arXiv:0803.4121v2 [math.

QA] 5 May 2008

A diagrammatic approach to categorication of quantum groups I

Mikhail Khovanov and Aaron D. Lauda March 31, 2008
Abstract To each graph without loops and multiple edges we assign a family of rings. Categories of projective modules over these rings categorify Uq (g), where g is the Kac-Moody Lie algebra associated with the graph.

Contents
1 Introduction 2 Rings R() and their properties 2.1 Denitions . . . . . . . . . . . . . . . . . . . 2.2 Examples . . . . . . . . . . . . . . . . . . . 2.3 A faithful representation and a basis of R() 2.4 Properties of R() . . . . . . . . . . . . . . 2.5 Representations . . . . . . . . . . . . . . . . 2.6 Induction and restriction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 5 5 10 15 22 25 31 34 34 39 45 48

3 Quantum groups and the Grothendieck ring of R 3.1 Homomorphism of twisted bialgebras . . . . . . . . . . . . . 3.2 Surjectivity of . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 Tight monomials and indecomposable projectives . . . . . . . 3.4 A conjecture on categorication of irreducible representations

INTRODUCTION

Introduction

The goal of this paper is to categorify U = Uq (g), for an arbitrary simplylaced Kac-Moody algebra g. Here U stands for the quantum deformation of the universal enveloping algebra of the lower-triangular subalgebra of g. Following the discovery of quantum groups Uq (g) by Drinfeld [12] and Jimbo [18], Ringel [40] found a Hall algebra interpretation of the negative half U of the quantum group in the simply-laced Dynkin case. Lusztig [31], [32], [33] gave a geometric interpretation of U and produced a canonical basis there via a sophisticated approach which required the full strength of the theory of l-adic perverse sheaves. Kashiwara [19] dened a crystal basis of U at 0, a graph equipped with extra data, and in [20] constructed the socalled global crystal basis of U . Grojnowski and Lusztig [16] proved that the global crystal basis and the canonical basis are the same. The canonical basis B of U gives rise to bases in all irreducible integrable U-representations. Lusztig [34] also produced an idempotented version U of U and dened a basis there. The work of Ariki [1] can be viewed as a categorication of the restricted dual of U (g) for g = slN and g = slN and a categorication of all irreducible integrable representations of these Lie algebras (see also [27], [2], [3], [37]). An integral version of the restricted dual of U (g) becomes the sum of Grothendieck groups of suitable blocks of ane Hecke algebra representations. An earlier work of Zelevinsky [47] can be understood in this context as a parametrization of basis elements of U (g) via certain irreducible representations of ane Hecke algebras. Irreducible integrable representations of U(g) become Grothendieck groups of Ariki-Koike cyclotomic Hecke algebras, which are certain nite-dimensional quotient algebras of ane Hecke algebras. Grojnowski [14] found a purely algebraic way to understand these categorications via a generalization of Kleshchevs methods for studying modular representations of the symmetric group [22], [23], [24]. This approach was further developed by Grojnowski-Vazirani [17], Vazirani [45], [46], BrundanKleshchev [8] and others. It is explained in Kleshchev [25] in the context of degenerate ane Hecke algebras. In this paper we introduce graded algebras categorifying Uq (g), for an arbitrary simply-laced g. We start with an unoriented graph without loops and multiple edges. Let I be the set of vertices of . The bilinear Cartan

INTRODUCTION form on

[I] is given on the basis elements i, j I by

The algebra U over (q), the negative (or positive) half of the quantum universal enveloping algebra, has generators i , i I, and dening relations i j = j i if i j = 0, 2 2 (q + q )i j i = i j + j i if i j = 1.
1

2 if i = j, i j = 1 if i and j are joined by an edge, 0 otherwise.

The algebra U contains a subring A f , which is the [q, q 1]-lattice generated (a) by all products of quantum divided powers i . The canonical basis B is a basis of A f viewed as a free [q, q 1 ]-module. In Section 2 of this paper to each graph as above we assign a family of graded rings R(), over [I]. The rings are dened geometrically, via braid-like plane diagrams which consist of interacting strings labelled by vertices of the graph. We prove basic results about these rings, then switch from the ground ring to a eld k to simplify the study of R()-modules. We show that the representation theory of R() categories the integral form A f of U . We consider the category R()pmod of nitely-generated graded left projective R()-modules and its Grothendieck group K0 (R()). Let R = R() and dene

K0 (R) =

[I]

K0 (R()).

Induction and restriction functors coming from the inclusions R()R( ) R( + ) give rise to the multiplication and comultiplication homomorphisms K0 (R) K0 (R) K0 (R), K0 (R) K0 (R) K0 (R) that satisfy the same properties as those for A f. We dene a homomorphism of [q, q 1 ]-algebras : A fK0 (R) that also respects comultiplication and (a ) (a ) takes a divided powers product element = i1 1 . . . ir r to the image of a certain projective module P in the Grothendieck group. The quantum Gabber-Kac theorem implies that is injective. By mirroring for the case of rings R() the methods of Kleshchev, Grojnowski, 3

INTRODUCTION

and Vazirani, who studied socles of induction and restriction applied to irreducible representations, we show that homomorphism is surjective for any graph and any eld k. The main result of the paper is the following theorem. Theorem 1.1. : bialgebras.
A f K0 (R)

is an isomorphism of

[I]-graded twisted

The term twisted bialgebras is used above, since the comultiplication in A f and K0 (R) becomes an algebra homomorphism only after the multiplication in the tensor squares A f 2 and K0 (R)2 is twisted by powers of q. We conjecture that, when is a tree and k = , isomorphism takes canonical basis elements to the images of indecomposable projective modules in K0 (R). When the graph is a single vertex, this conjecture is almost trivial. We verify the conjecture in a simple case of the graph with two vertices and one edge, with all canonical basis elements being monomials. The rings R() should be linked to Lusztigs geometric realization of A f : Conjecture 1.2. For k = and graph a tree, the algebra R() is Morita equivalent to the algebra of equivariant ext groups ExtG (L, L), where G = iI GL(i ) and L is the sum of simple perverse sheaves Lb , over all b B , in Lusztigs geometric realization of A f . This conjecture should follow from an isomorphism between R() and a suitable convolution algebra. When contains cycles, its possible to modify R() by introducing monodromies around the cycles, and the conjecture is likely to hold for a modied version of R(). Our results hint at the relation between representation theory of ane Hecke algebras for GL(n) when q is not a root of unity and representations of R() when the graph is a chain. We conjecture that completions of ane Hecke algebras along suitable maximal central ideals are Morita equivalent (or even isomorphic) to completions of R() along the grading. The above conjectures, if true, would link Lusztigs geometrization of U with Arikis categorication of U and its restricted dual for q = 1 and = An . We arrived at the denition of rings R() from computations involving homomorphisms of bimodules over cohomology rings of partial ag varieties. The bimodules themselves are the cohomology groups of partial and iterated ag varieties that give correspondences for the action of generators ei and 4

RINGS R() AND THEIR PROPERTIES

fi of quantum slN in the Beilinson-Lusztig-MacPherson geometric model [4] of quantum slN . This model was given a categorical interpretation by Grojnowski and Lusztig [15] and later reinterpreted, for N = 2, via translation and Zuckerman functors in [6], with various generalizations constructed in [13], [44], [48], and in a very recent striking work [49]. In Section 3.4 we dene certain quotient algebras of R() and conjecture that their categories of modules categorify irreducible integrable representations of Uq (g). A straightforward generalization of our constructions and results from algebras R() and their quotient algebras in the simply-laced case to that of an arbitrary symmetrizable Kac-Moody algebra g will be presented in a follow-up paper. In the paper [9] (see also [41]), Chuang and Rouquier dened sl(2)categorications, substantiated them with many diverse examples, and applied to the modular representation theory. Partially inspired by [9] and [10], the second author suggested and investigated a categorication of Lusztigs idempotent completion U of quantum sl(2) in [28], [29]. A denition of a categorication of Uq (g) for any simply-laced g can be obtained by combining the diagrammatic relations of R() with those of U-categorication in [28]. R. Rouquier [42], in his recent talk, dened sl(N)- and ane sl(N)categorications and outlined a conjectural program that aims to vastly generalize his prior work with J. Chuang [9] on sl(2)-categorications. We expect Rouquiers and our approaches to be closely related. R. Rouquier informed us that a signed version of rings R() appears in his categorication [43] of U(g) for a simply-laced g. Acknowledgments: This paper was written while the rst author was at the Institute for Advanced Study, and he would like to thank the Institute for its hospitality and the NSF for fully supporting him during that time through the grant DMS-0635607. Additional partial support came from the NSF grant DMS-0706924.

2
2.1

Rings R() and their properties

Denitions

We x a graph , not necessarily nite, with set of vertices I and unoriented edges E . We require that has no loops and multiple edges. By [I] we denote the commutative semigroup freely generated by vertices of and for

DEFINITIONS

[I] write
=
iI

i i , i

, = {0, 1, 2, . . . }.

