Lectures on the geometry of ag varieties

Transcription

1 Lectures on the geometry of ag varieties Michel Brion Introduction In these notes, we present some fundamental results concerning ag varieties and their Schubert varieties. By a ag variety, we mean a complex projective algebraic variety X, homogeneous under a complex linear algebraic group. The orbits of a Borel subgroup form a stratication of X into Schubert cells. These are isomorphic to ane spaces; their closures in X are the Schubert varieties, generally singular. The classes of the Schubert varieties form an additive basis of the cohomology ring H (X), and one easily shows that the structure constants of H (X) in this basis are all non-negative. Our main goal is to prove a related, but more hidden, statement in the Grothendieck ring K(X) of coherent sheaves on X. This ring admits an additive basis formed of structure sheaves of Schubert varieties, and the corresponding structure constants turn out to have alternating signs. These structure constants admit combinatorial expressions in the case of Grassmannians: those of H (X) (the Littlewood-Richardson coecients) have been known for many years, whereas those of K(X) were only recently determined by Buch [10]. This displayed their alternation of signs, and Buch conjectured that this property extends to all ag varieties. In this setting, the structure constants of the cohomology ring (a fortiori, those of the Grothendieck ring) are yet combinatorially elusive, and Buch's conjecture was proved in [6] by purely algebro-geometric methods. Here we have endeavoured to give a self-contained exposition of this proof. The main ingredients are geometric properties of Schubert varieties (e.g., their normality), and vanishing theorems for cohomology of line bundles on these varieties (these are deduced from the Kawamata-Viehweg theorem, a powerful generalization of the Kodaira vanishing theorem in complex geometry). Of importance are also the intersections of Schubert varieties with opposite Schubert varieties. These \Richardson varieties" are systematically used in these notes to provide geometric explanations for many formulae in the cohomology or Grothendieck ring of ag varieties. 1

2 The prerequisites are familiarity with algebraic geometry (for example, the contents of the rst three chapters of Hartshorne's book [30]) and with some algebraic topology (e.g., the book [26] by Greenberg and Harper). But no knowledge of algebraic groups is required. In fact, we have presented the notations and results in the case of the general linear group so that they may be extended readily to arbitrary connected, reductive algebraic groups by readers familiar with their structure theory. Thereby, we do not allow ourselves to use the rich algebraic and combinatorial tools which make Grassmannians and varieties of complete ags so special among all ag varieties. For these developments of Schubert calculus and its generalizations, the reader may consult the seminal article [43], the books [21], [23], [49], and the notes of Buch [11] and Tamvakis [68] in this volume. On the other hand, the notes of Duan in this volume [18] provide an introduction to the dierential topology of ag varieties, regarded as homogeneous spaces under compact Lie groups, with applications to Schubert calculus. The present text is organized as follows. The rst section discusses Schubert cells and varieties, their classes in the cohomology ring, and the Picard group of ag varieties. In the second section, we obtain restrictions on the singularities of Schubert varieties, and also vanishing theorems for the higher cohomology groups of line bundles on these varieties. The third section is devoted to a degeneration of the diagonal of a ag variety into unions of products of Schubert varieties, with applications to the Grothendieck group. In the fourth section, we obtain several \positivity" results in this group, including a solution of Buch's conjecture. Each section begins with a brief overview of its contents, and ends with bibliographical notes and open problems. These notes grew out of courses at the Institut Fourier (Grenoble) in the spring of 2003, and at the mini-school \Schubert Varieties" (Banach Center, Warsaw) in May I am grateful to the organizers of this school, Piotr Pragacz and Andrzej Weber, for their invitation and encouragements. I also thank the auditors of both courses, especially Dima Timashev, for their attention and comments. Conventions. Throughout these notes, we consider algebraic varieties over the eld C of complex numbers. We follow the notation and terminology of [30]; in particular, varieties are assumed to be irreducible. Unless otherwise stated, subvarieties are assumed to be closed. 2

3 1 Grassmannians and ag varieties We begin this section by reviewing the denitions and fundamental properties of Schubert varieties in Grassmannians and varieties of complete ags. Then we introduce the Schubert classes in the cohomology ring of ag varieties, and we study their multiplicative properties. Finally, we describe the Picard group of ag varieties, rst in terms of Schubert divisors, and then in terms of homogeneous line bundles; we also sketch the relation of the latter to representation theory. 1.1 Grassmannians The Grassmannian Gr(d; n) is the set of d-dimensional linear subspaces of C n. Given such a subspace E and a basis (v 1 ; : : : ; v d ) of E, the exterior product v 1 ^ ^ v d 2 V d C n only depends on E up to a non-zero scalar multiple. In other words, the point (E) := [v 1 ^ ^ v d ] of the projective space P( V d C n ) only depends on E. Further, (E) uniquely determines E, so that the map identies Gr(d; n) with the image in P( V d C n ) of the cone of decomposable d-vectors in V d C n. It follows that Gr(d; n) is a subvariety of the projective space P( V d C n ); the map is the Plucker embedding. The general linear group acts on the variety : Gr(d; n)! P( G := GL n (C) X := Gr(d; n) d^ C n ) via its natural action on C n. Clearly, X is a unique G-orbit, and the Plucker embedding is equivariant V V d with respect to the action of G on P( C n ) arising from its linear action d on C n. Let (e 1 ; : : : ; e n ) denote the standard basis of C n, then the isotropy group of the subspace he 1 ; : : : ; e d i is P := >< >: a 1;1 : : : a 1;d a 1;d+1 : : : a 1;n a d;1 : : : a d;d a d;d+1 : : : a d;n 0 : : : 0 a d+1;d+1 : : : a d+1;n : : : 0 a n;d+1 : : : a n;n 3 >= C A>;

