Affine Schubert VarietiesAffine Schubert Varieties

and the Variety of Loop Complexes

Peter Magyar

magyar@math.msu.edu

The variety of two-step loop-complexes isthe set of pairs of matrices {(X,Y)  |  XY = 0, YX = 0}.Using the construction of Lusztig [4], we show that this varietyis isomorphic to an open subset of a Schubert variety for the loop group of GLn(C). As an application, we give an explicitBott-Samelson resolution of the loop-complex variety.

1  Affine Schubert varieties

Let F: = C((t)), the field of formal Laurent seriesf(t) = åi ³ Naj tj with aj Î C; and A: = C[[t]], the ring of formal Taylor series.Fix a positive integer n, and define the loop groupG = GLn(C)Ù: = GLn(F),the group of invertible n×n matrices with coefficientsin F.

Let V: = Fn, a vector space over F with a naturalaction of G. Let e1,¼, en denote the standardF-basis of V, and for c Î Z,define ei+nj: = tj ei, so that {ei}i Î Zis a C-basis of V. With respect to these bases, we may write an element v Î V either as a column vectorv = [a1,¼,an]T with coefficients ai Î F,or as an infinite column vector[¼,a-1,a0,a1,a2,¼]T with coefficients ai Î C. (Here T means transpose.)We will sometimes think of a matrix g Î G as a listof column vectors: g = (v1,¼,vn).

An A-lattice L Ì V is the A-submodule L = Av1żÅAvn, where {v1,¼,vn} is an F-basis of V.The standard A-lattice is: E: = SpanAáe1, ¼enñ = SpanCáeiñi ³ 1,and we let Ek: = SpanCáeiñi ³ k.

The affine Grassmannian Gr(V) is the spaceof all A-lattices of V. Clearly Gr(V) is ahomogeneous space with respect to the obvious actionof G, and the stabilizer of the standard lattice Eis the maximal parabolic P: = GLn(A), the subgroup of matrices X with coefficients in A such that detX = a0+a1 t+¼ with a0 ¹ 0.Thus G/P @ Gr(V), and a matrix with column vectors(v1,¼,vn) Î G corresponds to the latticewith basis áv1,¼,vnñ.

Let a,b be positive integers with a+b = n.The partial affine flag varietyFl(a,b;V) is the space of all pairs of latticesL· = (L1,L2)such that L1 É L2 É tL1 anddimC(L1/L2) = a, dimC(L2/tL1) = b.We write these conditions concisely as:

L1 a
É
 
L2 b
É
 
tL1.
The standard flag in Fl(a,b,V)is E·: = (E(1) É E(2)),where E(1): = E1, E(2): = Ea+1. The stabilizer of E· isPa, is the subgroup of matrices g Î P which are block-lower-triangular modulo t:
Pa: = {g = (gij) Î GLn(A) | gij Î tA  " 1 £ i £ a < j £ n} .
Thus, G/Pa @ Fl(a,b;V), and a matrix(v1,¼,vn) Î G corresponds tothe flag:
L1 = SpanAáv1,¼,vnñ    É    L2 = SpanAáva+1,¼,vn, tv1,¼,tvañ .

Now let S¥ be the group of bijections p:Z®Z,and let t:i® i+n be a shift bijection. Define the disconnected Weyl group W ~ Ì S¥ to be the subgroupof bijections which commute with t: that is, W ~ : = {p Î S¥  |  pt = tp}.Clearly, any permutation of [1,n] extends in a unique way to an element of W ~ . Furthermore, any element p Î W ~ is equivalent to a sequence of integers[p(1),¼,p(n)] such that -p-:i® -p(i)-defines a permutation of [1,n],where -m-: = m mod n. For example, we denotet = [n+1,n+2,¼,2n].

We may embed the W ~ Ì G.For p Î W ~ we let p act F-linearly on V by p(ei): = ep(i). The correspondingmatrix is the affine permutation matrix (aij)with a-p-(i),i = tci, where p(i) = -p-(i)+nci.For example, we may identify t = diag(t,¼,t),

Define the simple reflections s0,¼,sn-1 in W ~ by si(i) = i+1, si(i+1) = i, and si(j) = j for j not º i, i+1 mod n. We denote sn: = s0,and more generally si+nj: = si.Then s0,¼,sn-1 are involutionsgenerating a subgroup W Ì W ~ and satisfying theCoxeter relations (si si+1)3 = id for 0 £ i £ n-1, and (si sj)2 = id otherwise.We have a semi-directproduct W ~ = ásñ |×W. Here s:i® i+1 is the shift operator, whichacts on W via the outer automorphism:ssis-1 = si+1.