(2.1)

Let || = i , and Supp() = {i | i = 0}. We dene a bilinear form on [I] by i i = 2, i j = 1 if i and j are connected by an edge, and i j = 0 otherwise. In the basis {i}iI of vertices the bilinear form is given by the Cartan matrix of . To we associate a diagrammatic calculus of planar diagrams. We consider collections of arcs on the plane connecting m points on one horizontal line with m points on another horizontal line. The position of m points on the line is xed once and for all (for instance, we could take points {1, 2, . . . , m} ). Arcs have no critical points when projected to the y-axis of the plane (a condition reminiscent of braids). Each arc is labelled by a vertex of . Arcs can intersect, but no triple intersections are allowed. An arc can carry dots. An example of such a diagram is shown below,

i j i

(2.2)

where i, j, k are vertices of and the label of an arc is written at the bottom end of the arc. We allow isotopies that do not change the combinatorial type of the diagram and do not create critical points for the projection onto the z-axis.

i j i k k i j i k k

We proceed by allowing nite linear combinations of these diagrams with

DEFINITIONS integral coecients, modulo the following local relations 0 if i = j, if i j = 0, = j i j i + if i j = 1.

i j i j

(2.3)

i j

=
i

j i

=
i

for i = j

(2.4)

i i

=
i i

(2.5)

i i

=
i i

(2.6)

=
i j k i j k

unless i = k and i j = 1

(2.7)

i j i i j i

=
i j i

if i j = 1

(2.8)

Fix [I]. Let Seq() be the set of all sequences of vertices i = i1 . . . im where ik I for each k and vertex i appears i times in the sequence. The 7

DEFINITIONS

length m of the sequence is equal to || and the cardinality of Seq() is equal , taken over all i I. For instance, to i , j , . . . Seq(2i + j) = {iij, iji, jii}. Each diagram D as described above determines two sequences bot(D) and top(D) in Seq() for some . The sequence bot(D) is given by reading the labels of arcs of D at the bottom position from left to right. We dene top(D) likewise. For instance, for the diagram in (2.2), bot(D) = ijik and top(D) = jiki. We often abbreviate sequences with many equal consecutive terms, and write in1 . . . inr for i1 . . . i1 i2 . . . i2 . . . ir . . . ir , where n1 + + nr = m. 1 r Dene the ring R() as follows: R() =
i ,j Seq() j R()i

(2.9)

as an abelian group, where j R()i is the abelian group of all linear combinations of diagrams with bot(D) = i and top(D) = j modulo the relations (2.3)(2.8). The product in R() is given by concatenation (see the left diagram in (2.11) below)
k R()j

j R()i k R()i ,

(2.10)

and xy = 0 for x l R()k and y j R()i if k = j .

k

...
i1 i2 im

(2.11)

By construction, R() is an associative ring. For each i Seq() the diagram 1i i R()i shown on the right of (2.11) is an idempotent, 12 = 1i , i x1i = x for all x j R()i and 1i x = x for all x i R()j , for all j . Furthermore, 1 = i Seq() 1i is the unit element of R(). We turn R() into a graded ring by declaring degrees of the generators to be deg
i

= 2,

deg 8

= i j.

(2.12)

DEFINITIONS Let Pi =
j Seq() j R()i , jP

=
i Seq()

j R()i .

(2.13)

Pi is a left graded projective R()-module and j P is a right graded projective R()-module. Flipping a diagram about a horizontal axis induces a gradingpreserving antiinvolution of R() which takes j R()i to i R()j and 1i to 1i . Flipping a diagram about a vertical axis and simultaneously taking to
i i

i i

(in other words, multiplying the diagram by (1)s where s is the number of times equally labelled strands intersect) is an involution of R() which commutes with . Sometimes it is convenient to convert from graphical to algebraic notation. For a sequence i = i1 i2 . . . im Seq() and 1 k m we denote xk,i :=
i1

... ...
ik im

(2.14)

with the dot positioned on the k-th strand counting from the left, and k,i :=
i1

...
ik ik+1

...
im

(2.15)

Other dening relations for R() can be similarly rewritten.

The symmetric group Sm acts on Seq(), m = || by permutations. Transposition sk = (k, k + 1) switches entries ik , ik+1 of i . Thus, k,i sk (i ) R()i . The relations (2.3) become 0 if ik = ik+1 . 1i if ik ik+1 = 0 k,sk (i ) k,i = (2.16) xk,i + xk+1,i if ik ik+1 = 1

EXAMPLES

10

2.2

Examples
, with the unit element given by the empty

1) = 0. We have R(0) = diagram.

2) = i for some vertex i. Then a diagram is a line with some number of dots on it. Hence, R(i) [x1,i ], where in our notation x1,i denotes a line = labelled i with one dot on it.
i

3) = mi for some vertex i. The only sequence in Seq(mi) is im = ii . . . i. Every strand in the diagram is labelled by i, and the local relations are

= 0
i i i i i

=
i i i

(2.17)

i i

=
i i

(2.18)

i i

=
i i

(2.19)

R(mi) is isomorphic to the nilHecke ring NHm , which is the unital ring of endomorphisms of the abelian group [x1 , . . . , xm ] generated by the endomorphisms of multiplication by x1 , . . . , xm and divided dierence operators f (x) sa f (x) a (f (x)) = , 1 a m 1, xa xa+1 where sa transposes xa and xa+1 in the polynomial f (x). The dening relations are xa xb = xb xa , a xb = xb a if |a b| > 1, a b = b a if |a b| > 1, 2 a = 0, a a+1 a = a+1 a a+1 , xa a a xa+1 = 1, a xa xa+1 a = 1. 10

EXAMPLES

11

In the above equations xa stands for the operator of multiplication by xa . The dening relations are exactly our graphical equations on identicallycolored strands, with the relations in the rst two rows corresponding to isotopies of diagrams. The nilHecke ring is related to the theory of Schubert varieties, see [26], [7]. The nilCoxeter ring is the subring of NHm generated by 1 , . . . , m1 , see [38, Chapter 2], [21]. Divided dierences go back to Newton; in the context of representation theory they appeared in [5], [11]. The center of NHm is the ring of symmetric polynomials in x1 , . . . , xm , and NHm is isomorphic to the ring of m! m! matrices with coecients in Z(NHm ), see [36]. Here we consider NHm as a graded ring, with deg(a ) = 2 and deg(xa ) = 2. The graded nilHecke ring plays a fundamental role in the categorication of Lusztigs quantum sl2 dened by the second author [28]. For each permutation w Sm let w = a1 . . . ar , where sa1 . . . sar is a minimal presentation of w, so that r = l(w). This element does not depend on the choice of presentation. Dene em = xm1 xm2 . . . xm1 w0 , where w0 is the longest permutation. 1 2 This element is an idempotent of degree 0. We will also use the idempotent (em ) given by reecting the diagram of em about the horizontal axis, (em ) = w0 xm1 xm2 . . . xm1 . 1 2 NHm (em ) is a left NHm -module isomorphic to the polynomial representation of NHm . The polynomial representation is the unique, up to isomorphism and grading shifts, graded indecomposable projective NHm module. The module NHm (em ) is nontrivial in even non-negative degrees only. The regular representation of NHm decomposes as the sum of m! copies of the polynomial one. Taking the grading into account and denoting by Pm the module NHm (em ) with the grading shifted down by m(m1) , we 2 get a direct sum decomposition of graded modules NHm Pm . = [m]!
q Here [m]! = [m][m 1] . . . [1] is the quantum factorial, [m] = q qq1 , and M f or M f , for a graded module M and a Laurent polynomial f =
m m

11

EXAMPLES fa q a M{a}. [q, q 1 ], denotes the direct sum over a

12 , of fa copies of

We denote by Pi,m the corresponding indecomposable graded projective module over R(mi) and by ei,m the idempotent corresponding to em under the isomorphism R(mi) NHm . As a graded abelian group, Pi,m is = m(m1) nontrivial in degrees 2 +2 .

Letting m P be the right graded projective module em NHm { m(m1) }, we 2 have a decomposition of graded right NHm -modules NHm m P [m]!. = We denote by i,m P the corresponding indecomposable graded projective right R(mi)-module. Idempotents ei,m and (ei,m ) have the following diagrammatic presentation for m = 3 ei,3 =
i i i

(ei,3 ) =
i

(2.20)
i

The quotient of [x1 , . . . , xm ] by the ideal of symmetric polynomials is a representation Lm of NHm which becomes irreducible upon tensoring with any eld k. Over k, any graded irreducible representation of NHm is isomorphic to Lm , up to a grading shift. We denote the corresponding irreducible representation of R(mi) by L(im ). Its nonzero in degrees 0, 2, . . . , m(m1) . The representation L(im ) is isomorphic to the 2 representation induced from the one-dimensional graded module L over k[x1 , . . . , xm ] (on which x1 , . . . , xm necessarily act trivially). Lemma 2.1. The common 0-eigenspace of operators x1 , . . . , xm on L(im ) = m Ind(L) is exactly 1 L. All Jordan blocks of xm on L(i ) are of size m. Proof. Due to the uniqueness of the irreducible module L(im ), it is isomorphic to the module induced from the trivial representation of the subring of NHm generated by the divided dierences 1 , . . . , m and symmetric polynomials in x1 , . . . , xm . This induced representation is isomorphic, as a k[x1 , . . . , xm ]-module, to the quotient of k[x1 , . . . , xm ] by the ideal generated by symmetric polynomials without the constant term, and to 12

EXAMPLES

13

the cohomology ring of the full ag variety. The lemma follows from the standard facts about this quotient ring. 4) = i+j and ij = 0. Seq(i+j) = {ij, ji}. The ring R() is isomorphic to the ring of 2 2 matrices with coecients in [x1 , x2 ]. The isomorphism is given on generators by
i j

1 0 0 0
j i

j i

0 0 0 1 0 1 0 0 x2 0 0 0

i j

0 0 1 0 x1 0 0 0
i

i j

5) = i1 + + im and ik i = 0 for all k = . Then R() is isomorphic to the ring of m! m! matrices with coecients in [x1 , . . . , xm ]. To see the isomorphism, enumerate the rows and columns by elements of Seq() and send the element j 1i to the elementary (j , i )-matrix. 6) = + such that i j = 0 for any i Supp( ) and j Supp( ). || In this case R() is isomorphic to the matrix algebra of size | |, | | with coecients in R( ) R( ). Indeed, except for crossings, there are no interactions between strands from and strands from . A pair i Seq( ), j Seq( ) denes the sequence i j Seq( + ) and
i j Ri j

R( ) R( ), =

(2.21)

since we can pull apart the i and j strands in any diagram D with i j = top(D) = bot(D), using that ik j = 0, for all k and . 7) = i + j and i j = 1. We can identify R(i + j) with the ring of 2 2 matrices with coecients in [x1 , x2 ] such that the bottom left coecient

13

EXAMPLES is divisible by x1 + x2 :
i j

14

1 0 0 0
j i

j i

0 0 0 1 0 1 0 0 x2 0 0 0

i j

0 0 x1 + x2 0 x1 0 0 0
i

i j

Remark 2.2. If i j = 1 then the elements

and
i j i

i j i

(2.22)

are mutually orthogonal idempotents in R(2i + j). For instance,

i j i

(2.3)

(2.3)

(2.5)

+
i j i

(2.3)

(2.23)
i j i

14

A FAITHFUL REPRESENTATION AND A BASIS OF R()

15

Equation (2.8) can be viewed as a decomposition of the idempotent 1iji into the sum of two orthogonal idempotents. Equation (2.7) can be thought of as allowing triple intersections for certain ijk.