4 (this is a maximal parabolic subgroup of G). Thus, X is the homogeneous space G=P. As a consequence, the algebraic variety X is nonsingular, of dimension dim(g) dim(p ) = d(n d). For any multi-index I := (i 1 ; : : : ; i d ), where 1 i 1 < : : : < i d n, we denote by E I the corresponding coordinate subspace of C n, i.e., E I = he i1 ; : : : ; e id i 2 X. In particular, E 1;2;:::;d is the standard coordinate subspace he 1 ; : : : ; e d i. We may now state the following result, whose proof is straightforward Proposition. (i) The E I are precisely the T -xed points in X, where T := 80 >< >: a 1;1 0 : : : 0 0 a 2;2 : : : : : : a n;n 19 >= C A>; GL n (C) is the subgroup of diagonal matrices (this is a maximal torus of G). (ii) X is the disjoint union of the orbits BE I, where B := 80 >< >: a 1;1 a 1;2 : : : a 1;n 0 a 2;2 : : : a 2;n : : : a n;n 19 >= C A>; GL n (C) is the subgroup of upper triangular matrices (this is a Borel subgroup of G) Denition. The Schubert cells in the Grassmannian are the orbits C I := BE I, i.e., the B-orbits in X. The closure in X of the Schubert cell C I (for the Zariski topology) is called the Schubert variety X I := C I. Note that B is the semi-direct product of T with the normal subgroup U := 80 >< >: 1 a 1;2 : : : a 1;n 0 1 : : : a 2;n : : : 1 19 >= C A>; (this is a maximal unipotent subgroup of G). Thus, we also have C I = UE I : the Schubert cells are just the U-orbits in X. Also, the isotropy group U EI is the subgroup of U where a ij = 0 whenever i =2 I and j 2 I. Let U I be the \complementary" subset of U, dened by a ij = 0 if i 2 I or j =2 I. Then one checks that U I is a subgroup of U, and the map U I! X, g 7! ge I is a locally 4

5 closed embedding with image C I. It follows that C I is a locally closed subvariety of X, isomorphic to the ane space C jij, where jij P d := (i j=1 j j). Thus, its closure X I is a projective variety of dimension jij. Next we present a geometric characterization of Schubert cells and varieties (see e.g. [21] 9.4) Proposition. (i) C I is the set of d-dimensional subspaces E C n such that dim(e \ he 1 ; : : : ; e j i) = # fk j 1 k d; i k < jg ; for j = 1; : : : ; n: (ii) X I is the set of d-dimensional subspaces E C n such that dim(e \ he 1 ; : : : ; e j i) # fk j 1 k d; i k < jg ; for j = 1; : : : ; n: Thus, we have X I = [ JI C J ; where J I if and only if j k i k for all k Examples. 1) For d = 1, the Grassmannian is just the projective space P n 1, and the Schubert varieties form a ag of linear subspaces X 1 X n, where X j = P j 1. 2) For d = 2 and n = 4 one gets the following poset of Schubert varieties: X point E Further, the Schubert variety X 24 is singular. Indeed, one checks that X P( V 2 C4 ) = P 5 is dened by one quadratic equation (the Plucker relation). Further, X 24 is the intersection of X with its tangent space at the point E 12. Thus, X 24 is a quadratic cone with vertex E 12, its unique singular point. 5

6 3) For arbitrary d and n, the Schubert variety X 1;2;:::;d is just the point E 1;2;:::;d, whereas X n d+1;n d+2;:::;n is the whole Grassmannian. On the other hand, X n d;n d+2;:::;n consists of those d-dimensional subspaces E that meet he 1 ; : : : ; e n d i: it is the intersection of X with the hyperplane of P( V d C n ) where the coordinate on e n d+1 ^ ^ e n vanishes. Since X is the disjoint union of the open Schubert cell C n d+1;n d+2;:::;n = C d(n d) with the irreducible divisor D := X n d;n d+2;:::;n, any divisor in X is linearly equivalent to a unique integer multiple of D. Equivalently, any line bundle on X is isomorphic to a unique tensor power of the line bundle L := O X (D), the pull-back of O(1) via the Plucker embedding. Thus, the Picard group Pic(X) is freely generated by the class of the very ample line bundle L. We may re-index Schubert varieties in two ways: 1. By partitions: with any multi-index I = (i 1 ; : : : ; i d ) we associate the partition = ( 1 ; : : : ; d ), where j := i j j for j = 1; : : : ; d. We then write X instead of X I. This yields a bijection between the set of multi-indices I = (i 1 ; : : : ; i d ) such that 1 i 1 < : : : < i d n, and the set of tuples of integers = ( 1 ; : : : ; d ) satisfying 0 1 : : : d n d. This is the set of partitions with d parts of size n d. The area of the partition is the number jj P d := j=1 j = jij. With this indexing, the dimension of X is the area of ; further, X X if and only if, that is, j j for all j. Alternatively, one may associate with any multi-index I = (i 1 ; : : : ; i d ) the dual partition (n i d ; n 1 i d 1 ; : : : ; n d + 1 i 1 ). This is still a partition with d parts of size n d, but now its area is the codimension of the corresponding Schubert variety. This indexing is used in the notes of Buch [11] and Tamvakis [68]. 2. By permutations: with a multi-index I = (i 1 ; : : : ; i d ) we associate the permutation w of the set f1; 2; : : : ; ng, dened as follows: w(k) = i k for k = 1; : : : ; d, whereas w(d + k) is the k-th element of the ordered set f1; : : : ; ng n I for k = 1; : : : ; n d. This sets up a bijection between the multi-indices and the permutations w such that w(1) < w(2) < < w(d) and w(d + 1) < < w(n). These permutations form a system of representatives of the coset space S n =(S d S n d ), where S n denotes the permutation group of the set f1; 2; : : : ; ng, and S d S n d is its subgroup stabilizing the subset f1; 2; : : : ; dg (and fd + 1; d + 2; : : : ; ng). Thus, we may parametrize the T -xed points of X, and hence the Schubert varieties, by the map S n =(S d S n d )! X, w(s d S n d ) 7! E w(1);:::;w(d). This parametrization will be generalized to all ag varieties in the next subsection. 1.2 Flag varieties Given a sequence (d 1 ; : : : ; d m ) of positive integers with sum n, a ag of type (d 1 ; : : : ; d m ) in C n is an increasing sequence of linear subspaces 0 = V 0 V 1 V 2 : : : V m = C n 6

