ORBIT CLASSIFICATION FOR THE REPRESENTATIONS ASSOCIATED TO GRADED LIE ALGEBRAS

Jerzy Weyman July 14-18, 2010

Content
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1. Root systems and simple Lie algebras 2. Gradings on simple Lie algebras related to simple roots 3. Kacs classication of representations with nitely many orbits. Examples involving tensors and
exterior powers

4. Vinberg method of classifying orbits, statements of the results 5. Vinberg method in practice. Working out of concrete examples 6. Dening ideals and syzygies 7. Applications References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 Introduction The irreducible representations of the reductive groups with nitely many orbits were classied by Kac in [K80]. They were divided into classes II, III and IV. All of these classes are related to gradings of the root systems, and to the corresponding groups. The representations of type II are parametrized by a pair (Xn , k ) where Xn is a Dynkin diagram with a distinguished node x Xn . This data denes a grading g = s i=s gi
The author was partially supported by NSF grant DMS-0600229 Typeset by AMS-TEX

JERZY WEYMAN

of a simple algebra g of type Xn such that the Cartan subalgebra h is contained in g0 and the root space g is contained in gi where i is the coecient of the simple root corresponding to the node x in the expression for as a linear combination of simple roots. The representation corresponding to (Xn , x) is the g1 with the action of the group G0 C where G0 is the adjoint group corresponding to g0 and C is the copy of C that occurs in maximal torus of G (the adjoint group corresponding to g) but not in maximal torus of G0 . The orbit closures for the representations of type II were described in two ways by Vinberg in [V75], [V87]. The rst description states that the orbits are the irreducible components of the intersections of the nilpotent orbits in g with the graded piece g1 . In the second paper Vinberg gave a more precise description in terms of some graded subalgebras of the graded algebra g = gi In these notes we give the introduction to these methods. We use mainly the examples of triple tensor products and third exterior powers. The notes are organized as follows. In section 1 we discuss root systems and simple Lie algebras. Section 2 introduces representations related to gradings on simple Lie algebras related to a choice of a simple root. Section 3 we state Kacs theorem and interpret special cases of triple tensor products and third exterior powers. In section 4 we expose the Vinberg method from [V87] that classes orbits in representations with nitely many orbits. In section 5 we work out some examples. In the following sections we give information about singularities and dening ideals of the orbit closures. We briey describe the geometric method of calculating syzygies and give examples of its use. We give some comments on how one should be able to calculate resolutions of all the coordinate rings of orbit closures.

1. Simple Lie algebras and root systems We recall the Cartan classication of simple Lie algebra and corresponding root systems. For a simple Lie algebra g we have the root decomposition g = h g where is the associated root system. The simple Lie algebra g has also a symmetric bilinear non-degenerate form (, ) : g g C given by the formula (X, Y ) = T r(ad(X)ad(Y )). Denition 1.1. Let V be an Euclidean space over R with a scalar product (, ). A root system is by denition a nite collection of vectors in V such that a) If then the only multiples of in are ,

GEOMETRY OF ORBIT CLOSURES

b) The values of , := 2(, ) (, )

are in Z for , in . c) If and is a reection in the hyperplane orthogonal to then () = . d) spans V . If 1 and 2 are two root systems in the Euclidean spaces V1 , V2 , then 1 2 is a root system in the orthogonal direct sum V1 V2 . A root system is irreducible if it cannot be decomposed into a direct sum of two root systems. In every root systems one can choose a basis = {1 , . . . , l } of simple roots which have the following property. The root system decomposes = + where positive roots can be written as nonnegative linear combinations of vectors in the basis, and negative roots, as non-positive linear combinations. Example 1.2. The root system if type An . V = {(a1 , . . . , an+1 ) Rn+1 | a1 + . . . + an+1 = 0} = {( i ={
1 j)

| 1 i < j n + 1}.
3, . . .

2, 2

n+1 }.

Corresponding Lie algebra is sln+1 . Example 1.3. The root system if type Bn . V = Rn = {( i ={
1 j)

| 1 i < j n, i , 1 i n}.
3, . . .

2, 2

n1

n , n }.

Corresponding Lie algebra is so2n+1 . Example 1.4. The root system if type Cn . V = Rn = {( i ={
1 j)

| 1 i < j n, 2 i , 1 i n}.
3, . . .

2, 2

n1

n , 2 n }.

Corresponding Lie algebra is sp2n .

JERZY WEYMAN

Example 1.5. The root system if type Dn . V = Rn = {( i ={

| 1 i < j n}. ,
n1

2, 2

3, . . .

n , n1

n }.

Corresponding Lie algebra is so2n . Example 1.6. The root system if type E6 . V = Rn = {( i 1 ( 2
1 j)

| 1 i < j 5,
8 ), with

even number of minus signs}.

1 ={ ( 2

...

8 ), 1

2, 2

4, 4

5, 4

5 }.

To every root system we associate its Dynkin diagram which is a graph whose nodes are the simple roots and the nodes i and j are joined by i , j j , i edges. Finally we mention that the basic representations of the Lie algebra corresponding to (or the corresponding simply connected group) are the fundamental representations. They correspond to fundamental weights i dened via i (j ) = i,j for all simple roots j , 1 j . The fundamental representations V (i ) are basic in that all others can be obtained from them as summands of tensor products. For classical groups most of them have very concrete descriptions. Example 1.7.
i

g = sln+1 , V (i ) = Example 1.8.

Cn+1 .

g = so2n+1 , V (i ) =

C2n+1 , 1 i n 1, V (n ) = spin(2n + 1).

GEOMETRY OF ORBIT CLOSURES

Example 1.9.
i i2

g = sp2n , V (i ) = Example 1.10.

C /
i

2n

C2n , 1 i n.

g = so2n , V (i ) =

C2n , 1 i n 2,

V (n1 ) = spin+ (2n), V (n ) = spin (2n). We will not need the description of spinor and half-spinor representations here. To a representation V of a reductive group G (like a triple tensor product or an exterior power) we can associate its diagram by choosing a black node (corresponding to a representation) and the white nodes corresponding to a Dynkin diagram of the group G. Let the Dynkin diagram of G have several connected components i . Then the representation V is the tensor product i V ((i) ) where (i) =
j

j j .

(I)

(i)

The diagram of V is obtained from by joining the black node with the vertex (I) (I) j with j edges. The main point is that (forgetting a few exceptions that will be listed) a representaton V has nitely many G-orbits if and only if the diagram of V is a Dynkin diagram. Exercises for section 1. 1. Let F be an orthogonal space of dimension 2n+1 with a non-degenerate quadratic form (, ). Choose a hyperbolic basis {e1 , . . . , en , e0 , en , . . . , e1 }. The orthogonal Lie algebra so2n+1 consists of endomorphisms of F such that ((u), v) + (u, (v)) = 0 u, v F.
2

Find the root decomposition of so(2n + 1) and identify it with the set F. 2. Let F be an symplectic space of dimension 2n with a non-degenerate skewsymmetric form (, ). Choose a hyperbolic basis {e1 , . . . , en , en , . . . , e1 }. The sym plectic Lie algebra sp2n consists of endomorphisms of F such that ((u), v) + (u, (v)) = 0 u, v F. Find the root decomposition of sp(2n) and identify it with the set S2 F . 3. Let F be an orthogonal space of dimension 2n with a non-degenerate quadratic form (, ). Choose a hyperbolic basis {e1 , . . . , en , en , . . . , e1 }. The orthogonal Lie algebra so2n consists of endomorphisms of F such that ((u), v) + (u, (v)) = 0 u, v F.
2

Find the root decomposition of so(2n) and identify it with the set

F.

