Journal of Algebra 277 (2004) 318–348

www.elsevier.com/locate/jalgebra

Classification of irreducible integrable modules

for toroidal Lie algebras with finite dimensional
weight spaces
S. Eswara Rao
School of Mathematics, Tata Institute of Fundamental Research, Homi Bhabha Road, Mumbai 400 005, India
Received 3 May 2003

Communicated by Efim Zelmanov

Dedicated to Professor P. Somaraju with admiration and gratitude

The study of Maps (X, G), the group of polynomial maps of a complex algebraic variety
X into a complex algebraic group G, and its representations is only well developed in
the case that X is a complex torus C∗ . In this case Maps (X, G) is a loop group and
the corresponding Lie algebra Maps (X, G̊) is the loop algebra C[t, t −1 ] ⊗ G̊. Here the
representation comes to life only after one replaces Maps (X, G̊) by its universal central
extension, the corresponding affine Lie algebra. One then obtains the well known theory of
highest weight modules, vertex representations, modular forms and character theory and
so on.
The next easiest case is presumably the case of n dimensional torus (C∗ )n . So we
consider the universal central extension of G̊ ⊗ C[t1±1 , . . . , tn±1 ] which is referred to as
the toroidal Lie algebra τ in [11] and [16]. The most interesting modules are the integrable
modules (where the real root space acts locally nilpotently (see Section 2)), as they lift
to the corresponding group. Unlike the affine case where the central extension is one
dimensional, the toroidal case has infinite dimensional centre which makes the theory more
complicated. For the first time a large number of integrable (reducible) modules for toroidal
Lie algebras (simply laced case) have been constructed through the use of vertex operators
in [11] and [16].
In this paper we construct two classes of (Examples 4.1 and 4.2) of irreducible integrable
modules for toroidal Lie algebras with finite dimensional weight spaces. In Sections 4 and
5 we prove that any irreducible integrable module where part of the center acts non-trivially
are the ones given in Example 4.2 up to an automorphism of τ . We have proved in [8] that

E-mail address: [email protected].

0021-8693/$ – see front matter  2004 Elsevier Inc. All rights reserved.
doi:10.1016/j.jalgebra.2004.03.016
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 319

the only modules with the above property where center acts trivially are the one given in
Example 4.1 (see Remark 5.5). In this case a similar classification is obtained in [18] with
a stronger assumption on the weight spaces. These results in the case n = 1 are due to [5]
and [6].
In Section 1, after recalling the construction of non-twisted affine Lie algebras from
[15], we establish the necessary terminology for root systems, non-degenerate bilinear
form and the Weyl group for toroidal Lie algebras. In Section 2 we recall the actual
definition of toroidal Lie algebras and integrable modules and prove some standard facts
about the Weyl group and weight systems (Lemma 2.3). Then we prove that an irreducible
integrable module for τ with finite dimensional weight spaces has a highest weight
vector in the following sense. Let Gaf = G̊ ⊗ C[t1 , t1−1 ] ⊕ CC1 ⊕ Cd1 be an affine Lie
algebra and let Gaf = N − ⊕ h1 ⊕ N + be the standard decomposition. Then we prove
that if some zero degree central operator acts non-trivially there exists a vector killed by
N + ⊗ An−1 (Propositions 2.4 and 4.8 after twisting the module up to an automorphism).
Let G̊ = n− ⊕ h◦ ⊕ n+ and suppose all zero degree central operators act trivially then we
prove that there is a vector killed by n+ ⊗ An (Proposition 2.12). These two results may be
seen as generalization of Theorem 2.4 of [5].
In Section 3 we define graded and non-graded highest weight modules in the generality
of loop Kac–Moody Lie algebras and prove (Proposition 3.5) that there is a one–one
correspondence between the graded and non-graded cases. The problem now reduces to the
classification of irreducible integrable highest weight modules (non-graded) (Lemma 3.6).
We prove (Remark 3.9) that any such module is actually a module for Gaf ⊗ An−1 /I
for a co-finite ideal I (we are in the case where some zero degree central elements act
non-trivially). In a combinatorial Lemma 3.11
which is of independent interest we prove
that such a Lie algebra is isomorphic to Gaf (direct sum of finitely many copies of
Gaf ).
Then it is very standard to classify irreducible integrable highest weight modules
for Gaf .
If the zero degree center acts trivially then the full center should act trivially
(Proposition 4.13) where we use an interesting result (Proposition 4.12) on Heisenberg
Lie algebras due to Futorny [13]. In this case the classification is given in [8] (see
Remark 5.5).
We prove in Lemma 4.6 in the generality of irreducible modules for τ with finite
dimensional weight spaces that most of the center acts trivially. In fact in each graded
component of the center at most one dimensional space acts non-trivially. See [10] for
some interesting results on general irreducible modules for τ . See [3] for some highest
weight module theory for toroidal Lie algebras with finite center.
In other papers [2] and [9] a more general toroidal Lie algebra is considered by
adding an infinite set of derivations and the center becomes finite dimensional. They have
constructed integrable irreducible (highest weight) modules for such a toroidal Lie algebra.
It remains to be seen which toroidal Lie algebra admits an interesting representation theory
in general. More general Lie algebras called extended affine Lie algebras (EALA), of which
toroidal Lie algebras are prime examples, are studied extensively. See for example [4] and
the references therein.
The classification problem for the twisted toroidal is under progress. The case where
the center is zero is done by Punita Batra [1].
320 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

1. Section

We first recall the some notation of the theory of affine Lie algebras from Chapter 6 of
[15]. We always denote N the non-negative integers, Z the integers and Z+ the positive
integers. All our vector spaces are over the complex numbers C.
Let G̊ be a simple finite dimensional Lie algebra. Let h̊ be a Cartan subalgebra of
finite dimension d. Let A = (aij ) be the Cartan matrix of the corresponding non-twisted
affine Lie algebra. Let a0 , a1 , . . . , ad be the numerical labels of the Dynkin diagram S(A).
Let a0∨ , . . . , ad∨ be the numerical labels of the Dynkin diagram S(AT ). Let bi−1 = ai∨ .
We know from [15] that for a non-twisted affine Lie algebra that a0 = b0 = 1. Write
A = Diag(a0 b0 , a1 b1 , . . . , ad bd )B where B is a symmetric matrix. Let h be the Cartan
subalgebra spanned by α0∨ , α1∨ , . . . , αd∨ , d0 such that h̊ is spanned by α1∨ , . . . , αd∨ . Define
αi ∈ h∗ by αi (αj∨ ) = aj i , αi (d0 ) = 0 for 1
i
d and α0 (d0 ) = 1. Define w in h∗ such
that w(αi∨ ) = δi0 and w(d0 ) = 0. Then h∗ is spanned by α0 , α1 , . . . , αd , w for dimension
reasons. Define nondegenerate symmetric bilinear form on h by (αi∨ , αj∨ ) = aj bj aij ,
(αi∨ , d0 ) = δi0 and (d0 , d0 ) = 0. Similarly h∗ carries a bilinear form such that (αi , αj ) =
bi−1 ai−1 aij , (αi , w) = δi0 and (w, w) = 0. As in [15] we normalize the form on h∗ by
(β, β) = 2 where β is the maximal root of G̊. Define


d 
d
C= bi−1 αi∨ + α0∨ , δ= ai αi + α0 .
i=1 i=1

Then δ − α0 = β. Let β ∨ = C − α0∨ . From [15] we have the following:

 
(δ, αi ) = 0, (δ, δ) = 0, (δ, w) = 1, C, αi∨ = 0, (C, C) = 0,
 
(C, d0 ) = 1, δ αi∨ = 0, αi (C) = 0 and w(C) = 1.

1.1. Lemma. For λ ∈ h∗ ,

(1) (λ, αi ) = bi−1 ai−1 λ(αi∨ ), 0
i
d,

(2) (λ, β) = λ(β ∨ ).

Proof. It is easy to see. 2

Define the ith fundamental reflection ri on h∗ such that ri λ = λ − λ(αi∨ )αi , 0
i
d.

Let Waf be the Weyl group generated by ri . Then ( , ) on h∗ is Waf -invariant.
We need the following simple lemma. Let Λ̊ be theroot lattice of G̊ spanned by
α1 , α2 , . . . , αd . For λ, µ ∈ h̊∗ define λ
µ if µ − λ = di=1 ni αi for ni ∈ N. Let Λ̊+
˚
be the set of weights in h̊∗ such that λ(αi∨ ) ∈ N for 1
i
d. A weight λ ∈ Λ̊+ is called
miniscule if µ
λ, µ ∈ Λ̊+ implies µ = λ.
˚
1.2. Lemma. Let λ ∈ h̊∗ be miniscule then λ(β ∨ ) = 0 or 1.
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 321

Proof. Follows from Exercise 13 of Section 13 of [14]. Just note that λ is dominant
integral. 2

We will now generalize this for toroidal algebras [11,16]. Fix a positive integer n. Let
h be the 2n + d dimensional vector space spanned by α1∨ , α2∨ , . . . , αd+n ∨ , d , . . . , d . Let
1 n
à = (Aij ) be a matrix of order n + d such that removal of n − 1 rows and the corresponding
columns in the last n rows and n columns should give A. We will describe the matrix
explicitly. Recall that A = (ai,j )0
i,j
d is the affine matrix. Then Ad+i,j = a0,j , 1

i
n, 1
j
d; Aj,d+i = aj,0 , 1
i
n, 1
j
d; Ad+i,d+j = 2, 1
i, j
n. Let
 = diag(a1 b1 , a2 b2 , . . . , ad bd , 1, . . . , 1) be a matrix of order d + n. Then à = D
D B  where
 is symmetric matrix.
B
Define αi ∈ h∗ such that αi (αj∨ ) = aj i for 1
i, j
d + n, αi (dj ) = 0 for 1
i
d,
1
j
n, αd+i (dj ) = δij for 1
i, j
n. Define wi ∈ h∗ (1
i
n) by wi (αj∨ ) = 0
for 1
j
d and wi (αj∨+d ) = δij for 1
j
n, wj (di ) = 0. Then we will see
that α1 , . . . , αd+n , w1 , . . . , wn is a basis of h∗ . Let ad+i = bd+j = 1 for 1
i, j
n.
Define a symmetric bilinear form h∗ by (αi , αj ) = bi−1 ai−1 aij , (αi , wj ) = 0, 1
i
d,
(αd+i , wj ) = δij for 1
i, j
n, (wi , wj ) = 0.
Define for 1
j
n,


d
δj = ai αi + αd+j = β + αj +d ,
i=1


d
Cj = bi−1 αi∨ + αd+j
∨
= β ∨ + αj∨+d .
i=1

As in the affine case we have the following:

 ∨   
(αi , δj ) = 0, (δi , δj ) = 0, αi , Cj = 0, (Ci , Cj ) = 0, δi αj∨ = 0,
δj (Ci ) = 0, αi (Cj ) = 0, δj (di ) = αd+j (di ) = δij and wi (Cj ) = δij .

