math0701750/sl.tex
1: \documentclass[twoside]{article}[11pt]
2: \markboth{\textsc{Some quasihomogeneous Legendrian varieties}}%
3: {\textsc{Jaros\l{}aw Buczy\'n{}ski}}
4: 
5: 
6: 
7: 
8: \pagestyle{myheadings}
9: \usepackage{a4}
10: \usepackage{verbatim}
11: \usepackage{amsmath}
12: \usepackage{amssymb}
13: \usepackage{amsthm}
14: \usepackage{definitions}
15: %\usepackage[all]{xy}
16: \usepackage[dvips]{graphicx}
17: 
18: 
19: \def\Inv{\mathcal{INV}}
20: 
21: \def\cDeg{\mathcal{DEG}}
22: \def\tr{\operatorname{tr}}
23: \def\Xdeg{X_{\operatorname{deg}}}
24: \def\Xinv{X_{\operatorname{inv}}}
25: 
26: 
27: 
28: 
29: \author{Jaros\l{}aw Buczy\'nski}
30: \date{6 July 2007}
31: 
32: 
33: \title{Some quasihomogeneous Legendrian varieties\footnote{
34: The article is a part of the research project 
35: N20103331/2715 funded by Polish financial 
36: means for science in years 2006-2008.
37: The article will be included in the author's PhD thesis.
38: Author's e-mail: jabu@mimuw.edu.pl
39: }
40: }
41: \hyphenation{
42: Le-gen-drian
43: %perp-en-di-cu-lar
44: }
45: 
46: \setcounter{tocdepth}{2}
47: 
48: 
49: \renewcommand{\labelenumi}{\theenumi.}
50: 
51: 
52: 
53: \begin{document}
54: \maketitle
55: 
56: %\section*{Things to correct}
57: 
58: %\textbf{
59: %}
60: 
61: 
62: \begin{abstract}
63: We construct a family of examples of Legendrian subvarieties in
64:  some projective spaces.
65: Although most of them are singular,
66: a new example of smooth Legendrian variety in dimension 8 is in this family.
67: The 8-fold has interesting properties:
68: it is a compactification of the special linear group,
69: a Fano manifold of index 5 and Picard number 1.
70: %A careful analysis of its properties leads us to conclusions     
71: %announced here and described precisely in some separate notes.
72: 
73: \end{abstract}
74: 
75: 
76: \tableofcontents
77: 
78: \section*{Acknowledgements}
79: 
80: The author is especially grateful to Sung Ho Wang for initiating the research. 
81: Also very special thanks to Insong Choe for his support and invitation
82: to KIAS (Korea Institute for Advanced Study)
83: where the author could (among many other attractions) meet Sung Ho Wang. 
84: The author acknowledges the help of Jaros\l{}aw Wi\'s{}niewski, Grzegorz Kapustka, 
85: Micha\l{} Kapustka, Michel Brion, Micha\l{} Krych, Joseph Landsberg, Laurent Manivel,
86: and Andrzej Weber.
87: Also thanks to an anonymous referee for pointing out some interesting
88: references, as well as for his other comments.
89: 
90: 
91: 
92: 
93: 
94: \section{Introduction}
95: 
96: Real Legendrian subvarieties are classical objects of differential
97: geometry and they have been investigated for ages.
98: However, complex Legendrian subvarieties in a projective space (see
99: \S\ref{notation_Legendrian} for the definition) are much
100: more rigid and only few smooth and compact examples were known
101: (see \cite{bryant}, \cite{landsbergmanivel04}, \cite{jabu_toric}):
102: \begin{enumerate}
103:   \item 
104:     linear subspaces;
105:   \item
106:     some homogeneous spaces called subadjoint varieties: 
107:     the product of a line and  a quadric $\P^1 \times Q^n$
108:     and five exceptional cases:
109:     \begin{itemize}
110:       \item
111:         twisted cubic curve  $\P^1 \subset \P^3$,
112:       \item
113:         Grassmannian $Gr_L(3,6) \subset \P^{13}$ of Lagrangian subspaces in $\C^6$,
114:       \item
115:         full Grassmannian  $Gr(3,6)\subset \P^{19}$,
116:       \item
117:         spinor variety $\mathbb{S}_6\subset \P^{31}$ 
118:          (i.e. the homogeneous $\mathbf{SO}(12)$-space 
119:          parametrising the vector subspaces of          dimension 6 contained 
120:          in a non-degenerate quadratic cone in $\C^{12}$) and 
121:       \item
122:         the 27-dimensional $E_7$-variety in $\P^{55}$ corresponding to the marked root: 
123:         \includegraphics[width=0.2\textwidth]{subadjoint_E7};
124:     \end{itemize}
125:   \item
126:     every smooth projective curve admits a Legendrian embedding in $\P^3$  \cite{bryant};
127:   \item
128:     a family of surfaces birational to the Kummer $K3$-surfaces \cite{landsbergmanivel04};
129:   \item
130:     the blow up of $\P^2$ in three general points \cite{jabu_toric}. 
131: \end{enumerate}
132: In this article we present a new example in dimension 8 (see theorem
133: \ref{theorem_classify_invertible}(b)). Also we show how does the
134: construction generalise to give new examples in dimensions 5 and 14 
135: (see section \ref{section_other_examples}) and
136: finally we announce a result which will produce plenty of such examples
137: (see section \ref{section_hyperplane}).
138: 
139:  The original motivation for studying Legendrian subvarieties in a complex
140: projective space comes from the studies of contact Fano manifolds%
141: \footnote{
142: A complex projective manifold $M$ of dimension $2n+1$ is called \textbf{a contact manifold},
143: if there exists a vector subbundle $F \subset TM$ of rank $2n$,
144: such that the map $F \otimes F \lra TM/F$ determined by the Lie bracket is nowhere degenerate.
145: In such a case $F$ is called \textbf{a contact distribution}.
146: A projective manifold is \textbf{Fano}, if the anticanonical bundle is ample.
147: }
148: (see \cite{wisniewski}, \cite{kebekus_lines1}, \cite{4authors}):
149: The variety of tangent directions to the minimal rational curves through a fixed point on a contact Fano 
150: manifold make a Legendrian subvariety in the projectivisation of the fibre of contact distribution.
151: The adjoint varieties 
152: (i.e.~the closed orbit of the adjoint action of a simple Lie group $G$ on $\P(\gotg)$)
153: are the only known examples of contact Fano manifolds 
154: and they give rise to the homogeneous Legendrian varieties%
155: \footnote{
156: The groups of types $B$ and $D$ give rise to $\P^1 \times Q^n$. 
157: The five exceptional groups  $G_2$, $F_4$, $E_6$,  $E_7$, $E_8$ 
158: make the exceptional homogeneous Legendrian varieties. The groups of types $A$ and $C$ 
159: are somewhat special --- see \cite{landsbergmanivel04}, \cite{jabu06}.
160: }. 
161: 
162: From our considerations here, some other potential applications come into
163: the view --- see sections \ref{section_other_examples} and
164: \ref{section_compactification}.
165: 
166: \medskip
167: 
168: Before we present our results precisely in section \ref{results}, we must
169: introduce some notation. We need the notation of 
170: \S\ref{notation_begins}-\S\ref{notation_middle_end} to state the
171: results and also \S\ref{notation_middle_begin}-\S\ref{notation_ends} to prove them.
172: 
173: 
174: 
175: 
176: 
177: \subsection{Notation and definitions}\label{section_basics}
178: 
179: For this article we fix an integer $m\ge 2$.
180: 
181: \subsubsection{Vector space $V$}\label{notation_V}
182: \label{notation_begins}
183: Let $V$ be a vector space over complex numbers $\C$ of dimension $2m^2$,
184:  which we interpret as a space of pairs of $m \times m$ matrices. 
185: The coordinates are: $a_{ij}$ and $b_{ij}$ for $i,j \in \{1,\ldots m\}$. 
186: By $A$ we denote the matrix $(a_{ij})$ and similarly for $B$ and $(b_{ij})$.
187: 
188: \medskip
189: 
190: By $\P(V)$ we mean the naive projectivisation of $V$, i.e.~the
191: quotient  $(V \backslash \{0\})/ \C^*$.
192: 
193: \medskip
194: 
195: Given two $m\times m$ matrices $A$ and $B$, by $(A,B)$ we denote the point of the vector space $V$, 
196: while by $[A,B]$ we denote the point of the projective space $\P(V)$.
197: 
198: \medskip
199: 
200: Sometimes, we will represent some linear maps $V\lra V$ and
201: some 2-linear forms $V\otimes V \lra \C$ as $2m^2 \times 2m^2$ matrices. In such a case
202: we will assume the coordinates on $V$ come in the lexicographical
203: order:
204: $$
205: a_{11}, \ldots, a_{1m}, a_{21}, \ldots, a_{mm}, b_{11}, \ldots, b_{1m},
206: b_{21}, \ldots, b_{mm}.
207: $$
208: 
209: \subsubsection{Symplectic form $\omega$}
210: \label{notation_omega}
211: On $V$ we consider the standard symplectic form 
212: %$\omega:= \sum \ud a_{ij} \wedge \ud b_{ij}$  so that:
213: \begin{equation}
214:  \label{properties_of_inner_product}
215: \omega\big((A,B), (A', B')\big):= 
216: \sum_{i,j} (a_{ij}b'_{ij} -a'_{ij}b_{ij}) = \tr \left( A(B')^T - A'B^T \right).
217: \end{equation}
218: Further we set $J$ to be the
219: matrix of $\omega$:
220: $$
221: J:=
222: \left[ 
223: \begin{array}{cc}
224: 0          & \Id_{m^2}\\
225: -\Id_{m^2} & 0
226: \end{array}
227: \right].
228: $$
229: 
230: 
231: 
232: 
233: \subsubsection{Lagrangian and Legendrian subvarieties}
234: \label{notation_Legendrian}
235: 
236: A linear subspace $W \subset V$ is called \emph{Lagrangian}
237: %(resp.~\emph{coisotropic})
238:  if the  $\omega$-perpen\-dicular subspace
239: $W^{\perp_\omega}$ is equal to %(resp.~contained in) 
240: $W$. 
241: Equivalently, $W$ is Lagrangian if and only if $\omega|_W \equiv 0$ and
242: $\dim W$ is maximal possible, i.e.~equal to $\half \dim V$.
243: 
244: \smallskip
245: 
246: A subvariety $Z \subset V$ is called \emph{Lagrangian} 
247: %(resp.~\emph{coisotropic})
248: if for every smooth point $z \in Z$ 
249: the tangent space $T_z Z \subset V$ is Lagrangian%
250: %(resp.~coisotropic)
251: .
252: In particular, if $Z$ is Lagrangian, then $\dim Z= \half \dim V= m^2$.
253: 
254: 
255: \smallskip
256: 
257: A subvariety $X \subset \P(V)$ is defined to be \emph{Legendrian} if its affine cone
258: $\hat{X}\subset V$ is Lagrangian.
259: In particular, if $X$ is Legendrian, then $\dim X= \half \dim V - 1= m^2-1$.
260: 
261: 
262: 
263: 
264: 
265: \subsubsection{Varieties $Y$, $\Xinv(m)$ and $\Xdeg(m,k)$}
266: \label{notation_Y}
267: 
268: We consider the following subvariety of $\P(V)$:
269: \begin{equation}
270: \label{defining_equations_of_Y}
271: Y:=\left\{[A,B]\in \P(V)  \mid AB^T = B^T A =
272: \lambda^2 \Id_m \textrm{ for some } \lambda \in \C\right\}.
