,Y}$ as the fiber above the zero section of this map.

809: 

810: Let $G$ be a semi-simple Chevalley group scheme over $\Z$, $P$ and $Q$

811: parabolic subgroups.

812: 

813: \begin{defi}

814: $P$ and $Q$ allow duality if the complex groups $P(\C),Q(\C)$ do.

815: \end{defi}

816: 

817: If $P$ and $Q$ allow duality, although the moment map 

818: $T^*G/P \rightarrow \g^*$

819: may fail to be birational in positive caracteristic, there is still a

820: well-defined birational map $T^*G/P \dasharrow T^*G/Q$, defined over

821: $\Z$~:

822: 

823: \begin{prop}

824: There is a $G$-equivariant birational map

825: $\mu : T^*G/P \dasharrow T^*G/Q$ defined over $\Z$.

826: \end{prop}

827: \pr

828: By \cite[theorem 6.1]{namikawa} and proposition \ref{prop_fibration},

829: any pair of parabolic subgroups allowing duality is

830: related by a chain of pairs $(P,Q)$ of parabolic subgroups for which the

831: birational map $T^* G/P \dasharrow T^* G/Q$ is locally isomorphic with

832: a family of birational maps given by a

833: fundamental stratified Mukai flop. It is therefore

834: enough to check the proposition

835: for fundamental cases. In these cases it is a consequence of the explicit

836: description of this flop recalled in \ref{subsection_notation}.

837: \qed

838: 

839: If $S$ is a scheme and $G,P,Q$ are as above, let $G_S,P_S,Q_S$ the

840: groups obtained by base change $S \rightarrow Spec\ \Z$.

841: 

842: \begin{defi}

843: Let $S$ be a reduced irreducible scheme, and

844: let $f: X \rightarrow G_S/P_S$ be an irreducible

845: closed $S$-subscheme. We say that

846: $X$ is suitable if $\mu$ is defined at the generic point

847: of $N^*_{X,G_S/P_S}$.

848: In this case, the dual scheme of $X$ is the

849: scheme-theoretic image of $N^*_{X,G_S/P_S}$ under $\pi_Q \circ \mu$.

850: \end{defi}

851: 

852: 

853: 

854: \subsectionplus{Functorial property of dual varieties}

855: 

856: \label{subsection_fonctoriel}

857: 

858: We come back to our setting of complex geometry. In the usual setting,

859: if $X_1,X_2 \subset \p V$ are subvarieties, with $X_1 \subset X_2$,

860: there is in general no relation of inclusion between the dual

861: varieties of $X_1$ and $X_2$. Thus dual varieties have bad functorial

862: properties. The only thing one can say is the following obvious result.

863: 

864: \begin{prop}

865: Let $P,Q \subset G$ allow duality.

866: Let $X \subset G/P$ be suitable, $g \in G$, and $Y=g(X)$. Then $Y$ is

867: suitable and $g(X^Q)=Y^Q$.

868: \label{foncteur}

869: \end{prop}

870: \pr

871: Let $x \in X^s$ and $f \in N_x^*X \subset \g^*$ an element in the open

872: $G$-orbit. Then

873: $\tr g^{-1}.f \in N_{g(x)}^*Y$ is also in the open $G$-orbit. Therefore,

874: $Y$ is suitable. Moreover, since $q$ is equivariant, 

875: $q(\tr g^{-1}.f)=g.q(f)$.

876: Therefore, $g(X^Q) \subset Y^Q$. By symmetry, we have

877: also $g^{-1}(Y^Q) \subset X^Q$, so $g(X^Q) = Y^Q$.

878: \qed

879: 

880: 

881: 

882: \subsectionplus{Dual varieties in type A}

883: 

884: \label{subsection_grassmannienne}

885: 

886: 

887: In this section, I give a description of the dual variety of a subvariety

888: $X \subset \G(r,V)$

889: using an analog of the quotient map $V \dasharrow \p V$ for

890: Grassmannians.

891: 

892: \lpara

893: 

894: If $A$ and $B$ are vector spaces, $Inj(A,B)$ will

895: denote the sets of linear (resp. linear and injective) maps from $A$ to $B$.

896: Let $\varpi : Hom(\C^r,V) \dasharrow \p Hom(\C^r,V)$ denote the

897: natural rational map, and let

898: $\pi : \p Hom(\C^r,V) \dasharrow \G(r,V)$ map $\varphi$ of rank $r$ on

899: its image.

900: Dually, consider $\varpi' : Hom(V,\C^r) \dasharrow \p Hom(V,\C^r)$

901: and $\pi' : \p Hom(V,\C^r) \dasharrow \G(r,V^*)$ mapping $\varphi'$

902: of rank $r$ on its kernel.

903: If $X \subset \G(r,V)$ is a subvariety, let $\overline X^o$ denote

904: the set $\pi^{-1}(X)$ and $\overline X$ its closure in $\p Hom(\C^r,V)$.

905: 

906: \begin{prop}

907: Let $X \subset \G(r,V)$ be a suitable variety. Then

908: $X^Q = \pi' [{(\overline X)}^*]$, where ${(\overline X)}^*$ is the usual dual

909: variety of the subvariety $\overline X \subset \p Hom(\C^r,V)$

910: of a projective space.

911: \label{type_A}

912: \end{prop}

913: \pr

914: We fix a smooth point $x \in X$ and $\overline f \in \p Hom(\C^r,V)$ 

915: such that $\pi(\overline f)=x$,

916: and start with two easy lemmas.

917: \begin{lemm}

918: $\overline X$ is smooth at $\overline f$.

919: \end{lemm}

920: \begin{proo}

921: In a neighbourhood of $\overline f$ we have $\overline X = \overline X^o$.

922: Moreover,

923: the map $\pi:\overline X^o \rightarrow X$ is locally a trivial fibration

924: with fiber at $x$ the smooth variety $Inj(\C^r,L_x)$.

925: \end{proo}

926: 

927: \noindent

928: We denote $f \in Hom(\C^r,V)$ such that 

929: $\varpi(f) = \overline f$.

930: \begin{lemm}

931: The affine tangent space $\widehat{T_{\overline f}\overline X}$ 

932: is the linear space of

933: maps $g:\C^r \rightarrow V$ such that the composition

934: $L_x \stackrel{f^{-1}}{\rightarrow} \C^r 

935: \stackrel{g}{\rightarrow}V \rightarrow V/L_x$

936: belongs to $T_xX$.

937: \label{lemme_TxX}

938: \end{lemm}

939: \noindent

940: Recall that for $Z \subset \p W$ a projective variety and $z \in Z$,

941: the affine tangent space $\widehat{T_zZ} \subset W$ is the tangent

942: space of the affine cone over $Z$ at a lift of $z$ in $W$.

943: \begin{proo}

944: Let $\widehat X \subset Hom(\C^r,V)$ be the affine cone over

945: $\overline X$.

946: Since $\widehat X$ is smooth at $f$, any tangent

947: vector is the direction of a curve included in $\widehat X$.

948: Let $\gamma : (C,0) \rightarrow (\widehat X,f)$ be a curve in

949: $\widehat X$ and let $g=\gamma'(0) \in Hom(\C^r,V)$. 

950: Under the well-known identification of

951: $T_x \G(r,V)$ with $Hom(L_x,V/L_x)$, the composition of the lemma

952: equals 

953: $(\pi \circ \varpi \circ \gamma)'(0)$. Therefore it belongs to

954: $T_xX$. By dimension count, the lemma follows.

955: \end{proo}

956: 

957: \noindent

958: \underline{\bf Proof of proposition \ref{type_A} :}

959: The linear subspace $(T_f\widehat X)^\bot \subset Hom(V,\C^r)$ is

960: the set of $g$'s such that for all $h \in T_f\widehat X$, the

961: composition 

962: $\C^r \stackrel{h}{\rightarrow} V \stackrel{g}{\rightarrow} \C^r$ is

963: traceless. Since $T_f \widehat X$ contains $Hom(\C^r,L_x)$, this means

964: that $g$ is induced by a morphism $\overline g: V/L_x \rightarrow \C^r$ such

965: that $f \circ \overline g$ is orthogonal to 

966: $T_xX \subset Hom(L_x,V/L_x)$, by lemma \ref{lemme_TxX}. Therefore,

967: for $h \in \G(r,V^*)$, we have 

968: $h \in \pi' \circ \varpi((T_f \widehat X)^\bot)$

969: \iff

970: $h \in q(N_x^*X)$.

971: \qed

972: 

973: 

974: 

975: 

976: 

977: 

978: %******************************************************************************

979: 

980: %******************************************************************************

981: 

982: %******************************************************************************

983: 

984: 

985: 

986: 

987: 

988: 

989: \sectionplus{Reduction to fundamental examples}

990: 

991: \label{section_reduction}

992: 

993: From section \ref{section_definition}, we see that there are a lot of pairs of

994: parabolic subgroups which allow duality. In this section, I will show

995: that to understand all the dual varieties, it is enough to

996: understand dual varieties for fundamental cases. 

997: 

998: For example, the

999: varieties corresponding to the marked diagrams

1000: $$

1001: \dynkineDeuxMarque{11.5}{4}{0}{2}

1002: \hspace{2cm}

1003: \dynkineDeuxMarque{11.5}{4}{6}{8}

1004: $$

1005: both have dimension 26. Using tables in \cite[p.202]{mcgovern}, we see

1006: that there is a unique nilpotent

1007: orbit of dimension 52 in ${\mathfrak e}_6$ and that the disconnected

1008: centralizer of an element of this orbit is trivial. Therefore, the two

1009: corresponding parabolic subgroups $P,Q \subset E_6$

1010: allow duality. It may seem at first that the corresponding

1011: duality

1012: $X \subset G/P \mapsto X^Q \subset G/Q$ has to do with the

1013: exceptional geometry of $E_6$. However, we will see that it is not

1014: the case; indeed, $X^Q$ can be described using dual varieties in four

1015: classical homogeneous spaces. Indeed, \cite[theorem 6.1]{namikawa}

1016: is verified

1017: in this case thanks to the sequence of parabolic subgroups

1018: $$

1019: \dynkineDeuxMarque{8.5}{4}{0}{2} \rightarrow

1020: \dynkineUneMarqueP{8.5}{4}{0} \rightarrow

1021: \dynkineUneMarqueP{8.5}{4}{8} \rightarrow

1022: \dynkineDeuxMarque{8.5}{4}{6}{8} \hspace{.5cm},

1023: $$

1024: and we will see in

1025: this section how to compute accordingly dual varieties.

1026: We will show that the computation of the dual variety

1027: $X^Q$ for $X \subset G/P$ can be

1028: done in three steps, the first and the last in a family of spinor varieties

1029: $\G_Q^+(5,10)$, and the second in a family $\p^5$'s.

1030: 

1031: 

1032: \subsectionplus{Biduality theorem}

1033: 

1034: 

1035: Let $G$ be as above, $P,Q,R \subset G$ be subgroups 

1036: such that $P$ and $Q$ allow

1037: duality, and $Q$ and $R$ allow duality. 

1038: Then, by definition $P$ and $R$ also allow duality.

1039: 

1040: \begin{theo}[Biduality theorem]

1041: Let $X \subset G/P$ be an suitable variety. Then $X^Q$ is suitable

1042: and $\mu(N^*X) = N^*X^Q$. In particular, 

1043: ${(X^Q)}^R = X^R$.

1044: \label{theo_bidualite}

1045: \end{theo}

1046: \noindent

1047: If $G=SL_n$, $P=R$ is the stabilisor of a line and $Q$ is the

1048: stabilisor of a hyperplane, we recover the

1049: usual biduality theorem.

1050: \begin{proo}

1051: We follow the argument of \cite[pp.27 to 30]{gkz}.\\

1052: Let $N=\mu(N^*X) \subset T^* G/Q$.

1053: Recall that $T^* G/Q$ is a symplectic variety. Moreover, it is proved

1054: in \cite{gkz} that $N^*X \subset T^* G/P$ is a lagrangien

1055: subvariety of $T^* G/P$. Let $\co \subset \g^*$ denote the 

1056: nilpotent orbit which

1057: closure is the image of $T^* G/P$. Since the birational morphisms

1058: $T^* G/P \stackrel{\sim}{\dasharrow} {\cal O}

1059: \stackrel{\sim}{\dasharrow} T^* G/Q$ are symplectic, it follows that

1060: $N$ is also lagrangien.

1061: 

1062: Moreover, it has the property that if 

1063: $(x,f) \in N$ and $\lambda \in \C$, then $(x,\lambda f) \in N$. 

1064: This follows from the fact that the image of $N^*X$ in

1065: $\overline {\cal O}$ is stable under multiplication by scalars.

1066: From \cite[proposition 3.1]{gkz}, we know that

1067: $N=N^*Z$ for $Z=\pi_Q(N)=X^Q$. Therefore, $\mu(N^*X)=N^*X^Q$

1068: and $X^Q$ is suitable.

1069: 

1070: Since ${(X^Q)}^R$ (resp. $X^R$) is the image of $N^*X_Q$

1071: (resp. $\mu(N^*X)$) under the

1072: rational map $T^* G/Q \dasharrow G/R$, these varieties are equal.

1073: \end{proo}

1074: 

1075: The following corollary shows that the name of biduality theorem for

1076: the above result is justified~:

1077: 

1078: \begin{coro}

1079: \label{coro_bidualite}

1080: Let $P,Q \subset G$ allow duality. If $X \subset G/P$ is suitable,

1081: then $X^Q$ is suitable and ${(X^Q)}^P = X$. Moreover,

1082: if $x \in X$ and $h \in X^Q$, then $h$ is

1083: tangent to $X$ at $x$ \iff $x$ is tangent to $X^Q$ at $h$.

1084: \end{coro}

1085: \begin{proo}

1086: To prove that ${(X^Q)}^P = X$, 

1087: it is enough to take $R=P$ in theorem \ref{theo_bidualite}, after

1088: observing that for suitable $X \subset G/P$, $X^P = X$.

1089: The second result, that $h$ is tangent to $X$

1090: at $x$ \iff $x$ is tangent to $X^Q$ at $h$ follows from the fact the

1091: first (resp. the second) affirmation means that $(x,h)$ lies in the

1092: image by $(p,\pi_Q)$ of an element in $\mu(N^*X)$ (resp. $N^*X^Q$).

1093: \end{proo}

1094: 

1095: 

1096: 

1097: \subsectionplus{Families of dual varieties}

1098: 

1099: \label{subsection_sequence}

1100: 

1101: 

1102: We consider the following situation :

1103: let $R \subset G$ be a parabolic subgroup.

1104: Let $P,Q \subset R \subset G$ be parabolic subgroups and recall

1105: notations of subsection \ref{subsection_fondamental}.

1106: If $X \subset G/P$ and $z \in G/R$, denote $X_z:=X \cap \cf_z$.

1107: Assume $P_L,Q_L \subset L$ allow duality.

1108: For $z \in G/R$ and suitable $Y \subset \cf_z \simeq L/P_L$, let 

1109: $Y^{Q_L} \subset \cg_z \simeq L/Q_L$ denote its generalised dual

1110: variety.

1111: 

1112: \begin{theo}

1113: With the previous notations, assume that $P,Q \subset G$ allow

1114: duality, and also $P_L,Q_L \subset L$.

1115: If $X \subset G/P$ is suitable, then

1116: for generic $x \in X$, $X_{f_P(x)} \subset \cf_{f_P(x)}$ is 

1117: suitable. Moreover, $X^Q$ is the closure of the union of

1118: the $X_{f_P(x)}^{Q_L}$, for such $x$ in $X$.

1119: \label{sous-groupe}

1120: \end{theo}

1121: \pr

1122: Let $f \in N^*_xX$ an element which $G$-orbit in $T^* G/P$ is dense

1123: and set $z = f_P(x)$.

1124: We have seen in the proof of proposition \ref{prop_fibration}

1125: that the restriction $f_x$ of $f$ to 

1126: $T_x \cf_z$ is a generic element in $T^* \cf_z \simeq T^* L/P_L$.

1127: Moreover, this restriction belongs to $N^*_x X_z$, so that $X_z$ is

1128: suitable.

1129: 

1130: Let $q_z : T^* \cf_z \dasharrow \cg_z$ be the composition of

1131: $\mu_z : T^* \cf_z \dasharrow T^* \cg_z$ and the projection

1132: $T^* \cg_z \rightarrow \cg_z$. Proposition \ref{prop_fibration}

1133: states that $q(f) = j_z \circ q_z(f_x) \in G/Q$. Therefore it follows

1134: that $q(N^*X_{|X_z}) = j_z(X_z^{Q_L})$.

1135: The description of $X^Q$ in the theorem follows.

1136: \qed

1137: 

1138: \lpara

1139: 

1140: As a consequence of theorems \ref{theo_bidualite} and \ref{sous-groupe},

1141: if $P=P_1 \times P_2$ and $Q=Q_1 \times Q_2$ are parabolic subgroups

1142: of $G=G_1 \times G_2$, 

1143: and if $X=X_1 \times X_2$, then we have

1144: $X^Q = X_1^{Q_1} \times X_2^{Q_2}$.

1145: 

1146: 

1147: 

1148: 

1149: 

1150: %******************************************************************************

1151: 

1152: %******************************************************************************

1153: 

1154: %******************************************************************************

1155: 

1156: 

1157: 

1158: \sectionplus{Tangency for fundamental examples}

1159: 

1160: \label{section_tangency}

1161: 

1162: In this section, if $x \in X \subset G/P$,

1163: I introduce a definition of the embedded tangent cone at $x$,

1164: $\overline{T_xX}$, which is a subvariety of $G/P$ and a cone

1165: at $x$ (in a suitable

1166: sense). I also introduce the cotangent

1167: variety at $x$, $\overline{N_xX}$, which is

1168: a subvariety of $G/Q$. Moreover a notion of ``linear varieties'' is defined

1169: and linear varieties are classified.

1170: 

1171: From now on, $P,Q \subset G$ are fundamental subgroups of $G$ allowing

1172: duality.

1173: 

1174: \subsectionplus{A tangent element is incident}

1175: 

1176: In this subsection, we prove that if $x \in X \subset G/P$ and 

1177: $h \in G/Q$ is tangent to $X$ at $x$ 

1178: (see definition \ref{i_h}), then $h$ is incident to $x$

1179: (in the sense that the stabilisors of $x$ and $h$ contain a common

1180: Borel subgroup). This only holds in fundamental cases.

1181: 

1182: \begin{nota}

1183: Let $x \in \g$ nilpotent. Then there exists $y,h \in \g$ such that $(x,y,h)$

1184: is a $\mathfrak {sl}_2$-triple. For $i \in \Z$, let $\g_i$ denote

1185: $\{X \in \g : [h,X]=iX\}$. The parabolic subalgebra 

1186: $\plie_x := \oplus _{i \geq 0} \g_i$ does not depend on $y$ and $h$

1187: \cite[theorem 3.8]{mcgovern}, and is

1188: called the canonical parabolic subalgebra of $x$.