For m = (m1 , . . . , mn ) ∈ Zn define δm = mi δi and note that (δm , δm ) = 0. δm are
called null roots.

1.3. Root and co-roots. Let ∆˚ be the finite root system. Let ∆ = {α + δ | α ∈ ∆˚ ∪ {0},
δ null root}. γ ∈ ∆ is called a real root if (γ , γ ) = 0. For α ∈ ∆˚ define a co-root

 |αi |2 ∨ 
d
α∨ = mi α for α = mi αi , α ∈ ∆.
˚
|α|2 i
i=1


˚ define γ ∨ = α ∨ +
For γ = α + δm , α ∈ ∆, 2
mj Cj .
|α|2

1.4. Lemma. For λ ∈ h∗ ,

322 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

(1) (λ, αi ) = bi−1 ai−1 λ(αi∨ ), 0
i
d + n,

|α|2 ∨
(2) (λ, α) = 2 λ(α ), α ∈ ∆re .

1.5. Weyl group [17]. For a real root γ define a reflection rγ on h∗ by

 
rγ (λ) = λ − λ γ ∨ γ .

From above one can see that it is a reflection and ( , ) is W -invariant where W is the Weyl
group generated by rγ , γ real.

Now the action of the linear functionals α1 , . . . , αd , δ1 , . . . , δn , w1 , w2 , . . . , wn on the

basis α1∨ , . . . , αd∨ , C1 , . . . , Cn , d1 , . . . , dn is given by the matrix
 
Å 0 0
0 0 I
0 I 0

which is invertible. Here Å is the finite Cartan matrix. In particular the linear functions are
a basis of h∗ .

2. Section [11,16]

Let G̊ be finite dimensional simple Lie algebra. Let n  2 be a positive integer. Let
A = An = C[t1±1 , . . . , tn±1 ] be a Laurent polynomials in n commuting variables. For
m = (m1 , . . . , mn ) ∈ Zn let t m = t1m1 · · · tnmn . For any vector space V let VA = V ⊗ A and
let v(m) = v ⊗ t m for v ∈ V . Let Z = ΩA /dA be the module of differentials. That is Z is
spanned by vectors t m Ki , 1
i
n, m ∈ Zn , together with the relations

(2.1) mi t m Ki = 0.

Let τ = G̊A ⊕ Z ⊕ D where D is a linear span of d1 , . . . , dn . We will now define a Lie

structure on τ called toroidal Lie algebra.
 r+s
(2.2) (1) [X(r), Y (s)] = [X, Y ](r + s) + (X, Y )d(t r )t s where d(t r )t s = ri t Ki
and ( , ) is non-degenerate G̊-invariant symmetric bilinear form on G̊ whose
restriction to h̊ is given in earlier section.
(2) Z is central in G̊A ⊕ Z.
(3) [di , X(r)] = ri X(r).
(4) [di , t m Kj ] = mi t m Kj .

Let h be the span of h̊ and K1 , . . . , Kn , d1 , . . . , dn . Clearly h can be identified with h

defined in Section 1 with Ki = Ci for 1
i
n. Let G̊ = α∈∆∪{0} ˚ G̊α be the root space
decomposition. Define
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 323

τα+δm = G̊α ⊗ t m , α ∈ ∆,
˚


n
τδm = h̊ ⊗ t m ⊕ Ct m Ki , m = 0,
i=1
τ0 = h.


Then clearly τ = α∈∆ τα is the root space decomposition with respect to h and consistent
with the root system ∆ defined in Section 1. For α ∈ ∆˚ let Xα be the root vector of root α.

Definition. A module V of τ is called integrable if

(1) (weight module) V = λ∈h∗ Vλ where

Vλ = v ∈ V | hv = λ(h)v, ∀h ∈ h ,

(2) for all α ∈ ∆,

˚ m ∈ Zn and v ∈ V there exists an integer k = k(α, m, v) such that
(Xα (m)) v = 0.
k

For a weight module V let P (V ) denote the set of all weights. The following is very
standard.

2.3. Lemma. Let V be irreducible integrable module for τ with finite dimensional weight
spaces.

(1) P (V ) is W -invariant.
(2) dim Vλ = dim Vwλ for w ∈ W , λ ∈ P (V ).
(3) α real in ∆, λ ∈ P (V ) then λ(α ∨ ) ∈ Z.
(4) If α is real, λ ∈ P (V ) and λ(α ∨ ) > 0 then λ − α ∈ P (V ).
(5) λ(Ci ) is a constant (integer) ∀λ ∈ P (V ).
  |α |2
Proof. Let α ∈ ∆˚ and let α = dj=1 cj αj . Define tα = cj 2j αj∨ so that α ∨ = 2tα /|α|2 .
One can easily check that (tα , tα ) = (α, α) and that α(h) = (tα , h) where tα is the unique
with that property because the form ( , ) on h̊ is nondegenerate. Fix Xα ∈ G̊α and choose
Yα ∈ G̊−α such that (Xα , Yα ) = (tα 2,tα ) .
1

Claim. [Xα , Yα ] = α ∨ . Consider (h, [Xα , Yα ]) = α(h)(Xα , Yα ) = α(h) (tα2,tα ) . By the

uniqueness of tα the claim follows.

Now [α ∨ , Xα ] = α(α ∨ )Xα = 2Xα , [α ∨ , Yα ] = −2Yα . Thus X ∨

α , Yα , α is an sl2 copy.
∨ ∨
Consider γ = α + δm and recall the definition of γ = α + |α|2 mj Cj . Consider
2

   2 
Xα ⊗ t m , Yα ⊗ t −m = α ∨ + (Xα , Yα ) mi Ci = α ∨ + 2 mi Ci = γ ∨ ,
|α|
324 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

 ∨ 
γ , Xα ⊗ t m = 2Xα ⊗ t m ,
 ∨ 
γ , Yα ⊗ t −m = −2Yα ⊗ t −m .

Thus Xα ⊗ t m , Yα ⊗ t −m , γ ∨ is a copy of sl2 .

Now (2), (3) and (4) follows from standard sl2 -representation theory. Since Ci is an
integral linear combination of αj∨ ’s it follows that λ(Ci ) is an integer. The fact that λ(Ci )
is constant follows from Lemma 4.3(2). 2

Consider the affine Lie algebra Gaf = G̊ ⊗ C[t1 , t1−1 ] ⊕ CC1 ⊕ Cd1 . Let Gaf = N − ⊕
h11 ⊕ N + be the standard decomposition into positive root spaces, negative root spaces and
a Cartan h11 spanned by α1∨ , . . . , αd∨ , αd+1
∨ ,d .
1

2.4. Proposition. Let V be irreducible integrable module for τ . Assume that C1 acts by a
non-zero scalar K1 and Ci , 2
i
n, act trivially. Then there exists a weight vector v in
V such that

N + ⊗ An−1 v = 0 or N − ⊗ An−1 v = 0

where An−1 = C[t2±1 , . . . , tn±1 ] and

N + ⊗ An−1 = Gα ⊗ t1m1 t m , m = (m2 , . . . , mn ) ∈ Zn−1 .
m1 0 if α∈∆˚ + , and
m1 >0 if α∈∆˚ − ∪{0}

We need some notation and a few lemmas for this. Let λ ∈ P (V ) then λ can be uniquely
written as
 
λ=λ+ gi δi + si wi

d
where λ = i=1 mi αi ∈ h̊∗ . Now for 2
i
n,

0 = λ(Ci ) = si wi (Ci ) = si ,
K1 = λ(C1 ) = s1 w1 (C1 ) = s1 .

(2.5) Thus λ = λ + gi δi + K1 w1 .

Let Γ0+ be non-negative integral linear combinations of α1 , . . . , αd .

2.6. Lemma. Let V be a module satisfying the conditions of the above proposition. Given
/ P (V ) for all η ∈ Γ0+ \{0} and
a fixed λ1 ∈ P (V ) there exists a λ ∈ P (V ) such that λ + η ∈
(λ + η)(di ) = λ(di ) = λ1 (di ) for all 1
i
n.
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 325

Proof. Let λ1 ∈ P (V ), let λ1 (di ) = gi . Let g = (g1 , . . . , gn ) and let

Vg = {v ∈ V : di v = gi v}.

Let Pg (V ) = {λ ∈ P (V ): Vλ ⊆ Vg }. Then from (2.5) for λ ∈ Pg (V ) we have


λ=λ+ gi δi + K1 w1 .

Furthermore λ | h̊ = λ. So if λ, µ ∈ Pg (V ) such that λ = µ then λ = µ. Now Vg is an

integrable G̊-module with finite dimensional weight spaces (with respect to h̊). Hence
 
Vg = V λ
λ∈h̊∗

 is an irreducible finite dimensional module for G̊. For λ, µ ∈ Pg (V ) we have

where V (λ)
λ − µ = di=1 mi αi . Since V is irreducible it follows that mi ’s are integers. So that λ’s
that occur in Pg (V ) determine a unique coset in the weight lattice of G̊ modulo its root
lattice. Let λ0 ∈ h̊∗ be the unique miniscule weight (see Exercise 13 of Chapter III of
[14]). Then by Lemma B, 13.4 of [14] it follows then λ0 is a weight of V (λ) that occur
in Vg . Since each weight space is finite dimensional the number of V (λ) that occurs in
Vg has to be finite. Thus Pg (V ) is finite. Let λ be maximal with respect to the ordering

. Then λ + η ∈ / P (V ) for any η ∈ Γ+0 \{0}. Just note that if λ + η ∈ P (V ) then λ + η ∈
˚
Pg (V ). Furthermore (λ + η)(di ) = gi = λ(di ). 2

(2.7) Now we will define a different ordering
on h∗ by λ
µ if µ − λ =

d+1 d+1
i=1 ni αi , ni ∈ N. Thus if we say λ  0 then λ = i=1 ni αi , ni ∈ N.

2.8. Lemma. Let V be as in the above proposition with the additional assumption that
K1 > 0. Suppose for all λ ∈ P (V ) there exists 0
η = 0 such that λ + η ∈ P (V ). Then
there exists infinitely many λi ∈ P (V ), i ∈ Z+ , such that

(1) There exists a weight vector ui of weight λi such that N + ui = 0.