273: \end{equation} 
274: The square at $\lambda$ seems to be irrelevant here, but it slightly simplifies the notation in the 
275: proofs  of theorem \ref{theorem_classify_invertible}(b)
276: and proposition \ref{proposition_orbits_of_G}(ii)
277: 
278: \label{notation_Xinv_Xdeg}
279: 
280: Further we define two types of subvarieties of $Y$:
281: $$
282: \Xinv(m):=\overline{
283: \bigg\{\left[g,\left(g^{-1}\right)^T\right] \in \P(V) \mid \det g = 1 \bigg\}}
284: $$
285: $$
286: \Xdeg(m,k):=\Big\{[A,B] \in \P(V) \mid  AB^T=B^TA=0, \ \rk A
287:   \le k, \ \rk B \le m-k  \Big\} 
288: $$
289: where $k \in {0,1,\ldots m}$.
290: The varieties $\Xdeg(m,k)$ have been also studied by \cite{strickland}
291: and \cite{mehta_trivedi}. $\Xinv(m)$ (especially $\Xinv(3)$) 
292: is the main object of this article.
293: 
294: \subsubsection{Automorphisms $\psi_{\mu}$} \label{notation_psi}
295: 
296: For any $\mu \in \C^*$ we let $\psi_{\mu}$ be the following linear
297: automorphism of $V$:
298: $$
299: \psi_{\mu}\big((A,B)\big) := (\mu A, \mu^{-1} B).
300: $$
301: 
302: Also the induced automorphism of $\P(V)$ will be denoted in the same
303: way:
304: $$
305: \psi_{\mu}\big([A,B]\big) := [\mu A, \mu^{-1} B].
306: $$
307: 
308: 
309: 
310: 
311: 
312: 
313: \medskip
314: 
315: The notation introduced so far is sufficient to state the results of this
316: paper (see section \ref{results}), but to prove them we need a few more
317: notions. 
318: %For the convenience of reader we group all the notation in
319: %one place.
320: 
321: 
322: 
323: \label{notation_middle_end}
324: 
325: \subsubsection{Groups $G$ and $\widetilde{G}$, Lie algebra $\gotg$ and
326:    their representation}
327: \label{notation_G}
328: 
329: \label{notation_middle_begin}
330: 
331: We set $\widetilde{G}:=\Gl_m \times \Gl_m$ and let it act on $V$ by:
332: $$
333: (g,h) \in \widetilde{G}, \ g,h \in \Gl_m, \ (A,B) \in V
334: $$
335: $$
336: (g,h) \cdot (A,B) := (g^T A h, g^{-1} B (h^{-1})^T).
337: $$
338: This action preserves the symplectic form $\omega$.
339: 
340: We will mostly consider the restricted action of 
341: $G:=\Sl_m \times \Sl_m < \widetilde{G}$.
342: 
343: We also set $\gotg:= \gotsl_m \times \gotsl_m$ to be the Lie algebra of $G$ and we have the
344: tangent action of $\gotg$ on $V$:
345: $$
346: (g,h) \cdot (A,B) = (g^T A + A h, - g B - B h^T).
347: $$
348: Though we denote the action of the groups $G$, $\widetilde{G}$ 
349: and the Lie algebra $\gotg$ by the same $\cdot$ 
350: we hope it will not lead to any confusion. 
351: Also the induced action of $G$ and $\widetilde{G}$ on $\P(V)$ will
352: be denoted by $\cdot$.
353: 
354: 
355: 
356: 
357: 
358: 
359: 
360: \subsubsection{Orbits $\Inv^m$ and $\cDeg^m_{k,l}$}
361: \label{notation_Inv_Deg}
362: 
363: We define the following sets:
364: $$
365: \Inv^m:=
366: \bigg\{\left[g,\left(g^{-1}\right)^T\right] \in \P(V) \mid \det g = 1 \bigg\},
367: $$
368: $$
369: \cDeg^m_{k,l}:= \Big\{[A,B] \in \P(V) \mid  AB^T=B^TA=0, \ \rk A = k, \
370: \rk B = l \Big\},
371: $$
372: so that $\Xinv(m)=\overline{\Inv^m}$ and $\Xdeg(m,k)= \overline{\cDeg^m_{k,m-k}}$. 
373: 
374: Clearly, if $k+l>m$ then $\cDeg^m_{k,l}$ is empty, so whenever
375: speaking of $\cDeg^m_{k,l}$ we will assume $k+l \le m$.
376: 
377: 
378: \subsubsection{Elementary matrices $E_{ij}$ and points $p_1$ and $p_2$}
379: \label{notation_Eij}
380: 
381: 
382: Let $E_{ij}$ be the elementary $m \times m$ matrix with
383: unit in the $i^{\textrm{th}}$ row and the $j^{\textrm{th}}$ column and zeroes
384: elsewhere. 
385: 
386: 
387: \label{notation_p1_p2}
388: 
389: We distinguish two points 
390: $p_1 \in \cDeg^m_{1,0}$ and
391: $p_2 \in \cDeg^m_{0,1}$:
392: $$
393: p_1:=
394: \left[ E_{mm}, 0 \right]
395: \
396: \textrm{ and }
397: \
398: p_2:=
399: \left[ 0, E_{mm} \right]
400: $$
401: 
402: These points will be usually chosen as nice representatives of the
403: closed orbits
404: $\cDeg^m_{1,0}$ and $\cDeg^m_{0,1}$.
405: 
406: 
407: 
408: 
409: \subsubsection{Tangent cone}\label{properties_of_tangent_cone}
410: 
411: We recall the notion of the tangent cone and a few among many of its
412: properties. For more details and the proofs we refer to 
413: \cite[lecture 20]{harris} and \cite[III.\S3,\S4]{mumford}.
414: 
415: 
416: For an irreducible Noetherian scheme $X$ over $\C$ and a closed point $x\in X$ 
417: we consider the local ring $\ccO_{X,x}$ and we 
418: let $\gotm_x$ to be the maximal ideal in $\ccO_{X,x}$.
419: Let 
420: $$
421: R:=\bigoplus_{i=0}^{\infty} \left(\gotm_x^i \slash \gotm_x^{i+1} \right)
422: $$
423: where $\gotm_x^0$ is just the whole $\ccO_{X,x}$. 
424: Now we define \emph{the tangent cone $TC_x X$ at $x$ to $X$} to be $\Spec R$.
425: 
426: 
427: If $X$ is a subscheme of an affine space $\A^n$ (which we will usually
428: assume to be an affine piece of a projective space)
429: the tangent cone at $x$ to $X$ can be understood as a subscheme of $\A^n$. 
430: Its equations can be derived from the ideal of $X$.
431: For simplicity assume $x=0 \in \A^n$ and then the polynomials defining $TC_0 X$ 
432: are the lowest degree homogeneous parts of the polynomials in the ideal of $X$.
433: 
434: Another interesting point-wise definition is that $v \in TC_0 X$ is a closed point if and only if 
435: there exists a holomorphic map $\varphi_v$ from
436: the disc $D_t:=\{ t\in \C  \ : \ \lvert t \rvert\ < \delta \}$ to $X$,
437: such that $\varphi_v(0) = 0$
438:  and the first non-zero coefficient in the Taylor expansion in $t$ of $\varphi_v(t) $
439: is $v$, i.e.:
440: $$ \begin{array}{rccl}
441: \varphi_v: & D_t & \lra & X\\
442:            &  t  & \mapsto & t^k v + t^{k+1}v_{k+1} +\ldots
443: \end {array}$$
444: 
445: \smallskip
446: 
447: We list some of the properties of the tangent cone, that will be used
448: freely in the proofs:
449: 
450: \begin{itemize}
451: \item[(1)]
452: The dimension of every component of $TC_x X$ is equal to the dimension of $X$. 
453: \item[(2)] $TC_x X$ is naturally embedded in the Zariski tangent space to $X$ at $x$ and 
454: $TC_x X$ spans the tangent space.
455: \item[(3)] 
456: $X$ is regular at $x$ if and only if $TC_x X$ is equal (as scheme) to
457: 	  the tangent space.
458: %\item[(4)]
459: %If $TC_x X$ is reduced, then $X$ is reduced at $x$.
460: \end{itemize}
461: 
462: 
463: %\begin{prf}
464: %The property (4) follows  from the definition and Krull Intersection Theorem 
465: %(see for example \cite[cor. 5.4]{eisenbud}).
466: %\end{prf}
467: 
468: 
469: 
470: \subsubsection{Submatrices - extracting rows and columns}\label{notation_submatrices}
471: 
472: 
473: Assume $A$ is an $m\times m$ matrix and $I,J$ are two sets of indices of
474: cardinality $k$ and $l$ respectively:
475: $$
476: I:=\{i_1,i_2, \ldots, i_k | 1 \le i_1 <  i_2 < \ldots < i_k \le m\},
477: $$
478: $$ 
479: J:=\{j_1,j_2, \ldots, j_l | 1 \le j_1 < j_2 < \ldots < j_l \le m\}.
480: $$
481: Then we denote by $A_{I,J}$ the $(m-k) \times (m-l)$ submatrix of $A$ obtained by 
482: removing rows of indices $I$ and columns of indices $J$.
483: Also for a set of indices $I$ we denote by $I'$ the set of $m-k$ indices complementary to $I$.
484: 
485: \smallskip
486: 
487: We will also use a simplified version of the above notation, when we
488: remove only a single column and single row:
489: $A_{ij}$ denotes the $(m-1) \times (m-1)$ submatrix of $A$ obtained by removing $i$-th row
490: and $j$-th column, i.e. $A_{ij} = A_{\{i\},\{j\}}$
491: 
492: \smallskip
493: 
494: Also in the simplest situation where we remove only the last row and
495: the last column, we simply write $A_m$, so that $A_m = A_{mm} = A_{\{m\},\{m\}}$.
496: 
497: 
498: 
499: 
500: 
501: 
502: 
503: 
504: \label{notation_ends}
505: 
506: \subsection{Main Results} \label{results}
507: 
508: 
509: 
510: In this note we give a classification%
511: \footnote{This problem was suggested by Sung Ho Wang.} 
512: of Legendrian subvarieties
513: in $\P(V)$ that are contained in $Y$.
514: 
515: 
516: 
517: \begin{theo}
518: \label{classification_theorem}
519: Let projective space $\P(V)$, varieties $Y$, $\Xinv(m)$,
520:  $\Xdeg(m,k)$ and automorphisms $\psi_{\mu}$ be defined as in
521:  \S\ref{notation_begins}-\S\ref{notation_middle_end}.
522: Assume $X\subset\P(V)$ is an irreducible subvariety.
523: Then $X$ is Legendrian and contained in $Y$ if and only if 
524: $X$ is one of the following varieties:
525: \begin{itemize}
526: \item[1.]
527: $X = \psi_{\mu}(\Xinv(m))$ for some $\mu \in \C^*$ or 
528: \item[2.]
529: $X=\Xdeg(m,k)$ for some $k \in \{0,1,\ldots m\}$.
530: 
531: \end{itemize}
532: \end{theo}
533: 
534: The idea of the proof of theorem \ref{classification_theorem} is based on
535: the observation that every Legendrian subvariety that is contained in $Y$
536: must be invariant under the action of group $G$. This is
537: explained in section \ref{section_action}. A proof of the theorem is
538: presented in section \ref{section_classification}.
539: 
540: Also we analyse which of the above varieties appearing in 1.~and 2.~are smooth:
541: 
542: 
543: \begin{theo}
544:     \label{theorem_classify_invertible}
545:   With the definition of $\Xinv(m)$ as in \S\ref{notation_Xinv_Xdeg},
546:   the family $\Xinv(m)$ contains the following varieties:
547:   \begin{itemize}
548:     \item[(a)] 
549:       $\Xinv(2)$ is a linear subspace.
550:     \item[(b)] 
551:       $\Xinv(3)$ is smooth, its  Picard group is generated by a hyperplane section.
552:       Moreover $\Xinv(3)$ is a compactification of $\Sl_3$ and it is
553:       isomorphic to a hyperplane section of Grassmannian $Gr(3,6)$. 