1189: \end{nota}

1190: 

1191: In the following lemma, I say that $\plie \subset \g$ is a maximal

1192: parabolic subalgebra of $\g$ of fundamental type if the pair

1193: $(\g,\plie)$ is the pair of Lie algebras of groups $(G,P)$ as in

1194: definition \ref{fondamental}. 

1195: 

1196: \begin{lemm}

1197: Let $x \in \g$ and $\plie$ be a polarisation of $x$. Assume that

1198: $\plie$ is a maximal parabolic subalgebra of fundamental type. 

1199: Then $\plie_x \subset \plie$.

1200: \end{lemm}

1201: \begin{proo}

1202: Let $\plie$ be a maximal parabolic subalgebra which is a polarisation of $x$.

1203: Let $(x,y,h)$ be a $\mathfrak {sl}_2$-triplet, $\h$ a Cartan

1204: subalgebra containing $h$ and $\Delta=\{\alpha_1,\ldots,\alpha_r\}$ 

1205: a basis of the root system such

1206: that $\forall \alpha \in \Delta,\alpha(h) \geq 0$. 

1207: 

1208: We denote $\plie_1$ the following maximal parabolic subalgebra~:

1209: $$

1210: \plie_1 := \h \oplus \bigoplus_{

1211: \begin{array}{c}

1212: \alpha = \sum_j k_j \alpha_j\\

1213: k_i \geq 0

1214: \end{array}

1215: } \g_{\alpha}\ \ ,

1216: $$

1217: where $i$ is chosen such that $\plie$ is conjugated to $\plie_1$

1218: (such $i$ exists because $\plie$ is a maximal parabolic subalgebra).

1219: 

1220: Let us prove that $x \in \u(\plie_1)$.

1221: According to the

1222: decomposition $\g = \h \oplus \bigoplus_{\alpha} \g_{\alpha}$, we can

1223: write $x = h_x + \sum_\alpha x_\alpha$, with $h_x \in \h$ and

1224: $x_\alpha \in \g_\alpha$. Now, since $[h,x]=2x$, we deduce that 

1225: $h_x = 0$ and that for any root $\alpha$, either $x_\alpha = 0$ or

1226: $\alpha(h) = 2$. 

1227: 

1228: \begin{clai}

1229: If $\alpha = \sum k_j \alpha_j$ is a root, then

1230: $\alpha(h)=2 \Longrightarrow k_i > 0$.

1231: \end{clai}

1232: \begin{proo}

1233: This is proved by ad hoc arguments in all cases. Assume first that 

1234: $\g = \liesl_n$ and that $\plie$ is the stabilisor of an $r$-dimensional

1235: subspace. Thus $i=r$.

1236: Recall that the weighted diagram of $x$ is by definition the list

1237: of the values $\alpha_j(h)$. The weighted diagrams of nilpotent

1238: elements in $\liesl_n$ are well-known; in our case, since

1239: $x$ is a generic element of $\u(\plie)$ with $\plie$ conjugated to

1240: $\plie_1$, we have 

1241: $\alpha_i(h)=\alpha_{n-i}(h)=1$ and the other values $\alpha_j(h)$ equal 0.

1242: The equality $\alpha(h)=2$ with $\alpha = \sum k_j \alpha_j$ amounts

1243: to $k_i + k_{n-i} = 2$, which implies $k_i = k_{n-i} = 1$.

1244: 

1245: Assume now that $\g = \spin_{4p+2}$. In this case, there is only one 

1246: possibility for the $G$-orbit in $\spin_{4p+2}$

1247: of $x$, and $\alpha_j(h)=1$ \iff $\alpha_j$

1248: is a spin root (ie $j \in \{2p,2p+1\}$); otherwise $\alpha_j(h)=0$. Therefore

1249: $\alpha(h) = 2$ implies that $\alpha$ is not less than the root

1250: $\alpha_{2p-1} + \alpha_{2p} + \alpha_{2p+1}$, which implies the claim.

1251: 

1252: If $\g$ is of type $\e_6$ and $\plie$ corresponds to the first root, then

1253: the weighted diagram of $x$ is $\poidsesix 100010$

1254: (see \cite[table p.202]{mcgovern}). Since for all roots

1255: $\sum k_j \alpha_j$ we have $-1\leq k_1,k_6\leq 1$, we again have

1256: $\sum k_j \alpha_j(h) = 2 \Rightarrow k_1=k_6=1$. In case $\plie$

1257: corresponds to the second root, the weighted diagram is

1258: $\poidsesix 010100$. The equality $\alpha(h)=2$ for

1259: $\alpha = \sum k_j \alpha_j$ implies that $k_3 + k_5 = 2$. If

1260: $k_3 = 2$, then necessarily $k_5 \geq 1$ (see the list of roots in

1261: \cite{bourbaki}), so we get a contradiction. Similarly $k_5 \leq 1$.

1262: So $k_3 = k_5 = 1$, and again the claim is proved.

1263: \end{proo}

1264: 

1265: This claim therefore proves that if $x_\alpha \not = 0$, with

1266: $\alpha = \sum k_j \alpha_j$, then $k_i > 0$. 

1267: This proves that $x$ belongs

1268: to 

1269: $$

1270: \bigoplus_{

1271: \begin{array}{c}

1272: \alpha = \sum_j k_j \alpha_j\\

1273: k_i \geq 1

1274: \end{array}

1275: } \g_{\alpha}\ ,

1276: $$

1277: which is readily seen to be $\plie_1^\bot = \u(\plie_1)$.

1278: Thus $x \in \u(\plie_1)$

1279: and $\plie_1$ is a polarisation of $x$.

1280: Now, since the map $T^* G/P \rightarrow \g$ is birational on its

1281: image, there is a unique polarisation of $x$ in the conjugacy class of

1282: $\plie$. Therefore $\plie = \plie_1$.

1283: 

1284: 

1285: \lpara

1286: 

1287: Let us now show that $\plie_x \subset \plie_1$. Since $\plie_x \supset \h$,

1288: it is the sum of $\h$ and some root spaces. Now, assume 

1289: $\g_\alpha \subset \plie_x$, with $\alpha = \sum k_j \alpha_j$.

1290: This means that $\sum k_j \alpha_j(h) \geq 0$.

1291: I claim that $k_i \geq 0$. In fact, if $k_i < 0$, then $\alpha$ is a negative

1292: root, so $k_j \leq 0$ for all $j$. We therefore have

1293: $\sum k_j \alpha_j(h) \leq k_i \alpha_i(h)$.

1294: In the proof of the above claim, we have seen that we

1295: allways have $\alpha_i(h)=1$. So we get a contradiction.

1296: 

1297: Therefore, we have $k_i \geq 0$, and so $\g_\alpha \subset \plie_1$.

1298: Since $\plie_1 = \plie$, we have proved that $\plie_x \subset \plie$, as

1299: claimed.

1300: \end{proo}

1301: 

1302: 

1303: \begin{coro}

1304: \label{coro_incident}

1305: If $\plie$ and $\q$ are polarisations of the same nilpotent element

1306: $x$, and are maximal parabolic subalgebras of

1307: fundamental type, then they contain a

1308: common Borel subalgebra.

1309: \end{coro}

1310: \pr

1311: They both contain the canonical parabolic subalgebra $\plie_x$.

1312: \qed

1313: 

1314: \lpara

1315: 

1316: We now show, with an example, that the above corollary is wrong if one

1317: considers non maximal parabolic subalgebras.

1318: 

1319: \begin{exem}

1320: Let $\g = {\mathfrak sl}_n$. Let $x \in \g$ be an element of rank 2,

1321: such that $x^3 = 0$ but $x^2 \not = 0$. Let $\plie$ (resp. $\q$) be the

1322: parabolic subalgebra preserving the image of $x^2$ and the image of

1323: $x$ (resp. the kernel of $x$ and the kernel of $x^2$). Then we have 

1324: $x \in  \u(\plie)$ and $x \in \u(\q)$. However, since

1325: $\im\ x \not \subset \ker x$, $\plie$ and $\q$ are

1326: not incident.

1327: \end{exem}

1328: \begin{proo}

1329: If $y \in \plie$ (resp. $y \in \q$), then the commutator

1330: $[x,y]$ is strictly upper triangular for the filtration

1331: $\im\ x^2 \subset \im\ x \subset \C^n$ (resp.

1332: $\ker x \subset \ker x^2 \subset \C^n$). Therefore, $[x,y]$ is

1333: traceless and so $x \in \plie^\bot$ (resp. $x \in \q^\bot$).

1334: \end{proo}

1335: 

1336: \lpara

1337: 

1338: The Schubert varieties $I_h$ (recall definition \ref{i_h})

1339: give a geometric understanding of the

1340: rational map $q : T^* G/P \dasharrow G/Q$ : 

1341: \begin{lemm}

1342: \label{lemm_q}

1343: Assume $P$ and $Q$ are maximal parabolic subgroups.

1344: Let $x \in G/P$ and $h \in G/Q$, and

1345: let $f$ be a generic element in 

1346: $T_x^* G/P$. Then $q(f) = h$ \iff $x \in I_h$ and the cotangent

1347: form $f$ vanishes on $T_x I_h$.

1348: \end{lemm}

1349: \noindent As a consequence of the lemma, there is a unique $h$ such that

1350: $x \in I_h$ and $f$ vanishes on $T_x I_h$. By definition, if $h$ is

1351: tangent to $X$ at $x$, then there exists $f \in N^*_xX$ such that $q$

1352: is defined at $f$ and $q(f)=h$. Thus the lemma implies that the

1353: intersection $I_h \cap X$ is not transverse at $x$, as was stated in 

1354: fact \ref{fait_tangent}.

1355: \begin{proo}

1356: Let $x \in G/P$; $t_P$ restricts to an isomorphism between

1357: $T^*_x G/P$ and $(\g/\plie_x)^* \subset \g^*$ if $\plie_x$ denotes the Lie

1358: algebra of the stabilisor of $x$. Conversely, given $\eta \in {\cal O}$,

1359: $\pi_P(t_P^{-1}(\eta))$ is the unique $x \in G/P$ such that the corresponding

1360: parabolic subalgebra $\plie_x$ is orthogonal to $\eta$.

1361: 

1362: Let $x \in G/P$, $f \in T_x^* G/P$ generic and 

1363: $\eta = t_P(f) \in \plie_x^\bot$, and let $h=q(f)$. 

1364: The previous argument shows that $h$ is the unique element in $G/Q$

1365: such that $\eta$ vanishes on $\q_h$. Moreover, we know by

1366: corollary \ref{coro_incident} that $x \in I_h$.

1367: Note that $T_xG/P = \g / \plie$ and

1368: $T_x I_h \simeq \q_h/(\plie_x \cap \q_h)$.

1369: Since $\eta$ vanishes on $\plie_x$, it will

1370: vanish on $\q_h$ \iff it vanishes on $\q_h/(\plie_x \cap \q_h)$, namely,

1371: \iff $f$ vanishes on $T_xI_h$.

1372: \end{proo}

1373: 

1374: \begin{exem}

1375: Let $h \in G/Q$ and let $X = I_h \subset G/P$. Then $X$ is suitable

1376: and $X^Q = \{h\}$. Moreover $p(T^*_hG/Q) = I_h$.

1377: \end{exem}

1378: \begin{proo}

1379: First, let $x \in X$, let $f \in T^*_xX$ be generic and let $h=q(f)$.

1380: By corollary \ref{coro_incident}, $x$ and $h$ are incident, and by

1381: lemma \ref{lemm_q}, $f$ vanishes on $T_x I_h$. Thus, $I_h$ is

1382: suitable. Since $G/Q$ is homogeneous, $I_h$ is suitable for all 

1383: $h \in G/Q$.

1384: 

1385: Let $x \in X$ and $f \in N^*_xX$ generic. Then by the above $q(f)=:h'$

1386: is well-defined, and by lemma \ref{lemm_q} again, $h'$ is the unique

1387: element in $G/Q$ such that $x \in I_{h'}$ and such that $f$ vanishes on

1388: $T_x I_{h'}$. Since $h$ satisfies these conditions, $h'=h$. Therefore,

1389: $X^Q = \{h\}$.

1390: 

1391: For the last point, we note that $p(T^*_h G/Q) = \{h\}^P = I_h$, by

1392: biduality theorem \ref{theo_bidualite}, 

1393: since we have proved that $I_h^Q = \{h\}$.

1394: \end{proo}

1395: 

1396: 

1397: \subsectionplus{Dual varieties and cones}

1398: 

1399: If $X \subset \p V$ is included in a hyperplane represented by

1400: $h \in \p V^*$, then the dual variety of $X$, which is a subvariety of

1401: $\p V^*$, is a cone over $h$. The aim of this subsection is to prove

1402: an analogous result for our generalised dual varieties. Our first goal

1403: is to define cones.

1404: 

1405: \begin{defi}Let $x_1,x_2 \in G/P$ 

1406: \begin{itemize}

1407: \item

1408: $x_1,x_2$ are linked if there exists $h \in G/Q$ such that 

1409: $x_1,x_2 \in I_h$.

1410: \item

1411: If $E \subset G/P$, let 

1412: $\displaystyle I_E := \bigcap_{x \in E} I_x \subset G/Q$.

1413: \item

1414: If $x_1,x_2$ are linked, denote 

1415: $\displaystyle L(x_1,x_2) = \bigcap_{h \in I_{\{x_1,x_2\}}} I_h$.

1416: \end{itemize}

1417: \end{defi}

1418: \noindent

1419: In $\p V$, all points are linked, and $L(x_1,x_2)$ is the line through

1420: $x_1$ and $x_2$. The difference between $\p V$ and our general

1421: situation is that in general $G$ does not 

1422: act transitively on pairs of distinct points $x,y \in G/P$, so that $L(x,y)$

1423: may depend, up to isomorphism, on $x$ and $y$.

1424: However, cones are defined in perfect analogy~:

1425: 

1426: \begin{defi}

1427: \label{def_cone}

1428: Let $X \subset G/P$ and $x \in X$. Then $X$ is a cone over $x$ if

1429: for all $y \in X$, $x$ and $y$ are linked and $L(x,y) \subset X$.

1430: \end{defi}

1431: \noindent

1432: An equivalent definition is that for generic $y \in X$ the same

1433: condition holds, as will be clear from the following description of

1434: $L(x,y)$~:

1435: 

1436: \begin{prop}

1437: \label{prop_l}

1438: Let $x \not = y \in G/P$. We have~:

1439: \begin{itemize}

1440: \item

1441: If $G/P = \G(r,V)$, then $(x,y)$ are linked \iff 

1442: $\codim_V(L_x+L_y) \geq r$, in which case $L(x,y) = \G(r,L_x+L_y)$.

1443: \item

1444: If $G/P = \G_Q^+(2p+1,4p+2)$, then $(x,y)$ are linked \iff

1445: we have $\dim(L_x \cap L_y)=2p-1$, in which case

1446: $L(x,y) = \{ z : L_z \supset L_x \cap L_y \} \simeq \p^1$.

1447: \item

1448: If $G/P = E_6/P_1$, then $(x,y)$ are allways linked. In case a line

1449: passes through $x$ and $y$ in $E_6/P_1$, then $L(x,y)$ is this line;

1450: otherwise, there is a unique smooth 8-dimensional quadric through $x$

1451: and $y$, and $L(x,y)$ is this quadric.

1452: \item

1453: If $G/P = E_6/P_3$, then $(x,y)$ are linked \iff there is a $\G(2,5)$

1454: through them. If $\dim(L_x \cap L_y) = 1$ then 

1455: $L(x,y)$ is equal to $\G(2,L_x + L_y) \simeq \p^2$, 

1456: otherwise $L(x,y) \simeq \G(2,5)$.

1457: \end{itemize}

1458: \end{prop}

1459: \noindent

1460: In this proposition, for the two exceptional cases, I use the minimal

1461: projective homogeneous embedding $E_6/P_i \subset \p V_i$. 

1462: For example, in the case of

1463: $E_6/P_3$, the condition that there is a $\G(2,5)$ through $x$ and $y$

1464: means that there is a linear 10-dimensional subspace $W \subset V_3$ 

1465: containing $x$ and $y$ and such that $\p W \cap E_6/P_3$ is projectively

1466: isomorphic with a Grassmanian $\G(2,5)$ in its Plücker

1467: embedding. Recall also that $E_6/P_3$ parametrises projective lines in

1468: $V_1$ which are included in $E_6/P_1$. For $x \in E_6/P_3$, the

1469: corresponding 2-dimensional subspace of $V_1$ has been denoted $L_x$.

1470: \begin{proo}

1471: The first case follows directly from the definition. In the second

1472: case, one only has to note that if there exists $h \in G/Q$ such that

1473: $(x,h),(y,h)$ are incident, then 

1474: $\dim(L_x\cap L_h) = \dim(L_y \cap L_h) = 2p$, so

1475: $\dim(L_x\cap L_h\cap L_y) = 2p-1$.

1476: 

1477: \lpara

1478: 

1479: For the exceptional cases one oviously has to use the geometry of the

1480: involved homogeneous spaces. Let us first consider $E_6/P_1$. For all

1481: $x \in E_6/P_1$, $I_x$ is a smooth 8-dimensional quadric. Moreover, for any

1482: $x\not = y \in E_6/P_1$, the intersection of the two quadrics

1483: $I_x$ and $I_y$ is either a point or a $\p^4$. In

1484: fact, this was proved in 

1485: \cite[propositions IV.3.2 and IV.3.3]{zak} in the context of Severi

1486: varieties, but also follows easily from the fact that there are three

1487: $E_6$-orbits in $E_6/P_1 \times E_6/P_1$ \cite[proposition 18]{quantique}.

1488: Given $x,y \in E_6/P_1$, we can have $x=y$, $x \not =

1489: y$ and there is a line through $x$ and $y$, or there is no line

1490: through $x$ and $y$. This describes the three orbits in 

1491: $E_6/P_1 \times E_6/P_1$.

1492: In the degenerate case when a line passes through $x$ and $y$, 

1493: $I_x \cap I_y$ is thus isomorphic with $\p^4$. Dually,

1494: the intersection of all the

1495: $I_h$ for $h$ in this $\p^4$ is a linear space (indeed, $x \in I_h$

1496: \iff $x \in E_6/P_1 \subset \p V_1$ is orthogonal to

1497: $\widehat{T_h E_6/P_6} \subset V_6=V_1^*$) and contains $x$ and $y$;

1498: a direct computation of dimension shows that it is exactly

1499: the line through $x$ and $y$.

1500: In the generic case, $I_x \cap I_y = \{h\}$; therefore 

1501: $L(x,y) = I_h$ is the unique 8-dimensional quadric through $x$ and $y$.