(2) λi (dj ) = λk (dj ), ∀i, k ∈ Z+ and 2
j
n.
(3) λi (d1 ) − λi−1 (d1 ) ∈ Z+ for i  2.
(4) There exists a common weight in V (λi ) ∀i where V (λi ) is the irreducible highest
weight module generated by ui .

Proof. Let λ be an in Lemma 2.6. Note that λ(αi∨ ) ∈ N for i = 1, 2, . . . , d which follows
from Lemma 2.3(3). Recall that ∆˚ is finite root system. Let ∆+a re = {α + mδ1 , α ∈ ∆, m 
˚
− +a ∨
0 and α ∈ ∆˚ , m > 0}. Let ∆(λ) = {γ ∈ ∆re : λ(γ )
0}. Since λ(C1 ) = K1 > 0 it is
easy to see that ∆(λ) is finite. This situation is very similar to the proof of Theorem 2.4(i)
of [5]. (Here we need our Lemma 2.6 as the arguments of Lemma 2.6 of [5] are not correct.
The 6th line from above on page 322 does not follows from earlier argument.) Now from
326 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

Claims 1, 2 and 3 in the proof of 2.4(i) of [5] we have a vector v in Vλ+p1 δ1 , p1  0, such
that

τrδ1 v = 0, r > 0 and τα+sδ1 v = 0

for all but finitely many roots α + sδ1 in ∆+a +

re . Since V is integrable it follows that U (N )v
is finite dimensional. Choose λ1 maximal in the ordering
among the weights of U (N + )v.
Then there exists a vector u1 of weight λ1 such that

N + u1 = 0.

Furthermore λ1 = λ + p1 δ1 + η1 , η1  0, and λ1 (dj ) = λ(dj ) for 2
j
n and

λ1 (d1 ) − λ(d1 ) = p1 + η1 (d1 )  0.

Now by Lemma 2.6 there exists λ̃1 ∈ P (V ) such that

(2.9) Vλ̃1 +η = 0, ∀η ∈ Γ0+ \{0} and λ̃1 (di ) = λ1 (di ), 1
i
n. By the assumption in the
lemma there exists η̃1 > 0 such that

λ̃1 + η̃1 ∈ P (V ).

By (2.9) it follows that η̃1 (d1 ) > 0 as η̃1  0 and we also have η̃1 (di ) = 0 for 2
i
n.
Repeating the above argument for λ̃1 + η̃1 in place of λ to get a non-zero weight vector u2
of weight λ2 = λ̃1 + η̃1 + p2 δ1 + η2 , η2  0, such that N + u2 = 0. Now λ2 (dj ) = λ̃1 (dj ) =
λ(dj ) for 2
j
n.

λ2 (d1 ) = λ1 (d1 ) + p2 + η2 (d1 ) + η̃1 (d1 )

so that λ2 (d1 ) − λ1 (d1 ) = p2 + η2 (d1 ) + η̃1 (d1 ) > 0. (Note the strict inequality.) Repeating
this process we have (1), (2) and (3). Clearly λi (αj∨ ) ∈ N, 1
j
1 + d. By irreducibility
of V we have


d
λi − λj = mi αi for mi ∈ Z.
i=1

Then the λ i s determine a unique coset of the weight lattice. As earlier let λ0 ∈ h̊∗ be the
unique miniscule weight. Thus λ0
λi , ∀i ∈ Z. (Note that each λi is a dominant integral
weight.) ˚
d
Let λ0 = λ0 + i=1 λ(di )δi + K1 w1
λi . This inequality is by construction of λi .

Recall λi = λi + λi (d1 )δ1 + di=2 λ(di )δi + K1 w1 and we have λi (di ) − λ(d1 ) > 0. Now
λ0 (αi∨ ) = λ0 (αi∨ ) ∈ N for i = 1, 2, . . . , d.
 ∨     
λ0 αd+1 = λ0 C1 − β ∨ = K1 − λ0 β ∨ .
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 327

Since K1 is integer and positive we have K1  1. By Lemma 1.2 we have λ0 (β ∨ ) = 0 or 1.

∨ ) ∈ N. Hence λ is dominant integral and λ
λ . By Proposition 12.5(a) of
Thus λ0 (αd+1 0 0 i
[15] it follows that λ0 is a weight of V (λi ) for all i. This proves (4). 2

Proof of Proposition 2.4. Assume K1 > 0. Suppose the conclusions of Lemma 2.8 are
true. Since the λi ’s are distinct, the V (λi ) are all non-isomorphic irreducible highest weight
modules. Hence their sum has to be direct. But Vλ0 ∩ V (λi ) = 0. Hence dim Vλ0 is infinite.
A contradiction.
So we conclude that there exists a λ ∈ P (V ) such that

(2.10) λ + η ∈ / P (V ) for α ∈ ∆+a

/ P (V ), for all η > 0. In particular λ + α ∈ re .

Thus from Lemma 2.3(4) it follows that λ(α ∨ ) ∈ N:

n
(2.11) Suppose Vλ+α+δ 1 = 0 for all α ∈ ∆+a re and for all δ =
1
i=2 mi δi . This means
Xα ⊗ t m Vλ = 0. But the positive real roots generates the positive null roots and hence
N + ⊗ An−1 V +a
λ = 0. So we are done. Now assume that Vλ+α+δ = 0 for some α ∈ ∆re and
n
for some δ = i=2 ni δi . Let µ = λ + α + δ.
n
Claim. Vµ+γ +δ 1 = 0 for γ ∈ ∆+a re and for all δ =
1
i=2 mi δi . Suppose it is false. Then
Vµ+γ +δ 1 = 0 for some γ and δ .
1

Case 1. (α + γ , α) > 0. We know by Lemma 1.4(2) that (α + γ )(α ∨ ) > 0. Put γ1 =

α + δ + δ 1 and consider
    
µ + γ + δ 1 γ1∨ = (λ + α + γ ) α ∨ > 0.

We are using the fact that λ(Ci ) = 0 for 2
i
n. By Lemma 2.3(4), λ + α + γ + δ +

δ 1 − (α + δ + δ 1 ) = λ + γ ∈ P (V ) a contradiction to (2.10).

Case 2. (α + γ , γ ) > 0 which can be done in the same way.

Since the symmetric bilinear form given by an affine matrix is semi positive definite on
the root lattice, we are left with

Case 3. (α + γ , α + γ ) = 0. This implies α + γ = δ1 for  > 0 and further (α + γ , α) = 0

and (α + γ , γ ) = 0. Then (λ + α + γ + δ + δ 1 )(γ1∨ ) = λ(α ∨ ) where γ1 = α + δ + δ 1 .

Subcase 1. λ(α ∨ ) > 0 (in any case λ(α ∨ )  0). By Lemma 2.3(4) we have
 
λ + α + γ + δ + δ 1 − α + δ + δ 1 = λ + γ ∈ P (V )

which is a contradiction to (2.10).

328 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

Subcase 2. λ(α ∨ ) = 0. Note that

 ∨  2
λ −α + δ1 + δ + δ 1 =  2 λ(C1 ) > 0.
|α|
Hence by Lemma 2.3(4) we have
 
λ + α + γ + δ + δ 1 − γ + δ + δ 1 = λ + α ∈ P (V )

a contradiction to (2.10).

Thus our claim follows. As earlier it follows that N + ⊗ An−1 Vµ = 0. Hence we are
done. The case K1 < 0 can be done similarly. 2

2.12. Proposition. Let V be integrable irreducible module for τ with finite dimensional
weight spaces. Suppose Ki = 0 for 1
i
n. Then there exists a weight vector v in V
such that

n+ ⊗ An v = 0 or n− ⊗ An v = 0

where G̊ = n+ ⊕ h̊ ⊕ n− is the standard decomposition.

Proof. This follows from the proof of Theorem 2.4(ii) of [5]. Use Lemma 2.6 of our paper
instead of Lemma 2.6 of [5]. The fact that the dimension of the space of null roots could
be greater than one does not matter. 2

3. Section

Let G be a Kac–Moody Lie algebra. Let h be a Cartan subalgebra of G. Let G  =

[G, G] and let h = h ∩ G  . Let G  = N + ⊕ h ⊕ N − be the standard decomposition into
positive roots spaces, negative roots spaces and a h . Fix a positive integer n and let
A = An = C[t1±1 , . . . , tn±1 ]. Let D be a linear span of derivations d1 , . . . , dn . Then let
A = G  ⊕ h ⊕ D where h is define as h = h ⊕ h (see §1.3 of [15]). Define Lie
G A
A by
algebra structure on G
 
X(r), Y (s) = [X, Y ](r + s),
 
di , X(r) = ri X(r), X, Y ∈ G, r, s ∈ Zn ,
 
b, X(r) = [b, X](r),
 
b, di = [di , dj ] = 0, b ∈ h.

Let h̃A = h ⊗ A ⊕ D ⊕ h be an abelian Lie algebra. For any Lie algebra G, let U (G)
be the universal enveloping algebra. Let H = h ⊗ A ⊕ h . Then U (H ) is clearly Zn -
graded abelian Lie algebra. Let ψ : U (H ) → A be a Zn -graded homomorphism. Then A is
a module for H via ψ defined as
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 329

 
b(m)t s = ψ b(m) t s , b ∈ h , m ∈ Zn ,
bt s = ψ(b)t s for b ∈ h .

Let Aψ = Image of ψ .

3.1. Lemma (Lemma 1.2, [7]). Aψ is an irreducible h̃A -module if and only if each non-zero
homogeneous element of Aψ is invertible.

Just note that h does not play any role in Lemma 3.1.
We need the notion of a highest weight module for G A . Let ψ be as above.

3.2. Definition. A module V of G A is called a highest weight module for G

A if there exist
a weight vector (with respect to h ⊕ D) v in V such that

(1) V = U (G A )v,
(2) NA+ v = 0,
(3) U (h̃A )v is an irreducible module for h̃A defined by use of ψ.

A . Let ψ be as above such

We will now define a universal highest weight module for G
+
that Aψ is irreducible. Let NA act trivially on Aψ . Now consider the following induced
A -module:
G
   
M ψ =U G A ⊗B A +
ψ where B = NA ⊕ h̃A .

We have the following standard.

3.3. Proposition.

(1) As h ⊕ D module, M(ψ) is a weight module.

(2) M(ψ) is a free NA− module and as a vector space
   
M ψ ∼= U NA− ⊗ Aψ .

(3) M(ψ) has a unique irreducible quotient called V (ψ).

Proof. See [8]. 2

We need the following non-graded highest weight module for GA ⊕ h .