554:       The connected component of $Aut(\Xinv(3))$ is equal to $G=\Sl_3 \times \Sl_3$ 
555:       and $\Xinv(3)$ is not a homogeneous space.
556:     \item[(c)]
557:       $\Xinv(4)$ is the 15 dimensional spinor variety $\mathbb{S}_6$.
558:     \item[(d)] 
559:       For $m\ge 5$, the variety $\Xinv(m)$ is singular.
560:   \end{itemize}
561: \end{theo}
562: 
563: A proof of the theorem is explained in section \ref{section_invertible}.
564: 
565: Variety $\Xinv(3)$ is not yet described as a Legendrian subvariety
566: variety, so it is our new smooth example of dimension 8. 
567: 
568: 
569: 
570: 
571: 
572: \begin{theo}
573: \label{theorem_smooth_degenerate}
574: With the definition of $\Xdeg(m)$ as in \S\ref{notation_Xinv_Xdeg},
575: variety $\Xdeg(m,k)$ is smooth if and only if $k=0$ , $k=m$ or $(m,k)=(2,1)$. 
576: In the first two cases, $\Xdeg(m,0)$ and $\Xdeg(m,m)$ are linear spaces, 
577: while $\Xdeg(2,1) \simeq \P^1 \times \P^1 \times \P^1 \subset \P^7$.
578: \end{theo}
579: 
580: A proof of the theorem is presented in section \ref{section_degenerate}.
581: 
582: \medskip
583: 
584: 
585: The results of theorems 
586: \ref{classification_theorem},
587: \ref{theorem_classify_invertible}
588: can be generalised in (at least) three different directions:
589: 
590: 
591: \subsection{Generalisation 1: Representation theory}
592:     \label{section_other_examples}
593: 
594: 
595: 
596: 
597: 
598: The interpretation of theorem
599: \ref{theorem_classify_invertible} (b) and (c) 
600: can be following:
601: We take the exceptional Legendrian variety $Gr(3,6)$, slice it with a linear section 
602: and we get a description, that generalised to matrices of bigger size gives 
603: the bigger exceptional Legendrian variety $\bS_6$.
604: Similar connection can be established between other exceptional Legendrian varieties.
605: 
606: \medskip
607: 
608: For instance, assume that $V^{sym}$ is a vector space of dimension $2 \binom{m+1}{2}$,
609: which we interpret as the space of pairs of $m \times m$ symmetric matrices $A,B$.
610: Now in $\P(V^{sym})$ consider the subvariety $\Xinv^{sym}(m)$, which is the closure of the following set:
611: $$
612: \{[A,A^{-1}]\in \P(V^{sym}) | A=A^T \textrm{ and } \det A=1\}.
613: $$
614: \begin{theo}
615: All the varieties $\Xinv^{sym}(m)$ are Legendrian and we have:
616: \begin{itemize}
617: \item[(a)] $\Xinv^{sym}(2)$ is a linear subspace.
618: \item[(b)] $\Xinv^{sym}(3)$ is smooth and it is isomorphic to a
619: 	   hyperplane section of Lagrangian Grassmannian $Gr_L(3,6)$.
620: \item[(c)] $\Xinv^{sym}(4)$ is smooth and it is the 9 dimensional Grassmannian variety $Gr(3,6)$.
621: \item[(d)] For $m\ge 5$, the variety $\Xinv^{sym}(m)$ is singular.
622: \end{itemize}
623: \end{theo}
624: 
625: The proof goes exactly as the proof of theorem \ref{theorem_classify_invertible}. 
626: 
627: 
628: \medskip
629: 
630: 
631: 
632: Similarly, we can take $V^{skew}$ to be a vector space of dimension $2 \binom{2m}{2}$,
633: which we interpret as the space of pairs of $2m \times 2m$ skew-symmetric matrices $A,B$.
634: Now in $\P(V^{skew})$ consider subvariety $\Xinv^{skew}(m)$, which is the closure of the following set:
635: $$
636: \{[A, - A^{-1}]\in \P(V^{skew}) | A= - A^T \textrm{ and } \pf A=1\}.
637: $$
638: \begin{theo}
639: All the varieties $\Xinv^{skew}(m)$ are Legendrian and we have:
640: \begin{itemize}
641: \item[(a)] $\Xinv^{skew}(2)$ is a linear subspace.
642: \item[(b)] $\Xinv^{skew}(3)$ is smooth and it is isomorphic to a hyperplane section of the spinor variety $\bS_6$.
643: \item[(c)] $\Xinv^{skew}(4)$ is smooth and it is the 27 dimensional $E_7$ variety.
644: \item[(d)] For $m\ge 5$, the variety $\Xinv^{skew}(m)$ is singular.
645: \end{itemize}
646: \end{theo}
647: 
648: Here the only difference is that we replace the determinants by the
649: Pfaffians of the appropriate submatrices
650: and also for the previous cases we will be picking some diagonal matrices
651: as nice representatives.
652: Since there is no non-zero skew-symmetric diagonal matrix,
653: we must modify a little bit our calculations, but there is no
654: essential difference in the technique.
655: 
656: 
657: \medskip
658: 
659: Neither $\Xinv^{sym}(3)$ nor $\Xinv^{skew}(3)$ 
660: have been described as smooth Legendrian subvarieties.
661: 
662: \medskip
663: 
664: Therefore we have established a new connection 
665: between the subadjoint varieties of the 4 exceptional
666: groups $F_4$, $E_6$, $E_7$ and $E_8$.
667: A similar connection was obtained by \cite{landsbergmanivel02}. 
668: 
669: 
670: 
671: %Also it can be interesting to understand why this construction fails for the last step.
672: %For $Gr_L (3,6)$, $Gr(3,6)$, $\bS_6$, we just took a rank 4 Jordan algebra (corresponding respectively 
673: %to reals, complex numbers and quaternions) 
674: %and related to it the appropriate variety.
675: %So what happens if we take a $4\times 4$ octonionic hermitian matrices?
676: %One obstruction for defining the variety is that the determinants of degree 3 are not well defined
677: %(because octonions are not associative),
678: %so it is hard to say what $A^{-1}$ should be. 
679: %But note that to define $\Xinv^{sym}(4)$, $\Xinv(4)$ and  $\Xinv^{skew}(4)$, 
680: %we only use quadratic equations which can be expressed in a form of some matrix multiplication
681: %or as some combination of some $2 \times 2$ minors (or $4\times 4$ Pfaffians).
682: %So the only obstruction now is the noncommutativity 
683: %and somehow  it is not an obstruction for the quaternion case.
684: 
685: 
686: 
687: \subsection{Generalisation 2: Hyperplane section}\label{section_hyperplane}
688: 
689: The variety $\Xinv(3)$ is the first described example of smooth non-homogeneous Legendrian variety of dimension
690: bigger than $2$ (see \cite{bryant}, \cite{landsbergmanivel04}, \cite{jabu_toric}).
691: But this example is very close to a homogeneous one, namely is isomorphic to a hyperplane section of $Gr(3,6)$, 
692: which is a well known Legendrian variety. 
693: So a natural question arises, 
694: whether a general hyperplane section of other Legendrian varieties admits Legendrian embedding. 
695: The answer is yes and we explain it (as well as many conclusions from this surprisingly simple observation)
696: in \cite{jabu_hyperplane}.
697: 
698: 
699: 
700: 
701: \subsection{Generalisation 3: Group compactification}\label{section_compactification}
702: 
703: Theorem \ref{theorem_classify_invertible}(b) says that $\Xinv(3)$ is a
704: smooth compactification of $\Sl_3$. 
705: In \cite{jabu_compactifications} we study a generalisation of this construction 
706: (which is not really related to Legendrian varieties) 
707: to find a family of compactifications of $\Sl_n$, 
708: which contains the smooth compactification of $\Sl_3$ 
709: and can be easily smoothened (by a single blow up of a closed orbit) for
710: $n=4$.
711: 
712: 
713: 
714: 
715: 
716: 
717: 
718: 
719: 
720: 
721: 
722: 
723: 
724: 
725: 
726: \section{$G$-action and its orbits}\label{section_action}
727: 
728: In \cite{jabu06} we prove:
729: 
730: \begin{theo} \label{theorem_ideal_and_group}
731:   Let $X \subset \P(V)$ be a Legendrian subvariety (see \S\ref{notation_Legendrian} for definition).
732:   Consider the following map:
733:   $$
734:     H^0(\ccO_{\P(V)}(2)) \simeq \Sym^2 V^* \ni  q = (x \mapsto x^T M(q) x) 
735:     \stackrel{\rho}{\mapsto} 2J \cdot M(q) \in \gotsp(V).
736:   $$
737:   where $M(q)$ is the $(2m^2) \times (2m^2)$ matrix of $q$ and $J$ is the matrix of the symplectic   form as in \S\ref{notation_V}.
738:   Let $\I_2(X) \subset \Sym^2 V^*$ be the vector space of quadrics containing $X$. Then:
739:   \begin{itemize} 
740:     \item
741:       $\rho(\I_2(X))$ is a Lie subalgebra of $\gotsp(V)$ tangent to a closed subgroup 
742:       $$
743:         \overline{\exp\Big(\rho\big(\I_2(X)\big)\Big)} < \Sp(V).
744:       $$
745:     \item
746:       We have the natural action of $\Sp(V)$ on $\P(V)$. 
747:       The group 
748:       $\overline{\exp\Big(\rho\big(\I_2(X)\big)\Big)}$
749:       is the maximal connected subgroup in $\Sp(V)$ 
750:       which under this action preserves $X \subset \P(V)$.
751:   \end{itemize}
752: \end{theo}
753: 
754: \begin{prf}
755:   See \cite[cor.~4.4, cor.~5.5, lem.~5.6]{jabu06}.
756: \end{prf}
757: 
758: 
759: Recall the definition of $Y$ in \S\ref{notation_Y}.
760: 
761: The following polynomials are in 
762: the homogeneous ideal of $Y$ (below $i,j$ are indices that run through $\{1,\ldots, m\}$, $k$ is a summation index):
763: \begin{subequations}\label{equations_of_Y}
764: \begin{align}
765: \sum_{k=1}^m a_{ik} b_{ik}  - \sum_{k=1}^m a_{1k} b_{1k}
766: \label{equations_of_Y1}\\
767: \sum_{k=1}^m a_{ik} b_{jk} \textrm{ for $i \ne j$}
768: \label{equations_of_Y2}\\
769: \sum_{k=1}^m a_{ki} b_{ki}  - \sum_{k=1}^m a_{k1} b_{k1}
770: \label{equations_of_Y3}\\
771: \sum_{k=1}^m a_{ki} b_{kj} \textrm{ for $i \ne j$}
772: \label{equations_of_Y4}
773: \end{align}
774: \end{subequations}
775: 
776: These equations simply come from eliminating $\lambda$ from the
777: defining equation of $Y$ --- see equation \eqref{defining_equations_of_Y}.
778: 
779: 
780: 
781: For the statement and proof of the following proposition, recall our
782: notation of 
783: \S\ref{notation_V}, 
784: \S\ref{notation_omega}, 
785: \S\ref{notation_Legendrian}, 
786: \S\ref{notation_G} and
787: \S\ref{notation_Eij}.
788: 
789: 
790: 
791: \begin{prop}
792: \label{proposition_action_of_G}
793: Let $X\subset \P(V)$ be a Legendrian subvariety.
794: If $X$ is contained in $Y$ 
795: then $X$ is preserved by the induced action of $G$ on $\P(V)$.
796: \end{prop}
797: 
798: 
799: 
800: 
801: \begin{prf}
802: Let $\I_2(X)$ be as in the theorem \ref{theorem_ideal_and_group} 
803: and define  $\I_2(Y)$ analogously.