1502: 

1503: \lpara

1504: 

1505: Let $\alpha,\beta \in E_6/P_3$ be linked, and

1506: denote $\kappa \in E_6/P_5$ an element such that 

1507: $\alpha,\beta \in I_\kappa$. According to subsection

1508: \ref{subsection_fondamental}, $\alpha$ and 

1509: $\beta$ represent 2-dimensional

1510: subspaces of a 5-dimensional subspace of $V_1$ denoted

1511: $W_\kappa$; we have denoted $L_\alpha,L_\beta$ these spaces.

1512: 

1513: Assume first that $\dim(L_\alpha \cap L_\beta) = 1$.

1514: It is proved in

1515: \cite[proposition 3.6]{flop_scorza} that the linear

1516: span of all the affine tangent spaces at the points of the projective plane

1517: generated by $L_\alpha$ and $L_\beta$

1518: is 24-dimensional and equal to the span of affine tangent spaces at

1519: points in $l_\alpha \cup l_\beta$. 

1520: Thus $(\alpha,\beta)$ defines a projective plane in $E_6/P_6$ and also in

1521: $E_6/P_5$. Moreover

1522: $I_{\alpha,\beta} = I_{\G(2,L_\alpha+L_\beta)} \simeq \p^2$ 

1523: and $L(\alpha,\beta) = \G(2,L_\alpha+L_\beta) \simeq \p^2$.

1524: 

1525: Assume finally that $L_\alpha$ and $L_\beta$ don't meet. 

1526: Let $L \subset L_\alpha \oplus L_\beta$ be any 3-dimensional subspace;

1527: the linear span $S_L$ of

1528: the affine tangent spaces at points of $\p L$ is again 24-dimensional,

1529: and any element in $I_{\alpha,\beta}$ must contain it.

1530: Assume that $I_{\alpha,\beta}$

1531: contains two points $\kappa,\lambda \in G/Q$. These points would correspond to

1532: codimension 2 subspaces $L_\kappa,L_\lambda$ of 

1533: $V_1$ containing $S_L$; therefore

1534: $L_\kappa$ and $L_\lambda$ would be contained 

1535: in a common hyperplane of $V_1$.

1536: Since $\alpha,\beta \in I_{\kappa,\lambda}$, by

1537: the case considered above, this would in turn imply that $L_\alpha$ and

1538: $L_\beta$ meet in dimension 1, which we have excluded.

1539: Therefore we have proved that $I_{\alpha,\beta} = \{ \kappa \}$, so

1540: $L(\alpha,\beta) = I_\kappa$ is isomorphic with $\G(2,5)$.

1541: \end{proo}

1542: 

1543: 

1544: \begin{theo}

1545: \label{theo_cone}

1546: Let $h \in G/Q$ and let $X \subset G/P$ such that $X \subset I_h$. 

1547: Then $X$ is suitable and $X^Q$ is a cone over $h$.

1548: \end{theo}

1549: \begin{rema}

1550: In fact, as the proof will show, in all the cases but in type $A$,

1551: a stronger result holds~: for any $k \in X^Q$, there is a certain

1552: homogeneous subvariety

1553: $q(\C.f + N^*_x I_h) \subset G/Q$, of type given by lemmas

1554: \ref{lemm_cone_gr}, \ref{lemm_cone_spin}, \ref{lemm_cone_e6_1},

1555: and \ref{lemm_cone_e6_2}, containing (eventually stricly) $L(h,k)$,

1556: and included in $X^Q$. Although the idea of proof of this theorem

1557: is uniform, this proof unfortunately ends up with a case by case

1558: analysis.

1559: \end{rema}

1560: \begin{proo}

1561: If $x \in X$, then $N^*_xX$ contains $N^*_x I_h$ on which $q$ is

1562: well-defined generically, so $X$ is suitable.

1563: Assume $X \subset I_h$ and let $k$ be a generic element in $X^Q$.

1564: By definition of $X^Q$ there is an element $x \in X$ and 

1565: $f \in N^*_xX$ such that $k = q(f)$. Since $x \in X \subset I_h$, we

1566: have $h \in I_x$. By corollary \ref{coro_incident}, $k \in I_x$;

1567: therefore $h$ and $k$ are linked. Moreover, we have 

1568: $f \not \in N_x^* I_h$ (otherwise we would have $q(f)=h$).

1569: Therefore it follows from

1570: the inclusion $q(\C.f + N^*_x I_h) \subset X^Q$ and the following

1571: lemmas \ref{lemm_cone_gr}, \ref{lemm_cone_spin}, \ref{lemm_cone_e6_1}

1572: and \ref{lemm_cone_e6_2} that $L(h,k) \subset X^Q$.

1573: \end{proo}

1574: 

1575: \begin{lemm}

1576: \label{lemm_cone_gr}

1577: Let $x \in \G(r,V),\ h\not = k \in \G(r,V^*)$ such that $h,k \in I_x$. Let

1578: $f \in N_x^* I_h$ such that $q$ is defined at $f$. Then

1579: $q(\C.f + N_x^* I_k) = L(h,k)$.

1580: \end{lemm}

1581: \begin{proo}

1582: Let $(e_i)$ be a base of $V$ and $(e_i^*)$ the dual base.

1583: Up to the action of $SL(V)$, we may assume that

1584: $L_x$ is the span of $e_1,\ldots,e_r$,

1585: $L_h$ is the span of $e_{n-r+1}^*,\ldots,e_n^*$, $L_k$ that

1586: of $e_{n-r+1-l}^*,\ldots,e_{n-r}^*,e_{n-r+1}^*,\ldots,e_{n-l}^*$, and

1587: finally that $f \in N_x^* I_h \simeq Hom(L_x^*,L_h)$ is defined by

1588: $f(e_j^*) = e_{n-r+j}^*$. Since

1589: $N_x^* I_k = Hom(L_x^*,L_k)$, a straightforward computation proves

1590: the lemma.

1591: \end{proo}

1592: 

1593: \begin{lemm}

1594: \label{lemm_cone_spin}

1595: Let $x \in \G_Q^+(2p+1,4p+2),\ h\not = k \in \G_Q^-(2p+1,4p+2)$ such that

1596: $h,k \in I_x$. Let

1597: $f \in N_x^* I_h$ such that $q$ is defined at $f$.

1598: Then $q(\C.f + N_x^* I_k) \simeq \p^{2p-1}$.

1599: \end{lemm}

1600: \noindent

1601: This lemma implies theorem \ref{theo_cone} in this case since

1602: $q(\C.f + N_x^* I_k)$ is a linear space containing $h$ and $k$, 

1603: and will therefore contain the line through $h$ and $k$.

1604: \begin{proo}

1605: We may assume that $x$ represents the isotropic subspace

1606: $L_x$ generated by $e_1^+,\ldots,e_{2p+1}^+$. Since $L_k$ meets $L_x$ along 

1607: a hyperplane, we may further assume that this hyperplane is generated by

1608: $e_2^+,\ldots,e_{2p+1}^+$. We therefore have

1609: $N^*_x I_k = \wedge^2 \scal{e_2^+,\ldots,e_{2p+1}^+} \subset

1610: \wedge^2 L_x = T^*_x \G_Q^+(2p+1,4p+2)$. Let 

1611: $f \in T^*_x \G_Q^+(2p+1,4p+2)$; since 

1612: $f \not \in N^*_x I_k$ (otherwise we would have $h=k$),

1613: the class of $f$ modulo $N^*_x I_h$ is the same as that of some form

1614: $e_1^+ \wedge e$, with $e \in \scal{e_2^+,\ldots,e_{2p+1}^+}$, and we may

1615: assume that $e=e_2^+$.

1616: 

1617: Recall that $I_x \simeq \p L_x^*$~:

1618: I claim that $q(\C.f + N_x^* I_k)$ is the orthogonal of $e_2^+$ in

1619: $\p L_x^*$. In fact, let $\wedge^p (\C.f + N_x^* I_k) \subset 

1620: \wedge^{2p} L_x \simeq L_x^*$ be the linear span of all the forms 

1621: in $\wedge^{2p} L_x$ which can

1622: be written as a wedge product of $p$ forms in 

1623: $\C.f \oplus N_x^* I_k$. We have

1624: $\wedge^p (\C.f + N_x^* I_k) \subset {(e_2^+)}^\bot$; therefore

1625: $q(\C.f + N_x^* I_k) \subset \p {(e_2^+)}^\bot$.

1626: 

1627: On the other hand, let 

1628: $\displaystyle f_0=\Sigma_{i=1}^p e_{2i-1}^+ \wedge e_{2i}^+$; we have

1629: $\displaystyle f_0^{\wedge(p-1)}=\Sigma_{i=1}^p {e_{2i-1}^+}^* 

1630: \wedge {e_{2i}^+}^* \wedge e_{2p+1}^*$, from which is follows that the 

1631: rational map $\C.f + N_x^* I_k \dasharrow \p {(e_2^+)}^\bot,

1632: g \mapsto [g^{\wedge p}]$

1633: is submersive at $f_0$, which implies the claim and the lemma. 

1634: \end{proo}

1635: 

1636: 

1637: \begin{lemm}

1638: \label{lemm_cone_e6_1}

1639: Let $x \in E_6/P_1,\ h \not = k \in E_6/P_6$ such that

1640: $h,k \in I_x$. Let

1641: $f \in N_x^* I_h$ such that $q$ is defined at $f$.

1642: If there passes a line through $h$ and $k$ in $E_6/P_6$

1643: then $q(\C.f + N_x^* I_k) \simeq \p^4$, otherwise

1644: $q(\C.f + N_x^* I_k) = I_x$

1645: \end{lemm}

1646: \begin{proo}

1647: We adopt the same strategy of proof as for lemma \ref{lemm_cone_spin}.

1648: Let $x \in E_6/P_1$ be fixed.

1649: In subsection \ref{subsection_fondamental}, we saw that

1650: $T^*_xX$ identifies with $\oc \oplus \oc$,

1651: $I_x$ with the projective quadric in $\p (\C \oplus \oc \oplus \C)$

1652: defined by $tu-N(z)=0$.

1653: We can assume that $k \in I_x$ is the class of $(0,0,1)$.

1654: Therefore, $N_x^* I_k = \overline{q^{-1}(k)} = \{(0,z) : z \in \oc\}$.

1655: 

1656: Write $f=(z_0,z_1)$. Since 

1657: $q((z_0,z_1)) = [N(z_0) : z_0 \overline z_1 : N(z_1)]$, 

1658: there will be a line through

1659: $q(f)$ and $k$ in the quadric $I_x$ \iff $N(z_0) = 0$. If this occurs, then

1660: $$q(\C.f + N_x^* I_k) = \{[(0,u,t)]:t\in \C,u\in L(z_0) \},$$

1661: where $L(z_0)$ denotes the set of right multiples of $z_0$~:

1662: $L(z_0)=\{z_0z:z\in \oc\}$.

1663: It is a linear subspace of $\oc$ of dimension 4, so

1664: $q(\C.f + N_x^* I_k)$ is isomorphic with $\p^4$, as desired. If

1665: $N(z_0) \not = 0$, then left multiplication by $z_0$ is invertible, so that

1666: $q : \C.f + N_x^* I_k \dasharrow I_x$ is dominant, and the lemma again

1667: holds.

1668: \end{proo}

1669: 

1670: 

1671: \begin{lemm}

1672: \label{lemm_cone_e6_2}

1673: Let $\alpha \in E_6/P_2,\ \kappa,\lambda \in E_6/P_5$ such that

1674: $\kappa,\lambda \in I_\alpha$. Let

1675: $f \in N_\alpha^* I_\kappa$ such that $q$ is defined at $f$.

1676: Then $q(\C.f + N_\alpha^* I_\kappa) = I_\alpha$.

1677: \end{lemm}

1678: \begin{proo}

1679: We fix $\alpha \in E_6/P_2$. Let $f_1^*,\ldots,f_5^*$ be a 

1680: base of $Q_\alpha^*$ and assume 

1681: that $\kappa$ corresponds to the linear subspace

1682: generated by $f_4^*,f_5^*$. Recall that there is a natural

1683: surjective map $\pi:T_\alpha^* E_6/P_2 

1684: \rightarrow Hom(L_\alpha^*,\wedge^2 Q_\alpha^*)$. Moreover,

1685: $\pi(N^*_\alpha I_\kappa) = Hom(L_\alpha^*,L) 

1686: \subset Hom(L_\alpha^*,\wedge^2 Q_\alpha^*)$, where 

1687: $L \subset \wedge^2 Q_\alpha^*$ is

1688: generated by $f_1^* \wedge f_4^*,f_1^* \wedge f_5^*,

1689: f_2^* \wedge f_4^*,f_2^* \wedge f_5^*,f_3^* \wedge f_4^*,f_3^* \wedge f_5^*,

1690: f_4^* \wedge f_5^*$ (for example, this follows from the fact that for any

1691: $\varphi \in Hom(L_\alpha^*,L)$,

1692: $\overline q(\varphi)$, if defined, equals $\kappa$).

1693: 

1694: Let $M \subset \wedge^2 Q_\alpha^*$ be generated by

1695: $f_1^* \wedge f_2^*,f_1^* \wedge f_3^*,f_2^* \wedge f_3^*$, so that

1696: $L \oplus M = \wedge^2 Q_\alpha^*$; the class of $\pi(f)$ modulo

1697: $\pi(N_\alpha^* I_\kappa)$ is the class of a unique

1698: $\overline f \in Hom(L_\alpha^*,M)$.

1699: Assume first that the rank of $\overline f$ is 1. We can therefore

1700: assume that $\overline f(e_1^*) = f_1^* \wedge f_2^*$ and

1701: $\overline f(e_2^*) = 0$, where $e_1^*,e_2^*$

1702: is a suitable basis of $L_\alpha^*$.

1703: 

1704: In the array below we give, for $\omega \in \wedge^2 Q_\alpha^*$, the value

1705: of the derivative $d\overline q_{\overline f}(\varphi)$, for 

1706: $\varphi:L_\alpha^* \rightarrow \wedge^2 Q_\alpha^*$ given by

1707: $\varphi(e_1^*) = \omega$ and $\varphi(e_2^*) = 0$~:

1708: $$

1709: \begin{array}{llll}

1710: f_1^* \wedge f_2^* \mapsto 0 & f_1^* \wedge f_3^* \mapsto 0 &

1711: f_1^* \wedge f_4^* \mapsto 0 & \hspace{-1cm}f_1^* \wedge f_5^* \mapsto 0 \\

1712: f_2^* \wedge f_3^* \mapsto 0 & f_2^* \wedge f_4^* \mapsto f_3^* \wedge f_4^* &

1713: f_2^* \wedge f_5^* \mapsto f_1^* \wedge f_4^* + f_3^* \wedge f_5^*\\

1714: f_3^* \wedge f_4^* \mapsto 0 & f_3^* \wedge f_5^* \mapsto 0 &

1715: f_4^* \wedge f_5^* \mapsto f_1^* \wedge f_5^*.

1716: \end{array}

1717: $$

1718: The following gives similar values for $\varphi$ defined by

1719: $\varphi(e_1^*) = 0$ and $\varphi(e_2^*) = \omega$~:

1720: $$

1721: \begin{array}{llll}

1722: f_1^* \wedge f_2^* \mapsto 0 & f_1^* \wedge f_3^* \mapsto 0 &

1723: f_1^* \wedge f_4^* \mapsto 0 & \hspace{-1cm}f_1^* \wedge f_5^* \mapsto 0 \\

1724: f_2^* \wedge f_3^* \mapsto 0 & f_2^* \wedge f_4^* \mapsto f_2^* \wedge f_3^* &

1725: f_2^* \wedge f_5^* \mapsto f_3^* \wedge f_4^* + f_1^* \wedge f_2^*\\

1726: f_3^* \wedge f_4^* \mapsto 0 & f_3^* \wedge f_5^* \mapsto 0 &

1727: f_4^* \wedge f_5^* \mapsto f_1^* \wedge f_4^*.

1728: \end{array}

1729: $$

1730: It follows from these computations that 

1731: $\overline q(\C.\overline f + \pi(N_\alpha^* I_\kappa))$ has

1732: dimension at least 6, so 

1733: $\overline q(\C.\overline f + \pi(N_\alpha^* I_\kappa)) = I_\alpha$ 

1734: in this case.

1735: 

1736: In case $\overline f$ has rank 2, the dimension of

1737: $\overline q(\C.\overline f + \pi(N_\alpha^* I_\kappa))$ 

1738: will not vary if $\overline f$ is replaced by

1739: $g.\overline f$, where $g \in SL(L_\alpha) \times SL(Q_\alpha)$ 

1740: preserves $\kappa$.

1741: Using a $\C^*$-action we can degenerate 

1742: $\overline f \in Hom(L_\alpha^*,M)$ 

1743: to some element $\overline f_0$ of

1744: rank one, for which we have already seen that 

1745: $\dim \overline q(\C.f_0 + \pi(N_\alpha^* I_\kappa)) = 6$.

1746: Since this dimension is

1747: lower semi-continuous, we have 

1748: $\dim \overline q(\C.\overline f + \pi(N_\alpha^* I_\kappa)) = 6$ and

1749: the lemma is proved.

1750: \end{proo}

1751: 

1752: \subsectionplus{The cotangent space and the tangent cone of a variety}

1753: 

1754: \label{subsection_cotangent}

1755: 

1756: \begin{defi} Let $x \in X$ be suitable.

1757: \begin{itemize}

1758: \label{cotangent}

1759: \item

1760: The embedded cotangent space of $X$ at $x$ is 

1761: $\overline{N_xX}:=q(N^*_xX)\subset G/Q$.

1762: \item

1763: The embedded tangent space of $X$ at $x$ is

1764: $\overline{T_xX}={\overline{N_xX}}^P$.

1765: \item

1766: $X \subset G/P$ is a linear subvariety if $\overline{T_xX}$ does not

1767: depend on $x$ suitable in $X$.

1768: \end{itemize}

1769: \end{defi}

1770: 

1771: \rks

1772: \begin{itemize}

1773: \item

1774: The notion of (co)-tangent space (and therefore of linear varieties)

1775: of $X \subset G/P$ could be defined for non maximal parabolic $P$, but

1776: then it would depend on the choice of a parabolic subgroup $Q$.

1777: \item

1778: An equivalent definition of linear subvarieties

1779: is that $\overline{N_xX}$ does not depend on

1780: suitable $x$ in $X$, since $\overline{N_xX} = {\overline{T_xX}}^Q$.

1781: \item

1782: By definition, $X^Q = \overline {\cup_{x \in X^s} \overline{N_xX}}$.

1783: \item

1784: In projective spaces, the tangent cone is the usual embedded tangent

1785: space and linear varieties are linear subspaces. Linear varieties will

1786: be classified in the next subsection.