3.4. Definition. A module W of GA ⊕h is said to be a (non-graded) highest weight module

if there exists a weight vector (with respect to h) v in W such that

(1) U (GA ⊕ h ) v = W .
(2) NA+ v = 0.
(3) There exists a ψ in H ∗ such that hv = ψ(h)v for all h ∈ H .
330 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

Let ψ ∈ H ∗ and let H act on the one dimensional vector space C(ψ) by ψ. Let NA+ act
trivially on C(ψ). Consider the induced module M(ψ) = U (GA ⊕ h ) ⊗B  C(ψ) where
B  = NA+ ⊕ H .
Now by standard arguments M(ψ) has a unique irreducible quotient V (ψ).
Let ψ be as above and let ψ = E(1) ◦ ψ where E(1) : Aψ → C defined by E(1)t m = 1.
We will make V (ψ)A into a (graded) G A -module by g(m)v(r) = (g(m)v)(m + r) for

g ∈ GA , r, m ∈ Z , v ∈ V (ψ) and
n

di v(r) = ri v(r),
 
b v(r) = b v (r), b ∈ h .

3.5. Proposition. Let ψ and ψ as above. Assume that Aψ is an irreducible h̃A -module. Let
G ⊂ Zn be such that {t m , m ∈ G} is a set of coset representatives for A/Aψ . Let v be a
highest weight vector of V (ψ). Then


(1) V (ψ)A = m∈G U v(m) as a G A module. U v(m) is a G
A -module generated by v(m).

(2) Each U v(m) is an irreducible GA -module.
(3) U v(0) ∼ A -module.
= V (ψ) as a G

Proof. Follows from Theorem 1.8 of [7]. It is stated for special ψ . But we have only used
the fact Aψ is an irreducible h̃A -module. More over our GA is smaller than the one in [7]
A . 2
but all the arguments take place inside G

3.6. Lemma (Lemma 1.10, [8]). Let ψ and ψ as above. Then V (ψ) has finite dimensional
weight spaces (with respect to h ⊕ D) if and only if V (ψ) has finite dimensional weight
spaces (with respect to h).

The following lemma tells us for which ψ, V (ψ) has finite dimensional weight spaces
with respect to h. There by giving conditions for V (ψ) to have finite dimensional weight
spaces.

3.7. Lemma. V (ψ) has finite dimensional weight spaces if and only if ψ factors through
h ⊗ A/I for some co-finite ideal I of A.

Proof. Assume that V (ψ) has finite dimensional weight spaces. Let ∆+ be a positive root
system for G. Let α be a simple root in ∆+ . Let Yα be a root vector for the root −α. For
fixed j, consider, for a highest weight vector v,

Yα ⊗ tjn v, n ∈ Z

 weight space Vλ−α (ψ), λ = ψ | h. Thus there exists a

which is contained in the same
non-zero polynomial Pj (α) = ai tji such that

Yα ⊗ Pj (α)v = 0,
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 331


where Yα ⊗ Pj (α) = ai Yα ⊗ tji . Let (P ) be the ideal generated by polynomial P
inside A.

Claim 1. Yα ⊗ (Pj (α))v = 0. Consider 0 = h(m)Yα ⊗ Pj (α)v = Yα ⊗ Pj (α)h(m)v −

α(h)Yα ⊗ t m Pj (α)v. Since h(m) acts by scalar on v, the first term is zero. Hence the
claim.

Put Pj = Pj (α) where α runs through all simple positive roots.

Claim 2. Yβ ⊗ (Pj )v = 0 for β ∈ ∆+ . First note that (Pj ) ⊂ (Pj (α)) and hence the claim
is true for any simple root. This reduces the proof to a proof by induction on the height β.
The claim is true for β such that height β = 1. Let α be a simple root in ∆+ . Let Xα be the
root vector of root α. Consider

Xα (m)Yβ ⊗ (Pj )v = Yβ ⊗ (Pj )Xα (m)v + [Xα , Yβ ] ⊗ t m (Pj )v.

The first term is zero as v is a highest weight vector. The second term is zero by induction.
Since Yβ ⊗ (Pj )v is killed by Xα (m) for α simple and for any m ∈ Zn it is easy to see
that Xα (m).Yβ ⊗ (Pj )v = 0 for all α ∈ ∆+ and m ∈ Zn . Hence Yβ ⊗ (Pj )v is a highest
weight vector of weight λ − β. But V (ψ) is an irreducible highest weight module and
hence Yβ ⊗ (Pj )v = 0. Hence Claim 2.

Claim 3. b ⊗ (Pj )v = 0, ∀b ∈ h . Consider hα ⊗ (Pj )v = Xα Yα ⊗ (Pj )v − Yα ⊗

(Pj )Xα v = 0. Since hα covers all h for α simple we have Claim 3.

Let I be an ideal generated by P1 , P2 , . . . , Pn inside A. It is elementary to see that

A/I is finite dimensional. Thus ψ factors through h ⊗ A/I . In fact we can prove that
G  ⊗ I.V (ψ) = 0 by considering W = {w ∈ V (ψ), G  ⊗ I w = 0}. We have just seen that
W is nonempty. It is easy to see that W is a sub module. Hence W = V (ψ).
Conversely suppose that ψ factors through h ⊗ A/I for a co-finite ideal I of A.

Claim 4. For β > 0, Yβ ⊗ I v = 0. Let α be a simple positive root. Consider

Xα (m)Yα ⊗ I v = Yα ⊗ I Xα (m)v + hα ⊗ I v = 0.

Let α1 be a simple positive root different from α. Then clearly Xα1 (m)Yα ⊗ I v = 0. Hence
Yα ⊗ I v is a highest weight vector. Since V (ψ) is an irreducible highest weight module
we conclude that Yα ⊗ I v = 0.
Now arguing as earlier by induction on the height β we have Yβ ⊗ I v = 0. Hence the
claim.

Thus we have G  ⊗ I v = 0. Now consider the non-zero submodule

W = w ∈ V (ψ), G  ⊗ I w = 0

of V (ψ). Since V (ψ) is irreducible we have W = V (ψ).

332 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

Let Vµ (ψ) be a weight space of V (ψ). Then by the PBW theorem any vector of Vµ (ψ)
is a linear combination of the vectors of the form

(3.8) Yα1 t m1 Yα2 t m2 · · · Yαk t mk such that − αi + λ = µ where λ = ψ | h and αi > 0.
Thus the number of αi that can occur are finite. Now for any finite dimensional
subspace W of V (ψ) and for any fixed αi the space Yαi ⊗ t m W , for m ∈ Zn is finite
dimensional.

Thus at every stage in equation of (3.8) we get a finite dimensional space. So we conclude
that Vµ (ψ) is finite dimensional. 2

3.9. Remark. (1) V (ψ) has finite dimensional weight space if and only if G  ⊗ I.V (ψ) = 0
for a co-finite ideal I of A.
(2) This lemma also gives new modules with finite dimensional weight spaces for affine
Lie algebras by taking G to be a finite dimensional simple Lie algebra and n = 1.

We will now give a continuous family of irreducible highest weight modules for G A
with finite dimensional weight spaces.
As earlier let n be a positive integer. For each i, 1
i
n, let Ni be a positive integer.
Let a i = (ai1 , . . . , aiNi ) be non-zero distinct complex numbers. Let N = N1 · · · Nn . Let
m m1 mn
I = (i1 , . . . , in ) where 1
ij
Nj . Let m = (m1 , . . . , mn ) ∈ Zn . Define aI = a1i1
· · · anin
.
Let Φ be the Lie algebra homomorphism defined by

m
(3.10) Φ : GA → N -copies G = GN , X ⊗ t m → (aI X). G could be any Lie algebra. This
map was first defined by Kac–Jakobsen for n = 1.

3.11. Lemma.

(a) Φ is surjective.
Nj
(b) Let Pj (tj ) = k=1 (tj − aj k ) and I be the ideal generated by P1 (t1 ), . . . , Pn (tn )
inside A. Then G ⊗ A/I ∼= GN .

Proof. (a) We will first prove that the following N × N matrix is invertible:
 m1 mn 
X = a1i 1
· · · a1in 0
mi
Ni −1
.
1
ij
Nj

The index m determines rows and the index (i1 , . . . , in ) determines columns. For n = 1,
X becomes a Vandermonde matrix as a1i = a1j . Hence X is invertible. We will now prove
this for n = 2 and then extend it for all n.
Given a square matrix A, we call a square matrix of the form
 
A 0 ··· 0
 0 A ··· 0 
 
··· ··· ··· ···
0 0 ··· A
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 333

j −1
a block diagonal matrix of A. Let B = aa1i 1
i,j
N1 and

j −1
C = a2i 1
i,j
N2 .

For K = 1, 2, since aKi are distinct non-zero complex numbers for distinct i, B and C are
Vandermonde matrices and hence they are invertible. Let B  be the N × N block diagonal
  
matrix of B. Similarly C for C. Clearly both B and C are invertible.
Let σ : 1, 2, . . . , N → 1, . . . , N be such that

σ (p) = (m − 1)N2 + ( + 1)

where p = N1  + m, 0

N2 − 1, 1
m
N1 (recall that N1 N2 = N). Clearly σ is

 = (c̃˜k, ) where c̃˜k, = c̃σ (k), , C
injective. Let D  = (c̃k, ) and B
 = (b̃k,). Then clearly D


is a product of permutation matrix and C and hence invertible. We now claim that up to a
permutation matrix X = B D which is invertible.
Consider (k, )th entry of BD:

N
(3.12) j =1 b̃kj c̃σ (j ) .

Claim. Exactly one and only one term in (3.12) is non-zero. Let

k = p1 N1 + q1 , 0
p1
N2 − 1, 1
q1
N1 ,
 = sN2 + t, 0
s
N1 − 1, 1
t
N2 .

 is a block diagonal matrix of B, for b̃kj to be non-zero in (3.12), j should be

Since B
equal to p1 N1 + q2 for some 1
q2
N1 .
Now σ (j ) = (q2 − 1)N2 + (p1 + 1). Now since C  is a block diagonal matrix, for C
σ (j ),
to be non-zero  should be equal to (q2 − 1)N2 + t which forces q2 − 1 = s. Thus (3.12)
is equal to

s−1 t −1
a1q a
1 2(p1 +1)
.