JERZY WEYMAN

Hint. The weight of ei is i , the weight of ei is i . The Cartan subalgebra h is the subspace of vectors of weight zero in g.

2. The representations of type II and groups Let Xn be a Dynkin diagram and let g be the corresponding simple Lie algebra. Let us distinguish a node x Xn . Let be a corresponding simple root in the root system corresponding to Xn . The choice of determines a Z- grading on by letting the degree of a root be equal to the coecient of when we write as a linear combination of simple roots. On the level of Lie algebras this corresponds to a Z-grading g = iZ gi . We dene the group G0 := (G, G) C where (G, G) is a connected semisimple group with the Dynkin diagram Xn \ x. A representation of type II is the representation of G0 on g1 . By construction the representations of type II correspond to the Dynkin diagrams Xn with distinguished nodes. Denoting by l the Levi factor g0 we have l = l z(l) where l denotes the Lie algebra associated to Xn with the omitted node x. Let us look at some interesting examples of these gradings. We look at all possible grading of Lie algebra of type E6 . Notice that in each case the bracket gi gj gi+j has to be G0 -equivariant, so it i determined up to a non-zero scalar. So in each case we get a model of Lie algebra of type E6 in terms of classical Lie algebras and their representations. In each case we exhibit only gi for i 0 because we always have gi = g . i In each case we exhibit the restriction of the Killing form of g to g1 . k = 1, 6. The representation in question is V = V (4 , D5 ) is a half-spinor rep1 resentation for the group G = Spin(10). Here 4 = ( 2 , 1 , 1 , 1 , 1 ). dim(X) = 16. 2 2 2 2 The weights of X are vectors in 5 dimensional space, with coordinates equal to 1 , 2 with even number of negative coordinates.

GEOMETRY OF ORBIT CLOSURES

We label the weight vectors in g1 by [I] where I is the subset of {1, 2, 3, 4, 5} of even cardinality where the sign of the component is negative. The graded Lie algebra of type E6 is g(E6 ) = g1 g0 g1 with g0 = C so(10). The invariant scalar product (, ) on g restricted to g1 is given by the formula 1 ([I], [J]) = 2 #({I \ J} {J \ I}). 2 Thus possible scalar products are only 2, 1, 0. So the possible root systems we can get are A1 A1 and A1 . Indeed, the triple product A1 A1 A1 is not possible because there are no three subsets I, J, K with cardinalities of three symmetric dierences being 4. This means we get k = 2. V =
3

F , F = C6 , G = GL(F ).

The graded Lie algebra of type E6 is g(E6 ) = g2 g1 g0 g1 g2 with g0 = C sl(6), g1 =

C6 , g2 =

C6 .

The weights of g1 are i + j + k for 1 i < j < k 6. We label the corresponding weight vector by [I] where I is a cardinality 3 subset of {1, 2, 3, 4, 5, 6}. The invariant scalar product on g restricted to g1 is ([I], [J]) = 1 where = #(I J). k = 3, 5. V = E
2

F , E = C2 , F = C5 , G = SL(E) SL(F ) C

The graded Lie algebra of type E6 is g(E6 ) = g2 g1 g0 g1 g2 with g0 = C sl(2) sl(5), g1 = C2 C5 , g2 = C2 C5 . Let {e1 , e2 } be a basis of E, {f1 , . . . , f5 } be a basis of F . We denote the tensor ea fi fj by [a; ij]. The invariant scalar product on g restricted to g1 is ([a; ij], [b; kl]) = 1 where = #({a} {b}) + #({i, j} {k, l}).
2 2 4

JERZY WEYMAN

k = 4. V = E F H, E = C2 , F = H = C3 , G = SL(E)SL(F )SL(H)C . The graded Lie algebra of type E6 is g(E6 ) = g3 g2 g1 g0 g1 g2 g3 with g0 = C sl(2) sl(3) sl(3), g1 = C2 C3 C3 , g2 = C2 C3 C3 , 3 3 3 3 g3 = S2,1 C2 C C . Let {e1 , e2 } be a basis of E, and {f1 , f2 , f3 }, {h1 , h2 , h3 } bases of F , H respectively. We label ea fi hu by [a; i; u]. The invariant scalar product on g restricted to g1 is ([a; i; u], [b; j; v]) = 1 where = #({a} {b}) + #({i} {j}) + #({u} {v}). It is wortwhile exhibiting explicitly the center z(l). Denote the 1 , . . . , n the simple roots, and let x = i0 . Dene the elements xi h by setting j (xi ) = i,j . Then 1.1. Proposition. a) We have z(l) = Cxi0 . b) For a representation V of g we get the weights of the restriction V |l by applying the induced epimorphism h h(l ) whose kernel is xi0 , and then expressing i0 by means of other i by modifying the wieght by a multiple of xi0 . c) Consider the irreducible representation V of g with the highest weight = i mi i . Let us describe the restriction Vl . We want to label every irreducible representation in this restriction by the labeling of the Dynkin diagram Xn with the marked node i0 . This is done as follows. We use b) to determine the weights of (V )l and identify the highest weights. This will give us the markings on all the nodes except i0 . To determine the marking at the node i0 we need to describe the action of z(l). We have (xi0 ) =
i 2 2 2

mi i (xi0 ) =
i

mi (

mi j )xi0 = j
i

mi mi0 , i

where we write i =
j

mi j . j

The numbers mi can be read from the Bourbaki tables. j Vinberg in [V75] gave a description of the G0 -orbits in the representations of type II in terms of conjugacy classes of nilpotent elements in g. Let e g1 be a

GEOMETRY OF ORBIT CLOSURES

nilpotent element in g. Consider the irreducible components of the intersection of the conjugacy class of e in g C(e) g1 = C1 (e) . . . Cn(e) (e). The sets Ci (e) are clearly G0 -stable. Vinbergs result shows that these are precisely the G0 -orbits in g1 . 1.2 Theorem. The G0 -orbits of the action of G0 on g1 are the components Ci (e), for all choices of the conjugacy classes C(e) and all i, 1 i n(e). Theorem 1.2. makes the connection between the orbits in g1 and the nilpotent orbits in g. This means we will be needing the classication of nilpotent orbits in simple Lie algebras. This was obtained by Bala and Carter in the papers [BC76a], [BC76b]. A good account of this theory is the book [CM93]. Here we recall that the nilpotent orbit of an element e in a simple Lie algebra g is characterized by the smallest Levi subalgebra l containing e. One must be careful because sometimes l is equal to g. If the element e is a principal element in l, then this orbit is denoted by the Dynkin diagram of l (but there might be dierent ways in which the root system R(l) sits as a subroot system of R(g). There are, however the nonprincipal nilpotent orbits that are not contained in a smaller reductive Lie algebra l. These are called the distinguished nilpotent orbits and are described in sections 8.2, 8.3, 8.4 of [CM93]. They are characterized by their associated parabolic subgroups (as their Dynkin characteristics are eve, see section 8 in [CM93]). Let us remark that for Lie algebras of classical types, for type An the only distinguished nilpotent orbits are the principal ones, and for types Bn , Cn , Dn these are orbits corresponding to the partitions with dierent parts. For exceptional Lie algebras the distinguished orbits can be read o the tables ins ection 8.4. of [CM93]. This theorem is not easy to use because it is not very explicit. In section 4 we describe more precise method from second Vinberg paper [V87]. Exercises for section 2. 1. Identify the groups G0 and the representations g1 for the gradings of An , Bn , Cn , Dn , E7 and E8 related to simple roots. 2. For (E7 , 2 ) and E8 , 2 ) nd the models of corresponding Lie algebras.

