arXiv:0807.3250v1 [math.

QA] 21 Jul 2008

A diagrammatic approach to categorication of quantum groups III

Mikhail Khovanov and Aaron D. Lauda July 22, 2008
Abstract We categorify the idempotented form of quantum sl(n).

Contents
1 Introduction 2 Graphical Interpretation of the bilinear form 2.1 Quantum groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Geometric interpretation of the bilinear form . . . . . . . . . . . . . . . . . 3 Graphical calculus for U categorication 3.1 The 2-category U . . . . . . . . . . . . . . 3.2 Spanning sets of HOMs in U . . . . . . . . 3.3 Properties and symmetries of 2-category U 3.4 Karoubi envelope, U, and 2-representations 3.5 Direct sum decompositions . . . . . . . . . 3.6 K0 (U) and homomorphism . . . . . . . . 3.7 Idempotented rings . . . . . . . . . . . . . 3.8 Surjectivity of . . . . . . . . . . . . . . . 3.9 Injectivity of in the nondegenerate case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 6 6 12 20 20 28 34 38 40 42 45 46 53 54 54 55 58 58 62 71

4 Categorication of U for sln 4.1 Forms of quantum sln . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The 2-category U (sln ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Iterated ag varieties 5.1 Cohomology of n-step varieties . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Graphical calculus for iterated ag varieties . . . . . . . . . . . . . . . . . 5.3 The 2-category FlagN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2
6 Representing U on the ag 2-category 6.1 Dening the 2-functor N . . . . . . . 6.2 Checking the relations of U . . . . . 6.3 Equivariant representation . . . . . . . 6.4 Nondegeneracy of U (sln ) . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

72 72 75 78 84 86

References

Introduction

In this paper we categorify the Beilinson-Lusztig-Macpherson idempotented modication U(sln ) of Uq (sln ) for any n, generalizing [21, 22] where such categorication was described for n = 2 and using constructions and results of [16, 17] which contain a categorication of U for any Cartan datum. In [24] Lusztig associates a quantum group U to any root datum; the latter consists of a perfect pairing , between two free abelian groups X and Y , embeddings of the set I of simple roots into X, Y , and a bilinear form on [I] subject to certain compatibility and integrality conditions. Lusztigs denition is slightly dierent from the original ones due to Drinfeld [9] and Jimbo [13]. Lusztig then modies U to the nonunital ring U which contains a system of idempotents {1 } over all weights X as a substitute for the unit element, U= 1 U1 .
,X

In the sln case U was originally dened by Beilinson, Lusztig, and Macpherson [2] and then appeared in [23, 14] in greater generality. It is clear from Lusztigs work that the (q)-algebra U is natural for at least the following reasons:

1. A U-module is the same as a U-module which has an integral weight decomposition. These modules are of prime importance in the representation theory of U. 2. U has analogues of the comultiplication, the antipode, and other standard symmetries of U. 3. U is a U-bimodule. 4. The Peter-Weyl theorem and the theory of cells can be intrinsically stated in terms of the algebra U. 5. U has an integral form A U, a [q, q 1 ]-lattice closed under multiplication and co multiplication. The integral form comes with a canonical basis B. Conjecturally, multiplication and comultiplication in this basis have coecients in [q, q 1] when the Cartan datum is symmetric.

6. The braid group associated to the Cartan datum acts on U. 2

3 Moreover, U appears throughout the categorication program for quantum groups. Representations of quantum groups that are known to have categorications all have in tegral weight decompositions, and thus automatically extend to representations of U. In most or all of these examples, see [3, 7, 10, 31, 33], the weight decomposition of representations lifts to a direct sum decomposition of categories, so one obtains a categorication of the idempotent 1 as the functor of projection onto the corresponding direct summand (the only possible exception is the categorication of tensor products via the ane Grassmannian [6, 5]). In the categorication of tensor powers of the fundamental Uq (sl2 )-representation [3, 10], each canonical basis element of U acts as an indecompos rather than U should be able projective functor or as the zero functor. The idea that U categoried goes back to Crane and Frenkel [8]. U is generated by elements Ei 1 , Fi 1 , and 1 , where X is an element of the weight lattice and i is a simple root. We will often write E+i instead of Ei and Ei instead of Fi . We have Ei 1 = 1 Ei 1 , where, in our notations, explained in Section 2.1, = iX , and iX is the element of X associated to the simple root i. Algebra U is spanned by products Ei 1 := Ei1 Ei2 . . . Eim 1 = 1 Ei1 Ei2 . . . Eim 1 , where i = (i1 , . . . , im ) is a signed sequence of simple roots, and = + i X . The integral form A U U is the [q, q 1 ]-algebra generated by divided powers Ei(a) 1 = 1 a E 1 . [a]i ! i

Note that U can, alternatively, be viewed as a pre-additive category with objects X and morphisms from to being 1 U1 . Of course, any ring with a collection of mutuallyorthogonal idempotents as a substitute for the unit element can be viewed as a pre-additive category and vice versa. From this perspective, though, we can expect the categorication of U to be a 2-category. In Section 3.1 we associate a 2-category U to a root datum. The objects of this 2category are integral weights X, the morphisms from to are nite formal sums of symbols Ei 1 {t}, where i = (i1 , . . . , im ) is a signed sequence of simple roots such that the left weight of the symbol is (Ei 1 = 1 Ei 1 ), and t is a grading shift. When i consists of a single term, we get 1-morphisms E+i 1 and Ei1 , which should be thought of as categorifying elements Ei 1 and Fi 1 of U, respectively. Grading shift {t} t categories multiplication by q . The one-morphism Ei 1 : should be thought of as a categorication of the element Ei 1 . When the sequence i is empty, we get the identity morphism 1 : , a categorication of the element 1 . Two-morphisms between Ei 1 {t} and Ej 1 {t } are given by linear combinations of degree t t diagrams drawn on the strip [0, 1] of the plane. The diagrams consist

4 of immersed oriented one-manifolds, with every component labelled by a simple root, and dots placed on the components. Labels and orientations at the lower and upper endpoints of the one-manifold must match the sequences i and j , respectively. Integral weights label regions of the plane cut out by the one-manifold, with the rightmost region labelled . Each diagram has an integer degree assigned to it. We work over a ground eld k, and dene a two-morphism between Ei 1 {t} and Ej 1 {t } as a linear combination of such diagrams of degree t t , with coecients in k, modulo isotopies and a collection of very carefully chosen local relations. The set of 2-morphisms U(Ei 1 {t}, Ej 1 {t }) is a k-vector space. We also form graded vector space HOMU (Ei 1 , Ej 1 ) :=
t

U(Ei 1 {t}, Ej 1 ).

(1.1)

Vertical composition of 2-morphisms is given by concatenation of diagrams, horizontal composition consists of placing diagrams next to each other. In each graded k-vector space HOMU (Ei 1 , Ej 1 ) we construct a homogeneous spanning set Bi ,j , which depends on extra choices. The Laurent power series in q, with the coecient at q r equal to the number of spanning set elements of degree r, is proportional to suitably normalized inner product Ei 1 , Ej 1 , where the semilinear form , is a mild modication of the Lusztig bilinear form on U. The proportionality coecient depends only on the root datum. We say that our graphical calculus is nondegenerate for a given root datum and eld k if for each i , j and the homogeneous spanning set Bi ,j , is a basis of the k-vector space HOMU (Ei 1 , Ej 1 ). Nondegeneracy will be crucial for our categorication constructions. The 2-category U is k-additive, and we form its Karoubian envelope U, the smallest 2-category which contains U and has splitting idempotents. Namely, for each , X, the category U(, ) of morphisms is dened as the Karoubian envelope of the additive k-linear category U(, ). The split Grothendieck category K0 (U) is a pre-additive category with objects , and the abelian group of morphisms from to is the split Grothendieck group K0 (U(, )) of the additive category U(, )). The grading shift functor on U(, ) turns K0 (U(, )) into a [q, q 1 ]-module. This module is free with the basis given by isomorphism classes of indecomposable objects of U (, ), up to grading shifts. The split can also be viewed as a nonunital [q, q 1 ]-algebra with a Grothendieck category K0 (U) collection of idempotents [1 ] as a substitute for the unit element. In Section 3.6 we set up a [q, q 1 ]-algebra homomorphism :
A UK0 (U)

which takes 1 to [1 ] and Ei 1 to [Ei 1 ], for any divided power signed sequence i . The main results of this paper are the following theorems. Theorem 1.1. The map is surjective for any root datum and eld k. Theorem 1.2. The map is injective if the graphical calculus for the root datum and eld k is nondegenerate. 4

5 Theorem 1.3. The graphical calculus is nondegenerate for the root datum of sln and any eld k. The three theorems together immediately imply Proposition 1.4. The map is an isomorphism for the root datum of sln and any eld k. The last result establishes a canonical isomorphism
A U(sln )

K0 U (sln ) =

and allows us to view U (sln ) as a categorication of U(sln ). Theorem 1.1, proved in Section 3.8, follows from the results of [16, 17, 21] and basic properties of Grothendieck groups and idempotents. Theorem 1.2, proved in Section 3.9, follows from the nondegeneracy of the semilinear form on U and its pictorial interpretation explained in Section 2.2. To prove theorem 1.3 we construct a family of 2-representations of U and check that the elements of each spanning set Bi ,j , act linearly independently on vector spaces in these 2-representations, implying nondegeneracy of the graphical calculus. Sections 46 are devoted to these constructions. Indecomposable one-morphisms, up to isomorphism and grading shifts, constitute a basis of K0 (U(sln )) A U(sln ), which might potentially depend on the ground eld k. The = multiplication in this basis has coecients in [q, q 1 ]. It is an open problem whether this basis coincides with the Lusztig canonical basis of A U(sln ). The answer is positive when n = 2, see [21]. Another major problem is to determine for which root data the graphical calculus is nondegenerate. Nondegeneracy immediately implies, via theorems 1.1 and 1.2, that U for a given root datum. categories U We believe that U will prove ubiquitous in representation theory. This 2-category or its mild modications is expected to act on parabolic-singular blocks of highest weight categories for slN in the context of categorication of sln representations [7, 10, 31], on derived categories of coherent sheaves on Kronheimer-Nakajima [20] and Nakajima [26] quiver varieties and on their Fukaya-Floer counterparts, on categories of modules over cyclotomic Hecke and degenerate Hecke algebras [1, 19], on categories of perverse sheaves in Zhengs categorications of tensor products [33], on categories of modules over cyclotomic quotients of rings R() in [16, 17], on categories of matrix factorizations that appear in [18, Section 11], et cetera. A possible approach to proving that the calculus is nondegenerate for other root systems is to show that U acts on a suciently large 2-category and verify that the spanning set elements act linearly independently. It would also be interesting to relate our constructions with those of Rouquier [29, 30]. Categories of projective modules over rings R(), dened in [16, 17], categorify U weight spaces. A subset of our dening local relations on two-morphisms gives the relations

6 for rings R(). This subset consists exactly of the relations whose diagrams have no critical points (U-turns) on strands and have all strand orientations going in the same direction. In other words, the relations on braid-like diagrams allow us to categorify U , while the relations without these restrictions lead to a categorication of the entire U, at least in the sln case. Informally, the passage from a categorication of U to a categorication of U is analogous to generalizing from braids to tangles. Acknowledgments. M.K. was partially supported by the NSF grant DMS-0706924 and, during the early stages of this work, by the Institute for Advanced Study.

2
2.1
2.1.1

Graphical Interpretation of the bilinear form

Quantum groups
Algebras f and U

We recall several denitions, following [24]. A Cartan datum (I, ) consists of a nite set I and a symmetric -valued bilinear form on [I], subject to conditions i i {2, 4, 6, . . . } for i I, dij := 2 ij {0, 1, 2, . . . } for any i = j in I. ii
a1 a3 1a Let qi = q 2 , [a]i = qi + qi + + qi , [a]i ! = [a]i [a 1]i . . . [1]i . Denote by f the free associative algebra over (q) with generators i , i I, and introduce q-divided powers (a) a i = i /[a]i !. The algebra f is [I]-graded, with i in degree i. The tensor square f f is an associative algebra with twisted multiplication
ii

(x1 x2 )(x1 x2 ) = q |x2||x1| x1 x1 x2 x2 for homogeneous x1 , x2 , x1 , x2 . The assignment r(i ) = i 1 + 1 i extends to a unique algebra homomorphism r : f f f. Algebra f carries a (q)-bilinear form determined by the conditions1

(1, 1) = 1,
2 (i , j ) = i,j (1 qi )1 for i, j I,

(x, yy ) = (r(x), y y ) for x, y, y f, (xx , y) = (x x , r(y)) for x, x , y f. The bilinear form (, ) is symmetric. Its radical I is a two-sided ideal of f. The form (, ) descends to a nondegenerate form on the associative (q)-algebra f = f/I.

Our bilinear form (, ) corresponds to Lusztigs bilinear form {, }, see Lusztig [24, 1.2.10].

Quantum groups Theorem 2.1. The ideal I is generated by the elements (1)a i j i
a+b=dij +1 (a) (b)

over all i, j I, i = j. It seems that the only known proof of this theorem, for a general Cartan datum, requires Lusztigs geometric realization of f via perverse sheaves. This proof is given in his book [24, Theorem 33.1.3]. Less sophisticated proofs exist when the Cartan datum is nite. We see that f is the quotient of f by the quantum Serre relations (1)a i j i = 0.
a+b=dij +1 (a) (a) (b)

(2.1)

Let A f be the [q, q 1 ]-subalgebra of f generated by the divided powers i , over all i I and a . A root datum of type (I, ) consists of

free nitely generated abelian groups X,Y and a perfect pairing , : Y X ; inclusions I X (i iX ) and I Y (i i) such that i, jX = 2 ij = dij for all i, j I. ii This implies i, iX = 2 for all i. We write iX , rather than Lusztigs i , to denote the image of i in X. The quantum group U associated to a root datum as above is the unital associative (q)-algebra given by generators Ei , Fi , K for i I and Y , subject to the relations: i) K0 = 1, K K = K+ for all , Y , ii) K Ei = q
,iX

Ei K for all i I, Y ,

iii) K Fi = q ,iX Fi K for all i I, Y ,

Ki iv) Ei Fj Fj Ei = ij Kiq1 , where Ki = K(ii/2)i , q
i i

v) For all i = j (1)a Ei Ej Ei = 0

a+b= i,jX +1 (a) (b)

and

(1)a Fi Fj Fi
a+b= i,jX +1

(a)

(b)

= 0.

If f (i ) I for a polynomial f in noncommutative variables i , i I, then f (Ei ) = 0 and f (Fi ) = 0 in U. This gives a pair of injective algebra homomorphisms f U.

Quantum groups 2.1.2 Some automorphisms of U

Let be the -linear involution of algebra (anti)automorphisms. The

(q) which maps q to q1. U has the following standard (q) and x U.

(q)-antilinear algebra involution : U U given by (q)-linear algebra involution : U U given by

(Ei ) = Fi , (Fi ) = Ei , (K ) = K .

(Ei ) = Ei , (Fi ) = Fi , (K ) = K , (f x) = f (x) for f The

The

(q)-linear algebra antiinvolution : U Uop given by

(Ei ) = Ei , (Fi ) = Fi , (K ) = K .

The

(q)-linear algebra antiinvolution : U Uop given by

(Ei ) = qi Ki Fi , (Fi ) = qi Ki Ei , (K ) = K .

We denote by the The

(q)-antilinear antiautomorphism : U Uop dened as the composite = :

1 (Ei ) = qi Ki Fi , 1 (Fi ) = qi Ki Ei ,

(q)-linear antiinvolution : U Uop .

(K ) = K .

2.1.3

U and the bilinear form

The (q)-algebra U is obtained from U by adjoining a collection of orthogonal idempotents 1 for each X, (2.2) 1 1 = , 1 , such that K 1 = 1 K = q
,

1 ,

Ei 1 = 1+iX Ei ,

Fi 1 = 1iX Fi .

(2.3)

The algebra U decomposes as direct sum of weight spaces U=

, X

1 U1 .

We say that , respectively , is the right, respectively left, weight of x 1 U1 . The is the [q, q 1 ]-subalgebra of U generated by products of divided powers E (a) 1 algebra A U i (a) and Fi 1 , and has a similar weight decomposition
AU

=
, X

1 (A U)1 . 8

Quantum groups The following identities hold in U and A U: (Ei Fj Fj Ei )1 = i,j [ i, ]i 1 , Ei 1 = 1+aiX Ei ,
min(a,b) (a) (b) Ei Fi 1 (a) (a)

(2.4)
(a)

Fi 1 = 1aiX Fi , Fi
i (at) (bt) (bt)

(a)

(2.5) 1 , (2.6)

=
t=0 min(a,b)

a b + i, t a + b i, t

Ei

(at)

(b) (a) Fi Ei 1

=
t=0

Ei
i

Fi

1 .

(2.7)

The (anti) automorphisms , , , and all naturally extend to U and A U if we set (1 ) = 1 , (1) = 1 , (1 ) = 1 , (1 ) = 1 , (1 ) = 1 .

Taking direct sums of the induced maps on each summand 1 U1 allows these maps to For example, the antiautomorphism induces for each and in X be extended to U. an isomorphism 1 U1 1 U1 . Restricting to the [q, q 1 ]-subalgebra A U and taking direct sums, we obtain an algebra antiautomorphism : A U A U such that (1 ) = 1 , 1 i, 1+ i, 1+iX Ei 1 for all X. (1+iX Ei 1 ) = qi 1 Fi 1+iX , and (1 Fi 1+iX ) = qi The following result is taken from Lusztig [24, 26.1.1], but our bilinear form is normalized slightly dierently from his. Proposition 2.2. There exists a unique pairing (, ) : U U (i)

(q) with the properties: (, ) is bilinear, i.e., (f x, y) = f (x, y), (x, f y) = f (x, y), for f (q) and x, y U,

(ii) (11 x12 , 11 y12 ) = 0 for all x, y U unless 1 = 1 and 2 = 2 , (iii) (ux, y) = (x, (u)y) for u U and x, y U, (iv) (x1 , x 1 ) = (x, x ) for all x, x f U+ and all (here (x, x ) is as in Section 2.1.1), = (v) (x, y) = (y, x) for all x, y U. Denition 2.3. Dene a semilinear form , : U U x, y := ((x), y) Proposition 2.4. The map , : U U (i) , is semilinear, i.e., f x, y

(q) by

for all x, y U.

(q) has the following properties = f x, y , x, f y = f x, y , for f (q) and x, y U,

(ii) 11 x12 , 11 y12 = 0 for all x, y U unless 1 = 1 and 2 = 2 , (iii) ux, y = x, (u)y for u U and x, y U, 9

Quantum groups (iv) x1 , x 1 = ((x), x ) for x, x f and X, (v) x, y = (y), (x) for all x, y U. Proof. Immediate.

10

Proposition 2.5. The bilinear form (, ) and the semilinear form , are both nondegenerate on U. Proof. Nondegeneracy of (, ) is implicit throughout [24, Chapter 26] and follows for instance from Theorem 26.3.1 of [24]. The bilinear and semilinear forms restrict to pairings (, ) : A U A U [q, q 1 ], 2.1.4 Signed sequences , : A U A U [q, q 1 ]. (2.8)

Throughout the paper we write E+i for Ei and Ei for Fi and need notation for products of these elements. By a signed sequence i = (1 i1 , 2 i2 , . . . , m im ), where 1 , . . . , m {+, } and i1 , . . . , im I, we mean a nite sequence of elements of I with signs. We may also write i as 1 i1 , 2 i2 , . . . , m im or even 1 i1 2 i2 . . . m im . Let |i| = m k ik , viewed as an k=1 element of [I]. The length of i , denoted i , is the number of elements in the sequence i (m in the above notations). Dene SSeq to be the set of signed sequences. Let Ei := E1 i1 E2 i2 . . . Em im U, Ei 1 := E1 i1 E2 i2 . . . Em im 1 U. Recall that iX X denotes the image of i under the embedding I X. Let i X := (1 i1 )X + + (m im )X X. (2.11) (2.9) (2.10)

Then Ei 1 = 1+i X Ei 1 , so that , respectively + i X , is the right, respectively left, weight of Ei 1 . A sequence i is positive if all signs 1 , . . . , m = +. We may write a positive sequence as (i1 , . . . , im ) or even i1 . . . im . Denote by SSeq+ the set of all positive sequences. As in [16], for [I], = i i, denote by Seq() the set of all sequences i = i1 im such that i appears i times in the sequence. There is an obvious bijection

SSeq+ =

[I]

Seq().