804: Clearly $\I_2(Y) \subset \I_2(X)$. 
805: By theorem \ref{theorem_ideal_and_group} it is enough to calculate that 
806: $\gotg \subset \rho\left(\I_2(Y)\right)$ or that the images of the quadrics 
807: \eqref{equations_of_Y1}--\eqref{equations_of_Y4} under $\rho$ generate $\gotg$.
808: 
809: We deal in details of the proof only for $m=2$. 
810: There is no difference between this case and the general
811: one, except for the complexity of notation%
812: %(which should be 3-dimensional at this point to be clear enough).
813: .
814: 
815: Let us take the quadric
816: $$
817: q_{ij}:=\sum_{k=1}^m a_{ik} b_{jk} = a_{i1} b_{j1} + a_{i2} b_{j2} 
818: $$
819: for any $i,j \in \{1,\ldots, m\} = \{1,2\}$.
820: Also let $Q_{ij}$ be the $2m^2 \times 2m^2$ symmetric matrix
821: corresponding to $q_{ij}$. For instance:
822: $$
823: Q_{12} =
824: \left[ 
825: \begin{array}{cccccccc}
826: 0&0 &0&0 &0&0 &\half & 0 \\
827: 0&0 &0&0 &0&0 &0 & \half \\
828: 0&0 &0&0 &0&0 &0 & 0 \\
829: 0&0 &0&0 &0&0 &0 & 0 \\
830: 0&0 &0&0 &0&0 &0 & 0 \\
831: 0&0 &0&0 &0&0 &0 & 0 \\
832: \half&0 &0&0 &0&0 &0 & 0 \\
833: 0&\half &0&0 &0&0 &0 & 0
834: \end{array}
835: \right] .
836: $$
837: 
838: 
839: 
840: So choose an arbitrary $(A,B) \in V$ and at the
841: moment we want to think of it as of a single vertical $2 m^2$-vector:
842: $(A,B) = [a_{11}, a_{12},a_{21}, a_{22},b_{11}, b_{12},b_{21},
843: b_{22}]^T$, so that the following multiplication makes sense:
844: $$
845: \rho(q_{12}) = 
846: 2 J \cdot Q_{12} \cdot (A,B) = 
847: $$
848: $$
849: = \left[ 
850: \begin{array}{cccccccc}
851: 0&0 &0&0 &1&0 &0 & 0 \\
852: 0&0 &0&0 &0&1 &0 & 0 \\
853: 0&0 &0&0 &0&0 &1 & 0 \\
854: 0&0 &0&0 &0&0 &0 & 1 \\
855: -1&0 &0&0 &0&0 &0 & 0 \\
856: 0&-1 &0&0 &0&0 &0 & 0 \\
857: 0&0 &-1&0 &0&0 &0 & 0 \\
858: 0&0 &0&-1 &0&0 &0 & 0
859: \end{array}
860: \right]
861: \left[ 
862: \begin{array}{cccccccc}
863: 0&0 &0&0 &0&0 &1 & 0 \\
864: 0&0 &0&0 &0&0 &0 & 1 \\
865: 0&0 &0&0 &0&0 &0 & 0 \\
866: 0&0 &0&0 &0&0 &0 & 0 \\
867: 0&0 &0&0 &0&0 &0 & 0 \\
868: 0&0 &0&0 &0&0 &0 & 0 \\
869: 1&0 &0&0 &0&0 &0 & 0 \\
870: 0&1 &0&0 &0&0 &0 & 0
871: \end{array}
872: \right]
873: \left[ 
874: \begin{array}{c}
875: a_{11}\\
876: a_{12}\\
877: a_{21}\\
878: a_{22}\\
879: b_{11}\\
880: b_{12}\\
881: b_{21}\\
882: b_{22}
883: \end{array}
884: \right]
885: =
886: $$
887: $$
888: =
889: \left[ 
890: \begin{array}{cccccccc}
891: 0&0 &0&0 &0&0 &0 & 0 \\
892: 0&0 &0&0 &0&0 &0 & 0 \\
893: 1&0 &0&0 &0&0 &0 & 0 \\
894: 0&1 &0&0 &0&0 &0 & 0 \\
895: 0&0 &0&0 &0&0 &-1 & 0 \\
896: 0&0 &0&0 &0&0 &0 & -1 \\
897: 0&0 &0&0 &0&0 &0 & 0 \\
898: 0&0 &0&0 &0&0 &0 & 0 
899: \end{array}
900: \right]
901: \left[ 
902: \begin{array}{c}
903: a_{11}\\
904: a_{12}\\
905: a_{21}\\
906: a_{22}\\
907: b_{11}\\
908: b_{12}\\
909: b_{21}\\
910: b_{22}
911: \end{array}
912: \right]
913: =
914: $$
915: $$
916: =
917: \left[ 
918: \begin{array}{c}
919: 0\\
920: 0\\
921: a_{11}\\
922: a_{12}\\
923: -b_{21}\\
924: -b_{22}\\
925: 0\\
926: 0
927: \end{array}
928: \right]
929: \stackrel{\textrm{back to the matrix notation} }{=}
930: \left(
931: \left[ 
932: \begin{array}{cc}
933: 0& 0 \\
934: a_{11} & a_{12}
935: \end{array}
936: \right] , \
937: \left[ 
938: \begin{array}{cc}
939: -b_{21} & -b_{22}\\
940: 0& 0
941: \end{array}
942: \right]
943: \right)
944: = 
945: $$
946: $$
947: =
948: \left(
949: \left[ 
950: \begin{array}{cc}
951: 0 &1 \\
952: 0 &0
953: \end{array}
954: \right]^T
955: \left[ 
956: \begin{array}{cc}
957: a_{11} & a_{12}\\
958: a_{21} & a_{22}\\
959: \end{array}
960: \right] , \ 
961: - \left[ 
962: \begin{array}{cc}
963: 0& 1 \\
964: 0 &0
965: \end{array}
966: \right]
967: \left[ 
968: \begin{array}{cc}
969: b_{11} & b_{12}\\
970: b_{21} & b_{22}
971: \end{array}
972: \right]
973: \right)
974:  \ = \ (E_{12}^T A, \  -E_{12} B)
975: $$
976: 
977: Going along exactly the same calculations, we see that:
978: 
979: %\begin{equation} \label{sl_elementary_action}
980: $$
981: 2 J \cdot Q_{ij} \cdot (A,B)  \ = \ (E_{ij}^T A, \  -E_{ij} B)
982: $$
983: %\end{equation}
984: 
985: Next in the ideal of $Y$ we have the following quadrics: $q_{ij}$ for
986: $i \ne j$ (see \eqref{equations_of_Y2}) and $q_{ii}-q_{11}$ (see
987: \eqref{equations_of_Y1}).
988: By taking images under $\rho$ of the linear combinations of those quadrics we can get an arbitrary
989: traceless matrix $g\in \gotsl_m$ acting on $V$ in the following way:
990: $$
991: g \cdot (A,B) = (g^T A, -g B).
992: $$
993: Exponentiate this action of $\gotsl_m$ to get the action of $\Sl_m$:
994: $$
995: g \cdot (A,B) = (g^T A, g^{-1} B).
996: $$
997: 
998: This proves of that the action of subgroup $\Sl_m \times 0 < G = \Sl_m \times \Sl_m$
999: indeed preserves $X$ as claimed in the lemma. 
1000: The action of the other component $0 \times \Sl_m$ is calculated in the same way, but
1001: using quadrics \eqref{equations_of_Y3}--\eqref{equations_of_Y4}.
1002: \end{prf}
1003: 
1004: 
1005: \subsection{Invariant subsets}\label{section_orbits}
1006: 
1007: 
1008: Recall our notation of 
1009: \S\ref{notation_V},
1010: \S\ref{notation_Y},
1011: \S\ref{notation_psi},
1012: \S\ref{notation_G} and
1013: \S\ref{notation_Inv_Deg}.
1014: Here we want to decompose $Y$ into a union of some $G$-invariant
1015: subsets, most of which are orbits. 
1016: 
1017: 
1018: \begin{prop}
1019: \label{proposition_orbits_of_G}
1020: \hfill
1021: \begin{itemize}
1022: \item[(i)]
1023: The sets $\Inv^m$, $\psi_{\mu}(\Inv^m)$ and $\cDeg^m_{k,l}$ 
1024: are $G$-invariant and they are all contained in $Y$.
1025: \item[(ii)]
1026: $Y$ is equal to the union of all $\psi_{\mu}(\Inv^m)$ (for
1027:   $\mu \in \C^*$) and all $\cDeg^m_{k,l}$ (for integers $k,l\ge
1028:   0$, $k+l\le m$).
1029: \item[(iii)]
1030: Every $\psi_{\mu}(\Inv^m)$ is an orbit of the action of $G$. 
1031: If $m$ is odd, then $\Inv^m$ is isomorphic (as algebraic variety) to
1032: $\Sl_m$. Otherwise if $m$ is even, then $\Inv^m$ is isomorphic to 
1033: $(\Sl_m / \Z_2) $. In both cases, $\dim \psi_{\mu}(\Inv^m) =  \dim
1034: \Inv^m = m^2 -1$. 
1035: \end{itemize}
1036: \end{prop}
1037: 
1038: \begin{prf}
1039: The proof of part (i) is an explicit verification from the definitions
1040: in \S\ref{section_basics}.
1041: 
1042: \medskip
1043: 
1044: To prove part (ii), assume $[A,B]$ is a point of $Y$, so $A B^T=B^TA=\lambda^2\Id_m$. 
1045: First assume that the ranks of both matrices are maximal: 
1046: $$
1047: \rk A = \rk B =m. 
1048: $$
1049: Then $\lambda$ must be non-zero and $B=\lambda^2 (A^{-1})^T$.
1050: Let $d:=(\det A)^{-\frac{1}{m}} $ so that 
1051: $$
1052: \det (dA)=1
1053: $$
1054: and let 
1055: $\mu:= \frac{1}{d \lambda} $.
1056: Then we have:
1057: $$
1058: [A,B] =
1059:  \left[A,\lambda^2 \left(A^{-1}\right)^T\right] =
1060:  \left[ \frac{dA}{d \lambda }, d\lambda \left((dA)^{-1}\right)^T\right] =
1061: $$
1062: $$
1063: % = [ dA, \mu^{-2} ((dA)^{-1})^T]
1064: = \left[ \mu (dA), \mu^{-1} \left((dA)^{-1}\right)^T\right]= 
1065:  \psi_{\mu} \left(\left[ (dA), \left((dA)^{-1}\right)^T\right]\right).
1066: $$
1067: Therefore $[A,B] \in \psi_{\mu}(\Inv^m)$.
1068: 
1069: Next, if either of the ranks is not maximal:
1070: $$
1071: \rk A < m \textrm{ or } \rk B < m
1072: $$
1073: then by \eqref{defining_equations_of_Y} we must have $AB^T = B^TA =
1074: 0$. 
1075: So $[A,B] \in \cDeg^m_{k,l}$ for $k=\rk A$ and $l=\rk B$. 
1076: 
1077: \medskip
1078: 
1079: Now we prove (iii).
1080: The action of $G$ commutes with $\psi_{\mu}$:
1081: $$
1082: (g,h)\cdot \psi_{\mu}\big([A,B]\big) = \psi_{\mu}\big((g,h) \cdot [A,B]\big).
1083: $$
1084: So to prove $\psi_{\mu}(\Inv^m)$ is an orbit it is enough to prove
1085: that $\Inv^m$ is an orbit, which follows from the definitions of the
1086: action and $\Inv^m$.