7 such that dim(v j =V j 1 ) = d j for j = 1; : : : ; m. The coordinate ags are those consisting of coordinate subspaces. Let X(d 1 ; : : : ; d m ) denote the set of ags of type (d 1 ; : : : ; d m ). For example, X(d; n d) is just the Grassmannian Gr(d; n). More generally, X(d 1 ; : : : ; d m ) is a subvariety of the product of the Grassmannians Gr(d i ; n), called the partial ag variety of type (d 1 ; : : : ; d m ). The group G = GL n (C) acts transitively on X(d 1 ; : : : ; d m ). Let P = P (d 1 ; : : : ; d m ) be the isotropy group of the standard ag (consisting of the standard coordinate subspaces). Then P (d 1 ; : : : ; d m ) consists of the block upper triangular invertible matrices with diagonal blocks of sizes d 1 ; : : : ; d m. In particular, P (d 1 ; : : : ; d m ) contains B; in fact, all subgroups of G containing B occur in this way. (These subgroups are the standard parabolic subgroups of G). Since X = G=P, it follows that X is nonsingular of dimension P1i<jm d id j. In particular, we have the variety X := X(1; : : : ; 1) of complete ags, also called the full ag variety; it is the homogeneous space G=B, of dimension n(n 1)=2. By sending any complete ag to the corresponding partial ag of a given type (d 1 ; : : : ; d m ), we obtain a morphism f : X = G=B! G=P (d 1 ; : : : ; d m ) = X(d 1 ; : : : ; d m ): Clearly, f is G-equivariant with ber P=B at the base point B=B (the standard complete ag). Thus, f is a bration with ber being the product of varieties of complete ags in C d 1, : : :, C dm. This allows us to reduce many questions regarding ag varieties to the case of the variety of complete ags; see Example below for details on this reduction. Therefore, we will mostly concentrate on the full ag variety. We now introduce Schubert cells and varieties in G=B. Observe that the complete coordinate ags correspond to the permutations of the set f1; : : : ; ng, by assigning to the ag 0 he i1 i he i1 ; e i2 ; : : : ; e ik i the permutation w such that w(k) = i k for all k. We regard the permutation group S n as a subgroup of GL n (C) via its natural action on the standard basis (e 1 ; : : : ; e n ). Then the (complete) coordinate ags are exactly the F w := wf, where F denotes the standard complete ag. Further, S n may be identied with the quotient W := N G (T )=T, where N G (T ) denotes the normalizer of T in G. (In other words, S n is the Weyl group of G with respect to T ). We may now formulate an analogue of Proposition (see e.g. [21] 10.2 for a proof) Proposition. (i) The xed points of T in X are the coordinate ags F w, w 2 W. (ii) X is the disjoint union of the orbits C w := BF w = UF w, where w 2 W. (iii) Let X w := C w (closure in the Zariski topology [ of X), then X w = C v ; v2w; vw 7

8 where v w if and only if we have (v(1); : : : ; v(d)) r.t.i.v. (w(1); : : : ; w(d)) r.t.i.v. for d = 1; : : : ; n 1 (here r.t.i.v. stands for \reordered to increasing values") Denition. C w := BF w is a Schubert cell, and X w := C w is the corresponding Schubert variety. The partial ordering on W is the Bruhat order. By the preceding proposition, we have X v X w if and only if this holds for the images of X v and X w in Gr(d; n), where d = 1; : : : ; n 1. Together with Proposition 1.1.3, this yields a geometric characterization of the Bruhat order on Schubert varieties. Also, note that the T -xed points in X w are the coordinate ags F v, where v 2 W and v w. We now describe the Schubert cells UF w. Note that the isotropy group U Fw = U \ wuw 1 =: U w is dened by a i;j = 0 whenever i < j and w 1 (i) < w 1 (j). Let U w be the \complementary" subset of U, dened by a ij = 0 whenever i < j and w 1 (i) > w 1 (j). Then U w = U \ wu w 1 is a subgroup, and one checks that the product map U w U w! U is an isomorphism of varieties. Hence the map U w! C w, g 7! gf w is an isomorphism as well. It follows that each C w is an ane space of dimension #f(i; j) j 1 i < j n; w 1 (i) > w 1 (j)g = #f(i; j) j 1 i < j n; w(i) > w(j)g: The latter set consists of the inversions of the permutation w; its cardinality is the length of w, denoted by `(w). Thus, C w = C`(w). More generally, we may dene Schubert cells and varieties in any partial ag variety X(d 1 ; : : : ; d m ) = G=P, where P = P (d 1 ; : : : ; P m ); these are parametrized by the coset space S n =(S d1 S dm ) =: W=W P. Specically, each right coset mod W P contains a unique permutation w such that we have w(1) < < w(d 1 ), w(d 1 + 1) < < w(d 1 + d 2 ), : : :, w(d d m 1 + 1) < < w(d d m ) = w(n). Equivalently, w wv for all v 2 W P. This denes the set W P of minimal representatives of W=W P. Now the Schubert cells in G=P are the orbits C wp := BwP=P = UwP=P G=P (w 2 W P ), and the Schubert varieties X wp are their closures. One checks that the map f : G=B! G=P restricts to an isomorphism C w = BwB=B = BwP=P = CwP, and hence to a birational morphism X w! X wp for any w 2 W P Examples. 1) The Bruhat order on S 2 is just (21) (12) 8

9 The picture of the Bruhat order on S 3 is (321) (231) (213) (312) (132) (123) 2) Let w o := (n; n 1; : : : ; 1), the order-reversing permutation. Then X = X wo, i.e., w o is the unique maximal element of the Bruhat order on W. Note that w 2 o = id, and `(w o w) = `(w o ) `(w) for any w 2 W. 3) The permutations of length 1 are exactly the elementary transpositions s 1 ; : : : ; s n 1, where each s i exchanges the indices i and i+1 and xes all other indices. The corresponding Schubert varieties are the Schubert curves X s1 ; : : : ; X sn 1. In fact, X si may be identied with the set of i-dimensional subspaces E C n such that he 1 ; : : : ; e i 1 i E he 1 ; : : : ; e i+1 i: Thus, X si is the projectivization of the quotient space he 1 ; : : : ; e i+1 i=he 1 ; : : : ; e i 1 i, so that X si = P1. 4) Likewise, the Schubert varieties of codimension 1 are X wos 1 ; : : : ; X wos n 1, also called the Schubert divisors. 5) Apart from the Grassmannians, the simplest partial ag variety is the incidence variety I = I n consisting of the pairs (V 1 ; V n 1 ), where V 1 C n is a line, and V n 1 C n is a hyperplane containing V 1. Denote by P n 1 = P(C n ) (resp. P n 1 = P((C n ) )) the projective space of lines (resp. hyperplanes) in C n, then I P n 1 P n 1 is dened by the bihomogeneous equation x 1 y x n y n = 0; where x 1 ; : : : ; x n are the standard coordinates on C n, and y 1 ; : : : ; y n are the dual coordinates on (C n ). One checks that the Schubert varieties in I are the I i;j := f(v 1 ; V n 1 ) 2 I j V 1 E 1;:::;i and E 1;:::;j 1 V n 1 g; where 1 i; j n and i 6= j. Thus, I i;j I is dened by the equations x i+1 = = x n = y 1 = = y j 1 = 0: 9