1787: \end{itemize}

1788: 

1789: \begin{exem}

1790: Let $x \in G/P$ and $X=\{x\}$. Then $\overline{T_xX} = \{x\}$.

1791: \end{exem}

1792: \begin{proo}

1793: In fact, $\overline{N_xX} = q(T^*_xG/P) = X^Q$, so this follows from

1794: theorem \ref{theo_bidualite}.

1795: \end{proo}

1796: 

1797: \begin{lemm}

1798: For $x \in X^s$,

1799: $\overline{T_xX}$ is a cone over $x$ and therefore $x \in \overline{T_xX}$.

1800: \end{lemm}

1801: \pr

1802: In fact, for $x \in X^s$, we have $\overline{N_xX} \subset I_x$, so 

1803: $\overline{T_xX} = {\overline{N_xX}}^P$ is a cone

1804: over $x$ by theorem

1805: \ref{theo_cone}.

1806: \qed

1807: 

1808: 

1809: 

1810: \subsectionplus{Linear subvarieties}

1811: 

1812: In this subsection, we classify linear subvarieties.

1813: 

1814: 

1815: \begin{prop}

1816: \label{prop_lineaire}

1817: The following array gives the list of all linear subvarieties~:

1818: $$

1819: \begin{array}{|c|c|c|}

1820: \hline

1821: \mbox{G/P} & \mbox{Linear varieties} \\

1822: \hline

1823: \G(r,n) & \G(r,p),\ r\leq p \leq n  \\

1824: \hline

1825: \G^+_Q(2p+1,4p+2) & \{pt\};I_h,h \in \G_Q^-(2p+1,4p+2) \\

1826: \hline

1827: E_6/P_1 & \{pt\};I_h,h\in E_6/P_6\\

1828: \hline

1829: E_6/P_2 & \{pt\};I_\kappa,\kappa\in E_6/P_5\\

1830: \hline

1831: \end{array}

1832: $$

1833: \end{prop}

1834: \begin{proo}

1835: Let $X \subset G/P$ be linear.

1836: First, we prove that

1837: $\forall x \in X,\overline{T_xX} = X$, and that $X^Q$

1838: is linear.

1839: Let $x \in X^s$. Then

1840: $X^Q = \overline {\cup_{y \in X^s} \overline{N_yX}} = \overline {N_xX}$,

1841: since for all $y \in X^s$, $\overline{N_yX} = \overline{N_xX}$.

1842: Therefore, $X = {\overline{N_xX}}^P = \overline{T_xX}$ by

1843: corollary \ref{coro_bidualite}.

1844: Let $h \in X^Q$ and $x \in X$. Then, by biduality theorem again,

1845: $x \in \overline{N_hX^Q}$ \iff $h \in \overline {N_xX} = X^Q$.

1846: Therefore, $\overline{N_hX^Q}  = X$ and $X^Q$ is linear and the claim

1847: is proved.

1848: Since $X = \overline{T_xX}$ for all $x \in X$, $X$ is a cone over all

1849: of its points by theorem \ref{theo_cone}.

1850: \lpara

1851: 

1852: We finish the proof case by case. 

1853: In the case of Grassmannians, if we denote 

1854: $W = \sum_{x \in X} L_x$, since $X$ is a cone over all of its points,

1855: we have $\G(r,W) \subset X$, and so $X = \G(r,W)$.

1856: 

1857: In the case of spinor varieties, any $x,y \in X$ must be linked, which

1858: implies that the line through $x$ and $y$ is in $X$, so $X$ is a

1859: linear subspace. As a consequence of the following proposition

1860: \ref{prop_spinoriel}, the only linear subspaces which dual

1861: variety is again a linear

1862: subspace are the point and maximal linear subspaces. Since we have seen

1863: that $X^Q$ must be a linear variety, the proposition follows in this

1864: case.

1865: 

1866: Let $X \subset E_6/P_1$ be linear. Let $h \in X^Q$. If there are two

1867: points $x,y \in X$ such that there is no line through $x$ and $y$,

1868: then by lemma \ref{lemm_cone_e6_1} $L(x,y) = I_h$. Since $X \subset I_h$

1869: and $L(x,y) \subset X$, we have $X = I_h$ (and $X^Q$ is a point).

1870: 

1871: Otherwise, by theorem \ref{theo_cone}, $X$ is a linear

1872: subspace. If $X^Q$ is not a linear subspace, by the argument above, 

1873: $X = {(X^Q)}^P$ is a point. Assume now that both $X$ and $X^Q$ are

1874: linear subspaces, not reduced to a point. By lemma 

1875: \ref{lemm_cone_e6_1}, $X^Q$ and $X={(X^Q)}^P$ contain a $\p^4$. But

1876: this implies that $X^Q \subset I_X$ is at most 1-dimensional

1877: (see the proof of theorem \ref{theo_cone}), and we get

1878: a contradiction.

1879: 

1880: \lpara

1881: 

1882: Let finally $X \subset E_6/P_3$ be linear. Assume $X$ is not reduced to

1883: a point. Let $h \in X^Q$; we have $X \subset I_h$.

1884: On the other hand, since

1885: $X$ is not a point, by lemma \ref{lemm_cone_e6_2}, it must

1886: contain $I_h$. Therefore, $X = I_h$.

1887: \end{proo}

1888: 

1889: 

1890: 

1891: 

1892: %******************************************************************************

1893: 

1894: %******************************************************************************

1895: 

1896: %******************************************************************************

1897: 

1898: 

1899: 

1900: 

1901: 

1902: 

1903: 

1904: 

1905: \sectionplus{Examples of dual varieties}

1906: 

1907: \label{section_exemple}

1908: 

1909: 

1910: \subsectionplus{Dual varieties of isotropic Grassmannians}

1911: 

1912: 

1913: Let $V$ be a vector space, $B:V \rightarrow V^*$ a bilinear form.

1914: If $\epsilon = \pm 1$ and

1915: $\tr B = \epsilon B$, we say that $B$ is $\epsilon$-symmetric.

1916: Assume that this is the case. Let $r$ be an integer; we 

1917: consider the variety $\G_B(r,V)$ of isotropic subspaces of $V$ of dimension

1918: $r$. The aim of this subsection is to describe the dual of

1919: $\G_B(r,V)$ in $G(r,V^*)$ in case $2r<\dim\ V$ (the other cases would be

1920: similar). Note that we don't assume that $B$ is an isomorphism.

1921: 

1922: We have a rational map $\G_B(r,V) \dasharrow \G(r,V^*)$ which maps a linear

1923: subspace to its orthogonal, and which is well-defined at the point $\alpha$

1924: \iff $L_\alpha$ does not meet the kernel of $B$. Assuming there are

1925: such points, we call co-isotropic Grassmannian the image of this

1926: rational map.

1927: 

1928: \begin{prop}

1929: \label{prop_symmetric}

1930: Assume $\epsilon=1$. Then $\G_B(r,V)$ is suitable \iff and

1931: $r \leq \rank(B)$. In this case, the dual variety of the isotropic Grassmannian

1932: $G_B(r,V)$ is the co-isotropic Grassmannian.

1933: \end{prop}

1934: 

1935: \begin{prop}

1936: \label{prop_skew}

1937: Assume $\epsilon=-1$. Then $\G_B(r,V)$ is suitable \iff $r$ is even

1938: and $r \leq \rank(B)$. In

1939: this case, the dual variety of the isotropic Grassmannian

1940: $G_B(r,V)$ is the co-isotropic Grassmannian.

1941: \end{prop}

1942: 

1943: \begin{proo}

1944: We prove propositions \ref{prop_symmetric} and \ref{prop_skew}

1945: simultaneously. Let $x \in \G_B(r,V)$ be generic. Under the natural

1946: isomorphism $T_x \G(r,V) \simeq Hom(L_x,V/L_x)$, we have

1947: the inclusion

1948: $T_x \G_B(r,V) \supset Hom(L_x,L_x^\bot/L_x)$,

1949: where $L_x^\bot$ denotes the orthogonal of $L_x$ with respect to

1950: $B$. It follows that if $\codim\ L_x^\bot < r$, then 

1951: $N_x^* \G_B(r,V)$ does not meet the open orbit in $T_x^* \G_B(r,V)$.

1952: If $r > \rank(B)$, this occurs for all $x \in \G_B(r,V)$, hence

1953: $\G_B(r,V)$ is not suitable.

1954: 

1955: Assume $r \leq \rank(B)$.

1956: Now, let $x \in \G_B(r,V)$ such that $\codim\ L_x^\bot = r$.

1957: Denote $Q_x = V/L_x$; we have a morphism 

1958: $Q_x \rightarrow L_x^*$,

1959: induced by $B$. Clearly, $T_x \G_B(r,V) \subset Hom(T_x,Q_x)$ 

1960: is the

1961: subspace of linear maps such that the composition 

1962: $L_x \rightarrow Q_x \rightarrow L_x^*$

1963: is $(-\epsilon)$-symmetric. Therefore,

1964: the normal space of $\G_B(r,V)$ at $x$ identifies with

1965: $\epsilon$-symmetric maps $L_x^* \rightarrow L_x$. 

1966: Since $\G_B(r,V)$ will be suitable \iff there are such maps of rank $r$,

1967: this occurs in all cases if $\epsilon = 1$ and exactly when $r$ is even

1968: when $\epsilon = -1$.

1969: 

1970: Now, the computation of the dual variety is straightforward : since

1971: we have already remarked that 

1972: $T_x \G_B(r,V) \supset Hom(L_x,L_x^\bot/L_x)$,

1973: the image of a generic conormal form at $x$ under the rational map

1974: $q:T^*_x \G(r,V) \dasharrow G(r,V^*)$ is the element in

1975: $\G(r,V^*)$ corresponding to $L_x^\bot$.

1976: \end{proo}

1977: 

1978: 

1979: \subsectionplus{Schubert varieties and quivers in the fundamental case}

1980: 

1981: \label{subsection_carquois}

1982: 

1983: In this subsection, I recall that to a cominuscule homogeneous space one

1984: can naturally associate a quiver, such that Schubert cells are parametrised

1985: by some subquivers. I also recall the Hasse diagram of a representation,

1986: and show how the quiver of a cominuscule homogeneous space can be

1987: identified with the Hasse diagram of a tangent space.

1988: This identification is

1989: due to Nicolas Perrin and

1990: Laurent Manivel.

1991: Then, I show that this identification behaves well as far as Schubert 

1992: subvarieties are concerned. Finally, I extend these results to

1993: $E_6/P_3$, which is not a cominuscule homogeneous space.

1994: 

1995: \lpara

1996: 

1997: The quiver of a cominuscule homogeneous space has been first introduced by

1998: N. Perrin \cite[definition 3.2]{carquois};

1999: here we use the slightly different definition

2000: \cite[definition 2.1]{quantique}. Recall that

2001: $\G(r,V),\G_Q^+(2p+1,4p+2)$ and $E_6/P_1$ are cominuscule spaces (in

2002: fact even minuscule). The quiver is defined using a reduced expression

2003: of $w_{G/P}$, the shortest element in the class of $w_0$ in $W/W_P$

2004: ($w_0$ is the longest element in $W$).

2005: Choose a reduced expression

2006: $w_{G/P} = s_{\beta_1} \cdots s_{\beta_N}$, with $N = \dim G/P$; the

2007: vertices of the quiver 

2008: $Q_{G/P}$ are in bijection with $[1,\ldots,N]$, and we refer

2009: to \cite[definition 2.1]{quantique} for the definition of the arrows.

2010: The quivers may be illustrated by relevant

2011: examples as follows~: \vspace{-.8cm}

2012: 

2013: $$

2014: \begin{array}{ccccc}

2015: \input{g37}  &&  \input{gq5}  &&  \input{e6}  \\

2016: \G(3,7) && \G_Q^+(5,10) && E_6/P_1

2017: \end{array}

2018: $$

2019: In these pictures, all arrows are going down. Moreover, we will use

2020: the definition of height of a vertex of such a quiver. More or less by

2021: definition (see \cite[definition 4.7]{carquois}), it is the height of

2022: the vertex in the above drawing, where by convention the lowest vertex

2023: has height 1 (so the highest vertex has height respectively 6,7,11 for

2024: $\G(3,7),\G_Q^+(5,10),E_6/P_1$).

2025: \lpara

2026: 

2027: Later we will have to identify this quiver with a Hasse diagram. 

2028: Let $V$ be a representation of a semi-simple group $\Lambda$.

2029: Let us recall that the Hasse diagram

2030: of $V$ is a quiver defined as follows.

2031: The vertices of this quiver are the weights of $V$, and

2032: there is an arrow from $\lambda_1$ to $\lambda_2$ if and only if

2033: $\lambda_2 - \lambda_1$ is a simple root. For example, the Hasse diagram

2034: of the $8$-dimensional representation of $Spin_8$ is given on the left~:

2035: 

2036: $$

2037: \begin{array}{ccc}

2038: \input{hasseq8} && \input{racineq10} \\

2039: Spin_8 & \hspace{1cm} & \mbox{Roots for }\Q^8

2040: \end{array}

2041: $$

2042: 

2043: \begin{prop}

2044: \label{hasse}

2045: Let $G/P$ be cominuscule and let $x \in G/P$ be the base point. 

2046: Let $\Lambda$ be a Levi factor

2047: of the stabilisor of $x$. Then the quiver $Q_{G/P}$

2048: of $G/P$ is isomorphic with the

2049: Hasse diagram $H_{G/P}$ of the $\Lambda$-module 

2050: $\widehat{T_xX}/L_x$.

2051: \end{prop}

2052: \noindent

2053: If $G/P \subset V$, recall that $\widehat{T_xG/P} \subset V$ 

2054: is the affine tangent space at $x$; it contains the

2055: line $L_x \subset V$ represented by $x \in \p V$,

2056: so that it makes sense to consider the quotient

2057: $\widehat{T_xG/P}/L_x$.

2058: We have stated this result without proof in

2059: \cite[proposition 7]{quantique}. In this

2060: article

2061: I need the explicit isomorphism, this is why I sketch the proof,

2062: leaving details to the reader.

2063: \begin{proo}

2064: It is known that to each vertex of the quiver one can associate a root of

2065: $G$. In fact, choose a reduced expression

2066: $w_{G/P} = s_{\beta_1} \ldots s_{\beta_N}$ and set

2067: $$ \alpha_i = s_{\beta_N} \circ \ldots \circ s_{\beta_{i+1}} (\beta_i). $$

2068: Since two different reduced expressions for

2069: $w_{G/P}$ only differ by commutation relations,

2070: it is easy to check that the induced map from the set of vertices of

2071: the quiver

2072: to the set of roots is well-defined (it does not depend on the reduced

2073: expression). In the following, we consider that a reduced expression is

2074: chosen, thus identifying this set of vertices with $[1,N]$.

2075: 

2076: For example, if $G/P$ is a smooth 8-dimensional quadric,

2077: then its quiver, and the corresponding roots, are given above

2078: (here we have shitfted the indices,

2079: denoting $(\epsilon_0,\ldots,\epsilon_4)$ a basis of the weight lattice of

2080: $Spin_{10}$). Note that the highest weight of the corresponding

2081: $Spin_{10}$-representation is $\epsilon_0$, 

2082: and that we recover the Hasse diagram

2083: of $Spin_8$ by considering the weights $\epsilon_0 - \alpha_i$.

2084: 

2085: By \cite[proposition 4.9]{carquois}, we may reduce the proof of our

2086: proposition to the particular case of a quadric

2087: of any dimension, as above, because if

2088: there is an arrow $i \rightarrow j$ in the quiver of $G/P$, then $i$

2089: and $j$ belong to a subquiver of $Q_{G/P}$ isomorphic with the quiver

2090: of a quadric. It is also possible (and probably shorter)

2091: to check directly in each case that if $\omega$ denotes the highest

2092: weight of $\Gamma(G/P,\co(1))$, then the set 

2093: $\{\omega - \alpha_i : 1\leq i \leq N\}$ is exactly the set of weights

2094: of the tangent space at the base point of $G/P$, and that the

2095: bijection $i \mapsto \omega - \alpha_i$ is an

2096: isomorphism of quivers $Q_{G/P} \rightarrow H_{G/P}$.

2097: \end{proo}

2098: 

2099: \lpara

2100: 

2101: Given $[w] \in W/W_P$, we associate the Schubert subvariety 

2102: $C_{[w]} \subset G/P$ which is the $B$-orbit closure of 

2103: $[w] \in G/P$. Assuming that $w$ is the minimal length representative

2104: of its class, we choose a reduced decomposition of $w$, and this

2105: defines a subquiver $Q_w$ of the quiver $Q_{G/P}$

2106: which is an order ideal

2107: (this means that if $i \rightarrow j$ is an arrow in $Q_{G/P}$ and $i

2108: \in Q_w$, then $j \in Q_w$~: 

2109: see \cite[proposition 4.5]{carquois}). We can also consider the

2110: subset $H_w$ of $H_{G/P}$ which elements are the weights of

2111: $w^{-1}.T_{[w]}C_{[w]} \subset T_{[e]} G/P$. The following proposition

2112: will be useful to compute the dual variety of $C_{[w]}$, because it

2113: describes the tangent bundle of $C_{[w]}$~:

2114: 

2115: \begin{prop}

2116: \label{prop_tangent_cw}

2117: Under the isomorphism $Q_{G/P} \simeq H_{G/P}$

2118: of proposition \ref{hasse}, we have

2119: $Q_w = H_w$.

2120: \end{prop}

2121: \begin{proo}

2122: Recall that $\omega$ denotes the highest weight of 

2123: $\Gamma(G/P,\co(1))$. All the weights of 

2124: $\widehat{T_x G/P}/L_x$ are of the

2125: form $\omega + \alpha$, where $\alpha$ are all the roots not in

2126: $\plie = Lie(P)$ (therefore $\alpha$ is a negative root). 

2127: All the weights of $T_{[w]} C_{[w]}$ are of the form

2128: $w.\omega + \beta$, with $\beta$ a positive root. Therefore, if

2129: $\omega + \alpha$ is a weight of $w^{-1}.T_{[w]}C_{[w]}$,

2130: $w.\alpha$ must be a positive root. So $\alpha$ must be a negative

2131: root sent by $w$ to a positive root. Denote $l(w)$ the length of $w$;

2132: there are $l(w)$ such roots,

2133: namely $\{-\alpha_i:1\leq i \leq l(w)\}$. Since $l(w)$ is also the

2134: dimension of $C_{[w]}$, it follows that

2135: $H_w$ is exactly the set of weights of the form 

2136: $\omega - \alpha_i,1\leq i \leq l(w)$, so the proposition follows.

2137: \end{proo}

2138: 

2139: \lpara

2140: 

2141: We now consider the case of $E_6/P_3$. Let 

2142: $[w] \in W/W_3$; we want to define a quiver $Q_{E_6/P_3}$ and a 

2143: subquiver $Q_w$ which pictures the tangent bundle of $C_{[w]}$.