1087: 
1088: We have the following epimorphic map:
1089: $$
1090: \begin{array}{rcl}
1091: \Sl_m &\lra & \Inv^m\\
1092: g &\mapsto& [g,(g^{-1})^T]
1093: \end{array}
1094: $$
1095: If $[g_1,(g_1^{-1})^T] = [g_2,(g_2^{-1})^T]$ then we must have
1096: $g_1 = \alpha g_2$ and $g_1 = \alpha^{-1} g_2$ for some 
1097: $\alpha \in \C^*$. 
1098: Hence $\alpha^2=1$ and $g_1 = \pm g_2$. If $m$ is odd and 
1099: $g_1\in \Sl_m$ then $-g_1 \notin \Sl_m$ so $g_1 = g_2$. So $\Inv^m$ is 
1100: either isomorphic to $\Sl_m$ or to $\Sl_m /\Z_2$ as stated.
1101: \end{prf}
1102: 
1103: 
1104: From proposition \ref{proposition_orbits_of_G}(ii) we conclude that 
1105: $\Xinv(m)$ is an equivariant compactification of $\Sl_m$ (if $m$ is odd) 
1106: or $\Sl_m / \Z_2$ (if $m$ is even).
1107: See \cite{timashev} and references therein 
1108: for the theory of equivariant compactifications. 
1109: In the setup of \cite[\S8]{timashev}, 
1110: this is the compactification corresponding to the representation $W \oplus W^*$,
1111: where $W$ is the standard representation of $\Sl_m$. 
1112: Therefore some properties of $\Xinv(m)$ could also be read from the general description 
1113: of group compactifications.
1114: 
1115: 
1116: \begin{prop}
1117: \label{degenerate_orbits_of_G}
1118: \hfill
1119: \begin{itemize}
1120: \item[(i)]
1121: The dimension of $\cDeg^m_{k,l}$ is $(k+l)(2m-k-l)-1$.
1122: In particular, if $k+l=m$ then the dimension is equal to $m^2-1$.
1123: \item[(ii)]
1124: $\cDeg^m_{k,l}$ is an orbit of the action of $G$, unless $m$ is even
1125:   and $k=l= \half m$. 
1126: \item[(iii)]
1127: If $m\ge 3$, then there are exactly two closed orbits of the action of
1128: $G$: $\cDeg^m_{1,0}$ and $\cDeg^m_{0,1}$.
1129: \end{itemize}
1130: \end{prop}
1131: 
1132: \begin{prf}
1133: Part (i) follows from \cite[prop 2.10]{strickland}.
1134: 
1135: \medskip
1136: 
1137: For part (ii) 
1138: let $[A,B] \in \cDeg^m_{k,l}$ be any point.
1139: By Gauss elimination and elementary linear algebra,
1140: we can prove that there exists 
1141: $(g,h) \in G$ such that $[A',B']:=(g,h) \cdot [A,B]$ is a pair of
1142: diagonal matrices. Moreover, if $k+l < m$ then we can choose $g$ and
1143: $h$ such that:
1144: $$
1145: A':=\diag(\underbrace{1,\ldots, 1}_{k},\underbrace{0,\ldots, 0}_{l},
1146: \underbrace{0,\ldots, 0}_{m-k-l}),
1147: $$\nopagebreak
1148: $$
1149: B':=\diag(\underbrace{0,\ldots, 0}_{k},\underbrace{1,\ldots, 1}_{l},
1150: \underbrace{0,\ldots, 0}_{m-k-l}).
1151: $$
1152: Hence $\cDeg^m_{k,l} = G\cdot [A',B']$  and this finishes the proof in the case
1153: $k+l < m$. 
1154: 
1155: So assume $k+l=m$. Then we can choose $(g,h)$ such that:
1156: $$
1157: A':=\diag(\underbrace{1,\ldots, 1}_{k},\underbrace{0,\ldots, 0}_{l}),
1158: $$
1159: $$
1160: B':=\diag(\underbrace{0,\ldots, 0}_{k},\underbrace{d,\ldots, d}_{l}),
1161: $$
1162: for some $d \in \C^*$. 
1163: If $k\ne l$, then set $e:=d^{\frac{1}{l-k}}$ and let
1164: $$
1165: g':=\diag(\underbrace{e^l,\ldots, e^l}_{k},\underbrace{e^{-k},\ldots, e^{-k}}_{l}).
1166: $$
1167: Clearly $\det (g')=1$ and:
1168: $$
1169: (g', \Id_m) \cdot [A',B'] = 
1170: \left[
1171: \diag(\underbrace{e^l,\ldots, e^l}_{k},\underbrace{0,\ldots,0}_{l}),
1172: \diag(\underbrace{0,\ldots, 0}_{k},\underbrace{d e^k,\ldots,d e^k}_{l})
1173: \right]
1174: $$
1175: where
1176: $$
1177: d e^k = d^{1+\frac{k}{l-k}} = d^{\frac{l}{l-k}} = e^l.
1178: $$
1179: So rescaling we get:
1180: $$
1181: (g', \Id_m) \cdot [A',B'] = 
1182: \left[
1183: \diag(\underbrace{1,\ldots, 1}_{k},\underbrace{0,\ldots,0}_{l}),
1184: \diag(\underbrace{0,\ldots, 0}_{k},\underbrace{1,\ldots,1}_{l})
1185: \right]
1186: $$
1187: and this finishes the proof of (ii).
1188: 
1189: 
1190: 
1191: \medskip
1192: 
1193: 
1194: For part (iii),
1195: denote by $W_1$ (respectively, $W_2$) the standard representation of 
1196: the first (respectively, the second) component of $G= \Sl_m \times \Sl_m$.
1197: Then our representation $V$ is isomorphic to $(W_1 \otimes W_2) \oplus (W_1^*\otimes W_2^*)$. 
1198: For $m\ge 3$ the representation $W_i$ is not isomorphic to $W_i^*$ and
1199: therefore $V$ is a union of two irreducible non-isomorphic
1200: representations, so there are exactly two closed orbits of this
1201: action on $\P(V)$.  These orbits are simply $\cDeg^m_{1,0}$
1202: and $\cDeg^m_{0,1}$.
1203: \end{prf}
1204: 
1205: 
1206: 
1207: 
1208: \subsection{Action of $\widetilde{G}$}
1209: 
1210: Recall the notation of 
1211: \S\ref{notation_V},
1212: \S\ref{notation_G} and
1213: \S\ref{notation_Inv_Deg}.
1214: 
1215: 
1216: The action of $\widetilde{G}$ extends the action of $G$, but it does
1217: not preserve $\Xinv(m)$. 
1218: So we will only consider the action of $\widetilde{G}$ when speaking of
1219: $\Xdeg(m,k)$.
1220: 
1221: We have properties analogous to proposition
1222: \ref{degenerate_orbits_of_G} (ii) and  (iii) but with no exceptional cases:
1223: 
1224: \begin{prop}
1225: \label{properties_of_Gtilde}
1226: \hfill
1227: \begin{itemize}
1228: \item[(i)]
1229: Every $\cDeg^m_{k,l}$ is an orbit of the action of $\widetilde{G}$.
1230: \item[(ii)]
1231: For every  $m$ there are exactly two closed orbits of the action of
1232: $\widetilde{G}$: $\cDeg^m_{1,0}$ and $\cDeg^m_{0,1}$.
1233: \end{itemize}
1234: \end{prop}
1235: 
1236: \begin{prf}
1237: It goes exactly as the proof of proposition
1238: \ref{degenerate_orbits_of_G} (ii) and (iii).
1239: \end{prf}
1240: 
1241: 
1242: 
1243: \section{Legendrian varieties in $Y$}\label{main_part}
1244: 
1245: In this section we prove the main results of the article.
1246: 
1247: \subsection{Classification}\label{section_classification}
1248: 
1249: We start with proving the theorem \ref{classification_theorem}. 
1250: For this we use our notation of section \ref{section_basics}.
1251: 
1252: 
1253: 
1254: \begin{prf}
1255: First assume $X$ is Legendrian and contained in $Y$.
1256: If $X$ contains a point $[A,B]$ where both
1257:  $A$ and $B$ are invertible,  then by proposition
1258: \ref{proposition_action_of_G}
1259:  it must contain the orbit of $[A,B]$, which by proposition 
1260: \ref{proposition_orbits_of_G}(ii) and (iii) is equal to 
1261: $\psi_{\mu}(\Inv^m)$ for some $\mu \in \C^*$. But dimension of $X$ is $m^2-1$ which is
1262:  exactly the dimension of $\psi_{\mu}(\Inv^m)$ (see proposition
1263:  \ref{proposition_orbits_of_G}(iii)), 
1264: so
1265: $$
1266: X =\overline{\psi_{\mu}(\Inv^m)} = \psi_{\mu}(\Xinv(m)).
1267: $$
1268: 
1269: \medskip
1270: 
1271: On the other hand if $X$ does not contain any point $[A,B]$ where both $A$ and $B$ are invertible then 
1272: in fact $X$ is contained in the locus $Y_0:=\{[A,B]:AB^T=B^T A=0\}$. 
1273: This locus is just the union of all $\cDeg^m_{k,l}$ and its irreducible components are
1274: the closures of $\cDeg^m_{k,m-k}$, which are exactly $\Xdeg(m,k)$.
1275:  So in particular every irreducible component has
1276:  dimension $m^2-1$ (see proposition \ref{degenerate_orbits_of_G}(i)) 
1277: and hence $X$ must be one of these components.
1278: 
1279: \medskip
1280: 
1281: Therefore it remains to show that all these varieties are Legendrian. 
1282: 
1283: The fact that $\Xdeg(m,k)$ is a Legendrian variety follows from
1284: \cite[pp524--525]{strickland}. Strickland proves there that the affine
1285: cone over $\Xdeg(m,k)$ (or $W(k, m-k)$ in the notation of
1286: \cite{strickland}) is the closure of a conormal bundle. Conormal bundles
1287: are classical examples of Lagrangian varieties.
1288: 
1289: \medskip
1290: 
1291: Since $\psi_{\mu}$ preserves the symplectic form $\omega$, it is
1292: enough to prove that $\Xinv(m)$ is Legendrian.
1293: 
1294: 
1295: 
1296: 
1297: 
1298: The group $G$ acts symplectically on $V$ and the action has an open
1299: orbit on $\Xinv(m)$ --- see proposition \ref{proposition_orbits_of_G} (iii).
1300: Thus the tangent spaces to the affine cone over $\Xinv(m)$ are Lagrangian if
1301:  and only if just one tangent space at a point of the open orbit is
1302:  Lagrangian.
1303: 
1304: So we take $[A,B]:=[\Id_m, \Id_m]$.
1305: Now the affine tangent space to $\Xinv(m)$ at $[\Id_m,\Id_m]$ 
1306: is the linear subspace of $V$ spanned by $(\Id_m,\Id_m)$ 
1307: and the image of the tangent action of the Lie algebra $\gotg$. 
1308: We must prove that for every four traceless matrices
1309: $g,h, g', h'$ we have:
1310: \begin{equation}
1311: \label{equation_for_Inv_Legendrian1}
1312: \omega \left((g,h) \cdot (\Id_m,\Id_m), \ (g',h') \cdot
1313: (\Id_m,\Id_m)\right) = 0 \textrm{ and}
1314: \end{equation}
1315: \begin{equation}
1316: \label{equation_for_Inv_Legendrian2}
1317: \omega \left((\Id_m,\Id_m),  \ (g,h)\cdot (\Id_m,\Id_m)\right) = 0
1318: \end{equation}
1319: 
1320: Equality \eqref{equation_for_Inv_Legendrian1} is true without the
1321: assumption that the matrices have trace $0$:
1322: $$
1323: \omega \big((g,h) \cdot (\Id_m,\Id_m),  \ (g',h') \cdot
1324: (\Id_m,\Id_m)\big) \ = 
1325: $$\nopagebreak
1326: $$
1327: = \ \omega\Big( \left( g^T+h, \ -(g+h^T)\right), \quad
1328: \left((g')^T+h', \ -(g' +(h')^T)\right)\Big) \ =
1329: $$\nopagebreak
1330: $$
1331: \stackrel{\textrm{by \eqref{properties_of_inner_product}}}{=}
1332:  \   \tr \Big( -\left(g^T+h\right) \left((g')^T + h'\right)   \  +
1333:  \  \left(g+h^T\right) \left( g'+(h')^T\right)   \Big)  \ =
1334: $$\nopagebreak
1335: $$
1336: = \ 0.