(2.12)

A sequence is negative if all signs 1 , . . . , m = . For a positive sequence i denote by i the corresponding negative sequence. For any signed sequence i denote by i the sequence given by reversing all signs in i . We write concatenation of sequences i and j as i j , attaching i to the left of i as ii , etc. 10

Quantum groups By a divided powers signed sequence (dpss for short) we mean a sequence
(a i = (1 i1 1 , 2 i2 2 , . . . , m imm ) ) (a ) (a )

11

(2.13)

where s and is are as before and a1 , . . . , am {1, 2, . . . }. Let |i | = m k ak ik [I] k=1 and dene i , the length of i , as m ak . For a dpss i let i be a signed sequence k=1 given by expanding i

i = (1 i1 , . . . , 1i1 , 2 i2 , . . . , 2 i2 , . . . , m im , . . . , m im ) = ((1 i1 )a1 (2 i2 )a2 . . . (m im )am ), with term 1 i1 repeating a1 times, term 2 i2 repeating a2 times, etc. Dene Ea (a) Ei(a) := Ei = i U, [a]i ! the quantum divided powers of generators Ei . Then (a) Ei(a) 1 = E 1 U
i

(2.14)

(2.15) (2.16) (2.17)

and has left weight + aiX . More generally, for a dpss i

1 2 Ei := E1 i1 E2 i2 . . . Emmm U i

(a )

(a )

(a )

and
m r=1

(a1 ) (a2 ) (a ) Ei 1 := E1 i1 E2 i2 . . . Emmm 1 U, i

with left weight + r ar (ir )X . Let SSeqd be the set of all dpss. Elements Ei 1 , over all i SSeqd and X, span [q, q 1 ]-module. For [I], Seqd() was dened in [16] as the set of all A U as a (a ) (a ) expressions i1 1 . . . imm such that m ar ir = . Each element of Seqd() gives a positive r=1 divided power sequence, and

SSeqd+ =

[I]

Seqd().

(2.18)

The length i of a dpss is dened as the sum a1 + a2 + + am in the above notation. For the readers convenience our notations for sequences are collected below:

Seqd() positive divided   powers sequences of weight y

SSeqd+ G positive divided power   signed sequences y

SSeqd G divided powers signed sequences y

?

Seq()  positive sequences  of weight

Seq+  G positive sequences

?

SSeq G signed sequences

?

11

Geometric interpretation of the bilinear form

12

2.2

Geometric interpretation of the bilinear form

Mark m points 1 {0}, 2 {0}, . . . , m {0} on the lower boundary {0} of the strip [0, 1] and k points 1 {1}, 2 {1}, . . . , k {1} on the upper boundary {1}. Assuming m + k is even, choose an immersion of m+k strands into [0, 1] with these 2 m + k points as the endpoints. Orient each strand and label it by an element of I. Then endpoints inherit orientations and labels from the strands: + + k=4 j i i i i u i 

m=4 + + Orientation and labels at lower and upper endpoints dene signed sequences i , j SSeq. In the above example, i = (+, j, +i, ) and j = (+i, j, i, +i). We consider immersions modulo boundarypreserving homotopies and call them pairings between sequences i and j , or simply (i , j )-pairings. Clearly, (i , j )-pairings are in a bijection with complete matchings of m + k points such that the two points in each matching pair share the same label and their orientations are compatible. Denote by p (i , j ) the set of all (i , j )-pairings. A minimal diagram D of a (i , j )-pairing is a generic immersion that realizes the pairing such that strands have no selfintersections and any two strands intersect at most once. We consider minimal diagrams up to boundarypreserving isotopies. A (i , j )-pairing has at least one minimal diagram, in the example below it has two:
i

e
j i

 (

 (

Here is an immersion which is not a minimal diagram

i

e i

y
i

Given X, color the regions of a minimal diagram D by elements of X such that the rightmost region is colored , and the two regions between the two sides on an i-labelled strand dier by iX y + iX
i

12

Geometric interpretation of the bilinear form

13

Isotope D so that the strands at each crossing are either both oriented up or down, and dene the degree deg(D, ) of D relative to as the integer which is the sum of contributions from all crossings, local maxima and minima of D: 
i


i


i

y
j

deg where

c+i,

ci,

c+i,

ci, ii (1 i, ). 2
1 i,

i j

i j

ci, := Notice that

(2.19) (2.20)

q ci, = qi

In the simply-laced case qi = q and ci, = 1 i, . The degrees of the other crossings are determined from the rules above j j i i y y y    y j := deg = deg deg  y  y i  deg  y
j i

=0

Since both of these crossings have degree zero, we will refer to them as balanced crossings. Proposition 2.6. deg(D, ) depends only on and on the (i , j )-pairing realized by D. Thus, deg(D, ) is independent of the choice of a minimal diagram for a pairing. Proof. Invariance under cancellation of U-turns can be shown as follows: + iX + iX ii ii y . deg y  y (1 + i, ) + (1 i, + iX ) = 0 = deg = 2 2

 := deg
j

y
i

y y

deg

 y 
j

y
i

=0

Invariance of deg(D, ) under other isotopies of D is straightforward and is left to the reader. Any two minimal diagrams of the same (i , j )-pairing are related by a sequence of isotopies and triplecrossing moves for various orientations, see examples below. y y y y y y y y y y y y y y  (2.21) y y y y  y y j j j k i i j  i i k k k The invariance under these moves is manifestly obvious. 13

Geometric interpretation of the bilinear form

14

From now on, for each (i , j )-pairing we choose one minimal diagram D representing this pairing and denote by p(i , j ) the set of these diagrams. Recall the pairings (, ) and , on U from Section 2.1. Theorem 2.7. For any i , j SSeq and , X (Ei 1 , Ej 1 ) = Ei 1 , Ej 1 =
Dp(i ,j )

q deg(D,)
iI

1 , 2 (1 qi )i

(2.22)

where i is the number of i-colored strands in D. For example,

i

,
i

'
j

p ((j, +i, +j, i), (+i, i))

so that (E(j,+i,+j,i)1 , E(+i,i) 1 ) = =

q 2ci, +c+j, ij + q c+j,iX

1 1 (2.23) 2 (1 q ) (1 qi ) j 1 1 2(1 i, ) (1+ j, ) ij (1+ j, ) ij qi qj q + qj q . (1 qi )2 (1 qj )

The product term in (2.22) is independent of D since 2i is the number of times label i appears in the sequence i j . Proof. The rst equality in (2.22) is obvious since Ei 1 is invariant under the involution . For i , j , , as in Theorem 2.7 dene (Ei 1 , Ej 1 ) =
Dp(i ,j )

q deg(D,)
iI

1 . 2 (1 qi )i

(2.24)

During the proof we view Ei 1 and Ej 1 in (Ei 1 , Ej 1 ) as formal symbols rather than elements of U, since we have not yet proved that (, ) descends to U. We want to show (Ei 1 , Ej 1 ) = (Ei 1 , Ej 1 ) for all i , j , and . The nontrivial case is = . We can also interpret the formula (Ei 1 , Ej 1 ) = 0 for = via the sum of diagrams, since then the rightmost region must be colored by both and , which is not allowed. i e
j i

 ( 14

Geometric interpretation of the bilinear form

15

Notice that in the absence of (i , j )-pairings the right hand side of (2.22) is zero, since there are no diagrams to sum over. If the left weights of Ei 1 and Ej 1 are dierent the inner product (Ei 1 , Ej 1 ) = 0 and (Ei 1 , Ej 1 ) = 0 as well since the set of (i , j )-pairings is empty in this case. Assume that both i and j are positive. Then any minimal (i , j )-diagram can be isotoped to be braidlike, with the strands going upward:
j1

y y

y y

jm

j1

y y

y y y

jm

y 
i1 im i1


im

Moreover, in this case k = m, otherwise p(i , j ) is empty. The sum on the right hand side of (2.22) describes the canonical bilinear form on U+ f evaluated on Ei = Ei1 Eim = and Ej = Ej1 . . . Ejm , see [25, 27, 16, 17]. In view of property (iv) of Lusztigs bilinear form, we obtain the following: Lemma 2.8. If i , j are positive, (Ei 1 , Ej 1 ) = (Ei 1 , Ej 1 ) for any X. Thus, for positive i , j , the geometrically dened bilinear form (, ) coincides with (, ). Lemma 2.9. Equation (2.22) holds for the pair (ii , j ) if and only if it holds for the pair (i , ij ). Proof. Attaching a U-turn gives a bijection between p(ii , j ) and p(i , ij ).
j

(2.25)

i f ( D

( D

i We have

deg( D, ) = deg(D, ) +

ii ii (1 i, iX ) = deg(D, ) (1 i, ), 2 2
1 i,

where the additional term matches the power of q in the formula (1 Ei ) = qi

Ei 1 .

15

Geometric interpretation of the bilinear form The previous lemma implies that it is enough to check the equality (Ei 1 , Ej 1 ) = (Ei 1 , Ej 1 ) when j is the empty sequence. The next three lemmas show that (Ei 1 , 1) = (Ei 1 , 1 ) for any i and . Lemma 2.10. Equation (2.22) holds for all pairs ((j )i , ) with i , j positive.

16

(2.26)

(2.27)

Proof. Bending up j by adding U-turns transforms the pair ((j )i , ) to (i , j ) for which (2.22) holds (see Lemma 2.8). We have (E(j )i 1 , 1 ) = q (Ei 1 , Ej 1 ), (E(j )i 1 , 1 ) = q (Ei 1 , Ej 1 ) , (2.28)

where is the power of q in the formula (Ej 1+i X ) = (Ej 1+i X ) = q 1+i X Ej . Lemma 2.11. For i, j I, i = j (Ei iji 1 , 1) = (Ei iji 1 , 1) if and only if (Ei jii 1 , 1 ) = (Ei jii 1 , 1 ). (2.30) Proof. Attach a crossing at the i j location to a diagram D in p(i i ji , ). D p # y
i i i

(2.29)

i j

D 
i

j i The resulting diagram D is minimal if the i and j strands of D do not intersect. Otherwise, it is not minimal, but the homotopy y  y  will make it minimal (the orientations in the picture are for the +i j case). We get a bijection p(i i ji , )
=

G p(i j ii , )

(2.31)

16

Geometric interpretation of the bilinear form y which preserves the degree of a diagram, for any , since deg  Hence, (Ei iji 1 , 1 ) = (Ei jii 1 , 1 ) . Since Ei iji 1 = Ei jii 1 in U, we see that (Ei iji 1 , 1 ) = (Ei jii 1 , 1 ) and the lemma follows. Lemma 2.12. Assume (Ei i 1 , 1 ) = (Ei i 1 , 1 ). Then (Ei +iii 1 , 1) = (Ei +iii 1 , 1 ) if and only if (Ei i+ii 1 , 1 ) = (Ei i+ii 1 , 1 ). Proof. Decompose the diagrams in p(i + i ii , ) into three classes (1) +i and i strands do not intersect

17 = 0 for any . (2.32)

(2.33)

(2.34) (2.35)

+i i

(2) +i and i strands intersect s

+i i (3) a strand connects +i and i s +i i Likewise, decompose the diagrams in p(i i + ii , ) into three classes: (1) i and +i strands do not intersect

s ! i +i

(2) i and +i strands intersect

i +i (3) a strand connects i and +i  i +i 17

Geometric interpretation of the bilinear form Set up a bijection p(i + i ii , )

=

18

G p(i i + ii , )

(2.36)

that takes diagrams from class (1) in the rst set to diagrams of class (2) in the second set by adding a ycrossing, diagrams of class (2) to diagrams of class (1) by removing the , and diagrams of class (3) to diagrams of class (3) by reversing the crossing +i  i orientation of the +i i strand. Let Ei +iii 1 = Ei +ii 1 Ei 1 , that is, = + i is the weight of the region to the X right of the strand near i We have in U Ei +iii 1 = Ei i+ii 1 + [ i, ]iEi i 1 . Therefore, (Ei +iii 1 , 1 ) = (Ei i+ii 1 , 1) + [ i, ]i (Ei i 1 , 1). (2.38) y = deg = 0, a diagram of class (1), respectively class Since deg   (2), contributes to (Ei +iii 1 , 1 ) as much as its image in class (2), respectively class (1), contributes to (Ei i+ii 1 , 1 ) . A diagrams of class (3) in p(i + i ii , ) comes from some diagram in p(i i , ) by adding a (+i i) cap to the correct position. Similarly, a diagram of class (3) in p(i i + ii , ) comes from a diagram in p(i i , ) by adding a (i + i) cap. Thus, the only contribution to the dierence (Ei +iii 1 , 1 ) (Ei i+ii 1 , 1 ) comes from class (3) and is given by y
B degB @ 0 i

+i i (2.37)

u 

C
C A

(Ei +iii 1 , 1) (Ei i+ii 1 , 1) = or

2 qi

(Ei i 1 , 1 )

B degB @

2 qi

C
C A

(Ei i 1 , 1 )

(Ei +iii 1 , 1 ) = (Ei i+ii 1 , 1) + = (Ei i+ii 1 , 1) +

qi

1 i,

qi 2 1 qi

1+ i,

(Ei i 1 , 1 ) (Ei i 1 , 1 )

qi

i,

qi 1 qi qi

i,

where the coecient of (Ei i 1 , 1 ) in the last term is easily recognized as [ i, ]i , completing the proof of the lemma.

18

Geometric interpretation of the bilinear form We can nish the proof that (Ei 1 , 1) = (Ei 1 , 1 )

19

(2.39)

by induction on i , the length of i . During the induction step we move all negative entries of i to the left of all positive entries, sometimes adding terms (Ej 1 , 1 ) , respectively (Ej 1 , 1 ) with j = i 2 to the equation. We can then reduce to the case when all negative entries of i precede all positive entries, which is Lemma 2.10. Theorem 2.7 follows. We can turn this proof around to dene U in a more geometric way than in Lusztig [24]. Start with the (q)-algebra U with mutuallyorthogonal idempotents 1 and basis {Ei 1 }i, , over all nite signed sequences i SSeq and X. The multiplication is

Ei 1 Ei 1 =

Ei i 1 0

if = + i X , otherwise.

(2.40)

When i is the empty sequence, E 1 = 1 . Dene a (q)-bilinear form (, ) on U via the sum over diagrams, formula (2.24). Let I U be the kernel of this bilinear form. Then

I=
,X

I ,

:= I U ,

(2.41)

where U is spanned by Ei 1 for all i such that = + i X . It follows from the denition of the bilinear form that I is an ideal of U. It is not hard to check that = U U/I and (, ) = (, ), (2.42)

following the above proof of Theorem 2.7. This denition is not too far o from Lusztigs original denition, which utilizes f U , dened as the quotient of the free associative = algebra f by the kernel of a bilinear form on f. The latter bilinear form is the restriction of (, ) on U to f U. Here we map f U by sending i = i1 . . . im to Ei 1 = E+i1 . . . E+im 1 for each positive sequence i (this map is not a homomorphism), alternatively we can send i to Ei 1 .

U y
?

GG U o y
?

?_

AU

? ?_
Af

GG f o

vertical arrows do not respect algebra structure

(q)-algebras

[q, q 1 ]-algebras

19

20 For each X there are two inclusions (corresponding to U+ and U ) of the lower half of the diagram to the upper half. Restricting to weight spaces, we get the following diagram
X

GG

y
?

X U

?_

X (A U)

y
?

y
? ?_
A f

G G f o

(q)-vector spaces
where
X U

[q, q 1 ]-modules

= 1X U1 , etc.

3
3.1
3.1.1

Graphical calculus for U categorication

The 2-category U
Denition

We dene a 2-category U for any root datum (Y, X, , , . . . ) of type (I, ). This 2-category has the structure of an additive k-linear 2-category, see [12] and [21, Section 5]. Thus the hom sets between any two objects form a k-linear category, and composition and identities are given by additive k-linear functors. The 2-morphisms in U are represented graphically using string diagrams, see [21, Section 4] and the references therein. Denition 3.1. Let (Y, X, , , . . . ) be a root datum of type (I, ). U is an additive k-linear 2-category. The 2-category U consists of objects: for X. The homs U(, ) between two objects , are additive k-linear categories consisting of: objects2 of U(, ): a 1-morphism in U from to is a formal nite direct sum of 1-morphisms Ei 1 {t} = 1 Ei 1 {t} for any t and signed sequence i SSeq such that = + i X .

morphisms of U(, ): for 1-morphisms Ei 1 {t}, Ej 1 {t } U, hom sets U(Ei 1 {t}, Ej 1 {t }) of U(, ) are graded k-vector spaces given by linear combinations of degree t t diagrams, modulo certain relations, built from composites of:
We refer to objects of the category U(, ) as 1-morphisms of U. Likewise, the morphisms of U(, ) are called 2-morphisms in U.
2

20

The 2-category U

21

i) Degree zero identity 2-morphisms 1x for each 1-morphism x in U; the identity 2-morphisms 1E+i 1 {t} and 1Ei 1 {t}, for i I, are represented graphically by 1E+i 1 {t} i + iX y i deg 0 1Ei 1 {t} i iX  i deg 0

and more generally, for a signed sequence i = 1 i1 2 i2 . . . m im , the identity 1Ei 1 {t} 2-morphism is represented as i1 + iX i1 i2 i2 im im

where the strand labelled i is oriented up if = + and oriented down if = . We will often place labels with no sign on the side of a strand and omit the labels at the top and bottom. The signs can be recovered from the orientations on the strands as explained in Section 2.2. ii) For each X the 2-morphisms Notation: 2-morphism: y  i,
i

i,

cccc   c  c i,j, y
i

c  c c c  c1 i,j,

+ iX y

 + iX

y
j

ii

 
j

Degree:

ii

i j

i j

Notation: 2-morphism: #

y
i,

y
i,


i

i,


i

i,

t
i

c+i,

q ci,

Degree:

c+i, 21

ci,

The 2-category U with ci, dened in (2.19), such that the following identities hold. The sl2 relations3 ( all of the strands are labelled by i): i) 1+iX E+i1 and 1 Ei 1+iX are biadjoint, up to grading shifts: + iX y  y y ii) 
i

22

+ iX y = 

y  + iX =

+ iX (3.1)

= y

+ iX y  = 

(3.2) + iX

+ iX

+ iX

y  + iX

=
i

 + iX = 

 y

+ iX (3.3)
i

iii) All dotted bubbles of negative degree are zero. That is,
i

 =0

if < i, 1 

= 0 if < i, 1

(3.4)

for all + , ywhere a dot carrying a label denotes the -fold iterated vertical composite of or  i, depending on the orientation. A dotted bubble of i, degree zero equals 1:
iw

i, 1

 = 1 for i, 1,

= 1 for i, 1.

i, 1

iv) For the following relations we employ the convention that all summations are increasing, so that =0 is zero if < 0. f y
i

y  y

i,

f =0 i

i, f y iw 
i, 1+f

y  y x

y =
i

i,


g=0

y
i

i, g

(3.5)

i, 1+g

The vertical ii-crossing was represented by the diagram

in [21, 22]. Here we use a standard

crossing for simplicity.