2144: Since $E_6/P_3$ is not cominuscule, the quiver defined as in

2145: \cite[definition 3.2]{carquois} is not well-defined

2146: (it depends on a reduced expression of $w_{E_6/P_3}$), and

2147: as we have already seen, the cotangent bundle $T^* E_6/P_3$ is no

2148: longer irreducible, so its Hasse diagram is not suitable neither.

2149: 

2150: But our luck is that for $f \in T^*_\alpha E_6/P_3$, $q(f)$ only

2151: depends on $\pi(f) \in L_\alpha \otimes \wedge^2 Q_\alpha^*$;

2152: therefore, what we care for is not really the conormal bundle of

2153: $C_{[w]}$, but rather its projection to the bundle

2154: $L \otimes \wedge^2 Q^*$. This is why we consider the following~:

2155: 

2156: \begin{defi}\ 

2157: \begin{itemize}

2158: \item

2159: Let $[e] \in E_6/P_3$ be the base point, and let $\Lambda$ denote a Levi

2160: factor of $P_3$.

2161: \item

2162: Let $Q_{E_6/P_3}$ denote the Hasse diagram of the $\Lambda$-module

2163: $L_{[e]}^* \otimes \wedge^2 Q_{[e]} \subset T_{[e]} E_6/P_3$.

2164: \item

2165: For $[w] \in W/W_3$ with $w$ the minimal length representative,

2166: let $Q_w \subset Q_{E_6/P_3}$ denote the set of

2167: weights of 

2168: $w^{-1}.T_{[w]} C_{[w]}\ \cap\ 

2169: (L_{[e]} \otimes \wedge^2 Q_{[e]}^*)$.

2170: \end{itemize}

2171: \end{defi}

2172: 

2173: \begin{prop}

2174: \label{prop_order_ideal}

2175: For $[w] \in W/W_3$, $Q_w \subset Q_{E_6/P_3}$ is an order ideal.

2176: \end{prop}

2177: \begin{proo}

2178: For $a \in \{-2,-1,0,1,2\}$, let $\g_k \subset \e_6$ denote

2179: $

2180: \displaystyle

2181: \bigoplus_{

2182: \begin{array}{c}

2183: \alpha = \sum_j k_j \alpha_j\\

2184: k_3 = a

2185: \end{array}

2186: } \g_{\alpha}

2187: $

2188: (by this I mean that the Cartan subalgebra is included in $\g_0$).

2189: We have $\g = \g_{-2} \oplus \g_{-1} \oplus \g_0 \oplus \g_1

2190: \oplus \g_2$ and $Lie(P_3) = \g_0 \oplus \g_1

2191: \oplus \g_2$. The tangent space $T_{[e]} E_6/P_3$ decomposes as

2192: $\g_{-2} \oplus \g_{-1}$; let

2193: ${\cal P}$ denote the weights of 

2194: $\widehat{T_{[e]} E_6/P_3}/L_{[e]}$ which are of the

2195: form $\omega + \alpha$, with $\alpha$ a root of $\g_{-1}$ (${\cal P}$

2196: is also the set of weights of $L_{[e]} \otimes \wedge^2 Q_{[e]}^*$).

2197: I claim that $w$ induces an increasing bijection between

2198: ${\cal P}$ and its image.

2199: The proposition follows from this claim because 

2200: $Q_{[w]}$ is the set of weights of 

2201: $w^{-1}.T_{[w]} C_{[w]}$ which are in ${\cal P}$; arguing as in the

2202: proof of proposition \ref{prop_tangent_cw}, this is the set of roots

2203: of $\g_{-1}$ which are mapped to a positive root by $w$, and this is

2204: obviously an order ideal since $w$ is increasing.

2205: 

2206: To prove the claim, we note that

2207: $L_{[e]} \otimes \wedge^2 Q_{[e]}^*$ is a minuscule 

2208: $\Lambda$-representation, since $\Lambda$ contains 

2209: $SL_2 \times SL_5$. Therefore $W_P$ permutes transitively the roots in 

2210: $\g_{-1}$. Let $\alpha_0$ be the highest root of $\g_{-1}$, let

2211: $\alpha_1,\alpha_2$ be roots of $\g_{-1}$ and assume

2212: $\alpha_1 \leq \alpha_2$. We can find $w_1,w_2 \in W_P$ such that

2213: $\alpha_i = w_i . \alpha_0$, assume moreover that $w_1,w_2$ are

2214: minimal such elements. Since $\alpha_1 \leq \alpha_2$, we

2215: have $w_1 \geq w_2$ for the Bruhat order, so that we may assume that a

2216: $w_2$ is a product of reflexions appearing in a reduced expression of

2217: $w_1$. Since $w$ is a minimal length representative in $W/W_3$, the

2218: product $w.w_1$ is still a reduced expression, so 

2219: $w.w_1 \geq w.w_2$, and so $w.\alpha_1 \leq w.\alpha_2$, as claimed.

2220: \end{proo}

2221: 

2222: \begin{rema}

2223: The same proof works for any $G/P$, as soon as $\g_{-1}$ is a

2224: minuscule $\Lambda$-representation, with the notations of the proof.

2225: \end{rema}

2226: 

2227: 

2228: 

2229: 

2230: %******************************************************************************

2231: 

2232: 

2233: 

2234: 

2235: 

2236: \subsectionplus{Schubert varieties and dual varieties}

2237: 

2238: \label{subsection_combinatoire}

2239: 

2240: The usual dual variety of a linear subspace is again a linear 

2241: subspace. The

2242: goal of this section is to generalise this result for Schubert varieties.

2243: 

2244: \begin{prop}

2245: \label{prop_schubert}

2246: Let $X \subset G/P$ be a suitable

2247: Schubert variety. Then $X^Q \subset G/Q$

2248: is a Schubert variety.

2249: \end{prop}

2250: \pr

2251: In fact, $X^Q$ is a $B$-stable (proposition \ref{foncteur})

2252: irreducible closed subvariety of $G/Q$.

2253: \qed

2254: 

2255: \lpara

2256: 

2257: Recall that $B$-stable

2258: Schubert varieties in $G/P$ are parametrised by the quotient set

2259: $W/W_P$. For $[w] \in W/W_P$ (resp. $[x] \in W/W_Q$),

2260: we denote as in the previous subsection

2261: $C_{[w]} = \overline {B.[w]} \subset G/P$ the corresponding Schubert

2262: subvariety (resp. $D_{[x]} = \overline {B.[x]} \subset G/Q$).

2263: In the rest of this article, I give a description of the $[w]$'s 

2264: such that the $C_{[w]}$ is suitable, and of the

2265: element in $W/W_Q$ corresponding to the dual Schubert variety, in the

2266: fundamental cases. According to

2267: section \ref{section_reduction}, this is enough to describe all dual varieties

2268: of Schubert varieties. The strategy for this description is first to use

2269: a $T$-fixed point argument, to reduce the task to a purely

2270: combinatorial one. In the types $A$ and $D$, I then give an

2271: explicit solution of this combinatorial problem. For the exceptional

2272: cases, my description of the dual Schubert varieties is not really

2273: explicit, but to compute them there is in principal only a finite

2274: number of computations to make.

2275: 

2276: \lpara

2277: 

2278: So we fix the minimal $G$-representation $V$ such that $G/P \subset \p V$. We

2279: denote $V = \oplus V_\lambda$ (resp. $V^* = \oplus V^*_\mu$)

2280: the weight decomposition of $V$ (resp. $V^*$).

2281: Let $C_{[w]} = \overline {B.[w]}$ be a Schubert

2282: variety. Recall that $\overline{N_{[w]} C_{[w]}}$

2283: denotes the variety of $y$'s in

2284: $G/Q$ which are tangent to $C_{[w]}$ at $[w]$, see definition

2285: \ref{cotangent}.

2286: 

2287: \begin{lemm}

2288: \label{t_fixe}

2289: Let $[x] \in W/W_Q$ such that 

2290: $C_{[w]}^Q = \overline {B.[x]}$. We have 

2291: $[x] \in \overline{N_{[w]} C_{[w]}}$.

2292: \end{lemm}

2293: \begin{proo}

2294: First, notice that $C_{[w]}^Q = \overline

2295: {B.\overline{N_{[w]}C_{[w]}}}$. In fact,

2296: $B.\overline{N_{[w]}C_{[w]}}$ contains the set of $y$'s in $G/Q$

2297: which are tangent at a point in $B.[w]$, therefore at a generic point

2298: of $C_{[w]}$.

2299: 

2300: Let $\mu_0$ be the highest weight of $V^*$ and denote $\mu = x.\mu_0$. 

2301: Let $y \in V^*$ such that $[y] \in C_{[w]}^Q \subset \p V^*$.

2302: Use the weight decomposition

2303: of $V^*$ to write $y = \sum_{\mu'} y_{\mu'}$.

2304: Since $[y] \in C_{[w]}^Q$ which is the closure of the $B$-orbit of the weight

2305: line of weight $\mu$,

2306: $y_{\mu'} = 0$ if $\mu' \not \geq \mu$. Assume that 

2307: $\forall [y] \in \overline{N_{[w]} C_{[w]}},y_{\mu}=0$.

2308: It would then follow from

2309: the first point that $\forall y \in C_{[w]}^Q, y_{\mu}=0$, contradicting

2310: $[x] \in C_{[w]}^Q$.

2311: 

2312: Therefore, there exists $[y]$ in $\overline{N_{[w]}C_{[w]}}$ such that 

2313: $y_\mu \not = 0$. Since $\overline{N_{[w]}C_{[w]}}$

2314: is $T$-stable and $V_\mu$ is 

2315: one-dimensional,

2316: we have $[x] \in \overline{N_{[w]} C_{[w]}}$.

2317: \end{proo}

2318: 

2319: I now explain how to compute

2320: the element $[x]$ of the previous lemma. 

2321: We want to take into account the case of $E_6/P_3$, which cotangent

2322: bundle is not irreducible. Recall that in this case there is a natural

2323: bundle morphism $\pi : T^* E_6/P_3 \rightarrow L \otimes \wedge^2 Q^*$. 

2324: To have uniform notations, in the other cases we denote

2325: $\pi : T^* G/P \rightarrow T^* G/P$ the identity and $\overline q = q$.

2326: 

2327: Decompose

2328: $\pi(T_{[e]}^* G/P)$ as a sum of weight spaces for the action of a Levi

2329: subgroup of $P$ : $\pi(T_{[e]}^* G/P) = \oplus T^*_\tau$, and write similarly

2330: $\scal{I_{[e]}} = \oplus_\nu N_\nu$, where if $[w'] \in W/W_P$,

2331: $\scal{I_{[w']}} \subset V^*$ denotes the linear span of the Schubert variety

2332: $I_{[w']} \subset G/Q \subset \p V^*$ (see notation \ref{i_h}).

2333: It can be easily checked directly on the

2334: examples that all the weight spaces $T^*_\tau$ and $N_\nu$ have dimension 1.

2335: The rational map 

2336: $\overline q : \pi(T_{[e]}^* G/P) \dasharrow G/Q \subset \p V^*$

2337: is given by a list

2338: of polynomial functions $\pi(T_{[e]}^* G/P) \rightarrow N_\nu$

2339: of the same degree $d$ and with values in the complex line $N_\nu$.

2340: The polarisations of these polynomials yield $d$-linear maps

2341: $T_{\tau_1}^* \times \cdots \times T_{\tau_d}^* \rightarrow N_\nu$, which

2342: will be denoted $P_{\tau_1,\ldots,\tau_d;\nu}$; remark that the space of

2343: such $d$-linear maps has dimension 1. Given $w \in W$, we denote

2344: ${\cal P}_w$ the set of weights $\nu$ such that there exist weights

2345: $\tau_1, \ldots , \tau_d$ of 

2346: $\pi(w^{-1}.N^*_{[w]} C_{[w]}) \subset \pi(T_{[e]}^* G/P)$

2347: such that $P_{\tau_1,\ldots,\tau_d;\nu}$ does not vanish.

2348: \lpara

2349: 

2350: \begin{prop}

2351: \label{prop_combinatoire}

2352: Let $[w] \in W/W_P$, with $w$ its minimal length representative.

2353: The variety $C_{[w]}$ is suitable \iff ${\cal P}_w$ is not empty. In this case,

2354: if we denote $[x] \in W/W_Q$ such that $C_{[w]}^Q = C_{[x]}$,

2355: then $x.\mu_0$ equals $w.\mu_1$, where $\mu_1$ is the

2356: lowest weight in ${\cal P}_w$.

2357: \end{prop}

2358: \begin{proo}

2359: In fact, by lemma \ref{t_fixe} and its proof, $x.\mu_0$ is the lowest weight

2360: $\mu$, if any,

2361: such that the $\mu$-component of the restriction of the rational

2362: map 

2363: $\pi(T^*_{[w]} G/P) \dasharrow G/Q$ to $N^* C_{[w]}$

2364: does not vanish identically. 

2365: 

2366: The weights of $\scal{I_{[w']}}$ for 

2367: $[w'] \in W/W_P$ are some weights of $V^*$,

2368: a set on which $W$ acts; therefore it makes sense to talk of

2369: $w''.\mu'$, for $w'' \in W$ and $\mu'$ a weight of $\scal{I_{[w']}}$.  

2370: I claim that $w$ induces an increasing bijection between the weights of

2371: $\scal{I_{[e]}}$ and those of $\scal{I_{[w]}}$. 

2372: The argument is similar to that of proposition \ref{prop_order_ideal}.

2373: In fact, a weight of $\scal{I_{[e]}}$ can be

2374: written as $v.\mu_0$, with $v \in W_P$ and $\mu_0$ the highest weight

2375: of $V^*$. Given two such weights $v_1.\mu_0 \geq v_2.\mu_0$, we can assume

2376: that $v_2$ is the minimal length representative of its class in 

2377: $W_P / W_{P \cap Q}$.

2378: Thus $v_1$ can be written as a product of some reflections which occur

2379: in a reduced expression of $v_2$. Since $v_2 \in W_P$ and $w$ is a minimal

2380: length representative, $l(wv_2) = l(w) + l(v_2)$. Therefore

2381: $wv_1 \leq wv_2$ for the Bruhat order, and thus

2382: $w.(v_1.\mu_0) \geq w.(v_2.\mu_0)$. This proves the claim.

2383: 

2384: Since the

2385: rational map $\overline q : \pi(T^* G/P) \dasharrow G/Q$

2386: is $G$-equivariant, the

2387: weight $\mu$

2388: of the proof of lemma \ref{t_fixe}

2389: is also $w.\mu_1$, where $\mu_1$ is the lowest weight such

2390: that the $\mu_1$-component of the restriction of the rational

2391: map $T^*_{[e]} G/P \dasharrow G/Q$ to 

2392: $w^{-1}.N^*_{[w]} C_{[w]}$ does not vanish identically.

2393: Obviously $\mu_1$ is the lowest weight $\nu$ such that some

2394: $P_{\tau_1,\ldots,\tau_d;\nu}$ 

2395: with $\tau_1,\ldots,\tau_d$ weights of $w^{-1}.N^*_{[w]} C_{[w]}$,

2396: does not vanish, so the proposition is proved.

2397: \end{proo}

2398: 

2399: We illustrate our method with the easy example $G/P = \p V$.

2400: Let $(e_1,\ldots,e_n)$ be a basis of $V$, let $k$ be an integer and let

2401: $L_k = Vect(e_1,\ldots,e_k),\ M_k = Vect(e_{k+1},\ldots,e_n)$.

2402: We consider the Schubert variety $X = \p L_k \subset \p V$ and compute

2403: its dual variety. The corresponding element of the Weyl group is the

2404: transposition $w=(1k)$. We have

2405: $T_{[w]} X \simeq Hom(e_k,L_k/e_k)$, so

2406: $w^{-1}. T_{[w]} X \simeq Hom(e_1,L_k/e_1)$ and so

2407: $w^{-1}. N^*_{[w]} X \simeq Hom(M_k,e_1)$. Since $q$ is defined taking the

2408: kernel, the lowest weight in $q(w^{-1}. N^*_{[w]})$ is 

2409: $\mu_1 = -\epsilon_{k+1}$. We have $w.\mu_1 = \mu_1$, so that the dual

2410: variety of $X$ is the $B$-orbit closure of $e_{k+1}^*$, as

2411: expected. Note that in this example it would have been easier to compute

2412: directly the lowest weight in

2413: $q(N^*_{[w]}C_{[w]})$, instead of applying first $w^{-1}$ and then $w$. In fact,

2414: this is what we will do to compute dual varieties of Schubert varieties

2415: in the cases $G/P = \G(r,V)$ and $G/P = \G_Q^+(2p+1,4p+2)$.

2416: \lpara

2417: 

2418: Recall from subsection \ref{subsection_carquois}

2419: the definition of height of a vertex of the quiver

2420: $Q_{G/P} = H_{G/P}$. We denote $h_0$ the maximal $h$ such that there exist

2421: $\tau_1,\ldots,\tau_d \in H_{G/P}$, $\nu$ a weight of $I_{[e]}$ such

2422: that $h(\tau_i) \geq h$ and $P_{\tau_1,\ldots,\tau_d;\nu} \not = 0$.

2423: We have the following values for $h_0$ (I have also indicated the

2424: height $h_{\max}$ of the heighest element of $Q_{G/P}$)~:

2425: $$

2426: \begin{array}{ccccc}

2427: G/P & \G(r,n) & \G_Q^+(2p+1,4p+2) & E_6/P_1 & E_6/P_3\\

2428: h_{\max} & n-1 & 4p-1 & 11 & 8\\

2429: h_0 & \max(r,n-r) & 2p+1 & 8 & 5

2430: \end{array}

2431: $$

2432: 

2433: \begin{theo}

2434: \label{theo_adapte}

2435: The Schubert subvariety $C_{[w]}$ is suitable \iff all the vertices of

2436: $Q_w$ have height at most $h_0-1$.

2437: \end{theo}

2438: \begin{proo}

2439: Unfortunately, I don't know how to prove in a uniform way this

2440: theorem. It will follow from propositions \ref{grass_adapte},

2441: \ref{prop_spinoriel}, \ref{prop_e6_1_adapte} and

2442: \ref{prop_e6_2_adapte}. The proof of these propositions also imply the

2443: above given values of $h_0$.

2444: \end{proo}

2445: 

2446: 

2447: 

2448: 

2449: %******************************************************************************

2450: 

2451: 

2452: 

2453: 

2454: \subsectionplus{Case of Grassmannians}

2455: 

2456: \label{subsection_schubert_grassmannienne}

2457: 

2458: Recall that $V$ is an $n$-dimensional vector space.