By Gaussian elimination, we establish theparabolic Bruhat decomposition of G into double Pa-cosets:G = -|-|-p Î Wa\W ~ /Wa PapPa, where we consider each p as an affine permutationmatrix, and Wa = ás1,¼,sa-1,sa+1,¼,snñ.We have the corresponding decomposition ofthe affine flag varietyFl(a,b;V) = -|-|-p Î Wa\W ~ /Wa Xpinto Schubert cells X°p: = Pa· pE·,where pE· = (pE(1) É pE(2)) is a translation of the standard flag.

The Schubert cells can be defined by Schubert conditions:

X°p = { L·  |   dimC(Li/LiÇ E(j)) = #(pZ(i)\ Z(j)) for i = 1,2, j Î Z},
where Z(1): = {1,2,3,¼}, Z(2): = {a+1,a+2,¼},  Z(j+2k): = tkZj; and p acts elementwise on setsof integers; also \ denotes set complement.The Schubert variety is thetopological closure Xp : = -X°p -.It can be described by replacing = in the above Schubert conditions with £ .

For any Schubert variety Xp,we define a certain affine open subset, the opposite cell X¢p Ì Xp.First we deal with the special case p Î W Ì W ~ .Let E¢(1): = SpanCáeiñi £ 0and E¢(2): = SpanCáeiñi £ acomplementary spaces to E(1),E(2), and define X¢p Ì Xp as the set of L· Î Xp such that LiÇE¢(i) = 0 for i = 1,2. For example, E· Î X¢p for any p Î W.Next, for the general case of p Î sk W, we letX¢p: = { L· Î Xp  |  LiÇsk E¢(i) = 0 for i = 1,2};so that skE· Î X¢p.

2  Loop-complexes and Lusztig's isomorphism

For positive integers a £ b, we consider the variety of two-step loop-complexes:

L = La,b: = { (X,Y) Î Mb×a(C)× Ma×b(C)  |  XY = 0, YX = 0}.
Recall that any finite chain-complex can be ``rolled up'' intosuch a two-step complex, by letting Ca (resp. Cb) bethe direct sum of all the odd-numbered (resp. even-numbered) spacesin the chain-complex.This gives a natural map from the variety of chain-complexesto L.

Now, L is a subvariety of the representations of the affine quiver AÙ1;a subvariety which is invariant under the natural action of the group GLa,b(C): = GLa(C)×GLb(C), namely(ga,gb)·(X,Y): = (gbXga-1,gaYgb-1).We easily see that L is a finite union of GLa,b(C)-orbits. In fact, Lhas exactly a+1 open orbits L°0,¼,L°a, whose closures give the a+1 irreducible components of L:

L°c: = {(X,Y) Î L  |   rankX = c, rankY = a-c},       Lc: = -L°c- .

We define an isomorphism from L to a union of opposite cells of Schubertvarieties in Fl(a,b;V), where V = Fn and n = a+b. Our notation will emphasizethe block decomposition V = FaÅFb,as well as:

E = CaÅCbÅtCaÅtCbż .
We define Lusztig's isomorphismY:L® G/Pa @ Fl(a,b;V) as:
Y(X,Y): = æ
ç
è
 
tIa
  tY
X
  tIb
 ö
÷
ø
mod Pa
where Im is an identity matrix of size m;or in terms of lattices, Y(X,Y) = (L1 É L2),where E É Li and:
L1  =  é
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
0
 0
X
 0
Ia
 Y
0
 Ib
0
 0
:
:
ù
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
  =  é
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
0
 0
X
 0
Ia
 0
0
 Ib
0
 0
:
:
ù
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
mod P ,      L2  =  é
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
0
 0
0
 0
Y
 0
Ib
 X
0
 Ia
0
 0
:
:
ù
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
  =  é
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
ê
0
 0
0
 0
Y
 0
Ib
 0
0
 Ia
0
 0
:
:
ù
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
ú
mod P .
Here the column vectors of each matrixgive an A-basis for Li Ì E, written with respect toE = SpanCáeiñi ³ 1. The blockshave sizes a,b,a,b,¼. The map Y is GLa,b(C)-equivariant,provided we embed GLa,b(C) Ì GLn(C) Ì Gas block-diagonal matrices with constant coefficients.