2.3

A faithful representation and a basis of R()

Action of R() on the sum of polynomial spaces. Choose an orientation of each edge of . For each [I] we dene an action of R() on the free abelian group

Po =
i Seq()

Poi ,

Poi = [x1 (i ), x2 (i ), . . . , xm (i )],

m = ||.

Its useful to think of the variable xk (i ) as labelled by the vertex of in the k-th position in the sequence i . The symmetric group Sm acts on Po by taking xa (i ) to xw(a) (w(i )), w Sm . The transposition sk maps xa (i ) to xa (sk (i )) if a = k, k + 1, xk (i ) to xk+1 (sk i ), and xk+1 (i ) to xk (sk i ). To dene the action of R(), we rst require that an element x j R()i acts by 0 on Pok if k = i and takes Poi to Poj . We describe the action of the generators. The dot in the k-th position xk,i (see (2.14)) acts by sending f Poi to xk (i )f Poi . The idempotent 1i act by the identity on Poi . The crossing k,i (see (2.15)) acts on f Poi by if ik ik+1 = 0, f sk f f if ik = ik+1 , xk (i ) xk+1 (i ) f sk f if ik ik+1 , f (xk (sk i ) + xk+1 (sk i ))(sk f ) if ik ik+1 . Notation ik ik+1 means that ik ik+1 = 1 and this edge of is oriented from ik+1 to ik . Note that when ik = ik+1 the crossing k,i acts by the divided dierence operator. When all strands have the same label i, the action reduces to the action of the nilHecke algebra on its polynomial representation. Proposition 2.3. These rules dene a left action of R() on Po . Proof. We check the dening relations for R(). The relation (2.3) with ij = 0 and ij = 1 and relation (2.4) are trivial to verify. The relation (2.3) with i = j and relations (2.5), (2.6), and (2.7) for i = j = k are just 15 f sk f

A FAITHFUL REPRESENTATION AND A BASIS OF R()

16

the nilHecke relations. The relation (2.7) with i, j, k all distinct, or with i = j = k, i j = 0, or with i = j = k, j k = 0 is easy to check. The same relation with i = j, i k = 1 or j = k, i j = 1 follows from the fact that the divided dierence operator annihilates symmetric polynomials. This leaves us with the last relation (2.8), reproduced below

i j i i j i

=
i j i

if i j = 1.

Its enough to check it on 3-stranded diagrams, with = 2i + j. To simplify formulas, we write x, y, and z instead of xk (i ), xk+1 (i ), and xk+2 (i ) for each i = {iji, iij, jii}. Assume that the ij edge is i j. The left hand side of the relation is 1,jii 2,jii 1,iji 2,iij 1,iij 2,iji , taking Poiji to Poiji . We compute the action of this element on each monomial xu y v z w , u, v, w :

1,jii 2,jii1,iji (xu y v z w ) = 1,jii2,jii (xv y u z w ) = 1,jii =

xv y u z w xv y w z u yz (2.24)

xu y v z w xw y v z u (x + y), xz 2,iij 1,iij 2,iji (xu y v z w ) = 2,iij 1,iij ((y + z)xu y w z v ) xu y w+1 xw+1 y u v xu y w xw y u v+1 = 2,iij z + z xy xy xu y w+1z v xw+1 y u z v + xu y w z v+1 xw y u z v+1 = 2,iij xy u v w+1 w+1 v u u v+1 w x y z x y z + x y z xw y v+1 z u . = xz

(2.25)

One can easily verify that the dierence of (2.24) and (2.25) is xu y v z w , prov-

16

A FAITHFUL REPRESENTATION AND A BASIS OF R() ing relation (2.8) in this case. When i j, we compute

17

1,jii 2,jii1,iji (xu y v z w ) = 1,jii2,jii ((x + y)xv y u z w ) y uzw y w zu y u+1z w y w z u+1 = 1,jii xv+1 + xv yz yz v+1 u w v+1 w u v u+1 w x y z x y z + x y z xv y w z u+1 = 1,jii yz u v+1 w w v+1 u u+1 v w x y z x y z + x y z xw y v z u+1 = , (2.26) xz xu y w z v xw y u z v 2,iij 1,iij 2,iji (xu y v z w ) = 2,iij 1,iij (xu y w z v ) = 2,iij xy u v w w v u x y z x y z = (y + z). (2.27) xz Again, the dierence of (2.26) and (2.27) is xu y v z w . Relation (2.8) and Proposition 2.3 follow. A spanning set. We look for a lower bound on the size of R(). An element of this ring is a linear combination of diagrams. If a diagram D contains two strands that intersect more than once, relations (2.3)(2.8) allow us to write D as a linear combination of diagrams, each with fewer intersections than D. Iterating, we can write any element of R() as a linear combination of diagrams with at most one intersection between any two strands. Furthermore, we can slide all dots in a diagram D all the way to the bottom of the diagram at the cost of adding a linear combination of diagrams with fewer crossings than D. These two operations together tell us that R() is spanned by diagrams having all dots at the bottom and with each pair of strands intersecting at most once:

Such D is determined by i = bot(D), a minimal presentation w = sk1 . . . skr , r = l(w) of a permutation w Sm and the number of dots at 17

A FAITHFUL REPRESENTATION AND A BASIS OF R()

18

each bottom endpoint of D. The dierence of two diagrams given by the same data except for dierent minimal presentations of w can be written as a linear combination of diagrams with fewer crossings than each of the original two diagrams. For each w Sm x its minimal presentation w. For i , j Seq() let j Si be the subset of Sm consisting of permutations w that take i to j via the standard action of permutations on sequences, dened earlier. For each w j Si we convert its minimal presentation w into an element of j R()i denoted wi . Denote the subset {wi }wj Si of j R()i by j Si . Example 2.4.
iji Siji

= {id, (13)}, and

idiji =
i j i

(13)iji =
i j i

or
i j

,
i

depending on whether we choose s2 s1 s2 or s1 s2 s1 as a minimal presentation of permutation (13). In general, j Si depends on our choices of minimal presentations for permutations. For instance, in the above example, (13)iji ijiSiji will depend nontrivially on whether the presentation s1 s2 s1 or s2 s1 s2 was chosen if i j = 1. Let j Bi be the set {y xu1 . . . xum } over all y j Si and ui . Here the m,i 1,i diagrams in j Si are multiplied by all possible monomials at the bottom. For example, the sets ij Bij and ji Bij consist of elements

u1

u2
i j

and

u1 i

u2
j

respectively, where we write

u u

Theorem 2.5. j R()i is a free graded abelian group with a homogeneous basis j Bi . 18

A FAITHFUL REPRESENTATION AND A BASIS OF R()

19

Proof. We have already observed that the set j Bi spans j R()i . This set consists of homogeneous elements relative to our grading on R(). To prove linear independence of elements of j Bi we check that they act on Po by linearly independent operators. Let j 1i be the diagram with the fewest number of crossings with bot(j 1i ) = i and top(j 1i ) = j . For example,
j jjiki1ijkij j i k i

=
i j k i j

Note that identically coloured lines do not intersect in j 1i , and that i 1i is just 1i . The product i 1j j 1i = (xa,i + xb,i ) where the product is over all pairs 1 a < b m such that the lines in j 1i ending at a and b bottom endpoints counting from the left intersect and are coloured by i, j with i j = 1. For instance, if i = ijj, j = jji with i j = 1, then
j j 1i j i

=
i j j

(2.28)

and the product

i 1j j 1i

=
i j j

(x1,i + x2,i )(x1,i + x3,i ).

(2.29)

Choose a complete order on the set of vertices of and orient so that for each edge i j we have i < j relative to the order. This order induces a lexicographic order on Seq(). We prove linear independence of j Bi by induction on j Seq() with respect to this order. 19

A FAITHFUL REPRESENTATION AND A BASIS OF R() Base of induction: We write

r

20

j = j1 . . . j1 j2 . . . j2 . . . jr . . . jr =

j1 1 j2 2

. . . jr r

Seq(),

=
k=1

k jk ,

(2.30) where j1 < j2 < < jr and r is the cardinality of Supp(). Clearly j is the lowest element in Seq() with respect to lexicographic order. For this j each w j Si can be written uniquely as w = w1 w0 where w1 S1 Sr and w0 is the unique minimal length element in j Si . Each minimal length representative w0 determines the same w0i = j 1i . Likewise, the element w1j does not depend on the choice of a minimal length representative w1 , since in the nilHecke algebra the element associated to a permutation does not depend on the minimal presentation of this permutation.
j1 j1 j1 j2 j2 jr jr jr

w1

w0

The set j Bi consists of elements w1j w0i xu , over all u xu = xu1 xu2 . . . xum . m,i 1,i 2,i It suces to check that induced maps w1j w0i xu : Poi Poj

m, where

(2.31)

are linearly independent, over all u m . Indeed, w0i xu takes xv Poi to xw0 (u+v) Poj , where w0 acts on v via the obvious permutation. This is due to peculiarities of our action, since the element k,i takes f Poi to sk f if ik > ik+1 . Elements w1j act on the monomials by products of divided dierence operators. Its known that the standard action of the nilHecke ring on polynomials is faithful [36], implying linear independence of all maps (2.31). 20

A FAITHFUL REPRESENTATION AND A BASIS OF R()

21

Induction step: Assume we proved that j Bi is independent, that jk < jk+1 , and set j = sk j = j1 . . . jk1 jk+1 jk jk+2 . . . jm . It suces to assume that jk jk+1 = 1, for otherwise jk jk+1 = 0 and the maps k,j , k,j set up bijections between j Bi and j Bi , implying linear independence of j Bi . To show that j Bi is independent, we examine its image under the map k,j : j R()i j R()i . (2.32) Dene a partial order on j Bi by requiring that w1 xu < w2 xv if (w1 ) < (w2 ), or if w1 = w2 , u1 = v1 , . . . , ut = vt , ut+1 < vt+1 for some t. Extend this partial order to a complete order on j Bi in some way. Dene the map : j Bi j Bi by y = k,j y if the strands of diagram y ending at the top endpoints numbered k, k + 1 from the left are disjoint
i1 ik ik+1 i1 ik ik+1

and y = y x,i if these two strands of y intersect. Here is the number, counting from the left, of the bottom endpoint of the strand with top endpoint k, and y is obtained from y by removing the intersection of these two strands. Graphically
i1 ik ik+1 i1 ik ik+1

We can write y = sk wi xu x,i if y = wi xu as above. The map : j Bi j Bi is clearly injective. It is not hard to compute that k,j y = y+ lower terms: (2.33) k,j y = y + nz z, nz , z j Bi ,
z<y

for any y j Bi , and by lower order terms we mean a linear combination of elements of j Bi less than y with respect to the order of j Bi . Induction step follows, since j Bi is a linearly independent set by induction hypothesis. This completes the proof of Theorem 2.5. 21

PROPERTIES OF R()

22

The representation Po has a grading. Choose any i Seq() and place the unit element 1 Poi in degree 0. This uniquely determines a grading on Po making it a graded module over the graded ring R(). Corollary 2.6. Po is a faithful graded module over the graded ring R().