10 It follows that I i;j is singular for 1 < j < i < n with singular locus I j 1;i+1, and is nonsingular otherwise. 6) For any partial ag variety G=P and any w 2 W P, the pull-back of the Schubert variety X wp under f : G=B! G=P is easily seen to be the Schubert variety X ww0;p, where w 0;P denotes the maximal element of W P. Specically, if P = P (d 1 ; : : : ; d m ) so that W P = S d1 S dm, then w 0;P = (w 0;d1 ; : : : ; w 0;dm ) with obvious notation. The products ww 0;P, where w 2 W P, are the maximal representatives of the cosets modulo W P. Thus, f restricts to a locally trivial bration X ww0;p! X wp with ber P=B. In particular, the preceding example yields many singular Schubert varieties in the variety of complete ags, by pull-back from the incidence variety Denition. The opposite Schubert cell (resp. variety) associated with w 2 W is C w := w o C wow (resp. X w := w o X wow). Observe that C w = B F w, where B := 80 >< >: a 1;1 0 : : : 0 a 2;1 a 2;2 : : : a n;1 a n;2 : : : a n;n 19 >= C A>; = w o Bw o (this is the opposite Borel subgroup to B containing the maximal torus T ). Also, X w has codimension `(w) in X. For example, C id = U via the map U! X, g 7! gf, where U := w o Uw o. Further, this map is an open immersion. Since X = G=B, this is equivalent to the fact that the product map U B! G is an open immersion (which, of course, may be checked directly). It follows that the quotient q : G! G=B, g 7! gb, is a trivial bration over C id ; thus, by G-equivariance, q is locally trivial for the Zariski topology. This also holds for any partial ag variety G=P with the same proof. Likewise, the map f : G=B! G=P is a locally trivial bration with ber P=B. 1.3 Schubert classes This subsection is devoted to the cohomology ring of the full ag variety. We begin by recalling some basic facts on the homology and cohomology of algebraic varieties, referring for details to [21] Appendix B or [23] Appendix A. We will consider (co)homology groups with integer coecients. Let X be a projective nonsingular algebraic variety of dimension n. Then X (viewed as a compact dierentiable manifold of dimension 2n) admits a canonical orientation, hence a canonical generator of the homology group H 2n (X) : the fundamental class [X]. By Poincare duality, the map H j (X)! H 2n j (X), 7! \ [X] is an isomorphism for all j. 10

11 Likewise, any nonsingular subvariety Y X of dimension p has a fundamental class in H 2p (Y ). Using Poincare duality, the image of this class in H 2p (X) yields the fundamental class [Y ] 2 H 2c (X), where c = n p is the codimension of Y. In particular, we obtain the fundamental class of a point [x], which is independent of x and generates the group H 2n (X). More generally, one denes the fundamental class [Y ] 2 H 2c (X) for any (possibly singular) subvariety Y of codimension c. Given, in the cohomology ring H (X), let h; i denote the coecient of thc class [x] in the cup product [. Then h; i is a bilinear form on H (X) called the Poincare duality pairing. It is non-degenerate over the rationals, resp. over the integers in the case where the group H (X) is torsion-free. For any two subvarieties Y, Z of X, each irreducible component C of Y \ Z satises dim(c) dim(y ) + dim(z), i.e., codim(c) codim(y ) + codim(z). We say that Y and Z meet properly in X, if codim(c) = codim(y ) + codim(z) for each C. Then we have in H (X): X [Y ] [ [Z] = m C [C]; C where the sum is over all irreducible components of Y \ Z, and m C is the intersection multiplicity of Y and Z along C, a positive integer. Further, m C = 1 if and only if Y and Z meet transversally along C, i.e., there exists a point x 2 C such that: x is a nonsingular point of Y and Z, and the tangent spaces at x satisfy T x Y + T x Z = T x X. Then x is a nonsingular point of C, and T x C = T x Y \ T x Z. In particular, if Y and Z are subvarieties such that dim(y )+dim(z) = dim(x), then Y meets Z properly if and only if their intersection is nite. In this case, we have h[y ]; [Z]i = P x2y \Z m x, where m x denotes the intersection multiplicity of Y and Z at x. In the case of transversal intersection, this simplies to h[y ]; [Z]i = #(Y \ Z). Returning to the case where X is a ag variety, we have the cohomology classes of the Schubert subvarieties, called the Schubert classes. Since X is the disjoint union of the Schubert cells, the Schubert classes form an additive basis of H (X); in particular, this group is torsion-free. To study the cup product of Schubert classes, we will need a version of Kleiman's transversality theorem, see [35] or [30] Theorem III Lemma. Let Y, Z be subvarieties of a ag variety X and let Y 0 Y (resp. Z 0 Z) be nonempty open subsets consisting of nonsingular points. Then there exists a nonempty is open subset of G such that: for any g 2, Y meets gz properly, and Y 0 \ gz 0 nonsingular and dense in Y \ gz. Thus, [Y ] [ [Z] = [Y \ gz] for all g 2. In particular, if dim(y ) + dim(z) = dim(x), then Y meets gz transversally for general g 2 G, that is, for all g in a nonempty open subset of G. Thus, Y \ gz is nite and h[y ]; [Z]i = #(Y \ gz), for general g 2 G. 11

12 Proof. Consider the map m : G Z! X, (g; z) 7! gz. This is a surjective morphism, equivariant for the action of G on G Z by left multiplication on the rst factor. Since X = G=P, it follows that m is a locally trivial bration for the Zariski topology. Thus, its scheme-theoretic bers are varieties of dimension dim(g) + dim(z) dim(x). Next consider the bered product V := (G Z) X Y and the pull-back : V! Y of m. Then is also a locally trivial bration with bers being varieties. It follows that the scheme V is a variety of dimension dim(g) + dim(z) dim(x) + dim(y ). Let : V! G be the composition of the projections (G Z) X Y! G Z! G. Then the ber of at any g 2 G may be identied with the scheme-theoretic intersection Y \ gz. Further, there exists a nonempty open subset of G such that the bers of at points of are either empty or equidimensional of dimension dim(y ) + dim(z) dim(x), i.e., of codimension codim(y ) + codim(z). This shows that Y meets gz properly for any g 2. Likewise, the restriction m 0 : G Z 0! X is a locally trivial bration with nonsingular bers, so that the bered product V 0 := (G Z 0 ) X Y 0 is a nonempty open subset of V consisting of nonsingular points. By generic smoothness, it follows that Y 0 \ gz 0 is nonsingular and dense in Y \ gz, for all g in a (possibly smaller) nonempty open subset of G. This implies, in turn, that all intersection multiplicities of Y \ Z are 1. Thus, we have [Y ] [ [gz] = [Y \ gz] for any g 2. Further, [Z] = [gz] as G is connected, so that [Y ] [ [Z] = [Y \ gz]. As a consequence, in the full ag variety X, any Schubert variety X w meets properly any opposite Schubert variety X v. (Indeed, the open subset meets the open subset BB = BU = B U of G; further, X w is B-invariant, and X v is B -invariant). Thus, X w \ X v is equidimensional of dimension dim(x w ) + dim(x v ) dim(x) = `(w) `(v). Moreover, the intersection C w \ C v is nonsingular and dense in X w \ X v. In fact, we have the following more precise result which may be proved by the argument of Lemma 1.3.1; see [9] for details Proposition. For any v; w 2 W, the intersection X w \ X v is non-empty if and only if v w; then X w \ X v is a variety Denition. Given v, w in W such that v w, the corresponding Richardson variety is X v w := X w \ X v. Note that X v w is T -invariant with xed points being the coordinate ags F x = xb=b, where x 2 W satises v x w. It follows that X v w X v0 w 0 if and only if v 0 v w w 0. Thus, the Richardson varieties may be viewed as geometric analogues of intervals for the Bruhat order Examples. 1) As special cases of Richardson varieties, we have the Schubert varieties X w = Xw id and the opposite Schubert varieties X v = Xw v o. Also, note that the Richardson variety Xw w is just the T -xed point F w, the transversal intersection of X w and X w. 12