We easily deduce:

Y(Lc)  =  ì
ï
ï
í
ï
ï
î
  L·Î Fl(a,b;V)  ê
ê
ê
ê
ê
ê
ê
 
E(2) ÉE(3)
ÈÈ
L1 a
É
L2b
É 
tL1
È  È
E(4)ÉE(5)
L1ÇE¢(3) = L2ÇE¢(4) = 0
dim(L1/L1ÇE(3))£ c
dim(L2/L2ÇE(4)) £a-c
ü
ï
ï
ý
ï
ï
þ

Here

E(1) = E1,  E(2) = Ea+1,  E(3) = Ea+b+1,  E(4) = E2a+b+1,  E(5) = E2a+2b+1 .
From this, we may also realize Lc as a subset of an ordinary flag variety Fl(b,a,b;Ca+2b), the variety of partial flags
Ca+2b b
É
 
U1 a
É
 
U2 b
É
 
0 .
In fact, let Ca+2b = E(2)/E(5), with C-basis {-e-a+1,¼, -e-2a+2b},where -e-i: = ei mod E(5).Thus -E-(3) É -E-(4) is the standardflag in Fl(b,a,b;Ca+2b).We also define the nilpotent linear operator -t-:Ca+2b®Ca+2b by -t-(-e-i) = -e-i+n mod E(5) .Then we have:
Lc @ Y(Lc) @ ì
ï
ï
ï
í
ï
ï
ï
î
U· Î Fl(b,a,b;Ca+2b) ê
ê
ê
ê
ê
ê
ê
ê
U1É-E-(4),  U2 Ì-E-(3),
dim(U1Ç-E-(3)) ³ a+b-c
dim(U2Ç-E-(4)) ³ b-a+c
U1Ç-E-¢(3) = U2Ç-E-¢(4) = 0
U2 É-t-(U1)
ü
ï
ï
ï
ý
ï
ï
ï
þ
This is precisely the opposite cell of a Schubertvariety in Fl(b,a,b;Ca+2b), but with theadditional algebraic incidence condition U2 É -t-(U1), which can be writtenin terms of the Plucker coordinatesof U1, U2.

3  Affine permutations for loop-complexes

We wish to identify the image of Lc as the opposite cell of an affine Schubert variety,

Y(Lc) = X¢p ,    for some   p = pc Î Wa\W ~ /Wa .
First, we construct the setspZ(1), pZ(2), which then determinep modulo Wa.These sets should contain numbers as small as possiblesubject to the conditions:
Z(2) 
É
 Z(3) 
b

 
È   
a

 
È   
pZ(1)
   a
É
 
  
pZ(2)
   b
É
 
  
tpZ(1)
a

 
È   
b

 
È   
Z(4) 
É
 Z(5) 
      
#(pZ(1)\ Z(3)) = c
#(pZ(2)\ Z(4)) = a-c
 ,
To construct p according to these constraints,we will divide [1,n] into intervals (blocks) of the form:


i,¼,j
(k) 
: = [i, i+1,¼,j],
where k = j-i+1 is the number of integers in theinterval. We perform two subdivisions as follows:
[1,¼,n]: = [


1,¼,c
i (c) 
,

c+1,¼,a
ii (a-c) 
,

a+1,¼,2a-c
iii (a-c) 
,


2a-c+1,¼,a+b-c
iv (b-a) 
,

a+b-c+1,¼,a+b
v (c) 
  ]
 ,
[1,¼,n]: = [


1,¼,a-c
i¢ (a-c) 
,

a-c+1,¼,a
ii¢ (c) 
,

a+1,¼,a+c
iii¢ (c) 
,


a+c+1,¼,b+c
iv¢ (b-a) 
,

b+c+1,¼,a+b
v¢ (a-c) 
  ]
 ,
where we have numbered the blocks with roman numerals.Now p takes the first set of blocks tothe second set, as well as shifting themby powers of t:
p: = [


a+1,¼,a+c
v¢ (c) 
,

a+2b+c+1,¼,2a+2b
t(iii¢) (a-c) 
,

a+b+1,¼,2a+b-c
t(ii¢) (a-c) 
,


2a+b+c+1,¼,a+2b+c
t(iv¢) (b-a) 
,

3a+2b-c+1,¼,3a+2b
t2(i¢) (c) 
  ]
 .
Recall that p represents the double coset Wap Wa: in fact, p is maximal with respect tothe left action of Wa, and minimal with respect tothe right action. This is the correct normalizationso that l(pc) = dimC(Lc).

To analyze the decomposition of p into simple reflections,we construct the loop wiring diagram of p.That is, the strip below represents a cylinder (identify thetop and bottom edges) with n = a+b dots on either end.For each i we write p(i) = -p-(i)+nj, and we draw a wire connecting the dot ion the right to the dot -p-(i) on the left,but looping upwards (around the cylinder) j times.We will group the wires into five cables correspondingto our blocks i,...,v (on the right)and i¢,¼,v¢ (on the left),so that the cable starting at i represents cnon-crossing wires, etc. As a final simplification,instead of drawing the diagram for p, we instead draw the diagram for t-1p(a harmless normalization, since t is in the center ofW ~ ).