22

The 2-category U
i

23 y  i, 1f

y y
i


i


i

y i

i, 1

+
f =0 g=0

1+g i, x   s i, 1f  1+g i,  (3.6)

f g

y 
i

y
i

y
i


i

i, 1

iw

+
f =0 g=0

f g

for all X. Notice that for some values ofy the dotted bubbles appearing above have negative labels. A composite of or with itself a negative  i, i, number of times does not make sense. These dotted bubbles with negative labels, called fake bubbles, are formal symbols inductively dened by the equation

t + + i,1+1t + + i, i, 1 1+
i

iw

++ i, 1

t + i, 1+

iw

= 1. (3.7)

and the additional condition

iw

 1

1 if i, = 0.

Although the labels are negative for fake bubbles, one can check that the overall degree of each fake bubble is still positive, so that these fake bubbles do not violate the positivity of dotted bubble axiom. The above equation, called the innite Grassmannian relation, remains valid even in high degree when most of the bubbles involved are not fake bubbles. See [21] for more details. v) NilHecke relations: y y y
i

y y y y
i i

y y =
i

y y y y
i

y = 0,
i i

y
i

(3.8)

y
i

y
i

y =
i

y
i

y
i

y
i

y =
i

y
i

y
i

y i

(3.9)

We will also include (3.10) for i = j as an sl2 -relation. 23

The 2-category U

24

All 2-morphisms are cyclic4 with respect to the above biadjoint structure. This is ensured by the relations (3.3), and the relations
i j i j


j


i

y y

y y

 

= 

y  y y


i

(3.10)

The cyclic condition on 2-morphisms expressed by (3.3) and (3.10) ensures that diagrams related by isotopy represent the same 2-morphism in U. It will be convenient to introduce degree zero 2-morphisms:
j j i i j i

y 

:= 

y y
i

=
i


j i j

  y

(3.11)

j i

:=
j

y
i

y y 

y 


j

y
i

(3.12)

where the second equality in (3.11) and (3.12) follow from (3.10). For i = j y 
i

y = j

y y
i

 y
i j

 = 
i j

y
j

(3.13)

The R()-relations: i) For i = j y =

i

y
i

y
j

if i j = 0,

y y
i

y
j

dij

y
j

+
i

dji y y j

if i j = 0. (3.14)

y
i
4

y
j

y
j

y
j

y =
i

y j

(3.15)

See [21] and the references therein for the denition of a cyclic 2-morphism with respect to a biadjoint structure.

24

The 2-category U ii) Unless i = k and i j = 0 y y y

i

25

y y y
j k

y =
i

y y y y
k

y
j

(3.16)

For i j = 0 y y y y y
i

y y y
dij 1 a

y y
j i

y
j

y =
i

y
a=0 i

y
j

dij 1a y
i

(3.17)

For example, for any shift t there are 2-morphisms i y y : E+i1 {t} E+i1 {t 2} i : 1 {t} Ei+i 1 {t c+i, } t #
i

y j
i

: E+i+j 1 {t} E+j+i1 {t i j} : Ei+i 1 {t} 1 {t ci, }

y y y
i i

in U, and the diagrammatic relation y y y y = y y

i i i

y y y
i

gives rise to relations in U Eiii 1 {t}, Eiii1 {t + 3i i} for all t . the additive k-linear composition functor U(, ) U( , ) U(, ) is given on 1-morphisms of U by Ej 1 {t } Ei 1 {t} Ej i 1 {t + t } for i X = , and on 2-morphisms of U by juxtaposition of diagrams y y y y y y y y w  w  w  w        25 (3.18)

The 2-category U

26

Remark 3.2. By choosing an orientation of the graph associated to Cartan datum (I, ) the R()-relations above can be modied by replacing them with signed R() relations determined by invertible elements ij , ji chosen for each edge of the graph (see [17] for more details). U has graded 2-homs dened by HOMU (x, y) :=
t

HomU (x{t}, y).

(3.19)

Also dene graded endomorphisms ENDU (x) := HOMU (x, x). (3.20)

The 2-category with the same objects and 1-morphisms as U and 2-homs given by HOMU (x, y) is denoted U , so that U (x, y) = HOMU (x, y). (3.21) U is a graded additive k-linear 2-category with translation [21, Section 5.1]. 3.1.2 Relations in U

We now collect some other relations that follow from those above. These relations simplify computations in the graphical calculus and can be used to reduce complex diagrams into simpler ones (see the proofs of Proposition 3.6 and Lemma 3.9). Proposition 3.3 (Bubble Slides). The following identities hold in U + jX f i y ( + 1 f )  f =0 i,+iX 1+f j + jX + jX i i y i y =  + y  i, 1+ i,+jX 2+ i,+j 1+ X j j j + jX i y  i, 1+
j

if i = j

if i j = 1

if i j = 0

26

The 2-category U f i ( + 1 f ) y w  f =0 ( i, 1)+f j i y w  + y iw  ( i, 1)+1 ( i, 1)+ j j i y w  ( i, 1)+

j

27

if i = j

y
j

if i j = 1

i,+jX 1+

if i j = 0

Proof. When i = j the proof appears in [21]. For i = j the equation follows from decomposing y y y y i i i i w y  w y   y  y = = j j j j
i,+jX 2+ i,+jX 2+ i,+jX 1+ i,+jX 1+

using the relations (3.13) and (3.14). Proposition 3.4 (More bubble slides). The following identities hold in U + iX + iX + iX 2 i i i y y   + w y 2 w w  ( i, +1)+(1) ( i, +1)+ ( i, +1)+(2) j j j y iw  = + jX f ( i, 1)+ i f y w  (1) j f =0 i,+jX 1+(f )
j

if i = j

if i j = 1

+ jX
i

y
j

i,+jX 1+

2 i y i + 2 y   ( i, ( i, 1)+(1) 1)+(2)
j j

if i = j

y
j

( i, 1)+

f =0

f i  (1)f y
j

if i j = 1

( i, 1)+(f )

27

Spanning sets of HOMs in U Proof. These equations follow from the previous Proposition. Proposition 3.5. Unless i = k = j we have y y y =  y j k i and when i = j = k we have y y y y y y  =  y y i i i i i  i y y y
i j

28

y 


k

(3.22)

f1 f2  y + i, 3+f4 f3 v
i

g2

u "g 1 y iw  g3 i, 1+g4 

(3.23)

where the rst sum is over all f1 , f2 , f3 , f4 0 with f1 + f2 + f3 + f4 = i, and the second sum is over all g1 , g2, g3 , g4 0 with g1 + g2 + g3 + g4 = i, 2. Recall that all summations in this paper are increasing, so that the rst summation is zero if i, < 0 and the second is zero when i, < 2. Reidemeister 3 like relations for all other orientations are determined from (3.16), (3.17), and the above relations using duality. Proof. For (3.22) if i = j post-compose both sides with the isomorphism y , 
j k i

then use that i = j = k to apply (3.16) on the right term and establish the equality. If i = j then we may assume that j = k. In this case we pre-compose with the isomorphism y  ,
i k j

then use (3.16) on the left side to establish the identity. The case i = j = k appears in [21, Section 5.4].

3.2

Spanning sets of HOMs in U

+ k=a

Given two Laurent power series f (q) = say that f (q) h(q) if fk hk for all k. 3.2.1 Endomorphisms of 1

fk q k and h(q) =

+ k=a

with fk , hk

we

For any root datum and X dene a graded commutative ring freely generated by symbols iw i   for i, 0 and for i, < 0 (3.24) i, 1+ i, 1+ of degree i i over all i I and > 0. 28

Spanning sets of HOMs in U

29

Proposition 3.6. Interpreting these generators of as elements of HOMU (1 , 1 ) induces a surjective graded k-algebra homomorphism HOMU (1 , 1 ). (3.25) Proof. By induction on the number of crossings of a closed diagram D representing an endomorphism of 1 one can reduce D to a linear combination of crossingless diagrams following the methods of [21, Section 8]. Crossingless diagrams that contain nested bubbles can be written as linear combinations of crossingless nonnested diagrams using the bubble slide equations in Propositions 3.3 and 3.4. Using the Grassmannian relations (3.7) all dotted bubbles with the same label i can be made to have the same orientation given by (3.24). The image of the monomial basis of under surjective homomorphism (3.25) is a homogeneous spanning set of the graded vector space HOMU (1 , 1 ). Denote this spanning set by B,, . Let 1 = . (3.26) 2a 1 qi iI a=1 depends only on the root datum (on the values of i i over all i I). The graded dimension of is . Corollary 3.7. gdim HOMU (1 , 1 ) and HOMU (1 , 1 ) is a local graded ring. Remark 3.8. is not Noetherian. We call monomials in this basis of and their images in HOMU (1 , 1 ) bubble monomials. 3.2.2 Homs between Ei 1 and Ej 1 for positive i and j

Recall rings R(), for

[I], from [16, 17] and decomposition

R() =
i ,j Seq() j R()i ,

(3.27)

where j R()i is spanned by braid-like diagrams D with i , j being the lower and upper sequences of D. Adding upward orientations to a diagram D in R(), placing it to the left of a collection of bubbles representing a monomial in
j

y
iw


1

i, 1+
iw

1+ j, 
j

 k, 1+

kw

i, 1+2 j, 1+4 i

 

29

Spanning sets of HOMs in U

30

and viewing the result as a 2-morphism from Ei 1 to Ej 1 induces a grading-preserving k-linear map i ,j , : j R()i k HOMU (Ei 1 , Ej 1 ). (3.28) Lemma 3.9. i ,j , is surjective. Proof. Start with a diagram D that represents an element in HOMU (Ei 1 , Ej 1 ). The relations in our graphical calculus allow us to inductively simplify D, by reducing the number of crossings if a strand or a circle of D has a self-intersection or if D contains two strands that intersect more than once. Bubble sliding rules allow moving bubbles to the far right of the diagram. Eventually, D reduces to a linear combination of diagrams which are products of diagrams representing elements of j R()i and monomials in . The elements in HOMU (Ei 1 , Ej 1 ) given by diagrams without circles, with no two strands intersecting more than once, and with all dots at the bottom are precisely the image under i ,j , of the basis j Bi of j R()i described in [16]. Denote by Bi ,j , the image under i ,j , of the product basis j Bi {monomials in } in j R()i . The lemma implies that Bi ,j , is a spanning set in HOMU (Ei 1 , Ej 1 ). Let E 1 := Ei 1 , E 1 := Ei 1 . (3.29)
i Seq() i Seq()

E 1 and E 1 are 1-morphisms in U. Summing i ,j , over all i , j Seq(), we obtain a homomorphism , : R() k ENDU (E 1 ) = EndU (E 1 ). Proposition 3.10. Homomorphism , of graded k-algebras is surjective. Adding a downward orientation to a diagram D in R(), multiplying it by (1)a where a is the number of crossings of identically colored lines, and placing it to the left of a collection of bubbles representing a monomial in induces a surjective homomorphism , : R() k ENDU (E 1 ). 3.2.3 Spanning sets for general i and j (3.31) (3.30)

We now describe a spanning set Bi ,j , in HOMU (Ei 1 , Ej 1 ) for any i , j , . Recall that p (i , j ) denotes the set of (i , j )-pairings, and p(i , j ) is a set of minimal diagrammatic representatives of these pairings. For each diagram D p(i , j ) choose an interval on each of the arcs, away from the intersections. The basis Bi ,j , consists of the union, over all D, of diagrams built out of D by putting an arbitrary number of dots on each of the intervals and placing any diagram representing a monomial in to the right of D decorated by these dots. Notice that Bi ,j , depends on extra choices, which we assume are made once and for all. For example, for i = (j, +i, +j, i) and j = (+i, i) the choice of minimal 30

Spanning sets of HOMs in U

31

diagrams is unique, see example following Theorem 2.7. The choice of intervals for the dots is not unique, though. Choosing these intervals as shown below
+i

+i

H g
j

{ g D

e m

+i +j i

+i +j i

results in the spanning set B(j,+i,+j,i),(+i,i), whose elements are

+i i iw j

a1 a2 j

a3

i, 1+1 j, 1+3 j iw i, 1+2 j, 1+4

  

 k, 1+

kw

+i +j i iw iw

bubble monomial in   1+ j,  1+ j, 
j j kw

+i

i, 1+ i, 1+

a1

a3

 k, 1+

a3 +i +j i

bubble monomial in

over all nonnegative integers a1 , a2 , a3 and over all diagrammatic monomials in (bubble orientations are for the case i, 0, j, < 0, k, 0). Proposition 3.11. For any intermediate choices made, Bi ,j , spans the k-vector space HOMU (Ei 1 , Ej 1 ). Proof. Relations on 2-morphisms in U(E 1 ) allow arbitrary homotopies of colored dotted diagrams modulo lower order terms, i.e., terms with fewer crossings, fewer circles, etc. Detailed discussion in [21, Section 8] generalizes to the present situation without diculty. For each s Bi ,j , its degree deg(s) is an integer, determined by the rules in Section 3.1. Proposition 3.12. For any i , j , and (Ei 1 , Ej 1 ) =
sBi ,j ,

q deg(s) .

(3.32)

31

Spanning sets of HOMs in U

32

Proof. The LHS of the equation (3.32) equals times the RHS of the formula (2.22). This is matched in the RHS of (3.32) by the summation over all monomials in , since q to the degree of these monomials add up to . The product term in the RHS of (2.22) is matched by the contribution to the RHS of (3.32) by all possible placements of dots. For each i-labelled strand dots contribute

q a(ii) =
a=0

1 1 = 2 ii 1q 1 qi

to the product, since the degree of a dot is i i, and the sum is over all ways to put some number a of dots on this strand. Finally, the sums over all minimal diagrams D p(i , j ) give equal contributions to the two sides of (3.32). Remark 3.13. In view of the rst equality in (2.22), we can restate the above proposition via , in place of (, ) on the LHS. Notice that gdimk (HOMU (Ei 1 , Ej 1 )) if and only if Bi ,j , is a basis of HOMU (Ei 1 , Ej 1 ). Corollary 3.14. For any sequences i , j and X gdimk (HOMU (Ei 1 , Ej 1 )) Ei 1 , Ej 1 . (3.34) =
sBi ,j ,

q deg(s)

(3.33)

Denition 3.15. We say that our graphical calculus is nondegenerate for a given root system and eld k if for all i , j , the set Bi ,j , is a basis of HOMU (Ei 1 , Ej 1 ). Thus, a calculus is nondegenerate if the equality holds in Corollary 3.14 for all i , j , . Remark 3.16. Nondegeneracy holds if the above condition is true for all X and all pairs of positive sequences i , j . 3.2.4 Endomorphisms of E, 1

For ,

[I] let
E, 1 :=
i Seq() j Seq( )

Ei j 1 .

(3.35)

Consider the graded ring ENDU (E, 1 ). A spanning set for this ring is given by dotted minimal diagrams of (i (i ), j (j ))-pairings over all i , j Seq(), i , j Seq( ) times

32

Spanning sets of HOMs in U bubble monomials in .

j j

33

y  
i

iw


j

i, 1+1 j, 1+3 iw

 k, 1+
kw
5

i, 1+2 j, 1+4

 

Let I, , be the subspace spanned by diagrams which contain a U-turn, i.e. an arc with both endpoints on {1} or on {0}.

Proposition 3.17. I, , is a 2-sided homogeneous ideal of ENDU (E, 1 ) which does not depend on choices of minimal diagrams for pairings. Proof. Left to the reader. Denote by R, , := ENDU (E, 1 )/I, , the graded quotient ring, and by the quotient map. There is a homomorphism : R() k R( ) k ENDU (E, 1 ) given by placing diagrams representing elements in R(), R( ), in parallel y (1)
a

(3.36)

y D

y
iw


1

D  

i, 1+
iw

1+ j, 
j

 , k, 1+
5

kw

i, 1+2 j, 1+4

orienting diagrams for R() upwards and diagrams for R( ) downwards, and multiplying by (1)a , where a is the number of crossings in D of equally labelled strands. The composition of this homomorphism with the quotient map is surjective. The diagram below contains an exact sequence of a ring, its 2-sided ideal, and the quotient ring: R() R( )

G I, ,

 G ENDU (E, 1 )

yyy yyy yyy yyy yyy yyy 99

G R, ,

G0.

(3.37)

33

Properties and symmetries of 2-category U

34

Remark 3.18. If the graphical calculus is nondegenerate, is an isomorphism, and the sequence splits 0 G I, , G ENDU (E, 1 ) o

G R() R( )

G 0.

(3.38)

A split ring homomorphism induces a split exact sequence of Grothendieck groups [28, Section 1.5] 0 G K0 (I, , ) G K0 ENDU (E, 1 ) o
K0 () K0 ()

G K R() R( ) 0

G0,

(3.39) leading to a canonical decomposition of the middle term as the sum of its two neighbors.