1337: $$
1338: 
1339: For equality \eqref{equation_for_Inv_Legendrian2} we calculate:
1340: $$
1341: \omega \big((\Id_m,\Id_m),  \ (g,h) \cdot (\Id_m,\Id_m)\big) \ = 
1342: $$\nopagebreak
1343: $$
1344: = \ \omega\Big((\Id_m,\Id_m) , \quad
1345: \left(g^T+h, \ -(g +h^T)\right)\Big) \ =
1346: $$\nopagebreak
1347: $$
1348: \stackrel{\textrm{by \eqref{properties_of_inner_product}}}{=}
1349: - \tr (g^T+h)  - \tr (g+h^T) \ = \ 0.
1350: $$
1351: 
1352: Hence we have proved that the closure of $\Inv^m$ is Legendrian.
1353: \end{prf}
1354: 
1355: 
1356: 
1357: \subsection{Degenerate matrices}\label{section_degenerate}
1358: 
1359: Recall our notation of 
1360: \S\ref{notation_V}, 
1361: \S\ref{notation_Xinv_Xdeg},
1362: \S\ref{notation_G},
1363: \S\ref{notation_p1_p2} and
1364: \S\ref{properties_of_tangent_cone}.
1365: 
1366: 
1367: By \cite[prop.~1.3]{strickland} the ideal of $\Xdeg(m,k)$ is generated by
1368: the coefficients of $AB^T$, the coefficients of $B^TA$, 
1369: the $(k+1)\times(k+1)$-minors of $A$ and the $(m-k+1)\times(m-k+1)$-minors of $B$.
1370: In short we will say that the equations of $\Xdeg(m,k)$ are given by:
1371: 
1372: \begin{equation}\label{equations_of_Xdeg}
1373: AB^T=0, \ B^TA=0, \quad \rk(A) \le k, \ \rk(B) \le m-k.
1374: \end{equation}
1375: 
1376: 
1377: 
1378: \begin{lem}
1379: Assume $m\ge 2$ and $1\le k \le m-1$. Then:
1380: \begin{itemize}
1381: \item[(i)]
1382: The tangent cone to $\Xdeg(m,k)$ at $p_1$ is a product of a linear space of dimension 
1383: $(2m-2)$ and the affine cone over $\Xdeg(m-1,k-1)$.
1384: \item[(i')]
1385: The tangent cone to $\Xdeg(m,k)$ at $p_2$ is a product of a linear space of dimension 
1386: $(2m-2)$ and the affine cone $\Xdeg(m-1,k)$.
1387: \item[(ii)]
1388: $\Xdeg(m,k)$ is smooth at $p_1$ if and only if $k=1$.
1389: \item[(ii')]
1390: $\Xdeg(m,k)$ is smooth at $p_2$ if and only if $k=m-1$.
1391: \end{itemize}
1392: \end{lem}
1393: 
1394: 
1395: \begin{prf}
1396: We only prove (i) and (ii), while (i') and (ii') follow in the same way by exchanging $a_{ij}$ and $b_{ij}$.
1397: Consider equations \eqref{equations_of_Xdeg} of $\Xdeg(m,k)$ 
1398: restricted to the affine neighbourhood of $p_1$ obtained by substituting $a_{mm}=1$. 
1399: Taking the lowest degree part of these equations we get some of the equations of the tangent cone at $p_1$
1400: (recall our convention on the notation of submatrices  --- see 
1401: \S\ref{notation_submatrices}):
1402: $$
1403: b_{im}=b_{mi}=0, \ A_m B_m^T =0, \ B_m^T A_m = 0, 
1404: $$
1405: $$ 
1406: \rk A_m \le k-1, \rk B_m \le m-k.
1407: $$
1408: These equations define the product of the linear subspace  $A_m=B_m=0, b_{im}=b_{mi}=0$ 
1409: and the affine cone over $\Xdeg(m-1,k-1)$ embedded in the set of those pairs of matrices,
1410:  whose last row and column are zero: $a_{im}=a_{mi}=0, b_{im}=b_{mi}=0$. 
1411: So the variety defined by those equations is irreducible and its dimension is equal to
1412: $(m-1)^2 + 2m-2 = m^2-1 = \dim \Xdeg(m,k)$. 
1413: Since it contains the tangent cone we are interested in
1414:  and by \S\ref{properties_of_tangent_cone}(1),
1415: they must coincide as claimed in (i).
1416: 
1417: 
1418: \medskip
1419: 
1420: Next (ii) follows immediately, since for $k=1$ the equations above reduce to 
1421: $$
1422: b_{im}=b_{mi}=0, \ \textrm{ and } A_m=0
1423: $$
1424: and hence the tangent cone is just the tangent space, so  $p_1$ is a smooth point of $\Xdeg(m,1)$.
1425: Conversely,  if $k>1$ then $\Xdeg(m-1, k-1)$ is not a linear space, so by (i) 
1426: the tangent cone is not a linear space either and $X$ is singular at $p_1$ --- see 
1427: \S\ref{properties_of_tangent_cone}(3).
1428: \end{prf}
1429: 
1430: 
1431: Now we can prove theorem \ref{theorem_smooth_degenerate}:
1432: 
1433: 
1434: \begin{prf}
1435: It is obvious from the definition of $\Xdeg(m,k)$, 
1436: that $\Xdeg(m,0)= \{A=0\}$ and $\Xdeg(m,m) = \{B=0\}$,
1437: so these are indeed linear spaces.
1438: 
1439: Therefore assume $1\le k \le m-1$. 
1440: But $\Xdeg(m,k)$ is
1441: $\widetilde{G}$ invariant (see proposition \ref{properties_of_Gtilde}(i)) and so is its singular locus $S$. 
1442: Hence $\Xdeg(m,k)$ is singular if and only if $S$ contains a closed orbit of $\widetilde{G}$. 
1443: 
1444: 
1445: 
1446: So $\Xdeg(m,k)$ is smooth, 
1447: if and only if it is smooth at both $p_1$ and $p_2$
1448: (see proposition \ref{properties_of_Gtilde}(ii)), which 
1449: (by lemma (ii) and (ii')) holds if and only if $k=1$ and $m=2$.
1450: 
1451: \smallskip
1452:  
1453: To finish the proof, 
1454: it remains to verify what kind of variety is $\Xdeg(2,1)$.
1455: Consider the following map:
1456: $$
1457: \P^1 \times \P^1 \times \P^1 \lra \P(V) \simeq \P^7
1458: $$
1459: $$
1460: [\mu_1, \mu_2],[\nu_1,\nu_2],[\xi_1,\xi_2] \longmapsto
1461: \left[
1462: \xi_1
1463: \left(
1464: \begin{array}{cc}
1465: \mu_1\nu_1 & \mu_1 \nu_2\\
1466: \mu_2\nu_1 & \mu_2 \nu_2
1467: \end{array}
1468: \right),
1469: \xi_2
1470: \left(
1471: \begin{array}{cc}
1472: \mu_2\nu_2 & -\mu_2 \nu_1\\
1473: -\mu_1\nu_2 & \mu_1 \nu_1
1474: \end{array}
1475: \right)
1476: \right]
1477: $$
1478: Clearly this is a Segre embedding in appropriate coordinates.
1479: The image of this embedding is contained in $\Xdeg(2,1)$ 
1480: (see equation \eqref{equations_of_Xdeg}) 
1481: and since dimension of $\Xdeg(2,1)$ 
1482: is equal to the dimension of  $\P^1 \times \P^1 \times \P^1$ 
1483: we conclude the above map gives an isomorphism of 
1484: $\Xdeg(2,1)$ and  $\P^1 \times \P^1 \times \P^1$.
1485: 
1486: 
1487: %It can be easily and explicitely seen, 
1488: %that it is $\P^1 \times \P^1 \times \P^1$, 
1489: %but let us just observe the following properties:
1490: %Some quadratic equations generate the ideal of $\Xdeg(2,1)$ 
1491: %(see \eqref{equations_of_Xdeg}), 
1492: %$\Xdeg(2,1)$ is Legendrian by theorem \ref{classification_theorem}
1493: %and it is smooth irreducible of dimension 3 
1494: %(see proposition \ref{degenerate_orbits_of_G}(i)),
1495: %so by \cite[thm 5.11]{jabu06} it must be exactly the product of $\P^1$'s.
1496: \end{prf}
1497: 
1498: 
1499: 
1500: 
1501: 
1502: 
1503: \subsection{Invertible matrices}\label{section_invertible}
1504: 
1505: Recall the notation of 
1506: \S\ref{notation_V},
1507: \S\ref{notation_Xinv_Xdeg},
1508: \S\ref{notation_G},
1509: \S\ref{notation_Inv_Deg} and
1510: \S\ref{properties_of_tangent_cone}.
1511: 
1512: 
1513: We wish to determine some of the equations of $\Xinv(m)$.
1514: Clearly the equations of $Y$ (see \eqref{equations_of_Y}) 
1515: are quadratic equations of $\Xinv(m)$.
1516: To find other equations, we recall, that 
1517: $$
1518: \Xinv(m):=\overline{
1519: \bigg\{\left[g,\left(g^{-1}\right)^T\right] \in \P(V) \mid \det g = 1 \bigg\}}
1520: $$
1521: But for a matrix $g$ with determinant 1 we know that the entries of
1522: $(g^{-1})^T$
1523: consist of the appropriate minors (up to sign)
1524: of $g$.
1525: Therefore we get many inhomogeneous equations satisfied by every pair
1526: $\left(g, (g^{-1})^T\right) \in V$
1527: (recall our convention on the notation of submatrices  --- see 
1528: \S\ref{notation_submatrices}):
1529: $$
1530: \det (A_{ij}) = (-1)^{i+j} b_{ij} \  \textrm{ and } \  a_{kl} = (-1)^{k+l} \det(B_{kl})  
1531: $$
1532: 
1533: To make them homogeneous, multiply two such equations appropriately:
1534: \begin{equation}\label{minors_1_of_X}
1535: \det(A_{ij}) a_{kl} = (-1)^{i+j+k+l} b_{ij} \det(B_{kl}).
1536: \end{equation}
1537: These are degree $m$ equations, 
1538: which are satisfied by the points of $\Xinv(m)$ 
1539: and we state the following theorem:
1540: 
1541: \begin{theo}
1542: \label{theorem_Xinv3_is_smooth}
1543: Let $m=3$. Then the quadratic equations \eqref{equations_of_Y1}--\eqref{equations_of_Y4} and
1544: the cubic equations \eqref{minors_1_of_X} generate the ideal of 
1545: $\Xinv(3)$. Moreover $\Xinv(3)$ is smooth.
1546: \end{theo}
1547: 
1548: \begin{prf}
1549: It is enough to prove that the scheme $X$ defined by equations
1550: \eqref{equations_of_Y1}--\eqref{equations_of_Y4} 
1551: and  \eqref{minors_1_of_X} is smooth, because the  reduced subscheme
1552: of $X$ coincides with $\Xinv(3)$.
1553: 
1554: 
1555: The scheme $X$ is $G$ invariant, 
1556: hence as in the proof of theorem \ref{theorem_smooth_degenerate} and by
1557: proposition \ref{degenerate_orbits_of_G}(iii) it is
1558:  enough to verify smoothness 
1559: at $p_1$ and $p_2$.