2459: We will

2460: parametrise Schubert varieties in $\G(r,V)$ by increasing lists of $r$ 

2461: integers,

2462: instead of partitions, because duality will appear easier to formulate

2463: in this way.

2464: The list $(l_i)$ will correspond to the Schubert variety 

2465: $C_l \subset \G(r,V)$ (resp. in $D_l \subset \G(r,V^*)$) which is the

2466: $B$-orbit closure of the linear space spanned by the

2467: $l_i$'s $T$-eigenvectors in $V$ (resp. in $V^*$).

2468: For $x \in \{1,\ldots,n\}$, we will write $x \in l$ to mean that there

2469: exists $i$ such that $x = l_i$.

2470: The $T$-fixed points in $V$ (resp. $V^*$) will be denoted $e_i$

2471: (resp. $e_i^*$).

2472: The $T$-fixed point whose $B$-orbit is dense in $C_l$ 

2473: (resp. $D_l$) will be denoted $x_l$

2474: (resp. $y_l$).

2475: The Bruhat order on Schubert cells is given by

2476: $l \leq m$ \iff $\forall i,l_i \leq m_i$. If $x_1,\ldots,x_r$ are 

2477: distinct integers

2478: not necessarily increasing, we denote the list obtained reordering the

2479: $x_i$ as $[x_1,\ldots,x_r]$.

2480: 

2481: Let $T_l$ denote $Vect(e_i : i \in l)$ and let

2482: $Q_l$ denote $Vect(e_i : i \not \in l)$.

2483: The tangent

2484: space at $x_l$ identifies with

2485: $Hom(Q_l,T_l)$.

2486: A weight in this space is given by a couple 

2487: $(x,y):x \in l,y \not \in l$.

2488: Recall that the rational map

2489: $$T_{x_l}^* \G(m,n) \simeq 

2490: Hom(Q_l,T_l) \dasharrow \G(r,V^*)$$

2491: is given by $\varphi \mapsto \ker \varphi$.

2492: Thus the degree of $q$ is $r$ and

2493: with the notations before proposition \ref{prop_combinatoire}, 

2494: we have~:

2495: 

2496: \begin{fait}

2497: \label{fait_P_grassmannienne}

2498: The multilinear form

2499: $P_{(x_1,y_1),\ldots,(x_r,y_r);l'}$ does not vanish \iff the $x_i$'s 

2500: and the $y_i$'s are all distinct, and $l'$ is the set of the $y_i$'s.

2501: \end{fait}

2502: 

2503: \noindent

2504: Given a list $l$, we consider the list $l^*$ defined inductively by

2505: $$

2506: l^*_i = \min \{ y : y>y_{i-1}, y>x_i, \forall j, y\not = x_j \}.

2507: $$

2508: 

2509: \begin{lemm}

2510: We have $\forall i,l_i^* \leq n$ \iff $\forall i \in \{1,\ldots,r\},

2511: l_i<n+2i-2r$.

2512: \end{lemm}

2513: \noindent

2514: In terms of partitions, this means that the $i$-th part

2515: must be at least $r+1-i$.

2516: \begin{proo}

2517: Let $i$ be an integer. The integers $l_j$ for $j>i$ and $l_j^*$ for

2518: $j \geq i$ are strictly greater than $l_i$

2519: and distinct, so the lemma follows.

2520: \end{proo}

2521: 

2522: \noindent

2523: As the following proposition

2524: shows, $l \mapsto l^*$ is the combinatorial model for the duality of

2525: Schubert varieties in Grassmannians~:

2526: 

2527: \begin{prop}

2528: \label{grass_adapte}

2529: $C_l$ is suitable \iff $\forall i \in \{1,\ldots,r\},l_i < n+2i-2r$.

2530: If $C_l$ is suitable then $C_l^Q = D_{l^*}$.

2531: \end{prop}

2532: \begin{proo}

2533: With the previous notations, the weights of the conormal

2534: space $N^*_{x_l} C_l$ are the couples

2535: $(x,y)$ with $x \in l,y \not \in l$, and $y>l_x$.

2536: 

2537: By proposition \ref{prop_combinatoire} and the comment after it,

2538: $C_l$ is suitable \iff there are lists $[y_1,\ldots,y_r]$

2539: and $x_1,\ldots,x_r$ with

2540: $(x_i,y_i)$ a weight of $N^*_{x_l} C_l$ and

2541: $P_{(x_1,y_1),\ldots,(x_r,y_r);l'} \not = 0$.

2542: In this case, if we denote $l'$ the list

2543: such that $C_l^Q = D_{l'}$, then $l'$ is the minimal possible such list.

2544: Moreover, in order that 

2545: $P_{(x_1,y_1),\ldots,(x_r,y_r);l'} \not = 0$,

2546: all $x_i$ must be distinct and we must have

2547: $\{x_i\} = l$, so we

2548: may assume by symmetry that $x_i = l_i$.

2549: It is easy to check that the set of such $l'$ is not

2550: empty \iff $\forall i \in \{1,\ldots,r\},l_i<n+2i-2r$. In fact, if

2551: $l_i \geq n+2i-2r$, then the values $y_j$ and $x_j$ for 

2552: $i \leq j \leq r$ must be distinct

2553: and between $n+2i-2r$ and $n$, a contradiction. Conversely, if

2554: $\forall i,l_i < n+2i-2r$, one may choose $y_i = l_i^*$.

2555: 

2556: We now show that $l^*$ is indeed the minimal list. Let

2557: $[y_1,\ldots,y_r]$ be any list with 

2558: $\forall i, y_i>l_i$ and $\forall i,j,y_i \not = l_j$.

2559: Let $(z_1,\ldots,z_r)$ be the corresponding ordered list

2560: (ie $\{y_1,\ldots,y_r\} = \{z_1,\ldots,z_r\}$ and

2561: $z_1 < z_2 < \cdots < z_r$). Then we have $z_1 > l_1$

2562: and $\forall j, z_1 \not = l_j$,

2563: so $z_1 \geq l_1^*$. Say $z_i = y_{\sigma(i)}$. If $z_1 < l_2$ then

2564: $\sigma(1)=1$. Thus in any case $z_2 > l_2$, so $z_2 \geq l^*_2$. By induction

2565: it follows that $\forall i,l^*_i \leq z_i$, so $l^*$ is the minimal possible

2566: list, and proposition \ref{prop_combinatoire} finishes the proof.

2567: \end{proo}

2568: 

2569: 

2570: We illustrate this proposition with two examples. The array below

2571: computes two dual varieties in $\G(3,8)$. It pictures the fact that

2572: for $l=(2,4,5)$ we have $l^*=\lambda=(3,6,7)$, and that for $m=(2,4,6)$ we have

2573: $m^*=\mu=(3,5,7)$~:

2574: $$

2575: \begin{array}{ccc}

2576: \young(\hfil l\lambda\hfil\hfil\hfil\hfil\hfil,\hfil\hfil\hfil l\hfil\lambda\hfil\hfil,\hfil\hfil\hfil\hfill l\hfil\lambda\hfil)

2577: &

2578: \hspace{1cm}

2579: &

2580: \young(\hfil m\mu\hfil\hfil\hfil\hfil\hfil,\hfil\hfil\hfil m\mu\hfil\hfil\hfil,\hfil\hfil\hfil\hfil\hfil m\mu\hfil)

2581: \end{array}

2582: $$

2583: 

2584: \noindent

2585: Note that we have $C_l \subset C_m$ but we don't have 

2586: $D_{l^*} \supset D_{m^*}$~: contrary to the case $G/P = \p V$, duality

2587: of Schubert cells is no longer decreasing.

2588: 

2589: 

2590: 

2591: %******************************************************************************

2592: 

2593: 

2594: 

2595: \subsectionplus{Case of spinor varieties}

2596: 

2597: \label{subsection_spinoriel}

2598: 

2599: Schubert cells in $\G_Q^+(2p+1,4p+2)$

2600: (resp. $\G_Q^-(2p+1,4p+2)$) are parametrised by lists of

2601: $+$ and $-$ signs, with an odd number of $+$ (resp. $-$) signs.

2602: The generic $T$-fixed point corresponding to the list

2603: $(\eta_i)$ is the subspace generated by $e_i^{\eta_i}$, and will be

2604: denoted $x_\eta$.

2605: Schubert cells are also parametrised by strict partitions of size $2p$

2606: (or subsets of $\{1,\ldots,2p\}$), the correspondance

2607: being that we set $x \in \lambda$ ($1 \leq x \leq 2p$)

2608: if $\eta_{2p+1-x} = -\ $.

2609: 

2610: \begin{defi}\

2611: \begin{itemize}

2612: \item

2613: If $(\eta_i),1\leq i \leq 2p+1,$ is a sequence of signs and $j$

2614: is an integer, we denote

2615: $\varphi(\eta,j)$ the sequence $\eta'$ of signs such that

2616: $\eta'_i = \eta_i$ for exactly all $i$'s but $j$.

2617: \item

2618: A sequence $(\eta_i),1\leq i \leq 2p+1,$ of signs is admissible if

2619: $$

2620: \forall i\in \{1,\ldots,p\},

2621: \#\{j : 1 \leq j \leq 2i, \eta_j = + \} \geq i.

2622: $$

2623: Assume that $\eta$ is admissible~:

2624: \item

2625: If there exists $i\leq p+1$ such that

2626: $\# \{j\ :\ 1\leq j\leq 2i-1,\eta_j=+\} = i-1$, then let $i_0$ be the

2627: minimal such $i$, and set $\eta^* = \varphi(\eta,2i_0-1)$.

2628: \item

2629: Otherwise there exists $i$ such that

2630: $$

2631: \forall k\geq i,\ \#\{j:j\leq k,\eta_j=+\} > \# \{j:j\leq

2632: k,\eta_j=-\}.

2633: $$

2634: Let $i_0$ be the minimal such $i$ and set $\eta^* =

2635: \varphi(\eta,i_0)$.

2636: \end{itemize}

2637: \end{defi}

2638: \noindent

2639: If there does not exist $i\leq p+1$ such that

2640: $\# \{j\ :\ 1\leq j\leq 2i-1,\eta_j=+\} = i-1$,

2641: then $\# \{ j\ :\ 1 \leq j \leq 2p+1 \} \geq p+1$, so $i=2p+1$

2642: satisfies the condition of the last point of this definition.

2643: 

2644: \lpara

2645: 

2646: Let $\eta$ be fixed.

2647: Since the positive roots are $\epsilon_i \pm \epsilon_j$ with $i<j$,

2648: the restriction of the Bruhat order on the set of $\varphi(\eta,j)$

2649: is given by~:

2650: 

2651: \begin{fait}

2652: \label{fait_ordre_spin}

2653: We have $\varphi(\eta,i) \leq \varphi(\eta,j)$ for the Bruhat order

2654: \iff

2655: $$

2656: \mbox{or }\left \{

2657: \begin{array}{l}

2658: \eta_i = \eta_j = + \mbox{ and } i\leq j\\

2659: \eta_i = + \mbox{ and } \eta_j = -\\

2660: \eta_i = \eta_j = - \mbox{ and } i \geq j.

2661: \end{array}

2662: \right .

2663: $$

2664: \end{fait}

2665: 

2666: 

2667: \begin{prop}

2668: \label{prop_spinoriel}

2669: $C_\eta$ is suitable \iff $\eta$ is admissible, and in this case

2670: $C_\eta^Q = D_{\eta^*}$.

2671: \end{prop}

2672: \begin{proo}

2673: Recall that $x_\eta \in G/P$ denotes the linear space spanned by

2674: $e_i^{\eta_i}$. It is well-known that $T_{x_\eta} G/P$ identifies

2675: with $\wedge^2 {Vect(e_i^{\epsilon_i})}^*$. Moreover,

2676: $N^*_{x_\eta} C_\eta \subset \wedge^2 {Vect(e_i^{\epsilon_i})}$ 

2677: is generated by $e_i^{\eta_i} \wedge e_j^{\eta_j}$ for $(i,j)$ such

2678: that $\eta_i = +$ and $i<j$. In fact, with the notations of 

2679: \cite[PLANCHE IV]{bourbaki}, the weight of $x_\eta$ is

2680: $\rho_\eta = \frac 12 \sum \eta_i \epsilon_i$, and the weights of

2681: $T_{x_\eta} C_\eta$ are the weights of the form

2682: $\frac 12 \sum \eta'_i \epsilon_i$ which can be expressed as

2683: $\rho_\eta + \alpha$, where $\alpha$ is a positive root. Therefore the

2684: claim follows from the fact that the positive roots are 

2685: $\epsilon_i \pm \epsilon_j$ with $i<j$.

2686: 

2687: The weights of $T^*_{x_\eta} G/P \simeq \wedge^2 {Vect(e_i^{\epsilon_i})}$

2688: are parametrised by couples $(x,y)$ of integers, with $x<y$.

2689: Now let $x_k,y_k,1 \leq k \leq p,$ be integers with $x_k < y_k$. With

2690: the notations of subsection \ref{subsection_combinatoire}, $\mu$ is given by a

2691: set of polynomials of degree $p$ and the $p$-multilinear map

2692: $P_{(x_k,y_k);\eta'}$ does not vanish \iff the $x_k$'s and the $y_k$'s

2693: are all distinct and $\eta'_i = \eta_i$ for exactly all $i$'s which

2694: belong to the set $U := \{x_k\} \cup \{ y_k \}$.

2695: 

2696: Given the previous description of 

2697: $N^*_{x_\eta} C_\eta$, the Schubert

2698: variety $C_\eta$ will be suitable \iff we can find

2699: $(x_k,y_k)$ such that

2700: \begin{equation}

2701: \label{condition_spin}

2702: x_k < y_k, \mbox{ the }x_k,y_k \mbox{ are all distinct, and }

2703: \eta_{x_k} = +.

2704: \end{equation}

2705: Therefore, for all

2706: $i$'s with $1\leq i \leq p$, we have the inequality

2707: $$

2708: 2i-1 \leq \#(U \cap [1,2i]) \leq 2\# \{j\ :\ 1\leq j\leq 2i,\eta_j = +

2709: \}.

2710: $$

2711: This implies that $\eta$ should be admissible.

2712: 

2713: Conversely, assuming that $\eta$ is admissible, let us consider the

2714: following algorithm which produces a list of distinct elements

2715: $(x_k,y_k)$ with $\eta_{x_k} = +$ and $x_k < y_k$.

2716: If $\forall i>1, \eta_i = +$, set $x_k = 2p$ and $y_k = 2p+1$.

2717: Otherwise, let $i_0$ be the minimal $i>1$ such that $\eta_i = -$;

2718: set $x_1 = i_0 - 1$ (the fact that $\eta$ is admissible garanties that

2719: even in the case $i_0=2$, we have $\eta_{i_0-1} = +$) and $y_1 =

2720: i_0$. Remove $\eta_{x_1}$ and $\eta_{y_1}$ from the list $\eta$ : this

2721: new list is again admissible, as one checks readily. Therefore, it is

2722: possible to define $(x_k,y_k)$ for $k \geq 2$ inductively.

2723: 

2724: \lpara

2725: 

2726: We therefore have proved that $C_\eta$ is suitable \iff $\eta$ is

2727: admissible. Let us now compute the dual variety. Assume first that

2728: there exists $i \in \{1,p+1\}$ such that

2729: \begin{equation}

2730: \label{condition}

2731: \# \{j : 1 \leq j \leq 2i-1, \eta_j = + \} = i-1.

2732: \end{equation}

2733: Let $i_0$ be the minimal such $i$. Admissibility of $\eta$ implies

2734: that $\# \{j\ :\ 1 \leq j \leq 2i_0-2, \eta_j = + \} = i_0 - 1$ and so

2735: $\eta_{2i_0-1} = -$. Therefore, if $(x_k,y_k)$ is any sequence

2736: satisfying (\ref{condition_spin})

2737: and $U = \{x_k\} \cup \{y_k\}$, there exists $j \leq 2i_0-1$

2738: such that $\eta_j = -$ and $j \not \in U$. Thus if there exists

2739: $(x_k,y_k)$ such that $P_{(x_k,y_k);\eta'} \not = 0$ for some $\eta'$,

2740: this implies $\eta' \geq \varphi(\eta,2i_0-1)$

2741: (recall fact \ref{fait_ordre_spin}).

2742: Conversely, the previous algorithm produces a sequence $(x_k,y_k)$ for

2743: which it is easy to see that

2744: $P_{(x_k,y_k);\varphi(\eta,2i_0-1)} \not = 0$. Thus

2745: $\eta^* = \varphi(\eta,2i_0-1)$ is the lowest

2746: list one can obtain in this way,

2747: so that $C_\eta^Q = D_{\eta^*}$ as claimed in this case.

2748: 

2749: Assume finaly that (\ref{condition}) holds for no $i \in \{1,p+1\}$.

2750: Therefore, as we have seen, there exists

2751: $i$ (for example $i=2p+1$) such that

2752: $$

2753: \forall k \geq i, \# \{j\ :\ j \leq k,\eta_j = +\}

2754: >

2755: \# \{j\ :\ j\leq k,\eta_j = -\}.

2756: $$

2757: Let $i_0$ be the minimal such $i$. Obviously, if $i$ is any integer,

2758: and $(x_k,y_k)$ satisfies (\ref{condition_spin})

2759: and $x_k,y_k \not = i$, then

2760: $$\forall k \geq i, \# \{j\ :\ j \leq k,\eta_j = +,j\not = i\}

2761: \geq

2762: \# \{j\ :\ j\leq k,\eta_j = -,j \not = i\},$$ so that

2763: $i \geq i_0$. So $P_{(x_k,y_k);\eta'} \not = 0$ implies

2764: $\eta' \geq \varphi(\eta,i_0)$. Again, the explicit algorithm provides

2765: a sequence $(x_y,y_k)$ such that

2766: $P_{(x_k,y_k);\varphi(\eta,i_0)} \not = 0$, and therefore

2767: $D_{\varphi(\eta,i_0)}$ is the dual variety of $C_\eta$.

2768: \end{proo}

2769: 

2770: 

2771: 

2772: \subsectionplus{Case of $E_{6,I}$}

2773: 

2774: \label{subsection_e6_1}

2775: 

2776: 

2777: We now consider the exceptional cases when $G$ is of type

2778: $E_6$. Recall that there are two possibilities for $(P,Q)$~: either

2779: they correspond to the roots $(\alpha_1,\alpha_6)$ or

2780: $(\alpha_3,\alpha_5)$. In each case I explain in which case

2781: $P_{\tau_1,\ldots,\tau_d;\nu}$ does not vanish. Using subsection

2782: \ref{subsection_combinatoire}, this describes in principle all dual

2783: varieties to Schubert varieties, although I will not give a simple

2784: combinatorial recipy for this correspondance (note however that to

2785: give such a description there is ``only'' a finite number of computations to

2786: do). My description of which $P_{\tau_1,\ldots,\tau_d;\nu}$ don't

2787: vanish will however yield a simple caracterisation of the suitable Schubert

2788: varieties.