3.3
3.3.1

Properties and symmetries of 2-category U

Almost biadjoints

The 1-morphism E+i 1 does not have a simultaneous left and right adjoint Ei1+iX because the units and counits which realize these biadjoints in U are not degree preserving. However, if we shift Ei 1+iX by {c+i, }, then the unit and counit for the adjunction E+i 1 Ei 1+iX {c+i, } become degree preserving. More generally, we have E+ i1 {t} Ei 1+iX {c+i, t} in U since the units and counits have degree: deg
i +i {tc+i, t}

#
i i

t + iX
i {tc+i, t}

= c+i, + (c+i, )

(3.40)

and still satisfy the zigzag identities. Similarly, E+i 1 {t} has a left adjoint Ei1+iX {c+i, t} in U. One can check that with these shifts the units and counits of the adjunction Ei 1+iX {c+i, t} E+i1 {t} become degree zero and are compatible with the zigzag identities. The left adjoint Ei 1+iX {c+i, t} and the right adjoint Ei 1+iX {c+i, t} of E+i 1 {t} only dier by a shift. We call morphisms with this property almost biadjoint. This situation is familiar to those studying derived categories of coherent sheaves on CalabiYau manifolds. Functors with these properties are called almost Frobenius functors in [15] where several other examples of this phenomenon are also given. It is then clear that E+i 1 {t} and Ei 1 {t} have almost biadjoints in U for all t and X with 1 Ei 1+iX {c+i, t} 1 E+i 1iX {ci, t} 1+iX E+i1 {t} 1iX Ei1 {t} 34 1 Ei1+iX {c+i, t} 1 E+i 1iX {ci, t}. (3.42)

deg

q
+i

= (ci,+iX )) (c+i, )

(3.41)

Properties and symmetries of 2-category U

35

Every 1-morphism in U is a direct sum of composites of E+i 1 {t}s and Ei 1 {t}s together with identities; by composing adjunctions as explained in [21, Section 5.5], the right and left adjoints of Ei 1 {t} can be computed. Thus, it is clear that all 1-morphisms in U have almost biadjoints. Positivity of bubbles: U(1 , 1 {t}) = 0 if t > 0, and U(1 , 1 ) is at most 1-dimensional (isomorphic to k if the calculus is nondegenerate). 3.3.2 Symmetries of U

We denote by U op the 2-category with the same objects as U but the 1-morphisms reversed. The direction of the 2-morphisms remain xed. The 2-category U co has the same objects and 1-morphism as U, but the directions of the 2-morphisms is reversed. That is, U co (x, y) = U(y, x) for 1-morphisms x and y. Finally, U coop denotes the 2-category with the same objects as U, but the directions of the 1-morphisms and 2-morphisms have been reversed. Using the symmetries of the diagrammatic relations imposed on U we construct 2functors on the various versions of U. In Proposition 3.28 we relate these 2-functors to various [q, q 1 ]-(anti)linear (anti)automorphisms of the algebra U. The various forms of contravariant behaviour for 2-functors on U translate into properties of the corresponding homomorphism in U as the following table summarizes: 2-functors U U U U op U U co U U coop Algebra maps [q, q 1 ]-linear homomorphisms [q, q 1 ]-linear antihomomorphisms [q, q 1 ]-antilinear homomorphisms [q, q 1 ]-antilinear antihomomorphisms

Rescale, invert the orientation, and send c: Consider c the operation on the ccc  c c c c c diagrammatic calculus that rescales the ii-crossing c i,i, c i,i, for all i I and X, inverts the orientation of each strand and sends : +j j k k +j +k +j y +k +k j k j y y y y iw i   = jw j        
+i +k +j i j j i k j +i +j +j

35

Properties and symmetries of 2-category U

36

This transformation preserves the degree of a diagram, so by extending to sums of diagrams we get a 2-functor : U U given by : U U 1 Ei 1 {t} 1 Ei 1 {t}

(3.43)

It is straight forward to check that is a strict 2-functor. In fact, it is a 2-isomorphism since its square is the identity. Rescale, reect across the y-axis, and send : The operation on diagrams cccc cccc   c c c c that rescales the ii-crossing  i,i,  i,i, for all i I and X, reects a diagram across the y-axis, and sends to leaves invariant the relations on the 2-morphisms of U. This operation +j j k j +k +k +j y +k +k j k j y y x y iw i   = jw j       
+i +k +j i j j j j i +j +k +i

is contravariant for composition of 1-morphisms, covariant for composition of 2-morphisms, and preserves the degree of a diagram. Hence, this symmetry gives a 2-isomorphism : U U op 1 Es1 Es2 Esm1 Esm 1 {t} 1 Esm Esm1 Es2 Es1 1 {t}

and on 2-morphisms maps linear combinations of diagrams to the linear combination of the diagrams obtained by applying the above transformation to each summand. The relations on U are symmetric under this transformation, and is a 2-functor. The square of is the identity. Reect across the x-axis and invert orientation: Here we are careful to keep track of what happens to the shifts of sources and targets +j +i +k +j i j j {t} y y y +k +k j k j {t } y jw iw   i = jw  w       
+i +k +j i j j {t} +j +k +k j k j{t }

36

Properties and symmetries of 2-category U

37

The degree shifts on the right hand side are required for this transformation to preserve the degree of a diagram. This transformation preserves the order of composition of 1morphisms, but is contravariant with respect to composition of 2-morphisms. Hence, by extending this transformation to sums of diagrams we get a 2-isomorphism given by : U U co 1 Ei 1 {t} 1 Ei 1 {t}

(3.44)

and on 2-morphisms reects the diagrams representing summands across the x-axis and inverts the orientation. Again, the relations on U possess this symmetry so it is not dicult to check that is a 2-functor. Furthermore, it is clear that is invertible since its square is the identity. It is easy to see that these 2-functors commute with each other on-the-nose. That is, we have equalities = , = , = (3.45) The composite 2-functor is given by : U U coop 1 Es1 Es2 Esm1 Esm 1 {t} 1 Esm Esm1 Es2 Es1 1 {t} and is given on 2-morphisms by rotating diagrams by 180 . The following transformation only diers from by a shift and is given by taking adjoints. Rotation by 180 (taking right adjoints): This transformation is a bit more subtle because it uses the almost biadjoint structure of U, in particular, the calculus of mates (see [21, Section 4.3]). For each 1 x1 U denote its right adjoint by 1 y1 . The symmetry of rotation by 180 can also be realized by the 2-functor that sends a 1-morphism 1 x1 to its right adjoint 1 y1 and each 2-morphism : 1 x1 1 x 1 to its mate under the adjunctions 1 x1 1 y1 and 1 x 1 1 y 1 . That is, is mapped to its right dual . Pictorially,
x y

=
y

37

 Karoubi envelope, U, and 2-representations

38

This transformation is contravariant with respect to composition of 1-morphisms and 2morphisms. We get a 2-functor : U 1 Es1 Es2 Esm1 Esm 1 {t} U coop 1 Esm Esm1 Es2 Es1 1 {t + t }

(3.46)

where the degree shift t for the right adjoint 1 Esm Esm1 Es2 Es1 1 {t + t }, determined by (3.42), ensures that is degree preserving. Inspection of the relations for U will reveal that they are invariant under this transformation so that is a 2-functor. We can dene an inverse for given by taking left adjoints. We record this 2-morphism here. 1 : U 1 Es1 Es2 Esm1 Esm 1 {t} U coop 1 Esm Esm1 Es2 Es1 1 {t + t } .

(3.47)

with degree shift t determined from (3.42) and the left dual of the 2-morphism dened in [21, Section 4.3]. Remark 3.19. The composition : U U gives 2-isomorphism that xes all diagrams and only eects the grading shifts. Remark 3.20. There are degree zero isomorphisms of graded k-vector spaces U (f x, y) U (x, (f )y), U (x, gy) U ( 1 (g)x, y), for all 1-morphisms f, g, x, y in U , dened at the end of Section 3.1.1. (3.48) (3.49)

3.4

Karoubi envelope, U , and 2-representations

The Karoubi envelope Kar(C) of a category C is an enlargement of the category C in which all idempotents split (see [21, Section 9] and references therein). There is a fully faithful functor C Kar(C) that is universal with respect to functors which split idempotents in C. This means that if F : C D is any functor where all idempotents split in D, then F extends uniquely (up to isomorphism) to a functor F : Kar(C) D (see for example [4], Proposition 6.5.9). Furthermore, for any functor G : C D and a natural transformation : F G, extends uniquely to a natural transformation : F G. When C is additive the inclusion C Kar(C) is an additive functor.

38

 Karoubi envelope, U, and 2-representations

39

Denition 3.21. Dene the additive k-linear 2-category U to have the same objects as U and hom additive k-linear categories given by U(, ) = Kar (U(, )). The fully faithful additive k-linear functors U(, ) U(, ) combine to form an additive k-linear 2-functor U U universal with respect to splitting idempotents in the hom categories U(, ). The composition functor U(, ) U ( , ) U(, ) is induced by the universal property of the Karoubi envelope from the composition functor for U. The 2-category U has graded 2-homs given by HOMU (x, y) :=
t

HomU (x{t}, y).

(3.50)

Denition 3.22. A 2-representation of U is a (weak) graded additive k-linear 2-functor : U M , where M is a graded additive k-linear 2-category with a translation. A 2-representation of U is an additive k-linear 2-functor : U M that respects the grading. This happens when there is an additive k-linear 2-functor M M , with M a graded additive k-linear 2-category, making the diagram U

G U

(3.51)

 M

 G M

weakly commutative. Thus, to study 2-representation of U it suces to study 2-representation of U and then restrict to degreepreserving 2-morphisms. A 2-representation of U is a additive k-linear 2-functor : U M that respects the grading. Let M be a 2-category as above in which idempotents split. Any 2-representation : U M gives a unique (up to isomorphism) 2-representation : U M. The 2-functor is obtained from by restricting to the degree preserving 2-morphisms of U and using the universal property of the Karoubi envelope. This is illustrated schematically below:
restrict to degree 0 2-morphisms

U o

G
?_

Karoubian envelope

 M o

?_

 M

ooo U ooo o ooo ooo oo ooo ooo o ooo wooo G

Remark 3.23. The 2-functors , , , on U extend to 2-functors on U, for which we

39

Direct sum decompositions use the same notations. For example, : U U Ei 1 {t}, e Ei 1 {t} , (e) ()

40

(3.52)

and the other 2-morphisms , , and are dened analogously. In particular, each 1 morphism in U has left and right adjoints.

3.5

Direct sum decompositions

Recall that E 1 is the direct sum of Ei 1 over all i Seq(): E 1 :=

i Seq()

Ei 1 .

(3.53)

Due to the existence of the homomorphism , in the formula (3.30) any degree 0 idempotent e of R() gives rise to the idempotent , (e) of E 1 and to the 1-morphism (E 1 , , (e)) of U. Introduce idempotents y y y y m(m1) 2 U(Emi 1 , Emi 1 ), ei,m = (1) U(Emi 1 , Emi 1 ) e+i,m =    

similar to the idempotent ei,m in [16, Section 2.2] and [17]. Dene 1-morphisms E+i(m) 1 and Ei(m) 1 in U by E+i(m) 1 := (E+im 1 , e+i,m ) Ei(m) 1 := (Eim 1 , ei,m ) As in [16, 17], we have direct sum decompositions E+im 1 (E+i(m) 1 )[m]i! , =
(a )

m(1 m) i i 2 2 m(1 m) i i 2 2

, .

(3.54) (3.55)

Eim 1 (Ei(m) 1 )[m]i ! . =

(a ) (a )

For any divided power sequence i = (1 i1 1 , 2i2 2 , . . . , m imm ) dene Ei 1 := (Eb1 , ei ), i 40

Direct sum decompositions where i is the sequence 1 i1 , . . . , 1 i1 , 2 i2 , . . . 2 i2 , . . . , m im . . . , m im = (1 i1 )a1 (2 i2 )a2 . . . (m im )am , with term 1 i1 repeating a1 times, term 2 i2 repeating a2 times, etc., and ei = e1 i1 ,a1 e2 i2 ,a2 . . . em im ,am

41

(3.56)

is the horizontal product of idempotents. When interested in only one part of a sequence i , we write . . . i . . . instead of i = i i i and E...i ... 1 instead of Ei 1 = Ei i i 1 . Proposition 3.24. For each i, j I, i = j, and X there are 2-isomorphisms of 1-morphisms in U
d+1 2

E...+i(2a) +j+i(d+12a) ... 1 = E...i(2a) ji(d+12a) ... 1 =

d 2

E...+i(2a+1) +j+i(d2a) ... 1 ,

a=0 d 2

a=0 d+1 2

E...i(2a+1) ji(d2a) ... 1 ,

a=0

(3.57)

a=0

where d = dij = i, jX . Proof. These isomorphisms follow from categoried quantum Serre relations [16, Proposition 2.13] and [17, Proposition 6] between idempotents in rings R(), via homomorphisms , and , . Proposition 3.25. For each i I, X there are 2-isomorphisms in U Ei +iii 1 Ei i+ii 1 [ i, ]i Ei i 1 = Ei i+ii 1 Ei +iii 1 [ i, ]i Ei i 1 = where = + i . X Proof. Set + i = . The decomposition Ei +iii 1 Ei i+ii 1 [ i, ]i Ei i 1 for = X i, 0 is given by 2-morphisms in U : :
1

if i, + i 0, X if i, + i 0, X

Ei +iii 1 Ei i+ii 1 [ i, ]i Ei i 1 Ei i+ii 1 [ i, ]i Ei i 1 Ei +iii 1

(3.58) (3.59)

where and 1 consist of matrices of diagrams y i i i s  0 for 0 s i, 1, = , s := i, 1+j . . . sj j=0 u  i, 1  y   i, 1 . . . i, 1s . . . . 1 = i  i 41

 K0 (U ) and homomorphism

42

Note that all bubbles that appear in s above are fake bubbles. One can check that 1 = IdEi +iii 1 and that 1 = IdEi i+ii 1 [ i, ] 1 using the sl2 -relations (for details see [21]). Here we have taken [ i, ] 1 = 1 {1 i, } 1 {2s + 1 i, } 1 { i, 1} (3.60)

so that and 1 have degree zero. The isomorphism Ei i+ii 1 Ei +iii 1 [ i, ]i = Ei i 1 for i, + i X 0 is given similarly (see [21]). Proposition 3.26. For each i, j I, i = j, X there are 2-isomorphisms E...+ij...1 E...j+i...1 = (3.61)

Proof. The degree zero 2-isomorphism E...+ij...1 E...j+i...1 for i = j is given by maps = y : E...+ij...1 E...j+i... 1 j i y : E...j+i...1 E...+ij... 1 j  i To see that these maps are isomorphisms use (3.13).

3.6

K0 (U) and homomorphism

K0 (U ) can be viewed as a pre-additive category or, alternatively, as an idempotented ring. When thought of as a category, it has objects , over all X. The abelian group of morphisms K0 U(, ) is dened as the (split) Grothendieck group of the additive category U(, ). The split Grothendieck group K0 (A) of an additive category A has generators [P ], over all objects P of A, and relations [P ] = [P ] + [P ] whenever P P P . In the = case of U(, ) the generators are [Ei 1 {t}, e], where = + i X , t , and e EndU (Ei 1 {t}) EndU (Ei 1 ) = EndU (Ei 1 ) = (3.62)

is an idempotent (degree zero idempotent when viewed as an element of the larger ring ENDU (Ei 1 )). The dening relations are [Ei 1 {t}, e] = [Ei 1 {t }, e ] + [Ei 1 {t }, e ] whenever there is an isomorphism in U(, ) (Ei 1 {t}, e) (Ei 1 {t }, e ) (Ei 1 {t }, e ). = (3.64) (3.63)

Moreover, K0 U(, ) is a [q, q 1 ]-module, with multiplication by q coming from the grading shift [Ei 1 {t + 1}, e{1}] = q[Ei 1 {t}, e]. (3.65) We write [Ei 1 ] instead of [Ei 1 , 1], where 1 is the identity 2-morphism of Ei 1 . 42

 K0 (U ) and homomorphism

43

The space of homs between any two objects in U(, ) is a nite-dimensional k-vector space. In particular, the Krull-Schmidt decomposition theorem holds, and an indecompos able object of U(, ) has the form (Ei 1 {t}, e) for some minimal/primitive idempotent e. Any presentation of 1 = e1 + + ek into the sum of minimal mutually-orthogonal idempotents gives rise to a decomposition
k

Ei 1 {t} =
r=1

(Ei 1 {t}, er )

(3.66)

into a direct sum of indecomposable objects of U (, ). Any object of U(, ) has a unique presentation, up to permutation of factors and isomorphisms, as a direct sum of indecomposables. Choose one representative b for each isomorphism class of indecomposables, up to grading shifts, and denote by B(, ) the set of these representatives. Then {[b]}b is a basis of K0 U(, ) , viewed as a free [q, q 1 ]-module. Composition bifunctors U(, ) U( , ) U (, ) induce [q, q 1 ]-bilinear maps K0 U(, ) K0 U ( , ) K0 U (, ) (3.68) (3.67)

turning K0 (U) into a [q, q 1 ]-linear additive category with objects X. Alternatively, we may view K0 (U) as a non-unital [q, q 1 ]-algebra K0 (U)(, )
,X

(3.69)

with a family of idempotents [1 ]. The set B := ,X B(, ) gives rise to a basis [B] := {[b]}bB of idempotented [q, q 1 ]-algebra K0 (U). Notice that basis elements are dened up to multiplication by powers of q; we will not try to choose a canonical grading normalization here. Multiplication in this basis has coecients in [q, q 1]. Both A U and K0 (U) are idempotented [q, q 1 ]-algebras, with the idempotents 1 and [1 ] labelled by X. To relate the two algebras, send 1 to [1 ] and, more generally, Ei 1 to [Ei 1 ] for all i SSeqd.

Proposition 3.27. The assignment Ei 1 [Ei 1 ] extends to morphism : A U K0 (U). Multiplication by q corresponds to the grading shift {1}.

[q, q 1]-algebra homo(3.70)

Proof. K0 (U) is a free [q, q 1 ]-module, so it is enough to check that the assignment above extends to a homomorphism of (q)-algebras

(q) : UK0 (U) 43

[q,q 1 ]

(q)

(3.71)

 K0 (U ) and homomorphism

44

(A U is also a free [q, q 1 ]-module, but this fact is not needed in the proof). Propo sitions 3.24, 3.25, and 3.26 show that dening relations of U lift to 2-isomorphisms of and, therefore, descend to relations in the Grothendieck group K0 (U). 1-morphisms in U 1 Restricting (q) to A U gives a homomorphism of [q, q ]-algebras with the image of the homomorphism lying in K0 (U). For each , X homomorphism restricts to a homomorphism of 1 (A U)1 K0 U (, ) . [q, q 1 ]-modules (3.72)

Proposition 3.28. Homomorphism intertwines (anti)automorphisms , , , of A U with (anti) automorphisms [], [], [], and [ ] of K0 (U ), respectively, i.e., the following diagrams commute.
AU

G K0 ( U )  G K0 ( U )
e []

AU

G K0 (U)
[e ]

AU

G K0 (U)
[e]

AU

G K0 (U)
[e]

 U A

 U A

 G K0 (U)

 U A

 G K0 (U)

 U A

 G K0 (U)

[] denotes the induced action of on the Grothendieck group, etc. Proof. The proof follows from denitions and our construction of . The 2-isomorphisms , , on U give isomorphisms of graded k-vector spaces U x, y U x, y U x, y U (x), (y) , = (U )op (x), (y) = (U )co (x), (y) =

= U (x), (y) , = U (y), (x) .

(3.73) (3.74) (3.75)

On Grothendieck rings these isomorphisms give equalities x, y x, y x, y = = = (x), (y) , (x), (y) , (y), (x) , (3.76) (3.77) (3.78) (3.79)

which should be compared with Propositions 26.1.4 and 26.1.6 in [24] and property (v) of the semilinear form. The last equality expressed in terms of the bilinear form (, ), with x replaced by (x), gives the identity (x, y) = (y, x).

44

Idempotented rings

45

3.7

Idempotented rings

An idempotented ring A is an associative ring, not necessarily unital, equipped with a system of idempotents {1x }, over elements x of some set Z. We require orthogonality 1x 1y = x,y 1x and decomposition A=
x,yZ

1y A1x .

(3.80)

By a (left) A-module we mean an A-module M such that M=

xZ

1x M.

(3.81)

In this paper three collections of idempotented rings appear: Lusztigs U and its integral form A U. Here Z = X, the weight lattice, U=
,X

1 U1 ,

AU

=
,X

1 (A U)1 .

(3.82)

The Grothendieck groups K0 (Kar(U )) and K0 (U), the latter dened in Section 3.6. Again, the parameterizing set Z = X. We only study K0 (U) =
,X

[1 ]K0 (U)[1 ],

(3.83)

with {[1 ]}X being the system of idempotents in K0 (U). Map : A U K0 (U) is a homomorphism of idempotented rings. For each , X the -graded ring
U

:=
i ,j

HOMU (Ei 1 , Ej 1 ),

(3.84)

where the sum is over all i , j SSeq with i X , j X = . Thus, the sum is over all sequences such that Ei 1 , Ej 1 have left weight . The parameterizing set Z = {i SSeq | i X = }. The category U (, ) is equivalent to the category of right nitely-generated graded projective U -modules and grading preserving homomorphisms. The equivalence functor
U (, ) pmod U

(3.85)

takes Ei 1 to
i , P

:=
j Z

HOMU (Ei 1 , Ej 1 ), 45

Surjectivity of and, more generally, an object (Ei 1 , e) to

i ,,e P

46

:=
j Z

HOMU (Ei 1 , e), Ej 1 .

The Grothendieck group K0 U(, ) is isomorphic to the Grothendieck group of pmod U .

Notice that we get idempotented rings from the 2-category U in various ways: 1) as the Grothendieck ring/pre-additive category K0 (U) of U , 2) as rings associated to categories U (, ). The 2-category U can itself be viewed as an idempotented monoidal category. We encode these observations into a diagram (small) pre-additive o 2-categories Grothendieck category/ring
!)