1560: Since we have the additional symmetry here (exchanging $a_{ij}$'s with $b_{ij}$'s)
1561: it is enough to verify the smoothness at $p_1$.
1562: 
1563: Now we calculate the tangent space to $X$ at $p_1$ by taking 
1564: linear parts of the equations evaluated at $a_{33}=1$. 
1565: From \eqref{equations_of_Y} we get that 
1566: $$
1567: b_{31}=b_{32}=b_{33}=b_{23}=b_{13}=0.
1568: $$
1569: Now from equations \eqref{minors_1_of_X} for $k=l=3$ and $i,j \ne 3$  we get the following evaluated equations:
1570: $$
1571: a_{i'j'} - a_{i'3}a_{3j'} = \pm b_{ij} B_{33} 
1572: $$
1573: (where $i'$ is either $1$ or $2$, which ever is different than $i$ and analogously for $j'$)
1574: so the linear part is just $a_{i'j'}=0$. Hence by varying $i$ and $j$ we can get 
1575: $$
1576: a_{11}= a_{12}=a_{21}=a_{22}=0.
1577: $$
1578: Therefore the tangent space has codimension at least $9$, which is exactly the codimension of 
1579: $\Xinv(3)$ --- see \ref{proposition_orbits_of_G}(iii). Hence $X$ is
1580: smooth (in particular reduced) and $X=\Xinv(3)$. 
1581: \end{prf}
1582: 
1583: To describe $\Xinv(m)$ for $m > 3$ we must find more equations. 
1584: 
1585: There is a more general version of the above  property of an inverse
1586: of a matrix with determinant 1, which is less popular.
1587: 
1588: 
1589: 
1590: 
1591: 
1592: 
1593: \begin{prop}\label{proposition_minors_2}
1594: \hfill
1595: \begin{itemize}
1596: \item[(i)]
1597: Assume $A$ is a $m \times m$ matrix of determinant $1$ and $I,J$ are
1598: two sets of indices, both of cardinality $k$
1599: (again recall our convention on indices and submatrices  --- see
1600: section \ref{notation_submatrices}). Denote by $B:=(A^{-1})^T$. 
1601: Then the appropriate minors are equal (up to sign):
1602: $$
1603: \det A_{I,J} = (-1)^{\Sigma I + \Sigma J} \det B_{I', J'}.
1604: $$
1605: \item[(ii)]
1606: The coordinate free way to express these equalities is following: 
1607: if  $W$ is a vector space of dimension $m$ 
1608: and $f$ is a linear automorphism of $W$,
1609: let $\Wedge{k} f$ be the induced automorphism of $\Wedge{k} W$.
1610: If $\Wedge{m} f = \Id_{\Wedge{m} W}$  then:
1611: $$
1612: \Wedge{m-k} f  = \Wedge{k} \left( \Wedge{m-1} f\right) .
1613: $$
1614: \item[(iii)]
1615: Consider the induced action of $G$ on the polynomials on $V$.
1616: Then the vector space spanned by the set of equations of (i) for a fixed
1617: 	  $k$ is $G$ invariant.
1618: \end{itemize}
1619: \end{prop}
1620: 
1621: \begin{prf}
1622: Part (ii) follows explicitly from (i), since 
1623: if $A$ is a matrix of $f$, then the terms of  the matrices of  the maps 
1624: $\Wedge{m-k} f$ and $\Wedge{k} ( \Wedge{m-1} f)$
1625:  are exactly the appropriate minors of $A$ and $B$.
1626: 
1627: Part (iii) follows easily from (ii).
1628: 
1629: As for (i), 
1630: %compare with a slightly similar statement in \cite[\S II.4.3]{sikorski}.
1631: %?? what is a slightly similar statement???
1632: %
1633: we only sketch the proof, leaving the details to the reader and his or her linear algebra students.
1634: Firstly, reduce to the case when $I$ and $J$ are just $\{1, \ldots k\}$ and the determinant of $A$
1635: is possibly $\pm 1$ (which is where the sign shows up in the equality).
1636: Secondly if both determinants $\det A_{I,J}$ and $\det B_{I', J'}$ are 
1637: zero, then the equality is clearly satisfied. Otherwise assume for example $\det A_{I,J} \ne 0$.
1638: Then performing the appropriate row and column operations we can change $A_{I,J}$ into a diagonal matrix, 
1639: $A_{I', J}$ and $A_{I, J'}$ into the zero matrices 
1640: and all these operations can be done without changing $B_{I',J'}$ 
1641: nor $\det A_{I,J}$. 
1642: Then the statement follows easily.
1643: \end{prf}
1644: 
1645: 
1646: 
1647: In particular we get:
1648: 
1649: \begin{cor}
1650: \label{corollary_equations_Xinv}
1651: Assume $k$, $I$ and $J$ are as in proposition \ref{proposition_minors_2}(i).
1652: \begin{itemize}
1653: \item[(a)]
1654: If $m$ is even and $k=\half m$, then the equation
1655: $$
1656: \det A_{I,J} = (-1)^{\Sigma I + \Sigma J} \det B_{I', J'}
1657: $$
1658: is homogeneous of degree $\half m$ and it is satisfied by points of $\Xinv(m)$.
1659: \item[(b)]
1660: If $0 \le k < \half m $ and $l= \half m - k$, then
1661: $$
1662: \left(\det A_{I,J}\right)^2 = \left(\det B_{I',J'}\right)^2 \cdot (a_{11}b_{11} + \ldots  a_{1m}b_{1m})^l
1663: $$
1664: is a homogeneous equation of degree $2(m-k)$ satisfied by points of $\Xinv(m)$.
1665: \end{itemize}
1666: \end{cor}
1667: 
1668: \begin{prf}
1669: Clearly both equations are homogeneous.
1670: If $\det A =1$ and $B=(A^{-1})^T$ then the following equations are satisfied: 
1671: \begin{equation}\label{equation_for_cor1}
1672: \det A_{I,J} = (-1)^{\Sigma I + \Sigma J} \det B_{I',J'},
1673: \end{equation}
1674: \begin{equation}\label{equation_for_cor2}
1675: 1 = (a_{11}b_{11} + \ldots  a_{1m}b_{1m})^l 
1676: \end{equation}
1677: (equation \eqref{equation_for_cor1} follows from proposition \ref{proposition_minors_2}(i)
1678: and \eqref{equation_for_cor2} follows from $AB^T=\Id_m$).
1679: Equation in (b) is just \eqref{equation_for_cor1} squared multiplied side-wise by \eqref{equation_for_cor2}.
1680: 
1681: So both equations in (a) and (b) are satisfied by every pair 
1682: $\left( A, (A^{-1})^T\right)$ 
1683: and by homogeneity also by
1684: $\left(\lambda A, \lambda (A^{-1})^T \right)$. 
1685: Hence (a) and (b) hold on an open dense subset of $\Xinv(m)$, so also on whole $\Xinv(m)$.
1686: \end{prf}
1687: 
1688: 
1689: 
1690: 
1691: We know enough equations of $\Xinv(m)$ to prove the theorem \ref{theorem_classify_invertible}:
1692: 
1693: 
1694: 
1695: \subsubsection{Case $m=2$ --- linear subspace}
1696: 
1697: \begin{prf}
1698: To prove (a) just take the linear equations from  proposition \ref{proposition_minors_2}(i) for $k=1$:
1699: $$
1700: a_{ij} = \pm b_{i'j'}
1701: $$
1702: where $\{i,i'\} = \{j,j'\} = \{1,2\}$.
1703: \end{prf}
1704: 
1705: 
1706: 
1707: \subsubsection{Case $m=3$ --- hyperplane section of $Gr(3,6)$}
1708: 
1709: \begin{prf}
1710: For (b), $\Xinv(3)$ is smooth by theorem \ref{theorem_Xinv3_is_smooth} 
1711: and it is a compactification of $\Inv^3 \simeq \Sl_3$ by proposition
1712: \ref{proposition_orbits_of_G}(i) and (iii).
1713: 
1714: 
1715: \paragraph{Picard group of $\Xinv(3)$.}
1716: 
1717: The complement of the open orbit 
1718: $$
1719: D:=\Xinv(3) \backslash \Inv^3
1720: $$
1721: must be a union of some orbits of $G$, each of them must have
1722: dimension smaller than $\dim \Inv^3 =8$. So by propositions
1723: \ref{proposition_orbits_of_G}(ii), (iii), \ref{degenerate_orbits_of_G}
1724: (i) and (ii) the only candidates are $\cDeg^3_{1,1}$, $\cDeg^3_{0,1}$
1725: and $\cDeg^3_{1,0}$.
1726: We claim they are all contained in $\Xinv(3)$. It is enough to prove
1727: that $\cDeg^3_{1,1} \subset \Xinv(3)$, since the other orbits are in
1728: the closure of $\cDeg^3_{1,1}$. Take the curve in $\Xinv(3)$
1729: parametrised by:
1730: $$
1731: \left[
1732: \left(
1733: \begin{array}{ccc}
1734: t  &0  &0\\
1735: 0  &1  &0\\
1736: 0  &0  &t^{-1}\\
1737: \end{array}
1738: \right),
1739: \left(
1740: \begin{array}{ccc}
1741: t^{-1} &0  &0\\
1742: 0      &1  &0\\
1743: 0      &0  &t\\
1744: \end{array}
1745: \right)
1746: \right].
1747: $$
1748: For $t=0$ the curve meets $\cDeg^3_{1,1}$, which finishes the proof of
1749: the claim.
1750: 
1751: 
1752: Since $\dim \cDeg^3_{1,1} = 7$  (see proposition
1753: \ref{degenerate_orbits_of_G}(i)), $D$ is a prime divisor.
1754: We have $\Pic(\Sl_3) = 0$ and by \cite[prop. II.6.5(c)]{hartshorne} the Picard group of 
1755:  $\Xinv(3)$ is isomorphic to $\Z$ with the ample generator $[D]$.
1756: 
1757: Next we check that $D$ is linearly equivalent 
1758: (as a divisor on $\Xinv(3)$) to a hyperplane section $H$ of $\Xinv(3)$. 
1759: Since we already know that $\Pic(\Xinv(3))=\Z \cdot [D]$, we must have
1760: $H \stackrel{lin}{\sim} k D$ for some positive integer $k$.
1761: But there are lines contained in $\Xinv(3)$ (for example those
1762: contained in $\cDeg^3_{1,0} \simeq \P^2 \times \P^2$)\footnote{
1763: Actually, the reader could also easily find explicitly some lines (or
1764: even planes) which intersect the open orbit 
1765: and conclude that $\Xinv(3)$ is covered by lines.
1766: }.
1767: So let $L\subset \Xinv(3)$ be any line and we intersect:
1768: $$
1769: D\cdot L = \frac{1}{k} H \cdot L = \frac{1}{k}.
1770: $$
1771: But the result must be an integer, so $k=1$ as claimed.
1772: 
1773: 
1774: \paragraph{Complete embedding.}
1775: 
1776: 
1777: Since $D$ itself is definitely not a hyperplane section of $\Xinv(3)$, 
1778: the conclusion is that the Legendrian embedding of $\Xinv(3)$ is not given by a complete linear system.
1779: The natural guess for a better embedding is the following:
1780: %
1781: %$$
1782: %\Xinv(m):=\overline{
1783: %\bigg\{\left[g,\left(g^{-1}\right)^T\right] \in \P(V) \mid \det g = 1 \bigg\}}
1784: %$$
1785: %
1786: $$
1787: X' := \overline{
1788: \bigg\{\left[1,g,\Wedge{2} g\right] \in \P^{18} = \P(\C \oplus V)
1789: \mid \det g = 1 \bigg\}}.