2.4

Properties of R()

From Theorem 2.5 we deduce several properties of R(). For each i Seq() the subring i R()i contains the polynomial ring Po(, i ) [x1,i , x2,i , . . . , xm,i ]. = We dierentiate between the ring Po(, i ) and the abelian group Poi on which we dened the action. The direct product Po() =
i Seq()

Po(, i )

(2.34)

is a commutative subring of R(). Proposition 2.7. R() is a free Po()-module of rank m! with respect to both left and right multiplication actions of Po(). Proof. For each permutation w Sm choose a minimal representative w and form w= wi .
i Seq()

Theorem 2.5 implies that the set {w}wSn is a basis of R() as a free graded module over Po() under the right multiplication action of the latter. The left multiplication case follows by applying the antiinvolution of R(). The symmetric group Sm acts on Po() by permuting strands (which carry labels and dots). Let Sym() = Po()Sm be the subring of Sm invariants. Its naturally isomorphic to the tensor product of rings of symmetric polynomials Sym() =
iSupp()

[x1 , . . . , xi ]Si ,

(2.35)

over vertices i in Supp(), with the number of variables i .

22

PROPERTIES OF R() Example 2.8. The following elements are generators of Sym(2i + j): 1=
i i j

23

+
i j i

+
j i i

i i j

+
i

i j

+
i j i

+
i j

+
i j

i i

+
j i

i i j

+
i j

+
i j

i i

+
i i j i

j i

+
j i i

Consider the inclusions of rings Sym() Po() R(). (2.36)

Each subsequent ring is a free rank m! module over the previous ring. Therefore, R() is a free module of rank (m!)2 over Sym(). Moreover, a simple computation shows that Sym() belongs to the center of R(). The converse is true as well. Theorem 2.9. Sym() is the center of R(). Proof. The multiplication map
i R()k

j R()k j 1i y

by j 1i is injective, since the composition i 1j j 1i is injective, being a certain product of sums of xa,i s (use Theorem 2.5). A central element z Z(R()) decomposes z=
i Seq() In particular, zi is a central element of i R()i . Let j = j1 1 j2 2 . . . jr r , for some order j1 . . . jr of vertices that appear in . The ring j R()j is the tensor product of nilHecke rings r

zi ,

zi = 1i z = z1i .

(2.37)

j R()j =
t=1

NHt

(2.38)

23

PROPERTIES OF R()

24

and its center is isomorphic to the tensor product of centers of nilHecke rings, which are known to be symmetric polynomials in t variables. Moreover, the composition Sym() // Z(R()) z // Z(j R()j ) // zj (2.39)

is an isomorphism. Subtracting an element of Sym(), we can assume that a central element z has zj = 0. Since for all i 0 = zj (j 1i ) = z(j 1i ) = (j 1i )z = (j 1i )zi , we get zi = 0 since the multiplication by j 1i is injective. Corollary 2.10. 1) R() is a free rank (m!)2 module over its center Sym(). 2) R() is free as a graded module over Sym(). The ring Sym() is + -graded. Any nitely-generated free graded Sym()module has a graded rank invariant which lies in [q, q 1]. The graded rank of the module Sym(){a} whose grading starts in degree a is q a , and graded rank is additive under direct sum. Its not hard to write a combinatorial formula for the graded rank of R(); we leave it to the reader as an exercise. (2.40)

Corollary 2.11. 1) R() is both left and right Noetherian. 2) R() is indecomposable. Indecomposability is equivalent to 1 being the only central idempotent in the ring. Note that R() is almost positively graded. Precisely, it is zero in degrees less that i i (i 1).

24

REPRESENTATIONS

25

2.5

Representations

In this and the following sections we assume that R() is dened over a eld k rather than over . All earlier results about R() remain valid over k. We view R() as a graded k-algebra with every element of k in degree 0. Let R()mod be the category of nitely-generated graded left R()-modules, R()fmod be the category of nite-dimensional graded R()-modules, and R()pmod be the category of projective objects in R()mod. The morphisms in each of these three categories are grading-preserving module homomorphisms. The rst two categories are abelian. We have a diagram of categories and inclusions R()fmod R()mod R()pmod. From now on, by an R()-module we mean a left graded nitely-generated R()-module, unless otherwise specied. For any two R()-modules M, N denote by Hom(M, N) or HomR() (M, N) the k-vector space of gradingpreserving homomorphisms, and by HOM(M, N) :=
a

Hom(M, N{a})

(2.41)

the -graded k-vector space of all R()-module morphisms. Here N{a} denotes N with the grading shifted up by a. By a simple R()-module we mean a simple object in the category R()mod. We denote by Sym+ () the unique graded maximal central ideal of Sym(). It is spanned by Sm invariant polynomials without the constant term. Proposition 2.12. A simple R()-module S is nite-dimensional and Sym+ () acts by 0 on it. Hom(S, S{a}) = 0 if a = 0, and S remains simple when viewed as an S-module without the grading. Proof. Obvious. Hence, S is a (graded) module over the nite-dimensional quotient algebra R () = R()/Sym+ ()R(). (2.42)

Note that dimk R () = (m!)2 , and, up to isomorphism and grading shifts, there are only nitely many simple R()-modules. We choose one representative Sb from each equivalence class, denote the set of equivalence classes 25

REPRESENTATIONS by B , and dene

26

B =

def [I]

B .

We expect a bijection between B and Lusztig-Kashiwara canonical basis B, hence use a similar notation. Thus, any simple R()-module is isomorphic to Sb {a} for a unique b B and a (recall that we are considering only graded modules). We do not specify the grading shift for Sb yet (but see the end of Section 3.2). Each module in R()fmod has nite length composition series with subsequent quotientssimple modules. The Grothendieck group G0 (R()) of R()fmod is a free [q, q 1 ]-module with the basis {[Sb ]}bB and the multiplication by q corresponding to the grading shift up by 1. The abelian category R()mod has the Krull-Schmidt unique direct sum decomposition property for modules. Objects Pi , i Seq(), belong to its subcategory R()pmod of projective modules. Each simple Sb has a unique (up to isomorphism) indecomposable projective cover, denoted Pb . We have HOM(Pb , Sb ) Hom(Pb , Sb ) End(Sb ). An = = indecomposable object of R()pmod is isomorphic to Pb {a} for a unique b B and a . Any object of R()pmod has a unique, up to isomorphism, direct sum decomposition into indecomposables. The Grothendieck group of R()pmod is a free [q, q 1 ]-module with the basis {[Pb ]}bB given by the images [Pb ] of indecomposable projectives. Denote this Grothendieck group by K0 (R()). Recall that for a right, respectively left, R()-module M we denote by M the left, respectively right R()-module M with the action twisted by . For P R()pmod, let P = HOM(P, R()) . This is a graded projective left R()-module and is a contravariant self-equivalence in R()pmod. We have Pi Pi for each i Seq(), and, more generally, Pi {a} Pi {a}. = = This self-equivalence induces a [q, q 1 ]-antilinear involution on K0 (R()), also denoted . There is a [q, q 1 ]-bilinear pairing (, ) : K0 (R()) G0 (R()) [q, q 1 ], ([P ], [M]) := gdimk (P R() M). (2.43) (2.44)

When the eld k is algebraically closed, End(Sb ) k, and the bases {[Pb ]}b = and {[Sb ]}b are dual, possibly up to rescaling by powers of q and permuta26

REPRESENTATIONS

27

tion of elements. In this case G0 (R()) and K0 (R()) are dual free [q, q 1 ]modules. We will show in Section 3.2 that, over any eld k, simples Sb are absolutely irreducible and the above pairing is perfect without any restrictions on k. There is a [q, q 1 ]-bilinear form, also denoted (, ), (, ) : K0 (R()) K0 (R()) [q 1 , q] ()q , where ()q = gdim(Sym()) =
i

(2.45)

1 1 q 2a a=1

(2.46)

and ([P ], [Q]) = gdimk (P R() Q). (2.47) Since P R() Q Q R() P , the form is symmetric. It follows from = Theorem 2.5 that i R()j i P R() Pj is a free graded Sym()-module for = any i , j . Therefore, P R() Q is a free graded Sym()-module of nite rank for any P , Q as above, and the form takes values in [q 1 , q] ()q . We have ([Pj ], [Pi ]) = gdim(j P R() Pi ) = gdim(j R()i ). Dene the character ch(M) of a graded nitely-generated R()-module M as ch(M) = gdim(1i M) i .
i Seq()

The character is an element of the free ((q))-module with the basis Seq(); when M is nite-dimensional, ch(M) is an element of the free [q, q 1 ]module with basis Seq(). We abbreviate gdim(1i M) to ch(M, i ), ch(M) =
i Seq()

ch(M, i ) i .
(n1 ) (n2 ) i2

Let Seqd() be the set of all expressions i1 and r na ia = . For instance, a=1

. . . ir

(nr )

such that n1 , . . . , nr

Seqd(2i + j) = {iij, iji, jij, i(2) j, ji(2) }. To i Seqd() we assign the idempotent 1i = ei1 ,n1 ei2 ,n2 eir ,nr , 27

REPRESENTATIONS

28

given by the tensor product of minimal idempotents ei,n in the nilHecke rings, see Section 2.2. Let i ! = [n1 ]! . . . [nr ]!, and i be the element of Seq() given by expanding i, i = i1 . . . i1 i2 . . . i2 . . . ir . . . ir . i = i i i Seq(). We have the equality of graded dimensions
r

gdim(1bM) = q i i ! gdim(1i M), i

i =
k=1

nk (nk 1) , 2

which follows from the structure of the nilHecke algebra. Let ch(M, i ) = q i gdim(1i M), then ch(M, i ) = i ! ch(M, i ). In particular, ch(M) determines ch(M, i ) for any i Seqd(). For i Seqd() dene the left graded projective module Pi = R()(1i ){ i }, and the right graded projective module
iP

= 1i R(){ i }.