2789: 

2790: As we have seen in subsection \ref{subsection_fondamental}, a Levi

2791: factor $L$ of $P_1$ is isomorphic with $\C^* \times Spin_{10}$, and

2792: $T_{[e]} E_6/P_1$ identifies with a

2793: $16$-dimensional spinor representation

2794: of $L$. Moreover, the closed $L$-orbit in $\p T^*_{[e]} G/P$

2795: identifies with a $L$-homogeneous spinor variety~: it is a connected

2796: component of the variety

2797: parametrising isotropic linear spaces of dimension 5 in a certain

2798: quadratic vector space of dimension 10 that we will denote $M$.

2799: It is proved in \cite[corollary 3.2]{flop_scorza}

2800: that $I_{[e]} \subset \p M$ is the

2801: corresponding 8-dimensional quadric acted upon by $L$ 

2802: and that the rational map 

2803: $T^*_{[e]} E_6/P_1 \dasharrow I_{[e]}$

2804: is induced by the unique $L$-equivariant

2805: quadratic map $T^*_{[e]} G/P \rightarrow M$. The polarisation 

2806: ${\cal P} : T^*_{[e]} E_6/P_1 \times T^*_{[e]} E_6/P_1 \rightarrow M$ of this

2807: equivariant map has the following geometric interpretation~: for 

2808: $x,y \in T^*_{[e]} E_6/P_1$ representing points of the spinor variety

2809: corresponding to the isotropic linear spaces $L_x,L_y$, the class

2810: of ${\cal P}(x,y)$ in $\p M$ is the intersection of $L_x$ and $L_y$ if

2811: this intersection has dimension 1, and ${\cal P}(x,y)=0$ otherwise.

2812: 

2813: Denote as in subsection \ref{subsection_spinoriel}

2814: $(e_1^+,\ldots,e_5^+,e_1^-,\ldots,e_5^-)$ a base of $M$ such that the

2815: quadratic form $Q$ satisfies

2816: $Q(\sum x_i^+ e_i^+ + \sum x_i^- e_i^-) = \sum x_i^+x_i^-$. An $L$-weight

2817: of $M$ can therefore be denoted 

2818: $\nu \in \{1^+,\ldots,5^+,1^-,\ldots,5^-\}$, and a weight 

2819: $\eta = (\eta_1,\ldots,\eta_5)$ in

2820: $T^*_{[e]} G/P$ corresponds to a list of plus or minus signs, with an

2821: odd number of plus signs.

2822: The condition for $P_{\eta,\eta';\nu}$ not to vanish is thus that

2823: $\eta$ and $\eta'$ have exactly one sign in common which is $\nu$.

2824: 

2825: From this description we can describe the suitable Schubert varieties.

2826: In the array below, I recall the quiver of $E_6/P_1$ and 

2827: I define an

2828: element $[w_{max}] \in W/W_P$ by its subquiver~:

2829: 

2830: $$

2831: \begin{array}{ccc}

2832: \input{e6} && \input{w_max_1}\\

2833: \mbox{Quiver of }E_6/P_1  &  \hspace{2cm}  &  \mbox{Subquiver of }[w_{max}]

2834: \end{array}

2835: $$

2836: 

2837: \noindent

2838: We have the following proposition~:

2839: 

2840: \begin{prop}

2841: \label{prop_e6_1_adapte}

2842: Let $[w] \in W/W_P$. Then the Schubert variety

2843: $C_{[w]}$ is suitable \iff $[w] \leq [w_{max}]$.

2844: \end{prop}

2845: \begin{proo}

2846: Below we give the Hasse diagram $H$ (resp. $H^*$) of the $L$-module

2847: $T_{[e]} E_6/P_1$ (resp.  $T^*_{[e]} E_6/P_1$), which, by proposition

2848: \ref{hasse}, is isomorphic with the quiver of $E_6/P_1$~:

2849: 

2850: $$

2851: \begin{array}{cc}

2852: \input{h} & \input {hd}\\

2853: \\

2854: H & H^*

2855: \end{array}

2856: $$

2857: 

2858: Let $\iota : H \rightarrow H^*$ be

2859: induced by the map $\eta \mapsto -\eta$ (in terms of quivers, this

2860: corresponds to the obvious symmetry).

2861: Let $[w] \in G/P$ and $Q_{[w]} \subset H$ the corresponding subquiver, marking

2862: the weights of $w^{-1}.T_{[w]} C_{[w]}$.

2863: Thanks to proposition \ref{prop_combinatoire},

2864: the proposition amounts to the fact that

2865: we can find two weights $\eta,\eta' \in H^* - \iota(Q_{[w]})$

2866: and which have only one sign in common \iff 

2867: $Q_{[w]} \subset Q_{[w_{max}]}$. This may be seen as follows~:

2868: 

2869: \begin{itemize}

2870: \item

2871: If $Q_{[w]} \subset Q_{[w_{max}]}$, we can set

2872: $\eta = (++--+)$ and $\eta' = (+-++-)$ to check that the corresponding

2873: Schubert variety is suitable (below the subset $\iota(Q_{[w_{max}]})$ 

2874: is drawn).

2875: \item

2876: If $Q_{[w]}$ contains the vertex corresponding to the

2877: weight $(-+--+)\ $, $\iota(Q_{[w]})$ contains the subset drawn below.

2878: Thus $\eta$ and $\eta'$ are weights which begin with $++$, so

2879: they have two common signs. The corresponding Schubert variety

2880: is not suitable.

2881: \item

2882: The last case is that the subquiver contains the vertex corresponding

2883: to the weight $(--++-)$. 

2884: Thus $\eta$ and $\eta'$ have at least 3 plus signs among the 

2885: 4 first signs, and

2886: therefore have at least 2 common signs.

2887: \end{itemize}

2888: 

2889: $$

2890: \begin{array}{ccccc}

2891: \input{cas0} && \input{cas1} && \input{cas2} \\

2892: \iota(Q_{[w_{max}]}) & \hspace{.5cm} & 

2893: (-+--+) \in Q_{[w]} & \hspace{.5cm} & (--++-) \in Q_{[w]}

2894: \end{array}

2895: $$

2896: 

2897: \end{proo}

2898: 

2899: \begin{exem}

2900: \label{exem_p5e6}

2901: Let $X \subset E_6/P_1$ be a linear subspace of maximal dimension 5.

2902: Then $X^Q \subset E_6/P_6$ is also a linear subspace of dimension 5.

2903: \end{exem}

2904: \noindent

2905: Therefore, this provides, in our setting, a new example of a variety

2906: which is isomorphic to its dual. Similar examples in the usual setting

2907: $X \subset \p^n$

2908: are projective subspaces, quadrics, 

2909: $\G(2,2p+1)$, and the spinor variety

2910: $\G_Q^+(5,10)$.

2911: \begin{proo}

2912: Let $X \subset E_6/P_1$ be a linear subspace of dimension 5.

2913: The variety parametrising linear subspaces of maximal dimension 5 is given by

2914: Tits shadows, according to \cite[theorem 4.3]{landsberg}.

2915: In particular, it is a

2916: homogeneous variety, and so we can assume that $X$ is the Schubert

2917: variety corresponding to the Weyl group element

2918: $w=s_6s_5s_4s_3s_1$. The corresponding quiver $Q_w$ and $\iota(Q_w)$

2919: follow; we have also drawn the quiver $Q_{w^*}$ of the dual variety.

2920: $$

2921: \begin{array}{ccccc}

2922: \input{p5}  &&  \input{ip5}  &&  \input{dp5}\\

2923: Q_w & \hspace{1cm} & \iota(Q_w) & \hspace{1cm} & Q_{w^*}

2924: \end{array}

2925: $$

2926: 

2927: According to proposition \ref{prop_combinatoire}, we must look for two

2928: weights of $H^*$ not in $\iota(Q_w)$ which have only one common

2929: sign. If this common sign is a minus sign, then among these two

2930: weights there are 6 minus signs. Therefore one of the weight has 4

2931: minus signs which is impossible given $\iota(Q_w)$ and the Hasse

2932: diagram $H^*$. Since $(--+++)$ and $(++--+)$ are weights not in

2933: $\iota(Q_w)$, the lowest weight is $5^+$. Note that this weight is

2934: obtained applying $w_6 = s_3s_4s_5s_6$ to the highest weight.

2935: Therefore, proposition

2936: \ref{prop_combinatoire} shows that the dual variety to $X$ corresponds

2937: to the class of $w.w_6 = s_6s_5s_4s_3s_1s_3s_4s_5s_6$ modulo $W_6$.

2938: Modulo $W_6$, we have

2939: $$

2940: \begin{array}{cccccc}

2941: & s_6s_5s_4s_3s_1s_3s_4s_5s_6 & = & s_6s_5s_4s_1s_3s_1s_4s_5s_6 & = &

2942: s_6s_5s_4s_1s_3s_4s_5s_6s_1\\

2943: \equiv & s_6s_5s_1s_4s_3s_4s_5s_6 & = & s_6s_5s_1s_3s_4s_3s_5s_6 &

2944: \equiv & s_6s_5s_1s_3s_4s_5s_6\\

2945: = & s_6s_1s_3s_5s_4s_5s_6 & \equiv & s_6s_1s_3s_4s_5s_6 &

2946: \equiv & s_1s_3s_4s_5s_6.

2947: \end{array}

2948: $$

2949: 

2950: This proves the claim.

2951: \end{proo}

2952: 

2953: \subsectionplus{Case of $E_{6,II}$}

2954: 

2955: \label{subsection_e6_2}

2956: 

2957: Let $\alpha \in E_6/P_3$ be the base point.

2958: Recall that there is a surjection

2959: $\pi : T^*_\alpha E_6/P_3 \rightarrow 

2960: L_\alpha \otimes \wedge^2 Q_\alpha^*$ and that the

2961: rational map $q:T^*_\alpha E_6/P_3 \dasharrow I_\alpha$ is induced by a

2962: rational map $\overline q : L_\alpha \otimes \wedge^2 Q^*_\alpha 

2963: \dasharrow I_\alpha = \G(2,Q_\alpha^*)$.

2964: 

2965: The description of $\overline q$ given in subsection

2966: \ref{subsection_fondamental} implies that

2967: $\overline q$ has degree 6. Consider as in subsection

2968: \ref{subsection_combinatoire} its polarisation, with

2969: coordinates denoted $P$. In order to give a non-vanishing

2970: criterium for $P$, let us introduce some notation. 

2971: Let $e_1,e_2$ be a basis of $L_\alpha$ and 

2972: $f_1^*,\ldots,f_5^*$ a basis of $Q_\alpha^*$.

2973: The weight of the

2974: vector $e_1 \otimes (f_i^* \wedge f_j^*)$ with $i<j$ will be denoted $ij$,

2975: and the weight of the vector $e_2 \otimes (f_i^* \wedge f_j^*)$ will be

2976: denoted $\underline{ij}$.

2977: Finally, the weight of 

2978: $f_i^* \wedge f_j^*$

2979: will be denoted $ij^*$.

2980: 

2981: Let $\tau_1,\ldots,\tau_6$ be weights of

2982: $L_\alpha \otimes \wedge^2 Q_\alpha^*$ and $\nu$ a weight of 

2983: $\wedge^2 Q_\alpha^*$,

2984: if $P_{\tau_1,\ldots,\tau_6,\nu} \not = 0$, then

2985: $\# \{k:\tau_k \in \{ij\}\ \} = 

2986: \# \{k:\tau_k \in \{\underline{ij}\}\ \} = 3$.

2987: So we assume that this is the case and that

2988: $\tau_1,\tau_2,\tau_3$ (resp. $\tau_4,\tau_5,\tau_6$) are of the form

2989: $ij$ (resp. $\underline{ij}$).

2990: 

2991: With this setting,

2992: $P_{i_1j_1,i_2j_2,i_3j_3,

2993: \underline{k_1l_1},\underline{k_2l_2},\underline{k_3l_3}

2994: ;mn^*}$

2995: will not vanish if and only if, up to permuting the three first weights and the

2996: three last, we have that $i_1,j_1,i_2,j_2$ (resp. $k_1,l_1,k_2,l_2$) 

2997: are all dinstinct; say they take all values in $\{1,\ldots,5\}$ except

2998: $u$ (resp. $v$). Moreover we must have $u \in \{k_3,l_3\}$ (resp.

2999: $v \in \{i_3,j_3\}$), say $\{k_3,l_3\} = \{u,u'\}$ (resp.

3000: $\{i_3,j_3\} = \{v,v'\}$). Finally, we must have $u' \not = v'$ and

3001: $\{u',v'\} = \{m,n\}$.

3002: 

3003: In principle, this combinatorial rule describes dual varieties of

3004: Schubert cells in this case. However, as in the case of $E_{6,I}$, we

3005: can be more precise as far as suitability is concerned. The Hasse

3006: diagram $H$ of

3007: $L_\alpha^* \otimes \wedge^2 Q_\alpha$ is given below, as well as a subquiver

3008: denoted $Q_{max}$. I have also indicated the Hasse diagram $H^*$ of

3009: $L_\alpha \otimes \wedge^2 Q_\alpha^*$~:

3010: 

3011: $$

3012: \begin{array}{ccccc}

3013: \input{hasse_e62} && \input{qmax} && \input{e6_2}\\

3014: H && Q_{max} && H^*

3015: \end{array}

3016: $$

3017: 

3018: In these pictures and the following, the weights $ij$ are drawn in red

3019: and the weights $\underline{ij}$ are drawn in blue.

3020: Given a Schubert cell $C_{[w]}$, with $w$ the minimal length

3021: representative of $[w] \in W/W_P$, recall that we associated

3022: (not injectively) to this Schubert cell

3023: a subquiver $Q_{[w]}$ of $H$ in subsection \ref{subsection_carquois}.

3024: 

3025: \begin{prop}

3026: \label{prop_e6_2_adapte}

3027: The Schubert cell $C_{[w]}$ is suitable \iff we

3028: have $Q_{[w]} \subset Q_{max}$.

3029: \end{prop}

3030: \begin{proo}

3031: As in the preceeding subsection, we define

3032: $\iota:H \rightarrow H^*$ given by

3033: $\eta \mapsto -\eta$.

3034: The weights of $H^*$ have been given above.

3035: 

3036: Since $q$ is induced by $\overline q$,

3037: a Schubert variety $C_{[w]}$ will be suitable \iff

3038: the rational map 

3039: $\overline q:L_\alpha \otimes \wedge^2 Q_\alpha^* \dasharrow I_\alpha$

3040: is defined generically on $\pi(w^{-1}.N^*_{[w]} C_{[w]})$; equivalently,

3041: $\overline q$ should be defined on the orthogonal of

3042: $w^{-1}.T_{[w]} C_{[w]} \cap L_\alpha^* \otimes \wedge^2 Q_\alpha$ in

3043: $L_\alpha \otimes \wedge^2 Q_\alpha^*$.

3044: Equivalently again, we should be able to

3045: find 6 weights $\tau_k$ not in $\iota(Q_{[w]})$ and some integers $i,j$

3046: such that $P_{\tau_k;ij^*}$ does not vanish.

3047: 

3048: In case $Q_{[w]}$ is

3049: included in $Q_{max}$, we can consider the weights

3050: $34, 25, 34$, $\underline{15}$, $\underline{24}$, $\underline{15}$

3051: (the corresponding subset $\iota(Q_{max})$ is drawn below), which satisfy

3052: the relation

3053: $P_{34,25,34,\underline{15},\underline{24},\underline{15};45^*} \not

3054: = 0$ and do not belong to $\iota(Q)$. Otherwise, there are four cases~:

3055: \begin{itemize}

3056: \item

3057: If $Q_{[w]}$ contains the weight $15$, by proposition

3058: \ref{prop_order_ideal}, $\iota(Q_{[w]})$ contains the

3059: corresponding subset in the array below. The remaining weights are of

3060: the form $ij$ or $\underline{ij}$ with $1<i<j$, so the Schubert variety cannot

3061: be suitable.

3062: \item

3063: If $Q_{[w]}$ contains the weight $\underline{14}$, 

3064: the remaining weights are of the form

3065: $ij$ or $\underline{kl}$ with $2<i<j$ (see the array below), so again

3066: the Schubert variety cannot be suitable.

3067: \item

3068: If $Q_{[w]}$ contains the weight $24$, let 

3069: $i_1j_1,i_2j_2,i_3j_3,\underline{i_4j_4},\underline{i_5j_5},

3070: \underline{i_6j_6}$\vspace{1mm} be a list of weights not in $\iota(Q_{[w]})$.

3071: Note that we have

3072: $i_kj_k \in \{12,13,14,15,34\}$ for all $k$. Assume that there exists

3073: $kl^*$ such that 

3074: $P_{i_1j_1,i_2j_2,i_3j_3,\underline{i_4j_4},\underline{i_5j_5},

3075: \underline{i_6j_6};kl^*} \not = 0$. This implies that the

3076: only integer $x$ (resp. $y$)

3077: which does not belong to the set $\{i_1,j_1,i_2,j_2\}$

3078: (resp. $\{i_4,j_4,i_5,j_5\}$) is

3079: either 2 or 5. This integer must

3080: therefore belong to the set $\{i_3,j_3\}$ (resp.

3081: $\{i_3,j_3\}$), so we must have $\{i_3,j_3\} = \{1,x\}$

3082: (resp. $\{i_6,j_6\} = \{1,y\}$). This implies that $k=l=1$, a

3083: contradiction.

3084: \item

3085: Assume finaly that $Q_{[w]}$ contains the weight $\underline{23}$.

3086: In this case all the

3087: weights which are not in $\iota(Q_{[w]})$

3088: and of the form $ij$ satisfy $j=5$. Again,

3089: the Schubert variety is not suitable.

3090: \end{itemize}

3091: 

3092: $$

3093: \begin{array}{ccccc}

3094: \input{Cas0} & \hspace{.9cm} & \input{Cas1} & \hspace{.9cm} & \input{Cas2}\\

3095: \mbox{Case } \iota(Q_{max}) && \mbox{Case } {15} \in Q_{[w]} &&

3096: \mbox{Case } \underline{14} \in Q_{[w]}

3097: \end{array}

3098: $$

3099: $$

3100: \begin{array}{ccc}

3101: \input{Cas3} && \input{Cas4}\\

3102: \mbox{Case } {24} \in Q_{[w]} & \hspace{1cm} & 

3103: \mbox{Case } \underline{23} \in Q_{[w]}

3104: \end{array}

3105: $$

3106: 

3107: 

3108: \end{proo}

3109: 

3110: 

3111: 

3112: 

3113: %******************************************************************************

3114: 

3115: %******************************************************************************

3116: 

3117: %******************************************************************************

3118: 

3119: 

3120: 

3121: \begin{thebibliography}{99}

3122:   \bibitem[Ar 99]{arrondo_severi} Arrondo, E.,

3123: {\it Projections of Grassmannians of lines and 

3124: characterization of Veronese varieties.}

3125: J. Algebraic Geom. {\bf 8} (1999), no. 1, 85--101. 