3. Kacs classification of representations with finitely many orbits

The representations of reductive groups with nitely many orbits were classied by Kac in [K80] and [DK85]. The result is

10

JERZY WEYMAN

Theorem 3.1. Let H be a reductive group operating on an irreducible representation with nitely many orbits. Then the pair (H, V ) is on the following list a) H = G0 , V = g1 where g = iZ g1 is a grading on a simple Lie algebra g related to a simple root k , b) H = SL(2) SL(3) SL(n) C operating on C2 C3 Cn , n 6, c) H = SL(2) Spin(7), V = C2 V (3 ). In fact Kac classies a more general class of representations, so-called visible representations. The nullcone N (V ) of a representation V of G is the set of v V for which f (v) = 0 for every G-invariant function f Sym(V ). Denition 3.2. The nullcone N (V ) of a representation V of G is the set of v V for which f (v) = 0 for every G-invariant function f Sym(V ). A representation V of G is visible if there are nitely many G-orbits in the null-cone N (V ) of V . A very important class of visible representations is related to nite order automorphisms of the root systems classied by Kac. Let Xn be an extended Dynkin graph corresponding to the Dynkin graph Xn . Let xk be a node in Xn . Let ak be the coecient of the isotropic root of the graph Xn at the vertex xk . Then there is a grading on g by Z/ak Z where g0 is a subalgebra corresponding to Xn \ k and g1 is given by the same recipy as for the cases in previous section. Alternative description os that the Kac-Moody Lie algebra g has a grading g = iZ gi and the components gi are periodic with period ak . The visible representations g1 have analogous properties to the adjoint representations. Thy have a Cartan subspace h which is the slice representation. The Weyl group W := N (h)/Z(h) acts on h as a reection group generated by psedoreections. Vinberg proves the analogue of Chevalley Theorem. The natural map Sym(g )G Sym(h )W 1 is an isomorphism. Both rings are polynomial rings. Vinberg proves only existence of h and identies W using Sheppard-Todd table and the knowledge of degrees of invariants. It would be very interesting to nd the direct and explicit proofs of these results. Example 3.1. Let us consider the case (E6 , 4 ). The order of is 3. We have
2 2 2

g0 = sl3 sl3 sl3 , g1 = C3 C3 C3 , g2 =

C3

C3

C3 .

GEOMETRY OF ORBIT CLOSURES

11

Exercises for section 3. 1. Determine the graded components gk for the case (E7 , 4 ). Answer: g0 = sl2 sl4 sl4 , g1 = C2 C4 C4 ,
2 2 2 3 3

g2 =

C , g3 = S2,1 C

C4 .

2. Do the same for the case (E7 , 5 ). 4. The Vinberg method for classifying orbits. All Lie algebras g we will consider will be Lie algebras of some algebraic group G. Let (Xn , k ) be one of the representations on our list. It denes the grading g = iZ gi where gi is the span of the roots which, written as a combination of simple roots, have k with coecient i. The component g0 contains in addition a Cartan subalgebra. G0 denotes the connected component of the subgroup of G corresponding to g0 . In the sequel Z(x) denotes the centalizer of an element x G. Z0 (x) is Z(x)G0 . The gothic letters, z, z0 denote corresponding Lie algebras. Similarly, N (x) denotes the normalizer of an element x G. N0 (x) is N (x) G0 . We denote R(g) the set of roots of a reductive Lie algebra g, and (g) denotes the set of simple roots. In order to state Vinberg Theorem we need some denitions. We will be dealing with the graded Lie subalgebras s = iZ si . Denition 4.1. A graded Lie subalgebra s of g is regular if it is normalized by a maximal torus in g0 . A reductive graded Lie algebra s of g is complete if it is not a proper graded Lie subalgebra of any regular reductive Z-graded Lie algebra of the same rank. Denition 4.2. A Z-graded Lie algebra g is locally at if one of the following equivalent conditions is satised, for e a point in general position in g1 : (1) The subgroup Z0 (e) is nite, (2) z0 (e) = 0, (3) dimg0 = dimg1 . Let g = iZ gi be a graded reductive Lie algebra. Let us x a nonzero nilpotent element e ga . Let us choose some maximal trous H in N0 (e). Its Lie algebra h is the accompanying torus of the element e. We denote the character of the torus H dened by the condition [u, e] = (u)e

12

JERZY WEYMAN

for u h. Consider the graded Lie subalgebra g(h, ) of g dened as follows g(h, ) = iZ g(h, )i where g(h, )i = {x gia | [u, x] = i(u)u H }. Denition 4.3. The support s of the nilpotent element e ga is the commutant of g(h, ) considered as a Z graded Lie algebra. Clearly e s1 . We are ready to state the main theorem of [V87]. Theorem 4.4. (Vinberg). The supports of nilpotent elements of the space gi are exactly the complete regular locally at semisimple Z-graded subalgebras of the algebra g. The nilpotent element e can be recovered from the support subalgebra s as the generic element in s1 . It follows from the theorem that the nilpotent e is dened uniquely (up to conjugation by an element of G0 ) by its support. This means it is enough to classify the regular semisimple Z-graded subalgebras s of g. Let us choose the maximal torus t of g0 . The Z-graded subalgebra s is standard if it is normalized by t, i.e. if for all i Z we have [t, si ] si . Vinberg also proves that every Z graded subalgebra s is conjugated to a standard subalgebra by an element of G0 . Moreover, he shows that if two standard Z-graded subalgebras are conjugated by an element of G0 , then they are conjugated by an element of N0 (t). This allows to give a combinatorial method for classifying regular semisimple Z-graded subalgebras of g. Let s be a standard semisimple Z-graded subalgebra of g. The subalgebra s denes the degree map deg : R(s) Z. For a standard Z-graded subalgebra s we also get the map f : R(s) R(g). The map f has to be additive, i.e. it satises f ( + ) = f () + f () , R(s), f () = f () R(s). Moreover we have

GEOMETRY OF ORBIT CLOSURES

13

Proposition 4.5. The map f satises the following properties: a) (f (), f ()) (, ) = , R(s). (f (), f ()) (, ) b) f () f () R(g) , (s), / c) deg f () = deg , (s). Conversely, every map satisfying a), b), c) denes a standard regular Z-graded subalgebra s of g. Remark 4.6. The subalgebra s corresponding to the map f is complete if and only if there exists an element w in the Weyl group W of g such that wf ((s)) (g) (see [V87], p.25) . Proposition 4.5 means that in order to classify the nilpotent elements e g1 we need to classify the possible maps f corresponding to its support, i.e. the corresponding complete regular Z-graded subalgebra s. Since we are interested in the nilpotents e g1 , we need to classify the maps f for which degf () {0, 1} for every (s). One should also mention the connection with the Bala-Carter classication. The Bala-Carter characteristic of a nilpotent element is given by the type of the algebra s. So to nd the components of a given nilpotent orbit with Bala-Carter characteristic s it is enough to see in how many ways the Lie algebra s can be embedded as a Z-graded Lie algebra into g. One can determine the orbits explicitly using the following strategy. Proposition 4.7. The following procedure allows to nd the components of the intersection of a nilpotent orbit in g with g1 . 1. Establish the restriction of the invariant scalar product (, ) on g to g1 . 2. Fix a nilpotent element e g with Bala-Carter characteristic s. 3. Find in how many ways s embeds into g as a standard Z-graded Lie subalgebra by exhibiting corresponding map f as in Proposition 4.5. 4. For a characteristic s such that e is a principal nilpotent in s the map f sends all simple roots of s to weight vectors in g1 . Thus it is enough to classify the subsets of weights vectors in g1 for which the pattern of scalar products is the same as the one for simple roots of s. 5. For non-principal orbits one needs to make a more detailed analysis, but still one reduces to nding sets of weight vectors in g1 with certain patterns of scalar products. Exercises for section 4. Classiify the subalgebras s of type A3 +2A1 and A2 +2A1 in bigwedge3 C8 (the case (E8 , 2 ). Prove that we do not have any cases of type A5 and D4 .