Whenever a cable with k wires crossesone with k¢ wires, we have a total of kk¢ wirecrossings. Thus the six cable-crossings of our picturegive a wire-crossing total of:

l(p) = (a-c)2+c2+(a-c)(b-a)+2c(a-c)+c(b-a) = ab ,
which we may confirm by checking directly that dim(Lc) = ab.

Now we may write a reduced decomposition for pas follows. For integers i,k, define the affine permutation si[k]: = si si-2¼si-2k+2, which has kmutually commuting factors. Recall our conventionsi+nj: = si. For each cable crossing:



i+1,¼,i+k
(k) 


i+1,¼,i+j
(j) 
  \
/  
  /
\  


i+k+1,¼,i+j+k
(j) 


i+j+1,¼,i+j+k
(k) 
we define the associated ``totally commutative'' permutation:
si+1[j,k] : =  si+k[1]si+k+1[2]¼si+k+m[min(m,j,k)]¼si+j+k[min(j,k)]¼si+j+m¢[min(m¢,j,k)]¼si+j+1[2]si+j[1].
Finally, we can write
p = t s1[a-c,c] sa+1[b-a,c] sb+1[a-c,c] sa+1[a-c,b-a] sb+a-c+1[c,c] sc+1[a-c,a-c] ,
where the six factors (other than t) correspond to the cable crossings, listed left to right.

4  Bott-Samelson resolution

We can use the above data to give a Bott-Samelsonresolution of singularities for Lc. Although this is clearly far from a minimal resolution,it brings the loop complexes into the framework of Frobenius splittings and other results forBott-Samelson varieties. In particular, thesingularities of Lc are rational, effective line bundles are acyclic, etc.

The construction of the affine Bott-Samelson varietyZi corresponding to a reduced word i = (i1,¼,il)is exactly analogous to (and includesas a special case) the construction for GLn(C).It is best illustrated by an example.

Let us take c = 1, a = 2, b = 3, n = 5 so that each ofthe blocks i,¼,v hassize 1, and our cable diagram in the previoussection is a simple wiring diagram.Then p = s1 s3 s4 s3 s0 s2,and the Bott-Samelson variety is:

Zi : =  ì
ï
ï
ï
í
ï
ï
ï
î
(L1,L2,L3,L4,L5,L¢4)
Î Gr(V)6
 ê
ê
ê
ê
ê
ê
ê
ê
 <
E4
/\
E1¬E2¬E3¬L¢4¬E5¬tE1
\/\\/
L1¬L2¬L3¬L4¬L5¬tL1
ü
ï
ï
ï
ý
ï
ï
ï
þ
.
Here Ej: = SpanCáeiñi ³ j,and each arrow U ¬ V or diagonal line U -- V indicatesthe conditions U É V, dimC(U/V) = 1.We construct a point of Zi by starting with the standard flag E1 É E3 É ¼,and successively choosing the spaces L2,L¢4, L5, L4, L1, L3,corresponding to the letters 1,3,4,3,0,2.Each such choice corresponds toa fibration with fiber P1, hence Ziis smooth of dimension l(p) = 6. As we did forXp, we can embedZi into a finite-dimensional flag varietyfor the C-vector space t-1E3/tE1,since t-1E3 É L É tE1for L = L1,¼,L5,L¢4.

We can definea regular, birational map of Zionto Xp byforgetting all the spaces except (L1,L3). This map is genericallyone-to-one because generically all the spacesare determined by L1, L3: that is,L2 = L1Ç E1L5 = tL1+tE1, etc.To desingularize the opposite cell X¢p @ Lc, we consider the subset of Zi whereL1, L3 are generic with respect to the opposite standard flag E¢1, E¢3.

For general a,b,c, the diagram of inclusionsdefining Zi is closely related to thedual graph of the wiring diagram of p.

References

[1]
A. Björner and F. Brenti,Affine permutations of type A,El. J. Combin. 3, Foata Festschrift.

[2]
V. Lakshmibai and P. Magyar,Degeneracy schemes, quiver schemes,and Schubert varieties, Internat. Math. Res. Notices12 (1988) 627-640.

[3]
G. Lusztig,Green polynomials and singularities of unipotentclasses, Adv. in Math. 42 (1981) 169-178.

[4]
G. Lusztig,Canonical bases arising from quantizeduniversal enveloping algebras,JAMS 3 (1990) 447-498.


File translated from TEX by TTH, version 2.34.
On 18 Sep 1999, 15:50.