This proves the claim. Notice that the exact placement of the entries are given by the
permutation matrix. But the entries of the rows and columns do not get mixed up. Hence
X=B D up to permutation.

j −1
Now we will prove the result for any n. Let Ak = (aki )1
i,j
Nk and let Ãk be a
block diagonal matrix of Ak . Consider Ã1 and Ã2 . Let E be a permutation matrix given
as in the case n = 2. The only difference is that E is actually a block diagonal matrix of
a permutation matrix of order N1 N2 . Now Ã1 E B 2 is a block diagonal matrix of a matrix
of order N1 N2 . Now take Ã1 E Ã2 in place of Ã1 and Ã3 in place of Ã2 in the case n = 2.
Repeat the process to get a matrix of the form Ã1 E Ã2 F Ã3 (F is a permutation matrix)
334 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

which is a block diagonal matrix of a matrix of order N1 N2 N3 . Each entry of this matrix
is of the form

j −1 j −1 j −1
a1i11 a2i22 a3i33 .

Repeating this process we get the desired matrix X which is invertible.

Let (X1 , . . . , XN ) ∈ GN . Now Φ(X−1 (X1 , . . . , XN )T ) = (X1 , . . . , XN ). This com-
pletes the proof of Lemma 3.11(a).
(b) G ⊗ Pj (tj ) ⊂ ker Φ, since the restriction of Φ is nothing but the evaluation map
at the roots of Pj (tj ). Thus we have G ⊗ I ⊆ ker Φ. Consider the space T = {G ⊗ t m ,
0
mi < Ni }. Since any G ⊗ P can be reduced to the linear combinations of elements
of T modulo G ⊗ I , it is a spanning set. In (a) we actually proved Φ is surjective on T .
Further Φ is injective on T as the corresponding matrix is invertible. Thus it follows that
G ⊗ A/I ∼ = GN . 2

3.13. Remark. The condition aij = aik is necessary. Otherwise Φ is not surjective.

Let V (λi ) be irreducible highest weight module for G.Then it is known by Lemma 9.10
of [15] that V (λi ) is irreducible G  -module. Thus V = N 
i=1 V (λi ) is an irreducible GN -
  
module. Restrict the map Φ in (3.10) to GA ⊕ h whose image contains GN . Thus V is an
irreducible GA ⊕ h module via Φ. Consider I = (i1 , . . . , in ) for 1
ij
Nj . There are
N of them. Give them an order say I1 , . . . , IN . The map Φ is defined in this order. Now
define
N m 

(3.14) ψ : U (H ) → C by ψ(b ⊗ t m ) = i=1 aIi λi (b), b ∈ h , and ψ(b) = λi (b), b ∈ h.

Then it is easy to see that V is a highest weight module with highest weight ψ. Thus
we have V (ψ) ∼= V . Now define

(3.15) ψ : U (H ) → A as Zn -graded homomorphism by ψ(b ⊗ t m ) = ψ(b ⊗ t m )t m .

A we need to know that A is an irreducible h̃A -module.

So in order to get modules for G ψ
(See Proposition 3.5 and Lemma 3.1.) It is well known that V (λi ) is integrable if and only
if λi is dominant integral. Since our interest is in constructing integrable modules we will
assume that each λi is dominant.

3.16. Lemma. Let ψ be as above. Assume each λi is dominant and not all of them are zero.
Then Aψ is an irreducible h̃A -module.

Proof. Let Ij = (i1 , . . . , in ) and let λj = λi1 i2 ···in . Consider Γ = {m ∈ Zn , t m ∈ Aψ }.


Since ψ is an algebra homomorphism, Γ is closed under addition. ψ(b ⊗ 1) = λi1 ···in (b)
and each weight is dominant and not all are zero. So ψ(b ⊗ 1) = 0 for some b. Hence
0 ∈ Γ . To prove the lemma it is sufficient to prove that Γ is a subgroup (see Lemma 3.1).
For 1
j
N consider ψ j = ψ | h ⊗ C[tj , tj−1 ].
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 335

k −kj
Claim. Image ψ j = C[tj j , tj ] for some kj
Nj . Consider

Nj
 m    m m
ψ j b ⊗ tj j = λi1 ···in (b)aj ijj tj j .
ij =1 (i1 ,...,in )

 mj
(i1 ,...,in ) λi1 ···in (b) is non-zero for some b. Now ψ j (b ⊗ tj )
Clearly for a fixed ij one of
mj
= 0 for 0 < mj
Nj cannot hold as (aj ij ) is a Vandermonde matrix. The same holds for
k −k
−Nj
mj < 0. Thus the image has to be C[tj j , tj j ] for some 0 < kj
Nj .

In fact kj | Nj see Lemma 4.4 of [6]. Thus we have

(3.17) 1 k1 Z ⊕ · · · ⊕ n kn Z ⊆ Γ for any 1 , . . . , n ∈ Z, 0 < ki
Ni .

Let Γ0 = {(m1 , . . . , mn ) ∈ Γ , 0
mi < ki } is a finite set. To prove the lemma it

is sufficient to prove that each element of Γ0 has an inverse in Γ . This is in view of
(3.17). Let m = (m1 , . . . , mn ) ∈ Γ0 be such that m1 is non-zero and minimal. (If the
all first co-ordinates if Γ0 are zero we take the next one.) We will first prove that there
exist m1 ∈ Γ such that m11 = −m1 . Let k1 = 1 m1 + s1 , 0
s1 < m1 , 1 > 0. Consider
x = (1 + 1)m − (k1 , . . . , kr ) ∈ Γ . Then (1 + 1)m1 − k1 = −s1 + m1
m1 . Further
0 < −s1 + m1 . Now we can assume that x ∈ Γ0 without changing the first co-ordinate
in view of (3.17). Now by minimality of m1 it will follow that −s1 + m1 = m1 . Now
consider m + 1 m − (k1 , . . . , kr ) = (0, ∗, ∗, . . .) ∈ Γ . Take any m1 ∈ Γ0 . Assume that
m11 = 0 write m11 = 11 m1 + s1 , 0
s1 < m1 . Arguing as earlier we see that s1 = 0. Hence
y 1 = m1 − 11 m = (0, ∗, . . . , ∗) ∈ Γ . We can further assume that y 1 ∈ Γ0 . Thus we have
proved that given m ∈ Γ0 there exists m1 ∈ Γ such that m + m1 = (0, ∗, . . . , ∗) ∈ Γ0 .
Repeating this process we get inverses of all elements of Γ0 . 2

3.18. Theorem. Let G be Kac–Moody Lie algebra. Let A = An = C[t1±1 , . . . , tn±1 ] be the
ring of Laurent polynomials in n commuting variables. Let ψ and ψ be defined as in (3.14)
and (3.15) with each λi dominant integrable. Then V (ψ) is an integral irreducible module
A with finite dimensional weight spaces. Furthermore V (ψ) is isomorphic to the first
for G
component of V (ψ) ⊗ A.

Proof. From Lemma 3.16, Aψ is irreducible. Hence V (ψ) is a irreducible highest weight
module from Proposition 3.3. From Proposition 3.5 it will follow that V (ψ) is isomorphic
to the first component of V (ψ) ⊗ A. Since each λi is dominant integral
 V (λi ) is integrable.
Since each V (λi ) is a highest weight module it is known that V = N i=1 V (λi ) is module
with

finite dimensional weight space with respect to h where h is included diagonally in
h. From Lemma 3.11 it follows that the map defined in (3.10) is surjective. Hence V ∼ =
V (ψ). Since V (ψ) has finite dimensional weight spaces it will follow from Lemma 3.6
that V (ψ) has finite dimensional weight spaces with respect to h̃ = h ⊕ D. 2

3.19. Remark. Most often Aψ = A. See (1.7) of [7]. In this case V (ψ) = V (ψ) ⊗ A.
336 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

A with finite
3.20. Proposition. Suppose V (ψ) is an integrable highest weight module for G
dimensional weight spaces. Then

  
N
for h ∈ h
m
ψ h⊗t =
m
aIj λj (h)t m
j =1

m
for some distinct non-zero scalar (ai1 , . . . , aiNi ) and aIj are as defined in (3.10) and
each λj is dominant integral.

Proof. Let ψ = E(1) ◦ ψ where E(1)t m = 1. Let v be a highest weight vector of V (ψ).
Then by Lemma 3.6, V (ψ) has finite dimensional weight spaces. Then from the proof of
Lemma 3.7, V (ψ) is a module for G  ⊗ A/I where I is a co-finite ideal of A. Further
I is generated by polynomials Pj (tj ) for 1
j
n. We can assume that Pj (tj ) has no
zero roots as one can multiply Pj (tj ) by tj− ,  > 0, and will get the same ideal I . Further
we can assume that each polynomial Pj is not a constant. In case Pj is constant then the
module will be trivial. Let aj 1 , . . . , aj Nj be distinct non-zero roots of Pj (tj ). Let

Nj

Pj1 (tj ) = (tj − aj k ).
k=1

Let I  be a co-finite ideal generated by Pj1 (tj ) inside A.

Claim. G  ⊗ I  is zero on V (ψ). Consider the map Φ : G  ⊗ A/I → G  ⊗ A/I  → 0. It is

easy to verify that ker Φ is solvable. Let α be a simple root of G. Let Xα , Yα , Hα be a sl2 -
triple and let Gα be the span of Xα , Yα and Hα . Consider G α = Gα ⊗ A/I which is finite
dimensional. Let W be G α module generated by v. Since V (ψ) is integrable and Xα ⊗ P
acts trivially on v and Hα ⊗ P acts as scalars, we conclude that W is finite dimensional.
By restricting the action of the solvable Lie algebra Gα ⊗ I  /I to W we get a vector w
in W (by Lie’s Theorem) such that Gα ⊗ I  /I acts as scalars on w. From the proof of the
Proposition 2.1 of [8] (we assumed the module is irreducible but for this conclusion we do
not need it) we get that

(3.21) Gα ⊗ (I  /I ) w = 0.

Subclaim. Gα ⊗ (I  /I ) v = 0. By definition of W we have a Y ∈ U (Yα ⊗ A/I ) such that

Y v = w. Since V (ψ) is irreducible there exists X ∈ U (τ ) such that

Xw = v = XY v

where X = X− H X+ , X+ ∈ U (N + ⊗ A), X− ∈ U (N − ⊗ A), H ∈ U (h ⊗ A). First note

that the weights of V (ψ) are of the form λ − ni αi where λ = ψ | h and α1 , . . . , αn the
simple positive roots and n1 , . . . , na are non-negative integers. The weight of w is λ − sα
for s  0. Thus for these reasons X+ ∈ U (Xα ⊗ A). Further X− has to be constant. Thus
X is linear combinations of products of the form
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 337

(3.22) U (h ⊗ A)Xα Q1 · · · Xα Q , Qi ∈ A.