14

JERZY WEYMAN

5. Vinberg method: examples. Example 5.1. Consider the diagram Xn = An . Let x be the note corresponding to m := m m+1 . We have g = sln+1 , G0 = SL(m, C) SL(n m + 1, C) C . The Z grading is g = g1 g0 g1 where g0 = C sl(m) sl(n m + 1), g1 = HomC (Cn+1m , Cm ). The weight vector ei fj corresponds to the root i m+j . Let us denote it by the label [i, j]. The invariant scalar product (, ) on g restricted to g1 is

([i1 , j1 ], [i2 , j2 ]) =

where = #([{i1 } {i2 }) + #([{j1 } {j2 }). This means the scalar products can take only values 2, 1, 0. So the only root systems whose simple roots can be embedded into g1 are (A1 )r for 1 r min(m, n). The corresponding set of weights has to be orthogonal, so (up to permutation of indices which means conjugation) we have one r-tuple {[1, 1], . . . , [r, r]}. The representative is the tensor e1 f1 + . . . + er fr . The corresponding orbit is the set of tensors of rank r. Consider a matrix of rank r in g1 . After changing the bases in Cm and in Cn+1m r we can assume that the corresponding nilpotent e(r) = j=1 Ej,m+j . The intersection of the orbit of e(r) with g1 is a determinantal variety of m (n + 1 m) matrices of rank r.

Lets go to triple tensor product cases.

Example 5.2. Consider the case (E6 , 4 ).

g(E6 ) = g3 g2 g1 g0 g1 g2 g3 with g0 = C sl(2) sl(3) sl(3), g1 = C2 C3 C3 , g2 = 3 3 3 3 g3 = S2,1 C2 C C .

C2

C3

C3 ,

GEOMETRY OF ORBIT CLOSURES

15

number 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17

s zero A1 2A1 2A1 2A1 3A1 3A1 A2 A2 + A1 A2 + A1 A2 + 2A1 A2 + 2A1 A2 + 2A1 2A2 A3 2A2 + A1 A3 + A1 D4 (a1 )

representative 0 [1; 1; 1] [1; 1; 1] + [2; 2; 1] [1; 1; 1] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] [1; 1; 1] + [1; 2; 2] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 1; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] [1; 1; 1] + [2; 1; 1] + [1; 2; 2] [2; 2; 2] + [1; 3; 3]

Example 5.3. Consider he case (E7 , 4 ). The graded Lie algebra of type E7 is g(E7 ) = g4 g3 g2 g1 g0 g1 g2 g3 g4 with g0 = C sl(2) sl(3) sl(4), g1 = C2 C3 C4 , g2 = C2 C3 C4 , 3 3 3 4 4 4 g3 = S2,1 C2 C C , g4 = S2,2 C2 S2,1,1 C3 C . Let {e1 , e2 } be a basis of E, and {f1 , f2 , f3 }, {h1 , h2 , h3 , h4 } bases of F , H respectively. We label ea fi hu by [a; i; u]. The invariant scalar product on g restricted to g1 is ([a; i; u], [b; j; v]) = 1 where = #({a} {b}) + #({i} {j}) + #({u} {v}). There are six H-nondegenerate orbits. They can be described by observing that the castling transform establishes a bijection between H-nondegenerate orbits and H -nondegenerate orbits for the 2 3 2 matrices corresponding to representation E F H . The six orbits in the representation E F H are: generic, hyperdeterminant hypersurface and four F -degenerate orbits, coming from 2 2 2 matrices: generic, hyperdeterminant and two determinantal varieties. Combining this knowledge with the case (E6 , 4) we get 23 orbits in our representation.
2 2 2

16

JERZY WEYMAN

number 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

s zero A1 2A1 2A1 2A1 3A1 3A1 A2 A2 + A1 A2 + A1 A2 + 2A1 A2 + 2A1 A2 + 2A1 2A2 A3 2A2 + A1 A3 + A1 D4 (a1 ) 2A2 A3 + A1 A3 + A2 A4 A4 + A1 A4 + A2

representative 0 [1; 1; 1] [1; 1; 1] + [2; 2; 1] [1; 1; 1] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] [1; 1; 1] + [1; 2; 2] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 1; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] [1; 1; 1] + [2; 1; 1] + [1; 2; 2] [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [2; 1; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 4] + [2; 3; 1]

Example 5.4. Consider the case (E8 , 4 ). The graded Lie algebra of type E8 is g(E8 ) = g6 g5 g4 g3 g2 g1 g0 g1 g2 g3 g4 g5 g6 with g0 = C sl(2) sl(3) sl(5), g1 = C2 C3 C5 , g2 = C2 C3 C5 , 3 3 3 5 4 5 g3 = S2,1 C2 C C , g4 = S2,2 C2 S2,1,1 C3 C , g5 = S3,2 C2 5 5 S2,2,1 C3 C , g6 = S3,3 C2 S2,2,2 C3 S2,14 C5 . Let {e1 , e2 } be a basis of E, and {f1 , f2 , f3 }, {h1 , h2 , h3 , h4 , h5 } bases of F , H respectively. We label ea fi hu by [a; i; u]. The invariant scalar product on g restricted to g1 is ([a; i; u], [b; j; v]) = 1 where = #({a} {b}) + #({i} {j}) + #({u} {v}).
2 2 2

GEOMETRY OF ORBIT CLOSURES

17

number 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

s zero A1 2A1 2A1 2A1 3A1 3A1 A2 A2 + A1 A2 + A1 A2 + 2A1 A2 + 2A1 A2 + 2A1 2A2 A3 2A2 + A1 A3 + A1 D4 (a1 ) 2A2 A3 + A1 A3 + A2 A4 A4 + A1 A4 + A2 A5 A6

representative 0 [1; 1; 1] [1; 1; 1] + [2; 2; 1] [1; 1; 1] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] [1; 1; 1] + [1; 2; 2] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 1; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] [1; 1; 1] + [2; 1; 1] + [1; 2; 2] [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [2; 1; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 4] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 4] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 5] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 4] + [1; 2; 5] + [2; 3; 1]

Exercises for section 5. Analyze one of the smaller cases, for example (D5 , 2 ).