We have Pb Pii! and bP i P i ! . For instance, = i i =

1 [2]! Pii Pi(2) = Piq+q = Pi(2) {1} Pi(2) {1}. = (2)

Moreover, ch(M, i ) = gdim(i P R() M) = gdim(HOM(Pi , M)). (2.48)

Given two or more sequences in Seqd() that dier only in several neighbouring terms, we denote identical parts in them via dots. For instance, . . . ij . . . and . . . ji . . . denote a pair of sequences i iji and i jii for some sequences i , i .

28

REPRESENTATIONS

29

Proposition 2.13. There are isomorphisms of graded projective right R()modules ...,ji...P if i j = 0, = P (2) P (2) if i j = 1, ...iji... P = ...i j... ...ji ...
...ij... P

and isomorphisms of graded projective left R()-modules P...ij... P...iji... P...ji..., if i j = 0, = P (2) P (2) if i j = 1. = ...i j... ...ji ...

Proof. It suces to show the isomorphisms for right projective modules; application of the antiinvolution would imply the corresponding isomorphisms for left projective modules. Multiplication by the diagram ...
i j ...ij... P

... and
...ji... P

is a grading-preserving isomorphism between Consider grading-preserving maps B0 : B1 :

...iji... P ...i(2) j... P

if i j = 0.

...ji(2) ... P ...i(2) j... P ...ji(2) ... P ...iji... P

given by matrices of diagrams

Notice that the top entry in B0 ends with the projector e2,i which takes 1...ii... to 1...i(2) ... , and we view this entry as a homomorphism
...iji... P ...iij... P ...i(2) j... P,

B0 =

B1 =

i j j i i

(2.49)

29

REPRESENTATIONS

30

ditto for the bottom entry in B0 . The degree of each diagram in B0 is 1, therefore the map B0 is grading-preserving since the grading shifts for the sequences dier by . . . i(2) j . . . . . . iji . . . = 1. We view the rst entry in B1 as the composition
...i(2) j... P

...iij...P ...iji...P,

and there is no need to write the corresponding idempotent (same for the second entry). We compute i j i j i i i i j B0 B1 = = j i i i i j j i i
i i j j i i

(2.3)

In the last matrix above the diagonal terms are idempotents 1i(2) j and 1ji(2) , giving identity maps of projectives ...i(2) j... P and ...ji(2) ... P , respectively.
j i

i i 0

0
j

(2.3),(2.5)

i i j

0 0
j

i i

B1 B0 =

i j i

= i

+
j i i j

30

INDUCTION AND RESTRICTION

31

(2.3)

i j i

+
i j i

(2.8)

Therefore, B0 , B1 are isomorphisms, and the second isomorphism follows. Corollary 2.14. For any M in R()mod there are isomorphisms of graded vector spaces 1...ij... M 1...iji...M{1} 1...ji...M if i j = 0, = 1 (2) M 1 (2) M if i j = 1. = ...i j... ...ji ...

Corollary 2.15. The following character equalities hold for any graded nitely-generated R()-module M ch(M, . . . ij . . . ) = ch(M, . . . , ji . . . ) if i j = 0, ch(M, . . . iji . . . ) = ch(M, . . . i(2) j . . . ) + ch(M, . . . ji(2) . . . ) if i j = 1, a+b ch(M, . . . i(a) i(b) . . . ) = ch(M, . . . i(a+b) . . . ). a

2.6

Induction and restriction

Suppose we have an inclusion of rings : B A which is not necessarily unital: e = (1) is only an idempotent in A. The induction functor between categories of unital modules Bmod Amod,
Ind

M A B M

is isomorphic to the functor M Ae B M. Its right adjoint Res : AmodBmod takes M to eM, viewed as a B-module. The inclusion of graded rings , : R() R( ) R( + ) is described by putting the diagrams next to each other. It takes the idempotent 1i 1j to 1i j and the unit element to an idempotent of R( + ) denoted 1, . 31

INDUCTION AND RESTRICTION

32

Proposition 2.16. 1, R( + ) is a free graded left R() R( )-module. Proof. The minimal representative w of a left S|| S| | -coset in S||+| | gives rise to the diagram i j w i j R( + )w 1 (i j ) of the minimal presentation of w with top ends of strands labelled by the sequence i j for i Seq() and j Seq( ). i1 i2 in j1 j2 jmn

The set of elements w=

i ,j i j w,

over all cosets, is a basis of 1, R( + ) as a free graded left R() R( )module. We denote the restriction and induction functors for the inclusion R() R( ) R( + ) by Res, and Ind, , respectively. Corollary 2.17. The restriction functor Res, takes projectives to projectives. Given a quadruple (, , , ) with + = + , let
, R ,

= 1 1 R( + )1 1 .

Proposition 2.18. Graded (R()R( ), R( )R( ))-bimodule , R , has a ltration by graded bimodules isomorphic to
R, R + ,

, + R , R

{( + )},

where R = R( ) R() R( + ) R( ), over all such that every term above is in [I].

[I]

32

INDUCTION AND RESTRICTION

33

Proof. This proposition is a version of the Mackeys induction-restriction theorem for inclusion of maximal parabolic subgroups Smn Sn Sm . The statement and its proof are best illustrated by the diagram +

These diagrams, over all (summing over all colorings of strands) will provide generators for the subquotient bimodules that appear in the proposition. The grading shift ( + ) is the degree of the intersection diagram of || parallel lines colored by any i Seq() and | + | parallel lines colored by any j Seq( + ). We have Ind, (Pi Pj ) Pi j =

for i Seq(), j Seq( ). By passing to direct summands, we see that the formula holds more generally, for i Seqd(), j Seqd( ). A shue k of a pair of sequences i Seq(), j Seq( ) is a sequence together with a choice of subsequence isomorphic to i such that j is the complementary subsequence. Shues of i , j are in a bijection with the minimal coset representatives of S|| S| | in S||+| | . We denote by deg(i , j , k ) the degree of the diagram in R( + ) naturally associated to the shue, see an example below. k

i 33

QUANTUM GROUPS AND THE GROTHENDIECK RING OF R

34

When the meaning is clear, we will also denote by k the underlying sequence of the shue k . Proposition 2.19. For any k Seq( + ) Res, Pk =
i ,j

Pi Pj {deg(i , j , k )},
iP i j =k

Res, (k P ) =

j P {deg(i , j , k )},

the sum over all ways to represent k as a shue of i Seq() and j Seq( ). The proposition follows immediately from the structure of bimodules R+ and + R, . Given two functions f and g on sets Seq() and Seq( ), respectively, with values in some commutative ring which contains [q, q 1 ], we dene their (quantum) shue product f g (see [30] and references therein) as a function on Seq( + ) given by
,

(f g)(k ) =
i ,j

q deg(i ,j ,k ) f (i )g(j ),

the sum is over all ways to represent k as a shue of i and j . Lemma 2.20. For M R()mod and N R( )mod we have ch(Ind, (M N)) = ch(M) ch(N). Proof. This lemma follows at once from the last proposition and formula (2.48).

3
3.1

Quantum groups and the Grothendieck ring of R

Homomorphism of twisted bialgebras

For a graph , we form the direct sum R=

[I]

R().

34

HOMOMORPHISM OF TWISTED BIALGEBRAS

35

This is a non-unital ring. By various categories of R-modules we will mean direct sums of corresponding categories of R()-modules: Rmod = Rfmod =
def [I] def [I]

R()mod, R()fmod, R()pmod.

Rpmod = The Grothendieck groups K0 (R) =

def [I]

[I]

K0 (R()),

G0 (R) =

[I]

G0 (R())

are the direct sums of Grothendieck groups of rings R(). We extend the pairings (2.43) and (2.45) to K0 (R) and G0 (R) by requiring that subspaces corresponding to dierent s be orthogonal. Induction and restriction functors for the inclusion R() R( ) R( + ), summed over all , , give functors Ind : R RmodRmod, Res : RmodR Rmod

where by RRmod we mean the direct sum of categories R()R( )mod, over all , . These functors restrict to subcategories of nite-dimensional modules and projective modules. Indeed, induction takes projectives to projectives. Restriction, in the case of these inclusions, also takes projectives to projectives, by Proposition 2.19 and the Krull-Shmidt property. Thus, these functors induce maps [Ind], [Res] on Grothendieck groups K0 (R) and G0 (R). Note that [Res] is the sum of maps K0 (R( + ))K0 (R()) K0 (R( )), or G0 (R( + ))G0 (R()) G0 (R( )), over all , ; the tensor products here and further are over [q, q 1 ]. Proposition 3.1. [Ind] turns K0 (R) and G0 (R) into associative unital [q, q 1 ]algebras. [Res] turns K0 (R) and G0 (R) into coassociative counital [q, q 1 ]coalgebras. 35

HOMOMORPHISM OF TWISTED BIALGEBRAS

36

Proof follows from the associativity of induction and restriction. The unit element is given by inducing with the one-dimensional module over R(). The counit is given by restricting to R() and taking the graded dimension. Denote the product [Ind](x1 , x2 ) for x1 , x2 K0 (R) simply by x1 x2 . We equip K0 (R) K0 (R) with the algebra structure via (x1 x2 )(x1 x2 ) = q |x2||x1 | x1 x1 x2 x2 for homogeneous x1 , x2 , x1 , x2 , where |x2 |

(3.1)

[I] is the weight of x2, etc.