13 2) Let Xw v be a Richardson variety of dimension 1, that is, v w and `(v) = `(w) 1. Then Xw v is isomorphic to the projective line, and v = ws for some transposition s = s ij (exchanging i and j, and xing all other indices). More generally, any T -invariant curve Y X is isomorphic to P 1 and contains exactly two T -xed points v, w, where v = ws for some transposition s. (Indeed, after multiplication by an element of W, we may assume that Y contains the standard ag F. Then Y \ C id is a T -invariant neighborhood of F in Y, and is also a T -invariant curve in C id = U (where T acts by conjugation). Now any such curve is a \coordinate line" given by a i;j = 0 for all (i; j) 6= (i 0 ; j 0 ), for some (i 0 ; j 0 ) such that 1 j 0 < i 0 n. The closure of this line in X has xed points F and s i0 ;j 0 F.) Richardson varieties may be used to describe the local structure of Schubert varieties along Schubert subvarieties, as follows Proposition. Let v, w 2 W such that v w. Then X w \vc id is an open T -invariant neighborhood of the point F v in X w, which meets X v w along X w \ C v. Further, the map (U \ vu v 1 ) (X w \ C v )! X w ; (g; x) 7! gx is an open immersion with image X w \ vc id. (Recall that U \ vu v 1 is isomorphic to C`(v) as a variety, and that the map U \ vu v 1! X, g 7! gf v is an isomorphism onto C v.) If, in addition, `(v) = `(w) 1, then X w \ C v is isomorphic to the ane line. As a consequence, X w is nonsingular along its Schubert divisor X v. Proof. Note that vc id is an open T -invariant neighborhood of F v in X, isomorphic to the variety vu v 1. In turn, the latter is isomorphic to (U \ vu v 1 ) (U \ vu v 1 ) via the product map; and the map U \ vu v 1! X, g 7! gf v is a locally closed immersion with image C v. It follows that the map (U \ vu v 1 ) C v! X; (g; x) 7! gx is an open immersion with image vf id, and that vf id \ X v = C v. Intersecting with the subvariety X w (invariant under the subgroup U \ vu v 1 ) completes the proof of the rst assertion. The second assertion follows from the preceding example. Richardson varieties also appear when multiplying Schubert classes. Indeed, by Proposition 1.3.2, we have in H (X): [X w ] [ [X v ] = [X v w]: Since dim(x v w) = `(w) `(v), it follows that the Poincare duality pairing h[x w ]; [X v ]i equals 1 if w = v, and 0 otherwise. This implies easily the following result. 13

14 1.3.6 Proposition. (i) The bases f[x w ]g and f[x w ]g = f[x wow]g of H (X) are dual for the Poincare duality pairing. (ii) For any subvariety Y X, we have [Y ] = X w2w a w (Y ) [X w ]; where a w (Y ) = h[y ]; [X w ]i = #(Y \ gx w ) for general g 2 G. In particular, the coecients of [Y ] in the basis of Schubert classes are non-negative. (iii) Let [X v ] [ [X w ] = X x2w a x vw [X x ] in H (X); then the structure constants a x vw are non-negative integers. Note nally that all these results adapt readily to any partial ag variety G=P. In fact, the map f : G=B! G=P induces a ring homomorphism f : H (G=P )! H (G=B) which sends any Schubert class [X wp ] to the Schubert class [X ww0;p ], where w 2 W P. In particular, f is injective. 1.4 The Picard group In this subsection, we study the Picard group of the full ag variety X = G=B. We rst give a very simple presentation of this group, viewed as the group of divisors modulo linear equivalence. The Picard group and divisor class group of Schubert varieties will be described in Subsection Proposition. The group Pic(X) is freely generated by the classes of the Schubert divisors X wosi where i = 1; : : : ; n 1. Any ample (resp. generated by its global sections) divisor on X is linearly equivalent to a positive (resp. non-negative) combination of these divisors. Further, any ample divisor is very ample. Proof. The open Schubert cell C wo has complement the union of the Schubert divisors. Since C wo is isomorphic to an ane space, its Picard group is trivial. Thus, the classes of X wos1 ; : : : ; X wosn 1 generate P the group Pic(X). n 1 If we have a relation a i=1 ix wosi = 0 in Pic(X), then there exists a rational function f on X having a zero or pole of order a i along each X wosi, and no other zero or pole. In particular, f is a regular, nowhere vanishing function on the ane space C wo. Hence f is constant, and a i = 0 for all i. Each Schubert divisor X wosd is the pull-back under the projection X! Gr(d; n) of the unique Schubert divisor in Gr(d; n). Since the latter divisor is a hyperplane section in the Plucker embedding, it follows that X wosd is generated by its global sections. 14