6. Triple tensor products: quiver approach. In the case of (E6 , 4 ), (E7 , 4 ), (E8 , 4 ) we are mostly interested in, there is another approach using representations of Kronecker quiver. The Kronecker quiver is the quiver Q: 1
b a

18

JERZY WEYMAN

A representation V of Q is just a pair of vector spaces {V (1), V (2)} and a pair of linear maps. Each representation V of Q has dimension vector = (dimV (1), dimV (2)) Z2 . The representations of Q form an Abelian category (in fact a category of KQmodules where KQ is a path algebra of Q). By Krull-Remak-Schmidt theorem each representation decomposes uniquely (up to permutation of factors) to a direct sum of indecomposable representations. In the case of Kronecker quiver one can list all indecomposable representations. The are as follows. In each dimension vector (n, n) there is P1 worth of representations, with the standard A1 is given by
n V (1) = K n , V (2) = K n , V (a) = Id, V (b) = J n where J is an n n Jordan block of dimension n. We denote these representations n by V . Moreover, there are two discrete innite families: {P n } of dimension vector (n, n + 1), n 0, and {Qn } of dimension vector (n + 1, n), n 0), dened as follows 1 0 ... 0 0 1 ... 0 0 0 n P (a) = . . . . . . . . . . . . . . . 0 0 ... 0 1 0 0 ... 0 0 0 0 ... 0 0 0 ... 0 0 1 n P (b) = . . . . . . . . . . . . . . . 0 0 ... 1 0 0 0 ... 0 1

The matrices of Qn are just transposes of the matrices of P n . Proof. The idea is to introduce reection functors C + (2) and C (1), and their compositions and 1 . They take Q into itself. Thus they take indecomposables into indecomposables. C + (2) takes dimension vector (m, n) into (2m n, m). So, the vector (p+1, p) goes to (p+2, p+1) and (p, p+1) goes to (p1, p). Respectively, C (1) takes (m, n) into (n, 2n m). This means (q, q + 1) goes to (q + 1, q + 2) and (q + 1, q) goes to (q, q 1). Notice that the action of both functors on dimenson vectors (m, n) preserve m n. This means there are no indecomposables in dimension vectors (m, n) with |m n| 2, as they would imply the existence of indecomposables in dimension vectors (|m n|, 0) or (0, |m n|). In the dimension vectors (p + 1, p) and (p, p + 1) there is one indecomposable, as it is the case in dimension vectors (1, 0) and (0, 1). Finally, in dimension vector (n, n), assuming one of the matrices is nonsingular, we can use it to identify V (1) and V (2) so the description of indecomposables follows

GEOMETRY OF ORBIT CLOSURES

19

from Jordan canonical form. This is the case if any linear combination of V (a) and V (b) is nonsingular. If not, then there has to be subspace of dimension p + 1 in V (1) that is taken to a subspace of dimension p in V (2) by both V (a) and V (b) (apply Hilbert-Mumford criterion-exercise), which means V s not indecomposable, as , after applying reection functors we reduce to a representation of dimension (n, n) having a subrepresentation of dimension (1, 0) which is injective, so it splits. To a tensor in C2 Cp Cq we can associate the representation V () by choosing a xed basis {e1 , e2 } of C2 , write = e1 (a) + e2 (b) and take V ()(a) = (a), V ((b) = (b). Alternatively, the group GL(2) acts on the set of representations of Q of dimension vector (p, q) by replacing V (a), V (b) by the linear combinations of these maps, and the orbits in C2 Cp Cq are the GL(2)-orbits of this action. In particular the P1 families become one GL(2)-orbit, and we get nitely many possibilities, as long as the numer of summands of that type is 3. In this language our three tables have the following meaning. (E6 , 4 ). number 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 representative 0 [1; 1; 1] [1; 1; 1] + [2; 2; 1] [1; 1; 1] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] [1; 1; 1] + [1; 2; 2] + [2; 1; 2] [1; 1; 1] + [1; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] [1; 1; 1] + [2; 2; 2] + [1; 3; 2] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 1; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] [1; 1; 1] + [2; 2; 2] + [1; 2; 3] + [2; 3; 1] + [1; 3; 2] [1; 1; 1] + [2; 2; 2] + [1; 3; 3] + [2; 1; 3] [1; 1; 1] + [2; 1; 1] + [1; 2; 2] [2; 2; 2] + [1; 3; 3] quiver 3P 0 3Q0 1 V 2P 0 2Q0 P 1 P 0 2Q0 Q1 2P 0 Q0 1 2V P 0 Q0

We end this section with some remarks on the castling transform. This is a version of the reection functors, suitable for the triple tensor products. The method allows to classify the cases when the triple tensor product has an open orbit. This method goes back at least to Sato-Kimura [SK77] who classied the irreducible representations of reductive groups with an open orbit.

20

JERZY WEYMAN

We will work with the triple tensor products Cp Cq Cr . The main point of the approach is the following observation. Assume that p q r. If r > pq then any orbit in Cp Cq Cr is inherited from Cp Cq Cpq . Moreover, if r = pq then Cp Cq Cpq has an open orbit. Proposition 6.1. Assume that p q r, r < pq. Then there are open subsets in Cp Cq Cr and in Cp Cq Cpqr such that there is a bijection between the orbits within these open subsets. Proof. Let U Cp Cq Cr be the open set of tensors
r

t=
i=1

ti ei

where ti are linearly independent in Cp Cq . Notice that the GLp GLq GLr orbits in U are in bijection with the GLp GLq -orbits of r-dimensional subspaces in CC p Cq . But from an r-dimensional subspace in Cp Cq we can produce a pq r-dimensional subspace in (Cp ) (Cq ) by taking the set of linear forms vanishing on a given r-dimensional subspace in Cp Cq . The move from the triple (p, q, r) to (p, q, pq r) is called castling. Let us introduce the invariant N (p, q, r) := p2 q 2 r2 pqr. Notice that the invariant N (p, q, r) does not change when we apply castling. Theorem 6.2. The triple tensor product Cp Cq Cr has an open orbit only if N (p, q, r) 2. This happens precisely when by a sequence of castling moves (exchanging (p, q, r) to (p, q, pq r)) the triple can be reduced to (1, q, r) or to (2, q, r). The open orbit occurs unless we are in the case linked to (2, r, r) with r 4. Proof. Obviously pqr = dim Cp Cq Cr and p2 +q 2 +r2 = dim(GLp GLq GLr ). We have a chance to have an open orbit only if N (p, q, r) 2 (the number 2 comes from the fact that homoteties from three linear groups act in the same way on Cp Cq Cr ). Let us assume that p q r. Recall that the number N (p, q, r) will not change if we replace (p, q, r) by (p, q, pq r). Thus we may assume that r pq . Now we 2 write rq pr pq N (p, q, r) = p(p ) + q(q ) + r(r ). 6 3 2 This shows that if p 3 and q 6 all three summands will be negative. Thus it remains to classify cases p = 1, p = 2 and 3 p q 6. When p = 1 we have N (p, q, r) = 1 + q 2 + r2 qr = 1 + (q r)2 + qr 2.

GEOMETRY OF ORBIT CLOSURES

21

When p = 2 we have N (p, q, r) = 4 + q 2 + r2 2qr = 4 + (q r)2 4. Detailed case by case analysis shows that in the cases 3 p q r the value of N is never 2. In the cases with p = 1 example 5 shows that the algebra A3 (1, q, r) is nite dimensional. Let us look at the examples with p = 2. We will think of C2 Cq Cr as of the set of representations of Kronecker quiver of dimension vector (q, r) this will allow us to identify the generic tensors. The generic tensors can be determined by the generic decomposition. The indecomposables occur in dimensions (n, n) (one parameter families) and (n, n + 1) and (n + 1, n). Thus canonical decomposition of (n, n + k) is (k v)(u, u + 1) + v(u + 1, u + 2) where n = uk + v is division with remainder. Similarly (n + k, n) = (k v)(u + 1, u) + v(u + 2, u + 1). This decomposition allows to write easily the generic tensors. Exercises for section 6. 1. Apply Hilbert-Mumford criterion to prove that if a representation V of the Kronecker quiver has the property that for each , the linear combination V (a) + V (b) is singular, then there exist p, 0 p n 1 and a subspace W of dimension p + 1 in V (1) such that dim (V (a)W + V (b)W ) p. 2. Fill the rest of the entries in the table (E6 , 4 )

7. Geometric technique. In this section we provide a quick description of main facts related to geometric technique of calculating syzygies. We work over an algebraically closed eld K. Let us consider the projective variety V of dimension m. Let X = AN be the C ane space. The space X V can be viewed as a total space of trivial vector bundle E of dimension n over V . Let us consider the subvariety Z in X V which is the total space of a subbundle S in E. We denote by q the projection q : X V X and by q the restriction of q to Z. Let Y = q(Z). We get the basic diagram Z q Y X V q X

22

JERZY WEYMAN

The projection from X V onto V is denoted by p and the quotient bundle E/S by T . Thus we have the exact sequence of vector bundles on V 0 S E T 0 The dimensions of S and T will be denoted by s, t respectively. The coordinate ring of X will be denoted by A. It is a polynomial ring in N variables over C. We will identify the sheaves on X with A-modules. The locally free resolution of the sheaf OZ as an OXV -module is given by the Koszul complex
t 2

K () : 0

(p ) . . .