G idempotented additive monoidal categories Categories/rings of homs between objects

(small) pre-additive categories o

G idempotented rings

3.8

Surjectivity of

To prove surjectivity of we will analyze the diagram of K-groups and their homomorphisms induced by the diagram (3.37) and trace minimal idempotents there, but rst recall some basics of Grothendieck groups of nite-dimensional algebras (as a model example) and graded algebras. 3.8.1 K0 of nite-dimensional algebras

A homomorphism of rings : A B induces a homomorphism of K0 -groups K0 () : K0 (A) K0 (B) of nitely-generated projective modules. For denition and properties of K0 we refer the reader to [28], [32, Chapter II]. Assume that A and B are nite-dimensional k-algebras, for a eld k, and is a kalgebra homomorphism. If is surjective then K0 () is surjective as well. On the level of idempotents, if 1 = e1 + + ek is a decomposition of 1 A into a sum of mutuallyorthogonal minimal idempotents, then Aes is an indecomposable projective A-module, 46

Surjectivity of

47

K0 (A) is a free abelian group with a basis {[Aer ]}rS , for a subset S {1, . . . , k}. S is any maximal subset with the property that Aes Aet as left A-modules for any s, t S, s = t. Applying to the above decomposition results in the equation B 1 = (e1 ) + + (ek ). (3.86)

where each (es ) is either 0 or a minimal idempotent in B. Relabel minimal idempotents so that (e1 ), . . . , (em ) = 0, (em+1 ) = = (ek ) = 0 (elements of S get permuted as well). Then 1 = (e1 )+ +(em ) is a decomposition of 1 B into a sum of mutually orthogonal minimal idempotents, B(er ) is an indecomposable projective B-module, 1 r m, and {[B(er )]}rS{1,...,m} is a basis of K0 (B). Remark 3.29. If A is a nite-dimensional k-algebra, the quotient map A A/J(A), where J(A) is the Jacobson radical of A, induces an isomorphism of K0 -groups K0 (A) K0 (A/J(A)). = (3.88) (3.87)

Indeed, J(A) is a nilpotent ideal, J(A)N = 0 for suciently large N, and the quotient by a nilpotent ideal induces an isomorphism of K0 s, see [32, Chapter II, Lemma 2.2]. Proposition 3.30. If A, B are nite-dimensional k-algebras such that all simple A- and B-modules are absolutely irreducible over k, then K0 (A) K0 (B) K0 (A k B) = (3.89)

via an isomorphism which takes [P ] [Q] for projective A, respectively B, modules P and Q to [P k Q]. Proof. By passing to A/J(A), B/J(B) and using the above remark, we reduce to the case of semisimple A and B. Then both A and B are nite products of the eld k and the proposition follows. 3.8.2 K0 of graded algebras

From here on we only consider -graded k-algebras, for a eld k. For a -graded k-algebra A = a Aa denote by K0 (A) the Grothendieck group of nitely-generated graded left projective A-modules. K0 (A) is a [q, q 1 ]-module. Throughout this subsection we assume that all weight spaces Aa are nite-dimensional, and the grading is bounded below: Aa = 0 for all a 0. Let P I(A) be the set of isomorphism classes of indecomposable graded projective Amodules, up to a grading shift. We can normalize the grading and choose one representative Q for each element of P I(A) so that 0 is the lowest nontrivial degree of Q. We write Q P I(A). 47

Surjectivity of Proposition 3.31. For A as above, K0 (A) is a free {[Q]}QP I(A) .

48 [q, q 1 ]-module with the basis

Proof. Since each weight space of A is nite-dimensional, the Krull-Schmidt property holds for graded projective nitely-generated A-modules. Any such module has a unique, up to isomorphism, decomposition as a direct sum of indecomposables, and K0 (A) is a free abelian group with a basis labelled by isomorphism classes of indecomposable projectives. Boundedness of A from below ensures that an indecomposable projective is not isomorphic to itself with a shifted grading, implying that K0 (A) is a free [q, q 1 ]-module and the rest of the proposition. We say that a 2-sided homogeneous ideal J of A is virtually nilpotent if for any a the weight space (J N )a = 0 for suciently large N. Proposition 3.32. For A as above and J a virtually nilpotent ideal of A the quotient map AA/J induces an isomorphism K0 (A) K0 (A/J). = Proof. Proposition follows from the lifting idempotents property. This is the graded version of Lemma 2.2 in [32, Chapter 2]. Proposition 3.33. Let : AB be a surjective homomorphism of nite-dimensional graded k-algebras. Then the induced map K0 () is surjective. Proof. The argument is essentially the same as in the nongraded case discussed earlier. In the bases of K0 (A) and K0 (B) given by indecomposable projective modules, the map K0 () sends some basis elements of K0 (A) to 0 and the rest go bijectively to the basis of K0 (B) (possibly after grading shifts). Finally, we discuss K0 of graded idempotented algebras. Let A be an associative graded k-algebra, possibly nonunital, with a family of mutually-orthogonal degree 0 idempotents 1x A, x Z, such that A= 1y A1x
x,yZ

(compare with the denition of idempotented ring in Section 3.7). We say that A is a graded idempotented k-algebra. By a graded nitely-generated projective A-module we mean a homogeneous direct summand of a nite direct sum (with nite multiplicities) of graded left A-modules A1x {t}, over x Z and t . By K0 (A) we denote the corresponding Grothendieck group, which is again a [q, q 1 ]-module. We assume that for each x, y Z the graded k-vector space 1y A1x is bounded below and has nite-dimensional weight spaces. Proposition 3.34. For A as above, K0 (A) is a free [q, q 1 ]-module with a basis given by isomorphism classes of indecomposables, up to grading shifts. Proof. The proof is essentially the same as that of Proposition 3.31. The dierence is in the absence of a canonical grading normalization for a representative Q of an isomorphism class of indecomposables up to grading shifts. This normalization can be chosen ad hoc, of course. 48

Surjectivity of 3.8.3 A triangle of K0 s -graded and K0 are

49

We will work in the graded case, so that the rings are modules. Consider the diagram of [q, q 1 ]-modules K0 R() R( )
K0 ()

[q, q 1]-

 K0 ENDU (E, 1 )

K0 () @
K0 ()

G K0 (R, , )

(3.90)

given by applying the K0 functor to the commutative triangle in (3.37). Recall that in [16, 17] we constructed an isomorphism of [q, q 1 ]-modules K0 (R()) A f , = where A f is the weight summand of the [q, q 1 ]-algebra A f. Likewise, (3.92) (3.91)

K0 (R( )) A f . =

R(), respectively R( ), is a free nite rank graded module over its center Z(R()) = Sym(), respectively Sym( ), isomorphic to the k-algebra of polynomials in several homogeneous generators, all of positive degree. Hence, R() R( ) is a free nite rank graded module over the central graded polynomial algebra Sym() Sym( ) . This algebra contains a homogeneous augmentation ideal Sym+ of codimension 1. Let J = R() R( ) Sym+ (3.93)

be the corresponding 2-sided ideal of R() R( ) , and consider the quotient algebra R := R() R( ) /J R()/ R() Sym+ () = R is a nite-dimensional k-algebra, and the quotient map : R() R( ) R induces an isomorphism of K0 -groups K0 ( ) : K0 (R() R( ) ) K0 (R), (3.96) (3.95) R( )/ R( ) Sym+ ( ) . (3.94)

since the ideal J is virtually nilpotent, see Proposition 3.32. We proved in [16, 17] that any simple graded R()-module is absolutely irreducible for any eld k, same for simple graded R()/ R()Sym+ -modules. Also note that K0 ( ) = 1 [q, q ], since is a graded local ring. The chain of isomorphisms K0 R() R( ) K0 (R) K0 (R()) K0 (R( )) A f A f = = = establishes the following result: 49 (3.97)

Surjectivity of Proposition 3.35. There is a canonical isomorphism K0 R() R( ) A f A f =

50

(3.98)

induced by isomorphisms K0 (R()) A f and K0 (R( )) A f constructed in [16, 17]. = = This proposition gives us a grip on the top term in the diagram (3.90). Proposition 3.36. K0 () is surjective. Proof. Since is surjective, (J) is a 2-sided ideal of R, ,. Start with a commutative square of surjective algebra homomorphisms R() R( )
/J

G G R, ,
/(J)

 R

 G G R, , /((J))

and apply functor K0 to obtain a commutative diagram K0 R() R( )

/J K0 ()

G K0 R, ,
/(J)

 K0 (R)

 G K0 R, , /((J))

The vertical arrows are isomorphisms since J and (J) are virtually nilpotent ideals. The bottom arrow is surjective, by Proposition 3.33, since R and R, , are nite-dimensional over k. Surjectivity of the top arrow follows. Corollary 3.37. K0 () is surjective. These observations are summarized in the following enhancement of (3.90) K0 (R() R( ) A f A f )=
K0 ()

 K0 (ENDU (E, 1 ))

0 () K @ @
K0 ()

G G K0 (R, ,) .

(3.99)

50

Surjectivity of 3.8.4 Idempotents in U

51

Let 1 = e1 + + ek , respectively 1 = e1 + + ek , be a decomposition of 1 EndU (E 1 ) R()0 , respectively 1 EndU (E 1 ) R( )0 , into the sum of minimal = = mutually-orthogonal idempotents. Here R()0 denotes the degree 0 subalgebra of R(). Each term in the decomposition
k k

1=
r=1 r =1

er er 1

(3.100)

of 1 R() R( ) is a minimal degree 0 idempotent, in view of discussion preceding Proposition 3.35. Let (3.101) er,r := (er er 1) EndU (E, 1 ) be the corresponding idempotent in the endomorphism algebra of E, 1 which may not be minimal. We can decompose
k(r,r )

er,r =
r =1

er,r ,r

(3.102)

into a sum of minimal mutually-orthogonal degree zero idempotents er,r ,r EndU (E, 1 ). Homomorphism induces a surjection of Grothendieck groups and maps each minimal idempotent er er 1 either to 0 or to a minimal degree 0 idempotent in R, , (Proposition 3.36). Consequently, for each (r, r , r ) the image (er,r ,r ) is either 0 or a minimal idempotent in R, , . Moreover, for each (r, r ) at most one of (er,r,r ) = 0 in R, , . We can relabel idempotents so that (er,r,1 ) = 0 and (er,r,r ) = 0 for r > 1 whenever (er er 1) = 0. Necessarily, (er,r,r ) = 0 for all r if (er er 1) = 0. If (er er 1) = 0 then er,r ,r I, , . A homogeneous element a I, , can be written as a nite sum a = u as as , where as , as are homogeneous, s=1 as U (E, 1 , Ei(s) 1 ), as U (Ei(s) 1 , E, 1 ), (3.103) and i(s) SSeq with i(s) < + . Indeed, an element of I, , can be written as a linear combination of diagrams with U-turns. Cutting each diagram in the middle allows us to view it as composition E, 1 Ei(s) 1 E, 1 , with the length + . i(s) of the sequence i(s) strictly less than the sum of lengths y y  51   i(s)

Surjectivity of Choose such a decomposition for each

u

52

er,r ,r I, , ,

er,r ,r =
s=1

as as ,

(3.104)

where in the notations we suppress dependence of u(r, r , r ), as (r, r , r ) and as (r, r , r ) on the three parameters. Multiplication by er,r ,r is the identity endomorphism of (E, 1 , er,r ,r ). We can view this indecomposable 1-morphism of U as an indecomposable projective mod ule (call it P ) over the graded idempotented ring U , = + X X . Then the identity endomorphism of P factors through projective module Q corresponding to the direct sum u s=1 Ei(s) 1 {ts }
Id

GQ

9G

for some ts . Therefore, P is isomorphic to a direct summand of Q, and the 1-morphism (E, 1 , er,r ,r ) of U is isomorphic to a direct summand of u Ei(s) 1 {ts }. s=1 Dene the width P of an indecomposable 1-morphism P Ob(U(, )) as the smallest m such that P is isomorphic to a direct summand of Ei 1 {t} for some i SSeq, i = m and t . For example, if P has width 0, then P is isomorphic to a direct summand of 1 {t} for some and t. One-morphism 1 {t} is indecomposable, since its endomorphism ring U(1 {t}, 1{t}) = k, or 0 (a possibility if the calculus is degenerate). This implies that any width zero 1-morphism is isomorphic to 1 {t}. Lemma 3.38. If P has width m then P is isomorphic to a direct summand of E, 1 {t} for some , [I], + = m, X, and t .

Proof. If i, j I and i = i i+ji has length m, then P is direct summand of Ei 1 {t} if an only if it is a direct summand of Ei +jii 1 {t}. Indeed, these two 1-morphisms are either isomorphic (if i = j) or dier by direct summands Ei i 1 {t }, all whose indecomposable summands have width at most m2, and thus cannot be isomorphic to P . By assumption, P is isomorphic to a direct summand of Ei 1 {t} with i = m. Moving all positive terms of i to the left of all negative terms produces a sequence j (j ) with j , j positive, j + j = m and P being a summand of Ej (j ) 1 {t}. But this 1-morphism is a direct summand of E, 1 {t} with , respectively , being the weight of j , respectively j . Proof of Theorem 1.1. We show that [P ] is in the image of : A U K0 (U ) by induction on the length of P . Let P have length m. Then P is a direct summand of E, 1 {t} for , as above. By shifting the degree of P down by t, P (E, 1 , er,r ,r ) for at least = one minimal idempotent er,r ,r . We must have r = 1 and (er,r ,r ) = 0, for otherwise P is isomorphic to a direct summand of s Ei(s) 1 {ts }, and since i(s) = m 2, the width of P is at most m 2, a contradiction. Thus r = 1 and (er,r,r ) = 0. We have (3.105) [P ] = [E, 1 , er,r,1 ], 52

Injectivity of in the nondegenerate case and

53

k(r,r )

[E, 1 , er,r ,r ] = [E, 1 , er,r ],

r =1

(3.106)

from (3.102). For r > 1 we have (er,r ,r ) = 0, and (E, 1 , er,r,r ) is isomorphic to a direct summand of a nite sum of Ei 1 {t}, for sequences i of length m 2. Each indecomposable summand of (E, 1 , er,r,r ) has length at most m 2. By induction hypothesis, [E, 1 , er,r ,r ] belongs to the image of for all 2 r k(r, r ). Thus,
k(r,r )

[P ] = [E, 1 , er,r ,1 ] = [E, 1 , er,r ]

r =2

[E, 1 , er,r ,r ] [E, 1 , er,r ] + (A U).

It now suces to show that [E, 1 , er,r ] belongs to image of . But the idempotent er,r is the image of er er 1 in R()R( ) , and the Grothendieck group of the latter is isomorphic to A f() A f( ). Therefore, [E, , er,r ] is in the image of A f() A f( ) under the composition map
A f()

A f( ) xy

G 1 (A U)1 G x+ y 1 , x x+ : y y :

G K0 U(, )

+ A f() U , A f( ) U .

This completes the proof of surjectivity of .

3.9

Injectivity of in the nondegenerate case

Assume that our graphical calculus is nondegenerate for a given root datum and eld k, so that Bi ,j , is a basis of HOMU (Ei 1 , Ej 1 ) for all i , j and . Then gdim HOMU (Ei 1 , Ej 1 ) =
t

q t dimk U (Ei 1 {t}, Ej 1 ).

(3.107)

Since the calculus is nondegenerate, gdim HOMU (Ei 1 , Ej 1 ) = Ei 1 , Ej 1 , and Ei 1 , over all i , , span U, the (3.108)

(q)-algebra homomorphism (q) : U K0 (U) [q,q ] (q)

1

(3.109)

53

54 intertwines the (q)-semilinear forms , on U and gdim HOMU (, ) on K0 . The latter form, which we denote , , is given by [P ], [Q]

:=
t

q t gdim U(P {t}, Q)

(3.110)

for any two 1-morphisms P, Q U (, ), and extends to the entire K0 (U) [q,q1 ] (q) via (q)-semilinearity and the orthogonality condition x, y = 0 for x K0 U(, ) , y K0 U( , ) unless = and = . By Proposition 2.5, , is nondegenerate on U. Therefore, (q) is injective, implying that is injective.

4
4.1

Categorication of U for sln

Forms of quantum sln

We consider various forms of the quantized enveloping algebra of sln corresponding to the root datum of the Dynkin graph
1

n1

For this root datum, any weight X can be written as = (1 , 2 , . . . , n1 ), where i = i, . The algebra Uq (sln ) is the (q)-algebra with 1 generated by the elements Ei , Fi and 1 Ki for i = 1, 2, . . . , n 1, with the dening relations

Ki Ki1 = Ki1 Ki = 1, Ki Ej Ki1 = q ij Ej ,

Ki K j = Kj K i ,

(4.1) (4.2)

Ki Fj Ki1 = q ij Fj ,

where i i = 2, i j = 1 if j = i 1, and i j = 0 otherwise, Ki Ki1 Ei Fj Fj Ei = ij , q q 1 Ei2 Ej (q + q 1 )Ei Ej Ei + Ej Ei2 = 0 if j = i 1, Fi2 Fj (q + q 1 )Fi Fj Fi + Fj Fi2 = 0 if j = i 1, Ei Ej = Ej Ei , Fi Fj = Fj Fi if |i j| > 1. (4.3) (4.4) (4.5) (4.6)

Recall that U(sln ) is obtained from Uq (sln ) by adjoining a collection of orthogonal idempotents 1 indexed by the weight lattice X of sln , 1 1 = 1 , (4.7)

54

The 2-category U (sln ) such that if = (1 , 2 , . . . , n1 ), then Ki 1 = 1 Ki = q i 1 , where + iX = (1 + 2, 2 1, 3 , . . . , n2 , n1) (1 , 2 , . . . , n2 , n1 1, n1 + 2) (1 , . . . , i1 1, i + 2, i+1 1, . . . , n1) Ei 1 = 1+iX Ei , Fi 1 = 1iX Fi ,

55

(4.8)

if i = 1 if i = n 1 (4.9) otherwise.

The [q, q 1 ]-algebra A U(sln ) is the integral form of U(sln ) generated by products of a Ei Fia (a) (a) divided powers Ei 1 := [a]! 1 , Fi 1 := [a]! 1 for X and i = 1, 2, . . . , n 1. The relationships are collected below: add idempotents integral form

Uq (sln )

G U(sln ) o

?_

A U(sln )

4.2

The 2-category U (sln )

We introduce a 2-category U that is dened analogously to U in the sln -case, but the R()-relations have been modied to the signed R()-relations given in [17]. Namely, the R()-relations in U are replaced in U by the signed R()-relations obtained from the oriented graph 1 2 3 G n1 , G G G (4.10) with vertices enumerated by the set {1, 2, . . . , n1}, using signs ij = ji = 1 for all edges. It was observed in [17] that the resulting ring R () is isomorphic to R(). In Section 4.2.1, following the denition of U , we extend this isomorphism to an isomorphism U U of 2-categories. The 2-category U is more convenient for constructing a representation on iterated ag varieties in Section 6. In general it is a poor practice to set up an isomorphism rather than an equivalence of categories, not to mention 2-categories. However, having an isomorphism U U is justied, since U and U have the same objects, morphisms, and generating 2-morphisms. Denition 4.1. U (sln ) is a additive k-linear 2-category with translation. The 2-category U (sln ) has objects, morphisms, and generating 2-morphisms as dened in (3.1), but some of the relations on 2-morphisms are modied. The sl2 relations and the shift isomorphism relations are the same as before, see equations (3.1)(3.9). All 2-morphisms are cyclic with respect to the biadjoint structure as before, see (3.3) and (3.10). The relations (3.13) hold. 55

The 2-category U (sln ) The signed R() relations are (a) For i = j, the relations y y y y = j i (i j) (b) For i = j, the relations y y =
i j

56

y
i

y
j

if i j = 0,

y y
i j

y y
j

if i j = 1. (4.11)

y
i

y
j

y
j

y =
i

y j

(4.12)

for all . (c) Unless i = k and j = i 1 y y y

i

y y y
j k

y =
i

y y y y
k

y
j

(4.13)

For i j = i 1 y = (i j) y y
j i i i

y
i

y
j

y y y

y y
i

y y y

y
j

y
i

(4.14)

56

The 2-category U (sln ) 4.2.1 The 2-isomorphism : U U

57

Dene an isomorphism of 2-categories : U U on objects by mapping , and on hom categories by graded additive k-linear functors : U(, ) U (, ) Ei 1 Ei 1 y
i1 i im

(4.15) (4.16) y
im

(1)i
i1

y
i1

y
im

i i+1

(1)i+1
i1 i1

y
i1

y
im

if i = i+1 , or i i+1

i i+1

y
im

im

otherwise

i i+1

i1

im

im

im

for all orientations for all orientations

i i+1

i i+1

i1

i i+1

i1

i i+1

Above, the i in (1)i refers to the enumeration of the vertex i in (4.10). One can check that the above transformations respects the sl2 -relations and the cyclic condition. Furthermore, maps the R()-relations to their signed analogs by rescaling the generators as above. is a 2-functor and an isomorphism of 2-categories. Remark 4.2. Since the 2-categories U and U are isomorphic, by the universal property of the Karoubi envelope, their Karoubi envelopes are isomorphic as well.
Dene U to be the graded additive k-linear category which has the same objects and 1-morphisms as U and 2-morphisms U (x, y) := t

U (x{t}, y).

(4.17)

4.2.2

Relation to rings R()

Regard the graded k-algebra R() with system of idempotents {1i } as a pre-additive klinear category whose objects are {i | i Seq()}. The k-vector space of morphisms from i to i is i R()i . The composition i R()i i R()i i R()i is given by multiplication in R(). For any weight there is a graded additive k-linear functor
: R() U (, + X )

(4.18)

57

58 that takes object i to Ei 1 , and is given on generators of homs by

j R()i U (Ei 1 , Ej 1 )

... ...
i1 i im

y (1)i y y y y (1)i+1 i1 i y y y
i1 i1 i im

y y
i+1 im

if i = i+1 , or i i+1

...
i1 i

...
i+1 im

y
im

otherwise.

i i+1

Linear combination of diagrams in R() get sent to the corresponding rescaled linear com bination in U (, + X ) with the weight labelling the far right region. In what follows we will refer to those diagrams in the image of as R() generators. The image of R() is spanned by diagrams with all strands upward pointing and no caps or cups.

5
5.1

Iterated ag varieties
Cohomology of n-step varieties

The material in this section generalizes that of [21, Section 6]. The reader is encouraged to start there for more examples and greater detail in the sl2 case. We enumerate by I = {1, 2, . . . , n 1} the vertex set of the Dynkin diagram of sln
1

n1

Fix N 0, and consider the variety F l(n) of n-step partial ags F F = (0 = F0 F1 . . . Fn = in

N N

. The dimensions of the subspaces Fi are conveniently expressed as a vector, dimF = (dim F0 , dim F1 , dim F2 , . . . , dim Fn ).

The connected components of F l(n) are parameterized by non-negative integers k = (k0 , k1 , k2 , . . . , kn ) such that 0 = k0 k1 k2 . . . kn = N. The connected component F l(k) corresponding to k consists of all ags F such that dimF = k. Throughout this section we refer to the terms k of k with the convention that k = 0 N 58 if 0, if n. (5.1)

Cohomology of n-step varieties The cohomology algebra of F l(k) is Hk := H (F l(k), k) =

0k1 (k2 k1 )(N kn1 ) + -graded,

59

H 2 (F l(k), k).

For 1 j n, 0 < kj kj1, let x(k)j, be a formal variable of degree 2. The ring Hk is isomorphic to the quotient ring
n

k[x(k)j,1 , x(k)j,2, . . . , x(k)j,kj kj1 ] /Ik,N ,

j=1

(5.2)

where Ik,N is the ideal generated by the homogeneous terms in the equation
n

(1 + x(k)j,1 t + x(k)j,2t2 + + x(k)j,kj kj1 tkj kj1 ) = 1.

j=1

(5.3)

Above, t is a formal variable used to keep track of the degrees. For notational convenience we add variables x(k)j,0 and set x(k)j,0 = 1. Furthermore, we set x(k)j, = 0 if > kj kj1 . (5.4)

It is helpful to express the above relation in an alternative form. Let x(k)i, denote the homogeneous term of degree 2 in the product
n

(1 + x(k)j,1 t + x(k)j,2t2 + + x(k)j,kj kj1 tkj kj1 ).

j=1, j=i

(5.5)

For example, if n = 4 and k = (1, 3, 4, 7), then equation (5.3) becomes 1 + x(k)1,1 t 1 + x(k)2,1 t + x(k)2,2 t2 1 + x(k)3,1 t 1 + x(k)4,1 t + x(k)4,2 t2 + x(k)4,3 t3 = 1 and the terms x(k)2, are given by omitting the second term and multiplying out the rest 1 + x(k)1,1 t 1 + x(k)3,1 t 1 + x(k)4,1 t + x(k)4,2 t2 + x(k)4,3 t3 so that x(k)2,0 = 1, x(k)2,1 = x(k)1,1 + x(k)3,1 + x(k)4,1 , x(k)2,2 = x(k)1,1 x(k)3,1 + x(k)1,1 x(k)4,1 + x(k)3,1 x(k)4,1 + x(k)4,2 , x(k)2,3 = x(k)1,1 x(k)3,1 x(k)4,1 + x(k)1,1 x(k)4,2 + x(k)3,1 x(k)4,2 + x(k)4,3 , x(k)2,4 = x(k)1,1 x(k)3,1 x(k)4,2 + x(k)1,1 x(k)4,3 + x(k)3,1 x(k)4,3 , x(k)2,5 = x(k)1,1 x(k)3,1 x(k)4,3 , 59 (5.6)

Cohomology of n-step varieties and x(k)2, = 0 for > 5. It is clear that (5.3) can be written

60

x(k)j,f x(k)j,f = ,0
f =0

(5.7)

for any 1 j n, where ,0 is the Kronecker delta. We call the elements x(k)j, dual generators in light of (5.7). For 1 i n 1 let
+i k

= (+i k)0 , (+ik)1 , (+ik)2 , . . . , (+ik)n ,

0 = (+i k)0 (+i k)1 (+i k)n = N,

be the sequence obtained from the sequence k by increasing the ith term by one
+i k

:= (k0 , k1 , k2 , . . . , ki1 , ki + 1, ki+1 , . . . , kn )

(5.8)

if ki + 1 ki+1 , or by setting the sequence to the empty sequence if ki = ki+1 . Namely, (+i k)j = kj if j = i and (+i k)i = ki + 1 if ki + 1 ki+1 . When ki +1 ki+1 then H+ik is the cohomology ring of the partial ag variety consisting of ags F with dimF = +i k. The ring H+i k is given by
n

H+i k =
j=1

k[x(+i k)j,1, x(+i k)j,2 , . . . , x(+i k)j,(+ik)j (+i k)j1 ] /I+ik,N k[x(+i k)j,1 , . . . , x(+i k)j,kj kj1 ] k[x(+i k)i,1 , . . . , x(+i k)ki ki1 +1 ]
j=i,i+1

k[x(+i k)i+1,1 , . . . , x(+i k)ki+1 ki 1 ]/I+i k,N where I+i k,N is the ideal generated by the homogeneous terms in
n

(5.9)

x(+i k)j, t
j=1 0

= 1.

(5.10)

We dene H = 0. Going back to the example of n = 4 and k = (1, 3, 4, 7), then

+3 k

= (1, 3, 5, 7), so that

H+3 k = k[x(+3 k)1,1 , x(+3 k)2,1 , x(+3 k)2,2 , x(+3 k)3,1 , x(+3 k)3,2 , x(+3 k)4,1 , x(+3 k)4,2 ]/I+3 k,7 (5.11) where I+3 k,7 is the ideal generated by the homogeneous terms in (1 + x(+3 k)1,1 )(1 + x(+3 k)2,1 + x(+3 k)2,2 )(1 + x(+3 k)3,1 + x(+3 k)3,2 ) (1 + x(+3 k)4,1 + x(+3 k)4,2 ) = 1. (5.12)

Similarly, we write i k = (k0 , k1, k2 , . . . , ki1 , ki 1, ki+1 , . . . , kn ) for the sequence k where we have subtracted one from the ith position whenever ki1 ki 1. The cohomology ring of the ag variety F l(i k) is Hi k , which can be expressed explicitly in terms of generators as above. When ki1 = ki then i k = and Hi k := H = 0. 60

Cohomology of n-step varieties 5.1.1 Flag varieties for the action of Ei and Fi

61

For 1 i n 1, dene k +i = k i = (k0 , k1 , k2 , . . . ki , ki + 1, ki+1 , . . . , kn1, kn ) (k0 , k1 , k2 , . . . ki1 , ki 1, ki , . . . , kn1 , kn ) if ki+1 ki + 1, otherwise. if ki 1 ki1 , otherwise.

For k i = the variety F l(k i ) is the component of F l(n + 1) consisting of ags F such that dimF = k i . The cohomology ring of F l(k i ) will be denoted by Hki . The cohomology ring Hk+i is Hk+i =
j=i+1

k[x(k +i )j,1, . . . , x(k +i )j,kj kj1 ]k[i ]k[x(k +i )i+1,1 , . . . , x(k +i )i+1,ki+1ki 1 ]/Ik+i ,N (5.13)

where Ik+i ,N is the ideal generated by the homogeneous terms in 1 + i t 1 + x(k +i )i+1,1 t + + x(k +i )i+1,ki+1 ki 1 tki+1 ki 1
kj kj1

(5.14)

j=i+1 f =0

x(k +i )j,f tf = 1.

The forgetful maps F l(k) o induce maps of cohomology rings Hk

p 1 p1

F l(k +i )

p2

G F l(+ik )

G Hk+i o

p 2

H+i k

that make Hk+i a right Hk H+i k -module. Since the algebra H+i k is commutative, we can turn a right H+i k -module into a left H+i k -module. Hence, we can make Hk+i into a (H+i k , Hk )-bimodule. In fact, Hk+i is free as a graded Hk -module and as a graded H+i k module. These inclusions making Hk+i a (H+ik , Hk )-bimodule are given explicitly as follows:
 Hk 

Hk+i for j = i + 1 , x(k +i )j, +i i x(k )i+1,1 + x(k +i )i+1, ,

(5.15)

x(k)j, x(k)i+1, and

 H+i k 

Hk+i for j = i , x(k +i )j, +i i x(k )i,1 + x(k +i )i, . 61

(5.16)

x(+i k)j, x(+i k)i,

Graphical calculus for iterated ag varieties

62

Notice that x(k)j, and x(+i k)j, for j = i, i + 1 are mapped to the same element of Hk+i . Using these inclusions we identify these elements of Hk and H+i k with their images in the bimodule Hk+i . Furthermore, we can also express the generators x(k +i )i, and x(k +i )i+1, of Hk+i as the images of certain generators in Hk or H+i k . Thus we can write Hk+i as Hk+i =
j=i+1

k x(k)j,1, . . . , x(k)j,kj kj1 k[x(+i k)i+1,1 , . . . , x(+i k)i+1,ki+1 ki 1 ] k[i ]/Ik+i ,N

or equivalently, Hk+i =
j=i

k x(+i k)j,1, . . . , x(+i k)j,kj kj1 k[x(k)i,1 , . . . , x(k)i,ki ki1 ] k[i ]/Ik+i ,N ,

where Ik+i ,N is the ideal described above. Therefore, i is the only generator of Hk+i that is not identied with a generator of Hk or H+i k under the above inclusions. Remark 5.1. The generators x(k)j, , x(+i k)j, , i of Hk+i correspond to Chern classes of tautological bundles over the variety F l(k +i ). The generator i corresponds to the line bundle Fki +1 /Fki associated to the subspaces Fki Fki +1 created by the subspace insertion k +i . Denition 5.2. The set of multiplicative i x(k)i,i x(+i k)i+1,i+1 x(k)j,j = x(+i k)j,j Hk+i generators for 0 < i ki ki1 , for 0 < i+1 ki+1 ki 1, for j = i, i + 1, and 0 < j kj kj1,

for the ring Hk+i , corresponding to the Chern class of the tautological line bundle, and to the canonical inclusions (5.15) and (5.16) of generators in H+i k and Hk into Hk+i , are called the canonical generators of Hk+i . Using commutativity we can regard Hk+i as an (Hk , H+ik )-bimodule. The generators of Hk and H+ik that are not mapped to canonical generators can be expressed in terms of canonical generators as follows: x(k)i+1, = i x(+i k)i+1,1 + x(+i k)i+1, , x(+i k)i, = i x(k)i,1 + x(k)i, , for all values of . (5.17)

5.2
5.2.1

Graphical calculus for iterated ag varieties

Rings Hk

All of what has been described can be easily visualized using a shorthand notation in which the generators of Hk are drawn as labelled bubbles oating in a region carrying a label 62

Graphical calculus for iterated ag varieties

63

called the weight. The label will be important when we relate partial ag varieties to the 2-category U categorifying A U(sln ). More precisely, will correspond to a weight of the irreducible representation of A U(sln ) with highest weight (N, 0, 0, . . . , 0). To a sequence k = (k0 , k1, k2 , . . . , kn ) as above associate = (k) = (1 , 2 , . . . , n1 ) n1 where = k+1 + 2k k1 . (5.18) The weight corresponding to the sequence +i k is dened analogously, where ki is replaced by ki + 1. Comparing with (4.9) it is clear that (+i k) = + iX . Similarly, (i k) = iX . With this convention, the generators of Hk and H+i k are depicted as follows: (5.19) x(k)j, := j, + iX x(+i k)j, :=
j,

(5.20)

where the identity is depicted by the empty region of the appropriate weight. Products of generators are depicted by a bubble in the plane for each generator present in the product. Diagrams are only considered up to planar isotopy. A generic element in Hk can be depicted as a formal linear combination of such diagrams. For example, if n = 4, k = (1, 3, 4, 7) then the element x(k)1,1 x(k)4,3 + 5 x(k)3,1 Hk is represented as
1,1 4,3

+5
3,1

If we depict the dual generators x(k)j, of Hk dened in (5.5) as j, x(k)j, = then the dening relations (5.7) for Hk become the equations
j,f f =0

j,f

j,g j,g

= ,0 (5.21)

=
g=0

for 0 and 1 j n. Notice that j,0 j,0 = = 1,

j,

=0

if < 0, or kj kj1 < ,

see (5.4) and the comments preceding (5.4).

63

Graphical calculus for iterated ag varieties 5.2.2 Bimodules Hk+i

64

The identity element in Hk+i is represented by a vertical line labelled i Hk+i 1 := + iX

i

where the orientation indicates that we are regarding Hk+i as an (H+ik , Hk )-bimodule. The on the right hand side is the weight corresponding to k. Hence, having on the right hand side of the diagram indicates the right action of Hk on Hk+i . Similarly, the + iX on the left indicates the left action of H+ik on Hk+i . When regarding Hk+i as an (Hk , H+ik )-bimodule we depict it in the graphical calculus with the opposite orientation (a downward pointing arrow). Hk+i 1 :=
i

+ iX

We will often omit the weights from all regions but one, with it understood that crossing an upward pointing arrow with label i from right to left changes the weight by iX , and crossing a downward pointing arrow from right to left changes the weight by iX . Equations (5.17) show that all of the generators from Hk+i except for i can be interpreted as either generators of Hk or H+i k under the natural inclusions. This fact is represented in the graphical calculus as follows: Hk+i x(k)j, :=
i j,

j =i+1 j=i

(5.22) (5.23)

Hk+i x(+i k)j, := Hk+i i :=

i

j, i

(5.24)

where each diagram inherits a grading from the Chern class it represents (deg x(k)j, = 2, deg x(+i k)j, = 2, and deg i = 2). Equation (5.22) is meant to depicts the generator x(k)j, Hk+i as the element x(k)j, Hk acting on the identity of Hk+i . Likewise, the generator x(+i k)j, Hk+i is depicted as the element x(+i k)j, H+ik acting on the identity of Hk+i . The generator i is represented by a dotted line so that i is represented by dots on a line, but for simplicity we write this using a single dot and a label to indicate the power. The identication (5.15) and (5.16) of x(k)j, with x(+i k)j, in Hk+i for j =, i, i + 1 leads to the graphical identity:
j, i

j, i

for j = i, i + 1.

(5.25)

64

Graphical calculus for iterated ag varieties Similarly, (5.17) provides the identities
i+1, i i, i

65

= =

i+1,1 i

i+1, i

(5.26) (5.27)

i,1 i

+
i

i,

expressing non-canonical generators in terms of canonical generators. It is sometimes helpful to express the canonical generators in terms of non-canonical generators:
i+1, i

=
f =0

(1)f
i

i+1,f

(5.28)

i, i

=
f =0

(1)f

i,f i

(5.29)

which can be veried using (5.26) and (5.27). From these equations we can derive other useful identities:

= (1)

f =0

i+1,f i i,g g=0 i

i+1,f

(5.30) (5.31)

= (1)

i,g

Proposition 5.3.
j, i i,1 i f =0

i f

if j = i, i + 1 if j = i

j, i

i,

(1)f

i+1,f i

if j = i + 1 (5.32)

65

Graphical calculus for iterated ag varieties

66

j, i f =0

if j = i, i + 1 if j = i

j, i

(1)f
i

i,f

i+1,1 i

+
i

i+1,

if j = i + 1 (5.33)

Proof. Recall that the elements x(k)j, are sums of homogeneous symmetric terms in all variables except for x(k)j, . By (5.25) all terms x(k), for = i, i + 1 can be slid across the line labelled i. The case when j = i, i + 1 then reduces to the problem of sliding symmetric combinations =0 x(k)i,f x(k)i+1,f across a line labelled i. Such slides are determined f by the following calculation in Hk+i :
f

x(k)i,f x(k)i+1,f
f =0

(5.29) f =0 g=0

(1)g x(+i k)i,f g ig x(k)i+1,f

(5.26)

(1)g x(+i k)i,f g x(+i k)i+1,f 1 ig+1

f =0 g=0 f

+
f =0 g=0

(1)g x(+i k)i,f g x(+i k)i+1,f ig

(5.34)

Change variables to f = f + 1, and g = g + 1 in the rst summation, so that all terms cancel except for g = 0 term in the second summation; this term is equal to

x(+i k)i,f x(+i k)i+1,f .

f =0

(5.35)

Hence, when j = i, i + 1 the elements x(k)j, and x(+i k)j, can be slid across a line labelled i. The dual generator x(k)j, contains symmetric homogeneous combinations of variables x(k), for = j. When j = i in (5.32) all terms in x(k)i, slide across lines labelled i except for the variables x(k)i+1, . Using (5.26) to slide these across establishes (5.32) for the case j = i. Similarly, when j = i + 1 all terms in x(k)i+1, slide from right to left across lines labelled i except for the variables x(k)i, . Using (5.29) to slide these completes the proof of (5.32). Equation (5.33) is proven similarly.

66

Graphical calculus for iterated ag varieties

67

By duality, analogous equations as those above hold for downward pointing arrows. For example, equation (5.30) implies
i

= (1)

f =0

i,f i

i,f

(5.36)

5.2.3

Bimodules Hki

For the remainder of this paper we write a signed sequence i = 1 i1 2 i2 . . . m im as i = s1 s2 . . . sm with s = i . For s = i write s k for the sequence obtained from k = (k0 , k1 , k2 , . . . , kn ) by increasing the ki by 1 if = + and i < i+1 , decreasing the sequence by 1 if = and i1 < i , and setting the sequence to otherwise. Then i k is either set to , or else it denotes the sequence obtained from k with the ki th term increased by one if s = +i , or decreased by one if s = i , sequentially for each s in the signed sequence i , reading from the right. It is clear that the sequence i k is equal to the sequence j k whenever i , j SSeq with i X = j X X and i k = and j k = . We write Hki for the (Hi k , Hk )-bimodule Hki := Hs2 s3 sm ks1 Hs
2 s3 sm k

Hs

m1 sm k

Hsm ksm1 Hsm k Hksm .

(5.37)

This bimodule can also be described as the cohomology ring of the variety of m + n step iterated partial ags corresponding to the sequence obtained from k by the ordered insertion of subspaces determined from the signed sequence i . The (Hi k , Hk )-bimodule Hki can be understood using the graphical calculus. A general signed sequence i = s1 s2 sm is represented by a sequence of lines coloured by the sequence i1 i2 im , where the line coloured by i is oriented upward if s = +i and oriented downward if s = i . Examples: 1. For the signed sequence +j + i SSeq consider the (H+j+i k , Hk )-bimodule Hk+j+i = H+i k+j H+i k Hk+i . As explained in the previous section, the identity elements of the (H+i k , Hk )-bimodule Hk+i and the (H+i+j k , H+ik )-bimodule H+i kj are depicted as + iX
i

and

+ iX + jX
j

+ iX

(5.38)

respectively. The identity element of Hk+j+i is represented by the diagram + iX + jX

j

+ iX
i

(5.39)

67

Graphical calculus for iterated ag varieties

68

The region in the middle of the two lines is labelled by the weight +iX corresponding to +i k. The tensor product over the action of H+i k is represented diagrammatically by the fact that a labelled bubble in the region with weight +iX can be equivalently regarded as an element of H+ik acting on the line corresponding to Hk+i , or the line corresponding to H+ikj . Action of Hk
, j i

Action of H+i k
j , i

Action of H+j+ik
, j i

The weight on the far right, and the weight of + iX + jX on the far left, indicate that this diagram is describing an (H+j+i k , Hk )-bimodule where the various actions are depicted as above. 2. For i = +i1 + i2 + i3 + ir the identity element of the bimodule Hki is depicted by a sequence of upward oriented labelled lines Hk i 1 =
i1 i2 i3

im1 im

(5.40)

The canonical generators i of each term in the tensor product (5.37) are represented graphically by a dot on the line labelled i . Hki 1 1 1 i 1 1 =
i1 i2

im

Likewise, the tensor product over the rings Hs s+1 sm k is represented by the regions between lines. Again, the weights on the far left and right of the diagram indicate the bimodule structure. Action of Hk
i1 i2 im j,

Action of Hi im k
i1

Action of H+i k
j, i1 i2

j, i

im

im

3. Consider the (Hk , Hk )-bimodule corresponding to the tensor product Hk+i H+i k Hk+i (5.41)

where in the rst factor we are regarding Hk+i as a (Hk , H+i k )-bimodule. The identity element of this bimodule is represented by the diagram
i

+ iX
i

(5.42)

68

Graphical calculus for iterated ag varieties 5.2.4 Identities arising from tensor products

69

Bubbles with a given label (j, ) can pass from right to left, or left to right, through a line coloured by i as long as j = i, i + 1. If j = i or j = i + 1 then a bubble can move through a line subject to the rules (5.26)(5.29). Furthermore, dots on a line can be exchanged for bubbles in the neighboring regions using (5.30) and (5.31). Dual bubbles corresponding to dual generators can be slid across lines using the rules (5.32) and (5.33). The following Lemma is needed for the denition of the 2-representation given in the next section. In particular, parts (i) and (ii) are used to provide two equivalent denitions of t , , # i i and (iii) and (iv) are used to show that t
i

Lemma 5.4. The following identities hold: i) Equivalent denitions of cups labelled i: in the ring Hk+i H+i k Hk+i we have
f =0

(1)f
i

i,f i

g=0

(1)g

g i,g i i

(5.43)

f =0

(1)f if x(k)i,f

g=0

(1)g x(k)i,g ig

for all

ii) In the ring Hi k+i Hi k Hi k+i we have

f =0

(1)f
i

i+1,f i

g=0

(1)g

g i+1,g i i

(5.44)

f =0

(1)f if x(k)i+1,f =

g=0

(1)g x(k)i+1,g ig

for all

.
(5.45)

iii) Dot slide formulas for cups: in the ring Hk+i H+i k Hk+i we have
ki ki1 f =0

(1)ki ki1 f if +1 x(k)i+1,ki ki1 f =

ki ki1 g=0

(1)ki ki1 g ig x(k)i+1,kiki1 g i .

69

Graphical calculus for iterated ag varieties iv) Dot slide formulas for cups: In the ring Hi k+i Hi k Hi k+i we have
ki+1 ki f =0

70

(1)ki+1 ki f if +1 x(k)i+1,ki+1 ki f =
ki+1 ki g=0

(5.46)

(1)ki+1 ki g ig x(k)i+1,ki+1 ki g i.

Proof. Part i) follows from the chain of equalities:

(1)
f =0

f i

f i,f i

(5.31)

(1)
f =0 g=0

i,g i

i,f g i

i,f

If we re-index by letting f = f and switch the order of summation we have

g

=
g=0 f =0

(1)

i,g i

i,(g)f i

i,f

(5.31)

(1)g
g=0

g i,g i i

which after re-indexing completes the proof of Part i). Part ii) is proven similarly. Part iii) follows from Part i) by letting f = f + 1 and adding and subtracting the term (1)+1 1 x(k)i,+1 , so that

(1)f if +1 x(k)i,f
f =0 +1

=
Part i)

(1)+1f if x(k)i,+1f (1)+1 1 x(k)i,+1

f =0 +1

(1)+1g x(k)i,+1g ig (1)+1 1 x(k)i,+1 .

g=0

(5.47) Pulling o the g = 0 term and re-indexing g = g 1 we have

(1)g x(k)i,g ig +1 + (1)+1 x(k)i,+1 1 1 x(k)i,+1 .

g =0

But the term with the summation is equal to

g =0

(1)g ig x(k)i,g i

(5.48)

by Part i), and the remaining terms (1)+1 x(k)i,+1 1 1 x(k)i,+1 , are zero when = ki ki1 since x(k)i,+1 is zero for ki ki+1 , see (5.4). Part iv) is proven similarly using Part ii). 70

The 2-category FlagN Corollary 5.5. The assignments (see Denition 6.1) y
i,

71

: Hk

1 and y
i,

Hk+i ki ki1
f =0

H+i k Hk+i {1 + ki1 ki+1 } (1)ki ki1 f if x(k)i,ki ki1 f

: Hk

Hi k+i ki+1 ki
g=0

Hi k Hi k+i {1 + ki1 ki+1 }

(1)ki+1 ki g ig x(k)i+1,ki+1 ki g

dene morphisms of graded bimodules of degree 1 + i = 1 ki1 + 2ki ki+1 and 1 i = 1 + ki1 2ki + ki+1 , respectively. Proof. For the rst claim it suces to check that the left action of each generator of Hk on y (1) Hk+i H+i k Hk+i is equal to right action of this generator. The Corollary i, follows since bubbles can slide across lines by (5.26)(5.29) at the cost of introducing powers of i on one of the tensor factors; by Lemma 5.4 (iii) and (iv), factors of i can be slid across tensor factors in the above sums. The second claim is proven similarly and the degrees of these bimodule maps are easily computed.

5.3

The 2-category FlagN

Bim is a 2-category whose objects are graded rings, whose 1-morphisms are graded bimodules, and 2-morphisms are degree-preserving bimodule homomorphisms. Idempotent bimodule homomorphisms split in Bim, so that any 2-representation : U Bim extends uniquely (up to isomorphisms) to a 2-representation U Bim, see Section 3.4. Graded 2-homs between graded bimodules M1 and M2 are given by HOMBim (M1 , M2 ) :=
t

HomBim (M1 {t}, M2 ).

(5.49)

Let Bim be the 2-category with the same objects and 1-morphisms as Bim and 2morphisms given by Bim (M1 , M2 ) := HOMBim (M1 , M2 ). (5.50) We now dene a sub 2-category FlagN of the 2-category Bim for each integer N
+.

Denition 5.6. The additive k-linear 2-category FlagN is the idempotent completion inside of Bim of the 2-category consisting of objects: the graded rings Hk for all k = (k0 , k1 , k2 , . . . , kn ) with 0 k1 k2 kn = N.

71

72 morphisms: generated by the graded (Hk ,Hk )-bimodule Hk , the graded (H+ik ,Hk )bimodules Hk+i and the graded (Hk ,H+i k )-bimodule Hk+i for all i I, together with their shifts Hk {t}, Hk+i {t}, and Hk+i {t} for t . The bimodules Hk = Hk {0} are the identity 1-morphisms. Thus, a morphism from Hk to Hi k is a nite direct sum of graded (Hj k , Hk )-bimodules of the form Hkj {t} := Hs2 s3 sm ks1 Hs
2 s3 sm k

Hs

m1 sm k

Hsm ksm1 Hsm k Hksm {t}

for signed sequences j = s1 s2 . . . sm with i X = j X X. 2-morphisms: degree-preserving bimodule maps. There is a graded additive subcategory Flag of Bim with the same objects and N 1-morphisms as FlagN , but with Flag (M1 , M2 ) := N
t In Section 6 we show that provides a 2-representation of U ; using the isomorphism : U U and restricting to degree zero 2-morphisms, the subcategory FlagN provides a 2-representation of U.

FlagN (M1 {t}, M2 ).

(5.51)

Flag N

Representing U on the ag 2-category

In this section we dene for each positive integer N a 2-representation N : U Flag . N The 2-functor N is degree preserving so that it restricts to a weak 2-functor : U FlagN . We will sometimes shorten N to for simplicity.

6.1

Dening the 2-functor N

On objects the 2-representation N : U Flag sends = (1 , 2 , . . . , n1 ) to the ring N Hk when = (k), i.e., when = k+1 + 2k k1 .

N : U Flag N Hk 0 N : U Flag N Hk {t} 1 {t} 0 E+i1 {t} Ei1 {t}

if = (k) otherwise.

Morphisms of U get mapped by N to graded bimodules: if = k+1 + 2k k1 otherwise. if = k+1 + 2k k1 otherwise.

Hk+i {t + 1 + ki1 + ki ki+1 } 0 Hki {t + 1 ki } 0

if = k+1 + 2k k1 otherwise. 72

Dening the 2-functor N

73

Here Hki {t + 1 ki } is the bimodule Hki with the grading shifted by t + 1 ki so that (Hk+i {t + 1 ki })j = (Hk+i )j+t+1ki . More generally, the 1-morphism Ei 1 {t} = Es1 1+(s2 )X ++(sm1 )X +(sm )X Esm1 1+(sm )X Esm 1 {t} (6.1)

is mapped by N to the graded bimodule Hei k with grading shift {t + t }, where t is the sum of the grading shifts for each terms of the composition in (6.1). Formal direct sums of morphisms of the above form are mapped to direct sums of the corresponding bimodules. 6.1.1 Biadjointness

Denition 6.1. The 2-morphisms generating biadjointness in U are mapped by to the following bimodule maps. Hk Hk+i H Hk+i {1 + ki1 ki+1 } +i k ki ki1 f (1)ki ki1 f i,ki ki1 f f =0 t (6.2) : # i i i ki ki1 1 (1)ki ki1 f f x(k)i,k k f

f =0 (6.3) i i ki+1 ki 1 (1)ki+1 ki f if x(k)i+1,ki+1 ki f f =0 Hk+i H k Hk+i {1 + ki1 ki+1 } Hk +i 1 2 (1)1 +2 +1+ki ki+1 i+1,1 +2 +1+ki ki+1 (6.4) : i i i1 i2 (1)1 +2 +1+ki ki+1 x(k)i+1,1 +2 +1+ki ki+1 Hi k+i H k Hi k+i {1 + ki1 ki+1 } Hk i 1 2 (1)1 +2 +1+ki1 ki i,1 +2 +1+ki1 ki : (6.5) i i i1 i2 (1)1 +2 +1+ki1 ki x(k)i,1 +2 +1+ki1 ki

Hk

f =0

i1

Hi k+i Hi k Hi k+i {1 + ki1 ki+1 } (1)ki+1 ki f

f i+1,ki+1 ki f

ki+1 ki

73

Dening the 2-functor N

74

Corollary 5.5 shows that the cups above are bimodule maps. It is clear that the caps are bimodule maps. These denitions preserve the degree of the 2-morphisms of U dened in Section 3.1. By Lemma 5.4, the clockwise oriented cap and cup have degree 1 i and the counter-clockwise oriented cap and cup have degree 1 + i so that these assignments are degree preserving. 6.1.2 R() generators are mapped by N to the graded

y Denition 6.2. The 2-morphisms and in U  i, i, bimodule maps: Hk+i {1 + ki1 ki+1 } y : + iX i i Hk+i {1 + ki1 ki+1 }  + iX : i i y
i

Hk+i {1 + ki1 ki+1 } (6.6) i+1 (6.7) i+1 Hk+i {1 + ki1 ki+1 }

Note that these assignment are degree preserving since these bimodule maps are degree 2. c  cccc c c  The 2-morphisms c i,j, and ccc1 i,j, are mapped by to the graded bimodule maps: y
j

H+j k+i H+j k Hk+j H+i k+j H+i k Hk+i

j 2 i1

(6.8) if i j = 0

1 1 f =0

i1

j 2

i1 +2 1f if
j 2

2 1

g=0 i1 +1 j 2 +1 i1 j 2 i1 {1}

i1 +2 1g ig {1}

if i = j if i if
j i G j G

Hj ki Hj k Hkj Hi kj Hi k Hki
j 2 i1

(6.9) if i j = 0

i1 j 2

2 1 f =0

i1 +2 1f if

1 1 g=0

i1 +2 1g ig

if i = j if i if
j i G j G

j 2 i1 {1} j 2 +1 i1 j 2 i1 +1 {1}

It is straightforward to see that these assignments dene bimodule maps of degree i j. In the case when i = j the bimodule map is just the divided dierence operator acting on i and j . 74

Checking the relations of U

75

6.2

Checking the relations of U

In this section we show that the relations on the 2-morphisms of U are satised in Flag , N thus establishing that is a graded additive k-linear 2-functor. From the denitions in the previous section it is clear that preserves the degrees associated to generators. For this reason, we often simplify our notation in this section by omitting the grading shifts {t} when no confusion is likely to arise. Proposition 6.3. preserves the sl2-relations of U . Proof. The proof in [21] that the sl2-relations of U are preserved by generalizes immediately with only minor renements to grading shifts and summation indices. The proof only makes use of (5.21) and (5.25)(5.33). For example, to prove biadjointness we must show + iX iX + iX iX y   y  = y  y =

+ iX iX iX for all strings labelled by i. We will prove the rst equality by computing the bimodules maps on elements i Hk+i . + iX + iX y  y : i (1)
ki ki1 f =0

(1)

i,(ki ki1 f ) i

i,ki ki1 f

But,

ki ki1 f =0

i,(ki ki1 f ) i

i,ki ki1 f

(5.31)

(6.10)

since the sum can be taken to by removing terms that are equal to zero when < ki ki1 , or by adding terms that are equal to zero if > ki ki1 . The others are proven similarly.

75

Checking the relations of U Lemma 6.4. i  y

j

76

+ iX =
i j

Hk+j Hk Hk+i H+i+j k+i H+i+j k H+i k+j j 1 i2 i2 j 1 (6.11) H+j+i+i k H+j+i k H+j k+i Hk+i Hk Hk+j j 1 i2 i2 j 1 (6.12)

 + iX y
j

y y
i

= 

These bimodule maps have degree zero for all i, j I and all weights . Proof. We compute the bimodule maps directly using the denitions in the previous section. The case when i = j appears in [21] so we will omit this case here. The map in (6.11), using (5.43) for the cup, is given by i j y y   y
j i

+ iX : j 1 i2
i,ki ki1 f i j 1

ki ki1

(1)2
f =0

i,2 (ki ki+1 f )

+ iX

(6.13)

if i j = 0 or

j G , and 1 i,2 (ki ki+1 1f ) i j 1 +1 i j i,2 (ki ki+1 f )

ki ki1 1

(1)2
f =0 ki ki1 1

i,ki ki1 1f

+ iX

f =0

(1)2 1

i,ki ki1 1f

+ iX (6.14)

if G . Careful calculation, keeping in mind the weights of each region, will show that these maps have degree zero. In both cases, the dual generators can slide across the line labelled j via (5.33). After changing indices, equations (6.13) and (6.14) both become
(+j k)i (+j k)i1

(1)
f =0

i,f i

i,2 f j

+ i

(6.15)

76

Checking the relations of U

77

where (+j k)i (+j k)i1 are the (i1)-st and i-th terms in +j k. By adding or removing terms that are equal to zero, depending on whether 2 is greater than or less than (+j k)i (+j k)i1 , the above summation can be taken to 2 , so that (5.31) completes the proof. Proposition 6.5. The equality 
i j

y
j

of graded bimodule maps holds in Flag for all weights . N

 

y y y


j

(6.16)

Proof. By direct calculation, using Lemma 6.4, we have that both the far left and right bimodule maps are given by if i j = 0 j 2 i1 2 1 1 1 1 +2 1f f 1 +2 1g g i i i i if i = j g=0 f =0 i1 j 2 j i j 2 i1 {1} if G j i j 2 +1 i1 j 2 i1 +1 {1} if G which agrees with
c  c c c  c1 i,j,+iX +jX

Proposition 6.6. The equalities y  y = y  , j i j i

y y
i

 
j

= 
i

y
j

(6.17)

of graded bimodule maps hold in Flag for all weights . N Proof. This follows immediately from Lemma 6.4. Proposition 6.7. The equalities y y y y , =
i j i j

y
i

y
j

y =
i

y j

(6.18)

hold in Flag for all weights . N cc c c c Proof. Using the denition of the bimodule map ( c i,j,) in (6.8) the proposition is easily veried.

77

Equivariant representation Proposition 6.8. The equalities y y y y = j i (i j) y y

j i

78

y
i

y
j

if i j = 0 (6.19)

y y
i j

y y
j

if i j = 1

y y

y
k

of bimodule maps hold in

y
i

y
j

y = (i j) y
i i

y =
i

y y y

y y y y

y
k

unless i = k and i j = 0 y
i

(6.20)

y y y

y
j i

y
j

Flag N

for all weights .

y
i

if i j = 1 (6.21)

Proof. To prove the proposition we compute the bimodule maps in (6.19) on the elements of the form i1 j 2 . Bimodule maps in (6.20) and (6.21) are computed on elements i1 j 2 k 3 and i1 j 2 i3 , respectively. The action on other elements in the cohomology rings is determined by the fact that the maps are bimodules morphisms. The proof of these remaining relations is the same as the proof that rings R() act on Po (notation as in [16]). Replacing the variables xk (i ) Poi with the Chern classes of line bundles k in the corresponding cohomology rings turns formulas in [16, Section 2.3] for the action of dots and crossings into formulas (6.6) and (6.8), with the signs taken into account. Thus, we have proven the following Theorem:
Theorem 6.9. N : U Flag is a 2-functor and a 2-representation. N

6.3
6.3.1

Equivariant representation
Reminders on equivariant cohomology

The GL(N)-equivariant cohomology of a point [11] is given by

HGL(N ) (pt) = H Gr(N, ) = k[x1 , x2 , . . . , xN , y1 , y2, . . .]/IN,

(6.22)

where IN, is the ideal generated by the homogeneous terms in 1 + x1 t + x2 t2 + xN tN 1 + y1 t + y2 t2 + yj tj + = 1. 78

Equivariant representation
Thus, HGL(N ) (pt) is isomorphic to the polynomial ring HGL(N ) (pt) k[x1 , x2 , . . . , xN 1 , xN ] =

79

(6.23)

with xi in degree 2i. Given a sequence k = (k0 , k1 , k2 , . . . , kn ) with 0 k1 k2 kn = N, GL(N) acts transitively on F l(k), so the equivariant cohomology of F l(k) is
HGL(N ) (F l(k)) = HStab(pt) pt F l(k)

(6.24)

where the stabilizer of a point 0 k1 kn = C N in F L(k) is the group of invertible block k1 (k2 k1 ) (N kn1 ) uppertriangular matrices. This group is contractible onto its subgroup GL(k1 ) GL(k2 k1 ) GL(N kn1 ). Hence,
G Hk HGL(k1 )GL(k2 k1 )GL(N kn1 ) (pt) = n

=
j=1 n

HGL(kj kj1 ) (pt)

=
j=1

k[x(k)j,1 , x(k)j,2, . . . , x(k)j,kj kj1 ]

(6.25)

with deg x(k)j, = 2. Thus, the equivariant cohomology of F l(k) has the same generators as the ordinary cohomology ring, but we do not mod out by the ideal Ik,N . G G The equivariant cohomology rings H+ik and Hi k of F l(+i k) and F l(i k) can be similarly computed. They have the same generators as the ordinary cohomology rings H+i k and Hi k , but no relations. G The equivariant cohomology Hk+i of F l(k +i ) also has the same generators as the ordinary cohomology ring, with no relations. Using the forgetful maps F l(k) F l(k +i ) F l(+i k) we get inclusions
 G Hk 

G Hk+i

(6.26)

x(k)j, x(k)i+1, and

for j = i + 1 , x(k +i )j, +i i x(k )i+1,1 + x(k +i )i+1, ,

 G H+i k 

G Hk+i

(6.27)

x(+i k)j, x(+i k)i,

for j = i , x(k +i )j, +i i x(k )i,1 + x(k +i )i, .

G G making Hk+i a graded (H+i k , Hk )-bimodule, just as in the non-equivariant case. Using G these inclusions we introduce canonical generators of Hk+i given by identifying certain

79

Equivariant representation

80

G G G generators of Hk and H+ik with their images in Hk+i . Thus, we can identify x(k)j, and G x(+i k)j, in Hk+i when j = i, i + 1, and G Hk+i = j=i+1

k x(k)j,1 . . . x(k)j,kj kj1 k[x(+i k)i+1,1 , . . . , x(+i k)i+1,ki+1ki 1 ] k[i ]

or equivalently,
G Hk+i = j=i G G G The generators of Hk and H+i k that are not mapped to canonical generators in Hk+i can be expressed in terms of canonical generators as follows:

k x(+i k)j,1 . . . x(+i k)j,kj kj1 k[x(k)i,1 , . . . , x(k)i,ki ki1 ] k[i ].

x(k)i+1, = i x(+i k)i+1,1 + x(+i k)i+1, x(+i k)i, = i x(k)i,1 + x(k)i, for all values of . 6.3.2 The 2-category EqFlagN

(6.28)

The 2-category EqFlagN is the equivariant analog of FlagN . Denition 6.10. The additive k-linear 2-category EqFlagN is the idempotent completion inside of Bim of the 2-category consisting of
G objects: the graded rings Hk for all k = (k0 , k1 , k2 , . . . , kn ) with 0 k1 k2 kn = N. G G G G G morphisms: generated by the graded (Hk ,Hk )-bimodule Hk , the graded (H+i k ,Hk )G G G G bimodules Hk+i and the graded (Hk ,H+i k )-bimodule Hk+i for all i I, together with G G G G G their shifts Hk {t}, Hk+i {t}, and Hk+i {t} for t . The bimodules Hk = Hk {0} are G the identity 1-morphisms. Thus, a morphism from Hk to HiG is a nite direct sum k G of graded (HjG , Hk )-bimodules of the form k G G Hkj {t} := Hs2 s3 sm ks1 H G

s2 s3 sm k

H G

sm1 sm k

G Hsm ksm1 H G

sm k

G Hksm {t}

for signed sequences j = s1 s2 . . . sm with i X = j X X. 2-morphisms: degree-preserving bimodule maps. There is a graded additive subcategory EqFlag of Bim with the same objects and N 1-morphisms as EqFlagN , and EqFlag (M1 , M2 ) := N
t

EqFlagN (M1 {t}, M2 ).

(6.29)

80

Equivariant representation 6.3.3 Equivariant representation G N

81

A 2-representation of the 2-category U (sl2 ) = U (sl2 ) was constructed in [22] using equivariant cohomology of partial ag varieties. Here we extend that construction to the sln -case and dene a 2-representation G : U EqFlag . N N The verication that N is 2-representation with the assignments given in Section 6.1 used only the relations (5.21) and (5.25)(5.33), together with Lemma 5.4 and Corollary 5.5 which both follow from these relations. To dene the equivariant 2-representation G ordinary cohomology rings are replaced N by equivariant cohomology rings and bimodule maps associated to 2-morphisms are dened the same way, except that the dual generators x(k)j, must be redened. Set x(k)j,0 = 1 and inductively dene

x(k)j, =
f =1

x(k)j,f x(k)j,f .

(6.30)

For example, x(k)j,1 = x(k)j,1 x(k)j,2 = x(k)j,1x(k)j,1 x(k)j,2 = x(k)2 x(k)j,2 j,1 x(k)j,3 = x(k)3 + 2 x(k)j,1 x(k)j,2 x(k)j,3 j,1 In the nonequivariant cohomology rings the two denitions (5.5) and (6.30) of dual generators agree, but not in the equivariant cohomology ring. Remark 6.11. We could have started with the denition of x(k)j, given in (6.30) and used this denition to construct the non-equivariant representation in the previous section. After all, the two denitions (5.5) and (6.30) are equivalent in the non-equivariant cohomology ring. However, the recursive denition (6.30) makes it more cumbersome to calculate with in practice and that is why we used (5.5) to construct the non-equivariant representation. Lemma 6.12. With x(k)j, redened as in (6.30) the relations (5.21) and (5.25)(5.33)

81

Equivariant representation hold in the equivariant cohomology rings:

82

x(k)j,f x(k)j,f = ,0
f =0

(6.31) (6.32) (6.33) (6.34) (6.35) (6.36)

for j = i, i + 1 x(k)j, = x(+i k)j, x(k)i+1, = i x(+i k)i+1,1 + x(+i k)i+1, x(+i k)i, = i x(k)i,1 + x(k)i,

x(+i k)i+1, =
f =0

(1)f if x(k)i+1,f x(+i k)i,f if

f =0

x(k)i, = (1) i =
f =0

x(+i k)i+1,f x(k)i+1,f =

g=0

x(+i k)ig x(k)i,g (6.37) (6.38)

x(k)j, = x(+i k)j,

if j = i, i + 1

x(k)j, =

x(+i k)i,1 i + x(+i k)i, if j = i

f =0 f =0

(6.39)

(1) x(+i k)i+1,f if if j = i + 1 (1) x(k)i,f if if j = i (6.40)

x(+i k)j, =

Proof. Equation (6.31) follows from the denition (6.30) of the dual elements x(k)j, . Equations (6.32)(6.34) equate two images of noncanonical generators under the inclusions (6.26) and (6.27). Equations (6.35) and (6.36) follow from (6.33) and (6.34). The rst equality in (6.37) is proven as follows:
f

x(k) i+1,1 i + x(k)i+1, if j = i + 1

(1)

f =0

x(+i k)i+1,f x(k)i+1,f

(6.35) f =0 g=0

(1)+g ig x(k)i+1,f g x(k)i+1,f

g

=
g=0 (6.31) g=0

(1)+g ig
f =0

x(k)i+1,(g)f x(k)i+1,f (6.41)

(1)+g ig g,0 = i .

82

Equivariant representation

83

The second equality above is just a re-indexing of the summation. The second equation in (6.37) is proven similarly. Equation (6.38) follows from (6.32) and the denition of x(k)j,. Equations (6.39) and (6.40) follow from (6.33) and (6.34) and the denition of x(k)j,.
Theorem 6.13. G : U EqFlag is a 2-representation. N N

Proof. It is clear that G preserves the degree associated to 2-morphisms since the 2N representation N does. The proof that G preserves the relations in U can be copied N line by line from the proof the N preserves the relations. By Lemma 6.12 all identities used in the proof of Theorem 6.9 hold in EqFlag with the dual elements redened according N to (6.30). Theorem 6.14. The 2-representations N and G categorify the irreducible representation N VN of U(sln ) with highest weight (N, 0, . . . , 0). Proof. Idempotent bimodule maps split in Bim, so by the universal property of the Karoubi envelope the additive 2-functors N : U FlagN and G : U EqFlagN , N obtained using the isomorphism : U U and restricting to degree preserving 2 morphisms, extend to 2-representations of U. G The rings Hk and Hk are graded local rings so that every projective module is free, and they both have (up to isomorphism and grading shift) a unique graded indecomposable projective module. The Grothendieck group of the category k Hk gmod (respectively G 1 G k Hk gmod) is a free [q, q ]-module with basis elements [Hk ] (respectively [Hk ]) over all k, where q i acts by shifting the grading degree by i. Thus, we have K0
k G Hk gmod K0 = k

Hk gmod A VN =

(6.42)

as [q, q 1 ]-modules, where the sums are over all sequences 0 k1 kn = N, and A VN is a representation of A U(sln ), an integral form of the representation VN of U(sln ). G G The bimodules (1 ), (E+i1 ) and (Ei 1 ) (or equivalently (1 ), (E+i 1 ) and G (Ei 1 )) induce functors on the graded module categories given by tensoring with these bimodules. The functors 1 := Hk Hk : Hk gmod Hk gmod Ei 1 := Hk+i Hk {1 + ki1 ki+1 } : Hk gmod H+ik gmod Fi1+iX := Hk+i H+i k {ki } : H+i k gmod Hk gmod (6.43) (6.44) (6.45)

have both left and right adjoints and commute with the shift functor, so they induce [q, q 1 ]-module maps on Grothendieck groups. Furthermore, the 2-functor N respects the relations of U , so by Propositions 3.243.26 these functors satisfy relations lifting those of U.

83

Nondegeneracy of U (sln )

84

6.4

Nondegeneracy of U (sln )
U (1 , 1 ) U (1 , 1 ) = (6.46)

Lemma 6.15. The surjective graded k-algebra homomorphism

of Proposition 3.6 is an isomorphism.

there exists some large Proof. Injectivity is established by showing that for each M N such that the images of bubble monomials in U (1 , 1 ) in variables of degree less than G M act by linearly independent operators on Hk for = (k), under the 2-functor G . By N a direct calculation min(,ki+1 ki ) G iw  : 1 (1) x(k)i,f x(k)i+1,f , f =0
i 1+

Only one orientation for dotted bubbles labelled by vertex i is allowed in a bubble monomial in , see (3.24). The image under G of a bubble monomial is composed of products of bimodule maps of the above form where, in the equivariant cohomology ring, sums of products of element of the form (x(k)i+1, x(k)i, ), respectively (x(k)i, x(k)i+1, ) are independent provided , < ki ki1 and , ki+1 ki so that (x(k)i+1, x(k)i+1, ), respectively (x(k)i, x(k)i+1, ), is nonzero. By taking N large, we can ensure this condition is satised for any xed M. Hence, images of bubble monomials in U (1 , 1 ) are independent. Lemma 6.16. There is an isomorphism of graded k-algebras
: R() U (E 1 , E 1 ) D D (D) . D ,

G as bimodule endomorphisms of Hk . After expanding the x(k)j, using (6.30) we have i  : 1 (1) (x(k)i+1, x(k)i, ) + products of lower , G w order terms i 1+ i : 1 (1) (x(k)i, x(k)i+1, ) + products of lower . G  order terms i 1+

i G 

(6.47)

i 1+

min(,ki ki1 )

1 (1)

g=0

x(k)i+1,g x(k)i,g ,

(6.48)

with given by (4.18) and D a bubble monomial in . 84

Nondegeneracy of U (sln )

85

Proof. Surjectivity of follows from Lemma 3.9 and the isomorphism from Section 4.2.1. Injectivity of is established by showing that for each M there exists some large N such that degree M elements (D) . D ,, as D and D run over a basis of R(), respectively , act by linear independent operators under the 2-representation G . The N 2-functor G must map horizontal composites to horizontal composites (given by tensor products) in EqFlag , so that N

G G G ( (D) . D ) = G (D) Hk G (D ) := fD Hk gD

(6.49)

G G G G for bimodule maps gD : Hk Hk and fD : Hki Hkj for some i , j Seq(). Let

Po () :=
i Seq()

Poi ,

Poi = k[i 1 , i 2 , . . . , i m ],

m = ||.

For large enough N, the bimodule G (E 1 ) contains Po () as a subspace. From the denition of (see (4.18)), together with the denitions of the bimodule maps associated to R() generators (see Section 6.1.2), it is clear that the action on Po () given by the 2functor G coincides with the action of R() on Po dened in [16]. In particular, bimodule maps fD corresponding to basis elements of R() must act by linear independent operators (see [16, proof of Theorem 2.5]) on the subspace Po () of G (E 1 ). Furthermore, from the denitions of bimodule maps associated to R() generators it is also clear that fD xes all other generators of G (E 1 ). By Lemma 6.15, for large N the bimodule maps gD act by linearly independent operG ators on Hk . Write
iw

 for i, 0, (6.50) for i, < 0.

i,

:=

i, 1+

i, 1+ 

G G G and let D = 1 ,1 2 ,2 r ,r . Then D acts on Hki Hki Hk Hk via G (Id1 . D ) = = G G 1 Hk gD . From (6.47) we have 1 r

1 Hk gD : 1 1 G
f1 =0

x(k)1 +1,f1 x(k)1 ,1 f1

fr =0

x(k)1 +1,fr x(k)r ,r fr

(6.51)

G in Hki . After expanding the x(k), using their denition (6.30), we have an expression for G the action of gD on Hki strictly in terms of variables x(k), . The bimodule maps fD Hk gD are linearly independent operators since the bimodG ule maps fD and gD are separately independent and act on algebraically independent

85

REFERENCES
G generators of Hki : 1 2 m

86

i1 i2 im

1 ,...,m ,1 ,...,k

1 j1

m 1 ,1 jm

j2

k ,k

where the sum over the a s is determined by the action of fD on 1 1 mm , and the sum over the x(k)b ,b , represented by labelled boxes on the far right, is determined from (6.51) by expanding the x(k)j, . In particular, fD xes variables x(k), represented by labelled boxes on the far right and gD acts only on such boxes.

This concludes the proof of Theorem 1.3 stated in the introduction, so that : A U(sln ) K0 (U(sln )) (6.52)

is an isomorphism. Proposition 3.28 allows us to view 2-functors , , , and as categori cations of symmetries , , , and of A U(sln ). The graded homs between 1-morphisms in U(sln ) categorify the semilinear form , given by (3.110). Simple nondegeneracy observations. We can assume that k is only a commutative ring, work over this ring from the start, and say that the calculus is nondegenerate over k if U (Ei 1 , Ej 1 ) is a free k-module with a basis Bi ,j , for all j , i and . We dont know any examples of a root datum and ring k when the calculus is degenerate. Over a eld the nondegeneracy of the calculus depends only on the root datum and the characteristic of k. If the calculus is nondegenerate over , it is nondegenerate over and over any commutative ring k.

Multigrading. The multigrading introduced at the end of [17] on rings R() extends to a multigrading on each hom space U (Ei 1 , Ej 1 ). The Karoubian envelope of the corresponding multigraded 2-category should categorify the multi-parameter deformation of A U.

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] A. Beilinson, G. Lusztig, and R. MacPherson. A geometric setting for the quantum deformation of GLn . Duke Math. J., 61(2):655677, 1990. [3] J. Bernstein, I. B. Frenkel, and M. Khovanov. A categorication of the TemperleyLieb algebra and Schur quotients of U(sl(2)) via projective and Zuckerman functors. Selecta Math. (N.S.), 5(2):199241, 1999, math.QA/0002087. 86

REFERENCES

87

[4] F. Borceux. Handbook of categorical algebra. 1, volume 50 of Encyclopedia of Mathematics and its Applications. Cambridge University Press, Cambridge, 1994. [5] S. Cautis and J. Kamnitzer. Knot homology via derived categories of coherent sheaves. II, sl(m) case, 2007, math.AG/0710.3216. [6] S. Cautis and J. Kamnitzer. Knot homology via derived categories of coherent sheaves. I, sl(2)-case. Duke Math. J., 142(3):511588, 2008, math.AG/0701194. [7] J. Chuang and R. Rouquier. Derived equivalences for symmetric groups and sl 2categorication. Ann. of Math., 167:245298, 2008, math.RT/0407205. [8] 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. [9] 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. [10] 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.QA/0511467. [11] W. Fulton. Equivariant cohomology in algebraic geometry. Eilenberg lectures, Columbia University. Notes by Dave Anderson, available at http://www.math.lsa.umich.edu/dandersn/eilenberg/, 2007. [12] N. Ganter and M. Kapranov. Representation and character theory in 2-categories. Adv. Math., 217(5):22682300, 2008, math.KT/0602510. [13] M. Jimbo. A q-dierence analogue of U(g) and the Yang-Baxter equation. Lett. Math. Phys., 10(1):6369, 1985. [14] M. Kashiwara. Global crystal bases of quantum groups. Duke Math. J., 69(2):455485, 1993. [15] M. Khovanov. A functor-valued invariant of tangles. Algebr. Geom. Topol., 2:665741 (electronic), 2002, math.QA/0103190. [16] M. Khovanov and A. Lauda. A diagrammatic approach to categorication of quantum groups I, 2008, math.QA/0803.4121. [17] M. Khovanov and A. Lauda. A diagrammatic approach to categorication of quantum groups II, 2008, math.QA/0804.2080. [18] M. Khovanov and L. Rozansky. Matrix factorizations and link homology. Fund. Math., 199(1):191, 2008, math.QA/0401268. 87

REFERENCES

88

[19] A. Kleshchev. Linear and projective representations of symmetric groups, volume 163 of Cambridge Tracts in Mathematics. Cambridge U. Press, 2005. [20] P. Kronheimer and H. Nakajima. Yang-Mills instantons on ALE gravitational instantons. Math. Ann., 288(2):263307, 1990. [21] A. D. Lauda. A categorication of quantum sl(2), 2008, math.QA/0803.3652. [22] A. D. Lauda. Categoried quantum sl(2) and equivariant cohomology of iterated ag varieties, 2008, math.QA/0803.3848. [23] G. Lusztig. Canonical bases in tensor products. 89(17):81778179, 1992. Proc. Nat. Acad. Sci. U.S.A.,

[24] G. Lusztig. Introduction to quantum groups, volume 110 of Progress in Mathematics. Birkhuser Boston Inc., Boston, MA, 1993. a [25] G. Lusztig. Canonical bases and Hall algebras. In Representation theories and algebraic geometry (Montreal, PQ, 1997), volume 514 of NATO Adv. Sci. Inst. Ser. C Math. Phys. Sci., pages 365399. Kluwer Acad. Publ., Dordrecht, 1998. [26] H. Nakajima. Instantons on ALE spaces, quiver varieties, and Kac-Moody algebras. Duke Math. J., 76(2):365416, 1994. [27] M. Reineke. Monomials in canonical bases of quantum groups and quadratic forms. J. Pure Appl. Algebra, 157(2-3):301309, 2001. [28] J. Rosenberg. Algebraic K-theory and its applications, volume 147 of Graduate Texts in Mathematics. Springer-Verlag, New York, 1994. [29] R. Rouquier. Higher representation theory. Talk at IAS, March 2008. [30] R. Rouquier. Higher representations of Kac-Moody algebras. Work in progress. [31] J. Sussan. Category O and sl(k) link invariants, 2007, math.QA/0701045. [32] C. Weibel. The K-book: An introduction to algebraic K-theory. available online, http://www.math.rutgers.edu/weibel/Kbook.html. [33] H. Zheng. Categorication of integrable representations of quantum groups, 2008, arXiv:0803.3668. M.K.: 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

88