15 As a consequence, any non-negative combination P of Schubert divisors is generated by n 1 its global sections. Further, the divisor X d=1 w os d is very ample, as the product map X! Q n 1 d=1 Gr(d; n) is a closed immersion. Thus, any positive combination of Schubert divisors is very ample. P n 1 Conversely, let D = a i=1 ix wosi be a globally generated (resp. ample) divisor on X. Then for any curve Y on X, the intersection number h[d]; [Y ]i is non-negative (resp. positive). Now take for Y a Schubert curve X sj, then This completes the proof. X n 1 h[d]; [Y ]i = h a i [X wosi ]; [X sj ]i = i=1 n X1 i=1 a i h[x s i ]; [X sj ]i = a j : Remark. We may assign to each divisor D on X, its cohomology class [D] 2 H 2 (X). Since linearly equivalent divisors are homologically equivalent, this denes the cycle map Pic(X)! H 2 (X), which is an isomorphism by Proposition More generally, assigning to each subvariety of X its cohomology class yields the cycle map A (X)! H 2 (X), where A (X) denotes the Chow ring of rational equivalence classes of algebraic cycles on X (graded by the codimension; in particular, A 1 (X) = Pic(X)). Since X has a \cellular decomposition" by Schubert cells, the cycle map is a ring isomorphism by [22] Example We will see in Section 4 that the ring H (X) is generated by H 2 (X) = Pic(X), over the rationals. (In fact, this holds over the integers for the variety of complete ags, as follows easily from its structure of iterated projective space bundle.) Next we obtain an alternative description of Pic(X) in terms of homogeneous line bundles on X; these can be dened as follows. Let be a character of B, i.e., a homomorphism of algebraic groups B! C. Let B act on the product G C by b(g; t) := (gb 1 ; (b)t). This action is free, and the quotient L = G B C := (G C)=B maps to G=B via (g; t)b 7! gb. This makes L the total space of a line bundle over G=B, the homogeneous line bundle associated to the weight. Note that G acts on L via g(h; t)b := (gh; t)b, and that the projection f : L! G=B is G-equivariant; further, any g 2 G induces a linear map from the ber f 1 (x) to f 1 (gx). In other words, L is a G-linearized line bundle on X. We now describe the characters of B. Note that any such character is uniquely determined by its restriction to T (since B = T U, and U is isomorphic to an ane space, so that any regular invertible function on U is constant). Further, one easily sees that the characters of the group T of diagonal invertible matrices are precisely the maps diag(t 1 ; : : : ; t n ) 7! t 1 1 t n n ; 15

16 where 1 ; : : : ; n are integers. This identies the multiplicative group of characters of B (also called weights) with the additive group Z n. Next we express the Chern classes c 1 (L ) 2 H 2 (X) = Pic(X) in the basis of Schubert divisors. More generally, we obtain the Chevalley formula which decomposes the products c 1 (L ) [ [X w ] in this basis Proposition. For any weight and any w 2 W, we have c 1 (L ) [ [X w ] = X ( i j ) [X wsij ]; where the sum is over the pairs (i; j) such that 1 i < j n, ws ij < w, and `(ws ij ) = `(w) 1 (that is, X wsij is a Schubert divisor in X w ). In particular, c 1 (L ) = n X1 i=1 ( i i+1 ) [X wos i ] = n X1 i=1 ( i i+1 ) [X s i ]: Thus, the map Z n! Pic(X), 7! c 1 (L ) is a surjective group homomorphism, and its kernel is generated by (1; : : : ; 1). Proof. We may write c 1 (L ) [ [X w ] = X v2w a v [X v ]; where the coecients a v are given by a v = hc 1 (L ) [ [X w ]; [X v ]i = hc 1 (L ); [X w ] [ [X v ]i = hc 1 (L ); [X v w]i: Thus, a v is the degree of the restriction of L to X v w if dim(x v w) = 1, and is 0 otherwise. Now dim(x v w) = 1 if and only if : v < w and `(v) = `(w) 1. Then v = ws ij for some transposition s ij, and X v w is isomorphic to P 1, by Example Further, one checks that the restriction of L to X v w is isomorphic to the line bundle O P 1( i j ) of degree i j. This relation between weights and line bundles motivates the following Denition. We say that the weight = ( 1 ; : : : ; n ) is dominant (resp. regular dominant), if 1 n (resp. 1 > > n ). The fundamental weights are the weights 1 ; : : : ; n 1 such that j := (1; : : : ; 1 (j times); 0; : : : ; 0 (n j times)): The determinant is the weight n := (1; : : : ; 1). We put := n 1 = (n 1; n 2; : : : ; 1; 0): 16

17 By Propositions and 1.4.3, the line bundle L is globally generated (resp. ample) if and only if the weight is dominant (resp. regular dominant). Further, the dominant weights are the combinations a a n 1 n 1 + a n n, where a 1 ; : : : ; a n 1 are nonnegative integers, and a n is an arbitrary integer; n is the restriction to T of the determinant function on G. For 1 d n 1, the line bundle L( d ) is the pull-back of O(1) under the composition X! Gr(d; n)! P( V d C n ). Further, we have by Proposition 1.4.3: c 1 (L d ) [ [X w ] = [X wos d ] [ [X w ] = X v [X v ]; the sum over the v 2 W such that v w, `(v) = `(w) 1, and v = ws ij with i < d < j. We now consider the spaces of global sections of homogeneous line bundles. For any weight, we put H 0 () := H 0 (X; L ): This is a nite-dimensional vector space, as X is projective. Further, since the line bundle L is G-linearized, the space H 0 () is a rational G-module, i.e., G acts linearly on this space and the corresponding homomorphism G! GL(H 0 ()) is algebraic. Further properties of this space and a renement of Proposition are given by the following: Proposition. The space H 0 () is non-zero if and only if is dominant. Then H 0 () contains a unique line of eigenvectors of the subgroup B, and the corresponding character of B is. The divisor of any such eigenvector p satises div(p ) = n X1 i=1 ( i i+1 ) X s i : More generally, for any w 2 W, the G-module H 0 () contains a unique line of eigenvectors of the subgroup wb w 1, and the corresponding weight is w. Any such eigenvector p w has a non-zero restriction to X w, with divisor div(p w j Xw ) = X ( i j ) X wsij ; the sum over the pairs (i; j) such that 1 i < j n and X wsij is a Schubert divisor in X w. (This makes sense as X w is nonsingular in codimension 1, see Proposition ) In particular, taking =, the zero locus of p w j Xw is exactly the union of all Schubert divisors in X w. Proof. If is dominant, then we know that L is generated by its global sections, and hence admits a non-zero section. Conversely, if H 0 () 6= 0 then L has a section which does not vanish at some point of X. Since X is homogeneous, the G-translates of generate L. Thus, L is dominant. 17