(p ) p () OXV

where = T . The dierentials in this complex are homogeneous of degree 1 in the coordinate functions on X. The direct image p (OZ ) can be identied with the the sheaf of algebras Sym() where = S . The idea of the geometric technique is to use the Koszul complex K() to construct for each vector bundle V on V the free complex F (V) of A-modules with the homology supported in Y . In many cases the complex F(OV ) gives the free resolution of the dening ideal of Y . For every vector bundle V on V we introduce the complex K(, V) := K() OXV p V This complex is a locally free resolution of the OXV -module M (V) := OZ p V. Now we are ready to state the basic theorem (Theorem (5.1.2) in [W]). Theorem 7.1. . For a vector bundle V on V we dene a free graded A-modules
i+j

F(V)i =
j0

H (V,

V) k A(i j)

a) There exist minimal dierentials di (V) : F(V)i F(V)i1 of degree 0 such that F(V) is a complex of graded free A-modules with Hi (F(V) ) = Ri q M (V) In particular the complex F(V) is exact in positive degrees.

GEOMETRY OF ORBIT CLOSURES

23

b) The sheaf Ri q M (V) is equal to H i (Z, M (V)) and it can be also identied with the graded A-module H i (V, Sym() V). c) If : M (V) M (V )(n) is a morphism of graded sheaves then there exists a morphism of complexes f () : F(V) F(V ) (n) Its induced map Hi (f ()) can be identied with the induced map H i (Z, M (V)) H i (Z, M (V ))(n). If V is a one dimensional trivial bundle on V then the complex F(V) is denoted simply by F . The next theorem gives the criterion for the complex F to be the free resolution of the coordinate ring of Y . Theorem 7.2.. Let us assume that the map q : Z Y is a birational isomorphism. Then the following properties hold. a) The module q OZ is the normalization of C[Y ]. b) If Ri q OZ = 0 for i > 0, then F is a nite free resolution of the normalization of C[Y ] treated as an A-module. 0 c) If Ri q OZ = 0 for i > 0 and F0 = H 0 (V, ) A = A then Y is normal and it has rational singularities. This is Theorem (5.1.3) in [W]. The complexes F(V) satisfy the Grothendieck type duality. Let V denote the canonical divisor on V . Theorem 7.3. Let V be a vector bundle on V . Let us introduce the dual bundle
t

V = V Then

V .

F(V ) = F(V) [m t] This is Theorem (5.1.4) in [W]. In all our applications the projective variety V will be a Grassmannian. To x the notation, let us work with the Grassmannian Grass(r, E) of subspaces of dimension r in a vector space F of dimension n. Let 0 R E Grass(r, E) Q 0 be a tautological sequence of the vector bundles on Grass(r, E). The vector bundle will be a direct sum of the bundles of the form S1 ,... ,nr Q S1 ,... ,r R. Thus all the exterior powers of will also be the direct sums of such bundles. We will apply repreatedly the following result to calculate cohomology of vector bundles S1 ,... ,nr Q S1 ,... ,r R.

24

JERZY WEYMAN

Proposition 7.4 (Botts algorithm). . The cohomology of the vector bundle S1 ,... ,nr Q S1 ,... ,r R on Grass(r, E) is calculated as follows. We look at the weight (, ) = (1 , . . . , nr , 1 , . . . , r ) and add to it = (n, n1, . . . , 1). Then one of two mutually exclusive cases occurs. a) The resulting sequence (, ) + = (1 + n, . . . nr + r + 1, 1 + r, . . . , r + 1) has repetitions. In such case H i (Grass(r, E), S Q S R) = 0 for all i 0. b) The sequence (, ) + has no repetitions. Then there is a unique permutation w n that makes this sequence decreasing. The sequence = w((, ) + ) is a non-increasing sequence. Then the only non-zero cohomology group of the sheaf S Q S R is the group H l where l = l(w) is the length of w. We have H l (Grass(r, E), S Q S R) = S E. Here S E denotes the highest weight representation S E of GL(E) corresponding to the highest weight (so-called Schur module).

8. Singularities, defining ideals and syzygies. The geometric method of calculating syzygies [W03] is applicable to analyze the singularities and dening ideals of orbit closures. We will look more closely at the three cases (E6 , 4 ), (E7 , 4 ) and (E8 , 4 ). Let us recall our basic notions. The coordinate ring of the triple tensor product is A = Sym(E F G ) = C[Xi,j,k ]. For an orbit closure Y = O(t) we dene its coordinate ring C[Y ] = A/J, J = {f A | f |O(t) = 0}. The ideal J is called the dening ideal of Y . In order to nd the data for some orbit closure we need to nd a desingularization of each orbit closure and then use the push-down of the Koszul complex to calculate the terms of a minimal free resolution. It is possible in most of the cases (and all cases for triple tensor products). Before we start talking about orbit closures for triple tensor products let us introduce the concept of a degenerate orbit.

GEOMETRY OF ORBIT CLOSURES

25

Denition 8.1. The GL(E) GL(F ) GL(G)-orbit of a tensor t E F G is E-degenerate (resp. F , G-degenerate) if there exists a proper subspace E E (resp. F F , G G) such that t E F G (resp. E F G, E F G ). The point is that for an E (resp. F , G)- degenerate obit closure we can get an estimate on the free resolution of the coordinate ring of the orbit closure as follows. Without loss of generality we can assume that dim E = dim E 1. Assume that we know the terms F of the free resolution of the coordinate ring of the orbit closure Y = O (t) where O (t) is the closure of the orbit of a tensor t treated as an element of E F G. Consider the desingularization of the subspace variety Z = {(t, R) E F G Grass(m 1, E) | r R F G} The tautological sequence associated to the projective space is 0 R E Grass(m 1, E) Q 0 Working over the sheaf of rings B = Sym(R F G ) we can produce the exact sequence F (R, F, G) of B modules modeled after F , whose terms are S R S F S G B(s). Taking sections we get an exact complex of A-modules supported in the subspace variety q(Z). Each of the modules H 0 (Grass(m 1, E), S R S F S G B) has a resolution with the terms

F (, , ) = H (Grass(m 1, E),

where = Q F G . We can use the cone construction to nd a (non-minimal) resolution of the coordinate ring of O(t) in E F G. Example 8.2. Look at the orbit O10 in the case (E6 , 4 ). It is clearly G-degenerate. In fact this is a general G-degenerate orbit, so we get its desingularization using the Grassmannian Grass(2, G) with the tautological sequence 0 R G Grass(2, G) Q 0 and the bundles = E F R , = E F Q . the resolution F has the terms
3