First we will see that

(3.23) Gα ⊗ I  U (h ⊗ A)w = 0 by (3.21).

Now consider

(3.24) Gα ⊗ I  Xα Q1 w = Xα QGα ⊗ I  w + [Xα , Gα ] ⊗ QI  w = 0.

Both terms are zero by (3.21). Now by induction on  we see that

Gα ⊗ I  Xα Q1 · · · Xα Q w = 0.

From (3.22), (3.23) and (3.24) it follows that Gα ⊗ I  Xw = Gα ⊗ I  v = 0. This proves the
subclaim.

Since h is spanned by Hα , α simple we have

h ⊗ I  v = 0.

We will now show that Yβ ⊗ I  v = 0 for any positive root β where Yβ is a root vector
of root −β. We do this by induction on the height of β. We clearly know this for β such
that height β = 1 by (3.21).
Let α1 be any simple root. Consider

Xα Yβ ⊗ I  v = Yβ ⊗ I  Xα v + [Xα , Yβ ] ⊗ I  v.

The first term is zero since v is a highest weight vector. The second term is zero by
induction. Thus we have proved that Yβ ⊗ I  v is a highest weight vector in an irreducible
highest weight module V (ψ). Hence for weight reasons Yβ ⊗ I  v = 0.
Now consider

 = w ∈ V (ψ), G  ⊗ I  w = 0
W

 = V (ψ). This proves the claim.

which is a non-zero submodule of V (ψ). Hence W

Thus we have a module for G  ⊗ A/I  . By by Lemma 3.11(b) we have G  ⊗ A/I  ∼ 

= GN
 
where N = N1 · · · Nr . Consider h sitting in the ith place of GN which acts as linear
function on the highest weight vector v of V (ψ). By standard theory of integrable modules
it follows that λi is dominant integral weight. From the map given in Lemma 3.11(a) it
follows that
   m  
ψ h ⊗ t m = aIi λi h for h ∈ h . 2
338 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

4. Section

We will now extract two classes of integrable irreducible modules for toroidal Lie
algebra τ with finite dimensional weight spaces.

4.1. Example. Z = 0 case. Take G to be finite dimensional as simple Lie algebra G̊ as

in Theorem 3.18. Then V (ψ) is an integrable irreducible module with finite dimensional
weight spaces for the toroidal Lie algebra τ where center Z acts trivially.

4.2. Example. Z = 0 case. Let G̊ be a simple finite dimensional Lie algebra. Let Gaf =
G̊ ⊗ C[t1 , t1−1 ] ⊕ CC1 ⊕ Cd1 be the non-twisted affine Kac–Moody Lie algebra. Consider
the following Lie algebra homomorphism


n 
n
Φ  : G̊ ⊗ A ⊕ Z ⊕ 
Cdi → Gaf ⊗ An−1 ⊕ Cdi
i=1 i=1

given by

(1) Φ  is the identity on G̊ ⊗ An ,

(2) Φ  (di ) = di , 
0 if m1 = 0,
(3) Φ  (t m Ki ) = 0 if m1 = 0, 2
i
n,
C1 ⊗ t m if m1 = 0, i = 1.

Here An−1 = C[t2±1 , . . . , tn±1 ] and G̊ ⊗ A is identified inside Gaf ⊗ An−1 by X ⊗ t m →

(X ⊗ t1m1 ) ⊗ t m where m = (m2 , . . . , mn ). Now take Gaf to be the Kac–Moody Lie algebra
in Theorem 3.18  and take n − 1 instead of n. Then V (ψ) is an irreducible integrable
Gaf ⊗ An−1 ⊕ ni=1 Cdi module. Then V (ψ) is a module for τ via the homomorphism Φ  .

Towards the end of this paper we prove that the above two classes are the only irre-
ducible integrable modules with finite dimensional weight spaces upto an automorphism
of τ .
We will now recall certain automorphisms of τ constructed in Section 4.3 of [7]. Let

A = (aij ) 1
i
n
1
j
n

be an element of GL(n, Z) the group of integral matrices of order n with determinate ±1.
Let r = (r1 , . . . , rn ), s = (s1 , . . . , sn ) ∈ Zn . Let ei = (0, . . . , 1, . . . , 0) be such that 1 in
the ith place and zero everywhere else. Let Ar T = mT , As T = d T where m, d ∈ Zn . Let
a(i) = (a1i , . . . , ani ) so that AeiT = a(i)T , where T denotes the transpose. We now define
an automorphism of τ again denoted by A:
       
A X ⊗ t r = X(m), A d tr ts = d tm · td .
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 339

Let (d11 , . . . , dn1 )T = (AT )−1 (d1 , . . . , dn )T . Define A(di ) = di1 . It is straightforward to
check that A defines an automorphism of τ . A does not preserve the natural Zn -gradation
of τ .
We note that Z does not commute with τ but commutes with G̊ ⊗ A ⊕ Z. In spite of
this we call them central operators since they are as good as central.
Let V be an irreducible τ module with finite dimensional weight spaces. We have the
following lemmas.

4.3. Lemma.

(1) Let z ∈ Z be a homogeneous element of degree m such that zv = 0 for some v in V .

Then zw = 0 for all non-zero w in V .
(2) The zero degree central operators Ki = Ci act by scalars.

Proof. Let W = {v ∈ V : zv = 0}. Consider zX ⊗ t s v = X ⊗ t s zv = 0 for v in W .

Further zdi v = (di z + mi z)v = 0 for v in V . Hence W is a submodule of V . Since V
is irreducible we should either have V = W or W = 0. But we know that W = V and
hence W = 0.
(2) The zero degree central operators Ki leaves each finite dimensional weight space
invariant. Thus Ki has a eigenvector v of eigenvalue . Since V is generated by v, Ki
(central) should act by the same scalar  everywhere. This argument holds good for any
operator on V of zero degree which commutes with τ and leaves a finite dimensional space
invariant. 2

We will now prove an important lemma which is crucial for the classification
result.

4.4. Lemma. Let z be a central operator in Z of degree m such that z = 0. Then there
exists a central operator T (need not be in Z) on V of degree −m such that

T z = zT = Id.

Proof. First consider zV a non-zero submodule of V . Hence zV = V . Now given a v in

V there exist a unique w in V such that zw = v. (Uniqueness follow from the fact that
zw1 = 0 implies w1 = 0 by Lemma 4.3(1).) Define T : V → V such that T (v) = w.
Then clearly zT v = zw = v. Now consider
 
z X(r)T v = X(r)zT v = X(r)v.

Hence by definition T (X(r)v) = X(r)T v. Thus T commutes with G ⊗ A and hence

commutes with z. In particular

T zv = zT v = zw = v. 2
340 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

4.5. Theorem. Let V be an irreducible τ module with finite dimensional weight spaces
with respect to h̊ ⊕ Z0 ⊕ D where Z0 is the linear span of K1 , . . . , Kn . Let

L = r, −r: t r Ki = 0 on V for some i .

Suppose the dimension of the C linear span of L is k. Then up to an automorphism of τ

(1) there exists non-zero integers m1 , m2 , . . . , mk and non-zero central operators z1 , . . . , zk

such that the degree of zi is equal to (0, . . . , mi , 0, . . . , 0),
(2) k < n,
(3) t r Ki = 0 for all i and for all r such that rj = 0 for some k + 1
j
n,
(4) t r Ki = 0 for 1
i
k and for all r ,
(5) there exists a proper submodule W of V for G̊A ⊕ Z ⊕ Dk such that V /W has finite
dimensional weight spaces with respect to h ⊕ Z0 ⊕ Dk where Dk is the linear span
of

{dk+1 , . . . , dn }.

Proof. Let T = {z1 z2 · · · z | where zi = t r Kj = 0 or zi = inverse of t r Kj = 0 for some j

and for some r}.
Let L1 = {s ∈ Zn | where s is the degree of some operator in T }. Then clearly
L is a subgroup of Zn . Now by the standard basis theorem there exists s 1 , . . . , s n ∈
1

Zn such that m1 s 1 , . . . , mk s k is a basis of L1 for some non-zero integers mi . Now

we can find an automorphism B such that Bs i = (0, . . . , 1, . . . , 0). Let r i = mi s i
for 1
i
k. Then Br i = (0, . . . , mi , . . . , 0) for 1
i
k. So after twisting the
automorphism we can assume that there exists non-zero central operators z1 , . . . , zk
of degree (m1 , 0, . . . , 0), . . . , (0, . . . , mk , 0, . . . , 0). Let T1 , . . . , Tk be the inverse of
z1 , . . . , zk .

Claim 1. W1 = {z1 v − v | v ∈ V } is a proper G̊A ⊕ Z ⊕ D1 -module.

Note that for i = 1, di commutes with z1 and hence W1 is di -invariant. Clearly W1 is
G̊A ⊕ Z invariant. To see that W1 is proper, just note that W1 cannot contain d1 weight
vectors. (Elementary but not obvious.) Now consider W2 = {z2 v − v: v ∈ V /W }. By the
above argument we can see that W2 is a proper G̊A ⊕ Z ⊕ D2 module. Continuing this
process we see that W = {zi v − v | v ∈ V , 1
i
k} is a proper G̊A ⊕ Z ⊕ Dk -module.

Claim 2. Ti W = W and v − Ti v ∈ W for all v in V .

Consider v − Ti v = zi Ti v − Ti v ∈ W . Now Ti (zj v − v) = zj (Ti v) − Ti v ∈ W . Hence
Claim 2.

Claim 3. Let λ be a weight of V . Let V λ = r Vλ+δr where the sum runs over all r such
that rk+1 = 0 = · · · = rn . Let W λ = V λ ∩ W . Then M = V λ /W λ is finite dimensional.
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 341

To see this first recall that zi is a central operator of degree m i = (0, . . . , mi , . . . , 0).
Consider

S= Vλ+δr , rk+1 = · · · = rn = 0.
|ri |<|mi |

Since each weight space is finite dimensional it follows that S is finite dimensional,
so to prove Claim 3 it is sufficient to prove that every v in Vλ is in S modulo Wλ . Let
v ∈ Vλ+δr , r = (r1 , . . . , rk , 0, . . . , 0). Let

ri = li mi + gi , |gi | < |mi |,

for gi , li ∈ Z. Note that the sign of li depends on the sign of ri and mi . Suppose ri > 0 and
mi < 0 then li
0. Suppose ri > 0 and mi > 0 then li  0.
Let
 −l  l
Z= zi i Ti i .
li <0 li 0

Then by Claim 2 and the definition of W we see that Zv = v modulo W for all v in V .
On the other hand it is easy to see that Zv ∈ S. Hence the claim.