Proposition 3.2. [Res] is an algebra homomorphism from K0 (R) to K0 (R) K0 (R) with the above algebra structure. Proof. This follows from Proposition 2.18. Recall the symmetric bilinear pairing (2.45) on K0 (R) taking values in [q, q 1 ] ()q . Proposition 3.3. The pairing (, ) has the following properties 1. (1, 1) = 1, 2. ([Pi ], [Pj ]) = i,j (1 q 2 )1 for i, j I, 3. (x, yy ) = ([Res](x), y y ), for x, y, y K0 (R), 4. (xx , y) = (x x , [Res](y)), for x, x , y K0 (R). Proof. Since 1 = [P ], where is the empty sequence, and P = k as a module over R() = k, the rst statement follows. When i = j, vectors [Pi ] and [Pj ] lie in mutually orthogonal subspaces K0 (R(i)) and K0 (R(j)), so that ([Pi ], [Pj ]) = 0. Also, ([Pi ], [Pi ]) = gdim(i R(i)i ) = gdim(k[x]) = (1 q 2 )1 . Let X R( + )pmod, Y R()pmod, and Y R( )pmod. Then ([X], [Y ][Y ]) = ([X], [Ind, Y Y ]) = gdim(X R(+ ) (+ R, ) R()R( ) Y Y ) = gdim(X (1 1 ) R()R( ) Y Y ) = ([Res, X], [Y ] [Y ]), and statement 3 follows. A similar computation establishes 4. 36

HOMOMORPHISM OF TWISTED BIALGEBRAS

37

We next recall (q)-algebras f and f from [34, Section 1] and A f, the integral form of f (our q is Lusztigs v 1 ). Algebra f is a free associative [I]-graded algebra on generators i . The degree of i is i. The tensor product f f is equipped with a algebra structure using the rule (3.1), and with a coalgebra structure r : f f f, determined by the conditions r(i ) = i 1 + 1 i and r being an algebra homomorphism. f comes equipped with a bilinear form (, ) uniquely determined by the same conditions as the ones in Proposition 3.3, with comultiplication r taking the place of comultiplication [Res] in K0 (R) and i , j taking place of [Pi ], [Pj ]. It has the weight space decomposition

f=

[I]

f .

Let I be the radical of the bilinear form (, ). It is a two-sided ideal of f and one forms the quotient algebra f = f/I, which also has the weight decomposition f . f=
[I]

The bilinear form and the comultiplication r descend to the quotient algebra. f and f come with a (q)-antilinear involution that takes q n to q n and i to i . It is not hard to check that the elements

i j i j for i j = 0 and
2 2 (q + q 1 )i j i i j j i for i j = 1

belong to the ideal I. The quantum version of the Gabber-Kac theorem says that I is generated by these elements over all pairs of vertices i = j of the graph (for instance, see Theorem 33.1.3 in [34]). Dene A f as the [q, q 1]-subalgebra generated by the divided powers (a) i , i I, a .

Proposition 3.4. There is an injective homomorphism of [q, q 1 ]-algebras (a ) (a ) (a ) (a ) : A fK0 (R) that takes i1 1 . . . ik k to [Pi ], where i = i1 1 . . . ik k . This homomorphism converts the comultiplication r of A f into the comultiplication [Res] in K0 (R). It takes the bilinear form on A f to the bilinear form on K0 (R): (x, y) = ((x), (y)). 37

HOMOMORPHISM OF TWISTED BIALGEBRAS The bar-involution of A f goes to the bar-involution of K0 (R) under . Proof. Start with the homomorphism of

38

(q)-algebras fK0(R)(q), where def K0 (R)(q) = K0 (R) [q,q ] (q),

1

dened by the condition that it takes i to [Pi ]. There are equalities in K0 (R)(q) : [Pij ] = [Pji], i j = 0, [Piji] = [Pi(2) j ] + [Pji(2) ], i j = 1, that come from isomorphisms of left projective modules in Proposition 2.13. These equalities match the generators of the ideal I. Therefore, the above homomorphism descends to a homomorphism (q) : fK0 (R)(q) . Under this homomorphism induction of projective R-modules corresponds to the multiplication in A f , so that (q) (i1 . . . ik ) = [Pi1 ...ik ]. Passing to the divided powers shows that (q) (i1 1 . . . ik k ) = [Pi(a1 ) ...i(ak ) ].
1 k

(a )

(a )

The bilinear forms on f and K0 (R)(q) satisfy the same properties, listed earlier, and these properties uniquely determine the form on f. Therefore, the homomorphism (q) respects the bilinear forms: ((q) (x), (q) (y)) = (x, y). Since the bilinear form on f is non-degenerate, homomorphism (q) is injective. The bar-involution on f is q-antilinear and xes each product element i1 . . . ik . The bar-involution on K0 (R)(q) is q-antilinear and xes [Pi ] for each i . Therefore, (q) (x) = (q) (x) for all x f. The image of the restriction of (q) to A f lies in K0 (R), therefore we get a homomorphism : A fK0 (R) by restriction. This homomorphism is injective and satises all the properties stated in the proposition. We will prove in the next section that is an isomorphism. 38

SURJECTIVITY OF

39

3.2

Surjectivity of

In this section we closely follow [25, Chapter 5]; all results there transfer directly to our case. For M in R()mod and i I let i M = (1i 1i )M = i,iR R() M, and, more generally, in M = (1ni 1ni )M = ni,niR R() M. We view in as the functor into the category R( ni) R(ni)mod. There are functorial isomorphisms HOMR() (Indni,niN L(in ), M) HOMR(ni)R(ni) (N L(in ), in M), = (3.2) for M as above and N R( ni)mod. The following lemma is obvious. Lemma 3.5. ch(in M) =
j Seq(ni)

ch(M, j in ) j ,

where we view in M as a module over the subalgebra R( ni) of R( ni) R(ni). Let i (M) = max{n 0|in M = 0}. This number is the length of the longest tail of is in sequences k with 1k M = 0. Lemma 3.6. If M R()mod is irreducible, and N L(in ){r} is an irreducible submodule of in (M) for some 0 n i (M) and r , then i (N) = i (M) n. Proof. This is our analogue of Lemma 5.1.2 of [25] and the proof is essentially the same. Let = i (M). Clearly, i(N) i (M) n. Isomorphisms (3.2) and the irreducibility of M imply that it is a quotient of Indni,ni N L(in ){r}. By exactness of i M, we get that i (M) = 0 is a quotient of i (Indni,ni N L(in )){r}. Hence, the latter module is nonzero, and the inequality i (N) i (M) n follows from the Shue lemma 2.20. 39

SURJECTIVITY OF

40

Lemma 3.7. Suppose N R()mod is irreducible and i (N) = 0. Let M = Ind,ni N L(in ). Then 1. in M N L(in ), = 2. hdM is irreducible and i (hdM) = n, 3. all other composition factors L of M have i (L) < n. Proof. This is the analogue of Lemma 5.1.3 in [25] for algebras R(). 1) is immediate from the Shue lemma and Lemma 3.5. 2) From (3.2) we see that a copy of N L(in ), possibly with a grading shift, appears in in Q for any non-zero quotient Q of M, including direct summands of hdM. Part 1, however, implies that N L(in ) appears only once in in M, so that hdM is irreducible. 3) From part 2) we have in (M) = in (hdM), so that in (L) = 0 for any other composition factor of M, since in is exact. Lemma 3.8. Let M R()mod be irreducible and = i (M). Then i M is isomorphic to N L(i ) for some irreducible N R( i)mod with i (N) = 0. Proof. The proof is identical to that of Lemma 5.1.4 in [25]. Lemma 3.9. Let N R()mod be irreducible and M = Ind,ni N L(in ). Then hdM is irreducible, i (hdM) = i (N) + n, and all other composition factors L of M have i(L) < i (N) + n. Proof. Same as proof of Lemma 5.1.5 in [25]. Proposition 3.10. For any irreducible M R()mod and 0 n i (M), socin M is an irreducible R( ni) R(ni)-module of the form L L(in ) with i (L) = i (M) n. Proof. Same as proof of Theorem 5.1.6 in [25]. The analogue of the Kato theorem in our framework is stated below (this theorem appears in the proof of Theorem 5.1.6). Proposition 3.11. Let be a composition of n. 1. The module L(in ) over the nilHecke algebra R(ni) is the only graded irreducible module, up to isomorphism and graded shifts. 40

SURJECTIVITY OF

41

2. All composition factors of Resn L(in ) are isomorphic to L(i1 ) L(ir ), up to grading shifts, and soc(Resn L(in )) is irreducible. 3. soc(Resn L(in )) L(in1 ), up to a grading shift. = n1 Here Resn denotes the restriction to the parabolic nilHecke subalgebra NH NH1 NHr . The proof in [25] works in this case as well, = with the equivalent of Lemma 4.3.1 being Lemma 2.1. i,i Let ei = Resi i be the functor of composition of i with the restriction from R( i) R(i) to R( i). Then i (M) = max{n 0|en M = 0} i and Res M = ei M. i
iI

Corollary 3.12. Let M R()mod be irreducible with i(M) > 0. Then soc(ei M) is irreducible and i (soc(ei M)) = i (M) 1. Socles of ei M are pairwise non-isomorphic for dierent i I. Proof is the same as for Corollaries 5.1.7 and 5.1.8 in [25]. For an irreducible M R()mod dene
+i ei M := soc(ei M), fi M := hd ind,i M L(i).