F0 = A, F1 =
4

(E F ) S1,1,1 G A(3),

F2 =

(E F ) S2,1,1 G A(4),

26
5

JERZY WEYMAN

F3 =
6

(E F ) S3,1,1 G A(5), (E F ) S4,1,1 G A(6),

F4 =

and our complex is just the Eagon-Northcott complex asscociated to the 33 minors of a 3 6 matrix we get when we atten our 3-dimensional matrix Example 8.3. Consider the orbit O8 in the case (E6 , 4 ). This is a G-degenerate orbit which corresponds to an orbit of codimension one in E F G . The resolution of this orbit closure is given by an invariant of degree 6, i.e. the complex 0 S3,3 E S2,2,2 F S3,3 G A (6) A . We construct a complex of sheaves F over B = Sym(E F R ) with F1 = S3,3 E S2,2,2 F S3,3 R B, F0 = B. Bott theorem implies both sheaves have no higher cohomology, so we get the exact sequence 0 H 0 (Grass(2, G), F1 ) H 0 (Grass(2, G), F0 ) C[O8 ]. Now we can nd the resolutions of both modules using the geometric construction from the previous example. The resolution of H 0 (Grass(2, G), F0 ) was worked out in the previous example. The resolution of H 0 (Grass(2, G), F1 ) is (use V = S3,3 E S2,2,2 F S3,3 R and from previous example). G0 = S3,3 E S2,2,2 F S3,3,0 G A(6), G1 = S3,3 E S2,2,2 F (E F ) S3,3,1 G A(7),
2

G2 = S3,3 E S2,2,2 F
3

(E F ) S3,3,2 G A(8), (E F ) S3,3,3 G A(9),

G3 = S3,3 E S2,2,2 F G4 = S3,3 E S2,2,2 F The mapping cone

(E F ) S4,4,4 G A(10). G F

gives a (possibly non-minimal) resolution of the coordinate ring C[O8 ]. In his case the length of our resolution is 5, i.e. equal to codimension of O8 so the orbit closure is Cohen-Macaulay.

GEOMETRY OF ORBIT CLOSURES

27

Example 8.4. The orbit closure O16 . This is the hypersurface given by the 233 matrices with vanishing hyperdeterminant. Its desingularization lives on P(E) P(F ) P(G). We treat each projective space as the set of 2-subspaces with the tautological subbundles RE , RF and RG respectively. The bundle is = Q QF G + E Q Q + Q F Q . E F G E G The complex F(16) is 0 (4, 2; 2, 2, 2; 2, 2, 2) A(6) (2, 1; 1, 1, 1; 1, 1, 1) A(3) A. The determinant of this matrix is the hyperdeterminant of our matrix which has degree 12. Example 8.5. Codimension 2 orbit closure O15 . The orbit closure has a geometric description as the set of pencils of 3 3 matrices whose determinant is a cube of a linear form. The bundle is the submodule with the weights (0, 1; 0, 0, 1; 0, 0, 1), (0, 1; 0, 1, 0; 0, 0, 1), (0, 1; 0, 0, 1; 0, 1, 0), (0, 1; 0, 1, 0; 0, 1, 0), (0, 1; 1, 0, 0; 0, 0, 1), (0, 1; 0, 0, 1; 1, 0, 0), (1, 0; 0, 0, 1; 0, 0, 1), (1, 0; 0, 1, 0; 0, 0, 1), (1, 0; 0, 0, 1; 0, 1, 0). This bundle can be thought of as the set of weights in the graphic form as follows. X X X X X O X O O X X O X O O O O O

Here the rst matrix represents the weights with (1, 0) on the rst coordinate, and the second matrix represents the weghts with (0, 1) on the rst coordinate. The symbol X denotes the weight in , the symbol O-the weight in . Our incidence space Z(15) has dimension 9 + 3 + 3 + 1 = 16, so it projects on the orbit of codimension 2. The calculation of cohomology reveals that we get a complex F(15) 0 (5, 4; 3, 3, 3; 3, 3, 3) A(9) (4, 2; 2, 2, 2; 2, 2, 2) A(6) A. This complex can be obtained also from the hyperdeterminant complex by looking at the kernel of the transpose of the cubic part of the complex F(16) . We see the complex is determinantal so it must give a resolution of the reduced ideal. This is

28

JERZY WEYMAN

in fact a proof that the incidence space Z(15) is a desingularization. Dividing by the regular sequence of 16 generic linear forms gives a resolution of the ring with Hilbert function 1 3t6 + 2t9 = 1 + 2t + 3t2 + 4t3 + 5t4 + 6t5 + 4t6 + 2t7 . (1 t)2 This means the degree of the orbit closure is 21. Example 8.6. The codimension 3 orbit O14 . This orbit closure has a nice geometric description. It consists of pencils of 3 3 matrices of linear forms containing a matrix of rank 1. The bundle is the submodule with the weights (0, 1; 0, 0, 1; 0, 0, 1), (0, 1; 0, 1, 0; 0, 0, 1), (0, 1; 0, 0, 1; 0, 1, 0), (0, 1; 0, 1, 0; 0, 1, 0), (0, 1; 1, 0, 0; 0, 0, 1), (0, 1; 0, 0, 1; 1, 0, 0), (0, 1; 1, 0, 0; 0, 1, 0), (0, 1; 0, 1, 0; 1, 0, 0). This bundle can be thought of as the set of weights in the graphic form as follows. X X X X X X X X X X O O O O O O O O

Here the rst matrix represents the weights with (1, 0) on the rst coordinate, and the second matrix represents the weghts with (0, 1) on the rst coordinate. The symbol X denotes the weight in , the symbol O-the weight in . Our incidence space has dimension 10 + 2 + 2 + 1 = 15, so it projects on the orbit of codimension 3. It is clearly O14 . The calculation of cohomology reveals that we get a complex F(14) 0 (5, 1; 2, 2, 2; 2, 2, 2) A(6) (3, 1; 2, 1, 1; 2, 1, 1) A(4) (2, 1; 1, 1, 1; 2, 1, 0) A(3) (2, 1; 2, 1, 0; 1, 1, 1) A(3) (1, 1; 1, 1, 0; 1, 1, 0) A(2) A. The terms of the complex F (14) can be understood by looking at the partitions on two last coordinates. They give terms of the Gulliksen-Nagard complex resolving 2 2 minors of the generic 3 3 matrix. The corresponding term on the rst coordnate comes from the symmetric power on the bundle Q . This means the E term comes from H 1 on the P(E) coordinate (except the trivial term), hence the shift in homological degree. The orbit is not normal. One still needs to resolve the cokernel module. This is an important problem, as similar situation occurs in several other cases.

GEOMETRY OF ORBIT CLOSURES

29

Example 8.7. The codimension 4 orbit O13 . The geometric description of this orbit can be understood from the quiver point of view. Our representation can be thought of as the set of representations of Kronecker quiver of dimension vector (3, 3). C3 3 C .