18 Now choose a dominant weight and put D := P n 1 i=1 ( i i+1 ) X s i. By Proposition 1.4.3, we have L = OX (D), so that L admits a section with divisor D. Since D is B -invariant, is a B -eigenvector; in particular, a T -eigenvector. And since D does not contain the standard ag F, it follows that (F ) 6= 0. Further, T acts on the ber of L at F by the weight, so that has weight. If 0 is another B -eigenvector in H 0 (), then the quotient 0 = is a rational function on X, which is U -invariant as and 0 are. Since the orbit U F is open in X, it follows that the function 0 = is constant, i.e., 0 is a scalar multiple of. By G-equivariance, it follows that H 0 () contains a unique line of eigenvectors of the subgroup wb w 1, with weight w. Let p w be such an eigenvector, then p w does not vanish at F w, hence (by T -equivariance) it has no zero on C w. So the zero locus of the restriction p w j Xw has support in X w n C w and hence is B-invariant. The desired formula follows by the above argument together with Proposition Remark. For any dominant weight, the G-module H 0 () contains a unique line of eigenvectors for B = w o B w o, of weight w o. On the other hand, the evaluation of sections at the base point B=B yields a non-zero linear map H 0 ()! C which is a B-eigenvector of weight. In other words, the dual G-module V () := H 0 () contains a canonical B-eigenvector of weight. One can show that both G-modules H 0 () and V () are simple, i.e., they admit no non-trivial proper submodules. Further, any simple rational G-module V is isomorphic to V () for a unique dominant weight, the highest weight of V. The T -module V () is the sum of its weight subspaces, and the corresponding weights lie in the convex hull of the orbit W Z n R n. For these results, see e.g. [21] 8.2 and Example. For d = 1; : : : ; n 1, the space V d C n has a basis consisting of the vectors e I := e i1 ^ ^ e id ; where I = (i 1 ; : : : ; i d ) and 1 i 1 < < i d n. These vectors are T -eigenvectors with pairwise distinct weights, and they form a unique orbit of W. It follows easily that the G-module V d C n is simple with highest weight d (the weight of the unique B-eigenvector e 1:::d ). In other words, we have V ( d ) = V d C n, so that H 0 ( d ) = ( V d C n ). Denote by p I 2 ( V d C n ) the elements of the dual basis of the basis fe I g of V d C n. The p I are homogeneous coordinates on Gr(d; n), the Plucker coordinates. From the previous remark, one readily obtains that div(p I j XI ) = X J; J<I; jjj=jij 1 18 X J :

19 This is a rened version of the formula c 1 (L) [ [X I ] = X J; J<I; jjj=jij 1 in H (Gr(d; n)), where L denotes the pull-back of O(1) via the Plucker embedding. Note that c 1 (L) is the class of the unique Schubert divisor. Notes. The results of this section are classical; they may be found in more detail in [21], [23] and [49], see also [68]. We refer to [66] Chapter 8 for an exposition of the theory of reductive algebraic groups with some fundamental results on their Schubert varieties. Further references are the survey [67] of Schubert varieties and their generalizations in this setting, and the book [39] regarding the general framework of Kac-Moody groups. The irreducibility of the intersections X w \ X v is due to Richardson [64], whereas the intersections C w \ C v have been studied by Deodhar [16]. In fact, the Richardson varieties in thc Grassmannians had appeared much earlier, in Hodge's geometric proof [32] of the Pieri formula which decomposes the product of an arbitrary Schubert class with the class of a \special" Schubert variety (consisting of those subspaces having a nontrivial intersection with a given standard coordinate subspace). The Richardson varieties play an important role in several recent articles, in relation to standard monomial theory; see [48], [41], [40], [9]. The decomposition of the products c 1 (L ) [ [X w ] in the basis of Schubert classes is due to Monk [56] for the variety of complete ags, and to Chevalley [12] in general; see [61] for an algebraic proof. The Chevalley formula is equivalent to the decomposition into Schubert classes of the products of classes of Schubert divisors with arbitrary Schubert classes. This yields a closed formula for certain structure constants a x vw of H (X); specically, those where v = w o s d for some elementary transposition s d. More generally, closed formulae for all structure constants have been obtained by several mathematicians, see [37], [17], [61]. The latter paper presents a general formula and applies it to give an algebro-combinatorial proof of the Pieri formula. We refer to [31], [58], [59], [60] for generalizations of the Pieri formula to the isotropic Grassmannians which yield combinatorial (in particular, positive) expressions for certain structure constants. However, the only known proof of the positivity of the general structure constants is geometric. In fact, an important problem in Schubert calculus is to nd a combinatorial expression of these constants which makes their positivity evident. [X J ] 19

20 2 Singularities of Schubert varieties As seen in Examples and 1.2.3, Schubert varieties are generally singular. In this section, we show that their singularities are rather mild. We begin by showing that they are normal. Then we introduce the Bott-Samelson desingularizations, and we establish the rationality of singularities of Schubert varieties. In particular, these are Cohen-Macaulay; we also describe their dualizing sheaf, Picard group, and divisor class group. Finally, we obtain the vanishing of all higher cohomology groups H j (X w ; L ), where is any dominant weight, and the surjectivity of the restriction map H 0 () = H 0 (X; L )! H 0 (X w ; L ). 2.1 Normality First we review an inductive construction of Schubert cells and varieties. Given w 2 W and an elementary transposition s i, we have either `(s i w) = `(w) 1 (and then s i w < w), or `(s i w) = `(w) + 1 (and then s i w > w). In the rst case, we have Bs i C w = C w [ C si w, whereas Bs i C w = C si w in the second case. Further, if w 6= id (resp. w 6= w o ), then there exists an index i such that the rst (resp. second) case occurs. (These properties of the Bruhat decomposition are easily checked in the case of the general linear group; for arbitrary reductive groups, see e.g. [66].) Next let P i be the subgroup of G = GL n (C) generated by B and s i. (This is a minimal parabolic subgroup of G.) Then P i is the stabilizer of the partial ag consisting of all standard coordinate subspaces, except he 1 ; : : : ; e i i. Further, P i =B is the Schubert curve X si = P1, and P i = B [ Bs i B is the closure in G of Bs i B. The group B acts on the product P i X w by b(g; x) := (gb 1 ; bx). This action is free; denote the quotient by P i B X w. Then the map P i X w! P i X; (g; x) 7! (g; gx) yields a map : P i B X w! P i =B X; (g; x)b 7! (gb; gx): Clearly, is injective and its image consists of those pairs (gb; x) 2 P i =B X such that g 1 x 2 X w ; this denes a closed subset of P i =B X. It follows that P i B X w is a projective variety equipped with a proper morphism with image P i X w, and with a morphism : P i B X w! X f : P i B X w! P i =B = P 1 : The action of P i by left multiplication on itself yields an action on P i B X w ; the maps and f are P i -equivariant. Further, f is a locally trivial bration with ber B B X w = Xw. 20