3126: 

3127:   \bibitem[AC 05]{arrondo_low} Arrondo, E., Caravantes, J.

3128: {\it On the Picard group of low-codimension subvarieties},

3129: math.AG/0511267.

3130: 

3131:   \bibitem[Bou 68]{bourbaki} Bourbaki, N., Groupes et alg\`ebres de Lie. 

3132: Chapitre 4,5,6.

3133: Actualit\'es Scientifiques et Industrielles, No. {\bf 1337} Hermann, Paris 1968

3134: 

3135:   \bibitem[Ch 06]{flop_scorza} Chaput, P.E.,

3136: {\it On Mukai flops for Scorza varieties}, math.AG/0601734

3137: 

3138:   \bibitem[Ch 07]{hermitian} Chaput, P.E.,

3139: {\it Geometry over composition algebras~: hermitian geometry}, in

3140: preparation.

3141: 

3142:   \bibitem[CMP 06]{quantique} Chaput, P.E., Manivel, L., Perrin, N.

3143: {\it Quantum cohomology of minuscule homogeneous spaces},

3144: math.AG/0607492, to appear in Transformation Groups.

3145: 

3146:   \bibitem[De 96a]{debarre1} Debarre, O.,

3147: {\it Théorèmes de connexité pour les produits d'espaces

3148: projectifs et les grassmanniennes},

3149: Amer. J. Math.  {\bf 118} (1996),  no. 6, 1347--1367

3150: 

3151:   \bibitem[De 96b]{debarre2} Debarre, O.,

3152: {\it Sur un théorème de connexité de Mumford pour les espaces homogènes},

3153: Manuscripta Math. {\bf 89} (1996), no. 4, 407--425. 

3154: 

3155:   \bibitem[Fa 81]{faltings} Faltings, G.

3156: {\it Formale Geometrie und homogene Räume},

3157: Invent. Math. {\bf 64} (1981), no. 1, 123--165.

3158: 

3159:   \bibitem[GKZ 94]{gkz} Gelfand, I. M.; Kapranov, M. M.; Zelevinsky, A. V. 

3160: Discriminants, resultants, and multidimensional determinants.

3161: Mathematics: Theory and Applications. Birkhäuser Boston, Inc., Boston, MA, 1994

3162: 

3163:   \bibitem[He 78]{hesselink} Hesselink, W. H.

3164: {\it Polarizations in the classical groups.}

3165: Math. Z. {\bf 160} (1978), no. 3, 217--234.

3166: 

3167:   \bibitem[LM 03]{landsberg} Landsberg, J., Manivel, L.,

3168: {\it On the projective geometry of rational homogeneous varieties},

3169: Comment. Math. Helv. {\bf 78} (2003), no. 1, 65--100.

3170: 

3171:   \bibitem[McG 02]{mcgovern} McGovern, W. M.

3172: {\it The adjoint representation and the adjoint action.} 

3173: Algebraic quotients. Torus actions and cohomology. 

3174: The adjoint representation and the adjoint action,  159--238, 

3175: Encyclopaedia Math. Sci., {\bf 131}, Springer, Berlin, 2002.

3176: 

3177:   \bibitem[Na 06]{namikawa} Namikawa, Y.,

3178: {\it Birational geometry of symplectic resolutions of nilpotent orbits},

3179: to appear in Advanced studies in pure mathematics {\bf 45},

3180: Moduli Spaces and Arithmetic Geometry (Kyoto 2004), p.75-116.

3181: 

3182:   \bibitem[Pe 05]{perrin_courbe}  Perrin N., {\it Rational curves on 

3183: minuscule Schubert varieties},

3184: J. Algebra  {\bf 294}  (2005),  no. 2, 431--462.

3185: 

3186:   \bibitem[Pe 06]{carquois} Perrin N., {\it Small resolutions 

3187: of minuscule Schubert varieties}, preprint

3188: arXiv:math.AG/0601117. 

3189: 

3190:   \bibitem[Pe 07]{perrin_low} Perrin N., {\it Small codimension smooth

3191: subvarieties in even-dimensional homogeneous spaces with Picard

3192: group $\Z$}, arXiv:math.AG/0702578.

3193: 

3194:   \bibitem[Ri 74]{richardson}  Richardson, R. W.,

3195: {\it Conjugacy classes in parabolic subgroups of semisimple algebraic groups.}

3196: Bull. London Math. Soc.  {\bf 6}  (1974), 21--24.

3197: 

3198:   \bibitem[Za 93]{zak} Zak, F.L.,

3199: Tangents and Secants of Algebraic Varieties.

3200:                         American Mathematical Society, Providence, RI

3201:                         (1993).

3202: 

3203: \end{thebibliography}

3204: 

3205: \vskip .2cm

3206: 

3207: \noindent

3208: Author's address :

3209: 

3210: \noindent

3211: Laboratoire de Math\'ematiques Jean Leray UMR 6629

3212: 

3213: \noindent

3214: 2 rue de la Houssini\`ere - BP 92208 - 44322 Nantes Cedex 3

3215: 

3216: \vskip .1cm

3217: \noindent

3218: Pierre-Emmanuel.Chaput@math.univ-nantes.fr

3219: 

3220: 

3221: 

3222: \end{document}

3223: 

3224: 

3225: 

3226: \lpara

3227: 

3228: Let us now see that fundamental cases are natural

3229: and in fact linked with usual duality (of $G$-representations)~:

3230: 

3231: \begin{prop}

3232: Let $P,Q \subset G$ be non-conjugated maximal parabolic subgroups. 

3233: Then there is a unique

3234: invariant hypersurface in $G/P \times G/Q$ \iff $P,Q,G$ are one of the

3235: fundamental examples.

3236: 

3237: In this case, if $L$ (resp. $M$) is the

3238: ample line bundle gnerating $Pic(G/P)$ (resp. $Pic(G/Q)$), then the

3239: $G$-representations $\Gamma(G/P,L)$ and $\Gamma(G/Q,M)$ are dual

3240: $G$-representations.

3241: \end{prop}

3242: \pr

3243: It is easy to check that if $P,Q\subset G$ is a fundamental example,

3244: then $\Gamma(G/P,L)$ and $\Gamma(G/Q,M)$ are dual

3245: $G$-representations. One also checks that the variety of orthogonal

3246: pairs in $G/P \times G/Q$ is the unique $G$-invariant hypersurface. In

3247: fact, the orbit of any non-orthogonal pair is dense.

3248: \lpara

3249: 

3250: Conversely, assume $H \subset G/P \times G/Q$ is a $G$-invariant

3251: hypersurface. Then we have a $G$-equivariant map which sends a point

3252: $x \in G/P$ on the divisor $\{y \in G/Q,(x,y) \in H\}$. The class of

3253: this divisor in $Pic(G/Q)$ does not depend on $y$; denote it $D$. We therefore

3254: get a map from $G/P$ to $\p \Gamma(G/Q,{\cal O}(D))$. Let $d$ denote

3255: the integer such that $\co(D)$ is $d$ times the ample generator of $Pic(G/Q)$.

3256: 

3257: Now, let $\omega_P$ (resp. $\omega_Q$) be the fundamental weight 

3258: corresponding to $P$ (resp. $Q$). Call $V_P$ (resp. $V_Q$)

3259: the representation with

3260: highest weight $\omega_P$ (resp. $\omega_Q$)

3261: and $W$ the representation with highest

3262: weight $d.\omega_Q$. Recall the map 

3263: $\varphi : G/P \rightarrow \p \Gamma(G/Q,{\cal O}(D)) = \p W^*$. This map is

3264: given by a $G$-invariant subspace

3265: $S\subset \Gamma(X,\varphi^* {\cal O}_{\p W^*}(1))$. There is an

3266: integer $e$ such that 

3267: $\varphi^* {\cal O}_{\p W^*}(1) \simeq \co_X(e)$. Thus

3268: $\Gamma(X,\varphi^* {\cal O}_{\p W^*}(1)) = S^{(e)}V^*$

3269: is an irreducible $G$-module and has $\Gamma(Y,\co_{\p W^*}(1))$ and

3270: $W=\Gamma(\p W^*,\co_{\p W^*}(1))$

3271: as subrepresentations, if $Y$ denotes $\varphi(X)$. Therefore,

3272: $S = \Gamma(Y,\co_{\p W^*}(1)) = W$. 

3273: So $S^{(d)} V_Q = W = S^{(e)} V_P^*$. It follows that

3274: $d=e$ and $V_P = V_Q^*$. From this, using the classification of the

3275: fundamental representations of simple groups, the proposition follows

3276: easily.

3277: \qed

3278: 

3279: Preuve du théo sur les variétés duales quand on a une fibration :

3280: 

3281: Let $z$ be any point in $\psi_Q^{-1}[\psi_P(x)]$.

3282: Let $\mathfrak{h}_x$ and $\mathfrak{q}_z \subset \g$ be the Lie

3283: subalgebras of $H_x$ and $Q_z$. Choose Levi subalgebras $\mathfrak{l,m}$

3284: of $\mathfrak{h}_x$ and $\mathfrak{q}_z$ such that 

3285: $\mathfrak m \subset \mathfrak l$. Then 

3286: $\u(\mathfrak q_z)=\u(\mathfrak h_x) 

3287: \oplus \u(\mathfrak q_z \cap \mathfrak l)$.

3288: Therefore, a linear form $f \in \g^*$ vanishes on $\mathfrak q_z$ \iff

3289: it vanishes on $\mathfrak m$, $\u(\mathfrak{h}_x)$ and

3290: $\u(\mathfrak{q}_z \cap \mathfrak{l})$.

3291: 

3292: Let $l \in \g^*$ be the image of a cotangent vector in $N^*_xX$. This

3293: implies that $l$ vanishes on $\mathfrak p_x$, and therefore on 

3294: $\mathfrak m$ and $\u(\mathfrak{h}_x)$.

3295: Let $Q_z$ be the parabolic group

3296: contained in $H_x$ and which image in $L_x$ is the parabolic subgroup

3297: stabilizing the image of $l_{|T_xF_x}$ in $L_x/Q_y$. 

3298: Let $\mathfrak q_z$ denote its

3299: Lie subalgebra. Since the image of $Q_z$ in $L_x$ corresponds to $l_{|T_xF_x}$,

3300: $l$ vanishes on $\u(\mathfrak{q}_z \cap \mathfrak{l})$.

3301: Therefore, it vanishes on $\mathfrak q_z$.

3302: 

3303: 

3304: 

3305: 

3306: 

3307: 

3308: 

3309: 

3310: 

3311: 

3312: preuve Nicolas :

3313: 

3314:  We will show

3315: that this is the case in general.

3316: 

3317: Let $\Lambda$ denote the set of $H$-weights of $T_xX$ and let $\omega$

3318: denote the weight of $V$.

3319: We have an injection (?)

3320: $\varphi : [1,N] \rightarrow \Lambda,i\mapsto \omega - \alpha_i$, 

3321: where the $G$-weight 

3322: $\omega - \alpha_i$ is considered by restriction as an $H$-weight.

3323: Let $ \beta$ denote the only simple root such that

3324: $\langle \beta^v,\omega \rangle = 1$. Because $\alpha_i$ is a root

3325: and, by \cite[lemma 2.13]{perrin_courbe}, $\alpha_i \geq \beta$,

3326: $\omega - \alpha_i$ is a weight of $T_xX$ (?)

3327: Since $\dim T_xX = \dim X = N$ and all $H$-weights in $T_xX$

3328: have multiplicity 1, $\varphi$ is thus a bijection.

3329: 

3330: According to lemmas \ref{fleche1} and \ref{fleche2}, $\varphi$

3331: induces an isomorphism of quivers between the quiver $Q_X$

3332: and the Hasse diagram of $T_xX$.

3333: \end{proo}

3334: 

3335: \begin{lemm}

3336: \label{fleche1}

3337: If there is an arrow from $i$ to $j$ in the quiver $Q_X$, then

3338: $\alpha_j - \alpha_i$ is a simple root.

3339: \end{lemm}

3340: \begin{proo}

3341: Let $i,j$ be such that there is an arrow $i \rightarrow j$. Assume first

3342: that $j$ has a predecessor $p(j)$. We consider the quiver $Q'_X$ defined

3343: as in \cite{carquois}. Then

3344: according to proposition 4.9 in \cite{carquois}, the interval

3345: $[p(j),j]$ for the order induced by $Q_X$ is of the form

3346: 

3347: 

3348: \begin{pspicture*}(-.5,-.5)(4.50,10.50)

3349: 

3350: \psline[linecolor=black](0.12,9.88)(0.88,9.12)

3351: \psline[linecolor=black](1.12,8.88)(1.88,8.12)

3352: \psline[linecolor=black](2.12,7.88)(2.88,7.12)

3353: \psline[linecolor=black](3.0,6.88)(3.0,6.12)

3354: \psline[linecolor=black](3.12,6.88)(3.88,6.12)

3355: \psline[linecolor=black](3.0,5.88)(3.0,5.12)

3356: \psline[linecolor=black](3.88,5.88)(3.12,5.12)

3357: \psline[linecolor=black](2.88,4.88)(2.12,4.12)

3358: \psline[linecolor=black](1.88,3.88)(1.12,3.12)

3359: \psline[linecolor=black](0.88,2.88)(0.12,2.12)

3360: \pscircle*(0.0,10.0){.17}

3361: \pscircle*(0.0,2.0){.17}

3362: \pscircle*(1.0,9.0){.17}

3363: \pscircle*(1.0,3.0){.17}

3364: \pscircle*(2.0,8.0){.17}

3365: \pscircle*(2.0,4.0){.17}

3366: \pscircle*(3.0,7.0){.17}

3367: \pscircle*(3.0,6.0){.17}

3368: \pscircle*(3.0,5.0){.17}

3369: \pscircle*(4.0,6.0){.17}

3370: \end{pspicture*}

3371: 

3372: (with an arbitrary even number of vertices denoted $L$).

3373: 

3374: We may moreover

3375: assume that $p(j)=i+L-1$. In fact, this is stated without proof as

3376: remark 4.10 in \cite{carquois}, and we sketch an argument for

3377: the convenience of the reader.

3378: 

3379: Consider first the case $L=4$. We have four vertices, three of which are

3380: $p(j),i,j$. Let $k$ denote the other vertex. For $p(j) \leq t \leq j$

3381: and $t \not \in \{i,k\}$,

3382: the reflexions $s_{\beta_t}$ commute with $\beta_j$ because there are exactly

3383: two arrows coming from $p(j)$ in $Q'_X$ and these arrow point to $i,k$.

3384: Therefore, we may already assume that $j=i+1$ and $k=p(j)+1$.

3385: 

3386: Let ${\cal D}$ denote the dynkin diagram of $G$; $\beta_i$ and

3387: $\beta_k$ are in two distinct connected components of ${\cal D}-\{\beta_j\}$.

3388: Since two reflexions corresponding to simple roots in different connected

3389: components of ${\cal D}-\{\beta_j\}$ commute, we may assume that there is an

3390: integer $u$ such that $\forall l \in ]k,u]$, $\beta_l$ is in the same

3391: connected component as $\beta_i$, and $\forall l \in ]u,i[$,

3392: $\beta_l$ is not. It follows that $\beta_l,l \in ]k,u]$ commute with

3393: $\beta_k$, and $\beta_l, l \in ]u,i[$ commute with $\beta_i$. So we are done

3394: in this case.

3395: 

3396: In the general case, since again there are at most two arrows from a vertex

3397: in $Q'_X$, all the reflexions $s_{\beta_t}$ with $t \in ]p(j),j[$

3398: commute with the $L/2-1$ first reflexions in the interval $]p(j),j[$ for

3399: the quiver order. Therefore, the only difficulty is to commute

3400: with the middle reflexions, and the argument is analogous to the first case.

3401: Therefore, we deduce that 

3402: $\alpha_j - \alpha_i = \alpha_{p(j)+1} - \alpha_{p(j)}$.

3403: 

3404: \smallskip

3405: 

3406: Assume now that $j$ has no predecessor. Then $i$ has also no predecessor.

3407: In fact, if $k$ is the predecessor of $i$, then by definition of $Q'_X$

3408: there is an arrow from $k$ to $j$, and from $i$ to $j$ in $Q'_X$. Therefore,

3409: $j$ is a hole in $Q'_X$ (according to definition 4.7 \cite{carquois}).

3410: This is a contradiction, because $Q'_X$ has no holes (this follows from

3411: \cite[proposition 4.11]{carquois}).

3412: Moreover, we can again assume that $j=i+1$.

3413: Therefore, direct computation shows that 

3414: $\alpha_j-\alpha_i = \beta_j$, so it is a simple

3415: root.

3416: It follows by induction that $\alpha_j - \alpha_i$ is allways a

3417: simple root, so that the lemma is proved.

3418: \end{proo}

3419: 

3420: \begin{lemm}

3421: \label{fleche2}

3422: If $\alpha_j-\alpha_i$ is a simple root, then there is an arrow from

3423: $i$ to $j$ in the quiver $Q_X$.

3424: \end{lemm}

3425: \begin{proo}

3426: Assume that $\gamma_i := \alpha_j - \alpha_i$ is a simple root.

3427: If $j$ has no predecessor, then

3428: there is a unique path from the first vertex in the quiver to $j$, so

3429: that all the roots $\alpha_t$ are either smaller or bigger than $\alpha_j$,

3430: according to $t < j$ or $t > j$ for the order in $Q'_X$.

3431: Therefore, there is a path from $i$ to

3432: $j$, actually an arrow.

3433: 

3434: We now assume that $j$ has a predecessor. Since $Q'_X$

3435: is connected, there is an arrow from a vertex, say $k$, to $j$. If $k=i$,

3436: we are done, so we assume $k \not = i$.

3437: By lemma

3438: \ref{fleche1}, $\gamma_k := \alpha_j - \alpha_k$ is a simple root. Moreover,

3439: the equality $\gamma_i = \gamma_k$ would imply $\alpha_i = \alpha_k$, which

3440: is not the case. Set $\alpha = \alpha_j - \beta_i - \beta_k$. Since

3441: $s_{\beta_i}$ and $s_{\beta_k}$ commute (?), $\alpha$ is a root. Therefore,

3442: there exists an integer $l$ such that $\alpha = \alpha_l$. Now, by induction,

3443: we can assume that there is an arrow from $l$ to $i$ and from $l$ to $k$ in

3444: $Q'_X$. Since $j$ has a predecessor,

3445: \cite[proposition 4.9]{carquois} implies that there is an

3446: arrow from $i$ to $j$. 

3447: \end{proo}

3448: