Quantum Grothendieck rings and derived Hall algebras

D. Hernandez, B. Leclerc
arXiv:1109.0862v3 [math.QA] 2 Mar 2013

Abstract
Let g = n ⊕ h ⊕ n− be a simple Lie algebra over C of type A, D, E, and let Uq (Lg) be the
associated quantum loop algebra. Following Nakajima [N3], Varagnolo-Vasserot [VV1], and
the first author [He2], we study a t-deformation Kt of the Grothendieck ring of a tensor cate-
gory CZ of finite-dimensional Uq (Lg)-modules. We obtain a presentation of Kt by generators
and relations.
Let Q be a Dynkin quiver of the same type as g. Let DH(Q) be the derived Hall algebra
of the bounded derived category Db (mod(FQ)) over a finite field
p F, introduced by Toën [T].
Our presentation shows that the specialization of Kt at t = |F| is isomorphic to DH(Q).
Under this isomorphism, the classes of fundamental Uq (Lg)-modules are mapped to scalar
multiples of the classes of indecomposable objects in DH(Q).
Our presentation of Kt is deduced from the preliminary study of a tensor subcategory
CQ of CZ analogous to the heart mod (FQ) of the triangulated category Db (mod(FQ)). We
show that the t-deformed Grothendieck ring Kt, Q of CQ is isomorphic to the positive part
of the quantum enveloping algebra of g, and that the basis of classes of simple objects of
Kt, Q corresponds to the dual of Lusztig’s canonical basis. The proof relies on the algebraic
characterizations of these bases, but we also give a geometric approach in the last section.
It follows that for every orientation Q of the Dynkin diagram, the category CQ gives a
new categorification of the coordinate ring C[N] of a unipotent group N with Lie algebra n,
together with its dual canonical basis.

Contents
1 Introduction 2

2 Cartan matrices and Auslander-Reiten quivers 5

3 Quantum tori 10

4 Quantum groups 16

5 Quantum Grothendieck rings 18

6 An isomorphism between quantum Grothendieck rings and quantum groups 24

7 A presentation of quantum Grothendieck rings 30

8 Derived Hall algebras 33

9 Quiver varieties 36

1
1 Introduction
1.1 Let g be a simple Lie algebra of type A, D, E over C. We denote by g = n⊕h⊕n− a triangular
decomposition of g. Let v be an indeterminate, and let
Uv (g) = Uv (n) ⊗Uv(h) ⊗Uv (n− )
be the corresponding Drinfeld-Jimbo quantum enveloping algebra over C(v), defined via a v-
analogue of the Chevalley-Serre presentation of U (g). Using a geometric realization of Uv (n) in
terms of perverse sheaves on varieties of representations of a quiver Q of the same Dynkin type
as g, Lusztig [Lu1] has defined a canonical basis B of Uv (n) with favorable positivity properties.
This was inspired by a seminal work of Ringel [Ri2], showing that the twisted Hall algebra of the
category mod (FQ)
p of representations of Q over a finite field F, is isomorphic to the specialization
of Uv (n) at v = |F|.

1.2 One can associate with g another quantum algebra. Let Lg = C[t,t −1 ] ⊗ g be the loop algebra
of g. Let q be a nonzero complex number, which is not a root of unity. Via a q-analogue of the
loop presentation of U (Lg), Drinfeld [D] has defined the quantum loop algebra Uq (Lg), an algebra
over C. The finite-dimensional representations of Uq (Lg) have attracted a lot of attention, because
of their connection with the trigonometric solutions of the quantum Yang-Baxter equation with
spectral parameter. In this paper we focus on a tensor subcategory CZ of the category of finite-
dimensional Uq (Lg)-modules, whose simple objects are parametrized by a discrete set (for the pre-
cise definition of CZ see [HL, §3.7] or §5.2 below). Denote by R the complexified Grothendieck
ring of CZ . Let t be another indeterminate. By works of Nakajima [N3] and Varagnolo-Vasserot
[VV1], the C-algebra R has an interesting t-deformation Rt over C(t). The first author [He2]
has introduced a slightly different deformation Kt . These t-deformations are important because
they contain for every simple object L of CZ a “class” [L]t which can be characterized by axioms
similar to those of Lusztig for the canonical basis B. As a consequence, Nakajima [N3] has shown
that one can calculate algorithmically the character of L.

1.3 Surprisingly, these deformed Grothendieck rings have not been much studied from the ring
theoretic point of view, and for instance, to the best of our knowledge, there is no available pre-
sentation by generators and relations in the literature. One of the main results of this paper (Theo-
rem 7.3) is a presentation of Kt , with a similar flavor as the familiar Drinfeld-Jimbo presentation
of Uv (n). More precisely, this presentation shows that Kt is obtained by taking an infinite number
of copies of Ut (n) labelled by m ∈ Z, and then imposing t-boson relations between generators of
copies sitting at adjacent integers, and t-commutation relations between generators of non-adjacent
copies.

1.4 Let Db (mod(FQ)) be the bounded derived category of mod(FQ). Toën [T] has attached to
this triangulated category an associative algebra called the derived Hall algebra of Db (mod(FQ))
(see also [XX]). Let DH(Q) denote the twisted derived Hall algebra obtained by twisting Toën’s
multiplication by means of the Ringel form, as in [S]. It follows from our presentation of Kt that:

p
Theorem 1.1 (a) The specialization of Kt at t = |F| is isomorphic to DH(Q).
(b) Under this isomorphism, the classes of fundamental Uq (Lg)-modules are mapped to scalar
multiples of the classes of indecomposable stalk complexes in DH(Q), and the basis of
classes of standard Uq (Lg)-modules is mapped to a rescaling of the natural basis of DH(Q)
indexed by isoclasses of objects of Db (mod(FQ)).

2
There is a similar result for the t-deformed Grothendieck ring Rt of [N3, VV1], but the twisted
derived Hall algebra should be replaced by a non-twisted one (Remark 8.4).

1.5 To obtain our presentation of Kt we first consider a tensor subcategory CQ of CZ which

“looks like mod(FQ) inside Db (mod(FQ))”. Recall that in [HL] we have introduced an increasing
sequence (Cℓ )ℓ>0 of subcategories of CZ . When Q is a bipartite orientation of the Dynkin diagram
and the Coxeter number h is even, CQ is just the subcategory Cℓ with ℓ = h/2 − 1. The general
definition of CQ for an arbitrary orientation Q will be given in §5.11 below. Let Kt,Q be the
subalgebra of Kt spanned by the elements [L]t associated with the simple objects L of CQ . Note
that Kt and Kt,Q are algebras over C(t 1/2 ), where t 1/2 is a square root of t.
The quantum algebra Uv (n) is endowed with a distinguished scalar product. Let B∗ be the
basis of Uv (n) adjoint to the canonical basis B with respect to this scalar product. Let v1/2 be a
square root of v, and set Uv (n) := C(v1/2 ) ⊗ Uv(n). The main step for obtaining the presentation
of Kt is:
∼
Theorem 1.2 (a) There is a C-algebra isomorphism Φ : Kt,Q → Uv (n) with Φ(t 1/2 ) = v1/2 .

(b) For every simple object L of CQ , the image Φ([L]t ) belongs to B∗ (up to some half-integral
power of v).

Nakajima obtained in [N4] similar results for the first subcategory C1 of [HL]. Namely, he
showed that the classes [L]t of simple objects of C1 can be identified with a subset of the basis Be∗
n). Here e
of Uv (e n denotes the positive part of the Kac-Moody algebra of rank 2 rk(g) attached to
the decorated Dynkin diagram of g, and B e is Lusztig’s canonical basis of Uv (e
n). For example, if g
has type A3 , e
g has type E6 . Note that in Theorem 1.2, we do not use en, but only n.

1.6 Let Av (n) be the graded dual of the vector space Uv (n). It can be endowed with a multipli-
cation coming from the comultiplication of Uv (g), and regarded as the quantum coordinate ring of
the unipotent group N with Lie algebra n (see e.g. [GLS]). The basis B∗ can be identified with
a basis of Av (n) called the dual canonical basis. It specializes when v 7→ 1 to a basis B of the
coordinate ring C[N].
By specializing v1/2 and t 1/2 to 1 in Theorem 1.2, we see that the complexified Grothendieck
ring RQ of CQ can be identified with the coordinate ring C[N] in such a way that the basis of RQ
consisting of the classes of simple objects becomes Lusztig’s dual canonical basis B of C[N]. We
can therefore state:

Theorem 1.3 The tensor category CQ is a categorification of the ring C[N] and its dual canonical
basis B.

Note that, by work of Khovanov-Lauda [KL], Rouquier [Ro], and Varagnolo-Vasserot [VV2],
(C[N], B) has another categorification in terms of KLR-algebras. In type An , KLR-algebras are
isomorphic to blocks of affine Hecke algebras, and the category CQ for an equi-oriented quiver Q
is related to a category of representations of affine Hecke algebras through the quantum affine
Schur-Weyl duality. It would be interesting to find for other Dynkin quivers Q similar functors
between CQ and the module categories of the corresponding KLR-algebras.

1.7 The first author [He4] has shown that tensor products of simple objects of CZ have the
following remarkable property: a tensor product L1 ⊗ · · · ⊗ Lk of simple objects Li is simple if and
only if for every pair 1 ≤ i < j ≤ k the tensor product Li ⊗ L j is simple. Using Theorem 1.2 this
yields the following:

3
Corollary 1.4 A product b1 · · · bk of elements bi of the dual canonical basis B∗ of Uv (n) belongs
to B∗ up to a power of v if and only if for every pair 1 ≤ i < j ≤ k the product bi b j belongs to B∗
up to a power of v.
Corollary 1.4 was expected in relation with the program of Berenstein and Zelevinsky [BZ1, BZ2]
of describing B∗ in terms of quantum cluster algebras. But it was only known in a few low rank
cases.

1.8 Theorem 1.2 also gives new supporting evidence for some conjectures formulated in [GLS]
and [HL]. It was conjectured in [HL, §13] that for every ℓ ∈ N, the Grothendieck ring Rℓ of Cℓ has
a particular cluster algebra structure for which all cluster monomials are classes of simple objects.
In [GLS], it is shown that Uv (n) has a quantum cluster algebra structure, and it is conjectured
that all quantum cluster monomials belong to B∗ . Suppose that Q is bipartite and h is even. Set
h′ = h/2 − 1. By comparing initial seeds, one sees that the quantum cluster structure of Kt,Q
obtained by transporting via Φ−1 the quantum cluster structure of Uv (n) is a t-analogue of the
cluster structure of Rh′ conjectured in [HL]. Thus, by Theorem 1.2, the two conjectures of [GLS]
for Uv (n) and of [HL] for Rh′ are essentially equivalent.
In [HL] and [N4], the conjecture for Rℓ was proved in the first non trivial case ℓ = 1. (In [HL]
some combinatorial steps of the proof were only verified for g of type An and D4 ; the proof of
[N4] is general and uses geometric representation theory.) Since C1 is a tensor subcategory of CQ
(for every g except sl2 and sl3 ), Kt,Q contains a subring Kt,1 corresponding to C1 . It is easy to
see that Φ(Kt,1 ) is equal to the subalgebra Uv (n(w)) of [GLS] where w = c2 is the square of the
Coxeter element of the Weyl group of g corresponding to the bipartite quiver Q. This is a quantum
cluster algebra of finite cluster type, equal to the Dynkin type of g in the classification of Fomin
and Zelevinsky. Thus, using [HL, N4, Q], Theorem 1.2 readily implies:
Corollary 1.5 Let w = c2 be as above. Then B∗ ∩ Uv (n(w)) is equal to the set of quantum cluster
monomials of Uv (n(w)).
For g of type An , Lampe [La] has given a direct proof of the fact that the quantum cluster
variables of Uv (n(w)) belong to B∗ .

1.9 Since the bases B∗ and {[L]t } have geometric origin, it is natural to ask for a geometric
explanation of Theorem 1.2 (b). In the final part of the paper, we show (Theorem 9.11) that
the quiver representation spaces Ed used by Lusztig to define the canonical basis of Uv (n) are
isomorphic to some particular graded quiver varieties M•0 (W d ) used by Nakajima for describing
the classes [L]t of the simple objects L of CQ . Moreover the intersection cohomology sheaves of
closures of Gd -orbits in Ed can be identified with the intersection cohomology sheaves of closures
of strata in M•0 (W d ). This is inspired by a similar result of Nakajima [N4] for the category C1 .

1.10 We now give an overview of the structure of the paper. In Section 2, we set up our notation
and introduce an important bijection ϕ between the set of fundamental modules of CZ and the
vertices of the Auslander-Reiten quiver of Db (KQ). We use this bijection to express the entries
of the inverse of the quantum Cartan matrix of g in terms of the Ringel form of Q, or in terms
of the scalar product of the weight lattice of g (Proposition 2.5). By construction, the quantum
Grothendieck ring Kt is a subring of a quantum torus Yt over C(t 1/2 ). The t-commutation relations
between generators of Yt are expressed in terms of entries of the inverse of the quantum Cartan
matrix of g [He2], hence by Proposition 2.5, in terms of scalar products of weights of g. The
quantum Grothendieck ring Kt,Q is a subring of a subtorus Yt,Q of Yt , of rank r equal to the
number of positive roots of g. On the other hand, by [GLS], Uv (n) has an explicit embedding

4
into a quantum torus Tv,Q of rank r over C(v1/2 ), whose generators are certain unipotent quantum
flag minors. The explicit v-commutation relations between these minors involve scalar products of
roots and weights of g. Comparing these two presentations, we show that there is an isomorphism
Φ : Yt,Q → Tv,Q mapping t 1/2 to v1/2 (Proposition 3.8).
The proof that Φ restricts to an isomorphism from Kt,Q to Uv (n) is based on some explicit
systems of algebraic identities satisfied by the generators of both algebras. In Section 4, we recall
from [GLS] a system of quantum determinantal identities occuring in Uv (n), and in Section 5 we
derive a quantum T -system for the (q,t)-characters of the Kirillov-Reshetikhin modules. (In [N2,
§4], a quantum T -system was already obtained for the t-deformed product used in [N3, VV1].
A quantum cluster algebra related to the quantum T -system of type A1 is also studied in [DFK].)
Comparing these two systems we obtain that Φ maps the classes of the Kirillov-Reshetikhin mod-
ules of CQ to certain quantum minors of Uv (n) (multiplied by explicit half-integral powers of v).
In particular, Φ maps the classes of the fundamental modules of CQ in Kt,Q to the generators of
the dual PBW-basis of Uv (n) associated with Q (up to powers of v1/2 ). This proves the first part
of Theorem 1.2. The second part is deduced from the algebraic characterizations of B∗ and of the
classes [L]t (Section 6). After some examples, we give the proof of Corollary 1.4.
The above-mentioned presentation of Kt (Theorem 7.3) is deduced from Theorem 1.2 in Sec-
tion 7, and in Section 8 we prove the isomorphism with the derived Hall algebra DH(Q) stated
in Theorem 1.1. Finally, in Section 9, we explain our geometric approach to Theorem 1.2 (b)
(Theorem 9.11).

2 Cartan matrices and Auslander-Reiten quivers

2.1 Cartan matrix
Let g be a simple Lie algebra of type A, D, E. We denote by I the set of vertices of its Dynkin
diagram, and we put n = |I|. The Cartan matrix of g is the I × I matrix C with entries

 2 if i = j,
Ci j = −1 if i and j are adjacent vertices of the Dynkin diagram,

0 otherwise.

We shall often use the shorthand notation i ∼ j to say that Ci j = −1.

We denote by P the weight lattice of g, and by ϖi (i ∈ I) its basis of fundamental weights. The
simple roots are defined by
αi = ∑ Ci j ϖ j , (i ∈ I).
j∈I

The set of simple roots is denoted by Π := {αi | i ∈ I}. We denote by (·, ·) the scalar product of
P defined by (αi , ϖ j ) = δi j . Equivalently (αi , α j ) = Ci j . The Weyl group W is generated by the
reflexions si acting on P by

si (λ ) = λ − (λ , αi )αi , (λ ∈ P, i ∈ I).

The root system of g is ∆ := W Π. It decomposes as ∆ = ∆+ ⊔ ∆− , where ∆+ = ∆ ∩ (⊕i∈I Nαi ) and

∆− = −∆+ . We write r := |∆+ |.
A Coxeter element of W is a product of the form c = si1 · · · sin where (i1 , . . . , in ) is an arbitrary
ordering of I. All Coxeter elements are conjugate in W . Their common order is called the Coxeter
number and denoted by h. We have hn = 2r.

5
 .. .. ..
. . .
4 ❤❤4 (3, 3) f▼▼
✁✁ ❤❤❤❤❤❤qqq8 ▼▼▼
✁✁ ❤❤❤ qq
❤❤❤❤
♣3 (1, 2) j❱❱❱ (2, 2) (4, 2)
♣♣♣✁♣✁✁ ❱❱❱❱ f▼ q8
♣
x ♣♣ ✁ ❱❱❱❱ ▼▼▼▼ qq
1
♣
2 ❱❱❱❱ qq
4 (3, 1)
❤❤❤❤8 f▼▼▼
❤❤❤❤❤❤❤ qqqq ▼▼
❤❤❤ ❤ q
(1, 0) j❱❱❱ (2, 0) (4, 0)
❱❱❱❱ f▼ qq8
❱❱❱❱ ▼▼▼▼ qq
❱❱❱ q
4 (3, −1)
❤❤❤❤q8 f▼▼▼
❤ ❤❤❤❤❤❤ qqqq ▼▼
❤❤
(1, −2) j❱❱ (2, −2) (4, −2)
❱❱❱❱ f▼ q8
❱❱❱❱ ▼▼▼▼ qq
❱❱❱ qq
(3, −3)
.. .. ..
. . .

b
Figure 1: A quiver Q of type D4 and its repetition quiver Q.

2.2 Quivers
Let Q be an orientation of the Dynkin diagram of g. In other words, Q is a Dynkin quiver of the
same Dynkin type as g.
For i ∈ I, we denote by si (Q) the quiver obtained from Q by changing the orientation of every
arrow with source i or target i. Let w = si1 · · · sik ∈ W be a reduced decomposition. We say that
i = (i1 , . . . , ik ) is adapted to Q if i1 is a source of Q, i2 is a source of si1 (Q), . . . , ik is a source of
sik−1 · · · si1 (Q). There is a unique Coxeter element having reduced expressions adapted to Q. We
shall denote it by τ .
We denote by Q1 the set of arrows of Q. A height function ξ : I → Z on Q is a function
satisfying
ξ j = ξi − 1 if i → j ∈ Q1 .
Since Q is connected, two height functions differ by a constant. We fix such a function ξ . Define
Ib := {(i, p) ∈ I × Z | p − ξi ∈ 2Z}.
b defined as the oriented graph with vertex set Ib and
We attach to Q the infinite repetition quiver Q,
two types of arrows:
b for all (i, p) ∈ Ib;
(i) if there is an arrow i → j in Q we have arrows (i, p) → ( j, p + 1) in Q
b for all ( j, q) ∈ Ib.
(ii) if there is an arrow i → j in Q we have arrows ( j, q) → (i, q + 1) in Q
Note that Q b depends only on the Dynkin diagram, and not on the choice of orientation Q. In fact, it
is well known that Q b is the quiver of a Z-covering of the preprojective algebra associated with Q.
In the literature, this quiver is often denoted by ZQ. An example is shown in Figure 1, where the
height function is ξ1 = ξ2 = 0, ξ3 = 1, ξ4 = 2.
Let ∆b := ∆+ × Z. We now describe a natural labelling of the vertices of Q b by ∆.
b For i ∈ I, let
B(i) be the subset of I consisting of all j’s such that there is a path from j to i in Q. Define
γi := ∑ α j, (i ∈ I).
j∈B(i)

6
 .. .. ..
4 . . (α +α +α ,1)
.
❡❡❡❡2 4 1 2 3 j❯❯❯❯❯
✁✁ ❡❡ ❡❡❡❡❡❥❡❥❥❥❥❥❥ ❯❯❯❯
✁✁ ❡❡❡❡❡❡
❡ ❡❡ ❡
❥❥ ❯
♣3 (α1 ,1) l❨❨❨❨❨❨ (α2 ,1) j❚❚ ✐4 (α4 ,0)
♣♣♣♣✁♣✁✁ ❨❨❨❨❨❨ ❚❚❚❚
❨❨❨❨❨❨ ❚❚❚ ✐✐✐✐✐✐✐
x ♣♣ ✁
♣ ❨❨❨❨❨❚ ✐✐✐
1 2 ❡❡ 2 (α3 +α4 ,0) j❯
❡❡❡❡❡ ❥❥4
❡ ❯❯❯❯
❡❡❡❡❡❡ ❥❥❥❥❥❥❥ ❯❯❯❯
❡❡❡❡❡❡ ❯❯
(α1 +α3 +α4 ,0) l❨❨❨ (α2 +α3 +α4 ,0) ✐4 (α3 ,0)
❨❨❨❨❨❨ j❚❚❚❚ ✐✐✐ ✐
✐
❨❨❨❨❨❨ ❚❚❚ ✐✐
❨❨ ❚ ✐✐✐
(α +α +2α3 +α4 ,0)
❡❡2 1 4 2 j❯❯❯❯
❡❡❡ ❡❡ ❡❡❡❡❡❥❡❥❥❥❥❥❥ ❯❯❯❯
❯
❡❡❡❡❡ ❥
(α2 +α3 ,0) l❨❨❨❨ (α1 +α3 ,0) (α1 +α2 +α3 +α4 ,0)
❨❨❨❨❨❨ j❚ ✐4
❨❨❨❨❨❚❨❚❚❚❚❚❚ ✐✐✐✐
❨❨❨❨ ❚ ✐✐✐✐
2 (α1 +α2 +α3 ,0)j❯❯
❡❡ ❡❡ ❡ ❡❡❡❡❥❡❥❡❥❥4 ❯❯❯❯
❡❡❡❡❡❡❡❡❡❡ ❥ ❥ ❥❥❥ ❯❯❯❯
❡
(α1 ,0) l❨❨❨❨❨❨ (α2 ,0) j❚❚ ✐4 (α4 ,−1)
❨❨❨❨❨❨❨ ❚ ✐✐✐✐
❨❨❨❨❨❚❨❚❚❚❚❚❚ ✐✐✐
❨❨❨❨ ✐✐✐
(α3 +α4 ,−1)
.. .. ..
. . .

b by ∆
Figure 2: The labelling of Q b for Q of type D4 .

We have γi ∈ ∆+ . There is a unique bijection ϕ : Ib → ∆

b defined inductively as follows:

(a) ϕ (i, ξi ) = (γi , 0) for i ∈ I;

(b) suppose that ϕ (i, p) = (β , m); then

• ϕ (i, p − 2) = (τ (β ), m) if τ (β ) ∈ ∆+ ;
• ϕ (i, p − 2) = (−τ (β ), m − 1) if τ (β ) ∈ ∆− ;
• ϕ (i, p + 2) = (τ −1 (β ), m) if τ −1 (β ) ∈ ∆+ ;
• ϕ (i, p + 2) = (−τ −1 (β ), m + 1) if τ −1 (β ) ∈ ∆− .
b depends on Q. This is illustrated in Figure 2.
Note that this second labelling of Q

2.3 Auslander-Reiten theory

The quiver Q b with its labelling by ∆
b arises in the representation theory of the path algebra KQ of Q
over a field K, as we shall now recall. We refer the reader to [ARS, ASS, GR, Ri1] for background
on quiver representations and Auslander-Reiten theory.
Let mod(KQ) be the abelian category of representations of Q over K. For an object X of
mod(KQ) we write dim(X ) for its dimension vector. We define the Ringel bilinear form

hX ,Y i := dim(Hom(X ,Y )) − dim(Ext1 (X ,Y )), (X ,Y ∈ mod(KQ)),

and the symmetric form (X ,Y ) := hX ,Y i + hY, X i. It is known that these forms depend only on
the dimension vectors dim(X ) and dim(Y ). Moreover, if we identify in the standard way dim(X )
and dim(Y ) with elements of the root lattice of g, then (X ,Y ) coincides with the natural scalar

7
product (dim(X ), dim(Y )). In this picture, αi is the dimension vector of the simple KQ-module Si
supported on vertex i, and γi is the dimension vector of its injective envelope Ii . Recall that, by
Gabriel’s theorem, the isoclasses of indecomposable KQ-modules are in natural bijection with ∆+ .
They form the vertices of the Auslander-Reiten quiver ΓQ of mod(KQ). The map β 7→ (β , 0)
identifies ΓQ with the full subgraph of Q b with set of vertices ∆+ × {0}. The map τ restricted to
the dimension vectors in ∆+ of non projective KQ-modules is the Auslander-Reiten translation of
mod(KQ) [ARS].
Let Db (mod(KQ)) be the bounded derived category of KQ. Its indecomposable objects are
the stalk complexes X [i], consisting of an indecomposable object X of mod(KQ) sitting in degree
i ∈ Z, and zero objects in all other degrees. Thus, the isoclasses of indecomposable objects of
Db (mod(KQ)) are naturally labelled by ∆. b Using this labelling, the quiver Qb is identified with the
b
Auslander-Reiten quiver of the triangulated category D (mod(KQ)) [Ha].

2.4 Quantum Cartan matrix

Let z be an indeterminate, and let C(z) be the matrix with entries

 z + z−1 if i = j,
Ci j (z) = −1 if i ∼ j,

0 otherwise.

e
Thus C(1) is just the Cartan matrix C of g. Since det(C) 6= 0, det(C(z)) 6= 0. We denote by C(z)
e
the inverse of the matrix C(z). This is a matrix with entries Ci j (z) ∈ Q(z). Denoting by A the
adjacency matrix of the Dynkin diagram we have

C(z) = (z + z−1 )I − A,

therefore
e =
C(z) ∑ (z + z−1)−k−1Ak .
k≥0

e have power series expansions in z of the form

Hence the entries of C(z)

Cei j (z) = ∑ Cei j (m) zm ,

m≥1

where Cei j (m) ∈ Z. Note that since C(z) is a symmetric matrix, we have Cei j (m) = Ceji (m).

2.5 Formula for Cei j (m)

We will now give several equivalent expressions for the coefficients Cei j (m). For other expressions
of Cei j (z) in type An and Dn , see [FR1, Appendix C].
Fix an orientation Q of the Dynkin diagram, and recall from §2.2 and §2.3 the associated
notation ξi , γi , the Coxeter transformation τ , and the Ringel form h·, ·i.
Proposition 2.1 Let m ≥ 1. If m + ξi − ξ j − 1 is odd then Cei j (m) = 0. Otherwise
 
Cei j (m) = τ (m+ξi−ξ j −1)/2 (γi ), ϖ j . (1)

Equivalently, D E
Cei j (m) = τ (m+ξi −ξ j −1)/2 (Ii ), I j . (2)

8
Proof — Let us denote temporarily by Di j (m) the value of Cei j (m) predicted by the proposition.
We want to show that
∑ Cik (z)Dk j (m)zm = δi j , (i, j ∈ I).
k∈I, m≥1

Using the definition of Cik (z), this is equivalent to show that

!


∑ z + z Di j (m) − ∑ Dk j (m) zm = δi j ,
−1
(i, j ∈ I). (3)
m≥1 k∼i

The coefficient of z0 in the left-hand side is equal to Di j (1). If ξi − ξ j is odd then by definition
Di j (1) = 0. Otherwise, if ξi − ξ j = 2l, then Di j (1) = (τ l (γi ), ϖ j ) is the coefficient of α j in τ l (γi ).
Let (β , m) be the vertex of Q b in the column of (γi , 0) and at the same height as (γ j , 0). Such a
vertex exists because ξi − ξ j is even, and clearly β = ±τ l (γi ). Now it is a well-known fact from
the combinatorics of Auslander-Reiten quivers that for all vertices (γ , s) of Q b at the same height as
(γ j , 0) the coefficient of α j in γ is 0, except if (γ , s) = (γ j , 0) in which case it is equal to 1. Hence
we have Di j (1) = δi j .
Consider now the coefficient of zm (m ≥ 1) in (3). We need to show that
Di j (m + 1) + Di j (m − 1) − ∑ Dk j (m) = 0, (i, j ∈ I, m ≥ 1). (4)
k:k∼i

Note that for k ∼ i we have ξk = ξi ± 1, hence if m + ξi − ξ j is odd, all summands of the left-hand
side are zero. Otherwise, writing m + ξi − ξ j = 2l, the left-hand side of (4) is
!
τ l (γi ) + τ l−1 (γi ) − ∑ τ l+(ξk −ξi +1)/2 (γk ) , ϖ j .
k∼i

Now it is again a familiar fact from the combinatorics of Auslander-Reiten quivers that
τ l (γi ) + τ l−1 (γi ) = ∑ τ l+(ξk −ξi +1)/2 (γk ),
k∼i

since the roots τ l (γi ), τ l−1 (γi ), and τ l+(ξk −ξi +1)/2 (γk ) with k ∼ i, form a mesh. This proves (1).
Finally, if β = dimX then (β , ϖ j ) is equal to the coefficient of α j in β , hence
(β , ϖ j ) = dim(Hom(X , I j )) = hX , I j i,
because I j is injective. This proves (2). ✷

Example 2.2 Take g of type A4 . One has for instance

Ce11 (z) = z1 − z9 + z11 − z19 + · · ·
Ce12 (z) = z2 − z8 + z12 − z18 + · · ·
Ce13 (z) = z3 − z7 + z13 − z17 + · · ·
Ce14 (z) = z4 − z6 + z14 − z16 + · · ·
Ce21 (z) = z2 − z8 + z12 − z18 + · · ·
Ce22 (z) = z1 + z3 − z7 − z9 + z11 + z13 − z17 − z19 + · · ·
Ce23 (z) = z2 + z4 − z6 − z8 + z12 + z14 − z16 − z18 · · ·
Ce24 (z) = z3 − z7 + z13 − z17 + · · ·

9
Let us choose the sink-source orientation Q with height function ξ1 = 0, ξ2 = 1, ξ3 = 0, ξ4 = 1.
Then τ = s2 s4 s1 s3 , and since τ 5 = 1, the roots τ l (γi ) are all determined by:

γ1 = α1 + α2 , γ2 = α2 , γ3 = α2 + α3 + α4 , γ4 = α4 ,
τ (γ1 ) = α3 + α4 , τ (γ2 ) = α1 + α2 + α3 + α4 , τ (γ3 ) = α1 + α2 + α3 , τ (γ4 ) = α2 + α3 ,
τ 2 (γ1 ) = −α4 , τ 2 (γ2 ) = α3 , τ 2 (γ3 ) = −α2 , τ 2 (γ4 ) = α1 ,
τ 3 (γ1 ) = −α2 − α3 , τ 3 (γ2 ) = −α2 − α3 − α4 , τ 3 (γ3 ) = −α1 − α2 − α3 − α4 , τ 3 (γ4 ) = −α1 − α2 ,
τ 4 (γ1 ) = −α1 , τ 4 (γ2 ) = −α1 − α2 − α3 , τ 4 (γ3 ) = −α3 , τ 4 (γ4 ) = −α3 − α4 .

For instance by Proposition 2.1, Ce23 (6) is equal to the coefficient of α3 in τ 3 (γ2 ) = −α2 − α3 − α4 ,
namely to −1.

Corollary 2.3 For i, j ∈ I and m ≥ 1 we have

Cei j (m + 2h) = Cei j (m).

Proof — Since τ h = 1, this follows immediately from Proposition 2.1. ✷

3 Quantum tori
3.1 The quantum torus Yt
b Define
Recall from §2.2 the labelling set Ib of Q.
h i
±1
Y := C Yi,p | (i, p) ∈ Ib

to be the Laurent polynomial ring generated by a collection of commutative variables Yi,p labelled
by Ib. This ring is related to a tensor subcategory CZ of the category of finite-dimensional Uq (Lg)-
modules considered in [HL] (see below §5.2).
Let t be an indeterminate. Following [He2] we introduce a t-deformed version (Yt , ∗) of Y ,
with noncommutative multiplication denoted by ∗. This is the C(t)-algebra generated by variables
still denoted by Yi,p , subject to the t-commutation relations

Yi,p ∗Y j,s := t N (i,p; j,s)Y j,s ∗Yi,p , ((i, p), ( j, s) ∈ Ib), (5)

where

N (i, p; j, s) := Cei j (p − s − 1) − Cei j (p − s + 1) − Cei j (s − p − 1) + Cei j (s − p + 1). (6)

Here we have extended the definition of Cei j (m) to every m ∈ Z by setting Cei j (m) = 0 if m ≤ 0.
e is symmetric, we have
Note that, since C(z)

N (i, p; j, s) = −N ( j, s; i, p), (i, j ∈ I, p, s ∈ Z). (7)

If p = s then N (i, p; j, s) = 0. Otherwise, without loss of generality we can assume that p < s.
Then, (6) simplifies as

N (i, p; j, s) = Cei j (s − p + 1) − Cei j (s − p − 1), (p < s). (8)

We regard the noncommutative ring (Yt , ∗) as a quantum torus of infinite rank.

10
Remark 3.1 In [VV1] and [N3], the construction of a t-deformed Grothendieck ring is based on
a slightly different quantum torus. Namely, in these papers the product is defined by:
′ (i,p; j,s)
Yi,p · Y j,s := t N Y j,s · Yi,p , ((i, p), ( j, s) ∈ Ib), (9)

where instead of (6) the following exponent is used:

 
N ′ (i, p; j, s) := −2 Cei j (p − s − 1) − Cei j (s − p − 1) . (10)

For instance, in type A3 , we have

Y1,0 ∗Y2,1 = tY2,1 ∗Y1,0 , whereas Y1,0 · Y2,1 = Y2,1 · Y1,0 .

In [N3, VV1], the definition of the product comes from a convolution operation for certain per-
verse sheaves on quiver varieties, and the deformation parameter t encodes the natural grading
of complexes of sheaves. Our product ∗ comes from [He2] and the original construction of q-
characters. Indeed in [FR2], the variables Yi,p ∈ Y are defined as formal power series in elements
of Uq (Lg), and they pairwise commute. In [He2], these formal power series are replaced by cer-
tain infinite sums Ỹi,p in elements of the quantum affine algebra Uq (b g) (with non trivial central
charge c), which can be seen as vertex operators. The original variable Yi,p is just one factor of the
complete variable Ỹi,p . The relations of the quantum affine algebra then give rise to t-commutation
relations between the Ỹi,p , where the parameter t appears as a formal power series with coefficients
in C[c±1 ] [He2, Theorem 3.11]. The defining relations (5) (6) of ∗ are obtained by replacing t by
t −1 in those t-commutation relations.

Recall from §2.2 the bijection ϕ : Ib → ∆.

b

Proposition 3.2 Let (i, p) and ( j, s) be elements of Ib with p < s. There holds
 
N (i, p; j, s) = τ (s−p+ξi−ξ j )/2 (γi ), γ j .

Moreover, if ϕ (i, p) = (β , m) and ϕ ( j, s) = (δ , l), then

N (i, p; j, s) = (−1)l−m (β , δ ).

Proof — First note that the definition of Ib implies that s − p + ξi − ξ j ∈ 2Z. By Proposition 2.1,
we have

N (i, p; j, s) = Cei j (s − p + 1) − Cei j (s − p − 1)

D E D E
= τ (s−p+ξi−ξ j )/2 (Ii ), I j − τ (s−p+ξi−ξ j )/2−1 (Ii ), I j .

Now recall the classical formula

hτ −1 (X ),Y i = −hY, X i, (X ,Y ∈ mod(FQ)).

It follows that
D E D E
N (i, p; j, s) = τ (s−p+ξi −ξ j )/2 (Ii ), I j + I j , τ (s−p+ξi−ξ j )/2 (Ii )
 
= τ (s−p+ξi−ξ j )/2 (Ii ), I j
 
= τ (s−p+ξi−ξ j )/2 (γi ), γ j .

11
This proves the first equality. The second equality is immediately deduced from the first one if we
note that, by definition of the bijection ϕ ,
τ (ξi −p)/2 (γi ) = (−1)m β , τ (ξ j −s)/2 (γi ) = (−1)l δ .
✷

Remark 3.3 For the product of [VV1] and [N3], we have, for p < s, a similar expression
D E D E
N ′ (i, p; j, s) = 2 τ (s−p+ξi−ξ j )/2−1 (γi ), γ j = −2 γ j , τ (s−p+ξi −ξ j )/2 (γi ) = (−1)l−m+1 2hδ , β i,

in which the symmetric scalar product (·, ·) is replaced by the non-symmetric Ringel form h·, ·i.

3.2 Commutative monomials

Let us adjoin a square root t 1/2 of t and extend the quantum torus (Yt , ∗) to
(Yt , ∗) := C(t 1/2 ) ⊗C(t) (Yt , ∗).
We notice that the expression
1 1
t 2 N ( j,s;i,p) Yi,p ∗Y j,s = t 2 N (i,p; j,s) Y j,s ∗Yi,p ∈ Yt
is invariant under permutation of (i, p) and ( j, s). We can then denote it as a commutative mono-
mial Yi,pY j,s = Y j,sYi,p , and write
1
Yi,p ∗Y j,s = t 2 N (i,p; j,s) Yi,pY j,s .
More generally, for a family (ui,p | (i, p) ∈ Ib) of integers with finitely many nonzero components,
the expression
−→
t 2 ∑(i,p)<( j,s) ui,p u j,s N ( j,s;i,p) (i,p)∈
1 u
* Ib Yi,pi,p
does not depend on the chosen ordering of Ib used to define it. We will denote it as a commutative
u
monomial ∏(i,p)∈Ib Yi,pi,p , and write
−→
Yi,pi,p = t 2 ∑(i,p)<( j,s) ui,p u j,s N (i,p; j,s) ∏
u 1 u
* Yi,pi,p .
(i,p)∈Ib (i,p)∈Ib

The commutative monomials form a basis of the C(t 1/2 )-vector space Yt . It will be convenient to
denote commutative monomials by

∏
u (m)
m= Yi,pi,p .
(i,p)∈Ib

A commutative monomial m is said to be dominant if ui,p (m) ≥ 0 for every (i, p) ∈ Ib.
The noncommutative product of two commutative monomials m1 and m2 is given by
1
m1 ∗ m2 = t 2 D(m1 ,m2 ) m1 m2 = t D(m1 ,m2 ) m2 ∗ m1 , (11)
where
D(m1 , m2 ) = ∑ ui,p (m1 )u j,s (m2 )N (i, p; j, s),
(i,p),( j,s)∈Ib
and
∏
u (m1 )+ui,p (m2 )
m1 m2 = Yi,pi,p ,
(i,p)∈Ib
denotes the commutative product.

12
3.3 The quantum torus Yt, Q
Recall the bijection ϕ : Ib → ∆
b of §2.2. Define

IbQ := ϕ −1 (∆+ × {0}) ⊂ Ib,

and let Yt,Q be the the C(t)-subalgebra of (Yt , ∗) generated by the variables Yi,p ((i, p) ∈ IbQ ). This
is a quantum torus of rank r = |∆+ |. We will also use the extended torus

(Yt,Q , ∗) := C(t 1/2 ) ⊗C(t) (Yt,Q , ∗).

Example 3.4 We take g of type D4 and choose Q as in Figure 1 and Figure 2. Comparing the two
figures we see that Yt,Q is generated by
±1 ±1 ±1 ±1 ±1 ±1 ±1 ±1 ±1 ±1 ±1 ±1
Y1,0 , Y1,−2 , Y1,−4 , Y2,0 , Y2,−2 , Y2,−4 , Y3,1 , Y3,−1 , Y3,−3 , Y4,2 , Y4,0 , Y4,−2 .

3.4 The quantum torus Tv, Q

Let w0 be the longest element of W . Let i = (i1 , . . . , ir ) be a reduced expression of w0 adapted to Q
(see §2.2). Following [GLS, §11], we introduce a quantum torus Tv,Q of rank r over C(v). (The
indeterminate v is denoted by q in [GLS]). Its generators are certain unipotent quantum minors

Dϖik , λk , (1 ≤ k ≤ r)

in the quantum coordinate ring Av (n). Here λk is the weight given by

λk = si1 · · · sik (ϖik ), (1 ≤ k ≤ r).

The definition of Av (n) will be recalled in §4.1 below. At this stage we only need to know the
explicit v-commutation relations satisfied by these minors. It is shown in [GLS, Lemma 11.2] that
for k < l there holds
Dϖik , λk Dϖil , λl = v(ϖik −λk , ϖil +λl ) Dϖil , λl Dϖik , λk . (12)
For 1 ≤ k ≤ r, set k− := max({s < k | is = ik } ∪ {0}). Define
 −1
Zk := Dϖik , λk Dϖi − , λk− , (13)
k

where if k− = 0 we understand Dϖi − , λk− = 1. Clearly, Zk (1 ≤ k ≤ r) is another set of generators

k
of Tv,Q . Let
βk = si1 · · · sik−1 (αik ), (1 ≤ k ≤ r). (14)
Note that we have
λk = λk − − βk , (1 ≤ k ≤ r), (15)
where if k− = 0 we use the convention λk− = ϖik .

Proposition 3.5 For 1 ≤ k < l ≤ r, we have:

Zk Zl = v−(βk ,βl ) Zl Zk . (16)

13
Proof — Let us introduce the integers µkl and νkl such that

Dϖik ,λk Dϖil ,λl = vµkl Dϖil ,λl Dϖik ,λk , Zk Zl = vνkl Zl Zk , (1 ≤ k, l ≤ r).

By definition of Zk we have

νkl = (µkl − µk− l ) − (µkl− − µk− l− ),

where we use the convention that µk− l = 0 if k− = 0, µkl− = 0 if l − = 0, and µk− l− = 0 if k− = 0

or l − = 0. Since k− < k < l, we have

µkl − µk− l = (ϖik − λk , ϖil + λl ) − (ϖik − λk− , ϖil + λl ) = (βk , ϖil + λl ). (17)

(a) If k < l − we have similarly

µkl− − µk− l− = ((βk , ϖil + λl− )

and so
νkl = (βk , ϖil + λl ) − (βk , ϖil + λl− ) = −(βk , βl ),
as required.
(b) If k = l − then µkl− = 0 and µk− l = (ϖik − λk− , ϖik + λk ). Hence

νkl = (βk , ϖil + λl ) + (ϖik − λk− , ϖik + λk )

= (βk , ϖik + λk ) − (βk , βl ) + (ϖik − λk− , ϖik + λk )
= −(βk , βl ) + (ϖik − λk− + βk , ϖik + λk )
= −(βk , βl ) + (ϖik − λk , ϖik + λk )
= −(βk , βl )

as required, because (ϖik − λk , ϖik + λk ) = (ϖik , ϖik ) − (λk , λk ) = 0.

(c) If k > l − then µkl− = −µl− k = −(ϖil − λl− , ϖik + λk ). Hence

−µkl− + µk− l− = (ϖil − λl− , ϖik + λk ) + (ϖik − λk− , ϖil + λl− )

= −(βk , ϖil + λl− ) + (ϖil − λl− , ϖik + λk ) + (ϖik − λk , ϖil + λl− )
= −(βk , ϖil + λl− ) + 2(ϖik , ϖil ) − 2(λk , λl− )
= −(βk , ϖil + λl− ).

Indeed, since l − < k < l, we have

(λk , λl− ) = (sil − +1 · · · sik (ϖik ), ϖil ) = (ϖik , sik · · · sil − +1 (ϖil )) = (ϖik , ϖil ).

It follows that again, νkl = (βk , ϖil + λl ) − (βk , ϖil + λl− ) = −(βk , βl ), as required. ✷

3.5 An isomorphism
It is well known that the roots βk (1 ≤ k ≤ r) give an enumeration of ∆+ . Therefore, for every
(i, p) ∈ IbQ there is a unique k such that ϕ (i, p) = (βk , 0).

14
Proposition 3.6 The assignment
t 7→ v,

Yi,p 7→ Zk , where (i, p) ∈ IbQ , and ϕ (i, p) = (βk , 0),

extends to an isomorphism of quantum tori from Yt,Q to Tv,Q .

Proof — This follows immediately from Proposition 3.2 and Proposition 3.5 if we note that when
ϕ (i, p) = (βk , 0) and ϕ ( j, s) = (βl , 0), p < s implies that k > l. ✷

3.6 The involution σ and the rescaled generators Xk

Let Tv,Q := C(v1/2 ) ⊗C(v) Tv,Q . For γ = ∑i ci αi in the root lattice of g, we set
(γ , γ )
deg γ := ∑ ci , N(γ ) := − deg γ . (18)
i 2
Following [GLS], we introduce an involution σ of Tv,Q , defined as the C-algebra anti-automor-
phism satisfying  
σ (v1/2 ) = v−1/2 , σ Dϖik ,λk = vN(ϖik −λk ) Dϖik ,λk . (19)
We rescale the generators Zk of Tv,Q by defining
(
vN(βk )/2 Zk if 1 ≤ k ≤ n,
Xk := (20)
N(βk )/2+(ϖik −λk−n , βk )
v Zk if n + 1 ≤ k ≤ r.
Note that these elements live in Tv,Q .
Lemma 3.7 For 1 ≤ k ≤ r we have:
σ (Xk ) = Xk .

Proof — For convenience, we set λk−n = ϖik if k − n ≤ 0. Using (13), (12), and the definition
of σ , we have
σ (Zk ) = vN(ϖik −λk )−N(ϖik −λk−n )−(ϖik −λk−n , ϖik +λk ) Zk .
A simple calculation using (15) shows that
N(ϖik − λk ) − N(ϖik − λk−n ) − (ϖik − λk−n , ϖik + λk ) = N(βk ) + 2(ϖik − λk−n , βk ),
and the lemma follows. ✷

Clearly, the rescaled generators Xk satisfy the same commutation relations as the Zk . Therefore,
if we define for a := (a1 , . . . , ar ) ∈ Zr ,

X a := v 2 ∑i< j ai a j (βi , β j ) X1a1 · · · Xrar ,

1
(21)
we have by Proposition 3.5,
σ (X a ) = v− 2 ∑i< j ai a j (βi , β j ) Xrar · · · X1a1 = X a .
1
(22)
Thus, X a is σ -invariant, and more generally an element of Tv,Q is σ -invariant if and only if all the
coefficients of its expansion with respect to the basis {X a | a ∈ Zr } are invariant under the map
v1/2 7→ v−1/2 . Moreover, one checks easily that
X a X b = v 2 ∑i< j (a j bi −ai b j )(βi , β j ) X a+b = v∑i< j (a j bi −ai b j )(βi , β j ) X b X a .
1
(23)

15
3.7 The isomorphism Φ
We can now state the main result of this section, which follows immediately from Proposition 3.6
and Equations (11), (23).

Proposition 3.8 There is a C-algebra isomorphism Φ : Yt,Q → Tv,Q given by

Φ(t 1/2 ) = v1/2 , Φ(Yi,p ) = Xk for (i, p) ∈ IbQ and ϕ (i, p) = (βk , 0).

More generally, let

∏
u (m)
m= Yi,pi,p
(i,p)∈IbQ

be a commutative monomial in Yt,Q , as in §3.2, and let a = (a1 , . . . , ar ) where ak = u(i,p) (m) if
ϕ (i, p) = (βk , 0). Then we have
Φ(m) = X a .
✷

4 Quantum groups
4.1 Background
Let n denote a maximal nilpotent subalgebra of g. Let Uv (n) be the Drinfeld-Jimbo quantum
enveloping algebra of n over C(v), with Chevalley generators ei (i ∈ I) subject to the quantum
Serre relations:
ei e j − e j ei = 0 if Ci j = 0,
e2i e j − (v + v−1 )ei e j ei + e j e2i = 0 if Ci j = −1.

It is endowed with a natural scalar product (·, ·) which we normalize by (ei , ei ) = 1 (see e.g. [GLS,
§4.3]). We denote by Av (n) the graded dual vector space of Uv (n). The map x 7→ (x, ·) is a vector
space isomorphism from Uv (n) to Av (n), which allows to define a multiplication on Av (n) by
transporting the multiplication of Uv (n).
Thus, Uv (n) and Av (n) are isomorphic algebras, but they have dual integral forms and therefore
they specialize differently at v = 1. One should regard Av (n) as a quantum coordinate ring of the
unipotent group N with Lie algebra n. For example, the elements Dϖik , λk of §3.4 are quantum
analogues of certain generalized minors on N. We set

Uv (n) := C(v1/2 ) ⊗C(v) Uv (n), Av (n) := C(v1/2 ) ⊗C(v) Av (n).

Since the basis involved in Theorem 1.2 (b) is the dual canonical basis B∗ , it is more natural to
think of the quantum algebra of Theorem 1.2 (a) as being Av (n) rather than Uv (n).
The algebra Uv (n) has a natural grading by the root lattice of g, given by deg(ei ) = αi . The
above isomorphism allows to transfer this grading to Av (n).

16
4.2 Determinantal identities
In [GLS], it is shown that Av (n) has a quantum cluster algebra structure. In particular, an explicit
realization of Av (n) as a subalgebra of the quantum torus Tv,Q is given. This goes as follows.
For u, w ∈ W and λ ∈ P+ , one has unipotent quantum minors Du(λ ),w(λ ) ∈ Av (n) (see [GLS,
§5.2]). They satisfy (
1 if u(λ ) = w(λ ),
Du(λ ),w(λ ) =
0 if u(λ ) 6≤ w(λ ).
Let i = (i1 , . . . , ir ) be as in §3.4. In [GLS, §5.4], a system of identities relating the quantum minors
D(b, d) := Dsi1 ···si
b
(ϖib ), si1 ···sid (ϖib ) , (0 ≤ b ≤ d ≤ r, ib = id ∈ I), (24)

is derived, which we now recall. By convention, we write D(0, b) = Dϖi , si1 ···si (ϖi ) . Note that the
b b b
minors D(0, b) (1 ≤ b ≤ r) form by definition a system of generators of Tv,Q . We will also use the
following shorthand notation:
b− ( j) := max ({s < b | is = j} ∪ {0}) , (25)
−
b := max ({s < b | is = ib } ∪ {0}) , (26)
µ (b, j) := si1 · · · sib (ϖ j ). (27)
In (27) we understand that µ (0, j) = ϖ j .
Proposition 4.1 ([GLS]) Let 1 ≤ b < d ≤ r be such that ib = id = i. There holds
−→
vA D(b, d)D(b− , d − ) = v−1+B D(b, d − )D(b− , d) + vC ∏ D(b− ( j), d − ( j)) (28)
j∼i

where
A = (µ (d, i), µ (b− , i) − µ (d − , i)), B = (µ (d − , i), µ (b− , i) − µ (d, i)),
and
C= ∑ (µ (d, j), µ (b, k) − µ (d, k)) .
j<k
j∼i, k∼i

This system of identities allows to express inductively every minor D(b, d) as a rational func-
tion of the flag minors D(0, c). Moreover, it follows from [GLS, Theorem 12.3] that all these
rational functions belong in fact to Tv,Q , and that Av (n) is the subalgebra of Tv,Q generated by the
minors D(b− , b) (1 ≤ b ≤ r).

4.3 The dual canonical basis B∗

Let us write
E ∗ (βk ) := D(k− , k), (1 ≤ k ≤ r), (29)
and for a = (a1 , . . . , ar ) ∈ Zr ,
E ∗ (a) = v− ∑k=1 ak (ak −1)/2 E ∗ (β1 )a1 · · · E ∗ (βr )ar .
r
(30)
Then E∗ = {E ∗ (a) | a ∈ Zr } is a C(v)-basis of Av (n), dual to a basis of Uv (n) of PBW-type, as
defined by Lusztig. The basis E∗ is called the dual PBW-basis of Av (n).
The involution σ of Tv,Q (see §3.6) can be restricted to Av (n). Lusztig [Lu1] has constructed
a canonical basis B of Uv (n). The dual basis B∗ = {B∗ (a) | a ∈ Nr } of Av (n) can be characterized
as follows (see e.g. [GLS]).

17
Proposition 4.2 For a = (a1 , . . . , ar ) ∈ Nr , the vector B∗ (a) is uniquely determined by the follow-
ing conditions:

(a) B∗ (a) ∈ E ∗ (a) + ∑c6=a v−1 Z[v−1 ]E ∗ (c);

(b) let β (a) := ∑1≤k≤r ak βk . Then σ (B∗ (a)) = vN(β (a)) B∗ (a).

The integer N(γ ) of (b) is defined in (18). Note that β (a) is just the weight of B∗ (a) or E ∗ (a) in
the natural grading of Av (n) by the root lattice of g. The basis B∗ is called the dual canonical basis
of Av (n).

5 Quantum Grothendieck rings

5.1 Background
For recent surveys on the representation theory of quantum loop algebras, we invite the reader to
consult [CH] or [Le].
Let Lg be the loop algebra attached to g, and let Uq (Lg) be the associated quantum enveloping
algebra. We assume that the deformation parameter q ∈ C∗ is not a root of unity.
By [FR2], every finite-dimensional Uq (Lg)-module M (of type 1) has a q-character χq (M).
These q-characters generate a commutative C-algebra isomorphic to the complexified Grothen-
dieck ring of the category of finite-dimensional irreducible Uq (Lg)-modules. Nakajima [N3],
Varagnolo and Vasserot [VV1], and Hernandez [He2], have studied t-deformations of the q-
characters of the standard modules and of the simple modules, as well as corresponding t-deforma-
tions of the Grothendieck ring. Although slightly different, these t-deformed Grothendieck rings
are essentially equivalent, and in particular they give rise to the same (q,t)-characters for the sim-
ple modules. In what follows, we will use the version of [He2]. Its definition will be recalled in
the next sections.

5.2 The subcategory CZ

The simple finite-dimensional irreducible Uq (Lg)-modules (of type 1) are usually labelled by Drin-
feld polynomials. Here we shall use an alternative labelling by dominant monomials (see [FR2]).
Moreover, as in [HL], we shall restrict our attention to a certain tensor subcategory CZ of the
category of finite-dimensional Uq (Lg)-modules. The simple modules in CZ are labelled by the
dominant monomials in Y , or equivalently, by the dominant commutative monomials in (Yt 1/2 , ∗)
(see §3.2), and their q-characters belong to Y . We shall denote by L(m) the simple module la-
belled by the dominant monomial m. When m = Yi,p is reduced to a single variable, L(m) is called
a fundamental module. When m is the only dominant monomial occuring in χq (L(m)), L(m) is
said to be minuscule. Fundamental modules are examples of minuscule modules [FM].

5.3 Standard modules

To a dominant commutative monomial m is also attached a tensor product of fundamental modules
called a standard module M(m) defined by
−→
O
M(m) := L(Yi,p )⊗ ui,p (m) , (31)
(i,p)∈Ib

18
where the product is ordered according to the following partial order on Ib:

(i, p) < ( j, s) ⇐⇒ p < s.

Note that for any fixed p ∈ Z and any total order on I, the tensor product
−→
O
L(Yi,p )⊗ ui,p (m)
i∈I
 
is irreducible, and its isomorphism class L ∏i∈I Yi,p
k does not depend on the order of the factors,
hence (31) is well defined up to isomorphism (see [FM, Proposition 6.15]). The classes [M(m)]
of the standard modules M(m) form a second basis of the Grothendieck group of CZ .

5.4 The ring Kt

We introduce the commutative monomials [FR2]

Ai,p = Yi,p+1Yi,p−1 ∏ Y j,p

−1
, ((i, p − 1) ∈ Ib). (32)
j∼i

Recall from §3.2 that commutative monomials in Y can be regarded as elements of (Yt , ∗). More
generally, the commutative polynomials
 
−1
Yi,p 1 + Ai,−1p+1 = Yi,p + Yi,p+2 ∏ Y j,p+1 , ((i, p) ∈ Ib)
j∼i

can be regarded as elements of (Yt , ∗). For i ∈ I, let Ki,t be the C(t 1/2 )-subalgebra of Yt (for the
noncommutative product ∗) generated by
   
Yi,p 1 + A−1 , Y ±1
, (i, p), ( j, s) ∈ b, j 6= i .
I
i, p+1 j,s

(In [He1], Ki,t is identified with the kernel of a t-deformed screening operator.) Define
\
Kt := Ki,t .
i∈I

It is shown in [He2] that an element of Kt is uniquely determined by the coefficients of its domi-
nant monomials. Moreover, for any dominant monomial m, there is a unique F(m) ∈ Kt such that
m occurs in F(m) with multiplicity 1 and no other dominant monomial occurs in F(m). These
F(m) form a C(t 1/2 )-basis of Kt .

5.5 Comparison with other t-deformations

The product ∗ used in this paper is the same as that of [He2], except that we have replaced t
by t −1 . The product of [He2] is slightly different from the products of [N3] and [VV1] (see
Remark 3.1). However, as shown in [He2, Proposition 3.16], for every (i, p), ( j, s) ∈ Ib the pairs
(Yi,p , A j,s ) and (Ai,p , A j,s ) are t-commutative with the same exponents of t for the three products of
[N3, VV1, He2]. This implies that the t-deformations of the Grothendieck ring R of CZ associated
with the three products are essentially equivalent, as will be explained below.

19
5.6 (q,t)-characters of standard modules
For a dominant commutative monomial m ∈ Yt 1/2 , define
!
←−
[M(m)]t := t α (m) ∏
u (m)
* F Yi,pi,p ∈ Kt . (33)
p∈Z i∈I

Here, α (m) ∈ 21 Z is chosen so that m occurs with coefficient 1 in the expansion of [M(m)]t on
the basis of commutative monomials of Yt 1/2 . The coefficients of [M(m)]t on this basis belong to
Z[t ±1 ] and may therefore be specialized at t = 1. The obtained specialization of [M(m)]t at t = 1
is equal to χq (M(m)), the q-character of the standard module M(m). Therefore we may use the
alternative notation
χq,t (M(m)) := [M(m)]t ,
and call this element of Kt the (q,t)-character of M(m).

5.7 The bar involution

One shows that there is a unique C-algebra anti-automorphism of (Yt , ∗) such that

t 1/2 = t −1/2 , Yi,p = Yi,p , ((i, p) ∈ Ib).

Clearly, the commutative monomials are bar-invariant, as in [N3, VV1]. The subring Kt is stable
under the bar involution, since each Ki,t is obviously stable. It follows that the elements F(m) are
bar-invariant (since m is the unique dominant monomial of F(m)). Hence the coefficients of the
expansion of F(m) on the basis of commutative monomials are unchanged under the replacement
of t by t −1 . Therefore, F(m) is the same as in [He2]. Since we have used in (33) the reverse
←
product ∗ , the elements χq,t (M(m)) also coincide with the corresponding elements of [N3, VV1,
He2], i.e. the coefficients of their expansion on the basis of commutative monomials are the same.

5.8 (q,t)-characters of simple modules

Proposition 5.1 ([N3]) For every dominant monomial m, there is a unique element [L(m)]t of Kt
satisfying
(a) [L(m)]t = [L(m)]t ,

(b) [L(m)]t ∈ [M(m)]t + ∑ t −1 Z[t −1 ] [M(m′ )]t .

m′ <m

Here m′ ≤ m means that m(m′ )−1 is a product of elements Ai,p in Y .

By §5.7, the elements [L(m)]t coincide with the corresponding elements of [N3, VV1, He2]. Using
the geometry of quiver varieties, Nakajima has shown:
Theorem 5.2 ([N3]) The specialization of [L(m)]t at t = 1 is equal to χq (L(m)), and the coeffi-
cients of the expansion of [L(m)]t as a linear combination of monomials in the Yi,p ’s belong to
N[t ±1 ].
Therefore we may use the alternative notation

χq,t (L(m)) := [L(m)]t ,

and call this element of Kt the (q,t)-character of L(m).

20
Corollary 5.3 (a) If L(m) is minuscule, χq,t (L(m)) = F(m).

(b) If χq (L(m)) is multiplicity-free, then χq,t (L(m)) = χq (L(m)) does not depend on t when
expressed on the basis of commutative monomials.

Proof — By the positivity statement of Theorem 5.2, every monomial occuring in χq,t (L(m))
already occurs in χq (L(m)). Thus, if L(m) is minuscule then χq,t (L(m)) is an element of Kt
containing the unique dominant monomial m, which proves (a). If χq (L(m)) is multiplicity-free,
then the coefficient of every commutative monomial in χq,t (L(m)) is of the form t k for some k ∈ Z.
But since χq,t (L(m)) is bar-invariant, we must have k = 0, which proves (b). ✷

5.9 Multiplicative structure

We shall regard the noncommutative ring (Kt , ∗) as a t-deformed version of the Grothendieck ring
R. But one should be aware that only the simple modules L(m) and the standard modules M(m)
have well-defined “classes” χq,t (L(m)) and χq,t (M(m)) in Kt .
For any dominant monomials m1 and m2 , write

χq,t (L(m1 )) ∗ χq,t (L(m2 )) = ∑ cm

m1 ,m2 (t
1/2
) χq,t (L(m)).
m

Note that every irreducible (q,t)-character is of the form χq,t (L(m)) = m(1 + ∑k Mk ), where the
−1
Mk are monomials in the Ai,p with coefficients in N[t,t −1 ] (see [He2]). So, by §5.5, the above
coefficients cm
m1 ,m2 (t
1/2 ) are obtained from the corresponding ones in [N3, VV1] by multiplying

by some t k with k ∈ Z/2. Varagnolo and Vasserot have shown the following positivity result:

Theorem 5.4 ([VV1]) The structure constants cm 1/2 ) belong to N[t 1/2 ,t −1/2 ].
m1 ,m2 (t

Corollary 5.5 L(m1 ) ⊗ L(m2 ) ≃ L(m) is a simple module if and only if

χq,t (L(m1 )) ∗ χq,t (L(m2 )) = t 2k χq,t (L(m2 )) ∗ χq,t (L(m1 )) = t k χq,t (L(m))

for some k ∈ Z/2.

Proof — If L(m1 ) ⊗ L(m2 ) ≃ L(m) then χq (L(m1 )) ∗ χq (L(m2 )) = χq (L(m)). Hence cm m1 ,m2 (1) =
1, and it follows from Theorem 5.4 that χq,t (L(m 1 )) ∗ χq,t (L(m 2 )) = t k χq,t (L(m)) for some k ∈
Z/2. Applying the bar involution, we get χq,t (L(m2 )) ∗ χq,t (L(m1 )) = t χq,t (L(m)). If conversely
−k

χq,t (L(m1 )) ∗ χq,t (L(m2 )) = t k χq,t (L(m)), then specializing t to 1 we get

χq (L(m1 ))χq (L(m2 )) = χq (L(m1 ) ⊗ L(m2 )) = χq (L(m)),

hence L(m1 ) ⊗ L(m2 ) ≃ L(m). ✷

21
5.10 Quantum T -system
(i)
For (i, p) ∈ Ib and k ∈ N, let mk,p := Yi,pYi,p+2 · · ·Yi,p+2k−2 . The simple Uq (Lg)-module
 
(i) (i)
Wk,p := L mk,p

(i)
is called a Kirillov-Reshetikhin module. (By convention, if k = 0 then Wk,p is the trivial one-
dimensional module.) The q-characters of the Kirillov-Reshetikhin modules satisfy the following
system of algebraic identities called T -system [KNS, N2, He3]. For every (i, p) ∈ Ib and k > 0,
there holds
         −Ci j
χq Wk,p χq Wk,p+2 = χq Wk−1,p+2 χq Wk+1,p + ∏ χq Wk,p+1
(i) (i) (i) (i) ( j)
.
j6=i

This can be lifted to a t-deformed T -system in Kt , as shown by the next proposition (see also
[N2, §4], where a different t-deformed product is used, as explained in Remark 3.1 and §5.5).
N ( j)
Before stating it, we note that j∼i Wk,p+1 is a simple module, hence by Corollary 5.5 the (q,t)-
( j)
characters χq,t (Wk,p+1
e
) pairwise t-commute in Kt . Moreover, it is easy to check that, since C(z)
( j) ( j′ ) ( j′ ) ( j)
is symmetric, for any j ∼ i and j′ ∼ i, one has mk,p+1 ∗ mk,p+1 = mk,p+1 ∗ mk,p+1 , hence the (q,t)-
 
( j) ( j)
characters χq,t (Wk,p+1 ) do in fact pairwise commute in Kt . So we may write ∗ χq,t Wk,p+1
j∼i
without specifying an ordering of the factors.

Proposition 5.6 In Kt there holds:

         
χq,t Wk,p ∗ χq,t Wk,p+2 = t α (i,k) χq,t Wk−1,p+2 ∗ χq,t Wk+1,p + t γ (i,k) ∗ χq,t Wk,p+1 ,
(i) (i) (i) (i) ( j)
j∼i

where
1 e 
α (i, k) = −1 + Cii (2k − 1) + Ceii (2k + 1) , γ (i, k) = α (i, k) + 1. (34)
2

Proof — Using Theorem 5.4, we see that the claimed identity holds for some integers α (i, k) and
γ (i, k), and we only have to check (34). To do so it is enough to compare the coefficients of some
particular monomials on both sides. We have mk,p ∗ mk,p+2 = t α mk−1,p+2 ∗ mk+1,p , where
(i) (i) (i) (i)

k−1
1
α = ∑ N (i, p; i, p + 2a) + 2 N (i, p; i, p + 2k)
a=1
k−1   1 
= ∑ Ceii (2a + 1) − Ceii (2a − 1) +
2
Ceii (2k + 1) − Ceii (2k − 1)
a=1

1 e 
= −Ceii (1) + Cii (2k − 1) + Ceii (2k + 1) .
2
Thus α (i, k) = α = −1 + (Ceii (2k − 1) + Ceii (2k + 1))/2, as claimed.
(i) (i)
Similarly, χq (Wk,p ) contains the monomial m := mk,p A−1 −1 −1
i,p+2k−1 · · · Ai,p+3 Ai,p+1 with coeffi-
(i) ( j)
cient 1, and we have m mk,p+2 = ∏ j∼i mk,p+1 . Now
!
 −1
∏ mk,p+1 ∗ mk,p+2 = t γ ∏ mk,p+1 ,
(i) (i) ( j) (i) ( j)
m ∗ mk,p+2 = mk,p+2
j∼i j∼i

22
where
k k
1
2∑ ∑ ∑ N ( j, p + 2a − 1; i, p + 2b)
γ =
j∼i a=1 b=1
k k  
1
2∑ ∑∑
= e
C ji (2(a − b) − 2) − e
C ji (2(a − b)) − e
C ji (2(b − a)) + e
C ji (2(b − a) + 2)
j∼i a=1 b=1
k  
1
2∑ ∑
= Ce ji (2(a − k) − 2) − e
C ji (2(a − 1)) − e
C ji (2(1 − a)) + e
C ji (2(k − a) + 2)
j∼i a=1
k  
1
2∑ ∑
= −Ce ji (2(a − 1)) + e
C ji (2(k − a) + 2)
j∼i a=1
1
2∑
= Ceji (2k).
j∼i
   
Thus γ (i, k) = γ = ∑ j∼i Ceji (2k) /2 = ∑ j∼i Cei j (2k) /2 = (Ceii (2k − 1) + Ceii (2k + 1))/2, as clai-
e (see the proof of Proposition 2.5).
med. Here, the last equality comes from the definition of C(z)
✷

Example 5.7 (a) Take g of type A1 . We have

e = z − z3 + z5 − z7 + z9 − · · · ,
C(z)

hence α (k) = −1 for all k > 0. Thus we get

χq,t Wk,p ∗ χq,t Wk,p+2 = t −1 χq,t Wk−1,p+2 ∗ χq,t Wk+1,p + 1.

(b) Take g of type A3 . Choose i = 1, k = 1, and p = 0. Using for example Proposition 2.5, we can

calculate

Ce11 (z) = z − z7 + z9 − z15 + · · · , Ce12 (z) = z2 − z6 + z10 − z14 + · · · ,

hence
1 e  1 1 1
α (1, 1) = −1 + C11 (1) + Ce11 (3) = − , γ (1, 1) = Ce12 (2) = .
2 2 2 2
Thus Proposition 5.6 gives
       
(1) (1) (1) (2)
χq,t W1,0 ∗ χq,t W1,2 = t −1/2 χq,t W2,0 + t 1/2 χq,t W1,1 .

5.11 The subcategory CQ

Recall the quantum torus Yt,Q of §3.3. The dominant commutative monomials in Yt,Q parametrize
the simple objects of an abelian subcategory CQ of CZ . More precisely, we define CQ as the full
subcategory of CZ whose objects have all their composition factors of the form L(m) where m is
a dominant commutative monomial in Yt,Q . When Q is a sink-source orientation of the Dynkin
diagram and the Coxeter number h is even, CQ is one of the subcategories Cℓ introduced in [HL];
namely, CQ = Ch′ where h′ = h/2 − 1.

23
Lemma 5.8 CQ is closed under tensor products, hence is a tensor subcategory of CZ .

Proof — This is a slight modification of the proof of [HL, Proposition 3.2]. Let L(m) and L(m′ )
be in CQ . This means that m and m′ are monomials in the variables Yi,p , (i, p) ∈ IˆQ . If L(m′′ ) is a
composition factor of L(m)⊗L(m′ ) then m′′ is a product of monomials of χq (L(m)) and χq (L(m′ )).
So we have m′′ = mm′ M where M is a monomial in the A−1 j,r . Then it is checked as in [HL, Section
′′
5.2.4] that, for m to be dominant, these ( j, r) have to satisfy ( j, r − 1) ∈ IˆQ and ( j, r + 1) ∈ IˆQ . It
follows that m′′ depends only on the variables Yi,p , (i, p) ∈ IˆQ , because IbQ is a “convex slice” of Ib,
that is, it satisfies:
(i) if (i, p), (i, p + 2k) ∈ IbQ for i ∈ I, p ∈ Z, k > 0, then (i, p + 2 j) ∈ IbQ for 1 ≤ j ≤ k − 1;

(ii) if (i, p), (i, p + 2) ∈ IbQ for i ∈ I, p ∈ Z, then for every j ∼ i we have ( j, p + 1) ∈ IbQ .
Hence the result. ✷

Example 5.9 We continue Example 3.4. We take g of type D4 and choose Q as in Figure 1. The
simple objects of CQ are of the form L(m), where
u u u u u u u u u u u u
m = Y1,01,0 Y1,−2
1,−2 1,−4
Y1,−4 Y2,02,0 Y2,−2
2,−2 2,−4
Y2,−4 Y3,13,1 Y3,−1
3,−1 3,−3
Y3,−3 Y4,24,2 Y4,04,0 Y4,−2
4,−2
.

and ui,p ∈ N.

5.12 The ring Kt, Q and the truncated (q,t)-characters

We denote by Kt,Q the C(t 1/2 )-subalgebra of Kt spanned by the (q,t)-characters χq,t (L(m)) of
the simple objects L(m) in CQ . We call Kt,Q the t-deformed Grothendieck ring of CQ .
The (q,t)-character of a simple object L(m) of CQ contains in general many monomials m′
which do not belong to Yt,Q . By discarding these monomials we obtain a truncated (q,t)-charac-
ter. We shall denote by χeq,t (L(m)) the truncated (q,t)-character of L(m). One checks that for
a simple object L(m) of CQ , all the dominant monomials occuring in χq,t (L(m)) belong to the
truncated (q,t)-character χeq,t (L(m)) (the proof is similar to that of [HL] for the category C1 , as
for the proof of Lemma 5.8 above). Therefore the truncation map

χq,t (L(m)) 7→ χeq,t (L(m))

extends to an injective algebra homomorphism from Kt,Q to Yt,Q . In the sequel we shall identify
Kt,Q with the subalgebra of Yt,Q given by the image of this homomorphism.

6 An isomorphism between quantum Grothendieck rings and quan-

tum groups
6.1 The isomorphism between Kt, Q and Av (n)
Recall the isomorphism Φ : Yt,Q → Tv,Q of Proposition 3.8, and the notation

Av (n) := C(v1/2 ) ⊗ Av (n).

Define the rescaled dual canonical basis of Av (n):

n o
e ∗ := Be∗ (a) := vN(β (a))/2 B∗ (a) | B∗ (a) ∈ B∗ .
B

24
 e ∗ are invariant under the involution σ . The next theorem is Theorem 1.2
Clearly, the elements of B
in a slightly more precise formulation.

Theorem 6.1 (a) Φ restricts to an isomorphism

∼
Kt,Q −→ Av (n).

(b) The basis of Kt,Q consisting of the irreducible truncated (q,t)-characters χeq,t (L(m)) is
mapped by Φ onto Be ∗.

Proof — We introduce some necessary notation. For 1 ≤ k ≤ r, let kmin := min{1 ≤ s ≤ r | is = ik }.

Set k(0) := k and, for a negative integer j, define k( j) = (k( j+1) )− , where the notation b− is as in
Equation (26). We also note that, by definition of λk and βk , if k− 6= 0 then τ −1 (λk ) = λk− and
τ −1 (βk ) = βk− .
Let us fix some (i, p) ∈ IbQ . By definition of Φ, we have:

Φ(Yi,p ) = Xk for (i, p) ∈ IbQ and ϕ (i, p) = (βk , 0).

Note that this relation between (i, p) and βk implies in particular that ik = i. Since if k− 6= 0,

ϕ (i, p + 2) = (τ −1 (βk ), 0) = (βk− , 0),

we deduce that Φ Yi,pYi,p+2Yi,p+4 · · ·Yi,ξi is equal up to a power of v to Xk Xk− Xk(−2) · · · Xkmin , that is,
up to a power of v, to D(0, k). Since the commutative monomial Yi,pYi,p+2 · · ·Yi,ξi is bar-invariant,
its image is σ -invariant, so it has to be equal to vN(ϖik −λk )/2 D(0, k). Now Yi,pYi,p+2 · · ·Yi,ξi is equal
(i)
to the truncated (q,t)-character of the Kirillov-Reshetikhin module W1+(ξi −p)/2, p . Hence we have
shown that   
Φ χe(q,t) W1+(ξ −p)/2, p = vN(ϖik −λk )/2 D(0, k).
(i)
i

We now want to show that, more generally, for 1 ≤ s ≤ (ξi − p)/2 + 1 we have
  
Φ χe(q,t) Ws, p = vN(λk(−s) −λk )/2 D(k(−s) , k).
(i)
(35)

This will be proved by comparing Proposition 4.1 and Proposition 5.6. Let us denote by

e d) := vN(λb −λd )/2 D(b, d)

D(b,

the rescaled quantum minors. Note that

1
N(λb − λd ) = (λb − λd , λb − λd ) − deg(λb − λd ) = (λb , λb − λd ) − deg(λb − λd ).
2
We can rewrite Proposition 4.1 as
−→
e − , d − ) = vX D(b,
e d)D(b
D(b, e − , d) + vY ∏ D(b
e d − )D(b e − ( j), d − ( j)) (36)
j∼i

where
1
X := −1 + B − A + ((λb , λb − λd ) + (λb− , λb− − λd − ) − (λb− , λb− − λd ) − (λb , λb − λd − )) ,
2

25
and
!
1

Y := C − A + (λb , λb − λd ) + (λb− , λb− − λd − ) − ∑ λb− ( j) , λb− ( j) − λd − ( j) .
2 j∼i

Replacing A and B by their values from Proposition 4.1, and simplifying the resulting expression,
we easily get
1
X = −1 + (λb + λb− , λd − − λd ).
2
Now, writing ib = id = i and b = d (−s) ,

(λb + λb− , λd − − λd ) = (λb , λd − ) − (λb− , λd ) = (ϖi , τ s−1 (ϖi )) − (ϖi , τ s+1 (ϖi )).

Hence, using that τ s−1 (ϖi ) − τ s+1(ϖi ) = τ s (γi ) + τ s−1(γi ), by Proposition 2.1 we get
1 e 
X = −1 + Cii (2s − 1) + Ceii (2s + 1) .
2
Similarly, replacing A and C by their values from Proposition 4.1, and simplifying the resulting
expression, we get
1
2∑ ∑(τ (ξ j −ξk)/2 ϖ j , ϖk − τ s(ϖk )).
Y = (ϖi , τ (ϖi ) − τ s+1 (ϖi )) + (ϖi , ϖi − τ s (ϖi )) −
j∼i k∼i

Using the identities

s−1
ϖi − τ s (ϖi ) = ∑ τ l (γi ), τ l (γi ) + τ l+1 (γi ) = ∑ τ l+(1+ξk −ξi )/2 (γk ),
l=0 k∼i

we get
 
(ϖi , τ (ϖi ) − τ s+1(ϖi )) + (ϖi , ϖi − τ s (ϖi )) = ∑ τ (ξi −ξk −1)/2 (ϖi ), ϖk − τ s (ϖk ) ,
k∼i

hence, ! !
1
Y=
2 ∑ τ (ξi −ξk −1)/2 2ϖi − ∑ τ (ξ j −ξi +1)/2 (ϖ j ) , ϖk − τ s (ϖk ) .
k∼i j∼i

Now,

2ϖi − ∑ τ (ξ j −ξi +1)/2 (ϖ j ) = 2ϖi − ∑ τ (ϖ j ) − ∑ ϖ j = αi + ∑ γ j = γi .

j∼i j∼i; ξ j −ξi =1 j∼i; ξ j −ξi =−1 j∼i; ξ j −ξi =1

Hence
1  
Y =
2 ∑ τ (ξi−ξk −1)/2 (γi ), ϖk − τ s (ϖk )
k∼i
1  
= −
2 ∑ τ −s+(ξi −ξk −1)/2
(γi ), ϖ k
k∼i
1  
=
2 ∑ τ s+(ξi−ξk −1)/2 (γi ), ϖk
k∼i
1
=
2 ∑ Ceik (2s).
k∼i

26
Therefore X = α (i, s), Y = γ (i, s), and by Proposition 5.6 we see that, for any (i, p′ ) ∈ IbQ , there
holds in Kt :
         
(i) (i) (i) (i) ( j)
χq,t Ws, p′ ∗ χq,t Ws, p′ +2 = t X χq,t Ws−1, p′ +2 ∗ χq,t Ws+1, p′ + tY ∗ χq,t Ws, p′ +1 . (37)
j∼i

It was shown in [GLS] that one can express every quantum minor D(b, d) as a (noncommutative)
Laurent polynomial in the quantum flag minors D(0, k), by means of an explicit sequence of appli-
cations of Proposition 4.1. Equivalently, every rescaled quantum minor D(b, e d) can be expressed
e
as a Laurent polynomial in the D(0, k)’s with coefficients in Z[v ±1/2 ], by means of an explicit
sequence of applications of (36). By comparing (36) and (37), we see that the (q,t)-character of
(i)
Ws, p (where ϕ (i, p) = (βd , 0) and b = d (−s) ) can be expressed by the same Laurent polynomial in
the (q,t)-characters  
( j)
χq,t W1+(ξ j −p′ )/2, p′ , (( j, p′ ) ∈ IbQ ),

where v1/2 is replaced by t 1/2 . This proves (35). In particular, we have

  
(i)
Φ (χeq,t (L(Yi,p ))) = Φ χeq,t W1,p = D(d e − , d) = vN(βd )/2 E ∗ (βd ), ((i, p) ∈ IbQ ).

Thus, Φ maps the truncated (q,t)-characters of the fundamental modules of CQ , that is, a set of
algebra generators of Kt,Q , to the rescaled dual PBW generators of Av (n). This proves (a).
It follows that Φ maps the truncated (q,t)-characters of the standard modules of CQ to the
elements of the dual PBW-basis of Av (n) up to some power of v. Let us calculate this power
of v. By Proposition 5.1, we have that [M(m)]t − [L(m)]t is a linear combination of [L(m′ )]t with
coefficients in t −1 Z[t −1 ], where [L(m)]t and the [L(m′ )]t are bar-invariant. On the other hand,
note that the rescaling factor vN(β (a))/2 of the dual canonical basis depends only on the weight
of the vector B∗ (a). Hence if we write Ee∗ (a) = vN(β (a))/2 E ∗ (a), the transition matrix between
the rescaled dual PBW-basis {Ee∗ (a)} and the rescaled dual canonical basis {Be∗ (a)} is identical
to the transition matrix between the original bases. Thus, by Proposition 4.2, Ee∗ (a) − Be∗ (a) is
a linear combination of Be∗ (a′ ) with coefficients in v−1 Z[v−1 ], where Be∗ (a) and the Be∗ (a′ ) are
σ -invariant. By Proposition 3.8, Φ maps the set of bar-invariant elements of Kt,Q to the set of
σ -invariant elements of Av (n). This implies that Φ maps the basis of Kt,Q given by the truncated
(q,t)-characters of the standard modules of CQ , to the rescaled dual PBW-basis {Ee∗ (a)} of Av (n).
Finally, using again Proposition 4.2 and Proposition 5.1, this yields (b). ✷

Example 6.2 Let g be of type A2 . Let Q be the quiver of type A2 with height function ξ1 = 2 and
ξ2 = 1. We have i = (1, 2, 1), and

D(0, 1) = Dϖ1 ,s1 (ϖ1 ) , D(0, 2) = Dϖ2 ,s1 s2 (ϖ2 ) , D(1, 3) = Ds1 (ϖ1 ),s1 s2 s1 (ϖ1 ) = Dϖ2 ,s2 (ϖ2 ) .

Let e1 and e2 be the Chevalley generators of Uv (n). In the identification Av (n) ≡ Uv (n) we have
D(0, 1) ≡ e1 and D(1, 3) ≡ e2 .
In this case the quantum torus Yt,Q is generated by Y1,0 ,Y1,2 ,Y2,1 , so Kt,Q is generated by
χeq,t (L(Y1,0 )), χeq,t (L(Y1,2 )), χeq,t (L(Y2,1 )). The isomorphism Φ of Theorem 6.1 satisfies

Φ ( χeq,t (L(Y1,2 ))) = D(0, 1), Φ (χeq,t (L(Y1,0 ))) = D(1, 3).

Thus Theorem 6.1 implies that χeq,t (L(Y1,2 )) and χeq,t (L(Y1,0 )) generate Kt,Q and satisfy the quan-
tum Serre relations.

27
 This can easily be checked by means of the quantum T -system. Indeed we have by Proposi-
tion 5.6:
χq,t (L(Y1,0 )) ∗ χq,t (L(Y1,2 )) = t −1/2 χq,t (L(Y1,0Y1,2 )) + t 1/2 χq,t (L(Y2,1 )),
and by applying the bar-involution

χq,t (L(Y1,2 )) ∗ χq,t (L(Y1,0 )) = t 1/2 χq,t (L(Y1,0Y1,2 )) + t −1/2 χq,t (L(Y2,1 )).

Eliminating χq,t (L(Y1,0Y1,2 )) we get

(t −1/2 − t 3/2 )χq,t (L(Y2,1 )) = χq,t (L(Y1,2 )) ∗ χq,t (L(Y1,0 )) − t χq,t (L(Y1,0 )) ∗ χq,t (L(Y1,2 )),

which shows that Kt,Q is generated by χeq,t (L(Y1,2 )) and χeq,t (L(Y1,0 )). Finally, using that

χq,t (L(Y2,1 )) ∗ χq,t (L(Y1,0 )) = t −1 χq,t (L(Y1,0 )) ∗ χq,t (L(Y2,1 ))

we obtain that

χq,t (L(Y1,2 )) ∗ χq,t (L(Y1,0 ))2 − t χq,t (L(Y1,0 )) ∗ χq,t (L(Y1,2 )) ∗ χq,t (L(Y1,0 ))

= t −1 χq,t (L(Y1,0 )) ∗ χq,t (L(Y1,2 )) ∗ χq,t (L(Y1,0 )) − χq,t (L(Y1,0 ))2 ∗ χq,t (L(Y1,2 )),

which is the first quantum Serre relation. The second one is obtained similarly.

Example 6.3 In this example, we illustrate the calculations behind the proof of Theorem 6.1.
Let g be of type A3 . Let Q be the quiver of type A3 with height function ξ1 = ξ3 = 2 and ξ2 = 3.
Thus Q has source 2 and sinks 1, 3. We take i = (2, 1, 3, 2, 1, 3), hence

β1 = α2 , β2 = α1 + α2 , β3 = α2 + α3 , β4 = α1 + α2 + α3 , β5 = α3 , β6 = α1 ,

and
λ1 = ϖ2 − α2 , λ2 = ϖ1 − α1 − α2 , λ3 = ϖ3 − α2 − α3 ,
λ4 = ϖ2 − α1 − 2α2 − α3 , λ5 = ϖ1 − α1 − α2 − α3 , λ6 = ϖ3 − α1 − α2 − α3 .

Note that in this case w0 = c2 where c = s2 s1 s3 is a Coxeter element. Thus, this example illustrates
also Corollary 1.5. The quantum unipotent minors generating Tv,Q are

D(0, 1) = Dϖ2 ,s2 (ϖ2 ) , D(0, 2) = Dϖ1 ,s2 s1 (ϖ1 ) , D(0, 3) = Dϖ3 ,s2 s1 s3 (ϖ3 ) ,
D(0, 4) = Dϖ2 ,s2 s1 s3 s2 (ϖ2 ) , D(0, 5) = Dϖ1 ,s2 s1 s3 s2 s1 (ϖ1 ) , D(0, 6) = Dϖ3 ,w0 (ϖ3 ) .

The generators of the dual PBW-basis are

E ∗ (β1 ) = D(0, 1), E ∗ (β2 ) = D(0, 2), E ∗ (β3 ) = D(0, 3),

E ∗ (β4 ) = Ds2 (ϖ2 ),s2 s1 s3 s2 (ϖ2 ) , E ∗ (β5 ) = Ds2 s1 (ϖ1 ),s2 s1 s3 s2 s1 (ϖ1 ) , E ∗ (β6 ) = Ds2 s1 s3 (ϖ3 ),w0 (ϖ3 ) .

The new generators Xi of Tv,Q are

X1 = D(0, 1), X2 = v−1/2 D(0, 2), X3 = v−1/2 D(0, 3),

X4 = v−1 D(0, 4)D(0, 1)−1 , X5 = v−1 D(0, 5)D(0, 2)−1 , X6 = v−1 D(0, 6)D(0, 3)−1 .

Let us define integers λi j and µi j by

D(0, i)D(0, j) = vλi j D(0, j)D(0, i), Xi X j = vµi j X j Xi , (1 ≤ i, j ≤ 6).

28
The matrices L = [λi j ] and M = [µi j ] are given by
   
0 −1 −1 0 0 0 0 −1 −1 0 1 1
1 0 0 0 1 −1 1 0 0 −1 1 −1
   
1 0 0 0 −1 1    1 0 0 −1 −1 1 
L=  , 
M= .
0 0 0 0 0 0 0 1 1 0 −1 −1

0 −1 1 0 0 0 −1 −1 1 1 0 0
0 1 −1 0 0 0 −1 1 −1 1 0 0

The generators of Yt,Q are Y1,0 ,Y3,0 ,Y2,1 ,Y1,2 ,Y3,2 ,Y2,3 . The isomorphism Φ is defined by

Φ(Y1,0 ) = X5 , Φ(Y3,0 ) = X6 , Φ(Y2,1 ) = X4 , Φ(Y1,2 ) = X2 , Φ(Y3,2 ) = X3 , Φ(Y2,3 ) = X1 .

The truncated (q,t)-characters of the fundamental modules of CQ are expressed in terms of com-
mutative monomials by

χeq,t (Y1,2 ) = Y1,2 , χeq,t (Y1,0 ) = Y1,0 +Y1,2

−1 −1
Y2,1 +Y2,3 Y3,2 ,
χeq,t (Y2,1 ) = Y2,1 +Y1,2Y2,3
−1
Y3,2 , χeq,t (Y2,3 ) = Y2,3 ,
χeq,t (Y3,2 ) = Y3,2 , χeq,t (Y3,0 ) = Y3,0 +Y3,2
−1 −1
Y2,1 +Y2,3 Y1,2 ,

Here, we have used the shorthand notation χ

eq,t (m) instead of χeq,t (L(m)). We also have

χeq,t (Y1,0Y1,2 ) = Y1,0Y1,2 , χeq,t (Y2,1Y2,3 ) = Y2,1Y2,3 , χeq,t (Y3,0Y3,2 ) = Y3,0Y3,2 .

Using the expression of D(0, k) in terms of X j ’s and the definition of Φ, one checks that

Φ(Y2,3 ) = D(0, 1), Φ(Y1,2 ) = v−1/2 D(0, 2), Φ(Y3,2 ) = v−1/2 D(0, 3),
Φ(Y2,1Y2,3 ) = v−1 D(0, 4), Φ(Y1,0Y1,2 ) = v−1 D(0, 5), Φ(Y3,0Y3,2 ) = v−1 D(0, 6),

in agreement with Theorem 6.1. By Proposition 4.1, we have

v−1 D(1, 4)D(0, 1) = v−1 D(1, 1)D(0, 4) + D(0, 2)D(0, 3),

hence
v−1 D(1, 4) = (v−1 D(0, 4) + D(0, 2)D(0, 3))D(0, 1)−1 .
Therefore


Φ−1 v−1 D(1, 4) = (Y2,1Y2,3 + tY1,2 ∗Y3,2 ) ∗Y2,3
−1 −1
= Y2,1 + tY1,2 ∗Y3,2 ∗Y2,3 −1
= Y2,1 +Y1,2Y3,2Y2,3 ,

where the last equality follows from (11). Thus we have

Φ−1 v−1 D(1, 4) = Φ−1 v−1 E ∗ (β4 ) = χeq,t (Y2,1 ),

in agreement with Theorem 6.1. Next, we have again by Proposition 4.1,

D(2, 5)D(0, 2) = v−1 D(0, 5) + D(1, 4).

Hence  
Φ−1 (D(2, 5)) = Y1,0Y1,2 + t(Y2,1 +Y1,2Y2,3
−1
Y3,2 ) ∗ (t 1/2Y1,2 )−1 .
Now,
−1 −1
(Y1,0Y1,2 ) ∗Y1,2 = t 1/2Y1,0 ∗Y1,2 ∗Y1,2 = t 1/2Y1,0 ,

29
and similarly
−1
Y2,1 ∗Y1,2 = t −1/2Y2,1Y1,2
−1
, −1
(Y1,2Y2,3 Y3,2 ) ∗Y1,2 )−1 = t −1/2Y2,3
−1
Y3,2 .

Therefore,

Φ−1 (D(2, 5)) = Φ−1 (E ∗ (β5 )) = Y1,0 +Y1,2

−1 −1
Y2,1 +Y2,3 Y3,2 = χeq,t (Y1,0 ).

Similarly, starting from the minor identity

D(3, 6)D(3, 0) = v−1 D(0, 6) + D(1, 4).

we deduce that

Φ−1 (D(3, 6)) = Φ−1 (E ∗ (β6 )) = Y3,0 +Y3,2

−1 −1
Y2,1 +Y2,3 Y1,2 = χeq,t (Y3,0 ).

Thus we have checked that Φ maps the fundamental characters

χeq,t (Y1,0 ), χeq,t (Y3,0 ), χeq,t (Y2,1 ), χeq,t (Y1,2 ), χeq,t (Y3,2 ), χeq,t (Y2,3 ),

to the rescaled dual PBW generators vN(βk )/2 E ∗ (βk ), in agreement with Theorem 6.1.

6.2 Proof of Corollary 1.4

Let b1 , . . . , bk ∈ B∗ , and let L1 , . . . , Lk be the simple objects of CQ such that

Φ(χeq,t (Li )) ∈ vZ/2 bi , (1 ≤ i ≤ k).

We have Φ(χeq,t (L1 ) ∗ · · · ∗ χeq,t (Lk )) ∈ vZ/2 b1 · · · bk , thus, by Theorem 6.1, b1 · · · bk ∈ vZ B∗ if and
only if χeq,t (L1 ) ∗ · · · ∗ χeq,t (Lk ) is the (q,t)-character of a simple module up to a power of v, that is
by Corollary 5.5, if and only if L1 ⊗ · · · ⊗ Lk is simple. Hence Corollary 1.4 follows from [He4].

7 A presentation of quantum Grothendieck rings

In the remaining sections we drop the symbol ∗ for the t-deformed product of Kt , and simply
write xy instead of x ∗ y.

7.1 The generators

Fix an orientation Q of the Dynkin diagram of g. Define an involution ν of I by w0 (αi ) = −αν (i) .
For i ∈ I write ϕ −1 (αi , 0) = (ki , pi ) ∈ IbQ . Define the following elements of Kt :



i,m := χq,t L Yν m (ki ),pi +mh
xQ , (i ∈ I, m ∈ Z). (38)

The elements xQ i,0 belong to Kt,Q and map to the Chevalley generators Dϖi ,si (ϖi ) ≡ ei of Av (n) ≡
Uv (n) under the isomorphism Φ of Theorem 6.1. Hence Kt,Q has a presentation given by the
generators xQ
i,0 (i ∈ I) subject to the relations (see §4.1)

xQ Q Q Q
i,0 x j,0 − x j,0 xi,0 = 0 if Ci j = 0,
(xQ 2 Q −1 Q Q Q Q Q 2
i,0 ) x j,0 − (t + t )xi,0 x j,0 xi,0 + x j,0 ( xi,0 ) = 0 if Ci j = −1.

30
In particular, every χq,t (L(Yi,p )) with (i, p) ∈ IbQ can be written as a noncommutative polynomial
in the xQ
i,0 ’s.
For m ∈ Z, let K (m) be the subalgebra of Kt generated by the xQ i,m (i ∈ I). Thus K
(0) = K ,
t,Q
and K (m) is isomorphic to K (0) for every m ∈ Z. This comes from the fact that Kt is generated
by the fundamental (q,t)-characters χq,t (L(Yi,p )) ((i, p) ∈ Ib), and that the assignment


χq,t (L(Yi,p )) 7→ χq,t L(Yν (i), p+h )
extends to an algebra automorphism Σ of Kt . (In fact, L(Yi,p ) is the Uq (Lg)-module dual to
L(Yν (i), p+h ) [CP, §5], see also [FM, Cor. 6.10].) Let IbQ,m := ϕ −1 (∆+ ×{m}). Thus χq,t (L(Yi,p )) ∈
K (m) for (i, p) ∈ IbQ,m . Therefore, we have proved:
Lemma 7.1 The elements xQ
i,m (i ∈ I, m ∈ Z) generate Kt . ✷

7.2 The presentation

We start with the following:
Lemma 7.2 Let (i, p) ∈ Ib and ( j, p + h) ∈ Ib. Write V := L(Yi,p ) ⊗ L(Y j, p+h ).
(a) If j 6= ν (i) then V is simple.


(b) If j = ν (i) then χq (V ) = χq L(Yi,pYν (i), p+h ) + 1.
(c) In general we have



χq,t (L(Yi,p )) χq,t L(Y j,p+h ) = t −(αi ,αν ( j) ) χq,t L(Y j,p+h ) χq,t (L(Yi,p )) + δi ν ( j) (1 − t −2 ),
where δik is the Kronecker symbol δ .


Proof — Consider the product π := χq (L(Yi,p )) χq L(Yν (i), p+h ) . By [FM, §6], χq (L(Yi,p )) con-
tains only one dominant monomial, namely Yi,p , one anti-dominant monomial, namely Yν−1 (i),p+h ,
±1
and all its other monomials involve only variables of the form Y j,m with p < m < p + h. It fol-
lows that, if j 6= ν (i), then π contains no other dominant monomial than Yi,pY j, p+h , hence V is
irreducible and isomorphic to L(Yi,pYν (i), p+h ). This proves (a).
If j = ν (i) then π contains only two dominant monomials, that is, Yi,pYν (i), p+h and 1. Therefore
V has at most two composition factors, L(Yi,pYν (i), p+h ) and the trivial one-dimensional represen-
tation. Since L(Yi,p ) = L(Yν (i), p+h )∗ , the trivial representation is indeed a composition factor of V
because Uq (Lg) is a Hopf algebra. This proves (b).
It follows that



χq (L(Yi,p )) χq L(Y j,p+h ) = χq L(Yi,pY j,p+h ) + δi ν ( j) .
In Kt , this identity gets t-deformed as

1

χq,t (L(Yi,p )) χq,t L(Y j,p+h ) = t 2 N (i,p; j,p+h) χq,t L(Yi,pY j,p+h ) + δi ν ( j) .
Now using Proposition 3.2 and a sink-source orientation Q where i is a source, we see that
N (i, p; j, p + h) = N (i, 0; j, h) = −(αi , αν ( j) ). Using the bar involution, we also have

1

χq,t L(Y j,p+h ) χq,t (L(Yi,p )) = t − 2 N (i,p; j,p+h) χq,t L(Yi,pY j,p+h ) + δi ν ( j) .


Then (c) follows by eliminating χq,t L(Yi,pY j,p+h ) between these two equations. ✷

We can now give a presentation of Kt .

31
Theorem 7.3 The algebra Kt is isomorphic to the C(t 1/2 )-algebra A presented by generators
yi,m (i ∈ I, m ∈ Z) subject only to the following relations:
(R1) for every m ∈ Z,

yi,m y j,m − y j,m yi,m = 0 if (αi , α j ) = 0,

y2i,m y j,m − (t + t −1 )yi,m y j,m yi,m + y j,m y2i,m = 0 if (αi , α j ) = −1;

(R2) for every m ∈ Z and every i, j ∈ I,

yi,m y j,m+1 = t −(αi ,α j ) y j,m+1 yi,m + δi j (1 − t −2 );

(R3) for every p > m + 1 and every i, j ∈ I,

p−m (α
i ,α j )
yi,m y j,p = t (−1) y j,p yi,m .

Proof — We fix a sink-source orientation Q. We first check that the xQ i,m satisfy the above relations.
The relations (R1) are the Drinfeld-Jimbo relations for the subalgebra K (m) , as explained in §7.1.
The relations (R2) follow from Lemma 7.2 (c) when ξi = ξ j . If ξi 6= ξ j , then xQ Q
i,m x j,m+1 corresponds
to a tensor product of the form L(Yi,p ) ⊗ L(Y j,p+1 ) or L(Yi,p ) ⊗ L(Y j,p+2h−1 ). These two types of
tensor products are always irreducible [FM, Proposition 6.15]. Using Corollary 5.5, it follows that
xi,m and x j,m+1 t-commute, and the exponent of t is calculated by means of Proposition 3.2. For
the relations (R3) we note that L(Yi,p ) ⊗ L(Y j,s ) is irreducible if s − p > h [FM, Proposition 6.15],
and we conclude similarly.
It follows that we have a surjective homomorphism F from A to Kt given by yi,m 7→ xQ i,m , and
we have to show that this is an isomorphism. Define A (m) as we have defined K (m) before. Then
(m) Q
A is presented by the relations (1) (with xi,m replaced by yi,m ), so F restricts to an isomorphism
from A (m) to K (m) . It follows from the relations (R2) and (R3) that every monomial M in the
yi,m ’s can be rewritten as a linear combination of monomials of the form Mk1 Mk2 · · · Mks with Mk j ∈
←
A (k j ) and k1 > k2 > · · · > ks . So we have A = ∏ A (m) . Now each K (m) has a basis B (m)
m∈Z
consisting of the (q,t)-characters of standard modules that it contains. Taking

B ′ := {bk1 bk2 · · · bks | bk j ∈ A (k j ) , F(bk j ) ∈ B (k j ) , k1 > · · · > ks , s ∈ N},

we get a spanning set of A such that F(B ′ ) is a basis of Kt , consisting of the (q,t)-characters of
all the standard modules of CZ . Hence B ′ is a basis of A and F is an isomorphism. ✷

Example 7.4 Let g = sl2 . By Theorem 7.3, Kt is presented by generators ym := χq,t (L(Y2m ))
indexed by m ∈ Z, subject to

ym ym+1 = t −2 ym+1 ym + 1 − t −2 ,
p−m
ym y p = t 2(−1) y p ym , if p > m + 1.

Remark 7.5 (a) It was shown by Frenkel and Reshetikhin [FR2, Corollary 2] that the (classical)
Grothendieck ring R of CZ is the polynomial ring in the classes of all fundamental modules
L(Yi,p ) ((i, p) ∈ Ib). More recently, a presentation of R in terms of Kirillov-Reshetikhin modules
and T -systems was given in [IIKNS, Corollary 2.9].

32
 Note that our presentation of Kt does not yield a new presentation of R. Indeed, in order to
obtain R from Kt by specializing t at 1, one needs to use the integral form Kt defined in §8.2
below, and the xQ
i,m are not generators of Kt if g 6= sl2 .

(b) For m ∈ Z, let K (m,m+1) denote the subalgebra of Kt generated by yi,m , yi,m+1 (i ∈ I). It
follows from Theorem 7.3 that K (m,m+1) is isomorphic to the t-deformed boson algebra Bt (g)
introduced by Kashiwara [K, §3.3].

8 Derived Hall algebras

8.1 The Hall algebra H(Q)
Let F be a finite field, and let u := |F|1/2 ∈ R>0 . Let mod(FQ) be the abelian category of rep-
resentations of Q over F. The twisted Hall algebra H(Q), introduced by Ringel, is the C-algebra
with basis {zX } labelled by the isoclasses of objects in mod(FQ), with multiplication
zX zY = uhY,Xi ∑ gW
X,Y zW ,
W

where gW X,Y is the number of submodules T of W such that T ≃ X and W /T ≃ Y . Ringel [Ri2,
Ri3, Ri4] has shown that H(Q) is isomorphic to the C-algebra Uu (n) obtained from Uv (n) by
specializing v at u. In this isomorphism, the basis {zX } is mapped to a PBW-basis of Uu (n). In
particular, if Si denotes the 1-dimensional simple supported on i ∈ I, zSi is mapped to the Chevalley
generator ei .

8.2 The derived Hall algebra DH(Q)

Let Db (mod(FQ)) be the bounded derived category of mod(FQ). Toën [T, §7] has associated
with this triangulated category an associative algebra DH(Q) with the following presentation. The
[m]
generators zX are labelled by all pairs (X , m) where X is an isoclass of mod(FQ) and m ∈ Z. (The
pair (X , m) corresponds to the stalk complex with X in degree m.) The relations are:
(D1) for every m ∈ Z,
zX zY = uhY,Xi ∑ gW
[m] [m] [m]
X,Y zW ;
W

(D2) for every m ∈ Z,

= u−hY,Xi ∑ u−hW,T i γX,Y
[m] [m+1] T,W [m+1] [m]
zX zY zT zW ;
W,T

(D3) for p > m + 1,

[m] [p] p−m (X,Y ) [p] [m]
zX zY = u(−1) zY zX .

Here, the Hall number γX,Y

T,W
is defined by Toën as
| Ex(W,Y, X , T )|
γX,Y
T,W
:= ,
| Aut(X )|| Aut(Y )|
where Ex(W,Y, X , T ) is the finite subset of Hom(W,Y ) × Hom(Y, X ) × Hom(X , T ) consisting of
exact sequences 0 → W → Y → X → T → 0. Note that, as in §8.1, we have twisted the multipli-
[m] [p] p−m
cation by inserting in the original Hall product zX zY of [T] a factor u(−1) hY,Xi , see [S].
[m]
Consider the elements zi,m := zSi for i ∈ I and m ∈ Z. As in §7.1, we see that the zi,m generate
DH(Q). More precisely, we have:

33
Proposition 8.1 The algebra DH(Q) is generated by the zi,m (i ∈ I, m ∈ Z) subject only to the
following relations:

(H1) for every m ∈ Z,

zi,m z j,m − z j,m zi,m = 0 if (αi , α j ) = 0,

z2i,m z j,m − (u + u−1 )zi,m z j,m zi,m + z j,m z2i,m = 0 if (αi , α j ) = −1;

(H2) for every m ∈ Z and every i, j ∈ I,

u−1
zi,m z j,m+1 = u−(αi ,α j ) z j,m+1 zi,m + δi j ;
u2 − 1

(H3) For every p > m + 1 and every i, j ∈ I,

p−m
(αi ,α j )
zi,m z j,p = u(−1) z j,p zi,m .

Proof — The relations (H1) follow immediately from (D1) and Ringel’s theorem. The relations
(H3) follow immediately from (D3). Let us deduce the relations (H2) from (D2).
If i 6= j, the only exact sequences 0 → W → S j → Si → T → 0 are of the form

f 0 g
0 → S j → S j → Si → Si → 0

where 0 means the zero map, and f and g are isomorphisms. Clearly there are (|F| − 1)2 such
S ,S
sequences, and since | Aut(Si )| = | Aut(S j )| = |F| − 1, we get that γSij,S ji = 1. Hence

zi,m z j,m+1 = u−hS j ,Si i u−hSi ,S j i z j,m+1 zi,m = u−(αi ,α j ) z j,m+1 zi,m .

If i = j, we have two types of exact sequences 0 → W → Si → Si → T → 0, namely

f 0 g 0 h 0
0 → Si → Si → Si → Si → 0, and 0 → 0 → Si → Si → 0 → 0,

where f , g, h are isomorphisms. It follows that

1 1
γSSii,S,Sii = 1, and γS0,0
i ,Si
= = 2 ,
|F| − 1 u − 1

hence
1
zi,m zi,m+1 = u−(Si ,Si ) zi,m+1 zi,m + u−hSi ,Si i .
u2 − 1
This proves (H2). Finally, the proof that relations (H1), (H2), (H3) give a presentation of DH(Q) is
entirely similar to the proof of the analogous statement in Theorem 7.3 (the basis B (m) is replaced
[m]
by {zX | X isoclass of mod(FQ)}), and we omit it. ✷

34
8.3 The isomorphism between Ku and DH(Q)
Define the integral form M
Kt := C[t 1/2 , t −1/2 ] χq,t (L) ⊂ Kt ,
L

where the sum runs over all isoclasses L of simple objects in CZ . By Theorem 5.4, this is a subring
of Kt . Set
Ku := C ⊗C[t 1/2 ,t −1/2 ] Kt ,

where C is regarded as a C[t 1/2 , t −1/2 ]-module via the specialization t 1/2 7→ u1/2 .
For β ∈ ∆b = ∆+ × Z, we denote by z[m] the basis element z[m] of DH(Q) with X ∈ mod(FQ)
β X
indecomposable of dimension vector β .
The following is a slightly more precise formulation of Theorem 1.1.
∼
Theorem 8.2 There is a C-algebra isomorphism ι : Ku → DH(Q) such that:

(a) the class of the fundamental Uq (Lg)-module L(Yi,p ) of CZ is mapped by ι to a scalar multiple
[m]
of zβ , where (β , m) = ϕ (i, p).

(b) the basis of classes of standard Uq (Lg)-modules of CZ is mapped by ι to a rescaling of the

natural basis of DH(Q) labelled by all isoclasses of objects of Db (mod(FQ)).

Proof — We first assume, as in the proof of Theorem 7.3, that Q is a sink-source orientation of
the Dynkin diagram. We can rescale the generators xQ
i,m of Kt by setting

1
xeQ
i,m := xQ
i,m , (i ∈ I, m ∈ Z).
u1/2 (u − u−1 )

Clearly the new generators xeQi,m still satisfy the homogeneous relations (R1) and (R3) of Theo-
rem 7.3, and the relations (R2) become

1 − t −2
xeQ eQj,m+1 = t −(αi ,α j ) xeQj,m+1 xeQ
i,m x i,m + δi j .
u(u − u−1 )2

Let xQ eQ
i,m = 1 ⊗ x eQ
i,m ∈ Ku . By Theorem 7.3, Proposition 8.1, the assignment xi,m 7→ zi,m extends to
an algebra isomorphism ι . Indeed, in the relations (R2) we have

1 − u−2 u−1
=
u(u − u−1 )2 u2 − 1

so the generators xQ
i,m of Ku and zi,m of DH(Q) give rise to identical presentations.
Since the PBW-basis of Uv (n) is orthogonal with respect to the bilinear form of §4.1, it only
differs from the dual PBW-basis E∗ by scalar multiples. Hence by Ringel’s theorem, it follows
from Theorem 6.1 that the classes of fundamental modules in CQ , which correspond under Φ to
[0]
the elements E ∗ (β ) (β ∈ ∆+ ) of Uv (n), are mapped by ι to scalar multiples of the zβ . So, if
[0]
ϕ (i, p) = (β , 0) we have ι ([L(Yi,p )]u ) = λi,p zβ for some λi,p ∈ C. Therefore, using on one side the
automorphism of Ku given by [L(Yi,p )]u 7→ [L(Yi,p−2 )]u , and on the other side the corresponding
automorphism of DH(Q) induced by the Auslander-Reiten translation τ of Db (mod (FQ)), we
get (a).

35
 Since the classes of standard modules are the ordered products of the [L(Yi,p )]u (up to powers
[m]
of u), and the basis elements of DH(Q) are the ordered products of the zβ (up to powers of u),
we get (b).
Therefore we have proved Theorem 8.2 in the case of a sink-source orientation. But the C-
algebras Ku and DH(Q) are both independent of Q. For Ku this is clear. On the other hand if Q′ is
another orientation of the Dynkin diagram, then Db (mod(FQ)) and Db (mod(FQ′ )) are equivalent
triangulated categories, so DH(Q) and DH(Q′ ) are isomorphic. Thus Ku is isomorphic to DH(Q)
for an arbitrary orientation. More precisely, recall that the map ϕ = ϕQ : Ib → ∆ b depends on the
choice of Q. There is a triangle equivalence FQQ′ : D (mod(FQ)) → D (mod(FQ′ )) such that the
b b

induced isomorphism fQQ′ : DH(Q) → DH(Q′ ) satisfies

 
[m] [m′ ]
fQQ′ zβ = zβ ′ where (β ′ , m′ ) = ϕQ′ ϕQ−1 (β , m).

Therefore (a) and (b) hold for an arbitrary orientation. ✷

In the proof of Theorem 7.3, we have shown that if Q is a sink-source orientation, the gen-
erators xQ
i,m of Kt satisfy the relations (R1), (R2), (R3). We can see now that this holds for any
orientation Q.

Corollary 8.3 The generators xQi,m of Kt satisfy the same relations for every orientation Q of the
Dynkin diagram, namely the relations (R1), (R2), (R3) of Theorem 7.3.

Proof — Let Q be any orientation, by Theorem 8.2, the elements xQ i,m of Ku are mapped by ι to
scalar multiples of the generators zi,m of DH(Q). Now the relations (H1), (H2), (H3) satisfied by
the zi,m are independent of Q. Moreover, they are all homogeneous except for (H2) with i = j.
Since scalar multiplication does not affect homogeneous relations, the elements 1 ⊗ xQ i,m of Ku
1/2
satisfy the relations (R1), (R2) (i 6= j), (R3) with t replaced by u = |F| . Since this is true for
every finite field F, it follows that the elements xQ
i,m of Kt satisfy the relations (R1), (R2) (i 6= j),
(R3) where t is an indeterminate.
Finally, the relations (R2) (i = j) follow from Lemma 7.2 (c) with i = ν ( j). ✷

Remark 8.4 Using Remark 3.3, one can modify the presentation of Kt to obtain a presentation of
the deformed Grothendieck ring Rt of [N3, VV1]. This presentation shows that the specialization
of Rt at t = u−1 is isomorphic to the non-twisted derived Hall algebra of Db (mod(FQ)) with the
opposite product.

9 Quiver varieties
In this section we show that the variety Ed of representations of Q with dimension vector d can
be regarded as a Nakajima graded quiver variety M•0 (W d ) for an appropriate Ib-graded vector
space W d . Moreover the stratification of Ed by Gd -orbits coincides with Nakajima’s stratifica-
tion of M•0 (W d ). It follows that the set of perverse sheaves used by Lusztig to define the (dual)
canonical basis of Uv (n) can be identified with a subset of the set of perverse sheaves used by
Nakajima for describing the classes [L]t of simple Uq (Lg)-modules. This gives a geometric way
of understanding Theorem 6.1 (b).

36
9.1 The quiver representation space Ed
Let d = (di )i∈I ∈ NI denote a dimension vector for Q. We will identify d with the element ∑i∈I di αi
of the root lattice of g. The variety Ed of representations of Q of dimension d is by definition
M
Ed := HomC (Cdi , Cd j ),
i→ j

the sum being over all arrows i → j of Q. This is just a C-vector space of dimension ∑i→ j di d j , but
the interesting geometry comes from the following stratification. Consider the algebraic group

Gd := ∏ GL(di , C).
i∈I

It acts on Ed by base change. There are finitely many orbits in one-to-one correspondence with
the isomorphism classes of representations of Q of dimension d. Thus, using Gabriel’s theorem,
these orbits have a natural labelling by the set
( )
r
Id := a = (ak ) ∈ Nr | ∑ ak βk = d ,
k=1

where the positive roots βk are enumerated as in (14). Let Oa denote the orbit labelled by the
element a of Id . Let IC(Oa ) be the intersection cohomology complex of Oa , extended by zero on
the complement of Oa . Let H i (IC(Oa )) be its ith cohomology sheaf, and H i (IC(Oa ))c the stalk
of this sheaf at a point of Oc .
Recall from §4.3 the dual PBW basis E∗ and the dual canonical basis B∗ of Av (n). Write

E ∗ (c) = ∑ κa,c (v) B∗ (a).

a∈Id

Lusztig has shown:

Theorem 9.1 [Lu1, §9, §10] The coefficients κa,c (v) are given by

κa,c (v) = vdim Oc −dim Oa ∑ vi dim H i (IC(Oa ))c . (39)

i≥0

9.2 Nakajima’s variety M•0 (W )

Let M
W= Wi (p)
(i,p)∈Ib

be a finite-dimensional Ib-graded C-vector space. In his geometric construction of representations

of Uq (Lg), Nakajima [N3] has associated with W an affine variety M•0 (W ) whose definition we
shall now recall.
Let Jb := {(i, p) ∈ I × Z | (i, p − 1) ∈ Ib}, and let
M
V= Vi (s)
(i,s)∈Jb

37
be a finite-dimensional Jb-graded C-vector space. Define
M
L• (V,W ) = Hom(Vi (s),Wi (s − 1)),
(i,s)∈Jb
M
•
L (W,V ) = Hom(Wi (p),Vi (p − 1)),
(i,p)∈Ib
M
•
E (V ) = Hom(Vi (s),V j (s − 1)).
(i,s)∈Jb; j∼i

Put M • (V,W ) = E • (V ) ⊕ L• (W,V ) ⊕ L• (V,W ). An element of M • (V,W ) is written (B, α , β ), and

its components are denoted by:

Bi j (s) ∈ Hom(Vi (s),V j (s − 1)),

αi (p) ∈ Hom(Wi (p),Vi (p − 1)),
βi (s) ∈ Hom(Vi (s),Wi (s − 1)).

We denote by Λ• (V,W ) the subvariety of the affine space M • (V,W ) defined by the equations

αi (s − 1)βi (s) + ∑ ε (i, j) B ji (s − 1)Bi j (s) = 0, ((i, s) ∈ Jb), (40)

j∼i

where ε (i, j) = 1 (resp. ε (i, j) = −1) if i → j is an arrow of Q (resp. i → j is not an arrow of Q).
The algebraic group
GV := ∏ GL(Vi (s))
(i,s)∈Jb

acts on M • (V,W ) by base change in V :

g · (B, α , β ) = (g j (s − 1)Bi j (s)gi (s)−1 ), (gi (p − 1)αi (p)), (βi (s)gi (s)−1 ) .

Note that there is no action on W . This action of GV preserves the subvariety Λ• (V,W ). One
defines the affine quotient
M•0 (V,W ) := Λ• (V,W )  GV .
By definition, the coordinate ring of M•0 (V,W ) is the ring of GV -invariant functions on Λ• (V,W ),
and M•0 (V,W ) parametrizes the closed GV -orbits. If Vi (s) ⊆ Vi′ (s) for every (i, s) ∈ Jb, then we
have a natural closed embedding M•0 (V,W ) ⊂ M•0 (V ′ ,W ). Finally, one defines
[
M•0 (W ) := M•0 (V,W ).
V

This is an affine variety, acted upon by the algebraic group

GW := ∏ GL(Wi (p)).
(i,p)∈Ib

Let M•0 reg (V,W ) be the open subset of M•0 (V,W ) parametrizing the closed free GV -orbits. For a
• reg
given W , we have M0 (V,W ) 6= 0/ only for a finite number of V ’s. Nakajima has shown that this
gives a stratification of M•0 (W ):
G • reg
M•0 (W ) = M0 (V,W ).
V

38
 • reg
A necessary condition for M0 (V,W ) to be nonempty is that
dimWi (p) − dimVi (p + 1) − dimVi (p − 1) + ∑ dimV j (p) ≥ 0
j∼i

for every (i, p) ∈ Ib. In this case we say that (V,W ) is a dominant pair. Equivalently, by (32), the
pair (V,W ) is dominant if and only if the monomial Y W AV ∈ Y is dominant, where we use the
shorthand notation

∏ ∏
dimWi (p) − dimVi (s)
Y W := Yi,p , AV := Ai,s .
(i,p)∈Ib (i,s)∈Jb

Note that this stratification of M•0 (W ) is GW -invariant. Hence each stratum is a union of GW -orbits.
• reg
Given a dominant pair (V,W ) such that M0 (V,W ) 6= 0, / we denote by ICW (V ) the intersec-
• reg
tion cohomology complex of the closure of the stratum M0 (V,W ). Let H i (ICW (V )) be its ith
• reg
cohomology sheaf, and H i (ICW (V ))V ′ be the stalk of this sheaf at a point of M0 (V ′ ,W ).
For a dominant monomial m ∈ Y , recall from §5.6 and §5.8 the (q,t)-characters χq,t (M(m))
and χq,t (L(m)) of the standard and of the simple Uq (Lg)-modules labelled by m. Write
χq,t (M(m′ )) = ∑ ζm,m′ (t) χq,t (L(m)).
m′

Nakajima has shown:

Theorem 9.2 [N3, §8] The coefficients ζm,m′ (t) are given by

∑ t i dim H i(ICW (V ))V ,

• reg • reg
(V ′ ,W )−dim M0
ζm,m′ (t) = t dim M0 (V,W )
′ (41)
i≥0

for any pair of strata M•0 reg (V,W ) and M•0 reg (V ′ ,W ) such that m = Y W AV and m′ = Y W AV .
′

Remark 9.3 (a) In order to make the comparison between Theorem 9.1 and Theorem 9.2 easier,
we stated Nakajima’s formula (41) in a different way from the original one. In [N3], Nakajima
writes
t dim M0 (V ,W ) ∑ t −i dim H i (i!xV ′ ICW (V ))
• reg ′

i≥0
for the right-hand side of (41), but in his degree convention the trivial local system on the open
• reg
stratum S = M0 (V,W ) appears in the intersection cohomology complex of S in degree dim S,
while in Lusztig’s convention it appears in degree 0. Here we follow Lusztig’s convention. More-
over Nakajima uses the costalk i!x at a point x instead of the stalk i∗x , which explains the change of
t i into t −i .
(b) A dominant monomial m can be written in several ways as m = Y W AV . The fact that the
right-hand side of (41) depends only on the monomials m and m′ , and not on the particular choices
of spaces W,V,V ′ , follows from a transversal slice argument [N1, §3].

9.3 An isomorphism
Let d = (di ) be a dimension vector, as in §9.1. Recall the bijection ϕ : Ib → ∆
b of §2.2. We define
b d
an I -graded space W by taking
W jd (p) := Cdi if ϕ ( j, p) = (αi , 0),

and W jd (p) := 0 for all others ( j, p) ∈ Ib. Clearly, the group GW d is isomorphic to Gd and we may
identify GW d ≡ Gd .

39
Proposition 9.4 There is a Gd -equivariant closed immersion of affine varieties

Ψ : M•0 (W d )−→Ed .

The proof of Proposition 9.4 will follow from the next two lemmas.
Lemma 9.5 Let i 6= i′ ∈ I and set ( j, p) = ϕ −1 (αi , 0), ( j′ , p′ ) = ϕ −1 (αi′ , 0). Assume that p′ ≤ p,
and write ϕ ( j′ , ξ j + p′ − p + 2) = (β , 0). Then, the coefficient of α j in the expansion of the root β
on the basis of simple roots is equal to 1 if there is an arrow i → i′ in Q, and to 0 otherwise.

Proof — By definition of ( j, p), ( j′ , p′ ), and of ϕ (see §2.2), we have

αi = τ (ξ j −p)/2 (γ j ), αi′ = τ (ξ j′ −p )/2 (γ j′ ).

′

It follows that
β = τ −1+(ξ j′ −ξ j −p +p)/2 (γ j′ ) = τ −1+(p−ξ j )/2 (αi′ ).
′

Recall the Ringel bilinear form h·, ·i. It may be characterized by

hαi , γ j i = δi j , (i, j ∈ I).

Hence, the coefficient of α j in β is equal to:

hβ , γ j i = hτ −1+(p−ξ j )/2 (αi′ ), γ j i = hτ −1 (αi′ ), τ (ξ j −p)/2 (γ j )i = hτ −1 (αi′ ), αi i = −hαi , αi′ i.

Now
−hαi , αi′ i = − dim(Hom(Si , Si′ )) + dim(Ext1 (Si , Si′ )) = dim(Ext1 (Si , Si′ )),
and this is equal to 1 if there is an arrow from i to i′ in Q, and to 0 otherwise. ✷

We now introduce an algebra Λ

e Q defined by a quiver Γ
eQ with relations. The vertices of Γ
eQ are
of two types:
• w j (p) for every ( j, p) = ϕ −1 (αi , 0) (i ∈ I);

• v j (p − 1) for every pair ( j, p) ∈ IbQ such that ( j, p − 2) ∈ IbQ .

The arrows of Γ
eQ are of three types:

• a j (p) : w j (p) → v j (p − 1);

• b j (p) : v j (p) → w j (p − 1);

• Bi j (p) : vi (p) → v j (p − 1) if j ∼ i.
The relations are:
ai (p − 1)bi (p) = ∑ ε (i, j)B ji (p − 1)Bi j (p).
j∼i

Obviously, as suggested by the notation, the definition of Λ e Q is so that the affine variety Λ• (V,W d )
is nothing but the representation variety of Λ e Q consisting of representations for which the spaces
d d
W j (p) = C are attached to the vertices w j (p), and the spaces V j (p−1) are attached to the vertices
i

v j (p − 1) (we assume that V j (p − 1) = 0 if v j (p − 1) is not a vertex of Γ eQ ).

For i ∈ I, we denote by εi the idempotent of Λ e Q associated with the vertex w j (p) such that
( j, p) = ϕ (αi , 0). We endow I with a total ordering such that i > i′ if p > p′ , where as above
−1

ϕ −1 (αi′ , 0) = ( j′ , p′ ). It is well-known that if there is an arrow i → i′ in Q then i > i′ .

40
Lemma 9.6 For i 6= i′ ∈ I, we have
(
  1 if there is a path from i to i′ in Q,
dim εi′ Λ
e Q εi =
0 otherwise.

Proof — If i < i′ then p ≤ p′ and clearly εi′ Λ e Q εi = 0. On the other hand if i < i′ there can be no
path from i to i′ in Q. Thus the lemma is clear in this case, and we may assume from now on that
i > i′ .
Let x ∈ εi′ Λ
e Q εi , and let i′′ ∈ I. Then x ∈ εi′ Λ
e Q εi′′ Λ
e Q εi if and only if x belongs to the two-sided
ideal of ΛQ generated by a j′′ (p − 1)b j′′ (p) where ϕ ( j′′ , p − 1) = (αi′′ , 0). This is because b j′′ (p)
e
is the only arrow entering in w j′′ (p − 1), and a j′′ (p − 1) is the only arrow exiting from w j′′ (p − 1).
Note that εi′ Λ
e Q εi′′ Λ
e Q εi 6= 0 implies that i > i′′ > i′ .
Let I be the the two-sided ideal of Λ e Q generated by all the degree two paths:

a j (p − 1)b j (p), (( j, p − 1) = ϕ −1 (αi , 0), (i ∈ I)).

Then, the algebra Λ e Q /I is defined by the same relations as the graded preprojective algebra Λ b
of [Le, §2.8] (but we have the additional vertices w j (p) and only a finite set of vertices v j (p)).
It follows that dim εi′ ΛQ εi / (εi′ I εi ) is equal to the v j′ (p′ + 1)-component of the dimension
e
vector of the indecomposable projective Λ-module
b with top v j (p − 1). Now it is well-known that
this dimension vector can be read off from the Auslander-Reiten quiver of Q. Namely, using our
notation, the v j (p − 1)-component of the dimension vector of the projective with top vi (ξi − 1) is
equal to the coefficient of αi in the root β such that ϕ ( j, p) = (β , 0). The dimension vectors of
the remaining indecomposable projectives are obtained from these particular ones by translation.
It follows that we can reformulate Lemma 9.5 as follows:
(
   1 if there is an arrow from i to i′ in Q,
dim εi ΛQ εi / (εi I εi ) =
′ e ′ (42)
0 otherwise.

In particular if i′ is the successor of i in the descending total order defined above, then εi′ I εi = 0,
and we have (
  1 if there is an arrow from i to i′ in Q,
dim εi′ Λ
e Q εi = (43)
0 otherwise.
Assume now that i > i′ are such that there is no path from i to i′ in Q. Then in particular there
is no arrow i → i′ , so by (42) we have
  
e Q εi = εi′ I εi = ∑ εi′ Λ
εi′ Λ eQε j ε jΛ
e Q εi .
i′ < j<i

For each summand, we have either no path from i to j or no path from j to i′ . So we can iterate
the splitting until we obtain an expression of the form
   
εi′ Λ
e Q εi = ∑ εi′ Λ
e Q εi · · · εi Λ
k 1 Q εi ,
e
i′ <i1 <···<ik <i

where in the right-hand side each factor εk Λ

e Q ε j is such that either we have a path from j to k or
k is the successor of j and there is no arrow from j to k. Moreover, since we have no path from

41
i to i′ each summand contains at least one factor of the second type, which is equal to 0 by (43).
Hence we have shown that εi′ Λe Q εi = 0.
Assume now that there is an arrow i → i′ in Q. Then we have as above
  
εi I εi = ∑ εi ΛQ ε j ε j ΛQ εi ,
′ ′ e e
i′ < j<i

where for each j we have either no path from i to j or no path from j to i′ (because  the Dynkin

diagram is a tree). Thus it follows from above that εi′ I εi = 0, so by (42) we get dim εi′ Λ
e Q εi = 1.
Finally, if there is a path i → i1 → · · · → ik → i′ in Q, with k ≥ 1, then there is no arrow from
i to i′ , and by (42) we have εi′ Λ
e Q εi = εi′ I εi . Moreover, this path is unique, and arguing as above
we can write    
εi′ Λ
e Q εi = εi′ Λe Q εi · · · εi Λ
k 1
e Q εi ,
 
where each factor has dimension 1, so again dim εi′ Λ e Q εi = 1. ✷

Proof of Proposition 9.4 — Let V be a Jb-graded space, and pick (B, α , β ) ∈ Λ• (V,W d ). As
explained above, we can regard (B, α , β ) as a representation of Λ
e Q . Choose two vertices i and i′
of Q, and set as before

( j, p) = ϕ −1 (αi , 0), ( j′ , p′ ) = ϕ −1 (αi′ , 0).

By Lemma 9.6 we have dim(εi′ Λ e Q εi ) ≤ 1. Let θii′ be a generator of εi′ Λ

e Q εi . By the proof of
Lemma 9.6, we can normalize the θii′ so that they verify θi′ i′′ θii′ = θii′′ for every i, i′ , i′′ ∈ I. More
precisely, if there is a path i → i1 → · · · → ik → i′ in Q, then θi,i′ = θik i′ · · · θii1 , and if there is no
path from i to i′ then θii′ = 0. Evaluating θii′ in the representation (B, α , β ) we obtain a linear map
ψii′ : W jd (p) → W jd′ (p′ ). The collection of maps (ψii′ ) for all arrows i → i′ of Q can be regarded as a
representation ψ of Q of dimension vector d. It follows easily from the definition of the GV -action
that ψ depends only on the GV -orbit of (B, α , β ). Hence the assignment (B, α , β ) 7→ ψ induces
a morphism of varieties ΨV : M•0 (V,W d ) → Ed . Moreover, it follows from the known description
of the generators of the coordinate ring of M•0 (V,W d ) (see [N4, §3.1]) that this coordinate ring is
generated by the matrix coefficients of the linear maps ψii′ for all pairs (i, i′ ). Hence ΨV induces
a surjective morphism from C[Ed ] to C[M•0 (V,W d )], thus ΨV is a closed immersion. Since for V
large enough we have M•0 (V,W d ) = M•0 (W d ), we obtain a closed immersion Ψ : M•0 (W d ) → Ed .
By construction, Ψ commutes with the actions of Gd on both varieties. ✷

Example 9.7 Take g of type D4 . We label the Dynkin diagram so that the central node is num-
bered 3, and we choose ξ1 = ξ2 = ξ4 = 4 and ξ3 = 5. Thus Q has a sink-source orientation with
source 3 and sinks 1, 2, 4. Given a dimension vector d = (d1 , d2 , d3 , d4 ), the corresponding graded
space W d is given by

W1d (0) = Cd1 , W2d (0) = Cd2 , W3d (5) = Cd3 , W4d (0) = Cd4 ,

and the other Wi (p)’s are zero (see the Auslander-Reiten quiver of Q in Figure 4). An element
(B, α , β ) of Λ(V,W d ) is represented in Figure 3. The defining equations (40) read

B13 (3)B31 (4) + B23 (3)B32 (4) + B43 (3)B34 (4) = 0,

B31 (2)B13 (3) = B32 (2)B23 (3) = B34 (2)B43 (3) = 0.

42
 W3 (5)
 α3 (5)
❢❢❢ V3 (4) ❏❏B34 (4)
❢❢ ❢ ❢❢❢❢❢❢ ttt ❏❏❏
❢❢❢❢B31 (4) yttB32 (4) %
s❢❢❢❢❢
V1 (3) ❳❳❳❳❳ V2 (3) V4 (3)
❳❳❳❳B❳13 (3) ❏❏❏B23 (3)
tt
❳❳❳❳❳ ❏
❳❳❳❳❳ ❏%
❳+ ytttB43 (3)
V (2)
❢❢❢❢❢ 3 ❏❏❏B34 (2)
❢❢ ❢❢ ❢❢❢❢❢ tttt ❏❏
❢❢❢❢ B31 (2) y
t B32 (2) %
s❢❢❢
V1 (1) V2 (1) V4 (1)
 β1 (1)  β2 (1)  β4 (1)
W1 (0) W2 (0) W4 (0)

Figure 3: (B, α , β ) in type D4 .

Thus,
β1 (1)B31 (2)B43 (3)B34 (4)α3 (5) = −β1 (1)B31 (2)B23 (3)B32 (4)α3 (5)
and
β1 (1)B31 (2)B13 (3)B31 (4)α3 (5) = 0.
Hence we can take
ψ31 := β1 (1)B31 (2)B43 (3)B34 (4)α3 (5),
and similarly

ψ32 := β2 (1)B32 (2)B13 (3)B31 (4)α3 (5), ψ34 := β4 (1)B34 (2)B23 (3)B32 (4)α3 (5).

We get a representation ψ := (ψ31 , ψ32 , ψ34 ) of Q on the space W d .

• reg
Proposition 9.8 There is a bijection between the set of (nonempty) strata M0 (V,W d ) and the
set Id of Gd -orbits.

• reg
Proof — Let us first consider a stratum M0 (V,W d ). By §9.2, the pair (V,W d ) is a dominant
pair. This means that we have nonnegative integers a j (1 ≤ j ≤ r) such that
r
Y W AV = ∏ Yi j ,pj j .
d a
(44)
j=1

Here for 1 ≤ j ≤ r, we have put (i j , p j ) = ϕ −1 (β j , 0). Indeed, by definition, every dominant

d
commutative monomial of the form Y W AV belongs to Yt,Q . Moreover we have a natural grading
of Yt,Q by the root lattice of g given by

deg(Yi j ,p j ) = β j , (1 ≤ j ≤ r).

It is easy to see that for every Ai,s ∈ Yt,Q we have deg Ai,s = 0. Therefore
r  d
∑ a j β j = deg Y W = d.
k=1

43
 Y3,5 ❡❡25 α i❙
❡ ❡❡ ❡❡ ❡❡❡❦❡❦❡❦❡❦❦❦ 3 ❙❙❙❙❙❙❙
❡❡❡❡❡❡
❡❡ ❦❦❦❦ ❙❙
Y1,4 Y2,4 Y4,4 α1 +α3 l❨❨❨❨❨ α2 +α3 i❙ 5 α3 + α4
❨❨❨❨❨❨ ❙❙❙❙ ❦❦❦❦
❨❨❨❨❨❨❙❙❙ ❦ ❦ ❦
❦
Y3,3 2 α1 +α2 +2α3 +α4
❡❡ ❡ ❦ 5 i❙❙❙❙
❡❡❡❡❡❦❦❦❦❦❦ ❙❙❙❙
❡❡❡❡❡❡ ❦
α2 +α3 +α4 l❨❨❨ α1 +α3 +α4
Y1,2 Y2,2 Y4,2 ❨❨❨❨❨❨ i❙❙❙ 5 α1 + α2 + α3
❨❨❨❨❨❨❙❙❙❙ ❦❦ ❦❦❦
❨ ❦❦
2 α1 + α2 + α3 + α4
Y3,1 ❡❡ ❡❡ ❡❡ ❦❦ 5 i❙❙❙❙
❡ ❡❡ ❡❡❡❡❡❡❡❡❦❦❦❦❦❦❦ ❙❙❙❙
❙
α α α
❡
1 2 4
Y1,0 Y2,0 Y4,0

Figure 4: The skeleton of CQ and the Auslander-Reiten quiver in type D4 .

Hence, to every stratum M•0 reg (V,W d ) corresponds an element a of Id given by (44).
a
Conversely, if a ∈ Id , we need to show that ma := ∏rj=1 Yi j ,pj j can be written in the form (44)
• reg
for some nonempty stratum M0 (V,W d ). By [N1, Th. 14.3.2], this is equivalent to the fact
d
that ma appears in the q-character of the standard module M(Y W ). For i ∈ I write as in §7.1
(ki , pi ) = ϕ −1 (αi , 0). Then we have by definition of W d

Y W = ∏ Ykdi ,p
d
i
i
.
i∈I

By §7.1 we know that the (q,t)-characters of the fundamental modules L(Yki ,pi ) (i ∈ I) generate
Kt,Q . Hence the simple module L(ma ), which is an object of CQ , is a composition factor of a
standard module of the form M(∏i∈I Ykeii,pi ) for some nonnegative integers ei . But, as before, we
must have
d = deg(ma ) = ∑ ei deg(Yki ,pi ) = ∑ ei αi ,
i∈I i∈I
d
hence ei = di for every i. Therefore ma is indeed a weight of M(Y W ). This proves the claim. ✷

• reg
Remark 9.9 The proof of Proposition 9.8 shows that M0 (V,W d ) is a nonempty stratum of
M•0 (W d ) if and only if (V,W d ) is a dominant pair, a purely combinatorial condition. In gen-
eral Nakajima [N1, Th. 14.3.2] only shows that this is a necessary condition. In representation-
d
theoretic terms, this means that every dominant monomial of the form Y W AV occurs in the q-
d
character of the standard module M(Y W ).

Example 9.10 We continue Example 9.7. There are 12 positive roots βk , which we identify with
the vertices of the Auslander-Reiten quiver of Q represented in Figure 4. The numbering is ob-
tained by reading this graph from top to bottom and left to right:

β1 = α3 , β2 = α1 + α3 , β3 = α2 + α3 , β4 = α3 + α4 , ··· , β12 = α4 .

The corresponding generators Yik ,pk of Yt,Q can be read at the corresponding place on the left side
of Figure 4. Let d = (d1 , d2 , d3 , d4 ) be a dimension vector for Q. Then
d d1 d2 d3 d4
Y W = Y1,0Y2,0Y3,5Y4,0 .

44
The elements of Id are 12-tuples a ∈ N12 encoding the decompositions of d into a sum of positive
roots. By Proposition 9.8, they are in one-to-one correspondence with the dominant monomials of
d
the form Y W AV . This bijection can be read immediately from Figure 4.
For example, if d = (1, 1, 1, 1) ≡ α1 + α2 + α3 + α4 , the correspondence is:

(α1 ) + (α2 ) + (α3 ) + (α4 ) ↔ Y1,0Y2,0Y3,5Y4,0 ↔ 1,

(α1 + α3 ) + (α2 ) + (α4 ) ↔ Y1,4Y2,0Y4,0 ↔ A1,1 A3,2 A2,3 A4,3 A3,4 ,
(α2 + α3 ) + (α1 ) + (α4 ) ↔ Y2,4Y1,0Y4,0 ↔ A2,1 A3,2 A1,3 A4,3 A3,4 ,
(α3 + α4 ) + (α1 ) + (α2 ) ↔ Y4,4Y1,0Y2,0 ↔ A4,1 A3,2 A1,3 A2,3 A3,4 ,
(α1 + α2 + α3 ) + (α4 ) ↔ Y4,2Y4,0 ↔ A1,1 A2,1 A23,2 A1,3 A2,3 A4,3 A3,4 ,
(α1 + α3 + α4 ) + (α2 ) ↔ Y2,2Y2,0 ↔ A1,1 A4,1 A23,2 A1,3 A2,3 A4,3 A3,4 ,
(α2 + α3 + α4 ) + (α1 ) ↔ Y1,2Y1,0 ↔ A2,1 A4,1 A23,2 A1,3 A2,3 A4,3 A3,4 ,
(α1 + α2 + α3 + α4 ) ↔ Y3,1 ↔ A1,1 A2,1 A4,1 A23,2 A1,3 A2,3 A4,3 A3,4 .

It is obtained by replacing each root βk in a decomposition of α1 + α2 + α3 + α4 by the corre-

d
 −1
sponding variable Yik ,pk . The third column gives the monomial Y W ∏rk=1 Yiak ,p
k
k
.

We can now state the main result of this section.

∼
Theorem 9.11 (a) We have a Gd -equivariant isomorphism of varieties Ψ : M•0 (W d ) −→ Ed .

(b) M•0 (W d ) is an affine space of dimension ∑i→ j di d j .

(c) Lusztig’s perverse sheaves IC(Oa ) on Ed are the same as Nakajima’s perverse sheaves
ICW d (V ) on M•0 (W d ).

Proof — In Proposition 9.4 we have constructed a Gd -equivariant closed immersion Ψ of M•0 (W d )

• reg
into Ed . Since each stratum M0 (V,W d ) is Gd -invariant, Ψ maps every stratum to a union of
orbits Oa . Since Ψ is injective and the number of strata is equal to the number of orbits (Proposi-
tion 9.8), it follows that Ψ maps each stratum of M•0 (W d ) to a single Gd -orbit in Ed , so every orbit
is contained in the image of Ψ. Thus Ψ is surjective. Since a surjective closed immersion between
reduced schemes is an isomorphism, this proves (a). Claim (b) follows immediately from (a), and
claim (c) is again a consequence of Proposition 9.8, which shows that the stratifications used for
defining the perverse sheaves are the same. ✷

Remark 9.12 (a) By the proof of Proposition 9.8, for every dominant monomial m in Yt,Q there
d
is a unique pair (V,W d ) such that m = Y W AV . Hence, even if the varieties M•0 (W d ) involve
∼
very particular spaces W d , the isomorphisms Ψ : M•0 (W d ) −→ Ed are enough to identify all the
irreducible (q,t)-characters of CQ .
Thus Theorem 9.1, Theorem 9.2, and Theorem 9.11 provide a geometric explanation of part (b)
of Theorem 6.1. By comparing convolution diagrams in [Lu2] and [VV1], it should also be
possible to understand in a geometric manner part (a) of Theorem 6.1, that is, the multiplicative
structure (see [N4, §3.5]).
(b) If we take for Q a quiver of type A with all arrows in the same direction, then M•0 (W d ) is just a
space of graded nilpotent endomorphisms as in the Ginzburg-Vasserot construction [GV] of type
A quantum loop algebras (see e.g. [Le, §2.5.3]). So Theorem 9.11 becomes tautological in this
case.

45
Acknowledgements
We thank H. Nakajima for helpful comments and discussions, in particular about the proof of
Theorem 9.11.

References
[ARS] M. AUSLANDER , I. R EITEN , S. S MALO, Representation theory of Artin algebras, Cambridge studies in advanced
mathematics 36, Cambridge 1995.

[ASS] I. A SSEM , D. S IMSON , A. S KOWRO ŃSKI , Elements of the representation theory of associative algebras. Vol.1, London
Math. Soc. Student Texts 65, Cambridge 2006.

[BZ1] A. B ERENSTEIN , A. Z ELEVINSKY, String bases for quantum groups of type Ar , in I.M. Gelfand seminar, Adv. Soviet
Math. 16, pp. 51–89, AMS 1993.

[BZ2] A. B ERENSTEIN , A. Z ELEVINSKY, Quantum cluster algebras, Adv. Math. 195 (2005), 405–455.

[CH] V. C HARI , D. H ERNANDEZ, Beyond Kirillov-Reshetikhin modules, Quantum affine algebras, extended affine Lie alge-
bras, and their applications, Contemp. Math., 506 (2010), 49–81.

[CP] V. C HARI , A. P RESSLEY, Minimal affinizations of representations of quantum groups: the simply laced case, J. Algebra
184 (1996), 1–30.

[DFK] P. D I F RANCESCO , R. K EDEM, The solution of the quantum A1 T -system for arbitrary boundary, Comm. Math. Phys.
313 (2012), 329–350.

[D] V.G. D RINFELD, A new realization of Yangians and quantized affine algebras, Soviet Math. Doklady 36 (1988), 212–
216.

[FM] E. F RENKEL , E. M UKHIN, Combinatorics of q-characters of finite-dimensional representations of quantum affine alge-
bras, Comm. Math. Phys. 216 (2001), 23–57.

[FR1] E. F RENKEL , N. R ESHETIKHIN, Deformations of W -algebras associated to simple Lie algebras, Commun. Math. Phys.
197 (1998), 1–32.

[FR2] E. F RENKEL , N. R ESHETIKHIN, The q-characters of representations of quantum affine algebras, Recent developments
in quantum affine algebras and related topics, Contemp. Math. 248 (1999), 163–205.

[GR] P. G ABRIEL , A.V. ROITER, Representations of finite-dimensional algebras, Encyclopaedia of Mathematical Sciences
73, Springer 1997.

[GLS] C. G EISS , B. L ECLERC , J. S CHR ÖER, Cluster structures on quantum coordinate rings, Selecta Math. New Series, DOI
10.1007/s00029-012-0099-x, http://link.springer.om/article/10.1007/s00029-012-0099-x (July 2012), arXiv:1104.0531
(2011).

[GV] V. G INZBURG , E. VASSEROT, Langlands reciprocity for affine quantum groups of type An , Internat. Math. Res. Notices
3 (1993), 67–85.

[Ha] D. H APPEL, Triangulated categories in the representation theory of finite-dimensional algebras, London Math. Soc.
Lecture Notes 119, Cambridge 1988.

[He1] D. H ERNANDEZ, t-analogues des opérateurs d’écrantage associés aux q-caractères, I. M. R. N. 8 (2003), 451–475.

[He2] D. H ERNANDEZ, Algebraic approach to q,t-characters, Adv. Math. 187 (2004), 1–52.

[He3] D. H ERNANDEZ, The Kirillov-Reshetikhin conjecture and solutions of T -systems, J. Reine Angew. Math. 596 (2006),
63–87.

[He4] D. H ERNANDEZ, Simple tensor products, Invent. Math. 181 (2010), 649–675.

[HL] D. H ERNANDEZ , B. L ECLERC, Cluster algebras and quantum affine algebras, Duke Math. J. 154 (2010), 265–341.

[IIKNS] R. I NOUE , O. I YAMA , A. K UNIBA , T. N AKANISHI , J. S UZUKI , Periodicities of T -systems and Y -systems, Nagoya
Math. J.197 (2010), 59—174.

[K] M. K ASHIWARA, On crystal bases of the q-analogue of universal enveloping algebras, Duke Math. J. 63 (1991), 465–
516.

46
[KL] M. K HOVANOV, A. L AUDA, A diagrammatic approach to categorification of quantum groups I, Represent. Theory 13
(2009), 309–347.

[KNS] A. K UNIBA , T. N AKANISHI , J. S UZUKI , Functional relations in solvable lattice models. I. Functional relations and
representation theory. Internat. J. Modern Phys. A 9 (1994), 5215–5266,

[La] P. L AMPE, Quantum cluster algebras of type A and the dual canonical basis, arXiv:1101.0580, (2011), 46 p.

[Le] B. L ECLERC, Quantum loop algebras, quiver varieties, and cluster algebras, in Representations of algebras and related
topics, EMS Series of Congress Reports, Ed. A. Skowroński, K. Yamagata, 2011, 117-152.

[Lu1] G. L USZTIG, Canonical bases arising from quantized enveloping algebras, J. Amer. Math. Soc. 3 (1990), 447-498.

[Lu2] G. L USZTIG, Quivers, perverse sheaves, and quantized enveloping algebras, J. Amer. Math. Soc. 4 (1991), no. 2, 365–
421.

[N1] H. N AKAJIMA, Quiver varieties and finite-dimensional representations of quantum affine algebras, J. Amer. Math. Soc.
14 (2001), 145–238.

[N2] H. N AKAJIMA, t-analogs of q-characters of Kirillov-Reshetikhin modules of quantum affine algebras, Represent. Theory
7 (2003), 259–274.

[N3] H. N AKAJIMA, Quiver varieties and t-analogs of q-characters of quantum affine algebras, Annals of Math. 160 (2004),
1057–1097.

[N4] H. N AKAJIMA, Quiver varieties and cluster algebras, Kyoto J. Math. 51 (2011), 71–126.

[Q] F. Q IN, Quantum Cluster Variables via Serre Polynomials, J. Reine Angew. Math., 668 (2012), 149–190.

[Ri1] C.-M. R INGEL, Tame algebras and quadratic forms, Lecture Notes in Math. 1099, Springer 1984.

[Ri2] C.-M. R INGEL, Hall algebras and quantum groups, Invent. Math. 101 (1990), 583–592.

[Ri3] C.-M. R INGEL, Hall algebras revisited, in Israel Math. Conf. Proc. 7, (1993), 171–176.

[Ri4] C.-M. R INGEL, PBW-bases of quantum groups, J. Reine Angew. Math. 470 (1996), 51–88.

[Ro] R. ROUQUIER, 2-Kac-Moody algebras, arXiv:0812.5023.

[S] J. S HENG, AR quivers, exceptional sequences and algorithms in derived Hall algebras, Science China (2010), doi:
10.1007/s11425-010-0069-8.

[T] B. T O ËN, Derived Hall algebras, Duke Math. J. 135 (2006), 587–615.

[VV1] M. VARAGNOLO , E. VASSEROT, Perverse sheaves and quantum Grothendieck rings, in Studies in memory of Issai
Schur, Prog. Math. 210, Birkhäuser, 2002, 345–365.

[VV2] M. VARAGNOLO , E. VASSEROT, Canonical bases and KLR algebras, J. Reine Angew. Math. 2011, 659, 67–100.

[XX] J. X IAO , F. X U, Hall algebras associated to triangulated categories, Duke Math. J. (2008) 143, 357–373.

David H ERNANDEZ : Université Paris Diderot-Paris 7,

Institut de Mathématiques de Jussieu - Paris Rive Gauche CNRS UMR 7586,
Bâtiment Sophie Germain, Case 7012,
75205 Paris Cedex 13, France.
email : [email protected]

Bernard L ECLERC : Normandie Univ, France

UNICAEN, LMNO F-14032 Caen, France
CNRS UMR 6139, F-14032 Caen, France
Institut Universitaire de France,
email : [email protected]

47