Thus we have proved that V /W is a G̊A ⊕ Z ⊕ Dk -module with finite dimensional

weight space with respect to h ⊕ Z0 ⊕ Dk . This proves (2).
(3) Suppose t r Ki = 0 for some r and such that rj = 0 for k + 1
j
n. This increases
the dimension to k + 1 which is wrong. So we have (3).
(4) Let hk be the linear span of h ⊗ t r , t r Ki for 1
i
k and rj = 0 for k + 1
j
n
and dk+1 , . . . , dn . One can check that hk is solvable. In fact [[hk , hk ], hk ] = 0. Now clearly
Vλ is a submodule for hk and Wλ is a submodule. Therefore M is a finite dimensional
module for hk . Thus by Lie’s Theorem there exists a non-zero vector v in M such
that h(r)v = λ(h, r)v for h(r) ∈ hk . Consider for m, d ∈ Zn such that dj = mj = 0 for
k + 1
j
n. Then
       
h, h d t m t d v = h(m), h (d) v = h(m)h (d) − h (d)h(m) v
     
= λ(h, m)λ h , d − λ h , d λ(h, m) v = 0

Let z = d(t m )t d and suppose z = 0 on V . Then we have just proved that there exists a
non-zero vector v in V /W (there is v ∈ / W ) such that zv ∈ W . But v = z−1 zv ⊆ z−1 W ⊆
W . A contradiction. Hence z = 0. In particular t r Ki = 0 for 1
i
k and rj = 0 for
k + 1
j
n. (See 2.2(1).) This together with (3) proves (4). The second part of (4)
follows from the above.
(5) Suppose k = n. Then by (4), Z has to act by zero which is a contradiction to the fact
that k = n > 0. 2

We record here the following lemma about the dimensions of central operators acting
on V . We will not need it anywhere but it is of independent interest.
342 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

4.6. Lemma. Let z1 , z2 be non-zero central operators of degree m on an irreducible

module V of τ with finite dimensional weight spaces. Then z2 = z1 for some constant .
In particular there is up to scalar at most one non-zero central operator in a given degree.

Proof. Let T1 be the inverse central operator of degree −m of z1 . Consider z2 T1 which is

of degree zero central operator and leaves a finite dimensional space Vg (see the proof of
Lemma 2.6) invariant.
Thus it has a eigenvector v say of eigenvalue . But v generates V and hence z2 T1 = 
on V . Now z1  = z1 z2 T1 = z2 . Hence we are done. 2

We need one more reduction modulo an automorphism of τ before we can take up the
classification problem. Recall that A ∈ GL(n, Z) defines an automorphism of τ such that
Ad(t r )t s = d(t Ar ).t As . (We are suppressing transpose T when there is no confusion.) It is
easy to see that if A = (aij ) then
n
(4.7) A(t s Ki ) = j =1 aij t
A(s) K .
j


By taking s = 0 we have A(Ki ) = nj=1 aij Kj . Let Ki act on V by ki . We know that
there exists k < n such
 that  ki = 0 for 1
i
k up to an automorphism of τ . Now choose
A ∈ GL(n, Z), A = 0I B0 such that I is identity matrix of order k × k and B ∈ GL(n − k, Z)
is such that B(kk+1 , . . . , kn )T = (0, . . . , 0, )T .

4.8. Proposition. Let V be an irreducible module for τ with finite dimensional weight
spaces. Let k be an integer as defined in Theorem 4.5. Then up to automorphism of τ the
assertions of Theorem 4.5 holds and further we can assume Ki = 0 on V for 1
i
n − 1.

Proof. We have Theorem 4.5. Now choose A such that (4.7) holds.
(1) Clearly holds.

(3) Claim. t A(r) Ki = 0 for all i and j such that A(r)j = 0 for some j such that
k + 1
j
n. First note that ri = 0 for some k + 1
i
n by the choice of A. Now
by Theorem 4.5(3) we have t r Ki = 0 for all i. Then from (4.7) and the fact that A in
invertible the claim follows.

(4) Claim. t A(r) Ki = 0, 1
i
k for all r. From (4.7) we have the claim by noting that
aij = 0 for j  k + 1.

(2) and (5) follows from arguments similar to Theorem 4.5(2) and (3). In addition we
can assume that A(Ki ) = 0, 1
i
n − 1. 2

We will now start eliminating several cases in order to classify integrable irreducible
modules for τ with finite dimensional weight spaces.

4.9. Proposition. Let V be an irreducible integrable module for τ with finite dimensional
weight spaces. Suppose k < n − 1 (see Theorem 4.5) and suppose Ki = 0 for some i, then
such a V does not exists.
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 343

Proof. By Proposition 4.8 we can assume that Ki = 0 for 1
i
n − 1 and Kn = 0. Let

Gaf = G̊ ⊗ C[tn , tn−1 ] ⊕ CKn ⊕ Cdn be an affine Lie algebra. Let Gaf = N + ⊕ h ⊕ N −
be a standard decomposition with h = h̊ ⊕ CKn ⊕ Cdn . Now by Proposition 2.4 we can
assume that there exists a weight vector v of weight λ in V such that

N + ⊗ An−1 v = 0.

(The other case can be dealt with similarly.) Here An−1 = C[t1±1 , . . . , tn−1
±1
]. (We have
 t mv = 0
chosen n instead of 1 in Proposition 2.4.) In particular for h ∈ h̊ we have h ⊗ tk+1 n
for all m > 0 and for all .

Claim. For any 0 = h ∈ h̊ the following vectors are linearly independent in V . Fix m > 0
and ,
−d −m

htk+1 tn · htk+1
d
tn−(m+1) v, d ∈ Z .

Suppose there exists non-zero scalars ad such that


−d −m d
B= ad htk+1 tn htk+1 tn−(m+1) v = 0.

Choose h ∈ h̊ such that (h, h ) = 0. Consider h tk+1

s
tnm+1 B = 0 for any s, which implies


−d −m  s
0= ad htk+1 tn h tk+1 tnm+1 htk+1
d
tn−(m+1) v
 
+ h, h −d+s
ad stk+1 d
tn Kk+1 htk+1 tn−(m+1) v
 
+ h, h −d+s
ad (m + 1)tk+1 d
tn Kn htk+1 tn−(m+1) v

(by 2.2(1)).
−d+s −d+s
Being central the term tk+1 tn Kk+1 and tk+1 tn Kn are zero (by Theorem 4.5(3)).
Thus the second and third term above are zero.
The first term is equal to

−d −m d
0= ad htk+1 tn htk+1 tn−(m+1) h tk+1
s
tnm+1 v
 
+ h, h −d −m s+d
ad htk+1 tn stk+1 Kk+1 v
  
+ h, h −d −m
ad htk+1 s+d
tn (m + 1)tk+1 Kn v

(by 2.2(1)).
s+d
Fix a d0 in the above and let s = −d0 . Now from (2.1), tk+1 Kk+1 = 0 for s + d = 0.
And Kk+1 = 0 by Proposition 4.8. Thus the second term is zero in the above. Now by
344 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

s+d
Theorem 4.5(3), tk+1 Kn = 0 for s + d = 0. Hence the third term is zero but for d0 . The
first term is zero as v is a highest weight vector. Hence we have

+s −m
a−s htk+1 tn · Kn v = 0.

+s −m
(4.10) Suppose htk+1 tn v = 0.

Consider
−(+s) m +s −m    
0 = h tk+1 tn htk+1 tn v = − h, h ( + s)Kk+1 v + m h, h Kn v

(by 2.2(1)). We also know that Kk+1 v = 0. But the second term is non-zero. Thus (4.10) is
false. This proves a−s = 0 a contradiction. This proves our claim.

Hence we have proved that under the conditions of Proposition 4.9. Vλ+δk+1 −(2m+1)δn
is infinite dimensional. This completes the proof of the proposition. 2

4.11. Remark. The case k = n − 1 where modules exists (see Example 4.2) will be dealt
in the next section. The case k = 0 and Ki = 0 for all i, in which case Z = 0 is dealt with
in [8].

Now we will deal the remaining case where k  1 and Ki = 0 for all i.
We will first recall an important result due to Futorny [13] on Heisenberg Lie Algebra.
Let H be a finite dimensional vector space with non-degenerate symmetric bilinear form
( , ). Then L(H ) = H ⊗ C[t, t −1 ] ⊕ Cc is called Heisenberg Lie algebra with the following
bracket
   
h ⊗ t m , h ⊗ t  = h, h mδm+,0 C.

4.12. Proposition (Proposition 4.3(i) [13]). Let V be any Z-graded, L(H )-module with
finite dimensional graded spaces and suppose the center acts by nonzero scalars. Then V
admits a graded vector v such that H ⊗ t n v = 0 for all n > 0 (or for all n < 0).

Proof. It is only proved for one dimensional H . But the proof works for any finite
dimensional H by choosing an orthogonal basis for H . There it is assumed that V is
irreducible, but it is not needed for getting a highest weight (or lowest weight) vector. 2

4.13. Proposition. Let V be an integrable irreducible module for τ with finite dimensional
weight spaces. Let k be as in Theorem 4.5. Suppose k  1 and Ki = 0 for all i. Then such
a module V does not exists.

Proof. Recall that G̊ is a finite dimensional simple Lie algebra with G̊ = n+ ⊕ h̊ ⊕ n− .

Then by Proposition 2.12 there exists a weight vector v of V such that n+ ⊗ An v = 0. (The
case n− ⊗ An v = 0 can be done similarly.) Suppose t m Ki = 0. Then by Theorem 4.5(3),
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 345

(4) we have i  k + 1 and mk+1 = 0 = · · · = mn . Let H be the Heisenberg Lie algebra

spanned by ht m tik , k > 0, hti−k , k > 0 and t m Ki with Lie bracket.
   
ht m tik , h ti = h, h kt m Ki δk+,0.

Consider M the H module generated by v. Then by Proposition 4.12 there exists w in M

such that

(a) ht m tik w = 0, k > 0, or

(b) htik w = 0 for k < 0.

Assume (a). Now w = Xv for X ∈ U (hA ). Then it is easy to see that n+ ⊗ An w = 0.

Let λ be the weight of w.

Claim. λ | h̊ = 0. Suppose it is zero. Let α be a simple root in ∆˚ and let Xα , Yα , hα =

[Xα , Yα ] be an sl2 copy inside G̊. Then Xα ⊗ t s , Yα ⊗ t −s , hα is an sl2 copy (because there
is no zero degree centre). Now by sl2 theory for integrable module Yα ⊗ t −s w = 0 for any
s as λ(hα ) = 0. Thus we have G̊ ⊗ An w = 0. This implies Z is zero which is not the case.
Thus λ | h̊ = 0.