21 In particular, P i X w is closed in X, and hence is the closure of Bs i C w. If s i w < w, then P i X w = X w. Then one checks that P i B X w may be identied with P i =B X w so that becomes the second projection. On the other hand, if s i w > w, then P i X w = X si w. Then one checks that restricts to an isomorphism so that is birational onto its image X si w. We are now in a position to prove Bs i B B C w! Bs i C w = C si w; Theorem. Any Schubert variety X w is normal. Proof. We argue by decreasing induction on dim(x w ) = `(w) =: `. In the case where ` = dim(x), the variety X w = X is nonsingular and hence normal. So we may assume that ` < dim(x) and that all Schubert varieties of dimension > ` are normal. Then we may choose an elementary transposition s i such that s i w > w. We divide the argument into three steps. Step 1. We show that the morphism : P i B X w! X si w satises R j O Pi B X w = 0 for all j 1. Indeed, factors as the closed immersion : P i B X w! P i =B X si w = P 1 X si w; (g; x)b 7! (gb; gx) followed by the projection p : P 1 X si w! X si w; (z; x) 7! x: Thus, the bers of are closed subschemes of P 1 and it follows that R j O P=B Xw = 0 for j > 1 = dim P 1. It remains to check the vanishing of R 1 O Pi B X w. For this, we consider the following short exact sequence of sheaves: 0! I! O P1 X si w! O Pi B X w! 0; where I denotes the ideal sheaf of the subvariety P i B X w of P 1 X si w. The derived long exact sequence for p yields an exact sequence R 1 p O P1 X si w! R1 p ( O Pi B X w )! R 2 p I: Further, R 1 p O P1 X si w = 0 as H 1 (P 1 ; O P 1) = 0; R 1 p ( O Pi B X w ) = R 1 O Pi B X w as is a closed immersion; and R 2 p I = 0 as all bers of p have dimension 1. This yields the desired vanishing. 21

22 Step 2. We now analyze the normalization map We have an exact sequence of sheaves : ~ Xw! X w : 0! O Xw! O ~ X w! F! 0; where F is a coherent sheaf with support the non-normal locus of X w. Further, the action of B on X w lifts to an action on ~ Xw so that is equivariant. Thus, both sheaves O Xw and O ~ X w are B-linearized; hence F is B-linearized as well. (See [8] x2 for details on linearized sheaves.) Now any B-linearized coherent sheaf G on X w yields an \induced " P i -linearized sheaf P i B G on P i B X w (namely, the unique P i -linearized sheaf which pulls back to the B- linearized sheaf G under the inclusion X w = B B X w! P i B X w ). Further, the assignment G 7! P i B G is exact. Therefore, one obtains a short exact sequence of P i -linearized sheaves on P i B X w : 0! O Pi B X w! (P i B ) O Pi B ~ X w! P i B F! 0: Apply, we obtain an exact sequence of sheaves on X si w: 0! O Pi B X w! (P i B ) O Pi B ~ X w! (P i B F)! R 1 O Pi B X w : Now O Pi B X w = O Xsi w by Zariski's main theorem, since : P i B X w! X si w is a proper birational morphism, and X si w is normal by the induction assumption. Likewise, (P i B ) O Pi B ~ X w = O Xsi w. Further, R 1 O Pi B X w = 0 by Step 1. It follows that (P i B F) = 0. Step 3. Finally, we assume that X w is non-normal and we derive a contradiction. Recall that the support of F is the non-normal locus of X w. By assumption, this is a non-empty B-invariant closed subset of X. Thus, the irreducible components of supp(f) are certain Schubert varieties X v. Choose such a v and let F v denote the subsheaf of F consisting of sections killed by the ideal sheaf of X v in X w. Then supp(f v ) = X v, since X v is an irreducible component of supp(f). Further, (P i B F v ) = 0, since F v is a subsheaf of F. Now choose the elementary transposition s i such that v < s i v. Then w < s i w (otherwise, P i X w = X w, so that P i stabilizes the non-normal locus of X w ; in particular, P i stabilizes X v, whence s i v < v). Thus, the morphism : P i B X v! X si v restricts to an isomorphism above C si v. Since supp(p i B F v ) = P i B X v, it follows that the support of (P i B F v ) contains C si v, i.e., this support is the whole X si v. In particular, (P i B F v ) is non-zero, which yields the desired contradiction. 22

23 2.2 Rationality of singularities Let w 2 W. If w 6= id then there exists a simple transposition s i1 such that `(s i1 w) = `(w) 1. Applying this to s i1 w and iterating this process, we obtain a decomposition w = s i1 s i2 s i`; where ` = `(w): We then say that the sequence of simple transpositions w := (s i1 ; s i2 ; : : : ; s i`) is a reduced decomposition of w. For such a decomposition, we have X w = P i1 X si1 w = P i1 P i2 P i`=b. We put v := s i1 w and v := (s i2 ; : : : ; s i`), so that w = (s i1 ; v) and X w = P i1 X v. We dene inductively the Bott-Samelson variety Z w by Z w := P i1 B Z v : Thus, Z w is equipped with an equivariant bration to P i1 =B = P1 with ber Z v at the base point. Further, Z w is the quotient of the product P i1 P i` by the action of B` via (b 1 ; b 2 ; : : : ; b` 1 ; b`)(g 1 ; g 2 ; : : : ; g`) = (g 1 b 1 1 ; b 1g 2 b 1 2 ; : : : ; b` 1g`b 1 ` ): The following statement is easily checked Proposition. (i) The space Z w is a nonsingular projective B-variety of dimension `, where B acts via g(g 1 ; g 2 ; : : : ; g`)b` := (gg 1 ; g 2 ; : : : ; g`)b`. For any subsequence v of w, we have a closed B-equivariant immersion Z v! Z w. (ii) The map Z w! (G=B)` = X`; (g 1 ; g 2 ; : : : ; g`)b` 7! (g 1 B; g 1 g 2 B; : : : ; g 1 g`b) is a closed B-equivariant embedding. (iii) The map ' : Z w = Z si1 ;:::;s i`! Z si1 ;:::;s i` 1 ; (g 1 ; : : : ; g` 1 ; g`)b` 7! (g 1 ; : : : ; g` 1 )B` 1 is a B-equivariant locally trivial bration with ber P i`=b = P 1. (iv) The map = w : Z w! P i1 P i`=b = X w ; (g 1 ; : : : ; g`)b` 7! g 1 g`b; is a proper B-equivariant morphism, and restricts to an isomorphism over C w. In particular, is birational. 23