(3.3)

The module fi M is irreducible by Lemma 3.9, while ei M is irreducible or 0 by Corollary 3.12, and i (M) = max{n 0|en M = 0}, i(fi M) = i (M) + 1. i In the statements below, isomorphisms of simple modules are allowed to be homogeneous (not necessarily degree-preserving). Lemma 3.13. For an irreducible M R()mod we have socin M (en M) L(in ), = i n n hd ind,ni (M L(i )) fi M. = (3.4) (3.5)

Lemma 3.14. For an irreducible M R()mod the socle of en M is isoi morphic to en M [n]! { n(n1) }. i 2 The proofs are equivalent to those of Lemmas 5.2.1 and 5.2.2 in [25]. 41

SURJECTIVITY OF

42

Lemma 3.15. For irreducible modules M R()mod and N R( + i)mod we have fi M N if and only if ei N M. = = Proof follows that of Lemma 5.2.3 in [25]. Corollary 3.16. Let M, N R()mod be irreducible. Then fi M fi N = if and only if M N. Assuming i (M), i (N) > 0, ei M ei N if and only if = = M N. = The character ch(M) of a nite-dimensional representation M R()mod takes values in [q, q 1]Seq(), the free [q, q 1 ]-module generated by Seq(), and descends to a homomorphism from the Grothendieck group G0 (R()) to [q, q 1 ]Seq(). Theorem 3.17. The character map ch : G0 (R()) [q, q 1 ]Seq() is injective. Equivalently, the characters of irreducible modules (one from each equivalence class up to grading shifts) are linearly independent functions on Seq(). The proof is identical to that of Theorem 5.3.1 in [25]. Note that in our case the character of a nite-dimensional graded module is a function on sequences with values in [q, q 1 ], while in the non-graded case of [25] its a function on sequences taking values in . This discrepancy has no eect on the proof. Passing to the fraction eld (q) of [q, q 1 ] and dualizing the map ch, which then becomes the composition

(q)Seq()f K0(R())(q),
(q)

we conclude that (q) , restricted to weight , is a surjective map of (q)vector spaces. We have already observed that and (q) are injective. By summing over all weights, we obtain the following result. Proposition 3.18. (q) : fK0 (R)(q) is an isomorphism. Therefore, the number of isomorphism classes of (graded) simple R()modules is the same for any eld k. Corollary 3.19. A (graded) irreducible R()-module is absolutely irreducible, for any , k and weight . 42

SURJECTIVITY OF

43

Next, assume that is nite. Choose a total order on I, i(0) < i(1) < i(k 1), k = |I|. For r > k dene i(r) = i(r ) where r is the residue of r modulo k. Fix one representative Sb from each isomorphism class b of irreducible R()-modules, up to grading shifts. Recall that we denoted this set of isomorphism classes by B . For all , to each b B assign the following sequence Yb = y0 y1 . . . of nonnegative integers: y0 = i(0) (M), and let M1 = ey0 M. Inductively, yr = i(r) (Mr ), and Mr+1 = eyr Mr . i(0) i(r) Note that y0 + y1 + = || and only nitely many terms in the sequence are non-zero. Introduce a lexicographic order on sequences of non-negative integers: y0 y1 > z0 z1 . . . if, for some t, y0 = z0 , y1 = z1 , . . . yt1 = zt1 and yt > zt . This order induces a total order on B , by b > c i Yb > Yc . To each sequence Yb = y0 y1 . . . we assign the projective R()-module PYbr associated to the divided powers sequence Ybr = . . . i(2)(y2 ) i(1)(y1 ) i(0)(y0 ) (the order of ys is reversed). Proposition 3.20. HOM(P (Yb), Sc ) = 0 if b < c and HOM(P (Yb ), Sb ) = k. This follows from the previous results and implies that the image [P ] of any (graded) projective R()-module in the Grothendieck group K0 (R()) can be written as a linear combination, with coecients in [q, q 1 ], of images of divided powers projectives [P ], for divided power sequences of the form Ybr . Therefore, : A f K0 (R()) is surjective. Since, is also injective, it is an isomorphism. The case of an innite follows by taking the direct limit of its nite subgraphs. This concludes the proof of the Theorem 1.1 stated in the introduction. It would be interesting to nd out if the theorem remains valid for rings R() over rather than over a eld k. For each divided power i(a) we have the corresponding projective Pi(a) . In(a) duction with this projective is an exact functor, denoted Fi , from R()mod to R( + ai)mod. Summing over all , form the functor Fi
(a)

: RmodRmod.

This functor restricts to the subcategory Rpmod of the category of projec(a ) (a ) tive modules. To any divided power sequence = i1 1 . . . ir r associate the functor (a ) (a ) F = Fi1 1 Fir r

43

SURJECTIVITY OF

44

on Rmod. To a nite sum k uk (k) where uk [q, q 1] and (k) are divided powers sequences associate the direct sum of shifted copies of F(k) :
uk F(k) . k

Theorem 3.21. For any relation uk (k) =

k

v ()

in A f with positive coecients uk , v projectives u P(k)k =

k

[q, q1] there is an isomorphism of

Pv ()

inducing an isomorphism of functors

uk F(k) = k v F () .

This result follows immediately from the earlier ones. We conclude that any relation in A f lifts to an isomorphism of functors. It is natural to view the category Rpmod, as well as the category of induction functors on Rmod it gives rise to, as a categorication of A f , the integral form of the quantum universal enveloping algebra of the negative half of the simply-laced Kac-Moody algebra associated to the graph . The semilinear hom form (, ) on K0 (R) dened by ([P ], [Q]) := gdimHOM(P, Q) is related to the tensor product bilinear form (, ) given by (2.45) via (x, y) = (x, y). Indeed, by surjectivity of , it suces to check this relation for x = [Pi ] and y = [Pj ], in which case both sides are equal to the graded dimension of i R()j . The involution of R() dened in Section 2.1 induces a self-equivalence of R()mod which takes projective Pi(a1 ) ...i(ar ) to Pi(ar ) ...i(a1 ) . The induced
1 r r 1

44

TIGHT MONOMIALS AND INDECOMPOSABLE PROJECTIVES

45

map [] on the Grothendieck group K0 (R) coincides, under the isomorphism (a) , the q-linear anti-involution of A f that xes each i . On the category R()fmod we have the contravariant duality functor which takes a nite-dimensional module M to its vector space dual M twisted by the antiinvolution . This duality functor leaves invariant the character evaluated at q = 1: ch(M )q=1 = ch(M)q=1 . Therefore, the contravariant duality preserves simples, up to overall shift:
Sb Sb {r}. =

Given i Seq(), we have ch(Sb , i ) [q 2 , q 2] or ch(Sb , i ) q [q 2 , q 2 ] for parity reasons (more generally, this is true for any indecomposable object of R()mod and can be used to decompose R()mod into the direct sum of two subcategories). Then ch(Sb , i ) [q 2 , q 2 ] if the same is true for ch(Sb , i ), and ch(Sb , i ) q [q 2 , q 2 ] if ch(Sb , i ) q [q 2 , q 2 ]. Hence, the shift r is an even number. r From now on we redene Sb by shifting its grading by 2 . We have Sb = Sb as graded modules. This normalization of Sb does not depend on the choice of i . The character of Sb is bar-invariant: ch(Sb , i ) = ch(Sb , i ) for all i Seq(), where q = q 1 . Extending the bar-involution to [q, q 1 ]Seq() by i = i , we have ch(Sb ) = ch(Sb ). This canonical (balanced) choice of grading for Sb allows us to x the grading on indecomposable projective Pb so that the quotient map Pb Sb is grading-preserving. In this way we obtain a basis {[Pb ]} in A f which depends only on the characteristic of k. Both the multiplication and the comultiplication in this basis have coecients in [q, q 1 ]. An example below shows this basis to be dierent from the Lusztig-Kashiwara basis when is an odd length cycle and k has any characteristic, and when is a cycle and k has characteristic 2.

3.3

Tight monomials and indecomposable projectives

(a ) (a )

Following Lusztig [35], we say that a monomial = 1 1 . . . k k is tight if it belongs to the canonical basis B of A f. It follows from the properties of the 45

TIGHT MONOMIALS AND INDECOMPOSABLE PROJECTIVES

46

canonical basis that a monomial is tight if and only if (, ) 1 q [q] (or see [39, Proposition 3.1]). Proposition 3.22. If a monomial is tight, the projective module P is indecomposable. Proof. Tightness of implies that HOM(P , P ) is a + -graded k-vector space which is one-dimensional in degree 0. Therefore, any degree 0 endomorphism of P is a multiple of the identity, and P is indecomposable. This argument works even over . Example 3.23. When the graph consists of a single vertex i, the weight (m) space A f mi is a rank one free [q, q 1 ]-module generated by i . The map takes it to [Pi(m) ], the generator of the Grothendieck group K0 (R(mi)). Projective module Pi(m) is indecomposable.
(a) (b) (c) . Tight monomials i j i (a, b, c Example 3.24. Let = (c) (b) (a) b a + c) and j i j (a, b, c , b a + c), with the identication i j

i j

(a) (a+c) (c) i

= j i

(c) (a+c) (a) j ,

constitute the canonical basis B of A f, see [34, Example 14.5.4]. Therefore, images of indecomposable projectives Pi(a) j (b) i(c) , b a + c and Pj (c) i(b) j (a) , b > a + c, constitute a basis in the free [q, q 1 ]-module K0 (R). Any indecomposable projective in Rmod is isomorphic to one of the above, up to a grading shift. Indecomposables Pi(a) j (a+c)i(c) and Pj (c) i(a+c) j (a) are isomorphic. /// be a cycle with n 3 vertices. Label the Example 3.25. Let =   vertices clockwise by 1, 2, . . . , n and let i = 12 . . . n. Then HOM(Pi i , Pii ) is + -graded and Hom(Pi i , Pi i ) is 2-dimensional with the basis {i i 1i i , }, where =  ///

3 n

3 n

46

TIGHT MONOMIALS AND INDECOMPOSABLE PROJECTIVES A computation shows that deg() = 0 and 2 = 0 if n is odd, 2 if n is even;

47

implying that i i 1ii is the only idempotent in Pi i if n is odd, or if n is even and char(k) = 2. Under these assumptions, Pi i is indecomposable, but [Pi i ], [Pi i ] 2 + q [q] = 1 + q [q], and [Pi i ] is not a canonical basis element. For an odd length cycle we found an indecomposable projective Pi i whose image in the Grothendieck group is not a canonical basis vector, while being invariant under the bar involution: [Pi i ] = [Pi i ]. When n is even and char(k) = 2, 0 = is an idempotent in Hom(Pi i , Pi i ), 2 and Pi i Pi i 0 Pi i (1 0 ) is isomorphic to the direct sum of two inde= composable projectives. Furthermore, = 1 0 where

1 =
1 1 2

3 n n

0 =

3 n

3 n

Module homomorphisms 1 : 0 : Pi i P1(2) 2(2) ...n(2) , P1(2) 2(2) ...n(2) Pi i ,

induced by these elements via right multiplication have degree 0, and 0 1 = 2 Id, so that Pi i 0 P1(2) 2(2) ...n(2) . =

47

A CONJECTURE ON CATEGORIFICATION OF IRREDUCIBLE REPRESENTATIONS

48

3.4

A conjecture on categorication of irreducible representations

Choose [I], = i i, i I. Let R(; ) be the quotient ring of R() by the ideal generated by all diagrams of the form

i1
i1 i2

...
im

where i1 . . . im Seq() and the leftmost string has i1 dots on it. The ring R(; ) inherits a grading from R(). These quotient rings should be the analogues of the Ariki-Koike cyclotomic Hecke algebras in our framework. Let def R(; ) = R(; ),
[I]

to a eld k. We expect that, for suciently nice and switch from and k, the category of graded modules over R(; ) categories the integrable irreducible Uq (g)-representation V with the highest weight . Let R(; )pmod be the category of nitely-generated graded projective left R(; )-modules and R(; )pmod = There should exist an isomorphism K0 (R(; )) V =
, , def [I]

R(; )pmod.

where V , is an integral version of V , a free [q, q 1 ]-module spanned by (a) the compositions of divided dierences Fi applied to the highest weight vector v V . Under this isomorphism indecomposable projectives should (a) (a) correspond to canonical basis vectors in V . The action of Ei and Fi (a) (a) should lift to exact functors Ei and Fi between categories R(; )pmod and R( + ai; )pmod as well as the categories R(; )mod and R( + (a) ai; )mod of all nitely-generated graded modules. These functors Ei 48

REFERENCES
(a)

49

and Fi will be direct summands of the induction and restriction functors between R(; ) and R( + ai; )-modules, dened la Ariki. We expect a them to be biadjoint, up to grading shifts.

References
[1] S. Ariki. On the decomposition numbers of the Hecke algebra of G(m, 1, n). J. Math. Kyoto Univ., 36(4):789808, 1996. [2] S. Ariki. Lectures math.QA/9908005. on cyclotomic Hecke algebras, 1999,

[3] S. Ariki. Representations of quantum algebras and combinatorics of Young tableaux, volume 26 of University Lecture Series. AMS, Providence, RI, 2002. [4] A. Beilinson, G. Lusztig, and R. MacPherson. A geometric setting for the quantum deformation of GLn . Duke Math. J., 61(2):655677, 1990. [5] I. Bernstein, I. Gelfand, and S. Gelfand. Schubert cells, and the cohomology of the spaces G/P . Russian Math. Surveys, 28:126, 1973. [6] J. Bernstein, I. B. Frenkel, and M. Khovanov. A categorication of the Temperley-Lieb algebra and Schur quotients of U(sl(2)) via projective and Zuckerman functors. Selecta Math. (N.S.), 5(2):199241, 1999. [7] S. Billey and V. Lakshmibai. Singular loci of Schubert varieties, volume 182 of Progress in Mathematics. Birkhuser Boston Inc., Boston, MA, a 2000. [8] J. Brundan and A. Kleshchev. Hecke-Cliord superalgebras, crystals (2) of type A2l and modular branching rules for Sn . Represent. Theory, 5:317403 (electronic), 2001, math.RT/0103060. [9] J. Chuang and R. Rouquier. Derived equivalences for symmetric groups and sl 2-categorication. Ann. of Math., 167:245298, 2008, math.RT/0407205. [10] L. Crane and I. B. Frenkel. Four-dimensional topological quantum eld theory, Hopf categories, and the canonical bases. J. Math. Phys., 35(10):51365154, 1994, hep-th/9405183. 49

REFERENCES

50

[11] M. Demazure. Invariants symtriques entiers des groupes de Weyl et e torsion. Invent. Math., 21:287301, 1973. [12] V. Drinfeld. Quantum groups. In Proceedings of the International Congress of Mathematicians, Vol. 1, 2 (Berkeley, Calif., 1986), pages 798820, Providence, RI, 1987. Amer. Math. Soc. [13] I. B. Frenkel, M. Khovanov, and C. Stroppel. A categorication of nite-dimensional irreducible representations of quantum sl(2) and their tensor products. Selecta Math. (N.S.), 12(3-4):379431, 2006, math/0511467. [14] I. Grojnowski. Ane slp controls the representation theory of the symmetric group and related Hecke algebras, 1999, math.RT/9907129. [15] I. Grojnowski and G. Lusztig. On bases of irreducible representations of quantum GLn . In Kazhdan-Lusztig theory and related topics (Chicago, IL, 1989), volume 139 of Contemp. Math., pages 167174. 1992. [16] I. Grojnowski and G. Lusztig. A comparison of bases of quantized enveloping algebras. In Linear algebraic groups and their representations (Los Angeles, CA, 1992), volume 153 of Contemp. Math., pages 1119. 1993. [17] I. Grojnowski and M. Vazirani. Strong multiplicity one theorems for ane Hecke algebras of type A. Transform. Groups, 6(2):143155, 2001. [18] M. Jimbo. A q-dierence analogue of U(g) and the Yang-Baxter equation. Lett. Math. Phys., 10(1):6369, 1985. [19] M. Kashiwara. Crystalizing the q-analogue of universal enveloping algebras. Comm. Math. Phys., 133(2):249260, 1990. [20] M. Kashiwara. On crystal bases of the Q-analogue of universal enveloping algebras. Duke Math. J., 63(2):465516, 1991. [21] M. Khovanov. Nilcoxeter algebras categorify the Weyl algebra. Comm. Algebra, 29(11):50335052, 2001, math.RT/9906166. [22] A. Kleshchev. Branching rules for modular representations of symmetric groups. II. J. Reine Angew. Math., 459:163212, 1995. 50

REFERENCES

51

[23] A. Kleshchev. Branching rules for modular representations of symmetric groups. III. Some corollaries and a problem of Mullineux. J. London Math. Soc. (2), 54(1):2538, 1996. [24] A. Kleshchev. On decomposition numbers and branching coecients for symmetric and special linear groups. Proc. London Math. Soc. (3), 75(3):497558, 1997. [25] A. Kleshchev. Linear and projective representations of symmetric groups, volume 163 of Cambridge Tracts in Mathematics. Cambridge U. Press, 2005. [26] B. Kostant and S. Kumar. The nil Hecke ring and cohomology of G/P for a Kac-Moody group G. Adv. in Math., 62(3):187237, 1986. [27] A. Lascoux, B. Leclerc, and J.-Y. Thibon. Hecke algebras at roots of unity and crystal bases of quantum ane algebras. Comm. Math. Phys., 181(1):205263, 1996. [28] A. D. Lauda. A categorication of quantum sl(2), 2008, arXiv:0803.3652. [29] A. D. Lauda. Categoried quantum sl(2) and equivariant cohomology of iterated ag variaties, 2008, arXiv:0803.3848. [30] B. Leclerc. Dual canonical bases, quantum shues and q-characters. Math. Z. 246(4):691732 (2004), no. 4. [31] G. Lusztig. Canonical bases arising from quantized enveloping algebras. J. Amer. Math. Soc., 3(2):447498, 1990. [32] G. Lusztig. Canonical bases arising from quantized enveloping algebras. II. Progr. Theoret. Phys. Suppl., 102:175201 (1991), 1990. Common trends in mathematics and quantum eld theories (Kyoto, 1990). [33] G. Lusztig. Quivers, perverse sheaves, and quantized enveloping algebras. J. Amer. Math. Soc., 4(2):365421, 1991. [34] G. Lusztig. Introduction to quantum groups, volume 110 of Progress in Mathematics. Birkhuser Boston Inc., Boston, MA, 1993. a

51

REFERENCES

52

[35] G. Lusztig. Tight monomials in quantized enveloping algebras. In Quantum deformations of algebras and their representations (Ramat-Gan, 1991/1992; Rehovot, 1991/1992), volume 7 of Israel Math. Conf. Proc., pages 117132. Bar-Ilan Univ., 1993. [36] L. Manivel. Symmetric functions, Schubert polynomials and degeneracy loci, volume 6 of SMF/AMS Texts and Monographs. AMS, Providence, RI, 2001. [37] A. Mathas. Iwahori-Hecke algebras and Schur algebras of the symmetric group, volume 15 of University Lecture Series. AMS, Providence, RI, 1999. [38] A. Postnikov. Enumeration in algebra and geometry. Ph.D. thesis, MIT, 1997 available at http://citeseer.ist.psu.edu/postnikov97enumeration.html, 1997. [39] M. Reineke. Monomials in canonical bases of quantum groups and quadratic forms. J. Pure Appl. Algebra, 157(2-3):301309, 2001. [40] C. Ringel. Hall algebras and quantum groups. 101(3):583591, 1990. Invent. Math.,

[41] R. Rouquier. Categorication of sl2 and braid groups. In Trends in representation theory of algebras and related topics, volume 406 of Contemp. Math., pages 137167. AMS, Providence, RI, 2006. [42] R. Rouquier. Higher representation theory. Talk at IAS, March 2008. [43] R. Rouquier. Higher representations of Kac-Moody algebras. Work in progress. [44] J. Sussan. Category math.QA/0701045. O and sl(k) link invariants, 2007,

[45] M. Vazirani. Irreducible modules over the ane Hecke algebra: a strong multiplicity one result. Ph.D. thesis, UC Berkeley, 1999, math.RT/0107052.

52

REFERENCES

53

[46] M. Vazirani. Parameterizing Hecke algebra modules: BernsteinZelevinsky multisegments, Kleshchev multipartitions, and crystal graphs. Transform. Groups, 7(3):267303, 2002, math.RT/0107052. [47] A. Zelevinsky. Induced representations of reductive p-adic groups. II. On irreducible representations of GL(n). Ann. Sci. Ecole Norm. Sup. (4), 13(2):165210, 1980. [48] H. Zheng. A geometric categorication of tensor products of Uq (sl2 )modules, 2007, arXiv:0705.2630. [49] H. Zheng. Categorication of integrable representations of quantum groups, 2008, arXiv:0803.3668. M.K.: School of Mathematics, Institute for Advanced Study, Princeton, NJ 08540, and Department of Mathematics, Columbia University, New York, NY 10027 email: khovanov@math.columbia.edu A.L: Department of Mathematics, Columbia University, New York, NY 10027
email: lauda@math.columbia.edu

53