Let α be a simple root such that λ(hα ) = 0. Let Xα = X, Yα = Y , h = hα be an sl2

copy. That is [X, Y ] = h, [h, Y ] = −2Y , [h, X] = 2X. Let 0 be such that

(4.14) (ht m − 0 (X, Y )t m Ki )w = 0 (where by Xt s we mean X ⊗ t s ).

Claim. The following infinite set of vectors are linearly independent in V :

−r r

Y ti Y ti w, r > 0, r = 0 , −0 .

Suppose there exists non-zero scalars ar such that


(4.15) ar Y ti−r Y tir w = 0.

We will be using the following in the calculation below:

(1) w is a highest weight vector,

(2) ti Ki = 0 which follows from Definition 2.2, for  = 0 and by assumption for  = 0,
(3) d(t m ti )ti−r = rt m ti−r Ki (by (2.2) and (2.1) and mi = 0).

Consider for r > 0,

Xtis Xt m ti Y ti−r Y tir w = htis−r ht m ti+r w − r(X, Y )htis−r · t m ti+r Ki w

+ htir+s ht m ti−r w − 2ht m ti+s w
− 2s(X, Y )t m ti+s Ki w + r(X, Y )t m ti−r Ki htir+s w.
346 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

Choose  such that  = ±0 and  − r > 0 which implies  + r > 0. Then first three
terms and the sixth term are zero. Thus applying Xtis Xt m ti to (4.15) such that  − r > 0,
we have for all r
  
ar ht m ti+s + s(X, Y )t m ti+s Ki w = 0.

Choose s such that  + s = 0. From (4.14) it follows that


(4.16) ar = 0.

Now choose r0 be the maximal among r that occur in (4.15). Now choose  such that
 − r0 = 0,  − r > 0 for r = r0 . This implies  + r > 0, ∀r. Again apply Xtis Xt m ti to
(4.15) we have

r +s
  
ar0 hti 0 ht m w − 2 ar ht m ti+s + s(X, Y )t m ti+s Ki w
r +s
+ r0 ar0 (X, Y )t m Ki hti 0 w.

Now choose s such that r0 + s = 0. Then ar0 (ht m + r0 (X, Y )t m ki )hw = 0. By choice of
r and the fact that hw = λ(h)w = 0, we conclude that ar0 = 0, which is a contradiction.
Thus Vλ+2α is infinite dimensional.

(b) htik w = 0, k < 0. Consider the set Y tir Y ti−r , r > 0, and apply Xt m tis Xti . Then we
get the desired linearly independent set. 2

5. Section

In this section we will deal with the last case k = n − 1, Kn = 0 and Ki = 0 for
1
i
n − 1. We will prove that in this case we will get all the modules defined in
Example 4.2.

5.1. Lemma. Any Zn−1 graded simple, commutative algebra M such that each graded
component is finite dimensional over C is isomorphic to a subalgebra of An−1 such that
each homogeneous element is invertible.


Proof. Let M = r∈Zn−1 Mr where each Mr is finite dimensional. Since M is graded
simple it follows that M0 is simple commutative algebra of finite dimension over C. Then
clearly M0 ∼ = C. Let 0 = w ∈ Mr . The ideal generated by w has to be M and hence
there exists an inverse say w−1 . Consider w−1 Mr ⊆ C. Since w−1 w = 1 it follows that
w−1 Mr = C. In particular each non-zero Mr is C. The lemma follows. 2

5.2. Theorem. Let V be an irreducible integrable models for τ with finite dimensional
weight spaces. Let k be as defined in Theorem 4.5. Assume k = n − 1 and Kn = 0. Then V
is isomorphic to V (ψ) as defined in Example 4.2.
 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348 347

Proof. First note that t r Ki = 0 for all r and 1
i
n − 1 and t r Kn = 0 for all r such that
rn = 0. Write

(5.3) τ = N − ⊗ An−1 ⊕ h̊ ⊗ An−1 ⊕ rn =0 t
rK
n ⊕ D + Z  ⊕ N + ⊗ An−1

where Z  is spanned by t r Ki for 1
i
n − 1 for all r and t r Ki for all r such that rn = 0.
We have N − ⊕ h̊ ⊕ CKn ⊕ Cdn ⊕ N + = Gaf = G̊ ⊗ C[tn , tn−1 ] ⊕ CKn ⊕ Cdn . We have
already noted that Z  acts trivially on V . By Lemma 2.3(5) we know that Kn acts by an
integer. By an automorphism we can assume that Kn acts by a positive integer. Now by
Proposition 2.4 we get a weight vector v in V such that

(5.4) N − ⊗ An−1 v = 0.

Let h be the abelian Lie algebra spanned by h̊ ⊗ An−1 ⊕ rn =0 t
rK .
n Here An−1 =
C[t1±1 , . . . , tn−1
±1
]. Let M be a h ⊕ D-module generated by v.

Claim. M is an irreducible h ⊕ D-module. Let w be a weight vector of M. Since V

is irreducible there exists X ∈ U (G̊A ⊕ Z  ) such that Xw = v. By the PBW theorem

X = X− H X+

where X+ ∈ U (N + ⊗ An−1 ), X− ∈ U (N − ⊗ An−1 ) and H ∈ U (h ) from (5.3). D and Z 

does not appear as w is a weight vector and Z  acts trivially. But since Y v = w for some
Y ∈ U (h ) we can see that N + ⊗ An−1 Y v = 0. Thus X+ w = 0, which means X+ cannot
appear and by weight reasons X− cannot appear. Thus X = H which belongs to U (h ).
This proves the claim.

Since v is a weight vector, D acts by a scalar and hence M = U (h )v. So we have M

a Zn−1 graded irreducible module for h . Thus M ∼ = U (h )/I for some graded ideal I .

Since h is abelian M is Z n−1 graded simple commutative algebra. Now by Lemma 5.1,
M is isomorphic to a subalgebra of An−1 . Let ψ : U (h ) → M = Aψ ⊆ An−1 be the
quotient map which is Zn−1 graded. By Proposition 3.3, we have V ∼ = V (ψ). As earlier
ψ = E(1) ◦ ψ where E(1) t m = 1.
 
 h = h̊ ⊕m CKn and identify t Kn as Kn ⊗ t (mn = 0). Thus h ⊗ An−1 = h̊ ⊗ An−1 ⊕
m m

mn =0 Ct Kn . From Proposition 3.20 we have

  N
for h ∈ h and m ∈ Zn−1 .
m
ψ h ⊗ tm = aIJ λj (h)
j =0


Now extend each λj to h that is give some value to λj (dn ) such that ψ(dn ) = λj (dn ). It
is well known that the irreducible integrable highest weight module V (λj ) for Gaf are all
 ∼
isomorphic for various values of λj (dn ) (see [15]). Now N j =1 V (λj ) = V (ψ) being the
unique irreducible module for the highest weight ψ. 2
348 S. Eswara Rao / Journal of Algebra 277 (2004) 318–348

5.5. Remark. We need to consider more general modules than in [8] where center acts
trivially like in our Example 4.1. Nevertheless we proved that an irreducible integrable
module (not graded) G̊A is actually a module for G̊ ⊗ A/I (Lemma 1.2 and Proposition 2.1
of [8]) where I is a cofinite ideal of A generated

by polynomials with distinct roots. But
our Lemma 3.11(b) says that G̊ ⊗ A/I ∼ = G̊. Thus in this case all irreducible integrable
modules with finite dimensional weight spaces for G̊A are given in Example 4.1.

Acknowledgments

I record my sincere thanks to Field’s Institute’s (Toronto) hospitality during the Fall of
2000 where some of the work has been done. I also thank V. Futorny for bringing to my
notice the references [12] and [13].

References

[1] P. Batra, Representations of twisted multi loop Lie algebra, J. Algebra 272 (2004) 404–416.
[2] S. Berman, Y. Billig, Irreducible representations for toroidal Lie algebras, J. Algebra 221 (1999) 188–231.
[3] S. Berman, B. Cox, Enveloping algebras and representations of toroidal Lie algebras, Pacific J. Math. 165 (2)
(1994) 239–267.
[4] S. Berman, Y. Gao, Y. Krylyuk, Quantum tori and the structure of elliptic quasi-simple Lie algebras, J. Funct.
Anal. 135 (1996) 339–389.
[5] V. Chari, Integrable representations of affine Lie algebras, Invent. Math. 85 (1986) 317–335.
[6] V. Chari, A.N. Pressley, New unitary representations of loop groups, Math. Ann. 275 (1986) 87–104.
[7] S. Eswara Rao, Iterated loop modules and a filtration for vertex representation of toroidal Lie algebras,
Pacific J. Math. 171 (2) (1995) 511–528.
[8] S. Eswara Rao, Classification of irreducible integrable modules for multi-loop algebras with finite
dimensional weight spaces, J. Algebra 246 (2001) 215–225.
[9] S. Eswara Rao, A generalization of irreducible modules for toroidal Lie algebras, TIFR preprint, 2000.
[10] S. Eswara Rao, Irreducible representations for toroidal Lie algebras, TIFR preprint, 2003, math.RT/0212137.
[11] S. Eswara Rao, R.V. Moody, Vertex representations for n-toroidal Lie algebras and a generalization of the
Virosoro algebra, Comm. Math. Phys. 159 (1994) 239–264.
[12] V. Futorny, Representations of Affine Lie Algebras, in: Queen’s Papers in Pure and Appl. Math., 1997.
(1)
[13] V. Futorny, Irreducible non-dense A1 -modules, Pacific J. Math. 172 (1996) 83–100.
[14] J.E. Humphreys, Introduction to Lie Algebras and Representations Theory, Springer, Berlin, Heidelberg,
New York, 1972.
[15] V. Kac, Infinite Dimensional Lie Algebras, 3rd ed., Cambridge University Press, 1990.
[16] R.V. Moody, S. Eswara Rao, T. Yokomuma, Toroidal Lie algebra and vertex representations, Geom.
Dedicata 35 (1990) 283–307.
[17] R.V. Moody, Z. Shi, Toroidal Weyl groups, Nova J. Algebra and Geom. 1 (1992) 317–337.
[18] Y. Yoon, On the polynomial representations of current algebras, J. Algebra 252 (2) (2002) 376–393.

Further reading

[19] S. Eswara Rao, Classification of loop modules with finite dimensional weight spaces, Math. Ann. 305 (1996)
651–663.