Our orbit closure then is the set of representations having a subrepresentation of dimension vector (2, 1). The desingularization Z(13) lives on Grass(1, F ) Grass(1, G) and the bundle is = E Q Q . F G Graphically we have X X X X O O X O O X X X X O O X O. O

Our incidence space has dimension 10 + 2 + 2 = 14, so it projects on the orbit of codimension 4. In this case one can see directly that Z(13) is a desingularization of O13 . The calculation of cohomology reveals that we get a complex F(13) 0 (4, 4; 3, 3, 2; 3, 3, 2) (4, 3; 3, 2, 2; 3, 2, 2) (2, 2; 2, 1, 1; 2, 1, 1) (3, 1; 2, 1, 1; 2, 1, 1) (3, 2; 2, 2, 2; 2, 2, 2) (2, 1; 1, 1, 1; 1, 1, 1) (2, 1; 1, 1, 1; 2, 1, 0) (2, 1; 2, 1, 0; 1, 1, 1) (3, 0; 1, 1, 1; 1, 1, 1) (1, 1; 1, 1, 0; 1, 1, 0)) (0, 0; 0, 0, 0; 0, 0, 0). The orbit closure is obviously not normal. The complex F(13) gives a minimal resolution of the normalization N (O13 ). We have an exact sequence 0 C[O13 ] C[N (O13 )] C(13) 0. The complex F(13) reveals that the A-module C(13) has the presentation (2, 1; 1, 1, 1; 1, 1, 1) (2, 1; 1, 1, 1; 2, 1, 0) (2, 1; 2, 1, 0; 1, 1, 1) (1, 1; 1, 1, 0; 1, 1, 0). Indeed, the representation (3, 0; 1, 1, 1; 1, 1, 1) can map only into the trivial term in the complex F(13) .

30

JERZY WEYMAN

We can look for the module with the above presentation among twisted modules supported in smaller orbits. It turns out the right choice is the orbit closure O7 . Its bundle has a diagram X X O X X O O O O X X O X X O O O. O

This bundle lives on the space Grass(2, F ) Grass(2, G). Consider the twisted 2 2 2 complex F( E R R )(7) . Its terms are: F G 0 (6, 6; 4, 4, 4; 4, 4, 4) (5, 4; 3, 3, 3; 4, 3, 2) (5, 4; 4, 3, 2; 3, 3, 3) (6, 3; 3, 3, 3; 3, 3, 3) (4, 4; 3, 3, 2; 3, 3, 2)(4, 4; 3, 3, 2; 4, 2, 2)(4, 4; 4, 2, 2; 3, 3, 2)2(5, 3; 3, 3, 2; 3, 3, 2) (4, 3; 3, 2, 2; 3, 2, 2) (4, 3; 3, 2, 2; 3, 3, 1) (4, 3; 3, 3, 1; 3, 2, 2) (5, 1; 2, 2, 2; 2, 2, 2) (2, 2; 2, 1, 1; 2, 1, 1) (3, 1; 2, 1, 1; 2, 1, 1) (3, 3; 2, 2, 2; 3, 3, 0) (3, 3; 3, 3, 0; 2, 2, 2) (2, 1; 1, 1, 1; 1, 1, 1) (2, 1; 1, 1, 1; 2, 1, 0) (2, 1; 2, 1, 0; 1, 1, 1) (1, 1; 1, 1, 0; 1, 1, 0). The resolution of he A-module C[O13 ] can be constructed as a mapping cone of the map
2 2 2

F(13) F(

QF

QH)(7)

covering the natural epimorphism of A-modules. The mapping cone is not a minimal resolution but the repeating representations might cancel out. Let us see that the pairs of representations (2, 2; 2, 1, 1; 2, 1, 1) and (3, 1; 2, 1, 1; 2, 1, 1) cancel out. From this we can deduce that the dening ideal of O13 is generated by the representations (3, 0; 1, 1, 1; 1, 1, 1) in degree 3 and the representations (3, 3; 2, 2, 2; 3, 3, 0) and (3, 3; 3, 3, 0; 2, 2, 2) in degree 6. Indeed, if (3, 1; 2, 1, 1; 2, 1, 1) would not cancel out, it would contribute to the minimal generators of the dening ideal. However it occurs once in S4 (E F G ) so that representation is already in the ideal generated by (3, 0; 1, 1, 1; 1, 1, 1). Regarding representation (2, 2; 2, 1, 1; 2, 1, 1), if it would occur in the dening ideal of O13 the analysis of the next section will show that then the dening ideal of O13 would contain the dening ideal of O12 . But this is impossible since both orbit closures have the same dimension. The homological dimension of C[O13 ] as an Amodule equals 5 because the top of the resolution of C(13) does not cancel out. So this coordinate ring is not Cohen-Macaulay. Also the non-normality locus of O13 equals to O7 .

GEOMETRY OF ORBIT CLOSURES

31

References
[ABW82] Akin, K., Buchsbaum, D., Weyman, J., Schur Functors and Schur Complexes, Adv. in Math. 44 (1982), 207-278. [BC76a] Bala, P., Carter, R., Classes of Unipotent Elements in Simple Algebraic Groups I,, Proc. Camb. Phil. Soc. 79 (1976), 401-425. [BC76b] Bala, P., Carter, R., Classes of Unipotent Elements in Simple Algebraic Groups II,, Proc. Camb. Phil. Soc. 80 (1976), 1-18. [CM93] Collingwood, D., McGovern, W., Nilpotent Orbits in Semisimple Lie Algebras, Van Nostrand Reinhold, New York, 1993, Van Nostrand Reinhold Mathematics Series. [DK85] Dadok, J., and Kac, V., Polar Reresentations, Journal of Algebra 92, (1985), 504-524. [EK94] Eisebud ,D., Koh J., Adv. in Math. 106 (1994), 1-35. [FH91] Fulton, W.; Harris, J., Representation Theory, Springer-Verlag, New York Heidelberg Berlin, 1991, Graduate Texts in Math. vol. 129. [GV] Gatti, V., Viniberghi E., Adv. in Math.. [GW98] Goodman, R., Wallach, N., Representations and Invariants of the Classical Groups, Cambridge University Press, Cambridge, UK, 1998, Encyclopedia of Mathematics and its Applications, vol. 68. [GY03] Grace, J.H., Young, A., The Algebra of Invariants, Chelsea, 1965. [Gu64] Gurevich, G.B., Foundations of the theory of algebraic invariants, Noordho, Groningen, 1964. [JPW81] Jzeak, T., Pragacz, P., Weyman, J., Resolutions of determinantal varieties and tensor o complexes associated with symmetric and antisymmetric matrices, Asterisque 87-88 (1981), 109-189. [K80] Kac, V., Some remarks on nilpotent orbits, J. of Algebra 64 (1980), 190-213. [Ko61] Kostant, B., Lie algebra cohomology and the generalized Borel-Weil theorem, Ann. of Mathematics 74 no. 2 (1961), 329387. [LM] Landsberg, J., Manivel, L.. [MM05] Manivel, L., Mezzetti, E, Manuscripta Mathematica 117 (2005), 319-331. [SK77] Sato, M, Kimura, T, A classication of irreducible prehomogeneous vector spaces and their relative invariants, Nagoya J. Math 65 (1977), 1-155. [V75] Vinberg, E. B., Weyl group of a graded Lie algebra, Izv. Akad. Nauk SSSR 40 (1975), 488-526. [V87] Vinberg, E.B., Classication of homogeneous nilpotent elements of a semisimple graded Lie algebra, Selecta Mathematica Sovietica 6 no.1 (1987). [W03] Weyman, J., Cohomology of vector bundles and syzygies, Cambridge University Press, Cambridge, UK, 2003, Cambridge Tracts in Mathematics, vol. 149. [WZ96] Weyman, J., Zelevinsky, A., Singularities of hyperdeterminants, Ann. Inst. Fourier (Grenoble) 46, 591644. Department of Mathematics, Northeastern University 360 Huntington Avenue, BOSTON, MA 02115, USA E-mail address: [email protected]