1790: $$
1791: (we note that $\Wedge{2} g = (g^{-1})^T$ for $g$ with $\det g=1$)
1792: and one can verify that the projection from the point $[1,0,0]\in \P^{18}$
1793: restricted to $X'$ gives an isomorphism with $\Xinv(3)$.
1794: 
1795: The Grassmannian $Gr(3,6)$ in its Pl\"u{}cker embedding can be described as the closure of:
1796: $$
1797: \bigg\{\left[1,g,\Wedge{2} g, \Wedge{3} g\right] \in \P^{19} = \P(\C \oplus V \oplus \C)
1798: \mid g \in M_{3 \times 3}\bigg\}
1799: $$
1800: and we immediately identify $X'$ as the section $H:=\left\{ \Wedge{3} g=1 \right\}$ of the Grassmannian. 
1801: 
1802: Though it is not essential, we note that $H^1(\ccO_{Gr(3,6)}) = 0$ 
1803: (see Kodaira vanishing theorem \cite[thm 4.2.1]{lazarsfeld}) and hence the above embedding of $\Xinv(3)$ 
1804: is given by the complete linear system.
1805: 
1806: 
1807: \paragraph{Automorphism group.}
1808: 
1809: It remains to calculate 
1810: $Aut\left(\Xinv(3)\right)^0$ 
1811: --- the connected component of the automorphism group.
1812: 
1813: The tangent Lie algebra of the group of automorphisms of a complex projective 
1814: manifold is equal to the global sections of the tangent bundle, see \cite{akhiezer}.
1815: A vector field on $\Xinv(3)$ is also a section of $T Gr(3,6)|_{\Xinv(3)}$ 
1816: and we have the following short exact sequence:
1817: $$
1818: 0 \lra T Gr(3,6)(-1) \lra T Gr(3,6) \lra T Gr(3,6)|_{\Xinv(3)} \lra 0
1819: $$
1820: The homogeneous vector bundle $T Gr(3,6)(-1)$ is isomorphic to 
1821: $U^* \otimes Q \otimes \Wedge{3} U$, where $U$ is the universal subbundle in  $Gr(3,6) \times \C^6$ 
1822: and $Q$ is the universal quotient bundle.
1823: This bundle corresponds to an irreducible module of the parabolic subgroup in $\Sl_6$.
1824: Calculating explicitly its highest weight and applying Bott formula 
1825: \cite{ottaviani} we get that $H^1\big(T Gr(3,6)(-1)\big) = 0$. 
1826: Hence every section of $T\Xinv(3)$ extends to a section of $TGr(3,6)$.
1827: In other words, if $P < Aut(Gr(3,6)) \simeq \P\Gl_6$ 
1828: is the subgroup preserving $\Xinv(3) \subset Gr(3,6)$,
1829: then the restriction map $P \lra Aut\left(\Xinv(3)\right)^0$ is epimorphic.
1830: 
1831: The action of $\Sl_6$ on $\Wedge{3} \C^6$ preserves 
1832: the natural symplectic form $\omega'$:
1833: $$
1834: \omega' : \Wedge{2}\left(\Wedge{3} \C^6\right) \lra  \Wedge{6} \C^6 \simeq \C.
1835: $$
1836: Since the action of $P$ on $\P\left(\Wedge{3} \C^6\right)$ 
1837: preserves the hyperplane $H$ containing $\Xinv(3)$, 
1838: it must also preserve $H^{\perp_{\omega'}}$, 
1839: i.e.~$P$ preserves $[1,0,0,1]\in \P^{19} = \P(\C\oplus V \oplus \C)$.
1840: Therefore $P$ acts on the quotient $H/(H^{\perp_{\omega'}}) = V$ 
1841: and hence the restriction map factorises:
1842: $$
1843: P \lra Aut(\P(V), \Xinv(3))^0 \epi Aut(\Xinv(3))^0.
1844: $$
1845: 
1846: By \cite{jabu_toric}, group $Aut(\P(V), \Xinv(3))^0$ 
1847: is contained in the image of $\Sp(V) \lra \P\Gl(V)$,
1848: so by theorem \ref{theorem_ideal_and_group},
1849: proposition \ref{proposition_action_of_G}
1850: and theorem \ref{theorem_Xinv3_is_smooth}
1851: \[
1852: Aut\left(\P(V), \Xinv(3)\right)^0= G.
1853: \]
1854: In particular $\Xinv(3)$ cannot be homogeneous as it contains 
1855: more than one orbit of the connected component of automorphism group.
1856: 
1857: \end{prf}
1858: 
1859: We note that the  fact that $\Xinv(3)$ is not homogeneous 
1860: can be also proved without calculating the automorphism group.
1861: Since $\Pic \Xinv(3) \simeq \Z$, 
1862: it follows from \cite[thm.~11]{landsbergmanivel04},
1863: that $\Xinv(3)$ could only be one of the subadjoint varieties.
1864: But none of them has $\Pic \simeq \Z$ and dimension 8.
1865: 
1866: 
1867: 
1868: 
1869: 
1870: 
1871: 
1872: 
1873: 
1874: 
1875: \subsubsection{Case m=4 --- spinor variety $\mathbb{S}_6$}
1876: 
1877: \begin{prf}
1878: To prove (c) we only need to take 30 quadratic equations of $Y$ as in
1879: \eqref{equations_of_Y} and 36 quadratic 
1880: equations from  corollary \ref{corollary_equations_Xinv} (a).
1881: By proposition \ref{proposition_minors_2}(iii)  
1882: the scheme $X$ defined by those quadratic equations is $G$ invariant.  
1883: As in the proofs of theorems  \ref{theorem_smooth_degenerate} and \ref{theorem_Xinv3_is_smooth},
1884: we only check that $X$ is smooth at $p_1$ and $p_2$ and 
1885: conclude it is smooth everywhere, hence those equations indeed define $\Xinv(4)$.
1886:  
1887: Therefore $\Xinv(4)$ is smooth, irreducible and its ideal is  generated by quadrics, 
1888: so  it falls into the classification of \cite[thm. 5.11]{jabu06}. 
1889: Hence we have two choices for $\Xinv(4)$, whose dimension is $15$:
1890: the product of a line and a quadric $\P^1\times Q_{14}$ or the spinor variety $\mathbb{S}_6$.
1891: The homogeneous ideal of polynomials vanishing on $\P^1\times Q_{14}\subset \P^{31}$ 
1892: is generated by $\dim (\Sl_2\times \mathbf{SO}_{16})=123$ 
1893: linearly independent quadratic polynomials 
1894: (see theorem \ref{theorem_ideal_and_group},
1895: alternatively, one can calculate the equations explicitly --- see \cite[\S7.2]{jabu_arxiv} ). 
1896: So $\Xinv(4)$, which by the above argument is generated by only 66 quadratic equations,
1897: must be isomorphic to $\mathbb{S}_6$.
1898: \end{prf}
1899: 
1900: 
1901: \subsubsection{Case m$\ge 5$ --- singular varieties} 
1902: 
1903: \begin{prf}
1904: Finally we prove (d).
1905: We want to prove, that for $m\ge 5$ variety $\Xinv(m)$ is singular at $p_1$. 
1906: To do that, we calculate the reduced tangent cone 
1907: $$
1908: T:=\big(TC_{p_1} \Xinv(m)\big)_{red}.
1909: $$ 
1910: From equations \eqref{equations_of_Y} we easily get the following 
1911: linear and quadratic equations of $T$
1912: (again we suggest to have a look at \S\ref{notation_submatrices}):
1913: $$
1914: b_{im} = b_{mi}=0, \quad A_m B_m^T = B_m^T A_m = \lambda^2 \Id_{m-1}
1915: $$
1916: for every $i\in \{1,\ldots m\}$  and some $\lambda\in \C^*$.
1917: 
1918: Next assume $I$ and $J$  are two sets of indices both of cardinality $k = \left\lfloor \half m \right\rfloor$
1919: and such that neither $I$ nor $J$ contains $m$. 
1920: Consider the equation of $\Xinv(m)$ as in corollary \ref{corollary_equations_Xinv}(b):
1921: $$
1922: \left(\det A_{I,J}\right)^2 = \left(\det B_{I',J'}\right)^2 \cdot (a_{11}b_{11} + \ldots  a_{1m}b_{1m})^l.
1923: $$
1924: To get an equation of $T$, we evaluate at $a_{mm}=1$ and take the lowest degree part,
1925: which is simply 
1926: $\left(\det \left((A_m)_{I,J}\right)\right)^2=0$.
1927: Since $T$ is reduced, by varying $I$ and $J$
1928: we get that: 
1929: $$
1930: \rk A_m \le m - 1 - k - 1 = \left\lceil \half m \right\rceil - 2 
1931: $$
1932: and therefore also:
1933: $$
1934:  A_m B_m^T = B_m^T A_m = 0.
1935: $$
1936: 
1937: Hence $T$ is contained in the product of the linear space 
1938: $W:=\{A_m=0, B=0\}$ and the affine cone $\hat{U}$ over the union of $\Xdeg(m-1,k)$ for 
1939: $k\le \left\lceil \half m \right\rceil - 2$. We claim that $T= W \times \hat{U}$.
1940: By proposition \ref{degenerate_orbits_of_G}(i), every component of $W \times \hat{U}$ 
1941: has dimension $2m-2+(m-1)^2 = m^2-1=\dim \Xinv(m)$,
1942: so by \S\ref{properties_of_tangent_cone}(1) the tangent cone must be a union of some of the components.
1943: Therefore to prove the claim it is enough to find for every $k \le \left\lceil \half m \right\rceil - 2$ 
1944: a single element of $\cDeg^{m-1}_{k, m-k-1}$ that is contained in the tangent cone. 
1945: 
1946: So take $\alpha$ and $\beta$ to be two strictly  positive integers such that 
1947: $$
1948: \alpha =  \left( \half m - k - 1 \right) \beta
1949: $$
1950: and consider the curve in $\P(V)$ with the following parametrisation:
1951: $$
1952: \left[
1953: \diag\{
1954: \underbrace{t^{\alpha},\ldots, t^{\alpha}}_{k},
1955: \underbrace{t^{\alpha+\beta},\ldots,t^{\alpha+\beta}}_{m-k-1}, 1
1956: \},
1957: \diag\{
1958: \underbrace{t^{\alpha+\beta},\ldots, t^{\alpha+\beta}}_{k},
1959: \underbrace{t^{\alpha},\ldots,t^{\alpha}}_{m-k-1}, t^{2\alpha + \beta}
1960: \}
1961: \right].
1962: $$
1963: It is easy to verify that this family is contained in $\Inv^m$ for $t\ne 0$
1964: and as $t$ converges to $0$, it gives rise to a tangent vector 
1965: (i.e.~an element of the reduced tangent cone - see point-wise definition in
1966: \S\ref{properties_of_tangent_cone}) that belongs to $\cDeg^{m-1}_{k, m-k-1}$.
1967: 
1968: So indeed $T= W \times \hat{U}$, which for $m\ge 5$ contains more than 1 component, 
1969: hence cannot be a linear space. 
1970: Therefore by \S\ref{properties_of_tangent_cone}(3) variety $\Xinv(m)$ is singular at $p_1$.
1971: \end{prf}
1972: 
1973: 
1974: 
1975: \begin{rem}
1976: Note that in both  cases of $\Xdeg(m,k)$ and $\Xinv(m)$, the reduced tangent cone 
1977: is a Lagrangian subvariety in the fibre of the contact distribution.
1978: This is not accidental as explained in \cite{jabu_tangent_cone}.
1979: \end{rem}
1980: 
1981: 
1982: 
1983: 
1984: 
1985: 
1986: 
1987: 
1988: 
1989: 
1990: 
1991: 
1992: 
1993: 
1994: \bibliography{references}
1995: 
1996: \bibliographystyle{alpha}
1997: 
1998: 
1999: 
2000: 
2001: 
2002: 
2003: \end{document}